]>
Commit | Line | Data |
---|---|---|
56e93d26 JQ |
1 | /* |
2 | * QEMU System Emulator | |
3 | * | |
4 | * Copyright (c) 2003-2008 Fabrice Bellard | |
76cc7b58 JQ |
5 | * Copyright (c) 2011-2015 Red Hat Inc |
6 | * | |
7 | * Authors: | |
8 | * Juan Quintela <quintela@redhat.com> | |
56e93d26 JQ |
9 | * |
10 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
11 | * of this software and associated documentation files (the "Software"), to deal | |
12 | * in the Software without restriction, including without limitation the rights | |
13 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
14 | * copies of the Software, and to permit persons to whom the Software is | |
15 | * furnished to do so, subject to the following conditions: | |
16 | * | |
17 | * The above copyright notice and this permission notice shall be included in | |
18 | * all copies or substantial portions of the Software. | |
19 | * | |
20 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
21 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
22 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
23 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
24 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
25 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
26 | * THE SOFTWARE. | |
27 | */ | |
e688df6b | 28 | |
1393a485 | 29 | #include "qemu/osdep.h" |
33c11879 | 30 | #include "cpu.h" |
56e93d26 | 31 | #include <zlib.h> |
f348b6d1 | 32 | #include "qemu/cutils.h" |
56e93d26 JQ |
33 | #include "qemu/bitops.h" |
34 | #include "qemu/bitmap.h" | |
7205c9ec | 35 | #include "qemu/main-loop.h" |
709e3fe8 | 36 | #include "xbzrle.h" |
7b1e1a22 | 37 | #include "ram.h" |
6666c96a | 38 | #include "migration.h" |
71bb07db | 39 | #include "socket.h" |
f2a8f0a6 | 40 | #include "migration/register.h" |
7b1e1a22 | 41 | #include "migration/misc.h" |
08a0aee1 | 42 | #include "qemu-file.h" |
be07b0ac | 43 | #include "postcopy-ram.h" |
56e93d26 | 44 | #include "migration/page_cache.h" |
56e93d26 | 45 | #include "qemu/error-report.h" |
e688df6b | 46 | #include "qapi/error.h" |
9af23989 | 47 | #include "qapi/qapi-events-migration.h" |
8acabf69 | 48 | #include "qapi/qmp/qerror.h" |
56e93d26 | 49 | #include "trace.h" |
56e93d26 | 50 | #include "exec/ram_addr.h" |
f9494614 | 51 | #include "exec/target_page.h" |
56e93d26 | 52 | #include "qemu/rcu_queue.h" |
a91246c9 | 53 | #include "migration/colo.h" |
9ac78b61 | 54 | #include "migration/block.h" |
af8b7d2b JQ |
55 | #include "sysemu/sysemu.h" |
56 | #include "qemu/uuid.h" | |
edd090c7 | 57 | #include "savevm.h" |
56e93d26 | 58 | |
56e93d26 JQ |
59 | /***********************************************************/ |
60 | /* ram save/restore */ | |
61 | ||
bb890ed5 JQ |
62 | /* RAM_SAVE_FLAG_ZERO used to be named RAM_SAVE_FLAG_COMPRESS, it |
63 | * worked for pages that where filled with the same char. We switched | |
64 | * it to only search for the zero value. And to avoid confusion with | |
65 | * RAM_SSAVE_FLAG_COMPRESS_PAGE just rename it. | |
66 | */ | |
67 | ||
56e93d26 | 68 | #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */ |
bb890ed5 | 69 | #define RAM_SAVE_FLAG_ZERO 0x02 |
56e93d26 JQ |
70 | #define RAM_SAVE_FLAG_MEM_SIZE 0x04 |
71 | #define RAM_SAVE_FLAG_PAGE 0x08 | |
72 | #define RAM_SAVE_FLAG_EOS 0x10 | |
73 | #define RAM_SAVE_FLAG_CONTINUE 0x20 | |
74 | #define RAM_SAVE_FLAG_XBZRLE 0x40 | |
75 | /* 0x80 is reserved in migration.h start with 0x100 next */ | |
76 | #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 | |
77 | ||
56e93d26 JQ |
78 | static inline bool is_zero_range(uint8_t *p, uint64_t size) |
79 | { | |
a1febc49 | 80 | return buffer_is_zero(p, size); |
56e93d26 JQ |
81 | } |
82 | ||
9360447d JQ |
83 | XBZRLECacheStats xbzrle_counters; |
84 | ||
56e93d26 JQ |
85 | /* struct contains XBZRLE cache and a static page |
86 | used by the compression */ | |
87 | static struct { | |
88 | /* buffer used for XBZRLE encoding */ | |
89 | uint8_t *encoded_buf; | |
90 | /* buffer for storing page content */ | |
91 | uint8_t *current_buf; | |
92 | /* Cache for XBZRLE, Protected by lock. */ | |
93 | PageCache *cache; | |
94 | QemuMutex lock; | |
c00e0928 JQ |
95 | /* it will store a page full of zeros */ |
96 | uint8_t *zero_target_page; | |
f265e0e4 JQ |
97 | /* buffer used for XBZRLE decoding */ |
98 | uint8_t *decoded_buf; | |
56e93d26 JQ |
99 | } XBZRLE; |
100 | ||
56e93d26 JQ |
101 | static void XBZRLE_cache_lock(void) |
102 | { | |
103 | if (migrate_use_xbzrle()) | |
104 | qemu_mutex_lock(&XBZRLE.lock); | |
105 | } | |
106 | ||
107 | static void XBZRLE_cache_unlock(void) | |
108 | { | |
109 | if (migrate_use_xbzrle()) | |
110 | qemu_mutex_unlock(&XBZRLE.lock); | |
111 | } | |
112 | ||
3d0684b2 JQ |
113 | /** |
114 | * xbzrle_cache_resize: resize the xbzrle cache | |
115 | * | |
116 | * This function is called from qmp_migrate_set_cache_size in main | |
117 | * thread, possibly while a migration is in progress. A running | |
118 | * migration may be using the cache and might finish during this call, | |
119 | * hence changes to the cache are protected by XBZRLE.lock(). | |
120 | * | |
c9dede2d | 121 | * Returns 0 for success or -1 for error |
3d0684b2 JQ |
122 | * |
123 | * @new_size: new cache size | |
8acabf69 | 124 | * @errp: set *errp if the check failed, with reason |
56e93d26 | 125 | */ |
c9dede2d | 126 | int xbzrle_cache_resize(int64_t new_size, Error **errp) |
56e93d26 JQ |
127 | { |
128 | PageCache *new_cache; | |
c9dede2d | 129 | int64_t ret = 0; |
56e93d26 | 130 | |
8acabf69 JQ |
131 | /* Check for truncation */ |
132 | if (new_size != (size_t)new_size) { | |
133 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", | |
134 | "exceeding address space"); | |
135 | return -1; | |
136 | } | |
137 | ||
2a313e5c JQ |
138 | if (new_size == migrate_xbzrle_cache_size()) { |
139 | /* nothing to do */ | |
c9dede2d | 140 | return 0; |
2a313e5c JQ |
141 | } |
142 | ||
56e93d26 JQ |
143 | XBZRLE_cache_lock(); |
144 | ||
145 | if (XBZRLE.cache != NULL) { | |
80f8dfde | 146 | new_cache = cache_init(new_size, TARGET_PAGE_SIZE, errp); |
56e93d26 | 147 | if (!new_cache) { |
56e93d26 JQ |
148 | ret = -1; |
149 | goto out; | |
150 | } | |
151 | ||
152 | cache_fini(XBZRLE.cache); | |
153 | XBZRLE.cache = new_cache; | |
154 | } | |
56e93d26 JQ |
155 | out: |
156 | XBZRLE_cache_unlock(); | |
157 | return ret; | |
158 | } | |
159 | ||
f9494614 AP |
160 | static void ramblock_recv_map_init(void) |
161 | { | |
162 | RAMBlock *rb; | |
163 | ||
164 | RAMBLOCK_FOREACH(rb) { | |
165 | assert(!rb->receivedmap); | |
166 | rb->receivedmap = bitmap_new(rb->max_length >> qemu_target_page_bits()); | |
167 | } | |
168 | } | |
169 | ||
170 | int ramblock_recv_bitmap_test(RAMBlock *rb, void *host_addr) | |
171 | { | |
172 | return test_bit(ramblock_recv_bitmap_offset(host_addr, rb), | |
173 | rb->receivedmap); | |
174 | } | |
175 | ||
1cba9f6e DDAG |
176 | bool ramblock_recv_bitmap_test_byte_offset(RAMBlock *rb, uint64_t byte_offset) |
177 | { | |
178 | return test_bit(byte_offset >> TARGET_PAGE_BITS, rb->receivedmap); | |
179 | } | |
180 | ||
f9494614 AP |
181 | void ramblock_recv_bitmap_set(RAMBlock *rb, void *host_addr) |
182 | { | |
183 | set_bit_atomic(ramblock_recv_bitmap_offset(host_addr, rb), rb->receivedmap); | |
184 | } | |
185 | ||
186 | void ramblock_recv_bitmap_set_range(RAMBlock *rb, void *host_addr, | |
187 | size_t nr) | |
188 | { | |
189 | bitmap_set_atomic(rb->receivedmap, | |
190 | ramblock_recv_bitmap_offset(host_addr, rb), | |
191 | nr); | |
192 | } | |
193 | ||
a335debb PX |
194 | #define RAMBLOCK_RECV_BITMAP_ENDING (0x0123456789abcdefULL) |
195 | ||
196 | /* | |
197 | * Format: bitmap_size (8 bytes) + whole_bitmap (N bytes). | |
198 | * | |
199 | * Returns >0 if success with sent bytes, or <0 if error. | |
200 | */ | |
201 | int64_t ramblock_recv_bitmap_send(QEMUFile *file, | |
202 | const char *block_name) | |
203 | { | |
204 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
205 | unsigned long *le_bitmap, nbits; | |
206 | uint64_t size; | |
207 | ||
208 | if (!block) { | |
209 | error_report("%s: invalid block name: %s", __func__, block_name); | |
210 | return -1; | |
211 | } | |
212 | ||
213 | nbits = block->used_length >> TARGET_PAGE_BITS; | |
214 | ||
215 | /* | |
216 | * Make sure the tmp bitmap buffer is big enough, e.g., on 32bit | |
217 | * machines we may need 4 more bytes for padding (see below | |
218 | * comment). So extend it a bit before hand. | |
219 | */ | |
220 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
221 | ||
222 | /* | |
223 | * Always use little endian when sending the bitmap. This is | |
224 | * required that when source and destination VMs are not using the | |
225 | * same endianess. (Note: big endian won't work.) | |
226 | */ | |
227 | bitmap_to_le(le_bitmap, block->receivedmap, nbits); | |
228 | ||
229 | /* Size of the bitmap, in bytes */ | |
230 | size = nbits / 8; | |
231 | ||
232 | /* | |
233 | * size is always aligned to 8 bytes for 64bit machines, but it | |
234 | * may not be true for 32bit machines. We need this padding to | |
235 | * make sure the migration can survive even between 32bit and | |
236 | * 64bit machines. | |
237 | */ | |
238 | size = ROUND_UP(size, 8); | |
239 | ||
240 | qemu_put_be64(file, size); | |
241 | qemu_put_buffer(file, (const uint8_t *)le_bitmap, size); | |
242 | /* | |
243 | * Mark as an end, in case the middle part is screwed up due to | |
244 | * some "misterious" reason. | |
245 | */ | |
246 | qemu_put_be64(file, RAMBLOCK_RECV_BITMAP_ENDING); | |
247 | qemu_fflush(file); | |
248 | ||
bf269906 | 249 | g_free(le_bitmap); |
a335debb PX |
250 | |
251 | if (qemu_file_get_error(file)) { | |
252 | return qemu_file_get_error(file); | |
253 | } | |
254 | ||
255 | return size + sizeof(size); | |
256 | } | |
257 | ||
ec481c6c JQ |
258 | /* |
259 | * An outstanding page request, on the source, having been received | |
260 | * and queued | |
261 | */ | |
262 | struct RAMSrcPageRequest { | |
263 | RAMBlock *rb; | |
264 | hwaddr offset; | |
265 | hwaddr len; | |
266 | ||
267 | QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; | |
268 | }; | |
269 | ||
6f37bb8b JQ |
270 | /* State of RAM for migration */ |
271 | struct RAMState { | |
204b88b8 JQ |
272 | /* QEMUFile used for this migration */ |
273 | QEMUFile *f; | |
6f37bb8b JQ |
274 | /* Last block that we have visited searching for dirty pages */ |
275 | RAMBlock *last_seen_block; | |
276 | /* Last block from where we have sent data */ | |
277 | RAMBlock *last_sent_block; | |
269ace29 JQ |
278 | /* Last dirty target page we have sent */ |
279 | ram_addr_t last_page; | |
6f37bb8b JQ |
280 | /* last ram version we have seen */ |
281 | uint32_t last_version; | |
282 | /* We are in the first round */ | |
283 | bool ram_bulk_stage; | |
8d820d6f JQ |
284 | /* How many times we have dirty too many pages */ |
285 | int dirty_rate_high_cnt; | |
f664da80 JQ |
286 | /* these variables are used for bitmap sync */ |
287 | /* last time we did a full bitmap_sync */ | |
288 | int64_t time_last_bitmap_sync; | |
eac74159 | 289 | /* bytes transferred at start_time */ |
c4bdf0cf | 290 | uint64_t bytes_xfer_prev; |
a66cd90c | 291 | /* number of dirty pages since start_time */ |
68908ed6 | 292 | uint64_t num_dirty_pages_period; |
b5833fde JQ |
293 | /* xbzrle misses since the beginning of the period */ |
294 | uint64_t xbzrle_cache_miss_prev; | |
36040d9c JQ |
295 | /* number of iterations at the beginning of period */ |
296 | uint64_t iterations_prev; | |
23b28c3c JQ |
297 | /* Iterations since start */ |
298 | uint64_t iterations; | |
9360447d | 299 | /* number of dirty bits in the bitmap */ |
2dfaf12e PX |
300 | uint64_t migration_dirty_pages; |
301 | /* protects modification of the bitmap */ | |
108cfae0 | 302 | QemuMutex bitmap_mutex; |
68a098f3 JQ |
303 | /* The RAMBlock used in the last src_page_requests */ |
304 | RAMBlock *last_req_rb; | |
ec481c6c JQ |
305 | /* Queue of outstanding page requests from the destination */ |
306 | QemuMutex src_page_req_mutex; | |
307 | QSIMPLEQ_HEAD(src_page_requests, RAMSrcPageRequest) src_page_requests; | |
6f37bb8b JQ |
308 | }; |
309 | typedef struct RAMState RAMState; | |
310 | ||
53518d94 | 311 | static RAMState *ram_state; |
6f37bb8b | 312 | |
9edabd4d | 313 | uint64_t ram_bytes_remaining(void) |
2f4fde93 | 314 | { |
bae416e5 DDAG |
315 | return ram_state ? (ram_state->migration_dirty_pages * TARGET_PAGE_SIZE) : |
316 | 0; | |
2f4fde93 JQ |
317 | } |
318 | ||
9360447d | 319 | MigrationStats ram_counters; |
96506894 | 320 | |
b8fb8cb7 DDAG |
321 | /* used by the search for pages to send */ |
322 | struct PageSearchStatus { | |
323 | /* Current block being searched */ | |
324 | RAMBlock *block; | |
a935e30f JQ |
325 | /* Current page to search from */ |
326 | unsigned long page; | |
b8fb8cb7 DDAG |
327 | /* Set once we wrap around */ |
328 | bool complete_round; | |
329 | }; | |
330 | typedef struct PageSearchStatus PageSearchStatus; | |
331 | ||
56e93d26 | 332 | struct CompressParam { |
56e93d26 | 333 | bool done; |
90e56fb4 | 334 | bool quit; |
56e93d26 JQ |
335 | QEMUFile *file; |
336 | QemuMutex mutex; | |
337 | QemuCond cond; | |
338 | RAMBlock *block; | |
339 | ram_addr_t offset; | |
34ab9e97 XG |
340 | |
341 | /* internally used fields */ | |
dcaf446e | 342 | z_stream stream; |
34ab9e97 | 343 | uint8_t *originbuf; |
56e93d26 JQ |
344 | }; |
345 | typedef struct CompressParam CompressParam; | |
346 | ||
347 | struct DecompressParam { | |
73a8912b | 348 | bool done; |
90e56fb4 | 349 | bool quit; |
56e93d26 JQ |
350 | QemuMutex mutex; |
351 | QemuCond cond; | |
352 | void *des; | |
d341d9f3 | 353 | uint8_t *compbuf; |
56e93d26 | 354 | int len; |
797ca154 | 355 | z_stream stream; |
56e93d26 JQ |
356 | }; |
357 | typedef struct DecompressParam DecompressParam; | |
358 | ||
359 | static CompressParam *comp_param; | |
360 | static QemuThread *compress_threads; | |
361 | /* comp_done_cond is used to wake up the migration thread when | |
362 | * one of the compression threads has finished the compression. | |
363 | * comp_done_lock is used to co-work with comp_done_cond. | |
364 | */ | |
0d9f9a5c LL |
365 | static QemuMutex comp_done_lock; |
366 | static QemuCond comp_done_cond; | |
56e93d26 JQ |
367 | /* The empty QEMUFileOps will be used by file in CompressParam */ |
368 | static const QEMUFileOps empty_ops = { }; | |
369 | ||
34ab9e97 | 370 | static QEMUFile *decomp_file; |
56e93d26 JQ |
371 | static DecompressParam *decomp_param; |
372 | static QemuThread *decompress_threads; | |
73a8912b LL |
373 | static QemuMutex decomp_done_lock; |
374 | static QemuCond decomp_done_cond; | |
56e93d26 | 375 | |
dcaf446e | 376 | static int do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
34ab9e97 | 377 | ram_addr_t offset, uint8_t *source_buf); |
56e93d26 JQ |
378 | |
379 | static void *do_data_compress(void *opaque) | |
380 | { | |
381 | CompressParam *param = opaque; | |
a7a9a88f LL |
382 | RAMBlock *block; |
383 | ram_addr_t offset; | |
56e93d26 | 384 | |
a7a9a88f | 385 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 386 | while (!param->quit) { |
a7a9a88f LL |
387 | if (param->block) { |
388 | block = param->block; | |
389 | offset = param->offset; | |
390 | param->block = NULL; | |
391 | qemu_mutex_unlock(¶m->mutex); | |
392 | ||
34ab9e97 XG |
393 | do_compress_ram_page(param->file, ¶m->stream, block, offset, |
394 | param->originbuf); | |
a7a9a88f | 395 | |
0d9f9a5c | 396 | qemu_mutex_lock(&comp_done_lock); |
a7a9a88f | 397 | param->done = true; |
0d9f9a5c LL |
398 | qemu_cond_signal(&comp_done_cond); |
399 | qemu_mutex_unlock(&comp_done_lock); | |
a7a9a88f LL |
400 | |
401 | qemu_mutex_lock(¶m->mutex); | |
402 | } else { | |
56e93d26 JQ |
403 | qemu_cond_wait(¶m->cond, ¶m->mutex); |
404 | } | |
56e93d26 | 405 | } |
a7a9a88f | 406 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
407 | |
408 | return NULL; | |
409 | } | |
410 | ||
411 | static inline void terminate_compression_threads(void) | |
412 | { | |
413 | int idx, thread_count; | |
414 | ||
415 | thread_count = migrate_compress_threads(); | |
3d0684b2 | 416 | |
56e93d26 JQ |
417 | for (idx = 0; idx < thread_count; idx++) { |
418 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 419 | comp_param[idx].quit = true; |
56e93d26 JQ |
420 | qemu_cond_signal(&comp_param[idx].cond); |
421 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
422 | } | |
423 | } | |
424 | ||
f0afa331 | 425 | static void compress_threads_save_cleanup(void) |
56e93d26 JQ |
426 | { |
427 | int i, thread_count; | |
428 | ||
429 | if (!migrate_use_compression()) { | |
430 | return; | |
431 | } | |
432 | terminate_compression_threads(); | |
433 | thread_count = migrate_compress_threads(); | |
434 | for (i = 0; i < thread_count; i++) { | |
dcaf446e XG |
