4 * Copyright (c) 2003-2008 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
28 #include <sys/types.h>
32 #include "monitor/monitor.h"
33 #include "sysemu/sysemu.h"
34 #include "qemu/bitops.h"
35 #include "qemu/bitmap.h"
36 #include "sysemu/arch_init.h"
37 #include "audio/audio.h"
38 #include "hw/i386/pc.h"
39 #include "hw/pci/pci.h"
40 #include "hw/audio/audio.h"
41 #include "sysemu/kvm.h"
42 #include "migration/migration.h"
43 #include "hw/i386/smbios.h"
44 #include "exec/address-spaces.h"
45 #include "hw/audio/pcspk.h"
46 #include "migration/page_cache.h"
47 #include "qemu/config-file.h"
48 #include "qemu/error-report.h"
49 #include "qmp-commands.h"
51 #include "exec/cpu-all.h"
52 #include "exec/ram_addr.h"
53 #include "hw/acpi/acpi.h"
54 #include "qemu/host-utils.h"
56 #ifdef DEBUG_ARCH_INIT
57 #define DPRINTF(fmt, ...) \
58 do { fprintf(stdout, "arch_init: " fmt, ## __VA_ARGS__); } while (0)
60 #define DPRINTF(fmt, ...) \
65 int graphic_width = 1024;
66 int graphic_height = 768;
67 int graphic_depth = 8;
69 int graphic_width = 800;
70 int graphic_height = 600;
71 int graphic_depth = 32;
75 #if defined(TARGET_ALPHA)
76 #define QEMU_ARCH QEMU_ARCH_ALPHA
77 #elif defined(TARGET_ARM)
78 #define QEMU_ARCH QEMU_ARCH_ARM
79 #elif defined(TARGET_CRIS)
80 #define QEMU_ARCH QEMU_ARCH_CRIS
81 #elif defined(TARGET_I386)
82 #define QEMU_ARCH QEMU_ARCH_I386
83 #elif defined(TARGET_M68K)
84 #define QEMU_ARCH QEMU_ARCH_M68K
85 #elif defined(TARGET_LM32)
86 #define QEMU_ARCH QEMU_ARCH_LM32
87 #elif defined(TARGET_MICROBLAZE)
88 #define QEMU_ARCH QEMU_ARCH_MICROBLAZE
89 #elif defined(TARGET_MIPS)
90 #define QEMU_ARCH QEMU_ARCH_MIPS
91 #elif defined(TARGET_MOXIE)
92 #define QEMU_ARCH QEMU_ARCH_MOXIE
93 #elif defined(TARGET_OPENRISC)
94 #define QEMU_ARCH QEMU_ARCH_OPENRISC
95 #elif defined(TARGET_PPC)
96 #define QEMU_ARCH QEMU_ARCH_PPC
97 #elif defined(TARGET_S390X)
98 #define QEMU_ARCH QEMU_ARCH_S390X
99 #elif defined(TARGET_SH4)
100 #define QEMU_ARCH QEMU_ARCH_SH4
101 #elif defined(TARGET_SPARC)
102 #define QEMU_ARCH QEMU_ARCH_SPARC
103 #elif defined(TARGET_XTENSA)
104 #define QEMU_ARCH QEMU_ARCH_XTENSA
105 #elif defined(TARGET_UNICORE32)
106 #define QEMU_ARCH QEMU_ARCH_UNICORE32
109 const uint32_t arch_type = QEMU_ARCH;
110 static bool mig_throttle_on;
111 static int dirty_rate_high_cnt;
112 static void check_guest_throttling(void);
114 /***********************************************************/
115 /* ram save/restore */
117 #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */
118 #define RAM_SAVE_FLAG_COMPRESS 0x02
119 #define RAM_SAVE_FLAG_MEM_SIZE 0x04
120 #define RAM_SAVE_FLAG_PAGE 0x08
121 #define RAM_SAVE_FLAG_EOS 0x10
122 #define RAM_SAVE_FLAG_CONTINUE 0x20
123 #define RAM_SAVE_FLAG_XBZRLE 0x40
124 /* 0x80 is reserved in migration.h start with 0x100 next */
126 static struct defconfig_file {
127 const char *filename;
128 /* Indicates it is an user config file (disabled by -no-user-config) */
130 } default_config_files[] = {
131 { CONFIG_QEMU_CONFDIR "/qemu.conf", true },
132 { CONFIG_QEMU_CONFDIR "/target-" TARGET_NAME ".conf", true },
133 { NULL }, /* end of list */
136 static const uint8_t ZERO_TARGET_PAGE[TARGET_PAGE_SIZE];
138 int qemu_read_default_config_files(bool userconfig)
141 struct defconfig_file *f;
143 for (f = default_config_files; f->filename; f++) {
144 if (!userconfig && f->userconfig) {
147 ret = qemu_read_config_file(f->filename);
