]> git.proxmox.com Git - qemu.git/blob - arch_init.c
Force auto-convegence of live migration
[qemu.git] / arch_init.c
1 /*
2 * QEMU System Emulator
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 *
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:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
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.
23 */
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 "qmp-commands.h"
49 #include "trace.h"
50 #include "exec/cpu-all.h"
51 #include "hw/acpi/acpi.h"
52
53 #ifdef DEBUG_ARCH_INIT
54 #define DPRINTF(fmt, ...) \
55 do { fprintf(stdout, "arch_init: " fmt, ## __VA_ARGS__); } while (0)
56 #else
57 #define DPRINTF(fmt, ...) \
58 do { } while (0)
59 #endif
60
61 #ifdef TARGET_SPARC
62 int graphic_width = 1024;
63 int graphic_height = 768;
64 int graphic_depth = 8;
65 #else
66 int graphic_width = 800;
67 int graphic_height = 600;
68 int graphic_depth = 32;
69 #endif
70
71
72 #if defined(TARGET_ALPHA)
73 #define QEMU_ARCH QEMU_ARCH_ALPHA
74 #elif defined(TARGET_ARM)
75 #define QEMU_ARCH QEMU_ARCH_ARM
76 #elif defined(TARGET_CRIS)
77 #define QEMU_ARCH QEMU_ARCH_CRIS
78 #elif defined(TARGET_I386)
79 #define QEMU_ARCH QEMU_ARCH_I386
80 #elif defined(TARGET_M68K)
81 #define QEMU_ARCH QEMU_ARCH_M68K
82 #elif defined(TARGET_LM32)
83 #define QEMU_ARCH QEMU_ARCH_LM32
84 #elif defined(TARGET_MICROBLAZE)
85 #define QEMU_ARCH QEMU_ARCH_MICROBLAZE
86 #elif defined(TARGET_MIPS)
87 #define QEMU_ARCH QEMU_ARCH_MIPS
88 #elif defined(TARGET_MOXIE)
89 #define QEMU_ARCH QEMU_ARCH_MOXIE
90 #elif defined(TARGET_OPENRISC)
91 #define QEMU_ARCH QEMU_ARCH_OPENRISC
92 #elif defined(TARGET_PPC)
93 #define QEMU_ARCH QEMU_ARCH_PPC
94 #elif defined(TARGET_S390X)
95 #define QEMU_ARCH QEMU_ARCH_S390X
96 #elif defined(TARGET_SH4)
97 #define QEMU_ARCH QEMU_ARCH_SH4
98 #elif defined(TARGET_SPARC)
99 #define QEMU_ARCH QEMU_ARCH_SPARC
100 #elif defined(TARGET_XTENSA)
101 #define QEMU_ARCH QEMU_ARCH_XTENSA
102 #elif defined(TARGET_UNICORE32)
103 #define QEMU_ARCH QEMU_ARCH_UNICORE32
104 #endif
105
106 const uint32_t arch_type = QEMU_ARCH;
107 static bool mig_throttle_on;
108 static int dirty_rate_high_cnt;
109 static void check_guest_throttling(void);
110
111 /***********************************************************/
112 /* ram save/restore */
113
114 #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */
115 #define RAM_SAVE_FLAG_COMPRESS 0x02
116 #define RAM_SAVE_FLAG_MEM_SIZE 0x04
117 #define RAM_SAVE_FLAG_PAGE 0x08
118 #define RAM_SAVE_FLAG_EOS 0x10
119 #define RAM_SAVE_FLAG_CONTINUE 0x20
120 #define RAM_SAVE_FLAG_XBZRLE 0x40
121
122
123 static struct defconfig_file {
124 const char *filename;
125 /* Indicates it is an user config file (disabled by -no-user-config) */
126 bool userconfig;
127 } default_config_files[] = {
128 { CONFIG_QEMU_CONFDIR "/qemu.conf", true },
129 { CONFIG_QEMU_CONFDIR "/target-" TARGET_NAME ".conf", true },
130 { NULL }, /* end of list */
131 };
132
133
134 int qemu_read_default_config_files(bool userconfig)
135 {
136 int ret;
137 struct defconfig_file *f;
138
139 for (f = default_config_files; f->filename; f++) {
140 if (!userconfig && f->userconfig) {
141 continue;
142 }
143 ret = qemu_read_config_file(f->filename);
144 if (ret < 0 && ret != -ENOENT) {
145 return ret;
146 }
147 }
148
149 return 0;
150 }
151
152 static inline bool is_zero_page(uint8_t *p)
153 {
154 return buffer_find_nonzero_offset(p, TARGET_PAGE_SIZE) ==
155 TARGET_PAGE_SIZE;
156 }
157
158 /* struct contains XBZRLE cache and a static page
