target/s390x/kvm/pv: Provide some more useful information if decryption fails
[qemu/kevin.git] / system / memory.c
bloba229a79988fce2aa3cb77e3a130db4c694e8cd49
1 /*
2 * Physical memory management
4 * Copyright 2011 Red Hat, Inc. and/or its affiliates
6 * Authors:
7 * Avi Kivity <avi@redhat.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
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.
16 #include "qemu/osdep.h"
17 #include "qemu/log.h"
18 #include "qapi/error.h"
19 #include "exec/memory.h"
20 #include "qapi/visitor.h"
21 #include "qemu/bitops.h"
22 #include "qemu/error-report.h"
23 #include "qemu/main-loop.h"
24 #include "qemu/qemu-print.h"
25 #include "qom/object.h"
26 #include "trace.h"
28 #include "exec/memory-internal.h"
29 #include "exec/ram_addr.h"
30 #include "sysemu/kvm.h"
31 #include "sysemu/runstate.h"
32 #include "sysemu/tcg.h"
33 #include "qemu/accel.h"
34 #include "hw/boards.h"
35 #include "migration/vmstate.h"
36 #include "exec/address-spaces.h"
38 //#define DEBUG_UNASSIGNED
40 static unsigned memory_region_transaction_depth;
41 static bool memory_region_update_pending;
42 static bool ioeventfd_update_pending;
43 unsigned int global_dirty_tracking;
45 static QTAILQ_HEAD(, MemoryListener) memory_listeners
46 = QTAILQ_HEAD_INITIALIZER(memory_listeners);
48 static QTAILQ_HEAD(, AddressSpace) address_spaces
49 = QTAILQ_HEAD_INITIALIZER(address_spaces);
51 static GHashTable *flat_views;
53 typedef struct AddrRange AddrRange;
56 * Note that signed integers are needed for negative offsetting in aliases
57 * (large MemoryRegion::alias_offset).
59 struct AddrRange {
60 Int128 start;
61 Int128 size;
64 static AddrRange addrrange_make(Int128 start, Int128 size)
66 return (AddrRange) { start, size };
69 static bool addrrange_equal(AddrRange r1, AddrRange r2)
71 return int128_eq(r1.start, r2.start) && int128_eq(r1.size, r2.size);
74 static Int128 addrrange_end(AddrRange r)
76 return int128_add(r.start, r.size);
79 static AddrRange addrrange_shift(AddrRange range, Int128 delta)
81 int128_addto(&range.start, delta);
82 return range;
85 static bool addrrange_contains(AddrRange range, Int128 addr)
87 return int128_ge(addr, range.start)
88 && int128_lt(addr, addrrange_end(range));
91 static bool addrrange_intersects(AddrRange r1, AddrRange r2)
93 return addrrange_contains(r1, r2.start)
94 || addrrange_contains(r2, r1.start);
97 static AddrRange addrrange_intersection(AddrRange r1, AddrRange r2)
99 Int128 start = int128_max(r1.start, r2.start);
100 Int128 end = int128_min(addrrange_end(r1), addrrange_end(r2));
101 return addrrange_make(start, int128_sub(end, start));
104 enum ListenerDirection { Forward, Reverse };
106 #define MEMORY_LISTENER_CALL_GLOBAL(_callback, _direction, _args...) \
107 do { \
108 MemoryListener *_listener; \
110 switch (_direction) { \
111 case Forward: \
112 QTAILQ_FOREACH(_listener, &memory_listeners, link) { \
113 if (_listener->_callback) { \
114 _listener->_callback(_listener, ##_args); \
117 break; \
118 case Reverse: \
119 QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners, link) { \
120 if (_listener->_callback) { \
121 _listener->_callback(_listener, ##_args); \
124 break; \
125 default: \
126 abort(); \
128 } while (0)
130 #define MEMORY_LISTENER_CALL(_as, _callback, _direction, _section, _args...) \
131 do { \
132 MemoryListener *_listener; \
134 switch (_direction) { \
135 case Forward: \
136 QTAILQ_FOREACH(_listener, &(_as)->listeners, link_as) { \
137 if (_listener->_callback) { \
138 _listener->_callback(_listener, _section, ##_args); \
141 break; \
142 case Reverse: \
143 QTAILQ_FOREACH_REVERSE(_listener, &(_as)->listeners, link_as) { \
144 if (_listener->_callback) { \
145 _listener->_callback(_listener, _section, ##_args); \
148 break; \
149 default: \
150 abort(); \
152 } while (0)
154 /* No need to ref/unref .mr, the FlatRange keeps it alive. */
155 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...) \
156 do { \
157 MemoryRegionSection mrs = section_from_flat_range(fr, \
158 address_space_to_flatview(as)); \
159 MEMORY_LISTENER_CALL(as, callback, dir, &mrs, ##_args); \
160 } while(0)
162 struct CoalescedMemoryRange {
163 AddrRange addr;
164 QTAILQ_ENTRY(CoalescedMemoryRange) link;
167 struct MemoryRegionIoeventfd {
168 AddrRange addr;
169 bool match_data;
170 uint64_t data;
171 EventNotifier *e;
174 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd *a,
175 MemoryRegionIoeventfd *b)
177 if (int128_lt(a->addr.start, b->addr.start)) {
178 return true;
179 } else if (int128_gt(a->addr.start, b->addr.start)) {
180 return false;
181 } else if (int128_lt(a->addr.size, b->addr.size)) {
182 return true;
183 } else if (int128_gt(a->addr.size, b->addr.size)) {
184 return false;
185 } else if (a->match_data < b->match_data) {
186 return true;
187 } else if (a->match_data > b->match_data) {
188 return false;
189 } else if (a->match_data) {
190 if (a->data < b->data) {
191 return true;
192 } else if (a->data > b->data) {
193 return false;
196 if (a->e < b->e) {
197 return true;
198 } else if (a->e > b->e) {
199 return false;
201 return false;
204 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd *a,
205 MemoryRegionIoeventfd *b)
207 if (int128_eq(a->addr.start, b->addr.start) &&
208 (!int128_nz(a->addr.size) || !int128_nz(b->addr.size) ||
209 (int128_eq(a->addr.size, b->addr.size) &&
210 (a->match_data == b->match_data) &&
211 ((a->match_data && (a->data == b->data)) || !a->match_data) &&
212 (a->e == b->e))))
213 return true;
215 return false;
218 /* Range of memory in the global map. Addresses are absolute. */
219 struct FlatRange {
220 MemoryRegion *mr;
221 hwaddr offset_in_region;
222 AddrRange addr;
223 uint8_t dirty_log_mask;
224 bool romd_mode;
225 bool readonly;
226 bool nonvolatile;
227 bool unmergeable;
230 #define FOR_EACH_FLAT_RANGE(var, view) \
231 for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
233 static inline MemoryRegionSection
234 section_from_flat_range(FlatRange *fr, FlatView *fv)
236 return (MemoryRegionSection) {
237 .mr = fr->mr,
238 .fv = fv,
239 .offset_within_region = fr->offset_in_region,
240 .size = fr->addr.size,
241 .offset_within_address_space = int128_get64(fr->addr.start),
242 .readonly = fr->readonly,
243 .nonvolatile = fr->nonvolatile,
244 .unmergeable = fr->unmergeable,
248 static bool flatrange_equal(FlatRange *a, FlatRange *b)
250 return a->mr == b->mr
251 && addrrange_equal(a->addr, b->addr)
252 && a->offset_in_region == b->offset_in_region
253 && a->romd_mode == b->romd_mode
254 && a->readonly == b->readonly
255 && a->nonvolatile == b->nonvolatile
256 && a->unmergeable == b->unmergeable;
259 static FlatView *flatview_new(MemoryRegion *mr_root)
261 FlatView *view;
263 view = g_new0(FlatView, 1);
264 view->ref = 1;
265 view->root = mr_root;
266 memory_region_ref(mr_root);
267 trace_flatview_new(view, mr_root);
269 return view;
272 /* Insert a range into a given position. Caller is responsible for maintaining
273 * sorting order.
275 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
277 if (view->nr == view->nr_allocated) {
278 view->nr_allocated = MAX(2 * view->nr, 10);
279 view->ranges = g_realloc(view->ranges,
280 view->nr_allocated * sizeof(*view->ranges));
282 memmove(view->ranges + pos + 1, view->ranges + pos,
283 (view->nr - pos) * sizeof(FlatRange));
284 view->ranges[pos] = *range;
285 memory_region_ref(range->mr);
286 ++view->nr;
289 static void flatview_destroy(FlatView *view)
291 int i;
293 trace_flatview_destroy(view, view->root);
294 if (view->dispatch) {
295 address_space_dispatch_free(view->dispatch);
297 for (i = 0; i < view->nr; i++) {
298 memory_region_unref(view->ranges[i].mr);
300 g_free(view->ranges);
301 memory_region_unref(view->root);
302 g_free(view);
305 static bool flatview_ref(FlatView *view)
307 return qatomic_fetch_inc_nonzero(&view->ref) > 0;
310 void flatview_unref(FlatView *view)
312 if (qatomic_fetch_dec(&view->ref) == 1) {
313 trace_flatview_destroy_rcu(view, view->root);
314 assert(view->root);
315 call_rcu(view, flatview_destroy, rcu);
319 static bool can_merge(FlatRange *r1, FlatRange *r2)
321 return int128_eq(addrrange_end(r1->addr), r2->addr.start)
322 && r1->mr == r2->mr
323 && int128_eq(int128_add(int128_make64(r1->offset_in_region),
324 r1->addr.size),
325 int128_make64(r2->offset_in_region))
326 && r1->dirty_log_mask == r2->dirty_log_mask
327 && r1->romd_mode == r2->romd_mode
328 && r1->readonly == r2->readonly
329 && r1->nonvolatile == r2->nonvolatile
330 && !r1->unmergeable && !r2->unmergeable;
333 /* Attempt to simplify a view by merging adjacent ranges */
334 static void flatview_simplify(FlatView *view)
336 unsigned i, j, k;
338 i = 0;
339 while (i < view->nr) {
340 j = i + 1;
341 while (j < view->nr
342 && can_merge(&view->ranges[j-1], &view->ranges[j])) {
343 int128_addto(&view->ranges[i].addr.size, view->ranges[j].addr.size);
344 ++j;
346 ++i;
347 for (k = i; k < j; k++) {
348 memory_region_unref(view->ranges[k].mr);
350 memmove(&view->ranges[i], &view->ranges[j],
351 (view->nr - j) * sizeof(view->ranges[j]));
352 view->nr -= j - i;
356 static bool memory_region_big_endian(MemoryRegion *mr)
358 #if TARGET_BIG_ENDIAN
359 return mr->ops->endianness != DEVICE_LITTLE_ENDIAN;
360 #else
361 return mr->ops->endianness == DEVICE_BIG_ENDIAN;
362 #endif
365 static void adjust_endianness(MemoryRegion *mr, uint64_t *data, MemOp op)
367 if ((op & MO_BSWAP) != devend_memop(mr->ops->endianness)) {
368 switch (op & MO_SIZE) {
369 case MO_8:
370 break;
371 case MO_16:
372 *data = bswap16(*data);
373 break;
374 case MO_32:
375 *data = bswap32(*data);
376 break;
377 case MO_64:
378 *data = bswap64(*data);
379 break;
380 default:
381 g_assert_not_reached();
386 static inline void memory_region_shift_read_access(uint64_t *value,
387 signed shift,
388 uint64_t mask,
389 uint64_t tmp)
391 if (shift >= 0) {
392 *value |= (tmp & mask) << shift;
393 } else {
394 *value |= (tmp & mask) >> -shift;
398 static inline uint64_t memory_region_shift_write_access(uint64_t *value,
399 signed shift,
400 uint64_t mask)
402 uint64_t tmp;
404 if (shift >= 0) {
405 tmp = (*value >> shift) & mask;
406 } else {
407 tmp = (*value << -shift) & mask;
410 return tmp;
413 static hwaddr memory_region_to_absolute_addr(MemoryRegion *mr, hwaddr offset)
415 MemoryRegion *root;
416 hwaddr abs_addr = offset;
418 abs_addr += mr->addr;
419 for (root = mr; root->container; ) {
420 root = root->container;
421 abs_addr += root->addr;
424 return abs_addr;
427 static int get_cpu_index(void)
429 if (current_cpu) {
430 return current_cpu->cpu_index;
432 return -1;
435 static MemTxResult memory_region_read_accessor(MemoryRegion *mr,
436 hwaddr addr,
437 uint64_t *value,
438 unsigned size,
439 signed shift,
440 uint64_t mask,
441 MemTxAttrs attrs)
443 uint64_t tmp;
445 tmp = mr->ops->read(mr->opaque, addr, size);
446 if (mr->subpage) {
447 trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
448 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_READ)) {
449 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
450 trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size,
451 memory_region_name(mr));
453 memory_region_shift_read_access(value, shift, mask, tmp);
454 return MEMTX_OK;
457 static MemTxResult memory_region_read_with_attrs_accessor(MemoryRegion *mr,
458 hwaddr addr,
459 uint64_t *value,
460 unsigned size,
461 signed shift,
462 uint64_t mask,
463 MemTxAttrs attrs)
465 uint64_t tmp = 0;
466 MemTxResult r;
468 r = mr->ops->read_with_attrs(mr->opaque, addr, &tmp, size, attrs);
469 if (mr->subpage) {
470 trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
471 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_READ)) {
472 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
473 trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size,
474 memory_region_name(mr));
476 memory_region_shift_read_access(value, shift, mask, tmp);
477 return r;
480 static MemTxResult memory_region_write_accessor(MemoryRegion *mr,
481 hwaddr addr,
482 uint64_t *value,
483 unsigned size,
484 signed shift,
485 uint64_t mask,
486 MemTxAttrs attrs)
488 uint64_t tmp = memory_region_shift_write_access(value, shift, mask);
490 if (mr->subpage) {
491 trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
492 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_WRITE)) {
493 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
494 trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size,
495 memory_region_name(mr));
497 mr->ops->write(mr->opaque, addr, tmp, size);
498 return MEMTX_OK;
501 static MemTxResult memory_region_write_with_attrs_accessor(MemoryRegion *mr,
502 hwaddr addr,
503 uint64_t *value,
504 unsigned size,
505 signed shift,
506 uint64_t mask,
507 MemTxAttrs attrs)
509 uint64_t tmp = memory_region_shift_write_access(value, shift, mask);
511 if (mr->subpage) {
512 trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
513 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_WRITE)) {
514 hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
515 trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size,
516 memory_region_name(mr));
518 return mr->ops->write_with_attrs(mr->opaque, addr, tmp, size, attrs);
521 static MemTxResult access_with_adjusted_size(hwaddr addr,
522 uint64_t *value,
523 unsigned size,
524 unsigned access_size_min,
525 unsigned access_size_max,
526 MemTxResult (*access_fn)
527 (MemoryRegion *mr,
528 hwaddr addr,
529 uint64_t *value,
530 unsigned size,
531 signed shift,
532 uint64_t mask,
533 MemTxAttrs attrs),
534 MemoryRegion *mr,
535 MemTxAttrs attrs)
537 uint64_t access_mask;
538 unsigned access_size;
539 unsigned i;
540 MemTxResult r = MEMTX_OK;
541 bool reentrancy_guard_applied = false;
543 if (!access_size_min) {
544 access_size_min = 1;
546 if (!access_size_max) {
547 access_size_max = 4;
550 /* Do not allow more than one simultaneous access to a device's IO Regions */
551 if (mr->dev && !mr->disable_reentrancy_guard &&
552 !mr->ram_device && !mr->ram && !mr->rom_device && !mr->readonly) {
553 if (mr->dev->mem_reentrancy_guard.engaged_in_io) {
554 warn_report_once("Blocked re-entrant IO on MemoryRegion: "
555 "%s at addr: 0x%" HWADDR_PRIX,
556 memory_region_name(mr), addr);
557 return MEMTX_ACCESS_ERROR;
559 mr->dev->mem_reentrancy_guard.engaged_in_io = true;
560 reentrancy_guard_applied = true;
563 /* FIXME: support unaligned access? */
564 access_size = MAX(MIN(size, access_size_max), access_size_min);
565 access_mask = MAKE_64BIT_MASK(0, access_size * 8);
566 if (memory_region_big_endian(mr)) {
567 for (i = 0; i < size; i += access_size) {
568 r |= access_fn(mr, addr + i, value, access_size,
569 (size - access_size - i) * 8, access_mask, attrs);
571 } else {
572 for (i = 0; i < size; i += access_size) {
573 r |= access_fn(mr, addr + i, value, access_size, i * 8,
574 access_mask, attrs);
577 if (mr->dev && reentrancy_guard_applied) {
578 mr->dev->mem_reentrancy_guard.engaged_in_io = false;
580 return r;
583 static AddressSpace *memory_region_to_address_space(MemoryRegion *mr)
585 AddressSpace *as;
587 while (mr->container) {
588 mr = mr->container;
590 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
591 if (mr == as->root) {
592 return as;
595 return NULL;
598 /* Render a memory region into the global view. Ranges in @view obscure
599 * ranges in @mr.
