Merge remote-tracking branch 'remotes/afaerber/tags/qom-cpu-for-peter' into staging
[qemu.git] / arch_init.c
blobc3f7d3f56d5359a091c465e138b5f14c57fd0f48
1 /*
2 * QEMU System Emulator
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
22 * THE SOFTWARE.
24 #include <stdint.h>
25 #include <stdarg.h>
26 #include <stdlib.h>
27 #ifndef _WIN32
28 #include <sys/types.h>
29 #include <sys/mman.h>
30 #endif
31 #include "config.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"
50 #include "trace.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)
60 #else
61 #define DPRINTF(fmt, ...) \
62 do { } while (0)
63 #endif
65 #ifdef TARGET_SPARC
66 int graphic_width = 1024;
67 int graphic_height = 768;
68 int graphic_depth = 8;
69 #else
70 int graphic_width = 800;
71 int graphic_height = 600;
72 int graphic_depth = 32;
73 #endif
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
110 #endif
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) */
134 bool userconfig;
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)
145 int ret;
146 struct defconfig_file *f;
148 for (f = default_config_files; f->filename; f++) {
149 if (!userconfig && f->userconfig) {
150 continue;
152 ret = qemu_read_config_file(f->filename);
153 if (ret < 0 && ret != -ENOENT) {
154 return ret;
158 return 0;
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 */
168 static struct {
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. */
174 PageCache *cache;
175 QemuMutex lock;
176 } XBZRLE;
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;
202 int64_t ret;
204 if (new_size < TARGET_PAGE_SIZE) {
205 return -1;
208 XBZRLE_cache_lock();
210 if (XBZRLE.cache != NULL) {
211 if (pow2floor(new_size) == migrate_xbzrle_cache_size()) {
212 goto out_new_size;
214 new_cache = cache_init(new_size / TARGET_PAGE_SIZE,
215 TARGET_PAGE_SIZE);
216 if (!new_cache) {
217 error_report("Error creating cache");
218 ret = -1;
219 goto out;
222 cache_fini(XBZRLE.cache);
223 XBZRLE.cache = new_cache;
226 out_new_size:
227 ret = pow2floor(new_size);
228 out:
229 XBZRLE_cache_unlock();
230 return ret;
233 /* accounting for migration statistics */
234 typedef struct AccountingInfo {
235 uint64_t dup_pages;
236 uint64_t skipped_pages;
237 uint64_t norm_pages;
238 uint64_t iterations;
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;
244 } AccountingInfo;
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)
333 size_t size;
335 if (block == last_sent_block) {
336 offset |= RAM_SAVE_FLAG_CONTINUE;
339 qemu_put_be64(f, offset);
340 size = 8;
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;
349 return size;
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
354 * by the new data.
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()) {
361 return;
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,
367 bitmap_sync_count);
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
380 * @current_data:
381 * @current_addr:
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++;
397 if (!last_stage) {
398 if (cache_insert(XBZRLE.cache, current_addr, *current_data,
399 bitmap_sync_count) == -1) {
400 return -1;
401 } else {
402 /* update *current_data when the page has been
403 inserted into cache */
404 *current_data = get_cached_data(XBZRLE.cache, current_addr);
407 return -1;
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,
418 TARGET_PAGE_SIZE);
419 if (encoded_len == 0) {
420 DPRINTF("Skipping unmodified page\n");
421 return 0;
422 } else if (encoded_len == -1) {
423 DPRINTF("Overflow\n");
424 acct_info.xbzrle_overflows++;
425 /* update data in the cache */
426 if (!last_stage) {
427 memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE);
428 *current_data = prev_cached_page;
430 return -1;
433 /* we need to update the data in the cache, in order to get the same data */
434 if (!last_stage) {
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;
448 return 1;
