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 static unsigned memory_region_transaction_depth
;
37 static bool memory_region_update_pending
;
38 static bool ioeventfd_update_pending
;
39 static bool global_dirty_log
= false;
41 static QTAILQ_HEAD(memory_listeners
, MemoryListener
) memory_listeners
42 = QTAILQ_HEAD_INITIALIZER(memory_listeners
);
44 static QTAILQ_HEAD(, AddressSpace
) address_spaces
45 = QTAILQ_HEAD_INITIALIZER(address_spaces
);
47 typedef struct AddrRange AddrRange
;
50 * Note that signed integers are needed for negative offsetting in aliases
51 * (large MemoryRegion::alias_offset).
58 static AddrRange
addrrange_make(Int128 start
, Int128 size
)
60 return (AddrRange
) { start
, size
};
63 static bool addrrange_equal(AddrRange r1
, AddrRange r2
)
65 return int128_eq(r1
.start
, r2
.start
) && int128_eq(r1
.size
, r2
.size
);
68 static Int128
addrrange_end(AddrRange r
)
70 return int128_add(r
.start
, r
.size
);
73 static AddrRange
addrrange_shift(AddrRange range
, Int128 delta
)
75 int128_addto(&range
.start
, delta
);
79 static bool addrrange_contains(AddrRange range
, Int128 addr
)
81 return int128_ge(addr
, range
.start
)
82 && int128_lt(addr
, addrrange_end(range
));
85 static bool addrrange_intersects(AddrRange r1
, AddrRange r2
)
87 return addrrange_contains(r1
, r2
.start
)
88 || addrrange_contains(r2
, r1
.start
);
91 static AddrRange
addrrange_intersection(AddrRange r1
, AddrRange r2
)
93 Int128 start
= int128_max(r1
.start
, r2
.start
);
94 Int128 end
= int128_min(addrrange_end(r1
), addrrange_end(r2
));
95 return addrrange_make(start
, int128_sub(end
, start
));
98 enum ListenerDirection
{ Forward
, Reverse
};
100 static bool memory_listener_match(MemoryListener
*listener
,
101 MemoryRegionSection
*section
)
103 return !listener
->address_space_filter
104 || listener
->address_space_filter
== section
->address_space
;
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, \
121 memory_listeners, link) { \
122 if (_listener->_callback) { \
123 _listener->_callback(_listener, ##_args); \
132 #define MEMORY_LISTENER_CALL(_callback, _direction, _section, _args...) \
134 MemoryListener *_listener; \
136 switch (_direction) { \
138 QTAILQ_FOREACH(_listener, &memory_listeners, link) { \
139 if (_listener->_callback \
140 && memory_listener_match(_listener, _section)) { \
141 _listener->_callback(_listener, _section, ##_args); \
146 QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners, \
147 memory_listeners, link) { \
148 if (_listener->_callback \
149 && memory_listener_match(_listener, _section)) { \
150 _listener->_callback(_listener, _section, ##_args); \
159 /* No need to ref/unref .mr, the FlatRange keeps it alive. */
160 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...) \
161 MEMORY_LISTENER_CALL(callback, dir, (&(MemoryRegionSection) { \
163 .address_space = (as), \
164 .offset_within_region = (fr)->offset_in_region, \
165 .size = (fr)->addr.size, \
166 .offset_within_address_space = int128_get64((fr)->addr.start), \
167 .readonly = (fr)->readonly, \
170 struct CoalescedMemoryRange
{
172 QTAILQ_ENTRY(CoalescedMemoryRange
) link
;
175 struct MemoryRegionIoeventfd
{
182 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd a
,
183 MemoryRegionIoeventfd b
)
185 if (int128_lt(a
.addr
.start
, b
.addr
.start
)) {
187 } else if (int128_gt(a
.addr
.start
, b
.addr
.start
)) {
189 } else if (int128_lt(a
.addr
.size
, b
.addr
.size
)) {
191 } else if (int128_gt(a
.addr
.size
, b
.addr
.size
)) {
193 } else if (a
.match_data
< b
.match_data
) {
195 } else if (a
.match_data
> b
.match_data
) {
197 } else if (a
.match_data
) {
198 if (a
.data
< b
.data
) {
200 } else if (a
.data
> b
.data
) {
206 } else if (a
.e
> b
.e
) {
212 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd a
,
213 MemoryRegionIoeventfd b
)
215 return !memory_region_ioeventfd_before(a
, b
)
216 && !memory_region_ioeventfd_before(b
, a
);
219 typedef struct FlatRange FlatRange
;
220 typedef struct FlatView FlatView
;
222 /* Range of memory in the global map. Addresses are absolute. */
225 hwaddr offset_in_region
;
227 uint8_t dirty_log_mask
;
232 /* Flattened global view of current active memory hierarchy. Kept in sorted
240 unsigned nr_allocated
;
243 typedef struct AddressSpaceOps AddressSpaceOps
;
245 #define FOR_EACH_FLAT_RANGE(var, view) \
246 for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
248 static bool flatrange_equal(FlatRange
*a
, FlatRange
*b
)
250 return a
->mr
== b
->mr
251 && addrrange_equal(a
->addr
, b
->addr
)
252 && a
->offset_in_region
== b
->offset_in_region
253 && a
->romd_mode
== b
->romd_mode
254 && a
->readonly
== b
->readonly
;
257 static void flatview_init(FlatView
*view
)
262 view
->nr_allocated
= 0;
265 /* Insert a range into a given position. Caller is responsible for maintaining
268 static void flatview_insert(FlatView
*view
, unsigned pos
, FlatRange
*range
)
270 if (view
->nr
== view
->nr_allocated
) {
271 view
->nr_allocated
= MAX(2 * view
->nr
, 10);
272 view
->ranges
= g_realloc(view
->ranges
,
273 view
->nr_allocated
* sizeof(*view
->ranges
));
275 memmove(view
->ranges
+ pos
+ 1, view
->ranges
+ pos
,
276 (view
->nr
- pos
) * sizeof(FlatRange
));
277 view
->ranges
[pos
] = *range
;
278 memory_region_ref(range
->mr
);
282 static void flatview_destroy(FlatView
*view
)
286 for (i
= 0; i
< view
->nr
; i
++) {
287 memory_region_unref(view
->ranges
[i
].mr
);
289 g_free(view
->ranges
);
293 static void flatview_ref(FlatView
*view
)
295 atomic_inc(&view
->ref
);
298 static void flatview_unref(FlatView
*view
)
300 if (atomic_fetch_dec(&view
->ref
) == 1) {
301 flatview_destroy(view
);
305 static bool can_merge(FlatRange
*r1
, FlatRange
*r2
)
307 return int128_eq(addrrange_end(r1
->addr
), r2
->addr
.start
)
309 && int128_eq(int128_add(int128_make64(r1
->offset_in_region
),
311 int128_make64(r2
->offset_in_region
))
312 && r1
->dirty_log_mask
== r2
->dirty_log_mask
313 && r1
->romd_mode
== r2
->romd_mode
314 && r1
->readonly
== r2
->readonly
;
317 /* Attempt to simplify a view by merging adjacent ranges */
318 static void flatview_simplify(FlatView
*view
)
323 while (i
< view
->nr
) {
326 && can_merge(&view
->ranges
[j
-1], &view
->ranges
[j
])) {
327 int128_addto(&view
->ranges
[i
].addr
.size
, view
->ranges
[j
].addr
.size
);
331 memmove(&view
->ranges
[i
], &view
->ranges
[j
],
332 (view
->nr
- j
) * sizeof(view
->ranges
[j
]));
337 static bool memory_region_big_endian(MemoryRegion
*mr
)
339 #ifdef TARGET_WORDS_BIGENDIAN
340 return mr
->ops
->endianness
!= DEVICE_LITTLE_ENDIAN
;
342 return mr
->ops
->endianness
== DEVICE_BIG_ENDIAN
;
346 static bool memory_region_wrong_endianness(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
, unsigned size
)
357 if (memory_region_wrong_endianness(mr
)) {
362 *data
= bswap16(*data
);
365 *data
= bswap32(*data
);
368 *data
= bswap64(*data
);
376 static hwaddr
memory_region_to_absolute_addr(MemoryRegion
*mr
, hwaddr offset
)
379 hwaddr abs_addr
= offset
;
381 abs_addr
+= mr
->addr
;
382 for (root
= mr
; root
->container
; ) {
383 root
= root
->container
;
384 abs_addr
+= root
->addr
;
390 static int get_cpu_index(void)
393 return current_cpu
->cpu_index
;
398 static MemTxResult
memory_region_oldmmio_read_accessor(MemoryRegion
*mr
,
408 tmp
= mr
->ops
->old_mmio
.read
[ctz32(size
)](mr
->opaque
, addr
);
410 trace_memory_region_subpage_read(get_cpu_index(), mr
, addr
, tmp
, size
);
411 } else if (mr
== &io_mem_notdirty
) {
412 /* Accesses to code which has previously been translated into a TB show
413 * up in the MMIO path, as accesses to the io_mem_notdirty
