]>
Commit | Line | Data |
---|---|---|
56e93d26 JQ |
1 | /* |
2 | * QEMU System Emulator | |
3 | * | |
4 | * Copyright (c) 2003-2008 Fabrice Bellard | |
76cc7b58 JQ |
5 | * Copyright (c) 2011-2015 Red Hat Inc |
6 | * | |
7 | * Authors: | |
8 | * Juan Quintela <quintela@redhat.com> | |
56e93d26 JQ |
9 | * |
10 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
11 | * of this software and associated documentation files (the "Software"), to deal | |
12 | * in the Software without restriction, including without limitation the rights | |
13 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
14 | * copies of the Software, and to permit persons to whom the Software is | |
15 | * furnished to do so, subject to the following conditions: | |
16 | * | |
17 | * The above copyright notice and this permission notice shall be included in | |
18 | * all copies or substantial portions of the Software. | |
19 | * | |
20 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
21 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
22 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
23 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
24 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
25 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
26 | * THE SOFTWARE. | |
27 | */ | |
e688df6b | 28 | |
1393a485 | 29 | #include "qemu/osdep.h" |
33c11879 | 30 | #include "cpu.h" |
56e93d26 | 31 | #include <zlib.h> |
f348b6d1 | 32 | #include "qemu/cutils.h" |
56e93d26 JQ |
33 | #include "qemu/bitops.h" |
34 | #include "qemu/bitmap.h" | |
7205c9ec | 35 | #include "qemu/main-loop.h" |
56eb90af | 36 | #include "qemu/pmem.h" |
709e3fe8 | 37 | #include "xbzrle.h" |
7b1e1a22 | 38 | #include "ram.h" |
6666c96a | 39 | #include "migration.h" |
71bb07db | 40 | #include "socket.h" |
f2a8f0a6 | 41 | #include "migration/register.h" |
7b1e1a22 | 42 | #include "migration/misc.h" |
08a0aee1 | 43 | #include "qemu-file.h" |
be07b0ac | 44 | #include "postcopy-ram.h" |
53d37d36 | 45 | #include "page_cache.h" |
56e93d26 | 46 | #include "qemu/error-report.h" |
e688df6b | 47 | #include "qapi/error.h" |
9af23989 | 48 | #include "qapi/qapi-events-migration.h" |
8acabf69 | 49 | #include "qapi/qmp/qerror.h" |
56e93d26 | 50 | #include "trace.h" |
56e93d26 | 51 | #include "exec/ram_addr.h" |
f9494614 | 52 | #include "exec/target_page.h" |
56e93d26 | 53 | #include "qemu/rcu_queue.h" |
a91246c9 | 54 | #include "migration/colo.h" |
53d37d36 | 55 | #include "block.h" |
af8b7d2b JQ |
56 | #include "sysemu/sysemu.h" |
57 | #include "qemu/uuid.h" | |
edd090c7 | 58 | #include "savevm.h" |
b9ee2f7d | 59 | #include "qemu/iov.h" |
56e93d26 | 60 | |
56e93d26 JQ |
61 | /***********************************************************/ |
62 | /* ram save/restore */ | |
63 | ||
bb890ed5 JQ |
64 | /* RAM_SAVE_FLAG_ZERO used to be named RAM_SAVE_FLAG_COMPRESS, it |
65 | * worked for pages that where filled with the same char. We switched | |
66 | * it to only search for the zero value. And to avoid confusion with | |
67 | * RAM_SSAVE_FLAG_COMPRESS_PAGE just rename it. | |
68 | */ | |
69 | ||
56e93d26 | 70 | #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */ |
bb890ed5 | 71 | #define RAM_SAVE_FLAG_ZERO 0x02 |
56e93d26 JQ |
72 | #define RAM_SAVE_FLAG_MEM_SIZE 0x04 |
73 | #define RAM_SAVE_FLAG_PAGE 0x08 | |
74 | #define RAM_SAVE_FLAG_EOS 0x10 | |
75 | #define RAM_SAVE_FLAG_CONTINUE 0x20 | |
76 | #define RAM_SAVE_FLAG_XBZRLE 0x40 | |
77 | /* 0x80 is reserved in migration.h start with 0x100 next */ | |
78 | #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 | |
79 | ||
56e93d26 JQ |
80 | static inline bool is_zero_range(uint8_t *p, uint64_t size) |
81 | { | |
a1febc49 | 82 | return buffer_is_zero(p, size); |
56e93d26 JQ |
83 | } |
84 | ||
9360447d JQ |
85 | XBZRLECacheStats xbzrle_counters; |
86 | ||
56e93d26 JQ |
87 | /* struct contains XBZRLE cache and a static page |
88 | used by the compression */ | |
89 | static struct { | |
90 | /* buffer used for XBZRLE encoding */ | |
91 | uint8_t *encoded_buf; | |
92 | /* buffer for storing page content */ | |
93 | uint8_t *current_buf; | |
94 | /* Cache for XBZRLE, Protected by lock. */ | |
95 | PageCache *cache; | |
96 | QemuMutex lock; | |
c00e0928 JQ |
97 | /* it will store a page full of zeros */ |
98 | uint8_t *zero_target_page; | |
f265e0e4 JQ |
99 | /* buffer used for XBZRLE decoding */ |
100 | uint8_t *decoded_buf; | |
56e93d26 JQ |
101 | } XBZRLE; |
102 | ||
56e93d26 JQ |
103 | static void XBZRLE_cache_lock(void) |
104 | { | |
105 | if (migrate_use_xbzrle()) | |
106 | qemu_mutex_lock(&XBZRLE.lock); | |
107 | } | |
108 | ||
109 | static void XBZRLE_cache_unlock(void) | |
110 | { | |
111 | if (migrate_use_xbzrle()) | |
112 | qemu_mutex_unlock(&XBZRLE.lock); | |
113 | } | |
114 | ||
3d0684b2 JQ |
115 | /** |
116 | * xbzrle_cache_resize: resize the xbzrle cache | |
117 | * | |
118 | * This function is called from qmp_migrate_set_cache_size in main | |
119 | * thread, possibly while a migration is in progress. A running | |
120 | * migration may be using the cache and might finish during this call, | |
121 | * hence changes to the cache are protected by XBZRLE.lock(). | |
122 | * | |
c9dede2d | 123 | * Returns 0 for success or -1 for error |
3d0684b2 JQ |
124 | * |
125 | * @new_size: new cache size | |
8acabf69 | 126 | * @errp: set *errp if the check failed, with reason |
56e93d26 | 127 | */ |
c9dede2d | 128 | int xbzrle_cache_resize(int64_t new_size, Error **errp) |
56e93d26 JQ |
129 | { |
130 | PageCache *new_cache; | |
c9dede2d | 131 | int64_t ret = 0; |
56e93d26 | 132 | |
8acabf69 JQ |
133 | /* Check for truncation */ |
134 | if (new_size != (size_t)new_size) { | |
135 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", | |
136 | "exceeding address space"); | |
137 | return -1; | |
138 | } | |
139 | ||
2a313e5c JQ |
140 | if (new_size == migrate_xbzrle_cache_size()) { |
141 | /* nothing to do */ | |
c9dede2d | 142 | return 0; |
2a313e5c JQ |
143 | } |
144 | ||
56e93d26 JQ |
145 | XBZRLE_cache_lock(); |
146 | ||
147 | if (XBZRLE.cache != NULL) { | |
80f8dfde | 148 | new_cache = cache_init(new_size, TARGET_PAGE_SIZE, errp); |
56e93d26 | 149 | if (!new_cache) { |
56e93d26 JQ |
150 | ret = -1; |
151 | goto out; | |
152 | } | |
153 | ||
154 | cache_fini(XBZRLE.cache); | |
155 | XBZRLE.cache = new_cache; | |
156 | } | |
56e93d26 JQ |
157 | out: |
158 | XBZRLE_cache_unlock(); | |
159 | return ret; | |
160 | } | |
161 | ||
fbd162e6 YK |
162 | static bool ramblock_is_ignored(RAMBlock *block) |
163 | { | |
164 | return !qemu_ram_is_migratable(block) || | |
165 | (migrate_ignore_shared() && qemu_ram_is_shared(block)); | |
166 | } | |
167 | ||
b895de50 | 168 | /* Should be holding either ram_list.mutex, or the RCU lock. */ |
fbd162e6 YK |
169 | #define RAMBLOCK_FOREACH_NOT_IGNORED(block) \ |
170 | INTERNAL_RAMBLOCK_FOREACH(block) \ | |
171 | if (ramblock_is_ignored(block)) {} else | |
172 | ||
b895de50 | 173 | #define RAMBLOCK_FOREACH_MIGRATABLE(block) \ |
343f632c | 174 | INTERNAL_RAMBLOCK_FOREACH(block) \ |
b895de50 CLG |
175 | if (!qemu_ram_is_migratable(block)) {} else |
176 | ||
343f632c DDAG |
177 | #undef RAMBLOCK_FOREACH |
178 | ||
fbd162e6 YK |
179 | int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque) |
180 | { | |
181 | RAMBlock *block; | |
182 | int ret = 0; | |
183 | ||
184 | rcu_read_lock(); | |
185 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
186 | ret = func(block, opaque); | |
187 | if (ret) { | |
188 | break; | |
189 | } | |
190 | } | |
191 | rcu_read_unlock(); | |
192 | return ret; | |
193 | } | |
194 | ||
f9494614 AP |
195 | static void ramblock_recv_map_init(void) |
196 | { | |
197 | RAMBlock *rb; | |
198 | ||
fbd162e6 | 199 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
f9494614 AP |
200 | assert(!rb->receivedmap); |
201 | rb->receivedmap = bitmap_new(rb->max_length >> qemu_target_page_bits()); | |
202 | } | |
203 | } | |
204 | ||
205 | int ramblock_recv_bitmap_test(RAMBlock *rb, void *host_addr) | |
206 | { | |
207 | return test_bit(ramblock_recv_bitmap_offset(host_addr, rb), | |
208 | rb->receivedmap); | |
209 | } | |
210 | ||
1cba9f6e DDAG |
211 | bool ramblock_recv_bitmap_test_byte_offset(RAMBlock *rb, uint64_t byte_offset) |
212 | { | |
213 | return test_bit(byte_offset >> TARGET_PAGE_BITS, rb->receivedmap); | |
214 | } | |
215 | ||
f9494614 AP |
216 | void ramblock_recv_bitmap_set(RAMBlock *rb, void *host_addr) |
217 | { | |
218 | set_bit_atomic(ramblock_recv_bitmap_offset(host_addr, rb), rb->receivedmap); | |
219 | } | |
220 | ||
221 | void ramblock_recv_bitmap_set_range(RAMBlock *rb, void *host_addr, | |
222 | size_t nr) | |
223 | { | |
224 | bitmap_set_atomic(rb->receivedmap, | |
225 | ramblock_recv_bitmap_offset(host_addr, rb), | |
226 | nr); | |
227 | } | |
228 | ||
a335debb PX |
229 | #define RAMBLOCK_RECV_BITMAP_ENDING (0x0123456789abcdefULL) |
230 | ||
231 | /* | |
232 | * Format: bitmap_size (8 bytes) + whole_bitmap (N bytes). | |
233 | * | |
234 | * Returns >0 if success with sent bytes, or <0 if error. | |
235 | */ | |
236 | int64_t ramblock_recv_bitmap_send(QEMUFile *file, | |
237 | const char *block_name) | |
238 | { | |
239 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
240 | unsigned long *le_bitmap, nbits; | |
241 | uint64_t size; | |
242 | ||
243 | if (!block) { | |
244 | error_report("%s: invalid block name: %s", __func__, block_name); | |
245 | return -1; | |
246 | } | |
247 | ||
248 | nbits = block->used_length >> TARGET_PAGE_BITS; | |
249 | ||
250 | /* | |
251 | * Make sure the tmp bitmap buffer is big enough, e.g., on 32bit | |
252 | * machines we may need 4 more bytes for padding (see below | |
253 | * comment). So extend it a bit before hand. | |
254 | */ | |
255 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
256 | ||
257 | /* | |
258 | * Always use little endian when sending the bitmap. This is | |
259 | * required that when source and destination VMs are not using the | |
260 | * same endianess. (Note: big endian won't work.) | |
261 | */ | |
262 | bitmap_to_le(le_bitmap, block->receivedmap, nbits); | |
263 | ||
264 | /* Size of the bitmap, in bytes */ | |
a725ef9f | 265 | size = DIV_ROUND_UP(nbits, 8); |
a335debb PX |
266 | |
267 | /* | |
268 | * size is always aligned to 8 bytes for 64bit machines, but it | |
269 | * may not be true for 32bit machines. We need this padding to | |
270 | * make sure the migration can survive even between 32bit and | |
271 | * 64bit machines. | |
272 | */ | |
273 | size = ROUND_UP(size, 8); | |
274 | ||
275 | qemu_put_be64(file, size); | |
276 | qemu_put_buffer(file, (const uint8_t *)le_bitmap, size); | |
277 | /* | |
278 | * Mark as an end, in case the middle part is screwed up due to | |
279 | * some "misterious" reason. | |
280 | */ | |
281 | qemu_put_be64(file, RAMBLOCK_RECV_BITMAP_ENDING); | |
282 | qemu_fflush(file); | |
283 | ||
bf269906 | 284 | g_free(le_bitmap); |
a335debb PX |
285 | |
286 | if (qemu_file_get_error(file)) { | |
287 | return qemu_file_get_error(file); | |
288 | } | |
289 | ||
290 | return size + sizeof(size); | |
291 | } | |
292 | ||
ec481c6c JQ |
293 | /* |
294 | * An outstanding page request, on the source, having been received | |
295 | * and queued | |
296 | */ | |
297 | struct RAMSrcPageRequest { | |
298 | RAMBlock *rb; | |
299 | hwaddr offset; | |
300 | hwaddr len; | |
301 | ||
302 | QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; | |
303 | }; | |
304 | ||
6f37bb8b JQ |
305 | /* State of RAM for migration */ |
306 | struct RAMState { | |
204b88b8 JQ |
307 | /* QEMUFile used for this migration */ |
308 | QEMUFile *f; | |
6f37bb8b JQ |
309 | /* Last block that we have visited searching for dirty pages */ |
310 | RAMBlock *last_seen_block; | |
311 | /* Last block from where we have sent data */ | |
312 | RAMBlock *last_sent_block; | |
269ace29 JQ |
313 | /* Last dirty target page we have sent */ |
314 | ram_addr_t last_page; | |
6f37bb8b JQ |
315 | /* last ram version we have seen */ |
316 | uint32_t last_version; | |
317 | /* We are in the first round */ | |
318 | bool ram_bulk_stage; | |
6eeb63f7 WW |
319 | /* The free page optimization is enabled */ |
320 | bool fpo_enabled; | |
8d820d6f JQ |
321 | /* How many times we have dirty too many pages */ |
322 | int dirty_rate_high_cnt; | |
f664da80 JQ |
323 | /* these variables are used for bitmap sync */ |
324 | /* last time we did a full bitmap_sync */ | |
325 | int64_t time_last_bitmap_sync; | |
eac74159 | 326 | /* bytes transferred at start_time */ |
c4bdf0cf | 327 | uint64_t bytes_xfer_prev; |
a66cd90c | 328 | /* number of dirty pages since start_time */ |
68908ed6 | 329 | uint64_t num_dirty_pages_period; |
b5833fde JQ |
330 | /* xbzrle misses since the beginning of the period */ |
331 | uint64_t xbzrle_cache_miss_prev; | |
76e03000 XG |
332 | |
333 | /* compression statistics since the beginning of the period */ | |
334 | /* amount of count that no free thread to compress data */ | |
335 | uint64_t compress_thread_busy_prev; | |
336 | /* amount bytes after compression */ | |
337 | uint64_t compressed_size_prev; | |
338 | /* amount of compressed pages */ | |
339 | uint64_t compress_pages_prev; | |
340 | ||
be8b02ed XG |
341 | /* total handled target pages at the beginning of period */ |
342 | uint64_t target_page_count_prev; | |
343 | /* total handled target pages since start */ | |
344 | uint64_t target_page_count; | |
9360447d | 345 | /* number of dirty bits in the bitmap */ |
2dfaf12e | 346 | uint64_t migration_dirty_pages; |
386a907b | 347 | /* Protects modification of the bitmap and migration dirty pages */ |
108cfae0 | 348 | QemuMutex bitmap_mutex; |
68a098f3 JQ |
349 | /* The RAMBlock used in the last src_page_requests */ |
350 | RAMBlock *last_req_rb; | |
ec481c6c JQ |
351 | /* Queue of outstanding page requests from the destination */ |
352 | QemuMutex src_page_req_mutex; | |
b58deb34 | 353 | QSIMPLEQ_HEAD(, RAMSrcPageRequest) src_page_requests; |
6f37bb8b JQ |
354 | }; |
355 | typedef struct RAMState RAMState; | |
356 | ||
53518d94 | 357 | static RAMState *ram_state; |
6f37bb8b | 358 | |
bd227060 WW |
359 | static NotifierWithReturnList precopy_notifier_list; |
360 | ||
361 | void precopy_infrastructure_init(void) | |
362 | { | |
363 | notifier_with_return_list_init(&precopy_notifier_list); | |
364 | } | |
365 | ||
366 | void precopy_add_notifier(NotifierWithReturn *n) | |
367 | { | |
368 | notifier_with_return_list_add(&precopy_notifier_list, n); | |
369 | } | |
370 | ||
371 | void precopy_remove_notifier(NotifierWithReturn *n) | |
372 | { | |
373 | notifier_with_return_remove(n); | |
374 | } | |
375 | ||
376 | int precopy_notify(PrecopyNotifyReason reason, Error **errp) | |
377 | { | |
378 | PrecopyNotifyData pnd; | |
379 | pnd.reason = reason; | |
380 | pnd.errp = errp; | |
381 | ||
382 | return notifier_with_return_list_notify(&precopy_notifier_list, &pnd); | |
383 | } | |
384 | ||
6eeb63f7 WW |
385 | void precopy_enable_free_page_optimization(void) |
386 | { | |
387 | if (!ram_state) { | |
388 | return; | |
389 | } | |
390 | ||
391 | ram_state->fpo_enabled = true; | |
392 | } | |
393 | ||
9edabd4d | 394 | uint64_t ram_bytes_remaining(void) |
2f4fde93 | 395 | { |
bae416e5 DDAG |
396 | return ram_state ? (ram_state->migration_dirty_pages * TARGET_PAGE_SIZE) : |
397 | 0; | |
2f4fde93 JQ |
398 | } |
399 | ||
9360447d | 400 | MigrationStats ram_counters; |
96506894 | 401 | |
b8fb8cb7 DDAG |
402 | /* used by the search for pages to send */ |
403 | struct PageSearchStatus { | |
404 | /* Current block being searched */ | |
405 | RAMBlock *block; | |
a935e30f JQ |
406 | /* Current page to search from */ |
407 | unsigned long page; | |
b8fb8cb7 DDAG |
408 | /* Set once we wrap around */ |
409 | bool complete_round; | |
410 | }; | |
411 | typedef struct PageSearchStatus PageSearchStatus; | |
412 | ||
76e03000 XG |
413 | CompressionStats compression_counters; |
414 | ||
56e93d26 | 415 | struct CompressParam { |
56e93d26 | 416 | bool done; |
90e56fb4 | 417 | bool quit; |
5e5fdcff | 418 | bool zero_page; |
56e93d26 JQ |
419 | QEMUFile *file; |
420 | QemuMutex mutex; | |
421 | QemuCond cond; | |
422 | RAMBlock *block; | |
423 | ram_addr_t offset; | |
34ab9e97 XG |
424 | |
425 | /* internally used fields */ | |
dcaf446e | 426 | z_stream stream; |
34ab9e97 | 427 | uint8_t *originbuf; |
56e93d26 JQ |
428 | }; |
429 | typedef struct CompressParam CompressParam; | |
430 | ||
431 | struct DecompressParam { | |
73a8912b | 432 | bool done; |
90e56fb4 | 433 | bool quit; |
56e93d26 JQ |
434 | QemuMutex mutex; |
435 | QemuCond cond; | |
436 | void *des; | |
d341d9f3 | 437 | uint8_t *compbuf; |
56e93d26 | 438 | int len; |
797ca154 | 439 | z_stream stream; |
56e93d26 JQ |
440 | }; |
441 | typedef struct DecompressParam DecompressParam; | |
442 | ||
443 | static CompressParam *comp_param; | |
444 | static QemuThread *compress_threads; | |
445 | /* comp_done_cond is used to wake up the migration thread when | |
446 | * one of the compression threads has finished the compression. | |
447 | * comp_done_lock is used to co-work with comp_done_cond. | |
448 | */ | |
0d9f9a5c LL |
449 | static QemuMutex comp_done_lock; |
450 | static QemuCond comp_done_cond; | |
56e93d26 JQ |
451 | /* The empty QEMUFileOps will be used by file in CompressParam */ |
452 | static const QEMUFileOps empty_ops = { }; | |
453 | ||
34ab9e97 | 454 | static QEMUFile *decomp_file; |
56e93d26 JQ |
455 | static DecompressParam *decomp_param; |
456 | static QemuThread *decompress_threads; | |
73a8912b LL |
457 | static QemuMutex decomp_done_lock; |
458 | static QemuCond decomp_done_cond; | |
56e93d26 | 459 | |
5e5fdcff | 460 | static bool do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
6ef3771c | 461 | ram_addr_t offset, uint8_t *source_buf); |
56e93d26 JQ |
462 | |
463 | static void *do_data_compress(void *opaque) | |
464 | { | |
465 | CompressParam *param = opaque; | |
a7a9a88f LL |
466 | RAMBlock *block; |
467 | ram_addr_t offset; | |
5e5fdcff | 468 | bool zero_page; |
56e93d26 | 469 | |
a7a9a88f | 470 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 471 | while (!param->quit) { |
a7a9a88f LL |
472 | if (param->block) { |
473 | block = param->block; | |
474 | offset = param->offset; | |
475 | param->block = NULL; | |
476 | qemu_mutex_unlock(¶m->mutex); | |
477 | ||
5e5fdcff XG |
478 | zero_page = do_compress_ram_page(param->file, ¶m->stream, |
479 | block, offset, param->originbuf); | |
a7a9a88f | 480 | |
0d9f9a5c | 481 | qemu_mutex_lock(&comp_done_lock); |
a7a9a88f | 482 | param->done = true; |
5e5fdcff | 483 | param->zero_page = zero_page; |
0d9f9a5c LL |
484 | qemu_cond_signal(&comp_done_cond); |
485 | qemu_mutex_unlock(&comp_done_lock); | |
a7a9a88f LL |
486 | |
487 | qemu_mutex_lock(¶m->mutex); | |
488 | } else { | |
56e93d26 JQ |
489 | qemu_cond_wait(¶m->cond, ¶m->mutex); |
490 | } | |
56e93d26 | 491 | } |
a7a9a88f | 492 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
493 | |
494 | return NULL; | |
495 | } | |
496 | ||
f0afa331 | 497 | static void compress_threads_save_cleanup(void) |
56e93d26 JQ |
498 | { |
499 | int i, thread_count; | |
500 | ||
05306935 | 501 | if (!migrate_use_compression() || !comp_param) { |
56e93d26 JQ |
502 | return; |
503 | } | |
05306935 | 504 | |
56e93d26 JQ |
505 | thread_count = migrate_compress_threads(); |
506 | for (i = 0; i < thread_count; i++) { | |
dcaf446e XG |
507 | /* |
508 | * we use it as a indicator which shows if the thread is | |
509 | * properly init'd or not | |
510 | */ | |
511 | if (!comp_param[i].file) { | |
512 | break; | |
513 | } | |
05306935 FL |
514 | |
515 | qemu_mutex_lock(&comp_param[i].mutex); | |
516 | comp_param[i].quit = true; | |
517 | qemu_cond_signal(&comp_param[i].cond); | |
518 | qemu_mutex_unlock(&comp_param[i].mutex); | |
519 | ||
56e93d26 | 520 | qemu_thread_join(compress_threads + i); |
56e93d26 JQ |
521 | qemu_mutex_destroy(&comp_param[i].mutex); |
522 | qemu_cond_destroy(&comp_param[i].cond); | |
dcaf446e | 523 | deflateEnd(&comp_param[i].stream); |
34ab9e97 | 524 | g_free(comp_param[i].originbuf); |
dcaf446e XG |
525 | qemu_fclose(comp_param[i].file); |
526 | comp_param[i].file = NULL; | |
56e93d26 | 527 | } |
0d9f9a5c LL |
528 | qemu_mutex_destroy(&comp_done_lock); |
529 | qemu_cond_destroy(&comp_done_cond); | |
56e93d26 JQ |
530 | g_free(compress_threads); |
531 | g_free(comp_param); | |
56e93d26 JQ |
532 | compress_threads = NULL; |
533 | comp_param = NULL; | |
56e93d26 JQ |
534 | } |
535 | ||
dcaf446e | 536 | static int compress_threads_save_setup(void) |
56e93d26 JQ |
537 | { |
538 | int i, thread_count; | |
539 | ||
540 | if (!migrate_use_compression()) { | |
dcaf446e | 541 | return 0; |
56e93d26 | 542 | } |
56e93d26 JQ |
543 | thread_count = migrate_compress_threads(); |
544 | compress_threads = g_new0(QemuThread, thread_count); | |
545 | comp_param = g_new0(CompressParam, thread_count); | |
0d9f9a5c LL |
546 | qemu_cond_init(&comp_done_cond); |
547 | qemu_mutex_init(&comp_done_lock); | |
56e93d26 | 548 | for (i = 0; i < thread_count; i++) { |
34ab9e97 XG |
549 | comp_param[i].originbuf = g_try_malloc(TARGET_PAGE_SIZE); |
550 | if (!comp_param[i].originbuf) { | |
551 | goto exit; | |
552 | } | |
553 | ||
dcaf446e XG |
554 | if (deflateInit(&comp_param[i].stream, |
555 | migrate_compress_level()) != Z_OK) { | |
34ab9e97 | 556 | g_free(comp_param[i].originbuf); |
dcaf446e XG |
557 | goto exit; |
558 | } | |
559 | ||
e110aa91 C |
560 | /* comp_param[i].file is just used as a dummy buffer to save data, |
561 | * set its ops to empty. | |
56e93d26 JQ |
562 | */ |
563 | comp_param[i].file = qemu_fopen_ops(NULL, &empty_ops); | |
564 | comp_param[i].done = true; | |
90e56fb4 | 565 | comp_param[i].quit = false; |
56e93d26 JQ |
566 | qemu_mutex_init(&comp_param[i].mutex); |
567 | qemu_cond_init(&comp_param[i].cond); | |
568 | qemu_thread_create(compress_threads + i, "compress", | |
