2 * Physical memory management
4 * Copyright 2011 Red Hat, Inc. and/or its affiliates
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"
18 #include "qapi/error.h"
20 #include "exec/memory.h"
21 #include "exec/address-spaces.h"
22 #include "qapi/visitor.h"
23 #include "qemu/bitops.h"
24 #include "qemu/error-report.h"
25 #include "qemu/main-loop.h"
26 #include "qemu/qemu-print.h"
27 #include "qom/object.h"
30 #include "exec/memory-internal.h"
31 #include "exec/ram_addr.h"
32 #include "sysemu/kvm.h"
33 #include "sysemu/runstate.h"
34 #include "sysemu/tcg.h"
35 #include "sysemu/accel.h"
36 #include "hw/boards.h"
37 #include "migration/vmstate.h"
39 //#define DEBUG_UNASSIGNED
41 static unsigned memory_region_transaction_depth
;
42 static bool memory_region_update_pending
;
43 static bool ioeventfd_update_pending
;
44 bool global_dirty_log
;
46 static QTAILQ_HEAD(, MemoryListener
) memory_listeners
47 = QTAILQ_HEAD_INITIALIZER(memory_listeners
);
49 static QTAILQ_HEAD(, AddressSpace
) address_spaces
50 = QTAILQ_HEAD_INITIALIZER(address_spaces
);
52 static GHashTable
*flat_views
;
54 typedef struct AddrRange AddrRange
;
57 * Note that signed integers are needed for negative offsetting in aliases
58 * (large MemoryRegion::alias_offset).
65 static AddrRange
addrrange_make(Int128 start
, Int128 size
)
67 return (AddrRange
) { start
, size
};
70 static bool addrrange_equal(AddrRange r1
, AddrRange r2
)
72 return int128_eq(r1
.start
, r2
.start
) && int128_eq(r1
.size
, r2
.size
);
75 static Int128
addrrange_end(AddrRange r
)
77 return int128_add(r
.start
, r
.size
);
80 static AddrRange
addrrange_shift(AddrRange range
, Int128 delta
)
82 int128_addto(&range
.start
, delta
);
86 static bool addrrange_contains(AddrRange range
, Int128 addr
)
88 return int128_ge(addr
, range
.start
)
89 && int128_lt(addr
, addrrange_end(range
));
92 static bool addrrange_intersects(AddrRange r1
, AddrRange r2
)
94 return addrrange_contains(r1
, r2
.start
)
95 || addrrange_contains(r2
, r1
.start
);
98 static AddrRange
addrrange_intersection(AddrRange r1
, AddrRange r2
)
100 Int128 start
= int128_max(r1
.start
, r2
.start
);
101 Int128 end
= int128_min(addrrange_end(r1
), addrrange_end(r2
));
102 return addrrange_make(start
, int128_sub(end
, start
));
105 enum ListenerDirection
{ Forward
, Reverse
};
107 #define MEMORY_LISTENER_CALL_GLOBAL(_callback, _direction, _args...) \
109 MemoryListener *_listener; \
111 switch (_direction) { \
113 QTAILQ_FOREACH(_listener, &memory_listeners, link) { \
114 if (_listener->_callback) { \
115 _listener->_callback(_listener, ##_args); \
120 QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners, link) { \
121 if (_listener->_callback) { \
122 _listener->_callback(_listener, ##_args); \
131 #define MEMORY_LISTENER_CALL(_as, _callback, _direction, _section, _args...) \
133 MemoryListener *_listener; \
135 switch (_direction) { \
137 QTAILQ_FOREACH(_listener, &(_as)->listeners, link_as) { \
138 if (_listener->_callback) { \
139 _listener->_callback(_listener, _section, ##_args); \
144 QTAILQ_FOREACH_REVERSE(_listener, &(_as)->listeners, link_as) { \
145 if (_listener->_callback) { \
146 _listener->_callback(_listener, _section, ##_args); \
155 /* No need to ref/unref .mr, the FlatRange keeps it alive. */
156 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...) \
158 MemoryRegionSection mrs = section_from_flat_range(fr, \
159 address_space_to_flatview(as)); \
160 MEMORY_LISTENER_CALL(as, callback, dir, &mrs, ##_args); \
163 struct CoalescedMemoryRange
{
165 QTAILQ_ENTRY(CoalescedMemoryRange
) link
;
168 struct MemoryRegionIoeventfd
{
175 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd
*a
,
176 MemoryRegionIoeventfd
*b
)
178 if (int128_lt(a
->addr
.start
, b
->addr
.start
)) {
180 } else if (int128_gt(a
->addr
.start
, b
->addr
.start
)) {
182 } else if (int128_lt(a
->addr
.size
, b
->addr
.size
)) {
184 } else if (int128_gt(a
->addr
.size
, b
->addr
.size
)) {
186 } else if (a
->match_data
< b
->match_data
) {
188 } else if (a
->match_data
> b
->match_data
) {
190 } else if (a
->match_data
) {
191 if (a
->data
< b
->data
) {
193 } else if (a
->data
> b
->data
) {
199 } else if (a
->e
> b
->e
) {
205 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd
*a
,
206 MemoryRegionIoeventfd
*b
)
208 return !memory_region_ioeventfd_before(a
, b
)
209 && !memory_region_ioeventfd_before(b
, a
);
212 /* Range of memory in the global map. Addresses are absolute. */
215 hwaddr offset_in_region
;
217 uint8_t dirty_log_mask
;
223 #define FOR_EACH_FLAT_RANGE(var, view) \
224 for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
226 static inline MemoryRegionSection
227 section_from_flat_range(FlatRange
*fr
, FlatView
*fv
)
229 return (MemoryRegionSection
) {
232 .offset_within_region
= fr
->offset_in_region
,
233 .size
= fr
->addr
.size
,
234 .offset_within_address_space
= int128_get64(fr
->addr
.start
),
235 .readonly
= fr
->readonly
,
236 .nonvolatile
= fr
->nonvolatile
,
240 static bool flatrange_equal(FlatRange
*a
, FlatRange
*b
)
242 return a
->mr
== b
->mr
243 && addrrange_equal(a
->addr
, b
->addr
)
244 && a
->offset_in_region
== b
->offset_in_region
245 && a
->romd_mode
== b
->romd_mode
246 && a
->readonly
== b
->readonly
247 && a
->nonvolatile
== b
->nonvolatile
;
250 static FlatView
*flatview_new(MemoryRegion
*mr_root
)
254 view
= g_new0(FlatView
, 1);
256 view
->root
= mr_root
;
257 memory_region_ref(mr_root
);
258 trace_flatview_new(view
, mr_root
);
263 /* Insert a range into a given position. Caller is responsible for maintaining
266 static void flatview_insert(FlatView
*view
, unsigned pos
, FlatRange
*range
)
268 if (view
->nr
== view
->nr_allocated
) {
269 view
->nr_allocated
= MAX(2 * view
->nr
, 10);
270 view
->ranges
= g_realloc(view
->ranges
,
271 view
->nr_allocated
* sizeof(*view
->ranges
));
273 memmove(view
->ranges
+ pos
+ 1, view
->ranges
+ pos
,
274 (view
->nr
- pos
) * sizeof(FlatRange
));
275 view
->ranges
[pos
] = *range
;
276 memory_region_ref(range
->mr
);
280 static void flatview_destroy(FlatView
*view
)
284 trace_flatview_destroy(view
, view
->root
);
285 if (view
->dispatch
) {
286 address_space_dispatch_free(view
->dispatch
);
288 for (i
= 0; i
< view
->nr
; i
++) {
289 memory_region_unref(view
->ranges
[i
].mr
);
291 g_free(view
->ranges
);
292 memory_region_unref(view
->root
);
296 static bool flatview_ref(FlatView
*view
)
298 return qatomic_fetch_inc_nonzero(&view
->ref
) > 0;
301 void flatview_unref(FlatView
*view
)
303 if (qatomic_fetch_dec(&view
->ref
) == 1) {
304 trace_flatview_destroy_rcu(view
, view
->root
);
306 call_rcu(view
, flatview_destroy
, rcu
);
310 static bool can_merge(FlatRange
*r1
, FlatRange
*r2
)
312 return int128_eq(addrrange_end(r1
->addr
), r2
->addr
.start
)
314 && int128_eq(int128_add(int128_make64(r1
->offset_in_region
),
316 int128_make64(r2
->offset_in_region
))
317 && r1
->dirty_log_mask
== r2
->dirty_log_mask
318 && r1
->romd_mode
== r2
->romd_mode
319 && r1
->readonly
== r2
->readonly
320 && r1
->nonvolatile
== r2
->nonvolatile
;
323 /* Attempt to simplify a view by merging adjacent ranges */
324 static void flatview_simplify(FlatView
*view
)
329 while (i
< view
->nr
) {
332 && can_merge(&view
->ranges
[j
-1], &view
->ranges
[j
])) {
333 int128_addto(&view
->ranges
[i
].addr
.size
, view
->ranges
[j
].addr
.size
);
337 for (k
= i
; k
< j
; k
++) {
338 memory_region_unref(view
->ranges
[k
].mr
);
340 memmove(&view
->ranges
[i
], &view
->ranges
[j
],
341 (view
->nr
- j
) * sizeof(view
->ranges
[j
]));
346 static bool memory_region_big_endian(MemoryRegion
*mr
)
348 #ifdef TARGET_WORDS_BIGENDIAN
349 return mr
->ops
->endianness
!= DEVICE_LITTLE_ENDIAN
;
351 return mr
->ops
->endianness
== DEVICE_BIG_ENDIAN
;
355 static void adjust_endianness(MemoryRegion
*mr
, uint64_t *data
, MemOp op
)
357 if ((op
& MO_BSWAP
) != devend_memop(mr
->ops
->endianness
)) {
358 switch (op
& MO_SIZE
) {
362 *data
= bswap16(*data
);
365 *data
= bswap32(*data
);
368 *data
= bswap64(*data
);
371 g_assert_not_reached();
376 static inline void memory_region_shift_read_access(uint64_t *value
,
382 *value
|= (tmp
& mask
) << shift
;
384 *value
|= (tmp
& mask
) >> -shift
;
388 static inline uint64_t memory_region_shift_write_access(uint64_t *value
,
395 tmp
= (*value
>> shift
) & mask
;
397 tmp
= (*value
<< -shift
) & mask
;
403 static hwaddr
memory_region_to_absolute_addr(MemoryRegion
*mr
, hwaddr offset
)
406 hwaddr abs_addr
= offset
;
408 abs_addr
+= mr
->addr
;
409 for (root
= mr
; root
->container
; ) {
410 root
= root
->container
;
411 abs_addr
+= root
->addr
;
417 static int get_cpu_index(void)
420 return current_cpu
->cpu_index
;
425 static MemTxResult
memory_region_read_accessor(MemoryRegion
*mr
,
435 tmp
= mr
->ops
->read(mr
->opaque
, addr
, size
);
437 trace_memory_region_subpage_read(get_cpu_index(), mr
, addr
, tmp
, size
);
438 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_READ
)) {
439 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
440 trace_memory_region_ops_read(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
442 memory_region_shift_read_access(value
, shift
, mask
, tmp
);
446 static MemTxResult
memory_region_read_with_attrs_accessor(MemoryRegion
*mr
,
457 r
= mr
->ops
->read_with_attrs(mr
->opaque
, addr
, &tmp
, size
, attrs
);
459 trace_memory_region_subpage_read(get_cpu_index(), mr
, addr
, tmp
, size
);
460 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_READ
)) {
461 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
462 trace_memory_region_ops_read(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
464 memory_region_shift_read_access(value
, shift
, mask
, tmp
);
468 static MemTxResult
memory_region_write_accessor(MemoryRegion
*mr
,
476 uint64_t tmp
= memory_region_shift_write_access(value
, shift
, mask
);
