]> git.proxmox.com Git - mirror_qemu.git/blame - arch_init.c
arch_init: Alloc and free data struct for compression
[mirror_qemu.git] / arch_init.c
CommitLineData
ad96090a
BS
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>
b2e0a138 26#include <stdlib.h>
3fcb38c2 27#include <zlib.h>
ad96090a 28#ifndef _WIN32
1c47cb16 29#include <sys/types.h>
ad96090a
BS
30#include <sys/mman.h>
31#endif
32#include "config.h"
83c9089e 33#include "monitor/monitor.h"
9c17d615 34#include "sysemu/sysemu.h"
1de7afc9
PB
35#include "qemu/bitops.h"
36#include "qemu/bitmap.h"
9c17d615 37#include "sysemu/arch_init.h"
ad96090a 38#include "audio/audio.h"
0d09e41a 39#include "hw/i386/pc.h"
a2cb15b0 40#include "hw/pci/pci.h"
0d09e41a 41#include "hw/audio/audio.h"
9c17d615 42#include "sysemu/kvm.h"
caf71f86 43#include "migration/migration.h"
0d09e41a 44#include "hw/i386/smbios.h"
022c62cb 45#include "exec/address-spaces.h"
0d09e41a 46#include "hw/audio/pcspk.h"
caf71f86 47#include "migration/page_cache.h"
1de7afc9 48#include "qemu/config-file.h"
d97326ee 49#include "qemu/error-report.h"
99afc91d 50#include "qmp-commands.h"
3c12193d 51#include "trace.h"
0d6d3c87 52#include "exec/cpu-all.h"
12291ec1 53#include "exec/ram_addr.h"
0445259b 54#include "hw/acpi/acpi.h"
aa8dc044 55#include "qemu/host-utils.h"
0dc3f44a 56#include "qemu/rcu_queue.h"
ad96090a 57
3a697f69
OW
58#ifdef DEBUG_ARCH_INIT
59#define DPRINTF(fmt, ...) \
60 do { fprintf(stdout, "arch_init: " fmt, ## __VA_ARGS__); } while (0)
61#else
62#define DPRINTF(fmt, ...) \
63 do { } while (0)
64#endif
65
ad96090a
BS
66#ifdef TARGET_SPARC
67int graphic_width = 1024;
68int graphic_height = 768;
69int graphic_depth = 8;
70#else
71int graphic_width = 800;
72int graphic_height = 600;
f1ff0e89 73int graphic_depth = 32;
ad96090a
BS
74#endif
75
ad96090a
BS
76
77#if defined(TARGET_ALPHA)
78#define QEMU_ARCH QEMU_ARCH_ALPHA
79#elif defined(TARGET_ARM)
80#define QEMU_ARCH QEMU_ARCH_ARM
81#elif defined(TARGET_CRIS)
82#define QEMU_ARCH QEMU_ARCH_CRIS
83#elif defined(TARGET_I386)
84#define QEMU_ARCH QEMU_ARCH_I386
85#elif defined(TARGET_M68K)
86#define QEMU_ARCH QEMU_ARCH_M68K
81ea0e13
MW
87#elif defined(TARGET_LM32)
88#define QEMU_ARCH QEMU_ARCH_LM32
ad96090a
BS
89#elif defined(TARGET_MICROBLAZE)
90#define QEMU_ARCH QEMU_ARCH_MICROBLAZE
91#elif defined(TARGET_MIPS)
92#define QEMU_ARCH QEMU_ARCH_MIPS
d15a9c23
AG
93#elif defined(TARGET_MOXIE)
94#define QEMU_ARCH QEMU_ARCH_MOXIE
e67db06e
JL
95#elif defined(TARGET_OPENRISC)
96#define QEMU_ARCH QEMU_ARCH_OPENRISC
ad96090a
BS
97#elif defined(TARGET_PPC)
98#define QEMU_ARCH QEMU_ARCH_PPC
99#elif defined(TARGET_S390X)
100#define QEMU_ARCH QEMU_ARCH_S390X
101#elif defined(TARGET_SH4)
102#define QEMU_ARCH QEMU_ARCH_SH4
103#elif defined(TARGET_SPARC)
104#define QEMU_ARCH QEMU_ARCH_SPARC
2328826b
MF
105#elif defined(TARGET_XTENSA)
106#define QEMU_ARCH QEMU_ARCH_XTENSA
4f23a1e6
GX
107#elif defined(TARGET_UNICORE32)
108#define QEMU_ARCH QEMU_ARCH_UNICORE32
48e06fe0
BK
109#elif defined(TARGET_TRICORE)
110#define QEMU_ARCH QEMU_ARCH_TRICORE
ad96090a
BS
111#endif
112
113const uint32_t arch_type = QEMU_ARCH;
7ca1dfad
CV
114static bool mig_throttle_on;
115static int dirty_rate_high_cnt;
116static void check_guest_throttling(void);
ad96090a 117
71411d35
C
118static uint64_t bitmap_sync_count;
119
ad96090a
BS
120/***********************************************************/
121/* ram save/restore */
122
d20878d2
YT
123#define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */
124#define RAM_SAVE_FLAG_COMPRESS 0x02
125#define RAM_SAVE_FLAG_MEM_SIZE 0x04
126#define RAM_SAVE_FLAG_PAGE 0x08
127#define RAM_SAVE_FLAG_EOS 0x10
128#define RAM_SAVE_FLAG_CONTINUE 0x20
17ad9b35 129#define RAM_SAVE_FLAG_XBZRLE 0x40
0033b8b4 130/* 0x80 is reserved in migration.h start with 0x100 next */
3fcb38c2 131#define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100
ad96090a 132
756557de
EH
133static struct defconfig_file {
134 const char *filename;
f29a5614
EH
135 /* Indicates it is an user config file (disabled by -no-user-config) */
136 bool userconfig;
756557de 137} default_config_files[] = {
f29a5614 138 { CONFIG_QEMU_CONFDIR "/qemu.conf", true },
2e59915d 139 { CONFIG_QEMU_CONFDIR "/target-" TARGET_NAME ".conf", true },
756557de
EH
140 { NULL }, /* end of list */
141};
142
6d3cb1f9 143static const uint8_t ZERO_TARGET_PAGE[TARGET_PAGE_SIZE];
756557de 144
f29a5614 145int qemu_read_default_config_files(bool userconfig)
b5a8fe5e
EH
146{
147 int ret;
756557de 148 struct defconfig_file *f;
b5a8fe5e 149
756557de 150 for (f = default_config_files; f->filename; f++) {
f29a5614
EH
151 if (!userconfig && f->userconfig) {
152 continue;
153 }
756557de
EH
154 ret = qemu_read_config_file(f->filename);
155 if (ret < 0 && ret != -ENOENT) {
156 return ret;
157 }
b5a8fe5e 158 }
4d8b3c63 159
b5a8fe5e
EH
160 return 0;
161}
162
dc3c26a4 163static inline bool is_zero_range(uint8_t *p, uint64_t size)
ad96090a 164{
dc3c26a4 165 return buffer_find_nonzero_offset(p, size) == size;
ad96090a
BS
166}
167
17ad9b35
OW
168/* struct contains XBZRLE cache and a static page
169 used by the compression */
170static struct {
171 /* buffer used for XBZRLE encoding */
172 uint8_t *encoded_buf;
173 /* buffer for storing page content */
174 uint8_t *current_buf;
fd8cec93 175 /* Cache for XBZRLE, Protected by lock. */
17ad9b35 176 PageCache *cache;
fd8cec93 177 QemuMutex lock;
d97326ee
DDAG
178} XBZRLE;
179
905f26f2
GA
180/* buffer used for XBZRLE decoding */
181static uint8_t *xbzrle_decoded_buf;
9e1ba4cc 182
fd8cec93
GA
183static void XBZRLE_cache_lock(void)
184{
185 if (migrate_use_xbzrle())
186 qemu_mutex_lock(&XBZRLE.lock);
187}
188
189static void XBZRLE_cache_unlock(void)
190{
191 if (migrate_use_xbzrle())
192 qemu_mutex_unlock(&XBZRLE.lock);
193}
194
d97326ee
DDAG
195/*
196 * called from qmp_migrate_set_cache_size in main thread, possibly while
197 * a migration is in progress.
