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
107 #elif defined(TARGET_TRICORE)
108 #define QEMU_ARCH QEMU_ARCH_TRICORE
111 const uint32_t arch_type
= QEMU_ARCH
;
112 static bool mig_throttle_on
;
113 static int dirty_rate_high_cnt
;
114 static void check_guest_throttling(void);
116 static uint64_t bitmap_sync_count
;
118 /***********************************************************/
119 /* ram save/restore */
121 #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */
122 #define RAM_SAVE_FLAG_COMPRESS 0x02
123 #define RAM_SAVE_FLAG_MEM_SIZE 0x04
124 #define RAM_SAVE_FLAG_PAGE 0x08
125 #define RAM_SAVE_FLAG_EOS 0x10
126 #define RAM_SAVE_FLAG_CONTINUE 0x20
127 #define RAM_SAVE_FLAG_XBZRLE 0x40
128 /* 0x80 is reserved in migration.h start with 0x100 next */
130 static struct defconfig_file
{
131 const char *filename
;
132 /* Indicates it is an user config file (disabled by -no-user-config) */
134 } default_config_files
[] = {
135 { CONFIG_QEMU_CONFDIR
"/qemu.conf", true },
136 { CONFIG_QEMU_CONFDIR
"/target-" TARGET_NAME
".conf", true },
137 { NULL
}, /* end of list */
140 static const uint8_t ZERO_TARGET_PAGE
[TARGET_PAGE_SIZE
];
142 int qemu_read_default_config_files(bool userconfig
)
145 struct defconfig_file
*f
;
147 for (f
= default_config_files
; f
->filename
; f
++) {
148 if (!userconfig
&& f
->userconfig
) {
151 ret
= qemu_read_config_file(f
->filename
);
152 if (ret
< 0 && ret
!= -ENOENT
) {
160 static inline bool is_zero_range(uint8_t *p
, uint64_t size
)
162 return buffer_find_nonzero_offset(p
, size
) == size
;
165 /* struct contains XBZRLE cache and a static page
166 used by the compression */
168 /* buffer used for XBZRLE encoding */
169 uint8_t *encoded_buf
;
170 /* buffer for storing page content */
171 uint8_t *current_buf
;
172 /* Cache for XBZRLE, Protected by lock. */
177 /* buffer used for XBZRLE decoding */
178 static uint8_t *xbzrle_decoded_buf
;
180 static void XBZRLE_cache_lock(void)
182 if (migrate_use_xbzrle())
183 qemu_mutex_lock(&XBZRLE
.lock
);
186 static void XBZRLE_cache_unlock(void)
188 if (migrate_use_xbzrle())
189 qemu_mutex_unlock(&XBZRLE
.lock
);
193 * called from qmp_migrate_set_cache_size in main thread, possibly while
194 * a migration is in progress.
195 * A running migration maybe using the cache and might finish during this
196 * call, hence changes to the cache are protected by XBZRLE.lock().
198 int64_t xbzrle_cache_resize(int64_t new_size
)
200 PageCache
*new_cache
;
203 if (new_size
< TARGET_PAGE_SIZE
) {
209 if (XBZRLE
.cache
!= NULL
) {
210 if (pow2floor(new_size
) == migrate_xbzrle_cache_size()) {
213 new_cache
= cache_init(new_size
/ TARGET_PAGE_SIZE
,
216 error_report("Error creating cache");
221 cache_fini(XBZRLE
.cache
);
222 XBZRLE
.cache
= new_cache
;
226 ret
= pow2floor(new_size
);
228 XBZRLE_cache_unlock();
232 /* accounting for migration statistics */
233 typedef struct AccountingInfo
{
235 uint64_t skipped_pages
;
238 uint64_t xbzrle_bytes
;
239 uint64_t xbzrle_pages
;
240 uint64_t xbzrle_cache_miss
;
241 double xbzrle_cache_miss_rate
;
242 uint64_t xbzrle_overflows
;
245 static AccountingInfo acct_info
;
247 static void acct_clear(void)
249 memset(&acct_info
, 0, sizeof(acct_info
));
252 uint64_t dup_mig_bytes_transferred(void)
254 return acct_info
.dup_pages
* TARGET_PAGE_SIZE
;
257 uint64_t dup_mig_pages_transferred(void)
259 return acct_info
.dup_pages
;
262 uint64_t skipped_mig_bytes_transferred(void)
