ssi: Convert legacy SSI_SLAVE -> DEVICE casts
[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 "exec/ram_addr.h"
52 #include "hw/acpi/acpi.h"
53 #include "qemu/host-utils.h"
54
55 #ifdef DEBUG_ARCH_INIT
56 #define DPRINTF(fmt, ...) \
57 do { fprintf(stdout, "arch_init: " fmt, ## __VA_ARGS__); } while (0)
58 #else
59 #define DPRINTF(fmt, ...) \
60 do { } while (0)
61 #endif
62
63 #ifdef TARGET_SPARC
64 int graphic_width = 1024;
65 int graphic_height = 768;
66 int graphic_depth = 8;
67 #else
68 int graphic_width = 800;
69 int graphic_height = 600;
70 int graphic_depth = 32;
71 #endif
72
73
74 #if defined(TARGET_ALPHA)
75 #define QEMU_ARCH QEMU_ARCH_ALPHA
76 #elif defined(TARGET_ARM)
77 #define QEMU_ARCH QEMU_ARCH_ARM
78 #elif defined(TARGET_CRIS)
79 #define QEMU_ARCH QEMU_ARCH_CRIS
80 #elif defined(TARGET_I386)
81 #define QEMU_ARCH QEMU_ARCH_I386
82 #elif defined(TARGET_M68K)
83 #define QEMU_ARCH QEMU_ARCH_M68K
84 #elif defined(TARGET_LM32)
85 #define QEMU_ARCH QEMU_ARCH_LM32
86 #elif defined(TARGET_MICROBLAZE)
87 #define QEMU_ARCH QEMU_ARCH_MICROBLAZE
88 #elif defined(TARGET_MIPS)
89 #define QEMU_ARCH QEMU_ARCH_MIPS
90 #elif defined(TARGET_MOXIE)
91 #define QEMU_ARCH QEMU_ARCH_MOXIE
92 #elif defined(TARGET_OPENRISC)
93 #define QEMU_ARCH QEMU_ARCH_OPENRISC
94 #elif defined(TARGET_PPC)
95 #define QEMU_ARCH QEMU_ARCH_PPC
96 #elif defined(TARGET_S390X)
97 #define QEMU_ARCH QEMU_ARCH_S390X
98 #elif defined(TARGET_SH4)
99 #define QEMU_ARCH QEMU_ARCH_SH4
100 #elif defined(TARGET_SPARC)
101 #define QEMU_ARCH QEMU_ARCH_SPARC
102 #elif defined(TARGET_XTENSA)
103 #define QEMU_ARCH QEMU_ARCH_XTENSA
104 #elif defined(TARGET_UNICORE32)
105 #define QEMU_ARCH QEMU_ARCH_UNICORE32
106 #endif
107
108 const uint32_t arch_type = QEMU_ARCH;
109 static bool mig_throttle_on;
110 static int dirty_rate_high_cnt;
111 static void check_guest_throttling(void);
112
113 /***********************************************************/
114 /* ram save/restore */
115
116 #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */
117 #define RAM_SAVE_FLAG_COMPRESS 0x02
118 #define RAM_SAVE_FLAG_MEM_SIZE 0x04
119 #define RAM_SAVE_FLAG_PAGE 0x08
120 #define RAM_SAVE_FLAG_EOS 0x10
121 #define RAM_SAVE_FLAG_CONTINUE 0x20
122 #define RAM_SAVE_FLAG_XBZRLE 0x40
123 /* 0x80 is reserved in migration.h start with 0x100 next */
124
125 static struct defconfig_file {
126 const char *filename;
127 /* Indicates it is an user config file (disabled by -no-user-config) */
128 bool userconfig;
129 } default_config_files[] = {
130 { CONFIG_QEMU_CONFDIR "/qemu.conf", true },
131 { CONFIG_QEMU_CONFDIR "/target-" TARGET_NAME ".conf", true },
132 { NULL }, /* end of list */
133 };
134
135 static const uint8_t ZERO_TARGET_PAGE[TARGET_PAGE_SIZE];
136
137 int qemu_read_default_config_files(bool userconfig)
138 {
139 int ret;
140 struct defconfig_file *f;
141
142 for (f = default_config_files; f->filename; f++) {
143 if (!userconfig && f->userconfig) {
144 continue;
145 }
146 ret = qemu_read_config_file(f->filename);
147 if (ret < 0 && ret != -ENOENT) {
148 return ret;
149 }
150 }
151
152 return 0;
153 }
154
155 static inline bool is_zero_range(uint8_t *p, uint64_t size)
156 {
157 return buffer_find_nonzero_offset(p, size) == size;
158 }
159
160 /* struct contains XBZRLE cache and a static page
161 used by the compression */
162 static struct {
163 /* buffer used for XBZRLE encoding */
164 uint8_t *encoded_buf;
165 /* buffer for storing page content */
166 uint8_t *current_buf;
167 /* Cache for XBZRLE, Protected by lock. */
168 PageCache *cache;
169 QemuMutex lock;
170 } XBZRLE = {
171 .encoded_buf = NULL,
172 .current_buf = NULL,
173 .cache = NULL,
174 };
175 /* buffer used for XBZRLE decoding */