451 static inline
452 ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr,
453 ram_addr_t start)
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);
460 unsigned long next;
462 if (ram_bulk_stage && nr > base) {
463 next = nr + 1;
464 } else {
465 next = find_next_bit(migration_bitmap, size, nr);
468 if (next < size) {
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)
477 bool ret;
478 int nr = addr >> TARGET_PAGE_BITS;
480 ret = test_and_set_bit(nr, migration_bitmap);
482 if (!ret) {
483 migration_dirty_pages++;
485 return ret;
488 static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length)
490 ram_addr_t addr;
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) {
495 int k;
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++) {
500 if (src[k]) {
501 unsigned long new_dirty;
502 new_dirty = ~migration_bitmap[k];
503 migration_bitmap[k] |= src[k];
504 new_dirty &= src[k];
505 migration_dirty_pages += ctpopl(new_dirty);
506 src[k] = 0;
509 } else {
510 for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) {
511 if (cpu_physical_memory_get_dirty(start + addr,
512 TARGET_PAGE_SIZE,
513 DIRTY_MEMORY_MIGRATION)) {
514 cpu_physical_memory_reset_dirty(start + addr,
515 TARGET_PAGE_SIZE,
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)
531 start_time = 0;
532 bytes_xfer_prev = 0;
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)
539 RAMBlock *block;
540 uint64_t num_dirty_pages_init = migration_dirty_pages;
541 MigrationState *s = migrate_get_current();
542 int64_t end_time;
543 int64_t bytes_xfer_now;
544 static uint64_t xbzrle_cache_miss_prev;
545 static uint64_t iterations_prev;
547 bitmap_sync_count++;
549 if (!bytes_xfer_prev) {
550 bytes_xfer_prev = ram_bytes_transferred();
553 if (!start_time) {
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);
560 rcu_read_lock();
561 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
562 migration_bitmap_sync_range(block->mr->ram_addr, block->used_length);
564 rcu_read_unlock();
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;
589 } else {
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)
625 int pages = -1;
626 uint64_t bytes_xmit;
627 ram_addr_t current_addr;
628 MemoryRegion *mr = block->mr;
629 uint8_t *p;
630 int ret;
631 bool send_async = true;
633 p = memory_region_get_ram_ptr(mr) + offset;
635 /* In doubt sent page as normal */
636 bytes_xmit = 0;
637 ret = ram_control_save_page(f, block->offset,
638 offset, TARGET_PAGE_SIZE, &bytes_xmit);
639 if (bytes_xmit) {
640 *bytes_transferred += bytes_xmit;
641 pages = 1;
644 XBZRLE_cache_lock();
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);
659 qemu_put_byte(f, 0);
660 *bytes_transferred += 1;
661 pages = 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);
669 if (!last_stage) {
670 /* Can't send this cached data async, since the cache page
671 * might get updated before it gets to the wire
673 send_async = false;
677 /* XBZRLE overflow or normal page */
678 if (pages == -1) {
679 *bytes_transferred += save_page_header(f, block,
680 offset | RAM_SAVE_FLAG_PAGE);
681 if (send_async) {
682 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE);
683 } else {
684 qemu_put_buffer(f, p, TARGET_PAGE_SIZE);
686 *bytes_transferred += TARGET_PAGE_SIZE;
687 pages = 1;
688 acct_info.norm_pages++;
691 XBZRLE_cache_unlock();
693 return pages;
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;
715 int pages = 0;
716 MemoryRegion *mr;
718 if (!block)
719 block = QLIST_FIRST_RCU(&ram_list.blocks);
721 while (true) {
722 mr = block->mr;
723 offset = migration_bitmap_find_and_reset_dirty(mr, offset);
724 if (complete_round && block == last_seen_block &&
725 offset >= last_offset) {
726 break;
728 if (offset >= block->used_length) {
729 offset = 0;
730 block = QLIST_NEXT_RCU(block, next);
731 if (!block) {
732 block = QLIST_FIRST_RCU(&ram_list.blocks);
733 complete_round = true;
734 ram_bulk_stage = false;