148 if (ret < 0 && ret != -ENOENT) {
156 static inline bool is_zero_range(uint8_t *p, uint64_t size)
158 return buffer_find_nonzero_offset(p, size) == size;
161 /* struct contains XBZRLE cache and a static page
162 used by the compression */
164 /* buffer used for XBZRLE encoding */
165 uint8_t *encoded_buf;
166 /* buffer for storing page content */
167 uint8_t *current_buf;
168 /* Cache for XBZRLE, Protected by lock. */
173 /* buffer used for XBZRLE decoding */
174 static uint8_t *xbzrle_decoded_buf;
176 static void XBZRLE_cache_lock(void)
178 if (migrate_use_xbzrle())
179 qemu_mutex_lock(&XBZRLE.lock);
182 static void XBZRLE_cache_unlock(void)
184 if (migrate_use_xbzrle())
185 qemu_mutex_unlock(&XBZRLE.lock);
189 * called from qmp_migrate_set_cache_size in main thread, possibly while
190 * a migration is in progress.
191 * A running migration maybe using the cache and might finish during this
192 * call, hence changes to the cache are protected by XBZRLE.lock().
194 int64_t xbzrle_cache_resize(int64_t new_size)
196 PageCache *new_cache;
199 if (new_size < TARGET_PAGE_SIZE) {
205 if (XBZRLE.cache != NULL) {
206 if (pow2floor(new_size) == migrate_xbzrle_cache_size()) {
209 new_cache = cache_init(new_size / TARGET_PAGE_SIZE,
212 error_report("Error creating cache");
217 cache_fini(XBZRLE.cache);
218 XBZRLE.cache = new_cache;
222 ret = pow2floor(new_size);
224 XBZRLE_cache_unlock();
228 /* accounting for migration statistics */
229 typedef struct AccountingInfo {
231 uint64_t skipped_pages;
234 uint64_t xbzrle_bytes;
235 uint64_t xbzrle_pages;
236 uint64_t xbzrle_cache_miss;
237 uint64_t xbzrle_overflows;
240 static AccountingInfo acct_info;
242 static void acct_clear(void)
244 memset(&acct_info, 0, sizeof(acct_info));
247 uint64_t dup_mig_bytes_transferred(void)
249 return acct_info.dup_pages * TARGET_PAGE_SIZE;
252 uint64_t dup_mig_pages_transferred(void)
254 return acct_info.dup_pages;
257 uint64_t skipped_mig_bytes_transferred(void)
259 return acct_info.skipped_pages * TARGET_PAGE_SIZE;
262 uint64_t skipped_mig_pages_transferred(void)
264 return acct_info.skipped_pages;
267 uint64_t norm_mig_bytes_transferred(void)
269 return acct_info.norm_pages * TARGET_PAGE_SIZE;
272 uint64_t norm_mig_pages_transferred(void)
274 return acct_info.norm_pages;
277 uint64_t xbzrle_mig_bytes_transferred(void)
279 return acct_info.xbzrle_bytes;
282 uint64_t xbzrle_mig_pages_transferred(void)
284 return acct_info.xbzrle_pages;
287 uint64_t xbzrle_mig_pages_cache_miss(void)
289 return acct_info.xbzrle_cache_miss;
292 uint64_t xbzrle_mig_pages_overflow(void)
294 return acct_info.xbzrle_overflows;
297 static size_t save_block_hdr(QEMUFile *f, RAMBlock *block, ram_addr_t offset,
302 qemu_put_be64(f, offset | cont | flag);
306 qemu_put_byte(f, strlen(block->idstr));
307 qemu_put_buffer(f, (uint8_t *)block->idstr,
308 strlen(block->idstr));
309 size += 1 + strlen(block->idstr);
314 /* This is the last block that we have visited serching for dirty pages
316 static RAMBlock *last_seen_block;
317 /* This is the last block from where we have sent data */
318 static RAMBlock *last_sent_block;
319 static ram_addr_t last_offset;
320 static unsigned long *migration_bitmap;
321 static uint64_t migration_dirty_pages;
322 static uint32_t last_version;
323 static bool ram_bulk_stage;
325 /* Update the xbzrle cache to reflect a page that's been sent as all 0.
