]>
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 PB |
29 | #include "qemu-common.h" |
30 | #include "cpu.h" | |
56e93d26 | 31 | #include <zlib.h> |
4addcd4f | 32 | #include "qapi-event.h" |
f348b6d1 | 33 | #include "qemu/cutils.h" |
56e93d26 JQ |
34 | #include "qemu/bitops.h" |
35 | #include "qemu/bitmap.h" | |
7205c9ec JQ |
36 | #include "qemu/timer.h" |
37 | #include "qemu/main-loop.h" | |
709e3fe8 | 38 | #include "xbzrle.h" |
7b1e1a22 | 39 | #include "ram.h" |
56e93d26 | 40 | #include "migration/migration.h" |
7b1e1a22 | 41 | #include "migration/misc.h" |
08a0aee1 | 42 | #include "qemu-file.h" |
987772d9 | 43 | #include "migration/vmstate.h" |
be07b0ac | 44 | #include "postcopy-ram.h" |
56e93d26 JQ |
45 | #include "exec/address-spaces.h" |
46 | #include "migration/page_cache.h" | |
56e93d26 | 47 | #include "qemu/error-report.h" |
56e93d26 | 48 | #include "trace.h" |
56e93d26 | 49 | #include "exec/ram_addr.h" |
56e93d26 | 50 | #include "qemu/rcu_queue.h" |
a91246c9 | 51 | #include "migration/colo.h" |
56e93d26 | 52 | |
56e93d26 JQ |
53 | /***********************************************************/ |
54 | /* ram save/restore */ | |
55 | ||
bb890ed5 JQ |
56 | /* RAM_SAVE_FLAG_ZERO used to be named RAM_SAVE_FLAG_COMPRESS, it |
57 | * worked for pages that where filled with the same char. We switched | |
58 | * it to only search for the zero value. And to avoid confusion with | |
59 | * RAM_SSAVE_FLAG_COMPRESS_PAGE just rename it. | |
60 | */ | |
61 | ||
56e93d26 | 62 | #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */ |
bb890ed5 | 63 | #define RAM_SAVE_FLAG_ZERO 0x02 |
56e93d26 JQ |
64 | #define RAM_SAVE_FLAG_MEM_SIZE 0x04 |
65 | #define RAM_SAVE_FLAG_PAGE 0x08 | |
66 | #define RAM_SAVE_FLAG_EOS 0x10 | |
67 | #define RAM_SAVE_FLAG_CONTINUE 0x20 | |
68 | #define RAM_SAVE_FLAG_XBZRLE 0x40 | |
69 | /* 0x80 is reserved in migration.h start with 0x100 next */ | |
70 | #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 | |
71 | ||
56e93d26 JQ |
72 | static inline bool is_zero_range(uint8_t *p, uint64_t size) |
73 | { | |
a1febc49 | 74 | return buffer_is_zero(p, size); |
56e93d26 JQ |
75 | } |
76 | ||
9360447d JQ |
77 | XBZRLECacheStats xbzrle_counters; |
78 | ||
56e93d26 JQ |
79 | /* struct contains XBZRLE cache and a static page |
80 | used by the compression */ | |
81 | static struct { | |
82 | /* buffer used for XBZRLE encoding */ | |
83 | uint8_t *encoded_buf; | |
84 | /* buffer for storing page content */ | |
85 | uint8_t *current_buf; | |
86 | /* Cache for XBZRLE, Protected by lock. */ | |
87 | PageCache *cache; | |
88 | QemuMutex lock; | |
c00e0928 JQ |
89 | /* it will store a page full of zeros */ |
90 | uint8_t *zero_target_page; | |
56e93d26 JQ |
91 | } XBZRLE; |
92 | ||
93 | /* buffer used for XBZRLE decoding */ | |
94 | static uint8_t *xbzrle_decoded_buf; | |
95 | ||
96 | static void XBZRLE_cache_lock(void) | |
97 | { | |
98 | if (migrate_use_xbzrle()) | |
99 | qemu_mutex_lock(&XBZRLE.lock); | |
100 | } | |
101 | ||
102 | static void XBZRLE_cache_unlock(void) | |
103 | { | |
104 | if (migrate_use_xbzrle()) | |
105 | qemu_mutex_unlock(&XBZRLE.lock); | |
106 | } | |
107 | ||
3d0684b2 JQ |
108 | /** |
109 | * xbzrle_cache_resize: resize the xbzrle cache | |
110 | * | |
111 | * This function is called from qmp_migrate_set_cache_size in main | |
112 | * thread, possibly while a migration is in progress. A running | |
113 | * migration may be using the cache and might finish during this call, | |
114 | * hence changes to the cache are protected by XBZRLE.lock(). | |
115 | * | |
116 | * Returns the new_size or negative in case of error. | |
117 | * | |
118 | * @new_size: new cache size | |
56e93d26 JQ |
119 | */ |
120 | int64_t xbzrle_cache_resize(int64_t new_size) | |
121 | { | |
122 | PageCache *new_cache; | |
123 | int64_t ret; | |
124 | ||
125 | if (new_size < TARGET_PAGE_SIZE) { | |
126 | return -1; | |
127 | } | |
128 | ||
129 | XBZRLE_cache_lock(); | |
130 | ||
131 | if (XBZRLE.cache != NULL) { | |
132 | if (pow2floor(new_size) == migrate_xbzrle_cache_size()) { | |
133 | goto out_new_size; | |
134 | } | |
135 | new_cache = cache_init(new_size / TARGET_PAGE_SIZE, | |
136 | TARGET_PAGE_SIZE); | |
137 | if (!new_cache) { | |
138 | error_report("Error creating cache"); | |
139 | ret = -1; | |
140 | goto out; | |
141 | } | |
142 | ||
143 | cache_fini(XBZRLE.cache); | |
144 | XBZRLE.cache = new_cache; | |
145 | } | |
146 | ||
147 | out_new_size: | |
148 | ret = pow2floor(new_size); | |
149 | out: | |
150 | XBZRLE_cache_unlock(); | |
151 | return ret; | |
152 | } | |
153 | ||
ec481c6c JQ |
154 | /* |
155 | * An outstanding page request, on the source, having been received | |
156 | * and queued | |
157 | */ | |
158 | struct RAMSrcPageRequest { | |
159 | RAMBlock *rb; | |
160 | hwaddr offset; | |
161 | hwaddr len; | |
162 | ||
163 | QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; | |
164 | }; | |
165 | ||
6f37bb8b JQ |
166 | /* State of RAM for migration */ |
167 | struct RAMState { | |
204b88b8 JQ |
168 | /* QEMUFile used for this migration */ |
169 | QEMUFile *f; | |
6f37bb8b JQ |
170 | /* Last block that we have visited searching for dirty pages */ |
171 | RAMBlock *last_seen_block; | |
172 | /* Last block from where we have sent data */ | |
173 | RAMBlock *last_sent_block; | |
269ace29 JQ |
174 | /* Last dirty target page we have sent */ |
175 | ram_addr_t last_page; | |
6f37bb8b JQ |
176 | /* last ram version we have seen */ |
177 | uint32_t last_version; | |
178 | /* We are in the first round */ | |
179 | bool ram_bulk_stage; | |
8d820d6f JQ |
180 | /* How many times we have dirty too many pages */ |
181 | int dirty_rate_high_cnt; | |
f664da80 JQ |
182 | /* these variables are used for bitmap sync */ |
183 | /* last time we did a full bitmap_sync */ | |
184 | int64_t time_last_bitmap_sync; | |
eac74159 | 185 | /* bytes transferred at start_time */ |
c4bdf0cf | 186 | uint64_t bytes_xfer_prev; |
a66cd90c | 187 | /* number of dirty pages since start_time */ |
68908ed6 | 188 | uint64_t num_dirty_pages_period; |
b5833fde JQ |
189 | /* xbzrle misses since the beginning of the period */ |
190 | uint64_t xbzrle_cache_miss_prev; | |
36040d9c JQ |
191 | /* number of iterations at the beginning of period */ |
192 | uint64_t iterations_prev; | |
23b28c3c JQ |
193 | /* Iterations since start */ |
194 | uint64_t iterations; | |
108cfae0 | 195 | /* protects modification of the bitmap */ |
9360447d JQ |
196 | uint64_t migration_dirty_pages; |
197 | /* number of dirty bits in the bitmap */ | |
108cfae0 | 198 | QemuMutex bitmap_mutex; |
68a098f3 JQ |
199 | /* The RAMBlock used in the last src_page_requests */ |
200 | RAMBlock *last_req_rb; | |
ec481c6c JQ |
201 | /* Queue of outstanding page requests from the destination */ |
202 | QemuMutex src_page_req_mutex; | |
203 | QSIMPLEQ_HEAD(src_page_requests, RAMSrcPageRequest) src_page_requests; | |
6f37bb8b JQ |
204 | }; |
205 | typedef struct RAMState RAMState; | |
206 | ||
53518d94 | 207 | static RAMState *ram_state; |
6f37bb8b | 208 | |
9edabd4d | 209 | uint64_t ram_bytes_remaining(void) |
2f4fde93 | 210 | { |
53518d94 | 211 | return ram_state->migration_dirty_pages * TARGET_PAGE_SIZE; |
2f4fde93 JQ |
212 | } |
213 | ||
9360447d | 214 | MigrationStats ram_counters; |
96506894 | 215 | |
b8fb8cb7 DDAG |
216 | /* used by the search for pages to send */ |
217 | struct PageSearchStatus { | |
218 | /* Current block being searched */ | |
219 | RAMBlock *block; | |
a935e30f JQ |
220 | /* Current page to search from */ |
221 | unsigned long page; | |
b8fb8cb7 DDAG |
222 | /* Set once we wrap around */ |
223 | bool complete_round; | |
224 | }; | |
225 | typedef struct PageSearchStatus PageSearchStatus; | |
226 | ||
56e93d26 | 227 | struct CompressParam { |
56e93d26 | 228 | bool done; |
90e56fb4 | 229 | bool quit; |
56e93d26 JQ |
230 | QEMUFile *file; |
231 | QemuMutex mutex; | |
232 | QemuCond cond; | |
233 | RAMBlock *block; | |
234 | ram_addr_t offset; | |
235 | }; | |
236 | typedef struct CompressParam CompressParam; | |
237 | ||
238 | struct DecompressParam { | |
73a8912b | 239 | bool done; |
90e56fb4 | 240 | bool quit; |
56e93d26 JQ |
241 | QemuMutex mutex; |
242 | QemuCond cond; | |
243 | void *des; | |
d341d9f3 | 244 | uint8_t *compbuf; |
56e93d26 JQ |
245 | int len; |
246 | }; | |
247 | typedef struct DecompressParam DecompressParam; | |
248 | ||
249 | static CompressParam *comp_param; | |
250 | static QemuThread *compress_threads; | |
251 | /* comp_done_cond is used to wake up the migration thread when | |
252 | * one of the compression threads has finished the compression. | |
253 | * comp_done_lock is used to co-work with comp_done_cond. | |
254 | */ | |
0d9f9a5c LL |
255 | static QemuMutex comp_done_lock; |
256 | static QemuCond comp_done_cond; | |
56e93d26 JQ |
257 | /* The empty QEMUFileOps will be used by file in CompressParam */ |
258 | static const QEMUFileOps empty_ops = { }; | |
259 | ||
56e93d26 JQ |
260 | static DecompressParam *decomp_param; |
261 | static QemuThread *decompress_threads; | |
73a8912b LL |
262 | static QemuMutex decomp_done_lock; |
263 | static QemuCond decomp_done_cond; | |
56e93d26 | 264 | |
a7a9a88f LL |
265 | static int do_compress_ram_page(QEMUFile *f, RAMBlock *block, |
266 | ram_addr_t offset); | |
56e93d26 JQ |
267 | |
268 | static void *do_data_compress(void *opaque) | |
269 | { | |
270 | CompressParam *param = opaque; | |
a7a9a88f LL |
271 | RAMBlock *block; |
272 | ram_addr_t offset; | |
56e93d26 | 273 | |
a7a9a88f | 274 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 275 | while (!param->quit) { |
a7a9a88f LL |
276 | if (param->block) { |
277 | block = param->block; | |
278 | offset = param->offset; | |
279 | param->block = NULL; | |
280 | qemu_mutex_unlock(¶m->mutex); | |
281 | ||
282 | do_compress_ram_page(param->file, block, offset); | |
283 | ||
0d9f9a5c | 284 | qemu_mutex_lock(&comp_done_lock); |
a7a9a88f | 285 | param->done = true; |
0d9f9a5c LL |
286 | qemu_cond_signal(&comp_done_cond); |
287 | qemu_mutex_unlock(&comp_done_lock); | |
a7a9a88f LL |
288 | |
289 | qemu_mutex_lock(¶m->mutex); | |
290 | } else { | |
56e93d26 JQ |
291 | qemu_cond_wait(¶m->cond, ¶m->mutex); |
292 | } | |
56e93d26 | 293 | } |
a7a9a88f | 294 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
295 | |
296 | return NULL; | |
297 | } | |
298 | ||
299 | static inline void terminate_compression_threads(void) | |
300 | { | |
301 | int idx, thread_count; | |
302 | ||
303 | thread_count = migrate_compress_threads(); | |
3d0684b2 | 304 | |
56e93d26 JQ |
305 | for (idx = 0; idx < thread_count; idx++) { |
306 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 307 | comp_param[idx].quit = true; |
56e93d26 JQ |
308 | qemu_cond_signal(&comp_param[idx].cond); |
309 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
310 | } | |
311 | } | |
312 | ||
313 | void migrate_compress_threads_join(void) | |
314 | { | |
315 | int i, thread_count; | |
316 | ||
317 | if (!migrate_use_compression()) { | |
318 | return; | |
319 | } | |
320 | terminate_compression_threads(); | |
321 | thread_count = migrate_compress_threads(); | |
322 | for (i = 0; i < thread_count; i++) { | |
323 | qemu_thread_join(compress_threads + i); | |
324 | qemu_fclose(comp_param[i].file); | |
325 | qemu_mutex_destroy(&comp_param[i].mutex); | |
326 | qemu_cond_destroy(&comp_param[i].cond); | |
327 | } | |
0d9f9a5c LL |
328 | qemu_mutex_destroy(&comp_done_lock); |
329 | qemu_cond_destroy(&comp_done_cond); | |
56e93d26 JQ |
330 | g_free(compress_threads); |
331 | g_free(comp_param); | |
56e93d26 JQ |
332 | compress_threads = NULL; |
333 | comp_param = NULL; | |
56e93d26 JQ |
334 | } |
335 | ||
336 | void migrate_compress_threads_create(void) | |
337 | { | |
338 | int i, thread_count; | |
339 | ||
340 | if (!migrate_use_compression()) { | |
341 | return; | |
342 | } | |
56e93d26 JQ |
343 | thread_count = migrate_compress_threads(); |
344 | compress_threads = g_new0(QemuThread, thread_count); | |
345 | comp_param = g_new0(CompressParam, thread_count); | |
0d9f9a5c LL |
346 | qemu_cond_init(&comp_done_cond); |
347 | qemu_mutex_init(&comp_done_lock); | |
56e93d26 | 348 | for (i = 0; i < thread_count; i++) { |
e110aa91 C |
349 | /* comp_param[i].file is just used as a dummy buffer to save data, |
350 | * set its ops to empty. | |
56e93d26 JQ |
351 | */ |
352 | comp_param[i].file = qemu_fopen_ops(NULL, &empty_ops); | |
353 | comp_param[i].done = true; | |
90e56fb4 | 354 | comp_param[i].quit = false; |
56e93d26 JQ |
355 | qemu_mutex_init(&comp_param[i].mutex); |
356 | qemu_cond_init(&comp_param[i].cond); | |
357 | qemu_thread_create(compress_threads + i, "compress", | |
358 | do_data_compress, comp_param + i, | |
