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 unsigned memory_region_transaction_depth = 0;
  22
  23 typedef struct AddrRange AddrRange;
  24
  25 /*
  26  * Note using signed integers limits us to physical addresses at most
  27  * 63 bits wide.  They are needed for negative offsetting in aliases
  28  * (large MemoryRegion::alias_offset).
  29  */
  30 struct AddrRange {
  31     int64_t start;
  32     int64_t size;
  33 };
  34
  35 static AddrRange addrrange_make(int64_t start, int64_t size)
  36 {
  37     return (AddrRange) { start, size };
  38 }
  39
  40 static bool addrrange_equal(AddrRange r1, AddrRange r2)
  41 {
  42     return r1.start == r2.start && r1.size == r2.size;
  43 }
  44
  45 static int64_t addrrange_end(AddrRange r)
  46 {
  47     return r.start + r.size;
  48 }
  49
  50 static AddrRange addrrange_shift(AddrRange range, int64_t delta)
  51 {
  52     range.start += delta;
  53     return range;
  54 }
  55
  56 static bool addrrange_intersects(AddrRange r1, AddrRange r2)
  57 {
  58     return (r1.start >= r2.start && r1.start < r2.start + r2.size)
  59         || (r2.start >= r1.start && r2.start < r1.start + r1.size);
  60 }
  61
  62 static AddrRange addrrange_intersection(AddrRange r1, AddrRange r2)
  63 {
  64     int64_t start = MAX(r1.start, r2.start);
  65     /* off-by-one arithmetic to prevent overflow */
  66     int64_t end = MIN(addrrange_end(r1) - 1, addrrange_end(r2) - 1);
  67     return addrrange_make(start, end - start + 1);
  68 }
  69
  70 struct CoalescedMemoryRange {
  71     AddrRange addr;
  72     QTAILQ_ENTRY(CoalescedMemoryRange) link;
  73 };
  74
  75 struct MemoryRegionIoeventfd {
  76     AddrRange addr;
  77     bool match_data;
  78     uint64_t data;
  79     int fd;
  80 };
  81
  82 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd a,
  83                                            MemoryRegionIoeventfd b)
  84 {
  85     if (a.addr.start < b.addr.start) {
  86         return true;
  87     } else if (a.addr.start > b.addr.start) {
  88         return false;
  89     } else if (a.addr.size < b.addr.size) {
  90         return true;
  91     } else if (a.addr.size > b.addr.size) {
  92         return false;
  93     } else if (a.match_data < b.match_data) {
  94         return true;
  95     } else  if (a.match_data > b.match_data) {
  96         return false;
  97     } else if (a.match_data) {
  98         if (a.data < b.data) {
  99             return true;
 100         } else if (a.data > b.data) {
 101             return false;
 102         }
 103     }
 104     if (a.fd < b.fd) {
 105         return true;
 106     } else if (a.fd > b.fd) {
 107         return false;
 108     }
 109     return false;
 110 }
 111
 112 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd a,
 113                                           MemoryRegionIoeventfd b)
 114 {
 115     return !memory_region_ioeventfd_before(a, b)
 116         && !memory_region_ioeventfd_before(b, a);
 117 }
 118
 119 typedef struct FlatRange FlatRange;
 120 typedef struct FlatView FlatView;
 121
 122 /* Range of memory in the global map.  Addresses are absolute. */
 123 struct FlatRange {
 124     MemoryRegion *mr;
 125     target_phys_addr_t offset_in_region;
 126     AddrRange addr;
 127     uint8_t dirty_log_mask;
 128     bool readable;
 129 };
 130
 131 /* Flattened global view of current active memory hierarchy.  Kept in sorted
 132  * order.
 133  */
 134 struct FlatView {
 135     FlatRange *ranges;
 136     unsigned nr;
 137     unsigned nr_allocated;
 138 };
 139
 140 typedef struct AddressSpace AddressSpace;
 141 typedef struct AddressSpaceOps AddressSpaceOps;
 142
 143 /* A system address space - I/O, memory, etc. */
 144 struct AddressSpace {
 145     const AddressSpaceOps *ops;
 146     MemoryRegion *root;
 147     FlatView current_map;
 148     int ioeventfd_nb;
 149     MemoryRegionIoeventfd *ioeventfds;
 150 };
 151
 152 struct AddressSpaceOps {
 153     void (*range_add)(AddressSpace *as, FlatRange *fr);
 154     void (*range_del)(AddressSpace *as, FlatRange *fr);
 155     void (*log_start)(AddressSpace *as, FlatRange *fr);
 156     void (*log_stop)(AddressSpace *as, FlatRange *fr);
 157     void (*ioeventfd_add)(AddressSpace *as, MemoryRegionIoeventfd *fd);
 158     void (*ioeventfd_del)(AddressSpace *as, MemoryRegionIoeventfd *fd);
 159 };
 160
 161 #define FOR_EACH_FLAT_RANGE(var, view)          \
 162     for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
 163
 164 static bool flatrange_equal(FlatRange *a, FlatRange *b)
 165 {
 166     return a->mr == b->mr
 167         && addrrange_equal(a->addr, b->addr)
 168         && a->offset_in_region == b->offset_in_region
 169         && a->readable == b->readable;
 170 }
 171
 172 static void flatview_init(FlatView *view)
 173 {
 174     view->ranges = NULL;
 175     view->nr = 0;
 176     view->nr_allocated = 0;
 177 }
 178
 179 /* Insert a range into a given position.  Caller is responsible for maintaining
 180  * sorting order.
