]>
Commit | Line | Data |
---|---|---|
56e93d26 JQ |
1 | /* |
2 | * QEMU System Emulator | |
3 | * | |
4 | * Copyright (c) 2003-2008 Fabrice Bellard | |
76cc7b58 JQ |
5 | * Copyright (c) 2011-2015 Red Hat Inc |
6 | * | |
7 | * Authors: | |
8 | * Juan Quintela <quintela@redhat.com> | |
56e93d26 JQ |
9 | * |
10 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
11 | * of this software and associated documentation files (the "Software"), to deal | |
12 | * in the Software without restriction, including without limitation the rights | |
13 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
14 | * copies of the Software, and to permit persons to whom the Software is | |
15 | * furnished to do so, subject to the following conditions: | |
16 | * | |
17 | * The above copyright notice and this permission notice shall be included in | |
18 | * all copies or substantial portions of the Software. | |
19 | * | |
20 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
21 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
22 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
23 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
24 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
25 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
26 | * THE SOFTWARE. | |
27 | */ | |
e688df6b | 28 | |
1393a485 | 29 | #include "qemu/osdep.h" |
33c11879 | 30 | #include "cpu.h" |
56e93d26 | 31 | #include <zlib.h> |
4addcd4f | 32 | #include "qapi-event.h" |
f348b6d1 | 33 | #include "qemu/cutils.h" |
56e93d26 JQ |
34 | #include "qemu/bitops.h" |
35 | #include "qemu/bitmap.h" | |
7205c9ec | 36 | #include "qemu/main-loop.h" |
709e3fe8 | 37 | #include "xbzrle.h" |
7b1e1a22 | 38 | #include "ram.h" |
6666c96a | 39 | #include "migration.h" |
f2a8f0a6 | 40 | #include "migration/register.h" |
7b1e1a22 | 41 | #include "migration/misc.h" |
08a0aee1 | 42 | #include "qemu-file.h" |
be07b0ac | 43 | #include "postcopy-ram.h" |
56e93d26 | 44 | #include "migration/page_cache.h" |
56e93d26 | 45 | #include "qemu/error-report.h" |
e688df6b | 46 | #include "qapi/error.h" |
8acabf69 | 47 | #include "qapi/qmp/qerror.h" |
56e93d26 | 48 | #include "trace.h" |
56e93d26 | 49 | #include "exec/ram_addr.h" |
f9494614 | 50 | #include "exec/target_page.h" |
56e93d26 | 51 | #include "qemu/rcu_queue.h" |
a91246c9 | 52 | #include "migration/colo.h" |
9ac78b61 | 53 | #include "migration/block.h" |
56e93d26 | 54 | |
56e93d26 JQ |
55 | /***********************************************************/ |
56 | /* ram save/restore */ | |
57 | ||
bb890ed5 JQ |
58 | /* RAM_SAVE_FLAG_ZERO used to be named RAM_SAVE_FLAG_COMPRESS, it |
59 | * worked for pages that where filled with the same char. We switched | |
60 | * it to only search for the zero value. And to avoid confusion with | |
61 | * RAM_SSAVE_FLAG_COMPRESS_PAGE just rename it. | |
62 | */ | |
63 | ||
56e93d26 | 64 | #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */ |
bb890ed5 | 65 | #define RAM_SAVE_FLAG_ZERO 0x02 |
56e93d26 JQ |
66 | #define RAM_SAVE_FLAG_MEM_SIZE 0x04 |
67 | #define RAM_SAVE_FLAG_PAGE 0x08 | |
68 | #define RAM_SAVE_FLAG_EOS 0x10 | |
69 | #define RAM_SAVE_FLAG_CONTINUE 0x20 | |
70 | #define RAM_SAVE_FLAG_XBZRLE 0x40 | |
71 | /* 0x80 is reserved in migration.h start with 0x100 next */ | |
72 | #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 | |
73 | ||
56e93d26 JQ |
74 | static inline bool is_zero_range(uint8_t *p, uint64_t size) |
75 | { | |
a1febc49 | 76 | return buffer_is_zero(p, size); |
56e93d26 JQ |
77 | } |
78 | ||
9360447d JQ |
79 | XBZRLECacheStats xbzrle_counters; |
80 | ||
56e93d26 JQ |
81 | /* struct contains XBZRLE cache and a static page |
82 | used by the compression */ | |
83 | static struct { | |
84 | /* buffer used for XBZRLE encoding */ | |
85 | uint8_t *encoded_buf; | |
86 | /* buffer for storing page content */ | |
87 | uint8_t *current_buf; | |
88 | /* Cache for XBZRLE, Protected by lock. */ | |
89 | PageCache *cache; | |
90 | QemuMutex lock; | |
c00e0928 JQ |
91 | /* it will store a page full of zeros */ |
92 | uint8_t *zero_target_page; | |
f265e0e4 JQ |
93 | /* buffer used for XBZRLE decoding */ |
94 | uint8_t *decoded_buf; | |
56e93d26 JQ |
95 | } XBZRLE; |
96 | ||
56e93d26 JQ |
97 | static void XBZRLE_cache_lock(void) |
98 | { | |
99 | if (migrate_use_xbzrle()) | |
100 | qemu_mutex_lock(&XBZRLE.lock); | |
101 | } | |
102 | ||
103 | static void XBZRLE_cache_unlock(void) | |
104 | { | |
105 | if (migrate_use_xbzrle()) | |
106 | qemu_mutex_unlock(&XBZRLE.lock); | |
107 | } | |
108 | ||
3d0684b2 JQ |
109 | /** |
110 | * xbzrle_cache_resize: resize the xbzrle cache | |
111 | * | |
112 | * This function is called from qmp_migrate_set_cache_size in main | |
113 | * thread, possibly while a migration is in progress. A running | |
114 | * migration may be using the cache and might finish during this call, | |
115 | * hence changes to the cache are protected by XBZRLE.lock(). | |
116 | * | |
c9dede2d | 117 | * Returns 0 for success or -1 for error |
3d0684b2 JQ |
118 | * |
119 | * @new_size: new cache size | |
8acabf69 | 120 | * @errp: set *errp if the check failed, with reason |
56e93d26 | 121 | */ |
c9dede2d | 122 | int xbzrle_cache_resize(int64_t new_size, Error **errp) |
56e93d26 JQ |
123 | { |
124 | PageCache *new_cache; | |
c9dede2d | 125 | int64_t ret = 0; |
56e93d26 | 126 | |
8acabf69 JQ |
127 | /* Check for truncation */ |
128 | if (new_size != (size_t)new_size) { | |
129 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", | |
130 | "exceeding address space"); | |
131 | return -1; | |
132 | } | |
133 | ||
2a313e5c JQ |
134 | if (new_size == migrate_xbzrle_cache_size()) { |
135 | /* nothing to do */ | |
c9dede2d | 136 | return 0; |
2a313e5c JQ |
137 | } |
138 | ||
56e93d26 JQ |
139 | XBZRLE_cache_lock(); |
140 | ||
141 | if (XBZRLE.cache != NULL) { | |
80f8dfde | 142 | new_cache = cache_init(new_size, TARGET_PAGE_SIZE, errp); |
56e93d26 | 143 | if (!new_cache) { |
56e93d26 JQ |
144 | ret = -1; |
145 | goto out; | |
146 | } | |
147 | ||
148 | cache_fini(XBZRLE.cache); | |
149 | XBZRLE.cache = new_cache; | |
150 | } | |
56e93d26 JQ |
151 | out: |
152 | XBZRLE_cache_unlock(); | |
153 | return ret; | |
154 | } | |
155 | ||
f9494614 AP |
156 | static void ramblock_recv_map_init(void) |
157 | { | |
158 | RAMBlock *rb; | |
159 | ||
160 | RAMBLOCK_FOREACH(rb) { | |
161 | assert(!rb->receivedmap); | |
162 | rb->receivedmap = bitmap_new(rb->max_length >> qemu_target_page_bits()); | |
163 | } | |
164 | } | |
165 | ||
166 | int ramblock_recv_bitmap_test(RAMBlock *rb, void *host_addr) | |
167 | { | |
168 | return test_bit(ramblock_recv_bitmap_offset(host_addr, rb), | |
169 | rb->receivedmap); | |
170 | } | |
171 | ||
172 | void ramblock_recv_bitmap_set(RAMBlock *rb, void *host_addr) | |
173 | { | |
174 | set_bit_atomic(ramblock_recv_bitmap_offset(host_addr, rb), rb->receivedmap); | |
175 | } | |
176 | ||
177 | void ramblock_recv_bitmap_set_range(RAMBlock *rb, void *host_addr, | |
178 | size_t nr) | |
179 | { | |
180 | bitmap_set_atomic(rb->receivedmap, | |
181 | ramblock_recv_bitmap_offset(host_addr, rb), | |
182 | nr); | |
183 | } | |
184 | ||
ec481c6c JQ |
185 | /* |
186 | * An outstanding page request, on the source, having been received | |
187 | * and queued | |
188 | */ | |
189 | struct RAMSrcPageRequest { | |
190 | RAMBlock *rb; | |
191 | hwaddr offset; | |
192 | hwaddr len; | |
193 | ||
194 | QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; | |
195 | }; | |
196 | ||
6f37bb8b JQ |
197 | /* State of RAM for migration */ |
198 | struct RAMState { | |
204b88b8 JQ |
199 | /* QEMUFile used for this migration */ |
200 | QEMUFile *f; | |
6f37bb8b JQ |
201 | /* Last block that we have visited searching for dirty pages */ |
202 | RAMBlock *last_seen_block; | |
203 | /* Last block from where we have sent data */ | |
204 | RAMBlock *last_sent_block; | |
269ace29 JQ |
205 | /* Last dirty target page we have sent */ |
206 | ram_addr_t last_page; | |
6f37bb8b JQ |
207 | /* last ram version we have seen */ |
208 | uint32_t last_version; | |
209 | /* We are in the first round */ | |
210 | bool ram_bulk_stage; | |
8d820d6f JQ |
211 | /* How many times we have dirty too many pages */ |
212 | int dirty_rate_high_cnt; | |
f664da80 JQ |
213 | /* these variables are used for bitmap sync */ |
214 | /* last time we did a full bitmap_sync */ | |
215 | int64_t time_last_bitmap_sync; | |
eac74159 | 216 | /* bytes transferred at start_time */ |
c4bdf0cf | 217 | uint64_t bytes_xfer_prev; |
a66cd90c | 218 | /* number of dirty pages since start_time */ |
68908ed6 | 219 | uint64_t num_dirty_pages_period; |
b5833fde JQ |
220 | /* xbzrle misses since the beginning of the period */ |
221 | uint64_t xbzrle_cache_miss_prev; | |
36040d9c JQ |
222 | /* number of iterations at the beginning of period */ |
223 | uint64_t iterations_prev; | |
23b28c3c JQ |
224 | /* Iterations since start */ |
225 | uint64_t iterations; | |
9360447d | 226 | /* number of dirty bits in the bitmap */ |
2dfaf12e PX |
227 | uint64_t migration_dirty_pages; |
228 | /* protects modification of the bitmap */ | |
108cfae0 | 229 | QemuMutex bitmap_mutex; |
68a098f3 JQ |
230 | /* The RAMBlock used in the last src_page_requests */ |
231 | RAMBlock *last_req_rb; | |
ec481c6c JQ |
232 | /* Queue of outstanding page requests from the destination */ |
233 | QemuMutex src_page_req_mutex; | |
234 | QSIMPLEQ_HEAD(src_page_requests, RAMSrcPageRequest) src_page_requests; | |
6f37bb8b JQ |
235 | }; |
236 | typedef struct RAMState RAMState; | |
237 | ||
53518d94 | 238 | static RAMState *ram_state; |
6f37bb8b | 239 | |
9edabd4d | 240 | uint64_t ram_bytes_remaining(void) |
2f4fde93 | 241 | { |
bae416e5 DDAG |
242 | return ram_state ? (ram_state->migration_dirty_pages * TARGET_PAGE_SIZE) : |
243 | 0; | |
2f4fde93 JQ |
244 | } |
245 | ||
9360447d | 246 | MigrationStats ram_counters; |
96506894 | 247 | |
b8fb8cb7 DDAG |
248 | /* used by the search for pages to send */ |
249 | struct PageSearchStatus { | |
250 | /* Current block being searched */ | |
251 | RAMBlock *block; | |
a935e30f JQ |
252 | /* Current page to search from */ |
253 | unsigned long page; | |
b8fb8cb7 DDAG |
254 | /* Set once we wrap around */ |
255 | bool complete_round; | |
256 | }; | |
257 | typedef struct PageSearchStatus PageSearchStatus; | |
258 | ||
56e93d26 | 259 | struct CompressParam { |
56e93d26 | 260 | bool done; |
90e56fb4 | 261 | bool quit; |
56e93d26 JQ |
262 | QEMUFile *file; |
263 | QemuMutex mutex; | |
264 | QemuCond cond; | |
265 | RAMBlock *block; | |
266 | ram_addr_t offset; | |
267 | }; | |
268 | typedef struct CompressParam CompressParam; | |
269 | ||
270 | struct DecompressParam { | |
73a8912b | 271 | bool done; |
90e56fb4 | 272 | bool quit; |
56e93d26 JQ |
273 | QemuMutex mutex; |
274 | QemuCond cond; | |
275 | void *des; | |
d341d9f3 | 276 | uint8_t *compbuf; |
56e93d26 JQ |
277 | int len; |
278 | }; | |
279 | typedef struct DecompressParam DecompressParam; | |
280 | ||
281 | static CompressParam *comp_param; | |
282 | static QemuThread *compress_threads; | |
283 | /* comp_done_cond is used to wake up the migration thread when | |
284 | * one of the compression threads has finished the compression. | |
285 | * comp_done_lock is used to co-work with comp_done_cond. | |
286 | */ | |
0d9f9a5c LL |
287 | static QemuMutex comp_done_lock; |
288 | static QemuCond comp_done_cond; | |
56e93d26 JQ |
289 | /* The empty QEMUFileOps will be used by file in CompressParam */ |
290 | static const QEMUFileOps empty_ops = { }; | |
291 | ||
56e93d26 JQ |
292 | static DecompressParam *decomp_param; |
293 | static QemuThread *decompress_threads; | |
73a8912b LL |
294 | static QemuMutex decomp_done_lock; |
295 | static QemuCond decomp_done_cond; | |
56e93d26 | 296 | |
a7a9a88f LL |
297 | static int do_compress_ram_page(QEMUFile *f, RAMBlock *block, |
298 | ram_addr_t offset); | |
56e93d26 JQ |
299 | |
300 | static void *do_data_compress(void *opaque) | |
301 | { | |
302 | CompressParam *param = opaque; | |
a7a9a88f LL |
303 | RAMBlock *block; |
304 | ram_addr_t offset; | |
56e93d26 | 305 | |
a7a9a88f | 306 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 307 | while (!param->quit) { |
a7a9a88f LL |
308 | if (param->block) { |
309 | block = param->block; | |
310 | offset = param->offset; | |
311 | param->block = NULL; | |
312 | qemu_mutex_unlock(¶m->mutex); | |
313 | ||
314 | do_compress_ram_page(param->file, block, offset); | |
315 | ||
0d9f9a5c | 316 | qemu_mutex_lock(&comp_done_lock); |
a7a9a88f | 317 | param->done = true; |
0d9f9a5c LL |
318 | qemu_cond_signal(&comp_done_cond); |
319 | qemu_mutex_unlock(&comp_done_lock); | |
a7a9a88f LL |
320 | |
321 | qemu_mutex_lock(¶m->mutex); | |
322 | } else { | |
56e93d26 JQ |
323 | qemu_cond_wait(¶m->cond, ¶m->mutex); |
324 | } | |
56e93d26 | 325 | } |
a7a9a88f | 326 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
327 | |
328 | return NULL; | |
329 | } | |
330 | ||
331 | static inline void terminate_compression_threads(void) | |
332 | { | |
333 | int idx, thread_count; | |
334 | ||
335 | thread_count = migrate_compress_threads(); | |
3d0684b2 | 336 | |
56e93d26 JQ |
337 | for (idx = 0; idx < thread_count; idx++) { |
338 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 339 | comp_param[idx].quit = true; |
56e93d26 JQ |
340 | qemu_cond_signal(&comp_param[idx].cond); |
341 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
342 | } | |
343 | } | |
344 | ||
f0afa331 | 345 | static void compress_threads_save_cleanup(void) |
56e93d26 JQ |
346 | { |
347 | int i, thread_count; | |
348 | ||
349 | if (!migrate_use_compression()) { | |
350 | return; | |
351 | } | |
352 | terminate_compression_threads(); | |
353 | thread_count = migrate_compress_threads(); | |
354 | for (i = 0; i < thread_count; i++) { | |
355 | qemu_thread_join(compress_threads + i); | |
356 | qemu_fclose(comp_param[i].file); | |
357 | qemu_mutex_destroy(&comp_param[i].mutex); | |
358 | qemu_cond_destroy(&comp_param[i].cond); | |
359 | } | |
0d9f9a5c LL |
360 | qemu_mutex_destroy(&comp_done_lock); |
361 | qemu_cond_destroy(&comp_done_cond); | |
56e93d26 JQ |
362 | g_free(compress_threads); |
363 | g_free(comp_param); | |
56e93d26 JQ |
364 | compress_threads = NULL; |
365 | comp_param = NULL; | |
56e93d26 JQ |
366 | } |
367 | ||
f0afa331 | 368 | static void compress_threads_save_setup(void) |
56e93d26 JQ |
369 | { |
370 | int i, thread_count; | |
371 | ||