415 trace_memory_region_tb_read(get_cpu_index(), addr
, tmp
, size
);
416 } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED
) {
417 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
418 trace_memory_region_ops_read(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
420 *value
|= (tmp
& mask
) << shift
;
424 static MemTxResult
memory_region_read_accessor(MemoryRegion
*mr
,
434 tmp
= mr
->ops
->read(mr
->opaque
, addr
, size
);
436 trace_memory_region_subpage_read(get_cpu_index(), mr
, addr
, tmp
, size
);
437 } else if (mr
== &io_mem_notdirty
) {
438 /* Accesses to code which has previously been translated into a TB show
439 * up in the MMIO path, as accesses to the io_mem_notdirty
441 trace_memory_region_tb_read(get_cpu_index(), addr
, tmp
, size
);
442 } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED
) {
443 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
444 trace_memory_region_ops_read(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
446 *value
|= (tmp
& mask
) << shift
;
450 static MemTxResult
memory_region_read_with_attrs_accessor(MemoryRegion
*mr
,
461 r
= mr
->ops
->read_with_attrs(mr
->opaque
, addr
, &tmp
, size
, attrs
);
463 trace_memory_region_subpage_read(get_cpu_index(), mr
, addr
, tmp
, size
);
464 } else if (mr
== &io_mem_notdirty
) {
465 /* Accesses to code which has previously been translated into a TB show
466 * up in the MMIO path, as accesses to the io_mem_notdirty
468 trace_memory_region_tb_read(get_cpu_index(), addr
, tmp
, size
);
469 } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED
) {
470 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
471 trace_memory_region_ops_read(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
473 *value
|= (tmp
& mask
) << shift
;
477 static MemTxResult
memory_region_oldmmio_write_accessor(MemoryRegion
*mr
,
487 tmp
= (*value
>> shift
) & mask
;
489 trace_memory_region_subpage_write(get_cpu_index(), mr
, addr
, tmp
, size
);
490 } else if (mr
== &io_mem_notdirty
) {
491 /* Accesses to code which has previously been translated into a TB show
492 * up in the MMIO path, as accesses to the io_mem_notdirty
494 trace_memory_region_tb_write(get_cpu_index(), addr
, tmp
, size
);
495 } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED
) {
496 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
497 trace_memory_region_ops_write(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
499 mr
->ops
->old_mmio
.write
[ctz32(size
)](mr
->opaque
, addr
, tmp
);
503 static MemTxResult
memory_region_write_accessor(MemoryRegion
*mr
,
513 tmp
= (*value
>> shift
) & mask
;
515 trace_memory_region_subpage_write(get_cpu_index(), mr
, addr
, tmp
, size
);
516 } else if (mr
== &io_mem_notdirty
) {
517 /* Accesses to code which has previously been translated into a TB show
518 * up in the MMIO path, as accesses to the io_mem_notdirty
520 trace_memory_region_tb_write(get_cpu_index(), addr
, tmp
, size
);
521 } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED
) {
522 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
523 trace_memory_region_ops_write(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
525 mr
->ops
->write(mr
->opaque
, addr
, tmp
, size
);
529 static MemTxResult
memory_region_write_with_attrs_accessor(MemoryRegion
*mr
,
539 tmp
= (*value
>> shift
) & mask
;
541 trace_memory_region_subpage_write(get_cpu_index(), mr
, addr
, tmp
, size
);
542 } else if (mr
== &io_mem_notdirty
) {
543 /* Accesses to code which has previously been translated into a TB show
544 * up in the MMIO path, as accesses to the io_mem_notdirty
546 trace_memory_region_tb_write(get_cpu_index(), addr
, tmp
, size
);
547 } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED
) {
548 hwaddr abs_addr
= memory_region_to_absolute_addr(mr
, addr
);
549 trace_memory_region_ops_write(get_cpu_index(), mr
, abs_addr
, tmp
, size
);
551 return mr
->ops
->write_with_attrs(mr
->opaque
, addr
, tmp
, size
, attrs
);
554 static MemTxResult
access_with_adjusted_size(hwaddr addr
,
557 unsigned access_size_min
,
558 unsigned access_size_max
,
559 MemTxResult (*access
)(MemoryRegion
*mr
,
569 uint64_t access_mask
;
570 unsigned access_size
;
572 MemTxResult r
= MEMTX_OK
;
574 if (!access_size_min
) {
577 if (!access_size_max
) {
581 /* FIXME: support unaligned access? */
582 access_size
= MAX(MIN(size
, access_size_max
), access_size_min
);
583 access_mask
= -1ULL >> (64 - access_size
* 8);
584 if (memory_region_big_endian(mr
)) {
585 for (i
= 0; i
< size
; i
+= access_size
) {
586 r
|= access(mr
, addr
+ i
, value
, access_size
,
587 (size
- access_size
- i
) * 8, access_mask
, attrs
);
590 for (i
= 0; i
< size
; i
+= access_size
) {
591 r
|= access(mr
, addr
+ i
, value
, access_size
, i
* 8,
598 static AddressSpace
*memory_region_to_address_space(MemoryRegion
*mr
)
602 while (mr
->container
) {
605 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
606 if (mr
== as
->root
) {
613 /* Render a memory region into the global view. Ranges in @view obscure
616 static void render_memory_region(FlatView
*view
,
622 MemoryRegion
*subregion
;
624 hwaddr offset_in_region
;
634 int128_addto(&base
, int128_make64(mr
->addr
));
635 readonly
|= mr
->readonly
;
637 tmp
= addrrange_make(base
, mr
->size
);
639 if (!addrrange_intersects(tmp
, clip
)) {
643 clip
= addrrange_intersection(tmp
, clip
);
646 int128_subfrom(&base
, int128_make64(mr
->alias
->addr
));
647 int128_subfrom(&base
, int128_make64(mr
->alias_offset
));
648 render_memory_region(view
, mr
->alias
, base
, clip
, readonly
);
652 /* Render subregions in priority order. */
653 QTAILQ_FOREACH(subregion
, &mr
->subregions
, subregions_link
) {
654 render_memory_region(view
, subregion
, base
, clip
, readonly
);
657 if (!mr
->terminates
) {
661 offset_in_region
= int128_get64(int128_sub(clip
.start
, base
));
666 fr
.dirty_log_mask
= memory_region_get_dirty_log_mask(mr
);
667 fr
.romd_mode
= mr
->romd_mode
;
668 fr
.readonly
= readonly
;
670 /* Render the region itself into any gaps left by the current view. */
671 for (i
= 0; i
< view
->nr
&& int128_nz(remain
); ++i
) {
672 if (int128_ge(base
, addrrange_end(view
->ranges
[i
].addr
))) {
675 if (int128_lt(base
, view
->ranges
[i
].addr
.start
)) {
676 now
= int128_min(remain
,
677 int128_sub(view
->ranges
[i
].addr
.start
, base
));
678 fr
.offset_in_region
= offset_in_region
;
679 fr
.addr
= addrrange_make(base
, now
);
680 flatview_insert(view
, i
, &fr
);
682 int128_addto(&base
, now
);
683 offset_in_region
+= int128_get64(now
);
684 int128_subfrom(&remain
, now
);
686 now
= int128_sub(int128_min(int128_add(base
, remain
),
687 addrrange_end(view
->ranges
[i
].addr
)),
689 int128_addto(&base
, now
);
690 offset_in_region
+= int128_get64(now
);
691 int128_subfrom(&remain
, now
);
693 if (int128_nz(remain
)) {
694 fr
.offset_in_region
= offset_in_region
;
695 fr
.addr
= addrrange_make(base
, remain
);
696 flatview_insert(view
, i
, &fr
);
700 /* Render a memory topology into a list of disjoint absolute ranges. */
701 static FlatView
*generate_memory_topology(MemoryRegion
*mr
)
705 view
= g_new(FlatView
, 1);
709 render_memory_region(view
, mr
, int128_zero(),
710 addrrange_make(int128_zero(), int128_2_64()), false);
712 flatview_simplify(view
);
717 static void address_space_add_del_ioeventfds(AddressSpace
*as
,
718 MemoryRegionIoeventfd
*fds_new
,
720 MemoryRegionIoeventfd
*fds_old
,
724 MemoryRegionIoeventfd
*fd
;
725 MemoryRegionSection section
;
727 /* Generate a symmetric difference of the old and new fd sets, adding
728 * and deleting as necessary.
