xref: /openbmc/qemu/system/memory.c (revision 4475a9b0)
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(bql_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 = ldn_he_p(mr->ram_block->host + addr, size);
1343 
1344     trace_memory_region_ram_device_read(get_cpu_index(), mr, addr, data, size);
1345 
1346     return data;
1347 }
1348 
1349 static void memory_region_ram_device_write(void *opaque, hwaddr addr,
1350                                            uint64_t data, unsigned size)
1351 {
1352     MemoryRegion *mr = opaque;
1353 
1354     trace_memory_region_ram_device_write(get_cpu_index(), mr, addr, data, size);
1355 
1356     stn_he_p(mr->ram_block->host + addr, size, data);
1357 }
1358 
1359 static const MemoryRegionOps ram_device_mem_ops = {
1360     .read = memory_region_ram_device_read,
1361     .write = memory_region_ram_device_write,
1362     .endianness = DEVICE_HOST_ENDIAN,
1363     .valid = {
1364         .min_access_size = 1,
1365         .max_access_size = 8,
1366         .unaligned = true,
1367     },
1368     .impl = {
1369         .min_access_size = 1,
1370         .max_access_size = 8,
1371         .unaligned = true,
1372     },
1373 };
1374 
1375 bool memory_region_access_valid(MemoryRegion *mr,
1376                                 hwaddr addr,
1377                                 unsigned size,
1378                                 bool is_write,
1379                                 MemTxAttrs attrs)
1380 {
1381     if (mr->ops->valid.accepts
1382         && !mr->ops->valid.accepts(mr->opaque, addr, size, is_write, attrs)) {
1383         qemu_log_mask(LOG_GUEST_ERROR, "Invalid %s at addr 0x%" HWADDR_PRIX
1384                       ", size %u, region '%s', reason: rejected\n",
1385                       is_write ? "write" : "read",
1386                       addr, size, memory_region_name(mr));
1387         return false;
1388     }
1389 
1390     if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1391         qemu_log_mask(LOG_GUEST_ERROR, "Invalid %s at addr 0x%" HWADDR_PRIX
1392                       ", size %u, region '%s', reason: unaligned\n",
1393                       is_write ? "write" : "read",
1394                       addr, size, memory_region_name(mr));
1395         return false;
1396     }
1397 
1398     /* Treat zero as compatibility all valid */
1399     if (!mr->ops->valid.max_access_size) {
1400         return true;
1401     }
1402 
1403     if (size > mr->ops->valid.max_access_size
1404         || size < mr->ops->valid.min_access_size) {
1405         qemu_log_mask(LOG_GUEST_ERROR, "Invalid %s at addr 0x%" HWADDR_PRIX
1406                       ", size %u, region '%s', reason: invalid size "
1407                       "(min:%u max:%u)\n",
1408                       is_write ? "write" : "read",
1409                       addr, size, memory_region_name(mr),
1410                       mr->ops->valid.min_access_size,
1411                       mr->ops->valid.max_access_size);
1412         return false;
1413     }
1414     return true;
1415 }
1416 
1417 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1418                                                 hwaddr addr,
1419                                                 uint64_t *pval,
1420                                                 unsigned size,
1421                                                 MemTxAttrs attrs)
1422 {
1423     *pval = 0;
1424 
1425     if (mr->ops->read) {
1426         return access_with_adjusted_size(addr, pval, size,
1427                                          mr->ops->impl.min_access_size,
1428                                          mr->ops->impl.max_access_size,
1429                                          memory_region_read_accessor,
1430                                          mr, attrs);
1431     } else {
1432         return access_with_adjusted_size(addr, pval, size,
1433                                          mr->ops->impl.min_access_size,
1434                                          mr->ops->impl.max_access_size,
1435                                          memory_region_read_with_attrs_accessor,
1436                                          mr, attrs);
1437     }
1438 }
1439 
1440 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1441                                         hwaddr addr,
1442                                         uint64_t *pval,
1443                                         MemOp op,
1444                                         MemTxAttrs attrs)
1445 {
1446     unsigned size = memop_size(op);
1447     MemTxResult r;
1448 
1449     if (mr->alias) {
1450         return memory_region_dispatch_read(mr->alias,
1451                                            mr->alias_offset + addr,
1452                                            pval, op, attrs);
1453     }
1454     if (!memory_region_access_valid(mr, addr, size, false, attrs)) {
1455         *pval = unassigned_mem_read(mr, addr, size);
1456         return MEMTX_DECODE_ERROR;
1457     }
1458 
1459     r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1460     adjust_endianness(mr, pval, op);
1461     return r;
1462 }
1463 
1464 /* Return true if an eventfd was signalled */
1465 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1466                                                     hwaddr addr,
1467                                                     uint64_t data,
1468                                                     unsigned size,
1469                                                     MemTxAttrs attrs)
1470 {
1471     MemoryRegionIoeventfd ioeventfd = {
1472         .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1473         .data = data,
1474     };
1475     unsigned i;
1476 
1477     for (i = 0; i < mr->ioeventfd_nb; i++) {
1478         ioeventfd.match_data = mr->ioeventfds[i].match_data;
1479         ioeventfd.e = mr->ioeventfds[i].e;
1480 
1481         if (memory_region_ioeventfd_equal(&ioeventfd, &mr->ioeventfds[i])) {
1482             event_notifier_set(ioeventfd.e);
1483             return true;
1484         }
1485     }
1486 
1487     return false;
1488 }
1489 
1490 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1491                                          hwaddr addr,
1492                                          uint64_t data,
1493                                          MemOp op,
1494                                          MemTxAttrs attrs)
1495 {
1496     unsigned size = memop_size(op);
1497 
1498     if (mr->alias) {
1499         return memory_region_dispatch_write(mr->alias,
1500                                             mr->alias_offset + addr,
1501                                             data, op, attrs);
1502     }
1503     if (!memory_region_access_valid(mr, addr, size, true, attrs)) {
1504         unassigned_mem_write(mr, addr, data, size);
1505         return MEMTX_DECODE_ERROR;
1506     }
1507 
1508     adjust_endianness(mr, &data, op);
1509 
1510     /*
1511      * FIXME: it's not clear why under KVM the write would be processed
1512      * directly, instead of going through eventfd.  This probably should
1513      * test "tcg_enabled() || qtest_enabled()", or should just go away.
1514      */
1515     if (!kvm_enabled() &&
1516         memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1517         return MEMTX_OK;
1518     }
1519 
1520     if (mr->ops->write) {
1521         return access_with_adjusted_size(addr, &data, size,
1522                                          mr->ops->impl.min_access_size,
1523                                          mr->ops->impl.max_access_size,
1524                                          memory_region_write_accessor, mr,
1525                                          attrs);
1526     } else {
1527         return
1528             access_with_adjusted_size(addr, &data, size,
1529                                       mr->ops->impl.min_access_size,
1530                                       mr->ops->impl.max_access_size,
1531                                       memory_region_write_with_attrs_accessor,
1532                                       mr, attrs);
1533     }
1534 }
1535 
1536 void memory_region_init_io(MemoryRegion *mr,
1537                            Object *owner,
1538                            const MemoryRegionOps *ops,
1539                            void *opaque,
1540                            const char *name,
1541                            uint64_t size)
1542 {
1543     memory_region_init(mr, owner, name, size);
1544     mr->ops = ops ? ops : &unassigned_mem_ops;
1545     mr->opaque = opaque;
1546     mr->terminates = true;
1547 }
1548 
1549 bool memory_region_init_ram_nomigrate(MemoryRegion *mr,
1550                                       Object *owner,
1551                                       const char *name,
1552                                       uint64_t size,
1553                                       Error **errp)
1554 {
1555     return memory_region_init_ram_flags_nomigrate(mr, owner, name,
1556                                                   size, 0, errp);
1557 }
1558 
1559 bool memory_region_init_ram_flags_nomigrate(MemoryRegion *mr,
1560                                             Object *owner,
1561                                             const char *name,
1562                                             uint64_t size,
1563                                             uint32_t ram_flags,
1564                                             Error **errp)
1565 {
1566     Error *err = NULL;
1567     memory_region_init(mr, owner, name, size);
1568     mr->ram = true;
1569     mr->terminates = true;
1570     mr->destructor = memory_region_destructor_ram;
1571     mr->ram_block = qemu_ram_alloc(size, ram_flags, mr, &err);
1572     if (err) {
1573         mr->size = int128_zero();
1574         object_unparent(OBJECT(mr));
1575         error_propagate(errp, err);
1576         return false;
1577     }
1578     return true;
1579 }
1580 
1581 bool memory_region_init_resizeable_ram(MemoryRegion *mr,
1582                                        Object *owner,
1583                                        const char *name,
1584                                        uint64_t size,
1585                                        uint64_t max_size,
1586                                        void (*resized)(const char*,
1587                                                        uint64_t length,
1588                                                        void *host),
1589                                        Error **errp)
1590 {
1591     Error *err = NULL;
1592     memory_region_init(mr, owner, name, size);
1593     mr->ram = true;
1594     mr->terminates = true;
1595     mr->destructor = memory_region_destructor_ram;
1596     mr->ram_block = qemu_ram_alloc_resizeable(size, max_size, resized,
1597                                               mr, &err);
1598     if (err) {
1599         mr->size = int128_zero();
1600         object_unparent(OBJECT(mr));
1601         error_propagate(errp, err);
1602         return false;
1603     }
1604     return true;
1605 }
1606 
1607 #ifdef CONFIG_POSIX
1608 bool memory_region_init_ram_from_file(MemoryRegion *mr,
1609                                       Object *owner,
1610                                       const char *name,
1611                                       uint64_t size,
1612                                       uint64_t align,
1613                                       uint32_t ram_flags,
1614                                       const char *path,
1615                                       ram_addr_t offset,
1616                                       Error **errp)
1617 {
1618     Error *err = NULL;
1619     memory_region_init(mr, owner, name, size);
1620     mr->ram = true;
1621     mr->readonly = !!(ram_flags & RAM_READONLY);
1622     mr->terminates = true;
1623     mr->destructor = memory_region_destructor_ram;
1624     mr->align = align;
1625     mr->ram_block = qemu_ram_alloc_from_file(size, mr, ram_flags, path,
1626                                              offset, &err);
1627     if (err) {
1628         mr->size = int128_zero();
1629         object_unparent(OBJECT(mr));
1630         error_propagate(errp, err);
1631         return false;
1632     }
1633     return true;
1634 }
1635 
1636 bool memory_region_init_ram_from_fd(MemoryRegion *mr,
1637                                     Object *owner,
1638                                     const char *name,
1639                                     uint64_t size,
1640                                     uint32_t ram_flags,
1641                                     int fd,
1642                                     ram_addr_t offset,
1643                                     Error **errp)