359 | QEMU_THREAD_JOINABLE); | |
360 | } | |
361 | } | |
362 | ||
363 | /** | |
3d0684b2 | 364 | * save_page_header: write page header to wire |
56e93d26 JQ |
365 | * |
366 | * If this is the 1st block, it also writes the block identification | |
367 | * | |
3d0684b2 | 368 | * Returns the number of bytes written |
56e93d26 JQ |
369 | * |
370 | * @f: QEMUFile where to send the data | |
371 | * @block: block that contains the page we want to send | |
372 | * @offset: offset inside the block for the page | |
373 | * in the lower bits, it contains flags | |
374 | */ | |
2bf3aa85 JQ |
375 | static size_t save_page_header(RAMState *rs, QEMUFile *f, RAMBlock *block, |
376 | ram_addr_t offset) | |
56e93d26 | 377 | { |
9f5f380b | 378 | size_t size, len; |
56e93d26 | 379 | |
24795694 JQ |
380 | if (block == rs->last_sent_block) { |
381 | offset |= RAM_SAVE_FLAG_CONTINUE; | |
382 | } | |
2bf3aa85 | 383 | qemu_put_be64(f, offset); |
56e93d26 JQ |
384 | size = 8; |
385 | ||
386 | if (!(offset & RAM_SAVE_FLAG_CONTINUE)) { | |
9f5f380b | 387 | len = strlen(block->idstr); |
2bf3aa85 JQ |
388 | qemu_put_byte(f, len); |
389 | qemu_put_buffer(f, (uint8_t *)block->idstr, len); | |
9f5f380b | 390 | size += 1 + len; |
24795694 | 391 | rs->last_sent_block = block; |
56e93d26 JQ |
392 | } |
393 | return size; | |
394 | } | |
395 | ||
3d0684b2 JQ |
396 | /** |
397 | * mig_throttle_guest_down: throotle down the guest | |
398 | * | |
399 | * Reduce amount of guest cpu execution to hopefully slow down memory | |
400 | * writes. If guest dirty memory rate is reduced below the rate at | |
401 | * which we can transfer pages to the destination then we should be | |
402 | * able to complete migration. Some workloads dirty memory way too | |
403 | * fast and will not effectively converge, even with auto-converge. | |
070afca2 JH |
404 | */ |
405 | static void mig_throttle_guest_down(void) | |
406 | { | |
407 | MigrationState *s = migrate_get_current(); | |
2594f56d DB |
408 | uint64_t pct_initial = s->parameters.cpu_throttle_initial; |
409 | uint64_t pct_icrement = s->parameters.cpu_throttle_increment; | |
070afca2 JH |
410 | |
411 | /* We have not started throttling yet. Let's start it. */ | |
412 | if (!cpu_throttle_active()) { | |
413 | cpu_throttle_set(pct_initial); | |
414 | } else { | |
415 | /* Throttling already on, just increase the rate */ | |
416 | cpu_throttle_set(cpu_throttle_get_percentage() + pct_icrement); | |
417 | } | |
418 | } | |
419 | ||
3d0684b2 JQ |
420 | /** |
421 | * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache | |
422 | * | |
6f37bb8b | 423 | * @rs: current RAM state |
3d0684b2 JQ |
424 | * @current_addr: address for the zero page |
425 | * | |
426 | * Update the xbzrle cache to reflect a page that's been sent as all 0. | |
56e93d26 JQ |
427 | * The important thing is that a stale (not-yet-0'd) page be replaced |
428 | * by the new data. | |
429 | * As a bonus, if the page wasn't in the cache it gets added so that | |
3d0684b2 | 430 | * when a small write is made into the 0'd page it gets XBZRLE sent. |
56e93d26 | 431 | */ |
6f37bb8b | 432 | static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr) |
56e93d26 | 433 | { |
6f37bb8b | 434 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { |
56e93d26 JQ |
435 | return; |
436 | } | |
437 | ||
438 | /* We don't care if this fails to allocate a new cache page | |
439 | * as long as it updated an old one */ | |
c00e0928 | 440 | cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page, |
9360447d | 441 | ram_counters.dirty_sync_count); |
56e93d26 JQ |
442 | } |
443 | ||
444 | #define ENCODING_FLAG_XBZRLE 0x1 | |
445 | ||
446 | /** | |
447 | * save_xbzrle_page: compress and send current page | |
448 | * | |
449 | * Returns: 1 means that we wrote the page | |
450 | * 0 means that page is identical to the one already sent | |
451 | * -1 means that xbzrle would be longer than normal | |
452 | * | |
5a987738 | 453 | * @rs: current RAM state |
3d0684b2 JQ |
454 | * @current_data: pointer to the address of the page contents |
455 | * @current_addr: addr of the page | |
56e93d26 JQ |
456 | * @block: block that contains the page we want to send |
457 | * @offset: offset inside the block for the page | |
458 | * @last_stage: if we are at the completion stage | |
56e93d26 | 459 | */ |
204b88b8 | 460 | static int save_xbzrle_page(RAMState *rs, uint8_t **current_data, |
56e93d26 | 461 | ram_addr_t current_addr, RAMBlock *block, |
072c2511 | 462 | ram_addr_t offset, bool last_stage) |
56e93d26 JQ |
463 | { |
464 | int encoded_len = 0, bytes_xbzrle; | |
465 | uint8_t *prev_cached_page; | |
466 | ||
9360447d JQ |
467 | if (!cache_is_cached(XBZRLE.cache, current_addr, |
468 | ram_counters.dirty_sync_count)) { | |
469 | xbzrle_counters.cache_miss++; | |
56e93d26 JQ |
470 | if (!last_stage) { |
471 | if (cache_insert(XBZRLE.cache, current_addr, *current_data, | |
9360447d | 472 | ram_counters.dirty_sync_count) == -1) { |
56e93d26 JQ |
473 | return -1; |
474 | } else { | |
475 | /* update *current_data when the page has been | |
476 | inserted into cache */ | |
477 | *current_data = get_cached_data(XBZRLE.cache, current_addr); | |
478 | } | |
479 | } | |
480 | return -1; | |
481 | } | |
482 | ||
483 | prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); | |
484 | ||
485 | /* save current buffer into memory */ | |
486 | memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); | |
487 | ||
488 | /* XBZRLE encoding (if there is no overflow) */ | |
489 | encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, | |
490 | TARGET_PAGE_SIZE, XBZRLE.encoded_buf, | |
491 | TARGET_PAGE_SIZE); | |
492 | if (encoded_len == 0) { | |
55c4446b | 493 | trace_save_xbzrle_page_skipping(); |
56e93d26 JQ |
494 | return 0; |
495 | } else if (encoded_len == -1) { | |
55c4446b | 496 | trace_save_xbzrle_page_overflow(); |
9360447d | 497 | xbzrle_counters.overflow++; |
56e93d26 JQ |
498 | /* update data in the cache */ |
499 | if (!last_stage) { | |
500 | memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE); | |
501 | *current_data = prev_cached_page; | |
502 | } | |
503 | return -1; | |
504 | } | |
505 | ||
506 | /* we need to update the data in the cache, in order to get the same data */ | |
507 | if (!last_stage) { | |
508 | memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); | |
509 | } | |
510 | ||
511 | /* Send XBZRLE based compressed page */ | |
2bf3aa85 | 512 | bytes_xbzrle = save_page_header(rs, rs->f, block, |
204b88b8 JQ |
513 | offset | RAM_SAVE_FLAG_XBZRLE); |
514 | qemu_put_byte(rs->f, ENCODING_FLAG_XBZRLE); | |
515 | qemu_put_be16(rs->f, encoded_len); | |
516 | qemu_put_buffer(rs->f, XBZRLE.encoded_buf, encoded_len); | |
56e93d26 | 517 | bytes_xbzrle += encoded_len + 1 + 2; |
9360447d JQ |
518 | xbzrle_counters.pages++; |
519 | xbzrle_counters.bytes += bytes_xbzrle; | |
520 | ram_counters.transferred += bytes_xbzrle; | |
56e93d26 JQ |
521 | |
522 | return 1; | |
523 | } | |
524 | ||
3d0684b2 JQ |
525 | /** |
526 | * migration_bitmap_find_dirty: find the next dirty page from start | |
f3f491fc | 527 | * |
3d0684b2 JQ |
528 | * Called with rcu_read_lock() to protect migration_bitmap |
529 | * | |
530 | * Returns the byte offset within memory region of the start of a dirty page | |
531 | * | |
6f37bb8b | 532 | * @rs: current RAM state |
3d0684b2 | 533 | * @rb: RAMBlock where to search for dirty pages |
a935e30f | 534 | * @start: page where we start the search |
f3f491fc | 535 | */ |
56e93d26 | 536 | static inline |
a935e30f | 537 | unsigned long migration_bitmap_find_dirty(RAMState *rs, RAMBlock *rb, |
f20e2865 | 538 | unsigned long start) |
56e93d26 | 539 | { |
6b6712ef JQ |
540 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; |
541 | unsigned long *bitmap = rb->bmap; | |
56e93d26 JQ |
542 | unsigned long next; |
543 | ||
6b6712ef JQ |
544 | if (rs->ram_bulk_stage && start > 0) { |
545 | next = start + 1; | |
56e93d26 | 546 | } else { |
6b6712ef | 547 | next = find_next_bit(bitmap, size, start); |
56e93d26 JQ |
548 | } |
549 | ||
6b6712ef | 550 | return next; |
56e93d26 JQ |
551 | } |
552 | ||
06b10688 | 553 | static inline bool migration_bitmap_clear_dirty(RAMState *rs, |
f20e2865 JQ |
554 | RAMBlock *rb, |
555 | unsigned long page) | |
a82d593b DDAG |
556 | { |
557 | bool ret; | |
a82d593b | 558 | |
6b6712ef | 559 | ret = test_and_clear_bit(page, rb->bmap); |
a82d593b DDAG |
560 | |
561 | if (ret) { | |
0d8ec885 | 562 | rs->migration_dirty_pages--; |
a82d593b DDAG |
563 | } |
564 | return ret; | |
565 | } | |
566 | ||
15440dd5 JQ |
567 | static void migration_bitmap_sync_range(RAMState *rs, RAMBlock *rb, |
568 | ram_addr_t start, ram_addr_t length) | |
56e93d26 | 569 | { |
0d8ec885 | 570 | rs->migration_dirty_pages += |
6b6712ef | 571 | cpu_physical_memory_sync_dirty_bitmap(rb, start, length, |
0d8ec885 | 572 | &rs->num_dirty_pages_period); |
56e93d26 JQ |
573 | } |
574 | ||
3d0684b2 JQ |
575 | /** |
576 | * ram_pagesize_summary: calculate all the pagesizes of a VM | |
577 | * | |
578 | * Returns a summary bitmap of the page sizes of all RAMBlocks | |
579 | * | |
580 | * For VMs with just normal pages this is equivalent to the host page | |
581 | * size. If it's got some huge pages then it's the OR of all the | |
582 | * different page sizes. | |
e8ca1db2 DDAG |
583 | */ |
584 | uint64_t ram_pagesize_summary(void) | |
585 | { | |
586 | RAMBlock *block; | |
587 | uint64_t summary = 0; | |
588 | ||
99e15582 | 589 | RAMBLOCK_FOREACH(block) { |
e8ca1db2 DDAG |
590 | summary |= block->page_size; |
591 | } | |
592 | ||
593 | return summary; | |
594 | } | |
595 | ||
8d820d6f | 596 | static void migration_bitmap_sync(RAMState *rs) |
56e93d26 JQ |
597 | { |
598 | RAMBlock *block; | |
56e93d26 | 599 | int64_t end_time; |
c4bdf0cf | 600 | uint64_t bytes_xfer_now; |
56e93d26 | 601 | |
9360447d | 602 | ram_counters.dirty_sync_count++; |
56e93d26 | 603 | |
f664da80 JQ |
604 | if (!rs->time_last_bitmap_sync) { |
605 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56e93d26 JQ |
606 | } |
607 | ||
608 | trace_migration_bitmap_sync_start(); | |
9c1f8f44 | 609 | memory_global_dirty_log_sync(); |
56e93d26 | 610 | |
108cfae0 | 611 | qemu_mutex_lock(&rs->bitmap_mutex); |
56e93d26 | 612 | rcu_read_lock(); |
99e15582 | 613 | RAMBLOCK_FOREACH(block) { |
15440dd5 | 614 | migration_bitmap_sync_range(rs, block, 0, block->used_length); |
56e93d26 JQ |
615 | } |
616 | rcu_read_unlock(); | |
108cfae0 | 617 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 618 | |
a66cd90c | 619 | trace_migration_bitmap_sync_end(rs->num_dirty_pages_period); |
1ffb5dfd | 620 | |
56e93d26 JQ |
621 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
622 | ||
623 | /* more than 1 second = 1000 millisecons */ | |
f664da80 | 624 | if (end_time > rs->time_last_bitmap_sync + 1000) { |
d693c6f1 | 625 | /* calculate period counters */ |
9360447d | 626 | ram_counters.dirty_pages_rate = rs->num_dirty_pages_period * 1000 |
d693c6f1 | 627 | / (end_time - rs->time_last_bitmap_sync); |
9360447d | 628 | bytes_xfer_now = ram_counters.transferred; |
d693c6f1 | 629 | |
56e93d26 JQ |
630 | if (migrate_auto_converge()) { |
631 | /* The following detection logic can be refined later. For now: | |
632 | Check to see if the dirtied bytes is 50% more than the approx. | |
633 | amount of bytes that just got transferred since the last time we | |
070afca2 JH |
634 | were in this routine. If that happens twice, start or increase |
635 | throttling */ | |
070afca2 | 636 | |
d693c6f1 | 637 | if ((rs->num_dirty_pages_period * TARGET_PAGE_SIZE > |
eac74159 | 638 | (bytes_xfer_now - rs->bytes_xfer_prev) / 2) && |
b4a3c64b | 639 | (++rs->dirty_rate_high_cnt >= 2)) { |
56e93d26 | 640 | trace_migration_throttle(); |
8d820d6f | 641 | rs->dirty_rate_high_cnt = 0; |
070afca2 | 642 | mig_throttle_guest_down(); |
d693c6f1 | 643 | } |
56e93d26 | 644 | } |
070afca2 | 645 | |
56e93d26 | 646 | if (migrate_use_xbzrle()) { |
23b28c3c | 647 | if (rs->iterations_prev != rs->iterations) { |
9360447d JQ |
648 | xbzrle_counters.cache_miss_rate = |
649 | (double)(xbzrle_counters.cache_miss - | |
b5833fde | 650 | rs->xbzrle_cache_miss_prev) / |
23b28c3c | 651 | (rs->iterations - rs->iterations_prev); |
56e93d26 | 652 | } |
23b28c3c | 653 | rs->iterations_prev = rs->iterations; |
9360447d | 654 | rs->xbzrle_cache_miss_prev = xbzrle_counters.cache_miss; |
56e93d26 | 655 | } |
d693c6f1 FF |
656 | |
657 | /* reset period counters */ | |
f664da80 | 658 | rs->time_last_bitmap_sync = end_time; |
a66cd90c | 659 | rs->num_dirty_pages_period = 0; |
d2a4d85a | 660 | rs->bytes_xfer_prev = bytes_xfer_now; |
56e93d26 | 661 | } |
4addcd4f | 662 | if (migrate_use_events()) { |
