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"
55 #include "qemu/rcu_queue.h"
57 #ifdef DEBUG_ARCH_INIT
58 #define DPRINTF(fmt, ...) \
59 do { fprintf(stdout, "arch_init: " fmt, ## __VA_ARGS__); } while (0)
61 #define DPRINTF(fmt, ...) \
66 int graphic_width = 1024;
67 int graphic_height = 768;
68 int graphic_depth = 8;
70 int graphic_width = 800;
71 int graphic_height = 600;
72 int graphic_depth = 32;
76 #if defined(TARGET_ALPHA)
77 #define QEMU_ARCH QEMU_ARCH_ALPHA
78 #elif defined(TARGET_ARM)
79 #define QEMU_ARCH QEMU_ARCH_ARM
80 #elif defined(TARGET_CRIS)
81 #define QEMU_ARCH QEMU_ARCH_CRIS
82 #elif defined(TARGET_I386)
83 #define QEMU_ARCH QEMU_ARCH_I386
84 #elif defined(TARGET_M68K)
85 #define QEMU_ARCH QEMU_ARCH_M68K
86 #elif defined(TARGET_LM32)
87 #define QEMU_ARCH QEMU_ARCH_LM32
88 #elif defined(TARGET_MICROBLAZE)
89 #define QEMU_ARCH QEMU_ARCH_MICROBLAZE
90 #elif defined(TARGET_MIPS)
91 #define QEMU_ARCH QEMU_ARCH_MIPS
92 #elif defined(TARGET_MOXIE)
93 #define QEMU_ARCH QEMU_ARCH_MOXIE
94 #elif defined(TARGET_OPENRISC)
95 #define QEMU_ARCH QEMU_ARCH_OPENRISC
96 #elif defined(TARGET_PPC)
97 #define QEMU_ARCH QEMU_ARCH_PPC
98 #elif defined(TARGET_S390X)
99 #define QEMU_ARCH QEMU_ARCH_S390X
100 #elif defined(TARGET_SH4)
101 #define QEMU_ARCH QEMU_ARCH_SH4
102 #elif defined(TARGET_SPARC)
103 #define QEMU_ARCH QEMU_ARCH_SPARC
104 #elif defined(TARGET_XTENSA)
105 #define QEMU_ARCH QEMU_ARCH_XTENSA
106 #elif defined(TARGET_UNICORE32)
107 #define QEMU_ARCH QEMU_ARCH_UNICORE32
108 #elif defined(TARGET_TRICORE)
109 #define QEMU_ARCH QEMU_ARCH_TRICORE
112 const uint32_t arch_type = QEMU_ARCH;
113 static bool mig_throttle_on;
114 static int dirty_rate_high_cnt;
115 static void check_guest_throttling(void);
117 static uint64_t bitmap_sync_count;
119 /***********************************************************/
120 /* ram save/restore */
122 #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */
123 #define RAM_SAVE_FLAG_COMPRESS 0x02
124 #define RAM_SAVE_FLAG_MEM_SIZE 0x04
125 #define RAM_SAVE_FLAG_PAGE 0x08
126 #define RAM_SAVE_FLAG_EOS 0x10
127 #define RAM_SAVE_FLAG_CONTINUE 0x20
128 #define RAM_SAVE_FLAG_XBZRLE 0x40
129 /* 0x80 is reserved in migration.h start with 0x100 next */
131 static struct defconfig_file {
132 const char *filename;
133 /* Indicates it is an user config file (disabled by -no-user-config) */
135 } default_config_files[] = {
136 { CONFIG_QEMU_CONFDIR "/qemu.conf", true },
137 { CONFIG_QEMU_CONFDIR "/target-" TARGET_NAME ".conf", true },
138 { NULL }, /* end of list */
141 static const uint8_t ZERO_TARGET_PAGE[TARGET_PAGE_SIZE];
143 int qemu_read_default_config_files(bool userconfig)
146 struct defconfig_file *f;
148 for (f = default_config_files; f->filename; f++) {
149 if (!userconfig && f->userconfig) {
152 ret = qemu_read_config_file(f->filename);
153 if (ret < 0 && ret != -ENOENT) {
161 static inline bool is_zero_range(uint8_t *p, uint64_t size)
163 return buffer_find_nonzero_offset(p, size) == size;
166 /* struct contains XBZRLE cache and a static page
167 used by the compression */
169 /* buffer used for XBZRLE encoding */
170 uint8_t *encoded_buf;
171 /* buffer for storing page content */
172 uint8_t *current_buf;
173 /* Cache for XBZRLE, Protected by lock. */
178 /* buffer used for XBZRLE decoding */
179 static uint8_t *xbzrle_decoded_buf;
181 static void XBZRLE_cache_lock(void)
183 if (migrate_use_xbzrle())
184 qemu_mutex_lock(&XBZRLE.lock);
187 static void XBZRLE_cache_unlock(void)
189 if (migrate_use_xbzrle())
190 qemu_mutex_unlock(&XBZRLE.lock);
194 * called from qmp_migrate_set_cache_size in main thread, possibly while
195 * a migration is in progress.
