]>
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 | */ | |
1393a485 | 28 | #include "qemu/osdep.h" |
33c11879 | 29 | #include "cpu.h" |
56e93d26 | 30 | #include <zlib.h> |
4addcd4f | 31 | #include "qapi-event.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" |
56e93d26 | 45 | #include "trace.h" |
56e93d26 | 46 | #include "exec/ram_addr.h" |
56e93d26 | 47 | #include "qemu/rcu_queue.h" |
a91246c9 | 48 | #include "migration/colo.h" |
56e93d26 | 49 | |
56e93d26 JQ |
50 | /***********************************************************/ |
51 | /* ram save/restore */ | |
52 | ||
bb890ed5 JQ |
53 | /* RAM_SAVE_FLAG_ZERO used to be named RAM_SAVE_FLAG_COMPRESS, it |
54 | * worked for pages that where filled with the same char. We switched | |
55 | * it to only search for the zero value. And to avoid confusion with | |
56 | * RAM_SSAVE_FLAG_COMPRESS_PAGE just rename it. | |
57 | */ | |
58 | ||
56e93d26 | 59 | #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */ |
bb890ed5 | 60 | #define RAM_SAVE_FLAG_ZERO 0x02 |
56e93d26 JQ |
61 | #define RAM_SAVE_FLAG_MEM_SIZE 0x04 |
62 | #define RAM_SAVE_FLAG_PAGE 0x08 | |
63 | #define RAM_SAVE_FLAG_EOS 0x10 | |
64 | #define RAM_SAVE_FLAG_CONTINUE 0x20 | |
65 | #define RAM_SAVE_FLAG_XBZRLE 0x40 | |
66 | /* 0x80 is reserved in migration.h start with 0x100 next */ | |
67 | #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 | |
68 | ||
56e93d26 JQ |
69 | static inline bool is_zero_range(uint8_t *p, uint64_t size) |
70 | { | |
a1febc49 | 71 | return buffer_is_zero(p, size); |
56e93d26 JQ |
72 | } |
73 | ||
9360447d JQ |
74 | XBZRLECacheStats xbzrle_counters; |
75 | ||
56e93d26 JQ |
76 | /* struct contains XBZRLE cache and a static page |
77 | used by the compression */ | |
78 | static struct { | |
79 | /* buffer used for XBZRLE encoding */ | |
80 | uint8_t *encoded_buf; | |
81 | /* buffer for storing page content */ | |
82 | uint8_t *current_buf; | |
83 | /* Cache for XBZRLE, Protected by lock. */ | |
84 | PageCache *cache; | |
85 | QemuMutex lock; | |
c00e0928 JQ |
86 | /* it will store a page full of zeros */ |
87 | uint8_t *zero_target_page; | |
f265e0e4 JQ |
88 | /* buffer used for XBZRLE decoding */ |
89 | uint8_t *decoded_buf; | |
56e93d26 JQ |
90 | } XBZRLE; |
91 | ||
56e93d26 JQ |
92 | static void XBZRLE_cache_lock(void) |
93 | { | |
94 | if (migrate_use_xbzrle()) | |
95 | qemu_mutex_lock(&XBZRLE.lock); | |
96 | } | |
97 | ||
98 | static void XBZRLE_cache_unlock(void) | |
99 | { | |
100 | if (migrate_use_xbzrle()) | |
101 | qemu_mutex_unlock(&XBZRLE.lock); | |
102 | } | |
103 | ||
3d0684b2 JQ |
104 | /** |
105 | * xbzrle_cache_resize: resize the xbzrle cache | |
106 | * | |
107 | * This function is called from qmp_migrate_set_cache_size in main | |
108 | * thread, possibly while a migration is in progress. A running | |
109 | * migration may be using the cache and might finish during this call, | |
110 | * hence changes to the cache are protected by XBZRLE.lock(). | |
111 | * | |
112 | * Returns the new_size or negative in case of error. | |
113 | * | |
114 | * @new_size: new cache size | |
56e93d26 JQ |
115 | */ |
116 | int64_t xbzrle_cache_resize(int64_t new_size) | |
117 | { | |
118 | PageCache *new_cache; | |
119 | int64_t ret; | |
120 | ||
121 | if (new_size < TARGET_PAGE_SIZE) { | |
122 | return -1; | |
123 | } | |
124 | ||
125 | XBZRLE_cache_lock(); | |
126 | ||
127 | if (XBZRLE.cache != NULL) { | |
128 | if (pow2floor(new_size) == migrate_xbzrle_cache_size()) { | |
129 | goto out_new_size; | |
130 | } | |
131 | new_cache = cache_init(new_size / TARGET_PAGE_SIZE, | |
132 | TARGET_PAGE_SIZE); | |
133 | if (!new_cache) { | |
134 | error_report("Error creating cache"); | |
135 | ret = -1; | |
136 | goto out; | |
137 | } | |
138 | ||
139 | cache_fini(XBZRLE.cache); | |
140 | XBZRLE.cache = new_cache; | |
141 | } | |
142 | ||
143 | out_new_size: | |
144 | ret = pow2floor(new_size); | |
145 | out: | |
146 | XBZRLE_cache_unlock(); | |
147 | return ret; | |
148 | } | |
149 | ||
ec481c6c JQ |
150 | /* |
151 | * An outstanding page request, on the source, having been received | |
152 | * and queued | |
153 | */ | |
154 | struct RAMSrcPageRequest { | |
155 | RAMBlock *rb; | |
156 | hwaddr offset; | |
157 | hwaddr len; | |
158 | ||
159 | QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; | |
160 | }; | |
161 | ||
6f37bb8b JQ |
162 | /* State of RAM for migration */ |
163 | struct RAMState { | |
204b88b8 JQ |
164 | /* QEMUFile used for this migration */ |
165 | QEMUFile *f; | |
6f37bb8b JQ |
166 | /* Last block that we have visited searching for dirty pages */ |
167 | RAMBlock *last_seen_block; | |
168 | /* Last block from where we have sent data */ | |
169 | RAMBlock *last_sent_block; | |
269ace29 JQ |
170 | /* Last dirty target page we have sent */ |
171 | ram_addr_t last_page; | |
6f37bb8b JQ |
172 | /* last ram version we have seen */ |
173 | uint32_t last_version; | |
174 | /* We are in the first round */ | |
175 | bool ram_bulk_stage; | |
8d820d6f JQ |
176 | /* How many times we have dirty too many pages */ |
177 | int dirty_rate_high_cnt; | |
f664da80 JQ |
178 | /* these variables are used for bitmap sync */ |
179 | /* last time we did a full bitmap_sync */ | |
180 | int64_t time_last_bitmap_sync; | |
eac74159 | 181 | /* bytes transferred at start_time */ |
c4bdf0cf | 182 | uint64_t bytes_xfer_prev; |
a66cd90c | 183 | /* number of dirty pages since start_time */ |
68908ed6 | 184 | uint64_t num_dirty_pages_period; |
b5833fde JQ |
185 | /* xbzrle misses since the beginning of the period */ |
186 | uint64_t xbzrle_cache_miss_prev; | |
36040d9c JQ |
187 | /* number of iterations at the beginning of period */ |
188 | uint64_t iterations_prev; | |
23b28c3c JQ |
189 | /* Iterations since start */ |
190 | uint64_t iterations; | |
108cfae0 | 191 | /* protects modification of the bitmap */ |
9360447d JQ |
192 | uint64_t migration_dirty_pages; |
193 | /* number of dirty bits in the bitmap */ | |
108cfae0 | 194 | QemuMutex bitmap_mutex; |
68a098f3 JQ |
195 | /* The RAMBlock used in the last src_page_requests */ |
196 | RAMBlock *last_req_rb; | |
ec481c6c JQ |
197 | /* Queue of outstanding page requests from the destination */ |
198 | QemuMutex src_page_req_mutex; | |
199 | QSIMPLEQ_HEAD(src_page_requests, RAMSrcPageRequest) src_page_requests; | |
6f37bb8b JQ |
200 | }; |
201 | typedef struct RAMState RAMState; | |
202 | ||
53518d94 | 203 | static RAMState *ram_state; |
6f37bb8b | 204 | |
9edabd4d | 205 | uint64_t ram_bytes_remaining(void) |
2f4fde93 | 206 | { |
53518d94 | 207 | return ram_state->migration_dirty_pages * TARGET_PAGE_SIZE; |
2f4fde93 JQ |
208 | } |
209 | ||
9360447d | 210 | MigrationStats ram_counters; |
96506894 | 211 | |
b8fb8cb7 DDAG |
212 | /* used by the search for pages to send */ |
213 | struct PageSearchStatus { | |
214 | /* Current block being searched */ | |
215 | RAMBlock *block; | |
a935e30f JQ |
216 | /* Current page to search from */ |
217 | unsigned long page; | |
b8fb8cb7 DDAG |
218 | /* Set once we wrap around */ |
219 | bool complete_round; | |
220 | }; | |
221 | typedef struct PageSearchStatus PageSearchStatus; | |
222 | ||
56e93d26 | 223 | struct CompressParam { |
56e93d26 | 224 | bool done; |
90e56fb4 | 225 | bool quit; |
56e93d26 JQ |
226 | QEMUFile *file; |
227 | QemuMutex mutex; | |
228 | QemuCond cond; | |
229 | RAMBlock *block; | |
230 | ram_addr_t offset; | |
231 | }; | |
232 | typedef struct CompressParam CompressParam; | |
233 | ||
234 | struct DecompressParam { | |
73a8912b | 235 | bool done; |
90e56fb4 | 236 | bool quit; |
56e93d26 JQ |
237 | QemuMutex mutex; |
238 | QemuCond cond; | |
239 | void *des; | |
d341d9f3 | 240 | uint8_t *compbuf; |
56e93d26 JQ |
241 | int len; |
242 | }; | |
243 | typedef struct DecompressParam DecompressParam; | |
244 | ||
245 | static CompressParam *comp_param; | |
246 | static QemuThread *compress_threads; | |
247 | /* comp_done_cond is used to wake up the migration thread when | |
248 | * one of the compression threads has finished the compression. | |
249 | * comp_done_lock is used to co-work with comp_done_cond. | |
250 | */ | |
0d9f9a5c LL |
251 | static QemuMutex comp_done_lock; |
252 | static QemuCond comp_done_cond; | |
56e93d26 JQ |
253 | /* The empty QEMUFileOps will be used by file in CompressParam */ |
254 | static const QEMUFileOps empty_ops = { }; | |
255 | ||
56e93d26 JQ |
256 | static DecompressParam *decomp_param; |
257 | static QemuThread *decompress_threads; | |
73a8912b LL |
258 | static QemuMutex decomp_done_lock; |
259 | static QemuCond decomp_done_cond; | |
56e93d26 | 260 | |
a7a9a88f LL |
261 | static int do_compress_ram_page(QEMUFile *f, RAMBlock *block, |
262 | ram_addr_t offset); | |
56e93d26 JQ |
263 | |
264 | static void *do_data_compress(void *opaque) | |
265 | { | |
266 | CompressParam *param = opaque; | |
a7a9a88f LL |
267 | RAMBlock *block; |
268 | ram_addr_t offset; | |
56e93d26 | 269 | |
a7a9a88f | 270 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 271 | while (!param->quit) { |
a7a9a88f LL |
272 | if (param->block) { |
273 | block = param->block; | |
274 | offset = param->offset; | |
275 | param->block = NULL; | |
276 | qemu_mutex_unlock(¶m->mutex); | |
277 | ||
278 | do_compress_ram_page(param->file, block, offset); | |
279 | ||
0d9f9a5c | 280 | qemu_mutex_lock(&comp_done_lock); |
a7a9a88f | 281 | param->done = true; |
0d9f9a5c LL |
282 | qemu_cond_signal(&comp_done_cond); |
283 | qemu_mutex_unlock(&comp_done_lock); | |
a7a9a88f LL |
284 | |
285 | qemu_mutex_lock(¶m->mutex); | |
286 | } else { | |
56e93d26 JQ |
287 | qemu_cond_wait(¶m->cond, ¶m->mutex); |
288 | } | |
56e93d26 | 289 | } |
a7a9a88f | 290 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
291 | |
292 | return NULL; | |
293 | } | |
294 | ||
295 | static inline void terminate_compression_threads(void) | |
296 | { | |
297 | int idx, thread_count; | |
298 | ||
299 | thread_count = migrate_compress_threads(); | |
3d0684b2 | 300 | |
56e93d26 JQ |
301 | for (idx = 0; idx < thread_count; idx++) { |
302 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 303 | comp_param[idx].quit = true; |
56e93d26 JQ |
304 | qemu_cond_signal(&comp_param[idx].cond); |
305 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
306 | } | |
307 | } | |
308 | ||
309 | void migrate_compress_threads_join(void) | |
310 | { | |
311 | int i, thread_count; | |
312 | ||
313 | if (!migrate_use_compression()) { | |
314 | return; | |
315 | } | |
316 | terminate_compression_threads(); | |
317 | thread_count = migrate_compress_threads(); | |
318 | for (i = 0; i < thread_count; i++) { | |
319 | qemu_thread_join(compress_threads + i); | |
320 | qemu_fclose(comp_param[i].file); | |
321 | qemu_mutex_destroy(&comp_param[i].mutex); | |
322 | qemu_cond_destroy(&comp_param[i].cond); | |
323 | } | |
0d9f9a5c LL |
324 | qemu_mutex_destroy(&comp_done_lock); |
325 | qemu_cond_destroy(&comp_done_cond); | |
56e93d26 JQ |
326 | g_free(compress_threads); |
327 | g_free(comp_param); | |
56e93d26 JQ |
328 | compress_threads = NULL; |
329 | comp_param = NULL; | |
56e93d26 JQ |
330 | } |
331 | ||
332 | void migrate_compress_threads_create(void) | |
333 | { | |
334 | int i, thread_count; | |
335 | ||
336 | if (!migrate_use_compression()) { | |
337 | return; | |
338 | } | |
56e93d26 JQ |
339 | thread_count = migrate_compress_threads(); |
340 | compress_threads = g_new0(QemuThread, thread_count); | |
341 | comp_param = g_new0(CompressParam, thread_count); | |
0d9f9a5c LL |
342 | qemu_cond_init(&comp_done_cond); |
343 | qemu_mutex_init(&comp_done_lock); | |
56e93d26 | 344 | for (i = 0; i < thread_count; i++) { |
e110aa91 C |
345 | /* comp_param[i].file is just used as a dummy buffer to save data, |
346 | * set its ops to empty. | |
56e93d26 JQ |
347 | */ |
348 | comp_param[i].file = qemu_fopen_ops(NULL, &empty_ops); | |
349 | comp_param[i].done = true; | |
90e56fb4 | 350 | comp_param[i].quit = false; |
56e93d26 JQ |
351 | qemu_mutex_init(&comp_param[i].mutex); |
352 | qemu_cond_init(&comp_param[i].cond); | |
353 | qemu_thread_create(compress_threads + i, "compress", | |
354 | do_data_compress, comp_param + i, | |
355 | QEMU_THREAD_JOINABLE); | |
356 | } | |
357 | } | |
358 | ||
359 | /** | |
3d0684b2 | 360 | * save_page_header: write page header to wire |
56e93d26 JQ |
361 | * |
362 | * If this is the 1st block, it also writes the block identification | |
363 | * | |
3d0684b2 | 364 | * Returns the number of bytes written |
56e93d26 JQ |
365 | * |
366 | * @f: QEMUFile where to send the data | |
367 | * @block: block that contains the page we want to send | |
368 | * @offset: offset inside the block for the page | |
369 | * in the lower bits, it contains flags | |
370 | */ | |
2bf3aa85 JQ |
371 | static size_t save_page_header(RAMState *rs, QEMUFile *f, RAMBlock *block, |
372 | ram_addr_t offset) | |
