]>
Commit | Line | Data |
---|---|---|
56e93d26 JQ |
1 | /* |
2 | * QEMU System Emulator | |
3 | * | |
4 | * Copyright (c) 2003-2008 Fabrice Bellard | |
76cc7b58 JQ |
5 | * Copyright (c) 2011-2015 Red Hat Inc |
6 | * | |
7 | * Authors: | |
8 | * Juan Quintela <quintela@redhat.com> | |
56e93d26 JQ |
9 | * |
10 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
11 | * of this software and associated documentation files (the "Software"), to deal | |
12 | * in the Software without restriction, including without limitation the rights | |
13 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
14 | * copies of the Software, and to permit persons to whom the Software is | |
15 | * furnished to do so, subject to the following conditions: | |
16 | * | |
17 | * The above copyright notice and this permission notice shall be included in | |
18 | * all copies or substantial portions of the Software. | |
19 | * | |
20 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
21 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
22 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
23 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
24 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
25 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
26 | * THE SOFTWARE. | |
27 | */ | |
e688df6b | 28 | |
1393a485 | 29 | #include "qemu/osdep.h" |
33c11879 | 30 | #include "cpu.h" |
56e93d26 | 31 | #include <zlib.h> |
f348b6d1 | 32 | #include "qemu/cutils.h" |
56e93d26 JQ |
33 | #include "qemu/bitops.h" |
34 | #include "qemu/bitmap.h" | |
7205c9ec | 35 | #include "qemu/main-loop.h" |
56eb90af | 36 | #include "qemu/pmem.h" |
709e3fe8 | 37 | #include "xbzrle.h" |
7b1e1a22 | 38 | #include "ram.h" |
6666c96a | 39 | #include "migration.h" |
71bb07db | 40 | #include "socket.h" |
f2a8f0a6 | 41 | #include "migration/register.h" |
7b1e1a22 | 42 | #include "migration/misc.h" |
08a0aee1 | 43 | #include "qemu-file.h" |
be07b0ac | 44 | #include "postcopy-ram.h" |
53d37d36 | 45 | #include "page_cache.h" |
56e93d26 | 46 | #include "qemu/error-report.h" |
e688df6b | 47 | #include "qapi/error.h" |
9af23989 | 48 | #include "qapi/qapi-events-migration.h" |
8acabf69 | 49 | #include "qapi/qmp/qerror.h" |
56e93d26 | 50 | #include "trace.h" |
56e93d26 | 51 | #include "exec/ram_addr.h" |
f9494614 | 52 | #include "exec/target_page.h" |
56e93d26 | 53 | #include "qemu/rcu_queue.h" |
a91246c9 | 54 | #include "migration/colo.h" |
53d37d36 | 55 | #include "block.h" |
af8b7d2b JQ |
56 | #include "sysemu/sysemu.h" |
57 | #include "qemu/uuid.h" | |
edd090c7 | 58 | #include "savevm.h" |
b9ee2f7d | 59 | #include "qemu/iov.h" |
56e93d26 | 60 | |
56e93d26 JQ |
61 | /***********************************************************/ |
62 | /* ram save/restore */ | |
63 | ||
bb890ed5 JQ |
64 | /* RAM_SAVE_FLAG_ZERO used to be named RAM_SAVE_FLAG_COMPRESS, it |
65 | * worked for pages that where filled with the same char. We switched | |
66 | * it to only search for the zero value. And to avoid confusion with | |
67 | * RAM_SSAVE_FLAG_COMPRESS_PAGE just rename it. | |
68 | */ | |
69 | ||
56e93d26 | 70 | #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */ |
bb890ed5 | 71 | #define RAM_SAVE_FLAG_ZERO 0x02 |
56e93d26 JQ |
72 | #define RAM_SAVE_FLAG_MEM_SIZE 0x04 |
73 | #define RAM_SAVE_FLAG_PAGE 0x08 | |
74 | #define RAM_SAVE_FLAG_EOS 0x10 | |
75 | #define RAM_SAVE_FLAG_CONTINUE 0x20 | |
76 | #define RAM_SAVE_FLAG_XBZRLE 0x40 | |
77 | /* 0x80 is reserved in migration.h start with 0x100 next */ | |
78 | #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 | |
79 | ||
56e93d26 JQ |
80 | static inline bool is_zero_range(uint8_t *p, uint64_t size) |
81 | { | |
a1febc49 | 82 | return buffer_is_zero(p, size); |
56e93d26 JQ |
83 | } |
84 | ||
9360447d JQ |
85 | XBZRLECacheStats xbzrle_counters; |
86 | ||
56e93d26 JQ |
87 | /* struct contains XBZRLE cache and a static page |
88 | used by the compression */ | |
89 | static struct { | |
90 | /* buffer used for XBZRLE encoding */ | |
91 | uint8_t *encoded_buf; | |
92 | /* buffer for storing page content */ | |
93 | uint8_t *current_buf; | |
94 | /* Cache for XBZRLE, Protected by lock. */ | |
95 | PageCache *cache; | |
96 | QemuMutex lock; | |
c00e0928 JQ |
97 | /* it will store a page full of zeros */ |
98 | uint8_t *zero_target_page; | |
f265e0e4 JQ |
99 | /* buffer used for XBZRLE decoding */ |
100 | uint8_t *decoded_buf; | |
56e93d26 JQ |
101 | } XBZRLE; |
102 | ||
56e93d26 JQ |
103 | static void XBZRLE_cache_lock(void) |
104 | { | |
105 | if (migrate_use_xbzrle()) | |
106 | qemu_mutex_lock(&XBZRLE.lock); | |
107 | } | |
108 | ||
109 | static void XBZRLE_cache_unlock(void) | |
110 | { | |
111 | if (migrate_use_xbzrle()) | |
112 | qemu_mutex_unlock(&XBZRLE.lock); | |
113 | } | |
114 | ||
3d0684b2 JQ |
115 | /** |
116 | * xbzrle_cache_resize: resize the xbzrle cache | |
117 | * | |
118 | * This function is called from qmp_migrate_set_cache_size in main | |
119 | * thread, possibly while a migration is in progress. A running | |
120 | * migration may be using the cache and might finish during this call, | |
121 | * hence changes to the cache are protected by XBZRLE.lock(). | |
122 | * | |
c9dede2d | 123 | * Returns 0 for success or -1 for error |
3d0684b2 JQ |
124 | * |
125 | * @new_size: new cache size | |
8acabf69 | 126 | * @errp: set *errp if the check failed, with reason |
56e93d26 | 127 | */ |
c9dede2d | 128 | int xbzrle_cache_resize(int64_t new_size, Error **errp) |
56e93d26 JQ |
129 | { |
130 | PageCache *new_cache; | |
c9dede2d | 131 | int64_t ret = 0; |
56e93d26 | 132 | |
8acabf69 JQ |
133 | /* Check for truncation */ |
134 | if (new_size != (size_t)new_size) { | |
135 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", | |
136 | "exceeding address space"); | |
137 | return -1; | |
138 | } | |
139 | ||
2a313e5c JQ |
140 | if (new_size == migrate_xbzrle_cache_size()) { |
141 | /* nothing to do */ | |
c9dede2d | 142 | return 0; |
2a313e5c JQ |
143 | } |
144 | ||
56e93d26 JQ |
145 | XBZRLE_cache_lock(); |
146 | ||
147 | if (XBZRLE.cache != NULL) { | |
80f8dfde | 148 | new_cache = cache_init(new_size, TARGET_PAGE_SIZE, errp); |
56e93d26 | 149 | if (!new_cache) { |
56e93d26 JQ |
150 | ret = -1; |
151 | goto out; | |
152 | } | |
153 | ||
154 | cache_fini(XBZRLE.cache); | |
155 | XBZRLE.cache = new_cache; | |
156 | } | |
56e93d26 JQ |
157 | out: |
158 | XBZRLE_cache_unlock(); | |
159 | return ret; | |
160 | } | |
161 | ||
b895de50 CLG |
162 | /* Should be holding either ram_list.mutex, or the RCU lock. */ |
163 | #define RAMBLOCK_FOREACH_MIGRATABLE(block) \ | |
343f632c | 164 | INTERNAL_RAMBLOCK_FOREACH(block) \ |
b895de50 CLG |
165 | if (!qemu_ram_is_migratable(block)) {} else |
166 | ||
343f632c DDAG |
167 | #undef RAMBLOCK_FOREACH |
168 | ||
f9494614 AP |
169 | static void ramblock_recv_map_init(void) |
170 | { | |
171 | RAMBlock *rb; | |
172 | ||
b895de50 | 173 | RAMBLOCK_FOREACH_MIGRATABLE(rb) { |
f9494614 AP |
174 | assert(!rb->receivedmap); |
175 | rb->receivedmap = bitmap_new(rb->max_length >> qemu_target_page_bits()); | |
176 | } | |
177 | } | |
178 | ||
179 | int ramblock_recv_bitmap_test(RAMBlock *rb, void *host_addr) | |
180 | { | |
181 | return test_bit(ramblock_recv_bitmap_offset(host_addr, rb), | |
182 | rb->receivedmap); | |
183 | } | |
184 | ||
1cba9f6e DDAG |
185 | bool ramblock_recv_bitmap_test_byte_offset(RAMBlock *rb, uint64_t byte_offset) |
186 | { | |
187 | return test_bit(byte_offset >> TARGET_PAGE_BITS, rb->receivedmap); | |
188 | } | |
189 | ||
f9494614 AP |
190 | void ramblock_recv_bitmap_set(RAMBlock *rb, void *host_addr) |
191 | { | |
192 | set_bit_atomic(ramblock_recv_bitmap_offset(host_addr, rb), rb->receivedmap); | |
193 | } | |
194 | ||
195 | void ramblock_recv_bitmap_set_range(RAMBlock *rb, void *host_addr, | |
196 | size_t nr) | |
197 | { | |
198 | bitmap_set_atomic(rb->receivedmap, | |
199 | ramblock_recv_bitmap_offset(host_addr, rb), | |
200 | nr); | |
201 | } | |
202 | ||
a335debb PX |
203 | #define RAMBLOCK_RECV_BITMAP_ENDING (0x0123456789abcdefULL) |
204 | ||
205 | /* | |
206 | * Format: bitmap_size (8 bytes) + whole_bitmap (N bytes). | |
207 | * | |
208 | * Returns >0 if success with sent bytes, or <0 if error. | |
209 | */ | |
210 | int64_t ramblock_recv_bitmap_send(QEMUFile *file, | |
211 | const char *block_name) | |
212 | { | |
213 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
214 | unsigned long *le_bitmap, nbits; | |
215 | uint64_t size; | |
216 | ||
217 | if (!block) { | |
218 | error_report("%s: invalid block name: %s", __func__, block_name); | |
219 | return -1; | |
220 | } | |
221 | ||
222 | nbits = block->used_length >> TARGET_PAGE_BITS; | |
223 | ||
224 | /* | |
225 | * Make sure the tmp bitmap buffer is big enough, e.g., on 32bit | |
226 | * machines we may need 4 more bytes for padding (see below | |
227 | * comment). So extend it a bit before hand. | |
228 | */ | |
229 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
230 | ||
231 | /* | |
232 | * Always use little endian when sending the bitmap. This is | |
233 | * required that when source and destination VMs are not using the | |
234 | * same endianess. (Note: big endian won't work.) | |
235 | */ | |
236 | bitmap_to_le(le_bitmap, block->receivedmap, nbits); | |
237 | ||
238 | /* Size of the bitmap, in bytes */ | |
a725ef9f | 239 | size = DIV_ROUND_UP(nbits, 8); |
a335debb PX |
240 | |
241 | /* | |
242 | * size is always aligned to 8 bytes for 64bit machines, but it | |
243 | * may not be true for 32bit machines. We need this padding to | |
244 | * make sure the migration can survive even between 32bit and | |
245 | * 64bit machines. | |
246 | */ | |
247 | size = ROUND_UP(size, 8); | |
248 | ||
249 | qemu_put_be64(file, size); | |
250 | qemu_put_buffer(file, (const uint8_t *)le_bitmap, size); | |
251 | /* | |
252 | * Mark as an end, in case the middle part is screwed up due to | |
253 | * some "misterious" reason. | |
254 | */ | |
255 | qemu_put_be64(file, RAMBLOCK_RECV_BITMAP_ENDING); | |
256 | qemu_fflush(file); | |
257 | ||
bf269906 | 258 | g_free(le_bitmap); |
a335debb PX |
259 | |
260 | if (qemu_file_get_error(file)) { | |
261 | return qemu_file_get_error(file); | |
262 | } | |
263 | ||
264 | return size + sizeof(size); | |
265 | } | |
266 | ||
ec481c6c JQ |
267 | /* |
268 | * An outstanding page request, on the source, having been received | |
269 | * and queued | |
270 | */ | |
271 | struct RAMSrcPageRequest { | |
272 | RAMBlock *rb; | |
273 | hwaddr offset; | |
274 | hwaddr len; | |
275 | ||
276 | QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; | |
277 | }; | |
278 | ||
6f37bb8b JQ |
279 | /* State of RAM for migration */ |
280 | struct RAMState { | |
204b88b8 JQ |
281 | /* QEMUFile used for this migration */ |
282 | QEMUFile *f; | |
6f37bb8b JQ |
283 | /* Last block that we have visited searching for dirty pages */ |
284 | RAMBlock *last_seen_block; | |
285 | /* Last block from where we have sent data */ | |
286 | RAMBlock *last_sent_block; | |
269ace29 JQ |
287 | /* Last dirty target page we have sent */ |
288 | ram_addr_t last_page; | |
6f37bb8b JQ |
289 | /* last ram version we have seen */ |
290 | uint32_t last_version; | |
291 | /* We are in the first round */ | |
292 | bool ram_bulk_stage; | |
8d820d6f JQ |
293 | /* How many times we have dirty too many pages */ |
294 | int dirty_rate_high_cnt; | |
f664da80 JQ |
295 | /* these variables are used for bitmap sync */ |
296 | /* last time we did a full bitmap_sync */ | |
297 | int64_t time_last_bitmap_sync; | |
eac74159 | 298 | /* bytes transferred at start_time */ |
c4bdf0cf | 299 | uint64_t bytes_xfer_prev; |
a66cd90c | 300 | /* number of dirty pages since start_time */ |
68908ed6 | 301 | uint64_t num_dirty_pages_period; |
b5833fde JQ |
302 | /* xbzrle misses since the beginning of the period */ |
303 | uint64_t xbzrle_cache_miss_prev; | |
36040d9c JQ |
304 | /* number of iterations at the beginning of period */ |
305 | uint64_t iterations_prev; | |
23b28c3c JQ |
306 | /* Iterations since start */ |
307 | uint64_t iterations; | |
9360447d | 308 | /* number of dirty bits in the bitmap */ |
2dfaf12e PX |
309 | uint64_t migration_dirty_pages; |
310 | /* protects modification of the bitmap */ | |
108cfae0 | 311 | QemuMutex bitmap_mutex; |
68a098f3 JQ |
312 | /* The RAMBlock used in the last src_page_requests */ |
313 | RAMBlock *last_req_rb; | |
ec481c6c JQ |
314 | /* Queue of outstanding page requests from the destination */ |
315 | QemuMutex src_page_req_mutex; | |
316 | QSIMPLEQ_HEAD(src_page_requests, RAMSrcPageRequest) src_page_requests; | |
6f37bb8b JQ |
317 | }; |
318 | typedef struct RAMState RAMState; | |
319 | ||
53518d94 | 320 | static RAMState *ram_state; |
6f37bb8b | 321 | |
9edabd4d | 322 | uint64_t ram_bytes_remaining(void) |
2f4fde93 | 323 | { |
bae416e5 DDAG |
324 | return ram_state ? (ram_state->migration_dirty_pages * TARGET_PAGE_SIZE) : |
325 | 0; | |
2f4fde93 JQ |
326 | } |
327 | ||
9360447d | 328 | MigrationStats ram_counters; |
96506894 | 329 | |
b8fb8cb7 DDAG |
330 | /* used by the search for pages to send */ |
331 | struct PageSearchStatus { | |
332 | /* Current block being searched */ | |
333 | RAMBlock *block; | |
a935e30f JQ |
334 | /* Current page to search from */ |
335 | unsigned long page; | |
b8fb8cb7 DDAG |
336 | /* Set once we wrap around */ |
337 | bool complete_round; | |
338 | }; | |
339 | typedef struct PageSearchStatus PageSearchStatus; | |
340 | ||
56e93d26 | 341 | struct CompressParam { |
56e93d26 | 342 | bool done; |
90e56fb4 | 343 | bool quit; |
56e93d26 JQ |
344 | QEMUFile *file; |
345 | QemuMutex mutex; | |
346 | QemuCond cond; | |
347 | RAMBlock *block; | |
348 | ram_addr_t offset; | |
34ab9e97 XG |
349 | |
350 | /* internally used fields */ | |
dcaf446e | 351 | z_stream stream; |
34ab9e97 | 352 | uint8_t *originbuf; |
56e93d26 JQ |
353 | }; |
354 | typedef struct CompressParam CompressParam; | |
355 | ||
356 | struct DecompressParam { | |
73a8912b | 357 | bool done; |
90e56fb4 | 358 | bool quit; |
56e93d26 JQ |
359 | QemuMutex mutex; |
360 | QemuCond cond; | |
361 | void *des; | |
d341d9f3 | 362 | uint8_t *compbuf; |
56e93d26 | 363 | int len; |
797ca154 | 364 | z_stream stream; |
56e93d26 JQ |
365 | }; |
366 | typedef struct DecompressParam DecompressParam; | |
367 | ||
368 | static CompressParam *comp_param; | |
369 | static QemuThread *compress_threads; | |
370 | /* comp_done_cond is used to wake up the migration thread when | |
371 | * one of the compression threads has finished the compression. | |
372 | * comp_done_lock is used to co-work with comp_done_cond. | |
373 | */ | |
0d9f9a5c LL |
374 | static QemuMutex comp_done_lock; |
375 | static QemuCond comp_done_cond; | |
56e93d26 JQ |
376 | /* The empty QEMUFileOps will be used by file in CompressParam */ |
377 | static const QEMUFileOps empty_ops = { }; | |
378 | ||
34ab9e97 | 379 | static QEMUFile *decomp_file; |
56e93d26 JQ |
380 | static DecompressParam *decomp_param; |
381 | static QemuThread *decompress_threads; | |
73a8912b LL |
382 | static QemuMutex decomp_done_lock; |
383 | static QemuCond decomp_done_cond; | |
56e93d26 | 384 | |
6ef3771c XG |
385 | static void do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
386 | ram_addr_t offset, uint8_t *source_buf); | |
56e93d26 JQ |
387 | |
388 | static void *do_data_compress(void *opaque) | |
389 | { | |
390 | CompressParam *param = opaque; | |
a7a9a88f LL |
391 | RAMBlock *block; |
392 | ram_addr_t offset; | |
56e93d26 | 393 | |
a7a9a88f | 394 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 395 | while (!param->quit) { |
a7a9a88f LL |
396 | if (param->block) { |
397 | block = param->block; | |
398 | offset = param->offset; | |
399 | param->block = NULL; | |
400 | qemu_mutex_unlock(¶m->mutex); | |
401 | ||
34ab9e97 XG |
402 | do_compress_ram_page(param->file, ¶m->stream, block, offset, |
403 | param->originbuf); | |
a7a9a88f | 404 | |
0d9f9a5c | 405 | qemu_mutex_lock(&comp_done_lock); |
a7a9a88f | 406 | param->done = true; |
0d9f9a5c LL |
407 | qemu_cond_signal(&comp_done_cond); |
408 | qemu_mutex_unlock(&comp_done_lock); | |
a7a9a88f LL |
409 | |
410 | qemu_mutex_lock(¶m->mutex); | |
411 | } else { | |
56e93d26 JQ |
412 | qemu_cond_wait(¶m->cond, ¶m->mutex); |
413 | } | |
56e93d26 | 414 | } |
a7a9a88f | 415 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
416 | |
417 | return NULL; | |
418 | } | |
419 | ||
420 | static inline void terminate_compression_threads(void) | |
421 | { | |
422 | int idx, thread_count; | |
423 | ||
424 | thread_count = migrate_compress_threads(); | |
3d0684b2 | 425 | |
56e93d26 JQ |
426 | for (idx = 0; idx < thread_count; idx++) { |
427 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 428 | comp_param[idx].quit = true; |
56e93d26 JQ |
429 | qemu_cond_signal(&comp_param[idx].cond); |
430 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
431 | } | |
432 | } | |
433 | ||
f0afa331 | 434 | static void compress_threads_save_cleanup(void) |
56e93d26 JQ |
435 | { |
436 | int i, thread_count; | |
437 | ||
438 | if (!migrate_use_compression()) { | |
439 | return; | |
440 | } | |
441 | terminate_compression_threads(); | |
442 | thread_count = migrate_compress_threads(); | |
443 | for (i = 0; i < thread_count; i++) { | |
dcaf446e XG |
444 | /* |
445 | * we use it as a indicator which shows if the thread is | |
446 | * properly init'd or not | |
447 | */ | |
448 | if (!comp_param[i].file) { | |
449 | break; | |
450 | } | |
56e93d26 | 451 | qemu_thread_join(compress_threads + i); |
56e93d26 JQ |
452 | qemu_mutex_destroy(&comp_param[i].mutex); |
453 | qemu_cond_destroy(&comp_param[i].cond); | |
dcaf446e | 454 | deflateEnd(&comp_param[i].stream); |
34ab9e97 | 455 | g_free(comp_param[i].originbuf); |
dcaf446e XG |
456 | qemu_fclose(comp_param[i].file); |
457 | comp_param[i].file = NULL; | |
56e93d26 | 458 | } |
0d9f9a5c LL |
459 | qemu_mutex_destroy(&comp_done_lock); |
460 | qemu_cond_destroy(&comp_done_cond); | |
56e93d26 JQ |
461 | g_free(compress_threads); |
462 | g_free(comp_param); | |
56e93d26 JQ |
463 | compress_threads = NULL; |
464 | comp_param = NULL; | |
56e93d26 JQ |
465 | } |
466 | ||
dcaf446e | 467 | static int compress_threads_save_setup(void) |
56e93d26 JQ |
468 | { |
469 | int i, thread_count; | |
470 | ||
471 | if (!migrate_use_compression()) { | |
dcaf446e | 472 | return 0; |
56e93d26 | 473 | } |
56e93d26 JQ |
474 | thread_count = migrate_compress_threads(); |
475 | compress_threads = g_new0(QemuThread, thread_count); | |
476 | comp_param = g_new0(CompressParam, thread_count); | |
0d9f9a5c LL |
477 | qemu_cond_init(&comp_done_cond); |
478 | qemu_mutex_init(&comp_done_lock); | |
56e93d26 | 479 | for (i = 0; i < thread_count; i++) { |
34ab9e97 XG |
480 | comp_param[i].originbuf = g_try_malloc(TARGET_PAGE_SIZE); |
481 | if (!comp_param[i].originbuf) { | |
482 | goto exit; | |
483 | } | |
484 | ||
dcaf446e XG |
485 | if (deflateInit(&comp_param[i].stream, |
486 | migrate_compress_level()) != Z_OK) { | |
34ab9e97 | 487 | g_free(comp_param[i].originbuf); |
dcaf446e XG |
488 | goto exit; |
489 | } | |
490 | ||
e110aa91 C |
491 | /* comp_param[i].file is just used as a dummy buffer to save data, |
492 | * set its ops to empty. | |
56e93d26 JQ |
493 | */ |
494 | comp_param[i].file = qemu_fopen_ops(NULL, &empty_ops); | |
495 | comp_param[i].done = true; | |
90e56fb4 | 496 | comp_param[i].quit = false; |
56e93d26 JQ |
497 | qemu_mutex_init(&comp_param[i].mutex); |
498 | qemu_cond_init(&comp_param[i].cond); | |
499 | qemu_thread_create(compress_threads + i, "compress", | |
500 | do_data_compress, comp_param + i, | |