196 * A running migration maybe using the cache and might finish during this
197 * call, hence changes to the cache are protected by XBZRLE.lock().
199 int64_t xbzrle_cache_resize(int64_t new_size)
201 PageCache *new_cache;
204 if (new_size < TARGET_PAGE_SIZE) {
210 if (XBZRLE.cache != NULL) {
211 if (pow2floor(new_size) == migrate_xbzrle_cache_size()) {
214 new_cache = cache_init(new_size / TARGET_PAGE_SIZE,
217 error_report("Error creating cache");
222 cache_fini(XBZRLE.cache);
223 XBZRLE.cache = new_cache;
227 ret = pow2floor(new_size);
229 XBZRLE_cache_unlock();
233 /* accounting for migration statistics */
234 typedef struct AccountingInfo {
236 uint64_t skipped_pages;
239 uint64_t xbzrle_bytes;
240 uint64_t xbzrle_pages;
241 uint64_t xbzrle_cache_miss;
242 double xbzrle_cache_miss_rate;
243 uint64_t xbzrle_overflows;
246 static AccountingInfo acct_info;
248 static void acct_clear(void)
250 memset(&acct_info, 0, sizeof(acct_info));
253 uint64_t dup_mig_bytes_transferred(void)
255 return acct_info.dup_pages * TARGET_PAGE_SIZE;
258 uint64_t dup_mig_pages_transferred(void)
260 return acct_info.dup_pages;
263 uint64_t skipped_mig_bytes_transferred(void)
265 return acct_info.skipped_pages * TARGET_PAGE_SIZE;
268 uint64_t skipped_mig_pages_transferred(void)
270 return acct_info.skipped_pages;
273 uint64_t norm_mig_bytes_transferred(void)
275 return acct_info.norm_pages * TARGET_PAGE_SIZE;
278 uint64_t norm_mig_pages_transferred(void)
280 return acct_info.norm_pages;
283 uint64_t xbzrle_mig_bytes_transferred(void)
285 return acct_info.xbzrle_bytes;
288 uint64_t xbzrle_mig_pages_transferred(void)
290 return acct_info.xbzrle_pages;
293 uint64_t xbzrle_mig_pages_cache_miss(void)
295 return acct_info.xbzrle_cache_miss;
298 double xbzrle_mig_cache_miss_rate(void)
300 return acct_info.xbzrle_cache_miss_rate;
303 uint64_t xbzrle_mig_pages_overflow(void)
305 return acct_info.xbzrle_overflows;
308 /* This is the last block that we have visited serching for dirty pages
310 static RAMBlock *last_seen_block;
311 /* This is the last block from where we have sent data */
312 static RAMBlock *last_sent_block;
313 static ram_addr_t last_offset;
314 static unsigned long *migration_bitmap;
315 static uint64_t migration_dirty_pages;
316 static uint32_t last_version;
317 static bool ram_bulk_stage;
320 * save_page_header: Write page header to wire
322 * If this is the 1st block, it also writes the block identification
324 * Returns: Number of bytes written
326 * @f: QEMUFile where to send the data
327 * @block: block that contains the page we want to send
328 * @offset: offset inside the block for the page
329 * in the lower bits, it contains flags
331 static size_t save_page_header(QEMUFile *f, RAMBlock *block, ram_addr_t offset)
335 if (block == last_sent_block) {
336 offset |= RAM_SAVE_FLAG_CONTINUE;
339 qemu_put_be64(f, offset);
342 if (block != last_sent_block) {
343 qemu_put_byte(f, strlen(block->idstr));
344 qemu_put_buffer(f, (uint8_t *)block->idstr,
345 strlen(block->idstr));
346 size += 1 + strlen(block->idstr);
347 last_sent_block = block;
352 /* Update the xbzrle cache to reflect a page that's been sent as all 0.
