?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
i386: only parse the initrd_filename once for multiboot modules
[qemu/ericb.git] / memory.c
blobe9cd446968835793c47a34f307f0419212813010
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 "qapi/visitor.h"
23 #include "qemu/bitops.h"
24 #include "qemu/error-report.h"
25 #include "qom/object.h"
26 #include "trace-root.h"
27
28 #include "exec/memory-internal.h"
29 #include "exec/ram_addr.h"
30 #include "sysemu/kvm.h"
31 #include "sysemu/sysemu.h"
32 #include "hw/misc/mmio_interface.h"
33 #include "hw/qdev-properties.h"
34 #include "migration/vmstate.h"
35
36 //#define DEBUG_UNASSIGNED
37
38 static unsigned memory_region_transaction_depth;
39 static bool memory_region_update_pending;
40 static bool ioeventfd_update_pending;
41 static bool global_dirty_log = false;
42
43 static QTAILQ_HEAD(memory_listeners, MemoryListener) memory_listeners
44 = QTAILQ_HEAD_INITIALIZER(memory_listeners);
45
46 static QTAILQ_HEAD(, AddressSpace) address_spaces
47 = QTAILQ_HEAD_INITIALIZER(address_spaces);
48
49 static GHashTable *flat_views;
50
51 typedef struct AddrRange AddrRange;
52
53 /*
54 * Note that signed integers are needed for negative offsetting in aliases
55 * (large MemoryRegion::alias_offset).
56 */
57 struct AddrRange {
58 Int128 start;
59 Int128 size;
60 };
61
62 static AddrRange addrrange_make(Int128 start, Int128 size)
63 {
64 return (AddrRange) { start, size };
65 }
66
67 static bool addrrange_equal(AddrRange r1, AddrRange r2)
68 {
69 return int128_eq(r1.start, r2.start) && int128_eq(r1.size, r2.size);
70 }
71
72 static Int128 addrrange_end(AddrRange r)
73 {
74 return int128_add(r.start, r.size);
75 }
76
77 static AddrRange addrrange_shift(AddrRange range, Int128 delta)
78 {
79 int128_addto(&range.start, delta);
80 return range;
81 }
82
83 static bool addrrange_contains(AddrRange range, Int128 addr)
84 {
85 return int128_ge(addr, range.start)
86 && int128_lt(addr, addrrange_end(range));
87 }
88
89 static bool addrrange_intersects(AddrRange r1, AddrRange r2)
90 {
91 return addrrange_contains(r1, r2.start)
92 || addrrange_contains(r2, r1.start);
93 }
94
95 static AddrRange addrrange_intersection(AddrRange r1, AddrRange r2)
96 {
97 Int128 start = int128_max(r1.start, r2.start);
98 Int128 end = int128_min(addrrange_end(r1), addrrange_end(r2));
99 return addrrange_make(start, int128_sub(end, start));
100 }
101
102 enum ListenerDirection { Forward, Reverse };
103
104 #define MEMORY_LISTENER_CALL_GLOBAL(_callback, _direction, _args...) \
105 do { \
106 MemoryListener *_listener; \
107 \
108 switch (_direction) { \
109 case Forward: \
110 QTAILQ_FOREACH(_listener, &memory_listeners, link) { \
111 if (_listener->_callback) { \
112 _listener->_callback(_listener, ##_args); \
113 } \
114 } \
115 break; \
116 case Reverse: \
117 QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners, \
118 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 struct memory_listeners_as *list = &(_as)->listeners; \
133 \
134 switch (_direction) { \
135 case Forward: \
136 QTAILQ_FOREACH(_listener, list, link_as) { \
137 if (_listener->_callback) { \
138 _listener->_callback(_listener, _section, ##_args); \
139 } \
140 } \
141 break; \
142 case Reverse: \
143 QTAILQ_FOREACH_REVERSE(_listener, list, memory_listeners_as, \
144 link_as) { \
145 if (_listener->_callback) { \
146 _listener->_callback(_listener, _section, ##_args); \
147 } \
148 } \
149 break; \
150 default: \
151 abort(); \
152 } \
153 } while (0)
154
155 /* No need to ref/unref .mr, the FlatRange keeps it alive. */
156 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...) \
157 do { \
158 MemoryRegionSection mrs = section_from_flat_range(fr, \
159 address_space_to_flatview(as)); \
160 MEMORY_LISTENER_CALL(as, callback, dir, &mrs, ##_args); \
161 } while(0)
162
163 struct CoalescedMemoryRange {
164 AddrRange addr;
165 QTAILQ_ENTRY(CoalescedMemoryRange) link;
166 };
167
168 struct MemoryRegionIoeventfd {
169 AddrRange addr;
170 bool match_data;
171 uint64_t data;
172 EventNotifier *e;
173 };
174
175 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd *a,
176 MemoryRegionIoeventfd *b)
177 {
178 if (int128_lt(a->addr.start, b->addr.start)) {
179 return true;
180 } else if (int128_gt(a->addr.start, b->addr.start)) {
181 return false;
182 } else if (int128_lt(a->addr.size, b->addr.size)) {
183 return true;
184 } else if (int128_gt(a->addr.size, b->addr.size)) {
185 return false;
186 } else if (a->match_data < b->match_data) {
187 return true;
188 } else if (a->match_data > b->match_data) {
189 return false;
190 } else if (a->match_data) {
191 if (a->data < b->data) {
192 return true;
193 } else if (a->data > b->data) {
194 return false;
195 }
196 }
197 if (a->e < b->e) {
198 return true;
199 } else if (a->e > b->e) {
200 return false;
201 }
202 return false;
203 }
204
205 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd *a,
206 MemoryRegionIoeventfd *b)
207 {
208 return !memory_region_ioeventfd_before(a, b)
209 && !memory_region_ioeventfd_before(b, a);
210 }
211
212 /* Range of memory in the global map. Addresses are absolute. */
213 struct FlatRange {
214 MemoryRegion *mr;
215 hwaddr offset_in_region;
216 AddrRange addr;
217 uint8_t dirty_log_mask;
218 bool romd_mode;
219 bool readonly;
220 };
221
222 #define FOR_EACH_FLAT_RANGE(var, view) \
223 for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
224
225 static inline MemoryRegionSection
226 section_from_flat_range(FlatRange *fr, FlatView *fv)
227 {
228 return (MemoryRegionSection) {
229 .mr = fr->mr,
230 .fv = fv,
231 .offset_within_region = fr->offset_in_region,
232 .size = fr->addr.size,
233 .offset_within_address_space = int128_get64(fr->addr.start),
234 .readonly = fr->readonly,
235 };
236 }
237
238 static bool flatrange_equal(FlatRange *a, FlatRange *b)
239 {
240 return a->mr == b->mr
241 && addrrange_equal(a->addr, b->addr)
242 && a->offset_in_region == b->offset_in_region
243 && a->romd_mode == b->romd_mode
244 && a->readonly == b->readonly;
245 }
246
247 static FlatView *flatview_new(MemoryRegion *mr_root)
248 {
249 FlatView *view;
250
251 view = g_new0(FlatView, 1);
252 view->ref = 1;
253 view->root = mr_root;
254 memory_region_ref(mr_root);
255 trace_flatview_new(view, mr_root);
256
257 return view;
258 }
259
260 /* Insert a range into a given position. Caller is responsible for maintaining
261 * sorting order.
262 */
263 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
264 {
265 if (view->nr == view->nr_allocated) {
266 view->nr_allocated = MAX(2 * view->nr, 10);
267 view->ranges = g_realloc(view->ranges,
268 view->nr_allocated * sizeof(*view->ranges));
269 }
270 memmove(view->ranges + pos + 1, view->ranges + pos,
271 (view->nr - pos) * sizeof(FlatRange));
272 view->ranges[pos] = *range;
273 memory_region_ref(range->mr);
274 ++view->nr;
275 }
276
277 static void flatview_destroy(FlatView *view)
278 {
279 int i;
280
281 trace_flatview_destroy(view, view->root);
282 if (view->dispatch) {
283 address_space_dispatch_free(view->dispatch);
284 }
285 for (i = 0; i < view->nr; i++) {
286 memory_region_unref(view->ranges[i].mr);
287 }
288 g_free(view->ranges);
289 memory_region_unref(view->root);
290 g_free(view);
291 }
292
293 static bool flatview_ref(FlatView *view)
294 {
295 return atomic_fetch_inc_nonzero(&view->ref) > 0;
296 }
297
298 void flatview_unref(FlatView *view)
299 {
300 if (atomic_fetch_dec(&view->ref) == 1) {
301 trace_flatview_destroy_rcu(view, view->root);
302 assert(view->root);
303 call_rcu(view, flatview_destroy, rcu);
304 }
305 }
306
307 static bool can_merge(FlatRange *r1, FlatRange *r2)
308 {
309 return int128_eq(addrrange_end(r1->addr), r2->addr.start)
310 && r1->mr == r2->mr
311 && int128_eq(int128_add(int128_make64(r1->offset_in_region),
312 r1->addr.size),
313 int128_make64(r2->offset_in_region))
314 && r1->dirty_log_mask == r2->dirty_log_mask
315 && r1->romd_mode == r2->romd_mode
316 && r1->readonly == r2->readonly;
317 }
318
319 /* Attempt to simplify a view by merging adjacent ranges */
320 static void flatview_simplify(FlatView *view)
321 {
322 unsigned i, j;
323
324 i = 0;
325 while (i < view->nr) {
326 j = i + 1;
327 while (j < view->nr
328 && can_merge(&view->ranges[j-1], &view->ranges[j])) {
329 int128_addto(&view->ranges[i].addr.size, view->ranges[j].addr.size);
330 ++j;
331 }
332 ++i;
333 memmove(&view->ranges[i], &view->ranges[j],
334 (view->nr - j) * sizeof(view->ranges[j]));
335 view->nr -= j - i;
336 }
337 }
338
339 static bool memory_region_big_endian(MemoryRegion *mr)
340 {
341 #ifdef TARGET_WORDS_BIGENDIAN
342 return mr->ops->endianness != DEVICE_LITTLE_ENDIAN;
343 #else
344 return mr->ops->endianness == DEVICE_BIG_ENDIAN;
345 #endif
346 }
347
348 static bool memory_region_wrong_endianness(MemoryRegion *mr)
349 {
350 #ifdef TARGET_WORDS_BIGENDIAN
351 return mr->ops->endianness == DEVICE_LITTLE_ENDIAN;
352 #else
353 return mr->ops->endianness == DEVICE_BIG_ENDIAN;
354 #endif
355 }
356
357 static void adjust_endianness(MemoryRegion *mr, uint64_t *data, unsigned size)
358 {
359 if (memory_region_wrong_endianness(mr)) {
360 switch (size) {
361 case 1:
362 break;
363 case 2:
364 *data = bswap16(*data);
365 break;
366 case 4:
367 *data = bswap32(*data);
368 break;
369 case 8:
370 *data = bswap64(*data);
371 break;
372 default:
373 abort();
374 }
375 }
376 }
377
378 static hwaddr memory_region_to_absolute_addr(MemoryRegion *mr, hwaddr offset)
379 {
380 MemoryRegion *root;
381 hwaddr abs_addr = offset;
382
383 abs_addr += mr->addr;
384 for (root = mr; root->container; ) {
385 root = root->container;
386 abs_addr += root->addr;
387 }
388
389 return abs_addr;
390 }
391
392 static int get_cpu_index(void)
393 {
394 if (current_cpu) {
395 return current_cpu->cpu_index;
396 }
397 return -1;
398 }
399
400 static MemTxResult memory_region_oldmmio_read_accessor(MemoryRegion *mr,
401 hwaddr addr,
402 uint64_t *value,
403 unsigned size,
404 unsigned shift,
405 uint64_t mask,
406 MemTxAttrs attrs)
407 {
408 uint64_t tmp;
409
410 tmp = mr->ops->old_mmio.read[ctz32(size)](mr->opaque, addr);
411 if (mr->subpage) {
412 trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
413 } else if (mr == &io_mem_notdirty) {
414 /* Accesses to code which has previously been translated into a TB show
415 * up in the MMIO path, as accesses to the io_mem_notdirty
416 * MemoryRegion. */
417 trace_memory_region_tb_read(get_cpu_index(), addr, tmp, size);
418 } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
419 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
420 trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
421 }
422 *value |= (tmp & mask) << shift;
423 return MEMTX_OK;
424 }
425
426 static MemTxResult memory_region_read_accessor(MemoryRegion *mr,
427 hwaddr addr,
428 uint64_t *value,
429 unsigned size,
430 unsigned shift,
431 uint64_t mask,
432 MemTxAttrs attrs)
433 {
434 uint64_t tmp;
435
436 tmp = mr->ops->read(mr->opaque, addr, size);
437 if (mr->subpage) {
438 trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
439 } else if (mr == &io_mem_notdirty) {
440 /* Accesses to code which has previously been translated into a TB show
441 * up in the MMIO path, as accesses to the io_mem_notdirty
