]>
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> |
f348b6d1 | 32 | #include "qemu/cutils.h" |
56e93d26 JQ |
33 | #include "qemu/bitops.h" |
34 | #include "qemu/bitmap.h" | |
7205c9ec | 35 | #include "qemu/main-loop.h" |
709e3fe8 | 36 | #include "xbzrle.h" |
7b1e1a22 | 37 | #include "ram.h" |
6666c96a | 38 | #include "migration.h" |
f2a8f0a6 | 39 | #include "migration/register.h" |
7b1e1a22 | 40 | #include "migration/misc.h" |
08a0aee1 | 41 | #include "qemu-file.h" |
be07b0ac | 42 | #include "postcopy-ram.h" |
56e93d26 | 43 | #include "migration/page_cache.h" |
56e93d26 | 44 | #include "qemu/error-report.h" |
e688df6b | 45 | #include "qapi/error.h" |
9af23989 | 46 | #include "qapi/qapi-events-migration.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 | ||
1cba9f6e DDAG |
172 | bool ramblock_recv_bitmap_test_byte_offset(RAMBlock *rb, uint64_t byte_offset) |
173 | { | |
174 | return test_bit(byte_offset >> TARGET_PAGE_BITS, rb->receivedmap); | |
175 | } | |
176 | ||
f9494614 AP |
177 | void ramblock_recv_bitmap_set(RAMBlock *rb, void *host_addr) |
178 | { | |
179 | set_bit_atomic(ramblock_recv_bitmap_offset(host_addr, rb), rb->receivedmap); | |
180 | } | |
181 | ||
182 | void ramblock_recv_bitmap_set_range(RAMBlock *rb, void *host_addr, | |
183 | size_t nr) | |
184 | { | |
185 | bitmap_set_atomic(rb->receivedmap, | |
186 | ramblock_recv_bitmap_offset(host_addr, rb), | |
187 | nr); | |
188 | } | |
189 | ||
ec481c6c JQ |
190 | /* |
191 | * An outstanding page request, on the source, having been received | |
192 | * and queued | |
193 | */ | |
194 | struct RAMSrcPageRequest { | |
195 | RAMBlock *rb; | |
196 | hwaddr offset; | |
197 | hwaddr len; | |
198 | ||
199 | QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; | |
200 | }; | |
201 | ||
6f37bb8b JQ |
202 | /* State of RAM for migration */ |
203 | struct RAMState { | |
204b88b8 JQ |
204 | /* QEMUFile used for this migration */ |
205 | QEMUFile *f; | |
6f37bb8b JQ |
206 | /* Last block that we have visited searching for dirty pages */ |
207 | RAMBlock *last_seen_block; | |
208 | /* Last block from where we have sent data */ | |
209 | RAMBlock *last_sent_block; | |
269ace29 JQ |
210 | /* Last dirty target page we have sent */ |
211 | ram_addr_t last_page; | |
6f37bb8b JQ |
212 | /* last ram version we have seen */ |
213 | uint32_t last_version; | |
214 | /* We are in the first round */ | |
215 | bool ram_bulk_stage; | |
8d820d6f JQ |
216 | /* How many times we have dirty too many pages */ |
217 | int dirty_rate_high_cnt; | |
f664da80 JQ |
218 | /* these variables are used for bitmap sync */ |
219 | /* last time we did a full bitmap_sync */ | |
220 | int64_t time_last_bitmap_sync; | |
eac74159 | 221 | /* bytes transferred at start_time */ |
c4bdf0cf | 222 | uint64_t bytes_xfer_prev; |
a66cd90c | 223 | /* number of dirty pages since start_time */ |
68908ed6 | 224 | uint64_t num_dirty_pages_period; |
b5833fde JQ |
225 | /* xbzrle misses since the beginning of the period */ |
226 | uint64_t xbzrle_cache_miss_prev; | |
36040d9c JQ |
227 | /* number of iterations at the beginning of period */ |
228 | uint64_t iterations_prev; | |
23b28c3c JQ |
229 | /* Iterations since start */ |
230 | uint64_t iterations; | |
9360447d | 231 | /* number of dirty bits in the bitmap */ |
2dfaf12e PX |
232 | uint64_t migration_dirty_pages; |
233 | /* protects modification of the bitmap */ | |
108cfae0 | 234 | QemuMutex bitmap_mutex; |
68a098f3 JQ |
235 | /* The RAMBlock used in the last src_page_requests */ |
236 | RAMBlock *last_req_rb; | |
ec481c6c JQ |
237 | /* Queue of outstanding page requests from the destination */ |
238 | QemuMutex src_page_req_mutex; | |
239 | QSIMPLEQ_HEAD(src_page_requests, RAMSrcPageRequest) src_page_requests; | |
6f37bb8b JQ |
240 | }; |
241 | typedef struct RAMState RAMState; | |
242 | ||
53518d94 | 243 | static RAMState *ram_state; |
6f37bb8b | 244 | |
9edabd4d | 245 | uint64_t ram_bytes_remaining(void) |
2f4fde93 | 246 | { |
bae416e5 DDAG |
247 | return ram_state ? (ram_state->migration_dirty_pages * TARGET_PAGE_SIZE) : |
248 | 0; | |
2f4fde93 JQ |
249 | } |
250 | ||
9360447d | 251 | MigrationStats ram_counters; |
96506894 | 252 | |
b8fb8cb7 DDAG |
253 | /* used by the search for pages to send */ |
254 | struct PageSearchStatus { | |
255 | /* Current block being searched */ | |
256 | RAMBlock *block; | |
a935e30f JQ |
257 | /* Current page to search from */ |
258 | unsigned long page; | |
b8fb8cb7 DDAG |
259 | /* Set once we wrap around */ |
260 | bool complete_round; | |
261 | }; | |
262 | typedef struct PageSearchStatus PageSearchStatus; | |
263 | ||
56e93d26 | 264 | struct CompressParam { |
56e93d26 | 265 | bool done; |
90e56fb4 | 266 | bool quit; |
56e93d26 JQ |
267 | QEMUFile *file; |
268 | QemuMutex mutex; | |
269 | QemuCond cond; | |
270 | RAMBlock *block; | |
271 | ram_addr_t offset; | |
dcaf446e | 272 | z_stream stream; |
56e93d26 JQ |
273 | }; |
274 | typedef struct CompressParam CompressParam; | |
275 | ||
276 | struct DecompressParam { | |
73a8912b | 277 | bool done; |
90e56fb4 | 278 | bool quit; |
56e93d26 JQ |
279 | QemuMutex mutex; |
280 | QemuCond cond; | |
281 | void *des; | |
d341d9f3 | 282 | uint8_t *compbuf; |
56e93d26 | 283 | int len; |
797ca154 | 284 | z_stream stream; |
56e93d26 JQ |
285 | }; |
286 | typedef struct DecompressParam DecompressParam; | |
287 | ||
288 | static CompressParam *comp_param; | |
289 | static QemuThread *compress_threads; | |
290 | /* comp_done_cond is used to wake up the migration thread when | |
291 | * one of the compression threads has finished the compression. | |
292 | * comp_done_lock is used to co-work with comp_done_cond. | |
293 | */ | |
0d9f9a5c LL |
294 | static QemuMutex comp_done_lock; |
295 | static QemuCond comp_done_cond; | |
56e93d26 JQ |
296 | /* The empty QEMUFileOps will be used by file in CompressParam */ |
297 | static const QEMUFileOps empty_ops = { }; | |
298 | ||
56e93d26 JQ |
299 | static DecompressParam *decomp_param; |
300 | static QemuThread *decompress_threads; | |
73a8912b LL |
301 | static QemuMutex decomp_done_lock; |
302 | static QemuCond decomp_done_cond; | |
56e93d26 | 303 | |
dcaf446e | 304 | static int do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
a7a9a88f | 305 | ram_addr_t offset); |
56e93d26 JQ |
306 | |
307 | static void *do_data_compress(void *opaque) | |
308 | { | |
309 | CompressParam *param = opaque; | |
a7a9a88f LL |
310 | RAMBlock *block; |
311 | ram_addr_t offset; | |
56e93d26 | 312 | |
a7a9a88f | 313 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 314 | while (!param->quit) { |
a7a9a88f LL |
315 | if (param->block) { |
316 | block = param->block; | |
317 | offset = param->offset; | |
318 | param->block = NULL; | |
319 | qemu_mutex_unlock(¶m->mutex); | |
320 | ||
dcaf446e | 321 | do_compress_ram_page(param->file, ¶m->stream, block, offset); |
a7a9a88f | 322 | |
0d9f9a5c | 323 | qemu_mutex_lock(&comp_done_lock); |
a7a9a88f | 324 | param->done = true; |
0d9f9a5c LL |
325 | qemu_cond_signal(&comp_done_cond); |
326 | qemu_mutex_unlock(&comp_done_lock); | |
a7a9a88f LL |
327 | |
328 | qemu_mutex_lock(¶m->mutex); | |
329 | } else { | |
56e93d26 JQ |
330 | qemu_cond_wait(¶m->cond, ¶m->mutex); |
331 | } | |
56e93d26 | 332 | } |
a7a9a88f | 333 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
334 | |
335 | return NULL; | |
336 | } | |
337 | ||
338 | static inline void terminate_compression_threads(void) | |
339 | { | |
340 | int idx, thread_count; | |
341 | ||
342 | thread_count = migrate_compress_threads(); | |
3d0684b2 | 343 | |
56e93d26 JQ |
344 | for (idx = 0; idx < thread_count; idx++) { |
345 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 346 | comp_param[idx].quit = true; |
56e93d26 JQ |
347 | qemu_cond_signal(&comp_param[idx].cond); |
348 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
349 | } | |
350 | } | |
351 | ||
f0afa331 | 352 | static void compress_threads_save_cleanup(void) |
56e93d26 JQ |
353 | { |
354 | int i, thread_count; | |
355 | ||
356 | if (!migrate_use_compression()) { | |
357 | return; | |
358 | } | |
359 | terminate_compression_threads(); | |
360 | thread_count = migrate_compress_threads(); | |
361 | for (i = 0; i < thread_count; i++) { | |
dcaf446e XG |
362 | /* |
363 | * we use it as a indicator which shows if the thread is | |
364 | * properly init'd or not | |
365 | */ | |
366 | if (!comp_param[i].file) { | |
367 | break; | |
368 | } | |
56e93d26 | 369 | qemu_thread_join(compress_threads + i); |
56e93d26 JQ |
370 | qemu_mutex_destroy(&comp_param[i].mutex); |
371 | qemu_cond_destroy(&comp_param[i].cond); | |
dcaf446e XG |
372 | deflateEnd(&comp_param[i].stream); |
373 | qemu_fclose(comp_param[i].file); | |
374 | comp_param[i].file = NULL; | |
56e93d26 | 375 | } |
0d9f9a5c LL |
376 | qemu_mutex_destroy(&comp_done_lock); |
377 | qemu_cond_destroy(&comp_done_cond); | |
56e93d26 JQ |
378 | g_free(compress_threads); |
379 | g_free(comp_param); | |
56e93d26 JQ |
380 | compress_threads = NULL; |
381 | comp_param = NULL; | |
56e93d26 JQ |
382 | } |
383 | ||
dcaf446e | 384 | static int compress_threads_save_setup(void) |
56e93d26 JQ |
385 | { |
386 | int i, thread_count; | |
387 | ||
388 | if (!migrate_use_compression()) { | |
dcaf446e | 389 | return 0; |
56e93d26 | 390 | } |
56e93d26 JQ |
391 | thread_count = migrate_compress_threads(); |
392 | compress_threads = g_new0(QemuThread, thread_count); | |
393 | comp_param = g_new0(CompressParam, thread_count); | |
0d9f9a5c LL |
394 | qemu_cond_init(&comp_done_cond); |
395 | qemu_mutex_init(&comp_done_lock); | |
56e93d26 | 396 | for (i = 0; i < thread_count; i++) { |
dcaf446e XG |
397 | if (deflateInit(&comp_param[i].stream, |
398 | migrate_compress_level()) != Z_OK) { | |
399 | goto exit; | |
400 | } | |
401 | ||
e110aa91 C |
402 | /* comp_param[i].file is just used as a dummy buffer to save data, |
403 | * set its ops to empty. | |
56e93d26 JQ |
404 | */ |
405 | comp_param[i].file = qemu_fopen_ops(NULL, &empty_ops); | |
406 | comp_param[i].done = true; | |
90e56fb4 | 407 | comp_param[i].quit = false; |
56e93d26 JQ |
408 | qemu_mutex_init(&comp_param[i].mutex); |
409 | qemu_cond_init(&comp_param[i].cond); | |
410 | qemu_thread_create(compress_threads + i, "compress", | |
411 | do_data_compress, comp_param + i, | |
412 | QEMU_THREAD_JOINABLE); | |
413 | } | |
dcaf446e XG |
414 | return 0; |
415 | ||
416 | exit: | |
417 | compress_threads_save_cleanup(); | |
418 | return -1; | |
56e93d26 JQ |
419 | } |
420 | ||
f986c3d2 JQ |
421 | /* Multiple fd's */ |
422 | ||
423 | struct MultiFDSendParams { | |
424 | uint8_t id; | |
425 | char *name; | |
426 | QemuThread thread; | |
427 | QemuSemaphore sem; | |
428 | QemuMutex mutex; | |
429 | bool quit; | |
430 | }; | |
431 | typedef struct MultiFDSendParams MultiFDSendParams; | |
432 | ||
433 | struct { | |
434 | MultiFDSendParams *params; | |
435 | /* number of created threads */ | |
436 | int count; | |
437 | } *multifd_send_state; | |
438 | ||
439 | static void terminate_multifd_send_threads(Error *errp) | |
440 | { | |
441 | int i; | |
442 | ||
443 | for (i = 0; i < multifd_send_state->count; i++) { | |
444 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
445 | ||
446 | qemu_mutex_lock(&p->mutex); | |
447 | p->quit = true; | |
448 | qemu_sem_post(&p->sem); | |
449 | qemu_mutex_unlock(&p->mutex); | |
450 | } | |
451 | } | |
452 | ||
453 | int multifd_save_cleanup(Error **errp) | |
454 | { | |
455 | int i; | |
456 | int ret = 0; | |
457 | ||
458 | if (!migrate_use_multifd()) { | |
459 | return 0; | |
460 | } | |
461 | terminate_multifd_send_threads(NULL); | |
462 | for (i = 0; i < multifd_send_state->count; i++) { | |
463 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
464 | ||
465 | qemu_thread_join(&p->thread); | |
466 | qemu_mutex_destroy(&p->mutex); | |
467 | qemu_sem_destroy(&p->sem); | |
468 | g_free(p->name); | |
469 | p->name = NULL; | |
470 | } | |
471 | g_free(multifd_send_state->params); | |
472 | multifd_send_state->params = NULL; | |
473 | g_free(multifd_send_state); | |
474 | multifd_send_state = NULL; | |
475 | return ret; | |
476 | } | |
477 | ||
478 | static void *multifd_send_thread(void *opaque) | |
479 | { | |
480 | MultiFDSendParams *p = opaque; | |
481 | ||
482 | while (true) { | |
483 | qemu_mutex_lock(&p->mutex); | |
484 | if (p->quit) { | |
485 | qemu_mutex_unlock(&p->mutex); | |
486 | break; | |
487 | } | |
488 | qemu_mutex_unlock(&p->mutex); | |
489 | qemu_sem_wait(&p->sem); | |
490 | } | |
491 | ||
492 | return NULL; | |
493 | } | |
494 | ||
495 | int multifd_save_setup(void) | |
496 | { | |
497 | int thread_count; | |
498 | uint8_t i; | |
499 | ||
500 | if (!migrate_use_multifd()) { | |
501 | return 0; | |
502 | } | |
503 | thread_count = migrate_multifd_channels(); | |
504 | multifd_send_state = g_malloc0(sizeof(*multifd_send_state)); | |
505 | multifd_send_state->params = g_new0(MultiFDSendParams, thread_count); | |
506 | multifd_send_state->count = 0; | |
507 | for (i = 0; i < thread_count; i++) { | |
508 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
509 | ||
510 | qemu_mutex_init(&p->mutex); | |
511 | qemu_sem_init(&p->sem, 0); | |
512 | p->quit = false; | |
513 | p->id = i; | |
514 | p->name = g_strdup_printf("multifdsend_%d", i); | |
515 | qemu_thread_create(&p->thread, p->name, multifd_send_thread, p, | |
516 | QEMU_THREAD_JOINABLE); | |
517 | ||
518 | multifd_send_state->count++; | |
519 | } | |
520 | return 0; | |
521 | } | |
522 | ||
523 | struct MultiFDRecvParams { | |
524 | uint8_t id; | |
525 | char *name; | |
526 | QemuThread thread; | |
527 | QemuSemaphore sem; | |
528 | QemuMutex mutex; | |
529 | bool quit; | |
530 | }; | |
531 | typedef struct MultiFDRecvParams MultiFDRecvParams; | |
532 | ||
533 | struct { | |
534 | MultiFDRecvParams *params; | |
535 | /* number of created threads */ | |
536 | int count; | |
537 | } *multifd_recv_state; | |
538 | ||
539 | static void terminate_multifd_recv_threads(Error *errp) | |
540 | { | |
541 | int i; | |
542 | ||
543 | for (i = 0; i < multifd_recv_state->count; i++) { | |
544 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