264 return acct_info
.skipped_pages
* TARGET_PAGE_SIZE
;
267 uint64_t skipped_mig_pages_transferred(void)
269 return acct_info
.skipped_pages
;
272 uint64_t norm_mig_bytes_transferred(void)
274 return acct_info
.norm_pages
* TARGET_PAGE_SIZE
;
277 uint64_t norm_mig_pages_transferred(void)
279 return acct_info
.norm_pages
;
282 uint64_t xbzrle_mig_bytes_transferred(void)
284 return acct_info
.xbzrle_bytes
;
287 uint64_t xbzrle_mig_pages_transferred(void)
289 return acct_info
.xbzrle_pages
;
292 uint64_t xbzrle_mig_pages_cache_miss(void)
294 return acct_info
.xbzrle_cache_miss
;
297 double xbzrle_mig_cache_miss_rate(void)
299 return acct_info
.xbzrle_cache_miss_rate
;
302 uint64_t xbzrle_mig_pages_overflow(void)
304 return acct_info
.xbzrle_overflows
;
307 static size_t save_block_hdr(QEMUFile
*f
, RAMBlock
*block
, ram_addr_t offset
,
312 qemu_put_be64(f
, offset
| cont
| flag
);
316 qemu_put_byte(f
, strlen(block
->idstr
));
317 qemu_put_buffer(f
, (uint8_t *)block
->idstr
,
318 strlen(block
->idstr
));
319 size
+= 1 + strlen(block
->idstr
);
324 /* This is the last block that we have visited serching for dirty pages
326 static RAMBlock
*last_seen_block
;
327 /* This is the last block from where we have sent data */
328 static RAMBlock
*last_sent_block
;
329 static ram_addr_t last_offset
;
330 static unsigned long *migration_bitmap
;
331 static uint64_t migration_dirty_pages
;
332 static uint32_t last_version
;
333 static bool ram_bulk_stage
;
335 /* Update the xbzrle cache to reflect a page that's been sent as all 0.
336 * The important thing is that a stale (not-yet-0'd) page be replaced
338 * As a bonus, if the page wasn't in the cache it gets added so that
339 * when a small write is made into the 0'd page it gets XBZRLE sent
341 static void xbzrle_cache_zero_page(ram_addr_t current_addr
)
343 if (ram_bulk_stage
|| !migrate_use_xbzrle()) {
347 /* We don't care if this fails to allocate a new cache page
348 * as long as it updated an old one */
349 cache_insert(XBZRLE
.cache
, current_addr
, ZERO_TARGET_PAGE
,
353 #define ENCODING_FLAG_XBZRLE 0x1
355 static int save_xbzrle_page(QEMUFile
*f
, uint8_t **current_data
,
356 ram_addr_t current_addr
, RAMBlock
*block
,
357 ram_addr_t offset
, int cont
, bool last_stage
)
359 int encoded_len
= 0, bytes_sent
= -1;
360 uint8_t *prev_cached_page
;
362 if (!cache_is_cached(XBZRLE
.cache
, current_addr
, bitmap_sync_count
)) {
363 acct_info
.xbzrle_cache_miss
++;
365 if (cache_insert(XBZRLE
.cache
, current_addr
, *current_data
,
366 bitmap_sync_count
) == -1) {
369 /* update *current_data when the page has been
370 inserted into cache */
371 *current_data
= get_cached_data(XBZRLE
.cache
, current_addr
);
377 prev_cached_page
= get_cached_data(XBZRLE
.cache
, current_addr
);
379 /* save current buffer into memory */
380 memcpy(XBZRLE
.current_buf
, *current_data
, TARGET_PAGE_SIZE
);
382 /* XBZRLE encoding (if there is no overflow) */
383 encoded_len
= xbzrle_encode_buffer(prev_cached_page
, XBZRLE
.current_buf
,
384 TARGET_PAGE_SIZE
, XBZRLE
.encoded_buf
,
386 if (encoded_len
== 0) {
387 DPRINTF("Skipping unmodified page\n");
389 } else if (encoded_len
== -1) {
390 DPRINTF("Overflow\n");
391 acct_info
.xbzrle_overflows
++;
392 /* update data in the cache */
394 memcpy(prev_cached_page
, *current_data
, TARGET_PAGE_SIZE
);
395 *current_data
= prev_cached_page
;
400 /* we need to update the data in the cache, in order to get the same data */