56e93d26 | 373 | { |
9f5f380b | 374 | size_t size, len; |
56e93d26 | 375 | |
24795694 JQ |
376 | if (block == rs->last_sent_block) { |
377 | offset |= RAM_SAVE_FLAG_CONTINUE; | |
378 | } | |
2bf3aa85 | 379 | qemu_put_be64(f, offset); |
56e93d26 JQ |
380 | size = 8; |
381 | ||
382 | if (!(offset & RAM_SAVE_FLAG_CONTINUE)) { | |
9f5f380b | 383 | len = strlen(block->idstr); |
2bf3aa85 JQ |
384 | qemu_put_byte(f, len); |
385 | qemu_put_buffer(f, (uint8_t *)block->idstr, len); | |
9f5f380b | 386 | size += 1 + len; |
24795694 | 387 | rs->last_sent_block = block; |
56e93d26 JQ |
388 | } |
389 | return size; | |
390 | } | |
391 | ||
3d0684b2 JQ |
392 | /** |
393 | * mig_throttle_guest_down: throotle down the guest | |
394 | * | |
395 | * Reduce amount of guest cpu execution to hopefully slow down memory | |
396 | * writes. If guest dirty memory rate is reduced below the rate at | |
397 | * which we can transfer pages to the destination then we should be | |
398 | * able to complete migration. Some workloads dirty memory way too | |
399 | * fast and will not effectively converge, even with auto-converge. | |
070afca2 JH |
400 | */ |
401 | static void mig_throttle_guest_down(void) | |
402 | { | |
403 | MigrationState *s = migrate_get_current(); | |
2594f56d DB |
404 | uint64_t pct_initial = s->parameters.cpu_throttle_initial; |
405 | uint64_t pct_icrement = s->parameters.cpu_throttle_increment; | |
070afca2 JH |
406 | |
407 | /* We have not started throttling yet. Let's start it. */ | |
408 | if (!cpu_throttle_active()) { | |
409 | cpu_throttle_set(pct_initial); | |
410 | } else { | |
411 | /* Throttling already on, just increase the rate */ | |
412 | cpu_throttle_set(cpu_throttle_get_percentage() + pct_icrement); | |
413 | } | |
414 | } | |
415 | ||
3d0684b2 JQ |
416 | /** |
417 | * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache | |
418 | * | |
6f37bb8b | 419 | * @rs: current RAM state |
3d0684b2 JQ |
420 | * @current_addr: address for the zero page |
421 | * | |
422 | * Update the xbzrle cache to reflect a page that's been sent as all 0. | |
56e93d26 JQ |
423 | * The important thing is that a stale (not-yet-0'd) page be replaced |
424 | * by the new data. | |
425 | * As a bonus, if the page wasn't in the cache it gets added so that | |
3d0684b2 | 426 | * when a small write is made into the 0'd page it gets XBZRLE sent. |
56e93d26 | 427 | */ |
6f37bb8b | 428 | static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr) |
56e93d26 | 429 | { |
6f37bb8b | 430 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { |
56e93d26 JQ |
431 | return; |
432 | } | |
433 | ||
434 | /* We don't care if this fails to allocate a new cache page | |
435 | * as long as it updated an old one */ | |
c00e0928 | 436 | cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page, |
9360447d | 437 | ram_counters.dirty_sync_count); |
56e93d26 JQ |
438 | } |
439 | ||
440 | #define ENCODING_FLAG_XBZRLE 0x1 | |
441 | ||
442 | /** | |
443 | * save_xbzrle_page: compress and send current page | |
444 | * | |
445 | * Returns: 1 means that we wrote the page | |
446 | * 0 means that page is identical to the one already sent | |
447 | * -1 means that xbzrle would be longer than normal | |
448 | * | |
5a987738 | 449 | * @rs: current RAM state |
3d0684b2 JQ |
450 | * @current_data: pointer to the address of the page contents |
451 | * @current_addr: addr of the page | |
56e93d26 JQ |
452 | * @block: block that contains the page we want to send |
453 | * @offset: offset inside the block for the page | |
454 | * @last_stage: if we are at the completion stage | |
56e93d26 | 455 | */ |
204b88b8 | 456 | static int save_xbzrle_page(RAMState *rs, uint8_t **current_data, |
56e93d26 | 457 | ram_addr_t current_addr, RAMBlock *block, |
072c2511 | 458 | ram_addr_t offset, bool last_stage) |
56e93d26 JQ |
459 | { |
460 | int encoded_len = 0, bytes_xbzrle; | |
461 | uint8_t *prev_cached_page; | |
462 | ||
9360447d JQ |
463 | if (!cache_is_cached(XBZRLE.cache, current_addr, |
464 | ram_counters.dirty_sync_count)) { | |
465 | xbzrle_counters.cache_miss++; | |
56e93d26 JQ |
466 | if (!last_stage) { |
467 | if (cache_insert(XBZRLE.cache, current_addr, *current_data, | |
9360447d | 468 | ram_counters.dirty_sync_count) == -1) { |
56e93d26 JQ |
469 | return -1; |
470 | } else { | |
471 | /* update *current_data when the page has been | |
472 | inserted into cache */ | |
473 | *current_data = get_cached_data(XBZRLE.cache, current_addr); | |
474 | } | |
475 | } | |
476 | return -1; | |
477 | } | |
478 | ||
479 | prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); | |
480 | ||
481 | /* save current buffer into memory */ | |
482 | memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); | |
483 | ||
484 | /* XBZRLE encoding (if there is no overflow) */ | |
485 | encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, | |
486 | TARGET_PAGE_SIZE, XBZRLE.encoded_buf, | |
487 | TARGET_PAGE_SIZE); | |
488 | if (encoded_len == 0) { | |
55c4446b | 489 | trace_save_xbzrle_page_skipping(); |
56e93d26 JQ |
490 | return 0; |
491 | } else if (encoded_len == -1) { | |
55c4446b | 492 | trace_save_xbzrle_page_overflow(); |
9360447d | 493 | xbzrle_counters.overflow++; |
56e93d26 JQ |
494 | /* update data in the cache */ |
495 | if (!last_stage) { | |
496 | memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE); | |
497 | *current_data = prev_cached_page; | |
498 | } | |
499 | return -1; | |
500 | } | |
501 | ||
502 | /* we need to update the data in the cache, in order to get the same data */ | |
503 | if (!last_stage) { | |
504 | memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); | |
505 | } | |
506 | ||
507 | /* Send XBZRLE based compressed page */ | |
2bf3aa85 | 508 | bytes_xbzrle = save_page_header(rs, rs->f, block, |
204b88b8 JQ |
509 | offset | RAM_SAVE_FLAG_XBZRLE); |
510 | qemu_put_byte(rs->f, ENCODING_FLAG_XBZRLE); | |
511 | qemu_put_be16(rs->f, encoded_len); | |
512 | qemu_put_buffer(rs->f, XBZRLE.encoded_buf, encoded_len); | |
56e93d26 | 513 | bytes_xbzrle += encoded_len + 1 + 2; |
9360447d JQ |
514 | xbzrle_counters.pages++; |
515 | xbzrle_counters.bytes += bytes_xbzrle; | |
516 | ram_counters.transferred += bytes_xbzrle; | |
56e93d26 JQ |
517 | |
518 | return 1; | |
519 | } | |
520 | ||
3d0684b2 JQ |
521 | /** |
522 | * migration_bitmap_find_dirty: find the next dirty page from start | |
f3f491fc | 523 | * |
3d0684b2 JQ |
524 | * Called with rcu_read_lock() to protect migration_bitmap |
525 | * | |
526 | * Returns the byte offset within memory region of the start of a dirty page | |
527 | * | |
6f37bb8b | 528 | * @rs: current RAM state |
3d0684b2 | 529 | * @rb: RAMBlock where to search for dirty pages |
a935e30f | 530 | * @start: page where we start the search |
f3f491fc | 531 | */ |
56e93d26 | 532 | static inline |
a935e30f | 533 | unsigned long migration_bitmap_find_dirty(RAMState *rs, RAMBlock *rb, |
f20e2865 | 534 | unsigned long start) |
56e93d26 | 535 | { |
6b6712ef JQ |
536 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; |
537 | unsigned long *bitmap = rb->bmap; | |
56e93d26 JQ |
538 | unsigned long next; |
539 | ||
6b6712ef JQ |
540 | if (rs->ram_bulk_stage && start > 0) { |
541 | next = start + 1; | |
56e93d26 | 542 | } else { |
6b6712ef | 543 | next = find_next_bit(bitmap, size, start); |
56e93d26 JQ |
544 | } |
545 | ||
6b6712ef | 546 | return next; |
56e93d26 JQ |
547 | } |
548 | ||
06b10688 | 549 | static inline bool migration_bitmap_clear_dirty(RAMState *rs, |
f20e2865 JQ |
550 | RAMBlock *rb, |
551 | unsigned long page) | |
a82d593b DDAG |
552 | { |
553 | bool ret; | |
a82d593b | 554 | |
6b6712ef | 555 | ret = test_and_clear_bit(page, rb->bmap); |
a82d593b DDAG |
556 | |
557 | if (ret) { | |
0d8ec885 | 558 | rs->migration_dirty_pages--; |
a82d593b DDAG |
559 | } |
560 | return ret; | |
561 | } | |
562 | ||
15440dd5 JQ |
563 | static void migration_bitmap_sync_range(RAMState *rs, RAMBlock *rb, |
564 | ram_addr_t start, ram_addr_t length) | |
56e93d26 | 565 | { |
0d8ec885 | 566 | rs->migration_dirty_pages += |
6b6712ef | 567 | cpu_physical_memory_sync_dirty_bitmap(rb, start, length, |
0d8ec885 | 568 | &rs->num_dirty_pages_period); |
56e93d26 JQ |
569 | } |
570 | ||
3d0684b2 JQ |
571 | /** |
572 | * ram_pagesize_summary: calculate all the pagesizes of a VM | |
573 | * | |
574 | * Returns a summary bitmap of the page sizes of all RAMBlocks | |
575 | * | |
576 | * For VMs with just normal pages this is equivalent to the host page | |
577 | * size. If it's got some huge pages then it's the OR of all the | |
578 | * different page sizes. | |
e8ca1db2 DDAG |
579 | */ |
580 | uint64_t ram_pagesize_summary(void) | |
581 | { | |
582 | RAMBlock *block; | |
583 | uint64_t summary = 0; | |
584 | ||
99e15582 | 585 | RAMBLOCK_FOREACH(block) { |
e8ca1db2 DDAG |
586 | summary |= block->page_size; |
587 | } | |
588 | ||
589 | return summary; | |
590 | } | |
591 | ||
8d820d6f | 592 | static void migration_bitmap_sync(RAMState *rs) |
56e93d26 JQ |
593 | { |
594 | RAMBlock *block; | |
56e93d26 | 595 | int64_t end_time; |
c4bdf0cf | 596 | uint64_t bytes_xfer_now; |
56e93d26 | 597 | |
9360447d | 598 | ram_counters.dirty_sync_count++; |
56e93d26 | 599 | |
f664da80 JQ |
600 | if (!rs->time_last_bitmap_sync) { |
601 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56e93d26 JQ |
602 | } |
603 | ||
604 | trace_migration_bitmap_sync_start(); | |
9c1f8f44 | 605 | memory_global_dirty_log_sync(); |
56e93d26 | 606 | |
108cfae0 | 607 | qemu_mutex_lock(&rs->bitmap_mutex); |
56e93d26 | 608 | rcu_read_lock(); |
99e15582 | 609 | RAMBLOCK_FOREACH(block) { |
15440dd5 | 610 | migration_bitmap_sync_range(rs, block, 0, block->used_length); |
56e93d26 JQ |
611 | } |
612 | rcu_read_unlock(); | |
108cfae0 | 613 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 614 | |
a66cd90c | 615 | trace_migration_bitmap_sync_end(rs->num_dirty_pages_period); |
1ffb5dfd | 616 | |
56e93d26 JQ |
617 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
618 | ||
619 | /* more than 1 second = 1000 millisecons */ | |
f664da80 | 620 | if (end_time > rs->time_last_bitmap_sync + 1000) { |
d693c6f1 | 621 | /* calculate period counters */ |
9360447d | 622 | ram_counters.dirty_pages_rate = rs->num_dirty_pages_period * 1000 |
d693c6f1 | 623 | / (end_time - rs->time_last_bitmap_sync); |
9360447d | 624 | bytes_xfer_now = ram_counters.transferred; |
d693c6f1 | 625 | |
56e93d26 JQ |
626 | if (migrate_auto_converge()) { |
627 | /* The following detection logic can be refined later. For now: | |
628 | Check to see if the dirtied bytes is 50% more than the approx. | |
629 | amount of bytes that just got transferred since the last time we | |
070afca2 JH |
630 | were in this routine. If that happens twice, start or increase |
631 | throttling */ | |
070afca2 | 632 | |
d693c6f1 | 633 | if ((rs->num_dirty_pages_period * TARGET_PAGE_SIZE > |
eac74159 | 634 | (bytes_xfer_now - rs->bytes_xfer_prev) / 2) && |
b4a3c64b | 635 | (++rs->dirty_rate_high_cnt >= 2)) { |
56e93d26 | 636 | trace_migration_throttle(); |
8d820d6f | 637 | rs->dirty_rate_high_cnt = 0; |
070afca2 | 638 | mig_throttle_guest_down(); |
d693c6f1 | 639 | } |
56e93d26 | 640 | } |
070afca2 | 641 | |
56e93d26 | 642 | if (migrate_use_xbzrle()) { |
23b28c3c | 643 | if (rs->iterations_prev != rs->iterations) { |
9360447d JQ |
644 | xbzrle_counters.cache_miss_rate = |
645 | (double)(xbzrle_counters.cache_miss - | |
b5833fde | 646 | rs->xbzrle_cache_miss_prev) / |
23b28c3c | 647 | (rs->iterations - rs->iterations_prev); |
56e93d26 | 648 | } |
23b28c3c | 649 | rs->iterations_prev = rs->iterations; |
9360447d | 650 | rs->xbzrle_cache_miss_prev = xbzrle_counters.cache_miss; |
56e93d26 | 651 | } |
d693c6f1 FF |
652 | |
653 | /* reset period counters */ | |
f664da80 | 654 | rs->time_last_bitmap_sync = end_time; |
a66cd90c | 655 | rs->num_dirty_pages_period = 0; |
d2a4d85a | 656 | rs->bytes_xfer_prev = bytes_xfer_now; |
56e93d26 | 657 | } |
4addcd4f | 658 | if (migrate_use_events()) { |
9360447d | 659 | qapi_event_send_migration_pass(ram_counters.dirty_sync_count, NULL); |
4addcd4f | 660 | } |
56e93d26 JQ |
661 | } |
662 | ||
663 | /** | |
3d0684b2 | 664 | * save_zero_page: send the zero page to the stream |
56e93d26 | 665 | * |
3d0684b2 | 666 | * Returns the number of pages written. |
56e93d26 | 667 | * |
f7ccd61b | 668 | * @rs: current RAM state |
56e93d26 JQ |
669 | * @block: block that contains the page we want to send |
670 | * @offset: offset inside the block for the page | |
671 | * @p: pointer to the page | |
56e93d26 | 672 | */ |
ce25d337 JQ |
673 | static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, |
674 | uint8_t *p) | |
56e93d26 JQ |
675 | { |
676 | int pages = -1; | |
677 | ||
678 | if (is_zero_range(p, TARGET_PAGE_SIZE)) { | |
9360447d JQ |
679 | ram_counters.duplicate++; |
680 | ram_counters.transferred += | |