501 | QEMU_THREAD_JOINABLE); | |
502 | } | |
dcaf446e XG |
503 | return 0; |
504 | ||
505 | exit: | |
506 | compress_threads_save_cleanup(); | |
507 | return -1; | |
56e93d26 JQ |
508 | } |
509 | ||
f986c3d2 JQ |
510 | /* Multiple fd's */ |
511 | ||
af8b7d2b JQ |
512 | #define MULTIFD_MAGIC 0x11223344U |
513 | #define MULTIFD_VERSION 1 | |
514 | ||
6df264ac JQ |
515 | #define MULTIFD_FLAG_SYNC (1 << 0) |
516 | ||
af8b7d2b JQ |
517 | typedef struct { |
518 | uint32_t magic; | |
519 | uint32_t version; | |
520 | unsigned char uuid[16]; /* QemuUUID */ | |
521 | uint8_t id; | |
522 | } __attribute__((packed)) MultiFDInit_t; | |
523 | ||
2a26c979 JQ |
524 | typedef struct { |
525 | uint32_t magic; | |
526 | uint32_t version; | |
527 | uint32_t flags; | |
528 | uint32_t size; | |
529 | uint32_t used; | |
530 | uint64_t packet_num; | |
531 | char ramblock[256]; | |
532 | uint64_t offset[]; | |
533 | } __attribute__((packed)) MultiFDPacket_t; | |
534 | ||
34c55a94 JQ |
535 | typedef struct { |
536 | /* number of used pages */ | |
537 | uint32_t used; | |
538 | /* number of allocated pages */ | |
539 | uint32_t allocated; | |
540 | /* global number of generated multifd packets */ | |
541 | uint64_t packet_num; | |
542 | /* offset of each page */ | |
543 | ram_addr_t *offset; | |
544 | /* pointer to each page */ | |
545 | struct iovec *iov; | |
546 | RAMBlock *block; | |
547 | } MultiFDPages_t; | |
548 | ||
8c4598f2 JQ |
549 | typedef struct { |
550 | /* this fields are not changed once the thread is created */ | |
551 | /* channel number */ | |
f986c3d2 | 552 | uint8_t id; |
8c4598f2 | 553 | /* channel thread name */ |
f986c3d2 | 554 | char *name; |
8c4598f2 | 555 | /* channel thread id */ |
f986c3d2 | 556 | QemuThread thread; |
8c4598f2 | 557 | /* communication channel */ |
60df2d4a | 558 | QIOChannel *c; |
8c4598f2 | 559 | /* sem where to wait for more work */ |
f986c3d2 | 560 | QemuSemaphore sem; |
8c4598f2 | 561 | /* this mutex protects the following parameters */ |
f986c3d2 | 562 | QemuMutex mutex; |
8c4598f2 | 563 | /* is this channel thread running */ |
66770707 | 564 | bool running; |
8c4598f2 | 565 | /* should this thread finish */ |
f986c3d2 | 566 | bool quit; |
0beb5ed3 JQ |
567 | /* thread has work to do */ |
568 | int pending_job; | |
34c55a94 JQ |
569 | /* array of pages to sent */ |
570 | MultiFDPages_t *pages; | |
2a26c979 JQ |
571 | /* packet allocated len */ |
572 | uint32_t packet_len; | |
573 | /* pointer to the packet */ | |
574 | MultiFDPacket_t *packet; | |
575 | /* multifd flags for each packet */ | |
576 | uint32_t flags; | |
577 | /* global number of generated multifd packets */ | |
578 | uint64_t packet_num; | |
408ea6ae JQ |
579 | /* thread local variables */ |
580 | /* packets sent through this channel */ | |
581 | uint64_t num_packets; | |
582 | /* pages sent through this channel */ | |
583 | uint64_t num_pages; | |
6df264ac JQ |
584 | /* syncs main thread and channels */ |
585 | QemuSemaphore sem_sync; | |
8c4598f2 JQ |
586 | } MultiFDSendParams; |
587 | ||
588 | typedef struct { | |
589 | /* this fields are not changed once the thread is created */ | |
590 | /* channel number */ | |
591 | uint8_t id; | |
592 | /* channel thread name */ | |
593 | char *name; | |
594 | /* channel thread id */ | |
595 | QemuThread thread; | |
596 | /* communication channel */ | |
597 | QIOChannel *c; | |
8c4598f2 JQ |
598 | /* this mutex protects the following parameters */ |
599 | QemuMutex mutex; | |
600 | /* is this channel thread running */ | |
601 | bool running; | |
34c55a94 JQ |
602 | /* array of pages to receive */ |
603 | MultiFDPages_t *pages; | |
2a26c979 JQ |
604 | /* packet allocated len */ |
605 | uint32_t packet_len; | |
606 | /* pointer to the packet */ | |
607 | MultiFDPacket_t *packet; | |
608 | /* multifd flags for each packet */ | |
609 | uint32_t flags; | |
610 | /* global number of generated multifd packets */ | |
611 | uint64_t packet_num; | |
408ea6ae JQ |
612 | /* thread local variables */ |
613 | /* packets sent through this channel */ | |
614 | uint64_t num_packets; | |
615 | /* pages sent through this channel */ | |
616 | uint64_t num_pages; | |
6df264ac JQ |
617 | /* syncs main thread and channels */ |
618 | QemuSemaphore sem_sync; | |
8c4598f2 | 619 | } MultiFDRecvParams; |
f986c3d2 | 620 | |
af8b7d2b JQ |
621 | static int multifd_send_initial_packet(MultiFDSendParams *p, Error **errp) |
622 | { | |
623 | MultiFDInit_t msg; | |
624 | int ret; | |
625 | ||
626 | msg.magic = cpu_to_be32(MULTIFD_MAGIC); | |
627 | msg.version = cpu_to_be32(MULTIFD_VERSION); | |
628 | msg.id = p->id; | |
629 | memcpy(msg.uuid, &qemu_uuid.data, sizeof(msg.uuid)); | |
630 | ||
631 | ret = qio_channel_write_all(p->c, (char *)&msg, sizeof(msg), errp); | |
632 | if (ret != 0) { | |
633 | return -1; | |
634 | } | |
635 | return 0; | |
636 | } | |
637 | ||
638 | static int multifd_recv_initial_packet(QIOChannel *c, Error **errp) | |
639 | { | |
640 | MultiFDInit_t msg; | |
641 | int ret; | |
642 | ||
643 | ret = qio_channel_read_all(c, (char *)&msg, sizeof(msg), errp); | |
644 | if (ret != 0) { | |
645 | return -1; | |
646 | } | |
647 | ||
648 | be32_to_cpus(&msg.magic); | |
649 | be32_to_cpus(&msg.version); | |
650 | ||
651 | if (msg.magic != MULTIFD_MAGIC) { | |
652 | error_setg(errp, "multifd: received packet magic %x " | |
653 | "expected %x", msg.magic, MULTIFD_MAGIC); | |
654 | return -1; | |
655 | } | |
656 | ||
657 | if (msg.version != MULTIFD_VERSION) { | |
658 | error_setg(errp, "multifd: received packet version %d " | |
659 | "expected %d", msg.version, MULTIFD_VERSION); | |
660 | return -1; | |
661 | } | |
662 | ||
663 | if (memcmp(msg.uuid, &qemu_uuid, sizeof(qemu_uuid))) { | |
664 | char *uuid = qemu_uuid_unparse_strdup(&qemu_uuid); | |
665 | char *msg_uuid = qemu_uuid_unparse_strdup((const QemuUUID *)msg.uuid); | |
666 | ||
667 | error_setg(errp, "multifd: received uuid '%s' and expected " | |
668 | "uuid '%s' for channel %hhd", msg_uuid, uuid, msg.id); | |
669 | g_free(uuid); | |
670 | g_free(msg_uuid); | |
671 | return -1; | |
672 | } | |
673 | ||
674 | if (msg.id > migrate_multifd_channels()) { | |
675 | error_setg(errp, "multifd: received channel version %d " | |
676 | "expected %d", msg.version, MULTIFD_VERSION); | |
677 | return -1; | |
678 | } | |
679 | ||
680 | return msg.id; | |
681 | } | |
682 | ||
34c55a94 JQ |
683 | static MultiFDPages_t *multifd_pages_init(size_t size) |
684 | { | |
685 | MultiFDPages_t *pages = g_new0(MultiFDPages_t, 1); | |
686 | ||
687 | pages->allocated = size; | |
688 | pages->iov = g_new0(struct iovec, size); | |
689 | pages->offset = g_new0(ram_addr_t, size); | |
690 | ||
691 | return pages; | |
692 | } | |
693 | ||
694 | static void multifd_pages_clear(MultiFDPages_t *pages) | |
695 | { | |
696 | pages->used = 0; | |
697 | pages->allocated = 0; | |
698 | pages->packet_num = 0; | |
699 | pages->block = NULL; | |
700 | g_free(pages->iov); | |
701 | pages->iov = NULL; | |
702 | g_free(pages->offset); | |
703 | pages->offset = NULL; | |
704 | g_free(pages); | |
705 | } | |
706 | ||
2a26c979 JQ |
707 | static void multifd_send_fill_packet(MultiFDSendParams *p) |
708 | { | |
709 | MultiFDPacket_t *packet = p->packet; | |
710 | int i; | |
711 | ||
712 | packet->magic = cpu_to_be32(MULTIFD_MAGIC); | |
713 | packet->version = cpu_to_be32(MULTIFD_VERSION); | |
714 | packet->flags = cpu_to_be32(p->flags); | |
715 | packet->size = cpu_to_be32(migrate_multifd_page_count()); | |
716 | packet->used = cpu_to_be32(p->pages->used); | |
717 | packet->packet_num = cpu_to_be64(p->packet_num); | |
718 | ||
719 | if (p->pages->block) { | |
720 | strncpy(packet->ramblock, p->pages->block->idstr, 256); | |
721 | } | |
722 | ||
723 | for (i = 0; i < p->pages->used; i++) { | |
724 | packet->offset[i] = cpu_to_be64(p->pages->offset[i]); | |
725 | } | |
726 | } | |
727 | ||
728 | static int multifd_recv_unfill_packet(MultiFDRecvParams *p, Error **errp) | |
729 | { | |
730 | MultiFDPacket_t *packet = p->packet; | |
731 | RAMBlock *block; | |
732 | int i; | |
733 | ||
2a26c979 JQ |
734 | be32_to_cpus(&packet->magic); |
735 | if (packet->magic != MULTIFD_MAGIC) { | |
736 | error_setg(errp, "multifd: received packet " | |
737 | "magic %x and expected magic %x", | |
738 | packet->magic, MULTIFD_MAGIC); | |
739 | return -1; | |
740 | } | |
741 | ||
742 | be32_to_cpus(&packet->version); | |
743 | if (packet->version != MULTIFD_VERSION) { | |
744 | error_setg(errp, "multifd: received packet " | |
745 | "version %d and expected version %d", | |
746 | packet->version, MULTIFD_VERSION); | |
747 | return -1; | |
748 | } | |
749 | ||
750 | p->flags = be32_to_cpu(packet->flags); | |
751 | ||
752 | be32_to_cpus(&packet->size); | |
753 | if (packet->size > migrate_multifd_page_count()) { | |
754 | error_setg(errp, "multifd: received packet " | |
755 | "with size %d and expected maximum size %d", | |
756 | packet->size, migrate_multifd_page_count()) ; | |
757 | return -1; | |
758 | } | |
759 | ||
760 | p->pages->used = be32_to_cpu(packet->used); | |
761 | if (p->pages->used > packet->size) { | |
762 | error_setg(errp, "multifd: received packet " | |
763 | "with size %d and expected maximum size %d", | |
764 | p->pages->used, packet->size) ; | |
765 | return -1; | |
766 | } | |
767 | ||
768 | p->packet_num = be64_to_cpu(packet->packet_num); | |
769 | ||
770 | if (p->pages->used) { | |
771 | /* make sure that ramblock is 0 terminated */ | |
772 | packet->ramblock[255] = 0; | |
773 | block = qemu_ram_block_by_name(packet->ramblock); | |
774 | if (!block) { | |
775 | error_setg(errp, "multifd: unknown ram block %s", | |
776 | packet->ramblock); | |
777 | return -1; | |
778 | } | |
779 | } | |
780 | ||
781 | for (i = 0; i < p->pages->used; i++) { | |
782 | ram_addr_t offset = be64_to_cpu(packet->offset[i]); | |
783 | ||
784 | if (offset > (block->used_length - TARGET_PAGE_SIZE)) { | |
785 | error_setg(errp, "multifd: offset too long " RAM_ADDR_FMT | |
786 | " (max " RAM_ADDR_FMT ")", | |
787 | offset, block->max_length); | |
788 | return -1; | |
789 | } | |
790 | p->pages->iov[i].iov_base = block->host + offset; | |
791 | p->pages->iov[i].iov_len = TARGET_PAGE_SIZE; | |
792 | } | |
793 | ||
794 | return 0; | |
795 | } | |
796 | ||
f986c3d2 JQ |
797 | struct { |
798 | MultiFDSendParams *params; | |
799 | /* number of created threads */ | |
800 | int count; | |
34c55a94 JQ |
801 | /* array of pages to sent */ |
802 | MultiFDPages_t *pages; | |
6df264ac JQ |
803 | /* syncs main thread and channels */ |
804 | QemuSemaphore sem_sync; | |
805 | /* global number of generated multifd packets */ | |
806 | uint64_t packet_num; | |
b9ee2f7d JQ |
807 | /* send channels ready */ |
808 | QemuSemaphore channels_ready; | |
f986c3d2 JQ |
809 | } *multifd_send_state; |
810 | ||
b9ee2f7d JQ |
811 | /* |
812 | * How we use multifd_send_state->pages and channel->pages? | |
813 | * | |
814 | * We create a pages for each channel, and a main one. Each time that | |
815 | * we need to send a batch of pages we interchange the ones between | |
816 | * multifd_send_state and the channel that is sending it. There are | |
817 | * two reasons for that: | |
818 | * - to not have to do so many mallocs during migration | |
819 | * - to make easier to know what to free at the end of migration | |
820 | * | |
821 | * This way we always know who is the owner of each "pages" struct, | |
822 | * and we don't need any loocking. It belongs to the migration thread | |
823 | * or to the channel thread. Switching is safe because the migration | |
824 | * thread is using the channel mutex when changing it, and the channel | |
825 | * have to had finish with its own, otherwise pending_job can't be | |
826 | * false. | |
827 | */ | |
828 | ||
829 | static void multifd_send_pages(void) | |
830 | { | |
831 | int i; | |
832 | static int next_channel; | |
833 | MultiFDSendParams *p = NULL; /* make happy gcc */ | |
834 | MultiFDPages_t *pages = multifd_send_state->pages; | |
835 | uint64_t transferred; | |
836 | ||
837 | qemu_sem_wait(&multifd_send_state->channels_ready); | |
838 | for (i = next_channel;; i = (i + 1) % migrate_multifd_channels()) { | |
839 | p = &multifd_send_state->params[i]; | |
840 | ||
841 | qemu_mutex_lock(&p->mutex); | |
842 | if (!p->pending_job) { | |
843 | p->pending_job++; | |
844 | next_channel = (i + 1) % migrate_multifd_channels(); | |
845 | break; | |
846 | } | |
847 | qemu_mutex_unlock(&p->mutex); | |
848 | } | |
849 | p->pages->used = 0; | |
850 | ||
851 | p->packet_num = multifd_send_state->packet_num++; | |
852 | p->pages->block = NULL; | |
853 | multifd_send_state->pages = p->pages; | |
854 | p->pages = pages; | |
4fcefd44 | 855 | transferred = ((uint64_t) pages->used) * TARGET_PAGE_SIZE + p->packet_len; |
b9ee2f7d JQ |
856 | ram_counters.multifd_bytes += transferred; |
857 | ram_counters.transferred += transferred;; | |
858 | qemu_mutex_unlock(&p->mutex); | |
859 | qemu_sem_post(&p->sem); | |
860 | } | |
861 | ||
862 | static void multifd_queue_page(RAMBlock *block, ram_addr_t offset) | |
863 | { | |
864 | MultiFDPages_t *pages = multifd_send_state->pages; | |
865 | ||
866 | if (!pages->block) { | |
867 | pages->block = block; | |
868 | } | |
869 | ||
870 | if (pages->block == block) { | |
871 | pages->offset[pages->used] = offset; | |
872 | pages->iov[pages->used].iov_base = block->host + offset; | |
873 | pages->iov[pages->used].iov_len = TARGET_PAGE_SIZE; | |
874 | pages->used++; | |
875 | ||
876 | if (pages->used < pages->allocated) { | |
877 | return; | |
878 | } | |
879 | } | |
880 | ||
881 | multifd_send_pages(); | |
882 | ||
883 | if (pages->block != block) { | |
884 | multifd_queue_page(block, offset); | |
885 | } | |
886 | } | |
887 | ||
66770707 | 888 | static void multifd_send_terminate_threads(Error *err) |
f986c3d2 JQ |
889 | { |
890 | int i; | |
891 | ||
7a169d74 JQ |
892 | if (err) { |
893 | MigrationState *s = migrate_get_current(); | |
894 | migrate_set_error(s, err); | |
895 | if (s->state == MIGRATION_STATUS_SETUP || | |
896 | s->state == MIGRATION_STATUS_PRE_SWITCHOVER || | |
897 | s->state == MIGRATION_STATUS_DEVICE || | |
898 | s->state == MIGRATION_STATUS_ACTIVE) { | |
899 | migrate_set_state(&s->state, s->state, | |
900 | MIGRATION_STATUS_FAILED); | |
901 | } | |
902 | } | |
903 | ||
66770707 | 904 | for (i = 0; i < migrate_multifd_channels(); i++) { |
f986c3d2 JQ |
905 | MultiFDSendParams *p = &multifd_send_state->params[i]; |
906 | ||
907 | qemu_mutex_lock(&p->mutex); | |
908 | p->quit = true; | |
909 | qemu_sem_post(&p->sem); | |
910 | qemu_mutex_unlock(&p->mutex); | |
911 | } | |
912 | } | |
913 | ||
914 | int multifd_save_cleanup(Error **errp) | |
915 | { | |
916 | int i; | |
917 | int ret = 0; | |
918 | ||
919 | if (!migrate_use_multifd()) { | |
920 | return 0; | |
921 | } | |
66770707 JQ |
922 | multifd_send_terminate_threads(NULL); |
923 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
f986c3d2 JQ |
924 | MultiFDSendParams *p = &multifd_send_state->params[i]; |
925 | ||
66770707 JQ |
926 | if (p->running) { |
927 | qemu_thread_join(&p->thread); | |
928 | } | |
60df2d4a JQ |
929 | socket_send_channel_destroy(p->c); |
930 | p->c = NULL; | |
f986c3d2 JQ |
931 | qemu_mutex_destroy(&p->mutex); |
932 | qemu_sem_destroy(&p->sem); | |
6df264ac | 933 | qemu_sem_destroy(&p->sem_sync); |
f986c3d2 JQ |
934 | g_free(p->name); |
935 | p->name = NULL; | |
34c55a94 JQ |
936 | multifd_pages_clear(p->pages); |
937 | p->pages = NULL; | |
2a26c979 JQ |
938 | p->packet_len = 0; |
939 | g_free(p->packet); | |
940 | p->packet = NULL; | |
f986c3d2 | 941 | } |
b9ee2f7d | 942 | qemu_sem_destroy(&multifd_send_state->channels_ready); |
6df264ac | 943 | qemu_sem_destroy(&multifd_send_state->sem_sync); |
f986c3d2 JQ |
944 | g_free(multifd_send_state->params); |
945 | multifd_send_state->params = NULL; | |
34c55a94 JQ |
946 | multifd_pages_clear(multifd_send_state->pages); |
947 | multifd_send_state->pages = NULL; | |
f986c3d2 JQ |
948 | g_free(multifd_send_state); |
949 | multifd_send_state = NULL; | |
950 | return ret; | |
951 | } | |
952 | ||
6df264ac JQ |
953 | static void multifd_send_sync_main(void) |
954 | { | |
955 | int i; | |
956 | ||
957 | if (!migrate_use_multifd()) { | |
958 | return; | |
959 | } | |
b9ee2f7d JQ |
960 | if (multifd_send_state->pages->used) { |
961 | multifd_send_pages(); | |
962 | } | |
6df264ac JQ |
963 | for (i = 0; i < migrate_multifd_channels(); i++) { |
964 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
965 | ||
966 | trace_multifd_send_sync_main_signal(p->id); | |
967 | ||
968 | qemu_mutex_lock(&p->mutex); | |
b9ee2f7d JQ |
969 | |
970 | p->packet_num = multifd_send_state->packet_num++; | |
6df264ac JQ |
971 | p->flags |= MULTIFD_FLAG_SYNC; |
972 | p->pending_job++; | |
973 | qemu_mutex_unlock(&p->mutex); | |
974 | qemu_sem_post(&p->sem); | |
975 | } | |
976 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
977 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
978 | ||
979 | trace_multifd_send_sync_main_wait(p->id); | |
980 | qemu_sem_wait(&multifd_send_state->sem_sync); | |
981 | } | |
982 | trace_multifd_send_sync_main(multifd_send_state->packet_num); | |
983 | } | |
984 | ||
f986c3d2 JQ |
985 | static void *multifd_send_thread(void *opaque) |
986 | { | |
987 | MultiFDSendParams *p = opaque; | |
af8b7d2b | 988 | Error *local_err = NULL; |
8b2db7f5 | 989 | int ret; |
af8b7d2b | 990 | |
408ea6ae | 991 | trace_multifd_send_thread_start(p->id); |
74637e6f | 992 | rcu_register_thread(); |
408ea6ae | 993 | |
af8b7d2b JQ |
994 | if (multifd_send_initial_packet(p, &local_err) < 0) { |
995 | goto out; | |
996 | } | |
408ea6ae JQ |
997 | /* initial packet */ |
998 | p->num_packets = 1; | |
f986c3d2 JQ |
999 | |
1000 | while (true) { | |
d82628e4 | 1001 | qemu_sem_wait(&p->sem); |
f986c3d2 | 1002 | qemu_mutex_lock(&p->mutex); |
0beb5ed3 JQ |
1003 | |
1004 | if (p->pending_job) { | |
1005 | uint32_t used = p->pages->used; | |
1006 | uint64_t packet_num = p->packet_num; | |
1007 | uint32_t flags = p->flags; | |
1008 | ||
1009 | multifd_send_fill_packet(p); | |
1010 | p->flags = 0; | |
1011 | p->num_packets++; | |
1012 | p->num_pages += used; | |
1013 | p->pages->used = 0; | |
1014 | qemu_mutex_unlock(&p->mutex); | |
1015 | ||
1016 | trace_multifd_send(p->id, packet_num, used, flags); | |
1017 | ||
8b2db7f5 JQ |
1018 | ret = qio_channel_write_all(p->c, (void *)p->packet, |
1019 | p->packet_len, &local_err); | |
1020 | if (ret != 0) { | |
1021 | break; | |
1022 | } | |
1023 | ||
1024 | ret = qio_channel_writev_all(p->c, p->pages->iov, used, &local_err); | |
1025 | if (ret != 0) { | |
1026 | break; | |
1027 | } | |
0beb5ed3 JQ |
1028 | |
1029 | qemu_mutex_lock(&p->mutex); | |
1030 | p->pending_job--; | |
1031 | qemu_mutex_unlock(&p->mutex); | |
6df264ac JQ |
1032 | |
1033 | if (flags & MULTIFD_FLAG_SYNC) { | |
1034 | qemu_sem_post(&multifd_send_state->sem_sync); | |
1035 | } | |
b9ee2f7d | 1036 | qemu_sem_post(&multifd_send_state->channels_ready); |
0beb5ed3 | 1037 | } else if (p->quit) { |
f986c3d2 JQ |
1038 | qemu_mutex_unlock(&p->mutex); |
1039 | break; | |
6df264ac JQ |
1040 | } else { |
1041 | qemu_mutex_unlock(&p->mutex); | |
1042 | /* sometimes there are spurious wakeups */ | |
f986c3d2 | 1043 | } |
f986c3d2 JQ |
1044 | } |
1045 | ||
af8b7d2b JQ |
1046 | out: |
1047 | if (local_err) { | |
1048 | multifd_send_terminate_threads(local_err); | |
1049 | } | |
1050 | ||
66770707 JQ |
1051 | qemu_mutex_lock(&p->mutex); |
1052 | p->running = false; | |
1053 | qemu_mutex_unlock(&p->mutex); | |
1054 | ||
74637e6f | 1055 | rcu_unregister_thread(); |