732 while (iold
< fds_old_nb
|| inew
< fds_new_nb
) {
733 if (iold
< fds_old_nb
734 && (inew
== fds_new_nb
735 || memory_region_ioeventfd_before(fds_old
[iold
],
738 section
= (MemoryRegionSection
) {
740 .offset_within_address_space
= int128_get64(fd
->addr
.start
),
741 .size
= fd
->addr
.size
,
743 MEMORY_LISTENER_CALL(eventfd_del
, Forward
, §ion
,
744 fd
->match_data
, fd
->data
, fd
->e
);
746 } else if (inew
< fds_new_nb
747 && (iold
== fds_old_nb
748 || memory_region_ioeventfd_before(fds_new
[inew
],
751 section
= (MemoryRegionSection
) {
753 .offset_within_address_space
= int128_get64(fd
->addr
.start
),
754 .size
= fd
->addr
.size
,
756 MEMORY_LISTENER_CALL(eventfd_add
, Reverse
, §ion
,
757 fd
->match_data
, fd
->data
, fd
->e
);
766 static FlatView
*address_space_get_flatview(AddressSpace
*as
)
771 view
= atomic_rcu_read(&as
->current_map
);
777 static void address_space_update_ioeventfds(AddressSpace
*as
)
781 unsigned ioeventfd_nb
= 0;
782 MemoryRegionIoeventfd
*ioeventfds
= NULL
;
786 view
= address_space_get_flatview(as
);
787 FOR_EACH_FLAT_RANGE(fr
, view
) {
788 for (i
= 0; i
< fr
->mr
->ioeventfd_nb
; ++i
) {
789 tmp
= addrrange_shift(fr
->mr
->ioeventfds
[i
].addr
,
790 int128_sub(fr
->addr
.start
,
791 int128_make64(fr
->offset_in_region
)));
792 if (addrrange_intersects(fr
->addr
, tmp
)) {
794 ioeventfds
= g_realloc(ioeventfds
,
795 ioeventfd_nb
* sizeof(*ioeventfds
));
796 ioeventfds
[ioeventfd_nb
-1] = fr
->mr
->ioeventfds
[i
];
797 ioeventfds
[ioeventfd_nb
-1].addr
= tmp
;
802 address_space_add_del_ioeventfds(as
, ioeventfds
, ioeventfd_nb
,
803 as
->ioeventfds
, as
->ioeventfd_nb
);
805 g_free(as
->ioeventfds
);
806 as
->ioeventfds
= ioeventfds
;
807 as
->ioeventfd_nb
= ioeventfd_nb
;
808 flatview_unref(view
);
811 static void address_space_update_topology_pass(AddressSpace
*as
,
812 const FlatView
*old_view
,
813 const FlatView
*new_view
,
817 FlatRange
*frold
, *frnew
;
819 /* Generate a symmetric difference of the old and new memory maps.
820 * Kill ranges in the old map, and instantiate ranges in the new map.
823 while (iold
< old_view
->nr
|| inew
< new_view
->nr
) {
824 if (iold
< old_view
->nr
) {
825 frold
= &old_view
->ranges
[iold
];
829 if (inew
< new_view
->nr
) {
830 frnew
= &new_view
->ranges
[inew
];
837 || int128_lt(frold
->addr
.start
, frnew
->addr
.start
)
838 || (int128_eq(frold
->addr
.start
, frnew
->addr
.start
)
839 && !flatrange_equal(frold
, frnew
)))) {
840 /* In old but not in new, or in both but attributes changed. */
843 MEMORY_LISTENER_UPDATE_REGION(frold
, as
, Reverse
, region_del
);
847 } else if (frold
&& frnew
&& flatrange_equal(frold
, frnew
)) {
848 /* In both and unchanged (except logging may have changed) */
851 MEMORY_LISTENER_UPDATE_REGION(frnew
, as
, Forward
, region_nop
);
852 if (frnew
->dirty_log_mask
& ~frold
->dirty_log_mask
) {
853 MEMORY_LISTENER_UPDATE_REGION(frnew
, as
, Forward
, log_start
,
854 frold
->dirty_log_mask
,
855 frnew
->dirty_log_mask
);
857 if (frold
->dirty_log_mask
& ~frnew
->dirty_log_mask
) {
858 MEMORY_LISTENER_UPDATE_REGION(frnew
, as
, Reverse
, log_stop
,
859 frold
->dirty_log_mask
,
860 frnew
->dirty_log_mask
);
870 MEMORY_LISTENER_UPDATE_REGION(frnew
, as
, Forward
, region_add
);
879 static void address_space_update_topology(AddressSpace
*as
)
881 FlatView
*old_view
= address_space_get_flatview(as
);
882 FlatView
*new_view
= generate_memory_topology(as
->root
);
884 address_space_update_topology_pass(as
, old_view
, new_view
, false);
885 address_space_update_topology_pass(as
, old_view
, new_view
, true);
887 /* Writes are protected by the BQL. */
888 atomic_rcu_set(&as
->current_map
, new_view
);
889 call_rcu(old_view
, flatview_unref
, rcu
);
891 /* Note that all the old MemoryRegions are still alive up to this
892 * point. This relieves most MemoryListeners from the need to
893 * ref/unref the MemoryRegions they get---unless they use them
894 * outside the iothread mutex, in which case precise reference
895 * counting is necessary.
897 flatview_unref(old_view
);
899 address_space_update_ioeventfds(as
);
902 void memory_region_transaction_begin(void)
904 qemu_flush_coalesced_mmio_buffer();
905 ++memory_region_transaction_depth
;
908 static void memory_region_clear_pending(void)
910 memory_region_update_pending
= false;
911 ioeventfd_update_pending
= false;
914 void memory_region_transaction_commit(void)
918 assert(memory_region_transaction_depth
);
919 --memory_region_transaction_depth
;
920 if (!memory_region_transaction_depth
) {
921 if (memory_region_update_pending
) {
922 MEMORY_LISTENER_CALL_GLOBAL(begin
, Forward
);
924 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
925 address_space_update_topology(as
);
928 MEMORY_LISTENER_CALL_GLOBAL(commit
, Forward
);
929 } else if (ioeventfd_update_pending
) {
930 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
931 address_space_update_ioeventfds(as
);
934 memory_region_clear_pending();
938 static void memory_region_destructor_none(MemoryRegion
*mr
)
942 static void memory_region_destructor_ram(MemoryRegion
*mr
)
944 qemu_ram_free(mr
->ram_block
);
947 static void memory_region_destructor_rom_device(MemoryRegion
*mr
)
949 qemu_ram_free(mr
->ram_block
);
952 static bool memory_region_need_escape(char c
)
954 return c
== '/' || c
== '[' || c
== '\\' || c
== ']';
957 static char *memory_region_escape_name(const char *name
)
964 for (p
= name
; *p
; p
++) {
965 bytes
+= memory_region_need_escape(*p
) ? 4 : 1;
967 if (bytes
== p
- name
) {
968 return g_memdup(name
, bytes
+ 1);
971 escaped
= g_malloc(bytes
+ 1);
972 for (p
= name
, q
= escaped
; *p
; p
++) {
974 if (unlikely(memory_region_need_escape(c
))) {
977 *q
++ = "0123456789abcdef"[c
>> 4];
978 c
= "0123456789abcdef"[c
& 15];
986 void memory_region_init(MemoryRegion
*mr
,
991 object_initialize(mr
, sizeof(*mr
), TYPE_MEMORY_REGION
);
992 mr
->size
= int128_make64(size
);
993 if (size
== UINT64_MAX
) {
994 mr
->size
= int128_2_64();
996 mr
->name
= g_strdup(name
);
998 mr
->ram_block
= NULL
;
1001 char *escaped_name
= memory_region_escape_name(name
);
1002 char *name_array
= g_strdup_printf("%s[*]", escaped_name
);
1005 owner
= container_get(qdev_get_machine(), "/unattached");
1008 object_property_add_child(owner
, name_array
, OBJECT(mr
), &error_abort
);
1009 object_unref(OBJECT(mr
));
1011 g_free(escaped_name
);
1015 static void memory_region_get_addr(Object
*obj
, Visitor
*v
, const char *name
,
1016 void *opaque
, Error
**errp
)
1018 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1019 uint64_t value
= mr
->addr
;
1021 visit_type_uint64(v
, name
, &value
, errp
);