 181  */
 182 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
 183 {
 184     if (view->nr == view->nr_allocated) {
 185         view->nr_allocated = MAX(2 * view->nr, 10);
 186         view->ranges = g_realloc(view->ranges,
 187                                     view->nr_allocated * sizeof(*view->ranges));
 188     }
 189     memmove(view->ranges + pos + 1, view->ranges + pos,
 190             (view->nr - pos) * sizeof(FlatRange));
 191     view->ranges[pos] = *range;
 192     ++view->nr;
 193 }
 194
 195 static void flatview_destroy(FlatView *view)
 196 {
 197     g_free(view->ranges);
 198 }
 199
 200 static bool can_merge(FlatRange *r1, FlatRange *r2)
 201 {
 202     return addrrange_end(r1->addr) == r2->addr.start
 203         && r1->mr == r2->mr
 204         && r1->offset_in_region + r1->addr.size == r2->offset_in_region
 205         && r1->dirty_log_mask == r2->dirty_log_mask
 206         && r1->readable == r2->readable;
 207 }
 208
 209 /* Attempt to simplify a view by merging ajacent ranges */
 210 static void flatview_simplify(FlatView *view)
 211 {
 212     unsigned i, j;
 213
 214     i = 0;
 215     while (i < view->nr) {
 216         j = i + 1;
 217         while (j < view->nr
 218                && can_merge(&view->ranges[j-1], &view->ranges[j])) {
 219             view->ranges[i].addr.size += view->ranges[j].addr.size;
 220             ++j;
 221         }
 222         ++i;
 223         memmove(&view->ranges[i], &view->ranges[j],
 224                 (view->nr - j) * sizeof(view->ranges[j]));
 225         view->nr -= j - i;
 226     }
 227 }
 228
 229 static void memory_region_read_accessor(void *opaque,
 230                                         target_phys_addr_t addr,
 231                                         uint64_t *value,
 232                                         unsigned size,
 233                                         unsigned shift,
 234                                         uint64_t mask)
 235 {
 236     MemoryRegion *mr = opaque;
 237     uint64_t tmp;
 238
 239     tmp = mr->ops->read(mr->opaque, addr, size);
 240     *value |= (tmp & mask) << shift;
 241 }
 242
 243 static void memory_region_write_accessor(void *opaque,
 244                                          target_phys_addr_t addr,
 245                                          uint64_t *value,
 246                                          unsigned size,
 247                                          unsigned shift,
 248                                          uint64_t mask)
 249 {
 250     MemoryRegion *mr = opaque;
 251     uint64_t tmp;
 252
 253     tmp = (*value >> shift) & mask;
 254     mr->ops->write(mr->opaque, addr, tmp, size);
 255 }
 256
 257 static void access_with_adjusted_size(target_phys_addr_t addr,
 258                                       uint64_t *value,
 259                                       unsigned size,
 260                                       unsigned access_size_min,
 261                                       unsigned access_size_max,
 262                                       void (*access)(void *opaque,
 263                                                      target_phys_addr_t addr,
 264                                                      uint64_t *value,
 265                                                      unsigned size,
 266                                                      unsigned shift,
 267                                                      uint64_t mask),
 268                                       void *opaque)
 269 {
 270     uint64_t access_mask;
 271     unsigned access_size;
 272     unsigned i;
 273
 274     if (!access_size_min) {
 275         access_size_min = 1;
 276     }
 277     if (!access_size_max) {
 278         access_size_max = 4;
 279     }
 280     access_size = MAX(MIN(size, access_size_max), access_size_min);
 281     access_mask = -1ULL >> (64 - access_size * 8);
 282     for (i = 0; i < size; i += access_size) {
 283         /* FIXME: big-endian support */
 284         access(opaque, addr + i, value, access_size, i * 8, access_mask);
 285     }
 286 }
 287
 288 static void memory_region_prepare_ram_addr(MemoryRegion *mr);
 289
 290 static void as_memory_range_add(AddressSpace *as, FlatRange *fr)
 291 {
 292     ram_addr_t phys_offset, region_offset;
 293
 294     memory_region_prepare_ram_addr(fr->mr);
 295
 296     phys_offset = fr->mr->ram_addr;
 297     region_offset = fr->offset_in_region;
 298     /* cpu_register_physical_memory_log() wants region_offset for
 299      * mmio, but prefers offseting phys_offset for RAM.  Humour it.
