xref: /openbmc/qemu/system/memory.c (revision 806f71ee)
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     /*
1539      * FIXME: it's not clear why under KVM the write would be processed
1540      * directly, instead of going through eventfd.  This probably should
1541      * test "tcg_enabled() || qtest_enabled()", or should just go away.
1542      */
1543     if (!kvm_enabled() &&
1544         memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1545         return MEMTX_OK;
1546     }
1547 
1548     if (mr->ops->write) {
1549         return access_with_adjusted_size(addr, &data, size,
1550                                          mr->ops->impl.min_access_size,
1551                                          mr->ops->impl.max_access_size,
1552                                          memory_region_write_accessor, mr,
1553                                          attrs);
1554     } else {
1555         return
1556             access_with_adjusted_size(addr, &data, size,
1557                                       mr->ops->impl.min_access_size,
1558                                       mr->ops->impl.max_access_size,
1559                                       memory_region_write_with_attrs_accessor,
1560                                       mr, attrs);
1561     }
1562 }
1563 
1564 void memory_region_init_io(MemoryRegion *mr,
1565                            Object *owner,
1566                            const MemoryRegionOps *ops,
1567                            void *opaque,
1568                            const char *name,
1569                            uint64_t size)
1570 {
1571     memory_region_init(mr, owner, name, size);
1572     mr->ops = ops ? ops : &unassigned_mem_ops;
1573     mr->opaque = opaque;
1574     mr->terminates = true;
1575 }
1576 
1577 void memory_region_init_ram_nomigrate(MemoryRegion *mr,
1578                                       Object *owner,
1579                                       const char *name,
1580                                       uint64_t size,
1581                                       Error **errp)
1582 {
1583     memory_region_init_ram_flags_nomigrate(mr, owner, name, size, 0, errp);
1584 }
1585 
1586 void memory_region_init_ram_flags_nomigrate(MemoryRegion *mr,
1587                                             Object *owner,
1588                                             const char *name,
1589                                             uint64_t size,
1590                                             uint32_t ram_flags,
1591                                             Error **errp)
1592 {
1593     Error *err = NULL;
1594     memory_region_init(mr, owner, name, size);
1595     mr->ram = true;
1596     mr->terminates = true;
1597     mr->destructor = memory_region_destructor_ram;
1598     mr->ram_block = qemu_ram_alloc(size, ram_flags, mr, &err);
1599     if (err) {
1600         mr->size = int128_zero();
1601         object_unparent(OBJECT(mr));
1602         error_propagate(errp, err);
1603     }
1604 }
1605 
1606 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1607                                        Object *owner,
1608                                        const char *name,
1609                                        uint64_t size,
1610                                        uint64_t max_size,
1611                                        void (*resized)(const char*,
1612                                                        uint64_t length,
1613                                                        void *host),
1614                                        Error **errp)
1615 {
1616     Error *err = NULL;
1617     memory_region_init(mr, owner, name, size);
1618     mr->ram = true;
1619     mr->terminates = true;
1620     mr->destructor = memory_region_destructor_ram;
1621     mr->ram_block = qemu_ram_alloc_resizeable(size, max_size, resized,
1622                                               mr, &err);
1623     if (err) {
1624         mr->size = int128_zero();
1625         object_unparent(OBJECT(mr));
1626         error_propagate(errp, err);
1627     }
1628 }
1629 
1630 #ifdef CONFIG_POSIX
1631 void memory_region_init_ram_from_file(MemoryRegion *mr,
1632                                       Object *owner,
1633                                       const char *name,
1634                                       uint64_t size,
1635                                       uint64_t align,
1636                                       uint32_t ram_flags,
1637                                       const char *path,
1638                                       ram_addr_t offset,
1639                                       Error **errp)
1640 {
1641     Error *err = NULL;
1642     memory_region_init(mr, owner, name, size);
1643     mr->ram = true;
1644     mr->readonly = !!(ram_flags & RAM_READONLY);
1645     mr->terminates = true;
1646     mr->destructor = memory_region_destructor_ram;
1647     mr->align = align;
1648     mr->ram_block = qemu_ram_alloc_from_file(size, mr, ram_flags, path,
1649                                              offset, &err);
1650     if (err) {
1651         mr->size = int128_zero();
1652         object_unparent(OBJECT(mr));
1653         error_propagate(errp, err);
1654     }
1655 }
1656 
1657 void memory_region_init_ram_from_fd(MemoryRegion *mr,
1658                                     Object *owner,
1659                                     const char *name,
1660                                     uint64_t size,
1661                                     uint32_t ram_flags,
1662                                     int fd,
1663                                     ram_addr_t offset,
1664                                     Error **errp)
1665 {
1666     Error *err = NULL;
1667     memory_region_init(mr, owner, name, size);
1668     mr->ram = true;
1669     mr->readonly = !!(ram_flags & RAM_READONLY);
1670     mr->terminates = true;
1671     mr->destructor = memory_region_destructor_ram;
1672     mr->ram_block = qemu_ram_alloc_from_fd(size, mr, ram_flags, fd, offset,
1673                                            &err);
1674     if (err) {
1675         mr->size = int128_zero();
1676         object_unparent(OBJECT(mr));
1677         error_propagate(errp, err);
1678     }
1679 }
1680 #endif
1681 
1682 void memory_region_init_ram_ptr(MemoryRegion *mr,
1683                                 Object *owner,
1684                                 const char *name,
1685                                 uint64_t size,
1686                                 void *ptr)
1687 {
1688     memory_region_init(mr, owner, name, size);
1689     mr->ram = true;
1690     mr->terminates = true;
1691     mr->destructor = memory_region_destructor_ram;
1692 
1693     /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL.  */
1694     assert(ptr != NULL);
1695     mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1696 }
1697 
1698 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
1699                                        Object *owner,
1700                                        const char *name,
1701                                        uint64_t size,
1702                                        void *ptr)
1703 {
1704     memory_region_init(mr, owner, name, size);
1705     mr->ram = true;
1706     mr->terminates = true;
1707     mr->ram_device = true;
1708     mr->ops = &ram_device_mem_ops;
1709     mr->opaque = mr;
1710     mr->destructor = memory_region_destructor_ram;
1711 
1712     /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL.  */
1713     assert(ptr != NULL);
1714     mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1715 }
1716 
1717 void memory_region_init_alias(MemoryRegion *mr,
1718                               Object *owner,
1719                               const char *name,
1720                               MemoryRegion *orig,
1721                               hwaddr offset,
1722                               uint64_t size)
1723 {
1724     memory_region_init(mr, owner, name, size);
1725     mr->alias = orig;
1726     mr->alias_offset = offset;
1727 }
1728 
1729 void memory_region_init_rom_nomigrate(MemoryRegion *mr,
1730                                       Object *owner,
1731                                       const char *name,
1732                                       uint64_t size,
1733                                       Error **errp)
1734 {
1735     memory_region_init_ram_flags_nomigrate(mr, owner, name, size, 0, errp);
1736     mr->readonly = true;
1737 }
1738 
1739 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1740                                              Object *owner,
1741                                              const MemoryRegionOps *ops,
1742                                              void *opaque,
1743                                              const char *name,
1744                                              uint64_t size,
1745                                              Error **errp)
1746 {
1747     Error *err = NULL;
1748     assert(ops);
1749     memory_region_init(mr, owner, name, size);
1750     mr->ops = ops;
1751     mr->opaque = opaque;
1752     mr->terminates = true;
1753     mr->rom_device = true;
1754     mr->destructor = memory_region_destructor_ram;
1755     mr->ram_block = qemu_ram_alloc(size, 0, mr, &err);
1756     if (err) {
1757         mr->size = int128_zero();
1758         object_unparent(OBJECT(mr));
1759         error_propagate(errp, err);
1760     }
1761 }
1762 
1763 void memory_region_init_iommu(void *_iommu_mr,
1764                               size_t instance_size,
1765                               const char *mrtypename,
1766                               Object *owner,
1767                               const char *name,
1768                               uint64_t size)
1769 {
1770     struct IOMMUMemoryRegion *iommu_mr;
1771     struct MemoryRegion *mr;
1772 
1773     object_initialize(_iommu_mr, instance_size, mrtypename);
1774     mr = MEMORY_REGION(_iommu_mr);
1775     memory_region_do_init(mr, owner, name, size);
1776     iommu_mr = IOMMU_MEMORY_REGION(mr);
1777     mr->terminates = true;  /* then re-forwards */
1778     QLIST_INIT(&iommu_mr->iommu_notify);
1779     iommu_mr->iommu_notify_flags = IOMMU_NOTIFIER_NONE;
1780 }
1781 
1782 static void memory_region_finalize(Object *obj)
1783 {
1784     MemoryRegion *mr = MEMORY_REGION(obj);
1785 
1786     assert(!mr->container);
1787 
1788     /* We know the region is not visible in any address space (it
1789      * does not have a container and cannot be a root either because
1790      * it has no references, so we can blindly clear mr->enabled.
1791      * memory_region_set_enabled instead could trigger a transaction
1792      * and cause an infinite loop.