408ea6ae JQ |
1056 | trace_multifd_send_thread_end(p->id, p->num_packets, p->num_pages); |
1057 | ||
f986c3d2 JQ |
1058 | return NULL; |
1059 | } | |
1060 | ||
60df2d4a JQ |
1061 | static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque) |
1062 | { | |
1063 | MultiFDSendParams *p = opaque; | |
1064 | QIOChannel *sioc = QIO_CHANNEL(qio_task_get_source(task)); | |
1065 | Error *local_err = NULL; | |
1066 | ||
1067 | if (qio_task_propagate_error(task, &local_err)) { | |
1068 | if (multifd_save_cleanup(&local_err) != 0) { | |
1069 | migrate_set_error(migrate_get_current(), local_err); | |
1070 | } | |
1071 | } else { | |
1072 | p->c = QIO_CHANNEL(sioc); | |
1073 | qio_channel_set_delay(p->c, false); | |
1074 | p->running = true; | |
1075 | qemu_thread_create(&p->thread, p->name, multifd_send_thread, p, | |
1076 | QEMU_THREAD_JOINABLE); | |
1077 | ||
1078 | atomic_inc(&multifd_send_state->count); | |
1079 | } | |
1080 | } | |
1081 | ||
f986c3d2 JQ |
1082 | int multifd_save_setup(void) |
1083 | { | |
1084 | int thread_count; | |
34c55a94 | 1085 | uint32_t page_count = migrate_multifd_page_count(); |
f986c3d2 JQ |
1086 | uint8_t i; |
1087 | ||
1088 | if (!migrate_use_multifd()) { | |
1089 | return 0; | |
1090 | } | |
1091 | thread_count = migrate_multifd_channels(); | |
1092 | multifd_send_state = g_malloc0(sizeof(*multifd_send_state)); | |
1093 | multifd_send_state->params = g_new0(MultiFDSendParams, thread_count); | |
66770707 | 1094 | atomic_set(&multifd_send_state->count, 0); |
34c55a94 | 1095 | multifd_send_state->pages = multifd_pages_init(page_count); |
6df264ac | 1096 | qemu_sem_init(&multifd_send_state->sem_sync, 0); |
b9ee2f7d | 1097 | qemu_sem_init(&multifd_send_state->channels_ready, 0); |
34c55a94 | 1098 | |
f986c3d2 JQ |
1099 | for (i = 0; i < thread_count; i++) { |
1100 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
1101 | ||
1102 | qemu_mutex_init(&p->mutex); | |
1103 | qemu_sem_init(&p->sem, 0); | |
6df264ac | 1104 | qemu_sem_init(&p->sem_sync, 0); |
f986c3d2 | 1105 | p->quit = false; |
0beb5ed3 | 1106 | p->pending_job = 0; |
f986c3d2 | 1107 | p->id = i; |
34c55a94 | 1108 | p->pages = multifd_pages_init(page_count); |
2a26c979 JQ |
1109 | p->packet_len = sizeof(MultiFDPacket_t) |
1110 | + sizeof(ram_addr_t) * page_count; | |
1111 | p->packet = g_malloc0(p->packet_len); | |
f986c3d2 | 1112 | p->name = g_strdup_printf("multifdsend_%d", i); |
60df2d4a | 1113 | socket_send_channel_create(multifd_new_send_channel_async, p); |
f986c3d2 JQ |
1114 | } |
1115 | return 0; | |
1116 | } | |
1117 | ||
f986c3d2 JQ |
1118 | struct { |
1119 | MultiFDRecvParams *params; | |
1120 | /* number of created threads */ | |
1121 | int count; | |
6df264ac JQ |
1122 | /* syncs main thread and channels */ |
1123 | QemuSemaphore sem_sync; | |
1124 | /* global number of generated multifd packets */ | |
1125 | uint64_t packet_num; | |
f986c3d2 JQ |
1126 | } *multifd_recv_state; |
1127 | ||
66770707 | 1128 | static void multifd_recv_terminate_threads(Error *err) |
f986c3d2 JQ |
1129 | { |
1130 | int i; | |
1131 | ||
7a169d74 JQ |
1132 | if (err) { |
1133 | MigrationState *s = migrate_get_current(); | |
1134 | migrate_set_error(s, err); | |
1135 | if (s->state == MIGRATION_STATUS_SETUP || | |
1136 | s->state == MIGRATION_STATUS_ACTIVE) { | |
1137 | migrate_set_state(&s->state, s->state, | |
1138 | MIGRATION_STATUS_FAILED); | |
1139 | } | |
1140 | } | |
1141 | ||
66770707 | 1142 | for (i = 0; i < migrate_multifd_channels(); i++) { |
f986c3d2 JQ |
1143 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; |
1144 | ||
1145 | qemu_mutex_lock(&p->mutex); | |
7a5cc33c JQ |
1146 | /* We could arrive here for two reasons: |
1147 | - normal quit, i.e. everything went fine, just finished | |
1148 | - error quit: We close the channels so the channel threads | |
1149 | finish the qio_channel_read_all_eof() */ | |
1150 | qio_channel_shutdown(p->c, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); | |
f986c3d2 JQ |
1151 | qemu_mutex_unlock(&p->mutex); |
1152 | } | |
1153 | } | |
1154 | ||
1155 | int multifd_load_cleanup(Error **errp) | |
1156 | { | |
1157 | int i; | |
1158 | int ret = 0; | |
1159 | ||
1160 | if (!migrate_use_multifd()) { | |
1161 | return 0; | |
1162 | } | |
66770707 JQ |
1163 | multifd_recv_terminate_threads(NULL); |
1164 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
f986c3d2 JQ |
1165 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; |
1166 | ||
66770707 JQ |
1167 | if (p->running) { |
1168 | qemu_thread_join(&p->thread); | |
1169 | } | |
60df2d4a JQ |
1170 | object_unref(OBJECT(p->c)); |
1171 | p->c = NULL; | |
f986c3d2 | 1172 | qemu_mutex_destroy(&p->mutex); |
6df264ac | 1173 | qemu_sem_destroy(&p->sem_sync); |
f986c3d2 JQ |
1174 | g_free(p->name); |
1175 | p->name = NULL; | |
34c55a94 JQ |
1176 | multifd_pages_clear(p->pages); |
1177 | p->pages = NULL; | |
2a26c979 JQ |
1178 | p->packet_len = 0; |
1179 | g_free(p->packet); | |
1180 | p->packet = NULL; | |
f986c3d2 | 1181 | } |
6df264ac | 1182 | qemu_sem_destroy(&multifd_recv_state->sem_sync); |
f986c3d2 JQ |
1183 | g_free(multifd_recv_state->params); |
1184 | multifd_recv_state->params = NULL; | |
1185 | g_free(multifd_recv_state); | |
1186 | multifd_recv_state = NULL; | |
1187 | ||
1188 | return ret; | |
1189 | } | |
1190 | ||
6df264ac JQ |
1191 | static void multifd_recv_sync_main(void) |
1192 | { | |
1193 | int i; | |
1194 | ||
1195 | if (!migrate_use_multifd()) { | |
1196 | return; | |
1197 | } | |
1198 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
1199 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
1200 | ||
6df264ac JQ |
1201 | trace_multifd_recv_sync_main_wait(p->id); |
1202 | qemu_sem_wait(&multifd_recv_state->sem_sync); | |
1203 | qemu_mutex_lock(&p->mutex); | |
1204 | if (multifd_recv_state->packet_num < p->packet_num) { | |
1205 | multifd_recv_state->packet_num = p->packet_num; | |
1206 | } | |
1207 | qemu_mutex_unlock(&p->mutex); | |
1208 | } | |
1209 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
1210 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
1211 | ||
1212 | trace_multifd_recv_sync_main_signal(p->id); | |
6df264ac JQ |
1213 | qemu_sem_post(&p->sem_sync); |
1214 | } | |
1215 | trace_multifd_recv_sync_main(multifd_recv_state->packet_num); | |
1216 | } | |
1217 | ||
f986c3d2 JQ |
1218 | static void *multifd_recv_thread(void *opaque) |
1219 | { | |
1220 | MultiFDRecvParams *p = opaque; | |
2a26c979 JQ |
1221 | Error *local_err = NULL; |
1222 | int ret; | |
f986c3d2 | 1223 | |
408ea6ae | 1224 | trace_multifd_recv_thread_start(p->id); |
74637e6f | 1225 | rcu_register_thread(); |
408ea6ae | 1226 | |
f986c3d2 | 1227 | while (true) { |
6df264ac JQ |
1228 | uint32_t used; |
1229 | uint32_t flags; | |
0beb5ed3 | 1230 | |
8b2db7f5 JQ |
1231 | ret = qio_channel_read_all_eof(p->c, (void *)p->packet, |
1232 | p->packet_len, &local_err); | |
1233 | if (ret == 0) { /* EOF */ | |
1234 | break; | |
1235 | } | |
1236 | if (ret == -1) { /* Error */ | |
1237 | break; | |
1238 | } | |
2a26c979 | 1239 | |
6df264ac JQ |
1240 | qemu_mutex_lock(&p->mutex); |
1241 | ret = multifd_recv_unfill_packet(p, &local_err); | |
1242 | if (ret) { | |
f986c3d2 JQ |
1243 | qemu_mutex_unlock(&p->mutex); |
1244 | break; | |
1245 | } | |
6df264ac JQ |
1246 | |
1247 | used = p->pages->used; | |
1248 | flags = p->flags; | |
1249 | trace_multifd_recv(p->id, p->packet_num, used, flags); | |
6df264ac JQ |
1250 | p->num_packets++; |
1251 | p->num_pages += used; | |
f986c3d2 | 1252 | qemu_mutex_unlock(&p->mutex); |
6df264ac | 1253 | |
8b2db7f5 JQ |
1254 | ret = qio_channel_readv_all(p->c, p->pages->iov, used, &local_err); |
1255 | if (ret != 0) { | |
1256 | break; | |
1257 | } | |
1258 | ||
6df264ac JQ |
1259 | if (flags & MULTIFD_FLAG_SYNC) { |
1260 | qemu_sem_post(&multifd_recv_state->sem_sync); | |
1261 | qemu_sem_wait(&p->sem_sync); | |
1262 | } | |
f986c3d2 JQ |
1263 | } |
1264 | ||
d82628e4 JQ |
1265 | if (local_err) { |
1266 | multifd_recv_terminate_threads(local_err); | |
1267 | } | |
66770707 JQ |
1268 | qemu_mutex_lock(&p->mutex); |
1269 | p->running = false; | |
1270 | qemu_mutex_unlock(&p->mutex); | |
1271 | ||
74637e6f | 1272 | rcu_unregister_thread(); |
408ea6ae JQ |
1273 | trace_multifd_recv_thread_end(p->id, p->num_packets, p->num_pages); |
1274 | ||
f986c3d2 JQ |
1275 | return NULL; |
1276 | } | |
1277 | ||
1278 | int multifd_load_setup(void) | |
1279 | { | |
1280 | int thread_count; | |
34c55a94 | 1281 | uint32_t page_count = migrate_multifd_page_count(); |
f986c3d2 JQ |
1282 | uint8_t i; |
1283 | ||
1284 | if (!migrate_use_multifd()) { | |
1285 | return 0; | |
1286 | } | |
1287 | thread_count = migrate_multifd_channels(); | |
1288 | multifd_recv_state = g_malloc0(sizeof(*multifd_recv_state)); | |
1289 | multifd_recv_state->params = g_new0(MultiFDRecvParams, thread_count); | |
66770707 | 1290 | atomic_set(&multifd_recv_state->count, 0); |
6df264ac | 1291 | qemu_sem_init(&multifd_recv_state->sem_sync, 0); |
34c55a94 | 1292 | |
f986c3d2 JQ |
1293 | for (i = 0; i < thread_count; i++) { |
1294 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
1295 | ||
1296 | qemu_mutex_init(&p->mutex); | |
6df264ac | 1297 | qemu_sem_init(&p->sem_sync, 0); |
f986c3d2 | 1298 | p->id = i; |
34c55a94 | 1299 | p->pages = multifd_pages_init(page_count); |
2a26c979 JQ |
1300 | p->packet_len = sizeof(MultiFDPacket_t) |
1301 | + sizeof(ram_addr_t) * page_count; | |
1302 | p->packet = g_malloc0(p->packet_len); | |
f986c3d2 | 1303 | p->name = g_strdup_printf("multifdrecv_%d", i); |
f986c3d2 JQ |
1304 | } |
1305 | return 0; | |
1306 | } | |
1307 | ||
62c1e0ca JQ |
1308 | bool multifd_recv_all_channels_created(void) |
1309 | { | |
1310 | int thread_count = migrate_multifd_channels(); | |
1311 | ||
1312 | if (!migrate_use_multifd()) { | |
1313 | return true; | |
1314 | } | |
1315 | ||
1316 | return thread_count == atomic_read(&multifd_recv_state->count); | |
1317 | } | |
1318 | ||
81e62053 PX |
1319 | /* Return true if multifd is ready for the migration, otherwise false */ |
1320 | bool multifd_recv_new_channel(QIOChannel *ioc) | |
71bb07db | 1321 | { |
60df2d4a | 1322 | MultiFDRecvParams *p; |
af8b7d2b JQ |
1323 | Error *local_err = NULL; |
1324 | int id; | |
60df2d4a | 1325 | |
af8b7d2b JQ |
1326 | id = multifd_recv_initial_packet(ioc, &local_err); |
1327 | if (id < 0) { | |
1328 | multifd_recv_terminate_threads(local_err); | |
81e62053 | 1329 | return false; |
af8b7d2b JQ |
1330 | } |
1331 | ||
1332 | p = &multifd_recv_state->params[id]; | |
1333 | if (p->c != NULL) { | |
1334 | error_setg(&local_err, "multifd: received id '%d' already setup'", | |
1335 | id); | |
1336 | multifd_recv_terminate_threads(local_err); | |
81e62053 | 1337 | return false; |
af8b7d2b | 1338 | } |
60df2d4a JQ |
1339 | p->c = ioc; |
1340 | object_ref(OBJECT(ioc)); | |
408ea6ae JQ |
1341 | /* initial packet */ |
1342 | p->num_packets = 1; | |
60df2d4a JQ |
1343 | |
1344 | p->running = true; | |
1345 | qemu_thread_create(&p->thread, p->name, multifd_recv_thread, p, | |
1346 | QEMU_THREAD_JOINABLE); | |
1347 | atomic_inc(&multifd_recv_state->count); | |
81e62053 | 1348 | return multifd_recv_state->count == migrate_multifd_channels(); |
71bb07db JQ |
1349 | } |
1350 | ||
56e93d26 | 1351 | /** |
3d0684b2 | 1352 | * save_page_header: write page header to wire |
56e93d26 JQ |
1353 | * |
1354 | * If this is the 1st block, it also writes the block identification | |
1355 | * | |
3d0684b2 | 1356 | * Returns the number of bytes written |
56e93d26 JQ |
1357 | * |
1358 | * @f: QEMUFile where to send the data | |
1359 | * @block: block that contains the page we want to send | |
1360 | * @offset: offset inside the block for the page | |
1361 | * in the lower bits, it contains flags | |
1362 | */ | |
2bf3aa85 JQ |
1363 | static size_t save_page_header(RAMState *rs, QEMUFile *f, RAMBlock *block, |
1364 | ram_addr_t offset) | |
56e93d26 | 1365 | { |
9f5f380b | 1366 | size_t size, len; |
56e93d26 | 1367 | |
24795694 JQ |
1368 | if (block == rs->last_sent_block) { |
1369 | offset |= RAM_SAVE_FLAG_CONTINUE; | |
1370 | } | |
2bf3aa85 | 1371 | qemu_put_be64(f, offset); |
56e93d26 JQ |
1372 | size = 8; |
1373 | ||
1374 | if (!(offset & RAM_SAVE_FLAG_CONTINUE)) { | |
9f5f380b | 1375 | len = strlen(block->idstr); |
2bf3aa85 JQ |
1376 | qemu_put_byte(f, len); |
1377 | qemu_put_buffer(f, (uint8_t *)block->idstr, len); | |
9f5f380b | 1378 | size += 1 + len; |
24795694 | 1379 | rs->last_sent_block = block; |
56e93d26 JQ |
1380 | } |
1381 | return size; | |
1382 | } | |
1383 | ||
3d0684b2 JQ |
1384 | /** |
1385 | * mig_throttle_guest_down: throotle down the guest | |
1386 | * | |
1387 | * Reduce amount of guest cpu execution to hopefully slow down memory | |
1388 | * writes. If guest dirty memory rate is reduced below the rate at | |
1389 | * which we can transfer pages to the destination then we should be | |
1390 | * able to complete migration. Some workloads dirty memory way too | |
1391 | * fast and will not effectively converge, even with auto-converge. | |
070afca2 JH |
1392 | */ |
1393 | static void mig_throttle_guest_down(void) | |
1394 | { | |
1395 | MigrationState *s = migrate_get_current(); | |
2594f56d DB |
1396 | uint64_t pct_initial = s->parameters.cpu_throttle_initial; |
1397 | uint64_t pct_icrement = s->parameters.cpu_throttle_increment; | |
4cbc9c7f | 1398 | int pct_max = s->parameters.max_cpu_throttle; |
070afca2 JH |
1399 | |
1400 | /* We have not started throttling yet. Let's start it. */ | |
1401 | if (!cpu_throttle_active()) { | |
1402 | cpu_throttle_set(pct_initial); | |
1403 | } else { | |
1404 | /* Throttling already on, just increase the rate */ | |
4cbc9c7f LQ |
1405 | cpu_throttle_set(MIN(cpu_throttle_get_percentage() + pct_icrement, |
1406 | pct_max)); | |
070afca2 JH |
1407 | } |
1408 | } | |
1409 | ||
3d0684b2 JQ |
1410 | /** |
1411 | * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache | |
1412 | * | |
6f37bb8b | 1413 | * @rs: current RAM state |
3d0684b2 JQ |
1414 | * @current_addr: address for the zero page |
1415 | * | |
1416 | * Update the xbzrle cache to reflect a page that's been sent as all 0. | |
56e93d26 JQ |
1417 | * The important thing is that a stale (not-yet-0'd) page be replaced |
1418 | * by the new data. | |
1419 | * As a bonus, if the page wasn't in the cache it gets added so that | |
3d0684b2 | 1420 | * when a small write is made into the 0'd page it gets XBZRLE sent. |
56e93d26 | 1421 | */ |
6f37bb8b | 1422 | static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr) |
56e93d26 | 1423 | { |
6f37bb8b | 1424 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { |
56e93d26 JQ |
1425 | return; |
1426 | } | |
1427 | ||
1428 | /* We don't care if this fails to allocate a new cache page | |
1429 | * as long as it updated an old one */ | |
c00e0928 | 1430 | cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page, |
9360447d | 1431 | ram_counters.dirty_sync_count); |
56e93d26 JQ |
1432 | } |
1433 | ||
1434 | #define ENCODING_FLAG_XBZRLE 0x1 | |
1435 | ||
1436 | /** | |
1437 | * save_xbzrle_page: compress and send current page | |
1438 | * | |
1439 | * Returns: 1 means that we wrote the page | |
1440 | * 0 means that page is identical to the one already sent | |
1441 | * -1 means that xbzrle would be longer than normal | |
1442 | * | |
5a987738 | 1443 | * @rs: current RAM state |
3d0684b2 JQ |
1444 | * @current_data: pointer to the address of the page contents |
1445 | * @current_addr: addr of the page | |
56e93d26 JQ |
1446 | * @block: block that contains the page we want to send |
1447 | * @offset: offset inside the block for the page | |
1448 | * @last_stage: if we are at the completion stage | |
56e93d26 | 1449 | */ |
204b88b8 | 1450 | static int save_xbzrle_page(RAMState *rs, uint8_t **current_data, |
56e93d26 | 1451 | ram_addr_t current_addr, RAMBlock *block, |
072c2511 | 1452 | ram_addr_t offset, bool last_stage) |
56e93d26 JQ |
1453 | { |
1454 | int encoded_len = 0, bytes_xbzrle; | |
1455 | uint8_t *prev_cached_page; | |
1456 | ||
9360447d JQ |
1457 | if (!cache_is_cached(XBZRLE.cache, current_addr, |
1458 | ram_counters.dirty_sync_count)) { | |
1459 | xbzrle_counters.cache_miss++; | |
56e93d26 JQ |
1460 | if (!last_stage) { |
1461 | if (cache_insert(XBZRLE.cache, current_addr, *current_data, | |
9360447d | 1462 | ram_counters.dirty_sync_count) == -1) { |
56e93d26 JQ |
1463 | return -1; |
1464 | } else { | |
1465 | /* update *current_data when the page has been | |
1466 | inserted into cache */ | |
1467 | *current_data = get_cached_data(XBZRLE.cache, current_addr); | |
1468 | } | |
1469 | } | |
1470 | return -1; | |
1471 | } | |
1472 | ||
1473 | prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); | |
1474 | ||
1475 | /* save current buffer into memory */ | |
1476 | memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); | |
1477 | ||
1478 | /* XBZRLE encoding (if there is no overflow) */ | |
1479 | encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, | |
1480 | TARGET_PAGE_SIZE, XBZRLE.encoded_buf, | |
1481 | TARGET_PAGE_SIZE); | |
1482 | if (encoded_len == 0) { | |
55c4446b | 1483 | trace_save_xbzrle_page_skipping(); |
56e93d26 JQ |
1484 | return 0; |
1485 | } else if (encoded_len == -1) { | |
55c4446b | 1486 | trace_save_xbzrle_page_overflow(); |
9360447d | 1487 | xbzrle_counters.overflow++; |
56e93d26 JQ |
1488 | /* update data in the cache */ |
1489 | if (!last_stage) { | |
1490 | memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE); | |
1491 | *current_data = prev_cached_page; | |
1492 | } | |
1493 | return -1; | |
1494 | } | |
1495 | ||
1496 | /* we need to update the data in the cache, in order to get the same data */ | |
1497 | if (!last_stage) { | |
1498 | memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); | |
1499 | } | |
1500 | ||
1501 | /* Send XBZRLE based compressed page */ | |
2bf3aa85 | 1502 | bytes_xbzrle = save_page_header(rs, rs->f, block, |
204b88b8 JQ |
1503 | offset | RAM_SAVE_FLAG_XBZRLE); |
1504 | qemu_put_byte(rs->f, ENCODING_FLAG_XBZRLE); | |
1505 | qemu_put_be16(rs->f, encoded_len); | |
1506 | qemu_put_buffer(rs->f, XBZRLE.encoded_buf, encoded_len); | |
56e93d26 | 1507 | bytes_xbzrle += encoded_len + 1 + 2; |
9360447d JQ |
1508 | xbzrle_counters.pages++; |
1509 | xbzrle_counters.bytes += bytes_xbzrle; | |
1510 | ram_counters.transferred += bytes_xbzrle; | |
56e93d26 JQ |
1511 | |
1512 | return 1; | |
1513 | } | |
1514 | ||
3d0684b2 JQ |
1515 | /** |
1516 | * migration_bitmap_find_dirty: find the next dirty page from start | |
f3f491fc | 1517 | * |
3d0684b2 JQ |
1518 | * Called with rcu_read_lock() to protect migration_bitmap |
1519 | * | |
1520 | * Returns the byte offset within memory region of the start of a dirty page | |
1521 | * | |
6f37bb8b | 1522 | * @rs: current RAM state |
3d0684b2 | 1523 | * @rb: RAMBlock where to search for dirty pages |
a935e30f | 1524 | * @start: page where we start the search |
f3f491fc | 1525 | */ |
56e93d26 | 1526 | static inline |
a935e30f | 1527 | unsigned long migration_bitmap_find_dirty(RAMState *rs, RAMBlock *rb, |
f20e2865 | 1528 | unsigned long start) |
56e93d26 | 1529 | { |
6b6712ef JQ |
1530 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; |
1531 | unsigned long *bitmap = rb->bmap; | |
56e93d26 JQ |
1532 | unsigned long next; |
1533 | ||
b895de50 CLG |
1534 | if (!qemu_ram_is_migratable(rb)) { |
1535 | return size; | |
1536 | } | |
1537 | ||
6b6712ef JQ |
1538 | if (rs->ram_bulk_stage && start > 0) { |
1539 | next = start + 1; | |
56e93d26 | 1540 | } else { |
6b6712ef | 1541 | next = find_next_bit(bitmap, size, start); |
