Merge remote-tracking branch 'remotes/rth/tags/pull-tcg-20200602' into staging
[qemu.git] / memory.c
1 /*
2 * Physical memory management
3 *
4 * Copyright 2011 Red Hat, Inc. and/or its affiliates
5 *
6 * Authors:
7 * Avi Kivity <avi@redhat.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 * 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 "qapi/error.h"
18 #include "cpu.h"
19 #include "exec/memory.h"
20 #include "exec/address-spaces.h"
21 #include "qapi/visitor.h"
22 #include "qemu/bitops.h"
23 #include "qemu/error-report.h"
24 #include "qemu/main-loop.h"
25 #include "qemu/qemu-print.h"
26 #include "qom/object.h"
27 #include "trace-root.h"
28
29 #include "exec/memory-internal.h"
30 #include "exec/ram_addr.h"
31 #include "sysemu/kvm.h"
32 #include "sysemu/runstate.h"
33 #include "sysemu/tcg.h"
34 #include "sysemu/accel.h"
35 #include "hw/boards.h"
36 #include "migration/vmstate.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 bool global_dirty_log;
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 return !memory_region_ioeventfd_before(a, b)
208 && !memory_region_ioeventfd_before(b, a);
209 }
210
211 /* Range of memory in the global map. Addresses are absolute. */
212 struct FlatRange {
213 MemoryRegion *mr;
214 hwaddr offset_in_region;
215 AddrRange addr;
216 uint8_t dirty_log_mask;
217 bool romd_mode;
218 bool readonly;
219 bool nonvolatile;
220 };
221
222 #define FOR_EACH_FLAT_RANGE(var, view) \
223 for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
224
225 static inline MemoryRegionSection
226 section_from_flat_range(FlatRange *fr, FlatView *fv)
227 {
228 return (MemoryRegionSection) {
229 .mr = fr->mr,
230 .fv = fv,
231 .offset_within_region = fr->offset_in_region,
232 .size = fr->addr.size,
233 .offset_within_address_space = int128_get64(fr->addr.start),
234 .readonly = fr->readonly,
235 .nonvolatile = fr->nonvolatile,
236 };
237 }
238
239 static bool flatrange_equal(FlatRange *a, FlatRange *b)
240 {
241 return a->mr == b->mr
242 && addrrange_equal(a->addr, b->addr)
243 && a->offset_in_region == b->offset_in_region
244 && a->romd_mode == b->romd_mode
245 && a->readonly == b->readonly
246 && a->nonvolatile == b->nonvolatile;
247 }
248
249 static FlatView *flatview_new(MemoryRegion *mr_root)
250 {
251 FlatView *view;
252
253 view = g_new0(FlatView, 1);
254 view->ref = 1;
255 view->root = mr_root;
256 memory_region_ref(mr_root);
257 trace_flatview_new(view, mr_root);
258
259 return view;
260 }
261
262 /* Insert a range into a given position. Caller is responsible for maintaining
263 * sorting order.
264 */
265 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
266 {
267 if (view->nr == view->nr_allocated) {
268 view->nr_allocated = MAX(2 * view->nr, 10);
269 view->ranges = g_realloc(view->ranges,
270 view->nr_allocated * sizeof(*view->ranges));
271 }
272 memmove(view->ranges + pos + 1, view->ranges + pos,
273 (view->nr - pos) * sizeof(FlatRange));
274 view->ranges[pos] = *range;
275 memory_region_ref(range->mr);
276 ++view->nr;
277 }
278
279 static void flatview_destroy(FlatView *view)
280 {
281 int i;
282
283 trace_flatview_destroy(view, view->root);
284 if (view->dispatch) {
285 address_space_dispatch_free(view->dispatch);
286 }
287 for (i = 0; i < view->nr; i++) {
288 memory_region_unref(view->ranges[i].mr);
289 }
290 g_free(view->ranges);
291 memory_region_unref(view->root);
292 g_free(view);
293 }
294
295 static bool flatview_ref(FlatView *view)
296 {
297 return atomic_fetch_inc_nonzero(&view->ref) > 0;
298 }
299
300 void flatview_unref(FlatView *view)
301 {
302 if (atomic_fetch_dec(&view->ref) == 1) {
303 trace_flatview_destroy_rcu(view, view->root);
304 assert(view->root);
305 call_rcu(view, flatview_destroy, rcu);
306 }
307 }
308
309 static bool can_merge(FlatRange *r1, FlatRange *r2)
310 {
311 return int128_eq(addrrange_end(r1->addr), r2->addr.start)
312 && r1->mr == r2->mr
313 && int128_eq(int128_add(int128_make64(r1->offset_in_region),
314 r1->addr.size),
315 int128_make64(r2->offset_in_region))
316 && r1->dirty_log_mask == r2->dirty_log_mask
317 && r1->romd_mode == r2->romd_mode
318 && r1->readonly == r2->readonly
319 && r1->nonvolatile == r2->nonvolatile;
320 }
321
322 /* Attempt to simplify a view by merging adjacent ranges */
323 static void flatview_simplify(FlatView *view)
324 {
325 unsigned i, j, k;
326
327 i = 0;
328 while (i < view->nr) {
329 j = i + 1;
330 while (j < view->nr
331 && can_merge(&view->ranges[j-1], &view->ranges[j])) {
332 int128_addto(&view->ranges[i].addr.size, view->ranges[j].addr.size);
333 ++j;
334 }
335 ++i;
336 for (k = i; k < j; k++) {
337 memory_region_unref(view->ranges[k].mr);
338 }
339 memmove(&view->ranges[i], &view->ranges[j],
340 (view->nr - j) * sizeof(view->ranges[j]));
341 view->nr -= j - i;
342 }
343 }
344
345 static bool memory_region_big_endian(MemoryRegion *mr)
346 {
347 #ifdef TARGET_WORDS_BIGENDIAN
348 return mr->ops->endianness != DEVICE_LITTLE_ENDIAN;
349 #else
350 return mr->ops->endianness == DEVICE_BIG_ENDIAN;
351 #endif
352 }
353
354 static void adjust_endianness(MemoryRegion *mr, uint64_t *data, MemOp op)
355 {
356 if ((op & MO_BSWAP) != devend_memop(mr->ops->endianness)) {
357 switch (op & MO_SIZE) {
358 case MO_8:
359 break;
360 case MO_16:
361 *data = bswap16(*data);
362 break;
363 case MO_32:
364 *data = bswap32(*data);
365 break;
366 case MO_64:
367 *data = bswap64(*data);
368 break;
369 default:
370 g_assert_not_reached();
371 }
372 }
373 }
374
375 static inline void memory_region_shift_read_access(uint64_t *value,
376 signed shift,
377 uint64_t mask,
378 uint64_t tmp)
379 {
380 if (shift >= 0) {
381 *value |= (tmp & mask) << shift;
382 } else {
383 *value |= (tmp & mask) >> -shift;
384 }
385 }
386
387 static inline uint64_t memory_region_shift_write_access(uint64_t *value,
388 signed shift,
389 uint64_t mask)
390 {
391 uint64_t tmp;
392
393 if (shift >= 0) {
394 tmp = (*value >> shift) & mask;
395 } else {
396 tmp = (*value << -shift) & mask;
397 }
398
399 return tmp;
400 }
401
402 static hwaddr memory_region_to_absolute_addr(MemoryRegion *mr, hwaddr offset)
403 {
404 MemoryRegion *root;
405 hwaddr abs_addr = offset;
406
407 abs_addr += mr->addr;
408 for (root = mr; root->container; ) {
409 root = root->container;
410 abs_addr += root->addr;
411 }
412
413 return abs_addr;
414 }
415
416 static int get_cpu_index(void)
417 {
418 if (current_cpu) {
419 return current_cpu->cpu_index;
420 }
421 return -1;
422 }
423
424 static MemTxResult memory_region_read_accessor(MemoryRegion *mr,
425 hwaddr addr,
426 uint64_t *value,
427 unsigned size,
428 signed shift,
429 uint64_t mask,
430 MemTxAttrs attrs)
431 {
432 uint64_t tmp;
433
434 tmp = mr->ops->read(mr->opaque, addr, size);
435 if (mr->subpage) {
436 trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
437 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_READ)) {
438 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
439 trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
440 }
441 memory_region_shift_read_access(value, shift, mask, tmp);
442 return MEMTX_OK;
443 }
444
445 static MemTxResult memory_region_read_with_attrs_accessor(MemoryRegion *mr,
446 hwaddr addr,
447 uint64_t *value,
448 unsigned size,
449 signed shift,
450 uint64_t mask,
451 MemTxAttrs attrs)
452 {
453 uint64_t tmp = 0;
454 MemTxResult r;
455
456 r = mr->ops->read_with_attrs(mr->opaque, addr, &tmp, size, attrs);
457 if (mr->subpage) {
458 trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
459 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_READ)) {
460 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
461 trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
462 }
463 memory_region_shift_read_access(value, shift, mask, tmp);
464 return r;
465 }
466
467 static MemTxResult memory_region_write_accessor(MemoryRegion *mr,
468 hwaddr addr,
469 uint64_t *value,
470 unsigned size,
471 signed shift,
472 uint64_t mask,
473 MemTxAttrs attrs)
474 {
475 uint64_t tmp = memory_region_shift_write_access(value, shift, mask);
476
477 if (mr->subpage) {
478 trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
479 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_WRITE)) {
480 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
481 trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
482 }
483 mr->ops->write(mr->opaque, addr, tmp, size);
484 return MEMTX_OK;
485 }
486
487 static MemTxResult memory_region_write_with_attrs_accessor(MemoryRegion *mr,
488 hwaddr addr,
489 uint64_t *value,
490 unsigned size,
491 signed shift,
492 uint64_t mask,
493 MemTxAttrs attrs)
494 {
495 uint64_t tmp = memory_region_shift_write_access(value, shift, mask);
496
497 if (mr->subpage) {
498 trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
499 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_WRITE)) {
500 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
501 trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
502 }
503 return mr->ops->write_with_attrs(mr->opaque, addr, tmp, size, attrs);
504 }
505
506 static MemTxResult access_with_adjusted_size(hwaddr addr,
507 uint64_t *value,
508 unsigned size,
509 unsigned access_size_min,
510 unsigned access_size_max,
511 MemTxResult (*access_fn)
512 (MemoryRegion *mr,
513 hwaddr addr,
514 uint64_t *value,
515 unsigned size,
516 signed shift,
517 uint64_t mask,
518 MemTxAttrs attrs),
519 MemoryRegion *mr,
520 MemTxAttrs attrs)
521 {
522 uint64_t access_mask;
523 unsigned access_size;
524 unsigned i;
525 MemTxResult r = MEMTX_OK;
526
527 if (!access_size_min) {
528 access_size_min = 1;
529 }
530 if (!access_size_max) {
531 access_size_max = 4;
532 }
533
534 /* FIXME: support unaligned access? */
535 access_size = MAX(MIN(size, access_size_max), access_size_min);
536 access_mask = MAKE_64BIT_MASK(0, access_size * 8);
537 if (memory_region_big_endian(mr)) {
538 for (i = 0; i < size; i += access_size) {
539 r |= access_fn(mr, addr + i, value, access_size,
540 (size - access_size - i) * 8, access_mask, attrs);
541 }
542 } else {
543 for (i = 0; i < size; i += access_size) {
544 r |= access_fn(mr, addr + i, value, access_size, i * 8,
545 access_mask, attrs);
546 }
547 }
548 return r;
549 }
550
551 static AddressSpace *memory_region_to_address_space(MemoryRegion *mr)
552 {
553 AddressSpace *as;
554
555 while (mr->container) {
556 mr = mr->container;
557 }
558 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
559 if (mr == as->root) {
560 return as;
561 }
562 }
563 return NULL;
564 }
565
566 /* Render a memory region into the global view. Ranges in @view obscure
567 * ranges in @mr.
