4 * Copyright (c) 2003-2008 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
28 #include <sys/types.h>
32 #include "monitor/monitor.h"
33 #include "sysemu/sysemu.h"
34 #include "qemu/bitops.h"
35 #include "qemu/bitmap.h"
36 #include "sysemu/arch_init.h"
37 #include "audio/audio.h"
38 #include "hw/i386/pc.h"
39 #include "hw/pci/pci.h"
40 #include "hw/audio/audio.h"
41 #include "sysemu/kvm.h"
42 #include "migration/migration.h"
43 #include "hw/i386/smbios.h"
44 #include "exec/address-spaces.h"
45 #include "hw/audio/pcspk.h"
46 #include "migration/page_cache.h"
47 #include "qemu/config-file.h"
48 #include "qemu/error-report.h"
49 #include "qmp-commands.h"
51 #include "exec/cpu-all.h"
52 #include "exec/ram_addr.h"
53 #include "hw/acpi/acpi.h"
54 #include "qemu/host-utils.h"
56 #ifdef DEBUG_ARCH_INIT
57 #define DPRINTF(fmt, ...) \
58 do { fprintf(stdout, "arch_init: " fmt, ## __VA_ARGS__); } while (0)
60 #define DPRINTF(fmt, ...) \
65 int graphic_width
= 1024;
66 int graphic_height
= 768;
67 int graphic_depth
= 8;
69 int graphic_width
= 800;
70 int graphic_height
= 600;
71 int graphic_depth
= 32;
75 #if defined(TARGET_ALPHA)
76 #define QEMU_ARCH QEMU_ARCH_ALPHA
77 #elif defined(TARGET_ARM)
78 #define QEMU_ARCH QEMU_ARCH_ARM
79 #elif defined(TARGET_CRIS)
80 #define QEMU_ARCH QEMU_ARCH_CRIS
81 #elif defined(TARGET_I386)
82 #define QEMU_ARCH QEMU_ARCH_I386
83 #elif defined(TARGET_M68K)
84 #define QEMU_ARCH QEMU_ARCH_M68K
85 #elif defined(TARGET_LM32)
86 #define QEMU_ARCH QEMU_ARCH_LM32
87 #elif defined(TARGET_MICROBLAZE)
88 #define QEMU_ARCH QEMU_ARCH_MICROBLAZE
89 #elif defined(TARGET_MIPS)
90 #define QEMU_ARCH QEMU_ARCH_MIPS
91 #elif defined(TARGET_MOXIE)
92 #define QEMU_ARCH QEMU_ARCH_MOXIE
93 #elif defined(TARGET_OPENRISC)
94 #define QEMU_ARCH QEMU_ARCH_OPENRISC
95 #elif defined(TARGET_PPC)
96 #define QEMU_ARCH QEMU_ARCH_PPC
97 #elif defined(TARGET_S390X)
98 #define QEMU_ARCH QEMU_ARCH_S390X
99 #elif defined(TARGET_SH4)
100 #define QEMU_ARCH QEMU_ARCH_SH4
101 #elif defined(TARGET_SPARC)
102 #define QEMU_ARCH QEMU_ARCH_SPARC
103 #elif defined(TARGET_XTENSA)
104 #define QEMU_ARCH QEMU_ARCH_XTENSA
105 #elif defined(TARGET_UNICORE32)
106 #define QEMU_ARCH QEMU_ARCH_UNICORE32
109 const uint32_t arch_type
= QEMU_ARCH
;
110 static bool mig_throttle_on
;
111 static int dirty_rate_high_cnt
;
112 static void check_guest_throttling(void);
114 static uint64_t bitmap_sync_count
;
116 /***********************************************************/
117 /* ram save/restore */
119 #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */
120 #define RAM_SAVE_FLAG_COMPRESS 0x02
121 #define RAM_SAVE_FLAG_MEM_SIZE 0x04
122 #define RAM_SAVE_FLAG_PAGE 0x08
123 #define RAM_SAVE_FLAG_EOS 0x10
124 #define RAM_SAVE_FLAG_CONTINUE 0x20
125 #define RAM_SAVE_FLAG_XBZRLE 0x40
126 /* 0x80 is reserved in migration.h start with 0x100 next */
128 static struct defconfig_file
{
129 const char *filename
;
130 /* Indicates it is an user config file (disabled by -no-user-config) */
132 } default_config_files
[] = {
133 { CONFIG_QEMU_CONFDIR
"/qemu.conf", true },
134 { CONFIG_QEMU_CONFDIR
"/target-" TARGET_NAME
".conf", true },
135 { NULL
}, /* end of list */
138 static const uint8_t ZERO_TARGET_PAGE
[TARGET_PAGE_SIZE
];
140 int qemu_read_default_config_files(bool userconfig
)
143 struct defconfig_file
*f
;
145 for (f
= default_config_files
; f
->filename
; f
++) {
146 if (!userconfig
&& f
->userconfig
) {
149 ret
= qemu_read_config_file(f
->filename
);
150 if (ret
< 0 && ret
!= -ENOENT
) {
158 static inline bool is_zero_range(uint8_t *p
, uint64_t size
)
160 return buffer_find_nonzero_offset(p
, size
) == size
;
163 /* struct contains XBZRLE cache and a static page
164 used by the compression */
166 /* buffer used for XBZRLE encoding */
167 uint8_t *encoded_buf
;
168 /* buffer for storing page content */
169 uint8_t *current_buf
;
170 /* Cache for XBZRLE, Protected by lock. */
175 /* buffer used for XBZRLE decoding */
176 static uint8_t *xbzrle_decoded_buf
;
178 static void XBZRLE_cache_lock(void)
180 if (migrate_use_xbzrle())
181 qemu_mutex_lock(&XBZRLE
.lock
);
184 static void XBZRLE_cache_unlock(void)
186 if (migrate_use_xbzrle())
187 qemu_mutex_unlock(&XBZRLE
.lock
);
191 * called from qmp_migrate_set_cache_size in main thread, possibly while
192 * a migration is in progress.