353 * The important thing is that a stale (not-yet-0'd) page be replaced
355 * As a bonus, if the page wasn't in the cache it gets added so that
356 * when a small write is made into the 0'd page it gets XBZRLE sent
358 static void xbzrle_cache_zero_page(ram_addr_t current_addr)
360 if (ram_bulk_stage || !migrate_use_xbzrle()) {
364 /* We don't care if this fails to allocate a new cache page
365 * as long as it updated an old one */
366 cache_insert(XBZRLE.cache, current_addr, ZERO_TARGET_PAGE,
370 #define ENCODING_FLAG_XBZRLE 0x1
373 * save_xbzrle_page: compress and send current page
375 * Returns: 1 means that we wrote the page
376 * 0 means that page is identical to the one already sent
377 * -1 means that xbzrle would be longer than normal
379 * @f: QEMUFile where to send the data
382 * @block: block that contains the page we want to send
383 * @offset: offset inside the block for the page
384 * @last_stage: if we are at the completion stage
385 * @bytes_transferred: increase it with the number of transferred bytes
387 static int save_xbzrle_page(QEMUFile *f, uint8_t **current_data,
388 ram_addr_t current_addr, RAMBlock *block,
389 ram_addr_t offset, bool last_stage,
390 uint64_t *bytes_transferred)
392 int encoded_len = 0, bytes_xbzrle;
393 uint8_t *prev_cached_page;
395 if (!cache_is_cached(XBZRLE.cache, current_addr, bitmap_sync_count)) {
396 acct_info.xbzrle_cache_miss++;
398 if (cache_insert(XBZRLE.cache, current_addr, *current_data,
399 bitmap_sync_count) == -1) {
402 /* update *current_data when the page has been
403 inserted into cache */
404 *current_data = get_cached_data(XBZRLE.cache, current_addr);
410 prev_cached_page = get_cached_data(XBZRLE.cache, current_addr);
412 /* save current buffer into memory */
413 memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE);
415 /* XBZRLE encoding (if there is no overflow) */
416 encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf,
417 TARGET_PAGE_SIZE, XBZRLE.encoded_buf,
419 if (encoded_len == 0) {
420 DPRINTF("Skipping unmodified page\n");
422 } else if (encoded_len == -1) {
423 DPRINTF("Overflow\n");
424 acct_info.xbzrle_overflows++;
425 /* update data in the cache */
427 memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE);
428 *current_data = prev_cached_page;
433 /* we need to update the data in the cache, in order to get the same data */
435 memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE);
438 /* Send XBZRLE based compressed page */
439 bytes_xbzrle = save_page_header(f, block, offset | RAM_SAVE_FLAG_XBZRLE);
440 qemu_put_byte(f, ENCODING_FLAG_XBZRLE);
441 qemu_put_be16(f, encoded_len);
442 qemu_put_buffer(f, XBZRLE.encoded_buf, encoded_len);
443 bytes_xbzrle += encoded_len + 1 + 2;
444 acct_info.xbzrle_pages++;
445 acct_info.xbzrle_bytes += bytes_xbzrle;
446 *bytes_transferred += bytes_xbzrle;
452 ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr,
455 unsigned long base = mr->ram_addr >> TARGET_PAGE_BITS;
456 unsigned long nr = base + (start >> TARGET_PAGE_BITS);
457 uint64_t mr_size = TARGET_PAGE_ALIGN(memory_region_size(mr));
458 unsigned long size = base + (mr_size >> TARGET_PAGE_BITS);
462 if (ram_bulk_stage && nr > base) {
465 next = find_next_bit(migration_bitmap, size, nr);
469 clear_bit(next, migration_bitmap);
470 migration_dirty_pages--;
472 return (next - base) << TARGET_PAGE_BITS;
475 static inline bool migration_bitmap_set_dirty(ram_addr_t addr)
478 int nr = addr >> TARGET_PAGE_BITS;
480 ret = test_and_set_bit(nr, migration_bitmap);
483 migration_dirty_pages++;