479 trace_memory_region_subpage_write(get_cpu_index(), mr
, addr
, tmp
, size
);
480 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_WRITE
)) {
481 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
482 trace_memory_region_ops_write(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
484 mr
->ops
->write(mr
->opaque
, addr
, tmp
, size
);
488 static MemTxResult
memory_region_write_with_attrs_accessor(MemoryRegion
*mr
,
496 uint64_t tmp
= memory_region_shift_write_access(value
, shift
, mask
);
499 trace_memory_region_subpage_write(get_cpu_index(), mr
, addr
, tmp
, size
);
500 } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_WRITE
)) {
501 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
502 trace_memory_region_ops_write(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
504 return mr
->ops
->write_with_attrs(mr
->opaque
, addr
, tmp
, size
, attrs
);
507 static MemTxResult
access_with_adjusted_size(hwaddr addr
,
510 unsigned access_size_min
,
511 unsigned access_size_max
,
512 MemTxResult (*access_fn
)
523 uint64_t access_mask
;
524 unsigned access_size
;
526 MemTxResult r
= MEMTX_OK
;
528 if (!access_size_min
) {
531 if (!access_size_max
) {
535 /* FIXME: support unaligned access? */
536 access_size
= MAX(MIN(size
, access_size_max
), access_size_min
);
537 access_mask
= MAKE_64BIT_MASK(0, access_size
* 8);
538 if (memory_region_big_endian(mr
)) {
539 for (i
= 0; i
< size
; i
+= access_size
) {
540 r
|= access_fn(mr
, addr
+ i
, value
, access_size
,
541 (size
- access_size
- i
) * 8, access_mask
, attrs
);
544 for (i
= 0; i
< size
; i
+= access_size
) {
545 r
|= access_fn(mr
, addr
+ i
, value
, access_size
, i
* 8,
552 static AddressSpace
*memory_region_to_address_space(MemoryRegion
*mr
)
556 while (mr
->container
) {
559 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
560 if (mr
== as
->root
) {
567 /* Render a memory region into the global view. Ranges in @view obscure
570 static void render_memory_region(FlatView
*view
,
577 MemoryRegion
*subregion
;
579 hwaddr offset_in_region
;
589 int128_addto(&base
, int128_make64(mr
->addr
));
590 readonly
|= mr
->readonly
;
591 nonvolatile
|= mr
->nonvolatile
;
593 tmp
= addrrange_make(base
, mr
->size
);
595 if (!addrrange_intersects(tmp
, clip
)) {
599 clip
= addrrange_intersection(tmp
, clip
);
602 int128_subfrom(&base
, int128_make64(mr
->alias
->addr
));
603 int128_subfrom(&base
, int128_make64(mr
->alias_offset
));
604 render_memory_region(view
, mr
->alias
, base
, clip
,
605 readonly
, nonvolatile
);
609 /* Render subregions in priority order. */
610 QTAILQ_FOREACH(subregion
, &mr
->subregions
, subregions_link
) {
611 render_memory_region(view
, subregion
, base
, clip
,
612 readonly
, nonvolatile
);
615 if (!mr
->terminates
) {
619 offset_in_region
= int128_get64(int128_sub(clip
.start
, base
));
624 fr
.dirty_log_mask
= memory_region_get_dirty_log_mask(mr
);
625 fr
.romd_mode
= mr
->romd_mode
;
626 fr
.readonly
= readonly
;
627 fr
.nonvolatile
= nonvolatile
;
629 /* Render the region itself into any gaps left by the current view. */
630 for (i
= 0; i
< view
->nr
&& int128_nz(remain
); ++i
) {
631 if (int128_ge(base
, addrrange_end(view
->ranges
[i
].addr
))) {
634 if (int128_lt(base
, view
->ranges
[i
].addr
.start
)) {
635 now
= int128_min(remain
,
636 int128_sub(view
->ranges
[i
].addr
.start
, base
));
637 fr
.offset_in_region
= offset_in_region
;
638 fr
.addr
= addrrange_make(base
, now
);
639 flatview_insert(view
, i
, &fr
);
641 int128_addto(&base
, now
);
642 offset_in_region
+= int128_get64(now
);
643 int128_subfrom(&remain
, now
);
645 now
= int128_sub(int128_min(int128_add(base
, remain
),
646 addrrange_end(view
->ranges
[i
].addr
)),
648 int128_addto(&base
, now
);
649 offset_in_region
+= int128_get64(now
);
650 int128_subfrom(&remain
, now
);
652 if (int128_nz(remain
)) {
653 fr
.offset_in_region
= offset_in_region
;
654 fr
.addr
= addrrange_make(base
, remain
);
655 flatview_insert(view
, i
, &fr
);
659 void flatview_for_each_range(FlatView
*fv
, flatview_cb cb
, void *opaque
)
666 FOR_EACH_FLAT_RANGE(fr
, fv
) {
667 if (cb(fr
->addr
.start
, fr
->addr
.size
, fr
->mr
, opaque
))
672 static MemoryRegion
*memory_region_get_flatview_root(MemoryRegion
*mr
)
674 while (mr
->enabled
) {
676 if (!mr
->alias_offset
&& int128_ge(mr
->size
, mr
->alias
->size
)) {
677 /* The alias is included in its entirety. Use it as
678 * the "real" root, so that we can share more FlatViews.
683 } else if (!mr
->terminates
) {
684 unsigned int found
= 0;
685 MemoryRegion
*child
, *next
= NULL
;
686 QTAILQ_FOREACH(child
, &mr
->subregions
, subregions_link
) {
687 if (child
->enabled
) {
692 if (!child
->addr
&& int128_ge(mr
->size
, child
->size
)) {
693 /* A child is included in its entirety. If it's the only
694 * enabled one, use it in the hope of finding an alias down the
695 * way. This will also let us share FlatViews.
716 /* Render a memory topology into a list of disjoint absolute ranges. */
717 static FlatView
*generate_memory_topology(MemoryRegion
*mr
)
722 view
= flatview_new(mr
);
725 render_memory_region(view
, mr
, int128_zero(),
726 addrrange_make(int128_zero(), int128_2_64()),
729 flatview_simplify(view
);
731 view
->dispatch
= address_space_dispatch_new(view
);
732 for (i
= 0; i
< view
->nr
; i
++) {
733 MemoryRegionSection mrs
=
734 section_from_flat_range(&view
->ranges
[i
], view
);
735 flatview_add_to_dispatch(view
, &mrs
);
737 address_space_dispatch_compact(view
->dispatch
);
738 g_hash_table_replace(flat_views
, mr
, view
);
743 static void address_space_add_del_ioeventfds(AddressSpace
*as
,
744 MemoryRegionIoeventfd
*fds_new
,
746 MemoryRegionIoeventfd
*fds_old
,
750 MemoryRegionIoeventfd
*fd
;
751 MemoryRegionSection section
;
753 /* Generate a symmetric difference of the old and new fd sets, adding
754 * and deleting as necessary.
758 while (iold
< fds_old_nb
|| inew
< fds_new_nb
) {
759 if (iold
< fds_old_nb
760 && (inew
== fds_new_nb
761 || memory_region_ioeventfd_before(&fds_old
[iold
],
764 section
= (MemoryRegionSection
) {
765 .fv
= address_space_to_flatview(as
),
766 .offset_within_address_space
= int128_get64(fd
->addr
.start
),
767 .size
= fd
->addr
.size
,
769 MEMORY_LISTENER_CALL(as
, eventfd_del
, Forward
, §ion
,
770 fd
->match_data
, fd
->data
, fd
->e
);
772 } else if (inew
< fds_new_nb
773 && (iold
== fds_old_nb
774 || memory_region_ioeventfd_before(&fds_new
[inew
],
777 section
= (MemoryRegionSection
) {
778 .fv
= address_space_to_flatview(as
),
779 .offset_within_address_space
= int128_get64(fd
->addr
.start
),
780 .size
= fd
->addr
.size
,
782 MEMORY_LISTENER_CALL(as
, eventfd_add
, Reverse
, §ion
,
783 fd
->match_data
, fd
->data
, fd
->e
);
792 FlatView
*address_space_get_flatview(AddressSpace
*as
)
796 RCU_READ_LOCK_GUARD();
798 view
= address_space_to_flatview(as
);
799 /* If somebody has replaced as->current_map concurrently,
800 * flatview_ref returns false.
802 } while (!flatview_ref(view
));
806 static void address_space_update_ioeventfds(AddressSpace
*as
)
810 unsigned ioeventfd_nb
= 0;
811 unsigned ioeventfd_max
;
812 MemoryRegionIoeventfd
*ioeventfds
;
817 * It is likely that the number of ioeventfds hasn't changed much, so use
818 * the previous size as the starting value, with some headroom to avoid
819 * gratuitous reallocations.
821 ioeventfd_max
= QEMU_ALIGN_UP(as
->ioeventfd_nb
, 4);
822 ioeventfds
= g_new(MemoryRegionIoeventfd
, ioeventfd_max
);
824 view
= address_space_get_flatview(as
);
825 FOR_EACH_FLAT_RANGE(fr
, view
) {
826 for (i
= 0; i
< fr
->mr
->ioeventfd_nb
; ++i
) {
827 tmp
= addrrange_shift(fr
->mr
->ioeventfds
[i
].addr
,
828 int128_sub(fr
->addr
.start
,
829 int128_make64(fr
->offset_in_region
)));
830 if (addrrange_intersects(fr
->addr
, tmp
)) {
832 if (ioeventfd_nb
> ioeventfd_max
) {
833 ioeventfd_max
= MAX(ioeventfd_max
* 2, 4);
834 ioeventfds
= g_realloc(ioeventfds
,
835 ioeventfd_max
* sizeof(*ioeventfds
));
837 ioeventfds
[ioeventfd_nb
-1] = fr
->mr
->ioeventfds
[i
];
838 ioeventfds
[ioeventfd_nb
-1].addr
= tmp
;
843 address_space_add_del_ioeventfds(as
, ioeventfds
, ioeventfd_nb
,
844 as
->ioeventfds
, as
->ioeventfd_nb
);
846 g_free(as
->ioeventfds
);
847 as
->ioeventfds
= ioeventfds
;
848 as
->ioeventfd_nb
= ioeventfd_nb
;
849 flatview_unref(view
);
853 * Notify the memory listeners about the coalesced IO change events of
854 * range `cmr'. Only the part that has intersection of the specified
855 * FlatRange will be sent.
857 static void flat_range_coalesced_io_notify(FlatRange
*fr
, AddressSpace
*as
,
858 CoalescedMemoryRange
*cmr
, bool add
)
862 tmp
= addrrange_shift(cmr
->addr
,
863 int128_sub(fr
->addr
.start
,
864 int128_make64(fr
->offset_in_region
)));
865 if (!addrrange_intersects(tmp
, fr
->addr
)) {
868 tmp
= addrrange_intersection(tmp
, fr
->addr
);
871 MEMORY_LISTENER_UPDATE_REGION(fr
, as
, Forward
, coalesced_io_add
,
872 int128_get64(tmp
.start
),
873 int128_get64(tmp
.size
));
875 MEMORY_LISTENER_UPDATE_REGION(fr
, as
, Reverse
, coalesced_io_del
,
876 int128_get64(tmp
.start
),
877 int128_get64(tmp
.size
));
881 static void flat_range_coalesced_io_del(FlatRange
*fr
, AddressSpace
*as
)
883 CoalescedMemoryRange
*cmr
;
885 QTAILQ_FOREACH(cmr
, &fr
->mr
->coalesced
, link
) {
886 flat_range_coalesced_io_notify(fr
, as
, cmr
, false);
890 static void flat_range_coalesced_io_add(FlatRange
*fr
, AddressSpace
*as
)
892 MemoryRegion
*mr
= fr
->mr
;
893 CoalescedMemoryRange
*cmr
;
895 if (QTAILQ_EMPTY(&mr
->coalesced
)) {
899 QTAILQ_FOREACH(cmr
, &mr
->coalesced
, link
) {
900 flat_range_coalesced_io_notify(fr
, as
, cmr
, true);
904 static void address_space_update_topology_pass(AddressSpace
*as
,
905 const FlatView
*old_view
,
906 const FlatView
*new_view
,
910 FlatRange
*frold
, *frnew
;
912 /* Generate a symmetric difference of the old and new memory maps.