bb890ed5 | 681 | save_page_header(rs, rs->f, block, offset | RAM_SAVE_FLAG_ZERO); |
ce25d337 | 682 | qemu_put_byte(rs->f, 0); |
9360447d | 683 | ram_counters.transferred += 1; |
56e93d26 JQ |
684 | pages = 1; |
685 | } | |
686 | ||
687 | return pages; | |
688 | } | |
689 | ||
5727309d | 690 | static void ram_release_pages(const char *rbname, uint64_t offset, int pages) |
53f09a10 | 691 | { |
5727309d | 692 | if (!migrate_release_ram() || !migration_in_postcopy()) { |
53f09a10 PB |
693 | return; |
694 | } | |
695 | ||
aaa2064c | 696 | ram_discard_range(rbname, offset, pages << TARGET_PAGE_BITS); |
53f09a10 PB |
697 | } |
698 | ||
56e93d26 | 699 | /** |
3d0684b2 | 700 | * ram_save_page: send the given page to the stream |
56e93d26 | 701 | * |
3d0684b2 | 702 | * Returns the number of pages written. |
3fd3c4b3 DDAG |
703 | * < 0 - error |
704 | * >=0 - Number of pages written - this might legally be 0 | |
705 | * if xbzrle noticed the page was the same. | |
56e93d26 | 706 | * |
6f37bb8b | 707 | * @rs: current RAM state |
56e93d26 JQ |
708 | * @block: block that contains the page we want to send |
709 | * @offset: offset inside the block for the page | |
710 | * @last_stage: if we are at the completion stage | |
56e93d26 | 711 | */ |
a0a8aa14 | 712 | static int ram_save_page(RAMState *rs, PageSearchStatus *pss, bool last_stage) |
56e93d26 JQ |
713 | { |
714 | int pages = -1; | |
715 | uint64_t bytes_xmit; | |
716 | ram_addr_t current_addr; | |
56e93d26 JQ |
717 | uint8_t *p; |
718 | int ret; | |
719 | bool send_async = true; | |
a08f6890 | 720 | RAMBlock *block = pss->block; |
a935e30f | 721 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
56e93d26 | 722 | |
2f68e399 | 723 | p = block->host + offset; |
1db9d8e5 | 724 | trace_ram_save_page(block->idstr, (uint64_t)offset, p); |
56e93d26 JQ |
725 | |
726 | /* In doubt sent page as normal */ | |
727 | bytes_xmit = 0; | |
ce25d337 | 728 | ret = ram_control_save_page(rs->f, block->offset, |
56e93d26 JQ |
729 | offset, TARGET_PAGE_SIZE, &bytes_xmit); |
730 | if (bytes_xmit) { | |
9360447d | 731 | ram_counters.transferred += bytes_xmit; |
56e93d26 JQ |
732 | pages = 1; |
733 | } | |
734 | ||
735 | XBZRLE_cache_lock(); | |
736 | ||
737 | current_addr = block->offset + offset; | |
738 | ||
56e93d26 JQ |
739 | if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { |
740 | if (ret != RAM_SAVE_CONTROL_DELAYED) { | |
741 | if (bytes_xmit > 0) { | |
9360447d | 742 | ram_counters.normal++; |
56e93d26 | 743 | } else if (bytes_xmit == 0) { |
9360447d | 744 | ram_counters.duplicate++; |
56e93d26 JQ |
745 | } |
746 | } | |
747 | } else { | |
ce25d337 | 748 | pages = save_zero_page(rs, block, offset, p); |
56e93d26 JQ |
749 | if (pages > 0) { |
750 | /* Must let xbzrle know, otherwise a previous (now 0'd) cached | |
751 | * page would be stale | |
752 | */ | |
6f37bb8b | 753 | xbzrle_cache_zero_page(rs, current_addr); |
a935e30f | 754 | ram_release_pages(block->idstr, offset, pages); |
6f37bb8b | 755 | } else if (!rs->ram_bulk_stage && |
5727309d | 756 | !migration_in_postcopy() && migrate_use_xbzrle()) { |
204b88b8 | 757 | pages = save_xbzrle_page(rs, &p, current_addr, block, |
072c2511 | 758 | offset, last_stage); |
56e93d26 JQ |
759 | if (!last_stage) { |
760 | /* Can't send this cached data async, since the cache page | |
761 | * might get updated before it gets to the wire | |
762 | */ | |
763 | send_async = false; | |
764 | } | |
765 | } | |
766 | } | |
767 | ||
768 | /* XBZRLE overflow or normal page */ | |
769 | if (pages == -1) { | |
9360447d JQ |
770 | ram_counters.transferred += |
771 | save_page_header(rs, rs->f, block, offset | RAM_SAVE_FLAG_PAGE); | |
56e93d26 | 772 | if (send_async) { |
ce25d337 | 773 | qemu_put_buffer_async(rs->f, p, TARGET_PAGE_SIZE, |
53f09a10 | 774 | migrate_release_ram() & |
5727309d | 775 | migration_in_postcopy()); |
56e93d26 | 776 | } else { |
ce25d337 | 777 | qemu_put_buffer(rs->f, p, TARGET_PAGE_SIZE); |
56e93d26 | 778 | } |
9360447d | 779 | ram_counters.transferred += TARGET_PAGE_SIZE; |
56e93d26 | 780 | pages = 1; |
9360447d | 781 | ram_counters.normal++; |
56e93d26 JQ |
782 | } |
783 | ||
784 | XBZRLE_cache_unlock(); | |
785 | ||
786 | return pages; | |
787 | } | |
788 | ||
a7a9a88f LL |
789 | static int do_compress_ram_page(QEMUFile *f, RAMBlock *block, |
790 | ram_addr_t offset) | |
56e93d26 | 791 | { |
53518d94 | 792 | RAMState *rs = ram_state; |
56e93d26 | 793 | int bytes_sent, blen; |
a7a9a88f | 794 | uint8_t *p = block->host + (offset & TARGET_PAGE_MASK); |
56e93d26 | 795 | |
2bf3aa85 | 796 | bytes_sent = save_page_header(rs, f, block, offset | |
56e93d26 | 797 | RAM_SAVE_FLAG_COMPRESS_PAGE); |
a7a9a88f | 798 | blen = qemu_put_compression_data(f, p, TARGET_PAGE_SIZE, |
56e93d26 | 799 | migrate_compress_level()); |
b3be2896 LL |
800 | if (blen < 0) { |
801 | bytes_sent = 0; | |
802 | qemu_file_set_error(migrate_get_current()->to_dst_file, blen); | |
803 | error_report("compressed data failed!"); | |
804 | } else { | |
805 | bytes_sent += blen; | |
5727309d | 806 | ram_release_pages(block->idstr, offset & TARGET_PAGE_MASK, 1); |
b3be2896 | 807 | } |
56e93d26 JQ |
808 | |
809 | return bytes_sent; | |
810 | } | |
811 | ||
ce25d337 | 812 | static void flush_compressed_data(RAMState *rs) |
56e93d26 JQ |
813 | { |
814 | int idx, len, thread_count; | |
815 | ||
816 | if (!migrate_use_compression()) { | |
817 | return; | |
818 | } | |
819 | thread_count = migrate_compress_threads(); | |
a7a9a88f | 820 | |
0d9f9a5c | 821 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 | 822 | for (idx = 0; idx < thread_count; idx++) { |
a7a9a88f | 823 | while (!comp_param[idx].done) { |
0d9f9a5c | 824 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 | 825 | } |
a7a9a88f | 826 | } |
0d9f9a5c | 827 | qemu_mutex_unlock(&comp_done_lock); |
a7a9a88f LL |
828 | |
829 | for (idx = 0; idx < thread_count; idx++) { | |
830 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 831 | if (!comp_param[idx].quit) { |
ce25d337 | 832 | len = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
9360447d | 833 | ram_counters.transferred += len; |
56e93d26 | 834 | } |
a7a9a88f | 835 | qemu_mutex_unlock(&comp_param[idx].mutex); |
56e93d26 JQ |
836 | } |
837 | } | |
838 | ||
839 | static inline void set_compress_params(CompressParam *param, RAMBlock *block, | |
840 | ram_addr_t offset) | |
841 | { | |
842 | param->block = block; | |
843 | param->offset = offset; | |
844 | } | |
845 | ||
ce25d337 JQ |
846 | static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block, |
847 | ram_addr_t offset) | |
56e93d26 JQ |
848 | { |
849 | int idx, thread_count, bytes_xmit = -1, pages = -1; | |
850 | ||
851 | thread_count = migrate_compress_threads(); | |
0d9f9a5c | 852 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 JQ |
853 | while (true) { |
854 | for (idx = 0; idx < thread_count; idx++) { | |
855 | if (comp_param[idx].done) { | |
a7a9a88f | 856 | comp_param[idx].done = false; |
ce25d337 | 857 | bytes_xmit = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
a7a9a88f | 858 | qemu_mutex_lock(&comp_param[idx].mutex); |
56e93d26 | 859 | set_compress_params(&comp_param[idx], block, offset); |
a7a9a88f LL |
860 | qemu_cond_signal(&comp_param[idx].cond); |
861 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
56e93d26 | 862 | pages = 1; |
9360447d JQ |
863 | ram_counters.normal++; |
864 | ram_counters.transferred += bytes_xmit; | |
56e93d26 JQ |
865 | break; |
866 | } | |
867 | } | |
868 | if (pages > 0) { | |
869 | break; | |
870 | } else { | |
0d9f9a5c | 871 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 JQ |
872 | } |
873 | } | |
0d9f9a5c | 874 | qemu_mutex_unlock(&comp_done_lock); |
56e93d26 JQ |
875 | |
876 | return pages; | |
877 | } | |
878 | ||
879 | /** | |
880 | * ram_save_compressed_page: compress the given page and send it to the stream | |
881 | * | |
3d0684b2 | 882 | * Returns the number of pages written. |
56e93d26 | 883 | * |
6f37bb8b | 884 | * @rs: current RAM state |
56e93d26 JQ |
885 | * @block: block that contains the page we want to send |
886 | * @offset: offset inside the block for the page | |
887 | * @last_stage: if we are at the completion stage | |
56e93d26 | 888 | */ |
a0a8aa14 JQ |
889 | static int ram_save_compressed_page(RAMState *rs, PageSearchStatus *pss, |
890 | bool last_stage) | |
56e93d26 JQ |
891 | { |
892 | int pages = -1; | |
fc50438e | 893 | uint64_t bytes_xmit = 0; |
56e93d26 | 894 | uint8_t *p; |
fc50438e | 895 | int ret, blen; |
a08f6890 | 896 | RAMBlock *block = pss->block; |
a935e30f | 897 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
56e93d26 | 898 | |
2f68e399 | 899 | p = block->host + offset; |
56e93d26 | 900 | |
ce25d337 | 901 | ret = ram_control_save_page(rs->f, block->offset, |
56e93d26 JQ |
902 | offset, TARGET_PAGE_SIZE, &bytes_xmit); |
903 | if (bytes_xmit) { | |
9360447d | 904 | ram_counters.transferred += bytes_xmit; |
56e93d26 JQ |
905 | pages = 1; |
906 | } | |
56e93d26 JQ |
907 | if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { |
908 | if (ret != RAM_SAVE_CONTROL_DELAYED) { | |
909 | if (bytes_xmit > 0) { | |
9360447d | 910 | ram_counters.normal++; |
56e93d26 | 911 | } else if (bytes_xmit == 0) { |
9360447d | 912 | ram_counters.duplicate++; |
56e93d26 JQ |
913 | } |
914 | } | |
915 | } else { | |
916 | /* When starting the process of a new block, the first page of | |
917 | * the block should be sent out before other pages in the same | |
918 | * block, and all the pages in last block should have been sent | |
919 | * out, keeping this order is important, because the 'cont' flag | |
920 | * is used to avoid resending the block name. | |
921 | */ | |
6f37bb8b | 922 | if (block != rs->last_sent_block) { |
ce25d337 JQ |
923 | flush_compressed_data(rs); |
924 | pages = save_zero_page(rs, block, offset, p); | |
56e93d26 | 925 | if (pages == -1) { |
fc50438e | 926 | /* Make sure the first page is sent out before other pages */ |
2bf3aa85 | 927 | bytes_xmit = save_page_header(rs, rs->f, block, offset | |
fc50438e | 928 | RAM_SAVE_FLAG_COMPRESS_PAGE); |
ce25d337 | 929 | blen = qemu_put_compression_data(rs->f, p, TARGET_PAGE_SIZE, |
fc50438e LL |
930 | migrate_compress_level()); |
931 | if (blen > 0) { | |
9360447d JQ |
932 | ram_counters.transferred += bytes_xmit + blen; |
933 | ram_counters.normal++; | |
b3be2896 | 934 | pages = 1; |
fc50438e | 935 | } else { |
ce25d337 | 936 | qemu_file_set_error(rs->f, blen); |
fc50438e | 937 | error_report("compressed data failed!"); |
b3be2896 | 938 | } |
56e93d26 | 939 | } |
53f09a10 | 940 | if (pages > 0) { |
a935e30f | 941 | ram_release_pages(block->idstr, offset, pages); |
53f09a10 | 942 | } |
56e93d26 | 943 | } else { |
ce25d337 | 944 | pages = save_zero_page(rs, block, offset, p); |
56e93d26 | 945 | if (pages == -1) { |
ce25d337 | 946 | pages = compress_page_with_multi_thread(rs, block, offset); |
53f09a10 | 947 | } else { |
a935e30f | 948 | ram_release_pages(block->idstr, offset, pages); |
56e93d26 JQ |
949 | } |
950 | } | |
951 | } | |
952 | ||
953 | return pages; | |
954 | } | |
955 | ||
3d0684b2 JQ |
956 | /** |
957 | * find_dirty_block: find the next dirty page and update any state | |
958 | * associated with the search process. | |
b9e60928 | 959 | * |
3d0684b2 | 960 | * Returns if a page is found |
b9e60928 | 961 | * |
6f37bb8b | 962 | * @rs: current RAM state |
3d0684b2 JQ |
963 | * @pss: data about the state of the current dirty page scan |
964 | * @again: set to false if the search has scanned the whole of RAM | |
b9e60928 | 965 | */ |
f20e2865 | 966 | static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again) |
b9e60928 | 967 | { |
f20e2865 | 968 | pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page); |
6f37bb8b | 969 | if (pss->complete_round && pss->block == rs->last_seen_block && |
a935e30f | 970 | pss->page >= rs->last_page) { |
b9e60928 DDAG |
971 | /* |
972 | * We've been once around the RAM and haven't found anything. | |
973 | * Give up. | |
974 | */ | |
975 | *again = false; | |
976 | return false; | |
977 | } | |
a935e30f | 978 | if ((pss->page << TARGET_PAGE_BITS) >= pss->block->used_length) { |
b9e60928 | 979 | /* Didn't find anything in this RAM Block */ |
a935e30f | 980 | pss->page = 0; |
b9e60928 DDAG |
981 | pss->block = QLIST_NEXT_RCU(pss->block, next); |
982 | if (!pss->block) { | |
983 | /* Hit the end of the list */ | |
984 | pss->block = QLIST_FIRST_RCU(&ram_list.blocks); | |
985 | /* Flag that we've looped */ | |
986 | pss->complete_round = true; | |
6f37bb8b | 987 | rs->ram_bulk_stage = false; |
b9e60928 DDAG |
988 | if (migrate_use_xbzrle()) { |
989 | /* If xbzrle is on, stop using the data compression at this | |
990 | * point. In theory, xbzrle can do better than compression. | |
991 | */ | |
ce25d337 | 992 | flush_compressed_data(rs); |