1644 {
1645     Error *err = NULL;
1646     memory_region_init(mr, owner, name, size);
1647     mr->ram = true;
1648     mr->readonly = !!(ram_flags & RAM_READONLY);
1649     mr->terminates = true;
1650     mr->destructor = memory_region_destructor_ram;
1651     mr->ram_block = qemu_ram_alloc_from_fd(size, mr, ram_flags, fd, offset,
1652                                            &err);
1653     if (err) {
1654         mr->size = int128_zero();
1655         object_unparent(OBJECT(mr));
1656         error_propagate(errp, err);
1657         return false;
1658     }
1659     return true;
1660 }
1661 #endif
1662 
1663 void memory_region_init_ram_ptr(MemoryRegion *mr,
1664                                 Object *owner,
1665                                 const char *name,
1666                                 uint64_t size,
1667                                 void *ptr)
1668 {
1669     memory_region_init(mr, owner, name, size);
1670     mr->ram = true;
1671     mr->terminates = true;
1672     mr->destructor = memory_region_destructor_ram;
1673 
1674     /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL.  */
1675     assert(ptr != NULL);
1676     mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_abort);
1677 }
1678 
1679 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
1680                                        Object *owner,
1681                                        const char *name,
1682                                        uint64_t size,
1683                                        void *ptr)
1684 {
1685     memory_region_init(mr, owner, name, size);
1686     mr->ram = true;
1687     mr->terminates = true;
1688     mr->ram_device = true;
1689     mr->ops = &ram_device_mem_ops;
1690     mr->opaque = mr;
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_abort);
1696 }
1697 
1698 void memory_region_init_alias(MemoryRegion *mr,
1699                               Object *owner,
1700                               const char *name,
1701                               MemoryRegion *orig,
1702                               hwaddr offset,
1703                               uint64_t size)
1704 {
1705     memory_region_init(mr, owner, name, size);
1706     mr->alias = orig;
1707     mr->alias_offset = offset;
1708 }
1709 
1710 bool memory_region_init_rom_nomigrate(MemoryRegion *mr,
1711                                       Object *owner,
1712                                       const char *name,
1713                                       uint64_t size,
1714                                       Error **errp)
1715 {
1716     if (!memory_region_init_ram_flags_nomigrate(mr, owner, name,
1717                                                 size, 0, errp)) {
1718          return false;
1719     }
1720     mr->readonly = true;
1721 
1722     return true;
1723 }
1724 
1725 bool memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1726                                              Object *owner,
1727                                              const MemoryRegionOps *ops,
1728                                              void *opaque,
1729                                              const char *name,
1730                                              uint64_t size,
1731                                              Error **errp)
1732 {
1733     Error *err = NULL;
1734     assert(ops);
1735     memory_region_init(mr, owner, name, size);
1736     mr->ops = ops;
1737     mr->opaque = opaque;
1738     mr->terminates = true;
1739     mr->rom_device = true;
1740     mr->destructor = memory_region_destructor_ram;
1741     mr->ram_block = qemu_ram_alloc(size, 0, mr, &err);
1742     if (err) {
1743         mr->size = int128_zero();
1744         object_unparent(OBJECT(mr));
1745         error_propagate(errp, err);
1746         return false;
1747     }
1748     return true;
1749 }
1750 
1751 void memory_region_init_iommu(void *_iommu_mr,
1752                               size_t instance_size,
1753                               const char *mrtypename,
1754                               Object *owner,
1755                               const char *name,
1756                               uint64_t size)
1757 {
1758     struct IOMMUMemoryRegion *iommu_mr;
1759     struct MemoryRegion *mr;
1760 
1761     object_initialize(_iommu_mr, instance_size, mrtypename);
1762     mr = MEMORY_REGION(_iommu_mr);
1763     memory_region_do_init(mr, owner, name, size);
1764     iommu_mr = IOMMU_MEMORY_REGION(mr);
1765     mr->terminates = true;  /* then re-forwards */
1766     QLIST_INIT(&iommu_mr->iommu_notify);
1767     iommu_mr->iommu_notify_flags = IOMMU_NOTIFIER_NONE;
1768 }
1769 
1770 static void memory_region_finalize(Object *obj)
1771 {
1772     MemoryRegion *mr = MEMORY_REGION(obj);
1773 
1774     assert(!mr->container);
1775 
1776     /* We know the region is not visible in any address space (it
1777      * does not have a container and cannot be a root either because
1778      * it has no references, so we can blindly clear mr->enabled.
1779      * memory_region_set_enabled instead could trigger a transaction
1780      * and cause an infinite loop.
1781      */
1782     mr->enabled = false;
1783     memory_region_transaction_begin();
1784     while (!QTAILQ_EMPTY(&mr->subregions)) {
1785         MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1786         memory_region_del_subregion(mr, subregion);
1787     }
1788     memory_region_transaction_commit();
1789 
1790     mr->destructor(mr);
1791     memory_region_clear_coalescing(mr);
1792     g_free((char *)mr->name);
1793     g_free(mr->ioeventfds);
1794 }
1795 
1796 Object *memory_region_owner(MemoryRegion *mr)
1797 {
1798     Object *obj = OBJECT(mr);
1799     return obj->parent;
1800 }
1801 
1802 void memory_region_ref(MemoryRegion *mr)
1803 {
1804     /* MMIO callbacks most likely will access data that belongs
1805      * to the owner, hence the need to ref/unref the owner whenever
1806      * the memory region is in use.
1807      *
1808      * The memory region is a child of its owner.  As long as the
1809      * owner doesn't call unparent itself on the memory region,
1810      * ref-ing the owner will also keep the memory region alive.
1811      * Memory regions without an owner are supposed to never go away;
1812      * we do not ref/unref them because it slows down DMA sensibly.
1813      */
1814     if (mr && mr->owner) {
1815         object_ref(mr->owner);
1816     }
1817 }
1818 
1819 void memory_region_unref(MemoryRegion *mr)
1820 {
1821     if (mr && mr->owner) {
1822         object_unref(mr->owner);
1823     }
1824 }
1825 
1826 uint64_t memory_region_size(MemoryRegion *mr)
1827 {
1828     if (int128_eq(mr->size, int128_2_64())) {
1829         return UINT64_MAX;
1830     }
1831     return int128_get64(mr->size);
1832 }
1833 
1834 const char *memory_region_name(const MemoryRegion *mr)
1835 {
1836     if (!mr->name) {
1837         ((MemoryRegion *)mr)->name =
1838             g_strdup(object_get_canonical_path_component(OBJECT(mr)));
1839     }
1840     return mr->name;
1841 }
1842 
1843 bool memory_region_is_ram_device(MemoryRegion *mr)
1844 {
1845     return mr->ram_device;
1846 }
1847 
1848 bool memory_region_is_protected(MemoryRegion *mr)
1849 {
1850     return mr->ram && (mr->ram_block->flags & RAM_PROTECTED);
1851 }
1852 
1853 bool memory_region_has_guest_memfd(MemoryRegion *mr)
1854 {
1855     return mr->ram_block && mr->ram_block->guest_memfd >= 0;
1856 }
1857 
1858 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1859 {
1860     uint8_t mask = mr->dirty_log_mask;
1861     RAMBlock *rb = mr->ram_block;
1862 
1863     if (global_dirty_tracking && ((rb && qemu_ram_is_migratable(rb)) ||
1864                              memory_region_is_iommu(mr))) {
1865         mask |= (1 << DIRTY_MEMORY_MIGRATION);
1866     }
1867 
1868     if (tcg_enabled() && rb) {
1869         /* TCG only cares about dirty memory logging for RAM, not IOMMU.  */
1870         mask |= (1 << DIRTY_MEMORY_CODE);
1871     }
1872     return mask;
1873 }
1874 
1875 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1876 {
1877     return memory_region_get_dirty_log_mask(mr) & (1 << client);
1878 }
1879 
1880 static int memory_region_update_iommu_notify_flags(IOMMUMemoryRegion *iommu_mr,
1881                                                    Error **errp)
1882 {
1883     IOMMUNotifierFlag flags = IOMMU_NOTIFIER_NONE;
1884     IOMMUNotifier *iommu_notifier;
1885     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1886     int ret = 0;
1887 
1888     IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
1889         flags |= iommu_notifier->notifier_flags;
1890     }
1891 
1892     if (flags != iommu_mr->iommu_notify_flags && imrc->notify_flag_changed) {
1893         ret = imrc->notify_flag_changed(iommu_mr,
1894                                         iommu_mr->iommu_notify_flags,
1895                                         flags, errp);
1896     }
1897 
1898     if (!ret) {
1899         iommu_mr->iommu_notify_flags = flags;
1900     }
1901     return ret;
1902 }
1903 
1904 int memory_region_iommu_set_page_size_mask(IOMMUMemoryRegion *iommu_mr,
1905                                            uint64_t page_size_mask,
1906                                            Error **errp)
1907 {
1908     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1909     int ret = 0;
1910 
1911     if (imrc->iommu_set_page_size_mask) {
1912         ret = imrc->iommu_set_page_size_mask(iommu_mr, page_size_mask, errp);
1913     }
1914     return ret;
1915 }
1916 
1917 int memory_region_register_iommu_notifier(MemoryRegion *mr,
1918                                           IOMMUNotifier *n, Error **errp)
1919 {
1920     IOMMUMemoryRegion *iommu_mr;
1921     int ret;
1922 
1923     if (mr->alias) {
1924         return memory_region_register_iommu_notifier(mr->alias, n, errp);
1925     }
1926 
1927     /* We need to register for at least one bitfield */
1928     iommu_mr = IOMMU_MEMORY_REGION(mr);
1929     assert(n->notifier_flags != IOMMU_NOTIFIER_NONE);
1930     assert(n->start <= n->end);
1931     assert(n->iommu_idx >= 0 &&
1932            n->iommu_idx < memory_region_iommu_num_indexes(iommu_mr));
1933 
1934     QLIST_INSERT_HEAD(&iommu_mr->iommu_notify, n, node);
1935     ret = memory_region_update_iommu_notify_flags(iommu_mr, errp);
1936     if (ret) {
1937         QLIST_REMOVE(n, node);
1938     }
1939     return ret;
1940 }
1941 
1942 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr)
1943 {
1944     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1945 
1946     if (imrc->get_min_page_size) {
1947         return imrc->get_min_page_size(iommu_mr);
1948     }
1949     return TARGET_PAGE_SIZE;
1950 }
1951 
1952 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
1953 {
1954     MemoryRegion *mr = MEMORY_REGION(iommu_mr);
1955     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1956     hwaddr addr, granularity;
1957     IOMMUTLBEntry iotlb;
1958 
1959     /* If the IOMMU has its own replay callback, override */
1960     if (imrc->replay) {
1961         imrc->replay(iommu_mr, n);
1962         return;
1963     }
1964 
1965     granularity = memory_region_iommu_get_min_page_size(iommu_mr);
1966 
1967     for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1968         iotlb = imrc->translate(iommu_mr, addr, IOMMU_NONE, n->iommu_idx);
1969         if (iotlb.perm != IOMMU_NONE) {
1970             n->notify(n, &iotlb);
1971         }
1972 
1973         /* if (2^64 - MR size) < granularity, it's possible to get an
1974          * infinite loop here.  This should catch such a wraparound */
1975         if ((addr + granularity) < addr) {
1976             break;
1977         }
1978     }
1979 }
1980 
1981 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1982                                              IOMMUNotifier *n)
1983 {
1984     IOMMUMemoryRegion *iommu_mr;
1985 
1986     if (mr->alias) {
1987         memory_region_unregister_iommu_notifier(mr->alias, n);
1988         return;
1989     }
1990     QLIST_REMOVE(n, node);
1991     iommu_mr = IOMMU_MEMORY_REGION(mr);
1992     memory_region_update_iommu_notify_flags(iommu_mr, NULL);
1993 }
1994 
1995 void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
1996                                     const IOMMUTLBEvent *event)
1997 {
1998     const IOMMUTLBEntry *entry = &event->entry;
1999     hwaddr entry_end = entry->iova + entry->addr_mask;
2000     IOMMUTLBEntry tmp = *entry;
2001 
2002     if (event->type == IOMMU_NOTIFIER_UNMAP) {
2003         assert(entry->perm == IOMMU_NONE);
2004     }
2005 
2006     /*
2007      * Skip the notification if the notification does not overlap
2008      * with registered range.