545 | ||
546 | qemu_mutex_lock(&p->mutex); | |
547 | p->quit = true; | |
548 | qemu_sem_post(&p->sem); | |
549 | qemu_mutex_unlock(&p->mutex); | |
550 | } | |
551 | } | |
552 | ||
553 | int multifd_load_cleanup(Error **errp) | |
554 | { | |
555 | int i; | |
556 | int ret = 0; | |
557 | ||
558 | if (!migrate_use_multifd()) { | |
559 | return 0; | |
560 | } | |
561 | terminate_multifd_recv_threads(NULL); | |
562 | for (i = 0; i < multifd_recv_state->count; i++) { | |
563 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
564 | ||
565 | qemu_thread_join(&p->thread); | |
566 | qemu_mutex_destroy(&p->mutex); | |
567 | qemu_sem_destroy(&p->sem); | |
568 | g_free(p->name); | |
569 | p->name = NULL; | |
570 | } | |
571 | g_free(multifd_recv_state->params); | |
572 | multifd_recv_state->params = NULL; | |
573 | g_free(multifd_recv_state); | |
574 | multifd_recv_state = NULL; | |
575 | ||
576 | return ret; | |
577 | } | |
578 | ||
579 | static void *multifd_recv_thread(void *opaque) | |
580 | { | |
581 | MultiFDRecvParams *p = opaque; | |
582 | ||
583 | while (true) { | |
584 | qemu_mutex_lock(&p->mutex); | |
585 | if (p->quit) { | |
586 | qemu_mutex_unlock(&p->mutex); | |
587 | break; | |
588 | } | |
589 | qemu_mutex_unlock(&p->mutex); | |
590 | qemu_sem_wait(&p->sem); | |
591 | } | |
592 | ||
593 | return NULL; | |
594 | } | |
595 | ||
596 | int multifd_load_setup(void) | |
597 | { | |
598 | int thread_count; | |
599 | uint8_t i; | |
600 | ||
601 | if (!migrate_use_multifd()) { | |
602 | return 0; | |
603 | } | |
604 | thread_count = migrate_multifd_channels(); | |
605 | multifd_recv_state = g_malloc0(sizeof(*multifd_recv_state)); | |
606 | multifd_recv_state->params = g_new0(MultiFDRecvParams, thread_count); | |
607 | multifd_recv_state->count = 0; | |
608 | for (i = 0; i < thread_count; i++) { | |
609 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
610 | ||
611 | qemu_mutex_init(&p->mutex); | |
612 | qemu_sem_init(&p->sem, 0); | |
613 | p->quit = false; | |
614 | p->id = i; | |
615 | p->name = g_strdup_printf("multifdrecv_%d", i); | |
616 | qemu_thread_create(&p->thread, p->name, multifd_recv_thread, p, | |
617 | QEMU_THREAD_JOINABLE); | |
618 | multifd_recv_state->count++; | |
619 | } | |
620 | return 0; | |
621 | } | |
622 | ||
56e93d26 | 623 | /** |
3d0684b2 | 624 | * save_page_header: write page header to wire |
56e93d26 JQ |
625 | * |
626 | * If this is the 1st block, it also writes the block identification | |
627 | * | |
3d0684b2 | 628 | * Returns the number of bytes written |
56e93d26 JQ |
629 | * |
630 | * @f: QEMUFile where to send the data | |
631 | * @block: block that contains the page we want to send | |
632 | * @offset: offset inside the block for the page | |
633 | * in the lower bits, it contains flags | |
634 | */ | |
2bf3aa85 JQ |
635 | static size_t save_page_header(RAMState *rs, QEMUFile *f, RAMBlock *block, |
636 | ram_addr_t offset) | |
56e93d26 | 637 | { |
9f5f380b | 638 | size_t size, len; |
56e93d26 | 639 | |
24795694 JQ |
640 | if (block == rs->last_sent_block) { |
641 | offset |= RAM_SAVE_FLAG_CONTINUE; | |
642 | } | |
2bf3aa85 | 643 | qemu_put_be64(f, offset); |
56e93d26 JQ |
644 | size = 8; |
645 | ||
646 | if (!(offset & RAM_SAVE_FLAG_CONTINUE)) { | |
9f5f380b | 647 | len = strlen(block->idstr); |
2bf3aa85 JQ |
648 | qemu_put_byte(f, len); |
649 | qemu_put_buffer(f, (uint8_t *)block->idstr, len); | |
9f5f380b | 650 | size += 1 + len; |
24795694 | 651 | rs->last_sent_block = block; |
56e93d26 JQ |
652 | } |
653 | return size; | |
654 | } | |
655 | ||
3d0684b2 JQ |
656 | /** |
657 | * mig_throttle_guest_down: throotle down the guest | |
658 | * | |
659 | * Reduce amount of guest cpu execution to hopefully slow down memory | |
660 | * writes. If guest dirty memory rate is reduced below the rate at | |
661 | * which we can transfer pages to the destination then we should be | |
662 | * able to complete migration. Some workloads dirty memory way too | |
663 | * fast and will not effectively converge, even with auto-converge. | |
070afca2 JH |
664 | */ |
665 | static void mig_throttle_guest_down(void) | |
666 | { | |
667 | MigrationState *s = migrate_get_current(); | |
2594f56d DB |
668 | uint64_t pct_initial = s->parameters.cpu_throttle_initial; |
669 | uint64_t pct_icrement = s->parameters.cpu_throttle_increment; | |
070afca2 JH |
670 | |
671 | /* We have not started throttling yet. Let's start it. */ | |
672 | if (!cpu_throttle_active()) { | |
673 | cpu_throttle_set(pct_initial); | |
674 | } else { | |
675 | /* Throttling already on, just increase the rate */ | |
676 | cpu_throttle_set(cpu_throttle_get_percentage() + pct_icrement); | |
677 | } | |
678 | } | |
679 | ||
3d0684b2 JQ |
680 | /** |
681 | * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache | |
682 | * | |
6f37bb8b | 683 | * @rs: current RAM state |
3d0684b2 JQ |
684 | * @current_addr: address for the zero page |
685 | * | |
686 | * Update the xbzrle cache to reflect a page that's been sent as all 0. | |
56e93d26 JQ |
687 | * The important thing is that a stale (not-yet-0'd) page be replaced |
688 | * by the new data. | |
689 | * As a bonus, if the page wasn't in the cache it gets added so that | |
3d0684b2 | 690 | * when a small write is made into the 0'd page it gets XBZRLE sent. |
56e93d26 | 691 | */ |
6f37bb8b | 692 | static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr) |
56e93d26 | 693 | { |
6f37bb8b | 694 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { |
56e93d26 JQ |
695 | return; |
696 | } | |
697 | ||
698 | /* We don't care if this fails to allocate a new cache page | |
699 | * as long as it updated an old one */ | |
c00e0928 | 700 | cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page, |
9360447d | 701 | ram_counters.dirty_sync_count); |
56e93d26 JQ |
702 | } |
703 | ||
704 | #define ENCODING_FLAG_XBZRLE 0x1 | |
705 | ||
706 | /** | |
707 | * save_xbzrle_page: compress and send current page | |
708 | * | |
709 | * Returns: 1 means that we wrote the page | |
710 | * 0 means that page is identical to the one already sent | |
711 | * -1 means that xbzrle would be longer than normal | |
712 | * | |
5a987738 | 713 | * @rs: current RAM state |
3d0684b2 JQ |
714 | * @current_data: pointer to the address of the page contents |
715 | * @current_addr: addr of the page | |
56e93d26 JQ |
716 | * @block: block that contains the page we want to send |
717 | * @offset: offset inside the block for the page | |
718 | * @last_stage: if we are at the completion stage | |
56e93d26 | 719 | */ |
204b88b8 | 720 | static int save_xbzrle_page(RAMState *rs, uint8_t **current_data, |
56e93d26 | 721 | ram_addr_t current_addr, RAMBlock *block, |
072c2511 | 722 | ram_addr_t offset, bool last_stage) |
56e93d26 JQ |
723 | { |
724 | int encoded_len = 0, bytes_xbzrle; | |
725 | uint8_t *prev_cached_page; | |
726 | ||
9360447d JQ |
727 | if (!cache_is_cached(XBZRLE.cache, current_addr, |
728 | ram_counters.dirty_sync_count)) { | |
729 | xbzrle_counters.cache_miss++; | |
56e93d26 JQ |
730 | if (!last_stage) { |
731 | if (cache_insert(XBZRLE.cache, current_addr, *current_data, | |
9360447d | 732 | ram_counters.dirty_sync_count) == -1) { |
56e93d26 JQ |
733 | return -1; |
734 | } else { | |
735 | /* update *current_data when the page has been | |
736 | inserted into cache */ | |
737 | *current_data = get_cached_data(XBZRLE.cache, current_addr); | |
738 | } | |
739 | } | |
740 | return -1; | |
741 | } | |
742 | ||
743 | prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); | |
744 | ||
745 | /* save current buffer into memory */ | |
746 | memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); | |
747 | ||
748 | /* XBZRLE encoding (if there is no overflow) */ | |
749 | encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, | |
750 | TARGET_PAGE_SIZE, XBZRLE.encoded_buf, | |
751 | TARGET_PAGE_SIZE); | |
752 | if (encoded_len == 0) { | |
55c4446b | 753 | trace_save_xbzrle_page_skipping(); |
56e93d26 JQ |
754 | return 0; |
755 | } else if (encoded_len == -1) { | |
55c4446b | 756 | trace_save_xbzrle_page_overflow(); |
9360447d | 757 | xbzrle_counters.overflow++; |
56e93d26 JQ |
758 | /* update data in the cache */ |
759 | if (!last_stage) { | |
760 | memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE); | |
761 | *current_data = prev_cached_page; | |
762 | } | |
763 | return -1; | |
764 | } | |
765 | ||
766 | /* we need to update the data in the cache, in order to get the same data */ | |
767 | if (!last_stage) { | |
768 | memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); | |
769 | } | |
770 | ||
771 | /* Send XBZRLE based compressed page */ | |
2bf3aa85 | 772 | bytes_xbzrle = save_page_header(rs, rs->f, block, |
204b88b8 JQ |
773 | offset | RAM_SAVE_FLAG_XBZRLE); |
774 | qemu_put_byte(rs->f, ENCODING_FLAG_XBZRLE); | |
775 | qemu_put_be16(rs->f, encoded_len); | |
776 | qemu_put_buffer(rs->f, XBZRLE.encoded_buf, encoded_len); | |
56e93d26 | 777 | bytes_xbzrle += encoded_len + 1 + 2; |
9360447d JQ |
778 | xbzrle_counters.pages++; |
779 | xbzrle_counters.bytes += bytes_xbzrle; | |
780 | ram_counters.transferred += bytes_xbzrle; | |
56e93d26 JQ |
781 | |
782 | return 1; | |
783 | } | |
784 | ||
3d0684b2 JQ |
785 | /** |
786 | * migration_bitmap_find_dirty: find the next dirty page from start | |
f3f491fc | 787 | * |
3d0684b2 JQ |
788 | * Called with rcu_read_lock() to protect migration_bitmap |
789 | * | |
790 | * Returns the byte offset within memory region of the start of a dirty page | |
791 | * | |
6f37bb8b | 792 | * @rs: current RAM state |
3d0684b2 | 793 | * @rb: RAMBlock where to search for dirty pages |
a935e30f | 794 | * @start: page where we start the search |
f3f491fc | 795 | */ |
56e93d26 | 796 | static inline |
a935e30f | 797 | unsigned long migration_bitmap_find_dirty(RAMState *rs, RAMBlock *rb, |
f20e2865 | 798 | unsigned long start) |
56e93d26 | 799 | { |
6b6712ef JQ |
800 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; |
801 | unsigned long *bitmap = rb->bmap; | |
56e93d26 JQ |
802 | unsigned long next; |
803 | ||
6b6712ef JQ |
804 | if (rs->ram_bulk_stage && start > 0) { |
805 | next = start + 1; | |
56e93d26 | 806 | } else { |
6b6712ef | 807 | next = find_next_bit(bitmap, size, start); |
56e93d26 JQ |
808 | } |
809 | ||
6b6712ef | 810 | return next; |
56e93d26 JQ |
811 | } |
812 | ||
06b10688 | 813 | static inline bool migration_bitmap_clear_dirty(RAMState *rs, |
f20e2865 JQ |
814 | RAMBlock *rb, |
815 | unsigned long page) | |
a82d593b DDAG |
816 | { |
817 | bool ret; | |
a82d593b | 818 | |
6b6712ef | 819 | ret = test_and_clear_bit(page, rb->bmap); |
a82d593b DDAG |
820 | |
821 | if (ret) { | |
0d8ec885 | 822 | rs->migration_dirty_pages--; |
a82d593b DDAG |
823 | } |
824 | return ret; | |
825 | } | |
826 | ||
15440dd5 JQ |
827 | static void migration_bitmap_sync_range(RAMState *rs, RAMBlock *rb, |
828 | ram_addr_t start, ram_addr_t length) | |
56e93d26 | 829 | { |
0d8ec885 | 830 | rs->migration_dirty_pages += |
6b6712ef | 831 | cpu_physical_memory_sync_dirty_bitmap(rb, start, length, |
0d8ec885 | 832 | &rs->num_dirty_pages_period); |
56e93d26 JQ |
833 | } |
834 | ||
3d0684b2 JQ |
835 | /** |
836 | * ram_pagesize_summary: calculate all the pagesizes of a VM | |
837 | * | |
838 | * Returns a summary bitmap of the page sizes of all RAMBlocks | |
839 | * | |
840 | * For VMs with just normal pages this is equivalent to the host page | |
841 | * size. If it's got some huge pages then it's the OR of all the | |
842 | * different page sizes. | |
e8ca1db2 DDAG |
843 | */ |
844 | uint64_t ram_pagesize_summary(void) | |
845 | { | |
846 | RAMBlock *block; | |
847 | uint64_t summary = 0; | |
848 | ||
99e15582 | 849 | RAMBLOCK_FOREACH(block) { |
e8ca1db2 DDAG |
850 | summary |= block->page_size; |
851 | } | |
852 | ||
853 | return summary; | |
854 | } | |
855 | ||
8d820d6f | 856 | static void migration_bitmap_sync(RAMState *rs) |
56e93d26 JQ |
857 | { |
858 | RAMBlock *block; | |
56e93d26 | 859 | int64_t end_time; |
c4bdf0cf | 860 | uint64_t bytes_xfer_now; |
56e93d26 | 861 | |
9360447d | 862 | ram_counters.dirty_sync_count++; |
56e93d26 | 863 | |
f664da80 JQ |
864 | if (!rs->time_last_bitmap_sync) { |
865 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56e93d26 JQ |
866 | } |
867 | ||
868 | trace_migration_bitmap_sync_start(); | |
9c1f8f44 | 869 | memory_global_dirty_log_sync(); |
56e93d26 | 870 | |
108cfae0 | 871 | qemu_mutex_lock(&rs->bitmap_mutex); |
56e93d26 | 872 | rcu_read_lock(); |
99e15582 | 873 | RAMBLOCK_FOREACH(block) { |
15440dd5 | 874 | migration_bitmap_sync_range(rs, block, 0, block->used_length); |
56e93d26 JQ |
875 | } |
876 | rcu_read_unlock(); | |
108cfae0 | 877 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 878 | |
a66cd90c | 879 | trace_migration_bitmap_sync_end(rs->num_dirty_pages_period); |
1ffb5dfd | 880 | |
56e93d26 JQ |
881 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
882 | ||
883 | /* more than 1 second = 1000 millisecons */ | |
f664da80 | 884 | if (end_time > rs->time_last_bitmap_sync + 1000) { |
d693c6f1 | 885 | /* calculate period counters */ |
9360447d | 886 | ram_counters.dirty_pages_rate = rs->num_dirty_pages_period * 1000 |
d693c6f1 | 887 | / (end_time - rs->time_last_bitmap_sync); |
9360447d | 888 | bytes_xfer_now = ram_counters.transferred; |
d693c6f1 | 889 | |
9ac78b61 PL |
890 | /* During block migration the auto-converge logic incorrectly detects |
891 | * that ram migration makes no progress. Avoid this by disabling the | |
892 | * throttling logic during the bulk phase of block migration. */ | |
893 | if (migrate_auto_converge() && !blk_mig_bulk_active()) { | |
56e93d26 JQ |
894 | /* The following detection logic can be refined later. For now: |
895 | Check to see if the dirtied bytes is 50% more than the approx. | |
896 | amount of bytes that just got transferred since the last time we | |
070afca2 JH |
897 | were in this routine. If that happens twice, start or increase |
898 | throttling */ | |
070afca2 | 899 | |
d693c6f1 | 900 | if ((rs->num_dirty_pages_period * TARGET_PAGE_SIZE > |