601 static void render_memory_region(FlatView *view,
602 MemoryRegion *mr,
603 Int128 base,
604 AddrRange clip,
605 bool readonly,
606 bool nonvolatile,
607 bool unmergeable)
609 MemoryRegion *subregion;
610 unsigned i;
611 hwaddr offset_in_region;
612 Int128 remain;
613 Int128 now;
614 FlatRange fr;
615 AddrRange tmp;
617 if (!mr->enabled) {
618 return;
621 int128_addto(&base, int128_make64(mr->addr));
622 readonly |= mr->readonly;
623 nonvolatile |= mr->nonvolatile;
624 unmergeable |= mr->unmergeable;
626 tmp = addrrange_make(base, mr->size);
628 if (!addrrange_intersects(tmp, clip)) {
629 return;
632 clip = addrrange_intersection(tmp, clip);
634 if (mr->alias) {
635 int128_subfrom(&base, int128_make64(mr->alias->addr));
636 int128_subfrom(&base, int128_make64(mr->alias_offset));
637 render_memory_region(view, mr->alias, base, clip,
638 readonly, nonvolatile, unmergeable);
639 return;
642 /* Render subregions in priority order. */
643 QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
644 render_memory_region(view, subregion, base, clip,
645 readonly, nonvolatile, unmergeable);
648 if (!mr->terminates) {
649 return;
652 offset_in_region = int128_get64(int128_sub(clip.start, base));
653 base = clip.start;
654 remain = clip.size;
656 fr.mr = mr;
657 fr.dirty_log_mask = memory_region_get_dirty_log_mask(mr);
658 fr.romd_mode = mr->romd_mode;
659 fr.readonly = readonly;
660 fr.nonvolatile = nonvolatile;
661 fr.unmergeable = unmergeable;
663 /* Render the region itself into any gaps left by the current view. */
664 for (i = 0; i < view->nr && int128_nz(remain); ++i) {
665 if (int128_ge(base, addrrange_end(view->ranges[i].addr))) {
666 continue;
668 if (int128_lt(base, view->ranges[i].addr.start)) {
669 now = int128_min(remain,
670 int128_sub(view->ranges[i].addr.start, base));
671 fr.offset_in_region = offset_in_region;
672 fr.addr = addrrange_make(base, now);
673 flatview_insert(view, i, &fr);
674 ++i;
675 int128_addto(&base, now);
676 offset_in_region += int128_get64(now);
677 int128_subfrom(&remain, now);
679 now = int128_sub(int128_min(int128_add(base, remain),
680 addrrange_end(view->ranges[i].addr)),
681 base);
682 int128_addto(&base, now);
683 offset_in_region += int128_get64(now);
684 int128_subfrom(&remain, now);
686 if (int128_nz(remain)) {
687 fr.offset_in_region = offset_in_region;
688 fr.addr = addrrange_make(base, remain);
689 flatview_insert(view, i, &fr);
693 void flatview_for_each_range(FlatView *fv, flatview_cb cb , void *opaque)
695 FlatRange *fr;
697 assert(fv);
698 assert(cb);
700 FOR_EACH_FLAT_RANGE(fr, fv) {
701 if (cb(fr->addr.start, fr->addr.size, fr->mr,
702 fr->offset_in_region, opaque)) {
703 break;
708 static MemoryRegion *memory_region_get_flatview_root(MemoryRegion *mr)
710 while (mr->enabled) {
711 if (mr->alias) {
712 if (!mr->alias_offset && int128_ge(mr->size, mr->alias->size)) {
713 /* The alias is included in its entirety. Use it as
714 * the "real" root, so that we can share more FlatViews.
716 mr = mr->alias;
717 continue;
719 } else if (!mr->terminates) {
720 unsigned int found = 0;
721 MemoryRegion *child, *next = NULL;
722 QTAILQ_FOREACH(child, &mr->subregions, subregions_link) {
723 if (child->enabled) {
724 if (++found > 1) {
725 next = NULL;
726 break;
728 if (!child->addr && int128_ge(mr->size, child->size)) {
729 /* A child is included in its entirety. If it's the only
730 * enabled one, use it in the hope of finding an alias down the
731 * way. This will also let us share FlatViews.
733 next = child;
737 if (found == 0) {
738 return NULL;
740 if (next) {
741 mr = next;
742 continue;
746 return mr;
749 return NULL;
752 /* Render a memory topology into a list of disjoint absolute ranges. */
753 static FlatView *generate_memory_topology(MemoryRegion *mr)
755 int i;
756 FlatView *view;
758 view = flatview_new(mr);
760 if (mr) {
761 render_memory_region(view, mr, int128_zero(),
762 addrrange_make(int128_zero(), int128_2_64()),
763 false, false, false);
765 flatview_simplify(view);
767 view->dispatch = address_space_dispatch_new(view);
768 for (i = 0; i < view->nr; i++) {
769 MemoryRegionSection mrs =
770 section_from_flat_range(&view->ranges[i], view);
771 flatview_add_to_dispatch(view, &mrs);
773 address_space_dispatch_compact(view->dispatch);
774 g_hash_table_replace(flat_views, mr, view);
776 return view;
779 static void address_space_add_del_ioeventfds(AddressSpace *as,
780 MemoryRegionIoeventfd *fds_new,
781 unsigned fds_new_nb,
782 MemoryRegionIoeventfd *fds_old,
783 unsigned fds_old_nb)
785 unsigned iold, inew;
786 MemoryRegionIoeventfd *fd;
787 MemoryRegionSection section;
789 /* Generate a symmetric difference of the old and new fd sets, adding
790 * and deleting as necessary.
793 iold = inew = 0;
794 while (iold < fds_old_nb || inew < fds_new_nb) {
795 if (iold < fds_old_nb
796 && (inew == fds_new_nb
797 || memory_region_ioeventfd_before(&fds_old[iold],
798 &fds_new[inew]))) {
799 fd = &fds_old[iold];
800 section = (MemoryRegionSection) {
801 .fv = address_space_to_flatview(as),
802 .offset_within_address_space = int128_get64(fd->addr.start),
803 .size = fd->addr.size,
805 MEMORY_LISTENER_CALL(as, eventfd_del, Forward, &section,
806 fd->match_data, fd->data, fd->e);
807 ++iold;
808 } else if (inew < fds_new_nb
809 && (iold == fds_old_nb
810 || memory_region_ioeventfd_before(&fds_new[inew],
811 &fds_old[iold]))) {
812 fd = &fds_new[inew];
813 section = (MemoryRegionSection) {
814 .fv = address_space_to_flatview(as),
815 .offset_within_address_space = int128_get64(fd->addr.start),
816 .size = fd->addr.size,
818 MEMORY_LISTENER_CALL(as, eventfd_add, Reverse, &section,
819 fd->match_data, fd->data, fd->e);
820 ++inew;
821 } else {
822 ++iold;
823 ++inew;
828 FlatView *address_space_get_flatview(AddressSpace *as)
830 FlatView *view;
832 RCU_READ_LOCK_GUARD();
833 do {
834 view = address_space_to_flatview(as);
835 /* If somebody has replaced as->current_map concurrently,
836 * flatview_ref returns false.
838 } while (!flatview_ref(view));
839 return view;
842 static void address_space_update_ioeventfds(AddressSpace *as)
844 FlatView *view;
845 FlatRange *fr;
846 unsigned ioeventfd_nb = 0;
847 unsigned ioeventfd_max;
848 MemoryRegionIoeventfd *ioeventfds;
849 AddrRange tmp;
850 unsigned i;
852 if (!as->ioeventfd_notifiers) {
853 return;
857 * It is likely that the number of ioeventfds hasn't changed much, so use
858 * the previous size as the starting value, with some headroom to avoid
859 * gratuitous reallocations.
861 ioeventfd_max = QEMU_ALIGN_UP(as->ioeventfd_nb, 4);
862 ioeventfds = g_new(MemoryRegionIoeventfd, ioeventfd_max);
864 view = address_space_get_flatview(as);
865 FOR_EACH_FLAT_RANGE(fr, view) {
866 for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
867 tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
868 int128_sub(fr->addr.start,
869 int128_make64(fr->offset_in_region)));
870 if (addrrange_intersects(fr->addr, tmp)) {
871 ++ioeventfd_nb;
872 if (ioeventfd_nb > ioeventfd_max) {
873 ioeventfd_max = MAX(ioeventfd_max * 2, 4);
874 ioeventfds = g_realloc(ioeventfds,
875 ioeventfd_max * sizeof(*ioeventfds));
877 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
878 ioeventfds[ioeventfd_nb-1].addr = tmp;
883 address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
884 as->ioeventfds, as->ioeventfd_nb);
886 g_free(as->ioeventfds);
887 as->ioeventfds = ioeventfds;
888 as->ioeventfd_nb = ioeventfd_nb;
889 flatview_unref(view);
893 * Notify the memory listeners about the coalesced IO change events of
894 * range `cmr'. Only the part that has intersection of the specified
895 * FlatRange will be sent.
897 static void flat_range_coalesced_io_notify(FlatRange *fr, AddressSpace *as,
898 CoalescedMemoryRange *cmr, bool add)
900 AddrRange tmp;
902 tmp = addrrange_shift(cmr->addr,
903 int128_sub(fr->addr.start,
904 int128_make64(fr->offset_in_region)));
905 if (!addrrange_intersects(tmp, fr->addr)) {
906 return;
908 tmp = addrrange_intersection(tmp, fr->addr);
910 if (add) {
911 MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, coalesced_io_add,
912 int128_get64(tmp.start),
913 int128_get64(tmp.size));
914 } else {
915 MEMORY_LISTENER_UPDATE_REGION(fr, as, Reverse, coalesced_io_del,
916 int128_get64(tmp.start),
917 int128_get64(tmp.size));
921 static void flat_range_coalesced_io_del(FlatRange *fr, AddressSpace *as)
923 CoalescedMemoryRange *cmr;
925 QTAILQ_FOREACH(cmr, &fr->mr->coalesced, link) {
926 flat_range_coalesced_io_notify(fr, as, cmr, false);
930 static void flat_range_coalesced_io_add(FlatRange *fr, AddressSpace *as)
932 MemoryRegion *mr = fr->mr;
933 CoalescedMemoryRange *cmr;
935 if (QTAILQ_EMPTY(&mr->coalesced)) {
936 return;
939 QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
940 flat_range_coalesced_io_notify(fr, as, cmr, true);
944 static void address_space_update_topology_pass(AddressSpace *as,
945 const FlatView *old_view,
946 const FlatView *new_view,
947 bool adding)
949 unsigned iold, inew;
950 FlatRange *frold, *frnew;
952 /* Generate a symmetric difference of the old and new memory maps.
