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