193 * A running migration maybe using the cache and might finish during this
194 * call, hence changes to the cache are protected by XBZRLE.lock().
196 int64_t xbzrle_cache_resize(int64_t new_size
)
198 PageCache
*new_cache
;
201 if (new_size
< TARGET_PAGE_SIZE
) {
207 if (XBZRLE
.cache
!= NULL
) {
208 if (pow2floor(new_size
) == migrate_xbzrle_cache_size()) {
211 new_cache
= cache_init(new_size
/ TARGET_PAGE_SIZE
,
214 error_report("Error creating cache");
219 cache_fini(XBZRLE
.cache
);
220 XBZRLE
.cache
= new_cache
;
224 ret
= pow2floor(new_size
);
226 XBZRLE_cache_unlock();
230 /* accounting for migration statistics */
231 typedef struct AccountingInfo
{
233 uint64_t skipped_pages
;
236 uint64_t xbzrle_bytes
;
237 uint64_t xbzrle_pages
;
238 uint64_t xbzrle_cache_miss
;
239 double xbzrle_cache_miss_rate
;
240 uint64_t xbzrle_overflows
;
243 static AccountingInfo acct_info
;
245 static void acct_clear(void)
247 memset(&acct_info
, 0, sizeof(acct_info
));
250 uint64_t dup_mig_bytes_transferred(void)
252 return acct_info
.dup_pages
* TARGET_PAGE_SIZE
;
255 uint64_t dup_mig_pages_transferred(void)
257 return acct_info
.dup_pages
;
260 uint64_t skipped_mig_bytes_transferred(void)
262 return acct_info
.skipped_pages
* TARGET_PAGE_SIZE
;
265 uint64_t skipped_mig_pages_transferred(void)
267 return acct_info
.skipped_pages
;
270 uint64_t norm_mig_bytes_transferred(void)
272 return acct_info
.norm_pages
* TARGET_PAGE_SIZE
;
275 uint64_t norm_mig_pages_transferred(void)
277 return acct_info
.norm_pages
;
280 uint64_t xbzrle_mig_bytes_transferred(void)
282 return acct_info
.xbzrle_bytes
;
285 uint64_t xbzrle_mig_pages_transferred(void)
287 return acct_info
.xbzrle_pages
;
290 uint64_t xbzrle_mig_pages_cache_miss(void)
292 return acct_info
.xbzrle_cache_miss
;
295 double xbzrle_mig_cache_miss_rate(void)
297 return acct_info
.xbzrle_cache_miss_rate
;
300 uint64_t xbzrle_mig_pages_overflow(void)
302 return acct_info
.xbzrle_overflows
;
305 static size_t save_block_hdr(QEMUFile
*f
, RAMBlock
*block
, ram_addr_t offset
,
310 qemu_put_be64(f
, offset
| cont
| flag
);
314 qemu_put_byte(f
, strlen(block
->idstr
));
315 qemu_put_buffer(f
, (uint8_t *)block
->idstr
,
316 strlen(block
->idstr
));
317 size
+= 1 + strlen(block
->idstr
);
322 /* This is the last block that we have visited serching for dirty pages
324 static RAMBlock
*last_seen_block
;
325 /* This is the last block from where we have sent data */
326 static RAMBlock
*last_sent_block
;
327 static ram_addr_t last_offset
;
328 static unsigned long *migration_bitmap
;
329 static uint64_t migration_dirty_pages
;
330 static uint32_t last_version
;
331 static bool ram_bulk_stage
;
333 /* Update the xbzrle cache to reflect a page that's been sent as all 0.