1793      */
1794     mr->enabled = false;
1795     memory_region_transaction_begin();
1796     while (!QTAILQ_EMPTY(&mr->subregions)) {
1797         MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1798         memory_region_del_subregion(mr, subregion);
1799     }
1800     memory_region_transaction_commit();
1801 
1802     mr->destructor(mr);
1803     memory_region_clear_coalescing(mr);
1804     g_free((char *)mr->name);
1805     g_free(mr->ioeventfds);
1806 }
1807 
1808 Object *memory_region_owner(MemoryRegion *mr)
1809 {
1810     Object *obj = OBJECT(mr);
1811     return obj->parent;
1812 }
1813 
1814 void memory_region_ref(MemoryRegion *mr)
1815 {
1816     /* MMIO callbacks most likely will access data that belongs
1817      * to the owner, hence the need to ref/unref the owner whenever
1818      * the memory region is in use.
1819      *
1820      * The memory region is a child of its owner.  As long as the
1821      * owner doesn't call unparent itself on the memory region,
1822      * ref-ing the owner will also keep the memory region alive.
1823      * Memory regions without an owner are supposed to never go away;
1824      * we do not ref/unref them because it slows down DMA sensibly.
1825      */
1826     if (mr && mr->owner) {
1827         object_ref(mr->owner);
1828     }
1829 }
1830 
1831 void memory_region_unref(MemoryRegion *mr)
1832 {
1833     if (mr && mr->owner) {
1834         object_unref(mr->owner);
1835     }
1836 }
1837 
1838 uint64_t memory_region_size(MemoryRegion *mr)
1839 {
1840     if (int128_eq(mr->size, int128_2_64())) {
1841         return UINT64_MAX;
1842     }
1843     return int128_get64(mr->size);
1844 }
1845 
1846 const char *memory_region_name(const MemoryRegion *mr)
1847 {
1848     if (!mr->name) {
1849         ((MemoryRegion *)mr)->name =
1850             g_strdup(object_get_canonical_path_component(OBJECT(mr)));
1851     }
1852     return mr->name;
1853 }
1854 
1855 bool memory_region_is_ram_device(MemoryRegion *mr)
1856 {
1857     return mr->ram_device;
1858 }
1859 
1860 bool memory_region_is_protected(MemoryRegion *mr)
1861 {
1862     return mr->ram && (mr->ram_block->flags & RAM_PROTECTED);
1863 }
1864 
1865 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1866 {
1867     uint8_t mask = mr->dirty_log_mask;
1868     RAMBlock *rb = mr->ram_block;
1869 
1870     if (global_dirty_tracking && ((rb && qemu_ram_is_migratable(rb)) ||
1871                              memory_region_is_iommu(mr))) {
1872         mask |= (1 << DIRTY_MEMORY_MIGRATION);
1873     }
1874 
1875     if (tcg_enabled() && rb) {
1876         /* TCG only cares about dirty memory logging for RAM, not IOMMU.  */
1877         mask |= (1 << DIRTY_MEMORY_CODE);
1878     }
1879     return mask;
1880 }
1881 
1882 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1883 {
1884     return memory_region_get_dirty_log_mask(mr) & (1 << client);
1885 }
1886 
1887 static int memory_region_update_iommu_notify_flags(IOMMUMemoryRegion *iommu_mr,
1888                                                    Error **errp)
1889 {
1890     IOMMUNotifierFlag flags = IOMMU_NOTIFIER_NONE;
1891     IOMMUNotifier *iommu_notifier;
1892     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1893     int ret = 0;
1894 
1895     IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
1896         flags |= iommu_notifier->notifier_flags;
1897     }
1898 
1899     if (flags != iommu_mr->iommu_notify_flags && imrc->notify_flag_changed) {
1900         ret = imrc->notify_flag_changed(iommu_mr,
1901                                         iommu_mr->iommu_notify_flags,
1902                                         flags, errp);
1903     }
1904 
1905     if (!ret) {
1906         iommu_mr->iommu_notify_flags = flags;
1907     }
1908     return ret;
1909 }
1910 
1911 int memory_region_iommu_set_page_size_mask(IOMMUMemoryRegion *iommu_mr,
1912                                            uint64_t page_size_mask,
1913                                            Error **errp)
1914 {
1915     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1916     int ret = 0;
1917 
1918     if (imrc->iommu_set_page_size_mask) {
1919         ret = imrc->iommu_set_page_size_mask(iommu_mr, page_size_mask, errp);
1920     }
1921     return ret;
1922 }
1923 
1924 int memory_region_register_iommu_notifier(MemoryRegion *mr,
1925                                           IOMMUNotifier *n, Error **errp)
1926 {
1927     IOMMUMemoryRegion *iommu_mr;
1928     int ret;
1929 
1930     if (mr->alias) {
1931         return memory_region_register_iommu_notifier(mr->alias, n, errp);
1932     }
1933 
1934     /* We need to register for at least one bitfield */
1935     iommu_mr = IOMMU_MEMORY_REGION(mr);
1936     assert(n->notifier_flags != IOMMU_NOTIFIER_NONE);
1937     assert(n->start <= n->end);
1938     assert(n->iommu_idx >= 0 &&
1939            n->iommu_idx < memory_region_iommu_num_indexes(iommu_mr));
1940 
1941     QLIST_INSERT_HEAD(&iommu_mr->iommu_notify, n, node);
1942     ret = memory_region_update_iommu_notify_flags(iommu_mr, errp);
1943     if (ret) {
1944         QLIST_REMOVE(n, node);
1945     }
1946     return ret;
1947 }
1948 
1949 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr)
1950 {
1951     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1952 
1953     if (imrc->get_min_page_size) {
1954         return imrc->get_min_page_size(iommu_mr);
1955     }
1956     return TARGET_PAGE_SIZE;
1957 }
1958 
1959 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
1960 {
1961     MemoryRegion *mr = MEMORY_REGION(iommu_mr);
1962     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1963     hwaddr addr, granularity;
1964     IOMMUTLBEntry iotlb;
1965 
1966     /* If the IOMMU has its own replay callback, override */
1967     if (imrc->replay) {
1968         imrc->replay(iommu_mr, n);
1969         return;
1970     }
1971 
1972     granularity = memory_region_iommu_get_min_page_size(iommu_mr);
1973 
1974     for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1975         iotlb = imrc->translate(iommu_mr, addr, IOMMU_NONE, n->iommu_idx);
1976         if (iotlb.perm != IOMMU_NONE) {
1977             n->notify(n, &iotlb);
1978         }
1979 
1980         /* if (2^64 - MR size) < granularity, it's possible to get an
1981          * infinite loop here.  This should catch such a wraparound */
1982         if ((addr + granularity) < addr) {
1983             break;
1984         }
1985     }
1986 }
1987 
1988 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1989                                              IOMMUNotifier *n)
1990 {
1991     IOMMUMemoryRegion *iommu_mr;
1992 
1993     if (mr->alias) {
1994         memory_region_unregister_iommu_notifier(mr->alias, n);
1995         return;
1996     }
1997     QLIST_REMOVE(n, node);
1998     iommu_mr = IOMMU_MEMORY_REGION(mr);
1999     memory_region_update_iommu_notify_flags(iommu_mr, NULL);
2000 }
2001 
2002 void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
2003                                     IOMMUTLBEvent *event)
2004 {
2005     IOMMUTLBEntry *entry = &event->entry;
2006     hwaddr entry_end = entry->iova + entry->addr_mask;
2007     IOMMUTLBEntry tmp = *entry;
2008 
2009     if (event->type == IOMMU_NOTIFIER_UNMAP) {
2010         assert(entry->perm == IOMMU_NONE);
2011     }
2012 
2013     /*
2014      * Skip the notification if the notification does not overlap
2015      * with registered range.