435 | /* |
436 | * we use it as a indicator which shows if the thread is | |
437 | * properly init'd or not | |
438 | */ | |
439 | if (!comp_param[i].file) { | |
440 | break; | |
441 | } | |
56e93d26 | 442 | qemu_thread_join(compress_threads + i); |
56e93d26 JQ |
443 | qemu_mutex_destroy(&comp_param[i].mutex); |
444 | qemu_cond_destroy(&comp_param[i].cond); | |
dcaf446e | 445 | deflateEnd(&comp_param[i].stream); |
34ab9e97 | 446 | g_free(comp_param[i].originbuf); |
dcaf446e XG |
447 | qemu_fclose(comp_param[i].file); |
448 | comp_param[i].file = NULL; | |
56e93d26 | 449 | } |
0d9f9a5c LL |
450 | qemu_mutex_destroy(&comp_done_lock); |
451 | qemu_cond_destroy(&comp_done_cond); | |
56e93d26 JQ |
452 | g_free(compress_threads); |
453 | g_free(comp_param); | |
56e93d26 JQ |
454 | compress_threads = NULL; |
455 | comp_param = NULL; | |
56e93d26 JQ |
456 | } |
457 | ||
dcaf446e | 458 | static int compress_threads_save_setup(void) |
56e93d26 JQ |
459 | { |
460 | int i, thread_count; | |
461 | ||
462 | if (!migrate_use_compression()) { | |
dcaf446e | 463 | return 0; |
56e93d26 | 464 | } |
56e93d26 JQ |
465 | thread_count = migrate_compress_threads(); |
466 | compress_threads = g_new0(QemuThread, thread_count); | |
467 | comp_param = g_new0(CompressParam, thread_count); | |
0d9f9a5c LL |
468 | qemu_cond_init(&comp_done_cond); |
469 | qemu_mutex_init(&comp_done_lock); | |
56e93d26 | 470 | for (i = 0; i < thread_count; i++) { |
34ab9e97 XG |
471 | comp_param[i].originbuf = g_try_malloc(TARGET_PAGE_SIZE); |
472 | if (!comp_param[i].originbuf) { | |
473 | goto exit; | |
474 | } | |
475 | ||
dcaf446e XG |
476 | if (deflateInit(&comp_param[i].stream, |
477 | migrate_compress_level()) != Z_OK) { | |
34ab9e97 | 478 | g_free(comp_param[i].originbuf); |
dcaf446e XG |
479 | goto exit; |
480 | } | |
481 | ||
e110aa91 C |
482 | /* comp_param[i].file is just used as a dummy buffer to save data, |
483 | * set its ops to empty. | |
56e93d26 JQ |
484 | */ |
485 | comp_param[i].file = qemu_fopen_ops(NULL, &empty_ops); | |
486 | comp_param[i].done = true; | |
90e56fb4 | 487 | comp_param[i].quit = false; |
56e93d26 JQ |
488 | qemu_mutex_init(&comp_param[i].mutex); |
489 | qemu_cond_init(&comp_param[i].cond); | |
490 | qemu_thread_create(compress_threads + i, "compress", | |
491 | do_data_compress, comp_param + i, | |
492 | QEMU_THREAD_JOINABLE); | |
493 | } | |
dcaf446e XG |
494 | return 0; |
495 | ||
496 | exit: | |
497 | compress_threads_save_cleanup(); | |
498 | return -1; | |
56e93d26 JQ |
499 | } |
500 | ||
f986c3d2 JQ |
501 | /* Multiple fd's */ |
502 | ||
af8b7d2b JQ |
503 | #define MULTIFD_MAGIC 0x11223344U |
504 | #define MULTIFD_VERSION 1 | |
505 | ||
506 | typedef struct { | |
507 | uint32_t magic; | |
508 | uint32_t version; | |
509 | unsigned char uuid[16]; /* QemuUUID */ | |
510 | uint8_t id; | |
511 | } __attribute__((packed)) MultiFDInit_t; | |
512 | ||
8c4598f2 JQ |
513 | typedef struct { |
514 | /* this fields are not changed once the thread is created */ | |
515 | /* channel number */ | |
f986c3d2 | 516 | uint8_t id; |
8c4598f2 | 517 | /* channel thread name */ |
f986c3d2 | 518 | char *name; |
8c4598f2 | 519 | /* channel thread id */ |
f986c3d2 | 520 | QemuThread thread; |
8c4598f2 | 521 | /* communication channel */ |
60df2d4a | 522 | QIOChannel *c; |
8c4598f2 | 523 | /* sem where to wait for more work */ |
f986c3d2 | 524 | QemuSemaphore sem; |
8c4598f2 | 525 | /* this mutex protects the following parameters */ |
f986c3d2 | 526 | QemuMutex mutex; |
8c4598f2 | 527 | /* is this channel thread running */ |
66770707 | 528 | bool running; |
8c4598f2 | 529 | /* should this thread finish */ |
f986c3d2 | 530 | bool quit; |
8c4598f2 JQ |
531 | } MultiFDSendParams; |
532 | ||
533 | typedef struct { | |
534 | /* this fields are not changed once the thread is created */ | |
535 | /* channel number */ | |
536 | uint8_t id; | |
537 | /* channel thread name */ | |
538 | char *name; | |
539 | /* channel thread id */ | |
540 | QemuThread thread; | |
541 | /* communication channel */ | |
542 | QIOChannel *c; | |
543 | /* sem where to wait for more work */ | |
544 | QemuSemaphore sem; | |
545 | /* this mutex protects the following parameters */ | |
546 | QemuMutex mutex; | |
547 | /* is this channel thread running */ | |
548 | bool running; | |
549 | /* should this thread finish */ | |
550 | bool quit; | |
551 | } MultiFDRecvParams; | |
f986c3d2 | 552 | |
af8b7d2b JQ |
553 | static int multifd_send_initial_packet(MultiFDSendParams *p, Error **errp) |
554 | { | |
555 | MultiFDInit_t msg; | |
556 | int ret; | |
557 | ||
558 | msg.magic = cpu_to_be32(MULTIFD_MAGIC); | |
559 | msg.version = cpu_to_be32(MULTIFD_VERSION); | |
560 | msg.id = p->id; | |
561 | memcpy(msg.uuid, &qemu_uuid.data, sizeof(msg.uuid)); | |
562 | ||
563 | ret = qio_channel_write_all(p->c, (char *)&msg, sizeof(msg), errp); | |
564 | if (ret != 0) { | |
565 | return -1; | |
566 | } | |
567 | return 0; | |
568 | } | |
569 | ||
570 | static int multifd_recv_initial_packet(QIOChannel *c, Error **errp) | |
571 | { | |
572 | MultiFDInit_t msg; | |
573 | int ret; | |
574 | ||
575 | ret = qio_channel_read_all(c, (char *)&msg, sizeof(msg), errp); | |
576 | if (ret != 0) { | |
577 | return -1; | |
578 | } | |
579 | ||
580 | be32_to_cpus(&msg.magic); | |
581 | be32_to_cpus(&msg.version); | |
582 | ||
583 | if (msg.magic != MULTIFD_MAGIC) { | |
584 | error_setg(errp, "multifd: received packet magic %x " | |
585 | "expected %x", msg.magic, MULTIFD_MAGIC); | |
586 | return -1; | |
587 | } | |
588 | ||
589 | if (msg.version != MULTIFD_VERSION) { | |
590 | error_setg(errp, "multifd: received packet version %d " | |
591 | "expected %d", msg.version, MULTIFD_VERSION); | |
592 | return -1; | |
593 | } | |
594 | ||
595 | if (memcmp(msg.uuid, &qemu_uuid, sizeof(qemu_uuid))) { | |
596 | char *uuid = qemu_uuid_unparse_strdup(&qemu_uuid); | |
597 | char *msg_uuid = qemu_uuid_unparse_strdup((const QemuUUID *)msg.uuid); | |
598 | ||
599 | error_setg(errp, "multifd: received uuid '%s' and expected " | |
600 | "uuid '%s' for channel %hhd", msg_uuid, uuid, msg.id); | |
601 | g_free(uuid); | |
602 | g_free(msg_uuid); | |
603 | return -1; | |
604 | } | |
605 | ||
606 | if (msg.id > migrate_multifd_channels()) { | |
607 | error_setg(errp, "multifd: received channel version %d " | |
608 | "expected %d", msg.version, MULTIFD_VERSION); | |
609 | return -1; | |
610 | } | |
611 | ||
612 | return msg.id; | |
613 | } | |
614 | ||
f986c3d2 JQ |
615 | struct { |
616 | MultiFDSendParams *params; | |
617 | /* number of created threads */ | |
618 | int count; | |
619 | } *multifd_send_state; | |
620 | ||
66770707 | 621 | static void multifd_send_terminate_threads(Error *err) |
f986c3d2 JQ |
622 | { |
623 | int i; | |
624 | ||
7a169d74 JQ |
625 | if (err) { |
626 | MigrationState *s = migrate_get_current(); | |
627 | migrate_set_error(s, err); | |
628 | if (s->state == MIGRATION_STATUS_SETUP || | |
629 | s->state == MIGRATION_STATUS_PRE_SWITCHOVER || | |
630 | s->state == MIGRATION_STATUS_DEVICE || | |
631 | s->state == MIGRATION_STATUS_ACTIVE) { | |
632 | migrate_set_state(&s->state, s->state, | |
633 | MIGRATION_STATUS_FAILED); | |
634 | } | |
635 | } | |
636 | ||
66770707 | 637 | for (i = 0; i < migrate_multifd_channels(); i++) { |
f986c3d2 JQ |
638 | MultiFDSendParams *p = &multifd_send_state->params[i]; |
639 | ||
640 | qemu_mutex_lock(&p->mutex); | |
641 | p->quit = true; | |
642 | qemu_sem_post(&p->sem); | |
643 | qemu_mutex_unlock(&p->mutex); | |
644 | } | |
645 | } | |
646 | ||
647 | int multifd_save_cleanup(Error **errp) | |
648 | { | |
649 | int i; | |
650 | int ret = 0; | |
651 | ||
652 | if (!migrate_use_multifd()) { | |
653 | return 0; | |
654 | } | |
66770707 JQ |
655 | multifd_send_terminate_threads(NULL); |
656 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
f986c3d2 JQ |
657 | MultiFDSendParams *p = &multifd_send_state->params[i]; |
658 | ||
66770707 JQ |
659 | if (p->running) { |
660 | qemu_thread_join(&p->thread); | |
661 | } | |
60df2d4a JQ |
662 | socket_send_channel_destroy(p->c); |
663 | p->c = NULL; | |
f986c3d2 JQ |
664 | qemu_mutex_destroy(&p->mutex); |
665 | qemu_sem_destroy(&p->sem); | |
666 | g_free(p->name); | |
667 | p->name = NULL; | |
668 | } | |
669 | g_free(multifd_send_state->params); | |
670 | multifd_send_state->params = NULL; | |
671 | g_free(multifd_send_state); | |
672 | multifd_send_state = NULL; | |
673 | return ret; | |
674 | } | |
675 | ||
676 | static void *multifd_send_thread(void *opaque) | |
677 | { | |
678 | MultiFDSendParams *p = opaque; | |
af8b7d2b JQ |
679 | Error *local_err = NULL; |
680 | ||
681 | if (multifd_send_initial_packet(p, &local_err) < 0) { | |
682 | goto out; | |
683 | } | |
f986c3d2 JQ |
684 | |
685 | while (true) { | |
686 | qemu_mutex_lock(&p->mutex); | |
687 | if (p->quit) { | |
688 | qemu_mutex_unlock(&p->mutex); | |
689 | break; | |
690 | } | |
691 | qemu_mutex_unlock(&p->mutex); | |
692 | qemu_sem_wait(&p->sem); | |
693 | } | |
694 | ||
af8b7d2b JQ |
695 | out: |
696 | if (local_err) { | |
697 | multifd_send_terminate_threads(local_err); | |
698 | } | |
699 | ||
66770707 JQ |
700 | qemu_mutex_lock(&p->mutex); |
701 | p->running = false; | |
702 | qemu_mutex_unlock(&p->mutex); | |
703 | ||
f986c3d2 JQ |
704 | return NULL; |
705 | } | |
706 | ||
60df2d4a JQ |
707 | static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque) |
708 | { | |
709 | MultiFDSendParams *p = opaque; | |
710 | QIOChannel *sioc = QIO_CHANNEL(qio_task_get_source(task)); | |
711 | Error *local_err = NULL; | |
712 | ||
713 | if (qio_task_propagate_error(task, &local_err)) { | |
714 | if (multifd_save_cleanup(&local_err) != 0) { | |
715 | migrate_set_error(migrate_get_current(), local_err); | |
716 | } | |
717 | } else { | |
718 | p->c = QIO_CHANNEL(sioc); | |
719 | qio_channel_set_delay(p->c, false); | |
720 | p->running = true; | |
721 | qemu_thread_create(&p->thread, p->name, multifd_send_thread, p, | |
722 | QEMU_THREAD_JOINABLE); | |
723 | ||
724 | atomic_inc(&multifd_send_state->count); | |
725 | } | |
726 | } | |
727 | ||
f986c3d2 JQ |
728 | int multifd_save_setup(void) |
729 | { | |
730 | int thread_count; | |
731 | uint8_t i; | |
732 | ||
733 | if (!migrate_use_multifd()) { | |
734 | return 0; | |
735 | } | |
736 | thread_count = migrate_multifd_channels(); | |
737 | multifd_send_state = g_malloc0(sizeof(*multifd_send_state)); | |
738 | multifd_send_state->params = g_new0(MultiFDSendParams, thread_count); | |
66770707 | 739 | atomic_set(&multifd_send_state->count, 0); |
f986c3d2 JQ |
740 | for (i = 0; i < thread_count; i++) { |
741 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
742 | ||
743 | qemu_mutex_init(&p->mutex); | |
744 | qemu_sem_init(&p->sem, 0); | |
745 | p->quit = false; | |
746 | p->id = i; | |
747 | p->name = g_strdup_printf("multifdsend_%d", i); | |
60df2d4a | 748 | socket_send_channel_create(multifd_new_send_channel_async, p); |
f986c3d2 JQ |
749 | } |
750 | return 0; | |
751 | } | |
752 | ||
f986c3d2 JQ |
753 | struct { |
754 | MultiFDRecvParams *params; | |
755 | /* number of created threads */ | |
756 | int count; | |
757 | } *multifd_recv_state; | |
758 | ||
66770707 | 759 | static void multifd_recv_terminate_threads(Error *err) |
f986c3d2 JQ |
760 | { |
761 | int i; | |
762 | ||
7a169d74 JQ |
763 | if (err) { |
764 | MigrationState *s = migrate_get_current(); | |
765 | migrate_set_error(s, err); | |
766 | if (s->state == MIGRATION_STATUS_SETUP || | |
767 | s->state == MIGRATION_STATUS_ACTIVE) { | |
768 | migrate_set_state(&s->state, s->state, | |
769 | MIGRATION_STATUS_FAILED); | |
770 | } | |
771 | } | |
772 | ||
66770707 | 773 | for (i = 0; i < migrate_multifd_channels(); i++) { |
f986c3d2 JQ |
774 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; |
775 | ||
776 | qemu_mutex_lock(&p->mutex); | |
777 | p->quit = true; | |
778 | qemu_sem_post(&p->sem); | |
779 | qemu_mutex_unlock(&p->mutex); | |
780 | } | |
781 | } | |
782 | ||
783 | int multifd_load_cleanup(Error **errp) | |
784 | { | |
785 | int i; | |
786 | int ret = 0; | |
787 | ||
788 | if (!migrate_use_multifd()) { | |
789 | return 0; | |
790 | } | |
66770707 JQ |
791 | multifd_recv_terminate_threads(NULL); |
792 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
f986c3d2 JQ |
793 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; |
794 | ||
66770707 JQ |
795 | if (p->running) { |
796 | qemu_thread_join(&p->thread); | |
797 | } | |
60df2d4a JQ |
798 | object_unref(OBJECT(p->c)); |
799 | p->c = NULL; | |
f986c3d2 JQ |
800 | qemu_mutex_destroy(&p->mutex); |
801 | qemu_sem_destroy(&p->sem); | |
802 | g_free(p->name); | |
803 | p->name = NULL; | |
804 | } | |
805 | g_free(multifd_recv_state->params); | |
806 | multifd_recv_state->params = NULL; | |
807 | g_free(multifd_recv_state); | |
808 | multifd_recv_state = NULL; | |
809 | ||
810 | return ret; | |
811 | } | |
812 | ||
813 | static void *multifd_recv_thread(void *opaque) | |
814 | { | |
815 | MultiFDRecvParams *p = opaque; | |
816 | ||
817 | while (true) { | |
818 | qemu_mutex_lock(&p->mutex); | |
819 | if (p->quit) { | |
820 | qemu_mutex_unlock(&p->mutex); | |
821 | break; | |
822 | } | |
823 | qemu_mutex_unlock(&p->mutex); | |
824 | qemu_sem_wait(&p->sem); | |
825 | } | |
826 | ||
66770707 JQ |
827 | qemu_mutex_lock(&p->mutex); |
828 | p->running = false; | |
829 | qemu_mutex_unlock(&p->mutex); | |
830 | ||
f986c3d2 JQ |
831 | return NULL; |
832 | } | |
833 | ||
834 | int multifd_load_setup(void) | |
835 | { | |
836 | int thread_count; | |
837 | uint8_t i; | |
838 | ||
839 | if (!migrate_use_multifd()) { | |
840 | return 0; | |
841 | } | |
842 | thread_count = migrate_multifd_channels(); | |
843 | multifd_recv_state = g_malloc0(sizeof(*multifd_recv_state)); | |
844 | multifd_recv_state->params = g_new0(MultiFDRecvParams, thread_count); | |
66770707 | 845 | atomic_set(&multifd_recv_state->count, 0); |
f986c3d2 JQ |
846 | for (i = 0; i < thread_count; i++) { |
847 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
848 | ||
849 | qemu_mutex_init(&p->mutex); | |
850 | qemu_sem_init(&p->sem, 0); | |
851 | p->quit = false; | |
852 | p->id = i; | |
853 | p->name = g_strdup_printf("multifdrecv_%d", i); | |
f986c3d2 JQ |
854 | } |
855 | return 0; | |
856 | } | |
857 | ||
62c1e0ca JQ |
858 | bool multifd_recv_all_channels_created(void) |
859 | { | |
860 | int thread_count = migrate_multifd_channels(); | |
861 | ||
862 | if (!migrate_use_multifd()) { | |
863 | return true; | |
864 | } | |
865 | ||
866 | return thread_count == atomic_read(&multifd_recv_state->count); | |
867 | } | |
868 | ||
71bb07db JQ |
869 | void multifd_recv_new_channel(QIOChannel *ioc) |
870 | { | |
60df2d4a | 871 | MultiFDRecvParams *p; |
af8b7d2b JQ |
872 | Error *local_err = NULL; |
873 | int id; | |
60df2d4a | 874 | |
af8b7d2b JQ |
875 | id = multifd_recv_initial_packet(ioc, &local_err); |
876 | if (id < 0) { | |
877 | multifd_recv_terminate_threads(local_err); | |
878 | return; | |
879 | } | |
880 | ||
881 | p = &multifd_recv_state->params[id]; | |
882 | if (p->c != NULL) { | |
883 | error_setg(&local_err, "multifd: received id '%d' already setup'", | |
884 | id); | |
885 | multifd_recv_terminate_threads(local_err); | |
886 | return; | |
887 | } | |
60df2d4a JQ |
888 | p->c = ioc; |
889 | object_ref(OBJECT(ioc)); | |
890 | ||
891 | p->running = true; | |
892 | qemu_thread_create(&p->thread, p->name, multifd_recv_thread, p, | |
893 | QEMU_THREAD_JOINABLE); | |
894 | atomic_inc(&multifd_recv_state->count); | |
36c2f8be JQ |
895 | if (multifd_recv_state->count == migrate_multifd_channels()) { |
896 | migration_incoming_process(); | |
897 | } | |
71bb07db JQ |
898 | } |
899 | ||
56e93d26 | 900 | /** |
3d0684b2 | 901 | * save_page_header: write page header to wire |
56e93d26 JQ |
902 | * |
903 | * If this is the 1st block, it also writes the block identification | |
904 | * | |
3d0684b2 | 905 | * Returns the number of bytes written |
56e93d26 JQ |
906 | * |
907 | * @f: QEMUFile where to send the data | |
908 | * @block: block that contains the page we want to send | |
909 | * @offset: offset inside the block for the page | |
910 | * in the lower bits, it contains flags | |
911 | */ | |
2bf3aa85 JQ |
912 | static size_t save_page_header(RAMState *rs, QEMUFile *f, RAMBlock *block, |
913 | ram_addr_t offset) | |
56e93d26 | 914 | { |
9f5f380b | 915 | size_t size, len; |
56e93d26 | 916 | |
24795694 JQ |
917 | if (block == rs->last_sent_block) { |
918 | offset |= RAM_SAVE_FLAG_CONTINUE; | |
919 | } | |
2bf3aa85 | 920 | qemu_put_be64(f, offset); |
56e93d26 JQ |
921 | size = 8; |
922 | ||
923 | if (!(offset & RAM_SAVE_FLAG_CONTINUE)) { | |
9f5f380b | 924 | len = strlen(block->idstr); |
2bf3aa85 JQ |
925 | qemu_put_byte(f, len); |
926 | qemu_put_buffer(f, (uint8_t *)block->idstr, len); | |