9360447d | 663 | qapi_event_send_migration_pass(ram_counters.dirty_sync_count, NULL); |
4addcd4f | 664 | } |
56e93d26 JQ |
665 | } |
666 | ||
667 | /** | |
3d0684b2 | 668 | * save_zero_page: send the zero page to the stream |
56e93d26 | 669 | * |
3d0684b2 | 670 | * Returns the number of pages written. |
56e93d26 | 671 | * |
f7ccd61b | 672 | * @rs: current RAM state |
56e93d26 JQ |
673 | * @block: block that contains the page we want to send |
674 | * @offset: offset inside the block for the page | |
675 | * @p: pointer to the page | |
56e93d26 | 676 | */ |
ce25d337 JQ |
677 | static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, |
678 | uint8_t *p) | |
56e93d26 JQ |
679 | { |
680 | int pages = -1; | |
681 | ||
682 | if (is_zero_range(p, TARGET_PAGE_SIZE)) { | |
9360447d JQ |
683 | ram_counters.duplicate++; |
684 | ram_counters.transferred += | |
bb890ed5 | 685 | save_page_header(rs, rs->f, block, offset | RAM_SAVE_FLAG_ZERO); |
ce25d337 | 686 | qemu_put_byte(rs->f, 0); |
9360447d | 687 | ram_counters.transferred += 1; |
56e93d26 JQ |
688 | pages = 1; |
689 | } | |
690 | ||
691 | return pages; | |
692 | } | |
693 | ||
5727309d | 694 | static void ram_release_pages(const char *rbname, uint64_t offset, int pages) |
53f09a10 | 695 | { |
5727309d | 696 | if (!migrate_release_ram() || !migration_in_postcopy()) { |
53f09a10 PB |
697 | return; |
698 | } | |
699 | ||
aaa2064c | 700 | ram_discard_range(rbname, offset, pages << TARGET_PAGE_BITS); |
53f09a10 PB |
701 | } |
702 | ||
56e93d26 | 703 | /** |
3d0684b2 | 704 | * ram_save_page: send the given page to the stream |
56e93d26 | 705 | * |
3d0684b2 | 706 | * Returns the number of pages written. |
3fd3c4b3 DDAG |
707 | * < 0 - error |
708 | * >=0 - Number of pages written - this might legally be 0 | |
709 | * if xbzrle noticed the page was the same. | |
56e93d26 | 710 | * |
6f37bb8b | 711 | * @rs: current RAM state |
56e93d26 JQ |
712 | * @block: block that contains the page we want to send |
713 | * @offset: offset inside the block for the page | |
714 | * @last_stage: if we are at the completion stage | |
56e93d26 | 715 | */ |
a0a8aa14 | 716 | static int ram_save_page(RAMState *rs, PageSearchStatus *pss, bool last_stage) |
56e93d26 JQ |
717 | { |
718 | int pages = -1; | |
719 | uint64_t bytes_xmit; | |
720 | ram_addr_t current_addr; | |
56e93d26 JQ |
721 | uint8_t *p; |
722 | int ret; | |
723 | bool send_async = true; | |
a08f6890 | 724 | RAMBlock *block = pss->block; |
a935e30f | 725 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
56e93d26 | 726 | |
2f68e399 | 727 | p = block->host + offset; |
1db9d8e5 | 728 | trace_ram_save_page(block->idstr, (uint64_t)offset, p); |
56e93d26 JQ |
729 | |
730 | /* In doubt sent page as normal */ | |
731 | bytes_xmit = 0; | |
ce25d337 | 732 | ret = ram_control_save_page(rs->f, block->offset, |
56e93d26 JQ |
733 | offset, TARGET_PAGE_SIZE, &bytes_xmit); |
734 | if (bytes_xmit) { | |
9360447d | 735 | ram_counters.transferred += bytes_xmit; |
56e93d26 JQ |
736 | pages = 1; |
737 | } | |
738 | ||
739 | XBZRLE_cache_lock(); | |
740 | ||
741 | current_addr = block->offset + offset; | |
742 | ||
56e93d26 JQ |
743 | if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { |
744 | if (ret != RAM_SAVE_CONTROL_DELAYED) { | |
745 | if (bytes_xmit > 0) { | |
9360447d | 746 | ram_counters.normal++; |
56e93d26 | 747 | } else if (bytes_xmit == 0) { |
9360447d | 748 | ram_counters.duplicate++; |
56e93d26 JQ |
749 | } |
750 | } | |
751 | } else { | |
ce25d337 | 752 | pages = save_zero_page(rs, block, offset, p); |
56e93d26 JQ |
753 | if (pages > 0) { |
754 | /* Must let xbzrle know, otherwise a previous (now 0'd) cached | |
755 | * page would be stale | |
756 | */ | |
6f37bb8b | 757 | xbzrle_cache_zero_page(rs, current_addr); |
a935e30f | 758 | ram_release_pages(block->idstr, offset, pages); |
6f37bb8b | 759 | } else if (!rs->ram_bulk_stage && |
5727309d | 760 | !migration_in_postcopy() && migrate_use_xbzrle()) { |
204b88b8 | 761 | pages = save_xbzrle_page(rs, &p, current_addr, block, |
072c2511 | 762 | offset, last_stage); |
56e93d26 JQ |
763 | if (!last_stage) { |
764 | /* Can't send this cached data async, since the cache page | |
765 | * might get updated before it gets to the wire | |
766 | */ | |
767 | send_async = false; | |
768 | } | |
769 | } | |
770 | } | |
771 | ||
772 | /* XBZRLE overflow or normal page */ | |
773 | if (pages == -1) { | |
9360447d JQ |
774 | ram_counters.transferred += |
775 | save_page_header(rs, rs->f, block, offset | RAM_SAVE_FLAG_PAGE); | |
56e93d26 | 776 | if (send_async) { |
ce25d337 | 777 | qemu_put_buffer_async(rs->f, p, TARGET_PAGE_SIZE, |
53f09a10 | 778 | migrate_release_ram() & |
5727309d | 779 | migration_in_postcopy()); |
56e93d26 | 780 | } else { |
ce25d337 | 781 | qemu_put_buffer(rs->f, p, TARGET_PAGE_SIZE); |
56e93d26 | 782 | } |
9360447d | 783 | ram_counters.transferred += TARGET_PAGE_SIZE; |
56e93d26 | 784 | pages = 1; |
9360447d | 785 | ram_counters.normal++; |
56e93d26 JQ |
786 | } |
787 | ||
788 | XBZRLE_cache_unlock(); | |
789 | ||
790 | return pages; | |
791 | } | |
792 | ||
a7a9a88f LL |
793 | static int do_compress_ram_page(QEMUFile *f, RAMBlock *block, |
794 | ram_addr_t offset) | |
56e93d26 | 795 | { |
53518d94 | 796 | RAMState *rs = ram_state; |
56e93d26 | 797 | int bytes_sent, blen; |
a7a9a88f | 798 | uint8_t *p = block->host + (offset & TARGET_PAGE_MASK); |
56e93d26 | 799 | |
2bf3aa85 | 800 | bytes_sent = save_page_header(rs, f, block, offset | |
56e93d26 | 801 | RAM_SAVE_FLAG_COMPRESS_PAGE); |
a7a9a88f | 802 | blen = qemu_put_compression_data(f, p, TARGET_PAGE_SIZE, |
56e93d26 | 803 | migrate_compress_level()); |
b3be2896 LL |
804 | if (blen < 0) { |
805 | bytes_sent = 0; | |
806 | qemu_file_set_error(migrate_get_current()->to_dst_file, blen); | |
807 | error_report("compressed data failed!"); | |
808 | } else { | |
809 | bytes_sent += blen; | |
5727309d | 810 | ram_release_pages(block->idstr, offset & TARGET_PAGE_MASK, 1); |
b3be2896 | 811 | } |
56e93d26 JQ |
812 | |
813 | return bytes_sent; | |
814 | } | |
815 | ||
ce25d337 | 816 | static void flush_compressed_data(RAMState *rs) |
56e93d26 JQ |
817 | { |
818 | int idx, len, thread_count; | |
819 | ||
820 | if (!migrate_use_compression()) { | |
821 | return; | |
822 | } | |
823 | thread_count = migrate_compress_threads(); | |
a7a9a88f | 824 | |
0d9f9a5c | 825 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 | 826 | for (idx = 0; idx < thread_count; idx++) { |
a7a9a88f | 827 | while (!comp_param[idx].done) { |
0d9f9a5c | 828 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 | 829 | } |
a7a9a88f | 830 | } |
0d9f9a5c | 831 | qemu_mutex_unlock(&comp_done_lock); |
a7a9a88f LL |
832 | |
833 | for (idx = 0; idx < thread_count; idx++) { | |
834 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 835 | if (!comp_param[idx].quit) { |
ce25d337 | 836 | len = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
9360447d | 837 | ram_counters.transferred += len; |
56e93d26 | 838 | } |
a7a9a88f | 839 | qemu_mutex_unlock(&comp_param[idx].mutex); |
56e93d26 JQ |
840 | } |
841 | } | |
842 | ||
843 | static inline void set_compress_params(CompressParam *param, RAMBlock *block, | |
844 | ram_addr_t offset) | |
845 | { | |
846 | param->block = block; | |
847 | param->offset = offset; | |
848 | } | |
849 | ||
ce25d337 JQ |
850 | static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block, |
851 | ram_addr_t offset) | |
56e93d26 JQ |
852 | { |
853 | int idx, thread_count, bytes_xmit = -1, pages = -1; | |
854 | ||
855 | thread_count = migrate_compress_threads(); | |
0d9f9a5c | 856 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 JQ |
857 | while (true) { |
858 | for (idx = 0; idx < thread_count; idx++) { | |
859 | if (comp_param[idx].done) { | |
a7a9a88f | 860 | comp_param[idx].done = false; |
ce25d337 | 861 | bytes_xmit = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
a7a9a88f | 862 | qemu_mutex_lock(&comp_param[idx].mutex); |
56e93d26 | 863 | set_compress_params(&comp_param[idx], block, offset); |
a7a9a88f LL |
864 | qemu_cond_signal(&comp_param[idx].cond); |
865 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
56e93d26 | 866 | pages = 1; |
9360447d JQ |
867 | ram_counters.normal++; |
868 | ram_counters.transferred += bytes_xmit; | |
56e93d26 JQ |
869 | break; |
870 | } | |
871 | } | |
872 | if (pages > 0) { | |
873 | break; | |
874 | } else { | |
0d9f9a5c | 875 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 JQ |
876 | } |
877 | } | |
0d9f9a5c | 878 | qemu_mutex_unlock(&comp_done_lock); |
56e93d26 JQ |
879 | |
880 | return pages; | |
881 | } | |
882 | ||
883 | /** | |
884 | * ram_save_compressed_page: compress the given page and send it to the stream | |
885 | * | |
3d0684b2 | 886 | * Returns the number of pages written. |
56e93d26 | 887 | * |
6f37bb8b | 888 | * @rs: current RAM state |
56e93d26 JQ |
889 | * @block: block that contains the page we want to send |
890 | * @offset: offset inside the block for the page | |
891 | * @last_stage: if we are at the completion stage | |
56e93d26 | 892 | */ |
a0a8aa14 JQ |
893 | static int ram_save_compressed_page(RAMState *rs, PageSearchStatus *pss, |
894 | bool last_stage) | |
56e93d26 JQ |
895 | { |
896 | int pages = -1; | |
fc50438e | 897 | uint64_t bytes_xmit = 0; |
56e93d26 | 898 | uint8_t *p; |
fc50438e | 899 | int ret, blen; |
a08f6890 | 900 | RAMBlock *block = pss->block; |
a935e30f | 901 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
56e93d26 | 902 | |
2f68e399 | 903 | p = block->host + offset; |
56e93d26 | 904 | |
ce25d337 | 905 | ret = ram_control_save_page(rs->f, block->offset, |
56e93d26 JQ |
906 | offset, TARGET_PAGE_SIZE, &bytes_xmit); |
907 | if (bytes_xmit) { | |
9360447d | 908 | ram_counters.transferred += bytes_xmit; |
56e93d26 JQ |
909 | pages = 1; |
910 | } | |
56e93d26 JQ |
911 | if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { |
912 | if (ret != RAM_SAVE_CONTROL_DELAYED) { | |
913 | if (bytes_xmit > 0) { | |
9360447d | 914 | ram_counters.normal++; |
56e93d26 | 915 | } else if (bytes_xmit == 0) { |
9360447d | 916 | ram_counters.duplicate++; |
56e93d26 JQ |
917 | } |
918 | } | |
919 | } else { | |
920 | /* When starting the process of a new block, the first page of | |
921 | * the block should be sent out before other pages in the same | |
922 | * block, and all the pages in last block should have been sent | |
923 | * out, keeping this order is important, because the 'cont' flag | |
924 | * is used to avoid resending the block name. | |
925 | */ | |
6f37bb8b | 926 | if (block != rs->last_sent_block) { |
ce25d337 JQ |
927 | flush_compressed_data(rs); |
928 | pages = save_zero_page(rs, block, offset, p); | |
56e93d26 | 929 | if (pages == -1) { |
fc50438e | 930 | /* Make sure the first page is sent out before other pages */ |
2bf3aa85 | 931 | bytes_xmit = save_page_header(rs, rs->f, block, offset | |
fc50438e | 932 | RAM_SAVE_FLAG_COMPRESS_PAGE); |
ce25d337 | 933 | blen = qemu_put_compression_data(rs->f, p, TARGET_PAGE_SIZE, |
fc50438e LL |
934 | migrate_compress_level()); |
935 | if (blen > 0) { | |
9360447d JQ |
936 | ram_counters.transferred += bytes_xmit + blen; |
937 | ram_counters.normal++; | |
b3be2896 | 938 | pages = 1; |
fc50438e | 939 | } else { |
ce25d337 | 940 | qemu_file_set_error(rs->f, blen); |
fc50438e | 941 | error_report("compressed data failed!"); |
b3be2896 | 942 | } |
56e93d26 | 943 | } |
53f09a10 | 944 | if (pages > 0) { |
a935e30f | 945 | ram_release_pages(block->idstr, offset, pages); |
53f09a10 | 946 | } |
56e93d26 | 947 | } else { |
ce25d337 | 948 | pages = save_zero_page(rs, block, offset, p); |
56e93d26 | 949 | if (pages == -1) { |
ce25d337 | 950 | pages = compress_page_with_multi_thread(rs, block, offset); |
53f09a10 | 951 | } else { |
a935e30f | 952 | ram_release_pages(block->idstr, offset, pages); |
56e93d26 JQ |
953 | } |
954 | } | |
955 | } | |
956 | ||
957 | return pages; | |
958 | } | |
959 | ||
3d0684b2 JQ |
960 | /** |
961 | * find_dirty_block: find the next dirty page and update any state | |
962 | * associated with the search process. | |
b9e60928 | 963 | * |
3d0684b2 | 964 | * Returns if a page is found |
b9e60928 | 965 | * |
6f37bb8b | 966 | * @rs: current RAM state |
3d0684b2 JQ |
967 | * @pss: data about the state of the current dirty page scan |
968 | * @again: set to false if the search has scanned the whole of RAM | |
b9e60928 | 969 | */ |
f20e2865 | 970 | static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again) |
b9e60928 | 971 | { |
f20e2865 | 972 | pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page); |
6f37bb8b | 973 | if (pss->complete_round && pss->block == rs->last_seen_block && |
a935e30f | 974 | pss->page >= rs->last_page) { |
b9e60928 DDAG |
975 | /* |