1024 static void memory_region_get_container(Object
*obj
, Visitor
*v
,
1025 const char *name
, void *opaque
,
1028 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1029 gchar
*path
= (gchar
*)"";
1031 if (mr
->container
) {
1032 path
= object_get_canonical_path(OBJECT(mr
->container
));
1034 visit_type_str(v
, name
, &path
, errp
);
1035 if (mr
->container
) {
1040 static Object
*memory_region_resolve_container(Object
*obj
, void *opaque
,
1043 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1045 return OBJECT(mr
->container
);
1048 static void memory_region_get_priority(Object
*obj
, Visitor
*v
,
1049 const char *name
, void *opaque
,
1052 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1053 int32_t value
= mr
->priority
;
1055 visit_type_int32(v
, name
, &value
, errp
);
1058 static void memory_region_get_size(Object
*obj
, Visitor
*v
, const char *name
,
1059 void *opaque
, Error
**errp
)
1061 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1062 uint64_t value
= memory_region_size(mr
);
1064 visit_type_uint64(v
, name
, &value
, errp
);
1067 static void memory_region_initfn(Object
*obj
)
1069 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1072 mr
->ops
= &unassigned_mem_ops
;
1074 mr
->romd_mode
= true;
1075 mr
->global_locking
= true;
1076 mr
->destructor
= memory_region_destructor_none
;
1077 QTAILQ_INIT(&mr
->subregions
);
1078 QTAILQ_INIT(&mr
->coalesced
);
1080 op
= object_property_add(OBJECT(mr
), "container",
1081 "link<" TYPE_MEMORY_REGION
">",
1082 memory_region_get_container
,
1083 NULL
, /* memory_region_set_container */
1084 NULL
, NULL
, &error_abort
);
1085 op
->resolve
= memory_region_resolve_container
;
1087 object_property_add(OBJECT(mr
), "addr", "uint64",
1088 memory_region_get_addr
,
1089 NULL
, /* memory_region_set_addr */
1090 NULL
, NULL
, &error_abort
);
1091 object_property_add(OBJECT(mr
), "priority", "uint32",
1092 memory_region_get_priority
,
1093 NULL
, /* memory_region_set_priority */
1094 NULL
, NULL
, &error_abort
);
1095 object_property_add(OBJECT(mr
), "size", "uint64",
1096 memory_region_get_size
,
1097 NULL
, /* memory_region_set_size, */
1098 NULL
, NULL
, &error_abort
);
1101 static uint64_t unassigned_mem_read(void *opaque
, hwaddr addr
,
1104 #ifdef DEBUG_UNASSIGNED
1105 printf("Unassigned mem read " TARGET_FMT_plx
"\n", addr
);
1107 if (current_cpu
!= NULL
) {
1108 cpu_unassigned_access(current_cpu
, addr
, false, false, 0, size
);
1113 static void unassigned_mem_write(void *opaque
, hwaddr addr
,
1114 uint64_t val
, unsigned size
)
1116 #ifdef DEBUG_UNASSIGNED
1117 printf("Unassigned mem write " TARGET_FMT_plx
" = 0x%"PRIx64
"\n", addr
, val
);
1119 if (current_cpu
!= NULL
) {
1120 cpu_unassigned_access(current_cpu
, addr
, true, false, 0, size
);
1124 static bool unassigned_mem_accepts(void *opaque
, hwaddr addr
,
1125 unsigned size
, bool is_write
)
1130 const MemoryRegionOps unassigned_mem_ops
= {
1131 .valid
.accepts
= unassigned_mem_accepts
,
1132 .endianness
= DEVICE_NATIVE_ENDIAN
,
1135 bool memory_region_access_valid(MemoryRegion
*mr
,
1140 int access_size_min
, access_size_max
;
1143 if (!mr
->ops
->valid
.unaligned
&& (addr
& (size
- 1))) {
1147 if (!mr
->ops
->valid
.accepts
) {
1151 access_size_min
= mr
->ops
->valid
.min_access_size
;
1152 if (!mr
->ops
->valid
.min_access_size
) {
1153 access_size_min
= 1;
1156 access_size_max
= mr
->ops
->valid
.max_access_size
;
1157 if (!mr
->ops
->valid
.max_access_size
) {
1158 access_size_max
= 4;
1161 access_size
= MAX(MIN(size
, access_size_max
), access_size_min
);
1162 for (i
= 0; i
< size
; i
+= access_size
) {
1163 if (!mr
->ops
->valid
.accepts(mr
->opaque
, addr
+ i
, access_size
,
1172 static MemTxResult
memory_region_dispatch_read1(MemoryRegion
*mr
,
1180 if (mr
->ops
->read
) {
1181 return access_with_adjusted_size(addr
, pval
, size
,
1182 mr
->ops
->impl
.min_access_size
,
1183 mr
->ops
->impl
.max_access_size
,
1184 memory_region_read_accessor
,
1186 } else if (mr
->ops
->read_with_attrs
) {
1187 return access_with_adjusted_size(addr
, pval
, size
,
1188 mr
->ops
->impl
.min_access_size
,
1189 mr
->ops
->impl
.max_access_size
,
1190 memory_region_read_with_attrs_accessor
,
1193 return access_with_adjusted_size(addr
, pval
, size
, 1, 4,
1194 memory_region_oldmmio_read_accessor
,
1199 MemTxResult
memory_region_dispatch_read(MemoryRegion
*mr
,
1207 if (!memory_region_access_valid(mr
, addr
, size
, false)) {
1208 *pval
= unassigned_mem_read(mr
, addr
, size
);
1209 return MEMTX_DECODE_ERROR
;
1212 r
= memory_region_dispatch_read1(mr
, addr
, pval
, size
, attrs
);
1213 adjust_endianness(mr
, pval
, size
);
1217 /* Return true if an eventfd was signalled */
1218 static bool memory_region_dispatch_write_eventfds(MemoryRegion
*mr
,
1224 MemoryRegionIoeventfd ioeventfd
= {
1225 .addr
= addrrange_make(int128_make64(addr
), int128_make64(size
)),
1230 for (i
= 0; i
< mr
->ioeventfd_nb
; i
++) {
1231 ioeventfd
.match_data
= mr
->ioeventfds
[i
].match_data
;
1232 ioeventfd
.e
= mr
->ioeventfds
[i
].e
;
1234 if (memory_region_ioeventfd_equal(ioeventfd
, mr
->ioeventfds
[i
])) {
1235 event_notifier_set(ioeventfd
.e
);
1243 MemTxResult
memory_region_dispatch_write(MemoryRegion
*mr
,
1249 if (!memory_region_access_valid(mr
, addr
, size
, true)) {
1250 unassigned_mem_write(mr
, addr
, data
, size
);
1251 return MEMTX_DECODE_ERROR
;
1254 adjust_endianness(mr
, &data
, size
);
1256 if ((!kvm_eventfds_enabled()) &&
1257 memory_region_dispatch_write_eventfds(mr
, addr
, data
, size
, attrs
)) {
1261 if (mr
->ops
->write
) {
1262 return access_with_adjusted_size(addr
, &data
, size
,
1263 mr
->ops
->impl
.min_access_size
,
1264 mr
->ops
->impl
.max_access_size
,
1265 memory_region_write_accessor
, mr
,
1267 } else if (mr
->ops
->write_with_attrs
) {
1269 access_with_adjusted_size(addr
, &data
, size
,
1270 mr
->ops
->impl
.min_access_size
,
1271 mr
->ops
->impl
.max_access_size
,
1272 memory_region_write_with_attrs_accessor
,
1275 return access_with_adjusted_size(addr
, &data
, size
, 1, 4,
1276 memory_region_oldmmio_write_accessor
,
1281 void memory_region_init_io(MemoryRegion
*mr
,
1283 const MemoryRegionOps
*ops
,
1288 memory_region_init(mr
, owner
, name
, size
);
1289 mr
->ops
= ops
? ops
: &unassigned_mem_ops
;
1290 mr
->opaque
= opaque
;
1291 mr
->terminates
= true;
1294 void memory_region_init_ram(MemoryRegion
*mr
,
1300 memory_region_init(mr
, owner
, name
, size
);
1302 mr
->terminates
= true;
1303 mr
->destructor
= memory_region_destructor_ram
;
1304 mr
->ram_block
= qemu_ram_alloc(size
, mr
, errp
);
1305 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1308 void memory_region_init_resizeable_ram(MemoryRegion
