]> git.proxmox.com Git - mirror_qemu.git/blob - softmmu/memory.c
projects / mirror_qemu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge remote-tracking branch 'remotes/dg-gitlab/tags/ppc-for-6.1-20210709' into staging
[mirror_qemu.git] / softmmu / 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 "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
37 //#define DEBUG_UNASSIGNED
38
39 static unsigned memory_region_transaction_depth;
40 static bool memory_region_update_pending;
41 static bool ioeventfd_update_pending;
42 bool global_dirty_log;
43
44 static QTAILQ_HEAD(, MemoryListener) memory_listeners
45 = QTAILQ_HEAD_INITIALIZER(memory_listeners);
46
47 static QTAILQ_HEAD(, AddressSpace) address_spaces
48 = QTAILQ_HEAD_INITIALIZER(address_spaces);
49
50 static GHashTable *flat_views;
51
52 typedef struct AddrRange AddrRange;
53
54 /*
55 * Note that signed integers are needed for negative offsetting in aliases
56 * (large MemoryRegion::alias_offset).
57 */
58 struct AddrRange {
59 Int128 start;
60 Int128 size;
61 };
62
63 static AddrRange addrrange_make(Int128 start, Int128 size)
64 {
65 return (AddrRange) { start, size };
66 }
67
68 static bool addrrange_equal(AddrRange r1, AddrRange r2)
69 {
70 return int128_eq(r1.start, r2.start) && int128_eq(r1.size, r2.size);
71 }
72
73 static Int128 addrrange_end(AddrRange r)
74 {
75 return int128_add(r.start, r.size);
76 }
77
78 static AddrRange addrrange_shift(AddrRange range, Int128 delta)
79 {
80 int128_addto(&range.start, delta);
81 return range;
82 }
83
84 static bool addrrange_contains(AddrRange range, Int128 addr)
85 {
86 return int128_ge(addr, range.start)
87 && int128_lt(addr, addrrange_end(range));
88 }
89
90 static bool addrrange_intersects(AddrRange r1, AddrRange r2)
91 {
92 return addrrange_contains(r1, r2.start)
93 || addrrange_contains(r2, r1.start);
94 }
95
96 static AddrRange addrrange_intersection(AddrRange r1, AddrRange r2)
97 {
98 Int128 start = int128_max(r1.start, r2.start);
99 Int128 end = int128_min(addrrange_end(r1), addrrange_end(r2));
100 return addrrange_make(start, int128_sub(end, start));
101 }
102
103 enum ListenerDirection { Forward, Reverse };
104
105 #define MEMORY_LISTENER_CALL_GLOBAL(_callback, _direction, _args...) \
106 do { \
107 MemoryListener *_listener; \
108 \
109 switch (_direction) { \
110 case Forward: \
111 QTAILQ_FOREACH(_listener, &memory_listeners, link) { \
112 if (_listener->_callback) { \
113 _listener->_callback(_listener, ##_args); \
114 } \
115 } \
116 break; \
117 case Reverse: \
118 QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners, link) { \
119 if (_listener->_callback) { \
120 _listener->_callback(_listener, ##_args); \
121 } \
122 } \
123 break; \
124 default: \
125 abort(); \
126 } \
127 } while (0)
128
129 #define MEMORY_LISTENER_CALL(_as, _callback, _direction, _section, _args...) \
130 do { \
131 MemoryListener *_listener; \
132 \
133 switch (_direction) { \
134 case Forward: \
135 QTAILQ_FOREACH(_listener, &(_as)->listeners, link_as) { \
136 if (_listener->_callback) { \
137 _listener->_callback(_listener, _section, ##_args); \
138 } \
139 } \
140 break; \
141 case Reverse: \
142 QTAILQ_FOREACH_REVERSE(_listener, &(_as)->listeners, link_as) { \
143 if (_listener->_callback) { \
144 _listener->_callback(_listener, _section, ##_args); \
145 } \
146 } \
147 break; \
148 default: \
149 abort(); \
150 } \
151 } while (0)
152
153 /* No need to ref/unref .mr, the FlatRange keeps it alive. */
154 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...) \
155 do { \
156 MemoryRegionSection mrs = section_from_flat_range(fr, \
157 address_space_to_flatview(as)); \
158 MEMORY_LISTENER_CALL(as, callback, dir, &mrs, ##_args); \
159 } while(0)
160
161 struct CoalescedMemoryRange {
162 AddrRange addr;
163 QTAILQ_ENTRY(CoalescedMemoryRange) link;
164 };
165
166 struct MemoryRegionIoeventfd {
167 AddrRange addr;
168 bool match_data;
169 uint64_t data;
170 EventNotifier *e;
171 };
172
173 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd *a,
174 MemoryRegionIoeventfd *b)
175 {
176 if (int128_lt(a->addr.start, b->addr.start)) {
177 return true;
178 } else if (int128_gt(a->addr.start, b->addr.start)) {
179 return false;
180 } else if (int128_lt(a->addr.size, b->addr.size)) {
181 return true;
182 } else if (int128_gt(a->addr.size, b->addr.size)) {
183 return false;
184 } else if (a->match_data < b->match_data) {
185 return true;
186 } else if (a->match_data > b->match_data) {
187 return false;
188 } else if (a->match_data) {
189 if (a->data < b->data) {
190 return true;
191 } else if (a->data > b->data) {
192 return false;
193 }
194 }
195 if (a->e < b->e) {
196 return true;
197 } else if (a->e > b->e) {
198 return false;
199 }
200 return false;
201 }
202
203 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd *a,
204 MemoryRegionIoeventfd *b)
205 {
206 if (int128_eq(a->addr.start, b->addr.start) &&
207 (!int128_nz(a->addr.size) || !int128_nz(b->addr.size) ||
208 (int128_eq(a->addr.size, b->addr.size) &&
209 (a->match_data == b->match_data) &&
210 ((a->match_data && (a->data == b->data)) || !a->match_data) &&
211 (a->e == b->e))))
212 return true;
213
214 return false;
215 }
216
217 /* Range of memory in the global map. Addresses are absolute. */
218 struct FlatRange {
219 MemoryRegion *mr;
220 hwaddr offset_in_region;
221 AddrRange addr;
222 uint8_t dirty_log_mask;
223 bool romd_mode;
224 bool readonly;
225 bool nonvolatile;
226 };
227
228 #define FOR_EACH_FLAT_RANGE(var, view) \
229 for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
230
231 static inline MemoryRegionSection
232 section_from_flat_range(FlatRange *fr, FlatView *fv)
233 {
234 return (MemoryRegionSection) {
235 .mr = fr->mr,
236 .fv = fv,
237 .offset_within_region = fr->offset_in_region,
238 .size = fr->addr.size,
239 .offset_within_address_space = int128_get64(fr->addr.start),
240 .readonly = fr->readonly,
241 .nonvolatile = fr->nonvolatile,
242 };
243 }
244
245 static bool flatrange_equal(FlatRange *a, FlatRange *b)
246 {
247 return a->mr == b->mr
248 && addrrange_equal(a->addr, b->addr)
249 && a->offset_in_region == b->offset_in_region
250 && a->romd_mode == b->romd_mode
251 && a->readonly == b->readonly
252 && a->nonvolatile == b->nonvolatile;
253 }
254
255 static FlatView *flatview_new(MemoryRegion *mr_root)
256 {
257 FlatView *view;
258
259 view = g_new0(FlatView, 1);
260 view->ref = 1;
261 view->root = mr_root;
262 memory_region_ref(mr_root);
263 trace_flatview_new(view, mr_root);
264
265 return view;
266 }
267
268 /* Insert a range into a given position. Caller is responsible for maintaining
269 * sorting order.
270 */
271 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
272 {
273 if (view->nr == view->nr_allocated) {
274 view->nr_allocated = MAX(2 * view->nr, 10);
275 view->ranges = g_realloc(view->ranges,
276 view->nr_allocated * sizeof(*view->ranges));
277 }
278 memmove(view->ranges + pos + 1, view->ranges + pos,
279 (view->nr - pos) * sizeof(FlatRange));
280 view->ranges[pos] = *range;
281 memory_region_ref(range->mr);
282 ++view->nr;
283 }
284
285 static void flatview_destroy(FlatView *view)
286 {
287 int i;
288
289 trace_flatview_destroy(view, view->root);
290 if (view->dispatch) {
291 address_space_dispatch_free(view->dispatch);
292 }
293 for (i = 0; i < view->nr; i++) {
294 memory_region_unref(view->ranges[i].mr);
295 }
296 g_free(view->ranges);
297 memory_region_unref(view->root);
298 g_free(view);
299 }
300
301 static bool flatview_ref(FlatView *view)
302 {
303 return qatomic_fetch_inc_nonzero(&view->ref) > 0;
304 }
305
306 void flatview_unref(FlatView *view)
307 {
308 if (qatomic_fetch_dec(&view->ref) == 1) {
309 trace_flatview_destroy_rcu(view, view->root);
310 assert(view->root);
311 call_rcu(view, flatview_destroy, rcu);
312 }
313 }
314
315 static bool can_merge(FlatRange *r1, FlatRange *r2)
316 {
317 return int128_eq(addrrange_end(r1->addr), r2->addr.start)
318 && r1->mr == r2->mr
319 && int128_eq(int128_add(int128_make64(r1->offset_in_region),
320 r1->addr.size),
321 int128_make64(r2->offset_in_region))
322 && r1->dirty_log_mask == r2->dirty_log_mask
323 && r1->romd_mode == r2->romd_mode
324 && r1->readonly == r2->readonly
325 && r1->nonvolatile == r2->nonvolatile;
326 }
327
328 /* Attempt to simplify a view by merging adjacent ranges */
329 static void flatview_simplify(FlatView *view)
330 {
331 unsigned i, j, k;
332
333 i = 0;
334 while (i < view->nr) {
335 j = i + 1;
336 while (j < view->nr
337 && can_merge(&view->ranges[j-1], &view->ranges[j])) {
338 int128_addto(&view->ranges[i].addr.size, view->ranges[j].addr.size);
339 ++j;
340 }
341 ++i;
342 for (k = i; k < j; k++) {
343 memory_region_unref(view->ranges[k].mr);
344 }
345 memmove(&view->ranges[i], &view->ranges[j],
346 (view->nr - j) * sizeof(view->ranges[j]));
347 view->nr -= j - i;
348 }
349 }
350
351 static bool memory_region_big_endian(MemoryRegion *mr)
352 {
353 #ifdef TARGET_WORDS_BIGENDIAN
354 return mr->ops->endianness != DEVICE_LITTLE_ENDIAN;
355 #else
356 return mr->ops->endianness == DEVICE_BIG_ENDIAN;
357 #endif
358 }
359
360 static void adjust_endianness(MemoryRegion *mr, uint64_t *data, MemOp op)
361 {
362 if ((op & MO_BSWAP) != devend_memop(mr->ops->endianness)) {
363 switch (op & MO_SIZE) {
364 case MO_8:
365 break;
366 case MO_16:
367 *data = bswap16(*data);
368 break;
369 case MO_32:
370 *data = bswap32(*data);
371 break;
372 case MO_64:
373 *data = bswap64(*data);
374 break;
375 default:
376 g_assert_not_reached();
377 }
378 }
379 }
380
381 static inline void memory_region_shift_read_access(uint64_t *value,
382 signed shift,
383 uint64_t mask,
384 uint64_t tmp)
385 {
386 if (shift >= 0) {
387 *value |= (tmp & mask) << shift;
388 } else {
389 *value |= (tmp & mask) >> -shift;
390 }
391 }
392
393 static inline uint64_t memory_region_shift_write_access(uint64_t *value,
394 signed shift,
395 uint64_t mask)
396 {
397 uint64_t tmp;
398
399 if (shift >= 0) {
400 tmp = (*value >> shift) & mask;
401 } else {
402 tmp = (*value << -shift) & mask;
403 }
404
405 return tmp;
406 }
407
408 static hwaddr memory_region_to_absolute_addr(MemoryRegion *mr, hwaddr offset)
409 {
410 MemoryRegion *root;
411 hwaddr abs_addr = offset;
412
413 abs_addr += mr->addr;
414 for (root = mr; root->container; ) {
415 root = root->container;
416 abs_addr += root->addr;
417 }
418
419 return abs_addr;
420 }
421
422 static int get_cpu_index(void)
423 {
424 if (current_cpu) {
425 return current_cpu->cpu_index;
426 }
427 return -1;
428 }
429
430 static MemTxResult memory_region_read_accessor(MemoryRegion *mr,
431 hwaddr addr,
432 uint64_t *value,
433 unsigned size,
434 signed shift,
435 uint64_t mask,
436 MemTxAttrs attrs)
437 {
438 uint64_t tmp;
439
440 tmp = mr->ops->read(mr->opaque, addr, size);
441 if (mr->subpage) {
442 trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
443 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_READ)) {
444 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
445 trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
446 }
447 memory_region_shift_read_access(value, shift, mask, tmp);
448 return MEMTX_OK;
449 }
450
451 static MemTxResult memory_region_read_with_attrs_accessor(MemoryRegion *mr,
452 hwaddr addr,
453 uint64_t *value,
454 unsigned size,
455 signed shift,
456 uint64_t mask,
457 MemTxAttrs attrs)
458 {
459 uint64_t tmp = 0;
460 MemTxResult r;
461
462 r = mr->ops->read_with_attrs(mr->opaque, addr, &tmp, size, attrs);
463 if (mr->subpage) {
464 trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
465 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_READ)) {
466 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
467 trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
468 }
469 memory_region_shift_read_access(value, shift, mask, tmp);
470 return r;
471 }
472
473 static MemTxResult memory_region_write_accessor(MemoryRegion *mr,
474 hwaddr addr,
475 uint64_t *value,
476 unsigned size,
477 signed shift,
478 uint64_t mask,
479 MemTxAttrs attrs)
480 {
481 uint64_t tmp = memory_region_shift_write_access(value, shift, mask);
482
483 if (mr->subpage) {
484 trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
485 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_WRITE)) {
486 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
487 trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
488 }
489 mr->ops->write(mr->opaque, addr, tmp, size);
490 return MEMTX_OK;
491 }
492
493 static MemTxResult memory_region_write_with_attrs_accessor(MemoryRegion *mr,
494 hwaddr addr,
495 uint64_t *value,
496 unsigned size,
497 signed shift,
498 uint64_t mask,
499 MemTxAttrs attrs)
500 {
501 uint64_t tmp = memory_region_shift_write_access(value, shift, mask);
502
503 if (mr->subpage) {
504 trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
505 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_WRITE)) {
506 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
507 trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
508 }
509 return mr->ops->write_with_attrs(mr->opaque, addr, tmp, size, attrs);
510 }
511
512 static MemTxResult access_with_adjusted_size(hwaddr addr,
513 uint64_t *value,
514 unsigned size,
515 unsigned access_size_min,
516 unsigned access_size_max,
517 MemTxResult (*access_fn)
518 (MemoryRegion *mr,
519 hwaddr addr,
520 uint64_t *value,
521 unsigned size,
522 signed shift,
523 uint64_t mask,
524 MemTxAttrs attrs),
525 MemoryRegion *mr,
526 MemTxAttrs attrs)
527 {
528 uint64_t access_mask;
529 unsigned access_size;
530 unsigned i;
531 MemTxResult r = MEMTX_OK;
532
533 if (!access_size_min) {
534 access_size_min = 1;
535 }
536 if (!access_size_max) {
537 access_size_max = 4;
538 }
539
540 /* FIXME: support unaligned access? */
541 access_size = MAX(MIN(size, access_size_max), access_size_min);
542 access_mask = MAKE_64BIT_MASK(0, access_size * 8);
543 if (memory_region_big_endian(mr)) {
544 for (i = 0; i < size; i += access_size) {
545 r |= access_fn(mr, addr + i, value, access_size,
546 (size - access_size - i) * 8, access_mask, attrs);
547 }
548 } else {
549 for (i = 0; i < size; i += access_size) {
550 r |= access_fn(mr, addr + i, value, access_size, i * 8,
551 access_mask, attrs);
552 }
553 }
554 return r;
555 }
556
557 static AddressSpace *memory_region_to_address_space(MemoryRegion *mr)
558 {
559 AddressSpace *as;
560
561 while (mr->container) {
562 mr = mr->container;
563 }
564 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
565 if (mr == as->root) {
566 return as;
567 }
568 }
569 return NULL;
570 }
571
572 /* Render a memory region into the global view. Ranges in @view obscure
573 * ranges in @mr.
