]> git.proxmox.com Git - mirror_qemu.git/blob - memory.c
projects / mirror_qemu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
memory: hide memory_region_sync_dirty_bitmap behind DirtyBitmapSnapshot
[mirror_qemu.git] / memory.c
1 /*
2 * Physical memory management
3 *
4 * Copyright 2011 Red Hat, Inc. and/or its affiliates
5 *
6 * Authors:
7 * Avi Kivity <avi@redhat.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
14 */
15
16 #include "qemu/osdep.h"
17 #include "qapi/error.h"
18 #include "qemu-common.h"
19 #include "cpu.h"
20 #include "exec/memory.h"
21 #include "exec/address-spaces.h"
22 #include "exec/ioport.h"
23 #include "qapi/visitor.h"
24 #include "qemu/bitops.h"
25 #include "qemu/error-report.h"
26 #include "qom/object.h"
27 #include "trace-root.h"
28
29 #include "exec/memory-internal.h"
30 #include "exec/ram_addr.h"
31 #include "sysemu/kvm.h"
32 #include "sysemu/sysemu.h"
33 #include "hw/misc/mmio_interface.h"
34 #include "hw/qdev-properties.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 static bool global_dirty_log = false;
43
44 static QTAILQ_HEAD(memory_listeners, 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, \
119 memory_listeners, link) { \
120 if (_listener->_callback) { \
121 _listener->_callback(_listener, ##_args); \
122 } \
123 } \
124 break; \
125 default: \
126 abort(); \
127 } \
128 } while (0)
129
130 #define MEMORY_LISTENER_CALL(_as, _callback, _direction, _section, _args...) \
131 do { \
132 MemoryListener *_listener; \
133 struct memory_listeners_as *list = &(_as)->listeners; \
134 \
135 switch (_direction) { \
136 case Forward: \
137 QTAILQ_FOREACH(_listener, list, link_as) { \
138 if (_listener->_callback) { \
139 _listener->_callback(_listener, _section, ##_args); \
140 } \
141 } \
142 break; \
143 case Reverse: \
144 QTAILQ_FOREACH_REVERSE(_listener, list, memory_listeners_as, \
145 link_as) { \
146 if (_listener->_callback) { \
147 _listener->_callback(_listener, _section, ##_args); \
148 } \
149 } \
150 break; \
151 default: \
152 abort(); \
153 } \
154 } while (0)
155
156 /* No need to ref/unref .mr, the FlatRange keeps it alive. */
157 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...) \
158 do { \
159 MemoryRegionSection mrs = section_from_flat_range(fr, \
160 address_space_to_flatview(as)); \
161 MEMORY_LISTENER_CALL(as, callback, dir, &mrs, ##_args); \
162 } while(0)
163
164 struct CoalescedMemoryRange {
165 AddrRange addr;
166 QTAILQ_ENTRY(CoalescedMemoryRange) link;
167 };
168
169 struct MemoryRegionIoeventfd {
170 AddrRange addr;
171 bool match_data;
172 uint64_t data;
173 EventNotifier *e;
174 };
175
176 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd a,
177 MemoryRegionIoeventfd b)
178 {
179 if (int128_lt(a.addr.start, b.addr.start)) {
180 return true;
181 } else if (int128_gt(a.addr.start, b.addr.start)) {
182 return false;
183 } else if (int128_lt(a.addr.size, b.addr.size)) {
184 return true;
185 } else if (int128_gt(a.addr.size, b.addr.size)) {
186 return false;
187 } else if (a.match_data < b.match_data) {
188 return true;
189 } else if (a.match_data > b.match_data) {
190 return false;
191 } else if (a.match_data) {
192 if (a.data < b.data) {
193 return true;
194 } else if (a.data > b.data) {
195 return false;
196 }
197 }
198 if (a.e < b.e) {
199 return true;
200 } else if (a.e > b.e) {
201 return false;
202 }
203 return false;
204 }
205
206 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd a,
207 MemoryRegionIoeventfd b)
208 {
209 return !memory_region_ioeventfd_before(a, b)
210 && !memory_region_ioeventfd_before(b, a);
211 }
212
213 typedef struct FlatRange FlatRange;
214
215 /* Range of memory in the global map. Addresses are absolute. */
216 struct FlatRange {
217 MemoryRegion *mr;
218 hwaddr offset_in_region;
219 AddrRange addr;
220 uint8_t dirty_log_mask;
221 bool romd_mode;
222 bool readonly;
223 };
224
225 /* Flattened global view of current active memory hierarchy. Kept in sorted
226 * order.
227 */
228 struct FlatView {
229 struct rcu_head rcu;
230 unsigned ref;
231 FlatRange *ranges;
232 unsigned nr;
233 unsigned nr_allocated;
234 struct AddressSpaceDispatch *dispatch;
235 MemoryRegion *root;
236 };
237
238 typedef struct AddressSpaceOps AddressSpaceOps;
239
240 #define FOR_EACH_FLAT_RANGE(var, view) \
241 for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
242
243 static inline MemoryRegionSection
244 section_from_flat_range(FlatRange *fr, FlatView *fv)
245 {
246 return (MemoryRegionSection) {
247 .mr = fr->mr,
248 .fv = fv,
249 .offset_within_region = fr->offset_in_region,
250 .size = fr->addr.size,
251 .offset_within_address_space = int128_get64(fr->addr.start),
252 .readonly = fr->readonly,
253 };
254 }
255
256 static bool flatrange_equal(FlatRange *a, FlatRange *b)
257 {
258 return a->mr == b->mr
259 && addrrange_equal(a->addr, b->addr)
260 && a->offset_in_region == b->offset_in_region
261 && a->romd_mode == b->romd_mode
262 && a->readonly == b->readonly;
263 }
264
265 static FlatView *flatview_new(MemoryRegion *mr_root)
266 {
267 FlatView *view;
268
269 view = g_new0(FlatView, 1);
270 view->ref = 1;
271 view->root = mr_root;
272 memory_region_ref(mr_root);
273 trace_flatview_new(view, mr_root);
274
275 return view;
276 }
277
278 /* Insert a range into a given position. Caller is responsible for maintaining
279 * sorting order.
280 */
281 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
282 {
283 if (view->nr == view->nr_allocated) {
284 view->nr_allocated = MAX(2 * view->nr, 10);
285 view->ranges = g_realloc(view->ranges,
286 view->nr_allocated * sizeof(*view->ranges));
287 }
288 memmove(view->ranges + pos + 1, view->ranges + pos,
289 (view->nr - pos) * sizeof(FlatRange));
290 view->ranges[pos] = *range;
291 memory_region_ref(range->mr);
292 ++view->nr;
293 }
294
295 static void flatview_destroy(FlatView *view)
296 {
297 int i;
298
299 trace_flatview_destroy(view, view->root);
300 if (view->dispatch) {
301 address_space_dispatch_free(view->dispatch);
302 }
303 for (i = 0; i < view->nr; i++) {
304 memory_region_unref(view->ranges[i].mr);
305 }
306 g_free(view->ranges);
307 memory_region_unref(view->root);
308 g_free(view);
309 }
310
311 static bool flatview_ref(FlatView *view)
312 {
313 return atomic_fetch_inc_nonzero(&view->ref) > 0;
314 }
315
316 static void flatview_unref(FlatView *view)
317 {
318 if (atomic_fetch_dec(&view->ref) == 1) {
319 trace_flatview_destroy_rcu(view, view->root);
320 assert(view->root);
321 call_rcu(view, flatview_destroy, rcu);
322 }
323 }
324
325 FlatView *address_space_to_flatview(AddressSpace *as)
326 {
327 return atomic_rcu_read(&as->current_map);
328 }
329
330 AddressSpaceDispatch *flatview_to_dispatch(FlatView *fv)
331 {
332 return fv->dispatch;
333 }
334
335 AddressSpaceDispatch *address_space_to_dispatch(AddressSpace *as)
336 {
337 return flatview_to_dispatch(address_space_to_flatview(as));
338 }
339
340 static bool can_merge(FlatRange *r1, FlatRange *r2)
341 {
342 return int128_eq(addrrange_end(r1->addr), r2->addr.start)
343 && r1->mr == r2->mr
344 && int128_eq(int128_add(int128_make64(r1->offset_in_region),
345 r1->addr.size),
346 int128_make64(r2->offset_in_region))
347 && r1->dirty_log_mask == r2->dirty_log_mask
348 && r1->romd_mode == r2->romd_mode
349 && r1->readonly == r2->readonly;
350 }
351
352 /* Attempt to simplify a view by merging adjacent ranges */
353 static void flatview_simplify(FlatView *view)
354 {
355 unsigned i, j;
356
357 i = 0;
358 while (i < view->nr) {
359 j = i + 1;
360 while (j < view->nr
361 && can_merge(&view->ranges[j-1], &view->ranges[j])) {
362 int128_addto(&view->ranges[i].addr.size, view->ranges[j].addr.size);
363 ++j;
364 }
365 ++i;
366 memmove(&view->ranges[i], &view->ranges[j],
367 (view->nr - j) * sizeof(view->ranges[j]));
368 view->nr -= j - i;
369 }
370 }
371
372 static bool memory_region_big_endian(MemoryRegion *mr)
373 {
374 #ifdef TARGET_WORDS_BIGENDIAN
375 return mr->ops->endianness != DEVICE_LITTLE_ENDIAN;
376 #else
377 return mr->ops->endianness == DEVICE_BIG_ENDIAN;
378 #endif
379 }
380
381 static bool memory_region_wrong_endianness(MemoryRegion *mr)
382 {
383 #ifdef TARGET_WORDS_BIGENDIAN
384 return mr->ops->endianness == DEVICE_LITTLE_ENDIAN;
385 #else
386 return mr->ops->endianness == DEVICE_BIG_ENDIAN;
387 #endif
388 }
389
390 static void adjust_endianness(MemoryRegion *mr, uint64_t *data, unsigned size)
391 {
392 if (memory_region_wrong_endianness(mr)) {
393 switch (size) {
394 case 1:
395 break;
396 case 2:
397 *data = bswap16(*data);
398 break;
399 case 4:
400 *data = bswap32(*data);
401 break;
402 case 8:
403 *data = bswap64(*data);
404 break;
405 default:
406 abort();
407 }
408 }
409 }
410
411 static hwaddr memory_region_to_absolute_addr(MemoryRegion *mr, hwaddr offset)
412 {
413 MemoryRegion *root;
414 hwaddr abs_addr = offset;
415
416 abs_addr += mr->addr;
417 for (root = mr; root->container; ) {
418 root = root->container;
419 abs_addr += root->addr;
420 }
421
422 return abs_addr;
423 }
424
425 static int get_cpu_index(void)
426 {
427 if (current_cpu) {
428 return current_cpu->cpu_index;
429 }
430 return -1;
431 }
432
433 static MemTxResult memory_region_oldmmio_read_accessor(MemoryRegion *mr,
434 hwaddr addr,
435 uint64_t *value,
436 unsigned size,
437 unsigned shift,
438 uint64_t mask,
439 MemTxAttrs attrs)
440 {
441 uint64_t tmp;
442
443 tmp = mr->ops->old_mmio.read[ctz32(size)](mr->opaque, addr);
444 if (mr->subpage) {
445 trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
446 } else if (mr == &io_mem_notdirty) {
447 /* Accesses to code which has previously been translated into a TB show
448 * up in the MMIO path, as accesses to the io_mem_notdirty
449 * MemoryRegion. */
450 trace_memory_region_tb_read(get_cpu_index(), addr, tmp, size);
451 } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
452 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
453 trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
454 }
455 *value |= (tmp & mask) << shift;
456 return MEMTX_OK;
457 }
458
459 static MemTxResult memory_region_read_accessor(MemoryRegion *mr,
460 hwaddr addr,
461 uint64_t *value,
462 unsigned size,
463 unsigned shift,
464 uint64_t mask,
465 MemTxAttrs attrs)
466 {
467 uint64_t tmp;
468
469 tmp = mr->ops->read(mr->opaque, addr, size);
470 if (mr->subpage) {
471 trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
472 } else if (mr == &io_mem_notdirty) {
473 /* Accesses to code which has previously been translated into a TB show
474 * up in the MMIO path, as accesses to the io_mem_notdirty
475 * MemoryRegion. */
476 trace_memory_region_tb_read(get_cpu_index(), addr, tmp, size);
477 } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
478 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
479 trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
480 }
481 *value |= (tmp & mask) << shift;
482 return MEMTX_OK;
483 }
484