913 * Kill ranges in the old map, and instantiate ranges in the new map.
916 while (iold
< old_view
->nr
|| inew
< new_view
->nr
) {
917 if (iold
< old_view
->nr
) {
918 frold
= &old_view
->ranges
[iold
];
922 if (inew
< new_view
->nr
) {
923 frnew
= &new_view
->ranges
[inew
];
930 || int128_lt(frold
->addr
.start
, frnew
->addr
.start
)
931 || (int128_eq(frold
->addr
.start
, frnew
->addr
.start
)
932 && !flatrange_equal(frold
, frnew
)))) {
933 /* In old but not in new, or in both but attributes changed. */
936 flat_range_coalesced_io_del(frold
, as
);
937 MEMORY_LISTENER_UPDATE_REGION(frold
, as
, Reverse
, region_del
);
941 } else if (frold
&& frnew
&& flatrange_equal(frold
, frnew
)) {
942 /* In both and unchanged (except logging may have changed) */
945 MEMORY_LISTENER_UPDATE_REGION(frnew
, as
, Forward
, region_nop
);
946 if (frnew
->dirty_log_mask
& ~frold
->dirty_log_mask
) {
947 MEMORY_LISTENER_UPDATE_REGION(frnew
, as
, Forward
, log_start
,
948 frold
->dirty_log_mask
,
949 frnew
->dirty_log_mask
);
951 if (frold
->dirty_log_mask
& ~frnew
->dirty_log_mask
) {
952 MEMORY_LISTENER_UPDATE_REGION(frnew
, as
, Reverse
, log_stop
,
953 frold
->dirty_log_mask
,
954 frnew
->dirty_log_mask
);
964 MEMORY_LISTENER_UPDATE_REGION(frnew
, as
, Forward
, region_add
);
965 flat_range_coalesced_io_add(frnew
, as
);
973 static void flatviews_init(void)
975 static FlatView
*empty_view
;
981 flat_views
= g_hash_table_new_full(g_direct_hash
, g_direct_equal
, NULL
,
982 (GDestroyNotify
) flatview_unref
);
984 empty_view
= generate_memory_topology(NULL
);
985 /* We keep it alive forever in the global variable. */
986 flatview_ref(empty_view
);
988 g_hash_table_replace(flat_views
, NULL
, empty_view
);
989 flatview_ref(empty_view
);
993 static void flatviews_reset(void)
998 g_hash_table_unref(flat_views
);
1003 /* Render unique FVs */
1004 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
1005 MemoryRegion
*physmr
= memory_region_get_flatview_root(as
->root
);
1007 if (g_hash_table_lookup(flat_views
, physmr
)) {
1011 generate_memory_topology(physmr
);
1015 static void address_space_set_flatview(AddressSpace
*as
)
1017 FlatView
*old_view
= address_space_to_flatview(as
);
1018 MemoryRegion
*physmr
= memory_region_get_flatview_root(as
->root
);
1019 FlatView
*new_view
= g_hash_table_lookup(flat_views
, physmr
);
1023 if (old_view
== new_view
) {
1028 flatview_ref(old_view
);
1031 flatview_ref(new_view
);
1033 if (!QTAILQ_EMPTY(&as
->listeners
)) {
1034 FlatView tmpview
= { .nr
= 0 }, *old_view2
= old_view
;
1037 old_view2
= &tmpview
;
1039 address_space_update_topology_pass(as
, old_view2
, new_view
, false);
1040 address_space_update_topology_pass(as
, old_view2
, new_view
, true);
1043 /* Writes are protected by the BQL. */
1044 qatomic_rcu_set(&as
->current_map
, new_view
);
1046 flatview_unref(old_view
);
1049 /* Note that all the old MemoryRegions are still alive up to this
1050 * point. This relieves most MemoryListeners from the need to
1051 * ref/unref the MemoryRegions they get---unless they use them
1052 * outside the iothread mutex, in which case precise reference
1053 * counting is necessary.
1056 flatview_unref(old_view
);
1060 static void address_space_update_topology(AddressSpace
*as
)
1062 MemoryRegion
*physmr
= memory_region_get_flatview_root(as
->root
);
1065 if (!g_hash_table_lookup(flat_views
, physmr
)) {
1066 generate_memory_topology(physmr
);
1068 address_space_set_flatview(as
);
1071 void memory_region_transaction_begin(void)
1073 qemu_flush_coalesced_mmio_buffer();
1074 ++memory_region_transaction_depth
;
1077 void memory_region_transaction_commit(void)
1081 assert(memory_region_transaction_depth
);
1082 assert(qemu_mutex_iothread_locked());
1084 --memory_region_transaction_depth
;
1085 if (!memory_region_transaction_depth
) {
1086 if (memory_region_update_pending
) {
1089 MEMORY_LISTENER_CALL_GLOBAL(begin
, Forward
);
1091 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
1092 address_space_set_flatview(as
);
1093 address_space_update_ioeventfds(as
);
1095 memory_region_update_pending
= false;
1096 ioeventfd_update_pending
= false;
1097 MEMORY_LISTENER_CALL_GLOBAL(commit
, Forward
);
1098 } else if (ioeventfd_update_pending
) {
1099 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
1100 address_space_update_ioeventfds(as
);
1102 ioeventfd_update_pending
= false;
1107 static void memory_region_destructor_none(MemoryRegion
*mr
)
1111 static void memory_region_destructor_ram(MemoryRegion
*mr
)
1113 qemu_ram_free(mr
->ram_block
);
1116 static bool memory_region_need_escape(char c
)
1118 return c
== '/' || c
== '[' || c
== '\\' || c
== ']';
1121 static char *memory_region_escape_name(const char *name
)
1128 for (p
= name
; *p
; p
++) {
1129 bytes
+= memory_region_need_escape(*p
) ? 4 : 1;
1131 if (bytes
== p
- name
) {
1132 return g_memdup(name
, bytes
+ 1);
1135 escaped
= g_malloc(bytes
+ 1);
1136 for (p
= name
, q
= escaped
; *p
; p
++) {
1138 if (unlikely(memory_region_need_escape(c
))) {
1141 *q
++ = "0123456789abcdef"[c
>> 4];
1142 c
= "0123456789abcdef"[c
& 15];
1150 static void memory_region_do_init(MemoryRegion
*mr
,
1155 mr
->size
= int128_make64(size
);
1156 if (size
== UINT64_MAX
) {
1157 mr
->size
= int128_2_64();
1159 mr
->name
= g_strdup(name
);
1161 mr
->ram_block
= NULL
;
1164 char *escaped_name
= memory_region_escape_name(name
);
1165 char *name_array
= g_strdup_printf("%s[*]", escaped_name
);
1168 owner
= container_get(qdev_get_machine(), "/unattached");
1171 object_property_add_child(owner
, name_array
, OBJECT(mr
));
1172 object_unref(OBJECT(mr
));
1174 g_free(escaped_name
);
1178 void memory_region_init(MemoryRegion
*mr
,
1183 object_initialize(mr
, sizeof(*mr
), TYPE_MEMORY_REGION
);
1184 memory_region_do_init(mr
, owner
, name
, size
);
1187 static void memory_region_get_container(Object
*obj
, Visitor
*v
,
1188 const char *name
, void *opaque
,
1191 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1192 char *path
= (char *)"";
1194 if (mr
->container
) {
1195 path
= object_get_canonical_path(OBJECT(mr
->container
));
1197 visit_type_str(v
, name
, &path
, errp
);
1198 if (mr
->container
) {
1203 static Object
*memory_region_resolve_container(Object
*obj
, void *opaque
,
1206 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1208 return OBJECT(mr
->container
);
1211 static void memory_region_get_priority(Object
*obj
, Visitor
*v
,
1212 const char *name
, void *opaque
,
1215 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1216 int32_t value
= mr
->priority
;
1218 visit_type_int32(v
, name
, &value
, errp
);
1221 static void memory_region_get_size(Object
*obj
, Visitor
*v
, const char *name
,
1222 void *opaque
, Error
**errp
)
1224 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1225 uint64_t value
= memory_region_size(mr
);
1227 visit_type_uint64(v
, name
, &value
, errp
);
1230 static void memory_region_initfn(Object
*obj
)
1232 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1235 mr
->ops
= &unassigned_mem_ops
;
1237 mr
->romd_mode
= true;
1238 mr
->destructor
= memory_region_destructor_none
;
1239 QTAILQ_INIT(&mr
->subregions
);
1240 QTAILQ_INIT(&mr
->coalesced
);
1242 op
= object_property_add(OBJECT(mr
), "container",
1243 "link<" TYPE_MEMORY_REGION
">",
1244 memory_region_get_container
,
1245 NULL
, /* memory_region_set_container */
1247 op
->resolve
= memory_region_resolve_container
;
1249 object_property_add_uint64_ptr(OBJECT(mr
), "addr",
1250 &mr
->addr
, OBJ_PROP_FLAG_READ
);
1251 object_property_add(OBJECT(mr
), "priority", "uint32",
1252 memory_region_get_priority
,
1253 NULL
, /* memory_region_set_priority */
1255 object_property_add(OBJECT(mr
), "size", "uint64",
1256 memory_region_get_size
,
1257 NULL
, /* memory_region_set_size, */
1261 static void iommu_memory_region_initfn(Object
*obj
)
1263 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1265 mr
->is_iommu
= true;
1268 static uint64_t unassigned_mem_read(void *opaque
, hwaddr addr
,
1271 #ifdef DEBUG_UNASSIGNED
1272 printf("Unassigned mem read " TARGET_FMT_plx
"\n", addr
);
1277 static void unassigned_mem_write(void *opaque
, hwaddr addr
,
1278 uint64_t val
, unsigned size
)
1280 #ifdef DEBUG_UNASSIGNED
1281 printf("Unassigned mem write " TARGET_FMT_plx
" = 0x%"PRIx64
"\n", addr
, val
);
1285 static bool unassigned_mem_accepts(void *opaque
, hwaddr addr
,
1286 unsigned size
, bool is_write
,
1292 const MemoryRegionOps unassigned_mem_ops
= {
1293 .valid
.accepts
= unassigned_mem_accepts
,
1294 .endianness
= DEVICE_NATIVE_ENDIAN
,
1297 static uint64_t memory_region_ram_device_read(void *opaque
,
1298 hwaddr addr
, unsigned size
)
1300 MemoryRegion
*mr
= opaque
;
1301 uint64_t data
= (uint64_t)~0;
1305 data
= *(uint8_t *)(mr
->ram_block
->host
+ addr
);
1308 data
= *(uint16_t *)(mr
->ram_block
->host
+ addr
);
1311 data
= *(uint32_t *)(mr
->ram_block
->host
+ addr
);
1314 data
= *(uint64_t *)(mr
->ram_block
->host
+ addr
);
1318 trace_memory_region_ram_device_read(get_cpu_index(), mr
, addr
, data
, size
);
1323 static void memory_region_ram_device_write(void *opaque
, hwaddr addr
,
1324 uint64_t data
, unsigned size
)
1326 MemoryRegion
*mr
= opaque
;
1328 trace_memory_region_ram_device_write(get_cpu_index(), mr
, addr
, data
, size
);
1332 *(uint8_t *)(mr
->ram_block
->host
+ addr
) = (uint8_t)data
;
1335 *(uint16_t *)(mr
->ram_block
->host
+ addr
) = (uint16_t)data