b9e60928 DDAG |
993 | } |
994 | } | |
995 | /* Didn't find anything this time, but try again on the new block */ | |
996 | *again = true; | |
997 | return false; | |
998 | } else { | |
999 | /* Can go around again, but... */ | |
1000 | *again = true; | |
1001 | /* We've found something so probably don't need to */ | |
1002 | return true; | |
1003 | } | |
1004 | } | |
1005 | ||
3d0684b2 JQ |
1006 | /** |
1007 | * unqueue_page: gets a page of the queue | |
1008 | * | |
a82d593b | 1009 | * Helper for 'get_queued_page' - gets a page off the queue |
a82d593b | 1010 | * |
3d0684b2 JQ |
1011 | * Returns the block of the page (or NULL if none available) |
1012 | * | |
ec481c6c | 1013 | * @rs: current RAM state |
3d0684b2 | 1014 | * @offset: used to return the offset within the RAMBlock |
a82d593b | 1015 | */ |
f20e2865 | 1016 | static RAMBlock *unqueue_page(RAMState *rs, ram_addr_t *offset) |
a82d593b DDAG |
1017 | { |
1018 | RAMBlock *block = NULL; | |
1019 | ||
ec481c6c JQ |
1020 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1021 | if (!QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
1022 | struct RAMSrcPageRequest *entry = | |
1023 | QSIMPLEQ_FIRST(&rs->src_page_requests); | |
a82d593b DDAG |
1024 | block = entry->rb; |
1025 | *offset = entry->offset; | |
a82d593b DDAG |
1026 | |
1027 | if (entry->len > TARGET_PAGE_SIZE) { | |
1028 | entry->len -= TARGET_PAGE_SIZE; | |
1029 | entry->offset += TARGET_PAGE_SIZE; | |
1030 | } else { | |
1031 | memory_region_unref(block->mr); | |
ec481c6c | 1032 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
a82d593b DDAG |
1033 | g_free(entry); |
1034 | } | |
1035 | } | |
ec481c6c | 1036 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
a82d593b DDAG |
1037 | |
1038 | return block; | |
1039 | } | |
1040 | ||
3d0684b2 JQ |
1041 | /** |
1042 | * get_queued_page: unqueue a page from the postocpy requests | |
1043 | * | |
1044 | * Skips pages that are already sent (!dirty) | |
a82d593b | 1045 | * |
3d0684b2 | 1046 | * Returns if a queued page is found |
a82d593b | 1047 | * |
6f37bb8b | 1048 | * @rs: current RAM state |
3d0684b2 | 1049 | * @pss: data about the state of the current dirty page scan |
a82d593b | 1050 | */ |
f20e2865 | 1051 | static bool get_queued_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b DDAG |
1052 | { |
1053 | RAMBlock *block; | |
1054 | ram_addr_t offset; | |
1055 | bool dirty; | |
1056 | ||
1057 | do { | |
f20e2865 | 1058 | block = unqueue_page(rs, &offset); |
a82d593b DDAG |
1059 | /* |
1060 | * We're sending this page, and since it's postcopy nothing else | |
1061 | * will dirty it, and we must make sure it doesn't get sent again | |
1062 | * even if this queue request was received after the background | |
1063 | * search already sent it. | |
1064 | */ | |
1065 | if (block) { | |
f20e2865 JQ |
1066 | unsigned long page; |
1067 | ||
6b6712ef JQ |
1068 | page = offset >> TARGET_PAGE_BITS; |
1069 | dirty = test_bit(page, block->bmap); | |
a82d593b | 1070 | if (!dirty) { |
06b10688 | 1071 | trace_get_queued_page_not_dirty(block->idstr, (uint64_t)offset, |
6b6712ef | 1072 | page, test_bit(page, block->unsentmap)); |
a82d593b | 1073 | } else { |
f20e2865 | 1074 | trace_get_queued_page(block->idstr, (uint64_t)offset, page); |
a82d593b DDAG |
1075 | } |
1076 | } | |
1077 | ||
1078 | } while (block && !dirty); | |
1079 | ||
1080 | if (block) { | |
1081 | /* | |
1082 | * As soon as we start servicing pages out of order, then we have | |
1083 | * to kill the bulk stage, since the bulk stage assumes | |
1084 | * in (migration_bitmap_find_and_reset_dirty) that every page is | |
1085 | * dirty, that's no longer true. | |
1086 | */ | |
6f37bb8b | 1087 | rs->ram_bulk_stage = false; |
a82d593b DDAG |
1088 | |
1089 | /* | |
1090 | * We want the background search to continue from the queued page | |
1091 | * since the guest is likely to want other pages near to the page | |
1092 | * it just requested. | |
1093 | */ | |
1094 | pss->block = block; | |
a935e30f | 1095 | pss->page = offset >> TARGET_PAGE_BITS; |
a82d593b DDAG |
1096 | } |
1097 | ||
1098 | return !!block; | |
1099 | } | |
1100 | ||
6c595cde | 1101 | /** |
5e58f968 JQ |
1102 | * migration_page_queue_free: drop any remaining pages in the ram |
1103 | * request queue | |
6c595cde | 1104 | * |
3d0684b2 JQ |
1105 | * It should be empty at the end anyway, but in error cases there may |
1106 | * be some left. in case that there is any page left, we drop it. | |
1107 | * | |
6c595cde | 1108 | */ |
83c13382 | 1109 | static void migration_page_queue_free(RAMState *rs) |
6c595cde | 1110 | { |
ec481c6c | 1111 | struct RAMSrcPageRequest *mspr, *next_mspr; |
6c595cde DDAG |
1112 | /* This queue generally should be empty - but in the case of a failed |
1113 | * migration might have some droppings in. | |
1114 | */ | |
1115 | rcu_read_lock(); | |
ec481c6c | 1116 | QSIMPLEQ_FOREACH_SAFE(mspr, &rs->src_page_requests, next_req, next_mspr) { |
6c595cde | 1117 | memory_region_unref(mspr->rb->mr); |
ec481c6c | 1118 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
6c595cde DDAG |
1119 | g_free(mspr); |
1120 | } | |
1121 | rcu_read_unlock(); | |
1122 | } | |
1123 | ||
1124 | /** | |
3d0684b2 JQ |
1125 | * ram_save_queue_pages: queue the page for transmission |
1126 | * | |
1127 | * A request from postcopy destination for example. | |
1128 | * | |
1129 | * Returns zero on success or negative on error | |
1130 | * | |
3d0684b2 JQ |
1131 | * @rbname: Name of the RAMBLock of the request. NULL means the |
1132 | * same that last one. | |
1133 | * @start: starting address from the start of the RAMBlock | |
1134 | * @len: length (in bytes) to send | |
6c595cde | 1135 | */ |
96506894 | 1136 | int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len) |
6c595cde DDAG |
1137 | { |
1138 | RAMBlock *ramblock; | |
53518d94 | 1139 | RAMState *rs = ram_state; |
6c595cde | 1140 | |
9360447d | 1141 | ram_counters.postcopy_requests++; |
6c595cde DDAG |
1142 | rcu_read_lock(); |
1143 | if (!rbname) { | |
1144 | /* Reuse last RAMBlock */ | |
68a098f3 | 1145 | ramblock = rs->last_req_rb; |
6c595cde DDAG |
1146 | |
1147 | if (!ramblock) { | |
1148 | /* | |
1149 | * Shouldn't happen, we can't reuse the last RAMBlock if | |
1150 | * it's the 1st request. | |
1151 | */ | |
1152 | error_report("ram_save_queue_pages no previous block"); | |
1153 | goto err; | |
1154 | } | |
1155 | } else { | |
1156 | ramblock = qemu_ram_block_by_name(rbname); | |
1157 | ||
1158 | if (!ramblock) { | |
1159 | /* We shouldn't be asked for a non-existent RAMBlock */ | |
1160 | error_report("ram_save_queue_pages no block '%s'", rbname); | |
1161 | goto err; | |
1162 | } | |
68a098f3 | 1163 | rs->last_req_rb = ramblock; |
6c595cde DDAG |
1164 | } |
1165 | trace_ram_save_queue_pages(ramblock->idstr, start, len); | |
1166 | if (start+len > ramblock->used_length) { | |
9458ad6b JQ |
1167 | error_report("%s request overrun start=" RAM_ADDR_FMT " len=" |
1168 | RAM_ADDR_FMT " blocklen=" RAM_ADDR_FMT, | |
6c595cde DDAG |
1169 | __func__, start, len, ramblock->used_length); |
1170 | goto err; | |
1171 | } | |
1172 | ||
ec481c6c JQ |
1173 | struct RAMSrcPageRequest *new_entry = |
1174 | g_malloc0(sizeof(struct RAMSrcPageRequest)); | |
6c595cde DDAG |
1175 | new_entry->rb = ramblock; |
1176 | new_entry->offset = start; | |
1177 | new_entry->len = len; | |
1178 | ||
1179 | memory_region_ref(ramblock->mr); | |
ec481c6c JQ |
1180 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1181 | QSIMPLEQ_INSERT_TAIL(&rs->src_page_requests, new_entry, next_req); | |
1182 | qemu_mutex_unlock(&rs->src_page_req_mutex); | |
6c595cde DDAG |
1183 | rcu_read_unlock(); |
1184 | ||
1185 | return 0; | |
1186 | ||
1187 | err: | |
1188 | rcu_read_unlock(); | |
1189 | return -1; | |
1190 | } | |
1191 | ||
a82d593b | 1192 | /** |
3d0684b2 | 1193 | * ram_save_target_page: save one target page |
a82d593b | 1194 | * |
3d0684b2 | 1195 | * Returns the number of pages written |
a82d593b | 1196 | * |
6f37bb8b | 1197 | * @rs: current RAM state |
3d0684b2 | 1198 | * @ms: current migration state |
3d0684b2 | 1199 | * @pss: data about the page we want to send |
a82d593b | 1200 | * @last_stage: if we are at the completion stage |
a82d593b | 1201 | */ |
a0a8aa14 | 1202 | static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 1203 | bool last_stage) |
a82d593b DDAG |
1204 | { |
1205 | int res = 0; | |
1206 | ||
1207 | /* Check the pages is dirty and if it is send it */ | |
f20e2865 | 1208 | if (migration_bitmap_clear_dirty(rs, pss->block, pss->page)) { |
6d358d94 JQ |
1209 | /* |
1210 | * If xbzrle is on, stop using the data compression after first | |
1211 | * round of migration even if compression is enabled. In theory, | |
1212 | * xbzrle can do better than compression. | |
1213 | */ | |
6b6712ef JQ |
1214 | if (migrate_use_compression() && |
1215 | (rs->ram_bulk_stage || !migrate_use_xbzrle())) { | |
a0a8aa14 | 1216 | res = ram_save_compressed_page(rs, pss, last_stage); |
a82d593b | 1217 | } else { |
a0a8aa14 | 1218 | res = ram_save_page(rs, pss, last_stage); |
a82d593b DDAG |
1219 | } |
1220 | ||
1221 | if (res < 0) { | |
1222 | return res; | |
1223 | } | |
6b6712ef JQ |
1224 | if (pss->block->unsentmap) { |
1225 | clear_bit(pss->page, pss->block->unsentmap); | |
a82d593b DDAG |
1226 | } |
1227 | } | |
1228 | ||
1229 | return res; | |
1230 | } | |
1231 | ||
1232 | /** | |
3d0684b2 | 1233 | * ram_save_host_page: save a whole host page |
a82d593b | 1234 | * |
3d0684b2 JQ |
1235 | * Starting at *offset send pages up to the end of the current host |
1236 | * page. It's valid for the initial offset to point into the middle of | |
1237 | * a host page in which case the remainder of the hostpage is sent. | |
1238 | * Only dirty target pages are sent. Note that the host page size may | |
1239 | * be a huge page for this block. | |
1eb3fc0a DDAG |
1240 | * The saving stops at the boundary of the used_length of the block |
1241 | * if the RAMBlock isn't a multiple of the host page size. | |
a82d593b | 1242 | * |
3d0684b2 JQ |
1243 | * Returns the number of pages written or negative on error |
1244 | * | |
6f37bb8b | 1245 | * @rs: current RAM state |
3d0684b2 | 1246 | * @ms: current migration state |
3d0684b2 | 1247 | * @pss: data about the page we want to send |
a82d593b | 1248 | * @last_stage: if we are at the completion stage |
a82d593b | 1249 | */ |
a0a8aa14 | 1250 | static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 1251 | bool last_stage) |
a82d593b DDAG |
1252 | { |
1253 | int tmppages, pages = 0; | |
a935e30f JQ |
1254 | size_t pagesize_bits = |
1255 | qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
4c011c37 | 1256 | |
a82d593b | 1257 | do { |
f20e2865 | 1258 | tmppages = ram_save_target_page(rs, pss, last_stage); |
a82d593b DDAG |
1259 | if (tmppages < 0) { |
1260 | return tmppages; | |
1261 | } | |
1262 | ||
1263 | pages += tmppages; | |
a935e30f | 1264 | pss->page++; |
1eb3fc0a DDAG |
1265 | } while ((pss->page & (pagesize_bits - 1)) && |
1266 | offset_in_ramblock(pss->block, pss->page << TARGET_PAGE_BITS)); | |
a82d593b DDAG |
1267 | |
1268 | /* The offset we leave with is the last one we looked at */ | |
a935e30f | 1269 | pss->page--; |
a82d593b DDAG |
1270 | return pages; |
1271 | } | |
6c595cde | 1272 | |
56e93d26 | 1273 | /** |
3d0684b2 | 1274 | * ram_find_and_save_block: finds a dirty page and sends it to f |
56e93d26 JQ |
1275 | * |
1276 | * Called within an RCU critical section. | |
1277 | * | |
3d0684b2 | 1278 | * Returns the number of pages written where zero means no dirty pages |
56e93d26 | 1279 | * |
6f37bb8b | 1280 | * @rs: current RAM state |
56e93d26 | 1281 | * @last_stage: if we are at the completion stage |
a82d593b DDAG |
1282 | * |
1283 | * On systems where host-page-size > target-page-size it will send all the | |
1284 | * pages in a host page that are dirty. | |
56e93d26 JQ |
1285 | */ |
1286 | ||
ce25d337 | 1287 | static int ram_find_and_save_block(RAMState *rs, bool last_stage) |
56e93d26 | 1288 | { |
b8fb8cb7 | 1289 | PageSearchStatus pss; |
56e93d26 | 1290 | int pages = 0; |
b9e60928 | 1291 | bool again, found; |
56e93d26 | 1292 | |
0827b9e9 AA |
1293 | /* No dirty page as there is zero RAM */ |
1294 | if (!ram_bytes_total()) { | |
1295 | return pages; | |
1296 | } | |
1297 | ||
6f37bb8b | 1298 | pss.block = rs->last_seen_block; |
a935e30f | 1299 | pss.page = rs->last_page; |
b8fb8cb7 DDAG |
1300 | pss.complete_round = false; |
1301 | ||
1302 | if (!pss.block) { | |
1303 | pss.block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1304 | } | |
56e93d26 | 1305 | |
b9e60928 | 1306 | do { |
a82d593b | 1307 | again = true; |
f20e2865 | 1308 | found = get_queued_page(rs, &pss); |
b9e60928 | 1309 | |
a82d593b DDAG |
1310 | if (!found) { |
1311 | /* priority queue empty, so just search for something dirty */ | |
