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  */
  13
  14 #include "memory.h"
  15 #include "exec-memory.h"
  16 #include "ioport.h"
  17 #include "bitops.h"
  18 #include "kvm.h"
  19 #include <assert.h>
  20
  21 #define WANT_EXEC_OBSOLETE
  22 #include "exec-obsolete.h"
  23
  24 unsigned memory_region_transaction_depth = 0;
  25 static bool memory_region_update_pending = false;
  26 static bool global_dirty_log = false;
  27
  28 static QLIST_HEAD(, MemoryListener) memory_listeners
  29     = QLIST_HEAD_INITIALIZER(memory_listeners);
  30
  31 typedef struct AddrRange AddrRange;
  32
  33 /*
  34  * Note using signed integers limits us to physical addresses at most
  35  * 63 bits wide.  They are needed for negative offsetting in aliases
  36  * (large MemoryRegion::alias_offset).
  37  */
  38 struct AddrRange {
  39     Int128 start;
  40     Int128 size;
  41 };
  42
  43 static AddrRange addrrange_make(Int128 start, Int128 size)
  44 {
  45     return (AddrRange) { start, size };
  46 }
  47
  48 static bool addrrange_equal(AddrRange r1, AddrRange r2)
  49 {
  50     return int128_eq(r1.start, r2.start) && int128_eq(r1.size, r2.size);
  51 }
  52
  53 static Int128 addrrange_end(AddrRange r)
  54 {
  55     return int128_add(r.start, r.size);
  56 }
  57
  58 static AddrRange addrrange_shift(AddrRange range, Int128 delta)
  59 {
  60     int128_addto(&range.start, delta);
  61     return range;
  62 }
  63
  64 static bool addrrange_contains(AddrRange range, Int128 addr)
  65 {
  66     return int128_ge(addr, range.start)
  67         && int128_lt(addr, addrrange_end(range));
  68 }
  69
  70 static bool addrrange_intersects(AddrRange r1, AddrRange r2)
  71 {
  72     return addrrange_contains(r1, r2.start)
  73         || addrrange_contains(r2, r1.start);
  74 }
  75
  76 static AddrRange addrrange_intersection(AddrRange r1, AddrRange r2)
  77 {
  78     Int128 start = int128_max(r1.start, r2.start);
  79     Int128 end = int128_min(addrrange_end(r1), addrrange_end(r2));
  80     return addrrange_make(start, int128_sub(end, start));
  81 }
  82
  83 struct CoalescedMemoryRange {
  84     AddrRange addr;
  85     QTAILQ_ENTRY(CoalescedMemoryRange) link;
  86 };
  87
  88 struct MemoryRegionIoeventfd {
  89     AddrRange addr;
  90     bool match_data;
  91     uint64_t data;
  92     int fd;
  93 };
  94
  95 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd a,
  96                                            MemoryRegionIoeventfd b)
  97 {
  98     if (int128_lt(a.addr.start, b.addr.start)) {
  99         return true;
 100     } else if (int128_gt(a.addr.start, b.addr.start)) {
 101         return false;
 102     } else if (int128_lt(a.addr.size, b.addr.size)) {
 103         return true;
 104     } else if (int128_gt(a.addr.size, b.addr.size)) {
 105         return false;
 106     } else if (a.match_data < b.match_data) {
 107         return true;
 108     } else  if (a.match_data > b.match_data) {
 109         return false;
 110     } else if (a.match_data) {
 111         if (a.data < b.data) {
 112             return true;
 113         } else if (a.data > b.data) {
 114             return false;
 115         }
 116     }
 117     if (a.fd < b.fd) {
 118         return true;
 119     } else if (a.fd > b.fd) {
 120         return false;
 121     }
 122     return false;
 123 }
 124
 125 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd a,
 126                                           MemoryRegionIoeventfd b)
 127 {
 128     return !memory_region_ioeventfd_before(a, b)
 129         && !memory_region_ioeventfd_before(b, a);
 130 }
 131
 132 typedef struct FlatRange FlatRange;
 133 typedef struct FlatView FlatView;
 134
 135 /* Range of memory in the global map.  Addresses are absolute. */
 136 struct FlatRange {
 137     MemoryRegion *mr;
 138     target_phys_addr_t offset_in_region;
 139     AddrRange addr;
 140     uint8_t dirty_log_mask;
 141     bool readable;
 142     bool readonly;
 143 };
 144
 145 /* Flattened global view of current active memory hierarchy.  Kept in sorted
 146  * order.
 147  */
 148 struct FlatView {
 149     FlatRange *ranges;
 150     unsigned nr;
 151     unsigned nr_allocated;
 152 };
 153
 154 typedef struct AddressSpace AddressSpace;
 155 typedef struct AddressSpaceOps AddressSpaceOps;
 156
 157 /* A system address space - I/O, memory, etc. */
 158 struct AddressSpace {
 159     const AddressSpaceOps *ops;
 160     MemoryRegion *root;
 161     FlatView current_map;
 162     int ioeventfd_nb;
 163     MemoryRegionIoeventfd *ioeventfds;
 164 };
 165
 166 struct AddressSpaceOps {
 167     void (*range_add)(AddressSpace *as, FlatRange *fr);
 168     void (*range_del)(AddressSpace *as, FlatRange *fr);
 169     void (*log_start)(AddressSpace *as, FlatRange *fr);
 170     void (*log_stop)(AddressSpace *as, FlatRange *fr);
 171     void (*ioeventfd_add)(AddressSpace *as, MemoryRegionIoeventfd *fd);
 172     void (*ioeventfd_del)(AddressSpace *as, MemoryRegionIoeventfd *fd);
 173 };
 174
 175 #define FOR_EACH_FLAT_RANGE(var, view)          \
 176     for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
 177
 178 static bool flatrange_equal(FlatRange *a, FlatRange *b)
 179 {
 180     return a->mr == b->mr
 181         && addrrange_equal(a->addr, b->addr)
 182         && a->offset_in_region == b->offset_in_region
 183         && a->readable == b->readable
 184         && a->readonly == b->readonly;
 185 }
 186
 187 static void flatview_init(FlatView *view)
 188 {
 189     view->ranges = NULL;
 190     view->nr = 0;
 191     view->nr_allocated = 0;
 192 }
 193
 194 /* Insert a range into a given position.  Caller is responsible for maintaining
 195  * sorting order.