442 * MemoryRegion. */
443 trace_memory_region_tb_read(get_cpu_index(), addr, tmp, size);
444 } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
445 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
446 trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
447 }
448 *value |= (tmp & mask) << shift;
449 return MEMTX_OK;
450 }
451
452 static MemTxResult memory_region_read_with_attrs_accessor(MemoryRegion *mr,
453 hwaddr addr,
454 uint64_t *value,
455 unsigned size,
456 unsigned shift,
457 uint64_t mask,
458 MemTxAttrs attrs)
459 {
460 uint64_t tmp = 0;
461 MemTxResult r;
462
463 r = mr->ops->read_with_attrs(mr->opaque, addr, &tmp, size, attrs);
464 if (mr->subpage) {
465 trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
466 } else if (mr == &io_mem_notdirty) {
467 /* Accesses to code which has previously been translated into a TB show
468 * up in the MMIO path, as accesses to the io_mem_notdirty
469 * MemoryRegion. */
470 trace_memory_region_tb_read(get_cpu_index(), addr, tmp, size);
471 } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
472 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
473 trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
474 }
475 *value |= (tmp & mask) << shift;
476 return r;
477 }
478
479 static MemTxResult memory_region_oldmmio_write_accessor(MemoryRegion *mr,
480 hwaddr addr,
481 uint64_t *value,
482 unsigned size,
483 unsigned shift,
484 uint64_t mask,
485 MemTxAttrs attrs)
486 {
487 uint64_t tmp;
488
489 tmp = (*value >> shift) & mask;
490 if (mr->subpage) {
491 trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
492 } else if (mr == &io_mem_notdirty) {
493 /* Accesses to code which has previously been translated into a TB show
494 * up in the MMIO path, as accesses to the io_mem_notdirty
495 * MemoryRegion. */
496 trace_memory_region_tb_write(get_cpu_index(), addr, tmp, size);
497 } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
498 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
499 trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
500 }
501 mr->ops->old_mmio.write[ctz32(size)](mr->opaque, addr, tmp);
502 return MEMTX_OK;
503 }
504
505 static MemTxResult memory_region_write_accessor(MemoryRegion *mr,
506 hwaddr addr,
507 uint64_t *value,
508 unsigned size,
509 unsigned shift,
510 uint64_t mask,
511 MemTxAttrs attrs)
512 {
513 uint64_t tmp;
514
515 tmp = (*value >> shift) & mask;
516 if (mr->subpage) {
517 trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
518 } else if (mr == &io_mem_notdirty) {
519 /* Accesses to code which has previously been translated into a TB show
520 * up in the MMIO path, as accesses to the io_mem_notdirty
521 * MemoryRegion. */
522 trace_memory_region_tb_write(get_cpu_index(), addr, tmp, size);
523 } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
524 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
525 trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
526 }
527 mr->ops->write(mr->opaque, addr, tmp, size);
528 return MEMTX_OK;
529 }
530
531 static MemTxResult memory_region_write_with_attrs_accessor(MemoryRegion *mr,
532 hwaddr addr,
533 uint64_t *value,
534 unsigned size,
535 unsigned shift,
536 uint64_t mask,
537 MemTxAttrs attrs)
538 {
539 uint64_t tmp;
540
541 tmp = (*value >> shift) & mask;
542 if (mr->subpage) {
543 trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
544 } else if (mr == &io_mem_notdirty) {
545 /* Accesses to code which has previously been translated into a TB show
546 * up in the MMIO path, as accesses to the io_mem_notdirty
547 * MemoryRegion. */
548 trace_memory_region_tb_write(get_cpu_index(), addr, tmp, size);
549 } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
550 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
551 trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
552 }
553 return mr->ops->write_with_attrs(mr->opaque, addr, tmp, size, attrs);
554 }
555
556 static MemTxResult access_with_adjusted_size(hwaddr addr,
557 uint64_t *value,
558 unsigned size,
559 unsigned access_size_min,
560 unsigned access_size_max,
561 MemTxResult (*access_fn)
562 (MemoryRegion *mr,
563 hwaddr addr,
564 uint64_t *value,
565 unsigned size,
566 unsigned shift,
567 uint64_t mask,
568 MemTxAttrs attrs),
569 MemoryRegion *mr,
570 MemTxAttrs attrs)
571 {
572 uint64_t access_mask;
573 unsigned access_size;
574 unsigned i;
575 MemTxResult r = MEMTX_OK;
576
577 if (!access_size_min) {
578 access_size_min = 1;
579 }
580 if (!access_size_max) {
581 access_size_max = 4;
582 }
583
584 /* FIXME: support unaligned access? */
585 access_size = MAX(MIN(size, access_size_max), access_size_min);
586 access_mask = -1ULL >> (64 - access_size * 8);
587 if (memory_region_big_endian(mr)) {
588 for (i = 0; i < size; i += access_size) {
589 r |= access_fn(mr, addr + i, value, access_size,
590 (size - access_size - i) * 8, access_mask, attrs);
591 }
592 } else {
593 for (i = 0; i < size; i += access_size) {
594 r |= access_fn(mr, addr + i, value, access_size, i * 8,
595 access_mask, attrs);
596 }
597 }
598 return r;
599 }
600
601 static AddressSpace *memory_region_to_address_space(MemoryRegion *mr)
602 {
603 AddressSpace *as;
604
605 while (mr->container) {
606 mr = mr->container;
607 }
608 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
609 if (mr == as->root) {
610 return as;
611 }
612 }
613 return NULL;
614 }
615
616 /* Render a memory region into the global view. Ranges in @view obscure
617 * ranges in @mr.
618 */
619 static void render_memory_region(FlatView *view,
620 MemoryRegion *mr,
621 Int128 base,
622 AddrRange clip,
623 bool readonly)
624 {
625 MemoryRegion *subregion;
626 unsigned i;
627 hwaddr offset_in_region;
628 Int128 remain;
629 Int128 now;
630 FlatRange fr;
631 AddrRange tmp;
632
633 if (!mr->enabled) {
634 return;
635 }
636
637 int128_addto(&base, int128_make64(mr->addr));
638 readonly |= mr->readonly;
639
640 tmp = addrrange_make(base, mr->size);
641
642 if (!addrrange_intersects(tmp, clip)) {
643 return;
644 }
645
646 clip = addrrange_intersection(tmp, clip);
647
648 if (mr->alias) {
649 int128_subfrom(&base, int128_make64(mr->alias->addr));
650 int128_subfrom(&base, int128_make64(mr->alias_offset));
651 render_memory_region(view, mr->alias, base, clip, readonly);
652 return;
653 }
654
655 /* Render subregions in priority order. */
656 QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
657 render_memory_region(view, subregion, base, clip, readonly);
658 }
659
660 if (!mr->terminates) {
661 return;
662 }
663
664 offset_in_region = int128_get64(int128_sub(clip.start, base));
665 base = clip.start;
666 remain = clip.size;
667
668 fr.mr = mr;
669 fr.dirty_log_mask = memory_region_get_dirty_log_mask(mr);
670 fr.romd_mode = mr->romd_mode;
671 fr.readonly = readonly;
672
673 /* Render the region itself into any gaps left by the current view. */
674 for (i = 0; i < view->nr && int128_nz(remain); ++i) {
675 if (int128_ge(base, addrrange_end(view->ranges[i].addr))) {
676 continue;
677 }
678 if (int128_lt(base, view->ranges[i].addr.start)) {
679 now = int128_min(remain,
680 int128_sub(view->ranges[i].addr.start, base));
681 fr.offset_in_region = offset_in_region;
682 fr.addr = addrrange_make(base, now);
683 flatview_insert(view, i, &fr);
684 ++i;
685 int128_addto(&base, now);
686 offset_in_region += int128_get64(now);
687 int128_subfrom(&remain, now);
688 }
689 now = int128_sub(int128_min(int128_add(base, remain),
690 addrrange_end(view->ranges[i].addr)),
691 base);
692 int128_addto(&base, now);
693 offset_in_region += int128_get64(now);
694 int128_subfrom(&remain, now);
695 }
696 if (int128_nz(remain)) {
697 fr.offset_in_region = offset_in_region;
698 fr.addr = addrrange_make(base, remain);
699 flatview_insert(view, i, &fr);
700 }
701 }
702
703 static MemoryRegion *memory_region_get_flatview_root(MemoryRegion *mr)
704 {
705 while (mr->enabled) {
706 if (mr->alias) {
707 if (!mr->alias_offset && int128_ge(mr->size, mr->alias->size)) {
708 /* The alias is included in its entirety. Use it as
709 * the "real" root, so that we can share more FlatViews.
710 */
711 mr = mr->alias;
712 continue;
713 }
714 } else if (!mr->terminates) {
715 unsigned int found = 0;
716 MemoryRegion *child, *next = NULL;
717 QTAILQ_FOREACH(child, &mr->subregions, subregions_link) {
718 if (child->enabled) {
719 if (++found > 1) {
720 next = NULL;
721 break;
722 }
723 if (!child->addr && int128_ge(mr->size, child->size)) {
724 /* A child is included in its entirety. If it's the only
725 * enabled one, use it in the hope of finding an alias down the
726 * way. This will also let us share FlatViews.
727 */
728 next = child;
729 }
730 }
731 }
732 if (found == 0) {
733 return NULL;
734 }
735 if (next) {
736 mr = next;
737 continue;
738 }
739 }
740
741 return mr;
742 }
743
744 return NULL;
745 }
746
747 /* Render a memory topology into a list of disjoint absolute ranges. */
748 static FlatView *generate_memory_topology(MemoryRegion *mr)
749 {
750 int i;
751 FlatView *view;
752
753 view = flatview_new(mr);
754
755 if (mr) {
756 render_memory_region(view, mr, int128_zero(),
757 addrrange_make(int128_zero(), int128_2_64()), false);
758 }
759 flatview_simplify(view);
760
761 view->dispatch = address_space_dispatch_new(view);
762 for (i = 0; i < view->nr; i++) {
763 MemoryRegionSection mrs =
764 section_from_flat_range(&view->ranges[i], view);
765 flatview_add_to_dispatch(view, &mrs);
766 }
767 address_space_dispatch_compact(view->dispatch);
768 g_hash_table_replace(flat_views, mr, view);
769
770 return view;
771 }
772
773 static void address_space_add_del_ioeventfds(AddressSpace *as,
774 MemoryRegionIoeventfd *fds_new,
775 unsigned fds_new_nb,
776 MemoryRegionIoeventfd *fds_old,
777 unsigned fds_old_nb)
778 {
779 unsigned iold, inew;
780 MemoryRegionIoeventfd *fd;
781 MemoryRegionSection section;
782
783 /* Generate a symmetric difference of the old and new fd sets, adding
784 * and deleting as necessary.
785 */
786
787 iold = inew = 0;
788 while (iold < fds_old_nb || inew < fds_new_nb) {
789 if (iold < fds_old_nb
790 && (inew == fds_new_nb
791 || memory_region_ioeventfd_before(&fds_old[iold],
792 &fds_new[inew]))) {
793 fd = &fds_old[iold];
794 section = (MemoryRegionSection) {
795 .fv = address_space_to_flatview(as),
796 .offset_within_address_space = int128_get64(fd->addr.start),
797 .size = fd->addr.size,
798 };
799 MEMORY_LISTENER_CALL(as, eventfd_del, Forward, &section,
800 fd->match_data, fd->data, fd->e);
801 ++iold;
802 } else if (inew < fds_new_nb
803 && (iold == fds_old_nb
804 || memory_region_ioeventfd_before(&fds_new[inew],
805 &fds_old[iold]))) {
806 fd = &fds_new[inew];
807 section = (MemoryRegionSection) {
808 .fv = address_space_to_flatview(as),
809 .offset_within_address_space = int128_get64(fd->addr.start),
810 .size = fd->addr.size,
811 };
812 MEMORY_LISTENER_CALL(as, eventfd_add, Reverse, &section,
813 fd->match_data, fd->data, fd->e);
814 ++inew;
815 } else {
816 ++iold;
817 ++inew;
818 }
819 }
820 }
821
822 FlatView *address_space_get_flatview(AddressSpace *as)
823 {
824 FlatView *view;
825
826 rcu_read_lock();
827 do {
828 view = address_space_to_flatview(as);
829 /* If somebody has replaced as->current_map concurrently,
830 * flatview_ref returns false.