568 */
569 static void render_memory_region(FlatView *view,
570 MemoryRegion *mr,
571 Int128 base,
572 AddrRange clip,
573 bool readonly,
574 bool nonvolatile)
575 {
576 MemoryRegion *subregion;
577 unsigned i;
578 hwaddr offset_in_region;
579 Int128 remain;
580 Int128 now;
581 FlatRange fr;
582 AddrRange tmp;
583
584 if (!mr->enabled) {
585 return;
586 }
587
588 int128_addto(&base, int128_make64(mr->addr));
589 readonly |= mr->readonly;
590 nonvolatile |= mr->nonvolatile;
591
592 tmp = addrrange_make(base, mr->size);
593
594 if (!addrrange_intersects(tmp, clip)) {
595 return;
596 }
597
598 clip = addrrange_intersection(tmp, clip);
599
600 if (mr->alias) {
601 int128_subfrom(&base, int128_make64(mr->alias->addr));
602 int128_subfrom(&base, int128_make64(mr->alias_offset));
603 render_memory_region(view, mr->alias, base, clip,
604 readonly, nonvolatile);
605 return;
606 }
607
608 /* Render subregions in priority order. */
609 QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
610 render_memory_region(view, subregion, base, clip,
611 readonly, nonvolatile);
612 }
613
614 if (!mr->terminates) {
615 return;
616 }
617
618 offset_in_region = int128_get64(int128_sub(clip.start, base));
619 base = clip.start;
620 remain = clip.size;
621
622 fr.mr = mr;
623 fr.dirty_log_mask = memory_region_get_dirty_log_mask(mr);
624 fr.romd_mode = mr->romd_mode;
625 fr.readonly = readonly;
626 fr.nonvolatile = nonvolatile;
627
628 /* Render the region itself into any gaps left by the current view. */
629 for (i = 0; i < view->nr && int128_nz(remain); ++i) {
630 if (int128_ge(base, addrrange_end(view->ranges[i].addr))) {
631 continue;
632 }
633 if (int128_lt(base, view->ranges[i].addr.start)) {
634 now = int128_min(remain,
635 int128_sub(view->ranges[i].addr.start, base));
636 fr.offset_in_region = offset_in_region;
637 fr.addr = addrrange_make(base, now);
638 flatview_insert(view, i, &fr);
639 ++i;
640 int128_addto(&base, now);
641 offset_in_region += int128_get64(now);
642 int128_subfrom(&remain, now);
643 }
644 now = int128_sub(int128_min(int128_add(base, remain),
645 addrrange_end(view->ranges[i].addr)),
646 base);
647 int128_addto(&base, now);
648 offset_in_region += int128_get64(now);
649 int128_subfrom(&remain, now);
650 }
651 if (int128_nz(remain)) {
652 fr.offset_in_region = offset_in_region;
653 fr.addr = addrrange_make(base, remain);
654 flatview_insert(view, i, &fr);
655 }
656 }
657
658 static MemoryRegion *memory_region_get_flatview_root(MemoryRegion *mr)
659 {
660 while (mr->enabled) {
661 if (mr->alias) {
662 if (!mr->alias_offset && int128_ge(mr->size, mr->alias->size)) {
663 /* The alias is included in its entirety. Use it as
664 * the "real" root, so that we can share more FlatViews.
665 */
666 mr = mr->alias;
667 continue;
668 }
669 } else if (!mr->terminates) {
670 unsigned int found = 0;
671 MemoryRegion *child, *next = NULL;
672 QTAILQ_FOREACH(child, &mr->subregions, subregions_link) {
673 if (child->enabled) {
674 if (++found > 1) {
675 next = NULL;
676 break;
677 }
678 if (!child->addr && int128_ge(mr->size, child->size)) {
679 /* A child is included in its entirety. If it's the only
680 * enabled one, use it in the hope of finding an alias down the
681 * way. This will also let us share FlatViews.
682 */
683 next = child;
684 }
685 }
686 }
687 if (found == 0) {
688 return NULL;
689 }
690 if (next) {
691 mr = next;
692 continue;
693 }
694 }
695
696 return mr;
697 }
698
699 return NULL;
700 }
701
702 /* Render a memory topology into a list of disjoint absolute ranges. */
703 static FlatView *generate_memory_topology(MemoryRegion *mr)
704 {
705 int i;
706 FlatView *view;
707
708 view = flatview_new(mr);
709
710 if (mr) {
711 render_memory_region(view, mr, int128_zero(),
712 addrrange_make(int128_zero(), int128_2_64()),
713 false, false);
714 }
715 flatview_simplify(view);
716
717 view->dispatch = address_space_dispatch_new(view);
718 for (i = 0; i < view->nr; i++) {
719 MemoryRegionSection mrs =
720 section_from_flat_range(&view->ranges[i], view);
721 flatview_add_to_dispatch(view, &mrs);
722 }
723 address_space_dispatch_compact(view->dispatch);
724 g_hash_table_replace(flat_views, mr, view);
725
726 return view;
727 }
728
729 static void address_space_add_del_ioeventfds(AddressSpace *as,
730 MemoryRegionIoeventfd *fds_new,
731 unsigned fds_new_nb,
732 MemoryRegionIoeventfd *fds_old,
733 unsigned fds_old_nb)
734 {
735 unsigned iold, inew;
736 MemoryRegionIoeventfd *fd;
737 MemoryRegionSection section;
738
739 /* Generate a symmetric difference of the old and new fd sets, adding
740 * and deleting as necessary.
741 */
742
743 iold = inew = 0;
744 while (iold < fds_old_nb || inew < fds_new_nb) {
745 if (iold < fds_old_nb
746 && (inew == fds_new_nb
747 || memory_region_ioeventfd_before(&fds_old[iold],
748 &fds_new[inew]))) {
749 fd = &fds_old[iold];
750 section = (MemoryRegionSection) {
751 .fv = address_space_to_flatview(as),
752 .offset_within_address_space = int128_get64(fd->addr.start),
753 .size = fd->addr.size,
754 };
755 MEMORY_LISTENER_CALL(as, eventfd_del, Forward, &section,
756 fd->match_data, fd->data, fd->e);
757 ++iold;
758 } else if (inew < fds_new_nb
759 && (iold == fds_old_nb
760 || memory_region_ioeventfd_before(&fds_new[inew],
761 &fds_old[iold]))) {
762 fd = &fds_new[inew];
763 section = (MemoryRegionSection) {
764 .fv = address_space_to_flatview(as),
765 .offset_within_address_space = int128_get64(fd->addr.start),
766 .size = fd->addr.size,
767 };
768 MEMORY_LISTENER_CALL(as, eventfd_add, Reverse, &section,
769 fd->match_data, fd->data, fd->e);
770 ++inew;
771 } else {
772 ++iold;
773 ++inew;
774 }
775 }
776 }
777
778 FlatView *address_space_get_flatview(AddressSpace *as)
779 {
780 FlatView *view;
781
782 RCU_READ_LOCK_GUARD();
783 do {
784 view = address_space_to_flatview(as);
785 /* If somebody has replaced as->current_map concurrently,
786 * flatview_ref returns false.
787 */
788 } while (!flatview_ref(view));
789 return view;
790 }
791
792 static void address_space_update_ioeventfds(AddressSpace *as)
793 {
794 FlatView *view;
795 FlatRange *fr;
796 unsigned ioeventfd_nb = 0;
797 unsigned ioeventfd_max;
798 MemoryRegionIoeventfd *ioeventfds;
799 AddrRange tmp;
800 unsigned i;
801
802 /*
803 * It is likely that the number of ioeventfds hasn't changed much, so use
804 * the previous size as the starting value, with some headroom to avoid
805 * gratuitous reallocations.
806 */
807 ioeventfd_max = QEMU_ALIGN_UP(as->ioeventfd_nb, 4);
808 ioeventfds = g_new(MemoryRegionIoeventfd, ioeventfd_max);
809
810 view = address_space_get_flatview(as);
811 FOR_EACH_FLAT_RANGE(fr, view) {
812 for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
813 tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
814 int128_sub(fr->addr.start,
815 int128_make64(fr->offset_in_region)));
816 if (addrrange_intersects(fr->addr, tmp)) {
817 ++ioeventfd_nb;
818 if (ioeventfd_nb > ioeventfd_max) {
819 ioeventfd_max = MAX(ioeventfd_max * 2, 4);
820 ioeventfds = g_realloc(ioeventfds,
821 ioeventfd_max * sizeof(*ioeventfds));
822 }
823 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
824 ioeventfds[ioeventfd_nb-1].addr = tmp;
825 }
826 }
827 }
828
829 address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
830 as->ioeventfds, as->ioeventfd_nb);
831
832 g_free(as->ioeventfds);
833 as->ioeventfds = ioeventfds;
834 as->ioeventfd_nb = ioeventfd_nb;
835 flatview_unref(view);
836 }
837
838 /*
839 * Notify the memory listeners about the coalesced IO change events of
840 * range `cmr'. Only the part that has intersection of the specified
841 * FlatRange will be sent.
842 */
843 static void flat_range_coalesced_io_notify(FlatRange *fr, AddressSpace *as,
844 CoalescedMemoryRange *cmr, bool add)
845 {
846 AddrRange tmp;
847
848 tmp = addrrange_shift(cmr->addr,
849 int128_sub(fr->addr.start,
850 int128_make64(fr->offset_in_region)));
851 if (!addrrange_intersects(tmp, fr->addr)) {
852 return;
853 }
854 tmp = addrrange_intersection(tmp, fr->addr);
855
856 if (add) {
857 MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, coalesced_io_add,
858 int128_get64(tmp.start),
859 int128_get64(tmp.size));
860 } else {
861 MEMORY_LISTENER_UPDATE_REGION(fr, as, Reverse, coalesced_io_del,
862 int128_get64(tmp.start),
863 int128_get64(tmp.size));
864 }
865 }
866
867 static void flat_range_coalesced_io_del(FlatRange *fr, AddressSpace *as)
868 {
869 CoalescedMemoryRange *cmr;
870
871 QTAILQ_FOREACH(cmr, &fr->mr->coalesced, link) {
872 flat_range_coalesced_io_notify(fr, as, cmr, false);
873 }
874 }
875
876 static void flat_range_coalesced_io_add(FlatRange *fr, AddressSpace *as)
877 {
878 MemoryRegion *mr = fr->mr;
879 CoalescedMemoryRange *cmr;
880
881 if (QTAILQ_EMPTY(&mr->coalesced)) {
882 return;
883 }
884
885 QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
886 flat_range_coalesced_io_notify(fr, as, cmr, true);
887 }
888 }
889
890 static void address_space_update_topology_pass(AddressSpace *as,
891 const FlatView *old_view,
892 const FlatView *new_view,
893 bool adding)
894 {
895 unsigned iold, inew;
896 FlatRange *frold, *frnew;
897
898 /* Generate a symmetric difference of the old and new memory maps.
899 * Kill ranges in the old map, and instantiate ranges in the new map.