334 * The important thing is that a stale (not-yet-0'd) page be replaced
336 * As a bonus, if the page wasn't in the cache it gets added so that
337 * when a small write is made into the 0'd page it gets XBZRLE sent
339 static void xbzrle_cache_zero_page(ram_addr_t current_addr
)
341 if (ram_bulk_stage
|| !migrate_use_xbzrle()) {
345 /* We don't care if this fails to allocate a new cache page
346 * as long as it updated an old one */
347 cache_insert(XBZRLE
.cache
, current_addr
, ZERO_TARGET_PAGE
);
350 #define ENCODING_FLAG_XBZRLE 0x1
352 static int save_xbzrle_page(QEMUFile
*f
, uint8_t **current_data
,
353 ram_addr_t current_addr
, RAMBlock
*block
,
354 ram_addr_t offset
, int cont
, bool last_stage
)
356 int encoded_len
= 0, bytes_sent
= -1;
357 uint8_t *prev_cached_page
;
359 if (!cache_is_cached(XBZRLE
.cache
, current_addr
)) {
360 acct_info
.xbzrle_cache_miss
++;
362 if (cache_insert(XBZRLE
.cache
, current_addr
, *current_data
) == -1) {
365 /* update *current_data when the page has been
366 inserted into cache */
367 *current_data
= get_cached_data(XBZRLE
.cache
, current_addr
);
373 prev_cached_page
= get_cached_data(XBZRLE
.cache
, current_addr
);
375 /* save current buffer into memory */
376 memcpy(XBZRLE
.current_buf
, *current_data
, TARGET_PAGE_SIZE
);
378 /* XBZRLE encoding (if there is no overflow) */
379 encoded_len
= xbzrle_encode_buffer(prev_cached_page
, XBZRLE
.current_buf
,
380 TARGET_PAGE_SIZE
, XBZRLE
.encoded_buf
,
382 if (encoded_len
== 0) {
383 DPRINTF("Skipping unmodified page\n");
385 } else if (encoded_len
== -1) {
386 DPRINTF("Overflow\n");
387 acct_info
.xbzrle_overflows
++;
388 /* update data in the cache */
390 memcpy(prev_cached_page
, *current_data
, TARGET_PAGE_SIZE
);
391 *current_data
= prev_cached_page
;
396 /* we need to update the data in the cache, in order to get the same data */
398 memcpy(prev_cached_page
, XBZRLE
.current_buf
, TARGET_PAGE_SIZE
);
401 /* Send XBZRLE based compressed page */
402 bytes_sent
= save_block_hdr(f
, block
, offset
, cont
, RAM_SAVE_FLAG_XBZRLE
);
403 qemu_put_byte(f
, ENCODING_FLAG_XBZRLE
);
404 qemu_put_be16(f
, encoded_len
);
405 qemu_put_buffer(f
, XBZRLE
.encoded_buf
, encoded_len
);
406 bytes_sent
+= encoded_len
+ 1 + 2;
407 acct_info
.xbzrle_pages
++;
408 acct_info
.xbzrle_bytes
+= bytes_sent
;
414 ram_addr_t
migration_bitmap_find_and_reset_dirty(MemoryRegion
*mr
,
417 unsigned long base
= mr
->ram_addr
>> TARGET_PAGE_BITS
;
418 unsigned long nr
= base
+ (start
>> TARGET_PAGE_BITS
);
419 uint64_t mr_size
= TARGET_PAGE_ALIGN(memory_region_size(mr
));
420 unsigned long size
= base
+ (mr_size
>> TARGET_PAGE_BITS
);
424 if (ram_bulk_stage
&& nr
> base
) {
427 next
= find_next_bit(migration_bitmap
, size
, nr
);
431 clear_bit(next
, migration_bitmap
);
432 migration_dirty_pages
--;
434 return (next
- base
) << TARGET_PAGE_BITS
;
437 static inline bool migration_bitmap_set_dirty(ram_addr_t addr
)
440 int nr
= addr
>> TARGET_PAGE_BITS
;
442 ret
= test_and_set_bit(nr
, migration_bitmap
);
445 migration_dirty_pages
++;
450 static void migration_bitmap_sync_range(ram_addr_t start
, ram_addr_t length
)
453 unsigned long page
= BIT_WORD(start
>> TARGET_PAGE_BITS
);
455 /* start address is aligned at the start of a word? */
456 if (((page
* BITS_PER_LONG
) << TARGET_PAGE_BITS
) == start
) {
458 int nr
= BITS_TO_LONGS(length
>> TARGET_PAGE_BITS
);
459 unsigned long *src
= ram_list
.dirty_memory
[DIRTY_MEMORY_MIGRATION
];
461 for (k
= page
; k
< page
+ nr
; k
++) {
463 unsigned long new_dirty
;
464 new_dirty
= ~migration_bitmap