 300      */
 301     if ((phys_offset & ~TARGET_PAGE_MASK) <= IO_MEM_ROM) {
 302         phys_offset += region_offset;
 303         region_offset = 0;
 304     }
 305
 306     if (!fr->readable) {
 307         phys_offset &= ~TARGET_PAGE_MASK & ~IO_MEM_ROMD;
 308     }
 309
 310     cpu_register_physical_memory_log(fr->addr.start,
 311                                      fr->addr.size,
 312                                      phys_offset,
 313                                      region_offset,
 314                                      fr->dirty_log_mask);
 315 }
 316
 317 static void as_memory_range_del(AddressSpace *as, FlatRange *fr)
 318 {
 319     if (fr->dirty_log_mask) {
 320         cpu_physical_sync_dirty_bitmap(fr->addr.start,
 321                                        fr->addr.start + fr->addr.size);
 322     }
 323     cpu_register_physical_memory(fr->addr.start, fr->addr.size,
 324                                  IO_MEM_UNASSIGNED);
 325 }
 326
 327 static void as_memory_log_start(AddressSpace *as, FlatRange *fr)
 328 {
 329     cpu_physical_log_start(fr->addr.start, fr->addr.size);
 330 }
 331
 332 static void as_memory_log_stop(AddressSpace *as, FlatRange *fr)
 333 {
 334     cpu_physical_log_stop(fr->addr.start, fr->addr.size);
 335 }
 336
 337 static void as_memory_ioeventfd_add(AddressSpace *as, MemoryRegionIoeventfd *fd)
 338 {
 339     int r;
 340
 341     assert(fd->match_data && fd->addr.size == 4);
 342
 343     r = kvm_set_ioeventfd_mmio_long(fd->fd, fd->addr.start, fd->data, true);
 344     if (r < 0) {
 345         abort();
 346     }
 347 }
 348
 349 static void as_memory_ioeventfd_del(AddressSpace *as, MemoryRegionIoeventfd *fd)
 350 {
 351     int r;
 352
 353     r = kvm_set_ioeventfd_mmio_long(fd->fd, fd->addr.start, fd->data, false);
 354     if (r < 0) {
 355         abort();
 356     }
 357 }
 358
 359 static const AddressSpaceOps address_space_ops_memory = {
 360     .range_add = as_memory_range_add,
 361     .range_del = as_memory_range_del,
 362     .log_start = as_memory_log_start,
 363     .log_stop = as_memory_log_stop,
 364     .ioeventfd_add = as_memory_ioeventfd_add,
 365     .ioeventfd_del = as_memory_ioeventfd_del,
 366 };
 367
 368 static AddressSpace address_space_memory = {
 369     .ops = &address_space_ops_memory,
 370 };
 371
 372 static const MemoryRegionPortio *find_portio(MemoryRegion *mr, uint64_t offset,
 373                                              unsigned width, bool write)
 374 {
 375     const MemoryRegionPortio *mrp;
 376
 377     for (mrp = mr->ops->old_portio; mrp->size; ++mrp) {
 378         if (offset >= mrp->offset && offset < mrp->offset + mrp->len
 379             && width == mrp->size
 380             && (write ? (bool)mrp->write : (bool)mrp->read)) {
 381             return mrp;
 382         }
 383     }
 384     return NULL;
 385 }
 386
 387 static void memory_region_iorange_read(IORange *iorange,
 388                                        uint64_t offset,
 389                                        unsigned width,
 390                                        uint64_t *data)
 391 {
 392     MemoryRegion *mr = container_of(iorange, MemoryRegion, iorange);
 393
 394     if (mr->ops->old_portio) {
 395         const MemoryRegionPortio *mrp = find_portio(mr, offset, width, false);
 396
 397         *data = ((uint64_t)1 << (width * 8)) - 1;
 398         if (mrp) {
 399             *data = mrp->read(mr->opaque, offset);
 400         }
 401         return;
 402     }
 403     *data = 0;
 404     access_with_adjusted_size(offset, data, width,
 405                               mr->ops->impl.min_access_size,
 406                               mr->ops->impl.max_access_size,
 407                               memory_region_read_accessor, mr);
 408 }
 409
 410 static void memory_region_iorange_write(IORange *iorange,
 411                                         uint64_t offset,
 412                                         unsigned width,
 413                                         uint64_t data)
 414 {
 415     MemoryRegion *mr = container_of(iorange, MemoryRegion, iorange);
 416
 417     if (mr->ops->old_portio) {
 418         const MemoryRegionPortio *mrp = find_portio(mr, offset, width, true);
 419
 420         if (mrp) {
 421             mrp->write(mr->opaque, offset, data);
 422         }
 423         return;
 424     }
 425     access_with_adjusted_size(offset, &data, width,
 426                               mr->ops->impl.min_access_size,
 427                               mr->ops->impl.max_access_size,
 428                               memory_region_write_accessor, mr);
 429 }
 430
 431 static const IORangeOps memory_region_iorange_ops = {
 432     .read = memory_region_iorange_read,
 433     .write = memory_region_iorange_write,
 434 };
 435
 436 static void as_io_range_add(AddressSpace *as, FlatRange *fr)
 437 {
 438     iorange_init(&fr->mr->iorange, &memory_region_iorange_ops,
 439                  fr->addr.start,fr->addr.size);
 440     ioport_register(&fr->mr->iorange);
 441 }
 442
 443 static void as_io_range_del(AddressSpace *as, FlatRange *fr)
 444 {
 445     isa_unassign_ioport(fr->addr.start, fr->addr.size);
 446 }
 447
 448 static void as_io_ioeventfd_add(AddressSpace *as, MemoryRegionIoeventfd *fd)
 449 {
 450     int r;
 451
 452     assert(fd->match_data && fd->addr.size == 2);
 453
 454     r = kvm_set_ioeventfd_pio_word(fd->fd, fd->addr.start, fd->data, true);
 455     if (r < 0) {
 456         abort();
 457     }
 458 }
 459
 460 static void as_io_ioeventfd_del(AddressSpace *as, MemoryRegionIoeventfd *fd)
 461 {
 462     int r;
 463
 464     r = kvm_set_ioeventfd_pio_word(fd->fd, fd->addr.start, fd->data, false);
 465     if (r < 0) {
 466         abort();
 467     }
 468 }
 469
 470 static const AddressSpaceOps address_space_ops_io = {
 471     .range_add = as_io_range_add,
 472     .range_del = as_io_range_del,
 473     .ioeventfd_add = as_io_ioeventfd_add,
 474     .ioeventfd_del = as_io_ioeventfd_del,
 475 };
 476
 477 static AddressSpace address_space_io = {
 478     .ops = &address_space_ops_io,
 479 };
 480
 481 /* Render a memory region into the global view.  Ranges in @view obscure
 482  * ranges in @mr.