372 | if (!migrate_use_compression()) { | |
373 | return; | |
374 | } | |
56e93d26 JQ |
375 | thread_count = migrate_compress_threads(); |
376 | compress_threads = g_new0(QemuThread, thread_count); | |
377 | comp_param = g_new0(CompressParam, thread_count); | |
0d9f9a5c LL |
378 | qemu_cond_init(&comp_done_cond); |
379 | qemu_mutex_init(&comp_done_lock); | |
56e93d26 | 380 | for (i = 0; i < thread_count; i++) { |
e110aa91 C |
381 | /* comp_param[i].file is just used as a dummy buffer to save data, |
382 | * set its ops to empty. | |
56e93d26 JQ |
383 | */ |
384 | comp_param[i].file = qemu_fopen_ops(NULL, &empty_ops); | |
385 | comp_param[i].done = true; | |
90e56fb4 | 386 | comp_param[i].quit = false; |
56e93d26 JQ |
387 | qemu_mutex_init(&comp_param[i].mutex); |
388 | qemu_cond_init(&comp_param[i].cond); | |
389 | qemu_thread_create(compress_threads + i, "compress", | |
390 | do_data_compress, comp_param + i, | |
391 | QEMU_THREAD_JOINABLE); | |
392 | } | |
393 | } | |
394 | ||
f986c3d2 JQ |
395 | /* Multiple fd's */ |
396 | ||
397 | struct MultiFDSendParams { | |
398 | uint8_t id; | |
399 | char *name; | |
400 | QemuThread thread; | |
401 | QemuSemaphore sem; | |
402 | QemuMutex mutex; | |
403 | bool quit; | |
404 | }; | |
405 | typedef struct MultiFDSendParams MultiFDSendParams; | |
406 | ||
407 | struct { | |
408 | MultiFDSendParams *params; | |
409 | /* number of created threads */ | |
410 | int count; | |
411 | } *multifd_send_state; | |
412 | ||
413 | static void terminate_multifd_send_threads(Error *errp) | |
414 | { | |
415 | int i; | |
416 | ||
417 | for (i = 0; i < multifd_send_state->count; i++) { | |
418 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
419 | ||
420 | qemu_mutex_lock(&p->mutex); | |
421 | p->quit = true; | |
422 | qemu_sem_post(&p->sem); | |
423 | qemu_mutex_unlock(&p->mutex); | |
424 | } | |
425 | } | |
426 | ||
427 | int multifd_save_cleanup(Error **errp) | |
428 | { | |
429 | int i; | |
430 | int ret = 0; | |
431 | ||
432 | if (!migrate_use_multifd()) { | |
433 | return 0; | |
434 | } | |
435 | terminate_multifd_send_threads(NULL); | |
436 | for (i = 0; i < multifd_send_state->count; i++) { | |
437 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
438 | ||
439 | qemu_thread_join(&p->thread); | |
440 | qemu_mutex_destroy(&p->mutex); | |
441 | qemu_sem_destroy(&p->sem); | |
442 | g_free(p->name); | |
443 | p->name = NULL; | |
444 | } | |
445 | g_free(multifd_send_state->params); | |
446 | multifd_send_state->params = NULL; | |
447 | g_free(multifd_send_state); | |
448 | multifd_send_state = NULL; | |
449 | return ret; | |
450 | } | |
451 | ||
452 | static void *multifd_send_thread(void *opaque) | |
453 | { | |
454 | MultiFDSendParams *p = opaque; | |
455 | ||
456 | while (true) { | |
457 | qemu_mutex_lock(&p->mutex); | |
458 | if (p->quit) { | |
459 | qemu_mutex_unlock(&p->mutex); | |
460 | break; | |
461 | } | |
462 | qemu_mutex_unlock(&p->mutex); | |
463 | qemu_sem_wait(&p->sem); | |
464 | } | |
465 | ||
466 | return NULL; | |
467 | } | |
468 | ||
469 | int multifd_save_setup(void) | |
470 | { | |
471 | int thread_count; | |
472 | uint8_t i; | |
473 | ||
474 | if (!migrate_use_multifd()) { | |
475 | return 0; | |
476 | } | |
477 | thread_count = migrate_multifd_channels(); | |
478 | multifd_send_state = g_malloc0(sizeof(*multifd_send_state)); | |
479 | multifd_send_state->params = g_new0(MultiFDSendParams, thread_count); | |
480 | multifd_send_state->count = 0; | |
481 | for (i = 0; i < thread_count; i++) { | |
482 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
483 | ||
484 | qemu_mutex_init(&p->mutex); | |
485 | qemu_sem_init(&p->sem, 0); | |
486 | p->quit = false; | |
487 | p->id = i; | |
488 | p->name = g_strdup_printf("multifdsend_%d", i); | |
489 | qemu_thread_create(&p->thread, p->name, multifd_send_thread, p, | |
490 | QEMU_THREAD_JOINABLE); | |
491 | ||
492 | multifd_send_state->count++; | |
493 | } | |
494 | return 0; | |
495 | } | |
496 | ||
497 | struct MultiFDRecvParams { | |
498 | uint8_t id; | |
499 | char *name; | |
500 | QemuThread thread; | |
501 | QemuSemaphore sem; | |
502 | QemuMutex mutex; | |
503 | bool quit; | |
504 | }; | |
505 | typedef struct MultiFDRecvParams MultiFDRecvParams; | |
506 | ||
507 | struct { | |
508 | MultiFDRecvParams *params; | |
509 | /* number of created threads */ | |
510 | int count; | |
511 | } *multifd_recv_state; | |
512 | ||
513 | static void terminate_multifd_recv_threads(Error *errp) | |
514 | { | |
515 | int i; | |
516 | ||
517 | for (i = 0; i < multifd_recv_state->count; i++) { | |
518 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
519 | ||
520 | qemu_mutex_lock(&p->mutex); | |
521 | p->quit = true; | |
522 | qemu_sem_post(&p->sem); | |
523 | qemu_mutex_unlock(&p->mutex); | |
524 | } | |
525 | } | |
526 | ||
527 | int multifd_load_cleanup(Error **errp) | |
528 | { | |
529 | int i; | |
530 | int ret = 0; | |
531 | ||
532 | if (!migrate_use_multifd()) { | |
533 | return 0; | |
534 | } | |
535 | terminate_multifd_recv_threads(NULL); | |
536 | for (i = 0; i < multifd_recv_state->count; i++) { | |
537 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
538 | ||
539 | qemu_thread_join(&p->thread); | |
540 | qemu_mutex_destroy(&p->mutex); | |
541 | qemu_sem_destroy(&p->sem); | |
542 | g_free(p->name); | |
543 | p->name = NULL; | |
544 | } | |
545 | g_free(multifd_recv_state->params); | |
546 | multifd_recv_state->params = NULL; | |
547 | g_free(multifd_recv_state); | |
548 | multifd_recv_state = NULL; | |
549 | ||
550 | return ret; | |
551 | } | |
552 | ||
553 | static void *multifd_recv_thread(void *opaque) | |
554 | { | |
555 | MultiFDRecvParams *p = opaque; | |
556 | ||
557 | while (true) { | |
558 | qemu_mutex_lock(&p->mutex); | |
559 | if (p->quit) { | |
560 | qemu_mutex_unlock(&p->mutex); | |
561 | break; | |
562 | } | |
563 | qemu_mutex_unlock(&p->mutex); | |
564 | qemu_sem_wait(&p->sem); | |
565 | } | |
566 | ||
567 | return NULL; | |
568 | } | |
569 | ||
570 | int multifd_load_setup(void) | |
571 | { | |
572 | int thread_count; | |
573 | uint8_t i; | |
574 | ||
575 | if (!migrate_use_multifd()) { | |
576 | return 0; | |
577 | } | |
578 | thread_count = migrate_multifd_channels(); | |
579 | multifd_recv_state = g_malloc0(sizeof(*multifd_recv_state)); | |
580 | multifd_recv_state->params = g_new0(MultiFDRecvParams, thread_count); | |
581 | multifd_recv_state->count = 0; | |
582 | for (i = 0; i < thread_count; i++) { | |
583 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
584 | ||
585 | qemu_mutex_init(&p->mutex); | |
586 | qemu_sem_init(&p->sem, 0); | |
587 | p->quit = false; | |
588 | p->id = i; | |
589 | p->name = g_strdup_printf("multifdrecv_%d", i); | |
590 | qemu_thread_create(&p->thread, p->name, multifd_recv_thread, p, | |
591 | QEMU_THREAD_JOINABLE); | |
592 | multifd_recv_state->count++; | |
593 | } | |
594 | return 0; | |
595 | } | |
596 | ||
56e93d26 | 597 | /** |
3d0684b2 | 598 | * save_page_header: write page header to wire |
56e93d26 JQ |
599 | * |
600 | * If this is the 1st block, it also writes the block identification | |
601 | * | |
3d0684b2 | 602 | * Returns the number of bytes written |
56e93d26 JQ |
603 | * |
604 | * @f: QEMUFile where to send the data | |
605 | * @block: block that contains the page we want to send | |
606 | * @offset: offset inside the block for the page | |
607 | * in the lower bits, it contains flags | |
608 | */ | |
2bf3aa85 JQ |
609 | static size_t save_page_header(RAMState *rs, QEMUFile *f, RAMBlock *block, |
610 | ram_addr_t offset) | |
56e93d26 | 611 | { |
9f5f380b | 612 | size_t size, len; |
56e93d26 | 613 | |
24795694 JQ |
614 | if (block == rs->last_sent_block) { |
615 | offset |= RAM_SAVE_FLAG_CONTINUE; | |
616 | } | |
2bf3aa85 | 617 | qemu_put_be64(f, offset); |
56e93d26 JQ |
618 | size = 8; |
619 | ||
620 | if (!(offset & RAM_SAVE_FLAG_CONTINUE)) { | |
9f5f380b | 621 | len = strlen(block->idstr); |
2bf3aa85 JQ |
622 | qemu_put_byte(f, len); |
623 | qemu_put_buffer(f, (uint8_t *)block->idstr, len); | |
9f5f380b | 624 | size += 1 + len; |
24795694 | 625 | rs->last_sent_block = block; |
56e93d26 JQ |
626 | } |
627 | return size; | |
628 | } | |
629 | ||
3d0684b2 JQ |
630 | /** |
631 | * mig_throttle_guest_down: throotle down the guest | |
632 | * | |
633 | * Reduce amount of guest cpu execution to hopefully slow down memory | |
634 | * writes. If guest dirty memory rate is reduced below the rate at | |
635 | * which we can transfer pages to the destination then we should be | |
636 | * able to complete migration. Some workloads dirty memory way too | |
637 | * fast and will not effectively converge, even with auto-converge. | |
070afca2 JH |
638 | */ |
639 | static void mig_throttle_guest_down(void) | |
640 | { | |
641 | MigrationState *s = migrate_get_current(); | |
2594f56d DB |
642 | uint64_t pct_initial = s->parameters.cpu_throttle_initial; |
643 | uint64_t pct_icrement = s->parameters.cpu_throttle_increment; | |
070afca2 JH |
644 | |
645 | /* We have not started throttling yet. Let's start it. */ | |
646 | if (!cpu_throttle_active()) { | |
647 | cpu_throttle_set(pct_initial); | |
648 | } else { | |
649 | /* Throttling already on, just increase the rate */ | |
650 | cpu_throttle_set(cpu_throttle_get_percentage() + pct_icrement); | |
651 | } | |
652 | } | |
653 | ||
3d0684b2 JQ |
654 | /** |
655 | * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache | |
656 | * | |
6f37bb8b | 657 | * @rs: current RAM state |
3d0684b2 JQ |
658 | * @current_addr: address for the zero page |
659 | * | |
660 | * Update the xbzrle cache to reflect a page that's been sent as all 0. | |
56e93d26 JQ |
661 | * The important thing is that a stale (not-yet-0'd) page be replaced |
662 | * by the new data. | |
663 | * As a bonus, if the page wasn't in the cache it gets added so that | |
3d0684b2 | 664 | * when a small write is made into the 0'd page it gets XBZRLE sent. |
56e93d26 | 665 | */ |
6f37bb8b | 666 | static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr) |
56e93d26 | 667 | { |
6f37bb8b | 668 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { |
56e93d26 JQ |
669 | return; |
670 | } | |
671 | ||
672 | /* We don't care if this fails to allocate a new cache page | |
673 | * as long as it updated an old one */ | |
c00e0928 | 674 | cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page, |
9360447d | 675 | ram_counters.dirty_sync_count); |
56e93d26 JQ |
676 | } |
677 | ||
678 | #define ENCODING_FLAG_XBZRLE 0x1 | |
679 | ||
680 | /** | |
681 | * save_xbzrle_page: compress and send current page | |
682 | * | |
683 | * Returns: 1 means that we wrote the page | |
684 | * 0 means that page is identical to the one already sent | |
685 | * -1 means that xbzrle would be longer than normal | |
686 | * | |
5a987738 | 687 | * @rs: current RAM state |
3d0684b2 JQ |
688 | * @current_data: pointer to the address of the page contents |
689 | * @current_addr: addr of the page | |
56e93d26 JQ |
690 | * @block: block that contains the page we want to send |
691 | * @offset: offset inside the block for the page | |
692 | * @last_stage: if we are at the completion stage | |
56e93d26 | 693 | */ |
204b88b8 | 694 | static int save_xbzrle_page(RAMState *rs, uint8_t **current_data, |
56e93d26 | 695 | ram_addr_t current_addr, RAMBlock *block, |
072c2511 | 696 | ram_addr_t offset, bool last_stage) |
56e93d26 JQ |
697 | { |
698 | int encoded_len = 0, bytes_xbzrle; | |
699 | uint8_t *prev_cached_page; | |
700 | ||
9360447d JQ |
701 | if (!cache_is_cached(XBZRLE.cache, current_addr, |
702 | ram_counters.dirty_sync_count)) { | |
703 | xbzrle_counters.cache_miss++; | |
56e93d26 JQ |
704 | if (!last_stage) { |
705 | if (cache_insert(XBZRLE.cache, current_addr, *current_data, | |
9360447d | 706 | ram_counters.dirty_sync_count) == -1) { |
56e93d26 JQ |
707 | return -1; |
708 | } else { | |
709 | /* update *current_data when the page has been | |
710 | inserted into cache */ | |
711 | *current_data = get_cached_data(XBZRLE.cache, current_addr); | |
712 | } | |
713 | } | |
714 | return -1; | |
715 | } | |
716 | ||
717 | prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); | |
718 | ||
719 | /* save current buffer into memory */ | |
720 | memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); | |
721 | ||
722 | /* XBZRLE encoding (if there is no overflow) */ | |
723 | encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, | |
724 | TARGET_PAGE_SIZE, XBZRLE.encoded_buf, | |
725 | TARGET_PAGE_SIZE); | |
726 | if (encoded_len == 0) { | |
55c4446b | 727 | trace_save_xbzrle_page_skipping(); |
56e93d26 JQ |
728 | return 0; |
729 | } else if (encoded_len == -1) { | |
55c4446b | 730 | trace_save_xbzrle_page_overflow(); |
9360447d | 731 | xbzrle_counters.overflow++; |
56e93d26 JQ |
732 | /* update data in the cache */ |
733 | if (!last_stage) { | |
734 | memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE); | |
735 | *current_data = prev_cached_page; | |
736 | } | |
737 | return -1; | |
738 | } | |
739 | ||
740 | /* we need to update the data in the cache, in order to get the same data */ | |
741 | if (!last_stage) { | |
742 | memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); | |
743 | } | |
744 | ||
745 | /* Send XBZRLE based compressed page */ | |
2bf3aa85 | 746 | bytes_xbzrle = save_page_header(rs, rs->f, block, |
204b88b8 JQ |
747 | offset | RAM_SAVE_FLAG_XBZRLE); |
748 | qemu_put_byte(rs->f, ENCODING_FLAG_XBZRLE); | |
749 | qemu_put_be16(rs->f, encoded_len); | |
750 | qemu_put_buffer(rs->f, XBZRLE.encoded_buf, encoded_len); | |
56e93d26 | 751 | bytes_xbzrle += encoded_len + 1 + 2; |
9360447d JQ |
752 | xbzrle_counters.pages++; |
753 | xbzrle_counters.bytes += bytes_xbzrle; | |
754 | ram_counters.transferred += bytes_xbzrle; | |
56e93d26 JQ |
755 | |
756 | return 1; | |
757 | } | |
758 | ||
3d0684b2 JQ |
759 | /** |
760 | * migration_bitmap_find_dirty: find the next dirty page from start | |
f3f491fc | 761 | * |
3d0684b2 JQ |
762 | * Called with rcu_read_lock() to protect migration_bitmap |
763 | * | |
764 | * Returns the byte offset within memory region of the start of a dirty page | |
765 | * | |
6f37bb8b | 766 | * @rs: current RAM state |
3d0684b2 | 767 | * @rb: RAMBlock where to search for dirty pages |
a935e30f | 768 | * @start: page where we start the search |
f3f491fc | 769 | */ |
56e93d26 | 770 | static inline |
a935e30f | 771 | unsigned long migration_bitmap_find_dirty(RAMState *rs, RAMBlock *rb, |