[k
];
465 migration_bitmap
[k
] |= src
[k
];
467 migration_dirty_pages
+= ctpopl(new_dirty
);
472 for (addr
= 0; addr
< length
; addr
+= TARGET_PAGE_SIZE
) {
473 if (cpu_physical_memory_get_dirty(start
+ addr
,
475 DIRTY_MEMORY_MIGRATION
)) {
476 cpu_physical_memory_reset_dirty(start
+ addr
,
478 DIRTY_MEMORY_MIGRATION
);
479 migration_bitmap_set_dirty(start
+ addr
);
486 /* Needs iothread lock! */
488 static void migration_bitmap_sync(void)
491 uint64_t num_dirty_pages_init
= migration_dirty_pages
;
492 MigrationState
*s
= migrate_get_current();
493 static int64_t start_time
;
494 static int64_t bytes_xfer_prev
;
495 static int64_t num_dirty_pages_period
;
497 int64_t bytes_xfer_now
;
498 static uint64_t xbzrle_cache_miss_prev
;
499 static uint64_t iterations_prev
;
503 if (!bytes_xfer_prev
) {
504 bytes_xfer_prev
= ram_bytes_transferred();
508 start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
511 trace_migration_bitmap_sync_start();
512 address_space_sync_dirty_bitmap(&address_space_memory
);
514 QTAILQ_FOREACH(block
, &ram_list
.blocks
, next
) {
515 migration_bitmap_sync_range(block
->mr
->ram_addr
, block
->length
);
517 trace_migration_bitmap_sync_end(migration_dirty_pages
518 - num_dirty_pages_init
);
519 num_dirty_pages_period
+= migration_dirty_pages
- num_dirty_pages_init
;
520 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
522 /* more than 1 second = 1000 millisecons */
523 if (end_time
> start_time
+ 1000) {
524 if (migrate_auto_converge()) {
525 /* The following detection logic can be refined later. For now:
526 Check to see if the dirtied bytes is 50% more than the approx.
527 amount of bytes that just got transferred since the last time we
528 were in this routine. If that happens >N times (for now N==4)
529 we turn on the throttle down logic */
530 bytes_xfer_now
= ram_bytes_transferred();
531 if (s
->dirty_pages_rate
&&
532 (num_dirty_pages_period
* TARGET_PAGE_SIZE
>
533 (bytes_xfer_now
- bytes_xfer_prev
)/2) &&
534 (dirty_rate_high_cnt
++ > 4)) {
535 trace_migration_throttle();
536 mig_throttle_on
= true;
537 dirty_rate_high_cnt
= 0;
539 bytes_xfer_prev
= bytes_xfer_now
;
541 mig_throttle_on
= false;
543 if (migrate_use_xbzrle()) {
544 if (iterations_prev
!= 0) {
545 acct_info
.xbzrle_cache_miss_rate
=
546 (double)(acct_info
.xbzrle_cache_miss
-
547 xbzrle_cache_miss_prev
) /
548 (acct_info
.iterations
- iterations_prev
);
550 iterations_prev
= acct_info
.iterations
;
551 xbzrle_cache_miss_prev
= acct_info
.xbzrle_cache_miss
;
553 s
->dirty_pages_rate
= num_dirty_pages_period
* 1000
554 / (end_time
- start_time
);
555 s
->dirty_bytes_rate
= s
->dirty_pages_rate
* TARGET_PAGE_SIZE
;
556 start_time
= end_time
;
557 num_dirty_pages_period
= 0;
558 s
->dirty_sync_count
= bitmap_sync_count
;
563 * ram_save_block: Writes a page of memory to the stream f
565 * Returns: The number of bytes written.
566 * 0 means no dirty pages
569 static int ram_save_block(QEMUFile
*f
, bool last_stage
)
571 RAMBlock
*block
= last_seen_block
;
572 ram_addr_t offset
= last_offset
;
573 bool complete_round
= false;
576 ram_addr_t current_addr
;
579 block
= QTAILQ_FIRST(&ram_list
.blocks
);
583 offset
= migration_bitmap_find_and_reset_dirty(mr
, offset
);
584 if (complete_round
&& block
== last_seen_block
&&
585 offset
>= last_offset
) {
588 if (offset
>= block
->length
) {
590 block
= QTAILQ_NEXT(block
, next
);
592 block
= QTAILQ_FIRST(&ram_list
.blocks
);
593 complete_round
= true;
594 ram_bulk_stage
= false;
599 bool send_async
= true;
600 int cont
= (block
== last_sent_block
) ?
