]>
Commit | Line | Data |
---|---|---|
56e93d26 JQ |
1 | /* |
2 | * QEMU System Emulator | |
3 | * | |
4 | * Copyright (c) 2003-2008 Fabrice Bellard | |
76cc7b58 JQ |
5 | * Copyright (c) 2011-2015 Red Hat Inc |
6 | * | |
7 | * Authors: | |
8 | * Juan Quintela <quintela@redhat.com> | |
56e93d26 JQ |
9 | * |
10 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
11 | * of this software and associated documentation files (the "Software"), to deal | |
12 | * in the Software without restriction, including without limitation the rights | |
13 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
14 | * copies of the Software, and to permit persons to whom the Software is | |
15 | * furnished to do so, subject to the following conditions: | |
16 | * | |
17 | * The above copyright notice and this permission notice shall be included in | |
18 | * all copies or substantial portions of the Software. | |
19 | * | |
20 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
21 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
22 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
23 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
24 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
25 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
26 | * THE SOFTWARE. | |
27 | */ | |
e688df6b | 28 | |
1393a485 | 29 | #include "qemu/osdep.h" |
33c11879 | 30 | #include "cpu.h" |
56e93d26 | 31 | #include <zlib.h> |
f348b6d1 | 32 | #include "qemu/cutils.h" |
56e93d26 JQ |
33 | #include "qemu/bitops.h" |
34 | #include "qemu/bitmap.h" | |
7205c9ec | 35 | #include "qemu/main-loop.h" |
709e3fe8 | 36 | #include "xbzrle.h" |
7b1e1a22 | 37 | #include "ram.h" |
6666c96a | 38 | #include "migration.h" |
71bb07db | 39 | #include "socket.h" |
f2a8f0a6 | 40 | #include "migration/register.h" |
7b1e1a22 | 41 | #include "migration/misc.h" |
08a0aee1 | 42 | #include "qemu-file.h" |
be07b0ac | 43 | #include "postcopy-ram.h" |
53d37d36 | 44 | #include "page_cache.h" |
56e93d26 | 45 | #include "qemu/error-report.h" |
e688df6b | 46 | #include "qapi/error.h" |
9af23989 | 47 | #include "qapi/qapi-events-migration.h" |
8acabf69 | 48 | #include "qapi/qmp/qerror.h" |
56e93d26 | 49 | #include "trace.h" |
56e93d26 | 50 | #include "exec/ram_addr.h" |
f9494614 | 51 | #include "exec/target_page.h" |
56e93d26 | 52 | #include "qemu/rcu_queue.h" |
a91246c9 | 53 | #include "migration/colo.h" |
53d37d36 | 54 | #include "block.h" |
af8b7d2b JQ |
55 | #include "sysemu/sysemu.h" |
56 | #include "qemu/uuid.h" | |
edd090c7 | 57 | #include "savevm.h" |
b9ee2f7d | 58 | #include "qemu/iov.h" |
56e93d26 | 59 | |
56e93d26 JQ |
60 | /***********************************************************/ |
61 | /* ram save/restore */ | |
62 | ||
bb890ed5 JQ |
63 | /* RAM_SAVE_FLAG_ZERO used to be named RAM_SAVE_FLAG_COMPRESS, it |
64 | * worked for pages that where filled with the same char. We switched | |
65 | * it to only search for the zero value. And to avoid confusion with | |
66 | * RAM_SSAVE_FLAG_COMPRESS_PAGE just rename it. | |
67 | */ | |
68 | ||
56e93d26 | 69 | #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */ |
bb890ed5 | 70 | #define RAM_SAVE_FLAG_ZERO 0x02 |
56e93d26 JQ |
71 | #define RAM_SAVE_FLAG_MEM_SIZE 0x04 |
72 | #define RAM_SAVE_FLAG_PAGE 0x08 | |
73 | #define RAM_SAVE_FLAG_EOS 0x10 | |
74 | #define RAM_SAVE_FLAG_CONTINUE 0x20 | |
75 | #define RAM_SAVE_FLAG_XBZRLE 0x40 | |
76 | /* 0x80 is reserved in migration.h start with 0x100 next */ | |
77 | #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 | |
78 | ||
56e93d26 JQ |
79 | static inline bool is_zero_range(uint8_t *p, uint64_t size) |
80 | { | |
a1febc49 | 81 | return buffer_is_zero(p, size); |
56e93d26 JQ |
82 | } |
83 | ||
9360447d JQ |
84 | XBZRLECacheStats xbzrle_counters; |
85 | ||
56e93d26 JQ |
86 | /* struct contains XBZRLE cache and a static page |
87 | used by the compression */ | |
88 | static struct { | |
89 | /* buffer used for XBZRLE encoding */ | |
90 | uint8_t *encoded_buf; | |
91 | /* buffer for storing page content */ | |
92 | uint8_t *current_buf; | |
93 | /* Cache for XBZRLE, Protected by lock. */ | |
94 | PageCache *cache; | |
95 | QemuMutex lock; | |
c00e0928 JQ |
96 | /* it will store a page full of zeros */ |
97 | uint8_t *zero_target_page; | |
f265e0e4 JQ |
98 | /* buffer used for XBZRLE decoding */ |
99 | uint8_t *decoded_buf; | |
56e93d26 JQ |
100 | } XBZRLE; |
101 | ||
56e93d26 JQ |
102 | static void XBZRLE_cache_lock(void) |
103 | { | |
104 | if (migrate_use_xbzrle()) | |
105 | qemu_mutex_lock(&XBZRLE.lock); | |
106 | } | |
107 | ||
108 | static void XBZRLE_cache_unlock(void) | |
109 | { | |
110 | if (migrate_use_xbzrle()) | |
111 | qemu_mutex_unlock(&XBZRLE.lock); | |
112 | } | |
113 | ||
3d0684b2 JQ |
114 | /** |
115 | * xbzrle_cache_resize: resize the xbzrle cache | |
116 | * | |
117 | * This function is called from qmp_migrate_set_cache_size in main | |
118 | * thread, possibly while a migration is in progress. A running | |
119 | * migration may be using the cache and might finish during this call, | |
120 | * hence changes to the cache are protected by XBZRLE.lock(). | |
121 | * | |
c9dede2d | 122 | * Returns 0 for success or -1 for error |
3d0684b2 JQ |
123 | * |
124 | * @new_size: new cache size | |
8acabf69 | 125 | * @errp: set *errp if the check failed, with reason |
56e93d26 | 126 | */ |
c9dede2d | 127 | int xbzrle_cache_resize(int64_t new_size, Error **errp) |
56e93d26 JQ |
128 | { |
129 | PageCache *new_cache; | |
c9dede2d | 130 | int64_t ret = 0; |
56e93d26 | 131 | |
8acabf69 JQ |
132 | /* Check for truncation */ |
133 | if (new_size != (size_t)new_size) { | |
134 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", | |
135 | "exceeding address space"); | |
136 | return -1; | |
137 | } | |
138 | ||
2a313e5c JQ |
139 | if (new_size == migrate_xbzrle_cache_size()) { |
140 | /* nothing to do */ | |
c9dede2d | 141 | return 0; |
2a313e5c JQ |
142 | } |
143 | ||
56e93d26 JQ |
144 | XBZRLE_cache_lock(); |
145 | ||
146 | if (XBZRLE.cache != NULL) { | |
80f8dfde | 147 | new_cache = cache_init(new_size, TARGET_PAGE_SIZE, errp); |
56e93d26 | 148 | if (!new_cache) { |
56e93d26 JQ |
149 | ret = -1; |
150 | goto out; | |
151 | } | |
152 | ||
153 | cache_fini(XBZRLE.cache); | |
154 | XBZRLE.cache = new_cache; | |
155 | } | |
56e93d26 JQ |
156 | out: |
157 | XBZRLE_cache_unlock(); | |
158 | return ret; | |
159 | } | |
160 | ||
fbd162e6 YK |
161 | static bool ramblock_is_ignored(RAMBlock *block) |
162 | { | |
163 | return !qemu_ram_is_migratable(block) || | |
164 | (migrate_ignore_shared() && qemu_ram_is_shared(block)); | |
165 | } | |
166 | ||
b895de50 | 167 | /* Should be holding either ram_list.mutex, or the RCU lock. */ |
fbd162e6 YK |
168 | #define RAMBLOCK_FOREACH_NOT_IGNORED(block) \ |
169 | INTERNAL_RAMBLOCK_FOREACH(block) \ | |
170 | if (ramblock_is_ignored(block)) {} else | |
171 | ||
b895de50 | 172 | #define RAMBLOCK_FOREACH_MIGRATABLE(block) \ |
343f632c | 173 | INTERNAL_RAMBLOCK_FOREACH(block) \ |
b895de50 CLG |
174 | if (!qemu_ram_is_migratable(block)) {} else |
175 | ||
343f632c DDAG |
176 | #undef RAMBLOCK_FOREACH |
177 | ||
fbd162e6 YK |
178 | int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque) |
179 | { | |
180 | RAMBlock *block; | |
181 | int ret = 0; | |
182 | ||
89ac5a1d DDAG |
183 | RCU_READ_LOCK_GUARD(); |
184 | ||
fbd162e6 YK |
185 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
186 | ret = func(block, opaque); | |
187 | if (ret) { | |
188 | break; | |
189 | } | |
190 | } | |
fbd162e6 YK |
191 | return ret; |
192 | } | |
193 | ||
f9494614 AP |
194 | static void ramblock_recv_map_init(void) |
195 | { | |
196 | RAMBlock *rb; | |
197 | ||
fbd162e6 | 198 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
f9494614 AP |
199 | assert(!rb->receivedmap); |
200 | rb->receivedmap = bitmap_new(rb->max_length >> qemu_target_page_bits()); | |
201 | } | |
202 | } | |
203 | ||
204 | int ramblock_recv_bitmap_test(RAMBlock *rb, void *host_addr) | |
205 | { | |
206 | return test_bit(ramblock_recv_bitmap_offset(host_addr, rb), | |
207 | rb->receivedmap); | |
208 | } | |
209 | ||
1cba9f6e DDAG |
210 | bool ramblock_recv_bitmap_test_byte_offset(RAMBlock *rb, uint64_t byte_offset) |
211 | { | |
212 | return test_bit(byte_offset >> TARGET_PAGE_BITS, rb->receivedmap); | |
213 | } | |
214 | ||
f9494614 AP |
215 | void ramblock_recv_bitmap_set(RAMBlock *rb, void *host_addr) |
216 | { | |
217 | set_bit_atomic(ramblock_recv_bitmap_offset(host_addr, rb), rb->receivedmap); | |
218 | } | |
219 | ||
220 | void ramblock_recv_bitmap_set_range(RAMBlock *rb, void *host_addr, | |
221 | size_t nr) | |
222 | { | |
223 | bitmap_set_atomic(rb->receivedmap, | |
224 | ramblock_recv_bitmap_offset(host_addr, rb), | |
225 | nr); | |
226 | } | |
227 | ||
a335debb PX |
228 | #define RAMBLOCK_RECV_BITMAP_ENDING (0x0123456789abcdefULL) |
229 | ||
230 | /* | |
231 | * Format: bitmap_size (8 bytes) + whole_bitmap (N bytes). | |
232 | * | |
233 | * Returns >0 if success with sent bytes, or <0 if error. | |
234 | */ | |
235 | int64_t ramblock_recv_bitmap_send(QEMUFile *file, | |
236 | const char *block_name) | |
237 | { | |
238 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
239 | unsigned long *le_bitmap, nbits; | |
240 | uint64_t size; | |
241 | ||
242 | if (!block) { | |
243 | error_report("%s: invalid block name: %s", __func__, block_name); | |
244 | return -1; | |
245 | } | |
246 | ||
247 | nbits = block->used_length >> TARGET_PAGE_BITS; | |
248 | ||
249 | /* | |
250 | * Make sure the tmp bitmap buffer is big enough, e.g., on 32bit | |
251 | * machines we may need 4 more bytes for padding (see below | |
252 | * comment). So extend it a bit before hand. | |
253 | */ | |
254 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
255 | ||
256 | /* | |
257 | * Always use little endian when sending the bitmap. This is | |
258 | * required that when source and destination VMs are not using the | |
259 | * same endianess. (Note: big endian won't work.) | |
260 | */ | |
261 | bitmap_to_le(le_bitmap, block->receivedmap, nbits); | |
262 | ||
263 | /* Size of the bitmap, in bytes */ | |
a725ef9f | 264 | size = DIV_ROUND_UP(nbits, 8); |
a335debb PX |
265 | |
266 | /* | |
267 | * size is always aligned to 8 bytes for 64bit machines, but it | |
268 | * may not be true for 32bit machines. We need this padding to | |
269 | * make sure the migration can survive even between 32bit and | |
270 | * 64bit machines. | |
271 | */ | |
272 | size = ROUND_UP(size, 8); | |
273 | ||
274 | qemu_put_be64(file, size); | |
275 | qemu_put_buffer(file, (const uint8_t *)le_bitmap, size); | |
276 | /* | |
277 | * Mark as an end, in case the middle part is screwed up due to | |
278 | * some "misterious" reason. | |
279 | */ | |
280 | qemu_put_be64(file, RAMBLOCK_RECV_BITMAP_ENDING); | |
281 | qemu_fflush(file); | |
282 | ||
bf269906 | 283 | g_free(le_bitmap); |
a335debb PX |
284 | |
285 | if (qemu_file_get_error(file)) { | |
286 | return qemu_file_get_error(file); | |
287 | } | |
288 | ||
289 | return size + sizeof(size); | |
290 | } | |
291 | ||
ec481c6c JQ |
292 | /* |
293 | * An outstanding page request, on the source, having been received | |
294 | * and queued | |
295 | */ | |
296 | struct RAMSrcPageRequest { | |
297 | RAMBlock *rb; | |
298 | hwaddr offset; | |
299 | hwaddr len; | |
300 | ||
301 | QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; | |
302 | }; | |
303 | ||
6f37bb8b JQ |
304 | /* State of RAM for migration */ |
305 | struct RAMState { | |
204b88b8 JQ |
306 | /* QEMUFile used for this migration */ |
307 | QEMUFile *f; | |
6f37bb8b JQ |
308 | /* Last block that we have visited searching for dirty pages */ |
309 | RAMBlock *last_seen_block; | |
310 | /* Last block from where we have sent data */ | |
311 | RAMBlock *last_sent_block; | |
269ace29 JQ |
312 | /* Last dirty target page we have sent */ |
313 | ram_addr_t last_page; | |
6f37bb8b JQ |
314 | /* last ram version we have seen */ |
315 | uint32_t last_version; | |
316 | /* We are in the first round */ | |
317 | bool ram_bulk_stage; | |
6eeb63f7 WW |
318 | /* The free page optimization is enabled */ |
319 | bool fpo_enabled; | |
8d820d6f JQ |
320 | /* How many times we have dirty too many pages */ |
321 | int dirty_rate_high_cnt; | |
f664da80 JQ |
322 | /* these variables are used for bitmap sync */ |
323 | /* last time we did a full bitmap_sync */ | |
324 | int64_t time_last_bitmap_sync; | |
eac74159 | 325 | /* bytes transferred at start_time */ |
c4bdf0cf | 326 | uint64_t bytes_xfer_prev; |
a66cd90c | 327 | /* number of dirty pages since start_time */ |
68908ed6 | 328 | uint64_t num_dirty_pages_period; |
b5833fde JQ |
329 | /* xbzrle misses since the beginning of the period */ |
330 | uint64_t xbzrle_cache_miss_prev; | |
76e03000 XG |
331 | |
332 | /* compression statistics since the beginning of the period */ | |
333 | /* amount of count that no free thread to compress data */ | |
334 | uint64_t compress_thread_busy_prev; | |
335 | /* amount bytes after compression */ | |
336 | uint64_t compressed_size_prev; | |
337 | /* amount of compressed pages */ | |
338 | uint64_t compress_pages_prev; | |
339 | ||
be8b02ed XG |
340 | /* total handled target pages at the beginning of period */ |
341 | uint64_t target_page_count_prev; | |
342 | /* total handled target pages since start */ | |
343 | uint64_t target_page_count; | |
9360447d | 344 | /* number of dirty bits in the bitmap */ |
2dfaf12e | 345 | uint64_t migration_dirty_pages; |
386a907b | 346 | /* Protects modification of the bitmap and migration dirty pages */ |
108cfae0 | 347 | QemuMutex bitmap_mutex; |
68a098f3 JQ |
348 | /* The RAMBlock used in the last src_page_requests */ |
349 | RAMBlock *last_req_rb; | |
ec481c6c JQ |
350 | /* Queue of outstanding page requests from the destination */ |
351 | QemuMutex src_page_req_mutex; | |
b58deb34 | 352 | QSIMPLEQ_HEAD(, RAMSrcPageRequest) src_page_requests; |
6f37bb8b JQ |
353 | }; |
354 | typedef struct RAMState RAMState; | |
355 | ||
53518d94 | 356 | static RAMState *ram_state; |
6f37bb8b | 357 | |
bd227060 WW |
358 | static NotifierWithReturnList precopy_notifier_list; |
359 | ||
360 | void precopy_infrastructure_init(void) | |
361 | { | |
362 | notifier_with_return_list_init(&precopy_notifier_list); | |
363 | } | |
364 | ||
365 | void precopy_add_notifier(NotifierWithReturn *n) | |
366 | { | |
367 | notifier_with_return_list_add(&precopy_notifier_list, n); | |
368 | } | |
369 | ||
370 | void precopy_remove_notifier(NotifierWithReturn *n) | |
371 | { | |
372 | notifier_with_return_remove(n); | |
373 | } | |
374 | ||
375 | int precopy_notify(PrecopyNotifyReason reason, Error **errp) | |
376 | { | |
377 | PrecopyNotifyData pnd; | |
378 | pnd.reason = reason; | |
379 | pnd.errp = errp; | |
380 | ||
381 | return notifier_with_return_list_notify(&precopy_notifier_list, &pnd); | |
382 | } | |
383 | ||
6eeb63f7 WW |
384 | void precopy_enable_free_page_optimization(void) |
385 | { | |
386 | if (!ram_state) { | |
387 | return; | |
388 | } | |
389 | ||
390 | ram_state->fpo_enabled = true; | |
391 | } | |
392 | ||
9edabd4d | 393 | uint64_t ram_bytes_remaining(void) |
2f4fde93 | 394 | { |
bae416e5 DDAG |
395 | return ram_state ? (ram_state->migration_dirty_pages * TARGET_PAGE_SIZE) : |
396 | 0; | |
2f4fde93 JQ |
397 | } |
398 | ||
9360447d | 399 | MigrationStats ram_counters; |
96506894 | 400 | |
b8fb8cb7 DDAG |
401 | /* used by the search for pages to send */ |
402 | struct PageSearchStatus { | |
403 | /* Current block being searched */ | |
404 | RAMBlock *block; | |
a935e30f JQ |
405 | /* Current page to search from */ |
406 | unsigned long page; | |
b8fb8cb7 DDAG |
407 | /* Set once we wrap around */ |
408 | bool complete_round; | |
409 | }; | |
410 | typedef struct PageSearchStatus PageSearchStatus; | |
411 | ||
76e03000 XG |
412 | CompressionStats compression_counters; |
413 | ||
56e93d26 | 414 | struct CompressParam { |
56e93d26 | 415 | bool done; |
90e56fb4 | 416 | bool quit; |
5e5fdcff | 417 | bool zero_page; |
56e93d26 JQ |
418 | QEMUFile *file; |
419 | QemuMutex mutex; | |
420 | QemuCond cond; | |
421 | RAMBlock *block; | |
422 | ram_addr_t offset; | |
34ab9e97 XG |
423 | |
424 | /* internally used fields */ | |
dcaf446e | 425 | z_stream stream; |
34ab9e97 | 426 | uint8_t *originbuf; |
56e93d26 JQ |
427 | }; |
428 | typedef struct CompressParam CompressParam; | |
429 | ||
430 | struct DecompressParam { | |
73a8912b | 431 | bool done; |
90e56fb4 | 432 | bool quit; |
56e93d26 JQ |
433 | QemuMutex mutex; |
434 | QemuCond cond; | |
435 | void *des; | |
d341d9f3 | 436 | uint8_t *compbuf; |
56e93d26 | 437 | int len; |
797ca154 | 438 | z_stream stream; |
56e93d26 JQ |
439 | }; |
440 | typedef struct DecompressParam DecompressParam; | |
441 | ||
442 | static CompressParam *comp_param; | |
443 | static QemuThread *compress_threads; | |
444 | /* comp_done_cond is used to wake up the migration thread when | |
445 | * one of the compression threads has finished the compression. | |
446 | * comp_done_lock is used to co-work with comp_done_cond. | |
447 | */ | |
0d9f9a5c LL |
448 | static QemuMutex comp_done_lock; |
449 | static QemuCond comp_done_cond; | |
56e93d26 JQ |
450 | /* The empty QEMUFileOps will be used by file in CompressParam */ |
451 | static const QEMUFileOps empty_ops = { }; | |
452 | ||
34ab9e97 | 453 | static QEMUFile *decomp_file; |
56e93d26 JQ |
454 | static DecompressParam *decomp_param; |
455 | static QemuThread *decompress_threads; | |
73a8912b LL |
456 | static QemuMutex decomp_done_lock; |
457 | static QemuCond decomp_done_cond; | |
56e93d26 | 458 | |
5e5fdcff | 459 | static bool do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
6ef3771c | 460 | ram_addr_t offset, uint8_t *source_buf); |
56e93d26 JQ |
461 | |
462 | static void *do_data_compress(void *opaque) | |
463 | { | |
464 | CompressParam *param = opaque; | |
a7a9a88f LL |
465 | RAMBlock *block; |
466 | ram_addr_t offset; | |
5e5fdcff | 467 | bool zero_page; |
56e93d26 | 468 | |
a7a9a88f | 469 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 470 | while (!param->quit) { |
a7a9a88f LL |
471 | if (param->block) { |
472 | block = param->block; | |
473 | offset = param->offset; | |
474 | param->block = NULL; | |
475 | qemu_mutex_unlock(¶m->mutex); | |
476 | ||
5e5fdcff XG |
477 | zero_page = do_compress_ram_page(param->file, ¶m->stream, |
478 | block, offset, param->originbuf); | |
a7a9a88f | 479 | |
0d9f9a5c | 480 | qemu_mutex_lock(&comp_done_lock); |
a7a9a88f | 481 | param->done = true; |
5e5fdcff | 482 | param->zero_page = zero_page; |
0d9f9a5c LL |
483 | qemu_cond_signal(&comp_done_cond); |
484 | qemu_mutex_unlock(&comp_done_lock); | |
a7a9a88f LL |
485 | |
486 | qemu_mutex_lock(¶m->mutex); | |
487 | } else { | |
56e93d26 JQ |
488 | qemu_cond_wait(¶m->cond, ¶m->mutex); |
489 | } | |
56e93d26 | 490 | } |
a7a9a88f | 491 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
492 | |
493 | return NULL; | |
494 | } | |
495 | ||
f0afa331 | 496 | static void compress_threads_save_cleanup(void) |
56e93d26 JQ |
497 | { |
498 | int i, thread_count; | |
499 | ||
05306935 | 500 | if (!migrate_use_compression() || !comp_param) { |
56e93d26 JQ |
501 | return; |
502 | } | |
05306935 | 503 | |
56e93d26 JQ |
504 | thread_count = migrate_compress_threads(); |
505 | for (i = 0; i < thread_count; i++) { | |
dcaf446e XG |
506 | /* |
507 | * we use it as a indicator which shows if the thread is | |
508 | * properly init'd or not | |
509 | */ | |
510 | if (!comp_param[i].file) { | |
511 | break; | |
512 | } | |
05306935 FL |
513 | |
514 | qemu_mutex_lock(&comp_param[i].mutex); | |
515 | comp_param[i].quit = true; | |
516 | qemu_cond_signal(&comp_param[i].cond); | |
517 | qemu_mutex_unlock(&comp_param[i].mutex); | |
518 | ||
56e93d26 | 519 | qemu_thread_join(compress_threads + i); |
56e93d26 JQ |
520 | qemu_mutex_destroy(&comp_param[i].mutex); |
521 | qemu_cond_destroy(&comp_param[i].cond); | |
dcaf446e | 522 | deflateEnd(&comp_param[i].stream); |
34ab9e97 | 523 | g_free(comp_param[i].originbuf); |
dcaf446e XG |
524 | qemu_fclose(comp_param[i].file); |
525 | comp_param[i].file = NULL; | |
56e93d26 | 526 | } |
0d9f9a5c LL |
527 | qemu_mutex_destroy(&comp_done_lock); |
528 | qemu_cond_destroy(&comp_done_cond); | |
56e93d26 JQ |
529 | g_free(compress_threads); |
530 | g_free(comp_param); | |
56e93d26 JQ |
531 | compress_threads = NULL; |
532 | comp_param = NULL; | |
56e93d26 JQ |
533 | } |
534 | ||
dcaf446e | 535 | static int compress_threads_save_setup(void) |
56e93d26 JQ |
536 | { |
537 | int i, thread_count; | |
538 | ||
539 | if (!migrate_use_compression()) { | |
dcaf446e | 540 | return 0; |
56e93d26 | 541 | } |
56e93d26 JQ |
542 | thread_count = migrate_compress_threads(); |
543 | compress_threads = g_new0(QemuThread, thread_count); | |
544 | comp_param = g_new0(CompressParam, thread_count); | |
0d9f9a5c LL |
545 | qemu_cond_init(&comp_done_cond); |
546 | qemu_mutex_init(&comp_done_lock); | |
56e93d26 | 547 | for (i = 0; i < thread_count; i++) { |
34ab9e97 XG |
548 | comp_param[i].originbuf = g_try_malloc(TARGET_PAGE_SIZE); |
549 | if (!comp_param[i].originbuf) { | |
550 | goto exit; | |
551 | } | |
552 | ||
dcaf446e XG |
553 | if (deflateInit(&comp_param[i].stream, |
554 | migrate_compress_level()) != Z_OK) { | |
34ab9e97 | 555 | g_free(comp_param[i].originbuf); |
dcaf446e XG |
556 | goto exit; |
557 | } | |
558 | ||
e110aa91 C |
559 | /* comp_param[i].file is just used as a dummy buffer to save data, |
560 | * set its ops to empty. | |
56e93d26 JQ |
561 | */ |
562 | comp_param[i].file = qemu_fopen_ops(NULL, &empty_ops); | |
563 | comp_param[i].done = true; | |
90e56fb4 | 564 | comp_param[i].quit = false; |
56e93d26 JQ |
565 | qemu_mutex_init(&comp_param[i].mutex); |
566 | qemu_cond_init(&comp_param[i].cond); | |
567 | qemu_thread_create(compress_threads + i, "compress", | |
568 | do_data_compress, comp_param + i, | |
569 | QEMU_THREAD_JOINABLE); | |
570 | } | |
dcaf446e XG |
571 | return 0; |
572 | ||
573 | exit: | |
574 | compress_threads_save_cleanup(); | |
575 | return -1; | |
56e93d26 JQ |
576 | } |
577 | ||
f986c3d2 JQ |
578 | /* Multiple fd's */ |
579 | ||
af8b7d2b JQ |
580 | #define MULTIFD_MAGIC 0x11223344U |
