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 static size_t save_block_hdr(QEMUFile *f, RAMBlock *block, ram_addr_t offset,
313 qemu_put_be64(f, offset | cont | flag);
317 qemu_put_byte(f, strlen(block->idstr));
318 qemu_put_buffer(f, (uint8_t *)block->idstr,
319 strlen(block->idstr));
320 size += 1 + strlen(block->idstr);
325 /* This is the last block that we have visited serching for dirty pages
327 static RAMBlock *last_seen_block;
328 /* This is the last block from where we have sent data */
329 static RAMBlock *last_sent_block;
330 static ram_addr_t last_offset;
331 static unsigned long *migration_bitmap;
332 static uint64_t migration_dirty_pages;
333 static uint32_t last_version;
334 static bool ram_bulk_stage;
336 /* Update the xbzrle cache to reflect a page that's been sent as all 0.
337 * The important thing is that a stale (not-yet-0'd) page be replaced
339 * As a bonus, if the page wasn't in the cache it gets added so that
340 * when a small write is made into the 0'd page it gets XBZRLE sent
342 static void xbzrle_cache_zero_page(ram_addr_t current_addr)
344 if (ram_bulk_stage || !migrate_use_xbzrle()) {
348 /* We don't care if this fails to allocate a new cache page
349 * as long as it updated an old one */
350 cache_insert(XBZRLE.cache, current_addr, ZERO_TARGET_PAGE,
354 #define ENCODING_FLAG_XBZRLE 0x1
356 static int save_xbzrle_page(QEMUFile *f, uint8_t **current_data,
357 ram_addr_t current_addr, RAMBlock *block,
358 ram_addr_t offset, int cont, bool last_stage)
360 int encoded_len = 0, bytes_sent = -1;
361 uint8_t *prev_cached_page;
363 if (!cache_is_cached(XBZRLE.cache, current_addr, bitmap_sync_count)) {
364 acct_info.xbzrle_cache_miss++;
366 if (cache_insert(XBZRLE.cache, current_addr, *current_data,
367 bitmap_sync_count) == -1) {
370 /* update *current_data when the page has been
371 inserted into cache */
372 *current_data = get_cached_data(XBZRLE.cache, current_addr);
378 prev_cached_page = get_cached_data(XBZRLE.cache, current_addr);
380 /* save current buffer into memory */
381 memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE);
383 /* XBZRLE encoding (if there is no overflow) */
384 encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf,
385 TARGET_PAGE_SIZE, XBZRLE.encoded_buf,
387 if (encoded_len == 0) {
388 DPRINTF("Skipping unmodified page\n");
390 } else if (encoded_len == -1) {
391 DPRINTF("Overflow\n");
392 acct_info.xbzrle_overflows++;
393 /* update data in the cache */
395 memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE);
396 *current_data = prev_cached_page;
401 /* we need to update the data in the cache, in order to get the same data */
403 memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE);
406 /* Send XBZRLE based compressed page */
407 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_XBZRLE);
408 qemu_put_byte(f, ENCODING_FLAG_XBZRLE);
409 qemu_put_be16(f, encoded_len);
410 qemu_put_buffer(f, XBZRLE.encoded_buf, encoded_len);
411 bytes_sent += encoded_len + 1 + 2;
412 acct_info.xbzrle_pages++;
413 acct_info.xbzrle_bytes += bytes_sent;
419 ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr,
422 unsigned long base = mr->ram_addr >> TARGET_PAGE_BITS;
423 unsigned long nr = base + (start >> TARGET_PAGE_BITS);
424 uint64_t mr_size = TARGET_PAGE_ALIGN(memory_region_size(mr));
425 unsigned long size = base + (mr_size >> TARGET_PAGE_BITS);
429 if (ram_bulk_stage && nr > base) {
432 next = find_next_bit(migration_bitmap, size, nr);
436 clear_bit(next, migration_bitmap);
437 migration_dirty_pages--;
439 return (next - base) << TARGET_PAGE_BITS;
442 static inline bool migration_bitmap_set_dirty(ram_addr_t addr)