485 static MemTxResult memory_region_read_with_attrs_accessor(MemoryRegion *mr,
486 hwaddr addr,
487 uint64_t *value,
488 unsigned size,
489 unsigned shift,
490 uint64_t mask,
491 MemTxAttrs attrs)
492 {
493 uint64_t tmp = 0;
494 MemTxResult r;
495
496 r = mr->ops->read_with_attrs(mr->opaque, addr, &tmp, size, attrs);
497 if (mr->subpage) {
498 trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
499 } else if (mr == &io_mem_notdirty) {
500 /* Accesses to code which has previously been translated into a TB show
501 * up in the MMIO path, as accesses to the io_mem_notdirty
502 * MemoryRegion. */
503 trace_memory_region_tb_read(get_cpu_index(), addr, tmp, size);
504 } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
505 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
506 trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
507 }
508 *value |= (tmp & mask) << shift;
509 return r;
510 }
511
512 static MemTxResult memory_region_oldmmio_write_accessor(MemoryRegion *mr,
513 hwaddr addr,
514 uint64_t *value,
515 unsigned size,
516 unsigned shift,
517 uint64_t mask,
518 MemTxAttrs attrs)
519 {
520 uint64_t tmp;
521
522 tmp = (*value >> shift) & mask;
523 if (mr->subpage) {
524 trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
525 } else if (mr == &io_mem_notdirty) {
526 /* Accesses to code which has previously been translated into a TB show
527 * up in the MMIO path, as accesses to the io_mem_notdirty
528 * MemoryRegion. */
529 trace_memory_region_tb_write(get_cpu_index(), addr, tmp, size);
530 } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
531 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
532 trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
533 }
534 mr->ops->old_mmio.write[ctz32(size)](mr->opaque, addr, tmp);
535 return MEMTX_OK;
536 }
537
538 static MemTxResult memory_region_write_accessor(MemoryRegion *mr,
539 hwaddr addr,
540 uint64_t *value,
541 unsigned size,
542 unsigned shift,
543 uint64_t mask,
544 MemTxAttrs attrs)
545 {
546 uint64_t tmp;
547
548 tmp = (*value >> shift) & mask;
549 if (mr->subpage) {
550 trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
551 } else if (mr == &io_mem_notdirty) {
552 /* Accesses to code which has previously been translated into a TB show
553 * up in the MMIO path, as accesses to the io_mem_notdirty
554 * MemoryRegion. */
555 trace_memory_region_tb_write(get_cpu_index(), addr, tmp, size);
556 } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
557 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
558 trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
559 }
560 mr->ops->write(mr->opaque, addr, tmp, size);
561 return MEMTX_OK;
562 }
563
564 static MemTxResult memory_region_write_with_attrs_accessor(MemoryRegion *mr,
565 hwaddr addr,
566 uint64_t *value,
567 unsigned size,
568 unsigned shift,
569 uint64_t mask,
570 MemTxAttrs attrs)
571 {
572 uint64_t tmp;
573
574 tmp = (*value >> shift) & mask;
575 if (mr->subpage) {
576 trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
577 } else if (mr == &io_mem_notdirty) {
578 /* Accesses to code which has previously been translated into a TB show
579 * up in the MMIO path, as accesses to the io_mem_notdirty
580 * MemoryRegion. */
581 trace_memory_region_tb_write(get_cpu_index(), addr, tmp, size);
582 } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
583 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
584 trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
585 }
586 return mr->ops->write_with_attrs(mr->opaque, addr, tmp, size, attrs);
587 }
588
589 static MemTxResult access_with_adjusted_size(hwaddr addr,
590 uint64_t *value,
591 unsigned size,
592 unsigned access_size_min,
593 unsigned access_size_max,
594 MemTxResult (*access_fn)
595 (MemoryRegion *mr,
596 hwaddr addr,
597 uint64_t *value,
598 unsigned size,
599 unsigned shift,
600 uint64_t mask,
601 MemTxAttrs attrs),
602 MemoryRegion *mr,
603 MemTxAttrs attrs)
604 {
605 uint64_t access_mask;
606 unsigned access_size;
607 unsigned i;
608 MemTxResult r = MEMTX_OK;
609
610 if (!access_size_min) {
611 access_size_min = 1;
612 }
613 if (!access_size_max) {
614 access_size_max = 4;
615 }
616
617 /* FIXME: support unaligned access? */
618 access_size = MAX(MIN(size, access_size_max), access_size_min);
619 access_mask = -1ULL >> (64 - access_size * 8);
620 if (memory_region_big_endian(mr)) {
621 for (i = 0; i < size; i += access_size) {
622 r |= access_fn(mr, addr + i, value, access_size,
623 (size - access_size - i) * 8, access_mask, attrs);
624 }
625 } else {
626 for (i = 0; i < size; i += access_size) {
627 r |= access_fn(mr, addr + i, value, access_size, i * 8,
628 access_mask, attrs);
629 }
630 }
631 return r;
632 }
633
634 static AddressSpace *memory_region_to_address_space(MemoryRegion *mr)
635 {
636 AddressSpace *as;
637
638 while (mr->container) {
639 mr = mr->container;
640 }
641 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
642 if (mr == as->root) {
643 return as;
644 }
645 }
646 return NULL;
647 }
648
649 /* Render a memory region into the global view. Ranges in @view obscure
650 * ranges in @mr.
651 */
652 static void render_memory_region(FlatView *view,
653 MemoryRegion *mr,
654 Int128 base,
655 AddrRange clip,
656 bool readonly)
657 {
658 MemoryRegion *subregion;
659 unsigned i;
660 hwaddr offset_in_region;
661 Int128 remain;
662 Int128 now;
663 FlatRange fr;
664 AddrRange tmp;
665
666 if (!mr->enabled) {
667 return;
668 }
669
670 int128_addto(&base, int128_make64(mr->addr));
671 readonly |= mr->readonly;
672
673 tmp = addrrange_make(base, mr->size);
674
675 if (!addrrange_intersects(tmp, clip)) {
676 return;
677 }
678
679 clip = addrrange_intersection(tmp, clip);
680
681 if (mr->alias) {
682 int128_subfrom(&base, int128_make64(mr->alias->addr));
683 int128_subfrom(&base, int128_make64(mr->alias_offset));
684 render_memory_region(view, mr->alias, base, clip, readonly);
685 return;
686 }
687
688 /* Render subregions in priority order. */
689 QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
690 render_memory_region(view, subregion, base, clip, readonly);
691 }
692
693 if (!mr->terminates) {
694 return;
695 }
696
697 offset_in_region = int128_get64(int128_sub(clip.start, base));
698 base = clip.start;
699 remain = clip.size;
700
701 fr.mr = mr;
702 fr.dirty_log_mask = memory_region_get_dirty_log_mask(mr);
703 fr.romd_mode = mr->romd_mode;
704 fr.readonly = readonly;
705
706 /* Render the region itself into any gaps left by the current view. */
707 for (i = 0; i < view->nr && int128_nz(remain); ++i) {
708 if (int128_ge(base, addrrange_end(view->ranges[i].addr))) {
709 continue;
710 }
711 if (int128_lt(base, view->ranges[i].addr.start)) {
712 now = int128_min(remain,
713 int128_sub(view->ranges[i].addr.start, base));
714 fr.offset_in_region = offset_in_region;
715 fr.addr = addrrange_make(base, now);
716 flatview_insert(view, i, &fr);
717 ++i;
718 int128_addto(&base, now);
719 offset_in_region += int128_get64(now);
720 int128_subfrom(&remain, now);
721 }
722 now = int128_sub(int128_min(int128_add(base, remain),
723 addrrange_end(view->ranges[i].addr)),
724 base);
725 int128_addto(&base, now);
726 offset_in_region += int128_get64(now);
727 int128_subfrom(&remain, now);
728 }
729 if (int128_nz(remain)) {
730 fr.offset_in_region = offset_in_region;
731 fr.addr = addrrange_make(base, remain);
732 flatview_insert(view, i, &fr);
733 }
734 }
735
736 static MemoryRegion *memory_region_get_flatview_root(MemoryRegion *mr)
737 {
738 while (mr->enabled) {
739 if (mr->alias) {
740 if (!mr->alias_offset && int128_ge(mr->size, mr->alias->size)) {
741 /* The alias is included in its entirety. Use it as
742 * the "real" root, so that we can share more FlatViews.
743 */
744 mr = mr->alias;
745 continue;
746 }
747 } else if (!mr->terminates) {
748 unsigned int found = 0;
749 MemoryRegion *child, *next = NULL;
750 QTAILQ_FOREACH(child, &mr->subregions, subregions_link) {
751 if (child->enabled) {
752 if (++found > 1) {
753 next = NULL;
754 break;
755 }
756 if (!child->addr && int128_ge(mr->size, child->size)) {
757 /* A child is included in its entirety. If it's the only
758 * enabled one, use it in the hope of finding an alias down the
759 * way. This will also let us share FlatViews.
760 */
761 next = child;
762 }
763 }
764 }
765 if (found == 0) {
766 return NULL;
767 }
768 if (next) {
769 mr = next;
770 continue;
771 }
772 }
773
774 return mr;
775 }
776
777 return NULL;
778 }
779
780 /* Render a memory topology into a list of disjoint absolute ranges. */
781 static FlatView *generate_memory_topology(MemoryRegion *mr)
782 {
783 int i;
784 FlatView *view;
785
786 view = flatview_new(mr);
787
788 if (mr) {
789 render_memory_region(view, mr, int128_zero(),
790 addrrange_make(int128_zero(), int128_2_64()), false);
791 }
792 flatview_simplify(view);
793
794 view->dispatch = address_space_dispatch_new(view);
795 for (i = 0; i < view->nr; i++) {
796 MemoryRegionSection mrs =
797 section_from_flat_range(&view->ranges[i], view);
798 flatview_add_to_dispatch(view, &mrs);
799 }
800 address_space_dispatch_compact(view->dispatch);
801 g_hash_table_replace(flat_views, mr, view);
802
803 return view;
804 }
805
806 static void address_space_add_del_ioeventfds(AddressSpace *as,
807 MemoryRegionIoeventfd *fds_new,
808 unsigned fds_new_nb,
809 MemoryRegionIoeventfd *fds_old,
810 unsigned fds_old_nb)
811 {
812 unsigned iold, inew;
813 MemoryRegionIoeventfd *fd;
814 MemoryRegionSection section;
815
816 /* Generate a symmetric difference of the old and new fd sets, adding
817 * and deleting as necessary.
818 */
819
820 iold = inew = 0;
821 while (iold < fds_old_nb || inew < fds_new_nb) {
822 if (iold < fds_old_nb
823 && (inew == fds_new_nb
824 || memory_region_ioeventfd_before(fds_old[iold],
825 fds_new[inew]))) {
826 fd = &fds_old[iold];
827 section = (MemoryRegionSection) {
828 .fv = address_space_to_flatview(as),
829 .offset_within_address_space = int128_get64(fd->addr.start),
830 .size = fd->addr.size,
831 };
832 MEMORY_LISTENER_CALL(as, eventfd_del, Forward, &section,
833 fd->match_data, fd->data, fd->e);
834 ++iold;
835 } else if (inew < fds_new_nb
836 && (iold == fds_old_nb
837 || memory_region_ioeventfd_before(fds_new[inew],
838 fds_old[iold]))) {
839 fd = &fds_new[inew];
840 section = (MemoryRegionSection) {
841 .fv = address_space_to_flatview(as),
842 .offset_within_address_space = int128_get64(fd->addr.start),
843 .size = fd->addr.size,
844 };
845 MEMORY_LISTENER_CALL(as, eventfd_add, Reverse, &section,
846 fd->match_data, fd->data, fd->e);
847 ++inew;
848 } else {
849 ++iold;
850 ++inew;
851 }
852 }
853 }
854
855 static FlatView *address_space_get_flatview(AddressSpace *as)
856 {
857 FlatView *view;
858
859 rcu_read_lock();
860 do {
861 view = address_space_to_flatview(as);
862 /* If somebody has replaced as->current_map concurrently,
863 * flatview_ref returns false.
