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