eac74159 | 901 | (bytes_xfer_now - rs->bytes_xfer_prev) / 2) && |
b4a3c64b | 902 | (++rs->dirty_rate_high_cnt >= 2)) { |
56e93d26 | 903 | trace_migration_throttle(); |
8d820d6f | 904 | rs->dirty_rate_high_cnt = 0; |
070afca2 | 905 | mig_throttle_guest_down(); |
d693c6f1 | 906 | } |
56e93d26 | 907 | } |
070afca2 | 908 | |
56e93d26 | 909 | if (migrate_use_xbzrle()) { |
23b28c3c | 910 | if (rs->iterations_prev != rs->iterations) { |
9360447d JQ |
911 | xbzrle_counters.cache_miss_rate = |
912 | (double)(xbzrle_counters.cache_miss - | |
b5833fde | 913 | rs->xbzrle_cache_miss_prev) / |
23b28c3c | 914 | (rs->iterations - rs->iterations_prev); |
56e93d26 | 915 | } |
23b28c3c | 916 | rs->iterations_prev = rs->iterations; |
9360447d | 917 | rs->xbzrle_cache_miss_prev = xbzrle_counters.cache_miss; |
56e93d26 | 918 | } |
d693c6f1 FF |
919 | |
920 | /* reset period counters */ | |
f664da80 | 921 | rs->time_last_bitmap_sync = end_time; |
a66cd90c | 922 | rs->num_dirty_pages_period = 0; |
d2a4d85a | 923 | rs->bytes_xfer_prev = bytes_xfer_now; |
56e93d26 | 924 | } |
4addcd4f | 925 | if (migrate_use_events()) { |
9360447d | 926 | qapi_event_send_migration_pass(ram_counters.dirty_sync_count, NULL); |
4addcd4f | 927 | } |
56e93d26 JQ |
928 | } |
929 | ||
930 | /** | |
3d0684b2 | 931 | * save_zero_page: send the zero page to the stream |
56e93d26 | 932 | * |
3d0684b2 | 933 | * Returns the number of pages written. |
56e93d26 | 934 | * |
f7ccd61b | 935 | * @rs: current RAM state |
56e93d26 JQ |
936 | * @block: block that contains the page we want to send |
937 | * @offset: offset inside the block for the page | |
56e93d26 | 938 | */ |
7faccdc3 | 939 | static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset) |
56e93d26 | 940 | { |
7faccdc3 | 941 | uint8_t *p = block->host + offset; |
56e93d26 JQ |
942 | int pages = -1; |
943 | ||
944 | if (is_zero_range(p, TARGET_PAGE_SIZE)) { | |
9360447d JQ |
945 | ram_counters.duplicate++; |
946 | ram_counters.transferred += | |
bb890ed5 | 947 | save_page_header(rs, rs->f, block, offset | RAM_SAVE_FLAG_ZERO); |
ce25d337 | 948 | qemu_put_byte(rs->f, 0); |
9360447d | 949 | ram_counters.transferred += 1; |
56e93d26 JQ |
950 | pages = 1; |
951 | } | |
952 | ||
953 | return pages; | |
954 | } | |
955 | ||
5727309d | 956 | static void ram_release_pages(const char *rbname, uint64_t offset, int pages) |
53f09a10 | 957 | { |
5727309d | 958 | if (!migrate_release_ram() || !migration_in_postcopy()) { |
53f09a10 PB |
959 | return; |
960 | } | |
961 | ||
aaa2064c | 962 | ram_discard_range(rbname, offset, pages << TARGET_PAGE_BITS); |
53f09a10 PB |
963 | } |
964 | ||
56e93d26 | 965 | /** |
3d0684b2 | 966 | * ram_save_page: send the given page to the stream |
56e93d26 | 967 | * |
3d0684b2 | 968 | * Returns the number of pages written. |
3fd3c4b3 DDAG |
969 | * < 0 - error |
970 | * >=0 - Number of pages written - this might legally be 0 | |
971 | * if xbzrle noticed the page was the same. | |
56e93d26 | 972 | * |
6f37bb8b | 973 | * @rs: current RAM state |
56e93d26 JQ |
974 | * @block: block that contains the page we want to send |
975 | * @offset: offset inside the block for the page | |
976 | * @last_stage: if we are at the completion stage | |
56e93d26 | 977 | */ |
a0a8aa14 | 978 | static int ram_save_page(RAMState *rs, PageSearchStatus *pss, bool last_stage) |
56e93d26 JQ |
979 | { |
980 | int pages = -1; | |
981 | uint64_t bytes_xmit; | |
982 | ram_addr_t current_addr; | |
56e93d26 JQ |
983 | uint8_t *p; |
984 | int ret; | |
985 | bool send_async = true; | |
a08f6890 | 986 | RAMBlock *block = pss->block; |
a935e30f | 987 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
56e93d26 | 988 | |
2f68e399 | 989 | p = block->host + offset; |
1db9d8e5 | 990 | trace_ram_save_page(block->idstr, (uint64_t)offset, p); |
56e93d26 JQ |
991 | |
992 | /* In doubt sent page as normal */ | |
993 | bytes_xmit = 0; | |
ce25d337 | 994 | ret = ram_control_save_page(rs->f, block->offset, |
56e93d26 JQ |
995 | offset, TARGET_PAGE_SIZE, &bytes_xmit); |
996 | if (bytes_xmit) { | |
9360447d | 997 | ram_counters.transferred += bytes_xmit; |
56e93d26 JQ |
998 | pages = 1; |
999 | } | |
1000 | ||
1001 | XBZRLE_cache_lock(); | |
1002 | ||
1003 | current_addr = block->offset + offset; | |
1004 | ||
56e93d26 JQ |
1005 | if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { |
1006 | if (ret != RAM_SAVE_CONTROL_DELAYED) { | |
1007 | if (bytes_xmit > 0) { | |
9360447d | 1008 | ram_counters.normal++; |
56e93d26 | 1009 | } else if (bytes_xmit == 0) { |
9360447d | 1010 | ram_counters.duplicate++; |
56e93d26 JQ |
1011 | } |
1012 | } | |
1013 | } else { | |
7faccdc3 | 1014 | pages = save_zero_page(rs, block, offset); |
56e93d26 JQ |
1015 | if (pages > 0) { |
1016 | /* Must let xbzrle know, otherwise a previous (now 0'd) cached | |
1017 | * page would be stale | |
1018 | */ | |
6f37bb8b | 1019 | xbzrle_cache_zero_page(rs, current_addr); |
a935e30f | 1020 | ram_release_pages(block->idstr, offset, pages); |
6f37bb8b | 1021 | } else if (!rs->ram_bulk_stage && |
5727309d | 1022 | !migration_in_postcopy() && migrate_use_xbzrle()) { |
204b88b8 | 1023 | pages = save_xbzrle_page(rs, &p, current_addr, block, |
072c2511 | 1024 | offset, last_stage); |
56e93d26 JQ |
1025 | if (!last_stage) { |
1026 | /* Can't send this cached data async, since the cache page | |
1027 | * might get updated before it gets to the wire | |
1028 | */ | |
1029 | send_async = false; | |
1030 | } | |
1031 | } | |
1032 | } | |
1033 | ||
1034 | /* XBZRLE overflow or normal page */ | |
1035 | if (pages == -1) { | |
9360447d JQ |
1036 | ram_counters.transferred += |
1037 | save_page_header(rs, rs->f, block, offset | RAM_SAVE_FLAG_PAGE); | |
56e93d26 | 1038 | if (send_async) { |
ce25d337 | 1039 | qemu_put_buffer_async(rs->f, p, TARGET_PAGE_SIZE, |
53f09a10 | 1040 | migrate_release_ram() & |
5727309d | 1041 | migration_in_postcopy()); |
56e93d26 | 1042 | } else { |
ce25d337 | 1043 | qemu_put_buffer(rs->f, p, TARGET_PAGE_SIZE); |
56e93d26 | 1044 | } |
9360447d | 1045 | ram_counters.transferred += TARGET_PAGE_SIZE; |
56e93d26 | 1046 | pages = 1; |
9360447d | 1047 | ram_counters.normal++; |
56e93d26 JQ |
1048 | } |
1049 | ||
1050 | XBZRLE_cache_unlock(); | |
1051 | ||
1052 | return pages; | |
1053 | } | |
1054 | ||
dcaf446e | 1055 | static int do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
a7a9a88f | 1056 | ram_addr_t offset) |
56e93d26 | 1057 | { |
53518d94 | 1058 | RAMState *rs = ram_state; |
56e93d26 | 1059 | int bytes_sent, blen; |
a7a9a88f | 1060 | uint8_t *p = block->host + (offset & TARGET_PAGE_MASK); |
56e93d26 | 1061 | |
2bf3aa85 | 1062 | bytes_sent = save_page_header(rs, f, block, offset | |
56e93d26 | 1063 | RAM_SAVE_FLAG_COMPRESS_PAGE); |
dcaf446e | 1064 | blen = qemu_put_compression_data(f, stream, p, TARGET_PAGE_SIZE); |
b3be2896 LL |
1065 | if (blen < 0) { |
1066 | bytes_sent = 0; | |
1067 | qemu_file_set_error(migrate_get_current()->to_dst_file, blen); | |
1068 | error_report("compressed data failed!"); | |
1069 | } else { | |
1070 | bytes_sent += blen; | |
5727309d | 1071 | ram_release_pages(block->idstr, offset & TARGET_PAGE_MASK, 1); |
b3be2896 | 1072 | } |
56e93d26 JQ |
1073 | |
1074 | return bytes_sent; | |
1075 | } | |
1076 | ||
ce25d337 | 1077 | static void flush_compressed_data(RAMState *rs) |
56e93d26 JQ |
1078 | { |
1079 | int idx, len, thread_count; | |
1080 | ||
1081 | if (!migrate_use_compression()) { | |
1082 | return; | |
1083 | } | |
1084 | thread_count = migrate_compress_threads(); | |
a7a9a88f | 1085 | |
0d9f9a5c | 1086 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 | 1087 | for (idx = 0; idx < thread_count; idx++) { |
a7a9a88f | 1088 | while (!comp_param[idx].done) { |
0d9f9a5c | 1089 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 | 1090 | } |
a7a9a88f | 1091 | } |
0d9f9a5c | 1092 | qemu_mutex_unlock(&comp_done_lock); |
a7a9a88f LL |
1093 | |
1094 | for (idx = 0; idx < thread_count; idx++) { | |
1095 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 1096 | if (!comp_param[idx].quit) { |
ce25d337 | 1097 | len = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
9360447d | 1098 | ram_counters.transferred += len; |
56e93d26 | 1099 | } |
a7a9a88f | 1100 | qemu_mutex_unlock(&comp_param[idx].mutex); |
56e93d26 JQ |
1101 | } |
1102 | } | |
1103 | ||
1104 | static inline void set_compress_params(CompressParam *param, RAMBlock *block, | |
1105 | ram_addr_t offset) | |
1106 | { | |
1107 | param->block = block; | |
1108 | param->offset = offset; | |
1109 | } | |
1110 | ||
ce25d337 JQ |
1111 | static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block, |
1112 | ram_addr_t offset) | |
56e93d26 JQ |
1113 | { |
1114 | int idx, thread_count, bytes_xmit = -1, pages = -1; | |
1115 | ||
1116 | thread_count = migrate_compress_threads(); | |
0d9f9a5c | 1117 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 JQ |
1118 | while (true) { |
1119 | for (idx = 0; idx < thread_count; idx++) { | |
1120 | if (comp_param[idx].done) { | |
a7a9a88f | 1121 | comp_param[idx].done = false; |
ce25d337 | 1122 | bytes_xmit = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
a7a9a88f | 1123 | qemu_mutex_lock(&comp_param[idx].mutex); |
56e93d26 | 1124 | set_compress_params(&comp_param[idx], block, offset); |
a7a9a88f LL |
1125 | qemu_cond_signal(&comp_param[idx].cond); |
1126 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
56e93d26 | 1127 | pages = 1; |
9360447d JQ |
1128 | ram_counters.normal++; |
1129 | ram_counters.transferred += bytes_xmit; | |
56e93d26 JQ |
1130 | break; |
1131 | } | |
1132 | } | |
1133 | if (pages > 0) { | |
1134 | break; | |
1135 | } else { | |
0d9f9a5c | 1136 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 JQ |
1137 | } |
1138 | } | |
0d9f9a5c | 1139 | qemu_mutex_unlock(&comp_done_lock); |
56e93d26 JQ |
1140 | |
1141 | return pages; | |
1142 | } | |
1143 | ||
1144 | /** | |
1145 | * ram_save_compressed_page: compress the given page and send it to the stream | |
1146 | * | |
3d0684b2 | 1147 | * Returns the number of pages written. |
56e93d26 | 1148 | * |
6f37bb8b | 1149 | * @rs: current RAM state |
56e93d26 JQ |
1150 | * @block: block that contains the page we want to send |
1151 | * @offset: offset inside the block for the page | |
1152 | * @last_stage: if we are at the completion stage | |
56e93d26 | 1153 | */ |
a0a8aa14 JQ |
1154 | static int ram_save_compressed_page(RAMState *rs, PageSearchStatus *pss, |
1155 | bool last_stage) | |
56e93d26 JQ |
1156 | { |
1157 | int pages = -1; | |
fc50438e | 1158 | uint64_t bytes_xmit = 0; |
56e93d26 | 1159 | uint8_t *p; |
263a289a | 1160 | int ret; |
a08f6890 | 1161 | RAMBlock *block = pss->block; |
a935e30f | 1162 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
56e93d26 | 1163 | |
2f68e399 | 1164 | p = block->host + offset; |
56e93d26 | 1165 | |
ce25d337 | 1166 | ret = ram_control_save_page(rs->f, block->offset, |
56e93d26 JQ |
1167 | offset, TARGET_PAGE_SIZE, &bytes_xmit); |
1168 | if (bytes_xmit) { | |
9360447d | 1169 | ram_counters.transferred += bytes_xmit; |
56e93d26 JQ |
1170 | pages = 1; |
1171 | } | |
56e93d26 JQ |
1172 | if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { |
1173 | if (ret != RAM_SAVE_CONTROL_DELAYED) { | |
1174 | if (bytes_xmit > 0) { | |
9360447d | 1175 | ram_counters.normal++; |
56e93d26 | 1176 | } else if (bytes_xmit == 0) { |
9360447d | 1177 | ram_counters.duplicate++; |
56e93d26 JQ |
1178 | } |
1179 | } | |
1180 | } else { | |
1181 | /* When starting the process of a new block, the first page of | |
1182 | * the block should be sent out before other pages in the same | |
1183 | * block, and all the pages in last block should have been sent | |
1184 | * out, keeping this order is important, because the 'cont' flag | |
1185 | * is used to avoid resending the block name. | |
1186 | */ | |
6f37bb8b | 1187 | if (block != rs->last_sent_block) { |
ce25d337 | 1188 | flush_compressed_data(rs); |
7faccdc3 | 1189 | pages = save_zero_page(rs, block, offset); |
53f09a10 | 1190 | if (pages > 0) { |
a935e30f | 1191 | ram_release_pages(block->idstr, offset, pages); |
263a289a XG |
1192 | } else { |
1193 | /* | |
1194 | * Make sure the first page is sent out before other pages. | |
1195 | * | |
1196 | * we post it as normal page as compression will take much | |
1197 | * CPU resource. | |
1198 | */ | |
1199 | ram_counters.transferred += save_page_header(rs, rs->f, block, | |
1200 | offset | RAM_SAVE_FLAG_PAGE); | |
1201 | qemu_put_buffer_async(rs->f, p, TARGET_PAGE_SIZE, | |
1202 | migrate_release_ram() & | |
1203 | migration_in_postcopy()); | |
1204 | ram_counters.transferred += TARGET_PAGE_SIZE; | |
1205 | ram_counters.normal++; | |
1206 | pages = 1; | |
53f09a10 | 1207 | } |
56e93d26 | 1208 | } else { |
7faccdc3 | 1209 | pages = save_zero_page(rs, block, offset); |
56e93d26 | 1210 | if (pages == -1) { |
ce25d337 | 1211 | pages = compress_page_with_multi_thread(rs, block, offset); |
53f09a10 | 1212 | } else { |
a935e30f | 1213 | ram_release_pages(block->idstr, offset, pages); |
56e93d26 JQ |
1214 | } |
1215 | } | |
1216 | } | |
1217 | ||
1218 | return pages; | |
1219 | } | |
1220 | ||
3d0684b2 JQ |
1221 | /** |
1222 | * find_dirty_block: find the next dirty page and update any state | |
1223 | * associated with the search process. | |
b9e60928 | 1224 | * |
3d0684b2 | 1225 | * Returns if a page is found |
b9e60928 | 1226 | * |
6f37bb8b | 1227 | * @rs: current RAM state |
3d0684b2 JQ |
1228 | * @pss: data about the state of the current dirty page scan |
1229 | * @again: set to false if the search has scanned the whole of RAM | |
b9e60928 | 1230 | */ |
f20e2865 | 1231 | static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again) |
b9e60928 | 1232 | { |
f20e2865 | 1233 | pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page); |
6f37bb8b | 1234 | if (pss->complete_round && pss->block == rs->last_seen_block && |
a935e30f | 1235 | pss->page >= rs->last_page) { |
b9e60928 DDAG |
1236 | /* |
1237 | * We've been once around the RAM and haven't found anything. | |
1238 | * Give up. | |
1239 | */ | |
1240 | *again = false; | |
1241 | return false; | |