569 | do_data_compress, comp_param + i, | |
570 | QEMU_THREAD_JOINABLE); | |
571 | } | |
dcaf446e XG |
572 | return 0; |
573 | ||
574 | exit: | |
575 | compress_threads_save_cleanup(); | |
576 | return -1; | |
56e93d26 JQ |
577 | } |
578 | ||
f986c3d2 JQ |
579 | /* Multiple fd's */ |
580 | ||
af8b7d2b JQ |
581 | #define MULTIFD_MAGIC 0x11223344U |
582 | #define MULTIFD_VERSION 1 | |
583 | ||
6df264ac JQ |
584 | #define MULTIFD_FLAG_SYNC (1 << 0) |
585 | ||
af8b7d2b JQ |
586 | typedef struct { |
587 | uint32_t magic; | |
588 | uint32_t version; | |
589 | unsigned char uuid[16]; /* QemuUUID */ | |
590 | uint8_t id; | |
591 | } __attribute__((packed)) MultiFDInit_t; | |
592 | ||
2a26c979 JQ |
593 | typedef struct { |
594 | uint32_t magic; | |
595 | uint32_t version; | |
596 | uint32_t flags; | |
6f862692 JQ |
597 | /* maximum number of allocated pages */ |
598 | uint32_t pages_alloc; | |
599 | uint32_t pages_used; | |
2a34ee59 JQ |
600 | /* size of the next packet that contains pages */ |
601 | uint32_t next_packet_size; | |
2a26c979 JQ |
602 | uint64_t packet_num; |
603 | char ramblock[256]; | |
604 | uint64_t offset[]; | |
605 | } __attribute__((packed)) MultiFDPacket_t; | |
606 | ||
34c55a94 JQ |
607 | typedef struct { |
608 | /* number of used pages */ | |
609 | uint32_t used; | |
610 | /* number of allocated pages */ | |
611 | uint32_t allocated; | |
612 | /* global number of generated multifd packets */ | |
613 | uint64_t packet_num; | |
614 | /* offset of each page */ | |
615 | ram_addr_t *offset; | |
616 | /* pointer to each page */ | |
617 | struct iovec *iov; | |
618 | RAMBlock *block; | |
619 | } MultiFDPages_t; | |
620 | ||
8c4598f2 JQ |
621 | typedef struct { |
622 | /* this fields are not changed once the thread is created */ | |
623 | /* channel number */ | |
f986c3d2 | 624 | uint8_t id; |
8c4598f2 | 625 | /* channel thread name */ |
f986c3d2 | 626 | char *name; |
8c4598f2 | 627 | /* channel thread id */ |
f986c3d2 | 628 | QemuThread thread; |
8c4598f2 | 629 | /* communication channel */ |
60df2d4a | 630 | QIOChannel *c; |
8c4598f2 | 631 | /* sem where to wait for more work */ |
f986c3d2 | 632 | QemuSemaphore sem; |
8c4598f2 | 633 | /* this mutex protects the following parameters */ |
f986c3d2 | 634 | QemuMutex mutex; |
8c4598f2 | 635 | /* is this channel thread running */ |
66770707 | 636 | bool running; |
8c4598f2 | 637 | /* should this thread finish */ |
f986c3d2 | 638 | bool quit; |
0beb5ed3 JQ |
639 | /* thread has work to do */ |
640 | int pending_job; | |
34c55a94 JQ |
641 | /* array of pages to sent */ |
642 | MultiFDPages_t *pages; | |
2a26c979 JQ |
643 | /* packet allocated len */ |
644 | uint32_t packet_len; | |
645 | /* pointer to the packet */ | |
646 | MultiFDPacket_t *packet; | |
647 | /* multifd flags for each packet */ | |
648 | uint32_t flags; | |
2a34ee59 JQ |
649 | /* size of the next packet that contains pages */ |
650 | uint32_t next_packet_size; | |
2a26c979 JQ |
651 | /* global number of generated multifd packets */ |
652 | uint64_t packet_num; | |
408ea6ae JQ |
653 | /* thread local variables */ |
654 | /* packets sent through this channel */ | |
655 | uint64_t num_packets; | |
656 | /* pages sent through this channel */ | |
657 | uint64_t num_pages; | |
6df264ac JQ |
658 | /* syncs main thread and channels */ |
659 | QemuSemaphore sem_sync; | |
8c4598f2 JQ |
660 | } MultiFDSendParams; |
661 | ||
662 | typedef struct { | |
663 | /* this fields are not changed once the thread is created */ | |
664 | /* channel number */ | |
665 | uint8_t id; | |
666 | /* channel thread name */ | |
667 | char *name; | |
668 | /* channel thread id */ | |
669 | QemuThread thread; | |
670 | /* communication channel */ | |
671 | QIOChannel *c; | |
8c4598f2 JQ |
672 | /* this mutex protects the following parameters */ |
673 | QemuMutex mutex; | |
674 | /* is this channel thread running */ | |
675 | bool running; | |
34c55a94 JQ |
676 | /* array of pages to receive */ |
677 | MultiFDPages_t *pages; | |
2a26c979 JQ |
678 | /* packet allocated len */ |
679 | uint32_t packet_len; | |
680 | /* pointer to the packet */ | |
681 | MultiFDPacket_t *packet; | |
682 | /* multifd flags for each packet */ | |
683 | uint32_t flags; | |
684 | /* global number of generated multifd packets */ | |
685 | uint64_t packet_num; | |
408ea6ae | 686 | /* thread local variables */ |
2a34ee59 JQ |
687 | /* size of the next packet that contains pages */ |
688 | uint32_t next_packet_size; | |
408ea6ae JQ |
689 | /* packets sent through this channel */ |
690 | uint64_t num_packets; | |
691 | /* pages sent through this channel */ | |
692 | uint64_t num_pages; | |
6df264ac JQ |
693 | /* syncs main thread and channels */ |
694 | QemuSemaphore sem_sync; | |
8c4598f2 | 695 | } MultiFDRecvParams; |
f986c3d2 | 696 | |
af8b7d2b JQ |
697 | static int multifd_send_initial_packet(MultiFDSendParams *p, Error **errp) |
698 | { | |
699 | MultiFDInit_t msg; | |
700 | int ret; | |
701 | ||
702 | msg.magic = cpu_to_be32(MULTIFD_MAGIC); | |
703 | msg.version = cpu_to_be32(MULTIFD_VERSION); | |
704 | msg.id = p->id; | |
705 | memcpy(msg.uuid, &qemu_uuid.data, sizeof(msg.uuid)); | |
706 | ||
707 | ret = qio_channel_write_all(p->c, (char *)&msg, sizeof(msg), errp); | |
708 | if (ret != 0) { | |
709 | return -1; | |
710 | } | |
711 | return 0; | |
712 | } | |
713 | ||
714 | static int multifd_recv_initial_packet(QIOChannel *c, Error **errp) | |
715 | { | |
716 | MultiFDInit_t msg; | |
717 | int ret; | |
718 | ||
719 | ret = qio_channel_read_all(c, (char *)&msg, sizeof(msg), errp); | |
720 | if (ret != 0) { | |
721 | return -1; | |
722 | } | |
723 | ||
341ba0df PM |
724 | msg.magic = be32_to_cpu(msg.magic); |
725 | msg.version = be32_to_cpu(msg.version); | |
af8b7d2b JQ |
726 | |
727 | if (msg.magic != MULTIFD_MAGIC) { | |
728 | error_setg(errp, "multifd: received packet magic %x " | |
729 | "expected %x", msg.magic, MULTIFD_MAGIC); | |
730 | return -1; | |
731 | } | |
732 | ||
733 | if (msg.version != MULTIFD_VERSION) { | |
734 | error_setg(errp, "multifd: received packet version %d " | |
735 | "expected %d", msg.version, MULTIFD_VERSION); | |
736 | return -1; | |
737 | } | |
738 | ||
739 | if (memcmp(msg.uuid, &qemu_uuid, sizeof(qemu_uuid))) { | |
740 | char *uuid = qemu_uuid_unparse_strdup(&qemu_uuid); | |
741 | char *msg_uuid = qemu_uuid_unparse_strdup((const QemuUUID *)msg.uuid); | |
742 | ||
743 | error_setg(errp, "multifd: received uuid '%s' and expected " | |
744 | "uuid '%s' for channel %hhd", msg_uuid, uuid, msg.id); | |
745 | g_free(uuid); | |
746 | g_free(msg_uuid); | |
747 | return -1; | |
748 | } | |
749 | ||
750 | if (msg.id > migrate_multifd_channels()) { | |
751 | error_setg(errp, "multifd: received channel version %d " | |
752 | "expected %d", msg.version, MULTIFD_VERSION); | |
753 | return -1; | |
754 | } | |
755 | ||
756 | return msg.id; | |
757 | } | |
758 | ||
34c55a94 JQ |
759 | static MultiFDPages_t *multifd_pages_init(size_t size) |
760 | { | |
761 | MultiFDPages_t *pages = g_new0(MultiFDPages_t, 1); | |
762 | ||
763 | pages->allocated = size; | |
764 | pages->iov = g_new0(struct iovec, size); | |
765 | pages->offset = g_new0(ram_addr_t, size); | |
766 | ||
767 | return pages; | |
768 | } | |
769 | ||
770 | static void multifd_pages_clear(MultiFDPages_t *pages) | |
771 | { | |
772 | pages->used = 0; | |
773 | pages->allocated = 0; | |
774 | pages->packet_num = 0; | |
775 | pages->block = NULL; | |
776 | g_free(pages->iov); | |
777 | pages->iov = NULL; | |
778 | g_free(pages->offset); | |
779 | pages->offset = NULL; | |
780 | g_free(pages); | |
781 | } | |
782 | ||
2a26c979 JQ |
783 | static void multifd_send_fill_packet(MultiFDSendParams *p) |
784 | { | |
785 | MultiFDPacket_t *packet = p->packet; | |
786 | int i; | |
787 | ||
788 | packet->magic = cpu_to_be32(MULTIFD_MAGIC); | |
789 | packet->version = cpu_to_be32(MULTIFD_VERSION); | |
790 | packet->flags = cpu_to_be32(p->flags); | |
6f862692 JQ |
791 | packet->pages_alloc = cpu_to_be32(migrate_multifd_page_count()); |
792 | packet->pages_used = cpu_to_be32(p->pages->used); | |
2a34ee59 | 793 | packet->next_packet_size = cpu_to_be32(p->next_packet_size); |
2a26c979 JQ |
794 | packet->packet_num = cpu_to_be64(p->packet_num); |
795 | ||
796 | if (p->pages->block) { | |
797 | strncpy(packet->ramblock, p->pages->block->idstr, 256); | |
798 | } | |
799 | ||
800 | for (i = 0; i < p->pages->used; i++) { | |
801 | packet->offset[i] = cpu_to_be64(p->pages->offset[i]); | |
802 | } | |
803 | } | |
804 | ||
805 | static int multifd_recv_unfill_packet(MultiFDRecvParams *p, Error **errp) | |
806 | { | |
807 | MultiFDPacket_t *packet = p->packet; | |
808 | RAMBlock *block; | |
809 | int i; | |
810 | ||
341ba0df | 811 | packet->magic = be32_to_cpu(packet->magic); |
2a26c979 JQ |
812 | if (packet->magic != MULTIFD_MAGIC) { |
813 | error_setg(errp, "multifd: received packet " | |
814 | "magic %x and expected magic %x", | |
815 | packet->magic, MULTIFD_MAGIC); | |
816 | return -1; | |
817 | } | |
818 | ||
341ba0df | 819 | packet->version = be32_to_cpu(packet->version); |
2a26c979 JQ |
820 | if (packet->version != MULTIFD_VERSION) { |
821 | error_setg(errp, "multifd: received packet " | |
822 | "version %d and expected version %d", | |
823 | packet->version, MULTIFD_VERSION); | |
824 | return -1; | |
825 | } | |
826 | ||
827 | p->flags = be32_to_cpu(packet->flags); | |
828 | ||
6f862692 JQ |
829 | packet->pages_alloc = be32_to_cpu(packet->pages_alloc); |
830 | if (packet->pages_alloc > migrate_multifd_page_count()) { | |
2a26c979 JQ |
831 | error_setg(errp, "multifd: received packet " |
832 | "with size %d and expected maximum size %d", | |
6f862692 | 833 | packet->pages_alloc, migrate_multifd_page_count()) ; |
2a26c979 JQ |
834 | return -1; |
835 | } | |
836 | ||
6f862692 JQ |
837 | p->pages->used = be32_to_cpu(packet->pages_used); |
838 | if (p->pages->used > packet->pages_alloc) { | |
2a26c979 | 839 | error_setg(errp, "multifd: received packet " |
6f862692 JQ |
840 | "with %d pages and expected maximum pages are %d", |
841 | p->pages->used, packet->pages_alloc) ; | |
2a26c979 JQ |
842 | return -1; |
843 | } | |
844 | ||
2a34ee59 | 845 | p->next_packet_size = be32_to_cpu(packet->next_packet_size); |
2a26c979 JQ |
846 | p->packet_num = be64_to_cpu(packet->packet_num); |
847 | ||
848 | if (p->pages->used) { | |
849 | /* make sure that ramblock is 0 terminated */ | |
850 | packet->ramblock[255] = 0; | |
851 | block = qemu_ram_block_by_name(packet->ramblock); | |
852 | if (!block) { | |
853 | error_setg(errp, "multifd: unknown ram block %s", | |
854 | packet->ramblock); | |
855 | return -1; | |
856 | } | |
857 | } | |
858 | ||
859 | for (i = 0; i < p->pages->used; i++) { | |
860 | ram_addr_t offset = be64_to_cpu(packet->offset[i]); | |
861 | ||
862 | if (offset > (block->used_length - TARGET_PAGE_SIZE)) { | |
863 | error_setg(errp, "multifd: offset too long " RAM_ADDR_FMT | |
864 | " (max " RAM_ADDR_FMT ")", | |
865 | offset, block->max_length); | |
866 | return -1; | |
867 | } | |
868 | p->pages->iov[i].iov_base = block->host + offset; | |
869 | p->pages->iov[i].iov_len = TARGET_PAGE_SIZE; | |
870 | } | |
871 | ||
872 | return 0; | |
873 | } | |
874 | ||
f986c3d2 JQ |
875 | struct { |
876 | MultiFDSendParams *params; | |
877 | /* number of created threads */ | |
878 | int count; | |
34c55a94 JQ |
879 | /* array of pages to sent */ |
880 | MultiFDPages_t *pages; | |
6df264ac JQ |
881 | /* syncs main thread and channels */ |
882 | QemuSemaphore sem_sync; | |
883 | /* global number of generated multifd packets */ | |
884 | uint64_t packet_num; | |
b9ee2f7d JQ |
885 | /* send channels ready */ |
886 | QemuSemaphore channels_ready; | |
f986c3d2 JQ |
887 | } *multifd_send_state; |
888 | ||
b9ee2f7d JQ |
889 | /* |
890 | * How we use multifd_send_state->pages and channel->pages? | |
891 | * | |
892 | * We create a pages for each channel, and a main one. Each time that | |
893 | * we need to send a batch of pages we interchange the ones between | |
894 | * multifd_send_state and the channel that is sending it. There are | |
895 | * two reasons for that: | |
896 | * - to not have to do so many mallocs during migration | |
897 | * - to make easier to know what to free at the end of migration | |
898 | * | |
899 | * This way we always know who is the owner of each "pages" struct, | |
900 | * and we don't need any loocking. It belongs to the migration thread | |
901 | * or to the channel thread. Switching is safe because the migration | |
902 | * thread is using the channel mutex when changing it, and the channel | |
903 | * have to had finish with its own, otherwise pending_job can't be | |
904 | * false. | |
905 | */ | |
906 | ||
907 | static void multifd_send_pages(void) | |
908 | { | |
909 | int i; | |
910 | static int next_channel; | |
911 | MultiFDSendParams *p = NULL; /* make happy gcc */ | |
912 | MultiFDPages_t *pages = multifd_send_state->pages; | |
913 | uint64_t transferred; | |
914 | ||
915 | qemu_sem_wait(&multifd_send_state->channels_ready); | |
916 | for (i = next_channel;; i = (i + 1) % migrate_multifd_channels()) { | |
917 | p = &multifd_send_state->params[i]; | |
918 | ||
919 | qemu_mutex_lock(&p->mutex); | |
920 | if (!p->pending_job) { | |
921 | p->pending_job++; | |
922 | next_channel = (i + 1) % migrate_multifd_channels(); | |
923 | break; | |
924 | } | |
925 | qemu_mutex_unlock(&p->mutex); | |
926 | } | |
927 | p->pages->used = 0; | |
928 | ||
929 | p->packet_num = multifd_send_state->packet_num++; | |
930 | p->pages->block = NULL; | |
931 | multifd_send_state->pages = p->pages; | |
932 | p->pages = pages; | |
4fcefd44 | 933 | transferred = ((uint64_t) pages->used) * TARGET_PAGE_SIZE + p->packet_len; |
b9ee2f7d JQ |
934 | ram_counters.multifd_bytes += transferred; |
935 | ram_counters.transferred += transferred;; | |
936 | qemu_mutex_unlock(&p->mutex); | |
937 | qemu_sem_post(&p->sem); | |
938 | } | |
939 | ||
940 | static void multifd_queue_page(RAMBlock *block, ram_addr_t offset) | |
941 | { | |
942 | MultiFDPages_t *pages = multifd_send_state->pages; | |
943 | ||
944 | if (!pages->block) { | |
945 | pages->block = block; | |
946 | } | |
947 | ||
948 | if (pages->block == block) { | |
949 | pages->offset[pages->used] = offset; | |
950 | pages->iov[pages->used].iov_base = block->host + offset; | |
951 | pages->iov[pages->used].iov_len = TARGET_PAGE_SIZE; | |
952 | pages->used++; | |
953 | ||
954 | if (pages->used < pages->allocated) { | |
955 | return; | |
956 | } | |
957 | } | |
958 | ||
959 | multifd_send_pages(); | |
960 | ||
961 | if (pages->block != block) { | |
962 | multifd_queue_page(block, offset); | |
963 | } | |
964 | } | |
965 | ||
66770707 | 966 | static void multifd_send_terminate_threads(Error *err) |
f986c3d2 JQ |
967 | { |
968 | int i; | |
969 | ||
7a169d74 JQ |
970 | if (err) { |
971 | MigrationState *s = migrate_get_current(); | |
972 | migrate_set_error(s, err); | |
973 | if (s->state == MIGRATION_STATUS_SETUP || | |
974 | s->state == MIGRATION_STATUS_PRE_SWITCHOVER || | |
975 | s->state == MIGRATION_STATUS_DEVICE || | |
976 | s->state == MIGRATION_STATUS_ACTIVE) { | |
977 | migrate_set_state(&s->state, s->state, | |
978 | MIGRATION_STATUS_FAILED); | |
979 | } | |
980 | } | |
981 | ||
66770707 | 982 | for (i = 0; i < migrate_multifd_channels(); i++) { |
f986c3d2 JQ |
983 | MultiFDSendParams *p = &multifd_send_state->params[i]; |
984 | ||
985 | qemu_mutex_lock(&p->mutex); | |
986 | p->quit = true; | |
987 | qemu_sem_post(&p->sem); | |
988 | qemu_mutex_unlock(&p->mutex); | |
989 | } | |
990 | } | |
991 | ||
1398b2e3 | 992 | void multifd_save_cleanup(void) |
f986c3d2 JQ |
993 | { |
994 | int i; | |
f986c3d2 JQ |
995 | |
996 | if (!migrate_use_multifd()) { | |
1398b2e3 | 997 | return; |
f986c3d2 | 998 | } |
66770707 JQ |
999 | multifd_send_terminate_threads(NULL); |
1000 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
f986c3d2 JQ |
1001 | MultiFDSendParams *p = &multifd_send_state->params[i]; |
1002 | ||
66770707 JQ |
1003 | if (p->running) { |
1004 | qemu_thread_join(&p->thread); | |
1005 | } | |
60df2d4a JQ |
1006 | socket_send_channel_destroy(p->c); |
1007 | p->c = NULL; | |
f986c3d2 JQ |
1008 | qemu_mutex_destroy(&p->mutex); |
1009 | qemu_sem_destroy(&p->sem); | |
6df264ac | 1010 | qemu_sem_destroy(&p->sem_sync); |
f986c3d2 JQ |
1011 | g_free(p->name); |
1012 | p->name = NULL; | |
34c55a94 JQ |
1013 | multifd_pages_clear(p->pages); |
1014 | p->pages = NULL; | |
2a26c979 JQ |
1015 | p->packet_len = 0; |
1016 | g_free(p->packet); | |
1017 | p->packet = NULL; | |
f986c3d2 | 1018 | } |
b9ee2f7d | 1019 | qemu_sem_destroy(&multifd_send_state->channels_ready); |
6df264ac | 1020 | qemu_sem_destroy(&multifd_send_state->sem_sync); |
f986c3d2 JQ |
1021 | g_free(multifd_send_state->params); |
1022 | multifd_send_state->params = NULL; | |
34c55a94 JQ |
1023 | multifd_pages_clear(multifd_send_state->pages); |
1024 | multifd_send_state->pages = NULL; | |
f986c3d2 JQ |
1025 | g_free(multifd_send_state); |
1026 | multifd_send_state = NULL; | |
f986c3d2 JQ |
1027 | } |
1028 | ||
6df264ac JQ |
1029 | static void multifd_send_sync_main(void) |
1030 | { | |
1031 | int i; | |
1032 | ||
1033 | if (!migrate_use_multifd()) { | |
1034 | return; | |
1035 | } | |
b9ee2f7d JQ |
1036 | if (multifd_send_state->pages->used) { |
1037 | multifd_send_pages(); | |
1038 | } | |
6df264ac JQ |
1039 | for (i = 0; i < migrate_multifd_channels(); i++) { |
1040 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
1041 | ||
1042 | trace_multifd_send_sync_main_signal(p->id); | |
1043 | ||
1044 | qemu_mutex_lock(&p->mutex); | |
b9ee2f7d JQ |
1045 | |
1046 | p->packet_num = multifd_send_state->packet_num++; | |
6df264ac JQ |
1047 | p->flags |= MULTIFD_FLAG_SYNC; |
1048 | p->pending_job++; | |
1049 | qemu_mutex_unlock(&p->mutex); | |
1050 | qemu_sem_post(&p->sem); | |
1051 | } | |
1052 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
1053 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
1054 | ||
1055 | trace_multifd_send_sync_main_wait(p->id); | |
1056 | qemu_sem_wait(&multifd_send_state->sem_sync); | |
1057 | } | |
1058 | trace_multifd_send_sync_main(multifd_send_state->packet_num); | |
1059 | } | |
1060 | ||
f986c3d2 JQ |
1061 | static void *multifd_send_thread(void *opaque) |
1062 | { | |
1063 | MultiFDSendParams *p = opaque; | |
af8b7d2b | 1064 | Error *local_err = NULL; |
8b2db7f5 | 1065 | int ret; |
af8b7d2b | 1066 | |
408ea6ae | 1067 | trace_multifd_send_thread_start(p->id); |
74637e6f | 1068 | rcu_register_thread(); |
408ea6ae | 1069 | |
af8b7d2b JQ |
1070 | if (multifd_send_initial_packet(p, &local_err) < 0) { |
1071 | goto out; | |
1072 | } | |
408ea6ae JQ |
1073 | /* initial packet */ |
1074 | p->num_packets = 1; | |
f986c3d2 JQ |
1075 | |
1076 | while (true) { | |
d82628e4 | 1077 | qemu_sem_wait(&p->sem); |
f986c3d2 | 1078 | qemu_mutex_lock(&p->mutex); |
0beb5ed3 JQ |
1079 | |
1080 | if (p->pending_job) { | |
1081 | uint32_t used = p->pages->used; | |
1082 | uint64_t packet_num = p->packet_num; | |
1083 | uint32_t flags = p->flags; | |
1084 | ||
2a34ee59 | 1085 | p->next_packet_size = used * qemu_target_page_size(); |
0beb5ed3 JQ |
1086 | multifd_send_fill_packet(p); |
1087 | p->flags = 0; | |
1088 | p->num_packets++; | |
1089 | p->num_pages += used; | |
1090 | p->pages->used = 0; | |
1091 | qemu_mutex_unlock(&p->mutex); | |
1092 | ||
2a34ee59 JQ |
1093 | trace_multifd_send(p->id, packet_num, used, flags, |
1094 | p->next_packet_size); | |
0beb5ed3 | 1095 | |
8b2db7f5 JQ |
1096 | ret = qio_channel_write_all(p->c, (void *)p->packet, |
1097 | p->packet_len, &local_err); | |
1098 | if (ret != 0) { | |
1099 | break; | |
1100 | } | |
1101 | ||
ad24c7cb JQ |
1102 | if (used) { |
1103 | ret = qio_channel_writev_all(p->c, p->pages->iov, | |
1104 | used, &local_err); | |
1105 | if (ret != 0) { | |
1106 | break; | |
1107 | } | |
8b2db7f5 | 1108 | } |
0beb5ed3 JQ |
1109 | |
1110 | qemu_mutex_lock(&p->mutex); | |
1111 | p->pending_job--; | |
1112 | qemu_mutex_unlock(&p->mutex); | |
6df264ac JQ |
1113 | |
1114 | if (flags & MULTIFD_FLAG_SYNC) { | |
1115 | qemu_sem_post(&multifd_send_state->sem_sync); | |
1116 | } | |
b9ee2f7d | 1117 | qemu_sem_post(&multifd_send_state->channels_ready); |
0beb5ed3 | 1118 | } else if (p->quit) { |
f986c3d2 JQ |
1119 | qemu_mutex_unlock(&p->mutex); |
1120 | break; | |
6df264ac JQ |
1121 | } else { |
1122 | qemu_mutex_unlock(&p->mutex); | |
1123 | /* sometimes there are spurious wakeups */ | |
f986c3d2 | 1124 | } |
f986c3d2 JQ |
1125 | } |
1126 | ||
af8b7d2b JQ |
1127 | out: |
1128 | if (local_err) { | |
1129 | multifd_send_terminate_threads(local_err); | |
1130 | } | |
1131 | ||
66770707 JQ |
1132 | qemu_mutex_lock(&p->mutex); |
1133 | p->running = false; | |
1134 | qemu_mutex_unlock(&p->mutex); | |
1135 | ||
74637e6f | 1136 | rcu_unregister_thread(); |
408ea6ae JQ |
1137 | trace_multifd_send_thread_end(p->id, p->num_packets, p->num_pages); |
1138 | ||
f986c3d2 JQ |
1139 | return NULL; |
1140 | } | |
1141 | ||
60df2d4a JQ |
1142 | static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque) |
1143 | { | |
1144 | MultiFDSendParams *p = opaque; | |
1145 | QIOChannel *sioc = QIO_CHANNEL(qio_task_get_source(task)); | |
1146 | Error *local_err = NULL; | |
1147 | ||
1148 | if (qio_task_propagate_error(task, &local_err)) { | |
1398b2e3 FL |
1149 | migrate_set_error(migrate_get_current(), local_err); |
1150 | multifd_save_cleanup(); | |
60df2d4a JQ |
1151 | } else { |
1152 | p->c = QIO_CHANNEL(sioc); | |
1153 | qio_channel_set_delay(p->c, false); | |
1154 | p->running = true; | |
1155 | qemu_thread_create(&p->thread, p->name, multifd_send_thread, p, | |
1156 | QEMU_THREAD_JOINABLE); | |
1157 | ||
1158 | atomic_inc(&multifd_send_state->count); | |
1159 | } | |
1160 | } | |
1161 | ||
f986c3d2 JQ |
1162 | int multifd_save_setup(void) |
1163 | { | |
1164 | int thread_count; | |