581 | #define MULTIFD_VERSION 1 | |
582 | ||
6df264ac JQ |
583 | #define MULTIFD_FLAG_SYNC (1 << 0) |
584 | ||
efd1a1d6 | 585 | /* This value needs to be a multiple of qemu_target_page_size() */ |
4b0c7264 | 586 | #define MULTIFD_PACKET_SIZE (512 * 1024) |
efd1a1d6 | 587 | |
af8b7d2b JQ |
588 | typedef struct { |
589 | uint32_t magic; | |
590 | uint32_t version; | |
591 | unsigned char uuid[16]; /* QemuUUID */ | |
592 | uint8_t id; | |
5fbd8b4b JQ |
593 | uint8_t unused1[7]; /* Reserved for future use */ |
594 | uint64_t unused2[4]; /* Reserved for future use */ | |
af8b7d2b JQ |
595 | } __attribute__((packed)) MultiFDInit_t; |
596 | ||
2a26c979 JQ |
597 | typedef struct { |
598 | uint32_t magic; | |
599 | uint32_t version; | |
600 | uint32_t flags; | |
6f862692 JQ |
601 | /* maximum number of allocated pages */ |
602 | uint32_t pages_alloc; | |
603 | uint32_t pages_used; | |
2a34ee59 JQ |
604 | /* size of the next packet that contains pages */ |
605 | uint32_t next_packet_size; | |
2a26c979 | 606 | uint64_t packet_num; |
5fbd8b4b | 607 | uint64_t unused[4]; /* Reserved for future use */ |
2a26c979 JQ |
608 | char ramblock[256]; |
609 | uint64_t offset[]; | |
610 | } __attribute__((packed)) MultiFDPacket_t; | |
611 | ||
34c55a94 JQ |
612 | typedef struct { |
613 | /* number of used pages */ | |
614 | uint32_t used; | |
615 | /* number of allocated pages */ | |
616 | uint32_t allocated; | |
617 | /* global number of generated multifd packets */ | |
618 | uint64_t packet_num; | |
619 | /* offset of each page */ | |
620 | ram_addr_t *offset; | |
621 | /* pointer to each page */ | |
622 | struct iovec *iov; | |
623 | RAMBlock *block; | |
624 | } MultiFDPages_t; | |
625 | ||
8c4598f2 JQ |
626 | typedef struct { |
627 | /* this fields are not changed once the thread is created */ | |
628 | /* channel number */ | |
f986c3d2 | 629 | uint8_t id; |
8c4598f2 | 630 | /* channel thread name */ |
f986c3d2 | 631 | char *name; |
8c4598f2 | 632 | /* channel thread id */ |
f986c3d2 | 633 | QemuThread thread; |
8c4598f2 | 634 | /* communication channel */ |
60df2d4a | 635 | QIOChannel *c; |
8c4598f2 | 636 | /* sem where to wait for more work */ |
f986c3d2 | 637 | QemuSemaphore sem; |
8c4598f2 | 638 | /* this mutex protects the following parameters */ |
f986c3d2 | 639 | QemuMutex mutex; |
8c4598f2 | 640 | /* is this channel thread running */ |
66770707 | 641 | bool running; |
8c4598f2 | 642 | /* should this thread finish */ |
f986c3d2 | 643 | bool quit; |
0beb5ed3 JQ |
644 | /* thread has work to do */ |
645 | int pending_job; | |
34c55a94 JQ |
646 | /* array of pages to sent */ |
647 | MultiFDPages_t *pages; | |
2a26c979 JQ |
648 | /* packet allocated len */ |
649 | uint32_t packet_len; | |
650 | /* pointer to the packet */ | |
651 | MultiFDPacket_t *packet; | |
652 | /* multifd flags for each packet */ | |
653 | uint32_t flags; | |
2a34ee59 JQ |
654 | /* size of the next packet that contains pages */ |
655 | uint32_t next_packet_size; | |
2a26c979 JQ |
656 | /* global number of generated multifd packets */ |
657 | uint64_t packet_num; | |
408ea6ae JQ |
658 | /* thread local variables */ |
659 | /* packets sent through this channel */ | |
660 | uint64_t num_packets; | |
661 | /* pages sent through this channel */ | |
662 | uint64_t num_pages; | |
18cdcea3 JQ |
663 | /* syncs main thread and channels */ |
664 | QemuSemaphore sem_sync; | |
8c4598f2 JQ |
665 | } MultiFDSendParams; |
666 | ||
667 | typedef struct { | |
668 | /* this fields are not changed once the thread is created */ | |
669 | /* channel number */ | |
670 | uint8_t id; | |
671 | /* channel thread name */ | |
672 | char *name; | |
673 | /* channel thread id */ | |
674 | QemuThread thread; | |
675 | /* communication channel */ | |
676 | QIOChannel *c; | |
8c4598f2 JQ |
677 | /* this mutex protects the following parameters */ |
678 | QemuMutex mutex; | |
679 | /* is this channel thread running */ | |
680 | bool running; | |
3c3ca25d JQ |
681 | /* should this thread finish */ |
682 | bool quit; | |
34c55a94 JQ |
683 | /* array of pages to receive */ |
684 | MultiFDPages_t *pages; | |
2a26c979 JQ |
685 | /* packet allocated len */ |
686 | uint32_t packet_len; | |
687 | /* pointer to the packet */ | |
688 | MultiFDPacket_t *packet; | |
689 | /* multifd flags for each packet */ | |
690 | uint32_t flags; | |
691 | /* global number of generated multifd packets */ | |
692 | uint64_t packet_num; | |
408ea6ae | 693 | /* thread local variables */ |
2a34ee59 JQ |
694 | /* size of the next packet that contains pages */ |
695 | uint32_t next_packet_size; | |
408ea6ae JQ |
696 | /* packets sent through this channel */ |
697 | uint64_t num_packets; | |
698 | /* pages sent through this channel */ | |
699 | uint64_t num_pages; | |
6df264ac JQ |
700 | /* syncs main thread and channels */ |
701 | QemuSemaphore sem_sync; | |
8c4598f2 | 702 | } MultiFDRecvParams; |
f986c3d2 | 703 | |
af8b7d2b JQ |
704 | static int multifd_send_initial_packet(MultiFDSendParams *p, Error **errp) |
705 | { | |
706 | MultiFDInit_t msg; | |
707 | int ret; | |
708 | ||
709 | msg.magic = cpu_to_be32(MULTIFD_MAGIC); | |
710 | msg.version = cpu_to_be32(MULTIFD_VERSION); | |
711 | msg.id = p->id; | |
712 | memcpy(msg.uuid, &qemu_uuid.data, sizeof(msg.uuid)); | |
713 | ||
714 | ret = qio_channel_write_all(p->c, (char *)&msg, sizeof(msg), errp); | |
715 | if (ret != 0) { | |
716 | return -1; | |
717 | } | |
718 | return 0; | |
719 | } | |
720 | ||
721 | static int multifd_recv_initial_packet(QIOChannel *c, Error **errp) | |
722 | { | |
723 | MultiFDInit_t msg; | |
724 | int ret; | |
725 | ||
726 | ret = qio_channel_read_all(c, (char *)&msg, sizeof(msg), errp); | |
727 | if (ret != 0) { | |
728 | return -1; | |
729 | } | |
730 | ||
341ba0df PM |
731 | msg.magic = be32_to_cpu(msg.magic); |
732 | msg.version = be32_to_cpu(msg.version); | |
af8b7d2b JQ |
733 | |
734 | if (msg.magic != MULTIFD_MAGIC) { | |
735 | error_setg(errp, "multifd: received packet magic %x " | |
736 | "expected %x", msg.magic, MULTIFD_MAGIC); | |
737 | return -1; | |
738 | } | |
739 | ||
740 | if (msg.version != MULTIFD_VERSION) { | |
741 | error_setg(errp, "multifd: received packet version %d " | |
742 | "expected %d", msg.version, MULTIFD_VERSION); | |
743 | return -1; | |
744 | } | |
745 | ||
746 | if (memcmp(msg.uuid, &qemu_uuid, sizeof(qemu_uuid))) { | |
747 | char *uuid = qemu_uuid_unparse_strdup(&qemu_uuid); | |
748 | char *msg_uuid = qemu_uuid_unparse_strdup((const QemuUUID *)msg.uuid); | |
749 | ||
750 | error_setg(errp, "multifd: received uuid '%s' and expected " | |
751 | "uuid '%s' for channel %hhd", msg_uuid, uuid, msg.id); | |
752 | g_free(uuid); | |
753 | g_free(msg_uuid); | |
754 | return -1; | |
755 | } | |
756 | ||
757 | if (msg.id > migrate_multifd_channels()) { | |
758 | error_setg(errp, "multifd: received channel version %d " | |
759 | "expected %d", msg.version, MULTIFD_VERSION); | |
760 | return -1; | |
761 | } | |
762 | ||
763 | return msg.id; | |
764 | } | |
765 | ||
34c55a94 JQ |
766 | static MultiFDPages_t *multifd_pages_init(size_t size) |
767 | { | |
768 | MultiFDPages_t *pages = g_new0(MultiFDPages_t, 1); | |
769 | ||
770 | pages->allocated = size; | |
771 | pages->iov = g_new0(struct iovec, size); | |
772 | pages->offset = g_new0(ram_addr_t, size); | |
773 | ||
774 | return pages; | |
775 | } | |
776 | ||
777 | static void multifd_pages_clear(MultiFDPages_t *pages) | |
778 | { | |
779 | pages->used = 0; | |
780 | pages->allocated = 0; | |
781 | pages->packet_num = 0; | |
782 | pages->block = NULL; | |
783 | g_free(pages->iov); | |
784 | pages->iov = NULL; | |
785 | g_free(pages->offset); | |
786 | pages->offset = NULL; | |
787 | g_free(pages); | |
788 | } | |
789 | ||
2a26c979 JQ |
790 | static void multifd_send_fill_packet(MultiFDSendParams *p) |
791 | { | |
792 | MultiFDPacket_t *packet = p->packet; | |
793 | int i; | |
794 | ||
2a26c979 | 795 | packet->flags = cpu_to_be32(p->flags); |
f2148c4c | 796 | packet->pages_alloc = cpu_to_be32(p->pages->allocated); |
6f862692 | 797 | packet->pages_used = cpu_to_be32(p->pages->used); |
2a34ee59 | 798 | packet->next_packet_size = cpu_to_be32(p->next_packet_size); |
2a26c979 JQ |
799 | packet->packet_num = cpu_to_be64(p->packet_num); |
800 | ||
801 | if (p->pages->block) { | |
802 | strncpy(packet->ramblock, p->pages->block->idstr, 256); | |
803 | } | |
804 | ||
805 | for (i = 0; i < p->pages->used; i++) { | |
806 | packet->offset[i] = cpu_to_be64(p->pages->offset[i]); | |
807 | } | |
808 | } | |
809 | ||
810 | static int multifd_recv_unfill_packet(MultiFDRecvParams *p, Error **errp) | |
811 | { | |
812 | MultiFDPacket_t *packet = p->packet; | |
7ed379b2 | 813 | uint32_t pages_max = MULTIFD_PACKET_SIZE / qemu_target_page_size(); |
2a26c979 JQ |
814 | RAMBlock *block; |
815 | int i; | |
816 | ||
341ba0df | 817 | packet->magic = be32_to_cpu(packet->magic); |
2a26c979 JQ |
818 | if (packet->magic != MULTIFD_MAGIC) { |
819 | error_setg(errp, "multifd: received packet " | |
820 | "magic %x and expected magic %x", | |
821 | packet->magic, MULTIFD_MAGIC); | |
822 | return -1; | |
823 | } | |
824 | ||
341ba0df | 825 | packet->version = be32_to_cpu(packet->version); |
2a26c979 JQ |
826 | if (packet->version != MULTIFD_VERSION) { |
827 | error_setg(errp, "multifd: received packet " | |
828 | "version %d and expected version %d", | |
829 | packet->version, MULTIFD_VERSION); | |
830 | return -1; | |
831 | } | |
832 | ||
833 | p->flags = be32_to_cpu(packet->flags); | |
834 | ||
6f862692 | 835 | packet->pages_alloc = be32_to_cpu(packet->pages_alloc); |
7ed379b2 | 836 | /* |
d884e77b | 837 | * If we received a packet that is 100 times bigger than expected |
7ed379b2 JQ |
838 | * just stop migration. It is a magic number. |
839 | */ | |
840 | if (packet->pages_alloc > pages_max * 100) { | |
2a26c979 | 841 | error_setg(errp, "multifd: received packet " |
7ed379b2 JQ |
842 | "with size %d and expected a maximum size of %d", |
843 | packet->pages_alloc, pages_max * 100) ; | |
2a26c979 JQ |
844 | return -1; |
845 | } | |
7ed379b2 JQ |
846 | /* |
847 | * We received a packet that is bigger than expected but inside | |
848 | * reasonable limits (see previous comment). Just reallocate. | |
849 | */ | |
850 | if (packet->pages_alloc > p->pages->allocated) { | |
851 | multifd_pages_clear(p->pages); | |
f151f8ac | 852 | p->pages = multifd_pages_init(packet->pages_alloc); |
7ed379b2 | 853 | } |
2a26c979 | 854 | |
6f862692 JQ |
855 | p->pages->used = be32_to_cpu(packet->pages_used); |
856 | if (p->pages->used > packet->pages_alloc) { | |
2a26c979 | 857 | error_setg(errp, "multifd: received packet " |
6f862692 JQ |
858 | "with %d pages and expected maximum pages are %d", |
859 | p->pages->used, packet->pages_alloc) ; | |
2a26c979 JQ |
860 | return -1; |
861 | } | |
862 | ||
2a34ee59 | 863 | p->next_packet_size = be32_to_cpu(packet->next_packet_size); |
2a26c979 JQ |
864 | p->packet_num = be64_to_cpu(packet->packet_num); |
865 | ||
866 | if (p->pages->used) { | |
867 | /* make sure that ramblock is 0 terminated */ | |
868 | packet->ramblock[255] = 0; | |
869 | block = qemu_ram_block_by_name(packet->ramblock); | |
870 | if (!block) { | |
871 | error_setg(errp, "multifd: unknown ram block %s", | |
872 | packet->ramblock); | |
873 | return -1; | |
874 | } | |
875 | } | |
876 | ||
877 | for (i = 0; i < p->pages->used; i++) { | |
878 | ram_addr_t offset = be64_to_cpu(packet->offset[i]); | |
879 | ||
880 | if (offset > (block->used_length - TARGET_PAGE_SIZE)) { | |
881 | error_setg(errp, "multifd: offset too long " RAM_ADDR_FMT | |
882 | " (max " RAM_ADDR_FMT ")", | |
883 | offset, block->max_length); | |
884 | return -1; | |
885 | } | |
886 | p->pages->iov[i].iov_base = block->host + offset; | |
887 | p->pages->iov[i].iov_len = TARGET_PAGE_SIZE; | |
888 | } | |
889 | ||
890 | return 0; | |
891 | } | |
892 | ||
f986c3d2 JQ |
893 | struct { |
894 | MultiFDSendParams *params; | |
34c55a94 JQ |
895 | /* array of pages to sent */ |
896 | MultiFDPages_t *pages; | |
6df264ac JQ |
897 | /* global number of generated multifd packets */ |
898 | uint64_t packet_num; | |
b9ee2f7d JQ |
899 | /* send channels ready */ |
900 | QemuSemaphore channels_ready; | |
f986c3d2 JQ |
901 | } *multifd_send_state; |
902 | ||
b9ee2f7d JQ |
903 | /* |
904 | * How we use multifd_send_state->pages and channel->pages? | |
905 | * | |
906 | * We create a pages for each channel, and a main one. Each time that | |
907 | * we need to send a batch of pages we interchange the ones between | |
908 | * multifd_send_state and the channel that is sending it. There are | |
909 | * two reasons for that: | |
910 | * - to not have to do so many mallocs during migration | |
911 | * - to make easier to know what to free at the end of migration | |
912 | * | |
913 | * This way we always know who is the owner of each "pages" struct, | |
a5f7b1a6 | 914 | * and we don't need any locking. It belongs to the migration thread |
b9ee2f7d JQ |
915 | * or to the channel thread. Switching is safe because the migration |
916 | * thread is using the channel mutex when changing it, and the channel | |
917 | * have to had finish with its own, otherwise pending_job can't be | |
918 | * false. | |
919 | */ | |
920 | ||
1b81c974 | 921 | static int multifd_send_pages(RAMState *rs) |
b9ee2f7d JQ |
922 | { |
923 | int i; | |
924 | static int next_channel; | |
925 | MultiFDSendParams *p = NULL; /* make happy gcc */ | |
926 | MultiFDPages_t *pages = multifd_send_state->pages; | |
927 | uint64_t transferred; | |
928 | ||
929 | qemu_sem_wait(&multifd_send_state->channels_ready); | |
930 | for (i = next_channel;; i = (i + 1) % migrate_multifd_channels()) { | |
931 | p = &multifd_send_state->params[i]; | |
932 | ||
933 | qemu_mutex_lock(&p->mutex); | |
713f762a IR |
934 | if (p->quit) { |
935 | error_report("%s: channel %d has already quit!", __func__, i); | |
936 | qemu_mutex_unlock(&p->mutex); | |
937 | return -1; | |
938 | } | |
b9ee2f7d JQ |
939 | if (!p->pending_job) { |
940 | p->pending_job++; | |
941 | next_channel = (i + 1) % migrate_multifd_channels(); | |
942 | break; | |
943 | } | |
944 | qemu_mutex_unlock(&p->mutex); | |
945 | } | |
946 | p->pages->used = 0; | |
947 | ||
948 | p->packet_num = multifd_send_state->packet_num++; | |
949 | p->pages->block = NULL; | |
950 | multifd_send_state->pages = p->pages; | |
951 | p->pages = pages; | |
4fcefd44 | 952 | transferred = ((uint64_t) pages->used) * TARGET_PAGE_SIZE + p->packet_len; |
1b81c974 | 953 | qemu_file_update_transfer(rs->f, transferred); |
b9ee2f7d JQ |
954 | ram_counters.multifd_bytes += transferred; |
955 | ram_counters.transferred += transferred;; | |
956 | qemu_mutex_unlock(&p->mutex); | |
957 | qemu_sem_post(&p->sem); | |
713f762a IR |
958 | |
959 | return 1; | |
b9ee2f7d JQ |
960 | } |
961 | ||
1b81c974 | 962 | static int multifd_queue_page(RAMState *rs, RAMBlock *block, ram_addr_t offset) |
b9ee2f7d JQ |
963 | { |
964 | MultiFDPages_t *pages = multifd_send_state->pages; | |
965 | ||
966 | if (!pages->block) { | |
967 | pages->block = block; | |
968 | } | |
969 | ||
970 | if (pages->block == block) { | |
971 | pages->offset[pages->used] = offset; | |
972 | pages->iov[pages->used].iov_base = block->host + offset; | |
973 | pages->iov[pages->used].iov_len = TARGET_PAGE_SIZE; | |
974 | pages->used++; | |
975 | ||
976 | if (pages->used < pages->allocated) { | |
713f762a | 977 | return 1; |
b9ee2f7d JQ |
978 | } |
979 | } | |
980 | ||
1b81c974 | 981 | if (multifd_send_pages(rs) < 0) { |
713f762a IR |
982 | return -1; |
983 | } | |
b9ee2f7d JQ |
984 | |
985 | if (pages->block != block) { | |
1b81c974 | 986 | return multifd_queue_page(rs, block, offset); |
b9ee2f7d | 987 | } |
713f762a IR |
988 | |
989 | return 1; | |
b9ee2f7d JQ |
990 | } |
991 | ||
66770707 | 992 | static void multifd_send_terminate_threads(Error *err) |
f986c3d2 JQ |
993 | { |
994 | int i; | |
995 | ||
5558c91a JQ |
996 | trace_multifd_send_terminate_threads(err != NULL); |
997 | ||
7a169d74 JQ |
998 | if (err) { |
999 | MigrationState *s = migrate_get_current(); | |
1000 | migrate_set_error(s, err); | |
1001 | if (s->state == MIGRATION_STATUS_SETUP || | |
1002 | s->state == MIGRATION_STATUS_PRE_SWITCHOVER || | |
1003 | s->state == MIGRATION_STATUS_DEVICE || | |
1004 | s->state == MIGRATION_STATUS_ACTIVE) { | |
1005 | migrate_set_state(&s->state, s->state, | |
1006 | MIGRATION_STATUS_FAILED); | |
1007 | } | |
1008 | } | |
1009 | ||
66770707 | 1010 | for (i = 0; i < migrate_multifd_channels(); i++) { |
f986c3d2 JQ |
1011 | MultiFDSendParams *p = &multifd_send_state->params[i]; |
1012 | ||
1013 | qemu_mutex_lock(&p->mutex); | |
1014 | p->quit = true; | |
1015 | qemu_sem_post(&p->sem); | |
1016 | qemu_mutex_unlock(&p->mutex); | |
1017 | } | |
1018 | } | |
1019 | ||
1398b2e3 | 1020 | void multifd_save_cleanup(void) |
f986c3d2 JQ |
1021 | { |
1022 | int i; | |
f986c3d2 JQ |
1023 | |
1024 | if (!migrate_use_multifd()) { | |
1398b2e3 | 1025 | return; |
f986c3d2 | 1026 | } |
66770707 JQ |
1027 | multifd_send_terminate_threads(NULL); |
1028 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
f986c3d2 JQ |
1029 | MultiFDSendParams *p = &multifd_send_state->params[i]; |
1030 | ||
66770707 JQ |
1031 | if (p->running) { |
1032 | qemu_thread_join(&p->thread); | |
1033 | } | |
60df2d4a JQ |
1034 | socket_send_channel_destroy(p->c); |
1035 | p->c = NULL; | |
f986c3d2 JQ |
1036 | qemu_mutex_destroy(&p->mutex); |
1037 | qemu_sem_destroy(&p->sem); | |
18cdcea3 | 1038 | qemu_sem_destroy(&p->sem_sync); |
f986c3d2 JQ |
1039 | g_free(p->name); |
1040 | p->name = NULL; | |
34c55a94 JQ |
1041 | multifd_pages_clear(p->pages); |
1042 | p->pages = NULL; | |
2a26c979 JQ |
1043 | p->packet_len = 0; |
1044 | g_free(p->packet); | |
1045 | p->packet = NULL; | |
f986c3d2 | 1046 | } |
b9ee2f7d | 1047 | qemu_sem_destroy(&multifd_send_state->channels_ready); |
f986c3d2 JQ |
1048 | g_free(multifd_send_state->params); |
1049 | multifd_send_state->params = NULL; | |
34c55a94 JQ |
1050 | multifd_pages_clear(multifd_send_state->pages); |
1051 | multifd_send_state->pages = NULL; | |
f986c3d2 JQ |
1052 | g_free(multifd_send_state); |
1053 | multifd_send_state = NULL; | |
f986c3d2 JQ |
1054 | } |
1055 | ||
1b81c974 | 1056 | static void multifd_send_sync_main(RAMState *rs) |
6df264ac JQ |
1057 | { |
1058 | int i; | |
1059 | ||
1060 | if (!migrate_use_multifd()) { | |
1061 | return; | |
1062 | } | |
b9ee2f7d | 1063 | if (multifd_send_state->pages->used) { |
1b81c974 | 1064 | if (multifd_send_pages(rs) < 0) { |
713f762a IR |
1065 | error_report("%s: multifd_send_pages fail", __func__); |
1066 | return; | |
1067 | } | |
b9ee2f7d | 1068 | } |
6df264ac JQ |
1069 | for (i = 0; i < migrate_multifd_channels(); i++) { |
1070 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
1071 | ||
1072 | trace_multifd_send_sync_main_signal(p->id); | |
1073 | ||
1074 | qemu_mutex_lock(&p->mutex); | |
b9ee2f7d | 1075 | |
713f762a IR |
1076 | if (p->quit) { |
1077 | error_report("%s: channel %d has already quit", __func__, i); | |
1078 | qemu_mutex_unlock(&p->mutex); | |
1079 | return; | |
1080 | } | |
1081 | ||
b9ee2f7d | 1082 | p->packet_num = multifd_send_state->packet_num++; |
6df264ac JQ |
1083 | p->flags |= MULTIFD_FLAG_SYNC; |
1084 | p->pending_job++; | |
1b81c974 | 1085 | qemu_file_update_transfer(rs->f, p->packet_len); |
81507f6b IR |
1086 | ram_counters.multifd_bytes += p->packet_len; |
1087 | ram_counters.transferred += p->packet_len; | |
6df264ac JQ |
1088 | qemu_mutex_unlock(&p->mutex); |
1089 | qemu_sem_post(&p->sem); | |
1090 | } | |
1091 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
1092 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
1093 | ||
1094 | trace_multifd_send_sync_main_wait(p->id); | |
18cdcea3 | 1095 | qemu_sem_wait(&p->sem_sync); |
6df264ac JQ |
1096 | } |
1097 | trace_multifd_send_sync_main(multifd_send_state->packet_num); | |
1098 | } | |
1099 | ||
f986c3d2 JQ |
1100 | static void *multifd_send_thread(void *opaque) |
1101 | { | |
1102 | MultiFDSendParams *p = opaque; | |
af8b7d2b | 1103 | Error *local_err = NULL; |
a3ec6b7d IR |
1104 | int ret = 0; |
1105 | uint32_t flags = 0; | |
af8b7d2b | 1106 | |
408ea6ae | 1107 | trace_multifd_send_thread_start(p->id); |
74637e6f | 1108 | rcu_register_thread(); |
408ea6ae | 1109 | |
af8b7d2b | 1110 | if (multifd_send_initial_packet(p, &local_err) < 0) { |
2f4aefd3 | 1111 | ret = -1; |
af8b7d2b JQ |
1112 | goto out; |
1113 | } | |
408ea6ae JQ |
1114 | /* initial packet */ |
1115 | p->num_packets = 1; | |
f986c3d2 JQ |
1116 | |
1117 | while (true) { | |
d82628e4 | 1118 | qemu_sem_wait(&p->sem); |
f986c3d2 | 1119 | qemu_mutex_lock(&p->mutex); |
0beb5ed3 JQ |
1120 | |
1121 | if (p->pending_job) { | |
1122 | uint32_t used = p->pages->used; | |
1123 | uint64_t packet_num = p->packet_num; | |
a3ec6b7d | 1124 | flags = p->flags; |
0beb5ed3 | 1125 | |
2a34ee59 | 1126 | p->next_packet_size = used * qemu_target_page_size(); |
0beb5ed3 JQ |
1127 | multifd_send_fill_packet(p); |
1128 | p->flags = 0; | |
1129 | p->num_packets++; | |
1130 | p->num_pages += used; | |
0beb5ed3 JQ |
1131 | qemu_mutex_unlock(&p->mutex); |
1132 | ||
2a34ee59 JQ |
1133 | trace_multifd_send(p->id, packet_num, used, flags, |
1134 | p->next_packet_size); | |
0beb5ed3 | 1135 | |
8b2db7f5 JQ |
1136 | ret = qio_channel_write_all(p->c, (void *)p->packet, |
1137 | p->packet_len, &local_err); | |
1138 | if (ret != 0) { | |
1139 | break; | |
1140 | } | |
1141 | ||
ad24c7cb JQ |
1142 | if (used) { |
1143 | ret = qio_channel_writev_all(p->c, p->pages->iov, | |
1144 | used, &local_err); | |
1145 | if (ret != 0) { | |
1146 | break; | |
1147 | } | |
8b2db7f5 | 1148 | } |
0beb5ed3 JQ |
1149 | |
1150 | qemu_mutex_lock(&p->mutex); | |
1151 | p->pending_job--; | |
1152 | qemu_mutex_unlock(&p->mutex); | |
6df264ac JQ |
1153 | |
1154 | if (flags & MULTIFD_FLAG_SYNC) { | |
18cdcea3 | 1155 | qemu_sem_post(&p->sem_sync); |
6df264ac | 1156 | } |
b9ee2f7d | 1157 | qemu_sem_post(&multifd_send_state->channels_ready); |
0beb5ed3 | 1158 | } else if (p->quit) { |
f986c3d2 JQ |
1159 | qemu_mutex_unlock(&p->mutex); |
1160 | break; | |
6df264ac JQ |
1161 | } else { |
1162 | qemu_mutex_unlock(&p->mutex); | |
1163 | /* sometimes there are spurious wakeups */ | |
