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"
17 #include "qapi/error.h"
18 #include "qemu-common.h"
20 #include "exec/memory.h"
21 #include "exec/address-spaces.h"
22 #include "exec/ioport.h"
23 #include "qapi/visitor.h"
24 #include "qemu/bitops.h"
25 #include "qemu/error-report.h"
26 #include "qom/object.h"
29 #include "exec/memory-internal.h"
30 #include "exec/ram_addr.h"
31 #include "sysemu/kvm.h"
32 #include "sysemu/sysemu.h"
34 //#define DEBUG_UNASSIGNED
36 #define RAM_ADDR_INVALID (~(ram_addr_t)0)
38 static unsigned memory_region_transaction_depth
;
39 static bool memory_region_update_pending
;
40 static bool ioeventfd_update_pending
;
41 static bool global_dirty_log
= false;
43 static QTAILQ_HEAD(memory_listeners
, MemoryListener
) memory_listeners
44 = QTAILQ_HEAD_INITIALIZER(memory_listeners
);
46 static QTAILQ_HEAD(, AddressSpace
) address_spaces
47 = QTAILQ_HEAD_INITIALIZER(address_spaces
);
49 typedef struct AddrRange AddrRange
;
52 * Note that signed integers are needed for negative offsetting in aliases
53 * (large MemoryRegion::alias_offset).
60 static AddrRange
addrrange_make(Int128 start
, Int128 size
)
62 return (AddrRange
) { start
, size
};
65 static bool addrrange_equal(AddrRange r1
, AddrRange r2
)
67 return int128_eq(r1
.start
, r2
.start
) && int128_eq(r1
.size
, r2
.size
);
70 static Int128
addrrange_end(AddrRange r
)
72 return int128_add(r
.start
, r
.size
);
75 static AddrRange
addrrange_shift(AddrRange range
, Int128 delta
)
77 int128_addto(&range
.start
, delta
);
81 static bool addrrange_contains(AddrRange range
, Int128 addr
)
83 return int128_ge(addr
, range
.start
)
84 && int128_lt(addr
, addrrange_end(range
));
87 static bool addrrange_intersects(AddrRange r1
, AddrRange r2
)
89 return addrrange_contains(r1
, r2
.start
)
90 || addrrange_contains(r2
, r1
.start
);
93 static AddrRange
addrrange_intersection(AddrRange r1
, AddrRange r2
)
95 Int128 start
= int128_max(r1
.start
, r2
.start
);
96 Int128 end
= int128_min(addrrange_end(r1
), addrrange_end(r2
));
97 return addrrange_make(start
, int128_sub(end
, start
));
100 enum ListenerDirection
{ Forward
, Reverse
};
102 static bool memory_listener_match(MemoryListener
*listener
,
103 MemoryRegionSection
*section
)
105 return !listener
->address_space_filter
106 || listener
->address_space_filter
== section
->address_space
;
109 #define MEMORY_LISTENER_CALL_GLOBAL(_callback, _direction, _args...) \
111 MemoryListener *_listener; \
113 switch (_direction) { \
115 QTAILQ_FOREACH(_listener, &memory_listeners, link) { \
116 if (_listener->_callback) { \
117 _listener->_callback(_listener, ##_args); \
122 QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners, \
123 memory_listeners, link) { \
124 if (_listener->_callback) { \
125 _listener->_callback(_listener, ##_args); \
134 #define MEMORY_LISTENER_CALL(_callback, _direction, _section, _args...) \
136 MemoryListener *_listener; \
138 switch (_direction) { \
140 QTAILQ_FOREACH(_listener, &memory_listeners, link) { \
141 if (_listener->_callback \
142 && memory_listener_match(_listener, _section)) { \
143 _listener->_callback(_listener, _section, ##_args); \
148 QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners, \
149 memory_listeners, link) { \
150 if (_listener->_callback \
151 && memory_listener_match(_listener, _section)) { \
152 _listener->_callback(_listener, _section, ##_args); \
161 /* No need to ref/unref .mr, the FlatRange keeps it alive. */
162 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...) \
163 MEMORY_LISTENER_CALL(callback, dir, (&(MemoryRegionSection) { \
165 .address_space = (as), \
166 .offset_within_region = (fr)->offset_in_region, \
167 .size = (fr)->addr.size, \
168 .offset_within_address_space = int128_get64((fr)->addr.start), \
169 .readonly = (fr)->readonly, \
172 struct CoalescedMemoryRange
{
174 QTAILQ_ENTRY(CoalescedMemoryRange
) link
;
177 struct MemoryRegionIoeventfd
{
184 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd a
,
185 MemoryRegionIoeventfd b
)
187 if (int128_lt(a
.addr
.start
, b
.addr
.start
)) {
189 } else if (int128_gt(a
.addr
.start
, b
.addr
.start
)) {
191 } else if (int128_lt(a
.addr
.size
, b
.addr
.size
)) {
193 } else if (int128_gt(a
.addr
.size
, b
.addr
.size
)) {
195 } else if (a
.match_data
< b
.match_data
) {
197 } else if (a
.match_data
> b
.match_data
) {
199 } else if (a
.match_data
) {
200 if (a
.data
< b
.data
) {
202 } else if (a
.data
> b
.data
) {
208 } else if (a
.e
> b
.e
) {
214 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd a
,
215 MemoryRegionIoeventfd b
)
217 return !memory_region_ioeventfd_before(a
, b
)
218 && !memory_region_ioeventfd_before(b
, a
);
221 typedef struct FlatRange FlatRange
;
222 typedef struct FlatView FlatView
;
224 /* Range of memory in the global map. Addresses are absolute. */
227 hwaddr offset_in_region
;
229 uint8_t dirty_log_mask
;
234 /* Flattened global view of current active memory hierarchy. Kept in sorted
242 unsigned nr_allocated
;
245 typedef struct AddressSpaceOps AddressSpaceOps
;
247 #define FOR_EACH_FLAT_RANGE(var, view) \
248 for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
250 static bool flatrange_equal(FlatRange
*a
, FlatRange
*b
)
252 return a
->mr
== b
->mr
253 && addrrange_equal(a
->addr
, b
->addr
)
254 && a
->offset_in_region
== b
->offset_in_region
255 && a
->romd_mode
== b
->romd_mode
256 && a
->readonly
== b
->readonly
;
259 static void flatview_init(FlatView
*view
)
264 view
->nr_allocated
= 0;
267 /* Insert a range into a given position. Caller is responsible for maintaining
270 static void flatview_insert(FlatView
*view
, unsigned pos
, FlatRange
*range
)
272 if (view
->nr
== view
->nr_allocated
) {
273 view
->nr_allocated
= MAX(2 * view
->nr
, 10);
274 view
->ranges
= g_realloc(view
->ranges
,
275 view
->nr_allocated
* sizeof(*view
->ranges
));
277 memmove(view
->ranges
+ pos
+ 1, view
->ranges
+ pos
,
278 (view
->nr
- pos
) * sizeof(FlatRange
));
279 view
->ranges
[pos
] = *range
;
280 memory_region_ref(range
->mr
);
284 static void flatview_destroy(FlatView
*view
)
288 for (i
= 0; i
< view
->nr
; i
++) {
289 memory_region_unref(view
->ranges
[i
].mr
);
291 g_free(view
->ranges
);
295 static void flatview_ref(FlatView
*view
)
297 atomic_inc(&view
->ref
);
300 static void flatview_unref(FlatView
*view
)
302 if (atomic_fetch_dec(&view
->ref
) == 1) {
303 flatview_destroy(view
);
307 static bool can_merge(FlatRange
*r1
, FlatRange
*r2
)
309 return int128_eq(addrrange_end(r1
->addr
), r2
->addr
.start
)
311 && int128_eq(int128_add(int128_make64(r1
->offset_in_region
),
313 int128_make64(r2
->offset_in_region
))
314 && r1
->dirty_log_mask
== r2
->dirty_log_mask
315 && r1
->romd_mode
== r2
->romd_mode
316 && r1
->readonly
== r2
->readonly
;
319 /* Attempt to simplify a view by merging adjacent ranges */
320 static void flatview_simplify(FlatView
*view
)
325 while (i
< view
->nr
) {
328 && can_merge(&view
->ranges
[j
-1], &view
->ranges
[j
])) {
329 int128_addto(&view
->ranges
[i
].addr
.size
, view
->ranges
[j
].addr
.size
);
333 memmove(&view
->ranges
[i
], &view
->ranges
[j
],
334 (view
->nr
- j
) * sizeof(view
->ranges
[j
]));
339 static bool memory_region_big_endian(MemoryRegion
*mr
)
341 #ifdef TARGET_WORDS_BIGENDIAN
342 return mr
->ops
->endianness
!= DEVICE_LITTLE_ENDIAN
;
344 return mr
->ops
->endianness
== DEVICE_BIG_ENDIAN
;
348 static bool memory_region_wrong_endianness(MemoryRegion
*mr
)
350 #ifdef TARGET_WORDS_BIGENDIAN
351 return mr
->ops
->endianness
== DEVICE_LITTLE_ENDIAN
;
353 return mr
->ops
->endianness
== DEVICE_BIG_ENDIAN
;
357 static void adjust_endianness(MemoryRegion
*mr
, uint64_t *data
, unsigned size
)
359 if (memory_region_wrong_endianness(mr
)) {
364 *data
= bswap16(*data
);
367 *data
= bswap32(*data
);
370 *data
= bswap64(*data
);
378 static hwaddr
memory_region_to_absolute_addr(MemoryRegion
*mr
, hwaddr offset
)
381 hwaddr abs_addr
= offset
;
383 abs_addr
+= mr
->addr
;
384 for (root
= mr
; root
->container
; ) {
385 root
= root
->container
;
386 abs_addr
+= root
->addr