34c55a94 | 1165 | uint32_t page_count = migrate_multifd_page_count(); |
f986c3d2 JQ |
1166 | uint8_t i; |
1167 | ||
1168 | if (!migrate_use_multifd()) { | |
1169 | return 0; | |
1170 | } | |
1171 | thread_count = migrate_multifd_channels(); | |
1172 | multifd_send_state = g_malloc0(sizeof(*multifd_send_state)); | |
1173 | multifd_send_state->params = g_new0(MultiFDSendParams, thread_count); | |
66770707 | 1174 | atomic_set(&multifd_send_state->count, 0); |
34c55a94 | 1175 | multifd_send_state->pages = multifd_pages_init(page_count); |
6df264ac | 1176 | qemu_sem_init(&multifd_send_state->sem_sync, 0); |
b9ee2f7d | 1177 | qemu_sem_init(&multifd_send_state->channels_ready, 0); |
34c55a94 | 1178 | |
f986c3d2 JQ |
1179 | for (i = 0; i < thread_count; i++) { |
1180 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
1181 | ||
1182 | qemu_mutex_init(&p->mutex); | |
1183 | qemu_sem_init(&p->sem, 0); | |
6df264ac | 1184 | qemu_sem_init(&p->sem_sync, 0); |
f986c3d2 | 1185 | p->quit = false; |
0beb5ed3 | 1186 | p->pending_job = 0; |
f986c3d2 | 1187 | p->id = i; |
34c55a94 | 1188 | p->pages = multifd_pages_init(page_count); |
2a26c979 JQ |
1189 | p->packet_len = sizeof(MultiFDPacket_t) |
1190 | + sizeof(ram_addr_t) * page_count; | |
1191 | p->packet = g_malloc0(p->packet_len); | |
f986c3d2 | 1192 | p->name = g_strdup_printf("multifdsend_%d", i); |
60df2d4a | 1193 | socket_send_channel_create(multifd_new_send_channel_async, p); |
f986c3d2 JQ |
1194 | } |
1195 | return 0; | |
1196 | } | |
1197 | ||
f986c3d2 JQ |
1198 | struct { |
1199 | MultiFDRecvParams *params; | |
1200 | /* number of created threads */ | |
1201 | int count; | |
6df264ac JQ |
1202 | /* syncs main thread and channels */ |
1203 | QemuSemaphore sem_sync; | |
1204 | /* global number of generated multifd packets */ | |
1205 | uint64_t packet_num; | |
f986c3d2 JQ |
1206 | } *multifd_recv_state; |
1207 | ||
66770707 | 1208 | static void multifd_recv_terminate_threads(Error *err) |
f986c3d2 JQ |
1209 | { |
1210 | int i; | |
1211 | ||
7a169d74 JQ |
1212 | if (err) { |
1213 | MigrationState *s = migrate_get_current(); | |
1214 | migrate_set_error(s, err); | |
1215 | if (s->state == MIGRATION_STATUS_SETUP || | |
1216 | s->state == MIGRATION_STATUS_ACTIVE) { | |
1217 | migrate_set_state(&s->state, s->state, | |
1218 | MIGRATION_STATUS_FAILED); | |
1219 | } | |
1220 | } | |
1221 | ||
66770707 | 1222 | for (i = 0; i < migrate_multifd_channels(); i++) { |
f986c3d2 JQ |
1223 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; |
1224 | ||
1225 | qemu_mutex_lock(&p->mutex); | |
7a5cc33c JQ |
1226 | /* We could arrive here for two reasons: |
1227 | - normal quit, i.e. everything went fine, just finished | |
1228 | - error quit: We close the channels so the channel threads | |
1229 | finish the qio_channel_read_all_eof() */ | |
1230 | qio_channel_shutdown(p->c, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); | |
f986c3d2 JQ |
1231 | qemu_mutex_unlock(&p->mutex); |
1232 | } | |
1233 | } | |
1234 | ||
1235 | int multifd_load_cleanup(Error **errp) | |
1236 | { | |
1237 | int i; | |
1238 | int ret = 0; | |
1239 | ||
1240 | if (!migrate_use_multifd()) { | |
1241 | return 0; | |
1242 | } | |
66770707 JQ |
1243 | multifd_recv_terminate_threads(NULL); |
1244 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
f986c3d2 JQ |
1245 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; |
1246 | ||
66770707 JQ |
1247 | if (p->running) { |
1248 | qemu_thread_join(&p->thread); | |
1249 | } | |
60df2d4a JQ |
1250 | object_unref(OBJECT(p->c)); |
1251 | p->c = NULL; | |
f986c3d2 | 1252 | qemu_mutex_destroy(&p->mutex); |
6df264ac | 1253 | qemu_sem_destroy(&p->sem_sync); |
f986c3d2 JQ |
1254 | g_free(p->name); |
1255 | p->name = NULL; | |
34c55a94 JQ |
1256 | multifd_pages_clear(p->pages); |
1257 | p->pages = NULL; | |
2a26c979 JQ |
1258 | p->packet_len = 0; |
1259 | g_free(p->packet); | |
1260 | p->packet = NULL; | |
f986c3d2 | 1261 | } |
6df264ac | 1262 | qemu_sem_destroy(&multifd_recv_state->sem_sync); |
f986c3d2 JQ |
1263 | g_free(multifd_recv_state->params); |
1264 | multifd_recv_state->params = NULL; | |
1265 | g_free(multifd_recv_state); | |
1266 | multifd_recv_state = NULL; | |
1267 | ||
1268 | return ret; | |
1269 | } | |
1270 | ||
6df264ac JQ |
1271 | static void multifd_recv_sync_main(void) |
1272 | { | |
1273 | int i; | |
1274 | ||
1275 | if (!migrate_use_multifd()) { | |
1276 | return; | |
1277 | } | |
1278 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
1279 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
1280 | ||
6df264ac JQ |
1281 | trace_multifd_recv_sync_main_wait(p->id); |
1282 | qemu_sem_wait(&multifd_recv_state->sem_sync); | |
1283 | qemu_mutex_lock(&p->mutex); | |
1284 | if (multifd_recv_state->packet_num < p->packet_num) { | |
1285 | multifd_recv_state->packet_num = p->packet_num; | |
1286 | } | |
1287 | qemu_mutex_unlock(&p->mutex); | |
1288 | } | |
1289 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
1290 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
1291 | ||
1292 | trace_multifd_recv_sync_main_signal(p->id); | |
6df264ac JQ |
1293 | qemu_sem_post(&p->sem_sync); |
1294 | } | |
1295 | trace_multifd_recv_sync_main(multifd_recv_state->packet_num); | |
1296 | } | |
1297 | ||
f986c3d2 JQ |
1298 | static void *multifd_recv_thread(void *opaque) |
1299 | { | |
1300 | MultiFDRecvParams *p = opaque; | |
2a26c979 JQ |
1301 | Error *local_err = NULL; |
1302 | int ret; | |
f986c3d2 | 1303 | |
408ea6ae | 1304 | trace_multifd_recv_thread_start(p->id); |
74637e6f | 1305 | rcu_register_thread(); |
408ea6ae | 1306 | |
f986c3d2 | 1307 | while (true) { |
6df264ac JQ |
1308 | uint32_t used; |
1309 | uint32_t flags; | |
0beb5ed3 | 1310 | |
8b2db7f5 JQ |
1311 | ret = qio_channel_read_all_eof(p->c, (void *)p->packet, |
1312 | p->packet_len, &local_err); | |
1313 | if (ret == 0) { /* EOF */ | |
1314 | break; | |
1315 | } | |
1316 | if (ret == -1) { /* Error */ | |
1317 | break; | |
1318 | } | |
2a26c979 | 1319 | |
6df264ac JQ |
1320 | qemu_mutex_lock(&p->mutex); |
1321 | ret = multifd_recv_unfill_packet(p, &local_err); | |
1322 | if (ret) { | |
f986c3d2 JQ |
1323 | qemu_mutex_unlock(&p->mutex); |
1324 | break; | |
1325 | } | |
6df264ac JQ |
1326 | |
1327 | used = p->pages->used; | |
1328 | flags = p->flags; | |
2a34ee59 JQ |
1329 | trace_multifd_recv(p->id, p->packet_num, used, flags, |
1330 | p->next_packet_size); | |
6df264ac JQ |
1331 | p->num_packets++; |
1332 | p->num_pages += used; | |
f986c3d2 | 1333 | qemu_mutex_unlock(&p->mutex); |
6df264ac | 1334 | |
ad24c7cb JQ |
1335 | if (used) { |
1336 | ret = qio_channel_readv_all(p->c, p->pages->iov, | |
1337 | used, &local_err); | |
1338 | if (ret != 0) { | |
1339 | break; | |
1340 | } | |
8b2db7f5 JQ |
1341 | } |
1342 | ||
6df264ac JQ |
1343 | if (flags & MULTIFD_FLAG_SYNC) { |
1344 | qemu_sem_post(&multifd_recv_state->sem_sync); | |
1345 | qemu_sem_wait(&p->sem_sync); | |
1346 | } | |
f986c3d2 JQ |
1347 | } |
1348 | ||
d82628e4 JQ |
1349 | if (local_err) { |
1350 | multifd_recv_terminate_threads(local_err); | |
1351 | } | |
66770707 JQ |
1352 | qemu_mutex_lock(&p->mutex); |
1353 | p->running = false; | |
1354 | qemu_mutex_unlock(&p->mutex); | |
1355 | ||
74637e6f | 1356 | rcu_unregister_thread(); |
408ea6ae JQ |
1357 | trace_multifd_recv_thread_end(p->id, p->num_packets, p->num_pages); |
1358 | ||
f986c3d2 JQ |
1359 | return NULL; |
1360 | } | |
1361 | ||
1362 | int multifd_load_setup(void) | |
1363 | { | |
1364 | int thread_count; | |
34c55a94 | 1365 | uint32_t page_count = migrate_multifd_page_count(); |
f986c3d2 JQ |
1366 | uint8_t i; |
1367 | ||
1368 | if (!migrate_use_multifd()) { | |
1369 | return 0; | |
1370 | } | |
1371 | thread_count = migrate_multifd_channels(); | |
1372 | multifd_recv_state = g_malloc0(sizeof(*multifd_recv_state)); | |
1373 | multifd_recv_state->params = g_new0(MultiFDRecvParams, thread_count); | |
66770707 | 1374 | atomic_set(&multifd_recv_state->count, 0); |
6df264ac | 1375 | qemu_sem_init(&multifd_recv_state->sem_sync, 0); |
34c55a94 | 1376 | |
f986c3d2 JQ |
1377 | for (i = 0; i < thread_count; i++) { |
1378 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
1379 | ||
1380 | qemu_mutex_init(&p->mutex); | |
6df264ac | 1381 | qemu_sem_init(&p->sem_sync, 0); |
f986c3d2 | 1382 | p->id = i; |
34c55a94 | 1383 | p->pages = multifd_pages_init(page_count); |
2a26c979 JQ |
1384 | p->packet_len = sizeof(MultiFDPacket_t) |
1385 | + sizeof(ram_addr_t) * page_count; | |
1386 | p->packet = g_malloc0(p->packet_len); | |
f986c3d2 | 1387 | p->name = g_strdup_printf("multifdrecv_%d", i); |
f986c3d2 JQ |
1388 | } |
1389 | return 0; | |
1390 | } | |
1391 | ||
62c1e0ca JQ |
1392 | bool multifd_recv_all_channels_created(void) |
1393 | { | |
1394 | int thread_count = migrate_multifd_channels(); | |
1395 | ||
1396 | if (!migrate_use_multifd()) { | |
1397 | return true; | |
1398 | } | |
1399 | ||
1400 | return thread_count == atomic_read(&multifd_recv_state->count); | |
1401 | } | |
1402 | ||
49ed0d24 FL |
1403 | /* |
1404 | * Try to receive all multifd channels to get ready for the migration. | |
1405 | * - Return true and do not set @errp when correctly receving all channels; | |
1406 | * - Return false and do not set @errp when correctly receiving the current one; | |
1407 | * - Return false and set @errp when failing to receive the current channel. | |
1408 | */ | |
1409 | bool multifd_recv_new_channel(QIOChannel *ioc, Error **errp) | |
71bb07db | 1410 | { |
60df2d4a | 1411 | MultiFDRecvParams *p; |
af8b7d2b JQ |
1412 | Error *local_err = NULL; |
1413 | int id; | |
60df2d4a | 1414 | |
af8b7d2b JQ |
1415 | id = multifd_recv_initial_packet(ioc, &local_err); |
1416 | if (id < 0) { | |
1417 | multifd_recv_terminate_threads(local_err); | |
49ed0d24 FL |
1418 | error_propagate_prepend(errp, local_err, |
1419 | "failed to receive packet" | |
1420 | " via multifd channel %d: ", | |
1421 | atomic_read(&multifd_recv_state->count)); | |
81e62053 | 1422 | return false; |
af8b7d2b JQ |
1423 | } |
1424 | ||
1425 | p = &multifd_recv_state->params[id]; | |
1426 | if (p->c != NULL) { | |
1427 | error_setg(&local_err, "multifd: received id '%d' already setup'", | |
1428 | id); | |
1429 | multifd_recv_terminate_threads(local_err); | |
49ed0d24 | 1430 | error_propagate(errp, local_err); |
81e62053 | 1431 | return false; |
af8b7d2b | 1432 | } |
60df2d4a JQ |
1433 | p->c = ioc; |
1434 | object_ref(OBJECT(ioc)); | |
408ea6ae JQ |
1435 | /* initial packet */ |
1436 | p->num_packets = 1; | |
60df2d4a JQ |
1437 | |
1438 | p->running = true; | |
1439 | qemu_thread_create(&p->thread, p->name, multifd_recv_thread, p, | |
1440 | QEMU_THREAD_JOINABLE); | |
1441 | atomic_inc(&multifd_recv_state->count); | |
49ed0d24 FL |
1442 | return atomic_read(&multifd_recv_state->count) == |
1443 | migrate_multifd_channels(); | |
71bb07db JQ |
1444 | } |
1445 | ||
56e93d26 | 1446 | /** |
3d0684b2 | 1447 | * save_page_header: write page header to wire |
56e93d26 JQ |
1448 | * |
1449 | * If this is the 1st block, it also writes the block identification | |
1450 | * | |
3d0684b2 | 1451 | * Returns the number of bytes written |
56e93d26 JQ |
1452 | * |
1453 | * @f: QEMUFile where to send the data | |
1454 | * @block: block that contains the page we want to send | |
1455 | * @offset: offset inside the block for the page | |
1456 | * in the lower bits, it contains flags | |
1457 | */ | |
2bf3aa85 JQ |
1458 | static size_t save_page_header(RAMState *rs, QEMUFile *f, RAMBlock *block, |
1459 | ram_addr_t offset) | |
56e93d26 | 1460 | { |
9f5f380b | 1461 | size_t size, len; |
56e93d26 | 1462 | |
24795694 JQ |
1463 | if (block == rs->last_sent_block) { |
1464 | offset |= RAM_SAVE_FLAG_CONTINUE; | |
1465 | } | |
2bf3aa85 | 1466 | qemu_put_be64(f, offset); |
56e93d26 JQ |
1467 | size = 8; |
1468 | ||
1469 | if (!(offset & RAM_SAVE_FLAG_CONTINUE)) { | |
9f5f380b | 1470 | len = strlen(block->idstr); |
2bf3aa85 JQ |
1471 | qemu_put_byte(f, len); |
1472 | qemu_put_buffer(f, (uint8_t *)block->idstr, len); | |
9f5f380b | 1473 | size += 1 + len; |
24795694 | 1474 | rs->last_sent_block = block; |
56e93d26 JQ |
1475 | } |
1476 | return size; | |
1477 | } | |
1478 | ||
3d0684b2 JQ |
1479 | /** |
1480 | * mig_throttle_guest_down: throotle down the guest | |
1481 | * | |
1482 | * Reduce amount of guest cpu execution to hopefully slow down memory | |
1483 | * writes. If guest dirty memory rate is reduced below the rate at | |
1484 | * which we can transfer pages to the destination then we should be | |
1485 | * able to complete migration. Some workloads dirty memory way too | |
1486 | * fast and will not effectively converge, even with auto-converge. | |
070afca2 JH |
1487 | */ |
1488 | static void mig_throttle_guest_down(void) | |
1489 | { | |
1490 | MigrationState *s = migrate_get_current(); | |
2594f56d DB |
1491 | uint64_t pct_initial = s->parameters.cpu_throttle_initial; |
1492 | uint64_t pct_icrement = s->parameters.cpu_throttle_increment; | |
4cbc9c7f | 1493 | int pct_max = s->parameters.max_cpu_throttle; |
070afca2 JH |
1494 | |
1495 | /* We have not started throttling yet. Let's start it. */ | |
1496 | if (!cpu_throttle_active()) { | |
1497 | cpu_throttle_set(pct_initial); | |
1498 | } else { | |
1499 | /* Throttling already on, just increase the rate */ | |
4cbc9c7f LQ |
1500 | cpu_throttle_set(MIN(cpu_throttle_get_percentage() + pct_icrement, |
1501 | pct_max)); | |
070afca2 JH |
1502 | } |
1503 | } | |
1504 | ||
3d0684b2 JQ |
1505 | /** |
1506 | * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache | |
1507 | * | |
6f37bb8b | 1508 | * @rs: current RAM state |
3d0684b2 JQ |
1509 | * @current_addr: address for the zero page |
1510 | * | |
1511 | * Update the xbzrle cache to reflect a page that's been sent as all 0. | |
56e93d26 JQ |
1512 | * The important thing is that a stale (not-yet-0'd) page be replaced |
1513 | * by the new data. | |
1514 | * As a bonus, if the page wasn't in the cache it gets added so that | |
3d0684b2 | 1515 | * when a small write is made into the 0'd page it gets XBZRLE sent. |
56e93d26 | 1516 | */ |
6f37bb8b | 1517 | static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr) |
56e93d26 | 1518 | { |
6f37bb8b | 1519 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { |
56e93d26 JQ |
1520 | return; |
1521 | } | |
1522 | ||
1523 | /* We don't care if this fails to allocate a new cache page | |
1524 | * as long as it updated an old one */ | |
c00e0928 | 1525 | cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page, |
9360447d | 1526 | ram_counters.dirty_sync_count); |
56e93d26 JQ |
1527 | } |
1528 | ||
1529 | #define ENCODING_FLAG_XBZRLE 0x1 | |
1530 | ||
1531 | /** | |
1532 | * save_xbzrle_page: compress and send current page | |
1533 | * | |
1534 | * Returns: 1 means that we wrote the page | |
1535 | * 0 means that page is identical to the one already sent | |
1536 | * -1 means that xbzrle would be longer than normal | |
1537 | * | |
5a987738 | 1538 | * @rs: current RAM state |
3d0684b2 JQ |
1539 | * @current_data: pointer to the address of the page contents |
1540 | * @current_addr: addr of the page | |
56e93d26 JQ |
1541 | * @block: block that contains the page we want to send |
1542 | * @offset: offset inside the block for the page | |
1543 | * @last_stage: if we are at the completion stage | |
56e93d26 | 1544 | */ |
204b88b8 | 1545 | static int save_xbzrle_page(RAMState *rs, uint8_t **current_data, |
56e93d26 | 1546 | ram_addr_t current_addr, RAMBlock *block, |
072c2511 | 1547 | ram_addr_t offset, bool last_stage) |
56e93d26 JQ |
1548 | { |
1549 | int encoded_len = 0, bytes_xbzrle; | |
1550 | uint8_t *prev_cached_page; | |
1551 | ||
9360447d JQ |
1552 | if (!cache_is_cached(XBZRLE.cache, current_addr, |
1553 | ram_counters.dirty_sync_count)) { | |
1554 | xbzrle_counters.cache_miss++; | |
56e93d26 JQ |
1555 | if (!last_stage) { |
1556 | if (cache_insert(XBZRLE.cache, current_addr, *current_data, | |
9360447d | 1557 | ram_counters.dirty_sync_count) == -1) { |
56e93d26 JQ |
1558 | return -1; |
1559 | } else { | |
1560 | /* update *current_data when the page has been | |
1561 | inserted into cache */ | |
1562 | *current_data = get_cached_data(XBZRLE.cache, current_addr); | |
1563 | } | |
1564 | } | |
1565 | return -1; | |
1566 | } | |
1567 | ||
1568 | prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); | |
1569 | ||
1570 | /* save current buffer into memory */ | |
1571 | memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); | |
1572 | ||
1573 | /* XBZRLE encoding (if there is no overflow) */ | |
1574 | encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, | |
1575 | TARGET_PAGE_SIZE, XBZRLE.encoded_buf, | |
1576 | TARGET_PAGE_SIZE); | |
1577 | if (encoded_len == 0) { | |
55c4446b | 1578 | trace_save_xbzrle_page_skipping(); |
56e93d26 JQ |
1579 | return 0; |
1580 | } else if (encoded_len == -1) { | |
55c4446b | 1581 | trace_save_xbzrle_page_overflow(); |
9360447d | 1582 | xbzrle_counters.overflow++; |
56e93d26 JQ |
1583 | /* update data in the cache */ |
1584 | if (!last_stage) { | |
1585 | memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE); | |
1586 | *current_data = prev_cached_page; | |
1587 | } | |
1588 | return -1; | |
1589 | } | |
1590 | ||
1591 | /* we need to update the data in the cache, in order to get the same data */ | |
1592 | if (!last_stage) { | |
1593 | memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); | |
1594 | } | |
1595 | ||
1596 | /* Send XBZRLE based compressed page */ | |
2bf3aa85 | 1597 | bytes_xbzrle = save_page_header(rs, rs->f, block, |
204b88b8 JQ |
1598 | offset | RAM_SAVE_FLAG_XBZRLE); |
1599 | qemu_put_byte(rs->f, ENCODING_FLAG_XBZRLE); | |
1600 | qemu_put_be16(rs->f, encoded_len); | |
1601 | qemu_put_buffer(rs->f, XBZRLE.encoded_buf, encoded_len); | |
56e93d26 | 1602 | bytes_xbzrle += encoded_len + 1 + 2; |
9360447d JQ |
1603 | xbzrle_counters.pages++; |
1604 | xbzrle_counters.bytes += bytes_xbzrle; | |
1605 | ram_counters.transferred += bytes_xbzrle; | |
56e93d26 JQ |
1606 | |
1607 | return 1; | |
1608 | } | |
1609 | ||
3d0684b2 JQ |
1610 | /** |
1611 | * migration_bitmap_find_dirty: find the next dirty page from start | |
f3f491fc | 1612 | * |
3d0684b2 JQ |
1613 | * Called with rcu_read_lock() to protect migration_bitmap |
1614 | * | |
1615 | * Returns the byte offset within memory region of the start of a dirty page | |
1616 | * | |
6f37bb8b | 1617 | * @rs: current RAM state |
3d0684b2 | 1618 | * @rb: RAMBlock where to search for dirty pages |
a935e30f | 1619 | * @start: page where we start the search |
f3f491fc | 1620 | */ |
56e93d26 | 1621 | static inline |
a935e30f | 1622 | unsigned long migration_bitmap_find_dirty(RAMState *rs, RAMBlock *rb, |
f20e2865 | 1623 | unsigned long start) |
56e93d26 | 1624 | { |
6b6712ef JQ |
1625 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; |
1626 | unsigned long *bitmap = rb->bmap; | |
56e93d26 JQ |
1627 | unsigned long next; |
1628 | ||
fbd162e6 | 1629 | if (ramblock_is_ignored(rb)) { |
b895de50 CLG |
1630 | return size; |
1631 | } | |
1632 | ||
6eeb63f7 WW |
1633 | /* |
1634 | * When the free page optimization is enabled, we need to check the bitmap | |
1635 | * to send the non-free pages rather than all the pages in the bulk stage. | |
1636 | */ | |
1637 | if (!rs->fpo_enabled && rs->ram_bulk_stage && start > 0) { | |
6b6712ef | 1638 | next = start + 1; |
56e93d26 | 1639 | } else { |
6b6712ef | 1640 | next = find_next_bit(bitmap, size, start); |
56e93d26 JQ |
1641 | } |
1642 | ||
6b6712ef | 1643 | return next; |
56e93d26 JQ |
1644 | } |
1645 | ||
06b10688 | 1646 | static inline bool migration_bitmap_clear_dirty(RAMState *rs, |
f20e2865 JQ |
1647 | RAMBlock *rb, |
1648 | unsigned long page) | |
a82d593b DDAG |
1649 | { |
1650 | bool ret; | |
a82d593b | 1651 | |
386a907b | 1652 | qemu_mutex_lock(&rs->bitmap_mutex); |
6b6712ef | 1653 | ret = test_and_clear_bit(page, rb->bmap); |
a82d593b DDAG |
1654 | |
1655 | if (ret) { | |
0d8ec885 | 1656 | rs->migration_dirty_pages--; |
a82d593b | 1657 | } |
386a907b WW |
1658 | qemu_mutex_unlock(&rs->bitmap_mutex); |
1659 | ||
a82d593b DDAG |
1660 | return ret; |
1661 | } | |
1662 | ||
15440dd5 JQ |
1663 | static void migration_bitmap_sync_range(RAMState *rs, RAMBlock *rb, |
1664 | ram_addr_t start, ram_addr_t length) | |
56e93d26 | 1665 | { |
0d8ec885 | 1666 | rs->migration_dirty_pages += |
6b6712ef | 1667 | cpu_physical_memory_sync_dirty_bitmap(rb, start, length, |
0d8ec885 | 1668 | &rs->num_dirty_pages_period); |
56e93d26 JQ |
1669 | } |
1670 | ||
3d0684b2 JQ |
1671 | /** |
1672 | * ram_pagesize_summary: calculate all the pagesizes of a VM | |
1673 | * | |
1674 | * Returns a summary bitmap of the page sizes of all RAMBlocks | |
1675 | * | |
1676 | * For VMs with just normal pages this is equivalent to the host page | |
1677 | * size. If it's got some huge pages then it's the OR of all the | |
1678 | * different page sizes. | |
e8ca1db2 DDAG |
1679 | */ |
1680 | uint64_t ram_pagesize_summary(void) | |
1681 | { | |
1682 | RAMBlock *block; | |
1683 | uint64_t summary = 0; | |
1684 | ||
fbd162e6 | 1685 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
e8ca1db2 DDAG |
1686 | summary |= block->page_size; |
1687 | } | |
1688 | ||
1689 | return summary; | |
1690 | } | |
1691 | ||
aecbfe9c XG |
1692 | uint64_t ram_get_total_transferred_pages(void) |
1693 | { | |
1694 | return ram_counters.normal + ram_counters.duplicate + | |
1695 | compression_counters.pages + xbzrle_counters.pages; | |
1696 | } | |
1697 | ||
b734035b XG |
1698 | static void migration_update_rates(RAMState *rs, int64_t end_time) |
1699 | { | |
be8b02ed | 1700 | uint64_t page_count = rs->target_page_count - rs->target_page_count_prev; |
76e03000 | 1701 | double compressed_size; |
b734035b XG |
1702 | |
1703 | /* calculate period counters */ | |
1704 | ram_counters.dirty_pages_rate = rs->num_dirty_pages_period * 1000 | |