2016      */
2017     if (notifier->start > entry_end || notifier->end < entry->iova) {
2018         return;
2019     }
2020 
2021     if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
2022         /* Crop (iova, addr_mask) to range */
2023         tmp.iova = MAX(tmp.iova, notifier->start);
2024         tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
2025     } else {
2026         assert(entry->iova >= notifier->start && entry_end <= notifier->end);
2027     }
2028 
2029     if (event->type & notifier->notifier_flags) {
2030         notifier->notify(notifier, &tmp);
2031     }
2032 }
2033 
2034 void memory_region_unmap_iommu_notifier_range(IOMMUNotifier *notifier)
2035 {
2036     IOMMUTLBEvent event;
2037 
2038     event.type = IOMMU_NOTIFIER_UNMAP;
2039     event.entry.target_as = &address_space_memory;
2040     event.entry.iova = notifier->start;
2041     event.entry.perm = IOMMU_NONE;
2042     event.entry.addr_mask = notifier->end - notifier->start;
2043 
2044     memory_region_notify_iommu_one(notifier, &event);
2045 }
2046 
2047 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
2048                                 int iommu_idx,
2049                                 IOMMUTLBEvent event)
2050 {
2051     IOMMUNotifier *iommu_notifier;
2052 
2053     assert(memory_region_is_iommu(MEMORY_REGION(iommu_mr)));
2054 
2055     IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
2056         if (iommu_notifier->iommu_idx == iommu_idx) {
2057             memory_region_notify_iommu_one(iommu_notifier, &event);
2058         }
2059     }
2060 }
2061 
2062 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
2063                                  enum IOMMUMemoryRegionAttr attr,
2064                                  void *data)
2065 {
2066     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2067 
2068     if (!imrc->get_attr) {
2069         return -EINVAL;
2070     }
2071 
2072     return imrc->get_attr(iommu_mr, attr, data);
2073 }
2074 
2075 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr,
2076                                        MemTxAttrs attrs)
2077 {
2078     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2079 
2080     if (!imrc->attrs_to_index) {
2081         return 0;
2082     }
2083 
2084     return imrc->attrs_to_index(iommu_mr, attrs);
2085 }
2086 
2087 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr)
2088 {
2089     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2090 
2091     if (!imrc->num_indexes) {
2092         return 1;
2093     }
2094 
2095     return imrc->num_indexes(iommu_mr);
2096 }
2097 
2098 RamDiscardManager *memory_region_get_ram_discard_manager(MemoryRegion *mr)
2099 {
2100     if (!memory_region_is_ram(mr)) {
2101         return NULL;
2102     }
2103     return mr->rdm;
2104 }
2105 
2106 void memory_region_set_ram_discard_manager(MemoryRegion *mr,
2107                                            RamDiscardManager *rdm)
2108 {
2109     g_assert(memory_region_is_ram(mr));
2110     g_assert(!rdm || !mr->rdm);
2111     mr->rdm = rdm;
2112 }
2113 
2114 uint64_t ram_discard_manager_get_min_granularity(const RamDiscardManager *rdm,
2115                                                  const MemoryRegion *mr)
2116 {
2117     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2118 
2119     g_assert(rdmc->get_min_granularity);
2120     return rdmc->get_min_granularity(rdm, mr);
2121 }
2122 
2123 bool ram_discard_manager_is_populated(const RamDiscardManager *rdm,
2124                                       const MemoryRegionSection *section)
2125 {
2126     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2127 
2128     g_assert(rdmc->is_populated);
2129     return rdmc->is_populated(rdm, section);
2130 }
2131 
2132 int ram_discard_manager_replay_populated(const RamDiscardManager *rdm,
2133                                          MemoryRegionSection *section,
2134                                          ReplayRamPopulate replay_fn,
2135                                          void *opaque)
2136 {
2137     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2138 
2139     g_assert(rdmc->replay_populated);
2140     return rdmc->replay_populated(rdm, section, replay_fn, opaque);
2141 }
2142 
2143 void ram_discard_manager_replay_discarded(const RamDiscardManager *rdm,
2144                                           MemoryRegionSection *section,
2145                                           ReplayRamDiscard replay_fn,
2146                                           void *opaque)
2147 {
2148     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2149 
2150     g_assert(rdmc->replay_discarded);
2151     rdmc->replay_discarded(rdm, section, replay_fn, opaque);
2152 }
2153 
2154 void ram_discard_manager_register_listener(RamDiscardManager *rdm,
2155                                            RamDiscardListener *rdl,
2156                                            MemoryRegionSection *section)
2157 {
2158     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2159 
2160     g_assert(rdmc->register_listener);
2161     rdmc->register_listener(rdm, rdl, section);
2162 }
2163 
2164 void ram_discard_manager_unregister_listener(RamDiscardManager *rdm,
2165                                              RamDiscardListener *rdl)
2166 {
2167     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2168 
2169     g_assert(rdmc->unregister_listener);
2170     rdmc->unregister_listener(rdm, rdl);
2171 }
2172 
2173 /* Called with rcu_read_lock held.  */
2174 bool memory_get_xlat_addr(IOMMUTLBEntry *iotlb, void **vaddr,
2175                           ram_addr_t *ram_addr, bool *read_only,
2176                           bool *mr_has_discard_manager)
2177 {
2178     MemoryRegion *mr;
2179     hwaddr xlat;
2180     hwaddr len = iotlb->addr_mask + 1;
2181     bool writable = iotlb->perm & IOMMU_WO;
2182 
2183     if (mr_has_discard_manager) {
2184         *mr_has_discard_manager = false;
2185     }
2186     /*
2187      * The IOMMU TLB entry we have just covers translation through
2188      * this IOMMU to its immediate target.  We need to translate
2189      * it the rest of the way through to memory.
2190      */
2191     mr = address_space_translate(&address_space_memory, iotlb->translated_addr,
2192                                  &xlat, &len, writable, MEMTXATTRS_UNSPECIFIED);
2193     if (!memory_region_is_ram(mr)) {
2194         error_report("iommu map to non memory area %" HWADDR_PRIx "", xlat);
2195         return false;
2196     } else if (memory_region_has_ram_discard_manager(mr)) {
2197         RamDiscardManager *rdm = memory_region_get_ram_discard_manager(mr);
2198         MemoryRegionSection tmp = {
2199             .mr = mr,
2200             .offset_within_region = xlat,
2201             .size = int128_make64(len),
2202         };
2203         if (mr_has_discard_manager) {
2204             *mr_has_discard_manager = true;
2205         }
2206         /*
2207          * Malicious VMs can map memory into the IOMMU, which is expected
2208          * to remain discarded. vfio will pin all pages, populating memory.
2209          * Disallow that. vmstate priorities make sure any RamDiscardManager
2210          * were already restored before IOMMUs are restored.
2211          */
2212         if (!ram_discard_manager_is_populated(rdm, &tmp)) {
2213             error_report("iommu map to discarded memory (e.g., unplugged via"
2214                          " virtio-mem): %" HWADDR_PRIx "",
2215                          iotlb->translated_addr);
2216             return false;
2217         }
2218     }
2219 
2220     /*
2221      * Translation truncates length to the IOMMU page size,
2222      * check that it did not truncate too much.
2223      */
2224     if (len & iotlb->addr_mask) {
2225         error_report("iommu has granularity incompatible with target AS");
2226         return false;
2227     }
2228 
2229     if (vaddr) {
2230         *vaddr = memory_region_get_ram_ptr(mr) + xlat;
2231     }
2232 
2233     if (ram_addr) {
2234         *ram_addr = memory_region_get_ram_addr(mr) + xlat;
2235     }
2236 
2237     if (read_only) {
2238         *read_only = !writable || mr->readonly;
2239     }
2240 
2241     return true;
2242 }
2243 
2244 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
2245 {
2246     uint8_t mask = 1 << client;
2247     uint8_t old_logging;
2248 
2249     assert(client == DIRTY_MEMORY_VGA);
2250     old_logging = mr->vga_logging_count;
2251     mr->vga_logging_count += log ? 1 : -1;
2252     if (!!old_logging == !!mr->vga_logging_count) {
2253         return;
2254     }
2255 
2256     memory_region_transaction_begin();
2257     mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
2258     memory_region_update_pending |= mr->enabled;
2259     memory_region_transaction_commit();
2260 }
2261 
2262 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
2263                              hwaddr size)
2264 {
2265     assert(mr->ram_block);
2266     cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr,
2267                                         size,
2268                                         memory_region_get_dirty_log_mask(mr));
2269 }
2270 
2271 /*
2272  * If memory region `mr' is NULL, do global sync.  Otherwise, sync
2273  * dirty bitmap for the specified memory region.
2274  */
2275 static void memory_region_sync_dirty_bitmap(MemoryRegion *mr, bool last_stage)
2276 {
2277     MemoryListener *listener;
2278     AddressSpace *as;
2279     FlatView *view;
2280     FlatRange *fr;
2281 
2282     /* If the same address space has multiple log_sync listeners, we
2283      * visit that address space's FlatView multiple times.  But because
2284      * log_sync listeners are rare, it's still cheaper than walking each
2285      * address space once.
2286      */
2287     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2288         if (listener->log_sync) {
2289             as = listener->address_space;
2290             view = address_space_get_flatview(as);
2291             FOR_EACH_FLAT_RANGE(fr, view) {
2292                 if (fr->dirty_log_mask && (!mr || fr->mr == mr)) {
2293                     MemoryRegionSection mrs = section_from_flat_range(fr, view);
2294                     listener->log_sync(listener, &mrs);
2295                 }
2296             }
2297             flatview_unref(view);
2298             trace_memory_region_sync_dirty(mr ? mr->name : "(all)", listener->name, 0);
2299         } else if (listener->log_sync_global) {
2300             /*
2301              * No matter whether MR is specified, what we can do here
2302              * is to do a global sync, because we are not capable to
2303              * sync in a finer granularity.
2304              */
2305             listener->log_sync_global(listener, last_stage);
2306             trace_memory_region_sync_dirty(mr ? mr->name : "(all)", listener->name, 1);
2307         }
2308     }
2309 }
2310 
2311 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start,
2312                                       hwaddr len)
2313 {
2314     MemoryRegionSection mrs;
2315     MemoryListener *listener;
2316     AddressSpace *as;
2317     FlatView *view;
2318     FlatRange *fr;
2319     hwaddr sec_start, sec_end, sec_size;
2320 
2321     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2322         if (!listener->log_clear) {
2323             continue;
2324         }
2325         as = listener->address_space;
2326         view = address_space_get_flatview(as);
2327         FOR_EACH_FLAT_RANGE(fr, view) {
2328             if (!fr->dirty_log_mask || fr->mr != mr) {
2329                 /*
2330                  * Clear dirty bitmap operation only applies to those
2331                  * regions whose dirty logging is at least enabled
2332                  */
2333                 continue;
2334             }
2335 
2336             mrs = section_from_flat_range(fr, view);
2337 
2338             sec_start = MAX(mrs.offset_within_region, start);
2339             sec_end = mrs.offset_within_region + int128_get64(mrs.size);
2340             sec_end = MIN(sec_end, start + len);
2341 
2342             if (sec_start >= sec_end) {
2343                 /*
2344                  * If this memory region section has no intersection
2345                  * with the requested range, skip.