574 */
575 static void render_memory_region(FlatView *view,
576 MemoryRegion *mr,
577 Int128 base,
578 AddrRange clip,
579 bool readonly,
580 bool nonvolatile)
581 {
582 MemoryRegion *subregion;
583 unsigned i;
584 hwaddr offset_in_region;
585 Int128 remain;
586 Int128 now;
587 FlatRange fr;
588 AddrRange tmp;
589
590 if (!mr->enabled) {
591 return;
592 }
593
594 int128_addto(&base, int128_make64(mr->addr));
595 readonly |= mr->readonly;
596 nonvolatile |= mr->nonvolatile;
597
598 tmp = addrrange_make(base, mr->size);
599
600 if (!addrrange_intersects(tmp, clip)) {
601 return;
602 }
603
604 clip = addrrange_intersection(tmp, clip);
605
606 if (mr->alias) {
607 int128_subfrom(&base, int128_make64(mr->alias->addr));
608 int128_subfrom(&base, int128_make64(mr->alias_offset));
609 render_memory_region(view, mr->alias, base, clip,
610 readonly, nonvolatile);
611 return;
612 }
613
614 /* Render subregions in priority order. */
615 QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
616 render_memory_region(view, subregion, base, clip,
617 readonly, nonvolatile);
618 }
619
620 if (!mr->terminates) {
621 return;
622 }
623
624 offset_in_region = int128_get64(int128_sub(clip.start, base));
625 base = clip.start;
626 remain = clip.size;
627
628 fr.mr = mr;
629 fr.dirty_log_mask = memory_region_get_dirty_log_mask(mr);
630 fr.romd_mode = mr->romd_mode;
631 fr.readonly = readonly;
632 fr.nonvolatile = nonvolatile;
633
634 /* Render the region itself into any gaps left by the current view. */
635 for (i = 0; i < view->nr && int128_nz(remain); ++i) {
636 if (int128_ge(base, addrrange_end(view->ranges[i].addr))) {
637 continue;
638 }
639 if (int128_lt(base, view->ranges[i].addr.start)) {
640 now = int128_min(remain,
641 int128_sub(view->ranges[i].addr.start, base));
642 fr.offset_in_region = offset_in_region;
643 fr.addr = addrrange_make(base, now);
644 flatview_insert(view, i, &fr);
645 ++i;
646 int128_addto(&base, now);
647 offset_in_region += int128_get64(now);
648 int128_subfrom(&remain, now);
649 }
650 now = int128_sub(int128_min(int128_add(base, remain),
651 addrrange_end(view->ranges[i].addr)),
652 base);
653 int128_addto(&base, now);
654 offset_in_region += int128_get64(now);
655 int128_subfrom(&remain, now);
656 }
657 if (int128_nz(remain)) {
658 fr.offset_in_region = offset_in_region;
659 fr.addr = addrrange_make(base, remain);
660 flatview_insert(view, i, &fr);
661 }
662 }
663
664 void flatview_for_each_range(FlatView *fv, flatview_cb cb , void *opaque)
665 {
666 FlatRange *fr;
667
668 assert(fv);
669 assert(cb);
670
671 FOR_EACH_FLAT_RANGE(fr, fv) {
672 if (cb(fr->addr.start, fr->addr.size, fr->mr,
673 fr->offset_in_region, opaque)) {
674 break;
675 }
676 }
677 }
678
679 static MemoryRegion *memory_region_get_flatview_root(MemoryRegion *mr)
680 {
681 while (mr->enabled) {
682 if (mr->alias) {
683 if (!mr->alias_offset && int128_ge(mr->size, mr->alias->size)) {
684 /* The alias is included in its entirety. Use it as
685 * the "real" root, so that we can share more FlatViews.
686 */
687 mr = mr->alias;
688 continue;
689 }
690 } else if (!mr->terminates) {
691 unsigned int found = 0;
692 MemoryRegion *child, *next = NULL;
693 QTAILQ_FOREACH(child, &mr->subregions, subregions_link) {
694 if (child->enabled) {
695 if (++found > 1) {
696 next = NULL;
697 break;
698 }
699 if (!child->addr && int128_ge(mr->size, child->size)) {
700 /* A child is included in its entirety. If it's the only
701 * enabled one, use it in the hope of finding an alias down the
702 * way. This will also let us share FlatViews.
703 */
704 next = child;
705 }
706 }
707 }
708 if (found == 0) {
709 return NULL;
710 }
711 if (next) {
712 mr = next;
713 continue;
714 }
715 }
716
717 return mr;
718 }
719
720 return NULL;
721 }
722
723 /* Render a memory topology into a list of disjoint absolute ranges. */
724 static FlatView *generate_memory_topology(MemoryRegion *mr)
725 {
726 int i;
727 FlatView *view;
728
729 view = flatview_new(mr);
730
731 if (mr) {
732 render_memory_region(view, mr, int128_zero(),
733 addrrange_make(int128_zero(), int128_2_64()),
734 false, false);
735 }
736 flatview_simplify(view);
737
738 view->dispatch = address_space_dispatch_new(view);
739 for (i = 0; i < view->nr; i++) {
740 MemoryRegionSection mrs =
741 section_from_flat_range(&view->ranges[i], view);
742 flatview_add_to_dispatch(view, &mrs);
743 }
744 address_space_dispatch_compact(view->dispatch);
745 g_hash_table_replace(flat_views, mr, view);
746
747 return view;
748 }
749
750 static void address_space_add_del_ioeventfds(AddressSpace *as,
751 MemoryRegionIoeventfd *fds_new,
752 unsigned fds_new_nb,
753 MemoryRegionIoeventfd *fds_old,
754 unsigned fds_old_nb)
755 {
756 unsigned iold, inew;
757 MemoryRegionIoeventfd *fd;
758 MemoryRegionSection section;
759
760 /* Generate a symmetric difference of the old and new fd sets, adding
761 * and deleting as necessary.
762 */
763
764 iold = inew = 0;
765 while (iold < fds_old_nb || inew < fds_new_nb) {
766 if (iold < fds_old_nb
767 && (inew == fds_new_nb
768 || memory_region_ioeventfd_before(&fds_old[iold],
769 &fds_new[inew]))) {
770 fd = &fds_old[iold];
771 section = (MemoryRegionSection) {
772 .fv = address_space_to_flatview(as),
773 .offset_within_address_space = int128_get64(fd->addr.start),
774 .size = fd->addr.size,
775 };
776 MEMORY_LISTENER_CALL(as, eventfd_del, Forward, &section,
777 fd->match_data, fd->data, fd->e);
778 ++iold;
779 } else if (inew < fds_new_nb
780 && (iold == fds_old_nb
781 || memory_region_ioeventfd_before(&fds_new[inew],
782 &fds_old[iold]))) {
783 fd = &fds_new[inew];
784 section = (MemoryRegionSection) {
785 .fv = address_space_to_flatview(as),
786 .offset_within_address_space = int128_get64(fd->addr.start),
787 .size = fd->addr.size,
788 };
789 MEMORY_LISTENER_CALL(as, eventfd_add, Reverse, &section,
790 fd->match_data, fd->data, fd->e);
791 ++inew;
792 } else {
793 ++iold;
794 ++inew;
795 }
796 }
797 }
798
799 FlatView *address_space_get_flatview(AddressSpace *as)
800 {
801 FlatView *view;
802
803 RCU_READ_LOCK_GUARD();
804 do {
805 view = address_space_to_flatview(as);
806 /* If somebody has replaced as->current_map concurrently,
807 * flatview_ref returns false.
808 */
809 } while (!flatview_ref(view));
810 return view;
811 }
812
813 static void address_space_update_ioeventfds(AddressSpace *as)
814 {
815 FlatView *view;
816 FlatRange *fr;
817 unsigned ioeventfd_nb = 0;
818 unsigned ioeventfd_max;
819 MemoryRegionIoeventfd *ioeventfds;
820 AddrRange tmp;
821 unsigned i;
822
823 /*
824 * It is likely that the number of ioeventfds hasn't changed much, so use
825 * the previous size as the starting value, with some headroom to avoid
826 * gratuitous reallocations.
827 */
828 ioeventfd_max = QEMU_ALIGN_UP(as->ioeventfd_nb, 4);
829 ioeventfds = g_new(MemoryRegionIoeventfd, ioeventfd_max);
830
831 view = address_space_get_flatview(as);
832 FOR_EACH_FLAT_RANGE(fr, view) {
833 for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
834 tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
835 int128_sub(fr->addr.start,
836 int128_make64(fr->offset_in_region)));
837 if (addrrange_intersects(fr->addr, tmp)) {
838 ++ioeventfd_nb;
839 if (ioeventfd_nb > ioeventfd_max) {
840 ioeventfd_max = MAX(ioeventfd_max * 2, 4);
841 ioeventfds = g_realloc(ioeventfds,
842 ioeventfd_max * sizeof(*ioeventfds));
843 }
844 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
845 ioeventfds[ioeventfd_nb-1].addr = tmp;
846 }
847 }
848 }
849
850 address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
851 as->ioeventfds, as->ioeventfd_nb);
852
853 g_free(as->ioeventfds);
854 as->ioeventfds = ioeventfds;
855 as->ioeventfd_nb = ioeventfd_nb;
856 flatview_unref(view);
857 }
858
859 /*
860 * Notify the memory listeners about the coalesced IO change events of
861 * range `cmr'. Only the part that has intersection of the specified
862 * FlatRange will be sent.
863 */
864 static void flat_range_coalesced_io_notify(FlatRange *fr, AddressSpace *as,
865 CoalescedMemoryRange *cmr, bool add)
866 {
867 AddrRange tmp;
868
869 tmp = addrrange_shift(cmr->addr,
870 int128_sub(fr->addr.start,
871 int128_make64(fr->offset_in_region)));
872 if (!addrrange_intersects(tmp, fr->addr)) {
873 return;
874 }
875 tmp = addrrange_intersection(tmp, fr->addr);
876
877 if (add) {
878 MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, coalesced_io_add,
879 int128_get64(tmp.start),
880 int128_get64(tmp.size));
881 } else {
882 MEMORY_LISTENER_UPDATE_REGION(fr, as, Reverse, coalesced_io_del,
883 int128_get64(tmp.start),
884 int128_get64(tmp.size));
885 }
886 }
887
888 static void flat_range_coalesced_io_del(FlatRange *fr, AddressSpace *as)
889 {
890 CoalescedMemoryRange *cmr;
891
892 QTAILQ_FOREACH(cmr, &fr->mr->coalesced, link) {
893 flat_range_coalesced_io_notify(fr, as, cmr, false);
894 }
895 }
896
897 static void flat_range_coalesced_io_add(FlatRange *fr, AddressSpace *as)
898 {
899 MemoryRegion *mr = fr->mr;
900 CoalescedMemoryRange *cmr;
901
902 if (QTAILQ_EMPTY(&mr->coalesced)) {
903 return;
904 }
905
906 QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
907 flat_range_coalesced_io_notify(fr, as, cmr, true);
908 }
909 }
910
911 static void address_space_update_topology_pass(AddressSpace *as,
912 const FlatView *old_view,
913 const FlatView *new_view,
914 bool adding)
915 {
916 unsigned iold, inew;
917 FlatRange *frold, *frnew;
918
919 /* Generate a symmetric difference of the old and new memory maps.