56e93d26 JQ |
1542 | } |
1543 | ||
6b6712ef | 1544 | return next; |
56e93d26 JQ |
1545 | } |
1546 | ||
06b10688 | 1547 | static inline bool migration_bitmap_clear_dirty(RAMState *rs, |
f20e2865 JQ |
1548 | RAMBlock *rb, |
1549 | unsigned long page) | |
a82d593b DDAG |
1550 | { |
1551 | bool ret; | |
a82d593b | 1552 | |
6b6712ef | 1553 | ret = test_and_clear_bit(page, rb->bmap); |
a82d593b DDAG |
1554 | |
1555 | if (ret) { | |
0d8ec885 | 1556 | rs->migration_dirty_pages--; |
a82d593b DDAG |
1557 | } |
1558 | return ret; | |
1559 | } | |
1560 | ||
15440dd5 JQ |
1561 | static void migration_bitmap_sync_range(RAMState *rs, RAMBlock *rb, |
1562 | ram_addr_t start, ram_addr_t length) | |
56e93d26 | 1563 | { |
0d8ec885 | 1564 | rs->migration_dirty_pages += |
6b6712ef | 1565 | cpu_physical_memory_sync_dirty_bitmap(rb, start, length, |
0d8ec885 | 1566 | &rs->num_dirty_pages_period); |
56e93d26 JQ |
1567 | } |
1568 | ||
3d0684b2 JQ |
1569 | /** |
1570 | * ram_pagesize_summary: calculate all the pagesizes of a VM | |
1571 | * | |
1572 | * Returns a summary bitmap of the page sizes of all RAMBlocks | |
1573 | * | |
1574 | * For VMs with just normal pages this is equivalent to the host page | |
1575 | * size. If it's got some huge pages then it's the OR of all the | |
1576 | * different page sizes. | |
e8ca1db2 DDAG |
1577 | */ |
1578 | uint64_t ram_pagesize_summary(void) | |
1579 | { | |
1580 | RAMBlock *block; | |
1581 | uint64_t summary = 0; | |
1582 | ||
b895de50 | 1583 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
e8ca1db2 DDAG |
1584 | summary |= block->page_size; |
1585 | } | |
1586 | ||
1587 | return summary; | |
1588 | } | |
1589 | ||
b734035b XG |
1590 | static void migration_update_rates(RAMState *rs, int64_t end_time) |
1591 | { | |
1592 | uint64_t iter_count = rs->iterations - rs->iterations_prev; | |
1593 | ||
1594 | /* calculate period counters */ | |
1595 | ram_counters.dirty_pages_rate = rs->num_dirty_pages_period * 1000 | |
1596 | / (end_time - rs->time_last_bitmap_sync); | |
1597 | ||
1598 | if (!iter_count) { | |
1599 | return; | |
1600 | } | |
1601 | ||
1602 | if (migrate_use_xbzrle()) { | |
1603 | xbzrle_counters.cache_miss_rate = (double)(xbzrle_counters.cache_miss - | |
1604 | rs->xbzrle_cache_miss_prev) / iter_count; | |
1605 | rs->xbzrle_cache_miss_prev = xbzrle_counters.cache_miss; | |
1606 | } | |
1607 | } | |
1608 | ||
8d820d6f | 1609 | static void migration_bitmap_sync(RAMState *rs) |
56e93d26 JQ |
1610 | { |
1611 | RAMBlock *block; | |
56e93d26 | 1612 | int64_t end_time; |
c4bdf0cf | 1613 | uint64_t bytes_xfer_now; |
56e93d26 | 1614 | |
9360447d | 1615 | ram_counters.dirty_sync_count++; |
56e93d26 | 1616 | |
f664da80 JQ |
1617 | if (!rs->time_last_bitmap_sync) { |
1618 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56e93d26 JQ |
1619 | } |
1620 | ||
1621 | trace_migration_bitmap_sync_start(); | |
9c1f8f44 | 1622 | memory_global_dirty_log_sync(); |
56e93d26 | 1623 | |
108cfae0 | 1624 | qemu_mutex_lock(&rs->bitmap_mutex); |
56e93d26 | 1625 | rcu_read_lock(); |
b895de50 | 1626 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
15440dd5 | 1627 | migration_bitmap_sync_range(rs, block, 0, block->used_length); |
56e93d26 | 1628 | } |
650af890 | 1629 | ram_counters.remaining = ram_bytes_remaining(); |
56e93d26 | 1630 | rcu_read_unlock(); |
108cfae0 | 1631 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 1632 | |
a66cd90c | 1633 | trace_migration_bitmap_sync_end(rs->num_dirty_pages_period); |
1ffb5dfd | 1634 | |
56e93d26 JQ |
1635 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1636 | ||
1637 | /* more than 1 second = 1000 millisecons */ | |
f664da80 | 1638 | if (end_time > rs->time_last_bitmap_sync + 1000) { |
9360447d | 1639 | bytes_xfer_now = ram_counters.transferred; |
d693c6f1 | 1640 | |
9ac78b61 PL |
1641 | /* During block migration the auto-converge logic incorrectly detects |
1642 | * that ram migration makes no progress. Avoid this by disabling the | |
1643 | * throttling logic during the bulk phase of block migration. */ | |
1644 | if (migrate_auto_converge() && !blk_mig_bulk_active()) { | |
56e93d26 JQ |
1645 | /* The following detection logic can be refined later. For now: |
1646 | Check to see if the dirtied bytes is 50% more than the approx. | |
1647 | amount of bytes that just got transferred since the last time we | |
070afca2 JH |
1648 | were in this routine. If that happens twice, start or increase |
1649 | throttling */ | |
070afca2 | 1650 | |
d693c6f1 | 1651 | if ((rs->num_dirty_pages_period * TARGET_PAGE_SIZE > |
eac74159 | 1652 | (bytes_xfer_now - rs->bytes_xfer_prev) / 2) && |
b4a3c64b | 1653 | (++rs->dirty_rate_high_cnt >= 2)) { |
56e93d26 | 1654 | trace_migration_throttle(); |
8d820d6f | 1655 | rs->dirty_rate_high_cnt = 0; |
070afca2 | 1656 | mig_throttle_guest_down(); |
d693c6f1 | 1657 | } |
56e93d26 | 1658 | } |
070afca2 | 1659 | |
b734035b XG |
1660 | migration_update_rates(rs, end_time); |
1661 | ||
1662 | rs->iterations_prev = rs->iterations; | |
d693c6f1 FF |
1663 | |
1664 | /* reset period counters */ | |
f664da80 | 1665 | rs->time_last_bitmap_sync = end_time; |
a66cd90c | 1666 | rs->num_dirty_pages_period = 0; |
d2a4d85a | 1667 | rs->bytes_xfer_prev = bytes_xfer_now; |
56e93d26 | 1668 | } |
4addcd4f | 1669 | if (migrate_use_events()) { |
9360447d | 1670 | qapi_event_send_migration_pass(ram_counters.dirty_sync_count, NULL); |
4addcd4f | 1671 | } |
56e93d26 JQ |
1672 | } |
1673 | ||
6c97ec5f XG |
1674 | /** |
1675 | * save_zero_page_to_file: send the zero page to the file | |
1676 | * | |
1677 | * Returns the size of data written to the file, 0 means the page is not | |
1678 | * a zero page | |
1679 | * | |
1680 | * @rs: current RAM state | |
1681 | * @file: the file where the data is saved | |
1682 | * @block: block that contains the page we want to send | |
1683 | * @offset: offset inside the block for the page | |
1684 | */ | |
1685 | static int save_zero_page_to_file(RAMState *rs, QEMUFile *file, | |
1686 | RAMBlock *block, ram_addr_t offset) | |
1687 | { | |
1688 | uint8_t *p = block->host + offset; | |
1689 | int len = 0; | |
1690 | ||
1691 | if (is_zero_range(p, TARGET_PAGE_SIZE)) { | |
1692 | len += save_page_header(rs, file, block, offset | RAM_SAVE_FLAG_ZERO); | |
1693 | qemu_put_byte(file, 0); | |
1694 | len += 1; | |
1695 | } | |
1696 | return len; | |
1697 | } | |
1698 | ||
56e93d26 | 1699 | /** |
3d0684b2 | 1700 | * save_zero_page: send the zero page to the stream |
56e93d26 | 1701 | * |
3d0684b2 | 1702 | * Returns the number of pages written. |
56e93d26 | 1703 | * |
f7ccd61b | 1704 | * @rs: current RAM state |
56e93d26 JQ |
1705 | * @block: block that contains the page we want to send |
1706 | * @offset: offset inside the block for the page | |
56e93d26 | 1707 | */ |
7faccdc3 | 1708 | static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset) |
56e93d26 | 1709 | { |
6c97ec5f | 1710 | int len = save_zero_page_to_file(rs, rs->f, block, offset); |
56e93d26 | 1711 | |
6c97ec5f | 1712 | if (len) { |
9360447d | 1713 | ram_counters.duplicate++; |
6c97ec5f XG |
1714 | ram_counters.transferred += len; |
1715 | return 1; | |
56e93d26 | 1716 | } |
6c97ec5f | 1717 | return -1; |
56e93d26 JQ |
1718 | } |
1719 | ||
5727309d | 1720 | static void ram_release_pages(const char *rbname, uint64_t offset, int pages) |
53f09a10 | 1721 | { |
5727309d | 1722 | if (!migrate_release_ram() || !migration_in_postcopy()) { |
53f09a10 PB |
1723 | return; |
1724 | } | |
1725 | ||
aaa2064c | 1726 | ram_discard_range(rbname, offset, pages << TARGET_PAGE_BITS); |
53f09a10 PB |
1727 | } |
1728 | ||
059ff0fb XG |
1729 | /* |
1730 | * @pages: the number of pages written by the control path, | |
1731 | * < 0 - error | |
1732 | * > 0 - number of pages written | |
1733 | * | |
1734 | * Return true if the pages has been saved, otherwise false is returned. | |
1735 | */ | |
1736 | static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, | |
1737 | int *pages) | |
1738 | { | |
1739 | uint64_t bytes_xmit = 0; | |
1740 | int ret; | |
1741 | ||
1742 | *pages = -1; | |
1743 | ret = ram_control_save_page(rs->f, block->offset, offset, TARGET_PAGE_SIZE, | |
1744 | &bytes_xmit); | |
1745 | if (ret == RAM_SAVE_CONTROL_NOT_SUPP) { | |
1746 | return false; | |
1747 | } | |
1748 | ||
1749 | if (bytes_xmit) { | |
1750 | ram_counters.transferred += bytes_xmit; | |
1751 | *pages = 1; | |
1752 | } | |
1753 | ||
1754 | if (ret == RAM_SAVE_CONTROL_DELAYED) { | |
1755 | return true; | |
1756 | } | |
1757 | ||
1758 | if (bytes_xmit > 0) { | |
1759 | ram_counters.normal++; | |
1760 | } else if (bytes_xmit == 0) { | |
1761 | ram_counters.duplicate++; | |
1762 | } | |
1763 | ||
1764 | return true; | |
1765 | } | |
1766 | ||
65dacaa0 XG |
1767 | /* |
1768 | * directly send the page to the stream | |
1769 | * | |
1770 | * Returns the number of pages written. | |
1771 | * | |
1772 | * @rs: current RAM state | |
1773 | * @block: block that contains the page we want to send | |
1774 | * @offset: offset inside the block for the page | |
1775 | * @buf: the page to be sent | |
1776 | * @async: send to page asyncly | |
1777 | */ | |
1778 | static int save_normal_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, | |
1779 | uint8_t *buf, bool async) | |
1780 | { | |
1781 | ram_counters.transferred += save_page_header(rs, rs->f, block, | |
1782 | offset | RAM_SAVE_FLAG_PAGE); | |
1783 | if (async) { | |
1784 | qemu_put_buffer_async(rs->f, buf, TARGET_PAGE_SIZE, | |
1785 | migrate_release_ram() & | |
1786 | migration_in_postcopy()); | |
1787 | } else { | |
1788 | qemu_put_buffer(rs->f, buf, TARGET_PAGE_SIZE); | |
1789 | } | |
1790 | ram_counters.transferred += TARGET_PAGE_SIZE; | |
1791 | ram_counters.normal++; | |
1792 | return 1; | |
1793 | } | |
1794 | ||
56e93d26 | 1795 | /** |
3d0684b2 | 1796 | * ram_save_page: send the given page to the stream |
56e93d26 | 1797 | * |
3d0684b2 | 1798 | * Returns the number of pages written. |
3fd3c4b3 DDAG |
1799 | * < 0 - error |
1800 | * >=0 - Number of pages written - this might legally be 0 | |
1801 | * if xbzrle noticed the page was the same. | |
56e93d26 | 1802 | * |
6f37bb8b | 1803 | * @rs: current RAM state |
56e93d26 JQ |
1804 | * @block: block that contains the page we want to send |
1805 | * @offset: offset inside the block for the page | |
1806 | * @last_stage: if we are at the completion stage | |
56e93d26 | 1807 | */ |
a0a8aa14 | 1808 | static int ram_save_page(RAMState *rs, PageSearchStatus *pss, bool last_stage) |
56e93d26 JQ |
1809 | { |
1810 | int pages = -1; | |
56e93d26 | 1811 | uint8_t *p; |
56e93d26 | 1812 | bool send_async = true; |
a08f6890 | 1813 | RAMBlock *block = pss->block; |
a935e30f | 1814 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
059ff0fb | 1815 | ram_addr_t current_addr = block->offset + offset; |
56e93d26 | 1816 | |
2f68e399 | 1817 | p = block->host + offset; |
1db9d8e5 | 1818 | trace_ram_save_page(block->idstr, (uint64_t)offset, p); |
56e93d26 | 1819 | |
56e93d26 | 1820 | XBZRLE_cache_lock(); |
d7400a34 XG |
1821 | if (!rs->ram_bulk_stage && !migration_in_postcopy() && |
1822 | migrate_use_xbzrle()) { | |
059ff0fb XG |
1823 | pages = save_xbzrle_page(rs, &p, current_addr, block, |
1824 | offset, last_stage); | |
1825 | if (!last_stage) { | |
1826 | /* Can't send this cached data async, since the cache page | |
1827 | * might get updated before it gets to the wire | |
56e93d26 | 1828 | */ |
059ff0fb | 1829 | send_async = false; |
56e93d26 JQ |
1830 | } |
1831 | } | |
1832 | ||
1833 | /* XBZRLE overflow or normal page */ | |
1834 | if (pages == -1) { | |
65dacaa0 | 1835 | pages = save_normal_page(rs, block, offset, p, send_async); |
56e93d26 JQ |
1836 | } |
1837 | ||
1838 | XBZRLE_cache_unlock(); | |
1839 | ||
1840 | return pages; | |
1841 | } | |
1842 | ||
b9ee2f7d JQ |
1843 | static int ram_save_multifd_page(RAMState *rs, RAMBlock *block, |
1844 | ram_addr_t offset) | |
1845 | { | |
b9ee2f7d | 1846 | multifd_queue_page(block, offset); |
b9ee2f7d JQ |
1847 | ram_counters.normal++; |
1848 | ||
1849 | return 1; | |
1850 | } | |
1851 | ||
6ef3771c XG |
1852 | static void do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
1853 | ram_addr_t offset, uint8_t *source_buf) | |
56e93d26 | 1854 | { |
53518d94 | 1855 | RAMState *rs = ram_state; |
a7a9a88f | 1856 | uint8_t *p = block->host + (offset & TARGET_PAGE_MASK); |
6ef3771c | 1857 | int ret; |
56e93d26 | 1858 | |
6ef3771c | 1859 | save_page_header(rs, f, block, offset | RAM_SAVE_FLAG_COMPRESS_PAGE); |
34ab9e97 XG |
1860 | |
1861 | /* | |
1862 | * copy it to a internal buffer to avoid it being modified by VM | |
1863 | * so that we can catch up the error during compression and | |
1864 | * decompression | |
1865 | */ | |
1866 | memcpy(source_buf, p, TARGET_PAGE_SIZE); | |
6ef3771c XG |
1867 | ret = qemu_put_compression_data(f, stream, source_buf, TARGET_PAGE_SIZE); |
1868 | if (ret < 0) { | |
1869 | qemu_file_set_error(migrate_get_current()->to_dst_file, ret); | |
b3be2896 | 1870 | error_report("compressed data failed!"); |
6ef3771c | 1871 | return; |
b3be2896 | 1872 | } |
56e93d26 | 1873 | |
6ef3771c | 1874 | ram_release_pages(block->idstr, offset & TARGET_PAGE_MASK, 1); |
56e93d26 JQ |
1875 | } |
1876 | ||
ce25d337 | 1877 | static void flush_compressed_data(RAMState *rs) |
56e93d26 JQ |
1878 | { |
1879 | int idx, len, thread_count; | |
1880 | ||
1881 | if (!migrate_use_compression()) { | |
1882 | return; | |
1883 | } | |
1884 | thread_count = migrate_compress_threads(); | |
a7a9a88f | 1885 | |
0d9f9a5c | 1886 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 | 1887 | for (idx = 0; idx < thread_count; idx++) { |
a7a9a88f | 1888 | while (!comp_param[idx].done) { |
0d9f9a5c | 1889 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 | 1890 | } |
a7a9a88f | 1891 | } |
0d9f9a5c | 1892 | qemu_mutex_unlock(&comp_done_lock); |
a7a9a88f LL |
1893 | |
1894 | for (idx = 0; idx < thread_count; idx++) { | |
1895 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 1896 | if (!comp_param[idx].quit) { |
ce25d337 | 1897 | len = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
9360447d | 1898 | ram_counters.transferred += len; |
56e93d26 | 1899 | } |
a7a9a88f | 1900 | qemu_mutex_unlock(&comp_param[idx].mutex); |
56e93d26 JQ |
1901 | } |
1902 | } | |
1903 | ||
1904 | static inline void set_compress_params(CompressParam *param, RAMBlock *block, | |
1905 | ram_addr_t offset) | |
1906 | { | |
1907 | param->block = block; | |
1908 | param->offset = offset; | |
1909 | } | |
1910 | ||
ce25d337 JQ |
1911 | static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block, |
1912 | ram_addr_t offset) | |
56e93d26 JQ |
1913 | { |
1914 | int idx, thread_count, bytes_xmit = -1, pages = -1; | |
1d58872a | 1915 | bool wait = migrate_compress_wait_thread(); |
56e93d26 JQ |
1916 | |
1917 | thread_count = migrate_compress_threads(); | |
0d9f9a5c | 1918 | qemu_mutex_lock(&comp_done_lock); |
1d58872a XG |
1919 | retry: |
1920 | for (idx = 0; idx < thread_count; idx++) { | |
1921 | if (comp_param[idx].done) { | |
1922 | comp_param[idx].done = false; | |
1923 | bytes_xmit = qemu_put_qemu_file(rs->f, comp_param[idx].file); | |
1924 | qemu_mutex_lock(&comp_param[idx].mutex); | |
1925 | set_compress_params(&comp_param[idx], block, offset); | |
1926 | qemu_cond_signal(&comp_param[idx].cond); | |
1927 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
1928 | pages = 1; | |
1d58872a | 1929 | ram_counters.transferred += bytes_xmit; |
56e93d26 | 1930 | break; |
56e93d26 JQ |
1931 | } |
1932 | } | |
1d58872a XG |
1933 | |
1934 | /* | |
1935 | * wait for the free thread if the user specifies 'compress-wait-thread', | |
1936 | * otherwise we will post the page out in the main thread as normal page. | |
1937 | */ | |
1938 | if (pages < 0 && wait) { | |
1939 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); | |
1940 | goto retry; | |
1941 | } | |
0d9f9a5c | 1942 | qemu_mutex_unlock(&comp_done_lock); |
56e93d26 JQ |
1943 | |
1944 | return pages; | |
1945 | } | |
1946 | ||
3d0684b2 JQ |
1947 | /** |
1948 | * find_dirty_block: find the next dirty page and update any state | |
1949 | * associated with the search process. | |
b9e60928 | 1950 | * |
3d0684b2 | 1951 | * Returns if a page is found |
b9e60928 | 1952 | * |
6f37bb8b | 1953 | * @rs: current RAM state |
3d0684b2 JQ |
1954 | * @pss: data about the state of the current dirty page scan |
1955 | * @again: set to false if the search has scanned the whole of RAM | |
b9e60928 | 1956 | */ |
f20e2865 | 1957 | static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again) |
b9e60928 | 1958 | { |
f20e2865 | 1959 | pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page); |
6f37bb8b | 1960 | if (pss->complete_round && pss->block == rs->last_seen_block && |
a935e30f | 1961 | pss->page >= rs->last_page) { |
b9e60928 DDAG |
1962 | /* |
1963 | * We've been once around the RAM and haven't found anything. | |
1964 | * Give up. | |
1965 | */ | |
1966 | *again = false; | |
1967 | return false; | |
1968 | } | |
a935e30f | 1969 | if ((pss->page << TARGET_PAGE_BITS) >= pss->block->used_length) { |
b9e60928 | 1970 | /* Didn't find anything in this RAM Block */ |
a935e30f | 1971 | pss->page = 0; |
b9e60928 DDAG |
1972 | pss->block = QLIST_NEXT_RCU(pss->block, next); |
1973 | if (!pss->block) { | |
1974 | /* Hit the end of the list */ | |
1975 | pss->block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1976 | /* Flag that we've looped */ | |
1977 | pss->complete_round = true; | |
6f37bb8b | 1978 | rs->ram_bulk_stage = false; |
b9e60928 DDAG |
1979 | if (migrate_use_xbzrle()) { |
1980 | /* If xbzrle is on, stop using the data compression at this | |
1981 | * point. In theory, xbzrle can do better than compression. | |
1982 | */ | |
ce25d337 | 1983 | flush_compressed_data(rs); |
b9e60928 DDAG |
1984 | } |
1985 | } | |
1986 | /* Didn't find anything this time, but try again on the new block */ | |
1987 | *again = true; | |
1988 | return false; | |
1989 | } else { | |
1990 | /* Can go around again, but... */ | |
1991 | *again = true; | |
1992 | /* We've found something so probably don't need to */ | |
1993 | return true; | |
1994 | } | |
1995 | } | |
1996 | ||
3d0684b2 JQ |
1997 | /** |
1998 | * unqueue_page: gets a page of the queue | |
1999 | * | |
a82d593b | 2000 | * Helper for 'get_queued_page' - gets a page off the queue |
a82d593b | 2001 | * |
3d0684b2 JQ |
2002 | * Returns the block of the page (or NULL if none available) |
2003 | * | |
ec481c6c | 2004 | * @rs: current RAM state |
3d0684b2 | 2005 | * @offset: used to return the offset within the RAMBlock |
a82d593b | 2006 | */ |
f20e2865 | 2007 | static RAMBlock *unqueue_page(RAMState *rs, ram_addr_t *offset) |
a82d593b DDAG |
2008 | { |
2009 | RAMBlock *block = NULL; | |
2010 | ||
ec481c6c JQ |
2011 | qemu_mutex_lock(&rs->src_page_req_mutex); |
2012 | if (!QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
2013 | struct RAMSrcPageRequest *entry = | |
2014 | QSIMPLEQ_FIRST(&rs->src_page_requests); | |
a82d593b DDAG |
2015 | block = entry->rb; |
2016 | *offset = entry->offset; | |
a82d593b DDAG |
2017 | |
2018 | if (entry->len > TARGET_PAGE_SIZE) { | |
2019 | entry->len -= TARGET_PAGE_SIZE; | |
2020 | entry->offset += TARGET_PAGE_SIZE; | |
2021 | } else { | |
2022 | memory_region_unref(block->mr); | |
ec481c6c | 2023 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
a82d593b | 2024 | g_free(entry); |
e03a34f8 | 2025 | migration_consume_urgent_request(); |
a82d593b DDAG |
2026 | } |
2027 | } | |
ec481c6c | 2028 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