159 used by the compression */
160 static struct {
161 /* buffer used for XBZRLE encoding */
162 uint8_t *encoded_buf;
163 /* buffer for storing page content */
164 uint8_t *current_buf;
165 /* buffer used for XBZRLE decoding */
166 uint8_t *decoded_buf;
167 /* Cache for XBZRLE */
168 PageCache *cache;
169 } XBZRLE = {
170 .encoded_buf = NULL,
171 .current_buf = NULL,
172 .decoded_buf = NULL,
173 .cache = NULL,
174 };
175
176
177 int64_t xbzrle_cache_resize(int64_t new_size)
178 {
179 if (XBZRLE.cache != NULL) {
180 return cache_resize(XBZRLE.cache, new_size / TARGET_PAGE_SIZE) *
181 TARGET_PAGE_SIZE;
182 }
183 return pow2floor(new_size);
184 }
185
186 /* accounting for migration statistics */
187 typedef struct AccountingInfo {
188 uint64_t dup_pages;
189 uint64_t skipped_pages;
190 uint64_t norm_pages;
191 uint64_t iterations;
192 uint64_t xbzrle_bytes;
193 uint64_t xbzrle_pages;
194 uint64_t xbzrle_cache_miss;
195 uint64_t xbzrle_overflows;
196 } AccountingInfo;
197
198 static AccountingInfo acct_info;
199
200 static void acct_clear(void)
201 {
202 memset(&acct_info, 0, sizeof(acct_info));
203 }
204
205 uint64_t dup_mig_bytes_transferred(void)
206 {
207 return acct_info.dup_pages * TARGET_PAGE_SIZE;
208 }
209
210 uint64_t dup_mig_pages_transferred(void)
211 {
212 return acct_info.dup_pages;
213 }
214
215 uint64_t skipped_mig_bytes_transferred(void)
216 {
217 return acct_info.skipped_pages * TARGET_PAGE_SIZE;
218 }
219
220 uint64_t skipped_mig_pages_transferred(void)
221 {
222 return acct_info.skipped_pages;
223 }
224
225 uint64_t norm_mig_bytes_transferred(void)
226 {
227 return acct_info.norm_pages * TARGET_PAGE_SIZE;
228 }
229
230 uint64_t norm_mig_pages_transferred(void)
231 {
232 return acct_info.norm_pages;
233 }
234
235 uint64_t xbzrle_mig_bytes_transferred(void)
236 {
237 return acct_info.xbzrle_bytes;
238 }
239
240 uint64_t xbzrle_mig_pages_transferred(void)
241 {
242 return acct_info.xbzrle_pages;
243 }
244
245 uint64_t xbzrle_mig_pages_cache_miss(void)
246 {
247 return acct_info.xbzrle_cache_miss;
248 }
249
250 uint64_t xbzrle_mig_pages_overflow(void)
251 {
252 return acct_info.xbzrle_overflows;
253 }
254
255 static size_t save_block_hdr(QEMUFile *f, RAMBlock *block, ram_addr_t offset,
256 int cont, int flag)
257 {
258 size_t size;
259
260 qemu_put_be64(f, offset | cont | flag);
261 size = 8;
262
263 if (!cont) {
264 qemu_put_byte(f, strlen(block->idstr));
265 qemu_put_buffer(f, (uint8_t *)block->idstr,
266 strlen(block->idstr));
267 size += 1 + strlen(block->idstr);
268 }
269 return size;
270 }
271
272 #define ENCODING_FLAG_XBZRLE 0x1
273
274 static int save_xbzrle_page(QEMUFile *f, uint8_t *current_data,
275 ram_addr_t current_addr, RAMBlock *block,
276 ram_addr_t offset, int cont, bool last_stage)
277 {
278 int encoded_len = 0, bytes_sent = -1;
279 uint8_t *prev_cached_page;
280
281 if (!cache_is_cached(XBZRLE.cache, current_addr)) {
282 if (!last_stage) {
283 cache_insert(XBZRLE.cache, current_addr, current_data);
284 }
285 acct_info.xbzrle_cache_miss++;
286 return -1;
287 }
288
289 prev_cached_page = get_cached_data(XBZRLE.cache, current_addr);
290
291 /* save current buffer into memory */
292 memcpy(XBZRLE.current_buf, current_data, TARGET_PAGE_SIZE);
293
294 /* XBZRLE encoding (if there is no overflow) */
295 encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf,
296 TARGET_PAGE_SIZE, XBZRLE.encoded_buf,
297 TARGET_PAGE_SIZE);
298 if (encoded_len == 0) {
299 DPRINTF("Skipping unmodified page\n");