176 static uint8_t *xbzrle_decoded_buf;
177
178 static void XBZRLE_cache_lock(void)
179 {
180 if (migrate_use_xbzrle())
181 qemu_mutex_lock(&XBZRLE.lock);
182 }
183
184 static void XBZRLE_cache_unlock(void)
185 {
186 if (migrate_use_xbzrle())
187 qemu_mutex_unlock(&XBZRLE.lock);
188 }
189
190 int64_t xbzrle_cache_resize(int64_t new_size)
191 {
192 PageCache *new_cache, *cache_to_free;
193
194 if (new_size < TARGET_PAGE_SIZE) {
195 return -1;
196 }
197
198 /* no need to lock, the current thread holds qemu big lock */
199 if (XBZRLE.cache != NULL) {
200 /* check XBZRLE.cache again later */
201 if (pow2floor(new_size) == migrate_xbzrle_cache_size()) {
202 return pow2floor(new_size);
203 }
204 new_cache = cache_init(new_size / TARGET_PAGE_SIZE,
205 TARGET_PAGE_SIZE);
206 if (!new_cache) {
207 DPRINTF("Error creating cache\n");
208 return -1;
209 }
210
211 XBZRLE_cache_lock();
212 /* the XBZRLE.cache may have be destroyed, check it again */
213 if (XBZRLE.cache != NULL) {
214 cache_to_free = XBZRLE.cache;
215 XBZRLE.cache = new_cache;
216 } else {
217 cache_to_free = new_cache;
218 }
219 XBZRLE_cache_unlock();
220
221 cache_fini(cache_to_free);
222 }
223
224 return pow2floor(new_size);
225 }
226
227 /* accounting for migration statistics */
228 typedef struct AccountingInfo {
229 uint64_t dup_pages;
230 uint64_t skipped_pages;
231 uint64_t norm_pages;
232 uint64_t iterations;
233 uint64_t xbzrle_bytes;
234 uint64_t xbzrle_pages;
235 uint64_t xbzrle_cache_miss;
236 uint64_t xbzrle_overflows;
237 } AccountingInfo;
238
239 static AccountingInfo acct_info;
240
241 static void acct_clear(void)
242 {
243 memset(&acct_info, 0, sizeof(acct_info));
244 }
245
246 uint64_t dup_mig_bytes_transferred(void)
247 {
248 return acct_info.dup_pages * TARGET_PAGE_SIZE;
249 }
250
251 uint64_t dup_mig_pages_transferred(void)
252 {
253 return acct_info.dup_pages;
254 }
255
256 uint64_t skipped_mig_bytes_transferred(void)
257 {
258 return acct_info.skipped_pages * TARGET_PAGE_SIZE;
259 }
260
261 uint64_t skipped_mig_pages_transferred(void)
262 {
263 return acct_info.skipped_pages;
264 }
265
266 uint64_t norm_mig_bytes_transferred(void)
267 {
268 return acct_info.norm_pages * TARGET_PAGE_SIZE;
269 }
270
271 uint64_t norm_mig_pages_transferred(void)
272 {
273 return acct_info.norm_pages;
274 }
275
276 uint64_t xbzrle_mig_bytes_transferred(void)
277 {
278 return acct_info.xbzrle_bytes;
279 }
280
281 uint64_t xbzrle_mig_pages_transferred(void)
282 {
283 return acct_info.xbzrle_pages;
284 }
285
286 uint64_t xbzrle_mig_pages_cache_miss(void)
287 {
288 return acct_info.xbzrle_cache_miss;
289 }
290
291 uint64_t xbzrle_mig_pages_overflow(void)
292 {
293 return acct_info.xbzrle_overflows;
294 }
295
296 static size_t save_block_hdr(QEMUFile *f, RAMBlock *block, ram_addr_t offset,
297 int cont, int flag)
298 {
299 size_t size;
300
301 qemu_put_be64(f, offset | cont | flag);
302 size = 8;
303
304 if (!cont) {
305 qemu_put_byte(f, strlen(block->idstr));
306 qemu_put_buffer(f, (uint8_t *)block->idstr,
307 strlen(block->idstr));
308 size += 1 + strlen(block->idstr);
309 }
310 return size;
311 }
312
313 /* This is the last block that we have visited serching for dirty pages
314 */
315 static RAMBlock *last_seen_block;
316 /* This is the last block from where we have sent data */
317 static RAMBlock *last_sent_block;
318 static ram_addr_t last_offset;
319 static unsigned long *migration_bitmap;
320 static uint64_t migration_dirty_pages;
321 static uint32_t last_version;
322 static bool ram_bulk_stage;
323
324 /* Update the xbzrle cache to reflect a page that's been sent as all 0.
325 * The important thing is that a stale (not-yet-0'd) page be replaced
326 * by the new data.
