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
29 #include <sys/types.h>
33 #include "monitor/monitor.h"
34 #include "sysemu/sysemu.h"
35 #include "qemu/bitops.h"
36 #include "qemu/bitmap.h"
37 #include "sysemu/arch_init.h"
38 #include "audio/audio.h"
39 #include "hw/i386/pc.h"
40 #include "hw/pci/pci.h"
41 #include "hw/audio/audio.h"
42 #include "sysemu/kvm.h"
43 #include "migration/migration.h"
44 #include "hw/i386/smbios.h"
45 #include "exec/address-spaces.h"
46 #include "hw/audio/pcspk.h"
47 #include "migration/page_cache.h"
48 #include "qemu/config-file.h"
49 #include "qemu/error-report.h"
50 #include "qmp-commands.h"
52 #include "exec/cpu-all.h"
53 #include "exec/ram_addr.h"
54 #include "hw/acpi/acpi.h"
55 #include "qemu/host-utils.h"
56 #include "qemu/rcu_queue.h"
58 #ifdef DEBUG_ARCH_INIT
59 #define DPRINTF(fmt, ...) \
60 do { fprintf(stdout, "arch_init: " fmt, ## __VA_ARGS__); } while (0)
62 #define DPRINTF(fmt, ...) \
67 int graphic_width
= 1024;
68 int graphic_height
= 768;
69 int graphic_depth
= 8;
71 int graphic_width
= 800;
72 int graphic_height
= 600;
73 int graphic_depth
= 32;
77 #if defined(TARGET_ALPHA)
78 #define QEMU_ARCH QEMU_ARCH_ALPHA
79 #elif defined(TARGET_ARM)
80 #define QEMU_ARCH QEMU_ARCH_ARM
81 #elif defined(TARGET_CRIS)
82 #define QEMU_ARCH QEMU_ARCH_CRIS
83 #elif defined(TARGET_I386)
84 #define QEMU_ARCH QEMU_ARCH_I386
85 #elif defined(TARGET_M68K)
86 #define QEMU_ARCH QEMU_ARCH_M68K
87 #elif defined(TARGET_LM32)
88 #define QEMU_ARCH QEMU_ARCH_LM32
89 #elif defined(TARGET_MICROBLAZE)
90 #define QEMU_ARCH QEMU_ARCH_MICROBLAZE
91 #elif defined(TARGET_MIPS)
92 #define QEMU_ARCH QEMU_ARCH_MIPS
93 #elif defined(TARGET_MOXIE)
94 #define QEMU_ARCH QEMU_ARCH_MOXIE
95 #elif defined(TARGET_OPENRISC)
96 #define QEMU_ARCH QEMU_ARCH_OPENRISC
97 #elif defined(TARGET_PPC)
98 #define QEMU_ARCH QEMU_ARCH_PPC
99 #elif defined(TARGET_S390X)
100 #define QEMU_ARCH QEMU_ARCH_S390X
101 #elif defined(TARGET_SH4)
102 #define QEMU_ARCH QEMU_ARCH_SH4
103 #elif defined(TARGET_SPARC)
104 #define QEMU_ARCH QEMU_ARCH_SPARC
105 #elif defined(TARGET_XTENSA)
106 #define QEMU_ARCH QEMU_ARCH_XTENSA
107 #elif defined(TARGET_UNICORE32)
108 #define QEMU_ARCH QEMU_ARCH_UNICORE32
109 #elif defined(TARGET_TRICORE)
110 #define QEMU_ARCH QEMU_ARCH_TRICORE
113 const uint32_t arch_type
= QEMU_ARCH
;
114 static bool mig_throttle_on
;
115 static int dirty_rate_high_cnt
;
116 static void check_guest_throttling(void);
118 static uint64_t bitmap_sync_count
;
120 /***********************************************************/
121 /* ram save/restore */
123 #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */
124 #define RAM_SAVE_FLAG_COMPRESS 0x02
125 #define RAM_SAVE_FLAG_MEM_SIZE 0x04
126 #define RAM_SAVE_FLAG_PAGE 0x08
127 #define RAM_SAVE_FLAG_EOS 0x10
128 #define RAM_SAVE_FLAG_CONTINUE 0x20
129 #define RAM_SAVE_FLAG_XBZRLE 0x40
130 /* 0x80 is reserved in migration.h start with 0x100 next */
131 #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100
133 static struct defconfig_file
{
134 const char *filename
;
135 /* Indicates it is an user config file (disabled by -no-user-config) */
137 } default_config_files
[] = {
138 { CONFIG_QEMU_CONFDIR
"/qemu.conf", true },
139 { CONFIG_QEMU_CONFDIR
"/target-" TARGET_NAME
".conf", true },
140 { NULL
}, /* end of list */
143 static const uint8_t ZERO_TARGET_PAGE
[TARGET_PAGE_SIZE
];
145 int qemu_read_default_config_files(bool userconfig
)
148 struct defconfig_file
*f
;
150 for (f
= default_config_files
; f
->filename
; f
++) {
151 if (!userconfig
&& f
->userconfig
) {
154 ret
= qemu_read_config_file(f
->filename
);
155 if (ret
< 0 && ret
!= -ENOENT
) {
163 static inline bool is_zero_range(uint8_t *p
, uint64_t size
)
165 return buffer_find_nonzero_offset(p
, size
) == size
;
168 /* struct contains XBZRLE cache and a static page
169 used by the compression */
171 /* buffer used for XBZRLE encoding */
172 uint8_t *encoded_buf
;
173 /* buffer for storing page content */
174 uint8_t *current_buf
;
175 /* Cache for XBZRLE, Protected by lock. */
180 /* buffer used for XBZRLE decoding */
181 static uint8_t *xbzrle_decoded_buf
;
183 static void XBZRLE_cache_lock(void)
185 if (migrate_use_xbzrle())
186 qemu_mutex_lock(&XBZRLE
.lock
);
189 static void XBZRLE_cache_unlock(void)
191 if (migrate_use_xbzrle())
192 qemu_mutex_unlock(&XBZRLE
.lock
);
196 * called from qmp_migrate_set_cache_size in main thread, possibly while
197 * a migration is in progress.