953 * Kill ranges in the old map, and instantiate ranges in the new map.
955 iold = inew = 0;
956 while (iold < old_view->nr || inew < new_view->nr) {
957 if (iold < old_view->nr) {
958 frold = &old_view->ranges[iold];
959 } else {
960 frold = NULL;
962 if (inew < new_view->nr) {
963 frnew = &new_view->ranges[inew];
964 } else {
965 frnew = NULL;
968 if (frold
969 && (!frnew
970 || int128_lt(frold->addr.start, frnew->addr.start)
971 || (int128_eq(frold->addr.start, frnew->addr.start)
972 && !flatrange_equal(frold, frnew)))) {
973 /* In old but not in new, or in both but attributes changed. */
975 if (!adding) {
976 flat_range_coalesced_io_del(frold, as);
977 MEMORY_LISTENER_UPDATE_REGION(frold, as, Reverse, region_del);
980 ++iold;
981 } else if (frold && frnew && flatrange_equal(frold, frnew)) {
982 /* In both and unchanged (except logging may have changed) */
984 if (adding) {
985 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_nop);
986 if (frnew->dirty_log_mask & ~frold->dirty_log_mask) {
987 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, log_start,
988 frold->dirty_log_mask,
989 frnew->dirty_log_mask);
991 if (frold->dirty_log_mask & ~frnew->dirty_log_mask) {
992 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Reverse, log_stop,
993 frold->dirty_log_mask,
994 frnew->dirty_log_mask);
998 ++iold;
999 ++inew;
1000 } else {
1001 /* In new */
1003 if (adding) {
1004 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_add);
1005 flat_range_coalesced_io_add(frnew, as);
1008 ++inew;
1013 static void flatviews_init(void)
1015 static FlatView *empty_view;
1017 if (flat_views) {
1018 return;
1021 flat_views = g_hash_table_new_full(g_direct_hash, g_direct_equal, NULL,
1022 (GDestroyNotify) flatview_unref);
1023 if (!empty_view) {
1024 empty_view = generate_memory_topology(NULL);
1025 /* We keep it alive forever in the global variable. */
1026 flatview_ref(empty_view);
1027 } else {
1028 g_hash_table_replace(flat_views, NULL, empty_view);
1029 flatview_ref(empty_view);
1033 static void flatviews_reset(void)
1035 AddressSpace *as;
1037 if (flat_views) {
1038 g_hash_table_unref(flat_views);
1039 flat_views = NULL;
1041 flatviews_init();
1043 /* Render unique FVs */
1044 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1045 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1047 if (g_hash_table_lookup(flat_views, physmr)) {
1048 continue;
1051 generate_memory_topology(physmr);
1055 static void address_space_set_flatview(AddressSpace *as)
1057 FlatView *old_view = address_space_to_flatview(as);
1058 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1059 FlatView *new_view = g_hash_table_lookup(flat_views, physmr);
1061 assert(new_view);
1063 if (old_view == new_view) {
1064 return;
1067 if (old_view) {
1068 flatview_ref(old_view);
1071 flatview_ref(new_view);
1073 if (!QTAILQ_EMPTY(&as->listeners)) {
1074 FlatView tmpview = { .nr = 0 }, *old_view2 = old_view;
1076 if (!old_view2) {
1077 old_view2 = &tmpview;
1079 address_space_update_topology_pass(as, old_view2, new_view, false);
1080 address_space_update_topology_pass(as, old_view2, new_view, true);
1083 /* Writes are protected by the BQL. */
1084 qatomic_rcu_set(&as->current_map, new_view);
1085 if (old_view) {
1086 flatview_unref(old_view);
1089 /* Note that all the old MemoryRegions are still alive up to this
1090 * point. This relieves most MemoryListeners from the need to
1091 * ref/unref the MemoryRegions they get---unless they use them
1092 * outside the iothread mutex, in which case precise reference
1093 * counting is necessary.
1095 if (old_view) {
1096 flatview_unref(old_view);
1100 static void address_space_update_topology(AddressSpace *as)
1102 MemoryRegion *physmr = memory_region_get_flatview_root(as->root);
1104 flatviews_init();
1105 if (!g_hash_table_lookup(flat_views, physmr)) {
1106 generate_memory_topology(physmr);
1108 address_space_set_flatview(as);
1111 void memory_region_transaction_begin(void)
1113 qemu_flush_coalesced_mmio_buffer();
1114 ++memory_region_transaction_depth;
1117 void memory_region_transaction_commit(void)
1119 AddressSpace *as;
1121 assert(memory_region_transaction_depth);
1122 assert(bql_locked());
1124 --memory_region_transaction_depth;
1125 if (!memory_region_transaction_depth) {
1126 if (memory_region_update_pending) {
1127 flatviews_reset();
1129 MEMORY_LISTENER_CALL_GLOBAL(begin, Forward);
1131 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1132 address_space_set_flatview(as);
1133 address_space_update_ioeventfds(as);
1135 memory_region_update_pending = false;
1136 ioeventfd_update_pending = false;
1137 MEMORY_LISTENER_CALL_GLOBAL(commit, Forward);
1138 } else if (ioeventfd_update_pending) {
1139 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1140 address_space_update_ioeventfds(as);
1142 ioeventfd_update_pending = false;
1147 static void memory_region_destructor_none(MemoryRegion *mr)
1151 static void memory_region_destructor_ram(MemoryRegion *mr)
1153 qemu_ram_free(mr->ram_block);
1156 static bool memory_region_need_escape(char c)
1158 return c == '/' || c == '[' || c == '\\' || c == ']';
1161 static char *memory_region_escape_name(const char *name)
1163 const char *p;
1164 char *escaped, *q;
1165 uint8_t c;
1166 size_t bytes = 0;
1168 for (p = name; *p; p++) {
1169 bytes += memory_region_need_escape(*p) ? 4 : 1;
1171 if (bytes == p - name) {
1172 return g_memdup(name, bytes + 1);
1175 escaped = g_malloc(bytes + 1);
1176 for (p = name, q = escaped; *p; p++) {
1177 c = *p;
1178 if (unlikely(memory_region_need_escape(c))) {
1179 *q++ = '\\';
1180 *q++ = 'x';
1181 *q++ = "0123456789abcdef"[c >> 4];
1182 c = "0123456789abcdef"[c & 15];
1184 *q++ = c;
1186 *q = 0;
1187 return escaped;
1190 static void memory_region_do_init(MemoryRegion *mr,
1191 Object *owner,
1192 const char *name,
1193 uint64_t size)
1195 mr->size = int128_make64(size);
1196 if (size == UINT64_MAX) {
1197 mr->size = int128_2_64();
1199 mr->name = g_strdup(name);
1200 mr->owner = owner;
1201 mr->dev = (DeviceState *) object_dynamic_cast(mr->owner, TYPE_DEVICE);
1202 mr->ram_block = NULL;
1204 if (name) {
1205 char *escaped_name = memory_region_escape_name(name);
1206 char *name_array = g_strdup_printf("%s[*]", escaped_name);
1208 if (!owner) {
1209 owner = container_get(qdev_get_machine(), "/unattached");
1212 object_property_add_child(owner, name_array, OBJECT(mr));
1213 object_unref(OBJECT(mr));
1214 g_free(name_array);
1215 g_free(escaped_name);
1219 void memory_region_init(MemoryRegion *mr,
1220 Object *owner,
1221 const char *name,
1222 uint64_t size)
1224 object_initialize(mr, sizeof(*mr), TYPE_MEMORY_REGION);
1225 memory_region_do_init(mr, owner, name, size);
1228 static void memory_region_get_container(Object *obj, Visitor *v,
1229 const char *name, void *opaque,
1230 Error **errp)
1232 MemoryRegion *mr = MEMORY_REGION(obj);
1233 char *path = (char *)"";
1235 if (mr->container) {
1236 path = object_get_canonical_path(OBJECT(mr->container));
1238 visit_type_str(v, name, &path, errp);
1239 if (mr->container) {
1240 g_free(path);
1244 static Object *memory_region_resolve_container(Object *obj, void *opaque,
1245 const char *part)
1247 MemoryRegion *mr = MEMORY_REGION(obj);
1249 return OBJECT(mr->container);
1252 static void memory_region_get_priority(Object *obj, Visitor *v,
1253 const char *name, void *opaque,
1254 Error **errp)
1256 MemoryRegion *mr = MEMORY_REGION(obj);
1257 int32_t value = mr->priority;
1259 visit_type_int32(v, name, &value, errp);
1262 static void memory_region_get_size(Object *obj, Visitor *v, const char *name,
1263 void *opaque, Error **errp)
1265 MemoryRegion *mr = MEMORY_REGION(obj);
1266 uint64_t value = memory_region_size(mr);
1268 visit_type_uint64(v, name, &value, errp);
1271 static void memory_region_initfn(Object *obj)
1273 MemoryRegion *mr = MEMORY_REGION(obj);
1274 ObjectProperty *op;
1276 mr->ops = &unassigned_mem_ops;
1277 mr->enabled = true;
1278 mr->romd_mode = true;
1279 mr->destructor = memory_region_destructor_none;
1280 QTAILQ_INIT(&mr->subregions);
1281 QTAILQ_INIT(&mr->coalesced);
1283 op = object_property_add(OBJECT(mr), "container",
1284 "link<" TYPE_MEMORY_REGION ">",
1285 memory_region_get_container,
1286 NULL, /* memory_region_set_container */
1287 NULL, NULL);
1288 op->resolve = memory_region_resolve_container;
1290 object_property_add_uint64_ptr(OBJECT(mr), "addr",
1291 &mr->addr, OBJ_PROP_FLAG_READ);
1292 object_property_add(OBJECT(mr), "priority", "uint32",
1293 memory_region_get_priority,
1294 NULL, /* memory_region_set_priority */
1295 NULL, NULL);
1296 object_property_add(OBJECT(mr), "size", "uint64",
1297 memory_region_get_size,
1298 NULL, /* memory_region_set_size, */
1299 NULL, NULL);
1302 static void iommu_memory_region_initfn(Object *obj)
1304 MemoryRegion *mr = MEMORY_REGION(obj);
1306 mr->is_iommu = true;
1309 static uint64_t unassigned_mem_read(void *opaque, hwaddr addr,
1310 unsigned size)
1312 #ifdef DEBUG_UNASSIGNED
1313 printf("Unassigned mem read " HWADDR_FMT_plx "\n", addr);
1314 #endif
1315 return 0;
1318 static void unassigned_mem_write(void *opaque, hwaddr addr,
1319 uint64_t val, unsigned size)
1321 #ifdef DEBUG_UNASSIGNED
1322 printf("Unassigned mem write " HWADDR_FMT_plx " = 0x%"PRIx64"\n", addr, val);
1323 #endif
1326 static bool unassigned_mem_accepts(void *opaque, hwaddr addr,
1327 unsigned size, bool is_write,
1328 MemTxAttrs attrs)
1330 return false;
1333 const MemoryRegionOps unassigned_mem_ops = {
1334 .valid.accepts = unassigned_mem_accepts,
1335 .endianness = DEVICE_NATIVE_ENDIAN,
1338 static uint64_t memory_region_ram_device_read(void *opaque,
1339 hwaddr addr, unsigned size)
1341 MemoryRegion *mr = opaque;
1342 uint64_t data = ldn_he_p(mr->ram_block->host + addr, size);
1344 trace_memory_region_ram_device_read(get_cpu_index(), mr, addr, data, size);
1346 return data;
1349 static void memory_region_ram_device_write(void *opaque, hwaddr addr,
1350 uint64_t data, unsigned size)
1352 MemoryRegion *mr = opaque;
1354 trace_memory_region_ram_device_write(get_cpu_index(), mr, addr, data, size);
1356 stn_he_p(mr->ram_block->host + addr, size, data);
1359 static const MemoryRegionOps ram_device_mem_ops = {
1360 .read = memory_region_ram_device_read,
1361 .write = memory_region_ram_device_write,
1362 .endianness = DEVICE_HOST_ENDIAN,
1363 .valid = {
1364 .min_access_size = 1,
1365 .max_access_size = 8,
1366 .unaligned = true,
1368 .impl = {
1369 .min_access_size = 1,
1370 .max_access_size = 8,
1371 .unaligned = true,
1375 bool memory_region_access_valid(MemoryRegion *mr,
1376 hwaddr addr,
1377 unsigned size,
1378 bool is_write,
1379 MemTxAttrs attrs)
1381 if (mr->ops->valid.accepts
1382 && !mr->ops->valid.accepts(mr->opaque, addr, size, is_write, attrs)) {
1383 qemu_log_mask(LOG_GUEST_ERROR, "Invalid %s at addr 0x%" HWADDR_PRIX
1384 ", size %u, region '%s', reason: rejected\n",
1385 is_write ? "write" : "read",
1386 addr, size, memory_region_name(mr));
1387 return false;
1390 if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1391 qemu_log_mask(LOG_GUEST_ERROR, "Invalid %s at addr 0x%" HWADDR_PRIX
1392 ", size %u, region '%s', reason: unaligned\n",
1393 is_write ? "write" : "read",
1394 addr, size, memory_region_name(mr));
1395 return false;
1398 /* Treat zero as compatibility all valid */
1399 if (!mr->ops->valid.max_access_size) {
1400 return true;
1403 if (size > mr->ops->valid.max_access_size
1404 || size < mr->ops->valid.min_access_size) {
1405 qemu_log_mask(LOG_GUEST_ERROR, "Invalid %s at addr 0x%" HWADDR_PRIX
1406 ", size %u, region '%s', reason: invalid size "
1407 "(min:%u max:%u)\n",
1408 is_write ? "write" : "read",
1409 addr, size, memory_region_name(mr),
1410 mr->ops->valid.min_access_size,
1411 mr->ops->valid.max_access_size);
1412 return false;
1414 return true;
1417 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1418 hwaddr addr,
1419 uint64_t *pval,
1420 unsigned size,
1421 MemTxAttrs attrs)
1423 *pval = 0;
1425 if (mr->ops->read) {
1426 return access_with_adjusted_size(addr, pval, size,
1427 mr->ops->impl.min_access_size,
1428 mr->ops->impl.max_access_size,
1429 memory_region_read_accessor,
1430 mr, attrs);
1431 } else {
1432 return access_with_adjusted_size(addr, pval, size,
1433 mr->ops->impl.min_access_size,
1434 mr->ops->impl.max_access_size,
1435 memory_region_read_with_attrs_accessor,
1436 mr, attrs);
1440 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1441 hwaddr addr,
1442 uint64_t *pval,
1443 MemOp op,
1444 MemTxAttrs attrs)
1446 unsigned size = memop_size(op);
1447 MemTxResult r;
1449 if (mr->alias) {
1450 return memory_region_dispatch_read(mr->alias,
1451 mr->alias_offset + addr,
1452 pval, op, attrs);
1454 if (!memory_region_access_valid(mr, addr, size, false, attrs)) {
1455 *pval = unassigned_mem_read(mr, addr, size);
1456 return MEMTX_DECODE_ERROR;
1459 r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1460 adjust_endianness(mr, pval, op);
1461 return r;
1464 /* Return true if an eventfd was signalled */
1465 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1466 hwaddr addr,
1467 uint64_t data,
1468 unsigned size,
1469 MemTxAttrs attrs)
1471 MemoryRegionIoeventfd ioeventfd = {
1472 .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1473 .data = data,
1475 unsigned i;
1477 for (i = 0; i < mr->ioeventfd_nb; i++) {
1478 ioeventfd.match_data = mr->ioeventfds[i].match_data;
1479 ioeventfd.e = mr->ioeventfds[i].e;
1481 if (memory_region_ioeventfd_equal(&ioeventfd, &mr->ioeventfds[i])) {
1482 event_notifier_set(ioeventfd.e);
1483 return true;
1487 return false;
1490 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1491 hwaddr addr,
1492 uint64_t data,
1493 MemOp op,
1494 MemTxAttrs attrs)
1496 unsigned size = memop_size(op);
1498 if (mr->alias) {
1499 return memory_region_dispatch_write(mr->alias,
1500 mr->alias_offset + addr,
1501 data, op, attrs);
1503 if (!memory_region_access_valid(mr, addr, size, true, attrs)) {
1504 unassigned_mem_write(mr, addr, data, size);
1505 return MEMTX_DECODE_ERROR;
1508 adjust_endianness(mr, &data, op);
1511 * FIXME: it's not clear why under KVM the write would be processed
1512 * directly, instead of going through eventfd. This probably should
1513 * test "tcg_enabled() || qtest_enabled()", or should just go away.