f986c3d2 | 1164 | } |
f986c3d2 JQ |
1165 | } |
1166 | ||
af8b7d2b JQ |
1167 | out: |
1168 | if (local_err) { | |
7dd59d01 | 1169 | trace_multifd_send_error(p->id); |
af8b7d2b JQ |
1170 | multifd_send_terminate_threads(local_err); |
1171 | } | |
1172 | ||
a3ec6b7d IR |
1173 | /* |
1174 | * Error happen, I will exit, but I can't just leave, tell | |
1175 | * who pay attention to me. | |
1176 | */ | |
1177 | if (ret != 0) { | |
2f4aefd3 | 1178 | qemu_sem_post(&p->sem_sync); |
a3ec6b7d IR |
1179 | qemu_sem_post(&multifd_send_state->channels_ready); |
1180 | } | |
1181 | ||
66770707 JQ |
1182 | qemu_mutex_lock(&p->mutex); |
1183 | p->running = false; | |
1184 | qemu_mutex_unlock(&p->mutex); | |
1185 | ||
74637e6f | 1186 | rcu_unregister_thread(); |
408ea6ae JQ |
1187 | trace_multifd_send_thread_end(p->id, p->num_packets, p->num_pages); |
1188 | ||
f986c3d2 JQ |
1189 | return NULL; |
1190 | } | |
1191 | ||
60df2d4a JQ |
1192 | static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque) |
1193 | { | |
1194 | MultiFDSendParams *p = opaque; | |
1195 | QIOChannel *sioc = QIO_CHANNEL(qio_task_get_source(task)); | |
1196 | Error *local_err = NULL; | |
1197 | ||
7dd59d01 | 1198 | trace_multifd_new_send_channel_async(p->id); |
60df2d4a | 1199 | if (qio_task_propagate_error(task, &local_err)) { |
1398b2e3 FL |
1200 | migrate_set_error(migrate_get_current(), local_err); |
1201 | multifd_save_cleanup(); | |
60df2d4a JQ |
1202 | } else { |
1203 | p->c = QIO_CHANNEL(sioc); | |
1204 | qio_channel_set_delay(p->c, false); | |
1205 | p->running = true; | |
1206 | qemu_thread_create(&p->thread, p->name, multifd_send_thread, p, | |
1207 | QEMU_THREAD_JOINABLE); | |
60df2d4a JQ |
1208 | } |
1209 | } | |
1210 | ||
f986c3d2 JQ |
1211 | int multifd_save_setup(void) |
1212 | { | |
1213 | int thread_count; | |
efd1a1d6 | 1214 | uint32_t page_count = MULTIFD_PACKET_SIZE / qemu_target_page_size(); |
f986c3d2 JQ |
1215 | uint8_t i; |
1216 | ||
1217 | if (!migrate_use_multifd()) { | |
1218 | return 0; | |
1219 | } | |
1220 | thread_count = migrate_multifd_channels(); | |
1221 | multifd_send_state = g_malloc0(sizeof(*multifd_send_state)); | |
1222 | multifd_send_state->params = g_new0(MultiFDSendParams, thread_count); | |
34c55a94 | 1223 | multifd_send_state->pages = multifd_pages_init(page_count); |
b9ee2f7d | 1224 | qemu_sem_init(&multifd_send_state->channels_ready, 0); |
34c55a94 | 1225 | |
f986c3d2 JQ |
1226 | for (i = 0; i < thread_count; i++) { |
1227 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
1228 | ||
1229 | qemu_mutex_init(&p->mutex); | |
1230 | qemu_sem_init(&p->sem, 0); | |
18cdcea3 | 1231 | qemu_sem_init(&p->sem_sync, 0); |
f986c3d2 | 1232 | p->quit = false; |
0beb5ed3 | 1233 | p->pending_job = 0; |
f986c3d2 | 1234 | p->id = i; |
34c55a94 | 1235 | p->pages = multifd_pages_init(page_count); |
2a26c979 JQ |
1236 | p->packet_len = sizeof(MultiFDPacket_t) |
1237 | + sizeof(ram_addr_t) * page_count; | |
1238 | p->packet = g_malloc0(p->packet_len); | |
9985e1f4 WY |
1239 | p->packet->magic = cpu_to_be32(MULTIFD_MAGIC); |
1240 | p->packet->version = cpu_to_be32(MULTIFD_VERSION); | |
f986c3d2 | 1241 | p->name = g_strdup_printf("multifdsend_%d", i); |
60df2d4a | 1242 | socket_send_channel_create(multifd_new_send_channel_async, p); |
f986c3d2 JQ |
1243 | } |
1244 | return 0; | |
1245 | } | |
1246 | ||
f986c3d2 JQ |
1247 | struct { |
1248 | MultiFDRecvParams *params; | |
1249 | /* number of created threads */ | |
1250 | int count; | |
6df264ac JQ |
1251 | /* syncs main thread and channels */ |
1252 | QemuSemaphore sem_sync; | |
1253 | /* global number of generated multifd packets */ | |
1254 | uint64_t packet_num; | |
f986c3d2 JQ |
1255 | } *multifd_recv_state; |
1256 | ||
66770707 | 1257 | static void multifd_recv_terminate_threads(Error *err) |
f986c3d2 JQ |
1258 | { |
1259 | int i; | |
1260 | ||
5558c91a JQ |
1261 | trace_multifd_recv_terminate_threads(err != NULL); |
1262 | ||
7a169d74 JQ |
1263 | if (err) { |
1264 | MigrationState *s = migrate_get_current(); | |
1265 | migrate_set_error(s, err); | |
1266 | if (s->state == MIGRATION_STATUS_SETUP || | |
1267 | s->state == MIGRATION_STATUS_ACTIVE) { | |
1268 | migrate_set_state(&s->state, s->state, | |
1269 | MIGRATION_STATUS_FAILED); | |
1270 | } | |
1271 | } | |
1272 | ||
66770707 | 1273 | for (i = 0; i < migrate_multifd_channels(); i++) { |
f986c3d2 JQ |
1274 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; |
1275 | ||
1276 | qemu_mutex_lock(&p->mutex); | |
3c3ca25d | 1277 | p->quit = true; |
7a5cc33c JQ |
1278 | /* We could arrive here for two reasons: |
1279 | - normal quit, i.e. everything went fine, just finished | |
1280 | - error quit: We close the channels so the channel threads | |
1281 | finish the qio_channel_read_all_eof() */ | |
1282 | qio_channel_shutdown(p->c, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); | |
f986c3d2 JQ |
1283 | qemu_mutex_unlock(&p->mutex); |
1284 | } | |
1285 | } | |
1286 | ||
1287 | int multifd_load_cleanup(Error **errp) | |
1288 | { | |
1289 | int i; | |
1290 | int ret = 0; | |
1291 | ||
1292 | if (!migrate_use_multifd()) { | |
1293 | return 0; | |
1294 | } | |
66770707 JQ |
1295 | multifd_recv_terminate_threads(NULL); |
1296 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
f986c3d2 JQ |
1297 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; |
1298 | ||
66770707 | 1299 | if (p->running) { |
3c3ca25d | 1300 | p->quit = true; |
f193bc0c IR |
1301 | /* |
1302 | * multifd_recv_thread may hung at MULTIFD_FLAG_SYNC handle code, | |
1303 | * however try to wakeup it without harm in cleanup phase. | |
1304 | */ | |
1305 | qemu_sem_post(&p->sem_sync); | |
66770707 JQ |
1306 | qemu_thread_join(&p->thread); |
1307 | } | |
60df2d4a JQ |
1308 | object_unref(OBJECT(p->c)); |
1309 | p->c = NULL; | |
f986c3d2 | 1310 | qemu_mutex_destroy(&p->mutex); |
6df264ac | 1311 | qemu_sem_destroy(&p->sem_sync); |
f986c3d2 JQ |
1312 | g_free(p->name); |
1313 | p->name = NULL; | |
34c55a94 JQ |
1314 | multifd_pages_clear(p->pages); |
1315 | p->pages = NULL; | |
2a26c979 JQ |
1316 | p->packet_len = 0; |
1317 | g_free(p->packet); | |
1318 | p->packet = NULL; | |
f986c3d2 | 1319 | } |
6df264ac | 1320 | qemu_sem_destroy(&multifd_recv_state->sem_sync); |
f986c3d2 JQ |
1321 | g_free(multifd_recv_state->params); |
1322 | multifd_recv_state->params = NULL; | |
1323 | g_free(multifd_recv_state); | |
1324 | multifd_recv_state = NULL; | |
1325 | ||
1326 | return ret; | |
1327 | } | |
1328 | ||
6df264ac JQ |
1329 | static void multifd_recv_sync_main(void) |
1330 | { | |
1331 | int i; | |
1332 | ||
1333 | if (!migrate_use_multifd()) { | |
1334 | return; | |
1335 | } | |
1336 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
1337 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
1338 | ||
6df264ac JQ |
1339 | trace_multifd_recv_sync_main_wait(p->id); |
1340 | qemu_sem_wait(&multifd_recv_state->sem_sync); | |
77568ea7 WY |
1341 | } |
1342 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
1343 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
1344 | ||
6df264ac JQ |
1345 | qemu_mutex_lock(&p->mutex); |
1346 | if (multifd_recv_state->packet_num < p->packet_num) { | |
1347 | multifd_recv_state->packet_num = p->packet_num; | |
1348 | } | |
1349 | qemu_mutex_unlock(&p->mutex); | |
6df264ac | 1350 | trace_multifd_recv_sync_main_signal(p->id); |
6df264ac JQ |
1351 | qemu_sem_post(&p->sem_sync); |
1352 | } | |
1353 | trace_multifd_recv_sync_main(multifd_recv_state->packet_num); | |
1354 | } | |
1355 | ||
f986c3d2 JQ |
1356 | static void *multifd_recv_thread(void *opaque) |
1357 | { | |
1358 | MultiFDRecvParams *p = opaque; | |
2a26c979 JQ |
1359 | Error *local_err = NULL; |
1360 | int ret; | |
f986c3d2 | 1361 | |
408ea6ae | 1362 | trace_multifd_recv_thread_start(p->id); |
74637e6f | 1363 | rcu_register_thread(); |
408ea6ae | 1364 | |
f986c3d2 | 1365 | while (true) { |
6df264ac JQ |
1366 | uint32_t used; |
1367 | uint32_t flags; | |
0beb5ed3 | 1368 | |
3c3ca25d JQ |
1369 | if (p->quit) { |
1370 | break; | |
1371 | } | |
1372 | ||
8b2db7f5 JQ |
1373 | ret = qio_channel_read_all_eof(p->c, (void *)p->packet, |
1374 | p->packet_len, &local_err); | |
1375 | if (ret == 0) { /* EOF */ | |
1376 | break; | |
1377 | } | |
1378 | if (ret == -1) { /* Error */ | |
1379 | break; | |
1380 | } | |
2a26c979 | 1381 | |
6df264ac JQ |
1382 | qemu_mutex_lock(&p->mutex); |
1383 | ret = multifd_recv_unfill_packet(p, &local_err); | |
1384 | if (ret) { | |
f986c3d2 JQ |
1385 | qemu_mutex_unlock(&p->mutex); |
1386 | break; | |
1387 | } | |
6df264ac JQ |
1388 | |
1389 | used = p->pages->used; | |
1390 | flags = p->flags; | |
2a34ee59 JQ |
1391 | trace_multifd_recv(p->id, p->packet_num, used, flags, |
1392 | p->next_packet_size); | |
6df264ac JQ |
1393 | p->num_packets++; |
1394 | p->num_pages += used; | |
f986c3d2 | 1395 | qemu_mutex_unlock(&p->mutex); |
6df264ac | 1396 | |
ad24c7cb JQ |
1397 | if (used) { |
1398 | ret = qio_channel_readv_all(p->c, p->pages->iov, | |
1399 | used, &local_err); | |
1400 | if (ret != 0) { | |
1401 | break; | |
1402 | } | |
8b2db7f5 JQ |
1403 | } |
1404 | ||
6df264ac JQ |
1405 | if (flags & MULTIFD_FLAG_SYNC) { |
1406 | qemu_sem_post(&multifd_recv_state->sem_sync); | |
1407 | qemu_sem_wait(&p->sem_sync); | |
1408 | } | |
f986c3d2 JQ |
1409 | } |
1410 | ||
d82628e4 JQ |
1411 | if (local_err) { |
1412 | multifd_recv_terminate_threads(local_err); | |
1413 | } | |
66770707 JQ |
1414 | qemu_mutex_lock(&p->mutex); |
1415 | p->running = false; | |
1416 | qemu_mutex_unlock(&p->mutex); | |
1417 | ||
74637e6f | 1418 | rcu_unregister_thread(); |
408ea6ae JQ |
1419 | trace_multifd_recv_thread_end(p->id, p->num_packets, p->num_pages); |
1420 | ||
f986c3d2 JQ |
1421 | return NULL; |
1422 | } | |
1423 | ||
1424 | int multifd_load_setup(void) | |
1425 | { | |
1426 | int thread_count; | |
efd1a1d6 | 1427 | uint32_t page_count = MULTIFD_PACKET_SIZE / qemu_target_page_size(); |
f986c3d2 JQ |
1428 | uint8_t i; |
1429 | ||
1430 | if (!migrate_use_multifd()) { | |
1431 | return 0; | |
1432 | } | |
1433 | thread_count = migrate_multifd_channels(); | |
1434 | multifd_recv_state = g_malloc0(sizeof(*multifd_recv_state)); | |
1435 | multifd_recv_state->params = g_new0(MultiFDRecvParams, thread_count); | |
66770707 | 1436 | atomic_set(&multifd_recv_state->count, 0); |
6df264ac | 1437 | qemu_sem_init(&multifd_recv_state->sem_sync, 0); |
34c55a94 | 1438 | |
f986c3d2 JQ |
1439 | for (i = 0; i < thread_count; i++) { |
1440 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
1441 | ||
1442 | qemu_mutex_init(&p->mutex); | |
6df264ac | 1443 | qemu_sem_init(&p->sem_sync, 0); |
3c3ca25d | 1444 | p->quit = false; |
f986c3d2 | 1445 | p->id = i; |
34c55a94 | 1446 | p->pages = multifd_pages_init(page_count); |
2a26c979 JQ |
1447 | p->packet_len = sizeof(MultiFDPacket_t) |
1448 | + sizeof(ram_addr_t) * page_count; | |
1449 | p->packet = g_malloc0(p->packet_len); | |
f986c3d2 | 1450 | p->name = g_strdup_printf("multifdrecv_%d", i); |
f986c3d2 JQ |
1451 | } |
1452 | return 0; | |
1453 | } | |
1454 | ||
62c1e0ca JQ |
1455 | bool multifd_recv_all_channels_created(void) |
1456 | { | |
1457 | int thread_count = migrate_multifd_channels(); | |
1458 | ||
1459 | if (!migrate_use_multifd()) { | |
1460 | return true; | |
1461 | } | |
1462 | ||
1463 | return thread_count == atomic_read(&multifd_recv_state->count); | |
1464 | } | |
1465 | ||
49ed0d24 FL |
1466 | /* |
1467 | * Try to receive all multifd channels to get ready for the migration. | |
1468 | * - Return true and do not set @errp when correctly receving all channels; | |
1469 | * - Return false and do not set @errp when correctly receiving the current one; | |
1470 | * - Return false and set @errp when failing to receive the current channel. | |
1471 | */ | |
1472 | bool multifd_recv_new_channel(QIOChannel *ioc, Error **errp) | |
71bb07db | 1473 | { |
60df2d4a | 1474 | MultiFDRecvParams *p; |
af8b7d2b JQ |
1475 | Error *local_err = NULL; |
1476 | int id; | |
60df2d4a | 1477 | |
af8b7d2b JQ |
1478 | id = multifd_recv_initial_packet(ioc, &local_err); |
1479 | if (id < 0) { | |
1480 | multifd_recv_terminate_threads(local_err); | |
49ed0d24 FL |
1481 | error_propagate_prepend(errp, local_err, |
1482 | "failed to receive packet" | |
1483 | " via multifd channel %d: ", | |
1484 | atomic_read(&multifd_recv_state->count)); | |
81e62053 | 1485 | return false; |
af8b7d2b | 1486 | } |
7dd59d01 | 1487 | trace_multifd_recv_new_channel(id); |
af8b7d2b JQ |
1488 | |
1489 | p = &multifd_recv_state->params[id]; | |
1490 | if (p->c != NULL) { | |
1491 | error_setg(&local_err, "multifd: received id '%d' already setup'", | |
1492 | id); | |
1493 | multifd_recv_terminate_threads(local_err); | |
49ed0d24 | 1494 | error_propagate(errp, local_err); |
81e62053 | 1495 | return false; |
af8b7d2b | 1496 | } |
60df2d4a JQ |
1497 | p->c = ioc; |
1498 | object_ref(OBJECT(ioc)); | |
408ea6ae JQ |
1499 | /* initial packet */ |
1500 | p->num_packets = 1; | |
60df2d4a JQ |
1501 | |
1502 | p->running = true; | |
1503 | qemu_thread_create(&p->thread, p->name, multifd_recv_thread, p, | |
1504 | QEMU_THREAD_JOINABLE); | |
1505 | atomic_inc(&multifd_recv_state->count); | |
49ed0d24 FL |
1506 | return atomic_read(&multifd_recv_state->count) == |
1507 | migrate_multifd_channels(); | |
71bb07db JQ |
1508 | } |
1509 | ||
56e93d26 | 1510 | /** |
3d0684b2 | 1511 | * save_page_header: write page header to wire |
56e93d26 JQ |
1512 | * |
1513 | * If this is the 1st block, it also writes the block identification | |
1514 | * | |
3d0684b2 | 1515 | * Returns the number of bytes written |
56e93d26 JQ |
1516 | * |
1517 | * @f: QEMUFile where to send the data | |
1518 | * @block: block that contains the page we want to send | |
1519 | * @offset: offset inside the block for the page | |
1520 | * in the lower bits, it contains flags | |
1521 | */ | |
2bf3aa85 JQ |
1522 | static size_t save_page_header(RAMState *rs, QEMUFile *f, RAMBlock *block, |
1523 | ram_addr_t offset) | |
56e93d26 | 1524 | { |
9f5f380b | 1525 | size_t size, len; |
56e93d26 | 1526 | |
24795694 JQ |
1527 | if (block == rs->last_sent_block) { |
1528 | offset |= RAM_SAVE_FLAG_CONTINUE; | |
1529 | } | |
2bf3aa85 | 1530 | qemu_put_be64(f, offset); |
56e93d26 JQ |
1531 | size = 8; |
1532 | ||
1533 | if (!(offset & RAM_SAVE_FLAG_CONTINUE)) { | |
9f5f380b | 1534 | len = strlen(block->idstr); |
2bf3aa85 JQ |
1535 | qemu_put_byte(f, len); |
1536 | qemu_put_buffer(f, (uint8_t *)block->idstr, len); | |
9f5f380b | 1537 | size += 1 + len; |
24795694 | 1538 | rs->last_sent_block = block; |
56e93d26 JQ |
1539 | } |
1540 | return size; | |
1541 | } | |
1542 | ||
3d0684b2 JQ |
1543 | /** |
1544 | * mig_throttle_guest_down: throotle down the guest | |
1545 | * | |
1546 | * Reduce amount of guest cpu execution to hopefully slow down memory | |
1547 | * writes. If guest dirty memory rate is reduced below the rate at | |
1548 | * which we can transfer pages to the destination then we should be | |
1549 | * able to complete migration. Some workloads dirty memory way too | |
1550 | * fast and will not effectively converge, even with auto-converge. | |
070afca2 JH |
1551 | */ |
1552 | static void mig_throttle_guest_down(void) | |
1553 | { | |
1554 | MigrationState *s = migrate_get_current(); | |
2594f56d DB |
1555 | uint64_t pct_initial = s->parameters.cpu_throttle_initial; |
1556 | uint64_t pct_icrement = s->parameters.cpu_throttle_increment; | |
4cbc9c7f | 1557 | int pct_max = s->parameters.max_cpu_throttle; |
070afca2 JH |
1558 | |
1559 | /* We have not started throttling yet. Let's start it. */ | |
1560 | if (!cpu_throttle_active()) { | |
1561 | cpu_throttle_set(pct_initial); | |
1562 | } else { | |
1563 | /* Throttling already on, just increase the rate */ | |
4cbc9c7f LQ |
1564 | cpu_throttle_set(MIN(cpu_throttle_get_percentage() + pct_icrement, |
1565 | pct_max)); | |
070afca2 JH |
1566 | } |
1567 | } | |
1568 | ||
3d0684b2 JQ |
1569 | /** |
1570 | * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache | |
1571 | * | |
6f37bb8b | 1572 | * @rs: current RAM state |
3d0684b2 JQ |
1573 | * @current_addr: address for the zero page |
1574 | * | |
1575 | * Update the xbzrle cache to reflect a page that's been sent as all 0. | |
56e93d26 JQ |
1576 | * The important thing is that a stale (not-yet-0'd) page be replaced |
1577 | * by the new data. | |
1578 | * As a bonus, if the page wasn't in the cache it gets added so that | |
3d0684b2 | 1579 | * when a small write is made into the 0'd page it gets XBZRLE sent. |
56e93d26 | 1580 | */ |
6f37bb8b | 1581 | static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr) |
56e93d26 | 1582 | { |
6f37bb8b | 1583 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { |
56e93d26 JQ |
1584 | return; |
1585 | } | |
1586 | ||
1587 | /* We don't care if this fails to allocate a new cache page | |
1588 | * as long as it updated an old one */ | |
c00e0928 | 1589 | cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page, |
9360447d | 1590 | ram_counters.dirty_sync_count); |
56e93d26 JQ |
1591 | } |
1592 | ||
1593 | #define ENCODING_FLAG_XBZRLE 0x1 | |
1594 | ||
1595 | /** | |
1596 | * save_xbzrle_page: compress and send current page | |
1597 | * | |
1598 | * Returns: 1 means that we wrote the page | |
1599 | * 0 means that page is identical to the one already sent | |
1600 | * -1 means that xbzrle would be longer than normal | |
1601 | * | |
5a987738 | 1602 | * @rs: current RAM state |
3d0684b2 JQ |
1603 | * @current_data: pointer to the address of the page contents |
1604 | * @current_addr: addr of the page | |
56e93d26 JQ |
1605 | * @block: block that contains the page we want to send |
1606 | * @offset: offset inside the block for the page | |
1607 | * @last_stage: if we are at the completion stage | |
56e93d26 | 1608 | */ |
204b88b8 | 1609 | static int save_xbzrle_page(RAMState *rs, uint8_t **current_data, |
56e93d26 | 1610 | ram_addr_t current_addr, RAMBlock *block, |
072c2511 | 1611 | ram_addr_t offset, bool last_stage) |
56e93d26 JQ |
1612 | { |
1613 | int encoded_len = 0, bytes_xbzrle; | |
1614 | uint8_t *prev_cached_page; | |
1615 | ||
9360447d JQ |
1616 | if (!cache_is_cached(XBZRLE.cache, current_addr, |
1617 | ram_counters.dirty_sync_count)) { | |
1618 | xbzrle_counters.cache_miss++; | |
56e93d26 JQ |
1619 | if (!last_stage) { |
1620 | if (cache_insert(XBZRLE.cache, current_addr, *current_data, | |
9360447d | 1621 | ram_counters.dirty_sync_count) == -1) { |
56e93d26 JQ |
1622 | return -1; |
1623 | } else { | |
1624 | /* update *current_data when the page has been | |
1625 | inserted into cache */ | |
1626 | *current_data = get_cached_data(XBZRLE.cache, current_addr); | |
1627 | } | |
1628 | } | |
1629 | return -1; | |
1630 | } | |
1631 | ||
1632 | prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); | |
1633 | ||
1634 | /* save current buffer into memory */ | |
1635 | memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); | |
1636 | ||
1637 | /* XBZRLE encoding (if there is no overflow) */ | |
1638 | encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, | |
1639 | TARGET_PAGE_SIZE, XBZRLE.encoded_buf, | |
1640 | TARGET_PAGE_SIZE); | |
ca353803 WY |
1641 | |
1642 | /* | |
1643 | * Update the cache contents, so that it corresponds to the data | |
1644 | * sent, in all cases except where we skip the page. | |
1645 | */ | |
1646 | if (!last_stage && encoded_len != 0) { | |
1647 | memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); | |
1648 | /* | |
1649 | * In the case where we couldn't compress, ensure that the caller | |
1650 | * sends the data from the cache, since the guest might have | |
1651 | * changed the RAM since we copied it. | |
1652 | */ | |
1653 | *current_data = prev_cached_page; | |
1654 | } | |
1655 | ||
56e93d26 | 1656 | if (encoded_len == 0) { |
55c4446b | 1657 | trace_save_xbzrle_page_skipping(); |
56e93d26 JQ |
1658 | return 0; |
1659 | } else if (encoded_len == -1) { | |
55c4446b | 1660 | trace_save_xbzrle_page_overflow(); |
9360447d | 1661 | xbzrle_counters.overflow++; |
56e93d26 JQ |
1662 | return -1; |
1663 | } | |
1664 | ||
56e93d26 | 1665 | /* Send XBZRLE based compressed page */ |
2bf3aa85 | 1666 | bytes_xbzrle = save_page_header(rs, rs->f, block, |
204b88b8 JQ |
1667 | offset | RAM_SAVE_FLAG_XBZRLE); |
1668 | qemu_put_byte(rs->f, ENCODING_FLAG_XBZRLE); | |
1669 | qemu_put_be16(rs->f, encoded_len); | |
1670 | qemu_put_buffer(rs->f, XBZRLE.encoded_buf, encoded_len); | |
56e93d26 | 1671 | bytes_xbzrle += encoded_len + 1 + 2; |
9360447d JQ |
1672 | xbzrle_counters.pages++; |
1673 | xbzrle_counters.bytes += bytes_xbzrle; | |
1674 | ram_counters.transferred += bytes_xbzrle; | |
56e93d26 JQ |
1675 | |
1676 | return 1; | |
1677 | } | |
1678 | ||
3d0684b2 JQ |
1679 | /** |
1680 | * migration_bitmap_find_dirty: find the next dirty page from start | |
f3f491fc | 1681 | * |
a5f7b1a6 | 1682 | * Returns the page offset within memory region of the start of a dirty page |
3d0684b2 | 1683 | * |
6f37bb8b | 1684 | * @rs: current RAM state |
3d0684b2 | 1685 | * @rb: RAMBlock where to search for dirty pages |
a935e30f | 1686 | * @start: page where we start the search |
f3f491fc | 1687 | */ |
56e93d26 | 1688 | static inline |
a935e30f | 1689 | unsigned long migration_bitmap_find_dirty(RAMState *rs, RAMBlock *rb, |
f20e2865 | 1690 | unsigned long start) |
56e93d26 | 1691 | { |
6b6712ef JQ |
1692 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; |
1693 | unsigned long *bitmap = rb->bmap; | |
56e93d26 JQ |
1694 | unsigned long next; |
1695 | ||
fbd162e6 | 1696 | if (ramblock_is_ignored(rb)) { |
b895de50 CLG |
1697 | return size; |
1698 | } | |
1699 | ||
6eeb63f7 WW |
1700 | /* |
1701 | * When the free page optimization is enabled, we need to check the bitmap | |
1702 | * to send the non-free pages rather than all the pages in the bulk stage. | |
1703 | */ | |
1704 | if (!rs->fpo_enabled && rs->ram_bulk_stage && start > 0) { | |
6b6712ef | 1705 | next = start + 1; |
56e93d26 | 1706 | } else { |
6b6712ef | 1707 | next = find_next_bit(bitmap, size, start); |
56e93d26 JQ |
1708 | } |
1709 | ||
6b6712ef | 1710 | return next; |
56e93d26 JQ |
1711 | } |
1712 | ||
06b10688 | 1713 | static inline bool migration_bitmap_clear_dirty(RAMState *rs, |