327 * As a bonus, if the page wasn't in the cache it gets added so that
328 * when a small write is made into the 0'd page it gets XBZRLE sent
329 */
330 static void xbzrle_cache_zero_page(ram_addr_t current_addr)
331 {
332 if (ram_bulk_stage || !migrate_use_xbzrle()) {
333 return;
334 }
335
336 /* We don't care if this fails to allocate a new cache page
337 * as long as it updated an old one */
338 cache_insert(XBZRLE.cache, current_addr, ZERO_TARGET_PAGE);
339 }
340
341 #define ENCODING_FLAG_XBZRLE 0x1
342
343 static int save_xbzrle_page(QEMUFile *f, uint8_t *current_data,
344 ram_addr_t current_addr, RAMBlock *block,
345 ram_addr_t offset, int cont, bool last_stage)
346 {
347 int encoded_len = 0, bytes_sent = -1;
348 uint8_t *prev_cached_page;
349
350 if (!cache_is_cached(XBZRLE.cache, current_addr)) {
351 if (!last_stage) {
352 if (cache_insert(XBZRLE.cache, current_addr, current_data) == -1) {
353 return -1;
354 }
355 }
356 acct_info.xbzrle_cache_miss++;
357 return -1;
358 }
359
360 prev_cached_page = get_cached_data(XBZRLE.cache, current_addr);
361
362 /* save current buffer into memory */
363 memcpy(XBZRLE.current_buf, current_data, TARGET_PAGE_SIZE);
364
365 /* XBZRLE encoding (if there is no overflow) */
366 encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf,
367 TARGET_PAGE_SIZE, XBZRLE.encoded_buf,
368 TARGET_PAGE_SIZE);
369 if (encoded_len == 0) {
370 DPRINTF("Skipping unmodified page\n");
371 return 0;
372 } else if (encoded_len == -1) {
373 DPRINTF("Overflow\n");
374 acct_info.xbzrle_overflows++;
375 /* update data in the cache */
376 memcpy(prev_cached_page, current_data, TARGET_PAGE_SIZE);
377 return -1;
378 }
379
380 /* we need to update the data in the cache, in order to get the same data */
381 if (!last_stage) {
382 memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE);
383 }
384
385 /* Send XBZRLE based compressed page */
386 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_XBZRLE);
387 qemu_put_byte(f, ENCODING_FLAG_XBZRLE);
388 qemu_put_be16(f, encoded_len);
389 qemu_put_buffer(f, XBZRLE.encoded_buf, encoded_len);
390 bytes_sent += encoded_len + 1 + 2;
391 acct_info.xbzrle_pages++;
392 acct_info.xbzrle_bytes += bytes_sent;
393
394 return bytes_sent;
395 }
396
397 static inline
398 ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr,
399 ram_addr_t start)
400 {
401 unsigned long base = mr->ram_addr >> TARGET_PAGE_BITS;
402 unsigned long nr = base + (start >> TARGET_PAGE_BITS);
403 uint64_t mr_size = TARGET_PAGE_ALIGN(memory_region_size(mr));
404 unsigned long size = base + (mr_size >> TARGET_PAGE_BITS);
405
406 unsigned long next;
407
408 if (ram_bulk_stage && nr > base) {
409 next = nr + 1;
410 } else {
411 next = find_next_bit(migration_bitmap, size, nr);
412 }
413
414 if (next < size) {
415 clear_bit(next, migration_bitmap);
416 migration_dirty_pages--;
417 }
418 return (next - base) << TARGET_PAGE_BITS;
419 }
420
421 static inline bool migration_bitmap_set_dirty(ram_addr_t addr)
422 {
423 bool ret;
424 int nr = addr >> TARGET_PAGE_BITS;
425
426 ret = test_and_set_bit(nr, migration_bitmap);
427
428 if (!ret) {
429 migration_dirty_pages++;
430 }
431 return ret;
432 }
433
434 static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length)
435 {
436 ram_addr_t addr;
437 unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
438
439 /* start address is aligned at the start of a word? */
440 if (((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) {
441 int k;
442 int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS);
443 unsigned long *src = ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION];
444
445 for (k = page; k < page + nr; k++) {
446 if (src[k]) {
447 unsigned long new_dirty;
448 new_dirty = ~migration_bitmap[k];
449 migration_bitmap[k] |= src[k];
450 new_dirty &= src[k];
451 migration_dirty_pages += ctpopl(new_dirty);
452 src[k] = 0;
453 }
454 }
455 } else {
456 for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) {
457 if (cpu_physical_memory_get_dirty(start + addr,
458 TARGET_PAGE_SIZE,
459 DIRTY_MEMORY_MIGRATION)) {
460 cpu_physical_memory_reset_dirty(start + addr,
461 TARGET_PAGE_SIZE,
462 DIRTY_MEMORY_MIGRATION);
463 migration_bitmap_set_dirty(start + addr);
464 }
465 }
466 }
467 }
468
469
470 /* Needs iothread lock! */
471
472 static void migration_bitmap_sync(void)
473 {
474 RAMBlock *block;
475 uint64_t num_dirty_pages_init = migration_dirty_pages;
476 MigrationState *s = migrate_get_current();
477 static int64_t start_time;
478 static int64_t bytes_xfer_prev;
479 static int64_t num_dirty_pages_period;
480 int64_t end_time;
481 int64_t bytes_xfer_now;
482
483 if (!bytes_xfer_prev) {
484 bytes_xfer_prev = ram_bytes_transferred();
485 }
486
487 if (!start_time) {
488 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
489 }
490
491 trace_migration_bitmap_sync_start();
492 address_space_sync_dirty_bitmap(&address_space_memory);
493
494 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
495 migration_bitmap_sync_range(block->mr->ram_addr, block->length);
496 }
497 trace_migration_bitmap_sync_end(migration_dirty_pages
498 - num_dirty_pages_init);
499 num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init;
500 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
501
502 /* more than 1 second = 1000 millisecons */
503 if (end_time > start_time + 1000) {
504 if (migrate_auto_converge()) {
505 /* The following detection logic can be refined later. For now:
506 Check to see if the dirtied bytes is 50% more than the approx.
507 amount of bytes that just got transferred since the last time we
508 were in this routine. If that happens >N times (for now N==4)
509 we turn on the throttle down logic */
510 bytes_xfer_now = ram_bytes_transferred();
511 if (s->dirty_pages_rate &&
512 (num_dirty_pages_period * TARGET_PAGE_SIZE >
513 (bytes_xfer_now - bytes_xfer_prev)/2) &&
514 (dirty_rate_high_cnt++ > 4)) {
515 trace_migration_throttle();
516 mig_throttle_on = true;
517 dirty_rate_high_cnt = 0;
518 }
519 bytes_xfer_prev = bytes_xfer_now;
520 } else {
521 mig_throttle_on = false;
522 }
523 s->dirty_pages_rate = num_dirty_pages_period * 1000
524 / (end_time - start_time);
525 s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE;
526 start_time = end_time;
527 num_dirty_pages_period = 0;
528 }
529 }
530
531 /*
532 * ram_save_block: Writes a page of memory to the stream f
533 *
534 * Returns: The number of bytes written.
535 * 0 means no dirty pages
536 */
537
538 static int ram_save_block(QEMUFile *f, bool last_stage)
539 {
540 RAMBlock *block = last_seen_block;
541 ram_addr_t offset = last_offset;
542 bool complete_round = false;
543 int bytes_sent = 0;
544 MemoryRegion *mr;
545 ram_addr_t current_addr;
546
547 if (!block)
548 block = QTAILQ_FIRST(&ram_list.blocks);
549
550 while (true) {
551 mr = block->mr;
552 offset = migration_bitmap_find_and_reset_dirty(mr, offset);
553 if (complete_round && block == last_seen_block &&
554 offset >= last_offset) {
555 break;
556 }
557 if (offset >= block->length) {
558 offset = 0;
559 block = QTAILQ_NEXT(block, next);
560 if (!block) {
561 block = QTAILQ_FIRST(&ram_list.blocks);
562 complete_round = true;
563 ram_bulk_stage = false;
564 }
565 } else {
566 int ret;
567 uint8_t *p;
568 bool send_async = true;
569 int cont = (block == last_sent_block) ?
570 RAM_SAVE_FLAG_CONTINUE : 0;
571
572 p = memory_region_get_ram_ptr(mr) + offset;
573
574 /* In doubt sent page as normal */
575 bytes_sent = -1;
576 ret = ram_control_save_page(f, block->offset,
577 offset, TARGET_PAGE_SIZE, &bytes_sent);
578
579 XBZRLE_cache_lock();
580
581 current_addr = block->offset + offset;
582 if (ret != RAM_SAVE_CONTROL_NOT_SUPP) {
583 if (ret != RAM_SAVE_CONTROL_DELAYED) {
584 if (bytes_sent > 0) {
585 acct_info.norm_pages++;
586 } else if (bytes_sent == 0) {
587 acct_info.dup_pages++;
588 }
589 }
590 } else if (is_zero_range(p, TARGET_PAGE_SIZE)) {
591 acct_info.dup_pages++;
592 bytes_sent = save_block_hdr(f, block, offset, cont,
593 RAM_SAVE_FLAG_COMPRESS);
594 qemu_put_byte(f, 0);
595 bytes_sent++;
596 /* Must let xbzrle know, otherwise a previous (now 0'd) cached
597 * page would be stale
598 */
599 xbzrle_cache_zero_page(current_addr);
600 } else if (!ram_bulk_stage && migrate_use_xbzrle()) {
601 bytes_sent = save_xbzrle_page(f, p, current_addr, block,
602 offset, cont, last_stage);
603 if (!last_stage) {
604 /* We must send exactly what's in the xbzrle cache
605 * even if the page wasn't xbzrle compressed, so that
606 * it's right next time.