9f5f380b | 927 | size += 1 + len; |
24795694 | 928 | rs->last_sent_block = block; |
56e93d26 JQ |
929 | } |
930 | return size; | |
931 | } | |
932 | ||
3d0684b2 JQ |
933 | /** |
934 | * mig_throttle_guest_down: throotle down the guest | |
935 | * | |
936 | * Reduce amount of guest cpu execution to hopefully slow down memory | |
937 | * writes. If guest dirty memory rate is reduced below the rate at | |
938 | * which we can transfer pages to the destination then we should be | |
939 | * able to complete migration. Some workloads dirty memory way too | |
940 | * fast and will not effectively converge, even with auto-converge. | |
070afca2 JH |
941 | */ |
942 | static void mig_throttle_guest_down(void) | |
943 | { | |
944 | MigrationState *s = migrate_get_current(); | |
2594f56d DB |
945 | uint64_t pct_initial = s->parameters.cpu_throttle_initial; |
946 | uint64_t pct_icrement = s->parameters.cpu_throttle_increment; | |
070afca2 JH |
947 | |
948 | /* We have not started throttling yet. Let's start it. */ | |
949 | if (!cpu_throttle_active()) { | |
950 | cpu_throttle_set(pct_initial); | |
951 | } else { | |
952 | /* Throttling already on, just increase the rate */ | |
953 | cpu_throttle_set(cpu_throttle_get_percentage() + pct_icrement); | |
954 | } | |
955 | } | |
956 | ||
3d0684b2 JQ |
957 | /** |
958 | * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache | |
959 | * | |
6f37bb8b | 960 | * @rs: current RAM state |
3d0684b2 JQ |
961 | * @current_addr: address for the zero page |
962 | * | |
963 | * Update the xbzrle cache to reflect a page that's been sent as all 0. | |
56e93d26 JQ |
964 | * The important thing is that a stale (not-yet-0'd) page be replaced |
965 | * by the new data. | |
966 | * As a bonus, if the page wasn't in the cache it gets added so that | |
3d0684b2 | 967 | * when a small write is made into the 0'd page it gets XBZRLE sent. |
56e93d26 | 968 | */ |
6f37bb8b | 969 | static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr) |
56e93d26 | 970 | { |
6f37bb8b | 971 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { |
56e93d26 JQ |
972 | return; |
973 | } | |
974 | ||
975 | /* We don't care if this fails to allocate a new cache page | |
976 | * as long as it updated an old one */ | |
c00e0928 | 977 | cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page, |
9360447d | 978 | ram_counters.dirty_sync_count); |
56e93d26 JQ |
979 | } |
980 | ||
981 | #define ENCODING_FLAG_XBZRLE 0x1 | |
982 | ||
983 | /** | |
984 | * save_xbzrle_page: compress and send current page | |
985 | * | |
986 | * Returns: 1 means that we wrote the page | |
987 | * 0 means that page is identical to the one already sent | |
988 | * -1 means that xbzrle would be longer than normal | |
989 | * | |
5a987738 | 990 | * @rs: current RAM state |
3d0684b2 JQ |
991 | * @current_data: pointer to the address of the page contents |
992 | * @current_addr: addr of the page | |
56e93d26 JQ |
993 | * @block: block that contains the page we want to send |
994 | * @offset: offset inside the block for the page | |
995 | * @last_stage: if we are at the completion stage | |
56e93d26 | 996 | */ |
204b88b8 | 997 | static int save_xbzrle_page(RAMState *rs, uint8_t **current_data, |
56e93d26 | 998 | ram_addr_t current_addr, RAMBlock *block, |
072c2511 | 999 | ram_addr_t offset, bool last_stage) |
56e93d26 JQ |
1000 | { |
1001 | int encoded_len = 0, bytes_xbzrle; | |
1002 | uint8_t *prev_cached_page; | |
1003 | ||
9360447d JQ |
1004 | if (!cache_is_cached(XBZRLE.cache, current_addr, |
1005 | ram_counters.dirty_sync_count)) { | |
1006 | xbzrle_counters.cache_miss++; | |
56e93d26 JQ |
1007 | if (!last_stage) { |
1008 | if (cache_insert(XBZRLE.cache, current_addr, *current_data, | |
9360447d | 1009 | ram_counters.dirty_sync_count) == -1) { |
56e93d26 JQ |
1010 | return -1; |
1011 | } else { | |
1012 | /* update *current_data when the page has been | |
1013 | inserted into cache */ | |
1014 | *current_data = get_cached_data(XBZRLE.cache, current_addr); | |
1015 | } | |
1016 | } | |
1017 | return -1; | |
1018 | } | |
1019 | ||
1020 | prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); | |
1021 | ||
1022 | /* save current buffer into memory */ | |
1023 | memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); | |
1024 | ||
1025 | /* XBZRLE encoding (if there is no overflow) */ | |
1026 | encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, | |
1027 | TARGET_PAGE_SIZE, XBZRLE.encoded_buf, | |
1028 | TARGET_PAGE_SIZE); | |
1029 | if (encoded_len == 0) { | |
55c4446b | 1030 | trace_save_xbzrle_page_skipping(); |
56e93d26 JQ |
1031 | return 0; |
1032 | } else if (encoded_len == -1) { | |
55c4446b | 1033 | trace_save_xbzrle_page_overflow(); |
9360447d | 1034 | xbzrle_counters.overflow++; |
56e93d26 JQ |
1035 | /* update data in the cache */ |
1036 | if (!last_stage) { | |
1037 | memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE); | |
1038 | *current_data = prev_cached_page; | |
1039 | } | |
1040 | return -1; | |
1041 | } | |
1042 | ||
1043 | /* we need to update the data in the cache, in order to get the same data */ | |
1044 | if (!last_stage) { | |
1045 | memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); | |
1046 | } | |
1047 | ||
1048 | /* Send XBZRLE based compressed page */ | |
2bf3aa85 | 1049 | bytes_xbzrle = save_page_header(rs, rs->f, block, |
204b88b8 JQ |
1050 | offset | RAM_SAVE_FLAG_XBZRLE); |
1051 | qemu_put_byte(rs->f, ENCODING_FLAG_XBZRLE); | |
1052 | qemu_put_be16(rs->f, encoded_len); | |
1053 | qemu_put_buffer(rs->f, XBZRLE.encoded_buf, encoded_len); | |
56e93d26 | 1054 | bytes_xbzrle += encoded_len + 1 + 2; |
9360447d JQ |
1055 | xbzrle_counters.pages++; |
1056 | xbzrle_counters.bytes += bytes_xbzrle; | |
1057 | ram_counters.transferred += bytes_xbzrle; | |
56e93d26 JQ |
1058 | |
1059 | return 1; | |
1060 | } | |
1061 | ||
3d0684b2 JQ |
1062 | /** |
1063 | * migration_bitmap_find_dirty: find the next dirty page from start | |
f3f491fc | 1064 | * |
3d0684b2 JQ |
1065 | * Called with rcu_read_lock() to protect migration_bitmap |
1066 | * | |
1067 | * Returns the byte offset within memory region of the start of a dirty page | |
1068 | * | |
6f37bb8b | 1069 | * @rs: current RAM state |
3d0684b2 | 1070 | * @rb: RAMBlock where to search for dirty pages |
a935e30f | 1071 | * @start: page where we start the search |
f3f491fc | 1072 | */ |
56e93d26 | 1073 | static inline |
a935e30f | 1074 | unsigned long migration_bitmap_find_dirty(RAMState *rs, RAMBlock *rb, |
f20e2865 | 1075 | unsigned long start) |
56e93d26 | 1076 | { |
6b6712ef JQ |
1077 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; |
1078 | unsigned long *bitmap = rb->bmap; | |
56e93d26 JQ |
1079 | unsigned long next; |
1080 | ||
6b6712ef JQ |
1081 | if (rs->ram_bulk_stage && start > 0) { |
1082 | next = start + 1; | |
56e93d26 | 1083 | } else { |
6b6712ef | 1084 | next = find_next_bit(bitmap, size, start); |
56e93d26 JQ |
1085 | } |
1086 | ||
6b6712ef | 1087 | return next; |
56e93d26 JQ |
1088 | } |
1089 | ||
06b10688 | 1090 | static inline bool migration_bitmap_clear_dirty(RAMState *rs, |
f20e2865 JQ |
1091 | RAMBlock *rb, |
1092 | unsigned long page) | |
a82d593b DDAG |
1093 | { |
1094 | bool ret; | |
a82d593b | 1095 | |
6b6712ef | 1096 | ret = test_and_clear_bit(page, rb->bmap); |
a82d593b DDAG |
1097 | |
1098 | if (ret) { | |
0d8ec885 | 1099 | rs->migration_dirty_pages--; |
a82d593b DDAG |
1100 | } |
1101 | return ret; | |
1102 | } | |
1103 | ||
15440dd5 JQ |
1104 | static void migration_bitmap_sync_range(RAMState *rs, RAMBlock *rb, |
1105 | ram_addr_t start, ram_addr_t length) | |
56e93d26 | 1106 | { |
0d8ec885 | 1107 | rs->migration_dirty_pages += |
6b6712ef | 1108 | cpu_physical_memory_sync_dirty_bitmap(rb, start, length, |
0d8ec885 | 1109 | &rs->num_dirty_pages_period); |
56e93d26 JQ |
1110 | } |
1111 | ||
3d0684b2 JQ |
1112 | /** |
1113 | * ram_pagesize_summary: calculate all the pagesizes of a VM | |
1114 | * | |
1115 | * Returns a summary bitmap of the page sizes of all RAMBlocks | |
1116 | * | |
1117 | * For VMs with just normal pages this is equivalent to the host page | |
1118 | * size. If it's got some huge pages then it's the OR of all the | |
1119 | * different page sizes. | |
e8ca1db2 DDAG |
1120 | */ |
1121 | uint64_t ram_pagesize_summary(void) | |
1122 | { | |
1123 | RAMBlock *block; | |
1124 | uint64_t summary = 0; | |
1125 | ||
99e15582 | 1126 | RAMBLOCK_FOREACH(block) { |
e8ca1db2 DDAG |
1127 | summary |= block->page_size; |
1128 | } | |
1129 | ||
1130 | return summary; | |
1131 | } | |
1132 | ||
8d820d6f | 1133 | static void migration_bitmap_sync(RAMState *rs) |
56e93d26 JQ |
1134 | { |
1135 | RAMBlock *block; | |
56e93d26 | 1136 | int64_t end_time; |
c4bdf0cf | 1137 | uint64_t bytes_xfer_now; |
56e93d26 | 1138 | |
9360447d | 1139 | ram_counters.dirty_sync_count++; |
56e93d26 | 1140 | |
f664da80 JQ |
1141 | if (!rs->time_last_bitmap_sync) { |
1142 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56e93d26 JQ |
1143 | } |
1144 | ||
1145 | trace_migration_bitmap_sync_start(); | |
9c1f8f44 | 1146 | memory_global_dirty_log_sync(); |
56e93d26 | 1147 | |
108cfae0 | 1148 | qemu_mutex_lock(&rs->bitmap_mutex); |
56e93d26 | 1149 | rcu_read_lock(); |
99e15582 | 1150 | RAMBLOCK_FOREACH(block) { |
15440dd5 | 1151 | migration_bitmap_sync_range(rs, block, 0, block->used_length); |
56e93d26 JQ |
1152 | } |
1153 | rcu_read_unlock(); | |
108cfae0 | 1154 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 1155 | |
a66cd90c | 1156 | trace_migration_bitmap_sync_end(rs->num_dirty_pages_period); |
1ffb5dfd | 1157 | |
56e93d26 JQ |
1158 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1159 | ||
1160 | /* more than 1 second = 1000 millisecons */ | |
f664da80 | 1161 | if (end_time > rs->time_last_bitmap_sync + 1000) { |
d693c6f1 | 1162 | /* calculate period counters */ |
9360447d | 1163 | ram_counters.dirty_pages_rate = rs->num_dirty_pages_period * 1000 |
d693c6f1 | 1164 | / (end_time - rs->time_last_bitmap_sync); |
9360447d | 1165 | bytes_xfer_now = ram_counters.transferred; |
d693c6f1 | 1166 | |
9ac78b61 PL |
1167 | /* During block migration the auto-converge logic incorrectly detects |
1168 | * that ram migration makes no progress. Avoid this by disabling the | |
1169 | * throttling logic during the bulk phase of block migration. */ | |
1170 | if (migrate_auto_converge() && !blk_mig_bulk_active()) { | |
56e93d26 JQ |
1171 | /* The following detection logic can be refined later. For now: |
1172 | Check to see if the dirtied bytes is 50% more than the approx. | |
1173 | amount of bytes that just got transferred since the last time we | |
070afca2 JH |
1174 | were in this routine. If that happens twice, start or increase |
1175 | throttling */ | |
070afca2 | 1176 | |
d693c6f1 | 1177 | if ((rs->num_dirty_pages_period * TARGET_PAGE_SIZE > |
eac74159 | 1178 | (bytes_xfer_now - rs->bytes_xfer_prev) / 2) && |
b4a3c64b | 1179 | (++rs->dirty_rate_high_cnt >= 2)) { |
56e93d26 | 1180 | trace_migration_throttle(); |
8d820d6f | 1181 | rs->dirty_rate_high_cnt = 0; |
070afca2 | 1182 | mig_throttle_guest_down(); |
d693c6f1 | 1183 | } |
56e93d26 | 1184 | } |
070afca2 | 1185 | |
56e93d26 | 1186 | if (migrate_use_xbzrle()) { |
23b28c3c | 1187 | if (rs->iterations_prev != rs->iterations) { |
9360447d JQ |
1188 | xbzrle_counters.cache_miss_rate = |
1189 | (double)(xbzrle_counters.cache_miss - | |
b5833fde | 1190 | rs->xbzrle_cache_miss_prev) / |
23b28c3c | 1191 | (rs->iterations - rs->iterations_prev); |
56e93d26 | 1192 | } |
23b28c3c | 1193 | rs->iterations_prev = rs->iterations; |
9360447d | 1194 | rs->xbzrle_cache_miss_prev = xbzrle_counters.cache_miss; |
56e93d26 | 1195 | } |
d693c6f1 FF |
1196 | |
1197 | /* reset period counters */ | |
f664da80 | 1198 | rs->time_last_bitmap_sync = end_time; |
a66cd90c | 1199 | rs->num_dirty_pages_period = 0; |
d2a4d85a | 1200 | rs->bytes_xfer_prev = bytes_xfer_now; |
56e93d26 | 1201 | } |
4addcd4f | 1202 | if (migrate_use_events()) { |
9360447d | 1203 | qapi_event_send_migration_pass(ram_counters.dirty_sync_count, NULL); |
4addcd4f | 1204 | } |
56e93d26 JQ |
1205 | } |
1206 | ||
1207 | /** | |
3d0684b2 | 1208 | * save_zero_page: send the zero page to the stream |
56e93d26 | 1209 | * |
3d0684b2 | 1210 | * Returns the number of pages written. |
56e93d26 | 1211 | * |
f7ccd61b | 1212 | * @rs: current RAM state |
56e93d26 JQ |
1213 | * @block: block that contains the page we want to send |
1214 | * @offset: offset inside the block for the page | |
56e93d26 | 1215 | */ |
7faccdc3 | 1216 | static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset) |
56e93d26 | 1217 | { |
7faccdc3 | 1218 | uint8_t *p = block->host + offset; |
56e93d26 JQ |
1219 | int pages = -1; |
1220 | ||
1221 | if (is_zero_range(p, TARGET_PAGE_SIZE)) { | |
9360447d JQ |
1222 | ram_counters.duplicate++; |
1223 | ram_counters.transferred += | |
bb890ed5 | 1224 | save_page_header(rs, rs->f, block, offset | RAM_SAVE_FLAG_ZERO); |
ce25d337 | 1225 | qemu_put_byte(rs->f, 0); |
9360447d | 1226 | ram_counters.transferred += 1; |
56e93d26 JQ |
1227 | pages = 1; |
1228 | } | |
1229 | ||
1230 | return pages; | |
1231 | } | |
1232 | ||
5727309d | 1233 | static void ram_release_pages(const char *rbname, uint64_t offset, int pages) |
53f09a10 | 1234 | { |
5727309d | 1235 | if (!migrate_release_ram() || !migration_in_postcopy()) { |
53f09a10 PB |
1236 | return; |
1237 | } | |
1238 | ||
aaa2064c | 1239 | ram_discard_range(rbname, offset, pages << TARGET_PAGE_BITS); |
53f09a10 PB |
1240 | } |
1241 | ||
059ff0fb XG |
1242 | /* |
1243 | * @pages: the number of pages written by the control path, | |
1244 | * < 0 - error | |
1245 | * > 0 - number of pages written | |
1246 | * | |
1247 | * Return true if the pages has been saved, otherwise false is returned. | |
1248 | */ | |
1249 | static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, | |
1250 | int *pages) | |
1251 | { | |
1252 | uint64_t bytes_xmit = 0; | |
1253 | int ret; | |
1254 | ||
1255 | *pages = -1; | |
1256 | ret = ram_control_save_page(rs->f, block->offset, offset, TARGET_PAGE_SIZE, | |
1257 | &bytes_xmit); | |
1258 | if (ret == RAM_SAVE_CONTROL_NOT_SUPP) { | |
1259 | return false; | |
1260 | } | |
1261 | ||
1262 | if (bytes_xmit) { | |
1263 | ram_counters.transferred += bytes_xmit; | |
1264 | *pages = 1; | |
1265 | } | |
1266 | ||
1267 | if (ret == RAM_SAVE_CONTROL_DELAYED) { | |
1268 | return true; | |
1269 | } | |
1270 | ||
1271 | if (bytes_xmit > 0) { | |
1272 | ram_counters.normal++; | |
1273 | } else if (bytes_xmit == 0) { | |
1274 | ram_counters.duplicate++; | |
1275 | } | |
1276 | ||
1277 | return true; | |
1278 | } | |
1279 | ||
65dacaa0 XG |
1280 | /* |
1281 | * directly send the page to the stream | |
1282 | * | |
1283 | * Returns the number of pages written. | |
1284 | * | |
1285 | * @rs: current RAM state | |
1286 | * @block: block that contains the page we want to send | |
1287 | * @offset: offset inside the block for the page | |
1288 | * @buf: the page to be sent | |
1289 | * @async: send to page asyncly | |
1290 | */ | |
1291 | static int save_normal_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, | |
1292 | uint8_t *buf, bool async) | |
1293 | { | |
1294 | ram_counters.transferred += save_page_header(rs, rs->f, block, | |
1295 | offset | RAM_SAVE_FLAG_PAGE); | |
1296 | if (async) { | |
1297 | qemu_put_buffer_async(rs->f, buf, TARGET_PAGE_SIZE, | |
1298 | migrate_release_ram() & | |
1299 | migration_in_postcopy()); | |
1300 | } else { | |
1301 | qemu_put_buffer(rs->f, buf, TARGET_PAGE_SIZE); | |
1302 | } | |
1303 | ram_counters.transferred += TARGET_PAGE_SIZE; | |
1304 | ram_counters.normal++; | |
1305 | return 1; | |
1306 | } | |
1307 | ||
56e93d26 | 1308 | /** |
3d0684b2 | 1309 | * ram_save_page: send the given page to the stream |
56e93d26 | 1310 | * |
3d0684b2 | 1311 | * Returns the number of pages written. |
3fd3c4b3 DDAG |
1312 | * < 0 - error |
1313 | * >=0 - Number of pages written - this might legally be 0 | |
1314 | * if xbzrle noticed the page was the same. | |
56e93d26 | 1315 | * |
6f37bb8b | 1316 | * @rs: current RAM state |
56e93d26 JQ |
1317 | * @block: block that contains the page we want to send |
1318 | * @offset: offset inside the block for the page | |
1319 | * @last_stage: if we are at the completion stage | |
56e93d26 | 1320 | */ |
a0a8aa14 | 1321 | static int ram_save_page(RAMState *rs, PageSearchStatus *pss, bool last_stage) |
56e93d26 JQ |
1322 | { |
1323 | int pages = -1; | |
56e93d26 | 1324 | uint8_t *p; |
56e93d26 | 1325 | bool send_async = true; |
a08f6890 | 1326 | RAMBlock *block = pss->block; |
a935e30f | 1327 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
059ff0fb | 1328 | ram_addr_t current_addr = block->offset + offset; |
56e93d26 | 1329 | |
2f68e399 | 1330 | p = block->host + offset; |