f20e2865 JQ |
1714 | RAMBlock *rb, |
1715 | unsigned long page) | |
a82d593b DDAG |
1716 | { |
1717 | bool ret; | |
a82d593b | 1718 | |
386a907b | 1719 | qemu_mutex_lock(&rs->bitmap_mutex); |
002cad6b PX |
1720 | |
1721 | /* | |
1722 | * Clear dirty bitmap if needed. This _must_ be called before we | |
1723 | * send any of the page in the chunk because we need to make sure | |
1724 | * we can capture further page content changes when we sync dirty | |
1725 | * log the next time. So as long as we are going to send any of | |
1726 | * the page in the chunk we clear the remote dirty bitmap for all. | |
1727 | * Clearing it earlier won't be a problem, but too late will. | |
1728 | */ | |
1729 | if (rb->clear_bmap && clear_bmap_test_and_clear(rb, page)) { | |
1730 | uint8_t shift = rb->clear_bmap_shift; | |
1731 | hwaddr size = 1ULL << (TARGET_PAGE_BITS + shift); | |
1732 | hwaddr start = (page << TARGET_PAGE_BITS) & (-size); | |
1733 | ||
1734 | /* | |
1735 | * CLEAR_BITMAP_SHIFT_MIN should always guarantee this... this | |
1736 | * can make things easier sometimes since then start address | |
1737 | * of the small chunk will always be 64 pages aligned so the | |
1738 | * bitmap will always be aligned to unsigned long. We should | |
1739 | * even be able to remove this restriction but I'm simply | |
1740 | * keeping it. | |
1741 | */ | |
1742 | assert(shift >= 6); | |
1743 | trace_migration_bitmap_clear_dirty(rb->idstr, start, size, page); | |
1744 | memory_region_clear_dirty_bitmap(rb->mr, start, size); | |
1745 | } | |
1746 | ||
6b6712ef | 1747 | ret = test_and_clear_bit(page, rb->bmap); |
a82d593b DDAG |
1748 | |
1749 | if (ret) { | |
0d8ec885 | 1750 | rs->migration_dirty_pages--; |
a82d593b | 1751 | } |
386a907b WW |
1752 | qemu_mutex_unlock(&rs->bitmap_mutex); |
1753 | ||
a82d593b DDAG |
1754 | return ret; |
1755 | } | |
1756 | ||
267691b6 | 1757 | /* Called with RCU critical section */ |
7a3e9571 | 1758 | static void ramblock_sync_dirty_bitmap(RAMState *rs, RAMBlock *rb) |
56e93d26 | 1759 | { |
0d8ec885 | 1760 | rs->migration_dirty_pages += |
5d0980a4 | 1761 | cpu_physical_memory_sync_dirty_bitmap(rb, 0, rb->used_length, |
0d8ec885 | 1762 | &rs->num_dirty_pages_period); |
56e93d26 JQ |
1763 | } |
1764 | ||
3d0684b2 JQ |
1765 | /** |
1766 | * ram_pagesize_summary: calculate all the pagesizes of a VM | |
1767 | * | |
1768 | * Returns a summary bitmap of the page sizes of all RAMBlocks | |
1769 | * | |
1770 | * For VMs with just normal pages this is equivalent to the host page | |
1771 | * size. If it's got some huge pages then it's the OR of all the | |
1772 | * different page sizes. | |
e8ca1db2 DDAG |
1773 | */ |
1774 | uint64_t ram_pagesize_summary(void) | |
1775 | { | |
1776 | RAMBlock *block; | |
1777 | uint64_t summary = 0; | |
1778 | ||
fbd162e6 | 1779 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
e8ca1db2 DDAG |
1780 | summary |= block->page_size; |
1781 | } | |
1782 | ||
1783 | return summary; | |
1784 | } | |
1785 | ||
aecbfe9c XG |
1786 | uint64_t ram_get_total_transferred_pages(void) |
1787 | { | |
1788 | return ram_counters.normal + ram_counters.duplicate + | |
1789 | compression_counters.pages + xbzrle_counters.pages; | |
1790 | } | |
1791 | ||
b734035b XG |
1792 | static void migration_update_rates(RAMState *rs, int64_t end_time) |
1793 | { | |
be8b02ed | 1794 | uint64_t page_count = rs->target_page_count - rs->target_page_count_prev; |
76e03000 | 1795 | double compressed_size; |
b734035b XG |
1796 | |
1797 | /* calculate period counters */ | |
1798 | ram_counters.dirty_pages_rate = rs->num_dirty_pages_period * 1000 | |
1799 | / (end_time - rs->time_last_bitmap_sync); | |
1800 | ||
be8b02ed | 1801 | if (!page_count) { |
b734035b XG |
1802 | return; |
1803 | } | |
1804 | ||
1805 | if (migrate_use_xbzrle()) { | |
1806 | xbzrle_counters.cache_miss_rate = (double)(xbzrle_counters.cache_miss - | |
be8b02ed | 1807 | rs->xbzrle_cache_miss_prev) / page_count; |
b734035b XG |
1808 | rs->xbzrle_cache_miss_prev = xbzrle_counters.cache_miss; |
1809 | } | |
76e03000 XG |
1810 | |
1811 | if (migrate_use_compression()) { | |
1812 | compression_counters.busy_rate = (double)(compression_counters.busy - | |
1813 | rs->compress_thread_busy_prev) / page_count; | |
1814 | rs->compress_thread_busy_prev = compression_counters.busy; | |
1815 | ||
1816 | compressed_size = compression_counters.compressed_size - | |
1817 | rs->compressed_size_prev; | |
1818 | if (compressed_size) { | |
1819 | double uncompressed_size = (compression_counters.pages - | |
1820 | rs->compress_pages_prev) * TARGET_PAGE_SIZE; | |
1821 | ||
1822 | /* Compression-Ratio = Uncompressed-size / Compressed-size */ | |
1823 | compression_counters.compression_rate = | |
1824 | uncompressed_size / compressed_size; | |
1825 | ||
1826 | rs->compress_pages_prev = compression_counters.pages; | |
1827 | rs->compressed_size_prev = compression_counters.compressed_size; | |
1828 | } | |
1829 | } | |
b734035b XG |
1830 | } |
1831 | ||
8d820d6f | 1832 | static void migration_bitmap_sync(RAMState *rs) |
56e93d26 JQ |
1833 | { |
1834 | RAMBlock *block; | |
56e93d26 | 1835 | int64_t end_time; |
c4bdf0cf | 1836 | uint64_t bytes_xfer_now; |
56e93d26 | 1837 | |
9360447d | 1838 | ram_counters.dirty_sync_count++; |
56e93d26 | 1839 | |
f664da80 JQ |
1840 | if (!rs->time_last_bitmap_sync) { |
1841 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56e93d26 JQ |
1842 | } |
1843 | ||
1844 | trace_migration_bitmap_sync_start(); | |
9c1f8f44 | 1845 | memory_global_dirty_log_sync(); |
56e93d26 | 1846 | |
108cfae0 | 1847 | qemu_mutex_lock(&rs->bitmap_mutex); |
89ac5a1d DDAG |
1848 | WITH_RCU_READ_LOCK_GUARD() { |
1849 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
1850 | ramblock_sync_dirty_bitmap(rs, block); | |
1851 | } | |
1852 | ram_counters.remaining = ram_bytes_remaining(); | |
56e93d26 | 1853 | } |
108cfae0 | 1854 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 1855 | |
9458a9a1 | 1856 | memory_global_after_dirty_log_sync(); |
a66cd90c | 1857 | trace_migration_bitmap_sync_end(rs->num_dirty_pages_period); |
1ffb5dfd | 1858 | |
56e93d26 JQ |
1859 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1860 | ||
1861 | /* more than 1 second = 1000 millisecons */ | |
f664da80 | 1862 | if (end_time > rs->time_last_bitmap_sync + 1000) { |
9360447d | 1863 | bytes_xfer_now = ram_counters.transferred; |
d693c6f1 | 1864 | |
9ac78b61 PL |
1865 | /* During block migration the auto-converge logic incorrectly detects |
1866 | * that ram migration makes no progress. Avoid this by disabling the | |
1867 | * throttling logic during the bulk phase of block migration. */ | |
1868 | if (migrate_auto_converge() && !blk_mig_bulk_active()) { | |
56e93d26 JQ |
1869 | /* The following detection logic can be refined later. For now: |
1870 | Check to see if the dirtied bytes is 50% more than the approx. | |
1871 | amount of bytes that just got transferred since the last time we | |
070afca2 JH |
1872 | were in this routine. If that happens twice, start or increase |
1873 | throttling */ | |
070afca2 | 1874 | |
d693c6f1 | 1875 | if ((rs->num_dirty_pages_period * TARGET_PAGE_SIZE > |
eac74159 | 1876 | (bytes_xfer_now - rs->bytes_xfer_prev) / 2) && |
b4a3c64b | 1877 | (++rs->dirty_rate_high_cnt >= 2)) { |
56e93d26 | 1878 | trace_migration_throttle(); |
8d820d6f | 1879 | rs->dirty_rate_high_cnt = 0; |
070afca2 | 1880 | mig_throttle_guest_down(); |
d693c6f1 | 1881 | } |
56e93d26 | 1882 | } |
070afca2 | 1883 | |
b734035b XG |
1884 | migration_update_rates(rs, end_time); |
1885 | ||
be8b02ed | 1886 | rs->target_page_count_prev = rs->target_page_count; |
d693c6f1 FF |
1887 | |
1888 | /* reset period counters */ | |
f664da80 | 1889 | rs->time_last_bitmap_sync = end_time; |
a66cd90c | 1890 | rs->num_dirty_pages_period = 0; |
d2a4d85a | 1891 | rs->bytes_xfer_prev = bytes_xfer_now; |
56e93d26 | 1892 | } |
4addcd4f | 1893 | if (migrate_use_events()) { |
3ab72385 | 1894 | qapi_event_send_migration_pass(ram_counters.dirty_sync_count); |
4addcd4f | 1895 | } |
56e93d26 JQ |
1896 | } |
1897 | ||
bd227060 WW |
1898 | static void migration_bitmap_sync_precopy(RAMState *rs) |
1899 | { | |
1900 | Error *local_err = NULL; | |
1901 | ||
1902 | /* | |
1903 | * The current notifier usage is just an optimization to migration, so we | |
1904 | * don't stop the normal migration process in the error case. | |
1905 | */ | |
1906 | if (precopy_notify(PRECOPY_NOTIFY_BEFORE_BITMAP_SYNC, &local_err)) { | |
1907 | error_report_err(local_err); | |
1908 | } | |
1909 | ||
1910 | migration_bitmap_sync(rs); | |
1911 | ||
1912 | if (precopy_notify(PRECOPY_NOTIFY_AFTER_BITMAP_SYNC, &local_err)) { | |
1913 | error_report_err(local_err); | |
1914 | } | |
1915 | } | |
1916 | ||
6c97ec5f XG |
1917 | /** |
1918 | * save_zero_page_to_file: send the zero page to the file | |
1919 | * | |
1920 | * Returns the size of data written to the file, 0 means the page is not | |
1921 | * a zero page | |
1922 | * | |
1923 | * @rs: current RAM state | |
1924 | * @file: the file where the data is saved | |
1925 | * @block: block that contains the page we want to send | |
1926 | * @offset: offset inside the block for the page | |
1927 | */ | |
1928 | static int save_zero_page_to_file(RAMState *rs, QEMUFile *file, | |
1929 | RAMBlock *block, ram_addr_t offset) | |
1930 | { | |
1931 | uint8_t *p = block->host + offset; | |
1932 | int len = 0; | |
1933 | ||
1934 | if (is_zero_range(p, TARGET_PAGE_SIZE)) { | |
1935 | len += save_page_header(rs, file, block, offset | RAM_SAVE_FLAG_ZERO); | |
1936 | qemu_put_byte(file, 0); | |
1937 | len += 1; | |
1938 | } | |
1939 | return len; | |
1940 | } | |
1941 | ||
56e93d26 | 1942 | /** |
3d0684b2 | 1943 | * save_zero_page: send the zero page to the stream |
56e93d26 | 1944 | * |
3d0684b2 | 1945 | * Returns the number of pages written. |
56e93d26 | 1946 | * |
f7ccd61b | 1947 | * @rs: current RAM state |
56e93d26 JQ |
1948 | * @block: block that contains the page we want to send |
1949 | * @offset: offset inside the block for the page | |
56e93d26 | 1950 | */ |
7faccdc3 | 1951 | static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset) |
56e93d26 | 1952 | { |
6c97ec5f | 1953 | int len = save_zero_page_to_file(rs, rs->f, block, offset); |
56e93d26 | 1954 | |
6c97ec5f | 1955 | if (len) { |
9360447d | 1956 | ram_counters.duplicate++; |
6c97ec5f XG |
1957 | ram_counters.transferred += len; |
1958 | return 1; | |
56e93d26 | 1959 | } |
6c97ec5f | 1960 | return -1; |
56e93d26 JQ |
1961 | } |
1962 | ||
5727309d | 1963 | static void ram_release_pages(const char *rbname, uint64_t offset, int pages) |
53f09a10 | 1964 | { |
5727309d | 1965 | if (!migrate_release_ram() || !migration_in_postcopy()) { |
53f09a10 PB |
1966 | return; |
1967 | } | |
1968 | ||
aaa2064c | 1969 | ram_discard_range(rbname, offset, pages << TARGET_PAGE_BITS); |
53f09a10 PB |
1970 | } |
1971 | ||
059ff0fb XG |
1972 | /* |
1973 | * @pages: the number of pages written by the control path, | |
1974 | * < 0 - error | |
1975 | * > 0 - number of pages written | |
1976 | * | |
1977 | * Return true if the pages has been saved, otherwise false is returned. | |
1978 | */ | |
1979 | static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, | |
1980 | int *pages) | |
1981 | { | |
1982 | uint64_t bytes_xmit = 0; | |
1983 | int ret; | |
1984 | ||
1985 | *pages = -1; | |
1986 | ret = ram_control_save_page(rs->f, block->offset, offset, TARGET_PAGE_SIZE, | |
1987 | &bytes_xmit); | |
1988 | if (ret == RAM_SAVE_CONTROL_NOT_SUPP) { | |
1989 | return false; | |
1990 | } | |
1991 | ||
1992 | if (bytes_xmit) { | |
1993 | ram_counters.transferred += bytes_xmit; | |
1994 | *pages = 1; | |
1995 | } | |
1996 | ||
1997 | if (ret == RAM_SAVE_CONTROL_DELAYED) { | |
1998 | return true; | |
1999 | } | |
2000 | ||
2001 | if (bytes_xmit > 0) { | |
2002 | ram_counters.normal++; | |
2003 | } else if (bytes_xmit == 0) { | |
2004 | ram_counters.duplicate++; | |
2005 | } | |
2006 | ||
2007 | return true; | |
2008 | } | |
2009 | ||
65dacaa0 XG |
2010 | /* |
2011 | * directly send the page to the stream | |
2012 | * | |
2013 | * Returns the number of pages written. | |
2014 | * | |
2015 | * @rs: current RAM state | |
2016 | * @block: block that contains the page we want to send | |
2017 | * @offset: offset inside the block for the page | |
2018 | * @buf: the page to be sent | |
2019 | * @async: send to page asyncly | |
2020 | */ | |
2021 | static int save_normal_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, | |
2022 | uint8_t *buf, bool async) | |
2023 | { | |
2024 | ram_counters.transferred += save_page_header(rs, rs->f, block, | |
2025 | offset | RAM_SAVE_FLAG_PAGE); | |
2026 | if (async) { | |
2027 | qemu_put_buffer_async(rs->f, buf, TARGET_PAGE_SIZE, | |
2028 | migrate_release_ram() & | |
2029 | migration_in_postcopy()); | |
2030 | } else { | |
2031 | qemu_put_buffer(rs->f, buf, TARGET_PAGE_SIZE); | |
2032 | } | |
2033 | ram_counters.transferred += TARGET_PAGE_SIZE; | |
2034 | ram_counters.normal++; | |
2035 | return 1; | |
2036 | } | |
2037 | ||
56e93d26 | 2038 | /** |
3d0684b2 | 2039 | * ram_save_page: send the given page to the stream |
56e93d26 | 2040 | * |
3d0684b2 | 2041 | * Returns the number of pages written. |
3fd3c4b3 DDAG |
2042 | * < 0 - error |
2043 | * >=0 - Number of pages written - this might legally be 0 | |
2044 | * if xbzrle noticed the page was the same. | |
56e93d26 | 2045 | * |
6f37bb8b | 2046 | * @rs: current RAM state |
56e93d26 JQ |
2047 | * @block: block that contains the page we want to send |
2048 | * @offset: offset inside the block for the page | |
2049 | * @last_stage: if we are at the completion stage | |
56e93d26 | 2050 | */ |
a0a8aa14 | 2051 | static int ram_save_page(RAMState *rs, PageSearchStatus *pss, bool last_stage) |
56e93d26 JQ |
2052 | { |
2053 | int pages = -1; | |
56e93d26 | 2054 | uint8_t *p; |
56e93d26 | 2055 | bool send_async = true; |
a08f6890 | 2056 | RAMBlock *block = pss->block; |
a935e30f | 2057 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
059ff0fb | 2058 | ram_addr_t current_addr = block->offset + offset; |
56e93d26 | 2059 | |
2f68e399 | 2060 | p = block->host + offset; |
1db9d8e5 | 2061 | trace_ram_save_page(block->idstr, (uint64_t)offset, p); |
56e93d26 | 2062 | |
56e93d26 | 2063 | XBZRLE_cache_lock(); |
d7400a34 XG |
2064 | if (!rs->ram_bulk_stage && !migration_in_postcopy() && |
2065 | migrate_use_xbzrle()) { | |
059ff0fb XG |
2066 | pages = save_xbzrle_page(rs, &p, current_addr, block, |
2067 | offset, last_stage); | |
2068 | if (!last_stage) { | |
2069 | /* Can't send this cached data async, since the cache page | |
2070 | * might get updated before it gets to the wire | |
56e93d26 | 2071 | */ |
059ff0fb | 2072 | send_async = false; |
56e93d26 JQ |
2073 | } |
2074 | } | |
2075 | ||
2076 | /* XBZRLE overflow or normal page */ | |
2077 | if (pages == -1) { | |
65dacaa0 | 2078 | pages = save_normal_page(rs, block, offset, p, send_async); |
56e93d26 JQ |
2079 | } |
2080 | ||
2081 | XBZRLE_cache_unlock(); | |
2082 | ||
2083 | return pages; | |
2084 | } | |
2085 | ||
b9ee2f7d JQ |
2086 | static int ram_save_multifd_page(RAMState *rs, RAMBlock *block, |
2087 | ram_addr_t offset) | |
2088 | { | |
1b81c974 | 2089 | if (multifd_queue_page(rs, block, offset) < 0) { |
713f762a IR |
2090 | return -1; |
2091 | } | |
b9ee2f7d JQ |
2092 | ram_counters.normal++; |
2093 | ||
2094 | return 1; | |
2095 | } | |
2096 | ||
5e5fdcff | 2097 | static bool do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
6ef3771c | 2098 | ram_addr_t offset, uint8_t *source_buf) |
56e93d26 | 2099 | { |
53518d94 | 2100 | RAMState *rs = ram_state; |
a7a9a88f | 2101 | uint8_t *p = block->host + (offset & TARGET_PAGE_MASK); |
5e5fdcff | 2102 | bool zero_page = false; |
6ef3771c | 2103 | int ret; |
56e93d26 | 2104 | |
5e5fdcff XG |
2105 | if (save_zero_page_to_file(rs, f, block, offset)) { |
2106 | zero_page = true; | |
2107 | goto exit; | |
2108 | } | |
2109 | ||
6ef3771c | 2110 | save_page_header(rs, f, block, offset | RAM_SAVE_FLAG_COMPRESS_PAGE); |
34ab9e97 XG |
2111 | |
2112 | /* | |
2113 | * copy it to a internal buffer to avoid it being modified by VM | |
2114 | * so that we can catch up the error during compression and | |
2115 | * decompression | |
2116 | */ | |
2117 | memcpy(source_buf, p, TARGET_PAGE_SIZE); | |
6ef3771c XG |
2118 | ret = qemu_put_compression_data(f, stream, source_buf, TARGET_PAGE_SIZE); |
2119 | if (ret < 0) { | |
2120 | qemu_file_set_error(migrate_get_current()->to_dst_file, ret); | |
b3be2896 | 2121 | error_report("compressed data failed!"); |
5e5fdcff | 2122 | return false; |
b3be2896 | 2123 | } |
56e93d26 | 2124 | |
5e5fdcff | 2125 | exit: |
6ef3771c | 2126 | ram_release_pages(block->idstr, offset & TARGET_PAGE_MASK, 1); |
5e5fdcff XG |
2127 | return zero_page; |
2128 | } | |
2129 | ||
2130 | static void | |
2131 | update_compress_thread_counts(const CompressParam *param, int bytes_xmit) | |
2132 | { | |
76e03000 XG |
2133 | ram_counters.transferred += bytes_xmit; |
2134 | ||
5e5fdcff XG |
2135 | if (param->zero_page) { |
2136 | ram_counters.duplicate++; | |
76e03000 | 2137 | return; |
5e5fdcff | 2138 | } |
76e03000 XG |
2139 | |
2140 | /* 8 means a header with RAM_SAVE_FLAG_CONTINUE. */ | |
2141 | compression_counters.compressed_size += bytes_xmit - 8; | |
2142 | compression_counters.pages++; | |
56e93d26 JQ |
2143 | } |
2144 | ||
32b05495 XG |
2145 | static bool save_page_use_compression(RAMState *rs); |
2146 | ||
ce25d337 | 2147 | static void flush_compressed_data(RAMState *rs) |
56e93d26 JQ |
2148 | { |
2149 | int idx, len, thread_count; | |
2150 | ||
32b05495 | 2151 | if (!save_page_use_compression(rs)) { |
56e93d26 JQ |
2152 | return; |
2153 | } | |
2154 | thread_count = migrate_compress_threads(); | |
a7a9a88f | 2155 | |
0d9f9a5c | 2156 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 | 2157 | for (idx = 0; idx < thread_count; idx++) { |
a7a9a88f | 2158 | while (!comp_param[idx].done) { |
0d9f9a5c | 2159 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 | 2160 | } |
a7a9a88f | 2161 | } |
0d9f9a5c | 2162 | qemu_mutex_unlock(&comp_done_lock); |
a7a9a88f LL |
2163 | |
2164 | for (idx = 0; idx < thread_count; idx++) { | |
2165 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 2166 | if (!comp_param[idx].quit) { |
ce25d337 | 2167 | len = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
5e5fdcff XG |
2168 | /* |
2169 | * it's safe to fetch zero_page without holding comp_done_lock | |
2170 | * as there is no further request submitted to the thread, | |
2171 | * i.e, the thread should be waiting for a request at this point. | |
2172 | */ | |
2173 | update_compress_thread_counts(&comp_param[idx], len); | |
56e93d26 | 2174 | } |
a7a9a88f | 2175 | qemu_mutex_unlock(&comp_param[idx].mutex); |
56e93d26 JQ |
2176 | } |
2177 | } | |
2178 | ||
2179 | static inline void set_compress_params(CompressParam *param, RAMBlock *block, | |
2180 | ram_addr_t offset) | |
2181 | { | |
2182 | param->block = block; | |
2183 | param->offset = offset; | |
2184 | } | |
2185 | ||
ce25d337 JQ |
2186 | static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block, |
2187 | ram_addr_t offset) | |
56e93d26 JQ |
2188 | { |
2189 | int idx, thread_count, bytes_xmit = -1, pages = -1; | |
1d58872a | 2190 | bool wait = migrate_compress_wait_thread(); |
56e93d26 JQ |
2191 | |
2192 | thread_count = migrate_compress_threads(); | |
0d9f9a5c | 2193 | qemu_mutex_lock(&comp_done_lock); |
1d58872a XG |
2194 | retry: |
2195 | for (idx = 0; idx < thread_count; idx++) { | |
2196 | if (comp_param[idx].done) { | |
2197 | comp_param[idx].done = false; | |
2198 | bytes_xmit = qemu_put_qemu_file(rs->f, comp_param[idx].file); | |
2199 | qemu_mutex_lock(&comp_param[idx].mutex); | |
2200 | set_compress_params(&comp_param[idx], block, offset); | |
2201 | qemu_cond_signal(&comp_param[idx].cond); | |
2202 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
2203 | pages = 1; | |
5e5fdcff | 2204 | update_compress_thread_counts(&comp_param[idx], bytes_xmit); |
56e93d26 | 2205 | break; |
56e93d26 JQ |
2206 | } |
2207 | } | |
1d58872a XG |
2208 | |
2209 | /* | |
2210 | * wait for the free thread if the user specifies 'compress-wait-thread', | |
2211 | * otherwise we will post the page out in the main thread as normal page. | |
2212 | */ | |
2213 | if (pages < 0 && wait) { | |
2214 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); | |
2215 | goto retry; | |
2216 | } | |
0d9f9a5c | 2217 | qemu_mutex_unlock(&comp_done_lock); |
56e93d26 JQ |
2218 | |
2219 | return pages; | |
2220 | } | |
2221 | ||
3d0684b2 JQ |
2222 | /** |
2223 | * find_dirty_block: find the next dirty page and update any state | |
2224 | * associated with the search process. | |
b9e60928 | 2225 | * |
a5f7b1a6 | 2226 | * Returns true if a page is found |
b9e60928 | 2227 | * |
6f37bb8b | 2228 | * @rs: current RAM state |
3d0684b2 JQ |
2229 | * @pss: data about the state of the current dirty page scan |
2230 | * @again: set to false if the search has scanned the whole of RAM | |
b9e60928 | 2231 | */ |
f20e2865 | 2232 | static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again) |
b9e60928 | 2233 | { |
f20e2865 | 2234 | pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page); |
6f37bb8b | 2235 | if (pss->complete_round && pss->block == rs->last_seen_block && |
a935e30f | 2236 | pss->page >= rs->last_page) { |
b9e60928 DDAG |
2237 | /* |
2238 | * We've been once around the RAM and haven't found anything. | |
2239 | * Give up. | |
2240 | */ | |
2241 | *again = false; | |
2242 | return false; | |
2243 | } | |
a935e30f | 2244 | if ((pss->page << TARGET_PAGE_BITS) >= pss->block->used_length) { |
b9e60928 | 2245 | /* Didn't find anything in this RAM Block */ |
a935e30f | 2246 | pss->page = 0; |
b9e60928 DDAG |
2247 | pss->block = QLIST_NEXT_RCU(pss->block, next); |
2248 | if (!pss->block) { | |
48df9d80 XG |
2249 | /* |
2250 | * If memory migration starts over, we will meet a dirtied page | |
2251 | * which may still exists in compression threads's ring, so we | |
2252 | * should flush the compressed data to make sure the new page | |
2253 | * is not overwritten by the old one in the destination. | |
2254 | * | |