a82d593b DDAG |
2029 | |
2030 | return block; | |
2031 | } | |
2032 | ||
3d0684b2 JQ |
2033 | /** |
2034 | * get_queued_page: unqueue a page from the postocpy requests | |
2035 | * | |
2036 | * Skips pages that are already sent (!dirty) | |
a82d593b | 2037 | * |
3d0684b2 | 2038 | * Returns if a queued page is found |
a82d593b | 2039 | * |
6f37bb8b | 2040 | * @rs: current RAM state |
3d0684b2 | 2041 | * @pss: data about the state of the current dirty page scan |
a82d593b | 2042 | */ |
f20e2865 | 2043 | static bool get_queued_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b DDAG |
2044 | { |
2045 | RAMBlock *block; | |
2046 | ram_addr_t offset; | |
2047 | bool dirty; | |
2048 | ||
2049 | do { | |
f20e2865 | 2050 | block = unqueue_page(rs, &offset); |
a82d593b DDAG |
2051 | /* |
2052 | * We're sending this page, and since it's postcopy nothing else | |
2053 | * will dirty it, and we must make sure it doesn't get sent again | |
2054 | * even if this queue request was received after the background | |
2055 | * search already sent it. | |
2056 | */ | |
2057 | if (block) { | |
f20e2865 JQ |
2058 | unsigned long page; |
2059 | ||
6b6712ef JQ |
2060 | page = offset >> TARGET_PAGE_BITS; |
2061 | dirty = test_bit(page, block->bmap); | |
a82d593b | 2062 | if (!dirty) { |
06b10688 | 2063 | trace_get_queued_page_not_dirty(block->idstr, (uint64_t)offset, |
6b6712ef | 2064 | page, test_bit(page, block->unsentmap)); |
a82d593b | 2065 | } else { |
f20e2865 | 2066 | trace_get_queued_page(block->idstr, (uint64_t)offset, page); |
a82d593b DDAG |
2067 | } |
2068 | } | |
2069 | ||
2070 | } while (block && !dirty); | |
2071 | ||
2072 | if (block) { | |
2073 | /* | |
2074 | * As soon as we start servicing pages out of order, then we have | |
2075 | * to kill the bulk stage, since the bulk stage assumes | |
2076 | * in (migration_bitmap_find_and_reset_dirty) that every page is | |
2077 | * dirty, that's no longer true. | |
2078 | */ | |
6f37bb8b | 2079 | rs->ram_bulk_stage = false; |
a82d593b DDAG |
2080 | |
2081 | /* | |
2082 | * We want the background search to continue from the queued page | |
2083 | * since the guest is likely to want other pages near to the page | |
2084 | * it just requested. | |
2085 | */ | |
2086 | pss->block = block; | |
a935e30f | 2087 | pss->page = offset >> TARGET_PAGE_BITS; |
a82d593b DDAG |
2088 | } |
2089 | ||
2090 | return !!block; | |
2091 | } | |
2092 | ||
6c595cde | 2093 | /** |
5e58f968 JQ |
2094 | * migration_page_queue_free: drop any remaining pages in the ram |
2095 | * request queue | |
6c595cde | 2096 | * |
3d0684b2 JQ |
2097 | * It should be empty at the end anyway, but in error cases there may |
2098 | * be some left. in case that there is any page left, we drop it. | |
2099 | * | |
6c595cde | 2100 | */ |
83c13382 | 2101 | static void migration_page_queue_free(RAMState *rs) |
6c595cde | 2102 | { |
ec481c6c | 2103 | struct RAMSrcPageRequest *mspr, *next_mspr; |
6c595cde DDAG |
2104 | /* This queue generally should be empty - but in the case of a failed |
2105 | * migration might have some droppings in. | |
2106 | */ | |
2107 | rcu_read_lock(); | |
ec481c6c | 2108 | QSIMPLEQ_FOREACH_SAFE(mspr, &rs->src_page_requests, next_req, next_mspr) { |
6c595cde | 2109 | memory_region_unref(mspr->rb->mr); |
ec481c6c | 2110 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
6c595cde DDAG |
2111 | g_free(mspr); |
2112 | } | |
2113 | rcu_read_unlock(); | |
2114 | } | |
2115 | ||
2116 | /** | |
3d0684b2 JQ |
2117 | * ram_save_queue_pages: queue the page for transmission |
2118 | * | |
2119 | * A request from postcopy destination for example. | |
2120 | * | |
2121 | * Returns zero on success or negative on error | |
2122 | * | |
3d0684b2 JQ |
2123 | * @rbname: Name of the RAMBLock of the request. NULL means the |
2124 | * same that last one. | |
2125 | * @start: starting address from the start of the RAMBlock | |
2126 | * @len: length (in bytes) to send | |
6c595cde | 2127 | */ |
96506894 | 2128 | int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len) |
6c595cde DDAG |
2129 | { |
2130 | RAMBlock *ramblock; | |
53518d94 | 2131 | RAMState *rs = ram_state; |
6c595cde | 2132 | |
9360447d | 2133 | ram_counters.postcopy_requests++; |
6c595cde DDAG |
2134 | rcu_read_lock(); |
2135 | if (!rbname) { | |
2136 | /* Reuse last RAMBlock */ | |
68a098f3 | 2137 | ramblock = rs->last_req_rb; |
6c595cde DDAG |
2138 | |
2139 | if (!ramblock) { | |
2140 | /* | |
2141 | * Shouldn't happen, we can't reuse the last RAMBlock if | |
2142 | * it's the 1st request. | |
2143 | */ | |
2144 | error_report("ram_save_queue_pages no previous block"); | |
2145 | goto err; | |
2146 | } | |
2147 | } else { | |
2148 | ramblock = qemu_ram_block_by_name(rbname); | |
2149 | ||
2150 | if (!ramblock) { | |
2151 | /* We shouldn't be asked for a non-existent RAMBlock */ | |
2152 | error_report("ram_save_queue_pages no block '%s'", rbname); | |
2153 | goto err; | |
2154 | } | |
68a098f3 | 2155 | rs->last_req_rb = ramblock; |
6c595cde DDAG |
2156 | } |
2157 | trace_ram_save_queue_pages(ramblock->idstr, start, len); | |
2158 | if (start+len > ramblock->used_length) { | |
9458ad6b JQ |
2159 | error_report("%s request overrun start=" RAM_ADDR_FMT " len=" |
2160 | RAM_ADDR_FMT " blocklen=" RAM_ADDR_FMT, | |
6c595cde DDAG |
2161 | __func__, start, len, ramblock->used_length); |
2162 | goto err; | |
2163 | } | |
2164 | ||
ec481c6c JQ |
2165 | struct RAMSrcPageRequest *new_entry = |
2166 | g_malloc0(sizeof(struct RAMSrcPageRequest)); | |
6c595cde DDAG |
2167 | new_entry->rb = ramblock; |
2168 | new_entry->offset = start; | |
2169 | new_entry->len = len; | |
2170 | ||
2171 | memory_region_ref(ramblock->mr); | |
ec481c6c JQ |
2172 | qemu_mutex_lock(&rs->src_page_req_mutex); |
2173 | QSIMPLEQ_INSERT_TAIL(&rs->src_page_requests, new_entry, next_req); | |
e03a34f8 | 2174 | migration_make_urgent_request(); |
ec481c6c | 2175 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
6c595cde DDAG |
2176 | rcu_read_unlock(); |
2177 | ||
2178 | return 0; | |
2179 | ||
2180 | err: | |
2181 | rcu_read_unlock(); | |
2182 | return -1; | |
2183 | } | |
2184 | ||
d7400a34 XG |
2185 | static bool save_page_use_compression(RAMState *rs) |
2186 | { | |
2187 | if (!migrate_use_compression()) { | |
2188 | return false; | |
2189 | } | |
2190 | ||
2191 | /* | |
2192 | * If xbzrle is on, stop using the data compression after first | |
2193 | * round of migration even if compression is enabled. In theory, | |
2194 | * xbzrle can do better than compression. | |
2195 | */ | |
2196 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { | |
2197 | return true; | |
2198 | } | |
2199 | ||
2200 | return false; | |
2201 | } | |
2202 | ||
a82d593b | 2203 | /** |
3d0684b2 | 2204 | * ram_save_target_page: save one target page |
a82d593b | 2205 | * |
3d0684b2 | 2206 | * Returns the number of pages written |
a82d593b | 2207 | * |
6f37bb8b | 2208 | * @rs: current RAM state |
3d0684b2 | 2209 | * @pss: data about the page we want to send |
a82d593b | 2210 | * @last_stage: if we are at the completion stage |
a82d593b | 2211 | */ |
a0a8aa14 | 2212 | static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 2213 | bool last_stage) |
a82d593b | 2214 | { |
a8ec91f9 XG |
2215 | RAMBlock *block = pss->block; |
2216 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; | |
2217 | int res; | |
2218 | ||
2219 | if (control_save_page(rs, block, offset, &res)) { | |
2220 | return res; | |
2221 | } | |
2222 | ||
1faa5665 | 2223 | /* |
d7400a34 XG |
2224 | * When starting the process of a new block, the first page of |
2225 | * the block should be sent out before other pages in the same | |
2226 | * block, and all the pages in last block should have been sent | |
2227 | * out, keeping this order is important, because the 'cont' flag | |
2228 | * is used to avoid resending the block name. | |
1faa5665 | 2229 | */ |
d7400a34 XG |
2230 | if (block != rs->last_sent_block && save_page_use_compression(rs)) { |
2231 | flush_compressed_data(rs); | |
2232 | } | |
2233 | ||
2234 | res = save_zero_page(rs, block, offset); | |
2235 | if (res > 0) { | |
2236 | /* Must let xbzrle know, otherwise a previous (now 0'd) cached | |
2237 | * page would be stale | |
2238 | */ | |
2239 | if (!save_page_use_compression(rs)) { | |
2240 | XBZRLE_cache_lock(); | |
2241 | xbzrle_cache_zero_page(rs, block->offset + offset); | |
2242 | XBZRLE_cache_unlock(); | |
2243 | } | |
2244 | ram_release_pages(block->idstr, offset, res); | |
2245 | return res; | |
2246 | } | |
2247 | ||
da3f56cb XG |
2248 | /* |
2249 | * Make sure the first page is sent out before other pages. | |
2250 | * | |
2251 | * we post it as normal page as compression will take much | |
2252 | * CPU resource. | |
2253 | */ | |
2254 | if (block == rs->last_sent_block && save_page_use_compression(rs)) { | |
1d58872a XG |
2255 | res = compress_page_with_multi_thread(rs, block, offset); |
2256 | if (res > 0) { | |
2257 | return res; | |
2258 | } | |
b9ee2f7d JQ |
2259 | } else if (migrate_use_multifd()) { |
2260 | return ram_save_multifd_page(rs, block, offset); | |
a82d593b DDAG |
2261 | } |
2262 | ||
1faa5665 | 2263 | return ram_save_page(rs, pss, last_stage); |
a82d593b DDAG |
2264 | } |
2265 | ||
2266 | /** | |
3d0684b2 | 2267 | * ram_save_host_page: save a whole host page |
a82d593b | 2268 | * |
3d0684b2 JQ |
2269 | * Starting at *offset send pages up to the end of the current host |
2270 | * page. It's valid for the initial offset to point into the middle of | |
2271 | * a host page in which case the remainder of the hostpage is sent. | |
2272 | * Only dirty target pages are sent. Note that the host page size may | |
2273 | * be a huge page for this block. | |
1eb3fc0a DDAG |
2274 | * The saving stops at the boundary of the used_length of the block |
2275 | * if the RAMBlock isn't a multiple of the host page size. | |
a82d593b | 2276 | * |
3d0684b2 JQ |
2277 | * Returns the number of pages written or negative on error |
2278 | * | |
6f37bb8b | 2279 | * @rs: current RAM state |
3d0684b2 | 2280 | * @ms: current migration state |
3d0684b2 | 2281 | * @pss: data about the page we want to send |
a82d593b | 2282 | * @last_stage: if we are at the completion stage |
a82d593b | 2283 | */ |
a0a8aa14 | 2284 | static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 2285 | bool last_stage) |
a82d593b DDAG |
2286 | { |
2287 | int tmppages, pages = 0; | |
a935e30f JQ |
2288 | size_t pagesize_bits = |
2289 | qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
4c011c37 | 2290 | |
b895de50 CLG |
2291 | if (!qemu_ram_is_migratable(pss->block)) { |
2292 | error_report("block %s should not be migrated !", pss->block->idstr); | |
2293 | return 0; | |
2294 | } | |
2295 | ||
a82d593b | 2296 | do { |
1faa5665 XG |
2297 | /* Check the pages is dirty and if it is send it */ |
2298 | if (!migration_bitmap_clear_dirty(rs, pss->block, pss->page)) { | |
2299 | pss->page++; | |
2300 | continue; | |
2301 | } | |
2302 | ||
f20e2865 | 2303 | tmppages = ram_save_target_page(rs, pss, last_stage); |
a82d593b DDAG |
2304 | if (tmppages < 0) { |
2305 | return tmppages; | |
2306 | } | |
2307 | ||
2308 | pages += tmppages; | |
1faa5665 XG |
2309 | if (pss->block->unsentmap) { |
2310 | clear_bit(pss->page, pss->block->unsentmap); | |
2311 | } | |
2312 | ||
a935e30f | 2313 | pss->page++; |
1eb3fc0a DDAG |
2314 | } while ((pss->page & (pagesize_bits - 1)) && |
2315 | offset_in_ramblock(pss->block, pss->page << TARGET_PAGE_BITS)); | |
a82d593b DDAG |
2316 | |
2317 | /* The offset we leave with is the last one we looked at */ | |
a935e30f | 2318 | pss->page--; |
a82d593b DDAG |
2319 | return pages; |
2320 | } | |
6c595cde | 2321 | |
56e93d26 | 2322 | /** |
3d0684b2 | 2323 | * ram_find_and_save_block: finds a dirty page and sends it to f |
56e93d26 JQ |
2324 | * |
2325 | * Called within an RCU critical section. | |
2326 | * | |
3d0684b2 | 2327 | * Returns the number of pages written where zero means no dirty pages |
56e93d26 | 2328 | * |
6f37bb8b | 2329 | * @rs: current RAM state |
56e93d26 | 2330 | * @last_stage: if we are at the completion stage |
a82d593b DDAG |
2331 | * |
2332 | * On systems where host-page-size > target-page-size it will send all the | |
2333 | * pages in a host page that are dirty. | |
56e93d26 JQ |
2334 | */ |
2335 | ||
ce25d337 | 2336 | static int ram_find_and_save_block(RAMState *rs, bool last_stage) |
56e93d26 | 2337 | { |
b8fb8cb7 | 2338 | PageSearchStatus pss; |
56e93d26 | 2339 | int pages = 0; |
b9e60928 | 2340 | bool again, found; |
56e93d26 | 2341 | |
0827b9e9 AA |
2342 | /* No dirty page as there is zero RAM */ |
2343 | if (!ram_bytes_total()) { | |
2344 | return pages; | |
2345 | } | |
2346 | ||
6f37bb8b | 2347 | pss.block = rs->last_seen_block; |
a935e30f | 2348 | pss.page = rs->last_page; |
b8fb8cb7 DDAG |
2349 | pss.complete_round = false; |
2350 | ||
2351 | if (!pss.block) { | |
2352 | pss.block = QLIST_FIRST_RCU(&ram_list.blocks); | |
2353 | } | |
56e93d26 | 2354 | |
b9e60928 | 2355 | do { |
a82d593b | 2356 | again = true; |
f20e2865 | 2357 | found = get_queued_page(rs, &pss); |
b9e60928 | 2358 | |
a82d593b DDAG |
2359 | if (!found) { |
2360 | /* priority queue empty, so just search for something dirty */ | |
f20e2865 | 2361 | found = find_dirty_block(rs, &pss, &again); |
a82d593b | 2362 | } |
f3f491fc | 2363 | |
a82d593b | 2364 | if (found) { |
f20e2865 | 2365 | pages = ram_save_host_page(rs, &pss, last_stage); |
56e93d26 | 2366 | } |
b9e60928 | 2367 | } while (!pages && again); |
56e93d26 | 2368 | |
6f37bb8b | 2369 | rs->last_seen_block = pss.block; |
a935e30f | 2370 | rs->last_page = pss.page; |
56e93d26 JQ |
2371 | |
2372 | return pages; | |
2373 | } | |
2374 | ||
2375 | void acct_update_position(QEMUFile *f, size_t size, bool zero) | |
2376 | { | |
2377 | uint64_t pages = size / TARGET_PAGE_SIZE; | |
f7ccd61b | 2378 | |
56e93d26 | 2379 | if (zero) { |
9360447d | 2380 | ram_counters.duplicate += pages; |
56e93d26 | 2381 | } else { |
9360447d JQ |
2382 | ram_counters.normal += pages; |
2383 | ram_counters.transferred += size; | |
56e93d26 JQ |
2384 | qemu_update_position(f, size); |
2385 | } | |
2386 | } | |
2387 | ||
56e93d26 JQ |
2388 | uint64_t ram_bytes_total(void) |
2389 | { | |
2390 | RAMBlock *block; | |
2391 | uint64_t total = 0; | |
2392 | ||
2393 | rcu_read_lock(); | |
b895de50 | 2394 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
56e93d26 | 2395 | total += block->used_length; |
99e15582 | 2396 | } |
56e93d26 JQ |
2397 | rcu_read_unlock(); |
2398 | return total; | |
2399 | } | |
2400 | ||
f265e0e4 | 2401 | static void xbzrle_load_setup(void) |
56e93d26 | 2402 | { |
f265e0e4 | 2403 | XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE); |
56e93d26 JQ |
2404 | } |
2405 | ||
f265e0e4 JQ |
2406 | static void xbzrle_load_cleanup(void) |
2407 | { | |
2408 | g_free(XBZRLE.decoded_buf); | |
2409 | XBZRLE.decoded_buf = NULL; | |
2410 | } | |
2411 | ||
7d7c96be PX |
2412 | static void ram_state_cleanup(RAMState **rsp) |
2413 | { | |
b9ccaf6d DDAG |
2414 | if (*rsp) { |
2415 | migration_page_queue_free(*rsp); | |
2416 | qemu_mutex_destroy(&(*rsp)->bitmap_mutex); | |
2417 | qemu_mutex_destroy(&(*rsp)->src_page_req_mutex); | |
2418 | g_free(*rsp); | |
2419 | *rsp = NULL; | |
2420 | } | |
7d7c96be PX |
2421 | } |
2422 | ||
84593a08 PX |
2423 | static void xbzrle_cleanup(void) |
2424 | { | |
2425 | XBZRLE_cache_lock(); | |
2426 | if (XBZRLE.cache) { | |
2427 | cache_fini(XBZRLE.cache); | |
2428 | g_free(XBZRLE.encoded_buf); | |
2429 | g_free(XBZRLE.current_buf); | |
2430 | g_free(XBZRLE.zero_target_page); | |
2431 | XBZRLE.cache = NULL; | |
2432 | XBZRLE.encoded_buf = NULL; | |
2433 | XBZRLE.current_buf = NULL; | |
2434 | XBZRLE.zero_target_page = NULL; | |
2435 | } | |
2436 | XBZRLE_cache_unlock(); | |
2437 | } | |
2438 | ||
f265e0e4 | 2439 | static void ram_save_cleanup(void *opaque) |
56e93d26 | 2440 | { |
53518d94 | 2441 | RAMState **rsp = opaque; |
6b6712ef | 2442 | RAMBlock *block; |
eb859c53 | 2443 | |
2ff64038 LZ |
2444 | /* caller have hold iothread lock or is in a bh, so there is |
2445 | * no writing race against this migration_bitmap | |
2446 | */ | |
6b6712ef JQ |
2447 | memory_global_dirty_log_stop(); |
2448 | ||
b895de50 | 2449 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
6b6712ef JQ |
2450 | g_free(block->bmap); |
2451 | block->bmap = NULL; | |
2452 | g_free(block->unsentmap); | |
2453 | block->unsentmap = NULL; | |
56e93d26 JQ |
2454 | } |
2455 | ||
84593a08 | 2456 | xbzrle_cleanup(); |
f0afa331 | 2457 | compress_threads_save_cleanup(); |
7d7c96be | 2458 | ram_state_cleanup(rsp); |
56e93d26 JQ |
2459 | } |
2460 | ||
6f37bb8b | 2461 | static void ram_state_reset(RAMState *rs) |
56e93d26 | 2462 | { |
6f37bb8b JQ |
2463 | rs->last_seen_block = NULL; |
2464 | rs->last_sent_block = NULL; | |
269ace29 | 2465 | rs->last_page = 0; |
6f37bb8b JQ |
2466 | rs->last_version = ram_list.version; |
2467 | rs->ram_bulk_stage = true; | |
56e93d26 JQ |
2468 | } |
2469 | ||
2470 | #define MAX_WAIT 50 /* ms, half buffered_file limit */ | |
2471 | ||
4f2e4252 DDAG |
2472 | /* |
2473 | * 'expected' is the value you expect the bitmap mostly to be full | |
2474 | * of; it won't bother printing lines that are all this value. | |
2475 | * If 'todump' is null the migration bitmap is dumped. | |
2476 | */ | |
6b6712ef JQ |
2477 | void ram_debug_dump_bitmap(unsigned long *todump, bool expected, |
2478 | unsigned long pages) | |
4f2e4252 | 2479 | { |
4f2e4252 DDAG |
2480 | int64_t cur; |
2481 | int64_t linelen = 128; | |
2482 | char linebuf[129]; | |
2483 | ||
6b6712ef | 2484 | for (cur = 0; cur < pages; cur += linelen) { |
4f2e4252 DDAG |
2485 | int64_t curb; |
2486 | bool found = false; | |
2487 | /* | |
2488 | * Last line; catch the case where the line length | |
2489 | * is longer than remaining ram | |
2490 | */ | |
6b6712ef JQ |
2491 | if (cur + linelen > pages) { |
2492 | linelen = pages - cur; | |
4f2e4252 DDAG |
2493 | } |
2494 | for (curb = 0; curb < linelen; curb++) { | |
2495 | bool thisbit = test_bit(cur + curb, todump); | |
2496 | linebuf[curb] = thisbit ? '1' : '.'; | |
2497 | found = found || (thisbit != expected); | |
2498 | } | |
2499 | if (found) { | |
2500 | linebuf[curb] = '\0'; | |
2501 | fprintf(stderr, "0x%08" PRIx64 " : %s\n", cur, linebuf); | |
2502 | } | |
2503 | } | |
2504 | } | |
2505 | ||
e0b266f0 DDAG |
2506 | /* **** functions for postcopy ***** */ |
2507 | ||
ced1c616 PB |
2508 | void ram_postcopy_migrated_memory_release(MigrationState *ms) |
2509 | { | |
2510 | struct RAMBlock *block; | |
ced1c616 | 2511 | |
b895de50 | 2512 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
6b6712ef JQ |
2513 | unsigned long *bitmap = block->bmap; |
2514 | unsigned long range = block->used_length >> TARGET_PAGE_BITS; | |
2515 | unsigned long run_start = find_next_zero_bit(bitmap, range, 0); | |
ced1c616 PB |
2516 | |
2517 | while (run_start < range) { | |
2518 | unsigned long run_end = find_next_bit(bitmap, range, run_start + 1); | |
aaa2064c | 2519 | ram_discard_range(block->idstr, run_start << TARGET_PAGE_BITS, |
ced1c616 PB |
2520 | (run_end - run_start) << TARGET_PAGE_BITS); |
2521 | run_start = find_next_zero_bit(bitmap, range, run_end + 1); | |
2522 | } | |
2523 | } | |
2524 | } | |
2525 | ||
3d0684b2 JQ |
2526 | /** |
2527 | * postcopy_send_discard_bm_ram: discard a RAMBlock | |
2528 | * | |
2529 | * Returns zero on success | |
2530 | * | |
e0b266f0 DDAG |
2531 | * Callback from postcopy_each_ram_send_discard for each RAMBlock |
2532 | * Note: At this point the 'unsentmap' is the processed bitmap combined | |