445 int nr = addr >> TARGET_PAGE_BITS;
447 ret = test_and_set_bit(nr, migration_bitmap);
450 migration_dirty_pages++;
455 static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length)
458 unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
460 /* start address is aligned at the start of a word? */
461 if (((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) {
463 int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS);
464 unsigned long *src = ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION];
466 for (k = page; k < page + nr; k++) {
468 unsigned long new_dirty;
469 new_dirty = ~migration_bitmap[k];
470 migration_bitmap[k] |= src[k];
472 migration_dirty_pages += ctpopl(new_dirty);
477 for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) {
478 if (cpu_physical_memory_get_dirty(start + addr,
480 DIRTY_MEMORY_MIGRATION)) {
481 cpu_physical_memory_reset_dirty(start + addr,
483 DIRTY_MEMORY_MIGRATION);
484 migration_bitmap_set_dirty(start + addr);
491 /* Fix me: there are too many global variables used in migration process. */
492 static int64_t start_time;
493 static int64_t bytes_xfer_prev;
494 static int64_t num_dirty_pages_period;
496 static void migration_bitmap_sync_init(void)
500 num_dirty_pages_period = 0;
503 /* Called with iothread lock held, to protect ram_list.dirty_memory[] */
504 static void migration_bitmap_sync(void)
507 uint64_t num_dirty_pages_init = migration_dirty_pages;
508 MigrationState *s = migrate_get_current();
510 int64_t bytes_xfer_now;
511 static uint64_t xbzrle_cache_miss_prev;
512 static uint64_t iterations_prev;
516 if (!bytes_xfer_prev) {
517 bytes_xfer_prev = ram_bytes_transferred();
521 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
524 trace_migration_bitmap_sync_start();
525 address_space_sync_dirty_bitmap(&address_space_memory);
528 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
529 migration_bitmap_sync_range(block->mr->ram_addr, block->used_length);
533 trace_migration_bitmap_sync_end(migration_dirty_pages
534 - num_dirty_pages_init);
535 num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init;
536 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
538 /* more than 1 second = 1000 millisecons */
539 if (end_time > start_time + 1000) {
540 if (migrate_auto_converge()) {
541 /* The following detection logic can be refined later. For now:
542 Check to see if the dirtied bytes is 50% more than the approx.
543 amount of bytes that just got transferred since the last time we
544 were in this routine. If that happens >N times (for now N==4)
545 we turn on the throttle down logic */
546 bytes_xfer_now = ram_bytes_transferred();
547 if (s->dirty_pages_rate &&
548 (num_dirty_pages_period * TARGET_PAGE_SIZE >
549 (bytes_xfer_now - bytes_xfer_prev)/2) &&
550 (dirty_rate_high_cnt++ > 4)) {
551 trace_migration_throttle();
552 mig_throttle_on = true;
553 dirty_rate_high_cnt = 0;
555 bytes_xfer_prev = bytes_xfer_now;
557 mig_throttle_on = false;
559 if (migrate_use_xbzrle()) {
560 if (iterations_prev != 0) {
561 acct_info.xbzrle_cache_miss_rate =
562 (double)(acct_info.xbzrle_cache_miss -
563 xbzrle_cache_miss_prev) /
564 (acct_info.iterations - iterations_prev);
566 iterations_prev = acct_info.iterations;
567 xbzrle_cache_miss_prev = acct_info.xbzrle_cache_miss;
569 s->dirty_pages_rate = num_dirty_pages_period * 1000
570 / (end_time - start_time);
571 s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE;
572 start_time = end_time;
573 num_dirty_pages_period = 0;
574 s->dirty_sync_count = bitmap_sync_count;
579 * ram_save_page: Send the given page to the stream
581 * Returns: Number of bytes written.