2346                  */
2347                 continue;
2348             }
2349 
2350             /* Valid case; shrink the section if needed */
2351             mrs.offset_within_address_space +=
2352                 sec_start - mrs.offset_within_region;
2353             mrs.offset_within_region = sec_start;
2354             sec_size = sec_end - sec_start;
2355             mrs.size = int128_make64(sec_size);
2356             listener->log_clear(listener, &mrs);
2357         }
2358         flatview_unref(view);
2359     }
2360 }
2361 
2362 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
2363                                                             hwaddr addr,
2364                                                             hwaddr size,
2365                                                             unsigned client)
2366 {
2367     DirtyBitmapSnapshot *snapshot;
2368     assert(mr->ram_block);
2369     memory_region_sync_dirty_bitmap(mr, false);
2370     snapshot = cpu_physical_memory_snapshot_and_clear_dirty(mr, addr, size, client);
2371     memory_global_after_dirty_log_sync();
2372     return snapshot;
2373 }
2374 
2375 bool memory_region_snapshot_get_dirty(MemoryRegion *mr, DirtyBitmapSnapshot *snap,
2376                                       hwaddr addr, hwaddr size)
2377 {
2378     assert(mr->ram_block);
2379     return cpu_physical_memory_snapshot_get_dirty(snap,
2380                 memory_region_get_ram_addr(mr) + addr, size);
2381 }
2382 
2383 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
2384 {
2385     if (mr->readonly != readonly) {
2386         memory_region_transaction_begin();
2387         mr->readonly = readonly;
2388         memory_region_update_pending |= mr->enabled;
2389         memory_region_transaction_commit();
2390     }
2391 }
2392 
2393 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile)
2394 {
2395     if (mr->nonvolatile != nonvolatile) {
2396         memory_region_transaction_begin();
2397         mr->nonvolatile = nonvolatile;
2398         memory_region_update_pending |= mr->enabled;
2399         memory_region_transaction_commit();
2400     }
2401 }
2402 
2403 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
2404 {
2405     if (mr->romd_mode != romd_mode) {
2406         memory_region_transaction_begin();
2407         mr->romd_mode = romd_mode;
2408         memory_region_update_pending |= mr->enabled;
2409         memory_region_transaction_commit();
2410     }
2411 }
2412 
2413 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
2414                                hwaddr size, unsigned client)
2415 {
2416     assert(mr->ram_block);
2417     cpu_physical_memory_test_and_clear_dirty(
2418         memory_region_get_ram_addr(mr) + addr, size, client);
2419 }
2420 
2421 int memory_region_get_fd(MemoryRegion *mr)
2422 {
2423     RCU_READ_LOCK_GUARD();
2424     while (mr->alias) {
2425         mr = mr->alias;
2426     }
2427     return mr->ram_block->fd;
2428 }
2429 
2430 void *memory_region_get_ram_ptr(MemoryRegion *mr)
2431 {
2432     uint64_t offset = 0;
2433 
2434     RCU_READ_LOCK_GUARD();
2435     while (mr->alias) {
2436         offset += mr->alias_offset;
2437         mr = mr->alias;
2438     }
2439     assert(mr->ram_block);
2440     return qemu_map_ram_ptr(mr->ram_block, offset);
2441 }
2442 
2443 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset)
2444 {
2445     RAMBlock *block;
2446 
2447     block = qemu_ram_block_from_host(ptr, false, offset);
2448     if (!block) {
2449         return NULL;
2450     }
2451 
2452     return block->mr;
2453 }
2454 
2455 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr)
2456 {
2457     return mr->ram_block ? mr->ram_block->offset : RAM_ADDR_INVALID;
2458 }
2459 
2460 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
2461 {
2462     assert(mr->ram_block);
2463 
2464     qemu_ram_resize(mr->ram_block, newsize, errp);
2465 }
2466 
2467 void memory_region_msync(MemoryRegion *mr, hwaddr addr, hwaddr size)
2468 {
2469     if (mr->ram_block) {
2470         qemu_ram_msync(mr->ram_block, addr, size);
2471     }
2472 }
2473 
2474 void memory_region_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size)
2475 {
2476     /*
2477      * Might be extended case needed to cover
2478      * different types of memory regions
2479      */
2480     if (mr->dirty_log_mask) {
2481         memory_region_msync(mr, addr, size);
2482     }
2483 }
2484 
2485 /*
2486  * Call proper memory listeners about the change on the newly
2487  * added/removed CoalescedMemoryRange.
2488  */
2489 static void memory_region_update_coalesced_range(MemoryRegion *mr,
2490                                                  CoalescedMemoryRange *cmr,
2491                                                  bool add)
2492 {
2493     AddressSpace *as;
2494     FlatView *view;
2495     FlatRange *fr;
2496 
2497     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2498         view = address_space_get_flatview(as);
2499         FOR_EACH_FLAT_RANGE(fr, view) {
2500             if (fr->mr == mr) {
2501                 flat_range_coalesced_io_notify(fr, as, cmr, add);
2502             }
2503         }
2504         flatview_unref(view);
2505     }
2506 }
2507 
2508 void memory_region_set_coalescing(MemoryRegion *mr)
2509 {
2510     memory_region_clear_coalescing(mr);
2511     memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
2512 }
2513 
2514 void memory_region_add_coalescing(MemoryRegion *mr,
2515                                   hwaddr offset,
2516                                   uint64_t size)
2517 {
2518     CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
2519 
2520     cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
2521     QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
2522     memory_region_update_coalesced_range(mr, cmr, true);
2523     memory_region_set_flush_coalesced(mr);
2524 }
2525 
2526 void memory_region_clear_coalescing(MemoryRegion *mr)
2527 {
2528     CoalescedMemoryRange *cmr;
2529 
2530     if (QTAILQ_EMPTY(&mr->coalesced)) {
2531         return;
2532     }
2533 
2534     qemu_flush_coalesced_mmio_buffer();
2535     mr->flush_coalesced_mmio = false;
2536 
2537     while (!QTAILQ_EMPTY(&mr->coalesced)) {
2538         cmr = QTAILQ_FIRST(&mr->coalesced);
2539         QTAILQ_REMOVE(&mr->coalesced, cmr, link);
2540         memory_region_update_coalesced_range(mr, cmr, false);
2541         g_free(cmr);
2542     }
2543 }
2544 
2545 void memory_region_set_flush_coalesced(MemoryRegion *mr)
2546 {
2547     mr->flush_coalesced_mmio = true;
2548 }
2549 
2550 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
2551 {
2552     qemu_flush_coalesced_mmio_buffer();
2553     if (QTAILQ_EMPTY(&mr->coalesced)) {
2554         mr->flush_coalesced_mmio = false;
2555     }
2556 }
2557 
2558 void memory_region_add_eventfd(MemoryRegion *mr,
2559                                hwaddr addr,
2560                                unsigned size,
2561                                bool match_data,
2562                                uint64_t data,
2563                                EventNotifier *e)
2564 {
2565     MemoryRegionIoeventfd mrfd = {
2566         .addr.start = int128_make64(addr),
2567         .addr.size = int128_make64(size),
2568         .match_data = match_data,
2569         .data = data,
2570         .e = e,
2571     };
2572     unsigned i;
2573 
2574     if (size) {
2575         adjust_endianness(mr, &mrfd.data, size_memop(size) | MO_TE);
2576     }
2577     memory_region_transaction_begin();
2578     for (i = 0; i < mr->ioeventfd_nb; ++i) {
2579         if (memory_region_ioeventfd_before(&mrfd, &mr->ioeventfds[i])) {
2580             break;
2581         }
2582     }
2583     ++mr->ioeventfd_nb;
2584     mr->ioeventfds = g_realloc(mr->ioeventfds,
2585                                   sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
2586     memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
2587             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
2588     mr->ioeventfds[i] = mrfd;
2589     ioeventfd_update_pending |= mr->enabled;
2590     memory_region_transaction_commit();
2591 }
2592 
2593 void memory_region_del_eventfd(MemoryRegion *mr,
2594                                hwaddr addr,
2595                                unsigned size,
2596                                bool match_data,
2597                                uint64_t data,
2598                                EventNotifier *e)
2599 {
2600     MemoryRegionIoeventfd mrfd = {
2601         .addr.start = int128_make64(addr),
2602         .addr.size = int128_make64(size),
2603         .match_data = match_data,
2604         .data = data,
2605         .e = e,
2606     };
2607     unsigned i;
2608 
2609     if (size) {
2610         adjust_endianness(mr, &mrfd.data, size_memop(size) | MO_TE);
2611     }
2612     memory_region_transaction_begin();
2613     for (i = 0; i < mr->ioeventfd_nb; ++i) {
2614         if (memory_region_ioeventfd_equal(&mrfd, &mr->ioeventfds[i])) {
2615             break;
2616         }
2617     }
2618     assert(i != mr->ioeventfd_nb);
2619     memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
2620             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