300 return 0;
301 } else if (encoded_len == -1) {
302 DPRINTF("Overflow\n");
303 acct_info.xbzrle_overflows++;
304 /* update data in the cache */
305 memcpy(prev_cached_page, current_data, TARGET_PAGE_SIZE);
306 return -1;
307 }
308
309 /* we need to update the data in the cache, in order to get the same data */
310 if (!last_stage) {
311 memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE);
312 }
313
314 /* Send XBZRLE based compressed page */
315 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_XBZRLE);
316 qemu_put_byte(f, ENCODING_FLAG_XBZRLE);
317 qemu_put_be16(f, encoded_len);
318 qemu_put_buffer(f, XBZRLE.encoded_buf, encoded_len);
319 bytes_sent += encoded_len + 1 + 2;
320 acct_info.xbzrle_pages++;
321 acct_info.xbzrle_bytes += bytes_sent;
322
323 return bytes_sent;
324 }
325
326
327 /* This is the last block that we have visited serching for dirty pages
328 */
329 static RAMBlock *last_seen_block;
330 /* This is the last block from where we have sent data */
331 static RAMBlock *last_sent_block;
332 static ram_addr_t last_offset;
333 static unsigned long *migration_bitmap;
334 static uint64_t migration_dirty_pages;
335 static uint32_t last_version;
336 static bool ram_bulk_stage;
337
338 static inline
339 ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr,
340 ram_addr_t start)
341 {
342 unsigned long base = mr->ram_addr >> TARGET_PAGE_BITS;
343 unsigned long nr = base + (start >> TARGET_PAGE_BITS);
344 unsigned long size = base + (int128_get64(mr->size) >> TARGET_PAGE_BITS);
345
346 unsigned long next;
347
348 if (ram_bulk_stage && nr > base) {
349 next = nr + 1;
350 } else {
351 next = find_next_bit(migration_bitmap, size, nr);
352 }
353
354 if (next < size) {
355 clear_bit(next, migration_bitmap);
356 migration_dirty_pages--;
357 }
358 return (next - base) << TARGET_PAGE_BITS;
359 }
360
361 static inline bool migration_bitmap_set_dirty(MemoryRegion *mr,
362 ram_addr_t offset)
363 {
364 bool ret;
365 int nr = (mr->ram_addr + offset) >> TARGET_PAGE_BITS;
366
367 ret = test_and_set_bit(nr, migration_bitmap);
368
369 if (!ret) {
370 migration_dirty_pages++;
371 }
372 return ret;
373 }
374
375 /* Needs iothread lock! */
376
377 static void migration_bitmap_sync(void)
378 {
379 RAMBlock *block;
380 ram_addr_t addr;
381 uint64_t num_dirty_pages_init = migration_dirty_pages;
382 MigrationState *s = migrate_get_current();
383 static int64_t start_time;
384 static int64_t bytes_xfer_prev;
385 static int64_t num_dirty_pages_period;
386 int64_t end_time;
387 int64_t bytes_xfer_now;
388
389 if (!bytes_xfer_prev) {
390 bytes_xfer_prev = ram_bytes_transferred();
391 }
392
393 if (!start_time) {
394 start_time = qemu_get_clock_ms(rt_clock);
395 }
396
397 trace_migration_bitmap_sync_start();
398 address_space_sync_dirty_bitmap(&address_space_memory);
399
400 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
401 for (addr = 0; addr < block->length; addr += TARGET_PAGE_SIZE) {
402 if (memory_region_test_and_clear_dirty(block->mr,
403 addr, TARGET_PAGE_SIZE,
404 DIRTY_MEMORY_MIGRATION)) {
405 migration_bitmap_set_dirty(block->mr, addr);
406 }
407 }
408 }
409 trace_migration_bitmap_sync_end(migration_dirty_pages
410 - num_dirty_pages_init);
411 num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init;
412 end_time = qemu_get_clock_ms(rt_clock);
413
414 /* more than 1 second = 1000 millisecons */
415 if (end_time > start_time + 1000) {
416 if (migrate_auto_converge()) {
417 /* The following detection logic can be refined later. For now:
418 Check to see if the dirtied bytes is 50% more than the approx.
419 amount of bytes that just got transferred since the last time we
420 were in this routine. If that happens >N times (for now N==4)
421 we turn on the throttle down logic */
422 bytes_xfer_now = ram_bytes_transferred();
423 if (s->dirty_pages_rate &&
424 (num_dirty_pages_period * TARGET_PAGE_SIZE >
425 (bytes_xfer_now - bytes_xfer_prev)/2) &&
426 (dirty_rate_high_cnt++ > 4)) {
427 trace_migration_throttle();
428 mig_throttle_on = true;
429 dirty_rate_high_cnt = 0;
430 }
431 bytes_xfer_prev = bytes_xfer_now;
432 } else {
433 mig_throttle_on = false;
434 }
435 s->dirty_pages_rate = num_dirty_pages_period * 1000
436 / (end_time - start_time);
437 s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE;
438 start_time = end_time;
439 num_dirty_pages_period = 0;
440 }
441 }
442
443 /*
444 * ram_save_block: Writes a page of memory to the stream f
445 *
446 * Returns: The number of bytes written.
447 * 0 means no dirty pages
448 */
449
450 static int ram_save_block(QEMUFile *f, bool last_stage)
451 {
452 RAMBlock *block = last_seen_block;
453 ram_addr_t offset = last_offset;
454 bool complete_round = false;
455 int bytes_sent = 0;
456 MemoryRegion *mr;
457 ram_addr_t current_addr;
458
459 if (!block)
460 block = QTAILQ_FIRST(&ram_list.blocks);
461
462 while (true) {
463 mr = block->mr;
464 offset = migration_bitmap_find_and_reset_dirty(mr, offset);
465 if (complete_round && block == last_seen_block &&
466 offset >= last_offset) {
467 break;
468 }
469 if (offset >= block->length) {
470 offset = 0;
471 block = QTAILQ_NEXT(block, next);
472 if (!block) {
473 block = QTAILQ_FIRST(&ram_list.blocks);
474 complete_round = true;
475 ram_bulk_stage = false;
476 }
477 } else {
478 uint8_t *p;
479 int cont = (block == last_sent_block) ?
480 RAM_SAVE_FLAG_CONTINUE : 0;
481
482 p = memory_region_get_ram_ptr(mr) + offset;
483
484 /* In doubt sent page as normal */
485 bytes_sent = -1;
486 if (is_zero_page(p)) {
487 acct_info.dup_pages++;
488 bytes_sent = save_block_hdr(f, block, offset, cont,
489 RAM_SAVE_FLAG_COMPRESS);
490 qemu_put_byte(f, 0);
491 bytes_sent++;
492 } else if (!ram_bulk_stage && migrate_use_xbzrle()) {
493 current_addr = block->offset + offset;
494 bytes_sent = save_xbzrle_page(f, p, current_addr, block,
495 offset, cont, last_stage);
496 if (!last_stage) {
497 p = get_cached_data(XBZRLE.cache, current_addr);
498 }
499 }
500
501 /* XBZRLE overflow or normal page */
502 if (bytes_sent == -1) {
503 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_PAGE);
504 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE);
505 bytes_sent += TARGET_PAGE_SIZE;
506 acct_info.norm_pages++;
507 }
508
509 /* if page is unmodified, continue to the next */
510 if (bytes_sent > 0) {
511 last_sent_block = block;
512 break;
513 }
514 }
515 }
516 last_seen_block = block;
517 last_offset = offset;
518
519 return bytes_sent;
520 }
521
522 static uint64_t bytes_transferred;
523
524 void acct_update_position(QEMUFile *f, size_t size, bool zero)
525 {
526 uint64_t pages = size / TARGET_PAGE_SIZE;
527 if (zero) {
528 acct_info.dup_pages += pages;
529 } else {
530 acct_info.norm_pages += pages;
531 bytes_transferred += size;
532 qemu_update_position(f, size);
533 }
534 }
535
536 static ram_addr_t ram_save_remaining(void)
537 {
538 return migration_dirty_pages;
539 }
540
541 uint64_t ram_bytes_remaining(void)
542 {
543 return ram_save_remaining() * TARGET_PAGE_SIZE;
544 }
545
546 uint64_t ram_bytes_transferred(void)
547 {
548 return bytes_transferred;
549 }
550
551 uint64_t ram_bytes_total(void)
552 {
553 RAMBlock *block;
554 uint64_t total = 0;
555
556 QTAILQ_FOREACH(block, &ram_list.blocks, next)
557 total += block->length;
558