976 | * We've been once around the RAM and haven't found anything. | |
977 | * Give up. | |
978 | */ | |
979 | *again = false; | |
980 | return false; | |
981 | } | |
a935e30f | 982 | if ((pss->page << TARGET_PAGE_BITS) >= pss->block->used_length) { |
b9e60928 | 983 | /* Didn't find anything in this RAM Block */ |
a935e30f | 984 | pss->page = 0; |
b9e60928 DDAG |
985 | pss->block = QLIST_NEXT_RCU(pss->block, next); |
986 | if (!pss->block) { | |
987 | /* Hit the end of the list */ | |
988 | pss->block = QLIST_FIRST_RCU(&ram_list.blocks); | |
989 | /* Flag that we've looped */ | |
990 | pss->complete_round = true; | |
6f37bb8b | 991 | rs->ram_bulk_stage = false; |
b9e60928 DDAG |
992 | if (migrate_use_xbzrle()) { |
993 | /* If xbzrle is on, stop using the data compression at this | |
994 | * point. In theory, xbzrle can do better than compression. | |
995 | */ | |
ce25d337 | 996 | flush_compressed_data(rs); |
b9e60928 DDAG |
997 | } |
998 | } | |
999 | /* Didn't find anything this time, but try again on the new block */ | |
1000 | *again = true; | |
1001 | return false; | |
1002 | } else { | |
1003 | /* Can go around again, but... */ | |
1004 | *again = true; | |
1005 | /* We've found something so probably don't need to */ | |
1006 | return true; | |
1007 | } | |
1008 | } | |
1009 | ||
3d0684b2 JQ |
1010 | /** |
1011 | * unqueue_page: gets a page of the queue | |
1012 | * | |
a82d593b | 1013 | * Helper for 'get_queued_page' - gets a page off the queue |
a82d593b | 1014 | * |
3d0684b2 JQ |
1015 | * Returns the block of the page (or NULL if none available) |
1016 | * | |
ec481c6c | 1017 | * @rs: current RAM state |
3d0684b2 | 1018 | * @offset: used to return the offset within the RAMBlock |
a82d593b | 1019 | */ |
f20e2865 | 1020 | static RAMBlock *unqueue_page(RAMState *rs, ram_addr_t *offset) |
a82d593b DDAG |
1021 | { |
1022 | RAMBlock *block = NULL; | |
1023 | ||
ec481c6c JQ |
1024 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1025 | if (!QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
1026 | struct RAMSrcPageRequest *entry = | |
1027 | QSIMPLEQ_FIRST(&rs->src_page_requests); | |
a82d593b DDAG |
1028 | block = entry->rb; |
1029 | *offset = entry->offset; | |
a82d593b DDAG |
1030 | |
1031 | if (entry->len > TARGET_PAGE_SIZE) { | |
1032 | entry->len -= TARGET_PAGE_SIZE; | |
1033 | entry->offset += TARGET_PAGE_SIZE; | |
1034 | } else { | |
1035 | memory_region_unref(block->mr); | |
ec481c6c | 1036 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
a82d593b DDAG |
1037 | g_free(entry); |
1038 | } | |
1039 | } | |
ec481c6c | 1040 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
a82d593b DDAG |
1041 | |
1042 | return block; | |
1043 | } | |
1044 | ||
3d0684b2 JQ |
1045 | /** |
1046 | * get_queued_page: unqueue a page from the postocpy requests | |
1047 | * | |
1048 | * Skips pages that are already sent (!dirty) | |
a82d593b | 1049 | * |
3d0684b2 | 1050 | * Returns if a queued page is found |
a82d593b | 1051 | * |
6f37bb8b | 1052 | * @rs: current RAM state |
3d0684b2 | 1053 | * @pss: data about the state of the current dirty page scan |
a82d593b | 1054 | */ |
f20e2865 | 1055 | static bool get_queued_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b DDAG |
1056 | { |
1057 | RAMBlock *block; | |
1058 | ram_addr_t offset; | |
1059 | bool dirty; | |
1060 | ||
1061 | do { | |
f20e2865 | 1062 | block = unqueue_page(rs, &offset); |
a82d593b DDAG |
1063 | /* |
1064 | * We're sending this page, and since it's postcopy nothing else | |
1065 | * will dirty it, and we must make sure it doesn't get sent again | |
1066 | * even if this queue request was received after the background | |
1067 | * search already sent it. | |
1068 | */ | |
1069 | if (block) { | |
f20e2865 JQ |
1070 | unsigned long page; |
1071 | ||
6b6712ef JQ |
1072 | page = offset >> TARGET_PAGE_BITS; |
1073 | dirty = test_bit(page, block->bmap); | |
a82d593b | 1074 | if (!dirty) { |
06b10688 | 1075 | trace_get_queued_page_not_dirty(block->idstr, (uint64_t)offset, |
6b6712ef | 1076 | page, test_bit(page, block->unsentmap)); |
a82d593b | 1077 | } else { |
f20e2865 | 1078 | trace_get_queued_page(block->idstr, (uint64_t)offset, page); |
a82d593b DDAG |
1079 | } |
1080 | } | |
1081 | ||
1082 | } while (block && !dirty); | |
1083 | ||
1084 | if (block) { | |
1085 | /* | |
1086 | * As soon as we start servicing pages out of order, then we have | |
1087 | * to kill the bulk stage, since the bulk stage assumes | |
1088 | * in (migration_bitmap_find_and_reset_dirty) that every page is | |
1089 | * dirty, that's no longer true. | |
1090 | */ | |
6f37bb8b | 1091 | rs->ram_bulk_stage = false; |
a82d593b DDAG |
1092 | |
1093 | /* | |
1094 | * We want the background search to continue from the queued page | |
1095 | * since the guest is likely to want other pages near to the page | |
1096 | * it just requested. | |
1097 | */ | |
1098 | pss->block = block; | |
a935e30f | 1099 | pss->page = offset >> TARGET_PAGE_BITS; |
a82d593b DDAG |
1100 | } |
1101 | ||
1102 | return !!block; | |
1103 | } | |
1104 | ||
6c595cde | 1105 | /** |
5e58f968 JQ |
1106 | * migration_page_queue_free: drop any remaining pages in the ram |
1107 | * request queue | |
6c595cde | 1108 | * |
3d0684b2 JQ |
1109 | * It should be empty at the end anyway, but in error cases there may |
1110 | * be some left. in case that there is any page left, we drop it. | |
1111 | * | |
6c595cde | 1112 | */ |
83c13382 | 1113 | static void migration_page_queue_free(RAMState *rs) |
6c595cde | 1114 | { |
ec481c6c | 1115 | struct RAMSrcPageRequest *mspr, *next_mspr; |
6c595cde DDAG |
1116 | /* This queue generally should be empty - but in the case of a failed |
1117 | * migration might have some droppings in. | |
1118 | */ | |
1119 | rcu_read_lock(); | |
ec481c6c | 1120 | QSIMPLEQ_FOREACH_SAFE(mspr, &rs->src_page_requests, next_req, next_mspr) { |
6c595cde | 1121 | memory_region_unref(mspr->rb->mr); |
ec481c6c | 1122 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
6c595cde DDAG |
1123 | g_free(mspr); |
1124 | } | |
1125 | rcu_read_unlock(); | |
1126 | } | |
1127 | ||
1128 | /** | |
3d0684b2 JQ |
1129 | * ram_save_queue_pages: queue the page for transmission |
1130 | * | |
1131 | * A request from postcopy destination for example. | |
1132 | * | |
1133 | * Returns zero on success or negative on error | |
1134 | * | |
3d0684b2 JQ |
1135 | * @rbname: Name of the RAMBLock of the request. NULL means the |
1136 | * same that last one. | |
1137 | * @start: starting address from the start of the RAMBlock | |
1138 | * @len: length (in bytes) to send | |
6c595cde | 1139 | */ |
96506894 | 1140 | int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len) |
6c595cde DDAG |
1141 | { |
1142 | RAMBlock *ramblock; | |
53518d94 | 1143 | RAMState *rs = ram_state; |
6c595cde | 1144 | |
9360447d | 1145 | ram_counters.postcopy_requests++; |
6c595cde DDAG |
1146 | rcu_read_lock(); |
1147 | if (!rbname) { | |
1148 | /* Reuse last RAMBlock */ | |
68a098f3 | 1149 | ramblock = rs->last_req_rb; |
6c595cde DDAG |
1150 | |
1151 | if (!ramblock) { | |
1152 | /* | |
1153 | * Shouldn't happen, we can't reuse the last RAMBlock if | |
1154 | * it's the 1st request. | |
1155 | */ | |
1156 | error_report("ram_save_queue_pages no previous block"); | |
1157 | goto err; | |
1158 | } | |
1159 | } else { | |
1160 | ramblock = qemu_ram_block_by_name(rbname); | |
1161 | ||
1162 | if (!ramblock) { | |
1163 | /* We shouldn't be asked for a non-existent RAMBlock */ | |
1164 | error_report("ram_save_queue_pages no block '%s'", rbname); | |
1165 | goto err; | |
1166 | } | |
68a098f3 | 1167 | rs->last_req_rb = ramblock; |
6c595cde DDAG |
1168 | } |
1169 | trace_ram_save_queue_pages(ramblock->idstr, start, len); | |
1170 | if (start+len > ramblock->used_length) { | |
9458ad6b JQ |
1171 | error_report("%s request overrun start=" RAM_ADDR_FMT " len=" |
1172 | RAM_ADDR_FMT " blocklen=" RAM_ADDR_FMT, | |
6c595cde DDAG |
1173 | __func__, start, len, ramblock->used_length); |
1174 | goto err; | |
1175 | } | |
1176 | ||
ec481c6c JQ |
1177 | struct RAMSrcPageRequest *new_entry = |
1178 | g_malloc0(sizeof(struct RAMSrcPageRequest)); | |
6c595cde DDAG |
1179 | new_entry->rb = ramblock; |
1180 | new_entry->offset = start; | |
1181 | new_entry->len = len; | |
1182 | ||
1183 | memory_region_ref(ramblock->mr); | |
ec481c6c JQ |
1184 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1185 | QSIMPLEQ_INSERT_TAIL(&rs->src_page_requests, new_entry, next_req); | |
1186 | qemu_mutex_unlock(&rs->src_page_req_mutex); | |
6c595cde DDAG |
1187 | rcu_read_unlock(); |
1188 | ||
1189 | return 0; | |
1190 | ||
1191 | err: | |
1192 | rcu_read_unlock(); | |
1193 | return -1; | |
1194 | } | |
1195 | ||
a82d593b | 1196 | /** |
3d0684b2 | 1197 | * ram_save_target_page: save one target page |
a82d593b | 1198 | * |
3d0684b2 | 1199 | * Returns the number of pages written |
a82d593b | 1200 | * |
6f37bb8b | 1201 | * @rs: current RAM state |
3d0684b2 | 1202 | * @ms: current migration state |
3d0684b2 | 1203 | * @pss: data about the page we want to send |
a82d593b | 1204 | * @last_stage: if we are at the completion stage |
a82d593b | 1205 | */ |
a0a8aa14 | 1206 | static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 1207 | bool last_stage) |
a82d593b DDAG |
1208 | { |
1209 | int res = 0; | |
1210 | ||
1211 | /* Check the pages is dirty and if it is send it */ | |
f20e2865 | 1212 | if (migration_bitmap_clear_dirty(rs, pss->block, pss->page)) { |
6d358d94 JQ |
1213 | /* |
1214 | * If xbzrle is on, stop using the data compression after first | |
1215 | * round of migration even if compression is enabled. In theory, | |
1216 | * xbzrle can do better than compression. | |
1217 | */ | |
6b6712ef JQ |
1218 | if (migrate_use_compression() && |
1219 | (rs->ram_bulk_stage || !migrate_use_xbzrle())) { | |
a0a8aa14 | 1220 | res = ram_save_compressed_page(rs, pss, last_stage); |
a82d593b | 1221 | } else { |
a0a8aa14 | 1222 | res = ram_save_page(rs, pss, last_stage); |
a82d593b DDAG |
1223 | } |
1224 | ||
1225 | if (res < 0) { | |
1226 | return res; | |
1227 | } | |
6b6712ef JQ |
1228 | if (pss->block->unsentmap) { |
1229 | clear_bit(pss->page, pss->block->unsentmap); | |
a82d593b DDAG |
1230 | } |
1231 | } | |
1232 | ||
1233 | return res; | |
1234 | } | |
1235 | ||
1236 | /** | |
3d0684b2 | 1237 | * ram_save_host_page: save a whole host page |
a82d593b | 1238 | * |
3d0684b2 JQ |
1239 | * Starting at *offset send pages up to the end of the current host |
1240 | * page. It's valid for the initial offset to point into the middle of | |
1241 | * a host page in which case the remainder of the hostpage is sent. | |
1242 | * Only dirty target pages are sent. Note that the host page size may | |
1243 | * be a huge page for this block. | |
1eb3fc0a DDAG |
1244 | * The saving stops at the boundary of the used_length of the block |
1245 | * if the RAMBlock isn't a multiple of the host page size. | |
a82d593b | 1246 | * |
3d0684b2 JQ |
1247 | * Returns the number of pages written or negative on error |
1248 | * | |
6f37bb8b | 1249 | * @rs: current RAM state |
3d0684b2 | 1250 | * @ms: current migration state |
3d0684b2 | 1251 | * @pss: data about the page we want to send |
a82d593b | 1252 | * @last_stage: if we are at the completion stage |
a82d593b | 1253 | */ |
a0a8aa14 | 1254 | static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 1255 | bool last_stage) |
a82d593b DDAG |
1256 | { |
1257 | int tmppages, pages = 0; | |
a935e30f JQ |
1258 | size_t pagesize_bits = |
1259 | qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
4c011c37 | 1260 | |
a82d593b | 1261 | do { |
f20e2865 | 1262 | tmppages = ram_save_target_page(rs, pss, last_stage); |
a82d593b DDAG |
1263 | if (tmppages < 0) { |
1264 | return tmppages; | |
1265 | } | |
1266 | ||
1267 | pages += tmppages; | |
a935e30f | 1268 | pss->page++; |
1eb3fc0a DDAG |
1269 | } while ((pss->page & (pagesize_bits - 1)) && |
1270 | offset_in_ramblock(pss->block, pss->page << TARGET_PAGE_BITS)); | |
a82d593b DDAG |
1271 | |
1272 | /* The offset we leave with is the last one we looked at */ | |
a935e30f | 1273 | pss->page--; |
a82d593b DDAG |
1274 | return pages; |
1275 | } | |
6c595cde | 1276 | |
56e93d26 | 1277 | /** |
3d0684b2 | 1278 | * ram_find_and_save_block: finds a dirty page and sends it to f |
56e93d26 JQ |
1279 | * |
1280 | * Called within an RCU critical section. | |
1281 | * | |
3d0684b2 | 1282 | * Returns the number of pages written where zero means no dirty pages |
56e93d26 | 1283 | * |
6f37bb8b | 1284 | * @rs: current RAM state |
56e93d26 | 1285 | * @last_stage: if we are at the completion stage |
a82d593b DDAG |
1286 | * |
1287 | * On systems where host-page-size > target-page-size it will send all the | |