;
1338 *(uint32_t *)(mr
->ram_block
->host
+ addr
) = (uint32_t)data
;
1341 *(uint64_t *)(mr
->ram_block
->host
+ addr
) = data
;
1346 static const MemoryRegionOps ram_device_mem_ops
= {
1347 .read
= memory_region_ram_device_read
,
1348 .write
= memory_region_ram_device_write
,
1349 .endianness
= DEVICE_HOST_ENDIAN
,
1351 .min_access_size
= 1,
1352 .max_access_size
= 8,
1356 .min_access_size
= 1,
1357 .max_access_size
= 8,
1362 bool memory_region_access_valid(MemoryRegion
*mr
,
1368 if (mr
->ops
->valid
.accepts
1369 && !mr
->ops
->valid
.accepts(mr
->opaque
, addr
, size
, is_write
, attrs
)) {
1370 qemu_log_mask(LOG_GUEST_ERROR
, "Invalid access at addr "
1371 "0x%" HWADDR_PRIX
", size %u, "
1372 "region '%s', reason: rejected\n",
1373 addr
, size
, memory_region_name(mr
));
1377 if (!mr
->ops
->valid
.unaligned
&& (addr
& (size
- 1))) {
1378 qemu_log_mask(LOG_GUEST_ERROR
, "Invalid access at addr "
1379 "0x%" HWADDR_PRIX
", size %u, "
1380 "region '%s', reason: unaligned\n",
1381 addr
, size
, memory_region_name(mr
));
1385 /* Treat zero as compatibility all valid */
1386 if (!mr
->ops
->valid
.max_access_size
) {
1390 if (size
> mr
->ops
->valid
.max_access_size
1391 || size
< mr
->ops
->valid
.min_access_size
) {
1392 qemu_log_mask(LOG_GUEST_ERROR
, "Invalid access at addr "
1393 "0x%" HWADDR_PRIX
", size %u, "
1394 "region '%s', reason: invalid size "
1395 "(min:%u max:%u)\n",
1396 addr
, size
, memory_region_name(mr
),
1397 mr
->ops
->valid
.min_access_size
,
1398 mr
->ops
->valid
.max_access_size
);
1404 static MemTxResult
memory_region_dispatch_read1(MemoryRegion
*mr
,
1412 if (mr
->ops
->read
) {
1413 return access_with_adjusted_size(addr
, pval
, size
,
1414 mr
->ops
->impl
.min_access_size
,
1415 mr
->ops
->impl
.max_access_size
,
1416 memory_region_read_accessor
,
1419 return access_with_adjusted_size(addr
, pval
, size
,
1420 mr
->ops
->impl
.min_access_size
,
1421 mr
->ops
->impl
.max_access_size
,
1422 memory_region_read_with_attrs_accessor
,
1427 MemTxResult
memory_region_dispatch_read(MemoryRegion
*mr
,
1433 unsigned size
= memop_size(op
);
1436 fuzz_dma_read_cb(addr
, size
, mr
, false);
1437 if (!memory_region_access_valid(mr
, addr
, size
, false, attrs
)) {
1438 *pval
= unassigned_mem_read(mr
, addr
, size
);
1439 return MEMTX_DECODE_ERROR
;
1442 r
= memory_region_dispatch_read1(mr
, addr
, pval
, size
, attrs
);
1443 adjust_endianness(mr
, pval
, op
);
1447 /* Return true if an eventfd was signalled */
1448 static bool memory_region_dispatch_write_eventfds(MemoryRegion
*mr
,
1454 MemoryRegionIoeventfd ioeventfd
= {
1455 .addr
= addrrange_make(int128_make64(addr
), int128_make64(size
)),
1460 for (i
= 0; i
< mr
->ioeventfd_nb
; i
++) {
1461 ioeventfd
.match_data
= mr
->ioeventfds
[i
].match_data
;
1462 ioeventfd
.e
= mr
->ioeventfds
[i
].e
;
1464 if (memory_region_ioeventfd_equal(&ioeventfd
, &mr
->ioeventfds
[i
])) {
1465 event_notifier_set(ioeventfd
.e
);
1473 MemTxResult
memory_region_dispatch_write(MemoryRegion
*mr
,
1479 unsigned size
= memop_size(op
);
1481 if (!memory_region_access_valid(mr
, addr
, size
, true, attrs
)) {
1482 unassigned_mem_write(mr
, addr
, data
, size
);
1483 return MEMTX_DECODE_ERROR
;
1486 adjust_endianness(mr
, &data
, op
);
1488 if ((!kvm_eventfds_enabled()) &&
1489 memory_region_dispatch_write_eventfds(mr
, addr
, data
, size
, attrs
)) {
1493 if (mr
->ops
->write
) {
1494 return access_with_adjusted_size(addr
, &data
, size
,
1495 mr
->ops
->impl
.min_access_size
,
1496 mr
->ops
->impl
.max_access_size
,
1497 memory_region_write_accessor
, mr
,
1501 access_with_adjusted_size(addr
, &data
, size
,
1502 mr
->ops
->impl
.min_access_size
,
1503 mr
->ops
->impl
.max_access_size
,
1504 memory_region_write_with_attrs_accessor
,
1509 void memory_region_init_io(MemoryRegion
*mr
,
1511 const MemoryRegionOps
*ops
,
1516 memory_region_init(mr
, owner
, name
, size
);
1517 mr
->ops
= ops
? ops
: &unassigned_mem_ops
;
1518 mr
->opaque
= opaque
;
1519 mr
->terminates
= true;
1522 void memory_region_init_ram_nomigrate(MemoryRegion
*mr
,
1528 memory_region_init_ram_shared_nomigrate(mr
, owner
, name
, size
, false, errp
);
1531 void memory_region_init_ram_shared_nomigrate(MemoryRegion
*mr
,
1539 memory_region_init(mr
, owner
, name
, size
);
1541 mr
->terminates
= true;
1542 mr
->destructor
= memory_region_destructor_ram
;
1543 mr
->ram_block
= qemu_ram_alloc(size
, share
, mr
, &err
);
1544 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1546 mr
->size
= int128_zero();
1547 object_unparent(OBJECT(mr
));
1548 error_propagate(errp
, err
);
1552 void memory_region_init_resizeable_ram(MemoryRegion
*mr
,
1557 void (*resized
)(const char*,
1563 memory_region_init(mr
, owner
, name
, size
);
1565 mr
->terminates
= true;
1566 mr
->destructor
= memory_region_destructor_ram
;
1567 mr
->ram_block
= qemu_ram_alloc_resizeable(size
, max_size
, resized
,
1569 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1571 mr
->size
= int128_zero();
1572 object_unparent(OBJECT(mr
));
1573 error_propagate(errp
, err
);
1578 void memory_region_init_ram_from_file(MemoryRegion
*mr
,
1579 struct Object
*owner
,
1588 memory_region_init(mr
, owner
, name
, size
);
1590 mr
->terminates
= true;
1591 mr
->destructor
= memory_region_destructor_ram
;
1593 mr
->ram_block
= qemu_ram_alloc_from_file(size
, mr
, ram_flags
, path
, &err
);
1594 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1596 mr
->size
= int128_zero();
1597 object_unparent(OBJECT(mr
));
1598 error_propagate(errp
, err
);
1602 void memory_region_init_ram_from_fd(MemoryRegion
*mr
,
1603 struct Object
*owner
,
1611 memory_region_init(mr
, owner
, name
, size
);
1613 mr
->terminates
= true;
1614 mr
->destructor
= memory_region_destructor_ram
;
1615 mr
->ram_block
= qemu_ram_alloc_from_fd(size
, mr
,
1616 share
? RAM_SHARED
: 0,
1618 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1620 mr
->size
= int128_zero();
1621 object_unparent(OBJECT(mr
));
1622 error_propagate(errp
, err
);
1627 void memory_region_init_ram_ptr(MemoryRegion
*mr
,
1633 memory_region_init(mr
, owner
, name
, size
);
1635 mr
->terminates
= true;
1636 mr
->destructor
= memory_region_destructor_ram
;
1637 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1639 /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL. */
1640 assert(ptr
!= NULL
);
1641 mr
->ram_block
= qemu_ram_alloc_from_ptr(size
, ptr
, mr
, &error_fatal
);
1644 void memory_region_init_ram_device_ptr(MemoryRegion
*mr
,
1650 memory_region_init(mr
, owner
, name
, size
);
1652 mr
->terminates
= true;
1653 mr
->ram_device
= true;
1654 mr
->ops
= &ram_device_mem_ops
;
1656 mr
->destructor
= memory_region_destructor_ram
;
1657 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1658 /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL. */
1659 assert(ptr
!= NULL
);
1660 mr
->ram_block
= qemu_ram_alloc_from_ptr(size
, ptr
, mr
, &error_fatal
);
1663 void memory_region_init_alias(MemoryRegion
*mr
,
1670 memory_region_init(mr
, owner
, name
, size
);
1672 mr
->alias_offset
= offset
;
1675 void memory_region_init_rom_nomigrate(MemoryRegion
*mr
,
1676 struct Object
*owner
,
1681 memory_region_init_ram_shared_nomigrate(mr
, owner
, name
, size
, false, errp
);
1682 mr
->readonly
= true;
1685 void memory_region_init_rom_device_nomigrate(MemoryRegion
*mr
,
1687 const MemoryRegionOps
*ops
,
1695 memory_region_init(mr
, owner
, name
, size
);
1697 mr
->opaque
= opaque
;
1698 mr
->terminates
= true;
1699 mr
->rom_device
= true;
1700 mr
->destructor
= memory_region_destructor_ram
;
1701 mr
->ram_block
= qemu_ram_alloc(size
, false, mr
, &err
);
1703 mr
->size
= int128_zero();
1704 object_unparent(OBJECT(mr
));
1705 error_propagate(errp
, err
);
1709 void memory_region_init_iommu(void *_iommu_mr
,
1710 size_t instance_size
,
1711 const char *mrtypename
,
1716 struct IOMMUMemoryRegion
*iommu_mr
;
1717 struct MemoryRegion
*mr
;
1719 object_initialize(_iommu_mr
, instance_size
, mrtypename
);
1720 mr
= MEMORY_REGION(_iommu_mr
);
1721 memory_region_do_init(mr
, owner
, name
, size
);
1722 iommu_mr
= IOMMU_MEMORY_REGION(mr
);
1723 mr
->terminates
= true; /* then re-forwards */
1724 QLIST_INIT(&iommu_mr
->iommu_notify
);
1725 iommu_mr
->iommu_notify_flags
= IOMMU_NOTIFIER_NONE
;
1728 static void memory_region_finalize(Object
*obj
)
1730 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1732 assert(!mr
->container
);
1734 /* We know the region is not visible in any address space (it
1735 * does not have a container and cannot be a root either because
1736 * it has no references, so we can blindly clear mr->enabled.
1737 * memory_region_set_enabled instead could trigger a transaction
1738 * and cause an infinite loop.
1740 mr
->enabled
= false;
1741 memory_region_transaction_begin();
1742 while (!QTAILQ_EMPTY(&mr
->subregions
)) {
1743 MemoryRegion
*subregion
= QTAILQ_FIRST(&mr
->subregions
);
1744 memory_region_del_subregion(mr
, subregion
);
1746 memory_region_transaction_commit();
1749 memory_region_clear_coalescing(mr
);
1750 g_free((char *)mr
->name
);
1751 g_free(mr
->ioeventfds
);
1754 Object
*memory_region_owner(MemoryRegion
*mr
)
1756 Object
*obj
= OBJECT(mr
);
1760 void memory_region_ref(MemoryRegion
*mr
)
1762 /* MMIO callbacks most likely will access data that belongs
1763 * to the owner, hence the need to ref/unref the owner whenever
1764 * the memory region is in use.