 196  */
 197 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
 198 {
 199     if (view->nr == view->nr_allocated) {
 200         view->nr_allocated = MAX(2 * view->nr, 10);
 201         view->ranges = g_realloc(view->ranges,
 202                                     view->nr_allocated * sizeof(*view->ranges));
 203     }
 204     memmove(view->ranges + pos + 1, view->ranges + pos,
 205             (view->nr - pos) * sizeof(FlatRange));
 206     view->ranges[pos] = *range;
 207     ++view->nr;
 208 }
 209
 210 static void flatview_destroy(FlatView *view)
 211 {
 212     g_free(view->ranges);
 213 }
 214
 215 static bool can_merge(FlatRange *r1, FlatRange *r2)
 216 {
 217     return int128_eq(addrrange_end(r1->addr), r2->addr.start)
 218         && r1->mr == r2->mr
 219         && int128_eq(int128_add(int128_make64(r1->offset_in_region),
 220                                 r1->addr.size),
 221                      int128_make64(r2->offset_in_region))
 222         && r1->dirty_log_mask == r2->dirty_log_mask
 223         && r1->readable == r2->readable
 224         && r1->readonly == r2->readonly;
 225 }
 226
 227 /* Attempt to simplify a view by merging ajacent ranges */
 228 static void flatview_simplify(FlatView *view)
 229 {
 230     unsigned i, j;
 231
 232     i = 0;
 233     while (i < view->nr) {
 234         j = i + 1;
 235         while (j < view->nr
 236                && can_merge(&view->ranges[j-1], &view->ranges[j])) {
 237             int128_addto(&view->ranges[i].addr.size, view->ranges[j].addr.size);
 238             ++j;
 239         }
 240         ++i;
 241         memmove(&view->ranges[i], &view->ranges[j],
 242                 (view->nr - j) * sizeof(view->ranges[j]));
 243         view->nr -= j - i;
 244     }
 245 }
 246
 247 static void memory_region_read_accessor(void *opaque,
 248                                         target_phys_addr_t addr,
 249                                         uint64_t *value,
 250                                         unsigned size,
 251                                         unsigned shift,
 252                                         uint64_t mask)
 253 {
 254     MemoryRegion *mr = opaque;
 255     uint64_t tmp;
 256
 257     tmp = mr->ops->read(mr->opaque, addr, size);
 258     *value |= (tmp & mask) << shift;
 259 }
 260
 261 static void memory_region_write_accessor(void *opaque,
 262                                          target_phys_addr_t addr,
 263                                          uint64_t *value,
 264                                          unsigned size,
 265                                          unsigned shift,
 266                                          uint64_t mask)
 267 {
 268     MemoryRegion *mr = opaque;
 269     uint64_t tmp;
 270
 271     tmp = (*value >> shift) & mask;
 272     mr->ops->write(mr->opaque, addr, tmp, size);
 273 }
 274
 275 static void access_with_adjusted_size(target_phys_addr_t addr,
 276                                       uint64_t *value,
 277                                       unsigned size,
 278                                       unsigned access_size_min,
 279                                       unsigned access_size_max,
 280                                       void (*access)(void *opaque,
 281                                                      target_phys_addr_t addr,
 282                                                      uint64_t *value,
 283                                                      unsigned size,
 284                                                      unsigned shift,
 285                                                      uint64_t mask),
 286                                       void *opaque)
 287 {
 288     uint64_t access_mask;
 289     unsigned access_size;
 290     unsigned i;
 291
 292     if (!access_size_min) {
 293         access_size_min = 1;
 294     }
 295     if (!access_size_max) {
 296         access_size_max = 4;
 297     }
 298     access_size = MAX(MIN(size, access_size_max), access_size_min);
 299     access_mask = -1ULL >> (64 - access_size * 8);
 300     for (i = 0; i < size; i += access_size) {
 301         /* FIXME: big-endian support */
 302         access(opaque, addr + i, value, access_size, i * 8, access_mask);
 303     }
 304 }
 305
 306 static void memory_region_prepare_ram_addr(MemoryRegion *mr);
 307
 308 static void as_memory_range_add(AddressSpace *as, FlatRange *fr)
 309 {
 310     ram_addr_t phys_offset, region_offset;
 311
 312     memory_region_prepare_ram_addr(fr->mr);
 313
 314     phys_offset = fr->mr->ram_addr;
 315     region_offset = fr->offset_in_region;
 316     /* cpu_register_physical_memory_log() wants region_offset for
 317      * mmio, but prefers offseting phys_offset for RAM.  Humour it.
 318      */
 319     if ((phys_offset & ~TARGET_PAGE_MASK) <= IO_MEM_ROM) {
 320         phys_offset += region_offset;
 321         region_offset = 0;
 322     }
 323
 324     if (!fr->readable) {
 325         phys_offset &= ~TARGET_PAGE_MASK & ~IO_MEM_ROMD;
 326     }
 327
 328     if (fr->readonly) {
 329         phys_offset |= IO_MEM_ROM;
 330     }
 331
 332     cpu_register_physical_memory_log(int128_get64(fr->addr.start),
 333                                      int128_get64(fr->addr.size),
 334                                      phys_offset,
 335                                      region_offset,
 336                                      fr->dirty_log_mask);
 337 }
 338
 339 static void as_memory_range_del(AddressSpace *as, FlatRange *fr)
 340 {
 341     if (fr->dirty_log_mask) {
 342         Int128 end = addrrange_end(fr->addr);
 343         cpu_physical_sync_dirty_bitmap(int128_get64(fr->addr.start),
 344                                        int128_get64(end));
 345     }
 346     cpu_register_physical_memory(int128_get64(fr->addr.start),
 347                                  int128_get64(fr->addr.size),
 348                                  IO_MEM_UNASSIGNED);
 349 }
 350
 351 static void as_memory_log_start(AddressSpace *as, FlatRange *fr)
 352 {
 353     cpu_physical_log_start(int128_get64(fr->addr.start),
 354                            int128_get64(fr->addr.size));
 355 }
 356
 357 static void as_memory_log_stop(AddressSpace *as, FlatRange *fr)
 358 {
 359     cpu_physical_log_stop(int128_get64(fr->addr.start),
 360                           int128_get64(fr->addr.size));
 361 }
 362
 363 static void as_memory_ioeventfd_add(AddressSpace *as, MemoryRegionIoeventfd *fd)
 364 {
 365     int r;
 366
 367     assert(fd->match_data && int128_get64(fd->addr.size) == 4);
 368
 369     r = kvm_set_ioeventfd_mmio_long(fd->fd, int128_get64(fd->addr.start),
 370                                     fd->data, true);
 371     if (r < 0) {
 372         abort();
 373     }
 374 }
 375
 376 static void as_memory_ioeventfd_del(AddressSpace *as, MemoryRegionIoeventfd *fd)
 377 {
 378     int r;
 379
 380     r = kvm_set_ioeventfd_mmio_long(fd->fd, int128_get64(fd->addr.start),
 381                                     fd->data, false);
 382     if (r < 0) {
 383         abort();
 384     }
 385 }
 386
 387 static const AddressSpaceOps address_space_ops_memory = {
 388     .range_add = as_memory_range_add,
 389     .range_del = as_memory_range_del,
 390     .log_start = as_memory_log_start,
 391     .log_stop = as_memory_log_stop,
 392     .ioeventfd_add = as_memory_ioeventfd_add,
 393     .ioeventfd_del = as_memory_ioeventfd_del,
 394 };
 395
 396 static AddressSpace address_space_memory = {
 397     .ops = &address_space_ops_memory,
 398 };
 399
 400 static const MemoryRegionPortio *find_portio(MemoryRegion *mr, uint64_t offset,
 401                                              unsigned width, bool write)
 402 {
 403     const MemoryRegionPortio *mrp;
 404
 405     for (mrp = mr->ops->old_portio; mrp->size; ++mrp) {
 406         if (offset >= mrp->offset && offset < mrp->offset + mrp->len
 407             && width == mrp->size
 408             && (write ? (bool)mrp->write : (bool)mrp->read)) {
 409             return mrp;
 410         }
 411     }
 412     return NULL;
 413 }
 414
 415 static void memory_region_iorange_read(IORange *iorange,
 416                                        uint64_t offset,
 417                                        unsigned width,
 418                                        uint64_t *data)
 419 {