2255 | * Also If xbzrle is on, stop using the data compression at this | |
2256 | * point. In theory, xbzrle can do better than compression. | |
2257 | */ | |
2258 | flush_compressed_data(rs); | |
2259 | ||
b9e60928 DDAG |
2260 | /* Hit the end of the list */ |
2261 | pss->block = QLIST_FIRST_RCU(&ram_list.blocks); | |
2262 | /* Flag that we've looped */ | |
2263 | pss->complete_round = true; | |
6f37bb8b | 2264 | rs->ram_bulk_stage = false; |
b9e60928 DDAG |
2265 | } |
2266 | /* Didn't find anything this time, but try again on the new block */ | |
2267 | *again = true; | |
2268 | return false; | |
2269 | } else { | |
2270 | /* Can go around again, but... */ | |
2271 | *again = true; | |
2272 | /* We've found something so probably don't need to */ | |
2273 | return true; | |
2274 | } | |
2275 | } | |
2276 | ||
3d0684b2 JQ |
2277 | /** |
2278 | * unqueue_page: gets a page of the queue | |
2279 | * | |
a82d593b | 2280 | * Helper for 'get_queued_page' - gets a page off the queue |
a82d593b | 2281 | * |
3d0684b2 JQ |
2282 | * Returns the block of the page (or NULL if none available) |
2283 | * | |
ec481c6c | 2284 | * @rs: current RAM state |
3d0684b2 | 2285 | * @offset: used to return the offset within the RAMBlock |
a82d593b | 2286 | */ |
f20e2865 | 2287 | static RAMBlock *unqueue_page(RAMState *rs, ram_addr_t *offset) |
a82d593b DDAG |
2288 | { |
2289 | RAMBlock *block = NULL; | |
2290 | ||
ae526e32 XG |
2291 | if (QSIMPLEQ_EMPTY_ATOMIC(&rs->src_page_requests)) { |
2292 | return NULL; | |
2293 | } | |
2294 | ||
ec481c6c JQ |
2295 | qemu_mutex_lock(&rs->src_page_req_mutex); |
2296 | if (!QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
2297 | struct RAMSrcPageRequest *entry = | |
2298 | QSIMPLEQ_FIRST(&rs->src_page_requests); | |
a82d593b DDAG |
2299 | block = entry->rb; |
2300 | *offset = entry->offset; | |
a82d593b DDAG |
2301 | |
2302 | if (entry->len > TARGET_PAGE_SIZE) { | |
2303 | entry->len -= TARGET_PAGE_SIZE; | |
2304 | entry->offset += TARGET_PAGE_SIZE; | |
2305 | } else { | |
2306 | memory_region_unref(block->mr); | |
ec481c6c | 2307 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
a82d593b | 2308 | g_free(entry); |
e03a34f8 | 2309 | migration_consume_urgent_request(); |
a82d593b DDAG |
2310 | } |
2311 | } | |
ec481c6c | 2312 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
a82d593b DDAG |
2313 | |
2314 | return block; | |
2315 | } | |
2316 | ||
3d0684b2 | 2317 | /** |
ff1543af | 2318 | * get_queued_page: unqueue a page from the postcopy requests |
3d0684b2 JQ |
2319 | * |
2320 | * Skips pages that are already sent (!dirty) | |
a82d593b | 2321 | * |
a5f7b1a6 | 2322 | * Returns true if a queued page is found |
a82d593b | 2323 | * |
6f37bb8b | 2324 | * @rs: current RAM state |
3d0684b2 | 2325 | * @pss: data about the state of the current dirty page scan |
a82d593b | 2326 | */ |
f20e2865 | 2327 | static bool get_queued_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b DDAG |
2328 | { |
2329 | RAMBlock *block; | |
2330 | ram_addr_t offset; | |
2331 | bool dirty; | |
2332 | ||
2333 | do { | |
f20e2865 | 2334 | block = unqueue_page(rs, &offset); |
a82d593b DDAG |
2335 | /* |
2336 | * We're sending this page, and since it's postcopy nothing else | |
2337 | * will dirty it, and we must make sure it doesn't get sent again | |
2338 | * even if this queue request was received after the background | |
2339 | * search already sent it. | |
2340 | */ | |
2341 | if (block) { | |
f20e2865 JQ |
2342 | unsigned long page; |
2343 | ||
6b6712ef JQ |
2344 | page = offset >> TARGET_PAGE_BITS; |
2345 | dirty = test_bit(page, block->bmap); | |
a82d593b | 2346 | if (!dirty) { |
06b10688 | 2347 | trace_get_queued_page_not_dirty(block->idstr, (uint64_t)offset, |
64737606 | 2348 | page); |
a82d593b | 2349 | } else { |
f20e2865 | 2350 | trace_get_queued_page(block->idstr, (uint64_t)offset, page); |
a82d593b DDAG |
2351 | } |
2352 | } | |
2353 | ||
2354 | } while (block && !dirty); | |
2355 | ||
2356 | if (block) { | |
2357 | /* | |
2358 | * As soon as we start servicing pages out of order, then we have | |
2359 | * to kill the bulk stage, since the bulk stage assumes | |
2360 | * in (migration_bitmap_find_and_reset_dirty) that every page is | |
2361 | * dirty, that's no longer true. | |
2362 | */ | |
6f37bb8b | 2363 | rs->ram_bulk_stage = false; |
a82d593b DDAG |
2364 | |
2365 | /* | |
2366 | * We want the background search to continue from the queued page | |
2367 | * since the guest is likely to want other pages near to the page | |
2368 | * it just requested. | |
2369 | */ | |
2370 | pss->block = block; | |
a935e30f | 2371 | pss->page = offset >> TARGET_PAGE_BITS; |
422314e7 WY |
2372 | |
2373 | /* | |
2374 | * This unqueued page would break the "one round" check, even is | |
2375 | * really rare. | |
2376 | */ | |
2377 | pss->complete_round = false; | |
a82d593b DDAG |
2378 | } |
2379 | ||
2380 | return !!block; | |
2381 | } | |
2382 | ||
6c595cde | 2383 | /** |
5e58f968 JQ |
2384 | * migration_page_queue_free: drop any remaining pages in the ram |
2385 | * request queue | |
6c595cde | 2386 | * |
3d0684b2 JQ |
2387 | * It should be empty at the end anyway, but in error cases there may |
2388 | * be some left. in case that there is any page left, we drop it. | |
2389 | * | |
6c595cde | 2390 | */ |
83c13382 | 2391 | static void migration_page_queue_free(RAMState *rs) |
6c595cde | 2392 | { |
ec481c6c | 2393 | struct RAMSrcPageRequest *mspr, *next_mspr; |
6c595cde DDAG |
2394 | /* This queue generally should be empty - but in the case of a failed |
2395 | * migration might have some droppings in. | |
2396 | */ | |
89ac5a1d | 2397 | RCU_READ_LOCK_GUARD(); |
ec481c6c | 2398 | QSIMPLEQ_FOREACH_SAFE(mspr, &rs->src_page_requests, next_req, next_mspr) { |
6c595cde | 2399 | memory_region_unref(mspr->rb->mr); |
ec481c6c | 2400 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
6c595cde DDAG |
2401 | g_free(mspr); |
2402 | } | |
6c595cde DDAG |
2403 | } |
2404 | ||
2405 | /** | |
3d0684b2 JQ |
2406 | * ram_save_queue_pages: queue the page for transmission |
2407 | * | |
2408 | * A request from postcopy destination for example. | |
2409 | * | |
2410 | * Returns zero on success or negative on error | |
2411 | * | |
3d0684b2 JQ |
2412 | * @rbname: Name of the RAMBLock of the request. NULL means the |
2413 | * same that last one. | |
2414 | * @start: starting address from the start of the RAMBlock | |
2415 | * @len: length (in bytes) to send | |
6c595cde | 2416 | */ |
96506894 | 2417 | int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len) |
6c595cde DDAG |
2418 | { |
2419 | RAMBlock *ramblock; | |
53518d94 | 2420 | RAMState *rs = ram_state; |
6c595cde | 2421 | |
9360447d | 2422 | ram_counters.postcopy_requests++; |
89ac5a1d DDAG |
2423 | RCU_READ_LOCK_GUARD(); |
2424 | ||
6c595cde DDAG |
2425 | if (!rbname) { |
2426 | /* Reuse last RAMBlock */ | |
68a098f3 | 2427 | ramblock = rs->last_req_rb; |
6c595cde DDAG |
2428 | |
2429 | if (!ramblock) { | |
2430 | /* | |
2431 | * Shouldn't happen, we can't reuse the last RAMBlock if | |
2432 | * it's the 1st request. | |
2433 | */ | |
2434 | error_report("ram_save_queue_pages no previous block"); | |
2435 | goto err; | |
2436 | } | |
2437 | } else { | |
2438 | ramblock = qemu_ram_block_by_name(rbname); | |
2439 | ||
2440 | if (!ramblock) { | |
2441 | /* We shouldn't be asked for a non-existent RAMBlock */ | |
2442 | error_report("ram_save_queue_pages no block '%s'", rbname); | |
2443 | goto err; | |
2444 | } | |
68a098f3 | 2445 | rs->last_req_rb = ramblock; |
6c595cde DDAG |
2446 | } |
2447 | trace_ram_save_queue_pages(ramblock->idstr, start, len); | |
2448 | if (start+len > ramblock->used_length) { | |
9458ad6b JQ |
2449 | error_report("%s request overrun start=" RAM_ADDR_FMT " len=" |
2450 | RAM_ADDR_FMT " blocklen=" RAM_ADDR_FMT, | |
6c595cde DDAG |
2451 | __func__, start, len, ramblock->used_length); |
2452 | goto err; | |
2453 | } | |
2454 | ||
ec481c6c JQ |
2455 | struct RAMSrcPageRequest *new_entry = |
2456 | g_malloc0(sizeof(struct RAMSrcPageRequest)); | |
6c595cde DDAG |
2457 | new_entry->rb = ramblock; |
2458 | new_entry->offset = start; | |
2459 | new_entry->len = len; | |
2460 | ||
2461 | memory_region_ref(ramblock->mr); | |
ec481c6c JQ |
2462 | qemu_mutex_lock(&rs->src_page_req_mutex); |
2463 | QSIMPLEQ_INSERT_TAIL(&rs->src_page_requests, new_entry, next_req); | |
e03a34f8 | 2464 | migration_make_urgent_request(); |
ec481c6c | 2465 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
6c595cde DDAG |
2466 | |
2467 | return 0; | |
2468 | ||
2469 | err: | |
6c595cde DDAG |
2470 | return -1; |
2471 | } | |
2472 | ||
d7400a34 XG |
2473 | static bool save_page_use_compression(RAMState *rs) |
2474 | { | |
2475 | if (!migrate_use_compression()) { | |
2476 | return false; | |
2477 | } | |
2478 | ||
2479 | /* | |
2480 | * If xbzrle is on, stop using the data compression after first | |
2481 | * round of migration even if compression is enabled. In theory, | |
2482 | * xbzrle can do better than compression. | |
2483 | */ | |
2484 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { | |
2485 | return true; | |
2486 | } | |
2487 | ||
2488 | return false; | |
2489 | } | |
2490 | ||
5e5fdcff XG |
2491 | /* |
2492 | * try to compress the page before posting it out, return true if the page | |
2493 | * has been properly handled by compression, otherwise needs other | |
2494 | * paths to handle it | |
2495 | */ | |
2496 | static bool save_compress_page(RAMState *rs, RAMBlock *block, ram_addr_t offset) | |
2497 | { | |
2498 | if (!save_page_use_compression(rs)) { | |
2499 | return false; | |
2500 | } | |
2501 | ||
2502 | /* | |
2503 | * When starting the process of a new block, the first page of | |
2504 | * the block should be sent out before other pages in the same | |
2505 | * block, and all the pages in last block should have been sent | |
2506 | * out, keeping this order is important, because the 'cont' flag | |
2507 | * is used to avoid resending the block name. | |
2508 | * | |
2509 | * We post the fist page as normal page as compression will take | |
2510 | * much CPU resource. | |
2511 | */ | |
2512 | if (block != rs->last_sent_block) { | |
2513 | flush_compressed_data(rs); | |
2514 | return false; | |
2515 | } | |
2516 | ||
2517 | if (compress_page_with_multi_thread(rs, block, offset) > 0) { | |
2518 | return true; | |
2519 | } | |
2520 | ||
76e03000 | 2521 | compression_counters.busy++; |
5e5fdcff XG |
2522 | return false; |
2523 | } | |
2524 | ||
a82d593b | 2525 | /** |
3d0684b2 | 2526 | * ram_save_target_page: save one target page |
a82d593b | 2527 | * |
3d0684b2 | 2528 | * Returns the number of pages written |
a82d593b | 2529 | * |
6f37bb8b | 2530 | * @rs: current RAM state |
3d0684b2 | 2531 | * @pss: data about the page we want to send |
a82d593b | 2532 | * @last_stage: if we are at the completion stage |
a82d593b | 2533 | */ |
a0a8aa14 | 2534 | static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 2535 | bool last_stage) |
a82d593b | 2536 | { |
a8ec91f9 XG |
2537 | RAMBlock *block = pss->block; |
2538 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; | |
2539 | int res; | |
2540 | ||
2541 | if (control_save_page(rs, block, offset, &res)) { | |
2542 | return res; | |
2543 | } | |
2544 | ||
5e5fdcff XG |
2545 | if (save_compress_page(rs, block, offset)) { |
2546 | return 1; | |
d7400a34 XG |
2547 | } |
2548 | ||
2549 | res = save_zero_page(rs, block, offset); | |
2550 | if (res > 0) { | |
2551 | /* Must let xbzrle know, otherwise a previous (now 0'd) cached | |
2552 | * page would be stale | |
2553 | */ | |
2554 | if (!save_page_use_compression(rs)) { | |
2555 | XBZRLE_cache_lock(); | |
2556 | xbzrle_cache_zero_page(rs, block->offset + offset); | |
2557 | XBZRLE_cache_unlock(); | |
2558 | } | |
2559 | ram_release_pages(block->idstr, offset, res); | |
2560 | return res; | |
2561 | } | |
2562 | ||
da3f56cb | 2563 | /* |
5e5fdcff XG |
2564 | * do not use multifd for compression as the first page in the new |
2565 | * block should be posted out before sending the compressed page | |
da3f56cb | 2566 | */ |
5e5fdcff | 2567 | if (!save_page_use_compression(rs) && migrate_use_multifd()) { |
b9ee2f7d | 2568 | return ram_save_multifd_page(rs, block, offset); |
a82d593b DDAG |
2569 | } |
2570 | ||
1faa5665 | 2571 | return ram_save_page(rs, pss, last_stage); |
a82d593b DDAG |
2572 | } |
2573 | ||
2574 | /** | |
3d0684b2 | 2575 | * ram_save_host_page: save a whole host page |
a82d593b | 2576 | * |
3d0684b2 JQ |
2577 | * Starting at *offset send pages up to the end of the current host |
2578 | * page. It's valid for the initial offset to point into the middle of | |
2579 | * a host page in which case the remainder of the hostpage is sent. | |
2580 | * Only dirty target pages are sent. Note that the host page size may | |
2581 | * be a huge page for this block. | |
1eb3fc0a DDAG |
2582 | * The saving stops at the boundary of the used_length of the block |
2583 | * if the RAMBlock isn't a multiple of the host page size. | |
a82d593b | 2584 | * |
3d0684b2 JQ |
2585 | * Returns the number of pages written or negative on error |
2586 | * | |
6f37bb8b | 2587 | * @rs: current RAM state |
3d0684b2 | 2588 | * @ms: current migration state |
3d0684b2 | 2589 | * @pss: data about the page we want to send |
a82d593b | 2590 | * @last_stage: if we are at the completion stage |
a82d593b | 2591 | */ |
a0a8aa14 | 2592 | static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 2593 | bool last_stage) |
a82d593b DDAG |
2594 | { |
2595 | int tmppages, pages = 0; | |
a935e30f JQ |
2596 | size_t pagesize_bits = |
2597 | qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
4c011c37 | 2598 | |
fbd162e6 | 2599 | if (ramblock_is_ignored(pss->block)) { |
b895de50 CLG |
2600 | error_report("block %s should not be migrated !", pss->block->idstr); |
2601 | return 0; | |
2602 | } | |
2603 | ||
a82d593b | 2604 | do { |
1faa5665 XG |
2605 | /* Check the pages is dirty and if it is send it */ |
2606 | if (!migration_bitmap_clear_dirty(rs, pss->block, pss->page)) { | |
2607 | pss->page++; | |
2608 | continue; | |
2609 | } | |
2610 | ||
f20e2865 | 2611 | tmppages = ram_save_target_page(rs, pss, last_stage); |
a82d593b DDAG |
2612 | if (tmppages < 0) { |
2613 | return tmppages; | |
2614 | } | |
2615 | ||
2616 | pages += tmppages; | |
a935e30f | 2617 | pss->page++; |
1eb3fc0a DDAG |
2618 | } while ((pss->page & (pagesize_bits - 1)) && |
2619 | offset_in_ramblock(pss->block, pss->page << TARGET_PAGE_BITS)); | |
a82d593b DDAG |
2620 | |
2621 | /* The offset we leave with is the last one we looked at */ | |
a935e30f | 2622 | pss->page--; |
a82d593b DDAG |
2623 | return pages; |
2624 | } | |
6c595cde | 2625 | |
56e93d26 | 2626 | /** |
3d0684b2 | 2627 | * ram_find_and_save_block: finds a dirty page and sends it to f |
56e93d26 JQ |
2628 | * |
2629 | * Called within an RCU critical section. | |
2630 | * | |
e8f3735f XG |
2631 | * Returns the number of pages written where zero means no dirty pages, |
2632 | * or negative on error | |
56e93d26 | 2633 | * |
6f37bb8b | 2634 | * @rs: current RAM state |
56e93d26 | 2635 | * @last_stage: if we are at the completion stage |
a82d593b DDAG |
2636 | * |
2637 | * On systems where host-page-size > target-page-size it will send all the | |
2638 | * pages in a host page that are dirty. | |
56e93d26 JQ |
2639 | */ |
2640 | ||
ce25d337 | 2641 | static int ram_find_and_save_block(RAMState *rs, bool last_stage) |
56e93d26 | 2642 | { |
b8fb8cb7 | 2643 | PageSearchStatus pss; |
56e93d26 | 2644 | int pages = 0; |
b9e60928 | 2645 | bool again, found; |
56e93d26 | 2646 | |
0827b9e9 AA |
2647 | /* No dirty page as there is zero RAM */ |
2648 | if (!ram_bytes_total()) { | |
2649 | return pages; | |
2650 | } | |
2651 | ||
6f37bb8b | 2652 | pss.block = rs->last_seen_block; |
a935e30f | 2653 | pss.page = rs->last_page; |
b8fb8cb7 DDAG |
2654 | pss.complete_round = false; |
2655 | ||
2656 | if (!pss.block) { | |
2657 | pss.block = QLIST_FIRST_RCU(&ram_list.blocks); | |
2658 | } | |
56e93d26 | 2659 | |
b9e60928 | 2660 | do { |
a82d593b | 2661 | again = true; |
f20e2865 | 2662 | found = get_queued_page(rs, &pss); |
b9e60928 | 2663 | |
a82d593b DDAG |
2664 | if (!found) { |
2665 | /* priority queue empty, so just search for something dirty */ | |
f20e2865 | 2666 | found = find_dirty_block(rs, &pss, &again); |
a82d593b | 2667 | } |
f3f491fc | 2668 | |
a82d593b | 2669 | if (found) { |
f20e2865 | 2670 | pages = ram_save_host_page(rs, &pss, last_stage); |
56e93d26 | 2671 | } |
b9e60928 | 2672 | } while (!pages && again); |
56e93d26 | 2673 | |
6f37bb8b | 2674 | rs->last_seen_block = pss.block; |
a935e30f | 2675 | rs->last_page = pss.page; |
56e93d26 JQ |
2676 | |
2677 | return pages; | |
2678 | } | |
2679 | ||
2680 | void acct_update_position(QEMUFile *f, size_t size, bool zero) | |
2681 | { | |
2682 | uint64_t pages = size / TARGET_PAGE_SIZE; | |
f7ccd61b | 2683 | |
56e93d26 | 2684 | if (zero) { |
9360447d | 2685 | ram_counters.duplicate += pages; |
56e93d26 | 2686 | } else { |
9360447d JQ |
2687 | ram_counters.normal += pages; |
2688 | ram_counters.transferred += size; | |
56e93d26 JQ |
2689 | qemu_update_position(f, size); |
2690 | } | |
2691 | } | |
2692 | ||
fbd162e6 | 2693 | static uint64_t ram_bytes_total_common(bool count_ignored) |
56e93d26 JQ |
2694 | { |
2695 | RAMBlock *block; | |
2696 | uint64_t total = 0; | |
2697 | ||
89ac5a1d DDAG |
2698 | RCU_READ_LOCK_GUARD(); |
2699 | ||
fbd162e6 YK |
2700 | if (count_ignored) { |
2701 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
2702 | total += block->used_length; | |
2703 | } | |
2704 | } else { | |
2705 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
2706 | total += block->used_length; | |
2707 | } | |
99e15582 | 2708 | } |
56e93d26 JQ |
2709 | return total; |
2710 | } | |
2711 | ||
fbd162e6 YK |
2712 | uint64_t ram_bytes_total(void) |
2713 | { | |
2714 | return ram_bytes_total_common(false); | |
2715 | } | |
2716 | ||
f265e0e4 | 2717 | static void xbzrle_load_setup(void) |
56e93d26 | 2718 | { |
f265e0e4 | 2719 | XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE); |
56e93d26 JQ |
2720 | } |
2721 | ||
f265e0e4 JQ |
2722 | static void xbzrle_load_cleanup(void) |
2723 | { | |
2724 | g_free(XBZRLE.decoded_buf); | |
2725 | XBZRLE.decoded_buf = NULL; | |
2726 | } | |
2727 | ||
7d7c96be PX |
2728 | static void ram_state_cleanup(RAMState **rsp) |
2729 | { | |
b9ccaf6d DDAG |
2730 | if (*rsp) { |
2731 | migration_page_queue_free(*rsp); | |
2732 | qemu_mutex_destroy(&(*rsp)->bitmap_mutex); | |
2733 | qemu_mutex_destroy(&(*rsp)->src_page_req_mutex); | |
2734 | g_free(*rsp); | |
2735 | *rsp = NULL; | |
2736 | } | |
7d7c96be PX |
2737 | } |
2738 | ||
84593a08 PX |
2739 | static void xbzrle_cleanup(void) |
2740 | { | |
2741 | XBZRLE_cache_lock(); | |
2742 | if (XBZRLE.cache) { | |
2743 | cache_fini(XBZRLE.cache); | |
2744 | g_free(XBZRLE.encoded_buf); | |
2745 | g_free(XBZRLE.current_buf); | |
2746 | g_free(XBZRLE.zero_target_page); | |
2747 | XBZRLE.cache = NULL; | |
2748 | XBZRLE.encoded_buf = NULL; | |
2749 | XBZRLE.current_buf = NULL; | |
2750 | XBZRLE.zero_target_page = NULL; | |
2751 | } | |
2752 | XBZRLE_cache_unlock(); | |
2753 | } | |
2754 | ||
f265e0e4 | 2755 | static void ram_save_cleanup(void *opaque) |
56e93d26 | 2756 | { |
53518d94 | 2757 | RAMState **rsp = opaque; |
6b6712ef | 2758 | RAMBlock *block; |
eb859c53 | 2759 | |
2ff64038 | 2760 | /* caller have hold iothread lock or is in a bh, so there is |
4633456c | 2761 | * no writing race against the migration bitmap |
2ff64038 | 2762 | */ |
6b6712ef JQ |
2763 | memory_global_dirty_log_stop(); |
2764 | ||
fbd162e6 | 2765 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
002cad6b PX |
2766 | g_free(block->clear_bmap); |
2767 | block->clear_bmap = NULL; | |
6b6712ef JQ |
2768 | g_free(block->bmap); |
2769 | block->bmap = NULL; | |
56e93d26 JQ |
2770 | } |
2771 | ||
84593a08 | 2772 | xbzrle_cleanup(); |
f0afa331 | 2773 | compress_threads_save_cleanup(); |
7d7c96be | 2774 | ram_state_cleanup(rsp); |
56e93d26 JQ |
2775 | } |
2776 | ||
6f37bb8b | 2777 | static void ram_state_reset(RAMState *rs) |
56e93d26 | 2778 | { |
6f37bb8b JQ |
2779 | rs->last_seen_block = NULL; |
2780 | rs->last_sent_block = NULL; | |
269ace29 | 2781 | rs->last_page = 0; |
6f37bb8b JQ |
2782 | rs->last_version = ram_list.version; |
2783 | rs->ram_bulk_stage = true; | |
6eeb63f7 | 2784 | rs->fpo_enabled = false; |
56e93d26 JQ |
2785 | } |
2786 | ||
2787 | #define MAX_WAIT 50 /* ms, half buffered_file limit */ | |
2788 | ||
4f2e4252 DDAG |
2789 | /* |
2790 | * 'expected' is the value you expect the bitmap mostly to be full | |
2791 | * of; it won't bother printing lines that are all this value. | |
2792 | * If 'todump' is null the migration bitmap is dumped. | |
2793 | */ | |
6b6712ef JQ |
2794 | void ram_debug_dump_bitmap(unsigned long *todump, bool expected, |
2795 | unsigned long pages) | |
4f2e4252 | 2796 | { |
4f2e4252 DDAG |
2797 | int64_t cur; |
2798 | int64_t linelen = 128; | |
2799 | char linebuf[129]; | |
2800 | ||
6b6712ef | 2801 | for (cur = 0; cur < pages; cur += linelen) { |
4f2e4252 DDAG |
2802 | int64_t curb; |
2803 | bool found = false; | |
2804 | /* | |
2805 | * Last line; catch the case where the line length | |
2806 | * is longer than remaining ram | |
2807 | */ | |
6b6712ef JQ |
2808 | if (cur + linelen > pages) { |
2809 | linelen = pages - cur; | |
4f2e4252 DDAG |
2810 | } |
2811 | for (curb = 0; curb < linelen; curb++) { | |
2812 | bool thisbit = test_bit(cur + curb, todump); | |
2813 | linebuf[curb] = thisbit ? '1' : '.'; | |
2814 | found = found || (thisbit != expected); | |
2815 | } | |
2816 | if (found) { | |
2817 | linebuf[curb] = '\0'; | |
2818 | fprintf(stderr, "0x%08" PRIx64 " : %s\n", cur, linebuf); | |
2819 | } | |
2820 | } | |
2821 | } | |
2822 | ||
e0b266f0 DDAG |
2823 | /* **** functions for postcopy ***** */ |
2824 | ||
ced1c616 PB |
2825 | void ram_postcopy_migrated_memory_release(MigrationState *ms) |
2826 | { | |
2827 | struct RAMBlock *block; | |
ced1c616 | 2828 | |
fbd162e6 | 2829 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
6b6712ef JQ |
2830 | unsigned long *bitmap = block->bmap; |