583 static int ram_save_page(QEMUFile *f, RAMBlock* block, ram_addr_t offset,
588 ram_addr_t current_addr;
589 MemoryRegion *mr = block->mr;
592 bool send_async = true;
594 cont = (block == last_sent_block) ? RAM_SAVE_FLAG_CONTINUE : 0;
596 p = memory_region_get_ram_ptr(mr) + offset;
598 /* In doubt sent page as normal */
600 ret = ram_control_save_page(f, block->offset,
601 offset, TARGET_PAGE_SIZE, &bytes_sent);
605 current_addr = block->offset + offset;
606 if (ret != RAM_SAVE_CONTROL_NOT_SUPP) {
607 if (ret != RAM_SAVE_CONTROL_DELAYED) {
608 if (bytes_sent > 0) {
609 acct_info.norm_pages++;
610 } else if (bytes_sent == 0) {
611 acct_info.dup_pages++;
614 } else if (is_zero_range(p, TARGET_PAGE_SIZE)) {
615 acct_info.dup_pages++;
616 bytes_sent = save_block_hdr(f, block, offset, cont,
617 RAM_SAVE_FLAG_COMPRESS);
620 /* Must let xbzrle know, otherwise a previous (now 0'd) cached
621 * page would be stale
623 xbzrle_cache_zero_page(current_addr);
624 } else if (!ram_bulk_stage && migrate_use_xbzrle()) {
625 bytes_sent = save_xbzrle_page(f, &p, current_addr, block,
626 offset, cont, last_stage);
628 /* Can't send this cached data async, since the cache page
629 * might get updated before it gets to the wire
635 /* XBZRLE overflow or normal page */
636 if (bytes_sent == -1) {
637 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_PAGE);
639 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE);
641 qemu_put_buffer(f, p, TARGET_PAGE_SIZE);
643 bytes_sent += TARGET_PAGE_SIZE;
644 acct_info.norm_pages++;
647 XBZRLE_cache_unlock();
653 * ram_find_and_save_block: Finds a page to send and sends it to f
655 * Called within an RCU critical section.
657 * Returns: The number of bytes written.
658 * 0 means no dirty pages
661 static int ram_find_and_save_block(QEMUFile *f, bool last_stage)
663 RAMBlock *block = last_seen_block;
664 ram_addr_t offset = last_offset;
665 bool complete_round = false;
670 block = QLIST_FIRST_RCU(&ram_list.blocks);
674 offset = migration_bitmap_find_and_reset_dirty(mr, offset);
675 if (complete_round && block == last_seen_block &&
676 offset >= last_offset) {
679 if (offset >= block->used_length) {
681 block = QLIST_NEXT_RCU(block, next);
683 block = QLIST_FIRST_RCU(&ram_list.blocks);
684 complete_round = true;
685 ram_bulk_stage = false;
688 bytes_sent = ram_save_page(f, block, offset, last_stage);
690 /* if page is unmodified, continue to the next */
691 if (bytes_sent > 0) {
692 last_sent_block = block;
698 last_seen_block = block;
699 last_offset = offset;
703 static uint64_t bytes_transferred;
705 void acct_update_position(QEMUFile *f, size_t size, bool zero)
707 uint64_t pages = size / TARGET_PAGE_SIZE;
709 acct_info.dup_pages += pages;
711 acct_info.norm_pages += pages;
712 bytes_transferred += size;
713 qemu_update_position(f, size);
717 static ram_addr_t ram_save_remaining(void)
719 return migration_dirty_pages;
722 uint64_t ram_bytes_remaining(void)
724 return ram_save_remaining() * TARGET_PAGE_SIZE;
727 uint64_t ram_bytes_transferred(void)
729 return bytes_transferred;
732 uint64_t ram_bytes_total(void)
738 QLIST_FOREACH_RCU(block, &ram_list.blocks, next)
739 total += block->used_length;
744 void free_xbzrle_decoded_buf(void)
746 g_free(xbzrle_decoded_buf);
747 xbzrle_decoded_buf = NULL;
750 static void migration_end(void)
752 if (migration_bitmap) {
753 memory_global_dirty_log_stop();
754 g_free(migration_bitmap);
755 migration_bitmap = NULL;
760 cache_fini(XBZRLE.cache);
761 g_free(XBZRLE.encoded_buf);
762 g_free(XBZRLE.current_buf);
764 XBZRLE.encoded_buf = NULL;
765 XBZRLE.current_buf = NULL;
767 XBZRLE_cache_unlock();
770 static void ram_migration_cancel(void *opaque)
775 static void reset_ram_globals(void)
777 last_seen_block = NULL;
778 last_sent_block = NULL;
780 last_version = ram_list.version;
781 ram_bulk_stage = true;
784 #define MAX_WAIT 50 /* ms, half buffered_file limit */
787 /* Each of ram_save_setup, ram_save_iterate and ram_save_complete has
788 * long-running RCU critical section. When rcu-reclaims in the code
789 * start to become numerous it will be necessary to reduce the
790 * granularity of these critical sections.
