]>
Commit | Line | Data |
---|---|---|
56e93d26 JQ |
1 | /* |
2 | * QEMU System Emulator | |
3 | * | |
4 | * Copyright (c) 2003-2008 Fabrice Bellard | |
76cc7b58 JQ |
5 | * Copyright (c) 2011-2015 Red Hat Inc |
6 | * | |
7 | * Authors: | |
8 | * Juan Quintela <quintela@redhat.com> | |
56e93d26 JQ |
9 | * |
10 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
11 | * of this software and associated documentation files (the "Software"), to deal | |
12 | * in the Software without restriction, including without limitation the rights | |
13 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
14 | * copies of the Software, and to permit persons to whom the Software is | |
15 | * furnished to do so, subject to the following conditions: | |
16 | * | |
17 | * The above copyright notice and this permission notice shall be included in | |
18 | * all copies or substantial portions of the Software. | |
19 | * | |
20 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
21 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
22 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
23 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
24 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
25 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
26 | * THE SOFTWARE. | |
27 | */ | |
e688df6b | 28 | |
1393a485 | 29 | #include "qemu/osdep.h" |
33c11879 | 30 | #include "cpu.h" |
56e93d26 | 31 | #include <zlib.h> |
f348b6d1 | 32 | #include "qemu/cutils.h" |
56e93d26 JQ |
33 | #include "qemu/bitops.h" |
34 | #include "qemu/bitmap.h" | |
7205c9ec | 35 | #include "qemu/main-loop.h" |
709e3fe8 | 36 | #include "xbzrle.h" |
7b1e1a22 | 37 | #include "ram.h" |
6666c96a | 38 | #include "migration.h" |
71bb07db | 39 | #include "socket.h" |
f2a8f0a6 | 40 | #include "migration/register.h" |
7b1e1a22 | 41 | #include "migration/misc.h" |
08a0aee1 | 42 | #include "qemu-file.h" |
be07b0ac | 43 | #include "postcopy-ram.h" |
53d37d36 | 44 | #include "page_cache.h" |
56e93d26 | 45 | #include "qemu/error-report.h" |
e688df6b | 46 | #include "qapi/error.h" |
9af23989 | 47 | #include "qapi/qapi-events-migration.h" |
8acabf69 | 48 | #include "qapi/qmp/qerror.h" |
56e93d26 | 49 | #include "trace.h" |
56e93d26 | 50 | #include "exec/ram_addr.h" |
f9494614 | 51 | #include "exec/target_page.h" |
56e93d26 | 52 | #include "qemu/rcu_queue.h" |
a91246c9 | 53 | #include "migration/colo.h" |
53d37d36 | 54 | #include "block.h" |
af8b7d2b JQ |
55 | #include "sysemu/sysemu.h" |
56 | #include "qemu/uuid.h" | |
edd090c7 | 57 | #include "savevm.h" |
56e93d26 | 58 | |
56e93d26 JQ |
59 | /***********************************************************/ |
60 | /* ram save/restore */ | |
61 | ||
bb890ed5 JQ |
62 | /* RAM_SAVE_FLAG_ZERO used to be named RAM_SAVE_FLAG_COMPRESS, it |
63 | * worked for pages that where filled with the same char. We switched | |
64 | * it to only search for the zero value. And to avoid confusion with | |
65 | * RAM_SSAVE_FLAG_COMPRESS_PAGE just rename it. | |
66 | */ | |
67 | ||
56e93d26 | 68 | #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */ |
bb890ed5 | 69 | #define RAM_SAVE_FLAG_ZERO 0x02 |
56e93d26 JQ |
70 | #define RAM_SAVE_FLAG_MEM_SIZE 0x04 |
71 | #define RAM_SAVE_FLAG_PAGE 0x08 | |
72 | #define RAM_SAVE_FLAG_EOS 0x10 | |
73 | #define RAM_SAVE_FLAG_CONTINUE 0x20 | |
74 | #define RAM_SAVE_FLAG_XBZRLE 0x40 | |
75 | /* 0x80 is reserved in migration.h start with 0x100 next */ | |
76 | #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 | |
77 | ||
56e93d26 JQ |
78 | static inline bool is_zero_range(uint8_t *p, uint64_t size) |
79 | { | |
a1febc49 | 80 | return buffer_is_zero(p, size); |
56e93d26 JQ |
81 | } |
82 | ||
9360447d JQ |
83 | XBZRLECacheStats xbzrle_counters; |
84 | ||
56e93d26 JQ |
85 | /* struct contains XBZRLE cache and a static page |
86 | used by the compression */ | |
87 | static struct { | |
88 | /* buffer used for XBZRLE encoding */ | |
89 | uint8_t *encoded_buf; | |
90 | /* buffer for storing page content */ | |
91 | uint8_t *current_buf; | |
92 | /* Cache for XBZRLE, Protected by lock. */ | |
93 | PageCache *cache; | |
94 | QemuMutex lock; | |
c00e0928 JQ |
95 | /* it will store a page full of zeros */ |
96 | uint8_t *zero_target_page; | |
f265e0e4 JQ |
97 | /* buffer used for XBZRLE decoding */ |
98 | uint8_t *decoded_buf; | |
56e93d26 JQ |
99 | } XBZRLE; |
100 | ||
56e93d26 JQ |
101 | static void XBZRLE_cache_lock(void) |
102 | { | |
103 | if (migrate_use_xbzrle()) | |
104 | qemu_mutex_lock(&XBZRLE.lock); | |
105 | } | |
106 | ||
107 | static void XBZRLE_cache_unlock(void) | |
108 | { | |
109 | if (migrate_use_xbzrle()) | |
110 | qemu_mutex_unlock(&XBZRLE.lock); | |
111 | } | |
112 | ||
3d0684b2 JQ |
113 | /** |
114 | * xbzrle_cache_resize: resize the xbzrle cache | |
115 | * | |
116 | * This function is called from qmp_migrate_set_cache_size in main | |
117 | * thread, possibly while a migration is in progress. A running | |
118 | * migration may be using the cache and might finish during this call, | |
119 | * hence changes to the cache are protected by XBZRLE.lock(). | |
120 | * | |
c9dede2d | 121 | * Returns 0 for success or -1 for error |
3d0684b2 JQ |
122 | * |
123 | * @new_size: new cache size | |
8acabf69 | 124 | * @errp: set *errp if the check failed, with reason |
56e93d26 | 125 | */ |
c9dede2d | 126 | int xbzrle_cache_resize(int64_t new_size, Error **errp) |
56e93d26 JQ |
127 | { |
128 | PageCache *new_cache; | |
c9dede2d | 129 | int64_t ret = 0; |
56e93d26 | 130 | |
8acabf69 JQ |
131 | /* Check for truncation */ |
132 | if (new_size != (size_t)new_size) { | |
133 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", | |
134 | "exceeding address space"); | |
135 | return -1; | |
136 | } | |
137 | ||
2a313e5c JQ |
138 | if (new_size == migrate_xbzrle_cache_size()) { |
139 | /* nothing to do */ | |
c9dede2d | 140 | return 0; |
2a313e5c JQ |
141 | } |
142 | ||
56e93d26 JQ |
143 | XBZRLE_cache_lock(); |
144 | ||
145 | if (XBZRLE.cache != NULL) { | |
80f8dfde | 146 | new_cache = cache_init(new_size, TARGET_PAGE_SIZE, errp); |
56e93d26 | 147 | if (!new_cache) { |
56e93d26 JQ |
148 | ret = -1; |
149 | goto out; | |
150 | } | |
151 | ||
152 | cache_fini(XBZRLE.cache); | |
153 | XBZRLE.cache = new_cache; | |
154 | } | |
56e93d26 JQ |
155 | out: |
156 | XBZRLE_cache_unlock(); | |
157 | return ret; | |
158 | } | |
159 | ||
b895de50 CLG |
160 | /* Should be holding either ram_list.mutex, or the RCU lock. */ |
161 | #define RAMBLOCK_FOREACH_MIGRATABLE(block) \ | |
343f632c | 162 | INTERNAL_RAMBLOCK_FOREACH(block) \ |
b895de50 CLG |
163 | if (!qemu_ram_is_migratable(block)) {} else |
164 | ||
343f632c DDAG |
165 | #undef RAMBLOCK_FOREACH |
166 | ||
f9494614 AP |
167 | static void ramblock_recv_map_init(void) |
168 | { | |
169 | RAMBlock *rb; | |
170 | ||
b895de50 | 171 | RAMBLOCK_FOREACH_MIGRATABLE(rb) { |
f9494614 AP |
172 | assert(!rb->receivedmap); |
173 | rb->receivedmap = bitmap_new(rb->max_length >> qemu_target_page_bits()); | |
174 | } | |
175 | } | |
176 | ||
177 | int ramblock_recv_bitmap_test(RAMBlock *rb, void *host_addr) | |
178 | { | |
179 | return test_bit(ramblock_recv_bitmap_offset(host_addr, rb), | |
180 | rb->receivedmap); | |
181 | } | |
182 | ||
1cba9f6e DDAG |
183 | bool ramblock_recv_bitmap_test_byte_offset(RAMBlock *rb, uint64_t byte_offset) |
184 | { | |
185 | return test_bit(byte_offset >> TARGET_PAGE_BITS, rb->receivedmap); | |
186 | } | |
187 | ||
f9494614 AP |
188 | void ramblock_recv_bitmap_set(RAMBlock *rb, void *host_addr) |
189 | { | |
190 | set_bit_atomic(ramblock_recv_bitmap_offset(host_addr, rb), rb->receivedmap); | |
191 | } | |
192 | ||
193 | void ramblock_recv_bitmap_set_range(RAMBlock *rb, void *host_addr, | |
194 | size_t nr) | |
195 | { | |
196 | bitmap_set_atomic(rb->receivedmap, | |
197 | ramblock_recv_bitmap_offset(host_addr, rb), | |
198 | nr); | |
199 | } | |
200 | ||
a335debb PX |
201 | #define RAMBLOCK_RECV_BITMAP_ENDING (0x0123456789abcdefULL) |
202 | ||
203 | /* | |
204 | * Format: bitmap_size (8 bytes) + whole_bitmap (N bytes). | |
205 | * | |
206 | * Returns >0 if success with sent bytes, or <0 if error. | |
207 | */ | |
208 | int64_t ramblock_recv_bitmap_send(QEMUFile *file, | |
209 | const char *block_name) | |
210 | { | |
211 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
212 | unsigned long *le_bitmap, nbits; | |
213 | uint64_t size; | |
214 | ||
215 | if (!block) { | |
216 | error_report("%s: invalid block name: %s", __func__, block_name); | |
217 | return -1; | |
218 | } | |
219 | ||
220 | nbits = block->used_length >> TARGET_PAGE_BITS; | |
221 | ||
222 | /* | |
223 | * Make sure the tmp bitmap buffer is big enough, e.g., on 32bit | |
224 | * machines we may need 4 more bytes for padding (see below | |
225 | * comment). So extend it a bit before hand. | |
226 | */ | |
227 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
228 | ||
229 | /* | |
230 | * Always use little endian when sending the bitmap. This is | |
231 | * required that when source and destination VMs are not using the | |
232 | * same endianess. (Note: big endian won't work.) | |
233 | */ | |
234 | bitmap_to_le(le_bitmap, block->receivedmap, nbits); | |
235 | ||
236 | /* Size of the bitmap, in bytes */ | |
237 | size = nbits / 8; | |
238 | ||
239 | /* | |
240 | * size is always aligned to 8 bytes for 64bit machines, but it | |
241 | * may not be true for 32bit machines. We need this padding to | |
242 | * make sure the migration can survive even between 32bit and | |
243 | * 64bit machines. | |
244 | */ | |
245 | size = ROUND_UP(size, 8); | |
246 | ||
247 | qemu_put_be64(file, size); | |
248 | qemu_put_buffer(file, (const uint8_t *)le_bitmap, size); | |
249 | /* | |
250 | * Mark as an end, in case the middle part is screwed up due to | |
251 | * some "misterious" reason. | |
252 | */ | |
253 | qemu_put_be64(file, RAMBLOCK_RECV_BITMAP_ENDING); | |
254 | qemu_fflush(file); | |
255 | ||
bf269906 | 256 | g_free(le_bitmap); |
a335debb PX |
257 | |
258 | if (qemu_file_get_error(file)) { | |
259 | return qemu_file_get_error(file); | |
260 | } | |
261 | ||
262 | return size + sizeof(size); | |
263 | } | |
264 | ||
ec481c6c JQ |
265 | /* |
266 | * An outstanding page request, on the source, having been received | |
267 | * and queued | |
268 | */ | |
269 | struct RAMSrcPageRequest { | |
270 | RAMBlock *rb; | |
271 | hwaddr offset; | |
272 | hwaddr len; | |
273 | ||
274 | QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; | |
275 | }; | |
276 | ||
6f37bb8b JQ |
277 | /* State of RAM for migration */ |
278 | struct RAMState { | |
204b88b8 JQ |
279 | /* QEMUFile used for this migration */ |
280 | QEMUFile *f; | |
6f37bb8b JQ |
281 | /* Last block that we have visited searching for dirty pages */ |
282 | RAMBlock *last_seen_block; | |
283 | /* Last block from where we have sent data */ | |
284 | RAMBlock *last_sent_block; | |
269ace29 JQ |
285 | /* Last dirty target page we have sent */ |
286 | ram_addr_t last_page; | |
6f37bb8b JQ |
287 | /* last ram version we have seen */ |
288 | uint32_t last_version; | |
289 | /* We are in the first round */ | |
290 | bool ram_bulk_stage; | |
8d820d6f JQ |
291 | /* How many times we have dirty too many pages */ |
292 | int dirty_rate_high_cnt; | |
f664da80 JQ |
293 | /* these variables are used for bitmap sync */ |
294 | /* last time we did a full bitmap_sync */ | |
295 | int64_t time_last_bitmap_sync; | |
eac74159 | 296 | /* bytes transferred at start_time */ |
c4bdf0cf | 297 | uint64_t bytes_xfer_prev; |
a66cd90c | 298 | /* number of dirty pages since start_time */ |
68908ed6 | 299 | uint64_t num_dirty_pages_period; |
b5833fde JQ |
300 | /* xbzrle misses since the beginning of the period */ |
301 | uint64_t xbzrle_cache_miss_prev; | |
36040d9c JQ |
302 | /* number of iterations at the beginning of period */ |
303 | uint64_t iterations_prev; | |
23b28c3c JQ |
304 | /* Iterations since start */ |
305 | uint64_t iterations; | |
9360447d | 306 | /* number of dirty bits in the bitmap */ |
2dfaf12e PX |
307 | uint64_t migration_dirty_pages; |
308 | /* protects modification of the bitmap */ | |
108cfae0 | 309 | QemuMutex bitmap_mutex; |
68a098f3 JQ |
310 | /* The RAMBlock used in the last src_page_requests */ |
311 | RAMBlock *last_req_rb; | |
ec481c6c JQ |
312 | /* Queue of outstanding page requests from the destination */ |
313 | QemuMutex src_page_req_mutex; | |
314 | QSIMPLEQ_HEAD(src_page_requests, RAMSrcPageRequest) src_page_requests; | |
6f37bb8b JQ |
315 | }; |
316 | typedef struct RAMState RAMState; | |
317 | ||
53518d94 | 318 | static RAMState *ram_state; |
6f37bb8b | 319 | |
9edabd4d | 320 | uint64_t ram_bytes_remaining(void) |
2f4fde93 | 321 | { |
bae416e5 DDAG |
322 | return ram_state ? (ram_state->migration_dirty_pages * TARGET_PAGE_SIZE) : |
323 | 0; | |
2f4fde93 JQ |
324 | } |
325 | ||
9360447d | 326 | MigrationStats ram_counters; |
96506894 | 327 | |
b8fb8cb7 DDAG |
328 | /* used by the search for pages to send */ |
329 | struct PageSearchStatus { | |
330 | /* Current block being searched */ | |
331 | RAMBlock *block; | |
a935e30f JQ |
332 | /* Current page to search from */ |
333 | unsigned long page; | |
b8fb8cb7 DDAG |
334 | /* Set once we wrap around */ |
335 | bool complete_round; | |
336 | }; | |
337 | typedef struct PageSearchStatus PageSearchStatus; | |
338 | ||
56e93d26 | 339 | struct CompressParam { |
56e93d26 | 340 | bool done; |
90e56fb4 | 341 | bool quit; |
56e93d26 JQ |
342 | QEMUFile *file; |
343 | QemuMutex mutex; | |
344 | QemuCond cond; | |
345 | RAMBlock *block; | |
346 | ram_addr_t offset; | |
34ab9e97 XG |
347 | |
348 | /* internally used fields */ | |
dcaf446e | 349 | z_stream stream; |
34ab9e97 | 350 | uint8_t *originbuf; |
56e93d26 JQ |
351 | }; |
352 | typedef struct CompressParam CompressParam; | |
353 | ||
354 | struct DecompressParam { | |