 420     MemoryRegion *mr = container_of(iorange, MemoryRegion, iorange);
 421
 422     if (mr->ops->old_portio) {
 423         const MemoryRegionPortio *mrp = find_portio(mr, offset, width, false);
 424
 425         *data = ((uint64_t)1 << (width * 8)) - 1;
 426         if (mrp) {
 427             *data = mrp->read(mr->opaque, offset + mr->offset);
 428         } else if (width == 2) {
 429             mrp = find_portio(mr, offset, 1, false);
 430             assert(mrp);
 431             *data = mrp->read(mr->opaque, offset + mr->offset) |
 432                     (mrp->read(mr->opaque, offset + mr->offset + 1) << 8);
 433         }
 434         return;
 435     }
 436     *data = 0;
 437     access_with_adjusted_size(offset + mr->offset, data, width,
 438                               mr->ops->impl.min_access_size,
 439                               mr->ops->impl.max_access_size,
 440                               memory_region_read_accessor, mr);
 441 }
 442
 443 static void memory_region_iorange_write(IORange *iorange,
 444                                         uint64_t offset,
 445                                         unsigned width,
 446                                         uint64_t data)
 447 {
 448     MemoryRegion *mr = container_of(iorange, MemoryRegion, iorange);
 449
 450     if (mr->ops->old_portio) {
 451         const MemoryRegionPortio *mrp = find_portio(mr, offset, width, true);
 452
 453         if (mrp) {
 454             mrp->write(mr->opaque, offset + mr->offset, data);
 455         } else if (width == 2) {
 456             mrp = find_portio(mr, offset, 1, false);
 457             assert(mrp);
 458             mrp->write(mr->opaque, offset + mr->offset, data & 0xff);
 459             mrp->write(mr->opaque, offset + mr->offset + 1, data >> 8);
 460         }
 461         return;
 462     }
 463     access_with_adjusted_size(offset + mr->offset, &data, width,
 464                               mr->ops->impl.min_access_size,
 465                               mr->ops->impl.max_access_size,
 466                               memory_region_write_accessor, mr);
 467 }
 468
 469 static const IORangeOps memory_region_iorange_ops = {
 470     .read = memory_region_iorange_read,
 471     .write = memory_region_iorange_write,
 472 };
 473
 474 static void as_io_range_add(AddressSpace *as, FlatRange *fr)
 475 {
 476     iorange_init(&fr->mr->iorange, &memory_region_iorange_ops,
 477                  int128_get64(fr->addr.start), int128_get64(fr->addr.size));
 478     ioport_register(&fr->mr->iorange);
 479 }
 480
 481 static void as_io_range_del(AddressSpace *as, FlatRange *fr)
 482 {
 483     isa_unassign_ioport(int128_get64(fr->addr.start),
 484                         int128_get64(fr->addr.size));
 485 }
 486
 487 static void as_io_ioeventfd_add(AddressSpace *as, MemoryRegionIoeventfd *fd)
 488 {
 489     int r;
 490
 491     assert(fd->match_data && int128_get64(fd->addr.size) == 2);
 492
 493     r = kvm_set_ioeventfd_pio_word(fd->fd, int128_get64(fd->addr.start),
 494                                    fd->data, true);
 495     if (r < 0) {
 496         abort();
 497     }
 498 }
 499
 500 static void as_io_ioeventfd_del(AddressSpace *as, MemoryRegionIoeventfd *fd)
 501 {
 502     int r;
 503
 504     r = kvm_set_ioeventfd_pio_word(fd->fd, int128_get64(fd->addr.start),
 505                                    fd->data, false);
 506     if (r < 0) {
 507         abort();
 508     }
 509 }
 510
 511 static const AddressSpaceOps address_space_ops_io = {
 512     .range_add = as_io_range_add,
 513     .range_del = as_io_range_del,
 514     .ioeventfd_add = as_io_ioeventfd_add,
 515     .ioeventfd_del = as_io_ioeventfd_del,
 516 };
 517
 518 static AddressSpace address_space_io = {
 519     .ops = &address_space_ops_io,
 520 };
 521
 522 static AddressSpace *memory_region_to_address_space(MemoryRegion *mr)
 523 {
 524     while (mr->parent) {
 525         mr = mr->parent;
 526     }
 527     if (mr == address_space_memory.root) {
 528         return &address_space_memory;
 529     }
 530     if (mr == address_space_io.root) {
 531         return &address_space_io;
 532     }
 533     abort();
 534 }
 535
 536 /* Render a memory region into the global view.  Ranges in @view obscure
 537  * ranges in @mr.
 538  */
 539 static void render_memory_region(FlatView *view,
 540                                  MemoryRegion *mr,
 541                                  Int128 base,
 542                                  AddrRange clip,
 543                                  bool readonly)
 544 {
 545     MemoryRegion *subregion;
 546     unsigned i;
 547     target_phys_addr_t offset_in_region;
 548     Int128 remain;
 549     Int128 now;
 550     FlatRange fr;
 551     AddrRange tmp;
 552
 553     if (!mr->enabled) {
 554         return;
 555     }
 556
 557     int128_addto(&base, int128_make64(mr->addr));
 558     readonly |= mr->readonly;
 559
 560     tmp = addrrange_make(base, mr->size);
 561
 562     if (!addrrange_intersects(tmp, clip)) {
 563         return;
 564     }
 565
 566     clip = addrrange_intersection(tmp, clip);
 567
 568     if (mr->alias) {
 569         int128_subfrom(&base, int128_make64(mr->alias->addr));
 570         int128_subfrom(&base, int128_make64(mr->alias_offset));
 571         render_memory_region(view, mr->alias, base, clip, readonly);
 572         return;
 573     }
 574
 575     /* Render subregions in priority order. */
 576     QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
 577         render_memory_region(view, subregion, base, clip, readonly);
 578     }
 579
 580     if (!mr->terminates) {
 581         return;
 582     }
 583
 584     offset_in_region = int128_get64(int128_sub(clip.start, base));
 585     base = clip.start;
 586     remain = clip.size;
 587
 588     /* Render the region itself into any gaps left by the current view. */
 589     for (i = 0; i < view->nr && int128_nz(remain); ++i) {
 590         if (int128_ge(base, addrrange_end(view->ranges[i].addr))) {
 591             continue;
 592         }
 593         if (int128_lt(base, view->ranges[i].addr.start)) {
 594             now = int128_min(remain,
 595                              int128_sub(view->ranges[i].addr.start, base));
 596             fr.mr = mr;
 597             fr.offset_in_region = offset_in_region;
 598             fr.addr = addrrange_make(base, now);
 599             fr.dirty_log_mask = mr->dirty_log_mask;
 600             fr.readable = mr->readable;
 601             fr.readonly = readonly;
 602             flatview_insert(view, i, &fr);
 603             ++i;
 604             int128_addto(&base, now);
 605             offset_in_region += int128_get64(now);
 606             int128_subfrom(&remain, now);
 607         }
 608         if (int128_eq(base, view->ranges[i].addr.start)) {
 609             now = int128_min(remain, view->ranges[i].addr.size);
 610             int128_addto(&base, now);
 611             offset_in_region += int128_get64(now);
 612             int128_subfrom(&remain, now);
 613         }
 614     }
 615     if (int128_nz(remain)) {
 616         fr.mr = mr;
 617         fr.offset_in_region = offset_in_region;
 618         fr.addr = addrrange_make(base, remain);
 619         fr.dirty_log_mask = mr->dirty_log_mask;
 620         fr.readable = mr->readable;
 621         fr.readonly = readonly;
 622         flatview_insert(view, i, &fr);
 623     }
 624 }
 625
 626 /* Render a memory topology into a list of disjoint absolute ranges. */
 627 static FlatView generate_memory_topology(MemoryRegion *mr)
 628 {
 629     FlatView view;
 630
 631     flatview_init(&view);
 632
 633     render_memory_region(&view, mr, int128_zero(),
 634                          addrrange_make(int128_zero(), int128_2_64()), false);
 635     flatview_simplify(&view);
 636
 637     return view;
 638 }
 639
 640 static void address_space_add_del_ioeventfds(AddressSpace *as,
 641                                              MemoryRegionIoeventfd *fds_new,
 642                                              unsigned fds_new_nb,
 643                                              MemoryRegionIoeventfd *fds_old,
 644                                              unsigned fds_old_nb)
 645 {
 646     unsigned iold, inew;
 647
 648     /* Generate a symmetric difference of the old and new fd sets, adding
 649      * and deleting as necessary.