920 * Kill ranges in the old map, and instantiate ranges in the new map.
921 */
922 iold = inew = 0;
923 while (iold < old_view->nr || inew < new_view->nr) {
924 if (iold < old_view->nr) {
925 frold = &old_view->ranges[iold];
926 } else {
927 frold = NULL;
928 }
929 if (inew < new_view->nr) {
930 frnew = &new_view->ranges[inew];
931 } else {
932 frnew = NULL;
933 }
934
935 if (frold
936 && (!frnew
937 || int128_lt(frold->addr.start, frnew->addr.start)
938 || (int128_eq(frold->addr.start, frnew->addr.start)
939 && !flatrange_equal(frold, frnew)))) {
940 /* In old but not in new, or in both but attributes changed. */
941
942 if (!adding) {
943 flat_range_coalesced_io_del(frold, as);
944 MEMORY_LISTENER_UPDATE_REGION(frold, as, Reverse, region_del);
945 }
946
947 ++iold;
948 } else if (frold && frnew && flatrange_equal(frold, frnew)) {
949 /* In both and unchanged (except logging may have changed) */
950
951 if (adding) {
952 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_nop);
953 if (frnew->dirty_log_mask & ~frold->dirty_log_mask) {
954 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, log_start,
955 frold->dirty_log_mask,
956 frnew->dirty_log_mask);
957 }
958 if (frold->dirty_log_mask & ~frnew->dirty_log_mask) {
959 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Reverse, log_stop,
960 frold->dirty_log_mask,
961 frnew->dirty_log_mask);
962 }
963 }
964
965 ++iold;
966 ++inew;
967 } else {
968 /* In new */
969
970 if (adding) {
971 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_add);
972 flat_range_coalesced_io_add(frnew, as);
973 }
974
975 ++inew;
976 }
977 }
978 }
979
980 static void flatviews_init(void)
981 {
982 static FlatView *empty_view;
983
984 if (flat_views) {
985 return;
986 }
987
988 flat_views = g_hash_table_new_full(g_direct_hash, g_direct_equal, NULL,
989 (GDestroyNotify) flatview_unref);
990 if (!empty_view) {
991 empty_view = generate_memory_topology(NULL);
992 /* We keep it alive forever in the global variable. */
993 flatview_ref(empty_view);
994 } else {
995 g_hash_table_replace(flat_views, NULL, empty_view);
996 flatview_ref(empty_view);
997 }
998 }
999
1000 static void flatviews_reset(void)
1001 {
1002 AddressSpace *as;
1003
1004 if (flat_views) {
1005 g_hash_table_unref(flat_views);
1006 flat_views = NULL;
1007 }
1008 flatviews_init();
1009
1010 /* Render unique FVs */
1011 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1012 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1013
1014 if (g_hash_table_lookup(flat_views, physmr)) {
1015 continue;
1016 }
1017
1018 generate_memory_topology(physmr);
1019 }
1020 }
1021
1022 static void address_space_set_flatview(AddressSpace *as)
1023 {
1024 FlatView *old_view = address_space_to_flatview(as);
1025 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1026 FlatView *new_view = g_hash_table_lookup(flat_views, physmr);
1027
1028 assert(new_view);
1029
1030 if (old_view == new_view) {
1031 return;
1032 }
1033
1034 if (old_view) {
1035 flatview_ref(old_view);
1036 }
1037
1038 flatview_ref(new_view);
1039
1040 if (!QTAILQ_EMPTY(&as->listeners)) {
1041 FlatView tmpview = { .nr = 0 }, *old_view2 = old_view;
1042
1043 if (!old_view2) {
1044 old_view2 = &tmpview;
1045 }
1046 address_space_update_topology_pass(as, old_view2, new_view, false);
1047 address_space_update_topology_pass(as, old_view2, new_view, true);
1048 }
1049
1050 /* Writes are protected by the BQL. */
1051 qatomic_rcu_set(&as->current_map, new_view);
1052 if (old_view) {
1053 flatview_unref(old_view);
1054 }
1055
1056 /* Note that all the old MemoryRegions are still alive up to this
1057 * point. This relieves most MemoryListeners from the need to
1058 * ref/unref the MemoryRegions they get---unless they use them
1059 * outside the iothread mutex, in which case precise reference
1060 * counting is necessary.
1061 */
1062 if (old_view) {
1063 flatview_unref(old_view);
1064 }
1065 }
1066
1067 static void address_space_update_topology(AddressSpace *as)
1068 {
1069 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1070
1071 flatviews_init();
1072 if (!g_hash_table_lookup(flat_views, physmr)) {
1073 generate_memory_topology(physmr);
1074 }
1075 address_space_set_flatview(as);
1076 }
1077
1078 void memory_region_transaction_begin(void)
1079 {
1080 qemu_flush_coalesced_mmio_buffer();
1081 ++memory_region_transaction_depth;
1082 }
1083
1084 void memory_region_transaction_commit(void)
1085 {
1086 AddressSpace *as;
1087
1088 assert(memory_region_transaction_depth);
1089 assert(qemu_mutex_iothread_locked());
1090
1091 --memory_region_transaction_depth;
1092 if (!memory_region_transaction_depth) {
1093 if (memory_region_update_pending) {
1094 flatviews_reset();
1095
1096 MEMORY_LISTENER_CALL_GLOBAL(begin, Forward);
1097
1098 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1099 address_space_set_flatview(as);
1100 address_space_update_ioeventfds(as);
1101 }
1102 memory_region_update_pending = false;
1103 ioeventfd_update_pending = false;
1104 MEMORY_LISTENER_CALL_GLOBAL(commit, Forward);
1105 } else if (ioeventfd_update_pending) {
1106 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1107 address_space_update_ioeventfds(as);
1108 }
1109 ioeventfd_update_pending = false;
1110 }
1111 }
1112 }
1113
1114 static void memory_region_destructor_none(MemoryRegion *mr)
1115 {
1116 }
1117
1118 static void memory_region_destructor_ram(MemoryRegion *mr)
1119 {
1120 qemu_ram_free(mr->ram_block);
1121 }
1122
1123 static bool memory_region_need_escape(char c)
1124 {
1125 return c == '/' || c == '[' || c == '\\' || c == ']';
1126 }
1127
1128 static char *memory_region_escape_name(const char *name)
1129 {
1130 const char *p;
1131 char *escaped, *q;
1132 uint8_t c;
1133 size_t bytes = 0;
1134
1135 for (p = name; *p; p++) {
1136 bytes += memory_region_need_escape(*p) ? 4 : 1;
1137 }
1138 if (bytes == p - name) {
1139 return g_memdup(name, bytes + 1);
1140 }
1141
1142 escaped = g_malloc(bytes + 1);
1143 for (p = name, q = escaped; *p; p++) {
1144 c = *p;
1145 if (unlikely(memory_region_need_escape(c))) {
1146 *q++ = '\\';
1147 *q++ = 'x';
1148 *q++ = "0123456789abcdef"[c >> 4];
1149 c = "0123456789abcdef"[c & 15];
1150 }
1151 *q++ = c;
1152 }
1153 *q = 0;
1154 return escaped;
1155 }
1156
1157 static void memory_region_do_init(MemoryRegion *mr,
1158 Object *owner,
1159 const char *name,
1160 uint64_t size)
1161 {
1162 mr->size = int128_make64(size);
1163 if (size == UINT64_MAX) {
1164 mr->size = int128_2_64();
1165 }
1166 mr->name = g_strdup(name);
1167 mr->owner = owner;
1168 mr->ram_block = NULL;
1169
1170 if (name) {
1171 char *escaped_name = memory_region_escape_name(name);
1172 char *name_array = g_strdup_printf("%s[*]", escaped_name);
1173
1174 if (!owner) {
1175 owner = container_get(qdev_get_machine(), "/unattached");
1176 }
1177
1178 object_property_add_child(owner, name_array, OBJECT(mr));
1179 object_unref(OBJECT(mr));
1180 g_free(name_array);
1181 g_free(escaped_name);
1182 }
1183 }
1184
1185 void memory_region_init(MemoryRegion *mr,
1186 Object *owner,
1187 const char *name,
1188 uint64_t size)
1189 {
1190 object_initialize(mr, sizeof(*mr), TYPE_MEMORY_REGION);
1191 memory_region_do_init(mr, owner, name, size);
1192 }
1193
1194 static void memory_region_get_container(Object *obj, Visitor *v,
1195 const char *name, void *opaque,
1196 Error **errp)
1197 {
1198 MemoryRegion *mr = MEMORY_REGION(obj);
1199 char *path = (char *)"";
1200
1201 if (mr->container) {
1202 path = object_get_canonical_path(OBJECT(mr->container));
1203 }
1204 visit_type_str(v, name, &path, errp);
1205 if (mr->container) {
1206 g_free(path);
1207 }
1208 }
1209
1210 static Object *memory_region_resolve_container(Object *obj, void *opaque,
1211 const char *part)
1212 {
1213 MemoryRegion *mr = MEMORY_REGION(obj);
1214
1215 return OBJECT(mr->container);
1216 }
1217
1218 static void memory_region_get_priority(Object *obj, Visitor *v,
1219 const char *name, void *opaque,
1220 Error **errp)
1221 {
1222 MemoryRegion *mr = MEMORY_REGION(obj);
1223 int32_t value = mr->priority;
1224
1225 visit_type_int32(v, name, &value, errp);
1226 }
1227
1228 static void memory_region_get_size(Object *obj, Visitor *v, const char *name,
1229 void *opaque, Error **errp)
1230 {
1231 MemoryRegion *mr = MEMORY_REGION(obj);
1232 uint64_t value = memory_region_size(mr);
1233
1234 visit_type_uint64(v, name, &value, errp);
1235 }
1236
1237 static void memory_region_initfn(Object *obj)
1238 {
1239 MemoryRegion *mr = MEMORY_REGION(obj);
1240 ObjectProperty *op;
1241
1242 mr->ops = &unassigned_mem_ops;
1243 mr->enabled = true;
1244 mr->romd_mode = true;
1245 mr->destructor = memory_region_destructor_none;
1246 QTAILQ_INIT(&mr->subregions);
1247 QTAILQ_INIT(&mr->coalesced);
1248
1249 op = object_property_add(OBJECT(mr), "container",
1250 "link<" TYPE_MEMORY_REGION ">",
1251 memory_region_get_container,
1252 NULL, /* memory_region_set_container */
1253 NULL, NULL);
1254 op->resolve = memory_region_resolve_container;
1255
1256 object_property_add_uint64_ptr(OBJECT(mr), "addr",
1257 &mr->addr, OBJ_PROP_FLAG_READ);
1258 object_property_add(OBJECT(mr), "priority", "uint32",
1259 memory_region_get_priority,
1260 NULL, /* memory_region_set_priority */
1261 NULL, NULL);
1262 object_property_add(OBJECT(mr), "size", "uint64",
1263 memory_region_get_size,
1264 NULL, /* memory_region_set_size, */
1265 NULL, NULL);
1266 }
1267
1268 static void iommu_memory_region_initfn(Object *obj)
1269 {
1270 MemoryRegion *mr = MEMORY_REGION(obj);
1271
1272 mr->is_iommu = true;
1273 }
1274
1275 static uint64_t unassigned_mem_read(void *opaque, hwaddr addr,
1276 unsigned size)
1277 {
1278 #ifdef DEBUG_UNASSIGNED
1279 printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
1280 #endif
1281 return 0;
1282 }
1283
1284 static void unassigned_mem_write(void *opaque, hwaddr addr,
1285 uint64_t val, unsigned size)
1286 {
1287 #ifdef DEBUG_UNASSIGNED
1288 printf("Unassigned mem write " TARGET_FMT_plx " = 0x%"PRIx64"\n", addr, val);
1289 #endif
1290 }
1291
1292 static bool unassigned_mem_accepts(void *opaque, hwaddr addr,
1293 unsigned size, bool is_write,
1294 MemTxAttrs attrs)
1295 {
1296 return false;
1297 }
1298
1299 const MemoryRegionOps unassigned_mem_ops = {
1300 .valid.accepts = unassigned_mem_accepts,
1301 .endianness = DEVICE_NATIVE_ENDIAN,
1302 };
1303
1304 static uint64_t memory_region_ram_device_read(void *opaque,
1305 hwaddr addr, unsigned size)
1306 {
1307 MemoryRegion *mr = opaque;
1308 uint64_t data = (uint64_t)~0;
1309
1310 switch (size) {
1311 case 1:
1312 data = *(uint8_t *)(mr->ram_block->host + addr);
1313 break;
1314 case 2:
1315 data = *(uint16_t *)(mr->ram_block->host + addr);
1316 break;
1317 case 4:
1318 data = *(uint32_t *)(mr->ram_block->host + addr);
1319 break;
1320 case 8:
1321 data = *(uint64_t *)(mr->ram_block->host + addr);
1322 break;
1323 }
1324
1325 trace_memory_region_ram_device_read(get_cpu_index(), mr, addr, data, size);
1326
1327 return data;
1328 }
1329
1330 static void memory_region_ram_device_write(void *opaque, hwaddr addr,
1331 uint64_t data, unsigned size)
1332 {
1333 MemoryRegion *mr = opaque;
1334
1335 trace_memory_region_ram_device_write(get_cpu_index(), mr, addr, data, size);
1336
1337 switch (size) {
1338 case 1:
1339 *(uint8_t *)(mr->ram_block->host + addr) = (uint8_t)data;
1340 break;
1341 case 2:
1342 *(uint16_t *)(mr->ram_block->host + addr) = (uint16_t)data;
1343 break;
1344 case 4:
1345 *(uint32_t *)(mr->ram_block->host + addr) = (uint32_t)data;
1346 break;
1347 case 8:
1348 *(uint64_t *)(mr->ram_block->host + addr) = data;
1349 break;
1350 }
1351 }
1352
1353 static const MemoryRegionOps ram_device_mem_ops = {
1354 .read = memory_region_ram_device_read,
1355 .write = memory_region_ram_device_write,
1356 .endianness = DEVICE_HOST_ENDIAN,
1357 .valid = {
1358 .min_access_size = 1,
1359 .max_access_size = 8,
1360 .unaligned = true,
1361 },
1362 .impl = {
1363 .min_access_size = 1,
1364 .max_access_size = 8,
1365 .unaligned = true,
1366 },
1367 };
1368
1369 bool memory_region_access_valid(MemoryRegion *mr,
1370 hwaddr addr,
1371 unsigned size,
1372 bool is_write,
1373 MemTxAttrs attrs)
1374 {
1375 if (mr->ops->valid.accepts
1376 && !mr->ops->valid.accepts(mr->opaque, addr, size, is_write, attrs)) {
1377 qemu_log_mask(LOG_GUEST_ERROR, "Invalid access at addr "
1378 "0x%" HWADDR_PRIX ", size %u, "
1379 "region '%s', reason: rejected\n",
1380 addr, size, memory_region_name(mr));
1381 return false;
1382 }
1383
1384 if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1385 qemu_log_mask(LOG_GUEST_ERROR, "Invalid access at addr "
1386 "0x%" HWADDR_PRIX ", size %u, "
1387 "region '%s', reason: unaligned\n",
1388 addr, size, memory_region_name(mr));
1389 return false;
1390 }
1391
1392 /* Treat zero as compatibility all valid */
1393 if (!mr->ops->valid.max_access_size) {
1394 return true;
1395 }
1396
1397 if (size > mr->ops->valid.max_access_size
1398 || size < mr->ops->valid.min_access_size) {
1399 qemu_log_mask(LOG_GUEST_ERROR, "Invalid access at addr "
1400 "0x%" HWADDR_PRIX ", size %u, "
1401 "region '%s', reason: invalid size "
1402 "(min:%u max:%u)\n",
1403 addr, size, memory_region_name(mr),
1404 mr->ops->valid.min_access_size,
1405 mr->ops->valid.max_access_size);
1406 return false;
1407 }
1408 return true;
1409 }
1410
1411 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1412 hwaddr addr,
1413 uint64_t *pval,
1414 unsigned size,
1415 MemTxAttrs attrs)
1416 {
1417 *pval = 0;
1418
1419 if (mr->ops->read) {
1420 return access_with_adjusted_size(addr, pval, size,
1421 mr->ops->impl.min_access_size,
1422 mr->ops->impl.max_access_size,
1423 memory_region_read_accessor,
1424 mr, attrs);
1425 } else {
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_with_attrs_accessor,
1430 mr, attrs);
1431 }
1432 }
1433
1434 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1435 hwaddr addr,
1436 uint64_t *pval,
1437 MemOp op,
1438 MemTxAttrs attrs)
1439 {
1440 unsigned size = memop_size(op);
1441 MemTxResult r;
1442
1443 if (!memory_region_access_valid(mr, addr, size, false, attrs)) {
1444 *pval = unassigned_mem_read(mr, addr, size);
1445 return MEMTX_DECODE_ERROR;
1446 }
1447
1448 r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1449 adjust_endianness(mr, pval, op);
1450 return r;
1451 }
1452
1453 /* Return true if an eventfd was signalled */
1454 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1455 hwaddr addr,
1456 uint64_t data,
1457 unsigned size,
1458 MemTxAttrs attrs)
1459 {
1460 MemoryRegionIoeventfd ioeventfd = {
1461 .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1462 .data = data,
1463 };
1464 unsigned i;
1465
1466 for (i = 0; i < mr->ioeventfd_nb; i++) {
1467 ioeventfd.match_data = mr->ioeventfds[i].match_data;
1468 ioeventfd.e = mr->ioeventfds[i].e;
1469
1470 if (memory_region_ioeventfd_equal(&ioeventfd, &mr->ioeventfds[i])) {
1471 event_notifier_set(ioeventfd.e);
1472 return true;
1473 }
1474 }
1475
1476 return false;
1477 }
1478
1479 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1480 hwaddr addr,
1481 uint64_t data,
1482 MemOp op,
1483 MemTxAttrs attrs)
1484 {
1485 unsigned size = memop_size(op);
1486
1487 if (!memory_region_access_valid(mr, addr, size, true, attrs)) {
1488 unassigned_mem_write(mr, addr, data, size);
1489 return MEMTX_DECODE_ERROR;
1490 }
1491
1492 adjust_endianness(mr, &data, op);
1493
1494 if ((!kvm_eventfds_enabled()) &&
1495 memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1496 return MEMTX_OK;
1497 }
1498
1499 if (mr->ops->write) {
1500 return access_with_adjusted_size(addr, &data, size,
1501 mr->ops->impl.min_access_size,
1502 mr->ops->impl.max_access_size,
1503 memory_region_write_accessor, mr,
1504 attrs);
1505 } else {
1506 return
1507 access_with_adjusted_size(addr, &data, size,
1508 mr->ops->impl.min_access_size,
1509 mr->ops->impl.max_access_size,
1510 memory_region_write_with_attrs_accessor,
1511 mr, attrs);
1512 }
1513 }
1514
1515 void memory_region_init_io(MemoryRegion *mr,
1516 Object *owner,
1517 const MemoryRegionOps *ops,
1518 void *opaque,
1519 const char *name,
1520 uint64_t size)
1521 {
1522 memory_region_init(mr, owner, name, size);
1523 mr->ops = ops ? ops : &unassigned_mem_ops;
1524 mr->opaque = opaque;
1525 mr->terminates = true;
1526 }
1527
1528 void memory_region_init_ram_nomigrate(MemoryRegion *mr,
1529 Object *owner,
1530 const char *name,
1531 uint64_t size,
1532 Error **errp)
1533 {
1534 memory_region_init_ram_flags_nomigrate(mr, owner, name, size, 0, errp);
1535 }
1536
1537 void memory_region_init_ram_flags_nomigrate(MemoryRegion *mr,
1538 Object *owner,
1539 const char *name,
1540 uint64_t size,
1541 uint32_t ram_flags,
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(size, ram_flags, mr, &err);
1550 if (err) {
1551 mr->size = int128_zero();
1552 object_unparent(OBJECT(mr));
1553 error_propagate(errp, err);
1554 }
1555 }
1556
1557 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1558 Object *owner,
1559 const char *name,
1560 uint64_t size,
1561 uint64_t max_size,
1562 void (*resized)(const char*,
1563 uint64_t length,
1564 void *host),
1565 Error **errp)
1566 {
1567 Error *err = NULL;
1568 memory_region_init(mr, owner, name, size);
1569 mr->ram = true;
1570 mr->terminates = true;
1571 mr->destructor = memory_region_destructor_ram;
1572 mr->ram_block = qemu_ram_alloc_resizeable(size, max_size, resized,
1573 mr, &err);
1574 if (err) {
1575 mr->size = int128_zero();
1576 object_unparent(OBJECT(mr));
1577 error_propagate(errp, err);
1578 }
1579 }
1580
1581 #ifdef CONFIG_POSIX
1582 void memory_region_init_ram_from_file(MemoryRegion *mr,
1583 Object *owner,
1584 const char *name,
1585 uint64_t size,
1586 uint64_t align,
1587 uint32_t ram_flags,
1588 const char *path,
1589 bool readonly,
1590 Error **errp)
1591 {
1592 Error *err = NULL;
1593 memory_region_init(mr, owner, name, size);
1594 mr->ram = true;
1595 mr->readonly = readonly;
1596 mr->terminates = true;
1597 mr->destructor = memory_region_destructor_ram;
1598 mr->align = align;
1599 mr->ram_block = qemu_ram_alloc_from_file(size, mr, ram_flags, path,
1600 readonly, &err);
1601 if (err) {
1602 mr->size = int128_zero();
1603 object_unparent(OBJECT(mr));
1604 error_propagate(errp, err);
1605 }
1606 }
1607
1608 void memory_region_init_ram_from_fd(MemoryRegion *mr,
1609 Object *owner,
1610 const char *name,
1611 uint64_t size,
1612 uint32_t ram_flags,
1613 int fd,
1614 ram_addr_t offset,
1615 Error **errp)
1616 {
1617 Error *err = NULL;
1618 memory_region_init(mr, owner, name, size);
1619 mr->ram = true;
1620 mr->terminates = true;
1621 mr->destructor = memory_region_destructor_ram;
1622 mr->ram_block = qemu_ram_alloc_from_fd(size, mr, ram_flags, fd, offset,
1623 false, &err);
1624 if (err) {
1625 mr->size = int128_zero();
1626 object_unparent(OBJECT(mr));
1627 error_propagate(errp, err);
1628 }
1629 }
1630 #endif
1631
1632 void memory_region_init_ram_ptr(MemoryRegion *mr,
1633 Object *owner,
1634 const char *name,
1635 uint64_t size,
1636 void *ptr)
1637 {
1638 memory_region_init(mr, owner, name, size);
1639 mr->ram = true;
1640 mr->terminates = true;
1641 mr->destructor = memory_region_destructor_ram;
1642
1643 /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL. */
1644 assert(ptr != NULL);
1645 mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1646 }
1647
1648 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
1649 Object *owner,
1650 const char *name,
1651 uint64_t size,
1652 void *ptr)
1653 {
1654 memory_region_init(mr, owner, name, size);
1655 mr->ram = true;
1656 mr->terminates = true;
1657 mr->ram_device = true;
1658 mr->ops = &ram_device_mem_ops;
1659 mr->opaque = mr;
1660 mr->destructor = memory_region_destructor_ram;
1661
1662 /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL. */
1663 assert(ptr != NULL);
1664 mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1665 }
1666
1667 void memory_region_init_alias(MemoryRegion *mr,
1668 Object *owner,
1669 const char *name,
1670 MemoryRegion *orig,
1671 hwaddr offset,
1672 uint64_t size)
1673 {
1674 memory_region_init(mr, owner, name, size);
1675 mr->alias = orig;
1676 mr->alias_offset = offset;
1677 }
1678
1679 void memory_region_init_rom_nomigrate(MemoryRegion *mr,
1680 Object *owner,
1681 const char *name,
1682 uint64_t size,
1683 Error **errp)
1684 {
1685 memory_region_init_ram_flags_nomigrate(mr, owner, name, size, 0, errp);
1686 mr->readonly = true;
1687 }
1688
1689 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1690 Object *owner,
1691 const MemoryRegionOps *ops,
1692 void *opaque,
1693 const char *name,
1694 uint64_t size,
1695 Error **errp)
1696 {
1697 Error *err = NULL;
1698 assert(ops);
1699 memory_region_init(mr, owner, name, size);
1700 mr->ops = ops;
1701 mr->opaque = opaque;
1702 mr->terminates = true;
1703 mr->rom_device = true;
1704 mr->destructor = memory_region_destructor_ram;
1705 mr->ram_block = qemu_ram_alloc(size, 0, mr, &err);
1706 if (err) {
1707 mr->size = int128_zero();
1708 object_unparent(OBJECT(mr));
1709 error_propagate(errp, err);
1710 }
1711 }
1712
1713 void memory_region_init_iommu(void *_iommu_mr,
1714 size_t instance_size,
1715 const char *mrtypename,
1716 Object *owner,
1717 const char *name,
1718 uint64_t size)
1719 {
1720 struct IOMMUMemoryRegion *iommu_mr;
1721 struct MemoryRegion *mr;
1722
1723 object_initialize(_iommu_mr, instance_size, mrtypename);
1724 mr = MEMORY_REGION(_iommu_mr);
1725 memory_region_do_init(mr, owner, name, size);
1726 iommu_mr = IOMMU_MEMORY_REGION(mr);
1727 mr->terminates = true; /* then re-forwards */
1728 QLIST_INIT(&iommu_mr->iommu_notify);
1729 iommu_mr->iommu_notify_flags = IOMMU_NOTIFIER_NONE;
1730 }
1731
1732 static void memory_region_finalize(Object *obj)
1733 {
1734 MemoryRegion *mr = MEMORY_REGION(obj);
1735
1736 assert(!mr->container);
1737
1738 /* We know the region is not visible in any address space (it
1739 * does not have a container and cannot be a root either because
1740 * it has no references, so we can blindly clear mr->enabled.
1741 * memory_region_set_enabled instead could trigger a transaction
1742 * and cause an infinite loop.
1743 */
1744 mr->enabled = false;
1745 memory_region_transaction_begin();
1746 while (!QTAILQ_EMPTY(&mr->subregions)) {
1747 MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1748 memory_region_del_subregion(mr, subregion);
1749 }
1750 memory_region_transaction_commit();
1751
1752 mr->destructor(mr);
1753 memory_region_clear_coalescing(mr);
1754 g_free((char *)mr->name);
1755 g_free(mr->ioeventfds);
1756 }
1757
1758 Object *memory_region_owner(MemoryRegion *mr)
1759 {
1760 Object *obj = OBJECT(mr);
1761 return obj->parent;
1762 }
1763
1764 void memory_region_ref(MemoryRegion *mr)
1765 {
1766 /* MMIO callbacks most likely will access data that belongs
1767 * to the owner, hence the need to ref/unref the owner whenever
1768 * the memory region is in use.