900 */
901 iold = inew = 0;
902 while (iold < old_view->nr || inew < new_view->nr) {
903 if (iold < old_view->nr) {
904 frold = &old_view->ranges[iold];
905 } else {
906 frold = NULL;
907 }
908 if (inew < new_view->nr) {
909 frnew = &new_view->ranges[inew];
910 } else {
911 frnew = NULL;
912 }
913
914 if (frold
915 && (!frnew
916 || int128_lt(frold->addr.start, frnew->addr.start)
917 || (int128_eq(frold->addr.start, frnew->addr.start)
918 && !flatrange_equal(frold, frnew)))) {
919 /* In old but not in new, or in both but attributes changed. */
920
921 if (!adding) {
922 flat_range_coalesced_io_del(frold, as);
923 MEMORY_LISTENER_UPDATE_REGION(frold, as, Reverse, region_del);
924 }
925
926 ++iold;
927 } else if (frold && frnew && flatrange_equal(frold, frnew)) {
928 /* In both and unchanged (except logging may have changed) */
929
930 if (adding) {
931 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_nop);
932 if (frnew->dirty_log_mask & ~frold->dirty_log_mask) {
933 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, log_start,
934 frold->dirty_log_mask,
935 frnew->dirty_log_mask);
936 }
937 if (frold->dirty_log_mask & ~frnew->dirty_log_mask) {
938 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Reverse, log_stop,
939 frold->dirty_log_mask,
940 frnew->dirty_log_mask);
941 }
942 }
943
944 ++iold;
945 ++inew;
946 } else {
947 /* In new */
948
949 if (adding) {
950 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_add);
951 flat_range_coalesced_io_add(frnew, as);
952 }
953
954 ++inew;
955 }
956 }
957 }
958
959 static void flatviews_init(void)
960 {
961 static FlatView *empty_view;
962
963 if (flat_views) {
964 return;
965 }
966
967 flat_views = g_hash_table_new_full(g_direct_hash, g_direct_equal, NULL,
968 (GDestroyNotify) flatview_unref);
969 if (!empty_view) {
970 empty_view = generate_memory_topology(NULL);
971 /* We keep it alive forever in the global variable. */
972 flatview_ref(empty_view);
973 } else {
974 g_hash_table_replace(flat_views, NULL, empty_view);
975 flatview_ref(empty_view);
976 }
977 }
978
979 static void flatviews_reset(void)
980 {
981 AddressSpace *as;
982
983 if (flat_views) {
984 g_hash_table_unref(flat_views);
985 flat_views = NULL;
986 }
987 flatviews_init();
988
989 /* Render unique FVs */
990 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
991 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
992
993 if (g_hash_table_lookup(flat_views, physmr)) {
994 continue;
995 }
996
997 generate_memory_topology(physmr);
998 }
999 }
1000
1001 static void address_space_set_flatview(AddressSpace *as)
1002 {
1003 FlatView *old_view = address_space_to_flatview(as);
1004 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1005 FlatView *new_view = g_hash_table_lookup(flat_views, physmr);
1006
1007 assert(new_view);
1008
1009 if (old_view == new_view) {
1010 return;
1011 }
1012
1013 if (old_view) {
1014 flatview_ref(old_view);
1015 }
1016
1017 flatview_ref(new_view);
1018
1019 if (!QTAILQ_EMPTY(&as->listeners)) {
1020 FlatView tmpview = { .nr = 0 }, *old_view2 = old_view;
1021
1022 if (!old_view2) {
1023 old_view2 = &tmpview;
1024 }
1025 address_space_update_topology_pass(as, old_view2, new_view, false);
1026 address_space_update_topology_pass(as, old_view2, new_view, true);
1027 }
1028
1029 /* Writes are protected by the BQL. */
1030 atomic_rcu_set(&as->current_map, new_view);
1031 if (old_view) {
1032 flatview_unref(old_view);
1033 }
1034
1035 /* Note that all the old MemoryRegions are still alive up to this
1036 * point. This relieves most MemoryListeners from the need to
1037 * ref/unref the MemoryRegions they get---unless they use them
1038 * outside the iothread mutex, in which case precise reference
1039 * counting is necessary.
1040 */
1041 if (old_view) {
1042 flatview_unref(old_view);
1043 }
1044 }
1045
1046 static void address_space_update_topology(AddressSpace *as)
1047 {
1048 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1049
1050 flatviews_init();
1051 if (!g_hash_table_lookup(flat_views, physmr)) {
1052 generate_memory_topology(physmr);
1053 }
1054 address_space_set_flatview(as);
1055 }
1056
1057 void memory_region_transaction_begin(void)
1058 {
1059 qemu_flush_coalesced_mmio_buffer();
1060 ++memory_region_transaction_depth;
1061 }
1062
1063 void memory_region_transaction_commit(void)
1064 {
1065 AddressSpace *as;
1066
1067 assert(memory_region_transaction_depth);
1068 assert(qemu_mutex_iothread_locked());
1069
1070 --memory_region_transaction_depth;
1071 if (!memory_region_transaction_depth) {
1072 if (memory_region_update_pending) {
1073 flatviews_reset();
1074
1075 MEMORY_LISTENER_CALL_GLOBAL(begin, Forward);
1076
1077 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1078 address_space_set_flatview(as);
1079 address_space_update_ioeventfds(as);
1080 }
1081 memory_region_update_pending = false;
1082 ioeventfd_update_pending = false;
1083 MEMORY_LISTENER_CALL_GLOBAL(commit, Forward);
1084 } else if (ioeventfd_update_pending) {
1085 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1086 address_space_update_ioeventfds(as);
1087 }
1088 ioeventfd_update_pending = false;
1089 }
1090 }
1091 }
1092
1093 static void memory_region_destructor_none(MemoryRegion *mr)
1094 {
1095 }
1096
1097 static void memory_region_destructor_ram(MemoryRegion *mr)
1098 {
1099 qemu_ram_free(mr->ram_block);
1100 }
1101
1102 static bool memory_region_need_escape(char c)
1103 {
1104 return c == '/' || c == '[' || c == '\\' || c == ']';
1105 }
1106
1107 static char *memory_region_escape_name(const char *name)
1108 {
1109 const char *p;
1110 char *escaped, *q;
1111 uint8_t c;
1112 size_t bytes = 0;
1113
1114 for (p = name; *p; p++) {
1115 bytes += memory_region_need_escape(*p) ? 4 : 1;
1116 }
1117 if (bytes == p - name) {
1118 return g_memdup(name, bytes + 1);
1119 }
1120
1121 escaped = g_malloc(bytes + 1);
1122 for (p = name, q = escaped; *p; p++) {
1123 c = *p;
1124 if (unlikely(memory_region_need_escape(c))) {
1125 *q++ = '\\';
1126 *q++ = 'x';
1127 *q++ = "0123456789abcdef"[c >> 4];
1128 c = "0123456789abcdef"[c & 15];
1129 }
1130 *q++ = c;
1131 }
1132 *q = 0;
1133 return escaped;
1134 }
1135
1136 static void memory_region_do_init(MemoryRegion *mr,
1137 Object *owner,
1138 const char *name,
1139 uint64_t size)
1140 {
1141 mr->size = int128_make64(size);
1142 if (size == UINT64_MAX) {
1143 mr->size = int128_2_64();
1144 }
1145 mr->name = g_strdup(name);
1146 mr->owner = owner;
1147 mr->ram_block = NULL;
1148
1149 if (name) {
1150 char *escaped_name = memory_region_escape_name(name);
1151 char *name_array = g_strdup_printf("%s[*]", escaped_name);
1152
1153 if (!owner) {
1154 owner = container_get(qdev_get_machine(), "/unattached");
1155 }
1156
1157 object_property_add_child(owner, name_array, OBJECT(mr));
1158 object_unref(OBJECT(mr));
1159 g_free(name_array);
1160 g_free(escaped_name);
1161 }
1162 }
1163
1164 void memory_region_init(MemoryRegion *mr,
1165 Object *owner,
1166 const char *name,
1167 uint64_t size)
1168 {
1169 object_initialize(mr, sizeof(*mr), TYPE_MEMORY_REGION);
1170 memory_region_do_init(mr, owner, name, size);
1171 }
1172
1173 static void memory_region_get_container(Object *obj, Visitor *v,
1174 const char *name, void *opaque,
1175 Error **errp)
1176 {
1177 MemoryRegion *mr = MEMORY_REGION(obj);
1178 char *path = (char *)"";
1179
1180 if (mr->container) {
1181 path = object_get_canonical_path(OBJECT(mr->container));
1182 }
1183 visit_type_str(v, name, &path, errp);
1184 if (mr->container) {
1185 g_free(path);
1186 }
1187 }
1188
1189 static Object *memory_region_resolve_container(Object *obj, void *opaque,
1190 const char *part)
1191 {
1192 MemoryRegion *mr = MEMORY_REGION(obj);
1193
1194 return OBJECT(mr->container);
1195 }
1196
1197 static void memory_region_get_priority(Object *obj, Visitor *v,
1198 const char *name, void *opaque,
1199 Error **errp)
1200 {
1201 MemoryRegion *mr = MEMORY_REGION(obj);
1202 int32_t value = mr->priority;
1203
1204 visit_type_int32(v, name, &value, errp);
1205 }
1206
1207 static void memory_region_get_size(Object *obj, Visitor *v, const char *name,
1208 void *opaque, Error **errp)
1209 {
1210 MemoryRegion *mr = MEMORY_REGION(obj);
1211 uint64_t value = memory_region_size(mr);
1212
1213 visit_type_uint64(v, name, &value, errp);
1214 }
1215
1216 static void memory_region_initfn(Object *obj)
1217 {
1218 MemoryRegion *mr = MEMORY_REGION(obj);
1219 ObjectProperty *op;
1220
1221 mr->ops = &unassigned_mem_ops;
1222 mr->enabled = true;
1223 mr->romd_mode = true;
1224 mr->global_locking = true;
1225 mr->destructor = memory_region_destructor_none;
1226 QTAILQ_INIT(&mr->subregions);
1227 QTAILQ_INIT(&mr->coalesced);
1228
1229 op = object_property_add(OBJECT(mr), "container",
1230 "link<" TYPE_MEMORY_REGION ">",
1231 memory_region_get_container,
1232 NULL, /* memory_region_set_container */
1233 NULL, NULL);
1234 op->resolve = memory_region_resolve_container;
1235
1236 object_property_add_uint64_ptr(OBJECT(mr), "addr",
1237 &mr->addr, OBJ_PROP_FLAG_READ);
1238 object_property_add(OBJECT(mr), "priority", "uint32",
1239 memory_region_get_priority,
1240 NULL, /* memory_region_set_priority */
1241 NULL, NULL);
1242 object_property_add(OBJECT(mr), "size", "uint64",
1243 memory_region_get_size,
1244 NULL, /* memory_region_set_size, */
1245 NULL, NULL);
1246 }
1247
1248 static void iommu_memory_region_initfn(Object *obj)
1249 {
1250 MemoryRegion *mr = MEMORY_REGION(obj);
1251
1252 mr->is_iommu = true;
1253 }
1254
1255 static uint64_t unassigned_mem_read(void *opaque, hwaddr addr,
1256 unsigned size)
1257 {
1258 #ifdef DEBUG_UNASSIGNED
1259 printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
1260 #endif
1261 return 0;
1262 }
1263
1264 static void unassigned_mem_write(void *opaque, hwaddr addr,
1265 uint64_t val, unsigned size)
1266 {
1267 #ifdef DEBUG_UNASSIGNED
1268 printf("Unassigned mem write " TARGET_FMT_plx " = 0x%"PRIx64"\n", addr, val);
1269 #endif
1270 }
1271
1272 static bool unassigned_mem_accepts(void *opaque, hwaddr addr,
1273 unsigned size, bool is_write,
1274 MemTxAttrs attrs)
1275 {
1276 return false;
1277 }
1278
1279 const MemoryRegionOps unassigned_mem_ops = {
1280 .valid.accepts = unassigned_mem_accepts,
1281 .endianness = DEVICE_NATIVE_ENDIAN,
1282 };
1283
1284 static uint64_t memory_region_ram_device_read(void *opaque,
1285 hwaddr addr, unsigned size)
1286 {
1287 MemoryRegion *mr = opaque;
1288 uint64_t data = (uint64_t)~0;
1289
1290 switch (size) {
1291 case 1:
1292 data = *(uint8_t *)(mr->ram_block->host + addr);
1293 break;
1294 case 2:
1295 data = *(uint16_t *)(mr->ram_block->host + addr);
1296 break;
1297 case 4:
1298 data = *(uint32_t *)(mr->ram_block->host + addr);
1299 break;
1300 case 8:
1301 data = *(uint64_t *)(mr->ram_block->host + addr);
1302 break;
1303 }
1304
1305 trace_memory_region_ram_device_read(get_cpu_index(), mr, addr, data, size);
1306
1307 return data;
1308 }
1309
1310 static void memory_region_ram_device_write(void *opaque, hwaddr addr,
1311 uint64_t data, unsigned size)
1312 {
1313 MemoryRegion *mr = opaque;
1314
1315 trace_memory_region_ram_device_write(get_cpu_index(), mr, addr, data, size);