 650      */
 651
 652     iold = inew = 0;
 653     while (iold < fds_old_nb || inew < fds_new_nb) {
 654         if (iold < fds_old_nb
 655             && (inew == fds_new_nb
 656                 || memory_region_ioeventfd_before(fds_old[iold],
 657                                                   fds_new[inew]))) {
 658             as->ops->ioeventfd_del(as, &fds_old[iold]);
 659             ++iold;
 660         } else if (inew < fds_new_nb
 661                    && (iold == fds_old_nb
 662                        || memory_region_ioeventfd_before(fds_new[inew],
 663                                                          fds_old[iold]))) {
 664             as->ops->ioeventfd_add(as, &fds_new[inew]);
 665             ++inew;
 666         } else {
 667             ++iold;
 668             ++inew;
 669         }
 670     }
 671 }
 672
 673 static void address_space_update_ioeventfds(AddressSpace *as)
 674 {
 675     FlatRange *fr;
 676     unsigned ioeventfd_nb = 0;
 677     MemoryRegionIoeventfd *ioeventfds = NULL;
 678     AddrRange tmp;
 679     unsigned i;
 680
 681     FOR_EACH_FLAT_RANGE(fr, &as->current_map) {
 682         for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
 683             tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
 684                                   int128_sub(fr->addr.start,
 685                                              int128_make64(fr->offset_in_region)));
 686             if (addrrange_intersects(fr->addr, tmp)) {
 687                 ++ioeventfd_nb;
 688                 ioeventfds = g_realloc(ioeventfds,
 689                                           ioeventfd_nb * sizeof(*ioeventfds));
 690                 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
 691                 ioeventfds[ioeventfd_nb-1].addr = tmp;
 692             }
 693         }
 694     }
 695
 696     address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
 697                                      as->ioeventfds, as->ioeventfd_nb);
 698
 699     g_free(as->ioeventfds);
 700     as->ioeventfds = ioeventfds;
 701     as->ioeventfd_nb = ioeventfd_nb;
 702 }
 703
 704 typedef void ListenerCallback(MemoryListener *listener,
 705                               MemoryRegionSection *mrs);
 706
 707 /* Want "void (&MemoryListener::*callback)(const MemoryRegionSection& s)" */
 708 static void memory_listener_update_region(FlatRange *fr, AddressSpace *as,
 709                                           size_t callback_offset)
 710 {
 711     MemoryRegionSection section = {
 712         .mr = fr->mr,
 713         .address_space = as->root,
 714         .offset_within_region = fr->offset_in_region,
 715         .size = int128_get64(fr->addr.size),
 716         .offset_within_address_space = int128_get64(fr->addr.start),
 717     };
 718     MemoryListener *listener;
 719
 720     QLIST_FOREACH(listener, &memory_listeners, link) {
 721         ListenerCallback *callback
 722             = *(ListenerCallback **)((void *)listener + callback_offset);
 723         callback(listener, &section);
 724     }
 725 }
 726
 727 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, callback) \
 728     memory_listener_update_region(fr, as, offsetof(MemoryListener, callback))
 729
 730 static void address_space_update_topology_pass(AddressSpace *as,
 731                                                FlatView old_view,
 732                                                FlatView new_view,
 733                                                bool adding)
 734 {
 735     unsigned iold, inew;
 736     FlatRange *frold, *frnew;
 737
 738     /* Generate a symmetric difference of the old and new memory maps.
 739      * Kill ranges in the old map, and instantiate ranges in the new map.
 740      */
 741     iold = inew = 0;
 742     while (iold < old_view.nr || inew < new_view.nr) {
 743         if (iold < old_view.nr) {
 744             frold = &old_view.ranges[iold];
 745         } else {
 746             frold = NULL;
 747         }
 748         if (inew < new_view.nr) {
 749             frnew = &new_view.ranges[inew];
 750         } else {
 751             frnew = NULL;
 752         }
 753
 754         if (frold
 755             && (!frnew
 756                 || int128_lt(frold->addr.start, frnew->addr.start)
 757                 || (int128_eq(frold->addr.start, frnew->addr.start)
 758                     && !flatrange_equal(frold, frnew)))) {
 759             /* In old, but (not in new, or in new but attributes changed). */
 760
 761             if (!adding) {
 762                 MEMORY_LISTENER_UPDATE_REGION(frold, as, region_del);
 763                 as->ops->range_del(as, frold);
 764             }
 765
 766             ++iold;
 767         } else if (frold && frnew && flatrange_equal(frold, frnew)) {
 768             /* In both (logging may have changed) */
 769
 770             if (adding) {
 771                 if (frold->dirty_log_mask && !frnew->dirty_log_mask) {
 772                     MEMORY_LISTENER_UPDATE_REGION(frold, as, log_stop);
 773                     as->ops->log_stop(as, frnew);
 774                 } else if (frnew->dirty_log_mask && !frold->dirty_log_mask) {
 775                     as->ops->log_start(as, frnew);
 776                     MEMORY_LISTENER_UPDATE_REGION(frold, as, log_start);
 777                 }
 778             }
 779
 780             ++iold;
 781             ++inew;
 782         } else {
 783             /* In new */
 784
 785             if (adding) {
 786                 as->ops->range_add(as, frnew);
 787                 MEMORY_LISTENER_UPDATE_REGION(frold, as, region_add);
 788             }
 789
 790             ++inew;
 791         }
 792     }
 793 }
 794
 795
 796 static void address_space_update_topology(AddressSpace *as)
 797 {
 798     FlatView old_view = as->current_map;
 799     FlatView new_view = generate_memory_topology(as->root);
 800
 801     address_space_update_topology_pass(as, old_view, new_view, false);
 802     address_space_update_topology_pass(as, old_view, new_view, true);
 803
 804     as->current_map = new_view;
 805     flatview_destroy(&old_view);
 806     address_space_update_ioeventfds(as);
 807 }
 808
 809 static void memory_region_update_topology(MemoryRegion *mr)
 810 {
 811     if (memory_region_transaction_depth) {
 812         memory_region_update_pending |= !mr || mr->enabled;
 813         return;
 814     }
 815
 816     if (mr && !mr->enabled) {
 817         return;
 818     }
 819
 820     if (address_space_memory.root) {
 821         address_space_update_topology(&address_space_memory);
 822     }
 823     if (address_space_io.root) {
 824         address_space_update_topology(&address_space_io);
 825     }
 826
 827     memory_region_update_pending = false;
 828 }
 829
 830 void memory_region_transaction_begin(void)
 831 {
 832     ++memory_region_transaction_depth;
 833 }
 834
 835 void memory_region_transaction_commit(void)
 836 {
 837     assert(memory_region_transaction_depth);
 838     --memory_region_transaction_depth;
 839     if (!memory_region_transaction_depth && memory_region_update_pending) {
 840         memory_region_update_topology(NULL);
 841     }
 842 }
 843
 844 static void memory_region_destructor_none(MemoryRegion *mr)
 845 {
 846 }
 847
 848 static void memory_region_destructor_ram(MemoryRegion *mr)