2831 | unsigned long range = block->used_length >> TARGET_PAGE_BITS; | |
2832 | unsigned long run_start = find_next_zero_bit(bitmap, range, 0); | |
ced1c616 PB |
2833 | |
2834 | while (run_start < range) { | |
2835 | unsigned long run_end = find_next_bit(bitmap, range, run_start + 1); | |
aaa2064c | 2836 | ram_discard_range(block->idstr, run_start << TARGET_PAGE_BITS, |
ced1c616 PB |
2837 | (run_end - run_start) << TARGET_PAGE_BITS); |
2838 | run_start = find_next_zero_bit(bitmap, range, run_end + 1); | |
2839 | } | |
2840 | } | |
2841 | } | |
2842 | ||
3d0684b2 JQ |
2843 | /** |
2844 | * postcopy_send_discard_bm_ram: discard a RAMBlock | |
2845 | * | |
2846 | * Returns zero on success | |
2847 | * | |
e0b266f0 | 2848 | * Callback from postcopy_each_ram_send_discard for each RAMBlock |
3d0684b2 JQ |
2849 | * |
2850 | * @ms: current migration state | |
89dab31b | 2851 | * @block: RAMBlock to discard |
e0b266f0 | 2852 | */ |
810cf2bb | 2853 | static int postcopy_send_discard_bm_ram(MigrationState *ms, RAMBlock *block) |
e0b266f0 | 2854 | { |
6b6712ef | 2855 | unsigned long end = block->used_length >> TARGET_PAGE_BITS; |
e0b266f0 | 2856 | unsigned long current; |
1e7cf8c3 | 2857 | unsigned long *bitmap = block->bmap; |
e0b266f0 | 2858 | |
6b6712ef | 2859 | for (current = 0; current < end; ) { |
1e7cf8c3 | 2860 | unsigned long one = find_next_bit(bitmap, end, current); |
33a5cb62 | 2861 | unsigned long zero, discard_length; |
e0b266f0 | 2862 | |
33a5cb62 WY |
2863 | if (one >= end) { |
2864 | break; | |
2865 | } | |
e0b266f0 | 2866 | |
1e7cf8c3 | 2867 | zero = find_next_zero_bit(bitmap, end, one + 1); |
33a5cb62 WY |
2868 | |
2869 | if (zero >= end) { | |
2870 | discard_length = end - one; | |
e0b266f0 | 2871 | } else { |
33a5cb62 WY |
2872 | discard_length = zero - one; |
2873 | } | |
810cf2bb | 2874 | postcopy_discard_send_range(ms, one, discard_length); |
33a5cb62 | 2875 | current = one + discard_length; |
e0b266f0 DDAG |
2876 | } |
2877 | ||
2878 | return 0; | |
2879 | } | |
2880 | ||
3d0684b2 JQ |
2881 | /** |
2882 | * postcopy_each_ram_send_discard: discard all RAMBlocks | |
2883 | * | |
2884 | * Returns 0 for success or negative for error | |
2885 | * | |
e0b266f0 DDAG |
2886 | * Utility for the outgoing postcopy code. |
2887 | * Calls postcopy_send_discard_bm_ram for each RAMBlock | |
2888 | * passing it bitmap indexes and name. | |
e0b266f0 DDAG |
2889 | * (qemu_ram_foreach_block ends up passing unscaled lengths |
2890 | * which would mean postcopy code would have to deal with target page) | |
3d0684b2 JQ |
2891 | * |
2892 | * @ms: current migration state | |
e0b266f0 DDAG |
2893 | */ |
2894 | static int postcopy_each_ram_send_discard(MigrationState *ms) | |
2895 | { | |
2896 | struct RAMBlock *block; | |
2897 | int ret; | |
2898 | ||
fbd162e6 | 2899 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
810cf2bb | 2900 | postcopy_discard_send_init(ms, block->idstr); |
e0b266f0 DDAG |
2901 | |
2902 | /* | |
2903 | * Postcopy sends chunks of bitmap over the wire, but it | |
2904 | * just needs indexes at this point, avoids it having | |
2905 | * target page specific code. | |
2906 | */ | |
810cf2bb WY |
2907 | ret = postcopy_send_discard_bm_ram(ms, block); |
2908 | postcopy_discard_send_finish(ms); | |
e0b266f0 DDAG |
2909 | if (ret) { |
2910 | return ret; | |
2911 | } | |
2912 | } | |
2913 | ||
2914 | return 0; | |
2915 | } | |
2916 | ||
3d0684b2 | 2917 | /** |
8324ef86 | 2918 | * postcopy_chunk_hostpages_pass: canonicalize bitmap in hostpages |
3d0684b2 JQ |
2919 | * |
2920 | * Helper for postcopy_chunk_hostpages; it's called twice to | |
2921 | * canonicalize the two bitmaps, that are similar, but one is | |
2922 | * inverted. | |
99e314eb | 2923 | * |
3d0684b2 JQ |
2924 | * Postcopy requires that all target pages in a hostpage are dirty or |
2925 | * clean, not a mix. This function canonicalizes the bitmaps. | |
99e314eb | 2926 | * |
3d0684b2 | 2927 | * @ms: current migration state |
3d0684b2 | 2928 | * @block: block that contains the page we want to canonicalize |
99e314eb | 2929 | */ |
1e7cf8c3 | 2930 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, RAMBlock *block) |
99e314eb | 2931 | { |
53518d94 | 2932 | RAMState *rs = ram_state; |
6b6712ef | 2933 | unsigned long *bitmap = block->bmap; |
29c59172 | 2934 | unsigned int host_ratio = block->page_size / TARGET_PAGE_SIZE; |
6b6712ef | 2935 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
99e314eb DDAG |
2936 | unsigned long run_start; |
2937 | ||
29c59172 DDAG |
2938 | if (block->page_size == TARGET_PAGE_SIZE) { |
2939 | /* Easy case - TPS==HPS for a non-huge page RAMBlock */ | |
2940 | return; | |
2941 | } | |
2942 | ||
1e7cf8c3 WY |
2943 | /* Find a dirty page */ |
2944 | run_start = find_next_bit(bitmap, pages, 0); | |
99e314eb | 2945 | |
6b6712ef | 2946 | while (run_start < pages) { |
99e314eb DDAG |
2947 | |
2948 | /* | |
2949 | * If the start of this run of pages is in the middle of a host | |
2950 | * page, then we need to fixup this host page. | |
2951 | */ | |
9dec3cc3 | 2952 | if (QEMU_IS_ALIGNED(run_start, host_ratio)) { |
99e314eb | 2953 | /* Find the end of this run */ |
1e7cf8c3 | 2954 | run_start = find_next_zero_bit(bitmap, pages, run_start + 1); |
99e314eb DDAG |
2955 | /* |
2956 | * If the end isn't at the start of a host page, then the | |
2957 | * run doesn't finish at the end of a host page | |
2958 | * and we need to discard. | |
2959 | */ | |
99e314eb DDAG |
2960 | } |
2961 | ||
9dec3cc3 | 2962 | if (!QEMU_IS_ALIGNED(run_start, host_ratio)) { |
99e314eb | 2963 | unsigned long page; |
dad45ab2 WY |
2964 | unsigned long fixup_start_addr = QEMU_ALIGN_DOWN(run_start, |
2965 | host_ratio); | |
2966 | run_start = QEMU_ALIGN_UP(run_start, host_ratio); | |
99e314eb | 2967 | |
99e314eb DDAG |
2968 | /* Clean up the bitmap */ |
2969 | for (page = fixup_start_addr; | |
2970 | page < fixup_start_addr + host_ratio; page++) { | |
99e314eb DDAG |
2971 | /* |
2972 | * Remark them as dirty, updating the count for any pages | |
2973 | * that weren't previously dirty. | |
2974 | */ | |
0d8ec885 | 2975 | rs->migration_dirty_pages += !test_and_set_bit(page, bitmap); |
99e314eb DDAG |
2976 | } |
2977 | } | |
2978 | ||
1e7cf8c3 WY |
2979 | /* Find the next dirty page for the next iteration */ |
2980 | run_start = find_next_bit(bitmap, pages, run_start); | |
99e314eb DDAG |
2981 | } |
2982 | } | |
2983 | ||
3d0684b2 | 2984 | /** |
89dab31b | 2985 | * postcopy_chunk_hostpages: discard any partially sent host page |
3d0684b2 | 2986 | * |
99e314eb DDAG |
2987 | * Utility for the outgoing postcopy code. |
2988 | * | |
2989 | * Discard any partially sent host-page size chunks, mark any partially | |
29c59172 DDAG |
2990 | * dirty host-page size chunks as all dirty. In this case the host-page |
2991 | * is the host-page for the particular RAMBlock, i.e. it might be a huge page | |
99e314eb | 2992 | * |
3d0684b2 JQ |
2993 | * Returns zero on success |
2994 | * | |
2995 | * @ms: current migration state | |
6b6712ef | 2996 | * @block: block we want to work with |
99e314eb | 2997 | */ |
6b6712ef | 2998 | static int postcopy_chunk_hostpages(MigrationState *ms, RAMBlock *block) |
99e314eb | 2999 | { |
810cf2bb | 3000 | postcopy_discard_send_init(ms, block->idstr); |
99e314eb | 3001 | |
6b6712ef | 3002 | /* |
1e7cf8c3 | 3003 | * Ensure that all partially dirty host pages are made fully dirty. |
6b6712ef | 3004 | */ |
1e7cf8c3 | 3005 | postcopy_chunk_hostpages_pass(ms, block); |
99e314eb | 3006 | |
810cf2bb | 3007 | postcopy_discard_send_finish(ms); |
99e314eb DDAG |
3008 | return 0; |
3009 | } | |
3010 | ||
3d0684b2 JQ |
3011 | /** |
3012 | * ram_postcopy_send_discard_bitmap: transmit the discard bitmap | |
3013 | * | |
3014 | * Returns zero on success | |
3015 | * | |
e0b266f0 DDAG |
3016 | * Transmit the set of pages to be discarded after precopy to the target |
3017 | * these are pages that: | |
3018 | * a) Have been previously transmitted but are now dirty again | |
3019 | * b) Pages that have never been transmitted, this ensures that | |
3020 | * any pages on the destination that have been mapped by background | |
3021 | * tasks get discarded (transparent huge pages is the specific concern) | |
3022 | * Hopefully this is pretty sparse | |
3d0684b2 JQ |
3023 | * |
3024 | * @ms: current migration state | |
e0b266f0 DDAG |
3025 | */ |
3026 | int ram_postcopy_send_discard_bitmap(MigrationState *ms) | |
3027 | { | |
53518d94 | 3028 | RAMState *rs = ram_state; |
6b6712ef | 3029 | RAMBlock *block; |
e0b266f0 | 3030 | int ret; |
e0b266f0 | 3031 | |
89ac5a1d | 3032 | RCU_READ_LOCK_GUARD(); |
e0b266f0 DDAG |
3033 | |
3034 | /* This should be our last sync, the src is now paused */ | |
eb859c53 | 3035 | migration_bitmap_sync(rs); |
e0b266f0 | 3036 | |
6b6712ef JQ |
3037 | /* Easiest way to make sure we don't resume in the middle of a host-page */ |
3038 | rs->last_seen_block = NULL; | |
3039 | rs->last_sent_block = NULL; | |
3040 | rs->last_page = 0; | |
e0b266f0 | 3041 | |
fbd162e6 | 3042 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
6b6712ef JQ |
3043 | /* Deal with TPS != HPS and huge pages */ |
3044 | ret = postcopy_chunk_hostpages(ms, block); | |
3045 | if (ret) { | |
6b6712ef JQ |
3046 | return ret; |
3047 | } | |
e0b266f0 | 3048 | |
e0b266f0 | 3049 | #ifdef DEBUG_POSTCOPY |
1e7cf8c3 WY |
3050 | ram_debug_dump_bitmap(block->bmap, true, |
3051 | block->used_length >> TARGET_PAGE_BITS); | |
e0b266f0 | 3052 | #endif |
6b6712ef JQ |
3053 | } |
3054 | trace_ram_postcopy_send_discard_bitmap(); | |
e0b266f0 DDAG |
3055 | |
3056 | ret = postcopy_each_ram_send_discard(ms); | |
e0b266f0 DDAG |
3057 | |
3058 | return ret; | |
3059 | } | |
3060 | ||
3d0684b2 JQ |
3061 | /** |
3062 | * ram_discard_range: discard dirtied pages at the beginning of postcopy | |
e0b266f0 | 3063 | * |
3d0684b2 | 3064 | * Returns zero on success |
e0b266f0 | 3065 | * |
36449157 JQ |
3066 | * @rbname: name of the RAMBlock of the request. NULL means the |
3067 | * same that last one. | |
3d0684b2 JQ |
3068 | * @start: RAMBlock starting page |
3069 | * @length: RAMBlock size | |
e0b266f0 | 3070 | */ |
aaa2064c | 3071 | int ram_discard_range(const char *rbname, uint64_t start, size_t length) |
e0b266f0 DDAG |
3072 | { |
3073 | int ret = -1; | |
3074 | ||
36449157 | 3075 | trace_ram_discard_range(rbname, start, length); |
d3a5038c | 3076 | |
89ac5a1d | 3077 | RCU_READ_LOCK_GUARD(); |
36449157 | 3078 | RAMBlock *rb = qemu_ram_block_by_name(rbname); |
e0b266f0 DDAG |
3079 | |
3080 | if (!rb) { | |
36449157 | 3081 | error_report("ram_discard_range: Failed to find block '%s'", rbname); |
e0b266f0 DDAG |
3082 | goto err; |
3083 | } | |
3084 | ||
814bb08f PX |
3085 | /* |
3086 | * On source VM, we don't need to update the received bitmap since | |
3087 | * we don't even have one. | |
3088 | */ | |
3089 | if (rb->receivedmap) { | |
3090 | bitmap_clear(rb->receivedmap, start >> qemu_target_page_bits(), | |
3091 | length >> qemu_target_page_bits()); | |
3092 | } | |
3093 | ||
d3a5038c | 3094 | ret = ram_block_discard_range(rb, start, length); |
e0b266f0 DDAG |
3095 | |
3096 | err: | |
e0b266f0 DDAG |
3097 | return ret; |
3098 | } | |
3099 | ||
84593a08 PX |
3100 | /* |
3101 | * For every allocation, we will try not to crash the VM if the | |
3102 | * allocation failed. | |
3103 | */ | |
3104 | static int xbzrle_init(void) | |
3105 | { | |
3106 | Error *local_err = NULL; | |
3107 | ||
3108 | if (!migrate_use_xbzrle()) { | |
3109 | return 0; | |
3110 | } | |
3111 | ||
3112 | XBZRLE_cache_lock(); | |
3113 | ||
3114 | XBZRLE.zero_target_page = g_try_malloc0(TARGET_PAGE_SIZE); | |
3115 | if (!XBZRLE.zero_target_page) { | |
3116 | error_report("%s: Error allocating zero page", __func__); | |
3117 | goto err_out; | |
3118 | } | |
3119 | ||
3120 | XBZRLE.cache = cache_init(migrate_xbzrle_cache_size(), | |
3121 | TARGET_PAGE_SIZE, &local_err); | |
3122 | if (!XBZRLE.cache) { | |
3123 | error_report_err(local_err); | |
3124 | goto free_zero_page; | |
3125 | } | |
3126 | ||
3127 | XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); | |
3128 | if (!XBZRLE.encoded_buf) { | |
3129 | error_report("%s: Error allocating encoded_buf", __func__); | |
3130 | goto free_cache; | |
3131 | } | |
3132 | ||
3133 | XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); | |
3134 | if (!XBZRLE.current_buf) { | |
3135 | error_report("%s: Error allocating current_buf", __func__); | |
3136 | goto free_encoded_buf; | |
3137 | } | |
3138 | ||
3139 | /* We are all good */ | |
3140 | XBZRLE_cache_unlock(); | |
3141 | return 0; | |
3142 | ||
3143 | free_encoded_buf: | |
3144 | g_free(XBZRLE.encoded_buf); | |
3145 | XBZRLE.encoded_buf = NULL; | |
3146 | free_cache: | |
3147 | cache_fini(XBZRLE.cache); | |
3148 | XBZRLE.cache = NULL; | |
3149 | free_zero_page: | |
3150 | g_free(XBZRLE.zero_target_page); | |
3151 | XBZRLE.zero_target_page = NULL; | |
3152 | err_out: | |
3153 | XBZRLE_cache_unlock(); | |
3154 | return -ENOMEM; | |
3155 | } | |
3156 | ||
53518d94 | 3157 | static int ram_state_init(RAMState **rsp) |
56e93d26 | 3158 | { |
7d00ee6a PX |
3159 | *rsp = g_try_new0(RAMState, 1); |
3160 | ||
3161 | if (!*rsp) { | |
3162 | error_report("%s: Init ramstate fail", __func__); | |
3163 | return -1; | |
3164 | } | |
53518d94 JQ |
3165 | |
3166 | qemu_mutex_init(&(*rsp)->bitmap_mutex); | |
3167 | qemu_mutex_init(&(*rsp)->src_page_req_mutex); | |
3168 | QSIMPLEQ_INIT(&(*rsp)->src_page_requests); | |
56e93d26 | 3169 | |
7d00ee6a | 3170 | /* |
40c4d4a8 IR |
3171 | * Count the total number of pages used by ram blocks not including any |
3172 | * gaps due to alignment or unplugs. | |
03158519 | 3173 | * This must match with the initial values of dirty bitmap. |
7d00ee6a | 3174 | */ |
40c4d4a8 | 3175 | (*rsp)->migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS; |
7d00ee6a PX |
3176 | ram_state_reset(*rsp); |
3177 | ||
3178 | return 0; | |
3179 | } | |
3180 | ||
d6eff5d7 | 3181 | static void ram_list_init_bitmaps(void) |
7d00ee6a | 3182 | { |
002cad6b | 3183 | MigrationState *ms = migrate_get_current(); |
d6eff5d7 PX |
3184 | RAMBlock *block; |
3185 | unsigned long pages; | |
002cad6b | 3186 | uint8_t shift; |
56e93d26 | 3187 | |
0827b9e9 AA |
3188 | /* Skip setting bitmap if there is no RAM */ |
3189 | if (ram_bytes_total()) { | |
002cad6b PX |
3190 | shift = ms->clear_bitmap_shift; |
3191 | if (shift > CLEAR_BITMAP_SHIFT_MAX) { | |
3192 | error_report("clear_bitmap_shift (%u) too big, using " | |
3193 | "max value (%u)", shift, CLEAR_BITMAP_SHIFT_MAX); | |
3194 | shift = CLEAR_BITMAP_SHIFT_MAX; | |
3195 | } else if (shift < CLEAR_BITMAP_SHIFT_MIN) { | |
3196 | error_report("clear_bitmap_shift (%u) too small, using " | |
3197 | "min value (%u)", shift, CLEAR_BITMAP_SHIFT_MIN); | |
3198 | shift = CLEAR_BITMAP_SHIFT_MIN; | |
3199 | } | |
3200 | ||
fbd162e6 | 3201 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
d6eff5d7 | 3202 | pages = block->max_length >> TARGET_PAGE_BITS; |
03158519 WY |
3203 | /* |
3204 | * The initial dirty bitmap for migration must be set with all | |
3205 | * ones to make sure we'll migrate every guest RAM page to | |
3206 | * destination. | |
40c4d4a8 IR |
3207 | * Here we set RAMBlock.bmap all to 1 because when rebegin a |
3208 | * new migration after a failed migration, ram_list. | |
3209 | * dirty_memory[DIRTY_MEMORY_MIGRATION] don't include the whole | |
3210 | * guest memory. | |
03158519 | 3211 | */ |
6b6712ef | 3212 | block->bmap = bitmap_new(pages); |
40c4d4a8 | 3213 | bitmap_set(block->bmap, 0, pages); |
002cad6b PX |
3214 | block->clear_bmap_shift = shift; |
3215 | block->clear_bmap = bitmap_new(clear_bmap_size(pages, shift)); | |
0827b9e9 | 3216 | } |
f3f491fc | 3217 | } |
d6eff5d7 PX |
3218 | } |
3219 | ||
3220 | static void ram_init_bitmaps(RAMState *rs) | |
3221 | { | |
3222 | /* For memory_global_dirty_log_start below. */ | |
3223 | qemu_mutex_lock_iothread(); | |
3224 | qemu_mutex_lock_ramlist(); | |
f3f491fc | 3225 | |
89ac5a1d DDAG |
3226 | WITH_RCU_READ_LOCK_GUARD() { |
3227 | ram_list_init_bitmaps(); | |
3228 | memory_global_dirty_log_start(); | |
3229 | migration_bitmap_sync_precopy(rs); | |
3230 | } | |
56e93d26 | 3231 | qemu_mutex_unlock_ramlist(); |
49877834 | 3232 | qemu_mutex_unlock_iothread(); |
d6eff5d7 PX |
3233 | } |
3234 | ||
3235 | static int ram_init_all(RAMState **rsp) | |
3236 | { | |
3237 | if (ram_state_init(rsp)) { | |
3238 | return -1; | |
3239 | } | |
3240 | ||
3241 | if (xbzrle_init()) { | |
3242 | ram_state_cleanup(rsp); | |
3243 | return -1; | |
3244 | } | |
3245 | ||
3246 | ram_init_bitmaps(*rsp); | |
a91246c9 HZ |
3247 | |
3248 | return 0; | |
3249 | } | |
3250 | ||
08614f34 PX |
3251 | static void ram_state_resume_prepare(RAMState *rs, QEMUFile *out) |
3252 | { | |
3253 | RAMBlock *block; | |
3254 | uint64_t pages = 0; | |
3255 | ||
3256 | /* | |
3257 | * Postcopy is not using xbzrle/compression, so no need for that. | |
3258 | * Also, since source are already halted, we don't need to care | |
3259 | * about dirty page logging as well. | |
3260 | */ | |
3261 | ||
fbd162e6 | 3262 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
08614f34 PX |
3263 | pages += bitmap_count_one(block->bmap, |
3264 | block->used_length >> TARGET_PAGE_BITS); | |
3265 | } | |
3266 | ||
3267 | /* This may not be aligned with current bitmaps. Recalculate. */ | |
3268 | rs->migration_dirty_pages = pages; | |
3269 | ||
3270 | rs->last_seen_block = NULL; | |
3271 | rs->last_sent_block = NULL; | |
3272 | rs->last_page = 0; | |
3273 | rs->last_version = ram_list.version; | |
3274 | /* | |
3275 | * Disable the bulk stage, otherwise we'll resend the whole RAM no | |
3276 | * matter what we have sent. | |
3277 | */ | |
3278 | rs->ram_bulk_stage = false; | |
3279 | ||
3280 | /* Update RAMState cache of output QEMUFile */ | |
3281 | rs->f = out; | |
3282 | ||
3283 | trace_ram_state_resume_prepare(pages); | |
3284 | } | |
3285 | ||
6bcb05fc WW |
3286 | /* |
3287 | * This function clears bits of the free pages reported by the caller from the | |
3288 | * migration dirty bitmap. @addr is the host address corresponding to the | |
3289 | * start of the continuous guest free pages, and @len is the total bytes of | |
3290 | * those pages. | |
3291 | */ | |
3292 | void qemu_guest_free_page_hint(void *addr, size_t len) | |
3293 | { | |
3294 | RAMBlock *block; | |
3295 | ram_addr_t offset; | |
3296 | size_t used_len, start, npages; | |
3297 | MigrationState *s = migrate_get_current(); | |
3298 | ||
3299 | /* This function is currently expected to be used during live migration */ | |
3300 | if (!migration_is_setup_or_active(s->state)) { | |
3301 | return; | |
3302 | } | |
3303 | ||
3304 | for (; len > 0; len -= used_len, addr += used_len) { | |
3305 | block = qemu_ram_block_from_host(addr, false, &offset); | |
3306 | if (unlikely(!block || offset >= block->used_length)) { | |
3307 | /* | |
3308 | * The implementation might not support RAMBlock resize during | |
3309 | * live migration, but it could happen in theory with future | |
3310 | * updates. So we add a check here to capture that case. | |
3311 | */ | |
3312 | error_report_once("%s unexpected error", __func__); | |
3313 | return; | |
3314 | } | |
3315 | ||
3316 | if (len <= block->used_length - offset) { | |
3317 | used_len = len; | |
3318 | } else { | |
3319 | used_len = block->used_length - offset; | |
3320 | } | |
3321 | ||
3322 | start = offset >> TARGET_PAGE_BITS; | |
3323 | npages = used_len >> TARGET_PAGE_BITS; | |
3324 | ||
3325 | qemu_mutex_lock(&ram_state->bitmap_mutex); | |
3326 | ram_state->migration_dirty_pages -= | |
3327 | bitmap_count_one_with_offset(block->bmap, start, npages); | |
3328 | bitmap_clear(block->bmap, start, npages); | |
3329 | qemu_mutex_unlock(&ram_state->bitmap_mutex); | |
3330 | } | |
3331 | } | |
3332 | ||
3d0684b2 JQ |
3333 | /* |
3334 | * Each of ram_save_setup, ram_save_iterate and ram_save_complete has | |
a91246c9 HZ |
3335 | * long-running RCU critical section. When rcu-reclaims in the code |
3336 | * start to become numerous it will be necessary to reduce the | |
3337 | * granularity of these critical sections. | |
3338 | */ | |
3339 | ||
3d0684b2 JQ |
3340 | /** |
3341 | * ram_save_setup: Setup RAM for migration | |
3342 | * | |
3343 | * Returns zero to indicate success and negative for error | |
3344 | * | |
3345 | * @f: QEMUFile where to send the data | |
3346 | * @opaque: RAMState pointer | |
3347 | */ | |
a91246c9 HZ |
3348 | static int ram_save_setup(QEMUFile *f, void *opaque) |
3349 | { | |
53518d94 | 3350 | RAMState **rsp = opaque; |
a91246c9 HZ |
3351 | RAMBlock *block; |
3352 | ||
dcaf446e XG |
3353 | if (compress_threads_save_setup()) { |
3354 | return -1; | |
3355 | } | |
3356 | ||
a91246c9 HZ |
3357 | /* migration has already setup the bitmap, reuse it. */ |
3358 | if (!migration_in_colo_state()) { | |
7d00ee6a | 3359 | if (ram_init_all(rsp) != 0) { |
dcaf446e | 3360 | compress_threads_save_cleanup(); |
a91246c9 | 3361 | return -1; |
53518d94 | 3362 | } |
a91246c9 | 3363 | } |
53518d94 | 3364 | (*rsp)->f = f; |
a91246c9 | 3365 | |
0e6ebd48 DDAG |
3366 | WITH_RCU_READ_LOCK_GUARD() { |
3367 | qemu_put_be64(f, ram_bytes_total_common(true) | RAM_SAVE_FLAG_MEM_SIZE); | |
56e93d26 | 3368 | |
0e6ebd48 DDAG |
3369 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
3370 | qemu_put_byte(f, strlen(block->idstr)); | |
3371 | qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); | |
3372 | qemu_put_be64(f, block->used_length); | |
3373 | if (migrate_postcopy_ram() && block->page_size != | |
3374 | qemu_host_page_size) { | |
3375 | qemu_put_be64(f, block->page_size); | |
3376 | } | |
3377 | if (migrate_ignore_shared()) { | |
3378 | qemu_put_be64(f, block->mr->addr); | |
3379 | } | |
fbd162e6 | 3380 | } |
56e93d26 JQ |
3381 | } |
3382 | ||
56e93d26 JQ |
3383 | ram_control_before_iterate(f, RAM_CONTROL_SETUP); |
3384 | ram_control_after_iterate(f, RAM_CONTROL_SETUP); | |
3385 | ||
1b81c974 | 3386 | multifd_send_sync_main(*rsp); |
56e93d26 | 3387 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
35374cbd | 3388 | qemu_fflush(f); |
56e93d26 JQ |
3389 | |
3390 | return 0; | |
3391 | } | |
3392 | ||
3d0684b2 JQ |
3393 | /** |