*mr
,
1313 void (*resized
)(const char*,
1318 memory_region_init(mr
, owner
, name
, size
);
1320 mr
->terminates
= true;
1321 mr
->destructor
= memory_region_destructor_ram
;
1322 mr
->ram_block
= qemu_ram_alloc_resizeable(size
, max_size
, resized
,
1324 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1328 void memory_region_init_ram_from_file(MemoryRegion
*mr
,
1329 struct Object
*owner
,
1336 memory_region_init(mr
, owner
, name
, size
);
1338 mr
->terminates
= true;
1339 mr
->destructor
= memory_region_destructor_ram
;
1340 mr
->ram_block
= qemu_ram_alloc_from_file(size
, mr
, share
, path
, errp
);
1341 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1345 void memory_region_init_ram_ptr(MemoryRegion
*mr
,
1351 memory_region_init(mr
, owner
, name
, size
);
1353 mr
->terminates
= true;
1354 mr
->destructor
= memory_region_destructor_ram
;
1355 mr
->dirty_log_mask
= tcg_enabled() ? (1 << DIRTY_MEMORY_CODE
) : 0;
1357 /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL. */
1358 assert(ptr
!= NULL
);
1359 mr
->ram_block
= qemu_ram_alloc_from_ptr(size
, ptr
, mr
, &error_fatal
);
1362 void memory_region_set_skip_dump(MemoryRegion
*mr
)
1364 mr
->skip_dump
= true;
1367 void memory_region_init_alias(MemoryRegion
*mr
,
1374 memory_region_init(mr
, owner
, name
, size
);
1376 mr
->alias_offset
= offset
;
1379 void memory_region_init_rom_device(MemoryRegion
*mr
,
1381 const MemoryRegionOps
*ops
,
1387 memory_region_init(mr
, owner
, name
, size
);
1389 mr
->opaque
= opaque
;
1390 mr
->terminates
= true;
1391 mr
->rom_device
= true;
1392 mr
->destructor
= memory_region_destructor_rom_device
;
1393 mr
->ram_block
= qemu_ram_alloc(size
, mr
, errp
);
1396 void memory_region_init_iommu(MemoryRegion
*mr
,
1398 const MemoryRegionIOMMUOps
*ops
,
1402 memory_region_init(mr
, owner
, name
, size
);
1403 mr
->iommu_ops
= ops
,
1404 mr
->terminates
= true; /* then re-forwards */
1405 notifier_list_init(&mr
->iommu_notify
);
1408 static void memory_region_finalize(Object
*obj
)
1410 MemoryRegion
*mr
= MEMORY_REGION(obj
);
1412 assert(!mr
->container
);
1414 /* We know the region is not visible in any address space (it
1415 * does not have a container and cannot be a root either because
1416 * it has no references, so we can blindly clear mr->enabled.
1417 * memory_region_set_enabled instead could trigger a transaction
1418 * and cause an infinite loop.
1420 mr
->enabled
= false;
1421 memory_region_transaction_begin();
1422 while (!QTAILQ_EMPTY(&mr
->subregions
)) {
1423 MemoryRegion
*subregion
= QTAILQ_FIRST(&mr
->subregions
);
1424 memory_region_del_subregion(mr
, subregion
);
1426 memory_region_transaction_commit();
1429 memory_region_clear_coalescing(mr
);
1430 g_free((char *)mr
->name
);
1431 g_free(mr
->ioeventfds
);
1434 Object
*memory_region_owner(MemoryRegion
*mr
)
1436 Object
*obj
= OBJECT(mr
);
1440 void memory_region_ref(MemoryRegion
*mr
)
1442 /* MMIO callbacks most likely will access data that belongs
1443 * to the owner, hence the need to ref/unref the owner whenever
1444 * the memory region is in use.
1446 * The memory region is a child of its owner. As long as the
1447 * owner doesn't call unparent itself on the memory region,
1448 * ref-ing the owner will also keep the memory region alive.
1449 * Memory regions without an owner are supposed to never go away;
1450 * we do not ref/unref them because it slows down DMA sensibly.
1452 if (mr
&& mr
->owner
) {
1453 object_ref(mr
->owner
);
1457 void memory_region_unref(MemoryRegion
*mr
)
1459 if (mr
&& mr
->owner
) {
1460 object_unref(mr
->owner
);
1464 uint64_t memory_region_size(MemoryRegion
*mr
)
1466 if (int128_eq(mr
->size
, int128_2_64())) {
1469 return int128_get64(mr
->size
);
1472 const char *memory_region_name(const MemoryRegion
*mr
)
1475 ((MemoryRegion
*)mr
)->name
=
1476 object_get_canonical_path_component(OBJECT(mr
));
1481 bool memory_region_is_skip_dump(MemoryRegion
*mr
)
1483 return mr
->skip_dump
;
1486 uint8_t memory_region_get_dirty_log_mask(MemoryRegion
*mr
)
1488 uint8_t mask
= mr
->dirty_log_mask
;
1489 if (global_dirty_log
) {
1490 mask
|= (1 << DIRTY_MEMORY_MIGRATION
);
1495 bool memory_region_is_logging(MemoryRegion
*mr
, uint8_t client
)
1497 return memory_region_get_dirty_log_mask(mr
) & (1 << client
);
1500 void memory_region_register_iommu_notifier(MemoryRegion
*mr
, Notifier
*n
)
1502 if (mr
->iommu_ops
->notify_started
&&
1503 QLIST_EMPTY(&mr
->iommu_notify
.notifiers
)) {
1504 mr
->iommu_ops
->notify_started(mr
);
1506 notifier_list_add(&mr
->iommu_notify
, n
);
1509 uint64_t memory_region_iommu_get_min_page_size(MemoryRegion
*mr
)
1511 assert(memory_region_is_iommu(mr
));
1512 if (mr
->iommu_ops
&& mr
->iommu_ops
->get_min_page_size
) {
1513 return mr
->iommu_ops
->get_min_page_size(mr
);
1515 return TARGET_PAGE_SIZE
;
1518 void memory_region_iommu_replay(MemoryRegion
*mr
, Notifier
*n
, bool is_write
)
1520 hwaddr addr
, granularity
;
1521 IOMMUTLBEntry iotlb
;
1523 granularity
= memory_region_iommu_get_min_page_size(mr
);
1525 for (addr
= 0; addr
< memory_region_size(mr
); addr
+= granularity
) {
1526 iotlb
= mr
->iommu_ops
->translate(mr
, addr
, is_write
);
1527 if (iotlb
.perm
!= IOMMU_NONE
) {
1528 n
->notify(n
, &iotlb
);
1531 /* if (2^64 - MR size) < granularity, it's possible to get an
1532 * infinite loop here. This should catch such a wraparound */
1533 if ((addr
+ granularity
) < addr
) {
1539 void memory_region_unregister_iommu_notifier(MemoryRegion
*mr
, Notifier
*n
)
1542 if (mr
->iommu_ops
->notify_stopped
&&
1543 QLIST_EMPTY(&mr
->iommu_notify
.notifiers
)) {
1544 mr
->iommu_ops
->notify_stopped(mr
);
1548 void memory_region_notify_iommu(MemoryRegion
*mr
,
1549 IOMMUTLBEntry entry
)
1551 assert(memory_region_is_iommu(mr
));
1552 notifier_list_notify(&mr
->iommu_notify
, &entry
);
1555 void memory_region_set_log(MemoryRegion
*mr
, bool log
, unsigned client
)
1557 uint8_t mask
= 1 << client
;
1558 uint8_t old_logging
;
1560 assert(client
== DIRTY_MEMORY_VGA
);
1561 old_logging
= mr
->vga_logging_count
;
1562 mr
->vga_logging_count
+= log
? 1 : -1;
1563 if (!!old_logging
== !!mr
->vga_logging_count
) {
1567 memory_region_transaction_begin();
1568 mr
->dirty_log_mask
= (mr
->dirty_log_mask
& ~mask
) | (log
* mask
);
1569 memory_region_update_pending
|= mr
->enabled
;
1570 memory_region_transaction_commit();
1573 bool memory_region_get_dirty(MemoryRegion
*mr
, hwaddr addr
,
1574 hwaddr size
, unsigned client
)
1576 assert(mr
->ram_block
);
1577 return cpu_physical_memory_get_dirty(memory_region_get_ram_addr(mr
) + addr
,
1581 void memory_region_set_dirty(MemoryRegion
