xref: /openbmc/qemu/system/memory.c (revision e0c72452)
1 /*
2  * Physical memory management
3  *
4  * Copyright 2011 Red Hat, Inc. and/or its affiliates
5  *
6  * Authors:
7  *  Avi Kivity <avi@redhat.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  * Contributions after 2012-01-13 are licensed under the terms of the
13  * GNU GPL, version 2 or (at your option) any later version.
14  */
15 
16 #include "qemu/osdep.h"
17 #include "qemu/log.h"
18 #include "qapi/error.h"
19 #include "exec/memory.h"
20 #include "qapi/visitor.h"
21 #include "qemu/bitops.h"
22 #include "qemu/error-report.h"
23 #include "qemu/main-loop.h"
24 #include "qemu/qemu-print.h"
25 #include "qom/object.h"
26 #include "trace.h"
27 
28 #include "exec/memory-internal.h"
29 #include "exec/ram_addr.h"
30 #include "sysemu/kvm.h"
31 #include "sysemu/runstate.h"
32 #include "sysemu/tcg.h"
33 #include "qemu/accel.h"
34 #include "hw/boards.h"
35 #include "migration/vmstate.h"
36 #include "exec/address-spaces.h"
37 
38 //#define DEBUG_UNASSIGNED
39 
40 static unsigned memory_region_transaction_depth;
41 static bool memory_region_update_pending;
42 static bool ioeventfd_update_pending;
43 unsigned int global_dirty_tracking;
44 
45 static QTAILQ_HEAD(, MemoryListener) memory_listeners
46     = QTAILQ_HEAD_INITIALIZER(memory_listeners);
47 
48 static QTAILQ_HEAD(, AddressSpace) address_spaces
49     = QTAILQ_HEAD_INITIALIZER(address_spaces);
50 
51 static GHashTable *flat_views;
52 
53 typedef struct AddrRange AddrRange;
54 
55 /*
56  * Note that signed integers are needed for negative offsetting in aliases
57  * (large MemoryRegion::alias_offset).
58  */
59 struct AddrRange {
60     Int128 start;
61     Int128 size;
62 };
63 
64 static AddrRange addrrange_make(Int128 start, Int128 size)
65 {
66     return (AddrRange) { start, size };
67 }
68 
69 static bool addrrange_equal(AddrRange r1, AddrRange r2)
70 {
71     return int128_eq(r1.start, r2.start) && int128_eq(r1.size, r2.size);
72 }
73 
74 static Int128 addrrange_end(AddrRange r)
75 {
76     return int128_add(r.start, r.size);
77 }
78 
79 static AddrRange addrrange_shift(AddrRange range, Int128 delta)
80 {
81     int128_addto(&range.start, delta);
82     return range;
83 }
84 
85 static bool addrrange_contains(AddrRange range, Int128 addr)
86 {
87     return int128_ge(addr, range.start)
88         && int128_lt(addr, addrrange_end(range));
89 }
90 
91 static bool addrrange_intersects(AddrRange r1, AddrRange r2)
92 {
93     return addrrange_contains(r1, r2.start)
94         || addrrange_contains(r2, r1.start);
95 }
96 
97 static AddrRange addrrange_intersection(AddrRange r1, AddrRange r2)
98 {
99     Int128 start = int128_max(r1.start, r2.start);
100     Int128 end = int128_min(addrrange_end(r1), addrrange_end(r2));
101     return addrrange_make(start, int128_sub(end, start));
102 }
103 
104 enum ListenerDirection { Forward, Reverse };
105 
106 #define MEMORY_LISTENER_CALL_GLOBAL(_callback, _direction, _args...)    \
107     do {                                                                \
108         MemoryListener *_listener;                                      \
109                                                                         \
110         switch (_direction) {                                           \
111         case Forward:                                                   \
112             QTAILQ_FOREACH(_listener, &memory_listeners, link) {        \
113                 if (_listener->_callback) {                             \
114                     _listener->_callback(_listener, ##_args);           \
115                 }                                                       \
116             }                                                           \
117             break;                                                      \
118         case Reverse:                                                   \
119             QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners, link) { \
120                 if (_listener->_callback) {                             \
121                     _listener->_callback(_listener, ##_args);           \
122                 }                                                       \
123             }                                                           \
124             break;                                                      \
125         default:                                                        \
126             abort();                                                    \
127         }                                                               \
128     } while (0)
129 
130 #define MEMORY_LISTENER_CALL(_as, _callback, _direction, _section, _args...) \
131     do {                                                                \
132         MemoryListener *_listener;                                      \
133                                                                         \
134         switch (_direction) {                                           \
135         case Forward:                                                   \
136             QTAILQ_FOREACH(_listener, &(_as)->listeners, link_as) {     \
137                 if (_listener->_callback) {                             \
138                     _listener->_callback(_listener, _section, ##_args); \
139                 }                                                       \
140             }                                                           \
141             break;                                                      \
142         case Reverse:                                                   \
143             QTAILQ_FOREACH_REVERSE(_listener, &(_as)->listeners, link_as) { \
144                 if (_listener->_callback) {                             \
145                     _listener->_callback(_listener, _section, ##_args); \
146                 }                                                       \
147             }                                                           \
148             break;                                                      \
149         default:                                                        \
150             abort();                                                    \
151         }                                                               \
152     } while (0)
153 
154 /* No need to ref/unref .mr, the FlatRange keeps it alive.  */
155 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...)  \
156     do {                                                                \
157         MemoryRegionSection mrs = section_from_flat_range(fr,           \
158                 address_space_to_flatview(as));                         \
159         MEMORY_LISTENER_CALL(as, callback, dir, &mrs, ##_args);         \
160     } while(0)
161 
162 struct CoalescedMemoryRange {
163     AddrRange addr;
164     QTAILQ_ENTRY(CoalescedMemoryRange) link;
165 };
166 
167 struct MemoryRegionIoeventfd {
168     AddrRange addr;
169     bool match_data;
170     uint64_t data;
171     EventNotifier *e;
172 };
173 
174 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd *a,
175                                            MemoryRegionIoeventfd *b)
176 {
177     if (int128_lt(a->addr.start, b->addr.start)) {
178         return true;
179     } else if (int128_gt(a->addr.start, b->addr.start)) {
180         return false;
181     } else if (int128_lt(a->addr.size, b->addr.size)) {
182         return true;
183     } else if (int128_gt(a->addr.size, b->addr.size)) {
184         return false;
185     } else if (a->match_data < b->match_data) {
186         return true;
187     } else  if (a->match_data > b->match_data) {
188         return false;
189     } else if (a->match_data) {
190         if (a->data < b->data) {
191             return true;
192         } else if (a->data > b->data) {
193             return false;
194         }
195     }
196     if (a->e < b->e) {
197         return true;
198     } else if (a->e > b->e) {
199         return false;
200     }
201     return false;
202 }
203 
204 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd *a,
205                                           MemoryRegionIoeventfd *b)
206 {
207     if (int128_eq(a->addr.start, b->addr.start) &&
208         (!int128_nz(a->addr.size) || !int128_nz(b->addr.size) ||
209          (int128_eq(a->addr.size, b->addr.size) &&
210           (a->match_data == b->match_data) &&
211           ((a->match_data && (a->data == b->data)) || !a->match_data) &&
212           (a->e == b->e))))
213         return true;
214 
215     return false;
216 }
217 
218 /* Range of memory in the global map.  Addresses are absolute. */
219 struct FlatRange {
220     MemoryRegion *mr;
221     hwaddr offset_in_region;
222     AddrRange addr;
223     uint8_t dirty_log_mask;
224     bool romd_mode;
225     bool readonly;
226     bool nonvolatile;
227     bool unmergeable;
228 };
229 
230 #define FOR_EACH_FLAT_RANGE(var, view)          \
231     for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
232 
233 static inline MemoryRegionSection
234 section_from_flat_range(FlatRange *fr, FlatView *fv)
235 {
236     return (MemoryRegionSection) {
237         .mr = fr->mr,
238         .fv = fv,
239         .offset_within_region = fr->offset_in_region,
240         .size = fr->addr.size,
241         .offset_within_address_space = int128_get64(fr->addr.start),
242         .readonly = fr->readonly,
243         .nonvolatile = fr->nonvolatile,
244         .unmergeable = fr->unmergeable,
245     };
246 }
247 
248 static bool flatrange_equal(FlatRange *a, FlatRange *b)
249 {
250     return a->mr == b->mr
251         && addrrange_equal(a->addr, b->addr)
252         && a->offset_in_region == b->offset_in_region
253         && a->romd_mode == b->romd_mode
254         && a->readonly == b->readonly
255         && a->nonvolatile == b->nonvolatile
256         && a->unmergeable == b->unmergeable;
257 }
258 
259 static FlatView *flatview_new(MemoryRegion *mr_root)
260 {
261     FlatView *view;
262 
263     view = g_new0(FlatView, 1);
264     view->ref = 1;
265     view->root = mr_root;
266     memory_region_ref(mr_root);
267     trace_flatview_new(view, mr_root);
268 
269     return view;
270 }
271 
272 /* Insert a range into a given position.  Caller is responsible for maintaining
273  * sorting order.
274  */
275 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
276 {
277     if (view->nr == view->nr_allocated) {
278         view->nr_allocated = MAX(2 * view->nr, 10);
279         view->ranges = g_realloc(view->ranges,
280                                     view->nr_allocated * sizeof(*view->ranges));
281     }
282     memmove(view->ranges + pos + 1, view->ranges + pos,
283             (view->nr - pos) * sizeof(FlatRange));
284     view->ranges[pos] = *range;
285     memory_region_ref(range->mr);
286     ++view->nr;
287 }
288 
289 static void flatview_destroy(FlatView *view)
290 {
291     int i;
292 
293     trace_flatview_destroy(view, view->root);
294     if (view->dispatch) {
295         address_space_dispatch_free(view->dispatch);
296     }
297     for (i = 0; i < view->nr; i++) {
298         memory_region_unref(view->ranges[i].mr);
299     }
300     g_free(view->ranges);
301     memory_region_unref(view->root);
302     g_free(view);
303 }
304 
305 static bool flatview_ref(FlatView *view)
306 {
307     return qatomic_fetch_inc_nonzero(&view->ref) > 0;
308 }
309 
310 void flatview_unref(FlatView *view)
311 {
312     if (qatomic_fetch_dec(&view->ref) == 1) {
313         trace_flatview_destroy_rcu(view, view->root);
314         assert(view->root);
315         call_rcu(view, flatview_destroy, rcu);
316     }
317 }
318 
319 static bool can_merge(FlatRange *r1, FlatRange *r2)
320 {
321     return int128_eq(addrrange_end(r1->addr), r2->addr.start)
322         && r1->mr == r2->mr
323         && int128_eq(int128_add(int128_make64(r1->offset_in_region),
324                                 r1->addr.size),
325                      int128_make64(r2->offset_in_region))
326         && r1->dirty_log_mask == r2->dirty_log_mask
327         && r1->romd_mode == r2->romd_mode
328         && r1->readonly == r2->readonly
329         && r1->nonvolatile == r2->nonvolatile
330         && !r1->unmergeable && !r2->unmergeable;
331 }
332 
333 /* Attempt to simplify a view by merging adjacent ranges */
334 static void flatview_simplify(FlatView *view)
335 {
336     unsigned i, j, k;
337 
338     i = 0;
339     while (i < view->nr) {
340         j = i + 1;
341         while (j < view->nr
342                && can_merge(&view->ranges[j-1], &view->ranges[j])) {
343             int128_addto(&view->ranges[i].addr.size, view->ranges[j].addr.size);
344             ++j;
345         }
346         ++i;
347         for (k = i; k < j; k++) {
348             memory_region_unref(view->ranges[k].mr);
349         }
350         memmove(&view->ranges[i], &view->ranges[j],
351                 (view->nr - j) * sizeof(view->ranges[j]));
352         view->nr -= j - i;
353     }
354 }
355 
356 static bool memory_region_big_endian(MemoryRegion *mr)
357 {
358 #if TARGET_BIG_ENDIAN
359     return mr->ops->endianness != DEVICE_LITTLE_ENDIAN;
360 #else
361     return mr->ops->endianness == DEVICE_BIG_ENDIAN;
362 #endif
363 }
364 
365 static void adjust_endianness(MemoryRegion *mr, uint64_t *data, MemOp op)
366 {
367     if ((op & MO_BSWAP) != devend_memop(mr->ops->endianness)) {
368         switch (op & MO_SIZE) {
369         case MO_8:
370             break;
371         case MO_16:
372             *data = bswap16(*data);
373             break;
374         case MO_32:
375             *data = bswap32(*data);
376             break;
377         case MO_64:
378             *data = bswap64(*data);
379             break;
380         default:
381             g_assert_not_reached();
382         }
383     }
384 }
385 
386 static inline void memory_region_shift_read_access(uint64_t *value,
387                                                    signed shift,
388                                                    uint64_t mask,
389                                                    uint64_t tmp)
390 {
391     if (shift >= 0) {
392         *value |= (tmp & mask) << shift;
393     } else {
394         *value |= (tmp & mask) >> -shift;
395     }
396 }
397 
398 static inline uint64_t memory_region_shift_write_access(uint64_t *value,
399                                                         signed shift,
400                                                         uint64_t mask)
401 {
402     uint64_t tmp;
403 
404     if (shift >= 0) {
405         tmp = (*value >> shift) & mask;
406     } else {
407         tmp = (*value << -shift) & mask;
408     }
409 
410     return tmp;
411 }
412 
413 static hwaddr memory_region_to_absolute_addr(MemoryRegion *mr, hwaddr offset)
414 {
415     MemoryRegion *root;
416     hwaddr abs_addr = offset;
417 
418     abs_addr += mr->addr;
419     for (root = mr; root->container; ) {
420         root = root->container;
421         abs_addr += root->addr;
422     }
423 
424     return abs_addr;
425 }
426 
427 static int get_cpu_index(void)
428 {
429     if (current_cpu) {
430         return current_cpu->cpu_index;
431     }
432     return -1;
433 }
434 
435 static MemTxResult  memory_region_read_accessor(MemoryRegion *mr,
436                                                 hwaddr addr,
437                                                 uint64_t *value,
438                                                 unsigned size,
439                                                 signed shift,
440                                                 uint64_t mask,
441                                                 MemTxAttrs attrs)
442 {
443     uint64_t tmp;
444 
445     tmp = mr->ops->read(mr->opaque, addr, size);
446     if (mr->subpage) {
447         trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
448     } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_READ)) {
449         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
450         trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size,
451                                      memory_region_name(mr));
452     }
453     memory_region_shift_read_access(value, shift, mask, tmp);
454     return MEMTX_OK;
455 }
456 
457 static MemTxResult memory_region_read_with_attrs_accessor(MemoryRegion *mr,
458                                                           hwaddr addr,
459                                                           uint64_t *value,
460                                                           unsigned size,
461                                                           signed shift,
462                                                           uint64_t mask,
463                                                           MemTxAttrs attrs)
464 {
465     uint64_t tmp = 0;
466     MemTxResult r;
467 
468     r = mr->ops->read_with_attrs(mr->opaque, addr, &tmp, size, attrs);
469     if (mr->subpage) {
470         trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
471     } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_READ)) {
472         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
473         trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size,
474                                      memory_region_name(mr));
475     }
476     memory_region_shift_read_access(value, shift, mask, tmp);
477     return r;
478 }
479 
480 static MemTxResult memory_region_write_accessor(MemoryRegion *mr,
481                                                 hwaddr addr,
482                                                 uint64_t *value,
483                                                 unsigned size,
484                                                 signed shift,
485                                                 uint64_t mask,
486                                                 MemTxAttrs attrs)
487 {
488     uint64_t tmp = memory_region_shift_write_access(value, shift, mask);
489 
490     if (mr->subpage) {
491         trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
492     } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_WRITE)) {
493         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
494         trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size,
495                                       memory_region_name(mr));
496     }
497     mr->ops->write(mr->opaque, addr, tmp, size);
498     return MEMTX_OK;
499 }
500 
501 static MemTxResult memory_region_write_with_attrs_accessor(MemoryRegion *mr,
502                                                            hwaddr addr,
503                                                            uint64_t *value,
504                                                            unsigned size,
505                                                            signed shift,
506                                                            uint64_t mask,
507                                                            MemTxAttrs attrs)
508 {
509     uint64_t tmp = memory_region_shift_write_access(value, shift, mask);
510 
511     if (mr->subpage) {
512         trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
513     } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_WRITE)) {
514         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
515         trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size,
516                                       memory_region_name(mr));
517     }
518     return mr->ops->write_with_attrs(mr->opaque, addr, tmp, size, attrs);
519 }
520 
521 static MemTxResult access_with_adjusted_size(hwaddr addr,
522                                       uint64_t *value,
523                                       unsigned size,
524                                       unsigned access_size_min,
525                                       unsigned access_size_max,
526                                       MemTxResult (*access_fn)
527                                                   (MemoryRegion *mr,
528                                                    hwaddr addr,
529                                                    uint64_t *value,
530                                                    unsigned size,
531                                                    signed shift,
532                                                    uint64_t mask,
533                                                    MemTxAttrs attrs),
534                                       MemoryRegion *mr,
535                                       MemTxAttrs attrs)
536 {
537     uint64_t access_mask;
538     unsigned access_size;
539     unsigned i;
540     MemTxResult r = MEMTX_OK;
541     bool reentrancy_guard_applied = false;
542 
543     if (!access_size_min) {
544         access_size_min = 1;
545     }
546     if (!access_size_max) {
547         access_size_max = 4;
548     }
549 
550     /* Do not allow more than one simultaneous access to a device's IO Regions */
551     if (mr->dev && !mr->disable_reentrancy_guard &&
552         !mr->ram_device && !mr->ram && !mr->rom_device && !mr->readonly) {
553         if (mr->dev->mem_reentrancy_guard.engaged_in_io) {
554             warn_report_once("Blocked re-entrant IO on MemoryRegion: "
555                              "%s at addr: 0x%" HWADDR_PRIX,
556                              memory_region_name(mr), addr);
557             return MEMTX_ACCESS_ERROR;
558         }
559         mr->dev->mem_reentrancy_guard.engaged_in_io = true;
560         reentrancy_guard_applied = true;
561     }
562 
563     /* FIXME: support unaligned access? */
564     access_size = MAX(MIN(size, access_size_max), access_size_min);
565     access_mask = MAKE_64BIT_MASK(0, access_size * 8);
566     if (memory_region_big_endian(mr)) {
567         for (i = 0; i < size; i += access_size) {
568             r |= access_fn(mr, addr + i, value, access_size,
569                         (size - access_size - i) * 8, access_mask, attrs);
570         }
571     } else {
572         for (i = 0; i < size; i += access_size) {
573             r |= access_fn(mr, addr + i, value, access_size, i * 8,
574                         access_mask, attrs);
575         }
576     }
577     if (mr->dev && reentrancy_guard_applied) {
578         mr->dev->mem_reentrancy_guard.engaged_in_io = false;
579     }
580     return r;
581 }
582 
583 static AddressSpace *memory_region_to_address_space(MemoryRegion *mr)
584 {
585     AddressSpace *as;
586 
587     while (mr->container) {
588         mr = mr->container;
589     }
590     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
591         if (mr == as->root) {
592             return as;
593         }
594     }
595     return NULL;
596 }
597 
598 /* Render a memory region into the global view.  Ranges in @view obscure
599  * ranges in @mr.