2533 | * with the dirtymap; so a '1' means it's either dirty or unsent. | |
3d0684b2 JQ |
2534 | * |
2535 | * @ms: current migration state | |
2536 | * @pds: state for postcopy | |
2537 | * @start: RAMBlock starting page | |
2538 | * @length: RAMBlock size | |
e0b266f0 DDAG |
2539 | */ |
2540 | static int postcopy_send_discard_bm_ram(MigrationState *ms, | |
2541 | PostcopyDiscardState *pds, | |
6b6712ef | 2542 | RAMBlock *block) |
e0b266f0 | 2543 | { |
6b6712ef | 2544 | unsigned long end = block->used_length >> TARGET_PAGE_BITS; |
e0b266f0 | 2545 | unsigned long current; |
6b6712ef | 2546 | unsigned long *unsentmap = block->unsentmap; |
e0b266f0 | 2547 | |
6b6712ef | 2548 | for (current = 0; current < end; ) { |
e0b266f0 DDAG |
2549 | unsigned long one = find_next_bit(unsentmap, end, current); |
2550 | ||
2551 | if (one <= end) { | |
2552 | unsigned long zero = find_next_zero_bit(unsentmap, end, one + 1); | |
2553 | unsigned long discard_length; | |
2554 | ||
2555 | if (zero >= end) { | |
2556 | discard_length = end - one; | |
2557 | } else { | |
2558 | discard_length = zero - one; | |
2559 | } | |
d688c62d DDAG |
2560 | if (discard_length) { |
2561 | postcopy_discard_send_range(ms, pds, one, discard_length); | |
2562 | } | |
e0b266f0 DDAG |
2563 | current = one + discard_length; |
2564 | } else { | |
2565 | current = one; | |
2566 | } | |
2567 | } | |
2568 | ||
2569 | return 0; | |
2570 | } | |
2571 | ||
3d0684b2 JQ |
2572 | /** |
2573 | * postcopy_each_ram_send_discard: discard all RAMBlocks | |
2574 | * | |
2575 | * Returns 0 for success or negative for error | |
2576 | * | |
e0b266f0 DDAG |
2577 | * Utility for the outgoing postcopy code. |
2578 | * Calls postcopy_send_discard_bm_ram for each RAMBlock | |
2579 | * passing it bitmap indexes and name. | |
e0b266f0 DDAG |
2580 | * (qemu_ram_foreach_block ends up passing unscaled lengths |
2581 | * which would mean postcopy code would have to deal with target page) | |
3d0684b2 JQ |
2582 | * |
2583 | * @ms: current migration state | |
e0b266f0 DDAG |
2584 | */ |
2585 | static int postcopy_each_ram_send_discard(MigrationState *ms) | |
2586 | { | |
2587 | struct RAMBlock *block; | |
2588 | int ret; | |
2589 | ||
b895de50 | 2590 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
6b6712ef JQ |
2591 | PostcopyDiscardState *pds = |
2592 | postcopy_discard_send_init(ms, block->idstr); | |
e0b266f0 DDAG |
2593 | |
2594 | /* | |
2595 | * Postcopy sends chunks of bitmap over the wire, but it | |
2596 | * just needs indexes at this point, avoids it having | |
2597 | * target page specific code. | |
2598 | */ | |
6b6712ef | 2599 | ret = postcopy_send_discard_bm_ram(ms, pds, block); |
e0b266f0 DDAG |
2600 | postcopy_discard_send_finish(ms, pds); |
2601 | if (ret) { | |
2602 | return ret; | |
2603 | } | |
2604 | } | |
2605 | ||
2606 | return 0; | |
2607 | } | |
2608 | ||
3d0684b2 JQ |
2609 | /** |
2610 | * postcopy_chunk_hostpages_pass: canocalize bitmap in hostpages | |
2611 | * | |
2612 | * Helper for postcopy_chunk_hostpages; it's called twice to | |
2613 | * canonicalize the two bitmaps, that are similar, but one is | |
2614 | * inverted. | |
99e314eb | 2615 | * |
3d0684b2 JQ |
2616 | * Postcopy requires that all target pages in a hostpage are dirty or |
2617 | * clean, not a mix. This function canonicalizes the bitmaps. | |
99e314eb | 2618 | * |
3d0684b2 JQ |
2619 | * @ms: current migration state |
2620 | * @unsent_pass: if true we need to canonicalize partially unsent host pages | |
2621 | * otherwise we need to canonicalize partially dirty host pages | |
2622 | * @block: block that contains the page we want to canonicalize | |
2623 | * @pds: state for postcopy | |
99e314eb DDAG |
2624 | */ |
2625 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, bool unsent_pass, | |
2626 | RAMBlock *block, | |
2627 | PostcopyDiscardState *pds) | |
2628 | { | |
53518d94 | 2629 | RAMState *rs = ram_state; |
6b6712ef JQ |
2630 | unsigned long *bitmap = block->bmap; |
2631 | unsigned long *unsentmap = block->unsentmap; | |
29c59172 | 2632 | unsigned int host_ratio = block->page_size / TARGET_PAGE_SIZE; |
6b6712ef | 2633 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
99e314eb DDAG |
2634 | unsigned long run_start; |
2635 | ||
29c59172 DDAG |
2636 | if (block->page_size == TARGET_PAGE_SIZE) { |
2637 | /* Easy case - TPS==HPS for a non-huge page RAMBlock */ | |
2638 | return; | |
2639 | } | |
2640 | ||
99e314eb DDAG |
2641 | if (unsent_pass) { |
2642 | /* Find a sent page */ | |
6b6712ef | 2643 | run_start = find_next_zero_bit(unsentmap, pages, 0); |
99e314eb DDAG |
2644 | } else { |
2645 | /* Find a dirty page */ | |
6b6712ef | 2646 | run_start = find_next_bit(bitmap, pages, 0); |
99e314eb DDAG |
2647 | } |
2648 | ||
6b6712ef | 2649 | while (run_start < pages) { |
99e314eb DDAG |
2650 | bool do_fixup = false; |
2651 | unsigned long fixup_start_addr; | |
2652 | unsigned long host_offset; | |
2653 | ||
2654 | /* | |
2655 | * If the start of this run of pages is in the middle of a host | |
2656 | * page, then we need to fixup this host page. | |
2657 | */ | |
2658 | host_offset = run_start % host_ratio; | |
2659 | if (host_offset) { | |
2660 | do_fixup = true; | |
2661 | run_start -= host_offset; | |
2662 | fixup_start_addr = run_start; | |
2663 | /* For the next pass */ | |
2664 | run_start = run_start + host_ratio; | |
2665 | } else { | |
2666 | /* Find the end of this run */ | |
2667 | unsigned long run_end; | |
2668 | if (unsent_pass) { | |
6b6712ef | 2669 | run_end = find_next_bit(unsentmap, pages, run_start + 1); |
99e314eb | 2670 | } else { |
6b6712ef | 2671 | run_end = find_next_zero_bit(bitmap, pages, run_start + 1); |
99e314eb DDAG |
2672 | } |
2673 | /* | |
2674 | * If the end isn't at the start of a host page, then the | |
2675 | * run doesn't finish at the end of a host page | |
2676 | * and we need to discard. | |
2677 | */ | |
2678 | host_offset = run_end % host_ratio; | |
2679 | if (host_offset) { | |
2680 | do_fixup = true; | |
2681 | fixup_start_addr = run_end - host_offset; | |
2682 | /* | |
2683 | * This host page has gone, the next loop iteration starts | |
2684 | * from after the fixup | |
2685 | */ | |
2686 | run_start = fixup_start_addr + host_ratio; | |
2687 | } else { | |
2688 | /* | |
2689 | * No discards on this iteration, next loop starts from | |
2690 | * next sent/dirty page | |
2691 | */ | |
2692 | run_start = run_end + 1; | |
2693 | } | |
2694 | } | |
2695 | ||
2696 | if (do_fixup) { | |
2697 | unsigned long page; | |
2698 | ||
2699 | /* Tell the destination to discard this page */ | |
2700 | if (unsent_pass || !test_bit(fixup_start_addr, unsentmap)) { | |
2701 | /* For the unsent_pass we: | |
2702 | * discard partially sent pages | |
2703 | * For the !unsent_pass (dirty) we: | |
2704 | * discard partially dirty pages that were sent | |
2705 | * (any partially sent pages were already discarded | |
2706 | * by the previous unsent_pass) | |
2707 | */ | |
2708 | postcopy_discard_send_range(ms, pds, fixup_start_addr, | |
2709 | host_ratio); | |
2710 | } | |
2711 | ||
2712 | /* Clean up the bitmap */ | |
2713 | for (page = fixup_start_addr; | |
2714 | page < fixup_start_addr + host_ratio; page++) { | |
2715 | /* All pages in this host page are now not sent */ | |
2716 | set_bit(page, unsentmap); | |
2717 | ||
2718 | /* | |
2719 | * Remark them as dirty, updating the count for any pages | |
2720 | * that weren't previously dirty. | |
2721 | */ | |
0d8ec885 | 2722 | rs->migration_dirty_pages += !test_and_set_bit(page, bitmap); |
99e314eb DDAG |
2723 | } |
2724 | } | |
2725 | ||
2726 | if (unsent_pass) { | |
2727 | /* Find the next sent page for the next iteration */ | |
6b6712ef | 2728 | run_start = find_next_zero_bit(unsentmap, pages, run_start); |
99e314eb DDAG |
2729 | } else { |
2730 | /* Find the next dirty page for the next iteration */ | |
6b6712ef | 2731 | run_start = find_next_bit(bitmap, pages, run_start); |
99e314eb DDAG |
2732 | } |
2733 | } | |
2734 | } | |
2735 | ||
3d0684b2 JQ |
2736 | /** |
2737 | * postcopy_chuck_hostpages: discrad any partially sent host page | |
2738 | * | |
99e314eb DDAG |
2739 | * Utility for the outgoing postcopy code. |
2740 | * | |
2741 | * Discard any partially sent host-page size chunks, mark any partially | |
29c59172 DDAG |
2742 | * dirty host-page size chunks as all dirty. In this case the host-page |
2743 | * is the host-page for the particular RAMBlock, i.e. it might be a huge page | |
99e314eb | 2744 | * |
3d0684b2 JQ |
2745 | * Returns zero on success |
2746 | * | |
2747 | * @ms: current migration state | |
6b6712ef | 2748 | * @block: block we want to work with |
99e314eb | 2749 | */ |
6b6712ef | 2750 | static int postcopy_chunk_hostpages(MigrationState *ms, RAMBlock *block) |
99e314eb | 2751 | { |
6b6712ef JQ |
2752 | PostcopyDiscardState *pds = |
2753 | postcopy_discard_send_init(ms, block->idstr); | |
99e314eb | 2754 | |
6b6712ef JQ |
2755 | /* First pass: Discard all partially sent host pages */ |
2756 | postcopy_chunk_hostpages_pass(ms, true, block, pds); | |
2757 | /* | |
2758 | * Second pass: Ensure that all partially dirty host pages are made | |
2759 | * fully dirty. | |
2760 | */ | |
2761 | postcopy_chunk_hostpages_pass(ms, false, block, pds); | |
99e314eb | 2762 | |
6b6712ef | 2763 | postcopy_discard_send_finish(ms, pds); |
99e314eb DDAG |
2764 | return 0; |
2765 | } | |
2766 | ||
3d0684b2 JQ |
2767 | /** |
2768 | * ram_postcopy_send_discard_bitmap: transmit the discard bitmap | |
2769 | * | |
2770 | * Returns zero on success | |
2771 | * | |
e0b266f0 DDAG |
2772 | * Transmit the set of pages to be discarded after precopy to the target |
2773 | * these are pages that: | |
2774 | * a) Have been previously transmitted but are now dirty again | |
2775 | * b) Pages that have never been transmitted, this ensures that | |
2776 | * any pages on the destination that have been mapped by background | |
2777 | * tasks get discarded (transparent huge pages is the specific concern) | |
2778 | * Hopefully this is pretty sparse | |
3d0684b2 JQ |
2779 | * |
2780 | * @ms: current migration state | |
e0b266f0 DDAG |
2781 | */ |
2782 | int ram_postcopy_send_discard_bitmap(MigrationState *ms) | |
2783 | { | |
53518d94 | 2784 | RAMState *rs = ram_state; |
6b6712ef | 2785 | RAMBlock *block; |
e0b266f0 | 2786 | int ret; |
e0b266f0 DDAG |
2787 | |
2788 | rcu_read_lock(); | |
2789 | ||
2790 | /* This should be our last sync, the src is now paused */ | |
eb859c53 | 2791 | migration_bitmap_sync(rs); |
e0b266f0 | 2792 | |
6b6712ef JQ |
2793 | /* Easiest way to make sure we don't resume in the middle of a host-page */ |
2794 | rs->last_seen_block = NULL; | |
2795 | rs->last_sent_block = NULL; | |
2796 | rs->last_page = 0; | |
e0b266f0 | 2797 | |
b895de50 | 2798 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
6b6712ef JQ |
2799 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
2800 | unsigned long *bitmap = block->bmap; | |
2801 | unsigned long *unsentmap = block->unsentmap; | |
2802 | ||
2803 | if (!unsentmap) { | |
2804 | /* We don't have a safe way to resize the sentmap, so | |
2805 | * if the bitmap was resized it will be NULL at this | |
2806 | * point. | |
2807 | */ | |
2808 | error_report("migration ram resized during precopy phase"); | |
2809 | rcu_read_unlock(); | |
2810 | return -EINVAL; | |
2811 | } | |
2812 | /* Deal with TPS != HPS and huge pages */ | |
2813 | ret = postcopy_chunk_hostpages(ms, block); | |
2814 | if (ret) { | |
2815 | rcu_read_unlock(); | |
2816 | return ret; | |
2817 | } | |
e0b266f0 | 2818 | |
6b6712ef JQ |
2819 | /* |
2820 | * Update the unsentmap to be unsentmap = unsentmap | dirty | |
2821 | */ | |
2822 | bitmap_or(unsentmap, unsentmap, bitmap, pages); | |
e0b266f0 | 2823 | #ifdef DEBUG_POSTCOPY |
6b6712ef | 2824 | ram_debug_dump_bitmap(unsentmap, true, pages); |
e0b266f0 | 2825 | #endif |
6b6712ef JQ |
2826 | } |
2827 | trace_ram_postcopy_send_discard_bitmap(); | |
e0b266f0 DDAG |
2828 | |
2829 | ret = postcopy_each_ram_send_discard(ms); | |
2830 | rcu_read_unlock(); | |
2831 | ||
2832 | return ret; | |
2833 | } | |
2834 | ||
3d0684b2 JQ |
2835 | /** |
2836 | * ram_discard_range: discard dirtied pages at the beginning of postcopy | |
e0b266f0 | 2837 | * |
3d0684b2 | 2838 | * Returns zero on success |
e0b266f0 | 2839 | * |
36449157 JQ |
2840 | * @rbname: name of the RAMBlock of the request. NULL means the |
2841 | * same that last one. | |
3d0684b2 JQ |
2842 | * @start: RAMBlock starting page |
2843 | * @length: RAMBlock size | |
e0b266f0 | 2844 | */ |
aaa2064c | 2845 | int ram_discard_range(const char *rbname, uint64_t start, size_t length) |
e0b266f0 DDAG |
2846 | { |
2847 | int ret = -1; | |
2848 | ||
36449157 | 2849 | trace_ram_discard_range(rbname, start, length); |
d3a5038c | 2850 | |
e0b266f0 | 2851 | rcu_read_lock(); |
36449157 | 2852 | RAMBlock *rb = qemu_ram_block_by_name(rbname); |
e0b266f0 DDAG |
2853 | |
2854 | if (!rb) { | |
36449157 | 2855 | error_report("ram_discard_range: Failed to find block '%s'", rbname); |
e0b266f0 DDAG |
2856 | goto err; |
2857 | } | |
2858 | ||
814bb08f PX |
2859 | /* |
2860 | * On source VM, we don't need to update the received bitmap since | |
2861 | * we don't even have one. | |
2862 | */ | |
2863 | if (rb->receivedmap) { | |
2864 | bitmap_clear(rb->receivedmap, start >> qemu_target_page_bits(), | |
2865 | length >> qemu_target_page_bits()); | |
2866 | } | |
2867 | ||
d3a5038c | 2868 | ret = ram_block_discard_range(rb, start, length); |
e0b266f0 DDAG |
2869 | |
2870 | err: | |
2871 | rcu_read_unlock(); | |
2872 | ||
2873 | return ret; | |
2874 | } | |
2875 | ||
84593a08 PX |
2876 | /* |
2877 | * For every allocation, we will try not to crash the VM if the | |
2878 | * allocation failed. | |
2879 | */ | |
2880 | static int xbzrle_init(void) | |
2881 | { | |
2882 | Error *local_err = NULL; | |
2883 | ||
2884 | if (!migrate_use_xbzrle()) { | |
2885 | return 0; | |
2886 | } | |
2887 | ||
2888 | XBZRLE_cache_lock(); | |
2889 | ||
2890 | XBZRLE.zero_target_page = g_try_malloc0(TARGET_PAGE_SIZE); | |
2891 | if (!XBZRLE.zero_target_page) { | |
2892 | error_report("%s: Error allocating zero page", __func__); | |
2893 | goto err_out; | |
2894 | } | |
2895 | ||
2896 | XBZRLE.cache = cache_init(migrate_xbzrle_cache_size(), | |
2897 | TARGET_PAGE_SIZE, &local_err); | |
2898 | if (!XBZRLE.cache) { | |
2899 | error_report_err(local_err); | |
2900 | goto free_zero_page; | |
2901 | } | |
2902 | ||
2903 | XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); | |
2904 | if (!XBZRLE.encoded_buf) { | |
2905 | error_report("%s: Error allocating encoded_buf", __func__); | |
2906 | goto free_cache; | |
2907 | } | |
2908 | ||
2909 | XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); | |
2910 | if (!XBZRLE.current_buf) { | |
2911 | error_report("%s: Error allocating current_buf", __func__); | |
2912 | goto free_encoded_buf; | |
2913 | } | |
2914 | ||
2915 | /* We are all good */ | |
2916 | XBZRLE_cache_unlock(); | |
2917 | return 0; | |
2918 | ||
2919 | free_encoded_buf: | |
2920 | g_free(XBZRLE.encoded_buf); | |
2921 | XBZRLE.encoded_buf = NULL; | |
2922 | free_cache: | |
2923 | cache_fini(XBZRLE.cache); | |
2924 | XBZRLE.cache = NULL; | |
2925 | free_zero_page: | |
2926 | g_free(XBZRLE.zero_target_page); | |
2927 | XBZRLE.zero_target_page = NULL; | |
2928 | err_out: | |
2929 | XBZRLE_cache_unlock(); | |
2930 | return -ENOMEM; | |
2931 | } | |
2932 | ||
53518d94 | 2933 | static int ram_state_init(RAMState **rsp) |
56e93d26 | 2934 | { |
7d00ee6a PX |
2935 | *rsp = g_try_new0(RAMState, 1); |
2936 | ||
2937 | if (!*rsp) { | |
2938 | error_report("%s: Init ramstate fail", __func__); | |
2939 | return -1; | |
2940 | } | |
53518d94 JQ |
2941 | |
2942 | qemu_mutex_init(&(*rsp)->bitmap_mutex); | |
2943 | qemu_mutex_init(&(*rsp)->src_page_req_mutex); | |
2944 | QSIMPLEQ_INIT(&(*rsp)->src_page_requests); | |
56e93d26 | 2945 | |
7d00ee6a PX |
2946 | /* |
2947 | * Count the total number of pages used by ram blocks not including any | |
2948 | * gaps due to alignment or unplugs. | |
2949 | */ | |
2950 | (*rsp)->migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS; | |
2951 | ||
2952 | ram_state_reset(*rsp); | |
2953 | ||
2954 | return 0; | |
2955 | } | |
2956 | ||
d6eff5d7 | 2957 | static void ram_list_init_bitmaps(void) |
7d00ee6a | 2958 | { |
d6eff5d7 PX |
2959 | RAMBlock *block; |
2960 | unsigned long pages; | |
56e93d26 | 2961 | |
0827b9e9 AA |
2962 | /* Skip setting bitmap if there is no RAM */ |
2963 | if (ram_bytes_total()) { | |
b895de50 | 2964 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
d6eff5d7 | 2965 | pages = block->max_length >> TARGET_PAGE_BITS; |
6b6712ef JQ |
2966 | block->bmap = bitmap_new(pages); |
2967 | bitmap_set(block->bmap, 0, pages); | |
2968 | if (migrate_postcopy_ram()) { | |
2969 | block->unsentmap = bitmap_new(pages); | |
2970 | bitmap_set(block->unsentmap, 0, pages); | |
2971 | } | |
0827b9e9 | 2972 | } |
f3f491fc | 2973 | } |
d6eff5d7 PX |
2974 | } |
2975 | ||
2976 | static void ram_init_bitmaps(RAMState *rs) | |
2977 | { | |
2978 | /* For memory_global_dirty_log_start below. */ | |
2979 | qemu_mutex_lock_iothread(); | |
2980 | qemu_mutex_lock_ramlist(); | |
2981 | rcu_read_lock(); | |
f3f491fc | 2982 | |
d6eff5d7 | 2983 | ram_list_init_bitmaps(); |
56e93d26 | 2984 | memory_global_dirty_log_start(); |
d6eff5d7 PX |
2985 | migration_bitmap_sync(rs); |
2986 | ||
2987 | rcu_read_unlock(); | |
56e93d26 | 2988 | qemu_mutex_unlock_ramlist(); |
49877834 | 2989 | qemu_mutex_unlock_iothread(); |
d6eff5d7 PX |
2990 | } |
2991 | ||
2992 | static int ram_init_all(RAMState **rsp) | |
2993 | { | |
2994 | if (ram_state_init(rsp)) { | |
2995 | return -1; | |
2996 | } | |
2997 | ||
2998 | if (xbzrle_init()) { | |
2999 | ram_state_cleanup(rsp); | |
3000 | return -1; | |
3001 | } | |
3002 | ||
3003 | ram_init_bitmaps(*rsp); | |
a91246c9 HZ |
3004 | |
3005 | return 0; | |
3006 | } | |
3007 | ||
08614f34 PX |
3008 | static void ram_state_resume_prepare(RAMState *rs, QEMUFile *out) |
3009 | { | |
3010 | RAMBlock *block; | |
3011 | uint64_t pages = 0; | |
3012 | ||
3013 | /* | |
3014 | * Postcopy is not using xbzrle/compression, so no need for that. | |
3015 | * Also, since source are already halted, we don't need to care | |
3016 | * about dirty page logging as well. | |
3017 | */ | |
3018 | ||
ff0769a4 | 3019 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
08614f34 PX |
3020 | pages += bitmap_count_one(block->bmap, |
3021 | block->used_length >> TARGET_PAGE_BITS); | |
3022 | } | |
3023 | ||
3024 | /* This may not be aligned with current bitmaps. Recalculate. */ | |
3025 | rs->migration_dirty_pages = pages; | |
3026 | ||
3027 | rs->last_seen_block = NULL; | |
3028 | rs->last_sent_block = NULL; | |
3029 | rs->last_page = 0; | |
3030 | rs->last_version = ram_list.version; | |
3031 | /* | |
3032 | * Disable the bulk stage, otherwise we'll resend the whole RAM no | |
3033 | * matter what we have sent. | |
3034 | */ | |
3035 | rs->ram_bulk_stage = false; | |
3036 | ||
3037 | /* Update RAMState cache of output QEMUFile */ | |
3038 | rs->f = out; | |
3039 | ||
3040 | trace_ram_state_resume_prepare(pages); | |
3041 | } | |
3042 | ||
3d0684b2 JQ |
3043 | /* |
3044 | * Each of ram_save_setup, ram_save_iterate and ram_save_complete has | |
a91246c9 HZ |
3045 | * long-running RCU critical section. When rcu-reclaims in the code |
3046 | * start to become numerous it will be necessary to reduce the | |
3047 | * granularity of these critical sections. | |
3048 | */ | |
3049 | ||
3d0684b2 JQ |
3050 | /** |