2621     --mr->ioeventfd_nb;
2622     mr->ioeventfds = g_realloc(mr->ioeventfds,
2623                                   sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
2624     ioeventfd_update_pending |= mr->enabled;
2625     memory_region_transaction_commit();
2626 }
2627 
2628 static void memory_region_update_container_subregions(MemoryRegion *subregion)
2629 {
2630     MemoryRegion *mr = subregion->container;
2631     MemoryRegion *other;
2632 
2633     memory_region_transaction_begin();
2634 
2635     memory_region_ref(subregion);
2636     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
2637         if (subregion->priority >= other->priority) {
2638             QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
2639             goto done;
2640         }
2641     }
2642     QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
2643 done:
2644     memory_region_update_pending |= mr->enabled && subregion->enabled;
2645     memory_region_transaction_commit();
2646 }
2647 
2648 static void memory_region_add_subregion_common(MemoryRegion *mr,
2649                                                hwaddr offset,
2650                                                MemoryRegion *subregion)
2651 {
2652     MemoryRegion *alias;
2653 
2654     assert(!subregion->container);
2655     subregion->container = mr;
2656     for (alias = subregion->alias; alias; alias = alias->alias) {
2657         alias->mapped_via_alias++;
2658     }
2659     subregion->addr = offset;
2660     memory_region_update_container_subregions(subregion);
2661 }
2662 
2663 void memory_region_add_subregion(MemoryRegion *mr,
2664                                  hwaddr offset,
2665                                  MemoryRegion *subregion)
2666 {
2667     subregion->priority = 0;
2668     memory_region_add_subregion_common(mr, offset, subregion);
2669 }
2670 
2671 void memory_region_add_subregion_overlap(MemoryRegion *mr,
2672                                          hwaddr offset,
2673                                          MemoryRegion *subregion,
2674                                          int priority)
2675 {
2676     subregion->priority = priority;
2677     memory_region_add_subregion_common(mr, offset, subregion);
2678 }
2679 
2680 void memory_region_del_subregion(MemoryRegion *mr,
2681                                  MemoryRegion *subregion)
2682 {
2683     MemoryRegion *alias;
2684 
2685     memory_region_transaction_begin();
2686     assert(subregion->container == mr);
2687     subregion->container = NULL;
2688     for (alias = subregion->alias; alias; alias = alias->alias) {
2689         alias->mapped_via_alias--;
2690         assert(alias->mapped_via_alias >= 0);
2691     }
2692     QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
2693     memory_region_unref(subregion);
2694     memory_region_update_pending |= mr->enabled && subregion->enabled;
2695     memory_region_transaction_commit();
2696 }
2697 
2698 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
2699 {
2700     if (enabled == mr->enabled) {
2701         return;
2702     }
2703     memory_region_transaction_begin();
2704     mr->enabled = enabled;
2705     memory_region_update_pending = true;
2706     memory_region_transaction_commit();
2707 }
2708 
2709 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
2710 {
2711     Int128 s = int128_make64(size);
2712 
2713     if (size == UINT64_MAX) {
2714         s = int128_2_64();
2715     }
2716     if (int128_eq(s, mr->size)) {
2717         return;
2718     }
2719     memory_region_transaction_begin();
2720     mr->size = s;
2721     memory_region_update_pending = true;
2722     memory_region_transaction_commit();
2723 }
2724 
2725 static void memory_region_readd_subregion(MemoryRegion *mr)
2726 {
2727     MemoryRegion *container = mr->container;
2728 
2729     if (container) {
2730         memory_region_transaction_begin();
2731         memory_region_ref(mr);
2732         memory_region_del_subregion(container, mr);
2733         memory_region_add_subregion_common(container, mr->addr, mr);
2734         memory_region_unref(mr);
2735         memory_region_transaction_commit();
2736     }
2737 }
2738 
2739 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
2740 {
2741     if (addr != mr->addr) {
2742         mr->addr = addr;
2743         memory_region_readd_subregion(mr);
2744     }
2745 }
2746 
2747 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
2748 {
2749     assert(mr->alias);
2750 
2751     if (offset == mr->alias_offset) {
2752         return;
2753     }
2754 
2755     memory_region_transaction_begin();
2756     mr->alias_offset = offset;
2757     memory_region_update_pending |= mr->enabled;
2758     memory_region_transaction_commit();
2759 }
2760 
2761 void memory_region_set_unmergeable(MemoryRegion *mr, bool unmergeable)
2762 {
2763     if (unmergeable == mr->unmergeable) {
2764         return;
2765     }
2766 
2767     memory_region_transaction_begin();
2768     mr->unmergeable = unmergeable;
2769     memory_region_update_pending |= mr->enabled;
2770     memory_region_transaction_commit();
2771 }
2772 
2773 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
2774 {
2775     return mr->align;
2776 }
2777 
2778 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
2779 {
2780     const AddrRange *addr = addr_;
2781     const FlatRange *fr = fr_;
2782 
2783     if (int128_le(addrrange_end(*addr), fr->addr.start)) {
2784         return -1;
2785     } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
2786         return 1;
2787     }
2788     return 0;
2789 }
2790 
2791 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
2792 {
2793     return bsearch(&addr, view->ranges, view->nr,
2794                    sizeof(FlatRange), cmp_flatrange_addr);
2795 }
2796 
2797 bool memory_region_is_mapped(MemoryRegion *mr)
2798 {
2799     return !!mr->container || mr->mapped_via_alias;
2800 }
2801 
2802 /* Same as memory_region_find, but it does not add a reference to the
2803  * returned region.  It must be called from an RCU critical section.
2804  */
2805 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2806                                                   hwaddr addr, uint64_t size)
2807 {
2808     MemoryRegionSection ret = { .mr = NULL };
2809     MemoryRegion *root;
2810     AddressSpace *as;
2811     AddrRange range;
2812     FlatView *view;
2813     FlatRange *fr;
2814 
2815     addr += mr->addr;
2816     for (root = mr; root->container; ) {
2817         root = root->container;
2818         addr += root->addr;
2819     }
2820 
2821     as = memory_region_to_address_space(root);
2822     if (!as) {
2823         return ret;
2824     }
2825     range = addrrange_make(int128_make64(addr), int128_make64(size));
2826 
2827     view = address_space_to_flatview(as);
2828     fr = flatview_lookup(view, range);
2829     if (!fr) {
2830         return ret;
2831     }
2832 
2833     while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2834         --fr;
2835     }
2836 
2837     ret.mr = fr->mr;
2838     ret.fv = view;
2839     range = addrrange_intersection(range, fr->addr);
2840     ret.offset_within_region = fr->offset_in_region;
2841     ret.offset_within_region += int128_get64(int128_sub(range.start,
2842                                                         fr->addr.start));
2843     ret.size = range.size;
2844     ret.offset_within_address_space = int128_get64(range.start);
2845     ret.readonly = fr->readonly;
2846     ret.nonvolatile = fr->nonvolatile;
2847     return ret;
2848 }
2849 
2850 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2851                                        hwaddr addr, uint64_t size)
2852 {
2853     MemoryRegionSection ret;
2854     RCU_READ_LOCK_GUARD();
2855     ret = memory_region_find_rcu(mr, addr, size);
2856     if (ret.mr) {
2857         memory_region_ref(ret.mr);
2858     }
2859     return ret;
2860 }
2861 
2862 MemoryRegionSection *memory_region_section_new_copy(MemoryRegionSection *s)
2863 {
2864     MemoryRegionSection *tmp = g_new(MemoryRegionSection, 1);
2865 
2866     *tmp = *s;
2867     if (tmp->mr) {
2868         memory_region_ref(tmp->mr);
2869     }
2870     if (tmp->fv) {
2871         bool ret  = flatview_ref(tmp->fv);
2872 
2873         g_assert(ret);
2874     }
2875     return tmp;
2876 }
2877 
2878 void memory_region_section_free_copy(MemoryRegionSection *s)
2879 {
2880     if (s->fv) {
2881         flatview_unref(s->fv);
2882     }
2883     if (s->mr) {
2884         memory_region_unref(s->mr);
2885     }
2886     g_free(s);
2887 }
2888 
2889 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2890 {
2891     MemoryRegion *mr;
2892 
2893     RCU_READ_LOCK_GUARD();
2894     mr = memory_region_find_rcu(container, addr, 1).mr;
2895     return mr && mr != container;
2896 }
2897 
2898 void memory_global_dirty_log_sync(bool last_stage)
2899 {
2900     memory_region_sync_dirty_bitmap(NULL, last_stage);
2901 }
2902 
2903 void memory_global_after_dirty_log_sync(void)
2904 {
2905     MEMORY_LISTENER_CALL_GLOBAL(log_global_after_sync, Forward);
2906 }
2907 
2908 /*
2909  * Dirty track stop flags that are postponed due to VM being stopped.  Should
2910  * only be used within vmstate_change hook.