1705 | / (end_time - rs->time_last_bitmap_sync); | |
1706 | ||
be8b02ed | 1707 | if (!page_count) { |
b734035b XG |
1708 | return; |
1709 | } | |
1710 | ||
1711 | if (migrate_use_xbzrle()) { | |
1712 | xbzrle_counters.cache_miss_rate = (double)(xbzrle_counters.cache_miss - | |
be8b02ed | 1713 | rs->xbzrle_cache_miss_prev) / page_count; |
b734035b XG |
1714 | rs->xbzrle_cache_miss_prev = xbzrle_counters.cache_miss; |
1715 | } | |
76e03000 XG |
1716 | |
1717 | if (migrate_use_compression()) { | |
1718 | compression_counters.busy_rate = (double)(compression_counters.busy - | |
1719 | rs->compress_thread_busy_prev) / page_count; | |
1720 | rs->compress_thread_busy_prev = compression_counters.busy; | |
1721 | ||
1722 | compressed_size = compression_counters.compressed_size - | |
1723 | rs->compressed_size_prev; | |
1724 | if (compressed_size) { | |
1725 | double uncompressed_size = (compression_counters.pages - | |
1726 | rs->compress_pages_prev) * TARGET_PAGE_SIZE; | |
1727 | ||
1728 | /* Compression-Ratio = Uncompressed-size / Compressed-size */ | |
1729 | compression_counters.compression_rate = | |
1730 | uncompressed_size / compressed_size; | |
1731 | ||
1732 | rs->compress_pages_prev = compression_counters.pages; | |
1733 | rs->compressed_size_prev = compression_counters.compressed_size; | |
1734 | } | |
1735 | } | |
b734035b XG |
1736 | } |
1737 | ||
8d820d6f | 1738 | static void migration_bitmap_sync(RAMState *rs) |
56e93d26 JQ |
1739 | { |
1740 | RAMBlock *block; | |
56e93d26 | 1741 | int64_t end_time; |
c4bdf0cf | 1742 | uint64_t bytes_xfer_now; |
56e93d26 | 1743 | |
9360447d | 1744 | ram_counters.dirty_sync_count++; |
56e93d26 | 1745 | |
f664da80 JQ |
1746 | if (!rs->time_last_bitmap_sync) { |
1747 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56e93d26 JQ |
1748 | } |
1749 | ||
1750 | trace_migration_bitmap_sync_start(); | |
9c1f8f44 | 1751 | memory_global_dirty_log_sync(); |
56e93d26 | 1752 | |
108cfae0 | 1753 | qemu_mutex_lock(&rs->bitmap_mutex); |
56e93d26 | 1754 | rcu_read_lock(); |
fbd162e6 | 1755 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
15440dd5 | 1756 | migration_bitmap_sync_range(rs, block, 0, block->used_length); |
56e93d26 | 1757 | } |
650af890 | 1758 | ram_counters.remaining = ram_bytes_remaining(); |
56e93d26 | 1759 | rcu_read_unlock(); |
108cfae0 | 1760 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 1761 | |
a66cd90c | 1762 | trace_migration_bitmap_sync_end(rs->num_dirty_pages_period); |
1ffb5dfd | 1763 | |
56e93d26 JQ |
1764 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1765 | ||
1766 | /* more than 1 second = 1000 millisecons */ | |
f664da80 | 1767 | if (end_time > rs->time_last_bitmap_sync + 1000) { |
9360447d | 1768 | bytes_xfer_now = ram_counters.transferred; |
d693c6f1 | 1769 | |
9ac78b61 PL |
1770 | /* During block migration the auto-converge logic incorrectly detects |
1771 | * that ram migration makes no progress. Avoid this by disabling the | |
1772 | * throttling logic during the bulk phase of block migration. */ | |
1773 | if (migrate_auto_converge() && !blk_mig_bulk_active()) { | |
56e93d26 JQ |
1774 | /* The following detection logic can be refined later. For now: |
1775 | Check to see if the dirtied bytes is 50% more than the approx. | |
1776 | amount of bytes that just got transferred since the last time we | |
070afca2 JH |
1777 | were in this routine. If that happens twice, start or increase |
1778 | throttling */ | |
070afca2 | 1779 | |
d693c6f1 | 1780 | if ((rs->num_dirty_pages_period * TARGET_PAGE_SIZE > |
eac74159 | 1781 | (bytes_xfer_now - rs->bytes_xfer_prev) / 2) && |
b4a3c64b | 1782 | (++rs->dirty_rate_high_cnt >= 2)) { |
56e93d26 | 1783 | trace_migration_throttle(); |
8d820d6f | 1784 | rs->dirty_rate_high_cnt = 0; |
070afca2 | 1785 | mig_throttle_guest_down(); |
d693c6f1 | 1786 | } |
56e93d26 | 1787 | } |
070afca2 | 1788 | |
b734035b XG |
1789 | migration_update_rates(rs, end_time); |
1790 | ||
be8b02ed | 1791 | rs->target_page_count_prev = rs->target_page_count; |
d693c6f1 FF |
1792 | |
1793 | /* reset period counters */ | |
f664da80 | 1794 | rs->time_last_bitmap_sync = end_time; |
a66cd90c | 1795 | rs->num_dirty_pages_period = 0; |
d2a4d85a | 1796 | rs->bytes_xfer_prev = bytes_xfer_now; |
56e93d26 | 1797 | } |
4addcd4f | 1798 | if (migrate_use_events()) { |
3ab72385 | 1799 | qapi_event_send_migration_pass(ram_counters.dirty_sync_count); |
4addcd4f | 1800 | } |
56e93d26 JQ |
1801 | } |
1802 | ||
bd227060 WW |
1803 | static void migration_bitmap_sync_precopy(RAMState *rs) |
1804 | { | |
1805 | Error *local_err = NULL; | |
1806 | ||
1807 | /* | |
1808 | * The current notifier usage is just an optimization to migration, so we | |
1809 | * don't stop the normal migration process in the error case. | |
1810 | */ | |
1811 | if (precopy_notify(PRECOPY_NOTIFY_BEFORE_BITMAP_SYNC, &local_err)) { | |
1812 | error_report_err(local_err); | |
1813 | } | |
1814 | ||
1815 | migration_bitmap_sync(rs); | |
1816 | ||
1817 | if (precopy_notify(PRECOPY_NOTIFY_AFTER_BITMAP_SYNC, &local_err)) { | |
1818 | error_report_err(local_err); | |
1819 | } | |
1820 | } | |
1821 | ||
6c97ec5f XG |
1822 | /** |
1823 | * save_zero_page_to_file: send the zero page to the file | |
1824 | * | |
1825 | * Returns the size of data written to the file, 0 means the page is not | |
1826 | * a zero page | |
1827 | * | |
1828 | * @rs: current RAM state | |
1829 | * @file: the file where the data is saved | |
1830 | * @block: block that contains the page we want to send | |
1831 | * @offset: offset inside the block for the page | |
1832 | */ | |
1833 | static int save_zero_page_to_file(RAMState *rs, QEMUFile *file, | |
1834 | RAMBlock *block, ram_addr_t offset) | |
1835 | { | |
1836 | uint8_t *p = block->host + offset; | |
1837 | int len = 0; | |
1838 | ||
1839 | if (is_zero_range(p, TARGET_PAGE_SIZE)) { | |
1840 | len += save_page_header(rs, file, block, offset | RAM_SAVE_FLAG_ZERO); | |
1841 | qemu_put_byte(file, 0); | |
1842 | len += 1; | |
1843 | } | |
1844 | return len; | |
1845 | } | |
1846 | ||
56e93d26 | 1847 | /** |
3d0684b2 | 1848 | * save_zero_page: send the zero page to the stream |
56e93d26 | 1849 | * |
3d0684b2 | 1850 | * Returns the number of pages written. |
56e93d26 | 1851 | * |
f7ccd61b | 1852 | * @rs: current RAM state |
56e93d26 JQ |
1853 | * @block: block that contains the page we want to send |
1854 | * @offset: offset inside the block for the page | |
56e93d26 | 1855 | */ |
7faccdc3 | 1856 | static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset) |
56e93d26 | 1857 | { |
6c97ec5f | 1858 | int len = save_zero_page_to_file(rs, rs->f, block, offset); |
56e93d26 | 1859 | |
6c97ec5f | 1860 | if (len) { |
9360447d | 1861 | ram_counters.duplicate++; |
6c97ec5f XG |
1862 | ram_counters.transferred += len; |
1863 | return 1; | |
56e93d26 | 1864 | } |
6c97ec5f | 1865 | return -1; |
56e93d26 JQ |
1866 | } |
1867 | ||
5727309d | 1868 | static void ram_release_pages(const char *rbname, uint64_t offset, int pages) |
53f09a10 | 1869 | { |
5727309d | 1870 | if (!migrate_release_ram() || !migration_in_postcopy()) { |
53f09a10 PB |
1871 | return; |
1872 | } | |
1873 | ||
aaa2064c | 1874 | ram_discard_range(rbname, offset, pages << TARGET_PAGE_BITS); |
53f09a10 PB |
1875 | } |
1876 | ||
059ff0fb XG |
1877 | /* |
1878 | * @pages: the number of pages written by the control path, | |
1879 | * < 0 - error | |
1880 | * > 0 - number of pages written | |
1881 | * | |
1882 | * Return true if the pages has been saved, otherwise false is returned. | |
1883 | */ | |
1884 | static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, | |
1885 | int *pages) | |
1886 | { | |
1887 | uint64_t bytes_xmit = 0; | |
1888 | int ret; | |
1889 | ||
1890 | *pages = -1; | |
1891 | ret = ram_control_save_page(rs->f, block->offset, offset, TARGET_PAGE_SIZE, | |
1892 | &bytes_xmit); | |
1893 | if (ret == RAM_SAVE_CONTROL_NOT_SUPP) { | |
1894 | return false; | |
1895 | } | |
1896 | ||
1897 | if (bytes_xmit) { | |
1898 | ram_counters.transferred += bytes_xmit; | |
1899 | *pages = 1; | |
1900 | } | |
1901 | ||
1902 | if (ret == RAM_SAVE_CONTROL_DELAYED) { | |
1903 | return true; | |
1904 | } | |
1905 | ||
1906 | if (bytes_xmit > 0) { | |
1907 | ram_counters.normal++; | |
1908 | } else if (bytes_xmit == 0) { | |
1909 | ram_counters.duplicate++; | |
1910 | } | |
1911 | ||
1912 | return true; | |
1913 | } | |
1914 | ||
65dacaa0 XG |
1915 | /* |
1916 | * directly send the page to the stream | |
1917 | * | |
1918 | * Returns the number of pages written. | |
1919 | * | |
1920 | * @rs: current RAM state | |
1921 | * @block: block that contains the page we want to send | |
1922 | * @offset: offset inside the block for the page | |
1923 | * @buf: the page to be sent | |
1924 | * @async: send to page asyncly | |
1925 | */ | |
1926 | static int save_normal_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, | |
1927 | uint8_t *buf, bool async) | |
1928 | { | |
1929 | ram_counters.transferred += save_page_header(rs, rs->f, block, | |
1930 | offset | RAM_SAVE_FLAG_PAGE); | |
1931 | if (async) { | |
1932 | qemu_put_buffer_async(rs->f, buf, TARGET_PAGE_SIZE, | |
1933 | migrate_release_ram() & | |
1934 | migration_in_postcopy()); | |
1935 | } else { | |
1936 | qemu_put_buffer(rs->f, buf, TARGET_PAGE_SIZE); | |
1937 | } | |
1938 | ram_counters.transferred += TARGET_PAGE_SIZE; | |
1939 | ram_counters.normal++; | |
1940 | return 1; | |
1941 | } | |
1942 | ||
56e93d26 | 1943 | /** |
3d0684b2 | 1944 | * ram_save_page: send the given page to the stream |
56e93d26 | 1945 | * |
3d0684b2 | 1946 | * Returns the number of pages written. |
3fd3c4b3 DDAG |
1947 | * < 0 - error |
1948 | * >=0 - Number of pages written - this might legally be 0 | |
1949 | * if xbzrle noticed the page was the same. | |
56e93d26 | 1950 | * |
6f37bb8b | 1951 | * @rs: current RAM state |
56e93d26 JQ |
1952 | * @block: block that contains the page we want to send |
1953 | * @offset: offset inside the block for the page | |
1954 | * @last_stage: if we are at the completion stage | |
56e93d26 | 1955 | */ |
a0a8aa14 | 1956 | static int ram_save_page(RAMState *rs, PageSearchStatus *pss, bool last_stage) |
56e93d26 JQ |
1957 | { |
1958 | int pages = -1; | |
56e93d26 | 1959 | uint8_t *p; |
56e93d26 | 1960 | bool send_async = true; |
a08f6890 | 1961 | RAMBlock *block = pss->block; |
a935e30f | 1962 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
059ff0fb | 1963 | ram_addr_t current_addr = block->offset + offset; |
56e93d26 | 1964 | |
2f68e399 | 1965 | p = block->host + offset; |
1db9d8e5 | 1966 | trace_ram_save_page(block->idstr, (uint64_t)offset, p); |
56e93d26 | 1967 | |
56e93d26 | 1968 | XBZRLE_cache_lock(); |
d7400a34 XG |
1969 | if (!rs->ram_bulk_stage && !migration_in_postcopy() && |
1970 | migrate_use_xbzrle()) { | |
059ff0fb XG |
1971 | pages = save_xbzrle_page(rs, &p, current_addr, block, |
1972 | offset, last_stage); | |
1973 | if (!last_stage) { | |
1974 | /* Can't send this cached data async, since the cache page | |
1975 | * might get updated before it gets to the wire | |
56e93d26 | 1976 | */ |
059ff0fb | 1977 | send_async = false; |
56e93d26 JQ |
1978 | } |
1979 | } | |
1980 | ||
1981 | /* XBZRLE overflow or normal page */ | |
1982 | if (pages == -1) { | |
65dacaa0 | 1983 | pages = save_normal_page(rs, block, offset, p, send_async); |
56e93d26 JQ |
1984 | } |
1985 | ||
1986 | XBZRLE_cache_unlock(); | |
1987 | ||
1988 | return pages; | |
1989 | } | |
1990 | ||
b9ee2f7d JQ |
1991 | static int ram_save_multifd_page(RAMState *rs, RAMBlock *block, |
1992 | ram_addr_t offset) | |
1993 | { | |
b9ee2f7d | 1994 | multifd_queue_page(block, offset); |
b9ee2f7d JQ |
1995 | ram_counters.normal++; |
1996 | ||
1997 | return 1; | |
1998 | } | |
1999 | ||
5e5fdcff | 2000 | static bool do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
6ef3771c | 2001 | ram_addr_t offset, uint8_t *source_buf) |
56e93d26 | 2002 | { |
53518d94 | 2003 | RAMState *rs = ram_state; |
a7a9a88f | 2004 | uint8_t *p = block->host + (offset & TARGET_PAGE_MASK); |
5e5fdcff | 2005 | bool zero_page = false; |
6ef3771c | 2006 | int ret; |
56e93d26 | 2007 | |
5e5fdcff XG |
2008 | if (save_zero_page_to_file(rs, f, block, offset)) { |
2009 | zero_page = true; | |
2010 | goto exit; | |
2011 | } | |
2012 | ||
6ef3771c | 2013 | save_page_header(rs, f, block, offset | RAM_SAVE_FLAG_COMPRESS_PAGE); |
34ab9e97 XG |
2014 | |
2015 | /* | |
2016 | * copy it to a internal buffer to avoid it being modified by VM | |
2017 | * so that we can catch up the error during compression and | |
2018 | * decompression | |
2019 | */ | |
2020 | memcpy(source_buf, p, TARGET_PAGE_SIZE); | |
6ef3771c XG |
2021 | ret = qemu_put_compression_data(f, stream, source_buf, TARGET_PAGE_SIZE); |
2022 | if (ret < 0) { | |
2023 | qemu_file_set_error(migrate_get_current()->to_dst_file, ret); | |
b3be2896 | 2024 | error_report("compressed data failed!"); |
5e5fdcff | 2025 | return false; |
b3be2896 | 2026 | } |
56e93d26 | 2027 | |
5e5fdcff | 2028 | exit: |
6ef3771c | 2029 | ram_release_pages(block->idstr, offset & TARGET_PAGE_MASK, 1); |
5e5fdcff XG |
2030 | return zero_page; |
2031 | } | |
2032 | ||
2033 | static void | |
2034 | update_compress_thread_counts(const CompressParam *param, int bytes_xmit) | |
2035 | { | |
76e03000 XG |
2036 | ram_counters.transferred += bytes_xmit; |
2037 | ||
5e5fdcff XG |
2038 | if (param->zero_page) { |
2039 | ram_counters.duplicate++; | |
76e03000 | 2040 | return; |
5e5fdcff | 2041 | } |
76e03000 XG |
2042 | |
2043 | /* 8 means a header with RAM_SAVE_FLAG_CONTINUE. */ | |
2044 | compression_counters.compressed_size += bytes_xmit - 8; | |
2045 | compression_counters.pages++; | |
56e93d26 JQ |
2046 | } |
2047 | ||
32b05495 XG |
2048 | static bool save_page_use_compression(RAMState *rs); |
2049 | ||
ce25d337 | 2050 | static void flush_compressed_data(RAMState *rs) |
56e93d26 JQ |
2051 | { |
2052 | int idx, len, thread_count; | |
2053 | ||
32b05495 | 2054 | if (!save_page_use_compression(rs)) { |
56e93d26 JQ |
2055 | return; |
2056 | } | |
2057 | thread_count = migrate_compress_threads(); | |
a7a9a88f | 2058 | |
0d9f9a5c | 2059 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 | 2060 | for (idx = 0; idx < thread_count; idx++) { |
a7a9a88f | 2061 | while (!comp_param[idx].done) { |
0d9f9a5c | 2062 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 | 2063 | } |
a7a9a88f | 2064 | } |
0d9f9a5c | 2065 | qemu_mutex_unlock(&comp_done_lock); |
a7a9a88f LL |
2066 | |
2067 | for (idx = 0; idx < thread_count; idx++) { | |
2068 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 2069 | if (!comp_param[idx].quit) { |
ce25d337 | 2070 | len = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
5e5fdcff XG |
2071 | /* |
2072 | * it's safe to fetch zero_page without holding comp_done_lock | |
2073 | * as there is no further request submitted to the thread, | |
2074 | * i.e, the thread should be waiting for a request at this point. | |
2075 | */ | |
2076 | update_compress_thread_counts(&comp_param[idx], len); | |
56e93d26 | 2077 | } |
a7a9a88f | 2078 | qemu_mutex_unlock(&comp_param[idx].mutex); |
56e93d26 JQ |
2079 | } |
2080 | } | |
2081 | ||
2082 | static inline void set_compress_params(CompressParam *param, RAMBlock *block, | |
2083 | ram_addr_t offset) | |
2084 | { | |
2085 | param->block = block; | |
2086 | param->offset = offset; | |
2087 | } | |
2088 | ||
ce25d337 JQ |
2089 | static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block, |
2090 | ram_addr_t offset) | |
56e93d26 JQ |
2091 | { |
2092 | int idx, thread_count, bytes_xmit = -1, pages = -1; | |
1d58872a | 2093 | bool wait = migrate_compress_wait_thread(); |
56e93d26 JQ |
2094 | |
2095 | thread_count = migrate_compress_threads(); | |
0d9f9a5c | 2096 | qemu_mutex_lock(&comp_done_lock); |
1d58872a XG |
2097 | retry: |
2098 | for (idx = 0; idx < thread_count; idx++) { | |
2099 | if (comp_param[idx].done) { | |
2100 | comp_param[idx].done = false; | |
2101 | bytes_xmit = qemu_put_qemu_file(rs->f, comp_param[idx].file); | |
2102 | qemu_mutex_lock(&comp_param[idx].mutex); | |
2103 | set_compress_params(&comp_param[idx], block, offset); | |
2104 | qemu_cond_signal(&comp_param[idx].cond); | |
2105 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
2106 | pages = 1; | |
5e5fdcff | 2107 | update_compress_thread_counts(&comp_param[idx], bytes_xmit); |
56e93d26 | 2108 | break; |
56e93d26 JQ |
2109 | } |
2110 | } | |
1d58872a XG |
2111 | |
2112 | /* | |
2113 | * wait for the free thread if the user specifies 'compress-wait-thread', | |
2114 | * otherwise we will post the page out in the main thread as normal page. | |
2115 | */ | |
2116 | if (pages < 0 && wait) { | |
2117 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); | |
2118 | goto retry; | |
2119 | } | |
0d9f9a5c | 2120 | qemu_mutex_unlock(&comp_done_lock); |
56e93d26 JQ |
2121 | |
2122 | return pages; | |
2123 | } | |
2124 | ||
3d0684b2 JQ |
2125 | /** |
2126 | * find_dirty_block: find the next dirty page and update any state | |
2127 | * associated with the search process. | |
b9e60928 | 2128 | * |
3d0684b2 | 2129 | * Returns if a page is found |
b9e60928 | 2130 | * |
6f37bb8b | 2131 | * @rs: current RAM state |
3d0684b2 JQ |
2132 | * @pss: data about the state of the current dirty page scan |
2133 | * @again: set to false if the search has scanned the whole of RAM | |
b9e60928 | 2134 | */ |
f20e2865 | 2135 | static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again) |
b9e60928 | 2136 | { |
f20e2865 | 2137 | pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page); |
6f37bb8b | 2138 | if (pss->complete_round && pss->block == rs->last_seen_block && |
a935e30f | 2139 | pss->page >= rs->last_page) { |
b9e60928 DDAG |
2140 | /* |
2141 | * We've been once around the RAM and haven't found anything. | |
2142 | * Give up. | |
2143 | */ | |
2144 | *again = false; | |
2145 | return false; | |
2146 | } | |
a935e30f | 2147 | if ((pss->page << TARGET_PAGE_BITS) >= pss->block->used_length) { |
b9e60928 | 2148 | /* Didn't find anything in this RAM Block */ |
a935e30f | 2149 | pss->page = 0; |
b9e60928 DDAG |
2150 | pss->block = QLIST_NEXT_RCU(pss->block, next); |
2151 | if (!pss->block) { | |
48df9d80 XG |
2152 | /* |
2153 | * If memory migration starts over, we will meet a dirtied page | |
2154 | * which may still exists in compression threads's ring, so we | |
2155 | * should flush the compressed data to make sure the new page | |
2156 | * is not overwritten by the old one in the destination. | |
2157 | * | |
2158 | * Also If xbzrle is on, stop using the data compression at this | |
2159 | * point. In theory, xbzrle can do better than compression. | |
2160 | */ | |
2161 | flush_compressed_data(rs); | |
2162 | ||
b9e60928 DDAG |
2163 | /* Hit the end of the list */ |
2164 | pss->block = QLIST_FIRST_RCU(&ram_list.blocks); | |
2165 | /* Flag that we've looped */ | |
2166 | pss->complete_round = true; | |
6f37bb8b | 2167 | rs->ram_bulk_stage = false; |
b9e60928 DDAG |
2168 | } |
2169 | /* Didn't find anything this time, but try again on the new block */ | |
2170 | *again = true; | |
2171 | return false; | |
2172 | } else { | |
2173 | /* Can go around again, but... */ | |
2174 | *again = true; | |
2175 | /* We've found something so probably don't need to */ | |
2176 | return true; | |
2177 | } | |
2178 | } | |
2179 | ||
3d0684b2 JQ |
2180 | /** |
2181 | * unqueue_page: gets a page of the queue | |
2182 | * | |
a82d593b | 2183 | * Helper for 'get_queued_page' - gets a page off the queue |
a82d593b | 2184 | * |
3d0684b2 JQ |
2185 | * Returns the block of the page (or NULL if none available) |
2186 | * | |
ec481c6c | 2187 | * @rs: current RAM state |
3d0684b2 | 2188 | * @offset: used to return the offset within the RAMBlock |
a82d593b | 2189 | */ |
f20e2865 | 2190 | static RAMBlock *unqueue_page(RAMState *rs, ram_addr_t *offset) |
a82d593b DDAG |
2191 | { |
2192 | RAMBlock *block = NULL; | |
2193 | ||
ae526e32 XG |
2194 | if (QSIMPLEQ_EMPTY_ATOMIC(&rs->src_page_requests)) { |
2195 | return NULL; | |
2196 | } | |
2197 | ||
ec481c6c JQ |
2198 | qemu_mutex_lock(&rs->src_page_req_mutex); |
2199 | if (!QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
2200 | struct RAMSrcPageRequest *entry = | |
2201 | QSIMPLEQ_FIRST(&rs->src_page_requests); | |
a82d593b DDAG |
2202 | block = entry->rb; |
2203 | *offset = entry->offset; | |
a82d593b DDAG |
2204 | |
2205 | if (entry->len > TARGET_PAGE_SIZE) { | |
2206 | entry->len -= TARGET_PAGE_SIZE; | |
2207 | entry->offset += TARGET_PAGE_SIZE; | |
2208 | } else { | |
2209 | memory_region_unref(block->mr); | |
ec481c6c | 2210 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
a82d593b | 2211 | g_free(entry); |
e03a34f8 | 2212 | migration_consume_urgent_request(); |
a82d593b DDAG |
2213 | } |
2214 | } | |
ec481c6c | 2215 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
a82d593b DDAG |
2216 | |
2217 | return block; | |
2218 | } | |
2219 | ||
3d0684b2 JQ |
2220 | /** |
2221 | * get_queued_page: unqueue a page from the postocpy requests | |
2222 | * | |
2223 | * Skips pages that are already sent (!dirty) | |
a82d593b | 2224 | * |
3d0684b2 | 2225 | * Returns if a queued page is found |
a82d593b | 2226 | * |
6f37bb8b | 2227 | * @rs: current RAM state |
3d0684b2 | 2228 | * @pss: data about the state of the current dirty page scan |
a82d593b | 2229 | */ |
f20e2865 | 2230 | static bool get_queued_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b DDAG |
2231 | { |
2232 | RAMBlock *block; | |
2233 | ram_addr_t offset; | |
2234 | bool dirty; | |
2235 | ||
2236 | do { | |
f20e2865 | 2237 | block = unqueue_page(rs, &offset); |
a82d593b DDAG |
2238 | /* |
2239 | * We're sending this page, and since it's postcopy nothing else | |
2240 | * will dirty it, and we must make sure it doesn't get sent again | |
2241 | * even if this queue request was received after the background | |