559 return total;
560 }
561
562 static void migration_end(void)
563 {
564 if (migration_bitmap) {
565 memory_global_dirty_log_stop();
566 g_free(migration_bitmap);
567 migration_bitmap = NULL;
568 }
569
570 if (XBZRLE.cache) {
571 cache_fini(XBZRLE.cache);
572 g_free(XBZRLE.cache);
573 g_free(XBZRLE.encoded_buf);
574 g_free(XBZRLE.current_buf);
575 g_free(XBZRLE.decoded_buf);
576 XBZRLE.cache = NULL;
577 }
578 }
579
580 static void ram_migration_cancel(void *opaque)
581 {
582 migration_end();
583 }
584
585 static void reset_ram_globals(void)
586 {
587 last_seen_block = NULL;
588 last_sent_block = NULL;
589 last_offset = 0;
590 last_version = ram_list.version;
591 ram_bulk_stage = true;
592 }
593
594 #define MAX_WAIT 50 /* ms, half buffered_file limit */
595
596 static int ram_save_setup(QEMUFile *f, void *opaque)
597 {
598 RAMBlock *block;
599 int64_t ram_pages = last_ram_offset() >> TARGET_PAGE_BITS;
600
601 migration_bitmap = bitmap_new(ram_pages);
602 bitmap_set(migration_bitmap, 0, ram_pages);
603 migration_dirty_pages = ram_pages;
604 mig_throttle_on = false;
605 dirty_rate_high_cnt = 0;
606
607 if (migrate_use_xbzrle()) {
608 XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() /
609 TARGET_PAGE_SIZE,
610 TARGET_PAGE_SIZE);
611 if (!XBZRLE.cache) {
612 DPRINTF("Error creating cache\n");
613 return -1;
614 }
615 XBZRLE.encoded_buf = g_malloc0(TARGET_PAGE_SIZE);
616 XBZRLE.current_buf = g_malloc(TARGET_PAGE_SIZE);
617 acct_clear();
618 }
619
620 qemu_mutex_lock_iothread();
621 qemu_mutex_lock_ramlist();
622 bytes_transferred = 0;
623 reset_ram_globals();
624
625 memory_global_dirty_log_start();
626 migration_bitmap_sync();
627 qemu_mutex_unlock_iothread();
628
629 qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE);
630
631 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
632 qemu_put_byte(f, strlen(block->idstr));
633 qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr));
634 qemu_put_be64(f, block->length);
635 }
636
637 qemu_mutex_unlock_ramlist();
638 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
639
640 return 0;
641 }
642
643 static int ram_save_iterate(QEMUFile *f, void *opaque)
644 {
645 int ret;
646 int i;
647 int64_t t0;
648 int total_sent = 0;
649
650 qemu_mutex_lock_ramlist();
651
652 if (ram_list.version != last_version) {
653 reset_ram_globals();
654 }
655
656 t0 = qemu_get_clock_ns(rt_clock);
657 i = 0;
658 while ((ret = qemu_file_rate_limit(f)) == 0) {
659 int bytes_sent;
660
661 bytes_sent = ram_save_block(f, false);
662 /* no more blocks to sent */
663 if (bytes_sent == 0) {
664 break;
665 }
666 total_sent += bytes_sent;
667 acct_info.iterations++;
668 check_guest_throttling();
669 /* we want to check in the 1st loop, just in case it was the 1st time
670 and we had to sync the dirty bitmap.
671 qemu_get_clock_ns() is a bit expensive, so we only check each some
672 iterations
673 */
674 if ((i & 63) == 0) {
675 uint64_t t1 = (qemu_get_clock_ns(rt_clock) - t0) / 1000000;
676 if (t1 > MAX_WAIT) {
677 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n",
678 t1, i);
679 break;
680 }
681 }
682 i++;
683 }
684
685 qemu_mutex_unlock_ramlist();
686
687 if (ret < 0) {
688 bytes_transferred += total_sent;
689 return ret;
690 }
691
692 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
693 total_sent += 8;
694 bytes_transferred += total_sent;
695
696 return total_sent;
697 }
698
699 static int ram_save_complete(QEMUFile *f, void *opaque)
700 {
701 qemu_mutex_lock_ramlist();
702 migration_bitmap_sync();
703
704 /* try transferring iterative blocks of memory */
705
706 /* flush all remaining blocks regardless of rate limiting */
707 while (true) {
708 int bytes_sent;
709
710 bytes_sent = ram_save_block(f, true);
711 /* no more blocks to sent */