2009      */
2010     if (notifier->start > entry_end || notifier->end < entry->iova) {
2011         return;
2012     }
2013 
2014     if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
2015         /* Crop (iova, addr_mask) to range */
2016         tmp.iova = MAX(tmp.iova, notifier->start);
2017         tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
2018     } else {
2019         assert(entry->iova >= notifier->start && entry_end <= notifier->end);
2020     }
2021 
2022     if (event->type & notifier->notifier_flags) {
2023         notifier->notify(notifier, &tmp);
2024     }
2025 }
2026 
2027 void memory_region_unmap_iommu_notifier_range(IOMMUNotifier *notifier)
2028 {
2029     IOMMUTLBEvent event;
2030 
2031     event.type = IOMMU_NOTIFIER_UNMAP;
2032     event.entry.target_as = &address_space_memory;
2033     event.entry.iova = notifier->start;
2034     event.entry.perm = IOMMU_NONE;
2035     event.entry.addr_mask = notifier->end - notifier->start;
2036 
2037     memory_region_notify_iommu_one(notifier, &event);
2038 }
2039 
2040 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
2041                                 int iommu_idx,
2042                                 const IOMMUTLBEvent event)
2043 {
2044     IOMMUNotifier *iommu_notifier;
2045 
2046     assert(memory_region_is_iommu(MEMORY_REGION(iommu_mr)));
2047 
2048     IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
2049         if (iommu_notifier->iommu_idx == iommu_idx) {
2050             memory_region_notify_iommu_one(iommu_notifier, &event);
2051         }
2052     }
2053 }
2054 
2055 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
2056                                  enum IOMMUMemoryRegionAttr attr,
2057                                  void *data)
2058 {
2059     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2060 
2061     if (!imrc->get_attr) {
2062         return -EINVAL;
2063     }
2064 
2065     return imrc->get_attr(iommu_mr, attr, data);
2066 }
2067 
2068 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr,
2069                                        MemTxAttrs attrs)
2070 {
2071     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2072 
2073     if (!imrc->attrs_to_index) {
2074         return 0;
2075     }
2076 
2077     return imrc->attrs_to_index(iommu_mr, attrs);
2078 }
2079 
2080 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr)
2081 {
2082     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2083 
2084     if (!imrc->num_indexes) {
2085         return 1;
2086     }
2087 
2088     return imrc->num_indexes(iommu_mr);
2089 }
2090 
2091 RamDiscardManager *memory_region_get_ram_discard_manager(MemoryRegion *mr)
2092 {
2093     if (!memory_region_is_ram(mr)) {
2094         return NULL;
2095     }
2096     return mr->rdm;
2097 }
2098 
2099 void memory_region_set_ram_discard_manager(MemoryRegion *mr,
2100                                            RamDiscardManager *rdm)
2101 {
2102     g_assert(memory_region_is_ram(mr));
2103     g_assert(!rdm || !mr->rdm);
2104     mr->rdm = rdm;
2105 }
2106 
2107 uint64_t ram_discard_manager_get_min_granularity(const RamDiscardManager *rdm,
2108                                                  const MemoryRegion *mr)
2109 {
2110     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2111 
2112     g_assert(rdmc->get_min_granularity);
2113     return rdmc->get_min_granularity(rdm, mr);
2114 }
2115 
2116 bool ram_discard_manager_is_populated(const RamDiscardManager *rdm,
2117                                       const MemoryRegionSection *section)
2118 {
2119     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2120 
2121     g_assert(rdmc->is_populated);
2122     return rdmc->is_populated(rdm, section);
2123 }
2124 
2125 int ram_discard_manager_replay_populated(const RamDiscardManager *rdm,
2126                                          MemoryRegionSection *section,
2127                                          ReplayRamPopulate replay_fn,
2128                                          void *opaque)
2129 {
2130     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2131 
2132     g_assert(rdmc->replay_populated);
2133     return rdmc->replay_populated(rdm, section, replay_fn, opaque);
2134 }
2135 
2136 void ram_discard_manager_replay_discarded(const RamDiscardManager *rdm,
2137                                           MemoryRegionSection *section,
2138                                           ReplayRamDiscard replay_fn,
2139                                           void *opaque)
2140 {
2141     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2142 
2143     g_assert(rdmc->replay_discarded);
2144     rdmc->replay_discarded(rdm, section, replay_fn, opaque);
2145 }
2146 
2147 void ram_discard_manager_register_listener(RamDiscardManager *rdm,
2148                                            RamDiscardListener *rdl,
2149                                            MemoryRegionSection *section)
2150 {
2151     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2152 
2153     g_assert(rdmc->register_listener);
2154     rdmc->register_listener(rdm, rdl, section);
2155 }
2156 
2157 void ram_discard_manager_unregister_listener(RamDiscardManager *rdm,
2158                                              RamDiscardListener *rdl)
2159 {
2160     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2161 
2162     g_assert(rdmc->unregister_listener);
2163     rdmc->unregister_listener(rdm, rdl);
2164 }
2165 
2166 /* Called with rcu_read_lock held.  */
2167 bool memory_get_xlat_addr(IOMMUTLBEntry *iotlb, void **vaddr,
2168                           ram_addr_t *ram_addr, bool *read_only,
2169                           bool *mr_has_discard_manager, Error **errp)
2170 {
2171     MemoryRegion *mr;
2172     hwaddr xlat;
2173     hwaddr len = iotlb->addr_mask + 1;
2174     bool writable = iotlb->perm & IOMMU_WO;
2175 
2176     if (mr_has_discard_manager) {
2177         *mr_has_discard_manager = false;
2178     }
2179     /*
2180      * The IOMMU TLB entry we have just covers translation through
2181      * this IOMMU to its immediate target.  We need to translate
2182      * it the rest of the way through to memory.
2183      */
2184     mr = address_space_translate(&address_space_memory, iotlb->translated_addr,
2185                                  &xlat, &len, writable, MEMTXATTRS_UNSPECIFIED);
2186     if (!memory_region_is_ram(mr)) {
2187         error_setg(errp, "iommu map to non memory area %" HWADDR_PRIx "", xlat);
2188         return false;
2189     } else if (memory_region_has_ram_discard_manager(mr)) {
2190         RamDiscardManager *rdm = memory_region_get_ram_discard_manager(mr);
2191         MemoryRegionSection tmp = {
2192             .mr = mr,
2193             .offset_within_region = xlat,
2194             .size = int128_make64(len),
2195         };
2196         if (mr_has_discard_manager) {
2197             *mr_has_discard_manager = true;
2198         }
2199         /*
2200          * Malicious VMs can map memory into the IOMMU, which is expected
2201          * to remain discarded. vfio will pin all pages, populating memory.
2202          * Disallow that. vmstate priorities make sure any RamDiscardManager
2203          * were already restored before IOMMUs are restored.
2204          */
2205         if (!ram_discard_manager_is_populated(rdm, &tmp)) {
2206             error_setg(errp, "iommu map to discarded memory (e.g., unplugged"
2207                          " via virtio-mem): %" HWADDR_PRIx "",
2208                          iotlb->translated_addr);
2209             return false;
2210         }
2211     }
2212 
2213     /*
2214      * Translation truncates length to the IOMMU page size,
2215      * check that it did not truncate too much.
2216      */
2217     if (len & iotlb->addr_mask) {
2218         error_setg(errp, "iommu has granularity incompatible with target AS");
2219         return false;
2220     }
2221 
2222     if (vaddr) {
2223         *vaddr = memory_region_get_ram_ptr(mr) + xlat;
2224     }
2225 
2226     if (ram_addr) {
2227         *ram_addr = memory_region_get_ram_addr(mr) + xlat;
2228     }
2229 
2230     if (read_only) {
2231         *read_only = !writable || mr->readonly;
2232     }
2233 
2234     return true;
2235 }
2236 
2237 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
2238 {
2239     uint8_t mask = 1 << client;
2240     uint8_t old_logging;
2241 
2242     assert(client == DIRTY_MEMORY_VGA);
2243     old_logging = mr->vga_logging_count;
2244     mr->vga_logging_count += log ? 1 : -1;
2245     if (!!old_logging == !!mr->vga_logging_count) {
2246         return;
2247     }
2248 
2249     memory_region_transaction_begin();
2250     mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
2251     memory_region_update_pending |= mr->enabled;
2252     memory_region_transaction_commit();
2253 }
2254 
2255 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
2256                              hwaddr size)
2257 {
2258     assert(mr->ram_block);
2259     cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr,
2260                                         size,
2261                                         memory_region_get_dirty_log_mask(mr));
2262 }
2263 
2264 /*
2265  * If memory region `mr' is NULL, do global sync.  Otherwise, sync
2266  * dirty bitmap for the specified memory region.