198 * A running migration maybe using the cache and might finish during this
199 * call, hence changes to the cache are protected by XBZRLE.lock().
201 int64_t xbzrle_cache_resize(int64_t new_size
)
203 PageCache
*new_cache
;
206 if (new_size
< TARGET_PAGE_SIZE
) {
212 if (XBZRLE
.cache
!= NULL
) {
213 if (pow2floor(new_size
) == migrate_xbzrle_cache_size()) {
216 new_cache
= cache_init(new_size
/ TARGET_PAGE_SIZE
,
219 error_report("Error creating cache");
224 cache_fini(XBZRLE
.cache
);
225 XBZRLE
.cache
= new_cache
;
229 ret
= pow2floor(new_size
);
231 XBZRLE_cache_unlock();
235 /* accounting for migration statistics */
236 typedef struct AccountingInfo
{
238 uint64_t skipped_pages
;
241 uint64_t xbzrle_bytes
;
242 uint64_t xbzrle_pages
;
243 uint64_t xbzrle_cache_miss
;
244 double xbzrle_cache_miss_rate
;
245 uint64_t xbzrle_overflows
;
248 static AccountingInfo acct_info
;
250 static void acct_clear(void)
252 memset(&acct_info
, 0, sizeof(acct_info
));
255 uint64_t dup_mig_bytes_transferred(void)
257 return acct_info
.dup_pages
* TARGET_PAGE_SIZE
;
260 uint64_t dup_mig_pages_transferred(void)
262 return acct_info
.dup_pages
;
265 uint64_t skipped_mig_bytes_transferred(void)
267 return acct_info
.skipped_pages
* TARGET_PAGE_SIZE
;
270 uint64_t skipped_mig_pages_transferred(void)
272 return acct_info
.skipped_pages
;
275 uint64_t norm_mig_bytes_transferred(void)
277 return acct_info
.norm_pages
* TARGET_PAGE_SIZE
;
280 uint64_t norm_mig_pages_transferred(void)
282 return acct_info
.norm_pages
;
285 uint64_t xbzrle_mig_bytes_transferred(void)
287 return acct_info
.xbzrle_bytes
;
290 uint64_t xbzrle_mig_pages_transferred(void)
292 return acct_info
.xbzrle_pages
;
295 uint64_t xbzrle_mig_pages_cache_miss(void)
297 return acct_info
.xbzrle_cache_miss
;
300 double xbzrle_mig_cache_miss_rate(void)
302 return acct_info
.xbzrle_cache_miss_rate
;
305 uint64_t xbzrle_mig_pages_overflow(void)
307 return acct_info
.xbzrle_overflows
;
310 /* This is the last block that we have visited serching for dirty pages
312 static RAMBlock
*last_seen_block
;
313 /* This is the last block from where we have sent data */
314 static RAMBlock
*last_sent_block
;
315 static ram_addr_t last_offset
;
316 static unsigned long *migration_bitmap
;
317 static uint64_t migration_dirty_pages
;
318 static uint32_t last_version
;
319 static bool ram_bulk_stage
;
321 struct CompressParam
{
324 typedef struct CompressParam CompressParam
;
326 struct DecompressParam
{
329 typedef struct DecompressParam DecompressParam
;
331 static CompressParam
*comp_param
;
332 static QemuThread
*compress_threads
;
333 static bool quit_comp_thread
;
334 static bool quit_decomp_thread
;
335 static DecompressParam
*decomp_param
;
336 static QemuThread
*decompress_threads
;
337 static uint8_t *compressed_data_buf
;
339 static void *do_data_compress(void *opaque
)
341 while (!quit_comp_thread
) {
350 static inline void terminate_compression_threads(void)
352 quit_comp_thread
= true;
357 void migrate_compress_threads_join(void)
361 if (!migrate_use_compression()) {
364 terminate_compression_threads();
365 thread_count
= migrate_compress_threads();
366 for (i
= 0; i
< thread_count
; i
++) {
367 qemu_thread_join(compress_threads
+ i
);
369 g_free(compress_threads
);
371 compress_threads
= NULL
;
375 void migrate_compress_threads_create(void)
379 if (!migrate_use_compression()) {
382 quit_comp_thread
= false;
383 thread_count
= migrate_compress_threads();
384 compress_threads
= g_new0(QemuThread
, thread_count
);
385 comp_param
= g_new0(CompressParam
, thread_count
);
386 for (i
= 0; i
< thread_count
; i
++) {
387 qemu_thread_create(compress_threads
+ i
, "compress",
388 do_data_compress
, comp_param
+ i
,
389 QEMU_THREAD_JOINABLE
);
394 * save_page_header: Write page header to wire
396 * If this is the 1st block, it also writes the block identification
398 * Returns: Number of bytes written
400 * @f: QEMUFile where to send the data
401 * @block: block that contains the page we want to send
402 * @offset: offset inside the block for the page
403 * in the lower bits, it contains flags
405 static size_t save_page_header(QEMUFile
*f
, RAMBlock
*block
, ram_addr_t offset
)
409 qemu_put_be64(f
, offset
);
412 if (!(offset
& RAM_SAVE_FLAG_CONTINUE
)) {
413 qemu_put_byte(f
, strlen(block
->idstr
));
414 qemu_put_buffer(f
, (uint8_t *)block
->idstr
,
415 strlen(block
->idstr
));
416 size
+= 1 + strlen(block
->idstr
);
421 /* Update the xbzrle cache to reflect a page that's been sent as all 0.