607 */
608 p = get_cached_data(XBZRLE.cache, current_addr);
609
610 /* Can't send this cached data async, since the cache page
611 * might get updated before it gets to the wire
612 */
613 send_async = false;
614 }
615 }
616
617 /* XBZRLE overflow or normal page */
618 if (bytes_sent == -1) {
619 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_PAGE);
620 if (send_async) {
621 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE);
622 } else {
623 qemu_put_buffer(f, p, TARGET_PAGE_SIZE);
624 }
625 bytes_sent += TARGET_PAGE_SIZE;
626 acct_info.norm_pages++;
627 }
628
629 XBZRLE_cache_unlock();
630 /* if page is unmodified, continue to the next */
631 if (bytes_sent > 0) {
632 last_sent_block = block;
633 break;
634 }
635 }
636 }
637 last_seen_block = block;
638 last_offset = offset;
639
640 return bytes_sent;
641 }
642
643 static uint64_t bytes_transferred;
644
645 void acct_update_position(QEMUFile *f, size_t size, bool zero)
646 {
647 uint64_t pages = size / TARGET_PAGE_SIZE;
648 if (zero) {
649 acct_info.dup_pages += pages;
650 } else {
651 acct_info.norm_pages += pages;
652 bytes_transferred += size;
653 qemu_update_position(f, size);
654 }
655 }
656
657 static ram_addr_t ram_save_remaining(void)
658 {
659 return migration_dirty_pages;
660 }
661
662 uint64_t ram_bytes_remaining(void)
663 {
664 return ram_save_remaining() * TARGET_PAGE_SIZE;
665 }
666
667 uint64_t ram_bytes_transferred(void)
668 {
669 return bytes_transferred;
670 }
671
672 uint64_t ram_bytes_total(void)
673 {
674 RAMBlock *block;
675 uint64_t total = 0;
676
677 QTAILQ_FOREACH(block, &ram_list.blocks, next)
678 total += block->length;
679
680 return total;
681 }
682
683 void free_xbzrle_decoded_buf(void)
684 {
685 g_free(xbzrle_decoded_buf);
686 xbzrle_decoded_buf = NULL;
687 }
688
689 static void migration_end(void)
690 {
691 if (migration_bitmap) {
692 memory_global_dirty_log_stop();
693 g_free(migration_bitmap);
694 migration_bitmap = NULL;
695 }
696
697 XBZRLE_cache_lock();
698 if (XBZRLE.cache) {
699 cache_fini(XBZRLE.cache);
700 g_free(XBZRLE.cache);
701 g_free(XBZRLE.encoded_buf);
702 g_free(XBZRLE.current_buf);
703 XBZRLE.cache = NULL;
704 XBZRLE.encoded_buf = NULL;
705 XBZRLE.current_buf = NULL;
706 }
707 XBZRLE_cache_unlock();
708 }
709
710 static void ram_migration_cancel(void *opaque)
711 {
712 migration_end();
713 }
714
715 static void reset_ram_globals(void)
716 {
717 last_seen_block = NULL;
718 last_sent_block = NULL;
719 last_offset = 0;
720 last_version = ram_list.version;
721 ram_bulk_stage = true;
722 }
723
724 #define MAX_WAIT 50 /* ms, half buffered_file limit */
725
726 static int ram_save_setup(QEMUFile *f, void *opaque)
727 {
728 RAMBlock *block;
729 int64_t ram_pages = last_ram_offset() >> TARGET_PAGE_BITS;
730
731 migration_bitmap = bitmap_new(ram_pages);
732 bitmap_set(migration_bitmap, 0, ram_pages);
733 migration_dirty_pages = ram_pages;
734 mig_throttle_on = false;
735 dirty_rate_high_cnt = 0;
736
737 if (migrate_use_xbzrle()) {
738 qemu_mutex_lock_iothread();
739 XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() /
740 TARGET_PAGE_SIZE,
741 TARGET_PAGE_SIZE);
742 if (!XBZRLE.cache) {
743 qemu_mutex_unlock_iothread();
744 DPRINTF("Error creating cache\n");
745 return -1;
746 }
747 qemu_mutex_init(&XBZRLE.lock);
748 qemu_mutex_unlock_iothread();
749
750 /* We prefer not to abort if there is no memory */
751 XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE);
752 if (!XBZRLE.encoded_buf) {
753 DPRINTF("Error allocating encoded_buf\n");
754 return -1;
755 }
756
757 XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE);
758 if (!XBZRLE.current_buf) {
759 DPRINTF("Error allocating current_buf\n");
760 g_free(XBZRLE.encoded_buf);
761 XBZRLE.encoded_buf = NULL;
762 return -1;
763 }
764
765 acct_clear();
766 }
767
768 qemu_mutex_lock_iothread();
769 qemu_mutex_lock_ramlist();
770 bytes_transferred = 0;
771 reset_ram_globals();
772
773 memory_global_dirty_log_start();
774 migration_bitmap_sync();
775 qemu_mutex_unlock_iothread();
776
777 qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE);
778
779 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
780 qemu_put_byte(f, strlen(block->idstr));
781 qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr));
782 qemu_put_be64(f, block->length);
783 }
784
785 qemu_mutex_unlock_ramlist();
786
787 ram_control_before_iterate(f, RAM_CONTROL_SETUP);
788 ram_control_after_iterate(f, RAM_CONTROL_SETUP);
789
790 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
791
792 return 0;
793 }
794
795 static int ram_save_iterate(QEMUFile *f, void *opaque)
796 {
797 int ret;
798 int i;
799 int64_t t0;
800 int total_sent = 0;
801
802 qemu_mutex_lock_ramlist();
803
804 if (ram_list.version != last_version) {
805 reset_ram_globals();
806 }
807
808 ram_control_before_iterate(f, RAM_CONTROL_ROUND);
809
810 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
811 i = 0;
812 while ((ret = qemu_file_rate_limit(f)) == 0) {
813 int bytes_sent;
814
815 bytes_sent = ram_save_block(f, false);
816 /* no more blocks to sent */
817 if (bytes_sent == 0) {
818 break;
819 }
820 total_sent += bytes_sent;
821 acct_info.iterations++;
822 check_guest_throttling();
823 /* we want to check in the 1st loop, just in case it was the 1st time
824 and we had to sync the dirty bitmap.