1288 | * pages in a host page that are dirty. | |
56e93d26 JQ |
1289 | */ |
1290 | ||
ce25d337 | 1291 | static int ram_find_and_save_block(RAMState *rs, bool last_stage) |
56e93d26 | 1292 | { |
b8fb8cb7 | 1293 | PageSearchStatus pss; |
56e93d26 | 1294 | int pages = 0; |
b9e60928 | 1295 | bool again, found; |
56e93d26 | 1296 | |
0827b9e9 AA |
1297 | /* No dirty page as there is zero RAM */ |
1298 | if (!ram_bytes_total()) { | |
1299 | return pages; | |
1300 | } | |
1301 | ||
6f37bb8b | 1302 | pss.block = rs->last_seen_block; |
a935e30f | 1303 | pss.page = rs->last_page; |
b8fb8cb7 DDAG |
1304 | pss.complete_round = false; |
1305 | ||
1306 | if (!pss.block) { | |
1307 | pss.block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1308 | } | |
56e93d26 | 1309 | |
b9e60928 | 1310 | do { |
a82d593b | 1311 | again = true; |
f20e2865 | 1312 | found = get_queued_page(rs, &pss); |
b9e60928 | 1313 | |
a82d593b DDAG |
1314 | if (!found) { |
1315 | /* priority queue empty, so just search for something dirty */ | |
f20e2865 | 1316 | found = find_dirty_block(rs, &pss, &again); |
a82d593b | 1317 | } |
f3f491fc | 1318 | |
a82d593b | 1319 | if (found) { |
f20e2865 | 1320 | pages = ram_save_host_page(rs, &pss, last_stage); |
56e93d26 | 1321 | } |
b9e60928 | 1322 | } while (!pages && again); |
56e93d26 | 1323 | |
6f37bb8b | 1324 | rs->last_seen_block = pss.block; |
a935e30f | 1325 | rs->last_page = pss.page; |
56e93d26 JQ |
1326 | |
1327 | return pages; | |
1328 | } | |
1329 | ||
1330 | void acct_update_position(QEMUFile *f, size_t size, bool zero) | |
1331 | { | |
1332 | uint64_t pages = size / TARGET_PAGE_SIZE; | |
f7ccd61b | 1333 | |
56e93d26 | 1334 | if (zero) { |
9360447d | 1335 | ram_counters.duplicate += pages; |
56e93d26 | 1336 | } else { |
9360447d JQ |
1337 | ram_counters.normal += pages; |
1338 | ram_counters.transferred += size; | |
56e93d26 JQ |
1339 | qemu_update_position(f, size); |
1340 | } | |
1341 | } | |
1342 | ||
56e93d26 JQ |
1343 | uint64_t ram_bytes_total(void) |
1344 | { | |
1345 | RAMBlock *block; | |
1346 | uint64_t total = 0; | |
1347 | ||
1348 | rcu_read_lock(); | |
99e15582 | 1349 | RAMBLOCK_FOREACH(block) { |
56e93d26 | 1350 | total += block->used_length; |
99e15582 | 1351 | } |
56e93d26 JQ |
1352 | rcu_read_unlock(); |
1353 | return total; | |
1354 | } | |
1355 | ||
1356 | void free_xbzrle_decoded_buf(void) | |
1357 | { | |
1358 | g_free(xbzrle_decoded_buf); | |
1359 | xbzrle_decoded_buf = NULL; | |
1360 | } | |
1361 | ||
6ad2a215 | 1362 | static void ram_migration_cleanup(void *opaque) |
56e93d26 | 1363 | { |
53518d94 | 1364 | RAMState **rsp = opaque; |
6b6712ef | 1365 | RAMBlock *block; |
eb859c53 | 1366 | |
2ff64038 LZ |
1367 | /* caller have hold iothread lock or is in a bh, so there is |
1368 | * no writing race against this migration_bitmap | |
1369 | */ | |
6b6712ef JQ |
1370 | memory_global_dirty_log_stop(); |
1371 | ||
1372 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1373 | g_free(block->bmap); | |
1374 | block->bmap = NULL; | |
1375 | g_free(block->unsentmap); | |
1376 | block->unsentmap = NULL; | |
56e93d26 JQ |
1377 | } |
1378 | ||
1379 | XBZRLE_cache_lock(); | |
1380 | if (XBZRLE.cache) { | |
1381 | cache_fini(XBZRLE.cache); | |
1382 | g_free(XBZRLE.encoded_buf); | |
1383 | g_free(XBZRLE.current_buf); | |
c00e0928 | 1384 | g_free(XBZRLE.zero_target_page); |
56e93d26 JQ |
1385 | XBZRLE.cache = NULL; |
1386 | XBZRLE.encoded_buf = NULL; | |
1387 | XBZRLE.current_buf = NULL; | |
c00e0928 | 1388 | XBZRLE.zero_target_page = NULL; |
56e93d26 JQ |
1389 | } |
1390 | XBZRLE_cache_unlock(); | |
53518d94 JQ |
1391 | migration_page_queue_free(*rsp); |
1392 | g_free(*rsp); | |
1393 | *rsp = NULL; | |
56e93d26 JQ |
1394 | } |
1395 | ||
6f37bb8b | 1396 | static void ram_state_reset(RAMState *rs) |
56e93d26 | 1397 | { |
6f37bb8b JQ |
1398 | rs->last_seen_block = NULL; |
1399 | rs->last_sent_block = NULL; | |
269ace29 | 1400 | rs->last_page = 0; |
6f37bb8b JQ |
1401 | rs->last_version = ram_list.version; |
1402 | rs->ram_bulk_stage = true; | |
56e93d26 JQ |
1403 | } |
1404 | ||
1405 | #define MAX_WAIT 50 /* ms, half buffered_file limit */ | |
1406 | ||
4f2e4252 DDAG |
1407 | /* |
1408 | * 'expected' is the value you expect the bitmap mostly to be full | |
1409 | * of; it won't bother printing lines that are all this value. | |
1410 | * If 'todump' is null the migration bitmap is dumped. | |
1411 | */ | |
6b6712ef JQ |
1412 | void ram_debug_dump_bitmap(unsigned long *todump, bool expected, |
1413 | unsigned long pages) | |
4f2e4252 | 1414 | { |
4f2e4252 DDAG |
1415 | int64_t cur; |
1416 | int64_t linelen = 128; | |
1417 | char linebuf[129]; | |
1418 | ||
6b6712ef | 1419 | for (cur = 0; cur < pages; cur += linelen) { |
4f2e4252 DDAG |
1420 | int64_t curb; |
1421 | bool found = false; | |
1422 | /* | |
1423 | * Last line; catch the case where the line length | |
1424 | * is longer than remaining ram | |
1425 | */ | |
6b6712ef JQ |
1426 | if (cur + linelen > pages) { |
1427 | linelen = pages - cur; | |
4f2e4252 DDAG |
1428 | } |
1429 | for (curb = 0; curb < linelen; curb++) { | |
1430 | bool thisbit = test_bit(cur + curb, todump); | |
1431 | linebuf[curb] = thisbit ? '1' : '.'; | |
1432 | found = found || (thisbit != expected); | |
1433 | } | |
1434 | if (found) { | |
1435 | linebuf[curb] = '\0'; | |
1436 | fprintf(stderr, "0x%08" PRIx64 " : %s\n", cur, linebuf); | |
1437 | } | |
1438 | } | |
1439 | } | |
1440 | ||
e0b266f0 DDAG |
1441 | /* **** functions for postcopy ***** */ |
1442 | ||
ced1c616 PB |
1443 | void ram_postcopy_migrated_memory_release(MigrationState *ms) |
1444 | { | |
1445 | struct RAMBlock *block; | |
ced1c616 | 1446 | |
99e15582 | 1447 | RAMBLOCK_FOREACH(block) { |
6b6712ef JQ |
1448 | unsigned long *bitmap = block->bmap; |
1449 | unsigned long range = block->used_length >> TARGET_PAGE_BITS; | |
1450 | unsigned long run_start = find_next_zero_bit(bitmap, range, 0); | |
ced1c616 PB |
1451 | |
1452 | while (run_start < range) { | |
1453 | unsigned long run_end = find_next_bit(bitmap, range, run_start + 1); | |
aaa2064c | 1454 | ram_discard_range(block->idstr, run_start << TARGET_PAGE_BITS, |
ced1c616 PB |
1455 | (run_end - run_start) << TARGET_PAGE_BITS); |
1456 | run_start = find_next_zero_bit(bitmap, range, run_end + 1); | |
1457 | } | |
1458 | } | |
1459 | } | |
1460 | ||
3d0684b2 JQ |
1461 | /** |
1462 | * postcopy_send_discard_bm_ram: discard a RAMBlock | |
1463 | * | |
1464 | * Returns zero on success | |
1465 | * | |
e0b266f0 DDAG |
1466 | * Callback from postcopy_each_ram_send_discard for each RAMBlock |
1467 | * Note: At this point the 'unsentmap' is the processed bitmap combined | |
1468 | * with the dirtymap; so a '1' means it's either dirty or unsent. | |
3d0684b2 JQ |
1469 | * |
1470 | * @ms: current migration state | |
1471 | * @pds: state for postcopy | |
1472 | * @start: RAMBlock starting page | |
1473 | * @length: RAMBlock size | |
e0b266f0 DDAG |
1474 | */ |
1475 | static int postcopy_send_discard_bm_ram(MigrationState *ms, | |
1476 | PostcopyDiscardState *pds, | |
6b6712ef | 1477 | RAMBlock *block) |
e0b266f0 | 1478 | { |
6b6712ef | 1479 | unsigned long end = block->used_length >> TARGET_PAGE_BITS; |
e0b266f0 | 1480 | unsigned long current; |
6b6712ef | 1481 | unsigned long *unsentmap = block->unsentmap; |
e0b266f0 | 1482 | |
6b6712ef | 1483 | for (current = 0; current < end; ) { |
e0b266f0 DDAG |
1484 | unsigned long one = find_next_bit(unsentmap, end, current); |
1485 | ||
1486 | if (one <= end) { | |
1487 | unsigned long zero = find_next_zero_bit(unsentmap, end, one + 1); | |
1488 | unsigned long discard_length; | |
1489 | ||
1490 | if (zero >= end) { | |
1491 | discard_length = end - one; | |
1492 | } else { | |
1493 | discard_length = zero - one; | |
1494 | } | |
d688c62d DDAG |
1495 | if (discard_length) { |
1496 | postcopy_discard_send_range(ms, pds, one, discard_length); | |
1497 | } | |
e0b266f0 DDAG |
1498 | current = one + discard_length; |
1499 | } else { | |
1500 | current = one; | |
1501 | } | |
1502 | } | |
1503 | ||
1504 | return 0; | |
1505 | } | |
1506 | ||
3d0684b2 JQ |
1507 | /** |
1508 | * postcopy_each_ram_send_discard: discard all RAMBlocks | |
1509 | * | |
1510 | * Returns 0 for success or negative for error | |
1511 | * | |
e0b266f0 DDAG |
1512 | * Utility for the outgoing postcopy code. |
1513 | * Calls postcopy_send_discard_bm_ram for each RAMBlock | |
1514 | * passing it bitmap indexes and name. | |
e0b266f0 DDAG |
1515 | * (qemu_ram_foreach_block ends up passing unscaled lengths |
1516 | * which would mean postcopy code would have to deal with target page) | |
3d0684b2 JQ |
1517 | * |
1518 | * @ms: current migration state | |
e0b266f0 DDAG |
1519 | */ |
1520 | static int postcopy_each_ram_send_discard(MigrationState *ms) | |
1521 | { | |
1522 | struct RAMBlock *block; | |
1523 | int ret; | |
1524 | ||
99e15582 | 1525 | RAMBLOCK_FOREACH(block) { |
6b6712ef JQ |
1526 | PostcopyDiscardState *pds = |
1527 | postcopy_discard_send_init(ms, block->idstr); | |
e0b266f0 DDAG |
1528 | |
1529 | /* | |
1530 | * Postcopy sends chunks of bitmap over the wire, but it | |
1531 | * just needs indexes at this point, avoids it having | |
1532 | * target page specific code. | |
1533 | */ | |
6b6712ef | 1534 | ret = postcopy_send_discard_bm_ram(ms, pds, block); |
e0b266f0 DDAG |
1535 | postcopy_discard_send_finish(ms, pds); |
1536 | if (ret) { | |
1537 | return ret; | |
1538 | } | |
1539 | } | |
1540 | ||
1541 | return 0; | |
1542 | } | |
1543 | ||
3d0684b2 JQ |
1544 | /** |
1545 | * postcopy_chunk_hostpages_pass: canocalize bitmap in hostpages | |
1546 | * | |
1547 | * Helper for postcopy_chunk_hostpages; it's called twice to | |
1548 | * canonicalize the two bitmaps, that are similar, but one is | |
1549 | * inverted. | |
99e314eb | 1550 | * |
3d0684b2 JQ |
1551 | * Postcopy requires that all target pages in a hostpage are dirty or |
1552 | * clean, not a mix. This function canonicalizes the bitmaps. | |
99e314eb | 1553 | * |
3d0684b2 JQ |
1554 | * @ms: current migration state |
1555 | * @unsent_pass: if true we need to canonicalize partially unsent host pages | |
1556 | * otherwise we need to canonicalize partially dirty host pages | |
1557 | * @block: block that contains the page we want to canonicalize | |
1558 | * @pds: state for postcopy | |
99e314eb DDAG |
1559 | */ |
1560 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, bool unsent_pass, | |
1561 | RAMBlock *block, | |
1562 | PostcopyDiscardState *pds) | |
1563 | { | |
53518d94 | 1564 | RAMState *rs = ram_state; |
6b6712ef JQ |
1565 | unsigned long *bitmap = block->bmap; |
1566 | unsigned long *unsentmap = block->unsentmap; | |
29c59172 | 1567 | unsigned int host_ratio = block->page_size / TARGET_PAGE_SIZE; |
6b6712ef | 1568 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
99e314eb DDAG |
1569 | unsigned long run_start; |
1570 | ||
29c59172 DDAG |
1571 | if (block->page_size == TARGET_PAGE_SIZE) { |
1572 | /* Easy case - TPS==HPS for a non-huge page RAMBlock */ | |
1573 | return; | |
1574 | } | |
1575 | ||
99e314eb DDAG |
1576 | if (unsent_pass) { |
1577 | /* Find a sent page */ | |
6b6712ef | 1578 | run_start = find_next_zero_bit(unsentmap, pages, 0); |
99e314eb DDAG |
1579 | } else { |
1580 | /* Find a dirty page */ | |
6b6712ef | 1581 | run_start = find_next_bit(bitmap, pages, 0); |
99e314eb DDAG |
1582 | } |
1583 | ||
6b6712ef | 1584 | while (run_start < pages) { |
99e314eb DDAG |
1585 | bool do_fixup = false; |
1586 | unsigned long fixup_start_addr; | |
1587 | unsigned long host_offset; | |
1588 | ||
1589 | /* | |
1590 | * If the start of this run of pages is in the middle of a host | |
1591 | * page, then we need to fixup this host page. | |
1592 | */ | |
1593 | host_offset = run_start % host_ratio; | |
1594 | if (host_offset) { | |
1595 | do_fixup = true; | |
1596 | run_start -= host_offset; | |
1597 | fixup_start_addr = run_start; | |
1598 | /* For the next pass */ | |
1599 | run_start = run_start + host_ratio; | |
1600 | } else { | |
1601 | /* Find the end of this run */ | |
1602 | unsigned long run_end; | |
1603 | if (unsent_pass) { | |
6b6712ef | 1604 | run_end = find_next_bit(unsentmap, pages, run_start + 1); |
99e314eb | 1605 | } else { |
6b6712ef | 1606 | run_end = find_next_zero_bit(bitmap, pages, run_start + 1); |
99e314eb DDAG |
1607 | } |
1608 | /* | |
1609 | * If the end isn't at the start of a host page, then the | |
1610 | * run doesn't finish at the end of a host page | |
1611 | * and we need to discard. | |
1612 | */ | |
1613 | host_offset = run_end % host_ratio; | |
1614 | if (host_offset) { | |
1615 | do_fixup = true; | |
1616 | fixup_start_addr = run_end - host_offset; | |
