2 * Declarations for cpu physical memory functions
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 or
10 * later. See the COPYING file in the top-level directory.
15 * This header is for use by exec.c and memory.c ONLY. Do not include it.
16 * The functions declared here will be removed soon.
22 #ifndef CONFIG_USER_ONLY
24 #include "hw/xen/xen.h"
25 #include "sysemu/tcg.h"
26 #include "exec/ramlist.h"
30 struct MemoryRegion
*mr
;
32 uint8_t *colo_cache
; /* For colo, VM's ram cache */
34 ram_addr_t used_length
;
35 ram_addr_t max_length
;
36 void (*resized
)(const char*, uint64_t length
, void *host
);
38 /* Protected by iothread lock. */
40 /* RCU-enabled, writes protected by the ramlist lock */
41 QLIST_ENTRY(RAMBlock
) next
;
42 QLIST_HEAD(, RAMBlockNotifier
) ramblock_notifiers
;
45 /* dirty bitmap used during migration */
47 /* bitmap of already received pages in postcopy */
48 unsigned long *receivedmap
;
51 * bitmap to track already cleared dirty bitmap. When the bit is
52 * set, it means the corresponding memory chunk needs a log-clear.
53 * Set this up to non-NULL to enable the capability to postpone
54 * and split clearing of dirty bitmap on the remote node (e.g.,
55 * KVM). The bitmap will be set only when doing global sync.
57 * NOTE: this bitmap is different comparing to the other bitmaps
58 * in that one bit can represent multiple guest pages (which is
59 * decided by the `clear_bmap_shift' variable below). On
60 * destination side, this should always be NULL, and the variable
61 * `clear_bmap_shift' is meaningless.
63 unsigned long *clear_bmap
;
64 uint8_t clear_bmap_shift
;
68 * clear_bmap_size: calculate clear bitmap size
70 * @pages: number of guest pages
71 * @shift: guest page number shift
73 * Returns: number of bits for the clear bitmap
75 static inline long clear_bmap_size(uint64_t pages
, uint8_t shift
)
77 return DIV_ROUND_UP(pages
, 1UL << shift
);
81 * clear_bmap_set: set clear bitmap for the page range
83 * @rb: the ramblock to operate on
84 * @start: the start page number
85 * @size: number of pages to set in the bitmap
89 static inline void clear_bmap_set(RAMBlock
*rb
, uint64_t start
,
92 uint8_t shift
= rb
->clear_bmap_shift
;
94 bitmap_set_atomic(rb
->clear_bmap
, start
>> shift
,
95 clear_bmap_size(npages
, shift
));
99 * clear_bmap_test_and_clear: test clear bitmap for the page, clear if set
101 * @rb: the ramblock to operate on
102 * @page: the page number to check
104 * Returns: true if the bit was set, false otherwise
106 static inline bool clear_bmap_test_and_clear(RAMBlock
*rb
, uint64_t page
)
108 uint8_t shift
= rb
->clear_bmap_shift
;
110 return bitmap_test_and_clear_atomic(rb
->clear_bmap
, page
>> shift
, 1);
113 static inline bool offset_in_ramblock(RAMBlock
*b
, ram_addr_t offset
)
115 return (b
&& b
->host
&& offset
< b
->used_length
) ? true : false;
118 static inline void *ramblock_ptr(RAMBlock
*block
, ram_addr_t offset
)
120 assert(offset_in_ramblock(block
, offset
));
121 return (char *)block
->host
+ offset
;
124 static inline unsigned long int ramblock_recv_bitmap_offset(void *host_addr
,
127 uint64_t host_addr_offset
=
128 (uint64_t)(uintptr_t)(host_addr
- (void *)rb
->host
);
129 return host_addr_offset
>> TARGET_PAGE_BITS
;
132 bool ramblock_is_pmem(RAMBlock
*rb
);
134 long qemu_minrampagesize(void);
135 long qemu_maxrampagesize(void);
138 * qemu_ram_alloc_from_file,
139 * qemu_ram_alloc_from_fd: Allocate a ram block from the specified backing
143 * @size: the size in bytes of the ram block
144 * @mr: the memory region where the ram block is
145 * @ram_flags: specify the properties of the ram block, which can be one
146 * or bit-or of following values
147 * - RAM_SHARED: mmap the backing file or device with MAP_SHARED
148 * - RAM_PMEM: the backend @mem_path or @fd is persistent memory
149 * Other bits are ignored.
150 * @mem_path or @fd: specify the backing file or device
151 * @errp: pointer to Error*, to store an error if it happens
154 * On success, return a pointer to the ram block.
155 * On failure, return NULL.