864 */
865 } while (!flatview_ref(view));
866 rcu_read_unlock();
867 return view;
868 }
869
870 static void address_space_update_ioeventfds(AddressSpace *as)
871 {
872 FlatView *view;
873 FlatRange *fr;
874 unsigned ioeventfd_nb = 0;
875 MemoryRegionIoeventfd *ioeventfds = NULL;
876 AddrRange tmp;
877 unsigned i;
878
879 view = address_space_get_flatview(as);
880 FOR_EACH_FLAT_RANGE(fr, view) {
881 for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
882 tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
883 int128_sub(fr->addr.start,
884 int128_make64(fr->offset_in_region)));
885 if (addrrange_intersects(fr->addr, tmp)) {
886 ++ioeventfd_nb;
887 ioeventfds = g_realloc(ioeventfds,
888 ioeventfd_nb * sizeof(*ioeventfds));
889 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
890 ioeventfds[ioeventfd_nb-1].addr = tmp;
891 }
892 }
893 }
894
895 address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
896 as->ioeventfds, as->ioeventfd_nb);
897
898 g_free(as->ioeventfds);
899 as->ioeventfds = ioeventfds;
900 as->ioeventfd_nb = ioeventfd_nb;
901 flatview_unref(view);
902 }
903
904 static void address_space_update_topology_pass(AddressSpace *as,
905 const FlatView *old_view,
906 const FlatView *new_view,
907 bool adding)
908 {
909 unsigned iold, inew;
910 FlatRange *frold, *frnew;
911
912 /* Generate a symmetric difference of the old and new memory maps.
913 * Kill ranges in the old map, and instantiate ranges in the new map.
914 */
915 iold = inew = 0;
916 while (iold < old_view->nr || inew < new_view->nr) {
917 if (iold < old_view->nr) {
918 frold = &old_view->ranges[iold];
919 } else {
920 frold = NULL;
921 }
922 if (inew < new_view->nr) {
923 frnew = &new_view->ranges[inew];
924 } else {
925 frnew = NULL;
926 }
927
928 if (frold
929 && (!frnew
930 || int128_lt(frold->addr.start, frnew->addr.start)
931 || (int128_eq(frold->addr.start, frnew->addr.start)
932 && !flatrange_equal(frold, frnew)))) {
933 /* In old but not in new, or in both but attributes changed. */
934
935 if (!adding) {
936 MEMORY_LISTENER_UPDATE_REGION(frold, as, Reverse, region_del);
937 }
938
939 ++iold;
940 } else if (frold && frnew && flatrange_equal(frold, frnew)) {
941 /* In both and unchanged (except logging may have changed) */
942
943 if (adding) {
944 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_nop);
945 if (frnew->dirty_log_mask & ~frold->dirty_log_mask) {
946 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, log_start,
947 frold->dirty_log_mask,
948 frnew->dirty_log_mask);
949 }
950 if (frold->dirty_log_mask & ~frnew->dirty_log_mask) {
951 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Reverse, log_stop,
952 frold->dirty_log_mask,
953 frnew->dirty_log_mask);
954 }
955 }
956
957 ++iold;
958 ++inew;
959 } else {
960 /* In new */
961
962 if (adding) {
963 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_add);
964 }
965
966 ++inew;
967 }
968 }
969 }
970
971 static void flatviews_init(void)
972 {
973 static FlatView *empty_view;
974
975 if (flat_views) {
976 return;
977 }
978
979 flat_views = g_hash_table_new_full(g_direct_hash, g_direct_equal, NULL,
980 (GDestroyNotify) flatview_unref);
981 if (!empty_view) {
982 empty_view = generate_memory_topology(NULL);
983 /* We keep it alive forever in the global variable. */
984 flatview_ref(empty_view);
985 } else {
986 g_hash_table_replace(flat_views, NULL, empty_view);
987 flatview_ref(empty_view);
988 }
989 }
990
991 static void flatviews_reset(void)
992 {
993 AddressSpace *as;
994
995 if (flat_views) {
996 g_hash_table_unref(flat_views);
997 flat_views = NULL;
998 }
999 flatviews_init();
1000
1001 /* Render unique FVs */
1002 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1003 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1004
1005 if (g_hash_table_lookup(flat_views, physmr)) {
1006 continue;
1007 }
1008
1009 generate_memory_topology(physmr);
1010 }
1011 }
1012
1013 static void address_space_set_flatview(AddressSpace *as)
1014 {
1015 FlatView *old_view = address_space_to_flatview(as);
1016 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1017 FlatView *new_view = g_hash_table_lookup(flat_views, physmr);
1018
1019 assert(new_view);
1020
1021 if (old_view == new_view) {
1022 return;
1023 }
1024
1025 if (old_view) {
1026 flatview_ref(old_view);
1027 }
1028
1029 flatview_ref(new_view);
1030
1031 if (!QTAILQ_EMPTY(&as->listeners)) {
1032 FlatView tmpview = { .nr = 0 }, *old_view2 = old_view;
1033
1034 if (!old_view2) {
1035 old_view2 = &tmpview;
1036 }
1037 address_space_update_topology_pass(as, old_view2, new_view, false);
1038 address_space_update_topology_pass(as, old_view2, new_view, true);
1039 }
1040
1041 /* Writes are protected by the BQL. */
1042 atomic_rcu_set(&as->current_map, new_view);
1043 if (old_view) {
1044 flatview_unref(old_view);
1045 }
1046
1047 /* Note that all the old MemoryRegions are still alive up to this
1048 * point. This relieves most MemoryListeners from the need to
1049 * ref/unref the MemoryRegions they get---unless they use them
1050 * outside the iothread mutex, in which case precise reference
1051 * counting is necessary.
1052 */
1053 if (old_view) {
1054 flatview_unref(old_view);
1055 }
1056 }
1057
1058 static void address_space_update_topology(AddressSpace *as)
1059 {
1060 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1061
1062 flatviews_init();
1063 if (!g_hash_table_lookup(flat_views, physmr)) {
1064 generate_memory_topology(physmr);
1065 }
1066 address_space_set_flatview(as);
1067 }
1068
1069 void memory_region_transaction_begin(void)
1070 {
1071 qemu_flush_coalesced_mmio_buffer();
1072 ++memory_region_transaction_depth;
1073 }
1074
1075 void memory_region_transaction_commit(void)
1076 {
1077 AddressSpace *as;
1078
1079 assert(memory_region_transaction_depth);
1080 assert(qemu_mutex_iothread_locked());
1081
1082 --memory_region_transaction_depth;
1083 if (!memory_region_transaction_depth) {
1084 if (memory_region_update_pending) {
1085 flatviews_reset();
1086
1087 MEMORY_LISTENER_CALL_GLOBAL(begin, Forward);
1088
1089 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1090 address_space_set_flatview(as);
1091 address_space_update_ioeventfds(as);
1092 }
1093 memory_region_update_pending = false;
1094 ioeventfd_update_pending = false;
1095 MEMORY_LISTENER_CALL_GLOBAL(commit, Forward);
1096 } else if (ioeventfd_update_pending) {
1097 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1098 address_space_update_ioeventfds(as);
1099 }
1100 ioeventfd_update_pending = false;
1101 }
1102 }
1103 }
1104
1105 static void memory_region_destructor_none(MemoryRegion *mr)
1106 {
1107 }
1108
1109 static void memory_region_destructor_ram(MemoryRegion *mr)
1110 {
1111 qemu_ram_free(mr->ram_block);
1112 }
1113
1114 static bool memory_region_need_escape(char c)
1115 {
1116 return c == '/' || c == '[' || c == '\\' || c == ']';
1117 }
1118
1119 static char *memory_region_escape_name(const char *name)
1120 {
1121 const char *p;
1122 char *escaped, *q;
1123 uint8_t c;
1124 size_t bytes = 0;
1125
1126 for (p = name; *p; p++) {
1127 bytes += memory_region_need_escape(*p) ? 4 : 1;
1128 }
1129 if (bytes == p - name) {
1130 return g_memdup(name, bytes + 1);
1131 }
1132
1133 escaped = g_malloc(bytes + 1);
1134 for (p = name, q = escaped; *p; p++) {
1135 c = *p;
1136 if (unlikely(memory_region_need_escape(c))) {
1137 *q++ = '\\';
1138 *q++ = 'x';
1139 *q++ = "0123456789abcdef"[c >> 4];
1140 c = "0123456789abcdef"[c & 15];
1141 }
1142 *q++ = c;
1143 }
1144 *q = 0;
1145 return escaped;
1146 }
1147
1148 static void memory_region_do_init(MemoryRegion *mr,
1149 Object *owner,
1150 const char *name,
1151 uint64_t size)
1152 {
1153 mr->size = int128_make64(size);
1154 if (size == UINT64_MAX) {
1155 mr->size = int128_2_64();
1156 }
1157 mr->name = g_strdup(name);
1158 mr->owner = owner;
1159 mr->ram_block = NULL;
1160
1161 if (name) {
1162 char *escaped_name = memory_region_escape_name(name);
1163 char *name_array = g_strdup_printf("%s[*]", escaped_name);
1164
1165 if (!owner) {
1166 owner = container_get(qdev_get_machine(), "/unattached");
1167 }
1168
1169 object_property_add_child(owner, name_array, OBJECT(mr), &error_abort);
1170 object_unref(OBJECT(mr));
1171 g_free(name_array);
1172 g_free(escaped_name);
1173 }
1174 }
1175
1176 void memory_region_init(MemoryRegion *mr,
1177 Object *owner,
1178 const char *name,
1179 uint64_t size)
1180 {
1181 object_initialize(mr, sizeof(*mr), TYPE_MEMORY_REGION);
1182 memory_region_do_init(mr, owner, name, size);
1183 }
1184
1185 static void memory_region_get_addr(Object *obj, Visitor *v, const char *name,
1186 void *opaque, Error **errp)
1187 {
1188 MemoryRegion *mr = MEMORY_REGION(obj);
1189 uint64_t value = mr->addr;
1190
1191 visit_type_uint64(v, name, &value, errp);
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 gchar *path = (gchar *)"";
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->global_locking = true;
1246 mr->destructor = memory_region_destructor_none;
1247 QTAILQ_INIT(&mr->subregions);
1248 QTAILQ_INIT(&mr->coalesced);
1249
1250 op = object_property_add(OBJECT(mr), "container",
1251 "link<" TYPE_MEMORY_REGION ">",
1252 memory_region_get_container,
1253 NULL, /* memory_region_set_container */
1254 NULL, NULL, &error_abort);
1255 op->resolve = memory_region_resolve_container;
1256
1257 object_property_add(OBJECT(mr), "addr", "uint64",
1258 memory_region_get_addr,
1259 NULL, /* memory_region_set_addr */
1260 NULL, NULL, &error_abort);
1261 object_property_add(OBJECT(mr), "priority", "uint32",
1262 memory_region_get_priority,
1263 NULL, /* memory_region_set_priority */
1264 NULL, NULL, &error_abort);
1265 object_property_add(OBJECT(mr), "size", "uint64",
1266 memory_region_get_size,
1267 NULL, /* memory_region_set_size, */
1268 NULL, NULL, &error_abort);
1269 }
1270
1271 static void iommu_memory_region_initfn(Object *obj)
1272 {
1273 MemoryRegion *mr = MEMORY_REGION(obj);
1274
1275 mr->is_iommu = true;
1276 }
1277
1278 static uint64_t unassigned_mem_read(void *opaque, hwaddr addr,
1279 unsigned size)
1280 {
1281 #ifdef DEBUG_UNASSIGNED
1282 printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
1283 #endif
1284 if (current_cpu != NULL) {
1285 cpu_unassigned_access(current_cpu, addr, false, false, 0, size);
1286 }
1287 return 0;
1288 }
1289
1290 static void unassigned_mem_write(void *opaque, hwaddr addr,
1291 uint64_t val, unsigned size)
1292 {
1293 #ifdef DEBUG_UNASSIGNED
1294 printf("Unassigned mem write " TARGET_FMT_plx " = 0x%"PRIx64"\n", addr, val);
1295 #endif
1296 if (current_cpu != NULL) {
1297 cpu_unassigned_access(current_cpu, addr, true, false, 0, size);
1298 }
1299 }
1300
1301 static bool unassigned_mem_accepts(void *opaque, hwaddr addr,
1302 unsigned size, bool is_write)
1303 {
1304 return false;
1305 }
1306
1307 const MemoryRegionOps unassigned_mem_ops = {
1308 .valid.accepts = unassigned_mem_accepts,
1309 .endianness = DEVICE_NATIVE_ENDIAN,
