]>
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" |
f348b6d1 | 30 | #include "qemu/cutils.h" |
56e93d26 JQ |
31 | #include "qemu/bitops.h" |
32 | #include "qemu/bitmap.h" | |
7205c9ec | 33 | #include "qemu/main-loop.h" |
709e3fe8 | 34 | #include "xbzrle.h" |
7b1e1a22 | 35 | #include "ram.h" |
6666c96a | 36 | #include "migration.h" |
f2a8f0a6 | 37 | #include "migration/register.h" |
7b1e1a22 | 38 | #include "migration/misc.h" |
08a0aee1 | 39 | #include "qemu-file.h" |
be07b0ac | 40 | #include "postcopy-ram.h" |
53d37d36 | 41 | #include "page_cache.h" |
56e93d26 | 42 | #include "qemu/error-report.h" |
e688df6b | 43 | #include "qapi/error.h" |
ab7cbb0b | 44 | #include "qapi/qapi-types-migration.h" |
9af23989 | 45 | #include "qapi/qapi-events-migration.h" |
8acabf69 | 46 | #include "qapi/qmp/qerror.h" |
56e93d26 | 47 | #include "trace.h" |
56e93d26 | 48 | #include "exec/ram_addr.h" |
f9494614 | 49 | #include "exec/target_page.h" |
56e93d26 | 50 | #include "qemu/rcu_queue.h" |
a91246c9 | 51 | #include "migration/colo.h" |
53d37d36 | 52 | #include "block.h" |
b0c3cf94 | 53 | #include "sysemu/cpu-throttle.h" |
edd090c7 | 54 | #include "savevm.h" |
b9ee2f7d | 55 | #include "qemu/iov.h" |
d32ca5ad | 56 | #include "multifd.h" |
278e2f55 AG |
57 | #include "sysemu/runstate.h" |
58 | ||
59 | #if defined(__linux__) | |
60 | #include "qemu/userfaultfd.h" | |
61 | #endif /* defined(__linux__) */ | |
56e93d26 | 62 | |
56e93d26 JQ |
63 | /***********************************************************/ |
64 | /* ram save/restore */ | |
65 | ||
bb890ed5 JQ |
66 | /* RAM_SAVE_FLAG_ZERO used to be named RAM_SAVE_FLAG_COMPRESS, it |
67 | * worked for pages that where filled with the same char. We switched | |
68 | * it to only search for the zero value. And to avoid confusion with | |
69 | * RAM_SSAVE_FLAG_COMPRESS_PAGE just rename it. | |
70 | */ | |
71 | ||
56e93d26 | 72 | #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */ |
bb890ed5 | 73 | #define RAM_SAVE_FLAG_ZERO 0x02 |
56e93d26 JQ |
74 | #define RAM_SAVE_FLAG_MEM_SIZE 0x04 |
75 | #define RAM_SAVE_FLAG_PAGE 0x08 | |
76 | #define RAM_SAVE_FLAG_EOS 0x10 | |
77 | #define RAM_SAVE_FLAG_CONTINUE 0x20 | |
78 | #define RAM_SAVE_FLAG_XBZRLE 0x40 | |
79 | /* 0x80 is reserved in migration.h start with 0x100 next */ | |
80 | #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 | |
81 | ||
56e93d26 JQ |
82 | static inline bool is_zero_range(uint8_t *p, uint64_t size) |
83 | { | |
a1febc49 | 84 | return buffer_is_zero(p, size); |
56e93d26 JQ |
85 | } |
86 | ||
9360447d JQ |
87 | XBZRLECacheStats xbzrle_counters; |
88 | ||
56e93d26 JQ |
89 | /* struct contains XBZRLE cache and a static page |
90 | used by the compression */ | |
91 | static struct { | |
92 | /* buffer used for XBZRLE encoding */ | |
93 | uint8_t *encoded_buf; | |
94 | /* buffer for storing page content */ | |
95 | uint8_t *current_buf; | |
96 | /* Cache for XBZRLE, Protected by lock. */ | |
97 | PageCache *cache; | |
98 | QemuMutex lock; | |
c00e0928 JQ |
99 | /* it will store a page full of zeros */ |
100 | uint8_t *zero_target_page; | |
f265e0e4 JQ |
101 | /* buffer used for XBZRLE decoding */ |
102 | uint8_t *decoded_buf; | |
56e93d26 JQ |
103 | } XBZRLE; |
104 | ||
56e93d26 JQ |
105 | static void XBZRLE_cache_lock(void) |
106 | { | |
f4c51a6b | 107 | if (migrate_use_xbzrle()) { |
56e93d26 | 108 | qemu_mutex_lock(&XBZRLE.lock); |
f4c51a6b | 109 | } |
56e93d26 JQ |
110 | } |
111 | ||
112 | static void XBZRLE_cache_unlock(void) | |
113 | { | |
f4c51a6b | 114 | if (migrate_use_xbzrle()) { |
56e93d26 | 115 | qemu_mutex_unlock(&XBZRLE.lock); |
f4c51a6b | 116 | } |
56e93d26 JQ |
117 | } |
118 | ||
3d0684b2 JQ |
119 | /** |
120 | * xbzrle_cache_resize: resize the xbzrle cache | |
121 | * | |
cbde7be9 | 122 | * This function is called from migrate_params_apply in main |
3d0684b2 JQ |
123 | * thread, possibly while a migration is in progress. A running |
124 | * migration may be using the cache and might finish during this call, | |
125 | * hence changes to the cache are protected by XBZRLE.lock(). | |
126 | * | |
c9dede2d | 127 | * Returns 0 for success or -1 for error |
3d0684b2 JQ |
128 | * |
129 | * @new_size: new cache size | |
8acabf69 | 130 | * @errp: set *errp if the check failed, with reason |
56e93d26 | 131 | */ |
8b9407a0 | 132 | int xbzrle_cache_resize(uint64_t new_size, Error **errp) |
56e93d26 JQ |
133 | { |
134 | PageCache *new_cache; | |
c9dede2d | 135 | int64_t ret = 0; |
56e93d26 | 136 | |
8acabf69 JQ |
137 | /* Check for truncation */ |
138 | if (new_size != (size_t)new_size) { | |
139 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", | |
140 | "exceeding address space"); | |
141 | return -1; | |
142 | } | |
143 | ||
2a313e5c JQ |
144 | if (new_size == migrate_xbzrle_cache_size()) { |
145 | /* nothing to do */ | |
c9dede2d | 146 | return 0; |
2a313e5c JQ |
147 | } |
148 | ||
56e93d26 JQ |
149 | XBZRLE_cache_lock(); |
150 | ||
151 | if (XBZRLE.cache != NULL) { | |
80f8dfde | 152 | new_cache = cache_init(new_size, TARGET_PAGE_SIZE, errp); |
56e93d26 | 153 | if (!new_cache) { |
56e93d26 JQ |
154 | ret = -1; |
155 | goto out; | |
156 | } | |
157 | ||
158 | cache_fini(XBZRLE.cache); | |
159 | XBZRLE.cache = new_cache; | |
160 | } | |
56e93d26 JQ |
161 | out: |
162 | XBZRLE_cache_unlock(); | |
163 | return ret; | |
164 | } | |
165 | ||
3ded54b1 | 166 | bool ramblock_is_ignored(RAMBlock *block) |
fbd162e6 YK |
167 | { |
168 | return !qemu_ram_is_migratable(block) || | |
169 | (migrate_ignore_shared() && qemu_ram_is_shared(block)); | |
170 | } | |
171 | ||
343f632c DDAG |
172 | #undef RAMBLOCK_FOREACH |
173 | ||
fbd162e6 YK |
174 | int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque) |
175 | { | |
176 | RAMBlock *block; | |
177 | int ret = 0; | |
178 | ||
89ac5a1d DDAG |
179 | RCU_READ_LOCK_GUARD(); |
180 | ||
fbd162e6 YK |
181 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
182 | ret = func(block, opaque); | |
183 | if (ret) { | |
184 | break; | |
185 | } | |
186 | } | |
fbd162e6 YK |
187 | return ret; |
188 | } | |
189 | ||
f9494614 AP |
190 | static void ramblock_recv_map_init(void) |
191 | { | |
192 | RAMBlock *rb; | |
193 | ||
fbd162e6 | 194 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
f9494614 AP |
195 | assert(!rb->receivedmap); |
196 | rb->receivedmap = bitmap_new(rb->max_length >> qemu_target_page_bits()); | |
197 | } | |
198 | } | |
199 | ||
200 | int ramblock_recv_bitmap_test(RAMBlock *rb, void *host_addr) | |
201 | { | |
202 | return test_bit(ramblock_recv_bitmap_offset(host_addr, rb), | |
203 | rb->receivedmap); | |
204 | } | |
205 | ||
1cba9f6e DDAG |
206 | bool ramblock_recv_bitmap_test_byte_offset(RAMBlock *rb, uint64_t byte_offset) |
207 | { | |
208 | return test_bit(byte_offset >> TARGET_PAGE_BITS, rb->receivedmap); | |
209 | } | |
210 | ||
f9494614 AP |
211 | void ramblock_recv_bitmap_set(RAMBlock *rb, void *host_addr) |
212 | { | |
213 | set_bit_atomic(ramblock_recv_bitmap_offset(host_addr, rb), rb->receivedmap); | |
214 | } | |
215 | ||
216 | void ramblock_recv_bitmap_set_range(RAMBlock *rb, void *host_addr, | |
217 | size_t nr) | |
218 | { | |
219 | bitmap_set_atomic(rb->receivedmap, | |
220 | ramblock_recv_bitmap_offset(host_addr, rb), | |
221 | nr); | |
222 | } | |
223 | ||
a335debb PX |
224 | #define RAMBLOCK_RECV_BITMAP_ENDING (0x0123456789abcdefULL) |
225 | ||
226 | /* | |
227 | * Format: bitmap_size (8 bytes) + whole_bitmap (N bytes). | |
228 | * | |
229 | * Returns >0 if success with sent bytes, or <0 if error. | |
230 | */ | |
231 | int64_t ramblock_recv_bitmap_send(QEMUFile *file, | |
232 | const char *block_name) | |
233 | { | |
234 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
235 | unsigned long *le_bitmap, nbits; | |
236 | uint64_t size; | |
237 | ||
238 | if (!block) { | |
239 | error_report("%s: invalid block name: %s", __func__, block_name); | |
240 | return -1; | |
241 | } | |
242 | ||
243 | nbits = block->used_length >> TARGET_PAGE_BITS; | |
244 | ||
245 | /* | |
246 | * Make sure the tmp bitmap buffer is big enough, e.g., on 32bit | |
247 | * machines we may need 4 more bytes for padding (see below | |
248 | * comment). So extend it a bit before hand. | |
249 | */ | |
250 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
251 | ||
252 | /* | |
253 | * Always use little endian when sending the bitmap. This is | |
254 | * required that when source and destination VMs are not using the | |
3a4452d8 | 255 | * same endianness. (Note: big endian won't work.) |
a335debb PX |
256 | */ |
257 | bitmap_to_le(le_bitmap, block->receivedmap, nbits); | |
258 | ||
259 | /* Size of the bitmap, in bytes */ | |
a725ef9f | 260 | size = DIV_ROUND_UP(nbits, 8); |
a335debb PX |
261 | |
262 | /* | |
263 | * size is always aligned to 8 bytes for 64bit machines, but it | |
264 | * may not be true for 32bit machines. We need this padding to | |
265 | * make sure the migration can survive even between 32bit and | |
266 | * 64bit machines. | |
267 | */ | |
268 | size = ROUND_UP(size, 8); | |
269 | ||
270 | qemu_put_be64(file, size); | |
271 | qemu_put_buffer(file, (const uint8_t *)le_bitmap, size); | |
272 | /* | |
273 | * Mark as an end, in case the middle part is screwed up due to | |
3a4452d8 | 274 | * some "mysterious" reason. |
a335debb PX |
275 | */ |
276 | qemu_put_be64(file, RAMBLOCK_RECV_BITMAP_ENDING); | |
277 | qemu_fflush(file); | |
278 | ||
bf269906 | 279 | g_free(le_bitmap); |
a335debb PX |
280 | |
281 | if (qemu_file_get_error(file)) { | |
282 | return qemu_file_get_error(file); | |
283 | } | |
284 | ||
285 | return size + sizeof(size); | |
286 | } | |
287 | ||
ec481c6c JQ |
288 | /* |
289 | * An outstanding page request, on the source, having been received | |
290 | * and queued | |
291 | */ | |
292 | struct RAMSrcPageRequest { | |
293 | RAMBlock *rb; | |
294 | hwaddr offset; | |
295 | hwaddr len; | |
296 | ||
297 | QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; | |
298 | }; | |
299 | ||
6f37bb8b JQ |
300 | /* State of RAM for migration */ |
301 | struct RAMState { | |
204b88b8 JQ |
302 | /* QEMUFile used for this migration */ |
303 | QEMUFile *f; | |
278e2f55 AG |
304 | /* UFFD file descriptor, used in 'write-tracking' migration */ |
305 | int uffdio_fd; | |
6f37bb8b JQ |
306 | /* Last block that we have visited searching for dirty pages */ |
307 | RAMBlock *last_seen_block; | |
308 | /* Last block from where we have sent data */ | |
309 | RAMBlock *last_sent_block; | |
269ace29 JQ |
310 | /* Last dirty target page we have sent */ |
311 | ram_addr_t last_page; | |
6f37bb8b JQ |
312 | /* last ram version we have seen */ |
313 | uint32_t last_version; | |
8d820d6f JQ |
314 | /* How many times we have dirty too many pages */ |
315 | int dirty_rate_high_cnt; | |
f664da80 JQ |
316 | /* these variables are used for bitmap sync */ |
317 | /* last time we did a full bitmap_sync */ | |
318 | int64_t time_last_bitmap_sync; | |
eac74159 | 319 | /* bytes transferred at start_time */ |
c4bdf0cf | 320 | uint64_t bytes_xfer_prev; |
a66cd90c | 321 | /* number of dirty pages since start_time */ |
68908ed6 | 322 | uint64_t num_dirty_pages_period; |
b5833fde JQ |
323 | /* xbzrle misses since the beginning of the period */ |
324 | uint64_t xbzrle_cache_miss_prev; | |
e460a4b1 WW |
325 | /* Amount of xbzrle pages since the beginning of the period */ |
326 | uint64_t xbzrle_pages_prev; | |
327 | /* Amount of xbzrle encoded bytes since the beginning of the period */ | |
328 | uint64_t xbzrle_bytes_prev; | |
1a373522 DH |
329 | /* Start using XBZRLE (e.g., after the first round). */ |
330 | bool xbzrle_enabled; | |
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 | ||
9edabd4d | 384 | uint64_t ram_bytes_remaining(void) |
2f4fde93 | 385 | { |
bae416e5 DDAG |
386 | return ram_state ? (ram_state->migration_dirty_pages * TARGET_PAGE_SIZE) : |
387 | 0; | |
2f4fde93 JQ |
388 | } |
389 | ||
9360447d | 390 | MigrationStats ram_counters; |
96506894 | 391 | |
b8fb8cb7 DDAG |
392 | /* used by the search for pages to send */ |
393 | struct PageSearchStatus { | |
394 | /* Current block being searched */ | |
395 | RAMBlock *block; | |
a935e30f JQ |
396 | /* Current page to search from */ |
397 | unsigned long page; | |
b8fb8cb7 DDAG |
398 | /* Set once we wrap around */ |
399 | bool complete_round; | |
400 | }; | |
401 | typedef struct PageSearchStatus PageSearchStatus; | |
402 | ||
76e03000 XG |
403 | CompressionStats compression_counters; |
404 | ||
56e93d26 | 405 | struct CompressParam { |
56e93d26 | 406 | bool done; |
90e56fb4 | 407 | bool quit; |
5e5fdcff | 408 | bool zero_page; |
56e93d26 JQ |
409 | QEMUFile *file; |
410 | QemuMutex mutex; | |
411 | QemuCond cond; | |
412 | RAMBlock *block; | |
413 | ram_addr_t offset; | |
34ab9e97 XG |
414 | |
415 | /* internally used fields */ | |
dcaf446e | 416 | z_stream stream; |
34ab9e97 | 417 | uint8_t *originbuf; |
56e93d26 JQ |
418 | }; |
419 | typedef struct CompressParam CompressParam; | |
420 | ||
421 | struct DecompressParam { | |
73a8912b | 422 | bool done; |
90e56fb4 | 423 | bool quit; |
56e93d26 JQ |
424 | QemuMutex mutex; |
425 | QemuCond cond; | |
426 | void *des; | |
d341d9f3 | 427 | uint8_t *compbuf; |
56e93d26 | 428 | int len; |
797ca154 | 429 | z_stream stream; |
56e93d26 JQ |
430 | }; |
431 | typedef struct DecompressParam DecompressParam; | |
432 | ||
433 | static CompressParam *comp_param; | |
434 | static QemuThread *compress_threads; | |
435 | /* comp_done_cond is used to wake up the migration thread when | |
436 | * one of the compression threads has finished the compression. | |
437 | * comp_done_lock is used to co-work with comp_done_cond. | |
438 | */ | |
0d9f9a5c LL |
439 | static QemuMutex comp_done_lock; |
440 | static QemuCond comp_done_cond; | |
56e93d26 JQ |
441 | /* The empty QEMUFileOps will be used by file in CompressParam */ |
442 | static const QEMUFileOps empty_ops = { }; | |
443 | ||
34ab9e97 | 444 | static QEMUFile *decomp_file; |
56e93d26 JQ |
445 | static DecompressParam *decomp_param; |
446 | static QemuThread *decompress_threads; | |
73a8912b LL |
447 | static QemuMutex decomp_done_lock; |
448 | static QemuCond decomp_done_cond; | |
56e93d26 | 449 | |
5e5fdcff | 450 | static bool do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
6ef3771c | 451 | ram_addr_t offset, uint8_t *source_buf); |
56e93d26 JQ |
452 | |
453 | static void *do_data_compress(void *opaque) | |
454 | { | |
455 | CompressParam *param = opaque; | |
a7a9a88f LL |
456 | RAMBlock *block; |
457 | ram_addr_t offset; | |
5e5fdcff | 458 | bool zero_page; |
56e93d26 | 459 | |
a7a9a88f | 460 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 461 | while (!param->quit) { |
a7a9a88f LL |
462 | if (param->block) { |
463 | block = param->block; | |
464 | offset = param->offset; | |
465 | param->block = NULL; | |
466 | qemu_mutex_unlock(¶m->mutex); | |
467 | ||
5e5fdcff XG |
468 | zero_page = do_compress_ram_page(param->file, ¶m->stream, |
469 | block, offset, param->originbuf); | |
a7a9a88f | 470 | |
0d9f9a5c | 471 | qemu_mutex_lock(&comp_done_lock); |
a7a9a88f | 472 | param->done = true; |
5e5fdcff | 473 | param->zero_page = zero_page; |
0d9f9a5c LL |
474 | qemu_cond_signal(&comp_done_cond); |
475 | qemu_mutex_unlock(&comp_done_lock); | |
a7a9a88f LL |
476 | |
477 | qemu_mutex_lock(¶m->mutex); | |
478 | } else { | |
56e93d26 JQ |
479 | qemu_cond_wait(¶m->cond, ¶m->mutex); |
480 | } | |
56e93d26 | 481 | } |
a7a9a88f | 482 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
483 | |
484 | return NULL; | |
485 | } | |
486 | ||
f0afa331 | 487 | static void compress_threads_save_cleanup(void) |
56e93d26 JQ |
488 | { |
489 | int i, thread_count; | |
490 | ||
05306935 | 491 | if (!migrate_use_compression() || !comp_param) { |
56e93d26 JQ |
492 | return; |
493 | } | |
05306935 | 494 | |
56e93d26 JQ |
495 | thread_count = migrate_compress_threads(); |
496 | for (i = 0; i < thread_count; i++) { | |
dcaf446e XG |
497 | /* |
498 | * we use it as a indicator which shows if the thread is | |
499 | * properly init'd or not | |
500 | */ | |
501 | if (!comp_param[i].file) { | |
502 | break; | |
503 | } | |
05306935 FL |
504 | |
505 | qemu_mutex_lock(&comp_param[i].mutex); | |
506 | comp_param[i].quit = true; | |
507 | qemu_cond_signal(&comp_param[i].cond); | |
508 | qemu_mutex_unlock(&comp_param[i].mutex); | |
509 | ||
56e93d26 | 510 | qemu_thread_join(compress_threads + i); |
56e93d26 JQ |
511 | qemu_mutex_destroy(&comp_param[i].mutex); |
512 | qemu_cond_destroy(&comp_param[i].cond); | |
dcaf446e | 513 | deflateEnd(&comp_param[i].stream); |
34ab9e97 | 514 | g_free(comp_param[i].originbuf); |
dcaf446e XG |
515 | qemu_fclose(comp_param[i].file); |
516 | comp_param[i].file = NULL; | |
56e93d26 | 517 | } |
0d9f9a5c LL |
518 | qemu_mutex_destroy(&comp_done_lock); |
519 | qemu_cond_destroy(&comp_done_cond); | |
56e93d26 JQ |
520 | g_free(compress_threads); |
521 | g_free(comp_param); | |
56e93d26 JQ |
522 | compress_threads = NULL; |
523 | comp_param = NULL; | |
56e93d26 JQ |
524 | } |
525 | ||
dcaf446e | 526 | static int compress_threads_save_setup(void) |
56e93d26 JQ |
527 | { |
528 | int i, thread_count; | |
529 | ||
530 | if (!migrate_use_compression()) { | |
dcaf446e | 531 | return 0; |
56e93d26 | 532 | } |
56e93d26 JQ |
533 | thread_count = migrate_compress_threads(); |
534 | compress_threads = g_new0(QemuThread, thread_count); | |
535 | comp_param = g_new0(CompressParam, thread_count); | |
0d9f9a5c LL |
536 | qemu_cond_init(&comp_done_cond); |
537 | qemu_mutex_init(&comp_done_lock); | |
56e93d26 | 538 | for (i = 0; i < thread_count; i++) { |
34ab9e97 XG |
539 | comp_param[i].originbuf = g_try_malloc(TARGET_PAGE_SIZE); |
540 | if (!comp_param[i].originbuf) { | |
541 | goto exit; | |
542 | } | |
543 | ||
dcaf446e XG |
544 | if (deflateInit(&comp_param[i].stream, |
545 | migrate_compress_level()) != Z_OK) { | |
34ab9e97 | 546 | g_free(comp_param[i].originbuf); |
dcaf446e XG |
547 | goto exit; |
548 | } | |
549 | ||
e110aa91 C |
550 | /* comp_param[i].file is just used as a dummy buffer to save data, |
551 | * set its ops to empty. | |
56e93d26 JQ |
552 | */ |
553 | comp_param[i].file = qemu_fopen_ops(NULL, &empty_ops); | |
554 | comp_param[i].done = true; | |
90e56fb4 | 555 | comp_param[i].quit = false; |
56e93d26 JQ |
556 | qemu_mutex_init(&comp_param[i].mutex); |
557 | qemu_cond_init(&comp_param[i].cond); | |
558 | qemu_thread_create(compress_threads + i, "compress", | |
559 | do_data_compress, comp_param + i, | |
560 | QEMU_THREAD_JOINABLE); | |
561 | } | |
dcaf446e XG |
562 | return 0; |
563 | ||
564 | exit: | |
565 | compress_threads_save_cleanup(); | |
566 | return -1; | |
56e93d26 JQ |
567 | } |
568 | ||
569 | /** | |
3d0684b2 | 570 | * save_page_header: write page header to wire |
56e93d26 JQ |
571 | * |
572 | * If this is the 1st block, it also writes the block identification | |
573 | * | |
3d0684b2 | 574 | * Returns the number of bytes written |
56e93d26 JQ |
575 | * |
576 | * @f: QEMUFile where to send the data | |
577 | * @block: block that contains the page we want to send | |
578 | * @offset: offset inside the block for the page | |
579 | * in the lower bits, it contains flags | |
580 | */ | |
2bf3aa85 JQ |
581 | static size_t save_page_header(RAMState *rs, QEMUFile *f, RAMBlock *block, |
582 | ram_addr_t offset) | |
56e93d26 | 583 | { |
9f5f380b | 584 | size_t size, len; |
56e93d26 | 585 | |
24795694 JQ |
586 | if (block == rs->last_sent_block) { |
587 | offset |= RAM_SAVE_FLAG_CONTINUE; | |
588 | } | |
2bf3aa85 | 589 | qemu_put_be64(f, offset); |
56e93d26 JQ |
590 | size = 8; |
591 | ||