326 * The important thing is that a stale (not-yet-0'd) page be replaced
328 * As a bonus, if the page wasn't in the cache it gets added so that
329 * when a small write is made into the 0'd page it gets XBZRLE sent
331 static void xbzrle_cache_zero_page(ram_addr_t current_addr)
333 if (ram_bulk_stage || !migrate_use_xbzrle()) {
337 /* We don't care if this fails to allocate a new cache page
338 * as long as it updated an old one */
339 cache_insert(XBZRLE.cache, current_addr, ZERO_TARGET_PAGE);
342 #define ENCODING_FLAG_XBZRLE 0x1
344 static int save_xbzrle_page(QEMUFile *f, uint8_t *current_data,
345 ram_addr_t current_addr, RAMBlock *block,
346 ram_addr_t offset, int cont, bool last_stage)
348 int encoded_len = 0, bytes_sent = -1;
349 uint8_t *prev_cached_page;
351 if (!cache_is_cached(XBZRLE.cache, current_addr)) {
353 if (cache_insert(XBZRLE.cache, current_addr, current_data) == -1) {
357 acct_info.xbzrle_cache_miss++;
361 prev_cached_page = get_cached_data(XBZRLE.cache, current_addr);
363 /* save current buffer into memory */
364 memcpy(XBZRLE.current_buf, current_data, TARGET_PAGE_SIZE);
366 /* XBZRLE encoding (if there is no overflow) */
367 encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf,
368 TARGET_PAGE_SIZE, XBZRLE.encoded_buf,
370 if (encoded_len == 0) {
371 DPRINTF("Skipping unmodified page\n");
373 } else if (encoded_len == -1) {
374 DPRINTF("Overflow\n");
375 acct_info.xbzrle_overflows++;
376 /* update data in the cache */
377 memcpy(prev_cached_page, current_data, TARGET_PAGE_SIZE);
381 /* we need to update the data in the cache, in order to get the same data */
383 memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE);
386 /* Send XBZRLE based compressed page */
387 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_XBZRLE);
388 qemu_put_byte(f, ENCODING_FLAG_XBZRLE);
389 qemu_put_be16(f, encoded_len);
390 qemu_put_buffer(f, XBZRLE.encoded_buf, encoded_len);
391 bytes_sent += encoded_len + 1 + 2;
392 acct_info.xbzrle_pages++;
393 acct_info.xbzrle_bytes += bytes_sent;
399 ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr,
402 unsigned long base = mr->ram_addr >> TARGET_PAGE_BITS;
403 unsigned long nr = base + (start >> TARGET_PAGE_BITS);
404 uint64_t mr_size = TARGET_PAGE_ALIGN(memory_region_size(mr));
405 unsigned long size = base + (mr_size >> TARGET_PAGE_BITS);
409 if (ram_bulk_stage && nr > base) {
412 next = find_next_bit(migration_bitmap, size, nr);
416 clear_bit(next, migration_bitmap);
417 migration_dirty_pages--;
419 return (next - base) << TARGET_PAGE_BITS;
422 static inline bool migration_bitmap_set_dirty(ram_addr_t addr)
425 int nr = addr >> TARGET_PAGE_BITS;
427 ret = test_and_set_bit(nr, migration_bitmap);
430 migration_dirty_pages++;
435 static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length)
438 unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
440 /* start address is aligned at the start of a word? */
441 if (((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) {
443 int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS);
444 unsigned long *src = ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION];