 483  */
 484 static void render_memory_region(FlatView *view,
 485                                  MemoryRegion *mr,
 486                                  target_phys_addr_t base,
 487                                  AddrRange clip)
 488 {
 489     MemoryRegion *subregion;
 490     unsigned i;
 491     target_phys_addr_t offset_in_region;
 492     int64_t remain;
 493     int64_t now;
 494     FlatRange fr;
 495     AddrRange tmp;
 496
 497     base += mr->addr;
 498
 499     tmp = addrrange_make(base, mr->size);
 500
 501     if (!addrrange_intersects(tmp, clip)) {
 502         return;
 503     }
 504
 505     clip = addrrange_intersection(tmp, clip);
 506
 507     if (mr->alias) {
 508         base -= mr->alias->addr;
 509         base -= mr->alias_offset;
 510         render_memory_region(view, mr->alias, base, clip);
 511         return;
 512     }
 513
 514     /* Render subregions in priority order. */
 515     QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
 516         render_memory_region(view, subregion, base, clip);
 517     }
 518
 519     if (!mr->terminates) {
 520         return;
 521     }
 522
 523     offset_in_region = clip.start - base;
 524     base = clip.start;
 525     remain = clip.size;
 526
 527     /* Render the region itself into any gaps left by the current view. */
 528     for (i = 0; i < view->nr && remain; ++i) {
 529         if (base >= addrrange_end(view->ranges[i].addr)) {
 530             continue;
 531         }
 532         if (base < view->ranges[i].addr.start) {
 533             now = MIN(remain, view->ranges[i].addr.start - base);
 534             fr.mr = mr;
 535             fr.offset_in_region = offset_in_region;
 536             fr.addr = addrrange_make(base, now);
 537             fr.dirty_log_mask = mr->dirty_log_mask;
 538             fr.readable = mr->readable;
 539             flatview_insert(view, i, &fr);
 540             ++i;
 541             base += now;
 542             offset_in_region += now;
 543             remain -= now;
 544         }
 545         if (base == view->ranges[i].addr.start) {
 546             now = MIN(remain, view->ranges[i].addr.size);
 547             base += now;
 548             offset_in_region += now;
 549             remain -= now;
 550         }
 551     }
 552     if (remain) {
 553         fr.mr = mr;
 554         fr.offset_in_region = offset_in_region;
 555         fr.addr = addrrange_make(base, remain);
 556         fr.dirty_log_mask = mr->dirty_log_mask;
 557         fr.readable = mr->readable;
 558         flatview_insert(view, i, &fr);
 559     }
 560 }
 561
 562 /* Render a memory topology into a list of disjoint absolute ranges. */
 563 static FlatView generate_memory_topology(MemoryRegion *mr)
 564 {
 565     FlatView view;
 566
 567     flatview_init(&view);
 568
 569     render_memory_region(&view, mr, 0, addrrange_make(0, INT64_MAX));
 570     flatview_simplify(&view);
 571
 572     return view;
 573 }
 574
 575 static void address_space_add_del_ioeventfds(AddressSpace *as,
 576                                              MemoryRegionIoeventfd *fds_new,
 577                                              unsigned fds_new_nb,
 578                                              MemoryRegionIoeventfd *fds_old,
 579                                              unsigned fds_old_nb)
 580 {
 581     unsigned iold, inew;
 582
 583     /* Generate a symmetric difference of the old and new fd sets, adding
 584      * and deleting as necessary.
 585      */
 586
 587     iold = inew = 0;
 588     while (iold < fds_old_nb || inew < fds_new_nb) {
 589         if (iold < fds_old_nb
 590             && (inew == fds_new_nb
 591                 || memory_region_ioeventfd_before(fds_old[iold],
 592                                                   fds_new[inew]))) {
 593             as->ops->ioeventfd_del(as, &fds_old[iold]);
 594             ++iold;
 595         } else if (inew < fds_new_nb
 596                    && (iold == fds_old_nb
 597                        || memory_region_ioeventfd_before(fds_new[inew],
 598                                                          fds_old[iold]))) {
 599             as->ops->ioeventfd_add(as, &fds_new[inew]);
 600             ++inew;
 601         } else {
 602             ++iold;
 603             ++inew;
 604         }
 605     }
 606 }
 607
 608 static void address_space_update_ioeventfds(AddressSpace *as)
 609 {
 610     FlatRange *fr;
 611     unsigned ioeventfd_nb = 0;
 612     MemoryRegionIoeventfd *ioeventfds = NULL;
 613     AddrRange tmp;
 614     unsigned i;
 615
 616     FOR_EACH_FLAT_RANGE(fr, &as->current_map) {
 617         for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
 618             tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
 619                                   fr->addr.start - fr->offset_in_region);
 620             if (addrrange_intersects(fr->addr, tmp)) {
 621                 ++ioeventfd_nb;
 622                 ioeventfds = g_realloc(ioeventfds,
 623                                           ioeventfd_nb * sizeof(*ioeventfds));
 624                 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
 625                 ioeventfds[ioeventfd_nb-1].addr = tmp;
 626             }
 627         }
 628     }
 629
 630     address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
 631                                      as->ioeventfds, as->ioeventfd_nb);
 632
 633     g_free(as->ioeventfds);
 634     as->ioeventfds = ioeventfds;
 635     as->ioeventfd_nb = ioeventfd_nb;
 636 }
 637
 638 static void address_space_update_topology_pass(AddressSpace *as,
 639                                                FlatView old_view,
 640                                                FlatView new_view,
 641                                                bool adding)
 642 {
 643     unsigned iold, inew;
 644     FlatRange *frold, *frnew;
 645
 646     /* Generate a symmetric difference of the old and new memory maps.