600  */
601 static void render_memory_region(FlatView *view,
602                                  MemoryRegion *mr,
603                                  Int128 base,
604                                  AddrRange clip,
605                                  bool readonly,
606                                  bool nonvolatile,
607                                  bool unmergeable)
608 {
609     MemoryRegion *subregion;
610     unsigned i;
611     hwaddr offset_in_region;
612     Int128 remain;
613     Int128 now;
614     FlatRange fr;
615     AddrRange tmp;
616 
617     if (!mr->enabled) {
618         return;
619     }
620 
621     int128_addto(&base, int128_make64(mr->addr));
622     readonly |= mr->readonly;
623     nonvolatile |= mr->nonvolatile;
624     unmergeable |= mr->unmergeable;
625 
626     tmp = addrrange_make(base, mr->size);
627 
628     if (!addrrange_intersects(tmp, clip)) {
629         return;
630     }
631 
632     clip = addrrange_intersection(tmp, clip);
633 
634     if (mr->alias) {
635         int128_subfrom(&base, int128_make64(mr->alias->addr));
636         int128_subfrom(&base, int128_make64(mr->alias_offset));
637         render_memory_region(view, mr->alias, base, clip,
638                              readonly, nonvolatile, unmergeable);
639         return;
640     }
641 
642     /* Render subregions in priority order. */
643     QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
644         render_memory_region(view, subregion, base, clip,
645                              readonly, nonvolatile, unmergeable);
646     }
647 
648     if (!mr->terminates) {
649         return;
650     }
651 
652     offset_in_region = int128_get64(int128_sub(clip.start, base));
653     base = clip.start;
654     remain = clip.size;
655 
656     fr.mr = mr;
657     fr.dirty_log_mask = memory_region_get_dirty_log_mask(mr);
658     fr.romd_mode = mr->romd_mode;
659     fr.readonly = readonly;
660     fr.nonvolatile = nonvolatile;
661     fr.unmergeable = unmergeable;
662 
663     /* Render the region itself into any gaps left by the current view. */
664     for (i = 0; i < view->nr && int128_nz(remain); ++i) {
665         if (int128_ge(base, addrrange_end(view->ranges[i].addr))) {
666             continue;
667         }
668         if (int128_lt(base, view->ranges[i].addr.start)) {
669             now = int128_min(remain,
670                              int128_sub(view->ranges[i].addr.start, base));
671             fr.offset_in_region = offset_in_region;
672             fr.addr = addrrange_make(base, now);
673             flatview_insert(view, i, &fr);
674             ++i;
675             int128_addto(&base, now);
676             offset_in_region += int128_get64(now);
677             int128_subfrom(&remain, now);
678         }
679         now = int128_sub(int128_min(int128_add(base, remain),
680                                     addrrange_end(view->ranges[i].addr)),
681                          base);
682         int128_addto(&base, now);
683         offset_in_region += int128_get64(now);
684         int128_subfrom(&remain, now);
685     }
686     if (int128_nz(remain)) {
687         fr.offset_in_region = offset_in_region;
688         fr.addr = addrrange_make(base, remain);
689         flatview_insert(view, i, &fr);
690     }
691 }
692 
693 void flatview_for_each_range(FlatView *fv, flatview_cb cb , void *opaque)
694 {
695     FlatRange *fr;
696 
697     assert(fv);
698     assert(cb);
699 
700     FOR_EACH_FLAT_RANGE(fr, fv) {
701         if (cb(fr->addr.start, fr->addr.size, fr->mr,
702                fr->offset_in_region, opaque)) {
703             break;
704         }
705     }
706 }
707 
708 static MemoryRegion *memory_region_get_flatview_root(MemoryRegion *mr)
709 {
710     while (mr->enabled) {
711         if (mr->alias) {
712             if (!mr->alias_offset && int128_ge(mr->size, mr->alias->size)) {
713                 /* The alias is included in its entirety.  Use it as
714                  * the "real" root, so that we can share more FlatViews.
715                  */
716                 mr = mr->alias;
717                 continue;
718             }
719         } else if (!mr->terminates) {
720             unsigned int found = 0;
721             MemoryRegion *child, *next = NULL;
722             QTAILQ_FOREACH(child, &mr->subregions, subregions_link) {
723                 if (child->enabled) {
724                     if (++found > 1) {
725                         next = NULL;
726                         break;
727                     }
728                     if (!child->addr && int128_ge(mr->size, child->size)) {
729                         /* A child is included in its entirety.  If it's the only
730                          * enabled one, use it in the hope of finding an alias down the
731                          * way. This will also let us share FlatViews.
732                          */
733                         next = child;
734                     }
735                 }
736             }
737             if (found == 0) {
738                 return NULL;
739             }
740             if (next) {
741                 mr = next;
742                 continue;
743             }
744         }
745 
746         return mr;
747     }
748 
749     return NULL;
750 }
751 
752 /* Render a memory topology into a list of disjoint absolute ranges. */
753 static FlatView *generate_memory_topology(MemoryRegion *mr)
754 {
755     int i;
756     FlatView *view;
757 
758     view = flatview_new(mr);
759 
760     if (mr) {
761         render_memory_region(view, mr, int128_zero(),
762                              addrrange_make(int128_zero(), int128_2_64()),
763                              false, false, false);
764     }
765     flatview_simplify(view);
766 
767     view->dispatch = address_space_dispatch_new(view);
768     for (i = 0; i < view->nr; i++) {
769         MemoryRegionSection mrs =
770             section_from_flat_range(&view->ranges[i], view);
771         flatview_add_to_dispatch(view, &mrs);
772     }
773     address_space_dispatch_compact(view->dispatch);
774     g_hash_table_replace(flat_views, mr, view);
775 
776     return view;
777 }
778 
779 static void address_space_add_del_ioeventfds(AddressSpace *as,
780                                              MemoryRegionIoeventfd *fds_new,
781                                              unsigned fds_new_nb,
782                                              MemoryRegionIoeventfd *fds_old,
783                                              unsigned fds_old_nb)
784 {
785     unsigned iold, inew;
786     MemoryRegionIoeventfd *fd;
787     MemoryRegionSection section;
788 
789     /* Generate a symmetric difference of the old and new fd sets, adding
790      * and deleting as necessary.
791      */
792 
793     iold = inew = 0;
794     while (iold < fds_old_nb || inew < fds_new_nb) {
795         if (iold < fds_old_nb
796             && (inew == fds_new_nb
797                 || memory_region_ioeventfd_before(&fds_old[iold],
798                                                   &fds_new[inew]))) {
799             fd = &fds_old[iold];
800             section = (MemoryRegionSection) {
801                 .fv = address_space_to_flatview(as),
802                 .offset_within_address_space = int128_get64(fd->addr.start),
803                 .size = fd->addr.size,
804             };
805             MEMORY_LISTENER_CALL(as, eventfd_del, Forward, &section,
806                                  fd->match_data, fd->data, fd->e);
807             ++iold;
808         } else if (inew < fds_new_nb
809                    && (iold == fds_old_nb
810                        || memory_region_ioeventfd_before(&fds_new[inew],
811                                                          &fds_old[iold]))) {
812             fd = &fds_new[inew];
813             section = (MemoryRegionSection) {
814                 .fv = address_space_to_flatview(as),
815                 .offset_within_address_space = int128_get64(fd->addr.start),
816                 .size = fd->addr.size,
817             };
818             MEMORY_LISTENER_CALL(as, eventfd_add, Reverse, &section,
819                                  fd->match_data, fd->data, fd->e);
820             ++inew;
821         } else {
822             ++iold;
823             ++inew;
824         }
825     }
826 }
827 
828 FlatView *address_space_get_flatview(AddressSpace *as)
829 {
830     FlatView *view;
831 
832     RCU_READ_LOCK_GUARD();
833     do {
834         view = address_space_to_flatview(as);
835         /* If somebody has replaced as->current_map concurrently,
836          * flatview_ref returns false.
837          */
838     } while (!flatview_ref(view));
839     return view;
840 }
841 
842 static void address_space_update_ioeventfds(AddressSpace *as)
843 {
844     FlatView *view;
845     FlatRange *fr;
846     unsigned ioeventfd_nb = 0;
847     unsigned ioeventfd_max;
848     MemoryRegionIoeventfd *ioeventfds;
849     AddrRange tmp;
850     unsigned i;
851 
852     if (!as->ioeventfd_notifiers) {
853         return;
854     }
855 
856     /*
857      * It is likely that the number of ioeventfds hasn't changed much, so use
858      * the previous size as the starting value, with some headroom to avoid
859      * gratuitous reallocations.
860      */
861     ioeventfd_max = QEMU_ALIGN_UP(as->ioeventfd_nb, 4);
862     ioeventfds = g_new(MemoryRegionIoeventfd, ioeventfd_max);
863 
864     view = address_space_get_flatview(as);
865     FOR_EACH_FLAT_RANGE(fr, view) {
866         for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
867             tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
868                                   int128_sub(fr->addr.start,
869                                              int128_make64(fr->offset_in_region)));
870             if (addrrange_intersects(fr->addr, tmp)) {
871                 ++ioeventfd_nb;
872                 if (ioeventfd_nb > ioeventfd_max) {
873                     ioeventfd_max = MAX(ioeventfd_max * 2, 4);
874                     ioeventfds = g_realloc(ioeventfds,
875                             ioeventfd_max * sizeof(*ioeventfds));
876                 }
877                 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
878                 ioeventfds[ioeventfd_nb-1].addr = tmp;
879             }
880         }
881     }
882 
883     address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
884                                      as->ioeventfds, as->ioeventfd_nb);
885 
886     g_free(as->ioeventfds);
887     as->ioeventfds = ioeventfds;
888     as->ioeventfd_nb = ioeventfd_nb;
889     flatview_unref(view);
890 }
891 
892 /*
893  * Notify the memory listeners about the coalesced IO change events of
894  * range `cmr'.  Only the part that has intersection of the specified
895  * FlatRange will be sent.
896  */
897 static void flat_range_coalesced_io_notify(FlatRange *fr, AddressSpace *as,
898                                            CoalescedMemoryRange *cmr, bool add)
899 {
900     AddrRange tmp;
901 
902     tmp = addrrange_shift(cmr->addr,
903                           int128_sub(fr->addr.start,
904                                      int128_make64(fr->offset_in_region)));
905     if (!addrrange_intersects(tmp, fr->addr)) {
906         return;
907     }
908     tmp = addrrange_intersection(tmp, fr->addr);
909 
910     if (add) {
911         MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, coalesced_io_add,
912                                       int128_get64(tmp.start),
913                                       int128_get64(tmp.size));
914     } else {
915         MEMORY_LISTENER_UPDATE_REGION(fr, as, Reverse, coalesced_io_del,
916                                       int128_get64(tmp.start),
917                                       int128_get64(tmp.size));
918     }
919 }
920 
921 static void flat_range_coalesced_io_del(FlatRange *fr, AddressSpace *as)
922 {
923     CoalescedMemoryRange *cmr;
924 
925     QTAILQ_FOREACH(cmr, &fr->mr->coalesced, link) {
926         flat_range_coalesced_io_notify(fr, as, cmr, false);
927     }
928 }
929 
930 static void flat_range_coalesced_io_add(FlatRange *fr, AddressSpace *as)
931 {
932     MemoryRegion *mr = fr->mr;
933     CoalescedMemoryRange *cmr;
934 
935     if (QTAILQ_EMPTY(&mr->coalesced)) {
936         return;
937     }
938 
939     QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
940         flat_range_coalesced_io_notify(fr, as, cmr, true);
941     }
942 }
943 
944 static void address_space_update_topology_pass(AddressSpace *as,
945                                                const FlatView *old_view,
946                                                const FlatView *new_view,
947                                                bool adding)
948 {
949     unsigned iold, inew;
950     FlatRange *frold, *frnew;
951 
952     /* Generate a symmetric difference of the old and new memory maps.