1766 * The memory region is a child of its owner. As long as the
1767 * owner doesn't call unparent itself on the memory region,
1768 * ref-ing the owner will also keep the memory region alive.
1769 * Memory regions without an owner are supposed to never go away;
1770 * we do not ref/unref them because it slows down DMA sensibly.
1772 if (mr
&& mr
->owner
) {
1773 object_ref(mr
->owner
);
1777 void memory_region_unref(MemoryRegion
*mr
)
1779 if (mr
&& mr
->owner
) {
1780 object_unref(mr
->owner
);
1784 uint64_t memory_region_size(MemoryRegion
*mr
)
1786 if (int128_eq(mr
->size
, int128_2_64())) {
1789 return int128_get64(mr
->size
);
1792 const char *memory_region_name(const MemoryRegion
*mr
)
1795 ((MemoryRegion
*)mr
)->name
=
1796 g_strdup(object_get_canonical_path_component(OBJECT(mr
)));
1801 bool memory_region_is_ram_device(MemoryRegion
*mr
)
1803 return mr
->ram_device
;
1806 uint8_t memory_region_get_dirty_log_mask(MemoryRegion
*mr
)
1808 uint8_t mask
= mr
->dirty_log_mask
;
1809 RAMBlock
*rb
= mr
->ram_block
;
1811 if (global_dirty_log
&& ((rb
&& qemu_ram_is_migratable(rb
)) ||
1812 memory_region_is_iommu(mr
))) {
1813 mask
|= (1 << DIRTY_MEMORY_MIGRATION
);
1818 bool memory_region_is_logging(MemoryRegion
*mr
, uint8_t client
)
1820 return memory_region_get_dirty_log_mask(mr
) & (1 << client
);
1823 static int memory_region_update_iommu_notify_flags(IOMMUMemoryRegion
*iommu_mr
,
1826 IOMMUNotifierFlag flags
= IOMMU_NOTIFIER_NONE
;
1827 IOMMUNotifier
*iommu_notifier
;
1828 IOMMUMemoryRegionClass
*imrc
= IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr
);
1831 IOMMU_NOTIFIER_FOREACH(iommu_notifier
, iommu_mr
) {
1832 flags
|= iommu_notifier
->notifier_flags
;
1835 if (flags
!= iommu_mr
->iommu_notify_flags
&& imrc
->notify_flag_changed
) {
1836 ret
= imrc
->notify_flag_changed(iommu_mr
,
1837 iommu_mr
->iommu_notify_flags
,
1842 iommu_mr
->iommu_notify_flags
= flags
;
1847 int memory_region_iommu_set_page_size_mask(IOMMUMemoryRegion
*iommu_mr
,
1848 uint64_t page_size_mask
,
1851 IOMMUMemoryRegionClass
*imrc
= IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr
);
1854 if (imrc
->iommu_set_page_size_mask
) {
1855 ret
= imrc
->iommu_set_page_size_mask(iommu_mr
, page_size_mask
, errp
);
1860 int memory_region_register_iommu_notifier(MemoryRegion
*mr
,
1861 IOMMUNotifier
*n
, Error
**errp
)
1863 IOMMUMemoryRegion
*iommu_mr
;
1867 return memory_region_register_iommu_notifier(mr
->alias
, n
, errp
);
1870 /* We need to register for at least one bitfield */
1871 iommu_mr
= IOMMU_MEMORY_REGION(mr
);
1872 assert(n
->notifier_flags
!= IOMMU_NOTIFIER_NONE
);
1873 assert(n
->start
<= n
->end
);
1874 assert(n
->iommu_idx
>= 0 &&
1875 n
->iommu_idx
< memory_region_iommu_num_indexes(iommu_mr
));
1877 QLIST_INSERT_HEAD(&iommu_mr
->iommu_notify
, n
, node
);
1878 ret
= memory_region_update_iommu_notify_flags(iommu_mr
, errp
);
1880 QLIST_REMOVE(n
, node
);
1885 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion
*iommu_mr
)
1887 IOMMUMemoryRegionClass
*imrc
= IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr
);
1889 if (imrc
->get_min_page_size
) {
1890 return imrc
->get_min_page_size(iommu_mr
);
1892 return TARGET_PAGE_SIZE
;
1895 void memory_region_iommu_replay(IOMMUMemoryRegion
*iommu_mr
, IOMMUNotifier
*n
)
1897 MemoryRegion
*mr
= MEMORY_REGION(iommu_mr
);
1898 IOMMUMemoryRegionClass
*imrc
= IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr
);
1899 hwaddr addr
, granularity
;
1900 IOMMUTLBEntry iotlb
;
1902 /* If the IOMMU has its own replay callback, override */
1904 imrc
->replay(iommu_mr
, n
);
1908 granularity
= memory_region_iommu_get_min_page_size(iommu_mr
);
1910 for (addr
= 0; addr
< memory_region_size(mr
); addr
+= granularity
) {
1911 iotlb
= imrc
->translate(iommu_mr
, addr
, IOMMU_NONE
, n
->iommu_idx
);
1912 if (iotlb
.perm
!= IOMMU_NONE
) {
1913 n
->notify(n
, &iotlb
);
1916 /* if (2^64 - MR size) < granularity, it's possible to get an
1917 * infinite loop here. This should catch such a wraparound */
1918 if ((addr
+ granularity
) < addr
) {
1924 void memory_region_unregister_iommu_notifier(MemoryRegion
*mr
,
1927 IOMMUMemoryRegion
*iommu_mr
;
1930 memory_region_unregister_iommu_notifier(mr
->alias
, n
);
1933 QLIST_REMOVE(n
, node
);
1934 iommu_mr
= IOMMU_MEMORY_REGION(mr
);
1935 memory_region_update_iommu_notify_flags(iommu_mr
, NULL
);
1938 void memory_region_notify_one(IOMMUNotifier
*notifier
,
1939 IOMMUTLBEntry
*entry
)
1941 IOMMUNotifierFlag request_flags
;
1942 hwaddr entry_end
= entry
->iova
+ entry
->addr_mask
;
1945 * Skip the notification if the notification does not overlap
1946 * with registered range.
1948 if (notifier
->start
> entry_end
|| notifier
->end
< entry
->iova
) {
1952 assert(entry
->iova
>= notifier
->start
&& entry_end
<= notifier
->end
);
1954 if (entry
->perm
& IOMMU_RW
) {
1955 request_flags
= IOMMU_NOTIFIER_MAP
;
1957 request_flags
= IOMMU_NOTIFIER_UNMAP
;
1960 if (notifier
->notifier_flags
& request_flags
) {
1961 notifier
->notify(notifier
, entry
);
1965 void memory_region_notify_iommu(IOMMUMemoryRegion
*iommu_mr
,
1967 IOMMUTLBEntry entry
)
1969 IOMMUNotifier
*iommu_notifier
;
1971 assert(memory_region_is_iommu(MEMORY_REGION(iommu_mr
)));
1973 IOMMU_NOTIFIER_FOREACH(iommu_notifier
, iommu_mr
) {
1974 if (iommu_notifier
->iommu_idx
== iommu_idx
) {
1975 memory_region_notify_one(iommu_notifier
, &entry
);
1980 int memory_region_iommu_get_attr(IOMMUMemoryRegion
*iommu_mr
,
1981 enum IOMMUMemoryRegionAttr attr
,
1984 IOMMUMemoryRegionClass
*imrc
= IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr
);
1986 if (!imrc
->get_attr
) {
1990 return imrc
->get_attr(iommu_mr
, attr
, data
);
1993 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion
*iommu_mr
,
1996 IOMMUMemoryRegionClass
*imrc
= IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr
);
1998 if (!imrc
->attrs_to_index
) {
2002 return imrc
->attrs_to_index(iommu_mr
, attrs
);
2005 int memory_region_iommu_num_indexes(IOMMUMemoryRegion
*iommu_mr
)
2007 IOMMUMemoryRegionClass
*imrc
= IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr
);
2009 if (!imrc
->num_indexes
) {
2013 return imrc
->num_indexes(iommu_mr
);
2016 void memory_region_set_log(MemoryRegion
*mr
, bool log
, unsigned client
)
2018 uint8_t mask
= 1 << client
;
2019 uint8_t old_logging
;
2021 assert(client
== DIRTY_MEMORY_VGA
);
2022 old_logging
= mr
->vga_logging_count
;
2023 mr
->vga_logging_count
+= log
? 1 : -1;
2024 if (!!old_logging
== !!mr
->vga_logging_count
) {
2028 memory_region_transaction_begin();
2029 mr
->dirty_log_mask
= (mr
->dirty_log_mask
& ~mask
) | (log
* mask
);
2030 memory_region_update_pending
|= mr
->enabled
;
2031 memory_region_transaction_commit();
2034 void memory_region_set_dirty(MemoryRegion
*mr
, hwaddr addr
,
2037 assert(mr
->ram_block
);
2038 cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr
) + addr
,
2040 memory_region_get_dirty_log_mask(mr
));
2043 static void memory_region_sync_dirty_bitmap(MemoryRegion
*mr
)
2045 MemoryListener
*listener
;
2050 /* If the same address space has multiple log_sync listeners, we
2051 * visit that address space's FlatView multiple times. But because
2052 * log_sync listeners are rare, it's still cheaper than walking each
2053 * address space once.
2055 QTAILQ_FOREACH(listener
, &memory_listeners
, link
) {
2056 if (!listener
->log_sync
) {
2059 as
= listener
->address_space
;
2060 view
= address_space_get_flatview(as
);
2061 FOR_EACH_FLAT_RANGE(fr
, view
) {
2062 if (fr
->dirty_log_mask
&& (!mr
|| fr
->mr
== mr
)) {
2063 MemoryRegionSection mrs
= section_from_flat_range(fr
, view
);
2064 listener
->log_sync(listener
, &mrs
);
2067 flatview_unref(view
);
2071 void memory_region_clear_dirty_bitmap(MemoryRegion
*mr
, hwaddr start
,
2074 MemoryRegionSection mrs
;
2075 MemoryListener
*listener
;
2079 hwaddr sec_start
, sec_end
, sec_size
;
2081 QTAILQ_FOREACH(listener
, &memory_listeners
, link
) {
2082 if (!listener
->log_clear
) {
2085 as
= listener
->address_space
;
2086 view
= address_space_get_flatview(as
);
2087 FOR_EACH_FLAT_RANGE(fr
, view
) {
2088 if (!fr
->dirty_log_mask
|| fr
->mr
!= mr
) {
2090 * Clear dirty bitmap operation only applies to those
2091 * regions whose dirty logging is at least enabled
2096 mrs
= section_from_flat_range(fr
, view
);
2098 sec_start
= MAX(mrs
.offset_within_region
, start
);
2099 sec_end
= mrs
.offset_within_region
+ int128_get64(mrs
.size
);
2100 sec_end
= MIN(sec_end
, start
+ len
);
2102 if (sec_start
>= sec_end
) {
2104 * If this memory region section has no intersection
2105 * with the requested range, skip.