1316
1317 switch (size) {
1318 case 1:
1319 *(uint8_t *)(mr->ram_block->host + addr) = (uint8_t)data;
1320 break;
1321 case 2:
1322 *(uint16_t *)(mr->ram_block->host + addr) = (uint16_t)data;
1323 break;
1324 case 4:
1325 *(uint32_t *)(mr->ram_block->host + addr) = (uint32_t)data;
1326 break;
1327 case 8:
1328 *(uint64_t *)(mr->ram_block->host + addr) = data;
1329 break;
1330 }
1331 }
1332
1333 static const MemoryRegionOps ram_device_mem_ops = {
1334 .read = memory_region_ram_device_read,
1335 .write = memory_region_ram_device_write,
1336 .endianness = DEVICE_HOST_ENDIAN,
1337 .valid = {
1338 .min_access_size = 1,
1339 .max_access_size = 8,
1340 .unaligned = true,
1341 },
1342 .impl = {
1343 .min_access_size = 1,
1344 .max_access_size = 8,
1345 .unaligned = true,
1346 },
1347 };
1348
1349 bool memory_region_access_valid(MemoryRegion *mr,
1350 hwaddr addr,
1351 unsigned size,
1352 bool is_write,
1353 MemTxAttrs attrs)
1354 {
1355 int access_size_min, access_size_max;
1356 int access_size, i;
1357
1358 if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1359 return false;
1360 }
1361
1362 if (!mr->ops->valid.accepts) {
1363 return true;
1364 }
1365
1366 access_size_min = mr->ops->valid.min_access_size;
1367 if (!mr->ops->valid.min_access_size) {
1368 access_size_min = 1;
1369 }
1370
1371 access_size_max = mr->ops->valid.max_access_size;
1372 if (!mr->ops->valid.max_access_size) {
1373 access_size_max = 4;
1374 }
1375
1376 access_size = MAX(MIN(size, access_size_max), access_size_min);
1377 for (i = 0; i < size; i += access_size) {
1378 if (!mr->ops->valid.accepts(mr->opaque, addr + i, access_size,
1379 is_write, attrs)) {
1380 return false;
1381 }
1382 }
1383
1384 return true;
1385 }
1386
1387 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1388 hwaddr addr,
1389 uint64_t *pval,
1390 unsigned size,
1391 MemTxAttrs attrs)
1392 {
1393 *pval = 0;
1394
1395 if (mr->ops->read) {
1396 return access_with_adjusted_size(addr, pval, size,
1397 mr->ops->impl.min_access_size,
1398 mr->ops->impl.max_access_size,
1399 memory_region_read_accessor,
1400 mr, attrs);
1401 } else {
1402 return access_with_adjusted_size(addr, pval, size,
1403 mr->ops->impl.min_access_size,
1404 mr->ops->impl.max_access_size,
1405 memory_region_read_with_attrs_accessor,
1406 mr, attrs);
1407 }
1408 }
1409
1410 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1411 hwaddr addr,
1412 uint64_t *pval,
1413 MemOp op,
1414 MemTxAttrs attrs)
1415 {
1416 unsigned size = memop_size(op);
1417 MemTxResult r;
1418
1419 if (!memory_region_access_valid(mr, addr, size, false, attrs)) {
1420 *pval = unassigned_mem_read(mr, addr, size);
1421 return MEMTX_DECODE_ERROR;
1422 }
1423
1424 r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1425 adjust_endianness(mr, pval, op);
1426 return r;
1427 }
1428
1429 /* Return true if an eventfd was signalled */
1430 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1431 hwaddr addr,
1432 uint64_t data,
1433 unsigned size,
1434 MemTxAttrs attrs)
1435 {
1436 MemoryRegionIoeventfd ioeventfd = {
1437 .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1438 .data = data,
1439 };
1440 unsigned i;
1441
1442 for (i = 0; i < mr->ioeventfd_nb; i++) {
1443 ioeventfd.match_data = mr->ioeventfds[i].match_data;
1444 ioeventfd.e = mr->ioeventfds[i].e;
1445
1446 if (memory_region_ioeventfd_equal(&ioeventfd, &mr->ioeventfds[i])) {
1447 event_notifier_set(ioeventfd.e);
1448 return true;
1449 }
1450 }
1451
1452 return false;
1453 }
1454
1455 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1456 hwaddr addr,
1457 uint64_t data,
1458 MemOp op,
1459 MemTxAttrs attrs)
1460 {
1461 unsigned size = memop_size(op);
1462
1463 if (!memory_region_access_valid(mr, addr, size, true, attrs)) {
1464 unassigned_mem_write(mr, addr, data, size);
1465 return MEMTX_DECODE_ERROR;
1466 }
1467
1468 adjust_endianness(mr, &data, op);
1469
1470 if ((!kvm_eventfds_enabled()) &&
1471 memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1472 return MEMTX_OK;
1473 }
1474
1475 if (mr->ops->write) {
1476 return access_with_adjusted_size(addr, &data, size,
1477 mr->ops->impl.min_access_size,
1478 mr->ops->impl.max_access_size,
1479 memory_region_write_accessor, mr,
1480 attrs);
1481 } else {
1482 return
1483 access_with_adjusted_size(addr, &data, size,
1484 mr->ops->impl.min_access_size,
1485 mr->ops->impl.max_access_size,
1486 memory_region_write_with_attrs_accessor,
1487 mr, attrs);
1488 }
1489 }
1490
1491 void memory_region_init_io(MemoryRegion *mr,
1492 Object *owner,
1493 const MemoryRegionOps *ops,
1494 void *opaque,
1495 const char *name,
1496 uint64_t size)
1497 {
1498 memory_region_init(mr, owner, name, size);
1499 mr->ops = ops ? ops : &unassigned_mem_ops;
1500 mr->opaque = opaque;
1501 mr->terminates = true;
1502 }
1503
1504 void memory_region_init_ram_nomigrate(MemoryRegion *mr,
1505 Object *owner,
1506 const char *name,
1507 uint64_t size,
1508 Error **errp)
1509 {
1510 memory_region_init_ram_shared_nomigrate(mr, owner, name, size, false, errp);
1511 }
1512
1513 void memory_region_init_ram_shared_nomigrate(MemoryRegion *mr,
1514 Object *owner,
1515 const char *name,
1516 uint64_t size,
1517 bool share,
1518 Error **errp)
1519 {
1520 Error *err = NULL;
1521 memory_region_init(mr, owner, name, size);
1522 mr->ram = true;
1523 mr->terminates = true;
1524 mr->destructor = memory_region_destructor_ram;
1525 mr->ram_block = qemu_ram_alloc(size, share, mr, &err);
1526 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1527 if (err) {
1528 mr->size = int128_zero();
1529 object_unparent(OBJECT(mr));
1530 error_propagate(errp, err);
1531 }
1532 }
1533
1534 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1535 Object *owner,
1536 const char *name,
1537 uint64_t size,
1538 uint64_t max_size,
1539 void (*resized)(const char*,
1540 uint64_t length,
1541 void *host),
1542 Error **errp)
1543 {
1544 Error *err = NULL;
1545 memory_region_init(mr, owner, name, size);
1546 mr->ram = true;
1547 mr->terminates = true;
1548 mr->destructor = memory_region_destructor_ram;
1549 mr->ram_block = qemu_ram_alloc_resizeable(size, max_size, resized,
1550 mr, &err);
1551 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1552 if (err) {
1553 mr->size = int128_zero();
1554 object_unparent(OBJECT(mr));
1555 error_propagate(errp, err);
1556 }
1557 }
1558
1559 #ifdef CONFIG_POSIX
1560 void memory_region_init_ram_from_file(MemoryRegion *mr,
1561 struct Object *owner,
1562 const char *name,
1563 uint64_t size,
1564 uint64_t align,
1565 uint32_t ram_flags,
1566 const char *path,
1567 Error **errp)
1568 {
1569 Error *err = NULL;
1570 memory_region_init(mr, owner, name, size);
1571 mr->ram = true;
1572 mr->terminates = true;
1573 mr->destructor = memory_region_destructor_ram;
1574 mr->align = align;
1575 mr->ram_block = qemu_ram_alloc_from_file(size, mr, ram_flags, path, &err);
1576 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1577 if (err) {
1578 mr->size = int128_zero();
1579 object_unparent(OBJECT(mr));
1580 error_propagate(errp, err);
1581 }
1582 }
1583
1584 void memory_region_init_ram_from_fd(MemoryRegion *mr,
1585 struct Object *owner,
1586 const char *name,
1587 uint64_t size,
1588 bool share,
1589 int fd,
1590 Error **errp)
1591 {
1592 Error *err = NULL;
1593 memory_region_init(mr, owner, name, size);
1594 mr->ram = true;
1595 mr->terminates = true;
1596 mr->destructor = memory_region_destructor_ram;
1597 mr->ram_block = qemu_ram_alloc_from_fd(size, mr,
1598 share ? RAM_SHARED : 0,
1599 fd, &err);
1600 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1601 if (err) {
1602 mr->size = int128_zero();
1603 object_unparent(OBJECT(mr));
1604 error_propagate(errp, err);
1605 }
1606 }
1607 #endif
1608
1609 void memory_region_init_ram_ptr(MemoryRegion *mr,
1610 Object *owner,
1611 const char *name,
1612 uint64_t size,
1613 void *ptr)
1614 {
1615 memory_region_init(mr, owner, name, size);
1616 mr->ram = true;
1617 mr->terminates = true;
1618 mr->destructor = memory_region_destructor_ram;
1619 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1620
1621 /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL. */
1622 assert(ptr != NULL);
1623 mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1624 }
1625
1626 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
1627 Object *owner,
1628 const char *name,
1629 uint64_t size,
1630 void *ptr)
1631 {
1632 memory_region_init(mr, owner, name, size);
1633 mr->ram = true;
1634 mr->terminates = true;
1635 mr->ram_device = true;
1636 mr->ops = &ram_device_mem_ops;
1637 mr->opaque = mr;
1638 mr->destructor = memory_region_destructor_ram;
1639 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1640 /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL. */
1641 assert(ptr != NULL);
1642 mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1643 }
1644
1645 void memory_region_init_alias(MemoryRegion *mr,
1646 Object *owner,
1647 const char *name,
1648 MemoryRegion *orig,
1649 hwaddr offset,
1650 uint64_t size)
1651 {
1652 memory_region_init(mr, owner, name, size);
1653 mr->alias = orig;
1654 mr->alias_offset = offset;
1655 }
1656
1657 void memory_region_init_rom_nomigrate(MemoryRegion *mr,
1658 struct Object *owner,
1659 const char *name,
1660 uint64_t size,
1661 Error **errp)
1662 {
1663 memory_region_init_ram_shared_nomigrate(mr, owner, name, size, false, errp);
1664 mr->readonly = true;
1665 }
1666
1667 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1668 Object *owner,
1669 const MemoryRegionOps *ops,
1670 void *opaque,
1671 const char *name,
1672 uint64_t size,
1673 Error **errp)
1674 {
1675 Error *err = NULL;
1676 assert(ops);
1677 memory_region_init(mr, owner, name, size);
1678 mr->ops = ops;
1679 mr->opaque = opaque;
1680 mr->terminates = true;
1681 mr->rom_device = true;
1682 mr->destructor = memory_region_destructor_ram;
1683 mr->ram_block = qemu_ram_alloc(size, false, mr, &err);
1684 if (err) {
1685 mr->size = int128_zero();
1686 object_unparent(OBJECT(mr));
1687 error_propagate(errp, err);
1688 }
1689 }
1690
1691 void memory_region_init_iommu(void *_iommu_mr,
1692 size_t instance_size,
1693 const char *mrtypename,
1694 Object *owner,
1695 const char *name,
1696 uint64_t size)
1697 {
1698 struct IOMMUMemoryRegion *iommu_mr;
1699 struct MemoryRegion *mr;
1700
1701 object_initialize(_iommu_mr, instance_size, mrtypename);
1702 mr = MEMORY_REGION(_iommu_mr);
1703 memory_region_do_init(mr, owner, name, size);
1704 iommu_mr = IOMMU_MEMORY_REGION(mr);
1705 mr->terminates = true; /* then re-forwards */
1706 QLIST_INIT(&iommu_mr->iommu_notify);
1707 iommu_mr->iommu_notify_flags = IOMMU_NOTIFIER_NONE;
1708 }
1709
1710 static void memory_region_finalize(Object *obj)
1711 {
1712 MemoryRegion *mr = MEMORY_REGION(obj);
1713
1714 assert(!mr->container);
1715
1716 /* We know the region is not visible in any address space (it
1717 * does not have a container and cannot be a root either because
1718 * it has no references, so we can blindly clear mr->enabled.
1719 * memory_region_set_enabled instead could trigger a transaction
1720 * and cause an infinite loop.