3051 | * ram_save_setup: Setup RAM for migration | |
3052 | * | |
3053 | * Returns zero to indicate success and negative for error | |
3054 | * | |
3055 | * @f: QEMUFile where to send the data | |
3056 | * @opaque: RAMState pointer | |
3057 | */ | |
a91246c9 HZ |
3058 | static int ram_save_setup(QEMUFile *f, void *opaque) |
3059 | { | |
53518d94 | 3060 | RAMState **rsp = opaque; |
a91246c9 HZ |
3061 | RAMBlock *block; |
3062 | ||
dcaf446e XG |
3063 | if (compress_threads_save_setup()) { |
3064 | return -1; | |
3065 | } | |
3066 | ||
a91246c9 HZ |
3067 | /* migration has already setup the bitmap, reuse it. */ |
3068 | if (!migration_in_colo_state()) { | |
7d00ee6a | 3069 | if (ram_init_all(rsp) != 0) { |
dcaf446e | 3070 | compress_threads_save_cleanup(); |
a91246c9 | 3071 | return -1; |
53518d94 | 3072 | } |
a91246c9 | 3073 | } |
53518d94 | 3074 | (*rsp)->f = f; |
a91246c9 HZ |
3075 | |
3076 | rcu_read_lock(); | |
56e93d26 JQ |
3077 | |
3078 | qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE); | |
3079 | ||
b895de50 | 3080 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
56e93d26 JQ |
3081 | qemu_put_byte(f, strlen(block->idstr)); |
3082 | qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); | |
3083 | qemu_put_be64(f, block->used_length); | |
ef08fb38 DDAG |
3084 | if (migrate_postcopy_ram() && block->page_size != qemu_host_page_size) { |
3085 | qemu_put_be64(f, block->page_size); | |
3086 | } | |
56e93d26 JQ |
3087 | } |
3088 | ||
3089 | rcu_read_unlock(); | |
3090 | ||
3091 | ram_control_before_iterate(f, RAM_CONTROL_SETUP); | |
3092 | ram_control_after_iterate(f, RAM_CONTROL_SETUP); | |
3093 | ||
6df264ac | 3094 | multifd_send_sync_main(); |
56e93d26 | 3095 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
35374cbd | 3096 | qemu_fflush(f); |
56e93d26 JQ |
3097 | |
3098 | return 0; | |
3099 | } | |
3100 | ||
3d0684b2 JQ |
3101 | /** |
3102 | * ram_save_iterate: iterative stage for migration | |
3103 | * | |
3104 | * Returns zero to indicate success and negative for error | |
3105 | * | |
3106 | * @f: QEMUFile where to send the data | |
3107 | * @opaque: RAMState pointer | |
3108 | */ | |
56e93d26 JQ |
3109 | static int ram_save_iterate(QEMUFile *f, void *opaque) |
3110 | { | |
53518d94 JQ |
3111 | RAMState **temp = opaque; |
3112 | RAMState *rs = *temp; | |
56e93d26 JQ |
3113 | int ret; |
3114 | int i; | |
3115 | int64_t t0; | |
5c90308f | 3116 | int done = 0; |
56e93d26 | 3117 | |
b2557345 PL |
3118 | if (blk_mig_bulk_active()) { |
3119 | /* Avoid transferring ram during bulk phase of block migration as | |
3120 | * the bulk phase will usually take a long time and transferring | |
3121 | * ram updates during that time is pointless. */ | |
3122 | goto out; | |
3123 | } | |
3124 | ||
56e93d26 | 3125 | rcu_read_lock(); |
6f37bb8b JQ |
3126 | if (ram_list.version != rs->last_version) { |
3127 | ram_state_reset(rs); | |
56e93d26 JQ |
3128 | } |
3129 | ||
3130 | /* Read version before ram_list.blocks */ | |
3131 | smp_rmb(); | |
3132 | ||
3133 | ram_control_before_iterate(f, RAM_CONTROL_ROUND); | |
3134 | ||
3135 | t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
3136 | i = 0; | |
e03a34f8 DDAG |
3137 | while ((ret = qemu_file_rate_limit(f)) == 0 || |
3138 | !QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
56e93d26 JQ |
3139 | int pages; |
3140 | ||
e03a34f8 DDAG |
3141 | if (qemu_file_get_error(f)) { |
3142 | break; | |
3143 | } | |
3144 | ||
ce25d337 | 3145 | pages = ram_find_and_save_block(rs, false); |
56e93d26 JQ |
3146 | /* no more pages to sent */ |
3147 | if (pages == 0) { | |
5c90308f | 3148 | done = 1; |
56e93d26 JQ |
3149 | break; |
3150 | } | |
23b28c3c | 3151 | rs->iterations++; |
070afca2 | 3152 | |
56e93d26 JQ |
3153 | /* we want to check in the 1st loop, just in case it was the 1st time |
3154 | and we had to sync the dirty bitmap. | |
3155 | qemu_get_clock_ns() is a bit expensive, so we only check each some | |
3156 | iterations | |
3157 | */ | |
3158 | if ((i & 63) == 0) { | |
3159 | uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000; | |
3160 | if (t1 > MAX_WAIT) { | |
55c4446b | 3161 | trace_ram_save_iterate_big_wait(t1, i); |
56e93d26 JQ |
3162 | break; |
3163 | } | |
3164 | } | |
3165 | i++; | |
3166 | } | |
ce25d337 | 3167 | flush_compressed_data(rs); |
56e93d26 JQ |
3168 | rcu_read_unlock(); |
3169 | ||
3170 | /* | |
3171 | * Must occur before EOS (or any QEMUFile operation) | |
3172 | * because of RDMA protocol. | |
3173 | */ | |
3174 | ram_control_after_iterate(f, RAM_CONTROL_ROUND); | |
3175 | ||
6df264ac | 3176 | multifd_send_sync_main(); |
b2557345 | 3177 | out: |
56e93d26 | 3178 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
35374cbd | 3179 | qemu_fflush(f); |
9360447d | 3180 | ram_counters.transferred += 8; |
56e93d26 JQ |
3181 | |
3182 | ret = qemu_file_get_error(f); | |
3183 | if (ret < 0) { | |
3184 | return ret; | |
3185 | } | |
3186 | ||
5c90308f | 3187 | return done; |
56e93d26 JQ |
3188 | } |
3189 | ||
3d0684b2 JQ |
3190 | /** |
3191 | * ram_save_complete: function called to send the remaining amount of ram | |
3192 | * | |
3193 | * Returns zero to indicate success | |
3194 | * | |
3195 | * Called with iothread lock | |
3196 | * | |
3197 | * @f: QEMUFile where to send the data | |
3198 | * @opaque: RAMState pointer | |
3199 | */ | |
56e93d26 JQ |
3200 | static int ram_save_complete(QEMUFile *f, void *opaque) |
3201 | { | |
53518d94 JQ |
3202 | RAMState **temp = opaque; |
3203 | RAMState *rs = *temp; | |
6f37bb8b | 3204 | |
56e93d26 JQ |
3205 | rcu_read_lock(); |
3206 | ||
5727309d | 3207 | if (!migration_in_postcopy()) { |
8d820d6f | 3208 | migration_bitmap_sync(rs); |
663e6c1d | 3209 | } |
56e93d26 JQ |
3210 | |
3211 | ram_control_before_iterate(f, RAM_CONTROL_FINISH); | |
3212 | ||
3213 | /* try transferring iterative blocks of memory */ | |
3214 | ||
3215 | /* flush all remaining blocks regardless of rate limiting */ | |
3216 | while (true) { | |
3217 | int pages; | |
3218 | ||
ce25d337 | 3219 | pages = ram_find_and_save_block(rs, !migration_in_colo_state()); |
56e93d26 JQ |
3220 | /* no more blocks to sent */ |
3221 | if (pages == 0) { | |
3222 | break; | |
3223 | } | |
3224 | } | |
3225 | ||
ce25d337 | 3226 | flush_compressed_data(rs); |
56e93d26 | 3227 | ram_control_after_iterate(f, RAM_CONTROL_FINISH); |
56e93d26 JQ |
3228 | |
3229 | rcu_read_unlock(); | |
d09a6fde | 3230 | |
6df264ac | 3231 | multifd_send_sync_main(); |
56e93d26 | 3232 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
35374cbd | 3233 | qemu_fflush(f); |
56e93d26 JQ |
3234 | |
3235 | return 0; | |
3236 | } | |
3237 | ||
c31b098f | 3238 | static void ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size, |
47995026 VSO |
3239 | uint64_t *res_precopy_only, |
3240 | uint64_t *res_compatible, | |
3241 | uint64_t *res_postcopy_only) | |
56e93d26 | 3242 | { |
53518d94 JQ |
3243 | RAMState **temp = opaque; |
3244 | RAMState *rs = *temp; | |
56e93d26 JQ |
3245 | uint64_t remaining_size; |
3246 | ||
9edabd4d | 3247 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 3248 | |
5727309d | 3249 | if (!migration_in_postcopy() && |
663e6c1d | 3250 | remaining_size < max_size) { |
56e93d26 JQ |
3251 | qemu_mutex_lock_iothread(); |
3252 | rcu_read_lock(); | |
8d820d6f | 3253 | migration_bitmap_sync(rs); |
56e93d26 JQ |
3254 | rcu_read_unlock(); |
3255 | qemu_mutex_unlock_iothread(); | |
9edabd4d | 3256 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 3257 | } |
c31b098f | 3258 | |
86e1167e VSO |
3259 | if (migrate_postcopy_ram()) { |
3260 | /* We can do postcopy, and all the data is postcopiable */ | |
47995026 | 3261 | *res_compatible += remaining_size; |
86e1167e | 3262 | } else { |
47995026 | 3263 | *res_precopy_only += remaining_size; |
86e1167e | 3264 | } |
56e93d26 JQ |
3265 | } |
3266 | ||
3267 | static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) | |
3268 | { | |
3269 | unsigned int xh_len; | |
3270 | int xh_flags; | |
063e760a | 3271 | uint8_t *loaded_data; |
56e93d26 | 3272 | |
56e93d26 JQ |
3273 | /* extract RLE header */ |
3274 | xh_flags = qemu_get_byte(f); | |
3275 | xh_len = qemu_get_be16(f); | |
3276 | ||
3277 | if (xh_flags != ENCODING_FLAG_XBZRLE) { | |
3278 | error_report("Failed to load XBZRLE page - wrong compression!"); | |
3279 | return -1; | |
3280 | } | |
3281 | ||
3282 | if (xh_len > TARGET_PAGE_SIZE) { | |
3283 | error_report("Failed to load XBZRLE page - len overflow!"); | |
3284 | return -1; | |
3285 | } | |
f265e0e4 | 3286 | loaded_data = XBZRLE.decoded_buf; |
56e93d26 | 3287 | /* load data and decode */ |
f265e0e4 | 3288 | /* it can change loaded_data to point to an internal buffer */ |
063e760a | 3289 | qemu_get_buffer_in_place(f, &loaded_data, xh_len); |
56e93d26 JQ |
3290 | |
3291 | /* decode RLE */ | |
063e760a | 3292 | if (xbzrle_decode_buffer(loaded_data, xh_len, host, |
56e93d26 JQ |
3293 | TARGET_PAGE_SIZE) == -1) { |
3294 | error_report("Failed to load XBZRLE page - decode error!"); | |
3295 | return -1; | |
3296 | } | |
3297 | ||
3298 | return 0; | |
3299 | } | |
3300 | ||
3d0684b2 JQ |
3301 | /** |
3302 | * ram_block_from_stream: read a RAMBlock id from the migration stream | |
3303 | * | |
3304 | * Must be called from within a rcu critical section. | |
3305 | * | |
56e93d26 | 3306 | * Returns a pointer from within the RCU-protected ram_list. |
a7180877 | 3307 | * |
3d0684b2 JQ |
3308 | * @f: QEMUFile where to read the data from |
3309 | * @flags: Page flags (mostly to see if it's a continuation of previous block) | |
a7180877 | 3310 | */ |
3d0684b2 | 3311 | static inline RAMBlock *ram_block_from_stream(QEMUFile *f, int flags) |
56e93d26 JQ |
3312 | { |
3313 | static RAMBlock *block = NULL; | |
3314 | char id[256]; | |
3315 | uint8_t len; | |
3316 | ||
3317 | if (flags & RAM_SAVE_FLAG_CONTINUE) { | |
4c4bad48 | 3318 | if (!block) { |
56e93d26 JQ |
3319 | error_report("Ack, bad migration stream!"); |
3320 | return NULL; | |
3321 | } | |
4c4bad48 | 3322 | return block; |
56e93d26 JQ |
3323 | } |
3324 | ||
3325 | len = qemu_get_byte(f); | |
3326 | qemu_get_buffer(f, (uint8_t *)id, len); | |
3327 | id[len] = 0; | |
3328 | ||
e3dd7493 | 3329 | block = qemu_ram_block_by_name(id); |
4c4bad48 HZ |
3330 | if (!block) { |
3331 | error_report("Can't find block %s", id); | |
3332 | return NULL; | |
56e93d26 JQ |
3333 | } |
3334 | ||
b895de50 CLG |
3335 | if (!qemu_ram_is_migratable(block)) { |
3336 | error_report("block %s should not be migrated !", id); | |
3337 | return NULL; | |
3338 | } | |
3339 | ||
4c4bad48 HZ |
3340 | return block; |
3341 | } | |
3342 | ||
3343 | static inline void *host_from_ram_block_offset(RAMBlock *block, | |
3344 | ram_addr_t offset) | |
3345 | { | |
3346 | if (!offset_in_ramblock(block, offset)) { | |
3347 | return NULL; | |
3348 | } | |
3349 | ||
3350 | return block->host + offset; | |
56e93d26 JQ |
3351 | } |
3352 | ||
3d0684b2 JQ |
3353 | /** |
3354 | * ram_handle_compressed: handle the zero page case | |
3355 | * | |
56e93d26 JQ |
3356 | * If a page (or a whole RDMA chunk) has been |
3357 | * determined to be zero, then zap it. | |
3d0684b2 JQ |
3358 | * |
3359 | * @host: host address for the zero page | |
3360 | * @ch: what the page is filled from. We only support zero | |
3361 | * @size: size of the zero page | |
56e93d26 JQ |
3362 | */ |
3363 | void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) | |
3364 | { | |
3365 | if (ch != 0 || !is_zero_range(host, size)) { | |
3366 | memset(host, ch, size); | |
3367 | } | |
3368 | } | |
3369 | ||
797ca154 XG |
3370 | /* return the size after decompression, or negative value on error */ |
3371 | static int | |
3372 | qemu_uncompress_data(z_stream *stream, uint8_t *dest, size_t dest_len, | |
3373 | const uint8_t *source, size_t source_len) | |
3374 | { | |
3375 | int err; | |
3376 | ||
3377 | err = inflateReset(stream); | |
3378 | if (err != Z_OK) { | |
3379 | return -1; | |
3380 | } | |
3381 | ||
3382 | stream->avail_in = source_len; | |
3383 | stream->next_in = (uint8_t *)source; | |
3384 | stream->avail_out = dest_len; | |
3385 | stream->next_out = dest; | |
3386 | ||
3387 | err = inflate(stream, Z_NO_FLUSH); | |
3388 | if (err != Z_STREAM_END) { | |
3389 | return -1; | |
3390 | } | |
3391 | ||
3392 | return stream->total_out; | |
3393 | } | |
3394 | ||
56e93d26 JQ |
3395 | static void *do_data_decompress(void *opaque) |
3396 | { | |
3397 | DecompressParam *param = opaque; | |
3398 | unsigned long pagesize; | |
33d151f4 | 3399 | uint8_t *des; |
34ab9e97 | 3400 | int len, ret; |
56e93d26 | 3401 | |
33d151f4 | 3402 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 3403 | while (!param->quit) { |
33d151f4 LL |
3404 | if (param->des) { |
3405 | des = param->des; | |
3406 | len = param->len; | |
3407 | param->des = 0; | |
3408 | qemu_mutex_unlock(¶m->mutex); | |
3409 | ||
56e93d26 | 3410 | pagesize = TARGET_PAGE_SIZE; |
34ab9e97 XG |
3411 | |
3412 | ret = qemu_uncompress_data(¶m->stream, des, pagesize, | |
3413 | param->compbuf, len); | |
f548222c | 3414 | if (ret < 0 && migrate_get_current()->decompress_error_check) { |
34ab9e97 XG |
3415 | error_report("decompress data failed"); |
3416 | qemu_file_set_error(decomp_file, ret); | |
3417 | } | |
73a8912b | 3418 | |
33d151f4 LL |
3419 | qemu_mutex_lock(&decomp_done_lock); |
3420 | param->done = true; | |
3421 | qemu_cond_signal(&decomp_done_cond); | |
3422 | qemu_mutex_unlock(&decomp_done_lock); | |
3423 | ||
3424 | qemu_mutex_lock(¶m->mutex); | |
3425 | } else { | |
3426 | qemu_cond_wait(¶m->cond, ¶m->mutex); | |
3427 | } | |
56e93d26 | 3428 | } |
33d151f4 | 3429 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
3430 | |
3431 | return NULL; | |
3432 | } | |
3433 | ||
34ab9e97 | 3434 | static int wait_for_decompress_done(void) |
5533b2e9 LL |
3435 | { |
3436 | int idx, thread_count; | |
3437 | ||
3438 | if (!migrate_use_compression()) { | |
34ab9e97 | 3439 | return 0; |
5533b2e9 LL |
3440 | } |
3441 | ||
3442 | thread_count = migrate_decompress_threads(); | |
3443 | qemu_mutex_lock(&decomp_done_lock); | |
3444 | for (idx = 0; idx < thread_count; idx++) { | |
3445 | while (!decomp_param[idx].done) { | |
3446 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
3447 | } | |
3448 | } | |
3449 | qemu_mutex_unlock(&decomp_done_lock); | |
34ab9e97 | 3450 | return qemu_file_get_error(decomp_file); |
5533b2e9 LL |
3451 | } |
3452 | ||
f0afa331 | 3453 | static void compress_threads_load_cleanup(void) |
56e93d26 JQ |
3454 | { |
3455 | int i, thread_count; | |
3456 | ||
3416ab5b JQ |
3457 | if (!migrate_use_compression()) { |
3458 | return; | |
3459 | } | |
56e93d26 JQ |
3460 | thread_count = migrate_decompress_threads(); |
3461 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
3462 | /* |
3463 | * we use it as a indicator which shows if the thread is | |
3464 | * properly init'd or not | |
3465 | */ | |
3466 | if (!decomp_param[i].compbuf) { | |
3467 | break; | |
3468 | } | |
3469 | ||
56e93d26 | 3470 | qemu_mutex_lock(&decomp_param[i].mutex); |
90e56fb4 | 3471 | decomp_param[i].quit = true; |
56e93d26 JQ |
3472 | qemu_cond_signal(&decomp_param[i].cond); |
3473 | qemu_mutex_unlock(&decomp_param[i].mutex); | |
3474 | } | |
3475 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
3476 | if (!decomp_param[i].compbuf) { |
3477 | break; | |
3478 | } | |
3479 | ||
56e93d26 JQ |
3480 | qemu_thread_join(decompress_threads + i); |
3481 | qemu_mutex_destroy(&decomp_param[i].mutex); | |
3482 | qemu_cond_destroy(&decomp_param[i].cond); | |
797ca154 | 3483 | inflateEnd(&decomp_param[i].stream); |
56e93d26 | 3484 | g_free(decomp_param[i].compbuf); |
797ca154 | 3485 | decomp_param[i].compbuf = NULL; |
56e93d26 JQ |
3486 | } |
3487 | g_free(decompress_threads); | |
3488 | g_free(decomp_param); | |
56e93d26 JQ |
3489 | decompress_threads = NULL; |
3490 | decomp_param = NULL; | |
34ab9e97 | 3491 | decomp_file = NULL; |
56e93d26 JQ |
3492 | } |
3493 | ||
34ab9e97 | 3494 | static int compress_threads_load_setup(QEMUFile *f) |
797ca154 XG |
3495 | { |
3496 | int i, thread_count; | |
3497 | ||
3498 | if (!migrate_use_compression()) { | |
3499 | return 0; | |
3500 | } | |
3501 | ||
3502 | thread_count = migrate_decompress_threads(); | |
3503 | decompress_threads = g_new0(QemuThread, thread_count); | |
3504 | decomp_param = g_new0(DecompressParam, thread_count); | |
3505 | qemu_mutex_init(&decomp_done_lock); | |
3506 | qemu_cond_init(&decomp_done_cond); | |
34ab9e97 | 3507 | decomp_file = f; |
797ca154 XG |
3508 | for (i = 0; i < thread_count; i++) { |
3509 | if (inflateInit(&decomp_param[i].stream) != Z_OK) { | |
3510 | goto exit; | |
3511 | } | |
3512 | ||
3513 | decomp_param[i].compbuf = g_malloc0(compressBound(TARGET_PAGE_SIZE)); | |
3514 | qemu_mutex_init(&decomp_param[i].mutex); | |
3515 | qemu_cond_init(&decomp_param[i].cond); | |
3516 | decomp_param[i].done = true; | |
3517 | decomp_param[i].quit = false; | |
3518 | qemu_thread_create(decompress_threads + i, "decompress", | |
3519 | do_data_decompress, decomp_param + i, | |
3520 | QEMU_THREAD_JOINABLE); | |
3521 | } | |
3522 | return 0; | |
3523 | exit: | |
3524 | compress_threads_load_cleanup(); | |
3525 | return -1; | |
3526 | } | |
3527 | ||
c1bc6626 | 3528 | static void decompress_data_with_multi_threads(QEMUFile *f, |
56e93d26 JQ |
3529 | void *host, int len) |
3530 | { | |
3531 | int idx, thread_count; | |
3532 | ||
3533 | thread_count = migrate_decompress_threads(); | |
73a8912b | 3534 | qemu_mutex_lock(&decomp_done_lock); |
56e93d26 JQ |
3535 | while (true) { |
3536 | for (idx = 0; idx < thread_count; idx++) { | |
73a8912b | 3537 | if (decomp_param[idx].done) { |
33d151f4 LL |
3538 | decomp_param[idx].done = false; |
3539 | qemu_mutex_lock(&decomp_param[idx].mutex); | |
c1bc6626 | 3540 | qemu_get_buffer(f, decomp_param[idx].compbuf, len); |
56e93d26 JQ |
3541 | decomp_param[idx].des = host; |
3542 | decomp_param[idx].len = len; | |
33d151f4 LL |
3543 | qemu_cond_signal(&decomp_param[idx].cond); |
3544 | qemu_mutex_unlock(&decomp_param[idx].mutex); | |
56e93d26 JQ |
3545 | break; |
3546 | } | |
3547 | } | |
3548 | if (idx < thread_count) { | |
3549 | break; | |
73a8912b LL |
3550 | } else { |
3551 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
56e93d26 JQ |
3552 | } |
3553 | } | |
73a8912b | 3554 | qemu_mutex_unlock(&decomp_done_lock); |
56e93d26 JQ |
3555 | } |
3556 | ||
f265e0e4 JQ |
3557 | /** |
3558 | * ram_load_setup: Setup RAM for migration incoming side | |
3559 | * | |
3560 | * Returns zero to indicate success and negative for error | |
3561 | * | |
3562 | * @f: QEMUFile where to receive the data | |
3563 | * @opaque: RAMState pointer | |
3564 | */ | |
3565 | static int ram_load_setup(QEMUFile *f, void *opaque) | |
3566 | { | |
34ab9e97 | 3567 | if (compress_threads_load_setup(f)) { |
797ca154 XG |
3568 | return -1; |
3569 | } | |
3570 | ||
f265e0e4 | 3571 | xbzrle_load_setup(); |
f9494614 | 3572 | ramblock_recv_map_init(); |
f265e0e4 JQ |
3573 | return 0; |
3574 | } | |
3575 | ||
3576 | static int ram_load_cleanup(void *opaque) | |
3577 | { | |
f9494614 | 3578 | RAMBlock *rb; |
56eb90af JH |
3579 | |
3580 | RAMBLOCK_FOREACH_MIGRATABLE(rb) { | |
3581 | if (ramblock_is_pmem(rb)) { | |
3582 | pmem_persist(rb->host, rb->used_length); | |
3583 | } | |
3584 | } | |
3585 | ||
f265e0e4 | 3586 | xbzrle_load_cleanup(); |
f0afa331 | 3587 | compress_threads_load_cleanup(); |
f9494614 | 3588 | |
b895de50 | 3589 | RAMBLOCK_FOREACH_MIGRATABLE(rb) { |
f9494614 AP |
3590 | g_free(rb->receivedmap); |
3591 | rb->receivedmap = NULL; | |
3592 | } | |
f265e0e4 JQ |
3593 | return 0; |
3594 | } | |
3595 | ||
3d0684b2 JQ |