953      * Kill ranges in the old map, and instantiate ranges in the new map.
954      */
955     iold = inew = 0;
956     while (iold < old_view->nr || inew < new_view->nr) {
957         if (iold < old_view->nr) {
958             frold = &old_view->ranges[iold];
959         } else {
960             frold = NULL;
961         }
962         if (inew < new_view->nr) {
963             frnew = &new_view->ranges[inew];
964         } else {
965             frnew = NULL;
966         }
967 
968         if (frold
969             && (!frnew
970                 || int128_lt(frold->addr.start, frnew->addr.start)
971                 || (int128_eq(frold->addr.start, frnew->addr.start)
972                     && !flatrange_equal(frold, frnew)))) {
973             /* In old but not in new, or in both but attributes changed. */
974 
975             if (!adding) {
976                 flat_range_coalesced_io_del(frold, as);
977                 MEMORY_LISTENER_UPDATE_REGION(frold, as, Reverse, region_del);
978             }
979 
980             ++iold;
981         } else if (frold && frnew && flatrange_equal(frold, frnew)) {
982             /* In both and unchanged (except logging may have changed) */
983 
984             if (adding) {
985                 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_nop);
986                 if (frnew->dirty_log_mask & ~frold->dirty_log_mask) {
987                     MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, log_start,
988                                                   frold->dirty_log_mask,
989                                                   frnew->dirty_log_mask);
990                 }
991                 if (frold->dirty_log_mask & ~frnew->dirty_log_mask) {
992                     MEMORY_LISTENER_UPDATE_REGION(frnew, as, Reverse, log_stop,
993                                                   frold->dirty_log_mask,
994                                                   frnew->dirty_log_mask);
995                 }
996             }
997 
998             ++iold;
999             ++inew;
1000         } else {
1001             /* In new */
1002 
1003             if (adding) {
1004                 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_add);
1005                 flat_range_coalesced_io_add(frnew, as);
1006             }
1007 
1008             ++inew;
1009         }
1010     }
1011 }
1012 
1013 static void flatviews_init(void)
1014 {
1015     static FlatView *empty_view;
1016 
1017     if (flat_views) {
1018         return;
1019     }
1020 
1021     flat_views = g_hash_table_new_full(g_direct_hash, g_direct_equal, NULL,
1022                                        (GDestroyNotify) flatview_unref);
1023     if (!empty_view) {
1024         empty_view = generate_memory_topology(NULL);
1025         /* We keep it alive forever in the global variable.  */
1026         flatview_ref(empty_view);
1027     } else {
1028         g_hash_table_replace(flat_views, NULL, empty_view);
1029         flatview_ref(empty_view);
1030     }
1031 }
1032 
1033 static void flatviews_reset(void)
1034 {
1035     AddressSpace *as;
1036 
1037     if (flat_views) {
1038         g_hash_table_unref(flat_views);
1039         flat_views = NULL;
1040     }
1041     flatviews_init();
1042 
1043     /* Render unique FVs */
1044     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1045         MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1046 
1047         if (g_hash_table_lookup(flat_views, physmr)) {
1048             continue;
1049         }
1050 
1051         generate_memory_topology(physmr);
1052     }
1053 }
1054 
1055 static void address_space_set_flatview(AddressSpace *as)
1056 {
1057     FlatView *old_view = address_space_to_flatview(as);
1058     MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1059     FlatView *new_view = g_hash_table_lookup(flat_views, physmr);
1060 
1061     assert(new_view);
1062 
1063     if (old_view == new_view) {
1064         return;
1065     }
1066 
1067     if (old_view) {
1068         flatview_ref(old_view);
1069     }
1070 
1071     flatview_ref(new_view);
1072 
1073     if (!QTAILQ_EMPTY(&as->listeners)) {
1074         FlatView tmpview = { .nr = 0 }, *old_view2 = old_view;
1075 
1076         if (!old_view2) {
1077             old_view2 = &tmpview;
1078         }
1079         address_space_update_topology_pass(as, old_view2, new_view, false);
1080         address_space_update_topology_pass(as, old_view2, new_view, true);
1081     }
1082 
1083     /* Writes are protected by the BQL.  */
1084     qatomic_rcu_set(&as->current_map, new_view);
1085     if (old_view) {
1086         flatview_unref(old_view);
1087     }
1088 
1089     /* Note that all the old MemoryRegions are still alive up to this
1090      * point.  This relieves most MemoryListeners from the need to
1091      * ref/unref the MemoryRegions they get---unless they use them
1092      * outside the iothread mutex, in which case precise reference
1093      * counting is necessary.
1094      */
1095     if (old_view) {
1096         flatview_unref(old_view);
1097     }
1098 }
1099 
1100 static void address_space_update_topology(AddressSpace *as)
1101 {
1102     MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1103 
1104     flatviews_init();
1105     if (!g_hash_table_lookup(flat_views, physmr)) {
1106         generate_memory_topology(physmr);
1107     }
1108     address_space_set_flatview(as);
1109 }
1110 
1111 void memory_region_transaction_begin(void)
1112 {
1113     qemu_flush_coalesced_mmio_buffer();
1114     ++memory_region_transaction_depth;
1115 }
1116 
1117 void memory_region_transaction_commit(void)
1118 {
1119     AddressSpace *as;
1120 
1121     assert(memory_region_transaction_depth);
1122     assert(qemu_mutex_iothread_locked());
1123 
1124     --memory_region_transaction_depth;
1125     if (!memory_region_transaction_depth) {
1126         if (memory_region_update_pending) {
1127             flatviews_reset();
1128 
1129             MEMORY_LISTENER_CALL_GLOBAL(begin, Forward);
1130 
1131             QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1132                 address_space_set_flatview(as);
1133                 address_space_update_ioeventfds(as);
1134             }
1135             memory_region_update_pending = false;
1136             ioeventfd_update_pending = false;
1137             MEMORY_LISTENER_CALL_GLOBAL(commit, Forward);
1138         } else if (ioeventfd_update_pending) {
1139             QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1140                 address_space_update_ioeventfds(as);
1141             }
1142             ioeventfd_update_pending = false;
1143         }
1144    }
1145 }
1146 
1147 static void memory_region_destructor_none(MemoryRegion *mr)
1148 {
1149 }
1150 
1151 static void memory_region_destructor_ram(MemoryRegion *mr)
1152 {
1153     qemu_ram_free(mr->ram_block);
1154 }
1155 
1156 static bool memory_region_need_escape(char c)
1157 {
1158     return c == '/' || c == '[' || c == '\\' || c == ']';
1159 }
1160 
1161 static char *memory_region_escape_name(const char *name)
1162 {
1163     const char *p;
1164     char *escaped, *q;
1165     uint8_t c;
1166     size_t bytes = 0;
1167 
1168     for (p = name; *p; p++) {
1169         bytes += memory_region_need_escape(*p) ? 4 : 1;
1170     }
1171     if (bytes == p - name) {
1172        return g_memdup(name, bytes + 1);
1173     }
1174 
1175     escaped = g_malloc(bytes + 1);
1176     for (p = name, q = escaped; *p; p++) {
1177         c = *p;
1178         if (unlikely(memory_region_need_escape(c))) {
1179             *q++ = '\\';
1180             *q++ = 'x';
1181             *q++ = "0123456789abcdef"[c >> 4];
1182             c = "0123456789abcdef"[c & 15];
1183         }
1184         *q++ = c;
1185     }
1186     *q = 0;
1187     return escaped;
1188 }
1189 
1190 static void memory_region_do_init(MemoryRegion *mr,
1191                                   Object *owner,
1192                                   const char *name,
1193                                   uint64_t size)
1194 {
1195     mr->size = int128_make64(size);
1196     if (size == UINT64_MAX) {
1197         mr->size = int128_2_64();
1198     }
1199     mr->name = g_strdup(name);
1200     mr->owner = owner;
1201     mr->dev = (DeviceState *) object_dynamic_cast(mr->owner, TYPE_DEVICE);
1202     mr->ram_block = NULL;
1203 
1204     if (name) {
1205         char *escaped_name = memory_region_escape_name(name);
1206         char *name_array = g_strdup_printf("%s[*]", escaped_name);
1207 
1208         if (!owner) {
1209             owner = container_get(qdev_get_machine(), "/unattached");
1210         }
1211 
1212         object_property_add_child(owner, name_array, OBJECT(mr));
1213         object_unref(OBJECT(mr));
1214         g_free(name_array);
1215         g_free(escaped_name);
1216     }
1217 }
1218 
1219 void memory_region_init(MemoryRegion *mr,
1220                         Object *owner,
1221                         const char *name,
1222                         uint64_t size)
1223 {
1224     object_initialize(mr, sizeof(*mr), TYPE_MEMORY_REGION);
1225     memory_region_do_init(mr, owner, name, size);
1226 }
1227 
1228 static void memory_region_get_container(Object *obj, Visitor *v,
1229                                         const char *name, void *opaque,
1230                                         Error **errp)
1231 {
1232     MemoryRegion *mr = MEMORY_REGION(obj);
1233     char *path = (char *)"";
1234 
1235     if (mr->container) {
1236         path = object_get_canonical_path(OBJECT(mr->container));
1237     }
1238     visit_type_str(v, name, &path, errp);
1239     if (mr->container) {
1240         g_free(path);
1241     }
1242 }
1243 
1244 static Object *memory_region_resolve_container(Object *obj, void *opaque,
1245                                                const char *part)
1246 {
1247     MemoryRegion *mr = MEMORY_REGION(obj);
1248 
1249     return OBJECT(mr->container);
1250 }
1251 
1252 static void memory_region_get_priority(Object *obj, Visitor *v,
1253                                        const char *name, void *opaque,
1254                                        Error **errp)
1255 {
1256     MemoryRegion *mr = MEMORY_REGION(obj);
1257     int32_t value = mr->priority;
1258 
1259     visit_type_int32(v, name, &value, errp);
1260 }
1261 
1262 static void memory_region_get_size(Object *obj, Visitor *v, const char *name,
1263                                    void *opaque, Error **errp)
1264 {
1265     MemoryRegion *mr = MEMORY_REGION(obj);
1266     uint64_t value = memory_region_size(mr);
1267 
1268     visit_type_uint64(v, name, &value, errp);
1269 }
1270 
1271 static void memory_region_initfn(Object *obj)
1272 {
1273     MemoryRegion *mr = MEMORY_REGION(obj);
1274     ObjectProperty *op;
1275 
1276     mr->ops = &unassigned_mem_ops;
1277     mr->enabled = true;
1278     mr->romd_mode = true;
1279     mr->destructor = memory_region_destructor_none;
1280     QTAILQ_INIT(&mr->subregions);
1281     QTAILQ_INIT(&mr->coalesced);
1282 
1283     op = object_property_add(OBJECT(mr), "container",
1284                              "link<" TYPE_MEMORY_REGION ">",
1285                              memory_region_get_container,
1286                              NULL, /* memory_region_set_container */
1287                              NULL, NULL);
1288     op->resolve = memory_region_resolve_container;
1289 
1290     object_property_add_uint64_ptr(OBJECT(mr), "addr",
1291                                    &mr->addr, OBJ_PROP_FLAG_READ);
1292     object_property_add(OBJECT(mr), "priority", "uint32",
1293                         memory_region_get_priority,
1294                         NULL, /* memory_region_set_priority */
1295                         NULL, NULL);
1296     object_property_add(OBJECT(mr), "size", "uint64",
1297                         memory_region_get_size,
1298                         NULL, /* memory_region_set_size, */
1299                         NULL, NULL);
1300 }
1301 
1302 static void iommu_memory_region_initfn(Object *obj)
1303 {
1304     MemoryRegion *mr = MEMORY_REGION(obj);
1305 
1306     mr->is_iommu = true;
1307 }
1308 
1309 static uint64_t unassigned_mem_read(void *opaque, hwaddr addr,
1310                                     unsigned size)
1311 {
1312 #ifdef DEBUG_UNASSIGNED
1313     printf("Unassigned mem read " HWADDR_FMT_plx "\n", addr);
1314 #endif
1315     return 0;
1316 }
1317 
1318 static void unassigned_mem_write(void *opaque, hwaddr addr,
1319                                  uint64_t val, unsigned size)
1320 {
1321 #ifdef DEBUG_UNASSIGNED
1322     printf("Unassigned mem write " HWADDR_FMT_plx " = 0x%"PRIx64"\n", addr, val);
1323 #endif
1324 }
1325 
1326 static bool unassigned_mem_accepts(void *opaque, hwaddr addr,
1327                                    unsigned size, bool is_write,
1328                                    MemTxAttrs attrs)
1329 {
1330     return false;
1331 }
1332 
1333 const MemoryRegionOps unassigned_mem_ops = {
1334     .valid.accepts = unassigned_mem_accepts,
1335     .endianness = DEVICE_NATIVE_ENDIAN,
1336 };
1337 
1338 static uint64_t memory_region_ram_device_read(void *opaque,
1339                                               hwaddr addr, unsigned size)
1340 {
1341     MemoryRegion *mr = opaque;
1342     uint64_t data = (uint64_t)~0;
1343 
1344     switch (size) {
1345     case 1:
1346         data = *(uint8_t *)(mr->ram_block->host + addr);
1347         break;
1348     case 2:
1349         data = *(uint16_t *)(mr->ram_block->host + addr);
1350         break;
1351     case 4:
1352         data = *(uint32_t *)(mr->ram_block->host + addr);
1353         break;
1354     case 8:
1355         data = *(uint64_t *)(mr->ram_block->host + addr);
1356         break;
1357     }
1358 
1359     trace_memory_region_ram_device_read(get_cpu_index(), mr, addr, data, size);
1360 
1361     return data;
1362 }
1363 
1364 static void memory_region_ram_device_write(void *opaque, hwaddr addr,
1365                                            uint64_t data, unsigned size)
1366 {
1367     MemoryRegion *mr = opaque;
1368 
1369     trace_memory_region_ram_device_write(get_cpu_index(), mr, addr, data, size);
1370 
1371     switch (size) {
1372     case 1:
1373         *(uint8_t *)(mr->ram_block->host + addr) = (uint8_t)data;
1374         break;
1375     case 2:
1376         *(uint16_t *)(mr->ram_block->host + addr) = (uint16_t)data;
1377         break;
1378     case 4:
1379         *(uint32_t *)(mr->ram_block->host + addr) = (uint32_t)data;
1380         break;
1381     case 8:
1382         *(uint64_t *)(mr->ram_block->host + addr) = data;
1383         break;
1384     }
1385 }
1386 
1387 static const MemoryRegionOps ram_device_mem_ops = {
1388     .read = memory_region_ram_device_read,
1389     .write = memory_region_ram_device_write,
1390     .endianness = DEVICE_HOST_ENDIAN,
1391     .valid = {
1392         .min_access_size = 1,
1393         .max_access_size = 8,
1394         .unaligned = true,
1395     },
1396     .impl = {
1397         .min_access_size = 1,
1398         .max_access_size = 8,
1399         .unaligned = true,
1400     },
1401 };
1402 
1403 bool memory_region_access_valid(MemoryRegion *mr,
1404                                 hwaddr addr,
1405                                 unsigned size,
1406                                 bool is_write,
1407                                 MemTxAttrs attrs)
1408 {
1409     if (mr->ops->valid.accepts
1410         && !mr->ops->valid.accepts(mr->opaque, addr, size, is_write, attrs)) {
1411         qemu_log_mask(LOG_GUEST_ERROR, "Invalid %s at addr 0x%" HWADDR_PRIX
1412                       ", size %u, region '%s', reason: rejected\n",
1413                       is_write ? "write" : "read",
1414                       addr, size, memory_region_name(mr));
1415         return false;
1416     }
1417 
1418     if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1419         qemu_log_mask(LOG_GUEST_ERROR, "Invalid %s at addr 0x%" HWADDR_PRIX
1420                       ", size %u, region '%s', reason: unaligned\n",
1421                       is_write ? "write" : "read",
1422                       addr, size, memory_region_name(mr));
1423         return false;
1424     }
1425 
