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