2267  */
2268 static void memory_region_sync_dirty_bitmap(MemoryRegion *mr, bool last_stage)
2269 {
2270     MemoryListener *listener;
2271     AddressSpace *as;
2272     FlatView *view;
2273     FlatRange *fr;
2274 
2275     /* If the same address space has multiple log_sync listeners, we
2276      * visit that address space's FlatView multiple times.  But because
2277      * log_sync listeners are rare, it's still cheaper than walking each
2278      * address space once.
2279      */
2280     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2281         if (listener->log_sync) {
2282             as = listener->address_space;
2283             view = address_space_get_flatview(as);
2284             FOR_EACH_FLAT_RANGE(fr, view) {
2285                 if (fr->dirty_log_mask && (!mr || fr->mr == mr)) {
2286                     MemoryRegionSection mrs = section_from_flat_range(fr, view);
2287                     listener->log_sync(listener, &mrs);
2288                 }
2289             }
2290             flatview_unref(view);
2291             trace_memory_region_sync_dirty(mr ? mr->name : "(all)", listener->name, 0);
2292         } else if (listener->log_sync_global) {
2293             /*
2294              * No matter whether MR is specified, what we can do here
2295              * is to do a global sync, because we are not capable to
2296              * sync in a finer granularity.
2297              */
2298             listener->log_sync_global(listener, last_stage);
2299             trace_memory_region_sync_dirty(mr ? mr->name : "(all)", listener->name, 1);
2300         }
2301     }
2302 }
2303 
2304 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start,
2305                                       hwaddr len)
2306 {
2307     MemoryRegionSection mrs;
2308     MemoryListener *listener;
2309     AddressSpace *as;
2310     FlatView *view;
2311     FlatRange *fr;
2312     hwaddr sec_start, sec_end, sec_size;
2313 
2314     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2315         if (!listener->log_clear) {
2316             continue;
2317         }
2318         as = listener->address_space;
2319         view = address_space_get_flatview(as);
2320         FOR_EACH_FLAT_RANGE(fr, view) {
2321             if (!fr->dirty_log_mask || fr->mr != mr) {
2322                 /*
2323                  * Clear dirty bitmap operation only applies to those
2324                  * regions whose dirty logging is at least enabled
2325                  */
2326                 continue;
2327             }
2328 
2329             mrs = section_from_flat_range(fr, view);
2330 
2331             sec_start = MAX(mrs.offset_within_region, start);
2332             sec_end = mrs.offset_within_region + int128_get64(mrs.size);
2333             sec_end = MIN(sec_end, start + len);
2334 
2335             if (sec_start >= sec_end) {
2336                 /*
2337                  * If this memory region section has no intersection
2338                  * with the requested range, skip.
2339                  */
2340                 continue;
2341             }
2342 
2343             /* Valid case; shrink the section if needed */
2344             mrs.offset_within_address_space +=
2345                 sec_start - mrs.offset_within_region;
2346             mrs.offset_within_region = sec_start;
2347             sec_size = sec_end - sec_start;
2348             mrs.size = int128_make64(sec_size);
2349             listener->log_clear(listener, &mrs);
2350         }
2351         flatview_unref(view);
2352     }
2353 }
2354 
2355 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
2356                                                             hwaddr addr,
2357                                                             hwaddr size,
2358                                                             unsigned client)
2359 {
2360     DirtyBitmapSnapshot *snapshot;
2361     assert(mr->ram_block);
2362     memory_region_sync_dirty_bitmap(mr, false);
2363     snapshot = cpu_physical_memory_snapshot_and_clear_dirty(mr, addr, size, client);
2364     memory_global_after_dirty_log_sync();
2365     return snapshot;
2366 }
2367 
2368 bool memory_region_snapshot_get_dirty(MemoryRegion *mr, DirtyBitmapSnapshot *snap,
2369                                       hwaddr addr, hwaddr size)
2370 {
2371     assert(mr->ram_block);
2372     return cpu_physical_memory_snapshot_get_dirty(snap,
2373                 memory_region_get_ram_addr(mr) + addr, size);
2374 }
2375 
2376 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
2377 {
2378     if (mr->readonly != readonly) {
2379         memory_region_transaction_begin();
2380         mr->readonly = readonly;
2381         memory_region_update_pending |= mr->enabled;
2382         memory_region_transaction_commit();
2383     }
2384 }
2385 
2386 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile)
2387 {
2388     if (mr->nonvolatile != nonvolatile) {
2389         memory_region_transaction_begin();
2390         mr->nonvolatile = nonvolatile;
2391         memory_region_update_pending |= mr->enabled;
2392         memory_region_transaction_commit();
2393     }
2394 }
2395 
2396 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
2397 {
2398     if (mr->romd_mode != romd_mode) {
2399         memory_region_transaction_begin();
2400         mr->romd_mode = romd_mode;
2401         memory_region_update_pending |= mr->enabled;
2402         memory_region_transaction_commit();
2403     }
2404 }
2405 
2406 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
2407                                hwaddr size, unsigned client)
2408 {
2409     assert(mr->ram_block);
2410     cpu_physical_memory_test_and_clear_dirty(
2411         memory_region_get_ram_addr(mr) + addr, size, client);
2412 }
2413 
2414 int memory_region_get_fd(MemoryRegion *mr)
2415 {
2416     RCU_READ_LOCK_GUARD();
2417     while (mr->alias) {
2418         mr = mr->alias;
2419     }
2420     return mr->ram_block->fd;
2421 }
2422 
2423 void *memory_region_get_ram_ptr(MemoryRegion *mr)
2424 {
2425     uint64_t offset = 0;
2426 
2427     RCU_READ_LOCK_GUARD();
2428     while (mr->alias) {
2429         offset += mr->alias_offset;
2430         mr = mr->alias;
2431     }
2432     assert(mr->ram_block);
2433     return qemu_map_ram_ptr(mr->ram_block, offset);
2434 }
2435 
2436 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset)
2437 {
2438     RAMBlock *block;
2439 
2440     block = qemu_ram_block_from_host(ptr, false, offset);
2441     if (!block) {
2442         return NULL;
2443     }
2444 
2445     return block->mr;
2446 }
2447 
2448 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr)
2449 {
2450     return mr->ram_block ? mr->ram_block->offset : RAM_ADDR_INVALID;
2451 }
2452 
2453 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
2454 {
2455     assert(mr->ram_block);
2456 
2457     qemu_ram_resize(mr->ram_block, newsize, errp);
2458 }
2459 
2460 void memory_region_msync(MemoryRegion *mr, hwaddr addr, hwaddr size)
2461 {
2462     if (mr->ram_block) {
2463         qemu_ram_msync(mr->ram_block, addr, size);
2464     }
2465 }
2466 
2467 void memory_region_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size)
2468 {
2469     /*
2470      * Might be extended case needed to cover
2471      * different types of memory regions
2472      */
2473     if (mr->dirty_log_mask) {
2474         memory_region_msync(mr, addr, size);
2475     }
2476 }
2477 
2478 /*
2479  * Call proper memory listeners about the change on the newly
2480  * added/removed CoalescedMemoryRange.