73a8912b | 355 | bool done; |
90e56fb4 | 356 | bool quit; |
56e93d26 JQ |
357 | QemuMutex mutex; |
358 | QemuCond cond; | |
359 | void *des; | |
d341d9f3 | 360 | uint8_t *compbuf; |
56e93d26 | 361 | int len; |
797ca154 | 362 | z_stream stream; |
56e93d26 JQ |
363 | }; |
364 | typedef struct DecompressParam DecompressParam; | |
365 | ||
366 | static CompressParam *comp_param; | |
367 | static QemuThread *compress_threads; | |
368 | /* comp_done_cond is used to wake up the migration thread when | |
369 | * one of the compression threads has finished the compression. | |
370 | * comp_done_lock is used to co-work with comp_done_cond. | |
371 | */ | |
0d9f9a5c LL |
372 | static QemuMutex comp_done_lock; |
373 | static QemuCond comp_done_cond; | |
56e93d26 JQ |
374 | /* The empty QEMUFileOps will be used by file in CompressParam */ |
375 | static const QEMUFileOps empty_ops = { }; | |
376 | ||
34ab9e97 | 377 | static QEMUFile *decomp_file; |
56e93d26 JQ |
378 | static DecompressParam *decomp_param; |
379 | static QemuThread *decompress_threads; | |
73a8912b LL |
380 | static QemuMutex decomp_done_lock; |
381 | static QemuCond decomp_done_cond; | |
56e93d26 | 382 | |
dcaf446e | 383 | static int do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
34ab9e97 | 384 | ram_addr_t offset, uint8_t *source_buf); |
56e93d26 JQ |
385 | |
386 | static void *do_data_compress(void *opaque) | |
387 | { | |
388 | CompressParam *param = opaque; | |
a7a9a88f LL |
389 | RAMBlock *block; |
390 | ram_addr_t offset; | |
56e93d26 | 391 | |
a7a9a88f | 392 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 393 | while (!param->quit) { |
a7a9a88f LL |
394 | if (param->block) { |
395 | block = param->block; | |
396 | offset = param->offset; | |
397 | param->block = NULL; | |
398 | qemu_mutex_unlock(¶m->mutex); | |
399 | ||
34ab9e97 XG |
400 | do_compress_ram_page(param->file, ¶m->stream, block, offset, |
401 | param->originbuf); | |
a7a9a88f | 402 | |
0d9f9a5c | 403 | qemu_mutex_lock(&comp_done_lock); |
a7a9a88f | 404 | param->done = true; |
0d9f9a5c LL |
405 | qemu_cond_signal(&comp_done_cond); |
406 | qemu_mutex_unlock(&comp_done_lock); | |
a7a9a88f LL |
407 | |
408 | qemu_mutex_lock(¶m->mutex); | |
409 | } else { | |
56e93d26 JQ |
410 | qemu_cond_wait(¶m->cond, ¶m->mutex); |
411 | } | |
56e93d26 | 412 | } |
a7a9a88f | 413 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
414 | |
415 | return NULL; | |
416 | } | |
417 | ||
418 | static inline void terminate_compression_threads(void) | |
419 | { | |
420 | int idx, thread_count; | |
421 | ||
422 | thread_count = migrate_compress_threads(); | |
3d0684b2 | 423 | |
56e93d26 JQ |
424 | for (idx = 0; idx < thread_count; idx++) { |
425 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 426 | comp_param[idx].quit = true; |
56e93d26 JQ |
427 | qemu_cond_signal(&comp_param[idx].cond); |
428 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
429 | } | |
430 | } | |
431 | ||
f0afa331 | 432 | static void compress_threads_save_cleanup(void) |
56e93d26 JQ |
433 | { |
434 | int i, thread_count; | |
435 | ||
436 | if (!migrate_use_compression()) { | |
437 | return; | |
438 | } | |
439 | terminate_compression_threads(); | |
440 | thread_count = migrate_compress_threads(); | |
441 | for (i = 0; i < thread_count; i++) { | |
dcaf446e XG |
442 | /* |
443 | * we use it as a indicator which shows if the thread is | |
444 | * properly init'd or not | |
445 | */ | |
446 | if (!comp_param[i].file) { | |
447 | break; | |
448 | } | |
56e93d26 | 449 | qemu_thread_join(compress_threads + i); |
56e93d26 JQ |
450 | qemu_mutex_destroy(&comp_param[i].mutex); |
451 | qemu_cond_destroy(&comp_param[i].cond); | |
dcaf446e | 452 | deflateEnd(&comp_param[i].stream); |
34ab9e97 | 453 | g_free(comp_param[i].originbuf); |
dcaf446e XG |
454 | qemu_fclose(comp_param[i].file); |
455 | comp_param[i].file = NULL; | |
56e93d26 | 456 | } |
0d9f9a5c LL |
457 | qemu_mutex_destroy(&comp_done_lock); |
458 | qemu_cond_destroy(&comp_done_cond); | |
56e93d26 JQ |
459 | g_free(compress_threads); |
460 | g_free(comp_param); | |
56e93d26 JQ |
461 | compress_threads = NULL; |
462 | comp_param = NULL; | |
56e93d26 JQ |
463 | } |
464 | ||
dcaf446e | 465 | static int compress_threads_save_setup(void) |
56e93d26 JQ |
466 | { |
467 | int i, thread_count; | |
468 | ||
469 | if (!migrate_use_compression()) { | |
dcaf446e | 470 | return 0; |
56e93d26 | 471 | } |
56e93d26 JQ |
472 | thread_count = migrate_compress_threads(); |
473 | compress_threads = g_new0(QemuThread, thread_count); | |
474 | comp_param = g_new0(CompressParam, thread_count); | |
0d9f9a5c LL |
475 | qemu_cond_init(&comp_done_cond); |
476 | qemu_mutex_init(&comp_done_lock); | |
56e93d26 | 477 | for (i = 0; i < thread_count; i++) { |
34ab9e97 XG |
478 | comp_param[i].originbuf = g_try_malloc(TARGET_PAGE_SIZE); |
479 | if (!comp_param[i].originbuf) { | |
480 | goto exit; | |
481 | } | |
482 | ||
dcaf446e XG |
483 | if (deflateInit(&comp_param[i].stream, |
484 | migrate_compress_level()) != Z_OK) { | |
34ab9e97 | 485 | g_free(comp_param[i].originbuf); |
dcaf446e XG |
486 | goto exit; |
487 | } | |
488 | ||
e110aa91 C |
489 | /* comp_param[i].file is just used as a dummy buffer to save data, |
490 | * set its ops to empty. | |
56e93d26 JQ |
491 | */ |
492 | comp_param[i].file = qemu_fopen_ops(NULL, &empty_ops); | |
493 | comp_param[i].done = true; | |
90e56fb4 | 494 | comp_param[i].quit = false; |
56e93d26 JQ |
495 | qemu_mutex_init(&comp_param[i].mutex); |
496 | qemu_cond_init(&comp_param[i].cond); | |
497 | qemu_thread_create(compress_threads + i, "compress", | |
498 | do_data_compress, comp_param + i, | |
499 | QEMU_THREAD_JOINABLE); | |
500 | } | |
dcaf446e XG |
501 | return 0; |
502 | ||
503 | exit: | |
504 | compress_threads_save_cleanup(); | |
505 | return -1; | |
56e93d26 JQ |
506 | } |
507 | ||
f986c3d2 JQ |
508 | /* Multiple fd's */ |
509 | ||
af8b7d2b JQ |
510 | #define MULTIFD_MAGIC 0x11223344U |
511 | #define MULTIFD_VERSION 1 | |
512 | ||
513 | typedef struct { | |
514 | uint32_t magic; | |
515 | uint32_t version; | |
516 | unsigned char uuid[16]; /* QemuUUID */ | |
517 | uint8_t id; | |
518 | } __attribute__((packed)) MultiFDInit_t; | |
519 | ||
2a26c979 JQ |
520 | typedef struct { |
521 | uint32_t magic; | |
522 | uint32_t version; | |
523 | uint32_t flags; | |
524 | uint32_t size; | |
525 | uint32_t used; | |
526 | uint64_t packet_num; | |
527 | char ramblock[256]; | |
528 | uint64_t offset[]; | |
529 | } __attribute__((packed)) MultiFDPacket_t; | |
530 | ||
34c55a94 JQ |
531 | typedef struct { |
532 | /* number of used pages */ | |
533 | uint32_t used; | |
534 | /* number of allocated pages */ | |
535 | uint32_t allocated; | |
536 | /* global number of generated multifd packets */ | |
537 | uint64_t packet_num; | |
538 | /* offset of each page */ | |
539 | ram_addr_t *offset; | |
540 | /* pointer to each page */ | |
541 | struct iovec *iov; | |
542 | RAMBlock *block; | |
543 | } MultiFDPages_t; | |
544 | ||
8c4598f2 JQ |
545 | typedef struct { |
546 | /* this fields are not changed once the thread is created */ | |
547 | /* channel number */ | |
f986c3d2 | 548 | uint8_t id; |
8c4598f2 | 549 | /* channel thread name */ |
f986c3d2 | 550 | char *name; |
8c4598f2 | 551 | /* channel thread id */ |
f986c3d2 | 552 | QemuThread thread; |
8c4598f2 | 553 | /* communication channel */ |
60df2d4a | 554 | QIOChannel *c; |
8c4598f2 | 555 | /* sem where to wait for more work */ |
f986c3d2 | 556 | QemuSemaphore sem; |
8c4598f2 | 557 | /* this mutex protects the following parameters */ |
f986c3d2 | 558 | QemuMutex mutex; |
8c4598f2 | 559 | /* is this channel thread running */ |
66770707 | 560 | bool running; |
8c4598f2 | 561 | /* should this thread finish */ |
f986c3d2 | 562 | bool quit; |
34c55a94 JQ |
563 | /* array of pages to sent */ |
564 | MultiFDPages_t *pages; | |
2a26c979 JQ |
565 | /* packet allocated len */ |
566 | uint32_t packet_len; | |
567 | /* pointer to the packet */ | |
568 | MultiFDPacket_t *packet; | |
569 | /* multifd flags for each packet */ | |
570 | uint32_t flags; | |
571 | /* global number of generated multifd packets */ | |
572 | uint64_t packet_num; | |
408ea6ae JQ |
573 | /* thread local variables */ |
574 | /* packets sent through this channel */ | |
575 | uint64_t num_packets; | |
576 | /* pages sent through this channel */ | |
577 | uint64_t num_pages; | |
8c4598f2 JQ |
578 | } MultiFDSendParams; |
579 | ||
580 | typedef struct { | |
581 | /* this fields are not changed once the thread is created */ | |
582 | /* channel number */ | |
583 | uint8_t id; | |
584 | /* channel thread name */ | |
585 | char *name; | |
586 | /* channel thread id */ | |
587 | QemuThread thread; | |
588 | /* communication channel */ | |
589 | QIOChannel *c; | |
590 | /* sem where to wait for more work */ | |
591 | QemuSemaphore sem; | |
592 | /* this mutex protects the following parameters */ | |
593 | QemuMutex mutex; | |
594 | /* is this channel thread running */ | |
595 | bool running; | |
596 | /* should this thread finish */ | |
597 | bool quit; | |
34c55a94 JQ |
598 | /* array of pages to receive */ |
599 | MultiFDPages_t *pages; | |
2a26c979 JQ |
600 | /* packet allocated len */ |
601 | uint32_t packet_len; | |
602 | /* pointer to the packet */ | |
603 | MultiFDPacket_t *packet; | |
604 | /* multifd flags for each packet */ | |
605 | uint32_t flags; | |
606 | /* global number of generated multifd packets */ | |
607 | uint64_t packet_num; | |
408ea6ae JQ |
608 | /* thread local variables */ |
609 | /* packets sent through this channel */ | |
610 | uint64_t num_packets; | |
611 | /* pages sent through this channel */ | |
612 | uint64_t num_pages; | |
8c4598f2 | 613 | } MultiFDRecvParams; |
f986c3d2 | 614 | |
af8b7d2b JQ |
615 | static int multifd_send_initial_packet(MultiFDSendParams *p, Error **errp) |
616 | { | |
617 | MultiFDInit_t msg; | |
618 | int ret; | |
619 | ||
620 | msg.magic = cpu_to_be32(MULTIFD_MAGIC); | |
621 | msg.version = cpu_to_be32(MULTIFD_VERSION); | |
622 | msg.id = p->id; | |
623 | memcpy(msg.uuid, &qemu_uuid.data, sizeof(msg.uuid)); | |
624 | ||
625 | ret = qio_channel_write_all(p->c, (char *)&msg, sizeof(msg), errp); | |
626 | if (ret != 0) { | |
627 | return -1; | |
628 | } | |
629 | return 0; | |
630 | } | |
631 | ||
632 | static int multifd_recv_initial_packet(QIOChannel *c, Error **errp) | |
633 | { | |
634 | MultiFDInit_t msg; | |
635 | int ret; | |
636 | ||
637 | ret = qio_channel_read_all(c, (char *)&msg, sizeof(msg), errp); | |
638 | if (ret != 0) { | |
639 | return -1; | |
640 | } | |
641 | ||
642 | be32_to_cpus(&msg.magic); | |
643 | be32_to_cpus(&msg.version); | |
644 | ||
645 | if (msg.magic != MULTIFD_MAGIC) { | |
646 | error_setg(errp, "multifd: received packet magic %x " | |
647 | "expected %x", msg.magic, MULTIFD_MAGIC); | |
648 | return -1; | |
649 | } | |
650 | ||
651 | if (msg.version != MULTIFD_VERSION) { | |
652 | error_setg(errp, "multifd: received packet version %d " | |
653 | "expected %d", msg.version, MULTIFD_VERSION); | |
654 | return -1; | |
655 | } | |
656 | ||
657 | if (memcmp(msg.uuid, &qemu_uuid, sizeof(qemu_uuid))) { | |
658 | char *uuid = qemu_uuid_unparse_strdup(&qemu_uuid); | |
659 | char *msg_uuid = qemu_uuid_unparse_strdup((const QemuUUID *)msg.uuid); | |
660 | ||
661 | error_setg(errp, "multifd: received uuid '%s' and expected " | |
662 | "uuid '%s' for channel %hhd", msg_uuid, uuid, msg.id); | |
663 | g_free(uuid); | |
664 | g_free(msg_uuid); | |
665 | return -1; | |
666 | } | |
667 | ||
668 | if (msg.id > migrate_multifd_channels()) { | |
669 | error_setg(errp, "multifd: received channel version %d " | |
670 | "expected %d", msg.version, MULTIFD_VERSION); | |
671 | return -1; | |
672 | } | |
673 | ||
674 | return msg.id; | |
675 | } | |
676 | ||
34c55a94 JQ |
677 | static MultiFDPages_t *multifd_pages_init(size_t size) |
678 | { | |
679 | MultiFDPages_t *pages = g_new0(MultiFDPages_t, 1); | |
680 | ||
681 | pages->allocated = size; | |
682 | pages->iov = g_new0(struct iovec, size); | |
683 | pages->offset = g_new0(ram_addr_t, size); | |
684 | ||
685 | return pages; | |
686 | } | |
687 | ||
688 | static void multifd_pages_clear(MultiFDPages_t *pages) | |
689 | { | |
690 | pages->used = 0; | |
691 | pages->allocated = 0; | |
692 | pages->packet_num = 0; | |
693 | pages->block = NULL; | |
694 | g_free(pages->iov); | |
695 | pages->iov = NULL; | |
696 | g_free(pages->offset); | |
697 | pages->offset = NULL; | |
698 | g_free(pages); | |
699 | } | |
700 | ||
2a26c979 JQ |
701 | static void multifd_send_fill_packet(MultiFDSendParams *p) |
702 | { | |
703 | MultiFDPacket_t *packet = p->packet; | |
704 | int i; | |
705 | ||
706 | packet->magic = cpu_to_be32(MULTIFD_MAGIC); | |
707 | packet->version = cpu_to_be32(MULTIFD_VERSION); | |
708 | packet->flags = cpu_to_be32(p->flags); | |
709 | packet->size = cpu_to_be32(migrate_multifd_page_count()); | |
710 | packet->used = cpu_to_be32(p->pages->used); | |
711 | packet->packet_num = cpu_to_be64(p->packet_num); | |
712 | ||
713 | if (p->pages->block) { | |
714 | strncpy(packet->ramblock, p->pages->block->idstr, 256); | |
715 | } | |
716 | ||
717 | for (i = 0; i < p->pages->used; i++) { | |
718 | packet->offset[i] = cpu_to_be64(p->pages->offset[i]); | |
719 | } | |
720 | } | |
721 | ||
722 | static int multifd_recv_unfill_packet(MultiFDRecvParams *p, Error **errp) | |
723 | { | |
724 | MultiFDPacket_t *packet = p->packet; | |
725 | RAMBlock *block; | |
726 | int i; | |
727 | ||
728 | /* ToDo: We can't use it until we haven't received a message */ | |
729 | return 0; | |
730 | ||
731 | be32_to_cpus(&packet->magic); | |
732 | if (packet->magic != MULTIFD_MAGIC) { | |
733 | error_setg(errp, "multifd: received packet " | |
734 | "magic %x and expected magic %x", | |
735 | packet->magic, MULTIFD_MAGIC); | |