;
392 static int get_cpu_index(void)
395 return current_cpu
->cpu_index
;
400 static MemTxResult
memory_region_oldmmio_read_accessor(MemoryRegion
*mr
,
410 tmp
= mr
->ops
->old_mmio
.read
[ctz32(size
)](mr
->opaque
, addr
);
412 trace_memory_region_subpage_read(get_cpu_index(), mr
, addr
, tmp
, size
);
413 } else if (mr
== &io_mem_notdirty
) {
414 /* Accesses to code which has previously been translated into a TB show
415 * up in the MMIO path, as accesses to the io_mem_notdirty
417 trace_memory_region_tb_read(get_cpu_index(), addr
, tmp
, size
);
418 } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED
) {
419 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
420 trace_memory_region_ops_read(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
422 *value
|= (tmp
& mask
) << shift
;
426 static MemTxResult
memory_region_read_accessor(MemoryRegion
*mr
,
436 tmp
= mr
->ops
->read(mr
->opaque
, addr
, size
);
438 trace_memory_region_subpage_read(get_cpu_index(), mr
, addr
, tmp
, size
);
439 } else if (mr
== &io_mem_notdirty
) {
440 /* Accesses to code which has previously been translated into a TB show
441 * up in the MMIO path, as accesses to the io_mem_notdirty
443 trace_memory_region_tb_read(get_cpu_index(), addr
, tmp
, size
);
444 } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED
) {
445 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
446 trace_memory_region_ops_read(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
448 *value
|= (tmp
& mask
) << shift
;
452 static MemTxResult
memory_region_read_with_attrs_accessor(MemoryRegion
*mr
,
463 r
= mr
->ops
->read_with_attrs(mr
->opaque
, addr
, &tmp
, size
, attrs
);
465 trace_memory_region_subpage_read(get_cpu_index(), mr
, addr
, tmp
, size
);
466 } else if (mr
== &io_mem_notdirty
) {
467 /* Accesses to code which has previously been translated into a TB show
468 * up in the MMIO path, as accesses to the io_mem_notdirty
470 trace_memory_region_tb_read(get_cpu_index(), addr
, tmp
, size
);
471 } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED
) {
472 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
473 trace_memory_region_ops_read(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
475 *value
|= (tmp
& mask
) << shift
;
479 static MemTxResult
memory_region_oldmmio_write_accessor(MemoryRegion
*mr
,
489 tmp
= (*value
>> shift
) & mask
;
491 trace_memory_region_subpage_write(get_cpu_index(), mr
, addr
, tmp
, size
);
492 } else if (mr
== &io_mem_notdirty
) {
493 /* Accesses to code which has previously been translated into a TB show
494 * up in the MMIO path, as accesses to the io_mem_notdirty
496 trace_memory_region_tb_write(get_cpu_index(), addr
, tmp
, size
);
497 } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED
) {
498 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
499 trace_memory_region_ops_write(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
501 mr
->ops
->old_mmio
.write
[ctz32(size
)](mr
->opaque
, addr
, tmp
);
505 static MemTxResult
memory_region_write_accessor(MemoryRegion
*mr
,
515 tmp
= (*value
>> shift
) & mask
;
517 trace_memory_region_subpage_write(get_cpu_index(), mr
, addr
, tmp
, size
);
518 } else if (mr
== &io_mem_notdirty
) {
519 /* Accesses to code which has previously been translated into a TB show
520 * up in the MMIO path, as accesses to the io_mem_notdirty
522 trace_memory_region_tb_write(get_cpu_index(), addr
, tmp
, size
);
523 } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED
) {
524 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
525 trace_memory_region_ops_write(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
527 mr
->ops
->write(mr
->opaque
, addr
, tmp
, size
);
531 static MemTxResult
memory_region_write_with_attrs_accessor(MemoryRegion
*mr
,
541 tmp
= (*value
>> shift
) & mask
;
543 trace_memory_region_subpage_write(get_cpu_index(), mr
, addr
, tmp
, size
);
544 } else if (mr
== &io_mem_notdirty
) {
545 /* Accesses to code which has previously been translated into a TB show
546 * up in the MMIO path, as accesses to the io_mem_notdirty
548 trace_memory_region_tb_write(get_cpu_index(), addr
, tmp
, size
);
549 } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED
) {
550 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
551 trace_memory_region_ops_write(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
553 return mr
->ops
->write_with_attrs(mr
->opaque
, addr
, tmp
, size
, attrs
);
556 static MemTxResult
access_with_adjusted_size(hwaddr addr
,
559 unsigned access_size_min
,
560 unsigned access_size_max
,
561 MemTxResult (*access
)(MemoryRegion
*mr
,
571 uint64_t access_mask
;
572 unsigned access_size
;
574 MemTxResult r
= MEMTX_OK
;
576 if (!access_size_min
) {
579 if (!access_size_max
) {
583 /* FIXME: support unaligned access? */
584 access_size
= MAX(MIN(size
, access_size_max
), access_size_min
);
585 access_mask
= -1ULL >> (64 - access_size
* 8);
586 if (memory_region_big_endian(mr
)) {
587 for (i
= 0; i
< size
; i
+= access_size
) {
588 r
|= access(mr
, addr
+ i
, value
, access_size
,
589 (size
- access_size
- i
) * 8, access_mask
, attrs
);
592 for (i
= 0; i
< size
; i
+= access_size
) {
593 r
|= access(mr
, addr
+ i
, value
, access_size
, i
* 8,
600 static AddressSpace
*memory_region_to_address_space(MemoryRegion
*mr
)
604 while (mr
->container
) {
607 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
608 if (mr
== as
->root
) {
615 /* Render a memory region into the global view. Ranges in @view obscure
618 static void render_memory_region(FlatView
*view
,
624 MemoryRegion
*subregion
;
626 hwaddr offset_in_region
;
636 int128_addto(&base
, int128_make64(mr
->addr
));
637 readonly
|= mr
->readonly
;
639 tmp
= addrrange_make(base
, mr
->size
);
641 if (!addrrange_intersects(tmp
, clip
)) {
645 clip
= addrrange_intersection(tmp
, clip
);
648 int128_subfrom(&base
, int128_make64(mr
->alias
->addr
));
649 int128_subfrom(&base
, int128_make64(mr
->alias_offset
));
650 render_memory_region(view
, mr
->alias
, base
, clip
, readonly
);
654 /* Render subregions in priority order. */
655 QTAILQ_FOREACH(subregion
, &mr
->subregions
, subregions_link
) {
656 render_memory_region(view
, subregion
, base
, clip
, readonly
);
659 if (!mr
->terminates
) {
663 offset_in_region
= int128_get64(int128_sub(clip
.start
, base
));
668 fr
.dirty_log_mask
= memory_region_get_dirty_log_mask(mr
);
669 fr
.romd_mode
= mr
->romd_mode
;
670 fr
.readonly
= readonly
;
672 /* Render the region itself into any gaps left by the current view. */
673 for (i
= 0; i
< view
->nr
&& int128_nz(remain
); ++i
) {
674 if (int128_ge(base
, addrrange_end(view
->ranges
[i
].addr
))) {
677 if (int128_lt(base
, view
->ranges
[i
].addr
.start
)) {
678 now
= int128_min(remain
,
679 int128_sub(view
->ranges
[i
].addr
.start
, base
));
680 fr
.offset_in_region
= offset_in_region
;
681 fr
.addr
= addrrange_make(base
, now
);
682 flatview_insert(view
, i
, &fr
);
684 int128_addto(&base
, now
);
685 offset_in_region
+= int128_get64(now
);
686 int128_subfrom(&remain
, now
);
688 now
= int128_sub(int128_min(int128_add(base
, remain
),
689 addrrange_end(view
->ranges
[i
].addr
)),
691 int128_addto(&base
, now
);
692 offset_in_region
+= int128_get64(now
);
693 int128_subfrom(&remain
, now
);
695 if (int128_nz(remain
)) {
696 fr
.offset_in_region
= offset_in_region
;
697 fr
.addr
= addrrange_make(base
, remain
);
698 flatview_insert(view
, i
, &fr
);
702 /* Render a memory topology into a list of disjoint absolute ranges. */
703 static FlatView
*generate_memory_topology(MemoryRegion
*mr
)
707 view
= g_new(FlatView
, 1);
711 render_memory_region(view
, mr
, int128_zero(),
712 addrrange_make(int128_zero(), int128_2_64()), false);
714 flatview_simplify(view
);
719 static void address_space_add_del_ioeventfds(AddressSpace
*as
,
720 MemoryRegionIoeventfd
*fds_new
,
722 MemoryRegionIoeventfd
*fds_old
,
726 MemoryRegionIoeventfd
*fd
;
727 MemoryRegionSection section
;
729 /* Generate a symmetric difference of the old and new fd sets, adding
730 * and deleting as necessary.