2481  */
2482 static void memory_region_update_coalesced_range(MemoryRegion *mr,
2483                                                  CoalescedMemoryRange *cmr,
2484                                                  bool add)
2485 {
2486     AddressSpace *as;
2487     FlatView *view;
2488     FlatRange *fr;
2489 
2490     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2491         view = address_space_get_flatview(as);
2492         FOR_EACH_FLAT_RANGE(fr, view) {
2493             if (fr->mr == mr) {
2494                 flat_range_coalesced_io_notify(fr, as, cmr, add);
2495             }
2496         }
2497         flatview_unref(view);
2498     }
2499 }
2500 
2501 void memory_region_set_coalescing(MemoryRegion *mr)
2502 {
2503     memory_region_clear_coalescing(mr);
2504     memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
2505 }
2506 
2507 void memory_region_add_coalescing(MemoryRegion *mr,
2508                                   hwaddr offset,
2509                                   uint64_t size)
2510 {
2511     CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
2512 
2513     cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
2514     QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
2515     memory_region_update_coalesced_range(mr, cmr, true);
2516     memory_region_set_flush_coalesced(mr);
2517 }
2518 
2519 void memory_region_clear_coalescing(MemoryRegion *mr)
2520 {
2521     CoalescedMemoryRange *cmr;
2522 
2523     if (QTAILQ_EMPTY(&mr->coalesced)) {
2524         return;
2525     }
2526 
2527     qemu_flush_coalesced_mmio_buffer();
2528     mr->flush_coalesced_mmio = false;
2529 
2530     while (!QTAILQ_EMPTY(&mr->coalesced)) {
2531         cmr = QTAILQ_FIRST(&mr->coalesced);
2532         QTAILQ_REMOVE(&mr->coalesced, cmr, link);
2533         memory_region_update_coalesced_range(mr, cmr, false);
2534         g_free(cmr);
2535     }
2536 }
2537 
2538 void memory_region_set_flush_coalesced(MemoryRegion *mr)
2539 {
2540     mr->flush_coalesced_mmio = true;
2541 }
2542 
2543 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
2544 {
2545     qemu_flush_coalesced_mmio_buffer();
2546     if (QTAILQ_EMPTY(&mr->coalesced)) {
2547         mr->flush_coalesced_mmio = false;
2548     }
2549 }
2550 
2551 void memory_region_add_eventfd(MemoryRegion *mr,
2552                                hwaddr addr,
2553                                unsigned size,
2554                                bool match_data,
2555                                uint64_t data,
2556                                EventNotifier *e)
2557 {
2558     MemoryRegionIoeventfd mrfd = {
2559         .addr.start = int128_make64(addr),
2560         .addr.size = int128_make64(size),
2561         .match_data = match_data,
2562         .data = data,
2563         .e = e,
2564     };
2565     unsigned i;
2566 
2567     if (size) {
2568         adjust_endianness(mr, &mrfd.data, size_memop(size) | MO_TE);
2569     }
2570     memory_region_transaction_begin();
2571     for (i = 0; i < mr->ioeventfd_nb; ++i) {
2572         if (memory_region_ioeventfd_before(&mrfd, &mr->ioeventfds[i])) {
2573             break;
2574         }
2575     }
2576     ++mr->ioeventfd_nb;
2577     mr->ioeventfds = g_realloc(mr->ioeventfds,
2578                                   sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
2579     memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
2580             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
2581     mr->ioeventfds[i] = mrfd;
2582     ioeventfd_update_pending |= mr->enabled;
2583     memory_region_transaction_commit();
2584 }
2585 
2586 void memory_region_del_eventfd(MemoryRegion *mr,
2587                                hwaddr addr,
2588                                unsigned size,
2589                                bool match_data,
2590                                uint64_t data,
2591                                EventNotifier *e)
2592 {
2593     MemoryRegionIoeventfd mrfd = {
2594         .addr.start = int128_make64(addr),
2595         .addr.size = int128_make64(size),
2596         .match_data = match_data,
2597         .data = data,
2598         .e = e,
2599     };
2600     unsigned i;
2601 
2602     if (size) {
2603         adjust_endianness(mr, &mrfd.data, size_memop(size) | MO_TE);
2604     }
2605     memory_region_transaction_begin();
2606     for (i = 0; i < mr->ioeventfd_nb; ++i) {
2607         if (memory_region_ioeventfd_equal(&mrfd, &mr->ioeventfds[i])) {
2608             break;
2609         }
2610     }
2611     assert(i != mr->ioeventfd_nb);
2612     memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
2613             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
2614     --mr->ioeventfd_nb;
2615     mr->ioeventfds = g_realloc(mr->ioeventfds,
2616                                   sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
2617     ioeventfd_update_pending |= mr->enabled;
2618     memory_region_transaction_commit();
2619 }
2620 
2621 static void memory_region_update_container_subregions(MemoryRegion *subregion)
2622 {
2623     MemoryRegion *mr = subregion->container;
2624     MemoryRegion *other;
2625 
2626     memory_region_transaction_begin();
2627 
2628     memory_region_ref(subregion);
2629     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
2630         if (subregion->priority >= other->priority) {
2631             QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
2632             goto done;
2633         }
2634     }
2635     QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
2636 done:
2637     memory_region_update_pending |= mr->enabled && subregion->enabled;
2638     memory_region_transaction_commit();
2639 }
2640 
2641 static void memory_region_add_subregion_common(MemoryRegion *mr,
2642                                                hwaddr offset,
2643                                                MemoryRegion *subregion)
2644 {
2645     MemoryRegion *alias;
2646 
2647     assert(!subregion->container);
2648     subregion->container = mr;
2649     for (alias = subregion->alias; alias; alias = alias->alias) {
2650         alias->mapped_via_alias++;
2651     }
2652     subregion->addr = offset;
2653     memory_region_update_container_subregions(subregion);
2654 }
2655 
2656 void memory_region_add_subregion(MemoryRegion *mr,
2657                                  hwaddr offset,
2658                                  MemoryRegion *subregion)
2659 {
2660     subregion->priority = 0;
2661     memory_region_add_subregion_common(mr, offset, subregion);
2662 }
2663 
2664 void memory_region_add_subregion_overlap(MemoryRegion *mr,
2665                                          hwaddr offset,
2666                                          MemoryRegion *subregion,
2667                                          int priority)
2668 {
2669     subregion->priority = priority;
2670     memory_region_add_subregion_common(mr, offset, subregion);
2671 }
2672 
2673 void memory_region_del_subregion(MemoryRegion *mr,
2674                                  MemoryRegion *subregion)
2675 {
2676     MemoryRegion *alias;
2677 
2678     memory_region_transaction_begin();
2679     assert(subregion->container == mr);
2680     subregion->container = NULL;
2681     for (alias = subregion->alias; alias; alias = alias->alias) {
2682         alias->mapped_via_alias--;
2683         assert(alias->mapped_via_alias >= 0);
2684     }
2685     QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
2686     memory_region_unref(subregion);
2687     memory_region_update_pending |= mr->enabled && subregion->enabled;
2688     memory_region_transaction_commit();
2689 }
2690 
2691 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
2692 {
2693     if (enabled == mr->enabled) {
2694         return;
2695     }
2696     memory_region_transaction_begin();
2697     mr->enabled = enabled;
2698     memory_region_update_pending = true;
2699     memory_region_transaction_commit();
2700 }
2701 
2702 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
2703 {
2704     Int128 s = int128_make64(size);
2705 
2706     if (size == UINT64_MAX) {
2707         s = int128_2_64();
2708     }
2709     if (int128_eq(s, mr->size)) {
2710         return;
2711     }
2712     memory_region_transaction_begin();
2713     mr->size = s;
2714     memory_region_update_pending = true;
2715     memory_region_transaction_commit();
2716 }
2717 
2718 static void memory_region_readd_subregion(MemoryRegion *mr)
2719 {
2720     MemoryRegion *container = mr->container;
2721 
2722     if (container) {
2723         memory_region_transaction_begin();
2724         memory_region_ref(mr);
2725         memory_region_del_subregion(container, mr);
2726         memory_region_add_subregion_common(container, mr->addr, mr);
2727         memory_region_unref(mr);
2728         memory_region_transaction_commit();
2729     }
2730 }
2731 
2732 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
2733 {
2734     if (addr != mr->addr) {
2735         mr->addr = addr;
2736         memory_region_readd_subregion(mr);
2737     }
2738 }
2739 
2740 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
2741 {
2742     assert(mr->alias);
2743 
2744     if (offset == mr->alias_offset) {
2745         return;
2746     }
2747 
2748     memory_region_transaction_begin();
2749     mr->alias_offset = offset;
2750     memory_region_update_pending |= mr->enabled;
2751     memory_region_transaction_commit();
2752 }
2753 
2754 void memory_region_set_unmergeable(MemoryRegion *mr, bool unmergeable)
2755 {
2756     if (unmergeable == mr->unmergeable) {
2757         return;
2758     }
2759 
2760     memory_region_transaction_begin();
2761     mr->unmergeable = unmergeable;
2762     memory_region_update_pending |= mr->enabled;
2763     memory_region_transaction_commit();
2764 }
2765 
2766 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
2767 {
2768     return mr->align;
2769 }
2770 
2771 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
2772 {
2773     const AddrRange *addr = addr_;
2774     const FlatRange *fr = fr_;
2775 
2776     if (int128_le(addrrange_end(*addr), fr->addr.start)) {
2777         return -1;
2778     } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
2779         return 1;
2780     }
2781     return 0;
2782 }
2783 
2784 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
2785 {
2786     return bsearch(&addr, view->ranges, view->nr,
2787                    sizeof(FlatRange), cmp_flatrange_addr);
2788 }
2789 
2790 bool memory_region_is_mapped(MemoryRegion *mr)
2791 {
2792     return !!mr->container || mr->mapped_via_alias;
2793 }
2794 
2795 /* Same as memory_region_find, but it does not add a reference to the
2796  * returned region.  It must be called from an RCU critical section.
2797  */
2798 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2799                                                   hwaddr addr, uint64_t size)
2800 {
2801     MemoryRegionSection ret = { .mr = NULL };
2802     MemoryRegion *root;
2803     AddressSpace *as;
2804     AddrRange range;
2805     FlatView *view;
2806     FlatRange *fr;
2807 
2808     addr += mr->addr;
2809     for (root = mr; root->container; ) {
2810         root = root->container;
2811         addr += root->addr;
2812     }
2813 
2814     as = memory_region_to_address_space(root);
2815     if (!as) {
2816         return ret;
2817     }
2818     range = addrrange_make(int128_make64(addr), int128_make64(size));
2819 
2820     view = address_space_to_flatview(as);
2821     fr = flatview_lookup(view, range);
2822     if (!fr) {
2823         return ret;
2824     }
2825 
2826     while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2827         --fr;
2828     }
2829 
2830     ret.mr = fr->mr;
2831     ret.fv = view;
2832     range = addrrange_intersection(range, fr->addr);
2833     ret.offset_within_region = fr->offset_in_region;
2834     ret.offset_within_region += int128_get64(int128_sub(range.start,
2835                                                         fr->addr.start));
2836     ret.size = range.size;
2837     ret.offset_within_address_space = int128_get64(range.start);
2838     ret.readonly = fr->readonly;
2839     ret.nonvolatile = fr->nonvolatile;
2840     return ret;
2841 }
2842 
2843 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2844                                        hwaddr addr, uint64_t size)
2845 {
2846     MemoryRegionSection ret;
2847     RCU_READ_LOCK_GUARD();
2848     ret = memory_region_find_rcu(mr, addr, size);
2849     if (ret.mr) {
2850         memory_region_ref(ret.mr);
2851     }
2852     return ret;
2853 }
2854 
2855 MemoryRegionSection *memory_region_section_new_copy(MemoryRegionSection *s)
2856 {
2857     MemoryRegionSection *tmp = g_new(MemoryRegionSection, 1);
2858 
2859     *tmp = *s;
2860     if (tmp->mr) {
2861         memory_region_ref(tmp->mr);
2862     }
2863     if (tmp->fv) {
2864         bool ret  = flatview_ref(tmp->fv);
2865 
2866         g_assert(ret);
2867     }
2868     return tmp;
2869 }
2870 
2871 void memory_region_section_free_copy(MemoryRegionSection *s)
2872 {
2873     if (s->fv) {
2874         flatview_unref(s->fv);
2875     }
2876     if (s->mr) {
2877         memory_region_unref(s->mr);
2878     }
2879     g_free(s);
2880 }
2881 
2882 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2883 {
2884     MemoryRegion *mr;
2885 
2886     RCU_READ_LOCK_GUARD();
2887     mr = memory_region_find_rcu(container, addr, 1).mr;
2888     return mr && mr != container;
2889 }
2890 
2891 void memory_global_dirty_log_sync(bool last_stage)
2892 {
2893     memory_region_sync_dirty_bitmap(NULL, last_stage);
2894 }
2895 
2896 void memory_global_after_dirty_log_sync(void)
2897 {
2898     MEMORY_LISTENER_CALL_GLOBAL(log_global_after_sync, Forward);
2899 }
2900 
2901 /*
2902  * Dirty track stop flags that are postponed due to VM being stopped.  Should
2903  * only be used within vmstate_change hook.