446 for (k = page; k < page + nr; k++) {
448 unsigned long new_dirty;
449 new_dirty = ~migration_bitmap[k];
450 migration_bitmap[k] |= src[k];
452 migration_dirty_pages += ctpopl(new_dirty);
457 for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) {
458 if (cpu_physical_memory_get_dirty(start + addr,
460 DIRTY_MEMORY_MIGRATION)) {
461 cpu_physical_memory_reset_dirty(start + addr,
463 DIRTY_MEMORY_MIGRATION);
464 migration_bitmap_set_dirty(start + addr);
471 /* Needs iothread lock! */
473 static void migration_bitmap_sync(void)
476 uint64_t num_dirty_pages_init = migration_dirty_pages;
477 MigrationState *s = migrate_get_current();
478 static int64_t start_time;
479 static int64_t bytes_xfer_prev;
480 static int64_t num_dirty_pages_period;
482 int64_t bytes_xfer_now;
484 if (!bytes_xfer_prev) {
485 bytes_xfer_prev = ram_bytes_transferred();
489 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
492 trace_migration_bitmap_sync_start();
493 address_space_sync_dirty_bitmap(&address_space_memory);
495 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
496 migration_bitmap_sync_range(block->mr->ram_addr, block->length);
498 trace_migration_bitmap_sync_end(migration_dirty_pages
499 - num_dirty_pages_init);
500 num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init;
501 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
503 /* more than 1 second = 1000 millisecons */
504 if (end_time > start_time + 1000) {
505 if (migrate_auto_converge()) {
506 /* The following detection logic can be refined later. For now:
507 Check to see if the dirtied bytes is 50% more than the approx.
508 amount of bytes that just got transferred since the last time we
509 were in this routine. If that happens >N times (for now N==4)
510 we turn on the throttle down logic */
511 bytes_xfer_now = ram_bytes_transferred();
512 if (s->dirty_pages_rate &&
513 (num_dirty_pages_period * TARGET_PAGE_SIZE >
514 (bytes_xfer_now - bytes_xfer_prev)/2) &&
515 (dirty_rate_high_cnt++ > 4)) {
516 trace_migration_throttle();
517 mig_throttle_on = true;
518 dirty_rate_high_cnt = 0;
520 bytes_xfer_prev = bytes_xfer_now;
522 mig_throttle_on = false;
524 s->dirty_pages_rate = num_dirty_pages_period * 1000
525 / (end_time - start_time);
526 s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE;
527 start_time = end_time;
528 num_dirty_pages_period = 0;
533 * ram_save_block: Writes a page of memory to the stream f
535 * Returns: The number of bytes written.
536 * 0 means no dirty pages
539 static int ram_save_block(QEMUFile *f, bool last_stage)
541 RAMBlock *block = last_seen_block;
542 ram_addr_t offset = last_offset;
543 bool complete_round = false;
546 ram_addr_t current_addr;
549 block = QTAILQ_FIRST(&ram_list.blocks);
553 offset = migration_bitmap_find_and_reset_dirty(mr, offset);
554 if (complete_round && block == last_seen_block &&
555 offset >= last_offset) {
558 if (offset >= block->length) {
560 block = QTAILQ_NEXT(block, next);
562 block = QTAILQ_FIRST(&ram_list.blocks);
563 complete_round = true;
564 ram_bulk_stage = false;
569 bool send_async = true;
570 int cont = (block == last_sent_block) ?