402 memcpy(prev_cached_page
, XBZRLE
.current_buf
, TARGET_PAGE_SIZE
);
405 /* Send XBZRLE based compressed page */
406 bytes_sent
= save_block_hdr(f
, block
, offset
, cont
, RAM_SAVE_FLAG_XBZRLE
);
407 qemu_put_byte(f
, ENCODING_FLAG_XBZRLE
);
408 qemu_put_be16(f
, encoded_len
);
409 qemu_put_buffer(f
, XBZRLE
.encoded_buf
, encoded_len
);
410 bytes_sent
+= encoded_len
+ 1 + 2;
411 acct_info
.xbzrle_pages
++;
412 acct_info
.xbzrle_bytes
+= bytes_sent
;
418 ram_addr_t
migration_bitmap_find_and_reset_dirty(MemoryRegion
*mr
,
421 unsigned long base
= mr
->ram_addr
>> TARGET_PAGE_BITS
;
422 unsigned long nr
= base
+ (start
>> TARGET_PAGE_BITS
);
423 uint64_t mr_size
= TARGET_PAGE_ALIGN(memory_region_size(mr
));
424 unsigned long size
= base
+ (mr_size
>> TARGET_PAGE_BITS
);
428 if (ram_bulk_stage
&& nr
> base
) {
431 next
= find_next_bit(migration_bitmap
, size
, nr
);
435 clear_bit(next
, migration_bitmap
);
436 migration_dirty_pages
--;
438 return (next
- base
) << TARGET_PAGE_BITS
;
441 static inline bool migration_bitmap_set_dirty(ram_addr_t addr
)
444 int nr
= addr
>> TARGET_PAGE_BITS
;
446 ret
= test_and_set_bit(nr
, migration_bitmap
);
449 migration_dirty_pages
++;
454 static void migration_bitmap_sync_range(ram_addr_t start
, ram_addr_t length
)
457 unsigned long page
= BIT_WORD(start
>> TARGET_PAGE_BITS
);
459 /* start address is aligned at the start of a word? */
460 if (((page
* BITS_PER_LONG
) << TARGET_PAGE_BITS
) == start
) {
462 int nr
= BITS_TO_LONGS(length
>> TARGET_PAGE_BITS
);
463 unsigned long *src
= ram_list
.dirty_memory
[DIRTY_MEMORY_MIGRATION
];
465 for (k
= page
; k
< page
+ nr
; k
++) {
467 unsigned long new_dirty
;
468 new_dirty
= ~migration_bitmap
[k
];
469 migration_bitmap
[k
] |= src
[k
];
471 migration_dirty_pages
+= ctpopl(new_dirty
);
476 for (addr
= 0; addr
< length
; addr
+= TARGET_PAGE_SIZE
) {
477 if (cpu_physical_memory_get_dirty(start
+ addr
,
479 DIRTY_MEMORY_MIGRATION
)) {
480 cpu_physical_memory_reset_dirty(start
+ addr
,
482 DIRTY_MEMORY_MIGRATION
);
483 migration_bitmap_set_dirty(start
+ addr
);
490 /* Needs iothread lock! */
491 /* Fix me: there are too many global variables used in migration process. */
492 static int64_t start_time
;
493 static int64_t bytes_xfer_prev
;
494 static int64_t num_dirty_pages_period
;
496 static void migration_bitmap_sync_init(void)
500 num_dirty_pages_period
= 0;
503 static void migration_bitmap_sync(void)
506 uint64_t num_dirty_pages_init
= migration_dirty_pages
;
507 MigrationState
*s
= migrate_get_current();
509 int64_t bytes_xfer_now
;
510 static uint64_t xbzrle_cache_miss_prev
;
511 static uint64_t iterations_prev
;
515 if (!bytes_xfer_prev
) {
516 bytes_xfer_prev
= ram_bytes_transferred();
520 start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
523 trace_migration_bitmap_sync_start();
524 address_space_sync_dirty_bitmap(&address_space_memory
);
526 QTAILQ_FOREACH(block
, &ram_list
.blocks
, next
) {
527 migration_bitmap_sync_range(block
->mr
->ram_addr
, block
->used_length
);
529 trace_migration_bitmap_sync_end(migration_dirty_pages
530 - num_dirty_pages_init
);
531 num_dirty_pages_period
+= migration_dirty_pages
- num_dirty_pages_init
;
532 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
534 /* more than 1 second = 1000 millisecons */
535 if (end_time
> start_time
+ 1000) {
536 if (migrate_auto_converge()) {
537 /* The following detection logic can be refined later. For now:
538 Check to see if the dirtied bytes is 50% more than the approx.
539 amount of bytes that just got transferred since the last time we
540 were in this routine. If that happens >N times (for now N==4)
541 we turn on the throttle down logic */
542 bytes_xfer_now
= ram_bytes_transferred();
543 if (s
->dirty_pages_rate
&&
544 (num_dirty_pages_period
* TARGET_PAGE_SIZE
>
545 (bytes_xfer_now
- bytes_xfer_prev
)/2) &&
546 (dirty_rate_high_cnt
++ > 4)) {
547 trace_migration_throttle();
548 mig_throttle_on
= true;
549 dirty_rate_high_cnt
= 0;
551 bytes_xfer_prev
= bytes_xfer_now
;
553 mig_throttle_on
= false;
555 if (migrate_use_xbzrle()) {
556 if (iterations_prev
!= 0) {
557 acct_info
.xbzrle_cache_miss_rate
=
558 (double)(acct_info
.xbzrle_cache_miss
-
559 xbzrle_cache_miss_prev
) /
560 (acct_info
.iterations
- iterations_prev
);
562 iterations_prev
= acct_info
.iterations
;
563 xbzrle_cache_miss_prev
= acct_info
.xbzrle_cache_miss
;
565 s
->dirty_pages_rate
= num_dirty_pages_period
* 1000
566 / (end_time
- start_time
);
567 s
->dirty_bytes_rate
= s
->dirty_pages_rate
* TARGET_PAGE_SIZE
;
568 start_time
= end_time
;
569 num_dirty_pages_period
= 0;
570 s
->dirty_sync_count
= bitmap_sync_count
;
575 * ram_save_page: Send the given page to the stream
577 * Returns: Number of bytes written.
579 static int ram_save_page(QEMUFile
*f
, RAMBlock
* block
, ram_addr_t offset
,
584 ram_addr_t current_addr
;
585 MemoryRegion
*mr
= block
->mr
;
588 bool send_async
= true;
590 cont
= (block
== last_sent_block
) ? RAM_SAVE_FLAG_CONTINUE
: 0;
592 p
= memory_region_get_ram_ptr(mr
) + offset
;
594 /* In doubt sent page as normal */
596 ret
= ram_control_save_page(f
, block
->offset
,
597 offset
, TARGET_PAGE_SIZE
, &bytes_sent
);
601 current_addr
= block
->offset
+ offset
;
602 if (ret
!= RAM_SAVE_CONTROL_NOT_SUPP
) {
603 if (ret
!= RAM_SAVE_CONTROL_DELAYED
) {
604 if (bytes_sent
> 0) {
605 acct_info
.norm_pages
++;
606 } else if (bytes_sent
== 0) {
607 acct_info
.dup_pages
++;
610 } else if (is_zero_range(p
, TARGET_PAGE_SIZE
)) {
611 acct_info
.dup_pages
++;
612 bytes_sent
= save_block_hdr(f
, block
, offset
, cont
,
613 RAM_SAVE_FLAG_COMPRESS
);
616 /* Must let xbzrle know, otherwise a previous (now 0'd) cached
617 * page would be stale
619 xbzrle_cache_zero_page(current_addr
);
620 } else if (!ram_bulk_stage
&& migrate_use_xbzrle()) {
621 bytes_sent
= save_xbzrle_page(f
, &p
, current_addr
, block
,
622 offset
, cont
, last_stage
);
624 /* Can't send this cached data async, since the cache page
625 * might get updated before it gets to the wire
631 /* XBZRLE overflow or normal page */
632 if (bytes_sent
== -1) {
633 bytes_sent
= save_block_hdr(f
, block
, offset
, cont
, RAM_SAVE_FLAG_PAGE
);
635 qemu_put_buffer_async(f
, p
, TARGET_PAGE_SIZE
);
637 qemu_put_buffer(f
, p
, TARGET_PAGE_SIZE
);
639 bytes_sent
+= TARGET_PAGE_SIZE
;
640 acct_info
.norm_pages
++;
643 XBZRLE_cache_unlock();
649 * ram_find_and_save_block: Finds a page to send and sends it to f
651 * Returns: The number of bytes written.