198 * A running migration maybe using the cache and might finish during this
199 * call, hence changes to the cache are protected by XBZRLE.lock().
200 */
9e1ba4cc
OW
201int64_t xbzrle_cache_resize(int64_t new_size)
202{
d97326ee
DDAG
203 PageCache *new_cache;
204 int64_t ret;
fd8cec93 205
c91e681a
OW
206 if (new_size < TARGET_PAGE_SIZE) {
207 return -1;
208 }
209
d97326ee
DDAG
210 XBZRLE_cache_lock();
211
9e1ba4cc 212 if (XBZRLE.cache != NULL) {
fd8cec93 213 if (pow2floor(new_size) == migrate_xbzrle_cache_size()) {
d97326ee 214 goto out_new_size;
fd8cec93
GA
215 }
216 new_cache = cache_init(new_size / TARGET_PAGE_SIZE,
217 TARGET_PAGE_SIZE);
218 if (!new_cache) {
d97326ee
DDAG
219 error_report("Error creating cache");
220 ret = -1;
221 goto out;
fd8cec93 222 }
fd8cec93 223
d97326ee
DDAG
224 cache_fini(XBZRLE.cache);
225 XBZRLE.cache = new_cache;
9e1ba4cc 226 }
fd8cec93 227
d97326ee
DDAG
228out_new_size:
229 ret = pow2floor(new_size);
230out:
231 XBZRLE_cache_unlock();
232 return ret;
9e1ba4cc
OW
233}
234
004d4c10
OW
235/* accounting for migration statistics */
236typedef struct AccountingInfo {
237 uint64_t dup_pages;
f1c72795 238 uint64_t skipped_pages;
004d4c10
OW
239 uint64_t norm_pages;
240 uint64_t iterations;
f36d55af
OW
241 uint64_t xbzrle_bytes;
242 uint64_t xbzrle_pages;
243 uint64_t xbzrle_cache_miss;
8bc39233 244 double xbzrle_cache_miss_rate;
f36d55af 245 uint64_t xbzrle_overflows;
004d4c10
OW
246} AccountingInfo;
247
248static AccountingInfo acct_info;
249
250static void acct_clear(void)
251{
252 memset(&acct_info, 0, sizeof(acct_info));
253}
254
255uint64_t dup_mig_bytes_transferred(void)
256{
257 return acct_info.dup_pages * TARGET_PAGE_SIZE;
258}
259
260uint64_t dup_mig_pages_transferred(void)
261{
262 return acct_info.dup_pages;
263}
264
f1c72795
PL
265uint64_t skipped_mig_bytes_transferred(void)
266{
267 return acct_info.skipped_pages * TARGET_PAGE_SIZE;
268}
269
270uint64_t skipped_mig_pages_transferred(void)
271{
272 return acct_info.skipped_pages;
273}
274
004d4c10
OW
275uint64_t norm_mig_bytes_transferred(void)
276{
277 return acct_info.norm_pages * TARGET_PAGE_SIZE;
278}
279
280uint64_t norm_mig_pages_transferred(void)
281{
282 return acct_info.norm_pages;
283}
284
f36d55af
OW
285uint64_t xbzrle_mig_bytes_transferred(void)
286{
287 return acct_info.xbzrle_bytes;
288}
289
290uint64_t xbzrle_mig_pages_transferred(void)
291{
292 return acct_info.xbzrle_pages;
293}
294
295uint64_t xbzrle_mig_pages_cache_miss(void)
296{
297 return acct_info.xbzrle_cache_miss;
298}
299
8bc39233
C
300double xbzrle_mig_cache_miss_rate(void)
301{
302 return acct_info.xbzrle_cache_miss_rate;
303}
304
f36d55af
OW
305uint64_t xbzrle_mig_pages_overflow(void)
306{
307 return acct_info.xbzrle_overflows;
308}
309
73bab2fc
JQ
310/* This is the last block that we have visited serching for dirty pages
311 */
312static RAMBlock *last_seen_block;
313/* This is the last block from where we have sent data */
314static RAMBlock *last_sent_block;
315static ram_addr_t last_offset;
316static unsigned long *migration_bitmap;
317static uint64_t migration_dirty_pages;
318static uint32_t last_version;
319static bool ram_bulk_stage;
320
8706d2d5 321struct CompressParam {
474ddaf6
LL
322 bool start;
323 bool done;
324 QEMUFile *file;
325 QemuMutex mutex;
326 QemuCond cond;
327 RAMBlock *block;
328 ram_addr_t offset;
8706d2d5
LL
329};
330typedef struct CompressParam CompressParam;
331
3fcb38c2
LL
332struct DecompressParam {
333 /* To be done */
334};
335typedef struct DecompressParam DecompressParam;
336
8706d2d5
LL
337static CompressParam *comp_param;
338static QemuThread *compress_threads;
474ddaf6
LL
339/* comp_done_cond is used to wake up the migration thread when
340 * one of the compression threads has finished the compression.
341 * comp_done_lock is used to co-work with comp_done_cond.
342 */
343static QemuMutex *comp_done_lock;
344static QemuCond *comp_done_cond;
345/* The empty QEMUFileOps will be used by file in CompressParam */
346static const QEMUFileOps empty_ops = { };
8706d2d5 347static bool quit_comp_thread;
3fcb38c2
LL
348static bool quit_decomp_thread;
349static DecompressParam *decomp_param;
350static QemuThread *decompress_threads;
351static uint8_t *compressed_data_buf;
8706d2d5
LL
352
353static void *do_data_compress(void *opaque)
354{
355 while (!quit_comp_thread) {
356
357 /* To be done */
358
359 }
360
361 return NULL;
362}
363
364static inline void terminate_compression_threads(void)
365{
366 quit_comp_thread = true;
367
368 /* To be done */
369}
370
371void migrate_compress_threads_join(void)
372{
373 int i, thread_count;
374
375 if (!migrate_use_compression()) {
376 return;
377 }
378 terminate_compression_threads();
379 thread_count = migrate_compress_threads();
380 for (i = 0; i < thread_count; i++) {
381 qemu_thread_join(compress_threads + i);
474ddaf6
LL
382 qemu_fclose(comp_param[i].file);
383 qemu_mutex_destroy(&comp_param[i].mutex);
384 qemu_cond_destroy(&comp_param[i].cond);
8706d2d5 385 }
474ddaf6
LL
386 qemu_mutex_destroy(comp_done_lock);
387 qemu_cond_destroy(comp_done_cond);
8706d2d5
LL
388 g_free(compress_threads);
389 g_free(comp_param);
474ddaf6
LL
390 g_free(comp_done_cond);
391 g_free(comp_done_lock);
8706d2d5
LL
392 compress_threads = NULL;
393 comp_param = NULL;
474ddaf6
LL
394 comp_done_cond = NULL;
395 comp_done_lock = NULL;
8706d2d5
LL
396}
397
398void migrate_compress_threads_create(void)
399{
400 int i, thread_count;
401
402 if (!migrate_use_compression()) {
403 return;
404 }
405 quit_comp_thread = false;
406 thread_count = migrate_compress_threads();
407 compress_threads = g_new0(QemuThread, thread_count);
408 comp_param = g_new0(CompressParam, thread_count);
474ddaf6
LL
409 comp_done_cond = g_new0(QemuCond, 1);
410 comp_done_lock = g_new0(QemuMutex, 1);
411 qemu_cond_init(comp_done_cond);
412 qemu_mutex_init(comp_done_lock);
8706d2d5 413 for (i = 0; i < thread_count; i++) {
474ddaf6
LL
414 /* com_param[i].file is just used as a dummy buffer to save data, set
415 * it's ops to empty.