1769 *
1770 * The memory region is a child of its owner. As long as the
1771 * owner doesn't call unparent itself on the memory region,
1772 * ref-ing the owner will also keep the memory region alive.
1773 * Memory regions without an owner are supposed to never go away;
1774 * we do not ref/unref them because it slows down DMA sensibly.
1775 */
1776 if (mr && mr->owner) {
1777 object_ref(mr->owner);
1778 }
1779 }
1780
1781 void memory_region_unref(MemoryRegion *mr)
1782 {
1783 if (mr && mr->owner) {
1784 object_unref(mr->owner);
1785 }
1786 }
1787
1788 uint64_t memory_region_size(MemoryRegion *mr)
1789 {
1790 if (int128_eq(mr->size, int128_2_64())) {
1791 return UINT64_MAX;
1792 }
1793 return int128_get64(mr->size);
1794 }
1795
1796 const char *memory_region_name(const MemoryRegion *mr)
1797 {
1798 if (!mr->name) {
1799 ((MemoryRegion *)mr)->name =
1800 g_strdup(object_get_canonical_path_component(OBJECT(mr)));
1801 }
1802 return mr->name;
1803 }
1804
1805 bool memory_region_is_ram_device(MemoryRegion *mr)
1806 {
1807 return mr->ram_device;
1808 }
1809
1810 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1811 {
1812 uint8_t mask = mr->dirty_log_mask;
1813 RAMBlock *rb = mr->ram_block;
1814
1815 if (global_dirty_log && ((rb && qemu_ram_is_migratable(rb)) ||
1816 memory_region_is_iommu(mr))) {
1817 mask |= (1 << DIRTY_MEMORY_MIGRATION);
1818 }
1819
1820 if (tcg_enabled() && rb) {
1821 /* TCG only cares about dirty memory logging for RAM, not IOMMU. */
1822 mask |= (1 << DIRTY_MEMORY_CODE);
1823 }
1824 return mask;
1825 }
1826
1827 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1828 {
1829 return memory_region_get_dirty_log_mask(mr) & (1 << client);
1830 }
1831
1832 static int memory_region_update_iommu_notify_flags(IOMMUMemoryRegion *iommu_mr,
1833 Error **errp)
1834 {
1835 IOMMUNotifierFlag flags = IOMMU_NOTIFIER_NONE;
1836 IOMMUNotifier *iommu_notifier;
1837 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1838 int ret = 0;
1839
1840 IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
1841 flags |= iommu_notifier->notifier_flags;
1842 }
1843
1844 if (flags != iommu_mr->iommu_notify_flags && imrc->notify_flag_changed) {
1845 ret = imrc->notify_flag_changed(iommu_mr,
1846 iommu_mr->iommu_notify_flags,
1847 flags, errp);
1848 }
1849
1850 if (!ret) {
1851 iommu_mr->iommu_notify_flags = flags;
1852 }
1853 return ret;
1854 }
1855
1856 int memory_region_iommu_set_page_size_mask(IOMMUMemoryRegion *iommu_mr,
1857 uint64_t page_size_mask,
1858 Error **errp)
1859 {
1860 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1861 int ret = 0;
1862
1863 if (imrc->iommu_set_page_size_mask) {
1864 ret = imrc->iommu_set_page_size_mask(iommu_mr, page_size_mask, errp);
1865 }
1866 return ret;
1867 }
1868
1869 int memory_region_register_iommu_notifier(MemoryRegion *mr,
1870 IOMMUNotifier *n, Error **errp)
1871 {
1872 IOMMUMemoryRegion *iommu_mr;
1873 int ret;
1874
1875 if (mr->alias) {
1876 return memory_region_register_iommu_notifier(mr->alias, n, errp);
1877 }
1878
1879 /* We need to register for at least one bitfield */
1880 iommu_mr = IOMMU_MEMORY_REGION(mr);
1881 assert(n->notifier_flags != IOMMU_NOTIFIER_NONE);
1882 assert(n->start <= n->end);
1883 assert(n->iommu_idx >= 0 &&
1884 n->iommu_idx < memory_region_iommu_num_indexes(iommu_mr));
1885
1886 QLIST_INSERT_HEAD(&iommu_mr->iommu_notify, n, node);
1887 ret = memory_region_update_iommu_notify_flags(iommu_mr, errp);
1888 if (ret) {
1889 QLIST_REMOVE(n, node);
1890 }
1891 return ret;
1892 }
1893
1894 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr)
1895 {
1896 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1897
1898 if (imrc->get_min_page_size) {
1899 return imrc->get_min_page_size(iommu_mr);
1900 }
1901 return TARGET_PAGE_SIZE;
1902 }
1903
1904 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
1905 {
1906 MemoryRegion *mr = MEMORY_REGION(iommu_mr);
1907 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1908 hwaddr addr, granularity;
1909 IOMMUTLBEntry iotlb;
1910
1911 /* If the IOMMU has its own replay callback, override */
1912 if (imrc->replay) {
1913 imrc->replay(iommu_mr, n);
1914 return;
1915 }
1916
1917 granularity = memory_region_iommu_get_min_page_size(iommu_mr);
1918
1919 for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1920 iotlb = imrc->translate(iommu_mr, addr, IOMMU_NONE, n->iommu_idx);
1921 if (iotlb.perm != IOMMU_NONE) {
1922 n->notify(n, &iotlb);
1923 }
1924
1925 /* if (2^64 - MR size) < granularity, it's possible to get an
1926 * infinite loop here. This should catch such a wraparound */
1927 if ((addr + granularity) < addr) {
1928 break;
1929 }
1930 }
1931 }
1932
1933 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1934 IOMMUNotifier *n)
1935 {
1936 IOMMUMemoryRegion *iommu_mr;
1937
1938 if (mr->alias) {
1939 memory_region_unregister_iommu_notifier(mr->alias, n);
1940 return;
1941 }
1942 QLIST_REMOVE(n, node);
1943 iommu_mr = IOMMU_MEMORY_REGION(mr);
1944 memory_region_update_iommu_notify_flags(iommu_mr, NULL);
1945 }
1946
1947 void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
1948 IOMMUTLBEvent *event)
1949 {
1950 IOMMUTLBEntry *entry = &event->entry;
1951 hwaddr entry_end = entry->iova + entry->addr_mask;
1952 IOMMUTLBEntry tmp = *entry;
1953
1954 if (event->type == IOMMU_NOTIFIER_UNMAP) {
1955 assert(entry->perm == IOMMU_NONE);
1956 }
1957
1958 /*
1959 * Skip the notification if the notification does not overlap
1960 * with registered range.
1961 */
1962 if (notifier->start > entry_end || notifier->end < entry->iova) {
1963 return;
1964 }
1965
1966 if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
1967 /* Crop (iova, addr_mask) to range */
1968 tmp.iova = MAX(tmp.iova, notifier->start);
1969 tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
1970 } else {
1971 assert(entry->iova >= notifier->start && entry_end <= notifier->end);
1972 }
1973
1974 if (event->type & notifier->notifier_flags) {
1975 notifier->notify(notifier, &tmp);
1976 }
1977 }
1978
1979 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
1980 int iommu_idx,
1981 IOMMUTLBEvent event)
1982 {
1983 IOMMUNotifier *iommu_notifier;
1984
1985 assert(memory_region_is_iommu(MEMORY_REGION(iommu_mr)));
1986
1987 IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
1988 if (iommu_notifier->iommu_idx == iommu_idx) {
1989 memory_region_notify_iommu_one(iommu_notifier, &event);
1990 }
1991 }
1992 }
1993
1994 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
1995 enum IOMMUMemoryRegionAttr attr,
1996 void *data)
1997 {
1998 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1999
2000 if (!imrc->get_attr) {
2001 return -EINVAL;
2002 }
2003
2004 return imrc->get_attr(iommu_mr, attr, data);
2005 }
2006
2007 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr,
2008 MemTxAttrs attrs)
2009 {
2010 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2011
2012 if (!imrc->attrs_to_index) {
2013 return 0;
2014 }
2015
2016 return imrc->attrs_to_index(iommu_mr, attrs);
2017 }
2018
2019 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr)
2020 {
2021 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2022
2023 if (!imrc->num_indexes) {
2024 return 1;
2025 }
2026
2027 return imrc->num_indexes(iommu_mr);
2028 }
2029
2030 RamDiscardManager *memory_region_get_ram_discard_manager(MemoryRegion *mr)
2031 {
2032 if (!memory_region_is_mapped(mr) || !memory_region_is_ram(mr)) {
2033 return NULL;
2034 }
2035 return mr->rdm;
2036 }
2037
2038 void memory_region_set_ram_discard_manager(MemoryRegion *mr,
2039 RamDiscardManager *rdm)
2040 {
2041 g_assert(memory_region_is_ram(mr) && !memory_region_is_mapped(mr));
2042 g_assert(!rdm || !mr->rdm);
2043 mr->rdm = rdm;
2044 }
2045
2046 uint64_t ram_discard_manager_get_min_granularity(const RamDiscardManager *rdm,
2047 const MemoryRegion *mr)
2048 {
2049 RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2050
2051 g_assert(rdmc->get_min_granularity);
2052 return rdmc->get_min_granularity(rdm, mr);
2053 }
2054
2055 bool ram_discard_manager_is_populated(const RamDiscardManager *rdm,
2056 const MemoryRegionSection *section)
2057 {
2058 RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2059
2060 g_assert(rdmc->is_populated);
2061 return rdmc->is_populated(rdm, section);
2062 }
2063
2064 int ram_discard_manager_replay_populated(const RamDiscardManager *rdm,
2065 MemoryRegionSection *section,
2066 ReplayRamPopulate replay_fn,
2067 void *opaque)
2068 {
2069 RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2070
2071 g_assert(rdmc->replay_populated);
2072 return rdmc->replay_populated(rdm, section, replay_fn, opaque);
2073 }
2074
2075 void ram_discard_manager_register_listener(RamDiscardManager *rdm,
2076 RamDiscardListener *rdl,
2077 MemoryRegionSection *section)
2078 {
2079 RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2080
2081 g_assert(rdmc->register_listener);
2082 rdmc->register_listener(rdm, rdl, section);
2083 }
2084
2085 void ram_discard_manager_unregister_listener(RamDiscardManager *rdm,
2086 RamDiscardListener *rdl)
2087 {
2088 RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2089
2090 g_assert(rdmc->unregister_listener);
2091 rdmc->unregister_listener(rdm, rdl);
2092 }
2093
2094 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
2095 {
2096 uint8_t mask = 1 << client;
2097 uint8_t old_logging;
2098
2099 assert(client == DIRTY_MEMORY_VGA);
2100 old_logging = mr->vga_logging_count;
2101 mr->vga_logging_count += log ? 1 : -1;
2102 if (!!old_logging == !!mr->vga_logging_count) {
2103 return;
2104 }
2105
2106 memory_region_transaction_begin();
2107 mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
2108 memory_region_update_pending |= mr->enabled;
2109 memory_region_transaction_commit();
2110 }
2111
2112 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
2113 hwaddr size)
2114 {
2115 assert(mr->ram_block);
2116 cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr,
2117 size,
2118 memory_region_get_dirty_log_mask(mr));
2119 }
2120
2121 /*
2122 * If memory region `mr' is NULL, do global sync. Otherwise, sync
2123 * dirty bitmap for the specified memory region.
2124 */
2125 static void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
2126 {
2127 MemoryListener *listener;
2128 AddressSpace *as;
2129 FlatView *view;
2130 FlatRange *fr;
2131
2132 /* If the same address space has multiple log_sync listeners, we
2133 * visit that address space's FlatView multiple times. But because
2134 * log_sync listeners are rare, it's still cheaper than walking each
2135 * address space once.
2136 */
2137 QTAILQ_FOREACH(listener, &memory_listeners, link) {
2138 if (listener->log_sync) {
2139 as = listener->address_space;
2140 view = address_space_get_flatview(as);
2141 FOR_EACH_FLAT_RANGE(fr, view) {
2142 if (fr->dirty_log_mask && (!mr || fr->mr == mr)) {
2143 MemoryRegionSection mrs = section_from_flat_range(fr, view);
2144 listener->log_sync(listener, &mrs);
2145 }
2146 }
2147 flatview_unref(view);
2148 } else if (listener->log_sync_global) {
2149 /*
2150 * No matter whether MR is specified, what we can do here
2151 * is to do a global sync, because we are not capable to
2152 * sync in a finer granularity.
2153 */
2154 listener->log_sync_global(listener);
2155 }
2156 }
2157 }
2158
2159 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start,
2160 hwaddr len)
2161 {
2162 MemoryRegionSection mrs;
2163 MemoryListener *listener;
2164 AddressSpace *as;
2165 FlatView *view;
2166 FlatRange *fr;
2167 hwaddr sec_start, sec_end, sec_size;
2168
2169 QTAILQ_FOREACH(listener, &memory_listeners, link) {
2170 if (!listener->log_clear) {
2171 continue;
2172 }
2173 as = listener->address_space;
2174 view = address_space_get_flatview(as);
2175 FOR_EACH_FLAT_RANGE(fr, view) {
2176 if (!fr->dirty_log_mask || fr->mr != mr) {
2177 /*
2178 * Clear dirty bitmap operation only applies to those
2179 * regions whose dirty logging is at least enabled
2180 */
2181 continue;
2182 }
2183
2184 mrs = section_from_flat_range(fr, view);
2185
2186 sec_start = MAX(mrs.offset_within_region, start);
2187 sec_end = mrs.offset_within_region + int128_get64(mrs.size);
2188 sec_end = MIN(sec_end, start + len);
2189
2190 if (sec_start >= sec_end) {
2191 /*
2192 * If this memory region section has no intersection
2193 * with the requested range, skip.