652 * 0 means no dirty pages
655 static int ram_find_and_save_block(QEMUFile
*f
, bool last_stage
)
657 RAMBlock
*block
= last_seen_block
;
658 ram_addr_t offset
= last_offset
;
659 bool complete_round
= false;
664 block
= QTAILQ_FIRST(&ram_list
.blocks
);
668 offset
= migration_bitmap_find_and_reset_dirty(mr
, offset
);
669 if (complete_round
&& block
== last_seen_block
&&
670 offset
>= last_offset
) {
673 if (offset
>= block
->used_length
) {
675 block
= QTAILQ_NEXT(block
, next
);
677 block
= QTAILQ_FIRST(&ram_list
.blocks
);
678 complete_round
= true;
679 ram_bulk_stage
= false;
682 bytes_sent
= ram_save_page(f
, block
, offset
, last_stage
);
684 /* if page is unmodified, continue to the next */
685 if (bytes_sent
> 0) {
686 last_sent_block
= block
;
691 last_seen_block
= block
;
692 last_offset
= offset
;
697 static uint64_t bytes_transferred
;
699 void acct_update_position(QEMUFile
*f
, size_t size
, bool zero
)
701 uint64_t pages
= size
/ TARGET_PAGE_SIZE
;
703 acct_info
.dup_pages
+= pages
;
705 acct_info
.norm_pages
+= pages
;
706 bytes_transferred
+= size
;
707 qemu_update_position(f
, size
);
711 static ram_addr_t
ram_save_remaining(void)
713 return migration_dirty_pages
;
716 uint64_t ram_bytes_remaining(void)
718 return ram_save_remaining() * TARGET_PAGE_SIZE
;
721 uint64_t ram_bytes_transferred(void)
723 return bytes_transferred
;
726 uint64_t ram_bytes_total(void)
731 QTAILQ_FOREACH(block
, &ram_list
.blocks
, next
)
732 total
+= block
->used_length
;
737 void free_xbzrle_decoded_buf(void)
739 g_free(xbzrle_decoded_buf
);
740 xbzrle_decoded_buf
= NULL
;
743 static void migration_end(void)
745 if (migration_bitmap
) {
746 memory_global_dirty_log_stop();
747 g_free(migration_bitmap
);
748 migration_bitmap
= NULL
;
753 cache_fini(XBZRLE
.cache
);
754 g_free(XBZRLE
.encoded_buf
);
755 g_free(XBZRLE
.current_buf
);
757 XBZRLE
.encoded_buf
= NULL
;
758 XBZRLE
.current_buf
= NULL
;
760 XBZRLE_cache_unlock();
763 static void ram_migration_cancel(void *opaque
)
768 static void reset_ram_globals(void)
770 last_seen_block
= NULL
;
771 last_sent_block
= NULL
;
773 last_version
= ram_list
.version
;
774 ram_bulk_stage
= true;
777 #define MAX_WAIT 50 /* ms, half buffered_file limit */
779 static int ram_save_setup(QEMUFile
*f
, void *opaque
)
782 int64_t ram_bitmap_pages
; /* Size of bitmap in pages, including gaps */
784 mig_throttle_on
= false;
785 dirty_rate_high_cnt
= 0;
786 bitmap_sync_count
= 0;
787 migration_bitmap_sync_init();
789 if (migrate_use_xbzrle()) {
791 XBZRLE
.cache
= cache_init(migrate_xbzrle_cache_size() /
795 XBZRLE_cache_unlock();
796 error_report("Error creating cache");
799 XBZRLE_cache_unlock();
801 /* We prefer not to abort if there is no memory */
802 XBZRLE
.encoded_buf
= g_try_malloc0(TARGET_PAGE_SIZE
);
803 if (!XBZRLE
.encoded_buf
) {
804 error_report("Error allocating encoded_buf");
808 XBZRLE
.current_buf
= g_try_malloc(TARGET_PAGE_SIZE
);
809 if (!XBZRLE
.current_buf
) {
810 error_report("Error allocating current_buf");
811 g_free(XBZRLE
.encoded_buf
);
812 XBZRLE
.encoded_buf
= NULL
;
819 qemu_mutex_lock_iothread();
820 qemu_mutex_lock_ramlist();
821 bytes_transferred
= 0;
824 ram_bitmap_pages
= last_ram_offset() >> TARGET_PAGE_BITS
;
825 migration_bitmap
= bitmap_new(ram_bitmap_pages
);
826 bitmap_set(migration_bitmap
, 0, ram_bitmap_pages
);
829 * Count the total number of pages used by ram blocks not including any
830 * gaps due to alignment or unplugs.
832 migration_dirty_pages
= 0;
833 QTAILQ_FOREACH(block
, &ram_list
.blocks
, next
) {
834 uint64_t block_pages
;
836 block_pages
= block
->used_length
>> TARGET_PAGE_BITS
;
837 migration_dirty_pages
+= block_pages
;
840 memory_global_dirty_log_start();
841 migration_bitmap_sync();
842 qemu_mutex_unlock_iothread();
844 qemu_put_be64(f
, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE
);
846 QTAILQ_FOREACH(block
, &ram_list
.blocks
, next
) {
847 qemu_put_byte(f
, strlen(block
->idstr
));
848 qemu_put_buffer(f
, (uint8_t *)block
->idstr
, strlen(block
->idstr
));
849 qemu_put_be64(f
, block
->used_length
);
852 qemu_mutex_unlock_ramlist();
854 ram_control_before_iterate(f
, RAM_CONTROL_SETUP
);
855 ram_control_after_iterate(f
, RAM_CONTROL_SETUP
);
857 qemu_put_be64(f
, RAM_SAVE_FLAG_EOS
);
862 static int ram_save_iterate(QEMUFile
*f
, void *opaque
)
869 qemu_mutex_lock_ramlist();
871 if (ram_list
.version
!= last_version
) {
875 ram_control_before_iterate(f
, RAM_CONTROL_ROUND
);
877 t0
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
879 while ((ret
= qemu_file_rate_limit(f
)) == 0) {
882 bytes_sent
= ram_find_and_save_block(f
, false);
883 /* no more blocks to sent */
884 if (bytes_sent
== 0) {
887 total_sent
+= bytes_sent
;
888 acct_info
.iterations
++;
889 check_guest_throttling();
890 /* we want to check in the 1st loop, just in case it was the 1st time
891 and we had to sync the dirty bitmap.