157 RAMBlock
*qemu_ram_alloc_from_file(ram_addr_t size
, MemoryRegion
*mr
,
158 uint32_t ram_flags
, const char *mem_path
,
160 RAMBlock
*qemu_ram_alloc_from_fd(ram_addr_t size
, MemoryRegion
*mr
,
161 uint32_t ram_flags
, int fd
,
164 RAMBlock
*qemu_ram_alloc_from_ptr(ram_addr_t size
, void *host
,
165 MemoryRegion
*mr
, Error
**errp
);
166 RAMBlock
*qemu_ram_alloc(ram_addr_t size
, bool share
, MemoryRegion
*mr
,
168 RAMBlock
*qemu_ram_alloc_resizeable(ram_addr_t size
, ram_addr_t max_size
,
169 void (*resized
)(const char*,
172 MemoryRegion
*mr
, Error
**errp
);
173 void qemu_ram_free(RAMBlock
*block
);
175 int qemu_ram_resize(RAMBlock
*block
, ram_addr_t newsize
, Error
**errp
);
177 #define DIRTY_CLIENTS_ALL ((1 << DIRTY_MEMORY_NUM) - 1)
178 #define DIRTY_CLIENTS_NOCODE (DIRTY_CLIENTS_ALL & ~(1 << DIRTY_MEMORY_CODE))
180 void tb_invalidate_phys_range(ram_addr_t start
, ram_addr_t end
);
182 static inline bool cpu_physical_memory_get_dirty(ram_addr_t start
,
186 DirtyMemoryBlocks
*blocks
;
187 unsigned long end
, page
;
188 unsigned long idx
, offset
, base
;
191 assert(client
< DIRTY_MEMORY_NUM
);
193 end
= TARGET_PAGE_ALIGN(start
+ length
) >> TARGET_PAGE_BITS
;
194 page
= start
>> TARGET_PAGE_BITS
;
198 blocks
= atomic_rcu_read(&ram_list
.dirty_memory
[client
]);
200 idx
= page
/ DIRTY_MEMORY_BLOCK_SIZE
;
201 offset
= page
% DIRTY_MEMORY_BLOCK_SIZE
;
202 base
= page
- offset
;
204 unsigned long next
= MIN(end
, base
+ DIRTY_MEMORY_BLOCK_SIZE
);
205 unsigned long num
= next
- base
;
206 unsigned long found
= find_next_bit(blocks
->blocks
[idx
], num
, offset
);
215 base
+= DIRTY_MEMORY_BLOCK_SIZE
;
223 static inline bool cpu_physical_memory_all_dirty(ram_addr_t start
,
227 DirtyMemoryBlocks
*blocks
;
228 unsigned long end
, page
;
229 unsigned long idx
, offset
, base
;
232 assert(client
< DIRTY_MEMORY_NUM
);
234 end
= TARGET_PAGE_ALIGN(start
+ length
) >> TARGET_PAGE_BITS
;
235 page
= start
>> TARGET_PAGE_BITS
;
239 blocks
= atomic_rcu_read(&ram_list
.dirty_memory
[client
]);
241 idx
= page
/ DIRTY_MEMORY_BLOCK_SIZE
;
242 offset
= page
% DIRTY_MEMORY_BLOCK_SIZE
;
243 base
= page
- offset
;
245 unsigned long next
= MIN(end
, base
+ DIRTY_MEMORY_BLOCK_SIZE
);
246 unsigned long num
= next
- base
;
247 unsigned long found
= find_next_zero_bit(blocks
->blocks
[idx
], num
, offset
);
256 base
+= DIRTY_MEMORY_BLOCK_SIZE
;
264 static inline bool cpu_physical_memory_get_dirty_flag(ram_addr_t addr
,
267 return cpu_physical_memory_get_dirty(addr
, 1, client
);
270 static inline bool cpu_physical_memory_is_clean(ram_addr_t addr
)
272 bool vga
= cpu_physical_memory_get_dirty_flag(addr
, DIRTY_MEMORY_VGA
);
273 bool code
= cpu_physical_memory_get_dirty_flag(addr
, DIRTY_MEMORY_CODE
);
275 cpu_physical_memory_get_dirty_flag(addr
, DIRTY_MEMORY_MIGRATION
);
276 return !(vga
&& code
&& migration
);
279 static inline uint8_t cpu_physical_memory_range_includes_clean(ram_addr_t start
,
285 if (mask
& (1 << DIRTY_MEMORY_VGA
) &&
286 !cpu_physical_memory_all_dirty(start
, length
, DIRTY_MEMORY_VGA
)) {
287 ret
|= (1 << DIRTY_MEMORY_VGA
);
289 if (mask
& (1 << DIRTY_MEMORY_CODE
) &&
290 !cpu_physical_memory_all_dirty(start
, length
, DIRTY_MEMORY_CODE
)) {
291 ret
|= (1 << DIRTY_MEMORY_CODE
);
293 if (mask