 647      * Kill ranges in the old map, and instantiate ranges in the new map.
 648      */
 649     iold = inew = 0;
 650     while (iold < old_view.nr || inew < new_view.nr) {
 651         if (iold < old_view.nr) {
 652             frold = &old_view.ranges[iold];
 653         } else {
 654             frold = NULL;
 655         }
 656         if (inew < new_view.nr) {
 657             frnew = &new_view.ranges[inew];
 658         } else {
 659             frnew = NULL;
 660         }
 661
 662         if (frold
 663             && (!frnew
 664                 || frold->addr.start < frnew->addr.start
 665                 || (frold->addr.start == frnew->addr.start
 666                     && !flatrange_equal(frold, frnew)))) {
 667             /* In old, but (not in new, or in new but attributes changed). */
 668
 669             if (!adding) {
 670                 as->ops->range_del(as, frold);
 671             }
 672
 673             ++iold;
 674         } else if (frold && frnew && flatrange_equal(frold, frnew)) {
 675             /* In both (logging may have changed) */
 676
 677             if (adding) {
 678                 if (frold->dirty_log_mask && !frnew->dirty_log_mask) {
 679                     as->ops->log_stop(as, frnew);
 680                 } else if (frnew->dirty_log_mask && !frold->dirty_log_mask) {
 681                     as->ops->log_start(as, frnew);
 682                 }
 683             }
 684
 685             ++iold;
 686             ++inew;
 687         } else {
 688             /* In new */
 689
 690             if (adding) {
 691                 as->ops->range_add(as, frnew);
 692             }
 693
 694             ++inew;
 695         }
 696     }
 697 }
 698
 699
 700 static void address_space_update_topology(AddressSpace *as)
 701 {
 702     FlatView old_view = as->current_map;
 703     FlatView new_view = generate_memory_topology(as->root);
 704
 705     address_space_update_topology_pass(as, old_view, new_view, false);
 706     address_space_update_topology_pass(as, old_view, new_view, true);
 707
 708     as->current_map = new_view;
 709     flatview_destroy(&old_view);
 710     address_space_update_ioeventfds(as);
 711 }
 712
 713 static void memory_region_update_topology(void)
 714 {
 715     if (memory_region_transaction_depth) {
 716         return;
 717     }
 718
 719     if (address_space_memory.root) {
 720         address_space_update_topology(&address_space_memory);
 721     }
 722     if (address_space_io.root) {
 723         address_space_update_topology(&address_space_io);
 724     }
 725 }
 726
 727 void memory_region_transaction_begin(void)
 728 {
 729     ++memory_region_transaction_depth;
 730 }
 731
 732 void memory_region_transaction_commit(void)
 733 {
 734     assert(memory_region_transaction_depth);
 735     --memory_region_transaction_depth;
 736     memory_region_update_topology();
 737 }
 738
 739 static void memory_region_destructor_none(MemoryRegion *mr)
 740 {
 741 }
 742
 743 static void memory_region_destructor_ram(MemoryRegion *mr)
 744 {
 745     qemu_ram_free(mr->ram_addr);
 746 }
 747
 748 static void memory_region_destructor_ram_from_ptr(MemoryRegion *mr)
 749 {
 750     qemu_ram_free_from_ptr(mr->ram_addr);
 751 }
 752
 753 static void memory_region_destructor_iomem(MemoryRegion *mr)
 754 {
 755     cpu_unregister_io_memory(mr->ram_addr);
 756 }
 757
 758 static void memory_region_destructor_rom_device(MemoryRegion *mr)
 759 {
 760     qemu_ram_free(mr->ram_addr & TARGET_PAGE_MASK);
 761     cpu_unregister_io_memory(mr->ram_addr & ~(TARGET_PAGE_MASK | IO_MEM_ROMD));
 762 }
 763
 764 void memory_region_init(MemoryRegion *mr,
 765                         const char *name,
 766                         uint64_t size)
 767 {
 768     mr->ops = NULL;
 769     mr->parent = NULL;
 770     mr->size = size;
 771     mr->addr = 0;
 772     mr->offset = 0;
 773     mr->terminates = false;
 774     mr->readable = true;
 775     mr->destructor = memory_region_destructor_none;
 776     mr->priority = 0;
 777     mr->may_overlap = false;
 778     mr->alias = NULL;
 779     QTAILQ_INIT(&mr->subregions);
 780     memset(&mr->subregions_link, 0, sizeof mr->subregions_link);
 781     QTAILQ_INIT(&mr->coalesced);
 782     mr->name = g_strdup(name);
 783     mr->dirty_log_mask = 0;
 784     mr->ioeventfd_nb = 0;
 785     mr->ioeventfds = NULL;
 786 }
 787
 788 static bool memory_region_access_valid(MemoryRegion *mr,
 789                                        target_phys_addr_t addr,
 790                                        unsigned size)
 791 {
 792     if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
 793         return false;
 794     }
 795
 796     /* Treat zero as compatibility all valid */
 797     if (!mr->ops->valid.max_access_size) {
 798         return true;
 799     }
 800
 801     if (size > mr->ops->valid.max_access_size
 802         || size < mr->ops->valid.min_access_size) {
 803         return false;
 804     }
 805     return true;
 806 }
 807
 808 static uint32_t memory_region_read_thunk_n(void *_mr,
 809                                            target_phys_addr_t addr,
 810                                            unsigned size)
 811 {
 812     MemoryRegion *mr = _mr;
 813     uint64_t data = 0;
 814