736 | return -1; | |
737 | } | |
738 | ||
739 | be32_to_cpus(&packet->version); | |
740 | if (packet->version != MULTIFD_VERSION) { | |
741 | error_setg(errp, "multifd: received packet " | |
742 | "version %d and expected version %d", | |
743 | packet->version, MULTIFD_VERSION); | |
744 | return -1; | |
745 | } | |
746 | ||
747 | p->flags = be32_to_cpu(packet->flags); | |
748 | ||
749 | be32_to_cpus(&packet->size); | |
750 | if (packet->size > migrate_multifd_page_count()) { | |
751 | error_setg(errp, "multifd: received packet " | |
752 | "with size %d and expected maximum size %d", | |
753 | packet->size, migrate_multifd_page_count()) ; | |
754 | return -1; | |
755 | } | |
756 | ||
757 | p->pages->used = be32_to_cpu(packet->used); | |
758 | if (p->pages->used > packet->size) { | |
759 | error_setg(errp, "multifd: received packet " | |
760 | "with size %d and expected maximum size %d", | |
761 | p->pages->used, packet->size) ; | |
762 | return -1; | |
763 | } | |
764 | ||
765 | p->packet_num = be64_to_cpu(packet->packet_num); | |
766 | ||
767 | if (p->pages->used) { | |
768 | /* make sure that ramblock is 0 terminated */ | |
769 | packet->ramblock[255] = 0; | |
770 | block = qemu_ram_block_by_name(packet->ramblock); | |
771 | if (!block) { | |
772 | error_setg(errp, "multifd: unknown ram block %s", | |
773 | packet->ramblock); | |
774 | return -1; | |
775 | } | |
776 | } | |
777 | ||
778 | for (i = 0; i < p->pages->used; i++) { | |
779 | ram_addr_t offset = be64_to_cpu(packet->offset[i]); | |
780 | ||
781 | if (offset > (block->used_length - TARGET_PAGE_SIZE)) { | |
782 | error_setg(errp, "multifd: offset too long " RAM_ADDR_FMT | |
783 | " (max " RAM_ADDR_FMT ")", | |
784 | offset, block->max_length); | |
785 | return -1; | |
786 | } | |
787 | p->pages->iov[i].iov_base = block->host + offset; | |
788 | p->pages->iov[i].iov_len = TARGET_PAGE_SIZE; | |
789 | } | |
790 | ||
791 | return 0; | |
792 | } | |
793 | ||
f986c3d2 JQ |
794 | struct { |
795 | MultiFDSendParams *params; | |
796 | /* number of created threads */ | |
797 | int count; | |
34c55a94 JQ |
798 | /* array of pages to sent */ |
799 | MultiFDPages_t *pages; | |
f986c3d2 JQ |
800 | } *multifd_send_state; |
801 | ||
66770707 | 802 | static void multifd_send_terminate_threads(Error *err) |
f986c3d2 JQ |
803 | { |
804 | int i; | |
805 | ||
7a169d74 JQ |
806 | if (err) { |
807 | MigrationState *s = migrate_get_current(); | |
808 | migrate_set_error(s, err); | |
809 | if (s->state == MIGRATION_STATUS_SETUP || | |
810 | s->state == MIGRATION_STATUS_PRE_SWITCHOVER || | |
811 | s->state == MIGRATION_STATUS_DEVICE || | |
812 | s->state == MIGRATION_STATUS_ACTIVE) { | |
813 | migrate_set_state(&s->state, s->state, | |
814 | MIGRATION_STATUS_FAILED); | |
815 | } | |
816 | } | |
817 | ||
66770707 | 818 | for (i = 0; i < migrate_multifd_channels(); i++) { |
f986c3d2 JQ |
819 | MultiFDSendParams *p = &multifd_send_state->params[i]; |
820 | ||
821 | qemu_mutex_lock(&p->mutex); | |
822 | p->quit = true; | |
823 | qemu_sem_post(&p->sem); | |
824 | qemu_mutex_unlock(&p->mutex); | |
825 | } | |
826 | } | |
827 | ||
828 | int multifd_save_cleanup(Error **errp) | |
829 | { | |
830 | int i; | |
831 | int ret = 0; | |
832 | ||
833 | if (!migrate_use_multifd()) { | |
834 | return 0; | |
835 | } | |
66770707 JQ |
836 | multifd_send_terminate_threads(NULL); |
837 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
f986c3d2 JQ |
838 | MultiFDSendParams *p = &multifd_send_state->params[i]; |
839 | ||
66770707 JQ |
840 | if (p->running) { |
841 | qemu_thread_join(&p->thread); | |
842 | } | |
60df2d4a JQ |
843 | socket_send_channel_destroy(p->c); |
844 | p->c = NULL; | |
f986c3d2 JQ |
845 | qemu_mutex_destroy(&p->mutex); |
846 | qemu_sem_destroy(&p->sem); | |
847 | g_free(p->name); | |
848 | p->name = NULL; | |
34c55a94 JQ |
849 | multifd_pages_clear(p->pages); |
850 | p->pages = NULL; | |
2a26c979 JQ |
851 | p->packet_len = 0; |
852 | g_free(p->packet); | |
853 | p->packet = NULL; | |
f986c3d2 JQ |
854 | } |
855 | g_free(multifd_send_state->params); | |
856 | multifd_send_state->params = NULL; | |
34c55a94 JQ |
857 | multifd_pages_clear(multifd_send_state->pages); |
858 | multifd_send_state->pages = NULL; | |
f986c3d2 JQ |
859 | g_free(multifd_send_state); |
860 | multifd_send_state = NULL; | |
861 | return ret; | |
862 | } | |
863 | ||
864 | static void *multifd_send_thread(void *opaque) | |
865 | { | |
866 | MultiFDSendParams *p = opaque; | |
af8b7d2b JQ |
867 | Error *local_err = NULL; |
868 | ||
408ea6ae JQ |
869 | trace_multifd_send_thread_start(p->id); |
870 | ||
af8b7d2b JQ |
871 | if (multifd_send_initial_packet(p, &local_err) < 0) { |
872 | goto out; | |
873 | } | |
408ea6ae JQ |
874 | /* initial packet */ |
875 | p->num_packets = 1; | |
f986c3d2 JQ |
876 | |
877 | while (true) { | |
d82628e4 | 878 | qemu_sem_wait(&p->sem); |
f986c3d2 | 879 | qemu_mutex_lock(&p->mutex); |
2a26c979 | 880 | multifd_send_fill_packet(p); |
f986c3d2 JQ |
881 | if (p->quit) { |
882 | qemu_mutex_unlock(&p->mutex); | |
883 | break; | |
884 | } | |
885 | qemu_mutex_unlock(&p->mutex); | |
d82628e4 JQ |
886 | /* this is impossible */ |
887 | error_setg(&local_err, "multifd_send_thread: Unknown command"); | |
888 | break; | |
f986c3d2 JQ |
889 | } |
890 | ||
af8b7d2b JQ |
891 | out: |
892 | if (local_err) { | |
893 | multifd_send_terminate_threads(local_err); | |
894 | } | |
895 | ||
66770707 JQ |
896 | qemu_mutex_lock(&p->mutex); |
897 | p->running = false; | |
898 | qemu_mutex_unlock(&p->mutex); | |
899 | ||
408ea6ae JQ |
900 | trace_multifd_send_thread_end(p->id, p->num_packets, p->num_pages); |
901 | ||
f986c3d2 JQ |
902 | return NULL; |
903 | } | |
904 | ||
60df2d4a JQ |
905 | static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque) |
906 | { | |
907 | MultiFDSendParams *p = opaque; | |
908 | QIOChannel *sioc = QIO_CHANNEL(qio_task_get_source(task)); | |
909 | Error *local_err = NULL; | |
910 | ||
911 | if (qio_task_propagate_error(task, &local_err)) { | |
912 | if (multifd_save_cleanup(&local_err) != 0) { | |
913 | migrate_set_error(migrate_get_current(), local_err); | |
914 | } | |
915 | } else { | |
916 | p->c = QIO_CHANNEL(sioc); | |
917 | qio_channel_set_delay(p->c, false); | |
918 | p->running = true; | |
919 | qemu_thread_create(&p->thread, p->name, multifd_send_thread, p, | |
920 | QEMU_THREAD_JOINABLE); | |
921 | ||
922 | atomic_inc(&multifd_send_state->count); | |
923 | } | |
924 | } | |
925 | ||
f986c3d2 JQ |
926 | int multifd_save_setup(void) |
927 | { | |
928 | int thread_count; | |
34c55a94 | 929 | uint32_t page_count = migrate_multifd_page_count(); |
f986c3d2 JQ |
930 | uint8_t i; |
931 | ||
932 | if (!migrate_use_multifd()) { | |
933 | return 0; | |
934 | } | |
935 | thread_count = migrate_multifd_channels(); | |
936 | multifd_send_state = g_malloc0(sizeof(*multifd_send_state)); | |
937 | multifd_send_state->params = g_new0(MultiFDSendParams, thread_count); | |
66770707 | 938 | atomic_set(&multifd_send_state->count, 0); |
34c55a94 JQ |
939 | multifd_send_state->pages = multifd_pages_init(page_count); |
940 | ||
f986c3d2 JQ |
941 | for (i = 0; i < thread_count; i++) { |
942 | MultiFDSendParams *p = &multifd_send_state->params[i]; | |
943 | ||
944 | qemu_mutex_init(&p->mutex); | |
945 | qemu_sem_init(&p->sem, 0); | |
946 | p->quit = false; | |
947 | p->id = i; | |
34c55a94 | 948 | p->pages = multifd_pages_init(page_count); |
2a26c979 JQ |
949 | p->packet_len = sizeof(MultiFDPacket_t) |
950 | + sizeof(ram_addr_t) * page_count; | |
951 | p->packet = g_malloc0(p->packet_len); | |
f986c3d2 | 952 | p->name = g_strdup_printf("multifdsend_%d", i); |
60df2d4a | 953 | socket_send_channel_create(multifd_new_send_channel_async, p); |
f986c3d2 JQ |
954 | } |
955 | return 0; | |
956 | } | |
957 | ||
f986c3d2 JQ |
958 | struct { |
959 | MultiFDRecvParams *params; | |
960 | /* number of created threads */ | |
961 | int count; | |
962 | } *multifd_recv_state; | |
963 | ||
66770707 | 964 | static void multifd_recv_terminate_threads(Error *err) |
f986c3d2 JQ |
965 | { |
966 | int i; | |
967 | ||
7a169d74 JQ |
968 | if (err) { |
969 | MigrationState *s = migrate_get_current(); | |
970 | migrate_set_error(s, err); | |
971 | if (s->state == MIGRATION_STATUS_SETUP || | |
972 | s->state == MIGRATION_STATUS_ACTIVE) { | |
973 | migrate_set_state(&s->state, s->state, | |
974 | MIGRATION_STATUS_FAILED); | |
975 | } | |
976 | } | |
977 | ||
66770707 | 978 | for (i = 0; i < migrate_multifd_channels(); i++) { |
f986c3d2 JQ |
979 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; |
980 | ||
981 | qemu_mutex_lock(&p->mutex); | |
982 | p->quit = true; | |
983 | qemu_sem_post(&p->sem); | |
984 | qemu_mutex_unlock(&p->mutex); | |
985 | } | |
986 | } | |
987 | ||
988 | int multifd_load_cleanup(Error **errp) | |
989 | { | |
990 | int i; | |
991 | int ret = 0; | |
992 | ||
993 | if (!migrate_use_multifd()) { | |
994 | return 0; | |
995 | } | |
66770707 JQ |
996 | multifd_recv_terminate_threads(NULL); |
997 | for (i = 0; i < migrate_multifd_channels(); i++) { | |
f986c3d2 JQ |
998 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; |
999 | ||
66770707 JQ |
1000 | if (p->running) { |
1001 | qemu_thread_join(&p->thread); | |
1002 | } | |
60df2d4a JQ |
1003 | object_unref(OBJECT(p->c)); |
1004 | p->c = NULL; | |
f986c3d2 JQ |
1005 | qemu_mutex_destroy(&p->mutex); |
1006 | qemu_sem_destroy(&p->sem); | |
1007 | g_free(p->name); | |
1008 | p->name = NULL; | |
34c55a94 JQ |
1009 | multifd_pages_clear(p->pages); |
1010 | p->pages = NULL; | |
2a26c979 JQ |
1011 | p->packet_len = 0; |
1012 | g_free(p->packet); | |
1013 | p->packet = NULL; | |
f986c3d2 JQ |
1014 | } |
1015 | g_free(multifd_recv_state->params); | |
1016 | multifd_recv_state->params = NULL; | |
1017 | g_free(multifd_recv_state); | |
1018 | multifd_recv_state = NULL; | |
1019 | ||
1020 | return ret; | |
1021 | } | |
1022 | ||
1023 | static void *multifd_recv_thread(void *opaque) | |
1024 | { | |
1025 | MultiFDRecvParams *p = opaque; | |
2a26c979 JQ |
1026 | Error *local_err = NULL; |
1027 | int ret; | |
f986c3d2 | 1028 | |
408ea6ae JQ |
1029 | trace_multifd_recv_thread_start(p->id); |
1030 | ||
f986c3d2 | 1031 | while (true) { |
d82628e4 | 1032 | qemu_sem_wait(&p->sem); |
f986c3d2 | 1033 | qemu_mutex_lock(&p->mutex); |
2a26c979 JQ |
1034 | if (false) { |
1035 | /* ToDo: Packet reception goes here */ | |
1036 | ||
1037 | ret = multifd_recv_unfill_packet(p, &local_err); | |
1038 | qemu_mutex_unlock(&p->mutex); | |
1039 | if (ret) { | |
1040 | break; | |
1041 | } | |
1042 | } else if (p->quit) { | |
f986c3d2 JQ |
1043 | qemu_mutex_unlock(&p->mutex); |
1044 | break; | |
1045 | } | |
1046 | qemu_mutex_unlock(&p->mutex); | |
d82628e4 JQ |
1047 | /* this is impossible */ |
1048 | error_setg(&local_err, "multifd_recv_thread: Unknown command"); | |
1049 | break; | |
f986c3d2 JQ |
1050 | } |
1051 | ||
d82628e4 JQ |
1052 | if (local_err) { |
1053 | multifd_recv_terminate_threads(local_err); | |
1054 | } | |
66770707 JQ |
1055 | qemu_mutex_lock(&p->mutex); |
1056 | p->running = false; | |
1057 | qemu_mutex_unlock(&p->mutex); | |
1058 | ||
408ea6ae JQ |
1059 | trace_multifd_recv_thread_end(p->id, p->num_packets, p->num_pages); |
1060 | ||
f986c3d2 JQ |
1061 | return NULL; |
1062 | } | |
1063 | ||
1064 | int multifd_load_setup(void) | |
1065 | { | |
1066 | int thread_count; | |
34c55a94 | 1067 | uint32_t page_count = migrate_multifd_page_count(); |
f986c3d2 JQ |
1068 | uint8_t i; |
1069 | ||
1070 | if (!migrate_use_multifd()) { | |
1071 | return 0; | |
1072 | } | |
1073 | thread_count = migrate_multifd_channels(); | |
1074 | multifd_recv_state = g_malloc0(sizeof(*multifd_recv_state)); | |
1075 | multifd_recv_state->params = g_new0(MultiFDRecvParams, thread_count); | |
66770707 | 1076 | atomic_set(&multifd_recv_state->count, 0); |
34c55a94 | 1077 | |
f986c3d2 JQ |
1078 | for (i = 0; i < thread_count; i++) { |
1079 | MultiFDRecvParams *p = &multifd_recv_state->params[i]; | |
1080 | ||
1081 | qemu_mutex_init(&p->mutex); | |
1082 | qemu_sem_init(&p->sem, 0); | |
1083 | p->quit = false; | |
1084 | p->id = i; | |
34c55a94 | 1085 | p->pages = multifd_pages_init(page_count); |
2a26c979 JQ |
1086 | p->packet_len = sizeof(MultiFDPacket_t) |
1087 | + sizeof(ram_addr_t) * page_count; | |
1088 | p->packet = g_malloc0(p->packet_len); | |
f986c3d2 | 1089 | p->name = g_strdup_printf("multifdrecv_%d", i); |
f986c3d2 JQ |
1090 | } |
1091 | return 0; | |
1092 | } | |
1093 | ||
62c1e0ca JQ |
1094 | bool multifd_recv_all_channels_created(void) |
1095 | { | |
1096 | int thread_count = migrate_multifd_channels(); | |
1097 | ||
1098 | if (!migrate_use_multifd()) { | |
1099 | return true; | |
1100 | } | |
1101 | ||
1102 | return thread_count == atomic_read(&multifd_recv_state->count); | |
1103 | } | |
1104 | ||
71bb07db JQ |
1105 | void multifd_recv_new_channel(QIOChannel *ioc) |
1106 | { | |
60df2d4a | 1107 | MultiFDRecvParams *p; |
af8b7d2b JQ |
1108 | Error *local_err = NULL; |
1109 | int id; | |
60df2d4a | 1110 | |
af8b7d2b JQ |
1111 | id = multifd_recv_initial_packet(ioc, &local_err); |
1112 | if (id < 0) { | |
1113 | multifd_recv_terminate_threads(local_err); | |
1114 | return; | |
1115 | } | |
1116 | ||
1117 | p = &multifd_recv_state->params[id]; | |
1118 | if (p->c != NULL) { | |
1119 | error_setg(&local_err, "multifd: received id '%d' already setup'", | |
1120 | id); | |
1121 | multifd_recv_terminate_threads(local_err); | |
1122 | return; | |
1123 | } | |
60df2d4a JQ |
1124 | p->c = ioc; |
1125 | object_ref(OBJECT(ioc)); | |
408ea6ae JQ |
1126 | /* initial packet */ |
1127 | p->num_packets = 1; | |
60df2d4a JQ |
1128 | |
1129 | p->running = true; | |
1130 | qemu_thread_create(&p->thread, p->name, multifd_recv_thread, p, | |
1131 | QEMU_THREAD_JOINABLE); | |
1132 | atomic_inc(&multifd_recv_state->count); | |