2110 /* Valid case; shrink the section if needed */
2111 mrs
.offset_within_address_space
+=
2112 sec_start
- mrs
.offset_within_region
;
2113 mrs
.offset_within_region
= sec_start
;
2114 sec_size
= sec_end
- sec_start
;
2115 mrs
.size
= int128_make64(sec_size
);
2116 listener
->log_clear(listener
, &mrs
);
2118 flatview_unref(view
);
2122 DirtyBitmapSnapshot
*memory_region_snapshot_and_clear_dirty(MemoryRegion
*mr
,
2127 DirtyBitmapSnapshot
*snapshot
;
2128 assert(mr
->ram_block
);
2129 memory_region_sync_dirty_bitmap(mr
);
2130 snapshot
= cpu_physical_memory_snapshot_and_clear_dirty(mr
, addr
, size
, client
);
2131 memory_global_after_dirty_log_sync();
2135 bool memory_region_snapshot_get_dirty(MemoryRegion
*mr
, DirtyBitmapSnapshot
*snap
,
2136 hwaddr addr
, hwaddr size
)
2138 assert(mr
->ram_block
);
2139 return cpu_physical_memory_snapshot_get_dirty(snap
,
2140 memory_region_get_ram_addr(mr
) + addr
, size
);
2143 void memory_region_set_readonly(MemoryRegion
*mr
, bool readonly
)
2145 if (mr
->readonly
!= readonly
) {
2146 memory_region_transaction_begin();
2147 mr
->readonly
= readonly
;
2148 memory_region_update_pending
|= mr
->enabled
;
2149 memory_region_transaction_commit();
2153 void memory_region_set_nonvolatile(MemoryRegion
*mr
, bool nonvolatile
)
2155 if (mr
->nonvolatile
!= nonvolatile
) {
2156 memory_region_transaction_begin();
2157 mr
->nonvolatile
= nonvolatile
;
2158 memory_region_update_pending
|= mr
->enabled
;
2159 memory_region_transaction_commit();
2163 void memory_region_rom_device_set_romd(MemoryRegion
*mr
, bool romd_mode
)
2165 if (mr
->romd_mode
!= romd_mode
) {
2166 memory_region_transaction_begin();
2167 mr
->romd_mode
= romd_mode
;
2168 memory_region_update_pending
|= mr
->enabled
;
2169 memory_region_transaction_commit();
2173 void memory_region_reset_dirty(MemoryRegion
*mr
, hwaddr addr
,
2174 hwaddr size
, unsigned client
)
2176 assert(mr
->ram_block
);
2177 cpu_physical_memory_test_and_clear_dirty(
2178 memory_region_get_ram_addr(mr
) + addr
, size
, client
);
2181 int memory_region_get_fd(MemoryRegion
*mr
)
2185 RCU_READ_LOCK_GUARD();
2189 fd
= mr
->ram_block
->fd
;
2194 void *memory_region_get_ram_ptr(MemoryRegion
*mr
)
2197 uint64_t offset
= 0;
2199 RCU_READ_LOCK_GUARD();
2201 offset
+= mr
->alias_offset
;
2204 assert(mr
->ram_block
);
2205 ptr
= qemu_map_ram_ptr(mr
->ram_block
, offset
);
2210 MemoryRegion
*memory_region_from_host(void *ptr
, ram_addr_t
*offset
)
2214 block
= qemu_ram_block_from_host(ptr
, false, offset
);
2222 ram_addr_t
memory_region_get_ram_addr(MemoryRegion
*mr
)
2224 return mr
->ram_block
? mr
->ram_block
->offset
: RAM_ADDR_INVALID
;
2227 void memory_region_ram_resize(MemoryRegion
*mr
, ram_addr_t newsize
, Error
**errp
)
2229 assert(mr
->ram_block
);
2231 qemu_ram_resize(mr
->ram_block
, newsize
, errp
);
2234 void memory_region_msync(MemoryRegion
*mr
, hwaddr addr
, hwaddr size
)
2236 if (mr
->ram_block
) {
2237 qemu_ram_msync(mr
->ram_block
, addr
, size
);
2241 void memory_region_writeback(MemoryRegion
*mr
, hwaddr addr
, hwaddr size
)
2244 * Might be extended case needed to cover
2245 * different types of memory regions
2247 if (mr
->dirty_log_mask
) {
2248 memory_region_msync(mr
, addr
, size
);
2253 * Call proper memory listeners about the change on the newly
2254 * added/removed CoalescedMemoryRange.
2256 static void memory_region_update_coalesced_range(MemoryRegion
*mr
,
2257 CoalescedMemoryRange
*cmr
,
2264 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
2265 view
= address_space_get_flatview(as
);
2266 FOR_EACH_FLAT_RANGE(fr
, view
) {
2268 flat_range_coalesced_io_notify(fr
, as
, cmr
, add
);
2271 flatview_unref(view
);
2275 void memory_region_set_coalescing(MemoryRegion
*mr
)
2277 memory_region_clear_coalescing(mr
);
2278 memory_region_add_coalescing(mr
, 0, int128_get64(mr
->size
));
2281 void memory_region_add_coalescing(MemoryRegion
*mr
,
2285 CoalescedMemoryRange
*cmr
= g_malloc(sizeof(*cmr
));
2287 cmr
->addr
= addrrange_make(int128_make64(offset
), int128_make64(size
));
2288 QTAILQ_INSERT_TAIL(&mr
->coalesced
, cmr
, link
);
2289 memory_region_update_coalesced_range(mr
, cmr
, true);
2290 memory_region_set_flush_coalesced(mr
);
2293 void memory_region_clear_coalescing(MemoryRegion
*mr
)
2295 CoalescedMemoryRange
*cmr
;
2297 if (QTAILQ_EMPTY(&mr
->coalesced
)) {
2301 qemu_flush_coalesced_mmio_buffer();
2302 mr
->flush_coalesced_mmio
= false;
2304 while (!QTAILQ_EMPTY(&mr
->coalesced
)) {
2305 cmr
= QTAILQ_FIRST(&mr
->coalesced
);
2306 QTAILQ_REMOVE(&mr
->coalesced
, cmr
, link
);
2307 memory_region_update_coalesced_range(mr
, cmr
, false);
2312 void memory_region_set_flush_coalesced(MemoryRegion
*mr
)
2314 mr
->flush_coalesced_mmio
= true;
2317 void memory_region_clear_flush_coalesced(MemoryRegion
*mr
)
2319 qemu_flush_coalesced_mmio_buffer();
2320 if (QTAILQ_EMPTY(&mr
->coalesced
)) {
2321 mr
->flush_coalesced_mmio
= false;
2325 static bool userspace_eventfd_warning
;
2327 void memory_region_add_eventfd(MemoryRegion
*mr
,
2334 MemoryRegionIoeventfd mrfd
= {
2335 .addr
.start
= int128_make64(addr
),
2336 .addr
.size
= int128_make64(size
),
2337 .match_data
= match_data
,
2343 if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
2344 userspace_eventfd_warning
))) {
2345 userspace_eventfd_warning
= true;
2346 error_report("Using eventfd without MMIO binding in KVM. "
2347 "Suboptimal performance expected");
2351 adjust_endianness(mr
, &mrfd
.data
, size_memop(size
) | MO_TE
);
2353 memory_region_transaction_begin();
2354 for (i
= 0; i
< mr
->ioeventfd_nb
; ++i
) {
2355 if (memory_region_ioeventfd_before(&mrfd
, &mr
->ioeventfds
[i
])) {
2360 mr
->ioeventfds
= g_realloc(mr
->ioeventfds
,
2361 sizeof(*mr
->ioeventfds
) * mr
->ioeventfd_nb
);
2362 memmove(&mr
->ioeventfds
[i
+1], &mr
->ioeventfds
[i
],
2363 sizeof(*mr
->ioeventfds
) * (mr
->ioeventfd_nb
-1 - i
));
2364 mr
->ioeventfds
[i
] = mrfd
;
2365 ioeventfd_update_pending
|= mr
->enabled
;
2366 memory_region_transaction_commit();
2369 void memory_region_del_eventfd(MemoryRegion
*mr
,
2376 MemoryRegionIoeventfd mrfd
= {
2377 .addr
.start
= int128_make64(addr
),
2378 .addr
.size
= int128_make64(size
),
2379 .match_data
= match_data
,
2386 adjust_endianness(mr
, &mrfd
.data
, size_memop(size
) | MO_TE
);
2388 memory_region_transaction_begin();
2389 for (i
= 0; i
< mr
->ioeventfd_nb
; ++i
) {
2390 if (memory_region_ioeventfd_equal(&mrfd
, &mr
->ioeventfds
[i
])) {
2394 assert(i
!= mr
->ioeventfd_nb
);
2395 memmove(&mr
->ioeventfds
[i
], &mr
->ioeventfds
[i
+1],
2396 sizeof(*mr
->ioeventfds
) * (mr
->ioeventfd_nb
- (i
+1)));
2398 mr
->ioeventfds
= g_realloc(mr
->ioeventfds
,
2399 sizeof(*mr
->ioeventfds
)*mr
->ioeventfd_nb
+ 1);
2400 ioeventfd_update_pending
|= mr
->enabled
;
2401 memory_region_transaction_commit();
2404 static void memory_region_update_container_subregions(MemoryRegion
*subregion
)
2406 MemoryRegion
*mr
= subregion
->container
;
2407 MemoryRegion
*other
;
2409 memory_region_transaction_begin();
2411 memory_region_ref(subregion
);
2412 QTAILQ_FOREACH(other
, &mr
->subregions
, subregions_link
) {
2413 if (subregion
->priority
>= other
->priority
) {
2414 QTAILQ_INSERT_BEFORE(other
, subregion
, subregions_link
);
2418 QTAILQ_INSERT_TAIL(&mr
->subregions
, subregion
, subregions_link
);
2420 memory_region_update_pending
|= mr
->enabled
&& subregion
->enabled
;
2421 memory_region_transaction_commit();
2424 static void memory_region_add_subregion_common(MemoryRegion
*mr
,
2426 MemoryRegion
*subregion
)
2428 assert(!subregion
->container
);
2429 subregion
->container
= mr
;
2430 subregion
->addr
= offset
;
2431 memory_region_update_container_subregions(subregion
);
2434 void memory_region_add_subregion(MemoryRegion
*mr
,
2436 MemoryRegion
*subregion
)
2438 subregion
->priority
= 0;
2439 memory_region_add_subregion_common(mr
, offset
, subregion
);
2442 void memory_region_add_subregion_overlap(MemoryRegion
*mr
,
2444 MemoryRegion
*subregion
,
2447 subregion
->priority
= priority
;
2448 memory_region_add_subregion_common(mr
, offset
, subregion
);
2451 void memory_region_del_subregion(MemoryRegion
*mr
,
2452 MemoryRegion
*subregion
)
2454 memory_region_transaction_begin();
2455 assert(subregion
->container
== mr
);
2456 subregion
->container
= NULL
;
2457 QTAILQ_REMOVE(&mr
->subregions
, subregion
, subregions_link
);
2458 memory_region_unref(subregion
);
2459 memory_region_update_pending
|= mr
->enabled
&& subregion
->enabled
;
2460 memory_region_transaction_commit();
2463 void memory_region_set_enabled(MemoryRegion
*mr
, bool enabled
)
2465 if (enabled
== mr
->enabled
) {
2468 memory_region_transaction_begin();
2469 mr
->enabled
= enabled
;
2470 memory_region_update_pending
= true;
2471 memory_region_transaction_commit();
2474 void memory_region_set_size(MemoryRegion
*mr
, uint64_t size
)
2476 Int128 s
= int128_make64(size
);
2478 if (size
== UINT64_MAX
) {
2481 if (int128_eq(s
, mr
->size
)) {
2484 memory_region_transaction_begin();
2486 memory_region_update_pending
= true;
2487 memory_region_transaction_commit();
2490 static void memory_region_readd_subregion(MemoryRegion
*mr
)
2492 MemoryRegion
*container
= mr
->container
;
2495 memory_region_transaction_begin();
2496 memory_region_ref(mr
);
2497 memory_region_del_subregion(container
, mr
);
2498 mr
->container
= container
;
2499 memory_region_update_container_subregions(mr
);
2500 memory_region_unref(mr
);
2501 memory_region_transaction_commit();
2505 void memory_region_set_address(MemoryRegion
*mr
, hwaddr addr
)
2507 if (addr
!= mr
->addr
) {
2509 memory_region_readd_subregion(mr
);
2513 void memory_region_set_alias_offset(MemoryRegion
*mr
, hwaddr offset
)
2517 if (offset
== mr
->alias_offset
) {
2521 memory_region_transaction_begin();
2522 mr
->alias_offset
= offset
;
2523 memory_region_update_pending
|= mr
->enabled
;
2524 memory_region_transaction_commit();
2527 uint64_t memory_region_get_alignment(const MemoryRegion
*mr
)
2532 static int cmp_flatrange_addr(const void *addr_
, const void *fr_
)
2534 const AddrRange
*addr
= addr_
;
2535 const FlatRange
*fr
= fr_
;
2537 if (int128_le(addrrange_end(*addr
), fr
->addr
.start
)) {
2539 } else if (int128_ge(addr
->start
, addrrange_end(fr
->addr
))) {
2545 static FlatRange
*flatview_lookup(FlatView
*view
, AddrRange addr
)
2547 return bsearch(&addr
, view
->ranges
, view
->nr
,
2548 sizeof(FlatRange
), cmp_flatrange_addr
);
2551 bool memory_region_is_mapped(MemoryRegion
*mr
)
2553 return mr
->container
? true : false;
2556 /* Same as memory_region_find, but it does not add a reference to the
2557 * returned region. It must be called from an RCU critical section.