1617 | /* | |
1618 | * This host page has gone, the next loop iteration starts | |
1619 | * from after the fixup | |
1620 | */ | |
1621 | run_start = fixup_start_addr + host_ratio; | |
1622 | } else { | |
1623 | /* | |
1624 | * No discards on this iteration, next loop starts from | |
1625 | * next sent/dirty page | |
1626 | */ | |
1627 | run_start = run_end + 1; | |
1628 | } | |
1629 | } | |
1630 | ||
1631 | if (do_fixup) { | |
1632 | unsigned long page; | |
1633 | ||
1634 | /* Tell the destination to discard this page */ | |
1635 | if (unsent_pass || !test_bit(fixup_start_addr, unsentmap)) { | |
1636 | /* For the unsent_pass we: | |
1637 | * discard partially sent pages | |
1638 | * For the !unsent_pass (dirty) we: | |
1639 | * discard partially dirty pages that were sent | |
1640 | * (any partially sent pages were already discarded | |
1641 | * by the previous unsent_pass) | |
1642 | */ | |
1643 | postcopy_discard_send_range(ms, pds, fixup_start_addr, | |
1644 | host_ratio); | |
1645 | } | |
1646 | ||
1647 | /* Clean up the bitmap */ | |
1648 | for (page = fixup_start_addr; | |
1649 | page < fixup_start_addr + host_ratio; page++) { | |
1650 | /* All pages in this host page are now not sent */ | |
1651 | set_bit(page, unsentmap); | |
1652 | ||
1653 | /* | |
1654 | * Remark them as dirty, updating the count for any pages | |
1655 | * that weren't previously dirty. | |
1656 | */ | |
0d8ec885 | 1657 | rs->migration_dirty_pages += !test_and_set_bit(page, bitmap); |
99e314eb DDAG |
1658 | } |
1659 | } | |
1660 | ||
1661 | if (unsent_pass) { | |
1662 | /* Find the next sent page for the next iteration */ | |
6b6712ef | 1663 | run_start = find_next_zero_bit(unsentmap, pages, run_start); |
99e314eb DDAG |
1664 | } else { |
1665 | /* Find the next dirty page for the next iteration */ | |
6b6712ef | 1666 | run_start = find_next_bit(bitmap, pages, run_start); |
99e314eb DDAG |
1667 | } |
1668 | } | |
1669 | } | |
1670 | ||
3d0684b2 JQ |
1671 | /** |
1672 | * postcopy_chuck_hostpages: discrad any partially sent host page | |
1673 | * | |
99e314eb DDAG |
1674 | * Utility for the outgoing postcopy code. |
1675 | * | |
1676 | * Discard any partially sent host-page size chunks, mark any partially | |
29c59172 DDAG |
1677 | * dirty host-page size chunks as all dirty. In this case the host-page |
1678 | * is the host-page for the particular RAMBlock, i.e. it might be a huge page | |
99e314eb | 1679 | * |
3d0684b2 JQ |
1680 | * Returns zero on success |
1681 | * | |
1682 | * @ms: current migration state | |
6b6712ef | 1683 | * @block: block we want to work with |
99e314eb | 1684 | */ |
6b6712ef | 1685 | static int postcopy_chunk_hostpages(MigrationState *ms, RAMBlock *block) |
99e314eb | 1686 | { |
6b6712ef JQ |
1687 | PostcopyDiscardState *pds = |
1688 | postcopy_discard_send_init(ms, block->idstr); | |
99e314eb | 1689 | |
6b6712ef JQ |
1690 | /* First pass: Discard all partially sent host pages */ |
1691 | postcopy_chunk_hostpages_pass(ms, true, block, pds); | |
1692 | /* | |
1693 | * Second pass: Ensure that all partially dirty host pages are made | |
1694 | * fully dirty. | |
1695 | */ | |
1696 | postcopy_chunk_hostpages_pass(ms, false, block, pds); | |
99e314eb | 1697 | |
6b6712ef | 1698 | postcopy_discard_send_finish(ms, pds); |
99e314eb DDAG |
1699 | return 0; |
1700 | } | |
1701 | ||
3d0684b2 JQ |
1702 | /** |
1703 | * ram_postcopy_send_discard_bitmap: transmit the discard bitmap | |
1704 | * | |
1705 | * Returns zero on success | |
1706 | * | |
e0b266f0 DDAG |
1707 | * Transmit the set of pages to be discarded after precopy to the target |
1708 | * these are pages that: | |
1709 | * a) Have been previously transmitted but are now dirty again | |
1710 | * b) Pages that have never been transmitted, this ensures that | |
1711 | * any pages on the destination that have been mapped by background | |
1712 | * tasks get discarded (transparent huge pages is the specific concern) | |
1713 | * Hopefully this is pretty sparse | |
3d0684b2 JQ |
1714 | * |
1715 | * @ms: current migration state | |
e0b266f0 DDAG |
1716 | */ |
1717 | int ram_postcopy_send_discard_bitmap(MigrationState *ms) | |
1718 | { | |
53518d94 | 1719 | RAMState *rs = ram_state; |
6b6712ef | 1720 | RAMBlock *block; |
e0b266f0 | 1721 | int ret; |
e0b266f0 DDAG |
1722 | |
1723 | rcu_read_lock(); | |
1724 | ||
1725 | /* This should be our last sync, the src is now paused */ | |
eb859c53 | 1726 | migration_bitmap_sync(rs); |
e0b266f0 | 1727 | |
6b6712ef JQ |
1728 | /* Easiest way to make sure we don't resume in the middle of a host-page */ |
1729 | rs->last_seen_block = NULL; | |
1730 | rs->last_sent_block = NULL; | |
1731 | rs->last_page = 0; | |
e0b266f0 | 1732 | |
6b6712ef JQ |
1733 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { |
1734 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; | |
1735 | unsigned long *bitmap = block->bmap; | |
1736 | unsigned long *unsentmap = block->unsentmap; | |
1737 | ||
1738 | if (!unsentmap) { | |
1739 | /* We don't have a safe way to resize the sentmap, so | |
1740 | * if the bitmap was resized it will be NULL at this | |
1741 | * point. | |
1742 | */ | |
1743 | error_report("migration ram resized during precopy phase"); | |
1744 | rcu_read_unlock(); | |
1745 | return -EINVAL; | |
1746 | } | |
1747 | /* Deal with TPS != HPS and huge pages */ | |
1748 | ret = postcopy_chunk_hostpages(ms, block); | |
1749 | if (ret) { | |
1750 | rcu_read_unlock(); | |
1751 | return ret; | |
1752 | } | |
e0b266f0 | 1753 | |
6b6712ef JQ |
1754 | /* |
1755 | * Update the unsentmap to be unsentmap = unsentmap | dirty | |
1756 | */ | |
1757 | bitmap_or(unsentmap, unsentmap, bitmap, pages); | |
e0b266f0 | 1758 | #ifdef DEBUG_POSTCOPY |
6b6712ef | 1759 | ram_debug_dump_bitmap(unsentmap, true, pages); |
e0b266f0 | 1760 | #endif |
6b6712ef JQ |
1761 | } |
1762 | trace_ram_postcopy_send_discard_bitmap(); | |
e0b266f0 DDAG |
1763 | |
1764 | ret = postcopy_each_ram_send_discard(ms); | |
1765 | rcu_read_unlock(); | |
1766 | ||
1767 | return ret; | |
1768 | } | |
1769 | ||
3d0684b2 JQ |
1770 | /** |
1771 | * ram_discard_range: discard dirtied pages at the beginning of postcopy | |
e0b266f0 | 1772 | * |
3d0684b2 | 1773 | * Returns zero on success |
e0b266f0 | 1774 | * |
36449157 JQ |
1775 | * @rbname: name of the RAMBlock of the request. NULL means the |
1776 | * same that last one. | |
3d0684b2 JQ |
1777 | * @start: RAMBlock starting page |
1778 | * @length: RAMBlock size | |
e0b266f0 | 1779 | */ |
aaa2064c | 1780 | int ram_discard_range(const char *rbname, uint64_t start, size_t length) |
e0b266f0 DDAG |
1781 | { |
1782 | int ret = -1; | |
1783 | ||
36449157 | 1784 | trace_ram_discard_range(rbname, start, length); |
d3a5038c | 1785 | |
e0b266f0 | 1786 | rcu_read_lock(); |
36449157 | 1787 | RAMBlock *rb = qemu_ram_block_by_name(rbname); |
e0b266f0 DDAG |
1788 | |
1789 | if (!rb) { | |
36449157 | 1790 | error_report("ram_discard_range: Failed to find block '%s'", rbname); |
e0b266f0 DDAG |
1791 | goto err; |
1792 | } | |
1793 | ||
d3a5038c | 1794 | ret = ram_block_discard_range(rb, start, length); |
e0b266f0 DDAG |
1795 | |
1796 | err: | |
1797 | rcu_read_unlock(); | |
1798 | ||
1799 | return ret; | |
1800 | } | |
1801 | ||
53518d94 | 1802 | static int ram_state_init(RAMState **rsp) |
56e93d26 | 1803 | { |
53518d94 JQ |
1804 | *rsp = g_new0(RAMState, 1); |
1805 | ||
1806 | qemu_mutex_init(&(*rsp)->bitmap_mutex); | |
1807 | qemu_mutex_init(&(*rsp)->src_page_req_mutex); | |
1808 | QSIMPLEQ_INIT(&(*rsp)->src_page_requests); | |
56e93d26 JQ |
1809 | |
1810 | if (migrate_use_xbzrle()) { | |
1811 | XBZRLE_cache_lock(); | |
c00e0928 | 1812 | XBZRLE.zero_target_page = g_malloc0(TARGET_PAGE_SIZE); |
56e93d26 JQ |
1813 | XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() / |
1814 | TARGET_PAGE_SIZE, | |
1815 | TARGET_PAGE_SIZE); | |
1816 | if (!XBZRLE.cache) { | |
1817 | XBZRLE_cache_unlock(); | |
1818 | error_report("Error creating cache"); | |
53518d94 JQ |
1819 | g_free(*rsp); |
1820 | *rsp = NULL; | |
56e93d26 JQ |
1821 | return -1; |
1822 | } | |
1823 | XBZRLE_cache_unlock(); | |
1824 | ||
1825 | /* We prefer not to abort if there is no memory */ | |
1826 | XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); | |
1827 | if (!XBZRLE.encoded_buf) { | |
1828 | error_report("Error allocating encoded_buf"); | |
53518d94 JQ |
1829 | g_free(*rsp); |
1830 | *rsp = NULL; | |
56e93d26 JQ |
1831 | return -1; |
1832 | } | |
1833 | ||
1834 | XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); | |
1835 | if (!XBZRLE.current_buf) { | |
1836 | error_report("Error allocating current_buf"); | |
1837 | g_free(XBZRLE.encoded_buf); | |
1838 | XBZRLE.encoded_buf = NULL; | |
53518d94 JQ |
1839 | g_free(*rsp); |
1840 | *rsp = NULL; | |
56e93d26 JQ |
1841 | return -1; |
1842 | } | |
56e93d26 JQ |
1843 | } |
1844 | ||
49877834 PB |
1845 | /* For memory_global_dirty_log_start below. */ |
1846 | qemu_mutex_lock_iothread(); | |
1847 | ||
56e93d26 JQ |
1848 | qemu_mutex_lock_ramlist(); |
1849 | rcu_read_lock(); | |
53518d94 | 1850 | ram_state_reset(*rsp); |
56e93d26 | 1851 | |
0827b9e9 AA |
1852 | /* Skip setting bitmap if there is no RAM */ |
1853 | if (ram_bytes_total()) { | |
6b6712ef JQ |
1854 | RAMBlock *block; |
1855 | ||
1856 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1857 | unsigned long pages = block->max_length >> TARGET_PAGE_BITS; | |
0827b9e9 | 1858 | |
6b6712ef JQ |
1859 | block->bmap = bitmap_new(pages); |
1860 | bitmap_set(block->bmap, 0, pages); | |
1861 | if (migrate_postcopy_ram()) { | |
1862 | block->unsentmap = bitmap_new(pages); | |
1863 | bitmap_set(block->unsentmap, 0, pages); | |
1864 | } | |
0827b9e9 | 1865 | } |
f3f491fc DDAG |
1866 | } |
1867 | ||
56e93d26 JQ |
1868 | /* |
1869 | * Count the total number of pages used by ram blocks not including any | |
1870 | * gaps due to alignment or unplugs. | |
1871 | */ | |
53518d94 | 1872 | (*rsp)->migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS; |
56e93d26 JQ |
1873 | |
1874 | memory_global_dirty_log_start(); | |
53518d94 | 1875 | migration_bitmap_sync(*rsp); |
56e93d26 | 1876 | qemu_mutex_unlock_ramlist(); |
49877834 | 1877 | qemu_mutex_unlock_iothread(); |
a91246c9 HZ |
1878 | rcu_read_unlock(); |
1879 | ||
1880 | return 0; | |
1881 | } | |
1882 | ||
3d0684b2 JQ |
1883 | /* |
1884 | * Each of ram_save_setup, ram_save_iterate and ram_save_complete has | |
a91246c9 HZ |
1885 | * long-running RCU critical section. When rcu-reclaims in the code |
1886 | * start to become numerous it will be necessary to reduce the | |
1887 | * granularity of these critical sections. | |
1888 | */ | |
1889 | ||
3d0684b2 JQ |
1890 | /** |
1891 | * ram_save_setup: Setup RAM for migration | |
1892 | * | |
1893 | * Returns zero to indicate success and negative for error | |
1894 | * | |
1895 | * @f: QEMUFile where to send the data | |
1896 | * @opaque: RAMState pointer | |
1897 | */ | |
a91246c9 HZ |
1898 | static int ram_save_setup(QEMUFile *f, void *opaque) |
1899 | { | |
53518d94 | 1900 | RAMState **rsp = opaque; |
a91246c9 HZ |
1901 | RAMBlock *block; |
1902 | ||
1903 | /* migration has already setup the bitmap, reuse it. */ | |
1904 | if (!migration_in_colo_state()) { | |
53518d94 | 1905 | if (ram_state_init(rsp) != 0) { |
a91246c9 | 1906 | return -1; |
53518d94 | 1907 | } |
a91246c9 | 1908 | } |
53518d94 | 1909 | (*rsp)->f = f; |
a91246c9 HZ |
1910 | |
1911 | rcu_read_lock(); | |
56e93d26 JQ |
1912 | |
1913 | qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE); | |
1914 | ||
99e15582 | 1915 | RAMBLOCK_FOREACH(block) { |
56e93d26 JQ |
1916 | qemu_put_byte(f, strlen(block->idstr)); |
1917 | qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); | |
1918 | qemu_put_be64(f, block->used_length); | |
ef08fb38 DDAG |
1919 | if (migrate_postcopy_ram() && block->page_size != qemu_host_page_size) { |
1920 | qemu_put_be64(f, block->page_size); | |
1921 | } | |
56e93d26 JQ |
1922 | } |
1923 | ||
1924 | rcu_read_unlock(); | |
1925 | ||
1926 | ram_control_before_iterate(f, RAM_CONTROL_SETUP); | |
1927 | ram_control_after_iterate(f, RAM_CONTROL_SETUP); | |
1928 | ||
1929 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); | |
1930 | ||
1931 | return 0; | |
1932 | } | |
1933 | ||
3d0684b2 JQ |
1934 | /** |
1935 | * ram_save_iterate: iterative stage for migration | |
1936 | * | |
1937 | * Returns zero to indicate success and negative for error | |
1938 | * | |
1939 | * @f: QEMUFile where to send the data | |
1940 | * @opaque: RAMState pointer | |
1941 | */ | |
56e93d26 JQ |
1942 | static int ram_save_iterate(QEMUFile *f, void *opaque) |
1943 | { | |
53518d94 JQ |
1944 | RAMState **temp = opaque; |
1945 | RAMState *rs = *temp; | |
56e93d26 JQ |