1721 */
1722 mr->enabled = false;
1723 memory_region_transaction_begin();
1724 while (!QTAILQ_EMPTY(&mr->subregions)) {
1725 MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1726 memory_region_del_subregion(mr, subregion);
1727 }
1728 memory_region_transaction_commit();
1729
1730 mr->destructor(mr);
1731 memory_region_clear_coalescing(mr);
1732 g_free((char *)mr->name);
1733 g_free(mr->ioeventfds);
1734 }
1735
1736 Object *memory_region_owner(MemoryRegion *mr)
1737 {
1738 Object *obj = OBJECT(mr);
1739 return obj->parent;
1740 }
1741
1742 void memory_region_ref(MemoryRegion *mr)
1743 {
1744 /* MMIO callbacks most likely will access data that belongs
1745 * to the owner, hence the need to ref/unref the owner whenever
1746 * the memory region is in use.
1747 *
1748 * The memory region is a child of its owner. As long as the
1749 * owner doesn't call unparent itself on the memory region,
1750 * ref-ing the owner will also keep the memory region alive.
1751 * Memory regions without an owner are supposed to never go away;
1752 * we do not ref/unref them because it slows down DMA sensibly.
1753 */
1754 if (mr && mr->owner) {
1755 object_ref(mr->owner);
1756 }
1757 }
1758
1759 void memory_region_unref(MemoryRegion *mr)
1760 {
1761 if (mr && mr->owner) {
1762 object_unref(mr->owner);
1763 }
1764 }
1765
1766 uint64_t memory_region_size(MemoryRegion *mr)
1767 {
1768 if (int128_eq(mr->size, int128_2_64())) {
1769 return UINT64_MAX;
1770 }
1771 return int128_get64(mr->size);
1772 }
1773
1774 const char *memory_region_name(const MemoryRegion *mr)
1775 {
1776 if (!mr->name) {
1777 ((MemoryRegion *)mr)->name =
1778 object_get_canonical_path_component(OBJECT(mr));
1779 }
1780 return mr->name;
1781 }
1782
1783 bool memory_region_is_ram_device(MemoryRegion *mr)
1784 {
1785 return mr->ram_device;
1786 }
1787
1788 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1789 {
1790 uint8_t mask = mr->dirty_log_mask;
1791 if (global_dirty_log && mr->ram_block) {
1792 mask |= (1 << DIRTY_MEMORY_MIGRATION);
1793 }
1794 return mask;
1795 }
1796
1797 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1798 {
1799 return memory_region_get_dirty_log_mask(mr) & (1 << client);
1800 }
1801
1802 static int memory_region_update_iommu_notify_flags(IOMMUMemoryRegion *iommu_mr,
1803 Error **errp)
1804 {
1805 IOMMUNotifierFlag flags = IOMMU_NOTIFIER_NONE;
1806 IOMMUNotifier *iommu_notifier;
1807 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1808 int ret = 0;
1809
1810 IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
1811 flags |= iommu_notifier->notifier_flags;
1812 }
1813
1814 if (flags != iommu_mr->iommu_notify_flags && imrc->notify_flag_changed) {
1815 ret = imrc->notify_flag_changed(iommu_mr,
1816 iommu_mr->iommu_notify_flags,
1817 flags, errp);
1818 }
1819
1820 if (!ret) {
1821 iommu_mr->iommu_notify_flags = flags;
1822 }
1823 return ret;
1824 }
1825
1826 int memory_region_register_iommu_notifier(MemoryRegion *mr,
1827 IOMMUNotifier *n, Error **errp)
1828 {
1829 IOMMUMemoryRegion *iommu_mr;
1830 int ret;
1831
1832 if (mr->alias) {
1833 return memory_region_register_iommu_notifier(mr->alias, n, errp);
1834 }
1835
1836 /* We need to register for at least one bitfield */
1837 iommu_mr = IOMMU_MEMORY_REGION(mr);
1838 assert(n->notifier_flags != IOMMU_NOTIFIER_NONE);
1839 assert(n->start <= n->end);
1840 assert(n->iommu_idx >= 0 &&
1841 n->iommu_idx < memory_region_iommu_num_indexes(iommu_mr));
1842
1843 QLIST_INSERT_HEAD(&iommu_mr->iommu_notify, n, node);
1844 ret = memory_region_update_iommu_notify_flags(iommu_mr, errp);
1845 if (ret) {
1846 QLIST_REMOVE(n, node);
1847 }
1848 return ret;
1849 }
1850
1851 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr)
1852 {
1853 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1854
1855 if (imrc->get_min_page_size) {
1856 return imrc->get_min_page_size(iommu_mr);
1857 }
1858 return TARGET_PAGE_SIZE;
1859 }
1860
1861 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
1862 {
1863 MemoryRegion *mr = MEMORY_REGION(iommu_mr);
1864 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1865 hwaddr addr, granularity;
1866 IOMMUTLBEntry iotlb;
1867
1868 /* If the IOMMU has its own replay callback, override */
1869 if (imrc->replay) {
1870 imrc->replay(iommu_mr, n);
1871 return;
1872 }
1873
1874 granularity = memory_region_iommu_get_min_page_size(iommu_mr);
1875
1876 for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1877 iotlb = imrc->translate(iommu_mr, addr, IOMMU_NONE, n->iommu_idx);
1878 if (iotlb.perm != IOMMU_NONE) {
1879 n->notify(n, &iotlb);
1880 }
1881
1882 /* if (2^64 - MR size) < granularity, it's possible to get an
1883 * infinite loop here. This should catch such a wraparound */
1884 if ((addr + granularity) < addr) {
1885 break;
1886 }
1887 }
1888 }
1889
1890 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1891 IOMMUNotifier *n)
1892 {
1893 IOMMUMemoryRegion *iommu_mr;
1894
1895 if (mr->alias) {
1896 memory_region_unregister_iommu_notifier(mr->alias, n);
1897 return;
1898 }
1899 QLIST_REMOVE(n, node);
1900 iommu_mr = IOMMU_MEMORY_REGION(mr);
1901 memory_region_update_iommu_notify_flags(iommu_mr, NULL);
1902 }
1903
1904 void memory_region_notify_one(IOMMUNotifier *notifier,
1905 IOMMUTLBEntry *entry)
1906 {
1907 IOMMUNotifierFlag request_flags;
1908 hwaddr entry_end = entry->iova + entry->addr_mask;
1909
1910 /*
1911 * Skip the notification if the notification does not overlap
1912 * with registered range.
1913 */
1914 if (notifier->start > entry_end || notifier->end < entry->iova) {
1915 return;
1916 }
1917
1918 assert(entry->iova >= notifier->start && entry_end <= notifier->end);
1919
1920 if (entry->perm & IOMMU_RW) {
1921 request_flags = IOMMU_NOTIFIER_MAP;
1922 } else {
1923 request_flags = IOMMU_NOTIFIER_UNMAP;
1924 }
1925
1926 if (notifier->notifier_flags & request_flags) {
1927 notifier->notify(notifier, entry);
1928 }
1929 }
1930
1931 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
1932 int iommu_idx,
1933 IOMMUTLBEntry entry)
1934 {
1935 IOMMUNotifier *iommu_notifier;
1936
1937 assert(memory_region_is_iommu(MEMORY_REGION(iommu_mr)));
1938
1939 IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
1940 if (iommu_notifier->iommu_idx == iommu_idx) {
1941 memory_region_notify_one(iommu_notifier, &entry);
1942 }
1943 }
1944 }
1945
1946 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
1947 enum IOMMUMemoryRegionAttr attr,
1948 void *data)
1949 {
1950 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1951
1952 if (!imrc->get_attr) {
1953 return -EINVAL;
1954 }
1955
1956 return imrc->get_attr(iommu_mr, attr, data);
1957 }
1958
1959 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr,
1960 MemTxAttrs attrs)
1961 {
1962 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1963
1964 if (!imrc->attrs_to_index) {
1965 return 0;
1966 }
1967
1968 return imrc->attrs_to_index(iommu_mr, attrs);
1969 }
1970
1971 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr)
1972 {
1973 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1974
1975 if (!imrc->num_indexes) {
1976 return 1;
1977 }
1978
1979 return imrc->num_indexes(iommu_mr);
1980 }
1981
1982 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1983 {
1984 uint8_t mask = 1 << client;
1985 uint8_t old_logging;
1986
1987 assert(client == DIRTY_MEMORY_VGA);
1988 old_logging = mr->vga_logging_count;
1989 mr->vga_logging_count += log ? 1 : -1;
1990 if (!!old_logging == !!mr->vga_logging_count) {
1991 return;
1992 }
1993
1994 memory_region_transaction_begin();
1995 mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1996 memory_region_update_pending |= mr->enabled;
1997 memory_region_transaction_commit();
1998 }
1999
2000 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
2001 hwaddr size)
2002 {
2003 assert(mr->ram_block);
2004 cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr,
2005 size,
2006 memory_region_get_dirty_log_mask(mr));
2007 }
2008
2009 static void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
2010 {
2011 MemoryListener *listener;
2012 AddressSpace *as;
2013 FlatView *view;
2014 FlatRange *fr;
2015
2016 /* If the same address space has multiple log_sync listeners, we
2017 * visit that address space's FlatView multiple times. But because
2018 * log_sync listeners are rare, it's still cheaper than walking each
2019 * address space once.
2020 */
2021 QTAILQ_FOREACH(listener, &memory_listeners, link) {
2022 if (!listener->log_sync) {
2023 continue;
2024 }
2025 as = listener->address_space;
2026 view = address_space_get_flatview(as);
2027 FOR_EACH_FLAT_RANGE(fr, view) {
2028 if (fr->dirty_log_mask && (!mr || fr->mr == mr)) {
2029 MemoryRegionSection mrs = section_from_flat_range(fr, view);
2030 listener->log_sync(listener, &mrs);
2031 }
2032 }
2033 flatview_unref(view);
2034 }
2035 }
2036
2037 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start,
2038 hwaddr len)
2039 {
2040 MemoryRegionSection mrs;
2041 MemoryListener *listener;
2042 AddressSpace *as;
2043 FlatView *view;
2044 FlatRange *fr;
2045 hwaddr sec_start, sec_end, sec_size;
2046
2047 QTAILQ_FOREACH(listener, &memory_listeners, link) {
2048 if (!listener->log_clear) {
2049 continue;
2050 }
2051 as = listener->address_space;
2052 view = address_space_get_flatview(as);
2053 FOR_EACH_FLAT_RANGE(fr, view) {
2054 if (!fr->dirty_log_mask || fr->mr != mr) {
2055 /*
2056 * Clear dirty bitmap operation only applies to those
2057 * regions whose dirty logging is at least enabled
2058 */
2059 continue;
2060 }
2061
2062 mrs = section_from_flat_range(fr, view);
2063
2064 sec_start = MAX(mrs.offset_within_region, start);
2065 sec_end = mrs.offset_within_region + int128_get64(mrs.size);
2066 sec_end = MIN(sec_end, start + len);
2067
2068 if (sec_start >= sec_end) {
2069 /*
2070 * If this memory region section has no intersection
2071 * with the requested range, skip.