488 static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length)
491 unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
493 /* start address is aligned at the start of a word? */
494 if (((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) {
496 int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS);
497 unsigned long *src = ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION];
499 for (k = page; k < page + nr; k++) {
501 unsigned long new_dirty;
502 new_dirty = ~migration_bitmap[k];
503 migration_bitmap[k] |= src[k];
505 migration_dirty_pages += ctpopl(new_dirty);
510 for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) {
511 if (cpu_physical_memory_get_dirty(start + addr,
513 DIRTY_MEMORY_MIGRATION)) {
514 cpu_physical_memory_reset_dirty(start + addr,
516 DIRTY_MEMORY_MIGRATION);
517 migration_bitmap_set_dirty(start + addr);
524 /* Fix me: there are too many global variables used in migration process. */
525 static int64_t start_time;
526 static int64_t bytes_xfer_prev;
527 static int64_t num_dirty_pages_period;
529 static void migration_bitmap_sync_init(void)
533 num_dirty_pages_period = 0;
536 /* Called with iothread lock held, to protect ram_list.dirty_memory[] */
537 static void migration_bitmap_sync(void)
540 uint64_t num_dirty_pages_init = migration_dirty_pages;
541 MigrationState *s = migrate_get_current();
543 int64_t bytes_xfer_now;
544 static uint64_t xbzrle_cache_miss_prev;
545 static uint64_t iterations_prev;
549 if (!bytes_xfer_prev) {
550 bytes_xfer_prev = ram_bytes_transferred();
554 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
557 trace_migration_bitmap_sync_start();
558 address_space_sync_dirty_bitmap(&address_space_memory);
561 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
562 migration_bitmap_sync_range(block->mr->ram_addr, block->used_length);
566 trace_migration_bitmap_sync_end(migration_dirty_pages
567 - num_dirty_pages_init);
568 num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init;
569 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
571 /* more than 1 second = 1000 millisecons */
572 if (end_time > start_time + 1000) {
573 if (migrate_auto_converge()) {
574 /* The following detection logic can be refined later. For now:
575 Check to see if the dirtied bytes is 50% more than the approx.
576 amount of bytes that just got transferred since the last time we
577 were in this routine. If that happens >N times (for now N==4)
578 we turn on the throttle down logic */
579 bytes_xfer_now = ram_bytes_transferred();
580 if (s->dirty_pages_rate &&
581 (num_dirty_pages_period * TARGET_PAGE_SIZE >
582 (bytes_xfer_now - bytes_xfer_prev)/2) &&
583 (dirty_rate_high_cnt++ > 4)) {
584 trace_migration_throttle();
585 mig_throttle_on = true;
586 dirty_rate_high_cnt = 0;
588 bytes_xfer_prev = bytes_xfer_now;
590 mig_throttle_on = false;
592 if (migrate_use_xbzrle()) {
593 if (iterations_prev != 0) {
594 acct_info.xbzrle_cache_miss_rate =
595 (double)(acct_info.xbzrle_cache_miss -
596 xbzrle_cache_miss_prev) /
597 (acct_info.iterations - iterations_prev);
599 iterations_prev = acct_info.iterations;
600 xbzrle_cache_miss_prev = acct_info.xbzrle_cache_miss;
602 s->dirty_pages_rate = num_dirty_pages_period * 1000
603 / (end_time - start_time);
604 s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE;
605 start_time = end_time;
606 num_dirty_pages_period = 0;
607 s->dirty_sync_count = bitmap_sync_count;
612 * ram_save_page: Send the given page to the stream
614 * Returns: Number of pages written.