2904  */
2905 static unsigned int postponed_stop_flags;
2906 static VMChangeStateEntry *vmstate_change;
2907 static void memory_global_dirty_log_stop_postponed_run(void);
2908 
2909 static bool memory_global_dirty_log_do_start(Error **errp)
2910 {
2911     MemoryListener *listener;
2912 
2913     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2914         if (listener->log_global_start) {
2915             if (!listener->log_global_start(listener, errp)) {
2916                 goto err;
2917             }
2918         }
2919     }
2920     return true;
2921 
2922 err:
2923     while ((listener = QTAILQ_PREV(listener, link)) != NULL) {
2924         if (listener->log_global_stop) {
2925             listener->log_global_stop(listener);
2926         }
2927     }
2928 
2929     return false;
2930 }
2931 
2932 bool memory_global_dirty_log_start(unsigned int flags, Error **errp)
2933 {
2934     unsigned int old_flags;
2935 
2936     assert(flags && !(flags & (~GLOBAL_DIRTY_MASK)));
2937 
2938     if (vmstate_change) {
2939         /* If there is postponed stop(), operate on it first */
2940         postponed_stop_flags &= ~flags;
2941         memory_global_dirty_log_stop_postponed_run();
2942     }
2943 
2944     flags &= ~global_dirty_tracking;
2945     if (!flags) {
2946         return true;
2947     }
2948 
2949     old_flags = global_dirty_tracking;
2950     global_dirty_tracking |= flags;
2951     trace_global_dirty_changed(global_dirty_tracking);
2952 
2953     if (!old_flags) {
2954         if (!memory_global_dirty_log_do_start(errp)) {
2955             global_dirty_tracking &= ~flags;
2956             trace_global_dirty_changed(global_dirty_tracking);
2957             return false;
2958         }
2959 
2960         memory_region_transaction_begin();
2961         memory_region_update_pending = true;
2962         memory_region_transaction_commit();
2963     }
2964     return true;
2965 }
2966 
2967 static void memory_global_dirty_log_do_stop(unsigned int flags)
2968 {
2969     assert(flags && !(flags & (~GLOBAL_DIRTY_MASK)));
2970     assert((global_dirty_tracking & flags) == flags);
2971     global_dirty_tracking &= ~flags;
2972 
2973     trace_global_dirty_changed(global_dirty_tracking);
2974 
2975     if (!global_dirty_tracking) {
2976         memory_region_transaction_begin();
2977         memory_region_update_pending = true;
2978         memory_region_transaction_commit();
2979         MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2980     }
2981 }
2982 
2983 /*
2984  * Execute the postponed dirty log stop operations if there is, then reset
2985  * everything (including the flags and the vmstate change hook).
2986  */
2987 static void memory_global_dirty_log_stop_postponed_run(void)
2988 {
2989     /* This must be called with the vmstate handler registered */
2990     assert(vmstate_change);
2991 
2992     /* Note: postponed_stop_flags can be cleared in log start routine */
2993     if (postponed_stop_flags) {
2994         memory_global_dirty_log_do_stop(postponed_stop_flags);
2995         postponed_stop_flags = 0;
2996     }
2997 
2998     qemu_del_vm_change_state_handler(vmstate_change);
2999     vmstate_change = NULL;
3000 }
3001 
3002 static void memory_vm_change_state_handler(void *opaque, bool running,
3003                                            RunState state)
3004 {
3005     if (running) {
3006         memory_global_dirty_log_stop_postponed_run();
3007     }
3008 }
3009 
3010 void memory_global_dirty_log_stop(unsigned int flags)
3011 {
3012     if (!runstate_is_running()) {
3013         /* Postpone the dirty log stop, e.g., to when VM starts again */
3014         if (vmstate_change) {
3015             /* Batch with previous postponed flags */
3016             postponed_stop_flags |= flags;
3017         } else {
3018             postponed_stop_flags = flags;
3019             vmstate_change = qemu_add_vm_change_state_handler(
3020                 memory_vm_change_state_handler, NULL);
3021         }
3022         return;
3023     }
3024 
3025     memory_global_dirty_log_do_stop(flags);
3026 }
3027 
3028 static void listener_add_address_space(MemoryListener *listener,
3029                                        AddressSpace *as)
3030 {
3031     FlatView *view;
3032     FlatRange *fr;
3033 
3034     if (listener->begin) {
3035         listener->begin(listener);
3036     }
3037     if (global_dirty_tracking) {
3038         /*
3039          * Currently only VFIO can fail log_global_start(), and it's not
3040          * yet allowed to hotplug any PCI device during migration. So this
3041          * should never fail when invoked, guard it with error_abort.  If
3042          * it can start to fail in the future, we need to be able to fail
3043          * the whole listener_add_address_space() and its callers.
3044          */
3045         if (listener->log_global_start) {
3046             listener->log_global_start(listener, &error_abort);
3047         }
3048     }
3049 
3050     view = address_space_get_flatview(as);
3051     FOR_EACH_FLAT_RANGE(fr, view) {
3052         MemoryRegionSection section = section_from_flat_range(fr, view);
3053 
3054         if (listener->region_add) {
3055             listener->region_add(listener, &section);
3056         }
3057         if (fr->dirty_log_mask && listener->log_start) {
3058             listener->log_start(listener, &section, 0, fr->dirty_log_mask);
3059         }
3060     }
3061     if (listener->commit) {
3062         listener->commit(listener);
3063     }
3064     flatview_unref(view);
3065 }
3066 
3067 static void listener_del_address_space(MemoryListener *listener,
3068                                        AddressSpace *as)
3069 {
3070     FlatView *view;
3071     FlatRange *fr;
3072 
3073     if (listener->begin) {
3074         listener->begin(listener);
3075     }
3076     view = address_space_get_flatview(as);
3077     FOR_EACH_FLAT_RANGE(fr, view) {
3078         MemoryRegionSection section = section_from_flat_range(fr, view);
3079 
3080         if (fr->dirty_log_mask && listener->log_stop) {
3081             listener->log_stop(listener, &section, fr->dirty_log_mask, 0);
3082         }
3083         if (listener->region_del) {
3084             listener->region_del(listener, &section);
3085         }
3086     }
3087     if (listener->commit) {
3088         listener->commit(listener);
3089     }
3090     flatview_unref(view);
3091 }
3092 
3093 void memory_listener_register(MemoryListener *listener, AddressSpace *as)
3094 {
3095     MemoryListener *other = NULL;
3096 
3097     /* Only one of them can be defined for a listener */
3098     assert(!(listener->log_sync && listener->log_sync_global));
3099 
3100     listener->address_space = as;
3101     if (QTAILQ_EMPTY(&memory_listeners)
3102         || listener->priority >= QTAILQ_LAST(&memory_listeners)->priority) {
3103         QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
3104     } else {
3105         QTAILQ_FOREACH(other, &memory_listeners, link) {
3106             if (listener->priority < other->priority) {
3107                 break;
3108             }
3109         }
3110         QTAILQ_INSERT_BEFORE(other, listener, link);
3111     }
3112 
3113     if (QTAILQ_EMPTY(&as->listeners)
3114         || listener->priority >= QTAILQ_LAST(&as->listeners)->priority) {
3115         QTAILQ_INSERT_TAIL(&as->listeners, listener, link_as);
3116     } else {
3117         QTAILQ_FOREACH(other, &as->listeners, link_as) {
3118             if (listener->priority < other->priority) {
3119                 break;
3120             }
3121         }
3122         QTAILQ_INSERT_BEFORE(other, listener, link_as);
3123     }
3124 
3125     listener_add_address_space(listener, as);
3126 
3127     if (listener->eventfd_add || listener->eventfd_del) {
3128         as->ioeventfd_notifiers++;
3129     }
3130 }
3131 
3132 void memory_listener_unregister(MemoryListener *listener)
3133 {
3134     if (!listener->address_space) {
3135         return;
3136     }
3137 
3138     if (listener->eventfd_add || listener->eventfd_del) {
3139         listener->address_space->ioeventfd_notifiers--;
3140     }
3141 
3142     listener_del_address_space(listener, listener->address_space);
3143     QTAILQ_REMOVE(&memory_listeners, listener, link);
3144     QTAILQ_REMOVE(&listener->address_space->listeners, listener, link_as);
3145     listener->address_space = NULL;
3146 }
3147 
3148 void address_space_remove_listeners(AddressSpace *as)
3149 {
3150     while (!QTAILQ_EMPTY(&as->listeners)) {
3151         memory_listener_unregister(QTAILQ_FIRST(&as->listeners));
3152     }
3153 }
3154 
3155 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
3156 {
3157     memory_region_ref(root);
3158     as->root = root;
3159     as->current_map = NULL;
3160     as->ioeventfd_nb = 0;
3161     as->ioeventfds = NULL;
3162     QTAILQ_INIT(&as->listeners);
3163     QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
3164     as->bounce.in_use = false;
3165     qemu_mutex_init(&as->map_client_list_lock);
3166     QLIST_INIT(&as->map_client_list);
3167     as->name = g_strdup(name ? name : "anonymous");
3168     address_space_update_topology(as);
3169     address_space_update_ioeventfds(as);
3170 }
3171 
3172 static void do_address_space_destroy(AddressSpace *as)
3173 {
3174     assert(!qatomic_read(&as->bounce.in_use));
3175     assert(QLIST_EMPTY(&as->map_client_list));
3176     qemu_mutex_destroy(&as->map_client_list_lock);
3177 
3178     assert(QTAILQ_EMPTY(&as->listeners));
3179 
3180     flatview_unref(as->current_map);
3181     g_free(as->name);
3182     g_free(as->ioeventfds);
3183     memory_region_unref(as->root);
3184 }
3185 
3186 void address_space_destroy(AddressSpace *as)
3187 {
3188     MemoryRegion *root = as->root;
3189 
3190     /* Flush out anything from MemoryListeners listening in on this */
3191     memory_region_transaction_begin();
3192     as->root = NULL;
3193     memory_region_transaction_commit();
3194     QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
3195 
3196     /* At this point, as->dispatch and as->current_map are dummy
3197      * entries that the guest should never use.  Wait for the old
3198      * values to expire before freeing the data.