892 qemu_get_clock_ns() is a bit expensive, so we only check each some
896 uint64_t t1
= (qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - t0
) / 1000000;
898 DPRINTF("big wait: %" PRIu64
" milliseconds, %d iterations\n",
906 qemu_mutex_unlock_ramlist();
909 * Must occur before EOS (or any QEMUFile operation)
910 * because of RDMA protocol.
912 ram_control_after_iterate(f
, RAM_CONTROL_ROUND
);
914 bytes_transferred
+= total_sent
;
917 * Do not count these 8 bytes into total_sent, so that we can
918 * return 0 if no page had been dirtied.
920 qemu_put_be64(f
, RAM_SAVE_FLAG_EOS
);
921 bytes_transferred
+= 8;
923 ret
= qemu_file_get_error(f
);
931 static int ram_save_complete(QEMUFile
*f
, void *opaque
)
933 qemu_mutex_lock_ramlist();
934 migration_bitmap_sync();
936 ram_control_before_iterate(f
, RAM_CONTROL_FINISH
);
938 /* try transferring iterative blocks of memory */
940 /* flush all remaining blocks regardless of rate limiting */
944 bytes_sent
= ram_find_and_save_block(f
, true);
945 /* no more blocks to sent */
946 if (bytes_sent
== 0) {
949 bytes_transferred
+= bytes_sent
;
952 ram_control_after_iterate(f
, RAM_CONTROL_FINISH
);
955 qemu_mutex_unlock_ramlist();
956 qemu_put_be64(f
, RAM_SAVE_FLAG_EOS
);
961 static uint64_t ram_save_pending(QEMUFile
*f
, void *opaque
, uint64_t max_size
)
963 uint64_t remaining_size
;
965 remaining_size
= ram_save_remaining() * TARGET_PAGE_SIZE
;
967 if (remaining_size
< max_size
) {
968 qemu_mutex_lock_iothread();
969 migration_bitmap_sync();
970 qemu_mutex_unlock_iothread();
971 remaining_size
= ram_save_remaining() * TARGET_PAGE_SIZE
;
973 return remaining_size
;
976 static int load_xbzrle(QEMUFile
*f
, ram_addr_t addr
, void *host
)
981 if (!xbzrle_decoded_buf
) {
982 xbzrle_decoded_buf
= g_malloc(TARGET_PAGE_SIZE
);
985 /* extract RLE header */
986 xh_flags
= qemu_get_byte(f
);
987 xh_len
= qemu_get_be16(f
);
989 if (xh_flags
!= ENCODING_FLAG_XBZRLE
) {
990 error_report("Failed to load XBZRLE page - wrong compression!");
994 if (xh_len
> TARGET_PAGE_SIZE
) {
995 error_report("Failed to load XBZRLE page - len overflow!");
998 /* load data and decode */
999 qemu_get_buffer(f
, xbzrle_decoded_buf
, xh_len
);
1002 if (xbzrle_decode_buffer(xbzrle_decoded_buf
, xh_len
, host
,
1003 TARGET_PAGE_SIZE
) == -1) {
1004 error_report("Failed to load XBZRLE page - decode error!");
1011 static inline void *host_from_stream_offset(QEMUFile
*f
,
1015 static RAMBlock
*block
= NULL
;
1019 if (flags
& RAM_SAVE_FLAG_CONTINUE
) {
1020 if (!block
|| block
->max_length
<= offset
) {
1021 error_report("Ack, bad migration stream!");
1025 return memory_region_get_ram_ptr(block
->mr
) + offset
;
1028 len
= qemu_get_byte(f
);
1029 qemu_get_buffer(f
, (uint8_t *)id
, len
);
1032 QTAILQ_FOREACH(block
, &ram_list
.blocks
, next
) {
1033 if (!strncmp(id
, block
->idstr
, sizeof(id
)) &&
1034 block
->max_length
> offset
) {
1035 return memory_region_get_ram_ptr(block
->mr
) + offset
;
1039 error_report("Can't find block %s!", id
);
1044 * If a page (or a whole RDMA chunk) has been
1045 * determined to be zero, then zap it.