712 if (bytes_sent == 0) {
713 break;
714 }
715 bytes_transferred += bytes_sent;
716 }
717 migration_end();
718
719 qemu_mutex_unlock_ramlist();
720 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
721
722 return 0;
723 }
724
725 static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
726 {
727 uint64_t remaining_size;
728
729 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
730
731 if (remaining_size < max_size) {
732 qemu_mutex_lock_iothread();
733 migration_bitmap_sync();
734 qemu_mutex_unlock_iothread();
735 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
736 }
737 return remaining_size;
738 }
739
740 static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host)
741 {
742 int ret, rc = 0;
743 unsigned int xh_len;
744 int xh_flags;
745
746 if (!XBZRLE.decoded_buf) {
747 XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE);
748 }
749
750 /* extract RLE header */
751 xh_flags = qemu_get_byte(f);
752 xh_len = qemu_get_be16(f);
753
754 if (xh_flags != ENCODING_FLAG_XBZRLE) {
755 fprintf(stderr, "Failed to load XBZRLE page - wrong compression!\n");
756 return -1;
757 }
758
759 if (xh_len > TARGET_PAGE_SIZE) {
760 fprintf(stderr, "Failed to load XBZRLE page - len overflow!\n");
761 return -1;
762 }
763 /* load data and decode */
764 qemu_get_buffer(f, XBZRLE.decoded_buf, xh_len);
765
766 /* decode RLE */
767 ret = xbzrle_decode_buffer(XBZRLE.decoded_buf, xh_len, host,
768 TARGET_PAGE_SIZE);
769 if (ret == -1) {
770 fprintf(stderr, "Failed to load XBZRLE page - decode error!\n");
771 rc = -1;
772 } else if (ret > TARGET_PAGE_SIZE) {
773 fprintf(stderr, "Failed to load XBZRLE page - size %d exceeds %d!\n",
774 ret, TARGET_PAGE_SIZE);
775 abort();
776 }
777
778 return rc;
779 }
780
781 static inline void *host_from_stream_offset(QEMUFile *f,
782 ram_addr_t offset,
783 int flags)
784 {
785 static RAMBlock *block = NULL;
786 char id[256];
787 uint8_t len;
788
789 if (flags & RAM_SAVE_FLAG_CONTINUE) {
790 if (!block) {
791 fprintf(stderr, "Ack, bad migration stream!\n");
792 return NULL;
793 }
794
795 return memory_region_get_ram_ptr(block->mr) + offset;
796 }
797
798 len = qemu_get_byte(f);
799 qemu_get_buffer(f, (uint8_t *)id, len);
800 id[len] = 0;
801
802 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
803 if (!strncmp(id, block->idstr, sizeof(id)))
804 return memory_region_get_ram_ptr(block->mr) + offset;
805 }
806
807 fprintf(stderr, "Can't find block %s!\n", id);
808 return NULL;
809 }
810
811 static int ram_load(QEMUFile *f, void *opaque, int version_id)
812 {
813 ram_addr_t addr;
814 int flags, ret = 0;
815 int error;
816 static uint64_t seq_iter;
817
818 seq_iter++;
819
820 if (version_id < 4 || version_id > 4) {
821 return -EINVAL;
822 }
823
824 do {
825 addr = qemu_get_be64(f);
826
827 flags = addr & ~TARGET_PAGE_MASK;
828 addr &= TARGET_PAGE_MASK;
829
830 if (flags & RAM_SAVE_FLAG_MEM_SIZE) {
831 if (version_id == 4) {
832 /* Synchronize RAM block list */
833 char id[256];
834 ram_addr_t length;
835 ram_addr_t total_ram_bytes = addr;
836
837 while (total_ram_bytes) {
838 RAMBlock *block;
839 uint8_t len;
840
841 len = qemu_get_byte(f);
842 qemu_get_buffer(f, (uint8_t *)id, len);
843 id[len] = 0;
844 length = qemu_get_be64(f);
845
846 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
847 if (!strncmp(id, block->idstr, sizeof(id))) {
848 if (block->length != length) {
849 fprintf(stderr,
850 "Length mismatch: %s: " RAM_ADDR_FMT
851 " in != " RAM_ADDR_FMT "\n", id, length,
852 block->length);
853 ret = -EINVAL;
854 goto done;
855 }
856 break;
857 }
858 }
859
860 if (!block) {
861 fprintf(stderr, "Unknown ramblock \"%s\", cannot "
862 "accept migration\n", id);
863 ret = -EINVAL;
864 goto done;
865 }
866
867 total_ram_bytes -= length;