1242 | } | |
a935e30f | 1243 | if ((pss->page << TARGET_PAGE_BITS) >= pss->block->used_length) { |
b9e60928 | 1244 | /* Didn't find anything in this RAM Block */ |
a935e30f | 1245 | pss->page = 0; |
b9e60928 DDAG |
1246 | pss->block = QLIST_NEXT_RCU(pss->block, next); |
1247 | if (!pss->block) { | |
1248 | /* Hit the end of the list */ | |
1249 | pss->block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1250 | /* Flag that we've looped */ | |
1251 | pss->complete_round = true; | |
6f37bb8b | 1252 | rs->ram_bulk_stage = false; |
b9e60928 DDAG |
1253 | if (migrate_use_xbzrle()) { |
1254 | /* If xbzrle is on, stop using the data compression at this | |
1255 | * point. In theory, xbzrle can do better than compression. | |
1256 | */ | |
ce25d337 | 1257 | flush_compressed_data(rs); |
b9e60928 DDAG |
1258 | } |
1259 | } | |
1260 | /* Didn't find anything this time, but try again on the new block */ | |
1261 | *again = true; | |
1262 | return false; | |
1263 | } else { | |
1264 | /* Can go around again, but... */ | |
1265 | *again = true; | |
1266 | /* We've found something so probably don't need to */ | |
1267 | return true; | |
1268 | } | |
1269 | } | |
1270 | ||
3d0684b2 JQ |
1271 | /** |
1272 | * unqueue_page: gets a page of the queue | |
1273 | * | |
a82d593b | 1274 | * Helper for 'get_queued_page' - gets a page off the queue |
a82d593b | 1275 | * |
3d0684b2 JQ |
1276 | * Returns the block of the page (or NULL if none available) |
1277 | * | |
ec481c6c | 1278 | * @rs: current RAM state |
3d0684b2 | 1279 | * @offset: used to return the offset within the RAMBlock |
a82d593b | 1280 | */ |
f20e2865 | 1281 | static RAMBlock *unqueue_page(RAMState *rs, ram_addr_t *offset) |
a82d593b DDAG |
1282 | { |
1283 | RAMBlock *block = NULL; | |
1284 | ||
ec481c6c JQ |
1285 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1286 | if (!QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
1287 | struct RAMSrcPageRequest *entry = | |
1288 | QSIMPLEQ_FIRST(&rs->src_page_requests); | |
a82d593b DDAG |
1289 | block = entry->rb; |
1290 | *offset = entry->offset; | |
a82d593b DDAG |
1291 | |
1292 | if (entry->len > TARGET_PAGE_SIZE) { | |
1293 | entry->len -= TARGET_PAGE_SIZE; | |
1294 | entry->offset += TARGET_PAGE_SIZE; | |
1295 | } else { | |
1296 | memory_region_unref(block->mr); | |
ec481c6c | 1297 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
a82d593b DDAG |
1298 | g_free(entry); |
1299 | } | |
1300 | } | |
ec481c6c | 1301 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
a82d593b DDAG |
1302 | |
1303 | return block; | |
1304 | } | |
1305 | ||
3d0684b2 JQ |
1306 | /** |
1307 | * get_queued_page: unqueue a page from the postocpy requests | |
1308 | * | |
1309 | * Skips pages that are already sent (!dirty) | |
a82d593b | 1310 | * |
3d0684b2 | 1311 | * Returns if a queued page is found |
a82d593b | 1312 | * |
6f37bb8b | 1313 | * @rs: current RAM state |
3d0684b2 | 1314 | * @pss: data about the state of the current dirty page scan |
a82d593b | 1315 | */ |
f20e2865 | 1316 | static bool get_queued_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b DDAG |
1317 | { |
1318 | RAMBlock *block; | |
1319 | ram_addr_t offset; | |
1320 | bool dirty; | |
1321 | ||
1322 | do { | |
f20e2865 | 1323 | block = unqueue_page(rs, &offset); |
a82d593b DDAG |
1324 | /* |
1325 | * We're sending this page, and since it's postcopy nothing else | |
1326 | * will dirty it, and we must make sure it doesn't get sent again | |
1327 | * even if this queue request was received after the background | |
1328 | * search already sent it. | |
1329 | */ | |
1330 | if (block) { | |
f20e2865 JQ |
1331 | unsigned long page; |
1332 | ||
6b6712ef JQ |
1333 | page = offset >> TARGET_PAGE_BITS; |
1334 | dirty = test_bit(page, block->bmap); | |
a82d593b | 1335 | if (!dirty) { |
06b10688 | 1336 | trace_get_queued_page_not_dirty(block->idstr, (uint64_t)offset, |
6b6712ef | 1337 | page, test_bit(page, block->unsentmap)); |
a82d593b | 1338 | } else { |
f20e2865 | 1339 | trace_get_queued_page(block->idstr, (uint64_t)offset, page); |
a82d593b DDAG |
1340 | } |
1341 | } | |
1342 | ||
1343 | } while (block && !dirty); | |
1344 | ||
1345 | if (block) { | |
1346 | /* | |
1347 | * As soon as we start servicing pages out of order, then we have | |
1348 | * to kill the bulk stage, since the bulk stage assumes | |
1349 | * in (migration_bitmap_find_and_reset_dirty) that every page is | |
1350 | * dirty, that's no longer true. | |
1351 | */ | |
6f37bb8b | 1352 | rs->ram_bulk_stage = false; |
a82d593b DDAG |
1353 | |
1354 | /* | |
1355 | * We want the background search to continue from the queued page | |
1356 | * since the guest is likely to want other pages near to the page | |
1357 | * it just requested. | |
1358 | */ | |
1359 | pss->block = block; | |
a935e30f | 1360 | pss->page = offset >> TARGET_PAGE_BITS; |
a82d593b DDAG |
1361 | } |
1362 | ||
1363 | return !!block; | |
1364 | } | |
1365 | ||
6c595cde | 1366 | /** |
5e58f968 JQ |
1367 | * migration_page_queue_free: drop any remaining pages in the ram |
1368 | * request queue | |
6c595cde | 1369 | * |
3d0684b2 JQ |
1370 | * It should be empty at the end anyway, but in error cases there may |
1371 | * be some left. in case that there is any page left, we drop it. | |
1372 | * | |
6c595cde | 1373 | */ |
83c13382 | 1374 | static void migration_page_queue_free(RAMState *rs) |
6c595cde | 1375 | { |
ec481c6c | 1376 | struct RAMSrcPageRequest *mspr, *next_mspr; |
6c595cde DDAG |
1377 | /* This queue generally should be empty - but in the case of a failed |
1378 | * migration might have some droppings in. | |
1379 | */ | |
1380 | rcu_read_lock(); | |
ec481c6c | 1381 | QSIMPLEQ_FOREACH_SAFE(mspr, &rs->src_page_requests, next_req, next_mspr) { |
6c595cde | 1382 | memory_region_unref(mspr->rb->mr); |
ec481c6c | 1383 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
6c595cde DDAG |
1384 | g_free(mspr); |
1385 | } | |
1386 | rcu_read_unlock(); | |
1387 | } | |
1388 | ||
1389 | /** | |
3d0684b2 JQ |
1390 | * ram_save_queue_pages: queue the page for transmission |
1391 | * | |
1392 | * A request from postcopy destination for example. | |
1393 | * | |
1394 | * Returns zero on success or negative on error | |
1395 | * | |
3d0684b2 JQ |
1396 | * @rbname: Name of the RAMBLock of the request. NULL means the |
1397 | * same that last one. | |
1398 | * @start: starting address from the start of the RAMBlock | |
1399 | * @len: length (in bytes) to send | |
6c595cde | 1400 | */ |
96506894 | 1401 | int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len) |
6c595cde DDAG |
1402 | { |
1403 | RAMBlock *ramblock; | |
53518d94 | 1404 | RAMState *rs = ram_state; |
6c595cde | 1405 | |
9360447d | 1406 | ram_counters.postcopy_requests++; |
6c595cde DDAG |
1407 | rcu_read_lock(); |
1408 | if (!rbname) { | |
1409 | /* Reuse last RAMBlock */ | |
68a098f3 | 1410 | ramblock = rs->last_req_rb; |
6c595cde DDAG |
1411 | |
1412 | if (!ramblock) { | |
1413 | /* | |
1414 | * Shouldn't happen, we can't reuse the last RAMBlock if | |
1415 | * it's the 1st request. | |
1416 | */ | |
1417 | error_report("ram_save_queue_pages no previous block"); | |
1418 | goto err; | |
1419 | } | |
1420 | } else { | |
1421 | ramblock = qemu_ram_block_by_name(rbname); | |
1422 | ||
1423 | if (!ramblock) { | |
1424 | /* We shouldn't be asked for a non-existent RAMBlock */ | |
1425 | error_report("ram_save_queue_pages no block '%s'", rbname); | |
1426 | goto err; | |
1427 | } | |
68a098f3 | 1428 | rs->last_req_rb = ramblock; |
6c595cde DDAG |
1429 | } |
1430 | trace_ram_save_queue_pages(ramblock->idstr, start, len); | |
1431 | if (start+len > ramblock->used_length) { | |
9458ad6b JQ |
1432 | error_report("%s request overrun start=" RAM_ADDR_FMT " len=" |
1433 | RAM_ADDR_FMT " blocklen=" RAM_ADDR_FMT, | |
6c595cde DDAG |
1434 | __func__, start, len, ramblock->used_length); |
1435 | goto err; | |
1436 | } | |
1437 | ||
ec481c6c JQ |
1438 | struct RAMSrcPageRequest *new_entry = |
1439 | g_malloc0(sizeof(struct RAMSrcPageRequest)); | |
6c595cde DDAG |
1440 | new_entry->rb = ramblock; |
1441 | new_entry->offset = start; | |
1442 | new_entry->len = len; | |
1443 | ||
1444 | memory_region_ref(ramblock->mr); | |
ec481c6c JQ |
1445 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1446 | QSIMPLEQ_INSERT_TAIL(&rs->src_page_requests, new_entry, next_req); | |
1447 | qemu_mutex_unlock(&rs->src_page_req_mutex); | |
6c595cde DDAG |
1448 | rcu_read_unlock(); |
1449 | ||
1450 | return 0; | |
1451 | ||
1452 | err: | |
1453 | rcu_read_unlock(); | |
1454 | return -1; | |
1455 | } | |
1456 | ||
a82d593b | 1457 | /** |
3d0684b2 | 1458 | * ram_save_target_page: save one target page |
a82d593b | 1459 | * |
3d0684b2 | 1460 | * Returns the number of pages written |
a82d593b | 1461 | * |
6f37bb8b | 1462 | * @rs: current RAM state |
3d0684b2 | 1463 | * @ms: current migration state |
3d0684b2 | 1464 | * @pss: data about the page we want to send |
a82d593b | 1465 | * @last_stage: if we are at the completion stage |
a82d593b | 1466 | */ |
a0a8aa14 | 1467 | static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 1468 | bool last_stage) |
a82d593b DDAG |
1469 | { |
1470 | int res = 0; | |
1471 | ||
1472 | /* Check the pages is dirty and if it is send it */ | |
f20e2865 | 1473 | if (migration_bitmap_clear_dirty(rs, pss->block, pss->page)) { |
6d358d94 JQ |
1474 | /* |
1475 | * If xbzrle is on, stop using the data compression after first | |
1476 | * round of migration even if compression is enabled. In theory, | |
1477 | * xbzrle can do better than compression. | |
1478 | */ | |
6b6712ef JQ |
1479 | if (migrate_use_compression() && |
1480 | (rs->ram_bulk_stage || !migrate_use_xbzrle())) { | |
a0a8aa14 | 1481 | res = ram_save_compressed_page(rs, pss, last_stage); |
a82d593b | 1482 | } else { |
a0a8aa14 | 1483 | res = ram_save_page(rs, pss, last_stage); |
a82d593b DDAG |
1484 | } |
1485 | ||
1486 | if (res < 0) { | |
1487 | return res; | |
1488 | } | |
6b6712ef JQ |
1489 | if (pss->block->unsentmap) { |
1490 | clear_bit(pss->page, pss->block->unsentmap); | |
a82d593b DDAG |
1491 | } |
1492 | } | |
1493 | ||
1494 | return res; | |
1495 | } | |
1496 | ||
1497 | /** | |
3d0684b2 | 1498 | * ram_save_host_page: save a whole host page |
a82d593b | 1499 | * |
3d0684b2 JQ |
1500 | * Starting at *offset send pages up to the end of the current host |
1501 | * page. It's valid for the initial offset to point into the middle of | |
1502 | * a host page in which case the remainder of the hostpage is sent. | |
1503 | * Only dirty target pages are sent. Note that the host page size may | |
1504 | * be a huge page for this block. | |
1eb3fc0a DDAG |
1505 | * The saving stops at the boundary of the used_length of the block |
1506 | * if the RAMBlock isn't a multiple of the host page size. | |
a82d593b | 1507 | * |
3d0684b2 JQ |
1508 | * Returns the number of pages written or negative on error |
1509 | * | |
6f37bb8b | 1510 | * @rs: current RAM state |
3d0684b2 | 1511 | * @ms: current migration state |
3d0684b2 | 1512 | * @pss: data about the page we want to send |
a82d593b | 1513 | * @last_stage: if we are at the completion stage |
a82d593b | 1514 | */ |
a0a8aa14 | 1515 | static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 1516 | bool last_stage) |
a82d593b DDAG |
1517 | { |
1518 | int tmppages, pages = 0; | |
a935e30f JQ |
1519 | size_t pagesize_bits = |
1520 | qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
4c011c37 | 1521 | |
a82d593b | 1522 | do { |
f20e2865 | 1523 | tmppages = ram_save_target_page(rs, pss, last_stage); |
a82d593b DDAG |
1524 | if (tmppages < 0) { |
1525 | return tmppages; | |
1526 | } | |
1527 | ||
1528 | pages += tmppages; | |
a935e30f | 1529 | pss->page++; |
1eb3fc0a DDAG |
1530 | } while ((pss->page & (pagesize_bits - 1)) && |
1531 | offset_in_ramblock(pss->block, pss->page << TARGET_PAGE_BITS)); | |
a82d593b DDAG |
1532 | |
1533 | /* The offset we leave with is the last one we looked at */ | |
a935e30f | 1534 | pss->page--; |
a82d593b DDAG |
1535 | return pages; |
1536 | } | |
6c595cde | 1537 | |
56e93d26 | 1538 | /** |
3d0684b2 | 1539 | * ram_find_and_save_block: finds a dirty page and sends it to f |
56e93d26 JQ |
1540 | * |
1541 | * Called within an RCU critical section. | |
1542 | * | |
3d0684b2 | 1543 | * Returns the number of pages written where zero means no dirty pages |
56e93d26 | 1544 | * |
6f37bb8b | 1545 | * @rs: current RAM state |
56e93d26 | 1546 | * @last_stage: if we are at the completion stage |
a82d593b DDAG |
1547 | * |
1548 | * On systems where host-page-size > target-page-size it will send all the | |
1549 | * pages in a host page that are dirty. | |
56e93d26 JQ |
1550 | */ |
1551 | ||
ce25d337 | 1552 | static int ram_find_and_save_block(RAMState *rs, bool last_stage) |
56e93d26 | 1553 | { |
b8fb8cb7 | 1554 | PageSearchStatus pss; |
56e93d26 | 1555 | int pages = 0; |
b9e60928 | 1556 | bool again, found; |
56e93d26 | 1557 | |
0827b9e9 AA |
1558 | /* No dirty page as there is zero RAM */ |
1559 | if (!ram_bytes_total()) { | |
1560 | return pages; | |
1561 | } | |
1562 | ||
6f37bb8b | 1563 | pss.block = rs->last_seen_block; |
a935e30f | 1564 | pss.page = rs->last_page; |
b8fb8cb7 DDAG |
1565 | pss.complete_round = false; |
1566 | ||
1567 | if (!pss.block) { | |
1568 | pss.block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1569 | } | |
56e93d26 | 1570 | |
b9e60928 | 1571 | do { |
a82d593b | 1572 | again = true; |
f20e2865 | 1573 | found = get_queued_page(rs, &pss); |
b9e60928 | 1574 | |
a82d593b DDAG |
1575 | if (!found) { |
1576 | /* priority queue empty, so just search for something dirty */ | |
f20e2865 | 1577 | found = find_dirty_block(rs, &pss, &again); |
a82d593b | 1578 | } |
f3f491fc | 1579 | |
a82d593b | 1580 | if (found) { |
f20e2865 | 1581 | pages = ram_save_host_page(rs, &pss, last_stage); |
56e93d26 | 1582 | } |
b9e60928 | 1583 | } while (!pages && again); |
56e93d26 | 1584 | |
6f37bb8b | 1585 | rs->last_seen_block = pss.block; |
a935e30f | 1586 | rs->last_page = pss.page; |
56e93d26 JQ |
1587 | |
1588 | return pages; | |
1589 | } | |