592 | if (!(offset & RAM_SAVE_FLAG_CONTINUE)) { | |
9f5f380b | 593 | len = strlen(block->idstr); |
2bf3aa85 JQ |
594 | qemu_put_byte(f, len); |
595 | qemu_put_buffer(f, (uint8_t *)block->idstr, len); | |
9f5f380b | 596 | size += 1 + len; |
24795694 | 597 | rs->last_sent_block = block; |
56e93d26 JQ |
598 | } |
599 | return size; | |
600 | } | |
601 | ||
3d0684b2 JQ |
602 | /** |
603 | * mig_throttle_guest_down: throotle down the guest | |
604 | * | |
605 | * Reduce amount of guest cpu execution to hopefully slow down memory | |
606 | * writes. If guest dirty memory rate is reduced below the rate at | |
607 | * which we can transfer pages to the destination then we should be | |
608 | * able to complete migration. Some workloads dirty memory way too | |
609 | * fast and will not effectively converge, even with auto-converge. | |
070afca2 | 610 | */ |
cbbf8182 KZ |
611 | static void mig_throttle_guest_down(uint64_t bytes_dirty_period, |
612 | uint64_t bytes_dirty_threshold) | |
070afca2 JH |
613 | { |
614 | MigrationState *s = migrate_get_current(); | |
2594f56d | 615 | uint64_t pct_initial = s->parameters.cpu_throttle_initial; |
cbbf8182 KZ |
616 | uint64_t pct_increment = s->parameters.cpu_throttle_increment; |
617 | bool pct_tailslow = s->parameters.cpu_throttle_tailslow; | |
4cbc9c7f | 618 | int pct_max = s->parameters.max_cpu_throttle; |
070afca2 | 619 | |
cbbf8182 KZ |
620 | uint64_t throttle_now = cpu_throttle_get_percentage(); |
621 | uint64_t cpu_now, cpu_ideal, throttle_inc; | |
622 | ||
070afca2 JH |
623 | /* We have not started throttling yet. Let's start it. */ |
624 | if (!cpu_throttle_active()) { | |
625 | cpu_throttle_set(pct_initial); | |
626 | } else { | |
627 | /* Throttling already on, just increase the rate */ | |
cbbf8182 KZ |
628 | if (!pct_tailslow) { |
629 | throttle_inc = pct_increment; | |
630 | } else { | |
631 | /* Compute the ideal CPU percentage used by Guest, which may | |
632 | * make the dirty rate match the dirty rate threshold. */ | |
633 | cpu_now = 100 - throttle_now; | |
634 | cpu_ideal = cpu_now * (bytes_dirty_threshold * 1.0 / | |
635 | bytes_dirty_period); | |
636 | throttle_inc = MIN(cpu_now - cpu_ideal, pct_increment); | |
637 | } | |
638 | cpu_throttle_set(MIN(throttle_now + throttle_inc, pct_max)); | |
070afca2 JH |
639 | } |
640 | } | |
641 | ||
3d0684b2 JQ |
642 | /** |
643 | * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache | |
644 | * | |
6f37bb8b | 645 | * @rs: current RAM state |
3d0684b2 JQ |
646 | * @current_addr: address for the zero page |
647 | * | |
648 | * Update the xbzrle cache to reflect a page that's been sent as all 0. | |
56e93d26 JQ |
649 | * The important thing is that a stale (not-yet-0'd) page be replaced |
650 | * by the new data. | |
651 | * As a bonus, if the page wasn't in the cache it gets added so that | |
3d0684b2 | 652 | * when a small write is made into the 0'd page it gets XBZRLE sent. |
56e93d26 | 653 | */ |
6f37bb8b | 654 | static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr) |
56e93d26 | 655 | { |
1a373522 | 656 | if (!rs->xbzrle_enabled) { |
56e93d26 JQ |
657 | return; |
658 | } | |
659 | ||
660 | /* We don't care if this fails to allocate a new cache page | |
661 | * as long as it updated an old one */ | |
c00e0928 | 662 | cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page, |
9360447d | 663 | ram_counters.dirty_sync_count); |
56e93d26 JQ |
664 | } |
665 | ||
666 | #define ENCODING_FLAG_XBZRLE 0x1 | |
667 | ||
668 | /** | |
669 | * save_xbzrle_page: compress and send current page | |
670 | * | |
671 | * Returns: 1 means that we wrote the page | |
672 | * 0 means that page is identical to the one already sent | |
673 | * -1 means that xbzrle would be longer than normal | |
674 | * | |
5a987738 | 675 | * @rs: current RAM state |
3d0684b2 JQ |
676 | * @current_data: pointer to the address of the page contents |
677 | * @current_addr: addr of the page | |
56e93d26 JQ |
678 | * @block: block that contains the page we want to send |
679 | * @offset: offset inside the block for the page | |
680 | * @last_stage: if we are at the completion stage | |
56e93d26 | 681 | */ |
204b88b8 | 682 | static int save_xbzrle_page(RAMState *rs, uint8_t **current_data, |
56e93d26 | 683 | ram_addr_t current_addr, RAMBlock *block, |
072c2511 | 684 | ram_addr_t offset, bool last_stage) |
56e93d26 JQ |
685 | { |
686 | int encoded_len = 0, bytes_xbzrle; | |
687 | uint8_t *prev_cached_page; | |
688 | ||
9360447d JQ |
689 | if (!cache_is_cached(XBZRLE.cache, current_addr, |
690 | ram_counters.dirty_sync_count)) { | |
691 | xbzrle_counters.cache_miss++; | |
56e93d26 JQ |
692 | if (!last_stage) { |
693 | if (cache_insert(XBZRLE.cache, current_addr, *current_data, | |
9360447d | 694 | ram_counters.dirty_sync_count) == -1) { |
56e93d26 JQ |
695 | return -1; |
696 | } else { | |
697 | /* update *current_data when the page has been | |
698 | inserted into cache */ | |
699 | *current_data = get_cached_data(XBZRLE.cache, current_addr); | |
700 | } | |
701 | } | |
702 | return -1; | |
703 | } | |
704 | ||
e460a4b1 WW |
705 | /* |
706 | * Reaching here means the page has hit the xbzrle cache, no matter what | |
707 | * encoding result it is (normal encoding, overflow or skipping the page), | |
3a4452d8 | 708 | * count the page as encoded. This is used to calculate the encoding rate. |
e460a4b1 WW |
709 | * |
710 | * Example: 2 pages (8KB) being encoded, first page encoding generates 2KB, | |
711 | * 2nd page turns out to be skipped (i.e. no new bytes written to the | |
712 | * page), the overall encoding rate will be 8KB / 2KB = 4, which has the | |
713 | * skipped page included. In this way, the encoding rate can tell if the | |
714 | * guest page is good for xbzrle encoding. | |
715 | */ | |
716 | xbzrle_counters.pages++; | |
56e93d26 JQ |
717 | prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); |
718 | ||
719 | /* save current buffer into memory */ | |
720 | memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); | |
721 | ||
722 | /* XBZRLE encoding (if there is no overflow) */ | |
723 | encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, | |
724 | TARGET_PAGE_SIZE, XBZRLE.encoded_buf, | |
725 | TARGET_PAGE_SIZE); | |
ca353803 WY |
726 | |
727 | /* | |
728 | * Update the cache contents, so that it corresponds to the data | |
729 | * sent, in all cases except where we skip the page. | |
730 | */ | |
731 | if (!last_stage && encoded_len != 0) { | |
732 | memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); | |
733 | /* | |
734 | * In the case where we couldn't compress, ensure that the caller | |
735 | * sends the data from the cache, since the guest might have | |
736 | * changed the RAM since we copied it. | |
737 | */ | |
738 | *current_data = prev_cached_page; | |
739 | } | |
740 | ||
56e93d26 | 741 | if (encoded_len == 0) { |
55c4446b | 742 | trace_save_xbzrle_page_skipping(); |
56e93d26 JQ |
743 | return 0; |
744 | } else if (encoded_len == -1) { | |
55c4446b | 745 | trace_save_xbzrle_page_overflow(); |
9360447d | 746 | xbzrle_counters.overflow++; |
e460a4b1 | 747 | xbzrle_counters.bytes += TARGET_PAGE_SIZE; |
56e93d26 JQ |
748 | return -1; |
749 | } | |
750 | ||
56e93d26 | 751 | /* Send XBZRLE based compressed page */ |
2bf3aa85 | 752 | bytes_xbzrle = save_page_header(rs, rs->f, block, |
204b88b8 JQ |
753 | offset | RAM_SAVE_FLAG_XBZRLE); |
754 | qemu_put_byte(rs->f, ENCODING_FLAG_XBZRLE); | |
755 | qemu_put_be16(rs->f, encoded_len); | |
756 | qemu_put_buffer(rs->f, XBZRLE.encoded_buf, encoded_len); | |
56e93d26 | 757 | bytes_xbzrle += encoded_len + 1 + 2; |
e460a4b1 WW |
758 | /* |
759 | * Like compressed_size (please see update_compress_thread_counts), | |
760 | * the xbzrle encoded bytes don't count the 8 byte header with | |
761 | * RAM_SAVE_FLAG_CONTINUE. | |
762 | */ | |
763 | xbzrle_counters.bytes += bytes_xbzrle - 8; | |
9360447d | 764 | ram_counters.transferred += bytes_xbzrle; |
56e93d26 JQ |
765 | |
766 | return 1; | |
767 | } | |
768 | ||
3d0684b2 JQ |
769 | /** |
770 | * migration_bitmap_find_dirty: find the next dirty page from start | |
f3f491fc | 771 | * |
a5f7b1a6 | 772 | * Returns the page offset within memory region of the start of a dirty page |
3d0684b2 | 773 | * |
6f37bb8b | 774 | * @rs: current RAM state |
3d0684b2 | 775 | * @rb: RAMBlock where to search for dirty pages |
a935e30f | 776 | * @start: page where we start the search |
f3f491fc | 777 | */ |
56e93d26 | 778 | static inline |
a935e30f | 779 | unsigned long migration_bitmap_find_dirty(RAMState *rs, RAMBlock *rb, |
f20e2865 | 780 | unsigned long start) |
56e93d26 | 781 | { |
6b6712ef JQ |
782 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; |
783 | unsigned long *bitmap = rb->bmap; | |
56e93d26 | 784 | |
fbd162e6 | 785 | if (ramblock_is_ignored(rb)) { |
b895de50 CLG |
786 | return size; |
787 | } | |
788 | ||
1a373522 | 789 | return find_next_bit(bitmap, size, start); |
56e93d26 JQ |
790 | } |
791 | ||
06b10688 | 792 | static inline bool migration_bitmap_clear_dirty(RAMState *rs, |
f20e2865 JQ |
793 | RAMBlock *rb, |
794 | unsigned long page) | |
a82d593b DDAG |
795 | { |
796 | bool ret; | |
a82d593b | 797 | |
37396950 | 798 | QEMU_LOCK_GUARD(&rs->bitmap_mutex); |
002cad6b PX |
799 | |
800 | /* | |
801 | * Clear dirty bitmap if needed. This _must_ be called before we | |
802 | * send any of the page in the chunk because we need to make sure | |
803 | * we can capture further page content changes when we sync dirty | |
804 | * log the next time. So as long as we are going to send any of | |
805 | * the page in the chunk we clear the remote dirty bitmap for all. | |
806 | * Clearing it earlier won't be a problem, but too late will. | |
807 | */ | |
808 | if (rb->clear_bmap && clear_bmap_test_and_clear(rb, page)) { | |
809 | uint8_t shift = rb->clear_bmap_shift; | |
810 | hwaddr size = 1ULL << (TARGET_PAGE_BITS + shift); | |
8bba004c | 811 | hwaddr start = (((ram_addr_t)page) << TARGET_PAGE_BITS) & (-size); |
002cad6b PX |
812 | |
813 | /* | |
814 | * CLEAR_BITMAP_SHIFT_MIN should always guarantee this... this | |
815 | * can make things easier sometimes since then start address | |
816 | * of the small chunk will always be 64 pages aligned so the | |
817 | * bitmap will always be aligned to unsigned long. We should | |
818 | * even be able to remove this restriction but I'm simply | |
819 | * keeping it. | |
820 | */ | |
821 | assert(shift >= 6); | |
822 | trace_migration_bitmap_clear_dirty(rb->idstr, start, size, page); | |
823 | memory_region_clear_dirty_bitmap(rb->mr, start, size); | |
824 | } | |
825 | ||
6b6712ef | 826 | ret = test_and_clear_bit(page, rb->bmap); |
a82d593b DDAG |
827 | |
828 | if (ret) { | |
0d8ec885 | 829 | rs->migration_dirty_pages--; |
a82d593b | 830 | } |
386a907b | 831 | |
a82d593b DDAG |
832 | return ret; |
833 | } | |
834 | ||
267691b6 | 835 | /* Called with RCU critical section */ |
7a3e9571 | 836 | static void ramblock_sync_dirty_bitmap(RAMState *rs, RAMBlock *rb) |
56e93d26 | 837 | { |
fb613580 KZ |
838 | uint64_t new_dirty_pages = |
839 | cpu_physical_memory_sync_dirty_bitmap(rb, 0, rb->used_length); | |
840 | ||
841 | rs->migration_dirty_pages += new_dirty_pages; | |
842 | rs->num_dirty_pages_period += new_dirty_pages; | |
56e93d26 JQ |
843 | } |
844 | ||
3d0684b2 JQ |
845 | /** |
846 | * ram_pagesize_summary: calculate all the pagesizes of a VM | |
847 | * | |
848 | * Returns a summary bitmap of the page sizes of all RAMBlocks | |
849 | * | |
850 | * For VMs with just normal pages this is equivalent to the host page | |
851 | * size. If it's got some huge pages then it's the OR of all the | |
852 | * different page sizes. | |
e8ca1db2 DDAG |
853 | */ |
854 | uint64_t ram_pagesize_summary(void) | |
855 | { | |
856 | RAMBlock *block; | |
857 | uint64_t summary = 0; | |
858 | ||
fbd162e6 | 859 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
e8ca1db2 DDAG |
860 | summary |= block->page_size; |
861 | } | |
862 | ||
863 | return summary; | |
864 | } | |
865 | ||
aecbfe9c XG |
866 | uint64_t ram_get_total_transferred_pages(void) |
867 | { | |
868 | return ram_counters.normal + ram_counters.duplicate + | |
869 | compression_counters.pages + xbzrle_counters.pages; | |
870 | } | |
871 | ||
b734035b XG |
872 | static void migration_update_rates(RAMState *rs, int64_t end_time) |
873 | { | |
be8b02ed | 874 | uint64_t page_count = rs->target_page_count - rs->target_page_count_prev; |
76e03000 | 875 | double compressed_size; |
b734035b XG |
876 | |
877 | /* calculate period counters */ | |
878 | ram_counters.dirty_pages_rate = rs->num_dirty_pages_period * 1000 | |
879 | / (end_time - rs->time_last_bitmap_sync); | |
880 | ||
be8b02ed | 881 | if (!page_count) { |
b734035b XG |
882 | return; |
883 | } | |
884 | ||
885 | if (migrate_use_xbzrle()) { | |
e460a4b1 WW |
886 | double encoded_size, unencoded_size; |
887 | ||
b734035b | 888 | xbzrle_counters.cache_miss_rate = (double)(xbzrle_counters.cache_miss - |
be8b02ed | 889 | rs->xbzrle_cache_miss_prev) / page_count; |
b734035b | 890 | rs->xbzrle_cache_miss_prev = xbzrle_counters.cache_miss; |
e460a4b1 WW |
891 | unencoded_size = (xbzrle_counters.pages - rs->xbzrle_pages_prev) * |
892 | TARGET_PAGE_SIZE; | |
893 | encoded_size = xbzrle_counters.bytes - rs->xbzrle_bytes_prev; | |
92271402 | 894 | if (xbzrle_counters.pages == rs->xbzrle_pages_prev || !encoded_size) { |
e460a4b1 | 895 | xbzrle_counters.encoding_rate = 0; |
e460a4b1 WW |
896 | } else { |
897 | xbzrle_counters.encoding_rate = unencoded_size / encoded_size; | |
898 | } | |
899 | rs->xbzrle_pages_prev = xbzrle_counters.pages; | |
900 | rs->xbzrle_bytes_prev = xbzrle_counters.bytes; | |
b734035b | 901 | } |
76e03000 XG |
902 | |
903 | if (migrate_use_compression()) { | |
904 | compression_counters.busy_rate = (double)(compression_counters.busy - | |
905 | rs->compress_thread_busy_prev) / page_count; | |
906 | rs->compress_thread_busy_prev = compression_counters.busy; | |
907 | ||
908 | compressed_size = compression_counters.compressed_size - | |
909 | rs->compressed_size_prev; | |
910 | if (compressed_size) { | |
911 | double uncompressed_size = (compression_counters.pages - | |
912 | rs->compress_pages_prev) * TARGET_PAGE_SIZE; | |
913 | ||
914 | /* Compression-Ratio = Uncompressed-size / Compressed-size */ | |
915 | compression_counters.compression_rate = | |
916 | uncompressed_size / compressed_size; | |
917 | ||
918 | rs->compress_pages_prev = compression_counters.pages; | |
919 | rs->compressed_size_prev = compression_counters.compressed_size; | |
920 | } | |
921 | } | |
b734035b XG |
922 | } |
923 | ||
dc14a470 KZ |
924 | static void migration_trigger_throttle(RAMState *rs) |
925 | { | |
926 | MigrationState *s = migrate_get_current(); | |
927 | uint64_t threshold = s->parameters.throttle_trigger_threshold; | |
928 | ||
929 | uint64_t bytes_xfer_period = ram_counters.transferred - rs->bytes_xfer_prev; | |
930 | uint64_t bytes_dirty_period = rs->num_dirty_pages_period * TARGET_PAGE_SIZE; | |
931 | uint64_t bytes_dirty_threshold = bytes_xfer_period * threshold / 100; | |
932 | ||
933 | /* During block migration the auto-converge logic incorrectly detects | |
934 | * that ram migration makes no progress. Avoid this by disabling the | |
935 | * throttling logic during the bulk phase of block migration. */ | |
936 | if (migrate_auto_converge() && !blk_mig_bulk_active()) { | |
937 | /* The following detection logic can be refined later. For now: | |
938 | Check to see if the ratio between dirtied bytes and the approx. | |
939 | amount of bytes that just got transferred since the last time | |
940 | we were in this routine reaches the threshold. If that happens | |
941 | twice, start or increase throttling. */ | |
942 | ||
943 | if ((bytes_dirty_period > bytes_dirty_threshold) && | |
944 | (++rs->dirty_rate_high_cnt >= 2)) { | |
945 | trace_migration_throttle(); | |
946 | rs->dirty_rate_high_cnt = 0; | |
cbbf8182 KZ |
947 | mig_throttle_guest_down(bytes_dirty_period, |
948 | bytes_dirty_threshold); | |
dc14a470 KZ |
949 | } |
950 | } | |
951 | } | |
952 | ||
8d820d6f | 953 | static void migration_bitmap_sync(RAMState *rs) |
56e93d26 JQ |
954 | { |
955 | RAMBlock *block; | |
56e93d26 | 956 | int64_t end_time; |
56e93d26 | 957 | |
9360447d | 958 | ram_counters.dirty_sync_count++; |
56e93d26 | 959 | |
f664da80 JQ |
960 | if (!rs->time_last_bitmap_sync) { |
961 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56e93d26 JQ |
962 | } |
963 | ||
964 | trace_migration_bitmap_sync_start(); | |
9c1f8f44 | 965 | memory_global_dirty_log_sync(); |
56e93d26 | 966 | |
108cfae0 | 967 | qemu_mutex_lock(&rs->bitmap_mutex); |
89ac5a1d DDAG |
968 | WITH_RCU_READ_LOCK_GUARD() { |
969 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
970 | ramblock_sync_dirty_bitmap(rs, block); | |
971 | } | |
972 | ram_counters.remaining = ram_bytes_remaining(); | |
56e93d26 | 973 | } |
108cfae0 | 974 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 975 | |
9458a9a1 | 976 | memory_global_after_dirty_log_sync(); |
a66cd90c | 977 | trace_migration_bitmap_sync_end(rs->num_dirty_pages_period); |
1ffb5dfd | 978 | |
56e93d26 JQ |
979 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
980 | ||
981 | /* more than 1 second = 1000 millisecons */ | |
f664da80 | 982 | if (end_time > rs->time_last_bitmap_sync + 1000) { |
dc14a470 | 983 | migration_trigger_throttle(rs); |
070afca2 | 984 | |
b734035b XG |
985 | migration_update_rates(rs, end_time); |
986 | ||
be8b02ed | 987 | rs->target_page_count_prev = rs->target_page_count; |
d693c6f1 FF |
988 | |
989 | /* reset period counters */ | |
f664da80 | 990 | rs->time_last_bitmap_sync = end_time; |
a66cd90c | 991 | rs->num_dirty_pages_period = 0; |
dc14a470 | 992 | rs->bytes_xfer_prev = ram_counters.transferred; |
56e93d26 | 993 | } |
4addcd4f | 994 | if (migrate_use_events()) { |
3ab72385 | 995 | qapi_event_send_migration_pass(ram_counters.dirty_sync_count); |
4addcd4f | 996 | } |
56e93d26 JQ |
997 | } |
998 | ||
bd227060 WW |
999 | static void migration_bitmap_sync_precopy(RAMState *rs) |
1000 | { | |
1001 | Error *local_err = NULL; | |
1002 | ||
1003 | /* | |
1004 | * The current notifier usage is just an optimization to migration, so we | |
1005 | * don't stop the normal migration process in the error case. | |
1006 | */ | |
1007 | if (precopy_notify(PRECOPY_NOTIFY_BEFORE_BITMAP_SYNC, &local_err)) { | |
1008 | error_report_err(local_err); | |
b4a1733c | 1009 | local_err = NULL; |
bd227060 WW |
1010 | } |
1011 | ||
1012 | migration_bitmap_sync(rs); | |
1013 | ||
1014 | if (precopy_notify(PRECOPY_NOTIFY_AFTER_BITMAP_SYNC, &local_err)) { | |
1015 | error_report_err(local_err); | |
1016 | } | |
1017 | } | |
1018 | ||
6c97ec5f XG |
1019 | /** |
1020 | * save_zero_page_to_file: send the zero page to the file | |
1021 | * | |
1022 | * Returns the size of data written to the file, 0 means the page is not | |
1023 | * a zero page | |
1024 | * | |
1025 | * @rs: current RAM state | |
1026 | * @file: the file where the data is saved | |
1027 | * @block: block that contains the page we want to send | |
1028 | * @offset: offset inside the block for the page | |
1029 | */ | |
1030 | static int save_zero_page_to_file(RAMState *rs, QEMUFile *file, | |
1031 | RAMBlock *block, ram_addr_t offset) | |
1032 | { | |
1033 | uint8_t *p = block->host + offset; | |
1034 | int len = 0; | |
1035 | ||
1036 | if (is_zero_range(p, TARGET_PAGE_SIZE)) { | |
1037 | len += save_page_header(rs, file, block, offset | RAM_SAVE_FLAG_ZERO); | |
1038 | qemu_put_byte(file, 0); | |
1039 | len += 1; | |
1040 | } | |
1041 | return len; | |
1042 | } | |