422 * The important thing is that a stale (not-yet-0'd) page be replaced
424 * As a bonus, if the page wasn't in the cache it gets added so that
425 * when a small write is made into the 0'd page it gets XBZRLE sent
427 static void xbzrle_cache_zero_page(ram_addr_t current_addr
)
429 if (ram_bulk_stage
|| !migrate_use_xbzrle()) {
433 /* We don't care if this fails to allocate a new cache page
434 * as long as it updated an old one */
435 cache_insert(XBZRLE
.cache
, current_addr
, ZERO_TARGET_PAGE
,
439 #define ENCODING_FLAG_XBZRLE 0x1
442 * save_xbzrle_page: compress and send current page
444 * Returns: 1 means that we wrote the page
445 * 0 means that page is identical to the one already sent
446 * -1 means that xbzrle would be longer than normal
448 * @f: QEMUFile where to send the data
451 * @block: block that contains the page we want to send
452 * @offset: offset inside the block for the page
453 * @last_stage: if we are at the completion stage
454 * @bytes_transferred: increase it with the number of transferred bytes
456 static int save_xbzrle_page(QEMUFile
*f
, uint8_t **current_data
,
457 ram_addr_t current_addr
, RAMBlock
*block
,
458 ram_addr_t offset
, bool last_stage
,
459 uint64_t *bytes_transferred
)
461 int encoded_len
= 0, bytes_xbzrle
;
462 uint8_t *prev_cached_page
;
464 if (!cache_is_cached(XBZRLE
.cache
, current_addr
, bitmap_sync_count
)) {
465 acct_info
.xbzrle_cache_miss
++;
467 if (cache_insert(XBZRLE
.cache
, current_addr
, *current_data
,
468 bitmap_sync_count
) == -1) {
471 /* update *current_data when the page has been
472 inserted into cache */
473 *current_data
= get_cached_data(XBZRLE
.cache
, current_addr
);
479 prev_cached_page
= get_cached_data(XBZRLE
.cache
, current_addr
);
481 /* save current buffer into memory */
482 memcpy(XBZRLE
.current_buf
, *current_data
, TARGET_PAGE_SIZE
);
484 /* XBZRLE encoding (if there is no overflow) */
485 encoded_len
= xbzrle_encode_buffer(prev_cached_page
, XBZRLE
.current_buf
,
486 TARGET_PAGE_SIZE
, XBZRLE
.encoded_buf
,
488 if (encoded_len
== 0) {
489 DPRINTF("Skipping unmodified page\n");
491 } else if (encoded_len
== -1) {
492 DPRINTF("Overflow\n");
493 acct_info
.xbzrle_overflows
++;
494 /* update data in the cache */
496 memcpy(prev_cached_page
, *current_data
, TARGET_PAGE_SIZE
);
497 *current_data
= prev_cached_page
;
502 /* we need to update the data in the cache, in order to get the same data */
504 memcpy(prev_cached_page
, XBZRLE
.current_buf
, TARGET_PAGE_SIZE
);
507 /* Send XBZRLE based compressed page */
508 bytes_xbzrle
= save_page_header(f
, block
, offset
| RAM_SAVE_FLAG_XBZRLE
);
509 qemu_put_byte(f
, ENCODING_FLAG_XBZRLE
);
510 qemu_put_be16(f
, encoded_len
);
511 qemu_put_buffer(f
, XBZRLE
.encoded_buf
, encoded_len
);
512 bytes_xbzrle
+= encoded_len
+ 1 + 2;
513 acct_info
.xbzrle_pages
++;
514 acct_info
.xbzrle_bytes
+= bytes_xbzrle
;
515 *bytes_transferred
+= bytes_xbzrle
;
521 ram_addr_t
migration_bitmap_find_and_reset_dirty(MemoryRegion
*mr
,
524 unsigned long base
= mr
->ram_addr
>> TARGET_PAGE_BITS
;
525 unsigned long nr
= base
+ (start
>> TARGET_PAGE_BITS
);
526 uint64_t mr_size
= TARGET_PAGE_ALIGN(memory_region_size(mr
));
527 unsigned long size
= base
+ (mr_size
>> TARGET_PAGE_BITS
);
531 if (ram_bulk_stage
&& nr
> base
) {
534 next
= find_next_bit(migration_bitmap
, size
, nr
);
538 clear_bit(next
, migration_bitmap
);
539 migration_dirty_pages
--;
541 return (next
- base
) << TARGET_PAGE_BITS
;
544 static inline bool migration_bitmap_set_dirty(ram_addr_t addr
)
547 int nr
= addr
>> TARGET_PAGE_BITS
;
549 ret
= test_and_set_bit(nr
, migration_bitmap
);
552 migration_dirty_pages
++;
557 static void migration_bitmap_sync_range(ram_addr_t start
, ram_addr_t length
)
560 unsigned long page
= BIT_WORD(start
>> TARGET_PAGE_BITS
);
562 /* start address is aligned at the start of a word? */
563 if (((page
* BITS_PER_LONG
) << TARGET_PAGE_BITS
) == start
) {
565 int nr
= BITS_TO_LONGS(length
>> TARGET_PAGE_BITS
);
566 unsigned long *src
= ram_list
.dirty_memory
[DIRTY_MEMORY_MIGRATION
];
568 for (k
= page
; k
< page
+ nr
; k
++) {
570 unsigned long new_dirty
;
571 new_dirty
= ~migration_bitmap
[k
];
572 migration_bitmap
[k
] |= src
[k
];
574 migration_dirty_pages
+= ctpopl(new_dirty
);
579 for (addr
= 0; addr
< length
; addr
+= TARGET_PAGE_SIZE
) {
580 if (cpu_physical_memory_get_dirty(start
+ addr
,
582 DIRTY_MEMORY_MIGRATION
)) {
583 cpu_physical_memory_reset_dirty(start
+ addr
,
585 DIRTY_MEMORY_MIGRATION
);
586 migration_bitmap_set_dirty(start
+ addr
);
593 /* Fix me: there are too many global variables used in migration process. */
594 static int64_t start_time
;
595 static int64_t bytes_xfer_prev
;