1515 if (!kvm_enabled() &&
1516 memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1517 return MEMTX_OK;
1520 if (mr->ops->write) {
1521 return access_with_adjusted_size(addr, &data, size,
1522 mr->ops->impl.min_access_size,
1523 mr->ops->impl.max_access_size,
1524 memory_region_write_accessor, mr,
1525 attrs);
1526 } else {
1527 return
1528 access_with_adjusted_size(addr, &data, size,
1529 mr->ops->impl.min_access_size,
1530 mr->ops->impl.max_access_size,
1531 memory_region_write_with_attrs_accessor,
1532 mr, attrs);
1536 void memory_region_init_io(MemoryRegion *mr,
1537 Object *owner,
1538 const MemoryRegionOps *ops,
1539 void *opaque,
1540 const char *name,
1541 uint64_t size)
1543 memory_region_init(mr, owner, name, size);
1544 mr->ops = ops ? ops : &unassigned_mem_ops;
1545 mr->opaque = opaque;
1546 mr->terminates = true;
1549 bool memory_region_init_ram_nomigrate(MemoryRegion *mr,
1550 Object *owner,
1551 const char *name,
1552 uint64_t size,
1553 Error **errp)
1555 return memory_region_init_ram_flags_nomigrate(mr, owner, name,
1556 size, 0, errp);
1559 bool memory_region_init_ram_flags_nomigrate(MemoryRegion *mr,
1560 Object *owner,
1561 const char *name,
1562 uint64_t size,
1563 uint32_t ram_flags,
1564 Error **errp)
1566 Error *err = NULL;
1567 memory_region_init(mr, owner, name, size);
1568 mr->ram = true;
1569 mr->terminates = true;
1570 mr->destructor = memory_region_destructor_ram;
1571 mr->ram_block = qemu_ram_alloc(size, ram_flags, mr, &err);
1572 if (err) {
1573 mr->size = int128_zero();
1574 object_unparent(OBJECT(mr));
1575 error_propagate(errp, err);
1576 return false;
1578 return true;
1581 bool memory_region_init_resizeable_ram(MemoryRegion *mr,
1582 Object *owner,
1583 const char *name,
1584 uint64_t size,
1585 uint64_t max_size,
1586 void (*resized)(const char*,
1587 uint64_t length,
1588 void *host),
1589 Error **errp)
1591 Error *err = NULL;
1592 memory_region_init(mr, owner, name, size);
1593 mr->ram = true;
1594 mr->terminates = true;
1595 mr->destructor = memory_region_destructor_ram;
1596 mr->ram_block = qemu_ram_alloc_resizeable(size, max_size, resized,
1597 mr, &err);
1598 if (err) {
1599 mr->size = int128_zero();
1600 object_unparent(OBJECT(mr));
1601 error_propagate(errp, err);
1602 return false;
1604 return true;
1607 #ifdef CONFIG_POSIX
1608 bool memory_region_init_ram_from_file(MemoryRegion *mr,
1609 Object *owner,
1610 const char *name,
1611 uint64_t size,
1612 uint64_t align,
1613 uint32_t ram_flags,
1614 const char *path,
1615 ram_addr_t offset,
1616 Error **errp)
1618 Error *err = NULL;
1619 memory_region_init(mr, owner, name, size);
1620 mr->ram = true;
1621 mr->readonly = !!(ram_flags & RAM_READONLY);
1622 mr->terminates = true;
1623 mr->destructor = memory_region_destructor_ram;
1624 mr->align = align;
1625 mr->ram_block = qemu_ram_alloc_from_file(size, mr, ram_flags, path,
1626 offset, &err);
1627 if (err) {
1628 mr->size = int128_zero();
1629 object_unparent(OBJECT(mr));
1630 error_propagate(errp, err);
1631 return false;
1633 return true;
1636 bool memory_region_init_ram_from_fd(MemoryRegion *mr,
1637 Object *owner,
1638 const char *name,
1639 uint64_t size,
1640 uint32_t ram_flags,
1641 int fd,
1642 ram_addr_t offset,
1643 Error **errp)
1645 Error *err = NULL;
1646 memory_region_init(mr, owner, name, size);
1647 mr->ram = true;
1648 mr->readonly = !!(ram_flags & RAM_READONLY);
1649 mr->terminates = true;
1650 mr->destructor = memory_region_destructor_ram;
1651 mr->ram_block = qemu_ram_alloc_from_fd(size, mr, ram_flags, fd, offset,
1652 &err);
1653 if (err) {
1654 mr->size = int128_zero();
1655 object_unparent(OBJECT(mr));
1656 error_propagate(errp, err);
1657 return false;
1659 return true;
1661 #endif
1663 void memory_region_init_ram_ptr(MemoryRegion *mr,
1664 Object *owner,
1665 const char *name,
1666 uint64_t size,
1667 void *ptr)
1669 memory_region_init(mr, owner, name, size);
1670 mr->ram = true;
1671 mr->terminates = true;
1672 mr->destructor = memory_region_destructor_ram;
1674 /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL. */
1675 assert(ptr != NULL);
1676 mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_abort);
1679 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
1680 Object *owner,
1681 const char *name,
1682 uint64_t size,
1683 void *ptr)
1685 memory_region_init(mr, owner, name, size);
1686 mr->ram = true;
1687 mr->terminates = true;
1688 mr->ram_device = true;
1689 mr->ops = &ram_device_mem_ops;
1690 mr->opaque = mr;
1691 mr->destructor = memory_region_destructor_ram;
1693 /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL. */
1694 assert(ptr != NULL);
1695 mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_abort);
1698 void memory_region_init_alias(MemoryRegion *mr,
1699 Object *owner,
1700 const char *name,
1701 MemoryRegion *orig,
1702 hwaddr offset,
1703 uint64_t size)
1705 memory_region_init(mr, owner, name, size);
1706 mr->alias = orig;
1707 mr->alias_offset = offset;
1710 bool memory_region_init_rom_nomigrate(MemoryRegion *mr,
1711 Object *owner,
1712 const char *name,
1713 uint64_t size,
1714 Error **errp)
1716 if (!memory_region_init_ram_flags_nomigrate(mr, owner, name,
1717 size, 0, errp)) {
1718 return false;
1720 mr->readonly = true;
1722 return true;
1725 bool memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1726 Object *owner,
1727 const MemoryRegionOps *ops,
1728 void *opaque,
1729 const char *name,
1730 uint64_t size,
1731 Error **errp)
1733 Error *err = NULL;
1734 assert(ops);
1735 memory_region_init(mr, owner, name, size);
1736 mr->ops = ops;
1737 mr->opaque = opaque;
1738 mr->terminates = true;
1739 mr->rom_device = true;
1740 mr->destructor = memory_region_destructor_ram;
1741 mr->ram_block = qemu_ram_alloc(size, 0, mr, &err);
1742 if (err) {
1743 mr->size = int128_zero();
1744 object_unparent(OBJECT(mr));
1745 error_propagate(errp, err);
1746 return false;
1748 return true;
1751 void memory_region_init_iommu(void *_iommu_mr,
1752 size_t instance_size,
1753 const char *mrtypename,
1754 Object *owner,
1755 const char *name,
1756 uint64_t size)
1758 struct IOMMUMemoryRegion *iommu_mr;
1759 struct MemoryRegion *mr;
1761 object_initialize(_iommu_mr, instance_size, mrtypename);
1762 mr = MEMORY_REGION(_iommu_mr);
1763 memory_region_do_init(mr, owner, name, size);
1764 iommu_mr = IOMMU_MEMORY_REGION(mr);
1765 mr->terminates = true; /* then re-forwards */
1766 QLIST_INIT(&iommu_mr->iommu_notify);
1767 iommu_mr->iommu_notify_flags = IOMMU_NOTIFIER_NONE;
1770 static void memory_region_finalize(Object *obj)
1772 MemoryRegion *mr = MEMORY_REGION(obj);
1774 assert(!mr->container);
1776 /* We know the region is not visible in any address space (it
1777 * does not have a container and cannot be a root either because
1778 * it has no references, so we can blindly clear mr->enabled.
1779 * memory_region_set_enabled instead could trigger a transaction
1780 * and cause an infinite loop.
1782 mr->enabled = false;
1783 memory_region_transaction_begin();
1784 while (!QTAILQ_EMPTY(&mr->subregions)) {
1785 MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1786 memory_region_del_subregion(mr, subregion);
1788 memory_region_transaction_commit();
1790 mr->destructor(mr);
1791 memory_region_clear_coalescing(mr);
1792 g_free((char *)mr->name);
1793 g_free(mr->ioeventfds);
1796 Object *memory_region_owner(MemoryRegion *mr)
1798 Object *obj = OBJECT(mr);
1799 return obj->parent;
1802 void memory_region_ref(MemoryRegion *mr)
1804 /* MMIO callbacks most likely will access data that belongs
1805 * to the owner, hence the need to ref/unref the owner whenever
1806 * the memory region is in use.
1808 * The memory region is a child of its owner. As long as the
1809 * owner doesn't call unparent itself on the memory region,
1810 * ref-ing the owner will also keep the memory region alive.
1811 * Memory regions without an owner are supposed to never go away;
1812 * we do not ref/unref them because it slows down DMA sensibly.