f20e2865 | 1312 | found = find_dirty_block(rs, &pss, &again); |
a82d593b | 1313 | } |
f3f491fc | 1314 | |
a82d593b | 1315 | if (found) { |
f20e2865 | 1316 | pages = ram_save_host_page(rs, &pss, last_stage); |
56e93d26 | 1317 | } |
b9e60928 | 1318 | } while (!pages && again); |
56e93d26 | 1319 | |
6f37bb8b | 1320 | rs->last_seen_block = pss.block; |
a935e30f | 1321 | rs->last_page = pss.page; |
56e93d26 JQ |
1322 | |
1323 | return pages; | |
1324 | } | |
1325 | ||
1326 | void acct_update_position(QEMUFile *f, size_t size, bool zero) | |
1327 | { | |
1328 | uint64_t pages = size / TARGET_PAGE_SIZE; | |
f7ccd61b | 1329 | |
56e93d26 | 1330 | if (zero) { |
9360447d | 1331 | ram_counters.duplicate += pages; |
56e93d26 | 1332 | } else { |
9360447d JQ |
1333 | ram_counters.normal += pages; |
1334 | ram_counters.transferred += size; | |
56e93d26 JQ |
1335 | qemu_update_position(f, size); |
1336 | } | |
1337 | } | |
1338 | ||
56e93d26 JQ |
1339 | uint64_t ram_bytes_total(void) |
1340 | { | |
1341 | RAMBlock *block; | |
1342 | uint64_t total = 0; | |
1343 | ||
1344 | rcu_read_lock(); | |
99e15582 | 1345 | RAMBLOCK_FOREACH(block) { |
56e93d26 | 1346 | total += block->used_length; |
99e15582 | 1347 | } |
56e93d26 JQ |
1348 | rcu_read_unlock(); |
1349 | return total; | |
1350 | } | |
1351 | ||
f265e0e4 | 1352 | static void xbzrle_load_setup(void) |
56e93d26 | 1353 | { |
f265e0e4 | 1354 | XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE); |
56e93d26 JQ |
1355 | } |
1356 | ||
f265e0e4 JQ |
1357 | static void xbzrle_load_cleanup(void) |
1358 | { | |
1359 | g_free(XBZRLE.decoded_buf); | |
1360 | XBZRLE.decoded_buf = NULL; | |
1361 | } | |
1362 | ||
1363 | static void ram_save_cleanup(void *opaque) | |
56e93d26 | 1364 | { |
53518d94 | 1365 | RAMState **rsp = opaque; |
6b6712ef | 1366 | RAMBlock *block; |
eb859c53 | 1367 | |
2ff64038 LZ |
1368 | /* caller have hold iothread lock or is in a bh, so there is |
1369 | * no writing race against this migration_bitmap | |
1370 | */ | |
6b6712ef JQ |
1371 | memory_global_dirty_log_stop(); |
1372 | ||
1373 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1374 | g_free(block->bmap); | |
1375 | block->bmap = NULL; | |
1376 | g_free(block->unsentmap); | |
1377 | block->unsentmap = NULL; | |
56e93d26 JQ |
1378 | } |
1379 | ||
1380 | XBZRLE_cache_lock(); | |
1381 | if (XBZRLE.cache) { | |
1382 | cache_fini(XBZRLE.cache); | |
1383 | g_free(XBZRLE.encoded_buf); | |
1384 | g_free(XBZRLE.current_buf); | |
c00e0928 | 1385 | g_free(XBZRLE.zero_target_page); |
56e93d26 JQ |
1386 | XBZRLE.cache = NULL; |
1387 | XBZRLE.encoded_buf = NULL; | |
1388 | XBZRLE.current_buf = NULL; | |
c00e0928 | 1389 | XBZRLE.zero_target_page = NULL; |
56e93d26 JQ |
1390 | } |
1391 | XBZRLE_cache_unlock(); | |
53518d94 JQ |
1392 | migration_page_queue_free(*rsp); |
1393 | g_free(*rsp); | |
1394 | *rsp = NULL; | |
56e93d26 JQ |
1395 | } |
1396 | ||
6f37bb8b | 1397 | static void ram_state_reset(RAMState *rs) |
56e93d26 | 1398 | { |
6f37bb8b JQ |
1399 | rs->last_seen_block = NULL; |
1400 | rs->last_sent_block = NULL; | |
269ace29 | 1401 | rs->last_page = 0; |
6f37bb8b JQ |
1402 | rs->last_version = ram_list.version; |
1403 | rs->ram_bulk_stage = true; | |
56e93d26 JQ |
1404 | } |
1405 | ||
1406 | #define MAX_WAIT 50 /* ms, half buffered_file limit */ | |
1407 | ||
4f2e4252 DDAG |
1408 | /* |
1409 | * 'expected' is the value you expect the bitmap mostly to be full | |
1410 | * of; it won't bother printing lines that are all this value. | |
1411 | * If 'todump' is null the migration bitmap is dumped. | |
1412 | */ | |
6b6712ef JQ |
1413 | void ram_debug_dump_bitmap(unsigned long *todump, bool expected, |
1414 | unsigned long pages) | |
4f2e4252 | 1415 | { |
4f2e4252 DDAG |
1416 | int64_t cur; |
1417 | int64_t linelen = 128; | |
1418 | char linebuf[129]; | |
1419 | ||
6b6712ef | 1420 | for (cur = 0; cur < pages; cur += linelen) { |
4f2e4252 DDAG |
1421 | int64_t curb; |
1422 | bool found = false; | |
1423 | /* | |
1424 | * Last line; catch the case where the line length | |
1425 | * is longer than remaining ram | |
1426 | */ | |
6b6712ef JQ |
1427 | if (cur + linelen > pages) { |
1428 | linelen = pages - cur; | |
4f2e4252 DDAG |
1429 | } |
1430 | for (curb = 0; curb < linelen; curb++) { | |
1431 | bool thisbit = test_bit(cur + curb, todump); | |
1432 | linebuf[curb] = thisbit ? '1' : '.'; | |
1433 | found = found || (thisbit != expected); | |
1434 | } | |
1435 | if (found) { | |
1436 | linebuf[curb] = '\0'; | |
1437 | fprintf(stderr, "0x%08" PRIx64 " : %s\n", cur, linebuf); | |
1438 | } | |
1439 | } | |
1440 | } | |
1441 | ||
e0b266f0 DDAG |
1442 | /* **** functions for postcopy ***** */ |
1443 | ||
ced1c616 PB |
1444 | void ram_postcopy_migrated_memory_release(MigrationState *ms) |
1445 | { | |
1446 | struct RAMBlock *block; | |
ced1c616 | 1447 | |
99e15582 | 1448 | RAMBLOCK_FOREACH(block) { |
6b6712ef JQ |
1449 | unsigned long *bitmap = block->bmap; |
1450 | unsigned long range = block->used_length >> TARGET_PAGE_BITS; | |
1451 | unsigned long run_start = find_next_zero_bit(bitmap, range, 0); | |
ced1c616 PB |
1452 | |
1453 | while (run_start < range) { | |
1454 | unsigned long run_end = find_next_bit(bitmap, range, run_start + 1); | |
aaa2064c | 1455 | ram_discard_range(block->idstr, run_start << TARGET_PAGE_BITS, |
ced1c616 PB |
1456 | (run_end - run_start) << TARGET_PAGE_BITS); |
1457 | run_start = find_next_zero_bit(bitmap, range, run_end + 1); | |
1458 | } | |
1459 | } | |
1460 | } | |
1461 | ||
3d0684b2 JQ |
1462 | /** |
1463 | * postcopy_send_discard_bm_ram: discard a RAMBlock | |
1464 | * | |
1465 | * Returns zero on success | |
1466 | * | |
e0b266f0 DDAG |
1467 | * Callback from postcopy_each_ram_send_discard for each RAMBlock |
1468 | * Note: At this point the 'unsentmap' is the processed bitmap combined | |
1469 | * with the dirtymap; so a '1' means it's either dirty or unsent. | |
3d0684b2 JQ |
1470 | * |
1471 | * @ms: current migration state | |
1472 | * @pds: state for postcopy | |
1473 | * @start: RAMBlock starting page | |
1474 | * @length: RAMBlock size | |
e0b266f0 DDAG |
1475 | */ |
1476 | static int postcopy_send_discard_bm_ram(MigrationState *ms, | |
1477 | PostcopyDiscardState *pds, | |
6b6712ef | 1478 | RAMBlock *block) |
e0b266f0 | 1479 | { |
6b6712ef | 1480 | unsigned long end = block->used_length >> TARGET_PAGE_BITS; |
e0b266f0 | 1481 | unsigned long current; |
6b6712ef | 1482 | unsigned long *unsentmap = block->unsentmap; |
e0b266f0 | 1483 | |
6b6712ef | 1484 | for (current = 0; current < end; ) { |
e0b266f0 DDAG |
1485 | unsigned long one = find_next_bit(unsentmap, end, current); |
1486 | ||
1487 | if (one <= end) { | |
1488 | unsigned long zero = find_next_zero_bit(unsentmap, end, one + 1); | |
1489 | unsigned long discard_length; | |
1490 | ||
1491 | if (zero >= end) { | |
1492 | discard_length = end - one; | |
1493 | } else { | |
1494 | discard_length = zero - one; | |
1495 | } | |
d688c62d DDAG |
1496 | if (discard_length) { |
1497 | postcopy_discard_send_range(ms, pds, one, discard_length); | |
1498 | } | |
e0b266f0 DDAG |
1499 | current = one + discard_length; |
1500 | } else { | |
1501 | current = one; | |
1502 | } | |
1503 | } | |
1504 | ||
1505 | return 0; | |
1506 | } | |
1507 | ||
3d0684b2 JQ |
1508 | /** |
1509 | * postcopy_each_ram_send_discard: discard all RAMBlocks | |
1510 | * | |
1511 | * Returns 0 for success or negative for error | |
1512 | * | |
e0b266f0 DDAG |
1513 | * Utility for the outgoing postcopy code. |
1514 | * Calls postcopy_send_discard_bm_ram for each RAMBlock | |
1515 | * passing it bitmap indexes and name. | |
e0b266f0 DDAG |
1516 | * (qemu_ram_foreach_block ends up passing unscaled lengths |
1517 | * which would mean postcopy code would have to deal with target page) | |
3d0684b2 JQ |
1518 | * |
1519 | * @ms: current migration state | |
e0b266f0 DDAG |
1520 | */ |
1521 | static int postcopy_each_ram_send_discard(MigrationState *ms) | |
1522 | { | |
1523 | struct RAMBlock *block; | |
1524 | int ret; | |
1525 | ||
99e15582 | 1526 | RAMBLOCK_FOREACH(block) { |
6b6712ef JQ |
1527 | PostcopyDiscardState *pds = |
1528 | postcopy_discard_send_init(ms, block->idstr); | |
e0b266f0 DDAG |
1529 | |
1530 | /* | |
1531 | * Postcopy sends chunks of bitmap over the wire, but it | |
1532 | * just needs indexes at this point, avoids it having | |
1533 | * target page specific code. | |
1534 | */ | |
6b6712ef | 1535 | ret = postcopy_send_discard_bm_ram(ms, pds, block); |
e0b266f0 DDAG |
1536 | postcopy_discard_send_finish(ms, pds); |
1537 | if (ret) { | |
1538 | return ret; | |
1539 | } | |
1540 | } | |
1541 | ||
1542 | return 0; | |
1543 | } | |
1544 | ||
3d0684b2 JQ |
1545 | /** |
1546 | * postcopy_chunk_hostpages_pass: canocalize bitmap in hostpages | |
1547 | * | |
1548 | * Helper for postcopy_chunk_hostpages; it's called twice to | |
1549 | * canonicalize the two bitmaps, that are similar, but one is | |
1550 | * inverted. | |
99e314eb | 1551 | * |
3d0684b2 JQ |
1552 | * Postcopy requires that all target pages in a hostpage are dirty or |
1553 | * clean, not a mix. This function canonicalizes the bitmaps. | |
99e314eb | 1554 | * |
3d0684b2 JQ |
1555 | * @ms: current migration state |
1556 | * @unsent_pass: if true we need to canonicalize partially unsent host pages | |
1557 | * otherwise we need to canonicalize partially dirty host pages | |
1558 | * @block: block that contains the page we want to canonicalize | |
1559 | * @pds: state for postcopy | |
99e314eb DDAG |
1560 | */ |
1561 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, bool unsent_pass, | |
1562 | RAMBlock *block, | |
1563 | PostcopyDiscardState *pds) | |
1564 | { | |
53518d94 | 1565 | RAMState *rs = ram_state; |
6b6712ef JQ |
1566 | unsigned long *bitmap = block->bmap; |
1567 | unsigned long *unsentmap = block->unsentmap; | |
29c59172 | 1568 | unsigned int host_ratio = block->page_size / TARGET_PAGE_SIZE; |
6b6712ef | 1569 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
99e314eb DDAG |
1570 | unsigned long run_start; |
1571 | ||
29c59172 DDAG |
1572 | if (block->page_size == TARGET_PAGE_SIZE) { |
1573 | /* Easy case - TPS==HPS for a non-huge page RAMBlock */ | |
1574 | return; | |
1575 | } | |
1576 | ||
99e314eb DDAG |
1577 | if (unsent_pass) { |
1578 | /* Find a sent page */ | |
6b6712ef | 1579 | run_start = find_next_zero_bit(unsentmap, pages, 0); |
99e314eb DDAG |
1580 | } else { |
1581 | /* Find a dirty page */ | |
6b6712ef | 1582 | run_start = find_next_bit(bitmap, pages, 0); |
99e314eb DDAG |
1583 | } |
1584 | ||
6b6712ef | 1585 | while (run_start < pages) { |
99e314eb DDAG |
1586 | bool do_fixup = false; |
1587 | unsigned long fixup_start_addr; | |
1588 | unsigned long host_offset; | |
1589 | ||
1590 | /* | |
1591 | * If the start of this run of pages is in the middle of a host | |
1592 | * page, then we need to fixup this host page. | |
1593 | */ | |
1594 | host_offset = run_start % host_ratio; | |
1595 | if (host_offset) { | |
1596 | do_fixup = true; | |
1597 | run_start -= host_offset; | |
1598 | fixup_start_addr = run_start; | |
1599 | /* For the next pass */ | |
1600 | run_start = run_start + host_ratio; | |
1601 | } else { | |
1602 | /* Find the end of this run */ | |
1603 | unsigned long run_end; | |
1604 | if (unsent_pass) { | |
6b6712ef | 1605 | run_end = find_next_bit(unsentmap, pages, run_start + 1); |
99e314eb | 1606 | } else { |
6b6712ef | 1607 | run_end = find_next_zero_bit(bitmap, pages, run_start + 1); |
99e314eb DDAG |
1608 | } |
1609 | /* | |
1610 | * If the end isn't at the start of a host page, then the | |
1611 | * run doesn't finish at the end of a host page | |
1612 | * and we need to discard. | |
1613 | */ | |
1614 | host_offset = run_end % host_ratio; | |
1615 | if (host_offset) { | |
1616 | do_fixup = true; | |
1617 | fixup_start_addr = run_end - host_offset; | |
1618 | /* | |
1619 | * This host page has gone, the next loop iteration starts | |
1620 | * from after the fixup | |
1621 | */ | |
1622 | run_start = fixup_start_addr + host_ratio; | |
1623 | } else { | |
1624 | /* | |
1625 | * No discards on this iteration, next loop starts from | |
1626 | * next sent/dirty page | |
1627 | */ | |
1628 | run_start = run_end + 1; | |
1629 | } | |
1630 | } | |
1631 | ||
1632 | if (do_fixup) { | |
1633 | unsigned long page; | |
1634 | ||
1635 | /* Tell the destination to discard this page */ | |
1636 | if (unsent_pass || !test_bit(fixup_start_addr, unsentmap)) { | |
1637 | /* For the unsent_pass we: | |
1638 | * discard partially sent pages | |
1639 | * For the !unsent_pass (dirty) we: | |
1640 | * discard partially dirty pages that were sent | |
1641 | * (any partially sent pages were already discarded | |
1642 | * by the previous unsent_pass) | |
1643 | */ | |
1644 | postcopy_discard_send_range(ms, pds, fixup_start_addr, | |