1426     /* Treat zero as compatibility all valid */
1427     if (!mr->ops->valid.max_access_size) {
1428         return true;
1429     }
1430 
1431     if (size > mr->ops->valid.max_access_size
1432         || size < mr->ops->valid.min_access_size) {
1433         qemu_log_mask(LOG_GUEST_ERROR, "Invalid %s at addr 0x%" HWADDR_PRIX
1434                       ", size %u, region '%s', reason: invalid size "
1435                       "(min:%u max:%u)\n",
1436                       is_write ? "write" : "read",
1437                       addr, size, memory_region_name(mr),
1438                       mr->ops->valid.min_access_size,
1439                       mr->ops->valid.max_access_size);
1440         return false;
1441     }
1442     return true;
1443 }
1444 
1445 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1446                                                 hwaddr addr,
1447                                                 uint64_t *pval,
1448                                                 unsigned size,
1449                                                 MemTxAttrs attrs)
1450 {
1451     *pval = 0;
1452 
1453     if (mr->ops->read) {
1454         return access_with_adjusted_size(addr, pval, size,
1455                                          mr->ops->impl.min_access_size,
1456                                          mr->ops->impl.max_access_size,
1457                                          memory_region_read_accessor,
1458                                          mr, attrs);
1459     } else {
1460         return access_with_adjusted_size(addr, pval, size,
1461                                          mr->ops->impl.min_access_size,
1462                                          mr->ops->impl.max_access_size,
1463                                          memory_region_read_with_attrs_accessor,
1464                                          mr, attrs);
1465     }
1466 }
1467 
1468 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1469                                         hwaddr addr,
1470                                         uint64_t *pval,
1471                                         MemOp op,
1472                                         MemTxAttrs attrs)
1473 {
1474     unsigned size = memop_size(op);
1475     MemTxResult r;
1476 
1477     if (mr->alias) {
1478         return memory_region_dispatch_read(mr->alias,
1479                                            mr->alias_offset + addr,
1480                                            pval, op, attrs);
1481     }
1482     if (!memory_region_access_valid(mr, addr, size, false, attrs)) {
1483         *pval = unassigned_mem_read(mr, addr, size);
1484         return MEMTX_DECODE_ERROR;
1485     }
1486 
1487     r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1488     adjust_endianness(mr, pval, op);
1489     return r;
1490 }
1491 
1492 /* Return true if an eventfd was signalled */
1493 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1494                                                     hwaddr addr,
1495                                                     uint64_t data,
1496                                                     unsigned size,
1497                                                     MemTxAttrs attrs)
1498 {
1499     MemoryRegionIoeventfd ioeventfd = {
1500         .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1501         .data = data,
1502     };
1503     unsigned i;
1504 
1505     for (i = 0; i < mr->ioeventfd_nb; i++) {
1506         ioeventfd.match_data = mr->ioeventfds[i].match_data;
1507         ioeventfd.e = mr->ioeventfds[i].e;
1508 
1509         if (memory_region_ioeventfd_equal(&ioeventfd, &mr->ioeventfds[i])) {
1510             event_notifier_set(ioeventfd.e);
1511             return true;
1512         }
1513     }
1514 
1515     return false;
1516 }
1517 
1518 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1519                                          hwaddr addr,
1520                                          uint64_t data,
1521                                          MemOp op,
1522                                          MemTxAttrs attrs)
1523 {
1524     unsigned size = memop_size(op);
1525 
1526     if (mr->alias) {
1527         return memory_region_dispatch_write(mr->alias,
1528                                             mr->alias_offset + addr,
1529                                             data, op, attrs);
1530     }
1531     if (!memory_region_access_valid(mr, addr, size, true, attrs)) {
1532         unassigned_mem_write(mr, addr, data, size);
1533         return MEMTX_DECODE_ERROR;
1534     }
1535 
1536     adjust_endianness(mr, &data, op);
1537 
1538     if ((!kvm_eventfds_enabled()) &&
1539         memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1540         return MEMTX_OK;
1541     }
1542 
1543     if (mr->ops->write) {
1544         return access_with_adjusted_size(addr, &data, size,
1545                                          mr->ops->impl.min_access_size,
1546                                          mr->ops->impl.max_access_size,
1547                                          memory_region_write_accessor, mr,
1548                                          attrs);
1549     } else {
1550         return
1551             access_with_adjusted_size(addr, &data, size,
1552                                       mr->ops->impl.min_access_size,
1553                                       mr->ops->impl.max_access_size,
1554                                       memory_region_write_with_attrs_accessor,
1555                                       mr, attrs);
1556     }
1557 }
1558 
1559 void memory_region_init_io(MemoryRegion *mr,
1560                            Object *owner,
1561                            const MemoryRegionOps *ops,
1562                            void *opaque,
1563                            const char *name,
1564                            uint64_t size)
1565 {
1566     memory_region_init(mr, owner, name, size);
1567     mr->ops = ops ? ops : &unassigned_mem_ops;
1568     mr->opaque = opaque;
1569     mr->terminates = true;
1570 }
1571 
1572 void memory_region_init_ram_nomigrate(MemoryRegion *mr,
1573                                       Object *owner,
1574                                       const char *name,
1575                                       uint64_t size,
1576                                       Error **errp)
1577 {
1578     memory_region_init_ram_flags_nomigrate(mr, owner, name, size, 0, errp);
1579 }
1580 
1581 void memory_region_init_ram_flags_nomigrate(MemoryRegion *mr,
1582                                             Object *owner,
1583                                             const char *name,
1584                                             uint64_t size,
1585                                             uint32_t ram_flags,
1586                                             Error **errp)
1587 {
1588     Error *err = NULL;
1589     memory_region_init(mr, owner, name, size);
1590     mr->ram = true;
1591     mr->terminates = true;
1592     mr->destructor = memory_region_destructor_ram;
1593     mr->ram_block = qemu_ram_alloc(size, ram_flags, mr, &err);
1594     if (err) {
1595         mr->size = int128_zero();
1596         object_unparent(OBJECT(mr));
1597         error_propagate(errp, err);
1598     }
1599 }
1600 
1601 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1602                                        Object *owner,
1603                                        const char *name,
1604                                        uint64_t size,
1605                                        uint64_t max_size,
1606                                        void (*resized)(const char*,
1607                                                        uint64_t length,
1608                                                        void *host),
1609                                        Error **errp)
1610 {
1611     Error *err = NULL;
1612     memory_region_init(mr, owner, name, size);
1613     mr->ram = true;
1614     mr->terminates = true;
1615     mr->destructor = memory_region_destructor_ram;
1616     mr->ram_block = qemu_ram_alloc_resizeable(size, max_size, resized,
1617                                               mr, &err);
1618     if (err) {
1619         mr->size = int128_zero();
1620         object_unparent(OBJECT(mr));
1621         error_propagate(errp, err);
1622     }
1623 }
1624 
1625 #ifdef CONFIG_POSIX
1626 void memory_region_init_ram_from_file(MemoryRegion *mr,
1627                                       Object *owner,
1628                                       const char *name,
1629                                       uint64_t size,
1630                                       uint64_t align,
1631                                       uint32_t ram_flags,
1632                                       const char *path,
1633                                       ram_addr_t offset,
1634                                       Error **errp)
1635 {
1636     Error *err = NULL;
1637     memory_region_init(mr, owner, name, size);
1638     mr->ram = true;
1639     mr->readonly = !!(ram_flags & RAM_READONLY);
1640     mr->terminates = true;
1641     mr->destructor = memory_region_destructor_ram;
1642     mr->align = align;
1643     mr->ram_block = qemu_ram_alloc_from_file(size, mr, ram_flags, path,
1644                                              offset, &err);
1645     if (err) {
1646         mr->size = int128_zero();
1647         object_unparent(OBJECT(mr));
1648         error_propagate(errp, err);
1649     }
1650 }
1651 
1652 void memory_region_init_ram_from_fd(MemoryRegion *mr,
1653                                     Object *owner,
1654                                     const char *name,
1655                                     uint64_t size,
1656                                     uint32_t ram_flags,
1657                                     int fd,
1658                                     ram_addr_t offset,
1659                                     Error **errp)
1660 {
1661     Error *err = NULL;
1662     memory_region_init(mr, owner, name, size);
1663     mr->ram = true;
1664     mr->readonly = !!(ram_flags & RAM_READONLY);
1665     mr->terminates = true;
1666     mr->destructor = memory_region_destructor_ram;
1667     mr->ram_block = qemu_ram_alloc_from_fd(size, mr, ram_flags, fd, offset,
1668                                            &err);
1669     if (err) {
1670         mr->size = int128_zero();
1671         object_unparent(OBJECT(mr));
1672         error_propagate(errp, err);
1673     }
1674 }
1675 #endif
1676 
1677 void memory_region_init_ram_ptr(MemoryRegion *mr,
1678                                 Object *owner,
1679                                 const char *name,
1680                                 uint64_t size,
1681                                 void *ptr)
1682 {
1683     memory_region_init(mr, owner, name, size);
1684     mr->ram = true;
1685     mr->terminates = true;
1686     mr->destructor = memory_region_destructor_ram;
1687 
1688     /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL.  */
1689     assert(ptr != NULL);
1690     mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1691 }
1692 
1693 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
1694                                        Object *owner,
1695                                        const char *name,
1696                                        uint64_t size,
1697                                        void *ptr)
1698 {
1699     memory_region_init(mr, owner, name, size);
1700     mr->ram = true;
1701     mr->terminates = true;
1702     mr->ram_device = true;
1703     mr->ops = &ram_device_mem_ops;
1704     mr->opaque = mr;
1705     mr->destructor = memory_region_destructor_ram;
1706 
1707     /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL.  */
1708     assert(ptr != NULL);
1709     mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1710 }
1711 
1712 void memory_region_init_alias(MemoryRegion *mr,
1713                               Object *owner,
1714                               const char *name,
1715                               MemoryRegion *orig,
1716                               hwaddr offset,
1717                               uint64_t size)
1718 {
1719     memory_region_init(mr, owner, name, size);
1720     mr->alias = orig;
1721     mr->alias_offset = offset;
1722 }
1723 
1724 void memory_region_init_rom_nomigrate(MemoryRegion *mr,
1725                                       Object *owner,
1726                                       const char *name,
1727                                       uint64_t size,
1728                                       Error **errp)
1729 {
1730     memory_region_init_ram_flags_nomigrate(mr, owner, name, size, 0, errp);
1731     mr->readonly = true;
1732 }
1733 
1734 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1735                                              Object *owner,
1736                                              const MemoryRegionOps *ops,
1737                                              void *opaque,
1738                                              const char *name,
1739                                              uint64_t size,
1740                                              Error **errp)
1741 {
1742     Error *err = NULL;
1743     assert(ops);
1744     memory_region_init(mr, owner, name, size);
1745     mr->ops = ops;
1746     mr->opaque = opaque;
1747     mr->terminates = true;
1748     mr->rom_device = true;
1749     mr->destructor = memory_region_destructor_ram;
1750     mr->ram_block = qemu_ram_alloc(size, 0, mr, &err);
1751     if (err) {
1752         mr->size = int128_zero();
1753         object_unparent(OBJECT(mr));
1754         error_propagate(errp, err);
1755     }
1756 }
1757 
1758 void memory_region_init_iommu(void *_iommu_mr,
1759                               size_t instance_size,
1760                               const char *mrtypename,
1761                               Object *owner,
1762                               const char *name,
1763                               uint64_t size)
1764 {
1765     struct IOMMUMemoryRegion *iommu_mr;
1766     struct MemoryRegion *mr;
1767 
1768     object_initialize(_iommu_mr, instance_size, mrtypename);
1769     mr = MEMORY_REGION(_iommu_mr);
1770     memory_region_do_init(mr, owner, name, size);
1771     iommu_mr = IOMMU_MEMORY_REGION(mr);
1772     mr->terminates = true;  /* then re-forwards */
1773     QLIST_INIT(&iommu_mr->iommu_notify);
1774     iommu_mr->iommu_notify_flags = IOMMU_NOTIFIER_NONE;
1775 }
1776 
1777 static void memory_region_finalize(Object *obj)
1778 {
1779     MemoryRegion *mr = MEMORY_REGION(obj);
1780 
1781     assert(!mr->container);
1782 
1783     /* We know the region is not visible in any address space (it
1784      * does not have a container and cannot be a root either because
1785      * it has no references, so we can blindly clear mr->enabled.
1786      * memory_region_set_enabled instead could trigger a transaction
1787      * and cause an infinite loop.
1788      */
1789     mr->enabled = false;
1790     memory_region_transaction_begin();
1791     while (!QTAILQ_EMPTY(&mr->subregions)) {
1792         MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1793         memory_region_del_subregion(mr, subregion);
1794     }
1795     memory_region_transaction_commit();
1796 
1797     mr->destructor(mr);
1798     memory_region_clear_coalescing(mr);
1799     g_free((char *)mr->name);
1800     g_free(mr->ioeventfds);
1801 }
1802 
1803 Object *memory_region_owner(MemoryRegion *mr)
1804 {
1805     Object *obj = OBJECT(mr);
1806     return obj->parent;
1807 }
1808 
1809 void memory_region_ref(MemoryRegion *mr)
1810 {
1811     /* MMIO callbacks most likely will access data that belongs
1812      * to the owner, hence the need to ref/unref the owner whenever
1813      * the memory region is in use.
1814      *
1815      * The memory region is a child of its owner.  As long as the
1816      * owner doesn't call unparent itself on the memory region,
1817      * ref-ing the owner will also keep the memory region alive.
1818      * Memory regions without an owner are supposed to never go away;
1819      * we do not ref/unref them because it slows down DMA sensibly.