 815     if (!memory_region_access_valid(mr, addr, size)) {
 816         return -1U; /* FIXME: better signalling */
 817     }
 818
 819     if (!mr->ops->read) {
 820         return mr->ops->old_mmio.read[bitops_ffsl(size)](mr->opaque, addr);
 821     }
 822
 823     /* FIXME: support unaligned access */
 824     access_with_adjusted_size(addr + mr->offset, &data, size,
 825                               mr->ops->impl.min_access_size,
 826                               mr->ops->impl.max_access_size,
 827                               memory_region_read_accessor, mr);
 828
 829     return data;
 830 }
 831
 832 static void memory_region_write_thunk_n(void *_mr,
 833                                         target_phys_addr_t addr,
 834                                         unsigned size,
 835                                         uint64_t data)
 836 {
 837     MemoryRegion *mr = _mr;
 838
 839     if (!memory_region_access_valid(mr, addr, size)) {
 840         return; /* FIXME: better signalling */
 841     }
 842
 843     if (!mr->ops->write) {
 844         mr->ops->old_mmio.write[bitops_ffsl(size)](mr->opaque, addr, data);
 845         return;
 846     }
 847
 848     /* FIXME: support unaligned access */
 849     access_with_adjusted_size(addr + mr->offset, &data, size,
 850                               mr->ops->impl.min_access_size,
 851                               mr->ops->impl.max_access_size,
 852                               memory_region_write_accessor, mr);
 853 }
 854
 855 static uint32_t memory_region_read_thunk_b(void *mr, target_phys_addr_t addr)
 856 {
 857     return memory_region_read_thunk_n(mr, addr, 1);
 858 }
 859
 860 static uint32_t memory_region_read_thunk_w(void *mr, target_phys_addr_t addr)
 861 {
 862     return memory_region_read_thunk_n(mr, addr, 2);
 863 }
 864
 865 static uint32_t memory_region_read_thunk_l(void *mr, target_phys_addr_t addr)
 866 {
 867     return memory_region_read_thunk_n(mr, addr, 4);
 868 }
 869
 870 static void memory_region_write_thunk_b(void *mr, target_phys_addr_t addr,
 871                                         uint32_t data)
 872 {
 873     memory_region_write_thunk_n(mr, addr, 1, data);
 874 }
 875
 876 static void memory_region_write_thunk_w(void *mr, target_phys_addr_t addr,
 877                                         uint32_t data)
 878 {
 879     memory_region_write_thunk_n(mr, addr, 2, data);
 880 }
 881
 882 static void memory_region_write_thunk_l(void *mr, target_phys_addr_t addr,
 883                                         uint32_t data)
 884 {
 885     memory_region_write_thunk_n(mr, addr, 4, data);
 886 }
 887
 888 static CPUReadMemoryFunc * const memory_region_read_thunk[] = {
 889     memory_region_read_thunk_b,
 890     memory_region_read_thunk_w,
 891     memory_region_read_thunk_l,
 892 };
 893
 894 static CPUWriteMemoryFunc * const memory_region_write_thunk[] = {
 895     memory_region_write_thunk_b,
 896     memory_region_write_thunk_w,
 897     memory_region_write_thunk_l,
 898 };
 899
 900 static void memory_region_prepare_ram_addr(MemoryRegion *mr)
 901 {
 902     if (mr->backend_registered) {
 903         return;
 904     }
 905
 906     mr->destructor = memory_region_destructor_iomem;
 907     mr->ram_addr = cpu_register_io_memory(memory_region_read_thunk,
 908                                           memory_region_write_thunk,
 909                                           mr,
 910                                           mr->ops->endianness);
 911     mr->backend_registered = true;
 912 }
 913
 914 void memory_region_init_io(MemoryRegion *mr,
 915                            const MemoryRegionOps *ops,
 916                            void *opaque,
 917                            const char *name,
 918                            uint64_t size)
 919 {
 920     memory_region_init(mr, name, size);
 921     mr->ops = ops;
 922     mr->opaque = opaque;
 923     mr->terminates = true;
 924     mr->backend_registered = false;
 925 }
 926
 927 void memory_region_init_ram(MemoryRegion *mr,
 928                             DeviceState *dev,
 929                             const char *name,
 930                             uint64_t size)
 931 {
 932     memory_region_init(mr, name, size);
 933     mr->terminates = true;
 934     mr->destructor = memory_region_destructor_ram;
 935     mr->ram_addr = qemu_ram_alloc(dev, name, size);
 936     mr->backend_registered = true;
 937 }
 938
 939 void memory_region_init_ram_ptr(MemoryRegion *mr,
 940                                 DeviceState *dev,
 941                                 const char *name,
 942                                 uint64_t size,
 943                                 void *ptr)
 944 {
 945     memory_region_init(mr, name, size);
 946     mr->terminates = true;
 947     mr->destructor = memory_region_destructor_ram_from_ptr;
 948     mr->ram_addr = qemu_ram_alloc_from_ptr(dev, name, size, ptr);
 949     mr->backend_registered = true;
 950 }
 951
 952 void memory_region_init_alias(MemoryRegion *mr,
 953                               const char *name,
 954                               MemoryRegion *orig,
 955                               target_phys_addr_t offset,
 956                               uint64_t size)
 957 {
 958     memory_region_init(mr, name, size);