*mr
, hwaddr addr
,
1584 assert(mr
->ram_block
);
1585 cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr
) + addr
,
1587 memory_region_get_dirty_log_mask(mr
));
1590 bool memory_region_test_and_clear_dirty(MemoryRegion
*mr
, hwaddr addr
,
1591 hwaddr size
, unsigned client
)
1593 assert(mr
->ram_block
);
1594 return cpu_physical_memory_test_and_clear_dirty(
1595 memory_region_get_ram_addr(mr
) + addr
, size
, client
);
1599 void memory_region_sync_dirty_bitmap(MemoryRegion
*mr
)
1604 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
1605 FlatView
*view
= address_space_get_flatview(as
);
1606 FOR_EACH_FLAT_RANGE(fr
, view
) {
1608 MEMORY_LISTENER_UPDATE_REGION(fr
, as
, Forward
, log_sync
);
1611 flatview_unref(view
);
1615 void memory_region_set_readonly(MemoryRegion
*mr
, bool readonly
)
1617 if (mr
->readonly
!= readonly
) {
1618 memory_region_transaction_begin();
1619 mr
->readonly
= readonly
;
1620 memory_region_update_pending
|= mr
->enabled
;
1621 memory_region_transaction_commit();
1625 void memory_region_rom_device_set_romd(MemoryRegion
*mr
, bool romd_mode
)
1627 if (mr
->romd_mode
!= romd_mode
) {
1628 memory_region_transaction_begin();
1629 mr
->romd_mode
= romd_mode
;
1630 memory_region_update_pending
|= mr
->enabled
;
1631 memory_region_transaction_commit();
1635 void memory_region_reset_dirty(MemoryRegion
*mr
, hwaddr addr
,
1636 hwaddr size
, unsigned client
)
1638 assert(mr
->ram_block
);
1639 cpu_physical_memory_test_and_clear_dirty(
1640 memory_region_get_ram_addr(mr
) + addr
, size
, client
);
1643 int memory_region_get_fd(MemoryRegion
*mr
)
1651 fd
= mr
->ram_block
->fd
;
1657 void memory_region_set_fd(MemoryRegion
*mr
, int fd
)
1663 mr
->ram_block
->fd
= fd
;
1667 void *memory_region_get_ram_ptr(MemoryRegion
*mr
)
1670 uint64_t offset
= 0;
1674 offset
+= mr
->alias_offset
;
1677 assert(mr
->ram_block
);
1678 ptr
= qemu_map_ram_ptr(mr
->ram_block
, offset
);
1684 MemoryRegion
*memory_region_from_host(void *ptr
, ram_addr_t
*offset
)
1688 block
= qemu_ram_block_from_host(ptr
, false, offset
);
1696 ram_addr_t
memory_region_get_ram_addr(MemoryRegion
*mr
)
1698 return mr
->ram_block
? mr
->ram_block
->offset
: RAM_ADDR_INVALID
;
1701 void memory_region_ram_resize(MemoryRegion
*mr
, ram_addr_t newsize
, Error
**errp
)
1703 assert(mr
->ram_block
);
1705 qemu_ram_resize(mr
->ram_block
, newsize
, errp
);
1708 static void memory_region_update_coalesced_range_as(MemoryRegion
*mr
, AddressSpace
*as
)
1712 CoalescedMemoryRange
*cmr
;
1714 MemoryRegionSection section
;
1716 view
= address_space_get_flatview(as
);
1717 FOR_EACH_FLAT_RANGE(fr
, view
) {
1719 section
= (MemoryRegionSection
) {
1720 .address_space
= as
,
1721 .offset_within_address_space
= int128_get64(fr
->addr
.start
),
1722 .size
= fr
->addr
.size
,
1725 MEMORY_LISTENER_CALL(coalesced_mmio_del
, Reverse
, §ion
,
1726 int128_get64(fr
->addr
.start
),
1727 int128_get64(fr
->addr
.size
));
1728 QTAILQ_FOREACH(cmr
, &mr
->coalesced
, link
) {
1729 tmp
= addrrange_shift(cmr
->addr
,
1730 int128_sub(fr
->addr
.start
,
1731 int128_make64(fr
->offset_in_region
)));
1732 if (!addrrange_intersects(tmp
, fr
->addr
)) {
1735 tmp
= addrrange_intersection(tmp
, fr
->addr
);
1736 MEMORY_LISTENER_CALL(coalesced_mmio_add
, Forward
, §ion
,
1737 int128_get64(tmp
.start
),
1738 int128_get64(tmp
.size
));
1742 flatview_unref(view
);
1745 static void memory_region_update_coalesced_range(MemoryRegion
*mr
)
1749 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
1750 memory_region_update_coalesced_range_as(mr
, as
);
1754 void memory_region_set_coalescing(MemoryRegion
*mr
)
1756 memory_region_clear_coalescing(mr
);
1757 memory_region_add_coalescing(mr
, 0, int128_get64(mr
->size
));
1760 void memory_region_add_coalescing(MemoryRegion
*mr
,
1764 CoalescedMemoryRange
*cmr
= g_malloc(sizeof(*cmr
));
1766 cmr
->addr
= addrrange_make(int128_make64(offset
), int128_make64(size
));
1767 QTAILQ_INSERT_TAIL(&mr
->coalesced
, cmr
, link
);
1768 memory_region_update_coalesced_range(mr
);
1769 memory_region_set_flush_coalesced(mr
);
1772 void memory_region_clear_coalescing(MemoryRegion
*mr
)
1774 CoalescedMemoryRange
*cmr
;
1775 bool updated
= false;
1777 qemu_flush_coalesced_mmio_buffer();
1778 mr
->flush_coalesced_mmio
= false;
1780 while (!QTAILQ_EMPTY(&mr
->coalesced
)) {
1781 cmr
= QTAILQ_FIRST(&mr
->coalesced
);
1782 QTAILQ_REMOVE(&mr
->coalesced
, cmr
, link
);
1788 memory_region_update_coalesced_range(mr
);
1792 void memory_region_set_flush_coalesced(MemoryRegion
*mr
)
1794 mr
->flush_coalesced_mmio
= true;
1797 void memory_region_clear_flush_coalesced(MemoryRegion
*mr
)
1799 qemu_flush_coalesced_mmio_buffer();
1800 if (QTAILQ_EMPTY(&mr
->coalesced
)) {
1801 mr
->flush_coalesced_mmio
= false;
1805 void memory_region_set_global_locking(MemoryRegion
*mr
)
1807 mr
->global_locking
= true;
1810 void memory_region_clear_global_locking(MemoryRegion
*mr
)
1812 mr
->global_locking
= false;
1815 static bool userspace_eventfd_warning
;
1817 void memory_region_add_eventfd(MemoryRegion
*mr
,
1824 MemoryRegionIoeventfd mrfd
= {
1825 .addr
.start
= int128_make64(addr
),
1826 .addr
.size
= int128_make64(size
),
1827 .match_data
= match_data
,
1833 if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
1834 userspace_eventfd_warning
))) {
1835 userspace_eventfd_warning
= true;
1836 error_report("Using eventfd without MMIO binding in KVM. "
1837 "Suboptimal performance expected");
1841 adjust_endianness(mr
, &mrfd
.data
, size
);
1843 memory_region_transaction_begin();
1844 for (i
= 0; i
< mr
->ioeventfd_nb
; ++i
) {
1845 if (memory_region_ioeventfd_before(mrfd
, mr
->ioeventfds
[i
])) {
1850 mr
->ioeventfds
= g_realloc(mr
->ioeventfds
,
1851 sizeof(*mr
->ioeventfds
) * mr
->ioeventfd_nb
);
1852 memmove(&mr
->ioeventfds
[i
+1], &mr
->ioeventfds
[i
],
1853 sizeof(*mr
->ioeventfds
) * (mr
->ioeventfd_nb
-1 - i
));
1854 mr
->ioeventfds
[i
] = mrfd
;
1855 ioeventfd_update_pending
|= mr
->enabled
;
1856 memory_region_transaction_commit();
1859 void memory_region_del_eventfd(MemoryRegion
*mr
,
1866 MemoryRegionIoeventfd mrfd
= {
1867 .addr
.start
= int128_make64(addr
),
1868 .addr
.size
= int128_make64(size
),
1869 .match_data
= match_data
,
1876 adjust_endianness(mr
, &mrfd
.data
, size
);
1878 memory_region_transaction_begin();
1879 for (i
= 0; i
< mr
->ioeventfd_nb
; ++i
) {
1880 if (memory_region_ioeventfd_equal(mrfd
, mr
->ioeventfds
[i
])) {
1884 assert(i
!= mr
->ioeventfd_nb
);
1885 memmove(&mr
->ioeventfds
[i
], &mr
->ioeventfds
[i
+1],
1886 sizeof(*mr
->ioeventfds
) * (mr
->ioeventfd_nb
- (i
+1)));
1888 mr
->ioeventfds