831 */
832 } while (!flatview_ref(view));
833 rcu_read_unlock();
834 return view;
835 }
836
837 static void address_space_update_ioeventfds(AddressSpace *as)
838 {
839 FlatView *view;
840 FlatRange *fr;
841 unsigned ioeventfd_nb = 0;
842 MemoryRegionIoeventfd *ioeventfds = NULL;
843 AddrRange tmp;
844 unsigned i;
845
846 view = address_space_get_flatview(as);
847 FOR_EACH_FLAT_RANGE(fr, view) {
848 for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
849 tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
850 int128_sub(fr->addr.start,
851 int128_make64(fr->offset_in_region)));
852 if (addrrange_intersects(fr->addr, tmp)) {
853 ++ioeventfd_nb;
854 ioeventfds = g_realloc(ioeventfds,
855 ioeventfd_nb * sizeof(*ioeventfds));
856 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
857 ioeventfds[ioeventfd_nb-1].addr = tmp;
858 }
859 }
860 }
861
862 address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
863 as->ioeventfds, as->ioeventfd_nb);
864
865 g_free(as->ioeventfds);
866 as->ioeventfds = ioeventfds;
867 as->ioeventfd_nb = ioeventfd_nb;
868 flatview_unref(view);
869 }
870
871 static void address_space_update_topology_pass(AddressSpace *as,
872 const FlatView *old_view,
873 const FlatView *new_view,
874 bool adding)
875 {
876 unsigned iold, inew;
877 FlatRange *frold, *frnew;
878
879 /* Generate a symmetric difference of the old and new memory maps.
880 * Kill ranges in the old map, and instantiate ranges in the new map.
881 */
882 iold = inew = 0;
883 while (iold < old_view->nr || inew < new_view->nr) {
884 if (iold < old_view->nr) {
885 frold = &old_view->ranges[iold];
886 } else {
887 frold = NULL;
888 }
889 if (inew < new_view->nr) {
890 frnew = &new_view->ranges[inew];
891 } else {
892 frnew = NULL;
893 }
894
895 if (frold
896 && (!frnew
897 || int128_lt(frold->addr.start, frnew->addr.start)
898 || (int128_eq(frold->addr.start, frnew->addr.start)
899 && !flatrange_equal(frold, frnew)))) {
900 /* In old but not in new, or in both but attributes changed. */
901
902 if (!adding) {
903 MEMORY_LISTENER_UPDATE_REGION(frold, as, Reverse, region_del);
904 }
905
906 ++iold;
907 } else if (frold && frnew && flatrange_equal(frold, frnew)) {
908 /* In both and unchanged (except logging may have changed) */
909
910 if (adding) {
911 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_nop);
912 if (frnew->dirty_log_mask & ~frold->dirty_log_mask) {
913 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, log_start,
914 frold->dirty_log_mask,
915 frnew->dirty_log_mask);
916 }
917 if (frold->dirty_log_mask & ~frnew->dirty_log_mask) {
918 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Reverse, log_stop,
919 frold->dirty_log_mask,
920 frnew->dirty_log_mask);
921 }
922 }
923
924 ++iold;
925 ++inew;
926 } else {
927 /* In new */
928
929 if (adding) {
930 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_add);
931 }
932
933 ++inew;
934 }
935 }
936 }
937
938 static void flatviews_init(void)
939 {
940 static FlatView *empty_view;
941
942 if (flat_views) {
943 return;
944 }
945
946 flat_views = g_hash_table_new_full(g_direct_hash, g_direct_equal, NULL,
947 (GDestroyNotify) flatview_unref);
948 if (!empty_view) {
949 empty_view = generate_memory_topology(NULL);
950 /* We keep it alive forever in the global variable. */
951 flatview_ref(empty_view);
952 } else {
953 g_hash_table_replace(flat_views, NULL, empty_view);
954 flatview_ref(empty_view);
955 }
956 }
957
958 static void flatviews_reset(void)
959 {
960 AddressSpace *as;
961
962 if (flat_views) {
963 g_hash_table_unref(flat_views);
964 flat_views = NULL;
965 }
966 flatviews_init();
967
968 /* Render unique FVs */
969 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
970 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
971
972 if (g_hash_table_lookup(flat_views, physmr)) {
973 continue;
974 }
975
976 generate_memory_topology(physmr);
977 }
978 }
979
980 static void address_space_set_flatview(AddressSpace *as)
981 {
982 FlatView *old_view = address_space_to_flatview(as);
983 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
984 FlatView *new_view = g_hash_table_lookup(flat_views, physmr);
985
986 assert(new_view);
987
988 if (old_view == new_view) {
989 return;
990 }
991
992 if (old_view) {
993 flatview_ref(old_view);
994 }
995
996 flatview_ref(new_view);
997
998 if (!QTAILQ_EMPTY(&as->listeners)) {
999 FlatView tmpview = { .nr = 0 }, *old_view2 = old_view;
1000
1001 if (!old_view2) {
1002 old_view2 = &tmpview;
1003 }
1004 address_space_update_topology_pass(as, old_view2, new_view, false);
1005 address_space_update_topology_pass(as, old_view2, new_view, true);
1006 }
1007
1008 /* Writes are protected by the BQL. */
1009 atomic_rcu_set(&as->current_map, new_view);
1010 if (old_view) {
1011 flatview_unref(old_view);
1012 }
1013
1014 /* Note that all the old MemoryRegions are still alive up to this
1015 * point. This relieves most MemoryListeners from the need to
1016 * ref/unref the MemoryRegions they get---unless they use them
1017 * outside the iothread mutex, in which case precise reference
1018 * counting is necessary.
1019 */
1020 if (old_view) {
1021 flatview_unref(old_view);
1022 }
1023 }
1024
1025 static void address_space_update_topology(AddressSpace *as)
1026 {
1027 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1028
1029 flatviews_init();
1030 if (!g_hash_table_lookup(flat_views, physmr)) {
1031 generate_memory_topology(physmr);
1032 }
1033 address_space_set_flatview(as);
1034 }
1035
1036 void memory_region_transaction_begin(void)
1037 {
1038 qemu_flush_coalesced_mmio_buffer();
1039 ++memory_region_transaction_depth;
1040 }
1041
1042 void memory_region_transaction_commit(void)
1043 {
1044 AddressSpace *as;
1045
1046 assert(memory_region_transaction_depth);
1047 assert(qemu_mutex_iothread_locked());
1048
1049 --memory_region_transaction_depth;
1050 if (!memory_region_transaction_depth) {
1051 if (memory_region_update_pending) {
1052 flatviews_reset();
1053
1054 MEMORY_LISTENER_CALL_GLOBAL(begin, Forward);
1055
1056 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1057 address_space_set_flatview(as);
1058 address_space_update_ioeventfds(as);
1059 }
1060 memory_region_update_pending = false;
1061 ioeventfd_update_pending = false;
1062 MEMORY_LISTENER_CALL_GLOBAL(commit, Forward);
1063 } else if (ioeventfd_update_pending) {
1064 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1065 address_space_update_ioeventfds(as);
1066 }
1067 ioeventfd_update_pending = false;
1068 }
1069 }
1070 }
1071
1072 static void memory_region_destructor_none(MemoryRegion *mr)
1073 {
1074 }
1075
1076 static void memory_region_destructor_ram(MemoryRegion *mr)
1077 {
1078 qemu_ram_free(mr->ram_block);
1079 }
1080
1081 static bool memory_region_need_escape(char c)
1082 {
1083 return c == '/' || c == '[' || c == '\\' || c == ']';
1084 }
1085
1086 static char *memory_region_escape_name(const char *name)
1087 {
1088 const char *p;
1089 char *escaped, *q;
1090 uint8_t c;
1091 size_t bytes = 0;
1092
1093 for (p = name; *p; p++) {
1094 bytes += memory_region_need_escape(*p) ? 4 : 1;
1095 }
1096 if (bytes == p - name) {
1097 return g_memdup(name, bytes + 1);
1098 }
1099
1100 escaped = g_malloc(bytes + 1);
1101 for (p = name, q = escaped; *p; p++) {
1102 c = *p;
1103 if (unlikely(memory_region_need_escape(c))) {
1104 *q++ = '\\';
1105 *q++ = 'x';
1106 *q++ = "0123456789abcdef"[c >> 4];
1107 c = "0123456789abcdef"[c & 15];
1108 }
1109 *q++ = c;
1110 }
1111 *q = 0;
1112 return escaped;
1113 }
1114
1115 static void memory_region_do_init(MemoryRegion *mr,
1116 Object *owner,
1117 const char *name,
1118 uint64_t size)
1119 {
1120 mr->size = int128_make64(size);
1121 if (size == UINT64_MAX) {
1122 mr->size = int128_2_64();
1123 }
1124 mr->name = g_strdup(name);
1125 mr->owner = owner;
1126 mr->ram_block = NULL;
1127
1128 if (name) {
1129 char *escaped_name = memory_region_escape_name(name);
1130 char *name_array = g_strdup_printf("%s[*]", escaped_name);
1131
1132 if (!owner) {
1133 owner = container_get(qdev_get_machine(), "/unattached");
1134 }
1135
1136 object_property_add_child(owner, name_array, OBJECT(mr), &error_abort);
1137 object_unref(OBJECT(mr));
1138 g_free(name_array);
1139 g_free(escaped_name);
1140 }
1141 }
1142
1143 void memory_region_init(MemoryRegion *mr,
1144 Object *owner,
1145 const char *name,
1146 uint64_t size)
1147 {
1148 object_initialize(mr, sizeof(*mr), TYPE_MEMORY_REGION);
1149 memory_region_do_init(mr, owner, name, size);
1150 }
1151
1152 static void memory_region_get_addr(Object *obj, Visitor *v, const char *name,
1153 void *opaque, Error **errp)
1154 {
1155 MemoryRegion *mr = MEMORY_REGION(obj);
1156 uint64_t value = mr->addr;
1157
1158 visit_type_uint64(v, name, &value, errp);
1159 }
1160
1161 static void memory_region_get_container(Object *obj, Visitor *v,
1162 const char *name, void *opaque,
1163 Error **errp)
1164 {
1165 MemoryRegion *mr = MEMORY_REGION(obj);
1166 gchar *path = (gchar *)"";
1167
1168 if (mr->container) {
1169 path = object_get_canonical_path(OBJECT(mr->container));
1170 }
1171 visit_type_str(v, name, &path, errp);
1172 if (mr->container) {
1173 g_free(path);
1174 }
1175 }
1176
1177 static Object *memory_region_resolve_container(Object *obj, void *opaque,
1178 const char *part)
1179 {
1180 MemoryRegion *mr = MEMORY_REGION(obj);
1181
1182 return OBJECT(mr->container);
1183 }
1184
1185 static void memory_region_get_priority(Object *obj, Visitor *v,
1186 const char *name, void *opaque,
1187 Error **errp)
1188 {
1189 MemoryRegion *mr = MEMORY_REGION(obj);
1190 int32_t value = mr->priority;
1191
1192 visit_type_int32(v, name, &value, errp);
1193 }
1194
1195 static void memory_region_get_size(Object *obj, Visitor *v, const char *name,
1196 void *opaque, Error **errp)
1197 {
1198 MemoryRegion *mr = MEMORY_REGION(obj);
1199 uint64_t value = memory_region_size(mr);
1200
1201 visit_type_uint64(v, name, &value, errp);
1202 }
1203
1204 static void memory_region_initfn(Object *obj)
1205 {
1206 MemoryRegion *mr = MEMORY_REGION(obj);
1207 ObjectProperty *op;
1208
1209 mr->ops = &unassigned_mem_ops;
1210 mr->enabled = true;
1211 mr->romd_mode = true;
1212 mr->global_locking = true;
1213 mr->destructor = memory_region_destructor_none;
1214 QTAILQ_INIT(&mr->subregions);
1215 QTAILQ_INIT(&mr->coalesced);
1216
1217 op = object_property_add(OBJECT(mr), "container",
1218 "link<" TYPE_MEMORY_REGION ">",
1219 memory_region_get_container,
1220 NULL, /* memory_region_set_container */
1221 NULL, NULL, &error_abort);
1222 op->resolve = memory_region_resolve_container;
1223
1224 object_property_add(OBJECT(mr), "addr", "uint64",
1225 memory_region_get_addr,
1226 NULL, /* memory_region_set_addr */
1227 NULL, NULL, &error_abort);
1228 object_property_add(OBJECT(mr), "priority", "uint32",
1229 memory_region_get_priority,
1230 NULL, /* memory_region_set_priority */
1231 NULL, NULL, &error_abort);
1232 object_property_add(OBJECT(mr), "size", "uint64",
1233 memory_region_get_size,
1234 NULL, /* memory_region_set_size, */
1235 NULL, NULL, &error_abort);
1236 }
1237
1238 static void iommu_memory_region_initfn(Object *obj)
1239 {
1240 MemoryRegion *mr = MEMORY_REGION(obj);
1241
1242 mr->is_iommu = true;
1243 }
1244
1245 static uint64_t unassigned_mem_read(void *opaque, hwaddr addr,
1246 unsigned size)
1247 {
1248 #ifdef DEBUG_UNASSIGNED
1249 printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
1250 #endif
1251 if (current_cpu != NULL) {
1252 cpu_unassigned_access(current_cpu, addr, false, false, 0, size);
1253 }
1254 return 0;
1255 }
1256
1257 static void unassigned_mem_write(void *opaque, hwaddr addr,
1258 uint64_t val, unsigned size)
1259 {
1260 #ifdef DEBUG_UNASSIGNED
1261 printf("Unassigned mem write " TARGET_FMT_plx " = 0x%"PRIx64"\n", addr, val);