416 */
417 comp_param[i].file = qemu_fopen_ops(NULL, &empty_ops);
418 qemu_mutex_init(&comp_param[i].mutex);
419 qemu_cond_init(&comp_param[i].cond);
8706d2d5
LL
420 qemu_thread_create(compress_threads + i, "compress",
421 do_data_compress, comp_param + i,
422 QEMU_THREAD_JOINABLE);
423 }
424}
425
f6f14c58
JQ
426/**
427 * save_page_header: Write page header to wire
428 *
429 * If this is the 1st block, it also writes the block identification
430 *
431 * Returns: Number of bytes written
432 *
433 * @f: QEMUFile where to send the data
434 * @block: block that contains the page we want to send
435 * @offset: offset inside the block for the page
436 * in the lower bits, it contains flags
437 */
438static size_t save_page_header(QEMUFile *f, RAMBlock *block, ram_addr_t offset)
0c51f43d 439{
3f7d7b09
JQ
440 size_t size;
441
f6f14c58 442 qemu_put_be64(f, offset);
3f7d7b09 443 size = 8;
0c51f43d 444
43edc0ed 445 if (!(offset & RAM_SAVE_FLAG_CONTINUE)) {
3f7d7b09
JQ
446 qemu_put_byte(f, strlen(block->idstr));
447 qemu_put_buffer(f, (uint8_t *)block->idstr,
448 strlen(block->idstr));
449 size += 1 + strlen(block->idstr);
450 }
451 return size;
0c51f43d
OW
452}
453
6d3cb1f9
DDAG
454/* Update the xbzrle cache to reflect a page that's been sent as all 0.
455 * The important thing is that a stale (not-yet-0'd) page be replaced
456 * by the new data.
457 * As a bonus, if the page wasn't in the cache it gets added so that
458 * when a small write is made into the 0'd page it gets XBZRLE sent
459 */
460static void xbzrle_cache_zero_page(ram_addr_t current_addr)
461{
462 if (ram_bulk_stage || !migrate_use_xbzrle()) {
463 return;
464 }
465
466 /* We don't care if this fails to allocate a new cache page
467 * as long as it updated an old one */
27af7d6e
C
468 cache_insert(XBZRLE.cache, current_addr, ZERO_TARGET_PAGE,
469 bitmap_sync_count);
6d3cb1f9
DDAG
470}
471
17ad9b35
OW
472#define ENCODING_FLAG_XBZRLE 0x1
473
f4be0f75
JQ
474/**
475 * save_xbzrle_page: compress and send current page
476 *
477 * Returns: 1 means that we wrote the page
478 * 0 means that page is identical to the one already sent
479 * -1 means that xbzrle would be longer than normal
480 *
481 * @f: QEMUFile where to send the data
482 * @current_data:
483 * @current_addr:
484 * @block: block that contains the page we want to send
485 * @offset: offset inside the block for the page
486 * @last_stage: if we are at the completion stage
487 * @bytes_transferred: increase it with the number of transferred bytes
488 */
1534ee93 489static int save_xbzrle_page(QEMUFile *f, uint8_t **current_data,
17ad9b35 490 ram_addr_t current_addr, RAMBlock *block,
73bab2fc 491 ram_addr_t offset, bool last_stage,
f4be0f75 492 uint64_t *bytes_transferred)
17ad9b35 493{
f4be0f75 494 int encoded_len = 0, bytes_xbzrle;
17ad9b35
OW
495 uint8_t *prev_cached_page;
496
27af7d6e 497 if (!cache_is_cached(XBZRLE.cache, current_addr, bitmap_sync_count)) {
1534ee93 498 acct_info.xbzrle_cache_miss++;
dd051c72 499 if (!last_stage) {
27af7d6e
C
500 if (cache_insert(XBZRLE.cache, current_addr, *current_data,
501 bitmap_sync_count) == -1) {
89db9987 502 return -1;
1534ee93
C
503 } else {
504 /* update *current_data when the page has been
505 inserted into cache */
506 *current_data = get_cached_data(XBZRLE.cache, current_addr);
89db9987 507 }
dd051c72 508 }
17ad9b35
OW
509 return -1;
510 }
511
512 prev_cached_page = get_cached_data(XBZRLE.cache, current_addr);
513
514 /* save current buffer into memory */
1534ee93 515 memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE);
17ad9b35
OW
516
517 /* XBZRLE encoding (if there is no overflow) */
518 encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf,
519 TARGET_PAGE_SIZE, XBZRLE.encoded_buf,
520 TARGET_PAGE_SIZE);
521 if (encoded_len == 0) {
522 DPRINTF("Skipping unmodified page\n");
523 return 0;
524 } else if (encoded_len == -1) {
525 DPRINTF("Overflow\n");
f36d55af 526 acct_info.xbzrle_overflows++;
17ad9b35 527 /* update data in the cache */
1534ee93
C
528 if (!last_stage) {
529 memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE);
530 *current_data = prev_cached_page;
531 }
17ad9b35
OW
532 return -1;
533 }
534
535 /* we need to update the data in the cache, in order to get the same data */
dd051c72
JQ
536 if (!last_stage) {
537 memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE);
538 }
17ad9b35
OW
539
540 /* Send XBZRLE based compressed page */
f6f14c58 541 bytes_xbzrle = save_page_header(f, block, offset | RAM_SAVE_FLAG_XBZRLE);
17ad9b35
OW
542 qemu_put_byte(f, ENCODING_FLAG_XBZRLE);
543 qemu_put_be16(f, encoded_len);
544 qemu_put_buffer(f, XBZRLE.encoded_buf, encoded_len);
f4be0f75 545 bytes_xbzrle += encoded_len + 1 + 2;
f36d55af 546 acct_info.xbzrle_pages++;
f4be0f75
JQ
547 acct_info.xbzrle_bytes += bytes_xbzrle;
548 *bytes_transferred += bytes_xbzrle;
17ad9b35 549
f4be0f75 550 return 1;
17ad9b35
OW
551}
552
4c8ae0f6
JQ
553static inline
554ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr,
555 ram_addr_t start)
69268cde 556{
4c8ae0f6
JQ
557 unsigned long base = mr->ram_addr >> TARGET_PAGE_BITS;
558 unsigned long nr = base + (start >> TARGET_PAGE_BITS);
0851c9f7
MT
559 uint64_t mr_size = TARGET_PAGE_ALIGN(memory_region_size(mr));
560 unsigned long size = base + (mr_size >> TARGET_PAGE_BITS);
c6bf8e0e 561
70c8652b
PL
562 unsigned long next;
563
564 if (ram_bulk_stage && nr > base) {
565 next = nr + 1;
566 } else {
567 next = find_next_bit(migration_bitmap, size, nr);
568 }
69268cde 569
4c8ae0f6
JQ
570 if (next < size) {
571 clear_bit(next, migration_bitmap);
c6bf8e0e 572 migration_dirty_pages--;
69268cde 573 }
4c8ae0f6 574 return (next - base) << TARGET_PAGE_BITS;
69268cde
JQ
575}
576
791fa2a2 577static inline bool migration_bitmap_set_dirty(ram_addr_t addr)
e44d26c8 578{
c6bf8e0e 579 bool ret;
791fa2a2 580 int nr = addr >> TARGET_PAGE_BITS;
e44d26c8 581
c6bf8e0e
JQ
582 ret = test_and_set_bit(nr, migration_bitmap);
583
584 if (!ret) {
585 migration_dirty_pages++;
e44d26c8 586 }
c6bf8e0e 587 return ret;
e44d26c8
JQ
588}
589
791fa2a2
JQ
590static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length)
591{
592 ram_addr_t addr;
aa8dc044
JQ
593 unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
594
595 /* start address is aligned at the start of a word? */
596 if (((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) {
597 int k;
598 int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS);
599 unsigned long *src = ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION];
600
601 for (k = page; k < page + nr; k++) {
602 if (src[k]) {
603 unsigned long new_dirty;
604 new_dirty = ~migration_bitmap[k];
605 migration_bitmap[k] |= src[k];
606 new_dirty &= src[k];
607 migration_dirty_pages += ctpopl(new_dirty);
608 src[k] = 0;
609 }
610 }
611 } else {
612 for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) {
613 if (cpu_physical_memory_get_dirty(start + addr,
614 TARGET_PAGE_SIZE,
615 DIRTY_MEMORY_MIGRATION)) {
616 cpu_physical_memory_reset_dirty(start + addr,
617 TARGET_PAGE_SIZE,
618 DIRTY_MEMORY_MIGRATION);
619 migration_bitmap_set_dirty(start + addr);
620 }
791fa2a2
JQ
621 }
622 }
623}
624
625
6c1b663c
C
626/* Fix me: there are too many global variables used in migration process. */
627static int64_t start_time;
628static int64_t bytes_xfer_prev;
629static int64_t num_dirty_pages_period;
630
631static void migration_bitmap_sync_init(void)
632{
633 start_time = 0;
634 bytes_xfer_prev = 0;
635 num_dirty_pages_period = 0;
636}
32c835ba 637
ae3a7047 638/* Called with iothread lock held, to protect ram_list.dirty_memory[] */
dd2df737
JQ
639static void migration_bitmap_sync(void)
640{
c6bf8e0e 641 RAMBlock *block;
c6bf8e0e 642 uint64_t num_dirty_pages_init = migration_dirty_pages;
8d017193 643 MigrationState *s = migrate_get_current();
8d017193 644 int64_t end_time;
7ca1dfad 645 int64_t bytes_xfer_now;
8bc39233
C
646 static uint64_t xbzrle_cache_miss_prev;
647 static uint64_t iterations_prev;
7ca1dfad 648
71411d35
C
649 bitmap_sync_count++;
650
7ca1dfad
CV
651 if (!bytes_xfer_prev) {
652 bytes_xfer_prev = ram_bytes_transferred();
653 }
8d017193
JQ
654
655 if (!start_time) {
bc72ad67 656 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
8d017193 657 }
3c12193d
JQ
658
659 trace_migration_bitmap_sync_start();
1d671369 660 address_space_sync_dirty_bitmap(&address_space_memory);
c6bf8e0e 661
0dc3f44a
MD
662 rcu_read_lock();
663 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
9b8424d5 664 migration_bitmap_sync_range(block->mr->ram_addr, block->used_length);
c6bf8e0e 665 }
0dc3f44a
MD
666 rcu_read_unlock();
667
c6bf8e0e 668 trace_migration_bitmap_sync_end(migration_dirty_pages
3c12193d 669 - num_dirty_pages_init);
8d017193 670 num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init;
bc72ad67 671 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
8d017193
JQ
672
673 /* more than 1 second = 1000 millisecons */
674 if (end_time > start_time + 1000) {
7ca1dfad
CV
675 if (migrate_auto_converge()) {
676 /* The following detection logic can be refined later. For now:
677 Check to see if the dirtied bytes is 50% more than the approx.
678 amount of bytes that just got transferred since the last time we
679 were in this routine. If that happens >N times (for now N==4)
680 we turn on the throttle down logic */
681 bytes_xfer_now = ram_bytes_transferred();
682 if (s->dirty_pages_rate &&
683 (num_dirty_pages_period * TARGET_PAGE_SIZE >
684 (bytes_xfer_now - bytes_xfer_prev)/2) &&
685 (dirty_rate_high_cnt++ > 4)) {
686 trace_migration_throttle();
687 mig_throttle_on = true;
688 dirty_rate_high_cnt = 0;
689 }
690 bytes_xfer_prev = bytes_xfer_now;
691 } else {
692 mig_throttle_on = false;
693 }
8bc39233
C
694 if (migrate_use_xbzrle()) {
695 if (iterations_prev != 0) {
696 acct_info.xbzrle_cache_miss_rate =
697 (double)(acct_info.xbzrle_cache_miss -
698 xbzrle_cache_miss_prev) /
699 (acct_info.iterations - iterations_prev);
700 }
701 iterations_prev = acct_info.iterations;
702 xbzrle_cache_miss_prev = acct_info.xbzrle_cache_miss;
703 }
8d017193
JQ
704 s->dirty_pages_rate = num_dirty_pages_period * 1000
705 / (end_time - start_time);
90f8ae72 706 s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE;
8d017193
JQ
707 start_time = end_time;
708 num_dirty_pages_period = 0;
58570ed8 709 s->dirty_sync_count = bitmap_sync_count;
8d017193 710 }
dd2df737
JQ
711}
712
87cf878b 713/**
14bcfdc7
DDAG
714 * ram_save_page: Send the given page to the stream
715 *
87cf878b
JQ
716 * Returns: Number of pages written.
717 *
718 * @f: QEMUFile where to send the data
719 * @block: block that contains the page we want to send
720 * @offset: offset inside the block for the page
721 * @last_stage: if we are at the completion stage
722 * @bytes_transferred: increase it with the number of transferred bytes
14bcfdc7
DDAG
723 */
724static int ram_save_page(QEMUFile *f, RAMBlock* block, ram_addr_t offset,
87cf878b 725 bool last_stage, uint64_t *bytes_transferred)
14bcfdc7 726{
87cf878b 727 int pages = -1;
6e1dea46 728 uint64_t bytes_xmit;
14bcfdc7
DDAG
729 ram_addr_t current_addr;
730 MemoryRegion *mr = block->mr;
731 uint8_t *p;
732 int ret;
733 bool send_async = true;
734
14bcfdc7
DDAG
735 p = memory_region_get_ram_ptr(mr) + offset;
736
737 /* In doubt sent page as normal */
6e1dea46 738 bytes_xmit = 0;
14bcfdc7 739 ret = ram_control_save_page(f, block->offset,
6e1dea46
JQ
740 offset, TARGET_PAGE_SIZE, &bytes_xmit);
741 if (bytes_xmit) {
87cf878b
JQ
742 *bytes_transferred += bytes_xmit;
743 pages = 1;
6e1dea46 744 }
14bcfdc7
DDAG
745
746 XBZRLE_cache_lock();
747
748 current_addr = block->offset + offset;
43edc0ed
JQ
749
750 if (block == last_sent_block) {
751 offset |= RAM_SAVE_FLAG_CONTINUE;
752 }
14bcfdc7
DDAG
753 if (ret != RAM_SAVE_CONTROL_NOT_SUPP) {
754 if (ret != RAM_SAVE_CONTROL_DELAYED) {
6e1dea46 755 if (bytes_xmit > 0) {
14bcfdc7 756 acct_info.norm_pages++;
6e1dea46 757 } else if (bytes_xmit == 0) {
14bcfdc7
DDAG
758 acct_info.dup_pages++;
759 }
760 }
761 } else if (is_zero_range(p, TARGET_PAGE_SIZE)) {
762 acct_info.dup_pages++;
f6f14c58
JQ
763 *bytes_transferred += save_page_header(f, block,
764 offset | RAM_SAVE_FLAG_COMPRESS);
14bcfdc7 765 qemu_put_byte(f, 0);
87cf878b
JQ
766 *bytes_transferred += 1;
767 pages = 1;
14bcfdc7
DDAG
768 /* Must let xbzrle know, otherwise a previous (now 0'd) cached
769 * page would be stale
770 */
771 xbzrle_cache_zero_page(current_addr);
772 } else if (!ram_bulk_stage && migrate_use_xbzrle()) {
f4be0f75 773 pages = save_xbzrle_page(f, &p, current_addr, block,
73bab2fc 774 offset, last_stage, bytes_transferred);
14bcfdc7
DDAG
775 if (!last_stage) {
776 /* Can't send this cached data async, since the cache page
777 * might get updated before it gets to the wire
778 */
779 send_async = false;
780 }
781 }
782
783 /* XBZRLE overflow or normal page */
87cf878b 784 if (pages == -1) {
f6f14c58
JQ
785 *bytes_transferred += save_page_header(f, block,
786 offset | RAM_SAVE_FLAG_PAGE);
14bcfdc7
DDAG
787 if (send_async) {
788 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE);
789 } else {
790 qemu_put_buffer(f, p, TARGET_PAGE_SIZE);
791 }
87cf878b
JQ
792 *bytes_transferred += TARGET_PAGE_SIZE;
793 pages = 1;
14bcfdc7
DDAG
794 acct_info.norm_pages++;
795 }
796
797 XBZRLE_cache_unlock();
798
87cf878b 799 return pages;
14bcfdc7
DDAG
800}
801
8706d2d5
LL
802/**
803 * ram_save_compressed_page: compress the given page and send it to the stream
804 *
805 * Returns: Number of pages written.
806 *
807 * @f: QEMUFile where to send the data
808 * @block: block that contains the page we want to send
809 * @offset: offset inside the block for the page
810 * @last_stage: if we are at the completion stage
811 * @bytes_transferred: increase it with the number of transferred bytes
812 */
813static int ram_save_compressed_page(QEMUFile *f, RAMBlock *block,
814 ram_addr_t offset, bool last_stage,
815 uint64_t *bytes_transferred)
816{
817 int pages = -1;
818
819 /* To be done*/
820
821 return pages;
822}
823
0fcd8d31
JQ
824/**
825 * ram_find_and_save_block: Finds a dirty page and sends it to f
6c779f22 826 *
0dc3f44a
MD
827 * Called within an RCU critical section.