2072 */
2073 continue;
2074 }
2075
2076 /* Valid case; shrink the section if needed */
2077 mrs.offset_within_address_space +=
2078 sec_start - mrs.offset_within_region;
2079 mrs.offset_within_region = sec_start;
2080 sec_size = sec_end - sec_start;
2081 mrs.size = int128_make64(sec_size);
2082 listener->log_clear(listener, &mrs);
2083 }
2084 flatview_unref(view);
2085 }
2086 }
2087
2088 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
2089 hwaddr addr,
2090 hwaddr size,
2091 unsigned client)
2092 {
2093 DirtyBitmapSnapshot *snapshot;
2094 assert(mr->ram_block);
2095 memory_region_sync_dirty_bitmap(mr);
2096 snapshot = cpu_physical_memory_snapshot_and_clear_dirty(mr, addr, size, client);
2097 memory_global_after_dirty_log_sync();
2098 return snapshot;
2099 }
2100
2101 bool memory_region_snapshot_get_dirty(MemoryRegion *mr, DirtyBitmapSnapshot *snap,
2102 hwaddr addr, hwaddr size)
2103 {
2104 assert(mr->ram_block);
2105 return cpu_physical_memory_snapshot_get_dirty(snap,
2106 memory_region_get_ram_addr(mr) + addr, size);
2107 }
2108
2109 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
2110 {
2111 if (mr->readonly != readonly) {
2112 memory_region_transaction_begin();
2113 mr->readonly = readonly;
2114 memory_region_update_pending |= mr->enabled;
2115 memory_region_transaction_commit();
2116 }
2117 }
2118
2119 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile)
2120 {
2121 if (mr->nonvolatile != nonvolatile) {
2122 memory_region_transaction_begin();
2123 mr->nonvolatile = nonvolatile;
2124 memory_region_update_pending |= mr->enabled;
2125 memory_region_transaction_commit();
2126 }
2127 }
2128
2129 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
2130 {
2131 if (mr->romd_mode != romd_mode) {
2132 memory_region_transaction_begin();
2133 mr->romd_mode = romd_mode;
2134 memory_region_update_pending |= mr->enabled;
2135 memory_region_transaction_commit();
2136 }
2137 }
2138
2139 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
2140 hwaddr size, unsigned client)
2141 {
2142 assert(mr->ram_block);
2143 cpu_physical_memory_test_and_clear_dirty(
2144 memory_region_get_ram_addr(mr) + addr, size, client);
2145 }
2146
2147 int memory_region_get_fd(MemoryRegion *mr)
2148 {
2149 int fd;
2150
2151 RCU_READ_LOCK_GUARD();
2152 while (mr->alias) {
2153 mr = mr->alias;
2154 }
2155 fd = mr->ram_block->fd;
2156
2157 return fd;
2158 }
2159
2160 void *memory_region_get_ram_ptr(MemoryRegion *mr)
2161 {
2162 void *ptr;
2163 uint64_t offset = 0;
2164
2165 RCU_READ_LOCK_GUARD();
2166 while (mr->alias) {
2167 offset += mr->alias_offset;
2168 mr = mr->alias;
2169 }
2170 assert(mr->ram_block);
2171 ptr = qemu_map_ram_ptr(mr->ram_block, offset);
2172
2173 return ptr;
2174 }
2175
2176 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset)
2177 {
2178 RAMBlock *block;
2179
2180 block = qemu_ram_block_from_host(ptr, false, offset);
2181 if (!block) {
2182 return NULL;
2183 }
2184
2185 return block->mr;
2186 }
2187
2188 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr)
2189 {
2190 return mr->ram_block ? mr->ram_block->offset : RAM_ADDR_INVALID;
2191 }
2192
2193 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
2194 {
2195 assert(mr->ram_block);
2196
2197 qemu_ram_resize(mr->ram_block, newsize, errp);
2198 }
2199
2200
2201 void memory_region_do_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size)
2202 {
2203 /*
2204 * Might be extended case needed to cover
2205 * different types of memory regions
2206 */
2207 if (mr->ram_block && mr->dirty_log_mask) {
2208 qemu_ram_writeback(mr->ram_block, addr, size);
2209 }
2210 }
2211
2212 /*
2213 * Call proper memory listeners about the change on the newly
2214 * added/removed CoalescedMemoryRange.
2215 */
2216 static void memory_region_update_coalesced_range(MemoryRegion *mr,
2217 CoalescedMemoryRange *cmr,
2218 bool add)
2219 {
2220 AddressSpace *as;
2221 FlatView *view;
2222 FlatRange *fr;
2223
2224 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2225 view = address_space_get_flatview(as);
2226 FOR_EACH_FLAT_RANGE(fr, view) {
2227 if (fr->mr == mr) {
2228 flat_range_coalesced_io_notify(fr, as, cmr, add);
2229 }
2230 }
2231 flatview_unref(view);
2232 }
2233 }
2234
2235 void memory_region_set_coalescing(MemoryRegion *mr)
2236 {
2237 memory_region_clear_coalescing(mr);
2238 memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
2239 }
2240
2241 void memory_region_add_coalescing(MemoryRegion *mr,
2242 hwaddr offset,
2243 uint64_t size)
2244 {
2245 CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
2246
2247 cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
2248 QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
2249 memory_region_update_coalesced_range(mr, cmr, true);
2250 memory_region_set_flush_coalesced(mr);
2251 }
2252
2253 void memory_region_clear_coalescing(MemoryRegion *mr)
2254 {
2255 CoalescedMemoryRange *cmr;
2256
2257 if (QTAILQ_EMPTY(&mr->coalesced)) {
2258 return;
2259 }
2260
2261 qemu_flush_coalesced_mmio_buffer();
2262 mr->flush_coalesced_mmio = false;
2263
2264 while (!QTAILQ_EMPTY(&mr->coalesced)) {
2265 cmr = QTAILQ_FIRST(&mr->coalesced);
2266 QTAILQ_REMOVE(&mr->coalesced, cmr, link);
2267 memory_region_update_coalesced_range(mr, cmr, false);
2268 g_free(cmr);
2269 }
2270 }
2271
2272 void memory_region_set_flush_coalesced(MemoryRegion *mr)
2273 {
2274 mr->flush_coalesced_mmio = true;
2275 }
2276
2277 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
2278 {
2279 qemu_flush_coalesced_mmio_buffer();
2280 if (QTAILQ_EMPTY(&mr->coalesced)) {
2281 mr->flush_coalesced_mmio = false;
2282 }
2283 }
2284
2285 void memory_region_clear_global_locking(MemoryRegion *mr)
2286 {
2287 mr->global_locking = false;
2288 }
2289
2290 static bool userspace_eventfd_warning;
2291
2292 void memory_region_add_eventfd(MemoryRegion *mr,
2293 hwaddr addr,
2294 unsigned size,
2295 bool match_data,
2296 uint64_t data,
2297 EventNotifier *e)
2298 {
2299 MemoryRegionIoeventfd mrfd = {
2300 .addr.start = int128_make64(addr),
2301 .addr.size = int128_make64(size),
2302 .match_data = match_data,
2303 .data = data,
2304 .e = e,
2305 };
2306 unsigned i;
2307
2308 if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
2309 userspace_eventfd_warning))) {
2310 userspace_eventfd_warning = true;
2311 error_report("Using eventfd without MMIO binding in KVM. "
2312 "Suboptimal performance expected");
2313 }
2314
2315 if (size) {
2316 adjust_endianness(mr, &mrfd.data, size_memop(size) | MO_TE);
2317 }
2318 memory_region_transaction_begin();
2319 for (i = 0; i < mr->ioeventfd_nb; ++i) {
2320 if (memory_region_ioeventfd_before(&mrfd, &mr->ioeventfds[i])) {
2321 break;
2322 }
2323 }
2324 ++mr->ioeventfd_nb;
2325 mr->ioeventfds = g_realloc(mr->ioeventfds,
2326 sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
2327 memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
2328 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
2329 mr->ioeventfds[i] = mrfd;
2330 ioeventfd_update_pending |= mr->enabled;
2331 memory_region_transaction_commit();
2332 }
2333
2334 void memory_region_del_eventfd(MemoryRegion *mr,
2335 hwaddr addr,
2336 unsigned size,
2337 bool match_data,
2338 uint64_t data,
2339 EventNotifier *e)
2340 {
2341 MemoryRegionIoeventfd mrfd = {
2342 .addr.start = int128_make64(addr),
2343 .addr.size = int128_make64(size),
2344 .match_data = match_data,
2345 .data = data,
2346 .e = e,
2347 };
2348 unsigned i;
2349
2350 if (size) {
2351 adjust_endianness(mr, &mrfd.data, size_memop(size) | MO_TE);
2352 }
2353 memory_region_transaction_begin();
2354 for (i = 0; i < mr->ioeventfd_nb; ++i) {
2355 if (memory_region_ioeventfd_equal(&mrfd, &mr->ioeventfds[i])) {
2356 break;
2357 }
2358 }
2359 assert(i != mr->ioeventfd_nb);
2360 memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
2361 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
2362 --mr->ioeventfd_nb;
2363 mr->ioeventfds = g_realloc(mr->ioeventfds,
2364 sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
2365 ioeventfd_update_pending |= mr->enabled;
2366 memory_region_transaction_commit();
2367 }
2368
2369 static void memory_region_update_container_subregions(MemoryRegion *subregion)
2370 {
2371 MemoryRegion *mr = subregion->container;
2372 MemoryRegion *other;
2373
2374 memory_region_transaction_begin();
2375
2376 memory_region_ref(subregion);
2377 QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
2378 if (subregion->priority >= other->priority) {
2379 QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
2380 goto done;
2381 }
2382 }
2383 QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
2384 done:
2385 memory_region_update_pending |= mr->enabled && subregion->enabled;
2386 memory_region_transaction_commit();
2387 }
2388
2389 static void memory_region_add_subregion_common(MemoryRegion *mr,
2390 hwaddr offset,
2391 MemoryRegion *subregion)
2392 {
2393 assert(!subregion->container);
2394 subregion->container = mr;
2395 subregion->addr = offset;
2396 memory_region_update_container_subregions(subregion);
2397 }
2398
2399 void memory_region_add_subregion(MemoryRegion *mr,
2400 hwaddr offset,
2401 MemoryRegion *subregion)
2402 {
2403 subregion->priority = 0;
2404 memory_region_add_subregion_common(mr, offset, subregion);
2405 }
2406
2407 void memory_region_add_subregion_overlap(MemoryRegion *mr,
2408 hwaddr offset,
2409 MemoryRegion *subregion,
2410 int priority)
2411 {
2412 subregion->priority = priority;
2413 memory_region_add_subregion_common(mr, offset, subregion);
2414 }
2415
2416 void memory_region_del_subregion(MemoryRegion *mr,
2417 MemoryRegion *subregion)
2418 {
2419 memory_region_transaction_begin();
2420 assert(subregion->container == mr);
2421 subregion->container = NULL;
2422 QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
2423 memory_region_unref(subregion);
2424 memory_region_update_pending |= mr->enabled && subregion->enabled;
2425 memory_region_transaction_commit();
2426 }
2427
2428 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
2429 {
2430 if (enabled == mr->enabled) {
2431 return;
2432 }
2433 memory_region_transaction_begin();
2434 mr->enabled = enabled;
2435 memory_region_update_pending = true;
2436 memory_region_transaction_commit();
2437 }
2438
2439 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
2440 {
2441 Int128 s = int128_make64(size);
2442
2443 if (size == UINT64_MAX) {
2444 s = int128_2_64();
2445 }
2446 if (int128_eq(s, mr->size)) {
2447 return;
2448 }
2449 memory_region_transaction_begin();
2450 mr->size = s;
2451 memory_region_update_pending = true;
2452 memory_region_transaction_commit();
2453 }
2454
2455 static void memory_region_readd_subregion(MemoryRegion *mr)
2456 {
2457 MemoryRegion *container = mr->container;
2458
2459 if (container) {
2460 memory_region_transaction_begin();
2461 memory_region_ref(mr);
2462 memory_region_del_subregion(container, mr);
2463 mr->container = container;
2464 memory_region_update_container_subregions(mr);
2465 memory_region_unref(mr);
2466 memory_region_transaction_commit();
2467 }
2468 }
2469
2470 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
2471 {
2472 if (addr != mr->addr) {
2473 mr->addr = addr;
2474 memory_region_readd_subregion(mr);
2475 }
2476 }
2477
2478 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
2479 {
2480 assert(mr->alias);
2481
2482 if (offset == mr->alias_offset) {
2483 return;
2484 }
2485
2486 memory_region_transaction_begin();
2487 mr->alias_offset = offset;
2488 memory_region_update_pending |= mr->enabled;
2489 memory_region_transaction_commit();
2490 }
2491
2492 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
2493 {
2494 return mr->align;
2495 }
2496
2497 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
2498 {
2499 const AddrRange *addr = addr_;
2500 const FlatRange *fr = fr_;
2501
2502 if (int128_le(addrrange_end(*addr), fr->addr.start)) {
2503 return -1;
2504 } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
2505 return 1;
2506 }
2507 return 0;
2508 }
2509
2510 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
2511 {
2512 return bsearch(&addr, view->ranges, view->nr,
2513 sizeof(FlatRange), cmp_flatrange_addr);
2514 }
2515
2516 bool memory_region_is_mapped(MemoryRegion *mr)
2517 {
2518 return mr->container ? true : false;
2519 }
2520
2521 /* Same as memory_region_find, but it does not add a reference to the
2522 * returned region. It must be called from an RCU critical section.