1043 | ||
56e93d26 | 1044 | /** |
3d0684b2 | 1045 | * save_zero_page: send the zero page to the stream |
56e93d26 | 1046 | * |
3d0684b2 | 1047 | * Returns the number of pages written. |
56e93d26 | 1048 | * |
f7ccd61b | 1049 | * @rs: current RAM state |
56e93d26 JQ |
1050 | * @block: block that contains the page we want to send |
1051 | * @offset: offset inside the block for the page | |
56e93d26 | 1052 | */ |
7faccdc3 | 1053 | static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset) |
56e93d26 | 1054 | { |
6c97ec5f | 1055 | int len = save_zero_page_to_file(rs, rs->f, block, offset); |
56e93d26 | 1056 | |
6c97ec5f | 1057 | if (len) { |
9360447d | 1058 | ram_counters.duplicate++; |
6c97ec5f XG |
1059 | ram_counters.transferred += len; |
1060 | return 1; | |
56e93d26 | 1061 | } |
6c97ec5f | 1062 | return -1; |
56e93d26 JQ |
1063 | } |
1064 | ||
5727309d | 1065 | static void ram_release_pages(const char *rbname, uint64_t offset, int pages) |
53f09a10 | 1066 | { |
5727309d | 1067 | if (!migrate_release_ram() || !migration_in_postcopy()) { |
53f09a10 PB |
1068 | return; |
1069 | } | |
1070 | ||
8bba004c | 1071 | ram_discard_range(rbname, offset, ((ram_addr_t)pages) << TARGET_PAGE_BITS); |
53f09a10 PB |
1072 | } |
1073 | ||
059ff0fb XG |
1074 | /* |
1075 | * @pages: the number of pages written by the control path, | |
1076 | * < 0 - error | |
1077 | * > 0 - number of pages written | |
1078 | * | |
1079 | * Return true if the pages has been saved, otherwise false is returned. | |
1080 | */ | |
1081 | static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, | |
1082 | int *pages) | |
1083 | { | |
1084 | uint64_t bytes_xmit = 0; | |
1085 | int ret; | |
1086 | ||
1087 | *pages = -1; | |
1088 | ret = ram_control_save_page(rs->f, block->offset, offset, TARGET_PAGE_SIZE, | |
1089 | &bytes_xmit); | |
1090 | if (ret == RAM_SAVE_CONTROL_NOT_SUPP) { | |
1091 | return false; | |
1092 | } | |
1093 | ||
1094 | if (bytes_xmit) { | |
1095 | ram_counters.transferred += bytes_xmit; | |
1096 | *pages = 1; | |
1097 | } | |
1098 | ||
1099 | if (ret == RAM_SAVE_CONTROL_DELAYED) { | |
1100 | return true; | |
1101 | } | |
1102 | ||
1103 | if (bytes_xmit > 0) { | |
1104 | ram_counters.normal++; | |
1105 | } else if (bytes_xmit == 0) { | |
1106 | ram_counters.duplicate++; | |
1107 | } | |
1108 | ||
1109 | return true; | |
1110 | } | |
1111 | ||
65dacaa0 XG |
1112 | /* |
1113 | * directly send the page to the stream | |
1114 | * | |
1115 | * Returns the number of pages written. | |
1116 | * | |
1117 | * @rs: current RAM state | |
1118 | * @block: block that contains the page we want to send | |
1119 | * @offset: offset inside the block for the page | |
1120 | * @buf: the page to be sent | |
1121 | * @async: send to page asyncly | |
1122 | */ | |
1123 | static int save_normal_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, | |
1124 | uint8_t *buf, bool async) | |
1125 | { | |
1126 | ram_counters.transferred += save_page_header(rs, rs->f, block, | |
1127 | offset | RAM_SAVE_FLAG_PAGE); | |
1128 | if (async) { | |
1129 | qemu_put_buffer_async(rs->f, buf, TARGET_PAGE_SIZE, | |
1130 | migrate_release_ram() & | |
1131 | migration_in_postcopy()); | |
1132 | } else { | |
1133 | qemu_put_buffer(rs->f, buf, TARGET_PAGE_SIZE); | |
1134 | } | |
1135 | ram_counters.transferred += TARGET_PAGE_SIZE; | |
1136 | ram_counters.normal++; | |
1137 | return 1; | |
1138 | } | |
1139 | ||
56e93d26 | 1140 | /** |
3d0684b2 | 1141 | * ram_save_page: send the given page to the stream |
56e93d26 | 1142 | * |
3d0684b2 | 1143 | * Returns the number of pages written. |
3fd3c4b3 DDAG |
1144 | * < 0 - error |
1145 | * >=0 - Number of pages written - this might legally be 0 | |
1146 | * if xbzrle noticed the page was the same. | |
56e93d26 | 1147 | * |
6f37bb8b | 1148 | * @rs: current RAM state |
56e93d26 JQ |
1149 | * @block: block that contains the page we want to send |
1150 | * @offset: offset inside the block for the page | |
1151 | * @last_stage: if we are at the completion stage | |
56e93d26 | 1152 | */ |
a0a8aa14 | 1153 | static int ram_save_page(RAMState *rs, PageSearchStatus *pss, bool last_stage) |
56e93d26 JQ |
1154 | { |
1155 | int pages = -1; | |
56e93d26 | 1156 | uint8_t *p; |
56e93d26 | 1157 | bool send_async = true; |
a08f6890 | 1158 | RAMBlock *block = pss->block; |
8bba004c | 1159 | ram_addr_t offset = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS; |
059ff0fb | 1160 | ram_addr_t current_addr = block->offset + offset; |
56e93d26 | 1161 | |
2f68e399 | 1162 | p = block->host + offset; |
1db9d8e5 | 1163 | trace_ram_save_page(block->idstr, (uint64_t)offset, p); |
56e93d26 | 1164 | |
56e93d26 | 1165 | XBZRLE_cache_lock(); |
1a373522 | 1166 | if (rs->xbzrle_enabled && !migration_in_postcopy()) { |
059ff0fb XG |
1167 | pages = save_xbzrle_page(rs, &p, current_addr, block, |
1168 | offset, last_stage); | |
1169 | if (!last_stage) { | |
1170 | /* Can't send this cached data async, since the cache page | |
1171 | * might get updated before it gets to the wire | |
56e93d26 | 1172 | */ |
059ff0fb | 1173 | send_async = false; |
56e93d26 JQ |
1174 | } |
1175 | } | |
1176 | ||
1177 | /* XBZRLE overflow or normal page */ | |
1178 | if (pages == -1) { | |
65dacaa0 | 1179 | pages = save_normal_page(rs, block, offset, p, send_async); |
56e93d26 JQ |
1180 | } |
1181 | ||
1182 | XBZRLE_cache_unlock(); | |
1183 | ||
1184 | return pages; | |
1185 | } | |
1186 | ||
b9ee2f7d JQ |
1187 | static int ram_save_multifd_page(RAMState *rs, RAMBlock *block, |
1188 | ram_addr_t offset) | |
1189 | { | |
67a4c891 | 1190 | if (multifd_queue_page(rs->f, block, offset) < 0) { |
713f762a IR |
1191 | return -1; |
1192 | } | |
b9ee2f7d JQ |
1193 | ram_counters.normal++; |
1194 | ||
1195 | return 1; | |
1196 | } | |
1197 | ||
5e5fdcff | 1198 | static bool do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
6ef3771c | 1199 | ram_addr_t offset, uint8_t *source_buf) |
56e93d26 | 1200 | { |
53518d94 | 1201 | RAMState *rs = ram_state; |
a7a9a88f | 1202 | uint8_t *p = block->host + (offset & TARGET_PAGE_MASK); |
5e5fdcff | 1203 | bool zero_page = false; |
6ef3771c | 1204 | int ret; |
56e93d26 | 1205 | |
5e5fdcff XG |
1206 | if (save_zero_page_to_file(rs, f, block, offset)) { |
1207 | zero_page = true; | |
1208 | goto exit; | |
1209 | } | |
1210 | ||
6ef3771c | 1211 | save_page_header(rs, f, block, offset | RAM_SAVE_FLAG_COMPRESS_PAGE); |
34ab9e97 XG |
1212 | |
1213 | /* | |
1214 | * copy it to a internal buffer to avoid it being modified by VM | |
1215 | * so that we can catch up the error during compression and | |
1216 | * decompression | |
1217 | */ | |
1218 | memcpy(source_buf, p, TARGET_PAGE_SIZE); | |
6ef3771c XG |
1219 | ret = qemu_put_compression_data(f, stream, source_buf, TARGET_PAGE_SIZE); |
1220 | if (ret < 0) { | |
1221 | qemu_file_set_error(migrate_get_current()->to_dst_file, ret); | |
b3be2896 | 1222 | error_report("compressed data failed!"); |
5e5fdcff | 1223 | return false; |
b3be2896 | 1224 | } |
56e93d26 | 1225 | |
5e5fdcff | 1226 | exit: |
6ef3771c | 1227 | ram_release_pages(block->idstr, offset & TARGET_PAGE_MASK, 1); |
5e5fdcff XG |
1228 | return zero_page; |
1229 | } | |
1230 | ||
1231 | static void | |
1232 | update_compress_thread_counts(const CompressParam *param, int bytes_xmit) | |
1233 | { | |
76e03000 XG |
1234 | ram_counters.transferred += bytes_xmit; |
1235 | ||
5e5fdcff XG |
1236 | if (param->zero_page) { |
1237 | ram_counters.duplicate++; | |
76e03000 | 1238 | return; |
5e5fdcff | 1239 | } |
76e03000 XG |
1240 | |
1241 | /* 8 means a header with RAM_SAVE_FLAG_CONTINUE. */ | |
1242 | compression_counters.compressed_size += bytes_xmit - 8; | |
1243 | compression_counters.pages++; | |
56e93d26 JQ |
1244 | } |
1245 | ||
32b05495 XG |
1246 | static bool save_page_use_compression(RAMState *rs); |
1247 | ||
ce25d337 | 1248 | static void flush_compressed_data(RAMState *rs) |
56e93d26 JQ |
1249 | { |
1250 | int idx, len, thread_count; | |
1251 | ||
32b05495 | 1252 | if (!save_page_use_compression(rs)) { |
56e93d26 JQ |
1253 | return; |
1254 | } | |
1255 | thread_count = migrate_compress_threads(); | |
a7a9a88f | 1256 | |
0d9f9a5c | 1257 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 | 1258 | for (idx = 0; idx < thread_count; idx++) { |
a7a9a88f | 1259 | while (!comp_param[idx].done) { |
0d9f9a5c | 1260 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 | 1261 | } |
a7a9a88f | 1262 | } |
0d9f9a5c | 1263 | qemu_mutex_unlock(&comp_done_lock); |
a7a9a88f LL |
1264 | |
1265 | for (idx = 0; idx < thread_count; idx++) { | |
1266 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 1267 | if (!comp_param[idx].quit) { |
ce25d337 | 1268 | len = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
5e5fdcff XG |
1269 | /* |
1270 | * it's safe to fetch zero_page without holding comp_done_lock | |
1271 | * as there is no further request submitted to the thread, | |
1272 | * i.e, the thread should be waiting for a request at this point. | |
1273 | */ | |
1274 | update_compress_thread_counts(&comp_param[idx], len); | |
56e93d26 | 1275 | } |
a7a9a88f | 1276 | qemu_mutex_unlock(&comp_param[idx].mutex); |
56e93d26 JQ |
1277 | } |
1278 | } | |
1279 | ||
1280 | static inline void set_compress_params(CompressParam *param, RAMBlock *block, | |
1281 | ram_addr_t offset) | |
1282 | { | |
1283 | param->block = block; | |
1284 | param->offset = offset; | |
1285 | } | |
1286 | ||
ce25d337 JQ |
1287 | static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block, |
1288 | ram_addr_t offset) | |
56e93d26 JQ |
1289 | { |
1290 | int idx, thread_count, bytes_xmit = -1, pages = -1; | |
1d58872a | 1291 | bool wait = migrate_compress_wait_thread(); |
56e93d26 JQ |
1292 | |
1293 | thread_count = migrate_compress_threads(); | |
0d9f9a5c | 1294 | qemu_mutex_lock(&comp_done_lock); |
1d58872a XG |
1295 | retry: |
1296 | for (idx = 0; idx < thread_count; idx++) { | |
1297 | if (comp_param[idx].done) { | |
1298 | comp_param[idx].done = false; | |
1299 | bytes_xmit = qemu_put_qemu_file(rs->f, comp_param[idx].file); | |
1300 | qemu_mutex_lock(&comp_param[idx].mutex); | |
1301 | set_compress_params(&comp_param[idx], block, offset); | |
1302 | qemu_cond_signal(&comp_param[idx].cond); | |
1303 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
1304 | pages = 1; | |
5e5fdcff | 1305 | update_compress_thread_counts(&comp_param[idx], bytes_xmit); |
56e93d26 | 1306 | break; |
56e93d26 JQ |
1307 | } |
1308 | } | |
1d58872a XG |
1309 | |
1310 | /* | |
1311 | * wait for the free thread if the user specifies 'compress-wait-thread', | |
1312 | * otherwise we will post the page out in the main thread as normal page. | |
1313 | */ | |
1314 | if (pages < 0 && wait) { | |
1315 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); | |
1316 | goto retry; | |
1317 | } | |
0d9f9a5c | 1318 | qemu_mutex_unlock(&comp_done_lock); |
56e93d26 JQ |
1319 | |
1320 | return pages; | |
1321 | } | |
1322 | ||
3d0684b2 JQ |
1323 | /** |
1324 | * find_dirty_block: find the next dirty page and update any state | |
1325 | * associated with the search process. | |
b9e60928 | 1326 | * |
a5f7b1a6 | 1327 | * Returns true if a page is found |
b9e60928 | 1328 | * |
6f37bb8b | 1329 | * @rs: current RAM state |
3d0684b2 JQ |
1330 | * @pss: data about the state of the current dirty page scan |
1331 | * @again: set to false if the search has scanned the whole of RAM | |
b9e60928 | 1332 | */ |
f20e2865 | 1333 | static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again) |
b9e60928 | 1334 | { |
f20e2865 | 1335 | pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page); |
6f37bb8b | 1336 | if (pss->complete_round && pss->block == rs->last_seen_block && |
a935e30f | 1337 | pss->page >= rs->last_page) { |
b9e60928 DDAG |
1338 | /* |
1339 | * We've been once around the RAM and haven't found anything. | |
1340 | * Give up. | |
1341 | */ | |
1342 | *again = false; | |
1343 | return false; | |
1344 | } | |
8bba004c AR |
1345 | if ((((ram_addr_t)pss->page) << TARGET_PAGE_BITS) |
1346 | >= pss->block->used_length) { | |
b9e60928 | 1347 | /* Didn't find anything in this RAM Block */ |
a935e30f | 1348 | pss->page = 0; |
b9e60928 DDAG |
1349 | pss->block = QLIST_NEXT_RCU(pss->block, next); |
1350 | if (!pss->block) { | |
48df9d80 XG |
1351 | /* |
1352 | * If memory migration starts over, we will meet a dirtied page | |
1353 | * which may still exists in compression threads's ring, so we | |
1354 | * should flush the compressed data to make sure the new page | |
1355 | * is not overwritten by the old one in the destination. | |
1356 | * | |
1357 | * Also If xbzrle is on, stop using the data compression at this | |
1358 | * point. In theory, xbzrle can do better than compression. | |
1359 | */ | |
1360 | flush_compressed_data(rs); | |
1361 | ||
b9e60928 DDAG |
1362 | /* Hit the end of the list */ |
1363 | pss->block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1364 | /* Flag that we've looped */ | |
1365 | pss->complete_round = true; | |
1a373522 DH |
1366 | /* After the first round, enable XBZRLE. */ |
1367 | if (migrate_use_xbzrle()) { | |
1368 | rs->xbzrle_enabled = true; | |
1369 | } | |
b9e60928 DDAG |
1370 | } |
1371 | /* Didn't find anything this time, but try again on the new block */ | |
1372 | *again = true; | |
1373 | return false; | |
1374 | } else { | |
1375 | /* Can go around again, but... */ | |
1376 | *again = true; | |
1377 | /* We've found something so probably don't need to */ | |
1378 | return true; | |
1379 | } | |
1380 | } | |
1381 | ||
3d0684b2 JQ |
1382 | /** |
1383 | * unqueue_page: gets a page of the queue | |
1384 | * | |
a82d593b | 1385 | * Helper for 'get_queued_page' - gets a page off the queue |
a82d593b | 1386 | * |
3d0684b2 JQ |
1387 | * Returns the block of the page (or NULL if none available) |
1388 | * | |
ec481c6c | 1389 | * @rs: current RAM state |
3d0684b2 | 1390 | * @offset: used to return the offset within the RAMBlock |
a82d593b | 1391 | */ |
f20e2865 | 1392 | static RAMBlock *unqueue_page(RAMState *rs, ram_addr_t *offset) |
a82d593b DDAG |
1393 | { |
1394 | RAMBlock *block = NULL; | |
1395 | ||
ae526e32 XG |
1396 | if (QSIMPLEQ_EMPTY_ATOMIC(&rs->src_page_requests)) { |
1397 | return NULL; | |
1398 | } | |
1399 | ||
6e8a355d | 1400 | QEMU_LOCK_GUARD(&rs->src_page_req_mutex); |
ec481c6c JQ |
1401 | if (!QSIMPLEQ_EMPTY(&rs->src_page_requests)) { |
1402 | struct RAMSrcPageRequest *entry = | |
1403 | QSIMPLEQ_FIRST(&rs->src_page_requests); | |
a82d593b DDAG |
1404 | block = entry->rb; |
1405 | *offset = entry->offset; | |
a82d593b DDAG |
1406 | |
1407 | if (entry->len > TARGET_PAGE_SIZE) { | |
1408 | entry->len -= TARGET_PAGE_SIZE; | |
1409 | entry->offset += TARGET_PAGE_SIZE; | |
1410 | } else { | |
1411 | memory_region_unref(block->mr); | |
ec481c6c | 1412 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
a82d593b | 1413 | g_free(entry); |
e03a34f8 | 1414 | migration_consume_urgent_request(); |
a82d593b DDAG |
1415 | } |
1416 | } | |
a82d593b DDAG |
1417 | |
1418 | return block; | |
1419 | } | |
1420 | ||
278e2f55 AG |
1421 | #if defined(__linux__) |
1422 | /** | |
1423 | * poll_fault_page: try to get next UFFD write fault page and, if pending fault | |
1424 | * is found, return RAM block pointer and page offset | |
1425 | * | |
1426 | * Returns pointer to the RAMBlock containing faulting page, | |
1427 | * NULL if no write faults are pending | |
1428 | * | |
1429 | * @rs: current RAM state | |
1430 | * @offset: page offset from the beginning of the block | |
1431 | */ | |
1432 | static RAMBlock *poll_fault_page(RAMState *rs, ram_addr_t *offset) | |
1433 | { | |
1434 | struct uffd_msg uffd_msg; | |
1435 | void *page_address; | |
82ea3e3b | 1436 | RAMBlock *block; |
278e2f55 AG |
1437 | int res; |
1438 | ||
1439 | if (!migrate_background_snapshot()) { | |
1440 | return NULL; | |
1441 | } | |
1442 | ||
1443 | res = uffd_read_events(rs->uffdio_fd, &uffd_msg, 1); | |
1444 | if (res <= 0) { | |
1445 | return NULL; | |
1446 | } | |
1447 | ||
1448 | page_address = (void *)(uintptr_t) uffd_msg.arg.pagefault.address; | |
82ea3e3b AG |
1449 | block = qemu_ram_block_from_host(page_address, false, offset); |
1450 | assert(block && (block->flags & RAM_UF_WRITEPROTECT) != 0); | |
1451 | return block; | |
278e2f55 AG |
1452 | } |
1453 | ||
1454 | /** | |
1455 | * ram_save_release_protection: release UFFD write protection after | |
1456 | * a range of pages has been saved | |
1457 | * | |
1458 | * @rs: current RAM state | |
1459 | * @pss: page-search-status structure | |
1460 | * @start_page: index of the first page in the range relative to pss->block | |
1461 | * | |
1462 | * Returns 0 on success, negative value in case of an error | |
1463 | */ | |
1464 | static int ram_save_release_protection(RAMState *rs, PageSearchStatus *pss, | |
1465 | unsigned long start_page) | |
1466 | { | |
1467 | int res = 0; | |
1468 | ||
1469 | /* Check if page is from UFFD-managed region. */ | |
1470 | if (pss->block->flags & RAM_UF_WRITEPROTECT) { | |
1471 | void *page_address = pss->block->host + (start_page << TARGET_PAGE_BITS); | |
1472 | uint64_t run_length = (pss->page - start_page + 1) << TARGET_PAGE_BITS; | |
1473 | ||
1474 | /* Flush async buffers before un-protect. */ | |
1475 | qemu_fflush(rs->f); | |
1476 | /* Un-protect memory range. */ | |
1477 | res = uffd_change_protection(rs->uffdio_fd, page_address, run_length, | |
1478 | false, false); | |
1479 | } | |
1480 | ||
1481 | return res; | |
1482 | } | |
1483 | ||
1484 | /* ram_write_tracking_available: check if kernel supports required UFFD features | |
1485 | * | |
1486 | * Returns true if supports, false otherwise | |
1487 | */ | |
1488 | bool ram_write_tracking_available(void) | |
1489 | { | |
1490 | uint64_t uffd_features; | |
1491 | int res; | |
1492 | ||
1493 | res = uffd_query_features(&uffd_features); | |
1494 | return (res == 0 && | |
1495 | (uffd_features & UFFD_FEATURE_PAGEFAULT_FLAG_WP) != 0); | |
1496 | } | |
1497 | ||
1498 | /* ram_write_tracking_compatible: check if guest configuration is | |
1499 | * compatible with 'write-tracking' | |
1500 | * | |
1501 | * Returns true if compatible, false otherwise | |
1502 | */ | |
1503 | bool ram_write_tracking_compatible(void) | |
1504 | { | |
1505 | const uint64_t uffd_ioctls_mask = BIT(_UFFDIO_WRITEPROTECT); | |
1506 | int uffd_fd; | |
82ea3e3b | 1507 | RAMBlock *block; |
278e2f55 AG |
1508 | bool ret = false; |
1509 | ||
1510 | /* Open UFFD file descriptor */ | |
1511 | uffd_fd = uffd_create_fd(UFFD_FEATURE_PAGEFAULT_FLAG_WP, false); | |
1512 | if (uffd_fd < 0) { | |
1513 | return false; | |
1514 | } | |
1515 | ||
1516 | RCU_READ_LOCK_GUARD(); | |
1517 | ||
82ea3e3b | 1518 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
278e2f55 AG |
1519 | uint64_t uffd_ioctls; |
1520 | ||
1521 | /* Nothing to do with read-only and MMIO-writable regions */ | |
82ea3e3b | 1522 | if (block->mr->readonly || block->mr->rom_device) { |
278e2f55 AG |
1523 | continue; |
1524 | } | |
1525 | /* Try to register block memory via UFFD-IO to track writes */ | |
82ea3e3b | 1526 | if (uffd_register_memory(uffd_fd, block->host, block->max_length, |
278e2f55 AG |
1527 | UFFDIO_REGISTER_MODE_WP, &uffd_ioctls)) { |
1528 | goto out; | |
1529 | } | |
1530 | if ((uffd_ioctls & uffd_ioctls_mask) != uffd_ioctls_mask) { | |
1531 | goto out; | |
1532 | } | |
1533 | } | |
1534 | ret = true; | |
1535 | ||
1536 | out: | |
1537 | uffd_close_fd(uffd_fd); | |
1538 | return ret; | |
1539 | } | |
1540 | ||
eeccb99c AG |
1541 | /* |
1542 | * ram_block_populate_pages: populate memory in the RAM block by reading | |