571 RAM_SAVE_FLAG_CONTINUE : 0;
573 p = memory_region_get_ram_ptr(mr) + offset;
575 /* In doubt sent page as normal */
577 ret = ram_control_save_page(f, block->offset,
578 offset, TARGET_PAGE_SIZE, &bytes_sent);
582 current_addr = block->offset + offset;
583 if (ret != RAM_SAVE_CONTROL_NOT_SUPP) {
584 if (ret != RAM_SAVE_CONTROL_DELAYED) {
585 if (bytes_sent > 0) {
586 acct_info.norm_pages++;
587 } else if (bytes_sent == 0) {
588 acct_info.dup_pages++;
591 } else if (is_zero_range(p, TARGET_PAGE_SIZE)) {
592 acct_info.dup_pages++;
593 bytes_sent = save_block_hdr(f, block, offset, cont,
594 RAM_SAVE_FLAG_COMPRESS);
597 /* Must let xbzrle know, otherwise a previous (now 0'd) cached
598 * page would be stale
600 xbzrle_cache_zero_page(current_addr);
601 } else if (!ram_bulk_stage && migrate_use_xbzrle()) {
602 bytes_sent = save_xbzrle_page(f, p, current_addr, block,
603 offset, cont, last_stage);
605 /* We must send exactly what's in the xbzrle cache
606 * even if the page wasn't xbzrle compressed, so that
607 * it's right next time.
609 p = get_cached_data(XBZRLE.cache, current_addr);
611 /* Can't send this cached data async, since the cache page
612 * might get updated before it gets to the wire
618 /* XBZRLE overflow or normal page */
619 if (bytes_sent == -1) {
620 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_PAGE);
622 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE);
624 qemu_put_buffer(f, p, TARGET_PAGE_SIZE);
626 bytes_sent += TARGET_PAGE_SIZE;
627 acct_info.norm_pages++;
630 XBZRLE_cache_unlock();
631 /* if page is unmodified, continue to the next */
632 if (bytes_sent > 0) {
633 last_sent_block = block;
638 last_seen_block = block;
639 last_offset = offset;
644 static uint64_t bytes_transferred;
646 void acct_update_position(QEMUFile *f, size_t size, bool zero)
648 uint64_t pages = size / TARGET_PAGE_SIZE;
650 acct_info.dup_pages += pages;
652 acct_info.norm_pages += pages;
653 bytes_transferred += size;
654 qemu_update_position(f, size);
658 static ram_addr_t ram_save_remaining(void)
660 return migration_dirty_pages;
663 uint64_t ram_bytes_remaining(void)
665 return ram_save_remaining() * TARGET_PAGE_SIZE;
668 uint64_t ram_bytes_transferred(void)
670 return bytes_transferred;
673 uint64_t ram_bytes_total(void)
678 QTAILQ_FOREACH(block, &ram_list.blocks, next)
679 total += block->length;
684 void free_xbzrle_decoded_buf(void)
686 g_free(xbzrle_decoded_buf);
687 xbzrle_decoded_buf = NULL;
690 static void migration_end(void)
692 if (migration_bitmap) {
693 memory_global_dirty_log_stop();
694 g_free(migration_bitmap);
695 migration_bitmap = NULL;
700 cache_fini(XBZRLE.cache);
701 g_free(XBZRLE.cache);
702 g_free(XBZRLE.encoded_buf);
703 g_free(XBZRLE.current_buf);
705 XBZRLE.encoded_buf = NULL;
706 XBZRLE.current_buf = NULL;
708 XBZRLE_cache_unlock();
711 static void ram_migration_cancel(void *opaque)
716 static void reset_ram_globals(void)
718 last_seen_block = NULL;
719 last_sent_block = NULL;
721 last_version = ram_list.version;
722 ram_bulk_stage = true;
725 #define MAX_WAIT 50 /* ms, half buffered_file limit */
727 static int ram_save_setup(QEMUFile *f, void *opaque)
730 int64_t ram_bitmap_pages; /* Size of bitmap in pages, including gaps */
732 mig_throttle_on = false;
733 dirty_rate_high_cnt = 0;
735 if (migrate_use_xbzrle()) {
737 XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() /
741 XBZRLE_cache_unlock();
742 error_report("Error creating cache");
745 XBZRLE_cache_unlock();
747 /* We prefer not to abort if there is no memory */
748 XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE);
749 if (!XBZRLE.encoded_buf) {
750 error_report("Error allocating encoded_buf");
754 XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE);
755 if (!XBZRLE.current_buf) {
756 error_report("Error allocating current_buf");
757 g_free(XBZRLE.encoded_buf);
758 XBZRLE.encoded_buf = NULL;
765 qemu_mutex_lock_iothread();
766 qemu_mutex_lock_ramlist();
767 bytes_transferred = 0;
770 ram_bitmap_pages = last_ram_offset() >> TARGET_PAGE_BITS;
771 migration_bitmap = bitmap_new(ram_bitmap_pages);
772 bitmap_set(migration_bitmap, 0, ram_bitmap_pages);
775 * Count the total number of pages used by ram blocks not including any
776 * gaps due to alignment or unplugs.