f20e2865 | 772 | unsigned long start) |
56e93d26 | 773 | { |
6b6712ef JQ |
774 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; |
775 | unsigned long *bitmap = rb->bmap; | |
56e93d26 JQ |
776 | unsigned long next; |
777 | ||
6b6712ef JQ |
778 | if (rs->ram_bulk_stage && start > 0) { |
779 | next = start + 1; | |
56e93d26 | 780 | } else { |
6b6712ef | 781 | next = find_next_bit(bitmap, size, start); |
56e93d26 JQ |
782 | } |
783 | ||
6b6712ef | 784 | return next; |
56e93d26 JQ |
785 | } |
786 | ||
06b10688 | 787 | static inline bool migration_bitmap_clear_dirty(RAMState *rs, |
f20e2865 JQ |
788 | RAMBlock *rb, |
789 | unsigned long page) | |
a82d593b DDAG |
790 | { |
791 | bool ret; | |
a82d593b | 792 | |
6b6712ef | 793 | ret = test_and_clear_bit(page, rb->bmap); |
a82d593b DDAG |
794 | |
795 | if (ret) { | |
0d8ec885 | 796 | rs->migration_dirty_pages--; |
a82d593b DDAG |
797 | } |
798 | return ret; | |
799 | } | |
800 | ||
15440dd5 JQ |
801 | static void migration_bitmap_sync_range(RAMState *rs, RAMBlock *rb, |
802 | ram_addr_t start, ram_addr_t length) | |
56e93d26 | 803 | { |
0d8ec885 | 804 | rs->migration_dirty_pages += |
6b6712ef | 805 | cpu_physical_memory_sync_dirty_bitmap(rb, start, length, |
0d8ec885 | 806 | &rs->num_dirty_pages_period); |
56e93d26 JQ |
807 | } |
808 | ||
3d0684b2 JQ |
809 | /** |
810 | * ram_pagesize_summary: calculate all the pagesizes of a VM | |
811 | * | |
812 | * Returns a summary bitmap of the page sizes of all RAMBlocks | |
813 | * | |
814 | * For VMs with just normal pages this is equivalent to the host page | |
815 | * size. If it's got some huge pages then it's the OR of all the | |
816 | * different page sizes. | |
e8ca1db2 DDAG |
817 | */ |
818 | uint64_t ram_pagesize_summary(void) | |
819 | { | |
820 | RAMBlock *block; | |
821 | uint64_t summary = 0; | |
822 | ||
99e15582 | 823 | RAMBLOCK_FOREACH(block) { |
e8ca1db2 DDAG |
824 | summary |= block->page_size; |
825 | } | |
826 | ||
827 | return summary; | |
828 | } | |
829 | ||
8d820d6f | 830 | static void migration_bitmap_sync(RAMState *rs) |
56e93d26 JQ |
831 | { |
832 | RAMBlock *block; | |
56e93d26 | 833 | int64_t end_time; |
c4bdf0cf | 834 | uint64_t bytes_xfer_now; |
56e93d26 | 835 | |
9360447d | 836 | ram_counters.dirty_sync_count++; |
56e93d26 | 837 | |
f664da80 JQ |
838 | if (!rs->time_last_bitmap_sync) { |
839 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56e93d26 JQ |
840 | } |
841 | ||
842 | trace_migration_bitmap_sync_start(); | |
9c1f8f44 | 843 | memory_global_dirty_log_sync(); |
56e93d26 | 844 | |
108cfae0 | 845 | qemu_mutex_lock(&rs->bitmap_mutex); |
56e93d26 | 846 | rcu_read_lock(); |
99e15582 | 847 | RAMBLOCK_FOREACH(block) { |
15440dd5 | 848 | migration_bitmap_sync_range(rs, block, 0, block->used_length); |
56e93d26 JQ |
849 | } |
850 | rcu_read_unlock(); | |
108cfae0 | 851 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 852 | |
a66cd90c | 853 | trace_migration_bitmap_sync_end(rs->num_dirty_pages_period); |
1ffb5dfd | 854 | |
56e93d26 JQ |
855 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
856 | ||
857 | /* more than 1 second = 1000 millisecons */ | |
f664da80 | 858 | if (end_time > rs->time_last_bitmap_sync + 1000) { |
d693c6f1 | 859 | /* calculate period counters */ |
9360447d | 860 | ram_counters.dirty_pages_rate = rs->num_dirty_pages_period * 1000 |
d693c6f1 | 861 | / (end_time - rs->time_last_bitmap_sync); |
9360447d | 862 | bytes_xfer_now = ram_counters.transferred; |
d693c6f1 | 863 | |
9ac78b61 PL |
864 | /* During block migration the auto-converge logic incorrectly detects |
865 | * that ram migration makes no progress. Avoid this by disabling the | |
866 | * throttling logic during the bulk phase of block migration. */ | |
867 | if (migrate_auto_converge() && !blk_mig_bulk_active()) { | |
56e93d26 JQ |
868 | /* The following detection logic can be refined later. For now: |
869 | Check to see if the dirtied bytes is 50% more than the approx. | |
870 | amount of bytes that just got transferred since the last time we | |
070afca2 JH |
871 | were in this routine. If that happens twice, start or increase |
872 | throttling */ | |
070afca2 | 873 | |
d693c6f1 | 874 | if ((rs->num_dirty_pages_period * TARGET_PAGE_SIZE > |
eac74159 | 875 | (bytes_xfer_now - rs->bytes_xfer_prev) / 2) && |
b4a3c64b | 876 | (++rs->dirty_rate_high_cnt >= 2)) { |
56e93d26 | 877 | trace_migration_throttle(); |
8d820d6f | 878 | rs->dirty_rate_high_cnt = 0; |
070afca2 | 879 | mig_throttle_guest_down(); |
d693c6f1 | 880 | } |
56e93d26 | 881 | } |
070afca2 | 882 | |
56e93d26 | 883 | if (migrate_use_xbzrle()) { |
23b28c3c | 884 | if (rs->iterations_prev != rs->iterations) { |
9360447d JQ |
885 | xbzrle_counters.cache_miss_rate = |
886 | (double)(xbzrle_counters.cache_miss - | |
b5833fde | 887 | rs->xbzrle_cache_miss_prev) / |
23b28c3c | 888 | (rs->iterations - rs->iterations_prev); |
56e93d26 | 889 | } |
23b28c3c | 890 | rs->iterations_prev = rs->iterations; |
9360447d | 891 | rs->xbzrle_cache_miss_prev = xbzrle_counters.cache_miss; |
56e93d26 | 892 | } |
d693c6f1 FF |
893 | |
894 | /* reset period counters */ | |
f664da80 | 895 | rs->time_last_bitmap_sync = end_time; |
a66cd90c | 896 | rs->num_dirty_pages_period = 0; |
d2a4d85a | 897 | rs->bytes_xfer_prev = bytes_xfer_now; |
56e93d26 | 898 | } |
4addcd4f | 899 | if (migrate_use_events()) { |
9360447d | 900 | qapi_event_send_migration_pass(ram_counters.dirty_sync_count, NULL); |
4addcd4f | 901 | } |
56e93d26 JQ |
902 | } |
903 | ||
904 | /** | |
3d0684b2 | 905 | * save_zero_page: send the zero page to the stream |
56e93d26 | 906 | * |
3d0684b2 | 907 | * Returns the number of pages written. |
56e93d26 | 908 | * |
f7ccd61b | 909 | * @rs: current RAM state |
56e93d26 JQ |
910 | * @block: block that contains the page we want to send |
911 | * @offset: offset inside the block for the page | |
56e93d26 | 912 | */ |
7faccdc3 | 913 | static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset) |
56e93d26 | 914 | { |
7faccdc3 | 915 | uint8_t *p = block->host + offset; |
56e93d26 JQ |
916 | int pages = -1; |
917 | ||
918 | if (is_zero_range(p, TARGET_PAGE_SIZE)) { | |
9360447d JQ |
919 | ram_counters.duplicate++; |
920 | ram_counters.transferred += | |
bb890ed5 | 921 | save_page_header(rs, rs->f, block, offset | RAM_SAVE_FLAG_ZERO); |
ce25d337 | 922 | qemu_put_byte(rs->f, 0); |
9360447d | 923 | ram_counters.transferred += 1; |
56e93d26 JQ |
924 | pages = 1; |
925 | } | |
926 | ||
927 | return pages; | |
928 | } | |
929 | ||
5727309d | 930 | static void ram_release_pages(const char *rbname, uint64_t offset, int pages) |
53f09a10 | 931 | { |
5727309d | 932 | if (!migrate_release_ram() || !migration_in_postcopy()) { |
53f09a10 PB |
933 | return; |
934 | } | |
935 | ||
aaa2064c | 936 | ram_discard_range(rbname, offset, pages << TARGET_PAGE_BITS); |
53f09a10 PB |
937 | } |
938 | ||
56e93d26 | 939 | /** |
3d0684b2 | 940 | * ram_save_page: send the given page to the stream |
56e93d26 | 941 | * |
3d0684b2 | 942 | * Returns the number of pages written. |
3fd3c4b3 DDAG |
943 | * < 0 - error |
944 | * >=0 - Number of pages written - this might legally be 0 | |
945 | * if xbzrle noticed the page was the same. | |
56e93d26 | 946 | * |
6f37bb8b | 947 | * @rs: current RAM state |
56e93d26 JQ |
948 | * @block: block that contains the page we want to send |
949 | * @offset: offset inside the block for the page | |
950 | * @last_stage: if we are at the completion stage | |
56e93d26 | 951 | */ |
a0a8aa14 | 952 | static int ram_save_page(RAMState *rs, PageSearchStatus *pss, bool last_stage) |
56e93d26 JQ |
953 | { |
954 | int pages = -1; | |
955 | uint64_t bytes_xmit; | |
956 | ram_addr_t current_addr; | |
56e93d26 JQ |
957 | uint8_t *p; |
958 | int ret; | |
959 | bool send_async = true; | |
a08f6890 | 960 | RAMBlock *block = pss->block; |
a935e30f | 961 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
56e93d26 | 962 | |
2f68e399 | 963 | p = block->host + offset; |
1db9d8e5 | 964 | trace_ram_save_page(block->idstr, (uint64_t)offset, p); |
56e93d26 JQ |
965 | |
966 | /* In doubt sent page as normal */ | |
967 | bytes_xmit = 0; | |
ce25d337 | 968 | ret = ram_control_save_page(rs->f, block->offset, |
56e93d26 JQ |
969 | offset, TARGET_PAGE_SIZE, &bytes_xmit); |
970 | if (bytes_xmit) { | |
9360447d | 971 | ram_counters.transferred += bytes_xmit; |
56e93d26 JQ |
972 | pages = 1; |
973 | } | |
974 | ||
975 | XBZRLE_cache_lock(); | |
976 | ||
977 | current_addr = block->offset + offset; | |
978 | ||
56e93d26 JQ |
979 | if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { |
980 | if (ret != RAM_SAVE_CONTROL_DELAYED) { | |
981 | if (bytes_xmit > 0) { | |
9360447d | 982 | ram_counters.normal++; |
56e93d26 | 983 | } else if (bytes_xmit == 0) { |
9360447d | 984 | ram_counters.duplicate++; |
56e93d26 JQ |
985 | } |
986 | } | |
987 | } else { | |
7faccdc3 | 988 | pages = save_zero_page(rs, block, offset); |
56e93d26 JQ |
989 | if (pages > 0) { |
990 | /* Must let xbzrle know, otherwise a previous (now 0'd) cached | |
991 | * page would be stale | |
992 | */ | |
6f37bb8b | 993 | xbzrle_cache_zero_page(rs, current_addr); |
a935e30f | 994 | ram_release_pages(block->idstr, offset, pages); |
6f37bb8b | 995 | } else if (!rs->ram_bulk_stage && |
5727309d | 996 | !migration_in_postcopy() && migrate_use_xbzrle()) { |
204b88b8 | 997 | pages = save_xbzrle_page(rs, &p, current_addr, block, |
072c2511 | 998 | offset, last_stage); |
56e93d26 JQ |
999 | if (!last_stage) { |
1000 | /* Can't send this cached data async, since the cache page | |
1001 | * might get updated before it gets to the wire | |
1002 | */ | |
1003 | send_async = false; | |
1004 | } | |
1005 | } | |
1006 | } | |
1007 | ||
1008 | /* XBZRLE overflow or normal page */ | |
1009 | if (pages == -1) { | |
9360447d JQ |
1010 | ram_counters.transferred += |
1011 | save_page_header(rs, rs->f, block, offset | RAM_SAVE_FLAG_PAGE); | |
56e93d26 | 1012 | if (send_async) { |
ce25d337 | 1013 | qemu_put_buffer_async(rs->f, p, TARGET_PAGE_SIZE, |
53f09a10 | 1014 | migrate_release_ram() & |
5727309d | 1015 | migration_in_postcopy()); |
56e93d26 | 1016 | } else { |
ce25d337 | 1017 | qemu_put_buffer(rs->f, p, TARGET_PAGE_SIZE); |
56e93d26 | 1018 | } |
9360447d | 1019 | ram_counters.transferred += TARGET_PAGE_SIZE; |
56e93d26 | 1020 | pages = 1; |
9360447d | 1021 | ram_counters.normal++; |
56e93d26 JQ |
1022 | } |
1023 | ||
1024 | XBZRLE_cache_unlock(); | |
1025 | ||
1026 | return pages; | |
1027 | } | |
1028 | ||
a7a9a88f LL |
1029 | static int do_compress_ram_page(QEMUFile *f, RAMBlock *block, |
1030 | ram_addr_t offset) | |
56e93d26 | 1031 | { |
53518d94 | 1032 | RAMState *rs = ram_state; |
56e93d26 | 1033 | int bytes_sent, blen; |
a7a9a88f | 1034 | uint8_t *p = block->host + (offset & TARGET_PAGE_MASK); |
56e93d26 | 1035 | |
2bf3aa85 | 1036 | bytes_sent = save_page_header(rs, f, block, offset | |
56e93d26 | 1037 | RAM_SAVE_FLAG_COMPRESS_PAGE); |
a7a9a88f | 1038 | blen = qemu_put_compression_data(f, p, TARGET_PAGE_SIZE, |
56e93d26 | 1039 | migrate_compress_level()); |
b3be2896 LL |
1040 | if (blen < 0) { |
1041 | bytes_sent = 0; | |
1042 | qemu_file_set_error(migrate_get_current()->to_dst_file, blen); | |
1043 | error_report("compressed data failed!"); | |
1044 | } else { | |
1045 | bytes_sent += blen; | |
5727309d | 1046 | ram_release_pages(block->idstr, offset & TARGET_PAGE_MASK, 1); |
b3be2896 | 1047 | } |
56e93d26 JQ |
1048 | |
1049 | return bytes_sent; | |
1050 | } | |
1051 | ||
ce25d337 | 1052 | static void flush_compressed_data(RAMState *rs) |
56e93d26 JQ |
1053 | { |
1054 | int idx, len, thread_count; | |
1055 | ||
1056 | if (!migrate_use_compression()) { | |
1057 | return; | |
1058 | } | |
1059 | thread_count = migrate_compress_threads(); | |
a7a9a88f | 1060 | |
0d9f9a5c | 1061 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 | 1062 | for (idx = 0; idx < thread_count; idx++) { |
a7a9a88f | 1063 | while (!comp_param[idx].done) { |
0d9f9a5c | 1064 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 | 1065 | } |
a7a9a88f | 1066 | } |
0d9f9a5c | 1067 | qemu_mutex_unlock(&comp_done_lock); |
a7a9a88f LL |
1068 | |
1069 | for (idx = 0; idx < thread_count; idx++) { | |
1070 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 1071 | if (!comp_param[idx].quit) { |
ce25d337 | 1072 | len = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
9360447d | 1073 | ram_counters.transferred += len; |
56e93d26 | 1074 | } |
a7a9a88f | 1075 | qemu_mutex_unlock(&comp_param[idx].mutex); |
56e93d26 JQ |
1076 | } |
1077 | } | |
1078 | ||
1079 | static inline void set_compress_params(CompressParam *param, RAMBlock *block, | |
1080 | ram_addr_t offset) | |
1081 | { | |
1082 | param->block = block; | |
1083 | param->offset = offset; | |
1084 | } | |
1085 | ||
ce25d337 JQ |
1086 | static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block, |
1087 | ram_addr_t offset) | |
56e93d26 JQ |
1088 | { |
1089 | int idx, thread_count, bytes_xmit = -1, pages = -1; | |
1090 | ||
1091 | thread_count = migrate_compress_threads(); | |
0d9f9a5c | 1092 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 JQ |
1093 | while (true) { |
1094 | for (idx = 0; idx < thread_count; idx++) { | |
1095 | if (comp_param[idx].done) { | |
a7a9a88f | 1096 | comp_param[idx].done = false; |
ce25d337 | 1097 | bytes_xmit = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
a7a9a88f | 1098 | qemu_mutex_lock(&comp_param[idx].mutex); |
56e93d26 | 1099 | set_compress_params(&comp_param[idx], block, offset); |
a7a9a88f LL |
1100 | qemu_cond_signal(&comp_param[idx].cond); |
1101 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
56e93d26 | 1102 | pages = 1; |
9360447d JQ |
1103 | ram_counters.normal++; |
1104 | ram_counters.transferred += bytes_xmit; | |
56e93d26 JQ |
1105 | break; |
1106 | } | |
1107 | } | |
1108 | if (pages > 0) { | |
1109 | break; | |
1110 | } else { | |
0d9f9a5c | 1111 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 JQ |
1112 | } |
1113 | } | |
0d9f9a5c | 1114 | qemu_mutex_unlock(&comp_done_lock); |
56e93d26 JQ |
1115 | |
1116 | return pages; | |
1117 | } | |
1118 | ||
1119 | /** | |
1120 | * ram_save_compressed_page: compress the given page and send it to the stream | |
1121 | * | |
3d0684b2 | 1122 | * Returns the number of pages written. |
56e93d26 | 1123 | * |
6f37bb8b | 1124 | * @rs: current RAM state |
56e93d26 JQ |