1645 | host_ratio); | |
1646 | } | |
1647 | ||
1648 | /* Clean up the bitmap */ | |
1649 | for (page = fixup_start_addr; | |
1650 | page < fixup_start_addr + host_ratio; page++) { | |
1651 | /* All pages in this host page are now not sent */ | |
1652 | set_bit(page, unsentmap); | |
1653 | ||
1654 | /* | |
1655 | * Remark them as dirty, updating the count for any pages | |
1656 | * that weren't previously dirty. | |
1657 | */ | |
0d8ec885 | 1658 | rs->migration_dirty_pages += !test_and_set_bit(page, bitmap); |
99e314eb DDAG |
1659 | } |
1660 | } | |
1661 | ||
1662 | if (unsent_pass) { | |
1663 | /* Find the next sent page for the next iteration */ | |
6b6712ef | 1664 | run_start = find_next_zero_bit(unsentmap, pages, run_start); |
99e314eb DDAG |
1665 | } else { |
1666 | /* Find the next dirty page for the next iteration */ | |
6b6712ef | 1667 | run_start = find_next_bit(bitmap, pages, run_start); |
99e314eb DDAG |
1668 | } |
1669 | } | |
1670 | } | |
1671 | ||
3d0684b2 JQ |
1672 | /** |
1673 | * postcopy_chuck_hostpages: discrad any partially sent host page | |
1674 | * | |
99e314eb DDAG |
1675 | * Utility for the outgoing postcopy code. |
1676 | * | |
1677 | * Discard any partially sent host-page size chunks, mark any partially | |
29c59172 DDAG |
1678 | * dirty host-page size chunks as all dirty. In this case the host-page |
1679 | * is the host-page for the particular RAMBlock, i.e. it might be a huge page | |
99e314eb | 1680 | * |
3d0684b2 JQ |
1681 | * Returns zero on success |
1682 | * | |
1683 | * @ms: current migration state | |
6b6712ef | 1684 | * @block: block we want to work with |
99e314eb | 1685 | */ |
6b6712ef | 1686 | static int postcopy_chunk_hostpages(MigrationState *ms, RAMBlock *block) |
99e314eb | 1687 | { |
6b6712ef JQ |
1688 | PostcopyDiscardState *pds = |
1689 | postcopy_discard_send_init(ms, block->idstr); | |
99e314eb | 1690 | |
6b6712ef JQ |
1691 | /* First pass: Discard all partially sent host pages */ |
1692 | postcopy_chunk_hostpages_pass(ms, true, block, pds); | |
1693 | /* | |
1694 | * Second pass: Ensure that all partially dirty host pages are made | |
1695 | * fully dirty. | |
1696 | */ | |
1697 | postcopy_chunk_hostpages_pass(ms, false, block, pds); | |
99e314eb | 1698 | |
6b6712ef | 1699 | postcopy_discard_send_finish(ms, pds); |
99e314eb DDAG |
1700 | return 0; |
1701 | } | |
1702 | ||
3d0684b2 JQ |
1703 | /** |
1704 | * ram_postcopy_send_discard_bitmap: transmit the discard bitmap | |
1705 | * | |
1706 | * Returns zero on success | |
1707 | * | |
e0b266f0 DDAG |
1708 | * Transmit the set of pages to be discarded after precopy to the target |
1709 | * these are pages that: | |
1710 | * a) Have been previously transmitted but are now dirty again | |
1711 | * b) Pages that have never been transmitted, this ensures that | |
1712 | * any pages on the destination that have been mapped by background | |
1713 | * tasks get discarded (transparent huge pages is the specific concern) | |
1714 | * Hopefully this is pretty sparse | |
3d0684b2 JQ |
1715 | * |
1716 | * @ms: current migration state | |
e0b266f0 DDAG |
1717 | */ |
1718 | int ram_postcopy_send_discard_bitmap(MigrationState *ms) | |
1719 | { | |
53518d94 | 1720 | RAMState *rs = ram_state; |
6b6712ef | 1721 | RAMBlock *block; |
e0b266f0 | 1722 | int ret; |
e0b266f0 DDAG |
1723 | |
1724 | rcu_read_lock(); | |
1725 | ||
1726 | /* This should be our last sync, the src is now paused */ | |
eb859c53 | 1727 | migration_bitmap_sync(rs); |
e0b266f0 | 1728 | |
6b6712ef JQ |
1729 | /* Easiest way to make sure we don't resume in the middle of a host-page */ |
1730 | rs->last_seen_block = NULL; | |
1731 | rs->last_sent_block = NULL; | |
1732 | rs->last_page = 0; | |
e0b266f0 | 1733 | |
6b6712ef JQ |
1734 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { |
1735 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; | |
1736 | unsigned long *bitmap = block->bmap; | |
1737 | unsigned long *unsentmap = block->unsentmap; | |
1738 | ||
1739 | if (!unsentmap) { | |
1740 | /* We don't have a safe way to resize the sentmap, so | |
1741 | * if the bitmap was resized it will be NULL at this | |
1742 | * point. | |
1743 | */ | |
1744 | error_report("migration ram resized during precopy phase"); | |
1745 | rcu_read_unlock(); | |
1746 | return -EINVAL; | |
1747 | } | |
1748 | /* Deal with TPS != HPS and huge pages */ | |
1749 | ret = postcopy_chunk_hostpages(ms, block); | |
1750 | if (ret) { | |
1751 | rcu_read_unlock(); | |
1752 | return ret; | |
1753 | } | |
e0b266f0 | 1754 | |
6b6712ef JQ |
1755 | /* |
1756 | * Update the unsentmap to be unsentmap = unsentmap | dirty | |
1757 | */ | |
1758 | bitmap_or(unsentmap, unsentmap, bitmap, pages); | |
e0b266f0 | 1759 | #ifdef DEBUG_POSTCOPY |
6b6712ef | 1760 | ram_debug_dump_bitmap(unsentmap, true, pages); |
e0b266f0 | 1761 | #endif |
6b6712ef JQ |
1762 | } |
1763 | trace_ram_postcopy_send_discard_bitmap(); | |
e0b266f0 DDAG |
1764 | |
1765 | ret = postcopy_each_ram_send_discard(ms); | |
1766 | rcu_read_unlock(); | |
1767 | ||
1768 | return ret; | |
1769 | } | |
1770 | ||
3d0684b2 JQ |
1771 | /** |
1772 | * ram_discard_range: discard dirtied pages at the beginning of postcopy | |
e0b266f0 | 1773 | * |
3d0684b2 | 1774 | * Returns zero on success |
e0b266f0 | 1775 | * |
36449157 JQ |
1776 | * @rbname: name of the RAMBlock of the request. NULL means the |
1777 | * same that last one. | |
3d0684b2 JQ |
1778 | * @start: RAMBlock starting page |
1779 | * @length: RAMBlock size | |
e0b266f0 | 1780 | */ |
aaa2064c | 1781 | int ram_discard_range(const char *rbname, uint64_t start, size_t length) |
e0b266f0 DDAG |
1782 | { |
1783 | int ret = -1; | |
1784 | ||
36449157 | 1785 | trace_ram_discard_range(rbname, start, length); |
d3a5038c | 1786 | |
e0b266f0 | 1787 | rcu_read_lock(); |
36449157 | 1788 | RAMBlock *rb = qemu_ram_block_by_name(rbname); |
e0b266f0 DDAG |
1789 | |
1790 | if (!rb) { | |
36449157 | 1791 | error_report("ram_discard_range: Failed to find block '%s'", rbname); |
e0b266f0 DDAG |
1792 | goto err; |
1793 | } | |
1794 | ||
d3a5038c | 1795 | ret = ram_block_discard_range(rb, start, length); |
e0b266f0 DDAG |
1796 | |
1797 | err: | |
1798 | rcu_read_unlock(); | |
1799 | ||
1800 | return ret; | |
1801 | } | |
1802 | ||
53518d94 | 1803 | static int ram_state_init(RAMState **rsp) |
56e93d26 | 1804 | { |
53518d94 JQ |
1805 | *rsp = g_new0(RAMState, 1); |
1806 | ||
1807 | qemu_mutex_init(&(*rsp)->bitmap_mutex); | |
1808 | qemu_mutex_init(&(*rsp)->src_page_req_mutex); | |
1809 | QSIMPLEQ_INIT(&(*rsp)->src_page_requests); | |
56e93d26 JQ |
1810 | |
1811 | if (migrate_use_xbzrle()) { | |
1812 | XBZRLE_cache_lock(); | |
c00e0928 | 1813 | XBZRLE.zero_target_page = g_malloc0(TARGET_PAGE_SIZE); |
56e93d26 JQ |
1814 | XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() / |
1815 | TARGET_PAGE_SIZE, | |
1816 | TARGET_PAGE_SIZE); | |
1817 | if (!XBZRLE.cache) { | |
1818 | XBZRLE_cache_unlock(); | |
1819 | error_report("Error creating cache"); | |
53518d94 JQ |
1820 | g_free(*rsp); |
1821 | *rsp = NULL; | |
56e93d26 JQ |
1822 | return -1; |
1823 | } | |
1824 | XBZRLE_cache_unlock(); | |
1825 | ||
1826 | /* We prefer not to abort if there is no memory */ | |
1827 | XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); | |
1828 | if (!XBZRLE.encoded_buf) { | |
1829 | error_report("Error allocating encoded_buf"); | |
53518d94 JQ |
1830 | g_free(*rsp); |
1831 | *rsp = NULL; | |
56e93d26 JQ |
1832 | return -1; |
1833 | } | |
1834 | ||
1835 | XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); | |
1836 | if (!XBZRLE.current_buf) { | |
1837 | error_report("Error allocating current_buf"); | |
1838 | g_free(XBZRLE.encoded_buf); | |
1839 | XBZRLE.encoded_buf = NULL; | |
53518d94 JQ |
1840 | g_free(*rsp); |
1841 | *rsp = NULL; | |
56e93d26 JQ |
1842 | return -1; |
1843 | } | |
56e93d26 JQ |
1844 | } |
1845 | ||
49877834 PB |
1846 | /* For memory_global_dirty_log_start below. */ |
1847 | qemu_mutex_lock_iothread(); | |
1848 | ||
56e93d26 JQ |
1849 | qemu_mutex_lock_ramlist(); |
1850 | rcu_read_lock(); | |
53518d94 | 1851 | ram_state_reset(*rsp); |
56e93d26 | 1852 | |
0827b9e9 AA |
1853 | /* Skip setting bitmap if there is no RAM */ |
1854 | if (ram_bytes_total()) { | |
6b6712ef JQ |
1855 | RAMBlock *block; |
1856 | ||
1857 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1858 | unsigned long pages = block->max_length >> TARGET_PAGE_BITS; | |
0827b9e9 | 1859 | |
6b6712ef JQ |
1860 | block->bmap = bitmap_new(pages); |
1861 | bitmap_set(block->bmap, 0, pages); | |
1862 | if (migrate_postcopy_ram()) { | |
1863 | block->unsentmap = bitmap_new(pages); | |
1864 | bitmap_set(block->unsentmap, 0, pages); | |
1865 | } | |
0827b9e9 | 1866 | } |
f3f491fc DDAG |
1867 | } |
1868 | ||
56e93d26 JQ |
1869 | /* |
1870 | * Count the total number of pages used by ram blocks not including any | |
1871 | * gaps due to alignment or unplugs. | |
1872 | */ | |
53518d94 | 1873 | (*rsp)->migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS; |
56e93d26 JQ |
1874 | |
1875 | memory_global_dirty_log_start(); | |
53518d94 | 1876 | migration_bitmap_sync(*rsp); |
56e93d26 | 1877 | qemu_mutex_unlock_ramlist(); |
49877834 | 1878 | qemu_mutex_unlock_iothread(); |
a91246c9 HZ |
1879 | rcu_read_unlock(); |
1880 | ||
1881 | return 0; | |
1882 | } | |
1883 | ||
3d0684b2 JQ |
1884 | /* |
1885 | * Each of ram_save_setup, ram_save_iterate and ram_save_complete has | |
a91246c9 HZ |
1886 | * long-running RCU critical section. When rcu-reclaims in the code |
1887 | * start to become numerous it will be necessary to reduce the | |
1888 | * granularity of these critical sections. | |
1889 | */ | |
1890 | ||
3d0684b2 JQ |
1891 | /** |
1892 | * ram_save_setup: Setup RAM for migration | |
1893 | * | |
1894 | * Returns zero to indicate success and negative for error | |
1895 | * | |
1896 | * @f: QEMUFile where to send the data | |
1897 | * @opaque: RAMState pointer | |
1898 | */ | |
a91246c9 HZ |
1899 | static int ram_save_setup(QEMUFile *f, void *opaque) |
1900 | { | |
53518d94 | 1901 | RAMState **rsp = opaque; |
a91246c9 HZ |
1902 | RAMBlock *block; |
1903 | ||
1904 | /* migration has already setup the bitmap, reuse it. */ | |
1905 | if (!migration_in_colo_state()) { | |
53518d94 | 1906 | if (ram_state_init(rsp) != 0) { |
a91246c9 | 1907 | return -1; |
53518d94 | 1908 | } |
a91246c9 | 1909 | } |
53518d94 | 1910 | (*rsp)->f = f; |
a91246c9 HZ |
1911 | |
1912 | rcu_read_lock(); | |
56e93d26 JQ |
1913 | |
1914 | qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE); | |
1915 | ||
99e15582 | 1916 | RAMBLOCK_FOREACH(block) { |
56e93d26 JQ |
1917 | qemu_put_byte(f, strlen(block->idstr)); |
1918 | qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); | |
1919 | qemu_put_be64(f, block->used_length); | |
ef08fb38 DDAG |
1920 | if (migrate_postcopy_ram() && block->page_size != qemu_host_page_size) { |
1921 | qemu_put_be64(f, block->page_size); | |
1922 | } | |
56e93d26 JQ |
1923 | } |
1924 | ||
1925 | rcu_read_unlock(); | |
1926 | ||
1927 | ram_control_before_iterate(f, RAM_CONTROL_SETUP); | |
1928 | ram_control_after_iterate(f, RAM_CONTROL_SETUP); | |
1929 | ||
1930 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); | |
1931 | ||
1932 | return 0; | |
1933 | } | |
1934 | ||
3d0684b2 JQ |
1935 | /** |
1936 | * ram_save_iterate: iterative stage for migration | |
1937 | * | |
1938 | * Returns zero to indicate success and negative for error | |
1939 | * | |
1940 | * @f: QEMUFile where to send the data | |
1941 | * @opaque: RAMState pointer | |
1942 | */ | |
56e93d26 JQ |
1943 | static int ram_save_iterate(QEMUFile *f, void *opaque) |
1944 | { | |
53518d94 JQ |
1945 | RAMState **temp = opaque; |
1946 | RAMState *rs = *temp; | |
56e93d26 JQ |
1947 | int ret; |
1948 | int i; | |
1949 | int64_t t0; | |
5c90308f | 1950 | int done = 0; |
56e93d26 JQ |
1951 | |
1952 | rcu_read_lock(); | |
6f37bb8b JQ |
1953 | if (ram_list.version != rs->last_version) { |
1954 | ram_state_reset(rs); | |
56e93d26 JQ |
1955 | } |
1956 | ||
1957 | /* Read version before ram_list.blocks */ | |
1958 | smp_rmb(); | |
1959 | ||
1960 | ram_control_before_iterate(f, RAM_CONTROL_ROUND); | |
1961 | ||
1962 | t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
1963 | i = 0; | |
1964 | while ((ret = qemu_file_rate_limit(f)) == 0) { | |
1965 | int pages; | |
1966 | ||
ce25d337 | 1967 | pages = ram_find_and_save_block(rs, false); |
56e93d26 JQ |
1968 | /* no more pages to sent */ |
1969 | if (pages == 0) { | |
5c90308f | 1970 | done = 1; |
56e93d26 JQ |
1971 | break; |
1972 | } | |
23b28c3c | 1973 | rs->iterations++; |
070afca2 | 1974 | |
56e93d26 JQ |
1975 | /* we want to check in the 1st loop, just in case it was the 1st time |
1976 | and we had to sync the dirty bitmap. | |
1977 | qemu_get_clock_ns() is a bit expensive, so we only check each some | |