828 *
0fcd8d31 829 * Returns: The number of pages written
b823ceaa 830 * 0 means no dirty pages
0fcd8d31
JQ
831 *
832 * @f: QEMUFile where to send the data
833 * @last_stage: if we are at the completion stage
834 * @bytes_transferred: increase it with the number of transferred bytes
6c779f22
OW
835 */
836
0fcd8d31
JQ
837static int ram_find_and_save_block(QEMUFile *f, bool last_stage,
838 uint64_t *bytes_transferred)
ad96090a 839{
b23a9a5c 840 RAMBlock *block = last_seen_block;
e44359c3 841 ram_addr_t offset = last_offset;
4c8ae0f6 842 bool complete_round = false;
87cf878b 843 int pages = 0;
71c510e2 844 MemoryRegion *mr;
ad96090a 845
e44359c3 846 if (!block)
0dc3f44a 847 block = QLIST_FIRST_RCU(&ram_list.blocks);
e44359c3 848
4c8ae0f6 849 while (true) {
71c510e2 850 mr = block->mr;
4c8ae0f6
JQ
851 offset = migration_bitmap_find_and_reset_dirty(mr, offset);
852 if (complete_round && block == last_seen_block &&
853 offset >= last_offset) {
854 break;
855 }
9b8424d5 856 if (offset >= block->used_length) {
4c8ae0f6 857 offset = 0;
0dc3f44a 858 block = QLIST_NEXT_RCU(block, next);
4c8ae0f6 859 if (!block) {
0dc3f44a 860 block = QLIST_FIRST_RCU(&ram_list.blocks);
4c8ae0f6 861 complete_round = true;
78d07ae7 862 ram_bulk_stage = false;
4c8ae0f6
JQ
863 }
864 } else {
8706d2d5
LL
865 if (migrate_use_compression()) {
866 pages = ram_save_compressed_page(f, block, offset, last_stage,
867 bytes_transferred);
868 } else {
869 pages = ram_save_page(f, block, offset, last_stage,
870 bytes_transferred);
871 }
17ad9b35 872
17ad9b35 873 /* if page is unmodified, continue to the next */
87cf878b 874 if (pages > 0) {
43edc0ed 875 last_sent_block = block;
17ad9b35
OW
876 break;
877 }
ad96090a 878 }
4c8ae0f6 879 }
ae3a7047 880
b23a9a5c 881 last_seen_block = block;
e44359c3 882 last_offset = offset;
0fcd8d31 883
87cf878b 884 return pages;
ad96090a
BS
885}
886
887static uint64_t bytes_transferred;
888
2b0ce079
MH
889void acct_update_position(QEMUFile *f, size_t size, bool zero)
890{
891 uint64_t pages = size / TARGET_PAGE_SIZE;
892 if (zero) {
893 acct_info.dup_pages += pages;
894 } else {
895 acct_info.norm_pages += pages;
896 bytes_transferred += size;
897 qemu_update_position(f, size);
898 }
899}
900
ad96090a
BS
901static ram_addr_t ram_save_remaining(void)
902{
c6bf8e0e 903 return migration_dirty_pages;
ad96090a
BS
904}
905
906uint64_t ram_bytes_remaining(void)
907{
908 return ram_save_remaining() * TARGET_PAGE_SIZE;
909}
910
911uint64_t ram_bytes_transferred(void)
912{
913 return bytes_transferred;
914}
915
916uint64_t ram_bytes_total(void)
917{
d17b5288
AW
918 RAMBlock *block;
919 uint64_t total = 0;
920
0dc3f44a
MD
921 rcu_read_lock();
922 QLIST_FOREACH_RCU(block, &ram_list.blocks, next)
9b8424d5 923 total += block->used_length;
0dc3f44a 924 rcu_read_unlock();
d17b5288 925 return total;
ad96090a
BS
926}
927
905f26f2
GA
928void free_xbzrle_decoded_buf(void)
929{
930 g_free(xbzrle_decoded_buf);
931 xbzrle_decoded_buf = NULL;
932}
933
8e21cd32
OW
934static void migration_end(void)
935{
244eaa75
PB
936 if (migration_bitmap) {
937 memory_global_dirty_log_stop();
938 g_free(migration_bitmap);
939 migration_bitmap = NULL;
940 }
17ad9b35 941
fd8cec93 942 XBZRLE_cache_lock();
244eaa75 943 if (XBZRLE.cache) {
17ad9b35 944 cache_fini(XBZRLE.cache);
17ad9b35
OW
945 g_free(XBZRLE.encoded_buf);
946 g_free(XBZRLE.current_buf);
17ad9b35 947 XBZRLE.cache = NULL;
f6c6483b
OW
948 XBZRLE.encoded_buf = NULL;
949 XBZRLE.current_buf = NULL;
17ad9b35 950 }
fd8cec93 951 XBZRLE_cache_unlock();
8e21cd32
OW
952}
953
9b5bfab0
JQ
954static void ram_migration_cancel(void *opaque)
955{
956 migration_end();
957}
958
5a170775
JQ
959static void reset_ram_globals(void)
960{
b23a9a5c 961 last_seen_block = NULL;
5f718a15 962 last_sent_block = NULL;
5a170775 963 last_offset = 0;
f798b07f 964 last_version = ram_list.version;
78d07ae7 965 ram_bulk_stage = true;
5a170775
JQ
966}
967
4508bd9e
JQ
968#define MAX_WAIT 50 /* ms, half buffered_file limit */
969
0dc3f44a
MD
970
971/* Each of ram_save_setup, ram_save_iterate and ram_save_complete has
972 * long-running RCU critical section. When rcu-reclaims in the code
973 * start to become numerous it will be necessary to reduce the
974 * granularity of these critical sections.
975 */
976
d1315aac 977static int ram_save_setup(QEMUFile *f, void *opaque)
ad96090a 978{
d1315aac 979 RAMBlock *block;
e30d1d8c 980 int64_t ram_bitmap_pages; /* Size of bitmap in pages, including gaps */
c6bf8e0e 981
7ca1dfad
CV
982 mig_throttle_on = false;
983 dirty_rate_high_cnt = 0;
71411d35 984 bitmap_sync_count = 0;
6c1b663c 985 migration_bitmap_sync_init();
ad96090a 986
17ad9b35 987 if (migrate_use_xbzrle()) {
d97326ee 988 XBZRLE_cache_lock();
17ad9b35
OW
989 XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() /
990 TARGET_PAGE_SIZE,
991 TARGET_PAGE_SIZE);
992 if (!XBZRLE.cache) {
d97326ee
DDAG
993 XBZRLE_cache_unlock();
994 error_report("Error creating cache");
17ad9b35
OW
995 return -1;
996 }
d97326ee 997 XBZRLE_cache_unlock();
a17b2fd3
OW
998
999 /* We prefer not to abort if there is no memory */
1000 XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE);
1001 if (!XBZRLE.encoded_buf) {
d97326ee 1002 error_report("Error allocating encoded_buf");
a17b2fd3
OW
1003 return -1;
1004 }
1005
1006 XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE);
1007 if (!XBZRLE.current_buf) {
d97326ee 1008 error_report("Error allocating current_buf");
a17b2fd3
OW
1009 g_free(XBZRLE.encoded_buf);
1010 XBZRLE.encoded_buf = NULL;
1011 return -1;
1012 }
1013
004d4c10 1014 acct_clear();
17ad9b35
OW
1015 }
1016
ae3a7047 1017 /* iothread lock needed for ram_list.dirty_memory[] */
9b095037
PB
1018 qemu_mutex_lock_iothread();
1019 qemu_mutex_lock_ramlist();
0dc3f44a 1020 rcu_read_lock();
9b095037
PB
1021 bytes_transferred = 0;
1022 reset_ram_globals();
1023
e30d1d8c
DDAG
1024 ram_bitmap_pages = last_ram_offset() >> TARGET_PAGE_BITS;
1025 migration_bitmap = bitmap_new(ram_bitmap_pages);
1026 bitmap_set(migration_bitmap, 0, ram_bitmap_pages);
1027
1028 /*
1029 * Count the total number of pages used by ram blocks not including any
1030 * gaps due to alignment or unplugs.