1543 | * an integer from the beginning of each page. | |
1544 | * | |
1545 | * Since it's solely used for userfault_fd WP feature, here we just | |
1546 | * hardcode page size to qemu_real_host_page_size. | |
1547 | * | |
82ea3e3b | 1548 | * @block: RAM block to populate |
eeccb99c | 1549 | */ |
82ea3e3b | 1550 | static void ram_block_populate_pages(RAMBlock *block) |
eeccb99c | 1551 | { |
82ea3e3b | 1552 | char *ptr = (char *) block->host; |
eeccb99c | 1553 | |
82ea3e3b | 1554 | for (ram_addr_t offset = 0; offset < block->used_length; |
eeccb99c AG |
1555 | offset += qemu_real_host_page_size) { |
1556 | char tmp = *(ptr + offset); | |
1557 | ||
1558 | /* Don't optimize the read out */ | |
1559 | asm volatile("" : "+r" (tmp)); | |
1560 | } | |
1561 | } | |
1562 | ||
1563 | /* | |
1564 | * ram_write_tracking_prepare: prepare for UFFD-WP memory tracking | |
1565 | */ | |
1566 | void ram_write_tracking_prepare(void) | |
1567 | { | |
82ea3e3b | 1568 | RAMBlock *block; |
eeccb99c AG |
1569 | |
1570 | RCU_READ_LOCK_GUARD(); | |
1571 | ||
82ea3e3b | 1572 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
eeccb99c | 1573 | /* Nothing to do with read-only and MMIO-writable regions */ |
82ea3e3b | 1574 | if (block->mr->readonly || block->mr->rom_device) { |
eeccb99c AG |
1575 | continue; |
1576 | } | |
1577 | ||
1578 | /* | |
1579 | * Populate pages of the RAM block before enabling userfault_fd | |
1580 | * write protection. | |
1581 | * | |
1582 | * This stage is required since ioctl(UFFDIO_WRITEPROTECT) with | |
1583 | * UFFDIO_WRITEPROTECT_MODE_WP mode setting would silently skip | |
1584 | * pages with pte_none() entries in page table. | |
1585 | */ | |
82ea3e3b | 1586 | ram_block_populate_pages(block); |
eeccb99c AG |
1587 | } |
1588 | } | |
1589 | ||
278e2f55 AG |
1590 | /* |
1591 | * ram_write_tracking_start: start UFFD-WP memory tracking | |
1592 | * | |
1593 | * Returns 0 for success or negative value in case of error | |
1594 | */ | |
1595 | int ram_write_tracking_start(void) | |
1596 | { | |
1597 | int uffd_fd; | |
1598 | RAMState *rs = ram_state; | |
82ea3e3b | 1599 | RAMBlock *block; |
278e2f55 AG |
1600 | |
1601 | /* Open UFFD file descriptor */ | |
1602 | uffd_fd = uffd_create_fd(UFFD_FEATURE_PAGEFAULT_FLAG_WP, true); | |
1603 | if (uffd_fd < 0) { | |
1604 | return uffd_fd; | |
1605 | } | |
1606 | rs->uffdio_fd = uffd_fd; | |
1607 | ||
1608 | RCU_READ_LOCK_GUARD(); | |
1609 | ||
82ea3e3b | 1610 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
278e2f55 | 1611 | /* Nothing to do with read-only and MMIO-writable regions */ |
82ea3e3b | 1612 | if (block->mr->readonly || block->mr->rom_device) { |
278e2f55 AG |
1613 | continue; |
1614 | } | |
1615 | ||
1616 | /* Register block memory with UFFD to track writes */ | |
82ea3e3b AG |
1617 | if (uffd_register_memory(rs->uffdio_fd, block->host, |
1618 | block->max_length, UFFDIO_REGISTER_MODE_WP, NULL)) { | |
278e2f55 AG |
1619 | goto fail; |
1620 | } | |
1621 | /* Apply UFFD write protection to the block memory range */ | |
82ea3e3b AG |
1622 | if (uffd_change_protection(rs->uffdio_fd, block->host, |
1623 | block->max_length, true, false)) { | |
278e2f55 AG |
1624 | goto fail; |
1625 | } | |
82ea3e3b AG |
1626 | block->flags |= RAM_UF_WRITEPROTECT; |
1627 | memory_region_ref(block->mr); | |
278e2f55 | 1628 | |
82ea3e3b AG |
1629 | trace_ram_write_tracking_ramblock_start(block->idstr, block->page_size, |
1630 | block->host, block->max_length); | |
278e2f55 AG |
1631 | } |
1632 | ||
1633 | return 0; | |
1634 | ||
1635 | fail: | |
1636 | error_report("ram_write_tracking_start() failed: restoring initial memory state"); | |
1637 | ||
82ea3e3b AG |
1638 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
1639 | if ((block->flags & RAM_UF_WRITEPROTECT) == 0) { | |
278e2f55 AG |
1640 | continue; |
1641 | } | |
1642 | /* | |
1643 | * In case some memory block failed to be write-protected | |
1644 | * remove protection and unregister all succeeded RAM blocks | |
1645 | */ | |
82ea3e3b AG |
1646 | uffd_change_protection(rs->uffdio_fd, block->host, block->max_length, |
1647 | false, false); | |
1648 | uffd_unregister_memory(rs->uffdio_fd, block->host, block->max_length); | |
278e2f55 | 1649 | /* Cleanup flags and remove reference */ |
82ea3e3b AG |
1650 | block->flags &= ~RAM_UF_WRITEPROTECT; |
1651 | memory_region_unref(block->mr); | |
278e2f55 AG |
1652 | } |
1653 | ||
1654 | uffd_close_fd(uffd_fd); | |
1655 | rs->uffdio_fd = -1; | |
1656 | return -1; | |
1657 | } | |
1658 | ||
1659 | /** | |
1660 | * ram_write_tracking_stop: stop UFFD-WP memory tracking and remove protection | |
1661 | */ | |
1662 | void ram_write_tracking_stop(void) | |
1663 | { | |
1664 | RAMState *rs = ram_state; | |
82ea3e3b | 1665 | RAMBlock *block; |
278e2f55 AG |
1666 | |
1667 | RCU_READ_LOCK_GUARD(); | |
1668 | ||
82ea3e3b AG |
1669 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
1670 | if ((block->flags & RAM_UF_WRITEPROTECT) == 0) { | |
278e2f55 AG |
1671 | continue; |
1672 | } | |
1673 | /* Remove protection and unregister all affected RAM blocks */ | |
82ea3e3b AG |
1674 | uffd_change_protection(rs->uffdio_fd, block->host, block->max_length, |
1675 | false, false); | |
1676 | uffd_unregister_memory(rs->uffdio_fd, block->host, block->max_length); | |
278e2f55 | 1677 | |
82ea3e3b AG |
1678 | trace_ram_write_tracking_ramblock_stop(block->idstr, block->page_size, |
1679 | block->host, block->max_length); | |
278e2f55 AG |
1680 | |
1681 | /* Cleanup flags and remove reference */ | |
82ea3e3b AG |
1682 | block->flags &= ~RAM_UF_WRITEPROTECT; |
1683 | memory_region_unref(block->mr); | |
278e2f55 AG |
1684 | } |
1685 | ||
1686 | /* Finally close UFFD file descriptor */ | |
1687 | uffd_close_fd(rs->uffdio_fd); | |
1688 | rs->uffdio_fd = -1; | |
1689 | } | |
1690 | ||
1691 | #else | |
1692 | /* No target OS support, stubs just fail or ignore */ | |
1693 | ||
1694 | static RAMBlock *poll_fault_page(RAMState *rs, ram_addr_t *offset) | |
1695 | { | |
1696 | (void) rs; | |
1697 | (void) offset; | |
1698 | ||
1699 | return NULL; | |
1700 | } | |
1701 | ||
1702 | static int ram_save_release_protection(RAMState *rs, PageSearchStatus *pss, | |
1703 | unsigned long start_page) | |
1704 | { | |
1705 | (void) rs; | |
1706 | (void) pss; | |
1707 | (void) start_page; | |
1708 | ||
1709 | return 0; | |
1710 | } | |
1711 | ||
1712 | bool ram_write_tracking_available(void) | |
1713 | { | |
1714 | return false; | |
1715 | } | |
1716 | ||
1717 | bool ram_write_tracking_compatible(void) | |
1718 | { | |
1719 | assert(0); | |
1720 | return false; | |
1721 | } | |
1722 | ||
1723 | int ram_write_tracking_start(void) | |
1724 | { | |
1725 | assert(0); | |
1726 | return -1; | |
1727 | } | |
1728 | ||
1729 | void ram_write_tracking_stop(void) | |
1730 | { | |
1731 | assert(0); | |
1732 | } | |
1733 | #endif /* defined(__linux__) */ | |
1734 | ||
3d0684b2 | 1735 | /** |
ff1543af | 1736 | * get_queued_page: unqueue a page from the postcopy requests |
3d0684b2 JQ |
1737 | * |
1738 | * Skips pages that are already sent (!dirty) | |
a82d593b | 1739 | * |
a5f7b1a6 | 1740 | * Returns true if a queued page is found |
a82d593b | 1741 | * |
6f37bb8b | 1742 | * @rs: current RAM state |
3d0684b2 | 1743 | * @pss: data about the state of the current dirty page scan |
a82d593b | 1744 | */ |
f20e2865 | 1745 | static bool get_queued_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b DDAG |
1746 | { |
1747 | RAMBlock *block; | |
1748 | ram_addr_t offset; | |
1749 | bool dirty; | |
1750 | ||
1751 | do { | |
f20e2865 | 1752 | block = unqueue_page(rs, &offset); |
a82d593b DDAG |
1753 | /* |
1754 | * We're sending this page, and since it's postcopy nothing else | |
1755 | * will dirty it, and we must make sure it doesn't get sent again | |
1756 | * even if this queue request was received after the background | |
1757 | * search already sent it. | |
1758 | */ | |
1759 | if (block) { | |
f20e2865 JQ |
1760 | unsigned long page; |
1761 | ||
6b6712ef JQ |
1762 | page = offset >> TARGET_PAGE_BITS; |
1763 | dirty = test_bit(page, block->bmap); | |
a82d593b | 1764 | if (!dirty) { |
06b10688 | 1765 | trace_get_queued_page_not_dirty(block->idstr, (uint64_t)offset, |
64737606 | 1766 | page); |
a82d593b | 1767 | } else { |
f20e2865 | 1768 | trace_get_queued_page(block->idstr, (uint64_t)offset, page); |
a82d593b DDAG |
1769 | } |
1770 | } | |
1771 | ||
1772 | } while (block && !dirty); | |
1773 | ||
278e2f55 AG |
1774 | if (!block) { |
1775 | /* | |
1776 | * Poll write faults too if background snapshot is enabled; that's | |
1777 | * when we have vcpus got blocked by the write protected pages. | |
1778 | */ | |
1779 | block = poll_fault_page(rs, &offset); | |
1780 | } | |
1781 | ||
a82d593b | 1782 | if (block) { |
a82d593b DDAG |
1783 | /* |
1784 | * We want the background search to continue from the queued page | |
1785 | * since the guest is likely to want other pages near to the page | |
1786 | * it just requested. | |
1787 | */ | |
1788 | pss->block = block; | |
a935e30f | 1789 | pss->page = offset >> TARGET_PAGE_BITS; |
422314e7 WY |
1790 | |
1791 | /* | |
1792 | * This unqueued page would break the "one round" check, even is | |
1793 | * really rare. | |
1794 | */ | |
1795 | pss->complete_round = false; | |
a82d593b DDAG |
1796 | } |
1797 | ||
1798 | return !!block; | |
1799 | } | |
1800 | ||
6c595cde | 1801 | /** |
5e58f968 JQ |
1802 | * migration_page_queue_free: drop any remaining pages in the ram |
1803 | * request queue | |
6c595cde | 1804 | * |
3d0684b2 JQ |
1805 | * It should be empty at the end anyway, but in error cases there may |
1806 | * be some left. in case that there is any page left, we drop it. | |
1807 | * | |
6c595cde | 1808 | */ |
83c13382 | 1809 | static void migration_page_queue_free(RAMState *rs) |
6c595cde | 1810 | { |
ec481c6c | 1811 | struct RAMSrcPageRequest *mspr, *next_mspr; |
6c595cde DDAG |
1812 | /* This queue generally should be empty - but in the case of a failed |
1813 | * migration might have some droppings in. | |
1814 | */ | |
89ac5a1d | 1815 | RCU_READ_LOCK_GUARD(); |
ec481c6c | 1816 | QSIMPLEQ_FOREACH_SAFE(mspr, &rs->src_page_requests, next_req, next_mspr) { |
6c595cde | 1817 | memory_region_unref(mspr->rb->mr); |
ec481c6c | 1818 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
6c595cde DDAG |
1819 | g_free(mspr); |
1820 | } | |
6c595cde DDAG |
1821 | } |
1822 | ||
1823 | /** | |
3d0684b2 JQ |
1824 | * ram_save_queue_pages: queue the page for transmission |
1825 | * | |
1826 | * A request from postcopy destination for example. | |
1827 | * | |
1828 | * Returns zero on success or negative on error | |
1829 | * | |
3d0684b2 JQ |
1830 | * @rbname: Name of the RAMBLock of the request. NULL means the |
1831 | * same that last one. | |
1832 | * @start: starting address from the start of the RAMBlock | |
1833 | * @len: length (in bytes) to send | |
6c595cde | 1834 | */ |
96506894 | 1835 | int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len) |
6c595cde DDAG |
1836 | { |
1837 | RAMBlock *ramblock; | |
53518d94 | 1838 | RAMState *rs = ram_state; |
6c595cde | 1839 | |
9360447d | 1840 | ram_counters.postcopy_requests++; |
89ac5a1d DDAG |
1841 | RCU_READ_LOCK_GUARD(); |
1842 | ||
6c595cde DDAG |
1843 | if (!rbname) { |
1844 | /* Reuse last RAMBlock */ | |
68a098f3 | 1845 | ramblock = rs->last_req_rb; |
6c595cde DDAG |
1846 | |
1847 | if (!ramblock) { | |
1848 | /* | |
1849 | * Shouldn't happen, we can't reuse the last RAMBlock if | |
1850 | * it's the 1st request. | |
1851 | */ | |
1852 | error_report("ram_save_queue_pages no previous block"); | |
03acb4e9 | 1853 | return -1; |
6c595cde DDAG |
1854 | } |
1855 | } else { | |
1856 | ramblock = qemu_ram_block_by_name(rbname); | |
1857 | ||
1858 | if (!ramblock) { | |
1859 | /* We shouldn't be asked for a non-existent RAMBlock */ | |
1860 | error_report("ram_save_queue_pages no block '%s'", rbname); | |
03acb4e9 | 1861 | return -1; |
6c595cde | 1862 | } |
68a098f3 | 1863 | rs->last_req_rb = ramblock; |
6c595cde DDAG |
1864 | } |
1865 | trace_ram_save_queue_pages(ramblock->idstr, start, len); | |
395cb450 | 1866 | if (start + len > ramblock->used_length) { |
9458ad6b JQ |
1867 | error_report("%s request overrun start=" RAM_ADDR_FMT " len=" |
1868 | RAM_ADDR_FMT " blocklen=" RAM_ADDR_FMT, | |
6c595cde | 1869 | __func__, start, len, ramblock->used_length); |
03acb4e9 | 1870 | return -1; |
6c595cde DDAG |
1871 | } |
1872 | ||
ec481c6c JQ |
1873 | struct RAMSrcPageRequest *new_entry = |
1874 | g_malloc0(sizeof(struct RAMSrcPageRequest)); | |
6c595cde DDAG |
1875 | new_entry->rb = ramblock; |
1876 | new_entry->offset = start; | |
1877 | new_entry->len = len; | |
1878 | ||
1879 | memory_region_ref(ramblock->mr); | |
ec481c6c JQ |
1880 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1881 | QSIMPLEQ_INSERT_TAIL(&rs->src_page_requests, new_entry, next_req); | |
e03a34f8 | 1882 | migration_make_urgent_request(); |
ec481c6c | 1883 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
6c595cde DDAG |
1884 | |
1885 | return 0; | |
6c595cde DDAG |
1886 | } |
1887 | ||
d7400a34 XG |
1888 | static bool save_page_use_compression(RAMState *rs) |
1889 | { | |
1890 | if (!migrate_use_compression()) { | |
1891 | return false; | |
1892 | } | |
1893 | ||
1894 | /* | |
1a373522 DH |
1895 | * If xbzrle is enabled (e.g., after first round of migration), stop |
1896 | * using the data compression. In theory, xbzrle can do better than | |
1897 | * compression. | |
d7400a34 | 1898 | */ |
1a373522 DH |
1899 | if (rs->xbzrle_enabled) { |
1900 | return false; | |
d7400a34 XG |
1901 | } |
1902 | ||
1a373522 | 1903 | return true; |
d7400a34 XG |
1904 | } |
1905 | ||
5e5fdcff XG |
1906 | /* |
1907 | * try to compress the page before posting it out, return true if the page | |
1908 | * has been properly handled by compression, otherwise needs other | |
1909 | * paths to handle it | |
1910 | */ | |
1911 | static bool save_compress_page(RAMState *rs, RAMBlock *block, ram_addr_t offset) | |
1912 | { | |
1913 | if (!save_page_use_compression(rs)) { | |
1914 | return false; | |
1915 | } | |
1916 | ||
1917 | /* | |
1918 | * When starting the process of a new block, the first page of | |
1919 | * the block should be sent out before other pages in the same | |
1920 | * block, and all the pages in last block should have been sent | |
1921 | * out, keeping this order is important, because the 'cont' flag | |
1922 | * is used to avoid resending the block name. | |
1923 | * | |
1924 | * We post the fist page as normal page as compression will take | |
1925 | * much CPU resource. | |
1926 | */ | |
1927 | if (block != rs->last_sent_block) { | |
1928 | flush_compressed_data(rs); | |
1929 | return false; | |
1930 | } | |
1931 | ||
1932 | if (compress_page_with_multi_thread(rs, block, offset) > 0) { | |
1933 | return true; | |
1934 | } | |
1935 | ||
76e03000 | 1936 | compression_counters.busy++; |
5e5fdcff XG |
1937 | return false; |
1938 | } | |
1939 | ||
a82d593b | 1940 | /** |
3d0684b2 | 1941 | * ram_save_target_page: save one target page |
a82d593b | 1942 | * |
3d0684b2 | 1943 | * Returns the number of pages written |
a82d593b | 1944 | * |
6f37bb8b | 1945 | * @rs: current RAM state |
3d0684b2 | 1946 | * @pss: data about the page we want to send |
a82d593b | 1947 | * @last_stage: if we are at the completion stage |
a82d593b | 1948 | */ |
a0a8aa14 | 1949 | static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 1950 | bool last_stage) |
a82d593b | 1951 | { |
a8ec91f9 | 1952 | RAMBlock *block = pss->block; |
8bba004c | 1953 | ram_addr_t offset = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS; |
a8ec91f9 XG |
1954 | int res; |
1955 | ||
1956 | if (control_save_page(rs, block, offset, &res)) { | |
1957 | return res; | |
1958 | } | |
1959 | ||
5e5fdcff XG |
1960 | if (save_compress_page(rs, block, offset)) { |
1961 | return 1; | |
d7400a34 XG |
1962 | } |
1963 | ||
1964 | res = save_zero_page(rs, block, offset); | |
1965 | if (res > 0) { | |
1966 | /* Must let xbzrle know, otherwise a previous (now 0'd) cached | |
1967 | * page would be stale | |
1968 | */ | |
1969 | if (!save_page_use_compression(rs)) { | |
1970 | XBZRLE_cache_lock(); | |
1971 | xbzrle_cache_zero_page(rs, block->offset + offset); | |
1972 | XBZRLE_cache_unlock(); | |
1973 | } | |
1974 | ram_release_pages(block->idstr, offset, res); | |
1975 | return res; | |
1976 | } | |
1977 | ||
da3f56cb | 1978 | /* |
c6b3a2e0 WY |
1979 | * Do not use multifd for: |
1980 | * 1. Compression as the first page in the new block should be posted out | |
1981 | * before sending the compressed page | |
1982 | * 2. In postcopy as one whole host page should be placed | |
da3f56cb | 1983 | */ |
c6b3a2e0 WY |
1984 | if (!save_page_use_compression(rs) && migrate_use_multifd() |
1985 | && !migration_in_postcopy()) { | |
b9ee2f7d | 1986 | return ram_save_multifd_page(rs, block, offset); |
a82d593b DDAG |
1987 | } |
1988 | ||
1faa5665 | 1989 | return ram_save_page(rs, pss, last_stage); |
a82d593b DDAG |
1990 | } |
1991 | ||
1992 | /** | |
3d0684b2 | 1993 | * ram_save_host_page: save a whole host page |
a82d593b | 1994 | * |
3d0684b2 JQ |
1995 | * Starting at *offset send pages up to the end of the current host |
1996 | * page. It's valid for the initial offset to point into the middle of | |
1997 | * a host page in which case the remainder of the hostpage is sent. | |
1998 | * Only dirty target pages are sent. Note that the host page size may | |
1999 | * be a huge page for this block. | |
1eb3fc0a DDAG |
2000 | * The saving stops at the boundary of the used_length of the block |
2001 | * if the RAMBlock isn't a multiple of the host page size. | |
a82d593b | 2002 | * |
3d0684b2 JQ |
2003 | * Returns the number of pages written or negative on error |
2004 | * | |
6f37bb8b | 2005 | * @rs: current RAM state |
3d0684b2 | 2006 | * @ms: current migration state |
3d0684b2 | 2007 | * @pss: data about the page we want to send |
a82d593b | 2008 | * @last_stage: if we are at the completion stage |
a82d593b | 2009 | */ |
a0a8aa14 | 2010 | static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 2011 | bool last_stage) |
a82d593b DDAG |
2012 | { |
2013 | int tmppages, pages = 0; | |
a935e30f JQ |
2014 | size_t pagesize_bits = |
2015 | qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
ba1b7c81 KJ |
2016 | unsigned long hostpage_boundary = |
2017 | QEMU_ALIGN_UP(pss->page + 1, pagesize_bits); | |
278e2f55 AG |
2018 | unsigned long start_page = pss->page; |
2019 | int res; | |
4c011c37 | 2020 | |
fbd162e6 | 2021 | if (ramblock_is_ignored(pss->block)) { |
b895de50 CLG |
2022 | error_report("block %s should not be migrated !", pss->block->idstr); |
2023 | return 0; | |
2024 | } | |
2025 | ||
a82d593b | 2026 | do { |
1faa5665 | 2027 | /* Check the pages is dirty and if it is send it */ |
ba1b7c81 KJ |
2028 | if (migration_bitmap_clear_dirty(rs, pss->block, pss->page)) { |
2029 | tmppages = ram_save_target_page(rs, pss, last_stage); | |
2030 | if (tmppages < 0) { | |
2031 | return tmppages; | |
2032 | } | |
a82d593b | 2033 | |
ba1b7c81 KJ |
2034 | pages += tmppages; |
2035 | /* | |
2036 | * Allow rate limiting to happen in the middle of huge pages if | |
2037 | * something is sent in the current iteration. | |
2038 | */ | |
2039 | if (pagesize_bits > 1 && tmppages > 0) { | |
2040 | migration_rate_limit(); | |
2041 | } | |
23feba90 | 2042 | } |
ba1b7c81 KJ |
2043 | pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page); |
2044 | } while ((pss->page < hostpage_boundary) && | |
8bba004c AR |
2045 | offset_in_ramblock(pss->block, |
2046 | ((ram_addr_t)pss->page) << TARGET_PAGE_BITS)); | |