793 static int ram_save_setup(QEMUFile *f, void *opaque)
796 int64_t ram_bitmap_pages; /* Size of bitmap in pages, including gaps */
798 mig_throttle_on = false;
799 dirty_rate_high_cnt = 0;
800 bitmap_sync_count = 0;
801 migration_bitmap_sync_init();
803 if (migrate_use_xbzrle()) {
805 XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() /
809 XBZRLE_cache_unlock();
810 error_report("Error creating cache");
813 XBZRLE_cache_unlock();
815 /* We prefer not to abort if there is no memory */
816 XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE);
817 if (!XBZRLE.encoded_buf) {
818 error_report("Error allocating encoded_buf");
822 XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE);
823 if (!XBZRLE.current_buf) {
824 error_report("Error allocating current_buf");
825 g_free(XBZRLE.encoded_buf);
826 XBZRLE.encoded_buf = NULL;
833 /* iothread lock needed for ram_list.dirty_memory[] */
834 qemu_mutex_lock_iothread();
835 qemu_mutex_lock_ramlist();
837 bytes_transferred = 0;
840 ram_bitmap_pages = last_ram_offset() >> TARGET_PAGE_BITS;
841 migration_bitmap = bitmap_new(ram_bitmap_pages);
842 bitmap_set(migration_bitmap, 0, ram_bitmap_pages);
845 * Count the total number of pages used by ram blocks not including any
846 * gaps due to alignment or unplugs.
848 migration_dirty_pages = 0;
849 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
850 uint64_t block_pages;
852 block_pages = block->used_length >> TARGET_PAGE_BITS;
853 migration_dirty_pages += block_pages;
856 memory_global_dirty_log_start();
857 migration_bitmap_sync();
858 qemu_mutex_unlock_ramlist();
859 qemu_mutex_unlock_iothread();
861 qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE);
863 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
864 qemu_put_byte(f, strlen(block->idstr));
865 qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr));
866 qemu_put_be64(f, block->used_length);
871 ram_control_before_iterate(f, RAM_CONTROL_SETUP);
872 ram_control_after_iterate(f, RAM_CONTROL_SETUP);
874 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
879 static int ram_save_iterate(QEMUFile *f, void *opaque)
887 if (ram_list.version != last_version) {
891 /* Read version before ram_list.blocks */
894 ram_control_before_iterate(f, RAM_CONTROL_ROUND);
896 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
898 while ((ret = qemu_file_rate_limit(f)) == 0) {
901 bytes_sent = ram_find_and_save_block(f, false);
902 /* no more blocks to sent */
903 if (bytes_sent == 0) {
906 total_sent += bytes_sent;
907 acct_info.iterations++;
908 check_guest_throttling();
909 /* we want to check in the 1st loop, just in case it was the 1st time
910 and we had to sync the dirty bitmap.
911 qemu_get_clock_ns() is a bit expensive, so we only check each some
915 uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000;
917 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n",
927 * Must occur before EOS (or any QEMUFile operation)
928 * because of RDMA protocol.
930 ram_control_after_iterate(f, RAM_CONTROL_ROUND);
932 bytes_transferred += total_sent;
935 * Do not count these 8 bytes into total_sent, so that we can
936 * return 0 if no page had been dirtied.
938 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
939 bytes_transferred += 8;
941 ret = qemu_file_get_error(f);
949 /* Called with iothread lock */
950 static int ram_save_complete(QEMUFile *f, void *opaque)
954 migration_bitmap_sync();
956 ram_control_before_iterate(f, RAM_CONTROL_FINISH);
958 /* try transferring iterative blocks of memory */
960 /* flush all remaining blocks regardless of rate limiting */
964 bytes_sent = ram_find_and_save_block(f, true);
965 /* no more blocks to sent */
966 if (bytes_sent == 0) {
969 bytes_transferred += bytes_sent;
972 ram_control_after_iterate(f, RAM_CONTROL_FINISH);
976 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
981 static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
983 uint64_t remaining_size;
985 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
987 if (remaining_size < max_size) {
988 qemu_mutex_lock_iothread();
990 migration_bitmap_sync();
992 qemu_mutex_unlock_iothread();
993 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
995 return remaining_size;
998 static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host)
1000 unsigned int xh_len;
1003 if (!xbzrle_decoded_buf) {
1004 xbzrle_decoded_buf = g_malloc(TARGET_PAGE_SIZE);
1007 /* extract RLE header */
1008 xh_flags = qemu_get_byte(f);
1009 xh_len = qemu_get_be16(f);
1011 if (xh_flags != ENCODING_FLAG_XBZRLE) {
1012 error_report("Failed to load XBZRLE page - wrong compression!");
1016 if (xh_len > TARGET_PAGE_SIZE) {
1017 error_report("Failed to load XBZRLE page - len overflow!");
1020 /* load data and decode */
1021 qemu_get_buffer(f, xbzrle_decoded_buf, xh_len);
1024 if (xbzrle_decode_buffer(xbzrle_decoded_buf, xh_len, host,
1025 TARGET_PAGE_SIZE) == -1) {
1026 error_report("Failed to load XBZRLE page - decode error!");
1033 /* Must be called from within a rcu critical section.