= g_realloc(mr
->ioeventfds
,
1889 sizeof(*mr
->ioeventfds
)*mr
->ioeventfd_nb
+ 1);
1890 ioeventfd_update_pending
|= mr
->enabled
;
1891 memory_region_transaction_commit();
1894 static void memory_region_update_container_subregions(MemoryRegion
*subregion
)
1896 MemoryRegion
*mr
= subregion
->container
;
1897 MemoryRegion
*other
;
1899 memory_region_transaction_begin();
1901 memory_region_ref(subregion
);
1902 QTAILQ_FOREACH(other
, &mr
->subregions
, subregions_link
) {
1903 if (subregion
->priority
>= other
->priority
) {
1904 QTAILQ_INSERT_BEFORE(other
, subregion
, subregions_link
);
1908 QTAILQ_INSERT_TAIL(&mr
->subregions
, subregion
, subregions_link
);
1910 memory_region_update_pending
|= mr
->enabled
&& subregion
->enabled
;
1911 memory_region_transaction_commit();
1914 static void memory_region_add_subregion_common(MemoryRegion
*mr
,
1916 MemoryRegion
*subregion
)
1918 assert(!subregion
->container
);
1919 subregion
->container
= mr
;
1920 subregion
->addr
= offset
;
1921 memory_region_update_container_subregions(subregion
);
1924 void memory_region_add_subregion(MemoryRegion
*mr
,
1926 MemoryRegion
*subregion
)
1928 subregion
->priority
= 0;
1929 memory_region_add_subregion_common(mr
, offset
, subregion
);
1932 void memory_region_add_subregion_overlap(MemoryRegion
*mr
,
1934 MemoryRegion
*subregion
,
1937 subregion
->priority
= priority
;
1938 memory_region_add_subregion_common(mr
, offset
, subregion
);
1941 void memory_region_del_subregion(MemoryRegion
*mr
,
1942 MemoryRegion
*subregion
)
1944 memory_region_transaction_begin();
1945 assert(subregion
->container
== mr
);
1946 subregion
->container
= NULL
;
1947 QTAILQ_REMOVE(&mr
->subregions
, subregion
, subregions_link
);
1948 memory_region_unref(subregion
);
1949 memory_region_update_pending
|= mr
->enabled
&& subregion
->enabled
;
1950 memory_region_transaction_commit();
1953 void memory_region_set_enabled(MemoryRegion
*mr
, bool enabled
)
1955 if (enabled
== mr
->enabled
) {
1958 memory_region_transaction_begin();
1959 mr
->enabled
= enabled
;
1960 memory_region_update_pending
= true;
1961 memory_region_transaction_commit();
1964 void memory_region_set_size(MemoryRegion
*mr
, uint64_t size
)
1966 Int128 s
= int128_make64(size
);
1968 if (size
== UINT64_MAX
) {
1971 if (int128_eq(s
, mr
->size
)) {
1974 memory_region_transaction_begin();
1976 memory_region_update_pending
= true;
1977 memory_region_transaction_commit();
1980 static void memory_region_readd_subregion(MemoryRegion
*mr
)
1982 MemoryRegion
*container
= mr
->container
;
1985 memory_region_transaction_begin();
1986 memory_region_ref(mr
);
1987 memory_region_del_subregion(container
, mr
);
1988 mr
->container
= container
;
1989 memory_region_update_container_subregions(mr
);
1990 memory_region_unref(mr
);
1991 memory_region_transaction_commit();
1995 void memory_region_set_address(MemoryRegion
*mr
, hwaddr addr
)
1997 if (addr
!= mr
->addr
) {
1999 memory_region_readd_subregion(mr
);
2003 void memory_region_set_alias_offset(MemoryRegion
*mr
, hwaddr offset
)
2007 if (offset
== mr
->alias_offset
) {
2011 memory_region_transaction_begin();
2012 mr
->alias_offset
= offset
;
2013 memory_region_update_pending
|= mr
->enabled
;
2014 memory_region_transaction_commit();
2017 uint64_t memory_region_get_alignment(const MemoryRegion
*mr
)
2022 static int cmp_flatrange_addr(const void *addr_
, const void *fr_
)
2024 const AddrRange
*addr
= addr_
;
2025 const FlatRange
*fr
= fr_
;
2027 if (int128_le(addrrange_end(*addr
), fr
->addr
.start
)) {
2029 } else if (int128_ge(addr
->start
, addrrange_end(fr
->addr
))) {
2035 static FlatRange
*flatview_lookup(FlatView
*view
, AddrRange addr
)
2037 return bsearch(&addr
, view
->ranges
, view
->nr
,
2038 sizeof(FlatRange
), cmp_flatrange_addr
);
2041 bool memory_region_is_mapped(MemoryRegion
*mr
)
2043 return mr
->container
? true : false;
2046 /* Same as memory_region_find, but it does not add a reference to the
2047 * returned region. It must be called from an RCU critical section.
2049 static MemoryRegionSection
memory_region_find_rcu(MemoryRegion
*mr
,
2050 hwaddr addr
, uint64_t size
)
2052 MemoryRegionSection ret
= { .mr
= NULL
};
2060 for (root
= mr
; root
->container
; ) {
2061 root
= root
->container
;
2065 as
= memory_region_to_address_space(root
);
2069 range
= addrrange_make(int128_make64(addr
), int128_make64(size
));
2071 view
= atomic_rcu_read(&as
->current_map
);
2072 fr
= flatview_lookup(view
, range
);
2077 while (fr
> view
->ranges
&& addrrange_intersects(fr
[-1].addr
, range
)) {
2082 ret
.address_space
= as
;
2083 range
= addrrange_intersection(range
, fr
->addr
);
2084 ret
.offset_within_region
= fr
->offset_in_region
;
2085 ret
.offset_within_region
+= int128_get64(int128_sub(range
.start
,
2087 ret
.size
= range
.size
;
2088 ret
.offset_within_address_space
= int128_get64(range
.start
);
2089 ret
.readonly
= fr
->readonly
;
2093 MemoryRegionSection
memory_region_find(MemoryRegion
*mr
,
2094 hwaddr addr
, uint64_t size
)
2096 MemoryRegionSection ret
;
2098 ret
= memory_region_find_rcu(mr
, addr
, size
);
2100 memory_region_ref(ret
.mr
);
2106 bool memory_region_present(MemoryRegion
*container
, hwaddr addr
)
2111 mr
= memory_region_find_rcu(container
, addr
, 1).mr
;
2113 return mr
&& mr
!= container
;
2116 void address_space_sync_dirty_bitmap(AddressSpace
*as
)
2121 view
= address_space_get_flatview(as
);
2122 FOR_EACH_FLAT_RANGE(fr
, view
) {
2123 MEMORY_LISTENER_UPDATE_REGION(fr
, as
, Forward
, log_sync
);
2125 flatview_unref(view
);
2128 void memory_global_dirty_log_start(void)
2130 global_dirty_log
= true;
2132 MEMORY_LISTENER_CALL_GLOBAL(log_global_start
, Forward
);
2134 /* Refresh DIRTY_LOG_MIGRATION bit. */
2135 memory_region_transaction_begin();
2136 memory_region_update_pending
= true;
2137 memory_region_transaction_commit();
2140 void memory_global_dirty_log_stop(void)
2142 global_dirty_log
= false;
2144 /* Refresh DIRTY_LOG_MIGRATION bit. */
2145 memory_region_transaction_begin();
2146 memory_region_update_pending
= true;
2147 memory_region_transaction_commit();
2149 MEMORY_LISTENER_CALL_GLOBAL(log_global_stop
, Reverse
);
2152 static void listener_add_address_space(MemoryListener
*listener
,
2158 if (listener
->address_space_filter
2159 && listener
->address_space_filter
!= as
) {
2163 if (listener
->begin
) {
2164 listener
->begin(listener
);
2166 if (global_dirty_log
) {
2167 if (listener
->log_global_start
) {
2168 listener
->log_global_start(listener
);
2172 view
= address_space_get_flatview(as
);
2173 FOR_EACH_FLAT_RANGE(fr
, view
) {
2174 MemoryRegionSection section
= {
2176 .address_space
= as
,