1820      */
1821     if (mr && mr->owner) {
1822         object_ref(mr->owner);
1823     }
1824 }
1825 
1826 void memory_region_unref(MemoryRegion *mr)
1827 {
1828     if (mr && mr->owner) {
1829         object_unref(mr->owner);
1830     }
1831 }
1832 
1833 uint64_t memory_region_size(MemoryRegion *mr)
1834 {
1835     if (int128_eq(mr->size, int128_2_64())) {
1836         return UINT64_MAX;
1837     }
1838     return int128_get64(mr->size);
1839 }
1840 
1841 const char *memory_region_name(const MemoryRegion *mr)
1842 {
1843     if (!mr->name) {
1844         ((MemoryRegion *)mr)->name =
1845             g_strdup(object_get_canonical_path_component(OBJECT(mr)));
1846     }
1847     return mr->name;
1848 }
1849 
1850 bool memory_region_is_ram_device(MemoryRegion *mr)
1851 {
1852     return mr->ram_device;
1853 }
1854 
1855 bool memory_region_is_protected(MemoryRegion *mr)
1856 {
1857     return mr->ram && (mr->ram_block->flags & RAM_PROTECTED);
1858 }
1859 
1860 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1861 {
1862     uint8_t mask = mr->dirty_log_mask;
1863     RAMBlock *rb = mr->ram_block;
1864 
1865     if (global_dirty_tracking && ((rb && qemu_ram_is_migratable(rb)) ||
1866                              memory_region_is_iommu(mr))) {
1867         mask |= (1 << DIRTY_MEMORY_MIGRATION);
1868     }
1869 
1870     if (tcg_enabled() && rb) {
1871         /* TCG only cares about dirty memory logging for RAM, not IOMMU.  */
1872         mask |= (1 << DIRTY_MEMORY_CODE);
1873     }
1874     return mask;
1875 }
1876 
1877 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1878 {
1879     return memory_region_get_dirty_log_mask(mr) & (1 << client);
1880 }
1881 
1882 static int memory_region_update_iommu_notify_flags(IOMMUMemoryRegion *iommu_mr,
1883                                                    Error **errp)
1884 {
1885     IOMMUNotifierFlag flags = IOMMU_NOTIFIER_NONE;
1886     IOMMUNotifier *iommu_notifier;
1887     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1888     int ret = 0;
1889 
1890     IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
1891         flags |= iommu_notifier->notifier_flags;
1892     }
1893 
1894     if (flags != iommu_mr->iommu_notify_flags && imrc->notify_flag_changed) {
1895         ret = imrc->notify_flag_changed(iommu_mr,
1896                                         iommu_mr->iommu_notify_flags,
1897                                         flags, errp);
1898     }
1899 
1900     if (!ret) {
1901         iommu_mr->iommu_notify_flags = flags;
1902     }
1903     return ret;
1904 }
1905 
1906 int memory_region_iommu_set_page_size_mask(IOMMUMemoryRegion *iommu_mr,
1907                                            uint64_t page_size_mask,
1908                                            Error **errp)
1909 {
1910     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1911     int ret = 0;
1912 
1913     if (imrc->iommu_set_page_size_mask) {
1914         ret = imrc->iommu_set_page_size_mask(iommu_mr, page_size_mask, errp);
1915     }
1916     return ret;
1917 }
1918 
1919 int memory_region_register_iommu_notifier(MemoryRegion *mr,
1920                                           IOMMUNotifier *n, Error **errp)
1921 {
1922     IOMMUMemoryRegion *iommu_mr;
1923     int ret;
1924 
1925     if (mr->alias) {
1926         return memory_region_register_iommu_notifier(mr->alias, n, errp);
1927     }
1928 
1929     /* We need to register for at least one bitfield */
1930     iommu_mr = IOMMU_MEMORY_REGION(mr);
1931     assert(n->notifier_flags != IOMMU_NOTIFIER_NONE);
1932     assert(n->start <= n->end);
1933     assert(n->iommu_idx >= 0 &&
1934            n->iommu_idx < memory_region_iommu_num_indexes(iommu_mr));
1935 
1936     QLIST_INSERT_HEAD(&iommu_mr->iommu_notify, n, node);
1937     ret = memory_region_update_iommu_notify_flags(iommu_mr, errp);
1938     if (ret) {
1939         QLIST_REMOVE(n, node);
1940     }
1941     return ret;
1942 }
1943 
1944 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr)
1945 {
1946     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1947 
1948     if (imrc->get_min_page_size) {
1949         return imrc->get_min_page_size(iommu_mr);
1950     }
1951     return TARGET_PAGE_SIZE;
1952 }
1953 
1954 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
1955 {
1956     MemoryRegion *mr = MEMORY_REGION(iommu_mr);
1957     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1958     hwaddr addr, granularity;
1959     IOMMUTLBEntry iotlb;
1960 
1961     /* If the IOMMU has its own replay callback, override */
1962     if (imrc->replay) {
1963         imrc->replay(iommu_mr, n);
1964         return;
1965     }
1966 
1967     granularity = memory_region_iommu_get_min_page_size(iommu_mr);
1968 
1969     for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1970         iotlb = imrc->translate(iommu_mr, addr, IOMMU_NONE, n->iommu_idx);
1971         if (iotlb.perm != IOMMU_NONE) {
1972             n->notify(n, &iotlb);
1973         }
1974 
1975         /* if (2^64 - MR size) < granularity, it's possible to get an
1976          * infinite loop here.  This should catch such a wraparound */
1977         if ((addr + granularity) < addr) {
1978             break;
1979         }
1980     }
1981 }
1982 
1983 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1984                                              IOMMUNotifier *n)
1985 {
1986     IOMMUMemoryRegion *iommu_mr;
1987 
1988     if (mr->alias) {
1989         memory_region_unregister_iommu_notifier(mr->alias, n);
1990         return;
1991     }
1992     QLIST_REMOVE(n, node);
1993     iommu_mr = IOMMU_MEMORY_REGION(mr);
1994     memory_region_update_iommu_notify_flags(iommu_mr, NULL);
1995 }
1996 
1997 void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
1998                                     IOMMUTLBEvent *event)
1999 {
2000     IOMMUTLBEntry *entry = &event->entry;
2001     hwaddr entry_end = entry->iova + entry->addr_mask;
2002     IOMMUTLBEntry tmp = *entry;
2003 
2004     if (event->type == IOMMU_NOTIFIER_UNMAP) {
2005         assert(entry->perm == IOMMU_NONE);
2006     }
2007 
2008     /*
2009      * Skip the notification if the notification does not overlap
2010      * with registered range.
2011      */
2012     if (notifier->start > entry_end || notifier->end < entry->iova) {
2013         return;
2014     }
2015 
2016     if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
2017         /* Crop (iova, addr_mask) to range */
2018         tmp.iova = MAX(tmp.iova, notifier->start);
2019         tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
2020     } else {
2021         assert(entry->iova >= notifier->start && entry_end <= notifier->end);
2022     }
2023 
2024     if (event->type & notifier->notifier_flags) {
2025         notifier->notify(notifier, &tmp);
2026     }
2027 }
2028 
2029 void memory_region_unmap_iommu_notifier_range(IOMMUNotifier *notifier)
2030 {
2031     IOMMUTLBEvent event;
2032 
2033     event.type = IOMMU_NOTIFIER_UNMAP;
2034     event.entry.target_as = &address_space_memory;
2035     event.entry.iova = notifier->start;
2036     event.entry.perm = IOMMU_NONE;
2037     event.entry.addr_mask = notifier->end - notifier->start;
2038 
2039     memory_region_notify_iommu_one(notifier, &event);
2040 }
2041 
2042 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
2043                                 int iommu_idx,
2044                                 IOMMUTLBEvent event)
2045 {
2046     IOMMUNotifier *iommu_notifier;
2047 
2048     assert(memory_region_is_iommu(MEMORY_REGION(iommu_mr)));
2049 
2050     IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
2051         if (iommu_notifier->iommu_idx == iommu_idx) {
2052             memory_region_notify_iommu_one(iommu_notifier, &event);
2053         }
2054     }
2055 }
2056 
2057 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
2058                                  enum IOMMUMemoryRegionAttr attr,
2059                                  void *data)
2060 {
2061     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2062 
2063     if (!imrc->get_attr) {
2064         return -EINVAL;
2065     }
2066 
2067     return imrc->get_attr(iommu_mr, attr, data);
2068 }
2069 
2070 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr,
2071                                        MemTxAttrs attrs)
2072 {
2073     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2074 
2075     if (!imrc->attrs_to_index) {
2076         return 0;
2077     }
2078 
2079     return imrc->attrs_to_index(iommu_mr, attrs);
2080 }
2081 
2082 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr)
2083 {
2084     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2085 
2086     if (!imrc->num_indexes) {
2087         return 1;
2088     }
2089 
2090     return imrc->num_indexes(iommu_mr);
2091 }
2092 
2093 RamDiscardManager *memory_region_get_ram_discard_manager(MemoryRegion *mr)
2094 {
2095     if (!memory_region_is_ram(mr)) {
2096         return NULL;
2097     }
2098     return mr->rdm;
2099 }
2100 
2101 void memory_region_set_ram_discard_manager(MemoryRegion *mr,
2102                                            RamDiscardManager *rdm)
2103 {
2104     g_assert(memory_region_is_ram(mr));
2105     g_assert(!rdm || !mr->rdm);
2106     mr->rdm = rdm;
2107 }
2108 
2109 uint64_t ram_discard_manager_get_min_granularity(const RamDiscardManager *rdm,
2110                                                  const MemoryRegion *mr)
2111 {
2112     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2113 
2114     g_assert(rdmc->get_min_granularity);
2115     return rdmc->get_min_granularity(rdm, mr);
2116 }
2117 
2118 bool ram_discard_manager_is_populated(const RamDiscardManager *rdm,
2119                                       const MemoryRegionSection *section)
2120 {
2121     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2122 
2123     g_assert(rdmc->is_populated);
2124     return rdmc->is_populated(rdm, section);
2125 }
2126 
2127 int ram_discard_manager_replay_populated(const RamDiscardManager *rdm,
2128                                          MemoryRegionSection *section,
2129                                          ReplayRamPopulate replay_fn,
2130                                          void *opaque)
2131 {
2132     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2133 
2134     g_assert(rdmc->replay_populated);
2135     return rdmc->replay_populated(rdm, section, replay_fn, opaque);
2136 }
2137 
2138 void ram_discard_manager_replay_discarded(const RamDiscardManager *rdm,
2139                                           MemoryRegionSection *section,
2140                                           ReplayRamDiscard replay_fn,
2141                                           void *opaque)
2142 {
2143     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2144 
2145     g_assert(rdmc->replay_discarded);
2146     rdmc->replay_discarded(rdm, section, replay_fn, opaque);
2147 }
2148 
2149 void ram_discard_manager_register_listener(RamDiscardManager *rdm,
2150                                            RamDiscardListener *rdl,
2151                                            MemoryRegionSection *section)
2152 {
2153     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2154 
2155     g_assert(rdmc->register_listener);
2156     rdmc->register_listener(rdm, rdl, section);
2157 }
2158 
2159 void ram_discard_manager_unregister_listener(RamDiscardManager *rdm,
2160                                              RamDiscardListener *rdl)
2161 {
2162     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2163 
2164     g_assert(rdmc->unregister_listener);
2165     rdmc->unregister_listener(rdm, rdl);
2166 }
2167 
2168 /* Called with rcu_read_lock held.  */
2169 bool memory_get_xlat_addr(IOMMUTLBEntry *iotlb, void **vaddr,
2170                           ram_addr_t *ram_addr, bool *read_only,
2171                           bool *mr_has_discard_manager)
2172 {
2173     MemoryRegion *mr;
2174     hwaddr xlat;
2175     hwaddr len = iotlb->addr_mask + 1;
2176     bool writable = iotlb->perm & IOMMU_WO;
2177 
2178     if (mr_has_discard_manager) {
2179         *mr_has_discard_manager = false;
2180     }
2181     /*
2182      * The IOMMU TLB entry we have just covers translation through
2183      * this IOMMU to its immediate target.  We need to translate
2184      * it the rest of the way through to memory.
2185      */
2186     mr = address_space_translate(&address_space_memory, iotlb->translated_addr,
2187                                  &xlat, &len, writable, MEMTXATTRS_UNSPECIFIED);
2188     if (!memory_region_is_ram(mr)) {
2189         error_report("iommu map to non memory area %" HWADDR_PRIx "", xlat);
2190         return false;
2191     } else if (memory_region_has_ram_discard_manager(mr)) {
2192         RamDiscardManager *rdm = memory_region_get_ram_discard_manager(mr);
2193         MemoryRegionSection tmp = {
2194             .mr = mr,
2195             .offset_within_region = xlat,
2196             .size = int128_make64(len),
2197         };
2198         if (mr_has_discard_manager) {
2199             *mr_has_discard_manager = true;
2200         }
2201         /*
2202          * Malicious VMs can map memory into the IOMMU, which is expected
2203          * to remain discarded. vfio will pin all pages, populating memory.
2204          * Disallow that. vmstate priorities make sure any RamDiscardManager
2205          * were already restored before IOMMUs are restored.
2206          */
2207         if (!ram_discard_manager_is_populated(rdm, &tmp)) {
2208             error_report("iommu map to discarded memory (e.g., unplugged via"
2209                          " virtio-mem): %" HWADDR_PRIx "",
2210                          iotlb->translated_addr);
2211             return false;
2212         }
2213     }
2214 
2215     /*
2216      * Translation truncates length to the IOMMU page size,
2217      * check that it did not truncate too much.
2218      */
2219     if (len & iotlb->addr_mask) {
2220         error_report("iommu has granularity incompatible with target AS");
2221         return false;
2222     }
2223 
2224     if (vaddr) {
2225         *vaddr = memory_region_get_ram_ptr(mr) + xlat;
2226     }
2227 
2228     if (ram_addr) {
2229         *ram_addr = memory_region_get_ram_addr(mr) + xlat;
2230     }
2231 
2232     if (read_only) {
2233         *read_only = !writable || mr->readonly;
2234     }
2235 
2236     return true;
2237 }
2238 
2239 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
2240 {
2241     uint8_t mask = 1 << client;
2242     uint8_t old_logging;
2243 
2244     assert(client == DIRTY_MEMORY_VGA);
2245     old_logging = mr->vga_logging_count;
2246     mr->vga_logging_count += log ? 1 : -1;
2247     if (!!old_logging == !!mr->vga_logging_count) {
2248         return;
2249     }
2250 
2251     memory_region_transaction_begin();
2252     mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
2253     memory_region_update_pending |= mr->enabled;
2254     memory_region_transaction_commit();
2255 }
2256 
2257 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
2258                              hwaddr size)
2259 {
2260     assert(mr->ram_block);
2261     cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr,
2262                                         size,
2263                                         memory_region_get_dirty_log_mask(mr));
2264 }
2265 
2266 /*
2267  * If memory region `mr' is NULL, do global sync.  Otherwise, sync
2268  * dirty bitmap for the specified memory region.
2269  */
2270 static void memory_region_sync_dirty_bitmap(MemoryRegion *mr, bool last_stage)
2271 {
2272     MemoryListener *listener;
2273     AddressSpace *as;
2274     FlatView *view;
2275     FlatRange *fr;
2276 
2277     /* If the same address space has multiple log_sync listeners, we
2278      * visit that address space's FlatView multiple times.  But because
2279      * log_sync listeners are rare, it's still cheaper than walking each
2280      * address space once.
2281      */
2282     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2283         if (listener->log_sync) {
2284             as = listener->address_space;
2285             view = address_space_get_flatview(as);
2286             FOR_EACH_FLAT_RANGE(fr, view) {
2287                 if (fr->dirty_log_mask && (!mr || fr->mr == mr)) {
2288                     MemoryRegionSection mrs = section_from_flat_range(fr, view);
2289                     listener->log_sync(listener, &mrs);
2290                 }
2291             }
2292             flatview_unref(view);
2293             trace_memory_region_sync_dirty(mr ? mr->name : "(all)", listener->name, 0);
2294         } else if (listener->log_sync_global) {
2295             /*
2296              * No matter whether MR is specified, what we can do here
2297              * is to do a global sync, because we are not capable to
2298              * sync in a finer granularity.
2299              */
2300             listener->log_sync_global(listener, last_stage);
2301             trace_memory_region_sync_dirty(mr ? mr->name : "(all)", listener->name, 1);
2302         }
2303     }
2304 }
2305 
2306 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start,
2307                                       hwaddr len)
2308 {
2309     MemoryRegionSection mrs;
2310     MemoryListener *listener;
2311     AddressSpace *as;
2312     FlatView *view;
2313     FlatRange *fr;
2314     hwaddr sec_start, sec_end, sec_size;
2315 
2316     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2317         if (!listener->log_clear) {
2318             continue;
2319         }
2320         as = listener->address_space;
2321         view = address_space_get_flatview(as);
2322         FOR_EACH_FLAT_RANGE(fr, view) {
2323             if (!fr->dirty_log_mask || fr->mr != mr) {
2324                 /*
2325                  * Clear dirty bitmap operation only applies to those
2326                  * regions whose dirty logging is at least enabled
2327                  */
2328                 continue;
2329             }
2330 
2331             mrs = section_from_flat_range(fr, view);
2332 
2333             sec_start = MAX(mrs.offset_within_region, start);
2334             sec_end = mrs.offset_within_region + int128_get64(mrs.size);
2335             sec_end = MIN(sec_end, start + len);
2336 
2337             if (sec_start >= sec_end) {
2338                 /*
2339                  * If this memory region section has no intersection
2340                  * with the requested range, skip.