616 * @f: QEMUFile where to send the data
617 * @block: block that contains the page we want to send
618 * @offset: offset inside the block for the page
619 * @last_stage: if we are at the completion stage
620 * @bytes_transferred: increase it with the number of transferred bytes
622 static int ram_save_page(QEMUFile *f, RAMBlock* block, ram_addr_t offset,
623 bool last_stage, uint64_t *bytes_transferred)
627 ram_addr_t current_addr;
628 MemoryRegion *mr = block->mr;
631 bool send_async = true;
633 p = memory_region_get_ram_ptr(mr) + offset;
635 /* In doubt sent page as normal */
637 ret = ram_control_save_page(f, block->offset,
638 offset, TARGET_PAGE_SIZE, &bytes_xmit);
640 *bytes_transferred += bytes_xmit;
646 current_addr = block->offset + offset;
647 if (ret != RAM_SAVE_CONTROL_NOT_SUPP) {
648 if (ret != RAM_SAVE_CONTROL_DELAYED) {
649 if (bytes_xmit > 0) {
650 acct_info.norm_pages++;
651 } else if (bytes_xmit == 0) {
652 acct_info.dup_pages++;
655 } else if (is_zero_range(p, TARGET_PAGE_SIZE)) {
656 acct_info.dup_pages++;
657 *bytes_transferred += save_page_header(f, block,
658 offset | RAM_SAVE_FLAG_COMPRESS);
660 *bytes_transferred += 1;
662 /* Must let xbzrle know, otherwise a previous (now 0'd) cached
663 * page would be stale
665 xbzrle_cache_zero_page(current_addr);
666 } else if (!ram_bulk_stage && migrate_use_xbzrle()) {
667 pages = save_xbzrle_page(f, &p, current_addr, block,
668 offset, last_stage, bytes_transferred);
670 /* Can't send this cached data async, since the cache page
671 * might get updated before it gets to the wire
677 /* XBZRLE overflow or normal page */
679 *bytes_transferred += save_page_header(f, block,
680 offset | RAM_SAVE_FLAG_PAGE);
682 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE);
684 qemu_put_buffer(f, p, TARGET_PAGE_SIZE);
686 *bytes_transferred += TARGET_PAGE_SIZE;
688 acct_info.norm_pages++;
691 XBZRLE_cache_unlock();
697 * ram_find_and_save_block: Finds a dirty page and sends it to f
699 * Called within an RCU critical section.
701 * Returns: The number of pages written
702 * 0 means no dirty pages
704 * @f: QEMUFile where to send the data
705 * @last_stage: if we are at the completion stage
706 * @bytes_transferred: increase it with the number of transferred bytes
709 static int ram_find_and_save_block(QEMUFile *f, bool last_stage,
710 uint64_t *bytes_transferred)
712 RAMBlock *block = last_seen_block;
713 ram_addr_t offset = last_offset;
714 bool complete_round = false;
719 block = QLIST_FIRST_RCU(&ram_list.blocks);
723 offset = migration_bitmap_find_and_reset_dirty(mr, offset);
724 if (complete_round && block == last_seen_block &&
725 offset >= last_offset) {
728 if (offset >= block->used_length) {
730 block = QLIST_NEXT_RCU(block, next);
732 block = QLIST_FIRST_RCU(&ram_list.blocks);
733 complete_round = true;
734 ram_bulk_stage = false;
737 pages = ram_save_page(f, block, offset, last_stage,
740 /* if page is unmodified, continue to the next */
747 last_seen_block = block;
748 last_offset = offset;
753 static uint64_t bytes_transferred;
755 void acct_update_position(QEMUFile *f, size_t size, bool zero)
757 uint64_t pages = size / TARGET_PAGE_SIZE;
759 acct_info.dup_pages += pages;
761 acct_info.norm_pages += pages;
762 bytes_transferred += size;
763 qemu_update_position(f, size);
767 static ram_addr_t ram_save_remaining(void)
769 return migration_dirty_pages;
772 uint64_t ram_bytes_remaining(void)
774 return ram_save_remaining() * TARGET_PAGE_SIZE;
777 uint64_t ram_bytes_transferred(void)
779 return bytes_transferred;
782 uint64_t ram_bytes_total(void)
788 QLIST_FOREACH_RCU(block, &ram_list.blocks, next)
789 total += block->used_length;
794 void free_xbzrle_decoded_buf(void)
796 g_free(xbzrle_decoded_buf);
797 xbzrle_decoded_buf = NULL;
800 static void migration_end(void)
802 if (migration_bitmap) {
803 memory_global_dirty_log_stop();
804 g_free(migration_bitmap);
805 migration_bitmap = NULL;
810 cache_fini(XBZRLE.cache);
811 g_free(XBZRLE.encoded_buf);
812 g_free(XBZRLE.current_buf);
814 XBZRLE.encoded_buf = NULL;
815 XBZRLE.current_buf = NULL;
817 XBZRLE_cache_unlock();
820 static void ram_migration_cancel(void *opaque)
825 static void reset_ram_globals(void)
827 last_seen_block = NULL;
828 last_sent_block = NULL;
830 last_version = ram_list.version;
831 ram_bulk_stage = true;
834 #define MAX_WAIT 50 /* ms, half buffered_file limit */
837 /* Each of ram_save_setup, ram_save_iterate and ram_save_complete has
838 * long-running RCU critical section. When rcu-reclaims in the code
839 * start to become numerous it will be necessary to reduce the
840 * granularity of these critical sections.