 849 {
 850     qemu_ram_free(mr->ram_addr);
 851 }
 852
 853 static void memory_region_destructor_ram_from_ptr(MemoryRegion *mr)
 854 {
 855     qemu_ram_free_from_ptr(mr->ram_addr);
 856 }
 857
 858 static void memory_region_destructor_iomem(MemoryRegion *mr)
 859 {
 860     cpu_unregister_io_memory(mr->ram_addr);
 861 }
 862
 863 static void memory_region_destructor_rom_device(MemoryRegion *mr)
 864 {
 865     qemu_ram_free(mr->ram_addr & TARGET_PAGE_MASK);
 866     cpu_unregister_io_memory(mr->ram_addr & ~(TARGET_PAGE_MASK | IO_MEM_ROMD));
 867 }
 868
 869 void memory_region_init(MemoryRegion *mr,
 870                         const char *name,
 871                         uint64_t size)
 872 {
 873     mr->ops = NULL;
 874     mr->parent = NULL;
 875     mr->size = int128_make64(size);
 876     if (size == UINT64_MAX) {
 877         mr->size = int128_2_64();
 878     }
 879     mr->addr = 0;
 880     mr->offset = 0;
 881     mr->enabled = true;
 882     mr->terminates = false;
 883     mr->ram = false;
 884     mr->readable = true;
 885     mr->readonly = false;
 886     mr->destructor = memory_region_destructor_none;
 887     mr->priority = 0;
 888     mr->may_overlap = false;
 889     mr->alias = NULL;
 890     QTAILQ_INIT(&mr->subregions);
 891     memset(&mr->subregions_link, 0, sizeof mr->subregions_link);
 892     QTAILQ_INIT(&mr->coalesced);
 893     mr->name = g_strdup(name);
 894     mr->dirty_log_mask = 0;
 895     mr->ioeventfd_nb = 0;
 896     mr->ioeventfds = NULL;
 897 }
 898
 899 static bool memory_region_access_valid(MemoryRegion *mr,
 900                                        target_phys_addr_t addr,
 901                                        unsigned size,
 902                                        bool is_write)
 903 {
 904     if (mr->ops->valid.accepts
 905         && !mr->ops->valid.accepts(mr->opaque, addr, size, is_write)) {
 906         return false;
 907     }
 908
 909     if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
 910         return false;
 911     }
 912
 913     /* Treat zero as compatibility all valid */
 914     if (!mr->ops->valid.max_access_size) {
 915         return true;
 916     }
 917
 918     if (size > mr->ops->valid.max_access_size
 919         || size < mr->ops->valid.min_access_size) {
 920         return false;
 921     }
 922     return true;
 923 }
 924
 925 static uint32_t memory_region_read_thunk_n(void *_mr,
 926                                            target_phys_addr_t addr,
 927                                            unsigned size)
 928 {
 929     MemoryRegion *mr = _mr;
 930     uint64_t data = 0;
 931
 932     if (!memory_region_access_valid(mr, addr, size, false)) {
 933         return -1U; /* FIXME: better signalling */
 934     }
 935
 936     if (!mr->ops->read) {
 937         return mr->ops->old_mmio.read[bitops_ffsl(size)](mr->opaque, addr);
 938     }
 939
 940     /* FIXME: support unaligned access */
 941     access_with_adjusted_size(addr + mr->offset, &data, size,
 942                               mr->ops->impl.min_access_size,
 943                               mr->ops->impl.max_access_size,
 944                               memory_region_read_accessor, mr);
 945
 946     return data;
 947 }
 948
 949 static void memory_region_write_thunk_n(void *_mr,
 950                                         target_phys_addr_t addr,
 951                                         unsigned size,
 952                                         uint64_t data)
 953 {
 954     MemoryRegion *mr = _mr;
 955
 956     if (!memory_region_access_valid(mr, addr, size, true)) {
 957         return; /* FIXME: better signalling */
 958     }
 959
 960     if (!mr->ops->write) {
 961         mr->ops->old_mmio.write[bitops_ffsl(size)](mr->opaque, addr, data);
 962         return;
 963     }
 964
 965     /* FIXME: support unaligned access */
 966     access_with_adjusted_size(addr + mr->offset, &data, size,
 967                               mr->ops->impl.min_access_size,
 968                               mr->ops->impl.max_access_size,
 969                               memory_region_write_accessor, mr);
 970 }
 971
 972 static uint32_t memory_region_read_thunk_b(void *mr, target_phys_addr_t addr)
 973 {
 974     return memory_region_read_thunk_n(mr, addr, 1);
 975 }
 976
 977 static uint32_t memory_region_read_thunk_w(void *mr, target_phys_addr_t addr)
 978 {
 979     return memory_region_read_thunk_n(mr, addr, 2);
 980 }
 981
 982 static uint32_t memory_region_read_thunk_l(void *mr, target_phys_addr_t addr)
 983 {
 984     return memory_region_read_thunk_n(mr, addr, 4);
 985 }
 986
 987 static void memory_region_write_thunk_b(void *mr, target_phys_addr_t addr,
 988                                         uint32_t data)
 989 {
 990     memory_region_write_thunk_n(mr, addr, 1, data);
 991 }
 992
 993 static void memory_region_write_thunk_w(void *mr, target_phys_addr_t addr,
 994                                         uint32_t data)
 995 {
 996     memory_region_write_thunk_n(mr, addr, 2, data);
 997 }
 998
 999 static void memory_region_write_thunk_l(void *mr, target_phys_addr_t addr,
1000                                         uint32_t data)
1001 {
1002     memory_region_write_thunk_n(mr, addr, 4, data);
1003 }
1004
1005 static CPUReadMemoryFunc * const memory_region_read_thunk[] = {
1006     memory_region_read_thunk_b,
1007     memory_region_read_thunk_w,
1008     memory_region_read_thunk_l,
1009 };
1010
1011 static CPUWriteMemoryFunc * const memory_region_write_thunk[] = {
1012     memory_region_write_thunk_b,
1013     memory_region_write_thunk_w,
1014     memory_region_write_thunk_l,
1015 };
1016
1017 static void memory_region_prepare_ram_addr(MemoryRegion *mr)
1018 {
1019     if (mr->backend_registered) {
1020         return;
1021     }
1022
1023     mr->destructor = memory_region_destructor_iomem;
1024     mr->ram_addr = cpu_register_io_memory(memory_region_read_thunk,
1025                                           memory_region_write_thunk,
1026                                           mr,
1027                                           mr->ops->endianness);
1028     mr->backend_registered = true;
1029 }
1030
1031 void memory_region_init_io(MemoryRegion *mr,
1032                            const MemoryRegionOps *ops,
1033                            void *opaque,
1034                            const char *name,
1035                            uint64_t size)
1036 {
1037     memory_region_init(mr, name, size);
1038     mr->ops = ops;
1039     mr->opaque = opaque;
1040     mr->terminates = true;
1041     mr->backend_registered = false;
1042 }
1043
1044 void memory_region_init_ram(MemoryRegion *mr,
1045                             DeviceState *dev,
1046                             const char *name,
1047                             uint64_t size)
1048 {
1049     memory_region_init(mr, name, size);
1050     mr->ram = true;
1051     mr->terminates = true;
1052     mr->destructor = memory_region_destructor_ram;
1053     mr->ram_addr = qemu_ram_alloc(dev, name, size, mr);
1054     mr->backend_registered = true;
1055 }
1056
1057 void memory_region_init_ram_ptr(MemoryRegion *mr,
1058                                 DeviceState *dev,
1059                                 const char *name,