825 qemu_get_clock_ns() is a bit expensive, so we only check each some
826 iterations
827 */
828 if ((i & 63) == 0) {
829 uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000;
830 if (t1 > MAX_WAIT) {
831 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n",
832 t1, i);
833 break;
834 }
835 }
836 i++;
837 }
838
839 qemu_mutex_unlock_ramlist();
840
841 /*
842 * Must occur before EOS (or any QEMUFile operation)
843 * because of RDMA protocol.
844 */
845 ram_control_after_iterate(f, RAM_CONTROL_ROUND);
846
847 bytes_transferred += total_sent;
848
849 /*
850 * Do not count these 8 bytes into total_sent, so that we can
851 * return 0 if no page had been dirtied.
852 */
853 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
854 bytes_transferred += 8;
855
856 ret = qemu_file_get_error(f);
857 if (ret < 0) {
858 return ret;
859 }
860
861 return total_sent;
862 }
863
864 static int ram_save_complete(QEMUFile *f, void *opaque)
865 {
866 qemu_mutex_lock_ramlist();
867 migration_bitmap_sync();
868
869 ram_control_before_iterate(f, RAM_CONTROL_FINISH);
870
871 /* try transferring iterative blocks of memory */
872
873 /* flush all remaining blocks regardless of rate limiting */
874 while (true) {
875 int bytes_sent;
876
877 bytes_sent = ram_save_block(f, true);
878 /* no more blocks to sent */
879 if (bytes_sent == 0) {
880 break;
881 }
882 bytes_transferred += bytes_sent;
883 }
884
885 ram_control_after_iterate(f, RAM_CONTROL_FINISH);
886 migration_end();
887
888 qemu_mutex_unlock_ramlist();
889 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
890
891 return 0;
892 }
893
894 static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
895 {
896 uint64_t remaining_size;
897
898 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
899
900 if (remaining_size < max_size) {
901 qemu_mutex_lock_iothread();
902 migration_bitmap_sync();
903 qemu_mutex_unlock_iothread();
904 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
905 }
906 return remaining_size;
907 }
908
909 static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host)
910 {
911 int ret, rc = 0;
912 unsigned int xh_len;
913 int xh_flags;
914
915 if (!xbzrle_decoded_buf) {
916 xbzrle_decoded_buf = g_malloc(TARGET_PAGE_SIZE);
917 }
918
919 /* extract RLE header */
920 xh_flags = qemu_get_byte(f);
921 xh_len = qemu_get_be16(f);
922
923 if (xh_flags != ENCODING_FLAG_XBZRLE) {
924 fprintf(stderr, "Failed to load XBZRLE page - wrong compression!\n");
925 return -1;
926 }
927
928 if (xh_len > TARGET_PAGE_SIZE) {
929 fprintf(stderr, "Failed to load XBZRLE page - len overflow!\n");
930 return -1;
931 }
932 /* load data and decode */
933 qemu_get_buffer(f, xbzrle_decoded_buf, xh_len);
934
935 /* decode RLE */
936 ret = xbzrle_decode_buffer(xbzrle_decoded_buf, xh_len, host,
937 TARGET_PAGE_SIZE);
938 if (ret == -1) {
939 fprintf(stderr, "Failed to load XBZRLE page - decode error!\n");
940 rc = -1;
941 } else if (ret > TARGET_PAGE_SIZE) {
942 fprintf(stderr, "Failed to load XBZRLE page - size %d exceeds %d!\n",
943 ret, TARGET_PAGE_SIZE);
944 abort();
945 }
946
947 return rc;
948 }
949
950 static inline void *host_from_stream_offset(QEMUFile *f,
951 ram_addr_t offset,
952 int flags)
953 {
954 static RAMBlock *block = NULL;
955 char id[256];
956 uint8_t len;
957
958 if (flags & RAM_SAVE_FLAG_CONTINUE) {
959 if (!block) {
960 fprintf(stderr, "Ack, bad migration stream!\n");
961 return NULL;
962 }
963
964 return memory_region_get_ram_ptr(block->mr) + offset;
965 }
966
967 len = qemu_get_byte(f);
968 qemu_get_buffer(f, (uint8_t *)id, len);
969 id[len] = 0;
970
971 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
972 if (!strncmp(id, block->idstr, sizeof(id)))
973 return memory_region_get_ram_ptr(block->mr) + offset;
974 }
975
976 fprintf(stderr, "Can't find block %s!\n", id);
977 return NULL;
978 }
979
980 /*
981 * If a page (or a whole RDMA chunk) has been
982 * determined to be zero, then zap it.