2911  */
2912 static unsigned int postponed_stop_flags;
2913 static VMChangeStateEntry *vmstate_change;
2914 static void memory_global_dirty_log_stop_postponed_run(void);
2915 
2916 void memory_global_dirty_log_start(unsigned int flags)
2917 {
2918     unsigned int old_flags;
2919 
2920     assert(flags && !(flags & (~GLOBAL_DIRTY_MASK)));
2921 
2922     if (vmstate_change) {
2923         /* If there is postponed stop(), operate on it first */
2924         postponed_stop_flags &= ~flags;
2925         memory_global_dirty_log_stop_postponed_run();
2926     }
2927 
2928     flags &= ~global_dirty_tracking;
2929     if (!flags) {
2930         return;
2931     }
2932 
2933     old_flags = global_dirty_tracking;
2934     global_dirty_tracking |= flags;
2935     trace_global_dirty_changed(global_dirty_tracking);
2936 
2937     if (!old_flags) {
2938         MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2939         memory_region_transaction_begin();
2940         memory_region_update_pending = true;
2941         memory_region_transaction_commit();
2942     }
2943 }
2944 
2945 static void memory_global_dirty_log_do_stop(unsigned int flags)
2946 {
2947     assert(flags && !(flags & (~GLOBAL_DIRTY_MASK)));
2948     assert((global_dirty_tracking & flags) == flags);
2949     global_dirty_tracking &= ~flags;
2950 
2951     trace_global_dirty_changed(global_dirty_tracking);
2952 
2953     if (!global_dirty_tracking) {
2954         memory_region_transaction_begin();
2955         memory_region_update_pending = true;
2956         memory_region_transaction_commit();
2957         MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2958     }
2959 }
2960 
2961 /*
2962  * Execute the postponed dirty log stop operations if there is, then reset
2963  * everything (including the flags and the vmstate change hook).
2964  */
2965 static void memory_global_dirty_log_stop_postponed_run(void)
2966 {
2967     /* This must be called with the vmstate handler registered */
2968     assert(vmstate_change);
2969 
2970     /* Note: postponed_stop_flags can be cleared in log start routine */
2971     if (postponed_stop_flags) {
2972         memory_global_dirty_log_do_stop(postponed_stop_flags);
2973         postponed_stop_flags = 0;
2974     }
2975 
2976     qemu_del_vm_change_state_handler(vmstate_change);
2977     vmstate_change = NULL;
2978 }
2979 
2980 static void memory_vm_change_state_handler(void *opaque, bool running,
2981                                            RunState state)
2982 {
2983     if (running) {
2984         memory_global_dirty_log_stop_postponed_run();
2985     }
2986 }
2987 
2988 void memory_global_dirty_log_stop(unsigned int flags)
2989 {
2990     if (!runstate_is_running()) {
2991         /* Postpone the dirty log stop, e.g., to when VM starts again */
2992         if (vmstate_change) {
2993             /* Batch with previous postponed flags */
2994             postponed_stop_flags |= flags;
2995         } else {
2996             postponed_stop_flags = flags;
2997             vmstate_change = qemu_add_vm_change_state_handler(
2998                 memory_vm_change_state_handler, NULL);
2999         }
3000         return;
3001     }
3002 
3003     memory_global_dirty_log_do_stop(flags);
3004 }
3005 
3006 static void listener_add_address_space(MemoryListener *listener,
3007                                        AddressSpace *as)
3008 {
3009     FlatView *view;
3010     FlatRange *fr;
3011 
3012     if (listener->begin) {
3013         listener->begin(listener);
3014     }
3015     if (global_dirty_tracking) {
3016         if (listener->log_global_start) {
3017             listener->log_global_start(listener);
3018         }
3019     }
3020 
3021     view = address_space_get_flatview(as);
3022     FOR_EACH_FLAT_RANGE(fr, view) {
3023         MemoryRegionSection section = section_from_flat_range(fr, view);
3024 
3025         if (listener->region_add) {
3026             listener->region_add(listener, &section);
3027         }
3028         if (fr->dirty_log_mask && listener->log_start) {
3029             listener->log_start(listener, &section, 0, fr->dirty_log_mask);
3030         }
3031     }
3032     if (listener->commit) {
3033         listener->commit(listener);
3034     }
3035     flatview_unref(view);
3036 }
3037 
3038 static void listener_del_address_space(MemoryListener *listener,
3039                                        AddressSpace *as)
3040 {
3041     FlatView *view;
3042     FlatRange *fr;
3043 
3044     if (listener->begin) {
3045         listener->begin(listener);
3046     }
3047     view = address_space_get_flatview(as);
3048     FOR_EACH_FLAT_RANGE(fr, view) {
3049         MemoryRegionSection section = section_from_flat_range(fr, view);
3050 
3051         if (fr->dirty_log_mask && listener->log_stop) {
3052             listener->log_stop(listener, &section, fr->dirty_log_mask, 0);
3053         }
3054         if (listener->region_del) {
3055             listener->region_del(listener, &section);
3056         }
3057     }
3058     if (listener->commit) {
3059         listener->commit(listener);
3060     }
3061     flatview_unref(view);
3062 }
3063 
3064 void memory_listener_register(MemoryListener *listener, AddressSpace *as)
3065 {
3066     MemoryListener *other = NULL;
3067 
3068     /* Only one of them can be defined for a listener */
3069     assert(!(listener->log_sync && listener->log_sync_global));
3070 
3071     listener->address_space = as;
3072     if (QTAILQ_EMPTY(&memory_listeners)
3073         || listener->priority >= QTAILQ_LAST(&memory_listeners)->priority) {
3074         QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
3075     } else {
3076         QTAILQ_FOREACH(other, &memory_listeners, link) {
3077             if (listener->priority < other->priority) {
3078                 break;
3079             }
3080         }
3081         QTAILQ_INSERT_BEFORE(other, listener, link);
3082     }
3083 
3084     if (QTAILQ_EMPTY(&as->listeners)
3085         || listener->priority >= QTAILQ_LAST(&as->listeners)->priority) {
3086         QTAILQ_INSERT_TAIL(&as->listeners, listener, link_as);
3087     } else {
3088         QTAILQ_FOREACH(other, &as->listeners, link_as) {
3089             if (listener->priority < other->priority) {
3090                 break;
3091             }
3092         }
3093         QTAILQ_INSERT_BEFORE(other, listener, link_as);
3094     }
3095 
3096     listener_add_address_space(listener, as);
3097 
3098     if (listener->eventfd_add || listener->eventfd_del) {
3099         as->ioeventfd_notifiers++;
3100     }
3101 }
3102 
3103 void memory_listener_unregister(MemoryListener *listener)
3104 {
3105     if (!listener->address_space) {
3106         return;
3107     }
3108 
3109     if (listener->eventfd_add || listener->eventfd_del) {
3110         listener->address_space->ioeventfd_notifiers--;
3111     }
3112 
3113     listener_del_address_space(listener, listener->address_space);
3114     QTAILQ_REMOVE(&memory_listeners, listener, link);
3115     QTAILQ_REMOVE(&listener->address_space->listeners, listener, link_as);
3116     listener->address_space = NULL;
3117 }
3118 
3119 void address_space_remove_listeners(AddressSpace *as)
3120 {
3121     while (!QTAILQ_EMPTY(&as->listeners)) {
3122         memory_listener_unregister(QTAILQ_FIRST(&as->listeners));
3123     }
3124 }
3125 
3126 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
3127 {
3128     memory_region_ref(root);
3129     as->root = root;
3130     as->current_map = NULL;
3131     as->ioeventfd_nb = 0;
3132     as->ioeventfds = NULL;
3133     QTAILQ_INIT(&as->listeners);
3134     QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
3135     as->name = g_strdup(name ? name : "anonymous");
3136     address_space_update_topology(as);
3137     address_space_update_ioeventfds(as);
3138 }
3139 
3140 static void do_address_space_destroy(AddressSpace *as)
3141 {
3142     assert(QTAILQ_EMPTY(&as->listeners));
3143 
3144     flatview_unref(as->current_map);
3145     g_free(as->name);
3146     g_free(as->ioeventfds);
3147     memory_region_unref(as->root);
3148 }
3149 
3150 void address_space_destroy(AddressSpace *as)
3151 {
3152     MemoryRegion *root = as->root;
3153 
3154     /* Flush out anything from MemoryListeners listening in on this */
3155     memory_region_transaction_begin();
3156     as->root = NULL;
3157     memory_region_transaction_commit();
3158     QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
3159 
3160     /* At this point, as->dispatch and as->current_map are dummy
3161      * entries that the guest should never use.  Wait for the old
3162      * values to expire before freeing the data.