2341                  */
2342                 continue;
2343             }
2344 
2345             /* Valid case; shrink the section if needed */
2346             mrs.offset_within_address_space +=
2347                 sec_start - mrs.offset_within_region;
2348             mrs.offset_within_region = sec_start;
2349             sec_size = sec_end - sec_start;
2350             mrs.size = int128_make64(sec_size);
2351             listener->log_clear(listener, &mrs);
2352         }
2353         flatview_unref(view);
2354     }
2355 }
2356 
2357 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
2358                                                             hwaddr addr,
2359                                                             hwaddr size,
2360                                                             unsigned client)
2361 {
2362     DirtyBitmapSnapshot *snapshot;
2363     assert(mr->ram_block);
2364     memory_region_sync_dirty_bitmap(mr, false);
2365     snapshot = cpu_physical_memory_snapshot_and_clear_dirty(mr, addr, size, client);
2366     memory_global_after_dirty_log_sync();
2367     return snapshot;
2368 }
2369 
2370 bool memory_region_snapshot_get_dirty(MemoryRegion *mr, DirtyBitmapSnapshot *snap,
2371                                       hwaddr addr, hwaddr size)
2372 {
2373     assert(mr->ram_block);
2374     return cpu_physical_memory_snapshot_get_dirty(snap,
2375                 memory_region_get_ram_addr(mr) + addr, size);
2376 }
2377 
2378 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
2379 {
2380     if (mr->readonly != readonly) {
2381         memory_region_transaction_begin();
2382         mr->readonly = readonly;
2383         memory_region_update_pending |= mr->enabled;
2384         memory_region_transaction_commit();
2385     }
2386 }
2387 
2388 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile)
2389 {
2390     if (mr->nonvolatile != nonvolatile) {
2391         memory_region_transaction_begin();
2392         mr->nonvolatile = nonvolatile;
2393         memory_region_update_pending |= mr->enabled;
2394         memory_region_transaction_commit();
2395     }
2396 }
2397 
2398 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
2399 {
2400     if (mr->romd_mode != romd_mode) {
2401         memory_region_transaction_begin();
2402         mr->romd_mode = romd_mode;
2403         memory_region_update_pending |= mr->enabled;
2404         memory_region_transaction_commit();
2405     }
2406 }
2407 
2408 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
2409                                hwaddr size, unsigned client)
2410 {
2411     assert(mr->ram_block);
2412     cpu_physical_memory_test_and_clear_dirty(
2413         memory_region_get_ram_addr(mr) + addr, size, client);
2414 }
2415 
2416 int memory_region_get_fd(MemoryRegion *mr)
2417 {
2418     RCU_READ_LOCK_GUARD();
2419     while (mr->alias) {
2420         mr = mr->alias;
2421     }
2422     return mr->ram_block->fd;
2423 }
2424 
2425 void *memory_region_get_ram_ptr(MemoryRegion *mr)
2426 {
2427     uint64_t offset = 0;
2428 
2429     RCU_READ_LOCK_GUARD();
2430     while (mr->alias) {
2431         offset += mr->alias_offset;
2432         mr = mr->alias;
2433     }
2434     assert(mr->ram_block);
2435     return qemu_map_ram_ptr(mr->ram_block, offset);
2436 }
2437 
2438 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset)
2439 {
2440     RAMBlock *block;
2441 
2442     block = qemu_ram_block_from_host(ptr, false, offset);
2443     if (!block) {
2444         return NULL;
2445     }
2446 
2447     return block->mr;
2448 }
2449 
2450 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr)
2451 {
2452     return mr->ram_block ? mr->ram_block->offset : RAM_ADDR_INVALID;
2453 }
2454 
2455 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
2456 {
2457     assert(mr->ram_block);
2458 
2459     qemu_ram_resize(mr->ram_block, newsize, errp);
2460 }
2461 
2462 void memory_region_msync(MemoryRegion *mr, hwaddr addr, hwaddr size)
2463 {
2464     if (mr->ram_block) {
2465         qemu_ram_msync(mr->ram_block, addr, size);
2466     }
2467 }
2468 
2469 void memory_region_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size)
2470 {
2471     /*
2472      * Might be extended case needed to cover
2473      * different types of memory regions
2474      */
2475     if (mr->dirty_log_mask) {
2476         memory_region_msync(mr, addr, size);
2477     }
2478 }
2479 
2480 /*
2481  * Call proper memory listeners about the change on the newly
2482  * added/removed CoalescedMemoryRange.
2483  */
2484 static void memory_region_update_coalesced_range(MemoryRegion *mr,
2485                                                  CoalescedMemoryRange *cmr,
2486                                                  bool add)
2487 {
2488     AddressSpace *as;
2489     FlatView *view;
2490     FlatRange *fr;
2491 
2492     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2493         view = address_space_get_flatview(as);
2494         FOR_EACH_FLAT_RANGE(fr, view) {
2495             if (fr->mr == mr) {
2496                 flat_range_coalesced_io_notify(fr, as, cmr, add);
2497             }
2498         }
2499         flatview_unref(view);
2500     }
2501 }
2502 
2503 void memory_region_set_coalescing(MemoryRegion *mr)
2504 {
2505     memory_region_clear_coalescing(mr);
2506     memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
2507 }
2508 
2509 void memory_region_add_coalescing(MemoryRegion *mr,
2510                                   hwaddr offset,
2511                                   uint64_t size)
2512 {
2513     CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
2514 
2515     cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
2516     QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
2517     memory_region_update_coalesced_range(mr, cmr, true);
2518     memory_region_set_flush_coalesced(mr);
2519 }
2520 
2521 void memory_region_clear_coalescing(MemoryRegion *mr)
2522 {
2523     CoalescedMemoryRange *cmr;
2524 
2525     if (QTAILQ_EMPTY(&mr->coalesced)) {
2526         return;
2527     }
2528 
2529     qemu_flush_coalesced_mmio_buffer();
2530     mr->flush_coalesced_mmio = false;
2531 
2532     while (!QTAILQ_EMPTY(&mr->coalesced)) {
2533         cmr = QTAILQ_FIRST(&mr->coalesced);
2534         QTAILQ_REMOVE(&mr->coalesced, cmr, link);
2535         memory_region_update_coalesced_range(mr, cmr, false);
2536         g_free(cmr);
2537     }
2538 }
2539 
2540 void memory_region_set_flush_coalesced(MemoryRegion *mr)
2541 {
2542     mr->flush_coalesced_mmio = true;
2543 }
2544 
2545 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
2546 {
2547     qemu_flush_coalesced_mmio_buffer();
2548     if (QTAILQ_EMPTY(&mr->coalesced)) {
2549         mr->flush_coalesced_mmio = false;
2550     }
2551 }
2552 
2553 static bool userspace_eventfd_warning;
2554 
2555 void memory_region_add_eventfd(MemoryRegion *mr,
2556                                hwaddr addr,
2557                                unsigned size,
2558                                bool match_data,
2559                                uint64_t data,
2560                                EventNotifier *e)
2561 {
2562     MemoryRegionIoeventfd mrfd = {
2563         .addr.start = int128_make64(addr),
2564         .addr.size = int128_make64(size),
2565         .match_data = match_data,
2566         .data = data,
2567         .e = e,
2568     };
2569     unsigned i;
2570 
2571     if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
2572                             userspace_eventfd_warning))) {
2573         userspace_eventfd_warning = true;
2574         error_report("Using eventfd without MMIO binding in KVM. "
2575                      "Suboptimal performance expected");
2576     }
2577 
2578     if (size) {
2579         adjust_endianness(mr, &mrfd.data, size_memop(size) | MO_TE);
2580     }
2581     memory_region_transaction_begin();
2582     for (i = 0; i < mr->ioeventfd_nb; ++i) {
2583         if (memory_region_ioeventfd_before(&mrfd, &mr->ioeventfds[i])) {
2584             break;
2585         }
2586     }
2587     ++mr->ioeventfd_nb;
2588     mr->ioeventfds = g_realloc(mr->ioeventfds,
2589                                   sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
2590     memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
2591             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
2592     mr->ioeventfds[i] = mrfd;
2593     ioeventfd_update_pending |= mr->enabled;
2594     memory_region_transaction_commit();
2595 }
2596 
2597 void memory_region_del_eventfd(MemoryRegion *mr,
2598                                hwaddr addr,
2599                                unsigned size,
2600                                bool match_data,
2601                                uint64_t data,
2602                                EventNotifier *e)
2603 {
2604     MemoryRegionIoeventfd mrfd = {
2605         .addr.start = int128_make64(addr),
2606         .addr.size = int128_make64(size),
2607         .match_data = match_data,
2608         .data = data,
2609         .e = e,
2610     };
2611     unsigned i;
2612 
2613     if (size) {
2614         adjust_endianness(mr, &mrfd.data, size_memop(size) | MO_TE);
2615     }
2616     memory_region_transaction_begin();
2617     for (i = 0; i < mr->ioeventfd_nb; ++i) {
2618         if (memory_region_ioeventfd_equal(&mrfd, &mr->ioeventfds[i])) {
2619             break;
2620         }
2621     }
2622     assert(i != mr->ioeventfd_nb);
2623     memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
2624             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
2625     --mr->ioeventfd_nb;
2626     mr->ioeventfds = g_realloc(mr->ioeventfds,
2627                                   sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
2628     ioeventfd_update_pending |= mr->enabled;
2629     memory_region_transaction_commit();
2630 }
2631 
2632 static void memory_region_update_container_subregions(MemoryRegion *subregion)
2633 {
2634     MemoryRegion *mr = subregion->container;
2635     MemoryRegion *other;
2636 
2637     memory_region_transaction_begin();
2638 
2639     memory_region_ref(subregion);
2640     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
2641         if (subregion->priority >= other->priority) {
2642             QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
2643             goto done;
2644         }
2645     }
2646     QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
2647 done:
2648     memory_region_update_pending |= mr->enabled && subregion->enabled;
2649     memory_region_transaction_commit();
2650 }
2651 
2652 static void memory_region_add_subregion_common(MemoryRegion *mr,
2653                                                hwaddr offset,
2654                                                MemoryRegion *subregion)
2655 {
2656     MemoryRegion *alias;
2657 
2658     assert(!subregion->container);
2659     subregion->container = mr;
2660     for (alias = subregion->alias; alias; alias = alias->alias) {
2661         alias->mapped_via_alias++;
2662     }
2663     subregion->addr = offset;
2664     memory_region_update_container_subregions(subregion);
2665 }
2666 
2667 void memory_region_add_subregion(MemoryRegion *mr,
2668                                  hwaddr offset,
2669                                  MemoryRegion *subregion)
2670 {
2671     subregion->priority = 0;
2672     memory_region_add_subregion_common(mr, offset, subregion);
2673 }
2674 
2675 void memory_region_add_subregion_overlap(MemoryRegion *mr,
2676                                          hwaddr offset,
2677                                          MemoryRegion *subregion,
2678                                          int priority)
2679 {
2680     subregion->priority = priority;
2681     memory_region_add_subregion_common(mr, offset, subregion);
2682 }
2683 
2684 void memory_region_del_subregion(MemoryRegion *mr,
2685                                  MemoryRegion *subregion)
2686 {
2687     MemoryRegion *alias;
2688 
2689     memory_region_transaction_begin();
2690     assert(subregion->container == mr);
2691     subregion->container = NULL;
2692     for (alias = subregion->alias; alias; alias = alias->alias) {
2693         alias->mapped_via_alias--;
2694         assert(alias->mapped_via_alias >= 0);
2695     }
2696     QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
2697     memory_region_unref(subregion);
2698     memory_region_update_pending |= mr->enabled && subregion->enabled;
2699     memory_region_transaction_commit();
2700 }
2701 
2702 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
2703 {
2704     if (enabled == mr->enabled) {
2705         return;
2706     }
2707     memory_region_transaction_begin();
2708     mr->enabled = enabled;
2709     memory_region_update_pending = true;
2710     memory_region_transaction_commit();
2711 }
2712 
2713 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
2714 {
2715     Int128 s = int128_make64(size);
2716 
2717     if (size == UINT64_MAX) {
2718         s = int128_2_64();
2719     }
2720     if (int128_eq(s, mr->size)) {
2721         return;
2722     }
2723     memory_region_transaction_begin();
2724     mr->size = s;
2725     memory_region_update_pending = true;
2726     memory_region_transaction_commit();
2727 }
2728 
2729 static void memory_region_readd_subregion(MemoryRegion *mr)
2730 {
2731     MemoryRegion *container = mr->container;
2732 
2733     if (container) {
2734         memory_region_transaction_begin();
2735         memory_region_ref(mr);
2736         memory_region_del_subregion(container, mr);
2737         memory_region_add_subregion_common(container, mr->addr, mr);
2738         memory_region_unref(mr);
2739         memory_region_transaction_commit();
2740     }
2741 }
2742 
2743 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
2744 {
2745     if (addr != mr->addr) {
2746         mr->addr = addr;
2747         memory_region_readd_subregion(mr);
2748     }
2749 }
2750 
2751 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
2752 {
2753     assert(mr->alias);
2754 
2755     if (offset == mr->alias_offset) {
2756         return;
2757     }
2758 
2759     memory_region_transaction_begin();
2760     mr->alias_offset = offset;
2761     memory_region_update_pending |= mr->enabled;
2762     memory_region_transaction_commit();
2763 }
2764 
2765 void memory_region_set_unmergeable(MemoryRegion *mr, bool unmergeable)
2766 {
2767     if (unmergeable == mr->unmergeable) {
2768         return;
2769     }
2770 
2771     memory_region_transaction_begin();
2772     mr->unmergeable = unmergeable;
2773     memory_region_update_pending |= mr->enabled;
2774     memory_region_transaction_commit();
2775 }
2776 
2777 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
2778 {
2779     return mr->align;
2780 }
2781 
2782 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
2783 {
2784     const AddrRange *addr = addr_;
2785     const FlatRange *fr = fr_;
2786 
2787     if (int128_le(addrrange_end(*addr), fr->addr.start)) {
2788         return -1;
2789     } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
2790         return 1;
2791     }
2792     return 0;
2793 }
2794 
2795 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
2796 {
2797     return bsearch(&addr, view->ranges, view->nr,
2798                    sizeof(FlatRange), cmp_flatrange_addr);
2799 }
2800 
2801 bool memory_region_is_mapped(MemoryRegion *mr)
2802 {
2803     return !!mr->container || mr->mapped_via_alias;
2804 }
2805 
2806 /* Same as memory_region_find, but it does not add a reference to the
2807  * returned region.  It must be called from an RCU critical section.