596 static int64_t num_dirty_pages_period
;
598 static void migration_bitmap_sync_init(void)
602 num_dirty_pages_period
= 0;
605 /* Called with iothread lock held, to protect ram_list.dirty_memory[] */
606 static void migration_bitmap_sync(void)
609 uint64_t num_dirty_pages_init
= migration_dirty_pages
;
610 MigrationState
*s
= migrate_get_current();
612 int64_t bytes_xfer_now
;
613 static uint64_t xbzrle_cache_miss_prev
;
614 static uint64_t iterations_prev
;
618 if (!bytes_xfer_prev
) {
619 bytes_xfer_prev
= ram_bytes_transferred();
623 start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
626 trace_migration_bitmap_sync_start();
627 address_space_sync_dirty_bitmap(&address_space_memory
);
630 QLIST_FOREACH_RCU(block
, &ram_list
.blocks
, next
) {
631 migration_bitmap_sync_range(block
->mr
->ram_addr
, block
->used_length
);
635 trace_migration_bitmap_sync_end(migration_dirty_pages
636 - num_dirty_pages_init
);
637 num_dirty_pages_period
+= migration_dirty_pages
- num_dirty_pages_init
;
638 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
640 /* more than 1 second = 1000 millisecons */
641 if (end_time
> start_time
+ 1000) {
642 if (migrate_auto_converge()) {
643 /* The following detection logic can be refined later. For now:
644 Check to see if the dirtied bytes is 50% more than the approx.
645 amount of bytes that just got transferred since the last time we
646 were in this routine. If that happens >N times (for now N==4)
647 we turn on the throttle down logic */
648 bytes_xfer_now
= ram_bytes_transferred();
649 if (s
->dirty_pages_rate
&&
650 (num_dirty_pages_period
* TARGET_PAGE_SIZE
>
651 (bytes_xfer_now
- bytes_xfer_prev
)/2) &&
652 (dirty_rate_high_cnt
++ > 4)) {
653 trace_migration_throttle();
654 mig_throttle_on
= true;
655 dirty_rate_high_cnt
= 0;
657 bytes_xfer_prev
= bytes_xfer_now
;
659 mig_throttle_on
= false;
661 if (migrate_use_xbzrle()) {
662 if (iterations_prev
!= 0) {
663 acct_info
.xbzrle_cache_miss_rate
=
664 (double)(acct_info
.xbzrle_cache_miss
-
665 xbzrle_cache_miss_prev
) /
666 (acct_info
.iterations
- iterations_prev
);
668 iterations_prev
= acct_info
.iterations
;
669 xbzrle_cache_miss_prev
= acct_info
.xbzrle_cache_miss
;
671 s
->dirty_pages_rate
= num_dirty_pages_period
* 1000
672 / (end_time
- start_time
);
673 s
->dirty_bytes_rate
= s
->dirty_pages_rate
* TARGET_PAGE_SIZE
;
674 start_time
= end_time
;
675 num_dirty_pages_period
= 0;
676 s
->dirty_sync_count
= bitmap_sync_count
;
681 * ram_save_page: Send the given page to the stream
683 * Returns: Number of pages written.
685 * @f: QEMUFile where to send the data
686 * @block: block that contains the page we want to send
687 * @offset: offset inside the block for the page
688 * @last_stage: if we are at the completion stage
689 * @bytes_transferred: increase it with the number of transferred bytes
691 static int ram_save_page(QEMUFile
*f
, RAMBlock
* block
, ram_addr_t offset
,
692 bool last_stage
, uint64_t *bytes_transferred
)
696 ram_addr_t current_addr
;
697 MemoryRegion
*mr
= block
->mr
;
700 bool send_async
= true;
702 p
= memory_region_get_ram_ptr(mr
) + offset
;
704 /* In doubt sent page as normal */
706 ret
= ram_control_save_page(f
, block
->offset
,
707 offset
, TARGET_PAGE_SIZE
, &bytes_xmit
);
709 *bytes_transferred
+= bytes_xmit
;
715 current_addr
= block
->offset
+ offset
;
717 if (block
== last_sent_block
) {
718 offset
|= RAM_SAVE_FLAG_CONTINUE
;
720 if (ret
!= RAM_SAVE_CONTROL_NOT_SUPP
) {
721 if (ret
!= RAM_SAVE_CONTROL_DELAYED
) {
722 if (bytes_xmit
> 0) {
723 acct_info
.norm_pages
++;
724 } else if (bytes_xmit
== 0) {
725 acct_info
.dup_pages
++;
728 } else if (is_zero_range(p
, TARGET_PAGE_SIZE
)) {
729 acct_info
.dup_pages
++;
730 *bytes_transferred
+= save_page_header(f
, block
,
731 offset
| RAM_SAVE_FLAG_COMPRESS
);
733 *bytes_transferred
+= 1;
735 /* Must let xbzrle know, otherwise a previous (now 0'd) cached
736 * page would be stale
738 xbzrle_cache_zero_page(current_addr
);
739 } else if (!ram_bulk_stage
&& migrate_use_xbzrle()) {
740 pages
= save_xbzrle_page(f
, &p
, current_addr
, block
,
741 offset
, last_stage
, bytes_transferred
);
743 /* Can't send this cached data async, since the cache page
744 * might get updated before it gets to the wire
750 /* XBZRLE overflow or normal page */
752 *bytes_transferred
+= save_page_header(f
, block
,
753 offset
| RAM_SAVE_FLAG_PAGE
);
755 qemu_put_buffer_async(f
, p
, TARGET_PAGE_SIZE
);
757 qemu_put_buffer(f
, p
, TARGET_PAGE_SIZE
);
759 *bytes_transferred
+= TARGET_PAGE_SIZE
;
761 acct_info
.norm_pages
++;
764 XBZRLE_cache_unlock();
770 * ram_save_compressed_page: compress the given page and send it to the stream
772 * Returns: Number of pages written.