843 static int ram_save_setup(QEMUFile *f, void *opaque)
846 int64_t ram_bitmap_pages; /* Size of bitmap in pages, including gaps */
848 mig_throttle_on = false;
849 dirty_rate_high_cnt = 0;
850 bitmap_sync_count = 0;
851 migration_bitmap_sync_init();
853 if (migrate_use_xbzrle()) {
855 XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() /
859 XBZRLE_cache_unlock();
860 error_report("Error creating cache");
863 XBZRLE_cache_unlock();
865 /* We prefer not to abort if there is no memory */
866 XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE);
867 if (!XBZRLE.encoded_buf) {
868 error_report("Error allocating encoded_buf");
872 XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE);
873 if (!XBZRLE.current_buf) {
874 error_report("Error allocating current_buf");
875 g_free(XBZRLE.encoded_buf);
876 XBZRLE.encoded_buf = NULL;
883 /* iothread lock needed for ram_list.dirty_memory[] */
884 qemu_mutex_lock_iothread();
885 qemu_mutex_lock_ramlist();
887 bytes_transferred = 0;
890 ram_bitmap_pages = last_ram_offset() >> TARGET_PAGE_BITS;
891 migration_bitmap = bitmap_new(ram_bitmap_pages);
892 bitmap_set(migration_bitmap, 0, ram_bitmap_pages);
895 * Count the total number of pages used by ram blocks not including any
896 * gaps due to alignment or unplugs.
898 migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS;
900 memory_global_dirty_log_start();
901 migration_bitmap_sync();
902 qemu_mutex_unlock_ramlist();
903 qemu_mutex_unlock_iothread();
905 qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE);
907 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
908 qemu_put_byte(f, strlen(block->idstr));
909 qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr));
910 qemu_put_be64(f, block->used_length);
915 ram_control_before_iterate(f, RAM_CONTROL_SETUP);
916 ram_control_after_iterate(f, RAM_CONTROL_SETUP);
918 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
923 static int ram_save_iterate(QEMUFile *f, void *opaque)
931 if (ram_list.version != last_version) {
935 /* Read version before ram_list.blocks */
938 ram_control_before_iterate(f, RAM_CONTROL_ROUND);
940 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
942 while ((ret = qemu_file_rate_limit(f)) == 0) {
945 pages = ram_find_and_save_block(f, false, &bytes_transferred);
946 /* no more pages to sent */
951 acct_info.iterations++;
952 check_guest_throttling();
953 /* we want to check in the 1st loop, just in case it was the 1st time
954 and we had to sync the dirty bitmap.
955 qemu_get_clock_ns() is a bit expensive, so we only check each some
959 uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000;
961 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n",
971 * Must occur before EOS (or any QEMUFile operation)
972 * because of RDMA protocol.
974 ram_control_after_iterate(f, RAM_CONTROL_ROUND);
976 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
977 bytes_transferred += 8;
979 ret = qemu_file_get_error(f);
987 /* Called with iothread lock */
988 static int ram_save_complete(QEMUFile *f, void *opaque)
992 migration_bitmap_sync();
994 ram_control_before_iterate(f, RAM_CONTROL_FINISH);
996 /* try transferring iterative blocks of memory */
998 /* flush all remaining blocks regardless of rate limiting */
1002 pages = ram_find_and_save_block(f, true, &bytes_transferred);
1003 /* no more blocks to sent */
1009 ram_control_after_iterate(f, RAM_CONTROL_FINISH);
1013 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
1018 static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
1020 uint64_t remaining_size;
1022 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
1024 if (remaining_size < max_size) {
1025 qemu_mutex_lock_iothread();
1027 migration_bitmap_sync();
1029 qemu_mutex_unlock_iothread();
1030 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
1032 return remaining_size;
1035 static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host)
1037 unsigned int xh_len;
1040 if (!xbzrle_decoded_buf) {
1041 xbzrle_decoded_buf = g_malloc(TARGET_PAGE_SIZE);
1044 /* extract RLE header */
1045 xh_flags = qemu_get_byte(f);
1046 xh_len = qemu_get_be16(f);
1048 if (xh_flags != ENCODING_FLAG_XBZRLE) {
1049 error_report("Failed to load XBZRLE page - wrong compression!");
1053 if (xh_len > TARGET_PAGE_SIZE) {
1054 error_report("Failed to load XBZRLE page - len overflow!");
1057 /* load data and decode */
1058 qemu_get_buffer(f, xbzrle_decoded_buf, xh_len);
1061 if (xbzrle_decode_buffer(xbzrle_decoded_buf, xh_len, host,
1062 TARGET_PAGE_SIZE) == -1) {
1063 error_report("Failed to load XBZRLE page - decode error!");
1070 /* Must be called from within a rcu critical section.