1060                                 uint64_t size,
1061                                 void *ptr)
1062 {
1063     memory_region_init(mr, name, size);
1064     mr->ram = true;
1065     mr->terminates = true;
1066     mr->destructor = memory_region_destructor_ram_from_ptr;
1067     mr->ram_addr = qemu_ram_alloc_from_ptr(dev, name, size, ptr, mr);
1068     mr->backend_registered = true;
1069 }
1070
1071 void memory_region_init_alias(MemoryRegion *mr,
1072                               const char *name,
1073                               MemoryRegion *orig,
1074                               target_phys_addr_t offset,
1075                               uint64_t size)
1076 {
1077     memory_region_init(mr, name, size);
1078     mr->alias = orig;
1079     mr->alias_offset = offset;
1080 }
1081
1082 void memory_region_init_rom_device(MemoryRegion *mr,
1083                                    const MemoryRegionOps *ops,
1084                                    void *opaque,
1085                                    DeviceState *dev,
1086                                    const char *name,
1087                                    uint64_t size)
1088 {
1089     memory_region_init(mr, name, size);
1090     mr->ops = ops;
1091     mr->opaque = opaque;
1092     mr->terminates = true;
1093     mr->destructor = memory_region_destructor_rom_device;
1094     mr->ram_addr = qemu_ram_alloc(dev, name, size, mr);
1095     mr->ram_addr |= cpu_register_io_memory(memory_region_read_thunk,
1096                                            memory_region_write_thunk,
1097                                            mr,
1098                                            mr->ops->endianness);
1099     mr->ram_addr |= IO_MEM_ROMD;
1100     mr->backend_registered = true;
1101 }
1102
1103 void memory_region_destroy(MemoryRegion *mr)
1104 {
1105     assert(QTAILQ_EMPTY(&mr->subregions));
1106     mr->destructor(mr);
1107     memory_region_clear_coalescing(mr);
1108     g_free((char *)mr->name);
1109     g_free(mr->ioeventfds);
1110 }
1111
1112 uint64_t memory_region_size(MemoryRegion *mr)
1113 {
1114     if (int128_eq(mr->size, int128_2_64())) {
1115         return UINT64_MAX;
1116     }
1117     return int128_get64(mr->size);
1118 }
1119
1120 bool memory_region_is_ram(MemoryRegion *mr)
1121 {
1122     return mr->ram;
1123 }
1124
1125 bool memory_region_is_logging(MemoryRegion *mr)
1126 {
1127     return mr->dirty_log_mask;
1128 }
1129
1130 bool memory_region_is_rom(MemoryRegion *mr)
1131 {
1132     return mr->ram && mr->readonly;
1133 }
1134
1135 void memory_region_set_offset(MemoryRegion *mr, target_phys_addr_t offset)
1136 {
1137     mr->offset = offset;
1138 }
1139
1140 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1141 {
1142     uint8_t mask = 1 << client;
1143
1144     mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1145     memory_region_update_topology(mr);
1146 }
1147
1148 bool memory_region_get_dirty(MemoryRegion *mr, target_phys_addr_t addr,
1149                              unsigned client)
1150 {
1151     assert(mr->terminates);
1152     return cpu_physical_memory_get_dirty(mr->ram_addr + addr, 1 << client);
1153 }
1154
1155 void memory_region_set_dirty(MemoryRegion *mr, target_phys_addr_t addr)
1156 {
1157     assert(mr->terminates);
1158     return cpu_physical_memory_set_dirty(mr->ram_addr + addr);
1159 }
1160
1161 void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
1162 {
1163     FlatRange *fr;
1164
1165     FOR_EACH_FLAT_RANGE(fr, &address_space_memory.current_map) {
1166         if (fr->mr == mr) {
1167             MEMORY_LISTENER_UPDATE_REGION(fr, &address_space_memory, log_sync);
1168             cpu_physical_sync_dirty_bitmap(int128_get64(fr->addr.start),
1169                                            int128_get64(addrrange_end(fr->addr)));
1170         }
1171     }
1172 }
1173
1174 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
1175 {
1176     if (mr->readonly != readonly) {
1177         mr->readonly = readonly;
1178         memory_region_update_topology(mr);
1179     }
1180 }
1181
1182 void memory_region_rom_device_set_readable(MemoryRegion *mr, bool readable)
1183 {
1184     if (mr->readable != readable) {
1185         mr->readable = readable;
1186         memory_region_update_topology(mr);
1187     }
1188 }
1189
1190 void memory_region_reset_dirty(MemoryRegion *mr, target_phys_addr_t addr,
1191                                target_phys_addr_t size, unsigned client)
1192 {
1193     assert(mr->terminates);
1194     cpu_physical_memory_reset_dirty(mr->ram_addr + addr,
1195                                     mr->ram_addr + addr + size,
1196                                     1 << client);
1197 }
1198
1199 void *memory_region_get_ram_ptr(MemoryRegion *mr)
1200 {
1201     if (mr->alias) {
1202         return memory_region_get_ram_ptr(mr->alias) + mr->alias_offset;
1203     }
1204
1205     assert(mr->terminates);
1206
1207     return qemu_get_ram_ptr(mr->ram_addr & TARGET_PAGE_MASK);
1208 }
1209
1210 static void memory_region_update_coalesced_range(MemoryRegion *mr)
1211 {
1212     FlatRange *fr;
1213     CoalescedMemoryRange *cmr;
1214     AddrRange tmp;
1215
1216     FOR_EACH_FLAT_RANGE(fr, &address_space_memory.current_map) {
1217         if (fr->mr == mr) {
1218             qemu_unregister_coalesced_mmio(int128_get64(fr->addr.start),
1219                                            int128_get64(fr->addr.size));
1220             QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
1221                 tmp = addrrange_shift(cmr->addr,
1222                                       int128_sub(fr->addr.start,
1223                                                  int128_make64(fr->offset_in_region)));
1224                 if (!addrrange_intersects(tmp, fr->addr)) {
1225                     continue;
1226                 }
1227                 tmp = addrrange_intersection(tmp, fr->addr);
1228                 qemu_register_coalesced_mmio(int128_get64(tmp.start),
1229                                              int128_get64(tmp.size));
1230             }
1231         }
1232     }
1233 }
1234
1235 void memory_region_set_coalescing(MemoryRegion *mr)
1236 {
1237     memory_region_clear_coalescing(mr);
1238     memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
1239 }
1240
1241 void memory_region_add_coalescing(MemoryRegion *mr,
1242                                   target_phys_addr_t offset,
1243                                   uint64_t size)
1244 {
1245     CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
1246
1247     cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
1248     QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
1249     memory_region_update_coalesced_range(mr);
1250 }
1251
1252 void memory_region_clear_coalescing(MemoryRegion *mr)
1253 {
1254     CoalescedMemoryRange *cmr;
1255
1256     while (!QTAILQ_EMPTY(&mr->coalesced)) {
1257         cmr = QTAILQ_FIRST(&mr->coalesced);
1258         QTAILQ_REMOVE(&mr->coalesced, cmr, link);
1259         g_free(cmr);
1260     }
1261     memory_region_update_coalesced_range(mr);
1262 }
1263
1264 void memory_region_add_eventfd(MemoryRegion *mr,
1265                                target_phys_addr_t addr,
1266                                unsigned size,
1267                                bool match_data,
1268                                uint64_t data,