1031 */
f54a235f 1032 migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS;
e30d1d8c 1033
d1315aac 1034 memory_global_dirty_log_start();
c6bf8e0e 1035 migration_bitmap_sync();
0dc3f44a 1036 qemu_mutex_unlock_ramlist();
9b095037 1037 qemu_mutex_unlock_iothread();
ad96090a 1038
d1315aac 1039 qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE);
97ab12d4 1040
0dc3f44a 1041 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
d1315aac
JQ
1042 qemu_put_byte(f, strlen(block->idstr));
1043 qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr));
9b8424d5 1044 qemu_put_be64(f, block->used_length);
ad96090a
BS
1045 }
1046
0dc3f44a 1047 rcu_read_unlock();
0033b8b4
MH
1048
1049 ram_control_before_iterate(f, RAM_CONTROL_SETUP);
1050 ram_control_after_iterate(f, RAM_CONTROL_SETUP);
1051
d1315aac
JQ
1052 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
1053
1054 return 0;
1055}
1056
16310a3c 1057static int ram_save_iterate(QEMUFile *f, void *opaque)
d1315aac 1058{
d1315aac
JQ
1059 int ret;
1060 int i;
e4ed1541 1061 int64_t t0;
0fcd8d31 1062 int pages_sent = 0;
d1315aac 1063
0dc3f44a 1064 rcu_read_lock();
f798b07f
UD
1065 if (ram_list.version != last_version) {
1066 reset_ram_globals();
1067 }
1068
0dc3f44a
MD
1069 /* Read version before ram_list.blocks */
1070 smp_rmb();
1071
0033b8b4
MH
1072 ram_control_before_iterate(f, RAM_CONTROL_ROUND);
1073
bc72ad67 1074 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
4508bd9e 1075 i = 0;
2975725f 1076 while ((ret = qemu_file_rate_limit(f)) == 0) {
0fcd8d31 1077 int pages;
ad96090a 1078
0fcd8d31
JQ
1079 pages = ram_find_and_save_block(f, false, &bytes_transferred);
1080 /* no more pages to sent */
1081 if (pages == 0) {
ad96090a
BS
1082 break;
1083 }
0fcd8d31 1084 pages_sent += pages;
004d4c10 1085 acct_info.iterations++;
7ca1dfad 1086 check_guest_throttling();
4508bd9e
JQ
1087 /* we want to check in the 1st loop, just in case it was the 1st time
1088 and we had to sync the dirty bitmap.
1089 qemu_get_clock_ns() is a bit expensive, so we only check each some
1090 iterations
1091 */
1092 if ((i & 63) == 0) {
bc72ad67 1093 uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000;
4508bd9e 1094 if (t1 > MAX_WAIT) {
ef37a699 1095 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n",
4508bd9e
JQ
1096 t1, i);
1097 break;
1098 }
1099 }
1100 i++;
ad96090a 1101 }
0dc3f44a 1102 rcu_read_unlock();
fb3409de 1103
0033b8b4
MH
1104 /*
1105 * Must occur before EOS (or any QEMUFile operation)
1106 * because of RDMA protocol.
1107 */
1108 ram_control_after_iterate(f, RAM_CONTROL_ROUND);
1109
6cd0beda
LL
1110 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
1111 bytes_transferred += 8;
1112
1113 ret = qemu_file_get_error(f);
2975725f
JQ
1114 if (ret < 0) {
1115 return ret;
1116 }
1117
0fcd8d31 1118 return pages_sent;
16310a3c
JQ
1119}
1120
ae3a7047 1121/* Called with iothread lock */
16310a3c
JQ
1122static int ram_save_complete(QEMUFile *f, void *opaque)
1123{
0dc3f44a
MD
1124 rcu_read_lock();
1125
9c339485 1126 migration_bitmap_sync();
b2a8658e 1127
0033b8b4
MH
1128 ram_control_before_iterate(f, RAM_CONTROL_FINISH);
1129
ad96090a 1130 /* try transferring iterative blocks of memory */
3a697f69 1131
16310a3c 1132 /* flush all remaining blocks regardless of rate limiting */
6c779f22 1133 while (true) {
0fcd8d31 1134 int pages;
3fc250b4 1135
0fcd8d31 1136 pages = ram_find_and_save_block(f, true, &bytes_transferred);
6c779f22 1137 /* no more blocks to sent */
0fcd8d31 1138 if (pages == 0) {
6c779f22 1139 break;
ad96090a 1140 }
ad96090a 1141 }
0033b8b4
MH
1142
1143 ram_control_after_iterate(f, RAM_CONTROL_FINISH);
244eaa75 1144 migration_end();
ad96090a 1145
0dc3f44a 1146 rcu_read_unlock();
ad96090a
BS
1147 qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
1148
5b3c9638 1149 return 0;
ad96090a
BS
1150}
1151
e4ed1541
JQ
1152static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
1153{
1154 uint64_t remaining_size;
1155
1156 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
1157
1158 if (remaining_size < max_size) {
32c835ba 1159 qemu_mutex_lock_iothread();
0dc3f44a 1160 rcu_read_lock();
e4ed1541 1161 migration_bitmap_sync();
0dc3f44a 1162 rcu_read_unlock();
32c835ba 1163 qemu_mutex_unlock_iothread();
e4ed1541
JQ
1164 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
1165 }
1166 return remaining_size;
1167}
1168
17ad9b35
OW
1169static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host)
1170{
17ad9b35
OW
1171 unsigned int xh_len;
1172 int xh_flags;
1173
905f26f2
GA
1174 if (!xbzrle_decoded_buf) {
1175 xbzrle_decoded_buf = g_malloc(TARGET_PAGE_SIZE);
17ad9b35
OW
1176 }
1177
1178 /* extract RLE header */
1179 xh_flags = qemu_get_byte(f);
1180 xh_len = qemu_get_be16(f);
1181
1182 if (xh_flags != ENCODING_FLAG_XBZRLE) {
0971f1be 1183 error_report("Failed to load XBZRLE page - wrong compression!");
17ad9b35
OW
1184 return -1;
1185 }
1186
1187 if (xh_len > TARGET_PAGE_SIZE) {
0971f1be 1188 error_report("Failed to load XBZRLE page - len overflow!");
17ad9b35
OW
1189 return -1;
1190 }
1191 /* load data and decode */
905f26f2 1192 qemu_get_buffer(f, xbzrle_decoded_buf, xh_len);
17ad9b35
OW
1193
1194 /* decode RLE */
fb626663
CG
1195 if (xbzrle_decode_buffer(xbzrle_decoded_buf, xh_len, host,
1196 TARGET_PAGE_SIZE) == -1) {
0971f1be 1197 error_report("Failed to load XBZRLE page - decode error!");
fb626663 1198 return -1;
17ad9b35
OW
1199 }
1200
fb626663 1201 return 0;
17ad9b35
OW
1202}
1203
0dc3f44a
MD
1204/* Must be called from within a rcu critical section.
1205 * Returns a pointer from within the RCU-protected ram_list.
1206 */
a55bbe31
AW
1207static inline void *host_from_stream_offset(QEMUFile *f,
1208 ram_addr_t offset,
1209 int flags)
1210{
1211 static RAMBlock *block = NULL;
1212 char id[256];
1213 uint8_t len;
1214
1215 if (flags & RAM_SAVE_FLAG_CONTINUE) {
9b8424d5 1216 if (!block || block->max_length <= offset) {
0971f1be 1217 error_report("Ack, bad migration stream!");
a55bbe31
AW
1218 return NULL;
1219 }
1220
dc94a7ed 1221 return memory_region_get_ram_ptr(block->mr) + offset;
a55bbe31
AW
1222 }
1223
1224 len = qemu_get_byte(f);
1225 qemu_get_buffer(f, (uint8_t *)id, len);
1226 id[len] = 0;
1227
0dc3f44a 1228 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
9b8424d5
MT
1229 if (!strncmp(id, block->idstr, sizeof(id)) &&
1230 block->max_length > offset) {
dc94a7ed 1231 return memory_region_get_ram_ptr(block->mr) + offset;
0be839a2 1232 }
a55bbe31
AW
1233 }
1234
0971f1be 1235 error_report("Can't find block %s!", id);
a55bbe31
AW
1236 return NULL;
1237}
1238
44c3b58c
MH
1239/*
1240 * If a page (or a whole RDMA chunk) has been
1241 * determined to be zero, then zap it.