1262 #endif
1263 if (current_cpu != NULL) {
1264 cpu_unassigned_access(current_cpu, addr, true, false, 0, size);
1265 }
1266 }
1267
1268 static bool unassigned_mem_accepts(void *opaque, hwaddr addr,
1269 unsigned size, bool is_write,
1270 MemTxAttrs attrs)
1271 {
1272 return false;
1273 }
1274
1275 const MemoryRegionOps unassigned_mem_ops = {
1276 .valid.accepts = unassigned_mem_accepts,
1277 .endianness = DEVICE_NATIVE_ENDIAN,
1278 };
1279
1280 static uint64_t memory_region_ram_device_read(void *opaque,
1281 hwaddr addr, unsigned size)
1282 {
1283 MemoryRegion *mr = opaque;
1284 uint64_t data = (uint64_t)~0;
1285
1286 switch (size) {
1287 case 1:
1288 data = *(uint8_t *)(mr->ram_block->host + addr);
1289 break;
1290 case 2:
1291 data = *(uint16_t *)(mr->ram_block->host + addr);
1292 break;
1293 case 4:
1294 data = *(uint32_t *)(mr->ram_block->host + addr);
1295 break;
1296 case 8:
1297 data = *(uint64_t *)(mr->ram_block->host + addr);
1298 break;
1299 }
1300
1301 trace_memory_region_ram_device_read(get_cpu_index(), mr, addr, data, size);
1302
1303 return data;
1304 }
1305
1306 static void memory_region_ram_device_write(void *opaque, hwaddr addr,
1307 uint64_t data, unsigned size)
1308 {
1309 MemoryRegion *mr = opaque;
1310
1311 trace_memory_region_ram_device_write(get_cpu_index(), mr, addr, data, size);
1312
1313 switch (size) {
1314 case 1:
1315 *(uint8_t *)(mr->ram_block->host + addr) = (uint8_t)data;
1316 break;
1317 case 2:
1318 *(uint16_t *)(mr->ram_block->host + addr) = (uint16_t)data;
1319 break;
1320 case 4:
1321 *(uint32_t *)(mr->ram_block->host + addr) = (uint32_t)data;
1322 break;
1323 case 8:
1324 *(uint64_t *)(mr->ram_block->host + addr) = data;
1325 break;
1326 }
1327 }
1328
1329 static const MemoryRegionOps ram_device_mem_ops = {
1330 .read = memory_region_ram_device_read,
1331 .write = memory_region_ram_device_write,
1332 .endianness = DEVICE_HOST_ENDIAN,
1333 .valid = {
1334 .min_access_size = 1,
1335 .max_access_size = 8,
1336 .unaligned = true,
1337 },
1338 .impl = {
1339 .min_access_size = 1,
1340 .max_access_size = 8,
1341 .unaligned = true,
1342 },
1343 };
1344
1345 bool memory_region_access_valid(MemoryRegion *mr,
1346 hwaddr addr,
1347 unsigned size,
1348 bool is_write,
1349 MemTxAttrs attrs)
1350 {
1351 int access_size_min, access_size_max;
1352 int access_size, i;
1353
1354 if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1355 return false;
1356 }
1357
1358 if (!mr->ops->valid.accepts) {
1359 return true;
1360 }
1361
1362 access_size_min = mr->ops->valid.min_access_size;
1363 if (!mr->ops->valid.min_access_size) {
1364 access_size_min = 1;
1365 }
1366
1367 access_size_max = mr->ops->valid.max_access_size;
1368 if (!mr->ops->valid.max_access_size) {
1369 access_size_max = 4;
1370 }
1371
1372 access_size = MAX(MIN(size, access_size_max), access_size_min);
1373 for (i = 0; i < size; i += access_size) {
1374 if (!mr->ops->valid.accepts(mr->opaque, addr + i, access_size,
1375 is_write, attrs)) {
1376 return false;
1377 }
1378 }
1379
1380 return true;
1381 }
1382
1383 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1384 hwaddr addr,
1385 uint64_t *pval,
1386 unsigned size,
1387 MemTxAttrs attrs)
1388 {
1389 *pval = 0;
1390
1391 if (mr->ops->read) {
1392 return access_with_adjusted_size(addr, pval, size,
1393 mr->ops->impl.min_access_size,
1394 mr->ops->impl.max_access_size,
1395 memory_region_read_accessor,
1396 mr, attrs);
1397 } else if (mr->ops->read_with_attrs) {
1398 return access_with_adjusted_size(addr, pval, size,
1399 mr->ops->impl.min_access_size,
1400 mr->ops->impl.max_access_size,
1401 memory_region_read_with_attrs_accessor,
1402 mr, attrs);
1403 } else {
1404 return access_with_adjusted_size(addr, pval, size, 1, 4,
1405 memory_region_oldmmio_read_accessor,
1406 mr, attrs);
1407 }
1408 }
1409
1410 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1411 hwaddr addr,
1412 uint64_t *pval,
1413 unsigned size,
1414 MemTxAttrs attrs)
1415 {
1416 MemTxResult r;
1417
1418 if (!memory_region_access_valid(mr, addr, size, false, attrs)) {
1419 *pval = unassigned_mem_read(mr, addr, size);
1420 return MEMTX_DECODE_ERROR;
1421 }
1422
1423 r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1424 adjust_endianness(mr, pval, size);
1425 return r;
1426 }
1427
1428 /* Return true if an eventfd was signalled */
1429 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1430 hwaddr addr,
1431 uint64_t data,
1432 unsigned size,
1433 MemTxAttrs attrs)
1434 {
1435 MemoryRegionIoeventfd ioeventfd = {
1436 .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1437 .data = data,
1438 };
1439 unsigned i;
1440
1441 for (i = 0; i < mr->ioeventfd_nb; i++) {
1442 ioeventfd.match_data = mr->ioeventfds[i].match_data;
1443 ioeventfd.e = mr->ioeventfds[i].e;
1444
1445 if (memory_region_ioeventfd_equal(&ioeventfd, &mr->ioeventfds[i])) {
1446 event_notifier_set(ioeventfd.e);
1447 return true;
1448 }
1449 }
1450
1451 return false;
1452 }
1453
1454 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1455 hwaddr addr,
1456 uint64_t data,
1457 unsigned size,
1458 MemTxAttrs attrs)
1459 {
1460 if (!memory_region_access_valid(mr, addr, size, true, attrs)) {
1461 unassigned_mem_write(mr, addr, data, size);
1462 return MEMTX_DECODE_ERROR;
1463 }
1464
1465 adjust_endianness(mr, &data, size);
1466
1467 if ((!kvm_eventfds_enabled()) &&
1468 memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1469 return MEMTX_OK;
1470 }
1471
1472 if (mr->ops->write) {
1473 return access_with_adjusted_size(addr, &data, size,
1474 mr->ops->impl.min_access_size,
1475 mr->ops->impl.max_access_size,
1476 memory_region_write_accessor, mr,
1477 attrs);
1478 } else if (mr->ops->write_with_attrs) {
1479 return
1480 access_with_adjusted_size(addr, &data, size,
1481 mr->ops->impl.min_access_size,
1482 mr->ops->impl.max_access_size,
1483 memory_region_write_with_attrs_accessor,
1484 mr, attrs);
1485 } else {
1486 return access_with_adjusted_size(addr, &data, size, 1, 4,
1487 memory_region_oldmmio_write_accessor,
1488 mr, attrs);
1489 }
1490 }
1491
1492 void memory_region_init_io(MemoryRegion *mr,
1493 Object *owner,
1494 const MemoryRegionOps *ops,
1495 void *opaque,
1496 const char *name,
1497 uint64_t size)
1498 {
1499 memory_region_init(mr, owner, name, size);
1500 mr->ops = ops ? ops : &unassigned_mem_ops;
1501 mr->opaque = opaque;
1502 mr->terminates = true;
1503 }
1504
1505 void memory_region_init_ram_nomigrate(MemoryRegion *mr,
1506 Object *owner,
1507 const char *name,
1508 uint64_t size,
1509 Error **errp)
1510 {
1511 memory_region_init_ram_shared_nomigrate(mr, owner, name, size, false, errp);
1512 }
1513
1514 void memory_region_init_ram_shared_nomigrate(MemoryRegion *mr,
1515 Object *owner,
1516 const char *name,
1517 uint64_t size,
1518 bool share,
1519 Error **errp)
1520 {
1521 memory_region_init(mr, owner, name, size);
1522 mr->ram = true;
1523 mr->terminates = true;
1524 mr->destructor = memory_region_destructor_ram;
1525 mr->ram_block = qemu_ram_alloc(size, share, mr, errp);
1526 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1527 }
1528
1529 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1530 Object *owner,
1531 const char *name,
1532 uint64_t size,
1533 uint64_t max_size,
1534 void (*resized)(const char*,
1535 uint64_t length,
1536 void *host),
1537 Error **errp)
1538 {
1539 memory_region_init(mr, owner, name, size);
1540 mr->ram = true;
1541 mr->terminates = true;
1542 mr->destructor = memory_region_destructor_ram;
1543 mr->ram_block = qemu_ram_alloc_resizeable(size, max_size, resized,
1544 mr, errp);
1545 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1546 }
1547
1548 #ifdef __linux__
1549 void memory_region_init_ram_from_file(MemoryRegion *mr,
1550 struct Object *owner,
1551 const char *name,
1552 uint64_t size,
1553 uint64_t align,
1554 bool share,
1555 const char *path,
1556 Error **errp)
1557 {
1558 memory_region_init(mr, owner, name, size);
1559 mr->ram = true;
1560 mr->terminates = true;
1561 mr->destructor = memory_region_destructor_ram;
1562 mr->align = align;
1563 mr->ram_block = qemu_ram_alloc_from_file(size, mr, share, path, errp);
1564 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1565 }
1566
1567 void memory_region_init_ram_from_fd(MemoryRegion *mr,
1568 struct Object *owner,
1569 const char *name,
1570 uint64_t size,
1571 bool share,
1572 int fd,
1573 Error **errp)
1574 {
1575 memory_region_init(mr, owner, name, size);
1576 mr->ram = true;
1577 mr->terminates = true;
1578 mr->destructor = memory_region_destructor_ram;
1579 mr->ram_block = qemu_ram_alloc_from_fd(size, mr, share, fd, errp);
1580 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1581 }
1582 #endif
1583
1584 void memory_region_init_ram_ptr(MemoryRegion *mr,
1585 Object *owner,
1586 const char *name,
1587 uint64_t size,
1588 void *ptr)
1589 {
1590 memory_region_init(mr, owner, name, size);
1591 mr->ram = true;
1592 mr->terminates = true;
1593 mr->destructor = memory_region_destructor_ram;
1594 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1595
1596 /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL. */
1597 assert(ptr != NULL);
1598 mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1599 }
1600
1601 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
1602 Object *owner,
1603 const char *name,
1604 uint64_t size,
1605 void *ptr)
1606 {
1607 memory_region_init_ram_ptr(mr, owner, name, size, ptr);
1608 mr->ram_device = true;
1609 mr->ops = &ram_device_mem_ops;
1610 mr->opaque = mr;
1611 }
1612
1613 void memory_region_init_alias(MemoryRegion *mr,
1614 Object *owner,
1615 const char *name,
1616 MemoryRegion *orig,
1617 hwaddr offset,
1618 uint64_t size)
1619 {
1620 memory_region_init(mr, owner, name, size);
1621 mr->alias = orig;
1622 mr->alias_offset = offset;
1623 }
1624
1625 void memory_region_init_rom_nomigrate(MemoryRegion *mr,
1626 struct Object *owner,
1627 const char *name,
1628 uint64_t size,
1629 Error **errp)
1630 {
1631 memory_region_init(mr, owner, name, size);
1632 mr->ram = true;
1633 mr->readonly = true;
1634 mr->terminates = true;
1635 mr->destructor = memory_region_destructor_ram;
1636 mr->ram_block = qemu_ram_alloc(size, false, mr, errp);
1637 mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1638 }
1639
1640 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1641 Object *owner,
1642 const MemoryRegionOps *ops,
1643 void *opaque,
1644 const char *name,
1645 uint64_t size,
1646 Error **errp)
1647 {
1648 assert(ops);
1649 memory_region_init(mr, owner, name, size);
1650 mr->ops = ops;
1651 mr->opaque = opaque;
1652 mr->terminates = true;
1653 mr->rom_device = true;
1654 mr->destructor = memory_region_destructor_ram;
1655 mr->ram_block = qemu_ram_alloc(size, false, mr, errp);
1656 }
1657
1658 void memory_region_init_iommu(void *_iommu_mr,
1659 size_t instance_size,
1660 const char *mrtypename,
1661 Object *owner,
1662 const char *name,
1663 uint64_t size)
1664 {
1665 struct IOMMUMemoryRegion *iommu_mr;
1666 struct MemoryRegion *mr;
1667
1668 object_initialize(_iommu_mr, instance_size, mrtypename);
1669 mr = MEMORY_REGION(_iommu_mr);
1670 memory_region_do_init(mr, owner, name, size);
1671 iommu_mr = IOMMU_MEMORY_REGION(mr);
1672 mr->terminates = true; /* then re-forwards */
1673 QLIST_INIT(&iommu_mr->iommu_notify);
1674 iommu_mr->iommu_notify_flags = IOMMU_NOTIFIER_NONE;
1675 }
1676
1677 static void memory_region_finalize(Object *obj)
1678 {
1679 MemoryRegion *mr = MEMORY_REGION(obj);
1680
1681 assert(!mr->container);
1682
1683 /* We know the region is not visible in any address space (it
1684 * does not have a container and cannot be a root either because
1685 * it has no references, so we can blindly clear mr->enabled.