 959     mr->alias = orig;
 960     mr->alias_offset = offset;
 961 }
 962
 963 void memory_region_init_rom_device(MemoryRegion *mr,
 964                                    const MemoryRegionOps *ops,
 965                                    void *opaque,
 966                                    DeviceState *dev,
 967                                    const char *name,
 968                                    uint64_t size)
 969 {
 970     memory_region_init(mr, name, size);
 971     mr->ops = ops;
 972     mr->opaque = opaque;
 973     mr->terminates = true;
 974     mr->destructor = memory_region_destructor_rom_device;
 975     mr->ram_addr = qemu_ram_alloc(dev, name, size);
 976     mr->ram_addr |= cpu_register_io_memory(memory_region_read_thunk,
 977                                            memory_region_write_thunk,
 978                                            mr,
 979                                            mr->ops->endianness);
 980     mr->ram_addr |= IO_MEM_ROMD;
 981     mr->backend_registered = true;
 982 }
 983
 984 void memory_region_destroy(MemoryRegion *mr)
 985 {
 986     assert(QTAILQ_EMPTY(&mr->subregions));
 987     mr->destructor(mr);
 988     memory_region_clear_coalescing(mr);
 989     g_free((char *)mr->name);
 990     g_free(mr->ioeventfds);
 991 }
 992
 993 uint64_t memory_region_size(MemoryRegion *mr)
 994 {
 995     return mr->size;
 996 }
 997
 998 void memory_region_set_offset(MemoryRegion *mr, target_phys_addr_t offset)
 999 {
1000     mr->offset = offset;
1001 }
1002
1003 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1004 {
1005     uint8_t mask = 1 << client;
1006
1007     mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1008     memory_region_update_topology();
1009 }
1010
1011 bool memory_region_get_dirty(MemoryRegion *mr, target_phys_addr_t addr,
1012                              unsigned client)
1013 {
1014     assert(mr->terminates);
1015     return cpu_physical_memory_get_dirty(mr->ram_addr + addr, 1 << client);
1016 }
1017
1018 void memory_region_set_dirty(MemoryRegion *mr, target_phys_addr_t addr)
1019 {
1020     assert(mr->terminates);
1021     return cpu_physical_memory_set_dirty(mr->ram_addr + addr);
1022 }
1023
1024 void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
1025 {
1026     FlatRange *fr;
1027
1028     FOR_EACH_FLAT_RANGE(fr, &address_space_memory.current_map) {
1029         if (fr->mr == mr) {
1030             cpu_physical_sync_dirty_bitmap(fr->addr.start,
1031                                            fr->addr.start + fr->addr.size);
1032         }
1033     }
1034 }
1035
1036 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
1037 {
1038     /* FIXME */
1039 }
1040
1041 void memory_region_rom_device_set_readable(MemoryRegion *mr, bool readable)
1042 {
1043     if (mr->readable != readable) {
1044         mr->readable = readable;
1045         memory_region_update_topology();
1046     }
1047 }
1048
1049 void memory_region_reset_dirty(MemoryRegion *mr, target_phys_addr_t addr,
1050                                target_phys_addr_t size, unsigned client)
1051 {
1052     assert(mr->terminates);
1053     cpu_physical_memory_reset_dirty(mr->ram_addr + addr,
1054                                     mr->ram_addr + addr + size,
1055                                     1 << client);
1056 }
1057
1058 void *memory_region_get_ram_ptr(MemoryRegion *mr)
1059 {
1060     if (mr->alias) {
1061         return memory_region_get_ram_ptr(mr->alias) + mr->alias_offset;
1062     }
1063
1064     assert(mr->terminates);
1065
1066     return qemu_get_ram_ptr(mr->ram_addr & TARGET_PAGE_MASK);
1067 }
1068
1069 static void memory_region_update_coalesced_range(MemoryRegion *mr)
1070 {
1071     FlatRange *fr;
1072     CoalescedMemoryRange *cmr;
1073     AddrRange tmp;
1074
1075     FOR_EACH_FLAT_RANGE(fr, &address_space_memory.current_map) {
1076         if (fr->mr == mr) {
1077             qemu_unregister_coalesced_mmio(fr->addr.start, fr->addr.size);
1078             QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
1079                 tmp = addrrange_shift(cmr->addr,
1080                                       fr->addr.start - fr->offset_in_region);
1081                 if (!addrrange_intersects(tmp, fr->addr)) {
1082                     continue;
1083                 }
1084                 tmp = addrrange_intersection(tmp, fr->addr);
1085                 qemu_register_coalesced_mmio(tmp.start, tmp.size);
1086             }
1087         }
1088     }
1089 }
1090
1091 void memory_region_set_coalescing(MemoryRegion *mr)
1092 {
1093     memory_region_clear_coalescing(mr);
1094     memory_region_add_coalescing(mr, 0, mr->size);
1095 }
1096
1097 void memory_region_add_coalescing(MemoryRegion *mr,
1098                                   target_phys_addr_t offset,
1099                                   uint64_t size)
1100 {
1101     CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
1102
1103     cmr->addr = addrrange_make(offset, size);
1104     QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
1105     memory_region_update_coalesced_range(mr);
1106 }
1107
1108 void memory_region_clear_coalescing(MemoryRegion *mr)
1109 {
1110     CoalescedMemoryRange *cmr;
1111
1112     while (!QTAILQ_EMPTY(&mr->coalesced)) {