1071 * Returns a pointer from within the RCU-protected ram_list.
1073 static inline void *host_from_stream_offset(QEMUFile *f,
1077 static RAMBlock *block = NULL;
1081 if (flags & RAM_SAVE_FLAG_CONTINUE) {
1082 if (!block || block->max_length <= offset) {
1083 error_report("Ack, bad migration stream!");
1087 return memory_region_get_ram_ptr(block->mr) + offset;
1090 len = qemu_get_byte(f);
1091 qemu_get_buffer(f, (uint8_t *)id, len);
1094 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
1095 if (!strncmp(id, block->idstr, sizeof(id)) &&
1096 block->max_length > offset) {
1097 return memory_region_get_ram_ptr(block->mr) + offset;
1101 error_report("Can't find block %s!", id);
1106 * If a page (or a whole RDMA chunk) has been
1107 * determined to be zero, then zap it.
1109 void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
1111 if (ch != 0 || !is_zero_range(host, size)) {
1112 memset(host, ch, size);
1116 static int ram_load(QEMUFile *f, void *opaque, int version_id)
1118 int flags = 0, ret = 0;
1119 static uint64_t seq_iter;
1123 if (version_id != 4) {
1127 /* This RCU critical section can be very long running.
1128 * When RCU reclaims in the code start to become numerous,
1129 * it will be necessary to reduce the granularity of this
1133 while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) {
1134 ram_addr_t addr, total_ram_bytes;
1138 addr = qemu_get_be64(f);
1139 flags = addr & ~TARGET_PAGE_MASK;
1140 addr &= TARGET_PAGE_MASK;
1142 switch (flags & ~RAM_SAVE_FLAG_CONTINUE) {
1143 case RAM_SAVE_FLAG_MEM_SIZE:
1144 /* Synchronize RAM block list */
1145 total_ram_bytes = addr;
1146 while (!ret && total_ram_bytes) {
1152 len = qemu_get_byte(f);
1153 qemu_get_buffer(f, (uint8_t *)id, len);
1155 length = qemu_get_be64(f);
1157 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
1158 if (!strncmp(id, block->idstr, sizeof(id))) {
1159 if (length != block->used_length) {
1160 Error *local_err = NULL;
1162 ret = qemu_ram_resize(block->offset, length, &local_err);
1164 error_report_err(local_err);
1172 error_report("Unknown ramblock \"%s\", cannot "
1173 "accept migration", id);
1177 total_ram_bytes -= length;
1180 case RAM_SAVE_FLAG_COMPRESS:
1181 host = host_from_stream_offset(f, addr, flags);
1183 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr);
1187 ch = qemu_get_byte(f);
1188 ram_handle_compressed(host, ch, TARGET_PAGE_SIZE);
1190 case RAM_SAVE_FLAG_PAGE:
1191 host = host_from_stream_offset(f, addr, flags);
1193 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr);
1197 qemu_get_buffer(f, host, TARGET_PAGE_SIZE);
1199 case RAM_SAVE_FLAG_XBZRLE:
1200 host = host_from_stream_offset(f, addr, flags);
1202 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr);
1206 if (load_xbzrle(f, addr, host) < 0) {
1207 error_report("Failed to decompress XBZRLE page at "
1208 RAM_ADDR_FMT, addr);
1213 case RAM_SAVE_FLAG_EOS:
1217 if (flags & RAM_SAVE_FLAG_HOOK) {
1218 ram_control_load_hook(f, flags);
1220 error_report("Unknown combination of migration flags: %#x",
1226 ret = qemu_file_get_error(f);
1231 DPRINTF("Completed load of VM with exit code %d seq iteration "
1232 "%" PRIu64 "\n", ret, seq_iter);
1236 static SaveVMHandlers savevm_ram_handlers = {
1237 .save_live_setup = ram_save_setup,
1238 .save_live_iterate = ram_save_iterate,
1239 .save_live_complete = ram_save_complete,
1240 .save_live_pending = ram_save_pending,
1241 .load_state = ram_load,
1242 .cancel = ram_migration_cancel,
1245 void ram_mig_init(void)
1247 qemu_mutex_init(&XBZRLE.lock);
1248 register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, NULL);
1257 int (*init_isa) (ISABus *bus);
1258 int (*init_pci) (PCIBus *bus);
1262 static struct soundhw soundhw[9];