601 RAM_SAVE_FLAG_CONTINUE
: 0;
603 p
= memory_region_get_ram_ptr(mr
) + offset
;
605 /* In doubt sent page as normal */
607 ret
= ram_control_save_page(f
, block
->offset
,
608 offset
, TARGET_PAGE_SIZE
, &bytes_sent
);
612 current_addr
= block
->offset
+ offset
;
613 if (ret
!= RAM_SAVE_CONTROL_NOT_SUPP
) {
614 if (ret
!= RAM_SAVE_CONTROL_DELAYED
) {
615 if (bytes_sent
> 0) {
616 acct_info
.norm_pages
++;
617 } else if (bytes_sent
== 0) {
618 acct_info
.dup_pages
++;
621 } else if (is_zero_range(p
, TARGET_PAGE_SIZE
)) {
622 acct_info
.dup_pages
++;
623 bytes_sent
= save_block_hdr(f
, block
, offset
, cont
,
624 RAM_SAVE_FLAG_COMPRESS
);
627 /* Must let xbzrle know, otherwise a previous (now 0'd) cached
628 * page would be stale
630 xbzrle_cache_zero_page(current_addr
);
631 } else if (!ram_bulk_stage
&& migrate_use_xbzrle()) {
632 bytes_sent
= save_xbzrle_page(f
, &p
, current_addr
, block
,
633 offset
, cont
, last_stage
);
635 /* Can't send this cached data async, since the cache page
636 * might get updated before it gets to the wire
642 /* XBZRLE overflow or normal page */
643 if (bytes_sent
== -1) {
644 bytes_sent
= save_block_hdr(f
, block
, offset
, cont
, RAM_SAVE_FLAG_PAGE
);
646 qemu_put_buffer_async(f
, p
, TARGET_PAGE_SIZE
);
648 qemu_put_buffer(f
, p
, TARGET_PAGE_SIZE
);
650 bytes_sent
+= TARGET_PAGE_SIZE
;
651 acct_info
.norm_pages
++;
654 XBZRLE_cache_unlock();
655 /* if page is unmodified, continue to the next */
656 if (bytes_sent
> 0) {
657 last_sent_block
= block
;
662 last_seen_block
= block
;
663 last_offset
= offset
;
668 static uint64_t bytes_transferred
;
670 void acct_update_position(QEMUFile
*f
, size_t size
, bool zero
)
672 uint64_t pages
= size
/ TARGET_PAGE_SIZE
;
674 acct_info
.dup_pages
+= pages
;
676 acct_info
.norm_pages
+= pages
;
677 bytes_transferred
+= size
;
678 qemu_update_position(f
, size
);
682 static ram_addr_t
ram_save_remaining(void)
684 return migration_dirty_pages
;
687 uint64_t ram_bytes_remaining(void)
689 return ram_save_remaining() * TARGET_PAGE_SIZE
;
692 uint64_t ram_bytes_transferred(void)
694 return bytes_transferred
;
697 uint64_t ram_bytes_total(void)
702 QTAILQ_FOREACH(block
, &ram_list
.blocks
, next
)
703 total
+= block
->length
;
708 void free_xbzrle_decoded_buf(void)
710 g_free(xbzrle_decoded_buf
);
711 xbzrle_decoded_buf
= NULL
;
714 static void migration_end(void)
716 if (migration_bitmap
) {
717 memory_global_dirty_log_stop();
718 g_free(migration_bitmap
);
719 migration_bitmap
= NULL
;
724 cache_fini(XBZRLE
.cache
);
725 g_free(XBZRLE
.cache
);
726 g_free(XBZRLE
.encoded_buf
);
727 g_free(XBZRLE
.current_buf
);
729 XBZRLE
.encoded_buf
= NULL
;
730 XBZRLE
.current_buf
= NULL
;
732 XBZRLE_cache_unlock();
735 static void ram_migration_cancel(void *opaque
)
740 static void reset_ram_globals(void)
742 last_seen_block
= NULL
;
743 last_sent_block
= NULL
;
745 last_version
= ram_list
.version
;
746 ram_bulk_stage
= true;
749 #define MAX_WAIT 50 /* ms, half buffered_file limit */
751 static int ram_save_setup(QEMUFile
*f
, void *opaque
)
754 int64_t ram_bitmap_pages
; /* Size of bitmap in pages, including gaps */
756 mig_throttle_on
= false;
757 dirty_rate_high_cnt
= 0;
758 bitmap_sync_count
= 0;
760 if (migrate_use_xbzrle()) {
762 XBZRLE
.cache
= cache_init(migrate_xbzrle_cache_size() /
766 XBZRLE_cache_unlock();
767 error_report("Error creating cache");
770 XBZRLE_cache_unlock();
772 /* We prefer not to abort if there is no memory */
773 XBZRLE
.encoded_buf
= g_try_malloc0(TARGET_PAGE_SIZE
);
774 if (!XBZRLE
.encoded_buf
) {
775 error_report("Error allocating encoded_buf");
779 XBZRLE
.current_buf
= g_try_malloc(TARGET_PAGE_SIZE
);
780 if (!XBZRLE
.current_buf
) {
781 error_report("Error allocating current_buf");
782 g_free(XBZRLE
.encoded_buf
);
783 XBZRLE
.encoded_buf
= NULL
;
790 qemu_mutex_lock_iothread();
791 qemu_mutex_lock_ramlist();
792 bytes_transferred
= 0;
795 ram_bitmap_pages
= last_ram_offset() >> TARGET_PAGE_BITS
;
796 migration_bitmap
= bitmap_new(ram_bitmap_pages
);
797 bitmap_set(migration_bitmap
, 0, ram_bitmap_pages
);
800 * Count the total number of pages used by ram blocks not including any
801 * gaps due to alignment or unplugs.