1814 if (mr && mr->owner) {
1815 object_ref(mr->owner);
1819 void memory_region_unref(MemoryRegion *mr)
1821 if (mr && mr->owner) {
1822 object_unref(mr->owner);
1826 uint64_t memory_region_size(MemoryRegion *mr)
1828 if (int128_eq(mr->size, int128_2_64())) {
1829 return UINT64_MAX;
1831 return int128_get64(mr->size);
1834 const char *memory_region_name(const MemoryRegion *mr)
1836 if (!mr->name) {
1837 ((MemoryRegion *)mr)->name =
1838 g_strdup(object_get_canonical_path_component(OBJECT(mr)));
1840 return mr->name;
1843 bool memory_region_is_ram_device(MemoryRegion *mr)
1845 return mr->ram_device;
1848 bool memory_region_is_protected(MemoryRegion *mr)
1850 return mr->ram && (mr->ram_block->flags & RAM_PROTECTED);
1853 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1855 uint8_t mask = mr->dirty_log_mask;
1856 RAMBlock *rb = mr->ram_block;
1858 if (global_dirty_tracking && ((rb && qemu_ram_is_migratable(rb)) ||
1859 memory_region_is_iommu(mr))) {
1860 mask |= (1 << DIRTY_MEMORY_MIGRATION);
1863 if (tcg_enabled() && rb) {
1864 /* TCG only cares about dirty memory logging for RAM, not IOMMU. */
1865 mask |= (1 << DIRTY_MEMORY_CODE);
1867 return mask;
1870 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1872 return memory_region_get_dirty_log_mask(mr) & (1 << client);
1875 static int memory_region_update_iommu_notify_flags(IOMMUMemoryRegion *iommu_mr,
1876 Error **errp)
1878 IOMMUNotifierFlag flags = IOMMU_NOTIFIER_NONE;
1879 IOMMUNotifier *iommu_notifier;
1880 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1881 int ret = 0;
1883 IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
1884 flags |= iommu_notifier->notifier_flags;
1887 if (flags != iommu_mr->iommu_notify_flags && imrc->notify_flag_changed) {
1888 ret = imrc->notify_flag_changed(iommu_mr,
1889 iommu_mr->iommu_notify_flags,
1890 flags, errp);
1893 if (!ret) {
1894 iommu_mr->iommu_notify_flags = flags;
1896 return ret;
1899 int memory_region_iommu_set_page_size_mask(IOMMUMemoryRegion *iommu_mr,
1900 uint64_t page_size_mask,
1901 Error **errp)
1903 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1904 int ret = 0;
1906 if (imrc->iommu_set_page_size_mask) {
1907 ret = imrc->iommu_set_page_size_mask(iommu_mr, page_size_mask, errp);
1909 return ret;
1912 int memory_region_iommu_set_iova_ranges(IOMMUMemoryRegion *iommu_mr,
1913 GList *iova_ranges,
1914 Error **errp)
1916 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1917 int ret = 0;
1919 if (imrc->iommu_set_iova_ranges) {
1920 ret = imrc->iommu_set_iova_ranges(iommu_mr, iova_ranges, errp);
1922 return ret;
1925 int memory_region_register_iommu_notifier(MemoryRegion *mr,
1926 IOMMUNotifier *n, Error **errp)
1928 IOMMUMemoryRegion *iommu_mr;
1929 int ret;
1931 if (mr->alias) {
1932 return memory_region_register_iommu_notifier(mr->alias, n, errp);
1935 /* We need to register for at least one bitfield */
1936 iommu_mr = IOMMU_MEMORY_REGION(mr);
1937 assert(n->notifier_flags != IOMMU_NOTIFIER_NONE);
1938 assert(n->start <= n->end);
1939 assert(n->iommu_idx >= 0 &&
1940 n->iommu_idx < memory_region_iommu_num_indexes(iommu_mr));
1942 QLIST_INSERT_HEAD(&iommu_mr->iommu_notify, n, node);
1943 ret = memory_region_update_iommu_notify_flags(iommu_mr, errp);
1944 if (ret) {
1945 QLIST_REMOVE(n, node);
1947 return ret;
1950 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr)
1952 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1954 if (imrc->get_min_page_size) {
1955 return imrc->get_min_page_size(iommu_mr);
1957 return TARGET_PAGE_SIZE;
1960 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
1962 MemoryRegion *mr = MEMORY_REGION(iommu_mr);
1963 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
1964 hwaddr addr, granularity;
1965 IOMMUTLBEntry iotlb;
1967 /* If the IOMMU has its own replay callback, override */
1968 if (imrc->replay) {
1969 imrc->replay(iommu_mr, n);
1970 return;
1973 granularity = memory_region_iommu_get_min_page_size(iommu_mr);
1975 for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1976 iotlb = imrc->translate(iommu_mr, addr, IOMMU_NONE, n->iommu_idx);
1977 if (iotlb.perm != IOMMU_NONE) {
1978 n->notify(n, &iotlb);
1981 /* if (2^64 - MR size) < granularity, it's possible to get an
1982 * infinite loop here. This should catch such a wraparound */
1983 if ((addr + granularity) < addr) {
1984 break;
1989 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1990 IOMMUNotifier *n)
1992 IOMMUMemoryRegion *iommu_mr;
1994 if (mr->alias) {
1995 memory_region_unregister_iommu_notifier(mr->alias, n);
1996 return;
1998 QLIST_REMOVE(n, node);
1999 iommu_mr = IOMMU_MEMORY_REGION(mr);
2000 memory_region_update_iommu_notify_flags(iommu_mr, NULL);
2003 void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
2004 IOMMUTLBEvent *event)
2006 IOMMUTLBEntry *entry = &event->entry;
2007 hwaddr entry_end = entry->iova + entry->addr_mask;
2008 IOMMUTLBEntry tmp = *entry;
2010 if (event->type == IOMMU_NOTIFIER_UNMAP) {
2011 assert(entry->perm == IOMMU_NONE);
2015 * Skip the notification if the notification does not overlap
2016 * with registered range.
2018 if (notifier->start > entry_end || notifier->end < entry->iova) {
2019 return;
2022 if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
2023 /* Crop (iova, addr_mask) to range */
2024 tmp.iova = MAX(tmp.iova, notifier->start);
2025 tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
2026 } else {
2027 assert(entry->iova >= notifier->start && entry_end <= notifier->end);
2030 if (event->type & notifier->notifier_flags) {
2031 notifier->notify(notifier, &tmp);
2035 void memory_region_unmap_iommu_notifier_range(IOMMUNotifier *notifier)
2037 IOMMUTLBEvent event;
2039 event.type = IOMMU_NOTIFIER_UNMAP;
2040 event.entry.target_as = &address_space_memory;
2041 event.entry.iova = notifier->start;
2042 event.entry.perm = IOMMU_NONE;
2043 event.entry.addr_mask = notifier->end - notifier->start;
2045 memory_region_notify_iommu_one(notifier, &event);
2048 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
2049 int iommu_idx,
2050 IOMMUTLBEvent event)
2052 IOMMUNotifier *iommu_notifier;
2054 assert(memory_region_is_iommu(MEMORY_REGION(iommu_mr)));
2056 IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
2057 if (iommu_notifier->iommu_idx == iommu_idx) {
2058 memory_region_notify_iommu_one(iommu_notifier, &event);
2063 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
2064 enum IOMMUMemoryRegionAttr attr,
2065 void *data)
2067 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2069 if (!imrc->get_attr) {
2070 return -EINVAL;
2073 return imrc->get_attr(iommu_mr, attr, data);
2076 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr,
2077 MemTxAttrs attrs)
2079 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2081 if (!imrc->attrs_to_index) {
2082 return 0;
2085 return imrc->attrs_to_index(iommu_mr, attrs);
2088 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr)
2090 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
2092 if (!imrc->num_indexes) {
2093 return 1;
2096 return imrc->num_indexes(iommu_mr);
2099 RamDiscardManager *memory_region_get_ram_discard_manager(MemoryRegion *mr)
2101 if (!memory_region_is_ram(mr)) {
2102 return NULL;
2104 return mr->rdm;
2107 void memory_region_set_ram_discard_manager(MemoryRegion *mr,
2108 RamDiscardManager *rdm)
2110 g_assert(memory_region_is_ram(mr));
2111 g_assert(!rdm || !mr->rdm);
2112 mr->rdm = rdm;
2115 uint64_t ram_discard_manager_get_min_granularity(const RamDiscardManager *rdm,
2116 const MemoryRegion *mr)
2118 RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2120 g_assert(rdmc->get_min_granularity);
2121 return rdmc->get_min_granularity(rdm, mr);
2124 bool ram_discard_manager_is_populated(const RamDiscardManager *rdm,
2125 const MemoryRegionSection *section)
2127 RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2129 g_assert(rdmc->is_populated);
2130 return rdmc->is_populated(rdm, section);
2133 int ram_discard_manager_replay_populated(const RamDiscardManager *rdm,
2134 MemoryRegionSection *section,
2135 ReplayRamPopulate replay_fn,
2136 void *opaque)
2138 RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2140 g_assert(rdmc->replay_populated);
2141 return rdmc->replay_populated(rdm, section, replay_fn, opaque);
2144 void ram_discard_manager_replay_discarded(const RamDiscardManager *rdm,
2145 MemoryRegionSection *section,
2146 ReplayRamDiscard replay_fn,
2147 void *opaque)
2149 RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2151 g_assert(rdmc->replay_discarded);
2152 rdmc->replay_discarded(rdm, section, replay_fn, opaque);
2155 void ram_discard_manager_register_listener(RamDiscardManager *rdm,
2156 RamDiscardListener *rdl,
2157 MemoryRegionSection *section)
2159 RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2161 g_assert(rdmc->register_listener);
2162 rdmc->register_listener(rdm, rdl, section);
2165 void ram_discard_manager_unregister_listener(RamDiscardManager *rdm,
2166 RamDiscardListener *rdl)
2168 RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_GET_CLASS(rdm);
2170 g_assert(rdmc->unregister_listener);
2171 rdmc->unregister_listener(rdm, rdl);
2174 /* Called with rcu_read_lock held. */
2175 bool memory_get_xlat_addr(IOMMUTLBEntry *iotlb, void **vaddr,
2176 ram_addr_t *ram_addr, bool *read_only,
2177 bool *mr_has_discard_manager)
2179 MemoryRegion *mr;
2180 hwaddr xlat;
2181 hwaddr len = iotlb->addr_mask + 1;
2182 bool writable = iotlb->perm & IOMMU_WO;
2184 if (mr_has_discard_manager) {
2185 *mr_has_discard_manager = false;
2188 * The IOMMU TLB entry we have just covers translation through
2189 * this IOMMU to its immediate target. We need to translate
2190 * it the rest of the way through to memory.
2192 mr = address_space_translate(&address_space_memory, iotlb->translated_addr,
2193 &xlat, &len, writable, MEMTXATTRS_UNSPECIFIED);
2194 if (!memory_region_is_ram(mr)) {
2195 error_report("iommu map to non memory area %" HWADDR_PRIx "", xlat);
2196 return false;
2197 } else if (memory_region_has_ram_discard_manager(mr)) {
2198 RamDiscardManager *rdm = memory_region_get_ram_discard_manager(mr);
2199 MemoryRegionSection tmp = {
2200 .mr = mr,
2201 .offset_within_region = xlat,
2202 .size = int128_make64(len),
2204 if (mr_has_discard_manager) {
2205 *mr_has_discard_manager = true;
2208 * Malicious VMs can map memory into the IOMMU, which is expected
2209 * to remain discarded. vfio will pin all pages, populating memory.
2210 * Disallow that. vmstate priorities make sure any RamDiscardManager
2211 * were already restored before IOMMUs are restored.
2213 if (!ram_discard_manager_is_populated(rdm, &tmp)) {
2214 error_report("iommu map to discarded memory (e.g., unplugged via"
2215 " virtio-mem): %" HWADDR_PRIx "",
2216 iotlb->translated_addr);
2217 return false;
2222 * Translation truncates length to the IOMMU page size,
2223 * check that it did not truncate too much.
2225 if (len & iotlb->addr_mask) {
2226 error_report("iommu has granularity incompatible with target AS");
2227 return false;
2230 if (vaddr) {
2231 *vaddr = memory_region_get_ram_ptr(mr) + xlat;
2234 if (ram_addr) {
2235 *ram_addr = memory_region_get_ram_addr(mr) + xlat;
2238 if (read_only) {
2239 *read_only = !writable || mr->readonly;
2242 return true;
2245 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
2247 uint8_t mask = 1 << client;
2248 uint8_t old_logging;
2250 assert(client == DIRTY_MEMORY_VGA);
2251 old_logging = mr->vga_logging_count;
2252 mr->vga_logging_count += log ? 1 : -1;
2253 if (!!old_logging == !!mr->vga_logging_count) {
2254 return;
2257 memory_region_transaction_begin();
2258 mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
2259 memory_region_update_pending |= mr->enabled;
2260 memory_region_transaction_commit();
2263 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
2264 hwaddr size)
2266 assert(mr->ram_block);
2267 cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr,
2268 size,
2269 memory_region_get_dirty_log_mask(mr));
2273 * If memory region `mr' is NULL, do global sync. Otherwise, sync
2274 * dirty bitmap for the specified memory region.
2276 static void memory_region_sync_dirty_bitmap(MemoryRegion *mr, bool last_stage)
2278 MemoryListener *listener;
2279 AddressSpace *as;
2280 FlatView *view;
2281 FlatRange *fr;
2283 /* If the same address space has multiple log_sync listeners, we
2284 * visit that address space's FlatView multiple times. But because
2285 * log_sync listeners are rare, it's still cheaper than walking each
2286 * address space once.
2288 QTAILQ_FOREACH(listener, &memory_listeners, link) {
2289 if (listener->log_sync) {
2290 as = listener->address_space;
2291 view = address_space_get_flatview(as);
2292 FOR_EACH_FLAT_RANGE(fr, view) {
2293 if (fr->dirty_log_mask && (!mr || fr->mr == mr)) {
2294 MemoryRegionSection mrs = section_from_flat_range(fr, view);
2295 listener->log_sync(listener, &mrs);
2298 flatview_unref(view);
2299 trace_memory_region_sync_dirty(mr ? mr->name : "(all)", listener->name, 0);
2300 } else if (listener->log_sync_global) {
2302 * No matter whether MR is specified, what we can do here
2303 * is to do a global sync, because we are not capable to
2304 * sync in a finer granularity.