1db9d8e5 | 1331 | trace_ram_save_page(block->idstr, (uint64_t)offset, p); |
56e93d26 | 1332 | |
56e93d26 | 1333 | XBZRLE_cache_lock(); |
d7400a34 XG |
1334 | if (!rs->ram_bulk_stage && !migration_in_postcopy() && |
1335 | migrate_use_xbzrle()) { | |
059ff0fb XG |
1336 | pages = save_xbzrle_page(rs, &p, current_addr, block, |
1337 | offset, last_stage); | |
1338 | if (!last_stage) { | |
1339 | /* Can't send this cached data async, since the cache page | |
1340 | * might get updated before it gets to the wire | |
56e93d26 | 1341 | */ |
059ff0fb | 1342 | send_async = false; |
56e93d26 JQ |
1343 | } |
1344 | } | |
1345 | ||
1346 | /* XBZRLE overflow or normal page */ | |
1347 | if (pages == -1) { | |
65dacaa0 | 1348 | pages = save_normal_page(rs, block, offset, p, send_async); |
56e93d26 JQ |
1349 | } |
1350 | ||
1351 | XBZRLE_cache_unlock(); | |
1352 | ||
1353 | return pages; | |
1354 | } | |
1355 | ||
dcaf446e | 1356 | static int do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
34ab9e97 | 1357 | ram_addr_t offset, uint8_t *source_buf) |
56e93d26 | 1358 | { |
53518d94 | 1359 | RAMState *rs = ram_state; |
56e93d26 | 1360 | int bytes_sent, blen; |
a7a9a88f | 1361 | uint8_t *p = block->host + (offset & TARGET_PAGE_MASK); |
56e93d26 | 1362 | |
2bf3aa85 | 1363 | bytes_sent = save_page_header(rs, f, block, offset | |
56e93d26 | 1364 | RAM_SAVE_FLAG_COMPRESS_PAGE); |
34ab9e97 XG |
1365 | |
1366 | /* | |
1367 | * copy it to a internal buffer to avoid it being modified by VM | |
1368 | * so that we can catch up the error during compression and | |
1369 | * decompression | |
1370 | */ | |
1371 | memcpy(source_buf, p, TARGET_PAGE_SIZE); | |
1372 | blen = qemu_put_compression_data(f, stream, source_buf, TARGET_PAGE_SIZE); | |
b3be2896 LL |
1373 | if (blen < 0) { |
1374 | bytes_sent = 0; | |
1375 | qemu_file_set_error(migrate_get_current()->to_dst_file, blen); | |
1376 | error_report("compressed data failed!"); | |
1377 | } else { | |
1378 | bytes_sent += blen; | |
5727309d | 1379 | ram_release_pages(block->idstr, offset & TARGET_PAGE_MASK, 1); |
b3be2896 | 1380 | } |
56e93d26 JQ |
1381 | |
1382 | return bytes_sent; | |
1383 | } | |
1384 | ||
ce25d337 | 1385 | static void flush_compressed_data(RAMState *rs) |
56e93d26 JQ |
1386 | { |
1387 | int idx, len, thread_count; | |
1388 | ||
1389 | if (!migrate_use_compression()) { | |
1390 | return; | |
1391 | } | |
1392 | thread_count = migrate_compress_threads(); | |
a7a9a88f | 1393 | |
0d9f9a5c | 1394 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 | 1395 | for (idx = 0; idx < thread_count; idx++) { |
a7a9a88f | 1396 | while (!comp_param[idx].done) { |
0d9f9a5c | 1397 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 | 1398 | } |
a7a9a88f | 1399 | } |
0d9f9a5c | 1400 | qemu_mutex_unlock(&comp_done_lock); |
a7a9a88f LL |
1401 | |
1402 | for (idx = 0; idx < thread_count; idx++) { | |
1403 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 1404 | if (!comp_param[idx].quit) { |
ce25d337 | 1405 | len = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
9360447d | 1406 | ram_counters.transferred += len; |
56e93d26 | 1407 | } |
a7a9a88f | 1408 | qemu_mutex_unlock(&comp_param[idx].mutex); |
56e93d26 JQ |
1409 | } |
1410 | } | |
1411 | ||
1412 | static inline void set_compress_params(CompressParam *param, RAMBlock *block, | |
1413 | ram_addr_t offset) | |
1414 | { | |
1415 | param->block = block; | |
1416 | param->offset = offset; | |
1417 | } | |
1418 | ||
ce25d337 JQ |
1419 | static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block, |
1420 | ram_addr_t offset) | |
56e93d26 JQ |
1421 | { |
1422 | int idx, thread_count, bytes_xmit = -1, pages = -1; | |
1423 | ||
1424 | thread_count = migrate_compress_threads(); | |
0d9f9a5c | 1425 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 JQ |
1426 | while (true) { |
1427 | for (idx = 0; idx < thread_count; idx++) { | |
1428 | if (comp_param[idx].done) { | |
a7a9a88f | 1429 | comp_param[idx].done = false; |
ce25d337 | 1430 | bytes_xmit = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
a7a9a88f | 1431 | qemu_mutex_lock(&comp_param[idx].mutex); |
56e93d26 | 1432 | set_compress_params(&comp_param[idx], block, offset); |
a7a9a88f LL |
1433 | qemu_cond_signal(&comp_param[idx].cond); |
1434 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
56e93d26 | 1435 | pages = 1; |
9360447d JQ |
1436 | ram_counters.normal++; |
1437 | ram_counters.transferred += bytes_xmit; | |
56e93d26 JQ |
1438 | break; |
1439 | } | |
1440 | } | |
1441 | if (pages > 0) { | |
1442 | break; | |
1443 | } else { | |
0d9f9a5c | 1444 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 JQ |
1445 | } |
1446 | } | |
0d9f9a5c | 1447 | qemu_mutex_unlock(&comp_done_lock); |
56e93d26 JQ |
1448 | |
1449 | return pages; | |
1450 | } | |
1451 | ||
3d0684b2 JQ |
1452 | /** |
1453 | * find_dirty_block: find the next dirty page and update any state | |
1454 | * associated with the search process. | |
b9e60928 | 1455 | * |
3d0684b2 | 1456 | * Returns if a page is found |
b9e60928 | 1457 | * |
6f37bb8b | 1458 | * @rs: current RAM state |
3d0684b2 JQ |
1459 | * @pss: data about the state of the current dirty page scan |
1460 | * @again: set to false if the search has scanned the whole of RAM | |
b9e60928 | 1461 | */ |
f20e2865 | 1462 | static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again) |
b9e60928 | 1463 | { |
f20e2865 | 1464 | pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page); |
6f37bb8b | 1465 | if (pss->complete_round && pss->block == rs->last_seen_block && |
a935e30f | 1466 | pss->page >= rs->last_page) { |
b9e60928 DDAG |
1467 | /* |
1468 | * We've been once around the RAM and haven't found anything. | |
1469 | * Give up. | |
1470 | */ | |
1471 | *again = false; | |
1472 | return false; | |
1473 | } | |
a935e30f | 1474 | if ((pss->page << TARGET_PAGE_BITS) >= pss->block->used_length) { |
b9e60928 | 1475 | /* Didn't find anything in this RAM Block */ |
a935e30f | 1476 | pss->page = 0; |
b9e60928 DDAG |
1477 | pss->block = QLIST_NEXT_RCU(pss->block, next); |
1478 | if (!pss->block) { | |
1479 | /* Hit the end of the list */ | |
1480 | pss->block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1481 | /* Flag that we've looped */ | |
1482 | pss->complete_round = true; | |
6f37bb8b | 1483 | rs->ram_bulk_stage = false; |
b9e60928 DDAG |
1484 | if (migrate_use_xbzrle()) { |
1485 | /* If xbzrle is on, stop using the data compression at this | |
1486 | * point. In theory, xbzrle can do better than compression. | |
1487 | */ | |
ce25d337 | 1488 | flush_compressed_data(rs); |
b9e60928 DDAG |
1489 | } |
1490 | } | |
1491 | /* Didn't find anything this time, but try again on the new block */ | |
1492 | *again = true; | |
1493 | return false; | |
1494 | } else { | |
1495 | /* Can go around again, but... */ | |
1496 | *again = true; | |
1497 | /* We've found something so probably don't need to */ | |
1498 | return true; | |
1499 | } | |
1500 | } | |
1501 | ||
3d0684b2 JQ |
1502 | /** |
1503 | * unqueue_page: gets a page of the queue | |
1504 | * | |
a82d593b | 1505 | * Helper for 'get_queued_page' - gets a page off the queue |
a82d593b | 1506 | * |
3d0684b2 JQ |
1507 | * Returns the block of the page (or NULL if none available) |
1508 | * | |
ec481c6c | 1509 | * @rs: current RAM state |
3d0684b2 | 1510 | * @offset: used to return the offset within the RAMBlock |
a82d593b | 1511 | */ |
f20e2865 | 1512 | static RAMBlock *unqueue_page(RAMState *rs, ram_addr_t *offset) |
a82d593b DDAG |
1513 | { |
1514 | RAMBlock *block = NULL; | |
1515 | ||
ec481c6c JQ |
1516 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1517 | if (!QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
1518 | struct RAMSrcPageRequest *entry = | |
1519 | QSIMPLEQ_FIRST(&rs->src_page_requests); | |
a82d593b DDAG |
1520 | block = entry->rb; |
1521 | *offset = entry->offset; | |
a82d593b DDAG |
1522 | |
1523 | if (entry->len > TARGET_PAGE_SIZE) { | |
1524 | entry->len -= TARGET_PAGE_SIZE; | |
1525 | entry->offset += TARGET_PAGE_SIZE; | |
1526 | } else { | |
1527 | memory_region_unref(block->mr); | |
ec481c6c | 1528 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
a82d593b DDAG |
1529 | g_free(entry); |
1530 | } | |
1531 | } | |
ec481c6c | 1532 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
a82d593b DDAG |
1533 | |
1534 | return block; | |
1535 | } | |
1536 | ||
3d0684b2 JQ |
1537 | /** |
1538 | * get_queued_page: unqueue a page from the postocpy requests | |
1539 | * | |
1540 | * Skips pages that are already sent (!dirty) | |
a82d593b | 1541 | * |
3d0684b2 | 1542 | * Returns if a queued page is found |
a82d593b | 1543 | * |
6f37bb8b | 1544 | * @rs: current RAM state |
3d0684b2 | 1545 | * @pss: data about the state of the current dirty page scan |
a82d593b | 1546 | */ |
f20e2865 | 1547 | static bool get_queued_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b DDAG |
1548 | { |
1549 | RAMBlock *block; | |
1550 | ram_addr_t offset; | |
1551 | bool dirty; | |
1552 | ||
1553 | do { | |
f20e2865 | 1554 | block = unqueue_page(rs, &offset); |
a82d593b DDAG |
1555 | /* |
1556 | * We're sending this page, and since it's postcopy nothing else | |
1557 | * will dirty it, and we must make sure it doesn't get sent again | |
1558 | * even if this queue request was received after the background | |
1559 | * search already sent it. | |
1560 | */ | |
1561 | if (block) { | |
f20e2865 JQ |
1562 | unsigned long page; |
1563 | ||
6b6712ef JQ |
1564 | page = offset >> TARGET_PAGE_BITS; |
1565 | dirty = test_bit(page, block->bmap); | |
a82d593b | 1566 | if (!dirty) { |
06b10688 | 1567 | trace_get_queued_page_not_dirty(block->idstr, (uint64_t)offset, |
6b6712ef | 1568 | page, test_bit(page, block->unsentmap)); |
a82d593b | 1569 | } else { |
f20e2865 | 1570 | trace_get_queued_page(block->idstr, (uint64_t)offset, page); |
a82d593b DDAG |
1571 | } |
1572 | } | |
1573 | ||
1574 | } while (block && !dirty); | |
1575 | ||
1576 | if (block) { | |
1577 | /* | |
1578 | * As soon as we start servicing pages out of order, then we have | |
1579 | * to kill the bulk stage, since the bulk stage assumes | |
1580 | * in (migration_bitmap_find_and_reset_dirty) that every page is | |
1581 | * dirty, that's no longer true. | |
1582 | */ | |
6f37bb8b | 1583 | rs->ram_bulk_stage = false; |
a82d593b DDAG |
1584 | |
1585 | /* | |
1586 | * We want the background search to continue from the queued page | |
1587 | * since the guest is likely to want other pages near to the page | |
1588 | * it just requested. | |
1589 | */ | |
1590 | pss->block = block; | |
a935e30f | 1591 | pss->page = offset >> TARGET_PAGE_BITS; |
a82d593b DDAG |
1592 | } |
1593 | ||
1594 | return !!block; | |
1595 | } | |
1596 | ||
6c595cde | 1597 | /** |
5e58f968 JQ |
1598 | * migration_page_queue_free: drop any remaining pages in the ram |
1599 | * request queue | |
6c595cde | 1600 | * |
3d0684b2 JQ |
1601 | * It should be empty at the end anyway, but in error cases there may |
1602 | * be some left. in case that there is any page left, we drop it. | |
1603 | * | |
6c595cde | 1604 | */ |
83c13382 | 1605 | static void migration_page_queue_free(RAMState *rs) |
6c595cde | 1606 | { |
ec481c6c | 1607 | struct RAMSrcPageRequest *mspr, *next_mspr; |
6c595cde DDAG |
1608 | /* This queue generally should be empty - but in the case of a failed |
1609 | * migration might have some droppings in. | |
1610 | */ | |
1611 | rcu_read_lock(); | |
ec481c6c | 1612 | QSIMPLEQ_FOREACH_SAFE(mspr, &rs->src_page_requests, next_req, next_mspr) { |
6c595cde | 1613 | memory_region_unref(mspr->rb->mr); |
ec481c6c | 1614 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
6c595cde DDAG |
1615 | g_free(mspr); |
1616 | } | |
1617 | rcu_read_unlock(); | |
1618 | } | |
1619 | ||
1620 | /** | |
3d0684b2 JQ |
1621 | * ram_save_queue_pages: queue the page for transmission |
1622 | * | |
1623 | * A request from postcopy destination for example. | |
1624 | * | |
1625 | * Returns zero on success or negative on error | |
1626 | * | |
3d0684b2 JQ |
1627 | * @rbname: Name of the RAMBLock of the request. NULL means the |
1628 | * same that last one. | |
1629 | * @start: starting address from the start of the RAMBlock | |
1630 | * @len: length (in bytes) to send | |
6c595cde | 1631 | */ |
96506894 | 1632 | int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len) |
6c595cde DDAG |
1633 | { |
1634 | RAMBlock *ramblock; | |
53518d94 | 1635 | RAMState *rs = ram_state; |
6c595cde | 1636 | |
9360447d | 1637 | ram_counters.postcopy_requests++; |
6c595cde DDAG |
1638 | rcu_read_lock(); |
1639 | if (!rbname) { | |
1640 | /* Reuse last RAMBlock */ | |
68a098f3 | 1641 | ramblock = rs->last_req_rb; |
6c595cde DDAG |
1642 | |
1643 | if (!ramblock) { | |
1644 | /* | |
1645 | * Shouldn't happen, we can't reuse the last RAMBlock if | |
1646 | * it's the 1st request. | |
1647 | */ | |
1648 | error_report("ram_save_queue_pages no previous block"); | |
1649 | goto err; | |
1650 | } | |
1651 | } else { | |
1652 | ramblock = qemu_ram_block_by_name(rbname); | |
1653 | ||
1654 | if (!ramblock) { | |
1655 | /* We shouldn't be asked for a non-existent RAMBlock */ | |
1656 | error_report("ram_save_queue_pages no block '%s'", rbname); | |
1657 | goto err; | |
1658 | } | |
68a098f3 | 1659 | rs->last_req_rb = ramblock; |
6c595cde DDAG |
1660 | } |
1661 | trace_ram_save_queue_pages(ramblock->idstr, start, len); | |
1662 | if (start+len > ramblock->used_length) { | |
9458ad6b JQ |
1663 | error_report("%s request overrun start=" RAM_ADDR_FMT " len=" |
1664 | RAM_ADDR_FMT " blocklen=" RAM_ADDR_FMT, | |
6c595cde DDAG |
1665 | __func__, start, len, ramblock->used_length); |
1666 | goto err; | |
1667 | } | |
1668 | ||
ec481c6c JQ |
1669 | struct RAMSrcPageRequest *new_entry = |
1670 | g_malloc0(sizeof(struct RAMSrcPageRequest)); | |
6c595cde DDAG |
1671 | new_entry->rb = ramblock; |
1672 | new_entry->offset = start; | |
1673 | new_entry->len = len; | |
1674 | ||
1675 | memory_region_ref(ramblock->mr); | |
ec481c6c JQ |
1676 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1677 | QSIMPLEQ_INSERT_TAIL(&rs->src_page_requests, new_entry, next_req); | |
1678 | qemu_mutex_unlock(&rs->src_page_req_mutex); | |
6c595cde DDAG |
1679 | rcu_read_unlock(); |
1680 | ||
1681 | return 0; | |
1682 | ||
1683 | err: | |
1684 | rcu_read_unlock(); | |
1685 | return -1; | |
1686 | } | |
1687 | ||
d7400a34 XG |
1688 | static bool save_page_use_compression(RAMState *rs) |
1689 | { | |
1690 | if (!migrate_use_compression()) { | |
1691 | return false; | |
1692 | } | |
1693 | ||
1694 | /* | |
1695 | * If xbzrle is on, stop using the data compression after first | |
1696 | * round of migration even if compression is enabled. In theory, | |
1697 | * xbzrle can do better than compression. | |
1698 | */ | |
1699 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { | |
1700 | return true; | |
1701 | } | |
1702 | ||
1703 | return false; | |
1704 | } | |
1705 | ||
a82d593b | 1706 | /** |
3d0684b2 | 1707 | * ram_save_target_page: save one target page |
a82d593b | 1708 | * |
3d0684b2 | 1709 | * Returns the number of pages written |
a82d593b | 1710 | * |
6f37bb8b | 1711 | * @rs: current RAM state |
3d0684b2 | 1712 | * @pss: data about the page we want to send |
a82d593b | 1713 | * @last_stage: if we are at the completion stage |
a82d593b | 1714 | */ |
a0a8aa14 | 1715 | static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 1716 | bool last_stage) |
a82d593b | 1717 | { |
a8ec91f9 XG |
1718 | RAMBlock *block = pss->block; |
1719 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; | |
1720 | int res; | |
1721 | ||
1722 | if (control_save_page(rs, block, offset, &res)) { | |
1723 | return res; | |
1724 | } | |
1725 | ||
1faa5665 | 1726 | /* |
d7400a34 XG |
1727 | * When starting the process of a new block, the first page of |
1728 | * the block should be sent out before other pages in the same | |
1729 | * block, and all the pages in last block should have been sent | |
1730 | * out, keeping this order is important, because the 'cont' flag | |
1731 | * is used to avoid resending the block name. | |
1faa5665 | 1732 | */ |
d7400a34 XG |
1733 | if (block != rs->last_sent_block && save_page_use_compression(rs)) { |
1734 | flush_compressed_data(rs); | |
1735 | } | |
1736 | ||
1737 | res = save_zero_page(rs, block, offset); | |
1738 | if (res > 0) { | |
1739 | /* Must let xbzrle know, otherwise a previous (now 0'd) cached | |
1740 | * page would be stale | |
1741 | */ | |
1742 | if (!save_page_use_compression(rs)) { | |
1743 | XBZRLE_cache_lock(); | |
1744 | xbzrle_cache_zero_page(rs, block->offset + offset); | |
1745 | XBZRLE_cache_unlock(); | |
1746 | } | |
1747 | ram_release_pages(block->idstr, offset, res); | |
1748 | return res; | |
1749 | } | |
1750 | ||
da3f56cb XG |
1751 | /* |