803 migration_dirty_pages
= 0;
804 QTAILQ_FOREACH(block
, &ram_list
.blocks
, next
) {
805 uint64_t block_pages
;
807 block_pages
= block
->length
>> TARGET_PAGE_BITS
;
808 migration_dirty_pages
+= block_pages
;
811 memory_global_dirty_log_start();
812 migration_bitmap_sync();
813 qemu_mutex_unlock_iothread();
815 qemu_put_be64(f
, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE
);
817 QTAILQ_FOREACH(block
, &ram_list
.blocks
, next
) {
818 qemu_put_byte(f
, strlen(block
->idstr
));
819 qemu_put_buffer(f
, (uint8_t *)block
->idstr
, strlen(block
->idstr
));
820 qemu_put_be64(f
, block
->length
);
823 qemu_mutex_unlock_ramlist();
825 ram_control_before_iterate(f
, RAM_CONTROL_SETUP
);
826 ram_control_after_iterate(f
, RAM_CONTROL_SETUP
);
828 qemu_put_be64(f
, RAM_SAVE_FLAG_EOS
);
833 static int ram_save_iterate(QEMUFile
*f
, void *opaque
)
840 qemu_mutex_lock_ramlist();
842 if (ram_list
.version
!= last_version
) {
846 ram_control_before_iterate(f
, RAM_CONTROL_ROUND
);
848 t0
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
850 while ((ret
= qemu_file_rate_limit(f
)) == 0) {
853 bytes_sent
= ram_save_block(f
, false);
854 /* no more blocks to sent */
855 if (bytes_sent
== 0) {
858 total_sent
+= bytes_sent
;
859 acct_info
.iterations
++;
860 check_guest_throttling();
861 /* we want to check in the 1st loop, just in case it was the 1st time
862 and we had to sync the dirty bitmap.
863 qemu_get_clock_ns() is a bit expensive, so we only check each some
867 uint64_t t1
= (qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - t0
) / 1000000;
869 DPRINTF("big wait: %" PRIu64
" milliseconds, %d iterations\n",
877 qemu_mutex_unlock_ramlist();
880 * Must occur before EOS (or any QEMUFile operation)
881 * because of RDMA protocol.
883 ram_control_after_iterate(f
, RAM_CONTROL_ROUND
);
885 bytes_transferred
+= total_sent
;
888 * Do not count these 8 bytes into total_sent, so that we can
889 * return 0 if no page had been dirtied.
891 qemu_put_be64(f
, RAM_SAVE_FLAG_EOS
);
892 bytes_transferred
+= 8;
894 ret
= qemu_file_get_error(f
);
902 static int ram_save_complete(QEMUFile
*f
, void *opaque
)
904 qemu_mutex_lock_ramlist();
905 migration_bitmap_sync();
907 ram_control_before_iterate(f
, RAM_CONTROL_FINISH
);
909 /* try transferring iterative blocks of memory */
911 /* flush all remaining blocks regardless of rate limiting */
915 bytes_sent
= ram_save_block(f
, true);
916 /* no more blocks to sent */
917 if (bytes_sent
== 0) {
920 bytes_transferred
+= bytes_sent
;
923 ram_control_after_iterate(f
, RAM_CONTROL_FINISH
);
926 qemu_mutex_unlock_ramlist();
927 qemu_put_be64(f
, RAM_SAVE_FLAG_EOS
);
932 static uint64_t ram_save_pending(QEMUFile
*f
, void *opaque
, uint64_t max_size
)
934 uint64_t remaining_size
;
936 remaining_size
= ram_save_remaining() * TARGET_PAGE_SIZE
;
938 if (remaining_size
< max_size
) {
939 qemu_mutex_lock_iothread();
940 migration_bitmap_sync();
941 qemu_mutex_unlock_iothread();
942 remaining_size
= ram_save_remaining() * TARGET_PAGE_SIZE
;
944 return remaining_size
;
947 static int load_xbzrle(QEMUFile
*f
, ram_addr_t addr
, void *host
)
953 if (!xbzrle_decoded_buf
) {
954 xbzrle_decoded_buf
= g_malloc(TARGET_PAGE_SIZE
);
957 /* extract RLE header */
958 xh_flags
= qemu_get_byte(f
);
959 xh_len
= qemu_get_be16(f
);
961 if (xh_flags
!= ENCODING_FLAG_XBZRLE
) {
962 fprintf(stderr
, "Failed to load XBZRLE page - wrong compression!\n");
966 if (xh_len
> TARGET_PAGE_SIZE
) {
967 fprintf(stderr
, "Failed to load XBZRLE page - len overflow!\n");
970 /* load data and decode */
971 qemu_get_buffer(f
, xbzrle_decoded_buf
, xh_len
);
974 ret
= xbzrle_decode_buffer(xbzrle_decoded_buf
, xh_len
, host
,
977 fprintf(stderr
, "Failed to load XBZRLE page - decode error!\n");
979 } else if (ret
> TARGET_PAGE_SIZE
) {
980 fprintf(stderr
, "Failed to load XBZRLE page - size %d exceeds %d!\n",
981 ret
, TARGET_PAGE_SIZE
);
988 static inline void *host_from_stream_offset(QEMUFile
*f
,
992 static RAMBlock
*block
= NULL
;
996 if (flags
& RAM_SAVE_FLAG_CONTINUE
) {
998 fprintf(stderr
, "Ack, bad migration stream!\n");
1002 return memory_region_get_ram_ptr(block
->mr
) + offset
;
1005 len
= qemu_get_byte(f
);
1006 qemu_get_buffer(f
, (uint8_t *)id
, len
);
1009 QTAILQ_FOREACH(block
, &ram_list
.blocks
, next
) {
1010 if (!strncmp(id
, block
->idstr
, sizeof(id
)))
1011 return memory_region_get_ram_ptr(block
->mr
) + offset
;
1014 fprintf(stderr
, "Can't find block %s!\n", id
);
1019 * If a page (or a whole RDMA chunk) has been
1020 * determined to be zero, then zap it.