36c2f8be JQ |
1133 | if (multifd_recv_state->count == migrate_multifd_channels()) { |
1134 | migration_incoming_process(); | |
1135 | } | |
71bb07db JQ |
1136 | } |
1137 | ||
56e93d26 | 1138 | /** |
3d0684b2 | 1139 | * save_page_header: write page header to wire |
56e93d26 JQ |
1140 | * |
1141 | * If this is the 1st block, it also writes the block identification | |
1142 | * | |
3d0684b2 | 1143 | * Returns the number of bytes written |
56e93d26 JQ |
1144 | * |
1145 | * @f: QEMUFile where to send the data | |
1146 | * @block: block that contains the page we want to send | |
1147 | * @offset: offset inside the block for the page | |
1148 | * in the lower bits, it contains flags | |
1149 | */ | |
2bf3aa85 JQ |
1150 | static size_t save_page_header(RAMState *rs, QEMUFile *f, RAMBlock *block, |
1151 | ram_addr_t offset) | |
56e93d26 | 1152 | { |
9f5f380b | 1153 | size_t size, len; |
56e93d26 | 1154 | |
24795694 JQ |
1155 | if (block == rs->last_sent_block) { |
1156 | offset |= RAM_SAVE_FLAG_CONTINUE; | |
1157 | } | |
2bf3aa85 | 1158 | qemu_put_be64(f, offset); |
56e93d26 JQ |
1159 | size = 8; |
1160 | ||
1161 | if (!(offset & RAM_SAVE_FLAG_CONTINUE)) { | |
9f5f380b | 1162 | len = strlen(block->idstr); |
2bf3aa85 JQ |
1163 | qemu_put_byte(f, len); |
1164 | qemu_put_buffer(f, (uint8_t *)block->idstr, len); | |
9f5f380b | 1165 | size += 1 + len; |
24795694 | 1166 | rs->last_sent_block = block; |
56e93d26 JQ |
1167 | } |
1168 | return size; | |
1169 | } | |
1170 | ||
3d0684b2 JQ |
1171 | /** |
1172 | * mig_throttle_guest_down: throotle down the guest | |
1173 | * | |
1174 | * Reduce amount of guest cpu execution to hopefully slow down memory | |
1175 | * writes. If guest dirty memory rate is reduced below the rate at | |
1176 | * which we can transfer pages to the destination then we should be | |
1177 | * able to complete migration. Some workloads dirty memory way too | |
1178 | * fast and will not effectively converge, even with auto-converge. | |
070afca2 JH |
1179 | */ |
1180 | static void mig_throttle_guest_down(void) | |
1181 | { | |
1182 | MigrationState *s = migrate_get_current(); | |
2594f56d DB |
1183 | uint64_t pct_initial = s->parameters.cpu_throttle_initial; |
1184 | uint64_t pct_icrement = s->parameters.cpu_throttle_increment; | |
070afca2 JH |
1185 | |
1186 | /* We have not started throttling yet. Let's start it. */ | |
1187 | if (!cpu_throttle_active()) { | |
1188 | cpu_throttle_set(pct_initial); | |
1189 | } else { | |
1190 | /* Throttling already on, just increase the rate */ | |
1191 | cpu_throttle_set(cpu_throttle_get_percentage() + pct_icrement); | |
1192 | } | |
1193 | } | |
1194 | ||
3d0684b2 JQ |
1195 | /** |
1196 | * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache | |
1197 | * | |
6f37bb8b | 1198 | * @rs: current RAM state |
3d0684b2 JQ |
1199 | * @current_addr: address for the zero page |
1200 | * | |
1201 | * Update the xbzrle cache to reflect a page that's been sent as all 0. | |
56e93d26 JQ |
1202 | * The important thing is that a stale (not-yet-0'd) page be replaced |
1203 | * by the new data. | |
1204 | * As a bonus, if the page wasn't in the cache it gets added so that | |
3d0684b2 | 1205 | * when a small write is made into the 0'd page it gets XBZRLE sent. |
56e93d26 | 1206 | */ |
6f37bb8b | 1207 | static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr) |
56e93d26 | 1208 | { |
6f37bb8b | 1209 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { |
56e93d26 JQ |
1210 | return; |
1211 | } | |
1212 | ||
1213 | /* We don't care if this fails to allocate a new cache page | |
1214 | * as long as it updated an old one */ | |
c00e0928 | 1215 | cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page, |
9360447d | 1216 | ram_counters.dirty_sync_count); |
56e93d26 JQ |
1217 | } |
1218 | ||
1219 | #define ENCODING_FLAG_XBZRLE 0x1 | |
1220 | ||
1221 | /** | |
1222 | * save_xbzrle_page: compress and send current page | |
1223 | * | |
1224 | * Returns: 1 means that we wrote the page | |
1225 | * 0 means that page is identical to the one already sent | |
1226 | * -1 means that xbzrle would be longer than normal | |
1227 | * | |
5a987738 | 1228 | * @rs: current RAM state |
3d0684b2 JQ |
1229 | * @current_data: pointer to the address of the page contents |
1230 | * @current_addr: addr of the page | |
56e93d26 JQ |
1231 | * @block: block that contains the page we want to send |
1232 | * @offset: offset inside the block for the page | |
1233 | * @last_stage: if we are at the completion stage | |
56e93d26 | 1234 | */ |
204b88b8 | 1235 | static int save_xbzrle_page(RAMState *rs, uint8_t **current_data, |
56e93d26 | 1236 | ram_addr_t current_addr, RAMBlock *block, |
072c2511 | 1237 | ram_addr_t offset, bool last_stage) |
56e93d26 JQ |
1238 | { |
1239 | int encoded_len = 0, bytes_xbzrle; | |
1240 | uint8_t *prev_cached_page; | |
1241 | ||
9360447d JQ |
1242 | if (!cache_is_cached(XBZRLE.cache, current_addr, |
1243 | ram_counters.dirty_sync_count)) { | |
1244 | xbzrle_counters.cache_miss++; | |
56e93d26 JQ |
1245 | if (!last_stage) { |
1246 | if (cache_insert(XBZRLE.cache, current_addr, *current_data, | |
9360447d | 1247 | ram_counters.dirty_sync_count) == -1) { |
56e93d26 JQ |
1248 | return -1; |
1249 | } else { | |
1250 | /* update *current_data when the page has been | |
1251 | inserted into cache */ | |
1252 | *current_data = get_cached_data(XBZRLE.cache, current_addr); | |
1253 | } | |
1254 | } | |
1255 | return -1; | |
1256 | } | |
1257 | ||
1258 | prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); | |
1259 | ||
1260 | /* save current buffer into memory */ | |
1261 | memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); | |
1262 | ||
1263 | /* XBZRLE encoding (if there is no overflow) */ | |
1264 | encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, | |
1265 | TARGET_PAGE_SIZE, XBZRLE.encoded_buf, | |
1266 | TARGET_PAGE_SIZE); | |
1267 | if (encoded_len == 0) { | |
55c4446b | 1268 | trace_save_xbzrle_page_skipping(); |
56e93d26 JQ |
1269 | return 0; |
1270 | } else if (encoded_len == -1) { | |
55c4446b | 1271 | trace_save_xbzrle_page_overflow(); |
9360447d | 1272 | xbzrle_counters.overflow++; |
56e93d26 JQ |
1273 | /* update data in the cache */ |
1274 | if (!last_stage) { | |
1275 | memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE); | |
1276 | *current_data = prev_cached_page; | |
1277 | } | |
1278 | return -1; | |
1279 | } | |
1280 | ||
1281 | /* we need to update the data in the cache, in order to get the same data */ | |
1282 | if (!last_stage) { | |
1283 | memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); | |
1284 | } | |
1285 | ||
1286 | /* Send XBZRLE based compressed page */ | |
2bf3aa85 | 1287 | bytes_xbzrle = save_page_header(rs, rs->f, block, |
204b88b8 JQ |
1288 | offset | RAM_SAVE_FLAG_XBZRLE); |
1289 | qemu_put_byte(rs->f, ENCODING_FLAG_XBZRLE); | |
1290 | qemu_put_be16(rs->f, encoded_len); | |
1291 | qemu_put_buffer(rs->f, XBZRLE.encoded_buf, encoded_len); | |
56e93d26 | 1292 | bytes_xbzrle += encoded_len + 1 + 2; |
9360447d JQ |
1293 | xbzrle_counters.pages++; |
1294 | xbzrle_counters.bytes += bytes_xbzrle; | |
1295 | ram_counters.transferred += bytes_xbzrle; | |
56e93d26 JQ |
1296 | |
1297 | return 1; | |
1298 | } | |
1299 | ||
3d0684b2 JQ |
1300 | /** |
1301 | * migration_bitmap_find_dirty: find the next dirty page from start | |
f3f491fc | 1302 | * |
3d0684b2 JQ |
1303 | * Called with rcu_read_lock() to protect migration_bitmap |
1304 | * | |
1305 | * Returns the byte offset within memory region of the start of a dirty page | |
1306 | * | |
6f37bb8b | 1307 | * @rs: current RAM state |
3d0684b2 | 1308 | * @rb: RAMBlock where to search for dirty pages |
a935e30f | 1309 | * @start: page where we start the search |
f3f491fc | 1310 | */ |
56e93d26 | 1311 | static inline |
a935e30f | 1312 | unsigned long migration_bitmap_find_dirty(RAMState *rs, RAMBlock *rb, |
f20e2865 | 1313 | unsigned long start) |
56e93d26 | 1314 | { |
6b6712ef JQ |
1315 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; |
1316 | unsigned long *bitmap = rb->bmap; | |
56e93d26 JQ |
1317 | unsigned long next; |
1318 | ||
b895de50 CLG |
1319 | if (!qemu_ram_is_migratable(rb)) { |
1320 | return size; | |
1321 | } | |
1322 | ||
6b6712ef JQ |
1323 | if (rs->ram_bulk_stage && start > 0) { |
1324 | next = start + 1; | |
56e93d26 | 1325 | } else { |
6b6712ef | 1326 | next = find_next_bit(bitmap, size, start); |
56e93d26 JQ |
1327 | } |
1328 | ||
6b6712ef | 1329 | return next; |
56e93d26 JQ |
1330 | } |
1331 | ||
06b10688 | 1332 | static inline bool migration_bitmap_clear_dirty(RAMState *rs, |
f20e2865 JQ |
1333 | RAMBlock *rb, |
1334 | unsigned long page) | |
a82d593b DDAG |
1335 | { |
1336 | bool ret; | |
a82d593b | 1337 | |
6b6712ef | 1338 | ret = test_and_clear_bit(page, rb->bmap); |
a82d593b DDAG |
1339 | |
1340 | if (ret) { | |
0d8ec885 | 1341 | rs->migration_dirty_pages--; |
a82d593b DDAG |
1342 | } |
1343 | return ret; | |
1344 | } | |
1345 | ||
15440dd5 JQ |
1346 | static void migration_bitmap_sync_range(RAMState *rs, RAMBlock *rb, |
1347 | ram_addr_t start, ram_addr_t length) | |
56e93d26 | 1348 | { |
0d8ec885 | 1349 | rs->migration_dirty_pages += |
6b6712ef | 1350 | cpu_physical_memory_sync_dirty_bitmap(rb, start, length, |
0d8ec885 | 1351 | &rs->num_dirty_pages_period); |
56e93d26 JQ |
1352 | } |
1353 | ||
3d0684b2 JQ |
1354 | /** |
1355 | * ram_pagesize_summary: calculate all the pagesizes of a VM | |
1356 | * | |
1357 | * Returns a summary bitmap of the page sizes of all RAMBlocks | |
1358 | * | |
1359 | * For VMs with just normal pages this is equivalent to the host page | |
1360 | * size. If it's got some huge pages then it's the OR of all the | |
1361 | * different page sizes. | |
e8ca1db2 DDAG |
1362 | */ |
1363 | uint64_t ram_pagesize_summary(void) | |
1364 | { | |
1365 | RAMBlock *block; | |
1366 | uint64_t summary = 0; | |
1367 | ||
b895de50 | 1368 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
e8ca1db2 DDAG |
1369 | summary |= block->page_size; |
1370 | } | |
1371 | ||
1372 | return summary; | |
1373 | } | |
1374 | ||
b734035b XG |
1375 | static void migration_update_rates(RAMState *rs, int64_t end_time) |
1376 | { | |
1377 | uint64_t iter_count = rs->iterations - rs->iterations_prev; | |
1378 | ||
1379 | /* calculate period counters */ | |
1380 | ram_counters.dirty_pages_rate = rs->num_dirty_pages_period * 1000 | |
1381 | / (end_time - rs->time_last_bitmap_sync); | |
1382 | ||
1383 | if (!iter_count) { | |
1384 | return; | |
1385 | } | |
1386 | ||
1387 | if (migrate_use_xbzrle()) { | |
1388 | xbzrle_counters.cache_miss_rate = (double)(xbzrle_counters.cache_miss - | |
1389 | rs->xbzrle_cache_miss_prev) / iter_count; | |
1390 | rs->xbzrle_cache_miss_prev = xbzrle_counters.cache_miss; | |
1391 | } | |
1392 | } | |
1393 | ||
8d820d6f | 1394 | static void migration_bitmap_sync(RAMState *rs) |
56e93d26 JQ |
1395 | { |
1396 | RAMBlock *block; | |
56e93d26 | 1397 | int64_t end_time; |
c4bdf0cf | 1398 | uint64_t bytes_xfer_now; |
56e93d26 | 1399 | |
9360447d | 1400 | ram_counters.dirty_sync_count++; |
56e93d26 | 1401 | |
f664da80 JQ |
1402 | if (!rs->time_last_bitmap_sync) { |
1403 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56e93d26 JQ |
1404 | } |
1405 | ||
1406 | trace_migration_bitmap_sync_start(); | |
9c1f8f44 | 1407 | memory_global_dirty_log_sync(); |
56e93d26 | 1408 | |
108cfae0 | 1409 | qemu_mutex_lock(&rs->bitmap_mutex); |
56e93d26 | 1410 | rcu_read_lock(); |
b895de50 | 1411 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
15440dd5 | 1412 | migration_bitmap_sync_range(rs, block, 0, block->used_length); |
56e93d26 | 1413 | } |
650af890 | 1414 | ram_counters.remaining = ram_bytes_remaining(); |
56e93d26 | 1415 | rcu_read_unlock(); |
108cfae0 | 1416 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 1417 | |
a66cd90c | 1418 | trace_migration_bitmap_sync_end(rs->num_dirty_pages_period); |
1ffb5dfd | 1419 | |
56e93d26 JQ |
1420 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1421 | ||
1422 | /* more than 1 second = 1000 millisecons */ | |
f664da80 | 1423 | if (end_time > rs->time_last_bitmap_sync + 1000) { |
9360447d | 1424 | bytes_xfer_now = ram_counters.transferred; |
d693c6f1 | 1425 | |
9ac78b61 PL |
1426 | /* During block migration the auto-converge logic incorrectly detects |
1427 | * that ram migration makes no progress. Avoid this by disabling the | |
1428 | * throttling logic during the bulk phase of block migration. */ | |
1429 | if (migrate_auto_converge() && !blk_mig_bulk_active()) { | |
56e93d26 JQ |
1430 | /* The following detection logic can be refined later. For now: |
1431 | Check to see if the dirtied bytes is 50% more than the approx. | |
1432 | amount of bytes that just got transferred since the last time we | |
070afca2 JH |
1433 | were in this routine. If that happens twice, start or increase |
1434 | throttling */ | |
070afca2 | 1435 | |
d693c6f1 | 1436 | if ((rs->num_dirty_pages_period * TARGET_PAGE_SIZE > |
eac74159 | 1437 | (bytes_xfer_now - rs->bytes_xfer_prev) / 2) && |
b4a3c64b | 1438 | (++rs->dirty_rate_high_cnt >= 2)) { |
56e93d26 | 1439 | trace_migration_throttle(); |
8d820d6f | 1440 | rs->dirty_rate_high_cnt = 0; |
070afca2 | 1441 | mig_throttle_guest_down(); |
d693c6f1 | 1442 | } |
56e93d26 | 1443 | } |
070afca2 | 1444 | |
b734035b XG |
1445 | migration_update_rates(rs, end_time); |
1446 | ||
1447 | rs->iterations_prev = rs->iterations; | |
d693c6f1 FF |
1448 | |
1449 | /* reset period counters */ | |
f664da80 | 1450 | rs->time_last_bitmap_sync = end_time; |
a66cd90c | 1451 | rs->num_dirty_pages_period = 0; |
d2a4d85a | 1452 | rs->bytes_xfer_prev = bytes_xfer_now; |
56e93d26 | 1453 | } |
4addcd4f | 1454 | if (migrate_use_events()) { |
9360447d | 1455 | qapi_event_send_migration_pass(ram_counters.dirty_sync_count, NULL); |