ba1b7c81 KJ |
2047 | /* The offset we leave with is the min boundary of host page and block */ |
2048 | pss->page = MIN(pss->page, hostpage_boundary) - 1; | |
278e2f55 AG |
2049 | |
2050 | res = ram_save_release_protection(rs, pss, start_page); | |
2051 | return (res < 0 ? res : pages); | |
a82d593b | 2052 | } |
6c595cde | 2053 | |
56e93d26 | 2054 | /** |
3d0684b2 | 2055 | * ram_find_and_save_block: finds a dirty page and sends it to f |
56e93d26 JQ |
2056 | * |
2057 | * Called within an RCU critical section. | |
2058 | * | |
e8f3735f XG |
2059 | * Returns the number of pages written where zero means no dirty pages, |
2060 | * or negative on error | |
56e93d26 | 2061 | * |
6f37bb8b | 2062 | * @rs: current RAM state |
56e93d26 | 2063 | * @last_stage: if we are at the completion stage |
a82d593b DDAG |
2064 | * |
2065 | * On systems where host-page-size > target-page-size it will send all the | |
2066 | * pages in a host page that are dirty. | |
56e93d26 JQ |
2067 | */ |
2068 | ||
ce25d337 | 2069 | static int ram_find_and_save_block(RAMState *rs, bool last_stage) |
56e93d26 | 2070 | { |
b8fb8cb7 | 2071 | PageSearchStatus pss; |
56e93d26 | 2072 | int pages = 0; |
b9e60928 | 2073 | bool again, found; |
56e93d26 | 2074 | |
0827b9e9 AA |
2075 | /* No dirty page as there is zero RAM */ |
2076 | if (!ram_bytes_total()) { | |
2077 | return pages; | |
2078 | } | |
2079 | ||
6f37bb8b | 2080 | pss.block = rs->last_seen_block; |
a935e30f | 2081 | pss.page = rs->last_page; |
b8fb8cb7 DDAG |
2082 | pss.complete_round = false; |
2083 | ||
2084 | if (!pss.block) { | |
2085 | pss.block = QLIST_FIRST_RCU(&ram_list.blocks); | |
2086 | } | |
56e93d26 | 2087 | |
b9e60928 | 2088 | do { |
a82d593b | 2089 | again = true; |
f20e2865 | 2090 | found = get_queued_page(rs, &pss); |
b9e60928 | 2091 | |
a82d593b DDAG |
2092 | if (!found) { |
2093 | /* priority queue empty, so just search for something dirty */ | |
f20e2865 | 2094 | found = find_dirty_block(rs, &pss, &again); |
a82d593b | 2095 | } |
f3f491fc | 2096 | |
a82d593b | 2097 | if (found) { |
f20e2865 | 2098 | pages = ram_save_host_page(rs, &pss, last_stage); |
56e93d26 | 2099 | } |
b9e60928 | 2100 | } while (!pages && again); |
56e93d26 | 2101 | |
6f37bb8b | 2102 | rs->last_seen_block = pss.block; |
a935e30f | 2103 | rs->last_page = pss.page; |
56e93d26 JQ |
2104 | |
2105 | return pages; | |
2106 | } | |
2107 | ||
2108 | void acct_update_position(QEMUFile *f, size_t size, bool zero) | |
2109 | { | |
2110 | uint64_t pages = size / TARGET_PAGE_SIZE; | |
f7ccd61b | 2111 | |
56e93d26 | 2112 | if (zero) { |
9360447d | 2113 | ram_counters.duplicate += pages; |
56e93d26 | 2114 | } else { |
9360447d JQ |
2115 | ram_counters.normal += pages; |
2116 | ram_counters.transferred += size; | |
56e93d26 JQ |
2117 | qemu_update_position(f, size); |
2118 | } | |
2119 | } | |
2120 | ||
fbd162e6 | 2121 | static uint64_t ram_bytes_total_common(bool count_ignored) |
56e93d26 JQ |
2122 | { |
2123 | RAMBlock *block; | |
2124 | uint64_t total = 0; | |
2125 | ||
89ac5a1d DDAG |
2126 | RCU_READ_LOCK_GUARD(); |
2127 | ||
fbd162e6 YK |
2128 | if (count_ignored) { |
2129 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
2130 | total += block->used_length; | |
2131 | } | |
2132 | } else { | |
2133 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
2134 | total += block->used_length; | |
2135 | } | |
99e15582 | 2136 | } |
56e93d26 JQ |
2137 | return total; |
2138 | } | |
2139 | ||
fbd162e6 YK |
2140 | uint64_t ram_bytes_total(void) |
2141 | { | |
2142 | return ram_bytes_total_common(false); | |
2143 | } | |
2144 | ||
f265e0e4 | 2145 | static void xbzrle_load_setup(void) |
56e93d26 | 2146 | { |
f265e0e4 | 2147 | XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE); |
56e93d26 JQ |
2148 | } |
2149 | ||
f265e0e4 JQ |
2150 | static void xbzrle_load_cleanup(void) |
2151 | { | |
2152 | g_free(XBZRLE.decoded_buf); | |
2153 | XBZRLE.decoded_buf = NULL; | |
2154 | } | |
2155 | ||
7d7c96be PX |
2156 | static void ram_state_cleanup(RAMState **rsp) |
2157 | { | |
b9ccaf6d DDAG |
2158 | if (*rsp) { |
2159 | migration_page_queue_free(*rsp); | |
2160 | qemu_mutex_destroy(&(*rsp)->bitmap_mutex); | |
2161 | qemu_mutex_destroy(&(*rsp)->src_page_req_mutex); | |
2162 | g_free(*rsp); | |
2163 | *rsp = NULL; | |
2164 | } | |
7d7c96be PX |
2165 | } |
2166 | ||
84593a08 PX |
2167 | static void xbzrle_cleanup(void) |
2168 | { | |
2169 | XBZRLE_cache_lock(); | |
2170 | if (XBZRLE.cache) { | |
2171 | cache_fini(XBZRLE.cache); | |
2172 | g_free(XBZRLE.encoded_buf); | |
2173 | g_free(XBZRLE.current_buf); | |
2174 | g_free(XBZRLE.zero_target_page); | |
2175 | XBZRLE.cache = NULL; | |
2176 | XBZRLE.encoded_buf = NULL; | |
2177 | XBZRLE.current_buf = NULL; | |
2178 | XBZRLE.zero_target_page = NULL; | |
2179 | } | |
2180 | XBZRLE_cache_unlock(); | |
2181 | } | |
2182 | ||
f265e0e4 | 2183 | static void ram_save_cleanup(void *opaque) |
56e93d26 | 2184 | { |
53518d94 | 2185 | RAMState **rsp = opaque; |
6b6712ef | 2186 | RAMBlock *block; |
eb859c53 | 2187 | |
278e2f55 AG |
2188 | /* We don't use dirty log with background snapshots */ |
2189 | if (!migrate_background_snapshot()) { | |
2190 | /* caller have hold iothread lock or is in a bh, so there is | |
2191 | * no writing race against the migration bitmap | |
2192 | */ | |
2193 | memory_global_dirty_log_stop(); | |
2194 | } | |
6b6712ef | 2195 | |
fbd162e6 | 2196 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
002cad6b PX |
2197 | g_free(block->clear_bmap); |
2198 | block->clear_bmap = NULL; | |
6b6712ef JQ |
2199 | g_free(block->bmap); |
2200 | block->bmap = NULL; | |
56e93d26 JQ |
2201 | } |
2202 | ||
84593a08 | 2203 | xbzrle_cleanup(); |
f0afa331 | 2204 | compress_threads_save_cleanup(); |
7d7c96be | 2205 | ram_state_cleanup(rsp); |
56e93d26 JQ |
2206 | } |
2207 | ||
6f37bb8b | 2208 | static void ram_state_reset(RAMState *rs) |
56e93d26 | 2209 | { |
6f37bb8b JQ |
2210 | rs->last_seen_block = NULL; |
2211 | rs->last_sent_block = NULL; | |
269ace29 | 2212 | rs->last_page = 0; |
6f37bb8b | 2213 | rs->last_version = ram_list.version; |
1a373522 | 2214 | rs->xbzrle_enabled = false; |
56e93d26 JQ |
2215 | } |
2216 | ||
2217 | #define MAX_WAIT 50 /* ms, half buffered_file limit */ | |
2218 | ||
4f2e4252 DDAG |
2219 | /* |
2220 | * 'expected' is the value you expect the bitmap mostly to be full | |
2221 | * of; it won't bother printing lines that are all this value. | |
2222 | * If 'todump' is null the migration bitmap is dumped. | |
2223 | */ | |
6b6712ef JQ |
2224 | void ram_debug_dump_bitmap(unsigned long *todump, bool expected, |
2225 | unsigned long pages) | |
4f2e4252 | 2226 | { |
4f2e4252 DDAG |
2227 | int64_t cur; |
2228 | int64_t linelen = 128; | |
2229 | char linebuf[129]; | |
2230 | ||
6b6712ef | 2231 | for (cur = 0; cur < pages; cur += linelen) { |
4f2e4252 DDAG |
2232 | int64_t curb; |
2233 | bool found = false; | |
2234 | /* | |
2235 | * Last line; catch the case where the line length | |
2236 | * is longer than remaining ram | |
2237 | */ | |
6b6712ef JQ |
2238 | if (cur + linelen > pages) { |
2239 | linelen = pages - cur; | |
4f2e4252 DDAG |
2240 | } |
2241 | for (curb = 0; curb < linelen; curb++) { | |
2242 | bool thisbit = test_bit(cur + curb, todump); | |
2243 | linebuf[curb] = thisbit ? '1' : '.'; | |
2244 | found = found || (thisbit != expected); | |
2245 | } | |
2246 | if (found) { | |
2247 | linebuf[curb] = '\0'; | |
2248 | fprintf(stderr, "0x%08" PRIx64 " : %s\n", cur, linebuf); | |
2249 | } | |
2250 | } | |
2251 | } | |
2252 | ||
e0b266f0 DDAG |
2253 | /* **** functions for postcopy ***** */ |
2254 | ||
ced1c616 PB |
2255 | void ram_postcopy_migrated_memory_release(MigrationState *ms) |
2256 | { | |
2257 | struct RAMBlock *block; | |
ced1c616 | 2258 | |
fbd162e6 | 2259 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
6b6712ef JQ |
2260 | unsigned long *bitmap = block->bmap; |
2261 | unsigned long range = block->used_length >> TARGET_PAGE_BITS; | |
2262 | unsigned long run_start = find_next_zero_bit(bitmap, range, 0); | |
ced1c616 PB |
2263 | |
2264 | while (run_start < range) { | |
2265 | unsigned long run_end = find_next_bit(bitmap, range, run_start + 1); | |
8bba004c AR |
2266 | ram_discard_range(block->idstr, |
2267 | ((ram_addr_t)run_start) << TARGET_PAGE_BITS, | |
2268 | ((ram_addr_t)(run_end - run_start)) | |
2269 | << TARGET_PAGE_BITS); | |
ced1c616 PB |
2270 | run_start = find_next_zero_bit(bitmap, range, run_end + 1); |
2271 | } | |
2272 | } | |
2273 | } | |
2274 | ||
3d0684b2 JQ |
2275 | /** |
2276 | * postcopy_send_discard_bm_ram: discard a RAMBlock | |
2277 | * | |
2278 | * Returns zero on success | |
2279 | * | |
e0b266f0 | 2280 | * Callback from postcopy_each_ram_send_discard for each RAMBlock |
3d0684b2 JQ |
2281 | * |
2282 | * @ms: current migration state | |
89dab31b | 2283 | * @block: RAMBlock to discard |
e0b266f0 | 2284 | */ |
810cf2bb | 2285 | static int postcopy_send_discard_bm_ram(MigrationState *ms, RAMBlock *block) |
e0b266f0 | 2286 | { |
6b6712ef | 2287 | unsigned long end = block->used_length >> TARGET_PAGE_BITS; |
e0b266f0 | 2288 | unsigned long current; |
1e7cf8c3 | 2289 | unsigned long *bitmap = block->bmap; |
e0b266f0 | 2290 | |
6b6712ef | 2291 | for (current = 0; current < end; ) { |
1e7cf8c3 | 2292 | unsigned long one = find_next_bit(bitmap, end, current); |
33a5cb62 | 2293 | unsigned long zero, discard_length; |
e0b266f0 | 2294 | |
33a5cb62 WY |
2295 | if (one >= end) { |
2296 | break; | |
2297 | } | |
e0b266f0 | 2298 | |
1e7cf8c3 | 2299 | zero = find_next_zero_bit(bitmap, end, one + 1); |
33a5cb62 WY |
2300 | |
2301 | if (zero >= end) { | |
2302 | discard_length = end - one; | |
e0b266f0 | 2303 | } else { |
33a5cb62 WY |
2304 | discard_length = zero - one; |
2305 | } | |
810cf2bb | 2306 | postcopy_discard_send_range(ms, one, discard_length); |
33a5cb62 | 2307 | current = one + discard_length; |
e0b266f0 DDAG |
2308 | } |
2309 | ||
2310 | return 0; | |
2311 | } | |
2312 | ||
3d0684b2 JQ |
2313 | /** |
2314 | * postcopy_each_ram_send_discard: discard all RAMBlocks | |
2315 | * | |
2316 | * Returns 0 for success or negative for error | |
2317 | * | |
e0b266f0 DDAG |
2318 | * Utility for the outgoing postcopy code. |
2319 | * Calls postcopy_send_discard_bm_ram for each RAMBlock | |
2320 | * passing it bitmap indexes and name. | |
e0b266f0 DDAG |
2321 | * (qemu_ram_foreach_block ends up passing unscaled lengths |
2322 | * which would mean postcopy code would have to deal with target page) | |
3d0684b2 JQ |
2323 | * |
2324 | * @ms: current migration state | |
e0b266f0 DDAG |
2325 | */ |
2326 | static int postcopy_each_ram_send_discard(MigrationState *ms) | |
2327 | { | |
2328 | struct RAMBlock *block; | |
2329 | int ret; | |
2330 | ||
fbd162e6 | 2331 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
810cf2bb | 2332 | postcopy_discard_send_init(ms, block->idstr); |
e0b266f0 DDAG |
2333 | |
2334 | /* | |
2335 | * Postcopy sends chunks of bitmap over the wire, but it | |
2336 | * just needs indexes at this point, avoids it having | |
2337 | * target page specific code. | |
2338 | */ | |
810cf2bb WY |
2339 | ret = postcopy_send_discard_bm_ram(ms, block); |
2340 | postcopy_discard_send_finish(ms); | |
e0b266f0 DDAG |
2341 | if (ret) { |
2342 | return ret; | |
2343 | } | |
2344 | } | |
2345 | ||
2346 | return 0; | |
2347 | } | |
2348 | ||
3d0684b2 | 2349 | /** |
8324ef86 | 2350 | * postcopy_chunk_hostpages_pass: canonicalize bitmap in hostpages |
3d0684b2 JQ |
2351 | * |
2352 | * Helper for postcopy_chunk_hostpages; it's called twice to | |
2353 | * canonicalize the two bitmaps, that are similar, but one is | |
2354 | * inverted. | |
99e314eb | 2355 | * |
3d0684b2 JQ |
2356 | * Postcopy requires that all target pages in a hostpage are dirty or |
2357 | * clean, not a mix. This function canonicalizes the bitmaps. | |
99e314eb | 2358 | * |
3d0684b2 | 2359 | * @ms: current migration state |
3d0684b2 | 2360 | * @block: block that contains the page we want to canonicalize |
99e314eb | 2361 | */ |
1e7cf8c3 | 2362 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, RAMBlock *block) |
99e314eb | 2363 | { |
53518d94 | 2364 | RAMState *rs = ram_state; |
6b6712ef | 2365 | unsigned long *bitmap = block->bmap; |
29c59172 | 2366 | unsigned int host_ratio = block->page_size / TARGET_PAGE_SIZE; |
6b6712ef | 2367 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
99e314eb DDAG |
2368 | unsigned long run_start; |
2369 | ||
29c59172 DDAG |
2370 | if (block->page_size == TARGET_PAGE_SIZE) { |
2371 | /* Easy case - TPS==HPS for a non-huge page RAMBlock */ | |
2372 | return; | |
2373 | } | |
2374 | ||
1e7cf8c3 WY |
2375 | /* Find a dirty page */ |
2376 | run_start = find_next_bit(bitmap, pages, 0); | |
99e314eb | 2377 | |
6b6712ef | 2378 | while (run_start < pages) { |
99e314eb DDAG |
2379 | |
2380 | /* | |
2381 | * If the start of this run of pages is in the middle of a host | |
2382 | * page, then we need to fixup this host page. | |
2383 | */ | |
9dec3cc3 | 2384 | if (QEMU_IS_ALIGNED(run_start, host_ratio)) { |
99e314eb | 2385 | /* Find the end of this run */ |
1e7cf8c3 | 2386 | run_start = find_next_zero_bit(bitmap, pages, run_start + 1); |
99e314eb DDAG |
2387 | /* |
2388 | * If the end isn't at the start of a host page, then the | |
2389 | * run doesn't finish at the end of a host page | |
2390 | * and we need to discard. | |
2391 | */ | |
99e314eb DDAG |
2392 | } |
2393 | ||
9dec3cc3 | 2394 | if (!QEMU_IS_ALIGNED(run_start, host_ratio)) { |
99e314eb | 2395 | unsigned long page; |
dad45ab2 WY |
2396 | unsigned long fixup_start_addr = QEMU_ALIGN_DOWN(run_start, |
2397 | host_ratio); | |
2398 | run_start = QEMU_ALIGN_UP(run_start, host_ratio); | |
99e314eb | 2399 | |
99e314eb DDAG |
2400 | /* Clean up the bitmap */ |
2401 | for (page = fixup_start_addr; | |
2402 | page < fixup_start_addr + host_ratio; page++) { | |
99e314eb DDAG |
2403 | /* |
2404 | * Remark them as dirty, updating the count for any pages | |
2405 | * that weren't previously dirty. | |
2406 | */ | |
0d8ec885 | 2407 | rs->migration_dirty_pages += !test_and_set_bit(page, bitmap); |
99e314eb DDAG |
2408 | } |
2409 | } | |
2410 | ||
1e7cf8c3 WY |
2411 | /* Find the next dirty page for the next iteration */ |
2412 | run_start = find_next_bit(bitmap, pages, run_start); | |
99e314eb DDAG |
2413 | } |
2414 | } | |
2415 | ||
3d0684b2 | 2416 | /** |
89dab31b | 2417 | * postcopy_chunk_hostpages: discard any partially sent host page |
3d0684b2 | 2418 | * |
99e314eb DDAG |
2419 | * Utility for the outgoing postcopy code. |
2420 | * | |
2421 | * Discard any partially sent host-page size chunks, mark any partially | |
29c59172 DDAG |
2422 | * dirty host-page size chunks as all dirty. In this case the host-page |
2423 | * is the host-page for the particular RAMBlock, i.e. it might be a huge page | |
99e314eb | 2424 | * |
3d0684b2 JQ |
2425 | * Returns zero on success |
2426 | * | |
2427 | * @ms: current migration state | |
6b6712ef | 2428 | * @block: block we want to work with |
99e314eb | 2429 | */ |
6b6712ef | 2430 | static int postcopy_chunk_hostpages(MigrationState *ms, RAMBlock *block) |
99e314eb | 2431 | { |
810cf2bb | 2432 | postcopy_discard_send_init(ms, block->idstr); |
99e314eb | 2433 | |
6b6712ef | 2434 | /* |
1e7cf8c3 | 2435 | * Ensure that all partially dirty host pages are made fully dirty. |
6b6712ef | 2436 | */ |
1e7cf8c3 | 2437 | postcopy_chunk_hostpages_pass(ms, block); |
99e314eb | 2438 | |
810cf2bb | 2439 | postcopy_discard_send_finish(ms); |
99e314eb DDAG |
2440 | return 0; |
2441 | } | |
2442 | ||
3d0684b2 JQ |
2443 | /** |
2444 | * ram_postcopy_send_discard_bitmap: transmit the discard bitmap | |
2445 | * | |
2446 | * Returns zero on success | |
2447 | * | |
e0b266f0 DDAG |
2448 | * Transmit the set of pages to be discarded after precopy to the target |
2449 | * these are pages that: | |
2450 | * a) Have been previously transmitted but are now dirty again | |
2451 | * b) Pages that have never been transmitted, this ensures that | |
2452 | * any pages on the destination that have been mapped by background | |
2453 | * tasks get discarded (transparent huge pages is the specific concern) | |
2454 | * Hopefully this is pretty sparse | |
3d0684b2 JQ |
2455 | * |
2456 | * @ms: current migration state | |
e0b266f0 DDAG |
2457 | */ |
2458 | int ram_postcopy_send_discard_bitmap(MigrationState *ms) | |
2459 | { | |
53518d94 | 2460 | RAMState *rs = ram_state; |
6b6712ef | 2461 | RAMBlock *block; |
e0b266f0 | 2462 | int ret; |
e0b266f0 | 2463 | |
89ac5a1d | 2464 | RCU_READ_LOCK_GUARD(); |
e0b266f0 DDAG |
2465 | |
2466 | /* This should be our last sync, the src is now paused */ | |
eb859c53 | 2467 | migration_bitmap_sync(rs); |
e0b266f0 | 2468 | |
6b6712ef JQ |
2469 | /* Easiest way to make sure we don't resume in the middle of a host-page */ |
2470 | rs->last_seen_block = NULL; | |
2471 | rs->last_sent_block = NULL; | |
2472 | rs->last_page = 0; | |
e0b266f0 | 2473 | |
fbd162e6 | 2474 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
6b6712ef JQ |
2475 | /* Deal with TPS != HPS and huge pages */ |
2476 | ret = postcopy_chunk_hostpages(ms, block); | |
2477 | if (ret) { | |
6b6712ef JQ |
2478 | return ret; |
2479 | } | |
e0b266f0 | 2480 | |
e0b266f0 | 2481 | #ifdef DEBUG_POSTCOPY |
1e7cf8c3 WY |
2482 | ram_debug_dump_bitmap(block->bmap, true, |
2483 | block->used_length >> TARGET_PAGE_BITS); | |
e0b266f0 | 2484 | #endif |
6b6712ef JQ |
2485 | } |
2486 | trace_ram_postcopy_send_discard_bitmap(); | |
e0b266f0 | 2487 | |
b3ac2b94 | 2488 | return postcopy_each_ram_send_discard(ms); |
e0b266f0 DDAG |
2489 | } |
2490 | ||
3d0684b2 JQ |
2491 | /** |
2492 | * ram_discard_range: discard dirtied pages at the beginning of postcopy | |
e0b266f0 | 2493 | * |
3d0684b2 | 2494 | * Returns zero on success |
e0b266f0 | 2495 | * |
36449157 JQ |
2496 | * @rbname: name of the RAMBlock of the request. NULL means the |
2497 | * same that last one. | |
3d0684b2 JQ |
2498 | * @start: RAMBlock starting page |
2499 | * @length: RAMBlock size | |
e0b266f0 | 2500 | */ |
aaa2064c | 2501 | int ram_discard_range(const char *rbname, uint64_t start, size_t length) |
e0b266f0 | 2502 | { |
36449157 | 2503 | trace_ram_discard_range(rbname, start, length); |
d3a5038c | 2504 | |
89ac5a1d | 2505 | RCU_READ_LOCK_GUARD(); |
36449157 | 2506 | RAMBlock *rb = qemu_ram_block_by_name(rbname); |
e0b266f0 DDAG |
2507 | |
2508 | if (!rb) { | |
36449157 | 2509 | error_report("ram_discard_range: Failed to find block '%s'", rbname); |
03acb4e9 | 2510 | return -1; |
e0b266f0 DDAG |
2511 | } |
2512 | ||
814bb08f PX |
2513 | /* |
2514 | * On source VM, we don't need to update the received bitmap since | |
2515 | * we don't even have one. | |
2516 | */ | |
2517 | if (rb->receivedmap) { | |
2518 | bitmap_clear(rb->receivedmap, start >> qemu_target_page_bits(), | |
2519 | length >> qemu_target_page_bits()); | |
2520 | } | |
2521 | ||
03acb4e9 | 2522 | return ram_block_discard_range(rb, start, length); |
e0b266f0 DDAG |
2523 | } |
2524 | ||
84593a08 PX |
2525 | /* |
2526 | * For every allocation, we will try not to crash the VM if the | |
2527 | * allocation failed. | |
2528 | */ | |
2529 | static int xbzrle_init(void) | |
2530 | { | |
2531 | Error *local_err = NULL; | |
2532 | ||
2533 | if (!migrate_use_xbzrle()) { | |
2534 | return 0; | |
2535 | } | |
2536 | ||
2537 | XBZRLE_cache_lock(); | |
2538 | ||
2539 | XBZRLE.zero_target_page = g_try_malloc0(TARGET_PAGE_SIZE); | |
2540 | if (!XBZRLE.zero_target_page) { | |
2541 | error_report("%s: Error allocating zero page", __func__); | |
2542 | goto err_out; | |
2543 | } | |