1946 | int ret; |
1947 | int i; | |
1948 | int64_t t0; | |
5c90308f | 1949 | int done = 0; |
56e93d26 JQ |
1950 | |
1951 | rcu_read_lock(); | |
6f37bb8b JQ |
1952 | if (ram_list.version != rs->last_version) { |
1953 | ram_state_reset(rs); | |
56e93d26 JQ |
1954 | } |
1955 | ||
1956 | /* Read version before ram_list.blocks */ | |
1957 | smp_rmb(); | |
1958 | ||
1959 | ram_control_before_iterate(f, RAM_CONTROL_ROUND); | |
1960 | ||
1961 | t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
1962 | i = 0; | |
1963 | while ((ret = qemu_file_rate_limit(f)) == 0) { | |
1964 | int pages; | |
1965 | ||
ce25d337 | 1966 | pages = ram_find_and_save_block(rs, false); |
56e93d26 JQ |
1967 | /* no more pages to sent */ |
1968 | if (pages == 0) { | |
5c90308f | 1969 | done = 1; |
56e93d26 JQ |
1970 | break; |
1971 | } | |
23b28c3c | 1972 | rs->iterations++; |
070afca2 | 1973 | |
56e93d26 JQ |
1974 | /* we want to check in the 1st loop, just in case it was the 1st time |
1975 | and we had to sync the dirty bitmap. | |
1976 | qemu_get_clock_ns() is a bit expensive, so we only check each some | |
1977 | iterations | |
1978 | */ | |
1979 | if ((i & 63) == 0) { | |
1980 | uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000; | |
1981 | if (t1 > MAX_WAIT) { | |
55c4446b | 1982 | trace_ram_save_iterate_big_wait(t1, i); |
56e93d26 JQ |
1983 | break; |
1984 | } | |
1985 | } | |
1986 | i++; | |
1987 | } | |
ce25d337 | 1988 | flush_compressed_data(rs); |
56e93d26 JQ |
1989 | rcu_read_unlock(); |
1990 | ||
1991 | /* | |
1992 | * Must occur before EOS (or any QEMUFile operation) | |
1993 | * because of RDMA protocol. | |
1994 | */ | |
1995 | ram_control_after_iterate(f, RAM_CONTROL_ROUND); | |
1996 | ||
1997 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); | |
9360447d | 1998 | ram_counters.transferred += 8; |
56e93d26 JQ |
1999 | |
2000 | ret = qemu_file_get_error(f); | |
2001 | if (ret < 0) { | |
2002 | return ret; | |
2003 | } | |
2004 | ||
5c90308f | 2005 | return done; |
56e93d26 JQ |
2006 | } |
2007 | ||
3d0684b2 JQ |
2008 | /** |
2009 | * ram_save_complete: function called to send the remaining amount of ram | |
2010 | * | |
2011 | * Returns zero to indicate success | |
2012 | * | |
2013 | * Called with iothread lock | |
2014 | * | |
2015 | * @f: QEMUFile where to send the data | |
2016 | * @opaque: RAMState pointer | |
2017 | */ | |
56e93d26 JQ |
2018 | static int ram_save_complete(QEMUFile *f, void *opaque) |
2019 | { | |
53518d94 JQ |
2020 | RAMState **temp = opaque; |
2021 | RAMState *rs = *temp; | |
6f37bb8b | 2022 | |
56e93d26 JQ |
2023 | rcu_read_lock(); |
2024 | ||
5727309d | 2025 | if (!migration_in_postcopy()) { |
8d820d6f | 2026 | migration_bitmap_sync(rs); |
663e6c1d | 2027 | } |
56e93d26 JQ |
2028 | |
2029 | ram_control_before_iterate(f, RAM_CONTROL_FINISH); | |
2030 | ||
2031 | /* try transferring iterative blocks of memory */ | |
2032 | ||
2033 | /* flush all remaining blocks regardless of rate limiting */ | |
2034 | while (true) { | |
2035 | int pages; | |
2036 | ||
ce25d337 | 2037 | pages = ram_find_and_save_block(rs, !migration_in_colo_state()); |
56e93d26 JQ |
2038 | /* no more blocks to sent */ |
2039 | if (pages == 0) { | |
2040 | break; | |
2041 | } | |
2042 | } | |
2043 | ||
ce25d337 | 2044 | flush_compressed_data(rs); |
56e93d26 | 2045 | ram_control_after_iterate(f, RAM_CONTROL_FINISH); |
56e93d26 JQ |
2046 | |
2047 | rcu_read_unlock(); | |
d09a6fde | 2048 | |
56e93d26 JQ |
2049 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
2050 | ||
2051 | return 0; | |
2052 | } | |
2053 | ||
c31b098f DDAG |
2054 | static void ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size, |
2055 | uint64_t *non_postcopiable_pending, | |
2056 | uint64_t *postcopiable_pending) | |
56e93d26 | 2057 | { |
53518d94 JQ |
2058 | RAMState **temp = opaque; |
2059 | RAMState *rs = *temp; | |
56e93d26 JQ |
2060 | uint64_t remaining_size; |
2061 | ||
9edabd4d | 2062 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2063 | |
5727309d | 2064 | if (!migration_in_postcopy() && |
663e6c1d | 2065 | remaining_size < max_size) { |
56e93d26 JQ |
2066 | qemu_mutex_lock_iothread(); |
2067 | rcu_read_lock(); | |
8d820d6f | 2068 | migration_bitmap_sync(rs); |
56e93d26 JQ |
2069 | rcu_read_unlock(); |
2070 | qemu_mutex_unlock_iothread(); | |
9edabd4d | 2071 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2072 | } |
c31b098f DDAG |
2073 | |
2074 | /* We can do postcopy, and all the data is postcopiable */ | |
2075 | *postcopiable_pending += remaining_size; | |
56e93d26 JQ |
2076 | } |
2077 | ||
2078 | static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) | |
2079 | { | |
2080 | unsigned int xh_len; | |
2081 | int xh_flags; | |
063e760a | 2082 | uint8_t *loaded_data; |
56e93d26 JQ |
2083 | |
2084 | if (!xbzrle_decoded_buf) { | |
2085 | xbzrle_decoded_buf = g_malloc(TARGET_PAGE_SIZE); | |
2086 | } | |
063e760a | 2087 | loaded_data = xbzrle_decoded_buf; |
56e93d26 JQ |
2088 | |
2089 | /* extract RLE header */ | |
2090 | xh_flags = qemu_get_byte(f); | |
2091 | xh_len = qemu_get_be16(f); | |
2092 | ||
2093 | if (xh_flags != ENCODING_FLAG_XBZRLE) { | |
2094 | error_report("Failed to load XBZRLE page - wrong compression!"); | |
2095 | return -1; | |
2096 | } | |
2097 | ||
2098 | if (xh_len > TARGET_PAGE_SIZE) { | |
2099 | error_report("Failed to load XBZRLE page - len overflow!"); | |
2100 | return -1; | |
2101 | } | |
2102 | /* load data and decode */ | |
063e760a | 2103 | qemu_get_buffer_in_place(f, &loaded_data, xh_len); |
56e93d26 JQ |
2104 | |
2105 | /* decode RLE */ | |
063e760a | 2106 | if (xbzrle_decode_buffer(loaded_data, xh_len, host, |
56e93d26 JQ |
2107 | TARGET_PAGE_SIZE) == -1) { |
2108 | error_report("Failed to load XBZRLE page - decode error!"); | |
2109 | return -1; | |
2110 | } | |
2111 | ||
2112 | return 0; | |
2113 | } | |
2114 | ||
3d0684b2 JQ |
2115 | /** |
2116 | * ram_block_from_stream: read a RAMBlock id from the migration stream | |
2117 | * | |
2118 | * Must be called from within a rcu critical section. | |
2119 | * | |
56e93d26 | 2120 | * Returns a pointer from within the RCU-protected ram_list. |
a7180877 | 2121 | * |
3d0684b2 JQ |
2122 | * @f: QEMUFile where to read the data from |
2123 | * @flags: Page flags (mostly to see if it's a continuation of previous block) | |
a7180877 | 2124 | */ |
3d0684b2 | 2125 | static inline RAMBlock *ram_block_from_stream(QEMUFile *f, int flags) |
56e93d26 JQ |
2126 | { |
2127 | static RAMBlock *block = NULL; | |
2128 | char id[256]; | |
2129 | uint8_t len; | |
2130 | ||
2131 | if (flags & RAM_SAVE_FLAG_CONTINUE) { | |
4c4bad48 | 2132 | if (!block) { |
56e93d26 JQ |
2133 | error_report("Ack, bad migration stream!"); |
2134 | return NULL; | |
2135 | } | |
4c4bad48 | 2136 | return block; |
56e93d26 JQ |
2137 | } |
2138 | ||
2139 | len = qemu_get_byte(f); | |
2140 | qemu_get_buffer(f, (uint8_t *)id, len); | |
2141 | id[len] = 0; | |
2142 | ||
e3dd7493 | 2143 | block = qemu_ram_block_by_name(id); |
4c4bad48 HZ |
2144 | if (!block) { |
2145 | error_report("Can't find block %s", id); | |
2146 | return NULL; | |
56e93d26 JQ |
2147 | } |
2148 | ||
4c4bad48 HZ |
2149 | return block; |
2150 | } | |
2151 | ||
2152 | static inline void *host_from_ram_block_offset(RAMBlock *block, | |
2153 | ram_addr_t offset) | |
2154 | { | |
2155 | if (!offset_in_ramblock(block, offset)) { | |
2156 | return NULL; | |
2157 | } | |
2158 | ||
2159 | return block->host + offset; | |
56e93d26 JQ |
2160 | } |
2161 | ||
3d0684b2 JQ |
2162 | /** |
2163 | * ram_handle_compressed: handle the zero page case | |
2164 | * | |
56e93d26 JQ |
2165 | * If a page (or a whole RDMA chunk) has been |
2166 | * determined to be zero, then zap it. | |
3d0684b2 JQ |
2167 | * |
2168 | * @host: host address for the zero page | |
2169 | * @ch: what the page is filled from. We only support zero | |
2170 | * @size: size of the zero page | |
56e93d26 JQ |
2171 | */ |
2172 | void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) | |
2173 | { | |
2174 | if (ch != 0 || !is_zero_range(host, size)) { | |
2175 | memset(host, ch, size); | |
2176 | } | |
2177 | } | |
2178 | ||
2179 | static void *do_data_decompress(void *opaque) | |
2180 | { | |
2181 | DecompressParam *param = opaque; | |
2182 | unsigned long pagesize; | |
33d151f4 LL |
2183 | uint8_t *des; |
2184 | int len; | |
56e93d26 | 2185 | |
33d151f4 | 2186 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 2187 | while (!param->quit) { |
33d151f4 LL |
2188 | if (param->des) { |
2189 | des = param->des; | |
2190 | len = param->len; | |
2191 | param->des = 0; | |
2192 | qemu_mutex_unlock(¶m->mutex); | |
2193 | ||
56e93d26 | 2194 | pagesize = TARGET_PAGE_SIZE; |
73a8912b LL |
2195 | /* uncompress() will return failed in some case, especially |
2196 | * when the page is dirted when doing the compression, it's | |
2197 | * not a problem because the dirty page will be retransferred | |
2198 | * and uncompress() won't break the data in other pages. | |
2199 | */ | |
33d151f4 LL |
2200 | uncompress((Bytef *)des, &pagesize, |
2201 | (const Bytef *)param->compbuf, len); | |
73a8912b | 2202 | |
33d151f4 LL |
2203 | qemu_mutex_lock(&decomp_done_lock); |
2204 | param->done = true; | |
2205 | qemu_cond_signal(&decomp_done_cond); | |
2206 | qemu_mutex_unlock(&decomp_done_lock); | |
2207 | ||
2208 | qemu_mutex_lock(¶m->mutex); | |
2209 | } else { | |
2210 | qemu_cond_wait(¶m->cond, ¶m->mutex); | |
2211 | } | |
56e93d26 | 2212 | } |
33d151f4 | 2213 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
2214 | |
2215 | return NULL; | |
2216 | } | |
2217 | ||
5533b2e9 LL |
2218 | static void wait_for_decompress_done(void) |
2219 | { | |
2220 | int idx, thread_count; | |
2221 | ||
2222 | if (!migrate_use_compression()) { | |
2223 | return; | |
2224 | } | |
2225 | ||
2226 | thread_count = migrate_decompress_threads(); | |
2227 | qemu_mutex_lock(&decomp_done_lock); | |
2228 | for (idx = 0; idx < thread_count; idx++) { | |
2229 | while (!decomp_param[idx].done) { | |
2230 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
2231 | } | |
2232 | } | |
2233 | qemu_mutex_unlock(&decomp_done_lock); | |
2234 | } | |
2235 | ||
56e93d26 JQ |
2236 | void migrate_decompress_threads_create(void) |
2237 | { | |
2238 | int i, thread_count; | |
2239 | ||
2240 | thread_count = migrate_decompress_threads(); | |
2241 | decompress_threads = g_new0(QemuThread, thread_count); | |
2242 | decomp_param = g_new0(DecompressParam, thread_count); | |
73a8912b LL |
2243 | qemu_mutex_init(&decomp_done_lock); |
2244 | qemu_cond_init(&decomp_done_cond); | |
56e93d26 JQ |
2245 | for (i = 0; i < thread_count; i++) { |
2246 | qemu_mutex_init(&decomp_param[i].mutex); | |
2247 | qemu_cond_init(&decomp_param[i].cond); | |
2248 | decomp_param[i].compbuf = g_malloc0(compressBound(TARGET_PAGE_SIZE)); | |
73a8912b | 2249 | decomp_param[i].done = true; |
90e56fb4 | 2250 | decomp_param[i].quit = false; |
56e93d26 JQ |
2251 | qemu_thread_create(decompress_threads + i, "decompress", |
2252 | do_data_decompress, decomp_param + i, | |
2253 | QEMU_THREAD_JOINABLE); | |
2254 | } | |
2255 | } | |
2256 | ||
2257 | void migrate_decompress_threads_join(void) | |
2258 | { | |
2259 | int i, thread_count; | |
2260 | ||
56e93d26 JQ |
2261 | thread_count = migrate_decompress_threads(); |
2262 | for (i = 0; i < thread_count; i++) { | |
2263 | qemu_mutex_lock(&decomp_param[i].mutex); | |
90e56fb4 | 2264 | decomp_param[i].quit = true; |
56e93d26 JQ |
2265 | qemu_cond_signal(&decomp_param[i].cond); |
2266 | qemu_mutex_unlock(&decomp_param[i].mutex); | |
2267 | } | |
2268 | for (i = 0; i < thread_count; i++) { | |
2269 | qemu_thread_join(decompress_threads + i); | |
2270 | qemu_mutex_destroy(&decomp_param[i].mutex); | |
2271 | qemu_cond_destroy(&decomp_param[i].cond); | |
2272 | g_free(decomp_param[i].compbuf); | |
2273 | } | |
2274 | g_free(decompress_threads); | |
2275 | g_free(decomp_param); | |
56e93d26 JQ |
2276 | decompress_threads = NULL; |
2277 | decomp_param = NULL; | |
56e93d26 JQ |
2278 | } |
2279 | ||
c1bc6626 | 2280 | static void decompress_data_with_multi_threads(QEMUFile *f, |
56e93d26 JQ |
2281 | void *host, int len) |
2282 | { | |
2283 | int idx, thread_count; | |
2284 | ||
2285 | thread_count = migrate_decompress_threads(); | |
73a8912b | 2286 | qemu_mutex_lock(&decomp_done_lock); |
56e93d26 JQ |
2287 | while (true) { |
2288 | for (idx = 0; idx < thread_count; idx++) { | |
73a8912b | 2289 | if (decomp_param[idx].done) { |
33d151f4 LL |
2290 | decomp_param[idx].done = false; |
2291 | qemu_mutex_lock(&decomp_param[idx].mutex); | |
c1bc6626 | 2292 | qemu_get_buffer(f, decomp_param[idx].compbuf, len); |
56e93d26 JQ |
2293 | decomp_param[idx].des = host; |
2294 | decomp_param[idx].len = len; | |