1125 | * @block: block that contains the page we want to send |
1126 | * @offset: offset inside the block for the page | |
1127 | * @last_stage: if we are at the completion stage | |
56e93d26 | 1128 | */ |
a0a8aa14 JQ |
1129 | static int ram_save_compressed_page(RAMState *rs, PageSearchStatus *pss, |
1130 | bool last_stage) | |
56e93d26 JQ |
1131 | { |
1132 | int pages = -1; | |
fc50438e | 1133 | uint64_t bytes_xmit = 0; |
56e93d26 | 1134 | uint8_t *p; |
fc50438e | 1135 | int ret, blen; |
a08f6890 | 1136 | RAMBlock *block = pss->block; |
a935e30f | 1137 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
56e93d26 | 1138 | |
2f68e399 | 1139 | p = block->host + offset; |
56e93d26 | 1140 | |
ce25d337 | 1141 | ret = ram_control_save_page(rs->f, block->offset, |
56e93d26 JQ |
1142 | offset, TARGET_PAGE_SIZE, &bytes_xmit); |
1143 | if (bytes_xmit) { | |
9360447d | 1144 | ram_counters.transferred += bytes_xmit; |
56e93d26 JQ |
1145 | pages = 1; |
1146 | } | |
56e93d26 JQ |
1147 | if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { |
1148 | if (ret != RAM_SAVE_CONTROL_DELAYED) { | |
1149 | if (bytes_xmit > 0) { | |
9360447d | 1150 | ram_counters.normal++; |
56e93d26 | 1151 | } else if (bytes_xmit == 0) { |
9360447d | 1152 | ram_counters.duplicate++; |
56e93d26 JQ |
1153 | } |
1154 | } | |
1155 | } else { | |
1156 | /* When starting the process of a new block, the first page of | |
1157 | * the block should be sent out before other pages in the same | |
1158 | * block, and all the pages in last block should have been sent | |
1159 | * out, keeping this order is important, because the 'cont' flag | |
1160 | * is used to avoid resending the block name. | |
1161 | */ | |
6f37bb8b | 1162 | if (block != rs->last_sent_block) { |
ce25d337 | 1163 | flush_compressed_data(rs); |
7faccdc3 | 1164 | pages = save_zero_page(rs, block, offset); |
56e93d26 | 1165 | if (pages == -1) { |
fc50438e | 1166 | /* Make sure the first page is sent out before other pages */ |
2bf3aa85 | 1167 | bytes_xmit = save_page_header(rs, rs->f, block, offset | |
fc50438e | 1168 | RAM_SAVE_FLAG_COMPRESS_PAGE); |
ce25d337 | 1169 | blen = qemu_put_compression_data(rs->f, p, TARGET_PAGE_SIZE, |
fc50438e LL |
1170 | migrate_compress_level()); |
1171 | if (blen > 0) { | |
9360447d JQ |
1172 | ram_counters.transferred += bytes_xmit + blen; |
1173 | ram_counters.normal++; | |
b3be2896 | 1174 | pages = 1; |
fc50438e | 1175 | } else { |
ce25d337 | 1176 | qemu_file_set_error(rs->f, blen); |
fc50438e | 1177 | error_report("compressed data failed!"); |
b3be2896 | 1178 | } |
56e93d26 | 1179 | } |
53f09a10 | 1180 | if (pages > 0) { |
a935e30f | 1181 | ram_release_pages(block->idstr, offset, pages); |
53f09a10 | 1182 | } |
56e93d26 | 1183 | } else { |
7faccdc3 | 1184 | pages = save_zero_page(rs, block, offset); |
56e93d26 | 1185 | if (pages == -1) { |
ce25d337 | 1186 | pages = compress_page_with_multi_thread(rs, block, offset); |
53f09a10 | 1187 | } else { |
a935e30f | 1188 | ram_release_pages(block->idstr, offset, pages); |
56e93d26 JQ |
1189 | } |
1190 | } | |
1191 | } | |
1192 | ||
1193 | return pages; | |
1194 | } | |
1195 | ||
3d0684b2 JQ |
1196 | /** |
1197 | * find_dirty_block: find the next dirty page and update any state | |
1198 | * associated with the search process. | |
b9e60928 | 1199 | * |
3d0684b2 | 1200 | * Returns if a page is found |
b9e60928 | 1201 | * |
6f37bb8b | 1202 | * @rs: current RAM state |
3d0684b2 JQ |
1203 | * @pss: data about the state of the current dirty page scan |
1204 | * @again: set to false if the search has scanned the whole of RAM | |
b9e60928 | 1205 | */ |
f20e2865 | 1206 | static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again) |
b9e60928 | 1207 | { |
f20e2865 | 1208 | pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page); |
6f37bb8b | 1209 | if (pss->complete_round && pss->block == rs->last_seen_block && |
a935e30f | 1210 | pss->page >= rs->last_page) { |
b9e60928 DDAG |
1211 | /* |
1212 | * We've been once around the RAM and haven't found anything. | |
1213 | * Give up. | |
1214 | */ | |
1215 | *again = false; | |
1216 | return false; | |
1217 | } | |
a935e30f | 1218 | if ((pss->page << TARGET_PAGE_BITS) >= pss->block->used_length) { |
b9e60928 | 1219 | /* Didn't find anything in this RAM Block */ |
a935e30f | 1220 | pss->page = 0; |
b9e60928 DDAG |
1221 | pss->block = QLIST_NEXT_RCU(pss->block, next); |
1222 | if (!pss->block) { | |
1223 | /* Hit the end of the list */ | |
1224 | pss->block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1225 | /* Flag that we've looped */ | |
1226 | pss->complete_round = true; | |
6f37bb8b | 1227 | rs->ram_bulk_stage = false; |
b9e60928 DDAG |
1228 | if (migrate_use_xbzrle()) { |
1229 | /* If xbzrle is on, stop using the data compression at this | |
1230 | * point. In theory, xbzrle can do better than compression. | |
1231 | */ | |
ce25d337 | 1232 | flush_compressed_data(rs); |
b9e60928 DDAG |
1233 | } |
1234 | } | |
1235 | /* Didn't find anything this time, but try again on the new block */ | |
1236 | *again = true; | |
1237 | return false; | |
1238 | } else { | |
1239 | /* Can go around again, but... */ | |
1240 | *again = true; | |
1241 | /* We've found something so probably don't need to */ | |
1242 | return true; | |
1243 | } | |
1244 | } | |
1245 | ||
3d0684b2 JQ |
1246 | /** |
1247 | * unqueue_page: gets a page of the queue | |
1248 | * | |
a82d593b | 1249 | * Helper for 'get_queued_page' - gets a page off the queue |
a82d593b | 1250 | * |
3d0684b2 JQ |
1251 | * Returns the block of the page (or NULL if none available) |
1252 | * | |
ec481c6c | 1253 | * @rs: current RAM state |
3d0684b2 | 1254 | * @offset: used to return the offset within the RAMBlock |
a82d593b | 1255 | */ |
f20e2865 | 1256 | static RAMBlock *unqueue_page(RAMState *rs, ram_addr_t *offset) |
a82d593b DDAG |
1257 | { |
1258 | RAMBlock *block = NULL; | |
1259 | ||
ec481c6c JQ |
1260 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1261 | if (!QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
1262 | struct RAMSrcPageRequest *entry = | |
1263 | QSIMPLEQ_FIRST(&rs->src_page_requests); | |
a82d593b DDAG |
1264 | block = entry->rb; |
1265 | *offset = entry->offset; | |
a82d593b DDAG |
1266 | |
1267 | if (entry->len > TARGET_PAGE_SIZE) { | |
1268 | entry->len -= TARGET_PAGE_SIZE; | |
1269 | entry->offset += TARGET_PAGE_SIZE; | |
1270 | } else { | |
1271 | memory_region_unref(block->mr); | |
ec481c6c | 1272 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
a82d593b DDAG |
1273 | g_free(entry); |
1274 | } | |
1275 | } | |
ec481c6c | 1276 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
a82d593b DDAG |
1277 | |
1278 | return block; | |
1279 | } | |
1280 | ||
3d0684b2 JQ |
1281 | /** |
1282 | * get_queued_page: unqueue a page from the postocpy requests | |
1283 | * | |
1284 | * Skips pages that are already sent (!dirty) | |
a82d593b | 1285 | * |
3d0684b2 | 1286 | * Returns if a queued page is found |
a82d593b | 1287 | * |
6f37bb8b | 1288 | * @rs: current RAM state |
3d0684b2 | 1289 | * @pss: data about the state of the current dirty page scan |
a82d593b | 1290 | */ |
f20e2865 | 1291 | static bool get_queued_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b DDAG |
1292 | { |
1293 | RAMBlock *block; | |
1294 | ram_addr_t offset; | |
1295 | bool dirty; | |
1296 | ||
1297 | do { | |
f20e2865 | 1298 | block = unqueue_page(rs, &offset); |
a82d593b DDAG |
1299 | /* |
1300 | * We're sending this page, and since it's postcopy nothing else | |
1301 | * will dirty it, and we must make sure it doesn't get sent again | |
1302 | * even if this queue request was received after the background | |
1303 | * search already sent it. | |
1304 | */ | |
1305 | if (block) { | |
f20e2865 JQ |
1306 | unsigned long page; |
1307 | ||
6b6712ef JQ |
1308 | page = offset >> TARGET_PAGE_BITS; |
1309 | dirty = test_bit(page, block->bmap); | |
a82d593b | 1310 | if (!dirty) { |
06b10688 | 1311 | trace_get_queued_page_not_dirty(block->idstr, (uint64_t)offset, |
6b6712ef | 1312 | page, test_bit(page, block->unsentmap)); |
a82d593b | 1313 | } else { |
f20e2865 | 1314 | trace_get_queued_page(block->idstr, (uint64_t)offset, page); |
a82d593b DDAG |
1315 | } |
1316 | } | |
1317 | ||
1318 | } while (block && !dirty); | |
1319 | ||
1320 | if (block) { | |
1321 | /* | |
1322 | * As soon as we start servicing pages out of order, then we have | |
1323 | * to kill the bulk stage, since the bulk stage assumes | |
1324 | * in (migration_bitmap_find_and_reset_dirty) that every page is | |
1325 | * dirty, that's no longer true. | |
1326 | */ | |
6f37bb8b | 1327 | rs->ram_bulk_stage = false; |
a82d593b DDAG |
1328 | |
1329 | /* | |
1330 | * We want the background search to continue from the queued page | |
1331 | * since the guest is likely to want other pages near to the page | |
1332 | * it just requested. | |
1333 | */ | |
1334 | pss->block = block; | |
a935e30f | 1335 | pss->page = offset >> TARGET_PAGE_BITS; |
a82d593b DDAG |
1336 | } |
1337 | ||
1338 | return !!block; | |
1339 | } | |
1340 | ||
6c595cde | 1341 | /** |
5e58f968 JQ |
1342 | * migration_page_queue_free: drop any remaining pages in the ram |
1343 | * request queue | |
6c595cde | 1344 | * |
3d0684b2 JQ |
1345 | * It should be empty at the end anyway, but in error cases there may |
1346 | * be some left. in case that there is any page left, we drop it. | |
1347 | * | |
6c595cde | 1348 | */ |
83c13382 | 1349 | static void migration_page_queue_free(RAMState *rs) |
6c595cde | 1350 | { |
ec481c6c | 1351 | struct RAMSrcPageRequest *mspr, *next_mspr; |
6c595cde DDAG |
1352 | /* This queue generally should be empty - but in the case of a failed |
1353 | * migration might have some droppings in. | |
1354 | */ | |
1355 | rcu_read_lock(); | |
ec481c6c | 1356 | QSIMPLEQ_FOREACH_SAFE(mspr, &rs->src_page_requests, next_req, next_mspr) { |
6c595cde | 1357 | memory_region_unref(mspr->rb->mr); |
ec481c6c | 1358 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
6c595cde DDAG |
1359 | g_free(mspr); |
1360 | } | |
1361 | rcu_read_unlock(); | |
1362 | } | |
1363 | ||
1364 | /** | |
3d0684b2 JQ |
1365 | * ram_save_queue_pages: queue the page for transmission |
1366 | * | |
1367 | * A request from postcopy destination for example. | |
1368 | * | |
1369 | * Returns zero on success or negative on error | |
1370 | * | |
3d0684b2 JQ |
1371 | * @rbname: Name of the RAMBLock of the request. NULL means the |
1372 | * same that last one. | |
1373 | * @start: starting address from the start of the RAMBlock | |
1374 | * @len: length (in bytes) to send | |
6c595cde | 1375 | */ |
96506894 | 1376 | int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len) |
6c595cde DDAG |
1377 | { |
1378 | RAMBlock *ramblock; | |
53518d94 | 1379 | RAMState *rs = ram_state; |
6c595cde | 1380 | |
9360447d | 1381 | ram_counters.postcopy_requests++; |
6c595cde DDAG |
1382 | rcu_read_lock(); |
1383 | if (!rbname) { | |
1384 | /* Reuse last RAMBlock */ | |
68a098f3 | 1385 | ramblock = rs->last_req_rb; |
6c595cde DDAG |
1386 | |
1387 | if (!ramblock) { | |
1388 | /* | |
1389 | * Shouldn't happen, we can't reuse the last RAMBlock if | |
1390 | * it's the 1st request. | |
1391 | */ | |
1392 | error_report("ram_save_queue_pages no previous block"); | |
1393 | goto err; | |
1394 | } | |
1395 | } else { | |
1396 | ramblock = qemu_ram_block_by_name(rbname); | |
1397 | ||
1398 | if (!ramblock) { | |
1399 | /* We shouldn't be asked for a non-existent RAMBlock */ | |
1400 | error_report("ram_save_queue_pages no block '%s'", rbname); | |
1401 | goto err; | |
1402 | } | |
68a098f3 | 1403 | rs->last_req_rb = ramblock; |
6c595cde DDAG |
1404 | } |
1405 | trace_ram_save_queue_pages(ramblock->idstr, start, len); | |
1406 | if (start+len > ramblock->used_length) { | |
9458ad6b JQ |
1407 | error_report("%s request overrun start=" RAM_ADDR_FMT " len=" |
1408 | RAM_ADDR_FMT " blocklen=" RAM_ADDR_FMT, | |
6c595cde DDAG |
1409 | __func__, start, len, ramblock->used_length); |
1410 | goto err; | |
1411 | } | |
1412 | ||
ec481c6c JQ |
1413 | struct RAMSrcPageRequest *new_entry = |
1414 | g_malloc0(sizeof(struct RAMSrcPageRequest)); | |
6c595cde DDAG |
1415 | new_entry->rb = ramblock; |
1416 | new_entry->offset = start; | |
1417 | new_entry->len = len; | |
1418 | ||
1419 | memory_region_ref(ramblock->mr); | |
ec481c6c JQ |
1420 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1421 | QSIMPLEQ_INSERT_TAIL(&rs->src_page_requests, new_entry, next_req); | |
1422 | qemu_mutex_unlock(&rs->src_page_req_mutex); | |
6c595cde DDAG |
1423 | rcu_read_unlock(); |
1424 | ||
1425 | return 0; | |
1426 | ||
1427 | err: | |
1428 | rcu_read_unlock(); | |
1429 | return -1; | |
1430 | } | |
1431 | ||
a82d593b | 1432 | /** |
3d0684b2 | 1433 | * ram_save_target_page: save one target page |
a82d593b | 1434 | * |
3d0684b2 | 1435 | * Returns the number of pages written |
a82d593b | 1436 | * |
6f37bb8b | 1437 | * @rs: current RAM state |
3d0684b2 | 1438 | * @ms: current migration state |
3d0684b2 | 1439 | * @pss: data about the page we want to send |
a82d593b | 1440 | * @last_stage: if we are at the completion stage |
a82d593b | 1441 | */ |
a0a8aa14 | 1442 | static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 1443 | bool last_stage) |
a82d593b DDAG |
1444 | { |
1445 | int res = 0; | |
1446 | ||
1447 | /* Check the pages is dirty and if it is send it */ | |
f20e2865 | 1448 | if (migration_bitmap_clear_dirty(rs, pss->block, pss->page)) { |
6d358d94 JQ |
1449 | /* |
1450 | * If xbzrle is on, stop using the data compression after first | |
1451 | * round of migration even if compression is enabled. In theory, | |
1452 | * xbzrle can do better than compression. | |
1453 | */ | |
6b6712ef JQ |
1454 | if (migrate_use_compression() && |
1455 | (rs->ram_bulk_stage || !migrate_use_xbzrle())) { | |
a0a8aa14 | 1456 | res = ram_save_compressed_page(rs, pss, last_stage); |
a82d593b | 1457 | } else { |
a0a8aa14 | 1458 | res = ram_save_page(rs, pss, last_stage); |
a82d593b DDAG |
1459 | } |
1460 | ||
1461 | if (res < 0) { | |
1462 | return res; | |
1463 | } | |
6b6712ef JQ |
1464 | if (pss->block->unsentmap) { |
1465 | clear_bit(pss->page, pss->block->unsentmap); | |
a82d593b DDAG |
1466 | } |
1467 | } | |
1468 | ||
1469 | return res; | |
1470 | } | |
1471 | ||
1472 | /** | |
3d0684b2 | 1473 | * ram_save_host_page: save a whole host page |
a82d593b | 1474 | * |
3d0684b2 JQ |