983 */
984 void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
985 {
986 if (ch != 0 || !is_zero_range(host, size)) {
987 memset(host, ch, size);
988 }
989 }
990
991 static int ram_load(QEMUFile *f, void *opaque, int version_id)
992 {
993 ram_addr_t addr;
994 int flags, ret = 0;
995 int error;
996 static uint64_t seq_iter;
997
998 seq_iter++;
999
1000 if (version_id < 4 || version_id > 4) {
1001 return -EINVAL;
1002 }
1003
1004 do {
1005 addr = qemu_get_be64(f);
1006
1007 flags = addr & ~TARGET_PAGE_MASK;
1008 addr &= TARGET_PAGE_MASK;
1009
1010 if (flags & RAM_SAVE_FLAG_MEM_SIZE) {
1011 if (version_id == 4) {
1012 /* Synchronize RAM block list */
1013 char id[256];
1014 ram_addr_t length;
1015 ram_addr_t total_ram_bytes = addr;
1016
1017 while (total_ram_bytes) {
1018 RAMBlock *block;
1019 uint8_t len;
1020
1021 len = qemu_get_byte(f);
1022 qemu_get_buffer(f, (uint8_t *)id, len);
1023 id[len] = 0;
1024 length = qemu_get_be64(f);
1025
1026 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1027 if (!strncmp(id, block->idstr, sizeof(id))) {
1028 if (block->length != length) {
1029 fprintf(stderr,
1030 "Length mismatch: %s: " RAM_ADDR_FMT
1031 " in != " RAM_ADDR_FMT "\n", id, length,
1032 block->length);
1033 ret = -EINVAL;
1034 goto done;
1035 }
1036 break;
1037 }
1038 }
1039
1040 if (!block) {
1041 fprintf(stderr, "Unknown ramblock \"%s\", cannot "
1042 "accept migration\n", id);
1043 ret = -EINVAL;
1044 goto done;
1045 }
1046
1047 total_ram_bytes -= length;
1048 }
1049 }
1050 }
1051
1052 if (flags & RAM_SAVE_FLAG_COMPRESS) {
1053 void *host;
1054 uint8_t ch;
1055
1056 host = host_from_stream_offset(f, addr, flags);
1057 if (!host) {
1058 return -EINVAL;
1059 }
1060
1061 ch = qemu_get_byte(f);
1062 ram_handle_compressed(host, ch, TARGET_PAGE_SIZE);
1063 } else if (flags & RAM_SAVE_FLAG_PAGE) {
1064 void *host;
1065
1066 host = host_from_stream_offset(f, addr, flags);
1067 if (!host) {
1068 return -EINVAL;
1069 }
1070
1071 qemu_get_buffer(f, host, TARGET_PAGE_SIZE);
1072 } else if (flags & RAM_SAVE_FLAG_XBZRLE) {
1073 void *host = host_from_stream_offset(f, addr, flags);
1074 if (!host) {
1075 return -EINVAL;
1076 }
1077
1078 if (load_xbzrle(f, addr, host) < 0) {
1079 ret = -EINVAL;
1080 goto done;
1081 }
1082 } else if (flags & RAM_SAVE_FLAG_HOOK) {
1083 ram_control_load_hook(f, flags);
1084 }
1085 error = qemu_file_get_error(f);
1086 if (error) {
1087 ret = error;
1088 goto done;
1089 }
1090 } while (!(flags & RAM_SAVE_FLAG_EOS));
1091
1092 done:
1093 DPRINTF("Completed load of VM with exit code %d seq iteration "
1094 "%" PRIu64 "\n", ret, seq_iter);
1095 return ret;
1096 }
1097
1098 SaveVMHandlers savevm_ram_handlers = {
1099 .save_live_setup = ram_save_setup,
1100 .save_live_iterate = ram_save_iterate,
1101 .save_live_complete = ram_save_complete,
1102 .save_live_pending = ram_save_pending,
1103 .load_state = ram_load,
1104 .cancel = ram_migration_cancel,
1105 };
1106
1107 struct soundhw {
1108 const char *name;
1109 const char *descr;
1110 int enabled;
1111 int isa;
1112 union {
1113 int (*init_isa) (ISABus *bus);
1114 int (*init_pci) (PCIBus *bus);
1115 } init;
1116 };
1117
1118 static struct soundhw soundhw[9];
1119 static int soundhw_count;
1120
1121 void isa_register_soundhw(const char *name, const char *descr,
1122 int (*init_isa)(ISABus *bus))
1123 {
1124 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1125 soundhw[soundhw_count].name = name;
1126 soundhw[soundhw_count].descr = descr;
1127 soundhw[soundhw_count].isa = 1;
1128 soundhw[soundhw_count].init.init_isa = init_isa;
1129 soundhw_count++;
1130 }
1131
1132 void pci_register_soundhw(const char *name, const char *descr,
1133 int (*init_pci)(PCIBus *bus))
1134 {
1135 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1136 soundhw[soundhw_count].name = name;
1137 soundhw[soundhw_count].descr = descr;
1138 soundhw[soundhw_count].isa = 0;
1139 soundhw[soundhw_count].init.init_pci = init_pci;
1140 soundhw_count++;
1141 }
1142
1143 void select_soundhw(const char *optarg)
1144 {
1145 struct soundhw *c;
1146
1147 if (is_help_option(optarg)) {
1148 show_valid_cards:
1149
1150 if (soundhw_count) {
1151 printf("Valid sound card names (comma separated):\n");
1152 for (c = soundhw; c->name; ++c) {
1153 printf ("%-11s %s\n", c->name, c->descr);
1154 }
1155 printf("\n-soundhw all will enable all of the above\n");
1156 } else {