2523 */
2524 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2525 hwaddr addr, uint64_t size)
2526 {
2527 MemoryRegionSection ret = { .mr = NULL };
2528 MemoryRegion *root;
2529 AddressSpace *as;
2530 AddrRange range;
2531 FlatView *view;
2532 FlatRange *fr;
2533
2534 addr += mr->addr;
2535 for (root = mr; root->container; ) {
2536 root = root->container;
2537 addr += root->addr;
2538 }
2539
2540 as = memory_region_to_address_space(root);
2541 if (!as) {
2542 return ret;
2543 }
2544 range = addrrange_make(int128_make64(addr), int128_make64(size));
2545
2546 view = address_space_to_flatview(as);
2547 fr = flatview_lookup(view, range);
2548 if (!fr) {
2549 return ret;
2550 }
2551
2552 while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2553 --fr;
2554 }
2555
2556 ret.mr = fr->mr;
2557 ret.fv = view;
2558 range = addrrange_intersection(range, fr->addr);
2559 ret.offset_within_region = fr->offset_in_region;
2560 ret.offset_within_region += int128_get64(int128_sub(range.start,
2561 fr->addr.start));
2562 ret.size = range.size;
2563 ret.offset_within_address_space = int128_get64(range.start);
2564 ret.readonly = fr->readonly;
2565 ret.nonvolatile = fr->nonvolatile;
2566 return ret;
2567 }
2568
2569 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2570 hwaddr addr, uint64_t size)
2571 {
2572 MemoryRegionSection ret;
2573 RCU_READ_LOCK_GUARD();
2574 ret = memory_region_find_rcu(mr, addr, size);
2575 if (ret.mr) {
2576 memory_region_ref(ret.mr);
2577 }
2578 return ret;
2579 }
2580
2581 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2582 {
2583 MemoryRegion *mr;
2584
2585 RCU_READ_LOCK_GUARD();
2586 mr = memory_region_find_rcu(container, addr, 1).mr;
2587 return mr && mr != container;
2588 }
2589
2590 void memory_global_dirty_log_sync(void)
2591 {
2592 memory_region_sync_dirty_bitmap(NULL);
2593 }
2594
2595 void memory_global_after_dirty_log_sync(void)
2596 {
2597 MEMORY_LISTENER_CALL_GLOBAL(log_global_after_sync, Forward);
2598 }
2599
2600 static VMChangeStateEntry *vmstate_change;
2601
2602 void memory_global_dirty_log_start(void)
2603 {
2604 if (vmstate_change) {
2605 qemu_del_vm_change_state_handler(vmstate_change);
2606 vmstate_change = NULL;
2607 }
2608
2609 global_dirty_log = true;
2610
2611 MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2612
2613 /* Refresh DIRTY_MEMORY_MIGRATION bit. */
2614 memory_region_transaction_begin();
2615 memory_region_update_pending = true;
2616 memory_region_transaction_commit();
2617 }
2618
2619 static void memory_global_dirty_log_do_stop(void)
2620 {
2621 global_dirty_log = false;
2622
2623 /* Refresh DIRTY_MEMORY_MIGRATION bit. */
2624 memory_region_transaction_begin();
2625 memory_region_update_pending = true;
2626 memory_region_transaction_commit();
2627
2628 MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2629 }
2630
2631 static void memory_vm_change_state_handler(void *opaque, int running,
2632 RunState state)
2633 {
2634 if (running) {
2635 memory_global_dirty_log_do_stop();
2636
2637 if (vmstate_change) {
2638 qemu_del_vm_change_state_handler(vmstate_change);
2639 vmstate_change = NULL;
2640 }
2641 }
2642 }
2643
2644 void memory_global_dirty_log_stop(void)
2645 {
2646 if (!runstate_is_running()) {
2647 if (vmstate_change) {
2648 return;
2649 }
2650 vmstate_change = qemu_add_vm_change_state_handler(
2651 memory_vm_change_state_handler, NULL);
2652 return;
2653 }
2654
2655 memory_global_dirty_log_do_stop();
2656 }
2657
2658 static void listener_add_address_space(MemoryListener *listener,
2659 AddressSpace *as)
2660 {
2661 FlatView *view;
2662 FlatRange *fr;
2663
2664 if (listener->begin) {
2665 listener->begin(listener);
2666 }
2667 if (global_dirty_log) {
2668 if (listener->log_global_start) {
2669 listener->log_global_start(listener);
2670 }
2671 }
2672
2673 view = address_space_get_flatview(as);
2674 FOR_EACH_FLAT_RANGE(fr, view) {
2675 MemoryRegionSection section = section_from_flat_range(fr, view);
2676
2677 if (listener->region_add) {
2678 listener->region_add(listener, &section);
2679 }
2680 if (fr->dirty_log_mask && listener->log_start) {
2681 listener->log_start(listener, &section, 0, fr->dirty_log_mask);
2682 }
2683 }
2684 if (listener->commit) {
2685 listener->commit(listener);
2686 }
2687 flatview_unref(view);
2688 }
2689
2690 static void listener_del_address_space(MemoryListener *listener,
2691 AddressSpace *as)
2692 {
2693 FlatView *view;
2694 FlatRange *fr;
2695
2696 if (listener->begin) {
2697 listener->begin(listener);
2698 }
2699 view = address_space_get_flatview(as);
2700 FOR_EACH_FLAT_RANGE(fr, view) {
2701 MemoryRegionSection section = section_from_flat_range(fr, view);
2702
2703 if (fr->dirty_log_mask && listener->log_stop) {
2704 listener->log_stop(listener, &section, fr->dirty_log_mask, 0);
2705 }
2706 if (listener->region_del) {
2707 listener->region_del(listener, &section);
2708 }
2709 }
2710 if (listener->commit) {
2711 listener->commit(listener);
2712 }
2713 flatview_unref(view);
2714 }
2715
2716 void memory_listener_register(MemoryListener *listener, AddressSpace *as)
2717 {
2718 MemoryListener *other = NULL;
2719
2720 listener->address_space = as;
2721 if (QTAILQ_EMPTY(&memory_listeners)
2722 || listener->priority >= QTAILQ_LAST(&memory_listeners)->priority) {
2723 QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
2724 } else {
2725 QTAILQ_FOREACH(other, &memory_listeners, link) {
2726 if (listener->priority < other->priority) {
2727 break;
2728 }
2729 }
2730 QTAILQ_INSERT_BEFORE(other, listener, link);
2731 }
2732
2733 if (QTAILQ_EMPTY(&as->listeners)
2734 || listener->priority >= QTAILQ_LAST(&as->listeners)->priority) {
2735 QTAILQ_INSERT_TAIL(&as->listeners, listener, link_as);
2736 } else {
2737 QTAILQ_FOREACH(other, &as->listeners, link_as) {
2738 if (listener->priority < other->priority) {
2739 break;
2740 }
2741 }
2742 QTAILQ_INSERT_BEFORE(other, listener, link_as);
2743 }
2744
2745 listener_add_address_space(listener, as);
2746 }
2747
2748 void memory_listener_unregister(MemoryListener *listener)
2749 {
2750 if (!listener->address_space) {
2751 return;
2752 }
2753
2754 listener_del_address_space(listener, listener->address_space);
2755 QTAILQ_REMOVE(&memory_listeners, listener, link);
2756 QTAILQ_REMOVE(&listener->address_space->listeners, listener, link_as);
2757 listener->address_space = NULL;
2758 }
2759
2760 void address_space_remove_listeners(AddressSpace *as)
2761 {
2762 while (!QTAILQ_EMPTY(&as->listeners)) {
2763 memory_listener_unregister(QTAILQ_FIRST(&as->listeners));
2764 }
2765 }
2766
2767 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
2768 {
2769 memory_region_ref(root);
2770 as->root = root;
2771 as->current_map = NULL;
2772 as->ioeventfd_nb = 0;
2773 as->ioeventfds = NULL;
2774 QTAILQ_INIT(&as->listeners);
2775 QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
2776 as->name = g_strdup(name ? name : "anonymous");
2777 address_space_update_topology(as);
2778 address_space_update_ioeventfds(as);
2779 }
2780
2781 static void do_address_space_destroy(AddressSpace *as)
2782 {
2783 assert(QTAILQ_EMPTY(&as->listeners));
2784
2785 flatview_unref(as->current_map);
2786 g_free(as->name);
2787 g_free(as->ioeventfds);
2788 memory_region_unref(as->root);
2789 }
2790
2791 void address_space_destroy(AddressSpace *as)
2792 {
2793 MemoryRegion *root = as->root;
2794
2795 /* Flush out anything from MemoryListeners listening in on this */
2796 memory_region_transaction_begin();
2797 as->root = NULL;
2798 memory_region_transaction_commit();
2799 QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
2800
2801 /* At this point, as->dispatch and as->current_map are dummy
2802 * entries that the guest should never use. Wait for the old
2803 * values to expire before freeing the data.
2804 */
2805 as->root = root;
2806 call_rcu(as, do_address_space_destroy, rcu);
2807 }
2808
2809 static const char *memory_region_type(MemoryRegion *mr)
2810 {
2811 if (mr->alias) {
2812 return memory_region_type(mr->alias);
2813 }
2814 if (memory_region_is_ram_device(mr)) {
2815 return "ramd";
2816 } else if (memory_region_is_romd(mr)) {
2817 return "romd";
2818 } else if (memory_region_is_rom(mr)) {
2819 return "rom";
2820 } else if (memory_region_is_ram(mr)) {
2821 return "ram";
2822 } else {
2823 return "i/o";
2824 }
2825 }
2826
2827 typedef struct MemoryRegionList MemoryRegionList;
2828
2829 struct MemoryRegionList {
2830 const MemoryRegion *mr;
2831 QTAILQ_ENTRY(MemoryRegionList) mrqueue;
2832 };
2833
2834 typedef QTAILQ_HEAD(, MemoryRegionList) MemoryRegionListHead;
2835
2836 #define MR_SIZE(size) (int128_nz(size) ? (hwaddr)int128_get64( \
2837 int128_sub((size), int128_one())) : 0)
2838 #define MTREE_INDENT " "
2839
2840 static void mtree_expand_owner(const char *label, Object *obj)
2841 {
2842 DeviceState *dev = (DeviceState *) object_dynamic_cast(obj, TYPE_DEVICE);
2843
2844 qemu_printf(" %s:{%s", label, dev ? "dev" : "obj");
2845 if (dev && dev->id) {
2846 qemu_printf(" id=%s", dev->id);
2847 } else {
2848 char *canonical_path = object_get_canonical_path(obj);
2849 if (canonical_path) {
2850 qemu_printf(" path=%s", canonical_path);
2851 g_free(canonical_path);
2852 } else {
2853 qemu_printf(" type=%s", object_get_typename(obj));
2854 }
2855 }
2856 qemu_printf("}");
2857 }
2858
2859 static void mtree_print_mr_owner(const MemoryRegion *mr)
2860 {
2861 Object *owner = mr->owner;
2862 Object *parent = memory_region_owner((MemoryRegion *)mr);
2863
2864 if (!owner && !parent) {
2865 qemu_printf(" orphan");
2866 return;
2867 }
2868 if (owner) {
2869 mtree_expand_owner("owner", owner);
2870 }
2871 if (parent && parent != owner) {
2872 mtree_expand_owner("parent", parent);
2873 }
2874 }
2875
2876 static void mtree_print_mr(const MemoryRegion *mr, unsigned int level,
2877 hwaddr base,
2878 MemoryRegionListHead *alias_print_queue,
2879 bool owner)
2880 {
2881 MemoryRegionList *new_ml, *ml, *next_ml;
2882 MemoryRegionListHead submr_print_queue;
2883 const MemoryRegion *submr;
2884 unsigned int i;
2885 hwaddr cur_start, cur_end;
2886
2887 if (!mr) {
2888 return;
2889 }
2890
2891 for (i = 0; i < level; i++) {
2892 qemu_printf(MTREE_INDENT);
2893 }
2894
2895 cur_start = base + mr->addr;
2896 cur_end = cur_start + MR_SIZE(mr->size);
2897
2898 /*
2899 * Try to detect overflow of memory region. This should never
2900 * happen normally. When it happens, we dump something to warn the
2901 * user who is observing this.