& (1 << DIRTY_MEMORY_MIGRATION
) &&
294 !cpu_physical_memory_all_dirty(start
, length
, DIRTY_MEMORY_MIGRATION
)) {
295 ret
|= (1 << DIRTY_MEMORY_MIGRATION
);
300 static inline void cpu_physical_memory_set_dirty_flag(ram_addr_t addr
,
303 unsigned long page
, idx
, offset
;
304 DirtyMemoryBlocks
*blocks
;
306 assert(client
< DIRTY_MEMORY_NUM
);
308 page
= addr
>> TARGET_PAGE_BITS
;
309 idx
= page
/ DIRTY_MEMORY_BLOCK_SIZE
;
310 offset
= page
% DIRTY_MEMORY_BLOCK_SIZE
;
314 blocks
= atomic_rcu_read(&ram_list
.dirty_memory
[client
]);
316 set_bit_atomic(offset
, blocks
->blocks
[idx
]);
321 static inline void cpu_physical_memory_set_dirty_range(ram_addr_t start
,
325 DirtyMemoryBlocks
*blocks
[DIRTY_MEMORY_NUM
];
326 unsigned long end
, page
;
327 unsigned long idx
, offset
, base
;
330 if (!mask
&& !xen_enabled()) {
334 end
= TARGET_PAGE_ALIGN(start
+ length
) >> TARGET_PAGE_BITS
;
335 page
= start
>> TARGET_PAGE_BITS
;
339 for (i
= 0; i
< DIRTY_MEMORY_NUM
; i
++) {
340 blocks
[i
] = atomic_rcu_read(&ram_list
.dirty_memory
[i
]);
343 idx
= page
/ DIRTY_MEMORY_BLOCK_SIZE
;
344 offset
= page
% DIRTY_MEMORY_BLOCK_SIZE
;
345 base
= page
- offset
;
347 unsigned long next
= MIN(end
, base
+ DIRTY_MEMORY_BLOCK_SIZE
);
349 if (likely(mask
& (1 << DIRTY_MEMORY_MIGRATION
))) {
350 bitmap_set_atomic(blocks
[DIRTY_MEMORY_MIGRATION
]->blocks
[idx
],
351 offset
, next
- page
);
353 if (unlikely(mask
& (1 << DIRTY_MEMORY_VGA
))) {
354 bitmap_set_atomic(blocks
[DIRTY_MEMORY_VGA
]->blocks
[idx
],
355 offset
, next
- page
);
357 if (unlikely(mask
& (1 << DIRTY_MEMORY_CODE
))) {
358 bitmap_set_atomic(blocks
[DIRTY_MEMORY_CODE
]->blocks
[idx
],
359 offset
, next
- page
);
365 base
+= DIRTY_MEMORY_BLOCK_SIZE
;
370 xen_hvm_modified_memory(start
, length
);
374 static inline void cpu_physical_memory_set_dirty_lebitmap(unsigned long *bitmap
,
379 unsigned long page_number
, c
;
382 unsigned long len
= (pages
+ HOST_LONG_BITS
- 1) / HOST_LONG_BITS
;
383 unsigned long hpratio
= getpagesize() / TARGET_PAGE_SIZE
;
384 unsigned long page
= BIT_WORD(start
>> TARGET_PAGE_BITS
);
386 /* start address is aligned at the start of a word? */
387 if ((((page
* BITS_PER_LONG
) << TARGET_PAGE_BITS
) == start
) &&
389 unsigned long **blocks
[DIRTY_MEMORY_NUM
];
391 unsigned long offset
;
393 long nr
= BITS_TO_LONGS(pages
);
395 idx
= (start
>> TARGET_PAGE_BITS
) / DIRTY_MEMORY_BLOCK_SIZE
;
396 offset
= BIT_WORD((start
>> TARGET_PAGE_BITS
) %
397 DIRTY_MEMORY_BLOCK_SIZE
);
401 for (i
= 0; i
< DIRTY_MEMORY_NUM
; i
++) {
402 blocks
[i
] = atomic_rcu_read(&ram_list
.dirty_memory
[i
])->blocks
;
405 for (k
= 0; k
< nr
; k
++) {
407 unsigned long temp
= leul_to_cpu(bitmap
[k
]);
409 atomic_or(&blocks
[DIRTY_MEMORY_VGA
][idx
][offset
], temp
);
411 if (global_dirty_log
) {
412 atomic_or(&blocks
[DIRTY_MEMORY_MIGRATION
][idx
][offset
],
417 atomic_or(&blocks
[DIRTY_MEMORY_CODE
][idx
][offset
], temp
);
421 if (++offset
>= BITS_TO_LONGS(DIRTY_MEMORY_BLOCK_SIZE
)) {
429 xen_hvm_modified_memory(start
, pages
<< TARGET_PAGE_BITS
);
431 uint8_t clients
= tcg_enabled() ? DIRTY_CLIENTS_ALL
: DIRTY_CLIENTS_NOCODE
;
433 if (!global_dirty_log
) {
434 clients
&= ~(1 << DIRTY_MEMORY_MIGRATION
);
438 * bitmap-traveling is faster than memory-traveling (for addr...)