868 }
869 }
870 }
871
872 if (flags & RAM_SAVE_FLAG_COMPRESS) {
873 void *host;
874 uint8_t ch;
875
876 host = host_from_stream_offset(f, addr, flags);
877 if (!host) {
878 return -EINVAL;
879 }
880
881 ch = qemu_get_byte(f);
882 if (ch != 0 || !is_zero_page(host)) {
883 memset(host, ch, TARGET_PAGE_SIZE);
884 #ifndef _WIN32
885 if (ch == 0 &&
886 (!kvm_enabled() || kvm_has_sync_mmu()) &&
887 getpagesize() <= TARGET_PAGE_SIZE) {
888 qemu_madvise(host, TARGET_PAGE_SIZE, QEMU_MADV_DONTNEED);
889 }
890 #endif
891 }
892 } else if (flags & RAM_SAVE_FLAG_PAGE) {
893 void *host;
894
895 host = host_from_stream_offset(f, addr, flags);
896 if (!host) {
897 return -EINVAL;
898 }
899
900 qemu_get_buffer(f, host, TARGET_PAGE_SIZE);
901 } else if (flags & RAM_SAVE_FLAG_XBZRLE) {
902 void *host = host_from_stream_offset(f, addr, flags);
903 if (!host) {
904 return -EINVAL;
905 }
906
907 if (load_xbzrle(f, addr, host) < 0) {
908 ret = -EINVAL;
909 goto done;
910 }
911 }
912 error = qemu_file_get_error(f);
913 if (error) {
914 ret = error;
915 goto done;
916 }
917 } while (!(flags & RAM_SAVE_FLAG_EOS));
918
919 done:
920 DPRINTF("Completed load of VM with exit code %d seq iteration "
921 "%" PRIu64 "\n", ret, seq_iter);
922 return ret;
923 }
924
925 SaveVMHandlers savevm_ram_handlers = {
926 .save_live_setup = ram_save_setup,
927 .save_live_iterate = ram_save_iterate,
928 .save_live_complete = ram_save_complete,
929 .save_live_pending = ram_save_pending,
930 .load_state = ram_load,
931 .cancel = ram_migration_cancel,
932 };
933
934 struct soundhw {
935 const char *name;
936 const char *descr;
937 int enabled;
938 int isa;
939 union {
940 int (*init_isa) (ISABus *bus);
941 int (*init_pci) (PCIBus *bus);
942 } init;
943 };
944
945 static struct soundhw soundhw[9];
946 static int soundhw_count;
947
948 void isa_register_soundhw(const char *name, const char *descr,
949 int (*init_isa)(ISABus *bus))
950 {
951 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
952 soundhw[soundhw_count].name = name;
953 soundhw[soundhw_count].descr = descr;
954 soundhw[soundhw_count].isa = 1;
955 soundhw[soundhw_count].init.init_isa = init_isa;
956 soundhw_count++;
957 }
958
959 void pci_register_soundhw(const char *name, const char *descr,
960 int (*init_pci)(PCIBus *bus))
961 {
962 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
963 soundhw[soundhw_count].name = name;
964 soundhw[soundhw_count].descr = descr;
965 soundhw[soundhw_count].isa = 0;
966 soundhw[soundhw_count].init.init_pci = init_pci;
967 soundhw_count++;
968 }
969
970 void select_soundhw(const char *optarg)
971 {
972 struct soundhw *c;
973
974 if (is_help_option(optarg)) {
975 show_valid_cards:
976
977 if (soundhw_count) {
978 printf("Valid sound card names (comma separated):\n");
979 for (c = soundhw; c->name; ++c) {
980 printf ("%-11s %s\n", c->name, c->descr);
981 }
982 printf("\n-soundhw all will enable all of the above\n");
983 } else {
984 printf("Machine has no user-selectable audio hardware "
985 "(it may or may not have always-present audio hardware).\n");
986 }
987 exit(!is_help_option(optarg));
988 }
989 else {
990 size_t l;
991 const char *p;
992 char *e;
993 int bad_card = 0;
994
995 if (!strcmp(optarg, "all")) {
996 for (c = soundhw; c->name; ++c) {
997 c->enabled = 1;
998 }
999 return;
1000 }
1001
1002 p = optarg;
1003 while (*p) {
1004 e = strchr(p, ',');
1005 l = !e ? strlen(p) : (size_t) (e - p);
1006
1007 for (c = soundhw; c->name; ++c) {
1008 if (!strncmp(c->name, p, l) && !c->name[l]) {
1009 c->enabled = 1;
1010 break;
1011 }
1012 }
1013
1014 if (!c->name) {
1015 if (l > 80) {
1016 fprintf(stderr,
1017 "Unknown sound card name (too big to show)\n");
1018 }
1019 else {
1020 fprintf(stderr, "Unknown sound card name `%.*s'\n",