2544 | ||
2545 | XBZRLE.cache = cache_init(migrate_xbzrle_cache_size(), | |
2546 | TARGET_PAGE_SIZE, &local_err); | |
2547 | if (!XBZRLE.cache) { | |
2548 | error_report_err(local_err); | |
2549 | goto free_zero_page; | |
2550 | } | |
2551 | ||
2552 | XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); | |
2553 | if (!XBZRLE.encoded_buf) { | |
2554 | error_report("%s: Error allocating encoded_buf", __func__); | |
2555 | goto free_cache; | |
2556 | } | |
2557 | ||
2558 | XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); | |
2559 | if (!XBZRLE.current_buf) { | |
2560 | error_report("%s: Error allocating current_buf", __func__); | |
2561 | goto free_encoded_buf; | |
2562 | } | |
2563 | ||
2564 | /* We are all good */ | |
2565 | XBZRLE_cache_unlock(); | |
2566 | return 0; | |
2567 | ||
2568 | free_encoded_buf: | |
2569 | g_free(XBZRLE.encoded_buf); | |
2570 | XBZRLE.encoded_buf = NULL; | |
2571 | free_cache: | |
2572 | cache_fini(XBZRLE.cache); | |
2573 | XBZRLE.cache = NULL; | |
2574 | free_zero_page: | |
2575 | g_free(XBZRLE.zero_target_page); | |
2576 | XBZRLE.zero_target_page = NULL; | |
2577 | err_out: | |
2578 | XBZRLE_cache_unlock(); | |
2579 | return -ENOMEM; | |
2580 | } | |
2581 | ||
53518d94 | 2582 | static int ram_state_init(RAMState **rsp) |
56e93d26 | 2583 | { |
7d00ee6a PX |
2584 | *rsp = g_try_new0(RAMState, 1); |
2585 | ||
2586 | if (!*rsp) { | |
2587 | error_report("%s: Init ramstate fail", __func__); | |
2588 | return -1; | |
2589 | } | |
53518d94 JQ |
2590 | |
2591 | qemu_mutex_init(&(*rsp)->bitmap_mutex); | |
2592 | qemu_mutex_init(&(*rsp)->src_page_req_mutex); | |
2593 | QSIMPLEQ_INIT(&(*rsp)->src_page_requests); | |
56e93d26 | 2594 | |
7d00ee6a | 2595 | /* |
40c4d4a8 IR |
2596 | * Count the total number of pages used by ram blocks not including any |
2597 | * gaps due to alignment or unplugs. | |
03158519 | 2598 | * This must match with the initial values of dirty bitmap. |
7d00ee6a | 2599 | */ |
40c4d4a8 | 2600 | (*rsp)->migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS; |
7d00ee6a PX |
2601 | ram_state_reset(*rsp); |
2602 | ||
2603 | return 0; | |
2604 | } | |
2605 | ||
d6eff5d7 | 2606 | static void ram_list_init_bitmaps(void) |
7d00ee6a | 2607 | { |
002cad6b | 2608 | MigrationState *ms = migrate_get_current(); |
d6eff5d7 PX |
2609 | RAMBlock *block; |
2610 | unsigned long pages; | |
002cad6b | 2611 | uint8_t shift; |
56e93d26 | 2612 | |
0827b9e9 AA |
2613 | /* Skip setting bitmap if there is no RAM */ |
2614 | if (ram_bytes_total()) { | |
002cad6b PX |
2615 | shift = ms->clear_bitmap_shift; |
2616 | if (shift > CLEAR_BITMAP_SHIFT_MAX) { | |
2617 | error_report("clear_bitmap_shift (%u) too big, using " | |
2618 | "max value (%u)", shift, CLEAR_BITMAP_SHIFT_MAX); | |
2619 | shift = CLEAR_BITMAP_SHIFT_MAX; | |
2620 | } else if (shift < CLEAR_BITMAP_SHIFT_MIN) { | |
2621 | error_report("clear_bitmap_shift (%u) too small, using " | |
2622 | "min value (%u)", shift, CLEAR_BITMAP_SHIFT_MIN); | |
2623 | shift = CLEAR_BITMAP_SHIFT_MIN; | |
2624 | } | |
2625 | ||
fbd162e6 | 2626 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
d6eff5d7 | 2627 | pages = block->max_length >> TARGET_PAGE_BITS; |
03158519 WY |
2628 | /* |
2629 | * The initial dirty bitmap for migration must be set with all | |
2630 | * ones to make sure we'll migrate every guest RAM page to | |
2631 | * destination. | |
40c4d4a8 IR |
2632 | * Here we set RAMBlock.bmap all to 1 because when rebegin a |
2633 | * new migration after a failed migration, ram_list. | |
2634 | * dirty_memory[DIRTY_MEMORY_MIGRATION] don't include the whole | |
2635 | * guest memory. | |
03158519 | 2636 | */ |
6b6712ef | 2637 | block->bmap = bitmap_new(pages); |
40c4d4a8 | 2638 | bitmap_set(block->bmap, 0, pages); |
002cad6b PX |
2639 | block->clear_bmap_shift = shift; |
2640 | block->clear_bmap = bitmap_new(clear_bmap_size(pages, shift)); | |
0827b9e9 | 2641 | } |
f3f491fc | 2642 | } |
d6eff5d7 PX |
2643 | } |
2644 | ||
2645 | static void ram_init_bitmaps(RAMState *rs) | |
2646 | { | |
2647 | /* For memory_global_dirty_log_start below. */ | |
2648 | qemu_mutex_lock_iothread(); | |
2649 | qemu_mutex_lock_ramlist(); | |
f3f491fc | 2650 | |
89ac5a1d DDAG |
2651 | WITH_RCU_READ_LOCK_GUARD() { |
2652 | ram_list_init_bitmaps(); | |
278e2f55 AG |
2653 | /* We don't use dirty log with background snapshots */ |
2654 | if (!migrate_background_snapshot()) { | |
2655 | memory_global_dirty_log_start(); | |
2656 | migration_bitmap_sync_precopy(rs); | |
2657 | } | |
89ac5a1d | 2658 | } |
56e93d26 | 2659 | qemu_mutex_unlock_ramlist(); |
49877834 | 2660 | qemu_mutex_unlock_iothread(); |
d6eff5d7 PX |
2661 | } |
2662 | ||
2663 | static int ram_init_all(RAMState **rsp) | |
2664 | { | |
2665 | if (ram_state_init(rsp)) { | |
2666 | return -1; | |
2667 | } | |
2668 | ||
2669 | if (xbzrle_init()) { | |
2670 | ram_state_cleanup(rsp); | |
2671 | return -1; | |
2672 | } | |
2673 | ||
2674 | ram_init_bitmaps(*rsp); | |
a91246c9 HZ |
2675 | |
2676 | return 0; | |
2677 | } | |
2678 | ||
08614f34 PX |
2679 | static void ram_state_resume_prepare(RAMState *rs, QEMUFile *out) |
2680 | { | |
2681 | RAMBlock *block; | |
2682 | uint64_t pages = 0; | |
2683 | ||
2684 | /* | |
2685 | * Postcopy is not using xbzrle/compression, so no need for that. | |
2686 | * Also, since source are already halted, we don't need to care | |
2687 | * about dirty page logging as well. | |
2688 | */ | |
2689 | ||
fbd162e6 | 2690 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
08614f34 PX |
2691 | pages += bitmap_count_one(block->bmap, |
2692 | block->used_length >> TARGET_PAGE_BITS); | |
2693 | } | |
2694 | ||
2695 | /* This may not be aligned with current bitmaps. Recalculate. */ | |
2696 | rs->migration_dirty_pages = pages; | |
2697 | ||
1a373522 | 2698 | ram_state_reset(rs); |
08614f34 PX |
2699 | |
2700 | /* Update RAMState cache of output QEMUFile */ | |
2701 | rs->f = out; | |
2702 | ||
2703 | trace_ram_state_resume_prepare(pages); | |
2704 | } | |
2705 | ||
6bcb05fc WW |
2706 | /* |
2707 | * This function clears bits of the free pages reported by the caller from the | |
2708 | * migration dirty bitmap. @addr is the host address corresponding to the | |
2709 | * start of the continuous guest free pages, and @len is the total bytes of | |
2710 | * those pages. | |
2711 | */ | |
2712 | void qemu_guest_free_page_hint(void *addr, size_t len) | |
2713 | { | |
2714 | RAMBlock *block; | |
2715 | ram_addr_t offset; | |
2716 | size_t used_len, start, npages; | |
2717 | MigrationState *s = migrate_get_current(); | |
2718 | ||
2719 | /* This function is currently expected to be used during live migration */ | |
2720 | if (!migration_is_setup_or_active(s->state)) { | |
2721 | return; | |
2722 | } | |
2723 | ||
2724 | for (; len > 0; len -= used_len, addr += used_len) { | |
2725 | block = qemu_ram_block_from_host(addr, false, &offset); | |
2726 | if (unlikely(!block || offset >= block->used_length)) { | |
2727 | /* | |
2728 | * The implementation might not support RAMBlock resize during | |
2729 | * live migration, but it could happen in theory with future | |
2730 | * updates. So we add a check here to capture that case. | |
2731 | */ | |
2732 | error_report_once("%s unexpected error", __func__); | |
2733 | return; | |
2734 | } | |
2735 | ||
2736 | if (len <= block->used_length - offset) { | |
2737 | used_len = len; | |
2738 | } else { | |
2739 | used_len = block->used_length - offset; | |
2740 | } | |
2741 | ||
2742 | start = offset >> TARGET_PAGE_BITS; | |
2743 | npages = used_len >> TARGET_PAGE_BITS; | |
2744 | ||
2745 | qemu_mutex_lock(&ram_state->bitmap_mutex); | |
2746 | ram_state->migration_dirty_pages -= | |
2747 | bitmap_count_one_with_offset(block->bmap, start, npages); | |
2748 | bitmap_clear(block->bmap, start, npages); | |
2749 | qemu_mutex_unlock(&ram_state->bitmap_mutex); | |
2750 | } | |
2751 | } | |
2752 | ||
3d0684b2 JQ |
2753 | /* |
2754 | * Each of ram_save_setup, ram_save_iterate and ram_save_complete has | |
a91246c9 HZ |
2755 | * long-running RCU critical section. When rcu-reclaims in the code |
2756 | * start to become numerous it will be necessary to reduce the | |
2757 | * granularity of these critical sections. | |
2758 | */ | |
2759 | ||
3d0684b2 JQ |
2760 | /** |
2761 | * ram_save_setup: Setup RAM for migration | |
2762 | * | |
2763 | * Returns zero to indicate success and negative for error | |
2764 | * | |
2765 | * @f: QEMUFile where to send the data | |
2766 | * @opaque: RAMState pointer | |
2767 | */ | |
a91246c9 HZ |
2768 | static int ram_save_setup(QEMUFile *f, void *opaque) |
2769 | { | |
53518d94 | 2770 | RAMState **rsp = opaque; |
a91246c9 HZ |
2771 | RAMBlock *block; |
2772 | ||
dcaf446e XG |
2773 | if (compress_threads_save_setup()) { |
2774 | return -1; | |
2775 | } | |
2776 | ||
a91246c9 HZ |
2777 | /* migration has already setup the bitmap, reuse it. */ |
2778 | if (!migration_in_colo_state()) { | |
7d00ee6a | 2779 | if (ram_init_all(rsp) != 0) { |
dcaf446e | 2780 | compress_threads_save_cleanup(); |
a91246c9 | 2781 | return -1; |
53518d94 | 2782 | } |
a91246c9 | 2783 | } |
53518d94 | 2784 | (*rsp)->f = f; |
a91246c9 | 2785 | |
0e6ebd48 DDAG |
2786 | WITH_RCU_READ_LOCK_GUARD() { |
2787 | qemu_put_be64(f, ram_bytes_total_common(true) | RAM_SAVE_FLAG_MEM_SIZE); | |
56e93d26 | 2788 | |
0e6ebd48 DDAG |
2789 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
2790 | qemu_put_byte(f, strlen(block->idstr)); | |
2791 | qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); | |
2792 | qemu_put_be64(f, block->used_length); | |
2793 | if (migrate_postcopy_ram() && block->page_size != | |
2794 | qemu_host_page_size) { | |
2795 | qemu_put_be64(f, block->page_size); | |
2796 | } | |
2797 | if (migrate_ignore_shared()) { | |
2798 | qemu_put_be64(f, block->mr->addr); | |
2799 | } | |
fbd162e6 | 2800 | } |
56e93d26 JQ |
2801 | } |
2802 | ||
56e93d26 JQ |
2803 | ram_control_before_iterate(f, RAM_CONTROL_SETUP); |
2804 | ram_control_after_iterate(f, RAM_CONTROL_SETUP); | |
2805 | ||
99f2c6fb | 2806 | multifd_send_sync_main(f); |
56e93d26 | 2807 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
35374cbd | 2808 | qemu_fflush(f); |
56e93d26 JQ |
2809 | |
2810 | return 0; | |
2811 | } | |
2812 | ||
3d0684b2 JQ |
2813 | /** |
2814 | * ram_save_iterate: iterative stage for migration | |
2815 | * | |
2816 | * Returns zero to indicate success and negative for error | |
2817 | * | |
2818 | * @f: QEMUFile where to send the data | |
2819 | * @opaque: RAMState pointer | |
2820 | */ | |
56e93d26 JQ |
2821 | static int ram_save_iterate(QEMUFile *f, void *opaque) |
2822 | { | |
53518d94 JQ |
2823 | RAMState **temp = opaque; |
2824 | RAMState *rs = *temp; | |
3d4095b2 | 2825 | int ret = 0; |
56e93d26 JQ |
2826 | int i; |
2827 | int64_t t0; | |
5c90308f | 2828 | int done = 0; |
56e93d26 | 2829 | |
b2557345 PL |
2830 | if (blk_mig_bulk_active()) { |
2831 | /* Avoid transferring ram during bulk phase of block migration as | |
2832 | * the bulk phase will usually take a long time and transferring | |
2833 | * ram updates during that time is pointless. */ | |
2834 | goto out; | |
2835 | } | |
2836 | ||
89ac5a1d DDAG |
2837 | WITH_RCU_READ_LOCK_GUARD() { |
2838 | if (ram_list.version != rs->last_version) { | |
2839 | ram_state_reset(rs); | |
2840 | } | |
56e93d26 | 2841 | |
89ac5a1d DDAG |
2842 | /* Read version before ram_list.blocks */ |
2843 | smp_rmb(); | |
56e93d26 | 2844 | |
89ac5a1d | 2845 | ram_control_before_iterate(f, RAM_CONTROL_ROUND); |
56e93d26 | 2846 | |
89ac5a1d DDAG |
2847 | t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
2848 | i = 0; | |
2849 | while ((ret = qemu_file_rate_limit(f)) == 0 || | |
2850 | !QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
2851 | int pages; | |
e03a34f8 | 2852 | |
89ac5a1d DDAG |
2853 | if (qemu_file_get_error(f)) { |
2854 | break; | |
2855 | } | |
e8f3735f | 2856 | |
89ac5a1d DDAG |
2857 | pages = ram_find_and_save_block(rs, false); |
2858 | /* no more pages to sent */ | |
2859 | if (pages == 0) { | |
2860 | done = 1; | |
2861 | break; | |
2862 | } | |
e8f3735f | 2863 | |
89ac5a1d DDAG |
2864 | if (pages < 0) { |
2865 | qemu_file_set_error(f, pages); | |
56e93d26 JQ |
2866 | break; |
2867 | } | |
89ac5a1d DDAG |
2868 | |
2869 | rs->target_page_count += pages; | |
2870 | ||
644acf99 WY |
2871 | /* |
2872 | * During postcopy, it is necessary to make sure one whole host | |
2873 | * page is sent in one chunk. | |
2874 | */ | |
2875 | if (migrate_postcopy_ram()) { | |
2876 | flush_compressed_data(rs); | |
2877 | } | |
2878 | ||
89ac5a1d DDAG |
2879 | /* |
2880 | * we want to check in the 1st loop, just in case it was the 1st | |
2881 | * time and we had to sync the dirty bitmap. | |
2882 | * qemu_clock_get_ns() is a bit expensive, so we only check each | |
2883 | * some iterations | |
2884 | */ | |
2885 | if ((i & 63) == 0) { | |
2886 | uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / | |
2887 | 1000000; | |
2888 | if (t1 > MAX_WAIT) { | |
2889 | trace_ram_save_iterate_big_wait(t1, i); | |
2890 | break; | |
2891 | } | |
2892 | } | |
2893 | i++; | |
56e93d26 | 2894 | } |
56e93d26 | 2895 | } |
56e93d26 JQ |
2896 | |
2897 | /* | |
2898 | * Must occur before EOS (or any QEMUFile operation) | |
2899 | * because of RDMA protocol. | |
2900 | */ | |
2901 | ram_control_after_iterate(f, RAM_CONTROL_ROUND); | |
2902 | ||
b2557345 | 2903 | out: |
b69a0227 JQ |
2904 | if (ret >= 0 |
2905 | && migration_is_setup_or_active(migrate_get_current()->state)) { | |
99f2c6fb | 2906 | multifd_send_sync_main(rs->f); |
3d4095b2 JQ |
2907 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
2908 | qemu_fflush(f); | |
2909 | ram_counters.transferred += 8; | |
56e93d26 | 2910 | |
3d4095b2 JQ |
2911 | ret = qemu_file_get_error(f); |
2912 | } | |
56e93d26 JQ |
2913 | if (ret < 0) { |
2914 | return ret; | |
2915 | } | |
2916 | ||
5c90308f | 2917 | return done; |
56e93d26 JQ |
2918 | } |
2919 | ||
3d0684b2 JQ |
2920 | /** |
2921 | * ram_save_complete: function called to send the remaining amount of ram | |
2922 | * | |
e8f3735f | 2923 | * Returns zero to indicate success or negative on error |
3d0684b2 JQ |
2924 | * |
2925 | * Called with iothread lock | |
2926 | * | |
2927 | * @f: QEMUFile where to send the data | |
2928 | * @opaque: RAMState pointer | |
2929 | */ | |
56e93d26 JQ |
2930 | static int ram_save_complete(QEMUFile *f, void *opaque) |
2931 | { | |
53518d94 JQ |
2932 | RAMState **temp = opaque; |
2933 | RAMState *rs = *temp; | |
e8f3735f | 2934 | int ret = 0; |
6f37bb8b | 2935 | |
89ac5a1d DDAG |
2936 | WITH_RCU_READ_LOCK_GUARD() { |
2937 | if (!migration_in_postcopy()) { | |
2938 | migration_bitmap_sync_precopy(rs); | |
2939 | } | |
56e93d26 | 2940 | |
89ac5a1d | 2941 | ram_control_before_iterate(f, RAM_CONTROL_FINISH); |
56e93d26 | 2942 | |
89ac5a1d | 2943 | /* try transferring iterative blocks of memory */ |
56e93d26 | 2944 | |
89ac5a1d DDAG |
2945 | /* flush all remaining blocks regardless of rate limiting */ |
2946 | while (true) { | |
2947 | int pages; | |
56e93d26 | 2948 | |
89ac5a1d DDAG |
2949 | pages = ram_find_and_save_block(rs, !migration_in_colo_state()); |
2950 | /* no more blocks to sent */ | |
2951 | if (pages == 0) { | |
2952 | break; | |
2953 | } | |
2954 | if (pages < 0) { | |
2955 | ret = pages; | |
2956 | break; | |
2957 | } | |
e8f3735f | 2958 | } |
56e93d26 | 2959 | |
89ac5a1d DDAG |
2960 | flush_compressed_data(rs); |
2961 | ram_control_after_iterate(f, RAM_CONTROL_FINISH); | |
2962 | } | |
d09a6fde | 2963 | |
3d4095b2 | 2964 | if (ret >= 0) { |
99f2c6fb | 2965 | multifd_send_sync_main(rs->f); |
3d4095b2 JQ |
2966 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
2967 | qemu_fflush(f); | |
2968 | } | |
56e93d26 | 2969 | |
e8f3735f | 2970 | return ret; |
56e93d26 JQ |
2971 | } |
2972 | ||
c31b098f | 2973 | static void ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size, |
47995026 VSO |
2974 | uint64_t *res_precopy_only, |
2975 | uint64_t *res_compatible, | |
2976 | uint64_t *res_postcopy_only) | |
56e93d26 | 2977 | { |
53518d94 JQ |
2978 | RAMState **temp = opaque; |
2979 | RAMState *rs = *temp; | |
56e93d26 JQ |
2980 | uint64_t remaining_size; |
2981 | ||
9edabd4d | 2982 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2983 | |
5727309d | 2984 | if (!migration_in_postcopy() && |
663e6c1d | 2985 | remaining_size < max_size) { |
56e93d26 | 2986 | qemu_mutex_lock_iothread(); |
89ac5a1d DDAG |
2987 | WITH_RCU_READ_LOCK_GUARD() { |
2988 | migration_bitmap_sync_precopy(rs); | |
2989 | } | |
56e93d26 | 2990 | qemu_mutex_unlock_iothread(); |
9edabd4d | 2991 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2992 | } |
c31b098f | 2993 | |
86e1167e VSO |
2994 | if (migrate_postcopy_ram()) { |
2995 | /* We can do postcopy, and all the data is postcopiable */ | |
47995026 | 2996 | *res_compatible += remaining_size; |
86e1167e | 2997 | } else { |
47995026 | 2998 | *res_precopy_only += remaining_size; |
86e1167e | 2999 | } |
56e93d26 JQ |
3000 | } |
3001 | ||
3002 | static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) | |
3003 | { | |
3004 | unsigned int xh_len; | |
3005 | int xh_flags; | |
063e760a | 3006 | uint8_t *loaded_data; |
56e93d26 | 3007 | |
56e93d26 JQ |
3008 | /* extract RLE header */ |
3009 | xh_flags = qemu_get_byte(f); | |
3010 | xh_len = qemu_get_be16(f); | |
3011 | ||
3012 | if (xh_flags != ENCODING_FLAG_XBZRLE) { | |
3013 | error_report("Failed to load XBZRLE page - wrong compression!"); | |
3014 | return -1; | |
3015 | } | |
3016 | ||
3017 | if (xh_len > TARGET_PAGE_SIZE) { | |
3018 | error_report("Failed to load XBZRLE page - len overflow!"); | |
3019 | return -1; | |
3020 | } | |
f265e0e4 | 3021 | loaded_data = XBZRLE.decoded_buf; |
56e93d26 | 3022 | /* load data and decode */ |
f265e0e4 | 3023 | /* it can change loaded_data to point to an internal buffer */ |
063e760a | 3024 | qemu_get_buffer_in_place(f, &loaded_data, xh_len); |
56e93d26 JQ |
3025 | |
3026 | /* decode RLE */ | |
063e760a | 3027 | if (xbzrle_decode_buffer(loaded_data, xh_len, host, |
56e93d26 JQ |
3028 | TARGET_PAGE_SIZE) == -1) { |
3029 | error_report("Failed to load XBZRLE page - decode error!"); | |
3030 | return -1; | |
3031 | } | |
3032 | ||
3033 | return 0; | |
3034 | } | |
3035 | ||
3d0684b2 JQ |
3036 | /** |
3037 | * ram_block_from_stream: read a RAMBlock id from the migration stream | |
3038 | * | |
3039 | * Must be called from within a rcu critical section. | |
3040 | * | |
56e93d26 | 3041 | * Returns a pointer from within the RCU-protected ram_list. |
a7180877 | 3042 | * |
3d0684b2 JQ |
3043 | * @f: QEMUFile where to read the data from |
3044 | * @flags: Page flags (mostly to see if it's a continuation of previous block) | |
a7180877 | 3045 | */ |
3d0684b2 | 3046 | static inline RAMBlock *ram_block_from_stream(QEMUFile *f, int flags) |
56e93d26 | 3047 | { |
49324e93 | 3048 | static RAMBlock *block; |
56e93d26 JQ |
3049 | char id[256]; |
3050 | uint8_t len; | |
3051 | ||
3052 | if (flags & RAM_SAVE_FLAG_CONTINUE) { | |
4c4bad48 | 3053 | if (!block) { |
56e93d26 JQ |
3054 | error_report("Ack, bad migration stream!"); |
3055 | return NULL; | |
3056 | } | |
4c4bad48 | 3057 | return block; |
56e93d26 JQ |
3058 | } |
3059 | ||
3060 | len = qemu_get_byte(f); | |
3061 | qemu_get_buffer(f, (uint8_t *)id, len); | |
3062 | id[len] = 0; | |
3063 | ||
e3dd7493 | 3064 | block = qemu_ram_block_by_name(id); |
4c4bad48 HZ |
3065 | if (!block) { |
3066 | error_report("Can't find block %s", id); | |
3067 | return NULL; | |
56e93d26 JQ |
3068 | } |
3069 | ||
fbd162e6 | 3070 | if (ramblock_is_ignored(block)) { |
