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