736 } else {
737 pages = ram_save_page(f, block, offset, last_stage,
738 bytes_transferred);
740 /* if page is unmodified, continue to the next */
741 if (pages > 0) {
742 break;
747 last_seen_block = block;
748 last_offset = offset;
750 return pages;
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;
758 if (zero) {
759 acct_info.dup_pages += pages;
760 } else {
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)
784 RAMBlock *block;
785 uint64_t total = 0;
787 rcu_read_lock();
788 QLIST_FOREACH_RCU(block, &ram_list.blocks, next)
789 total += block->used_length;
790 rcu_read_unlock();
791 return total;
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;
808 XBZRLE_cache_lock();
809 if (XBZRLE.cache) {
810 cache_fini(XBZRLE.cache);
811 g_free(XBZRLE.encoded_buf);
812 g_free(XBZRLE.current_buf);
813 XBZRLE.cache = NULL;
814 XBZRLE.encoded_buf = NULL;
815 XBZRLE.current_buf = NULL;
817 XBZRLE_cache_unlock();
820 static void ram_migration_cancel(void *opaque)
822 migration_end();
825 static void reset_ram_globals(void)
827 last_seen_block = NULL;
828 last_sent_block = NULL;
829 last_offset = 0;
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)
845 RAMBlock *block;
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()) {
854 XBZRLE_cache_lock();
855 XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() /
856 TARGET_PAGE_SIZE,
857 TARGET_PAGE_SIZE);
858 if (!XBZRLE.cache) {
859 XBZRLE_cache_unlock();
860 error_report("Error creating cache");
861 return -1;
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");
869 return -1;
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;
877 return -1;
880 acct_clear();
883 /* iothread lock needed for ram_list.dirty_memory[] */
884 qemu_mutex_lock_iothread();
885 qemu_mutex_lock_ramlist();
886 rcu_read_lock();
887 bytes_transferred = 0;
888 reset_ram_globals();
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 = 0;
899 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
900 uint64_t block_pages;
902 block_pages = block->used_length >> TARGET_PAGE_BITS;
903 migration_dirty_pages += block_pages;
906 memory_global_dirty_log_start();
907 migration_bitmap_sync();
908 qemu_mutex_unlock_ramlist();
909 qemu_mutex_unlock_iothread();
911 qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE);
913 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
914 qemu_put_byte(f, strlen(block->idstr));
915 qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr));
916 qemu_put_be64(f, block->used_length);
919 rcu_read_unlock();
921 ram_control_before_iterate(f, RAM_CONTROL_SETUP);
922 ram_control_after_iterate(f, RAM_CONTROL_SETUP);
924 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
926 return 0;
929 static int ram_save_iterate(QEMUFile *f, void *opaque)
931 int ret;
932 int i;
933 int64_t t0;
934 int pages_sent = 0;
936 rcu_read_lock();
937 if (ram_list.version != last_version) {
938 reset_ram_globals();
941 /* Read version before ram_list.blocks */
942 smp_rmb();
944 ram_control_before_iterate(f, RAM_CONTROL_ROUND);
946 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
947 i = 0;
948 while ((ret = qemu_file_rate_limit(f)) == 0) {
949 int pages;
951 pages = ram_find_and_save_block(f, false, &bytes_transferred);
952 /* no more pages to sent */
953 if (pages == 0) {
954 break;
956 pages_sent += pages;
957 acct_info.iterations++;
958 check_guest_throttling();
959 /* we want to check in the 1st loop, just in case it was the 1st time
960 and we had to sync the dirty bitmap.
961 qemu_get_clock_ns() is a bit expensive, so we only check each some
962 iterations
964 if ((i & 63) == 0) {
965 uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000;
966 if (t1 > MAX_WAIT) {
967 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n",
968 t1, i);
969 break;
972 i++;
974 rcu_read_unlock();
977 * Must occur before EOS (or any QEMUFile operation)
978 * because of RDMA protocol.