1978 | iterations | |
1979 | */ | |
1980 | if ((i & 63) == 0) { | |
1981 | uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000; | |
1982 | if (t1 > MAX_WAIT) { | |
55c4446b | 1983 | trace_ram_save_iterate_big_wait(t1, i); |
56e93d26 JQ |
1984 | break; |
1985 | } | |
1986 | } | |
1987 | i++; | |
1988 | } | |
ce25d337 | 1989 | flush_compressed_data(rs); |
56e93d26 JQ |
1990 | rcu_read_unlock(); |
1991 | ||
1992 | /* | |
1993 | * Must occur before EOS (or any QEMUFile operation) | |
1994 | * because of RDMA protocol. | |
1995 | */ | |
1996 | ram_control_after_iterate(f, RAM_CONTROL_ROUND); | |
1997 | ||
1998 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); | |
9360447d | 1999 | ram_counters.transferred += 8; |
56e93d26 JQ |
2000 | |
2001 | ret = qemu_file_get_error(f); | |
2002 | if (ret < 0) { | |
2003 | return ret; | |
2004 | } | |
2005 | ||
5c90308f | 2006 | return done; |
56e93d26 JQ |
2007 | } |
2008 | ||
3d0684b2 JQ |
2009 | /** |
2010 | * ram_save_complete: function called to send the remaining amount of ram | |
2011 | * | |
2012 | * Returns zero to indicate success | |
2013 | * | |
2014 | * Called with iothread lock | |
2015 | * | |
2016 | * @f: QEMUFile where to send the data | |
2017 | * @opaque: RAMState pointer | |
2018 | */ | |
56e93d26 JQ |
2019 | static int ram_save_complete(QEMUFile *f, void *opaque) |
2020 | { | |
53518d94 JQ |
2021 | RAMState **temp = opaque; |
2022 | RAMState *rs = *temp; | |
6f37bb8b | 2023 | |
56e93d26 JQ |
2024 | rcu_read_lock(); |
2025 | ||
5727309d | 2026 | if (!migration_in_postcopy()) { |
8d820d6f | 2027 | migration_bitmap_sync(rs); |
663e6c1d | 2028 | } |
56e93d26 JQ |
2029 | |
2030 | ram_control_before_iterate(f, RAM_CONTROL_FINISH); | |
2031 | ||
2032 | /* try transferring iterative blocks of memory */ | |
2033 | ||
2034 | /* flush all remaining blocks regardless of rate limiting */ | |
2035 | while (true) { | |
2036 | int pages; | |
2037 | ||
ce25d337 | 2038 | pages = ram_find_and_save_block(rs, !migration_in_colo_state()); |
56e93d26 JQ |
2039 | /* no more blocks to sent */ |
2040 | if (pages == 0) { | |
2041 | break; | |
2042 | } | |
2043 | } | |
2044 | ||
ce25d337 | 2045 | flush_compressed_data(rs); |
56e93d26 | 2046 | ram_control_after_iterate(f, RAM_CONTROL_FINISH); |
56e93d26 JQ |
2047 | |
2048 | rcu_read_unlock(); | |
d09a6fde | 2049 | |
56e93d26 JQ |
2050 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
2051 | ||
2052 | return 0; | |
2053 | } | |
2054 | ||
c31b098f DDAG |
2055 | static void ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size, |
2056 | uint64_t *non_postcopiable_pending, | |
2057 | uint64_t *postcopiable_pending) | |
56e93d26 | 2058 | { |
53518d94 JQ |
2059 | RAMState **temp = opaque; |
2060 | RAMState *rs = *temp; | |
56e93d26 JQ |
2061 | uint64_t remaining_size; |
2062 | ||
9edabd4d | 2063 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2064 | |
5727309d | 2065 | if (!migration_in_postcopy() && |
663e6c1d | 2066 | remaining_size < max_size) { |
56e93d26 JQ |
2067 | qemu_mutex_lock_iothread(); |
2068 | rcu_read_lock(); | |
8d820d6f | 2069 | migration_bitmap_sync(rs); |
56e93d26 JQ |
2070 | rcu_read_unlock(); |
2071 | qemu_mutex_unlock_iothread(); | |
9edabd4d | 2072 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2073 | } |
c31b098f DDAG |
2074 | |
2075 | /* We can do postcopy, and all the data is postcopiable */ | |
2076 | *postcopiable_pending += remaining_size; | |
56e93d26 JQ |
2077 | } |
2078 | ||
2079 | static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) | |
2080 | { | |
2081 | unsigned int xh_len; | |
2082 | int xh_flags; | |
063e760a | 2083 | uint8_t *loaded_data; |
56e93d26 | 2084 | |
56e93d26 JQ |
2085 | /* extract RLE header */ |
2086 | xh_flags = qemu_get_byte(f); | |
2087 | xh_len = qemu_get_be16(f); | |
2088 | ||
2089 | if (xh_flags != ENCODING_FLAG_XBZRLE) { | |
2090 | error_report("Failed to load XBZRLE page - wrong compression!"); | |
2091 | return -1; | |
2092 | } | |
2093 | ||
2094 | if (xh_len > TARGET_PAGE_SIZE) { | |
2095 | error_report("Failed to load XBZRLE page - len overflow!"); | |
2096 | return -1; | |
2097 | } | |
f265e0e4 | 2098 | loaded_data = XBZRLE.decoded_buf; |
56e93d26 | 2099 | /* load data and decode */ |
f265e0e4 | 2100 | /* it can change loaded_data to point to an internal buffer */ |
063e760a | 2101 | qemu_get_buffer_in_place(f, &loaded_data, xh_len); |
56e93d26 JQ |
2102 | |
2103 | /* decode RLE */ | |
063e760a | 2104 | if (xbzrle_decode_buffer(loaded_data, xh_len, host, |
56e93d26 JQ |
2105 | TARGET_PAGE_SIZE) == -1) { |
2106 | error_report("Failed to load XBZRLE page - decode error!"); | |
2107 | return -1; | |
2108 | } | |
2109 | ||
2110 | return 0; | |
2111 | } | |
2112 | ||
3d0684b2 JQ |
2113 | /** |
2114 | * ram_block_from_stream: read a RAMBlock id from the migration stream | |
2115 | * | |
2116 | * Must be called from within a rcu critical section. | |
2117 | * | |
56e93d26 | 2118 | * Returns a pointer from within the RCU-protected ram_list. |
a7180877 | 2119 | * |
3d0684b2 JQ |
2120 | * @f: QEMUFile where to read the data from |
2121 | * @flags: Page flags (mostly to see if it's a continuation of previous block) | |
a7180877 | 2122 | */ |
3d0684b2 | 2123 | static inline RAMBlock *ram_block_from_stream(QEMUFile *f, int flags) |
56e93d26 JQ |
2124 | { |
2125 | static RAMBlock *block = NULL; | |
2126 | char id[256]; | |
2127 | uint8_t len; | |
2128 | ||
2129 | if (flags & RAM_SAVE_FLAG_CONTINUE) { | |
4c4bad48 | 2130 | if (!block) { |
56e93d26 JQ |
2131 | error_report("Ack, bad migration stream!"); |
2132 | return NULL; | |
2133 | } | |
4c4bad48 | 2134 | return block; |
56e93d26 JQ |
2135 | } |
2136 | ||
2137 | len = qemu_get_byte(f); | |
2138 | qemu_get_buffer(f, (uint8_t *)id, len); | |
2139 | id[len] = 0; | |
2140 | ||
e3dd7493 | 2141 | block = qemu_ram_block_by_name(id); |
4c4bad48 HZ |
2142 | if (!block) { |
2143 | error_report("Can't find block %s", id); | |
2144 | return NULL; | |
56e93d26 JQ |
2145 | } |
2146 | ||
4c4bad48 HZ |
2147 | return block; |
2148 | } | |
2149 | ||
2150 | static inline void *host_from_ram_block_offset(RAMBlock *block, | |
2151 | ram_addr_t offset) | |
2152 | { | |
2153 | if (!offset_in_ramblock(block, offset)) { | |
2154 | return NULL; | |
2155 | } | |
2156 | ||
2157 | return block->host + offset; | |
56e93d26 JQ |
2158 | } |
2159 | ||
3d0684b2 JQ |
2160 | /** |
2161 | * ram_handle_compressed: handle the zero page case | |
2162 | * | |
56e93d26 JQ |
2163 | * If a page (or a whole RDMA chunk) has been |
2164 | * determined to be zero, then zap it. | |
3d0684b2 JQ |
2165 | * |
2166 | * @host: host address for the zero page | |
2167 | * @ch: what the page is filled from. We only support zero | |
2168 | * @size: size of the zero page | |
56e93d26 JQ |
2169 | */ |
2170 | void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) | |
2171 | { | |
2172 | if (ch != 0 || !is_zero_range(host, size)) { | |
2173 | memset(host, ch, size); | |
2174 | } | |
2175 | } | |
2176 | ||
2177 | static void *do_data_decompress(void *opaque) | |
2178 | { | |
2179 | DecompressParam *param = opaque; | |
2180 | unsigned long pagesize; | |
33d151f4 LL |
2181 | uint8_t *des; |
2182 | int len; | |
56e93d26 | 2183 | |
33d151f4 | 2184 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 2185 | while (!param->quit) { |
33d151f4 LL |
2186 | if (param->des) { |
2187 | des = param->des; | |
2188 | len = param->len; | |
2189 | param->des = 0; | |
2190 | qemu_mutex_unlock(¶m->mutex); | |
2191 | ||
56e93d26 | 2192 | pagesize = TARGET_PAGE_SIZE; |
73a8912b LL |
2193 | /* uncompress() will return failed in some case, especially |
2194 | * when the page is dirted when doing the compression, it's | |
2195 | * not a problem because the dirty page will be retransferred | |
2196 | * and uncompress() won't break the data in other pages. | |
2197 | */ | |
33d151f4 LL |
2198 | uncompress((Bytef *)des, &pagesize, |
2199 | (const Bytef *)param->compbuf, len); | |
73a8912b | 2200 | |
33d151f4 LL |
2201 | qemu_mutex_lock(&decomp_done_lock); |
2202 | param->done = true; | |
2203 | qemu_cond_signal(&decomp_done_cond); | |
2204 | qemu_mutex_unlock(&decomp_done_lock); | |
2205 | ||
2206 | qemu_mutex_lock(¶m->mutex); | |
2207 | } else { | |
2208 | qemu_cond_wait(¶m->cond, ¶m->mutex); | |
2209 | } | |
56e93d26 | 2210 | } |
33d151f4 | 2211 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
2212 | |
2213 | return NULL; | |
2214 | } | |
2215 | ||
5533b2e9 LL |
2216 | static void wait_for_decompress_done(void) |
2217 | { | |
2218 | int idx, thread_count; | |
2219 | ||
2220 | if (!migrate_use_compression()) { | |
2221 | return; | |
2222 | } | |
2223 | ||
2224 | thread_count = migrate_decompress_threads(); | |
2225 | qemu_mutex_lock(&decomp_done_lock); | |
2226 | for (idx = 0; idx < thread_count; idx++) { | |
2227 | while (!decomp_param[idx].done) { | |
2228 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
2229 | } | |
2230 | } | |
2231 | qemu_mutex_unlock(&decomp_done_lock); | |
2232 | } | |
2233 | ||
56e93d26 JQ |
2234 | void migrate_decompress_threads_create(void) |
2235 | { | |
2236 | int i, thread_count; | |
2237 | ||
3416ab5b JQ |
2238 | if (!migrate_use_compression()) { |
2239 | return; | |
2240 | } | |
56e93d26 JQ |
2241 | thread_count = migrate_decompress_threads(); |
2242 | decompress_threads = g_new0(QemuThread, thread_count); | |
2243 | decomp_param = g_new0(DecompressParam, thread_count); | |
73a8912b LL |
2244 | qemu_mutex_init(&decomp_done_lock); |
2245 | qemu_cond_init(&decomp_done_cond); | |
56e93d26 JQ |
2246 | for (i = 0; i < thread_count; i++) { |
2247 | qemu_mutex_init(&decomp_param[i].mutex); | |
2248 | qemu_cond_init(&decomp_param[i].cond); | |
2249 | decomp_param[i].compbuf = g_malloc0(compressBound(TARGET_PAGE_SIZE)); | |
73a8912b | 2250 | decomp_param[i].done = true; |
90e56fb4 | 2251 | decomp_param[i].quit = false; |
56e93d26 JQ |
2252 | qemu_thread_create(decompress_threads + i, "decompress", |
2253 | do_data_decompress, decomp_param + i, | |
2254 | QEMU_THREAD_JOINABLE); | |
2255 | } | |
2256 | } | |
2257 | ||
2258 | void migrate_decompress_threads_join(void) | |
2259 | { | |
2260 | int i, thread_count; | |
2261 | ||
3416ab5b JQ |
2262 | if (!migrate_use_compression()) { |
2263 | return; | |
2264 | } | |
56e93d26 JQ |
2265 | thread_count = migrate_decompress_threads(); |
2266 | for (i = 0; i < thread_count; i++) { | |
2267 | qemu_mutex_lock(&decomp_param[i].mutex); | |
90e56fb4 | 2268 | decomp_param[i].quit = true; |
56e93d26 JQ |
2269 | qemu_cond_signal(&decomp_param[i].cond); |
2270 | qemu_mutex_unlock(&decomp_param[i].mutex); | |
2271 | } | |
2272 | for (i = 0; i < thread_count; i++) { | |
2273 | qemu_thread_join(decompress_threads + i); | |
2274 | qemu_mutex_destroy(&decomp_param[i].mutex); | |
2275 | qemu_cond_destroy(&decomp_param[i].cond); | |
2276 | g_free(decomp_param[i].compbuf); | |
2277 | } | |
2278 | g_free(decompress_threads); | |
2279 | g_free(decomp_param); | |
56e93d26 JQ |
2280 | decompress_threads = NULL; |
2281 | decomp_param = NULL; | |
56e93d26 JQ |
2282 | } |
2283 | ||
c1bc6626 | 2284 | static void decompress_data_with_multi_threads(QEMUFile *f, |
56e93d26 JQ |
2285 | void *host, int len) |
2286 | { | |
2287 | int idx, thread_count; | |
2288 | ||
2289 | thread_count = migrate_decompress_threads(); | |
73a8912b | 2290 | qemu_mutex_lock(&decomp_done_lock); |
56e93d26 JQ |
2291 | while (true) { |
2292 | for (idx = 0; idx < thread_count; idx++) { | |
73a8912b | 2293 | if (decomp_param[idx].done) { |
33d151f4 LL |
2294 | decomp_param[idx].done = false; |
2295 | qemu_mutex_lock(&decomp_param[idx].mutex); | |
c1bc6626 | 2296 | qemu_get_buffer(f, decomp_param[idx].compbuf, len); |
56e93d26 JQ |
2297 | decomp_param[idx].des = host; |
2298 | decomp_param[idx].len = len; | |
33d151f4 LL |
2299 | qemu_cond_signal(&decomp_param[idx].cond); |
2300 | qemu_mutex_unlock(&decomp_param[idx].mutex); | |
56e93d26 JQ |
2301 | break; |
2302 | } | |
2303 | } | |
2304 | if (idx < thread_count) { | |
2305 | break; | |
73a8912b LL |
2306 | } else { |
2307 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
56e93d26 JQ |
2308 | } |
2309 | } | |
73a8912b | 2310 | qemu_mutex_unlock(&decomp_done_lock); |
56e93d26 JQ |
2311 | } |
2312 | ||
f265e0e4 JQ |
2313 | /** |
2314 | * ram_load_setup: Setup RAM for migration incoming side | |
2315 | * | |
2316 | * Returns zero to indicate success and negative for error | |
2317 | * | |
2318 | * @f: QEMUFile where to receive the data | |
2319 | * @opaque: RAMState pointer | |
2320 | */ | |
2321 | static int ram_load_setup(QEMUFile *f, void *opaque) | |
2322 | { | |
2323 | xbzrle_load_setup(); | |
2324 | return 0; | |
2325 | } | |
2326 | ||
2327 | static int ram_load_cleanup(void *opaque) | |
2328 | { | |
2329 | xbzrle_load_cleanup(); | |
2330 | return 0; | |