1686 * memory_region_set_enabled instead could trigger a transaction
1687 * and cause an infinite loop.
1688 */
1689 mr->enabled = false;
1690 memory_region_transaction_begin();
1691 while (!QTAILQ_EMPTY(&mr->subregions)) {
1692 MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1693 memory_region_del_subregion(mr, subregion);
1694 }
1695 memory_region_transaction_commit();
1696
1697 mr->destructor(mr);
1698 memory_region_clear_coalescing(mr);
1699 g_free((char *)mr->name);
1700 g_free(mr->ioeventfds);
1701 }
1702
1703 Object *memory_region_owner(MemoryRegion *mr)
1704 {
1705 Object *obj = OBJECT(mr);
1706 return obj->parent;
1707 }
1708
1709 void memory_region_ref(MemoryRegion *mr)
1710 {
1711 /* MMIO callbacks most likely will access data that belongs
1712 * to the owner, hence the need to ref/unref the owner whenever
1713 * the memory region is in use.
1714 *
1715 * The memory region is a child of its owner. As long as the
1716 * owner doesn't call unparent itself on the memory region,
1717 * ref-ing the owner will also keep the memory region alive.
1718 * Memory regions without an owner are supposed to never go away;
1719 * we do not ref/unref them because it slows down DMA sensibly.
1720 */
1721 if (mr && mr->owner) {
1722 object_ref(mr->owner);
1723 }
1724 }
1725
1726 void memory_region_unref(MemoryRegion *mr)
1727 {
1728 if (mr && mr->owner) {
1729 object_unref(mr->owner);
1730 }
1731 }
1732
1733 uint64_t memory_region_size(MemoryRegion *mr)
1734 {
1735 if (int128_eq(mr->size, int128_2_64())) {
1736 return UINT64_MAX;
1737 }
1738 return int128_get64(mr->size);
1739 }
1740
1741 const char *memory_region_name(const MemoryRegion *mr)
1742 {
1743 if (!mr->name) {
1744 ((MemoryRegion *)mr)->name =
1745 object_get_canonical_path_component(OBJECT(mr));
1746 }
1747 return mr->name;
1748 }
1749
1750 bool memory_region_is_ram_device(MemoryRegion *mr)
1751 {
1752 return mr->ram_device;
1753 }
1754
1755 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1756 {
1757 uint8_t mask = mr->dirty_log_mask;
1758 if (global_dirty_log && mr->ram_block) {
1759 mask |= (1 << DIRTY_MEMORY_MIGRATION);
1760 }
1761 return mask;
1762 }
1763
1764 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1765 {
1766 return memory_region_get_dirty_log_mask(mr) & (1 << client);
1767 }
1768
1769 static void memory_region_update_iommu_notify_flags(IOMMUMemoryRegion *iommu_mr)
1770 {
1771 IOMMUNotifierFlag flags = IOMMU_NOTIFIER_NONE;
1772 IOMMUNotifier *iommu_notifier;
1773 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1774
1775 IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
1776 flags |= iommu_notifier->notifier_flags;
1777 }
1778
1779 if (flags != iommu_mr->iommu_notify_flags && imrc->notify_flag_changed) {
1780 imrc->notify_flag_changed(iommu_mr,
1781 iommu_mr->iommu_notify_flags,
1782 flags);
1783 }
1784
1785 iommu_mr->iommu_notify_flags = flags;
1786 }
1787
1788 void memory_region_register_iommu_notifier(MemoryRegion *mr,
1789 IOMMUNotifier *n)
1790 {
1791 IOMMUMemoryRegion *iommu_mr;
1792
1793 if (mr->alias) {
1794 memory_region_register_iommu_notifier(mr->alias, n);
1795 return;
1796 }
1797
1798 /* We need to register for at least one bitfield */
1799 iommu_mr = IOMMU_MEMORY_REGION(mr);
1800 assert(n->notifier_flags != IOMMU_NOTIFIER_NONE);
1801 assert(n->start <= n->end);
1802 assert(n->iommu_idx >= 0 &&
1803 n->iommu_idx < memory_region_iommu_num_indexes(iommu_mr));
1804
1805 QLIST_INSERT_HEAD(&iommu_mr->iommu_notify, n, node);
1806 memory_region_update_iommu_notify_flags(iommu_mr);
1807 }
1808
1809 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr)
1810 {
1811 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1812
1813 if (imrc->get_min_page_size) {
1814 return imrc->get_min_page_size(iommu_mr);
1815 }
1816 return TARGET_PAGE_SIZE;
1817 }
1818
1819 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
1820 {
1821 MemoryRegion *mr = MEMORY_REGION(iommu_mr);
1822 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1823 hwaddr addr, granularity;
1824 IOMMUTLBEntry iotlb;
1825
1826 /* If the IOMMU has its own replay callback, override */
1827 if (imrc->replay) {
1828 imrc->replay(iommu_mr, n);
1829 return;
1830 }
1831
1832 granularity = memory_region_iommu_get_min_page_size(iommu_mr);
1833
1834 for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1835 iotlb = imrc->translate(iommu_mr, addr, IOMMU_NONE, n->iommu_idx);
1836 if (iotlb.perm != IOMMU_NONE) {
1837 n->notify(n, &iotlb);
1838 }
1839
1840 /* if (2^64 - MR size) < granularity, it's possible to get an
1841 * infinite loop here. This should catch such a wraparound */
1842 if ((addr + granularity) < addr) {
1843 break;
1844 }
1845 }
1846 }
1847
1848 void memory_region_iommu_replay_all(IOMMUMemoryRegion *iommu_mr)
1849 {
1850 IOMMUNotifier *notifier;
1851
1852 IOMMU_NOTIFIER_FOREACH(notifier, iommu_mr) {
1853 memory_region_iommu_replay(iommu_mr, notifier);
1854 }
1855 }
1856
1857 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1858 IOMMUNotifier *n)
1859 {
1860 IOMMUMemoryRegion *iommu_mr;
1861
1862 if (mr->alias) {
1863 memory_region_unregister_iommu_notifier(mr->alias, n);
1864 return;
1865 }
1866 QLIST_REMOVE(n, node);
1867 iommu_mr = IOMMU_MEMORY_REGION(mr);
1868 memory_region_update_iommu_notify_flags(iommu_mr);
1869 }
1870
1871 void memory_region_notify_one(IOMMUNotifier *notifier,
1872 IOMMUTLBEntry *entry)
1873 {
1874 IOMMUNotifierFlag request_flags;
1875
1876 /*
1877 * Skip the notification if the notification does not overlap
1878 * with registered range.
1879 */
1880 if (notifier->start > entry->iova + entry->addr_mask ||
1881 notifier->end < entry->iova) {
1882 return;
1883 }
1884
1885 if (entry->perm & IOMMU_RW) {
1886 request_flags = IOMMU_NOTIFIER_MAP;
1887 } else {
1888 request_flags = IOMMU_NOTIFIER_UNMAP;
1889 }
1890
1891 if (notifier->notifier_flags & request_flags) {
1892 notifier->notify(notifier, entry);
1893 }
1894 }
1895
1896 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
1897 int iommu_idx,
1898 IOMMUTLBEntry entry)
1899 {
1900 IOMMUNotifier *iommu_notifier;
1901
1902 assert(memory_region_is_iommu(MEMORY_REGION(iommu_mr)));
1903
1904 IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
1905 if (iommu_notifier->iommu_idx == iommu_idx) {
1906 memory_region_notify_one(iommu_notifier, &entry);
1907 }
1908 }
1909 }
1910
1911 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
1912 enum IOMMUMemoryRegionAttr attr,
1913 void *data)
1914 {
1915 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1916
1917 if (!imrc->get_attr) {
1918 return -EINVAL;
1919 }
1920
1921 return imrc->get_attr(iommu_mr, attr, data);
1922 }
1923
1924 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr,
1925 MemTxAttrs attrs)
1926 {
1927 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1928
1929 if (!imrc->attrs_to_index) {
1930 return 0;
1931 }
1932
1933 return imrc->attrs_to_index(iommu_mr, attrs);
1934 }
1935
1936 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr)
1937 {
1938 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1939
1940 if (!imrc->num_indexes) {
1941 return 1;
1942 }
1943
1944 return imrc->num_indexes(iommu_mr);
1945 }
1946
1947 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1948 {
1949 uint8_t mask = 1 << client;
1950 uint8_t old_logging;
1951
1952 assert(client == DIRTY_MEMORY_VGA);
1953 old_logging = mr->vga_logging_count;
1954 mr->vga_logging_count += log ? 1 : -1;
1955 if (!!old_logging == !!mr->vga_logging_count) {
1956 return;
1957 }
1958
1959 memory_region_transaction_begin();
1960 mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1961 memory_region_update_pending |= mr->enabled;
1962 memory_region_transaction_commit();
1963 }
1964
1965 bool memory_region_get_dirty(MemoryRegion *mr, hwaddr addr,
1966 hwaddr size, unsigned client)
1967 {
1968 assert(mr->ram_block);
1969 return cpu_physical_memory_get_dirty(memory_region_get_ram_addr(mr) + addr,
1970 size, client);
1971 }
1972
1973 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1974 hwaddr size)
1975 {
1976 assert(mr->ram_block);
1977 cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr,
1978 size,
1979 memory_region_get_dirty_log_mask(mr));
1980 }
1981
1982 static void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
1983 {
1984 MemoryListener *listener;
1985 AddressSpace *as;
1986 FlatView *view;
1987 FlatRange *fr;
1988
1989 /* If the same address space has multiple log_sync listeners, we
1990 * visit that address space's FlatView multiple times. But because
1991 * log_sync listeners are rare, it's still cheaper than walking each
1992 * address space once.