1113         cmr = QTAILQ_FIRST(&mr->coalesced);
1114         QTAILQ_REMOVE(&mr->coalesced, cmr, link);
1115         g_free(cmr);
1116     }
1117     memory_region_update_coalesced_range(mr);
1118 }
1119
1120 void memory_region_add_eventfd(MemoryRegion *mr,
1121                                target_phys_addr_t addr,
1122                                unsigned size,
1123                                bool match_data,
1124                                uint64_t data,
1125                                int fd)
1126 {
1127     MemoryRegionIoeventfd mrfd = {
1128         .addr.start = addr,
1129         .addr.size = size,
1130         .match_data = match_data,
1131         .data = data,
1132         .fd = fd,
1133     };
1134     unsigned i;
1135
1136     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1137         if (memory_region_ioeventfd_before(mrfd, mr->ioeventfds[i])) {
1138             break;
1139         }
1140     }
1141     ++mr->ioeventfd_nb;
1142     mr->ioeventfds = g_realloc(mr->ioeventfds,
1143                                   sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
1144     memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
1145             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
1146     mr->ioeventfds[i] = mrfd;
1147     memory_region_update_topology();
1148 }
1149
1150 void memory_region_del_eventfd(MemoryRegion *mr,
1151                                target_phys_addr_t addr,
1152                                unsigned size,
1153                                bool match_data,
1154                                uint64_t data,
1155                                int fd)
1156 {
1157     MemoryRegionIoeventfd mrfd = {
1158         .addr.start = addr,
1159         .addr.size = size,
1160         .match_data = match_data,
1161         .data = data,
1162         .fd = fd,
1163     };
1164     unsigned i;
1165
1166     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1167         if (memory_region_ioeventfd_equal(mrfd, mr->ioeventfds[i])) {
1168             break;
1169         }
1170     }
1171     assert(i != mr->ioeventfd_nb);
1172     memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
1173             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
1174     --mr->ioeventfd_nb;
1175     mr->ioeventfds = g_realloc(mr->ioeventfds,
1176                                   sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
1177     memory_region_update_topology();
1178 }
1179
1180 static void memory_region_add_subregion_common(MemoryRegion *mr,
1181                                                target_phys_addr_t offset,
1182                                                MemoryRegion *subregion)
1183 {
1184     MemoryRegion *other;
1185
1186     assert(!subregion->parent);
1187     subregion->parent = mr;
1188     subregion->addr = offset;
1189     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1190         if (subregion->may_overlap || other->may_overlap) {
1191             continue;
1192         }
1193         if (offset >= other->offset + other->size
1194             || offset + subregion->size <= other->offset) {
1195             continue;
1196         }
1197 #if 0
1198         printf("warning: subregion collision %llx/%llx vs %llx/%llx\n",
1199                (unsigned long long)offset,
1200                (unsigned long long)subregion->size,
1201                (unsigned long long)other->offset,
1202                (unsigned long long)other->size);
1203 #endif
1204     }
1205     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1206         if (subregion->priority >= other->priority) {
1207             QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
1208             goto done;
1209         }
1210     }
1211     QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
1212 done:
1213     memory_region_update_topology();
1214 }
1215
1216
1217 void memory_region_add_subregion(MemoryRegion *mr,
1218                                  target_phys_addr_t offset,
1219                                  MemoryRegion *subregion)
1220 {
1221     subregion->may_overlap = false;
1222     subregion->priority = 0;
1223     memory_region_add_subregion_common(mr, offset, subregion);
1224 }
1225
1226 void memory_region_add_subregion_overlap(MemoryRegion *mr,
1227                                          target_phys_addr_t offset,
1228                                          MemoryRegion *subregion,
1229                                          unsigned priority)
1230 {
1231     subregion->may_overlap = true;
1232     subregion->priority = priority;
1233     memory_region_add_subregion_common(mr, offset, subregion);
1234 }
1235
1236 void memory_region_del_subregion(MemoryRegion *mr,
1237                                  MemoryRegion *subregion)
1238 {
1239     assert(subregion->parent == mr);
1240     subregion->parent = NULL;
1241     QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
1242     memory_region_update_topology();
1243 }
1244
1245 void set_system_memory_map(MemoryRegion *mr)
1246 {
1247     address_space_memory.root = mr;
1248     memory_region_update_topology();
1249 }
1250
1251 void set_system_io_map(MemoryRegion *mr)
1252 {
1253     address_space_io.root = mr;
1254     memory_region_update_topology();
1255 }