1310 };
1311
1312 static uint64_t memory_region_ram_device_read(void *opaque,
1313 hwaddr addr, unsigned size)
1314 {
1315 MemoryRegion *mr = opaque;
1316 uint64_t data = (uint64_t)~0;
1317
1318 switch (size) {
1319 case 1:
1320 data = *(uint8_t *)(mr->ram_block->host + addr);
1321 break;
1322 case 2:
1323 data = *(uint16_t *)(mr->ram_block->host + addr);
1324 break;
1325 case 4:
1326 data = *(uint32_t *)(mr->ram_block->host + addr);
1327 break;
1328 case 8:
1329 data = *(uint64_t *)(mr->ram_block->host + addr);
1330 break;
1331 }
1332
1333 trace_memory_region_ram_device_read(get_cpu_index(), mr, addr, data, size);
1334
1335 return data;
1336 }
1337
1338 static void memory_region_ram_device_write(void *opaque, hwaddr addr,
1339 uint64_t data, unsigned size)
1340 {
1341 MemoryRegion *mr = opaque;
1342
1343 trace_memory_region_ram_device_write(get_cpu_index(), mr, addr, data, size);
1344
1345 switch (size) {
1346 case 1:
1347 *(uint8_t *)(mr->ram_block->host + addr) = (uint8_t)data;
1348 break;
1349 case 2:
1350 *(uint16_t *)(mr->ram_block->host + addr) = (uint16_t)data;
1351 break;
1352 case 4:
1353 *(uint32_t *)(mr->ram_block->host + addr) = (uint32_t)data;
1354 break;
1355 case 8:
1356 *(uint64_t *)(mr->ram_block->host + addr) = data;
1357 break;
1358 }
1359 }
1360
1361 static const MemoryRegionOps ram_device_mem_ops = {
1362 .read = memory_region_ram_device_read,
1363 .write = memory_region_ram_device_write,
1364 .endianness = DEVICE_HOST_ENDIAN,
1365 .valid = {
1366 .min_access_size = 1,
1367 .max_access_size = 8,
1368 .unaligned = true,
1369 },
1370 .impl = {
1371 .min_access_size = 1,
1372 .max_access_size = 8,
1373 .unaligned = true,
1374 },
1375 };
1376
1377 bool memory_region_access_valid(MemoryRegion *mr,
1378 hwaddr addr,
1379 unsigned size,
1380 bool is_write)
1381 {
1382 int access_size_min, access_size_max;
1383 int access_size, i;
1384
1385 if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1386 return false;
1387 }
1388
1389 if (!mr->ops->valid.accepts) {
1390 return true;
1391 }
1392
1393 access_size_min = mr->ops->valid.min_access_size;
1394 if (!mr->ops->valid.min_access_size) {
1395 access_size_min = 1;
1396 }
1397
1398 access_size_max = mr->ops->valid.max_access_size;
1399 if (!mr->ops->valid.max_access_size) {
1400 access_size_max = 4;
1401 }
1402
1403 access_size = MAX(MIN(size, access_size_max), access_size_min);
1404 for (i = 0; i < size; i += access_size) {
1405 if (!mr->ops->valid.accepts(mr->opaque, addr + i, access_size,
1406 is_write)) {
1407 return false;
1408 }
1409 }
1410
1411 return true;
1412 }
1413
1414 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1415 hwaddr addr,
1416 uint64_t *pval,
1417 unsigned size,
1418 MemTxAttrs attrs)
1419 {
1420 *pval = 0;
1421
1422 if (mr->ops->read) {
1423 return access_with_adjusted_size(addr, pval, size,
1424 mr->ops->impl.min_access_size,
1425 mr->ops->impl.max_access_size,
1426 memory_region_read_accessor,
1427 mr, attrs);
1428 } else if (mr->ops->read_with_attrs) {
1429 return access_with_adjusted_size(addr, pval, size,
1430 mr->ops->impl.min_access_size,
1431 mr->ops->impl.max_access_size,
1432 memory_region_read_with_attrs_accessor,
1433 mr, attrs);
1434 } else {
1435 return access_with_adjusted_size(addr, pval, size, 1, 4,
1436 memory_region_oldmmio_read_accessor,
1437 mr, attrs);
1438 }
1439 }
1440
1441 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1442 hwaddr addr,
1443 uint64_t *pval,
1444 unsigned size,
1445 MemTxAttrs attrs)
1446 {
1447 MemTxResult r;
1448
1449 if (!memory_region_access_valid(mr, addr, size, false)) {
1450 *pval = unassigned_mem_read(mr, addr, size);
1451 return MEMTX_DECODE_ERROR;
1452 }
1453
1454 r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1455 adjust_endianness(mr, pval, size);
1456 return r;
1457 }
1458
1459 /* Return true if an eventfd was signalled */
1460 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1461 hwaddr addr,
1462 uint64_t data,
1463 unsigned size,
1464 MemTxAttrs attrs)
1465 {
1466 MemoryRegionIoeventfd ioeventfd = {
1467 .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1468 .data = data,
1469 };
1470 unsigned i;
1471
1472 for (i = 0; i < mr->ioeventfd_nb; i++) {
1473 ioeventfd.match_data = mr->ioeventfds[i].match_data;
1474 ioeventfd.e = mr->ioeventfds[i].e;
1475
1476 if (memory_region_ioeventfd_equal(ioeventfd, mr->ioeventfds[i])) {
1477 event_notifier_set(ioeventfd.e);
1478 return true;
1479 }
1480 }
1481
1482 return false;
1483 }
1484
1485 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1486 hwaddr addr,
1487 uint64_t data,
1488 unsigned size,
1489 MemTxAttrs attrs)
1490 {
1491 if (!memory_region_access_valid(mr, addr, size, true)) {
1492 unassigned_mem_write(mr, addr, data, size);
1493 return MEMTX_DECODE_ERROR;
1494 }
1495
1496 adjust_endianness(mr, &data, size);
1497
1498 if ((!kvm_eventfds_enabled()) &&
1499 memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1500 return MEMTX_OK;
1501 }
1502
1503 if (mr->ops->write) {
1504 return access_with_adjusted_size(addr, &data, size,
1505 mr->ops->impl.min_access_size,
1506 mr->ops->impl.max_access_size,
1507 memory_region_write_accessor, mr,
1508 attrs);
1509 } else if (mr->ops->write_with_attrs) {
1510 return
1511 access_with_adjusted_size(addr, &data, size,
1512 mr->ops->impl.min_access_size,
1513 mr->ops->impl.max_access_size,
1514 memory_region_write_with_attrs_accessor,
1515 mr, attrs);
1516 } else {
1517 return access_with_adjusted_size(addr, &data, size, 1, 4,
1518 memory_region_oldmmio_write_accessor,
1519 mr, attrs);
1520 }
1521 }
1522
1523 void memory_region_init_io(MemoryRegion *mr,
1524 Object *owner,
1525 const MemoryRegionOps *ops,
1526 void *opaque,
1527 const char *name,
1528 uint64_t size)
1529 {
1530 memory_region_init(mr, owner, name, size);
1531 mr->ops = ops ? ops : &unassigned_mem_ops;
1532 mr->opaque = opaque;
1533 mr->terminates = true;
1534 }
1535
1536 void memory_region_init_ram_nomigrate(MemoryRegion *mr,
1537 Object *owner,
1538 const char *name,
1539 uint64_t size,
1540 Error **errp)
1541 {
1542 memory_region_init(mr, owner, name, size);
1543 mr->ram = true;
1544 mr->terminates = true;
1545 mr->destructor = memory_region_destructor_ram;
1546 mr->ram_block = qemu_ram_alloc(size, mr, errp);
1547 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1548 }
1549
1550 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1551 Object *owner,
1552 const char *name,
1553 uint64_t size,
1554 uint64_t max_size,
1555 void (*resized)(const char*,
1556 uint64_t length,
1557 void *host),
1558 Error **errp)
1559 {
1560 memory_region_init(mr, owner, name, size);
1561 mr->ram = true;
1562 mr->terminates = true;
1563 mr->destructor = memory_region_destructor_ram;
1564 mr->ram_block = qemu_ram_alloc_resizeable(size, max_size, resized,
1565 mr, errp);
1566 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1567 }
1568
1569 #ifdef __linux__
1570 void memory_region_init_ram_from_file(MemoryRegion *mr,
1571 struct Object *owner,
1572 const char *name,
1573 uint64_t size,
1574 uint64_t align,
1575 bool share,
1576 const char *path,
1577 Error **errp)
1578 {
1579 memory_region_init(mr, owner, name, size);
1580 mr->ram = true;
1581 mr->terminates = true;
1582 mr->destructor = memory_region_destructor_ram;
1583 mr->align = align;
1584 mr->ram_block = qemu_ram_alloc_from_file(size, mr, share, path, errp);
1585 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1586 }
1587
1588 void memory_region_init_ram_from_fd(MemoryRegion *mr,
1589 struct Object *owner,
1590 const char *name,
1591 uint64_t size,
1592 bool share,
1593 int fd,
1594 Error **errp)
1595 {
1596 memory_region_init(mr, owner, name, size);
1597 mr->ram = true;
1598 mr->terminates = true;
1599 mr->destructor = memory_region_destructor_ram;
1600 mr->ram_block = qemu_ram_alloc_from_fd(size, mr, share, fd, errp);
1601 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1602 }
1603 #endif
1604
1605 void memory_region_init_ram_ptr(MemoryRegion *mr,
1606 Object *owner,
1607 const char *name,
1608 uint64_t size,
1609 void *ptr)
1610 {
1611 memory_region_init(mr, owner, name, size);
1612 mr->ram = true;
1613 mr->terminates = true;
1614 mr->destructor = memory_region_destructor_ram;
1615 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1616
1617 /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL. */
1618 assert(ptr != NULL);
1619 mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1620 }
1621
1622 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
1623 Object *owner,
1624 const char *name,
1625 uint64_t size,
1626 void *ptr)
1627 {
1628 memory_region_init_ram_ptr(mr, owner, name, size, ptr);
1629 mr->ram_device = true;
1630 mr->ops = &ram_device_mem_ops;
1631 mr->opaque = mr;
1632 }
1633
1634 void memory_region_init_alias(MemoryRegion *mr,
1635 Object *owner,
1636 const char *name,
1637 MemoryRegion *orig,
1638 hwaddr offset,
1639 uint64_t size)
1640 {
1641 memory_region_init(mr, owner, name, size);
1642 mr->alias = orig;
1643 mr->alias_offset = offset;
1644 }
1645
1646 void memory_region_init_rom_nomigrate(MemoryRegion *mr,
1647 struct Object *owner,
1648 const char *name,
1649 uint64_t size,
1650 Error **errp)
1651 {
1652 memory_region_init(mr, owner, name, size);
1653 mr->ram = true;
1654 mr->readonly = true;
1655 mr->terminates = true;
1656 mr->destructor = memory_region_destructor_ram;
1657 mr->ram_block = qemu_ram_alloc(size, mr, errp);
1658 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1659 }
1660
1661 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1662 Object *owner,
1663 const MemoryRegionOps *ops,
1664 void *opaque,
1665 const char *name,
1666 uint64_t size,
1667 Error **errp)
1668 {
1669 assert(ops);
1670 memory_region_init(mr, owner, name, size);
1671 mr->ops = ops;
1672 mr->opaque = opaque;
1673 mr->terminates = true;
1674 mr->rom_device = true;
1675 mr->destructor = memory_region_destructor_ram;
1676 mr->ram_block = qemu_ram_alloc(size, mr, errp);
1677 }
1678
1679 void memory_region_init_iommu(void *_iommu_mr,
1680 size_t instance_size,
1681 const char *mrtypename,
1682 Object *owner,
1683 const char *name,
1684 uint64_t size)
1685 {
1686 struct IOMMUMemoryRegion *iommu_mr;
1687 struct MemoryRegion *mr;
1688
1689 object_initialize(_iommu_mr, instance_size, mrtypename);
1690 mr = MEMORY_REGION(_iommu_mr);
1691 memory_region_do_init(mr, owner, name, size);
1692 iommu_mr = IOMMU_MEMORY_REGION(mr);
1693 mr->terminates = true; /* then re-forwards */
1694 QLIST_INIT(&iommu_mr->iommu_notify);
1695 iommu_mr->iommu_notify_flags = IOMMU_NOTIFIER_NONE;
1696 }
1697
1698 static void memory_region_finalize(Object *obj)
1699 {
1700 MemoryRegion *mr = MEMORY_REGION(obj);
1701
1702 assert(!mr->container);
1703
1704 /* We know the region is not visible in any address space (it
1705 * does not have a container and cannot be a root either because
1706 * it has no references, so we can blindly clear mr->enabled.