2242 | * search already sent it. | |
2243 | */ | |
2244 | if (block) { | |
f20e2865 JQ |
2245 | unsigned long page; |
2246 | ||
6b6712ef JQ |
2247 | page = offset >> TARGET_PAGE_BITS; |
2248 | dirty = test_bit(page, block->bmap); | |
a82d593b | 2249 | if (!dirty) { |
06b10688 | 2250 | trace_get_queued_page_not_dirty(block->idstr, (uint64_t)offset, |
6b6712ef | 2251 | page, test_bit(page, block->unsentmap)); |
a82d593b | 2252 | } else { |
f20e2865 | 2253 | trace_get_queued_page(block->idstr, (uint64_t)offset, page); |
a82d593b DDAG |
2254 | } |
2255 | } | |
2256 | ||
2257 | } while (block && !dirty); | |
2258 | ||
2259 | if (block) { | |
2260 | /* | |
2261 | * As soon as we start servicing pages out of order, then we have | |
2262 | * to kill the bulk stage, since the bulk stage assumes | |
2263 | * in (migration_bitmap_find_and_reset_dirty) that every page is | |
2264 | * dirty, that's no longer true. | |
2265 | */ | |
6f37bb8b | 2266 | rs->ram_bulk_stage = false; |
a82d593b DDAG |
2267 | |
2268 | /* | |
2269 | * We want the background search to continue from the queued page | |
2270 | * since the guest is likely to want other pages near to the page | |
2271 | * it just requested. | |
2272 | */ | |
2273 | pss->block = block; | |
a935e30f | 2274 | pss->page = offset >> TARGET_PAGE_BITS; |
a82d593b DDAG |
2275 | } |
2276 | ||
2277 | return !!block; | |
2278 | } | |
2279 | ||
6c595cde | 2280 | /** |
5e58f968 JQ |
2281 | * migration_page_queue_free: drop any remaining pages in the ram |
2282 | * request queue | |
6c595cde | 2283 | * |
3d0684b2 JQ |
2284 | * It should be empty at the end anyway, but in error cases there may |
2285 | * be some left. in case that there is any page left, we drop it. | |
2286 | * | |
6c595cde | 2287 | */ |
83c13382 | 2288 | static void migration_page_queue_free(RAMState *rs) |
6c595cde | 2289 | { |
ec481c6c | 2290 | struct RAMSrcPageRequest *mspr, *next_mspr; |
6c595cde DDAG |
2291 | /* This queue generally should be empty - but in the case of a failed |
2292 | * migration might have some droppings in. | |
2293 | */ | |
2294 | rcu_read_lock(); | |
ec481c6c | 2295 | QSIMPLEQ_FOREACH_SAFE(mspr, &rs->src_page_requests, next_req, next_mspr) { |
6c595cde | 2296 | memory_region_unref(mspr->rb->mr); |
ec481c6c | 2297 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
6c595cde DDAG |
2298 | g_free(mspr); |
2299 | } | |
2300 | rcu_read_unlock(); | |
2301 | } | |
2302 | ||
2303 | /** | |
3d0684b2 JQ |
2304 | * ram_save_queue_pages: queue the page for transmission |
2305 | * | |
2306 | * A request from postcopy destination for example. | |
2307 | * | |
2308 | * Returns zero on success or negative on error | |
2309 | * | |
3d0684b2 JQ |
2310 | * @rbname: Name of the RAMBLock of the request. NULL means the |
2311 | * same that last one. | |
2312 | * @start: starting address from the start of the RAMBlock | |
2313 | * @len: length (in bytes) to send | |
6c595cde | 2314 | */ |
96506894 | 2315 | int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len) |
6c595cde DDAG |
2316 | { |
2317 | RAMBlock *ramblock; | |
53518d94 | 2318 | RAMState *rs = ram_state; |
6c595cde | 2319 | |
9360447d | 2320 | ram_counters.postcopy_requests++; |
6c595cde DDAG |
2321 | rcu_read_lock(); |
2322 | if (!rbname) { | |
2323 | /* Reuse last RAMBlock */ | |
68a098f3 | 2324 | ramblock = rs->last_req_rb; |
6c595cde DDAG |
2325 | |
2326 | if (!ramblock) { | |
2327 | /* | |
2328 | * Shouldn't happen, we can't reuse the last RAMBlock if | |
2329 | * it's the 1st request. | |
2330 | */ | |
2331 | error_report("ram_save_queue_pages no previous block"); | |
2332 | goto err; | |
2333 | } | |
2334 | } else { | |
2335 | ramblock = qemu_ram_block_by_name(rbname); | |
2336 | ||
2337 | if (!ramblock) { | |
2338 | /* We shouldn't be asked for a non-existent RAMBlock */ | |
2339 | error_report("ram_save_queue_pages no block '%s'", rbname); | |
2340 | goto err; | |
2341 | } | |
68a098f3 | 2342 | rs->last_req_rb = ramblock; |
6c595cde DDAG |
2343 | } |
2344 | trace_ram_save_queue_pages(ramblock->idstr, start, len); | |
2345 | if (start+len > ramblock->used_length) { | |
9458ad6b JQ |
2346 | error_report("%s request overrun start=" RAM_ADDR_FMT " len=" |
2347 | RAM_ADDR_FMT " blocklen=" RAM_ADDR_FMT, | |
6c595cde DDAG |
2348 | __func__, start, len, ramblock->used_length); |
2349 | goto err; | |
2350 | } | |
2351 | ||
ec481c6c JQ |
2352 | struct RAMSrcPageRequest *new_entry = |
2353 | g_malloc0(sizeof(struct RAMSrcPageRequest)); | |
6c595cde DDAG |
2354 | new_entry->rb = ramblock; |
2355 | new_entry->offset = start; | |
2356 | new_entry->len = len; | |
2357 | ||
2358 | memory_region_ref(ramblock->mr); | |
ec481c6c JQ |
2359 | qemu_mutex_lock(&rs->src_page_req_mutex); |
2360 | QSIMPLEQ_INSERT_TAIL(&rs->src_page_requests, new_entry, next_req); | |
e03a34f8 | 2361 | migration_make_urgent_request(); |
ec481c6c | 2362 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
6c595cde DDAG |
2363 | rcu_read_unlock(); |
2364 | ||
2365 | return 0; | |
2366 | ||
2367 | err: | |
2368 | rcu_read_unlock(); | |
2369 | return -1; | |
2370 | } | |
2371 | ||
d7400a34 XG |
2372 | static bool save_page_use_compression(RAMState *rs) |
2373 | { | |
2374 | if (!migrate_use_compression()) { | |
2375 | return false; | |
2376 | } | |
2377 | ||
2378 | /* | |
2379 | * If xbzrle is on, stop using the data compression after first | |
2380 | * round of migration even if compression is enabled. In theory, | |
2381 | * xbzrle can do better than compression. | |
2382 | */ | |
2383 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { | |
2384 | return true; | |
2385 | } | |
2386 | ||
2387 | return false; | |
2388 | } | |
2389 | ||
5e5fdcff XG |
2390 | /* |
2391 | * try to compress the page before posting it out, return true if the page | |
2392 | * has been properly handled by compression, otherwise needs other | |
2393 | * paths to handle it | |
2394 | */ | |
2395 | static bool save_compress_page(RAMState *rs, RAMBlock *block, ram_addr_t offset) | |
2396 | { | |
2397 | if (!save_page_use_compression(rs)) { | |
2398 | return false; | |
2399 | } | |
2400 | ||
2401 | /* | |
2402 | * When starting the process of a new block, the first page of | |
2403 | * the block should be sent out before other pages in the same | |
2404 | * block, and all the pages in last block should have been sent | |
2405 | * out, keeping this order is important, because the 'cont' flag | |
2406 | * is used to avoid resending the block name. | |
2407 | * | |
2408 | * We post the fist page as normal page as compression will take | |
2409 | * much CPU resource. | |
2410 | */ | |
2411 | if (block != rs->last_sent_block) { | |
2412 | flush_compressed_data(rs); | |
2413 | return false; | |
2414 | } | |
2415 | ||
2416 | if (compress_page_with_multi_thread(rs, block, offset) > 0) { | |
2417 | return true; | |
2418 | } | |
2419 | ||
76e03000 | 2420 | compression_counters.busy++; |
5e5fdcff XG |
2421 | return false; |
2422 | } | |
2423 | ||
a82d593b | 2424 | /** |
3d0684b2 | 2425 | * ram_save_target_page: save one target page |
a82d593b | 2426 | * |
3d0684b2 | 2427 | * Returns the number of pages written |
a82d593b | 2428 | * |
6f37bb8b | 2429 | * @rs: current RAM state |
3d0684b2 | 2430 | * @pss: data about the page we want to send |
a82d593b | 2431 | * @last_stage: if we are at the completion stage |
a82d593b | 2432 | */ |
a0a8aa14 | 2433 | static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 2434 | bool last_stage) |
a82d593b | 2435 | { |
a8ec91f9 XG |
2436 | RAMBlock *block = pss->block; |
2437 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; | |
2438 | int res; | |
2439 | ||
2440 | if (control_save_page(rs, block, offset, &res)) { | |
2441 | return res; | |
2442 | } | |
2443 | ||
5e5fdcff XG |
2444 | if (save_compress_page(rs, block, offset)) { |
2445 | return 1; | |
d7400a34 XG |
2446 | } |
2447 | ||
2448 | res = save_zero_page(rs, block, offset); | |
2449 | if (res > 0) { | |
2450 | /* Must let xbzrle know, otherwise a previous (now 0'd) cached | |
2451 | * page would be stale | |
2452 | */ | |
2453 | if (!save_page_use_compression(rs)) { | |
2454 | XBZRLE_cache_lock(); | |
2455 | xbzrle_cache_zero_page(rs, block->offset + offset); | |
2456 | XBZRLE_cache_unlock(); | |
2457 | } | |
2458 | ram_release_pages(block->idstr, offset, res); | |
2459 | return res; | |
2460 | } | |
2461 | ||
da3f56cb | 2462 | /* |
5e5fdcff XG |
2463 | * do not use multifd for compression as the first page in the new |
2464 | * block should be posted out before sending the compressed page | |
da3f56cb | 2465 | */ |
5e5fdcff | 2466 | if (!save_page_use_compression(rs) && migrate_use_multifd()) { |
b9ee2f7d | 2467 | return ram_save_multifd_page(rs, block, offset); |
a82d593b DDAG |
2468 | } |
2469 | ||
1faa5665 | 2470 | return ram_save_page(rs, pss, last_stage); |
a82d593b DDAG |
2471 | } |
2472 | ||
2473 | /** | |
3d0684b2 | 2474 | * ram_save_host_page: save a whole host page |
a82d593b | 2475 | * |
3d0684b2 JQ |
2476 | * Starting at *offset send pages up to the end of the current host |
2477 | * page. It's valid for the initial offset to point into the middle of | |
2478 | * a host page in which case the remainder of the hostpage is sent. | |
2479 | * Only dirty target pages are sent. Note that the host page size may | |
2480 | * be a huge page for this block. | |
1eb3fc0a DDAG |
2481 | * The saving stops at the boundary of the used_length of the block |
2482 | * if the RAMBlock isn't a multiple of the host page size. | |
a82d593b | 2483 | * |
3d0684b2 JQ |
2484 | * Returns the number of pages written or negative on error |
2485 | * | |
6f37bb8b | 2486 | * @rs: current RAM state |
3d0684b2 | 2487 | * @ms: current migration state |
3d0684b2 | 2488 | * @pss: data about the page we want to send |
a82d593b | 2489 | * @last_stage: if we are at the completion stage |
a82d593b | 2490 | */ |
a0a8aa14 | 2491 | static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 2492 | bool last_stage) |
a82d593b DDAG |
2493 | { |
2494 | int tmppages, pages = 0; | |
a935e30f JQ |
2495 | size_t pagesize_bits = |
2496 | qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
4c011c37 | 2497 | |
fbd162e6 | 2498 | if (ramblock_is_ignored(pss->block)) { |
b895de50 CLG |
2499 | error_report("block %s should not be migrated !", pss->block->idstr); |
2500 | return 0; | |
2501 | } | |
2502 | ||
a82d593b | 2503 | do { |
1faa5665 XG |
2504 | /* Check the pages is dirty and if it is send it */ |
2505 | if (!migration_bitmap_clear_dirty(rs, pss->block, pss->page)) { | |
2506 | pss->page++; | |
2507 | continue; | |
2508 | } | |
2509 | ||
f20e2865 | 2510 | tmppages = ram_save_target_page(rs, pss, last_stage); |
a82d593b DDAG |
2511 | if (tmppages < 0) { |
2512 | return tmppages; | |
2513 | } | |
2514 | ||
2515 | pages += tmppages; | |
1faa5665 XG |
2516 | if (pss->block->unsentmap) { |
2517 | clear_bit(pss->page, pss->block->unsentmap); | |
2518 | } | |
2519 | ||
a935e30f | 2520 | pss->page++; |
1eb3fc0a DDAG |
2521 | } while ((pss->page & (pagesize_bits - 1)) && |
2522 | offset_in_ramblock(pss->block, pss->page << TARGET_PAGE_BITS)); | |
a82d593b DDAG |
2523 | |
2524 | /* The offset we leave with is the last one we looked at */ | |
a935e30f | 2525 | pss->page--; |
a82d593b DDAG |
2526 | return pages; |
2527 | } | |
6c595cde | 2528 | |
56e93d26 | 2529 | /** |
3d0684b2 | 2530 | * ram_find_and_save_block: finds a dirty page and sends it to f |
56e93d26 JQ |
2531 | * |
2532 | * Called within an RCU critical section. | |
2533 | * | |
e8f3735f XG |
2534 | * Returns the number of pages written where zero means no dirty pages, |
2535 | * or negative on error | |
56e93d26 | 2536 | * |
6f37bb8b | 2537 | * @rs: current RAM state |
56e93d26 | 2538 | * @last_stage: if we are at the completion stage |
a82d593b DDAG |
2539 | * |
2540 | * On systems where host-page-size > target-page-size it will send all the | |
2541 | * pages in a host page that are dirty. | |
56e93d26 JQ |
2542 | */ |
2543 | ||
ce25d337 | 2544 | static int ram_find_and_save_block(RAMState *rs, bool last_stage) |
56e93d26 | 2545 | { |
b8fb8cb7 | 2546 | PageSearchStatus pss; |
56e93d26 | 2547 | int pages = 0; |
b9e60928 | 2548 | bool again, found; |
56e93d26 | 2549 | |
0827b9e9 AA |
2550 | /* No dirty page as there is zero RAM */ |
2551 | if (!ram_bytes_total()) { | |
2552 | return pages; | |
2553 | } | |
2554 | ||
6f37bb8b | 2555 | pss.block = rs->last_seen_block; |
a935e30f | 2556 | pss.page = rs->last_page; |
b8fb8cb7 DDAG |
2557 | pss.complete_round = false; |
2558 | ||
2559 | if (!pss.block) { | |
2560 | pss.block = QLIST_FIRST_RCU(&ram_list.blocks); | |
2561 | } | |
56e93d26 | 2562 | |
b9e60928 | 2563 | do { |
a82d593b | 2564 | again = true; |
f20e2865 | 2565 | found = get_queued_page(rs, &pss); |
b9e60928 | 2566 | |
a82d593b DDAG |
2567 | if (!found) { |
2568 | /* priority queue empty, so just search for something dirty */ | |
f20e2865 | 2569 | found = find_dirty_block(rs, &pss, &again); |
a82d593b | 2570 | } |
f3f491fc | 2571 | |
a82d593b | 2572 | if (found) { |
f20e2865 | 2573 | pages = ram_save_host_page(rs, &pss, last_stage); |
56e93d26 | 2574 | } |
b9e60928 | 2575 | } while (!pages && again); |
56e93d26 | 2576 | |
6f37bb8b | 2577 | rs->last_seen_block = pss.block; |
a935e30f | 2578 | rs->last_page = pss.page; |
56e93d26 JQ |
2579 | |
2580 | return pages; | |
2581 | } | |
2582 | ||
2583 | void acct_update_position(QEMUFile *f, size_t size, bool zero) | |
2584 | { | |
2585 | uint64_t pages = size / TARGET_PAGE_SIZE; | |
f7ccd61b | 2586 | |
56e93d26 | 2587 | if (zero) { |
9360447d | 2588 | ram_counters.duplicate += pages; |
56e93d26 | 2589 | } else { |
9360447d JQ |
2590 | ram_counters.normal += pages; |
2591 | ram_counters.transferred += size; | |
56e93d26 JQ |
2592 | qemu_update_position(f, size); |
2593 | } | |
2594 | } | |
2595 | ||
fbd162e6 | 2596 | static uint64_t ram_bytes_total_common(bool count_ignored) |
56e93d26 JQ |
2597 | { |
2598 | RAMBlock *block; | |
2599 | uint64_t total = 0; | |
2600 | ||
2601 | rcu_read_lock(); | |
fbd162e6 YK |
2602 | if (count_ignored) { |
2603 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
2604 | total += block->used_length; | |
2605 | } | |
2606 | } else { | |
2607 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
2608 | total += block->used_length; | |
2609 | } | |
99e15582 | 2610 | } |
56e93d26 JQ |
2611 | rcu_read_unlock(); |
2612 | return total; | |
2613 | } | |
2614 | ||
fbd162e6 YK |
2615 | uint64_t ram_bytes_total(void) |
2616 | { | |
2617 | return ram_bytes_total_common(false); | |
2618 | } | |
2619 | ||
f265e0e4 | 2620 | static void xbzrle_load_setup(void) |
56e93d26 | 2621 | { |
f265e0e4 | 2622 | XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE); |
56e93d26 JQ |
2623 | } |
2624 | ||
f265e0e4 JQ |
2625 | static void xbzrle_load_cleanup(void) |
2626 | { | |
2627 | g_free(XBZRLE.decoded_buf); | |
2628 | XBZRLE.decoded_buf = NULL; | |
2629 | } | |
2630 | ||
7d7c96be PX |
2631 | static void ram_state_cleanup(RAMState **rsp) |
2632 | { | |
b9ccaf6d DDAG |
2633 | if (*rsp) { |
2634 | migration_page_queue_free(*rsp); | |
2635 | qemu_mutex_destroy(&(*rsp)->bitmap_mutex); | |
2636 | qemu_mutex_destroy(&(*rsp)->src_page_req_mutex); | |
2637 | g_free(*rsp); | |
2638 | *rsp = NULL; | |
2639 | } | |
7d7c96be PX |
2640 | } |
2641 | ||
84593a08 PX |
2642 | static void xbzrle_cleanup(void) |
2643 | { | |
2644 | XBZRLE_cache_lock(); | |
2645 | if (XBZRLE.cache) { | |
2646 | cache_fini(XBZRLE.cache); | |
2647 | g_free(XBZRLE.encoded_buf); | |
2648 | g_free(XBZRLE.current_buf); | |
2649 | g_free(XBZRLE.zero_target_page); | |
2650 | XBZRLE.cache = NULL; | |
2651 | XBZRLE.encoded_buf = NULL; | |
2652 | XBZRLE.current_buf = NULL; | |
2653 | XBZRLE.zero_target_page = NULL; | |
2654 | } | |
2655 | XBZRLE_cache_unlock(); | |
2656 | } | |
2657 | ||
f265e0e4 | 2658 | static void ram_save_cleanup(void *opaque) |
56e93d26 | 2659 | { |
53518d94 | 2660 | RAMState **rsp = opaque; |
6b6712ef | 2661 | RAMBlock *block; |
eb859c53 | 2662 | |
2ff64038 LZ |
2663 | /* caller have hold iothread lock or is in a bh, so there is |
2664 | * no writing race against this migration_bitmap | |
2665 | */ | |
6b6712ef JQ |
2666 | memory_global_dirty_log_stop(); |
2667 | ||
fbd162e6 | 2668 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
6b6712ef JQ |
2669 | g_free(block->bmap); |
2670 | block->bmap = NULL; | |
2671 | g_free(block->unsentmap); | |
2672 | block->unsentmap = NULL; | |
56e93d26 JQ |
2673 | } |
2674 | ||
84593a08 | 2675 | xbzrle_cleanup(); |
f0afa331 | 2676 | compress_threads_save_cleanup(); |
7d7c96be | 2677 | ram_state_cleanup(rsp); |
56e93d26 JQ |
2678 | } |
2679 | ||
6f37bb8b | 2680 | static void ram_state_reset(RAMState *rs) |
56e93d26 | 2681 | { |
6f37bb8b JQ |
2682 | rs->last_seen_block = NULL; |
2683 | rs->last_sent_block = NULL; | |
269ace29 | 2684 | rs->last_page = 0; |
6f37bb8b JQ |
2685 | rs->last_version = ram_list.version; |
2686 | rs->ram_bulk_stage = true; | |
6eeb63f7 | 2687 | rs->fpo_enabled = false; |
56e93d26 JQ |
2688 | } |
2689 | ||
2690 | #define MAX_WAIT 50 /* ms, half buffered_file limit */ | |
2691 | ||
4f2e4252 DDAG |
2692 | /* |
2693 | * 'expected' is the value you expect the bitmap mostly to be full | |
2694 | * of; it won't bother printing lines that are all this value. | |
2695 | * If 'todump' is null the migration bitmap is dumped. | |
2696 | */ | |
6b6712ef JQ |
2697 | void ram_debug_dump_bitmap(unsigned long *todump, bool expected, |
2698 | unsigned long pages) | |
4f2e4252 | 2699 | { |
4f2e4252 DDAG |
2700 | int64_t cur; |
2701 | int64_t linelen = 128; | |
2702 | char linebuf[129]; | |
2703 | ||
6b6712ef | 2704 | for (cur = 0; cur < pages; cur += linelen) { |
4f2e4252 DDAG |
2705 | int64_t curb; |
2706 | bool found = false; | |
2707 | /* | |
2708 | * Last line; catch the case where the line length | |
2709 | * is longer than remaining ram | |
2710 | */ | |
6b6712ef JQ |
2711 | if (cur + linelen > pages) { |
2712 | linelen = pages - cur; | |
4f2e4252 DDAG |
2713 | } |
2714 | for (curb = 0; curb < linelen; curb++) { | |
2715 | bool thisbit = test_bit(cur + curb, todump); | |
2716 | linebuf[curb] = thisbit ? '1' : '.'; | |
2717 | found = found || (thisbit != expected); | |
2718 | } | |
2719 | if (found) { | |
2720 | linebuf[curb] = '\0'; | |
2721 | fprintf(stderr, "0x%08" PRIx64 " : %s\n", cur, linebuf); | |
2722 | } | |
2723 | } | |
2724 | } | |
2725 | ||
e0b266f0 DDAG |
2726 | /* **** functions for postcopy ***** */ |
2727 | ||
ced1c616 PB |
2728 | void ram_postcopy_migrated_memory_release(MigrationState *ms) |
2729 | { | |
2730 | struct RAMBlock *block; | |
ced1c616 | 2731 | |
fbd162e6 | 2732 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
6b6712ef JQ |
2733 | unsigned long *bitmap = block->bmap; |
2734 | unsigned long range = block->used_length >> TARGET_PAGE_BITS; | |
2735 | unsigned long run_start = find_next_zero_bit(bitmap, range, 0); | |
ced1c616 PB |
2736 | |
2737 | while (run_start < range) { | |
2738 | unsigned long run_end = find_next_bit(bitmap, range, run_start + 1); | |
aaa2064c | 2739 | ram_discard_range(block->idstr, run_start << TARGET_PAGE_BITS, |
ced1c616 PB |
2740 | (run_end - run_start) << TARGET_PAGE_BITS); |
2741 | run_start = find_next_zero_bit(bitmap, range, run_end + 1); | |
2742 | } | |
2743 | } | |
2744 | } | |
2745 | ||
3d0684b2 JQ |
2746 | /** |
2747 | * postcopy_send_discard_bm_ram: discard a RAMBlock | |
2748 | * | |
2749 | * Returns zero on success | |
2750 | * | |
e0b266f0 DDAG |
2751 | * Callback from postcopy_each_ram_send_discard for each RAMBlock |
2752 | * Note: At this point the 'unsentmap' is the processed bitmap combined | |
2753 | * with the dirtymap; so a '1' means it's either dirty or unsent. | |
3d0684b2 JQ |
2754 | * |
2755 | * @ms: current migration state | |
2756 | * @pds: state for postcopy | |
2757 | * @start: RAMBlock starting page | |
2758 | * @length: RAMBlock size | |
e0b266f0 DDAG |
2759 | */ |
2760 | static int postcopy_send_discard_bm_ram(MigrationState *ms, | |
2761 | PostcopyDiscardState *pds, | |
6b6712ef | 2762 | RAMBlock *block) |
e0b266f0 | 2763 | { |
6b6712ef | 2764 | unsigned long end = block->used_length >> TARGET_PAGE_BITS; |
e0b266f0 | 2765 | unsigned long current; |
6b6712ef | 2766 | unsigned long *unsentmap = block->unsentmap; |
e0b266f0 | 2767 | |
6b6712ef | 2768 | for (current = 0; current < end; ) { |
e0b266f0 DDAG |
2769 | unsigned long one = find_next_bit(unsentmap, end, current); |
2770 | ||
2771 | if (one <= end) { | |
2772 | unsigned long zero = find_next_zero_bit(unsentmap, end, one + 1); | |
2773 | unsigned long discard_length; | |
2774 | ||
2775 | if (zero >= end) { | |
2776 | discard_length = end - one; | |
2777 | } else { | |
2778 | discard_length = zero - one; | |
2779 | } | |
d688c62d DDAG |
2780 | if (discard_length) { |
2781 | postcopy_discard_send_range(ms, pds, one, discard_length); | |
2782 | } | |
e0b266f0 DDAG |
2783 | current = one + discard_length; |
2784 | } else { | |
2785 | current = one; | |
2786 | } | |
2787 | } | |
2788 | ||
2789 | return 0; | |
2790 | } | |
2791 | ||
3d0684b2 JQ |
2792 | /** |
2793 | * postcopy_each_ram_send_discard: discard all RAMBlocks | |
2794 | * | |
2795 | * Returns 0 for success or negative for error | |
2796 | * | |
e0b266f0 DDAG |
2797 | * Utility for the outgoing postcopy code. |
2798 | * Calls postcopy_send_discard_bm_ram for each RAMBlock | |
2799 | * passing it bitmap indexes and name. | |
e0b266f0 DDAG |
2800 | * (qemu_ram_foreach_block ends up passing unscaled lengths |
2801 | * which would mean postcopy code would have to deal with target page) | |
3d0684b2 JQ |
2802 | * |
2803 | * @ms: current migration state | |
e0b266f0 DDAG |
2804 | */ |
2805 | static int postcopy_each_ram_send_discard(MigrationState *ms) | |
2806 | { | |
2807 | struct RAMBlock *block; | |