1752 | * Make sure the first page is sent out before other pages. | |
1753 | * | |
1754 | * we post it as normal page as compression will take much | |
1755 | * CPU resource. | |
1756 | */ | |
1757 | if (block == rs->last_sent_block && save_page_use_compression(rs)) { | |
701b1876 | 1758 | return compress_page_with_multi_thread(rs, block, offset); |
a82d593b DDAG |
1759 | } |
1760 | ||
1faa5665 | 1761 | return ram_save_page(rs, pss, last_stage); |
a82d593b DDAG |
1762 | } |
1763 | ||
1764 | /** | |
3d0684b2 | 1765 | * ram_save_host_page: save a whole host page |
a82d593b | 1766 | * |
3d0684b2 JQ |
1767 | * Starting at *offset send pages up to the end of the current host |
1768 | * page. It's valid for the initial offset to point into the middle of | |
1769 | * a host page in which case the remainder of the hostpage is sent. | |
1770 | * Only dirty target pages are sent. Note that the host page size may | |
1771 | * be a huge page for this block. | |
1eb3fc0a DDAG |
1772 | * The saving stops at the boundary of the used_length of the block |
1773 | * if the RAMBlock isn't a multiple of the host page size. | |
a82d593b | 1774 | * |
3d0684b2 JQ |
1775 | * Returns the number of pages written or negative on error |
1776 | * | |
6f37bb8b | 1777 | * @rs: current RAM state |
3d0684b2 | 1778 | * @ms: current migration state |
3d0684b2 | 1779 | * @pss: data about the page we want to send |
a82d593b | 1780 | * @last_stage: if we are at the completion stage |
a82d593b | 1781 | */ |
a0a8aa14 | 1782 | static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 1783 | bool last_stage) |
a82d593b DDAG |
1784 | { |
1785 | int tmppages, pages = 0; | |
a935e30f JQ |
1786 | size_t pagesize_bits = |
1787 | qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
4c011c37 | 1788 | |
a82d593b | 1789 | do { |
1faa5665 XG |
1790 | /* Check the pages is dirty and if it is send it */ |
1791 | if (!migration_bitmap_clear_dirty(rs, pss->block, pss->page)) { | |
1792 | pss->page++; | |
1793 | continue; | |
1794 | } | |
1795 | ||
f20e2865 | 1796 | tmppages = ram_save_target_page(rs, pss, last_stage); |
a82d593b DDAG |
1797 | if (tmppages < 0) { |
1798 | return tmppages; | |
1799 | } | |
1800 | ||
1801 | pages += tmppages; | |
1faa5665 XG |
1802 | if (pss->block->unsentmap) { |
1803 | clear_bit(pss->page, pss->block->unsentmap); | |
1804 | } | |
1805 | ||
a935e30f | 1806 | pss->page++; |
1eb3fc0a DDAG |
1807 | } while ((pss->page & (pagesize_bits - 1)) && |
1808 | offset_in_ramblock(pss->block, pss->page << TARGET_PAGE_BITS)); | |
a82d593b DDAG |
1809 | |
1810 | /* The offset we leave with is the last one we looked at */ | |
a935e30f | 1811 | pss->page--; |
a82d593b DDAG |
1812 | return pages; |
1813 | } | |
6c595cde | 1814 | |
56e93d26 | 1815 | /** |
3d0684b2 | 1816 | * ram_find_and_save_block: finds a dirty page and sends it to f |
56e93d26 JQ |
1817 | * |
1818 | * Called within an RCU critical section. | |
1819 | * | |
3d0684b2 | 1820 | * Returns the number of pages written where zero means no dirty pages |
56e93d26 | 1821 | * |
6f37bb8b | 1822 | * @rs: current RAM state |
56e93d26 | 1823 | * @last_stage: if we are at the completion stage |
a82d593b DDAG |
1824 | * |
1825 | * On systems where host-page-size > target-page-size it will send all the | |
1826 | * pages in a host page that are dirty. | |
56e93d26 JQ |
1827 | */ |
1828 | ||
ce25d337 | 1829 | static int ram_find_and_save_block(RAMState *rs, bool last_stage) |
56e93d26 | 1830 | { |
b8fb8cb7 | 1831 | PageSearchStatus pss; |
56e93d26 | 1832 | int pages = 0; |
b9e60928 | 1833 | bool again, found; |
56e93d26 | 1834 | |
0827b9e9 AA |
1835 | /* No dirty page as there is zero RAM */ |
1836 | if (!ram_bytes_total()) { | |
1837 | return pages; | |
1838 | } | |
1839 | ||
6f37bb8b | 1840 | pss.block = rs->last_seen_block; |
a935e30f | 1841 | pss.page = rs->last_page; |
b8fb8cb7 DDAG |
1842 | pss.complete_round = false; |
1843 | ||
1844 | if (!pss.block) { | |
1845 | pss.block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1846 | } | |
56e93d26 | 1847 | |
b9e60928 | 1848 | do { |
a82d593b | 1849 | again = true; |
f20e2865 | 1850 | found = get_queued_page(rs, &pss); |
b9e60928 | 1851 | |
a82d593b DDAG |
1852 | if (!found) { |
1853 | /* priority queue empty, so just search for something dirty */ | |
f20e2865 | 1854 | found = find_dirty_block(rs, &pss, &again); |
a82d593b | 1855 | } |
f3f491fc | 1856 | |
a82d593b | 1857 | if (found) { |
f20e2865 | 1858 | pages = ram_save_host_page(rs, &pss, last_stage); |
56e93d26 | 1859 | } |
b9e60928 | 1860 | } while (!pages && again); |
56e93d26 | 1861 | |
6f37bb8b | 1862 | rs->last_seen_block = pss.block; |
a935e30f | 1863 | rs->last_page = pss.page; |
56e93d26 JQ |
1864 | |
1865 | return pages; | |
1866 | } | |
1867 | ||
1868 | void acct_update_position(QEMUFile *f, size_t size, bool zero) | |
1869 | { | |
1870 | uint64_t pages = size / TARGET_PAGE_SIZE; | |
f7ccd61b | 1871 | |
56e93d26 | 1872 | if (zero) { |
9360447d | 1873 | ram_counters.duplicate += pages; |
56e93d26 | 1874 | } else { |
9360447d JQ |
1875 | ram_counters.normal += pages; |
1876 | ram_counters.transferred += size; | |
56e93d26 JQ |
1877 | qemu_update_position(f, size); |
1878 | } | |
1879 | } | |
1880 | ||
56e93d26 JQ |
1881 | uint64_t ram_bytes_total(void) |
1882 | { | |
1883 | RAMBlock *block; | |
1884 | uint64_t total = 0; | |
1885 | ||
1886 | rcu_read_lock(); | |
99e15582 | 1887 | RAMBLOCK_FOREACH(block) { |
56e93d26 | 1888 | total += block->used_length; |
99e15582 | 1889 | } |
56e93d26 JQ |
1890 | rcu_read_unlock(); |
1891 | return total; | |
1892 | } | |
1893 | ||
f265e0e4 | 1894 | static void xbzrle_load_setup(void) |
56e93d26 | 1895 | { |
f265e0e4 | 1896 | XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE); |
56e93d26 JQ |
1897 | } |
1898 | ||
f265e0e4 JQ |
1899 | static void xbzrle_load_cleanup(void) |
1900 | { | |
1901 | g_free(XBZRLE.decoded_buf); | |
1902 | XBZRLE.decoded_buf = NULL; | |
1903 | } | |
1904 | ||
7d7c96be PX |
1905 | static void ram_state_cleanup(RAMState **rsp) |
1906 | { | |
b9ccaf6d DDAG |
1907 | if (*rsp) { |
1908 | migration_page_queue_free(*rsp); | |
1909 | qemu_mutex_destroy(&(*rsp)->bitmap_mutex); | |
1910 | qemu_mutex_destroy(&(*rsp)->src_page_req_mutex); | |
1911 | g_free(*rsp); | |
1912 | *rsp = NULL; | |
1913 | } | |
7d7c96be PX |
1914 | } |
1915 | ||
84593a08 PX |
1916 | static void xbzrle_cleanup(void) |
1917 | { | |
1918 | XBZRLE_cache_lock(); | |
1919 | if (XBZRLE.cache) { | |
1920 | cache_fini(XBZRLE.cache); | |
1921 | g_free(XBZRLE.encoded_buf); | |
1922 | g_free(XBZRLE.current_buf); | |
1923 | g_free(XBZRLE.zero_target_page); | |
1924 | XBZRLE.cache = NULL; | |
1925 | XBZRLE.encoded_buf = NULL; | |
1926 | XBZRLE.current_buf = NULL; | |
1927 | XBZRLE.zero_target_page = NULL; | |
1928 | } | |
1929 | XBZRLE_cache_unlock(); | |
1930 | } | |
1931 | ||
f265e0e4 | 1932 | static void ram_save_cleanup(void *opaque) |
56e93d26 | 1933 | { |
53518d94 | 1934 | RAMState **rsp = opaque; |
6b6712ef | 1935 | RAMBlock *block; |
eb859c53 | 1936 | |
2ff64038 LZ |
1937 | /* caller have hold iothread lock or is in a bh, so there is |
1938 | * no writing race against this migration_bitmap | |
1939 | */ | |
6b6712ef JQ |
1940 | memory_global_dirty_log_stop(); |
1941 | ||
1942 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1943 | g_free(block->bmap); | |
1944 | block->bmap = NULL; | |
1945 | g_free(block->unsentmap); | |
1946 | block->unsentmap = NULL; | |
56e93d26 JQ |
1947 | } |
1948 | ||
84593a08 | 1949 | xbzrle_cleanup(); |
f0afa331 | 1950 | compress_threads_save_cleanup(); |
7d7c96be | 1951 | ram_state_cleanup(rsp); |
56e93d26 JQ |
1952 | } |
1953 | ||
6f37bb8b | 1954 | static void ram_state_reset(RAMState *rs) |
56e93d26 | 1955 | { |
6f37bb8b JQ |
1956 | rs->last_seen_block = NULL; |
1957 | rs->last_sent_block = NULL; | |
269ace29 | 1958 | rs->last_page = 0; |
6f37bb8b JQ |
1959 | rs->last_version = ram_list.version; |
1960 | rs->ram_bulk_stage = true; | |
56e93d26 JQ |
1961 | } |
1962 | ||
1963 | #define MAX_WAIT 50 /* ms, half buffered_file limit */ | |
1964 | ||
4f2e4252 DDAG |
1965 | /* |
1966 | * 'expected' is the value you expect the bitmap mostly to be full | |
1967 | * of; it won't bother printing lines that are all this value. | |
1968 | * If 'todump' is null the migration bitmap is dumped. | |
1969 | */ | |
6b6712ef JQ |
1970 | void ram_debug_dump_bitmap(unsigned long *todump, bool expected, |
1971 | unsigned long pages) | |
4f2e4252 | 1972 | { |
4f2e4252 DDAG |
1973 | int64_t cur; |
1974 | int64_t linelen = 128; | |
1975 | char linebuf[129]; | |
1976 | ||
6b6712ef | 1977 | for (cur = 0; cur < pages; cur += linelen) { |
4f2e4252 DDAG |
1978 | int64_t curb; |
1979 | bool found = false; | |
1980 | /* | |
1981 | * Last line; catch the case where the line length | |
1982 | * is longer than remaining ram | |
1983 | */ | |
6b6712ef JQ |
1984 | if (cur + linelen > pages) { |
1985 | linelen = pages - cur; | |
4f2e4252 DDAG |
1986 | } |
1987 | for (curb = 0; curb < linelen; curb++) { | |
1988 | bool thisbit = test_bit(cur + curb, todump); | |
1989 | linebuf[curb] = thisbit ? '1' : '.'; | |
1990 | found = found || (thisbit != expected); | |
1991 | } | |
1992 | if (found) { | |
1993 | linebuf[curb] = '\0'; | |
1994 | fprintf(stderr, "0x%08" PRIx64 " : %s\n", cur, linebuf); | |
1995 | } | |
1996 | } | |
1997 | } | |
1998 | ||
e0b266f0 DDAG |
1999 | /* **** functions for postcopy ***** */ |
2000 | ||
ced1c616 PB |
2001 | void ram_postcopy_migrated_memory_release(MigrationState *ms) |
2002 | { | |
2003 | struct RAMBlock *block; | |
ced1c616 | 2004 | |
99e15582 | 2005 | RAMBLOCK_FOREACH(block) { |
6b6712ef JQ |
2006 | unsigned long *bitmap = block->bmap; |
2007 | unsigned long range = block->used_length >> TARGET_PAGE_BITS; | |
2008 | unsigned long run_start = find_next_zero_bit(bitmap, range, 0); | |
ced1c616 PB |
2009 | |
2010 | while (run_start < range) { | |
2011 | unsigned long run_end = find_next_bit(bitmap, range, run_start + 1); | |
aaa2064c | 2012 | ram_discard_range(block->idstr, run_start << TARGET_PAGE_BITS, |
ced1c616 PB |
2013 | (run_end - run_start) << TARGET_PAGE_BITS); |
2014 | run_start = find_next_zero_bit(bitmap, range, run_end + 1); | |
2015 | } | |
2016 | } | |
2017 | } | |
2018 | ||
3d0684b2 JQ |
2019 | /** |
2020 | * postcopy_send_discard_bm_ram: discard a RAMBlock | |
2021 | * | |
2022 | * Returns zero on success | |
2023 | * | |
e0b266f0 DDAG |
2024 | * Callback from postcopy_each_ram_send_discard for each RAMBlock |
2025 | * Note: At this point the 'unsentmap' is the processed bitmap combined | |
2026 | * with the dirtymap; so a '1' means it's either dirty or unsent. | |
3d0684b2 JQ |
2027 | * |
2028 | * @ms: current migration state | |
2029 | * @pds: state for postcopy | |
2030 | * @start: RAMBlock starting page | |
2031 | * @length: RAMBlock size | |
e0b266f0 DDAG |
2032 | */ |
2033 | static int postcopy_send_discard_bm_ram(MigrationState *ms, | |
2034 | PostcopyDiscardState *pds, | |
6b6712ef | 2035 | RAMBlock *block) |
e0b266f0 | 2036 | { |
6b6712ef | 2037 | unsigned long end = block->used_length >> TARGET_PAGE_BITS; |
e0b266f0 | 2038 | unsigned long current; |
6b6712ef | 2039 | unsigned long *unsentmap = block->unsentmap; |
e0b266f0 | 2040 | |
6b6712ef | 2041 | for (current = 0; current < end; ) { |
e0b266f0 DDAG |
2042 | unsigned long one = find_next_bit(unsentmap, end, current); |
2043 | ||
2044 | if (one <= end) { | |
2045 | unsigned long zero = find_next_zero_bit(unsentmap, end, one + 1); | |
2046 | unsigned long discard_length; | |
2047 | ||
2048 | if (zero >= end) { | |
2049 | discard_length = end - one; | |
2050 | } else { | |
2051 | discard_length = zero - one; | |
2052 | } | |
d688c62d DDAG |
2053 | if (discard_length) { |
2054 | postcopy_discard_send_range(ms, pds, one, discard_length); | |
2055 | } | |
e0b266f0 DDAG |
2056 | current = one + discard_length; |
2057 | } else { | |
2058 | current = one; | |
2059 | } | |
2060 | } | |
2061 | ||
2062 | return 0; | |
2063 | } | |
2064 | ||
3d0684b2 JQ |
2065 | /** |
2066 | * postcopy_each_ram_send_discard: discard all RAMBlocks | |
2067 | * | |
2068 | * Returns 0 for success or negative for error | |
2069 | * | |
e0b266f0 DDAG |
2070 | * Utility for the outgoing postcopy code. |
2071 | * Calls postcopy_send_discard_bm_ram for each RAMBlock | |
2072 | * passing it bitmap indexes and name. | |
e0b266f0 DDAG |
2073 | * (qemu_ram_foreach_block ends up passing unscaled lengths |
2074 | * which would mean postcopy code would have to deal with target page) | |
3d0684b2 JQ |
2075 | * |
2076 | * @ms: current migration state | |
e0b266f0 DDAG |
2077 | */ |
2078 | static int postcopy_each_ram_send_discard(MigrationState *ms) | |
2079 | { | |
2080 | struct RAMBlock *block; | |
2081 | int ret; | |
2082 | ||
99e15582 | 2083 | RAMBLOCK_FOREACH(block) { |
6b6712ef JQ |
2084 | PostcopyDiscardState *pds = |
2085 | postcopy_discard_send_init(ms, block->idstr); | |
e0b266f0 DDAG |
2086 | |
2087 | /* | |
2088 | * Postcopy sends chunks of bitmap over the wire, but it | |
2089 | * just needs indexes at this point, avoids it having | |
2090 | * target page specific code. | |
2091 | */ | |
6b6712ef | 2092 | ret = postcopy_send_discard_bm_ram(ms, pds, block); |
e0b266f0 DDAG |
2093 | postcopy_discard_send_finish(ms, pds); |
2094 | if (ret) { | |
2095 | return ret; | |
2096 | } | |
2097 | } | |
2098 | ||
2099 | return 0; | |
2100 | } | |
2101 | ||
3d0684b2 JQ |
2102 | /** |
2103 | * postcopy_chunk_hostpages_pass: canocalize bitmap in hostpages | |
2104 | * | |
2105 | * Helper for postcopy_chunk_hostpages; it's called twice to | |
2106 | * canonicalize the two bitmaps, that are similar, but one is | |
2107 | * inverted. | |
99e314eb | 2108 | * |
3d0684b2 JQ |
2109 | * Postcopy requires that all target pages in a hostpage are dirty or |
2110 | * clean, not a mix. This function canonicalizes the bitmaps. | |
99e314eb | 2111 | * |
3d0684b2 JQ |
2112 | * @ms: current migration state |
2113 | * @unsent_pass: if true we need to canonicalize partially unsent host pages | |
2114 | * otherwise we need to canonicalize partially dirty host pages | |
2115 | * @block: block that contains the page we want to canonicalize | |
2116 | * @pds: state for postcopy | |
99e314eb DDAG |
2117 | */ |
2118 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, bool unsent_pass, | |
2119 | RAMBlock *block, | |
2120 | PostcopyDiscardState *pds) | |
2121 | { | |
53518d94 | 2122 | RAMState *rs = ram_state; |
6b6712ef JQ |
2123 | unsigned long *bitmap = block->bmap; |
2124 | unsigned long *unsentmap = block->unsentmap; | |
29c59172 | 2125 | unsigned int host_ratio = block->page_size / TARGET_PAGE_SIZE; |
6b6712ef | 2126 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
99e314eb DDAG |
2127 | unsigned long run_start; |
2128 | ||
29c59172 DDAG |
2129 | if (block->page_size == TARGET_PAGE_SIZE) { |
2130 | /* Easy case - TPS==HPS for a non-huge page RAMBlock */ | |
2131 | return; | |
2132 | } | |
2133 | ||
99e314eb DDAG |
2134 | if (unsent_pass) { |
2135 | /* Find a sent page */ | |
6b6712ef | 2136 | run_start = find_next_zero_bit(unsentmap, pages, 0); |
99e314eb DDAG |
2137 | } else { |
2138 | /* Find a dirty page */ | |
6b6712ef | 2139 | run_start = find_next_bit(bitmap, pages, 0); |
99e314eb DDAG |
2140 | } |
2141 | ||
6b6712ef | 2142 | while (run_start < pages) { |
99e314eb DDAG |
2143 | bool do_fixup = false; |
2144 | unsigned long fixup_start_addr; | |
2145 | unsigned long host_offset; | |
2146 | ||
2147 | /* | |
2148 | * If the start of this run of pages is in the middle of a host | |
2149 | * page, then we need to fixup this host page. | |
2150 | */ | |
2151 | host_offset = run_start % host_ratio; | |
2152 | if (host_offset) { | |
2153 | do_fixup = true; | |
2154 | run_start -= host_offset; | |
2155 | fixup_start_addr = run_start; | |
2156 | /* For the next pass */ | |
2157 | run_start = run_start + host_ratio; | |
2158 | } else { | |
2159 | /* Find the end of this run */ | |
2160 | unsigned long run_end; | |
2161 | if (unsent_pass) { | |
6b6712ef | 2162 | run_end = find_next_bit(unsentmap, pages, run_start + 1); |
99e314eb | 2163 | } else { |
6b6712ef | 2164 | run_end = find_next_zero_bit(bitmap, pages, run_start + 1); |
99e314eb DDAG |
2165 | } |
2166 | /* | |
2167 | * If the end isn't at the start of a host page, then the | |
2168 | * run doesn't finish at the end of a host page | |
2169 | * and we need to discard. | |
2170 | */ | |
2171 | host_offset = run_end % host_ratio; | |
2172 | if (host_offset) { | |
2173 | do_fixup = true; | |
2174 | fixup_start_addr = run_end - host_offset; | |
2175 | /* | |
2176 | * This host page has gone, the next loop iteration starts | |
2177 | * from after the fixup | |
2178 | */ | |
2179 | run_start = fixup_start_addr + host_ratio; | |