439 * especially when most of the memory is not dirty.
441 for (i
= 0; i
< len
; i
++) {
442 if (bitmap
[i
] != 0) {
443 c
= leul_to_cpu(bitmap
[i
]);
447 page_number
= (i
* HOST_LONG_BITS
+ j
) * hpratio
;
448 addr
= page_number
* TARGET_PAGE_SIZE
;
449 ram_addr
= start
+ addr
;
450 cpu_physical_memory_set_dirty_range(ram_addr
,
451 TARGET_PAGE_SIZE
* hpratio
, clients
);
457 #endif /* not _WIN32 */
459 bool cpu_physical_memory_test_and_clear_dirty(ram_addr_t start
,
463 DirtyBitmapSnapshot
*cpu_physical_memory_snapshot_and_clear_dirty
464 (MemoryRegion
*mr
, hwaddr offset
, hwaddr length
, unsigned client
);
466 bool cpu_physical_memory_snapshot_get_dirty(DirtyBitmapSnapshot
*snap
,
470 static inline void cpu_physical_memory_clear_dirty_range(ram_addr_t start
,
473 cpu_physical_memory_test_and_clear_dirty(start
, length
, DIRTY_MEMORY_MIGRATION
);
474 cpu_physical_memory_test_and_clear_dirty(start
, length
, DIRTY_MEMORY_VGA
);
475 cpu_physical_memory_test_and_clear_dirty(start
, length
, DIRTY_MEMORY_CODE
);
479 /* Called with RCU critical section */
481 uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock
*rb
,
484 uint64_t *real_dirty_pages
)
487 unsigned long word
= BIT_WORD((start
+ rb
->offset
) >> TARGET_PAGE_BITS
);
488 uint64_t num_dirty
= 0;
489 unsigned long *dest
= rb
->bmap
;
491 /* start address and length is aligned at the start of a word? */
492 if (((word
* BITS_PER_LONG
) << TARGET_PAGE_BITS
) ==
493 (start
+ rb
->offset
) &&
494 !(length
& ((BITS_PER_LONG
<< TARGET_PAGE_BITS
) - 1))) {
496 int nr
= BITS_TO_LONGS(length
>> TARGET_PAGE_BITS
);
497 unsigned long * const *src
;
498 unsigned long idx
= (word
* BITS_PER_LONG
) / DIRTY_MEMORY_BLOCK_SIZE
;
499 unsigned long offset
= BIT_WORD((word
* BITS_PER_LONG
) %
500 DIRTY_MEMORY_BLOCK_SIZE
);
501 unsigned long page
= BIT_WORD(start
>> TARGET_PAGE_BITS
);
503 src
= atomic_rcu_read(
504 &ram_list
.dirty_memory
[DIRTY_MEMORY_MIGRATION
])->blocks
;
506 for (k
= page
; k
< page
+ nr
; k
++) {
507 if (src
[idx
][offset
]) {
508 unsigned long bits
= atomic_xchg(&src
[idx
][offset
], 0);
509 unsigned long new_dirty
;
510 *real_dirty_pages
+= ctpopl(bits
);
511 new_dirty
= ~dest
[k
];
514 num_dirty
+= ctpopl(new_dirty
);
517 if (++offset
>= BITS_TO_LONGS(DIRTY_MEMORY_BLOCK_SIZE
)) {
523 if (rb
->clear_bmap
) {
525 * Postpone the dirty bitmap clear to the point before we
526 * really send the pages, also we will split the clear
527 * dirty procedure into smaller chunks.
529 clear_bmap_set(rb
, start
>> TARGET_PAGE_BITS
,
530 length
>> TARGET_PAGE_BITS
);
532 /* Slow path - still do that in a huge chunk */
533 memory_region_clear_dirty_bitmap(rb
->mr
, start
, length
);
536 ram_addr_t offset
= rb
->offset
;
538 for (addr
= 0; addr
< length
; addr
+= TARGET_PAGE_SIZE
) {
539 if (cpu_physical_memory_test_and_clear_dirty(
540 start
+ addr
+ offset
,
542 DIRTY_MEMORY_MIGRATION
)) {
543 *real_dirty_pages
+= 1;
544 long k
= (start
+ addr
) >> TARGET_PAGE_BITS
;
545 if (!test_and_set_bit(k
, dest
)) {