3163      */
3164     as->root = root;
3165     call_rcu(as, do_address_space_destroy, rcu);
3166 }
3167 
3168 static const char *memory_region_type(MemoryRegion *mr)
3169 {
3170     if (mr->alias) {
3171         return memory_region_type(mr->alias);
3172     }
3173     if (memory_region_is_ram_device(mr)) {
3174         return "ramd";
3175     } else if (memory_region_is_romd(mr)) {
3176         return "romd";
3177     } else if (memory_region_is_rom(mr)) {
3178         return "rom";
3179     } else if (memory_region_is_ram(mr)) {
3180         return "ram";
3181     } else {
3182         return "i/o";
3183     }
3184 }
3185 
3186 typedef struct MemoryRegionList MemoryRegionList;
3187 
3188 struct MemoryRegionList {
3189     const MemoryRegion *mr;
3190     QTAILQ_ENTRY(MemoryRegionList) mrqueue;
3191 };
3192 
3193 typedef QTAILQ_HEAD(, MemoryRegionList) MemoryRegionListHead;
3194 
3195 #define MR_SIZE(size) (int128_nz(size) ? (hwaddr)int128_get64( \
3196                            int128_sub((size), int128_one())) : 0)
3197 #define MTREE_INDENT "  "
3198 
3199 static void mtree_expand_owner(const char *label, Object *obj)
3200 {
3201     DeviceState *dev = (DeviceState *) object_dynamic_cast(obj, TYPE_DEVICE);
3202 
3203     qemu_printf(" %s:{%s", label, dev ? "dev" : "obj");
3204     if (dev && dev->id) {
3205         qemu_printf(" id=%s", dev->id);
3206     } else {
3207         char *canonical_path = object_get_canonical_path(obj);
3208         if (canonical_path) {
3209             qemu_printf(" path=%s", canonical_path);
3210             g_free(canonical_path);
3211         } else {
3212             qemu_printf(" type=%s", object_get_typename(obj));
3213         }
3214     }
3215     qemu_printf("}");
3216 }
3217 
3218 static void mtree_print_mr_owner(const MemoryRegion *mr)
3219 {
3220     Object *owner = mr->owner;
3221     Object *parent = memory_region_owner((MemoryRegion *)mr);
3222 
3223     if (!owner && !parent) {
3224         qemu_printf(" orphan");
3225         return;
3226     }
3227     if (owner) {
3228         mtree_expand_owner("owner", owner);
3229     }
3230     if (parent && parent != owner) {
3231         mtree_expand_owner("parent", parent);
3232     }
3233 }
3234 
3235 static void mtree_print_mr(const MemoryRegion *mr, unsigned int level,
3236                            hwaddr base,
3237                            MemoryRegionListHead *alias_print_queue,
3238                            bool owner, bool display_disabled)
3239 {
3240     MemoryRegionList *new_ml, *ml, *next_ml;
3241     MemoryRegionListHead submr_print_queue;
3242     const MemoryRegion *submr;
3243     unsigned int i;
3244     hwaddr cur_start, cur_end;
3245 
3246     if (!mr) {
3247         return;
3248     }
3249 
3250     cur_start = base + mr->addr;
3251     cur_end = cur_start + MR_SIZE(mr->size);
3252 
3253     /*
3254      * Try to detect overflow of memory region. This should never
3255      * happen normally. When it happens, we dump something to warn the
3256      * user who is observing this.
3257      */
3258     if (cur_start < base || cur_end < cur_start) {
3259         qemu_printf("[DETECTED OVERFLOW!] ");
3260     }
3261 
3262     if (mr->alias) {
3263         bool found = false;
3264 
3265         /* check if the alias is already in the queue */
3266         QTAILQ_FOREACH(ml, alias_print_queue, mrqueue) {
3267             if (ml->mr == mr->alias) {
3268                 found = true;
3269             }
3270         }
3271 
3272         if (!found) {
3273             ml = g_new(MemoryRegionList, 1);
3274             ml->mr = mr->alias;
3275             QTAILQ_INSERT_TAIL(alias_print_queue, ml, mrqueue);
3276         }
3277         if (mr->enabled || display_disabled) {
3278             for (i = 0; i < level; i++) {
3279                 qemu_printf(MTREE_INDENT);
3280             }
3281             qemu_printf(HWADDR_FMT_plx "-" HWADDR_FMT_plx
3282                         " (prio %d, %s%s): alias %s @%s " HWADDR_FMT_plx
3283                         "-" HWADDR_FMT_plx "%s",
3284                         cur_start, cur_end,
3285                         mr->priority,
3286                         mr->nonvolatile ? "nv-" : "",
3287                         memory_region_type((MemoryRegion *)mr),
3288                         memory_region_name(mr),
3289                         memory_region_name(mr->alias),
3290                         mr->alias_offset,
3291                         mr->alias_offset + MR_SIZE(mr->size),
3292                         mr->enabled ? "" : " [disabled]");
3293             if (owner) {
3294                 mtree_print_mr_owner(mr);
3295             }
3296             qemu_printf("\n");
3297         }
3298     } else {
3299         if (mr->enabled || display_disabled) {
3300             for (i = 0; i < level; i++) {
3301                 qemu_printf(MTREE_INDENT);
3302             }
3303             qemu_printf(HWADDR_FMT_plx "-" HWADDR_FMT_plx
3304                         " (prio %d, %s%s): %s%s",
3305                         cur_start, cur_end,
3306                         mr->priority,
3307                         mr->nonvolatile ? "nv-" : "",
3308                         memory_region_type((MemoryRegion *)mr),
3309                         memory_region_name(mr),
3310                         mr->enabled ? "" : " [disabled]");
3311             if (owner) {
3312                 mtree_print_mr_owner(mr);
3313             }
3314             qemu_printf("\n");
3315         }
3316     }
3317 
3318     QTAILQ_INIT(&submr_print_queue);
3319 
3320     QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
3321         new_ml = g_new(MemoryRegionList, 1);
3322         new_ml->mr = submr;
3323         QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
3324             if (new_ml->mr->addr < ml->mr->addr ||
3325                 (new_ml->mr->addr == ml->mr->addr &&
3326                  new_ml->mr->priority > ml->mr->priority)) {
3327                 QTAILQ_INSERT_BEFORE(ml, new_ml, mrqueue);
3328                 new_ml = NULL;
3329                 break;
3330             }
3331         }
3332         if (new_ml) {
3333             QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, mrqueue);
3334         }
3335     }
3336 
3337     QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
3338         mtree_print_mr(ml->mr, level + 1, cur_start,
3339                        alias_print_queue, owner, display_disabled);
3340     }
3341 
3342     QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, mrqueue, next_ml) {
3343         g_free(ml);
3344     }
3345 }
3346 
3347 struct FlatViewInfo {
3348     int counter;
3349     bool dispatch_tree;
3350     bool owner;
3351     AccelClass *ac;
3352 };
3353 
3354 static void mtree_print_flatview(gpointer key, gpointer value,
3355                                  gpointer user_data)
3356 {
3357     FlatView *view = key;
3358     GArray *fv_address_spaces = value;
3359     struct FlatViewInfo *fvi = user_data;
3360     FlatRange *range = &view->ranges[0];
3361     MemoryRegion *mr;
3362     int n = view->nr;
3363     int i;
3364     AddressSpace *as;
3365 
3366     qemu_printf("FlatView #%d\n", fvi->counter);
3367     ++fvi->counter;
3368 
3369     for (i = 0; i < fv_address_spaces->len; ++i) {
3370         as = g_array_index(fv_address_spaces, AddressSpace*, i);
3371         qemu_printf(" AS \"%s\", root: %s",
3372                     as->name, memory_region_name(as->root));
3373         if (as->root->alias) {
3374             qemu_printf(", alias %s", memory_region_name(as->root->alias));
3375         }
3376         qemu_printf("\n");
3377     }
3378 
3379     qemu_printf(" Root memory region: %s\n",
3380       view->root ? memory_region_name(view->root) : "(none)");
3381 
3382     if (n <= 0) {
3383         qemu_printf(MTREE_INDENT "No rendered FlatView\n\n");
3384         return;
3385     }
3386 
3387     while (n--) {
3388         mr = range->mr;
3389         if (range->offset_in_region) {
3390             qemu_printf(MTREE_INDENT HWADDR_FMT_plx "-" HWADDR_FMT_plx
3391                         " (prio %d, %s%s): %s @" HWADDR_FMT_plx,
3392                         int128_get64(range->addr.start),
3393                         int128_get64(range->addr.start)
3394                         + MR_SIZE(range->addr.size),
3395                         mr->priority,
3396                         range->nonvolatile ? "nv-" : "",
3397                         range->readonly ? "rom" : memory_region_type(mr),
3398                         memory_region_name(mr),
3399                         range->offset_in_region);
3400         } else {
3401             qemu_printf(MTREE_INDENT HWADDR_FMT_plx "-" HWADDR_FMT_plx
3402                         " (prio %d, %s%s): %s",
3403                         int128_get64(range->addr.start),
3404                         int128_get64(range->addr.start)
3405                         + MR_SIZE(range->addr.size),
3406                         mr->priority,
3407                         range->nonvolatile ? "nv-" : "",
3408                         range->readonly ? "rom" : memory_region_type(mr),
3409                         memory_region_name(mr));
3410         }
3411         if (fvi->owner) {
3412             mtree_print_mr_owner(mr);
3413         }
3414 
3415         if (fvi->ac) {
3416             for (i = 0; i < fv_address_spaces->len; ++i) {
3417                 as = g_array_index(fv_address_spaces, AddressSpace*, i);
3418                 if (fvi->ac->has_memory(current_machine, as,
3419                                         int128_get64(range->addr.start),
3420                                         MR_SIZE(range->addr.size) + 1)) {
3421                     qemu_printf(" %s", fvi->ac->name);
3422                 }