1242 */
1243void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
1244{
d613a56f 1245 if (ch != 0 || !is_zero_range(host, size)) {
44c3b58c 1246 memset(host, ch, size);
44c3b58c
MH
1247 }
1248}
1249
3fcb38c2
LL
1250static void *do_data_decompress(void *opaque)
1251{
1252 while (!quit_decomp_thread) {
1253 /* To be done */
1254 }
1255
1256 return NULL;
1257}
1258
1259void migrate_decompress_threads_create(void)
1260{
1261 int i, thread_count;
1262
1263 thread_count = migrate_decompress_threads();
1264 decompress_threads = g_new0(QemuThread, thread_count);
1265 decomp_param = g_new0(DecompressParam, thread_count);
1266 compressed_data_buf = g_malloc0(compressBound(TARGET_PAGE_SIZE));
1267 quit_decomp_thread = false;
1268 for (i = 0; i < thread_count; i++) {
1269 qemu_thread_create(decompress_threads + i, "decompress",
1270 do_data_decompress, decomp_param + i,
1271 QEMU_THREAD_JOINABLE);
1272 }
1273}
1274
1275void migrate_decompress_threads_join(void)
1276{
1277 int i, thread_count;
1278
1279 quit_decomp_thread = true;
1280 thread_count = migrate_decompress_threads();
1281 for (i = 0; i < thread_count; i++) {
1282 qemu_thread_join(decompress_threads + i);
1283 }
1284 g_free(decompress_threads);
1285 g_free(decomp_param);
1286 g_free(compressed_data_buf);
1287 decompress_threads = NULL;
1288 decomp_param = NULL;
1289 compressed_data_buf = NULL;
1290}
1291
1292static void decompress_data_with_multi_threads(uint8_t *compbuf,
1293 void *host, int len)
1294{
1295 /* To be done */
1296}
1297
7908c78d 1298static int ram_load(QEMUFile *f, void *opaque, int version_id)
ad96090a 1299{
5b0e9dd4 1300 int flags = 0, ret = 0;
3a697f69 1301 static uint64_t seq_iter;
3fcb38c2 1302 int len = 0;
3a697f69
OW
1303
1304 seq_iter++;
ad96090a 1305
21a246a4 1306 if (version_id != 4) {
4798fe55 1307 ret = -EINVAL;
ad96090a
BS
1308 }
1309
0dc3f44a
MD
1310 /* This RCU critical section can be very long running.
1311 * When RCU reclaims in the code start to become numerous,
1312 * it will be necessary to reduce the granularity of this
1313 * critical section.
1314 */
1315 rcu_read_lock();
5b0e9dd4
PL
1316 while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) {
1317 ram_addr_t addr, total_ram_bytes;
1318 void *host;
1319 uint8_t ch;
ad96090a 1320
5b0e9dd4 1321 addr = qemu_get_be64(f);
ad96090a
BS
1322 flags = addr & ~TARGET_PAGE_MASK;
1323 addr &= TARGET_PAGE_MASK;
1324
5b0e9dd4
PL
1325 switch (flags & ~RAM_SAVE_FLAG_CONTINUE) {
1326 case RAM_SAVE_FLAG_MEM_SIZE:
21a246a4 1327 /* Synchronize RAM block list */
5b0e9dd4
PL
1328 total_ram_bytes = addr;
1329 while (!ret && total_ram_bytes) {
21a246a4
C
1330 RAMBlock *block;
1331 uint8_t len;
5b0e9dd4
PL
1332 char id[256];
1333 ram_addr_t length;
21a246a4
C
1334
1335 len = qemu_get_byte(f);
1336 qemu_get_buffer(f, (uint8_t *)id, len);
1337 id[len] = 0;
1338 length = qemu_get_be64(f);
1339
0dc3f44a 1340 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
21a246a4 1341 if (!strncmp(id, block->idstr, sizeof(id))) {
b0cc3f83
MT
1342 if (length != block->used_length) {
1343 Error *local_err = NULL;
1344
1345 ret = qemu_ram_resize(block->offset, length, &local_err);
1346 if (local_err) {
565f65d2 1347 error_report_err(local_err);
b0cc3f83 1348 }
97ab12d4 1349 }
21a246a4 1350 break;
97ab12d4 1351 }
21a246a4 1352 }
97ab12d4 1353
21a246a4 1354 if (!block) {
0971f1be
LT
1355 error_report("Unknown ramblock \"%s\", cannot "
1356 "accept migration", id);
21a246a4 1357 ret = -EINVAL;
db80face 1358 }
21a246a4
C
1359
1360 total_ram_bytes -= length;
ad96090a 1361 }
5b0e9dd4
PL
1362 break;
1363 case RAM_SAVE_FLAG_COMPRESS:
f09f2189 1364 host = host_from_stream_offset(f, addr, flags);
492fb99c 1365 if (!host) {
db80face 1366 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr);
4798fe55 1367 ret = -EINVAL;
db80face 1368 break;
492fb99c 1369 }
97ab12d4 1370 ch = qemu_get_byte(f);
44c3b58c 1371 ram_handle_compressed(host, ch, TARGET_PAGE_SIZE);
5b0e9dd4
PL
1372 break;
1373 case RAM_SAVE_FLAG_PAGE:
f09f2189 1374 host = host_from_stream_offset(f, addr, flags);
0ff1f9f5 1375 if (!host) {
db80face 1376 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr);
4798fe55 1377 ret = -EINVAL;
db80face 1378 break;
0ff1f9f5 1379 }
97ab12d4 1380 qemu_get_buffer(f, host, TARGET_PAGE_SIZE);
5b0e9dd4 1381 break;
3fcb38c2
LL
1382 case RAM_SAVE_FLAG_COMPRESS_PAGE:
1383 host = host_from_stream_offset(f, addr, flags);
1384 if (!host) {
1385 error_report("Invalid RAM offset " RAM_ADDR_FMT, addr);
1386 ret = -EINVAL;
1387 break;
1388 }
1389
1390 len = qemu_get_be32(f);
1391 if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) {
1392 error_report("Invalid compressed data length: %d", len);
1393 ret = -EINVAL;
1394 break;
1395 }
1396 qemu_get_buffer(f, compressed_data_buf, len);
1397 decompress_data_with_multi_threads(compressed_data_buf, host, len);
1398 break;
5b0e9dd4
PL
1399 case RAM_SAVE_FLAG_XBZRLE:
1400 host = host_from_stream_offset(f, addr, flags);
17ad9b35 1401 if (!host) {
db80face 1402 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr);
4798fe55 1403 ret = -EINVAL;
db80face 1404 break;
17ad9b35 1405 }
17ad9b35 1406 if (load_xbzrle(f, addr, host) < 0) {
db80face
PL
1407 error_report("Failed to decompress XBZRLE page at "
1408 RAM_ADDR_FMT, addr);
17ad9b35 1409 ret = -EINVAL;
db80face 1410 break;
17ad9b35 1411 }
db80face 1412 break;
5b0e9dd4
PL
1413 case RAM_SAVE_FLAG_EOS:
1414 /* normal exit */
db80face 1415 break;
5b0e9dd4
PL
1416 default:
1417 if (flags & RAM_SAVE_FLAG_HOOK) {
1418 ram_control_load_hook(f, flags);
1419 } else {
1420 error_report("Unknown combination of migration flags: %#x",
1421 flags);
1422 ret = -EINVAL;
1423 }
1424 }
1425 if (!ret) {
1426 ret = qemu_file_get_error(f);
ad96090a 1427 }
db80face 1428 }
ad96090a 1429
0dc3f44a 1430 rcu_read_unlock();
ef37a699
IM
1431 DPRINTF("Completed load of VM with exit code %d seq iteration "
1432 "%" PRIu64 "\n", ret, seq_iter);
3a697f69 1433 return ret;
ad96090a
BS
1434}
1435
0d6ab3ab 1436static SaveVMHandlers savevm_ram_handlers = {
d1315aac 1437 .save_live_setup = ram_save_setup,
16310a3c
JQ
1438 .save_live_iterate = ram_save_iterate,
1439 .save_live_complete = ram_save_complete,
e4ed1541 1440 .save_live_pending = ram_save_pending,
7908c78d 1441 .load_state = ram_load,
9b5bfab0 1442 .cancel = ram_migration_cancel,
7908c78d
JQ
1443};
1444
0d6ab3ab
DDAG
1445void ram_mig_init(void)
1446{
d97326ee 1447 qemu_mutex_init(&XBZRLE.lock);