1475 | * Starting at *offset send pages up to the end of the current host |
1476 | * page. It's valid for the initial offset to point into the middle of | |
1477 | * a host page in which case the remainder of the hostpage is sent. | |
1478 | * Only dirty target pages are sent. Note that the host page size may | |
1479 | * be a huge page for this block. | |
1eb3fc0a DDAG |
1480 | * The saving stops at the boundary of the used_length of the block |
1481 | * if the RAMBlock isn't a multiple of the host page size. | |
a82d593b | 1482 | * |
3d0684b2 JQ |
1483 | * Returns the number of pages written or negative on error |
1484 | * | |
6f37bb8b | 1485 | * @rs: current RAM state |
3d0684b2 | 1486 | * @ms: current migration state |
3d0684b2 | 1487 | * @pss: data about the page we want to send |
a82d593b | 1488 | * @last_stage: if we are at the completion stage |
a82d593b | 1489 | */ |
a0a8aa14 | 1490 | static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 1491 | bool last_stage) |
a82d593b DDAG |
1492 | { |
1493 | int tmppages, pages = 0; | |
a935e30f JQ |
1494 | size_t pagesize_bits = |
1495 | qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
4c011c37 | 1496 | |
a82d593b | 1497 | do { |
f20e2865 | 1498 | tmppages = ram_save_target_page(rs, pss, last_stage); |
a82d593b DDAG |
1499 | if (tmppages < 0) { |
1500 | return tmppages; | |
1501 | } | |
1502 | ||
1503 | pages += tmppages; | |
a935e30f | 1504 | pss->page++; |
1eb3fc0a DDAG |
1505 | } while ((pss->page & (pagesize_bits - 1)) && |
1506 | offset_in_ramblock(pss->block, pss->page << TARGET_PAGE_BITS)); | |
a82d593b DDAG |
1507 | |
1508 | /* The offset we leave with is the last one we looked at */ | |
a935e30f | 1509 | pss->page--; |
a82d593b DDAG |
1510 | return pages; |
1511 | } | |
6c595cde | 1512 | |
56e93d26 | 1513 | /** |
3d0684b2 | 1514 | * ram_find_and_save_block: finds a dirty page and sends it to f |
56e93d26 JQ |
1515 | * |
1516 | * Called within an RCU critical section. | |
1517 | * | |
3d0684b2 | 1518 | * Returns the number of pages written where zero means no dirty pages |
56e93d26 | 1519 | * |
6f37bb8b | 1520 | * @rs: current RAM state |
56e93d26 | 1521 | * @last_stage: if we are at the completion stage |
a82d593b DDAG |
1522 | * |
1523 | * On systems where host-page-size > target-page-size it will send all the | |
1524 | * pages in a host page that are dirty. | |
56e93d26 JQ |
1525 | */ |
1526 | ||
ce25d337 | 1527 | static int ram_find_and_save_block(RAMState *rs, bool last_stage) |
56e93d26 | 1528 | { |
b8fb8cb7 | 1529 | PageSearchStatus pss; |
56e93d26 | 1530 | int pages = 0; |
b9e60928 | 1531 | bool again, found; |
56e93d26 | 1532 | |
0827b9e9 AA |
1533 | /* No dirty page as there is zero RAM */ |
1534 | if (!ram_bytes_total()) { | |
1535 | return pages; | |
1536 | } | |
1537 | ||
6f37bb8b | 1538 | pss.block = rs->last_seen_block; |
a935e30f | 1539 | pss.page = rs->last_page; |
b8fb8cb7 DDAG |
1540 | pss.complete_round = false; |
1541 | ||
1542 | if (!pss.block) { | |
1543 | pss.block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1544 | } | |
56e93d26 | 1545 | |
b9e60928 | 1546 | do { |
a82d593b | 1547 | again = true; |
f20e2865 | 1548 | found = get_queued_page(rs, &pss); |
b9e60928 | 1549 | |
a82d593b DDAG |
1550 | if (!found) { |
1551 | /* priority queue empty, so just search for something dirty */ | |
f20e2865 | 1552 | found = find_dirty_block(rs, &pss, &again); |
a82d593b | 1553 | } |
f3f491fc | 1554 | |
a82d593b | 1555 | if (found) { |
f20e2865 | 1556 | pages = ram_save_host_page(rs, &pss, last_stage); |
56e93d26 | 1557 | } |
b9e60928 | 1558 | } while (!pages && again); |
56e93d26 | 1559 | |
6f37bb8b | 1560 | rs->last_seen_block = pss.block; |
a935e30f | 1561 | rs->last_page = pss.page; |
56e93d26 JQ |
1562 | |
1563 | return pages; | |
1564 | } | |
1565 | ||
1566 | void acct_update_position(QEMUFile *f, size_t size, bool zero) | |
1567 | { | |
1568 | uint64_t pages = size / TARGET_PAGE_SIZE; | |
f7ccd61b | 1569 | |
56e93d26 | 1570 | if (zero) { |
9360447d | 1571 | ram_counters.duplicate += pages; |
56e93d26 | 1572 | } else { |
9360447d JQ |
1573 | ram_counters.normal += pages; |
1574 | ram_counters.transferred += size; | |
56e93d26 JQ |
1575 | qemu_update_position(f, size); |
1576 | } | |
1577 | } | |
1578 | ||
56e93d26 JQ |
1579 | uint64_t ram_bytes_total(void) |
1580 | { | |
1581 | RAMBlock *block; | |
1582 | uint64_t total = 0; | |
1583 | ||
1584 | rcu_read_lock(); | |
99e15582 | 1585 | RAMBLOCK_FOREACH(block) { |
56e93d26 | 1586 | total += block->used_length; |
99e15582 | 1587 | } |
56e93d26 JQ |
1588 | rcu_read_unlock(); |
1589 | return total; | |
1590 | } | |
1591 | ||
f265e0e4 | 1592 | static void xbzrle_load_setup(void) |
56e93d26 | 1593 | { |
f265e0e4 | 1594 | XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE); |
56e93d26 JQ |
1595 | } |
1596 | ||
f265e0e4 JQ |
1597 | static void xbzrle_load_cleanup(void) |
1598 | { | |
1599 | g_free(XBZRLE.decoded_buf); | |
1600 | XBZRLE.decoded_buf = NULL; | |
1601 | } | |
1602 | ||
7d7c96be PX |
1603 | static void ram_state_cleanup(RAMState **rsp) |
1604 | { | |
1605 | migration_page_queue_free(*rsp); | |
1606 | qemu_mutex_destroy(&(*rsp)->bitmap_mutex); | |
1607 | qemu_mutex_destroy(&(*rsp)->src_page_req_mutex); | |
1608 | g_free(*rsp); | |
1609 | *rsp = NULL; | |
1610 | } | |
1611 | ||
84593a08 PX |
1612 | static void xbzrle_cleanup(void) |
1613 | { | |
1614 | XBZRLE_cache_lock(); | |
1615 | if (XBZRLE.cache) { | |
1616 | cache_fini(XBZRLE.cache); | |
1617 | g_free(XBZRLE.encoded_buf); | |
1618 | g_free(XBZRLE.current_buf); | |
1619 | g_free(XBZRLE.zero_target_page); | |
1620 | XBZRLE.cache = NULL; | |
1621 | XBZRLE.encoded_buf = NULL; | |
1622 | XBZRLE.current_buf = NULL; | |
1623 | XBZRLE.zero_target_page = NULL; | |
1624 | } | |
1625 | XBZRLE_cache_unlock(); | |
1626 | } | |
1627 | ||
f265e0e4 | 1628 | static void ram_save_cleanup(void *opaque) |
56e93d26 | 1629 | { |
53518d94 | 1630 | RAMState **rsp = opaque; |
6b6712ef | 1631 | RAMBlock *block; |
eb859c53 | 1632 | |
2ff64038 LZ |
1633 | /* caller have hold iothread lock or is in a bh, so there is |
1634 | * no writing race against this migration_bitmap | |
1635 | */ | |
6b6712ef JQ |
1636 | memory_global_dirty_log_stop(); |
1637 | ||
1638 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1639 | g_free(block->bmap); | |
1640 | block->bmap = NULL; | |
1641 | g_free(block->unsentmap); | |
1642 | block->unsentmap = NULL; | |
56e93d26 JQ |
1643 | } |
1644 | ||
84593a08 | 1645 | xbzrle_cleanup(); |
f0afa331 | 1646 | compress_threads_save_cleanup(); |
7d7c96be | 1647 | ram_state_cleanup(rsp); |
56e93d26 JQ |
1648 | } |
1649 | ||
6f37bb8b | 1650 | static void ram_state_reset(RAMState *rs) |
56e93d26 | 1651 | { |
6f37bb8b JQ |
1652 | rs->last_seen_block = NULL; |
1653 | rs->last_sent_block = NULL; | |
269ace29 | 1654 | rs->last_page = 0; |
6f37bb8b JQ |
1655 | rs->last_version = ram_list.version; |
1656 | rs->ram_bulk_stage = true; | |
56e93d26 JQ |
1657 | } |
1658 | ||
1659 | #define MAX_WAIT 50 /* ms, half buffered_file limit */ | |
1660 | ||
4f2e4252 DDAG |
1661 | /* |
1662 | * 'expected' is the value you expect the bitmap mostly to be full | |
1663 | * of; it won't bother printing lines that are all this value. | |
1664 | * If 'todump' is null the migration bitmap is dumped. | |
1665 | */ | |
6b6712ef JQ |
1666 | void ram_debug_dump_bitmap(unsigned long *todump, bool expected, |
1667 | unsigned long pages) | |
4f2e4252 | 1668 | { |
4f2e4252 DDAG |
1669 | int64_t cur; |
1670 | int64_t linelen = 128; | |
1671 | char linebuf[129]; | |
1672 | ||
6b6712ef | 1673 | for (cur = 0; cur < pages; cur += linelen) { |
4f2e4252 DDAG |
1674 | int64_t curb; |
1675 | bool found = false; | |
1676 | /* | |
1677 | * Last line; catch the case where the line length | |
1678 | * is longer than remaining ram | |
1679 | */ | |
6b6712ef JQ |
1680 | if (cur + linelen > pages) { |
1681 | linelen = pages - cur; | |
4f2e4252 DDAG |
1682 | } |
1683 | for (curb = 0; curb < linelen; curb++) { | |
1684 | bool thisbit = test_bit(cur + curb, todump); | |
1685 | linebuf[curb] = thisbit ? '1' : '.'; | |
1686 | found = found || (thisbit != expected); | |
1687 | } | |
1688 | if (found) { | |
1689 | linebuf[curb] = '\0'; | |
1690 | fprintf(stderr, "0x%08" PRIx64 " : %s\n", cur, linebuf); | |
1691 | } | |
1692 | } | |
1693 | } | |
1694 | ||
e0b266f0 DDAG |
1695 | /* **** functions for postcopy ***** */ |
1696 | ||
ced1c616 PB |
1697 | void ram_postcopy_migrated_memory_release(MigrationState *ms) |
1698 | { | |
1699 | struct RAMBlock *block; | |
ced1c616 | 1700 | |
99e15582 | 1701 | RAMBLOCK_FOREACH(block) { |
6b6712ef JQ |
1702 | unsigned long *bitmap = block->bmap; |
1703 | unsigned long range = block->used_length >> TARGET_PAGE_BITS; | |
1704 | unsigned long run_start = find_next_zero_bit(bitmap, range, 0); | |
ced1c616 PB |
1705 | |
1706 | while (run_start < range) { | |
1707 | unsigned long run_end = find_next_bit(bitmap, range, run_start + 1); | |
aaa2064c | 1708 | ram_discard_range(block->idstr, run_start << TARGET_PAGE_BITS, |
ced1c616 PB |
1709 | (run_end - run_start) << TARGET_PAGE_BITS); |
1710 | run_start = find_next_zero_bit(bitmap, range, run_end + 1); | |
1711 | } | |
1712 | } | |
1713 | } | |
1714 | ||
3d0684b2 JQ |
1715 | /** |
1716 | * postcopy_send_discard_bm_ram: discard a RAMBlock | |
1717 | * | |
1718 | * Returns zero on success | |
1719 | * | |
e0b266f0 DDAG |
1720 | * Callback from postcopy_each_ram_send_discard for each RAMBlock |
1721 | * Note: At this point the 'unsentmap' is the processed bitmap combined | |
1722 | * with the dirtymap; so a '1' means it's either dirty or unsent. | |
3d0684b2 JQ |
1723 | * |
1724 | * @ms: current migration state | |
1725 | * @pds: state for postcopy | |
1726 | * @start: RAMBlock starting page | |
1727 | * @length: RAMBlock size | |
e0b266f0 DDAG |
1728 | */ |
1729 | static int postcopy_send_discard_bm_ram(MigrationState *ms, | |
1730 | PostcopyDiscardState *pds, | |
6b6712ef | 1731 | RAMBlock *block) |
e0b266f0 | 1732 | { |
6b6712ef | 1733 | unsigned long end = block->used_length >> TARGET_PAGE_BITS; |
e0b266f0 | 1734 | unsigned long current; |
6b6712ef | 1735 | unsigned long *unsentmap = block->unsentmap; |
e0b266f0 | 1736 | |
6b6712ef | 1737 | for (current = 0; current < end; ) { |
e0b266f0 DDAG |
1738 | unsigned long one = find_next_bit(unsentmap, end, current); |
1739 | ||
1740 | if (one <= end) { | |
1741 | unsigned long zero = find_next_zero_bit(unsentmap, end, one + 1); | |
1742 | unsigned long discard_length; | |
1743 | ||
1744 | if (zero >= end) { | |
1745 | discard_length = end - one; | |
1746 | } else { | |
1747 | discard_length = zero - one; | |
1748 | } | |
d688c62d DDAG |
1749 | if (discard_length) { |
1750 | postcopy_discard_send_range(ms, pds, one, discard_length); | |
1751 | } | |
e0b266f0 DDAG |
1752 | current = one + discard_length; |
1753 | } else { | |
1754 | current = one; | |
1755 | } | |
1756 | } | |
1757 | ||
1758 | return 0; | |
1759 | } | |
1760 | ||
3d0684b2 JQ |
1761 | /** |
1762 | * postcopy_each_ram_send_discard: discard all RAMBlocks | |
1763 | * | |
1764 | * Returns 0 for success or negative for error | |
1765 | * | |
e0b266f0 DDAG |
1766 | * Utility for the outgoing postcopy code. |
1767 | * Calls postcopy_send_discard_bm_ram for each RAMBlock | |
1768 | * passing it bitmap indexes and name. | |
e0b266f0 DDAG |
1769 | * (qemu_ram_foreach_block ends up passing unscaled lengths |
1770 | * which would mean postcopy code would have to deal with target page) | |
3d0684b2 JQ |
1771 | * |
1772 | * @ms: current migration state | |
e0b266f0 DDAG |
1773 | */ |
1774 | static int postcopy_each_ram_send_discard(MigrationState *ms) | |
1775 | { | |
1776 | struct RAMBlock *block; | |
1777 | int ret; | |
1778 | ||
99e15582 | 1779 | RAMBLOCK_FOREACH(block) { |
6b6712ef JQ |
1780 | PostcopyDiscardState *pds = |
1781 | postcopy_discard_send_init(ms, block->idstr); | |
e0b266f0 DDAG |
1782 | |
1783 | /* | |
1784 | * Postcopy sends chunks of bitmap over the wire, but it | |
1785 | * just needs indexes at this point, avoids it having | |
1786 | * target page specific code. | |
1787 | */ | |
6b6712ef | 1788 | ret = postcopy_send_discard_bm_ram(ms, pds, block); |
e0b266f0 DDAG |
1789 | postcopy_discard_send_finish(ms, pds); |
1790 | if (ret) { | |
1791 | return ret; | |
1792 | } | |
1793 | } | |
1794 | ||
1795 | return 0; | |
1796 | } | |
1797 | ||
3d0684b2 JQ |
1798 | /** |
1799 | * postcopy_chunk_hostpages_pass: canocalize bitmap in hostpages | |
1800 | * | |
1801 | * Helper for postcopy_chunk_hostpages; it's called twice to | |
1802 | * canonicalize the two bitmaps, that are similar, but one is | |
1803 | * inverted. | |
99e314eb | 1804 | * |
3d0684b2 JQ |
1805 | * Postcopy requires that all target pages in a hostpage are dirty or |
1806 | * clean, not a mix. This function canonicalizes the bitmaps. | |
99e314eb | 1807 | * |
3d0684b2 JQ |
1808 | * @ms: current migration state |
1809 | * @unsent_pass: if true we need to canonicalize partially unsent host pages | |
1810 | * otherwise we need to canonicalize partially dirty host pages | |
1811 | * @block: block that contains the page we want to canonicalize | |
1812 | * @pds: state for postcopy | |
99e314eb DDAG |
1813 | */ |
1814 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, bool unsent_pass, | |
1815 | RAMBlock *block, | |
1816 | PostcopyDiscardState *pds) | |
1817 | { | |
53518d94 | 1818 | RAMState *rs = ram_state; |
6b6712ef JQ |
1819 | unsigned long *bitmap = block->bmap; |
1820 | unsigned long *unsentmap = block->unsentmap; | |
29c59172 | 1821 | unsigned int host_ratio = block->page_size / TARGET_PAGE_SIZE; |
6b6712ef | 1822 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
99e314eb DDAG |
1823 | unsigned long run_start; |
1824 | ||
29c59172 DDAG |
1825 | if (block->page_size == TARGET_PAGE_SIZE) { |
1826 | /* Easy case - TPS==HPS for a non-huge page RAMBlock */ | |
1827 | return; | |
1828 | } | |
1829 | ||
99e314eb DDAG |
1830 | if (unsent_pass) { |
1831 | /* Find a sent page */ | |
6b6712ef | 1832 | run_start = find_next_zero_bit(unsentmap, pages, 0); |
99e314eb DDAG |
1833 | } else { |
1834 | /* Find a dirty page */ | |
6b6712ef | 1835 | run_start = find_next_bit(bitmap, pages, 0); |
99e314eb DDAG |
1836 | } |
1837 | ||
6b6712ef | 1838 | while (run_start < pages) { |
99e314eb DDAG |
1839 | bool do_fixup = false; |
1840 | unsigned long fixup_start_addr; | |