778 migration_dirty_pages = 0;
779 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
780 uint64_t block_pages;
782 block_pages = block->length >> TARGET_PAGE_BITS;
783 migration_dirty_pages += block_pages;
786 memory_global_dirty_log_start();
787 migration_bitmap_sync();
788 qemu_mutex_unlock_iothread();
790 qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE);
792 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
793 qemu_put_byte(f, strlen(block->idstr));
794 qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr));
795 qemu_put_be64(f, block->length);
798 qemu_mutex_unlock_ramlist();
800 ram_control_before_iterate(f, RAM_CONTROL_SETUP);
801 ram_control_after_iterate(f, RAM_CONTROL_SETUP);
803 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
808 static int ram_save_iterate(QEMUFile *f, void *opaque)
815 qemu_mutex_lock_ramlist();
817 if (ram_list.version != last_version) {
821 ram_control_before_iterate(f, RAM_CONTROL_ROUND);
823 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
825 while ((ret = qemu_file_rate_limit(f)) == 0) {
828 bytes_sent = ram_save_block(f, false);
829 /* no more blocks to sent */
830 if (bytes_sent == 0) {
833 total_sent += bytes_sent;
834 acct_info.iterations++;
835 check_guest_throttling();
836 /* we want to check in the 1st loop, just in case it was the 1st time
837 and we had to sync the dirty bitmap.
838 qemu_get_clock_ns() is a bit expensive, so we only check each some
842 uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000;
844 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n",
852 qemu_mutex_unlock_ramlist();
855 * Must occur before EOS (or any QEMUFile operation)
856 * because of RDMA protocol.
858 ram_control_after_iterate(f, RAM_CONTROL_ROUND);
860 bytes_transferred += total_sent;
863 * Do not count these 8 bytes into total_sent, so that we can
864 * return 0 if no page had been dirtied.
866 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
867 bytes_transferred += 8;
869 ret = qemu_file_get_error(f);
877 static int ram_save_complete(QEMUFile *f, void *opaque)
879 qemu_mutex_lock_ramlist();
880 migration_bitmap_sync();
882 ram_control_before_iterate(f, RAM_CONTROL_FINISH);
884 /* try transferring iterative blocks of memory */
886 /* flush all remaining blocks regardless of rate limiting */
890 bytes_sent = ram_save_block(f, true);
891 /* no more blocks to sent */
892 if (bytes_sent == 0) {
895 bytes_transferred += bytes_sent;
898 ram_control_after_iterate(f, RAM_CONTROL_FINISH);
901 qemu_mutex_unlock_ramlist();
902 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
907 static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
909 uint64_t remaining_size;
911 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
913 if (remaining_size < max_size) {
914 qemu_mutex_lock_iothread();
915 migration_bitmap_sync();
916 qemu_mutex_unlock_iothread();
917 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
919 return remaining_size;
922 static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host)
928 if (!xbzrle_decoded_buf) {
929 xbzrle_decoded_buf = g_malloc(TARGET_PAGE_SIZE);
932 /* extract RLE header */
933 xh_flags = qemu_get_byte(f);
934 xh_len = qemu_get_be16(f);
936 if (xh_flags != ENCODING_FLAG_XBZRLE) {
937 fprintf(stderr, "Failed to load XBZRLE page - wrong compression!\n");
941 if (xh_len > TARGET_PAGE_SIZE) {
942 fprintf(stderr, "Failed to load XBZRLE page - len overflow!\n");
945 /* load data and decode */
946 qemu_get_buffer(f, xbzrle_decoded_buf, xh_len);
949 ret = xbzrle_decode_buffer(xbzrle_decoded_buf, xh_len, host,
952 fprintf(stderr, "Failed to load XBZRLE page - decode error!\n");
954 } else if (ret > TARGET_PAGE_SIZE) {
955 fprintf(stderr, "Failed to load XBZRLE page - size %d exceeds %d!\n",
956 ret, TARGET_PAGE_SIZE);
963 static inline void *host_from_stream_offset(QEMUFile *f,
967 static RAMBlock *block = NULL;
971 if (flags & RAM_SAVE_FLAG_CONTINUE) {
973 fprintf(stderr, "Ack, bad migration stream!\n");
977 return memory_region_get_ram_ptr(block->mr) + offset;
980 len = qemu_get_byte(f);
981 qemu_get_buffer(f, (uint8_t *)id, len);
984 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
985 if (!strncmp(id, block->idstr, sizeof(id)))
986 return memory_region_get_ram_ptr(block->mr) + offset;
989 fprintf(stderr, "Can't find block %s!\n", id);
994 * If a page (or a whole RDMA chunk) has been
995 * determined to be zero, then zap it.