2808  */
2809 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2810                                                   hwaddr addr, uint64_t size)
2811 {
2812     MemoryRegionSection ret = { .mr = NULL };
2813     MemoryRegion *root;
2814     AddressSpace *as;
2815     AddrRange range;
2816     FlatView *view;
2817     FlatRange *fr;
2818 
2819     addr += mr->addr;
2820     for (root = mr; root->container; ) {
2821         root = root->container;
2822         addr += root->addr;
2823     }
2824 
2825     as = memory_region_to_address_space(root);
2826     if (!as) {
2827         return ret;
2828     }
2829     range = addrrange_make(int128_make64(addr), int128_make64(size));
2830 
2831     view = address_space_to_flatview(as);
2832     fr = flatview_lookup(view, range);
2833     if (!fr) {
2834         return ret;
2835     }
2836 
2837     while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2838         --fr;
2839     }
2840 
2841     ret.mr = fr->mr;
2842     ret.fv = view;
2843     range = addrrange_intersection(range, fr->addr);
2844     ret.offset_within_region = fr->offset_in_region;
2845     ret.offset_within_region += int128_get64(int128_sub(range.start,
2846                                                         fr->addr.start));
2847     ret.size = range.size;
2848     ret.offset_within_address_space = int128_get64(range.start);
2849     ret.readonly = fr->readonly;
2850     ret.nonvolatile = fr->nonvolatile;
2851     return ret;
2852 }
2853 
2854 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2855                                        hwaddr addr, uint64_t size)
2856 {
2857     MemoryRegionSection ret;
2858     RCU_READ_LOCK_GUARD();
2859     ret = memory_region_find_rcu(mr, addr, size);
2860     if (ret.mr) {
2861         memory_region_ref(ret.mr);
2862     }
2863     return ret;
2864 }
2865 
2866 MemoryRegionSection *memory_region_section_new_copy(MemoryRegionSection *s)
2867 {
2868     MemoryRegionSection *tmp = g_new(MemoryRegionSection, 1);
2869 
2870     *tmp = *s;
2871     if (tmp->mr) {
2872         memory_region_ref(tmp->mr);
2873     }
2874     if (tmp->fv) {
2875         bool ret  = flatview_ref(tmp->fv);
2876 
2877         g_assert(ret);
2878     }
2879     return tmp;
2880 }
2881 
2882 void memory_region_section_free_copy(MemoryRegionSection *s)
2883 {
2884     if (s->fv) {
2885         flatview_unref(s->fv);
2886     }
2887     if (s->mr) {
2888         memory_region_unref(s->mr);
2889     }
2890     g_free(s);
2891 }
2892 
2893 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2894 {
2895     MemoryRegion *mr;
2896 
2897     RCU_READ_LOCK_GUARD();
2898     mr = memory_region_find_rcu(container, addr, 1).mr;
2899     return mr && mr != container;
2900 }
2901 
2902 void memory_global_dirty_log_sync(bool last_stage)
2903 {
2904     memory_region_sync_dirty_bitmap(NULL, last_stage);
2905 }
2906 
2907 void memory_global_after_dirty_log_sync(void)
2908 {
2909     MEMORY_LISTENER_CALL_GLOBAL(log_global_after_sync, Forward);
2910 }
2911 
2912 /*
2913  * Dirty track stop flags that are postponed due to VM being stopped.  Should
2914  * only be used within vmstate_change hook.
2915  */
2916 static unsigned int postponed_stop_flags;
2917 static VMChangeStateEntry *vmstate_change;
2918 static void memory_global_dirty_log_stop_postponed_run(void);
2919 
2920 void memory_global_dirty_log_start(unsigned int flags)
2921 {
2922     unsigned int old_flags;
2923 
2924     assert(flags && !(flags & (~GLOBAL_DIRTY_MASK)));
2925 
2926     if (vmstate_change) {
2927         /* If there is postponed stop(), operate on it first */
2928         postponed_stop_flags &= ~flags;
2929         memory_global_dirty_log_stop_postponed_run();
2930     }
2931 
2932     flags &= ~global_dirty_tracking;
2933     if (!flags) {
2934         return;
2935     }
2936 
2937     old_flags = global_dirty_tracking;
2938     global_dirty_tracking |= flags;
2939     trace_global_dirty_changed(global_dirty_tracking);
2940 
2941     if (!old_flags) {
2942         MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2943         memory_region_transaction_begin();
2944         memory_region_update_pending = true;
2945         memory_region_transaction_commit();
2946     }
2947 }
2948 
2949 static void memory_global_dirty_log_do_stop(unsigned int flags)
2950 {
2951     assert(flags && !(flags & (~GLOBAL_DIRTY_MASK)));
2952     assert((global_dirty_tracking & flags) == flags);
2953     global_dirty_tracking &= ~flags;
2954 
2955     trace_global_dirty_changed(global_dirty_tracking);
2956 
2957     if (!global_dirty_tracking) {
2958         memory_region_transaction_begin();
2959         memory_region_update_pending = true;
2960         memory_region_transaction_commit();
2961         MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2962     }
2963 }
2964 
2965 /*
2966  * Execute the postponed dirty log stop operations if there is, then reset
2967  * everything (including the flags and the vmstate change hook).
2968  */
2969 static void memory_global_dirty_log_stop_postponed_run(void)
2970 {
2971     /* This must be called with the vmstate handler registered */
2972     assert(vmstate_change);
2973 
2974     /* Note: postponed_stop_flags can be cleared in log start routine */
2975     if (postponed_stop_flags) {
2976         memory_global_dirty_log_do_stop(postponed_stop_flags);
2977         postponed_stop_flags = 0;
2978     }
2979 
2980     qemu_del_vm_change_state_handler(vmstate_change);
2981     vmstate_change = NULL;
2982 }
2983 
2984 static void memory_vm_change_state_handler(void *opaque, bool running,
2985                                            RunState state)
2986 {
2987     if (running) {
2988         memory_global_dirty_log_stop_postponed_run();
2989     }
2990 }
2991 
2992 void memory_global_dirty_log_stop(unsigned int flags)
2993 {
2994     if (!runstate_is_running()) {
2995         /* Postpone the dirty log stop, e.g., to when VM starts again */
2996         if (vmstate_change) {
2997             /* Batch with previous postponed flags */
2998             postponed_stop_flags |= flags;
2999         } else {
3000             postponed_stop_flags = flags;
3001             vmstate_change = qemu_add_vm_change_state_handler(
3002                 memory_vm_change_state_handler, NULL);
3003         }
3004         return;
3005     }
3006 
3007     memory_global_dirty_log_do_stop(flags);
3008 }
3009 
3010 static void listener_add_address_space(MemoryListener *listener,
3011                                        AddressSpace *as)
3012 {
3013     FlatView *view;
3014     FlatRange *fr;
3015 
3016     if (listener->begin) {
3017         listener->begin(listener);
3018     }
3019     if (global_dirty_tracking) {
3020         if (listener->log_global_start) {
3021             listener->log_global_start(listener);
3022         }
3023     }
3024 
3025     view = address_space_get_flatview(as);
3026     FOR_EACH_FLAT_RANGE(fr, view) {
3027         MemoryRegionSection section = section_from_flat_range(fr, view);
3028 
3029         if (listener->region_add) {
3030             listener->region_add(listener, &section);
3031         }
3032         if (fr->dirty_log_mask && listener->log_start) {
3033             listener->log_start(listener, &section, 0, fr->dirty_log_mask);
3034         }
3035     }
3036     if (listener->commit) {
3037         listener->commit(listener);
3038     }
3039     flatview_unref(view);
3040 }
3041 
3042 static void listener_del_address_space(MemoryListener *listener,
3043                                        AddressSpace *as)
3044 {
3045     FlatView *view;
3046     FlatRange *fr;
3047 
3048     if (listener->begin) {
3049         listener->begin(listener);
3050     }
3051     view = address_space_get_flatview(as);
3052     FOR_EACH_FLAT_RANGE(fr, view) {
3053         MemoryRegionSection section = section_from_flat_range(fr, view);
3054 
3055         if (fr->dirty_log_mask && listener->log_stop) {
3056             listener->log_stop(listener, &section, fr->dirty_log_mask, 0);
3057         }
3058         if (listener->region_del) {
3059             listener->region_del(listener, &section);
3060         }
3061     }
3062     if (listener->commit) {
3063         listener->commit(listener);
3064     }
3065     flatview_unref(view);
3066 }
3067 
3068 void memory_listener_register(MemoryListener *listener, AddressSpace *as)
3069 {
3070     MemoryListener *other = NULL;
3071 
3072     /* Only one of them can be defined for a listener */
3073     assert(!(listener->log_sync && listener->log_sync_global));
3074 
3075     listener->address_space = as;
3076     if (QTAILQ_EMPTY(&memory_listeners)
3077         || listener->priority >= QTAILQ_LAST(&memory_listeners)->priority) {
3078         QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
3079     } else {
3080         QTAILQ_FOREACH(other, &memory_listeners, link) {
3081             if (listener->priority < other->priority) {
3082                 break;
3083             }
3084         }
3085         QTAILQ_INSERT_BEFORE(other, listener, link);
3086     }
3087 
3088     if (QTAILQ_EMPTY(&as->listeners)
3089         || listener->priority >= QTAILQ_LAST(&as->listeners)->priority) {
3090         QTAILQ_INSERT_TAIL(&as->listeners, listener, link_as);
3091     } else {
3092         QTAILQ_FOREACH(other, &as->listeners, link_as) {
3093             if (listener->priority < other->priority) {
3094                 break;
3095             }
3096         }
3097         QTAILQ_INSERT_BEFORE(other, listener, link_as);
3098     }
3099 
3100     listener_add_address_space(listener, as);
3101 
3102     if (listener->eventfd_add || listener->eventfd_del) {
3103         as->ioeventfd_notifiers++;
3104     }
3105 }
3106 
3107 void memory_listener_unregister(MemoryListener *listener)
3108 {
3109     if (!listener->address_space) {
3110         return;
3111     }
3112 
3113     if (listener->eventfd_add || listener->eventfd_del) {
3114         listener->address_space->ioeventfd_notifiers--;
3115     }
3116 
3117     listener_del_address_space(listener, listener->address_space);
3118     QTAILQ_REMOVE(&memory_listeners, listener, link);
3119     QTAILQ_REMOVE(&listener->address_space->listeners, listener, link_as);
3120     listener->address_space = NULL;
3121 }
3122 
3123 void address_space_remove_listeners(AddressSpace *as)
3124 {
3125     while (!QTAILQ_EMPTY(&as->listeners)) {
3126         memory_listener_unregister(QTAILQ_FIRST(&as->listeners));
3127     }
3128 }
3129 
3130 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
3131 {
3132     memory_region_ref(root);
3133     as->root = root;
3134     as->current_map = NULL;
3135     as->ioeventfd_nb = 0;
3136     as->ioeventfds = NULL;
3137     QTAILQ_INIT(&as->listeners);
3138     QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
3139     as->name = g_strdup(name ? name : "anonymous");
3140     address_space_update_topology(as);
3141     address_space_update_ioeventfds(as);
3142 }
3143 
3144 static void do_address_space_destroy(AddressSpace *as)
3145 {
3146     assert(QTAILQ_EMPTY(&as->listeners));
3147 
3148     flatview_unref(as->current_map);
3149     g_free(as->name);
3150     g_free(as->ioeventfds);
3151     memory_region_unref(as->root);
3152 }
3153 
3154 void address_space_destroy(AddressSpace *as)
3155 {
3156     MemoryRegion *root = as->root;
3157 
3158     /* Flush out anything from MemoryListeners listening in on this */
3159     memory_region_transaction_begin();
3160     as->root = NULL;
3161     memory_region_transaction_commit();
3162     QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
3163 
3164     /* At this point, as->dispatch and as->current_map are dummy
3165      * entries that the guest should never use.  Wait for the old
3166      * values to expire before freeing the data.
3167      */
3168     as->root = root;
3169     call_rcu(as, do_address_space_destroy, rcu);
3170 }
3171 
3172 static const char *memory_region_type(MemoryRegion *mr)
3173 {
3174     if (mr->alias) {
3175         return memory_region_type(mr->alias);
3176     }
3177     if (memory_region_is_ram_device(mr)) {
3178         return "ramd";
3179     } else if (memory_region_is_romd(mr)) {
3180         return "romd";
3181     } else if (memory_region_is_rom(mr)) {
3182         return "rom";
3183     } else if (memory_region_is_ram(mr)) {
3184         return "ram";
3185     } else {
3186         return "i/o";
3187     }
3188 }
3189 
3190 typedef struct MemoryRegionList MemoryRegionList;
3191 
3192 struct MemoryRegionList {
3193     const MemoryRegion *mr;
3194     QTAILQ_ENTRY(MemoryRegionList) mrqueue;
3195 };
3196 
3197 typedef QTAILQ_HEAD(, MemoryRegionList) MemoryRegionListHead;
3198 
3199 #define MR_SIZE(size) (int128_nz(size) ? (hwaddr)int128_get64( \
3200                            int128_sub((size), int128_one())) : 0)
3201 #define MTREE_INDENT "  "
3202 
3203 static void mtree_expand_owner(const char *label, Object *obj)
3204 {
3205     DeviceState *dev = (DeviceState *) object_dynamic_cast(obj, TYPE_DEVICE);
3206 
3207     qemu_printf(" %s:{%s", label, dev ? "dev" : "obj");
3208     if (dev && dev->id) {
3209         qemu_printf(" id=%s", dev->id);
3210     } else {
3211         char *canonical_path = object_get_canonical_path(obj);
3212         if (canonical_path) {
3213             qemu_printf(" path=%s", canonical_path);
3214             g_free(canonical_path);
3215         } else {
3216             qemu_printf(" type=%s", object_get_typename(obj));
3217         }
3218     }
3219     qemu_printf("}");
3220 }
3221 
3222 static void mtree_print_mr_owner(const MemoryRegion *mr)
3223 {
3224     Object *owner = mr->owner;
3225     Object *parent = memory_region_owner((MemoryRegion *)mr);
3226 
3227     if (!owner && !parent) {
3228         qemu_printf(" orphan");
3229         return;
3230     }
3231     if (owner) {
3232         mtree_expand_owner("owner", owner);
3233     }
3234     if (parent && parent != owner) {
3235         mtree_expand_owner("parent", parent);
3236     }
3237 }
3238 
3239 static void mtree_print_mr(const MemoryRegion *mr, unsigned int level,
3240                            hwaddr base,
3241                            MemoryRegionListHead *alias_print_queue,
3242                            bool owner, bool display_disabled)
3243 {
3244     MemoryRegionList *new_ml, *ml, *next_ml;
3245     MemoryRegionListHead submr_print_queue;
3246     const MemoryRegion *submr;
3247     unsigned int i;
3248     hwaddr cur_start, cur_end;
3249 
3250     if (!mr) {
3251         return;
3252     }
3253 
3254     cur_start = base + mr->addr;
3255     cur_end = cur_start + MR_SIZE(mr->size);
3256 
3257     /*
3258      * Try to detect overflow of memory region. This should never
3259      * happen normally. When it happens, we dump something to warn the
3260      * user who is observing this.