0d6ab3ab
DDAG
1448 register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, NULL);
1449}
1450
0dfa5ef9
IY
1451struct soundhw {
1452 const char *name;
1453 const char *descr;
1454 int enabled;
1455 int isa;
1456 union {
4a0f031d 1457 int (*init_isa) (ISABus *bus);
0dfa5ef9
IY
1458 int (*init_pci) (PCIBus *bus);
1459 } init;
1460};
1461
36cd6f6f
PB
1462static struct soundhw soundhw[9];
1463static int soundhw_count;
ad96090a 1464
36cd6f6f
PB
1465void isa_register_soundhw(const char *name, const char *descr,
1466 int (*init_isa)(ISABus *bus))
1467{
1468 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1469 soundhw[soundhw_count].name = name;
1470 soundhw[soundhw_count].descr = descr;
1471 soundhw[soundhw_count].isa = 1;
1472 soundhw[soundhw_count].init.init_isa = init_isa;
1473 soundhw_count++;
1474}
ad96090a 1475
36cd6f6f
PB
1476void pci_register_soundhw(const char *name, const char *descr,
1477 int (*init_pci)(PCIBus *bus))
1478{
1479 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1480 soundhw[soundhw_count].name = name;
1481 soundhw[soundhw_count].descr = descr;
1482 soundhw[soundhw_count].isa = 0;
1483 soundhw[soundhw_count].init.init_pci = init_pci;
1484 soundhw_count++;
1485}
ad96090a
BS
1486
1487void select_soundhw(const char *optarg)
1488{
1489 struct soundhw *c;
1490
c8057f95 1491 if (is_help_option(optarg)) {
ad96090a
BS
1492 show_valid_cards:
1493
36cd6f6f
PB
1494 if (soundhw_count) {
1495 printf("Valid sound card names (comma separated):\n");
1496 for (c = soundhw; c->name; ++c) {
1497 printf ("%-11s %s\n", c->name, c->descr);
1498 }
1499 printf("\n-soundhw all will enable all of the above\n");
1500 } else {
1501 printf("Machine has no user-selectable audio hardware "
1502 "(it may or may not have always-present audio hardware).\n");
ad96090a 1503 }
c8057f95 1504 exit(!is_help_option(optarg));
ad96090a
BS
1505 }
1506 else {
1507 size_t l;
1508 const char *p;
1509 char *e;
1510 int bad_card = 0;
1511
1512 if (!strcmp(optarg, "all")) {
1513 for (c = soundhw; c->name; ++c) {
1514 c->enabled = 1;
1515 }
1516 return;
1517 }
1518
1519 p = optarg;
1520 while (*p) {
1521 e = strchr(p, ',');
1522 l = !e ? strlen(p) : (size_t) (e - p);
1523
1524 for (c = soundhw; c->name; ++c) {
1525 if (!strncmp(c->name, p, l) && !c->name[l]) {
1526 c->enabled = 1;
1527 break;
1528 }
1529 }
1530
1531 if (!c->name) {
1532 if (l > 80) {
0971f1be 1533 error_report("Unknown sound card name (too big to show)");
ad96090a
BS
1534 }
1535 else {
0971f1be
LT
1536 error_report("Unknown sound card name `%.*s'",
1537 (int) l, p);
ad96090a
BS
1538 }
1539 bad_card = 1;
1540 }
1541 p += l + (e != NULL);
1542 }
1543
1544 if (bad_card) {
1545 goto show_valid_cards;
1546 }
1547 }
1548}
0dfa5ef9 1549
f81222bc 1550void audio_init(void)
0dfa5ef9
IY
1551{
1552 struct soundhw *c;
f81222bc
PB
1553 ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL);
1554 PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL);
0dfa5ef9
IY
1555
1556 for (c = soundhw; c->name; ++c) {
1557 if (c->enabled) {
1558 if (c->isa) {
f81222bc 1559 if (!isa_bus) {
0971f1be 1560 error_report("ISA bus not available for %s", c->name);
f81222bc 1561 exit(1);
0dfa5ef9 1562 }
f81222bc 1563 c->init.init_isa(isa_bus);
0dfa5ef9 1564 } else {
f81222bc 1565 if (!pci_bus) {
0971f1be 1566 error_report("PCI bus not available for %s", c->name);
f81222bc 1567 exit(1);
0dfa5ef9 1568 }
f81222bc 1569 c->init.init_pci(pci_bus);
0dfa5ef9
IY
1570 }
1571 }
1572 }
1573}
ad96090a
BS
1574
1575int qemu_uuid_parse(const char *str, uint8_t *uuid)
1576{
1577 int ret;
1578
1579 if (strlen(str) != 36) {
1580 return -1;
1581 }
1582
1583 ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3],
1584 &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9],
1585 &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14],
1586 &uuid[15]);
1587
1588 if (ret != 16) {
1589 return -1;
1590 }
ad96090a
BS
1591 return 0;
1592}
1593
0c764a9d 1594void do_acpitable_option(const QemuOpts *opts)
ad96090a
BS
1595{
1596#ifdef TARGET_I386
23084327
LE
1597 Error *err = NULL;
1598
1599 acpi_table_add(opts, &err);
1600 if (err) {
4a44d85e
SA
1601 error_report("Wrong acpi table provided: %s",
1602 error_get_pretty(err));
23084327 1603 error_free(err);
ad96090a
BS
1604 exit(1);
1605 }
1606#endif
1607}
1608
4f953d2f 1609void do_smbios_option(QemuOpts *opts)
ad96090a
BS
1610{
1611#ifdef TARGET_I386
4f953d2f 1612 smbios_entry_add(opts);
ad96090a
BS
1613#endif
1614}
1615
1616void cpudef_init(void)
1617{
1618#if defined(cpudef_setup)
1619 cpudef_setup(); /* parse cpu definitions in target config file */
1620#endif
1621}
1622
ad96090a
BS
1623int kvm_available(void)
1624{
1625#ifdef CONFIG_KVM
1626 return 1;
1627#else
1628 return 0;
1629#endif
1630}
1631
1632int xen_available(void)
1633{
1634#ifdef CONFIG_XEN
1635 return 1;
1636#else
1637 return 0;
1638#endif
1639}
99afc91d
DB
1640
1641
1642TargetInfo *qmp_query_target(Error **errp)
1643{
1644 TargetInfo *info = g_malloc0(sizeof(*info));
1645
c02a9552 1646 info->arch = g_strdup(TARGET_NAME);
99afc91d
DB
1647
1648 return info;
1649}
7ca1dfad
CV
1650
1651/* Stub function that's gets run on the vcpu when its brought out of the
1652 VM to run inside qemu via async_run_on_cpu()*/
1653static void mig_sleep_cpu(void *opq)
1654{
1655 qemu_mutex_unlock_iothread();
1656 g_usleep(30*1000);
1657 qemu_mutex_lock_iothread();
1658}
1659
1660/* To reduce the dirty rate explicitly disallow the VCPUs from spending
1661 much time in the VM. The migration thread will try to catchup.
1662 Workload will experience a performance drop.
1663*/
7ca1dfad
CV
1664static void mig_throttle_guest_down(void)
1665{
38fcbd3f
AF
1666 CPUState *cpu;
1667
7ca1dfad 1668 qemu_mutex_lock_iothread();
38fcbd3f
AF
1669 CPU_FOREACH(cpu) {
1670 async_run_on_cpu(cpu, mig_sleep_cpu, NULL);
1671 }
7ca1dfad
CV
1672 qemu_mutex_unlock_iothread();
1673}
1674
1675static void check_guest_throttling(void)
1676{
1677 static int64_t t0;
1678 int64_t t1;
1679
1680 if (!mig_throttle_on) {
1681 return;
1682 }
1683
1684 if (!t0) {
bc72ad67 1685 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
7ca1dfad
CV
1686 return;
1687 }
1688
bc72ad67 1689 t1 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
7ca1dfad
CV
1690
1691 /* If it has been more than 40 ms since the last time the guest
1692 * was throttled then do it again.
1693 */
1694 if (40 < (t1-t0)/1000000) {
1695 mig_throttle_guest_down();
1696 t0 = t1;
1697 }
1698}