2180 | } else { | |
2181 | /* | |
2182 | * No discards on this iteration, next loop starts from | |
2183 | * next sent/dirty page | |
2184 | */ | |
2185 | run_start = run_end + 1; | |
2186 | } | |
2187 | } | |
2188 | ||
2189 | if (do_fixup) { | |
2190 | unsigned long page; | |
2191 | ||
2192 | /* Tell the destination to discard this page */ | |
2193 | if (unsent_pass || !test_bit(fixup_start_addr, unsentmap)) { | |
2194 | /* For the unsent_pass we: | |
2195 | * discard partially sent pages | |
2196 | * For the !unsent_pass (dirty) we: | |
2197 | * discard partially dirty pages that were sent | |
2198 | * (any partially sent pages were already discarded | |
2199 | * by the previous unsent_pass) | |
2200 | */ | |
2201 | postcopy_discard_send_range(ms, pds, fixup_start_addr, | |
2202 | host_ratio); | |
2203 | } | |
2204 | ||
2205 | /* Clean up the bitmap */ | |
2206 | for (page = fixup_start_addr; | |
2207 | page < fixup_start_addr + host_ratio; page++) { | |
2208 | /* All pages in this host page are now not sent */ | |
2209 | set_bit(page, unsentmap); | |
2210 | ||
2211 | /* | |
2212 | * Remark them as dirty, updating the count for any pages | |
2213 | * that weren't previously dirty. | |
2214 | */ | |
0d8ec885 | 2215 | rs->migration_dirty_pages += !test_and_set_bit(page, bitmap); |
99e314eb DDAG |
2216 | } |
2217 | } | |
2218 | ||
2219 | if (unsent_pass) { | |
2220 | /* Find the next sent page for the next iteration */ | |
6b6712ef | 2221 | run_start = find_next_zero_bit(unsentmap, pages, run_start); |
99e314eb DDAG |
2222 | } else { |
2223 | /* Find the next dirty page for the next iteration */ | |
6b6712ef | 2224 | run_start = find_next_bit(bitmap, pages, run_start); |
99e314eb DDAG |
2225 | } |
2226 | } | |
2227 | } | |
2228 | ||
3d0684b2 JQ |
2229 | /** |
2230 | * postcopy_chuck_hostpages: discrad any partially sent host page | |
2231 | * | |
99e314eb DDAG |
2232 | * Utility for the outgoing postcopy code. |
2233 | * | |
2234 | * Discard any partially sent host-page size chunks, mark any partially | |
29c59172 DDAG |
2235 | * dirty host-page size chunks as all dirty. In this case the host-page |
2236 | * is the host-page for the particular RAMBlock, i.e. it might be a huge page | |
99e314eb | 2237 | * |
3d0684b2 JQ |
2238 | * Returns zero on success |
2239 | * | |
2240 | * @ms: current migration state | |
6b6712ef | 2241 | * @block: block we want to work with |
99e314eb | 2242 | */ |
6b6712ef | 2243 | static int postcopy_chunk_hostpages(MigrationState *ms, RAMBlock *block) |
99e314eb | 2244 | { |
6b6712ef JQ |
2245 | PostcopyDiscardState *pds = |
2246 | postcopy_discard_send_init(ms, block->idstr); | |
99e314eb | 2247 | |
6b6712ef JQ |
2248 | /* First pass: Discard all partially sent host pages */ |
2249 | postcopy_chunk_hostpages_pass(ms, true, block, pds); | |
2250 | /* | |
2251 | * Second pass: Ensure that all partially dirty host pages are made | |
2252 | * fully dirty. | |
2253 | */ | |
2254 | postcopy_chunk_hostpages_pass(ms, false, block, pds); | |
99e314eb | 2255 | |
6b6712ef | 2256 | postcopy_discard_send_finish(ms, pds); |
99e314eb DDAG |
2257 | return 0; |
2258 | } | |
2259 | ||
3d0684b2 JQ |
2260 | /** |
2261 | * ram_postcopy_send_discard_bitmap: transmit the discard bitmap | |
2262 | * | |
2263 | * Returns zero on success | |
2264 | * | |
e0b266f0 DDAG |
2265 | * Transmit the set of pages to be discarded after precopy to the target |
2266 | * these are pages that: | |
2267 | * a) Have been previously transmitted but are now dirty again | |
2268 | * b) Pages that have never been transmitted, this ensures that | |
2269 | * any pages on the destination that have been mapped by background | |
2270 | * tasks get discarded (transparent huge pages is the specific concern) | |
2271 | * Hopefully this is pretty sparse | |
3d0684b2 JQ |
2272 | * |
2273 | * @ms: current migration state | |
e0b266f0 DDAG |
2274 | */ |
2275 | int ram_postcopy_send_discard_bitmap(MigrationState *ms) | |
2276 | { | |
53518d94 | 2277 | RAMState *rs = ram_state; |
6b6712ef | 2278 | RAMBlock *block; |
e0b266f0 | 2279 | int ret; |
e0b266f0 DDAG |
2280 | |
2281 | rcu_read_lock(); | |
2282 | ||
2283 | /* This should be our last sync, the src is now paused */ | |
eb859c53 | 2284 | migration_bitmap_sync(rs); |
e0b266f0 | 2285 | |
6b6712ef JQ |
2286 | /* Easiest way to make sure we don't resume in the middle of a host-page */ |
2287 | rs->last_seen_block = NULL; | |
2288 | rs->last_sent_block = NULL; | |
2289 | rs->last_page = 0; | |
e0b266f0 | 2290 | |
6b6712ef JQ |
2291 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { |
2292 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; | |
2293 | unsigned long *bitmap = block->bmap; | |
2294 | unsigned long *unsentmap = block->unsentmap; | |
2295 | ||
2296 | if (!unsentmap) { | |
2297 | /* We don't have a safe way to resize the sentmap, so | |
2298 | * if the bitmap was resized it will be NULL at this | |
2299 | * point. | |
2300 | */ | |
2301 | error_report("migration ram resized during precopy phase"); | |
2302 | rcu_read_unlock(); | |
2303 | return -EINVAL; | |
2304 | } | |
2305 | /* Deal with TPS != HPS and huge pages */ | |
2306 | ret = postcopy_chunk_hostpages(ms, block); | |
2307 | if (ret) { | |
2308 | rcu_read_unlock(); | |
2309 | return ret; | |
2310 | } | |
e0b266f0 | 2311 | |
6b6712ef JQ |
2312 | /* |
2313 | * Update the unsentmap to be unsentmap = unsentmap | dirty | |
2314 | */ | |
2315 | bitmap_or(unsentmap, unsentmap, bitmap, pages); | |
e0b266f0 | 2316 | #ifdef DEBUG_POSTCOPY |
6b6712ef | 2317 | ram_debug_dump_bitmap(unsentmap, true, pages); |
e0b266f0 | 2318 | #endif |
6b6712ef JQ |
2319 | } |
2320 | trace_ram_postcopy_send_discard_bitmap(); | |
e0b266f0 DDAG |
2321 | |
2322 | ret = postcopy_each_ram_send_discard(ms); | |
2323 | rcu_read_unlock(); | |
2324 | ||
2325 | return ret; | |
2326 | } | |
2327 | ||
3d0684b2 JQ |
2328 | /** |
2329 | * ram_discard_range: discard dirtied pages at the beginning of postcopy | |
e0b266f0 | 2330 | * |
3d0684b2 | 2331 | * Returns zero on success |
e0b266f0 | 2332 | * |
36449157 JQ |
2333 | * @rbname: name of the RAMBlock of the request. NULL means the |
2334 | * same that last one. | |
3d0684b2 JQ |
2335 | * @start: RAMBlock starting page |
2336 | * @length: RAMBlock size | |
e0b266f0 | 2337 | */ |
aaa2064c | 2338 | int ram_discard_range(const char *rbname, uint64_t start, size_t length) |
e0b266f0 DDAG |
2339 | { |
2340 | int ret = -1; | |
2341 | ||
36449157 | 2342 | trace_ram_discard_range(rbname, start, length); |
d3a5038c | 2343 | |
e0b266f0 | 2344 | rcu_read_lock(); |
36449157 | 2345 | RAMBlock *rb = qemu_ram_block_by_name(rbname); |
e0b266f0 DDAG |
2346 | |
2347 | if (!rb) { | |
36449157 | 2348 | error_report("ram_discard_range: Failed to find block '%s'", rbname); |
e0b266f0 DDAG |
2349 | goto err; |
2350 | } | |
2351 | ||
f9494614 AP |
2352 | bitmap_clear(rb->receivedmap, start >> qemu_target_page_bits(), |
2353 | length >> qemu_target_page_bits()); | |
d3a5038c | 2354 | ret = ram_block_discard_range(rb, start, length); |
e0b266f0 DDAG |
2355 | |
2356 | err: | |
2357 | rcu_read_unlock(); | |
2358 | ||
2359 | return ret; | |
2360 | } | |
2361 | ||
84593a08 PX |
2362 | /* |
2363 | * For every allocation, we will try not to crash the VM if the | |
2364 | * allocation failed. | |
2365 | */ | |
2366 | static int xbzrle_init(void) | |
2367 | { | |
2368 | Error *local_err = NULL; | |
2369 | ||
2370 | if (!migrate_use_xbzrle()) { | |
2371 | return 0; | |
2372 | } | |
2373 | ||
2374 | XBZRLE_cache_lock(); | |
2375 | ||
2376 | XBZRLE.zero_target_page = g_try_malloc0(TARGET_PAGE_SIZE); | |
2377 | if (!XBZRLE.zero_target_page) { | |
2378 | error_report("%s: Error allocating zero page", __func__); | |
2379 | goto err_out; | |
2380 | } | |
2381 | ||
2382 | XBZRLE.cache = cache_init(migrate_xbzrle_cache_size(), | |
2383 | TARGET_PAGE_SIZE, &local_err); | |
2384 | if (!XBZRLE.cache) { | |
2385 | error_report_err(local_err); | |
2386 | goto free_zero_page; | |
2387 | } | |
2388 | ||
2389 | XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); | |
2390 | if (!XBZRLE.encoded_buf) { | |
2391 | error_report("%s: Error allocating encoded_buf", __func__); | |
2392 | goto free_cache; | |
2393 | } | |
2394 | ||
2395 | XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); | |
2396 | if (!XBZRLE.current_buf) { | |
2397 | error_report("%s: Error allocating current_buf", __func__); | |
2398 | goto free_encoded_buf; | |
2399 | } | |
2400 | ||
2401 | /* We are all good */ | |
2402 | XBZRLE_cache_unlock(); | |
2403 | return 0; | |
2404 | ||
2405 | free_encoded_buf: | |
2406 | g_free(XBZRLE.encoded_buf); | |
2407 | XBZRLE.encoded_buf = NULL; | |
2408 | free_cache: | |
2409 | cache_fini(XBZRLE.cache); | |
2410 | XBZRLE.cache = NULL; | |
2411 | free_zero_page: | |
2412 | g_free(XBZRLE.zero_target_page); | |
2413 | XBZRLE.zero_target_page = NULL; | |
2414 | err_out: | |
2415 | XBZRLE_cache_unlock(); | |
2416 | return -ENOMEM; | |
2417 | } | |
2418 | ||
53518d94 | 2419 | static int ram_state_init(RAMState **rsp) |
56e93d26 | 2420 | { |
7d00ee6a PX |
2421 | *rsp = g_try_new0(RAMState, 1); |
2422 | ||
2423 | if (!*rsp) { | |
2424 | error_report("%s: Init ramstate fail", __func__); | |
2425 | return -1; | |
2426 | } | |
53518d94 JQ |
2427 | |
2428 | qemu_mutex_init(&(*rsp)->bitmap_mutex); | |
2429 | qemu_mutex_init(&(*rsp)->src_page_req_mutex); | |
2430 | QSIMPLEQ_INIT(&(*rsp)->src_page_requests); | |
56e93d26 | 2431 | |
7d00ee6a PX |
2432 | /* |
2433 | * Count the total number of pages used by ram blocks not including any | |
2434 | * gaps due to alignment or unplugs. | |
2435 | */ | |
2436 | (*rsp)->migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS; | |
2437 | ||
2438 | ram_state_reset(*rsp); | |
2439 | ||
2440 | return 0; | |
2441 | } | |
2442 | ||
d6eff5d7 | 2443 | static void ram_list_init_bitmaps(void) |
7d00ee6a | 2444 | { |
d6eff5d7 PX |
2445 | RAMBlock *block; |
2446 | unsigned long pages; | |
56e93d26 | 2447 | |
0827b9e9 AA |
2448 | /* Skip setting bitmap if there is no RAM */ |
2449 | if (ram_bytes_total()) { | |
6b6712ef | 2450 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { |
d6eff5d7 | 2451 | pages = block->max_length >> TARGET_PAGE_BITS; |
6b6712ef JQ |
2452 | block->bmap = bitmap_new(pages); |
2453 | bitmap_set(block->bmap, 0, pages); | |
2454 | if (migrate_postcopy_ram()) { | |
2455 | block->unsentmap = bitmap_new(pages); | |
2456 | bitmap_set(block->unsentmap, 0, pages); | |
2457 | } | |
0827b9e9 | 2458 | } |
f3f491fc | 2459 | } |
d6eff5d7 PX |
2460 | } |
2461 | ||
2462 | static void ram_init_bitmaps(RAMState *rs) | |
2463 | { | |
2464 | /* For memory_global_dirty_log_start below. */ | |
2465 | qemu_mutex_lock_iothread(); | |
2466 | qemu_mutex_lock_ramlist(); | |
2467 | rcu_read_lock(); | |
f3f491fc | 2468 | |
d6eff5d7 | 2469 | ram_list_init_bitmaps(); |
56e93d26 | 2470 | memory_global_dirty_log_start(); |
d6eff5d7 PX |
2471 | migration_bitmap_sync(rs); |
2472 | ||
2473 | rcu_read_unlock(); | |
56e93d26 | 2474 | qemu_mutex_unlock_ramlist(); |
49877834 | 2475 | qemu_mutex_unlock_iothread(); |
d6eff5d7 PX |
2476 | } |
2477 | ||
2478 | static int ram_init_all(RAMState **rsp) | |
2479 | { | |
2480 | if (ram_state_init(rsp)) { | |
2481 | return -1; | |
2482 | } | |
2483 | ||
2484 | if (xbzrle_init()) { | |
2485 | ram_state_cleanup(rsp); | |
2486 | return -1; | |
2487 | } | |
2488 | ||
2489 | ram_init_bitmaps(*rsp); | |
a91246c9 HZ |
2490 | |
2491 | return 0; | |
2492 | } | |
2493 | ||
08614f34 PX |
2494 | static void ram_state_resume_prepare(RAMState *rs, QEMUFile *out) |
2495 | { | |
2496 | RAMBlock *block; | |
2497 | uint64_t pages = 0; | |
2498 | ||
2499 | /* | |
2500 | * Postcopy is not using xbzrle/compression, so no need for that. | |
2501 | * Also, since source are already halted, we don't need to care | |
2502 | * about dirty page logging as well. | |
2503 | */ | |
2504 | ||
2505 | RAMBLOCK_FOREACH(block) { | |
2506 | pages += bitmap_count_one(block->bmap, | |
2507 | block->used_length >> TARGET_PAGE_BITS); | |
2508 | } | |
2509 | ||
2510 | /* This may not be aligned with current bitmaps. Recalculate. */ | |
2511 | rs->migration_dirty_pages = pages; | |
2512 | ||
2513 | rs->last_seen_block = NULL; | |
2514 | rs->last_sent_block = NULL; | |
2515 | rs->last_page = 0; | |
2516 | rs->last_version = ram_list.version; | |
2517 | /* | |
2518 | * Disable the bulk stage, otherwise we'll resend the whole RAM no | |
2519 | * matter what we have sent. | |
2520 | */ | |
2521 | rs->ram_bulk_stage = false; | |
2522 | ||
2523 | /* Update RAMState cache of output QEMUFile */ | |
2524 | rs->f = out; | |
2525 | ||
2526 | trace_ram_state_resume_prepare(pages); | |
2527 | } | |
2528 | ||
3d0684b2 JQ |
2529 | /* |
2530 | * Each of ram_save_setup, ram_save_iterate and ram_save_complete has | |
a91246c9 HZ |
2531 | * long-running RCU critical section. When rcu-reclaims in the code |
2532 | * start to become numerous it will be necessary to reduce the | |
2533 | * granularity of these critical sections. | |
2534 | */ | |
2535 | ||
3d0684b2 JQ |
2536 | /** |
2537 | * ram_save_setup: Setup RAM for migration | |
2538 | * | |
2539 | * Returns zero to indicate success and negative for error | |
2540 | * | |
2541 | * @f: QEMUFile where to send the data | |
2542 | * @opaque: RAMState pointer | |
2543 | */ | |
a91246c9 HZ |
2544 | static int ram_save_setup(QEMUFile *f, void *opaque) |
2545 | { | |
53518d94 | 2546 | RAMState **rsp = opaque; |
a91246c9 HZ |
2547 | RAMBlock *block; |
2548 | ||
dcaf446e XG |
2549 | if (compress_threads_save_setup()) { |
2550 | return -1; | |
2551 | } | |
2552 | ||
a91246c9 HZ |
2553 | /* migration has already setup the bitmap, reuse it. */ |
2554 | if (!migration_in_colo_state()) { | |
7d00ee6a | 2555 | if (ram_init_all(rsp) != 0) { |
dcaf446e | 2556 | compress_threads_save_cleanup(); |
a91246c9 | 2557 | return -1; |
53518d94 | 2558 | } |
a91246c9 | 2559 | } |
53518d94 | 2560 | (*rsp)->f = f; |
a91246c9 HZ |
2561 | |
2562 | rcu_read_lock(); | |
56e93d26 JQ |
2563 | |
2564 | qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE); | |
2565 | ||
99e15582 | 2566 | RAMBLOCK_FOREACH(block) { |
56e93d26 JQ |
2567 | qemu_put_byte(f, strlen(block->idstr)); |
2568 | qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); | |
2569 | qemu_put_be64(f, block->used_length); | |
ef08fb38 DDAG |
2570 | if (migrate_postcopy_ram() && block->page_size != qemu_host_page_size) { |
2571 | qemu_put_be64(f, block->page_size); | |
2572 | } | |
56e93d26 JQ |
2573 | } |
2574 | ||
2575 | rcu_read_unlock(); | |
2576 | ||
2577 | ram_control_before_iterate(f, RAM_CONTROL_SETUP); | |
2578 | ram_control_after_iterate(f, RAM_CONTROL_SETUP); | |
2579 | ||
2580 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); | |
2581 | ||
2582 | return 0; | |
2583 | } | |
2584 | ||
3d0684b2 JQ |
2585 | /** |
2586 | * ram_save_iterate: iterative stage for migration | |
2587 | * | |
2588 | * Returns zero to indicate success and negative for error | |
2589 | * | |
2590 | * @f: QEMUFile where to send the data | |
2591 | * @opaque: RAMState pointer | |
2592 | */ | |
56e93d26 JQ |
2593 | static int ram_save_iterate(QEMUFile *f, void *opaque) |
2594 | { | |
53518d94 JQ |
2595 | RAMState **temp = opaque; |
2596 | RAMState *rs = *temp; | |
56e93d26 JQ |
2597 | int ret; |
2598 | int i; | |
2599 | int64_t t0; | |
5c90308f | 2600 | int done = 0; |
56e93d26 | 2601 | |
b2557345 PL |
2602 | if (blk_mig_bulk_active()) { |
2603 | /* Avoid transferring ram during bulk phase of block migration as | |
2604 | * the bulk phase will usually take a long time and transferring | |
2605 | * ram updates during that time is pointless. */ | |
2606 | goto out; | |
2607 | } | |
2608 | ||
56e93d26 | 2609 | rcu_read_lock(); |
6f37bb8b JQ |
2610 | if (ram_list.version != rs->last_version) { |
2611 | ram_state_reset(rs); | |
56e93d26 JQ |
2612 | } |
2613 | ||
2614 | /* Read version before ram_list.blocks */ | |
2615 | smp_rmb(); | |
2616 | ||
2617 | ram_control_before_iterate(f, RAM_CONTROL_ROUND); | |
2618 | ||
2619 | t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
2620 | i = 0; | |
2621 | while ((ret = qemu_file_rate_limit(f)) == 0) { | |
2622 | int pages; | |
2623 | ||
ce25d337 | 2624 | pages = ram_find_and_save_block(rs, false); |
56e93d26 JQ |
2625 | /* no more pages to sent */ |
2626 | if (pages == 0) { | |
5c90308f | 2627 | done = 1; |
56e93d26 JQ |
2628 | break; |
2629 | } | |
23b28c3c | 2630 | rs->iterations++; |
070afca2 | 2631 | |
56e93d26 JQ |
2632 | /* we want to check in the 1st loop, just in case it was the 1st time |
2633 | and we had to sync the dirty bitmap. | |
2634 | qemu_get_clock_ns() is a bit expensive, so we only check each some | |
2635 | iterations | |
2636 | */ | |
2637 | if ((i & 63) == 0) { | |