1993 */
1994 QTAILQ_FOREACH(listener, &memory_listeners, link) {
1995 if (!listener->log_sync) {
1996 continue;
1997 }
1998 as = listener->address_space;
1999 view = address_space_get_flatview(as);
2000 FOR_EACH_FLAT_RANGE(fr, view) {
2001 if (fr->dirty_log_mask && (!mr || fr->mr == mr)) {
2002 MemoryRegionSection mrs = section_from_flat_range(fr, view);
2003 listener->log_sync(listener, &mrs);
2004 }
2005 }
2006 flatview_unref(view);
2007 }
2008 }
2009
2010 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
2011 hwaddr addr,
2012 hwaddr size,
2013 unsigned client)
2014 {
2015 assert(mr->ram_block);
2016 memory_region_sync_dirty_bitmap(mr);
2017 return cpu_physical_memory_snapshot_and_clear_dirty(
2018 memory_region_get_ram_addr(mr) + addr, size, client);
2019 }
2020
2021 bool memory_region_snapshot_get_dirty(MemoryRegion *mr, DirtyBitmapSnapshot *snap,
2022 hwaddr addr, hwaddr size)
2023 {
2024 assert(mr->ram_block);
2025 return cpu_physical_memory_snapshot_get_dirty(snap,
2026 memory_region_get_ram_addr(mr) + addr, size);
2027 }
2028
2029 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
2030 {
2031 if (mr->readonly != readonly) {
2032 memory_region_transaction_begin();
2033 mr->readonly = readonly;
2034 memory_region_update_pending |= mr->enabled;
2035 memory_region_transaction_commit();
2036 }
2037 }
2038
2039 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
2040 {
2041 if (mr->romd_mode != romd_mode) {
2042 memory_region_transaction_begin();
2043 mr->romd_mode = romd_mode;
2044 memory_region_update_pending |= mr->enabled;
2045 memory_region_transaction_commit();
2046 }
2047 }
2048
2049 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
2050 hwaddr size, unsigned client)
2051 {
2052 assert(mr->ram_block);
2053 cpu_physical_memory_test_and_clear_dirty(
2054 memory_region_get_ram_addr(mr) + addr, size, client);
2055 }
2056
2057 int memory_region_get_fd(MemoryRegion *mr)
2058 {
2059 int fd;
2060
2061 rcu_read_lock();
2062 while (mr->alias) {
2063 mr = mr->alias;
2064 }
2065 fd = mr->ram_block->fd;
2066 rcu_read_unlock();
2067
2068 return fd;
2069 }
2070
2071 void *memory_region_get_ram_ptr(MemoryRegion *mr)
2072 {
2073 void *ptr;
2074 uint64_t offset = 0;
2075
2076 rcu_read_lock();
2077 while (mr->alias) {
2078 offset += mr->alias_offset;
2079 mr = mr->alias;
2080 }
2081 assert(mr->ram_block);
2082 ptr = qemu_map_ram_ptr(mr->ram_block, offset);
2083 rcu_read_unlock();
2084
2085 return ptr;
2086 }
2087
2088 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset)
2089 {
2090 RAMBlock *block;
2091
2092 block = qemu_ram_block_from_host(ptr, false, offset);
2093 if (!block) {
2094 return NULL;
2095 }
2096
2097 return block->mr;
2098 }
2099
2100 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr)
2101 {
2102 return mr->ram_block ? mr->ram_block->offset : RAM_ADDR_INVALID;
2103 }
2104
2105 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
2106 {
2107 assert(mr->ram_block);
2108
2109 qemu_ram_resize(mr->ram_block, newsize, errp);
2110 }
2111
2112 static void memory_region_update_coalesced_range_as(MemoryRegion *mr, AddressSpace *as)
2113 {
2114 FlatView *view;
2115 FlatRange *fr;
2116 CoalescedMemoryRange *cmr;
2117 AddrRange tmp;
2118 MemoryRegionSection section;
2119
2120 view = address_space_get_flatview(as);
2121 FOR_EACH_FLAT_RANGE(fr, view) {
2122 if (fr->mr == mr) {
2123 section = (MemoryRegionSection) {
2124 .fv = view,
2125 .offset_within_address_space = int128_get64(fr->addr.start),
2126 .size = fr->addr.size,
2127 };
2128
2129 MEMORY_LISTENER_CALL(as, coalesced_mmio_del, Reverse, &section,
2130 int128_get64(fr->addr.start),
2131 int128_get64(fr->addr.size));
2132 QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
2133 tmp = addrrange_shift(cmr->addr,
2134 int128_sub(fr->addr.start,
2135 int128_make64(fr->offset_in_region)));
2136 if (!addrrange_intersects(tmp, fr->addr)) {
2137 continue;
2138 }
2139 tmp = addrrange_intersection(tmp, fr->addr);
2140 MEMORY_LISTENER_CALL(as, coalesced_mmio_add, Forward, &section,
2141 int128_get64(tmp.start),
2142 int128_get64(tmp.size));
2143 }
2144 }
2145 }
2146 flatview_unref(view);
2147 }
2148
2149 static void memory_region_update_coalesced_range(MemoryRegion *mr)
2150 {
2151 AddressSpace *as;
2152
2153 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2154 memory_region_update_coalesced_range_as(mr, as);
2155 }
2156 }
2157
2158 void memory_region_set_coalescing(MemoryRegion *mr)
2159 {
2160 memory_region_clear_coalescing(mr);
2161 memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
2162 }
2163
2164 void memory_region_add_coalescing(MemoryRegion *mr,
2165 hwaddr offset,
2166 uint64_t size)
2167 {
2168 CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
2169
2170 cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
2171 QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
2172 memory_region_update_coalesced_range(mr);
2173 memory_region_set_flush_coalesced(mr);
2174 }
2175
2176 void memory_region_clear_coalescing(MemoryRegion *mr)
2177 {
2178 CoalescedMemoryRange *cmr;
2179 bool updated = false;
2180
2181 qemu_flush_coalesced_mmio_buffer();
2182 mr->flush_coalesced_mmio = false;
2183
2184 while (!QTAILQ_EMPTY(&mr->coalesced)) {
2185 cmr = QTAILQ_FIRST(&mr->coalesced);
2186 QTAILQ_REMOVE(&mr->coalesced, cmr, link);
2187 g_free(cmr);
2188 updated = true;
2189 }
2190
2191 if (updated) {
2192 memory_region_update_coalesced_range(mr);
2193 }
2194 }
2195
2196 void memory_region_set_flush_coalesced(MemoryRegion *mr)
2197 {
2198 mr->flush_coalesced_mmio = true;
2199 }
2200
2201 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
2202 {
2203 qemu_flush_coalesced_mmio_buffer();
2204 if (QTAILQ_EMPTY(&mr->coalesced)) {
2205 mr->flush_coalesced_mmio = false;
2206 }
2207 }
2208
2209 void memory_region_clear_global_locking(MemoryRegion *mr)
2210 {
2211 mr->global_locking = false;
2212 }
2213
2214 static bool userspace_eventfd_warning;
2215
2216 void memory_region_add_eventfd(MemoryRegion *mr,
2217 hwaddr addr,
2218 unsigned size,
2219 bool match_data,
2220 uint64_t data,
2221 EventNotifier *e)
2222 {
2223 MemoryRegionIoeventfd mrfd = {
2224 .addr.start = int128_make64(addr),
2225 .addr.size = int128_make64(size),
2226 .match_data = match_data,
2227 .data = data,
2228 .e = e,
2229 };
2230 unsigned i;
2231
2232 if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
2233 userspace_eventfd_warning))) {
2234 userspace_eventfd_warning = true;
2235 error_report("Using eventfd without MMIO binding in KVM. "
2236 "Suboptimal performance expected");
2237 }
2238
2239 if (size) {
2240 adjust_endianness(mr, &mrfd.data, size);
2241 }
2242 memory_region_transaction_begin();
2243 for (i = 0; i < mr->ioeventfd_nb; ++i) {
2244 if (memory_region_ioeventfd_before(&mrfd, &mr->ioeventfds[i])) {
2245 break;
2246 }
2247 }
2248 ++mr->ioeventfd_nb;
2249 mr->ioeventfds = g_realloc(mr->ioeventfds,
2250 sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
2251 memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
2252 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
2253 mr->ioeventfds[i] = mrfd;
2254 ioeventfd_update_pending |= mr->enabled;
2255 memory_region_transaction_commit();
2256 }
2257
2258 void memory_region_del_eventfd(MemoryRegion *mr,
2259 hwaddr addr,
2260 unsigned size,
2261 bool match_data,
2262 uint64_t data,
2263 EventNotifier *e)
2264 {
2265 MemoryRegionIoeventfd mrfd = {
2266 .addr.start = int128_make64(addr),
2267 .addr.size = int128_make64(size),
2268 .match_data = match_data,
2269 .data = data,
2270 .e = e,
2271 };
2272 unsigned i;
2273
2274 if (size) {
2275 adjust_endianness(mr, &mrfd.data, size);
2276 }
2277 memory_region_transaction_begin();
2278 for (i = 0; i < mr->ioeventfd_nb; ++i) {
2279 if (memory_region_ioeventfd_equal(&mrfd, &mr->ioeventfds[i])) {
2280 break;
2281 }
2282 }
2283 assert(i != mr->ioeventfd_nb);
2284 memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
2285 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
2286 --mr->ioeventfd_nb;
2287 mr->ioeventfds = g_realloc(mr->ioeventfds,
2288 sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
2289 ioeventfd_update_pending |= mr->enabled;
2290 memory_region_transaction_commit();
2291 }
2292
2293 static void memory_region_update_container_subregions(MemoryRegion *subregion)
2294 {
2295 MemoryRegion *mr = subregion->container;
2296 MemoryRegion *other;
2297
2298 memory_region_transaction_begin();
2299
2300 memory_region_ref(subregion);
2301 QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
2302 if (subregion->priority >= other->priority) {
2303 QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
2304 goto done;
2305 }
2306 }
2307 QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
2308 done:
2309 memory_region_update_pending |= mr->enabled && subregion->enabled;
2310 memory_region_transaction_commit();
2311 }
2312
2313 static void memory_region_add_subregion_common(MemoryRegion *mr,
2314 hwaddr offset,
2315 MemoryRegion *subregion)
2316 {
2317 assert(!subregion->container);
2318 subregion->container = mr;
2319 subregion->addr = offset;
2320 memory_region_update_container_subregions(subregion);
2321 }
2322
2323 void memory_region_add_subregion(MemoryRegion *mr,
2324 hwaddr offset,
2325 MemoryRegion *subregion)
2326 {
2327 subregion->priority = 0;
2328 memory_region_add_subregion_common(mr, offset, subregion);
2329 }
2330
2331 void memory_region_add_subregion_overlap(MemoryRegion *mr,
2332 hwaddr offset,
2333 MemoryRegion *subregion,
2334 int priority)
2335 {
2336 subregion->priority = priority;
2337 memory_region_add_subregion_common(mr, offset, subregion);
2338 }
2339
2340 void memory_region_del_subregion(MemoryRegion *mr,
2341 MemoryRegion *subregion)
2342 {
2343 memory_region_transaction_begin();
2344 assert(subregion->container == mr);
2345 subregion->container = NULL;
2346 QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
2347 memory_region_unref(subregion);
2348 memory_region_update_pending |= mr->enabled && subregion->enabled;
2349 memory_region_transaction_commit();
2350 }
2351
2352 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
2353 {
2354 if (enabled == mr->enabled) {
2355 return;
2356 }
2357 memory_region_transaction_begin();
2358 mr->enabled = enabled;
2359 memory_region_update_pending = true;
2360 memory_region_transaction_commit();
2361 }
2362
2363 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
2364 {
2365 Int128 s = int128_make64(size);
2366
2367 if (size == UINT64_MAX) {
2368 s = int128_2_64();
2369 }
2370 if (int128_eq(s, mr->size)) {
2371 return;
2372 }
2373 memory_region_transaction_begin();
2374 mr->size = s;
2375 memory_region_update_pending = true;
2376 memory_region_transaction_commit();
2377 }
2378
2379 static void memory_region_readd_subregion(MemoryRegion *mr)
2380 {
2381 MemoryRegion *container = mr->container;
2382
2383 if (container) {
2384 memory_region_transaction_begin();
2385 memory_region_ref(mr);
2386 memory_region_del_subregion(container, mr);
2387 mr->container = container;
2388 memory_region_update_container_subregions(mr);
2389 memory_region_unref(mr);
2390 memory_region_transaction_commit();
2391 }
2392 }
2393
2394 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
2395 {
2396 if (addr != mr->addr) {
2397 mr->addr = addr;
2398 memory_region_readd_subregion(mr);
2399 }
2400 }
2401
2402 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
2403 {
2404 assert(mr->alias);
2405
2406 if (offset == mr->alias_offset) {
2407 return;
2408 }
2409
2410 memory_region_transaction_begin();
2411 mr->alias_offset = offset;
2412 memory_region_update_pending |= mr->enabled;
2413 memory_region_transaction_commit();
2414 }
2415
2416 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
2417 {
2418 return mr->align;
2419 }
2420
2421 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
2422 {
2423 const AddrRange *addr = addr_;
2424 const FlatRange *fr = fr_;
2425
2426 if (int128_le(addrrange_end(*addr), fr->addr.start)) {
2427 return -1;
2428 } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
2429 return 1;
2430 }
2431 return 0;
2432 }
2433
2434 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
2435 {
2436 return bsearch(&addr, view->ranges, view->nr,
2437 sizeof(FlatRange), cmp_flatrange_addr);
2438 }
2439
2440 bool memory_region_is_mapped(MemoryRegion *mr)
2441 {
2442 return mr->container ? true : false;
2443 }
2444
2445 /* Same as memory_region_find, but it does not add a reference to the
2446 * returned region. It must be called from an RCU critical section.