1841 | unsigned long host_offset; | |
1842 | ||
1843 | /* | |
1844 | * If the start of this run of pages is in the middle of a host | |
1845 | * page, then we need to fixup this host page. | |
1846 | */ | |
1847 | host_offset = run_start % host_ratio; | |
1848 | if (host_offset) { | |
1849 | do_fixup = true; | |
1850 | run_start -= host_offset; | |
1851 | fixup_start_addr = run_start; | |
1852 | /* For the next pass */ | |
1853 | run_start = run_start + host_ratio; | |
1854 | } else { | |
1855 | /* Find the end of this run */ | |
1856 | unsigned long run_end; | |
1857 | if (unsent_pass) { | |
6b6712ef | 1858 | run_end = find_next_bit(unsentmap, pages, run_start + 1); |
99e314eb | 1859 | } else { |
6b6712ef | 1860 | run_end = find_next_zero_bit(bitmap, pages, run_start + 1); |
99e314eb DDAG |
1861 | } |
1862 | /* | |
1863 | * If the end isn't at the start of a host page, then the | |
1864 | * run doesn't finish at the end of a host page | |
1865 | * and we need to discard. | |
1866 | */ | |
1867 | host_offset = run_end % host_ratio; | |
1868 | if (host_offset) { | |
1869 | do_fixup = true; | |
1870 | fixup_start_addr = run_end - host_offset; | |
1871 | /* | |
1872 | * This host page has gone, the next loop iteration starts | |
1873 | * from after the fixup | |
1874 | */ | |
1875 | run_start = fixup_start_addr + host_ratio; | |
1876 | } else { | |
1877 | /* | |
1878 | * No discards on this iteration, next loop starts from | |
1879 | * next sent/dirty page | |
1880 | */ | |
1881 | run_start = run_end + 1; | |
1882 | } | |
1883 | } | |
1884 | ||
1885 | if (do_fixup) { | |
1886 | unsigned long page; | |
1887 | ||
1888 | /* Tell the destination to discard this page */ | |
1889 | if (unsent_pass || !test_bit(fixup_start_addr, unsentmap)) { | |
1890 | /* For the unsent_pass we: | |
1891 | * discard partially sent pages | |
1892 | * For the !unsent_pass (dirty) we: | |
1893 | * discard partially dirty pages that were sent | |
1894 | * (any partially sent pages were already discarded | |
1895 | * by the previous unsent_pass) | |
1896 | */ | |
1897 | postcopy_discard_send_range(ms, pds, fixup_start_addr, | |
1898 | host_ratio); | |
1899 | } | |
1900 | ||
1901 | /* Clean up the bitmap */ | |
1902 | for (page = fixup_start_addr; | |
1903 | page < fixup_start_addr + host_ratio; page++) { | |
1904 | /* All pages in this host page are now not sent */ | |
1905 | set_bit(page, unsentmap); | |
1906 | ||
1907 | /* | |
1908 | * Remark them as dirty, updating the count for any pages | |
1909 | * that weren't previously dirty. | |
1910 | */ | |
0d8ec885 | 1911 | rs->migration_dirty_pages += !test_and_set_bit(page, bitmap); |
99e314eb DDAG |
1912 | } |
1913 | } | |
1914 | ||
1915 | if (unsent_pass) { | |
1916 | /* Find the next sent page for the next iteration */ | |
6b6712ef | 1917 | run_start = find_next_zero_bit(unsentmap, pages, run_start); |
99e314eb DDAG |
1918 | } else { |
1919 | /* Find the next dirty page for the next iteration */ | |
6b6712ef | 1920 | run_start = find_next_bit(bitmap, pages, run_start); |
99e314eb DDAG |
1921 | } |
1922 | } | |
1923 | } | |
1924 | ||
3d0684b2 JQ |
1925 | /** |
1926 | * postcopy_chuck_hostpages: discrad any partially sent host page | |
1927 | * | |
99e314eb DDAG |
1928 | * Utility for the outgoing postcopy code. |
1929 | * | |
1930 | * Discard any partially sent host-page size chunks, mark any partially | |
29c59172 DDAG |
1931 | * dirty host-page size chunks as all dirty. In this case the host-page |
1932 | * is the host-page for the particular RAMBlock, i.e. it might be a huge page | |
99e314eb | 1933 | * |
3d0684b2 JQ |
1934 | * Returns zero on success |
1935 | * | |
1936 | * @ms: current migration state | |
6b6712ef | 1937 | * @block: block we want to work with |
99e314eb | 1938 | */ |
6b6712ef | 1939 | static int postcopy_chunk_hostpages(MigrationState *ms, RAMBlock *block) |
99e314eb | 1940 | { |
6b6712ef JQ |
1941 | PostcopyDiscardState *pds = |
1942 | postcopy_discard_send_init(ms, block->idstr); | |
99e314eb | 1943 | |
6b6712ef JQ |
1944 | /* First pass: Discard all partially sent host pages */ |
1945 | postcopy_chunk_hostpages_pass(ms, true, block, pds); | |
1946 | /* | |
1947 | * Second pass: Ensure that all partially dirty host pages are made | |
1948 | * fully dirty. | |
1949 | */ | |
1950 | postcopy_chunk_hostpages_pass(ms, false, block, pds); | |
99e314eb | 1951 | |
6b6712ef | 1952 | postcopy_discard_send_finish(ms, pds); |
99e314eb DDAG |
1953 | return 0; |
1954 | } | |
1955 | ||
3d0684b2 JQ |
1956 | /** |
1957 | * ram_postcopy_send_discard_bitmap: transmit the discard bitmap | |
1958 | * | |
1959 | * Returns zero on success | |
1960 | * | |
e0b266f0 DDAG |
1961 | * Transmit the set of pages to be discarded after precopy to the target |
1962 | * these are pages that: | |
1963 | * a) Have been previously transmitted but are now dirty again | |
1964 | * b) Pages that have never been transmitted, this ensures that | |
1965 | * any pages on the destination that have been mapped by background | |
1966 | * tasks get discarded (transparent huge pages is the specific concern) | |
1967 | * Hopefully this is pretty sparse | |
3d0684b2 JQ |
1968 | * |
1969 | * @ms: current migration state | |
e0b266f0 DDAG |
1970 | */ |
1971 | int ram_postcopy_send_discard_bitmap(MigrationState *ms) | |
1972 | { | |
53518d94 | 1973 | RAMState *rs = ram_state; |
6b6712ef | 1974 | RAMBlock *block; |
e0b266f0 | 1975 | int ret; |
e0b266f0 DDAG |
1976 | |
1977 | rcu_read_lock(); | |
1978 | ||
1979 | /* This should be our last sync, the src is now paused */ | |
eb859c53 | 1980 | migration_bitmap_sync(rs); |
e0b266f0 | 1981 | |
6b6712ef JQ |
1982 | /* Easiest way to make sure we don't resume in the middle of a host-page */ |
1983 | rs->last_seen_block = NULL; | |
1984 | rs->last_sent_block = NULL; | |
1985 | rs->last_page = 0; | |
e0b266f0 | 1986 | |
6b6712ef JQ |
1987 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { |
1988 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; | |
1989 | unsigned long *bitmap = block->bmap; | |
1990 | unsigned long *unsentmap = block->unsentmap; | |
1991 | ||
1992 | if (!unsentmap) { | |
1993 | /* We don't have a safe way to resize the sentmap, so | |
1994 | * if the bitmap was resized it will be NULL at this | |
1995 | * point. | |
1996 | */ | |
1997 | error_report("migration ram resized during precopy phase"); | |
1998 | rcu_read_unlock(); | |
1999 | return -EINVAL; | |
2000 | } | |
2001 | /* Deal with TPS != HPS and huge pages */ | |
2002 | ret = postcopy_chunk_hostpages(ms, block); | |
2003 | if (ret) { | |
2004 | rcu_read_unlock(); | |
2005 | return ret; | |
2006 | } | |
e0b266f0 | 2007 | |
6b6712ef JQ |
2008 | /* |
2009 | * Update the unsentmap to be unsentmap = unsentmap | dirty | |
2010 | */ | |
2011 | bitmap_or(unsentmap, unsentmap, bitmap, pages); | |
e0b266f0 | 2012 | #ifdef DEBUG_POSTCOPY |
6b6712ef | 2013 | ram_debug_dump_bitmap(unsentmap, true, pages); |
e0b266f0 | 2014 | #endif |
6b6712ef JQ |
2015 | } |
2016 | trace_ram_postcopy_send_discard_bitmap(); | |
e0b266f0 DDAG |
2017 | |
2018 | ret = postcopy_each_ram_send_discard(ms); | |
2019 | rcu_read_unlock(); | |
2020 | ||
2021 | return ret; | |
2022 | } | |
2023 | ||
3d0684b2 JQ |
2024 | /** |
2025 | * ram_discard_range: discard dirtied pages at the beginning of postcopy | |
e0b266f0 | 2026 | * |
3d0684b2 | 2027 | * Returns zero on success |
e0b266f0 | 2028 | * |
36449157 JQ |
2029 | * @rbname: name of the RAMBlock of the request. NULL means the |
2030 | * same that last one. | |
3d0684b2 JQ |
2031 | * @start: RAMBlock starting page |
2032 | * @length: RAMBlock size | |
e0b266f0 | 2033 | */ |
aaa2064c | 2034 | int ram_discard_range(const char *rbname, uint64_t start, size_t length) |
e0b266f0 DDAG |
2035 | { |
2036 | int ret = -1; | |
2037 | ||
36449157 | 2038 | trace_ram_discard_range(rbname, start, length); |
d3a5038c | 2039 | |
e0b266f0 | 2040 | rcu_read_lock(); |
36449157 | 2041 | RAMBlock *rb = qemu_ram_block_by_name(rbname); |
e0b266f0 DDAG |
2042 | |
2043 | if (!rb) { | |
36449157 | 2044 | error_report("ram_discard_range: Failed to find block '%s'", rbname); |
e0b266f0 DDAG |
2045 | goto err; |
2046 | } | |
2047 | ||
f9494614 AP |
2048 | bitmap_clear(rb->receivedmap, start >> qemu_target_page_bits(), |
2049 | length >> qemu_target_page_bits()); | |
d3a5038c | 2050 | ret = ram_block_discard_range(rb, start, length); |
e0b266f0 DDAG |
2051 | |
2052 | err: | |
2053 | rcu_read_unlock(); | |
2054 | ||
2055 | return ret; | |
2056 | } | |
2057 | ||
84593a08 PX |
2058 | /* |
2059 | * For every allocation, we will try not to crash the VM if the | |
2060 | * allocation failed. | |
2061 | */ | |
2062 | static int xbzrle_init(void) | |
2063 | { | |
2064 | Error *local_err = NULL; | |
2065 | ||
2066 | if (!migrate_use_xbzrle()) { | |
2067 | return 0; | |
2068 | } | |
2069 | ||
2070 | XBZRLE_cache_lock(); | |
2071 | ||
2072 | XBZRLE.zero_target_page = g_try_malloc0(TARGET_PAGE_SIZE); | |
2073 | if (!XBZRLE.zero_target_page) { | |
2074 | error_report("%s: Error allocating zero page", __func__); | |
2075 | goto err_out; | |
2076 | } | |
2077 | ||
2078 | XBZRLE.cache = cache_init(migrate_xbzrle_cache_size(), | |
2079 | TARGET_PAGE_SIZE, &local_err); | |
2080 | if (!XBZRLE.cache) { | |
2081 | error_report_err(local_err); | |
2082 | goto free_zero_page; | |
2083 | } | |
2084 | ||
2085 | XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); | |
2086 | if (!XBZRLE.encoded_buf) { | |
2087 | error_report("%s: Error allocating encoded_buf", __func__); | |
2088 | goto free_cache; | |
2089 | } | |
2090 | ||
2091 | XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); | |
2092 | if (!XBZRLE.current_buf) { | |
2093 | error_report("%s: Error allocating current_buf", __func__); | |
2094 | goto free_encoded_buf; | |
2095 | } | |
2096 | ||
2097 | /* We are all good */ | |
2098 | XBZRLE_cache_unlock(); | |
2099 | return 0; | |
2100 | ||
2101 | free_encoded_buf: | |
2102 | g_free(XBZRLE.encoded_buf); | |
2103 | XBZRLE.encoded_buf = NULL; | |
2104 | free_cache: | |
2105 | cache_fini(XBZRLE.cache); | |
2106 | XBZRLE.cache = NULL; | |
2107 | free_zero_page: | |
2108 | g_free(XBZRLE.zero_target_page); | |
2109 | XBZRLE.zero_target_page = NULL; | |
2110 | err_out: | |
2111 | XBZRLE_cache_unlock(); | |
2112 | return -ENOMEM; | |
2113 | } | |
2114 | ||
53518d94 | 2115 | static int ram_state_init(RAMState **rsp) |
56e93d26 | 2116 | { |
7d00ee6a PX |
2117 | *rsp = g_try_new0(RAMState, 1); |
2118 | ||
2119 | if (!*rsp) { | |
2120 | error_report("%s: Init ramstate fail", __func__); | |
2121 | return -1; | |
2122 | } | |
53518d94 JQ |
2123 | |
2124 | qemu_mutex_init(&(*rsp)->bitmap_mutex); | |
2125 | qemu_mutex_init(&(*rsp)->src_page_req_mutex); | |
2126 | QSIMPLEQ_INIT(&(*rsp)->src_page_requests); | |
56e93d26 | 2127 | |
7d00ee6a PX |
2128 | /* |
2129 | * Count the total number of pages used by ram blocks not including any | |
2130 | * gaps due to alignment or unplugs. | |
2131 | */ | |
2132 | (*rsp)->migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS; | |
2133 | ||
2134 | ram_state_reset(*rsp); | |
2135 | ||
2136 | return 0; | |
2137 | } | |
2138 | ||
d6eff5d7 | 2139 | static void ram_list_init_bitmaps(void) |
7d00ee6a | 2140 | { |
d6eff5d7 PX |
2141 | RAMBlock *block; |
2142 | unsigned long pages; | |
56e93d26 | 2143 | |
0827b9e9 AA |
2144 | /* Skip setting bitmap if there is no RAM */ |
2145 | if (ram_bytes_total()) { | |
6b6712ef | 2146 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { |
d6eff5d7 | 2147 | pages = block->max_length >> TARGET_PAGE_BITS; |
6b6712ef JQ |
2148 | block->bmap = bitmap_new(pages); |
2149 | bitmap_set(block->bmap, 0, pages); | |
2150 | if (migrate_postcopy_ram()) { | |
2151 | block->unsentmap = bitmap_new(pages); | |
2152 | bitmap_set(block->unsentmap, 0, pages); | |
2153 | } | |
0827b9e9 | 2154 | } |
f3f491fc | 2155 | } |
d6eff5d7 PX |
2156 | } |
2157 | ||
2158 | static void ram_init_bitmaps(RAMState *rs) | |
2159 | { | |
2160 | /* For memory_global_dirty_log_start below. */ | |
2161 | qemu_mutex_lock_iothread(); | |
2162 | qemu_mutex_lock_ramlist(); | |
2163 | rcu_read_lock(); | |
f3f491fc | 2164 | |
d6eff5d7 | 2165 | ram_list_init_bitmaps(); |
56e93d26 | 2166 | memory_global_dirty_log_start(); |
d6eff5d7 PX |
2167 | migration_bitmap_sync(rs); |
2168 | ||
2169 | rcu_read_unlock(); | |
56e93d26 | 2170 | qemu_mutex_unlock_ramlist(); |
49877834 | 2171 | qemu_mutex_unlock_iothread(); |
d6eff5d7 PX |
2172 | } |
2173 | ||
2174 | static int ram_init_all(RAMState **rsp) | |
2175 | { | |
2176 | if (ram_state_init(rsp)) { | |
2177 | return -1; | |
2178 | } | |
2179 | ||
2180 | if (xbzrle_init()) { | |
2181 | ram_state_cleanup(rsp); | |
2182 | return -1; | |
2183 | } | |
2184 | ||
2185 | ram_init_bitmaps(*rsp); | |
a91246c9 HZ |
2186 | |
2187 | return 0; | |
2188 | } | |
2189 | ||
3d0684b2 JQ |
2190 | /* |
2191 | * Each of ram_save_setup, ram_save_iterate and ram_save_complete has | |
a91246c9 HZ |
2192 | * long-running RCU critical section. When rcu-reclaims in the code |
2193 | * start to become numerous it will be necessary to reduce the | |
2194 | * granularity of these critical sections. | |
2195 | */ | |
2196 | ||
3d0684b2 JQ |
2197 | /** |
2198 | * ram_save_setup: Setup RAM for migration | |
2199 | * | |
2200 | * Returns zero to indicate success and negative for error | |
2201 | * | |
2202 | * @f: QEMUFile where to send the data | |
2203 | * @opaque: RAMState pointer | |
2204 | */ | |
a91246c9 HZ |
2205 | static int ram_save_setup(QEMUFile *f, void *opaque) |
2206 | { | |
53518d94 | 2207 | RAMState **rsp = opaque; |
a91246c9 HZ |
2208 | RAMBlock *block; |
2209 | ||
2210 | /* migration has already setup the bitmap, reuse it. */ | |
2211 | if (!migration_in_colo_state()) { | |
7d00ee6a | 2212 | if (ram_init_all(rsp) != 0) { |
a91246c9 | 2213 | return -1; |
53518d94 | 2214 | } |
a91246c9 | 2215 | } |
53518d94 | 2216 | (*rsp)->f = f; |
a91246c9 HZ |
2217 | |
2218 | rcu_read_lock(); | |
56e93d26 JQ |
2219 | |
2220 | qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE); | |
2221 | ||
99e15582 | 2222 | RAMBLOCK_FOREACH(block) { |
56e93d26 JQ |
2223 | qemu_put_byte(f, strlen(block->idstr)); |
2224 | qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); | |
2225 | qemu_put_be64(f, block->used_length); | |
ef08fb38 DDAG |
2226 | if (migrate_postcopy_ram() && block->page_size != qemu_host_page_size) { |