1021 (int) l, p);
1022 }
1023 bad_card = 1;
1024 }
1025 p += l + (e != NULL);
1026 }
1027
1028 if (bad_card) {
1029 goto show_valid_cards;
1030 }
1031 }
1032 }
1033
1034 void audio_init(void)
1035 {
1036 struct soundhw *c;
1037 ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL);
1038 PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL);
1039
1040 for (c = soundhw; c->name; ++c) {
1041 if (c->enabled) {
1042 if (c->isa) {
1043 if (!isa_bus) {
1044 fprintf(stderr, "ISA bus not available for %s\n", c->name);
1045 exit(1);
1046 }
1047 c->init.init_isa(isa_bus);
1048 } else {
1049 if (!pci_bus) {
1050 fprintf(stderr, "PCI bus not available for %s\n", c->name);
1051 exit(1);
1052 }
1053 c->init.init_pci(pci_bus);
1054 }
1055 }
1056 }
1057 }
1058
1059 int qemu_uuid_parse(const char *str, uint8_t *uuid)
1060 {
1061 int ret;
1062
1063 if (strlen(str) != 36) {
1064 return -1;
1065 }
1066
1067 ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3],
1068 &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9],
1069 &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14],
1070 &uuid[15]);
1071
1072 if (ret != 16) {
1073 return -1;
1074 }
1075 #ifdef TARGET_I386
1076 smbios_add_field(1, offsetof(struct smbios_type_1, uuid), uuid, 16);
1077 #endif
1078 return 0;
1079 }
1080
1081 void do_acpitable_option(const QemuOpts *opts)
1082 {
1083 #ifdef TARGET_I386
1084 Error *err = NULL;
1085
1086 acpi_table_add(opts, &err);
1087 if (err) {
1088 fprintf(stderr, "Wrong acpi table provided: %s\n",
1089 error_get_pretty(err));
1090 error_free(err);
1091 exit(1);
1092 }
1093 #endif
1094 }
1095
1096 void do_smbios_option(const char *optarg)
1097 {
1098 #ifdef TARGET_I386
1099 if (smbios_entry_add(optarg) < 0) {
1100 exit(1);
1101 }
1102 #endif
1103 }
1104
1105 void cpudef_init(void)
1106 {
1107 #if defined(cpudef_setup)
1108 cpudef_setup(); /* parse cpu definitions in target config file */
1109 #endif
1110 }
1111
1112 int tcg_available(void)
1113 {
1114 return 1;
1115 }
1116
1117 int kvm_available(void)
1118 {
1119 #ifdef CONFIG_KVM
1120 return 1;
1121 #else
1122 return 0;
1123 #endif
1124 }
1125
1126 int xen_available(void)
1127 {
1128 #ifdef CONFIG_XEN
1129 return 1;
1130 #else
1131 return 0;
1132 #endif
1133 }
1134
1135
1136 TargetInfo *qmp_query_target(Error **errp)
1137 {
1138 TargetInfo *info = g_malloc0(sizeof(*info));
1139
1140 info->arch = g_strdup(TARGET_NAME);
1141
1142 return info;
1143 }
1144
1145 /* Stub function that's gets run on the vcpu when its brought out of the
1146 VM to run inside qemu via async_run_on_cpu()*/
1147 static void mig_sleep_cpu(void *opq)
1148 {
1149 qemu_mutex_unlock_iothread();
1150 g_usleep(30*1000);
1151 qemu_mutex_lock_iothread();
1152 }
1153
1154 /* To reduce the dirty rate explicitly disallow the VCPUs from spending
1155 much time in the VM. The migration thread will try to catchup.
1156 Workload will experience a performance drop.
1157 */
1158 static void mig_throttle_cpu_down(CPUState *cpu, void *data)
1159 {
1160 async_run_on_cpu(cpu, mig_sleep_cpu, NULL);
1161 }
1162
1163 static void mig_throttle_guest_down(void)
1164 {
1165 qemu_mutex_lock_iothread();
1166 qemu_for_each_cpu(mig_throttle_cpu_down, NULL);
1167 qemu_mutex_unlock_iothread();
1168 }
1169
1170 static void check_guest_throttling(void)
1171 {
1172 static int64_t t0;
1173 int64_t t1;
1174
1175 if (!mig_throttle_on) {
1176 return;
1177 }
1178
1179 if (!t0) {
1180 t0 = qemu_get_clock_ns(rt_clock);
1181 return;
1182 }
1183
1184 t1 = qemu_get_clock_ns(rt_clock);
1185
1186 /* If it has been more than 40 ms since the last time the guest
1187 * was throttled then do it again.
1188 */
1189 if (40 < (t1-t0)/1000000) {
1190 mig_throttle_guest_down();
1191 t0 = t1;
1192 }
1193 }