4addcd4f | 1456 | } |
56e93d26 JQ |
1457 | } |
1458 | ||
1459 | /** | |
3d0684b2 | 1460 | * save_zero_page: send the zero page to the stream |
56e93d26 | 1461 | * |
3d0684b2 | 1462 | * Returns the number of pages written. |
56e93d26 | 1463 | * |
f7ccd61b | 1464 | * @rs: current RAM state |
56e93d26 JQ |
1465 | * @block: block that contains the page we want to send |
1466 | * @offset: offset inside the block for the page | |
56e93d26 | 1467 | */ |
7faccdc3 | 1468 | static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset) |
56e93d26 | 1469 | { |
7faccdc3 | 1470 | uint8_t *p = block->host + offset; |
56e93d26 JQ |
1471 | int pages = -1; |
1472 | ||
1473 | if (is_zero_range(p, TARGET_PAGE_SIZE)) { | |
9360447d JQ |
1474 | ram_counters.duplicate++; |
1475 | ram_counters.transferred += | |
bb890ed5 | 1476 | save_page_header(rs, rs->f, block, offset | RAM_SAVE_FLAG_ZERO); |
ce25d337 | 1477 | qemu_put_byte(rs->f, 0); |
9360447d | 1478 | ram_counters.transferred += 1; |
56e93d26 JQ |
1479 | pages = 1; |
1480 | } | |
1481 | ||
1482 | return pages; | |
1483 | } | |
1484 | ||
5727309d | 1485 | static void ram_release_pages(const char *rbname, uint64_t offset, int pages) |
53f09a10 | 1486 | { |
5727309d | 1487 | if (!migrate_release_ram() || !migration_in_postcopy()) { |
53f09a10 PB |
1488 | return; |
1489 | } | |
1490 | ||
aaa2064c | 1491 | ram_discard_range(rbname, offset, pages << TARGET_PAGE_BITS); |
53f09a10 PB |
1492 | } |
1493 | ||
059ff0fb XG |
1494 | /* |
1495 | * @pages: the number of pages written by the control path, | |
1496 | * < 0 - error | |
1497 | * > 0 - number of pages written | |
1498 | * | |
1499 | * Return true if the pages has been saved, otherwise false is returned. | |
1500 | */ | |
1501 | static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, | |
1502 | int *pages) | |
1503 | { | |
1504 | uint64_t bytes_xmit = 0; | |
1505 | int ret; | |
1506 | ||
1507 | *pages = -1; | |
1508 | ret = ram_control_save_page(rs->f, block->offset, offset, TARGET_PAGE_SIZE, | |
1509 | &bytes_xmit); | |
1510 | if (ret == RAM_SAVE_CONTROL_NOT_SUPP) { | |
1511 | return false; | |
1512 | } | |
1513 | ||
1514 | if (bytes_xmit) { | |
1515 | ram_counters.transferred += bytes_xmit; | |
1516 | *pages = 1; | |
1517 | } | |
1518 | ||
1519 | if (ret == RAM_SAVE_CONTROL_DELAYED) { | |
1520 | return true; | |
1521 | } | |
1522 | ||
1523 | if (bytes_xmit > 0) { | |
1524 | ram_counters.normal++; | |
1525 | } else if (bytes_xmit == 0) { | |
1526 | ram_counters.duplicate++; | |
1527 | } | |
1528 | ||
1529 | return true; | |
1530 | } | |
1531 | ||
65dacaa0 XG |
1532 | /* |
1533 | * directly send the page to the stream | |
1534 | * | |
1535 | * Returns the number of pages written. | |
1536 | * | |
1537 | * @rs: current RAM state | |
1538 | * @block: block that contains the page we want to send | |
1539 | * @offset: offset inside the block for the page | |
1540 | * @buf: the page to be sent | |
1541 | * @async: send to page asyncly | |
1542 | */ | |
1543 | static int save_normal_page(RAMState *rs, RAMBlock *block, ram_addr_t offset, | |
1544 | uint8_t *buf, bool async) | |
1545 | { | |
1546 | ram_counters.transferred += save_page_header(rs, rs->f, block, | |
1547 | offset | RAM_SAVE_FLAG_PAGE); | |
1548 | if (async) { | |
1549 | qemu_put_buffer_async(rs->f, buf, TARGET_PAGE_SIZE, | |
1550 | migrate_release_ram() & | |
1551 | migration_in_postcopy()); | |
1552 | } else { | |
1553 | qemu_put_buffer(rs->f, buf, TARGET_PAGE_SIZE); | |
1554 | } | |
1555 | ram_counters.transferred += TARGET_PAGE_SIZE; | |
1556 | ram_counters.normal++; | |
1557 | return 1; | |
1558 | } | |
1559 | ||
56e93d26 | 1560 | /** |
3d0684b2 | 1561 | * ram_save_page: send the given page to the stream |
56e93d26 | 1562 | * |
3d0684b2 | 1563 | * Returns the number of pages written. |
3fd3c4b3 DDAG |
1564 | * < 0 - error |
1565 | * >=0 - Number of pages written - this might legally be 0 | |
1566 | * if xbzrle noticed the page was the same. | |
56e93d26 | 1567 | * |
6f37bb8b | 1568 | * @rs: current RAM state |
56e93d26 JQ |
1569 | * @block: block that contains the page we want to send |
1570 | * @offset: offset inside the block for the page | |
1571 | * @last_stage: if we are at the completion stage | |
56e93d26 | 1572 | */ |
a0a8aa14 | 1573 | static int ram_save_page(RAMState *rs, PageSearchStatus *pss, bool last_stage) |
56e93d26 JQ |
1574 | { |
1575 | int pages = -1; | |
56e93d26 | 1576 | uint8_t *p; |
56e93d26 | 1577 | bool send_async = true; |
a08f6890 | 1578 | RAMBlock *block = pss->block; |
a935e30f | 1579 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; |
059ff0fb | 1580 | ram_addr_t current_addr = block->offset + offset; |
56e93d26 | 1581 | |
2f68e399 | 1582 | p = block->host + offset; |
1db9d8e5 | 1583 | trace_ram_save_page(block->idstr, (uint64_t)offset, p); |
56e93d26 | 1584 | |
56e93d26 | 1585 | XBZRLE_cache_lock(); |
d7400a34 XG |
1586 | if (!rs->ram_bulk_stage && !migration_in_postcopy() && |
1587 | migrate_use_xbzrle()) { | |
059ff0fb XG |
1588 | pages = save_xbzrle_page(rs, &p, current_addr, block, |
1589 | offset, last_stage); | |
1590 | if (!last_stage) { | |
1591 | /* Can't send this cached data async, since the cache page | |
1592 | * might get updated before it gets to the wire | |
56e93d26 | 1593 | */ |
059ff0fb | 1594 | send_async = false; |
56e93d26 JQ |
1595 | } |
1596 | } | |
1597 | ||
1598 | /* XBZRLE overflow or normal page */ | |
1599 | if (pages == -1) { | |
65dacaa0 | 1600 | pages = save_normal_page(rs, block, offset, p, send_async); |
56e93d26 JQ |
1601 | } |
1602 | ||
1603 | XBZRLE_cache_unlock(); | |
1604 | ||
1605 | return pages; | |
1606 | } | |
1607 | ||
dcaf446e | 1608 | static int do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block, |
34ab9e97 | 1609 | ram_addr_t offset, uint8_t *source_buf) |
56e93d26 | 1610 | { |
53518d94 | 1611 | RAMState *rs = ram_state; |
56e93d26 | 1612 | int bytes_sent, blen; |
a7a9a88f | 1613 | uint8_t *p = block->host + (offset & TARGET_PAGE_MASK); |
56e93d26 | 1614 | |
2bf3aa85 | 1615 | bytes_sent = save_page_header(rs, f, block, offset | |
56e93d26 | 1616 | RAM_SAVE_FLAG_COMPRESS_PAGE); |
34ab9e97 XG |
1617 | |
1618 | /* | |
1619 | * copy it to a internal buffer to avoid it being modified by VM | |
1620 | * so that we can catch up the error during compression and | |
1621 | * decompression | |
1622 | */ | |
1623 | memcpy(source_buf, p, TARGET_PAGE_SIZE); | |
1624 | blen = qemu_put_compression_data(f, stream, source_buf, TARGET_PAGE_SIZE); | |
b3be2896 LL |
1625 | if (blen < 0) { |
1626 | bytes_sent = 0; | |
1627 | qemu_file_set_error(migrate_get_current()->to_dst_file, blen); | |
1628 | error_report("compressed data failed!"); | |
1629 | } else { | |
1630 | bytes_sent += blen; | |
5727309d | 1631 | ram_release_pages(block->idstr, offset & TARGET_PAGE_MASK, 1); |
b3be2896 | 1632 | } |
56e93d26 JQ |
1633 | |
1634 | return bytes_sent; | |
1635 | } | |
1636 | ||
ce25d337 | 1637 | static void flush_compressed_data(RAMState *rs) |
56e93d26 JQ |
1638 | { |
1639 | int idx, len, thread_count; | |
1640 | ||
1641 | if (!migrate_use_compression()) { | |
1642 | return; | |
1643 | } | |
1644 | thread_count = migrate_compress_threads(); | |
a7a9a88f | 1645 | |
0d9f9a5c | 1646 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 | 1647 | for (idx = 0; idx < thread_count; idx++) { |
a7a9a88f | 1648 | while (!comp_param[idx].done) { |
0d9f9a5c | 1649 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 | 1650 | } |
a7a9a88f | 1651 | } |
0d9f9a5c | 1652 | qemu_mutex_unlock(&comp_done_lock); |
a7a9a88f LL |
1653 | |
1654 | for (idx = 0; idx < thread_count; idx++) { | |
1655 | qemu_mutex_lock(&comp_param[idx].mutex); | |
90e56fb4 | 1656 | if (!comp_param[idx].quit) { |
ce25d337 | 1657 | len = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
9360447d | 1658 | ram_counters.transferred += len; |
56e93d26 | 1659 | } |
a7a9a88f | 1660 | qemu_mutex_unlock(&comp_param[idx].mutex); |
56e93d26 JQ |
1661 | } |
1662 | } | |
1663 | ||
1664 | static inline void set_compress_params(CompressParam *param, RAMBlock *block, | |
1665 | ram_addr_t offset) | |
1666 | { | |
1667 | param->block = block; | |
1668 | param->offset = offset; | |
1669 | } | |
1670 | ||
ce25d337 JQ |
1671 | static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block, |
1672 | ram_addr_t offset) | |
56e93d26 JQ |
1673 | { |
1674 | int idx, thread_count, bytes_xmit = -1, pages = -1; | |
1675 | ||
1676 | thread_count = migrate_compress_threads(); | |
0d9f9a5c | 1677 | qemu_mutex_lock(&comp_done_lock); |
56e93d26 JQ |
1678 | while (true) { |
1679 | for (idx = 0; idx < thread_count; idx++) { | |
1680 | if (comp_param[idx].done) { | |
a7a9a88f | 1681 | comp_param[idx].done = false; |
ce25d337 | 1682 | bytes_xmit = qemu_put_qemu_file(rs->f, comp_param[idx].file); |
a7a9a88f | 1683 | qemu_mutex_lock(&comp_param[idx].mutex); |
56e93d26 | 1684 | set_compress_params(&comp_param[idx], block, offset); |
a7a9a88f LL |
1685 | qemu_cond_signal(&comp_param[idx].cond); |
1686 | qemu_mutex_unlock(&comp_param[idx].mutex); | |
56e93d26 | 1687 | pages = 1; |
9360447d JQ |
1688 | ram_counters.normal++; |
1689 | ram_counters.transferred += bytes_xmit; | |
56e93d26 JQ |
1690 | break; |
1691 | } | |
1692 | } | |
1693 | if (pages > 0) { | |
1694 | break; | |
1695 | } else { | |
0d9f9a5c | 1696 | qemu_cond_wait(&comp_done_cond, &comp_done_lock); |
56e93d26 JQ |
1697 | } |
1698 | } | |
0d9f9a5c | 1699 | qemu_mutex_unlock(&comp_done_lock); |
56e93d26 JQ |
1700 | |
1701 | return pages; | |
1702 | } | |
1703 | ||
3d0684b2 JQ |
1704 | /** |
1705 | * find_dirty_block: find the next dirty page and update any state | |
1706 | * associated with the search process. | |
b9e60928 | 1707 | * |
3d0684b2 | 1708 | * Returns if a page is found |
b9e60928 | 1709 | * |
6f37bb8b | 1710 | * @rs: current RAM state |
3d0684b2 JQ |
1711 | * @pss: data about the state of the current dirty page scan |
1712 | * @again: set to false if the search has scanned the whole of RAM | |
b9e60928 | 1713 | */ |
f20e2865 | 1714 | static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again) |
b9e60928 | 1715 | { |
f20e2865 | 1716 | pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page); |
6f37bb8b | 1717 | if (pss->complete_round && pss->block == rs->last_seen_block && |
a935e30f | 1718 | pss->page >= rs->last_page) { |
b9e60928 DDAG |
1719 | /* |
1720 | * We've been once around the RAM and haven't found anything. | |
1721 | * Give up. | |
1722 | */ | |
1723 | *again = false; | |
1724 | return false; | |
1725 | } | |
a935e30f | 1726 | if ((pss->page << TARGET_PAGE_BITS) >= pss->block->used_length) { |
b9e60928 | 1727 | /* Didn't find anything in this RAM Block */ |
a935e30f | 1728 | pss->page = 0; |
b9e60928 DDAG |
1729 | pss->block = QLIST_NEXT_RCU(pss->block, next); |
1730 | if (!pss->block) { | |
1731 | /* Hit the end of the list */ | |
1732 | pss->block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1733 | /* Flag that we've looped */ | |
1734 | pss->complete_round = true; | |
6f37bb8b | 1735 | rs->ram_bulk_stage = false; |
b9e60928 DDAG |
1736 | if (migrate_use_xbzrle()) { |
1737 | /* If xbzrle is on, stop using the data compression at this | |
1738 | * point. In theory, xbzrle can do better than compression. | |
1739 | */ | |
ce25d337 | 1740 | flush_compressed_data(rs); |
b9e60928 DDAG |
1741 | } |
1742 | } | |
1743 | /* Didn't find anything this time, but try again on the new block */ | |
1744 | *again = true; | |
1745 | return false; | |
1746 | } else { | |
1747 | /* Can go around again, but... */ | |
1748 | *again = true; | |
1749 | /* We've found something so probably don't need to */ | |
1750 | return true; | |
1751 | } | |
1752 | } | |
1753 | ||
3d0684b2 JQ |
1754 | /** |
1755 | * unqueue_page: gets a page of the queue | |
1756 | * | |
a82d593b | 1757 | * Helper for 'get_queued_page' - gets a page off the queue |
a82d593b | 1758 | * |
3d0684b2 JQ |
1759 | * Returns the block of the page (or NULL if none available) |
1760 | * | |
ec481c6c | 1761 | * @rs: current RAM state |
3d0684b2 | 1762 | * @offset: used to return the offset within the RAMBlock |
a82d593b | 1763 | */ |
f20e2865 | 1764 | static RAMBlock *unqueue_page(RAMState *rs, ram_addr_t *offset) |
a82d593b DDAG |
1765 | { |
1766 | RAMBlock *block = NULL; | |
1767 | ||
ec481c6c JQ |
1768 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1769 | if (!QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
1770 | struct RAMSrcPageRequest *entry = | |
1771 | QSIMPLEQ_FIRST(&rs->src_page_requests); | |
a82d593b DDAG |
1772 | block = entry->rb; |
1773 | *offset = entry->offset; | |
a82d593b DDAG |
1774 | |
1775 | if (entry->len > TARGET_PAGE_SIZE) { | |
1776 | entry->len -= TARGET_PAGE_SIZE; | |
1777 | entry->offset += TARGET_PAGE_SIZE; | |
1778 | } else { | |
1779 | memory_region_unref(block->mr); | |
ec481c6c | 1780 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
a82d593b | 1781 | g_free(entry); |
e03a34f8 | 1782 | migration_consume_urgent_request(); |
a82d593b DDAG |
1783 | } |
1784 | } | |
ec481c6c | 1785 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
a82d593b DDAG |
1786 | |
1787 | return block; | |
1788 | } | |
1789 | ||
3d0684b2 JQ |
1790 | /** |
1791 | * get_queued_page: unqueue a page from the postocpy requests | |
1792 | * | |
1793 | * Skips pages that are already sent (!dirty) | |
a82d593b | 1794 | * |
3d0684b2 | 1795 | * Returns if a queued page is found |
a82d593b | 1796 | * |
6f37bb8b | 1797 | * @rs: current RAM state |
3d0684b2 | 1798 | * @pss: data about the state of the current dirty page scan |
a82d593b | 1799 | */ |
f20e2865 | 1800 | static bool get_queued_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b DDAG |
1801 | { |
1802 | RAMBlock *block; | |
1803 | ram_addr_t offset; | |
1804 | bool dirty; | |
1805 | ||