1707 * memory_region_set_enabled instead could trigger a transaction
1708 * and cause an infinite loop.
1709 */
1710 mr->enabled = false;
1711 memory_region_transaction_begin();
1712 while (!QTAILQ_EMPTY(&mr->subregions)) {
1713 MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1714 memory_region_del_subregion(mr, subregion);
1715 }
1716 memory_region_transaction_commit();
1717
1718 mr->destructor(mr);
1719 memory_region_clear_coalescing(mr);
1720 g_free((char *)mr->name);
1721 g_free(mr->ioeventfds);
1722 }
1723
1724 Object *memory_region_owner(MemoryRegion *mr)
1725 {
1726 Object *obj = OBJECT(mr);
1727 return obj->parent;
1728 }
1729
1730 void memory_region_ref(MemoryRegion *mr)
1731 {
1732 /* MMIO callbacks most likely will access data that belongs
1733 * to the owner, hence the need to ref/unref the owner whenever
1734 * the memory region is in use.
1735 *
1736 * The memory region is a child of its owner. As long as the
1737 * owner doesn't call unparent itself on the memory region,
1738 * ref-ing the owner will also keep the memory region alive.
1739 * Memory regions without an owner are supposed to never go away;
1740 * we do not ref/unref them because it slows down DMA sensibly.
1741 */
1742 if (mr && mr->owner) {
1743 object_ref(mr->owner);
1744 }
1745 }
1746
1747 void memory_region_unref(MemoryRegion *mr)
1748 {
1749 if (mr && mr->owner) {
1750 object_unref(mr->owner);
1751 }
1752 }
1753
1754 uint64_t memory_region_size(MemoryRegion *mr)
1755 {
1756 if (int128_eq(mr->size, int128_2_64())) {
1757 return UINT64_MAX;
1758 }
1759 return int128_get64(mr->size);
1760 }
1761
1762 const char *memory_region_name(const MemoryRegion *mr)
1763 {
1764 if (!mr->name) {
1765 ((MemoryRegion *)mr)->name =
1766 object_get_canonical_path_component(OBJECT(mr));
1767 }
1768 return mr->name;
1769 }
1770
1771 bool memory_region_is_ram_device(MemoryRegion *mr)
1772 {
1773 return mr->ram_device;
1774 }
1775
1776 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1777 {
1778 uint8_t mask = mr->dirty_log_mask;
1779 if (global_dirty_log && mr->ram_block) {
1780 mask |= (1 << DIRTY_MEMORY_MIGRATION);
1781 }
1782 return mask;
1783 }
1784
1785 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1786 {
1787 return memory_region_get_dirty_log_mask(mr) & (1 << client);
1788 }
1789
1790 static void memory_region_update_iommu_notify_flags(IOMMUMemoryRegion *iommu_mr)
1791 {
1792 IOMMUNotifierFlag flags = IOMMU_NOTIFIER_NONE;
1793 IOMMUNotifier *iommu_notifier;
1794 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1795
1796 IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
1797 flags |= iommu_notifier->notifier_flags;
1798 }
1799
1800 if (flags != iommu_mr->iommu_notify_flags && imrc->notify_flag_changed) {
1801 imrc->notify_flag_changed(iommu_mr,
1802 iommu_mr->iommu_notify_flags,
1803 flags);
1804 }
1805
1806 iommu_mr->iommu_notify_flags = flags;
1807 }
1808
1809 void memory_region_register_iommu_notifier(MemoryRegion *mr,
1810 IOMMUNotifier *n)
1811 {
1812 IOMMUMemoryRegion *iommu_mr;
1813
1814 if (mr->alias) {
1815 memory_region_register_iommu_notifier(mr->alias, n);
1816 return;
1817 }
1818
1819 /* We need to register for at least one bitfield */
1820 iommu_mr = IOMMU_MEMORY_REGION(mr);
1821 assert(n->notifier_flags != IOMMU_NOTIFIER_NONE);
1822 assert(n->start <= n->end);
1823 QLIST_INSERT_HEAD(&iommu_mr->iommu_notify, n, node);
1824 memory_region_update_iommu_notify_flags(iommu_mr);
1825 }
1826
1827 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr)
1828 {
1829 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1830
1831 if (imrc->get_min_page_size) {
1832 return imrc->get_min_page_size(iommu_mr);
1833 }
1834 return TARGET_PAGE_SIZE;
1835 }
1836
1837 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
1838 {
1839 MemoryRegion *mr = MEMORY_REGION(iommu_mr);
1840 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1841 hwaddr addr, granularity;
1842 IOMMUTLBEntry iotlb;
1843
1844 /* If the IOMMU has its own replay callback, override */
1845 if (imrc->replay) {
1846 imrc->replay(iommu_mr, n);
1847 return;
1848 }
1849
1850 granularity = memory_region_iommu_get_min_page_size(iommu_mr);
1851
1852 for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1853 iotlb = imrc->translate(iommu_mr, addr, IOMMU_NONE);
1854 if (iotlb.perm != IOMMU_NONE) {
1855 n->notify(n, &iotlb);
1856 }
1857
1858 /* if (2^64 - MR size) < granularity, it's possible to get an
1859 * infinite loop here. This should catch such a wraparound */
1860 if ((addr + granularity) < addr) {
1861 break;
1862 }
1863 }
1864 }
1865
1866 void memory_region_iommu_replay_all(IOMMUMemoryRegion *iommu_mr)
1867 {
1868 IOMMUNotifier *notifier;
1869
1870 IOMMU_NOTIFIER_FOREACH(notifier, iommu_mr) {
1871 memory_region_iommu_replay(iommu_mr, notifier);
1872 }
1873 }
1874
1875 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1876 IOMMUNotifier *n)
1877 {
1878 IOMMUMemoryRegion *iommu_mr;
1879
1880 if (mr->alias) {
1881 memory_region_unregister_iommu_notifier(mr->alias, n);
1882 return;
1883 }
1884 QLIST_REMOVE(n, node);
1885 iommu_mr = IOMMU_MEMORY_REGION(mr);
1886 memory_region_update_iommu_notify_flags(iommu_mr);
1887 }
1888
1889 void memory_region_notify_one(IOMMUNotifier *notifier,
1890 IOMMUTLBEntry *entry)
1891 {
1892 IOMMUNotifierFlag request_flags;
1893
1894 /*
1895 * Skip the notification if the notification does not overlap
1896 * with registered range.
1897 */
1898 if (notifier->start > entry->iova + entry->addr_mask ||
1899 notifier->end < entry->iova) {
1900 return;
1901 }
1902
1903 if (entry->perm & IOMMU_RW) {
1904 request_flags = IOMMU_NOTIFIER_MAP;
1905 } else {
1906 request_flags = IOMMU_NOTIFIER_UNMAP;
1907 }
1908
1909 if (notifier->notifier_flags & request_flags) {
1910 notifier->notify(notifier, entry);
1911 }
1912 }
1913
1914 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
1915 IOMMUTLBEntry entry)
1916 {
1917 IOMMUNotifier *iommu_notifier;
1918
1919 assert(memory_region_is_iommu(MEMORY_REGION(iommu_mr)));
1920
1921 IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
1922 memory_region_notify_one(iommu_notifier, &entry);
1923 }
1924 }
1925
1926 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
1927 enum IOMMUMemoryRegionAttr attr,
1928 void *data)
1929 {
1930 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1931
1932 if (!imrc->get_attr) {
1933 return -EINVAL;
1934 }
1935
1936 return imrc->get_attr(iommu_mr, attr, data);
1937 }
1938
1939 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1940 {
1941 uint8_t mask = 1 << client;
1942 uint8_t old_logging;
1943
1944 assert(client == DIRTY_MEMORY_VGA);
1945 old_logging = mr->vga_logging_count;
1946 mr->vga_logging_count += log ? 1 : -1;
1947 if (!!old_logging == !!mr->vga_logging_count) {
1948 return;
1949 }
1950
1951 memory_region_transaction_begin();
1952 mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1953 memory_region_update_pending |= mr->enabled;
1954 memory_region_transaction_commit();
1955 }
1956
1957 bool memory_region_get_dirty(MemoryRegion *mr, hwaddr addr,
1958 hwaddr size, unsigned client)
1959 {
1960 assert(mr->ram_block);
1961 return cpu_physical_memory_get_dirty(memory_region_get_ram_addr(mr) + addr,
1962 size, client);
1963 }
1964
1965 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1966 hwaddr size)
1967 {
1968 assert(mr->ram_block);
1969 cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr,
1970 size,
1971 memory_region_get_dirty_log_mask(mr));
1972 }
1973
1974 static void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
1975 {
1976 MemoryListener *listener;
1977 AddressSpace *as;
1978 FlatView *view;
1979 FlatRange *fr;
1980
1981 /* If the same address space has multiple log_sync listeners, we
1982 * visit that address space's FlatView multiple times. But because
1983 * log_sync listeners are rare, it's still cheaper than walking each
1984 * address space once.