1022 void ram_handle_compressed(void *host
, uint8_t ch
, uint64_t size
)
1024 if (ch
!= 0 || !is_zero_range(host
, size
)) {
1025 memset(host
, ch
, size
);
1029 static int ram_load(QEMUFile
*f
, void *opaque
, int version_id
)
1034 static uint64_t seq_iter
;
1038 if (version_id
!= 4) {
1044 addr
= qemu_get_be64(f
);
1046 flags
= addr
& ~TARGET_PAGE_MASK
;
1047 addr
&= TARGET_PAGE_MASK
;
1049 if (flags
& RAM_SAVE_FLAG_MEM_SIZE
) {
1050 /* Synchronize RAM block list */
1053 ram_addr_t total_ram_bytes
= addr
;
1055 while (total_ram_bytes
) {
1059 len
= qemu_get_byte(f
);
1060 qemu_get_buffer(f
, (uint8_t *)id
, len
);
1062 length
= qemu_get_be64(f
);
1064 QTAILQ_FOREACH(block
, &ram_list
.blocks
, next
) {
1065 if (!strncmp(id
, block
->idstr
, sizeof(id
))) {
1066 if (block
->length
!= length
) {
1068 "Length mismatch: %s: " RAM_ADDR_FMT
1069 " in != " RAM_ADDR_FMT
"\n", id
, length
,
1079 fprintf(stderr
, "Unknown ramblock \"%s\", cannot "
1080 "accept migration\n", id
);
1085 total_ram_bytes
-= length
;
1089 if (flags
& RAM_SAVE_FLAG_COMPRESS
) {
1093 host
= host_from_stream_offset(f
, addr
, flags
);
1099 ch
= qemu_get_byte(f
);
1100 ram_handle_compressed(host
, ch
, TARGET_PAGE_SIZE
);
1101 } else if (flags
& RAM_SAVE_FLAG_PAGE
) {
1104 host
= host_from_stream_offset(f
, addr
, flags
);
1110 qemu_get_buffer(f
, host
, TARGET_PAGE_SIZE
);
1111 } else if (flags
& RAM_SAVE_FLAG_XBZRLE
) {
1112 void *host
= host_from_stream_offset(f
, addr
, flags
);
1118 if (load_xbzrle(f
, addr
, host
) < 0) {
1122 } else if (flags
& RAM_SAVE_FLAG_HOOK
) {
1123 ram_control_load_hook(f
, flags
);
1125 error
= qemu_file_get_error(f
);
1130 } while (!(flags
& RAM_SAVE_FLAG_EOS
));
1133 DPRINTF("Completed load of VM with exit code %d seq iteration "
1134 "%" PRIu64
"\n", ret
, seq_iter
);
1138 static SaveVMHandlers savevm_ram_handlers
= {
1139 .save_live_setup
= ram_save_setup
,
1140 .save_live_iterate
= ram_save_iterate
,
1141 .save_live_complete
= ram_save_complete
,
1142 .save_live_pending
= ram_save_pending
,
1143 .load_state
= ram_load
,
1144 .cancel
= ram_migration_cancel
,
1147 void ram_mig_init(void)
1149 qemu_mutex_init(&XBZRLE
.lock
);
1150 register_savevm_live(NULL
, "ram", 0, 4, &savevm_ram_handlers
, NULL
);
1159 int (*init_isa
) (ISABus
*bus
);
1160 int (*init_pci
) (PCIBus
*bus
);
1164 static struct soundhw soundhw
[9];
1165 static int soundhw_count
;
1167 void isa_register_soundhw(const char *name
, const char *descr
,
1168 int (*init_isa
)(ISABus
*bus
))
1170 assert(soundhw_count
< ARRAY_SIZE(soundhw
) - 1);
1171 soundhw
[soundhw_count
].name
= name
;
1172 soundhw
[soundhw_count
].descr
= descr
;
1173 soundhw
[soundhw_count
].isa
= 1;
1174 soundhw
[soundhw_count
].init
.init_isa
= init_isa
;
1178 void pci_register_soundhw(const char *name
, const char *descr
,
1179 int (*init_pci
)(PCIBus
*bus
))
1181 assert(soundhw_count
< ARRAY_SIZE(soundhw
) - 1);
1182 soundhw
[soundhw_count
].name
= name
;
1183 soundhw
[soundhw_count
].descr
= descr
;
1184 soundhw
[soundhw_count
].isa
= 0;
1185 soundhw
[soundhw_count
].init
.init_pci
= init_pci
;
1189 void select_soundhw(const char *optarg
)
1193 if (is_help_option(optarg