734 while (iold
< fds_old_nb
|| inew
< fds_new_nb
) {
735 if (iold
< fds_old_nb
736 && (inew
== fds_new_nb
737 || memory_region_ioeventfd_before(fds_old
[iold
],
740 section
= (MemoryRegionSection
) {
742 .offset_within_address_space
= int128_get64(fd
->addr
.start
),
743 .size
= fd
->addr
.size
,
745 MEMORY_LISTENER_CALL(eventfd_del
, Forward
, §ion
,
746 fd
->match_data
, fd
->data
, fd
->e
);
748 } else if (inew
< fds_new_nb
749 && (iold
== fds_old_nb
750 || memory_region_ioeventfd_before(fds_new
[inew
],
753 section
= (MemoryRegionSection
) {
755 .offset_within_address_space
= int128_get64(fd
->addr
.start
),
756 .size
= fd
->addr
.size
,
758 MEMORY_LISTENER_CALL(eventfd_add
, Reverse
, §ion
,
759 fd
->match_data
, fd
->data
, fd
->e
);
768 static FlatView
*address_space_get_flatview(AddressSpace
*as
)
773 view
= atomic_rcu_read(&as
->current_map
);
779 static void address_space_update_ioeventfds(AddressSpace
*as
)
783 unsigned ioeventfd_nb
= 0;
784 MemoryRegionIoeventfd
*ioeventfds
= NULL
;
788 view
= address_space_get_flatview(as
);
789 FOR_EACH_FLAT_RANGE(fr
, view
) {
790 for (i
= 0; i
< fr
->mr
->ioeventfd_nb
; ++i
) {
791 tmp
= addrrange_shift(fr
->mr
->ioeventfds
[i
].addr
,
792 int128_sub(fr
->addr
.start
,
793 int128_make64(fr
->offset_in_region
)));
794 if (addrrange_intersects(fr
->addr
, tmp
)) {
796 ioeventfds
= g_realloc(ioeventfds
,
797 ioeventfd_nb
* sizeof(*ioeventfds
));
798 ioeventfds
[ioeventfd_nb
-1] = fr
->mr
->ioeventfds
[i
];
799 ioeventfds
[ioeventfd_nb
-1].addr
= tmp
;
804 address_space_add_del_ioeventfds(as
, ioeventfds
, ioeventfd_nb
,
805 as
->ioeventfds
, as
->ioeventfd_nb
);
807 g_free(as
->ioeventfds
);
808 as
->ioeventfds
= ioeventfds
;
809 as
->ioeventfd_nb
= ioeventfd_nb
;
810 flatview_unref(view
);
813 static void address_space_update_topology_pass(AddressSpace
*as
,
814 const FlatView
*old_view
,
815 const FlatView
*new_view
,
819 FlatRange
*frold
, *frnew
;
821 /* Generate a symmetric difference of the old and new memory maps.
822 * Kill ranges in the old map, and instantiate ranges in the new map.
825 while (iold
< old_view
->nr
|| inew
< new_view
->nr
) {
826 if (iold
< old_view
->nr
) {
827 frold
= &old_view
->ranges
[iold
];
831 if (inew
< new_view
->nr
) {
832 frnew
= &new_view
->ranges
[inew
];
839 || int128_lt(frold
->addr
.start
, frnew
->addr
.start
)
840 || (int128_eq(frold
->addr
.start
, frnew
->addr
.start
)
841 && !flatrange_equal(frold
, frnew
)))) {
842 /* In old but not in new, or in both but attributes changed. */
845 MEMORY_LISTENER_UPDATE_REGION(frold
, as
, Reverse
, region_del
);
849 } else if (frold
&& frnew
&& flatrange_equal(frold
, frnew
)) {
850 /* In both and unchanged (except logging may have changed) */
853 MEMORY_LISTENER_UPDATE_REGION(frnew
, as
, Forward
, region_nop
);
854 if (frnew
->dirty_log_mask
& ~frold
->dirty_log_mask
) {
855 MEMORY_LISTENER_UPDATE_REGION(frnew
, as
, Forward
, log_start
,
856 frold
->dirty_log_mask
,
857 frnew
->dirty_log_mask
);
859 if (frold
->dirty_log_mask
& ~frnew
->dirty_log_mask
) {
860 MEMORY_LISTENER_UPDATE_REGION(frnew
, as
, Reverse
, log_stop
,
861 frold
->dirty_log_mask
,
862 frnew
->dirty_log_mask
);
872 MEMORY_LISTENER_UPDATE_REGION(frnew
, as
, Forward
, region_add
);
881 static void address_space_update_topology(AddressSpace
*as
)
883 FlatView
*old_view
= address_space_get_flatview(as
);
884 FlatView
*new_view
= generate_memory_topology(as
->root
);
886 address_space_update_topology_pass(as
, old_view
, new_view
, false);
887 address_space_update_topology_pass(as
, old_view
, new_view
, true);
889 /* Writes are protected by the BQL. */
890 atomic_rcu_set(&as
->current_map
, new_view
);
891 call_rcu(old_view
, flatview_unref
, rcu
);
893 /* Note that all the old MemoryRegions are still alive up to this
894 * point. This relieves most MemoryListeners from the need to
895 * ref/unref the MemoryRegions they get---unless they use them
896 * outside the iothread mutex, in which case precise reference
897 * counting is necessary.