1047 void ram_handle_compressed(void *host
, uint8_t ch
, uint64_t size
)
1049 if (ch
!= 0 || !is_zero_range(host
, size
)) {
1050 memset(host
, ch
, size
);
1054 static int ram_load(QEMUFile
*f
, void *opaque
, int version_id
)
1056 int flags
= 0, ret
= 0;
1057 static uint64_t seq_iter
;
1061 if (version_id
!= 4) {
1065 while (!ret
&& !(flags
& RAM_SAVE_FLAG_EOS
)) {
1066 ram_addr_t addr
, total_ram_bytes
;
1070 addr
= qemu_get_be64(f
);
1071 flags
= addr
& ~TARGET_PAGE_MASK
;
1072 addr
&= TARGET_PAGE_MASK
;
1074 switch (flags
& ~RAM_SAVE_FLAG_CONTINUE
) {
1075 case RAM_SAVE_FLAG_MEM_SIZE
:
1076 /* Synchronize RAM block list */
1077 total_ram_bytes
= addr
;
1078 while (!ret
&& total_ram_bytes
) {
1084 len
= qemu_get_byte(f
);
1085 qemu_get_buffer(f
, (uint8_t *)id
, len
);
1087 length
= qemu_get_be64(f
);
1089 QTAILQ_FOREACH(block
, &ram_list
.blocks
, next
) {
1090 if (!strncmp(id
, block
->idstr
, sizeof(id
))) {
1091 if (length
!= block
->used_length
) {
1092 Error
*local_err
= NULL
;
1094 ret
= qemu_ram_resize(block
->offset
, length
, &local_err
);
1096 error_report("%s", error_get_pretty(local_err
));
1097 error_free(local_err
);
1105 error_report("Unknown ramblock \"%s\", cannot "
1106 "accept migration", id
);
1110 total_ram_bytes
-= length
;
1113 case RAM_SAVE_FLAG_COMPRESS
:
1114 host
= host_from_stream_offset(f
, addr
, flags
);
1116 error_report("Illegal RAM offset " RAM_ADDR_FMT
, addr
);
1121 ch
= qemu_get_byte(f
);
1122 ram_handle_compressed(host
, ch
, TARGET_PAGE_SIZE
);
1124 case RAM_SAVE_FLAG_PAGE
:
1125 host
= host_from_stream_offset(f
, addr
, flags
);
1127 error_report("Illegal RAM offset " RAM_ADDR_FMT
, addr
);
1132 qemu_get_buffer(f
, host
, TARGET_PAGE_SIZE
);
1134 case RAM_SAVE_FLAG_XBZRLE
:
1135 host
= host_from_stream_offset(f
, addr
, flags
);
1137 error_report("Illegal RAM offset " RAM_ADDR_FMT
, addr
);
1142 if (load_xbzrle(f
, addr
, host
) < 0) {
1143 error_report("Failed to decompress XBZRLE page at "
1144 RAM_ADDR_FMT
, addr
);
1149 case RAM_SAVE_FLAG_EOS
:
1153 if (flags
& RAM_SAVE_FLAG_HOOK
) {
1154 ram_control_load_hook(f
, flags
);
1156 error_report("Unknown combination of migration flags: %#x",
1162 ret
= qemu_file_get_error(f
);
1166 DPRINTF("Completed load of VM with exit code %d seq iteration "
1167 "%" PRIu64
"\n", ret
, seq_iter
);
1171 static SaveVMHandlers savevm_ram_handlers
= {
1172 .save_live_setup
= ram_save_setup
,
1173 .save_live_iterate
= ram_save_iterate
,
1174 .save_live_complete
= ram_save_complete
,
1175 .save_live_pending
= ram_save_pending
,
1176 .load_state
= ram_load
,
1177 .cancel
= ram_migration_cancel
,
1180 void ram_mig_init(void)
1182 qemu_mutex_init(&XBZRLE
.lock
);
1183 register_savevm_live(NULL
, "ram", 0, 4, &savevm_ram_handlers
, NULL
);
1192 int (*init_isa
) (ISABus
*bus
);
1193 int (*init_pci
) (PCIBus
*bus
);
1197 static struct soundhw soundhw
[9];
1198 static int soundhw_count
;
1200 void isa_register_soundhw(const char *name
, const char *descr
,
1201 int (*init_isa
)(ISABus
*bus
))
1203 assert(soundhw_count
< ARRAY_SIZE(soundhw
) - 1);
1204 soundhw
[soundhw_count
].name
= name
;
1205 soundhw
[soundhw_count
].descr
= descr
;
1206 soundhw
[soundhw_count
].isa
= 1;
1207 soundhw
[soundhw_count
].init
.init_isa
= init_isa
;
1211 void pci_register_soundhw(const char *name
, const char *descr
,
1212 int (*init_pci
)(PCIBus
*bus
))
1214 assert(soundhw_count