980 ram_control_after_iterate(f, RAM_CONTROL_ROUND);
982 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
983 bytes_transferred += 8;
985 ret = qemu_file_get_error(f);
986 if (ret < 0) {
987 return ret;
990 return pages_sent;
993 /* Called with iothread lock */
994 static int ram_save_complete(QEMUFile *f, void *opaque)
996 rcu_read_lock();
998 migration_bitmap_sync();
1000 ram_control_before_iterate(f, RAM_CONTROL_FINISH);
1002 /* try transferring iterative blocks of memory */
1004 /* flush all remaining blocks regardless of rate limiting */
1005 while (true) {
1006 int pages;
1008 pages = ram_find_and_save_block(f, true, &bytes_transferred);
1009 /* no more blocks to sent */
1010 if (pages == 0) {
1011 break;
1015 ram_control_after_iterate(f, RAM_CONTROL_FINISH);
1016 migration_end();
1018 rcu_read_unlock();
1019 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
1021 return 0;
1024 static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
1026 uint64_t remaining_size;
1028 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
1030 if (remaining_size < max_size) {
1031 qemu_mutex_lock_iothread();
1032 rcu_read_lock();
1033 migration_bitmap_sync();
1034 rcu_read_unlock();
1035 qemu_mutex_unlock_iothread();
1036 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
1038 return remaining_size;
1041 static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host)
1043 unsigned int xh_len;
1044 int xh_flags;
1046 if (!xbzrle_decoded_buf) {
1047 xbzrle_decoded_buf = g_malloc(TARGET_PAGE_SIZE);
1050 /* extract RLE header */
1051 xh_flags = qemu_get_byte(f);
1052 xh_len = qemu_get_be16(f);
1054 if (xh_flags != ENCODING_FLAG_XBZRLE) {
1055 error_report("Failed to load XBZRLE page - wrong compression!");
1056 return -1;
1059 if (xh_len > TARGET_PAGE_SIZE) {
1060 error_report("Failed to load XBZRLE page - len overflow!");
1061 return -1;
1063 /* load data and decode */
1064 qemu_get_buffer(f, xbzrle_decoded_buf, xh_len);
1066 /* decode RLE */
1067 if (xbzrle_decode_buffer(xbzrle_decoded_buf, xh_len, host,
1068 TARGET_PAGE_SIZE) == -1) {
1069 error_report("Failed to load XBZRLE page - decode error!");
1070 return -1;
1073 return 0;
1076 /* Must be called from within a rcu critical section.
1077 * Returns a pointer from within the RCU-protected ram_list.
1079 static inline void *host_from_stream_offset(QEMUFile *f,
1080 ram_addr_t offset,
1081 int flags)
1083 static RAMBlock *block = NULL;
1084 char id[256];
1085 uint8_t len;
1087 if (flags & RAM_SAVE_FLAG_CONTINUE) {
1088 if (!block || block->max_length <= offset) {
1089 error_report("Ack, bad migration stream!");
1090 return NULL;
1093 return memory_region_get_ram_ptr(block->mr) + offset;
1096 len = qemu_get_byte(f);
1097 qemu_get_buffer(f, (uint8_t *)id, len);
1098 id[len] = 0;
1100 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
1101 if (!strncmp(id, block->idstr, sizeof(id)) &&
1102 block->max_length > offset) {
1103 return memory_region_get_ram_ptr(block->mr) + offset;
1107 error_report("Can't find block %s!", id);
1108 return NULL;
1112 * If a page (or a whole RDMA chunk) has been
1113 * determined to be zero, then zap it.
1115 void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
1117 if (ch != 0 || !is_zero_range(host, size)) {
1118 memset(host, ch, size);
1122 static int ram_load(QEMUFile *f, void *opaque, int version_id)
1124 int flags = 0, ret = 0;
1125 static uint64_t seq_iter;
1127 seq_iter++;
1129 if (version_id != 4) {
1130 ret = -EINVAL;
1133 /* This RCU critical section can be very long running.
1134 * When RCU reclaims in the code start to become numerous,
1135 * it will be necessary to reduce the granularity of this
1136 * critical section.