899 flatview_unref(old_view
);
901 address_space_update_ioeventfds(as
);
904 void memory_region_transaction_begin(void)
906 qemu_flush_coalesced_mmio_buffer();
907 ++memory_region_transaction_depth
;
910 static void memory_region_clear_pending(void)
912 memory_region_update_pending
= false;
913 ioeventfd_update_pending
= false;
916 void memory_region_transaction_commit(void)
920 assert(memory_region_transaction_depth
);
921 --memory_region_transaction_depth
;
922 if (!memory_region_transaction_depth
) {
923 if (memory_region_update_pending
) {
924 MEMORY_LISTENER_CALL_GLOBAL(begin
, Forward
);
926 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
927 address_space_update_topology(as
);
930 MEMORY_LISTENER_CALL_GLOBAL(commit
, Forward
);
931 } else if (ioeventfd_update_pending
) {
932 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
933 address_space_update_ioeventfds(as
);
936 memory_region_clear_pending();
940 static void memory_region_destructor_none(MemoryRegion
*mr
)
944 static void memory_region_destructor_ram(MemoryRegion
*mr
)
946 qemu_ram_free(mr
->ram_block
);
949 static void memory_region_destructor_rom_device(MemoryRegion
*mr
)
951 qemu_ram_free(mr
->ram_block
);
954 static bool memory_region_need_escape(char c
)
956 return c
== '/' || c
== '[' || c
== '\\' || c
== ']';
959 static char *memory_region_escape_name(const char *name
)
966 for (p
= name
; *p
; p
++) {
967 bytes
+= memory_region_need_escape(*p
) ? 4 : 1;
969 if (bytes
== p
- name
) {
970 return g_memdup(name
, bytes
+ 1);
973 escaped
= g_malloc(bytes
+ 1);
974 for (p
= name
, q
= escaped
; *p
; p
++) {
976 if (unlikely(memory_region_need_escape(c
))) {
979 *q
++ = "0123456789abcdef"[c
>> 4];
980 c
= "0123456789abcdef"[c
& 15];
988 void memory_region_init(MemoryRegion
*mr
,
993 object_initialize(mr
, sizeof(*mr
), TYPE_MEMORY_REGION
);
994 mr
->size
= int128_make64(size
);
995 if (size
== UINT64_MAX
) {
996 mr
->size
= int128_2_64();
998 mr
->name
= g_strdup(name
);
1000 mr
->ram_block
= NULL
;
1003 char *escaped_name
= memory_region_escape_name(name
);
1004 char *name_array
= g_strdup_printf("%s[*]", escaped_name
);
1007 owner
= container_get(qdev_get_machine(), "/unattached");
1010 object_property_add_child(owner
, name_array
, OBJECT(mr
), &error_abort
);
1011 object_unref(OBJECT(mr
));
1013 g_free(escaped_name
);
1017 static void memory_region_get_addr(Object
*obj
, Visitor
*v
, const char *name
,
1018 void *opaque
, Error
**errp
)
1020 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1021 uint64_t value
= mr
->addr
;
1023 visit_type_uint64(v
, name
, &value
, errp
);
1026 static void memory_region_get_container(Object
*obj
, Visitor
*v
,
1027 const char *name
, void *opaque
,
1030 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1031 gchar
*path
= (gchar
*)"";
1033 if (mr
->container
) {
1034 path
= object_get_canonical_path(OBJECT(mr
->container
));
1036 visit_type_str(v
, name
, &path
, errp
);
1037 if (mr
->container
) {
1042 static Object
*memory_region_resolve_container(Object
*obj
, void *opaque
,
1045 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1047 return OBJECT(mr
->container
);
1050 static void memory_region_get_priority(Object
*obj
, Visitor
*v
,
1051 const char *name
, void *opaque
,
1054 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1055 int32_t value
= mr
->priority
;
1057 visit_type_int32(v
, name
, &value
, errp
);
1060 static void memory_region_get_size(Object
*obj
, Visitor
*v
, const char *name
,
1061 void *opaque
, Error
**errp
)
1063 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1064 uint64_t value
= memory_region_size(mr
);
1066 visit_type_uint64(v
, name
, &value
, errp
);
1069 static void memory_region_initfn(Object
*obj
)
1071 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1074 mr
->ops
= &unassigned_mem_ops
;
1076 mr
->romd_mode
= true;
1077 mr
->global_locking
= true;
1078 mr
->destructor
= memory_region_destructor_none
;
1079 QTAILQ_INIT(&mr
->subregions
);
1080 QTAILQ_INIT(&mr
->coalesced
);
1082 op
= object_property_add(OBJECT(mr
), "container",
1083 "link<" TYPE_MEMORY_REGION
">",
1084 memory_region_get_container
,
1085 NULL
, /* memory_region_set_container */
1086 NULL
, NULL
, &error_abort
);
1087 op
->resolve
= memory_region_resolve_container
;
1089 object_property_add(OBJECT(mr
), "addr", "uint64",
1090 memory_region_get_addr
,
1091 NULL
, /* memory_region_set_addr */
1092 NULL
, NULL
, &error_abort
);
1093 object_property_add(OBJECT(mr
), "priority", "uint32",
1094 memory_region_get_priority
,
1095 NULL
, /* memory_region_set_priority */
1096 NULL
, NULL
, &error_abort
);
1097 object_property_add(OBJECT(mr
), "size", "uint64",
1098 memory_region_get_size
,
1099 NULL
, /* memory_region_set_size, */
1100 NULL
, NULL
, &error_abort
);
1103 static uint64_t unassigned_mem_read(void *opaque
, hwaddr addr
,
1106 #ifdef DEBUG_UNASSIGNED
1107 printf("Unassigned mem read " TARGET_FMT_plx
"\n", addr
);
1109 if (current_cpu
!= NULL
) {
1110 cpu_unassigned_access(current_cpu
, addr
, false, false, 0, size
);
1115 static void unassigned_mem_write(void *opaque
, hwaddr addr
,
1116 uint64_t val
, unsigned size
)
1118 #ifdef DEBUG_UNASSIGNED
1119 printf("Unassigned mem write " TARGET_FMT_plx
" = 0x%"PRIx64
"\n", addr
, val
);
1121 if (current_cpu
!= NULL
) {
1122 cpu_unassigned_access(current_cpu
, addr
, true, false, 0, size
);
1126 static bool unassigned_mem_accepts(void *opaque
, hwaddr addr
,
1127 unsigned size
, bool is_write
)
1132 const MemoryRegionOps unassigned_mem_ops
= {
1133 .valid
.accepts
= unassigned_mem_accepts
,
1134 .endianness
= DEVICE_NATIVE_ENDIAN
,
1137 bool memory_region_access_valid(MemoryRegion
*mr
,
1142 int access_size_min
, access_size_max
;
1145 if (!mr
->ops
->valid
.unaligned
&& (addr
& (size
- 1))) {
1149 if (!mr
->ops
->valid
.accepts
) {
1153 access_size_min
= mr
->ops
->valid
.min_access_size
;
1154 if (!mr
->ops
->valid
.min_access_size
) {
1155 access_size_min
= 1;
1158 access_size_max
= mr
->ops
->valid
.max_access_size
;
1159 if (!mr
->ops
->valid
.max_access_size
) {
1160 access_size_max
= 4;
1163 access_size
= MAX(MIN(size
, access_size_max
), access_size_min
);
1164 for (i
= 0; i
< size
; i
+= access_size
) {
1165 if (!mr
->ops
->valid
.accepts(mr
->opaque
, addr
+ i
, access_size
,
1174 static MemTxResult
memory_region_dispatch_read1(MemoryRegion
*mr
,
1182 if (mr
->ops
->read
) {
1183 return access_with_adjusted_size(addr
, pval
, size
,
1184 mr
->ops
->impl
.min_access_size
,
1185 mr
->ops
->impl
.max_access_size
,
1186 memory_region_read_accessor
,
1188 } else if (mr
->ops
->read_with_attrs
) {
1189 return access_with_adjusted_size(addr
, pval
, size
,
1190 mr
->ops
->impl
.min_access_size
,
1191 mr
->ops
->impl
.max_access_size
,
1192 memory_region_read_with_attrs_accessor
,
1195 return access_with_adjusted_size(addr
, pval
, size
, 1, 4,
1196 memory_region_oldmmio_read_accessor
,
1201 MemTxResult
memory_region_dispatch_read(MemoryRegion
*mr
,
1209 if (!memory_region_access_valid(mr