3199      */
3200     as->root = root;
3201     call_rcu(as, do_address_space_destroy, rcu);
3202 }
3203 
3204 static const char *memory_region_type(MemoryRegion *mr)
3205 {
3206     if (mr->alias) {
3207         return memory_region_type(mr->alias);
3208     }
3209     if (memory_region_is_ram_device(mr)) {
3210         return "ramd";
3211     } else if (memory_region_is_romd(mr)) {
3212         return "romd";
3213     } else if (memory_region_is_rom(mr)) {
3214         return "rom";
3215     } else if (memory_region_is_ram(mr)) {
3216         return "ram";
3217     } else {
3218         return "i/o";
3219     }
3220 }
3221 
3222 typedef struct MemoryRegionList MemoryRegionList;
3223 
3224 struct MemoryRegionList {
3225     const MemoryRegion *mr;
3226     QTAILQ_ENTRY(MemoryRegionList) mrqueue;
3227 };
3228 
3229 typedef QTAILQ_HEAD(, MemoryRegionList) MemoryRegionListHead;
3230 
3231 #define MR_SIZE(size) (int128_nz(size) ? (hwaddr)int128_get64( \
3232                            int128_sub((size), int128_one())) : 0)
3233 #define MTREE_INDENT "  "
3234 
3235 static void mtree_expand_owner(const char *label, Object *obj)
3236 {
3237     DeviceState *dev = (DeviceState *) object_dynamic_cast(obj, TYPE_DEVICE);
3238 
3239     qemu_printf(" %s:{%s", label, dev ? "dev" : "obj");
3240     if (dev && dev->id) {
3241         qemu_printf(" id=%s", dev->id);
3242     } else {
3243         char *canonical_path = object_get_canonical_path(obj);
3244         if (canonical_path) {
3245             qemu_printf(" path=%s", canonical_path);
3246             g_free(canonical_path);
3247         } else {
3248             qemu_printf(" type=%s", object_get_typename(obj));
3249         }
3250     }
3251     qemu_printf("}");
3252 }
3253 
3254 static void mtree_print_mr_owner(const MemoryRegion *mr)
3255 {
3256     Object *owner = mr->owner;
3257     Object *parent = memory_region_owner((MemoryRegion *)mr);
3258 
3259     if (!owner && !parent) {
3260         qemu_printf(" orphan");
3261         return;
3262     }
3263     if (owner) {
3264         mtree_expand_owner("owner", owner);
3265     }
3266     if (parent && parent != owner) {
3267         mtree_expand_owner("parent", parent);
3268     }
3269 }
3270 
3271 static void mtree_print_mr(const MemoryRegion *mr, unsigned int level,
3272                            hwaddr base,
3273                            MemoryRegionListHead *alias_print_queue,
3274                            bool owner, bool display_disabled)
3275 {
3276     MemoryRegionList *new_ml, *ml, *next_ml;
3277     MemoryRegionListHead submr_print_queue;
3278     const MemoryRegion *submr;
3279     unsigned int i;
3280     hwaddr cur_start, cur_end;
3281 
3282     if (!mr) {
3283         return;
3284     }
3285 
3286     cur_start = base + mr->addr;
3287     cur_end = cur_start + MR_SIZE(mr->size);
3288 
3289     /*
3290      * Try to detect overflow of memory region. This should never
3291      * happen normally. When it happens, we dump something to warn the
3292      * user who is observing this.
3293      */
3294     if (cur_start < base || cur_end < cur_start) {
3295         qemu_printf("[DETECTED OVERFLOW!] ");
3296     }
3297 
3298     if (mr->alias) {
3299         bool found = false;
3300 
3301         /* check if the alias is already in the queue */
3302         QTAILQ_FOREACH(ml, alias_print_queue, mrqueue) {
3303             if (ml->mr == mr->alias) {
3304                 found = true;
3305             }
3306         }
3307 
3308         if (!found) {
3309             ml = g_new(MemoryRegionList, 1);
3310             ml->mr = mr->alias;
3311             QTAILQ_INSERT_TAIL(alias_print_queue, ml, mrqueue);
3312         }
3313         if (mr->enabled || display_disabled) {
3314             for (i = 0; i < level; i++) {
3315                 qemu_printf(MTREE_INDENT);
3316             }
3317             qemu_printf(HWADDR_FMT_plx "-" HWADDR_FMT_plx
3318                         " (prio %d, %s%s): alias %s @%s " HWADDR_FMT_plx
3319                         "-" HWADDR_FMT_plx "%s",
3320                         cur_start, cur_end,
3321                         mr->priority,
3322                         mr->nonvolatile ? "nv-" : "",
3323                         memory_region_type((MemoryRegion *)mr),
3324                         memory_region_name(mr),
3325                         memory_region_name(mr->alias),
3326                         mr->alias_offset,
3327                         mr->alias_offset + MR_SIZE(mr->size),
3328                         mr->enabled ? "" : " [disabled]");
3329             if (owner) {
3330                 mtree_print_mr_owner(mr);
3331             }
3332             qemu_printf("\n");
3333         }
3334     } else {
3335         if (mr->enabled || display_disabled) {
3336             for (i = 0; i < level; i++) {
3337                 qemu_printf(MTREE_INDENT);
3338             }
3339             qemu_printf(HWADDR_FMT_plx "-" HWADDR_FMT_plx
3340                         " (prio %d, %s%s): %s%s",
3341                         cur_start, cur_end,
3342                         mr->priority,
3343                         mr->nonvolatile ? "nv-" : "",
3344                         memory_region_type((MemoryRegion *)mr),
3345                         memory_region_name(mr),
3346                         mr->enabled ? "" : " [disabled]");
3347             if (owner) {
3348                 mtree_print_mr_owner(mr);
3349             }
3350             qemu_printf("\n");
3351         }
3352     }
3353 
3354     QTAILQ_INIT(&submr_print_queue);
3355 
3356     QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
3357         new_ml = g_new(MemoryRegionList, 1);
3358         new_ml->mr = submr;
3359         QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
3360             if (new_ml->mr->addr < ml->mr->addr ||
3361                 (new_ml->mr->addr == ml->mr->addr &&
3362                  new_ml->mr->priority > ml->mr->priority)) {
3363                 QTAILQ_INSERT_BEFORE(ml, new_ml, mrqueue);
3364                 new_ml = NULL;
3365                 break;
3366             }
3367         }
3368         if (new_ml) {
3369             QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, mrqueue);
3370         }
3371     }
3372 
3373     QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
3374         mtree_print_mr(ml->mr, level + 1, cur_start,
3375                        alias_print_queue, owner, display_disabled);
3376     }
3377 
3378     QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, mrqueue, next_ml) {
3379         g_free(ml);
3380     }
3381 }
3382 
3383 struct FlatViewInfo {
3384     int counter;
3385     bool dispatch_tree;
3386     bool owner;
3387     AccelClass *ac;
3388 };
3389 
3390 static void mtree_print_flatview(gpointer key, gpointer value,
3391                                  gpointer user_data)
3392 {
3393     FlatView *view = key;
3394     GArray *fv_address_spaces = value;
3395     struct FlatViewInfo *fvi = user_data;
3396     FlatRange *range = &view->ranges[0];
3397     MemoryRegion *mr;
3398     int n = view->nr;
3399     int i;
3400     AddressSpace *as;
3401 
3402     qemu_printf("FlatView #%d\n", fvi->counter);
3403     ++fvi->counter;
3404 
3405     for (i = 0; i < fv_address_spaces->len; ++i) {
3406         as = g_array_index(fv_address_spaces, AddressSpace*, i);
3407         qemu_printf(" AS \"%s\", root: %s",
3408                     as->name, memory_region_name(as->root));
3409         if (as->root->alias) {
3410             qemu_printf(", alias %s", memory_region_name(as->root->alias));
3411         }
3412         qemu_printf("\n");
3413     }
3414 
3415     qemu_printf(" Root memory region: %s\n",
3416       view->root ? memory_region_name(view->root) : "(none)");
3417 
3418     if (n <= 0) {
3419         qemu_printf(MTREE_INDENT "No rendered FlatView\n\n");
3420         return;
3421     }
3422 
3423     while (n--) {
3424         mr = range->mr;
3425         if (range->offset_in_region) {
3426             qemu_printf(MTREE_INDENT HWADDR_FMT_plx "-" HWADDR_FMT_plx
3427                         " (prio %d, %s%s): %s @" HWADDR_FMT_plx,
3428                         int128_get64(range->addr.start),
3429                         int128_get64(range->addr.start)
3430                         + MR_SIZE(range->addr.size),
3431                         mr->priority,
3432                         range->nonvolatile ? "nv-" : "",
3433                         range->readonly ? "rom" : memory_region_type(mr),
3434                         memory_region_name(mr),
3435                         range->offset_in_region);
3436         } else {
3437             qemu_printf(MTREE_INDENT HWADDR_FMT_plx "-" HWADDR_FMT_plx
3438                         " (prio %d, %s%s): %s",
3439                         int128_get64(range->addr.start),
3440                         int128_get64(range->addr.start)
3441                         + MR_SIZE(range->addr.size),
3442                         mr->priority,
3443                         range->nonvolatile ? "nv-" : "",
3444                         range->readonly ? "rom" : memory_region_type(mr),
3445                         memory_region_name(mr));
3446         }
3447         if (fvi->owner) {
3448             mtree_print_mr_owner(mr);
3449         }
3450 
3451         if (fvi->ac) {
3452             for (i = 0; i < fv_address_spaces->len; ++i) {
3453                 as = g_array_index(fv_address_spaces, AddressSpace*, i);
3454                 if (fvi->ac->has_memory(current_machine, as,
3455                                         int128_get64(range->addr.start),
3456                                         MR_SIZE(range->addr.size) + 1)) {
3457                     qemu_printf(" %s", fvi->ac->name);
3458                 }
3459             }
3460         }
3461         qemu_printf("\n");
3462         range++;
3463     }
3464 
3465 #if !defined(CONFIG_USER_ONLY)
3466     if (fvi->dispatch_tree && view->root) {
3467         mtree_print_dispatch(view->dispatch, view->root);
3468     }
3469 #endif
3470 
3471     qemu_printf("\n");
3472 }
3473 
3474 static gboolean mtree_info_flatview_free(gpointer key, gpointer value,
3475                                       gpointer user_data)