< ARRAY_SIZE(soundhw
) - 1);
1215 soundhw
[soundhw_count
].name
= name
;
1216 soundhw
[soundhw_count
].descr
= descr
;
1217 soundhw
[soundhw_count
].isa
= 0;
1218 soundhw
[soundhw_count
].init
.init_pci
= init_pci
;
1222 void select_soundhw(const char *optarg
)
1226 if (is_help_option(optarg
)) {
1229 if (soundhw_count
) {
1230 printf("Valid sound card names (comma separated):\n");
1231 for (c
= soundhw
; c
->name
; ++c
) {
1232 printf ("%-11s %s\n", c
->name
, c
->descr
);
1234 printf("\n-soundhw all will enable all of the above\n");
1236 printf("Machine has no user-selectable audio hardware "
1237 "(it may or may not have always-present audio hardware).\n");
1239 exit(!is_help_option(optarg
));
1247 if (!strcmp(optarg
, "all")) {
1248 for (c
= soundhw
; c
->name
; ++c
) {
1257 l
= !e
? strlen(p
) : (size_t) (e
- p
);
1259 for (c
= soundhw
; c
->name
; ++c
) {
1260 if (!strncmp(c
->name
, p
, l
) && !c
->name
[l
]) {
1268 error_report("Unknown sound card name (too big to show)");
1271 error_report("Unknown sound card name `%.*s'",
1276 p
+= l
+ (e
!= NULL
);
1280 goto show_valid_cards
;
1285 void audio_init(void)
1288 ISABus
*isa_bus
= (ISABus
*) object_resolve_path_type("", TYPE_ISA_BUS
, NULL
);
1289 PCIBus
*pci_bus
= (PCIBus
*) object_resolve_path_type("", TYPE_PCI_BUS
, NULL
);
1291 for (c
= soundhw
; c
->name
; ++c
) {
1295 error_report("ISA bus not available for %s", c
->name
);
1298 c
->init
.init_isa(isa_bus
);
1301 error_report("PCI bus not available for %s", c
->name
);
1304 c
->init
.init_pci(pci_bus
);
1310 int qemu_uuid_parse(const char *str
, uint8_t *uuid
)
1314 if (strlen(str
) != 36) {
1318 ret
= sscanf(str
, UUID_FMT
, &uuid
[0], &uuid
[1], &uuid
[2], &uuid
[3],
1319 &uuid
[4], &uuid
[5], &uuid
[6], &uuid
[7], &uuid
[8], &uuid
[9],
1320 &uuid
[10], &uuid
[11], &uuid
[12], &uuid
[13], &uuid
[14],
1329 void do_acpitable_option(const QemuOpts
*opts
)
1334 acpi_table_add(opts
, &err
);
1336 error_report("Wrong acpi table provided: %s",
1337 error_get_pretty(err
));
1344 void do_smbios_option(QemuOpts
*opts
)
1347 smbios_entry_add(opts
);
1351 void cpudef_init(void)
1353 #if defined(cpudef_setup)
1354 cpudef_setup(); /* parse cpu definitions in target config file */
1358 int kvm_available(void)
1367 int xen_available(void)
1377 TargetInfo
*qmp_query_target(Error
**errp
)
1379 TargetInfo
*info
= g_malloc0(sizeof(*info
));
1381 info
->arch
= g_strdup(TARGET_NAME
);
1386 /* Stub function that's gets run on the vcpu when its brought out of the
1387 VM to run inside qemu via async_run_on_cpu()*/
1388 static void mig_sleep_cpu(void *opq
)
1390 qemu_mutex_unlock_iothread();
1392 qemu_mutex_lock_iothread();
1395 /* To reduce the dirty rate explicitly disallow the VCPUs from spending
1396 much time in the VM. The migration thread will try to catchup.
1397 Workload will experience a performance drop.
1399 static void mig_throttle_guest_down(void)
1403 qemu_mutex_lock_iothread();
1405 async_run_on_cpu(cpu
, mig_sleep_cpu
, NULL
);
1407 qemu_mutex_unlock_iothread();
1410 static void check_guest_throttling(void)
1415 if (!mig_throttle_on
) {
1420 t0
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
1424 t1
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
1426 /* If it has been more than 40 ms since the last time the guest
1427 * was throttled then do it again.
1429 if (40 < (t1
-t0
)/1000000) {
1430 mig_throttle_guest_down();