2447 */
2448 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2449 hwaddr addr, uint64_t size)
2450 {
2451 MemoryRegionSection ret = { .mr = NULL };
2452 MemoryRegion *root;
2453 AddressSpace *as;
2454 AddrRange range;
2455 FlatView *view;
2456 FlatRange *fr;
2457
2458 addr += mr->addr;
2459 for (root = mr; root->container; ) {
2460 root = root->container;
2461 addr += root->addr;
2462 }
2463
2464 as = memory_region_to_address_space(root);
2465 if (!as) {
2466 return ret;
2467 }
2468 range = addrrange_make(int128_make64(addr), int128_make64(size));
2469
2470 view = address_space_to_flatview(as);
2471 fr = flatview_lookup(view, range);
2472 if (!fr) {
2473 return ret;
2474 }
2475
2476 while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2477 --fr;
2478 }
2479
2480 ret.mr = fr->mr;
2481 ret.fv = view;
2482 range = addrrange_intersection(range, fr->addr);
2483 ret.offset_within_region = fr->offset_in_region;
2484 ret.offset_within_region += int128_get64(int128_sub(range.start,
2485 fr->addr.start));
2486 ret.size = range.size;
2487 ret.offset_within_address_space = int128_get64(range.start);
2488 ret.readonly = fr->readonly;
2489 return ret;
2490 }
2491
2492 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2493 hwaddr addr, uint64_t size)
2494 {
2495 MemoryRegionSection ret;
2496 rcu_read_lock();
2497 ret = memory_region_find_rcu(mr, addr, size);
2498 if (ret.mr) {
2499 memory_region_ref(ret.mr);
2500 }
2501 rcu_read_unlock();
2502 return ret;
2503 }
2504
2505 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2506 {
2507 MemoryRegion *mr;
2508
2509 rcu_read_lock();
2510 mr = memory_region_find_rcu(container, addr, 1).mr;
2511 rcu_read_unlock();
2512 return mr && mr != container;
2513 }
2514
2515 void memory_global_dirty_log_sync(void)
2516 {
2517 memory_region_sync_dirty_bitmap(NULL);
2518 }
2519
2520 static VMChangeStateEntry *vmstate_change;
2521
2522 void memory_global_dirty_log_start(void)
2523 {
2524 if (vmstate_change) {
2525 qemu_del_vm_change_state_handler(vmstate_change);
2526 vmstate_change = NULL;
2527 }
2528
2529 global_dirty_log = true;
2530
2531 MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2532
2533 /* Refresh DIRTY_LOG_MIGRATION bit. */
2534 memory_region_transaction_begin();
2535 memory_region_update_pending = true;
2536 memory_region_transaction_commit();
2537 }
2538
2539 static void memory_global_dirty_log_do_stop(void)
2540 {
2541 global_dirty_log = false;
2542
2543 /* Refresh DIRTY_LOG_MIGRATION bit. */
2544 memory_region_transaction_begin();
2545 memory_region_update_pending = true;
2546 memory_region_transaction_commit();
2547
2548 MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2549 }
2550
2551 static void memory_vm_change_state_handler(void *opaque, int running,
2552 RunState state)
2553 {
2554 if (running) {
2555 memory_global_dirty_log_do_stop();
2556
2557 if (vmstate_change) {
2558 qemu_del_vm_change_state_handler(vmstate_change);
2559 vmstate_change = NULL;
2560 }
2561 }
2562 }
2563
2564 void memory_global_dirty_log_stop(void)
2565 {
2566 if (!runstate_is_running()) {
2567 if (vmstate_change) {
2568 return;
2569 }
2570 vmstate_change = qemu_add_vm_change_state_handler(
2571 memory_vm_change_state_handler, NULL);
2572 return;
2573 }
2574
2575 memory_global_dirty_log_do_stop();
2576 }
2577
2578 static void listener_add_address_space(MemoryListener *listener,
2579 AddressSpace *as)
2580 {
2581 FlatView *view;
2582 FlatRange *fr;
2583
2584 if (listener->begin) {
2585 listener->begin(listener);
2586 }
2587 if (global_dirty_log) {
2588 if (listener->log_global_start) {
2589 listener->log_global_start(listener);
2590 }
2591 }
2592
2593 view = address_space_get_flatview(as);
2594 FOR_EACH_FLAT_RANGE(fr, view) {
2595 MemoryRegionSection section = section_from_flat_range(fr, view);
2596
2597 if (listener->region_add) {
2598 listener->region_add(listener, &section);
2599 }
2600 if (fr->dirty_log_mask && listener->log_start) {
2601 listener->log_start(listener, &section, 0, fr->dirty_log_mask);
2602 }
2603 }
2604 if (listener->commit) {
2605 listener->commit(listener);
2606 }
2607 flatview_unref(view);
2608 }
2609
2610 static void listener_del_address_space(MemoryListener *listener,
2611 AddressSpace *as)
2612 {
2613 FlatView *view;
2614 FlatRange *fr;
2615
2616 if (listener->begin) {
2617 listener->begin(listener);
2618 }
2619 view = address_space_get_flatview(as);
2620 FOR_EACH_FLAT_RANGE(fr, view) {
2621 MemoryRegionSection section = section_from_flat_range(fr, view);
2622
2623 if (fr->dirty_log_mask && listener->log_stop) {
2624 listener->log_stop(listener, &section, fr->dirty_log_mask, 0);
2625 }
2626 if (listener->region_del) {
2627 listener->region_del(listener, &section);
2628 }
2629 }
2630 if (listener->commit) {
2631 listener->commit(listener);
2632 }
2633 flatview_unref(view);
2634 }
2635
2636 void memory_listener_register(MemoryListener *listener, AddressSpace *as)
2637 {
2638 MemoryListener *other = NULL;
2639
2640 listener->address_space = as;
2641 if (QTAILQ_EMPTY(&memory_listeners)
2642 || listener->priority >= QTAILQ_LAST(&memory_listeners,
2643 memory_listeners)->priority) {
2644 QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
2645 } else {
2646 QTAILQ_FOREACH(other, &memory_listeners, link) {
2647 if (listener->priority < other->priority) {
2648 break;
2649 }
2650 }
2651 QTAILQ_INSERT_BEFORE(other, listener, link);
2652 }
2653
2654 if (QTAILQ_EMPTY(&as->listeners)
2655 || listener->priority >= QTAILQ_LAST(&as->listeners,
2656 memory_listeners)->priority) {
2657 QTAILQ_INSERT_TAIL(&as->listeners, listener, link_as);
2658 } else {
2659 QTAILQ_FOREACH(other, &as->listeners, link_as) {
2660 if (listener->priority < other->priority) {
2661 break;
2662 }
2663 }
2664 QTAILQ_INSERT_BEFORE(other, listener, link_as);
2665 }
2666
2667 listener_add_address_space(listener, as);
2668 }
2669
2670 void memory_listener_unregister(MemoryListener *listener)
2671 {
2672 if (!listener->address_space) {
2673 return;
2674 }
2675
2676 listener_del_address_space(listener, listener->address_space);
2677 QTAILQ_REMOVE(&memory_listeners, listener, link);
2678 QTAILQ_REMOVE(&listener->address_space->listeners, listener, link_as);
2679 listener->address_space = NULL;
2680 }
2681
2682 bool memory_region_request_mmio_ptr(MemoryRegion *mr, hwaddr addr)
2683 {
2684 void *host;
2685 unsigned size = 0;
2686 unsigned offset = 0;
2687 Object *new_interface;
2688
2689 if (!mr || !mr->ops->request_ptr) {
2690 return false;
2691 }
2692
2693 /*
2694 * Avoid an update if the request_ptr call
2695 * memory_region_invalidate_mmio_ptr which seems to be likely when we use
2696 * a cache.
2697 */
2698 memory_region_transaction_begin();
2699
2700 host = mr->ops->request_ptr(mr->opaque, addr - mr->addr, &size, &offset);
2701
2702 if (!host || !size) {
2703 memory_region_transaction_commit();
2704 return false;
2705 }
2706
2707 new_interface = object_new("mmio_interface");
2708 qdev_prop_set_uint64(DEVICE(new_interface), "start", offset);
2709 qdev_prop_set_uint64(DEVICE(new_interface), "end", offset + size - 1);
2710 qdev_prop_set_bit(DEVICE(new_interface), "ro", true);
2711 qdev_prop_set_ptr(DEVICE(new_interface), "host_ptr", host);
2712 qdev_prop_set_ptr(DEVICE(new_interface), "subregion", mr);
2713 object_property_set_bool(OBJECT(new_interface), true, "realized", NULL);
2714
2715 memory_region_transaction_commit();
2716 return true;
2717 }
2718
2719 typedef struct MMIOPtrInvalidate {
2720 MemoryRegion *mr;
2721 hwaddr offset;
2722 unsigned size;
2723 int busy;
2724 int allocated;
2725 } MMIOPtrInvalidate;
2726
2727 #define MAX_MMIO_INVALIDATE 10
2728 static MMIOPtrInvalidate mmio_ptr_invalidate_list[MAX_MMIO_INVALIDATE];
2729
2730 static void memory_region_do_invalidate_mmio_ptr(CPUState *cpu,
2731 run_on_cpu_data data)
2732 {
2733 MMIOPtrInvalidate *invalidate_data = (MMIOPtrInvalidate *)data.host_ptr;
2734 MemoryRegion *mr = invalidate_data->mr;
2735 hwaddr offset = invalidate_data->offset;
2736 unsigned size = invalidate_data->size;
2737 MemoryRegionSection section = memory_region_find(mr, offset, size);
2738
2739 qemu_mutex_lock_iothread();
2740
2741 /* Reset dirty so this doesn't happen later. */
2742 cpu_physical_memory_test_and_clear_dirty(offset, size, 1);
2743
2744 if (section.mr != mr) {
2745 /* memory_region_find add a ref on section.mr */
2746 memory_region_unref(section.mr);
2747 if (MMIO_INTERFACE(section.mr->owner)) {
2748 /* We found the interface just drop it. */
2749 object_property_set_bool(section.mr->owner, false, "realized",
2750 NULL);
2751 object_unref(section.mr->owner);
2752 object_unparent(section.mr->owner);
2753 }
2754 }
2755
2756 qemu_mutex_unlock_iothread();
2757
2758 if (invalidate_data->allocated) {
2759 g_free(invalidate_data);
2760 } else {
2761 invalidate_data->busy = 0;
2762 }
2763 }
2764
2765 void memory_region_invalidate_mmio_ptr(MemoryRegion *mr, hwaddr offset,
2766 unsigned size)
2767 {
2768 size_t i;
2769 MMIOPtrInvalidate *invalidate_data = NULL;
2770
2771 for (i = 0; i < MAX_MMIO_INVALIDATE; i++) {
2772 if (atomic_cmpxchg(&(mmio_ptr_invalidate_list[i].busy), 0, 1) == 0) {
2773 invalidate_data = &mmio_ptr_invalidate_list[i];
2774 break;
2775 }
2776 }
2777
2778 if (!invalidate_data) {
2779 invalidate_data = g_malloc0(sizeof(MMIOPtrInvalidate));
2780 invalidate_data->allocated = 1;
2781 }
2782
2783 invalidate_data->mr = mr;
2784 invalidate_data->offset = offset;
2785 invalidate_data->size = size;
2786
2787 async_safe_run_on_cpu(first_cpu, memory_region_do_invalidate_mmio_ptr,
2788 RUN_ON_CPU_HOST_PTR(invalidate_data));
2789 }
2790
2791 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
2792 {
2793 memory_region_ref(root);
2794 as->root = root;
2795 as->current_map = NULL;
2796 as->ioeventfd_nb = 0;
2797 as->ioeventfds = NULL;
2798 QTAILQ_INIT(&as->listeners);
2799 QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
2800 as->name = g_strdup(name ? name : "anonymous");
2801 address_space_update_topology(as);
2802 address_space_update_ioeventfds(as);
2803 }
2804
2805 static void do_address_space_destroy(AddressSpace *as)
2806 {
2807 assert(QTAILQ_EMPTY(&as->listeners));
2808
2809 flatview_unref(as->current_map);
2810 g_free(as->name);
2811 g_free(as->ioeventfds);
2812 memory_region_unref(as->root);
2813 }
2814
2815 void address_space_destroy(AddressSpace *as)
2816 {
2817 MemoryRegion *root = as->root;
2818
2819 /* Flush out anything from MemoryListeners listening in on this */
2820 memory_region_transaction_begin();
2821 as->root = NULL;
2822 memory_region_transaction_commit();
2823 QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
2824
2825 /* At this point, as->dispatch and as->current_map are dummy
2826 * entries that the guest should never use. Wait for the old
2827 * values to expire before freeing the data.
2828 */
2829 as->root = root;
2830 call_rcu(as, do_address_space_destroy, rcu);
2831 }
2832
2833 static const char *memory_region_type(MemoryRegion *mr)
2834 {
2835 if (memory_region_is_ram_device(mr)) {
2836 return "ramd";
2837 } else if (memory_region_is_romd(mr)) {
2838 return "romd";
2839 } else if (memory_region_is_rom(mr)) {
2840 return "rom";
2841 } else if (memory_region_is_ram(mr)) {
2842 return "ram";
2843 } else {
2844 return "i/o";
2845 }
2846 }
2847
2848 typedef struct MemoryRegionList MemoryRegionList;
2849
2850 struct MemoryRegionList {
2851 const MemoryRegion *mr;
2852 QTAILQ_ENTRY(MemoryRegionList) mrqueue;
2853 };
2854
2855 typedef QTAILQ_HEAD(mrqueue, MemoryRegionList) MemoryRegionListHead;
2856
2857 #define MR_SIZE(size) (int128_nz(size) ? (hwaddr)int128_get64( \
2858 int128_sub((size), int128_one())) : 0)
2859 #define MTREE_INDENT " "
2860
2861 static void mtree_expand_owner(fprintf_function mon_printf, void *f,
2862 const char *label, Object *obj)
2863 {
2864 DeviceState *dev = (DeviceState *) object_dynamic_cast(obj, TYPE_DEVICE);
2865
2866 mon_printf(f, " %s:{%s", label, dev ? "dev" : "obj");
2867 if (dev && dev->id) {
2868 mon_printf(f, " id=%s", dev->id);
2869 } else {
2870 gchar *canonical_path = object_get_canonical_path(obj);
2871 if (canonical_path) {
2872 mon_printf(f, " path=%s", canonical_path);
2873 g_free(canonical_path);
2874 } else {
2875 mon_printf(f, " type=%s", object_get_typename(obj));
2876 }
2877 }
2878 mon_printf(f, "}");
2879 }
2880
2881 static void mtree_print_mr_owner(fprintf_function mon_printf, void *f,
2882 const MemoryRegion *mr)
2883 {
2884 Object *owner = mr->owner;
2885 Object *parent = memory_region_owner((MemoryRegion *)mr);
2886
2887 if (!owner && !parent) {
2888 mon_printf(f, " orphan");
2889 return;
2890 }
2891 if (owner) {
2892 mtree_expand_owner(mon_printf, f, "owner", owner);
2893 }
2894 if (parent && parent != owner) {
2895 mtree_expand_owner(mon_printf, f, "parent", parent);
2896 }
2897 }
2898
2899 static void mtree_print_mr(fprintf_function mon_printf, void *f,
2900 const MemoryRegion *mr, unsigned int level,
2901 hwaddr base,
2902 MemoryRegionListHead *alias_print_queue,
2903 bool owner)
2904 {
2905 MemoryRegionList *new_ml, *ml, *next_ml;
2906 MemoryRegionListHead submr_print_queue;
2907 const MemoryRegion *submr;
2908 unsigned int i;
2909 hwaddr cur_start, cur_end;
2910
2911 if (!mr) {
2912 return;
2913 }
2914
2915 for (i = 0; i < level; i++) {
2916 mon_printf(f, MTREE_INDENT);
2917 }
2918
2919 cur_start = base + mr->addr;
2920 cur_end = cur_start + MR_SIZE(mr->size);
2921
2922 /*
2923 * Try to detect overflow of memory region. This should never
2924 * happen normally. When it happens, we dump something to warn the
2925 * user who is observing this.