1269                                int fd)
1270 {
1271     MemoryRegionIoeventfd mrfd = {
1272         .addr.start = int128_make64(addr),
1273         .addr.size = int128_make64(size),
1274         .match_data = match_data,
1275         .data = data,
1276         .fd = fd,
1277     };
1278     unsigned i;
1279
1280     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1281         if (memory_region_ioeventfd_before(mrfd, mr->ioeventfds[i])) {
1282             break;
1283         }
1284     }
1285     ++mr->ioeventfd_nb;
1286     mr->ioeventfds = g_realloc(mr->ioeventfds,
1287                                   sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
1288     memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
1289             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
1290     mr->ioeventfds[i] = mrfd;
1291     memory_region_update_topology(mr);
1292 }
1293
1294 void memory_region_del_eventfd(MemoryRegion *mr,
1295                                target_phys_addr_t addr,
1296                                unsigned size,
1297                                bool match_data,
1298                                uint64_t data,
1299                                int fd)
1300 {
1301     MemoryRegionIoeventfd mrfd = {
1302         .addr.start = int128_make64(addr),
1303         .addr.size = int128_make64(size),
1304         .match_data = match_data,
1305         .data = data,
1306         .fd = fd,
1307     };
1308     unsigned i;
1309
1310     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1311         if (memory_region_ioeventfd_equal(mrfd, mr->ioeventfds[i])) {
1312             break;
1313         }
1314     }
1315     assert(i != mr->ioeventfd_nb);
1316     memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
1317             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
1318     --mr->ioeventfd_nb;
1319     mr->ioeventfds = g_realloc(mr->ioeventfds,
1320                                   sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
1321     memory_region_update_topology(mr);
1322 }
1323
1324 static void memory_region_add_subregion_common(MemoryRegion *mr,
1325                                                target_phys_addr_t offset,
1326                                                MemoryRegion *subregion)
1327 {
1328     MemoryRegion *other;
1329
1330     assert(!subregion->parent);
1331     subregion->parent = mr;
1332     subregion->addr = offset;
1333     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1334         if (subregion->may_overlap || other->may_overlap) {
1335             continue;
1336         }
1337         if (int128_gt(int128_make64(offset),
1338                       int128_add(int128_make64(other->addr), other->size))
1339             || int128_le(int128_add(int128_make64(offset), subregion->size),
1340                          int128_make64(other->addr))) {
1341             continue;
1342         }
1343 #if 0
1344         printf("warning: subregion collision %llx/%llx (%s) "
1345                "vs %llx/%llx (%s)\n",
1346                (unsigned long long)offset,
1347                (unsigned long long)int128_get64(subregion->size),
1348                subregion->name,
1349                (unsigned long long)other->addr,
1350                (unsigned long long)int128_get64(other->size),
1351                other->name);
1352 #endif
1353     }
1354     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1355         if (subregion->priority >= other->priority) {
1356             QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
1357             goto done;
1358         }
1359     }
1360     QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
1361 done:
1362     memory_region_update_topology(mr);
1363 }
1364
1365
1366 void memory_region_add_subregion(MemoryRegion *mr,
1367                                  target_phys_addr_t offset,
1368                                  MemoryRegion *subregion)
1369 {
1370     subregion->may_overlap = false;
1371     subregion->priority = 0;
1372     memory_region_add_subregion_common(mr, offset, subregion);
1373 }
1374
1375 void memory_region_add_subregion_overlap(MemoryRegion *mr,
1376                                          target_phys_addr_t offset,
1377                                          MemoryRegion *subregion,
1378                                          unsigned priority)
1379 {
1380     subregion->may_overlap = true;
1381     subregion->priority = priority;
1382     memory_region_add_subregion_common(mr, offset, subregion);
1383 }
1384
1385 void memory_region_del_subregion(MemoryRegion *mr,
1386                                  MemoryRegion *subregion)
1387 {
1388     assert(subregion->parent == mr);
1389     subregion->parent = NULL;
1390     QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
1391     memory_region_update_topology(mr);
1392 }
1393
1394 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
1395 {
1396     if (enabled == mr->enabled) {
1397         return;
1398     }
1399     mr->enabled = enabled;
1400     memory_region_update_topology(NULL);
1401 }
1402
1403 void memory_region_set_address(MemoryRegion *mr, target_phys_addr_t addr)
1404 {
1405     MemoryRegion *parent = mr->parent;
1406     unsigned priority = mr->priority;
1407     bool may_overlap = mr->may_overlap;
1408
1409     if (addr == mr->addr || !parent) {
1410         mr->addr = addr;
1411         return;
1412     }
1413
1414     memory_region_transaction_begin();
1415     memory_region_del_subregion(parent, mr);
1416     if (may_overlap) {
1417         memory_region_add_subregion_overlap(parent, addr, mr, priority);
1418     } else {
1419         memory_region_add_subregion(parent, addr, mr);
1420     }
1421     memory_region_transaction_commit();
1422 }
1423
1424 void memory_region_set_alias_offset(MemoryRegion *mr, target_phys_addr_t offset)
1425 {
1426     target_phys_addr_t old_offset = mr->alias_offset;
1427
1428     assert(mr->alias);
1429     mr->alias_offset = offset;
1430
1431     if (offset == old_offset || !mr->parent) {
1432         return;
1433     }
1434
1435     memory_region_update_topology(mr);
1436 }
1437
1438 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
1439 {
1440     const AddrRange *addr = addr_;
1441     const FlatRange *fr = fr_;
1442
1443     if (int128_le(addrrange_end(*addr), fr->addr.start)) {
1444         return -1;
1445     } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
1446         return 1;
1447     }
1448     return 0;
1449 }
1450
1451 static FlatRange *address_space_lookup(AddressSpace *as, AddrRange addr)
1452 {
1453     return bsearch(&addr, as->current_map.ranges, as->current_map.nr,
1454                    sizeof(FlatRange), cmp_flatrange_addr);
1455 }
1456
1457 MemoryRegionSection memory_region_find(MemoryRegion *address_space,
1458                                        target_phys_addr_t addr, uint64_t size)
1459 {
1460     AddressSpace *as = memory_region_to_address_space(address_space);
1461     AddrRange range = addrrange_make(int128_make64(addr),
1462                                      int128_make64(size));
1463     FlatRange *fr = address_space_lookup(as, range);
1464     MemoryRegionSection ret = { .mr = NULL, .size = 0 };
1465
1466     if (!fr) {
1467         return ret;
1468     }
1469