3394 | * ram_save_iterate: iterative stage for migration | |
3395 | * | |
3396 | * Returns zero to indicate success and negative for error | |
3397 | * | |
3398 | * @f: QEMUFile where to send the data | |
3399 | * @opaque: RAMState pointer | |
3400 | */ | |
56e93d26 JQ |
3401 | static int ram_save_iterate(QEMUFile *f, void *opaque) |
3402 | { | |
53518d94 JQ |
3403 | RAMState **temp = opaque; |
3404 | RAMState *rs = *temp; | |
56e93d26 JQ |
3405 | int ret; |
3406 | int i; | |
3407 | int64_t t0; | |
5c90308f | 3408 | int done = 0; |
56e93d26 | 3409 | |
b2557345 PL |
3410 | if (blk_mig_bulk_active()) { |
3411 | /* Avoid transferring ram during bulk phase of block migration as | |
3412 | * the bulk phase will usually take a long time and transferring | |
3413 | * ram updates during that time is pointless. */ | |
3414 | goto out; | |
3415 | } | |
3416 | ||
89ac5a1d DDAG |
3417 | WITH_RCU_READ_LOCK_GUARD() { |
3418 | if (ram_list.version != rs->last_version) { | |
3419 | ram_state_reset(rs); | |
3420 | } | |
56e93d26 | 3421 | |
89ac5a1d DDAG |
3422 | /* Read version before ram_list.blocks */ |
3423 | smp_rmb(); | |
56e93d26 | 3424 | |
89ac5a1d | 3425 | ram_control_before_iterate(f, RAM_CONTROL_ROUND); |
56e93d26 | 3426 | |
89ac5a1d DDAG |
3427 | t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
3428 | i = 0; | |
3429 | while ((ret = qemu_file_rate_limit(f)) == 0 || | |
3430 | !QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
3431 | int pages; | |
e03a34f8 | 3432 | |
89ac5a1d DDAG |
3433 | if (qemu_file_get_error(f)) { |
3434 | break; | |
3435 | } | |
e8f3735f | 3436 | |
89ac5a1d DDAG |
3437 | pages = ram_find_and_save_block(rs, false); |
3438 | /* no more pages to sent */ | |
3439 | if (pages == 0) { | |
3440 | done = 1; | |
3441 | break; | |
3442 | } | |
e8f3735f | 3443 | |
89ac5a1d DDAG |
3444 | if (pages < 0) { |
3445 | qemu_file_set_error(f, pages); | |
56e93d26 JQ |
3446 | break; |
3447 | } | |
89ac5a1d DDAG |
3448 | |
3449 | rs->target_page_count += pages; | |
3450 | ||
3451 | /* | |
3452 | * we want to check in the 1st loop, just in case it was the 1st | |
3453 | * time and we had to sync the dirty bitmap. | |
3454 | * qemu_clock_get_ns() is a bit expensive, so we only check each | |
3455 | * some iterations | |
3456 | */ | |
3457 | if ((i & 63) == 0) { | |
3458 | uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / | |
3459 | 1000000; | |
3460 | if (t1 > MAX_WAIT) { | |
3461 | trace_ram_save_iterate_big_wait(t1, i); | |
3462 | break; | |
3463 | } | |
3464 | } | |
3465 | i++; | |
56e93d26 | 3466 | } |
56e93d26 | 3467 | } |
56e93d26 JQ |
3468 | |
3469 | /* | |
3470 | * Must occur before EOS (or any QEMUFile operation) | |
3471 | * because of RDMA protocol. | |
3472 | */ | |
3473 | ram_control_after_iterate(f, RAM_CONTROL_ROUND); | |
3474 | ||
b2557345 | 3475 | out: |
1b81c974 | 3476 | multifd_send_sync_main(rs); |
56e93d26 | 3477 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
35374cbd | 3478 | qemu_fflush(f); |
9360447d | 3479 | ram_counters.transferred += 8; |
56e93d26 JQ |
3480 | |
3481 | ret = qemu_file_get_error(f); | |
3482 | if (ret < 0) { | |
3483 | return ret; | |
3484 | } | |
3485 | ||
5c90308f | 3486 | return done; |
56e93d26 JQ |
3487 | } |
3488 | ||
3d0684b2 JQ |
3489 | /** |
3490 | * ram_save_complete: function called to send the remaining amount of ram | |
3491 | * | |
e8f3735f | 3492 | * Returns zero to indicate success or negative on error |
3d0684b2 JQ |
3493 | * |
3494 | * Called with iothread lock | |
3495 | * | |
3496 | * @f: QEMUFile where to send the data | |
3497 | * @opaque: RAMState pointer | |
3498 | */ | |
56e93d26 JQ |
3499 | static int ram_save_complete(QEMUFile *f, void *opaque) |
3500 | { | |
53518d94 JQ |
3501 | RAMState **temp = opaque; |
3502 | RAMState *rs = *temp; | |
e8f3735f | 3503 | int ret = 0; |
6f37bb8b | 3504 | |
89ac5a1d DDAG |
3505 | WITH_RCU_READ_LOCK_GUARD() { |
3506 | if (!migration_in_postcopy()) { | |
3507 | migration_bitmap_sync_precopy(rs); | |
3508 | } | |
56e93d26 | 3509 | |
89ac5a1d | 3510 | ram_control_before_iterate(f, RAM_CONTROL_FINISH); |
56e93d26 | 3511 | |
89ac5a1d | 3512 | /* try transferring iterative blocks of memory */ |
56e93d26 | 3513 | |
89ac5a1d DDAG |
3514 | /* flush all remaining blocks regardless of rate limiting */ |
3515 | while (true) { | |
3516 | int pages; | |
56e93d26 | 3517 | |
89ac5a1d DDAG |
3518 | pages = ram_find_and_save_block(rs, !migration_in_colo_state()); |
3519 | /* no more blocks to sent */ | |
3520 | if (pages == 0) { | |
3521 | break; | |
3522 | } | |
3523 | if (pages < 0) { | |
3524 | ret = pages; | |
3525 | break; | |
3526 | } | |
e8f3735f | 3527 | } |
56e93d26 | 3528 | |
89ac5a1d DDAG |
3529 | flush_compressed_data(rs); |
3530 | ram_control_after_iterate(f, RAM_CONTROL_FINISH); | |
3531 | } | |
d09a6fde | 3532 | |
1b81c974 | 3533 | multifd_send_sync_main(rs); |
56e93d26 | 3534 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
35374cbd | 3535 | qemu_fflush(f); |
56e93d26 | 3536 | |
e8f3735f | 3537 | return ret; |
56e93d26 JQ |
3538 | } |
3539 | ||
c31b098f | 3540 | static void ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size, |
47995026 VSO |
3541 | uint64_t *res_precopy_only, |
3542 | uint64_t *res_compatible, | |
3543 | uint64_t *res_postcopy_only) | |
56e93d26 | 3544 | { |
53518d94 JQ |
3545 | RAMState **temp = opaque; |
3546 | RAMState *rs = *temp; | |
56e93d26 JQ |
3547 | uint64_t remaining_size; |
3548 | ||
9edabd4d | 3549 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 3550 | |
5727309d | 3551 | if (!migration_in_postcopy() && |
663e6c1d | 3552 | remaining_size < max_size) { |
56e93d26 | 3553 | qemu_mutex_lock_iothread(); |
89ac5a1d DDAG |
3554 | WITH_RCU_READ_LOCK_GUARD() { |
3555 | migration_bitmap_sync_precopy(rs); | |
3556 | } | |
56e93d26 | 3557 | qemu_mutex_unlock_iothread(); |
9edabd4d | 3558 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 3559 | } |
c31b098f | 3560 | |
86e1167e VSO |
3561 | if (migrate_postcopy_ram()) { |
3562 | /* We can do postcopy, and all the data is postcopiable */ | |
47995026 | 3563 | *res_compatible += remaining_size; |
86e1167e | 3564 | } else { |
47995026 | 3565 | *res_precopy_only += remaining_size; |
86e1167e | 3566 | } |
56e93d26 JQ |
3567 | } |
3568 | ||
3569 | static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) | |
3570 | { | |
3571 | unsigned int xh_len; | |
3572 | int xh_flags; | |
063e760a | 3573 | uint8_t *loaded_data; |
56e93d26 | 3574 | |
56e93d26 JQ |
3575 | /* extract RLE header */ |
3576 | xh_flags = qemu_get_byte(f); | |
3577 | xh_len = qemu_get_be16(f); | |
3578 | ||
3579 | if (xh_flags != ENCODING_FLAG_XBZRLE) { | |
3580 | error_report("Failed to load XBZRLE page - wrong compression!"); | |
3581 | return -1; | |
3582 | } | |
3583 | ||
3584 | if (xh_len > TARGET_PAGE_SIZE) { | |
3585 | error_report("Failed to load XBZRLE page - len overflow!"); | |
3586 | return -1; | |
3587 | } | |
f265e0e4 | 3588 | loaded_data = XBZRLE.decoded_buf; |
56e93d26 | 3589 | /* load data and decode */ |
f265e0e4 | 3590 | /* it can change loaded_data to point to an internal buffer */ |
063e760a | 3591 | qemu_get_buffer_in_place(f, &loaded_data, xh_len); |
56e93d26 JQ |
3592 | |
3593 | /* decode RLE */ | |
063e760a | 3594 | if (xbzrle_decode_buffer(loaded_data, xh_len, host, |
56e93d26 JQ |
3595 | TARGET_PAGE_SIZE) == -1) { |
3596 | error_report("Failed to load XBZRLE page - decode error!"); | |
3597 | return -1; | |
3598 | } | |
3599 | ||
3600 | return 0; | |
3601 | } | |
3602 | ||
3d0684b2 JQ |
3603 | /** |
3604 | * ram_block_from_stream: read a RAMBlock id from the migration stream | |
3605 | * | |
3606 | * Must be called from within a rcu critical section. | |
3607 | * | |
56e93d26 | 3608 | * Returns a pointer from within the RCU-protected ram_list. |
a7180877 | 3609 | * |
3d0684b2 JQ |
3610 | * @f: QEMUFile where to read the data from |
3611 | * @flags: Page flags (mostly to see if it's a continuation of previous block) | |
a7180877 | 3612 | */ |
3d0684b2 | 3613 | static inline RAMBlock *ram_block_from_stream(QEMUFile *f, int flags) |
56e93d26 JQ |
3614 | { |
3615 | static RAMBlock *block = NULL; | |
3616 | char id[256]; | |
3617 | uint8_t len; | |
3618 | ||
3619 | if (flags & RAM_SAVE_FLAG_CONTINUE) { | |
4c4bad48 | 3620 | if (!block) { |
56e93d26 JQ |
3621 | error_report("Ack, bad migration stream!"); |
3622 | return NULL; | |
3623 | } | |
4c4bad48 | 3624 | return block; |
56e93d26 JQ |
3625 | } |
3626 | ||
3627 | len = qemu_get_byte(f); | |
3628 | qemu_get_buffer(f, (uint8_t *)id, len); | |
3629 | id[len] = 0; | |
3630 | ||
e3dd7493 | 3631 | block = qemu_ram_block_by_name(id); |
4c4bad48 HZ |
3632 | if (!block) { |
3633 | error_report("Can't find block %s", id); | |
3634 | return NULL; | |
56e93d26 JQ |
3635 | } |
3636 | ||
fbd162e6 | 3637 | if (ramblock_is_ignored(block)) { |
b895de50 CLG |
3638 | error_report("block %s should not be migrated !", id); |
3639 | return NULL; | |
3640 | } | |
3641 | ||
4c4bad48 HZ |
3642 | return block; |
3643 | } | |
3644 | ||
3645 | static inline void *host_from_ram_block_offset(RAMBlock *block, | |
3646 | ram_addr_t offset) | |
3647 | { | |
3648 | if (!offset_in_ramblock(block, offset)) { | |
3649 | return NULL; | |
3650 | } | |
3651 | ||
3652 | return block->host + offset; | |
56e93d26 JQ |
3653 | } |
3654 | ||
13af18f2 ZC |
3655 | static inline void *colo_cache_from_block_offset(RAMBlock *block, |
3656 | ram_addr_t offset) | |
3657 | { | |
3658 | if (!offset_in_ramblock(block, offset)) { | |
3659 | return NULL; | |
3660 | } | |
3661 | if (!block->colo_cache) { | |
3662 | error_report("%s: colo_cache is NULL in block :%s", | |
3663 | __func__, block->idstr); | |
3664 | return NULL; | |
3665 | } | |
7d9acafa ZC |
3666 | |
3667 | /* | |
3668 | * During colo checkpoint, we need bitmap of these migrated pages. | |
3669 | * It help us to decide which pages in ram cache should be flushed | |
3670 | * into VM's RAM later. | |
3671 | */ | |
3672 | if (!test_and_set_bit(offset >> TARGET_PAGE_BITS, block->bmap)) { | |
3673 | ram_state->migration_dirty_pages++; | |
3674 | } | |
13af18f2 ZC |
3675 | return block->colo_cache + offset; |
3676 | } | |
3677 | ||
3d0684b2 JQ |
3678 | /** |
3679 | * ram_handle_compressed: handle the zero page case | |
3680 | * | |
56e93d26 JQ |
3681 | * If a page (or a whole RDMA chunk) has been |
3682 | * determined to be zero, then zap it. | |
3d0684b2 JQ |
3683 | * |
3684 | * @host: host address for the zero page | |
3685 | * @ch: what the page is filled from. We only support zero | |
3686 | * @size: size of the zero page | |
56e93d26 JQ |
3687 | */ |
3688 | void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) | |
3689 | { | |
3690 | if (ch != 0 || !is_zero_range(host, size)) { | |
3691 | memset(host, ch, size); | |
3692 | } | |
3693 | } | |
3694 | ||
797ca154 XG |
3695 | /* return the size after decompression, or negative value on error */ |
3696 | static int | |
3697 | qemu_uncompress_data(z_stream *stream, uint8_t *dest, size_t dest_len, | |
3698 | const uint8_t *source, size_t source_len) | |
3699 | { | |
3700 | int err; | |
3701 | ||
3702 | err = inflateReset(stream); | |
3703 | if (err != Z_OK) { | |
3704 | return -1; | |
3705 | } | |
3706 | ||
3707 | stream->avail_in = source_len; | |
3708 | stream->next_in = (uint8_t *)source; | |
3709 | stream->avail_out = dest_len; | |
3710 | stream->next_out = dest; | |
3711 | ||
3712 | err = inflate(stream, Z_NO_FLUSH); | |
3713 | if (err != Z_STREAM_END) { | |
3714 | return -1; | |
3715 | } | |
3716 | ||
3717 | return stream->total_out; | |
3718 | } | |
3719 | ||
56e93d26 JQ |
3720 | static void *do_data_decompress(void *opaque) |
3721 | { | |
3722 | DecompressParam *param = opaque; | |
3723 | unsigned long pagesize; | |
33d151f4 | 3724 | uint8_t *des; |
34ab9e97 | 3725 | int len, ret; |
56e93d26 | 3726 | |
33d151f4 | 3727 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 3728 | while (!param->quit) { |
33d151f4 LL |
3729 | if (param->des) { |
3730 | des = param->des; | |
3731 | len = param->len; | |
3732 | param->des = 0; | |
3733 | qemu_mutex_unlock(¶m->mutex); | |
3734 | ||
56e93d26 | 3735 | pagesize = TARGET_PAGE_SIZE; |
34ab9e97 XG |
3736 | |
3737 | ret = qemu_uncompress_data(¶m->stream, des, pagesize, | |
3738 | param->compbuf, len); | |
f548222c | 3739 | if (ret < 0 && migrate_get_current()->decompress_error_check) { |
34ab9e97 XG |
3740 | error_report("decompress data failed"); |
3741 | qemu_file_set_error(decomp_file, ret); | |
3742 | } | |
73a8912b | 3743 | |
33d151f4 LL |
3744 | qemu_mutex_lock(&decomp_done_lock); |
3745 | param->done = true; | |
3746 | qemu_cond_signal(&decomp_done_cond); | |
3747 | qemu_mutex_unlock(&decomp_done_lock); | |
3748 | ||
3749 | qemu_mutex_lock(¶m->mutex); | |
3750 | } else { | |
3751 | qemu_cond_wait(¶m->cond, ¶m->mutex); | |
3752 | } | |
56e93d26 | 3753 | } |
33d151f4 | 3754 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
3755 | |
3756 | return NULL; | |
3757 | } | |
3758 | ||
34ab9e97 | 3759 | static int wait_for_decompress_done(void) |
5533b2e9 LL |
3760 | { |
3761 | int idx, thread_count; | |
3762 | ||
3763 | if (!migrate_use_compression()) { | |
34ab9e97 | 3764 | return 0; |
5533b2e9 LL |
3765 | } |
3766 | ||
3767 | thread_count = migrate_decompress_threads(); | |
3768 | qemu_mutex_lock(&decomp_done_lock); | |
3769 | for (idx = 0; idx < thread_count; idx++) { | |
3770 | while (!decomp_param[idx].done) { | |
3771 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
3772 | } | |
3773 | } | |
3774 | qemu_mutex_unlock(&decomp_done_lock); | |
34ab9e97 | 3775 | return qemu_file_get_error(decomp_file); |
5533b2e9 LL |
3776 | } |
3777 | ||
f0afa331 | 3778 | static void compress_threads_load_cleanup(void) |
56e93d26 JQ |
3779 | { |
3780 | int i, thread_count; | |
3781 | ||
3416ab5b JQ |
3782 | if (!migrate_use_compression()) { |
3783 | return; | |
3784 | } | |
56e93d26 JQ |
3785 | thread_count = migrate_decompress_threads(); |
3786 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
3787 | /* |
3788 | * we use it as a indicator which shows if the thread is | |
3789 | * properly init'd or not | |
3790 | */ | |
3791 | if (!decomp_param[i].compbuf) { | |
3792 | break; | |
3793 | } | |
3794 | ||
56e93d26 | 3795 | qemu_mutex_lock(&decomp_param[i].mutex); |
90e56fb4 | 3796 | decomp_param[i].quit = true; |
56e93d26 JQ |
3797 | qemu_cond_signal(&decomp_param[i].cond); |
3798 | qemu_mutex_unlock(&decomp_param[i].mutex); | |
3799 | } | |
3800 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
3801 | if (!decomp_param[i].compbuf) { |
3802 | break; | |
3803 | } | |
3804 | ||
56e93d26 JQ |
3805 | qemu_thread_join(decompress_threads + i); |
3806 | qemu_mutex_destroy(&decomp_param[i].mutex); | |
3807 | qemu_cond_destroy(&decomp_param[i].cond); | |
797ca154 | 3808 | inflateEnd(&decomp_param[i].stream); |
56e93d26 | 3809 | g_free(decomp_param[i].compbuf); |
797ca154 | 3810 | decomp_param[i].compbuf = NULL; |
56e93d26 JQ |
3811 | } |
3812 | g_free(decompress_threads); | |
3813 | g_free(decomp_param); | |
56e93d26 JQ |
3814 | decompress_threads = NULL; |
3815 | decomp_param = NULL; | |
34ab9e97 | 3816 | decomp_file = NULL; |
56e93d26 JQ |
3817 | } |
3818 | ||
34ab9e97 | 3819 | static int compress_threads_load_setup(QEMUFile *f) |
797ca154 XG |
3820 | { |
3821 | int i, thread_count; | |
3822 | ||
3823 | if (!migrate_use_compression()) { | |
3824 | return 0; | |
3825 | } | |
3826 | ||
3827 | thread_count = migrate_decompress_threads(); | |
3828 | decompress_threads = g_new0(QemuThread, thread_count); | |
3829 | decomp_param = g_new0(DecompressParam, thread_count); | |
3830 | qemu_mutex_init(&decomp_done_lock); | |
3831 | qemu_cond_init(&decomp_done_cond); | |
34ab9e97 | 3832 | decomp_file = f; |
797ca154 XG |
3833 | for (i = 0; i < thread_count; i++) { |
3834 | if (inflateInit(&decomp_param[i].stream) != Z_OK) { | |
3835 | goto exit; | |
3836 | } | |
3837 | ||
3838 | decomp_param[i].compbuf = g_malloc0(compressBound(TARGET_PAGE_SIZE)); | |
3839 | qemu_mutex_init(&decomp_param[i].mutex); | |
3840 | qemu_cond_init(&decomp_param[i].cond); | |
3841 | decomp_param[i].done = true; | |
3842 | decomp_param[i].quit = false; | |
3843 | qemu_thread_create(decompress_threads + i, "decompress", | |
3844 | do_data_decompress, decomp_param + i, | |
3845 | QEMU_THREAD_JOINABLE); | |
3846 | } | |
3847 | return 0; | |
3848 | exit: | |
3849 | compress_threads_load_cleanup(); | |
3850 | return -1; | |
3851 | } | |
3852 | ||
c1bc6626 | 3853 | static void decompress_data_with_multi_threads(QEMUFile *f, |
56e93d26 JQ |
3854 | void *host, int len) |
3855 | { | |
3856 | int idx, thread_count; | |
3857 | ||
3858 | thread_count = migrate_decompress_threads(); | |
73a8912b | 3859 | qemu_mutex_lock(&decomp_done_lock); |
56e93d26 JQ |
3860 | while (true) { |
3861 | for (idx = 0; idx < thread_count; idx++) { | |
73a8912b | 3862 | if (decomp_param[idx].done) { |
33d151f4 LL |
3863 | decomp_param[idx].done = false; |
3864 | qemu_mutex_lock(&decomp_param[idx].mutex); | |
c1bc6626 | 3865 | qemu_get_buffer(f, decomp_param[idx].compbuf, len); |
56e93d26 JQ |
3866 | decomp_param[idx].des = host; |
3867 | decomp_param[idx].len = len; | |
33d151f4 LL |
3868 | qemu_cond_signal(&decomp_param[idx].cond); |
3869 | qemu_mutex_unlock(&decomp_param[idx].mutex); | |
56e93d26 JQ |
3870 | break; |
3871 | } | |
3872 | } | |
3873 | if (idx < thread_count) { | |
3874 | break; | |
73a8912b LL |
3875 | } else { |
3876 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
56e93d26 JQ |
3877 | } |
3878 | } | |
73a8912b | 3879 | qemu_mutex_unlock(&decomp_done_lock); |
56e93d26 JQ |
3880 | } |
3881 | ||
13af18f2 ZC |
3882 | /* |
3883 | * colo cache: this is for secondary VM, we cache the whole | |
3884 | * memory of the secondary VM, it is need to hold the global lock | |
3885 | * to call this helper. | |
3886 | */ | |
3887 | int colo_init_ram_cache(void) | |
3888 | { | |
3889 | RAMBlock *block; | |
3890 | ||
3891 | rcu_read_lock(); | |
fbd162e6 | 3892 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
13af18f2 ZC |
3893 | block->colo_cache = qemu_anon_ram_alloc(block->used_length, |
3894 | NULL, | |
3895 | false); | |
3896 | if (!block->colo_cache) { | |
3897 | error_report("%s: Can't alloc memory for COLO cache of block %s," | |
3898 | "size 0x" RAM_ADDR_FMT, __func__, block->idstr, | |
3899 | block->used_length); | |
89ac5a1d DDAG |
3900 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
3901 | if (block->colo_cache) { | |
3902 | qemu_anon_ram_free(block->colo_cache, block->used_length); | |
3903 | block->colo_cache = NULL; | |
3904 | } | |
3905 | } | |
3906 | return -errno; | |
13af18f2 ZC |
3907 | } |
3908 | memcpy(block->colo_cache, block->host, block->used_length); | |
3909 | } | |
3910 | rcu_read_unlock(); | |
7d9acafa ZC |
3911 | /* |
3912 | * Record the dirty pages that sent by PVM, we use this dirty bitmap together | |
3913 | * with to decide which page in cache should be flushed into SVM's RAM. Here | |
3914 | * we use the same name 'ram_bitmap' as for migration. | |
3915 | */ | |
3916 | if (ram_bytes_total()) { | |
3917 | RAMBlock *block; | |
3918 | ||
fbd162e6 | 3919 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
7d9acafa ZC |
3920 | unsigned long pages = block->max_length >> TARGET_PAGE_BITS; |
3921 | ||
3922 | block->bmap = bitmap_new(pages); | |
3923 | bitmap_set(block->bmap, 0, pages); | |
3924 | } | |
3925 | } | |
3926 | ram_state = g_new0(RAMState, 1); | |
3927 | ram_state->migration_dirty_pages = 0; | |
c6e5bafb | 3928 | qemu_mutex_init(&ram_state->bitmap_mutex); |
d1955d22 | 3929 | memory_global_dirty_log_start(); |
7d9acafa | 3930 | |
13af18f2 | 3931 | return 0; |
13af18f2 ZC |
3932 | } |
3933 | ||
3934 | /* It is need to hold the global lock to call this helper */ | |
3935 | void colo_release_ram_cache(void) | |
3936 | { | |
3937 | RAMBlock *block; | |
3938 | ||
d1955d22 | 3939 | memory_global_dirty_log_stop(); |
fbd162e6 | 3940 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
7d9acafa ZC |
3941 | g_free(block->bmap); |
3942 | block->bmap = NULL; | |
3943 | } | |
3944 | ||
89ac5a1d DDAG |
3945 | WITH_RCU_READ_LOCK_GUARD() { |
3946 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3947 | if (block->colo_cache) { | |
3948 | qemu_anon_ram_free(block->colo_cache, block->used_length); | |
3949 | block->colo_cache = NULL; | |
3950 | } | |
13af18f2 ZC |
3951 | } |
3952 | } | |
c6e5bafb | 3953 | qemu_mutex_destroy(&ram_state->bitmap_mutex); |
7d9acafa ZC |
3954 | g_free(ram_state); |
3955 | ram_state = NULL; | |
13af18f2 ZC |
3956 | } |
3957 | ||
f265e0e4 JQ |
3958 | /** |
3959 | * ram_load_setup: Setup RAM for migration incoming side | |
3960 | * | |
3961 | * Returns zero to indicate success and negative for error | |
3962 | * | |
3963 | * @f: QEMUFile where to receive the data | |
3964 | * @opaque: RAMState pointer | |
3965 | */ | |
3966 | static int ram_load_setup(QEMUFile *f, void *opaque) | |
3967 | { | |
34ab9e97 | 3968 | if (compress_threads_load_setup(f)) { |
797ca154 XG |
3969 | return -1; |
3970 | } | |
3971 | ||
f265e0e4 | 3972 | xbzrle_load_setup(); |
f9494614 | 3973 | ramblock_recv_map_init(); |
13af18f2 | 3974 | |
f265e0e4 JQ |
3975 | return 0; |
3976 | } | |
3977 | ||
3978 | static int ram_load_cleanup(void *opaque) | |
3979 | { | |
f9494614 | 3980 | RAMBlock *rb; |
56eb90af | 3981 | |
fbd162e6 | 3982 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
bd108a44 | 3983 | qemu_ram_block_writeback(rb); |
56eb90af JH |
3984 | } |
3985 | ||
f265e0e4 | 3986 | xbzrle_load_cleanup(); |
f0afa331 | 3987 | compress_threads_load_cleanup(); |
f9494614 | 3988 | |
fbd162e6 | 3989 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
f9494614 AP |
3990 | g_free(rb->receivedmap); |
3991 | rb->receivedmap = NULL; | |
3992 | } | |
13af18f2 | 3993 | |
f265e0e4 JQ |
3994 | return 0; |
3995 | } | |
3996 | ||
3d0684b2 JQ |
3997 | /** |
3998 | * ram_postcopy_incoming_init: allocate postcopy data structures | |
3999 | * | |
4000 | * Returns 0 for success and negative if there was one error | |
4001 | * | |
4002 | * @mis: current migration incoming state | |
4003 | * | |
4004 | * Allocate data structures etc needed by incoming migration with | |
4005 | * postcopy-ram. postcopy-ram's similarly names | |