, addr
, size
, false)) {
1210 *pval
= unassigned_mem_read(mr
, addr
, size
);
1211 return MEMTX_DECODE_ERROR
;
1214 r
= memory_region_dispatch_read1(mr
, addr
, pval
, size
, attrs
);
1215 adjust_endianness(mr
, pval
, size
);
1219 /* Return true if an eventfd was signalled */
1220 static bool memory_region_dispatch_write_eventfds(MemoryRegion
*mr
,
1226 MemoryRegionIoeventfd ioeventfd
= {
1227 .addr
= addrrange_make(int128_make64(addr
), int128_make64(size
)),
1232 for (i
= 0; i
< mr
->ioeventfd_nb
; i
++) {
1233 ioeventfd
.match_data
= mr
->ioeventfds
[i
].match_data
;
1234 ioeventfd
.e
= mr
->ioeventfds
[i
].e
;
1236 if (memory_region_ioeventfd_equal(ioeventfd
, mr
->ioeventfds
[i
])) {
1237 event_notifier_set(ioeventfd
.e
);
1245 MemTxResult
memory_region_dispatch_write(MemoryRegion
*mr
,
1251 if (!memory_region_access_valid(mr
, addr
, size
, true)) {
1252 unassigned_mem_write(mr
, addr
, data
, size
);
1253 return MEMTX_DECODE_ERROR
;
1256 adjust_endianness(mr
, &data
, size
);
1258 if ((!kvm_eventfds_enabled()) &&
1259 memory_region_dispatch_write_eventfds(mr
, addr
, data
, size
, attrs
)) {
1263 if (mr
->ops
->write
) {
1264 return access_with_adjusted_size(addr
, &data
, size
,
1265 mr
->ops
->impl
.min_access_size
,
1266 mr
->ops
->impl
.max_access_size
,
1267 memory_region_write_accessor
, mr
,
1269 } else if (mr
->ops
->write_with_attrs
) {
1271 access_with_adjusted_size(addr
, &data
, size
,
1272 mr
->ops
->impl
.min_access_size
,
1273 mr
->ops
->impl
.max_access_size
,
1274 memory_region_write_with_attrs_accessor
,
1277 return access_with_adjusted_size(addr
, &data
, size
, 1, 4,
1278 memory_region_oldmmio_write_accessor
,
1283 void memory_region_init_io(MemoryRegion
*mr
,
1285 const MemoryRegionOps
*ops
,
1290 memory_region_init(mr
, owner
, name
, size
);
1291 mr
->ops
= ops
? ops
: &unassigned_mem_ops
;
1292 mr
->opaque
= opaque
;
1293 mr
->terminates
= true;
1296 void memory_region_init_ram(MemoryRegion
*mr
,
1302 memory_region_init(mr
, owner
, name
, size
);
1304 mr
->terminates
= true;
1305 mr
->destructor
= memory_region_destructor_ram
;
1306 mr
->ram_block
= qemu_ram_alloc(size
, mr
, errp
);
1307 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1310 void memory_region_init_resizeable_ram(MemoryRegion
*mr
,
1315 void (*resized
)(const char*,
1320 memory_region_init(mr
, owner
, name
, size
);
1322 mr
->terminates
= true;
1323 mr
->destructor
= memory_region_destructor_ram
;
1324 mr
->ram_block
= qemu_ram_alloc_resizeable(size
, max_size
, resized
,
1326 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1330 void memory_region_init_ram_from_file(MemoryRegion
*mr
,
1331 struct Object
*owner
,
1338 memory_region_init(mr
, owner
, name
, size
);
1340 mr
->terminates
= true;
1341 mr
->destructor
= memory_region_destructor_ram
;
1342 mr
->ram_block
= qemu_ram_alloc_from_file(size
, mr
, share
, path
, errp
);
1343 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1347 void memory_region_init_ram_ptr(MemoryRegion
*mr
,
1353 memory_region_init(mr
, owner
, name
, size
);
1355 mr
->terminates
= true;
1356 mr
->destructor
= memory_region_destructor_ram
;
1357 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1359 /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL. */
1360 assert(ptr
!= NULL
);
1361 mr
->ram_block
= qemu_ram_alloc_from_ptr(size
, ptr
, mr
, &error_fatal
);
1364 void memory_region_set_skip_dump(MemoryRegion
*mr
)
1366 mr
->skip_dump
= true;
1369 void memory_region_init_alias(MemoryRegion
*mr
,
1376 memory_region_init(mr
, owner
, name
, size
);
1378 mr
->alias_offset
= offset
;
1381 void memory_region_init_rom_device(MemoryRegion
*mr
,
1383 const MemoryRegionOps
*ops
,
1389 memory_region_init(mr
, owner
, name
, size
);
1391 mr
->opaque
= opaque
;
1392 mr
->terminates
= true;
1393 mr
->rom_device
= true;
1394 mr
->destructor
= memory_region_destructor_rom_device
;
1395 mr
->ram_block
= qemu_ram_alloc(size
, mr
, errp
);
1398 void memory_region_init_iommu(MemoryRegion
*mr
,
1400 const MemoryRegionIOMMUOps
*ops
,
1404 memory_region_init(mr
, owner
, name
, size
);
1405 mr
->iommu_ops
= ops
,
1406 mr
->terminates
= true; /* then re-forwards */
1407 notifier_list_init(&mr
->iommu_notify
);
1410 static void memory_region_finalize(Object
*obj
)
1412 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1414 assert(!mr
->container
);
1416 /* We know the region is not visible in any address space (it
1417 * does not have a container and cannot be a root either because
1418 * it has no references, so we can blindly clear mr->enabled.
1419 * memory_region_set_enabled instead could trigger a transaction
1420 * and cause an infinite loop.
1422 mr
->enabled
= false;
1423 memory_region_transaction_begin();
1424 while (!QTAILQ_EMPTY(&mr
->subregions
)) {
1425 MemoryRegion
*subregion
= QTAILQ_FIRST(&mr
->subregions
);
1426 memory_region_del_subregion(mr
, subregion
);
1428 memory_region_transaction_commit();
1431 memory_region_clear_coalescing(mr
);
1432 g_free((char *)mr
->name
);
1433 g_free(mr
->ioeventfds
);
1436 Object
*memory_region_owner(MemoryRegion
*mr
)
1438 Object
*obj
= OBJECT(mr
);
1442 void memory_region_ref(MemoryRegion
*mr
)
1444 /* MMIO callbacks most likely will access data that belongs
1445 * to the owner, hence the need to ref/unref the owner whenever
1446 * the memory region is in use.
1448 * The memory region is a child of its owner. As long as the
1449 * owner doesn't call unparent itself on the memory region,
1450 * ref-ing the owner will also keep the memory region alive.
1451 * Memory regions without an owner are supposed to never go away;
1452 * we do not ref/unref them because it slows down DMA sensibly.
1454 if (mr
&& mr
->owner
) {
1455 object_ref(mr
->owner
);
1459 void memory_region_unref(MemoryRegion
*mr
)
1461 if (mr
&& mr
->owner
) {
1462 object_unref(mr
->owner
);
1466 uint64_t memory_region_size(MemoryRegion
*mr
)
1468 if (int128_eq(mr
->size
, int128_2_64())) {
1471 return int128_get64(mr
->size
);
1474 const char *memory_region_name(const MemoryRegion
*mr
)
1477 ((MemoryRegion
*)mr
)->name
=
1478 object_get_canonical_path_component(OBJECT(mr
));
1483 bool memory_region_is_skip_dump(MemoryRegion
*mr
)
1485 return mr
->skip_dump
;
1488 uint8_t memory_region_get_dirty_log_mask(MemoryRegion
*mr
)
1490 uint8_t mask
= mr
->dirty_log_mask
;
1491 if (global_dirty_log
) {
1492 mask
|= (1 << DIRTY_MEMORY_MIGRATION
);
1497 bool memory_region_is_logging(MemoryRegion
*mr
, uint8_t client
)
1499 return memory_region_get_dirty_log_mask(mr
) & (1 << client
);
1502 void memory_region_register_iommu_notifier(MemoryRegion
*mr
, Notifier
*n
)
1504 notifier_list_add(&mr
->iommu_notify
, n
);
1507 void memory_region_iommu_replay(MemoryRegion
*mr
, Notifier
*n
,
1508 hwaddr granularity
, bool is_write
)
1511 IOMMUTLBEntry iotlb
;
1513 for (addr
= 0; addr
< memory_region_size(mr
); addr
+= granularity
) {
1514 iotlb
= mr
->iommu_ops
->translate(mr
, addr
, is_write
);
1515 if (iotlb
.perm
!= IOMMU_NONE
) {
1516 n
->notify(n