1470     while (fr > as->current_map.ranges
1471            && addrrange_intersects(fr[-1].addr, range)) {
1472         --fr;
1473     }
1474
1475     ret.mr = fr->mr;
1476     range = addrrange_intersection(range, fr->addr);
1477     ret.offset_within_region = fr->offset_in_region;
1478     ret.offset_within_region += int128_get64(int128_sub(range.start,
1479                                                         fr->addr.start));
1480     ret.size = int128_get64(range.size);
1481     ret.offset_within_address_space = int128_get64(range.start);
1482     return ret;
1483 }
1484
1485 void memory_global_sync_dirty_bitmap(MemoryRegion *address_space)
1486 {
1487     AddressSpace *as = memory_region_to_address_space(address_space);
1488     FlatRange *fr;
1489
1490     cpu_physical_sync_dirty_bitmap(0, TARGET_PHYS_ADDR_MAX);
1491     FOR_EACH_FLAT_RANGE(fr, &as->current_map) {
1492         MEMORY_LISTENER_UPDATE_REGION(fr, as, log_sync);
1493     }
1494 }
1495
1496 void memory_global_dirty_log_start(void)
1497 {
1498     MemoryListener *listener;
1499
1500     global_dirty_log = true;
1501     QLIST_FOREACH(listener, &memory_listeners, link) {
1502         listener->log_global_start(listener);
1503     }
1504 }
1505
1506 void memory_global_dirty_log_stop(void)
1507 {
1508     MemoryListener *listener;
1509
1510     global_dirty_log = false;
1511     QLIST_FOREACH(listener, &memory_listeners, link) {
1512         listener->log_global_stop(listener);
1513     }
1514 }
1515
1516 static void listener_add_address_space(MemoryListener *listener,
1517                                        AddressSpace *as)
1518 {
1519     FlatRange *fr;
1520
1521     if (global_dirty_log) {
1522         listener->log_global_start(listener);
1523     }
1524     FOR_EACH_FLAT_RANGE(fr, &as->current_map) {
1525         MemoryRegionSection section = {
1526             .mr = fr->mr,
1527             .address_space = as->root,
1528             .offset_within_region = fr->offset_in_region,
1529             .size = int128_get64(fr->addr.size),
1530             .offset_within_address_space = int128_get64(fr->addr.start),
1531         };
1532         listener->region_add(listener, &section);
1533     }
1534 }
1535
1536 void memory_listener_register(MemoryListener *listener)
1537 {
1538     QLIST_INSERT_HEAD(&memory_listeners, listener, link);
1539     listener_add_address_space(listener, &address_space_memory);
1540     listener_add_address_space(listener, &address_space_io);
1541 }
1542
1543 void memory_listener_unregister(MemoryListener *listener)
1544 {
1545     QLIST_REMOVE(listener, link);
1546 }
1547
1548 void set_system_memory_map(MemoryRegion *mr)
1549 {
1550     address_space_memory.root = mr;
1551     memory_region_update_topology(NULL);
1552 }
1553
1554 void set_system_io_map(MemoryRegion *mr)
1555 {
1556     address_space_io.root = mr;
1557     memory_region_update_topology(NULL);
1558 }
1559
1560 typedef struct MemoryRegionList MemoryRegionList;
1561
1562 struct MemoryRegionList {
1563     const MemoryRegion *mr;
1564     bool printed;
1565     QTAILQ_ENTRY(MemoryRegionList) queue;
1566 };
1567
1568 typedef QTAILQ_HEAD(queue, MemoryRegionList) MemoryRegionListHead;
1569
1570 static void mtree_print_mr(fprintf_function mon_printf, void *f,
1571                            const MemoryRegion *mr, unsigned int level,
1572                            target_phys_addr_t base,
1573                            MemoryRegionListHead *alias_print_queue)
1574 {
1575     MemoryRegionList *new_ml, *ml, *next_ml;
1576     MemoryRegionListHead submr_print_queue;
1577     const MemoryRegion *submr;
1578     unsigned int i;
1579
1580     if (!mr) {
1581         return;
1582     }
1583
1584     for (i = 0; i < level; i++) {
1585         mon_printf(f, "  ");
1586     }
1587
1588     if (mr->alias) {
1589         MemoryRegionList *ml;
1590         bool found = false;
1591
1592         /* check if the alias is already in the queue */
1593         QTAILQ_FOREACH(ml, alias_print_queue, queue) {
1594             if (ml->mr == mr->alias && !ml->printed) {
1595                 found = true;
1596             }
1597         }
1598
1599         if (!found) {
1600             ml = g_new(MemoryRegionList, 1);
1601             ml->mr = mr->alias;
1602             ml->printed = false;
1603             QTAILQ_INSERT_TAIL(alias_print_queue, ml, queue);
1604         }
1605         mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d): alias %s @%s "
1606                    TARGET_FMT_plx "-" TARGET_FMT_plx "\n",
1607                    base + mr->addr,
1608                    base + mr->addr
1609                    + (target_phys_addr_t)int128_get64(mr->size) - 1,
1610                    mr->priority,
1611                    mr->name,
1612                    mr->alias->name,
1613                    mr->alias_offset,
1614                    mr->alias_offset
1615                    + (target_phys_addr_t)int128_get64(mr->size) - 1);
1616     } else {
1617         mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d): %s\n",
1618                    base + mr->addr,
1619                    base + mr->addr
1620                    + (target_phys_addr_t)int128_get64(mr->size) - 1,
1621                    mr->priority,
1622                    mr->name);
1623     }
1624
1625     QTAILQ_INIT(&submr_print_queue);
1626
1627     QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
1628         new_ml = g_new(MemoryRegionList, 1);
1629         new_ml->mr = submr;
1630         QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
1631             if (new_ml->mr->addr < ml->mr->addr ||
1632                 (new_ml->mr->addr == ml->mr->addr &&
1633                  new_ml->mr->priority > ml->mr->priority)) {
1634                 QTAILQ_INSERT_BEFORE(ml, new_ml, queue);
1635                 new_ml = NULL;
1636                 break;
1637             }
1638         }
1639         if (new_ml) {
1640             QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, queue);
1641         }
1642     }
1643
1644     QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
1645         mtree_print_mr(mon_printf, f, ml->mr, level + 1, base + mr->addr,
1646                        alias_print_queue);
1647     }
1648
1649     QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, queue, next_ml) {
1650         g_free(ml);
1651     }
1652 }
1653
1654 void mtree_info(fprintf_function mon_printf, void *f)
1655 {
1656     MemoryRegionListHead ml_head;
1657     MemoryRegionList *ml, *ml2;
1658
1659     QTAILQ_INIT(&ml_head);
1660
1661     mon_printf(f, "memory\n");
1662     mtree_print_mr(mon_printf, f, address_space_memory.root, 0, 0, &ml_head);
1663
1664     /* print aliased regions */
1665     QTAILQ_FOREACH(ml, &ml_head, queue) {
1666         if (!ml->printed) {
1667             mon_printf(f, "%s\n", ml->mr->name);
1668             mtree_print_mr(mon_printf, f, ml->mr, 0, 0, &ml_head);
1669         }
1670     }
1671
1672     QTAILQ_FOREACH_SAFE(ml, &ml_head, queue, ml2) {
1673         g_free(ml);
1674     }
1675
1676     if (address_space_io.root &&
1677         !QTAILQ_EMPTY(&address_space_io.root->subregions)) {
1678         QTAILQ_INIT(&ml_head);
1679         mon_printf(f, "I/O\n");
1680         mtree_print_mr(mon_printf, f, address_space_io.root, 0, 0, &ml_head);
1681     }
1682 }