3476 {
3477     FlatView *view = key;
3478     GArray *fv_address_spaces = value;
3479 
3480     g_array_unref(fv_address_spaces);
3481     flatview_unref(view);
3482 
3483     return true;
3484 }
3485 
3486 static void mtree_info_flatview(bool dispatch_tree, bool owner)
3487 {
3488     struct FlatViewInfo fvi = {
3489         .counter = 0,
3490         .dispatch_tree = dispatch_tree,
3491         .owner = owner,
3492     };
3493     AddressSpace *as;
3494     FlatView *view;
3495     GArray *fv_address_spaces;
3496     GHashTable *views = g_hash_table_new(g_direct_hash, g_direct_equal);
3497     AccelClass *ac = ACCEL_GET_CLASS(current_accel());
3498 
3499     if (ac->has_memory) {
3500         fvi.ac = ac;
3501     }
3502 
3503     /* Gather all FVs in one table */
3504     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3505         view = address_space_get_flatview(as);
3506 
3507         fv_address_spaces = g_hash_table_lookup(views, view);
3508         if (!fv_address_spaces) {
3509             fv_address_spaces = g_array_new(false, false, sizeof(as));
3510             g_hash_table_insert(views, view, fv_address_spaces);
3511         }
3512 
3513         g_array_append_val(fv_address_spaces, as);
3514     }
3515 
3516     /* Print */
3517     g_hash_table_foreach(views, mtree_print_flatview, &fvi);
3518 
3519     /* Free */
3520     g_hash_table_foreach_remove(views, mtree_info_flatview_free, 0);
3521     g_hash_table_unref(views);
3522 }
3523 
3524 struct AddressSpaceInfo {
3525     MemoryRegionListHead *ml_head;
3526     bool owner;
3527     bool disabled;
3528 };
3529 
3530 /* Returns negative value if a < b; zero if a = b; positive value if a > b. */
3531 static gint address_space_compare_name(gconstpointer a, gconstpointer b)
3532 {
3533     const AddressSpace *as_a = a;
3534     const AddressSpace *as_b = b;
3535 
3536     return g_strcmp0(as_a->name, as_b->name);
3537 }
3538 
3539 static void mtree_print_as_name(gpointer data, gpointer user_data)
3540 {
3541     AddressSpace *as = data;
3542 
3543     qemu_printf("address-space: %s\n", as->name);
3544 }
3545 
3546 static void mtree_print_as(gpointer key, gpointer value, gpointer user_data)
3547 {
3548     MemoryRegion *mr = key;
3549     GSList *as_same_root_mr_list = value;
3550     struct AddressSpaceInfo *asi = user_data;
3551 
3552     g_slist_foreach(as_same_root_mr_list, mtree_print_as_name, NULL);
3553     mtree_print_mr(mr, 1, 0, asi->ml_head, asi->owner, asi->disabled);
3554     qemu_printf("\n");
3555 }
3556 
3557 static gboolean mtree_info_as_free(gpointer key, gpointer value,
3558                                    gpointer user_data)
3559 {
3560     GSList *as_same_root_mr_list = value;
3561 
3562     g_slist_free(as_same_root_mr_list);
3563 
3564     return true;
3565 }
3566 
3567 static void mtree_info_as(bool dispatch_tree, bool owner, bool disabled)
3568 {
3569     MemoryRegionListHead ml_head;
3570     MemoryRegionList *ml, *ml2;
3571     AddressSpace *as;
3572     GHashTable *views = g_hash_table_new(g_direct_hash, g_direct_equal);
3573     GSList *as_same_root_mr_list;
3574     struct AddressSpaceInfo asi = {
3575         .ml_head = &ml_head,
3576         .owner = owner,
3577         .disabled = disabled,
3578     };
3579 
3580     QTAILQ_INIT(&ml_head);
3581 
3582     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3583         /* Create hashtable, key=AS root MR, value = list of AS */
3584         as_same_root_mr_list = g_hash_table_lookup(views, as->root);
3585         as_same_root_mr_list = g_slist_insert_sorted(as_same_root_mr_list, as,
3586                                                      address_space_compare_name);
3587         g_hash_table_insert(views, as->root, as_same_root_mr_list);
3588     }
3589 
3590     /* print address spaces */
3591     g_hash_table_foreach(views, mtree_print_as, &asi);
3592     g_hash_table_foreach_remove(views, mtree_info_as_free, 0);
3593     g_hash_table_unref(views);
3594 
3595     /* print aliased regions */
3596     QTAILQ_FOREACH(ml, &ml_head, mrqueue) {
3597         qemu_printf("memory-region: %s\n", memory_region_name(ml->mr));
3598         mtree_print_mr(ml->mr, 1, 0, &ml_head, owner, disabled);
3599         qemu_printf("\n");
3600     }
3601 
3602     QTAILQ_FOREACH_SAFE(ml, &ml_head, mrqueue, ml2) {
3603         g_free(ml);
3604     }
3605 }
3606 
3607 void mtree_info(bool flatview, bool dispatch_tree, bool owner, bool disabled)
3608 {
3609     if (flatview) {
3610         mtree_info_flatview(dispatch_tree, owner);
3611     } else {
3612         mtree_info_as(dispatch_tree, owner, disabled);
3613     }
3614 }
3615 
3616 bool memory_region_init_ram(MemoryRegion *mr,
3617                             Object *owner,
3618                             const char *name,
3619                             uint64_t size,
3620                             Error **errp)
3621 {
3622     DeviceState *owner_dev;
3623 
3624     if (!memory_region_init_ram_nomigrate(mr, owner, name, size, errp)) {
3625         return false;
3626     }
3627     /* This will assert if owner is neither NULL nor a DeviceState.
3628      * We only want the owner here for the purposes of defining a
3629      * unique name for migration. TODO: Ideally we should implement
3630      * a naming scheme for Objects which are not DeviceStates, in
3631      * which case we can relax this restriction.
3632      */
3633     owner_dev = DEVICE(owner);
3634     vmstate_register_ram(mr, owner_dev);
3635 
3636     return true;
3637 }
3638 
3639 bool memory_region_init_ram_guest_memfd(MemoryRegion *mr,
3640                                         Object *owner,
3641                                         const char *name,
3642                                         uint64_t size,
3643                                         Error **errp)
3644 {
3645     DeviceState *owner_dev;
3646 
3647     if (!memory_region_init_ram_flags_nomigrate(mr, owner, name, size,
3648                                                 RAM_GUEST_MEMFD, errp)) {
3649         return false;
3650     }
3651     /* This will assert if owner is neither NULL nor a DeviceState.
3652      * We only want the owner here for the purposes of defining a
3653      * unique name for migration. TODO: Ideally we should implement
3654      * a naming scheme for Objects which are not DeviceStates, in
3655      * which case we can relax this restriction.
3656      */
3657     owner_dev = DEVICE(owner);
3658     vmstate_register_ram(mr, owner_dev);
3659 
3660     return true;
3661 }
3662 
3663 bool memory_region_init_rom(MemoryRegion *mr,
3664                             Object *owner,
3665                             const char *name,
3666                             uint64_t size,
3667                             Error **errp)
3668 {
3669     DeviceState *owner_dev;
3670 
3671     if (!memory_region_init_rom_nomigrate(mr, owner, name, size, errp)) {
3672         return false;
3673     }
3674     /* This will assert if owner is neither NULL nor a DeviceState.
3675      * We only want the owner here for the purposes of defining a
3676      * unique name for migration. TODO: Ideally we should implement
3677      * a naming scheme for Objects which are not DeviceStates, in
3678      * which case we can relax this restriction.
3679      */
3680     owner_dev = DEVICE(owner);
3681     vmstate_register_ram(mr, owner_dev);
3682 
3683     return true;
3684 }
3685 
3686 bool memory_region_init_rom_device(MemoryRegion *mr,
3687                                    Object *owner,
3688                                    const MemoryRegionOps *ops,
3689                                    void *opaque,
3690                                    const char *name,
3691                                    uint64_t size,
3692                                    Error **errp)
3693 {
3694     DeviceState *owner_dev;
3695 
3696     if (!memory_region_init_rom_device_nomigrate(mr, owner, ops, opaque,
3697                                                  name, size, errp)) {
3698         return false;
3699     }
3700     /* This will assert if owner is neither NULL nor a DeviceState.
3701      * We only want the owner here for the purposes of defining a
3702      * unique name for migration. TODO: Ideally we should implement
3703      * a naming scheme for Objects which are not DeviceStates, in
3704      * which case we can relax this restriction.
3705      */
3706     owner_dev = DEVICE(owner);
3707     vmstate_register_ram(mr, owner_dev);
3708 
3709     return true;
3710 }
3711 
3712 /*
3713  * Support system builds with CONFIG_FUZZ using a weak symbol and a stub for
3714  * the fuzz_dma_read_cb callback
3715  */
3716 #ifdef CONFIG_FUZZ
3717 void __attribute__((weak)) fuzz_dma_read_cb(size_t addr,
3718                       size_t len,
3719                       MemoryRegion *mr)
3720 {
3721 }
3722 #endif
3723 
3724 static const TypeInfo memory_region_info = {
3725     .parent             = TYPE_OBJECT,
3726     .name               = TYPE_MEMORY_REGION,
3727     .class_size         = sizeof(MemoryRegionClass),
3728     .instance_size      = sizeof(MemoryRegion),
3729     .instance_init      = memory_region_initfn,
3730     .instance_finalize  = memory_region_finalize,
3731 };
3732 
3733 static const TypeInfo iommu_memory_region_info = {
3734     .parent             = TYPE_MEMORY_REGION,
3735     .name               = TYPE_IOMMU_MEMORY_REGION,
3736     .class_size         = sizeof(IOMMUMemoryRegionClass),
3737     .instance_size      = sizeof(IOMMUMemoryRegion),
3738     .instance_init      = iommu_memory_region_initfn,
3739     .abstract           = true,
3740 };
3741 
3742 static const TypeInfo ram_discard_manager_info = {
3743     .parent             = TYPE_INTERFACE,
3744     .name               = TYPE_RAM_DISCARD_MANAGER,
3745     .class_size         = sizeof(RamDiscardManagerClass),
3746 };
3747 
3748 static void memory_register_types(void)
3749 {
3750     type_register_static(&memory_region_info);
3751     type_register_static(&iommu_memory_region_info);
3752     type_register_static(&ram_discard_manager_info);
3753 }
3754 
3755 type_init(memory_register_types)
3756