33d151f4 LL |
2295 | qemu_cond_signal(&decomp_param[idx].cond); |
2296 | qemu_mutex_unlock(&decomp_param[idx].mutex); | |
56e93d26 JQ |
2297 | break; |
2298 | } | |
2299 | } | |
2300 | if (idx < thread_count) { | |
2301 | break; | |
73a8912b LL |
2302 | } else { |
2303 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
56e93d26 JQ |
2304 | } |
2305 | } | |
73a8912b | 2306 | qemu_mutex_unlock(&decomp_done_lock); |
56e93d26 JQ |
2307 | } |
2308 | ||
3d0684b2 JQ |
2309 | /** |
2310 | * ram_postcopy_incoming_init: allocate postcopy data structures | |
2311 | * | |
2312 | * Returns 0 for success and negative if there was one error | |
2313 | * | |
2314 | * @mis: current migration incoming state | |
2315 | * | |
2316 | * Allocate data structures etc needed by incoming migration with | |
2317 | * postcopy-ram. postcopy-ram's similarly names | |
2318 | * postcopy_ram_incoming_init does the work. | |
1caddf8a DDAG |
2319 | */ |
2320 | int ram_postcopy_incoming_init(MigrationIncomingState *mis) | |
2321 | { | |
b8c48993 | 2322 | unsigned long ram_pages = last_ram_page(); |
1caddf8a DDAG |
2323 | |
2324 | return postcopy_ram_incoming_init(mis, ram_pages); | |
2325 | } | |
2326 | ||
3d0684b2 JQ |
2327 | /** |
2328 | * ram_load_postcopy: load a page in postcopy case | |
2329 | * | |
2330 | * Returns 0 for success or -errno in case of error | |
2331 | * | |
a7180877 DDAG |
2332 | * Called in postcopy mode by ram_load(). |
2333 | * rcu_read_lock is taken prior to this being called. | |
3d0684b2 JQ |
2334 | * |
2335 | * @f: QEMUFile where to send the data | |
a7180877 DDAG |
2336 | */ |
2337 | static int ram_load_postcopy(QEMUFile *f) | |
2338 | { | |
2339 | int flags = 0, ret = 0; | |
2340 | bool place_needed = false; | |
28abd200 | 2341 | bool matching_page_sizes = false; |
a7180877 DDAG |
2342 | MigrationIncomingState *mis = migration_incoming_get_current(); |
2343 | /* Temporary page that is later 'placed' */ | |
2344 | void *postcopy_host_page = postcopy_get_tmp_page(mis); | |
c53b7ddc | 2345 | void *last_host = NULL; |
a3b6ff6d | 2346 | bool all_zero = false; |
a7180877 DDAG |
2347 | |
2348 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
2349 | ram_addr_t addr; | |
2350 | void *host = NULL; | |
2351 | void *page_buffer = NULL; | |
2352 | void *place_source = NULL; | |
df9ff5e1 | 2353 | RAMBlock *block = NULL; |
a7180877 | 2354 | uint8_t ch; |
a7180877 DDAG |
2355 | |
2356 | addr = qemu_get_be64(f); | |
2357 | flags = addr & ~TARGET_PAGE_MASK; | |
2358 | addr &= TARGET_PAGE_MASK; | |
2359 | ||
2360 | trace_ram_load_postcopy_loop((uint64_t)addr, flags); | |
2361 | place_needed = false; | |
bb890ed5 | 2362 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE)) { |
df9ff5e1 | 2363 | block = ram_block_from_stream(f, flags); |
4c4bad48 HZ |
2364 | |
2365 | host = host_from_ram_block_offset(block, addr); | |
a7180877 DDAG |
2366 | if (!host) { |
2367 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
2368 | ret = -EINVAL; | |
2369 | break; | |
2370 | } | |
28abd200 | 2371 | matching_page_sizes = block->page_size == TARGET_PAGE_SIZE; |
a7180877 | 2372 | /* |
28abd200 DDAG |
2373 | * Postcopy requires that we place whole host pages atomically; |
2374 | * these may be huge pages for RAMBlocks that are backed by | |
2375 | * hugetlbfs. | |
a7180877 DDAG |
2376 | * To make it atomic, the data is read into a temporary page |
2377 | * that's moved into place later. | |
2378 | * The migration protocol uses, possibly smaller, target-pages | |
2379 | * however the source ensures it always sends all the components | |
2380 | * of a host page in order. | |
2381 | */ | |
2382 | page_buffer = postcopy_host_page + | |
28abd200 | 2383 | ((uintptr_t)host & (block->page_size - 1)); |
a7180877 | 2384 | /* If all TP are zero then we can optimise the place */ |
28abd200 | 2385 | if (!((uintptr_t)host & (block->page_size - 1))) { |
a7180877 | 2386 | all_zero = true; |
c53b7ddc DDAG |
2387 | } else { |
2388 | /* not the 1st TP within the HP */ | |
2389 | if (host != (last_host + TARGET_PAGE_SIZE)) { | |
9af9e0fe | 2390 | error_report("Non-sequential target page %p/%p", |
c53b7ddc DDAG |
2391 | host, last_host); |
2392 | ret = -EINVAL; | |
2393 | break; | |
2394 | } | |
a7180877 DDAG |
2395 | } |
2396 | ||
c53b7ddc | 2397 | |
a7180877 DDAG |
2398 | /* |
2399 | * If it's the last part of a host page then we place the host | |
2400 | * page | |
2401 | */ | |
2402 | place_needed = (((uintptr_t)host + TARGET_PAGE_SIZE) & | |
28abd200 | 2403 | (block->page_size - 1)) == 0; |
a7180877 DDAG |
2404 | place_source = postcopy_host_page; |
2405 | } | |
c53b7ddc | 2406 | last_host = host; |
a7180877 DDAG |
2407 | |
2408 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { | |
bb890ed5 | 2409 | case RAM_SAVE_FLAG_ZERO: |
a7180877 DDAG |
2410 | ch = qemu_get_byte(f); |
2411 | memset(page_buffer, ch, TARGET_PAGE_SIZE); | |
2412 | if (ch) { | |
2413 | all_zero = false; | |
2414 | } | |
2415 | break; | |
2416 | ||
2417 | case RAM_SAVE_FLAG_PAGE: | |
2418 | all_zero = false; | |
2419 | if (!place_needed || !matching_page_sizes) { | |
2420 | qemu_get_buffer(f, page_buffer, TARGET_PAGE_SIZE); | |
2421 | } else { | |
2422 | /* Avoids the qemu_file copy during postcopy, which is | |
2423 | * going to do a copy later; can only do it when we | |
2424 | * do this read in one go (matching page sizes) | |
2425 | */ | |
2426 | qemu_get_buffer_in_place(f, (uint8_t **)&place_source, | |
2427 | TARGET_PAGE_SIZE); | |
2428 | } | |
2429 | break; | |
2430 | case RAM_SAVE_FLAG_EOS: | |
2431 | /* normal exit */ | |
2432 | break; | |
2433 | default: | |
2434 | error_report("Unknown combination of migration flags: %#x" | |
2435 | " (postcopy mode)", flags); | |
2436 | ret = -EINVAL; | |
2437 | } | |
2438 | ||
2439 | if (place_needed) { | |
2440 | /* This gets called at the last target page in the host page */ | |
df9ff5e1 DDAG |
2441 | void *place_dest = host + TARGET_PAGE_SIZE - block->page_size; |
2442 | ||
a7180877 | 2443 | if (all_zero) { |
df9ff5e1 DDAG |
2444 | ret = postcopy_place_page_zero(mis, place_dest, |
2445 | block->page_size); | |
a7180877 | 2446 | } else { |
df9ff5e1 DDAG |
2447 | ret = postcopy_place_page(mis, place_dest, |
2448 | place_source, block->page_size); | |
a7180877 DDAG |
2449 | } |
2450 | } | |
2451 | if (!ret) { | |
2452 | ret = qemu_file_get_error(f); | |
2453 | } | |
2454 | } | |
2455 | ||
2456 | return ret; | |
2457 | } | |
2458 | ||
56e93d26 JQ |
2459 | static int ram_load(QEMUFile *f, void *opaque, int version_id) |
2460 | { | |
2461 | int flags = 0, ret = 0; | |
2462 | static uint64_t seq_iter; | |
2463 | int len = 0; | |
a7180877 DDAG |
2464 | /* |
2465 | * If system is running in postcopy mode, page inserts to host memory must | |
2466 | * be atomic | |
2467 | */ | |
2468 | bool postcopy_running = postcopy_state_get() >= POSTCOPY_INCOMING_LISTENING; | |
ef08fb38 DDAG |
2469 | /* ADVISE is earlier, it shows the source has the postcopy capability on */ |
2470 | bool postcopy_advised = postcopy_state_get() >= POSTCOPY_INCOMING_ADVISE; | |
56e93d26 JQ |
2471 | |
2472 | seq_iter++; | |
2473 | ||
2474 | if (version_id != 4) { | |
2475 | ret = -EINVAL; | |
2476 | } | |
2477 | ||
2478 | /* This RCU critical section can be very long running. | |
2479 | * When RCU reclaims in the code start to become numerous, | |
2480 | * it will be necessary to reduce the granularity of this | |
2481 | * critical section. | |
2482 | */ | |
2483 | rcu_read_lock(); | |
a7180877 DDAG |
2484 | |
2485 | if (postcopy_running) { | |
2486 | ret = ram_load_postcopy(f); | |
2487 | } | |
2488 | ||
2489 | while (!postcopy_running && !ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
56e93d26 | 2490 | ram_addr_t addr, total_ram_bytes; |
a776aa15 | 2491 | void *host = NULL; |
56e93d26 JQ |
2492 | uint8_t ch; |
2493 | ||
2494 | addr = qemu_get_be64(f); | |
2495 | flags = addr & ~TARGET_PAGE_MASK; | |
2496 | addr &= TARGET_PAGE_MASK; | |
2497 | ||
bb890ed5 | 2498 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
a776aa15 | 2499 | RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) { |
4c4bad48 HZ |
2500 | RAMBlock *block = ram_block_from_stream(f, flags); |
2501 | ||
2502 | host = host_from_ram_block_offset(block, addr); | |
a776aa15 DDAG |
2503 | if (!host) { |
2504 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
2505 | ret = -EINVAL; | |
2506 | break; | |
2507 | } | |
1db9d8e5 | 2508 | trace_ram_load_loop(block->idstr, (uint64_t)addr, flags, host); |
a776aa15 DDAG |
2509 | } |
2510 | ||
56e93d26 JQ |
2511 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { |
2512 | case RAM_SAVE_FLAG_MEM_SIZE: | |
2513 | /* Synchronize RAM block list */ | |
2514 | total_ram_bytes = addr; | |
2515 | while (!ret && total_ram_bytes) { | |
2516 | RAMBlock *block; | |
56e93d26 JQ |
2517 | char id[256]; |
2518 | ram_addr_t length; | |
2519 | ||
2520 | len = qemu_get_byte(f); | |
2521 | qemu_get_buffer(f, (uint8_t *)id, len); | |
2522 | id[len] = 0; | |
2523 | length = qemu_get_be64(f); | |
2524 | ||
e3dd7493 DDAG |
2525 | block = qemu_ram_block_by_name(id); |
2526 | if (block) { | |
2527 | if (length != block->used_length) { | |
2528 | Error *local_err = NULL; | |
56e93d26 | 2529 | |
fa53a0e5 | 2530 | ret = qemu_ram_resize(block, length, |
e3dd7493 DDAG |
2531 | &local_err); |
2532 | if (local_err) { | |
2533 | error_report_err(local_err); | |
56e93d26 | 2534 | } |
56e93d26 | 2535 | } |
ef08fb38 DDAG |
2536 | /* For postcopy we need to check hugepage sizes match */ |
2537 | if (postcopy_advised && | |
2538 | block->page_size != qemu_host_page_size) { | |
2539 | uint64_t remote_page_size = qemu_get_be64(f); | |
2540 | if (remote_page_size != block->page_size) { | |
2541 | error_report("Mismatched RAM page size %s " | |
2542 | "(local) %zd != %" PRId64, | |
2543 | id, block->page_size, | |
2544 | remote_page_size); | |
2545 | ret = -EINVAL; | |
2546 | } | |
2547 | } | |
e3dd7493 DDAG |
2548 | ram_control_load_hook(f, RAM_CONTROL_BLOCK_REG, |
2549 | block->idstr); | |
2550 | } else { | |
56e93d26 JQ |
2551 | error_report("Unknown ramblock \"%s\", cannot " |
2552 | "accept migration", id); | |
2553 | ret = -EINVAL; | |
2554 | } | |
2555 | ||
2556 | total_ram_bytes -= length; | |
2557 | } | |
2558 | break; | |
a776aa15 | 2559 | |
bb890ed5 | 2560 | case RAM_SAVE_FLAG_ZERO: |
56e93d26 JQ |
2561 | ch = qemu_get_byte(f); |
2562 | ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); | |
2563 | break; | |
a776aa15 | 2564 | |
56e93d26 | 2565 | case RAM_SAVE_FLAG_PAGE: |
56e93d26 JQ |
2566 | qemu_get_buffer(f, host, TARGET_PAGE_SIZE); |
2567 | break; | |
56e93d26 | 2568 | |
a776aa15 | 2569 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
56e93d26 JQ |
2570 | len = qemu_get_be32(f); |
2571 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
2572 | error_report("Invalid compressed data length: %d", len); | |
2573 | ret = -EINVAL; | |
2574 | break; | |
2575 | } | |
c1bc6626 | 2576 | decompress_data_with_multi_threads(f, host, len); |
56e93d26 | 2577 | break; |
a776aa15 | 2578 | |
56e93d26 | 2579 | case RAM_SAVE_FLAG_XBZRLE: |
56e93d26 JQ |
2580 | if (load_xbzrle(f, addr, host) < 0) { |
2581 | error_report("Failed to decompress XBZRLE page at " | |
2582 | RAM_ADDR_FMT, addr); | |
2583 | ret = -EINVAL; | |
2584 | break; | |
2585 | } | |
2586 | break; | |
2587 | case RAM_SAVE_FLAG_EOS: | |
2588 | /* normal exit */ | |
2589 | break; | |
2590 | default: | |
2591 | if (flags & RAM_SAVE_FLAG_HOOK) { | |
632e3a5c | 2592 | ram_control_load_hook(f, RAM_CONTROL_HOOK, NULL); |
56e93d26 JQ |
2593 | } else { |
2594 | error_report("Unknown combination of migration flags: %#x", | |
2595 | flags); | |
2596 | ret = -EINVAL; | |
2597 | } | |
2598 | } | |
2599 | if (!ret) { | |
2600 | ret = qemu_file_get_error(f); | |
2601 | } | |
2602 | } | |
2603 | ||
5533b2e9 | 2604 | wait_for_decompress_done(); |
56e93d26 | 2605 | rcu_read_unlock(); |
55c4446b | 2606 | trace_ram_load_complete(ret, seq_iter); |
56e93d26 JQ |
2607 | return ret; |
2608 | } | |
2609 | ||
2610 | static SaveVMHandlers savevm_ram_handlers = { | |
2611 | .save_live_setup = ram_save_setup, | |
2612 | .save_live_iterate = ram_save_iterate, | |
763c906b | 2613 | .save_live_complete_postcopy = ram_save_complete, |
a3e06c3d | 2614 | .save_live_complete_precopy = ram_save_complete, |
56e93d26 JQ |
2615 | .save_live_pending = ram_save_pending, |
2616 | .load_state = ram_load, | |
6ad2a215 | 2617 | .cleanup = ram_migration_cleanup, |
56e93d26 JQ |
2618 | }; |
2619 | ||
2620 | void ram_mig_init(void) | |
2621 | { | |
2622 | qemu_mutex_init(&XBZRLE.lock); | |
6f37bb8b | 2623 | register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, &ram_state); |
56e93d26 | 2624 | } |