1138 rcu_read_lock();
1139 while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) {
1140 ram_addr_t addr, total_ram_bytes;
1141 void *host;
1142 uint8_t ch;
1144 addr = qemu_get_be64(f);
1145 flags = addr & ~TARGET_PAGE_MASK;
1146 addr &= TARGET_PAGE_MASK;
1148 switch (flags & ~RAM_SAVE_FLAG_CONTINUE) {
1149 case RAM_SAVE_FLAG_MEM_SIZE:
1150 /* Synchronize RAM block list */
1151 total_ram_bytes = addr;
1152 while (!ret && total_ram_bytes) {
1153 RAMBlock *block;
1154 uint8_t len;
1155 char id[256];
1156 ram_addr_t length;
1158 len = qemu_get_byte(f);
1159 qemu_get_buffer(f, (uint8_t *)id, len);
1160 id[len] = 0;
1161 length = qemu_get_be64(f);
1163 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
1164 if (!strncmp(id, block->idstr, sizeof(id))) {
1165 if (length != block->used_length) {
1166 Error *local_err = NULL;
1168 ret = qemu_ram_resize(block->offset, length, &local_err);
1169 if (local_err) {
1170 error_report_err(local_err);
1173 break;
1177 if (!block) {
1178 error_report("Unknown ramblock \"%s\", cannot "
1179 "accept migration", id);
1180 ret = -EINVAL;
1183 total_ram_bytes -= length;
1185 break;
1186 case RAM_SAVE_FLAG_COMPRESS:
1187 host = host_from_stream_offset(f, addr, flags);
1188 if (!host) {
1189 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr);
1190 ret = -EINVAL;
1191 break;
1193 ch = qemu_get_byte(f);
1194 ram_handle_compressed(host, ch, TARGET_PAGE_SIZE);
1195 break;
1196 case RAM_SAVE_FLAG_PAGE:
1197 host = host_from_stream_offset(f, addr, flags);
1198 if (!host) {
1199 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr);
1200 ret = -EINVAL;
1201 break;
1203 qemu_get_buffer(f, host, TARGET_PAGE_SIZE);
1204 break;
1205 case RAM_SAVE_FLAG_XBZRLE:
1206 host = host_from_stream_offset(f, addr, flags);
1207 if (!host) {
1208 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr);
1209 ret = -EINVAL;
1210 break;
1212 if (load_xbzrle(f, addr, host) < 0) {
1213 error_report("Failed to decompress XBZRLE page at "
1214 RAM_ADDR_FMT, addr);
1215 ret = -EINVAL;
1216 break;
1218 break;
1219 case RAM_SAVE_FLAG_EOS:
1220 /* normal exit */
1221 break;
1222 default:
1223 if (flags & RAM_SAVE_FLAG_HOOK) {
1224 ram_control_load_hook(f, flags);
1225 } else {
1226 error_report("Unknown combination of migration flags: %#x",
1227 flags);
1228 ret = -EINVAL;
1231 if (!ret) {
1232 ret = qemu_file_get_error(f);
1236 rcu_read_unlock();
1237 DPRINTF("Completed load of VM with exit code %d seq iteration "
1238 "%" PRIu64 "\n", ret, seq_iter);
1239 return ret;
1242 static SaveVMHandlers savevm_ram_handlers = {
1243 .save_live_setup = ram_save_setup,
1244 .save_live_iterate = ram_save_iterate,
1245 .save_live_complete = ram_save_complete,
1246 .save_live_pending = ram_save_pending,
1247 .load_state = ram_load,
1248 .cancel = ram_migration_cancel,
1251 void ram_mig_init(void)
1253 qemu_mutex_init(&XBZRLE.lock);
1254 register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, NULL);
1257 struct soundhw {
1258 const char *name;
1259 const char *descr;
1260 int enabled;
1261 int isa;
1262 union {
1263 int (*init_isa) (ISABus *bus);
1264 int (*init_pci) (PCIBus *bus);
1265 } init;
1268 static struct soundhw soundhw[9];
1269 static int soundhw_count;
1271 void isa_register_soundhw(const char *name, const char *descr,
1272 int (*init_isa)(ISABus *bus))
1274 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1275 soundhw[soundhw_count].name = name;
1276 soundhw[soundhw_count].descr = descr;
1277 soundhw[soundhw_count].isa = 1;
1278 soundhw[soundhw_count].init.init_isa = init_isa;
1279 soundhw_count++;
1282 void pci_register_soundhw(const char *name, const char *descr,
1283 int (*init_pci)(PCIBus *bus))
1285 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1286 soundhw[soundhw_count].name = name;
1287 soundhw[soundhw_count].descr = descr;
1288 soundhw[soundhw_count].isa = 0;
1289 soundhw[soundhw_count].init.init_pci = init_pci;
1290 soundhw_count++;
1293 void select_soundhw(const char *optarg)
1295 struct soundhw *c;
1297 if (is_help_option(optarg)) {