774 * @f: QEMUFile where to send the data
775 * @block: block that contains the page we want to send
776 * @offset: offset inside the block for the page
777 * @last_stage: if we are at the completion stage
778 * @bytes_transferred: increase it with the number of transferred bytes
780 static int ram_save_compressed_page(QEMUFile
*f
, RAMBlock
*block
,
781 ram_addr_t offset
, bool last_stage
,
782 uint64_t *bytes_transferred
)
792 * ram_find_and_save_block: Finds a dirty page and sends it to f
794 * Called within an RCU critical section.
796 * Returns: The number of pages written
797 * 0 means no dirty pages
799 * @f: QEMUFile where to send the data
800 * @last_stage: if we are at the completion stage
801 * @bytes_transferred: increase it with the number of transferred bytes
804 static int ram_find_and_save_block(QEMUFile
*f
, bool last_stage
,
805 uint64_t *bytes_transferred
)
807 RAMBlock
*block
= last_seen_block
;
808 ram_addr_t offset
= last_offset
;
809 bool complete_round
= false;
814 block
= QLIST_FIRST_RCU(&ram_list
.blocks
);
818 offset
= migration_bitmap_find_and_reset_dirty(mr
, offset
);
819 if (complete_round
&& block
== last_seen_block
&&
820 offset
>= last_offset
) {
823 if (offset
>= block
->used_length
) {
825 block
= QLIST_NEXT_RCU(block
, next
);
827 block
= QLIST_FIRST_RCU(&ram_list
.blocks
);
828 complete_round
= true;
829 ram_bulk_stage
= false;
832 if (migrate_use_compression()) {
833 pages
= ram_save_compressed_page(f
, block
, offset
, last_stage
,
836 pages
= ram_save_page(f
, block
, offset
, last_stage
,
840 /* if page is unmodified, continue to the next */
842 last_sent_block
= block
;
848 last_seen_block
= block
;
849 last_offset
= offset
;
854 static uint64_t bytes_transferred
;
856 void acct_update_position(QEMUFile
*f
, size_t size
, bool zero
)
858 uint64_t pages
= size
/ TARGET_PAGE_SIZE
;
860 acct_info
.dup_pages
+= pages
;
862 acct_info
.norm_pages
+= pages
;
863 bytes_transferred
+= size
;
864 qemu_update_position(f
, size
);
868 static ram_addr_t
ram_save_remaining(void)
870 return migration_dirty_pages
;
873 uint64_t ram_bytes_remaining(void)
875 return ram_save_remaining() * TARGET_PAGE_SIZE
;
878 uint64_t ram_bytes_transferred(void)
880 return bytes_transferred
;
883 uint64_t ram_bytes_total(void)
889 QLIST_FOREACH_RCU(block
, &ram_list
.blocks
, next
)
890 total
+= block
->used_length
;
895 void free_xbzrle_decoded_buf(void)
897 g_free(xbzrle_decoded_buf
);
898 xbzrle_decoded_buf
= NULL
;
901 static void migration_end(void)
903 if (migration_bitmap
) {
904 memory_global_dirty_log_stop();
905 g_free(migration_bitmap
);
906 migration_bitmap
= NULL
;
911 cache_fini(XBZRLE
.cache
);
912 g_free(XBZRLE
.encoded_buf
);
913 g_free(XBZRLE
.current_buf
);
915 XBZRLE
.encoded_buf
= NULL
;
916 XBZRLE
.current_buf
= NULL
;
918 XBZRLE_cache_unlock();
921 static void ram_migration_cancel(void *opaque
)
926 static void reset_ram_globals(void)
928 last_seen_block
= NULL
;
929 last_sent_block
= NULL
;
931 last_version
= ram_list
.version
;
932 ram_bulk_stage
= true;
935 #define MAX_WAIT 50 /* ms, half buffered_file limit */
938 /* Each of ram_save_setup, ram_save_iterate and ram_save_complete has
939 * long-running RCU critical section. When rcu-reclaims in the code
940 * start to become numerous it will be necessary to reduce the
941 * granularity of these critical sections.
944 static int ram_save_setup(QEMUFile
*f
, void *opaque
)
947 int64_t ram_bitmap_pages
; /* Size of bitmap in pages, including gaps */
949 mig_throttle_on
= false;
950 dirty_rate_high_cnt
= 0;
951 bitmap_sync_count
= 0;
952 migration_bitmap_sync_init();
954 if (migrate_use_xbzrle()) {
956 XBZRLE
.cache
= cache_init(migrate_xbzrle_cache_size() /
960 XBZRLE_cache_unlock();
961 error_report("Error creating cache");
964 XBZRLE_cache_unlock();
966 /* We prefer not to abort if there is no memory */
967 XBZRLE
.encoded_buf
= g_try_malloc0(TARGET_PAGE_SIZE
);
968 if (!XBZRLE
.encoded_buf
) {
969 error_report("Error allocating encoded_buf");
973 XBZRLE
.current_buf
= g_try_malloc(TARGET_PAGE_SIZE
);
974 if (!XBZRLE
.current_buf
) {
975 error_report("Error allocating current_buf");
976 g_free(XBZRLE
.encoded_buf
);
977 XBZRLE
.encoded_buf
= NULL
;
984 /* iothread lock needed for ram_list.dirty_memory[] */
985 qemu_mutex_lock_iothread();
986 qemu_mutex_lock_ramlist();
988 bytes_transferred
= 0;
991 ram_bitmap_pages
= last_ram_offset() >> TARGET_PAGE_BITS
;
992 migration_bitmap
= bitmap_new(ram_bitmap_pages
);
993 bitmap_set(migration_bitmap
, 0, ram_bitmap_pages
);
996 * Count the total number of pages used by ram blocks not including any
997 * gaps due to alignment or unplugs.