2306 listener->log_sync_global(listener, last_stage);
2307 trace_memory_region_sync_dirty(mr ? mr->name : "(all)", listener->name, 1);
2312 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start,
2313 hwaddr len)
2315 MemoryRegionSection mrs;
2316 MemoryListener *listener;
2317 AddressSpace *as;
2318 FlatView *view;
2319 FlatRange *fr;
2320 hwaddr sec_start, sec_end, sec_size;
2322 QTAILQ_FOREACH(listener, &memory_listeners, link) {
2323 if (!listener->log_clear) {
2324 continue;
2326 as = listener->address_space;
2327 view = address_space_get_flatview(as);
2328 FOR_EACH_FLAT_RANGE(fr, view) {
2329 if (!fr->dirty_log_mask || fr->mr != mr) {
2331 * Clear dirty bitmap operation only applies to those
2332 * regions whose dirty logging is at least enabled
2334 continue;
2337 mrs = section_from_flat_range(fr, view);
2339 sec_start = MAX(mrs.offset_within_region, start);
2340 sec_end = mrs.offset_within_region + int128_get64(mrs.size);
2341 sec_end = MIN(sec_end, start + len);
2343 if (sec_start >= sec_end) {
2345 * If this memory region section has no intersection
2346 * with the requested range, skip.
2348 continue;
2351 /* Valid case; shrink the section if needed */
2352 mrs.offset_within_address_space +=
2353 sec_start - mrs.offset_within_region;
2354 mrs.offset_within_region = sec_start;
2355 sec_size = sec_end - sec_start;
2356 mrs.size = int128_make64(sec_size);
2357 listener->log_clear(listener, &mrs);
2359 flatview_unref(view);
2363 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
2364 hwaddr addr,
2365 hwaddr size,
2366 unsigned client)
2368 DirtyBitmapSnapshot *snapshot;
2369 assert(mr->ram_block);
2370 memory_region_sync_dirty_bitmap(mr, false);
2371 snapshot = cpu_physical_memory_snapshot_and_clear_dirty(mr, addr, size, client);
2372 memory_global_after_dirty_log_sync();
2373 return snapshot;
2376 bool memory_region_snapshot_get_dirty(MemoryRegion *mr, DirtyBitmapSnapshot *snap,
2377 hwaddr addr, hwaddr size)
2379 assert(mr->ram_block);
2380 return cpu_physical_memory_snapshot_get_dirty(snap,
2381 memory_region_get_ram_addr(mr) + addr, size);
2384 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
2386 if (mr->readonly != readonly) {
2387 memory_region_transaction_begin();
2388 mr->readonly = readonly;
2389 memory_region_update_pending |= mr->enabled;
2390 memory_region_transaction_commit();
2394 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile)
2396 if (mr->nonvolatile != nonvolatile) {
2397 memory_region_transaction_begin();
2398 mr->nonvolatile = nonvolatile;
2399 memory_region_update_pending |= mr->enabled;
2400 memory_region_transaction_commit();
2404 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
2406 if (mr->romd_mode != romd_mode) {
2407 memory_region_transaction_begin();
2408 mr->romd_mode = romd_mode;
2409 memory_region_update_pending |= mr->enabled;
2410 memory_region_transaction_commit();
2414 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
2415 hwaddr size, unsigned client)
2417 assert(mr->ram_block);
2418 cpu_physical_memory_test_and_clear_dirty(
2419 memory_region_get_ram_addr(mr) + addr, size, client);
2422 int memory_region_get_fd(MemoryRegion *mr)
2424 RCU_READ_LOCK_GUARD();
2425 while (mr->alias) {
2426 mr = mr->alias;
2428 return mr->ram_block->fd;
2431 void *memory_region_get_ram_ptr(MemoryRegion *mr)
2433 uint64_t offset = 0;
2435 RCU_READ_LOCK_GUARD();
2436 while (mr->alias) {
2437 offset += mr->alias_offset;
2438 mr = mr->alias;
2440 assert(mr->ram_block);
2441 return qemu_map_ram_ptr(mr->ram_block, offset);
2444 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset)
2446 RAMBlock *block;
2448 block = qemu_ram_block_from_host(ptr, false, offset);
2449 if (!block) {
2450 return NULL;
2453 return block->mr;
2456 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr)
2458 return mr->ram_block ? mr->ram_block->offset : RAM_ADDR_INVALID;
2461 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
2463 assert(mr->ram_block);
2465 qemu_ram_resize(mr->ram_block, newsize, errp);
2468 void memory_region_msync(MemoryRegion *mr, hwaddr addr, hwaddr size)
2470 if (mr->ram_block) {
2471 qemu_ram_msync(mr->ram_block, addr, size);
2475 void memory_region_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size)
2478 * Might be extended case needed to cover
2479 * different types of memory regions
2481 if (mr->dirty_log_mask) {
2482 memory_region_msync(mr, addr, size);
2487 * Call proper memory listeners about the change on the newly
2488 * added/removed CoalescedMemoryRange.
2490 static void memory_region_update_coalesced_range(MemoryRegion *mr,
2491 CoalescedMemoryRange *cmr,
2492 bool add)
2494 AddressSpace *as;
2495 FlatView *view;
2496 FlatRange *fr;
2498 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2499 view = address_space_get_flatview(as);
2500 FOR_EACH_FLAT_RANGE(fr, view) {
2501 if (fr->mr == mr) {
2502 flat_range_coalesced_io_notify(fr, as, cmr, add);
2505 flatview_unref(view);
2509 void memory_region_set_coalescing(MemoryRegion *mr)
2511 memory_region_clear_coalescing(mr);
2512 memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
2515 void memory_region_add_coalescing(MemoryRegion *mr,
2516 hwaddr offset,
2517 uint64_t size)
2519 CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
2521 cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
2522 QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
2523 memory_region_update_coalesced_range(mr, cmr, true);
2524 memory_region_set_flush_coalesced(mr);
2527 void memory_region_clear_coalescing(MemoryRegion *mr)
2529 CoalescedMemoryRange *cmr;
2531 if (QTAILQ_EMPTY(&mr->coalesced)) {
2532 return;
2535 qemu_flush_coalesced_mmio_buffer();
2536 mr->flush_coalesced_mmio = false;
2538 while (!QTAILQ_EMPTY(&mr->coalesced)) {
2539 cmr = QTAILQ_FIRST(&mr->coalesced);
2540 QTAILQ_REMOVE(&mr->coalesced, cmr, link);
2541 memory_region_update_coalesced_range(mr, cmr, false);
2542 g_free(cmr);
2546 void memory_region_set_flush_coalesced(MemoryRegion *mr)
2548 mr->flush_coalesced_mmio = true;
2551 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
2553 qemu_flush_coalesced_mmio_buffer();
2554 if (QTAILQ_EMPTY(&mr->coalesced)) {
2555 mr->flush_coalesced_mmio = false;
2559 void memory_region_add_eventfd(MemoryRegion *mr,
2560 hwaddr addr,
2561 unsigned size,
2562 bool match_data,
2563 uint64_t data,
2564 EventNotifier *e)
2566 MemoryRegionIoeventfd mrfd = {
2567 .addr.start = int128_make64(addr),
2568 .addr.size = int128_make64(size),
2569 .match_data = match_data,
2570 .data = data,
2571 .e = e,
2573 unsigned i;
2575 if (size) {
2576 adjust_endianness(mr, &mrfd.data, size_memop(size) | MO_TE);
2578 memory_region_transaction_begin();
2579 for (i = 0; i < mr->ioeventfd_nb; ++i) {
2580 if (memory_region_ioeventfd_before(&mrfd, &mr->ioeventfds[i])) {
2581 break;
2584 ++mr->ioeventfd_nb;
2585 mr->ioeventfds = g_realloc(mr->ioeventfds,
2586 sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
2587 memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
2588 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
2589 mr->ioeventfds[i] = mrfd;
2590 ioeventfd_update_pending |= mr->enabled;
2591 memory_region_transaction_commit();
2594 void memory_region_del_eventfd(MemoryRegion *mr,
2595 hwaddr addr,
2596 unsigned size,
2597 bool match_data,
2598 uint64_t data,
2599 EventNotifier *e)
2601 MemoryRegionIoeventfd mrfd = {
2602 .addr.start = int128_make64(addr),
2603 .addr.size = int128_make64(size),
2604 .match_data = match_data,
2605 .data = data,
2606 .e = e,
2608 unsigned i;
2610 if (size) {
2611 adjust_endianness(mr, &mrfd.data, size_memop(size) | MO_TE);
2613 memory_region_transaction_begin();
2614 for (i = 0; i < mr->ioeventfd_nb; ++i) {
2615 if (memory_region_ioeventfd_equal(&mrfd, &mr->ioeventfds[i])) {
2616 break;
2619 assert(i != mr->ioeventfd_nb);
2620 memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
2621 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
2622 --mr->ioeventfd_nb;
2623 mr->ioeventfds = g_realloc(mr->ioeventfds,
2624 sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
2625 ioeventfd_update_pending |= mr->enabled;
2626 memory_region_transaction_commit();
2629 static void memory_region_update_container_subregions(MemoryRegion *subregion)
2631 MemoryRegion *mr = subregion->container;
2632 MemoryRegion *other;
2634 memory_region_transaction_begin();
2636 memory_region_ref(subregion);
2637 QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
2638 if (subregion->priority >= other->priority) {
2639 QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
2640 goto done;
2643 QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
2644 done:
2645 memory_region_update_pending |= mr->enabled && subregion->enabled;
2646 memory_region_transaction_commit();
2649 static void memory_region_add_subregion_common(MemoryRegion *mr,
2650 hwaddr offset,
2651 MemoryRegion *subregion)
2653 MemoryRegion *alias;
2655 assert(!subregion->container);
2656 subregion->container = mr;
2657 for (alias = subregion->alias; alias; alias = alias->alias) {
2658 alias->mapped_via_alias++;
2660 subregion->addr = offset;
2661 memory_region_update_container_subregions(subregion);
2664 void memory_region_add_subregion(MemoryRegion *mr,
2665 hwaddr offset,
2666 MemoryRegion *subregion)
2668 subregion->priority = 0;
2669 memory_region_add_subregion_common(mr, offset, subregion);
2672 void memory_region_add_subregion_overlap(MemoryRegion *mr,
2673 hwaddr offset,
2674 MemoryRegion *subregion,
2675 int priority)
2677 subregion->priority = priority;
2678 memory_region_add_subregion_common(mr, offset, subregion);
2681 void memory_region_del_subregion(MemoryRegion *mr,
2682 MemoryRegion *subregion)
2684 MemoryRegion *alias;
2686 memory_region_transaction_begin();
2687 assert(subregion->container == mr);
2688 subregion->container = NULL;
2689 for (alias = subregion->alias; alias; alias = alias->alias) {
2690 alias->mapped_via_alias--;
2691 assert(alias->mapped_via_alias >= 0);
2693 QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
2694 memory_region_unref(subregion);
2695 memory_region_update_pending |= mr->enabled && subregion->enabled;
2696 memory_region_transaction_commit();
2699 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
2701 if (enabled == mr->enabled) {
2702 return;
2704 memory_region_transaction_begin();
2705 mr->enabled = enabled;
2706 memory_region_update_pending = true;
2707 memory_region_transaction_commit();
2710 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
2712 Int128 s = int128_make64(size);
2714 if (size == UINT64_MAX) {
2715 s = int128_2_64();
2717 if (int128_eq(s, mr->size)) {
2718 return;
2720 memory_region_transaction_begin();
2721 mr->size = s;
2722 memory_region_update_pending = true;
2723 memory_region_transaction_commit();
2726 static void memory_region_readd_subregion(MemoryRegion *mr)
2728 MemoryRegion *container = mr->container;
2730 if (container) {
2731 memory_region_transaction_begin();
2732 memory_region_ref(mr);
2733 memory_region_del_subregion(container, mr);
2734 memory_region_add_subregion_common(container, mr->addr, mr);
2735 memory_region_unref(mr);
2736 memory_region_transaction_commit();
2740 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
2742 if (addr != mr->addr) {
2743 mr->addr = addr;
2744 memory_region_readd_subregion(mr);
2748 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
2750 assert(mr->alias);
2752 if (offset == mr->alias_offset) {
2753 return;
2756 memory_region_transaction_begin();
2757 mr->alias_offset = offset;
2758 memory_region_update_pending |= mr->enabled;
2759 memory_region_transaction_commit();
2762 void memory_region_set_unmergeable(MemoryRegion *mr, bool unmergeable)
2764 if (unmergeable == mr->unmergeable) {
2765 return;
2768 memory_region_transaction_begin();
2769 mr->unmergeable = unmergeable;
2770 memory_region_update_pending |= mr->enabled;
2771 memory_region_transaction_commit();
2774 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
2776 return mr->align;
2779 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
2781 const AddrRange *addr = addr_;
2782 const FlatRange *fr = fr_;
2784 if (int128_le(addrrange_end(*addr), fr->addr.start)) {
2785 return -1;
2786 } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
2787 return 1;
2789 return 0;
2792 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
2794 return bsearch(&addr, view->ranges, view->nr,
2795 sizeof(FlatRange), cmp_flatrange_addr);
2798 bool memory_region_is_mapped(MemoryRegion *mr)
2800 return !!mr->container || mr->mapped_via_alias;
2803 /* Same as memory_region_find, but it does not add a reference to the
2804 * returned region. It must be called from an RCU critical section.