2227 | qemu_put_be64(f, block->page_size); | |
2228 | } | |
56e93d26 JQ |
2229 | } |
2230 | ||
2231 | rcu_read_unlock(); | |
f0afa331 | 2232 | compress_threads_save_setup(); |
56e93d26 JQ |
2233 | |
2234 | ram_control_before_iterate(f, RAM_CONTROL_SETUP); | |
2235 | ram_control_after_iterate(f, RAM_CONTROL_SETUP); | |
2236 | ||
2237 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); | |
2238 | ||
2239 | return 0; | |
2240 | } | |
2241 | ||
3d0684b2 JQ |
2242 | /** |
2243 | * ram_save_iterate: iterative stage for migration | |
2244 | * | |
2245 | * Returns zero to indicate success and negative for error | |
2246 | * | |
2247 | * @f: QEMUFile where to send the data | |
2248 | * @opaque: RAMState pointer | |
2249 | */ | |
56e93d26 JQ |
2250 | static int ram_save_iterate(QEMUFile *f, void *opaque) |
2251 | { | |
53518d94 JQ |
2252 | RAMState **temp = opaque; |
2253 | RAMState *rs = *temp; | |
56e93d26 JQ |
2254 | int ret; |
2255 | int i; | |
2256 | int64_t t0; | |
5c90308f | 2257 | int done = 0; |
56e93d26 JQ |
2258 | |
2259 | rcu_read_lock(); | |
6f37bb8b JQ |
2260 | if (ram_list.version != rs->last_version) { |
2261 | ram_state_reset(rs); | |
56e93d26 JQ |
2262 | } |
2263 | ||
2264 | /* Read version before ram_list.blocks */ | |
2265 | smp_rmb(); | |
2266 | ||
2267 | ram_control_before_iterate(f, RAM_CONTROL_ROUND); | |
2268 | ||
2269 | t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
2270 | i = 0; | |
2271 | while ((ret = qemu_file_rate_limit(f)) == 0) { | |
2272 | int pages; | |
2273 | ||
ce25d337 | 2274 | pages = ram_find_and_save_block(rs, false); |
56e93d26 JQ |
2275 | /* no more pages to sent */ |
2276 | if (pages == 0) { | |
5c90308f | 2277 | done = 1; |
56e93d26 JQ |
2278 | break; |
2279 | } | |
23b28c3c | 2280 | rs->iterations++; |
070afca2 | 2281 | |
56e93d26 JQ |
2282 | /* we want to check in the 1st loop, just in case it was the 1st time |
2283 | and we had to sync the dirty bitmap. | |
2284 | qemu_get_clock_ns() is a bit expensive, so we only check each some | |
2285 | iterations | |
2286 | */ | |
2287 | if ((i & 63) == 0) { | |
2288 | uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000; | |
2289 | if (t1 > MAX_WAIT) { | |
55c4446b | 2290 | trace_ram_save_iterate_big_wait(t1, i); |
56e93d26 JQ |
2291 | break; |
2292 | } | |
2293 | } | |
2294 | i++; | |
2295 | } | |
ce25d337 | 2296 | flush_compressed_data(rs); |
56e93d26 JQ |
2297 | rcu_read_unlock(); |
2298 | ||
2299 | /* | |
2300 | * Must occur before EOS (or any QEMUFile operation) | |
2301 | * because of RDMA protocol. | |
2302 | */ | |
2303 | ram_control_after_iterate(f, RAM_CONTROL_ROUND); | |
2304 | ||
2305 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); | |
9360447d | 2306 | ram_counters.transferred += 8; |
56e93d26 JQ |
2307 | |
2308 | ret = qemu_file_get_error(f); | |
2309 | if (ret < 0) { | |
2310 | return ret; | |
2311 | } | |
2312 | ||
5c90308f | 2313 | return done; |
56e93d26 JQ |
2314 | } |
2315 | ||
3d0684b2 JQ |
2316 | /** |
2317 | * ram_save_complete: function called to send the remaining amount of ram | |
2318 | * | |
2319 | * Returns zero to indicate success | |
2320 | * | |
2321 | * Called with iothread lock | |
2322 | * | |
2323 | * @f: QEMUFile where to send the data | |
2324 | * @opaque: RAMState pointer | |
2325 | */ | |
56e93d26 JQ |
2326 | static int ram_save_complete(QEMUFile *f, void *opaque) |
2327 | { | |
53518d94 JQ |
2328 | RAMState **temp = opaque; |
2329 | RAMState *rs = *temp; | |
6f37bb8b | 2330 | |
56e93d26 JQ |
2331 | rcu_read_lock(); |
2332 | ||
5727309d | 2333 | if (!migration_in_postcopy()) { |
8d820d6f | 2334 | migration_bitmap_sync(rs); |
663e6c1d | 2335 | } |
56e93d26 JQ |
2336 | |
2337 | ram_control_before_iterate(f, RAM_CONTROL_FINISH); | |
2338 | ||
2339 | /* try transferring iterative blocks of memory */ | |
2340 | ||
2341 | /* flush all remaining blocks regardless of rate limiting */ | |
2342 | while (true) { | |
2343 | int pages; | |
2344 | ||
ce25d337 | 2345 | pages = ram_find_and_save_block(rs, !migration_in_colo_state()); |
56e93d26 JQ |
2346 | /* no more blocks to sent */ |
2347 | if (pages == 0) { | |
2348 | break; | |
2349 | } | |
2350 | } | |
2351 | ||
ce25d337 | 2352 | flush_compressed_data(rs); |
56e93d26 | 2353 | ram_control_after_iterate(f, RAM_CONTROL_FINISH); |
56e93d26 JQ |
2354 | |
2355 | rcu_read_unlock(); | |
d09a6fde | 2356 | |
56e93d26 JQ |
2357 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
2358 | ||
2359 | return 0; | |
2360 | } | |
2361 | ||
c31b098f DDAG |
2362 | static void ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size, |
2363 | uint64_t *non_postcopiable_pending, | |
2364 | uint64_t *postcopiable_pending) | |
56e93d26 | 2365 | { |
53518d94 JQ |
2366 | RAMState **temp = opaque; |
2367 | RAMState *rs = *temp; | |
56e93d26 JQ |
2368 | uint64_t remaining_size; |
2369 | ||
9edabd4d | 2370 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2371 | |
5727309d | 2372 | if (!migration_in_postcopy() && |
663e6c1d | 2373 | remaining_size < max_size) { |
56e93d26 JQ |
2374 | qemu_mutex_lock_iothread(); |
2375 | rcu_read_lock(); | |
8d820d6f | 2376 | migration_bitmap_sync(rs); |
56e93d26 JQ |
2377 | rcu_read_unlock(); |
2378 | qemu_mutex_unlock_iothread(); | |
9edabd4d | 2379 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2380 | } |
c31b098f | 2381 | |
86e1167e VSO |
2382 | if (migrate_postcopy_ram()) { |
2383 | /* We can do postcopy, and all the data is postcopiable */ | |
2384 | *postcopiable_pending += remaining_size; | |
2385 | } else { | |
2386 | *non_postcopiable_pending += remaining_size; | |
2387 | } | |
56e93d26 JQ |
2388 | } |
2389 | ||
2390 | static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) | |
2391 | { | |
2392 | unsigned int xh_len; | |
2393 | int xh_flags; | |
063e760a | 2394 | uint8_t *loaded_data; |
56e93d26 | 2395 | |
56e93d26 JQ |
2396 | /* extract RLE header */ |
2397 | xh_flags = qemu_get_byte(f); | |
2398 | xh_len = qemu_get_be16(f); | |
2399 | ||
2400 | if (xh_flags != ENCODING_FLAG_XBZRLE) { | |
2401 | error_report("Failed to load XBZRLE page - wrong compression!"); | |
2402 | return -1; | |
2403 | } | |
2404 | ||
2405 | if (xh_len > TARGET_PAGE_SIZE) { | |
2406 | error_report("Failed to load XBZRLE page - len overflow!"); | |
2407 | return -1; | |
2408 | } | |
f265e0e4 | 2409 | loaded_data = XBZRLE.decoded_buf; |
56e93d26 | 2410 | /* load data and decode */ |
f265e0e4 | 2411 | /* it can change loaded_data to point to an internal buffer */ |
063e760a | 2412 | qemu_get_buffer_in_place(f, &loaded_data, xh_len); |
56e93d26 JQ |
2413 | |
2414 | /* decode RLE */ | |
063e760a | 2415 | if (xbzrle_decode_buffer(loaded_data, xh_len, host, |
56e93d26 JQ |
2416 | TARGET_PAGE_SIZE) == -1) { |
2417 | error_report("Failed to load XBZRLE page - decode error!"); | |
2418 | return -1; | |
2419 | } | |
2420 | ||
2421 | return 0; | |
2422 | } | |
2423 | ||
3d0684b2 JQ |
2424 | /** |
2425 | * ram_block_from_stream: read a RAMBlock id from the migration stream | |
2426 | * | |
2427 | * Must be called from within a rcu critical section. | |
2428 | * | |
56e93d26 | 2429 | * Returns a pointer from within the RCU-protected ram_list. |
a7180877 | 2430 | * |
3d0684b2 JQ |
2431 | * @f: QEMUFile where to read the data from |
2432 | * @flags: Page flags (mostly to see if it's a continuation of previous block) | |
a7180877 | 2433 | */ |
3d0684b2 | 2434 | static inline RAMBlock *ram_block_from_stream(QEMUFile *f, int flags) |
56e93d26 JQ |
2435 | { |
2436 | static RAMBlock *block = NULL; | |
2437 | char id[256]; | |
2438 | uint8_t len; | |
2439 | ||
2440 | if (flags & RAM_SAVE_FLAG_CONTINUE) { | |
4c4bad48 | 2441 | if (!block) { |
56e93d26 JQ |
2442 | error_report("Ack, bad migration stream!"); |
2443 | return NULL; | |
2444 | } | |
4c4bad48 | 2445 | return block; |
56e93d26 JQ |
2446 | } |
2447 | ||
2448 | len = qemu_get_byte(f); | |
2449 | qemu_get_buffer(f, (uint8_t *)id, len); | |
2450 | id[len] = 0; | |
2451 | ||
e3dd7493 | 2452 | block = qemu_ram_block_by_name(id); |
4c4bad48 HZ |
2453 | if (!block) { |
2454 | error_report("Can't find block %s", id); | |
2455 | return NULL; | |
56e93d26 JQ |
2456 | } |
2457 | ||
4c4bad48 HZ |
2458 | return block; |
2459 | } | |
2460 | ||
2461 | static inline void *host_from_ram_block_offset(RAMBlock *block, | |
2462 | ram_addr_t offset) | |
2463 | { | |
2464 | if (!offset_in_ramblock(block, offset)) { | |
2465 | return NULL; | |
2466 | } | |
2467 | ||
2468 | return block->host + offset; | |
56e93d26 JQ |
2469 | } |
2470 | ||
3d0684b2 JQ |
2471 | /** |
2472 | * ram_handle_compressed: handle the zero page case | |
2473 | * | |
56e93d26 JQ |
2474 | * If a page (or a whole RDMA chunk) has been |
2475 | * determined to be zero, then zap it. | |
3d0684b2 JQ |
2476 | * |
2477 | * @host: host address for the zero page | |
2478 | * @ch: what the page is filled from. We only support zero | |
2479 | * @size: size of the zero page | |
56e93d26 JQ |
2480 | */ |
2481 | void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) | |
2482 | { | |
2483 | if (ch != 0 || !is_zero_range(host, size)) { | |
2484 | memset(host, ch, size); | |
2485 | } | |
2486 | } | |
2487 | ||
2488 | static void *do_data_decompress(void *opaque) | |
2489 | { | |
2490 | DecompressParam *param = opaque; | |
2491 | unsigned long pagesize; | |
33d151f4 LL |
2492 | uint8_t *des; |
2493 | int len; | |
56e93d26 | 2494 | |
33d151f4 | 2495 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 2496 | while (!param->quit) { |
33d151f4 LL |
2497 | if (param->des) { |
2498 | des = param->des; | |
2499 | len = param->len; | |
2500 | param->des = 0; | |
2501 | qemu_mutex_unlock(¶m->mutex); | |
2502 | ||
56e93d26 | 2503 | pagesize = TARGET_PAGE_SIZE; |
73a8912b LL |
2504 | /* uncompress() will return failed in some case, especially |
2505 | * when the page is dirted when doing the compression, it's | |
2506 | * not a problem because the dirty page will be retransferred | |
2507 | * and uncompress() won't break the data in other pages. | |
2508 | */ | |
33d151f4 LL |
2509 | uncompress((Bytef *)des, &pagesize, |
2510 | (const Bytef *)param->compbuf, len); | |
73a8912b | 2511 | |
33d151f4 LL |
2512 | qemu_mutex_lock(&decomp_done_lock); |
2513 | param->done = true; | |
2514 | qemu_cond_signal(&decomp_done_cond); | |
2515 | qemu_mutex_unlock(&decomp_done_lock); | |
2516 | ||
2517 | qemu_mutex_lock(¶m->mutex); | |
2518 | } else { | |
2519 | qemu_cond_wait(¶m->cond, ¶m->mutex); | |
2520 | } | |
56e93d26 | 2521 | } |
33d151f4 | 2522 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
2523 | |
2524 | return NULL; | |
2525 | } | |
2526 | ||
5533b2e9 LL |
2527 | static void wait_for_decompress_done(void) |
2528 | { | |
2529 | int idx, thread_count; | |
2530 | ||
2531 | if (!migrate_use_compression()) { | |
2532 | return; | |
2533 | } | |
2534 | ||
2535 | thread_count = migrate_decompress_threads(); | |
2536 | qemu_mutex_lock(&decomp_done_lock); | |
2537 | for (idx = 0; idx < thread_count; idx++) { | |
2538 | while (!decomp_param[idx].done) { | |
2539 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
2540 | } | |
2541 | } | |
2542 | qemu_mutex_unlock(&decomp_done_lock); | |
2543 | } | |
2544 | ||
f0afa331 | 2545 | static void compress_threads_load_setup(void) |
56e93d26 JQ |
2546 | { |
2547 | int i, thread_count; | |
2548 | ||
3416ab5b JQ |
2549 | if (!migrate_use_compression()) { |
2550 | return; | |
2551 | } | |
56e93d26 JQ |
2552 | thread_count = migrate_decompress_threads(); |
2553 | decompress_threads = g_new0(QemuThread, thread_count); | |
2554 | decomp_param = g_new0(DecompressParam, thread_count); | |
73a8912b LL |
2555 | qemu_mutex_init(&decomp_done_lock); |
2556 | qemu_cond_init(&decomp_done_cond); | |
56e93d26 JQ |
2557 | for (i = 0; i < thread_count; i++) { |
2558 | qemu_mutex_init(&decomp_param[i].mutex); | |
2559 | qemu_cond_init(&decomp_param[i].cond); | |
2560 | decomp_param[i].compbuf = g_malloc0(compressBound(TARGET_PAGE_SIZE)); | |
73a8912b | 2561 | decomp_param[i].done = true; |
90e56fb4 | 2562 | decomp_param[i].quit = false; |
56e93d26 JQ |
2563 | qemu_thread_create(decompress_threads + i, "decompress", |
2564 | do_data_decompress, decomp_param + i, | |
2565 | QEMU_THREAD_JOINABLE); | |
2566 | } | |
2567 | } | |
2568 | ||
f0afa331 | 2569 | static void compress_threads_load_cleanup(void) |
56e93d26 JQ |
2570 | { |
2571 | int i, thread_count; | |
2572 | ||
3416ab5b JQ |
2573 | if (!migrate_use_compression()) { |
2574 | return; | |
2575 | } | |
56e93d26 JQ |
2576 | thread_count = migrate_decompress_threads(); |
2577 | for (i = 0; i < thread_count; i++) { | |
2578 | qemu_mutex_lock(&decomp_param[i].mutex); | |
90e56fb4 | 2579 | decomp_param[i].quit = true; |
56e93d26 JQ |
2580 | qemu_cond_signal(&decomp_param[i].cond); |
2581 | qemu_mutex_unlock(&decomp_param[i].mutex); | |
2582 | } | |
2583 | for (i = 0; i < thread_count; i++) { | |
2584 | qemu_thread_join(decompress_threads + i); | |
2585 | qemu_mutex_destroy(&decomp_param[i].mutex); | |
2586 | qemu_cond_destroy(&decomp_param[i].cond); | |
2587 | g_free(decomp_param[i].compbuf); | |
2588 | } | |
2589 | g_free(decompress_threads); | |
2590 | g_free(decomp_param); | |
56e93d26 JQ |
2591 | decompress_threads = NULL; |
2592 | decomp_param = NULL; | |
56e93d26 JQ |
2593 | } |
2594 | ||
c1bc6626 | 2595 | static void decompress_data_with_multi_threads(QEMUFile *f, |
56e93d26 JQ |
2596 | void *host, int len) |
2597 | { | |
2598 | int idx, thread_count; | |
2599 | ||
2600 | thread_count = migrate_decompress_threads(); | |
73a8912b | 2601 | qemu_mutex_lock(&decomp_done_lock); |
56e93d26 JQ |
2602 | while (true) { |
2603 | for (idx = 0; idx < thread_count; idx++) { | |
73a8912b | 2604 | if (decomp_param[idx].done) { |
33d151f4 LL |
2605 | decomp_param[idx].done = false; |
2606 | qemu_mutex_lock(&decomp_param[idx].mutex); | |
c1bc6626 | 2607 | qemu_get_buffer(f, decomp_param[idx].compbuf, len); |
56e93d26 JQ |
2608 | decomp_param[idx].des = host; |
2609 | decomp_param[idx].len = len; | |
33d151f4 LL |
2610 | qemu_cond_signal(&decomp_param[idx].cond); |
2611 | qemu_mutex_unlock(&decomp_param[idx].mutex); | |
56e93d26 JQ |
2612 | break; |
2613 | } | |
2614 | } | |
2615 | if (idx < thread_count) { | |
2616 | break; | |
73a8912b LL |
2617 | } else { |
2618 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
56e93d26 JQ |
2619 | } |
2620 | } | |
73a8912b | 2621 | qemu_mutex_unlock(&decomp_done_lock); |
56e93d26 JQ |