1985 */
1986 QTAILQ_FOREACH(listener, &memory_listeners, link) {
1987 if (!listener->log_sync) {
1988 continue;
1989 }
1990 as = listener->address_space;
1991 view = address_space_get_flatview(as);
1992 FOR_EACH_FLAT_RANGE(fr, view) {
1993 if (fr->mr == mr) {
1994 MemoryRegionSection mrs = section_from_flat_range(fr, view);
1995 listener->log_sync(listener, &mrs);
1996 }
1997 }
1998 flatview_unref(view);
1999 }
2000 }
2001
2002 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
2003 hwaddr addr,
2004 hwaddr size,
2005 unsigned client)
2006 {
2007 assert(mr->ram_block);
2008 memory_region_sync_dirty_bitmap(mr);
2009 return cpu_physical_memory_snapshot_and_clear_dirty(
2010 memory_region_get_ram_addr(mr) + addr, size, client);
2011 }
2012
2013 bool memory_region_snapshot_get_dirty(MemoryRegion *mr, DirtyBitmapSnapshot *snap,
2014 hwaddr addr, hwaddr size)
2015 {
2016 assert(mr->ram_block);
2017 return cpu_physical_memory_snapshot_get_dirty(snap,
2018 memory_region_get_ram_addr(mr) + addr, size);
2019 }
2020
2021 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
2022 {
2023 if (mr->readonly != readonly) {
2024 memory_region_transaction_begin();
2025 mr->readonly = readonly;
2026 memory_region_update_pending |= mr->enabled;
2027 memory_region_transaction_commit();
2028 }
2029 }
2030
2031 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
2032 {
2033 if (mr->romd_mode != romd_mode) {
2034 memory_region_transaction_begin();
2035 mr->romd_mode = romd_mode;
2036 memory_region_update_pending |= mr->enabled;
2037 memory_region_transaction_commit();
2038 }
2039 }
2040
2041 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
2042 hwaddr size, unsigned client)
2043 {
2044 assert(mr->ram_block);
2045 cpu_physical_memory_test_and_clear_dirty(
2046 memory_region_get_ram_addr(mr) + addr, size, client);
2047 }
2048
2049 int memory_region_get_fd(MemoryRegion *mr)
2050 {
2051 int fd;
2052
2053 rcu_read_lock();
2054 while (mr->alias) {
2055 mr = mr->alias;
2056 }
2057 fd = mr->ram_block->fd;
2058 rcu_read_unlock();
2059
2060 return fd;
2061 }
2062
2063 void *memory_region_get_ram_ptr(MemoryRegion *mr)
2064 {
2065 void *ptr;
2066 uint64_t offset = 0;
2067
2068 rcu_read_lock();
2069 while (mr->alias) {
2070 offset += mr->alias_offset;
2071 mr = mr->alias;
2072 }
2073 assert(mr->ram_block);
2074 ptr = qemu_map_ram_ptr(mr->ram_block, offset);
2075 rcu_read_unlock();
2076
2077 return ptr;
2078 }
2079
2080 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset)
2081 {
2082 RAMBlock *block;
2083
2084 block = qemu_ram_block_from_host(ptr, false, offset);
2085 if (!block) {
2086 return NULL;
2087 }
2088
2089 return block->mr;
2090 }
2091
2092 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr)
2093 {
2094 return mr->ram_block ? mr->ram_block->offset : RAM_ADDR_INVALID;
2095 }
2096
2097 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
2098 {
2099 assert(mr->ram_block);
2100
2101 qemu_ram_resize(mr->ram_block, newsize, errp);
2102 }
2103
2104 static void memory_region_update_coalesced_range_as(MemoryRegion *mr, AddressSpace *as)
2105 {
2106 FlatView *view;
2107 FlatRange *fr;
2108 CoalescedMemoryRange *cmr;
2109 AddrRange tmp;
2110 MemoryRegionSection section;
2111
2112 view = address_space_get_flatview(as);
2113 FOR_EACH_FLAT_RANGE(fr, view) {
2114 if (fr->mr == mr) {
2115 section = (MemoryRegionSection) {
2116 .fv = view,
2117 .offset_within_address_space = int128_get64(fr->addr.start),
2118 .size = fr->addr.size,
2119 };
2120
2121 MEMORY_LISTENER_CALL(as, coalesced_mmio_del, Reverse, &section,
2122 int128_get64(fr->addr.start),
2123 int128_get64(fr->addr.size));
2124 QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
2125 tmp = addrrange_shift(cmr->addr,
2126 int128_sub(fr->addr.start,
2127 int128_make64(fr->offset_in_region)));
2128 if (!addrrange_intersects(tmp, fr->addr)) {
2129 continue;
2130 }
2131 tmp = addrrange_intersection(tmp, fr->addr);
2132 MEMORY_LISTENER_CALL(as, coalesced_mmio_add, Forward, &section,
2133 int128_get64(tmp.start),
2134 int128_get64(tmp.size));
2135 }
2136 }
2137 }
2138 flatview_unref(view);
2139 }
2140
2141 static void memory_region_update_coalesced_range(MemoryRegion *mr)
2142 {
2143 AddressSpace *as;
2144
2145 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2146 memory_region_update_coalesced_range_as(mr, as);
2147 }
2148 }
2149
2150 void memory_region_set_coalescing(MemoryRegion *mr)
2151 {
2152 memory_region_clear_coalescing(mr);
2153 memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
2154 }
2155
2156 void memory_region_add_coalescing(MemoryRegion *mr,
2157 hwaddr offset,
2158 uint64_t size)
2159 {
2160 CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
2161
2162 cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
2163 QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
2164 memory_region_update_coalesced_range(mr);
2165 memory_region_set_flush_coalesced(mr);
2166 }
2167
2168 void memory_region_clear_coalescing(MemoryRegion *mr)
2169 {
2170 CoalescedMemoryRange *cmr;
2171 bool updated = false;
2172
2173 qemu_flush_coalesced_mmio_buffer();
2174 mr->flush_coalesced_mmio = false;
2175
2176 while (!QTAILQ_EMPTY(&mr->coalesced)) {
2177 cmr = QTAILQ_FIRST(&mr->coalesced);
2178 QTAILQ_REMOVE(&mr->coalesced, cmr, link);
2179 g_free(cmr);
2180 updated = true;
2181 }
2182
2183 if (updated) {
2184 memory_region_update_coalesced_range(mr);
2185 }
2186 }
2187
2188 void memory_region_set_flush_coalesced(MemoryRegion *mr)
2189 {
2190 mr->flush_coalesced_mmio = true;
2191 }
2192
2193 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
2194 {
2195 qemu_flush_coalesced_mmio_buffer();
2196 if (QTAILQ_EMPTY(&mr->coalesced)) {
2197 mr->flush_coalesced_mmio = false;
2198 }
2199 }
2200
2201 void memory_region_clear_global_locking(MemoryRegion *mr)
2202 {
2203 mr->global_locking = false;
2204 }
2205
2206 static bool userspace_eventfd_warning;
2207
2208 void memory_region_add_eventfd(MemoryRegion *mr,
2209 hwaddr addr,
2210 unsigned size,
2211 bool match_data,
2212 uint64_t data,
2213 EventNotifier *e)
2214 {
2215 MemoryRegionIoeventfd mrfd = {
2216 .addr.start = int128_make64(addr),
2217 .addr.size = int128_make64(size),
2218 .match_data = match_data,
2219 .data = data,
2220 .e = e,
2221 };
2222 unsigned i;
2223
2224 if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
2225 userspace_eventfd_warning))) {
2226 userspace_eventfd_warning = true;
2227 error_report("Using eventfd without MMIO binding in KVM. "
2228 "Suboptimal performance expected");
2229 }
2230
2231 if (size) {
2232 adjust_endianness(mr, &mrfd.data, size);
2233 }
2234 memory_region_transaction_begin();
2235 for (i = 0; i < mr->ioeventfd_nb; ++i) {
2236 if (memory_region_ioeventfd_before(mrfd, mr->ioeventfds[i])) {
2237 break;
2238 }
2239 }
2240 ++mr->ioeventfd_nb;
2241 mr->ioeventfds = g_realloc(mr->ioeventfds,
2242 sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
2243 memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
2244 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
2245 mr->ioeventfds[i] = mrfd;
2246 ioeventfd_update_pending |= mr->enabled;
2247 memory_region_transaction_commit();
2248 }
2249
2250 void memory_region_del_eventfd(MemoryRegion *mr,
2251 hwaddr addr,
2252 unsigned size,
2253 bool match_data,
2254 uint64_t data,
2255 EventNotifier *e)
2256 {
2257 MemoryRegionIoeventfd mrfd = {
2258 .addr.start = int128_make64(addr),
2259 .addr.size = int128_make64(size),
2260 .match_data = match_data,
2261 .data = data,
2262 .e = e,
2263 };
2264 unsigned i;
2265
2266 if (size) {
2267 adjust_endianness(mr, &mrfd.data, size);
2268 }
2269 memory_region_transaction_begin();
2270 for (i = 0; i < mr->ioeventfd_nb; ++i) {
2271 if (memory_region_ioeventfd_equal(mrfd, mr->ioeventfds[i])) {
2272 break;
2273 }
2274 }
2275 assert(i != mr->ioeventfd_nb);
2276 memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
2277 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
2278 --mr->ioeventfd_nb;
2279 mr->ioeventfds = g_realloc(mr->ioeventfds,
2280 sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
2281 ioeventfd_update_pending |= mr->enabled;
2282 memory_region_transaction_commit();
2283 }
2284
2285 static void memory_region_update_container_subregions(MemoryRegion *subregion)
2286 {
2287 MemoryRegion *mr = subregion->container;
2288 MemoryRegion *other;
2289
2290 memory_region_transaction_begin();
2291
2292 memory_region_ref(subregion);
2293 QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
2294 if (subregion->priority >= other->priority) {
2295 QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
2296 goto done;
2297 }
2298 }
2299 QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
2300 done:
2301 memory_region_update_pending |= mr->enabled && subregion->enabled;
2302 memory_region_transaction_commit();
2303 }
2304
2305 static void memory_region_add_subregion_common(MemoryRegion *mr,
2306 hwaddr offset,
2307 MemoryRegion *subregion)
2308 {
2309 assert(!subregion->container);
2310 subregion->container = mr;
2311 subregion->addr = offset;
2312 memory_region_update_container_subregions(subregion);
2313 }
2314
2315 void memory_region_add_subregion(MemoryRegion *mr,
2316 hwaddr offset,
2317 MemoryRegion *subregion)
2318 {
2319 subregion->priority = 0;
2320 memory_region_add_subregion_common(mr, offset, subregion);
2321 }
2322
2323 void memory_region_add_subregion_overlap(MemoryRegion *mr,
2324 hwaddr offset,
2325 MemoryRegion *subregion,
2326 int priority)
2327 {
2328 subregion->priority = priority;
2329 memory_region_add_subregion_common(mr, offset, subregion);
2330 }
2331
2332 void memory_region_del_subregion(MemoryRegion *mr,
2333 MemoryRegion *subregion)
2334 {
2335 memory_region_transaction_begin();
2336 assert(subregion->container == mr);
2337 subregion->container = NULL;
2338 QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
2339 memory_region_unref(subregion);
2340 memory_region_update_pending |= mr->enabled && subregion->enabled;
2341 memory_region_transaction_commit();
2342 }
2343
2344 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
2345 {
2346 if (enabled == mr->enabled) {
2347 return;
2348 }
2349 memory_region_transaction_begin();
2350 mr->enabled = enabled;
2351 memory_region_update_pending = true;
2352 memory_region_transaction_commit();
2353 }
2354
2355 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
2356 {
2357 Int128 s = int128_make64(size);
2358
2359 if (size == UINT64_MAX) {
2360 s = int128_2_64();
2361 }
2362 if (int128_eq(s, mr->size)) {
2363 return;
2364 }
2365 memory_region_transaction_begin();
2366 mr->size = s;
2367 memory_region_update_pending = true;
2368 memory_region_transaction_commit();
2369 }
2370
2371 static void memory_region_readd_subregion(MemoryRegion *mr)
2372 {
2373 MemoryRegion *container = mr->container;
2374
2375 if (container) {
2376 memory_region_transaction_begin();
2377 memory_region_ref(mr);
2378 memory_region_del_subregion(container, mr);
2379 mr->container = container;
2380 memory_region_update_container_subregions(mr);
2381 memory_region_unref(mr);
2382 memory_region_transaction_commit();
2383 }
2384 }
2385
2386 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
2387 {
2388 if (addr != mr->addr) {
2389 mr->addr = addr;
2390 memory_region_readd_subregion(mr);
2391 }
2392 }
2393
2394 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
2395 {
2396 assert(mr->alias);
2397
2398 if (offset == mr->alias_offset) {
2399 return;
2400 }
2401
2402 memory_region_transaction_begin();
2403 mr->alias_offset = offset;
2404 memory_region_update_pending |= mr->enabled;
2405 memory_region_transaction_commit();
2406 }
2407
2408 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
2409 {
2410 return mr->align;
2411 }
2412
2413 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
2414 {
2415 const AddrRange *addr = addr_;
2416 const FlatRange *fr = fr_;
2417
2418 if (int128_le(addrrange_end(*addr), fr->addr.start)) {
2419 return -1;
2420 } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
2421 return 1;
2422 }
2423 return 0;
2424 }
2425
2426 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
2427 {
2428 return bsearch(&addr, view->ranges, view->nr,
2429 sizeof(FlatRange), cmp_flatrange_addr);
2430 }
2431
2432 bool memory_region_is_mapped(MemoryRegion *mr)
2433 {
2434 return mr->container ? true : false;
2435 }
2436
2437 /* Same as memory_region_find, but it does not add a reference to the
2438 * returned region. It must be called from an RCU critical section.