1590 | ||
1591 | void acct_update_position(QEMUFile *f, size_t size, bool zero) | |
1592 | { | |
1593 | uint64_t pages = size / TARGET_PAGE_SIZE; | |
f7ccd61b | 1594 | |
56e93d26 | 1595 | if (zero) { |
9360447d | 1596 | ram_counters.duplicate += pages; |
56e93d26 | 1597 | } else { |
9360447d JQ |
1598 | ram_counters.normal += pages; |
1599 | ram_counters.transferred += size; | |
56e93d26 JQ |
1600 | qemu_update_position(f, size); |
1601 | } | |
1602 | } | |
1603 | ||
56e93d26 JQ |
1604 | uint64_t ram_bytes_total(void) |
1605 | { | |
1606 | RAMBlock *block; | |
1607 | uint64_t total = 0; | |
1608 | ||
1609 | rcu_read_lock(); | |
99e15582 | 1610 | RAMBLOCK_FOREACH(block) { |
56e93d26 | 1611 | total += block->used_length; |
99e15582 | 1612 | } |
56e93d26 JQ |
1613 | rcu_read_unlock(); |
1614 | return total; | |
1615 | } | |
1616 | ||
f265e0e4 | 1617 | static void xbzrle_load_setup(void) |
56e93d26 | 1618 | { |
f265e0e4 | 1619 | XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE); |
56e93d26 JQ |
1620 | } |
1621 | ||
f265e0e4 JQ |
1622 | static void xbzrle_load_cleanup(void) |
1623 | { | |
1624 | g_free(XBZRLE.decoded_buf); | |
1625 | XBZRLE.decoded_buf = NULL; | |
1626 | } | |
1627 | ||
7d7c96be PX |
1628 | static void ram_state_cleanup(RAMState **rsp) |
1629 | { | |
b9ccaf6d DDAG |
1630 | if (*rsp) { |
1631 | migration_page_queue_free(*rsp); | |
1632 | qemu_mutex_destroy(&(*rsp)->bitmap_mutex); | |
1633 | qemu_mutex_destroy(&(*rsp)->src_page_req_mutex); | |
1634 | g_free(*rsp); | |
1635 | *rsp = NULL; | |
1636 | } | |
7d7c96be PX |
1637 | } |
1638 | ||
84593a08 PX |
1639 | static void xbzrle_cleanup(void) |
1640 | { | |
1641 | XBZRLE_cache_lock(); | |
1642 | if (XBZRLE.cache) { | |
1643 | cache_fini(XBZRLE.cache); | |
1644 | g_free(XBZRLE.encoded_buf); | |
1645 | g_free(XBZRLE.current_buf); | |
1646 | g_free(XBZRLE.zero_target_page); | |
1647 | XBZRLE.cache = NULL; | |
1648 | XBZRLE.encoded_buf = NULL; | |
1649 | XBZRLE.current_buf = NULL; | |
1650 | XBZRLE.zero_target_page = NULL; | |
1651 | } | |
1652 | XBZRLE_cache_unlock(); | |
1653 | } | |
1654 | ||
f265e0e4 | 1655 | static void ram_save_cleanup(void *opaque) |
56e93d26 | 1656 | { |
53518d94 | 1657 | RAMState **rsp = opaque; |
6b6712ef | 1658 | RAMBlock *block; |
eb859c53 | 1659 | |
2ff64038 LZ |
1660 | /* caller have hold iothread lock or is in a bh, so there is |
1661 | * no writing race against this migration_bitmap | |
1662 | */ | |
6b6712ef JQ |
1663 | memory_global_dirty_log_stop(); |
1664 | ||
1665 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1666 | g_free(block->bmap); | |
1667 | block->bmap = NULL; | |
1668 | g_free(block->unsentmap); | |
1669 | block->unsentmap = NULL; | |
56e93d26 JQ |
1670 | } |
1671 | ||
84593a08 | 1672 | xbzrle_cleanup(); |
f0afa331 | 1673 | compress_threads_save_cleanup(); |
7d7c96be | 1674 | ram_state_cleanup(rsp); |
56e93d26 JQ |
1675 | } |
1676 | ||
6f37bb8b | 1677 | static void ram_state_reset(RAMState *rs) |
56e93d26 | 1678 | { |
6f37bb8b JQ |
1679 | rs->last_seen_block = NULL; |
1680 | rs->last_sent_block = NULL; | |
269ace29 | 1681 | rs->last_page = 0; |
6f37bb8b JQ |
1682 | rs->last_version = ram_list.version; |
1683 | rs->ram_bulk_stage = true; | |
56e93d26 JQ |
1684 | } |
1685 | ||
1686 | #define MAX_WAIT 50 /* ms, half buffered_file limit */ | |
1687 | ||
4f2e4252 DDAG |
1688 | /* |
1689 | * 'expected' is the value you expect the bitmap mostly to be full | |
1690 | * of; it won't bother printing lines that are all this value. | |
1691 | * If 'todump' is null the migration bitmap is dumped. | |
1692 | */ | |
6b6712ef JQ |
1693 | void ram_debug_dump_bitmap(unsigned long *todump, bool expected, |
1694 | unsigned long pages) | |
4f2e4252 | 1695 | { |
4f2e4252 DDAG |
1696 | int64_t cur; |
1697 | int64_t linelen = 128; | |
1698 | char linebuf[129]; | |
1699 | ||
6b6712ef | 1700 | for (cur = 0; cur < pages; cur += linelen) { |
4f2e4252 DDAG |
1701 | int64_t curb; |
1702 | bool found = false; | |
1703 | /* | |
1704 | * Last line; catch the case where the line length | |
1705 | * is longer than remaining ram | |
1706 | */ | |
6b6712ef JQ |
1707 | if (cur + linelen > pages) { |
1708 | linelen = pages - cur; | |
4f2e4252 DDAG |
1709 | } |
1710 | for (curb = 0; curb < linelen; curb++) { | |
1711 | bool thisbit = test_bit(cur + curb, todump); | |
1712 | linebuf[curb] = thisbit ? '1' : '.'; | |
1713 | found = found || (thisbit != expected); | |
1714 | } | |
1715 | if (found) { | |
1716 | linebuf[curb] = '\0'; | |
1717 | fprintf(stderr, "0x%08" PRIx64 " : %s\n", cur, linebuf); | |
1718 | } | |
1719 | } | |
1720 | } | |
1721 | ||
e0b266f0 DDAG |
1722 | /* **** functions for postcopy ***** */ |
1723 | ||
ced1c616 PB |
1724 | void ram_postcopy_migrated_memory_release(MigrationState *ms) |
1725 | { | |
1726 | struct RAMBlock *block; | |
ced1c616 | 1727 | |
99e15582 | 1728 | RAMBLOCK_FOREACH(block) { |
6b6712ef JQ |
1729 | unsigned long *bitmap = block->bmap; |
1730 | unsigned long range = block->used_length >> TARGET_PAGE_BITS; | |
1731 | unsigned long run_start = find_next_zero_bit(bitmap, range, 0); | |
ced1c616 PB |
1732 | |
1733 | while (run_start < range) { | |
1734 | unsigned long run_end = find_next_bit(bitmap, range, run_start + 1); | |
aaa2064c | 1735 | ram_discard_range(block->idstr, run_start << TARGET_PAGE_BITS, |
ced1c616 PB |
1736 | (run_end - run_start) << TARGET_PAGE_BITS); |
1737 | run_start = find_next_zero_bit(bitmap, range, run_end + 1); | |
1738 | } | |
1739 | } | |
1740 | } | |
1741 | ||
3d0684b2 JQ |
1742 | /** |
1743 | * postcopy_send_discard_bm_ram: discard a RAMBlock | |
1744 | * | |
1745 | * Returns zero on success | |
1746 | * | |
e0b266f0 DDAG |
1747 | * Callback from postcopy_each_ram_send_discard for each RAMBlock |
1748 | * Note: At this point the 'unsentmap' is the processed bitmap combined | |
1749 | * with the dirtymap; so a '1' means it's either dirty or unsent. | |
3d0684b2 JQ |
1750 | * |
1751 | * @ms: current migration state | |
1752 | * @pds: state for postcopy | |
1753 | * @start: RAMBlock starting page | |
1754 | * @length: RAMBlock size | |
e0b266f0 DDAG |
1755 | */ |
1756 | static int postcopy_send_discard_bm_ram(MigrationState *ms, | |
1757 | PostcopyDiscardState *pds, | |
6b6712ef | 1758 | RAMBlock *block) |
e0b266f0 | 1759 | { |
6b6712ef | 1760 | unsigned long end = block->used_length >> TARGET_PAGE_BITS; |
e0b266f0 | 1761 | unsigned long current; |
6b6712ef | 1762 | unsigned long *unsentmap = block->unsentmap; |
e0b266f0 | 1763 | |
6b6712ef | 1764 | for (current = 0; current < end; ) { |
e0b266f0 DDAG |
1765 | unsigned long one = find_next_bit(unsentmap, end, current); |
1766 | ||
1767 | if (one <= end) { | |
1768 | unsigned long zero = find_next_zero_bit(unsentmap, end, one + 1); | |
1769 | unsigned long discard_length; | |
1770 | ||
1771 | if (zero >= end) { | |
1772 | discard_length = end - one; | |
1773 | } else { | |
1774 | discard_length = zero - one; | |
1775 | } | |
d688c62d DDAG |
1776 | if (discard_length) { |
1777 | postcopy_discard_send_range(ms, pds, one, discard_length); | |
1778 | } | |
e0b266f0 DDAG |
1779 | current = one + discard_length; |
1780 | } else { | |
1781 | current = one; | |
1782 | } | |
1783 | } | |
1784 | ||
1785 | return 0; | |
1786 | } | |
1787 | ||
3d0684b2 JQ |
1788 | /** |
1789 | * postcopy_each_ram_send_discard: discard all RAMBlocks | |
1790 | * | |
1791 | * Returns 0 for success or negative for error | |
1792 | * | |
e0b266f0 DDAG |
1793 | * Utility for the outgoing postcopy code. |
1794 | * Calls postcopy_send_discard_bm_ram for each RAMBlock | |
1795 | * passing it bitmap indexes and name. | |
e0b266f0 DDAG |
1796 | * (qemu_ram_foreach_block ends up passing unscaled lengths |
1797 | * which would mean postcopy code would have to deal with target page) | |
3d0684b2 JQ |
1798 | * |
1799 | * @ms: current migration state | |
e0b266f0 DDAG |
1800 | */ |
1801 | static int postcopy_each_ram_send_discard(MigrationState *ms) | |
1802 | { | |
1803 | struct RAMBlock *block; | |
1804 | int ret; | |
1805 | ||
99e15582 | 1806 | RAMBLOCK_FOREACH(block) { |
6b6712ef JQ |
1807 | PostcopyDiscardState *pds = |
1808 | postcopy_discard_send_init(ms, block->idstr); | |
e0b266f0 DDAG |
1809 | |
1810 | /* | |
1811 | * Postcopy sends chunks of bitmap over the wire, but it | |
1812 | * just needs indexes at this point, avoids it having | |
1813 | * target page specific code. | |
1814 | */ | |
6b6712ef | 1815 | ret = postcopy_send_discard_bm_ram(ms, pds, block); |
e0b266f0 DDAG |
1816 | postcopy_discard_send_finish(ms, pds); |
1817 | if (ret) { | |
1818 | return ret; | |
1819 | } | |
1820 | } | |
1821 | ||
1822 | return 0; | |
1823 | } | |
1824 | ||
3d0684b2 JQ |
1825 | /** |
1826 | * postcopy_chunk_hostpages_pass: canocalize bitmap in hostpages | |
1827 | * | |
1828 | * Helper for postcopy_chunk_hostpages; it's called twice to | |
1829 | * canonicalize the two bitmaps, that are similar, but one is | |
1830 | * inverted. | |
99e314eb | 1831 | * |
3d0684b2 JQ |
1832 | * Postcopy requires that all target pages in a hostpage are dirty or |
1833 | * clean, not a mix. This function canonicalizes the bitmaps. | |
99e314eb | 1834 | * |
3d0684b2 JQ |
1835 | * @ms: current migration state |
1836 | * @unsent_pass: if true we need to canonicalize partially unsent host pages | |
1837 | * otherwise we need to canonicalize partially dirty host pages | |
1838 | * @block: block that contains the page we want to canonicalize | |
1839 | * @pds: state for postcopy | |
99e314eb DDAG |
1840 | */ |
1841 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, bool unsent_pass, | |
1842 | RAMBlock *block, | |
1843 | PostcopyDiscardState *pds) | |
1844 | { | |
53518d94 | 1845 | RAMState *rs = ram_state; |
6b6712ef JQ |
1846 | unsigned long *bitmap = block->bmap; |
1847 | unsigned long *unsentmap = block->unsentmap; | |
29c59172 | 1848 | unsigned int host_ratio = block->page_size / TARGET_PAGE_SIZE; |
6b6712ef | 1849 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
99e314eb DDAG |
1850 | unsigned long run_start; |
1851 | ||
29c59172 DDAG |
1852 | if (block->page_size == TARGET_PAGE_SIZE) { |
1853 | /* Easy case - TPS==HPS for a non-huge page RAMBlock */ | |
1854 | return; | |
1855 | } | |
1856 | ||
99e314eb DDAG |
1857 | if (unsent_pass) { |
1858 | /* Find a sent page */ | |
6b6712ef | 1859 | run_start = find_next_zero_bit(unsentmap, pages, 0); |
99e314eb DDAG |
1860 | } else { |
1861 | /* Find a dirty page */ | |
6b6712ef | 1862 | run_start = find_next_bit(bitmap, pages, 0); |
99e314eb DDAG |
1863 | } |
1864 | ||
6b6712ef | 1865 | while (run_start < pages) { |
99e314eb DDAG |
1866 | bool do_fixup = false; |
1867 | unsigned long fixup_start_addr; | |
1868 | unsigned long host_offset; | |
1869 | ||
1870 | /* | |
1871 | * If the start of this run of pages is in the middle of a host | |
1872 | * page, then we need to fixup this host page. | |
1873 | */ | |
1874 | host_offset = run_start % host_ratio; | |
1875 | if (host_offset) { | |
1876 | do_fixup = true; | |
1877 | run_start -= host_offset; | |
1878 | fixup_start_addr = run_start; | |
1879 | /* For the next pass */ | |
1880 | run_start = run_start + host_ratio; | |
1881 | } else { | |
1882 | /* Find the end of this run */ | |
1883 | unsigned long run_end; | |
1884 | if (unsent_pass) { | |
6b6712ef | 1885 | run_end = find_next_bit(unsentmap, pages, run_start + 1); |
99e314eb | 1886 | } else { |
6b6712ef | 1887 | run_end = find_next_zero_bit(bitmap, pages, run_start + 1); |
99e314eb DDAG |
1888 | } |
1889 | /* | |
1890 | * If the end isn't at the start of a host page, then the | |
1891 | * run doesn't finish at the end of a host page | |
1892 | * and we need to discard. | |
1893 | */ | |
1894 | host_offset = run_end % host_ratio; | |
1895 | if (host_offset) { | |
1896 | do_fixup = true; | |
1897 | fixup_start_addr = run_end - host_offset; | |
1898 | /* | |
1899 | * This host page has gone, the next loop iteration starts | |
1900 | * from after the fixup | |
1901 | */ | |
1902 | run_start = fixup_start_addr + host_ratio; | |
1903 | } else { | |
1904 | /* | |
1905 | * No discards on this iteration, next loop starts from | |
1906 | * next sent/dirty page | |
1907 | */ | |
1908 | run_start = run_end + 1; | |
1909 | } | |
1910 | } | |
1911 | ||
1912 | if (do_fixup) { | |
1913 | unsigned long page; | |
1914 | ||
1915 | /* Tell the destination to discard this page */ | |
1916 | if (unsent_pass || !test_bit(fixup_start_addr, unsentmap)) { | |
1917 | /* For the unsent_pass we: | |
1918 | * discard partially sent pages | |
1919 | * For the !unsent_pass (dirty) we: | |
1920 | * discard partially dirty pages that were sent | |
1921 | * (any partially sent pages were already discarded | |
1922 | * by the previous unsent_pass) | |
1923 | */ | |
1924 | postcopy_discard_send_range(ms, pds, fixup_start_addr, | |
1925 | host_ratio); | |
1926 | } | |
1927 | ||
1928 | /* Clean up the bitmap */ | |
1929 | for (page = fixup_start_addr; | |
1930 | page < fixup_start_addr + host_ratio; page++) { | |
1931 | /* All pages in this host page are now not sent */ | |
1932 | set_bit(page, unsentmap); | |
1933 | ||
1934 | /* | |
1935 | * Remark them as dirty, updating the count for any pages | |
1936 | * that weren't previously dirty. | |
1937 | */ | |
0d8ec885 | 1938 | rs->migration_dirty_pages += !test_and_set_bit(page, bitmap); |
99e314eb DDAG |
1939 | } |
1940 | } | |
1941 | ||
1942 | if (unsent_pass) { | |
1943 | /* Find the next sent page for the next iteration */ | |
6b6712ef | 1944 | run_start = find_next_zero_bit(unsentmap, pages, run_start); |
99e314eb DDAG |
1945 | } else { |
1946 | /* Find the next dirty page for the next iteration */ | |
6b6712ef | 1947 | run_start = find_next_bit(bitmap, pages, run_start); |
99e314eb DDAG |
1948 | } |
1949 | } | |
1950 | } | |
1951 | ||
3d0684b2 JQ |
1952 | /** |
1953 | * postcopy_chuck_hostpages: discrad any partially sent host page | |
1954 | * | |
99e314eb DDAG |
1955 | * Utility for the outgoing postcopy code. |
1956 | * | |
1957 | * Discard any partially sent host-page size chunks, mark any partially | |