1806 | do { | |
f20e2865 | 1807 | block = unqueue_page(rs, &offset); |
a82d593b DDAG |
1808 | /* |
1809 | * We're sending this page, and since it's postcopy nothing else | |
1810 | * will dirty it, and we must make sure it doesn't get sent again | |
1811 | * even if this queue request was received after the background | |
1812 | * search already sent it. | |
1813 | */ | |
1814 | if (block) { | |
f20e2865 JQ |
1815 | unsigned long page; |
1816 | ||
6b6712ef JQ |
1817 | page = offset >> TARGET_PAGE_BITS; |
1818 | dirty = test_bit(page, block->bmap); | |
a82d593b | 1819 | if (!dirty) { |
06b10688 | 1820 | trace_get_queued_page_not_dirty(block->idstr, (uint64_t)offset, |
6b6712ef | 1821 | page, test_bit(page, block->unsentmap)); |
a82d593b | 1822 | } else { |
f20e2865 | 1823 | trace_get_queued_page(block->idstr, (uint64_t)offset, page); |
a82d593b DDAG |
1824 | } |
1825 | } | |
1826 | ||
1827 | } while (block && !dirty); | |
1828 | ||
1829 | if (block) { | |
1830 | /* | |
1831 | * As soon as we start servicing pages out of order, then we have | |
1832 | * to kill the bulk stage, since the bulk stage assumes | |
1833 | * in (migration_bitmap_find_and_reset_dirty) that every page is | |
1834 | * dirty, that's no longer true. | |
1835 | */ | |
6f37bb8b | 1836 | rs->ram_bulk_stage = false; |
a82d593b DDAG |
1837 | |
1838 | /* | |
1839 | * We want the background search to continue from the queued page | |
1840 | * since the guest is likely to want other pages near to the page | |
1841 | * it just requested. | |
1842 | */ | |
1843 | pss->block = block; | |
a935e30f | 1844 | pss->page = offset >> TARGET_PAGE_BITS; |
a82d593b DDAG |
1845 | } |
1846 | ||
1847 | return !!block; | |
1848 | } | |
1849 | ||
6c595cde | 1850 | /** |
5e58f968 JQ |
1851 | * migration_page_queue_free: drop any remaining pages in the ram |
1852 | * request queue | |
6c595cde | 1853 | * |
3d0684b2 JQ |
1854 | * It should be empty at the end anyway, but in error cases there may |
1855 | * be some left. in case that there is any page left, we drop it. | |
1856 | * | |
6c595cde | 1857 | */ |
83c13382 | 1858 | static void migration_page_queue_free(RAMState *rs) |
6c595cde | 1859 | { |
ec481c6c | 1860 | struct RAMSrcPageRequest *mspr, *next_mspr; |
6c595cde DDAG |
1861 | /* This queue generally should be empty - but in the case of a failed |
1862 | * migration might have some droppings in. | |
1863 | */ | |
1864 | rcu_read_lock(); | |
ec481c6c | 1865 | QSIMPLEQ_FOREACH_SAFE(mspr, &rs->src_page_requests, next_req, next_mspr) { |
6c595cde | 1866 | memory_region_unref(mspr->rb->mr); |
ec481c6c | 1867 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
6c595cde DDAG |
1868 | g_free(mspr); |
1869 | } | |
1870 | rcu_read_unlock(); | |
1871 | } | |
1872 | ||
1873 | /** | |
3d0684b2 JQ |
1874 | * ram_save_queue_pages: queue the page for transmission |
1875 | * | |
1876 | * A request from postcopy destination for example. | |
1877 | * | |
1878 | * Returns zero on success or negative on error | |
1879 | * | |
3d0684b2 JQ |
1880 | * @rbname: Name of the RAMBLock of the request. NULL means the |
1881 | * same that last one. | |
1882 | * @start: starting address from the start of the RAMBlock | |
1883 | * @len: length (in bytes) to send | |
6c595cde | 1884 | */ |
96506894 | 1885 | int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len) |
6c595cde DDAG |
1886 | { |
1887 | RAMBlock *ramblock; | |
53518d94 | 1888 | RAMState *rs = ram_state; |
6c595cde | 1889 | |
9360447d | 1890 | ram_counters.postcopy_requests++; |
6c595cde DDAG |
1891 | rcu_read_lock(); |
1892 | if (!rbname) { | |
1893 | /* Reuse last RAMBlock */ | |
68a098f3 | 1894 | ramblock = rs->last_req_rb; |
6c595cde DDAG |
1895 | |
1896 | if (!ramblock) { | |
1897 | /* | |
1898 | * Shouldn't happen, we can't reuse the last RAMBlock if | |
1899 | * it's the 1st request. | |
1900 | */ | |
1901 | error_report("ram_save_queue_pages no previous block"); | |
1902 | goto err; | |
1903 | } | |
1904 | } else { | |
1905 | ramblock = qemu_ram_block_by_name(rbname); | |
1906 | ||
1907 | if (!ramblock) { | |
1908 | /* We shouldn't be asked for a non-existent RAMBlock */ | |
1909 | error_report("ram_save_queue_pages no block '%s'", rbname); | |
1910 | goto err; | |
1911 | } | |
68a098f3 | 1912 | rs->last_req_rb = ramblock; |
6c595cde DDAG |
1913 | } |
1914 | trace_ram_save_queue_pages(ramblock->idstr, start, len); | |
1915 | if (start+len > ramblock->used_length) { | |
9458ad6b JQ |
1916 | error_report("%s request overrun start=" RAM_ADDR_FMT " len=" |
1917 | RAM_ADDR_FMT " blocklen=" RAM_ADDR_FMT, | |
6c595cde DDAG |
1918 | __func__, start, len, ramblock->used_length); |
1919 | goto err; | |
1920 | } | |
1921 | ||
ec481c6c JQ |
1922 | struct RAMSrcPageRequest *new_entry = |
1923 | g_malloc0(sizeof(struct RAMSrcPageRequest)); | |
6c595cde DDAG |
1924 | new_entry->rb = ramblock; |
1925 | new_entry->offset = start; | |
1926 | new_entry->len = len; | |
1927 | ||
1928 | memory_region_ref(ramblock->mr); | |
ec481c6c JQ |
1929 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1930 | QSIMPLEQ_INSERT_TAIL(&rs->src_page_requests, new_entry, next_req); | |
e03a34f8 | 1931 | migration_make_urgent_request(); |
ec481c6c | 1932 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
6c595cde DDAG |
1933 | rcu_read_unlock(); |
1934 | ||
1935 | return 0; | |
1936 | ||
1937 | err: | |
1938 | rcu_read_unlock(); | |
1939 | return -1; | |
1940 | } | |
1941 | ||
d7400a34 XG |
1942 | static bool save_page_use_compression(RAMState *rs) |
1943 | { | |
1944 | if (!migrate_use_compression()) { | |
1945 | return false; | |
1946 | } | |
1947 | ||
1948 | /* | |
1949 | * If xbzrle is on, stop using the data compression after first | |
1950 | * round of migration even if compression is enabled. In theory, | |
1951 | * xbzrle can do better than compression. | |
1952 | */ | |
1953 | if (rs->ram_bulk_stage || !migrate_use_xbzrle()) { | |
1954 | return true; | |
1955 | } | |
1956 | ||
1957 | return false; | |
1958 | } | |
1959 | ||
a82d593b | 1960 | /** |
3d0684b2 | 1961 | * ram_save_target_page: save one target page |
a82d593b | 1962 | * |
3d0684b2 | 1963 | * Returns the number of pages written |
a82d593b | 1964 | * |
6f37bb8b | 1965 | * @rs: current RAM state |
3d0684b2 | 1966 | * @pss: data about the page we want to send |
a82d593b | 1967 | * @last_stage: if we are at the completion stage |
a82d593b | 1968 | */ |
a0a8aa14 | 1969 | static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 1970 | bool last_stage) |
a82d593b | 1971 | { |
a8ec91f9 XG |
1972 | RAMBlock *block = pss->block; |
1973 | ram_addr_t offset = pss->page << TARGET_PAGE_BITS; | |
1974 | int res; | |
1975 | ||
1976 | if (control_save_page(rs, block, offset, &res)) { | |
1977 | return res; | |
1978 | } | |
1979 | ||
1faa5665 | 1980 | /* |
d7400a34 XG |
1981 | * When starting the process of a new block, the first page of |
1982 | * the block should be sent out before other pages in the same | |
1983 | * block, and all the pages in last block should have been sent | |
1984 | * out, keeping this order is important, because the 'cont' flag | |
1985 | * is used to avoid resending the block name. | |
1faa5665 | 1986 | */ |
d7400a34 XG |
1987 | if (block != rs->last_sent_block && save_page_use_compression(rs)) { |
1988 | flush_compressed_data(rs); | |
1989 | } | |
1990 | ||
1991 | res = save_zero_page(rs, block, offset); | |
1992 | if (res > 0) { | |
1993 | /* Must let xbzrle know, otherwise a previous (now 0'd) cached | |
1994 | * page would be stale | |
1995 | */ | |
1996 | if (!save_page_use_compression(rs)) { | |
1997 | XBZRLE_cache_lock(); | |
1998 | xbzrle_cache_zero_page(rs, block->offset + offset); | |
1999 | XBZRLE_cache_unlock(); | |
2000 | } | |
2001 | ram_release_pages(block->idstr, offset, res); | |
2002 | return res; | |
2003 | } | |
2004 | ||
da3f56cb XG |
2005 | /* |
2006 | * Make sure the first page is sent out before other pages. | |
2007 | * | |
2008 | * we post it as normal page as compression will take much | |
2009 | * CPU resource. | |
2010 | */ | |
2011 | if (block == rs->last_sent_block && save_page_use_compression(rs)) { | |
701b1876 | 2012 | return compress_page_with_multi_thread(rs, block, offset); |
a82d593b DDAG |
2013 | } |
2014 | ||
1faa5665 | 2015 | return ram_save_page(rs, pss, last_stage); |
a82d593b DDAG |
2016 | } |
2017 | ||
2018 | /** | |
3d0684b2 | 2019 | * ram_save_host_page: save a whole host page |
a82d593b | 2020 | * |
3d0684b2 JQ |
2021 | * Starting at *offset send pages up to the end of the current host |
2022 | * page. It's valid for the initial offset to point into the middle of | |
2023 | * a host page in which case the remainder of the hostpage is sent. | |
2024 | * Only dirty target pages are sent. Note that the host page size may | |
2025 | * be a huge page for this block. | |
1eb3fc0a DDAG |
2026 | * The saving stops at the boundary of the used_length of the block |
2027 | * if the RAMBlock isn't a multiple of the host page size. | |
a82d593b | 2028 | * |
3d0684b2 JQ |
2029 | * Returns the number of pages written or negative on error |
2030 | * | |
6f37bb8b | 2031 | * @rs: current RAM state |
3d0684b2 | 2032 | * @ms: current migration state |
3d0684b2 | 2033 | * @pss: data about the page we want to send |
a82d593b | 2034 | * @last_stage: if we are at the completion stage |
a82d593b | 2035 | */ |
a0a8aa14 | 2036 | static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss, |
f20e2865 | 2037 | bool last_stage) |
a82d593b DDAG |
2038 | { |
2039 | int tmppages, pages = 0; | |
a935e30f JQ |
2040 | size_t pagesize_bits = |
2041 | qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
4c011c37 | 2042 | |
b895de50 CLG |
2043 | if (!qemu_ram_is_migratable(pss->block)) { |
2044 | error_report("block %s should not be migrated !", pss->block->idstr); | |
2045 | return 0; | |
2046 | } | |
2047 | ||
a82d593b | 2048 | do { |
1faa5665 XG |
2049 | /* Check the pages is dirty and if it is send it */ |
2050 | if (!migration_bitmap_clear_dirty(rs, pss->block, pss->page)) { | |
2051 | pss->page++; | |
2052 | continue; | |
2053 | } | |
2054 | ||
f20e2865 | 2055 | tmppages = ram_save_target_page(rs, pss, last_stage); |
a82d593b DDAG |
2056 | if (tmppages < 0) { |
2057 | return tmppages; | |
2058 | } | |
2059 | ||
2060 | pages += tmppages; | |
1faa5665 XG |
2061 | if (pss->block->unsentmap) { |
2062 | clear_bit(pss->page, pss->block->unsentmap); | |
2063 | } | |
2064 | ||
a935e30f | 2065 | pss->page++; |
1eb3fc0a DDAG |
2066 | } while ((pss->page & (pagesize_bits - 1)) && |
2067 | offset_in_ramblock(pss->block, pss->page << TARGET_PAGE_BITS)); | |
a82d593b DDAG |
2068 | |
2069 | /* The offset we leave with is the last one we looked at */ | |
a935e30f | 2070 | pss->page--; |
a82d593b DDAG |
2071 | return pages; |
2072 | } | |
6c595cde | 2073 | |
56e93d26 | 2074 | /** |
3d0684b2 | 2075 | * ram_find_and_save_block: finds a dirty page and sends it to f |
56e93d26 JQ |
2076 | * |
2077 | * Called within an RCU critical section. | |
2078 | * | |
3d0684b2 | 2079 | * Returns the number of pages written where zero means no dirty pages |
56e93d26 | 2080 | * |
6f37bb8b | 2081 | * @rs: current RAM state |
56e93d26 | 2082 | * @last_stage: if we are at the completion stage |
a82d593b DDAG |
2083 | * |
2084 | * On systems where host-page-size > target-page-size it will send all the | |
2085 | * pages in a host page that are dirty. | |
56e93d26 JQ |
2086 | */ |
2087 | ||
ce25d337 | 2088 | static int ram_find_and_save_block(RAMState *rs, bool last_stage) |
56e93d26 | 2089 | { |
b8fb8cb7 | 2090 | PageSearchStatus pss; |
56e93d26 | 2091 | int pages = 0; |
b9e60928 | 2092 | bool again, found; |
56e93d26 | 2093 | |
0827b9e9 AA |
2094 | /* No dirty page as there is zero RAM */ |
2095 | if (!ram_bytes_total()) { | |
2096 | return pages; | |
2097 | } | |
2098 | ||
6f37bb8b | 2099 | pss.block = rs->last_seen_block; |
a935e30f | 2100 | pss.page = rs->last_page; |
b8fb8cb7 DDAG |
2101 | pss.complete_round = false; |
2102 | ||
2103 | if (!pss.block) { | |
2104 | pss.block = QLIST_FIRST_RCU(&ram_list.blocks); | |
2105 | } | |
56e93d26 | 2106 | |
b9e60928 | 2107 | do { |
a82d593b | 2108 | again = true; |
f20e2865 | 2109 | found = get_queued_page(rs, &pss); |
b9e60928 | 2110 | |
a82d593b DDAG |
2111 | if (!found) { |
2112 | /* priority queue empty, so just search for something dirty */ | |
f20e2865 | 2113 | found = find_dirty_block(rs, &pss, &again); |
a82d593b | 2114 | } |
f3f491fc | 2115 | |
a82d593b | 2116 | if (found) { |
f20e2865 | 2117 | pages = ram_save_host_page(rs, &pss, last_stage); |
56e93d26 | 2118 | } |
b9e60928 | 2119 | } while (!pages && again); |
56e93d26 | 2120 | |
6f37bb8b | 2121 | rs->last_seen_block = pss.block; |
a935e30f | 2122 | rs->last_page = pss.page; |
56e93d26 JQ |
2123 | |
2124 | return pages; | |
2125 | } | |
2126 | ||
2127 | void acct_update_position(QEMUFile *f, size_t size, bool zero) | |
2128 | { | |
2129 | uint64_t pages = size / TARGET_PAGE_SIZE; | |
f7ccd61b | 2130 | |
56e93d26 | 2131 | if (zero) { |
9360447d | 2132 | ram_counters.duplicate += pages; |
56e93d26 | 2133 | } else { |
9360447d JQ |
2134 | ram_counters.normal += pages; |
2135 | ram_counters.transferred += size; | |
56e93d26 JQ |
2136 | qemu_update_position(f, size); |
2137 | } | |
2138 | } | |
2139 | ||
56e93d26 JQ |
2140 | uint64_t ram_bytes_total(void) |
2141 | { | |
2142 | RAMBlock *block; | |
2143 | uint64_t total = 0; | |
2144 | ||
2145 | rcu_read_lock(); | |
b895de50 | 2146 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
56e93d26 | 2147 | total += block->used_length; |
99e15582 | 2148 | } |
56e93d26 JQ |
2149 | rcu_read_unlock(); |
2150 | return total; | |
2151 | } | |
2152 | ||
f265e0e4 | 2153 | static void xbzrle_load_setup(void) |
56e93d26 | 2154 | { |
f265e0e4 | 2155 | XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE); |