2806 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2807 hwaddr addr, uint64_t size)
2809 MemoryRegionSection ret = { .mr = NULL };
2810 MemoryRegion *root;
2811 AddressSpace *as;
2812 AddrRange range;
2813 FlatView *view;
2814 FlatRange *fr;
2816 addr += mr->addr;
2817 for (root = mr; root->container; ) {
2818 root = root->container;
2819 addr += root->addr;
2822 as = memory_region_to_address_space(root);
2823 if (!as) {
2824 return ret;
2826 range = addrrange_make(int128_make64(addr), int128_make64(size));
2828 view = address_space_to_flatview(as);
2829 fr = flatview_lookup(view, range);
2830 if (!fr) {
2831 return ret;
2834 while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2835 --fr;
2838 ret.mr = fr->mr;
2839 ret.fv = view;
2840 range = addrrange_intersection(range, fr->addr);
2841 ret.offset_within_region = fr->offset_in_region;
2842 ret.offset_within_region += int128_get64(int128_sub(range.start,
2843 fr->addr.start));
2844 ret.size = range.size;
2845 ret.offset_within_address_space = int128_get64(range.start);
2846 ret.readonly = fr->readonly;
2847 ret.nonvolatile = fr->nonvolatile;
2848 return ret;
2851 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2852 hwaddr addr, uint64_t size)
2854 MemoryRegionSection ret;
2855 RCU_READ_LOCK_GUARD();
2856 ret = memory_region_find_rcu(mr, addr, size);
2857 if (ret.mr) {
2858 memory_region_ref(ret.mr);
2860 return ret;
2863 MemoryRegionSection *memory_region_section_new_copy(MemoryRegionSection *s)
2865 MemoryRegionSection *tmp = g_new(MemoryRegionSection, 1);
2867 *tmp = *s;
2868 if (tmp->mr) {
2869 memory_region_ref(tmp->mr);
2871 if (tmp->fv) {
2872 bool ret = flatview_ref(tmp->fv);
2874 g_assert(ret);
2876 return tmp;
2879 void memory_region_section_free_copy(MemoryRegionSection *s)
2881 if (s->fv) {
2882 flatview_unref(s->fv);
2884 if (s->mr) {
2885 memory_region_unref(s->mr);
2887 g_free(s);
2890 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2892 MemoryRegion *mr;
2894 RCU_READ_LOCK_GUARD();
2895 mr = memory_region_find_rcu(container, addr, 1).mr;
2896 return mr && mr != container;
2899 void memory_global_dirty_log_sync(bool last_stage)
2901 memory_region_sync_dirty_bitmap(NULL, last_stage);
2904 void memory_global_after_dirty_log_sync(void)
2906 MEMORY_LISTENER_CALL_GLOBAL(log_global_after_sync, Forward);
2910 * Dirty track stop flags that are postponed due to VM being stopped. Should
2911 * only be used within vmstate_change hook.
2913 static unsigned int postponed_stop_flags;
2914 static VMChangeStateEntry *vmstate_change;
2915 static void memory_global_dirty_log_stop_postponed_run(void);
2917 void memory_global_dirty_log_start(unsigned int flags)
2919 unsigned int old_flags;
2921 assert(flags && !(flags & (~GLOBAL_DIRTY_MASK)));
2923 if (vmstate_change) {
2924 /* If there is postponed stop(), operate on it first */
2925 postponed_stop_flags &= ~flags;
2926 memory_global_dirty_log_stop_postponed_run();
2929 flags &= ~global_dirty_tracking;
2930 if (!flags) {
2931 return;
2934 old_flags = global_dirty_tracking;
2935 global_dirty_tracking |= flags;
2936 trace_global_dirty_changed(global_dirty_tracking);
2938 if (!old_flags) {
2939 MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2940 memory_region_transaction_begin();
2941 memory_region_update_pending = true;
2942 memory_region_transaction_commit();
2946 static void memory_global_dirty_log_do_stop(unsigned int flags)
2948 assert(flags && !(flags & (~GLOBAL_DIRTY_MASK)));
2949 assert((global_dirty_tracking & flags) == flags);
2950 global_dirty_tracking &= ~flags;
2952 trace_global_dirty_changed(global_dirty_tracking);
2954 if (!global_dirty_tracking) {
2955 memory_region_transaction_begin();
2956 memory_region_update_pending = true;
2957 memory_region_transaction_commit();
2958 MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2963 * Execute the postponed dirty log stop operations if there is, then reset
2964 * everything (including the flags and the vmstate change hook).
2966 static void memory_global_dirty_log_stop_postponed_run(void)
2968 /* This must be called with the vmstate handler registered */
2969 assert(vmstate_change);
2971 /* Note: postponed_stop_flags can be cleared in log start routine */
2972 if (postponed_stop_flags) {
2973 memory_global_dirty_log_do_stop(postponed_stop_flags);
2974 postponed_stop_flags = 0;
2977 qemu_del_vm_change_state_handler(vmstate_change);
2978 vmstate_change = NULL;
2981 static void memory_vm_change_state_handler(void *opaque, bool running,
2982 RunState state)
2984 if (running) {
2985 memory_global_dirty_log_stop_postponed_run();
2989 void memory_global_dirty_log_stop(unsigned int flags)
2991 if (!runstate_is_running()) {
2992 /* Postpone the dirty log stop, e.g., to when VM starts again */
2993 if (vmstate_change) {
2994 /* Batch with previous postponed flags */
2995 postponed_stop_flags |= flags;
2996 } else {
2997 postponed_stop_flags = flags;
2998 vmstate_change = qemu_add_vm_change_state_handler(
2999 memory_vm_change_state_handler, NULL);
3001 return;
3004 memory_global_dirty_log_do_stop(flags);
3007 static void listener_add_address_space(MemoryListener *listener,
3008 AddressSpace *as)
3010 FlatView *view;
3011 FlatRange *fr;
3013 if (listener->begin) {
3014 listener->begin(listener);
3016 if (global_dirty_tracking) {
3017 if (listener->log_global_start) {
3018 listener->log_global_start(listener);
3022 view = address_space_get_flatview(as);
3023 FOR_EACH_FLAT_RANGE(fr, view) {
3024 MemoryRegionSection section = section_from_flat_range(fr, view);
3026 if (listener->region_add) {
3027 listener->region_add(listener, &section);
3029 if (fr->dirty_log_mask && listener->log_start) {
3030 listener->log_start(listener, &section, 0, fr->dirty_log_mask);
3033 if (listener->commit) {
3034 listener->commit(listener);
3036 flatview_unref(view);
3039 static void listener_del_address_space(MemoryListener *listener,
3040 AddressSpace *as)
3042 FlatView *view;
3043 FlatRange *fr;
3045 if (listener->begin) {
3046 listener->begin(listener);
3048 view = address_space_get_flatview(as);
3049 FOR_EACH_FLAT_RANGE(fr, view) {
3050 MemoryRegionSection section = section_from_flat_range(fr, view);
3052 if (fr->dirty_log_mask && listener->log_stop) {
3053 listener->log_stop(listener, &section, fr->dirty_log_mask, 0);
3055 if (listener->region_del) {
3056 listener->region_del(listener, &section);
3059 if (listener->commit) {
3060 listener->commit(listener);
3062 flatview_unref(view);
3065 void memory_listener_register(MemoryListener *listener, AddressSpace *as)
3067 MemoryListener *other = NULL;
3069 /* Only one of them can be defined for a listener */
3070 assert(!(listener->log_sync && listener->log_sync_global));
3072 listener->address_space = as;
3073 if (QTAILQ_EMPTY(&memory_listeners)
3074 || listener->priority >= QTAILQ_LAST(&memory_listeners)->priority) {
3075 QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
3076 } else {
3077 QTAILQ_FOREACH(other, &memory_listeners, link) {
3078 if (listener->priority < other->priority) {
3079 break;
3082 QTAILQ_INSERT_BEFORE(other, listener, link);
3085 if (QTAILQ_EMPTY(&as->listeners)
3086 || listener->priority >= QTAILQ_LAST(&as->listeners)->priority) {
3087 QTAILQ_INSERT_TAIL(&as->listeners, listener, link_as);
3088 } else {
3089 QTAILQ_FOREACH(other, &as->listeners, link_as) {
3090 if (listener->priority < other->priority) {
3091 break;
3094 QTAILQ_INSERT_BEFORE(other, listener, link_as);
3097 listener_add_address_space(listener, as);
3099 if (listener->eventfd_add || listener->eventfd_del) {
3100 as->ioeventfd_notifiers++;
3104 void memory_listener_unregister(MemoryListener *listener)
3106 if (!listener->address_space) {
3107 return;
3110 if (listener->eventfd_add || listener->eventfd_del) {
3111 listener->address_space->ioeventfd_notifiers--;
3114 listener_del_address_space(listener, listener->address_space);
3115 QTAILQ_REMOVE(&memory_listeners, listener, link);
3116 QTAILQ_REMOVE(&listener->address_space->listeners, listener, link_as);
3117 listener->address_space = NULL;
3120 void address_space_remove_listeners(AddressSpace *as)
3122 while (!QTAILQ_EMPTY(&as->listeners)) {
3123 memory_listener_unregister(QTAILQ_FIRST(&as->listeners));
3127 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
3129 memory_region_ref(root);
3130 as->root = root;
3131 as->current_map = NULL;
3132 as->ioeventfd_nb = 0;
3133 as->ioeventfds = NULL;
3134 QTAILQ_INIT(&as->listeners);
3135 QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
3136 as->name = g_strdup(name ? name : "anonymous");
3137 address_space_update_topology(as);
3138 address_space_update_ioeventfds(as);
3141 static void do_address_space_destroy(AddressSpace *as)
3143 assert(QTAILQ_EMPTY(&as->listeners));
3145 flatview_unref(as->current_map);
3146 g_free(as->name);
3147 g_free(as->ioeventfds);
3148 memory_region_unref(as->root);
3151 void address_space_destroy(AddressSpace *as)
3153 MemoryRegion *root = as->root;
3155 /* Flush out anything from MemoryListeners listening in on this */
3156 memory_region_transaction_begin();
3157 as->root = NULL;
3158 memory_region_transaction_commit();
3159 QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
3161 /* At this point, as->dispatch and as->current_map are dummy
3162 * entries that the guest should never use. Wait for the old
3163 * values to expire before freeing the data.
3165 as->root = root;
3166 call_rcu(as, do_address_space_destroy, rcu);
3169 static const char *memory_region_type(MemoryRegion *mr)
3171 if (mr->alias) {
3172 return memory_region_type(mr->alias);
3174 if (memory_region_is_ram_device(mr)) {
3175 return "ramd";
3176 } else if (memory_region_is_romd(mr)) {
3177 return "romd";
3178 } else if (memory_region_is_rom(mr)) {
3179 return "rom";
3180 } else if (memory_region_is_ram(mr)) {
3181 return "ram";
3182 } else {
3183 return "i/o";
3187 typedef struct MemoryRegionList MemoryRegionList;
3189 struct MemoryRegionList {
3190 const MemoryRegion *mr;
3191 QTAILQ_ENTRY(MemoryRegionList) mrqueue;
3194 typedef QTAILQ_HEAD(, MemoryRegionList) MemoryRegionListHead;
3196 #define MR_SIZE(size) (int128_nz(size) ? (hwaddr)int128_get64( \
3197 int128_sub((size), int128_one())) : 0)
3198 #define MTREE_INDENT " "
3200 static void mtree_expand_owner(const char *label, Object *obj)
3202 DeviceState *dev = (DeviceState *) object_dynamic_cast(obj, TYPE_DEVICE);
3204 qemu_printf(" %s:{%s", label, dev ? "dev" : "obj");
3205 if (dev && dev->id) {
3206 qemu_printf(" id=%s", dev->id);
3207 } else {
3208 char *canonical_path = object_get_canonical_path(obj);
3209 if (canonical_path) {
3210 qemu_printf(" path=%s", canonical_path);
3211 g_free(canonical_path);
3212 } else {
3213 qemu_printf(" type=%s", object_get_typename(obj));
3216 qemu_printf("}");
3219 static void mtree_print_mr_owner(const MemoryRegion *mr)
3221 Object *owner = mr->owner;
3222 Object *parent = memory_region_owner((MemoryRegion *)mr);
3224 if (!owner && !parent) {
3225 qemu_printf(" orphan");
3226 return;
3228 if (owner) {
3229 mtree_expand_owner("owner", owner);
3231 if (parent && parent != owner) {
3232 mtree_expand_owner("parent", parent);
3236 static void mtree_print_mr(const MemoryRegion *mr, unsigned int level,
3237 hwaddr base,
3238 MemoryRegionListHead *alias_print_queue,
3239 bool owner, bool display_disabled)
3241 MemoryRegionList *new_ml, *ml, *next_ml;
3242 MemoryRegionListHead submr_print_queue;
3243 const MemoryRegion *submr;
3244 unsigned int i;
3245 hwaddr cur_start, cur_end;
3247 if (!mr) {
3248 return;
3251 cur_start = base + mr->addr;
3252 cur_end = cur_start + MR_SIZE(mr->size);
3255 * Try to detect overflow of memory region. This should never
3256 * happen normally. When it happens, we dump something to warn the
3257 * user who is observing this.