2177 .offset_within_region
= fr
->offset_in_region
,
2178 .size
= fr
->addr
.size
,
2179 .offset_within_address_space
= int128_get64(fr
->addr
.start
),
2180 .readonly
= fr
->readonly
,
2182 if (fr
->dirty_log_mask
&& listener
->log_start
) {
2183 listener
->log_start(listener
, §ion
, 0, fr
->dirty_log_mask
);
2185 if (listener
->region_add
) {
2186 listener
->region_add(listener
, §ion
);
2189 if (listener
->commit
) {
2190 listener
->commit(listener
);
2192 flatview_unref(view
);
2195 void memory_listener_register(MemoryListener
*listener
, AddressSpace
*filter
)
2197 MemoryListener
*other
= NULL
;
2200 listener
->address_space_filter
= filter
;
2201 if (QTAILQ_EMPTY(&memory_listeners
)
2202 || listener
->priority
>= QTAILQ_LAST(&memory_listeners
,
2203 memory_listeners
)->priority
) {
2204 QTAILQ_INSERT_TAIL(&memory_listeners
, listener
, link
);
2206 QTAILQ_FOREACH(other
, &memory_listeners
, link
) {
2207 if (listener
->priority
< other
->priority
) {
2211 QTAILQ_INSERT_BEFORE(other
, listener
, link
);
2214 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
2215 listener_add_address_space(listener
, as
);
2219 void memory_listener_unregister(MemoryListener
*listener
)
2221 QTAILQ_REMOVE(&memory_listeners
, listener
, link
);
2224 void address_space_init(AddressSpace
*as
, MemoryRegion
*root
, const char *name
)
2226 memory_region_ref(root
);
2227 memory_region_transaction_begin();
2230 as
->malloced
= false;
2231 as
->current_map
= g_new(FlatView
, 1);
2232 flatview_init(as
->current_map
);
2233 as
->ioeventfd_nb
= 0;
2234 as
->ioeventfds
= NULL
;
2235 QTAILQ_INSERT_TAIL(&address_spaces
, as
, address_spaces_link
);
2236 as
->name
= g_strdup(name
? name
: "anonymous");
2237 address_space_init_dispatch(as
);
2238 memory_region_update_pending
|= root
->enabled
;
2239 memory_region_transaction_commit();
2242 static void do_address_space_destroy(AddressSpace
*as
)
2244 MemoryListener
*listener
;
2245 bool do_free
= as
->malloced
;
2247 address_space_destroy_dispatch(as
);
2249 QTAILQ_FOREACH(listener
, &memory_listeners
, link
) {
2250 assert(listener
->address_space_filter
!= as
);
2253 flatview_unref(as
->current_map
);
2255 g_free(as
->ioeventfds
);
2256 memory_region_unref(as
->root
);
2262 AddressSpace
*address_space_init_shareable(MemoryRegion
*root
, const char *name
)
2266 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
2267 if (root
== as
->root
&& as
->malloced
) {
2273 as
= g_malloc0(sizeof *as
);
2274 address_space_init(as
, root
, name
);
2275 as
->malloced
= true;
2279 void address_space_destroy(AddressSpace
*as
)
2281 MemoryRegion
*root
= as
->root
;
2284 if (as
->ref_count
) {
2287 /* Flush out anything from MemoryListeners listening in on this */
2288 memory_region_transaction_begin();
2290 memory_region_transaction_commit();
2291 QTAILQ_REMOVE(&address_spaces
, as
, address_spaces_link
);
2292 address_space_unregister(as
);
2294 /* At this point, as->dispatch and as->current_map are dummy
2295 * entries that the guest should never use. Wait for the old
2296 * values to expire before freeing the data.
2299 call_rcu(as
, do_address_space_destroy
, rcu
);
2302 typedef struct MemoryRegionList MemoryRegionList
;
2304 struct MemoryRegionList
{
2305 const MemoryRegion
*mr
;
2306 QTAILQ_ENTRY(MemoryRegionList
) queue
;
2309 typedef QTAILQ_HEAD(queue
, MemoryRegionList
) MemoryRegionListHead
;
2311 static void mtree_print_mr(fprintf_function mon_printf
, void *f
,
2312 const MemoryRegion
*mr
, unsigned int level
,
2314 MemoryRegionListHead
*alias_print_queue
)
2316 MemoryRegionList
*new_ml
, *ml
, *next_ml
;
2317 MemoryRegionListHead submr_print_queue
;
2318 const MemoryRegion
*submr
;
2325 for (i
= 0; i
< level
; i
++) {
2330 MemoryRegionList
*ml
;
2333 /* check if the alias is already in the queue */
2334 QTAILQ_FOREACH(ml
, alias_print_queue
, queue
) {
2335 if (ml
->mr
== mr
->alias
) {
2341 ml
= g_new(MemoryRegionList
, 1);
2343 QTAILQ_INSERT_TAIL(alias_print_queue
, ml
, queue
);
2345 mon_printf(f
, TARGET_FMT_plx
"-" TARGET_FMT_plx
2346 " (prio %d, %c%c): alias %s @%s " TARGET_FMT_plx
2347 "-" TARGET_FMT_plx
"%s\n",
2350 + (int128_nz(mr
->size
) ?
2351 (hwaddr
)int128_get64(int128_sub(mr
->size
,
2352 int128_one())) : 0),
2354 mr
->romd_mode
? 'R' : '-',
2355 !mr
->readonly
&& !(mr
->rom_device
&& mr
->romd_mode
) ? 'W'
2357 memory_region_name(mr
),
2358 memory_region_name(mr
->alias
),
2361 + (int128_nz(mr
->size
) ?
2362 (hwaddr
)int128_get64(int128_sub(mr
->size
,
2363 int128_one())) : 0),
2364 mr
->enabled
? "" : " [disabled]");
2367 TARGET_FMT_plx
"-" TARGET_FMT_plx
" (prio %d, %c%c): %s%s\n",
2370 + (int128_nz(mr
->size
) ?
2371 (hwaddr
)int128_get64(int128_sub(mr
->size
,
2372 int128_one())) : 0),
2374 mr
->romd_mode
? 'R' : '-',
2375 !mr
->readonly
&& !(mr
->rom_device
&& mr
->romd_mode
) ? 'W'
2377 memory_region_name(mr
),
2378 mr
->enabled
? "" : " [disabled]");
2381 QTAILQ_INIT(&submr_print_queue
);
2383 QTAILQ_FOREACH(submr
, &mr
->subregions
, subregions_link
) {
2384 new_ml
= g_new(MemoryRegionList
, 1);
2386 QTAILQ_FOREACH(ml
, &submr_print_queue
, queue
) {
2387 if (new_ml
->mr
->addr
< ml
->mr
->addr
||
2388 (new_ml
->mr
->addr
== ml
->mr
->addr
&&
2389 new_ml
->mr
->priority
> ml
->mr
->priority
)) {
2390 QTAILQ_INSERT_BEFORE(ml
, new_ml
, queue
);
2396 QTAILQ_INSERT_TAIL(&submr_print_queue
, new_ml
, queue
);
2400 QTAILQ_FOREACH(ml
, &submr_print_queue
, queue
) {
2401 mtree_print_mr(mon_printf
, f
, ml
->mr
, level
+ 1, base
+ mr
->addr
,
2405 QTAILQ_FOREACH_SAFE(ml
, &submr_print_queue
, queue
, next_ml
) {
2410 void mtree_info(fprintf_function mon_printf
, void *f
)
2412 MemoryRegionListHead ml_head
;
2413 MemoryRegionList
*ml
, *ml2
;
2416 QTAILQ_INIT(&ml_head
);
2418 QTAILQ_FOREACH(as
, &address_spaces
, address_spaces_link
) {
2419 mon_printf(f
, "address-space: %s\n", as
->name
);
2420 mtree_print_mr(mon_printf
, f
, as
->root
, 1, 0, &ml_head
);
2421 mon_printf(f
, "\n");
2424 /* print aliased regions */
2425 QTAILQ_FOREACH(ml
, &ml_head
, queue
) {
2426 mon_printf(f
, "memory-region: %s\n", memory_region_name(ml
->mr
));
2427 mtree_print_mr(mon_printf
, f
, ml
->mr
, 1, 0, &ml_head
);
2428 mon_printf(f
, "\n");
2431 QTAILQ_FOREACH_SAFE(ml
, &ml_head
, queue
, ml2
) {
2436 static const TypeInfo memory_region_info
= {
2437 .parent
= TYPE_OBJECT
,
2438 .name
= TYPE_MEMORY_REGION
,
2439 .instance_size
= sizeof(MemoryRegion
),
2440 .instance_init
= memory_region_initfn
,
2441 .instance_finalize
= memory_region_finalize
,
2444 static void memory_register_types(void)
2446 type_register_static(&memory_region_info
);
2449 type_init(memory_register_types
)