2902 */
2903 if (cur_start < base || cur_end < cur_start) {
2904 qemu_printf("[DETECTED OVERFLOW!] ");
2905 }
2906
2907 if (mr->alias) {
2908 MemoryRegionList *ml;
2909 bool found = false;
2910
2911 /* check if the alias is already in the queue */
2912 QTAILQ_FOREACH(ml, alias_print_queue, mrqueue) {
2913 if (ml->mr == mr->alias) {
2914 found = true;
2915 }
2916 }
2917
2918 if (!found) {
2919 ml = g_new(MemoryRegionList, 1);
2920 ml->mr = mr->alias;
2921 QTAILQ_INSERT_TAIL(alias_print_queue, ml, mrqueue);
2922 }
2923 qemu_printf(TARGET_FMT_plx "-" TARGET_FMT_plx
2924 " (prio %d, %s%s): alias %s @%s " TARGET_FMT_plx
2925 "-" TARGET_FMT_plx "%s",
2926 cur_start, cur_end,
2927 mr->priority,
2928 mr->nonvolatile ? "nv-" : "",
2929 memory_region_type((MemoryRegion *)mr),
2930 memory_region_name(mr),
2931 memory_region_name(mr->alias),
2932 mr->alias_offset,
2933 mr->alias_offset + MR_SIZE(mr->size),
2934 mr->enabled ? "" : " [disabled]");
2935 if (owner) {
2936 mtree_print_mr_owner(mr);
2937 }
2938 } else {
2939 qemu_printf(TARGET_FMT_plx "-" TARGET_FMT_plx
2940 " (prio %d, %s%s): %s%s",
2941 cur_start, cur_end,
2942 mr->priority,
2943 mr->nonvolatile ? "nv-" : "",
2944 memory_region_type((MemoryRegion *)mr),
2945 memory_region_name(mr),
2946 mr->enabled ? "" : " [disabled]");
2947 if (owner) {
2948 mtree_print_mr_owner(mr);
2949 }
2950 }
2951 qemu_printf("\n");
2952
2953 QTAILQ_INIT(&submr_print_queue);
2954
2955 QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
2956 new_ml = g_new(MemoryRegionList, 1);
2957 new_ml->mr = submr;
2958 QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
2959 if (new_ml->mr->addr < ml->mr->addr ||
2960 (new_ml->mr->addr == ml->mr->addr &&
2961 new_ml->mr->priority > ml->mr->priority)) {
2962 QTAILQ_INSERT_BEFORE(ml, new_ml, mrqueue);
2963 new_ml = NULL;
2964 break;
2965 }
2966 }
2967 if (new_ml) {
2968 QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, mrqueue);
2969 }
2970 }
2971
2972 QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
2973 mtree_print_mr(ml->mr, level + 1, cur_start,
2974 alias_print_queue, owner);
2975 }
2976
2977 QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, mrqueue, next_ml) {
2978 g_free(ml);
2979 }
2980 }
2981
2982 struct FlatViewInfo {
2983 int counter;
2984 bool dispatch_tree;
2985 bool owner;
2986 AccelClass *ac;
2987 };
2988
2989 static void mtree_print_flatview(gpointer key, gpointer value,
2990 gpointer user_data)
2991 {
2992 FlatView *view = key;
2993 GArray *fv_address_spaces = value;
2994 struct FlatViewInfo *fvi = user_data;
2995 FlatRange *range = &view->ranges[0];
2996 MemoryRegion *mr;
2997 int n = view->nr;
2998 int i;
2999 AddressSpace *as;
3000
3001 qemu_printf("FlatView #%d\n", fvi->counter);
3002 ++fvi->counter;
3003
3004 for (i = 0; i < fv_address_spaces->len; ++i) {
3005 as = g_array_index(fv_address_spaces, AddressSpace*, i);
3006 qemu_printf(" AS \"%s\", root: %s",
3007 as->name, memory_region_name(as->root));
3008 if (as->root->alias) {
3009 qemu_printf(", alias %s", memory_region_name(as->root->alias));
3010 }
3011 qemu_printf("\n");
3012 }
3013
3014 qemu_printf(" Root memory region: %s\n",
3015 view->root ? memory_region_name(view->root) : "(none)");
3016
3017 if (n <= 0) {
3018 qemu_printf(MTREE_INDENT "No rendered FlatView\n\n");
3019 return;
3020 }
3021
3022 while (n--) {
3023 mr = range->mr;
3024 if (range->offset_in_region) {
3025 qemu_printf(MTREE_INDENT TARGET_FMT_plx "-" TARGET_FMT_plx
3026 " (prio %d, %s%s): %s @" TARGET_FMT_plx,
3027 int128_get64(range->addr.start),
3028 int128_get64(range->addr.start)
3029 + MR_SIZE(range->addr.size),
3030 mr->priority,
3031 range->nonvolatile ? "nv-" : "",
3032 range->readonly ? "rom" : memory_region_type(mr),
3033 memory_region_name(mr),
3034 range->offset_in_region);
3035 } else {
3036 qemu_printf(MTREE_INDENT TARGET_FMT_plx "-" TARGET_FMT_plx
3037 " (prio %d, %s%s): %s",
3038 int128_get64(range->addr.start),
3039 int128_get64(range->addr.start)
3040 + MR_SIZE(range->addr.size),
3041 mr->priority,
3042 range->nonvolatile ? "nv-" : "",
3043 range->readonly ? "rom" : memory_region_type(mr),
3044 memory_region_name(mr));
3045 }
3046 if (fvi->owner) {
3047 mtree_print_mr_owner(mr);
3048 }
3049
3050 if (fvi->ac) {
3051 for (i = 0; i < fv_address_spaces->len; ++i) {
3052 as = g_array_index(fv_address_spaces, AddressSpace*, i);
3053 if (fvi->ac->has_memory(current_machine, as,
3054 int128_get64(range->addr.start),
3055 MR_SIZE(range->addr.size) + 1)) {
3056 qemu_printf(" %s", fvi->ac->name);
3057 }
3058 }
3059 }
3060 qemu_printf("\n");
3061 range++;
3062 }
3063
3064 #if !defined(CONFIG_USER_ONLY)
3065 if (fvi->dispatch_tree && view->root) {
3066 mtree_print_dispatch(view->dispatch, view->root);
3067 }
3068 #endif
3069
3070 qemu_printf("\n");
3071 }
3072
3073 static gboolean mtree_info_flatview_free(gpointer key, gpointer value,
3074 gpointer user_data)
3075 {
3076 FlatView *view = key;
3077 GArray *fv_address_spaces = value;
3078
3079 g_array_unref(fv_address_spaces);
3080 flatview_unref(view);
3081
3082 return true;
3083 }
3084
3085 void mtree_info(bool flatview, bool dispatch_tree, bool owner)
3086 {
3087 MemoryRegionListHead ml_head;
3088 MemoryRegionList *ml, *ml2;
3089 AddressSpace *as;
3090
3091 if (flatview) {
3092 FlatView *view;
3093 struct FlatViewInfo fvi = {
3094 .counter = 0,
3095 .dispatch_tree = dispatch_tree,
3096 .owner = owner,
3097 };
3098 GArray *fv_address_spaces;
3099 GHashTable *views = g_hash_table_new(g_direct_hash, g_direct_equal);
3100 AccelClass *ac = ACCEL_GET_CLASS(current_accel());
3101
3102 if (ac->has_memory) {
3103 fvi.ac = ac;
3104 }
3105
3106 /* Gather all FVs in one table */
3107 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3108 view = address_space_get_flatview(as);
3109
3110 fv_address_spaces = g_hash_table_lookup(views, view);
3111 if (!fv_address_spaces) {
3112 fv_address_spaces = g_array_new(false, false, sizeof(as));
3113 g_hash_table_insert(views, view, fv_address_spaces);
3114 }
3115
3116 g_array_append_val(fv_address_spaces, as);
3117 }
3118
3119 /* Print */
3120 g_hash_table_foreach(views, mtree_print_flatview, &fvi);
3121
3122 /* Free */
3123 g_hash_table_foreach_remove(views, mtree_info_flatview_free, 0);
3124 g_hash_table_unref(views);
3125
3126 return;
3127 }
3128
3129 QTAILQ_INIT(&ml_head);
3130
3131 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3132 qemu_printf("address-space: %s\n", as->name);
3133 mtree_print_mr(as->root, 1, 0, &ml_head, owner);
3134 qemu_printf("\n");
3135 }
3136
3137 /* print aliased regions */
3138 QTAILQ_FOREACH(ml, &ml_head, mrqueue) {
3139 qemu_printf("memory-region: %s\n", memory_region_name(ml->mr));
3140 mtree_print_mr(ml->mr, 1, 0, &ml_head, owner);
3141 qemu_printf("\n");
3142 }
3143
3144 QTAILQ_FOREACH_SAFE(ml, &ml_head, mrqueue, ml2) {
3145 g_free(ml);
3146 }
3147 }
3148
3149 void memory_region_init_ram(MemoryRegion *mr,
3150 struct Object *owner,
3151 const char *name,
3152 uint64_t size,
3153 Error **errp)
3154 {
3155 DeviceState *owner_dev;
3156 Error *err = NULL;
3157
3158 memory_region_init_ram_nomigrate(mr, owner, name, size, &err);
3159 if (err) {
3160 error_propagate(errp, err);
3161 return;
3162 }
3163 /* This will assert if owner is neither NULL nor a DeviceState.
3164 * We only want the owner here for the purposes of defining a
3165 * unique name for migration. TODO: Ideally we should implement
3166 * a naming scheme for Objects which are not DeviceStates, in
3167 * which case we can relax this restriction.
3168 */
3169 owner_dev = DEVICE(owner);
3170 vmstate_register_ram(mr, owner_dev);
3171 }
3172
3173 void memory_region_init_rom(MemoryRegion *mr,
3174 struct Object *owner,
3175 const char *name,
3176 uint64_t size,
3177 Error **errp)
3178 {
3179 DeviceState *owner_dev;
3180 Error *err = NULL;
3181
3182 memory_region_init_rom_nomigrate(mr, owner, name, size, &err);
3183 if (err) {
3184 error_propagate(errp, err);
3185 return;
3186 }
3187 /* This will assert if owner is neither NULL nor a DeviceState.
3188 * We only want the owner here for the purposes of defining a
3189 * unique name for migration. TODO: Ideally we should implement
3190 * a naming scheme for Objects which are not DeviceStates, in
3191 * which case we can relax this restriction.
3192 */
3193 owner_dev = DEVICE(owner);
3194 vmstate_register_ram(mr, owner_dev);
3195 }
3196
3197 void memory_region_init_rom_device(MemoryRegion *mr,
3198 struct Object *owner,
3199 const MemoryRegionOps *ops,
3200 void *opaque,
3201 const char *name,
3202 uint64_t size,
3203 Error **errp)
3204 {
3205 DeviceState *owner_dev;
3206 Error *err = NULL;
3207
3208 memory_region_init_rom_device_nomigrate(mr, owner, ops, opaque,
3209 name, size, &err);
3210 if (err) {
3211 error_propagate(errp, err);
3212 return;
3213 }
3214 /* This will assert if owner is neither NULL nor a DeviceState.
3215 * We only want the owner here for the purposes of defining a
3216 * unique name for migration. TODO: Ideally we should implement
3217 * a naming scheme for Objects which are not DeviceStates, in
3218 * which case we can relax this restriction.
3219 */
3220 owner_dev = DEVICE(owner);
3221 vmstate_register_ram(mr, owner_dev);
3222 }
3223
3224 static const TypeInfo memory_region_info = {
3225 .parent = TYPE_OBJECT,
3226 .name = TYPE_MEMORY_REGION,
3227 .class_size = sizeof(MemoryRegionClass),
3228 .instance_size = sizeof(MemoryRegion),
3229 .instance_init = memory_region_initfn,
3230 .instance_finalize = memory_region_finalize,
3231 };
3232
3233 static const TypeInfo iommu_memory_region_info = {
3234 .parent = TYPE_MEMORY_REGION,
3235 .name = TYPE_IOMMU_MEMORY_REGION,
3236 .class_size = sizeof(IOMMUMemoryRegionClass),
3237 .instance_size = sizeof(IOMMUMemoryRegion),
3238 .instance_init = iommu_memory_region_initfn,
3239 .abstract = true,
3240 };
3241
3242 static void memory_register_types(void)
3243 {
3244 type_register_static(&memory_region_info);
3245 type_register_static(&iommu_memory_region_info);
3246 }
3247
3248 type_init(memory_register_types)