1034 * Returns a pointer from within the RCU-protected ram_list.
1036 static inline void *host_from_stream_offset(QEMUFile *f,
1040 static RAMBlock *block = NULL;
1044 if (flags & RAM_SAVE_FLAG_CONTINUE) {
1045 if (!block || block->max_length <= offset) {
1046 error_report("Ack, bad migration stream!");
1050 return memory_region_get_ram_ptr(block->mr) + offset;
1053 len = qemu_get_byte(f);
1054 qemu_get_buffer(f, (uint8_t *)id, len);
1057 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
1058 if (!strncmp(id, block->idstr, sizeof(id)) &&
1059 block->max_length > offset) {
1060 return memory_region_get_ram_ptr(block->mr) + offset;
1064 error_report("Can't find block %s!", id);
1069 * If a page (or a whole RDMA chunk) has been
1070 * determined to be zero, then zap it.
1072 void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
1074 if (ch != 0 || !is_zero_range(host, size)) {
1075 memset(host, ch, size);
1079 static int ram_load(QEMUFile *f, void *opaque, int version_id)
1081 int flags = 0, ret = 0;
1082 static uint64_t seq_iter;
1086 if (version_id != 4) {
1090 /* This RCU critical section can be very long running.
1091 * When RCU reclaims in the code start to become numerous,
1092 * it will be necessary to reduce the granularity of this
1096 while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) {
1097 ram_addr_t addr, total_ram_bytes;
1101 addr = qemu_get_be64(f);
1102 flags = addr & ~TARGET_PAGE_MASK;
1103 addr &= TARGET_PAGE_MASK;
1105 switch (flags & ~RAM_SAVE_FLAG_CONTINUE) {
1106 case RAM_SAVE_FLAG_MEM_SIZE:
1107 /* Synchronize RAM block list */
1108 total_ram_bytes = addr;
1109 while (!ret && total_ram_bytes) {
1115 len = qemu_get_byte(f);
1116 qemu_get_buffer(f, (uint8_t *)id, len);
1118 length = qemu_get_be64(f);
1120 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
1121 if (!strncmp(id, block->idstr, sizeof(id))) {
1122 if (length != block->used_length) {
1123 Error *local_err = NULL;
1125 ret = qemu_ram_resize(block->offset, length, &local_err);
1127 error_report("%s", error_get_pretty(local_err));
1128 error_free(local_err);
1136 error_report("Unknown ramblock \"%s\", cannot "
1137 "accept migration", id);
1141 total_ram_bytes -= length;
1144 case RAM_SAVE_FLAG_COMPRESS:
1145 host = host_from_stream_offset(f, addr, flags);
1147 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr);
1151 ch = qemu_get_byte(f);
1152 ram_handle_compressed(host, ch, TARGET_PAGE_SIZE);
1154 case RAM_SAVE_FLAG_PAGE:
1155 host = host_from_stream_offset(f, addr, flags);
1157 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr);
1161 qemu_get_buffer(f, host, TARGET_PAGE_SIZE);
1163 case RAM_SAVE_FLAG_XBZRLE:
1164 host = host_from_stream_offset(f, addr, flags);
1166 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr);
1170 if (load_xbzrle(f, addr, host) < 0) {
1171 error_report("Failed to decompress XBZRLE page at "
1172 RAM_ADDR_FMT, addr);
1177 case RAM_SAVE_FLAG_EOS:
1181 if (flags & RAM_SAVE_FLAG_HOOK) {
1182 ram_control_load_hook(f, flags);
1184 error_report("Unknown combination of migration flags: %#x",
1190 ret = qemu_file_get_error(f);
1195 DPRINTF("Completed load of VM with exit code %d seq iteration "
1196 "%" PRIu64 "\n", ret, seq_iter);
1200 static SaveVMHandlers savevm_ram_handlers = {
1201 .save_live_setup = ram_save_setup,
1202 .save_live_iterate = ram_save_iterate,
1203 .save_live_complete = ram_save_complete,
1204 .save_live_pending = ram_save_pending,
1205 .load_state = ram_load,
1206 .cancel = ram_migration_cancel,
1209 void ram_mig_init(void)
1211 qemu_mutex_init(&XBZRLE.lock);
1212 register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, NULL);
1221 int (*init_isa) (ISABus *bus);
1222 int (*init_pci) (PCIBus *bus);
1226 static struct soundhw soundhw[9];