999 migration_dirty_pages
= ram_bytes_total() >> TARGET_PAGE_BITS
;
1001 memory_global_dirty_log_start();
1002 migration_bitmap_sync();
1003 qemu_mutex_unlock_ramlist();
1004 qemu_mutex_unlock_iothread();
1006 qemu_put_be64(f
, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE
);
1008 QLIST_FOREACH_RCU(block
, &ram_list
.blocks
, next
) {
1009 qemu_put_byte(f
, strlen(block
->idstr
));
1010 qemu_put_buffer(f
, (uint8_t *)block
->idstr
, strlen(block
->idstr
));
1011 qemu_put_be64(f
, block
->used_length
);
1016 ram_control_before_iterate(f
, RAM_CONTROL_SETUP
);
1017 ram_control_after_iterate(f
, RAM_CONTROL_SETUP
);
1019 qemu_put_be64(f
, RAM_SAVE_FLAG_EOS
);
1024 static int ram_save_iterate(QEMUFile
*f
, void *opaque
)
1032 if (ram_list
.version
!= last_version
) {
1033 reset_ram_globals();
1036 /* Read version before ram_list.blocks */
1039 ram_control_before_iterate(f
, RAM_CONTROL_ROUND
);
1041 t0
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
1043 while ((ret
= qemu_file_rate_limit(f
)) == 0) {
1046 pages
= ram_find_and_save_block(f
, false, &bytes_transferred
);
1047 /* no more pages to sent */
1051 pages_sent
+= pages
;
1052 acct_info
.iterations
++;
1053 check_guest_throttling();
1054 /* we want to check in the 1st loop, just in case it was the 1st time
1055 and we had to sync the dirty bitmap.
1056 qemu_get_clock_ns() is a bit expensive, so we only check each some
1059 if ((i
& 63) == 0) {
1060 uint64_t t1
= (qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - t0
) / 1000000;
1061 if (t1
> MAX_WAIT
) {
1062 DPRINTF("big wait: %" PRIu64
" milliseconds, %d iterations\n",
1072 * Must occur before EOS (or any QEMUFile operation)
1073 * because of RDMA protocol.
1075 ram_control_after_iterate(f
, RAM_CONTROL_ROUND
);
1077 qemu_put_be64(f
, RAM_SAVE_FLAG_EOS
);
1078 bytes_transferred
+= 8;
1080 ret
= qemu_file_get_error(f
);
1088 /* Called with iothread lock */
1089 static int ram_save_complete(QEMUFile
*f
, void *opaque
)
1093 migration_bitmap_sync();
1095 ram_control_before_iterate(f
, RAM_CONTROL_FINISH
);
1097 /* try transferring iterative blocks of memory */
1099 /* flush all remaining blocks regardless of rate limiting */
1103 pages
= ram_find_and_save_block(f
, true, &bytes_transferred
);
1104 /* no more blocks to sent */
1110 ram_control_after_iterate(f
, RAM_CONTROL_FINISH
);
1114 qemu_put_be64(f
, RAM_SAVE_FLAG_EOS
);
1119 static uint64_t ram_save_pending(QEMUFile
*f
, void *opaque
, uint64_t max_size
)
1121 uint64_t remaining_size
;
1123 remaining_size
= ram_save_remaining() * TARGET_PAGE_SIZE
;
1125 if (remaining_size
< max_size
) {
1126 qemu_mutex_lock_iothread();
1128 migration_bitmap_sync();
1130 qemu_mutex_unlock_iothread();
1131 remaining_size
= ram_save_remaining() * TARGET_PAGE_SIZE
;
1133 return remaining_size
;
1136 static int load_xbzrle(QEMUFile
*f
, ram_addr_t addr
, void *host
)
1138 unsigned int xh_len
;
1141 if (!xbzrle_decoded_buf
) {
1142 xbzrle_decoded_buf
= g_malloc(TARGET_PAGE_SIZE
);
1145 /* extract RLE header */
1146 xh_flags
= qemu_get_byte(f
);
1147 xh_len
= qemu_get_be16(f
);
1149 if (xh_flags
!= ENCODING_FLAG_XBZRLE
) {
1150 error_report("Failed to load XBZRLE page - wrong compression!");
1154 if (xh_len
> TARGET_PAGE_SIZE
) {
1155 error_report("Failed to load XBZRLE page - len overflow!");
1158 /* load data and decode */
1159 qemu_get_buffer(f
, xbzrle_decoded_buf
, xh_len
);
1162 if (xbzrle_decode_buffer(xbzrle_decoded_buf
, xh_len
, host
,
1163 TARGET_PAGE_SIZE
) == -1) {
1164 error_report("Failed to load XBZRLE page - decode error!");
1171 /* Must be called from within a rcu critical section.
1172 * Returns a pointer from within the RCU-protected ram_list.
1174 static inline void *host_from_stream_offset(QEMUFile
*f
,
1178 static RAMBlock
*block
= NULL
;
1182 if (flags
& RAM_SAVE_FLAG_CONTINUE
) {
1183 if (!block
|| block
->max_length
<= offset
) {
1184 error_report("Ack, bad migration stream!");
1188 return memory_region_get_ram_ptr(block
->mr
) + offset
;
1191 len
= qemu_get_byte(f
);
1192 qemu_get_buffer(f
, (uint8_t *)id
, len
);
1195 QLIST_FOREACH_RCU(block
, &ram_list
.blocks
, next
) {
1196 if (!strncmp(id
, block
->idstr
, sizeof(id
)) &&
1197 block
->max_length
> offset
) {
1198 return memory_region_get_ram_ptr(block
->mr
) + offset
;
1202 error_report("Can't find block %s!", id
);
1207 * If a page (or a whole RDMA chunk) has been
1208 * determined to be zero, then zap it.