29c59172 DDAG |
1958 | * dirty host-page size chunks as all dirty. In this case the host-page |
1959 | * is the host-page for the particular RAMBlock, i.e. it might be a huge page | |
99e314eb | 1960 | * |
3d0684b2 JQ |
1961 | * Returns zero on success |
1962 | * | |
1963 | * @ms: current migration state | |
6b6712ef | 1964 | * @block: block we want to work with |
99e314eb | 1965 | */ |
6b6712ef | 1966 | static int postcopy_chunk_hostpages(MigrationState *ms, RAMBlock *block) |
99e314eb | 1967 | { |
6b6712ef JQ |
1968 | PostcopyDiscardState *pds = |
1969 | postcopy_discard_send_init(ms, block->idstr); | |
99e314eb | 1970 | |
6b6712ef JQ |
1971 | /* First pass: Discard all partially sent host pages */ |
1972 | postcopy_chunk_hostpages_pass(ms, true, block, pds); | |
1973 | /* | |
1974 | * Second pass: Ensure that all partially dirty host pages are made | |
1975 | * fully dirty. | |
1976 | */ | |
1977 | postcopy_chunk_hostpages_pass(ms, false, block, pds); | |
99e314eb | 1978 | |
6b6712ef | 1979 | postcopy_discard_send_finish(ms, pds); |
99e314eb DDAG |
1980 | return 0; |
1981 | } | |
1982 | ||
3d0684b2 JQ |
1983 | /** |
1984 | * ram_postcopy_send_discard_bitmap: transmit the discard bitmap | |
1985 | * | |
1986 | * Returns zero on success | |
1987 | * | |
e0b266f0 DDAG |
1988 | * Transmit the set of pages to be discarded after precopy to the target |
1989 | * these are pages that: | |
1990 | * a) Have been previously transmitted but are now dirty again | |
1991 | * b) Pages that have never been transmitted, this ensures that | |
1992 | * any pages on the destination that have been mapped by background | |
1993 | * tasks get discarded (transparent huge pages is the specific concern) | |
1994 | * Hopefully this is pretty sparse | |
3d0684b2 JQ |
1995 | * |
1996 | * @ms: current migration state | |
e0b266f0 DDAG |
1997 | */ |
1998 | int ram_postcopy_send_discard_bitmap(MigrationState *ms) | |
1999 | { | |
53518d94 | 2000 | RAMState *rs = ram_state; |
6b6712ef | 2001 | RAMBlock *block; |
e0b266f0 | 2002 | int ret; |
e0b266f0 DDAG |
2003 | |
2004 | rcu_read_lock(); | |
2005 | ||
2006 | /* This should be our last sync, the src is now paused */ | |
eb859c53 | 2007 | migration_bitmap_sync(rs); |
e0b266f0 | 2008 | |
6b6712ef JQ |
2009 | /* Easiest way to make sure we don't resume in the middle of a host-page */ |
2010 | rs->last_seen_block = NULL; | |
2011 | rs->last_sent_block = NULL; | |
2012 | rs->last_page = 0; | |
e0b266f0 | 2013 | |
6b6712ef JQ |
2014 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { |
2015 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; | |
2016 | unsigned long *bitmap = block->bmap; | |
2017 | unsigned long *unsentmap = block->unsentmap; | |
2018 | ||
2019 | if (!unsentmap) { | |
2020 | /* We don't have a safe way to resize the sentmap, so | |
2021 | * if the bitmap was resized it will be NULL at this | |
2022 | * point. | |
2023 | */ | |
2024 | error_report("migration ram resized during precopy phase"); | |
2025 | rcu_read_unlock(); | |
2026 | return -EINVAL; | |
2027 | } | |
2028 | /* Deal with TPS != HPS and huge pages */ | |
2029 | ret = postcopy_chunk_hostpages(ms, block); | |
2030 | if (ret) { | |
2031 | rcu_read_unlock(); | |
2032 | return ret; | |
2033 | } | |
e0b266f0 | 2034 | |
6b6712ef JQ |
2035 | /* |
2036 | * Update the unsentmap to be unsentmap = unsentmap | dirty | |
2037 | */ | |
2038 | bitmap_or(unsentmap, unsentmap, bitmap, pages); | |
e0b266f0 | 2039 | #ifdef DEBUG_POSTCOPY |
6b6712ef | 2040 | ram_debug_dump_bitmap(unsentmap, true, pages); |
e0b266f0 | 2041 | #endif |
6b6712ef JQ |
2042 | } |
2043 | trace_ram_postcopy_send_discard_bitmap(); | |
e0b266f0 DDAG |
2044 | |
2045 | ret = postcopy_each_ram_send_discard(ms); | |
2046 | rcu_read_unlock(); | |
2047 | ||
2048 | return ret; | |
2049 | } | |
2050 | ||
3d0684b2 JQ |
2051 | /** |
2052 | * ram_discard_range: discard dirtied pages at the beginning of postcopy | |
e0b266f0 | 2053 | * |
3d0684b2 | 2054 | * Returns zero on success |
e0b266f0 | 2055 | * |
36449157 JQ |
2056 | * @rbname: name of the RAMBlock of the request. NULL means the |
2057 | * same that last one. | |
3d0684b2 JQ |
2058 | * @start: RAMBlock starting page |
2059 | * @length: RAMBlock size | |
e0b266f0 | 2060 | */ |
aaa2064c | 2061 | int ram_discard_range(const char *rbname, uint64_t start, size_t length) |
e0b266f0 DDAG |
2062 | { |
2063 | int ret = -1; | |
2064 | ||
36449157 | 2065 | trace_ram_discard_range(rbname, start, length); |
d3a5038c | 2066 | |
e0b266f0 | 2067 | rcu_read_lock(); |
36449157 | 2068 | RAMBlock *rb = qemu_ram_block_by_name(rbname); |
e0b266f0 DDAG |
2069 | |
2070 | if (!rb) { | |
36449157 | 2071 | error_report("ram_discard_range: Failed to find block '%s'", rbname); |
e0b266f0 DDAG |
2072 | goto err; |
2073 | } | |
2074 | ||
f9494614 AP |
2075 | bitmap_clear(rb->receivedmap, start >> qemu_target_page_bits(), |
2076 | length >> qemu_target_page_bits()); | |
d3a5038c | 2077 | ret = ram_block_discard_range(rb, start, length); |
e0b266f0 DDAG |
2078 | |
2079 | err: | |
2080 | rcu_read_unlock(); | |
2081 | ||
2082 | return ret; | |
2083 | } | |
2084 | ||
84593a08 PX |
2085 | /* |
2086 | * For every allocation, we will try not to crash the VM if the | |
2087 | * allocation failed. | |
2088 | */ | |
2089 | static int xbzrle_init(void) | |
2090 | { | |
2091 | Error *local_err = NULL; | |
2092 | ||
2093 | if (!migrate_use_xbzrle()) { | |
2094 | return 0; | |
2095 | } | |
2096 | ||
2097 | XBZRLE_cache_lock(); | |
2098 | ||
2099 | XBZRLE.zero_target_page = g_try_malloc0(TARGET_PAGE_SIZE); | |
2100 | if (!XBZRLE.zero_target_page) { | |
2101 | error_report("%s: Error allocating zero page", __func__); | |
2102 | goto err_out; | |
2103 | } | |
2104 | ||
2105 | XBZRLE.cache = cache_init(migrate_xbzrle_cache_size(), | |
2106 | TARGET_PAGE_SIZE, &local_err); | |
2107 | if (!XBZRLE.cache) { | |
2108 | error_report_err(local_err); | |
2109 | goto free_zero_page; | |
2110 | } | |
2111 | ||
2112 | XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); | |
2113 | if (!XBZRLE.encoded_buf) { | |
2114 | error_report("%s: Error allocating encoded_buf", __func__); | |
2115 | goto free_cache; | |
2116 | } | |
2117 | ||
2118 | XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); | |
2119 | if (!XBZRLE.current_buf) { | |
2120 | error_report("%s: Error allocating current_buf", __func__); | |
2121 | goto free_encoded_buf; | |
2122 | } | |
2123 | ||
2124 | /* We are all good */ | |
2125 | XBZRLE_cache_unlock(); | |
2126 | return 0; | |
2127 | ||
2128 | free_encoded_buf: | |
2129 | g_free(XBZRLE.encoded_buf); | |
2130 | XBZRLE.encoded_buf = NULL; | |
2131 | free_cache: | |
2132 | cache_fini(XBZRLE.cache); | |
2133 | XBZRLE.cache = NULL; | |
2134 | free_zero_page: | |
2135 | g_free(XBZRLE.zero_target_page); | |
2136 | XBZRLE.zero_target_page = NULL; | |
2137 | err_out: | |
2138 | XBZRLE_cache_unlock(); | |
2139 | return -ENOMEM; | |
2140 | } | |
2141 | ||
53518d94 | 2142 | static int ram_state_init(RAMState **rsp) |
56e93d26 | 2143 | { |
7d00ee6a PX |
2144 | *rsp = g_try_new0(RAMState, 1); |
2145 | ||
2146 | if (!*rsp) { | |
2147 | error_report("%s: Init ramstate fail", __func__); | |
2148 | return -1; | |
2149 | } | |
53518d94 JQ |
2150 | |
2151 | qemu_mutex_init(&(*rsp)->bitmap_mutex); | |
2152 | qemu_mutex_init(&(*rsp)->src_page_req_mutex); | |
2153 | QSIMPLEQ_INIT(&(*rsp)->src_page_requests); | |
56e93d26 | 2154 | |
7d00ee6a PX |
2155 | /* |
2156 | * Count the total number of pages used by ram blocks not including any | |
2157 | * gaps due to alignment or unplugs. | |
2158 | */ | |
2159 | (*rsp)->migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS; | |
2160 | ||
2161 | ram_state_reset(*rsp); | |
2162 | ||
2163 | return 0; | |
2164 | } | |
2165 | ||
d6eff5d7 | 2166 | static void ram_list_init_bitmaps(void) |
7d00ee6a | 2167 | { |
d6eff5d7 PX |
2168 | RAMBlock *block; |
2169 | unsigned long pages; | |
56e93d26 | 2170 | |
0827b9e9 AA |
2171 | /* Skip setting bitmap if there is no RAM */ |
2172 | if (ram_bytes_total()) { | |
6b6712ef | 2173 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { |
d6eff5d7 | 2174 | pages = block->max_length >> TARGET_PAGE_BITS; |
6b6712ef JQ |
2175 | block->bmap = bitmap_new(pages); |
2176 | bitmap_set(block->bmap, 0, pages); | |
2177 | if (migrate_postcopy_ram()) { | |
2178 | block->unsentmap = bitmap_new(pages); | |
2179 | bitmap_set(block->unsentmap, 0, pages); | |
2180 | } | |
0827b9e9 | 2181 | } |
f3f491fc | 2182 | } |
d6eff5d7 PX |
2183 | } |
2184 | ||
2185 | static void ram_init_bitmaps(RAMState *rs) | |
2186 | { | |
2187 | /* For memory_global_dirty_log_start below. */ | |
2188 | qemu_mutex_lock_iothread(); | |
2189 | qemu_mutex_lock_ramlist(); | |
2190 | rcu_read_lock(); | |
f3f491fc | 2191 | |
d6eff5d7 | 2192 | ram_list_init_bitmaps(); |
56e93d26 | 2193 | memory_global_dirty_log_start(); |
d6eff5d7 PX |
2194 | migration_bitmap_sync(rs); |
2195 | ||
2196 | rcu_read_unlock(); | |
56e93d26 | 2197 | qemu_mutex_unlock_ramlist(); |
49877834 | 2198 | qemu_mutex_unlock_iothread(); |
d6eff5d7 PX |
2199 | } |
2200 | ||
2201 | static int ram_init_all(RAMState **rsp) | |
2202 | { | |
2203 | if (ram_state_init(rsp)) { | |
2204 | return -1; | |
2205 | } | |
2206 | ||
2207 | if (xbzrle_init()) { | |
2208 | ram_state_cleanup(rsp); | |
2209 | return -1; | |
2210 | } | |
2211 | ||
2212 | ram_init_bitmaps(*rsp); | |
a91246c9 HZ |
2213 | |
2214 | return 0; | |
2215 | } | |
2216 | ||
3d0684b2 JQ |
2217 | /* |
2218 | * Each of ram_save_setup, ram_save_iterate and ram_save_complete has | |
a91246c9 HZ |
2219 | * long-running RCU critical section. When rcu-reclaims in the code |
2220 | * start to become numerous it will be necessary to reduce the | |
2221 | * granularity of these critical sections. | |
2222 | */ | |
2223 | ||
3d0684b2 JQ |
2224 | /** |
2225 | * ram_save_setup: Setup RAM for migration | |
2226 | * | |
2227 | * Returns zero to indicate success and negative for error | |
2228 | * | |
2229 | * @f: QEMUFile where to send the data | |
2230 | * @opaque: RAMState pointer | |
2231 | */ | |
a91246c9 HZ |
2232 | static int ram_save_setup(QEMUFile *f, void *opaque) |
2233 | { | |
53518d94 | 2234 | RAMState **rsp = opaque; |
a91246c9 HZ |
2235 | RAMBlock *block; |
2236 | ||
dcaf446e XG |
2237 | if (compress_threads_save_setup()) { |
2238 | return -1; | |
2239 | } | |
2240 | ||
a91246c9 HZ |
2241 | /* migration has already setup the bitmap, reuse it. */ |
2242 | if (!migration_in_colo_state()) { | |
7d00ee6a | 2243 | if (ram_init_all(rsp) != 0) { |
dcaf446e | 2244 | compress_threads_save_cleanup(); |
a91246c9 | 2245 | return -1; |
53518d94 | 2246 | } |
a91246c9 | 2247 | } |
53518d94 | 2248 | (*rsp)->f = f; |
a91246c9 HZ |
2249 | |
2250 | rcu_read_lock(); | |
56e93d26 JQ |
2251 | |
2252 | qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE); | |
2253 | ||
99e15582 | 2254 | RAMBLOCK_FOREACH(block) { |
56e93d26 JQ |
2255 | qemu_put_byte(f, strlen(block->idstr)); |
2256 | qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); | |
2257 | qemu_put_be64(f, block->used_length); | |
ef08fb38 DDAG |
2258 | if (migrate_postcopy_ram() && block->page_size != qemu_host_page_size) { |
2259 | qemu_put_be64(f, block->page_size); | |
2260 | } | |
56e93d26 JQ |
2261 | } |
2262 | ||
2263 | rcu_read_unlock(); | |
2264 | ||
2265 | ram_control_before_iterate(f, RAM_CONTROL_SETUP); | |
2266 | ram_control_after_iterate(f, RAM_CONTROL_SETUP); | |
2267 | ||
2268 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); | |
2269 | ||
2270 | return 0; | |
2271 | } | |
2272 | ||
3d0684b2 JQ |
2273 | /** |
2274 | * ram_save_iterate: iterative stage for migration | |
2275 | * | |
2276 | * Returns zero to indicate success and negative for error | |
2277 | * | |
2278 | * @f: QEMUFile where to send the data | |
2279 | * @opaque: RAMState pointer | |
2280 | */ | |
56e93d26 JQ |
2281 | static int ram_save_iterate(QEMUFile *f, void *opaque) |
2282 | { | |
53518d94 JQ |
2283 | RAMState **temp = opaque; |
2284 | RAMState *rs = *temp; | |
56e93d26 JQ |
2285 | int ret; |
2286 | int i; | |
2287 | int64_t t0; | |
5c90308f | 2288 | int done = 0; |
56e93d26 | 2289 | |
b2557345 PL |
2290 | if (blk_mig_bulk_active()) { |
2291 | /* Avoid transferring ram during bulk phase of block migration as | |
2292 | * the bulk phase will usually take a long time and transferring | |
2293 | * ram updates during that time is pointless. */ | |
2294 | goto out; | |
2295 | } | |
2296 | ||
56e93d26 | 2297 | rcu_read_lock(); |
6f37bb8b JQ |
2298 | if (ram_list.version != rs->last_version) { |
2299 | ram_state_reset(rs); | |
56e93d26 JQ |
2300 | } |
2301 | ||
2302 | /* Read version before ram_list.blocks */ | |
2303 | smp_rmb(); | |
2304 | ||
2305 | ram_control_before_iterate(f, RAM_CONTROL_ROUND); | |
2306 | ||
2307 | t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
2308 | i = 0; | |
2309 | while ((ret = qemu_file_rate_limit(f)) == 0) { | |
2310 | int pages; | |
2311 | ||
ce25d337 | 2312 | pages = ram_find_and_save_block(rs, false); |
56e93d26 JQ |
2313 | /* no more pages to sent */ |
2314 | if (pages == 0) { | |
5c90308f | 2315 | done = 1; |
56e93d26 JQ |
2316 | break; |
2317 | } | |
23b28c3c | 2318 | rs->iterations++; |
070afca2 | 2319 | |
56e93d26 JQ |
2320 | /* we want to check in the 1st loop, just in case it was the 1st time |
2321 | and we had to sync the dirty bitmap. | |
2322 | qemu_get_clock_ns() is a bit expensive, so we only check each some | |
2323 | iterations | |
2324 | */ | |
2325 | if ((i & 63) == 0) { | |
2326 | uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000; | |
2327 | if (t1 > MAX_WAIT) { | |
55c4446b | 2328 | trace_ram_save_iterate_big_wait(t1, i); |
56e93d26 JQ |
2329 | break; |
2330 | } | |
2331 | } | |
2332 | i++; | |
2333 | } | |
ce25d337 | 2334 | flush_compressed_data(rs); |
56e93d26 JQ |
2335 | rcu_read_unlock(); |
2336 | ||
2337 | /* | |
2338 | * Must occur before EOS (or any QEMUFile operation) | |
2339 | * because of RDMA protocol. | |
2340 | */ | |
2341 | ram_control_after_iterate(f, RAM_CONTROL_ROUND); | |
2342 | ||
b2557345 | 2343 | out: |