997 void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
999 if (ch != 0 || !is_zero_range(host, size)) {
1000 memset(host, ch, size);
1004 static int ram_load(QEMUFile *f, void *opaque, int version_id)
1009 static uint64_t seq_iter;
1013 if (version_id != 4) {
1018 addr = qemu_get_be64(f);
1020 flags = addr & ~TARGET_PAGE_MASK;
1021 addr &= TARGET_PAGE_MASK;
1023 if (flags & RAM_SAVE_FLAG_MEM_SIZE) {
1024 /* Synchronize RAM block list */
1027 ram_addr_t total_ram_bytes = addr;
1029 while (total_ram_bytes) {
1033 len = qemu_get_byte(f);
1034 qemu_get_buffer(f, (uint8_t *)id, len);
1036 length = qemu_get_be64(f);
1038 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1039 if (!strncmp(id, block->idstr, sizeof(id))) {
1040 if (block->length != length) {
1042 "Length mismatch: %s: " RAM_ADDR_FMT
1043 " in != " RAM_ADDR_FMT "\n", id, length,
1053 fprintf(stderr, "Unknown ramblock \"%s\", cannot "
1054 "accept migration\n", id);
1059 total_ram_bytes -= length;
1063 if (flags & RAM_SAVE_FLAG_COMPRESS) {
1067 host = host_from_stream_offset(f, addr, flags);
1072 ch = qemu_get_byte(f);
1073 ram_handle_compressed(host, ch, TARGET_PAGE_SIZE);
1074 } else if (flags & RAM_SAVE_FLAG_PAGE) {
1077 host = host_from_stream_offset(f, addr, flags);
1082 qemu_get_buffer(f, host, TARGET_PAGE_SIZE);
1083 } else if (flags & RAM_SAVE_FLAG_XBZRLE) {
1084 void *host = host_from_stream_offset(f, addr, flags);
1089 if (load_xbzrle(f, addr, host) < 0) {
1093 } else if (flags & RAM_SAVE_FLAG_HOOK) {
1094 ram_control_load_hook(f, flags);
1096 error = qemu_file_get_error(f);
1101 } while (!(flags & RAM_SAVE_FLAG_EOS));
1104 DPRINTF("Completed load of VM with exit code %d seq iteration "
1105 "%" PRIu64 "\n", ret, seq_iter);
1109 static SaveVMHandlers savevm_ram_handlers = {
1110 .save_live_setup = ram_save_setup,
1111 .save_live_iterate = ram_save_iterate,
1112 .save_live_complete = ram_save_complete,
1113 .save_live_pending = ram_save_pending,
1114 .load_state = ram_load,
1115 .cancel = ram_migration_cancel,
1118 void ram_mig_init(void)
1120 qemu_mutex_init(&XBZRLE.lock);
1121 register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, NULL);
1130 int (*init_isa) (ISABus *bus);
1131 int (*init_pci) (PCIBus *bus);
1135 static struct soundhw soundhw[9];
1136 static int soundhw_count;
1138 void isa_register_soundhw(const char *name, const char *descr,
1139 int (*init_isa)(ISABus *bus))
1141 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1142 soundhw[soundhw_count].name = name;
1143 soundhw[soundhw_count].descr = descr;
1144 soundhw[soundhw_count].isa = 1;
1145 soundhw[soundhw_count].init.init_isa = init_isa;
1149 void pci_register_soundhw(const char *name, const char *descr,
1150 int (*init_pci)(PCIBus *bus))
1152 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1153 soundhw[soundhw_count].name = name;
1154 soundhw[soundhw_count].descr = descr;
1155 soundhw[soundhw_count].isa = 0;
1156 soundhw[soundhw_count].init.init_pci = init_pci;
1160 void select_soundhw(const char *optarg)