2194 */
2195 continue;
2196 }
2197
2198 /* Valid case; shrink the section if needed */
2199 mrs.offset_within_address_space +=
2200 sec_start - mrs.offset_within_region;
2201 mrs.offset_within_region = sec_start;
2202 sec_size = sec_end - sec_start;
2203 mrs.size = int128_make64(sec_size);
2204 listener->log_clear(listener, &mrs);
2205 }
2206 flatview_unref(view);
2207 }
2208 }
2209
2210 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
2211 hwaddr addr,
2212 hwaddr size,
2213 unsigned client)
2214 {
2215 DirtyBitmapSnapshot *snapshot;
2216 assert(mr->ram_block);
2217 memory_region_sync_dirty_bitmap(mr);
2218 snapshot = cpu_physical_memory_snapshot_and_clear_dirty(mr, addr, size, client);
2219 memory_global_after_dirty_log_sync();
2220 return snapshot;
2221 }
2222
2223 bool memory_region_snapshot_get_dirty(MemoryRegion *mr, DirtyBitmapSnapshot *snap,
2224 hwaddr addr, hwaddr size)
2225 {
2226 assert(mr->ram_block);
2227 return cpu_physical_memory_snapshot_get_dirty(snap,
2228 memory_region_get_ram_addr(mr) + addr, size);
2229 }
2230
2231 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
2232 {
2233 if (mr->readonly != readonly) {
2234 memory_region_transaction_begin();
2235 mr->readonly = readonly;
2236 memory_region_update_pending |= mr->enabled;
2237 memory_region_transaction_commit();
2238 }
2239 }
2240
2241 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile)
2242 {
2243 if (mr->nonvolatile != nonvolatile) {
2244 memory_region_transaction_begin();
2245 mr->nonvolatile = nonvolatile;
2246 memory_region_update_pending |= mr->enabled;
2247 memory_region_transaction_commit();
2248 }
2249 }
2250
2251 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
2252 {
2253 if (mr->romd_mode != romd_mode) {
2254 memory_region_transaction_begin();
2255 mr->romd_mode = romd_mode;
2256 memory_region_update_pending |= mr->enabled;
2257 memory_region_transaction_commit();
2258 }
2259 }
2260
2261 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
2262 hwaddr size, unsigned client)
2263 {
2264 assert(mr->ram_block);
2265 cpu_physical_memory_test_and_clear_dirty(
2266 memory_region_get_ram_addr(mr) + addr, size, client);
2267 }
2268
2269 int memory_region_get_fd(MemoryRegion *mr)
2270 {
2271 int fd;
2272
2273 RCU_READ_LOCK_GUARD();
2274 while (mr->alias) {
2275 mr = mr->alias;
2276 }
2277 fd = mr->ram_block->fd;
2278
2279 return fd;
2280 }
2281
2282 void *memory_region_get_ram_ptr(MemoryRegion *mr)
2283 {
2284 void *ptr;
2285 uint64_t offset = 0;
2286
2287 RCU_READ_LOCK_GUARD();
2288 while (mr->alias) {
2289 offset += mr->alias_offset;
2290 mr = mr->alias;
2291 }
2292 assert(mr->ram_block);
2293 ptr = qemu_map_ram_ptr(mr->ram_block, offset);
2294
2295 return ptr;
2296 }
2297
2298 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset)
2299 {
2300 RAMBlock *block;
2301
2302 block = qemu_ram_block_from_host(ptr, false, offset);
2303 if (!block) {
2304 return NULL;
2305 }
2306
2307 return block->mr;
2308 }
2309
2310 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr)
2311 {
2312 return mr->ram_block ? mr->ram_block->offset : RAM_ADDR_INVALID;
2313 }
2314
2315 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
2316 {
2317 assert(mr->ram_block);
2318
2319 qemu_ram_resize(mr->ram_block, newsize, errp);
2320 }
2321
2322 void memory_region_msync(MemoryRegion *mr, hwaddr addr, hwaddr size)
2323 {
2324 if (mr->ram_block) {
2325 qemu_ram_msync(mr->ram_block, addr, size);
2326 }
2327 }
2328
2329 void memory_region_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size)
2330 {
2331 /*
2332 * Might be extended case needed to cover
2333 * different types of memory regions
2334 */
2335 if (mr->dirty_log_mask) {
2336 memory_region_msync(mr, addr, size);
2337 }
2338 }
2339
2340 /*
2341 * Call proper memory listeners about the change on the newly
2342 * added/removed CoalescedMemoryRange.
2343 */
2344 static void memory_region_update_coalesced_range(MemoryRegion *mr,
2345 CoalescedMemoryRange *cmr,
2346 bool add)
2347 {
2348 AddressSpace *as;
2349 FlatView *view;
2350 FlatRange *fr;
2351
2352 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2353 view = address_space_get_flatview(as);
2354 FOR_EACH_FLAT_RANGE(fr, view) {
2355 if (fr->mr == mr) {
2356 flat_range_coalesced_io_notify(fr, as, cmr, add);
2357 }
2358 }
2359 flatview_unref(view);
2360 }
2361 }
2362
2363 void memory_region_set_coalescing(MemoryRegion *mr)
2364 {
2365 memory_region_clear_coalescing(mr);
2366 memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
2367 }
2368
2369 void memory_region_add_coalescing(MemoryRegion *mr,
2370 hwaddr offset,
2371 uint64_t size)
2372 {
2373 CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
2374
2375 cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
2376 QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
2377 memory_region_update_coalesced_range(mr, cmr, true);
2378 memory_region_set_flush_coalesced(mr);
2379 }
2380
2381 void memory_region_clear_coalescing(MemoryRegion *mr)
2382 {
2383 CoalescedMemoryRange *cmr;
2384
2385 if (QTAILQ_EMPTY(&mr->coalesced)) {
2386 return;
2387 }
2388
2389 qemu_flush_coalesced_mmio_buffer();
2390 mr->flush_coalesced_mmio = false;
2391
2392 while (!QTAILQ_EMPTY(&mr->coalesced)) {
2393 cmr = QTAILQ_FIRST(&mr->coalesced);
2394 QTAILQ_REMOVE(&mr->coalesced, cmr, link);
2395 memory_region_update_coalesced_range(mr, cmr, false);
2396 g_free(cmr);
2397 }
2398 }
2399
2400 void memory_region_set_flush_coalesced(MemoryRegion *mr)
2401 {
2402 mr->flush_coalesced_mmio = true;
2403 }
2404
2405 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
2406 {
2407 qemu_flush_coalesced_mmio_buffer();
2408 if (QTAILQ_EMPTY(&mr->coalesced)) {
2409 mr->flush_coalesced_mmio = false;
2410 }
2411 }
2412
2413 static bool userspace_eventfd_warning;
2414
2415 void memory_region_add_eventfd(MemoryRegion *mr,
2416 hwaddr addr,
2417 unsigned size,
2418 bool match_data,
2419 uint64_t data,
2420 EventNotifier *e)
2421 {
2422 MemoryRegionIoeventfd mrfd = {
2423 .addr.start = int128_make64(addr),
2424 .addr.size = int128_make64(size),
2425 .match_data = match_data,
2426 .data = data,
2427 .e = e,
2428 };
2429 unsigned i;
2430
2431 if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
2432 userspace_eventfd_warning))) {
2433 userspace_eventfd_warning = true;
2434 error_report("Using eventfd without MMIO binding in KVM. "
2435 "Suboptimal performance expected");
2436 }
2437
2438 if (size) {
2439 adjust_endianness(mr, &mrfd.data, size_memop(size) | MO_TE);
2440 }
2441 memory_region_transaction_begin();
2442 for (i = 0; i < mr->ioeventfd_nb; ++i) {
2443 if (memory_region_ioeventfd_before(&mrfd, &mr->ioeventfds[i])) {
2444 break;
2445 }
2446 }
2447 ++mr->ioeventfd_nb;
2448 mr->ioeventfds = g_realloc(mr->ioeventfds,
2449 sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
2450 memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
2451 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
2452 mr->ioeventfds[i] = mrfd;
2453 ioeventfd_update_pending |= mr->enabled;
2454 memory_region_transaction_commit();
2455 }
2456
2457 void memory_region_del_eventfd(MemoryRegion *mr,
2458 hwaddr addr,
2459 unsigned size,
2460 bool match_data,
2461 uint64_t data,
2462 EventNotifier *e)
2463 {
2464 MemoryRegionIoeventfd mrfd = {
2465 .addr.start = int128_make64(addr),
2466 .addr.size = int128_make64(size),
2467 .match_data = match_data,
2468 .data = data,
2469 .e = e,
2470 };
2471 unsigned i;
2472
2473 if (size) {
2474 adjust_endianness(mr, &mrfd.data, size_memop(size) | MO_TE);
2475 }
2476 memory_region_transaction_begin();
2477 for (i = 0; i < mr->ioeventfd_nb; ++i) {
2478 if (memory_region_ioeventfd_equal(&mrfd, &mr->ioeventfds[i])) {
2479 break;
2480 }
2481 }
2482 assert(i != mr->ioeventfd_nb);
2483 memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
2484 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
2485 --mr->ioeventfd_nb;
2486 mr->ioeventfds = g_realloc(mr->ioeventfds,
2487 sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
2488 ioeventfd_update_pending |= mr->enabled;
2489 memory_region_transaction_commit();
2490 }
2491
2492 static void memory_region_update_container_subregions(MemoryRegion *subregion)
2493 {
2494 MemoryRegion *mr = subregion->container;
2495 MemoryRegion *other;
2496
2497 memory_region_transaction_begin();
2498
2499 memory_region_ref(subregion);
2500 QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
2501 if (subregion->priority >= other->priority) {
2502 QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
2503 goto done;
2504 }
2505 }
2506 QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
2507 done:
2508 memory_region_update_pending |= mr->enabled && subregion->enabled;
2509 memory_region_transaction_commit();
2510 }
2511
2512 static void memory_region_add_subregion_common(MemoryRegion *mr,
2513 hwaddr offset,
2514 MemoryRegion *subregion)
2515 {
2516 assert(!subregion->container);
2517 subregion->container = mr;
2518 subregion->addr = offset;
2519 memory_region_update_container_subregions(subregion);
2520 }
2521
2522 void memory_region_add_subregion(MemoryRegion *mr,
2523 hwaddr offset,
2524 MemoryRegion *subregion)
2525 {
2526 subregion->priority = 0;
2527 memory_region_add_subregion_common(mr, offset, subregion);
2528 }
2529
2530 void memory_region_add_subregion_overlap(MemoryRegion *mr,
2531 hwaddr offset,
2532 MemoryRegion *subregion,
2533 int priority)
2534 {
2535 subregion->priority = priority;
2536 memory_region_add_subregion_common(mr, offset, subregion);
2537 }
2538
2539 void memory_region_del_subregion(MemoryRegion *mr,
2540 MemoryRegion *subregion)
2541 {
2542 memory_region_transaction_begin();
2543 assert(subregion->container == mr);
2544 subregion->container = NULL;
2545 QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
2546 memory_region_unref(subregion);
2547 memory_region_update_pending |= mr->enabled && subregion->enabled;
2548 memory_region_transaction_commit();
2549 }
2550
2551 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
2552 {
2553 if (enabled == mr->enabled) {
2554 return;
2555 }
2556 memory_region_transaction_begin();
2557 mr->enabled = enabled;
2558 memory_region_update_pending = true;
2559 memory_region_transaction_commit();
2560 }
2561
2562 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
2563 {
2564 Int128 s = int128_make64(size);
2565
2566 if (size == UINT64_MAX) {
2567 s = int128_2_64();
2568 }
2569 if (int128_eq(s, mr->size)) {
2570 return;
2571 }
2572 memory_region_transaction_begin();
2573 mr->size = s;
2574 memory_region_update_pending = true;
2575 memory_region_transaction_commit();
2576 }
2577
2578 static void memory_region_readd_subregion(MemoryRegion *mr)
2579 {
2580 MemoryRegion *container = mr->container;
2581
2582 if (container) {
2583 memory_region_transaction_begin();
2584 memory_region_ref(mr);
2585 memory_region_del_subregion(container, mr);
2586 mr->container = container;
2587 memory_region_update_container_subregions(mr);
2588 memory_region_unref(mr);
2589 memory_region_transaction_commit();
2590 }
2591 }
2592
2593 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
2594 {
2595 if (addr != mr->addr) {
2596 mr->addr = addr;
2597 memory_region_readd_subregion(mr);
2598 }
2599 }
2600
2601 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
2602 {
2603 assert(mr->alias);
2604
2605 if (offset == mr->alias_offset) {
2606 return;
2607 }
2608
2609 memory_region_transaction_begin();
2610 mr->alias_offset = offset;
2611 memory_region_update_pending |= mr->enabled;
2612 memory_region_transaction_commit();
2613 }
2614
2615 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
2616 {
2617 return mr->align;
2618 }
2619
2620 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
2621 {
2622 const AddrRange *addr = addr_;
2623 const FlatRange *fr = fr_;
2624
2625 if (int128_le(addrrange_end(*addr), fr->addr.start)) {
2626 return -1;
2627 } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
2628 return 1;
2629 }
2630 return 0;
2631 }
2632
2633 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
2634 {
2635 return bsearch(&addr, view->ranges, view->nr,
2636 sizeof(FlatRange), cmp_flatrange_addr);
2637 }
2638
2639 bool memory_region_is_mapped(MemoryRegion *mr)
2640 {
2641 return mr->container ? true : false;
2642 }
2643
2644 /* Same as memory_region_find, but it does not add a reference to the
2645 * returned region. It must be called from an RCU critical section.