2638 | uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000; | |
2639 | if (t1 > MAX_WAIT) { | |
55c4446b | 2640 | trace_ram_save_iterate_big_wait(t1, i); |
56e93d26 JQ |
2641 | break; |
2642 | } | |
2643 | } | |
2644 | i++; | |
2645 | } | |
ce25d337 | 2646 | flush_compressed_data(rs); |
56e93d26 JQ |
2647 | rcu_read_unlock(); |
2648 | ||
2649 | /* | |
2650 | * Must occur before EOS (or any QEMUFile operation) | |
2651 | * because of RDMA protocol. | |
2652 | */ | |
2653 | ram_control_after_iterate(f, RAM_CONTROL_ROUND); | |
2654 | ||
b2557345 | 2655 | out: |
56e93d26 | 2656 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
9360447d | 2657 | ram_counters.transferred += 8; |
56e93d26 JQ |
2658 | |
2659 | ret = qemu_file_get_error(f); | |
2660 | if (ret < 0) { | |
2661 | return ret; | |
2662 | } | |
2663 | ||
5c90308f | 2664 | return done; |
56e93d26 JQ |
2665 | } |
2666 | ||
3d0684b2 JQ |
2667 | /** |
2668 | * ram_save_complete: function called to send the remaining amount of ram | |
2669 | * | |
2670 | * Returns zero to indicate success | |
2671 | * | |
2672 | * Called with iothread lock | |
2673 | * | |
2674 | * @f: QEMUFile where to send the data | |
2675 | * @opaque: RAMState pointer | |
2676 | */ | |
56e93d26 JQ |
2677 | static int ram_save_complete(QEMUFile *f, void *opaque) |
2678 | { | |
53518d94 JQ |
2679 | RAMState **temp = opaque; |
2680 | RAMState *rs = *temp; | |
6f37bb8b | 2681 | |
56e93d26 JQ |
2682 | rcu_read_lock(); |
2683 | ||
5727309d | 2684 | if (!migration_in_postcopy()) { |
8d820d6f | 2685 | migration_bitmap_sync(rs); |
663e6c1d | 2686 | } |
56e93d26 JQ |
2687 | |
2688 | ram_control_before_iterate(f, RAM_CONTROL_FINISH); | |
2689 | ||
2690 | /* try transferring iterative blocks of memory */ | |
2691 | ||
2692 | /* flush all remaining blocks regardless of rate limiting */ | |
2693 | while (true) { | |
2694 | int pages; | |
2695 | ||
ce25d337 | 2696 | pages = ram_find_and_save_block(rs, !migration_in_colo_state()); |
56e93d26 JQ |
2697 | /* no more blocks to sent */ |
2698 | if (pages == 0) { | |
2699 | break; | |
2700 | } | |
2701 | } | |
2702 | ||
ce25d337 | 2703 | flush_compressed_data(rs); |
56e93d26 | 2704 | ram_control_after_iterate(f, RAM_CONTROL_FINISH); |
56e93d26 JQ |
2705 | |
2706 | rcu_read_unlock(); | |
d09a6fde | 2707 | |
56e93d26 JQ |
2708 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
2709 | ||
2710 | return 0; | |
2711 | } | |
2712 | ||
c31b098f | 2713 | static void ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size, |
47995026 VSO |
2714 | uint64_t *res_precopy_only, |
2715 | uint64_t *res_compatible, | |
2716 | uint64_t *res_postcopy_only) | |
56e93d26 | 2717 | { |
53518d94 JQ |
2718 | RAMState **temp = opaque; |
2719 | RAMState *rs = *temp; | |
56e93d26 JQ |
2720 | uint64_t remaining_size; |
2721 | ||
9edabd4d | 2722 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2723 | |
5727309d | 2724 | if (!migration_in_postcopy() && |
663e6c1d | 2725 | remaining_size < max_size) { |
56e93d26 JQ |
2726 | qemu_mutex_lock_iothread(); |
2727 | rcu_read_lock(); | |
8d820d6f | 2728 | migration_bitmap_sync(rs); |
56e93d26 JQ |
2729 | rcu_read_unlock(); |
2730 | qemu_mutex_unlock_iothread(); | |
9edabd4d | 2731 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2732 | } |
c31b098f | 2733 | |
86e1167e VSO |
2734 | if (migrate_postcopy_ram()) { |
2735 | /* We can do postcopy, and all the data is postcopiable */ | |
47995026 | 2736 | *res_compatible += remaining_size; |
86e1167e | 2737 | } else { |
47995026 | 2738 | *res_precopy_only += remaining_size; |
86e1167e | 2739 | } |
56e93d26 JQ |
2740 | } |
2741 | ||
2742 | static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) | |
2743 | { | |
2744 | unsigned int xh_len; | |
2745 | int xh_flags; | |
063e760a | 2746 | uint8_t *loaded_data; |
56e93d26 | 2747 | |
56e93d26 JQ |
2748 | /* extract RLE header */ |
2749 | xh_flags = qemu_get_byte(f); | |
2750 | xh_len = qemu_get_be16(f); | |
2751 | ||
2752 | if (xh_flags != ENCODING_FLAG_XBZRLE) { | |
2753 | error_report("Failed to load XBZRLE page - wrong compression!"); | |
2754 | return -1; | |
2755 | } | |
2756 | ||
2757 | if (xh_len > TARGET_PAGE_SIZE) { | |
2758 | error_report("Failed to load XBZRLE page - len overflow!"); | |
2759 | return -1; | |
2760 | } | |
f265e0e4 | 2761 | loaded_data = XBZRLE.decoded_buf; |
56e93d26 | 2762 | /* load data and decode */ |
f265e0e4 | 2763 | /* it can change loaded_data to point to an internal buffer */ |
063e760a | 2764 | qemu_get_buffer_in_place(f, &loaded_data, xh_len); |
56e93d26 JQ |
2765 | |
2766 | /* decode RLE */ | |
063e760a | 2767 | if (xbzrle_decode_buffer(loaded_data, xh_len, host, |
56e93d26 JQ |
2768 | TARGET_PAGE_SIZE) == -1) { |
2769 | error_report("Failed to load XBZRLE page - decode error!"); | |
2770 | return -1; | |
2771 | } | |
2772 | ||
2773 | return 0; | |
2774 | } | |
2775 | ||
3d0684b2 JQ |
2776 | /** |
2777 | * ram_block_from_stream: read a RAMBlock id from the migration stream | |
2778 | * | |
2779 | * Must be called from within a rcu critical section. | |
2780 | * | |
56e93d26 | 2781 | * Returns a pointer from within the RCU-protected ram_list. |
a7180877 | 2782 | * |
3d0684b2 JQ |
2783 | * @f: QEMUFile where to read the data from |
2784 | * @flags: Page flags (mostly to see if it's a continuation of previous block) | |
a7180877 | 2785 | */ |
3d0684b2 | 2786 | static inline RAMBlock *ram_block_from_stream(QEMUFile *f, int flags) |
56e93d26 JQ |
2787 | { |
2788 | static RAMBlock *block = NULL; | |
2789 | char id[256]; | |
2790 | uint8_t len; | |
2791 | ||
2792 | if (flags & RAM_SAVE_FLAG_CONTINUE) { | |
4c4bad48 | 2793 | if (!block) { |
56e93d26 JQ |
2794 | error_report("Ack, bad migration stream!"); |
2795 | return NULL; | |
2796 | } | |
4c4bad48 | 2797 | return block; |
56e93d26 JQ |
2798 | } |
2799 | ||
2800 | len = qemu_get_byte(f); | |
2801 | qemu_get_buffer(f, (uint8_t *)id, len); | |
2802 | id[len] = 0; | |
2803 | ||
e3dd7493 | 2804 | block = qemu_ram_block_by_name(id); |
4c4bad48 HZ |
2805 | if (!block) { |
2806 | error_report("Can't find block %s", id); | |
2807 | return NULL; | |
56e93d26 JQ |
2808 | } |
2809 | ||
4c4bad48 HZ |
2810 | return block; |
2811 | } | |
2812 | ||
2813 | static inline void *host_from_ram_block_offset(RAMBlock *block, | |
2814 | ram_addr_t offset) | |
2815 | { | |
2816 | if (!offset_in_ramblock(block, offset)) { | |
2817 | return NULL; | |
2818 | } | |
2819 | ||
2820 | return block->host + offset; | |
56e93d26 JQ |
2821 | } |
2822 | ||
3d0684b2 JQ |
2823 | /** |
2824 | * ram_handle_compressed: handle the zero page case | |
2825 | * | |
56e93d26 JQ |
2826 | * If a page (or a whole RDMA chunk) has been |
2827 | * determined to be zero, then zap it. | |
3d0684b2 JQ |
2828 | * |
2829 | * @host: host address for the zero page | |
2830 | * @ch: what the page is filled from. We only support zero | |
2831 | * @size: size of the zero page | |
56e93d26 JQ |
2832 | */ |
2833 | void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) | |
2834 | { | |
2835 | if (ch != 0 || !is_zero_range(host, size)) { | |
2836 | memset(host, ch, size); | |
2837 | } | |
2838 | } | |
2839 | ||
797ca154 XG |
2840 | /* return the size after decompression, or negative value on error */ |
2841 | static int | |
2842 | qemu_uncompress_data(z_stream *stream, uint8_t *dest, size_t dest_len, | |
2843 | const uint8_t *source, size_t source_len) | |
2844 | { | |
2845 | int err; | |
2846 | ||
2847 | err = inflateReset(stream); | |
2848 | if (err != Z_OK) { | |
2849 | return -1; | |
2850 | } | |
2851 | ||
2852 | stream->avail_in = source_len; | |
2853 | stream->next_in = (uint8_t *)source; | |
2854 | stream->avail_out = dest_len; | |
2855 | stream->next_out = dest; | |
2856 | ||
2857 | err = inflate(stream, Z_NO_FLUSH); | |
2858 | if (err != Z_STREAM_END) { | |
2859 | return -1; | |
2860 | } | |
2861 | ||
2862 | return stream->total_out; | |
2863 | } | |
2864 | ||
56e93d26 JQ |
2865 | static void *do_data_decompress(void *opaque) |
2866 | { | |
2867 | DecompressParam *param = opaque; | |
2868 | unsigned long pagesize; | |
33d151f4 | 2869 | uint8_t *des; |
34ab9e97 | 2870 | int len, ret; |
56e93d26 | 2871 | |
33d151f4 | 2872 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 2873 | while (!param->quit) { |
33d151f4 LL |
2874 | if (param->des) { |
2875 | des = param->des; | |
2876 | len = param->len; | |
2877 | param->des = 0; | |
2878 | qemu_mutex_unlock(¶m->mutex); | |
2879 | ||
56e93d26 | 2880 | pagesize = TARGET_PAGE_SIZE; |
34ab9e97 XG |
2881 | |
2882 | ret = qemu_uncompress_data(¶m->stream, des, pagesize, | |
2883 | param->compbuf, len); | |
2884 | if (ret < 0) { | |
2885 | error_report("decompress data failed"); | |
2886 | qemu_file_set_error(decomp_file, ret); | |
2887 | } | |
73a8912b | 2888 | |
33d151f4 LL |
2889 | qemu_mutex_lock(&decomp_done_lock); |
2890 | param->done = true; | |
2891 | qemu_cond_signal(&decomp_done_cond); | |
2892 | qemu_mutex_unlock(&decomp_done_lock); | |
2893 | ||
2894 | qemu_mutex_lock(¶m->mutex); | |
2895 | } else { | |
2896 | qemu_cond_wait(¶m->cond, ¶m->mutex); | |
2897 | } | |
56e93d26 | 2898 | } |
33d151f4 | 2899 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
2900 | |
2901 | return NULL; | |
2902 | } | |
2903 | ||
34ab9e97 | 2904 | static int wait_for_decompress_done(void) |
5533b2e9 LL |
2905 | { |
2906 | int idx, thread_count; | |
2907 | ||
2908 | if (!migrate_use_compression()) { | |
34ab9e97 | 2909 | return 0; |
5533b2e9 LL |
2910 | } |
2911 | ||
2912 | thread_count = migrate_decompress_threads(); | |
2913 | qemu_mutex_lock(&decomp_done_lock); | |
2914 | for (idx = 0; idx < thread_count; idx++) { | |
2915 | while (!decomp_param[idx].done) { | |
2916 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
2917 | } | |
2918 | } | |
2919 | qemu_mutex_unlock(&decomp_done_lock); | |
34ab9e97 | 2920 | return qemu_file_get_error(decomp_file); |
5533b2e9 LL |
2921 | } |
2922 | ||
f0afa331 | 2923 | static void compress_threads_load_cleanup(void) |
56e93d26 JQ |
2924 | { |
2925 | int i, thread_count; | |
2926 | ||
3416ab5b JQ |
2927 | if (!migrate_use_compression()) { |
2928 | return; | |
2929 | } | |
56e93d26 JQ |
2930 | thread_count = migrate_decompress_threads(); |
2931 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
2932 | /* |
2933 | * we use it as a indicator which shows if the thread is | |
2934 | * properly init'd or not | |
2935 | */ | |
2936 | if (!decomp_param[i].compbuf) { | |
2937 | break; | |
2938 | } | |
2939 | ||
56e93d26 | 2940 | qemu_mutex_lock(&decomp_param[i].mutex); |
90e56fb4 | 2941 | decomp_param[i].quit = true; |
56e93d26 JQ |
2942 | qemu_cond_signal(&decomp_param[i].cond); |
2943 | qemu_mutex_unlock(&decomp_param[i].mutex); | |
2944 | } | |
2945 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
2946 | if (!decomp_param[i].compbuf) { |
2947 | break; | |
2948 | } | |
2949 | ||
56e93d26 JQ |
2950 | qemu_thread_join(decompress_threads + i); |
2951 | qemu_mutex_destroy(&decomp_param[i].mutex); | |
2952 | qemu_cond_destroy(&decomp_param[i].cond); | |
797ca154 | 2953 | inflateEnd(&decomp_param[i].stream); |
56e93d26 | 2954 | g_free(decomp_param[i].compbuf); |
797ca154 | 2955 | decomp_param[i].compbuf = NULL; |
56e93d26 JQ |
2956 | } |
2957 | g_free(decompress_threads); | |
2958 | g_free(decomp_param); | |
56e93d26 JQ |
2959 | decompress_threads = NULL; |
2960 | decomp_param = NULL; | |
34ab9e97 | 2961 | decomp_file = NULL; |
56e93d26 JQ |
2962 | } |
2963 | ||
34ab9e97 | 2964 | static int compress_threads_load_setup(QEMUFile *f) |
797ca154 XG |
2965 | { |
2966 | int i, thread_count; | |
2967 | ||
2968 | if (!migrate_use_compression()) { | |
2969 | return 0; | |
2970 | } | |
2971 | ||
2972 | thread_count = migrate_decompress_threads(); | |
2973 | decompress_threads = g_new0(QemuThread, thread_count); | |
2974 | decomp_param = g_new0(DecompressParam, thread_count); | |
2975 | qemu_mutex_init(&decomp_done_lock); | |
2976 | qemu_cond_init(&decomp_done_cond); | |
34ab9e97 | 2977 | decomp_file = f; |
797ca154 XG |
2978 | for (i = 0; i < thread_count; i++) { |
2979 | if (inflateInit(&decomp_param[i].stream) != Z_OK) { | |
2980 | goto exit; | |
2981 | } | |
2982 | ||
2983 | decomp_param[i].compbuf = g_malloc0(compressBound(TARGET_PAGE_SIZE)); | |
2984 | qemu_mutex_init(&decomp_param[i].mutex); | |
2985 | qemu_cond_init(&decomp_param[i].cond); | |
2986 | decomp_param[i].done = true; | |
2987 | decomp_param[i].quit = false; | |
2988 | qemu_thread_create(decompress_threads + i, "decompress", | |
2989 | do_data_decompress, decomp_param + i, | |
2990 | QEMU_THREAD_JOINABLE); | |
2991 | } | |
2992 | return 0; | |
2993 | exit: | |
2994 | compress_threads_load_cleanup(); | |
2995 | return -1; | |
2996 | } | |
2997 | ||
c1bc6626 | 2998 | static void decompress_data_with_multi_threads(QEMUFile *f, |
56e93d26 JQ |
2999 | void *host, int len) |
3000 | { | |
3001 | int idx, thread_count; | |
3002 | ||
3003 | thread_count = migrate_decompress_threads(); | |
73a8912b | 3004 | qemu_mutex_lock(&decomp_done_lock); |
56e93d26 JQ |
3005 | while (true) { |
3006 | for (idx = 0; idx < thread_count; idx++) { | |
73a8912b | 3007 | if (decomp_param[idx].done) { |
33d151f4 LL |
3008 | decomp_param[idx].done = false; |
3009 | qemu_mutex_lock(&decomp_param[idx].mutex); | |
c1bc6626 | 3010 | qemu_get_buffer(f, decomp_param[idx].compbuf, len); |
56e93d26 JQ |
3011 | decomp_param[idx].des = host; |
3012 | decomp_param[idx].len = len; | |
33d151f4 LL |
3013 | qemu_cond_signal(&decomp_param[idx].cond); |
3014 | qemu_mutex_unlock(&decomp_param[idx].mutex); | |
56e93d26 JQ |
3015 | break; |
3016 | } | |
3017 | } | |
3018 | if (idx < thread_count) { | |
3019 | break; | |
73a8912b LL |
3020 | } else { |
3021 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
56e93d26 JQ |
3022 | } |
3023 | } | |
73a8912b | 3024 | qemu_mutex_unlock(&decomp_done_lock); |
56e93d26 JQ |
3025 | } |
3026 | ||
f265e0e4 JQ |
3027 | /** |
3028 | * ram_load_setup: Setup RAM for migration incoming side | |
3029 | * | |
3030 | * Returns zero to indicate success and negative for error | |
3031 | * | |
3032 | * @f: QEMUFile where to receive the data | |
3033 | * @opaque: RAMState pointer | |
3034 | */ | |
3035 | static int ram_load_setup(QEMUFile *f, void *opaque) | |
3036 | { | |
34ab9e97 | 3037 | if (compress_threads_load_setup(f)) { |
797ca154 XG |
3038 | return -1; |
3039 | } | |
3040 | ||
f265e0e4 | 3041 | xbzrle_load_setup(); |
f9494614 | 3042 | ramblock_recv_map_init(); |
f265e0e4 JQ |
3043 | return 0; |
3044 | } | |
3045 | ||
3046 | static int ram_load_cleanup(void *opaque) | |
3047 | { | |
f9494614 | 3048 | RAMBlock *rb; |
f265e0e4 | 3049 | xbzrle_load_cleanup(); |
f0afa331 | 3050 | compress_threads_load_cleanup(); |
f9494614 AP |
3051 | |
3052 | RAMBLOCK_FOREACH(rb) { | |
3053 | g_free(rb->receivedmap); | |
3054 | rb->receivedmap = NULL; | |
3055 | } | |
f265e0e4 JQ |
3056 | return 0; |
3057 | } | |
3058 | ||
3d0684b2 JQ |
3059 | /** |
3060 | * ram_postcopy_incoming_init: allocate postcopy data structures | |
3061 | * | |
3062 | * Returns 0 for success and negative if there was one error | |
3063 | * | |
3064 | * @mis: current migration incoming state | |
3065 | * | |
3066 | * Allocate data structures etc needed by incoming migration with | |
3067 | * postcopy-ram. postcopy-ram's similarly names | |
3068 | * postcopy_ram_incoming_init does the work. | |
1caddf8a DDAG |
3069 | */ |
3070 | int ram_postcopy_incoming_init(MigrationIncomingState *mis) | |
3071 | { | |
b8c48993 | 3072 | unsigned long ram_pages = last_ram_page(); |
1caddf8a DDAG |
3073 | |
3074 | return postcopy_ram_incoming_init(mis, ram_pages); | |
3075 | } | |
3076 | ||
3d0684b2 JQ |
3077 | /** |
3078 | * ram_load_postcopy: load a page in postcopy case | |
3079 | * | |
3080 | * Returns 0 for success or -errno in case of error | |
3081 | * | |
a7180877 DDAG |
3082 | * Called in postcopy mode by ram_load(). |
3083 | * rcu_read_lock is taken prior to this being called. | |
3d0684b2 JQ |
3084 | * |
3085 | * @f: QEMUFile where to send the data | |
a7180877 DDAG |
3086 | */ |
3087 | static int ram_load_postcopy(QEMUFile *f) | |
3088 | { | |
3089 | int flags = 0, ret = 0; | |
3090 | bool place_needed = false; | |
28abd200 | 3091 | bool matching_page_sizes = false; |
a7180877 DDAG |
3092 | MigrationIncomingState *mis = migration_incoming_get_current(); |
3093 | /* Temporary page that is later 'placed' */ | |
3094 | void *postcopy_host_page = postcopy_get_tmp_page(mis); | |
c53b7ddc | 3095 | void *last_host = NULL; |
a3b6ff6d | 3096 | bool all_zero = false; |
a7180877 DDAG |
3097 | |
3098 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
3099 | ram_addr_t addr; | |
3100 | void *host = NULL; | |
3101 | void *page_buffer = NULL; | |
3102 | void *place_source = NULL; | |
df9ff5e1 | 3103 | RAMBlock *block = NULL; |