4006 | * postcopy_ram_incoming_init does the work. | |
1caddf8a DDAG |
4007 | */ |
4008 | int ram_postcopy_incoming_init(MigrationIncomingState *mis) | |
4009 | { | |
c136180c | 4010 | return postcopy_ram_incoming_init(mis); |
1caddf8a DDAG |
4011 | } |
4012 | ||
3d0684b2 JQ |
4013 | /** |
4014 | * ram_load_postcopy: load a page in postcopy case | |
4015 | * | |
4016 | * Returns 0 for success or -errno in case of error | |
4017 | * | |
a7180877 DDAG |
4018 | * Called in postcopy mode by ram_load(). |
4019 | * rcu_read_lock is taken prior to this being called. | |
3d0684b2 JQ |
4020 | * |
4021 | * @f: QEMUFile where to send the data | |
a7180877 DDAG |
4022 | */ |
4023 | static int ram_load_postcopy(QEMUFile *f) | |
4024 | { | |
4025 | int flags = 0, ret = 0; | |
4026 | bool place_needed = false; | |
1aa83678 | 4027 | bool matches_target_page_size = false; |
a7180877 DDAG |
4028 | MigrationIncomingState *mis = migration_incoming_get_current(); |
4029 | /* Temporary page that is later 'placed' */ | |
3414322a | 4030 | void *postcopy_host_page = mis->postcopy_tmp_page; |
c53b7ddc | 4031 | void *last_host = NULL; |
a3b6ff6d | 4032 | bool all_zero = false; |
a7180877 DDAG |
4033 | |
4034 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
4035 | ram_addr_t addr; | |
4036 | void *host = NULL; | |
4037 | void *page_buffer = NULL; | |
4038 | void *place_source = NULL; | |
df9ff5e1 | 4039 | RAMBlock *block = NULL; |
a7180877 | 4040 | uint8_t ch; |
a7180877 DDAG |
4041 | |
4042 | addr = qemu_get_be64(f); | |
7a9ddfbf PX |
4043 | |
4044 | /* | |
4045 | * If qemu file error, we should stop here, and then "addr" | |
4046 | * may be invalid | |
4047 | */ | |
4048 | ret = qemu_file_get_error(f); | |
4049 | if (ret) { | |
4050 | break; | |
4051 | } | |
4052 | ||
a7180877 DDAG |
4053 | flags = addr & ~TARGET_PAGE_MASK; |
4054 | addr &= TARGET_PAGE_MASK; | |
4055 | ||
4056 | trace_ram_load_postcopy_loop((uint64_t)addr, flags); | |
4057 | place_needed = false; | |
bb890ed5 | 4058 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE)) { |
df9ff5e1 | 4059 | block = ram_block_from_stream(f, flags); |
4c4bad48 HZ |
4060 | |
4061 | host = host_from_ram_block_offset(block, addr); | |
a7180877 DDAG |
4062 | if (!host) { |
4063 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
4064 | ret = -EINVAL; | |
4065 | break; | |
4066 | } | |
1aa83678 | 4067 | matches_target_page_size = block->page_size == TARGET_PAGE_SIZE; |
a7180877 | 4068 | /* |
28abd200 DDAG |
4069 | * Postcopy requires that we place whole host pages atomically; |
4070 | * these may be huge pages for RAMBlocks that are backed by | |
4071 | * hugetlbfs. | |
a7180877 DDAG |
4072 | * To make it atomic, the data is read into a temporary page |
4073 | * that's moved into place later. | |
4074 | * The migration protocol uses, possibly smaller, target-pages | |
4075 | * however the source ensures it always sends all the components | |
4076 | * of a host page in order. | |
4077 | */ | |
4078 | page_buffer = postcopy_host_page + | |
28abd200 | 4079 | ((uintptr_t)host & (block->page_size - 1)); |
a7180877 | 4080 | /* If all TP are zero then we can optimise the place */ |
28abd200 | 4081 | if (!((uintptr_t)host & (block->page_size - 1))) { |
a7180877 | 4082 | all_zero = true; |
c53b7ddc DDAG |
4083 | } else { |
4084 | /* not the 1st TP within the HP */ | |
4085 | if (host != (last_host + TARGET_PAGE_SIZE)) { | |
9af9e0fe | 4086 | error_report("Non-sequential target page %p/%p", |
c53b7ddc DDAG |
4087 | host, last_host); |
4088 | ret = -EINVAL; | |
4089 | break; | |
4090 | } | |
a7180877 DDAG |
4091 | } |
4092 | ||
c53b7ddc | 4093 | |
a7180877 DDAG |
4094 | /* |
4095 | * If it's the last part of a host page then we place the host | |
4096 | * page | |
4097 | */ | |
4098 | place_needed = (((uintptr_t)host + TARGET_PAGE_SIZE) & | |
28abd200 | 4099 | (block->page_size - 1)) == 0; |
a7180877 DDAG |
4100 | place_source = postcopy_host_page; |
4101 | } | |
c53b7ddc | 4102 | last_host = host; |
a7180877 DDAG |
4103 | |
4104 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { | |
bb890ed5 | 4105 | case RAM_SAVE_FLAG_ZERO: |
a7180877 DDAG |
4106 | ch = qemu_get_byte(f); |
4107 | memset(page_buffer, ch, TARGET_PAGE_SIZE); | |
4108 | if (ch) { | |
4109 | all_zero = false; | |
4110 | } | |
4111 | break; | |
4112 | ||
4113 | case RAM_SAVE_FLAG_PAGE: | |
4114 | all_zero = false; | |
1aa83678 PX |
4115 | if (!matches_target_page_size) { |
4116 | /* For huge pages, we always use temporary buffer */ | |
a7180877 DDAG |
4117 | qemu_get_buffer(f, page_buffer, TARGET_PAGE_SIZE); |
4118 | } else { | |
1aa83678 PX |
4119 | /* |
4120 | * For small pages that matches target page size, we | |
4121 | * avoid the qemu_file copy. Instead we directly use | |
4122 | * the buffer of QEMUFile to place the page. Note: we | |
4123 | * cannot do any QEMUFile operation before using that | |
4124 | * buffer to make sure the buffer is valid when | |
4125 | * placing the page. | |
a7180877 DDAG |
4126 | */ |
4127 | qemu_get_buffer_in_place(f, (uint8_t **)&place_source, | |
4128 | TARGET_PAGE_SIZE); | |
4129 | } | |
4130 | break; | |
4131 | case RAM_SAVE_FLAG_EOS: | |
4132 | /* normal exit */ | |
6df264ac | 4133 | multifd_recv_sync_main(); |
a7180877 DDAG |
4134 | break; |
4135 | default: | |
4136 | error_report("Unknown combination of migration flags: %#x" | |
4137 | " (postcopy mode)", flags); | |
4138 | ret = -EINVAL; | |
7a9ddfbf PX |
4139 | break; |
4140 | } | |
4141 | ||
4142 | /* Detect for any possible file errors */ | |
4143 | if (!ret && qemu_file_get_error(f)) { | |
4144 | ret = qemu_file_get_error(f); | |
a7180877 DDAG |
4145 | } |
4146 | ||
7a9ddfbf | 4147 | if (!ret && place_needed) { |
a7180877 | 4148 | /* This gets called at the last target page in the host page */ |
df9ff5e1 DDAG |
4149 | void *place_dest = host + TARGET_PAGE_SIZE - block->page_size; |
4150 | ||
a7180877 | 4151 | if (all_zero) { |
df9ff5e1 | 4152 | ret = postcopy_place_page_zero(mis, place_dest, |
8be4620b | 4153 | block); |
a7180877 | 4154 | } else { |
df9ff5e1 | 4155 | ret = postcopy_place_page(mis, place_dest, |
8be4620b | 4156 | place_source, block); |
a7180877 DDAG |
4157 | } |
4158 | } | |
a7180877 DDAG |
4159 | } |
4160 | ||
4161 | return ret; | |
4162 | } | |
4163 | ||
acab30b8 DHB |
4164 | static bool postcopy_is_advised(void) |
4165 | { | |
4166 | PostcopyState ps = postcopy_state_get(); | |
4167 | return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END; | |
4168 | } | |
4169 | ||
4170 | static bool postcopy_is_running(void) | |
4171 | { | |
4172 | PostcopyState ps = postcopy_state_get(); | |
4173 | return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; | |
4174 | } | |
4175 | ||
e6f4aa18 ZC |
4176 | /* |
4177 | * Flush content of RAM cache into SVM's memory. | |
4178 | * Only flush the pages that be dirtied by PVM or SVM or both. | |
4179 | */ | |
4180 | static void colo_flush_ram_cache(void) | |
4181 | { | |
4182 | RAMBlock *block = NULL; | |
4183 | void *dst_host; | |
4184 | void *src_host; | |
4185 | unsigned long offset = 0; | |
4186 | ||
d1955d22 | 4187 | memory_global_dirty_log_sync(); |
89ac5a1d DDAG |
4188 | WITH_RCU_READ_LOCK_GUARD() { |
4189 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
4190 | ramblock_sync_dirty_bitmap(ram_state, block); | |
4191 | } | |
d1955d22 | 4192 | } |
d1955d22 | 4193 | |
e6f4aa18 | 4194 | trace_colo_flush_ram_cache_begin(ram_state->migration_dirty_pages); |
89ac5a1d DDAG |
4195 | WITH_RCU_READ_LOCK_GUARD() { |
4196 | block = QLIST_FIRST_RCU(&ram_list.blocks); | |
e6f4aa18 | 4197 | |
89ac5a1d DDAG |
4198 | while (block) { |
4199 | offset = migration_bitmap_find_dirty(ram_state, block, offset); | |
e6f4aa18 | 4200 | |
89ac5a1d DDAG |
4201 | if (offset << TARGET_PAGE_BITS >= block->used_length) { |
4202 | offset = 0; | |
4203 | block = QLIST_NEXT_RCU(block, next); | |
4204 | } else { | |
4205 | migration_bitmap_clear_dirty(ram_state, block, offset); | |
4206 | dst_host = block->host + (offset << TARGET_PAGE_BITS); | |
4207 | src_host = block->colo_cache + (offset << TARGET_PAGE_BITS); | |
4208 | memcpy(dst_host, src_host, TARGET_PAGE_SIZE); | |
4209 | } | |
e6f4aa18 ZC |
4210 | } |
4211 | } | |
e6f4aa18 ZC |
4212 | trace_colo_flush_ram_cache_end(); |
4213 | } | |
4214 | ||
10da4a36 WY |
4215 | /** |
4216 | * ram_load_precopy: load pages in precopy case | |
4217 | * | |
4218 | * Returns 0 for success or -errno in case of error | |
4219 | * | |
4220 | * Called in precopy mode by ram_load(). | |
4221 | * rcu_read_lock is taken prior to this being called. | |
4222 | * | |
4223 | * @f: QEMUFile where to send the data | |
4224 | */ | |
4225 | static int ram_load_precopy(QEMUFile *f) | |
56e93d26 | 4226 | { |
10da4a36 | 4227 | int flags = 0, ret = 0, invalid_flags = 0, len = 0; |
ef08fb38 | 4228 | /* ADVISE is earlier, it shows the source has the postcopy capability on */ |
acab30b8 | 4229 | bool postcopy_advised = postcopy_is_advised(); |
edc60127 JQ |
4230 | if (!migrate_use_compression()) { |
4231 | invalid_flags |= RAM_SAVE_FLAG_COMPRESS_PAGE; | |
4232 | } | |
a7180877 | 4233 | |
10da4a36 | 4234 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { |
56e93d26 | 4235 | ram_addr_t addr, total_ram_bytes; |
a776aa15 | 4236 | void *host = NULL; |
56e93d26 JQ |
4237 | uint8_t ch; |
4238 | ||
4239 | addr = qemu_get_be64(f); | |
4240 | flags = addr & ~TARGET_PAGE_MASK; | |
4241 | addr &= TARGET_PAGE_MASK; | |
4242 | ||
edc60127 JQ |
4243 | if (flags & invalid_flags) { |
4244 | if (flags & invalid_flags & RAM_SAVE_FLAG_COMPRESS_PAGE) { | |
4245 | error_report("Received an unexpected compressed page"); | |
4246 | } | |
4247 | ||
4248 | ret = -EINVAL; | |
4249 | break; | |
4250 | } | |
4251 | ||
bb890ed5 | 4252 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
a776aa15 | 4253 | RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) { |
4c4bad48 HZ |
4254 | RAMBlock *block = ram_block_from_stream(f, flags); |
4255 | ||
13af18f2 ZC |
4256 | /* |
4257 | * After going into COLO, we should load the Page into colo_cache. | |
4258 | */ | |
4259 | if (migration_incoming_in_colo_state()) { | |
4260 | host = colo_cache_from_block_offset(block, addr); | |
4261 | } else { | |
4262 | host = host_from_ram_block_offset(block, addr); | |
4263 | } | |
a776aa15 DDAG |
4264 | if (!host) { |
4265 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
4266 | ret = -EINVAL; | |
4267 | break; | |
4268 | } | |
13af18f2 ZC |
4269 | |
4270 | if (!migration_incoming_in_colo_state()) { | |
4271 | ramblock_recv_bitmap_set(block, host); | |
4272 | } | |
4273 | ||
1db9d8e5 | 4274 | trace_ram_load_loop(block->idstr, (uint64_t)addr, flags, host); |
a776aa15 DDAG |
4275 | } |
4276 | ||
56e93d26 JQ |
4277 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { |
4278 | case RAM_SAVE_FLAG_MEM_SIZE: | |
4279 | /* Synchronize RAM block list */ | |
4280 | total_ram_bytes = addr; | |
4281 | while (!ret && total_ram_bytes) { | |
4282 | RAMBlock *block; | |
56e93d26 JQ |
4283 | char id[256]; |
4284 | ram_addr_t length; | |
4285 | ||
4286 | len = qemu_get_byte(f); | |
4287 | qemu_get_buffer(f, (uint8_t *)id, len); | |
4288 | id[len] = 0; | |
4289 | length = qemu_get_be64(f); | |
4290 | ||
e3dd7493 | 4291 | block = qemu_ram_block_by_name(id); |
b895de50 CLG |
4292 | if (block && !qemu_ram_is_migratable(block)) { |
4293 | error_report("block %s should not be migrated !", id); | |
4294 | ret = -EINVAL; | |
4295 | } else if (block) { | |
e3dd7493 DDAG |
4296 | if (length != block->used_length) { |
4297 | Error *local_err = NULL; | |
56e93d26 | 4298 | |
fa53a0e5 | 4299 | ret = qemu_ram_resize(block, length, |
e3dd7493 DDAG |
4300 | &local_err); |
4301 | if (local_err) { | |
4302 | error_report_err(local_err); | |
56e93d26 | 4303 | } |
56e93d26 | 4304 | } |
ef08fb38 DDAG |
4305 | /* For postcopy we need to check hugepage sizes match */ |
4306 | if (postcopy_advised && | |
4307 | block->page_size != qemu_host_page_size) { | |
4308 | uint64_t remote_page_size = qemu_get_be64(f); | |
4309 | if (remote_page_size != block->page_size) { | |
4310 | error_report("Mismatched RAM page size %s " | |
4311 | "(local) %zd != %" PRId64, | |
4312 | id, block->page_size, | |
4313 | remote_page_size); | |
4314 | ret = -EINVAL; | |
4315 | } | |
4316 | } | |
fbd162e6 YK |
4317 | if (migrate_ignore_shared()) { |
4318 | hwaddr addr = qemu_get_be64(f); | |
fbd162e6 YK |
4319 | if (ramblock_is_ignored(block) && |
4320 | block->mr->addr != addr) { | |
4321 | error_report("Mismatched GPAs for block %s " | |
4322 | "%" PRId64 "!= %" PRId64, | |
4323 | id, (uint64_t)addr, | |
4324 | (uint64_t)block->mr->addr); | |
4325 | ret = -EINVAL; | |
4326 | } | |
4327 | } | |
e3dd7493 DDAG |
4328 | ram_control_load_hook(f, RAM_CONTROL_BLOCK_REG, |
4329 | block->idstr); | |
4330 | } else { | |
56e93d26 JQ |
4331 | error_report("Unknown ramblock \"%s\", cannot " |
4332 | "accept migration", id); | |
4333 | ret = -EINVAL; | |
4334 | } | |
4335 | ||
4336 | total_ram_bytes -= length; | |
4337 | } | |
4338 | break; | |
a776aa15 | 4339 | |
bb890ed5 | 4340 | case RAM_SAVE_FLAG_ZERO: |
56e93d26 JQ |
4341 | ch = qemu_get_byte(f); |
4342 | ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); | |
4343 | break; | |
a776aa15 | 4344 | |
56e93d26 | 4345 | case RAM_SAVE_FLAG_PAGE: |
56e93d26 JQ |
4346 | qemu_get_buffer(f, host, TARGET_PAGE_SIZE); |
4347 | break; | |
56e93d26 | 4348 | |
a776aa15 | 4349 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
56e93d26 JQ |
4350 | len = qemu_get_be32(f); |
4351 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
4352 | error_report("Invalid compressed data length: %d", len); | |
4353 | ret = -EINVAL; | |
4354 | break; | |
4355 | } | |
c1bc6626 | 4356 | decompress_data_with_multi_threads(f, host, len); |
56e93d26 | 4357 | break; |
a776aa15 | 4358 | |
56e93d26 | 4359 | case RAM_SAVE_FLAG_XBZRLE: |
56e93d26 JQ |
4360 | if (load_xbzrle(f, addr, host) < 0) { |
4361 | error_report("Failed to decompress XBZRLE page at " | |
4362 | RAM_ADDR_FMT, addr); | |
4363 | ret = -EINVAL; | |
4364 | break; | |
4365 | } | |
4366 | break; | |
4367 | case RAM_SAVE_FLAG_EOS: | |
4368 | /* normal exit */ | |
6df264ac | 4369 | multifd_recv_sync_main(); |
56e93d26 JQ |
4370 | break; |
4371 | default: | |
4372 | if (flags & RAM_SAVE_FLAG_HOOK) { | |
632e3a5c | 4373 | ram_control_load_hook(f, RAM_CONTROL_HOOK, NULL); |
56e93d26 JQ |
4374 | } else { |
4375 | error_report("Unknown combination of migration flags: %#x", | |
4376 | flags); | |
4377 | ret = -EINVAL; | |
4378 | } | |
4379 | } | |
4380 | if (!ret) { | |
4381 | ret = qemu_file_get_error(f); | |
4382 | } | |
4383 | } | |
4384 | ||
10da4a36 WY |
4385 | return ret; |
4386 | } | |
4387 | ||
4388 | static int ram_load(QEMUFile *f, void *opaque, int version_id) | |
4389 | { | |
4390 | int ret = 0; | |
4391 | static uint64_t seq_iter; | |
4392 | /* | |
4393 | * If system is running in postcopy mode, page inserts to host memory must | |
4394 | * be atomic | |
4395 | */ | |
4396 | bool postcopy_running = postcopy_is_running(); | |
4397 | ||
4398 | seq_iter++; | |
4399 | ||
4400 | if (version_id != 4) { | |
4401 | return -EINVAL; | |
4402 | } | |
4403 | ||
4404 | /* | |
4405 | * This RCU critical section can be very long running. | |
4406 | * When RCU reclaims in the code start to become numerous, | |
4407 | * it will be necessary to reduce the granularity of this | |
4408 | * critical section. | |
4409 | */ | |
89ac5a1d DDAG |
4410 | WITH_RCU_READ_LOCK_GUARD() { |
4411 | if (postcopy_running) { | |
4412 | ret = ram_load_postcopy(f); | |
4413 | } else { | |
4414 | ret = ram_load_precopy(f); | |
4415 | } | |
10da4a36 | 4416 | |
89ac5a1d | 4417 | ret |= wait_for_decompress_done(); |
10da4a36 | 4418 | } |
55c4446b | 4419 | trace_ram_load_complete(ret, seq_iter); |
e6f4aa18 ZC |
4420 | |
4421 | if (!ret && migration_incoming_in_colo_state()) { | |
4422 | colo_flush_ram_cache(); | |
4423 | } | |
56e93d26 JQ |
4424 | return ret; |
4425 | } | |
4426 | ||
c6467627 VSO |
4427 | static bool ram_has_postcopy(void *opaque) |
4428 | { | |
469dd51b | 4429 | RAMBlock *rb; |
fbd162e6 | 4430 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
469dd51b JH |
4431 | if (ramblock_is_pmem(rb)) { |
4432 | info_report("Block: %s, host: %p is a nvdimm memory, postcopy" | |
4433 | "is not supported now!", rb->idstr, rb->host); | |
4434 | return false; | |
4435 | } | |
4436 | } | |
4437 | ||
c6467627 VSO |
4438 | return migrate_postcopy_ram(); |
4439 | } | |
4440 | ||
edd090c7 PX |
4441 | /* Sync all the dirty bitmap with destination VM. */ |
4442 | static int ram_dirty_bitmap_sync_all(MigrationState *s, RAMState *rs) | |
4443 | { | |
4444 | RAMBlock *block; | |
4445 | QEMUFile *file = s->to_dst_file; | |
4446 | int ramblock_count = 0; | |
4447 | ||
4448 | trace_ram_dirty_bitmap_sync_start(); | |
4449 | ||
fbd162e6 | 4450 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
edd090c7 PX |
4451 | qemu_savevm_send_recv_bitmap(file, block->idstr); |
4452 | trace_ram_dirty_bitmap_request(block->idstr); | |
4453 | ramblock_count++; | |
4454 | } | |
4455 | ||
4456 | trace_ram_dirty_bitmap_sync_wait(); | |
4457 | ||
4458 | /* Wait until all the ramblocks' dirty bitmap synced */ | |
4459 | while (ramblock_count--) { | |
4460 | qemu_sem_wait(&s->rp_state.rp_sem); | |
4461 | } | |
4462 | ||
4463 | trace_ram_dirty_bitmap_sync_complete(); | |
4464 | ||
4465 | return 0; | |
4466 | } | |
4467 | ||
4468 | static void ram_dirty_bitmap_reload_notify(MigrationState *s) | |
4469 | { | |
4470 | qemu_sem_post(&s->rp_state.rp_sem); | |
4471 | } | |
4472 | ||
a335debb PX |
4473 | /* |
4474 | * Read the received bitmap, revert it as the initial dirty bitmap. | |
4475 | * This is only used when the postcopy migration is paused but wants | |
4476 | * to resume from a middle point. | |
4477 | */ | |
4478 | int ram_dirty_bitmap_reload(MigrationState *s, RAMBlock *block) | |
4479 | { | |
4480 | int ret = -EINVAL; | |
4481 | QEMUFile *file = s->rp_state.from_dst_file; | |
4482 | unsigned long *le_bitmap, nbits = block->used_length >> TARGET_PAGE_BITS; | |
a725ef9f | 4483 | uint64_t local_size = DIV_ROUND_UP(nbits, 8); |
a335debb PX |
4484 | uint64_t size, end_mark; |
4485 | ||
4486 | trace_ram_dirty_bitmap_reload_begin(block->idstr); | |
4487 | ||
4488 | if (s->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
4489 | error_report("%s: incorrect state %s", __func__, | |
4490 | MigrationStatus_str(s->state)); | |
4491 | return -EINVAL; | |
4492 | } | |
4493 | ||
4494 | /* | |
4495 | * Note: see comments in ramblock_recv_bitmap_send() on why we | |
4496 | * need the endianess convertion, and the paddings. | |
4497 | */ | |
4498 | local_size = ROUND_UP(local_size, 8); | |
4499 | ||
4500 | /* Add paddings */ | |
4501 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
4502 | ||
4503 | size = qemu_get_be64(file); | |
4504 | ||
4505 | /* The size of the bitmap should match with our ramblock */ | |
4506 | if (size != local_size) { | |
4507 | error_report("%s: ramblock '%s' bitmap size mismatch " | |
4508 | "(0x%"PRIx64" != 0x%"PRIx64")", __func__, | |
4509 | block->idstr, size, local_size); | |
4510 | ret = -EINVAL; | |
4511 | goto out; | |
4512 | } | |
4513 | ||
4514 | size = qemu_get_buffer(file, (uint8_t *)le_bitmap, local_size); | |
4515 | end_mark = qemu_get_be64(file); | |
4516 | ||
4517 | ret = qemu_file_get_error(file); | |
4518 | if (ret || size != local_size) { | |
4519 | error_report("%s: read bitmap failed for ramblock '%s': %d" | |
4520 | " (size 0x%"PRIx64", got: 0x%"PRIx64")", | |
4521 | __func__, block->idstr, ret, local_size, size); | |
4522 | ret = -EIO; | |
4523 | goto out; | |
4524 | } | |
4525 | ||
4526 | if (end_mark != RAMBLOCK_RECV_BITMAP_ENDING) { | |
4527 | error_report("%s: ramblock '%s' end mark incorrect: 0x%"PRIu64, | |
4528 | __func__, block->idstr, end_mark); | |
4529 | ret = -EINVAL; | |
4530 | goto out; | |
4531 | } | |
4532 | ||
4533 | /* | |
4534 | * Endianess convertion. We are during postcopy (though paused). | |
4535 | * The dirty bitmap won't change. We can directly modify it. | |
4536 | */ | |
4537 | bitmap_from_le(block->bmap, le_bitmap, nbits); | |
4538 | ||
4539 | /* | |
4540 | * What we received is "received bitmap". Revert it as the initial | |
4541 | * dirty bitmap for this ramblock. | |
4542 | */ | |
4543 | bitmap_complement(block->bmap, block->bmap, nbits); | |
4544 | ||
4545 | trace_ram_dirty_bitmap_reload_complete(block->idstr); | |
4546 | ||
edd090c7 PX |
4547 | /* |
4548 | * We succeeded to sync bitmap for current ramblock. If this is | |
4549 | * the last one to sync, we need to notify the main send thread. | |
4550 | */ | |
4551 | ram_dirty_bitmap_reload_notify(s); | |
4552 | ||
a335debb PX |
4553 | ret = 0; |
4554 | out: | |
bf269906 | 4555 | g_free(le_bitmap); |
a335debb PX |
4556 | return ret; |
4557 | } | |
4558 | ||
edd090c7 PX |
4559 | static int ram_resume_prepare(MigrationState *s, void *opaque) |
4560 | { | |
4561 | RAMState *rs = *(RAMState **)opaque; | |
08614f34 | 4562 | int ret; |
edd090c7 | 4563 | |
08614f34 PX |
4564 | ret = ram_dirty_bitmap_sync_all(s, rs); |
4565 | if (ret) { | |
4566 | return ret; | |
4567 | } | |
4568 | ||
4569 | ram_state_resume_prepare(rs, s->to_dst_file); | |
4570 | ||
4571 | return 0; | |
edd090c7 PX |
4572 | } |
4573 | ||
56e93d26 | 4574 | static SaveVMHandlers savevm_ram_handlers = { |
9907e842 | 4575 | .save_setup = ram_save_setup, |
56e93d26 | 4576 | .save_live_iterate = ram_save_iterate, |
763c906b | 4577 | .save_live_complete_postcopy = ram_save_complete, |
a3e06c3d | 4578 | .save_live_complete_precopy = ram_save_complete, |
c6467627 | 4579 | .has_postcopy = ram_has_postcopy, |
56e93d26 JQ |
4580 | .save_live_pending = ram_save_pending, |
4581 | .load_state = ram_load, | |
f265e0e4 JQ |
4582 | .save_cleanup = ram_save_cleanup, |
4583 | .load_setup = ram_load_setup, | |
4584 | .load_cleanup = ram_load_cleanup, | |
edd090c7 | 4585 | .resume_prepare = ram_resume_prepare, |
56e93d26 JQ |
4586 | }; |
4587 | ||
4588 | void ram_mig_init(void) | |
4589 | { | |
4590 | qemu_mutex_init(&XBZRLE.lock); | |
ce62df53 | 4591 | register_savevm_live("ram", 0, 4, &savevm_ram_handlers, &ram_state); |
56e93d26 | 4592 | } |