56e93d26 | 2344 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
9360447d | 2345 | ram_counters.transferred += 8; |
56e93d26 JQ |
2346 | |
2347 | ret = qemu_file_get_error(f); | |
2348 | if (ret < 0) { | |
2349 | return ret; | |
2350 | } | |
2351 | ||
5c90308f | 2352 | return done; |
56e93d26 JQ |
2353 | } |
2354 | ||
3d0684b2 JQ |
2355 | /** |
2356 | * ram_save_complete: function called to send the remaining amount of ram | |
2357 | * | |
2358 | * Returns zero to indicate success | |
2359 | * | |
2360 | * Called with iothread lock | |
2361 | * | |
2362 | * @f: QEMUFile where to send the data | |
2363 | * @opaque: RAMState pointer | |
2364 | */ | |
56e93d26 JQ |
2365 | static int ram_save_complete(QEMUFile *f, void *opaque) |
2366 | { | |
53518d94 JQ |
2367 | RAMState **temp = opaque; |
2368 | RAMState *rs = *temp; | |
6f37bb8b | 2369 | |
56e93d26 JQ |
2370 | rcu_read_lock(); |
2371 | ||
5727309d | 2372 | if (!migration_in_postcopy()) { |
8d820d6f | 2373 | migration_bitmap_sync(rs); |
663e6c1d | 2374 | } |
56e93d26 JQ |
2375 | |
2376 | ram_control_before_iterate(f, RAM_CONTROL_FINISH); | |
2377 | ||
2378 | /* try transferring iterative blocks of memory */ | |
2379 | ||
2380 | /* flush all remaining blocks regardless of rate limiting */ | |
2381 | while (true) { | |
2382 | int pages; | |
2383 | ||
ce25d337 | 2384 | pages = ram_find_and_save_block(rs, !migration_in_colo_state()); |
56e93d26 JQ |
2385 | /* no more blocks to sent */ |
2386 | if (pages == 0) { | |
2387 | break; | |
2388 | } | |
2389 | } | |
2390 | ||
ce25d337 | 2391 | flush_compressed_data(rs); |
56e93d26 | 2392 | ram_control_after_iterate(f, RAM_CONTROL_FINISH); |
56e93d26 JQ |
2393 | |
2394 | rcu_read_unlock(); | |
d09a6fde | 2395 | |
56e93d26 JQ |
2396 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
2397 | ||
2398 | return 0; | |
2399 | } | |
2400 | ||
c31b098f | 2401 | static void ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size, |
47995026 VSO |
2402 | uint64_t *res_precopy_only, |
2403 | uint64_t *res_compatible, | |
2404 | uint64_t *res_postcopy_only) | |
56e93d26 | 2405 | { |
53518d94 JQ |
2406 | RAMState **temp = opaque; |
2407 | RAMState *rs = *temp; | |
56e93d26 JQ |
2408 | uint64_t remaining_size; |
2409 | ||
9edabd4d | 2410 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2411 | |
5727309d | 2412 | if (!migration_in_postcopy() && |
663e6c1d | 2413 | remaining_size < max_size) { |
56e93d26 JQ |
2414 | qemu_mutex_lock_iothread(); |
2415 | rcu_read_lock(); | |
8d820d6f | 2416 | migration_bitmap_sync(rs); |
56e93d26 JQ |
2417 | rcu_read_unlock(); |
2418 | qemu_mutex_unlock_iothread(); | |
9edabd4d | 2419 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2420 | } |
c31b098f | 2421 | |
86e1167e VSO |
2422 | if (migrate_postcopy_ram()) { |
2423 | /* We can do postcopy, and all the data is postcopiable */ | |
47995026 | 2424 | *res_compatible += remaining_size; |
86e1167e | 2425 | } else { |
47995026 | 2426 | *res_precopy_only += remaining_size; |
86e1167e | 2427 | } |
56e93d26 JQ |
2428 | } |
2429 | ||
2430 | static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) | |
2431 | { | |
2432 | unsigned int xh_len; | |
2433 | int xh_flags; | |
063e760a | 2434 | uint8_t *loaded_data; |
56e93d26 | 2435 | |
56e93d26 JQ |
2436 | /* extract RLE header */ |
2437 | xh_flags = qemu_get_byte(f); | |
2438 | xh_len = qemu_get_be16(f); | |
2439 | ||
2440 | if (xh_flags != ENCODING_FLAG_XBZRLE) { | |
2441 | error_report("Failed to load XBZRLE page - wrong compression!"); | |
2442 | return -1; | |
2443 | } | |
2444 | ||
2445 | if (xh_len > TARGET_PAGE_SIZE) { | |
2446 | error_report("Failed to load XBZRLE page - len overflow!"); | |
2447 | return -1; | |
2448 | } | |
f265e0e4 | 2449 | loaded_data = XBZRLE.decoded_buf; |
56e93d26 | 2450 | /* load data and decode */ |
f265e0e4 | 2451 | /* it can change loaded_data to point to an internal buffer */ |
063e760a | 2452 | qemu_get_buffer_in_place(f, &loaded_data, xh_len); |
56e93d26 JQ |
2453 | |
2454 | /* decode RLE */ | |
063e760a | 2455 | if (xbzrle_decode_buffer(loaded_data, xh_len, host, |
56e93d26 JQ |
2456 | TARGET_PAGE_SIZE) == -1) { |
2457 | error_report("Failed to load XBZRLE page - decode error!"); | |
2458 | return -1; | |
2459 | } | |
2460 | ||
2461 | return 0; | |
2462 | } | |
2463 | ||
3d0684b2 JQ |
2464 | /** |
2465 | * ram_block_from_stream: read a RAMBlock id from the migration stream | |
2466 | * | |
2467 | * Must be called from within a rcu critical section. | |
2468 | * | |
56e93d26 | 2469 | * Returns a pointer from within the RCU-protected ram_list. |
a7180877 | 2470 | * |
3d0684b2 JQ |
2471 | * @f: QEMUFile where to read the data from |
2472 | * @flags: Page flags (mostly to see if it's a continuation of previous block) | |
a7180877 | 2473 | */ |
3d0684b2 | 2474 | static inline RAMBlock *ram_block_from_stream(QEMUFile *f, int flags) |
56e93d26 JQ |
2475 | { |
2476 | static RAMBlock *block = NULL; | |
2477 | char id[256]; | |
2478 | uint8_t len; | |
2479 | ||
2480 | if (flags & RAM_SAVE_FLAG_CONTINUE) { | |
4c4bad48 | 2481 | if (!block) { |
56e93d26 JQ |
2482 | error_report("Ack, bad migration stream!"); |
2483 | return NULL; | |
2484 | } | |
4c4bad48 | 2485 | return block; |
56e93d26 JQ |
2486 | } |
2487 | ||
2488 | len = qemu_get_byte(f); | |
2489 | qemu_get_buffer(f, (uint8_t *)id, len); | |
2490 | id[len] = 0; | |
2491 | ||
e3dd7493 | 2492 | block = qemu_ram_block_by_name(id); |
4c4bad48 HZ |
2493 | if (!block) { |
2494 | error_report("Can't find block %s", id); | |
2495 | return NULL; | |
56e93d26 JQ |
2496 | } |
2497 | ||
4c4bad48 HZ |
2498 | return block; |
2499 | } | |
2500 | ||
2501 | static inline void *host_from_ram_block_offset(RAMBlock *block, | |
2502 | ram_addr_t offset) | |
2503 | { | |
2504 | if (!offset_in_ramblock(block, offset)) { | |
2505 | return NULL; | |
2506 | } | |
2507 | ||
2508 | return block->host + offset; | |
56e93d26 JQ |
2509 | } |
2510 | ||
3d0684b2 JQ |
2511 | /** |
2512 | * ram_handle_compressed: handle the zero page case | |
2513 | * | |
56e93d26 JQ |
2514 | * If a page (or a whole RDMA chunk) has been |
2515 | * determined to be zero, then zap it. | |
3d0684b2 JQ |
2516 | * |
2517 | * @host: host address for the zero page | |
2518 | * @ch: what the page is filled from. We only support zero | |
2519 | * @size: size of the zero page | |
56e93d26 JQ |
2520 | */ |
2521 | void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) | |
2522 | { | |
2523 | if (ch != 0 || !is_zero_range(host, size)) { | |
2524 | memset(host, ch, size); | |
2525 | } | |
2526 | } | |
2527 | ||
797ca154 XG |
2528 | /* return the size after decompression, or negative value on error */ |
2529 | static int | |
2530 | qemu_uncompress_data(z_stream *stream, uint8_t *dest, size_t dest_len, | |
2531 | const uint8_t *source, size_t source_len) | |
2532 | { | |
2533 | int err; | |
2534 | ||
2535 | err = inflateReset(stream); | |
2536 | if (err != Z_OK) { | |
2537 | return -1; | |
2538 | } | |
2539 | ||
2540 | stream->avail_in = source_len; | |
2541 | stream->next_in = (uint8_t *)source; | |
2542 | stream->avail_out = dest_len; | |
2543 | stream->next_out = dest; | |
2544 | ||
2545 | err = inflate(stream, Z_NO_FLUSH); | |
2546 | if (err != Z_STREAM_END) { | |
2547 | return -1; | |
2548 | } | |
2549 | ||
2550 | return stream->total_out; | |
2551 | } | |
2552 | ||
56e93d26 JQ |
2553 | static void *do_data_decompress(void *opaque) |
2554 | { | |
2555 | DecompressParam *param = opaque; | |
2556 | unsigned long pagesize; | |
33d151f4 LL |
2557 | uint8_t *des; |
2558 | int len; | |
56e93d26 | 2559 | |
33d151f4 | 2560 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 2561 | while (!param->quit) { |
33d151f4 LL |
2562 | if (param->des) { |
2563 | des = param->des; | |
2564 | len = param->len; | |
2565 | param->des = 0; | |
2566 | qemu_mutex_unlock(¶m->mutex); | |
2567 | ||
56e93d26 | 2568 | pagesize = TARGET_PAGE_SIZE; |
797ca154 XG |
2569 | /* qemu_uncompress_data() will return failed in some case, |
2570 | * especially when the page is dirtied when doing the compression, | |
2571 | * it's not a problem because the dirty page will be retransferred | |
73a8912b LL |
2572 | * and uncompress() won't break the data in other pages. |
2573 | */ | |
797ca154 XG |
2574 | qemu_uncompress_data(¶m->stream, des, pagesize, param->compbuf, |
2575 | len); | |
73a8912b | 2576 | |
33d151f4 LL |
2577 | qemu_mutex_lock(&decomp_done_lock); |
2578 | param->done = true; | |
2579 | qemu_cond_signal(&decomp_done_cond); | |
2580 | qemu_mutex_unlock(&decomp_done_lock); | |
2581 | ||
2582 | qemu_mutex_lock(¶m->mutex); | |
2583 | } else { | |
2584 | qemu_cond_wait(¶m->cond, ¶m->mutex); | |
2585 | } | |
56e93d26 | 2586 | } |
33d151f4 | 2587 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
2588 | |
2589 | return NULL; | |
2590 | } | |
2591 | ||
5533b2e9 LL |
2592 | static void wait_for_decompress_done(void) |
2593 | { | |
2594 | int idx, thread_count; | |
2595 | ||
2596 | if (!migrate_use_compression()) { | |
2597 | return; | |
2598 | } | |
2599 | ||
2600 | thread_count = migrate_decompress_threads(); | |
2601 | qemu_mutex_lock(&decomp_done_lock); | |
2602 | for (idx = 0; idx < thread_count; idx++) { | |
2603 | while (!decomp_param[idx].done) { | |
2604 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
2605 | } | |
2606 | } | |
2607 | qemu_mutex_unlock(&decomp_done_lock); | |
2608 | } | |
2609 | ||
f0afa331 | 2610 | static void compress_threads_load_cleanup(void) |
56e93d26 JQ |
2611 | { |
2612 | int i, thread_count; | |
2613 | ||
3416ab5b JQ |
2614 | if (!migrate_use_compression()) { |
2615 | return; | |
2616 | } | |
56e93d26 JQ |
2617 | thread_count = migrate_decompress_threads(); |
2618 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
2619 | /* |
2620 | * we use it as a indicator which shows if the thread is | |
2621 | * properly init'd or not | |
2622 | */ | |
2623 | if (!decomp_param[i].compbuf) { | |
2624 | break; | |
2625 | } | |
2626 | ||
56e93d26 | 2627 | qemu_mutex_lock(&decomp_param[i].mutex); |
90e56fb4 | 2628 | decomp_param[i].quit = true; |
56e93d26 JQ |
2629 | qemu_cond_signal(&decomp_param[i].cond); |
2630 | qemu_mutex_unlock(&decomp_param[i].mutex); | |
2631 | } | |
2632 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
2633 | if (!decomp_param[i].compbuf) { |
2634 | break; | |
2635 | } | |
2636 | ||
56e93d26 JQ |
2637 | qemu_thread_join(decompress_threads + i); |
2638 | qemu_mutex_destroy(&decomp_param[i].mutex); | |
2639 | qemu_cond_destroy(&decomp_param[i].cond); | |
797ca154 | 2640 | inflateEnd(&decomp_param[i].stream); |
56e93d26 | 2641 | g_free(decomp_param[i].compbuf); |
797ca154 | 2642 | decomp_param[i].compbuf = NULL; |
56e93d26 JQ |
2643 | } |
2644 | g_free(decompress_threads); | |
2645 | g_free(decomp_param); | |
56e93d26 JQ |
2646 | decompress_threads = NULL; |
2647 | decomp_param = NULL; | |
56e93d26 JQ |
2648 | } |
2649 | ||
797ca154 XG |
2650 | static int compress_threads_load_setup(void) |
2651 | { | |
2652 | int i, thread_count; | |
2653 | ||
2654 | if (!migrate_use_compression()) { | |
2655 | return 0; | |
2656 | } | |
2657 | ||
2658 | thread_count = migrate_decompress_threads(); | |
2659 | decompress_threads = g_new0(QemuThread, thread_count); | |
2660 | decomp_param = g_new0(DecompressParam, thread_count); | |
2661 | qemu_mutex_init(&decomp_done_lock); | |
2662 | qemu_cond_init(&decomp_done_cond); | |
2663 | for (i = 0; i < thread_count; i++) { | |
2664 | if (inflateInit(&decomp_param[i].stream) != Z_OK) { | |
2665 | goto exit; | |
2666 | } | |
2667 | ||
2668 | decomp_param[i].compbuf = g_malloc0(compressBound(TARGET_PAGE_SIZE)); | |
2669 | qemu_mutex_init(&decomp_param[i].mutex); | |
2670 | qemu_cond_init(&decomp_param[i].cond); | |
2671 | decomp_param[i].done = true; | |
2672 | decomp_param[i].quit = false; | |
2673 | qemu_thread_create(decompress_threads + i, "decompress", | |
2674 | do_data_decompress, decomp_param + i, | |
2675 | QEMU_THREAD_JOINABLE); | |
2676 | } | |
2677 | return 0; | |
2678 | exit: | |
2679 | compress_threads_load_cleanup(); | |
2680 | return -1; | |
2681 | } | |
2682 | ||
c1bc6626 | 2683 | static void decompress_data_with_multi_threads(QEMUFile *f, |
56e93d26 JQ |
2684 | void *host, int len) |
2685 | { | |
2686 | int idx, thread_count; | |
2687 | ||
2688 | thread_count = migrate_decompress_threads(); | |
73a8912b | 2689 | qemu_mutex_lock(&decomp_done_lock); |
56e93d26 JQ |
2690 | while (true) { |
2691 | for (idx = 0; idx < thread_count; idx++) { | |
73a8912b | 2692 | if (decomp_param[idx].done) { |
33d151f4 LL |
2693 | decomp_param[idx].done = false; |
2694 | qemu_mutex_lock(&decomp_param[idx].mutex); | |
c1bc6626 | 2695 | qemu_get_buffer(f, decomp_param[idx].compbuf, len); |
56e93d26 JQ |
2696 | decomp_param[idx].des = host; |
2697 | decomp_param[idx].len = len; | |
33d151f4 LL |
2698 | qemu_cond_signal(&decomp_param[idx].cond); |
2699 | qemu_mutex_unlock(&decomp_param[idx].mutex); | |
56e93d26 JQ |
2700 | break; |
2701 | } | |
2702 | } | |
2703 | if (idx < thread_count) { | |
2704 | break; | |
73a8912b LL |
2705 | } else { |
2706 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
56e93d26 JQ |
2707 | } |
2708 | } | |
73a8912b | 2709 | qemu_mutex_unlock(&decomp_done_lock); |
56e93d26 JQ |
2710 | } |
2711 | ||
f265e0e4 JQ |
2712 | /** |
2713 | * ram_load_setup: Setup RAM for migration incoming side | |
2714 | * | |
2715 | * Returns zero to indicate success and negative for error | |
2716 | * | |
2717 | * @f: QEMUFile where to receive the data | |
2718 | * @opaque: RAMState pointer | |
2719 | */ | |
2720 | static int ram_load_setup(QEMUFile *f, void *opaque) | |
2721 | { | |
797ca154 XG |
2722 | if (compress_threads_load_setup()) { |
2723 | return -1; | |
2724 | } | |
2725 | ||
f265e0e4 | 2726 | xbzrle_load_setup(); |
f9494614 | 2727 | ramblock_recv_map_init(); |
f265e0e4 JQ |
2728 | return 0; |
2729 | } | |
2730 | ||
2731 | static int ram_load_cleanup(void *opaque) | |
2732 | { | |
f9494614 | 2733 | RAMBlock *rb; |
f265e0e4 | 2734 | xbzrle_load_cleanup(); |
f0afa331 | 2735 | compress_threads_load_cleanup(); |