2808 | int ret; | |
2809 | ||
fbd162e6 | 2810 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
6b6712ef JQ |
2811 | PostcopyDiscardState *pds = |
2812 | postcopy_discard_send_init(ms, block->idstr); | |
e0b266f0 DDAG |
2813 | |
2814 | /* | |
2815 | * Postcopy sends chunks of bitmap over the wire, but it | |
2816 | * just needs indexes at this point, avoids it having | |
2817 | * target page specific code. | |
2818 | */ | |
6b6712ef | 2819 | ret = postcopy_send_discard_bm_ram(ms, pds, block); |
e0b266f0 DDAG |
2820 | postcopy_discard_send_finish(ms, pds); |
2821 | if (ret) { | |
2822 | return ret; | |
2823 | } | |
2824 | } | |
2825 | ||
2826 | return 0; | |
2827 | } | |
2828 | ||
3d0684b2 JQ |
2829 | /** |
2830 | * postcopy_chunk_hostpages_pass: canocalize bitmap in hostpages | |
2831 | * | |
2832 | * Helper for postcopy_chunk_hostpages; it's called twice to | |
2833 | * canonicalize the two bitmaps, that are similar, but one is | |
2834 | * inverted. | |
99e314eb | 2835 | * |
3d0684b2 JQ |
2836 | * Postcopy requires that all target pages in a hostpage are dirty or |
2837 | * clean, not a mix. This function canonicalizes the bitmaps. | |
99e314eb | 2838 | * |
3d0684b2 JQ |
2839 | * @ms: current migration state |
2840 | * @unsent_pass: if true we need to canonicalize partially unsent host pages | |
2841 | * otherwise we need to canonicalize partially dirty host pages | |
2842 | * @block: block that contains the page we want to canonicalize | |
2843 | * @pds: state for postcopy | |
99e314eb DDAG |
2844 | */ |
2845 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, bool unsent_pass, | |
2846 | RAMBlock *block, | |
2847 | PostcopyDiscardState *pds) | |
2848 | { | |
53518d94 | 2849 | RAMState *rs = ram_state; |
6b6712ef JQ |
2850 | unsigned long *bitmap = block->bmap; |
2851 | unsigned long *unsentmap = block->unsentmap; | |
29c59172 | 2852 | unsigned int host_ratio = block->page_size / TARGET_PAGE_SIZE; |
6b6712ef | 2853 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
99e314eb DDAG |
2854 | unsigned long run_start; |
2855 | ||
29c59172 DDAG |
2856 | if (block->page_size == TARGET_PAGE_SIZE) { |
2857 | /* Easy case - TPS==HPS for a non-huge page RAMBlock */ | |
2858 | return; | |
2859 | } | |
2860 | ||
99e314eb DDAG |
2861 | if (unsent_pass) { |
2862 | /* Find a sent page */ | |
6b6712ef | 2863 | run_start = find_next_zero_bit(unsentmap, pages, 0); |
99e314eb DDAG |
2864 | } else { |
2865 | /* Find a dirty page */ | |
6b6712ef | 2866 | run_start = find_next_bit(bitmap, pages, 0); |
99e314eb DDAG |
2867 | } |
2868 | ||
6b6712ef | 2869 | while (run_start < pages) { |
99e314eb DDAG |
2870 | bool do_fixup = false; |
2871 | unsigned long fixup_start_addr; | |
2872 | unsigned long host_offset; | |
2873 | ||
2874 | /* | |
2875 | * If the start of this run of pages is in the middle of a host | |
2876 | * page, then we need to fixup this host page. | |
2877 | */ | |
2878 | host_offset = run_start % host_ratio; | |
2879 | if (host_offset) { | |
2880 | do_fixup = true; | |
2881 | run_start -= host_offset; | |
2882 | fixup_start_addr = run_start; | |
2883 | /* For the next pass */ | |
2884 | run_start = run_start + host_ratio; | |
2885 | } else { | |
2886 | /* Find the end of this run */ | |
2887 | unsigned long run_end; | |
2888 | if (unsent_pass) { | |
6b6712ef | 2889 | run_end = find_next_bit(unsentmap, pages, run_start + 1); |
99e314eb | 2890 | } else { |
6b6712ef | 2891 | run_end = find_next_zero_bit(bitmap, pages, run_start + 1); |
99e314eb DDAG |
2892 | } |
2893 | /* | |
2894 | * If the end isn't at the start of a host page, then the | |
2895 | * run doesn't finish at the end of a host page | |
2896 | * and we need to discard. | |
2897 | */ | |
2898 | host_offset = run_end % host_ratio; | |
2899 | if (host_offset) { | |
2900 | do_fixup = true; | |
2901 | fixup_start_addr = run_end - host_offset; | |
2902 | /* | |
2903 | * This host page has gone, the next loop iteration starts | |
2904 | * from after the fixup | |
2905 | */ | |
2906 | run_start = fixup_start_addr + host_ratio; | |
2907 | } else { | |
2908 | /* | |
2909 | * No discards on this iteration, next loop starts from | |
2910 | * next sent/dirty page | |
2911 | */ | |
2912 | run_start = run_end + 1; | |
2913 | } | |
2914 | } | |
2915 | ||
2916 | if (do_fixup) { | |
2917 | unsigned long page; | |
2918 | ||
2919 | /* Tell the destination to discard this page */ | |
2920 | if (unsent_pass || !test_bit(fixup_start_addr, unsentmap)) { | |
2921 | /* For the unsent_pass we: | |
2922 | * discard partially sent pages | |
2923 | * For the !unsent_pass (dirty) we: | |
2924 | * discard partially dirty pages that were sent | |
2925 | * (any partially sent pages were already discarded | |
2926 | * by the previous unsent_pass) | |
2927 | */ | |
2928 | postcopy_discard_send_range(ms, pds, fixup_start_addr, | |
2929 | host_ratio); | |
2930 | } | |
2931 | ||
2932 | /* Clean up the bitmap */ | |
2933 | for (page = fixup_start_addr; | |
2934 | page < fixup_start_addr + host_ratio; page++) { | |
2935 | /* All pages in this host page are now not sent */ | |
2936 | set_bit(page, unsentmap); | |
2937 | ||
2938 | /* | |
2939 | * Remark them as dirty, updating the count for any pages | |
2940 | * that weren't previously dirty. | |
2941 | */ | |
0d8ec885 | 2942 | rs->migration_dirty_pages += !test_and_set_bit(page, bitmap); |
99e314eb DDAG |
2943 | } |
2944 | } | |
2945 | ||
2946 | if (unsent_pass) { | |
2947 | /* Find the next sent page for the next iteration */ | |
6b6712ef | 2948 | run_start = find_next_zero_bit(unsentmap, pages, run_start); |
99e314eb DDAG |
2949 | } else { |
2950 | /* Find the next dirty page for the next iteration */ | |
6b6712ef | 2951 | run_start = find_next_bit(bitmap, pages, run_start); |
99e314eb DDAG |
2952 | } |
2953 | } | |
2954 | } | |
2955 | ||
3d0684b2 JQ |
2956 | /** |
2957 | * postcopy_chuck_hostpages: discrad any partially sent host page | |
2958 | * | |
99e314eb DDAG |
2959 | * Utility for the outgoing postcopy code. |
2960 | * | |
2961 | * Discard any partially sent host-page size chunks, mark any partially | |
29c59172 DDAG |
2962 | * dirty host-page size chunks as all dirty. In this case the host-page |
2963 | * is the host-page for the particular RAMBlock, i.e. it might be a huge page | |
99e314eb | 2964 | * |
3d0684b2 JQ |
2965 | * Returns zero on success |
2966 | * | |
2967 | * @ms: current migration state | |
6b6712ef | 2968 | * @block: block we want to work with |
99e314eb | 2969 | */ |
6b6712ef | 2970 | static int postcopy_chunk_hostpages(MigrationState *ms, RAMBlock *block) |
99e314eb | 2971 | { |
6b6712ef JQ |
2972 | PostcopyDiscardState *pds = |
2973 | postcopy_discard_send_init(ms, block->idstr); | |
99e314eb | 2974 | |
6b6712ef JQ |
2975 | /* First pass: Discard all partially sent host pages */ |
2976 | postcopy_chunk_hostpages_pass(ms, true, block, pds); | |
2977 | /* | |
2978 | * Second pass: Ensure that all partially dirty host pages are made | |
2979 | * fully dirty. | |
2980 | */ | |
2981 | postcopy_chunk_hostpages_pass(ms, false, block, pds); | |
99e314eb | 2982 | |
6b6712ef | 2983 | postcopy_discard_send_finish(ms, pds); |
99e314eb DDAG |
2984 | return 0; |
2985 | } | |
2986 | ||
3d0684b2 JQ |
2987 | /** |
2988 | * ram_postcopy_send_discard_bitmap: transmit the discard bitmap | |
2989 | * | |
2990 | * Returns zero on success | |
2991 | * | |
e0b266f0 DDAG |
2992 | * Transmit the set of pages to be discarded after precopy to the target |
2993 | * these are pages that: | |
2994 | * a) Have been previously transmitted but are now dirty again | |
2995 | * b) Pages that have never been transmitted, this ensures that | |
2996 | * any pages on the destination that have been mapped by background | |
2997 | * tasks get discarded (transparent huge pages is the specific concern) | |
2998 | * Hopefully this is pretty sparse | |
3d0684b2 JQ |
2999 | * |
3000 | * @ms: current migration state | |
e0b266f0 DDAG |
3001 | */ |
3002 | int ram_postcopy_send_discard_bitmap(MigrationState *ms) | |
3003 | { | |
53518d94 | 3004 | RAMState *rs = ram_state; |
6b6712ef | 3005 | RAMBlock *block; |
e0b266f0 | 3006 | int ret; |
e0b266f0 DDAG |
3007 | |
3008 | rcu_read_lock(); | |
3009 | ||
3010 | /* This should be our last sync, the src is now paused */ | |
eb859c53 | 3011 | migration_bitmap_sync(rs); |
e0b266f0 | 3012 | |
6b6712ef JQ |
3013 | /* Easiest way to make sure we don't resume in the middle of a host-page */ |
3014 | rs->last_seen_block = NULL; | |
3015 | rs->last_sent_block = NULL; | |
3016 | rs->last_page = 0; | |
e0b266f0 | 3017 | |
fbd162e6 | 3018 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
6b6712ef JQ |
3019 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
3020 | unsigned long *bitmap = block->bmap; | |
3021 | unsigned long *unsentmap = block->unsentmap; | |
3022 | ||
3023 | if (!unsentmap) { | |
3024 | /* We don't have a safe way to resize the sentmap, so | |
3025 | * if the bitmap was resized it will be NULL at this | |
3026 | * point. | |
3027 | */ | |
3028 | error_report("migration ram resized during precopy phase"); | |
3029 | rcu_read_unlock(); | |
3030 | return -EINVAL; | |
3031 | } | |
3032 | /* Deal with TPS != HPS and huge pages */ | |
3033 | ret = postcopy_chunk_hostpages(ms, block); | |
3034 | if (ret) { | |
3035 | rcu_read_unlock(); | |
3036 | return ret; | |
3037 | } | |
e0b266f0 | 3038 | |
6b6712ef JQ |
3039 | /* |
3040 | * Update the unsentmap to be unsentmap = unsentmap | dirty | |
3041 | */ | |
3042 | bitmap_or(unsentmap, unsentmap, bitmap, pages); | |
e0b266f0 | 3043 | #ifdef DEBUG_POSTCOPY |
6b6712ef | 3044 | ram_debug_dump_bitmap(unsentmap, true, pages); |
e0b266f0 | 3045 | #endif |
6b6712ef JQ |
3046 | } |
3047 | trace_ram_postcopy_send_discard_bitmap(); | |
e0b266f0 DDAG |
3048 | |
3049 | ret = postcopy_each_ram_send_discard(ms); | |
3050 | rcu_read_unlock(); | |
3051 | ||
3052 | return ret; | |
3053 | } | |
3054 | ||
3d0684b2 JQ |
3055 | /** |
3056 | * ram_discard_range: discard dirtied pages at the beginning of postcopy | |
e0b266f0 | 3057 | * |
3d0684b2 | 3058 | * Returns zero on success |
e0b266f0 | 3059 | * |
36449157 JQ |
3060 | * @rbname: name of the RAMBlock of the request. NULL means the |
3061 | * same that last one. | |
3d0684b2 JQ |
3062 | * @start: RAMBlock starting page |
3063 | * @length: RAMBlock size | |
e0b266f0 | 3064 | */ |
aaa2064c | 3065 | int ram_discard_range(const char *rbname, uint64_t start, size_t length) |
e0b266f0 DDAG |
3066 | { |
3067 | int ret = -1; | |
3068 | ||
36449157 | 3069 | trace_ram_discard_range(rbname, start, length); |
d3a5038c | 3070 | |
e0b266f0 | 3071 | rcu_read_lock(); |
36449157 | 3072 | RAMBlock *rb = qemu_ram_block_by_name(rbname); |
e0b266f0 DDAG |
3073 | |
3074 | if (!rb) { | |
36449157 | 3075 | error_report("ram_discard_range: Failed to find block '%s'", rbname); |
e0b266f0 DDAG |
3076 | goto err; |
3077 | } | |
3078 | ||
814bb08f PX |
3079 | /* |
3080 | * On source VM, we don't need to update the received bitmap since | |
3081 | * we don't even have one. | |
3082 | */ | |
3083 | if (rb->receivedmap) { | |
3084 | bitmap_clear(rb->receivedmap, start >> qemu_target_page_bits(), | |
3085 | length >> qemu_target_page_bits()); | |
3086 | } | |
3087 | ||
d3a5038c | 3088 | ret = ram_block_discard_range(rb, start, length); |
e0b266f0 DDAG |
3089 | |
3090 | err: | |
3091 | rcu_read_unlock(); | |
3092 | ||
3093 | return ret; | |
3094 | } | |
3095 | ||
84593a08 PX |
3096 | /* |
3097 | * For every allocation, we will try not to crash the VM if the | |
3098 | * allocation failed. | |
3099 | */ | |
3100 | static int xbzrle_init(void) | |
3101 | { | |
3102 | Error *local_err = NULL; | |
3103 | ||
3104 | if (!migrate_use_xbzrle()) { | |
3105 | return 0; | |
3106 | } | |
3107 | ||
3108 | XBZRLE_cache_lock(); | |
3109 | ||
3110 | XBZRLE.zero_target_page = g_try_malloc0(TARGET_PAGE_SIZE); | |
3111 | if (!XBZRLE.zero_target_page) { | |
3112 | error_report("%s: Error allocating zero page", __func__); | |
3113 | goto err_out; | |
3114 | } | |
3115 | ||
3116 | XBZRLE.cache = cache_init(migrate_xbzrle_cache_size(), | |
3117 | TARGET_PAGE_SIZE, &local_err); | |
3118 | if (!XBZRLE.cache) { | |
3119 | error_report_err(local_err); | |
3120 | goto free_zero_page; | |
3121 | } | |
3122 | ||
3123 | XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); | |
3124 | if (!XBZRLE.encoded_buf) { | |
3125 | error_report("%s: Error allocating encoded_buf", __func__); | |
3126 | goto free_cache; | |
3127 | } | |
3128 | ||
3129 | XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); | |
3130 | if (!XBZRLE.current_buf) { | |
3131 | error_report("%s: Error allocating current_buf", __func__); | |
3132 | goto free_encoded_buf; | |
3133 | } | |
3134 | ||
3135 | /* We are all good */ | |
3136 | XBZRLE_cache_unlock(); | |
3137 | return 0; | |
3138 | ||
3139 | free_encoded_buf: | |
3140 | g_free(XBZRLE.encoded_buf); | |
3141 | XBZRLE.encoded_buf = NULL; | |
3142 | free_cache: | |
3143 | cache_fini(XBZRLE.cache); | |
3144 | XBZRLE.cache = NULL; | |
3145 | free_zero_page: | |
3146 | g_free(XBZRLE.zero_target_page); | |
3147 | XBZRLE.zero_target_page = NULL; | |
3148 | err_out: | |
3149 | XBZRLE_cache_unlock(); | |
3150 | return -ENOMEM; | |
3151 | } | |
3152 | ||
53518d94 | 3153 | static int ram_state_init(RAMState **rsp) |
56e93d26 | 3154 | { |
7d00ee6a PX |
3155 | *rsp = g_try_new0(RAMState, 1); |
3156 | ||
3157 | if (!*rsp) { | |
3158 | error_report("%s: Init ramstate fail", __func__); | |
3159 | return -1; | |
3160 | } | |
53518d94 JQ |
3161 | |
3162 | qemu_mutex_init(&(*rsp)->bitmap_mutex); | |
3163 | qemu_mutex_init(&(*rsp)->src_page_req_mutex); | |
3164 | QSIMPLEQ_INIT(&(*rsp)->src_page_requests); | |
56e93d26 | 3165 | |
7d00ee6a PX |
3166 | /* |
3167 | * Count the total number of pages used by ram blocks not including any | |
3168 | * gaps due to alignment or unplugs. | |
3169 | */ | |
3170 | (*rsp)->migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS; | |
3171 | ||
3172 | ram_state_reset(*rsp); | |
3173 | ||
3174 | return 0; | |
3175 | } | |
3176 | ||
d6eff5d7 | 3177 | static void ram_list_init_bitmaps(void) |
7d00ee6a | 3178 | { |
d6eff5d7 PX |
3179 | RAMBlock *block; |
3180 | unsigned long pages; | |
56e93d26 | 3181 | |
0827b9e9 AA |
3182 | /* Skip setting bitmap if there is no RAM */ |
3183 | if (ram_bytes_total()) { | |
fbd162e6 | 3184 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
d6eff5d7 | 3185 | pages = block->max_length >> TARGET_PAGE_BITS; |
6b6712ef JQ |
3186 | block->bmap = bitmap_new(pages); |
3187 | bitmap_set(block->bmap, 0, pages); | |
3188 | if (migrate_postcopy_ram()) { | |
3189 | block->unsentmap = bitmap_new(pages); | |
3190 | bitmap_set(block->unsentmap, 0, pages); | |
3191 | } | |
0827b9e9 | 3192 | } |
f3f491fc | 3193 | } |
d6eff5d7 PX |
3194 | } |
3195 | ||
3196 | static void ram_init_bitmaps(RAMState *rs) | |
3197 | { | |
3198 | /* For memory_global_dirty_log_start below. */ | |
3199 | qemu_mutex_lock_iothread(); | |
3200 | qemu_mutex_lock_ramlist(); | |
3201 | rcu_read_lock(); | |
f3f491fc | 3202 | |
d6eff5d7 | 3203 | ram_list_init_bitmaps(); |
56e93d26 | 3204 | memory_global_dirty_log_start(); |
bd227060 | 3205 | migration_bitmap_sync_precopy(rs); |
d6eff5d7 PX |
3206 | |
3207 | rcu_read_unlock(); | |
56e93d26 | 3208 | qemu_mutex_unlock_ramlist(); |
49877834 | 3209 | qemu_mutex_unlock_iothread(); |
d6eff5d7 PX |
3210 | } |
3211 | ||
3212 | static int ram_init_all(RAMState **rsp) | |
3213 | { | |
3214 | if (ram_state_init(rsp)) { | |
3215 | return -1; | |
3216 | } | |
3217 | ||
3218 | if (xbzrle_init()) { | |
3219 | ram_state_cleanup(rsp); | |
3220 | return -1; | |
3221 | } | |
3222 | ||
3223 | ram_init_bitmaps(*rsp); | |
a91246c9 HZ |
3224 | |
3225 | return 0; | |
3226 | } | |
3227 | ||
08614f34 PX |
3228 | static void ram_state_resume_prepare(RAMState *rs, QEMUFile *out) |
3229 | { | |
3230 | RAMBlock *block; | |
3231 | uint64_t pages = 0; | |
3232 | ||
3233 | /* | |
3234 | * Postcopy is not using xbzrle/compression, so no need for that. | |
3235 | * Also, since source are already halted, we don't need to care | |
3236 | * about dirty page logging as well. | |
3237 | */ | |
3238 | ||
fbd162e6 | 3239 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
08614f34 PX |
3240 | pages += bitmap_count_one(block->bmap, |
3241 | block->used_length >> TARGET_PAGE_BITS); | |
3242 | } | |
3243 | ||
3244 | /* This may not be aligned with current bitmaps. Recalculate. */ | |
3245 | rs->migration_dirty_pages = pages; | |
3246 | ||
3247 | rs->last_seen_block = NULL; | |
3248 | rs->last_sent_block = NULL; | |
3249 | rs->last_page = 0; | |
3250 | rs->last_version = ram_list.version; | |
3251 | /* | |
3252 | * Disable the bulk stage, otherwise we'll resend the whole RAM no | |
3253 | * matter what we have sent. | |
3254 | */ | |
3255 | rs->ram_bulk_stage = false; | |
3256 | ||
3257 | /* Update RAMState cache of output QEMUFile */ | |
3258 | rs->f = out; | |
3259 | ||
3260 | trace_ram_state_resume_prepare(pages); | |
3261 | } | |
3262 | ||
6bcb05fc WW |
3263 | /* |
3264 | * This function clears bits of the free pages reported by the caller from the | |
3265 | * migration dirty bitmap. @addr is the host address corresponding to the | |
3266 | * start of the continuous guest free pages, and @len is the total bytes of | |
3267 | * those pages. | |
3268 | */ | |
3269 | void qemu_guest_free_page_hint(void *addr, size_t len) | |
3270 | { | |
3271 | RAMBlock *block; | |
3272 | ram_addr_t offset; | |
3273 | size_t used_len, start, npages; | |
3274 | MigrationState *s = migrate_get_current(); | |
3275 | ||
3276 | /* This function is currently expected to be used during live migration */ | |
3277 | if (!migration_is_setup_or_active(s->state)) { | |
3278 | return; | |
3279 | } | |
3280 | ||
3281 | for (; len > 0; len -= used_len, addr += used_len) { | |
3282 | block = qemu_ram_block_from_host(addr, false, &offset); | |
3283 | if (unlikely(!block || offset >= block->used_length)) { | |
3284 | /* | |
3285 | * The implementation might not support RAMBlock resize during | |
3286 | * live migration, but it could happen in theory with future | |
3287 | * updates. So we add a check here to capture that case. | |
3288 | */ | |
3289 | error_report_once("%s unexpected error", __func__); | |
3290 | return; | |
3291 | } | |
3292 | ||
3293 | if (len <= block->used_length - offset) { | |
3294 | used_len = len; | |
3295 | } else { | |
3296 | used_len = block->used_length - offset; | |
3297 | } | |
3298 | ||
3299 | start = offset >> TARGET_PAGE_BITS; | |
3300 | npages = used_len >> TARGET_PAGE_BITS; | |
3301 | ||
3302 | qemu_mutex_lock(&ram_state->bitmap_mutex); | |
3303 | ram_state->migration_dirty_pages -= | |
3304 | bitmap_count_one_with_offset(block->bmap, start, npages); | |
3305 | bitmap_clear(block->bmap, start, npages); | |
3306 | qemu_mutex_unlock(&ram_state->bitmap_mutex); | |
3307 | } | |
3308 | } | |
3309 | ||
3d0684b2 JQ |
3310 | /* |
3311 | * Each of ram_save_setup, ram_save_iterate and ram_save_complete has | |
a91246c9 HZ |
3312 | * long-running RCU critical section. When rcu-reclaims in the code |
3313 | * start to become numerous it will be necessary to reduce the | |
3314 | * granularity of these critical sections. | |
3315 | */ | |
3316 | ||
3d0684b2 JQ |
3317 | /** |
3318 | * ram_save_setup: Setup RAM for migration | |
3319 | * | |
3320 | * Returns zero to indicate success and negative for error | |
3321 | * | |
3322 | * @f: QEMUFile where to send the data | |
3323 | * @opaque: RAMState pointer | |
3324 | */ | |
a91246c9 HZ |
3325 | static int ram_save_setup(QEMUFile *f, void *opaque) |
3326 | { | |
53518d94 | 3327 | RAMState **rsp = opaque; |
a91246c9 HZ |
3328 | RAMBlock *block; |
3329 | ||
dcaf446e XG |
3330 | if (compress_threads_save_setup()) { |
3331 | return -1; | |
3332 | } | |
3333 | ||
a91246c9 HZ |
3334 | /* migration has already setup the bitmap, reuse it. */ |
3335 | if (!migration_in_colo_state()) { | |
7d00ee6a | 3336 | if (ram_init_all(rsp) != 0) { |
dcaf446e | 3337 | compress_threads_save_cleanup(); |
a91246c9 | 3338 | return -1; |
53518d94 | 3339 | } |
a91246c9 | 3340 | } |
53518d94 | 3341 | (*rsp)->f = f; |
a91246c9 HZ |
3342 | |
3343 | rcu_read_lock(); | |
56e93d26 | 3344 | |
fbd162e6 | 3345 | qemu_put_be64(f, ram_bytes_total_common(true) | RAM_SAVE_FLAG_MEM_SIZE); |
56e93d26 | 3346 | |
b895de50 | 3347 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
56e93d26 JQ |
3348 | qemu_put_byte(f, strlen(block->idstr)); |
3349 | qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); | |
3350 | qemu_put_be64(f, block->used_length); | |
ef08fb38 DDAG |
3351 | if (migrate_postcopy_ram() && block->page_size != qemu_host_page_size) { |
3352 | qemu_put_be64(f, block->page_size); | |
3353 | } | |
fbd162e6 YK |
3354 | if (migrate_ignore_shared()) { |
3355 | qemu_put_be64(f, block->mr->addr); | |
3356 | qemu_put_byte(f, ramblock_is_ignored(block) ? 1 : 0); | |
3357 | } | |
56e93d26 JQ |
3358 | } |
3359 | ||
3360 | rcu_read_unlock(); | |
3361 | ||
3362 | ram_control_before_iterate(f, RAM_CONTROL_SETUP); | |
3363 | ram_control_after_iterate(f, RAM_CONTROL_SETUP); | |
3364 | ||
6df264ac | 3365 | multifd_send_sync_main(); |
56e93d26 | 3366 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
35374cbd | 3367 | qemu_fflush(f); |
56e93d26 JQ |
3368 | |
3369 | return 0; | |
3370 | } | |
3371 | ||
3d0684b2 JQ |
3372 | /** |
3373 | * ram_save_iterate: iterative stage for migration | |
3374 | * | |
3375 | * Returns zero to indicate success and negative for error | |
3376 | * | |
3377 | * @f: QEMUFile where to send the data | |
3378 | * @opaque: RAMState pointer | |
3379 | */ | |
56e93d26 JQ |
3380 | static int ram_save_iterate(QEMUFile *f, void *opaque) |
3381 | { | |
53518d94 JQ |
3382 | RAMState **temp = opaque; |
3383 | RAMState *rs = *temp; | |
56e93d26 JQ |
3384 | int ret; |