56e93d26 JQ |
2156 | } |
2157 | ||
f265e0e4 JQ |
2158 | static void xbzrle_load_cleanup(void) |
2159 | { | |
2160 | g_free(XBZRLE.decoded_buf); | |
2161 | XBZRLE.decoded_buf = NULL; | |
2162 | } | |
2163 | ||
7d7c96be PX |
2164 | static void ram_state_cleanup(RAMState **rsp) |
2165 | { | |
b9ccaf6d DDAG |
2166 | if (*rsp) { |
2167 | migration_page_queue_free(*rsp); | |
2168 | qemu_mutex_destroy(&(*rsp)->bitmap_mutex); | |
2169 | qemu_mutex_destroy(&(*rsp)->src_page_req_mutex); | |
2170 | g_free(*rsp); | |
2171 | *rsp = NULL; | |
2172 | } | |
7d7c96be PX |
2173 | } |
2174 | ||
84593a08 PX |
2175 | static void xbzrle_cleanup(void) |
2176 | { | |
2177 | XBZRLE_cache_lock(); | |
2178 | if (XBZRLE.cache) { | |
2179 | cache_fini(XBZRLE.cache); | |
2180 | g_free(XBZRLE.encoded_buf); | |
2181 | g_free(XBZRLE.current_buf); | |
2182 | g_free(XBZRLE.zero_target_page); | |
2183 | XBZRLE.cache = NULL; | |
2184 | XBZRLE.encoded_buf = NULL; | |
2185 | XBZRLE.current_buf = NULL; | |
2186 | XBZRLE.zero_target_page = NULL; | |
2187 | } | |
2188 | XBZRLE_cache_unlock(); | |
2189 | } | |
2190 | ||
f265e0e4 | 2191 | static void ram_save_cleanup(void *opaque) |
56e93d26 | 2192 | { |
53518d94 | 2193 | RAMState **rsp = opaque; |
6b6712ef | 2194 | RAMBlock *block; |
eb859c53 | 2195 | |
2ff64038 LZ |
2196 | /* caller have hold iothread lock or is in a bh, so there is |
2197 | * no writing race against this migration_bitmap | |
2198 | */ | |
6b6712ef JQ |
2199 | memory_global_dirty_log_stop(); |
2200 | ||
b895de50 | 2201 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
6b6712ef JQ |
2202 | g_free(block->bmap); |
2203 | block->bmap = NULL; | |
2204 | g_free(block->unsentmap); | |
2205 | block->unsentmap = NULL; | |
56e93d26 JQ |
2206 | } |
2207 | ||
84593a08 | 2208 | xbzrle_cleanup(); |
f0afa331 | 2209 | compress_threads_save_cleanup(); |
7d7c96be | 2210 | ram_state_cleanup(rsp); |
56e93d26 JQ |
2211 | } |
2212 | ||
6f37bb8b | 2213 | static void ram_state_reset(RAMState *rs) |
56e93d26 | 2214 | { |
6f37bb8b JQ |
2215 | rs->last_seen_block = NULL; |
2216 | rs->last_sent_block = NULL; | |
269ace29 | 2217 | rs->last_page = 0; |
6f37bb8b JQ |
2218 | rs->last_version = ram_list.version; |
2219 | rs->ram_bulk_stage = true; | |
56e93d26 JQ |
2220 | } |
2221 | ||
2222 | #define MAX_WAIT 50 /* ms, half buffered_file limit */ | |
2223 | ||
4f2e4252 DDAG |
2224 | /* |
2225 | * 'expected' is the value you expect the bitmap mostly to be full | |
2226 | * of; it won't bother printing lines that are all this value. | |
2227 | * If 'todump' is null the migration bitmap is dumped. | |
2228 | */ | |
6b6712ef JQ |
2229 | void ram_debug_dump_bitmap(unsigned long *todump, bool expected, |
2230 | unsigned long pages) | |
4f2e4252 | 2231 | { |
4f2e4252 DDAG |
2232 | int64_t cur; |
2233 | int64_t linelen = 128; | |
2234 | char linebuf[129]; | |
2235 | ||
6b6712ef | 2236 | for (cur = 0; cur < pages; cur += linelen) { |
4f2e4252 DDAG |
2237 | int64_t curb; |
2238 | bool found = false; | |
2239 | /* | |
2240 | * Last line; catch the case where the line length | |
2241 | * is longer than remaining ram | |
2242 | */ | |
6b6712ef JQ |
2243 | if (cur + linelen > pages) { |
2244 | linelen = pages - cur; | |
4f2e4252 DDAG |
2245 | } |
2246 | for (curb = 0; curb < linelen; curb++) { | |
2247 | bool thisbit = test_bit(cur + curb, todump); | |
2248 | linebuf[curb] = thisbit ? '1' : '.'; | |
2249 | found = found || (thisbit != expected); | |
2250 | } | |
2251 | if (found) { | |
2252 | linebuf[curb] = '\0'; | |
2253 | fprintf(stderr, "0x%08" PRIx64 " : %s\n", cur, linebuf); | |
2254 | } | |
2255 | } | |
2256 | } | |
2257 | ||
e0b266f0 DDAG |
2258 | /* **** functions for postcopy ***** */ |
2259 | ||
ced1c616 PB |
2260 | void ram_postcopy_migrated_memory_release(MigrationState *ms) |
2261 | { | |
2262 | struct RAMBlock *block; | |
ced1c616 | 2263 | |
b895de50 | 2264 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
6b6712ef JQ |
2265 | unsigned long *bitmap = block->bmap; |
2266 | unsigned long range = block->used_length >> TARGET_PAGE_BITS; | |
2267 | unsigned long run_start = find_next_zero_bit(bitmap, range, 0); | |
ced1c616 PB |
2268 | |
2269 | while (run_start < range) { | |
2270 | unsigned long run_end = find_next_bit(bitmap, range, run_start + 1); | |
aaa2064c | 2271 | ram_discard_range(block->idstr, run_start << TARGET_PAGE_BITS, |
ced1c616 PB |
2272 | (run_end - run_start) << TARGET_PAGE_BITS); |
2273 | run_start = find_next_zero_bit(bitmap, range, run_end + 1); | |
2274 | } | |
2275 | } | |
2276 | } | |
2277 | ||
3d0684b2 JQ |
2278 | /** |
2279 | * postcopy_send_discard_bm_ram: discard a RAMBlock | |
2280 | * | |
2281 | * Returns zero on success | |
2282 | * | |
e0b266f0 DDAG |
2283 | * Callback from postcopy_each_ram_send_discard for each RAMBlock |
2284 | * Note: At this point the 'unsentmap' is the processed bitmap combined | |
2285 | * with the dirtymap; so a '1' means it's either dirty or unsent. | |
3d0684b2 JQ |
2286 | * |
2287 | * @ms: current migration state | |
2288 | * @pds: state for postcopy | |
2289 | * @start: RAMBlock starting page | |
2290 | * @length: RAMBlock size | |
e0b266f0 DDAG |
2291 | */ |
2292 | static int postcopy_send_discard_bm_ram(MigrationState *ms, | |
2293 | PostcopyDiscardState *pds, | |
6b6712ef | 2294 | RAMBlock *block) |
e0b266f0 | 2295 | { |
6b6712ef | 2296 | unsigned long end = block->used_length >> TARGET_PAGE_BITS; |
e0b266f0 | 2297 | unsigned long current; |
6b6712ef | 2298 | unsigned long *unsentmap = block->unsentmap; |
e0b266f0 | 2299 | |
6b6712ef | 2300 | for (current = 0; current < end; ) { |
e0b266f0 DDAG |
2301 | unsigned long one = find_next_bit(unsentmap, end, current); |
2302 | ||
2303 | if (one <= end) { | |
2304 | unsigned long zero = find_next_zero_bit(unsentmap, end, one + 1); | |
2305 | unsigned long discard_length; | |
2306 | ||
2307 | if (zero >= end) { | |
2308 | discard_length = end - one; | |
2309 | } else { | |
2310 | discard_length = zero - one; | |
2311 | } | |
d688c62d DDAG |
2312 | if (discard_length) { |
2313 | postcopy_discard_send_range(ms, pds, one, discard_length); | |
2314 | } | |
e0b266f0 DDAG |
2315 | current = one + discard_length; |
2316 | } else { | |
2317 | current = one; | |
2318 | } | |
2319 | } | |
2320 | ||
2321 | return 0; | |
2322 | } | |
2323 | ||
3d0684b2 JQ |
2324 | /** |
2325 | * postcopy_each_ram_send_discard: discard all RAMBlocks | |
2326 | * | |
2327 | * Returns 0 for success or negative for error | |
2328 | * | |
e0b266f0 DDAG |
2329 | * Utility for the outgoing postcopy code. |
2330 | * Calls postcopy_send_discard_bm_ram for each RAMBlock | |
2331 | * passing it bitmap indexes and name. | |
e0b266f0 DDAG |
2332 | * (qemu_ram_foreach_block ends up passing unscaled lengths |
2333 | * which would mean postcopy code would have to deal with target page) | |
3d0684b2 JQ |
2334 | * |
2335 | * @ms: current migration state | |
e0b266f0 DDAG |
2336 | */ |
2337 | static int postcopy_each_ram_send_discard(MigrationState *ms) | |
2338 | { | |
2339 | struct RAMBlock *block; | |
2340 | int ret; | |
2341 | ||
b895de50 | 2342 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
6b6712ef JQ |
2343 | PostcopyDiscardState *pds = |
2344 | postcopy_discard_send_init(ms, block->idstr); | |
e0b266f0 DDAG |
2345 | |
2346 | /* | |
2347 | * Postcopy sends chunks of bitmap over the wire, but it | |
2348 | * just needs indexes at this point, avoids it having | |
2349 | * target page specific code. | |
2350 | */ | |
6b6712ef | 2351 | ret = postcopy_send_discard_bm_ram(ms, pds, block); |
e0b266f0 DDAG |
2352 | postcopy_discard_send_finish(ms, pds); |
2353 | if (ret) { | |
2354 | return ret; | |
2355 | } | |
2356 | } | |
2357 | ||
2358 | return 0; | |
2359 | } | |
2360 | ||
3d0684b2 JQ |
2361 | /** |
2362 | * postcopy_chunk_hostpages_pass: canocalize bitmap in hostpages | |
2363 | * | |
2364 | * Helper for postcopy_chunk_hostpages; it's called twice to | |
2365 | * canonicalize the two bitmaps, that are similar, but one is | |
2366 | * inverted. | |
99e314eb | 2367 | * |
3d0684b2 JQ |
2368 | * Postcopy requires that all target pages in a hostpage are dirty or |
2369 | * clean, not a mix. This function canonicalizes the bitmaps. | |
99e314eb | 2370 | * |
3d0684b2 JQ |
2371 | * @ms: current migration state |
2372 | * @unsent_pass: if true we need to canonicalize partially unsent host pages | |
2373 | * otherwise we need to canonicalize partially dirty host pages | |
2374 | * @block: block that contains the page we want to canonicalize | |
2375 | * @pds: state for postcopy | |
99e314eb DDAG |
2376 | */ |
2377 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, bool unsent_pass, | |
2378 | RAMBlock *block, | |
2379 | PostcopyDiscardState *pds) | |
2380 | { | |
53518d94 | 2381 | RAMState *rs = ram_state; |
6b6712ef JQ |
2382 | unsigned long *bitmap = block->bmap; |
2383 | unsigned long *unsentmap = block->unsentmap; | |
29c59172 | 2384 | unsigned int host_ratio = block->page_size / TARGET_PAGE_SIZE; |
6b6712ef | 2385 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
99e314eb DDAG |
2386 | unsigned long run_start; |
2387 | ||
29c59172 DDAG |
2388 | if (block->page_size == TARGET_PAGE_SIZE) { |
2389 | /* Easy case - TPS==HPS for a non-huge page RAMBlock */ | |
2390 | return; | |
2391 | } | |
2392 | ||
99e314eb DDAG |
2393 | if (unsent_pass) { |
2394 | /* Find a sent page */ | |
6b6712ef | 2395 | run_start = find_next_zero_bit(unsentmap, pages, 0); |
99e314eb DDAG |
2396 | } else { |
2397 | /* Find a dirty page */ | |
6b6712ef | 2398 | run_start = find_next_bit(bitmap, pages, 0); |
99e314eb DDAG |
2399 | } |
2400 | ||
6b6712ef | 2401 | while (run_start < pages) { |
99e314eb DDAG |
2402 | bool do_fixup = false; |
2403 | unsigned long fixup_start_addr; | |
2404 | unsigned long host_offset; | |
2405 | ||
2406 | /* | |
2407 | * If the start of this run of pages is in the middle of a host | |
2408 | * page, then we need to fixup this host page. | |
2409 | */ | |
2410 | host_offset = run_start % host_ratio; | |
2411 | if (host_offset) { | |
2412 | do_fixup = true; | |
2413 | run_start -= host_offset; | |
2414 | fixup_start_addr = run_start; | |
2415 | /* For the next pass */ | |
2416 | run_start = run_start + host_ratio; | |
2417 | } else { | |
2418 | /* Find the end of this run */ | |
2419 | unsigned long run_end; | |
2420 | if (unsent_pass) { | |
6b6712ef | 2421 | run_end = find_next_bit(unsentmap, pages, run_start + 1); |
99e314eb | 2422 | } else { |
6b6712ef | 2423 | run_end = find_next_zero_bit(bitmap, pages, run_start + 1); |
99e314eb DDAG |
2424 | } |
2425 | /* | |
2426 | * If the end isn't at the start of a host page, then the | |
2427 | * run doesn't finish at the end of a host page | |
2428 | * and we need to discard. | |
2429 | */ | |
2430 | host_offset = run_end % host_ratio; | |
2431 | if (host_offset) { | |
2432 | do_fixup = true; | |
2433 | fixup_start_addr = run_end - host_offset; | |
2434 | /* | |
2435 | * This host page has gone, the next loop iteration starts | |
2436 | * from after the fixup | |
2437 | */ | |
2438 | run_start = fixup_start_addr + host_ratio; | |
2439 | } else { | |
2440 | /* | |
2441 | * No discards on this iteration, next loop starts from | |
2442 | * next sent/dirty page | |
2443 | */ | |
2444 | run_start = run_end + 1; | |
2445 | } | |
2446 | } | |
2447 | ||
2448 | if (do_fixup) { | |
2449 | unsigned long page; | |
2450 | ||
2451 | /* Tell the destination to discard this page */ | |
2452 | if (unsent_pass || !test_bit(fixup_start_addr, unsentmap)) { | |
2453 | /* For the unsent_pass we: | |
2454 | * discard partially sent pages | |
2455 | * For the !unsent_pass (dirty) we: | |
2456 | * discard partially dirty pages that were sent | |
2457 | * (any partially sent pages were already discarded | |
2458 | * by the previous unsent_pass) | |
2459 | */ | |
2460 | postcopy_discard_send_range(ms, pds, fixup_start_addr, | |
2461 | host_ratio); | |
2462 | } | |
2463 | ||
2464 | /* Clean up the bitmap */ | |
2465 | for (page = fixup_start_addr; | |
2466 | page < fixup_start_addr + host_ratio; page++) { | |
2467 | /* All pages in this host page are now not sent */ | |
2468 | set_bit(page, unsentmap); | |
2469 | ||
2470 | /* | |
2471 | * Remark them as dirty, updating the count for any pages | |
2472 | * that weren't previously dirty. | |
2473 | */ | |
0d8ec885 | 2474 | rs->migration_dirty_pages += !test_and_set_bit(page, bitmap); |
99e314eb DDAG |
2475 | } |
2476 | } | |
2477 | ||
2478 | if (unsent_pass) { | |
2479 | /* Find the next sent page for the next iteration */ | |
6b6712ef | 2480 | run_start = find_next_zero_bit(unsentmap, pages, run_start); |
99e314eb DDAG |
2481 | } else { |
2482 | /* Find the next dirty page for the next iteration */ | |
6b6712ef | 2483 | run_start = find_next_bit(bitmap, pages, run_start); |
99e314eb DDAG |
2484 | } |
2485 | } | |
2486 | } | |
2487 | ||
3d0684b2 JQ |
2488 | /** |
2489 | * postcopy_chuck_hostpages: discrad any partially sent host page | |
2490 | * | |
99e314eb DDAG |
2491 | * Utility for the outgoing postcopy code. |
2492 | * | |
2493 | * Discard any partially sent host-page size chunks, mark any partially | |
29c59172 DDAG |
2494 | * dirty host-page size chunks as all dirty. In this case the host-page |
2495 | * is the host-page for the particular RAMBlock, i.e. it might be a huge page | |
99e314eb | 2496 | * |
3d0684b2 JQ |
2497 | * Returns zero on success |
2498 | * | |
2499 | * @ms: current migration state | |
6b6712ef | 2500 | * @block: block we want to work with |
99e314eb | 2501 | */ |
6b6712ef | 2502 | static int postcopy_chunk_hostpages(MigrationState *ms, RAMBlock *block) |
99e314eb | 2503 | { |
6b6712ef JQ |
2504 | PostcopyDiscardState *pds = |
2505 | postcopy_discard_send_init(ms, block->idstr); | |
99e314eb | 2506 | |