2926 */
2927 if (cur_start < base || cur_end < cur_start) {
2928 mon_printf(f, "[DETECTED OVERFLOW!] ");
2929 }
2930
2931 if (mr->alias) {
2932 MemoryRegionList *ml;
2933 bool found = false;
2934
2935 /* check if the alias is already in the queue */
2936 QTAILQ_FOREACH(ml, alias_print_queue, mrqueue) {
2937 if (ml->mr == mr->alias) {
2938 found = true;
2939 }
2940 }
2941
2942 if (!found) {
2943 ml = g_new(MemoryRegionList, 1);
2944 ml->mr = mr->alias;
2945 QTAILQ_INSERT_TAIL(alias_print_queue, ml, mrqueue);
2946 }
2947 mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx
2948 " (prio %d, %s): alias %s @%s " TARGET_FMT_plx
2949 "-" TARGET_FMT_plx "%s",
2950 cur_start, cur_end,
2951 mr->priority,
2952 memory_region_type((MemoryRegion *)mr),
2953 memory_region_name(mr),
2954 memory_region_name(mr->alias),
2955 mr->alias_offset,
2956 mr->alias_offset + MR_SIZE(mr->size),
2957 mr->enabled ? "" : " [disabled]");
2958 if (owner) {
2959 mtree_print_mr_owner(mon_printf, f, mr);
2960 }
2961 } else {
2962 mon_printf(f,
2963 TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d, %s): %s%s",
2964 cur_start, cur_end,
2965 mr->priority,
2966 memory_region_type((MemoryRegion *)mr),
2967 memory_region_name(mr),
2968 mr->enabled ? "" : " [disabled]");
2969 if (owner) {
2970 mtree_print_mr_owner(mon_printf, f, mr);
2971 }
2972 }
2973 mon_printf(f, "\n");
2974
2975 QTAILQ_INIT(&submr_print_queue);
2976
2977 QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
2978 new_ml = g_new(MemoryRegionList, 1);
2979 new_ml->mr = submr;
2980 QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
2981 if (new_ml->mr->addr < ml->mr->addr ||
2982 (new_ml->mr->addr == ml->mr->addr &&
2983 new_ml->mr->priority > ml->mr->priority)) {
2984 QTAILQ_INSERT_BEFORE(ml, new_ml, mrqueue);
2985 new_ml = NULL;
2986 break;
2987 }
2988 }
2989 if (new_ml) {
2990 QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, mrqueue);
2991 }
2992 }
2993
2994 QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
2995 mtree_print_mr(mon_printf, f, ml->mr, level + 1, cur_start,
2996 alias_print_queue, owner);
2997 }
2998
2999 QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, mrqueue, next_ml) {
3000 g_free(ml);
3001 }
3002 }
3003
3004 struct FlatViewInfo {
3005 fprintf_function mon_printf;
3006 void *f;
3007 int counter;
3008 bool dispatch_tree;
3009 bool owner;
3010 };
3011
3012 static void mtree_print_flatview(gpointer key, gpointer value,
3013 gpointer user_data)
3014 {
3015 FlatView *view = key;
3016 GArray *fv_address_spaces = value;
3017 struct FlatViewInfo *fvi = user_data;
3018 fprintf_function p = fvi->mon_printf;
3019 void *f = fvi->f;
3020 FlatRange *range = &view->ranges[0];
3021 MemoryRegion *mr;
3022 int n = view->nr;
3023 int i;
3024 AddressSpace *as;
3025
3026 p(f, "FlatView #%d\n", fvi->counter);
3027 ++fvi->counter;
3028
3029 for (i = 0; i < fv_address_spaces->len; ++i) {
3030 as = g_array_index(fv_address_spaces, AddressSpace*, i);
3031 p(f, " AS \"%s\", root: %s", as->name, memory_region_name(as->root));
3032 if (as->root->alias) {
3033 p(f, ", alias %s", memory_region_name(as->root->alias));
3034 }
3035 p(f, "\n");
3036 }
3037
3038 p(f, " Root memory region: %s\n",
3039 view->root ? memory_region_name(view->root) : "(none)");
3040
3041 if (n <= 0) {
3042 p(f, MTREE_INDENT "No rendered FlatView\n\n");
3043 return;
3044 }
3045
3046 while (n--) {
3047 mr = range->mr;
3048 if (range->offset_in_region) {
3049 p(f, MTREE_INDENT TARGET_FMT_plx "-"
3050 TARGET_FMT_plx " (prio %d, %s): %s @" TARGET_FMT_plx,
3051 int128_get64(range->addr.start),
3052 int128_get64(range->addr.start) + MR_SIZE(range->addr.size),
3053 mr->priority,
3054 range->readonly ? "rom" : memory_region_type(mr),
3055 memory_region_name(mr),
3056 range->offset_in_region);
3057 } else {
3058 p(f, MTREE_INDENT TARGET_FMT_plx "-"
3059 TARGET_FMT_plx " (prio %d, %s): %s",
3060 int128_get64(range->addr.start),
3061 int128_get64(range->addr.start) + MR_SIZE(range->addr.size),
3062 mr->priority,
3063 range->readonly ? "rom" : memory_region_type(mr),
3064 memory_region_name(mr));
3065 }
3066 if (fvi->owner) {
3067 mtree_print_mr_owner(p, f, mr);
3068 }
3069 p(f, "\n");
3070 range++;
3071 }
3072
3073 #if !defined(CONFIG_USER_ONLY)
3074 if (fvi->dispatch_tree && view->root) {
3075 mtree_print_dispatch(p, f, view->dispatch, view->root);
3076 }
3077 #endif
3078
3079 p(f, "\n");
3080 }
3081
3082 static gboolean mtree_info_flatview_free(gpointer key, gpointer value,
3083 gpointer user_data)
3084 {
3085 FlatView *view = key;
3086 GArray *fv_address_spaces = value;
3087
3088 g_array_unref(fv_address_spaces);
3089 flatview_unref(view);
3090
3091 return true;
3092 }
3093
3094 void mtree_info(fprintf_function mon_printf, void *f, bool flatview,
3095 bool dispatch_tree, bool owner)
3096 {
3097 MemoryRegionListHead ml_head;
3098 MemoryRegionList *ml, *ml2;
3099 AddressSpace *as;
3100
3101 if (flatview) {
3102 FlatView *view;
3103 struct FlatViewInfo fvi = {
3104 .mon_printf = mon_printf,
3105 .f = f,
3106 .counter = 0,
3107 .dispatch_tree = dispatch_tree,
3108 .owner = owner,
3109 };
3110 GArray *fv_address_spaces;
3111 GHashTable *views = g_hash_table_new(g_direct_hash, g_direct_equal);
3112
3113 /* Gather all FVs in one table */
3114 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3115 view = address_space_get_flatview(as);
3116
3117 fv_address_spaces = g_hash_table_lookup(views, view);
3118 if (!fv_address_spaces) {
3119 fv_address_spaces = g_array_new(false, false, sizeof(as));
3120 g_hash_table_insert(views, view, fv_address_spaces);
3121 }
3122
3123 g_array_append_val(fv_address_spaces, as);
3124 }
3125
3126 /* Print */
3127 g_hash_table_foreach(views, mtree_print_flatview, &fvi);
3128
3129 /* Free */
3130 g_hash_table_foreach_remove(views, mtree_info_flatview_free, 0);
3131 g_hash_table_unref(views);
3132
3133 return;
3134 }
3135
3136 QTAILQ_INIT(&ml_head);
3137
3138 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3139 mon_printf(f, "address-space: %s\n", as->name);
3140 mtree_print_mr(mon_printf, f, as->root, 1, 0, &ml_head, owner);
3141 mon_printf(f, "\n");
3142 }
3143
3144 /* print aliased regions */
3145 QTAILQ_FOREACH(ml, &ml_head, mrqueue) {
3146 mon_printf(f, "memory-region: %s\n", memory_region_name(ml->mr));
3147 mtree_print_mr(mon_printf, f, ml->mr, 1, 0, &ml_head, owner);
3148 mon_printf(f, "\n");
3149 }
3150
3151 QTAILQ_FOREACH_SAFE(ml, &ml_head, mrqueue, ml2) {
3152 g_free(ml);
3153 }
3154 }
3155
3156 void memory_region_init_ram(MemoryRegion *mr,
3157 struct Object *owner,
3158 const char *name,
3159 uint64_t size,
3160 Error **errp)
3161 {
3162 DeviceState *owner_dev;
3163 Error *err = NULL;
3164
3165 memory_region_init_ram_nomigrate(mr, owner, name, size, &err);
3166 if (err) {
3167 error_propagate(errp, err);
3168 return;
3169 }
3170 /* This will assert if owner is neither NULL nor a DeviceState.
3171 * We only want the owner here for the purposes of defining a
3172 * unique name for migration. TODO: Ideally we should implement
3173 * a naming scheme for Objects which are not DeviceStates, in
3174 * which case we can relax this restriction.
3175 */
3176 owner_dev = DEVICE(owner);
3177 vmstate_register_ram(mr, owner_dev);
3178 }
3179
3180 void memory_region_init_rom(MemoryRegion *mr,
3181 struct Object *owner,
3182 const char *name,
3183 uint64_t size,
3184 Error **errp)
3185 {
3186 DeviceState *owner_dev;
3187 Error *err = NULL;
3188
3189 memory_region_init_rom_nomigrate(mr, owner, name, size, &err);
3190 if (err) {
3191 error_propagate(errp, err);
3192 return;
3193 }
3194 /* This will assert if owner is neither NULL nor a DeviceState.
3195 * We only want the owner here for the purposes of defining a
3196 * unique name for migration. TODO: Ideally we should implement
3197 * a naming scheme for Objects which are not DeviceStates, in
3198 * which case we can relax this restriction.
3199 */
3200 owner_dev = DEVICE(owner);
3201 vmstate_register_ram(mr, owner_dev);
3202 }
3203
3204 void memory_region_init_rom_device(MemoryRegion *mr,
3205 struct Object *owner,
3206 const MemoryRegionOps *ops,
3207 void *opaque,
3208 const char *name,
3209 uint64_t size,
3210 Error **errp)
3211 {
3212 DeviceState *owner_dev;
3213 Error *err = NULL;
3214
3215 memory_region_init_rom_device_nomigrate(mr, owner, ops, opaque,
3216 name, size, &err);
3217 if (err) {
3218 error_propagate(errp, err);
3219 return;
3220 }
3221 /* This will assert if owner is neither NULL nor a DeviceState.
3222 * We only want the owner here for the purposes of defining a
3223 * unique name for migration. TODO: Ideally we should implement
3224 * a naming scheme for Objects which are not DeviceStates, in
3225 * which case we can relax this restriction.
3226 */
3227 owner_dev = DEVICE(owner);
3228 vmstate_register_ram(mr, owner_dev);
3229 }
3230
3231 static const TypeInfo memory_region_info = {
3232 .parent = TYPE_OBJECT,
3233 .name = TYPE_MEMORY_REGION,
3234 .instance_size = sizeof(MemoryRegion),
3235 .instance_init = memory_region_initfn,
3236 .instance_finalize = memory_region_finalize,
3237 };
3238
3239 static const TypeInfo iommu_memory_region_info = {
3240 .parent = TYPE_MEMORY_REGION,
3241 .name = TYPE_IOMMU_MEMORY_REGION,
3242 .class_size = sizeof(IOMMUMemoryRegionClass),
3243 .instance_size = sizeof(IOMMUMemoryRegion),
3244 .instance_init = iommu_memory_region_initfn,
3245 .abstract = true,
3246 };
3247
3248 static void memory_register_types(void)
3249 {
3250 type_register_static(&memory_region_info);
3251 type_register_static(&iommu_memory_region_info);
3252 }
3253
3254 type_init(memory_register_types)
Eric Blake's pending patches to qemu