)) {
1196 if (soundhw_count
) {
1197 printf("Valid sound card names (comma separated):\n");
1198 for (c
= soundhw
; c
->name
; ++c
) {
1199 printf ("%-11s %s\n", c
->name
, c
->descr
);
1201 printf("\n-soundhw all will enable all of the above\n");
1203 printf("Machine has no user-selectable audio hardware "
1204 "(it may or may not have always-present audio hardware).\n");
1206 exit(!is_help_option(optarg
));
1214 if (!strcmp(optarg
, "all")) {
1215 for (c
= soundhw
; c
->name
; ++c
) {
1224 l
= !e
? strlen(p
) : (size_t) (e
- p
);
1226 for (c
= soundhw
; c
->name
; ++c
) {
1227 if (!strncmp(c
->name
, p
, l
) && !c
->name
[l
]) {
1236 "Unknown sound card name (too big to show)\n");
1239 fprintf(stderr
, "Unknown sound card name `%.*s'\n",
1244 p
+= l
+ (e
!= NULL
);
1248 goto show_valid_cards
;
1253 void audio_init(void)
1256 ISABus
*isa_bus
= (ISABus
*) object_resolve_path_type("", TYPE_ISA_BUS
, NULL
);
1257 PCIBus
*pci_bus
= (PCIBus
*) object_resolve_path_type("", TYPE_PCI_BUS
, NULL
);
1259 for (c
= soundhw
; c
->name
; ++c
) {
1263 fprintf(stderr
, "ISA bus not available for %s\n", c
->name
);
1266 c
->init
.init_isa(isa_bus
);
1269 fprintf(stderr
, "PCI bus not available for %s\n", c
->name
);
1272 c
->init
.init_pci(pci_bus
);
1278 int qemu_uuid_parse(const char *str
, uint8_t *uuid
)
1282 if (strlen(str
) != 36) {
1286 ret
= sscanf(str
, UUID_FMT
, &uuid
[0], &uuid
[1], &uuid
[2], &uuid
[3],
1287 &uuid
[4], &uuid
[5], &uuid
[6], &uuid
[7], &uuid
[8], &uuid
[9],
1288 &uuid
[10], &uuid
[11], &uuid
[12], &uuid
[13], &uuid
[14],
1297 void do_acpitable_option(const QemuOpts
*opts
)
1302 acpi_table_add(opts
, &err
);
1304 error_report("Wrong acpi table provided: %s",
1305 error_get_pretty(err
));
1312 void do_smbios_option(QemuOpts
*opts
)
1315 smbios_entry_add(opts
);
1319 void cpudef_init(void)
1321 #if defined(cpudef_setup)
1322 cpudef_setup(); /* parse cpu definitions in target config file */
1326 int tcg_available(void)
1331 int kvm_available(void)
1340 int xen_available(void)
1350 TargetInfo
*qmp_query_target(Error
**errp
)
1352 TargetInfo
*info
= g_malloc0(sizeof(*info
));
1354 info
->arch
= g_strdup(TARGET_NAME
);
1359 /* Stub function that's gets run on the vcpu when its brought out of the
1360 VM to run inside qemu via async_run_on_cpu()*/
1361 static void mig_sleep_cpu(void *opq
)
1363 qemu_mutex_unlock_iothread();
1365 qemu_mutex_lock_iothread();
1368 /* To reduce the dirty rate explicitly disallow the VCPUs from spending
1369 much time in the VM. The migration thread will try to catchup.
1370 Workload will experience a performance drop.
1372 static void mig_throttle_guest_down(void)
1376 qemu_mutex_lock_iothread();
1378 async_run_on_cpu(cpu
, mig_sleep_cpu
, NULL
);
1380 qemu_mutex_unlock_iothread();
1383 static void check_guest_throttling(void)
1388 if (!mig_throttle_on
) {
1393 t0
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
1397 t1
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
1399 /* If it has been more than 40 ms since the last time the guest
1400 * was throttled then do it again.
1402 if (40 < (t1
-t0
)/1000000) {
1403 mig_throttle_guest_down();