1298 show_valid_cards:
1300 if (soundhw_count) {
1301 printf("Valid sound card names (comma separated):\n");
1302 for (c = soundhw; c->name; ++c) {
1303 printf ("%-11s %s\n", c->name, c->descr);
1305 printf("\n-soundhw all will enable all of the above\n");
1306 } else {
1307 printf("Machine has no user-selectable audio hardware "
1308 "(it may or may not have always-present audio hardware).\n");
1310 exit(!is_help_option(optarg));
1312 else {
1313 size_t l;
1314 const char *p;
1315 char *e;
1316 int bad_card = 0;
1318 if (!strcmp(optarg, "all")) {
1319 for (c = soundhw; c->name; ++c) {
1320 c->enabled = 1;
1322 return;
1325 p = optarg;
1326 while (*p) {
1327 e = strchr(p, ',');
1328 l = !e ? strlen(p) : (size_t) (e - p);
1330 for (c = soundhw; c->name; ++c) {
1331 if (!strncmp(c->name, p, l) && !c->name[l]) {
1332 c->enabled = 1;
1333 break;
1337 if (!c->name) {
1338 if (l > 80) {
1339 error_report("Unknown sound card name (too big to show)");
1341 else {
1342 error_report("Unknown sound card name `%.*s'",
1343 (int) l, p);
1345 bad_card = 1;
1347 p += l + (e != NULL);
1350 if (bad_card) {
1351 goto show_valid_cards;
1356 void audio_init(void)
1358 struct soundhw *c;
1359 ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL);
1360 PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL);
1362 for (c = soundhw; c->name; ++c) {
1363 if (c->enabled) {
1364 if (c->isa) {
1365 if (!isa_bus) {
1366 error_report("ISA bus not available for %s", c->name);
1367 exit(1);
1369 c->init.init_isa(isa_bus);
1370 } else {
1371 if (!pci_bus) {
1372 error_report("PCI bus not available for %s", c->name);
1373 exit(1);
1375 c->init.init_pci(pci_bus);
1381 int qemu_uuid_parse(const char *str, uint8_t *uuid)
1383 int ret;
1385 if (strlen(str) != 36) {
1386 return -1;
1389 ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3],
1390 &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9],
1391 &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14],
1392 &uuid[15]);
1394 if (ret != 16) {
1395 return -1;
1397 return 0;
1400 void do_acpitable_option(const QemuOpts *opts)
1402 #ifdef TARGET_I386
1403 Error *err = NULL;
1405 acpi_table_add(opts, &err);
1406 if (err) {
1407 error_report("Wrong acpi table provided: %s",
1408 error_get_pretty(err));
1409 error_free(err);
1410 exit(1);
1412 #endif
1415 void do_smbios_option(QemuOpts *opts)
1417 #ifdef TARGET_I386
1418 smbios_entry_add(opts);
1419 #endif
1422 void cpudef_init(void)
1424 #if defined(cpudef_setup)
1425 cpudef_setup(); /* parse cpu definitions in target config file */
1426 #endif
1429 int kvm_available(void)
1431 #ifdef CONFIG_KVM
1432 return 1;
1433 #else
1434 return 0;
1435 #endif
1438 int xen_available(void)
1440 #ifdef CONFIG_XEN
1441 return 1;
1442 #else
1443 return 0;
1444 #endif
1448 TargetInfo *qmp_query_target(Error **errp)
1450 TargetInfo *info = g_malloc0(sizeof(*info));
1452 info->arch = g_strdup(TARGET_NAME);
1454 return info;
1457 /* Stub function that's gets run on the vcpu when its brought out of the
1458 VM to run inside qemu via async_run_on_cpu()*/
1459 static void mig_sleep_cpu(void *opq)
1461 qemu_mutex_unlock_iothread();
1462 g_usleep(30*1000);
1463 qemu_mutex_lock_iothread();
1466 /* To reduce the dirty rate explicitly disallow the VCPUs from spending
1467 much time in the VM. The migration thread will try to catchup.
1468 Workload will experience a performance drop.
1470 static void mig_throttle_guest_down(void)
1472 CPUState *cpu;
1474 qemu_mutex_lock_iothread();
1475 CPU_FOREACH(cpu) {
1476 async_run_on_cpu(cpu, mig_sleep_cpu, NULL);
1478 qemu_mutex_unlock_iothread();
1481 static void check_guest_throttling(void)
1483 static int64_t t0;
1484 int64_t t1;
1486 if (!mig_throttle_on) {
1487 return;
1490 if (!t0) {
1491 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1492 return;
1495 t1 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1497 /* If it has been more than 40 ms since the last time the guest
1498 * was throttled then do it again.
1500 if (40 < (t1-t0)/1000000) {
1501 mig_throttle_guest_down();
1502 t0 = t1;