, &iotlb
);
1519 /* if (2^64 - MR size) < granularity, it's possible to get an
1520 * infinite loop here. This should catch such a wraparound */
1521 if ((addr
+ granularity
) < addr
) {
1527 void memory_region_unregister_iommu_notifier(Notifier
*n
)
1532 void memory_region_notify_iommu(MemoryRegion
*mr
,
1533 IOMMUTLBEntry entry
)
1535 assert(memory_region_is_iommu(mr
));
1536 notifier_list_notify(&mr
->iommu_notify
, &entry
);
1539 void memory_region_set_log(MemoryRegion
*mr
, bool log
, unsigned client
)
1541 uint8_t mask
= 1 << client
;
1542 uint8_t old_logging
;
1544 assert(client
== DIRTY_MEMORY_VGA
);
1545 old_logging
= mr
->vga_logging_count
;
1546 mr
->vga_logging_count
+= log
? 1 : -1;
1547 if (!!old_logging
== !!mr
->vga_logging_count
) {
1551 memory_region_transaction_begin();
1552 mr
->dirty_log_mask
= (mr
->dirty_log_mask
& ~mask
) | (log
* mask
);
1553 memory_region_update_pending
|= mr
->enabled
;
1554 memory_region_transaction_commit();
1557 bool memory_region_get_dirty(MemoryRegion
*mr
, hwaddr addr
,
1558 hwaddr size
, unsigned client
)
1560 assert(mr
->ram_block
);
1561 return cpu_physical_memory_get_dirty(memory_region_get_ram_addr(mr
) + addr
,
1565 void memory_region_set_dirty(MemoryRegion
*mr
, hwaddr addr
,
1568 assert(mr
->ram_block
);
1569 cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr
) + addr
,
1571 memory_region_get_dirty_log_mask(mr
));
1574 bool memory_region_test_and_clear_dirty(MemoryRegion
*mr
, hwaddr addr
,
1575 hwaddr size
, unsigned client
)
1577 assert(mr
->ram_block
);
1578 return cpu_physical_memory_test_and_clear_dirty(
1579 memory_region_get_ram_addr(mr
) + addr
, size
, client
);
1583 void memory_region_sync_dirty_bitmap(MemoryRegion
*mr
)
1588 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
1589 FlatView
*view
= address_space_get_flatview(as
);
1590 FOR_EACH_FLAT_RANGE(fr
, view
) {
1592 MEMORY_LISTENER_UPDATE_REGION(fr
, as
, Forward
, log_sync
);
1595 flatview_unref(view
);
1599 void memory_region_set_readonly(MemoryRegion
*mr
, bool readonly
)
1601 if (mr
->readonly
!= readonly
) {
1602 memory_region_transaction_begin();
1603 mr
->readonly
= readonly
;
1604 memory_region_update_pending
|= mr
->enabled
;
1605 memory_region_transaction_commit();
1609 void memory_region_rom_device_set_romd(MemoryRegion
*mr
, bool romd_mode
)
1611 if (mr
->romd_mode
!= romd_mode
) {
1612 memory_region_transaction_begin();
1613 mr
->romd_mode
= romd_mode
;
1614 memory_region_update_pending
|= mr
->enabled
;
1615 memory_region_transaction_commit();
1619 void memory_region_reset_dirty(MemoryRegion
*mr
, hwaddr addr
,
1620 hwaddr size
, unsigned client
)
1622 assert(mr
->ram_block
);
1623 cpu_physical_memory_test_and_clear_dirty(
1624 memory_region_get_ram_addr(mr
) + addr
, size
, client
);
1627 int memory_region_get_fd(MemoryRegion
*mr
)
1630 return memory_region_get_fd(mr
->alias
);
1633 assert(mr
->ram_block
);
1635 return qemu_get_ram_fd(memory_region_get_ram_addr(mr
));
1638 void *memory_region_get_ram_ptr(MemoryRegion
*mr
)
1641 uint64_t offset
= 0;
1645 offset
+= mr
->alias_offset
;
1648 assert(mr
->ram_block
);
1649 ptr
= qemu_get_ram_ptr(mr
->ram_block
, memory_region_get_ram_addr(mr
));
1652 return ptr
+ offset
;
1655 ram_addr_t
memory_region_get_ram_addr(MemoryRegion
*mr
)
1657 return mr
->ram_block
? mr
->ram_block
->offset
: RAM_ADDR_INVALID
;
1660 void memory_region_ram_resize(MemoryRegion
*mr
, ram_addr_t newsize
, Error
**errp
)
1662 assert(mr
->ram_block
);
1664 qemu_ram_resize(mr
->ram_block
, newsize
, errp
);
1667 static void memory_region_update_coalesced_range_as(MemoryRegion
*mr
, AddressSpace
*as
)
1671 CoalescedMemoryRange
*cmr
;
1673 MemoryRegionSection section
;
1675 view
= address_space_get_flatview(as
);
1676 FOR_EACH_FLAT_RANGE(fr
, view
) {
1678 section
= (MemoryRegionSection
) {
1679 .address_space
= as
,
1680 .offset_within_address_space
= int128_get64(fr
->addr
.start
),
1681 .size
= fr
->addr
.size
,
1684 MEMORY_LISTENER_CALL(coalesced_mmio_del
, Reverse
, §ion
,
1685 int128_get64(fr
->addr
.start
),
1686 int128_get64(fr
->addr
.size
));
1687 QTAILQ_FOREACH(cmr
, &mr
->coalesced
, link
) {
1688 tmp
= addrrange_shift(cmr
->addr
,
1689 int128_sub(fr
->addr
.start
,
1690 int128_make64(fr
->offset_in_region
)));
1691 if (!addrrange_intersects(tmp
, fr
->addr
)) {
1694 tmp
= addrrange_intersection(tmp
, fr
->addr
);
1695 MEMORY_LISTENER_CALL(coalesced_mmio_add
, Forward
, §ion
,
1696 int128_get64(tmp
.start
),
1697 int128_get64(tmp
.size
));
1701 flatview_unref(view
);
1704 static void memory_region_update_coalesced_range(MemoryRegion
*mr
)
1708 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
1709 memory_region_update_coalesced_range_as(mr
, as
);
1713 void memory_region_set_coalescing(MemoryRegion
*mr
)
1715 memory_region_clear_coalescing(mr
);
1716 memory_region_add_coalescing(mr
, 0, int128_get64(mr
->size
));
1719 void memory_region_add_coalescing(MemoryRegion
*mr
,
1723 CoalescedMemoryRange
*cmr
= g_malloc(sizeof(*cmr
));
1725 cmr
->addr
= addrrange_make(int128_make64(offset
), int128_make64(size
));
1726 QTAILQ_INSERT_TAIL(&mr
->coalesced
, cmr
, link
);
1727 memory_region_update_coalesced_range(mr
);
1728 memory_region_set_flush_coalesced(mr
);
1731 void memory_region_clear_coalescing(MemoryRegion
*mr
)
1733 CoalescedMemoryRange
*cmr
;
1734 bool updated
= false;
1736 qemu_flush_coalesced_mmio_buffer();
1737 mr
->flush_coalesced_mmio
= false;
1739 while (!QTAILQ_EMPTY(&mr
->coalesced
)) {
1740 cmr
= QTAILQ_FIRST(&mr
->coalesced
);
1741 QTAILQ_REMOVE(&mr
->coalesced
, cmr
, link
);
1747 memory_region_update_coalesced_range(mr
);
1751 void memory_region_set_flush_coalesced(MemoryRegion
*mr
)
1753 mr
->flush_coalesced_mmio
= true;
1756 void memory_region_clear_flush_coalesced(MemoryRegion
*mr
)
1758 qemu_flush_coalesced_mmio_buffer();
1759 if (QTAILQ_EMPTY(&mr
->coalesced
)) {
1760 mr
->flush_coalesced_mmio
= false;
1764 void memory_region_set_global_locking(MemoryRegion
*mr
)
1766 mr
->global_locking
= true;
1769 void memory_region_clear_global_locking(MemoryRegion
*mr
)
1771 mr
->global_locking
= false;
1774 static bool userspace_eventfd_warning
;
1776 void memory_region_add_eventfd(MemoryRegion
*mr
,
1783 MemoryRegionIoeventfd mrfd
= {
1784 .addr
.start
= int128_make64(addr
),
1785 .addr
.size
= int128_make64(size
),
1786 .match_data
= match_data
,
1792 if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
1793 userspace_eventfd_warning
))) {
1794 userspace_eventfd_warning
= true;
1795 error_report("Using eventfd without MMIO binding in KVM. "
1796 "Suboptimal performance expected");
1800 adjust_endianness(mr
, &mrfd
.data
, size
);
1802 memory_region_transaction_begin();
1803 for (i
= 0; i
< mr
->ioeventfd_nb
; ++i
) {
1804 if (memory_region_ioeventfd_before(mrfd
, mr
->ioeventfds
[i
])) {
1809 mr
->ioeventfds
= g_realloc(mr
->ioeventfds
,
1810 sizeof(*mr
->ioeventfds
) * mr
->ioeventfd_nb
);
1811 memmove(&mr
->ioeventfds
[i
+1], &mr
->ioeventfds
[i
],
1812 sizeof(*mr
->ioeventfds