3423             }
3424         }
3425         qemu_printf("\n");
3426         range++;
3427     }
3428 
3429 #if !defined(CONFIG_USER_ONLY)
3430     if (fvi->dispatch_tree && view->root) {
3431         mtree_print_dispatch(view->dispatch, view->root);
3432     }
3433 #endif
3434 
3435     qemu_printf("\n");
3436 }
3437 
3438 static gboolean mtree_info_flatview_free(gpointer key, gpointer value,
3439                                       gpointer user_data)
3440 {
3441     FlatView *view = key;
3442     GArray *fv_address_spaces = value;
3443 
3444     g_array_unref(fv_address_spaces);
3445     flatview_unref(view);
3446 
3447     return true;
3448 }
3449 
3450 static void mtree_info_flatview(bool dispatch_tree, bool owner)
3451 {
3452     struct FlatViewInfo fvi = {
3453         .counter = 0,
3454         .dispatch_tree = dispatch_tree,
3455         .owner = owner,
3456     };
3457     AddressSpace *as;
3458     FlatView *view;
3459     GArray *fv_address_spaces;
3460     GHashTable *views = g_hash_table_new(g_direct_hash, g_direct_equal);
3461     AccelClass *ac = ACCEL_GET_CLASS(current_accel());
3462 
3463     if (ac->has_memory) {
3464         fvi.ac = ac;
3465     }
3466 
3467     /* Gather all FVs in one table */
3468     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3469         view = address_space_get_flatview(as);
3470 
3471         fv_address_spaces = g_hash_table_lookup(views, view);
3472         if (!fv_address_spaces) {
3473             fv_address_spaces = g_array_new(false, false, sizeof(as));
3474             g_hash_table_insert(views, view, fv_address_spaces);
3475         }
3476 
3477         g_array_append_val(fv_address_spaces, as);
3478     }
3479 
3480     /* Print */
3481     g_hash_table_foreach(views, mtree_print_flatview, &fvi);
3482 
3483     /* Free */
3484     g_hash_table_foreach_remove(views, mtree_info_flatview_free, 0);
3485     g_hash_table_unref(views);
3486 }
3487 
3488 struct AddressSpaceInfo {
3489     MemoryRegionListHead *ml_head;
3490     bool owner;
3491     bool disabled;
3492 };
3493 
3494 /* Returns negative value if a < b; zero if a = b; positive value if a > b. */
3495 static gint address_space_compare_name(gconstpointer a, gconstpointer b)
3496 {
3497     const AddressSpace *as_a = a;
3498     const AddressSpace *as_b = b;
3499 
3500     return g_strcmp0(as_a->name, as_b->name);
3501 }
3502 
3503 static void mtree_print_as_name(gpointer data, gpointer user_data)
3504 {
3505     AddressSpace *as = data;
3506 
3507     qemu_printf("address-space: %s\n", as->name);
3508 }
3509 
3510 static void mtree_print_as(gpointer key, gpointer value, gpointer user_data)
3511 {
3512     MemoryRegion *mr = key;
3513     GSList *as_same_root_mr_list = value;
3514     struct AddressSpaceInfo *asi = user_data;
3515 
3516     g_slist_foreach(as_same_root_mr_list, mtree_print_as_name, NULL);
3517     mtree_print_mr(mr, 1, 0, asi->ml_head, asi->owner, asi->disabled);
3518     qemu_printf("\n");
3519 }
3520 
3521 static gboolean mtree_info_as_free(gpointer key, gpointer value,
3522                                    gpointer user_data)
3523 {
3524     GSList *as_same_root_mr_list = value;
3525 
3526     g_slist_free(as_same_root_mr_list);
3527 
3528     return true;
3529 }
3530 
3531 static void mtree_info_as(bool dispatch_tree, bool owner, bool disabled)
3532 {
3533     MemoryRegionListHead ml_head;
3534     MemoryRegionList *ml, *ml2;
3535     AddressSpace *as;
3536     GHashTable *views = g_hash_table_new(g_direct_hash, g_direct_equal);
3537     GSList *as_same_root_mr_list;
3538     struct AddressSpaceInfo asi = {
3539         .ml_head = &ml_head,
3540         .owner = owner,
3541         .disabled = disabled,
3542     };
3543 
3544     QTAILQ_INIT(&ml_head);
3545 
3546     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3547         /* Create hashtable, key=AS root MR, value = list of AS */
3548         as_same_root_mr_list = g_hash_table_lookup(views, as->root);
3549         as_same_root_mr_list = g_slist_insert_sorted(as_same_root_mr_list, as,
3550                                                      address_space_compare_name);
3551         g_hash_table_insert(views, as->root, as_same_root_mr_list);
3552     }
3553 
3554     /* print address spaces */
3555     g_hash_table_foreach(views, mtree_print_as, &asi);
3556     g_hash_table_foreach_remove(views, mtree_info_as_free, 0);
3557     g_hash_table_unref(views);
3558 
3559     /* print aliased regions */
3560     QTAILQ_FOREACH(ml, &ml_head, mrqueue) {
3561         qemu_printf("memory-region: %s\n", memory_region_name(ml->mr));
3562         mtree_print_mr(ml->mr, 1, 0, &ml_head, owner, disabled);
3563         qemu_printf("\n");
3564     }
3565 
3566     QTAILQ_FOREACH_SAFE(ml, &ml_head, mrqueue, ml2) {
3567         g_free(ml);
3568     }
3569 }
3570 
3571 void mtree_info(bool flatview, bool dispatch_tree, bool owner, bool disabled)
3572 {
3573     if (flatview) {
3574         mtree_info_flatview(dispatch_tree, owner);
3575     } else {
3576         mtree_info_as(dispatch_tree, owner, disabled);
3577     }
3578 }
3579 
3580 void memory_region_init_ram(MemoryRegion *mr,
3581                             Object *owner,
3582                             const char *name,
3583                             uint64_t size,
3584                             Error **errp)
3585 {
3586     DeviceState *owner_dev;
3587     Error *err = NULL;
3588 
3589     memory_region_init_ram_nomigrate(mr, owner, name, size, &err);
3590     if (err) {
3591         error_propagate(errp, err);
3592         return;
3593     }
3594     /* This will assert if owner is neither NULL nor a DeviceState.
3595      * We only want the owner here for the purposes of defining a
3596      * unique name for migration. TODO: Ideally we should implement
3597      * a naming scheme for Objects which are not DeviceStates, in
3598      * which case we can relax this restriction.
3599      */
3600     owner_dev = DEVICE(owner);
3601     vmstate_register_ram(mr, owner_dev);
3602 }
3603 
3604 void memory_region_init_rom(MemoryRegion *mr,
3605                             Object *owner,
3606                             const char *name,
3607                             uint64_t size,
3608                             Error **errp)
3609 {
3610     DeviceState *owner_dev;
3611     Error *err = NULL;
3612 
3613     memory_region_init_rom_nomigrate(mr, owner, name, size, &err);
3614     if (err) {
3615         error_propagate(errp, err);
3616         return;
3617     }
3618     /* This will assert if owner is neither NULL nor a DeviceState.
3619      * We only want the owner here for the purposes of defining a
3620      * unique name for migration. TODO: Ideally we should implement
3621      * a naming scheme for Objects which are not DeviceStates, in
3622      * which case we can relax this restriction.
3623      */
3624     owner_dev = DEVICE(owner);
3625     vmstate_register_ram(mr, owner_dev);
3626 }
3627 
3628 void memory_region_init_rom_device(MemoryRegion *mr,
3629                                    Object *owner,
3630                                    const MemoryRegionOps *ops,
3631                                    void *opaque,
3632                                    const char *name,
3633                                    uint64_t size,
3634                                    Error **errp)
3635 {
3636     DeviceState *owner_dev;
3637     Error *err = NULL;
3638 
3639     memory_region_init_rom_device_nomigrate(mr, owner, ops, opaque,
3640                                             name, size, &err);
3641     if (err) {
3642         error_propagate(errp, err);
3643         return;
3644     }
3645     /* This will assert if owner is neither NULL nor a DeviceState.
3646      * We only want the owner here for the purposes of defining a
3647      * unique name for migration. TODO: Ideally we should implement
3648      * a naming scheme for Objects which are not DeviceStates, in
3649      * which case we can relax this restriction.
3650      */
3651     owner_dev = DEVICE(owner);
3652     vmstate_register_ram(mr, owner_dev);
3653 }
3654 
3655 /*
3656  * Support system builds with CONFIG_FUZZ using a weak symbol and a stub for
3657  * the fuzz_dma_read_cb callback
3658  */
3659 #ifdef CONFIG_FUZZ
3660 void __attribute__((weak)) fuzz_dma_read_cb(size_t addr,
3661                       size_t len,
3662                       MemoryRegion *mr)
3663 {
3664 }
3665 #endif
3666 
3667 static const TypeInfo memory_region_info = {
3668     .parent             = TYPE_OBJECT,
3669     .name               = TYPE_MEMORY_REGION,
3670     .class_size         = sizeof(MemoryRegionClass),
3671     .instance_size      = sizeof(MemoryRegion),
3672     .instance_init      = memory_region_initfn,
3673     .instance_finalize  = memory_region_finalize,
3674 };
3675 
3676 static const TypeInfo iommu_memory_region_info = {
3677     .parent             = TYPE_MEMORY_REGION,
3678     .name               = TYPE_IOMMU_MEMORY_REGION,
3679     .class_size         = sizeof(IOMMUMemoryRegionClass),
3680     .instance_size      = sizeof(IOMMUMemoryRegion),
3681     .instance_init      = iommu_memory_region_initfn,
3682     .abstract           = true,
3683 };
3684 
3685 static const TypeInfo ram_discard_manager_info = {
3686     .parent             = TYPE_INTERFACE,
3687     .name               = TYPE_RAM_DISCARD_MANAGER,
3688     .class_size         = sizeof(RamDiscardManagerClass),
3689 };
3690 
3691 static void memory_register_types(void)
3692 {
3693     type_register_static(&memory_region_info);
3694     type_register_static(&iommu_memory_region_info);
3695     type_register_static(&ram_discard_manager_info);
3696 }
3697 
3698 type_init(memory_register_types)
3699