2646 */
2647 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2648 hwaddr addr, uint64_t size)
2649 {
2650 MemoryRegionSection ret = { .mr = NULL };
2651 MemoryRegion *root;
2652 AddressSpace *as;
2653 AddrRange range;
2654 FlatView *view;
2655 FlatRange *fr;
2656
2657 addr += mr->addr;
2658 for (root = mr; root->container; ) {
2659 root = root->container;
2660 addr += root->addr;
2661 }
2662
2663 as = memory_region_to_address_space(root);
2664 if (!as) {
2665 return ret;
2666 }
2667 range = addrrange_make(int128_make64(addr), int128_make64(size));
2668
2669 view = address_space_to_flatview(as);
2670 fr = flatview_lookup(view, range);
2671 if (!fr) {
2672 return ret;
2673 }
2674
2675 while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2676 --fr;
2677 }
2678
2679 ret.mr = fr->mr;
2680 ret.fv = view;
2681 range = addrrange_intersection(range, fr->addr);
2682 ret.offset_within_region = fr->offset_in_region;
2683 ret.offset_within_region += int128_get64(int128_sub(range.start,
2684 fr->addr.start));
2685 ret.size = range.size;
2686 ret.offset_within_address_space = int128_get64(range.start);
2687 ret.readonly = fr->readonly;
2688 ret.nonvolatile = fr->nonvolatile;
2689 return ret;
2690 }
2691
2692 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2693 hwaddr addr, uint64_t size)
2694 {
2695 MemoryRegionSection ret;
2696 RCU_READ_LOCK_GUARD();
2697 ret = memory_region_find_rcu(mr, addr, size);
2698 if (ret.mr) {
2699 memory_region_ref(ret.mr);
2700 }
2701 return ret;
2702 }
2703
2704 MemoryRegionSection *memory_region_section_new_copy(MemoryRegionSection *s)
2705 {
2706 MemoryRegionSection *tmp = g_new(MemoryRegionSection, 1);
2707
2708 *tmp = *s;
2709 if (tmp->mr) {
2710 memory_region_ref(tmp->mr);
2711 }
2712 if (tmp->fv) {
2713 bool ret = flatview_ref(tmp->fv);
2714
2715 g_assert(ret);
2716 }
2717 return tmp;
2718 }
2719
2720 void memory_region_section_free_copy(MemoryRegionSection *s)
2721 {
2722 if (s->fv) {
2723 flatview_unref(s->fv);
2724 }
2725 if (s->mr) {
2726 memory_region_unref(s->mr);
2727 }
2728 g_free(s);
2729 }
2730
2731 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2732 {
2733 MemoryRegion *mr;
2734
2735 RCU_READ_LOCK_GUARD();
2736 mr = memory_region_find_rcu(container, addr, 1).mr;
2737 return mr && mr != container;
2738 }
2739
2740 void memory_global_dirty_log_sync(void)
2741 {
2742 memory_region_sync_dirty_bitmap(NULL);
2743 }
2744
2745 void memory_global_after_dirty_log_sync(void)
2746 {
2747 MEMORY_LISTENER_CALL_GLOBAL(log_global_after_sync, Forward);
2748 }
2749
2750 static VMChangeStateEntry *vmstate_change;
2751
2752 void memory_global_dirty_log_start(void)
2753 {
2754 if (vmstate_change) {
2755 qemu_del_vm_change_state_handler(vmstate_change);
2756 vmstate_change = NULL;
2757 }
2758
2759 global_dirty_log = true;
2760
2761 MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2762
2763 /* Refresh DIRTY_MEMORY_MIGRATION bit. */
2764 memory_region_transaction_begin();
2765 memory_region_update_pending = true;
2766 memory_region_transaction_commit();
2767 }
2768
2769 static void memory_global_dirty_log_do_stop(void)
2770 {
2771 global_dirty_log = false;
2772
2773 /* Refresh DIRTY_MEMORY_MIGRATION bit. */
2774 memory_region_transaction_begin();
2775 memory_region_update_pending = true;
2776 memory_region_transaction_commit();
2777
2778 MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2779 }
2780
2781 static void memory_vm_change_state_handler(void *opaque, bool running,
2782 RunState state)
2783 {
2784 if (running) {
2785 memory_global_dirty_log_do_stop();
2786
2787 if (vmstate_change) {
2788 qemu_del_vm_change_state_handler(vmstate_change);
2789 vmstate_change = NULL;
2790 }
2791 }
2792 }
2793
2794 void memory_global_dirty_log_stop(void)
2795 {
2796 if (!runstate_is_running()) {
2797 if (vmstate_change) {
2798 return;
2799 }
2800 vmstate_change = qemu_add_vm_change_state_handler(
2801 memory_vm_change_state_handler, NULL);
2802 return;
2803 }
2804
2805 memory_global_dirty_log_do_stop();
2806 }
2807
2808 static void listener_add_address_space(MemoryListener *listener,
2809 AddressSpace *as)
2810 {
2811 FlatView *view;
2812 FlatRange *fr;
2813
2814 if (listener->begin) {
2815 listener->begin(listener);
2816 }
2817 if (global_dirty_log) {
2818 if (listener->log_global_start) {
2819 listener->log_global_start(listener);
2820 }
2821 }
2822
2823 view = address_space_get_flatview(as);
2824 FOR_EACH_FLAT_RANGE(fr, view) {
2825 MemoryRegionSection section = section_from_flat_range(fr, view);
2826
2827 if (listener->region_add) {
2828 listener->region_add(listener, &section);
2829 }
2830 if (fr->dirty_log_mask && listener->log_start) {
2831 listener->log_start(listener, &section, 0, fr->dirty_log_mask);
2832 }
2833 }
2834 if (listener->commit) {
2835 listener->commit(listener);
2836 }
2837 flatview_unref(view);
2838 }
2839
2840 static void listener_del_address_space(MemoryListener *listener,
2841 AddressSpace *as)
2842 {
2843 FlatView *view;
2844 FlatRange *fr;
2845
2846 if (listener->begin) {
2847 listener->begin(listener);
2848 }
2849 view = address_space_get_flatview(as);
2850 FOR_EACH_FLAT_RANGE(fr, view) {
2851 MemoryRegionSection section = section_from_flat_range(fr, view);
2852
2853 if (fr->dirty_log_mask && listener->log_stop) {
2854 listener->log_stop(listener, &section, fr->dirty_log_mask, 0);
2855 }
2856 if (listener->region_del) {
2857 listener->region_del(listener, &section);
2858 }
2859 }
2860 if (listener->commit) {
2861 listener->commit(listener);
2862 }
2863 flatview_unref(view);
2864 }
2865
2866 void memory_listener_register(MemoryListener *listener, AddressSpace *as)
2867 {
2868 MemoryListener *other = NULL;
2869
2870 /* Only one of them can be defined for a listener */
2871 assert(!(listener->log_sync && listener->log_sync_global));
2872
2873 listener->address_space = as;
2874 if (QTAILQ_EMPTY(&memory_listeners)
2875 || listener->priority >= QTAILQ_LAST(&memory_listeners)->priority) {
2876 QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
2877 } else {
2878 QTAILQ_FOREACH(other, &memory_listeners, link) {
2879 if (listener->priority < other->priority) {
2880 break;
2881 }
2882 }
2883 QTAILQ_INSERT_BEFORE(other, listener, link);
2884 }
2885
2886 if (QTAILQ_EMPTY(&as->listeners)
2887 || listener->priority >= QTAILQ_LAST(&as->listeners)->priority) {
2888 QTAILQ_INSERT_TAIL(&as->listeners, listener, link_as);
2889 } else {
2890 QTAILQ_FOREACH(other, &as->listeners, link_as) {
2891 if (listener->priority < other->priority) {
2892 break;
2893 }
2894 }
2895 QTAILQ_INSERT_BEFORE(other, listener, link_as);
2896 }
2897
2898 listener_add_address_space(listener, as);
2899 }
2900
2901 void memory_listener_unregister(MemoryListener *listener)
2902 {
2903 if (!listener->address_space) {
2904 return;
2905 }
2906
2907 listener_del_address_space(listener, listener->address_space);
2908 QTAILQ_REMOVE(&memory_listeners, listener, link);
2909 QTAILQ_REMOVE(&listener->address_space->listeners, listener, link_as);
2910 listener->address_space = NULL;
2911 }
2912
2913 void address_space_remove_listeners(AddressSpace *as)
2914 {
2915 while (!QTAILQ_EMPTY(&as->listeners)) {
2916 memory_listener_unregister(QTAILQ_FIRST(&as->listeners));
2917 }
2918 }
2919
2920 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
2921 {
2922 memory_region_ref(root);
2923 as->root = root;
2924 as->current_map = NULL;
2925 as->ioeventfd_nb = 0;
2926 as->ioeventfds = NULL;
2927 QTAILQ_INIT(&as->listeners);
2928 QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
2929 as->name = g_strdup(name ? name : "anonymous");
2930 address_space_update_topology(as);
2931 address_space_update_ioeventfds(as);
2932 }
2933
2934 static void do_address_space_destroy(AddressSpace *as)
2935 {
2936 assert(QTAILQ_EMPTY(&as->listeners));
2937
2938 flatview_unref(as->current_map);
2939 g_free(as->name);
2940 g_free(as->ioeventfds);
2941 memory_region_unref(as->root);
2942 }
2943
2944 void address_space_destroy(AddressSpace *as)
2945 {
2946 MemoryRegion *root = as->root;
2947
2948 /* Flush out anything from MemoryListeners listening in on this */
2949 memory_region_transaction_begin();
2950 as->root = NULL;
2951 memory_region_transaction_commit();
2952 QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
2953
2954 /* At this point, as->dispatch and as->current_map are dummy
2955 * entries that the guest should never use. Wait for the old
2956 * values to expire before freeing the data.
2957 */
2958 as->root = root;
2959 call_rcu(as, do_address_space_destroy, rcu);
2960 }
2961
2962 static const char *memory_region_type(MemoryRegion *mr)
2963 {
2964 if (mr->alias) {
2965 return memory_region_type(mr->alias);
2966 }
2967 if (memory_region_is_ram_device(mr)) {
2968 return "ramd";
2969 } else if (memory_region_is_romd(mr)) {
2970 return "romd";
2971 } else if (memory_region_is_rom(mr)) {
2972 return "rom";
2973 } else if (memory_region_is_ram(mr)) {
2974 return "ram";
2975 } else {
2976 return "i/o";
2977 }
2978 }
2979
2980 typedef struct MemoryRegionList MemoryRegionList;
2981
2982 struct MemoryRegionList {
2983 const MemoryRegion *mr;
2984 QTAILQ_ENTRY(MemoryRegionList) mrqueue;
2985 };
2986
2987 typedef QTAILQ_HEAD(, MemoryRegionList) MemoryRegionListHead;
2988
2989 #define MR_SIZE(size) (int128_nz(size) ? (hwaddr)int128_get64( \
2990 int128_sub((size), int128_one())) : 0)
2991 #define MTREE_INDENT " "
2992
2993 static void mtree_expand_owner(const char *label, Object *obj)
2994 {
2995 DeviceState *dev = (DeviceState *) object_dynamic_cast(obj, TYPE_DEVICE);
2996
2997 qemu_printf(" %s:{%s", label, dev ? "dev" : "obj");
2998 if (dev && dev->id) {
2999 qemu_printf(" id=%s", dev->id);
3000 } else {
3001 char *canonical_path = object_get_canonical_path(obj);
3002 if (canonical_path) {
3003 qemu_printf(" path=%s", canonical_path);
3004 g_free(canonical_path);
3005 } else {
3006 qemu_printf(" type=%s", object_get_typename(obj));
3007 }
3008 }
3009 qemu_printf("}");
3010 }
3011
3012 static void mtree_print_mr_owner(const MemoryRegion *mr)
3013 {
3014 Object *owner = mr->owner;
3015 Object *parent = memory_region_owner((MemoryRegion *)mr);
3016
3017 if (!owner && !parent) {
3018 qemu_printf(" orphan");
3019 return;
3020 }
3021 if (owner) {
3022 mtree_expand_owner("owner", owner);
3023 }
3024 if (parent && parent != owner) {
3025 mtree_expand_owner("parent", parent);
3026 }
3027 }
3028
3029 static void mtree_print_mr(const MemoryRegion *mr, unsigned int level,
3030 hwaddr base,
3031 MemoryRegionListHead *alias_print_queue,
3032 bool owner, bool display_disabled)
3033 {
3034 MemoryRegionList *new_ml, *ml, *next_ml;
3035 MemoryRegionListHead submr_print_queue;
3036 const MemoryRegion *submr;
3037 unsigned int i;
3038 hwaddr cur_start, cur_end;
3039
3040 if (!mr) {
3041 return;
3042 }
3043
3044 cur_start = base + mr->addr;
3045 cur_end = cur_start + MR_SIZE(mr->size);
3046
3047 /*
3048 * Try to detect overflow of memory region. This should never
3049 * happen normally. When it happens, we dump something to warn the
3050 * user who is observing this.