1210 void ram_handle_compressed(void *host
, uint8_t ch
, uint64_t size
)
1212 if (ch
!= 0 || !is_zero_range(host
, size
)) {
1213 memset(host
, ch
, size
);
1217 static void *do_data_decompress(void *opaque
)
1219 while (!quit_decomp_thread
) {
1226 void migrate_decompress_threads_create(void)
1228 int i
, thread_count
;
1230 thread_count
= migrate_decompress_threads();
1231 decompress_threads
= g_new0(QemuThread
, thread_count
);
1232 decomp_param
= g_new0(DecompressParam
, thread_count
);
1233 compressed_data_buf
= g_malloc0(compressBound(TARGET_PAGE_SIZE
));
1234 quit_decomp_thread
= false;
1235 for (i
= 0; i
< thread_count
; i
++) {
1236 qemu_thread_create(decompress_threads
+ i
, "decompress",
1237 do_data_decompress
, decomp_param
+ i
,
1238 QEMU_THREAD_JOINABLE
);
1242 void migrate_decompress_threads_join(void)
1244 int i
, thread_count
;
1246 quit_decomp_thread
= true;
1247 thread_count
= migrate_decompress_threads();
1248 for (i
= 0; i
< thread_count
; i
++) {
1249 qemu_thread_join(decompress_threads
+ i
);
1251 g_free(decompress_threads
);
1252 g_free(decomp_param
);
1253 g_free(compressed_data_buf
);
1254 decompress_threads
= NULL
;
1255 decomp_param
= NULL
;
1256 compressed_data_buf
= NULL
;
1259 static void decompress_data_with_multi_threads(uint8_t *compbuf
,
1260 void *host
, int len
)
1265 static int ram_load(QEMUFile
*f
, void *opaque
, int version_id
)
1267 int flags
= 0, ret
= 0;
1268 static uint64_t seq_iter
;
1273 if (version_id
!= 4) {
1277 /* This RCU critical section can be very long running.
1278 * When RCU reclaims in the code start to become numerous,
1279 * it will be necessary to reduce the granularity of this
1283 while (!ret
&& !(flags
& RAM_SAVE_FLAG_EOS
)) {
1284 ram_addr_t addr
, total_ram_bytes
;
1288 addr
= qemu_get_be64(f
);
1289 flags
= addr
& ~TARGET_PAGE_MASK
;
1290 addr
&= TARGET_PAGE_MASK
;
1292 switch (flags
& ~RAM_SAVE_FLAG_CONTINUE
) {
1293 case RAM_SAVE_FLAG_MEM_SIZE
:
1294 /* Synchronize RAM block list */
1295 total_ram_bytes
= addr
;
1296 while (!ret
&& total_ram_bytes
) {
1302 len
= qemu_get_byte(f
);
1303 qemu_get_buffer(f
, (uint8_t *)id
, len
);
1305 length
= qemu_get_be64(f
);
1307 QLIST_FOREACH_RCU(block
, &ram_list
.blocks
, next
) {
1308 if (!strncmp(id
, block
->idstr
, sizeof(id
))) {
1309 if (length
!= block
->used_length
) {
1310 Error
*local_err
= NULL
;
1312 ret
= qemu_ram_resize(block
->offset
, length
, &local_err
);
1314 error_report_err(local_err
);
1322 error_report("Unknown ramblock \"%s\", cannot "
1323 "accept migration", id
);
1327 total_ram_bytes
-= length
;
1330 case RAM_SAVE_FLAG_COMPRESS
:
1331 host
= host_from_stream_offset(f
, addr
, flags
);
1333 error_report("Illegal RAM offset " RAM_ADDR_FMT
, addr
);
1337 ch
= qemu_get_byte(f
);
1338 ram_handle_compressed(host
, ch
, TARGET_PAGE_SIZE
);
1340 case RAM_SAVE_FLAG_PAGE
:
1341 host
= host_from_stream_offset(f
, addr
, flags
);
1343 error_report("Illegal RAM offset " RAM_ADDR_FMT
, addr
);
1347 qemu_get_buffer(f
, host
, TARGET_PAGE_SIZE
);
1349 case RAM_SAVE_FLAG_COMPRESS_PAGE
:
1350 host
= host_from_stream_offset(f
, addr
, flags
);
1352 error_report("Invalid RAM offset " RAM_ADDR_FMT
, addr
);
1357 len
= qemu_get_be32(f
);
1358 if (len
< 0 || len
> compressBound(TARGET_PAGE_SIZE
)) {
1359 error_report("Invalid compressed data length: %d", len
);
1363 qemu_get_buffer(f
, compressed_data_buf
, len
);
1364 decompress_data_with_multi_threads(compressed_data_buf
, host
, len
);
1366 case RAM_SAVE_FLAG_XBZRLE
:
1367 host
= host_from_stream_offset(f
, addr
, flags
);
1369 error_report("Illegal RAM offset " RAM_ADDR_FMT
, addr
);
1373 if (load_xbzrle(f
, addr
, host
) < 0) {
1374 error_report("Failed to decompress XBZRLE page at "
1375 RAM_ADDR_FMT
, addr
);
1380 case RAM_SAVE_FLAG_EOS
:
1384 if (flags
& RAM_SAVE_FLAG_HOOK
) {
1385 ram_control_load_hook(f
, flags
);
1387 error_report("Unknown combination of migration flags: %#x",
1393 ret
= qemu_file_get_error(f
);
1398 DPRINTF("Completed load of VM with exit code %d seq iteration "
1399 "%" PRIu64
"\n", ret
, seq_iter
);
1403 static SaveVMHandlers savevm_ram_handlers
= {
1404 .save_live_setup
= ram_save_setup
,
1405 .save_live_iterate
= ram_save_iterate
,
1406 .save_live_complete
= ram_save_complete
,
1407 .save_live_pending