b895de50 CLG |
3071 | error_report("block %s should not be migrated !", id); |
3072 | return NULL; | |
3073 | } | |
3074 | ||
4c4bad48 HZ |
3075 | return block; |
3076 | } | |
3077 | ||
3078 | static inline void *host_from_ram_block_offset(RAMBlock *block, | |
3079 | ram_addr_t offset) | |
3080 | { | |
3081 | if (!offset_in_ramblock(block, offset)) { | |
3082 | return NULL; | |
3083 | } | |
3084 | ||
3085 | return block->host + offset; | |
56e93d26 JQ |
3086 | } |
3087 | ||
13af18f2 | 3088 | static inline void *colo_cache_from_block_offset(RAMBlock *block, |
8af66371 | 3089 | ram_addr_t offset, bool record_bitmap) |
13af18f2 ZC |
3090 | { |
3091 | if (!offset_in_ramblock(block, offset)) { | |
3092 | return NULL; | |
3093 | } | |
3094 | if (!block->colo_cache) { | |
3095 | error_report("%s: colo_cache is NULL in block :%s", | |
3096 | __func__, block->idstr); | |
3097 | return NULL; | |
3098 | } | |
7d9acafa ZC |
3099 | |
3100 | /* | |
3101 | * During colo checkpoint, we need bitmap of these migrated pages. | |
3102 | * It help us to decide which pages in ram cache should be flushed | |
3103 | * into VM's RAM later. | |
3104 | */ | |
8af66371 HZ |
3105 | if (record_bitmap && |
3106 | !test_and_set_bit(offset >> TARGET_PAGE_BITS, block->bmap)) { | |
7d9acafa ZC |
3107 | ram_state->migration_dirty_pages++; |
3108 | } | |
13af18f2 ZC |
3109 | return block->colo_cache + offset; |
3110 | } | |
3111 | ||
3d0684b2 JQ |
3112 | /** |
3113 | * ram_handle_compressed: handle the zero page case | |
3114 | * | |
56e93d26 JQ |
3115 | * If a page (or a whole RDMA chunk) has been |
3116 | * determined to be zero, then zap it. | |
3d0684b2 JQ |
3117 | * |
3118 | * @host: host address for the zero page | |
3119 | * @ch: what the page is filled from. We only support zero | |
3120 | * @size: size of the zero page | |
56e93d26 JQ |
3121 | */ |
3122 | void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) | |
3123 | { | |
3124 | if (ch != 0 || !is_zero_range(host, size)) { | |
3125 | memset(host, ch, size); | |
3126 | } | |
3127 | } | |
3128 | ||
797ca154 XG |
3129 | /* return the size after decompression, or negative value on error */ |
3130 | static int | |
3131 | qemu_uncompress_data(z_stream *stream, uint8_t *dest, size_t dest_len, | |
3132 | const uint8_t *source, size_t source_len) | |
3133 | { | |
3134 | int err; | |
3135 | ||
3136 | err = inflateReset(stream); | |
3137 | if (err != Z_OK) { | |
3138 | return -1; | |
3139 | } | |
3140 | ||
3141 | stream->avail_in = source_len; | |
3142 | stream->next_in = (uint8_t *)source; | |
3143 | stream->avail_out = dest_len; | |
3144 | stream->next_out = dest; | |
3145 | ||
3146 | err = inflate(stream, Z_NO_FLUSH); | |
3147 | if (err != Z_STREAM_END) { | |
3148 | return -1; | |
3149 | } | |
3150 | ||
3151 | return stream->total_out; | |
3152 | } | |
3153 | ||
56e93d26 JQ |
3154 | static void *do_data_decompress(void *opaque) |
3155 | { | |
3156 | DecompressParam *param = opaque; | |
3157 | unsigned long pagesize; | |
33d151f4 | 3158 | uint8_t *des; |
34ab9e97 | 3159 | int len, ret; |
56e93d26 | 3160 | |
33d151f4 | 3161 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 3162 | while (!param->quit) { |
33d151f4 LL |
3163 | if (param->des) { |
3164 | des = param->des; | |
3165 | len = param->len; | |
3166 | param->des = 0; | |
3167 | qemu_mutex_unlock(¶m->mutex); | |
3168 | ||
56e93d26 | 3169 | pagesize = TARGET_PAGE_SIZE; |
34ab9e97 XG |
3170 | |
3171 | ret = qemu_uncompress_data(¶m->stream, des, pagesize, | |
3172 | param->compbuf, len); | |
f548222c | 3173 | if (ret < 0 && migrate_get_current()->decompress_error_check) { |
34ab9e97 XG |
3174 | error_report("decompress data failed"); |
3175 | qemu_file_set_error(decomp_file, ret); | |
3176 | } | |
73a8912b | 3177 | |
33d151f4 LL |
3178 | qemu_mutex_lock(&decomp_done_lock); |
3179 | param->done = true; | |
3180 | qemu_cond_signal(&decomp_done_cond); | |
3181 | qemu_mutex_unlock(&decomp_done_lock); | |
3182 | ||
3183 | qemu_mutex_lock(¶m->mutex); | |
3184 | } else { | |
3185 | qemu_cond_wait(¶m->cond, ¶m->mutex); | |
3186 | } | |
56e93d26 | 3187 | } |
33d151f4 | 3188 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
3189 | |
3190 | return NULL; | |
3191 | } | |
3192 | ||
34ab9e97 | 3193 | static int wait_for_decompress_done(void) |
5533b2e9 LL |
3194 | { |
3195 | int idx, thread_count; | |
3196 | ||
3197 | if (!migrate_use_compression()) { | |
34ab9e97 | 3198 | return 0; |
5533b2e9 LL |
3199 | } |
3200 | ||
3201 | thread_count = migrate_decompress_threads(); | |
3202 | qemu_mutex_lock(&decomp_done_lock); | |
3203 | for (idx = 0; idx < thread_count; idx++) { | |
3204 | while (!decomp_param[idx].done) { | |
3205 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
3206 | } | |
3207 | } | |
3208 | qemu_mutex_unlock(&decomp_done_lock); | |
34ab9e97 | 3209 | return qemu_file_get_error(decomp_file); |
5533b2e9 LL |
3210 | } |
3211 | ||
f0afa331 | 3212 | static void compress_threads_load_cleanup(void) |
56e93d26 JQ |
3213 | { |
3214 | int i, thread_count; | |
3215 | ||
3416ab5b JQ |
3216 | if (!migrate_use_compression()) { |
3217 | return; | |
3218 | } | |
56e93d26 JQ |
3219 | thread_count = migrate_decompress_threads(); |
3220 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
3221 | /* |
3222 | * we use it as a indicator which shows if the thread is | |
3223 | * properly init'd or not | |
3224 | */ | |
3225 | if (!decomp_param[i].compbuf) { | |
3226 | break; | |
3227 | } | |
3228 | ||
56e93d26 | 3229 | qemu_mutex_lock(&decomp_param[i].mutex); |
90e56fb4 | 3230 | decomp_param[i].quit = true; |
56e93d26 JQ |
3231 | qemu_cond_signal(&decomp_param[i].cond); |
3232 | qemu_mutex_unlock(&decomp_param[i].mutex); | |
3233 | } | |
3234 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
3235 | if (!decomp_param[i].compbuf) { |
3236 | break; | |
3237 | } | |
3238 | ||
56e93d26 JQ |
3239 | qemu_thread_join(decompress_threads + i); |
3240 | qemu_mutex_destroy(&decomp_param[i].mutex); | |
3241 | qemu_cond_destroy(&decomp_param[i].cond); | |
797ca154 | 3242 | inflateEnd(&decomp_param[i].stream); |
56e93d26 | 3243 | g_free(decomp_param[i].compbuf); |
797ca154 | 3244 | decomp_param[i].compbuf = NULL; |
56e93d26 JQ |
3245 | } |
3246 | g_free(decompress_threads); | |
3247 | g_free(decomp_param); | |
56e93d26 JQ |
3248 | decompress_threads = NULL; |
3249 | decomp_param = NULL; | |
34ab9e97 | 3250 | decomp_file = NULL; |
56e93d26 JQ |
3251 | } |
3252 | ||
34ab9e97 | 3253 | static int compress_threads_load_setup(QEMUFile *f) |
797ca154 XG |
3254 | { |
3255 | int i, thread_count; | |
3256 | ||
3257 | if (!migrate_use_compression()) { | |
3258 | return 0; | |
3259 | } | |
3260 | ||
3261 | thread_count = migrate_decompress_threads(); | |
3262 | decompress_threads = g_new0(QemuThread, thread_count); | |
3263 | decomp_param = g_new0(DecompressParam, thread_count); | |
3264 | qemu_mutex_init(&decomp_done_lock); | |
3265 | qemu_cond_init(&decomp_done_cond); | |
34ab9e97 | 3266 | decomp_file = f; |
797ca154 XG |
3267 | for (i = 0; i < thread_count; i++) { |
3268 | if (inflateInit(&decomp_param[i].stream) != Z_OK) { | |
3269 | goto exit; | |
3270 | } | |
3271 | ||
3272 | decomp_param[i].compbuf = g_malloc0(compressBound(TARGET_PAGE_SIZE)); | |
3273 | qemu_mutex_init(&decomp_param[i].mutex); | |
3274 | qemu_cond_init(&decomp_param[i].cond); | |
3275 | decomp_param[i].done = true; | |
3276 | decomp_param[i].quit = false; | |
3277 | qemu_thread_create(decompress_threads + i, "decompress", | |
3278 | do_data_decompress, decomp_param + i, | |
3279 | QEMU_THREAD_JOINABLE); | |
3280 | } | |
3281 | return 0; | |
3282 | exit: | |
3283 | compress_threads_load_cleanup(); | |
3284 | return -1; | |
3285 | } | |
3286 | ||
c1bc6626 | 3287 | static void decompress_data_with_multi_threads(QEMUFile *f, |
56e93d26 JQ |
3288 | void *host, int len) |
3289 | { | |
3290 | int idx, thread_count; | |
3291 | ||
3292 | thread_count = migrate_decompress_threads(); | |
37396950 | 3293 | QEMU_LOCK_GUARD(&decomp_done_lock); |
56e93d26 JQ |
3294 | while (true) { |
3295 | for (idx = 0; idx < thread_count; idx++) { | |
73a8912b | 3296 | if (decomp_param[idx].done) { |
33d151f4 LL |
3297 | decomp_param[idx].done = false; |
3298 | qemu_mutex_lock(&decomp_param[idx].mutex); | |
c1bc6626 | 3299 | qemu_get_buffer(f, decomp_param[idx].compbuf, len); |
56e93d26 JQ |
3300 | decomp_param[idx].des = host; |
3301 | decomp_param[idx].len = len; | |
33d151f4 LL |
3302 | qemu_cond_signal(&decomp_param[idx].cond); |
3303 | qemu_mutex_unlock(&decomp_param[idx].mutex); | |
56e93d26 JQ |
3304 | break; |
3305 | } | |
3306 | } | |
3307 | if (idx < thread_count) { | |
3308 | break; | |
73a8912b LL |
3309 | } else { |
3310 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
56e93d26 JQ |
3311 | } |
3312 | } | |
3313 | } | |
3314 | ||
b70cb3b4 RL |
3315 | static void colo_init_ram_state(void) |
3316 | { | |
3317 | ram_state_init(&ram_state); | |
b70cb3b4 RL |
3318 | } |
3319 | ||
13af18f2 ZC |
3320 | /* |
3321 | * colo cache: this is for secondary VM, we cache the whole | |
3322 | * memory of the secondary VM, it is need to hold the global lock | |
3323 | * to call this helper. | |
3324 | */ | |
3325 | int colo_init_ram_cache(void) | |
3326 | { | |
3327 | RAMBlock *block; | |
3328 | ||
44901b5a PB |
3329 | WITH_RCU_READ_LOCK_GUARD() { |
3330 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3331 | block->colo_cache = qemu_anon_ram_alloc(block->used_length, | |
3332 | NULL, | |
3333 | false); | |
3334 | if (!block->colo_cache) { | |
3335 | error_report("%s: Can't alloc memory for COLO cache of block %s," | |
3336 | "size 0x" RAM_ADDR_FMT, __func__, block->idstr, | |
3337 | block->used_length); | |
3338 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3339 | if (block->colo_cache) { | |
3340 | qemu_anon_ram_free(block->colo_cache, block->used_length); | |
3341 | block->colo_cache = NULL; | |
3342 | } | |
89ac5a1d | 3343 | } |
44901b5a | 3344 | return -errno; |
89ac5a1d | 3345 | } |
13af18f2 | 3346 | } |
13af18f2 | 3347 | } |
44901b5a | 3348 | |
7d9acafa ZC |
3349 | /* |
3350 | * Record the dirty pages that sent by PVM, we use this dirty bitmap together | |
3351 | * with to decide which page in cache should be flushed into SVM's RAM. Here | |
3352 | * we use the same name 'ram_bitmap' as for migration. | |
3353 | */ | |
3354 | if (ram_bytes_total()) { | |
3355 | RAMBlock *block; | |
3356 | ||
fbd162e6 | 3357 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
7d9acafa | 3358 | unsigned long pages = block->max_length >> TARGET_PAGE_BITS; |
7d9acafa | 3359 | block->bmap = bitmap_new(pages); |
7d9acafa ZC |
3360 | } |
3361 | } | |
7d9acafa | 3362 | |
b70cb3b4 | 3363 | colo_init_ram_state(); |
13af18f2 | 3364 | return 0; |
13af18f2 ZC |
3365 | } |
3366 | ||
0393031a HZ |
3367 | /* TODO: duplicated with ram_init_bitmaps */ |
3368 | void colo_incoming_start_dirty_log(void) | |
3369 | { | |
3370 | RAMBlock *block = NULL; | |
3371 | /* For memory_global_dirty_log_start below. */ | |
3372 | qemu_mutex_lock_iothread(); | |
3373 | qemu_mutex_lock_ramlist(); | |
3374 | ||
3375 | memory_global_dirty_log_sync(); | |
3376 | WITH_RCU_READ_LOCK_GUARD() { | |
3377 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3378 | ramblock_sync_dirty_bitmap(ram_state, block); | |
3379 | /* Discard this dirty bitmap record */ | |
3380 | bitmap_zero(block->bmap, block->max_length >> TARGET_PAGE_BITS); | |
3381 | } | |
3382 | memory_global_dirty_log_start(); | |
3383 | } | |
3384 | ram_state->migration_dirty_pages = 0; | |
3385 | qemu_mutex_unlock_ramlist(); | |
3386 | qemu_mutex_unlock_iothread(); | |
3387 | } | |
3388 | ||
13af18f2 ZC |
3389 | /* It is need to hold the global lock to call this helper */ |
3390 | void colo_release_ram_cache(void) | |
3391 | { | |
3392 | RAMBlock *block; | |
3393 | ||
d1955d22 | 3394 | memory_global_dirty_log_stop(); |
fbd162e6 | 3395 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
7d9acafa ZC |
3396 | g_free(block->bmap); |
3397 | block->bmap = NULL; | |
3398 | } | |
3399 | ||
89ac5a1d DDAG |
3400 | WITH_RCU_READ_LOCK_GUARD() { |
3401 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3402 | if (block->colo_cache) { | |
3403 | qemu_anon_ram_free(block->colo_cache, block->used_length); | |
3404 | block->colo_cache = NULL; | |
3405 | } | |
13af18f2 ZC |
3406 | } |
3407 | } | |
0393031a | 3408 | ram_state_cleanup(&ram_state); |
13af18f2 ZC |
3409 | } |
3410 | ||
f265e0e4 JQ |
3411 | /** |
3412 | * ram_load_setup: Setup RAM for migration incoming side | |
3413 | * | |
3414 | * Returns zero to indicate success and negative for error | |
3415 | * | |
3416 | * @f: QEMUFile where to receive the data | |
3417 | * @opaque: RAMState pointer | |
3418 | */ | |
3419 | static int ram_load_setup(QEMUFile *f, void *opaque) | |
3420 | { | |
34ab9e97 | 3421 | if (compress_threads_load_setup(f)) { |
797ca154 XG |
3422 | return -1; |
3423 | } | |
3424 | ||
f265e0e4 | 3425 | xbzrle_load_setup(); |
f9494614 | 3426 | ramblock_recv_map_init(); |
13af18f2 | 3427 | |
f265e0e4 JQ |
3428 | return 0; |
3429 | } | |
3430 | ||
3431 | static int ram_load_cleanup(void *opaque) | |
3432 | { | |
f9494614 | 3433 | RAMBlock *rb; |
56eb90af | 3434 | |
fbd162e6 | 3435 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
bd108a44 | 3436 | qemu_ram_block_writeback(rb); |
56eb90af JH |
3437 | } |
3438 | ||
f265e0e4 | 3439 | xbzrle_load_cleanup(); |
f0afa331 | 3440 | compress_threads_load_cleanup(); |
f9494614 | 3441 | |
fbd162e6 | 3442 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
f9494614 AP |
3443 | g_free(rb->receivedmap); |
3444 | rb->receivedmap = NULL; | |
3445 | } | |
13af18f2 | 3446 | |
f265e0e4 JQ |
3447 | return 0; |
3448 | } | |
3449 | ||
3d0684b2 JQ |
3450 | /** |
3451 | * ram_postcopy_incoming_init: allocate postcopy data structures | |
3452 | * | |
3453 | * Returns 0 for success and negative if there was one error | |
3454 | * | |
3455 | * @mis: current migration incoming state | |
3456 | * | |
3457 | * Allocate data structures etc needed by incoming migration with | |
3458 | * postcopy-ram. postcopy-ram's similarly names | |
3459 | * postcopy_ram_incoming_init does the work. | |
1caddf8a DDAG |
3460 | */ |
3461 | int ram_postcopy_incoming_init(MigrationIncomingState *mis) | |
3462 | { | |
c136180c | 3463 | return postcopy_ram_incoming_init(mis); |
1caddf8a DDAG |
3464 | } |
3465 | ||
3d0684b2 JQ |
3466 | /** |
3467 | * ram_load_postcopy: load a page in postcopy case | |
3468 | * | |
3469 | * Returns 0 for success or -errno in case of error | |
3470 | * | |
a7180877 DDAG |
3471 | * Called in postcopy mode by ram_load(). |
3472 | * rcu_read_lock is taken prior to this being called. | |
3d0684b2 JQ |
3473 | * |
3474 | * @f: QEMUFile where to send the data | |
a7180877 DDAG |
3475 | */ |
3476 | static int ram_load_postcopy(QEMUFile *f) | |
3477 | { | |
3478 | int flags = 0, ret = 0; | |
3479 | bool place_needed = false; | |
1aa83678 | 3480 | bool matches_target_page_size = false; |
a7180877 DDAG |
3481 | MigrationIncomingState *mis = migration_incoming_get_current(); |
3482 | /* Temporary page that is later 'placed' */ | |
3414322a | 3483 | void *postcopy_host_page = mis->postcopy_tmp_page; |
91ba442f | 3484 | void *this_host = NULL; |
ddf35bdf | 3485 | bool all_zero = true; |
4cbb3c63 | 3486 | int target_pages = 0; |
a7180877 DDAG |
3487 | |
3488 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
3489 | ram_addr_t addr; | |
3490 | void *host = NULL; | |
3491 | void *page_buffer = NULL; | |
3492 | void *place_source = NULL; | |
df9ff5e1 | 3493 | RAMBlock *block = NULL; |
a7180877 | 3494 | uint8_t ch; |
644acf99 | 3495 | int len; |
a7180877 DDAG |
3496 | |
3497 | addr = qemu_get_be64(f); | |
7a9ddfbf PX |
3498 | |
3499 | /* | |
3500 | * If qemu file error, we should stop here, and then "addr" | |
3501 | * may be invalid | |
3502 | */ | |
3503 | ret = qemu_file_get_error(f); | |
3504 | if (ret) { | |
3505 | break; | |
3506 | } | |
3507 | ||
a7180877 DDAG |
3508 | flags = addr & ~TARGET_PAGE_MASK; |
3509 | addr &= TARGET_PAGE_MASK; | |
3510 | ||
3511 | trace_ram_load_postcopy_loop((uint64_t)addr, flags); | |
644acf99 WY |
3512 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
3513 | RAM_SAVE_FLAG_COMPRESS_PAGE)) { | |
df9ff5e1 | 3514 | block = ram_block_from_stream(f, flags); |
4c4bad48 HZ |
3515 | |
3516 | host = host_from_ram_block_offset(block, addr); | |
a7180877 DDAG |
3517 | if (!host) { |
3518 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
3519 | ret = -EINVAL; | |
3520 | break; | |
3521 | } | |
4cbb3c63 | 3522 | target_pages++; |
1aa83678 | 3523 | matches_target_page_size = block->page_size == TARGET_PAGE_SIZE; |
a7180877 | 3524 | /* |
28abd200 DDAG |
3525 | * Postcopy requires that we place whole host pages atomically; |
3526 | * these may be huge pages for RAMBlocks that are backed by | |
3527 | * hugetlbfs. | |
a7180877 DDAG |
3528 | * To make it atomic, the data is read into a temporary page |
3529 | * that's moved into place later. | |
3530 | * The migration protocol uses, possibly smaller, target-pages | |
3531 | * however the source ensures it always sends all the components | |
91ba442f | 3532 | * of a host page in one chunk. |
a7180877 DDAG |
3533 | */ |
3534 | page_buffer = postcopy_host_page + | |
28abd200 | 3535 | ((uintptr_t)host & (block->page_size - 1)); |
e5e73b0f | 3536 | if (target_pages == 1) { |
91ba442f WY |
3537 | this_host = (void *)QEMU_ALIGN_DOWN((uintptr_t)host, |
3538 | block->page_size); | |
c53b7ddc DDAG |
3539 | } else { |
3540 | /* not the 1st TP within the HP */ | |
91ba442f WY |
3541 | if (QEMU_ALIGN_DOWN((uintptr_t)host, block->page_size) != |
3542 | (uintptr_t)this_host) { | |
3543 | error_report("Non-same host page %p/%p", | |
3544 | host, this_host); | |
c53b7ddc DDAG |
3545 | ret = -EINVAL; |
3546 | break; | |
3547 | } | |
a7180877 DDAG |
3548 | } |
3549 | ||
3550 | /* | |
3551 | * If it's the last part of a host page then we place the host | |
3552 | * page | |
3553 | */ | |
4cbb3c63 WY |
3554 | if (target_pages == (block->page_size / TARGET_PAGE_SIZE)) { |
3555 | place_needed = true; | |
4cbb3c63 | 3556 | } |
a7180877 DDAG |
3557 | place_source = postcopy_host_page; |
3558 | } | |
3559 | ||
3560 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { | |
bb890ed5 | 3561 | case RAM_SAVE_FLAG_ZERO: |
a7180877 | 3562 | ch = qemu_get_byte(f); |
2e36bc1b WY |
3563 | /* |
3564 | * Can skip to set page_buffer when | |
3565 | * this is a zero page and (block->page_size == TARGET_PAGE_SIZE). | |
3566 | */ | |
3567 | if (ch || !matches_target_page_size) { | |
3568 | memset(page_buffer, ch, TARGET_PAGE_SIZE); | |
3569 | } | |
a7180877 DDAG |
3570 | if (ch) { |
3571 | all_zero = false; | |
3572 | } | |
3573 | break; | |
3574 | ||
3575 | case RAM_SAVE_FLAG_PAGE: | |
3576 | all_zero = false; | |
1aa83678 PX |
3577 | if (!matches_target_page_size) { |
3578 | /* For huge pages, we always use temporary buffer */ | |
a7180877 DDAG |
3579 | qemu_get_buffer(f, page_buffer, TARGET_PAGE_SIZE); |
3580 | } else { | |
1aa83678 PX |
3581 | /* |
3582 | * For small pages that matches target page size, we | |
3583 | * avoid the qemu_file copy. Instead we directly use | |
3584 | * the buffer of QEMUFile to place the page. Note: we | |
3585 | * cannot do any QEMUFile operation before using that | |
3586 | * buffer to make sure the buffer is valid when | |
3587 | * placing the page. | |
a7180877 DDAG |
3588 | */ |
3589 | qemu_get_buffer_in_place(f, (uint8_t **)&place_source, | |
3590 | TARGET_PAGE_SIZE); | |
3591 | } | |
3592 | break; | |
644acf99 WY |
3593 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