3051 */
3052 if (cur_start < base || cur_end < cur_start) {
3053 qemu_printf("[DETECTED OVERFLOW!] ");
3054 }
3055
3056 if (mr->alias) {
3057 MemoryRegionList *ml;
3058 bool found = false;
3059
3060 /* check if the alias is already in the queue */
3061 QTAILQ_FOREACH(ml, alias_print_queue, mrqueue) {
3062 if (ml->mr == mr->alias) {
3063 found = true;
3064 }
3065 }
3066
3067 if (!found) {
3068 ml = g_new(MemoryRegionList, 1);
3069 ml->mr = mr->alias;
3070 QTAILQ_INSERT_TAIL(alias_print_queue, ml, mrqueue);
3071 }
3072 if (mr->enabled || display_disabled) {
3073 for (i = 0; i < level; i++) {
3074 qemu_printf(MTREE_INDENT);
3075 }
3076 qemu_printf(TARGET_FMT_plx "-" TARGET_FMT_plx
3077 " (prio %d, %s%s): alias %s @%s " TARGET_FMT_plx
3078 "-" TARGET_FMT_plx "%s",
3079 cur_start, cur_end,
3080 mr->priority,
3081 mr->nonvolatile ? "nv-" : "",
3082 memory_region_type((MemoryRegion *)mr),
3083 memory_region_name(mr),
3084 memory_region_name(mr->alias),
3085 mr->alias_offset,
3086 mr->alias_offset + MR_SIZE(mr->size),
3087 mr->enabled ? "" : " [disabled]");
3088 if (owner) {
3089 mtree_print_mr_owner(mr);
3090 }
3091 qemu_printf("\n");
3092 }
3093 } else {
3094 if (mr->enabled || display_disabled) {
3095 for (i = 0; i < level; i++) {
3096 qemu_printf(MTREE_INDENT);
3097 }
3098 qemu_printf(TARGET_FMT_plx "-" TARGET_FMT_plx
3099 " (prio %d, %s%s): %s%s",
3100 cur_start, cur_end,
3101 mr->priority,
3102 mr->nonvolatile ? "nv-" : "",
3103 memory_region_type((MemoryRegion *)mr),
3104 memory_region_name(mr),
3105 mr->enabled ? "" : " [disabled]");
3106 if (owner) {
3107 mtree_print_mr_owner(mr);
3108 }
3109 qemu_printf("\n");
3110 }
3111 }
3112
3113 QTAILQ_INIT(&submr_print_queue);
3114
3115 QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
3116 new_ml = g_new(MemoryRegionList, 1);
3117 new_ml->mr = submr;
3118 QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
3119 if (new_ml->mr->addr < ml->mr->addr ||
3120 (new_ml->mr->addr == ml->mr->addr &&
3121 new_ml->mr->priority > ml->mr->priority)) {
3122 QTAILQ_INSERT_BEFORE(ml, new_ml, mrqueue);
3123 new_ml = NULL;
3124 break;
3125 }
3126 }
3127 if (new_ml) {
3128 QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, mrqueue);
3129 }
3130 }
3131
3132 QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
3133 mtree_print_mr(ml->mr, level + 1, cur_start,
3134 alias_print_queue, owner, display_disabled);
3135 }
3136
3137 QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, mrqueue, next_ml) {
3138 g_free(ml);
3139 }
3140 }
3141
3142 struct FlatViewInfo {
3143 int counter;
3144 bool dispatch_tree;
3145 bool owner;
3146 AccelClass *ac;
3147 };
3148
3149 static void mtree_print_flatview(gpointer key, gpointer value,
3150 gpointer user_data)
3151 {
3152 FlatView *view = key;
3153 GArray *fv_address_spaces = value;
3154 struct FlatViewInfo *fvi = user_data;
3155 FlatRange *range = &view->ranges[0];
3156 MemoryRegion *mr;
3157 int n = view->nr;
3158 int i;
3159 AddressSpace *as;
3160
3161 qemu_printf("FlatView #%d\n", fvi->counter);
3162 ++fvi->counter;
3163
3164 for (i = 0; i < fv_address_spaces->len; ++i) {
3165 as = g_array_index(fv_address_spaces, AddressSpace*, i);
3166 qemu_printf(" AS \"%s\", root: %s",
3167 as->name, memory_region_name(as->root));
3168 if (as->root->alias) {
3169 qemu_printf(", alias %s", memory_region_name(as->root->alias));
3170 }
3171 qemu_printf("\n");
3172 }
3173
3174 qemu_printf(" Root memory region: %s\n",
3175 view->root ? memory_region_name(view->root) : "(none)");
3176
3177 if (n <= 0) {
3178 qemu_printf(MTREE_INDENT "No rendered FlatView\n\n");
3179 return;
3180 }
3181
3182 while (n--) {
3183 mr = range->mr;
3184 if (range->offset_in_region) {
3185 qemu_printf(MTREE_INDENT TARGET_FMT_plx "-" TARGET_FMT_plx
3186 " (prio %d, %s%s): %s @" TARGET_FMT_plx,
3187 int128_get64(range->addr.start),
3188 int128_get64(range->addr.start)
3189 + MR_SIZE(range->addr.size),
3190 mr->priority,
3191 range->nonvolatile ? "nv-" : "",
3192 range->readonly ? "rom" : memory_region_type(mr),
3193 memory_region_name(mr),
3194 range->offset_in_region);
3195 } else {
3196 qemu_printf(MTREE_INDENT TARGET_FMT_plx "-" TARGET_FMT_plx
3197 " (prio %d, %s%s): %s",
3198 int128_get64(range->addr.start),
3199 int128_get64(range->addr.start)
3200 + MR_SIZE(range->addr.size),
3201 mr->priority,
3202 range->nonvolatile ? "nv-" : "",
3203 range->readonly ? "rom" : memory_region_type(mr),
3204 memory_region_name(mr));
3205 }
3206 if (fvi->owner) {
3207 mtree_print_mr_owner(mr);
3208 }
3209
3210 if (fvi->ac) {
3211 for (i = 0; i < fv_address_spaces->len; ++i) {
3212 as = g_array_index(fv_address_spaces, AddressSpace*, i);
3213 if (fvi->ac->has_memory(current_machine, as,
3214 int128_get64(range->addr.start),
3215 MR_SIZE(range->addr.size) + 1)) {
3216 qemu_printf(" %s", fvi->ac->name);
3217 }
3218 }
3219 }
3220 qemu_printf("\n");
3221 range++;
3222 }
3223
3224 #if !defined(CONFIG_USER_ONLY)
3225 if (fvi->dispatch_tree && view->root) {
3226 mtree_print_dispatch(view->dispatch, view->root);
3227 }
3228 #endif
3229
3230 qemu_printf("\n");
3231 }
3232
3233 static gboolean mtree_info_flatview_free(gpointer key, gpointer value,
3234 gpointer user_data)
3235 {
3236 FlatView *view = key;
3237 GArray *fv_address_spaces = value;
3238
3239 g_array_unref(fv_address_spaces);
3240 flatview_unref(view);
3241
3242 return true;
3243 }
3244
3245 void mtree_info(bool flatview, bool dispatch_tree, bool owner, bool disabled)
3246 {
3247 MemoryRegionListHead ml_head;
3248 MemoryRegionList *ml, *ml2;
3249 AddressSpace *as;
3250
3251 if (flatview) {
3252 FlatView *view;
3253 struct FlatViewInfo fvi = {
3254 .counter = 0,
3255 .dispatch_tree = dispatch_tree,
3256 .owner = owner,
3257 };
3258 GArray *fv_address_spaces;
3259 GHashTable *views = g_hash_table_new(g_direct_hash, g_direct_equal);
3260 AccelClass *ac = ACCEL_GET_CLASS(current_accel());
3261
3262 if (ac->has_memory) {
3263 fvi.ac = ac;
3264 }
3265
3266 /* Gather all FVs in one table */
3267 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3268 view = address_space_get_flatview(as);
3269
3270 fv_address_spaces = g_hash_table_lookup(views, view);
3271 if (!fv_address_spaces) {
3272 fv_address_spaces = g_array_new(false, false, sizeof(as));
3273 g_hash_table_insert(views, view, fv_address_spaces);
3274 }
3275
3276 g_array_append_val(fv_address_spaces, as);
3277 }
3278
3279 /* Print */
3280 g_hash_table_foreach(views, mtree_print_flatview, &fvi);
3281
3282 /* Free */
3283 g_hash_table_foreach_remove(views, mtree_info_flatview_free, 0);
3284 g_hash_table_unref(views);
3285
3286 return;
3287 }
3288
3289 QTAILQ_INIT(&ml_head);
3290
3291 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3292 qemu_printf("address-space: %s\n", as->name);
3293 mtree_print_mr(as->root, 1, 0, &ml_head, owner, disabled);
3294 qemu_printf("\n");
3295 }
3296
3297 /* print aliased regions */
3298 QTAILQ_FOREACH(ml, &ml_head, mrqueue) {
3299 qemu_printf("memory-region: %s\n", memory_region_name(ml->mr));
3300 mtree_print_mr(ml->mr, 1, 0, &ml_head, owner, disabled);
3301 qemu_printf("\n");
3302 }
3303
3304 QTAILQ_FOREACH_SAFE(ml, &ml_head, mrqueue, ml2) {
3305 g_free(ml);
3306 }
3307 }
3308
3309 void memory_region_init_ram(MemoryRegion *mr,
3310 Object *owner,
3311 const char *name,
3312 uint64_t size,
3313 Error **errp)
3314 {
3315 DeviceState *owner_dev;
3316 Error *err = NULL;
3317
3318 memory_region_init_ram_nomigrate(mr, owner, name, size, &err);
3319 if (err) {
3320 error_propagate(errp, err);
3321 return;
3322 }
3323 /* This will assert if owner is neither NULL nor a DeviceState.
3324 * We only want the owner here for the purposes of defining a
3325 * unique name for migration. TODO: Ideally we should implement
3326 * a naming scheme for Objects which are not DeviceStates, in
3327 * which case we can relax this restriction.
3328 */
3329 owner_dev = DEVICE(owner);
3330 vmstate_register_ram(mr, owner_dev);
3331 }
3332
3333 void memory_region_init_rom(MemoryRegion *mr,
3334 Object *owner,
3335 const char *name,
3336 uint64_t size,
3337 Error **errp)
3338 {
3339 DeviceState *owner_dev;
3340 Error *err = NULL;
3341
3342 memory_region_init_rom_nomigrate(mr, owner, name, size, &err);
3343 if (err) {
3344 error_propagate(errp, err);
3345 return;
3346 }
3347 /* This will assert if owner is neither NULL nor a DeviceState.
3348 * We only want the owner here for the purposes of defining a
3349 * unique name for migration. TODO: Ideally we should implement
3350 * a naming scheme for Objects which are not DeviceStates, in
3351 * which case we can relax this restriction.
3352 */
3353 owner_dev = DEVICE(owner);
3354 vmstate_register_ram(mr, owner_dev);
3355 }
3356
3357 void memory_region_init_rom_device(MemoryRegion *mr,
3358 Object *owner,
3359 const MemoryRegionOps *ops,
3360 void *opaque,
3361 const char *name,
3362 uint64_t size,
3363 Error **errp)
3364 {
3365 DeviceState *owner_dev;
3366 Error *err = NULL;
3367
3368 memory_region_init_rom_device_nomigrate(mr, owner, ops, opaque,
3369 name, size, &err);
3370 if (err) {
3371 error_propagate(errp, err);
3372 return;
3373 }
3374 /* This will assert if owner is neither NULL nor a DeviceState.
3375 * We only want the owner here for the purposes of defining a
3376 * unique name for migration. TODO: Ideally we should implement
3377 * a naming scheme for Objects which are not DeviceStates, in
3378 * which case we can relax this restriction.
3379 */
3380 owner_dev = DEVICE(owner);
3381 vmstate_register_ram(mr, owner_dev);
3382 }
3383
3384 /*
3385 * Support softmmu builds with CONFIG_FUZZ using a weak symbol and a stub for
3386 * the fuzz_dma_read_cb callback
3387 */
3388 #ifdef CONFIG_FUZZ
3389 void __attribute__((weak)) fuzz_dma_read_cb(size_t addr,
3390 size_t len,
3391 MemoryRegion *mr)
3392 {
3393 }
3394 #endif
3395
3396 static const TypeInfo memory_region_info = {
3397 .parent = TYPE_OBJECT,
3398 .name = TYPE_MEMORY_REGION,
3399 .class_size = sizeof(MemoryRegionClass),
3400 .instance_size = sizeof(MemoryRegion),
3401 .instance_init = memory_region_initfn,
3402 .instance_finalize = memory_region_finalize,
3403 };
3404
3405 static const TypeInfo iommu_memory_region_info = {
3406 .parent = TYPE_MEMORY_REGION,
3407 .name = TYPE_IOMMU_MEMORY_REGION,
3408 .class_size = sizeof(IOMMUMemoryRegionClass),
3409 .instance_size = sizeof(IOMMUMemoryRegion),
3410 .instance_init = iommu_memory_region_initfn,
3411 .abstract = true,
3412 };
3413
3414 static const TypeInfo ram_discard_manager_info = {
3415 .parent = TYPE_INTERFACE,
3416 .name = TYPE_RAM_DISCARD_MANAGER,
3417 .class_size = sizeof(RamDiscardManagerClass),
3418 };
3419
3420 static void memory_register_types(void)
3421 {
3422 type_register_static(&memory_region_info);
3423 type_register_static(&iommu_memory_region_info);
3424 type_register_static(&ram_discard_manager_info);
3425 }
3426
3427 type_init(memory_register_types)
QEMU mirror