2559 static MemoryRegionSection
memory_region_find_rcu(MemoryRegion
*mr
,
2560 hwaddr addr
, uint64_t size
)
2562 MemoryRegionSection ret
= { .mr
= NULL
};
2570 for (root
= mr
; root
->container
; ) {
2571 root
= root
->container
;
2575 as
= memory_region_to_address_space(root
);
2579 range
= addrrange_make(int128_make64(addr
), int128_make64(size
));
2581 view
= address_space_to_flatview(as
);
2582 fr
= flatview_lookup(view
, range
);
2587 while (fr
> view
->ranges
&& addrrange_intersects(fr
[-1].addr
, range
)) {
2593 range
= addrrange_intersection(range
, fr
->addr
);
2594 ret
.offset_within_region
= fr
->offset_in_region
;
2595 ret
.offset_within_region
+= int128_get64(int128_sub(range
.start
,
2597 ret
.size
= range
.size
;
2598 ret
.offset_within_address_space
= int128_get64(range
.start
);
2599 ret
.readonly
= fr
->readonly
;
2600 ret
.nonvolatile
= fr
->nonvolatile
;
2604 MemoryRegionSection
memory_region_find(MemoryRegion
*mr
,
2605 hwaddr addr
, uint64_t size
)
2607 MemoryRegionSection ret
;
2608 RCU_READ_LOCK_GUARD();
2609 ret
= memory_region_find_rcu(mr
, addr
, size
);
2611 memory_region_ref(ret
.mr
);
2616 bool memory_region_present(MemoryRegion
*container
, hwaddr addr
)
2620 RCU_READ_LOCK_GUARD();
2621 mr
= memory_region_find_rcu(container
, addr
, 1).mr
;
2622 return mr
&& mr
!= container
;
2625 void memory_global_dirty_log_sync(void)
2627 memory_region_sync_dirty_bitmap(NULL
);
2630 void memory_global_after_dirty_log_sync(void)
2632 MEMORY_LISTENER_CALL_GLOBAL(log_global_after_sync
, Forward
);
2635 static VMChangeStateEntry
*vmstate_change
;
2637 void memory_global_dirty_log_start(void)
2639 if (vmstate_change
) {
2640 qemu_del_vm_change_state_handler(vmstate_change
);
2641 vmstate_change
= NULL
;
2644 global_dirty_log
= true;
2646 MEMORY_LISTENER_CALL_GLOBAL(log_global_start
, Forward
);
2648 /* Refresh DIRTY_MEMORY_MIGRATION bit. */
2649 memory_region_transaction_begin();
2650 memory_region_update_pending
= true;
2651 memory_region_transaction_commit();
2654 static void memory_global_dirty_log_do_stop(void)
2656 global_dirty_log
= false;
2658 /* Refresh DIRTY_MEMORY_MIGRATION bit. */
2659 memory_region_transaction_begin();
2660 memory_region_update_pending
= true;
2661 memory_region_transaction_commit();
2663 MEMORY_LISTENER_CALL_GLOBAL(log_global_stop
, Reverse
);
2666 static void memory_vm_change_state_handler(void *opaque
, int running
,
2670 memory_global_dirty_log_do_stop();
2672 if (vmstate_change
) {
2673 qemu_del_vm_change_state_handler(vmstate_change
);
2674 vmstate_change
= NULL
;
2679 void memory_global_dirty_log_stop(void)
2681 if (!runstate_is_running()) {
2682 if (vmstate_change
) {
2685 vmstate_change
= qemu_add_vm_change_state_handler(
2686 memory_vm_change_state_handler
, NULL
);
2690 memory_global_dirty_log_do_stop();
2693 static void listener_add_address_space(MemoryListener
*listener
,
2699 if (listener
->begin
) {
2700 listener
->begin(listener
);
2702 if (global_dirty_log
) {
2703 if (listener
->log_global_start
) {
2704 listener
->log_global_start(listener
);
2708 view
= address_space_get_flatview(as
);
2709 FOR_EACH_FLAT_RANGE(fr
, view
) {
2710 MemoryRegionSection section
= section_from_flat_range(fr
, view
);
2712 if (listener
->region_add
) {
2713 listener
->region_add(listener
, §ion
);
2715 if (fr
->dirty_log_mask
&& listener
->log_start
) {
2716 listener
->log_start(listener
, §ion
, 0, fr
->dirty_log_mask
);
2719 if (listener
->commit
) {
2720 listener
->commit(listener
);
2722 flatview_unref(view
);
2725 static void listener_del_address_space(MemoryListener
*listener
,
2731 if (listener
->begin
) {
2732 listener
->begin(listener
);
2734 view
= address_space_get_flatview(as
);
2735 FOR_EACH_FLAT_RANGE(fr
, view
) {
2736 MemoryRegionSection section
= section_from_flat_range(fr
, view
);
2738 if (fr
->dirty_log_mask
&& listener
->log_stop
) {
2739 listener
->log_stop(listener
, §ion
, fr
->dirty_log_mask
, 0);
2741 if (listener
->region_del
) {
2742 listener
->region_del(listener
, §ion
);
2745 if (listener
->commit
) {
2746 listener
->commit(listener
);
2748 flatview_unref(view
);
2751 void memory_listener_register(MemoryListener
*listener
, AddressSpace
*as
)
2753 MemoryListener
*other
= NULL
;
2755 listener
->address_space
= as
;
2756 if (QTAILQ_EMPTY(&memory_listeners
)
2757 || listener
->priority
>= QTAILQ_LAST(&memory_listeners
)->priority
) {
2758 QTAILQ_INSERT_TAIL(&memory_listeners
, listener
, link
);
2760 QTAILQ_FOREACH(other
, &memory_listeners
, link
) {
2761 if (listener
->priority
< other
->priority
) {
2765 QTAILQ_INSERT_BEFORE(other
, listener
, link
);
2768 if (QTAILQ_EMPTY(&as
->listeners
)
2769 || listener
->priority
>= QTAILQ_LAST(&as
->listeners
)->priority
) {
2770 QTAILQ_INSERT_TAIL(&as
->listeners
, listener
, link_as
);
2772 QTAILQ_FOREACH(other
, &as
->listeners
, link_as
) {
2773 if (listener
->priority
< other
->priority
) {
2777 QTAILQ_INSERT_BEFORE(other
, listener
, link_as
);
2780 listener_add_address_space(listener
, as
);
2783 void memory_listener_unregister(MemoryListener
*listener
)
2785 if (!listener
->address_space
) {
2789 listener_del_address_space(listener
, listener
->address_space
);
2790 QTAILQ_REMOVE(&memory_listeners
, listener
, link
);
2791 QTAILQ_REMOVE(&listener
->address_space
->listeners
, listener
, link_as
);
2792 listener
->address_space
= NULL
;
2795 void address_space_remove_listeners(AddressSpace
*as
)
2797 while (!QTAILQ_EMPTY(&as
->listeners
)) {
2798 memory_listener_unregister(QTAILQ_FIRST(&as
->listeners
));
2802 void address_space_init(AddressSpace
*as
, MemoryRegion
*root
, const char *name
)
2804 memory_region_ref(root
);
2806 as
->current_map
= NULL
;
2807 as
->ioeventfd_nb
= 0;
2808 as
->ioeventfds
= NULL
;
2809 QTAILQ_INIT(&as
->listeners
);
2810 QTAILQ_INSERT_TAIL(&address_spaces
, as
, address_spaces_link
);
2811 as
->name
= g_strdup(name
? name
: "anonymous");
2812 address_space_update_topology(as
);
2813 address_space_update_ioeventfds(as
);
2816 static void do_address_space_destroy(AddressSpace
*as
)
2818 assert(QTAILQ_EMPTY(&as
->listeners
));
2820 flatview_unref(as
->current_map
);
2822 g_free(as
->ioeventfds
);
2823 memory_region_unref(as
->root
);
2826 void address_space_destroy(AddressSpace
*as
)
2828 MemoryRegion
*root
= as
->root
;
2830 /* Flush out anything from MemoryListeners listening in on this */
2831 memory_region_transaction_begin();
2833 memory_region_transaction_commit();
2834 QTAILQ_REMOVE(&address_spaces
, as
, address_spaces_link
);
2836 /* At this point, as->dispatch and as->current_map are dummy
2837 * entries that the guest should never use. Wait for the old
2838 * values to expire before freeing the data.
2841 call_rcu(as
, do_address_space_destroy
, rcu
);
2844 static const char *memory_region_type(MemoryRegion
*mr
)
2847 return memory_region_type(mr
->alias
);
2849 if (memory_region_is_ram_device(mr
)) {
2851 } else if (memory_region_is_romd(mr
)) {
2853 } else if (memory_region_is_rom(mr
)) {
2855 } else if (memory_region_is_ram(mr
)) {
2862 typedef struct MemoryRegionList MemoryRegionList
;
2864 struct MemoryRegionList
{
2865 const MemoryRegion
*mr
;
2866 QTAILQ_ENTRY(MemoryRegionList
) mrqueue
;
2869 typedef QTAILQ_HEAD(, MemoryRegionList
) MemoryRegionListHead
;
2871 #define MR_SIZE(size) (int128_nz(size) ? (hwaddr)int128_get64( \
2872 int128_sub((size), int128_one())) : 0)
2873 #define MTREE_INDENT " "
2875 static void mtree_expand_owner(const char *label
, Object
*obj
)
2877 DeviceState
*dev
= (DeviceState
*) object_dynamic_cast(obj
, TYPE_DEVICE
);
2879 qemu_printf(" %s:{%s", label
, dev
? "dev" : "obj");
2880 if (dev
&& dev
->id
) {
2881 qemu_printf(" id=%s", dev
->id
);
2883 char *canonical_path
= object_get_canonical_path(obj
);
2884 if (canonical_path
) {
2885 qemu_printf(" path=%s", canonical_path
);
2886 g_free(canonical_path
);
2888 qemu_printf(" type=%s", object_get_typename(obj
));
2894 static void mtree_print_mr_owner(const MemoryRegion
*mr
)
2896 Object
*owner
= mr
->owner
;
2897 Object
*parent
= memory_region_owner((MemoryRegion
*)mr
);
2899 if (!owner
&& !parent
) {
2900 qemu_printf(" orphan");
2904 mtree_expand_owner("owner", owner
);
2906 if (parent
&& parent
!= owner
) {
2907 mtree_expand_owner("parent", parent
);
2911 static void mtree_print_mr(const MemoryRegion
*mr
, unsigned int level
,
2913 MemoryRegionListHead
*alias_print_queue
,
2914 bool owner
, bool display_disabled
)
2916 MemoryRegionList
*new_ml
, *ml
, *next_ml
;
2917 MemoryRegionListHead submr_print_queue
;
2918 const MemoryRegion
*submr
;
2920 hwaddr cur_start
, cur_end
;
2926 cur_start
= base
+ mr
->addr
;
2927 cur_end
= cur_start
+ MR_SIZE(mr
->size
);
2930 * Try to detect overflow of memory region. This should never
2931 * happen normally. When it happens, we dump something to warn the
2932 * user who is observing this.