3596 | /** |
3597 | * ram_postcopy_incoming_init: allocate postcopy data structures | |
3598 | * | |
3599 | * Returns 0 for success and negative if there was one error | |
3600 | * | |
3601 | * @mis: current migration incoming state | |
3602 | * | |
3603 | * Allocate data structures etc needed by incoming migration with | |
3604 | * postcopy-ram. postcopy-ram's similarly names | |
3605 | * postcopy_ram_incoming_init does the work. | |
1caddf8a DDAG |
3606 | */ |
3607 | int ram_postcopy_incoming_init(MigrationIncomingState *mis) | |
3608 | { | |
c136180c | 3609 | return postcopy_ram_incoming_init(mis); |
1caddf8a DDAG |
3610 | } |
3611 | ||
3d0684b2 JQ |
3612 | /** |
3613 | * ram_load_postcopy: load a page in postcopy case | |
3614 | * | |
3615 | * Returns 0 for success or -errno in case of error | |
3616 | * | |
a7180877 DDAG |
3617 | * Called in postcopy mode by ram_load(). |
3618 | * rcu_read_lock is taken prior to this being called. | |
3d0684b2 JQ |
3619 | * |
3620 | * @f: QEMUFile where to send the data | |
a7180877 DDAG |
3621 | */ |
3622 | static int ram_load_postcopy(QEMUFile *f) | |
3623 | { | |
3624 | int flags = 0, ret = 0; | |
3625 | bool place_needed = false; | |
1aa83678 | 3626 | bool matches_target_page_size = false; |
a7180877 DDAG |
3627 | MigrationIncomingState *mis = migration_incoming_get_current(); |
3628 | /* Temporary page that is later 'placed' */ | |
3629 | void *postcopy_host_page = postcopy_get_tmp_page(mis); | |
c53b7ddc | 3630 | void *last_host = NULL; |
a3b6ff6d | 3631 | bool all_zero = false; |
a7180877 DDAG |
3632 | |
3633 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
3634 | ram_addr_t addr; | |
3635 | void *host = NULL; | |
3636 | void *page_buffer = NULL; | |
3637 | void *place_source = NULL; | |
df9ff5e1 | 3638 | RAMBlock *block = NULL; |
a7180877 | 3639 | uint8_t ch; |
a7180877 DDAG |
3640 | |
3641 | addr = qemu_get_be64(f); | |
7a9ddfbf PX |
3642 | |
3643 | /* | |
3644 | * If qemu file error, we should stop here, and then "addr" | |
3645 | * may be invalid | |
3646 | */ | |
3647 | ret = qemu_file_get_error(f); | |
3648 | if (ret) { | |
3649 | break; | |
3650 | } | |
3651 | ||
a7180877 DDAG |
3652 | flags = addr & ~TARGET_PAGE_MASK; |
3653 | addr &= TARGET_PAGE_MASK; | |
3654 | ||
3655 | trace_ram_load_postcopy_loop((uint64_t)addr, flags); | |
3656 | place_needed = false; | |
bb890ed5 | 3657 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE)) { |
df9ff5e1 | 3658 | block = ram_block_from_stream(f, flags); |
4c4bad48 HZ |
3659 | |
3660 | host = host_from_ram_block_offset(block, addr); | |
a7180877 DDAG |
3661 | if (!host) { |
3662 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
3663 | ret = -EINVAL; | |
3664 | break; | |
3665 | } | |
1aa83678 | 3666 | matches_target_page_size = block->page_size == TARGET_PAGE_SIZE; |
a7180877 | 3667 | /* |
28abd200 DDAG |
3668 | * Postcopy requires that we place whole host pages atomically; |
3669 | * these may be huge pages for RAMBlocks that are backed by | |
3670 | * hugetlbfs. | |
a7180877 DDAG |
3671 | * To make it atomic, the data is read into a temporary page |
3672 | * that's moved into place later. | |
3673 | * The migration protocol uses, possibly smaller, target-pages | |
3674 | * however the source ensures it always sends all the components | |
3675 | * of a host page in order. | |
3676 | */ | |
3677 | page_buffer = postcopy_host_page + | |
28abd200 | 3678 | ((uintptr_t)host & (block->page_size - 1)); |
a7180877 | 3679 | /* If all TP are zero then we can optimise the place */ |
28abd200 | 3680 | if (!((uintptr_t)host & (block->page_size - 1))) { |
a7180877 | 3681 | all_zero = true; |
c53b7ddc DDAG |
3682 | } else { |
3683 | /* not the 1st TP within the HP */ | |
3684 | if (host != (last_host + TARGET_PAGE_SIZE)) { | |
9af9e0fe | 3685 | error_report("Non-sequential target page %p/%p", |
c53b7ddc DDAG |
3686 | host, last_host); |
3687 | ret = -EINVAL; | |
3688 | break; | |
3689 | } | |
a7180877 DDAG |
3690 | } |
3691 | ||
c53b7ddc | 3692 | |
a7180877 DDAG |
3693 | /* |
3694 | * If it's the last part of a host page then we place the host | |
3695 | * page | |
3696 | */ | |
3697 | place_needed = (((uintptr_t)host + TARGET_PAGE_SIZE) & | |
28abd200 | 3698 | (block->page_size - 1)) == 0; |
a7180877 DDAG |
3699 | place_source = postcopy_host_page; |
3700 | } | |
c53b7ddc | 3701 | last_host = host; |
a7180877 DDAG |
3702 | |
3703 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { | |
bb890ed5 | 3704 | case RAM_SAVE_FLAG_ZERO: |
a7180877 DDAG |
3705 | ch = qemu_get_byte(f); |
3706 | memset(page_buffer, ch, TARGET_PAGE_SIZE); | |
3707 | if (ch) { | |
3708 | all_zero = false; | |
3709 | } | |
3710 | break; | |
3711 | ||
3712 | case RAM_SAVE_FLAG_PAGE: | |
3713 | all_zero = false; | |
1aa83678 PX |
3714 | if (!matches_target_page_size) { |
3715 | /* For huge pages, we always use temporary buffer */ | |
a7180877 DDAG |
3716 | qemu_get_buffer(f, page_buffer, TARGET_PAGE_SIZE); |
3717 | } else { | |
1aa83678 PX |
3718 | /* |
3719 | * For small pages that matches target page size, we | |
3720 | * avoid the qemu_file copy. Instead we directly use | |
3721 | * the buffer of QEMUFile to place the page. Note: we | |
3722 | * cannot do any QEMUFile operation before using that | |
3723 | * buffer to make sure the buffer is valid when | |
3724 | * placing the page. | |
a7180877 DDAG |
3725 | */ |
3726 | qemu_get_buffer_in_place(f, (uint8_t **)&place_source, | |
3727 | TARGET_PAGE_SIZE); | |
3728 | } | |
3729 | break; | |
3730 | case RAM_SAVE_FLAG_EOS: | |
3731 | /* normal exit */ | |
6df264ac | 3732 | multifd_recv_sync_main(); |
a7180877 DDAG |
3733 | break; |
3734 | default: | |
3735 | error_report("Unknown combination of migration flags: %#x" | |
3736 | " (postcopy mode)", flags); | |
3737 | ret = -EINVAL; | |
7a9ddfbf PX |
3738 | break; |
3739 | } | |
3740 | ||
3741 | /* Detect for any possible file errors */ | |
3742 | if (!ret && qemu_file_get_error(f)) { | |
3743 | ret = qemu_file_get_error(f); | |
a7180877 DDAG |
3744 | } |
3745 | ||
7a9ddfbf | 3746 | if (!ret && place_needed) { |
a7180877 | 3747 | /* This gets called at the last target page in the host page */ |
df9ff5e1 DDAG |
3748 | void *place_dest = host + TARGET_PAGE_SIZE - block->page_size; |
3749 | ||
a7180877 | 3750 | if (all_zero) { |
df9ff5e1 | 3751 | ret = postcopy_place_page_zero(mis, place_dest, |
8be4620b | 3752 | block); |
a7180877 | 3753 | } else { |
df9ff5e1 | 3754 | ret = postcopy_place_page(mis, place_dest, |
8be4620b | 3755 | place_source, block); |
a7180877 DDAG |
3756 | } |
3757 | } | |
a7180877 DDAG |
3758 | } |
3759 | ||
3760 | return ret; | |
3761 | } | |
3762 | ||
acab30b8 DHB |
3763 | static bool postcopy_is_advised(void) |
3764 | { | |
3765 | PostcopyState ps = postcopy_state_get(); | |
3766 | return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END; | |
3767 | } | |
3768 | ||
3769 | static bool postcopy_is_running(void) | |
3770 | { | |
3771 | PostcopyState ps = postcopy_state_get(); | |
3772 | return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; | |
3773 | } | |
3774 | ||
56e93d26 JQ |
3775 | static int ram_load(QEMUFile *f, void *opaque, int version_id) |
3776 | { | |
edc60127 | 3777 | int flags = 0, ret = 0, invalid_flags = 0; |
56e93d26 JQ |
3778 | static uint64_t seq_iter; |
3779 | int len = 0; | |
a7180877 DDAG |
3780 | /* |
3781 | * If system is running in postcopy mode, page inserts to host memory must | |
3782 | * be atomic | |
3783 | */ | |
acab30b8 | 3784 | bool postcopy_running = postcopy_is_running(); |
ef08fb38 | 3785 | /* ADVISE is earlier, it shows the source has the postcopy capability on */ |
acab30b8 | 3786 | bool postcopy_advised = postcopy_is_advised(); |
56e93d26 JQ |
3787 | |
3788 | seq_iter++; | |
3789 | ||
3790 | if (version_id != 4) { | |
3791 | ret = -EINVAL; | |
3792 | } | |
3793 | ||
edc60127 JQ |
3794 | if (!migrate_use_compression()) { |
3795 | invalid_flags |= RAM_SAVE_FLAG_COMPRESS_PAGE; | |
3796 | } | |
56e93d26 JQ |
3797 | /* This RCU critical section can be very long running. |
3798 | * When RCU reclaims in the code start to become numerous, | |
3799 | * it will be necessary to reduce the granularity of this | |
3800 | * critical section. | |
3801 | */ | |
3802 | rcu_read_lock(); | |
a7180877 DDAG |
3803 | |
3804 | if (postcopy_running) { | |
3805 | ret = ram_load_postcopy(f); | |
3806 | } | |
3807 | ||
3808 | while (!postcopy_running && !ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
56e93d26 | 3809 | ram_addr_t addr, total_ram_bytes; |
a776aa15 | 3810 | void *host = NULL; |
56e93d26 JQ |
3811 | uint8_t ch; |
3812 | ||
3813 | addr = qemu_get_be64(f); | |
3814 | flags = addr & ~TARGET_PAGE_MASK; | |
3815 | addr &= TARGET_PAGE_MASK; | |
3816 | ||
edc60127 JQ |
3817 | if (flags & invalid_flags) { |
3818 | if (flags & invalid_flags & RAM_SAVE_FLAG_COMPRESS_PAGE) { | |
3819 | error_report("Received an unexpected compressed page"); | |
3820 | } | |
3821 | ||
3822 | ret = -EINVAL; | |
3823 | break; | |
3824 | } | |
3825 | ||
bb890ed5 | 3826 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
a776aa15 | 3827 | RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) { |
4c4bad48 HZ |
3828 | RAMBlock *block = ram_block_from_stream(f, flags); |
3829 | ||
3830 | host = host_from_ram_block_offset(block, addr); | |
a776aa15 DDAG |
3831 | if (!host) { |
3832 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
3833 | ret = -EINVAL; | |
3834 | break; | |
3835 | } | |
f9494614 | 3836 | ramblock_recv_bitmap_set(block, host); |
1db9d8e5 | 3837 | trace_ram_load_loop(block->idstr, (uint64_t)addr, flags, host); |
a776aa15 DDAG |
3838 | } |
3839 | ||
56e93d26 JQ |
3840 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { |
3841 | case RAM_SAVE_FLAG_MEM_SIZE: | |
3842 | /* Synchronize RAM block list */ | |
3843 | total_ram_bytes = addr; | |
3844 | while (!ret && total_ram_bytes) { | |
3845 | RAMBlock *block; | |
56e93d26 JQ |
3846 | char id[256]; |
3847 | ram_addr_t length; | |
3848 | ||
3849 | len = qemu_get_byte(f); | |
3850 | qemu_get_buffer(f, (uint8_t *)id, len); | |
3851 | id[len] = 0; | |
3852 | length = qemu_get_be64(f); | |
3853 | ||
e3dd7493 | 3854 | block = qemu_ram_block_by_name(id); |
b895de50 CLG |
3855 | if (block && !qemu_ram_is_migratable(block)) { |
3856 | error_report("block %s should not be migrated !", id); | |
3857 | ret = -EINVAL; | |
3858 | } else if (block) { | |
e3dd7493 DDAG |
3859 | if (length != block->used_length) { |
3860 | Error *local_err = NULL; | |
56e93d26 | 3861 | |
fa53a0e5 | 3862 | ret = qemu_ram_resize(block, length, |
e3dd7493 DDAG |
3863 | &local_err); |
3864 | if (local_err) { | |
3865 | error_report_err(local_err); | |
56e93d26 | 3866 | } |
56e93d26 | 3867 | } |
ef08fb38 DDAG |
3868 | /* For postcopy we need to check hugepage sizes match */ |
3869 | if (postcopy_advised && | |
3870 | block->page_size != qemu_host_page_size) { | |
3871 | uint64_t remote_page_size = qemu_get_be64(f); | |
3872 | if (remote_page_size != block->page_size) { | |
3873 | error_report("Mismatched RAM page size %s " | |
3874 | "(local) %zd != %" PRId64, | |
3875 | id, block->page_size, | |
3876 | remote_page_size); | |
3877 | ret = -EINVAL; | |
3878 | } | |
3879 | } | |
e3dd7493 DDAG |
3880 | ram_control_load_hook(f, RAM_CONTROL_BLOCK_REG, |
3881 | block->idstr); | |
3882 | } else { | |
56e93d26 JQ |
3883 | error_report("Unknown ramblock \"%s\", cannot " |
3884 | "accept migration", id); | |
3885 | ret = -EINVAL; | |
3886 | } | |
3887 | ||
3888 | total_ram_bytes -= length; | |
3889 | } | |
3890 | break; | |
a776aa15 | 3891 | |
bb890ed5 | 3892 | case RAM_SAVE_FLAG_ZERO: |
56e93d26 JQ |
3893 | ch = qemu_get_byte(f); |
3894 | ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); | |
3895 | break; | |
a776aa15 | 3896 | |
56e93d26 | 3897 | case RAM_SAVE_FLAG_PAGE: |
56e93d26 JQ |
3898 | qemu_get_buffer(f, host, TARGET_PAGE_SIZE); |
3899 | break; | |
56e93d26 | 3900 | |
a776aa15 | 3901 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
56e93d26 JQ |
3902 | len = qemu_get_be32(f); |
3903 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
3904 | error_report("Invalid compressed data length: %d", len); | |
3905 | ret = -EINVAL; | |
3906 | break; | |
3907 | } | |
c1bc6626 | 3908 | decompress_data_with_multi_threads(f, host, len); |
56e93d26 | 3909 | break; |
a776aa15 | 3910 | |
56e93d26 | 3911 | case RAM_SAVE_FLAG_XBZRLE: |
56e93d26 JQ |
3912 | if (load_xbzrle(f, addr, host) < 0) { |
3913 | error_report("Failed to decompress XBZRLE page at " | |
3914 | RAM_ADDR_FMT, addr); | |
3915 | ret = -EINVAL; | |
3916 | break; | |
3917 | } | |
3918 | break; | |
3919 | case RAM_SAVE_FLAG_EOS: | |
3920 | /* normal exit */ | |
6df264ac | 3921 | multifd_recv_sync_main(); |
56e93d26 JQ |
3922 | break; |
3923 | default: | |
3924 | if (flags & RAM_SAVE_FLAG_HOOK) { | |
632e3a5c | 3925 | ram_control_load_hook(f, RAM_CONTROL_HOOK, NULL); |
56e93d26 JQ |
3926 | } else { |
3927 | error_report("Unknown combination of migration flags: %#x", | |
3928 | flags); | |
3929 | ret = -EINVAL; | |
3930 | } | |
3931 | } | |
3932 | if (!ret) { | |
3933 | ret = qemu_file_get_error(f); | |
3934 | } | |
3935 | } | |
3936 | ||
34ab9e97 | 3937 | ret |= wait_for_decompress_done(); |
56e93d26 | 3938 | rcu_read_unlock(); |
55c4446b | 3939 | trace_ram_load_complete(ret, seq_iter); |
56e93d26 JQ |
3940 | return ret; |
3941 | } | |
3942 | ||
c6467627 VSO |
3943 | static bool ram_has_postcopy(void *opaque) |
3944 | { | |
469dd51b JH |
3945 | RAMBlock *rb; |
3946 | RAMBLOCK_FOREACH_MIGRATABLE(rb) { | |
3947 | if (ramblock_is_pmem(rb)) { | |
3948 | info_report("Block: %s, host: %p is a nvdimm memory, postcopy" | |
3949 | "is not supported now!", rb->idstr, rb->host); | |
3950 | return false; | |
3951 | } | |
3952 | } | |
3953 | ||
c6467627 VSO |
3954 | return migrate_postcopy_ram(); |
3955 | } | |
3956 | ||
edd090c7 PX |
3957 | /* Sync all the dirty bitmap with destination VM. */ |
3958 | static int ram_dirty_bitmap_sync_all(MigrationState *s, RAMState *rs) | |
3959 | { | |
3960 | RAMBlock *block; | |
3961 | QEMUFile *file = s->to_dst_file; | |
3962 | int ramblock_count = 0; | |
3963 | ||
3964 | trace_ram_dirty_bitmap_sync_start(); | |
3965 | ||
ff0769a4 | 3966 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
edd090c7 PX |
3967 | qemu_savevm_send_recv_bitmap(file, block->idstr); |
3968 | trace_ram_dirty_bitmap_request(block->idstr); | |
3969 | ramblock_count++; | |
3970 | } | |
3971 | ||
3972 | trace_ram_dirty_bitmap_sync_wait(); | |
3973 | ||
3974 | /* Wait until all the ramblocks' dirty bitmap synced */ | |
3975 | while (ramblock_count--) { | |
3976 | qemu_sem_wait(&s->rp_state.rp_sem); | |
3977 | } | |
3978 | ||
3979 | trace_ram_dirty_bitmap_sync_complete(); | |
3980 | ||
3981 | return 0; | |
3982 | } | |
3983 | ||
3984 | static void ram_dirty_bitmap_reload_notify(MigrationState *s) | |
3985 | { | |
3986 | qemu_sem_post(&s->rp_state.rp_sem); | |
3987 | } | |
3988 | ||
a335debb PX |
3989 | /* |
3990 | * Read the received bitmap, revert it as the initial dirty bitmap. | |
3991 | * This is only used when the postcopy migration is paused but wants | |
3992 | * to resume from a middle point. | |
3993 | */ | |
3994 | int ram_dirty_bitmap_reload(MigrationState *s, RAMBlock *block) | |
3995 | { | |
3996 | int ret = -EINVAL; | |
3997 | QEMUFile *file = s->rp_state.from_dst_file; | |
3998 | unsigned long *le_bitmap, nbits = block->used_length >> TARGET_PAGE_BITS; | |
a725ef9f | 3999 | uint64_t local_size = DIV_ROUND_UP(nbits, 8); |
a335debb PX |
4000 | uint64_t size, end_mark; |
4001 | ||
4002 | trace_ram_dirty_bitmap_reload_begin(block->idstr); | |
4003 | ||
4004 | if (s->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
4005 | error_report("%s: incorrect state %s", __func__, | |
4006 | MigrationStatus_str(s->state)); | |
4007 | return -EINVAL; | |
4008 | } | |
4009 | ||
4010 | /* | |
4011 | * Note: see comments in ramblock_recv_bitmap_send() on why we | |
4012 | * need the endianess convertion, and the paddings. | |
4013 | */ | |
4014 | local_size = ROUND_UP(local_size, 8); | |
4015 | ||
4016 | /* Add paddings */ | |
4017 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
4018 | ||
4019 | size = qemu_get_be64(file); | |
4020 | ||
4021 | /* The size of the bitmap should match with our ramblock */ | |
4022 | if (size != local_size) { | |
4023 | error_report("%s: ramblock '%s' bitmap size mismatch " | |
4024 | "(0x%"PRIx64" != 0x%"PRIx64")", __func__, | |
4025 | block->idstr, size, local_size); | |
4026 | ret = -EINVAL; | |
4027 | goto out; | |
4028 | } | |
4029 | ||
4030 | size = qemu_get_buffer(file, (uint8_t *)le_bitmap, local_size); | |
4031 | end_mark = qemu_get_be64(file); | |
4032 | ||
4033 | ret = qemu_file_get_error(file); | |
4034 | if (ret || size != local_size) { | |
4035 | error_report("%s: read bitmap failed for ramblock '%s': %d" | |
4036 | " (size 0x%"PRIx64", got: 0x%"PRIx64")", | |
4037 | __func__, block->idstr, ret, local_size, size); | |
4038 | ret = -EIO; | |
4039 | goto out; | |
4040 | } | |
4041 | ||
4042 | if (end_mark != RAMBLOCK_RECV_BITMAP_ENDING) { | |
4043 | error_report("%s: ramblock '%s' end mark incorrect: 0x%"PRIu64, | |
4044 | __func__, block->idstr, end_mark); | |
4045 | ret = -EINVAL; | |
4046 | goto out; | |
4047 | } | |
4048 | ||
4049 | /* | |
4050 | * Endianess convertion. We are during postcopy (though paused). | |
4051 | * The dirty bitmap won't change. We can directly modify it. | |
4052 | */ | |
4053 | bitmap_from_le(block->bmap, le_bitmap, nbits); | |
4054 | ||
4055 | /* | |
4056 | * What we received is "received bitmap". Revert it as the initial | |
4057 | * dirty bitmap for this ramblock. | |
4058 | */ | |
4059 | bitmap_complement(block->bmap, block->bmap, nbits); | |
4060 | ||
4061 | trace_ram_dirty_bitmap_reload_complete(block->idstr); | |
4062 | ||
edd090c7 PX |
4063 | /* |
4064 | * We succeeded to sync bitmap for current ramblock. If this is | |
4065 | * the last one to sync, we need to notify the main send thread. | |
4066 | */ | |
4067 | ram_dirty_bitmap_reload_notify(s); | |
4068 | ||
a335debb PX |
4069 | ret = 0; |
4070 | out: | |
bf269906 | 4071 | g_free(le_bitmap); |
a335debb PX |
4072 | return ret; |
4073 | } | |
4074 | ||
edd090c7 PX |
4075 | static int ram_resume_prepare(MigrationState *s, void *opaque) |
4076 | { | |
4077 | RAMState *rs = *(RAMState **)opaque; | |
08614f34 | 4078 | int ret; |
edd090c7 | 4079 | |
08614f34 PX |
4080 | ret = ram_dirty_bitmap_sync_all(s, rs); |
4081 | if (ret) { | |
4082 | return ret; | |
4083 | } | |
4084 | ||
4085 | ram_state_resume_prepare(rs, s->to_dst_file); | |
4086 | ||
4087 | return 0; | |
edd090c7 PX |
4088 | } |
4089 | ||
56e93d26 | 4090 | static SaveVMHandlers savevm_ram_handlers = { |
9907e842 | 4091 | .save_setup = ram_save_setup, |
56e93d26 | 4092 | .save_live_iterate = ram_save_iterate, |
763c906b | 4093 | .save_live_complete_postcopy = ram_save_complete, |
a3e06c3d | 4094 | .save_live_complete_precopy = ram_save_complete, |
c6467627 | 4095 | .has_postcopy = ram_has_postcopy, |
56e93d26 JQ |
4096 | .save_live_pending = ram_save_pending, |
4097 | .load_state = ram_load, | |
f265e0e4 JQ |
4098 | .save_cleanup = ram_save_cleanup, |
4099 | .load_setup = ram_load_setup, | |
4100 | .load_cleanup = ram_load_cleanup, | |
edd090c7 | 4101 | .resume_prepare = ram_resume_prepare, |
56e93d26 JQ |
4102 | }; |
4103 | ||
4104 | void ram_mig_init(void) | |
4105 | { | |
4106 | qemu_mutex_init(&XBZRLE.lock); | |
6f37bb8b | 4107 | register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, &ram_state); |
56e93d26 | 4108 | } |