= ram_save_pending
,
1408 .load_state
= ram_load
,
1409 .cancel
= ram_migration_cancel
,
1412 void ram_mig_init(void)
1414 qemu_mutex_init(&XBZRLE
.lock
);
1415 register_savevm_live(NULL
, "ram", 0, 4, &savevm_ram_handlers
, NULL
);
1424 int (*init_isa
) (ISABus
*bus
);
1425 int (*init_pci
) (PCIBus
*bus
);
1429 static struct soundhw soundhw
[9];
1430 static int soundhw_count
;
1432 void isa_register_soundhw(const char *name
, const char *descr
,
1433 int (*init_isa
)(ISABus
*bus
))
1435 assert(soundhw_count
< ARRAY_SIZE(soundhw
) - 1);
1436 soundhw
[soundhw_count
].name
= name
;
1437 soundhw
[soundhw_count
].descr
= descr
;
1438 soundhw
[soundhw_count
].isa
= 1;
1439 soundhw
[soundhw_count
].init
.init_isa
= init_isa
;
1443 void pci_register_soundhw(const char *name
, const char *descr
,
1444 int (*init_pci
)(PCIBus
*bus
))
1446 assert(soundhw_count
< ARRAY_SIZE(soundhw
) - 1);
1447 soundhw
[soundhw_count
].name
= name
;
1448 soundhw
[soundhw_count
].descr
= descr
;
1449 soundhw
[soundhw_count
].isa
= 0;
1450 soundhw
[soundhw_count
].init
.init_pci
= init_pci
;
1454 void select_soundhw(const char *optarg
)
1458 if (is_help_option(optarg
)) {
1461 if (soundhw_count
) {
1462 printf("Valid sound card names (comma separated):\n");
1463 for (c
= soundhw
; c
->name
; ++c
) {
1464 printf ("%-11s %s\n", c
->name
, c
->descr
);
1466 printf("\n-soundhw all will enable all of the above\n");
1468 printf("Machine has no user-selectable audio hardware "
1469 "(it may or may not have always-present audio hardware).\n");
1471 exit(!is_help_option(optarg
));
1479 if (!strcmp(optarg
, "all")) {
1480 for (c
= soundhw
; c
->name
; ++c
) {
1489 l
= !e
? strlen(p
) : (size_t) (e
- p
);
1491 for (c
= soundhw
; c
->name
; ++c
) {
1492 if (!strncmp(c
->name
, p
, l
) && !c
->name
[l
]) {
1500 error_report("Unknown sound card name (too big to show)");
1503 error_report("Unknown sound card name `%.*s'",
1508 p
+= l
+ (e
!= NULL
);
1512 goto show_valid_cards
;
1517 void audio_init(void)
1520 ISABus
*isa_bus
= (ISABus
*) object_resolve_path_type("", TYPE_ISA_BUS
, NULL
);
1521 PCIBus
*pci_bus
= (PCIBus
*) object_resolve_path_type("", TYPE_PCI_BUS
, NULL
);
1523 for (c
= soundhw
; c
->name
; ++c
) {
1527 error_report("ISA bus not available for %s", c
->name
);
1530 c
->init
.init_isa(isa_bus
);
1533 error_report("PCI bus not available for %s", c
->name
);
1536 c
->init
.init_pci(pci_bus
);
1542 int qemu_uuid_parse(const char *str
, uint8_t *uuid
)
1546 if (strlen(str
) != 36) {
1550 ret
= sscanf(str
, UUID_FMT
, &uuid
[0], &uuid
[1], &uuid
[2], &uuid
[3],
1551 &uuid
[4], &uuid
[5], &uuid
[6], &uuid
[7], &uuid
[8], &uuid
[9],
1552 &uuid
[10], &uuid
[11], &uuid
[12], &uuid
[13], &uuid
[14],
1561 void do_acpitable_option(const QemuOpts
*opts
)
1566 acpi_table_add(opts
, &err
);
1568 error_report("Wrong acpi table provided: %s",
1569 error_get_pretty(err
));
1576 void do_smbios_option(QemuOpts
*opts
)
1579 smbios_entry_add(opts
);
1583 void cpudef_init(void)
1585 #if defined(cpudef_setup)
1586 cpudef_setup(); /* parse cpu definitions in target config file */
1590 int kvm_available(void)
1599 int xen_available(void)
1609 TargetInfo
*qmp_query_target(Error
**errp
)
1611 TargetInfo
*info
= g_malloc0(sizeof(*info
));
1613 info
->arch
= g_strdup(TARGET_NAME
);
1618 /* Stub function that's gets run on the vcpu when its brought out of the
1619 VM to run inside qemu via async_run_on_cpu()*/
1620 static void mig_sleep_cpu(void *opq
)
1622 qemu_mutex_unlock_iothread();
1624 qemu_mutex_lock_iothread();
1627 /* To reduce the dirty rate explicitly disallow the VCPUs from spending
1628 much time in the VM. The migration thread will try to catchup.
1629 Workload will experience a performance drop.
1631 static void mig_throttle_guest_down(void)
1635 qemu_mutex_lock_iothread();
1637 async_run_on_cpu(cpu
, mig_sleep_cpu
, NULL
);
1639 qemu_mutex_unlock_iothread();
1642 static void check_guest_throttling(void)
1647 if (!mig_throttle_on
) {
1652 t0
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
1656 t1
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
1658 /* If it has been more than 40 ms since the last time the guest
1659 * was throttled then do it again.
1661 if (40 < (t1
-t0
)/1000000) {
1662 mig_throttle_guest_down();