1263 static int soundhw_count;
1265 void isa_register_soundhw(const char *name, const char *descr,
1266 int (*init_isa)(ISABus *bus))
1268 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1269 soundhw[soundhw_count].name = name;
1270 soundhw[soundhw_count].descr = descr;
1271 soundhw[soundhw_count].isa = 1;
1272 soundhw[soundhw_count].init.init_isa = init_isa;
1276 void pci_register_soundhw(const char *name, const char *descr,
1277 int (*init_pci)(PCIBus *bus))
1279 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1280 soundhw[soundhw_count].name = name;
1281 soundhw[soundhw_count].descr = descr;
1282 soundhw[soundhw_count].isa = 0;
1283 soundhw[soundhw_count].init.init_pci = init_pci;
1287 void select_soundhw(const char *optarg)
1291 if (is_help_option(optarg)) {
1294 if (soundhw_count) {
1295 printf("Valid sound card names (comma separated):\n");
1296 for (c = soundhw; c->name; ++c) {
1297 printf ("%-11s %s\n", c->name, c->descr);
1299 printf("\n-soundhw all will enable all of the above\n");
1301 printf("Machine has no user-selectable audio hardware "
1302 "(it may or may not have always-present audio hardware).\n");
1304 exit(!is_help_option(optarg));
1312 if (!strcmp(optarg, "all")) {
1313 for (c = soundhw; c->name; ++c) {
1322 l = !e ? strlen(p) : (size_t) (e - p);
1324 for (c = soundhw; c->name; ++c) {
1325 if (!strncmp(c->name, p, l) && !c->name[l]) {
1333 error_report("Unknown sound card name (too big to show)");
1336 error_report("Unknown sound card name `%.*s'",
1341 p += l + (e != NULL);
1345 goto show_valid_cards;
1350 void audio_init(void)
1353 ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL);
1354 PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL);
1356 for (c = soundhw; c->name; ++c) {
1360 error_report("ISA bus not available for %s", c->name);
1363 c->init.init_isa(isa_bus);
1366 error_report("PCI bus not available for %s", c->name);
1369 c->init.init_pci(pci_bus);
1375 int qemu_uuid_parse(const char *str, uint8_t *uuid)
1379 if (strlen(str) != 36) {
1383 ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3],
1384 &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9],
1385 &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14],
1394 void do_acpitable_option(const QemuOpts *opts)
1399 acpi_table_add(opts, &err);
1401 error_report("Wrong acpi table provided: %s",
1402 error_get_pretty(err));
1409 void do_smbios_option(QemuOpts *opts)
1412 smbios_entry_add(opts);
1416 void cpudef_init(void)
1418 #if defined(cpudef_setup)
1419 cpudef_setup(); /* parse cpu definitions in target config file */
1423 int kvm_available(void)
1432 int xen_available(void)
1442 TargetInfo *qmp_query_target(Error **errp)
1444 TargetInfo *info = g_malloc0(sizeof(*info));
1446 info->arch = g_strdup(TARGET_NAME);
1451 /* Stub function that's gets run on the vcpu when its brought out of the
1452 VM to run inside qemu via async_run_on_cpu()*/
1453 static void mig_sleep_cpu(void *opq)
1455 qemu_mutex_unlock_iothread();
1457 qemu_mutex_lock_iothread();
1460 /* To reduce the dirty rate explicitly disallow the VCPUs from spending
1461 much time in the VM. The migration thread will try to catchup.
1462 Workload will experience a performance drop.
1464 static void mig_throttle_guest_down(void)
1468 qemu_mutex_lock_iothread();
1470 async_run_on_cpu(cpu, mig_sleep_cpu, NULL);
1472 qemu_mutex_unlock_iothread();
1475 static void check_guest_throttling(void)
1480 if (!mig_throttle_on) {
1485 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1489 t1 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1491 /* If it has been more than 40 ms since the last time the guest
1492 * was throttled then do it again.
1494 if (40 < (t1-t0)/1000000) {
1495 mig_throttle_guest_down();