1157 printf("Machine has no user-selectable audio hardware "
1158 "(it may or may not have always-present audio hardware).\n");
1159 }
1160 exit(!is_help_option(optarg));
1161 }
1162 else {
1163 size_t l;
1164 const char *p;
1165 char *e;
1166 int bad_card = 0;
1167
1168 if (!strcmp(optarg, "all")) {
1169 for (c = soundhw; c->name; ++c) {
1170 c->enabled = 1;
1171 }
1172 return;
1173 }
1174
1175 p = optarg;
1176 while (*p) {
1177 e = strchr(p, ',');
1178 l = !e ? strlen(p) : (size_t) (e - p);
1179
1180 for (c = soundhw; c->name; ++c) {
1181 if (!strncmp(c->name, p, l) && !c->name[l]) {
1182 c->enabled = 1;
1183 break;
1184 }
1185 }
1186
1187 if (!c->name) {
1188 if (l > 80) {
1189 fprintf(stderr,
1190 "Unknown sound card name (too big to show)\n");
1191 }
1192 else {
1193 fprintf(stderr, "Unknown sound card name `%.*s'\n",
1194 (int) l, p);
1195 }
1196 bad_card = 1;
1197 }
1198 p += l + (e != NULL);
1199 }
1200
1201 if (bad_card) {
1202 goto show_valid_cards;
1203 }
1204 }
1205 }
1206
1207 void audio_init(void)
1208 {
1209 struct soundhw *c;
1210 ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL);
1211 PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL);
1212
1213 for (c = soundhw; c->name; ++c) {
1214 if (c->enabled) {
1215 if (c->isa) {
1216 if (!isa_bus) {
1217 fprintf(stderr, "ISA bus not available for %s\n", c->name);
1218 exit(1);
1219 }
1220 c->init.init_isa(isa_bus);
1221 } else {
1222 if (!pci_bus) {
1223 fprintf(stderr, "PCI bus not available for %s\n", c->name);
1224 exit(1);
1225 }
1226 c->init.init_pci(pci_bus);
1227 }
1228 }
1229 }
1230 }
1231
1232 int qemu_uuid_parse(const char *str, uint8_t *uuid)
1233 {
1234 int ret;
1235
1236 if (strlen(str) != 36) {
1237 return -1;
1238 }
1239
1240 ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3],
1241 &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9],
1242 &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14],
1243 &uuid[15]);
1244
1245 if (ret != 16) {
1246 return -1;
1247 }
1248 return 0;
1249 }
1250
1251 void do_acpitable_option(const QemuOpts *opts)
1252 {
1253 #ifdef TARGET_I386
1254 Error *err = NULL;
1255
1256 acpi_table_add(opts, &err);
1257 if (err) {
1258 error_report("Wrong acpi table provided: %s",
1259 error_get_pretty(err));
1260 error_free(err);
1261 exit(1);
1262 }
1263 #endif
1264 }
1265
1266 void do_smbios_option(QemuOpts *opts)
1267 {
1268 #ifdef TARGET_I386
1269 smbios_entry_add(opts);
1270 #endif
1271 }
1272
1273 void cpudef_init(void)
1274 {
1275 #if defined(cpudef_setup)
1276 cpudef_setup(); /* parse cpu definitions in target config file */
1277 #endif
1278 }
1279
1280 int tcg_available(void)
1281 {
1282 return 1;
1283 }
1284
1285 int kvm_available(void)
1286 {
1287 #ifdef CONFIG_KVM
1288 return 1;
1289 #else
1290 return 0;
1291 #endif
1292 }
1293
1294 int xen_available(void)
1295 {
1296 #ifdef CONFIG_XEN
1297 return 1;
1298 #else
1299 return 0;
1300 #endif
1301 }
1302
1303
1304 TargetInfo *qmp_query_target(Error **errp)
1305 {
1306 TargetInfo *info = g_malloc0(sizeof(*info));
1307
1308 info->arch = g_strdup(TARGET_NAME);
1309
1310 return info;
1311 }
1312
1313 /* Stub function that's gets run on the vcpu when its brought out of the
1314 VM to run inside qemu via async_run_on_cpu()*/
1315 static void mig_sleep_cpu(void *opq)
1316 {
1317 qemu_mutex_unlock_iothread();
1318 g_usleep(30*1000);
1319 qemu_mutex_lock_iothread();
1320 }
1321
1322 /* To reduce the dirty rate explicitly disallow the VCPUs from spending
1323 much time in the VM. The migration thread will try to catchup.
1324 Workload will experience a performance drop.
1325 */
1326 static void mig_throttle_guest_down(void)
1327 {
1328 CPUState *cpu;
1329
1330 qemu_mutex_lock_iothread();
1331 CPU_FOREACH(cpu) {
1332 async_run_on_cpu(cpu, mig_sleep_cpu, NULL);
1333 }
1334 qemu_mutex_unlock_iothread();
1335 }
1336
1337 static void check_guest_throttling(void)
1338 {
1339 static int64_t t0;
1340 int64_t t1;
1341
1342 if (!mig_throttle_on) {
1343 return;
1344 }
1345
1346 if (!t0) {
1347 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1348 return;
1349 }
1350
1351 t1 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1352
1353 /* If it has been more than 40 ms since the last time the guest
1354 * was throttled then do it again.
1355 */
1356 if (40 < (t1-t0)/1000000) {
1357 mig_throttle_guest_down();
1358 t0 = t1;
1359 }
1360 }