) * (mr
->ioeventfd_nb
-1 - i
));
1813 mr
->ioeventfds
[i
] = mrfd
;
1814 ioeventfd_update_pending
|= mr
->enabled
;
1815 memory_region_transaction_commit();
1818 void memory_region_del_eventfd(MemoryRegion
*mr
,
1825 MemoryRegionIoeventfd mrfd
= {
1826 .addr
.start
= int128_make64(addr
),
1827 .addr
.size
= int128_make64(size
),
1828 .match_data
= match_data
,
1835 adjust_endianness(mr
, &mrfd
.data
, size
);
1837 memory_region_transaction_begin();
1838 for (i
= 0; i
< mr
->ioeventfd_nb
; ++i
) {
1839 if (memory_region_ioeventfd_equal(mrfd
, mr
->ioeventfds
[i
])) {
1843 assert(i
!= mr
->ioeventfd_nb
);
1844 memmove(&mr
->ioeventfds
[i
], &mr
->ioeventfds
[i
+1],
1845 sizeof(*mr
->ioeventfds
) * (mr
->ioeventfd_nb
- (i
+1)));
1847 mr
->ioeventfds
= g_realloc(mr
->ioeventfds
,
1848 sizeof(*mr
->ioeventfds
)*mr
->ioeventfd_nb
+ 1);
1849 ioeventfd_update_pending
|= mr
->enabled
;
1850 memory_region_transaction_commit();
1853 static void memory_region_update_container_subregions(MemoryRegion
*subregion
)
1855 MemoryRegion
*mr
= subregion
->container
;
1856 MemoryRegion
*other
;
1858 memory_region_transaction_begin();
1860 memory_region_ref(subregion
);
1861 QTAILQ_FOREACH(other
, &mr
->subregions
, subregions_link
) {
1862 if (subregion
->priority
>= other
->priority
) {
1863 QTAILQ_INSERT_BEFORE(other
, subregion
, subregions_link
);
1867 QTAILQ_INSERT_TAIL(&mr
->subregions
, subregion
, subregions_link
);
1869 memory_region_update_pending
|= mr
->enabled
&& subregion
->enabled
;
1870 memory_region_transaction_commit();
1873 static void memory_region_add_subregion_common(MemoryRegion
*mr
,
1875 MemoryRegion
*subregion
)
1877 assert(!subregion
->container
);
1878 subregion
->container
= mr
;
1879 subregion
->addr
= offset
;
1880 memory_region_update_container_subregions(subregion
);
1883 void memory_region_add_subregion(MemoryRegion
*mr
,
1885 MemoryRegion
*subregion
)
1887 subregion
->priority
= 0;
1888 memory_region_add_subregion_common(mr
, offset
, subregion
);
1891 void memory_region_add_subregion_overlap(MemoryRegion
*mr
,
1893 MemoryRegion
*subregion
,
1896 subregion
->priority
= priority
;
1897 memory_region_add_subregion_common(mr
, offset
, subregion
);
1900 void memory_region_del_subregion(MemoryRegion
*mr
,
1901 MemoryRegion
*subregion
)
1903 memory_region_transaction_begin();
1904 assert(subregion
->container
== mr
);
1905 subregion
->container
= NULL
;
1906 QTAILQ_REMOVE(&mr
->subregions
, subregion
, subregions_link
);
1907 memory_region_unref(subregion
);
1908 memory_region_update_pending
|= mr
->enabled
&& subregion
->enabled
;
1909 memory_region_transaction_commit();
1912 void memory_region_set_enabled(MemoryRegion
*mr
, bool enabled
)
1914 if (enabled
== mr
->enabled
) {
1917 memory_region_transaction_begin();
1918 mr
->enabled
= enabled
;
1919 memory_region_update_pending
= true;
1920 memory_region_transaction_commit();
1923 void memory_region_set_size(MemoryRegion
*mr
, uint64_t size
)
1925 Int128 s
= int128_make64(size
);
1927 if (size
== UINT64_MAX
) {
1930 if (int128_eq(s
, mr
->size
)) {
1933 memory_region_transaction_begin();
1935 memory_region_update_pending
= true;
1936 memory_region_transaction_commit();
1939 static void memory_region_readd_subregion(MemoryRegion
*mr
)
1941 MemoryRegion
*container
= mr
->container
;
1944 memory_region_transaction_begin();
1945 memory_region_ref(mr
);
1946 memory_region_del_subregion(container
, mr
);
1947 mr
->container
= container
;
1948 memory_region_update_container_subregions(mr
);
1949 memory_region_unref(mr
);
1950 memory_region_transaction_commit();
1954 void memory_region_set_address(MemoryRegion
*mr
, hwaddr addr
)
1956 if (addr
!= mr
->addr
) {
1958 memory_region_readd_subregion(mr
);
1962 void memory_region_set_alias_offset(MemoryRegion
*mr
, hwaddr offset
)
1966 if (offset
== mr
->alias_offset
) {
1970 memory_region_transaction_begin();
1971 mr
->alias_offset
= offset
;
1972 memory_region_update_pending
|= mr
->enabled
;
1973 memory_region_transaction_commit();
1976 uint64_t memory_region_get_alignment(const MemoryRegion
*mr
)
1981 static int cmp_flatrange_addr(const void *addr_
, const void *fr_
)
1983 const AddrRange
*addr
= addr_
;
1984 const FlatRange
*fr
= fr_
;
1986 if (int128_le(addrrange_end(*addr
), fr
->addr
.start
)) {
1988 } else if (int128_ge(addr
->start
, addrrange_end(fr
->addr
))) {
1994 static FlatRange
*flatview_lookup(FlatView
*view
, AddrRange addr
)
1996 return bsearch(&addr
, view
->ranges
, view
->nr
,
1997 sizeof(FlatRange
), cmp_flatrange_addr
);
2000 bool memory_region_is_mapped(MemoryRegion
*mr
)
2002 return mr
->container
? true : false;
2005 /* Same as memory_region_find, but it does not add a reference to the
2006 * returned region. It must be called from an RCU critical section.
2008 static MemoryRegionSection
memory_region_find_rcu(MemoryRegion
*mr
,
2009 hwaddr addr
, uint64_t size
)
2011 MemoryRegionSection ret
= { .mr
= NULL
};
2019 for (root
= mr
; root
->container
; ) {
2020 root
= root
->container
;
2024 as
= memory_region_to_address_space(root
);
2028 range
= addrrange_make(int128_make64(addr
), int128_make64(size
));
2030 view
= atomic_rcu_read(&as
->current_map
);
2031 fr
= flatview_lookup(view
, range
);
2036 while (fr
> view
->ranges
&& addrrange_intersects(fr
[-1].addr
, range
)) {
2041 ret
.address_space
= as
;
2042 range
= addrrange_intersection(range
, fr
->addr
);
2043 ret
.offset_within_region
= fr
->offset_in_region
;
2044 ret
.offset_within_region
+= int128_get64(int128_sub(range
.start
,
2046 ret
.size
= range
.size
;
2047 ret
.offset_within_address_space
= int128_get64(range
.start
);
2048 ret
.readonly
= fr
->readonly
;
2052 MemoryRegionSection
memory_region_find(MemoryRegion
*mr
,
2053 hwaddr addr
, uint64_t size
)
2055 MemoryRegionSection ret
;
2057 ret
= memory_region_find_rcu(mr
, addr
, size
);
2059 memory_region_ref(ret
.mr
);
2065 bool memory_region_present(MemoryRegion
*container
, hwaddr addr
)
2070 mr
= memory_region_find_rcu(container
, addr
, 1).mr
;
2072 return mr
&& mr
!= container
;
2075 void address_space_sync_dirty_bitmap(AddressSpace
*as
)
2080 view
= address_space_get_flatview(as
);
2081 FOR_EACH_FLAT_RANGE(fr
, view
) {
2082 MEMORY_LISTENER_UPDATE_REGION(fr
, as
, Forward
, log_sync
);
2084 flatview_unref(view
);
2087 void memory_global_dirty_log_start(void)
2089 global_dirty_log
= true;
2091 MEMORY_LISTENER_CALL_GLOBAL(log_global_start
, Forward
);
2093 /* Refresh DIRTY_LOG_MIGRATION bit. */
2094 memory_region_transaction_begin();
2095 memory_region_update_pending
= true;
2096 memory_region_transaction_commit();
2099 void memory_global_dirty_log_stop(void)
2101 global_dirty_log
= false;
2103 /* Refresh DIRTY_LOG_MIGRATION bit. */
2104 memory_region_transaction_begin();
2105 memory_region_update_pending
= true;
2106 memory_region_transaction_commit();
2108 MEMORY_LISTENER_CALL_GLOBAL(log_global_stop
, Reverse
);
2111 static void listener_add_address_space(MemoryListener
*listener
,
2117 if (listener
->address_space_filter