3385 | int i; | |
3386 | int64_t t0; | |
5c90308f | 3387 | int done = 0; |
56e93d26 | 3388 | |
b2557345 PL |
3389 | if (blk_mig_bulk_active()) { |
3390 | /* Avoid transferring ram during bulk phase of block migration as | |
3391 | * the bulk phase will usually take a long time and transferring | |
3392 | * ram updates during that time is pointless. */ | |
3393 | goto out; | |
3394 | } | |
3395 | ||
56e93d26 | 3396 | rcu_read_lock(); |
6f37bb8b JQ |
3397 | if (ram_list.version != rs->last_version) { |
3398 | ram_state_reset(rs); | |
56e93d26 JQ |
3399 | } |
3400 | ||
3401 | /* Read version before ram_list.blocks */ | |
3402 | smp_rmb(); | |
3403 | ||
3404 | ram_control_before_iterate(f, RAM_CONTROL_ROUND); | |
3405 | ||
3406 | t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
3407 | i = 0; | |
e03a34f8 DDAG |
3408 | while ((ret = qemu_file_rate_limit(f)) == 0 || |
3409 | !QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
56e93d26 JQ |
3410 | int pages; |
3411 | ||
e03a34f8 DDAG |
3412 | if (qemu_file_get_error(f)) { |
3413 | break; | |
3414 | } | |
3415 | ||
ce25d337 | 3416 | pages = ram_find_and_save_block(rs, false); |
56e93d26 JQ |
3417 | /* no more pages to sent */ |
3418 | if (pages == 0) { | |
5c90308f | 3419 | done = 1; |
56e93d26 JQ |
3420 | break; |
3421 | } | |
e8f3735f XG |
3422 | |
3423 | if (pages < 0) { | |
3424 | qemu_file_set_error(f, pages); | |
3425 | break; | |
3426 | } | |
3427 | ||
be8b02ed | 3428 | rs->target_page_count += pages; |
070afca2 | 3429 | |
56e93d26 JQ |
3430 | /* we want to check in the 1st loop, just in case it was the 1st time |
3431 | and we had to sync the dirty bitmap. | |
3432 | qemu_get_clock_ns() is a bit expensive, so we only check each some | |
3433 | iterations | |
3434 | */ | |
3435 | if ((i & 63) == 0) { | |
3436 | uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000; | |
3437 | if (t1 > MAX_WAIT) { | |
55c4446b | 3438 | trace_ram_save_iterate_big_wait(t1, i); |
56e93d26 JQ |
3439 | break; |
3440 | } | |
3441 | } | |
3442 | i++; | |
3443 | } | |
56e93d26 JQ |
3444 | rcu_read_unlock(); |
3445 | ||
3446 | /* | |
3447 | * Must occur before EOS (or any QEMUFile operation) | |
3448 | * because of RDMA protocol. | |
3449 | */ | |
3450 | ram_control_after_iterate(f, RAM_CONTROL_ROUND); | |
3451 | ||
6df264ac | 3452 | multifd_send_sync_main(); |
b2557345 | 3453 | out: |
56e93d26 | 3454 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
35374cbd | 3455 | qemu_fflush(f); |
9360447d | 3456 | ram_counters.transferred += 8; |
56e93d26 JQ |
3457 | |
3458 | ret = qemu_file_get_error(f); | |
3459 | if (ret < 0) { | |
3460 | return ret; | |
3461 | } | |
3462 | ||
5c90308f | 3463 | return done; |
56e93d26 JQ |
3464 | } |
3465 | ||
3d0684b2 JQ |
3466 | /** |
3467 | * ram_save_complete: function called to send the remaining amount of ram | |
3468 | * | |
e8f3735f | 3469 | * Returns zero to indicate success or negative on error |
3d0684b2 JQ |
3470 | * |
3471 | * Called with iothread lock | |
3472 | * | |
3473 | * @f: QEMUFile where to send the data | |
3474 | * @opaque: RAMState pointer | |
3475 | */ | |
56e93d26 JQ |
3476 | static int ram_save_complete(QEMUFile *f, void *opaque) |
3477 | { | |
53518d94 JQ |
3478 | RAMState **temp = opaque; |
3479 | RAMState *rs = *temp; | |
e8f3735f | 3480 | int ret = 0; |
6f37bb8b | 3481 | |
56e93d26 JQ |
3482 | rcu_read_lock(); |
3483 | ||
5727309d | 3484 | if (!migration_in_postcopy()) { |
bd227060 | 3485 | migration_bitmap_sync_precopy(rs); |
663e6c1d | 3486 | } |
56e93d26 JQ |
3487 | |
3488 | ram_control_before_iterate(f, RAM_CONTROL_FINISH); | |
3489 | ||
3490 | /* try transferring iterative blocks of memory */ | |
3491 | ||
3492 | /* flush all remaining blocks regardless of rate limiting */ | |
3493 | while (true) { | |
3494 | int pages; | |
3495 | ||
ce25d337 | 3496 | pages = ram_find_and_save_block(rs, !migration_in_colo_state()); |
56e93d26 JQ |
3497 | /* no more blocks to sent */ |
3498 | if (pages == 0) { | |
3499 | break; | |
3500 | } | |
e8f3735f XG |
3501 | if (pages < 0) { |
3502 | ret = pages; | |
3503 | break; | |
3504 | } | |
56e93d26 JQ |
3505 | } |
3506 | ||
ce25d337 | 3507 | flush_compressed_data(rs); |
56e93d26 | 3508 | ram_control_after_iterate(f, RAM_CONTROL_FINISH); |
56e93d26 JQ |
3509 | |
3510 | rcu_read_unlock(); | |
d09a6fde | 3511 | |
6df264ac | 3512 | multifd_send_sync_main(); |
56e93d26 | 3513 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
35374cbd | 3514 | qemu_fflush(f); |
56e93d26 | 3515 | |
e8f3735f | 3516 | return ret; |
56e93d26 JQ |
3517 | } |
3518 | ||
c31b098f | 3519 | static void ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size, |
47995026 VSO |
3520 | uint64_t *res_precopy_only, |
3521 | uint64_t *res_compatible, | |
3522 | uint64_t *res_postcopy_only) | |
56e93d26 | 3523 | { |
53518d94 JQ |
3524 | RAMState **temp = opaque; |
3525 | RAMState *rs = *temp; | |
56e93d26 JQ |
3526 | uint64_t remaining_size; |
3527 | ||
9edabd4d | 3528 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 3529 | |
5727309d | 3530 | if (!migration_in_postcopy() && |
663e6c1d | 3531 | remaining_size < max_size) { |
56e93d26 JQ |
3532 | qemu_mutex_lock_iothread(); |
3533 | rcu_read_lock(); | |
bd227060 | 3534 | migration_bitmap_sync_precopy(rs); |
56e93d26 JQ |
3535 | rcu_read_unlock(); |
3536 | qemu_mutex_unlock_iothread(); | |
9edabd4d | 3537 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 3538 | } |
c31b098f | 3539 | |
86e1167e VSO |
3540 | if (migrate_postcopy_ram()) { |
3541 | /* We can do postcopy, and all the data is postcopiable */ | |
47995026 | 3542 | *res_compatible += remaining_size; |
86e1167e | 3543 | } else { |
47995026 | 3544 | *res_precopy_only += remaining_size; |
86e1167e | 3545 | } |
56e93d26 JQ |
3546 | } |
3547 | ||
3548 | static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) | |
3549 | { | |
3550 | unsigned int xh_len; | |
3551 | int xh_flags; | |
063e760a | 3552 | uint8_t *loaded_data; |
56e93d26 | 3553 | |
56e93d26 JQ |
3554 | /* extract RLE header */ |
3555 | xh_flags = qemu_get_byte(f); | |
3556 | xh_len = qemu_get_be16(f); | |
3557 | ||
3558 | if (xh_flags != ENCODING_FLAG_XBZRLE) { | |
3559 | error_report("Failed to load XBZRLE page - wrong compression!"); | |
3560 | return -1; | |
3561 | } | |
3562 | ||
3563 | if (xh_len > TARGET_PAGE_SIZE) { | |
3564 | error_report("Failed to load XBZRLE page - len overflow!"); | |
3565 | return -1; | |
3566 | } | |
f265e0e4 | 3567 | loaded_data = XBZRLE.decoded_buf; |
56e93d26 | 3568 | /* load data and decode */ |
f265e0e4 | 3569 | /* it can change loaded_data to point to an internal buffer */ |
063e760a | 3570 | qemu_get_buffer_in_place(f, &loaded_data, xh_len); |
56e93d26 JQ |
3571 | |
3572 | /* decode RLE */ | |
063e760a | 3573 | if (xbzrle_decode_buffer(loaded_data, xh_len, host, |
56e93d26 JQ |
3574 | TARGET_PAGE_SIZE) == -1) { |
3575 | error_report("Failed to load XBZRLE page - decode error!"); | |
3576 | return -1; | |
3577 | } | |
3578 | ||
3579 | return 0; | |
3580 | } | |
3581 | ||
3d0684b2 JQ |
3582 | /** |
3583 | * ram_block_from_stream: read a RAMBlock id from the migration stream | |
3584 | * | |
3585 | * Must be called from within a rcu critical section. | |
3586 | * | |
56e93d26 | 3587 | * Returns a pointer from within the RCU-protected ram_list. |
a7180877 | 3588 | * |
3d0684b2 JQ |
3589 | * @f: QEMUFile where to read the data from |
3590 | * @flags: Page flags (mostly to see if it's a continuation of previous block) | |
a7180877 | 3591 | */ |
3d0684b2 | 3592 | static inline RAMBlock *ram_block_from_stream(QEMUFile *f, int flags) |
56e93d26 JQ |
3593 | { |
3594 | static RAMBlock *block = NULL; | |
3595 | char id[256]; | |
3596 | uint8_t len; | |
3597 | ||
3598 | if (flags & RAM_SAVE_FLAG_CONTINUE) { | |
4c4bad48 | 3599 | if (!block) { |
56e93d26 JQ |
3600 | error_report("Ack, bad migration stream!"); |
3601 | return NULL; | |
3602 | } | |
4c4bad48 | 3603 | return block; |
56e93d26 JQ |
3604 | } |
3605 | ||
3606 | len = qemu_get_byte(f); | |
3607 | qemu_get_buffer(f, (uint8_t *)id, len); | |
3608 | id[len] = 0; | |
3609 | ||
e3dd7493 | 3610 | block = qemu_ram_block_by_name(id); |
4c4bad48 HZ |
3611 | if (!block) { |
3612 | error_report("Can't find block %s", id); | |
3613 | return NULL; | |
56e93d26 JQ |
3614 | } |
3615 | ||
fbd162e6 | 3616 | if (ramblock_is_ignored(block)) { |
b895de50 CLG |
3617 | error_report("block %s should not be migrated !", id); |
3618 | return NULL; | |
3619 | } | |
3620 | ||
4c4bad48 HZ |
3621 | return block; |
3622 | } | |
3623 | ||
3624 | static inline void *host_from_ram_block_offset(RAMBlock *block, | |
3625 | ram_addr_t offset) | |
3626 | { | |
3627 | if (!offset_in_ramblock(block, offset)) { | |
3628 | return NULL; | |
3629 | } | |
3630 | ||
3631 | return block->host + offset; | |
56e93d26 JQ |
3632 | } |
3633 | ||
13af18f2 ZC |
3634 | static inline void *colo_cache_from_block_offset(RAMBlock *block, |
3635 | ram_addr_t offset) | |
3636 | { | |
3637 | if (!offset_in_ramblock(block, offset)) { | |
3638 | return NULL; | |
3639 | } | |
3640 | if (!block->colo_cache) { | |
3641 | error_report("%s: colo_cache is NULL in block :%s", | |
3642 | __func__, block->idstr); | |
3643 | return NULL; | |
3644 | } | |
7d9acafa ZC |
3645 | |
3646 | /* | |
3647 | * During colo checkpoint, we need bitmap of these migrated pages. | |
3648 | * It help us to decide which pages in ram cache should be flushed | |
3649 | * into VM's RAM later. | |
3650 | */ | |
3651 | if (!test_and_set_bit(offset >> TARGET_PAGE_BITS, block->bmap)) { | |
3652 | ram_state->migration_dirty_pages++; | |
3653 | } | |
13af18f2 ZC |
3654 | return block->colo_cache + offset; |
3655 | } | |
3656 | ||
3d0684b2 JQ |
3657 | /** |
3658 | * ram_handle_compressed: handle the zero page case | |
3659 | * | |
56e93d26 JQ |
3660 | * If a page (or a whole RDMA chunk) has been |
3661 | * determined to be zero, then zap it. | |
3d0684b2 JQ |
3662 | * |
3663 | * @host: host address for the zero page | |
3664 | * @ch: what the page is filled from. We only support zero | |
3665 | * @size: size of the zero page | |
56e93d26 JQ |
3666 | */ |
3667 | void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) | |
3668 | { | |
3669 | if (ch != 0 || !is_zero_range(host, size)) { | |
3670 | memset(host, ch, size); | |
3671 | } | |
3672 | } | |
3673 | ||
797ca154 XG |
3674 | /* return the size after decompression, or negative value on error */ |
3675 | static int | |
3676 | qemu_uncompress_data(z_stream *stream, uint8_t *dest, size_t dest_len, | |
3677 | const uint8_t *source, size_t source_len) | |
3678 | { | |
3679 | int err; | |
3680 | ||
3681 | err = inflateReset(stream); | |
3682 | if (err != Z_OK) { | |
3683 | return -1; | |
3684 | } | |
3685 | ||
3686 | stream->avail_in = source_len; | |
3687 | stream->next_in = (uint8_t *)source; | |
3688 | stream->avail_out = dest_len; | |
3689 | stream->next_out = dest; | |
3690 | ||
3691 | err = inflate(stream, Z_NO_FLUSH); | |
3692 | if (err != Z_STREAM_END) { | |
3693 | return -1; | |
3694 | } | |
3695 | ||
3696 | return stream->total_out; | |
3697 | } | |
3698 | ||
56e93d26 JQ |
3699 | static void *do_data_decompress(void *opaque) |
3700 | { | |
3701 | DecompressParam *param = opaque; | |
3702 | unsigned long pagesize; | |
33d151f4 | 3703 | uint8_t *des; |
34ab9e97 | 3704 | int len, ret; |
56e93d26 | 3705 | |
33d151f4 | 3706 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 3707 | while (!param->quit) { |
33d151f4 LL |
3708 | if (param->des) { |
3709 | des = param->des; | |
3710 | len = param->len; | |
3711 | param->des = 0; | |
3712 | qemu_mutex_unlock(¶m->mutex); | |
3713 | ||
56e93d26 | 3714 | pagesize = TARGET_PAGE_SIZE; |
34ab9e97 XG |
3715 | |
3716 | ret = qemu_uncompress_data(¶m->stream, des, pagesize, | |
3717 | param->compbuf, len); | |
f548222c | 3718 | if (ret < 0 && migrate_get_current()->decompress_error_check) { |
34ab9e97 XG |
3719 | error_report("decompress data failed"); |
3720 | qemu_file_set_error(decomp_file, ret); | |
3721 | } | |
73a8912b | 3722 | |
33d151f4 LL |
3723 | qemu_mutex_lock(&decomp_done_lock); |
3724 | param->done = true; | |
3725 | qemu_cond_signal(&decomp_done_cond); | |
3726 | qemu_mutex_unlock(&decomp_done_lock); | |
3727 | ||
3728 | qemu_mutex_lock(¶m->mutex); | |
3729 | } else { | |
3730 | qemu_cond_wait(¶m->cond, ¶m->mutex); | |
3731 | } | |
56e93d26 | 3732 | } |
33d151f4 | 3733 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
3734 | |
3735 | return NULL; | |
3736 | } | |
3737 | ||
34ab9e97 | 3738 | static int wait_for_decompress_done(void) |
5533b2e9 LL |
3739 | { |
3740 | int idx, thread_count; | |
3741 | ||
3742 | if (!migrate_use_compression()) { | |
34ab9e97 | 3743 | return 0; |
5533b2e9 LL |
3744 | } |
3745 | ||
3746 | thread_count = migrate_decompress_threads(); | |
3747 | qemu_mutex_lock(&decomp_done_lock); | |
3748 | for (idx = 0; idx < thread_count; idx++) { | |
3749 | while (!decomp_param[idx].done) { | |
3750 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
3751 | } | |
3752 | } | |
3753 | qemu_mutex_unlock(&decomp_done_lock); | |
34ab9e97 | 3754 | return qemu_file_get_error(decomp_file); |
5533b2e9 LL |
3755 | } |
3756 | ||
f0afa331 | 3757 | static void compress_threads_load_cleanup(void) |
56e93d26 JQ |
3758 | { |
3759 | int i, thread_count; | |
3760 | ||
3416ab5b JQ |
3761 | if (!migrate_use_compression()) { |
3762 | return; | |
3763 | } | |
56e93d26 JQ |
3764 | thread_count = migrate_decompress_threads(); |
3765 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
3766 | /* |
3767 | * we use it as a indicator which shows if the thread is | |
3768 | * properly init'd or not | |
3769 | */ | |
3770 | if (!decomp_param[i].compbuf) { | |
3771 | break; | |
3772 | } | |
3773 | ||
56e93d26 | 3774 | qemu_mutex_lock(&decomp_param[i].mutex); |
90e56fb4 | 3775 | decomp_param[i].quit = true; |
56e93d26 JQ |
3776 | qemu_cond_signal(&decomp_param[i].cond); |
3777 | qemu_mutex_unlock(&decomp_param[i].mutex); | |
3778 | } | |
3779 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
3780 | if (!decomp_param[i].compbuf) { |
3781 | break; | |
3782 | } | |
3783 | ||
56e93d26 JQ |
3784 | qemu_thread_join(decompress_threads + i); |
3785 | qemu_mutex_destroy(&decomp_param[i].mutex); | |
3786 | qemu_cond_destroy(&decomp_param[i].cond); | |
797ca154 | 3787 | inflateEnd(&decomp_param[i].stream); |
56e93d26 | 3788 | g_free(decomp_param[i].compbuf); |
797ca154 | 3789 | decomp_param[i].compbuf = NULL; |
56e93d26 JQ |
3790 | } |
3791 | g_free(decompress_threads); | |
3792 | g_free(decomp_param); | |
56e93d26 JQ |
3793 | decompress_threads = NULL; |
3794 | decomp_param = NULL; | |
34ab9e97 | 3795 | decomp_file = NULL; |
56e93d26 JQ |
3796 | } |
3797 | ||
34ab9e97 | 3798 | static int compress_threads_load_setup(QEMUFile *f) |
797ca154 XG |
3799 | { |
3800 | int i, thread_count; | |
3801 | ||
3802 | if (!migrate_use_compression()) { | |
3803 | return 0; | |
3804 | } | |
3805 | ||
3806 | thread_count = migrate_decompress_threads(); | |
3807 | decompress_threads = g_new0(QemuThread, thread_count); | |
3808 | decomp_param = g_new0(DecompressParam, thread_count); | |
3809 | qemu_mutex_init(&decomp_done_lock); | |
3810 | qemu_cond_init(&decomp_done_cond); | |
34ab9e97 | 3811 | decomp_file = f; |
797ca154 XG |
3812 | for (i = 0; i < thread_count; i++) { |
3813 | if (inflateInit(&decomp_param[i].stream) != Z_OK) { | |
3814 | goto exit; | |
3815 | } | |
3816 | ||
3817 | decomp_param[i].compbuf = g_malloc0(compressBound(TARGET_PAGE_SIZE)); | |
3818 | qemu_mutex_init(&decomp_param[i].mutex); | |
3819 | qemu_cond_init(&decomp_param[i].cond); | |
3820 | decomp_param[i].done = true; | |
3821 | decomp_param[i].quit = false; | |
3822 | qemu_thread_create(decompress_threads + i, "decompress", | |
3823 | do_data_decompress, decomp_param + i, | |
3824 | QEMU_THREAD_JOINABLE); | |
3825 | } | |
3826 | return 0; | |
3827 | exit: | |
3828 | compress_threads_load_cleanup(); | |
3829 | return -1; | |
3830 | } | |
3831 | ||
c1bc6626 | 3832 | static void decompress_data_with_multi_threads(QEMUFile *f, |
56e93d26 JQ |
3833 | void *host, int len) |
3834 | { | |
3835 | int idx, thread_count; | |
3836 | ||
3837 | thread_count = migrate_decompress_threads(); | |
73a8912b | 3838 | qemu_mutex_lock(&decomp_done_lock); |
56e93d26 JQ |
3839 | while (true) { |
3840 | for (idx = 0; idx < thread_count; idx++) { | |
73a8912b | 3841 | if (decomp_param[idx].done) { |
33d151f4 LL |
3842 | decomp_param[idx].done = false; |
3843 | qemu_mutex_lock(&decomp_param[idx].mutex); | |
c1bc6626 | 3844 | qemu_get_buffer(f, decomp_param[idx].compbuf, len); |
56e93d26 JQ |
3845 | decomp_param[idx].des = host; |
3846 | decomp_param[idx].len = len; | |
33d151f4 LL |
3847 | qemu_cond_signal(&decomp_param[idx].cond); |
3848 | qemu_mutex_unlock(&decomp_param[idx].mutex); | |
56e93d26 JQ |
3849 | break; |
3850 | } | |
3851 | } | |
3852 | if (idx < thread_count) { | |
3853 | break; | |
73a8912b LL |
3854 | } else { |
3855 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
56e93d26 JQ |
3856 | } |
3857 | } | |
73a8912b | 3858 | qemu_mutex_unlock(&decomp_done_lock); |
56e93d26 JQ |
3859 | } |
3860 | ||
13af18f2 ZC |
3861 | /* |
3862 | * colo cache: this is for secondary VM, we cache the whole | |
3863 | * memory of the secondary VM, it is need to hold the global lock | |
3864 | * to call this helper. | |
3865 | */ | |
3866 | int colo_init_ram_cache(void) | |
3867 | { | |
3868 | RAMBlock *block; | |
3869 | ||
3870 | rcu_read_lock(); | |
fbd162e6 | 3871 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
13af18f2 ZC |
3872 | block->colo_cache = qemu_anon_ram_alloc(block->used_length, |
3873 | NULL, | |
3874 | false); | |
3875 | if (!block->colo_cache) { | |
3876 | error_report("%s: Can't alloc memory for COLO cache of block %s," | |
3877 | "size 0x" RAM_ADDR_FMT, __func__, block->idstr, | |
3878 | block->used_length); | |
3879 | goto out_locked; | |
3880 | } | |
3881 | memcpy(block->colo_cache, block->host, block->used_length); | |
3882 | } | |
3883 | rcu_read_unlock(); | |
7d9acafa ZC |
3884 | /* |
3885 | * Record the dirty pages that sent by PVM, we use this dirty bitmap together | |
3886 | * with to decide which page in cache should be flushed into SVM's RAM. Here | |
3887 | * we use the same name 'ram_bitmap' as for migration. | |
3888 | */ | |
3889 | if (ram_bytes_total()) { | |
3890 | RAMBlock *block; | |
3891 | ||
fbd162e6 | 3892 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
7d9acafa ZC |
3893 | unsigned long pages = block->max_length >> TARGET_PAGE_BITS; |
3894 | ||
3895 | block->bmap = bitmap_new(pages); | |
3896 | bitmap_set(block->bmap, 0, pages); | |
3897 | } | |
3898 | } | |
3899 | ram_state = g_new0(RAMState, 1); | |
3900 | ram_state->migration_dirty_pages = 0; | |
d1955d22 | 3901 | memory_global_dirty_log_start(); |
7d9acafa | 3902 | |
13af18f2 ZC |
3903 | return 0; |
3904 | ||
3905 | out_locked: | |
7d9acafa | 3906 | |
fbd162e6 | 3907 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
13af18f2 ZC |
3908 | if (block->colo_cache) { |
3909 | qemu_anon_ram_free(block->colo_cache, block->used_length); | |
3910 | block->colo_cache = NULL; | |
3911 | } | |
3912 | } | |
3913 | ||
3914 | rcu_read_unlock(); | |
3915 | return -errno; | |
3916 | } | |
3917 | ||
3918 | /* It is need to hold the global lock to call this helper */ | |
3919 | void colo_release_ram_cache(void) | |
3920 | { | |
3921 | RAMBlock *block; | |
3922 | ||
d1955d22 | 3923 | memory_global_dirty_log_stop(); |
fbd162e6 | 3924 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
7d9acafa ZC |
3925 | g_free(block->bmap); |
3926 | block->bmap = NULL; | |
3927 | } | |
3928 | ||
13af18f2 | 3929 | rcu_read_lock(); |
7d9acafa | 3930 | |
fbd162e6 | 3931 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
13af18f2 ZC |
3932 | if (block->colo_cache) { |
3933 | qemu_anon_ram_free(block->colo_cache, block->used_length); | |
3934 | block->colo_cache = NULL; | |
3935 | } | |
3936 | } | |
7d9acafa | 3937 | |
13af18f2 | 3938 | rcu_read_unlock(); |
7d9acafa ZC |
3939 | g_free(ram_state); |
3940 | ram_state = NULL; | |
13af18f2 ZC |
3941 | } |
3942 | ||
f265e0e4 JQ |
3943 | /** |
3944 | * ram_load_setup: Setup RAM for migration incoming side | |
3945 | * | |
3946 | * Returns zero to indicate success and negative for error | |
3947 | * | |
3948 | * @f: QEMUFile where to receive the data | |
3949 | * @opaque: RAMState pointer | |
3950 | */ | |
3951 | static int ram_load_setup(QEMUFile *f, void *opaque) | |
3952 | { | |
34ab9e97 | 3953 | if (compress_threads_load_setup(f)) { |
797ca154 XG |
3954 | return -1; |
3955 | } | |
3956 | ||
f265e0e4 | 3957 | xbzrle_load_setup(); |
f9494614 | 3958 | ramblock_recv_map_init(); |
13af18f2 | 3959 | |
f265e0e4 JQ |
3960 | return 0; |
3961 | } | |
3962 | ||
3963 | static int ram_load_cleanup(void *opaque) | |
3964 | { | |
f9494614 | 3965 | RAMBlock *rb; |
56eb90af | 3966 | |
fbd162e6 | 3967 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
56eb90af JH |
3968 | if (ramblock_is_pmem(rb)) { |
3969 | pmem_persist(rb->host, rb->used_length); | |
3970 | } | |
3971 | } | |
3972 | ||
f265e0e4 | 3973 | xbzrle_load_cleanup(); |
f0afa331 | 3974 | compress_threads_load_cleanup(); |
f9494614 | 3975 | |
fbd162e6 | 3976 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
f9494614 AP |
3977 | g_free(rb->receivedmap); |
3978 | rb->receivedmap = NULL; | |
3979 | } | |
13af18f2 | 3980 | |
f265e0e4 JQ |
3981 | return 0; |
3982 | } | |
3983 | ||
3d0684b2 JQ |
3984 | /** |
3985 | * ram_postcopy_incoming_init: allocate postcopy data structures | |
3986 | * | |
3987 | * Returns 0 for success and negative if there was one error | |
3988 | * | |
3989 | * @mis: current migration incoming state | |
3990 | * | |
3991 | * Allocate data structures etc needed by incoming migration with | |
3992 | * postcopy-ram. postcopy-ram's similarly names | |