3259 if (cur_start < base || cur_end < cur_start) {
3260 qemu_printf("[DETECTED OVERFLOW!] ");
3263 if (mr->alias) {
3264 bool found = false;
3266 /* check if the alias is already in the queue */
3267 QTAILQ_FOREACH(ml, alias_print_queue, mrqueue) {
3268 if (ml->mr == mr->alias) {
3269 found = true;
3273 if (!found) {
3274 ml = g_new(MemoryRegionList, 1);
3275 ml->mr = mr->alias;
3276 QTAILQ_INSERT_TAIL(alias_print_queue, ml, mrqueue);
3278 if (mr->enabled || display_disabled) {
3279 for (i = 0; i < level; i++) {
3280 qemu_printf(MTREE_INDENT);
3282 qemu_printf(HWADDR_FMT_plx "-" HWADDR_FMT_plx
3283 " (prio %d, %s%s): alias %s @%s " HWADDR_FMT_plx
3284 "-" HWADDR_FMT_plx "%s",
3285 cur_start, cur_end,
3286 mr->priority,
3287 mr->nonvolatile ? "nv-" : "",
3288 memory_region_type((MemoryRegion *)mr),
3289 memory_region_name(mr),
3290 memory_region_name(mr->alias),
3291 mr->alias_offset,
3292 mr->alias_offset + MR_SIZE(mr->size),
3293 mr->enabled ? "" : " [disabled]");
3294 if (owner) {
3295 mtree_print_mr_owner(mr);
3297 qemu_printf("\n");
3299 } else {
3300 if (mr->enabled || display_disabled) {
3301 for (i = 0; i < level; i++) {
3302 qemu_printf(MTREE_INDENT);
3304 qemu_printf(HWADDR_FMT_plx "-" HWADDR_FMT_plx
3305 " (prio %d, %s%s): %s%s",
3306 cur_start, cur_end,
3307 mr->priority,
3308 mr->nonvolatile ? "nv-" : "",
3309 memory_region_type((MemoryRegion *)mr),
3310 memory_region_name(mr),
3311 mr->enabled ? "" : " [disabled]");
3312 if (owner) {
3313 mtree_print_mr_owner(mr);
3315 qemu_printf("\n");
3319 QTAILQ_INIT(&submr_print_queue);
3321 QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
3322 new_ml = g_new(MemoryRegionList, 1);
3323 new_ml->mr = submr;
3324 QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
3325 if (new_ml->mr->addr < ml->mr->addr ||
3326 (new_ml->mr->addr == ml->mr->addr &&
3327 new_ml->mr->priority > ml->mr->priority)) {
3328 QTAILQ_INSERT_BEFORE(ml, new_ml, mrqueue);
3329 new_ml = NULL;
3330 break;
3333 if (new_ml) {
3334 QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, mrqueue);
3338 QTAILQ_FOREACH(ml, &submr_print_queue, mrqueue) {
3339 mtree_print_mr(ml->mr, level + 1, cur_start,
3340 alias_print_queue, owner, display_disabled);
3343 QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, mrqueue, next_ml) {
3344 g_free(ml);
3348 struct FlatViewInfo {
3349 int counter;
3350 bool dispatch_tree;
3351 bool owner;
3352 AccelClass *ac;
3355 static void mtree_print_flatview(gpointer key, gpointer value,
3356 gpointer user_data)
3358 FlatView *view = key;
3359 GArray *fv_address_spaces = value;
3360 struct FlatViewInfo *fvi = user_data;
3361 FlatRange *range = &view->ranges[0];
3362 MemoryRegion *mr;
3363 int n = view->nr;
3364 int i;
3365 AddressSpace *as;
3367 qemu_printf("FlatView #%d\n", fvi->counter);
3368 ++fvi->counter;
3370 for (i = 0; i < fv_address_spaces->len; ++i) {
3371 as = g_array_index(fv_address_spaces, AddressSpace*, i);
3372 qemu_printf(" AS \"%s\", root: %s",
3373 as->name, memory_region_name(as->root));
3374 if (as->root->alias) {
3375 qemu_printf(", alias %s", memory_region_name(as->root->alias));
3377 qemu_printf("\n");
3380 qemu_printf(" Root memory region: %s\n",
3381 view->root ? memory_region_name(view->root) : "(none)");
3383 if (n <= 0) {
3384 qemu_printf(MTREE_INDENT "No rendered FlatView\n\n");
3385 return;
3388 while (n--) {
3389 mr = range->mr;
3390 if (range->offset_in_region) {
3391 qemu_printf(MTREE_INDENT HWADDR_FMT_plx "-" HWADDR_FMT_plx
3392 " (prio %d, %s%s): %s @" HWADDR_FMT_plx,
3393 int128_get64(range->addr.start),
3394 int128_get64(range->addr.start)
3395 + MR_SIZE(range->addr.size),
3396 mr->priority,
3397 range->nonvolatile ? "nv-" : "",
3398 range->readonly ? "rom" : memory_region_type(mr),
3399 memory_region_name(mr),
3400 range->offset_in_region);
3401 } else {
3402 qemu_printf(MTREE_INDENT HWADDR_FMT_plx "-" HWADDR_FMT_plx
3403 " (prio %d, %s%s): %s",
3404 int128_get64(range->addr.start),
3405 int128_get64(range->addr.start)
3406 + MR_SIZE(range->addr.size),
3407 mr->priority,
3408 range->nonvolatile ? "nv-" : "",
3409 range->readonly ? "rom" : memory_region_type(mr),
3410 memory_region_name(mr));
3412 if (fvi->owner) {
3413 mtree_print_mr_owner(mr);
3416 if (fvi->ac) {
3417 for (i = 0; i < fv_address_spaces->len; ++i) {
3418 as = g_array_index(fv_address_spaces, AddressSpace*, i);
3419 if (fvi->ac->has_memory(current_machine, as,
3420 int128_get64(range->addr.start),
3421 MR_SIZE(range->addr.size) + 1)) {
3422 qemu_printf(" %s", fvi->ac->name);
3426 qemu_printf("\n");
3427 range++;
3430 #if !defined(CONFIG_USER_ONLY)
3431 if (fvi->dispatch_tree && view->root) {
3432 mtree_print_dispatch(view->dispatch, view->root);
3434 #endif
3436 qemu_printf("\n");
3439 static gboolean mtree_info_flatview_free(gpointer key, gpointer value,
3440 gpointer user_data)
3442 FlatView *view = key;
3443 GArray *fv_address_spaces = value;
3445 g_array_unref(fv_address_spaces);
3446 flatview_unref(view);
3448 return true;
3451 static void mtree_info_flatview(bool dispatch_tree, bool owner)
3453 struct FlatViewInfo fvi = {
3454 .counter = 0,
3455 .dispatch_tree = dispatch_tree,
3456 .owner = owner,
3458 AddressSpace *as;
3459 FlatView *view;
3460 GArray *fv_address_spaces;
3461 GHashTable *views = g_hash_table_new(g_direct_hash, g_direct_equal);
3462 AccelClass *ac = ACCEL_GET_CLASS(current_accel());
3464 if (ac->has_memory) {
3465 fvi.ac = ac;
3468 /* Gather all FVs in one table */
3469 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3470 view = address_space_get_flatview(as);
3472 fv_address_spaces = g_hash_table_lookup(views, view);
3473 if (!fv_address_spaces) {
3474 fv_address_spaces = g_array_new(false, false, sizeof(as));
3475 g_hash_table_insert(views, view, fv_address_spaces);
3478 g_array_append_val(fv_address_spaces, as);
3481 /* Print */
3482 g_hash_table_foreach(views, mtree_print_flatview, &fvi);
3484 /* Free */
3485 g_hash_table_foreach_remove(views, mtree_info_flatview_free, 0);
3486 g_hash_table_unref(views);
3489 struct AddressSpaceInfo {
3490 MemoryRegionListHead *ml_head;
3491 bool owner;
3492 bool disabled;
3495 /* Returns negative value if a < b; zero if a = b; positive value if a > b. */
3496 static gint address_space_compare_name(gconstpointer a, gconstpointer b)
3498 const AddressSpace *as_a = a;
3499 const AddressSpace *as_b = b;
3501 return g_strcmp0(as_a->name, as_b->name);
3504 static void mtree_print_as_name(gpointer data, gpointer user_data)
3506 AddressSpace *as = data;
3508 qemu_printf("address-space: %s\n", as->name);
3511 static void mtree_print_as(gpointer key, gpointer value, gpointer user_data)
3513 MemoryRegion *mr = key;
3514 GSList *as_same_root_mr_list = value;
3515 struct AddressSpaceInfo *asi = user_data;
3517 g_slist_foreach(as_same_root_mr_list, mtree_print_as_name, NULL);
3518 mtree_print_mr(mr, 1, 0, asi->ml_head, asi->owner, asi->disabled);
3519 qemu_printf("\n");
3522 static gboolean mtree_info_as_free(gpointer key, gpointer value,
3523 gpointer user_data)
3525 GSList *as_same_root_mr_list = value;
3527 g_slist_free(as_same_root_mr_list);
3529 return true;
3532 static void mtree_info_as(bool dispatch_tree, bool owner, bool disabled)
3534 MemoryRegionListHead ml_head;
3535 MemoryRegionList *ml, *ml2;
3536 AddressSpace *as;
3537 GHashTable *views = g_hash_table_new(g_direct_hash, g_direct_equal);
3538 GSList *as_same_root_mr_list;
3539 struct AddressSpaceInfo asi = {
3540 .ml_head = &ml_head,
3541 .owner = owner,
3542 .disabled = disabled,
3545 QTAILQ_INIT(&ml_head);
3547 QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
3548 /* Create hashtable, key=AS root MR, value = list of AS */
3549 as_same_root_mr_list = g_hash_table_lookup(views, as->root);
3550 as_same_root_mr_list = g_slist_insert_sorted(as_same_root_mr_list, as,
3551 address_space_compare_name);
3552 g_hash_table_insert(views, as->root, as_same_root_mr_list);
3555 /* print address spaces */
3556 g_hash_table_foreach(views, mtree_print_as, &asi);
3557 g_hash_table_foreach_remove(views, mtree_info_as_free, 0);
3558 g_hash_table_unref(views);
3560 /* print aliased regions */
3561 QTAILQ_FOREACH(ml, &ml_head, mrqueue) {
3562 qemu_printf("memory-region: %s\n", memory_region_name(ml->mr));
3563 mtree_print_mr(ml->mr, 1, 0, &ml_head, owner, disabled);
3564 qemu_printf("\n");
3567 QTAILQ_FOREACH_SAFE(ml, &ml_head, mrqueue, ml2) {
3568 g_free(ml);
3572 void mtree_info(bool flatview, bool dispatch_tree, bool owner, bool disabled)
3574 if (flatview) {
3575 mtree_info_flatview(dispatch_tree, owner);
3576 } else {
3577 mtree_info_as(dispatch_tree, owner, disabled);
3581 bool memory_region_init_ram(MemoryRegion *mr,
3582 Object *owner,
3583 const char *name,
3584 uint64_t size,
3585 Error **errp)
3587 DeviceState *owner_dev;
3589 if (!memory_region_init_ram_nomigrate(mr, owner, name, size, errp)) {
3590 return false;
3592 /* This will assert if owner is neither NULL nor a DeviceState.
3593 * We only want the owner here for the purposes of defining a
3594 * unique name for migration. TODO: Ideally we should implement
3595 * a naming scheme for Objects which are not DeviceStates, in
3596 * which case we can relax this restriction.
3598 owner_dev = DEVICE(owner);
3599 vmstate_register_ram(mr, owner_dev);
3601 return true;
3604 bool memory_region_init_rom(MemoryRegion *mr,
3605 Object *owner,
3606 const char *name,
3607 uint64_t size,
3608 Error **errp)
3610 DeviceState *owner_dev;
3612 if (!memory_region_init_rom_nomigrate(mr, owner, name, size, errp)) {
3613 return false;
3615 /* This will assert if owner is neither NULL nor a DeviceState.
3616 * We only want the owner here for the purposes of defining a
3617 * unique name for migration. TODO: Ideally we should implement
3618 * a naming scheme for Objects which are not DeviceStates, in
3619 * which case we can relax this restriction.
3621 owner_dev = DEVICE(owner);
3622 vmstate_register_ram(mr, owner_dev);
3624 return true;
3627 bool memory_region_init_rom_device(MemoryRegion *mr,
3628 Object *owner,
3629 const MemoryRegionOps *ops,
3630 void *opaque,
3631 const char *name,
3632 uint64_t size,
3633 Error **errp)
3635 DeviceState *owner_dev;
3637 if (!memory_region_init_rom_device_nomigrate(mr, owner, ops, opaque,
3638 name, size, errp)) {
3639 return false;
3641 /* This will assert if owner is neither NULL nor a DeviceState.
3642 * We only want the owner here for the purposes of defining a
3643 * unique name for migration. TODO: Ideally we should implement
3644 * a naming scheme for Objects which are not DeviceStates, in
3645 * which case we can relax this restriction.
3647 owner_dev = DEVICE(owner);
3648 vmstate_register_ram(mr, owner_dev);
3650 return true;
3654 * Support system builds with CONFIG_FUZZ using a weak symbol and a stub for
3655 * the fuzz_dma_read_cb callback
3657 #ifdef CONFIG_FUZZ
3658 void __attribute__((weak)) fuzz_dma_read_cb(size_t addr,
3659 size_t len,
3660 MemoryRegion *mr)
3663 #endif
3665 static const TypeInfo memory_region_info = {
3666 .parent = TYPE_OBJECT,
3667 .name = TYPE_MEMORY_REGION,
3668 .class_size = sizeof(MemoryRegionClass),
3669 .instance_size = sizeof(MemoryRegion),
3670 .instance_init = memory_region_initfn,
3671 .instance_finalize = memory_region_finalize,
3674 static const TypeInfo iommu_memory_region_info = {
3675 .parent = TYPE_MEMORY_REGION,
3676 .name = TYPE_IOMMU_MEMORY_REGION,
3677 .class_size = sizeof(IOMMUMemoryRegionClass),
3678 .instance_size = sizeof(IOMMUMemoryRegion),
3679 .instance_init = iommu_memory_region_initfn,
3680 .abstract = true,
3683 static const TypeInfo ram_discard_manager_info = {
3684 .parent = TYPE_INTERFACE,
3685 .name = TYPE_RAM_DISCARD_MANAGER,
3686 .class_size = sizeof(RamDiscardManagerClass),
3689 static void memory_register_types(void)
3691 type_register_static(&memory_region_info);
3692 type_register_static(&iommu_memory_region_info);
3693 type_register_static(&ram_discard_manager_info);
3696 type_init(memory_register_types)