a7180877 | 3104 | uint8_t ch; |
a7180877 DDAG |
3105 | |
3106 | addr = qemu_get_be64(f); | |
7a9ddfbf PX |
3107 | |
3108 | /* | |
3109 | * If qemu file error, we should stop here, and then "addr" | |
3110 | * may be invalid | |
3111 | */ | |
3112 | ret = qemu_file_get_error(f); | |
3113 | if (ret) { | |
3114 | break; | |
3115 | } | |
3116 | ||
a7180877 DDAG |
3117 | flags = addr & ~TARGET_PAGE_MASK; |
3118 | addr &= TARGET_PAGE_MASK; | |
3119 | ||
3120 | trace_ram_load_postcopy_loop((uint64_t)addr, flags); | |
3121 | place_needed = false; | |
bb890ed5 | 3122 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE)) { |
df9ff5e1 | 3123 | block = ram_block_from_stream(f, flags); |
4c4bad48 HZ |
3124 | |
3125 | host = host_from_ram_block_offset(block, addr); | |
a7180877 DDAG |
3126 | if (!host) { |
3127 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
3128 | ret = -EINVAL; | |
3129 | break; | |
3130 | } | |
28abd200 | 3131 | matching_page_sizes = block->page_size == TARGET_PAGE_SIZE; |
a7180877 | 3132 | /* |
28abd200 DDAG |
3133 | * Postcopy requires that we place whole host pages atomically; |
3134 | * these may be huge pages for RAMBlocks that are backed by | |
3135 | * hugetlbfs. | |
a7180877 DDAG |
3136 | * To make it atomic, the data is read into a temporary page |
3137 | * that's moved into place later. | |
3138 | * The migration protocol uses, possibly smaller, target-pages | |
3139 | * however the source ensures it always sends all the components | |
3140 | * of a host page in order. | |
3141 | */ | |
3142 | page_buffer = postcopy_host_page + | |
28abd200 | 3143 | ((uintptr_t)host & (block->page_size - 1)); |
a7180877 | 3144 | /* If all TP are zero then we can optimise the place */ |
28abd200 | 3145 | if (!((uintptr_t)host & (block->page_size - 1))) { |
a7180877 | 3146 | all_zero = true; |
c53b7ddc DDAG |
3147 | } else { |
3148 | /* not the 1st TP within the HP */ | |
3149 | if (host != (last_host + TARGET_PAGE_SIZE)) { | |
9af9e0fe | 3150 | error_report("Non-sequential target page %p/%p", |
c53b7ddc DDAG |
3151 | host, last_host); |
3152 | ret = -EINVAL; | |
3153 | break; | |
3154 | } | |
a7180877 DDAG |
3155 | } |
3156 | ||
c53b7ddc | 3157 | |
a7180877 DDAG |
3158 | /* |
3159 | * If it's the last part of a host page then we place the host | |
3160 | * page | |
3161 | */ | |
3162 | place_needed = (((uintptr_t)host + TARGET_PAGE_SIZE) & | |
28abd200 | 3163 | (block->page_size - 1)) == 0; |
a7180877 DDAG |
3164 | place_source = postcopy_host_page; |
3165 | } | |
c53b7ddc | 3166 | last_host = host; |
a7180877 DDAG |
3167 | |
3168 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { | |
bb890ed5 | 3169 | case RAM_SAVE_FLAG_ZERO: |
a7180877 DDAG |
3170 | ch = qemu_get_byte(f); |
3171 | memset(page_buffer, ch, TARGET_PAGE_SIZE); | |
3172 | if (ch) { | |
3173 | all_zero = false; | |
3174 | } | |
3175 | break; | |
3176 | ||
3177 | case RAM_SAVE_FLAG_PAGE: | |
3178 | all_zero = false; | |
3179 | if (!place_needed || !matching_page_sizes) { | |
3180 | qemu_get_buffer(f, page_buffer, TARGET_PAGE_SIZE); | |
3181 | } else { | |
3182 | /* Avoids the qemu_file copy during postcopy, which is | |
3183 | * going to do a copy later; can only do it when we | |
3184 | * do this read in one go (matching page sizes) | |
3185 | */ | |
3186 | qemu_get_buffer_in_place(f, (uint8_t **)&place_source, | |
3187 | TARGET_PAGE_SIZE); | |
3188 | } | |
3189 | break; | |
3190 | case RAM_SAVE_FLAG_EOS: | |
3191 | /* normal exit */ | |
3192 | break; | |
3193 | default: | |
3194 | error_report("Unknown combination of migration flags: %#x" | |
3195 | " (postcopy mode)", flags); | |
3196 | ret = -EINVAL; | |
7a9ddfbf PX |
3197 | break; |
3198 | } | |
3199 | ||
3200 | /* Detect for any possible file errors */ | |
3201 | if (!ret && qemu_file_get_error(f)) { | |
3202 | ret = qemu_file_get_error(f); | |
a7180877 DDAG |
3203 | } |
3204 | ||
7a9ddfbf | 3205 | if (!ret && place_needed) { |
a7180877 | 3206 | /* This gets called at the last target page in the host page */ |
df9ff5e1 DDAG |
3207 | void *place_dest = host + TARGET_PAGE_SIZE - block->page_size; |
3208 | ||
a7180877 | 3209 | if (all_zero) { |
df9ff5e1 | 3210 | ret = postcopy_place_page_zero(mis, place_dest, |
8be4620b | 3211 | block); |
a7180877 | 3212 | } else { |
df9ff5e1 | 3213 | ret = postcopy_place_page(mis, place_dest, |
8be4620b | 3214 | place_source, block); |
a7180877 DDAG |
3215 | } |
3216 | } | |
a7180877 DDAG |
3217 | } |
3218 | ||
3219 | return ret; | |
3220 | } | |
3221 | ||
acab30b8 DHB |
3222 | static bool postcopy_is_advised(void) |
3223 | { | |
3224 | PostcopyState ps = postcopy_state_get(); | |
3225 | return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END; | |
3226 | } | |
3227 | ||
3228 | static bool postcopy_is_running(void) | |
3229 | { | |
3230 | PostcopyState ps = postcopy_state_get(); | |
3231 | return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; | |
3232 | } | |
3233 | ||
56e93d26 JQ |
3234 | static int ram_load(QEMUFile *f, void *opaque, int version_id) |
3235 | { | |
edc60127 | 3236 | int flags = 0, ret = 0, invalid_flags = 0; |
56e93d26 JQ |
3237 | static uint64_t seq_iter; |
3238 | int len = 0; | |
a7180877 DDAG |
3239 | /* |
3240 | * If system is running in postcopy mode, page inserts to host memory must | |
3241 | * be atomic | |
3242 | */ | |
acab30b8 | 3243 | bool postcopy_running = postcopy_is_running(); |
ef08fb38 | 3244 | /* ADVISE is earlier, it shows the source has the postcopy capability on */ |
acab30b8 | 3245 | bool postcopy_advised = postcopy_is_advised(); |
56e93d26 JQ |
3246 | |
3247 | seq_iter++; | |
3248 | ||
3249 | if (version_id != 4) { | |
3250 | ret = -EINVAL; | |
3251 | } | |
3252 | ||
edc60127 JQ |
3253 | if (!migrate_use_compression()) { |
3254 | invalid_flags |= RAM_SAVE_FLAG_COMPRESS_PAGE; | |
3255 | } | |
56e93d26 JQ |
3256 | /* This RCU critical section can be very long running. |
3257 | * When RCU reclaims in the code start to become numerous, | |
3258 | * it will be necessary to reduce the granularity of this | |
3259 | * critical section. | |
3260 | */ | |
3261 | rcu_read_lock(); | |
a7180877 DDAG |
3262 | |
3263 | if (postcopy_running) { | |
3264 | ret = ram_load_postcopy(f); | |
3265 | } | |
3266 | ||
3267 | while (!postcopy_running && !ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
56e93d26 | 3268 | ram_addr_t addr, total_ram_bytes; |
a776aa15 | 3269 | void *host = NULL; |
56e93d26 JQ |
3270 | uint8_t ch; |
3271 | ||
3272 | addr = qemu_get_be64(f); | |
3273 | flags = addr & ~TARGET_PAGE_MASK; | |
3274 | addr &= TARGET_PAGE_MASK; | |
3275 | ||
edc60127 JQ |
3276 | if (flags & invalid_flags) { |
3277 | if (flags & invalid_flags & RAM_SAVE_FLAG_COMPRESS_PAGE) { | |
3278 | error_report("Received an unexpected compressed page"); | |
3279 | } | |
3280 | ||
3281 | ret = -EINVAL; | |
3282 | break; | |
3283 | } | |
3284 | ||
bb890ed5 | 3285 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
a776aa15 | 3286 | RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) { |
4c4bad48 HZ |
3287 | RAMBlock *block = ram_block_from_stream(f, flags); |
3288 | ||
3289 | host = host_from_ram_block_offset(block, addr); | |
a776aa15 DDAG |
3290 | if (!host) { |
3291 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
3292 | ret = -EINVAL; | |
3293 | break; | |
3294 | } | |
f9494614 | 3295 | ramblock_recv_bitmap_set(block, host); |
1db9d8e5 | 3296 | trace_ram_load_loop(block->idstr, (uint64_t)addr, flags, host); |
a776aa15 DDAG |
3297 | } |
3298 | ||
56e93d26 JQ |
3299 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { |
3300 | case RAM_SAVE_FLAG_MEM_SIZE: | |
3301 | /* Synchronize RAM block list */ | |
3302 | total_ram_bytes = addr; | |
3303 | while (!ret && total_ram_bytes) { | |
3304 | RAMBlock *block; | |
56e93d26 JQ |
3305 | char id[256]; |
3306 | ram_addr_t length; | |
3307 | ||
3308 | len = qemu_get_byte(f); | |
3309 | qemu_get_buffer(f, (uint8_t *)id, len); | |
3310 | id[len] = 0; | |
3311 | length = qemu_get_be64(f); | |
3312 | ||
e3dd7493 DDAG |
3313 | block = qemu_ram_block_by_name(id); |
3314 | if (block) { | |
3315 | if (length != block->used_length) { | |
3316 | Error *local_err = NULL; | |
56e93d26 | 3317 | |
fa53a0e5 | 3318 | ret = qemu_ram_resize(block, length, |
e3dd7493 DDAG |
3319 | &local_err); |
3320 | if (local_err) { | |
3321 | error_report_err(local_err); | |
56e93d26 | 3322 | } |
56e93d26 | 3323 | } |
ef08fb38 DDAG |
3324 | /* For postcopy we need to check hugepage sizes match */ |
3325 | if (postcopy_advised && | |
3326 | block->page_size != qemu_host_page_size) { | |
3327 | uint64_t remote_page_size = qemu_get_be64(f); | |
3328 | if (remote_page_size != block->page_size) { | |
3329 | error_report("Mismatched RAM page size %s " | |
3330 | "(local) %zd != %" PRId64, | |
3331 | id, block->page_size, | |
3332 | remote_page_size); | |
3333 | ret = -EINVAL; | |
3334 | } | |
3335 | } | |
e3dd7493 DDAG |
3336 | ram_control_load_hook(f, RAM_CONTROL_BLOCK_REG, |
3337 | block->idstr); | |
3338 | } else { | |
56e93d26 JQ |
3339 | error_report("Unknown ramblock \"%s\", cannot " |
3340 | "accept migration", id); | |
3341 | ret = -EINVAL; | |
3342 | } | |
3343 | ||
3344 | total_ram_bytes -= length; | |
3345 | } | |
3346 | break; | |
a776aa15 | 3347 | |
bb890ed5 | 3348 | case RAM_SAVE_FLAG_ZERO: |
56e93d26 JQ |
3349 | ch = qemu_get_byte(f); |
3350 | ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); | |
3351 | break; | |
a776aa15 | 3352 | |
56e93d26 | 3353 | case RAM_SAVE_FLAG_PAGE: |
56e93d26 JQ |
3354 | qemu_get_buffer(f, host, TARGET_PAGE_SIZE); |
3355 | break; | |
56e93d26 | 3356 | |
a776aa15 | 3357 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
56e93d26 JQ |
3358 | len = qemu_get_be32(f); |
3359 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
3360 | error_report("Invalid compressed data length: %d", len); | |
3361 | ret = -EINVAL; | |
3362 | break; | |
3363 | } | |
c1bc6626 | 3364 | decompress_data_with_multi_threads(f, host, len); |
56e93d26 | 3365 | break; |
a776aa15 | 3366 | |
56e93d26 | 3367 | case RAM_SAVE_FLAG_XBZRLE: |
56e93d26 JQ |
3368 | if (load_xbzrle(f, addr, host) < 0) { |
3369 | error_report("Failed to decompress XBZRLE page at " | |
3370 | RAM_ADDR_FMT, addr); | |
3371 | ret = -EINVAL; | |
3372 | break; | |
3373 | } | |
3374 | break; | |
3375 | case RAM_SAVE_FLAG_EOS: | |
3376 | /* normal exit */ | |
3377 | break; | |
3378 | default: | |
3379 | if (flags & RAM_SAVE_FLAG_HOOK) { | |
632e3a5c | 3380 | ram_control_load_hook(f, RAM_CONTROL_HOOK, NULL); |
56e93d26 JQ |
3381 | } else { |
3382 | error_report("Unknown combination of migration flags: %#x", | |
3383 | flags); | |
3384 | ret = -EINVAL; | |
3385 | } | |
3386 | } | |
3387 | if (!ret) { | |
3388 | ret = qemu_file_get_error(f); | |
3389 | } | |
3390 | } | |
3391 | ||
34ab9e97 | 3392 | ret |= wait_for_decompress_done(); |
56e93d26 | 3393 | rcu_read_unlock(); |
55c4446b | 3394 | trace_ram_load_complete(ret, seq_iter); |
56e93d26 JQ |
3395 | return ret; |
3396 | } | |
3397 | ||
c6467627 VSO |
3398 | static bool ram_has_postcopy(void *opaque) |
3399 | { | |
3400 | return migrate_postcopy_ram(); | |
3401 | } | |
3402 | ||
edd090c7 PX |
3403 | /* Sync all the dirty bitmap with destination VM. */ |
3404 | static int ram_dirty_bitmap_sync_all(MigrationState *s, RAMState *rs) | |
3405 | { | |
3406 | RAMBlock *block; | |
3407 | QEMUFile *file = s->to_dst_file; | |
3408 | int ramblock_count = 0; | |
3409 | ||
3410 | trace_ram_dirty_bitmap_sync_start(); | |
3411 | ||
3412 | RAMBLOCK_FOREACH(block) { | |
3413 | qemu_savevm_send_recv_bitmap(file, block->idstr); | |
3414 | trace_ram_dirty_bitmap_request(block->idstr); | |
3415 | ramblock_count++; | |
3416 | } | |
3417 | ||
3418 | trace_ram_dirty_bitmap_sync_wait(); | |
3419 | ||
3420 | /* Wait until all the ramblocks' dirty bitmap synced */ | |
3421 | while (ramblock_count--) { | |
3422 | qemu_sem_wait(&s->rp_state.rp_sem); | |
3423 | } | |
3424 | ||
3425 | trace_ram_dirty_bitmap_sync_complete(); | |
3426 | ||
3427 | return 0; | |
3428 | } | |
3429 | ||
3430 | static void ram_dirty_bitmap_reload_notify(MigrationState *s) | |
3431 | { | |
3432 | qemu_sem_post(&s->rp_state.rp_sem); | |
3433 | } | |
3434 | ||
a335debb PX |
3435 | /* |
3436 | * Read the received bitmap, revert it as the initial dirty bitmap. | |
3437 | * This is only used when the postcopy migration is paused but wants | |
3438 | * to resume from a middle point. | |
3439 | */ | |
3440 | int ram_dirty_bitmap_reload(MigrationState *s, RAMBlock *block) | |
3441 | { | |
3442 | int ret = -EINVAL; | |
3443 | QEMUFile *file = s->rp_state.from_dst_file; | |
3444 | unsigned long *le_bitmap, nbits = block->used_length >> TARGET_PAGE_BITS; | |
3445 | uint64_t local_size = nbits / 8; | |
3446 | uint64_t size, end_mark; | |
3447 | ||
3448 | trace_ram_dirty_bitmap_reload_begin(block->idstr); | |
3449 | ||
3450 | if (s->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
3451 | error_report("%s: incorrect state %s", __func__, | |
3452 | MigrationStatus_str(s->state)); | |
3453 | return -EINVAL; | |
3454 | } | |
3455 | ||
3456 | /* | |
3457 | * Note: see comments in ramblock_recv_bitmap_send() on why we | |
3458 | * need the endianess convertion, and the paddings. | |
3459 | */ | |
3460 | local_size = ROUND_UP(local_size, 8); | |
3461 | ||
3462 | /* Add paddings */ | |
3463 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
3464 | ||
3465 | size = qemu_get_be64(file); | |
3466 | ||
3467 | /* The size of the bitmap should match with our ramblock */ | |
3468 | if (size != local_size) { | |
3469 | error_report("%s: ramblock '%s' bitmap size mismatch " | |
3470 | "(0x%"PRIx64" != 0x%"PRIx64")", __func__, | |
3471 | block->idstr, size, local_size); | |
3472 | ret = -EINVAL; | |
3473 | goto out; | |
3474 | } | |
3475 | ||
3476 | size = qemu_get_buffer(file, (uint8_t *)le_bitmap, local_size); | |
3477 | end_mark = qemu_get_be64(file); | |
3478 | ||
3479 | ret = qemu_file_get_error(file); | |
3480 | if (ret || size != local_size) { | |
3481 | error_report("%s: read bitmap failed for ramblock '%s': %d" | |
3482 | " (size 0x%"PRIx64", got: 0x%"PRIx64")", | |
3483 | __func__, block->idstr, ret, local_size, size); | |
3484 | ret = -EIO; | |
3485 | goto out; | |
3486 | } | |
3487 | ||
3488 | if (end_mark != RAMBLOCK_RECV_BITMAP_ENDING) { | |
3489 | error_report("%s: ramblock '%s' end mark incorrect: 0x%"PRIu64, | |
3490 | __func__, block->idstr, end_mark); | |
3491 | ret = -EINVAL; | |
3492 | goto out; | |
3493 | } | |
3494 | ||
3495 | /* | |
3496 | * Endianess convertion. We are during postcopy (though paused). | |
3497 | * The dirty bitmap won't change. We can directly modify it. | |
3498 | */ | |
3499 | bitmap_from_le(block->bmap, le_bitmap, nbits); | |
3500 | ||
3501 | /* | |
3502 | * What we received is "received bitmap". Revert it as the initial | |
3503 | * dirty bitmap for this ramblock. | |
3504 | */ | |
3505 | bitmap_complement(block->bmap, block->bmap, nbits); | |
3506 | ||
3507 | trace_ram_dirty_bitmap_reload_complete(block->idstr); | |
3508 | ||
edd090c7 PX |
3509 | /* |
3510 | * We succeeded to sync bitmap for current ramblock. If this is | |
3511 | * the last one to sync, we need to notify the main send thread. | |
3512 | */ | |
3513 | ram_dirty_bitmap_reload_notify(s); | |
3514 | ||
a335debb PX |
3515 | ret = 0; |
3516 | out: | |
bf269906 | 3517 | g_free(le_bitmap); |
a335debb PX |
3518 | return ret; |
3519 | } | |
3520 | ||
edd090c7 PX |
3521 | static int ram_resume_prepare(MigrationState *s, void *opaque) |
3522 | { | |
3523 | RAMState *rs = *(RAMState **)opaque; | |
08614f34 | 3524 | int ret; |
edd090c7 | 3525 | |
08614f34 PX |
3526 | ret = ram_dirty_bitmap_sync_all(s, rs); |
3527 | if (ret) { | |
3528 | return ret; | |
3529 | } | |
3530 | ||
3531 | ram_state_resume_prepare(rs, s->to_dst_file); | |
3532 | ||
3533 | return 0; | |
edd090c7 PX |
3534 | } |
3535 | ||
56e93d26 | 3536 | static SaveVMHandlers savevm_ram_handlers = { |
9907e842 | 3537 | .save_setup = ram_save_setup, |
56e93d26 | 3538 | .save_live_iterate = ram_save_iterate, |
763c906b | 3539 | .save_live_complete_postcopy = ram_save_complete, |
a3e06c3d | 3540 | .save_live_complete_precopy = ram_save_complete, |
c6467627 | 3541 | .has_postcopy = ram_has_postcopy, |
56e93d26 JQ |
3542 | .save_live_pending = ram_save_pending, |
3543 | .load_state = ram_load, | |
f265e0e4 JQ |
3544 | .save_cleanup = ram_save_cleanup, |
3545 | .load_setup = ram_load_setup, | |
3546 | .load_cleanup = ram_load_cleanup, | |
edd090c7 | 3547 | .resume_prepare = ram_resume_prepare, |
56e93d26 JQ |
3548 | }; |
3549 | ||
3550 | void ram_mig_init(void) | |
3551 | { | |
3552 | qemu_mutex_init(&XBZRLE.lock); | |
6f37bb8b | 3553 | register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, &ram_state); |
56e93d26 | 3554 | } |