1227 static int soundhw_count;
1229 void isa_register_soundhw(const char *name, const char *descr,
1230 int (*init_isa)(ISABus *bus))
1232 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1233 soundhw[soundhw_count].name = name;
1234 soundhw[soundhw_count].descr = descr;
1235 soundhw[soundhw_count].isa = 1;
1236 soundhw[soundhw_count].init.init_isa = init_isa;
1240 void pci_register_soundhw(const char *name, const char *descr,
1241 int (*init_pci)(PCIBus *bus))
1243 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1244 soundhw[soundhw_count].name = name;
1245 soundhw[soundhw_count].descr = descr;
1246 soundhw[soundhw_count].isa = 0;
1247 soundhw[soundhw_count].init.init_pci = init_pci;
1251 void select_soundhw(const char *optarg)
1255 if (is_help_option(optarg)) {
1258 if (soundhw_count) {
1259 printf("Valid sound card names (comma separated):\n");
1260 for (c = soundhw; c->name; ++c) {
1261 printf ("%-11s %s\n", c->name, c->descr);
1263 printf("\n-soundhw all will enable all of the above\n");
1265 printf("Machine has no user-selectable audio hardware "
1266 "(it may or may not have always-present audio hardware).\n");
1268 exit(!is_help_option(optarg));
1276 if (!strcmp(optarg, "all")) {
1277 for (c = soundhw; c->name; ++c) {
1286 l = !e ? strlen(p) : (size_t) (e - p);
1288 for (c = soundhw; c->name; ++c) {
1289 if (!strncmp(c->name, p, l) && !c->name[l]) {
1297 error_report("Unknown sound card name (too big to show)");
1300 error_report("Unknown sound card name `%.*s'",
1305 p += l + (e != NULL);
1309 goto show_valid_cards;
1314 void audio_init(void)
1317 ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL);
1318 PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL);
1320 for (c = soundhw; c->name; ++c) {
1324 error_report("ISA bus not available for %s", c->name);
1327 c->init.init_isa(isa_bus);
1330 error_report("PCI bus not available for %s", c->name);
1333 c->init.init_pci(pci_bus);
1339 int qemu_uuid_parse(const char *str, uint8_t *uuid)
1343 if (strlen(str) != 36) {
1347 ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3],
1348 &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9],
1349 &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14],
1358 void do_acpitable_option(const QemuOpts *opts)
1363 acpi_table_add(opts, &err);
1365 error_report("Wrong acpi table provided: %s",
1366 error_get_pretty(err));
1373 void do_smbios_option(QemuOpts *opts)
1376 smbios_entry_add(opts);
1380 void cpudef_init(void)
1382 #if defined(cpudef_setup)
1383 cpudef_setup(); /* parse cpu definitions in target config file */
1387 int kvm_available(void)
1396 int xen_available(void)
1406 TargetInfo *qmp_query_target(Error **errp)
1408 TargetInfo *info = g_malloc0(sizeof(*info));
1410 info->arch = g_strdup(TARGET_NAME);
1415 /* Stub function that's gets run on the vcpu when its brought out of the
1416 VM to run inside qemu via async_run_on_cpu()*/
1417 static void mig_sleep_cpu(void *opq)
1419 qemu_mutex_unlock_iothread();
1421 qemu_mutex_lock_iothread();
1424 /* To reduce the dirty rate explicitly disallow the VCPUs from spending
1425 much time in the VM. The migration thread will try to catchup.
1426 Workload will experience a performance drop.
1428 static void mig_throttle_guest_down(void)
1432 qemu_mutex_lock_iothread();
1434 async_run_on_cpu(cpu, mig_sleep_cpu, NULL);
1436 qemu_mutex_unlock_iothread();
1439 static void check_guest_throttling(void)
1444 if (!mig_throttle_on) {
1449 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1453 t1 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1455 /* If it has been more than 40 ms since the last time the guest
1456 * was throttled then do it again.
1458 if (40 < (t1-t0)/1000000) {
1459 mig_throttle_guest_down();