1164 if (is_help_option(optarg)) {
1167 if (soundhw_count) {
1168 printf("Valid sound card names (comma separated):\n");
1169 for (c = soundhw; c->name; ++c) {
1170 printf ("%-11s %s\n", c->name, c->descr);
1172 printf("\n-soundhw all will enable all of the above\n");
1174 printf("Machine has no user-selectable audio hardware "
1175 "(it may or may not have always-present audio hardware).\n");
1177 exit(!is_help_option(optarg));
1185 if (!strcmp(optarg, "all")) {
1186 for (c = soundhw; c->name; ++c) {
1195 l = !e ? strlen(p) : (size_t) (e - p);
1197 for (c = soundhw; c->name; ++c) {
1198 if (!strncmp(c->name, p, l) && !c->name[l]) {
1207 "Unknown sound card name (too big to show)\n");
1210 fprintf(stderr, "Unknown sound card name `%.*s'\n",
1215 p += l + (e != NULL);
1219 goto show_valid_cards;
1224 void audio_init(void)
1227 ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL);
1228 PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL);
1230 for (c = soundhw; c->name; ++c) {
1234 fprintf(stderr, "ISA bus not available for %s\n", c->name);
1237 c->init.init_isa(isa_bus);
1240 fprintf(stderr, "PCI bus not available for %s\n", c->name);
1243 c->init.init_pci(pci_bus);
1249 int qemu_uuid_parse(const char *str, uint8_t *uuid)
1253 if (strlen(str) != 36) {
1257 ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3],
1258 &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9],
1259 &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14],
1268 void do_acpitable_option(const QemuOpts *opts)
1273 acpi_table_add(opts, &err);
1275 error_report("Wrong acpi table provided: %s",
1276 error_get_pretty(err));
1283 void do_smbios_option(QemuOpts *opts)
1286 smbios_entry_add(opts);
1290 void cpudef_init(void)
1292 #if defined(cpudef_setup)
1293 cpudef_setup(); /* parse cpu definitions in target config file */
1297 int tcg_available(void)
1302 int kvm_available(void)
1311 int xen_available(void)
1321 TargetInfo *qmp_query_target(Error **errp)
1323 TargetInfo *info = g_malloc0(sizeof(*info));
1325 info->arch = g_strdup(TARGET_NAME);
1330 /* Stub function that's gets run on the vcpu when its brought out of the
1331 VM to run inside qemu via async_run_on_cpu()*/
1332 static void mig_sleep_cpu(void *opq)
1334 qemu_mutex_unlock_iothread();
1336 qemu_mutex_lock_iothread();
1339 /* To reduce the dirty rate explicitly disallow the VCPUs from spending
1340 much time in the VM. The migration thread will try to catchup.
1341 Workload will experience a performance drop.
1343 static void mig_throttle_guest_down(void)
1347 qemu_mutex_lock_iothread();
1349 async_run_on_cpu(cpu, mig_sleep_cpu, NULL);
1351 qemu_mutex_unlock_iothread();
1354 static void check_guest_throttling(void)
1359 if (!mig_throttle_on) {
1364 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1368 t1 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1370 /* If it has been more than 40 ms since the last time the guest
1371 * was throttled then do it again.
1373 if (40 < (t1-t0)/1000000) {
1374 mig_throttle_guest_down();