2934 if (cur_start
< base
|| cur_end
< cur_start
) {
2935 qemu_printf("[DETECTED OVERFLOW!] ");
2939 MemoryRegionList
*ml
;
2942 /* check if the alias is already in the queue */
2943 QTAILQ_FOREACH(ml
, alias_print_queue
, mrqueue
) {
2944 if (ml
->mr
== mr
->alias
) {
2950 ml
= g_new(MemoryRegionList
, 1);
2952 QTAILQ_INSERT_TAIL(alias_print_queue
, ml
, mrqueue
);
2954 if (mr
->enabled
|| display_disabled
) {
2955 for (i
= 0; i
< level
; i
++) {
2956 qemu_printf(MTREE_INDENT
);
2958 qemu_printf(TARGET_FMT_plx
"-" TARGET_FMT_plx
2959 " (prio %d, %s%s): alias %s @%s " TARGET_FMT_plx
2960 "-" TARGET_FMT_plx
"%s",
2963 mr
->nonvolatile
? "nv-" : "",
2964 memory_region_type((MemoryRegion
*)mr
),
2965 memory_region_name(mr
),
2966 memory_region_name(mr
->alias
),
2968 mr
->alias_offset
+ MR_SIZE(mr
->size
),
2969 mr
->enabled
? "" : " [disabled]");
2971 mtree_print_mr_owner(mr
);
2976 if (mr
->enabled
|| display_disabled
) {
2977 for (i
= 0; i
< level
; i
++) {
2978 qemu_printf(MTREE_INDENT
);
2980 qemu_printf(TARGET_FMT_plx
"-" TARGET_FMT_plx
2981 " (prio %d, %s%s): %s%s",
2984 mr
->nonvolatile
? "nv-" : "",
2985 memory_region_type((MemoryRegion
*)mr
),
2986 memory_region_name(mr
),
2987 mr
->enabled
? "" : " [disabled]");
2989 mtree_print_mr_owner(mr
);
2995 QTAILQ_INIT(&submr_print_queue
);
2997 QTAILQ_FOREACH(submr
, &mr
->subregions
, subregions_link
) {
2998 new_ml
= g_new(MemoryRegionList
, 1);
3000 QTAILQ_FOREACH(ml
, &submr_print_queue
, mrqueue
) {
3001 if (new_ml
->mr
->addr
< ml
->mr
->addr
||
3002 (new_ml
->mr
->addr
== ml
->mr
->addr
&&
3003 new_ml
->mr
->priority
> ml
->mr
->priority
)) {
3004 QTAILQ_INSERT_BEFORE(ml
, new_ml
, mrqueue
);
3010 QTAILQ_INSERT_TAIL(&submr_print_queue
, new_ml
, mrqueue
);
3014 QTAILQ_FOREACH(ml
, &submr_print_queue
, mrqueue
) {
3015 mtree_print_mr(ml
->mr
, level
+ 1, cur_start
,
3016 alias_print_queue
, owner
, display_disabled
);
3019 QTAILQ_FOREACH_SAFE(ml
, &submr_print_queue
, mrqueue
, next_ml
) {
3024 struct FlatViewInfo
{
3031 static void mtree_print_flatview(gpointer key
, gpointer value
,
3034 FlatView
*view
= key
;
3035 GArray
*fv_address_spaces
= value
;
3036 struct FlatViewInfo
*fvi
= user_data
;
3037 FlatRange
*range
= &view
->ranges
[0];
3043 qemu_printf("FlatView #%d\n", fvi
->counter
);
3046 for (i
= 0; i
< fv_address_spaces
->len
; ++i
) {
3047 as
= g_array_index(fv_address_spaces
, AddressSpace
*, i
);
3048 qemu_printf(" AS \"%s\", root: %s",
3049 as
->name
, memory_region_name(as
->root
));
3050 if (as
->root
->alias
) {
3051 qemu_printf(", alias %s", memory_region_name(as
->root
->alias
));
3056 qemu_printf(" Root memory region: %s\n",
3057 view
->root
? memory_region_name(view
->root
) : "(none)");
3060 qemu_printf(MTREE_INDENT
"No rendered FlatView\n\n");
3066 if (range
->offset_in_region
) {
3067 qemu_printf(MTREE_INDENT TARGET_FMT_plx
"-" TARGET_FMT_plx
3068 " (prio %d, %s%s): %s @" TARGET_FMT_plx
,
3069 int128_get64(range
->addr
.start
),
3070 int128_get64(range
->addr
.start
)
3071 + MR_SIZE(range
->addr
.size
),
3073 range
->nonvolatile
? "nv-" : "",
3074 range
->readonly
? "rom" : memory_region_type(mr
),
3075 memory_region_name(mr
),
3076 range
->offset_in_region
);
3078 qemu_printf(MTREE_INDENT TARGET_FMT_plx
"-" TARGET_FMT_plx
3079 " (prio %d, %s%s): %s",
3080 int128_get64(range
->addr
.start
),
3081 int128_get64(range
->addr
.start
)
3082 + MR_SIZE(range
->addr
.size
),
3084 range
->nonvolatile
? "nv-" : "",
3085 range
->readonly
? "rom" : memory_region_type(mr
),
3086 memory_region_name(mr
));
3089 mtree_print_mr_owner(mr
);
3093 for (i
= 0; i
< fv_address_spaces
->len
; ++i
) {
3094 as
= g_array_index(fv_address_spaces
, AddressSpace
*, i
);
3095 if (fvi
->ac
->has_memory(current_machine
, as
,
3096 int128_get64(range
->addr
.start
),
3097 MR_SIZE(range
->addr
.size
) + 1)) {
3098 qemu_printf(" %s", fvi
->ac
->name
);
3106 #if !defined(CONFIG_USER_ONLY)
3107 if (fvi
->dispatch_tree
&& view
->root
) {
3108 mtree_print_dispatch(view
->dispatch
, view
->root
);
3115 static gboolean
mtree_info_flatview_free(gpointer key
, gpointer value
,
3118 FlatView
*view
= key
;
3119 GArray
*fv_address_spaces
= value
;
3121 g_array_unref(fv_address_spaces
);
3122 flatview_unref(view
);
3127 void mtree_info(bool flatview
, bool dispatch_tree
, bool owner
, bool disabled
)
3129 MemoryRegionListHead ml_head
;
3130 MemoryRegionList
*ml
, *ml2
;
3135 struct FlatViewInfo fvi
= {
3137 .dispatch_tree
= dispatch_tree
,
3140 GArray
*fv_address_spaces
;
3141 GHashTable
*views
= g_hash_table_new(g_direct_hash
, g_direct_equal
);
3142 AccelClass
*ac
= ACCEL_GET_CLASS(current_accel());
3144 if (ac
->has_memory
) {
3148 /* Gather all FVs in one table */
3149 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
3150 view
= address_space_get_flatview(as
);
3152 fv_address_spaces
= g_hash_table_lookup(views
, view
);
3153 if (!fv_address_spaces
) {
3154 fv_address_spaces
= g_array_new(false, false, sizeof(as
));
3155 g_hash_table_insert(views
, view
, fv_address_spaces
);
3158 g_array_append_val(fv_address_spaces
, as
);
3162 g_hash_table_foreach(views
, mtree_print_flatview
, &fvi
);
3165 g_hash_table_foreach_remove(views
, mtree_info_flatview_free
, 0);
3166 g_hash_table_unref(views
);
3171 QTAILQ_INIT(&ml_head
);
3173 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
3174 qemu_printf("address-space: %s\n", as
->name
);
3175 mtree_print_mr(as
->root
, 1, 0, &ml_head
, owner
, disabled
);
3179 /* print aliased regions */
3180 QTAILQ_FOREACH(ml
, &ml_head
, mrqueue
) {
3181 qemu_printf("memory-region: %s\n", memory_region_name(ml
->mr
));
3182 mtree_print_mr(ml
->mr
, 1, 0, &ml_head
, owner
, disabled
);
3186 QTAILQ_FOREACH_SAFE(ml
, &ml_head
, mrqueue
, ml2
) {
3191 void memory_region_init_ram(MemoryRegion
*mr
,
3192 struct Object
*owner
,
3197 DeviceState
*owner_dev
;
3200 memory_region_init_ram_nomigrate(mr
, owner
, name
, size
, &err
);
3202 error_propagate(errp
, err
);
3205 /* This will assert if owner is neither NULL nor a DeviceState.
3206 * We only want the owner here for the purposes of defining a
3207 * unique name for migration. TODO: Ideally we should implement
3208 * a naming scheme for Objects which are not DeviceStates, in
3209 * which case we can relax this restriction.
3211 owner_dev
= DEVICE(owner
);
3212 vmstate_register_ram(mr
, owner_dev
);
3215 void memory_region_init_rom(MemoryRegion
*mr
,
3216 struct Object
*owner
,
3221 DeviceState
*owner_dev
;
3224 memory_region_init_rom_nomigrate(mr
, owner
, name
, size
, &err
);
3226 error_propagate(errp
, err
);
3229 /* This will assert if owner is neither NULL nor a DeviceState.
3230 * We only want the owner here for the purposes of defining a
3231 * unique name for migration. TODO: Ideally we should implement
3232 * a naming scheme for Objects which are not DeviceStates, in
3233 * which case we can relax this restriction.
3235 owner_dev
= DEVICE(owner
);
3236 vmstate_register_ram(mr
, owner_dev
);
3239 void memory_region_init_rom_device(MemoryRegion
*mr
,
3240 struct Object
*owner
,
3241 const MemoryRegionOps
*ops
,
3247 DeviceState
*owner_dev
;
3250 memory_region_init_rom_device_nomigrate(mr
, owner
, ops
, opaque
,
3253 error_propagate(errp
, err
);
3256 /* This will assert if owner is neither NULL nor a DeviceState.
3257 * We only want the owner here for the purposes of defining a
3258 * unique name for migration. TODO: Ideally we should implement
3259 * a naming scheme for Objects which are not DeviceStates, in
3260 * which case we can relax this restriction.
3262 owner_dev
= DEVICE(owner
);
3263 vmstate_register_ram(mr
, owner_dev
);
3267 * Support softmmu builds with CONFIG_FUZZ using a weak symbol and a stub for
3268 * the fuzz_dma_read_cb callback
3271 void __attribute__((weak
)) fuzz_dma_read_cb(size_t addr
,
3279 static const TypeInfo memory_region_info
= {
3280 .parent
= TYPE_OBJECT
,
3281 .name
= TYPE_MEMORY_REGION
,
3282 .class_size
= sizeof(MemoryRegionClass
),
3283 .instance_size
= sizeof(MemoryRegion
),
3284 .instance_init
= memory_region_initfn
,
3285 .instance_finalize
= memory_region_finalize
,
3288 static const TypeInfo iommu_memory_region_info
= {
3289 .parent
= TYPE_MEMORY_REGION
,
3290 .name
= TYPE_IOMMU_MEMORY_REGION
,
3291 .class_size
= sizeof(IOMMUMemoryRegionClass
),
3292 .instance_size
= sizeof(IOMMUMemoryRegion
),
3293 .instance_init
= iommu_memory_region_initfn
,
3297 static void memory_register_types(void)
3299 type_register_static(&memory_region_info
);
3300 type_register_static(&iommu_memory_region_info
);
3303 type_init(memory_register_types
)