2331 | } | |
2332 | ||
3d0684b2 JQ |
2333 | /** |
2334 | * ram_postcopy_incoming_init: allocate postcopy data structures | |
2335 | * | |
2336 | * Returns 0 for success and negative if there was one error | |
2337 | * | |
2338 | * @mis: current migration incoming state | |
2339 | * | |
2340 | * Allocate data structures etc needed by incoming migration with | |
2341 | * postcopy-ram. postcopy-ram's similarly names | |
2342 | * postcopy_ram_incoming_init does the work. | |
1caddf8a DDAG |
2343 | */ |
2344 | int ram_postcopy_incoming_init(MigrationIncomingState *mis) | |
2345 | { | |
b8c48993 | 2346 | unsigned long ram_pages = last_ram_page(); |
1caddf8a DDAG |
2347 | |
2348 | return postcopy_ram_incoming_init(mis, ram_pages); | |
2349 | } | |
2350 | ||
3d0684b2 JQ |
2351 | /** |
2352 | * ram_load_postcopy: load a page in postcopy case | |
2353 | * | |
2354 | * Returns 0 for success or -errno in case of error | |
2355 | * | |
a7180877 DDAG |
2356 | * Called in postcopy mode by ram_load(). |
2357 | * rcu_read_lock is taken prior to this being called. | |
3d0684b2 JQ |
2358 | * |
2359 | * @f: QEMUFile where to send the data | |
a7180877 DDAG |
2360 | */ |
2361 | static int ram_load_postcopy(QEMUFile *f) | |
2362 | { | |
2363 | int flags = 0, ret = 0; | |
2364 | bool place_needed = false; | |
28abd200 | 2365 | bool matching_page_sizes = false; |
a7180877 DDAG |
2366 | MigrationIncomingState *mis = migration_incoming_get_current(); |
2367 | /* Temporary page that is later 'placed' */ | |
2368 | void *postcopy_host_page = postcopy_get_tmp_page(mis); | |
c53b7ddc | 2369 | void *last_host = NULL; |
a3b6ff6d | 2370 | bool all_zero = false; |
a7180877 DDAG |
2371 | |
2372 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
2373 | ram_addr_t addr; | |
2374 | void *host = NULL; | |
2375 | void *page_buffer = NULL; | |
2376 | void *place_source = NULL; | |
df9ff5e1 | 2377 | RAMBlock *block = NULL; |
a7180877 | 2378 | uint8_t ch; |
a7180877 DDAG |
2379 | |
2380 | addr = qemu_get_be64(f); | |
2381 | flags = addr & ~TARGET_PAGE_MASK; | |
2382 | addr &= TARGET_PAGE_MASK; | |
2383 | ||
2384 | trace_ram_load_postcopy_loop((uint64_t)addr, flags); | |
2385 | place_needed = false; | |
bb890ed5 | 2386 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE)) { |
df9ff5e1 | 2387 | block = ram_block_from_stream(f, flags); |
4c4bad48 HZ |
2388 | |
2389 | host = host_from_ram_block_offset(block, addr); | |
a7180877 DDAG |
2390 | if (!host) { |
2391 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
2392 | ret = -EINVAL; | |
2393 | break; | |
2394 | } | |
28abd200 | 2395 | matching_page_sizes = block->page_size == TARGET_PAGE_SIZE; |
a7180877 | 2396 | /* |
28abd200 DDAG |
2397 | * Postcopy requires that we place whole host pages atomically; |
2398 | * these may be huge pages for RAMBlocks that are backed by | |
2399 | * hugetlbfs. | |
a7180877 DDAG |
2400 | * To make it atomic, the data is read into a temporary page |
2401 | * that's moved into place later. | |
2402 | * The migration protocol uses, possibly smaller, target-pages | |
2403 | * however the source ensures it always sends all the components | |
2404 | * of a host page in order. | |
2405 | */ | |
2406 | page_buffer = postcopy_host_page + | |
28abd200 | 2407 | ((uintptr_t)host & (block->page_size - 1)); |
a7180877 | 2408 | /* If all TP are zero then we can optimise the place */ |
28abd200 | 2409 | if (!((uintptr_t)host & (block->page_size - 1))) { |
a7180877 | 2410 | all_zero = true; |
c53b7ddc DDAG |
2411 | } else { |
2412 | /* not the 1st TP within the HP */ | |
2413 | if (host != (last_host + TARGET_PAGE_SIZE)) { | |
9af9e0fe | 2414 | error_report("Non-sequential target page %p/%p", |
c53b7ddc DDAG |
2415 | host, last_host); |
2416 | ret = -EINVAL; | |
2417 | break; | |
2418 | } | |
a7180877 DDAG |
2419 | } |
2420 | ||
c53b7ddc | 2421 | |
a7180877 DDAG |
2422 | /* |
2423 | * If it's the last part of a host page then we place the host | |
2424 | * page | |
2425 | */ | |
2426 | place_needed = (((uintptr_t)host + TARGET_PAGE_SIZE) & | |
28abd200 | 2427 | (block->page_size - 1)) == 0; |
a7180877 DDAG |
2428 | place_source = postcopy_host_page; |
2429 | } | |
c53b7ddc | 2430 | last_host = host; |
a7180877 DDAG |
2431 | |
2432 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { | |
bb890ed5 | 2433 | case RAM_SAVE_FLAG_ZERO: |
a7180877 DDAG |
2434 | ch = qemu_get_byte(f); |
2435 | memset(page_buffer, ch, TARGET_PAGE_SIZE); | |
2436 | if (ch) { | |
2437 | all_zero = false; | |
2438 | } | |
2439 | break; | |
2440 | ||
2441 | case RAM_SAVE_FLAG_PAGE: | |
2442 | all_zero = false; | |
2443 | if (!place_needed || !matching_page_sizes) { | |
2444 | qemu_get_buffer(f, page_buffer, TARGET_PAGE_SIZE); | |
2445 | } else { | |
2446 | /* Avoids the qemu_file copy during postcopy, which is | |
2447 | * going to do a copy later; can only do it when we | |
2448 | * do this read in one go (matching page sizes) | |
2449 | */ | |
2450 | qemu_get_buffer_in_place(f, (uint8_t **)&place_source, | |
2451 | TARGET_PAGE_SIZE); | |
2452 | } | |
2453 | break; | |
2454 | case RAM_SAVE_FLAG_EOS: | |
2455 | /* normal exit */ | |
2456 | break; | |
2457 | default: | |
2458 | error_report("Unknown combination of migration flags: %#x" | |
2459 | " (postcopy mode)", flags); | |
2460 | ret = -EINVAL; | |
2461 | } | |
2462 | ||
2463 | if (place_needed) { | |
2464 | /* This gets called at the last target page in the host page */ | |
df9ff5e1 DDAG |
2465 | void *place_dest = host + TARGET_PAGE_SIZE - block->page_size; |
2466 | ||
a7180877 | 2467 | if (all_zero) { |
df9ff5e1 DDAG |
2468 | ret = postcopy_place_page_zero(mis, place_dest, |
2469 | block->page_size); | |
a7180877 | 2470 | } else { |
df9ff5e1 DDAG |
2471 | ret = postcopy_place_page(mis, place_dest, |
2472 | place_source, block->page_size); | |
a7180877 DDAG |
2473 | } |
2474 | } | |
2475 | if (!ret) { | |
2476 | ret = qemu_file_get_error(f); | |
2477 | } | |
2478 | } | |
2479 | ||
2480 | return ret; | |
2481 | } | |
2482 | ||
56e93d26 JQ |
2483 | static int ram_load(QEMUFile *f, void *opaque, int version_id) |
2484 | { | |
edc60127 | 2485 | int flags = 0, ret = 0, invalid_flags = 0; |
56e93d26 JQ |
2486 | static uint64_t seq_iter; |
2487 | int len = 0; | |
a7180877 DDAG |
2488 | /* |
2489 | * If system is running in postcopy mode, page inserts to host memory must | |
2490 | * be atomic | |
2491 | */ | |
2492 | bool postcopy_running = postcopy_state_get() >= POSTCOPY_INCOMING_LISTENING; | |
ef08fb38 DDAG |
2493 | /* ADVISE is earlier, it shows the source has the postcopy capability on */ |
2494 | bool postcopy_advised = postcopy_state_get() >= POSTCOPY_INCOMING_ADVISE; | |
56e93d26 JQ |
2495 | |
2496 | seq_iter++; | |
2497 | ||
2498 | if (version_id != 4) { | |
2499 | ret = -EINVAL; | |
2500 | } | |
2501 | ||
edc60127 JQ |
2502 | if (!migrate_use_compression()) { |
2503 | invalid_flags |= RAM_SAVE_FLAG_COMPRESS_PAGE; | |
2504 | } | |
56e93d26 JQ |
2505 | /* This RCU critical section can be very long running. |
2506 | * When RCU reclaims in the code start to become numerous, | |
2507 | * it will be necessary to reduce the granularity of this | |
2508 | * critical section. | |
2509 | */ | |
2510 | rcu_read_lock(); | |
a7180877 DDAG |
2511 | |
2512 | if (postcopy_running) { | |
2513 | ret = ram_load_postcopy(f); | |
2514 | } | |
2515 | ||
2516 | while (!postcopy_running && !ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
56e93d26 | 2517 | ram_addr_t addr, total_ram_bytes; |
a776aa15 | 2518 | void *host = NULL; |
56e93d26 JQ |
2519 | uint8_t ch; |
2520 | ||
2521 | addr = qemu_get_be64(f); | |
2522 | flags = addr & ~TARGET_PAGE_MASK; | |
2523 | addr &= TARGET_PAGE_MASK; | |
2524 | ||
edc60127 JQ |
2525 | if (flags & invalid_flags) { |
2526 | if (flags & invalid_flags & RAM_SAVE_FLAG_COMPRESS_PAGE) { | |
2527 | error_report("Received an unexpected compressed page"); | |
2528 | } | |
2529 | ||
2530 | ret = -EINVAL; | |
2531 | break; | |
2532 | } | |
2533 | ||
bb890ed5 | 2534 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
a776aa15 | 2535 | RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) { |
4c4bad48 HZ |
2536 | RAMBlock *block = ram_block_from_stream(f, flags); |
2537 | ||
2538 | host = host_from_ram_block_offset(block, addr); | |
a776aa15 DDAG |
2539 | if (!host) { |
2540 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
2541 | ret = -EINVAL; | |
2542 | break; | |
2543 | } | |
1db9d8e5 | 2544 | trace_ram_load_loop(block->idstr, (uint64_t)addr, flags, host); |
a776aa15 DDAG |
2545 | } |
2546 | ||
56e93d26 JQ |
2547 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { |
2548 | case RAM_SAVE_FLAG_MEM_SIZE: | |
2549 | /* Synchronize RAM block list */ | |
2550 | total_ram_bytes = addr; | |
2551 | while (!ret && total_ram_bytes) { | |
2552 | RAMBlock *block; | |
56e93d26 JQ |
2553 | char id[256]; |
2554 | ram_addr_t length; | |
2555 | ||
2556 | len = qemu_get_byte(f); | |
2557 | qemu_get_buffer(f, (uint8_t *)id, len); | |
2558 | id[len] = 0; | |
2559 | length = qemu_get_be64(f); | |
2560 | ||
e3dd7493 DDAG |
2561 | block = qemu_ram_block_by_name(id); |
2562 | if (block) { | |
2563 | if (length != block->used_length) { | |
2564 | Error *local_err = NULL; | |
56e93d26 | 2565 | |
fa53a0e5 | 2566 | ret = qemu_ram_resize(block, length, |
e3dd7493 DDAG |
2567 | &local_err); |
2568 | if (local_err) { | |
2569 | error_report_err(local_err); | |
56e93d26 | 2570 | } |
56e93d26 | 2571 | } |
ef08fb38 DDAG |
2572 | /* For postcopy we need to check hugepage sizes match */ |
2573 | if (postcopy_advised && | |
2574 | block->page_size != qemu_host_page_size) { | |
2575 | uint64_t remote_page_size = qemu_get_be64(f); | |
2576 | if (remote_page_size != block->page_size) { | |
2577 | error_report("Mismatched RAM page size %s " | |
2578 | "(local) %zd != %" PRId64, | |
2579 | id, block->page_size, | |
2580 | remote_page_size); | |
2581 | ret = -EINVAL; | |
2582 | } | |
2583 | } | |
e3dd7493 DDAG |
2584 | ram_control_load_hook(f, RAM_CONTROL_BLOCK_REG, |
2585 | block->idstr); | |
2586 | } else { | |
56e93d26 JQ |
2587 | error_report("Unknown ramblock \"%s\", cannot " |
2588 | "accept migration", id); | |
2589 | ret = -EINVAL; | |
2590 | } | |
2591 | ||
2592 | total_ram_bytes -= length; | |
2593 | } | |
2594 | break; | |
a776aa15 | 2595 | |
bb890ed5 | 2596 | case RAM_SAVE_FLAG_ZERO: |
56e93d26 JQ |
2597 | ch = qemu_get_byte(f); |
2598 | ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); | |
2599 | break; | |
a776aa15 | 2600 | |
56e93d26 | 2601 | case RAM_SAVE_FLAG_PAGE: |
56e93d26 JQ |
2602 | qemu_get_buffer(f, host, TARGET_PAGE_SIZE); |
2603 | break; | |
56e93d26 | 2604 | |
a776aa15 | 2605 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
56e93d26 JQ |
2606 | len = qemu_get_be32(f); |
2607 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
2608 | error_report("Invalid compressed data length: %d", len); | |
2609 | ret = -EINVAL; | |
2610 | break; | |
2611 | } | |
c1bc6626 | 2612 | decompress_data_with_multi_threads(f, host, len); |
56e93d26 | 2613 | break; |
a776aa15 | 2614 | |
56e93d26 | 2615 | case RAM_SAVE_FLAG_XBZRLE: |
56e93d26 JQ |
2616 | if (load_xbzrle(f, addr, host) < 0) { |
2617 | error_report("Failed to decompress XBZRLE page at " | |
2618 | RAM_ADDR_FMT, addr); | |
2619 | ret = -EINVAL; | |
2620 | break; | |
2621 | } | |
2622 | break; | |
2623 | case RAM_SAVE_FLAG_EOS: | |
2624 | /* normal exit */ | |
2625 | break; | |
2626 | default: | |
2627 | if (flags & RAM_SAVE_FLAG_HOOK) { | |
632e3a5c | 2628 | ram_control_load_hook(f, RAM_CONTROL_HOOK, NULL); |
56e93d26 JQ |
2629 | } else { |
2630 | error_report("Unknown combination of migration flags: %#x", | |
2631 | flags); | |
2632 | ret = -EINVAL; | |
2633 | } | |
2634 | } | |
2635 | if (!ret) { | |
2636 | ret = qemu_file_get_error(f); | |
2637 | } | |
2638 | } | |
2639 | ||
5533b2e9 | 2640 | wait_for_decompress_done(); |
56e93d26 | 2641 | rcu_read_unlock(); |
55c4446b | 2642 | trace_ram_load_complete(ret, seq_iter); |
56e93d26 JQ |
2643 | return ret; |
2644 | } | |
2645 | ||
2646 | static SaveVMHandlers savevm_ram_handlers = { | |
9907e842 | 2647 | .save_setup = ram_save_setup, |
56e93d26 | 2648 | .save_live_iterate = ram_save_iterate, |
763c906b | 2649 | .save_live_complete_postcopy = ram_save_complete, |
a3e06c3d | 2650 | .save_live_complete_precopy = ram_save_complete, |
56e93d26 JQ |
2651 | .save_live_pending = ram_save_pending, |
2652 | .load_state = ram_load, | |
f265e0e4 JQ |
2653 | .save_cleanup = ram_save_cleanup, |
2654 | .load_setup = ram_load_setup, | |
2655 | .load_cleanup = ram_load_cleanup, | |
56e93d26 JQ |
2656 | }; |
2657 | ||
2658 | void ram_mig_init(void) | |
2659 | { | |
2660 | qemu_mutex_init(&XBZRLE.lock); | |
6f37bb8b | 2661 | register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, &ram_state); |
56e93d26 | 2662 | } |