2439 */
2440 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2441 hwaddr addr, uint64_t size)
2442 {
2443 MemoryRegionSection ret = { .mr = NULL };
2444 MemoryRegion *root;
2445 AddressSpace *as;
2446 AddrRange range;
2447 FlatView *view;
2448 FlatRange *fr;
2449
2450 addr += mr->addr;
2451 for (root = mr; root->container; ) {
2452 root = root->container;
2453 addr += root->addr;
2454 }
2455
2456 as = memory_region_to_address_space(root);
2457 if (!as) {
2458 return ret;
2459 }
2460 range = addrrange_make(int128_make64(addr), int128_make64(size));
2461
2462 view = address_space_to_flatview(as);
2463 fr = flatview_lookup(view, range);
2464 if (!fr) {
2465 return ret;
2466 }
2467
2468 while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2469 --fr;
2470 }
2471
2472 ret.mr = fr->mr;
2473 ret.fv = view;
2474 range = addrrange_intersection(range, fr->addr);
2475 ret.offset_within_region = fr->offset_in_region;
2476 ret.offset_within_region += int128_get64(int128_sub(range.start,
2477 fr->addr.start));
2478 ret.size = range.size;
2479 ret.offset_within_address_space = int128_get64(range.start);
2480 ret.readonly = fr->readonly;
2481 return ret;
2482 }
2483
2484 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2485 hwaddr addr, uint64_t size)
2486 {
2487 MemoryRegionSection ret;
2488 rcu_read_lock();
2489 ret = memory_region_find_rcu(mr, addr, size);
2490 if (ret.mr) {
2491 memory_region_ref(ret.mr);
2492 }
2493 rcu_read_unlock();
2494 return ret;
2495 }
2496
2497 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2498 {
2499 MemoryRegion *mr;
2500
2501 rcu_read_lock();
2502 mr = memory_region_find_rcu(container, addr, 1).mr;
2503 rcu_read_unlock();
2504 return mr && mr != container;
2505 }
2506
2507 void memory_global_dirty_log_sync(void)
2508 {
2509 MemoryListener *listener;
2510 AddressSpace *as;
2511 FlatView *view;
2512 FlatRange *fr;
2513
2514 QTAILQ_FOREACH(listener, &memory_listeners, link) {
2515 if (!listener->log_sync) {
2516 continue;
2517 }
2518 as = listener->address_space;
2519 view = address_space_get_flatview(as);
2520 FOR_EACH_FLAT_RANGE(fr, view) {
2521 if (fr->dirty_log_mask) {
2522 MemoryRegionSection mrs = section_from_flat_range(fr, view);
2523
2524 listener->log_sync(listener, &mrs);
2525 }
2526 }
2527 flatview_unref(view);
2528 }
2529 }
2530
2531 static VMChangeStateEntry *vmstate_change;
2532
2533 void memory_global_dirty_log_start(void)
2534 {
2535 if (vmstate_change) {
2536 qemu_del_vm_change_state_handler(vmstate_change);
2537 vmstate_change = NULL;
2538 }
2539
2540 global_dirty_log = true;
2541
2542 MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2543
2544 /* Refresh DIRTY_LOG_MIGRATION bit. */
2545 memory_region_transaction_begin();
2546 memory_region_update_pending = true;
2547 memory_region_transaction_commit();
2548 }
2549
2550 static void memory_global_dirty_log_do_stop(void)
2551 {
2552 global_dirty_log = false;
2553
2554 /* Refresh DIRTY_LOG_MIGRATION bit. */
2555 memory_region_transaction_begin();
2556 memory_region_update_pending = true;
2557 memory_region_transaction_commit();
2558
2559 MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2560 }
2561
2562 static void memory_vm_change_state_handler(void *opaque, int running,
2563 RunState state)
2564 {
2565 if (running) {
2566 memory_global_dirty_log_do_stop();
2567
2568 if (vmstate_change) {
2569 qemu_del_vm_change_state_handler(vmstate_change);
2570 vmstate_change = NULL;
2571 }
2572 }
2573 }
2574
2575 void memory_global_dirty_log_stop(void)
2576 {
2577 if (!runstate_is_running()) {
2578 if (vmstate_change) {
2579 return;
2580 }
2581 vmstate_change = qemu_add_vm_change_state_handler(
2582 memory_vm_change_state_handler, NULL);
2583 return;
2584 }
2585
2586 memory_global_dirty_log_do_stop();
2587 }
2588
2589 static void listener_add_address_space(MemoryListener *listener,
2590 AddressSpace *as)
2591 {
2592 FlatView *view;
2593 FlatRange *fr;
2594
2595 if (listener->begin) {
2596 listener->begin(listener);
2597 }
2598 if (global_dirty_log) {
2599 if (listener->log_global_start) {
2600 listener->log_global_start(listener);
2601 }
2602 }
2603
2604 view = address_space_get_flatview(as);
2605 FOR_EACH_FLAT_RANGE(fr, view) {
2606 MemoryRegionSection section = section_from_flat_range(fr, view);
2607
2608 if (listener->region_add) {
2609 listener->region_add(listener, &section);
2610 }
2611 if (fr->dirty_log_mask && listener->log_start) {
2612 listener->log_start(listener, &section, 0, fr->dirty_log_mask);
2613 }
2614 }
2615 if (listener->commit) {
2616 listener->commit(listener);
2617 }
2618 flatview_unref(view);
2619 }
2620
2621 static void listener_del_address_space(MemoryListener *listener,
2622 AddressSpace *as)
2623 {
2624 FlatView *view;
2625 FlatRange *fr;
2626
2627 if (listener->begin) {
2628 listener->begin(listener);
2629 }
2630 view = address_space_get_flatview(as);
2631 FOR_EACH_FLAT_RANGE(fr, view) {
2632 MemoryRegionSection section = section_from_flat_range(fr, view);
2633
2634 if (fr->dirty_log_mask && listener->log_stop) {
2635 listener->log_stop(listener, &section, fr->dirty_log_mask, 0);
2636 }
2637 if (listener->region_del) {
2638 listener->region_del(listener, &section);
2639 }
2640 }
2641 if (listener->commit) {
2642 listener->commit(listener);
2643 }
2644 flatview_unref(view);
2645 }
2646
2647 void memory_listener_register(MemoryListener *listener, AddressSpace *as)
2648 {
2649 MemoryListener *other = NULL;
2650
2651 listener->address_space = as;
2652 if (QTAILQ_EMPTY(&memory_listeners)
2653 || listener->priority >= QTAILQ_LAST(&memory_listeners,
2654 memory_listeners)->priority) {
2655 QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
2656 } else {
2657 QTAILQ_FOREACH(other, &memory_listeners, link) {
2658 if (listener->priority < other->priority) {
2659 break;
2660 }
2661 }
2662 QTAILQ_INSERT_BEFORE(other, listener, link);
2663 }
2664
2665 if (QTAILQ_EMPTY(&as->listeners)
2666 || listener->priority >= QTAILQ_LAST(&as->listeners,
2667 memory_listeners)->priority) {
2668 QTAILQ_INSERT_TAIL(&as->listeners, listener, link_as);
2669 } else {
2670 QTAILQ_FOREACH(other, &as->listeners, link_as) {
2671 if (listener->priority < other->priority) {
2672 break;
2673 }
2674 }
2675 QTAILQ_INSERT_BEFORE(other, listener, link_as);
2676 }
2677
2678 listener_add_address_space(listener, as);
2679 }
2680
2681 void memory_listener_unregister(MemoryListener *listener)
2682 {
2683 if (!listener->address_space) {
2684 return;
2685 }
2686
2687 listener_del_address_space(listener, listener->address_space);
2688 QTAILQ_REMOVE(&memory_listeners, listener, link);
2689 QTAILQ_REMOVE(&listener->address_space->listeners, listener, link_as);
2690 listener->address_space = NULL;
2691 }
2692
2693 bool memory_region_request_mmio_ptr(MemoryRegion *mr, hwaddr addr)
2694 {
2695 void *host;
2696 unsigned size = 0;
2697 unsigned offset = 0;
2698 Object *new_interface;
2699
2700 if (!mr || !mr->ops->request_ptr) {
2701 return false;
2702 }
2703
2704 /*
2705 * Avoid an update if the request_ptr call
2706 * memory_region_invalidate_mmio_ptr which seems to be likely when we use
2707 * a cache.
2708 */
2709 memory_region_transaction_begin();
2710
2711 host = mr->ops->request_ptr(mr->opaque, addr - mr->addr, &size, &offset);
2712
2713 if (!host || !size) {
2714 memory_region_transaction_commit();
2715 return false;
2716 }
2717
2718 new_interface = object_new("mmio_interface");
2719 qdev_prop_set_uint64(DEVICE(new_interface), "start", offset);
2720 qdev_prop_set_uint64(DEVICE(new_interface), "end", offset + size - 1);
2721 qdev_prop_set_bit(DEVICE(new_interface), "ro", true);
2722 qdev_prop_set_ptr(DEVICE(new_interface), "host_ptr", host);
2723 qdev_prop_set_ptr(DEVICE(new_interface), "subregion", mr);
2724 object_property_set_bool(OBJECT(new_interface), true, "realized", NULL);
2725
2726 memory_region_transaction_commit();
2727 return true;
2728 }
2729
2730 typedef struct MMIOPtrInvalidate {
2731 MemoryRegion *mr;
2732 hwaddr offset;
2733 unsigned size;
2734 int busy;
2735 int allocated;
2736 } MMIOPtrInvalidate;
2737
2738 #define MAX_MMIO_INVALIDATE 10
2739 static MMIOPtrInvalidate mmio_ptr_invalidate_list[MAX_MMIO_INVALIDATE];
2740
2741 static void memory_region_do_invalidate_mmio_ptr(CPUState *cpu,
2742 run_on_cpu_data data)
2743 {
2744 MMIOPtrInvalidate *invalidate_data = (MMIOPtrInvalidate *)data.host_ptr;
2745 MemoryRegion *mr = invalidate_data->mr;
2746 hwaddr offset = invalidate_data->offset;
2747 unsigned size = invalidate_data->size;
2748 MemoryRegionSection section = memory_region_find(mr, offset, size);
2749
2750 qemu_mutex_lock_iothread();
2751
2752 /* Reset dirty so this doesn't happen later. */
2753 cpu_physical_memory_test_and_clear_dirty(offset, size, 1);
2754
2755 if (section.mr != mr) {
2756 /* memory_region_find add a ref on section.mr */
2757 memory_region_unref(section.mr);
2758 if (MMIO_INTERFACE(section.mr->owner)) {
2759 /* We found the interface just drop it. */
2760 object_property_set_bool(section.mr->owner, false, "realized",
2761 NULL);
2762 object_unref(section.mr->owner);
2763 object_unparent(section.mr->owner);
2764 }
2765 }
2766
2767 qemu_mutex_unlock_iothread();
2768
2769 if (invalidate_data->allocated) {
2770 g_free(invalidate_data);
2771 } else {
2772 invalidate_data->busy = 0;
2773 }
2774 }
2775
2776 void memory_region_invalidate_mmio_ptr(MemoryRegion *mr, hwaddr offset,
2777 unsigned size)
2778 {
2779 size_t i;
2780 MMIOPtrInvalidate *invalidate_data = NULL;
2781
2782 for (i = 0; i < MAX_MMIO_INVALIDATE; i++) {
2783 if (atomic_cmpxchg(&(mmio_ptr_invalidate_list[i].busy), 0, 1) == 0) {
2784 invalidate_data = &mmio_ptr_invalidate_list[i];
2785 break;
2786 }
2787 }
2788
2789 if (!invalidate_data) {
2790 invalidate_data = g_malloc0(sizeof(MMIOPtrInvalidate));
2791 invalidate_data->allocated = 1;
2792 }
2793
2794 invalidate_data->mr = mr;
2795 invalidate_data->offset = offset;
2796 invalidate_data->size = size;
2797
2798 async_safe_run_on_cpu(first_cpu, memory_region_do_invalidate_mmio_ptr,
2799 RUN_ON_CPU_HOST_PTR(invalidate_data));
2800 }
2801
2802 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
2803 {
2804 memory_region_ref(root);
2805 as->root = root;
2806 as->current_map = NULL;
2807 as->ioeventfd_nb = 0;
2808 as->ioeventfds = NULL;
2809 QTAILQ_INIT(&as->listeners);
2810 QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
2811 as->name = g_strdup(name ? name : "anonymous");
2812 address_space_update_topology(as);
2813 address_space_update_ioeventfds(as);
2814 }
2815
2816 static void do_address_space_destroy(AddressSpace *as)
2817 {
2818 assert(QTAILQ_EMPTY(&as->listeners));
2819
2820 flatview_unref(as->current_map);
2821 g_free(as->name);
2822 g_free(as->ioeventfds);
2823 memory_region_unref(as->root);
2824 }
2825
2826 void address_space_destroy(AddressSpace *as)
2827 {
2828 MemoryRegion *root = as->root;
2829
2830 /* Flush out anything from MemoryListeners listening in on this */
2831 memory_region_transaction_begin();
2832 as->root = NULL;
2833 memory_region_transaction_commit();
2834 QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
2835
2836 /* At this point, as->dispatch and as->current_map are dummy
2837 * entries that the guest should never use. Wait for the old
2838 * values to expire before freeing the data.