2118 && listener
->address_space_filter
!= as
) {
2122 if (listener
->begin
) {
2123 listener
->begin(listener
);
2125 if (global_dirty_log
) {
2126 if (listener
->log_global_start
) {
2127 listener
->log_global_start(listener
);
2131 view
= address_space_get_flatview(as
);
2132 FOR_EACH_FLAT_RANGE(fr
, view
) {
2133 MemoryRegionSection section
= {
2135 .address_space
= as
,
2136 .offset_within_region
= fr
->offset_in_region
,
2137 .size
= fr
->addr
.size
,
2138 .offset_within_address_space
= int128_get64(fr
->addr
.start
),
2139 .readonly
= fr
->readonly
,
2141 if (fr
->dirty_log_mask
&& listener
->log_start
) {
2142 listener
->log_start(listener
, §ion
, 0, fr
->dirty_log_mask
);
2144 if (listener
->region_add
) {
2145 listener
->region_add(listener
, §ion
);
2148 if (listener
->commit
) {
2149 listener
->commit(listener
);
2151 flatview_unref(view
);
2154 void memory_listener_register(MemoryListener
*listener
, AddressSpace
*filter
)
2156 MemoryListener
*other
= NULL
;
2159 listener
->address_space_filter
= filter
;
2160 if (QTAILQ_EMPTY(&memory_listeners
)
2161 || listener
->priority
>= QTAILQ_LAST(&memory_listeners
,
2162 memory_listeners
)->priority
) {
2163 QTAILQ_INSERT_TAIL(&memory_listeners
, listener
, link
);
2165 QTAILQ_FOREACH(other
, &memory_listeners
, link
) {
2166 if (listener
->priority
< other
->priority
) {
2170 QTAILQ_INSERT_BEFORE(other
, listener
, link
);
2173 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
2174 listener_add_address_space(listener
, as
);
2178 void memory_listener_unregister(MemoryListener
*listener
)
2180 QTAILQ_REMOVE(&memory_listeners
, listener
, link
);
2183 void address_space_init(AddressSpace
*as
, MemoryRegion
*root
, const char *name
)
2185 memory_region_ref(root
);
2186 memory_region_transaction_begin();
2189 as
->malloced
= false;
2190 as
->current_map
= g_new(FlatView
, 1);
2191 flatview_init(as
->current_map
);
2192 as
->ioeventfd_nb
= 0;
2193 as
->ioeventfds
= NULL
;
2194 QTAILQ_INSERT_TAIL(&address_spaces
, as
, address_spaces_link
);
2195 as
->name
= g_strdup(name
? name
: "anonymous");
2196 address_space_init_dispatch(as
);
2197 memory_region_update_pending
|= root
->enabled
;
2198 memory_region_transaction_commit();
2201 static void do_address_space_destroy(AddressSpace
*as
)
2203 MemoryListener
*listener
;
2204 bool do_free
= as
->malloced
;
2206 address_space_destroy_dispatch(as
);
2208 QTAILQ_FOREACH(listener
, &memory_listeners
, link
) {
2209 assert(listener
->address_space_filter
!= as
);
2212 flatview_unref(as
->current_map
);
2214 g_free(as
->ioeventfds
);
2215 memory_region_unref(as
->root
);
2221 AddressSpace
*address_space_init_shareable(MemoryRegion
*root
, const char *name
)
2225 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
2226 if (root
== as
->root
&& as
->malloced
) {
2232 as
= g_malloc0(sizeof *as
);
2233 address_space_init(as
, root
, name
);
2234 as
->malloced
= true;
2238 void address_space_destroy(AddressSpace
*as
)
2240 MemoryRegion
*root
= as
->root
;
2243 if (as
->ref_count
) {
2246 /* Flush out anything from MemoryListeners listening in on this */
2247 memory_region_transaction_begin();
2249 memory_region_transaction_commit();
2250 QTAILQ_REMOVE(&address_spaces
, as
, address_spaces_link
);
2251 address_space_unregister(as
);
2253 /* At this point, as->dispatch and as->current_map are dummy
2254 * entries that the guest should never use. Wait for the old
2255 * values to expire before freeing the data.
2258 call_rcu(as
, do_address_space_destroy
, rcu
);
2261 typedef struct MemoryRegionList MemoryRegionList
;
2263 struct MemoryRegionList
{
2264 const MemoryRegion
*mr
;
2265 QTAILQ_ENTRY(MemoryRegionList
) queue
;
2268 typedef QTAILQ_HEAD(queue
, MemoryRegionList
) MemoryRegionListHead
;
2270 static void mtree_print_mr(fprintf_function mon_printf
, void *f
,
2271 const MemoryRegion
*mr
, unsigned int level
,
2273 MemoryRegionListHead
*alias_print_queue
)
2275 MemoryRegionList
*new_ml
, *ml
, *next_ml
;
2276 MemoryRegionListHead submr_print_queue
;
2277 const MemoryRegion
*submr
;
2284 for (i
= 0; i
< level
; i
++) {
2289 MemoryRegionList
*ml
;
2292 /* check if the alias is already in the queue */
2293 QTAILQ_FOREACH(ml
, alias_print_queue
, queue
) {
2294 if (ml
->mr
== mr
->alias
) {
2300 ml
= g_new(MemoryRegionList
, 1);
2302 QTAILQ_INSERT_TAIL(alias_print_queue
, ml
, queue
);
2304 mon_printf(f
, TARGET_FMT_plx
"-" TARGET_FMT_plx
2305 " (prio %d, %c%c): alias %s @%s " TARGET_FMT_plx
2306 "-" TARGET_FMT_plx
"%s\n",
2309 + (int128_nz(mr
->size
) ?
2310 (hwaddr
)int128_get64(int128_sub(mr
->size
,
2311 int128_one())) : 0),
2313 mr
->romd_mode
? 'R' : '-',
2314 !mr
->readonly
&& !(mr
->rom_device
&& mr
->romd_mode
) ? 'W'
2316 memory_region_name(mr
),
2317 memory_region_name(mr
->alias
),
2320 + (int128_nz(mr
->size
) ?
2321 (hwaddr
)int128_get64(int128_sub(mr
->size
,
2322 int128_one())) : 0),
2323 mr
->enabled
? "" : " [disabled]");
2326 TARGET_FMT_plx
"-" TARGET_FMT_plx
" (prio %d, %c%c): %s%s\n",
2329 + (int128_nz(mr
->size
) ?
2330 (hwaddr
)int128_get64(int128_sub(mr
->size
,
2331 int128_one())) : 0),
2333 mr
->romd_mode
? 'R' : '-',
2334 !mr
->readonly
&& !(mr
->rom_device
&& mr
->romd_mode
) ? 'W'
2336 memory_region_name(mr
),
2337 mr
->enabled
? "" : " [disabled]");
2340 QTAILQ_INIT(&submr_print_queue
);
2342 QTAILQ_FOREACH(submr
, &mr
->subregions
, subregions_link
) {
2343 new_ml
= g_new(MemoryRegionList
, 1);
2345 QTAILQ_FOREACH(ml
, &submr_print_queue
, queue
) {
2346 if (new_ml
->mr
->addr
< ml
->mr
->addr
||
2347 (new_ml
->mr
->addr
== ml
->mr
->addr
&&
2348 new_ml
->mr
->priority
> ml
->mr
->priority
)) {
2349 QTAILQ_INSERT_BEFORE(ml
, new_ml
, queue
);
2355 QTAILQ_INSERT_TAIL(&submr_print_queue
, new_ml
, queue
);
2359 QTAILQ_FOREACH(ml
, &submr_print_queue
, queue
) {
2360 mtree_print_mr(mon_printf
, f
, ml
->mr
, level
+ 1, base
+ mr
->addr
,
2364 QTAILQ_FOREACH_SAFE(ml
, &submr_print_queue
, queue
, next_ml
) {
2369 void mtree_info(fprintf_function mon_printf
, void *f
)
2371 MemoryRegionListHead ml_head
;
2372 MemoryRegionList
*ml
, *ml2
;
2375 QTAILQ_INIT(&ml_head
);
2377 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
2378 mon_printf(f
, "address-space: %s\n", as
->name
);
2379 mtree_print_mr(mon_printf
, f
, as
->root
, 1, 0, &ml_head
);
2380 mon_printf(f
, "\n");
2383 /* print aliased regions */
2384 QTAILQ_FOREACH(ml
, &ml_head
, queue
) {
2385 mon_printf(f
, "memory-region: %s\n", memory_region_name(ml
->mr
));
2386 mtree_print_mr(mon_printf
, f
, ml
->mr
, 1, 0, &ml_head
);
2387 mon_printf(f
, "\n");
2390 QTAILQ_FOREACH_SAFE(ml
, &ml_head
, queue
, ml2
) {
2395 static const TypeInfo memory_region_info
= {
2396 .parent
= TYPE_OBJECT
,
2397 .name
= TYPE_MEMORY_REGION
,
2398 .instance_size
= sizeof(MemoryRegion
),
2399 .instance_init
= memory_region_initfn
,
2400 .instance_finalize
= memory_region_finalize
,
2403 static void memory_register_types(void)
2405 type_register_static(&memory_region_info
);
2408 type_init(memory_register_types
)