2622 | } |
2623 | ||
f265e0e4 JQ |
2624 | /** |
2625 | * ram_load_setup: Setup RAM for migration incoming side | |
2626 | * | |
2627 | * Returns zero to indicate success and negative for error | |
2628 | * | |
2629 | * @f: QEMUFile where to receive the data | |
2630 | * @opaque: RAMState pointer | |
2631 | */ | |
2632 | static int ram_load_setup(QEMUFile *f, void *opaque) | |
2633 | { | |
2634 | xbzrle_load_setup(); | |
f0afa331 | 2635 | compress_threads_load_setup(); |
f9494614 | 2636 | ramblock_recv_map_init(); |
f265e0e4 JQ |
2637 | return 0; |
2638 | } | |
2639 | ||
2640 | static int ram_load_cleanup(void *opaque) | |
2641 | { | |
f9494614 | 2642 | RAMBlock *rb; |
f265e0e4 | 2643 | xbzrle_load_cleanup(); |
f0afa331 | 2644 | compress_threads_load_cleanup(); |
f9494614 AP |
2645 | |
2646 | RAMBLOCK_FOREACH(rb) { | |
2647 | g_free(rb->receivedmap); | |
2648 | rb->receivedmap = NULL; | |
2649 | } | |
f265e0e4 JQ |
2650 | return 0; |
2651 | } | |
2652 | ||
3d0684b2 JQ |
2653 | /** |
2654 | * ram_postcopy_incoming_init: allocate postcopy data structures | |
2655 | * | |
2656 | * Returns 0 for success and negative if there was one error | |
2657 | * | |
2658 | * @mis: current migration incoming state | |
2659 | * | |
2660 | * Allocate data structures etc needed by incoming migration with | |
2661 | * postcopy-ram. postcopy-ram's similarly names | |
2662 | * postcopy_ram_incoming_init does the work. | |
1caddf8a DDAG |
2663 | */ |
2664 | int ram_postcopy_incoming_init(MigrationIncomingState *mis) | |
2665 | { | |
b8c48993 | 2666 | unsigned long ram_pages = last_ram_page(); |
1caddf8a DDAG |
2667 | |
2668 | return postcopy_ram_incoming_init(mis, ram_pages); | |
2669 | } | |
2670 | ||
3d0684b2 JQ |
2671 | /** |
2672 | * ram_load_postcopy: load a page in postcopy case | |
2673 | * | |
2674 | * Returns 0 for success or -errno in case of error | |
2675 | * | |
a7180877 DDAG |
2676 | * Called in postcopy mode by ram_load(). |
2677 | * rcu_read_lock is taken prior to this being called. | |
3d0684b2 JQ |
2678 | * |
2679 | * @f: QEMUFile where to send the data | |
a7180877 DDAG |
2680 | */ |
2681 | static int ram_load_postcopy(QEMUFile *f) | |
2682 | { | |
2683 | int flags = 0, ret = 0; | |
2684 | bool place_needed = false; | |
28abd200 | 2685 | bool matching_page_sizes = false; |
a7180877 DDAG |
2686 | MigrationIncomingState *mis = migration_incoming_get_current(); |
2687 | /* Temporary page that is later 'placed' */ | |
2688 | void *postcopy_host_page = postcopy_get_tmp_page(mis); | |
c53b7ddc | 2689 | void *last_host = NULL; |
a3b6ff6d | 2690 | bool all_zero = false; |
a7180877 DDAG |
2691 | |
2692 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
2693 | ram_addr_t addr; | |
2694 | void *host = NULL; | |
2695 | void *page_buffer = NULL; | |
2696 | void *place_source = NULL; | |
df9ff5e1 | 2697 | RAMBlock *block = NULL; |
a7180877 | 2698 | uint8_t ch; |
a7180877 DDAG |
2699 | |
2700 | addr = qemu_get_be64(f); | |
2701 | flags = addr & ~TARGET_PAGE_MASK; | |
2702 | addr &= TARGET_PAGE_MASK; | |
2703 | ||
2704 | trace_ram_load_postcopy_loop((uint64_t)addr, flags); | |
2705 | place_needed = false; | |
bb890ed5 | 2706 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE)) { |
df9ff5e1 | 2707 | block = ram_block_from_stream(f, flags); |
4c4bad48 HZ |
2708 | |
2709 | host = host_from_ram_block_offset(block, addr); | |
a7180877 DDAG |
2710 | if (!host) { |
2711 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
2712 | ret = -EINVAL; | |
2713 | break; | |
2714 | } | |
28abd200 | 2715 | matching_page_sizes = block->page_size == TARGET_PAGE_SIZE; |
a7180877 | 2716 | /* |
28abd200 DDAG |
2717 | * Postcopy requires that we place whole host pages atomically; |
2718 | * these may be huge pages for RAMBlocks that are backed by | |
2719 | * hugetlbfs. | |
a7180877 DDAG |
2720 | * To make it atomic, the data is read into a temporary page |
2721 | * that's moved into place later. | |
2722 | * The migration protocol uses, possibly smaller, target-pages | |
2723 | * however the source ensures it always sends all the components | |
2724 | * of a host page in order. | |
2725 | */ | |
2726 | page_buffer = postcopy_host_page + | |
28abd200 | 2727 | ((uintptr_t)host & (block->page_size - 1)); |
a7180877 | 2728 | /* If all TP are zero then we can optimise the place */ |
28abd200 | 2729 | if (!((uintptr_t)host & (block->page_size - 1))) { |
a7180877 | 2730 | all_zero = true; |
c53b7ddc DDAG |
2731 | } else { |
2732 | /* not the 1st TP within the HP */ | |
2733 | if (host != (last_host + TARGET_PAGE_SIZE)) { | |
9af9e0fe | 2734 | error_report("Non-sequential target page %p/%p", |
c53b7ddc DDAG |
2735 | host, last_host); |
2736 | ret = -EINVAL; | |
2737 | break; | |
2738 | } | |
a7180877 DDAG |
2739 | } |
2740 | ||
c53b7ddc | 2741 | |
a7180877 DDAG |
2742 | /* |
2743 | * If it's the last part of a host page then we place the host | |
2744 | * page | |
2745 | */ | |
2746 | place_needed = (((uintptr_t)host + TARGET_PAGE_SIZE) & | |
28abd200 | 2747 | (block->page_size - 1)) == 0; |
a7180877 DDAG |
2748 | place_source = postcopy_host_page; |
2749 | } | |
c53b7ddc | 2750 | last_host = host; |
a7180877 DDAG |
2751 | |
2752 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { | |
bb890ed5 | 2753 | case RAM_SAVE_FLAG_ZERO: |
a7180877 DDAG |
2754 | ch = qemu_get_byte(f); |
2755 | memset(page_buffer, ch, TARGET_PAGE_SIZE); | |
2756 | if (ch) { | |
2757 | all_zero = false; | |
2758 | } | |
2759 | break; | |
2760 | ||
2761 | case RAM_SAVE_FLAG_PAGE: | |
2762 | all_zero = false; | |
2763 | if (!place_needed || !matching_page_sizes) { | |
2764 | qemu_get_buffer(f, page_buffer, TARGET_PAGE_SIZE); | |
2765 | } else { | |
2766 | /* Avoids the qemu_file copy during postcopy, which is | |
2767 | * going to do a copy later; can only do it when we | |
2768 | * do this read in one go (matching page sizes) | |
2769 | */ | |
2770 | qemu_get_buffer_in_place(f, (uint8_t **)&place_source, | |
2771 | TARGET_PAGE_SIZE); | |
2772 | } | |
2773 | break; | |
2774 | case RAM_SAVE_FLAG_EOS: | |
2775 | /* normal exit */ | |
2776 | break; | |
2777 | default: | |
2778 | error_report("Unknown combination of migration flags: %#x" | |
2779 | " (postcopy mode)", flags); | |
2780 | ret = -EINVAL; | |
2781 | } | |
2782 | ||
2783 | if (place_needed) { | |
2784 | /* This gets called at the last target page in the host page */ | |
df9ff5e1 DDAG |
2785 | void *place_dest = host + TARGET_PAGE_SIZE - block->page_size; |
2786 | ||
a7180877 | 2787 | if (all_zero) { |
df9ff5e1 | 2788 | ret = postcopy_place_page_zero(mis, place_dest, |
8be4620b | 2789 | block); |
a7180877 | 2790 | } else { |
df9ff5e1 | 2791 | ret = postcopy_place_page(mis, place_dest, |
8be4620b | 2792 | place_source, block); |
a7180877 DDAG |
2793 | } |
2794 | } | |
2795 | if (!ret) { | |
2796 | ret = qemu_file_get_error(f); | |
2797 | } | |
2798 | } | |
2799 | ||
2800 | return ret; | |
2801 | } | |
2802 | ||
acab30b8 DHB |
2803 | static bool postcopy_is_advised(void) |
2804 | { | |
2805 | PostcopyState ps = postcopy_state_get(); | |
2806 | return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END; | |
2807 | } | |
2808 | ||
2809 | static bool postcopy_is_running(void) | |
2810 | { | |
2811 | PostcopyState ps = postcopy_state_get(); | |
2812 | return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; | |
2813 | } | |
2814 | ||
56e93d26 JQ |
2815 | static int ram_load(QEMUFile *f, void *opaque, int version_id) |
2816 | { | |
edc60127 | 2817 | int flags = 0, ret = 0, invalid_flags = 0; |
56e93d26 JQ |
2818 | static uint64_t seq_iter; |
2819 | int len = 0; | |
a7180877 DDAG |
2820 | /* |
2821 | * If system is running in postcopy mode, page inserts to host memory must | |
2822 | * be atomic | |
2823 | */ | |
acab30b8 | 2824 | bool postcopy_running = postcopy_is_running(); |
ef08fb38 | 2825 | /* ADVISE is earlier, it shows the source has the postcopy capability on */ |
acab30b8 | 2826 | bool postcopy_advised = postcopy_is_advised(); |
56e93d26 JQ |
2827 | |
2828 | seq_iter++; | |
2829 | ||
2830 | if (version_id != 4) { | |
2831 | ret = -EINVAL; | |
2832 | } | |
2833 | ||
edc60127 JQ |
2834 | if (!migrate_use_compression()) { |
2835 | invalid_flags |= RAM_SAVE_FLAG_COMPRESS_PAGE; | |
2836 | } | |
56e93d26 JQ |
2837 | /* This RCU critical section can be very long running. |
2838 | * When RCU reclaims in the code start to become numerous, | |
2839 | * it will be necessary to reduce the granularity of this | |
2840 | * critical section. | |
2841 | */ | |
2842 | rcu_read_lock(); | |
a7180877 DDAG |
2843 | |
2844 | if (postcopy_running) { | |
2845 | ret = ram_load_postcopy(f); | |
2846 | } | |
2847 | ||
2848 | while (!postcopy_running && !ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
56e93d26 | 2849 | ram_addr_t addr, total_ram_bytes; |
a776aa15 | 2850 | void *host = NULL; |
56e93d26 JQ |
2851 | uint8_t ch; |
2852 | ||
2853 | addr = qemu_get_be64(f); | |
2854 | flags = addr & ~TARGET_PAGE_MASK; | |
2855 | addr &= TARGET_PAGE_MASK; | |
2856 | ||
edc60127 JQ |
2857 | if (flags & invalid_flags) { |
2858 | if (flags & invalid_flags & RAM_SAVE_FLAG_COMPRESS_PAGE) { | |
2859 | error_report("Received an unexpected compressed page"); | |
2860 | } | |
2861 | ||
2862 | ret = -EINVAL; | |
2863 | break; | |
2864 | } | |
2865 | ||
bb890ed5 | 2866 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
a776aa15 | 2867 | RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) { |
4c4bad48 HZ |
2868 | RAMBlock *block = ram_block_from_stream(f, flags); |
2869 | ||
2870 | host = host_from_ram_block_offset(block, addr); | |
a776aa15 DDAG |
2871 | if (!host) { |
2872 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
2873 | ret = -EINVAL; | |
2874 | break; | |
2875 | } | |
f9494614 | 2876 | ramblock_recv_bitmap_set(block, host); |
1db9d8e5 | 2877 | trace_ram_load_loop(block->idstr, (uint64_t)addr, flags, host); |
a776aa15 DDAG |
2878 | } |
2879 | ||
56e93d26 JQ |
2880 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { |
2881 | case RAM_SAVE_FLAG_MEM_SIZE: | |
2882 | /* Synchronize RAM block list */ | |
2883 | total_ram_bytes = addr; | |
2884 | while (!ret && total_ram_bytes) { | |
2885 | RAMBlock *block; | |
56e93d26 JQ |
2886 | char id[256]; |
2887 | ram_addr_t length; | |
2888 | ||
2889 | len = qemu_get_byte(f); | |
2890 | qemu_get_buffer(f, (uint8_t *)id, len); | |
2891 | id[len] = 0; | |
2892 | length = qemu_get_be64(f); | |
2893 | ||
e3dd7493 DDAG |
2894 | block = qemu_ram_block_by_name(id); |
2895 | if (block) { | |
2896 | if (length != block->used_length) { | |
2897 | Error *local_err = NULL; | |
56e93d26 | 2898 | |
fa53a0e5 | 2899 | ret = qemu_ram_resize(block, length, |
e3dd7493 DDAG |
2900 | &local_err); |
2901 | if (local_err) { | |
2902 | error_report_err(local_err); | |
56e93d26 | 2903 | } |
56e93d26 | 2904 | } |
ef08fb38 DDAG |
2905 | /* For postcopy we need to check hugepage sizes match */ |
2906 | if (postcopy_advised && | |
2907 | block->page_size != qemu_host_page_size) { | |
2908 | uint64_t remote_page_size = qemu_get_be64(f); | |
2909 | if (remote_page_size != block->page_size) { | |
2910 | error_report("Mismatched RAM page size %s " | |
2911 | "(local) %zd != %" PRId64, | |
2912 | id, block->page_size, | |
2913 | remote_page_size); | |
2914 | ret = -EINVAL; | |
2915 | } | |
2916 | } | |
e3dd7493 DDAG |
2917 | ram_control_load_hook(f, RAM_CONTROL_BLOCK_REG, |
2918 | block->idstr); | |
2919 | } else { | |
56e93d26 JQ |
2920 | error_report("Unknown ramblock \"%s\", cannot " |
2921 | "accept migration", id); | |
2922 | ret = -EINVAL; | |
2923 | } | |
2924 | ||
2925 | total_ram_bytes -= length; | |
2926 | } | |
2927 | break; | |
a776aa15 | 2928 | |
bb890ed5 | 2929 | case RAM_SAVE_FLAG_ZERO: |
56e93d26 JQ |
2930 | ch = qemu_get_byte(f); |
2931 | ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); | |
2932 | break; | |
a776aa15 | 2933 | |
56e93d26 | 2934 | case RAM_SAVE_FLAG_PAGE: |
56e93d26 JQ |
2935 | qemu_get_buffer(f, host, TARGET_PAGE_SIZE); |
2936 | break; | |
56e93d26 | 2937 | |
a776aa15 | 2938 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
56e93d26 JQ |
2939 | len = qemu_get_be32(f); |
2940 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
2941 | error_report("Invalid compressed data length: %d", len); | |
2942 | ret = -EINVAL; | |
2943 | break; | |
2944 | } | |
c1bc6626 | 2945 | decompress_data_with_multi_threads(f, host, len); |
56e93d26 | 2946 | break; |
a776aa15 | 2947 | |
56e93d26 | 2948 | case RAM_SAVE_FLAG_XBZRLE: |
56e93d26 JQ |
2949 | if (load_xbzrle(f, addr, host) < 0) { |
2950 | error_report("Failed to decompress XBZRLE page at " | |
2951 | RAM_ADDR_FMT, addr); | |
2952 | ret = -EINVAL; | |
2953 | break; | |
2954 | } | |
2955 | break; | |
2956 | case RAM_SAVE_FLAG_EOS: | |
2957 | /* normal exit */ | |
2958 | break; | |
2959 | default: | |
2960 | if (flags & RAM_SAVE_FLAG_HOOK) { | |
632e3a5c | 2961 | ram_control_load_hook(f, RAM_CONTROL_HOOK, NULL); |
56e93d26 JQ |
2962 | } else { |
2963 | error_report("Unknown combination of migration flags: %#x", | |
2964 | flags); | |
2965 | ret = -EINVAL; | |
2966 | } | |
2967 | } | |
2968 | if (!ret) { | |
2969 | ret = qemu_file_get_error(f); | |
2970 | } | |
2971 | } | |
2972 | ||
5533b2e9 | 2973 | wait_for_decompress_done(); |
56e93d26 | 2974 | rcu_read_unlock(); |
55c4446b | 2975 | trace_ram_load_complete(ret, seq_iter); |
56e93d26 JQ |
2976 | return ret; |
2977 | } | |
2978 | ||
c6467627 VSO |
2979 | static bool ram_has_postcopy(void *opaque) |
2980 | { | |
2981 | return migrate_postcopy_ram(); | |
2982 | } | |
2983 | ||
56e93d26 | 2984 | static SaveVMHandlers savevm_ram_handlers = { |
9907e842 | 2985 | .save_setup = ram_save_setup, |
56e93d26 | 2986 | .save_live_iterate = ram_save_iterate, |
763c906b | 2987 | .save_live_complete_postcopy = ram_save_complete, |
a3e06c3d | 2988 | .save_live_complete_precopy = ram_save_complete, |
c6467627 | 2989 | .has_postcopy = ram_has_postcopy, |
56e93d26 JQ |
2990 | .save_live_pending = ram_save_pending, |
2991 | .load_state = ram_load, | |
f265e0e4 JQ |
2992 | .save_cleanup = ram_save_cleanup, |
2993 | .load_setup = ram_load_setup, | |
2994 | .load_cleanup = ram_load_cleanup, | |
56e93d26 JQ |
2995 | }; |
2996 | ||
2997 | void ram_mig_init(void) | |
2998 | { | |
2999 | qemu_mutex_init(&XBZRLE.lock); | |
6f37bb8b | 3000 | register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, &ram_state); |
56e93d26 | 3001 | } |