3594 | all_zero = false; | |
3595 | len = qemu_get_be32(f); | |
3596 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
3597 | error_report("Invalid compressed data length: %d", len); | |
3598 | ret = -EINVAL; | |
3599 | break; | |
3600 | } | |
3601 | decompress_data_with_multi_threads(f, page_buffer, len); | |
3602 | break; | |
3603 | ||
a7180877 DDAG |
3604 | case RAM_SAVE_FLAG_EOS: |
3605 | /* normal exit */ | |
6df264ac | 3606 | multifd_recv_sync_main(); |
a7180877 DDAG |
3607 | break; |
3608 | default: | |
29fccade | 3609 | error_report("Unknown combination of migration flags: 0x%x" |
a7180877 DDAG |
3610 | " (postcopy mode)", flags); |
3611 | ret = -EINVAL; | |
7a9ddfbf PX |
3612 | break; |
3613 | } | |
3614 | ||
644acf99 WY |
3615 | /* Got the whole host page, wait for decompress before placing. */ |
3616 | if (place_needed) { | |
3617 | ret |= wait_for_decompress_done(); | |
3618 | } | |
3619 | ||
7a9ddfbf PX |
3620 | /* Detect for any possible file errors */ |
3621 | if (!ret && qemu_file_get_error(f)) { | |
3622 | ret = qemu_file_get_error(f); | |
a7180877 DDAG |
3623 | } |
3624 | ||
7a9ddfbf | 3625 | if (!ret && place_needed) { |
a7180877 | 3626 | /* This gets called at the last target page in the host page */ |
91ba442f WY |
3627 | void *place_dest = (void *)QEMU_ALIGN_DOWN((uintptr_t)host, |
3628 | block->page_size); | |
df9ff5e1 | 3629 | |
a7180877 | 3630 | if (all_zero) { |
df9ff5e1 | 3631 | ret = postcopy_place_page_zero(mis, place_dest, |
8be4620b | 3632 | block); |
a7180877 | 3633 | } else { |
df9ff5e1 | 3634 | ret = postcopy_place_page(mis, place_dest, |
8be4620b | 3635 | place_source, block); |
a7180877 | 3636 | } |
ddf35bdf DH |
3637 | place_needed = false; |
3638 | target_pages = 0; | |
3639 | /* Assume we have a zero page until we detect something different */ | |
3640 | all_zero = true; | |
a7180877 | 3641 | } |
a7180877 DDAG |
3642 | } |
3643 | ||
3644 | return ret; | |
3645 | } | |
3646 | ||
acab30b8 DHB |
3647 | static bool postcopy_is_advised(void) |
3648 | { | |
3649 | PostcopyState ps = postcopy_state_get(); | |
3650 | return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END; | |
3651 | } | |
3652 | ||
3653 | static bool postcopy_is_running(void) | |
3654 | { | |
3655 | PostcopyState ps = postcopy_state_get(); | |
3656 | return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; | |
3657 | } | |
3658 | ||
e6f4aa18 ZC |
3659 | /* |
3660 | * Flush content of RAM cache into SVM's memory. | |
3661 | * Only flush the pages that be dirtied by PVM or SVM or both. | |
3662 | */ | |
24fa16f8 | 3663 | void colo_flush_ram_cache(void) |
e6f4aa18 ZC |
3664 | { |
3665 | RAMBlock *block = NULL; | |
3666 | void *dst_host; | |
3667 | void *src_host; | |
3668 | unsigned long offset = 0; | |
3669 | ||
d1955d22 | 3670 | memory_global_dirty_log_sync(); |
89ac5a1d DDAG |
3671 | WITH_RCU_READ_LOCK_GUARD() { |
3672 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3673 | ramblock_sync_dirty_bitmap(ram_state, block); | |
3674 | } | |
d1955d22 | 3675 | } |
d1955d22 | 3676 | |
e6f4aa18 | 3677 | trace_colo_flush_ram_cache_begin(ram_state->migration_dirty_pages); |
89ac5a1d DDAG |
3678 | WITH_RCU_READ_LOCK_GUARD() { |
3679 | block = QLIST_FIRST_RCU(&ram_list.blocks); | |
e6f4aa18 | 3680 | |
89ac5a1d DDAG |
3681 | while (block) { |
3682 | offset = migration_bitmap_find_dirty(ram_state, block, offset); | |
e6f4aa18 | 3683 | |
8bba004c AR |
3684 | if (((ram_addr_t)offset) << TARGET_PAGE_BITS |
3685 | >= block->used_length) { | |
89ac5a1d DDAG |
3686 | offset = 0; |
3687 | block = QLIST_NEXT_RCU(block, next); | |
3688 | } else { | |
3689 | migration_bitmap_clear_dirty(ram_state, block, offset); | |
8bba004c AR |
3690 | dst_host = block->host |
3691 | + (((ram_addr_t)offset) << TARGET_PAGE_BITS); | |
3692 | src_host = block->colo_cache | |
3693 | + (((ram_addr_t)offset) << TARGET_PAGE_BITS); | |
89ac5a1d DDAG |
3694 | memcpy(dst_host, src_host, TARGET_PAGE_SIZE); |
3695 | } | |
e6f4aa18 ZC |
3696 | } |
3697 | } | |
e6f4aa18 ZC |
3698 | trace_colo_flush_ram_cache_end(); |
3699 | } | |
3700 | ||
10da4a36 WY |
3701 | /** |
3702 | * ram_load_precopy: load pages in precopy case | |
3703 | * | |
3704 | * Returns 0 for success or -errno in case of error | |
3705 | * | |
3706 | * Called in precopy mode by ram_load(). | |
3707 | * rcu_read_lock is taken prior to this being called. | |
3708 | * | |
3709 | * @f: QEMUFile where to send the data | |
3710 | */ | |
3711 | static int ram_load_precopy(QEMUFile *f) | |
56e93d26 | 3712 | { |
e65cec5e | 3713 | int flags = 0, ret = 0, invalid_flags = 0, len = 0, i = 0; |
ef08fb38 | 3714 | /* ADVISE is earlier, it shows the source has the postcopy capability on */ |
acab30b8 | 3715 | bool postcopy_advised = postcopy_is_advised(); |
edc60127 JQ |
3716 | if (!migrate_use_compression()) { |
3717 | invalid_flags |= RAM_SAVE_FLAG_COMPRESS_PAGE; | |
3718 | } | |
a7180877 | 3719 | |
10da4a36 | 3720 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { |
56e93d26 | 3721 | ram_addr_t addr, total_ram_bytes; |
0393031a | 3722 | void *host = NULL, *host_bak = NULL; |
56e93d26 JQ |
3723 | uint8_t ch; |
3724 | ||
e65cec5e YK |
3725 | /* |
3726 | * Yield periodically to let main loop run, but an iteration of | |
3727 | * the main loop is expensive, so do it each some iterations | |
3728 | */ | |
3729 | if ((i & 32767) == 0 && qemu_in_coroutine()) { | |
3730 | aio_co_schedule(qemu_get_current_aio_context(), | |
3731 | qemu_coroutine_self()); | |
3732 | qemu_coroutine_yield(); | |
3733 | } | |
3734 | i++; | |
3735 | ||
56e93d26 JQ |
3736 | addr = qemu_get_be64(f); |
3737 | flags = addr & ~TARGET_PAGE_MASK; | |
3738 | addr &= TARGET_PAGE_MASK; | |
3739 | ||
edc60127 JQ |
3740 | if (flags & invalid_flags) { |
3741 | if (flags & invalid_flags & RAM_SAVE_FLAG_COMPRESS_PAGE) { | |
3742 | error_report("Received an unexpected compressed page"); | |
3743 | } | |
3744 | ||
3745 | ret = -EINVAL; | |
3746 | break; | |
3747 | } | |
3748 | ||
bb890ed5 | 3749 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
a776aa15 | 3750 | RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) { |
4c4bad48 HZ |
3751 | RAMBlock *block = ram_block_from_stream(f, flags); |
3752 | ||
0393031a | 3753 | host = host_from_ram_block_offset(block, addr); |
13af18f2 | 3754 | /* |
0393031a HZ |
3755 | * After going into COLO stage, we should not load the page |
3756 | * into SVM's memory directly, we put them into colo_cache firstly. | |
3757 | * NOTE: We need to keep a copy of SVM's ram in colo_cache. | |
3758 | * Previously, we copied all these memory in preparing stage of COLO | |
3759 | * while we need to stop VM, which is a time-consuming process. | |
3760 | * Here we optimize it by a trick, back-up every page while in | |
3761 | * migration process while COLO is enabled, though it affects the | |
3762 | * speed of the migration, but it obviously reduce the downtime of | |
3763 | * back-up all SVM'S memory in COLO preparing stage. | |
13af18f2 | 3764 | */ |
0393031a HZ |
3765 | if (migration_incoming_colo_enabled()) { |
3766 | if (migration_incoming_in_colo_state()) { | |
3767 | /* In COLO stage, put all pages into cache temporarily */ | |
8af66371 | 3768 | host = colo_cache_from_block_offset(block, addr, true); |
0393031a HZ |
3769 | } else { |
3770 | /* | |
3771 | * In migration stage but before COLO stage, | |
3772 | * Put all pages into both cache and SVM's memory. | |
3773 | */ | |
8af66371 | 3774 | host_bak = colo_cache_from_block_offset(block, addr, false); |
0393031a | 3775 | } |
13af18f2 | 3776 | } |
a776aa15 DDAG |
3777 | if (!host) { |
3778 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
3779 | ret = -EINVAL; | |
3780 | break; | |
3781 | } | |
13af18f2 ZC |
3782 | if (!migration_incoming_in_colo_state()) { |
3783 | ramblock_recv_bitmap_set(block, host); | |
3784 | } | |
3785 | ||
1db9d8e5 | 3786 | trace_ram_load_loop(block->idstr, (uint64_t)addr, flags, host); |
a776aa15 DDAG |
3787 | } |
3788 | ||
56e93d26 JQ |
3789 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { |
3790 | case RAM_SAVE_FLAG_MEM_SIZE: | |
3791 | /* Synchronize RAM block list */ | |
3792 | total_ram_bytes = addr; | |
3793 | while (!ret && total_ram_bytes) { | |
3794 | RAMBlock *block; | |
56e93d26 JQ |
3795 | char id[256]; |
3796 | ram_addr_t length; | |
3797 | ||
3798 | len = qemu_get_byte(f); | |
3799 | qemu_get_buffer(f, (uint8_t *)id, len); | |
3800 | id[len] = 0; | |
3801 | length = qemu_get_be64(f); | |
3802 | ||
e3dd7493 | 3803 | block = qemu_ram_block_by_name(id); |
b895de50 CLG |
3804 | if (block && !qemu_ram_is_migratable(block)) { |
3805 | error_report("block %s should not be migrated !", id); | |
3806 | ret = -EINVAL; | |
3807 | } else if (block) { | |
e3dd7493 DDAG |
3808 | if (length != block->used_length) { |
3809 | Error *local_err = NULL; | |
56e93d26 | 3810 | |
fa53a0e5 | 3811 | ret = qemu_ram_resize(block, length, |
e3dd7493 DDAG |
3812 | &local_err); |
3813 | if (local_err) { | |
3814 | error_report_err(local_err); | |
56e93d26 | 3815 | } |
56e93d26 | 3816 | } |
ef08fb38 | 3817 | /* For postcopy we need to check hugepage sizes match */ |
e846b746 | 3818 | if (postcopy_advised && migrate_postcopy_ram() && |
ef08fb38 DDAG |
3819 | block->page_size != qemu_host_page_size) { |
3820 | uint64_t remote_page_size = qemu_get_be64(f); | |
3821 | if (remote_page_size != block->page_size) { | |
3822 | error_report("Mismatched RAM page size %s " | |
3823 | "(local) %zd != %" PRId64, | |
3824 | id, block->page_size, | |
3825 | remote_page_size); | |
3826 | ret = -EINVAL; | |
3827 | } | |
3828 | } | |
fbd162e6 YK |
3829 | if (migrate_ignore_shared()) { |
3830 | hwaddr addr = qemu_get_be64(f); | |
fbd162e6 YK |
3831 | if (ramblock_is_ignored(block) && |
3832 | block->mr->addr != addr) { | |
3833 | error_report("Mismatched GPAs for block %s " | |
3834 | "%" PRId64 "!= %" PRId64, | |
3835 | id, (uint64_t)addr, | |
3836 | (uint64_t)block->mr->addr); | |
3837 | ret = -EINVAL; | |
3838 | } | |
3839 | } | |
e3dd7493 DDAG |
3840 | ram_control_load_hook(f, RAM_CONTROL_BLOCK_REG, |
3841 | block->idstr); | |
3842 | } else { | |
56e93d26 JQ |
3843 | error_report("Unknown ramblock \"%s\", cannot " |
3844 | "accept migration", id); | |
3845 | ret = -EINVAL; | |
3846 | } | |
3847 | ||
3848 | total_ram_bytes -= length; | |
3849 | } | |
3850 | break; | |
a776aa15 | 3851 | |
bb890ed5 | 3852 | case RAM_SAVE_FLAG_ZERO: |
56e93d26 JQ |
3853 | ch = qemu_get_byte(f); |
3854 | ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); | |
3855 | break; | |
a776aa15 | 3856 | |
56e93d26 | 3857 | case RAM_SAVE_FLAG_PAGE: |
56e93d26 JQ |
3858 | qemu_get_buffer(f, host, TARGET_PAGE_SIZE); |
3859 | break; | |
56e93d26 | 3860 | |
a776aa15 | 3861 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
56e93d26 JQ |
3862 | len = qemu_get_be32(f); |
3863 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
3864 | error_report("Invalid compressed data length: %d", len); | |
3865 | ret = -EINVAL; | |
3866 | break; | |
3867 | } | |
c1bc6626 | 3868 | decompress_data_with_multi_threads(f, host, len); |
56e93d26 | 3869 | break; |
a776aa15 | 3870 | |
56e93d26 | 3871 | case RAM_SAVE_FLAG_XBZRLE: |
56e93d26 JQ |
3872 | if (load_xbzrle(f, addr, host) < 0) { |
3873 | error_report("Failed to decompress XBZRLE page at " | |
3874 | RAM_ADDR_FMT, addr); | |
3875 | ret = -EINVAL; | |
3876 | break; | |
3877 | } | |
3878 | break; | |
3879 | case RAM_SAVE_FLAG_EOS: | |
3880 | /* normal exit */ | |
6df264ac | 3881 | multifd_recv_sync_main(); |
56e93d26 JQ |
3882 | break; |
3883 | default: | |
3884 | if (flags & RAM_SAVE_FLAG_HOOK) { | |
632e3a5c | 3885 | ram_control_load_hook(f, RAM_CONTROL_HOOK, NULL); |
56e93d26 | 3886 | } else { |
29fccade | 3887 | error_report("Unknown combination of migration flags: 0x%x", |
56e93d26 JQ |
3888 | flags); |
3889 | ret = -EINVAL; | |
3890 | } | |
3891 | } | |
3892 | if (!ret) { | |
3893 | ret = qemu_file_get_error(f); | |
3894 | } | |
0393031a HZ |
3895 | if (!ret && host_bak) { |
3896 | memcpy(host_bak, host, TARGET_PAGE_SIZE); | |
3897 | } | |
56e93d26 JQ |
3898 | } |
3899 | ||
ca1a6b70 | 3900 | ret |= wait_for_decompress_done(); |
10da4a36 WY |
3901 | return ret; |
3902 | } | |
3903 | ||
3904 | static int ram_load(QEMUFile *f, void *opaque, int version_id) | |
3905 | { | |
3906 | int ret = 0; | |
3907 | static uint64_t seq_iter; | |
3908 | /* | |
3909 | * If system is running in postcopy mode, page inserts to host memory must | |
3910 | * be atomic | |
3911 | */ | |
3912 | bool postcopy_running = postcopy_is_running(); | |
3913 | ||
3914 | seq_iter++; | |
3915 | ||
3916 | if (version_id != 4) { | |
3917 | return -EINVAL; | |
3918 | } | |
3919 | ||
3920 | /* | |
3921 | * This RCU critical section can be very long running. | |
3922 | * When RCU reclaims in the code start to become numerous, | |
3923 | * it will be necessary to reduce the granularity of this | |
3924 | * critical section. | |
3925 | */ | |
89ac5a1d DDAG |
3926 | WITH_RCU_READ_LOCK_GUARD() { |
3927 | if (postcopy_running) { | |
3928 | ret = ram_load_postcopy(f); | |
3929 | } else { | |
3930 | ret = ram_load_precopy(f); | |
3931 | } | |
10da4a36 | 3932 | } |
55c4446b | 3933 | trace_ram_load_complete(ret, seq_iter); |
e6f4aa18 | 3934 | |
56e93d26 JQ |
3935 | return ret; |
3936 | } | |
3937 | ||
c6467627 VSO |
3938 | static bool ram_has_postcopy(void *opaque) |
3939 | { | |
469dd51b | 3940 | RAMBlock *rb; |
fbd162e6 | 3941 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
469dd51b JH |
3942 | if (ramblock_is_pmem(rb)) { |
3943 | info_report("Block: %s, host: %p is a nvdimm memory, postcopy" | |
3944 | "is not supported now!", rb->idstr, rb->host); | |
3945 | return false; | |
3946 | } | |
3947 | } | |
3948 | ||
c6467627 VSO |
3949 | return migrate_postcopy_ram(); |
3950 | } | |
3951 | ||
edd090c7 PX |
3952 | /* Sync all the dirty bitmap with destination VM. */ |
3953 | static int ram_dirty_bitmap_sync_all(MigrationState *s, RAMState *rs) | |
3954 | { | |
3955 | RAMBlock *block; | |
3956 | QEMUFile *file = s->to_dst_file; | |
3957 | int ramblock_count = 0; | |
3958 | ||
3959 | trace_ram_dirty_bitmap_sync_start(); | |
3960 | ||
fbd162e6 | 3961 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
edd090c7 PX |
3962 | qemu_savevm_send_recv_bitmap(file, block->idstr); |
3963 | trace_ram_dirty_bitmap_request(block->idstr); | |
3964 | ramblock_count++; | |
3965 | } | |
3966 | ||
3967 | trace_ram_dirty_bitmap_sync_wait(); | |
3968 | ||
3969 | /* Wait until all the ramblocks' dirty bitmap synced */ | |
3970 | while (ramblock_count--) { | |
3971 | qemu_sem_wait(&s->rp_state.rp_sem); | |
3972 | } | |
3973 | ||
3974 | trace_ram_dirty_bitmap_sync_complete(); | |
3975 | ||
3976 | return 0; | |
3977 | } | |
3978 | ||
3979 | static void ram_dirty_bitmap_reload_notify(MigrationState *s) | |
3980 | { | |
3981 | qemu_sem_post(&s->rp_state.rp_sem); | |
3982 | } | |
3983 | ||
a335debb PX |
3984 | /* |
3985 | * Read the received bitmap, revert it as the initial dirty bitmap. | |
3986 | * This is only used when the postcopy migration is paused but wants | |
3987 | * to resume from a middle point. | |
3988 | */ | |
3989 | int ram_dirty_bitmap_reload(MigrationState *s, RAMBlock *block) | |
3990 | { | |
3991 | int ret = -EINVAL; | |
3992 | QEMUFile *file = s->rp_state.from_dst_file; | |
3993 | unsigned long *le_bitmap, nbits = block->used_length >> TARGET_PAGE_BITS; | |
a725ef9f | 3994 | uint64_t local_size = DIV_ROUND_UP(nbits, 8); |
a335debb PX |
3995 | uint64_t size, end_mark; |
3996 | ||
3997 | trace_ram_dirty_bitmap_reload_begin(block->idstr); | |
3998 | ||
3999 | if (s->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
4000 | error_report("%s: incorrect state %s", __func__, | |
4001 | MigrationStatus_str(s->state)); | |
4002 | return -EINVAL; | |
4003 | } | |
4004 | ||
4005 | /* | |
4006 | * Note: see comments in ramblock_recv_bitmap_send() on why we | |
3a4452d8 | 4007 | * need the endianness conversion, and the paddings. |
a335debb PX |
4008 | */ |
4009 | local_size = ROUND_UP(local_size, 8); | |
4010 | ||
4011 | /* Add paddings */ | |
4012 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
4013 | ||
4014 | size = qemu_get_be64(file); | |
4015 | ||
4016 | /* The size of the bitmap should match with our ramblock */ | |
4017 | if (size != local_size) { | |
4018 | error_report("%s: ramblock '%s' bitmap size mismatch " | |
4019 | "(0x%"PRIx64" != 0x%"PRIx64")", __func__, | |
4020 | block->idstr, size, local_size); | |
4021 | ret = -EINVAL; | |
4022 | goto out; | |
4023 | } | |
4024 | ||
4025 | size = qemu_get_buffer(file, (uint8_t *)le_bitmap, local_size); | |
4026 | end_mark = qemu_get_be64(file); | |
4027 | ||
4028 | ret = qemu_file_get_error(file); | |
4029 | if (ret || size != local_size) { | |
4030 | error_report("%s: read bitmap failed for ramblock '%s': %d" | |
4031 | " (size 0x%"PRIx64", got: 0x%"PRIx64")", | |
4032 | __func__, block->idstr, ret, local_size, size); | |
4033 | ret = -EIO; | |
4034 | goto out; | |
4035 | } | |
4036 | ||
4037 | if (end_mark != RAMBLOCK_RECV_BITMAP_ENDING) { | |
af3bbbe9 | 4038 | error_report("%s: ramblock '%s' end mark incorrect: 0x%"PRIx64, |
a335debb PX |
4039 | __func__, block->idstr, end_mark); |
4040 | ret = -EINVAL; | |
4041 | goto out; | |
4042 | } | |
4043 | ||
4044 | /* | |
3a4452d8 | 4045 | * Endianness conversion. We are during postcopy (though paused). |
a335debb PX |
4046 | * The dirty bitmap won't change. We can directly modify it. |
4047 | */ | |
4048 | bitmap_from_le(block->bmap, le_bitmap, nbits); | |
4049 | ||
4050 | /* | |
4051 | * What we received is "received bitmap". Revert it as the initial | |
4052 | * dirty bitmap for this ramblock. | |
4053 | */ | |
4054 | bitmap_complement(block->bmap, block->bmap, nbits); | |
4055 | ||
4056 | trace_ram_dirty_bitmap_reload_complete(block->idstr); | |
4057 | ||
edd090c7 PX |
4058 | /* |
4059 | * We succeeded to sync bitmap for current ramblock. If this is | |
4060 | * the last one to sync, we need to notify the main send thread. | |
4061 | */ | |
4062 | ram_dirty_bitmap_reload_notify(s); | |
4063 | ||
a335debb PX |
4064 | ret = 0; |
4065 | out: | |
bf269906 | 4066 | g_free(le_bitmap); |
a335debb PX |
4067 | return ret; |
4068 | } | |
4069 | ||
edd090c7 PX |
4070 | static int ram_resume_prepare(MigrationState *s, void *opaque) |
4071 | { | |
4072 | RAMState *rs = *(RAMState **)opaque; | |
08614f34 | 4073 | int ret; |
edd090c7 | 4074 | |
08614f34 PX |
4075 | ret = ram_dirty_bitmap_sync_all(s, rs); |
4076 | if (ret) { | |
4077 | return ret; | |
4078 | } | |
4079 | ||
4080 | ram_state_resume_prepare(rs, s->to_dst_file); | |
4081 | ||
4082 | return 0; | |
edd090c7 PX |
4083 | } |
4084 | ||
56e93d26 | 4085 | static SaveVMHandlers savevm_ram_handlers = { |
9907e842 | 4086 | .save_setup = ram_save_setup, |
56e93d26 | 4087 | .save_live_iterate = ram_save_iterate, |
763c906b | 4088 | .save_live_complete_postcopy = ram_save_complete, |
a3e06c3d | 4089 | .save_live_complete_precopy = ram_save_complete, |
c6467627 | 4090 | .has_postcopy = ram_has_postcopy, |
56e93d26 JQ |
4091 | .save_live_pending = ram_save_pending, |
4092 | .load_state = ram_load, | |
f265e0e4 JQ |
4093 | .save_cleanup = ram_save_cleanup, |
4094 | .load_setup = ram_load_setup, | |
4095 | .load_cleanup = ram_load_cleanup, | |
edd090c7 | 4096 | .resume_prepare = ram_resume_prepare, |
56e93d26 JQ |
4097 | }; |
4098 | ||
4099 | void ram_mig_init(void) | |
4100 | { | |
4101 | qemu_mutex_init(&XBZRLE.lock); | |
ce62df53 | 4102 | register_savevm_live("ram", 0, 4, &savevm_ram_handlers, &ram_state); |
56e93d26 | 4103 | } |