2839 */
2840 as->root = root;
2841 call_rcu(as, do_address_space_destroy, rcu);
2842 }
2843
2844 static const char *memory_region_type(MemoryRegion *mr)
2845 {
2846 if (memory_region_is_ram_device(mr)) {
2847 return "ramd";
2848 } else if (memory_region_is_romd(mr)) {
2849 return "romd";
2850 } else if (memory_region_is_rom(mr)) {
2851 return "rom";
2852 } else if (memory_region_is_ram(mr)) {
2853 return "ram";
2854 } else {
2855 return "i/o";
2856 }
2857 }
2858
2859 typedef struct MemoryRegionList MemoryRegionList;
2860
2861 struct MemoryRegionList {
2862 const MemoryRegion *mr;
2863 QTAILQ_ENTRY(MemoryRegionList) mrqueue;
2864 };
2865
2866 typedef QTAILQ_HEAD(mrqueue, MemoryRegionList) MemoryRegionListHead;
2867
2868 #define MR_SIZE(size) (int128_nz(size) ? (hwaddr)int128_get64( \
2869 int128_sub((size), int128_one())) : 0)
2870 #define MTREE_INDENT " "
2871
2872 static void mtree_print_mr(fprintf_function mon_printf, void *f,
2873 const MemoryRegion *mr, unsigned int level,
2874 hwaddr base,
2875 MemoryRegionListHead *alias_print_queue)
2876 {
2877 MemoryRegionList *new_ml, *ml, *next_ml;
2878 MemoryRegionListHead submr_print_queue;
2879 const MemoryRegion *submr;
2880 unsigned int i;
2881 hwaddr cur_start, cur_end;
2882
2883 if (!mr) {
2884 return;
2885 }
2886
2887 for (i = 0; i < level; i++) {
2888 mon_printf(f, MTREE_INDENT);
2889 }
2890
2891 cur_start = base + mr->addr;
2892 cur_end = cur_start + MR_SIZE(mr->size);
2893
2894 /*
2895 * Try to detect overflow of memory region. This should never
2896 * happen normally. When it happens, we dump something to warn the
2897 * user who is observing this.
2898 */
2899 if (cur_start < base || cur_end < cur_start) {
2900 mon_printf(f, "[DETECTED OVERFLOW!] ");
2901 }
2902
2903 if (mr->alias) {
2904 MemoryRegionList *ml;
2905 bool found = false;
2906
2907 /* check if the alias is already in the queue */
2908 QTAILQ_FOREACH(ml, alias_print_queue, mrqueue) {
2909 if (ml->mr == mr->alias) {
2910 found = true;
2911 }
2912 }
2913
2914 if (!found) {
2915 ml = g_new(MemoryRegionList, 1);
2916 ml->mr = mr->alias;
2917 QTAILQ_INSERT_TAIL(alias_print_queue, ml, mrqueue);
2918 }
2919 mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx
2920 " (prio %d, %s): alias %s @%s " TARGET_FMT_plx
2921 "-" TARGET_FMT_plx "%s\n",
2922 cur_start, cur_end,
2923 mr->priority,
2924 memory_region_type((MemoryRegion *)mr),
2925 memory_region_name(mr),
2926 memory_region_name(mr->alias),
2927 mr->alias_offset,
2928 mr->alias_offset + MR_SIZE(mr->size),
2929 mr->enabled ? "" : " [disabled]");
2930 } else {
2931 mon_printf(f,
2932 TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d, %s): %s%s\n",
2933 cur_start, cur_end,
2934 mr->priority,
2935 memory_region_type((MemoryRegion *)mr),
2936 memory_region_name(mr),
2937 mr->enabled ? "" : " [disabled]");
2938 }
2939
2940 QTAILQ_INIT(&submr_print_queue);
2941
2942 QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
2943 new_ml = g_new(MemoryRegionList, 1);
2944 new_ml->mr = submr;
2945 QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
2946 if (new_ml->mr->addr < ml->mr->addr ||
2947 (new_ml->mr->addr == ml->mr->addr &&
2948 new_ml->mr->priority > ml->mr->priority)) {
2949 QTAILQ_INSERT_BEFORE(ml, new_ml, mrqueue);
2950 new_ml = NULL;
2951 break;
2952 }
2953 }
2954 if (new_ml) {
2955 QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, mrqueue);
2956 }
2957 }
2958
2959 QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
2960 mtree_print_mr(mon_printf, f, ml->mr, level + 1, cur_start,
2961 alias_print_queue);
2962 }
2963
2964 QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, mrqueue, next_ml) {
2965 g_free(ml);
2966 }
2967 }
2968
2969 struct FlatViewInfo {
2970 fprintf_function mon_printf;
2971 void *f;
2972 int counter;
2973 bool dispatch_tree;
2974 };
2975
2976 static void mtree_print_flatview(gpointer key, gpointer value,
2977 gpointer user_data)
2978 {
2979 FlatView *view = key;
2980 GArray *fv_address_spaces = value;
2981 struct FlatViewInfo *fvi = user_data;
2982 fprintf_function p = fvi->mon_printf;
2983 void *f = fvi->f;
2984 FlatRange *range = &view->ranges[0];
2985 MemoryRegion *mr;
2986 int n = view->nr;
2987 int i;
2988 AddressSpace *as;
2989
2990 p(f, "FlatView #%d\n", fvi->counter);
2991 ++fvi->counter;
2992
2993 for (i = 0; i < fv_address_spaces->len; ++i) {
2994 as = g_array_index(fv_address_spaces, AddressSpace*, i);
2995 p(f, " AS \"%s\", root: %s", as->name, memory_region_name(as->root));
2996 if (as->root->alias) {
2997 p(f, ", alias %s", memory_region_name(as->root->alias));
2998 }
2999 p(f, "\n");
3000 }
3001
3002 p(f, " Root memory region: %s\n",
3003 view->root ? memory_region_name(view->root) : "(none)");
3004
3005 if (n <= 0) {
3006 p(f, MTREE_INDENT "No rendered FlatView\n\n");
3007 return;
3008 }
3009
3010 while (n--) {
3011 mr = range->mr;
3012 if (range->offset_in_region) {
3013 p(f, MTREE_INDENT TARGET_FMT_plx "-"
3014 TARGET_FMT_plx " (prio %d, %s): %s @" TARGET_FMT_plx "\n",
3015 int128_get64(range->addr.start),
3016 int128_get64(range->addr.start) + MR_SIZE(range->addr.size),
3017 mr->priority,
3018 range->readonly ? "rom" : memory_region_type(mr),
3019 memory_region_name(mr),
3020 range->offset_in_region);
3021 } else {
3022 p(f, MTREE_INDENT TARGET_FMT_plx "-"
3023 TARGET_FMT_plx " (prio %d, %s): %s\n",
3024 int128_get64(range->addr.start),
3025 int128_get64(range->addr.start) + MR_SIZE(range->addr.size),
3026 mr->priority,
3027 range->readonly ? "rom" : memory_region_type(mr),
3028 memory_region_name(mr));
3029 }
3030 range++;
3031 }
3032
3033 #if !defined(CONFIG_USER_ONLY)
3034 if (fvi->dispatch_tree && view->root) {
3035 mtree_print_dispatch(p, f, view->dispatch, view->root);
3036 }
3037 #endif
3038
3039 p(f, "\n");
3040 }
3041
3042 static gboolean mtree_info_flatview_free(gpointer key, gpointer value,
3043 gpointer user_data)
3044 {
3045 FlatView *view = key;
3046 GArray *fv_address_spaces = value;
3047
3048 g_array_unref(fv_address_spaces);
3049 flatview_unref(view);
3050
3051 return true;
3052 }
3053
3054 void mtree_info(fprintf_function mon_printf, void *f, bool flatview,
3055 bool dispatch_tree)
3056 {
3057 MemoryRegionListHead ml_head;
3058 MemoryRegionList *ml, *ml2;
3059 AddressSpace *as;
3060
3061 if (flatview) {
3062 FlatView *view;
3063 struct FlatViewInfo fvi = {
3064 .mon_printf = mon_printf,
3065 .f = f,
3066 .counter = 0,
3067 .dispatch_tree = dispatch_tree
3068 };
3069 GArray *fv_address_spaces;
3070 GHashTable *views = g_hash_table_new(g_direct_hash, g_direct_equal);
3071
3072 /* Gather all FVs in one table */
3073 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3074 view = address_space_get_flatview(as);
3075
3076 fv_address_spaces = g_hash_table_lookup(views, view);
3077 if (!fv_address_spaces) {
3078 fv_address_spaces = g_array_new(false, false, sizeof(as));
3079 g_hash_table_insert(views, view, fv_address_spaces);
3080 }
3081
3082 g_array_append_val(fv_address_spaces, as);
3083 }
3084
3085 /* Print */
3086 g_hash_table_foreach(views, mtree_print_flatview, &fvi);
3087
3088 /* Free */
3089 g_hash_table_foreach_remove(views, mtree_info_flatview_free, 0);
3090 g_hash_table_unref(views);
3091
3092 return;
3093 }
3094
3095 QTAILQ_INIT(&ml_head);
3096
3097 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3098 mon_printf(f, "address-space: %s\n", as->name);
3099 mtree_print_mr(mon_printf, f, as->root, 1, 0, &ml_head);
3100 mon_printf(f, "\n");
3101 }
3102
3103 /* print aliased regions */
3104 QTAILQ_FOREACH(ml, &ml_head, mrqueue) {
3105 mon_printf(f, "memory-region: %s\n", memory_region_name(ml->mr));
3106 mtree_print_mr(mon_printf, f, ml->mr, 1, 0, &ml_head);
3107 mon_printf(f, "\n");
3108 }
3109
3110 QTAILQ_FOREACH_SAFE(ml, &ml_head, mrqueue, ml2) {
3111 g_free(ml);
3112 }
3113 }
3114
3115 void memory_region_init_ram(MemoryRegion *mr,
3116 struct Object *owner,
3117 const char *name,
3118 uint64_t size,
3119 Error **errp)
3120 {
3121 DeviceState *owner_dev;
3122 Error *err = NULL;
3123
3124 memory_region_init_ram_nomigrate(mr, owner, name, size, &err);
3125 if (err) {
3126 error_propagate(errp, err);
3127 return;
3128 }
3129 /* This will assert if owner is neither NULL nor a DeviceState.
3130 * We only want the owner here for the purposes of defining a
3131 * unique name for migration. TODO: Ideally we should implement
3132 * a naming scheme for Objects which are not DeviceStates, in
3133 * which case we can relax this restriction.
3134 */
3135 owner_dev = DEVICE(owner);
3136 vmstate_register_ram(mr, owner_dev);
3137 }
3138
3139 void memory_region_init_rom(MemoryRegion *mr,
3140 struct Object *owner,
3141 const char *name,
3142 uint64_t size,
3143 Error **errp)
3144 {
3145 DeviceState *owner_dev;
3146 Error *err = NULL;
3147
3148 memory_region_init_rom_nomigrate(mr, owner, name, size, &err);
3149 if (err) {
3150 error_propagate(errp, err);
3151 return;
3152 }
3153 /* This will assert if owner is neither NULL nor a DeviceState.
3154 * We only want the owner here for the purposes of defining a
3155 * unique name for migration. TODO: Ideally we should implement
3156 * a naming scheme for Objects which are not DeviceStates, in
3157 * which case we can relax this restriction.
3158 */
3159 owner_dev = DEVICE(owner);
3160 vmstate_register_ram(mr, owner_dev);
3161 }
3162
3163 void memory_region_init_rom_device(MemoryRegion *mr,
3164 struct Object *owner,
3165 const MemoryRegionOps *ops,
3166 void *opaque,
3167 const char *name,
3168 uint64_t size,
3169 Error **errp)
3170 {
3171 DeviceState *owner_dev;
3172 Error *err = NULL;
3173
3174 memory_region_init_rom_device_nomigrate(mr, owner, ops, opaque,
3175 name, size, &err);
3176 if (err) {
3177 error_propagate(errp, err);
3178 return;
3179 }
3180 /* This will assert if owner is neither NULL nor a DeviceState.
3181 * We only want the owner here for the purposes of defining a
3182 * unique name for migration. TODO: Ideally we should implement
3183 * a naming scheme for Objects which are not DeviceStates, in
3184 * which case we can relax this restriction.
3185 */
3186 owner_dev = DEVICE(owner);
3187 vmstate_register_ram(mr, owner_dev);
3188 }
3189
3190 static const TypeInfo memory_region_info = {
3191 .parent = TYPE_OBJECT,
3192 .name = TYPE_MEMORY_REGION,
3193 .instance_size = sizeof(MemoryRegion),
3194 .instance_init = memory_region_initfn,
3195 .instance_finalize = memory_region_finalize,
3196 };
3197
3198 static const TypeInfo iommu_memory_region_info = {
3199 .parent = TYPE_MEMORY_REGION,
3200 .name = TYPE_IOMMU_MEMORY_REGION,
3201 .class_size = sizeof(IOMMUMemoryRegionClass),
3202 .instance_size = sizeof(IOMMUMemoryRegion),
3203 .instance_init = iommu_memory_region_initfn,
3204 .abstract = true,
3205 };
3206
3207 static void memory_register_types(void)
3208 {
3209 type_register_static(&memory_region_info);
3210 type_register_static(&iommu_memory_region_info);
3211 }
3212
3213 type_init(memory_register_types)
QEMU mirror