3261      */
3262     if (cur_start < base || cur_end < cur_start) {
3263         qemu_printf("[DETECTED OVERFLOW!] ");
3264     }
3265 
3266     if (mr->alias) {
3267         bool found = false;
3268 
3269         /* check if the alias is already in the queue */
3270         QTAILQ_FOREACH(ml, alias_print_queue, mrqueue) {
3271             if (ml->mr == mr->alias) {
3272                 found = true;
3273             }
3274         }
3275 
3276         if (!found) {
3277             ml = g_new(MemoryRegionList, 1);
3278             ml->mr = mr->alias;
3279             QTAILQ_INSERT_TAIL(alias_print_queue, ml, mrqueue);
3280         }
3281         if (mr->enabled || display_disabled) {
3282             for (i = 0; i < level; i++) {
3283                 qemu_printf(MTREE_INDENT);
3284             }
3285             qemu_printf(HWADDR_FMT_plx "-" HWADDR_FMT_plx
3286                         " (prio %d, %s%s): alias %s @%s " HWADDR_FMT_plx
3287                         "-" HWADDR_FMT_plx "%s",
3288                         cur_start, cur_end,
3289                         mr->priority,
3290                         mr->nonvolatile ? "nv-" : "",
3291                         memory_region_type((MemoryRegion *)mr),
3292                         memory_region_name(mr),
3293                         memory_region_name(mr->alias),
3294                         mr->alias_offset,
3295                         mr->alias_offset + MR_SIZE(mr->size),
3296                         mr->enabled ? "" : " [disabled]");
3297             if (owner) {
3298                 mtree_print_mr_owner(mr);
3299             }
3300             qemu_printf("\n");
3301         }
3302     } else {
3303         if (mr->enabled || display_disabled) {
3304             for (i = 0; i < level; i++) {
3305                 qemu_printf(MTREE_INDENT);
3306             }
3307             qemu_printf(HWADDR_FMT_plx "-" HWADDR_FMT_plx
3308                         " (prio %d, %s%s): %s%s",
3309                         cur_start, cur_end,
3310                         mr->priority,
3311                         mr->nonvolatile ? "nv-" : "",
3312                         memory_region_type((MemoryRegion *)mr),
3313                         memory_region_name(mr),
3314                         mr->enabled ? "" : " [disabled]");
3315             if (owner) {
3316                 mtree_print_mr_owner(mr);
3317             }
3318             qemu_printf("\n");
3319         }
3320     }
3321 
3322     QTAILQ_INIT(&submr_print_queue);
3323 
3324     QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
3325         new_ml = g_new(MemoryRegionList, 1);
3326         new_ml->mr = submr;
3327         QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
3328             if (new_ml->mr->addr < ml->mr->addr ||
3329                 (new_ml->mr->addr == ml->mr->addr &&
3330                  new_ml->mr->priority > ml->mr->priority)) {
3331                 QTAILQ_INSERT_BEFORE(ml, new_ml, mrqueue);
3332                 new_ml = NULL;
3333                 break;
3334             }
3335         }
3336         if (new_ml) {
3337             QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, mrqueue);
3338         }
3339     }
3340 
3341     QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
3342         mtree_print_mr(ml->mr, level + 1, cur_start,
3343                        alias_print_queue, owner, display_disabled);
3344     }
3345 
3346     QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, mrqueue, next_ml) {
3347         g_free(ml);
3348     }
3349 }
3350 
3351 struct FlatViewInfo {
3352     int counter;
3353     bool dispatch_tree;
3354     bool owner;
3355     AccelClass *ac;
3356 };
3357 
3358 static void mtree_print_flatview(gpointer key, gpointer value,
3359                                  gpointer user_data)
3360 {
3361     FlatView *view = key;
3362     GArray *fv_address_spaces = value;
3363     struct FlatViewInfo *fvi = user_data;
3364     FlatRange *range = &view->ranges[0];
3365     MemoryRegion *mr;
3366     int n = view->nr;
3367     int i;
3368     AddressSpace *as;
3369 
3370     qemu_printf("FlatView #%d\n", fvi->counter);
3371     ++fvi->counter;
3372 
3373     for (i = 0; i < fv_address_spaces->len; ++i) {
3374         as = g_array_index(fv_address_spaces, AddressSpace*, i);
3375         qemu_printf(" AS \"%s\", root: %s",
3376                     as->name, memory_region_name(as->root));
3377         if (as->root->alias) {
3378             qemu_printf(", alias %s", memory_region_name(as->root->alias));
3379         }
3380         qemu_printf("\n");
3381     }
3382 
3383     qemu_printf(" Root memory region: %s\n",
3384       view->root ? memory_region_name(view->root) : "(none)");
3385 
3386     if (n <= 0) {
3387         qemu_printf(MTREE_INDENT "No rendered FlatView\n\n");
3388         return;
3389     }
3390 
3391     while (n--) {
3392         mr = range->mr;
3393         if (range->offset_in_region) {
3394             qemu_printf(MTREE_INDENT HWADDR_FMT_plx "-" HWADDR_FMT_plx
3395                         " (prio %d, %s%s): %s @" HWADDR_FMT_plx,
3396                         int128_get64(range->addr.start),
3397                         int128_get64(range->addr.start)
3398                         + MR_SIZE(range->addr.size),
3399                         mr->priority,
3400                         range->nonvolatile ? "nv-" : "",
3401                         range->readonly ? "rom" : memory_region_type(mr),
3402                         memory_region_name(mr),
3403                         range->offset_in_region);
3404         } else {
3405             qemu_printf(MTREE_INDENT HWADDR_FMT_plx "-" HWADDR_FMT_plx
3406                         " (prio %d, %s%s): %s",
3407                         int128_get64(range->addr.start),
3408                         int128_get64(range->addr.start)
3409                         + MR_SIZE(range->addr.size),
3410                         mr->priority,
3411                         range->nonvolatile ? "nv-" : "",
3412                         range->readonly ? "rom" : memory_region_type(mr),
3413                         memory_region_name(mr));
3414         }
3415         if (fvi->owner) {
3416             mtree_print_mr_owner(mr);
3417         }
3418 
3419         if (fvi->ac) {
3420             for (i = 0; i < fv_address_spaces->len; ++i) {
3421                 as = g_array_index(fv_address_spaces, AddressSpace*, i);
3422                 if (fvi->ac->has_memory(current_machine, as,
3423                                         int128_get64(range->addr.start),
3424                                         MR_SIZE(range->addr.size) + 1)) {
3425                     qemu_printf(" %s", fvi->ac->name);
3426                 }
3427             }
3428         }
3429         qemu_printf("\n");
3430         range++;
3431     }
3432 
3433 #if !defined(CONFIG_USER_ONLY)
3434     if (fvi->dispatch_tree && view->root) {
3435         mtree_print_dispatch(view->dispatch, view->root);
3436     }
3437 #endif
3438 
3439     qemu_printf("\n");
3440 }
3441 
3442 static gboolean mtree_info_flatview_free(gpointer key, gpointer value,
3443                                       gpointer user_data)
3444 {
3445     FlatView *view = key;
3446     GArray *fv_address_spaces = value;
3447 
3448     g_array_unref(fv_address_spaces);
3449     flatview_unref(view);
3450 
3451     return true;
3452 }
3453 
3454 static void mtree_info_flatview(bool dispatch_tree, bool owner)
3455 {
3456     struct FlatViewInfo fvi = {
3457         .counter = 0,
3458         .dispatch_tree = dispatch_tree,
3459         .owner = owner,
3460     };
3461     AddressSpace *as;
3462     FlatView *view;
3463     GArray *fv_address_spaces;
3464     GHashTable *views = g_hash_table_new(g_direct_hash, g_direct_equal);
3465     AccelClass *ac = ACCEL_GET_CLASS(current_accel());
3466 
3467     if (ac->has_memory) {
3468         fvi.ac = ac;
3469     }
3470 
3471     /* Gather all FVs in one table */
3472     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3473         view = address_space_get_flatview(as);
3474 
3475         fv_address_spaces = g_hash_table_lookup(views, view);
3476         if (!fv_address_spaces) {
3477             fv_address_spaces = g_array_new(false, false, sizeof(as));
3478             g_hash_table_insert(views, view, fv_address_spaces);
3479         }
3480 
3481         g_array_append_val(fv_address_spaces, as);
3482     }
3483 
3484     /* Print */
3485     g_hash_table_foreach(views, mtree_print_flatview, &fvi);
3486 
3487     /* Free */
3488     g_hash_table_foreach_remove(views, mtree_info_flatview_free, 0);
3489     g_hash_table_unref(views);
3490 }
3491 
3492 struct AddressSpaceInfo {
3493     MemoryRegionListHead *ml_head;
3494     bool owner;
3495     bool disabled;
3496 };
3497 
3498 /* Returns negative value if a < b; zero if a = b; positive value if a > b. */
3499 static gint address_space_compare_name(gconstpointer a, gconstpointer b)
3500 {
3501     const AddressSpace *as_a = a;
3502     const AddressSpace *as_b = b;
3503 
3504     return g_strcmp0(as_a->name, as_b->name);
3505 }
3506 
3507 static void mtree_print_as_name(gpointer data, gpointer user_data)
3508 {
3509     AddressSpace *as = data;
3510 
3511     qemu_printf("address-space: %s\n", as->name);
3512 }
3513 
3514 static void mtree_print_as(gpointer key, gpointer value, gpointer user_data)
3515 {
3516     MemoryRegion *mr = key;
3517     GSList *as_same_root_mr_list = value;
3518     struct AddressSpaceInfo *asi = user_data;
3519 
3520     g_slist_foreach(as_same_root_mr_list, mtree_print_as_name, NULL);
3521     mtree_print_mr(mr, 1, 0, asi->ml_head, asi->owner, asi->disabled);
3522     qemu_printf("\n");
3523 }
3524 
3525 static gboolean mtree_info_as_free(gpointer key, gpointer value,
3526                                    gpointer user_data)
3527 {
3528     GSList *as_same_root_mr_list = value;
3529 
3530     g_slist_free(as_same_root_mr_list);
3531 
3532     return true;
3533 }
3534 
3535 static void mtree_info_as(bool dispatch_tree, bool owner, bool disabled)
3536 {
3537     MemoryRegionListHead ml_head;
3538     MemoryRegionList *ml, *ml2;
3539     AddressSpace *as;
3540     GHashTable *views = g_hash_table_new(g_direct_hash, g_direct_equal);
3541     GSList *as_same_root_mr_list;
3542     struct AddressSpaceInfo asi = {
3543         .ml_head = &ml_head,
3544         .owner = owner,
3545         .disabled = disabled,
3546     };
3547 
3548     QTAILQ_INIT(&ml_head);
3549 
3550     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3551         /* Create hashtable, key=AS root MR, value = list of AS */
3552         as_same_root_mr_list = g_hash_table_lookup(views, as->root);
3553         as_same_root_mr_list = g_slist_insert_sorted(as_same_root_mr_list, as,
3554                                                      address_space_compare_name);
3555         g_hash_table_insert(views, as->root, as_same_root_mr_list);
3556     }
3557 
3558     /* print address spaces */
3559     g_hash_table_foreach(views, mtree_print_as, &asi);
3560     g_hash_table_foreach_remove(views, mtree_info_as_free, 0);
3561     g_hash_table_unref(views);
3562 
3563     /* print aliased regions */
3564     QTAILQ_FOREACH(ml, &ml_head, mrqueue) {
3565         qemu_printf("memory-region: %s\n", memory_region_name(ml->mr));
3566         mtree_print_mr(ml->mr, 1, 0, &ml_head, owner, disabled);
3567         qemu_printf("\n");
3568     }
3569 
3570     QTAILQ_FOREACH_SAFE(ml, &ml_head, mrqueue, ml2) {
3571         g_free(ml);
3572     }
3573 }
3574 
3575 void mtree_info(bool flatview, bool dispatch_tree, bool owner, bool disabled)
3576 {
3577     if (flatview) {
3578         mtree_info_flatview(dispatch_tree, owner);
3579     } else {
3580         mtree_info_as(dispatch_tree, owner, disabled);
3581     }
3582 }
3583 
3584 void memory_region_init_ram(MemoryRegion *mr,
3585                             Object *owner,
3586                             const char *name,
3587                             uint64_t size,
3588                             Error **errp)
3589 {
3590     DeviceState *owner_dev;
3591     Error *err = NULL;
3592 
3593     memory_region_init_ram_nomigrate(mr, owner, name, size, &err);
3594     if (err) {
3595         error_propagate(errp, err);
3596         return;
3597     }
3598     /* This will assert if owner is neither NULL nor a DeviceState.
3599      * We only want the owner here for the purposes of defining a
3600      * unique name for migration. TODO: Ideally we should implement
3601      * a naming scheme for Objects which are not DeviceStates, in
3602      * which case we can relax this restriction.
3603      */
3604     owner_dev = DEVICE(owner);
3605     vmstate_register_ram(mr, owner_dev);
3606 }
3607 
3608 void memory_region_init_rom(MemoryRegion *mr,
3609                             Object *owner,
3610                             const char *name,
3611                             uint64_t size,
3612                             Error **errp)
3613 {
3614     DeviceState *owner_dev;
3615     Error *err = NULL;
3616 
3617     memory_region_init_rom_nomigrate(mr, owner, name, size, &err);
3618     if (err) {
3619         error_propagate(errp, err);
3620         return;
3621     }
3622     /* This will assert if owner is neither NULL nor a DeviceState.
3623      * We only want the owner here for the purposes of defining a
3624      * unique name for migration. TODO: Ideally we should implement
3625      * a naming scheme for Objects which are not DeviceStates, in
3626      * which case we can relax this restriction.
3627      */
3628     owner_dev = DEVICE(owner);
3629     vmstate_register_ram(mr, owner_dev);
3630 }
3631 
3632 void memory_region_init_rom_device(MemoryRegion *mr,
3633                                    Object *owner,
3634                                    const MemoryRegionOps *ops,
3635                                    void *opaque,
3636                                    const char *name,
3637                                    uint64_t size,
3638                                    Error **errp)
3639 {
3640     DeviceState *owner_dev;
3641     Error *err = NULL;
3642 
3643     memory_region_init_rom_device_nomigrate(mr, owner, ops, opaque,
3644                                             name, size, &err);
3645     if (err) {
3646         error_propagate(errp, err);
3647         return;
3648     }
3649     /* This will assert if owner is neither NULL nor a DeviceState.
3650      * We only want the owner here for the purposes of defining a
3651      * unique name for migration. TODO: Ideally we should implement
3652      * a naming scheme for Objects which are not DeviceStates, in
3653      * which case we can relax this restriction.
3654      */
3655     owner_dev = DEVICE(owner);
3656     vmstate_register_ram(mr, owner_dev);
3657 }
3658 
3659 /*
3660  * Support system builds with CONFIG_FUZZ using a weak symbol and a stub for
3661  * the fuzz_dma_read_cb callback
3662  */
3663 #ifdef CONFIG_FUZZ
3664 void __attribute__((weak)) fuzz_dma_read_cb(size_t addr,
3665                       size_t len,
3666                       MemoryRegion *mr)
3667 {
3668 }
3669 #endif
3670 
3671 static const TypeInfo memory_region_info = {
3672     .parent             = TYPE_OBJECT,
3673     .name               = TYPE_MEMORY_REGION,
3674     .class_size         = sizeof(MemoryRegionClass),
3675     .instance_size      = sizeof(MemoryRegion),
3676     .instance_init      = memory_region_initfn,
3677     .instance_finalize  = memory_region_finalize,
3678 };
3679 
3680 static const TypeInfo iommu_memory_region_info = {
3681     .parent             = TYPE_MEMORY_REGION,
3682     .name               = TYPE_IOMMU_MEMORY_REGION,
3683     .class_size         = sizeof(IOMMUMemoryRegionClass),
3684     .instance_size      = sizeof(IOMMUMemoryRegion),
3685     .instance_init      = iommu_memory_region_initfn,
3686     .abstract           = true,
3687 };
3688 
3689 static const TypeInfo ram_discard_manager_info = {
3690     .parent             = TYPE_INTERFACE,
3691     .name               = TYPE_RAM_DISCARD_MANAGER,
3692     .class_size         = sizeof(RamDiscardManagerClass),
3693 };
3694 
3695 static void memory_register_types(void)
3696 {
3697     type_register_static(&memory_region_info);
3698     type_register_static(&iommu_memory_region_info);
3699     type_register_static(&ram_discard_manager_info);
3700 }
3701 
3702 type_init(memory_register_types)
3703