f9494614 AP |
2736 | |
2737 | RAMBLOCK_FOREACH(rb) { | |
2738 | g_free(rb->receivedmap); | |
2739 | rb->receivedmap = NULL; | |
2740 | } | |
f265e0e4 JQ |
2741 | return 0; |
2742 | } | |
2743 | ||
3d0684b2 JQ |
2744 | /** |
2745 | * ram_postcopy_incoming_init: allocate postcopy data structures | |
2746 | * | |
2747 | * Returns 0 for success and negative if there was one error | |
2748 | * | |
2749 | * @mis: current migration incoming state | |
2750 | * | |
2751 | * Allocate data structures etc needed by incoming migration with | |
2752 | * postcopy-ram. postcopy-ram's similarly names | |
2753 | * postcopy_ram_incoming_init does the work. | |
1caddf8a DDAG |
2754 | */ |
2755 | int ram_postcopy_incoming_init(MigrationIncomingState *mis) | |
2756 | { | |
b8c48993 | 2757 | unsigned long ram_pages = last_ram_page(); |
1caddf8a DDAG |
2758 | |
2759 | return postcopy_ram_incoming_init(mis, ram_pages); | |
2760 | } | |
2761 | ||
3d0684b2 JQ |
2762 | /** |
2763 | * ram_load_postcopy: load a page in postcopy case | |
2764 | * | |
2765 | * Returns 0 for success or -errno in case of error | |
2766 | * | |
a7180877 DDAG |
2767 | * Called in postcopy mode by ram_load(). |
2768 | * rcu_read_lock is taken prior to this being called. | |
3d0684b2 JQ |
2769 | * |
2770 | * @f: QEMUFile where to send the data | |
a7180877 DDAG |
2771 | */ |
2772 | static int ram_load_postcopy(QEMUFile *f) | |
2773 | { | |
2774 | int flags = 0, ret = 0; | |
2775 | bool place_needed = false; | |
28abd200 | 2776 | bool matching_page_sizes = false; |
a7180877 DDAG |
2777 | MigrationIncomingState *mis = migration_incoming_get_current(); |
2778 | /* Temporary page that is later 'placed' */ | |
2779 | void *postcopy_host_page = postcopy_get_tmp_page(mis); | |
c53b7ddc | 2780 | void *last_host = NULL; |
a3b6ff6d | 2781 | bool all_zero = false; |
a7180877 DDAG |
2782 | |
2783 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
2784 | ram_addr_t addr; | |
2785 | void *host = NULL; | |
2786 | void *page_buffer = NULL; | |
2787 | void *place_source = NULL; | |
df9ff5e1 | 2788 | RAMBlock *block = NULL; |
a7180877 | 2789 | uint8_t ch; |
a7180877 DDAG |
2790 | |
2791 | addr = qemu_get_be64(f); | |
7a9ddfbf PX |
2792 | |
2793 | /* | |
2794 | * If qemu file error, we should stop here, and then "addr" | |
2795 | * may be invalid | |
2796 | */ | |
2797 | ret = qemu_file_get_error(f); | |
2798 | if (ret) { | |
2799 | break; | |
2800 | } | |
2801 | ||
a7180877 DDAG |
2802 | flags = addr & ~TARGET_PAGE_MASK; |
2803 | addr &= TARGET_PAGE_MASK; | |
2804 | ||
2805 | trace_ram_load_postcopy_loop((uint64_t)addr, flags); | |
2806 | place_needed = false; | |
bb890ed5 | 2807 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE)) { |
df9ff5e1 | 2808 | block = ram_block_from_stream(f, flags); |
4c4bad48 HZ |
2809 | |
2810 | host = host_from_ram_block_offset(block, addr); | |
a7180877 DDAG |
2811 | if (!host) { |
2812 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
2813 | ret = -EINVAL; | |
2814 | break; | |
2815 | } | |
28abd200 | 2816 | matching_page_sizes = block->page_size == TARGET_PAGE_SIZE; |
a7180877 | 2817 | /* |
28abd200 DDAG |
2818 | * Postcopy requires that we place whole host pages atomically; |
2819 | * these may be huge pages for RAMBlocks that are backed by | |
2820 | * hugetlbfs. | |
a7180877 DDAG |
2821 | * To make it atomic, the data is read into a temporary page |
2822 | * that's moved into place later. | |
2823 | * The migration protocol uses, possibly smaller, target-pages | |
2824 | * however the source ensures it always sends all the components | |
2825 | * of a host page in order. | |
2826 | */ | |
2827 | page_buffer = postcopy_host_page + | |
28abd200 | 2828 | ((uintptr_t)host & (block->page_size - 1)); |
a7180877 | 2829 | /* If all TP are zero then we can optimise the place */ |
28abd200 | 2830 | if (!((uintptr_t)host & (block->page_size - 1))) { |
a7180877 | 2831 | all_zero = true; |
c53b7ddc DDAG |
2832 | } else { |
2833 | /* not the 1st TP within the HP */ | |
2834 | if (host != (last_host + TARGET_PAGE_SIZE)) { | |
9af9e0fe | 2835 | error_report("Non-sequential target page %p/%p", |
c53b7ddc DDAG |
2836 | host, last_host); |
2837 | ret = -EINVAL; | |
2838 | break; | |
2839 | } | |
a7180877 DDAG |
2840 | } |
2841 | ||
c53b7ddc | 2842 | |
a7180877 DDAG |
2843 | /* |
2844 | * If it's the last part of a host page then we place the host | |
2845 | * page | |
2846 | */ | |
2847 | place_needed = (((uintptr_t)host + TARGET_PAGE_SIZE) & | |
28abd200 | 2848 | (block->page_size - 1)) == 0; |
a7180877 DDAG |
2849 | place_source = postcopy_host_page; |
2850 | } | |
c53b7ddc | 2851 | last_host = host; |
a7180877 DDAG |
2852 | |
2853 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { | |
bb890ed5 | 2854 | case RAM_SAVE_FLAG_ZERO: |
a7180877 DDAG |
2855 | ch = qemu_get_byte(f); |
2856 | memset(page_buffer, ch, TARGET_PAGE_SIZE); | |
2857 | if (ch) { | |
2858 | all_zero = false; | |
2859 | } | |
2860 | break; | |
2861 | ||
2862 | case RAM_SAVE_FLAG_PAGE: | |
2863 | all_zero = false; | |
2864 | if (!place_needed || !matching_page_sizes) { | |
2865 | qemu_get_buffer(f, page_buffer, TARGET_PAGE_SIZE); | |
2866 | } else { | |
2867 | /* Avoids the qemu_file copy during postcopy, which is | |
2868 | * going to do a copy later; can only do it when we | |
2869 | * do this read in one go (matching page sizes) | |
2870 | */ | |
2871 | qemu_get_buffer_in_place(f, (uint8_t **)&place_source, | |
2872 | TARGET_PAGE_SIZE); | |
2873 | } | |
2874 | break; | |
2875 | case RAM_SAVE_FLAG_EOS: | |
2876 | /* normal exit */ | |
2877 | break; | |
2878 | default: | |
2879 | error_report("Unknown combination of migration flags: %#x" | |
2880 | " (postcopy mode)", flags); | |
2881 | ret = -EINVAL; | |
7a9ddfbf PX |
2882 | break; |
2883 | } | |
2884 | ||
2885 | /* Detect for any possible file errors */ | |
2886 | if (!ret && qemu_file_get_error(f)) { | |
2887 | ret = qemu_file_get_error(f); | |
a7180877 DDAG |
2888 | } |
2889 | ||
7a9ddfbf | 2890 | if (!ret && place_needed) { |
a7180877 | 2891 | /* This gets called at the last target page in the host page */ |
df9ff5e1 DDAG |
2892 | void *place_dest = host + TARGET_PAGE_SIZE - block->page_size; |
2893 | ||
a7180877 | 2894 | if (all_zero) { |
df9ff5e1 | 2895 | ret = postcopy_place_page_zero(mis, place_dest, |
8be4620b | 2896 | block); |
a7180877 | 2897 | } else { |
df9ff5e1 | 2898 | ret = postcopy_place_page(mis, place_dest, |
8be4620b | 2899 | place_source, block); |
a7180877 DDAG |
2900 | } |
2901 | } | |
a7180877 DDAG |
2902 | } |
2903 | ||
2904 | return ret; | |
2905 | } | |
2906 | ||
acab30b8 DHB |
2907 | static bool postcopy_is_advised(void) |
2908 | { | |
2909 | PostcopyState ps = postcopy_state_get(); | |
2910 | return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END; | |
2911 | } | |
2912 | ||
2913 | static bool postcopy_is_running(void) | |
2914 | { | |
2915 | PostcopyState ps = postcopy_state_get(); | |
2916 | return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; | |
2917 | } | |
2918 | ||
56e93d26 JQ |
2919 | static int ram_load(QEMUFile *f, void *opaque, int version_id) |
2920 | { | |
edc60127 | 2921 | int flags = 0, ret = 0, invalid_flags = 0; |
56e93d26 JQ |
2922 | static uint64_t seq_iter; |
2923 | int len = 0; | |
a7180877 DDAG |
2924 | /* |
2925 | * If system is running in postcopy mode, page inserts to host memory must | |
2926 | * be atomic | |
2927 | */ | |
acab30b8 | 2928 | bool postcopy_running = postcopy_is_running(); |
ef08fb38 | 2929 | /* ADVISE is earlier, it shows the source has the postcopy capability on */ |
acab30b8 | 2930 | bool postcopy_advised = postcopy_is_advised(); |
56e93d26 JQ |
2931 | |
2932 | seq_iter++; | |
2933 | ||
2934 | if (version_id != 4) { | |
2935 | ret = -EINVAL; | |
2936 | } | |
2937 | ||
edc60127 JQ |
2938 | if (!migrate_use_compression()) { |
2939 | invalid_flags |= RAM_SAVE_FLAG_COMPRESS_PAGE; | |
2940 | } | |
56e93d26 JQ |
2941 | /* This RCU critical section can be very long running. |
2942 | * When RCU reclaims in the code start to become numerous, | |
2943 | * it will be necessary to reduce the granularity of this | |
2944 | * critical section. | |
2945 | */ | |
2946 | rcu_read_lock(); | |
a7180877 DDAG |
2947 | |
2948 | if (postcopy_running) { | |
2949 | ret = ram_load_postcopy(f); | |
2950 | } | |
2951 | ||
2952 | while (!postcopy_running && !ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
56e93d26 | 2953 | ram_addr_t addr, total_ram_bytes; |
a776aa15 | 2954 | void *host = NULL; |
56e93d26 JQ |
2955 | uint8_t ch; |
2956 | ||
2957 | addr = qemu_get_be64(f); | |
2958 | flags = addr & ~TARGET_PAGE_MASK; | |
2959 | addr &= TARGET_PAGE_MASK; | |
2960 | ||
edc60127 JQ |
2961 | if (flags & invalid_flags) { |
2962 | if (flags & invalid_flags & RAM_SAVE_FLAG_COMPRESS_PAGE) { | |
2963 | error_report("Received an unexpected compressed page"); | |
2964 | } | |
2965 | ||
2966 | ret = -EINVAL; | |
2967 | break; | |
2968 | } | |
2969 | ||
bb890ed5 | 2970 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
a776aa15 | 2971 | RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) { |
4c4bad48 HZ |
2972 | RAMBlock *block = ram_block_from_stream(f, flags); |
2973 | ||
2974 | host = host_from_ram_block_offset(block, addr); | |
a776aa15 DDAG |
2975 | if (!host) { |
2976 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
2977 | ret = -EINVAL; | |
2978 | break; | |
2979 | } | |
f9494614 | 2980 | ramblock_recv_bitmap_set(block, host); |
1db9d8e5 | 2981 | trace_ram_load_loop(block->idstr, (uint64_t)addr, flags, host); |
a776aa15 DDAG |
2982 | } |
2983 | ||
56e93d26 JQ |
2984 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { |
2985 | case RAM_SAVE_FLAG_MEM_SIZE: | |
2986 | /* Synchronize RAM block list */ | |
2987 | total_ram_bytes = addr; | |
2988 | while (!ret && total_ram_bytes) { | |
2989 | RAMBlock *block; | |
56e93d26 JQ |
2990 | char id[256]; |
2991 | ram_addr_t length; | |
2992 | ||
2993 | len = qemu_get_byte(f); | |
2994 | qemu_get_buffer(f, (uint8_t *)id, len); | |
2995 | id[len] = 0; | |
2996 | length = qemu_get_be64(f); | |
2997 | ||
e3dd7493 DDAG |
2998 | block = qemu_ram_block_by_name(id); |
2999 | if (block) { | |
3000 | if (length != block->used_length) { | |
3001 | Error *local_err = NULL; | |
56e93d26 | 3002 | |
fa53a0e5 | 3003 | ret = qemu_ram_resize(block, length, |
e3dd7493 DDAG |
3004 | &local_err); |
3005 | if (local_err) { | |
3006 | error_report_err(local_err); | |
56e93d26 | 3007 | } |
56e93d26 | 3008 | } |
ef08fb38 DDAG |
3009 | /* For postcopy we need to check hugepage sizes match */ |
3010 | if (postcopy_advised && | |
3011 | block->page_size != qemu_host_page_size) { | |
3012 | uint64_t remote_page_size = qemu_get_be64(f); | |
3013 | if (remote_page_size != block->page_size) { | |
3014 | error_report("Mismatched RAM page size %s " | |
3015 | "(local) %zd != %" PRId64, | |
3016 | id, block->page_size, | |
3017 | remote_page_size); | |
3018 | ret = -EINVAL; | |
3019 | } | |
3020 | } | |
e3dd7493 DDAG |
3021 | ram_control_load_hook(f, RAM_CONTROL_BLOCK_REG, |
3022 | block->idstr); | |
3023 | } else { | |
56e93d26 JQ |
3024 | error_report("Unknown ramblock \"%s\", cannot " |
3025 | "accept migration", id); | |
3026 | ret = -EINVAL; | |
3027 | } | |
3028 | ||
3029 | total_ram_bytes -= length; | |
3030 | } | |
3031 | break; | |
a776aa15 | 3032 | |
bb890ed5 | 3033 | case RAM_SAVE_FLAG_ZERO: |
56e93d26 JQ |
3034 | ch = qemu_get_byte(f); |
3035 | ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); | |
3036 | break; | |
a776aa15 | 3037 | |
56e93d26 | 3038 | case RAM_SAVE_FLAG_PAGE: |
56e93d26 JQ |
3039 | qemu_get_buffer(f, host, TARGET_PAGE_SIZE); |
3040 | break; | |
56e93d26 | 3041 | |
a776aa15 | 3042 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
56e93d26 JQ |
3043 | len = qemu_get_be32(f); |
3044 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
3045 | error_report("Invalid compressed data length: %d", len); | |
3046 | ret = -EINVAL; | |
3047 | break; | |
3048 | } | |
c1bc6626 | 3049 | decompress_data_with_multi_threads(f, host, len); |
56e93d26 | 3050 | break; |
a776aa15 | 3051 | |
56e93d26 | 3052 | case RAM_SAVE_FLAG_XBZRLE: |
56e93d26 JQ |
3053 | if (load_xbzrle(f, addr, host) < 0) { |
3054 | error_report("Failed to decompress XBZRLE page at " | |
3055 | RAM_ADDR_FMT, addr); | |
3056 | ret = -EINVAL; | |
3057 | break; | |
3058 | } | |
3059 | break; | |
3060 | case RAM_SAVE_FLAG_EOS: | |
3061 | /* normal exit */ | |
3062 | break; | |
3063 | default: | |
3064 | if (flags & RAM_SAVE_FLAG_HOOK) { | |
632e3a5c | 3065 | ram_control_load_hook(f, RAM_CONTROL_HOOK, NULL); |
56e93d26 JQ |
3066 | } else { |
3067 | error_report("Unknown combination of migration flags: %#x", | |
3068 | flags); | |
3069 | ret = -EINVAL; | |
3070 | } | |
3071 | } | |
3072 | if (!ret) { | |
3073 | ret = qemu_file_get_error(f); | |
3074 | } | |
3075 | } | |
3076 | ||
5533b2e9 | 3077 | wait_for_decompress_done(); |
56e93d26 | 3078 | rcu_read_unlock(); |
55c4446b | 3079 | trace_ram_load_complete(ret, seq_iter); |
56e93d26 JQ |
3080 | return ret; |
3081 | } | |
3082 | ||
c6467627 VSO |
3083 | static bool ram_has_postcopy(void *opaque) |
3084 | { | |
3085 | return migrate_postcopy_ram(); | |
3086 | } | |
3087 | ||
56e93d26 | 3088 | static SaveVMHandlers savevm_ram_handlers = { |
9907e842 | 3089 | .save_setup = ram_save_setup, |
56e93d26 | 3090 | .save_live_iterate = ram_save_iterate, |
763c906b | 3091 | .save_live_complete_postcopy = ram_save_complete, |
a3e06c3d | 3092 | .save_live_complete_precopy = ram_save_complete, |
c6467627 | 3093 | .has_postcopy = ram_has_postcopy, |
56e93d26 JQ |
3094 | .save_live_pending = ram_save_pending, |
3095 | .load_state = ram_load, | |
f265e0e4 JQ |
3096 | .save_cleanup = ram_save_cleanup, |
3097 | .load_setup = ram_load_setup, | |
3098 | .load_cleanup = ram_load_cleanup, | |
56e93d26 JQ |
3099 | }; |
3100 | ||
3101 | void ram_mig_init(void) | |
3102 | { | |
3103 | qemu_mutex_init(&XBZRLE.lock); | |
6f37bb8b | 3104 | register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, &ram_state); |
56e93d26 | 3105 | } |