3993 | * postcopy_ram_incoming_init does the work. | |
1caddf8a DDAG |
3994 | */ |
3995 | int ram_postcopy_incoming_init(MigrationIncomingState *mis) | |
3996 | { | |
c136180c | 3997 | return postcopy_ram_incoming_init(mis); |
1caddf8a DDAG |
3998 | } |
3999 | ||
3d0684b2 JQ |
4000 | /** |
4001 | * ram_load_postcopy: load a page in postcopy case | |
4002 | * | |
4003 | * Returns 0 for success or -errno in case of error | |
4004 | * | |
a7180877 DDAG |
4005 | * Called in postcopy mode by ram_load(). |
4006 | * rcu_read_lock is taken prior to this being called. | |
3d0684b2 JQ |
4007 | * |
4008 | * @f: QEMUFile where to send the data | |
a7180877 DDAG |
4009 | */ |
4010 | static int ram_load_postcopy(QEMUFile *f) | |
4011 | { | |
4012 | int flags = 0, ret = 0; | |
4013 | bool place_needed = false; | |
1aa83678 | 4014 | bool matches_target_page_size = false; |
a7180877 DDAG |
4015 | MigrationIncomingState *mis = migration_incoming_get_current(); |
4016 | /* Temporary page that is later 'placed' */ | |
4017 | void *postcopy_host_page = postcopy_get_tmp_page(mis); | |
c53b7ddc | 4018 | void *last_host = NULL; |
a3b6ff6d | 4019 | bool all_zero = false; |
a7180877 DDAG |
4020 | |
4021 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
4022 | ram_addr_t addr; | |
4023 | void *host = NULL; | |
4024 | void *page_buffer = NULL; | |
4025 | void *place_source = NULL; | |
df9ff5e1 | 4026 | RAMBlock *block = NULL; |
a7180877 | 4027 | uint8_t ch; |
a7180877 DDAG |
4028 | |
4029 | addr = qemu_get_be64(f); | |
7a9ddfbf PX |
4030 | |
4031 | /* | |
4032 | * If qemu file error, we should stop here, and then "addr" | |
4033 | * may be invalid | |
4034 | */ | |
4035 | ret = qemu_file_get_error(f); | |
4036 | if (ret) { | |
4037 | break; | |
4038 | } | |
4039 | ||
a7180877 DDAG |
4040 | flags = addr & ~TARGET_PAGE_MASK; |
4041 | addr &= TARGET_PAGE_MASK; | |
4042 | ||
4043 | trace_ram_load_postcopy_loop((uint64_t)addr, flags); | |
4044 | place_needed = false; | |
bb890ed5 | 4045 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE)) { |
df9ff5e1 | 4046 | block = ram_block_from_stream(f, flags); |
4c4bad48 HZ |
4047 | |
4048 | host = host_from_ram_block_offset(block, addr); | |
a7180877 DDAG |
4049 | if (!host) { |
4050 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
4051 | ret = -EINVAL; | |
4052 | break; | |
4053 | } | |
1aa83678 | 4054 | matches_target_page_size = block->page_size == TARGET_PAGE_SIZE; |
a7180877 | 4055 | /* |
28abd200 DDAG |
4056 | * Postcopy requires that we place whole host pages atomically; |
4057 | * these may be huge pages for RAMBlocks that are backed by | |
4058 | * hugetlbfs. | |
a7180877 DDAG |
4059 | * To make it atomic, the data is read into a temporary page |
4060 | * that's moved into place later. | |
4061 | * The migration protocol uses, possibly smaller, target-pages | |
4062 | * however the source ensures it always sends all the components | |
4063 | * of a host page in order. | |
4064 | */ | |
4065 | page_buffer = postcopy_host_page + | |
28abd200 | 4066 | ((uintptr_t)host & (block->page_size - 1)); |
a7180877 | 4067 | /* If all TP are zero then we can optimise the place */ |
28abd200 | 4068 | if (!((uintptr_t)host & (block->page_size - 1))) { |
a7180877 | 4069 | all_zero = true; |
c53b7ddc DDAG |
4070 | } else { |
4071 | /* not the 1st TP within the HP */ | |
4072 | if (host != (last_host + TARGET_PAGE_SIZE)) { | |
9af9e0fe | 4073 | error_report("Non-sequential target page %p/%p", |
c53b7ddc DDAG |
4074 | host, last_host); |
4075 | ret = -EINVAL; | |
4076 | break; | |
4077 | } | |
a7180877 DDAG |
4078 | } |
4079 | ||
c53b7ddc | 4080 | |
a7180877 DDAG |
4081 | /* |
4082 | * If it's the last part of a host page then we place the host | |
4083 | * page | |
4084 | */ | |
4085 | place_needed = (((uintptr_t)host + TARGET_PAGE_SIZE) & | |
28abd200 | 4086 | (block->page_size - 1)) == 0; |
a7180877 DDAG |
4087 | place_source = postcopy_host_page; |
4088 | } | |
c53b7ddc | 4089 | last_host = host; |
a7180877 DDAG |
4090 | |
4091 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { | |
bb890ed5 | 4092 | case RAM_SAVE_FLAG_ZERO: |
a7180877 DDAG |
4093 | ch = qemu_get_byte(f); |
4094 | memset(page_buffer, ch, TARGET_PAGE_SIZE); | |
4095 | if (ch) { | |
4096 | all_zero = false; | |
4097 | } | |
4098 | break; | |
4099 | ||
4100 | case RAM_SAVE_FLAG_PAGE: | |
4101 | all_zero = false; | |
1aa83678 PX |
4102 | if (!matches_target_page_size) { |
4103 | /* For huge pages, we always use temporary buffer */ | |
a7180877 DDAG |
4104 | qemu_get_buffer(f, page_buffer, TARGET_PAGE_SIZE); |
4105 | } else { | |
1aa83678 PX |
4106 | /* |
4107 | * For small pages that matches target page size, we | |
4108 | * avoid the qemu_file copy. Instead we directly use | |
4109 | * the buffer of QEMUFile to place the page. Note: we | |
4110 | * cannot do any QEMUFile operation before using that | |
4111 | * buffer to make sure the buffer is valid when | |
4112 | * placing the page. | |
a7180877 DDAG |
4113 | */ |
4114 | qemu_get_buffer_in_place(f, (uint8_t **)&place_source, | |
4115 | TARGET_PAGE_SIZE); | |
4116 | } | |
4117 | break; | |
4118 | case RAM_SAVE_FLAG_EOS: | |
4119 | /* normal exit */ | |
6df264ac | 4120 | multifd_recv_sync_main(); |
a7180877 DDAG |
4121 | break; |
4122 | default: | |
4123 | error_report("Unknown combination of migration flags: %#x" | |
4124 | " (postcopy mode)", flags); | |
4125 | ret = -EINVAL; | |
7a9ddfbf PX |
4126 | break; |
4127 | } | |
4128 | ||
4129 | /* Detect for any possible file errors */ | |
4130 | if (!ret && qemu_file_get_error(f)) { | |
4131 | ret = qemu_file_get_error(f); | |
a7180877 DDAG |
4132 | } |
4133 | ||
7a9ddfbf | 4134 | if (!ret && place_needed) { |
a7180877 | 4135 | /* This gets called at the last target page in the host page */ |
df9ff5e1 DDAG |
4136 | void *place_dest = host + TARGET_PAGE_SIZE - block->page_size; |
4137 | ||
a7180877 | 4138 | if (all_zero) { |
df9ff5e1 | 4139 | ret = postcopy_place_page_zero(mis, place_dest, |
8be4620b | 4140 | block); |
a7180877 | 4141 | } else { |
df9ff5e1 | 4142 | ret = postcopy_place_page(mis, place_dest, |
8be4620b | 4143 | place_source, block); |
a7180877 DDAG |
4144 | } |
4145 | } | |
a7180877 DDAG |
4146 | } |
4147 | ||
4148 | return ret; | |
4149 | } | |
4150 | ||
acab30b8 DHB |
4151 | static bool postcopy_is_advised(void) |
4152 | { | |
4153 | PostcopyState ps = postcopy_state_get(); | |
4154 | return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END; | |
4155 | } | |
4156 | ||
4157 | static bool postcopy_is_running(void) | |
4158 | { | |
4159 | PostcopyState ps = postcopy_state_get(); | |
4160 | return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; | |
4161 | } | |
4162 | ||
e6f4aa18 ZC |
4163 | /* |
4164 | * Flush content of RAM cache into SVM's memory. | |
4165 | * Only flush the pages that be dirtied by PVM or SVM or both. | |
4166 | */ | |
4167 | static void colo_flush_ram_cache(void) | |
4168 | { | |
4169 | RAMBlock *block = NULL; | |
4170 | void *dst_host; | |
4171 | void *src_host; | |
4172 | unsigned long offset = 0; | |
4173 | ||
d1955d22 HZ |
4174 | memory_global_dirty_log_sync(); |
4175 | rcu_read_lock(); | |
fbd162e6 | 4176 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
d1955d22 HZ |
4177 | migration_bitmap_sync_range(ram_state, block, 0, block->used_length); |
4178 | } | |
4179 | rcu_read_unlock(); | |
4180 | ||
e6f4aa18 ZC |
4181 | trace_colo_flush_ram_cache_begin(ram_state->migration_dirty_pages); |
4182 | rcu_read_lock(); | |
4183 | block = QLIST_FIRST_RCU(&ram_list.blocks); | |
4184 | ||
4185 | while (block) { | |
4186 | offset = migration_bitmap_find_dirty(ram_state, block, offset); | |
4187 | ||
4188 | if (offset << TARGET_PAGE_BITS >= block->used_length) { | |
4189 | offset = 0; | |
4190 | block = QLIST_NEXT_RCU(block, next); | |
4191 | } else { | |
4192 | migration_bitmap_clear_dirty(ram_state, block, offset); | |
4193 | dst_host = block->host + (offset << TARGET_PAGE_BITS); | |
4194 | src_host = block->colo_cache + (offset << TARGET_PAGE_BITS); | |
4195 | memcpy(dst_host, src_host, TARGET_PAGE_SIZE); | |
4196 | } | |
4197 | } | |
4198 | ||
4199 | rcu_read_unlock(); | |
4200 | trace_colo_flush_ram_cache_end(); | |
4201 | } | |
4202 | ||
56e93d26 JQ |
4203 | static int ram_load(QEMUFile *f, void *opaque, int version_id) |
4204 | { | |
edc60127 | 4205 | int flags = 0, ret = 0, invalid_flags = 0; |
56e93d26 JQ |
4206 | static uint64_t seq_iter; |
4207 | int len = 0; | |
a7180877 DDAG |
4208 | /* |
4209 | * If system is running in postcopy mode, page inserts to host memory must | |
4210 | * be atomic | |
4211 | */ | |
acab30b8 | 4212 | bool postcopy_running = postcopy_is_running(); |
ef08fb38 | 4213 | /* ADVISE is earlier, it shows the source has the postcopy capability on */ |
acab30b8 | 4214 | bool postcopy_advised = postcopy_is_advised(); |
56e93d26 JQ |
4215 | |
4216 | seq_iter++; | |
4217 | ||
4218 | if (version_id != 4) { | |
4219 | ret = -EINVAL; | |
4220 | } | |
4221 | ||
edc60127 JQ |
4222 | if (!migrate_use_compression()) { |
4223 | invalid_flags |= RAM_SAVE_FLAG_COMPRESS_PAGE; | |
4224 | } | |
56e93d26 JQ |
4225 | /* This RCU critical section can be very long running. |
4226 | * When RCU reclaims in the code start to become numerous, | |
4227 | * it will be necessary to reduce the granularity of this | |
4228 | * critical section. | |
4229 | */ | |
4230 | rcu_read_lock(); | |
a7180877 DDAG |
4231 | |
4232 | if (postcopy_running) { | |
4233 | ret = ram_load_postcopy(f); | |
4234 | } | |
4235 | ||
4236 | while (!postcopy_running && !ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
56e93d26 | 4237 | ram_addr_t addr, total_ram_bytes; |
a776aa15 | 4238 | void *host = NULL; |
56e93d26 JQ |
4239 | uint8_t ch; |
4240 | ||
4241 | addr = qemu_get_be64(f); | |
4242 | flags = addr & ~TARGET_PAGE_MASK; | |
4243 | addr &= TARGET_PAGE_MASK; | |
4244 | ||
edc60127 JQ |
4245 | if (flags & invalid_flags) { |
4246 | if (flags & invalid_flags & RAM_SAVE_FLAG_COMPRESS_PAGE) { | |
4247 | error_report("Received an unexpected compressed page"); | |
4248 | } | |
4249 | ||
4250 | ret = -EINVAL; | |
4251 | break; | |
4252 | } | |
4253 | ||
bb890ed5 | 4254 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
a776aa15 | 4255 | RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) { |
4c4bad48 HZ |
4256 | RAMBlock *block = ram_block_from_stream(f, flags); |
4257 | ||
13af18f2 ZC |
4258 | /* |
4259 | * After going into COLO, we should load the Page into colo_cache. | |
4260 | */ | |
4261 | if (migration_incoming_in_colo_state()) { | |
4262 | host = colo_cache_from_block_offset(block, addr); | |
4263 | } else { | |
4264 | host = host_from_ram_block_offset(block, addr); | |
4265 | } | |
a776aa15 DDAG |
4266 | if (!host) { |
4267 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
4268 | ret = -EINVAL; | |
4269 | break; | |
4270 | } | |
13af18f2 ZC |
4271 | |
4272 | if (!migration_incoming_in_colo_state()) { | |
4273 | ramblock_recv_bitmap_set(block, host); | |
4274 | } | |
4275 | ||
1db9d8e5 | 4276 | trace_ram_load_loop(block->idstr, (uint64_t)addr, flags, host); |
a776aa15 DDAG |
4277 | } |
4278 | ||
56e93d26 JQ |
4279 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { |
4280 | case RAM_SAVE_FLAG_MEM_SIZE: | |
4281 | /* Synchronize RAM block list */ | |
4282 | total_ram_bytes = addr; | |
4283 | while (!ret && total_ram_bytes) { | |
4284 | RAMBlock *block; | |
56e93d26 JQ |
4285 | char id[256]; |
4286 | ram_addr_t length; | |
4287 | ||
4288 | len = qemu_get_byte(f); | |
4289 | qemu_get_buffer(f, (uint8_t *)id, len); | |
4290 | id[len] = 0; | |
4291 | length = qemu_get_be64(f); | |
4292 | ||
e3dd7493 | 4293 | block = qemu_ram_block_by_name(id); |
b895de50 CLG |
4294 | if (block && !qemu_ram_is_migratable(block)) { |
4295 | error_report("block %s should not be migrated !", id); | |
4296 | ret = -EINVAL; | |
4297 | } else if (block) { | |
e3dd7493 DDAG |
4298 | if (length != block->used_length) { |
4299 | Error *local_err = NULL; | |
56e93d26 | 4300 | |
fa53a0e5 | 4301 | ret = qemu_ram_resize(block, length, |
e3dd7493 DDAG |
4302 | &local_err); |
4303 | if (local_err) { | |
4304 | error_report_err(local_err); | |
56e93d26 | 4305 | } |
56e93d26 | 4306 | } |
ef08fb38 DDAG |
4307 | /* For postcopy we need to check hugepage sizes match */ |
4308 | if (postcopy_advised && | |
4309 | block->page_size != qemu_host_page_size) { | |
4310 | uint64_t remote_page_size = qemu_get_be64(f); | |
4311 | if (remote_page_size != block->page_size) { | |
4312 | error_report("Mismatched RAM page size %s " | |
4313 | "(local) %zd != %" PRId64, | |
4314 | id, block->page_size, | |
4315 | remote_page_size); | |
4316 | ret = -EINVAL; | |
4317 | } | |
4318 | } | |
fbd162e6 YK |
4319 | if (migrate_ignore_shared()) { |
4320 | hwaddr addr = qemu_get_be64(f); | |
4321 | bool ignored = qemu_get_byte(f); | |
4322 | if (ignored != ramblock_is_ignored(block)) { | |
4323 | error_report("RAM block %s should %s be migrated", | |
4324 | id, ignored ? "" : "not"); | |
4325 | ret = -EINVAL; | |
4326 | } | |
4327 | if (ramblock_is_ignored(block) && | |
4328 | block->mr->addr != addr) { | |
4329 | error_report("Mismatched GPAs for block %s " | |
4330 | "%" PRId64 "!= %" PRId64, | |
4331 | id, (uint64_t)addr, | |
4332 | (uint64_t)block->mr->addr); | |
4333 | ret = -EINVAL; | |
4334 | } | |
4335 | } | |
e3dd7493 DDAG |
4336 | ram_control_load_hook(f, RAM_CONTROL_BLOCK_REG, |
4337 | block->idstr); | |
4338 | } else { | |
56e93d26 JQ |
4339 | error_report("Unknown ramblock \"%s\", cannot " |
4340 | "accept migration", id); | |
4341 | ret = -EINVAL; | |
4342 | } | |
4343 | ||
4344 | total_ram_bytes -= length; | |
4345 | } | |
4346 | break; | |
a776aa15 | 4347 | |
bb890ed5 | 4348 | case RAM_SAVE_FLAG_ZERO: |
56e93d26 JQ |
4349 | ch = qemu_get_byte(f); |
4350 | ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); | |
4351 | break; | |
a776aa15 | 4352 | |
56e93d26 | 4353 | case RAM_SAVE_FLAG_PAGE: |
56e93d26 JQ |
4354 | qemu_get_buffer(f, host, TARGET_PAGE_SIZE); |
4355 | break; | |
56e93d26 | 4356 | |
a776aa15 | 4357 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
56e93d26 JQ |
4358 | len = qemu_get_be32(f); |
4359 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
4360 | error_report("Invalid compressed data length: %d", len); | |
4361 | ret = -EINVAL; | |
4362 | break; | |
4363 | } | |
c1bc6626 | 4364 | decompress_data_with_multi_threads(f, host, len); |
56e93d26 | 4365 | break; |
a776aa15 | 4366 | |
56e93d26 | 4367 | case RAM_SAVE_FLAG_XBZRLE: |
56e93d26 JQ |
4368 | if (load_xbzrle(f, addr, host) < 0) { |
4369 | error_report("Failed to decompress XBZRLE page at " | |
4370 | RAM_ADDR_FMT, addr); | |
4371 | ret = -EINVAL; | |
4372 | break; | |
4373 | } | |
4374 | break; | |
4375 | case RAM_SAVE_FLAG_EOS: | |
4376 | /* normal exit */ | |
6df264ac | 4377 | multifd_recv_sync_main(); |
56e93d26 JQ |
4378 | break; |
4379 | default: | |
4380 | if (flags & RAM_SAVE_FLAG_HOOK) { | |
632e3a5c | 4381 | ram_control_load_hook(f, RAM_CONTROL_HOOK, NULL); |
56e93d26 JQ |
4382 | } else { |
4383 | error_report("Unknown combination of migration flags: %#x", | |
4384 | flags); | |
4385 | ret = -EINVAL; | |
4386 | } | |
4387 | } | |
4388 | if (!ret) { | |
4389 | ret = qemu_file_get_error(f); | |
4390 | } | |
4391 | } | |
4392 | ||
34ab9e97 | 4393 | ret |= wait_for_decompress_done(); |
56e93d26 | 4394 | rcu_read_unlock(); |
55c4446b | 4395 | trace_ram_load_complete(ret, seq_iter); |
e6f4aa18 ZC |
4396 | |
4397 | if (!ret && migration_incoming_in_colo_state()) { | |
4398 | colo_flush_ram_cache(); | |
4399 | } | |
56e93d26 JQ |
4400 | return ret; |
4401 | } | |
4402 | ||
c6467627 VSO |
4403 | static bool ram_has_postcopy(void *opaque) |
4404 | { | |
469dd51b | 4405 | RAMBlock *rb; |
fbd162e6 | 4406 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
469dd51b JH |
4407 | if (ramblock_is_pmem(rb)) { |
4408 | info_report("Block: %s, host: %p is a nvdimm memory, postcopy" | |
4409 | "is not supported now!", rb->idstr, rb->host); | |
4410 | return false; | |
4411 | } | |
4412 | } | |
4413 | ||
c6467627 VSO |
4414 | return migrate_postcopy_ram(); |
4415 | } | |
4416 | ||
edd090c7 PX |
4417 | /* Sync all the dirty bitmap with destination VM. */ |
4418 | static int ram_dirty_bitmap_sync_all(MigrationState *s, RAMState *rs) | |
4419 | { | |
4420 | RAMBlock *block; | |
4421 | QEMUFile *file = s->to_dst_file; | |
4422 | int ramblock_count = 0; | |
4423 | ||
4424 | trace_ram_dirty_bitmap_sync_start(); | |
4425 | ||
fbd162e6 | 4426 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
edd090c7 PX |
4427 | qemu_savevm_send_recv_bitmap(file, block->idstr); |
4428 | trace_ram_dirty_bitmap_request(block->idstr); | |
4429 | ramblock_count++; | |
4430 | } | |
4431 | ||
4432 | trace_ram_dirty_bitmap_sync_wait(); | |
4433 | ||
4434 | /* Wait until all the ramblocks' dirty bitmap synced */ | |
4435 | while (ramblock_count--) { | |
4436 | qemu_sem_wait(&s->rp_state.rp_sem); | |
4437 | } | |
4438 | ||
4439 | trace_ram_dirty_bitmap_sync_complete(); | |
4440 | ||
4441 | return 0; | |
4442 | } | |
4443 | ||
4444 | static void ram_dirty_bitmap_reload_notify(MigrationState *s) | |
4445 | { | |
4446 | qemu_sem_post(&s->rp_state.rp_sem); | |
4447 | } | |
4448 | ||
a335debb PX |
4449 | /* |
4450 | * Read the received bitmap, revert it as the initial dirty bitmap. | |
4451 | * This is only used when the postcopy migration is paused but wants | |
4452 | * to resume from a middle point. | |
4453 | */ | |
4454 | int ram_dirty_bitmap_reload(MigrationState *s, RAMBlock *block) | |
4455 | { | |
4456 | int ret = -EINVAL; | |
4457 | QEMUFile *file = s->rp_state.from_dst_file; | |
4458 | unsigned long *le_bitmap, nbits = block->used_length >> TARGET_PAGE_BITS; | |
a725ef9f | 4459 | uint64_t local_size = DIV_ROUND_UP(nbits, 8); |
a335debb PX |
4460 | uint64_t size, end_mark; |
4461 | ||
4462 | trace_ram_dirty_bitmap_reload_begin(block->idstr); | |
4463 | ||
4464 | if (s->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
4465 | error_report("%s: incorrect state %s", __func__, | |
4466 | MigrationStatus_str(s->state)); | |
4467 | return -EINVAL; | |
4468 | } | |
4469 | ||
4470 | /* | |
4471 | * Note: see comments in ramblock_recv_bitmap_send() on why we | |
4472 | * need the endianess convertion, and the paddings. | |
4473 | */ | |
4474 | local_size = ROUND_UP(local_size, 8); | |
4475 | ||
4476 | /* Add paddings */ | |
4477 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
4478 | ||
4479 | size = qemu_get_be64(file); | |
4480 | ||
4481 | /* The size of the bitmap should match with our ramblock */ | |
4482 | if (size != local_size) { | |
4483 | error_report("%s: ramblock '%s' bitmap size mismatch " | |
4484 | "(0x%"PRIx64" != 0x%"PRIx64")", __func__, | |
4485 | block->idstr, size, local_size); | |
4486 | ret = -EINVAL; | |
4487 | goto out; | |
4488 | } | |
4489 | ||
4490 | size = qemu_get_buffer(file, (uint8_t *)le_bitmap, local_size); | |
4491 | end_mark = qemu_get_be64(file); | |
4492 | ||
4493 | ret = qemu_file_get_error(file); | |
4494 | if (ret || size != local_size) { | |
4495 | error_report("%s: read bitmap failed for ramblock '%s': %d" | |
4496 | " (size 0x%"PRIx64", got: 0x%"PRIx64")", | |
4497 | __func__, block->idstr, ret, local_size, size); | |
4498 | ret = -EIO; | |
4499 | goto out; | |
4500 | } | |
4501 | ||
4502 | if (end_mark != RAMBLOCK_RECV_BITMAP_ENDING) { | |
4503 | error_report("%s: ramblock '%s' end mark incorrect: 0x%"PRIu64, | |
4504 | __func__, block->idstr, end_mark); | |
4505 | ret = -EINVAL; | |
4506 | goto out; | |
4507 | } | |
4508 | ||
4509 | /* | |
4510 | * Endianess convertion. We are during postcopy (though paused). | |
4511 | * The dirty bitmap won't change. We can directly modify it. | |
4512 | */ | |
4513 | bitmap_from_le(block->bmap, le_bitmap, nbits); | |
4514 | ||
4515 | /* | |
4516 | * What we received is "received bitmap". Revert it as the initial | |
4517 | * dirty bitmap for this ramblock. | |
4518 | */ | |
4519 | bitmap_complement(block->bmap, block->bmap, nbits); | |
4520 | ||
4521 | trace_ram_dirty_bitmap_reload_complete(block->idstr); | |
4522 | ||
edd090c7 PX |
4523 | /* |
4524 | * We succeeded to sync bitmap for current ramblock. If this is | |
4525 | * the last one to sync, we need to notify the main send thread. | |
4526 | */ | |
4527 | ram_dirty_bitmap_reload_notify(s); | |
4528 | ||
a335debb PX |
4529 | ret = 0; |
4530 | out: | |
bf269906 | 4531 | g_free(le_bitmap); |
a335debb PX |
4532 | return ret; |
4533 | } | |
4534 | ||
edd090c7 PX |
4535 | static int ram_resume_prepare(MigrationState *s, void *opaque) |
4536 | { | |
4537 | RAMState *rs = *(RAMState **)opaque; | |
08614f34 | 4538 | int ret; |
edd090c7 | 4539 | |
08614f34 PX |
4540 | ret = ram_dirty_bitmap_sync_all(s, rs); |
4541 | if (ret) { | |
4542 | return ret; | |
4543 | } | |
4544 | ||
4545 | ram_state_resume_prepare(rs, s->to_dst_file); | |
4546 | ||
4547 | return 0; | |
edd090c7 PX |
4548 | } |
4549 | ||
56e93d26 | 4550 | static SaveVMHandlers savevm_ram_handlers = { |
9907e842 | 4551 | .save_setup = ram_save_setup, |
56e93d26 | 4552 | .save_live_iterate = ram_save_iterate, |
763c906b | 4553 | .save_live_complete_postcopy = ram_save_complete, |
a3e06c3d | 4554 | .save_live_complete_precopy = ram_save_complete, |
c6467627 | 4555 | .has_postcopy = ram_has_postcopy, |
56e93d26 JQ |
4556 | .save_live_pending = ram_save_pending, |
4557 | .load_state = ram_load, | |
f265e0e4 JQ |
4558 | .save_cleanup = ram_save_cleanup, |
4559 | .load_setup = ram_load_setup, | |
4560 | .load_cleanup = ram_load_cleanup, | |
edd090c7 | 4561 | .resume_prepare = ram_resume_prepare, |
56e93d26 JQ |
4562 | }; |
4563 | ||
4564 | void ram_mig_init(void) | |
4565 | { | |
4566 | qemu_mutex_init(&XBZRLE.lock); | |
6f37bb8b | 4567 | register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, &ram_state); |
56e93d26 | 4568 | } |