6b6712ef JQ |
2507 | /* First pass: Discard all partially sent host pages */ |
2508 | postcopy_chunk_hostpages_pass(ms, true, block, pds); | |
2509 | /* | |
2510 | * Second pass: Ensure that all partially dirty host pages are made | |
2511 | * fully dirty. | |
2512 | */ | |
2513 | postcopy_chunk_hostpages_pass(ms, false, block, pds); | |
99e314eb | 2514 | |
6b6712ef | 2515 | postcopy_discard_send_finish(ms, pds); |
99e314eb DDAG |
2516 | return 0; |
2517 | } | |
2518 | ||
3d0684b2 JQ |
2519 | /** |
2520 | * ram_postcopy_send_discard_bitmap: transmit the discard bitmap | |
2521 | * | |
2522 | * Returns zero on success | |
2523 | * | |
e0b266f0 DDAG |
2524 | * Transmit the set of pages to be discarded after precopy to the target |
2525 | * these are pages that: | |
2526 | * a) Have been previously transmitted but are now dirty again | |
2527 | * b) Pages that have never been transmitted, this ensures that | |
2528 | * any pages on the destination that have been mapped by background | |
2529 | * tasks get discarded (transparent huge pages is the specific concern) | |
2530 | * Hopefully this is pretty sparse | |
3d0684b2 JQ |
2531 | * |
2532 | * @ms: current migration state | |
e0b266f0 DDAG |
2533 | */ |
2534 | int ram_postcopy_send_discard_bitmap(MigrationState *ms) | |
2535 | { | |
53518d94 | 2536 | RAMState *rs = ram_state; |
6b6712ef | 2537 | RAMBlock *block; |
e0b266f0 | 2538 | int ret; |
e0b266f0 DDAG |
2539 | |
2540 | rcu_read_lock(); | |
2541 | ||
2542 | /* This should be our last sync, the src is now paused */ | |
eb859c53 | 2543 | migration_bitmap_sync(rs); |
e0b266f0 | 2544 | |
6b6712ef JQ |
2545 | /* Easiest way to make sure we don't resume in the middle of a host-page */ |
2546 | rs->last_seen_block = NULL; | |
2547 | rs->last_sent_block = NULL; | |
2548 | rs->last_page = 0; | |
e0b266f0 | 2549 | |
b895de50 | 2550 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
6b6712ef JQ |
2551 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
2552 | unsigned long *bitmap = block->bmap; | |
2553 | unsigned long *unsentmap = block->unsentmap; | |
2554 | ||
2555 | if (!unsentmap) { | |
2556 | /* We don't have a safe way to resize the sentmap, so | |
2557 | * if the bitmap was resized it will be NULL at this | |
2558 | * point. | |
2559 | */ | |
2560 | error_report("migration ram resized during precopy phase"); | |
2561 | rcu_read_unlock(); | |
2562 | return -EINVAL; | |
2563 | } | |
2564 | /* Deal with TPS != HPS and huge pages */ | |
2565 | ret = postcopy_chunk_hostpages(ms, block); | |
2566 | if (ret) { | |
2567 | rcu_read_unlock(); | |
2568 | return ret; | |
2569 | } | |
e0b266f0 | 2570 | |
6b6712ef JQ |
2571 | /* |
2572 | * Update the unsentmap to be unsentmap = unsentmap | dirty | |
2573 | */ | |
2574 | bitmap_or(unsentmap, unsentmap, bitmap, pages); | |
e0b266f0 | 2575 | #ifdef DEBUG_POSTCOPY |
6b6712ef | 2576 | ram_debug_dump_bitmap(unsentmap, true, pages); |
e0b266f0 | 2577 | #endif |
6b6712ef JQ |
2578 | } |
2579 | trace_ram_postcopy_send_discard_bitmap(); | |
e0b266f0 DDAG |
2580 | |
2581 | ret = postcopy_each_ram_send_discard(ms); | |
2582 | rcu_read_unlock(); | |
2583 | ||
2584 | return ret; | |
2585 | } | |
2586 | ||
3d0684b2 JQ |
2587 | /** |
2588 | * ram_discard_range: discard dirtied pages at the beginning of postcopy | |
e0b266f0 | 2589 | * |
3d0684b2 | 2590 | * Returns zero on success |
e0b266f0 | 2591 | * |
36449157 JQ |
2592 | * @rbname: name of the RAMBlock of the request. NULL means the |
2593 | * same that last one. | |
3d0684b2 JQ |
2594 | * @start: RAMBlock starting page |
2595 | * @length: RAMBlock size | |
e0b266f0 | 2596 | */ |
aaa2064c | 2597 | int ram_discard_range(const char *rbname, uint64_t start, size_t length) |
e0b266f0 DDAG |
2598 | { |
2599 | int ret = -1; | |
2600 | ||
36449157 | 2601 | trace_ram_discard_range(rbname, start, length); |
d3a5038c | 2602 | |
e0b266f0 | 2603 | rcu_read_lock(); |
36449157 | 2604 | RAMBlock *rb = qemu_ram_block_by_name(rbname); |
e0b266f0 DDAG |
2605 | |
2606 | if (!rb) { | |
36449157 | 2607 | error_report("ram_discard_range: Failed to find block '%s'", rbname); |
e0b266f0 DDAG |
2608 | goto err; |
2609 | } | |
2610 | ||
f9494614 AP |
2611 | bitmap_clear(rb->receivedmap, start >> qemu_target_page_bits(), |
2612 | length >> qemu_target_page_bits()); | |
d3a5038c | 2613 | ret = ram_block_discard_range(rb, start, length); |
e0b266f0 DDAG |
2614 | |
2615 | err: | |
2616 | rcu_read_unlock(); | |
2617 | ||
2618 | return ret; | |
2619 | } | |
2620 | ||
84593a08 PX |
2621 | /* |
2622 | * For every allocation, we will try not to crash the VM if the | |
2623 | * allocation failed. | |
2624 | */ | |
2625 | static int xbzrle_init(void) | |
2626 | { | |
2627 | Error *local_err = NULL; | |
2628 | ||
2629 | if (!migrate_use_xbzrle()) { | |
2630 | return 0; | |
2631 | } | |
2632 | ||
2633 | XBZRLE_cache_lock(); | |
2634 | ||
2635 | XBZRLE.zero_target_page = g_try_malloc0(TARGET_PAGE_SIZE); | |
2636 | if (!XBZRLE.zero_target_page) { | |
2637 | error_report("%s: Error allocating zero page", __func__); | |
2638 | goto err_out; | |
2639 | } | |
2640 | ||
2641 | XBZRLE.cache = cache_init(migrate_xbzrle_cache_size(), | |
2642 | TARGET_PAGE_SIZE, &local_err); | |
2643 | if (!XBZRLE.cache) { | |
2644 | error_report_err(local_err); | |
2645 | goto free_zero_page; | |
2646 | } | |
2647 | ||
2648 | XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); | |
2649 | if (!XBZRLE.encoded_buf) { | |
2650 | error_report("%s: Error allocating encoded_buf", __func__); | |
2651 | goto free_cache; | |
2652 | } | |
2653 | ||
2654 | XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); | |
2655 | if (!XBZRLE.current_buf) { | |
2656 | error_report("%s: Error allocating current_buf", __func__); | |
2657 | goto free_encoded_buf; | |
2658 | } | |
2659 | ||
2660 | /* We are all good */ | |
2661 | XBZRLE_cache_unlock(); | |
2662 | return 0; | |
2663 | ||
2664 | free_encoded_buf: | |
2665 | g_free(XBZRLE.encoded_buf); | |
2666 | XBZRLE.encoded_buf = NULL; | |
2667 | free_cache: | |
2668 | cache_fini(XBZRLE.cache); | |
2669 | XBZRLE.cache = NULL; | |
2670 | free_zero_page: | |
2671 | g_free(XBZRLE.zero_target_page); | |
2672 | XBZRLE.zero_target_page = NULL; | |
2673 | err_out: | |
2674 | XBZRLE_cache_unlock(); | |
2675 | return -ENOMEM; | |
2676 | } | |
2677 | ||
53518d94 | 2678 | static int ram_state_init(RAMState **rsp) |
56e93d26 | 2679 | { |
7d00ee6a PX |
2680 | *rsp = g_try_new0(RAMState, 1); |
2681 | ||
2682 | if (!*rsp) { | |
2683 | error_report("%s: Init ramstate fail", __func__); | |
2684 | return -1; | |
2685 | } | |
53518d94 JQ |
2686 | |
2687 | qemu_mutex_init(&(*rsp)->bitmap_mutex); | |
2688 | qemu_mutex_init(&(*rsp)->src_page_req_mutex); | |
2689 | QSIMPLEQ_INIT(&(*rsp)->src_page_requests); | |
56e93d26 | 2690 | |
7d00ee6a PX |
2691 | /* |
2692 | * Count the total number of pages used by ram blocks not including any | |
2693 | * gaps due to alignment or unplugs. | |
2694 | */ | |
2695 | (*rsp)->migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS; | |
2696 | ||
2697 | ram_state_reset(*rsp); | |
2698 | ||
2699 | return 0; | |
2700 | } | |
2701 | ||
d6eff5d7 | 2702 | static void ram_list_init_bitmaps(void) |
7d00ee6a | 2703 | { |
d6eff5d7 PX |
2704 | RAMBlock *block; |
2705 | unsigned long pages; | |
56e93d26 | 2706 | |
0827b9e9 AA |
2707 | /* Skip setting bitmap if there is no RAM */ |
2708 | if (ram_bytes_total()) { | |
b895de50 | 2709 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
d6eff5d7 | 2710 | pages = block->max_length >> TARGET_PAGE_BITS; |
6b6712ef JQ |
2711 | block->bmap = bitmap_new(pages); |
2712 | bitmap_set(block->bmap, 0, pages); | |
2713 | if (migrate_postcopy_ram()) { | |
2714 | block->unsentmap = bitmap_new(pages); | |
2715 | bitmap_set(block->unsentmap, 0, pages); | |
2716 | } | |
0827b9e9 | 2717 | } |
f3f491fc | 2718 | } |
d6eff5d7 PX |
2719 | } |
2720 | ||
2721 | static void ram_init_bitmaps(RAMState *rs) | |
2722 | { | |
2723 | /* For memory_global_dirty_log_start below. */ | |
2724 | qemu_mutex_lock_iothread(); | |
2725 | qemu_mutex_lock_ramlist(); | |
2726 | rcu_read_lock(); | |
f3f491fc | 2727 | |
d6eff5d7 | 2728 | ram_list_init_bitmaps(); |
56e93d26 | 2729 | memory_global_dirty_log_start(); |
d6eff5d7 PX |
2730 | migration_bitmap_sync(rs); |
2731 | ||
2732 | rcu_read_unlock(); | |
56e93d26 | 2733 | qemu_mutex_unlock_ramlist(); |
49877834 | 2734 | qemu_mutex_unlock_iothread(); |
d6eff5d7 PX |
2735 | } |
2736 | ||
2737 | static int ram_init_all(RAMState **rsp) | |
2738 | { | |
2739 | if (ram_state_init(rsp)) { | |
2740 | return -1; | |
2741 | } | |
2742 | ||
2743 | if (xbzrle_init()) { | |
2744 | ram_state_cleanup(rsp); | |
2745 | return -1; | |
2746 | } | |
2747 | ||
2748 | ram_init_bitmaps(*rsp); | |
a91246c9 HZ |
2749 | |
2750 | return 0; | |
2751 | } | |
2752 | ||
08614f34 PX |
2753 | static void ram_state_resume_prepare(RAMState *rs, QEMUFile *out) |
2754 | { | |
2755 | RAMBlock *block; | |
2756 | uint64_t pages = 0; | |
2757 | ||
2758 | /* | |
2759 | * Postcopy is not using xbzrle/compression, so no need for that. | |
2760 | * Also, since source are already halted, we don't need to care | |
2761 | * about dirty page logging as well. | |
2762 | */ | |
2763 | ||
ff0769a4 | 2764 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
08614f34 PX |
2765 | pages += bitmap_count_one(block->bmap, |
2766 | block->used_length >> TARGET_PAGE_BITS); | |
2767 | } | |
2768 | ||
2769 | /* This may not be aligned with current bitmaps. Recalculate. */ | |
2770 | rs->migration_dirty_pages = pages; | |
2771 | ||
2772 | rs->last_seen_block = NULL; | |
2773 | rs->last_sent_block = NULL; | |
2774 | rs->last_page = 0; | |
2775 | rs->last_version = ram_list.version; | |
2776 | /* | |
2777 | * Disable the bulk stage, otherwise we'll resend the whole RAM no | |
2778 | * matter what we have sent. | |
2779 | */ | |
2780 | rs->ram_bulk_stage = false; | |
2781 | ||
2782 | /* Update RAMState cache of output QEMUFile */ | |
2783 | rs->f = out; | |
2784 | ||
2785 | trace_ram_state_resume_prepare(pages); | |
2786 | } | |
2787 | ||
3d0684b2 JQ |
2788 | /* |
2789 | * Each of ram_save_setup, ram_save_iterate and ram_save_complete has | |
a91246c9 HZ |
2790 | * long-running RCU critical section. When rcu-reclaims in the code |
2791 | * start to become numerous it will be necessary to reduce the | |
2792 | * granularity of these critical sections. | |
2793 | */ | |
2794 | ||
3d0684b2 JQ |
2795 | /** |
2796 | * ram_save_setup: Setup RAM for migration | |
2797 | * | |
2798 | * Returns zero to indicate success and negative for error | |
2799 | * | |
2800 | * @f: QEMUFile where to send the data | |
2801 | * @opaque: RAMState pointer | |
2802 | */ | |
a91246c9 HZ |
2803 | static int ram_save_setup(QEMUFile *f, void *opaque) |
2804 | { | |
53518d94 | 2805 | RAMState **rsp = opaque; |
a91246c9 HZ |
2806 | RAMBlock *block; |
2807 | ||
dcaf446e XG |
2808 | if (compress_threads_save_setup()) { |
2809 | return -1; | |
2810 | } | |
2811 | ||
a91246c9 HZ |
2812 | /* migration has already setup the bitmap, reuse it. */ |
2813 | if (!migration_in_colo_state()) { | |
7d00ee6a | 2814 | if (ram_init_all(rsp) != 0) { |
dcaf446e | 2815 | compress_threads_save_cleanup(); |
a91246c9 | 2816 | return -1; |
53518d94 | 2817 | } |
a91246c9 | 2818 | } |
53518d94 | 2819 | (*rsp)->f = f; |
a91246c9 HZ |
2820 | |
2821 | rcu_read_lock(); | |
56e93d26 JQ |
2822 | |
2823 | qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE); | |
2824 | ||
b895de50 | 2825 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
56e93d26 JQ |
2826 | qemu_put_byte(f, strlen(block->idstr)); |
2827 | qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); | |
2828 | qemu_put_be64(f, block->used_length); | |
ef08fb38 DDAG |
2829 | if (migrate_postcopy_ram() && block->page_size != qemu_host_page_size) { |
2830 | qemu_put_be64(f, block->page_size); | |
2831 | } | |
56e93d26 JQ |
2832 | } |
2833 | ||
2834 | rcu_read_unlock(); | |
2835 | ||
2836 | ram_control_before_iterate(f, RAM_CONTROL_SETUP); | |
2837 | ram_control_after_iterate(f, RAM_CONTROL_SETUP); | |
2838 | ||
2839 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); | |
2840 | ||
2841 | return 0; | |
2842 | } | |
2843 | ||
3d0684b2 JQ |
2844 | /** |
2845 | * ram_save_iterate: iterative stage for migration | |
2846 | * | |
2847 | * Returns zero to indicate success and negative for error | |
2848 | * | |
2849 | * @f: QEMUFile where to send the data | |
2850 | * @opaque: RAMState pointer | |
2851 | */ | |
56e93d26 JQ |
2852 | static int ram_save_iterate(QEMUFile *f, void *opaque) |
2853 | { | |
53518d94 JQ |
2854 | RAMState **temp = opaque; |
2855 | RAMState *rs = *temp; | |
56e93d26 JQ |
2856 | int ret; |
2857 | int i; | |
2858 | int64_t t0; | |
5c90308f | 2859 | int done = 0; |
56e93d26 | 2860 | |
b2557345 PL |
2861 | if (blk_mig_bulk_active()) { |
2862 | /* Avoid transferring ram during bulk phase of block migration as | |
2863 | * the bulk phase will usually take a long time and transferring | |
2864 | * ram updates during that time is pointless. */ | |
2865 | goto out; | |
2866 | } | |
2867 | ||
56e93d26 | 2868 | rcu_read_lock(); |
6f37bb8b JQ |
2869 | if (ram_list.version != rs->last_version) { |
2870 | ram_state_reset(rs); | |
56e93d26 JQ |
2871 | } |
2872 | ||
2873 | /* Read version before ram_list.blocks */ | |
2874 | smp_rmb(); | |
2875 | ||
2876 | ram_control_before_iterate(f, RAM_CONTROL_ROUND); | |
2877 | ||
2878 | t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
2879 | i = 0; | |
e03a34f8 DDAG |
2880 | while ((ret = qemu_file_rate_limit(f)) == 0 || |
2881 | !QSIMPLEQ_EMPTY(&rs->src_page_requests)) { | |
56e93d26 JQ |
2882 | int pages; |
2883 | ||
e03a34f8 DDAG |
2884 | if (qemu_file_get_error(f)) { |
2885 | break; | |
2886 | } | |
2887 | ||
ce25d337 | 2888 | pages = ram_find_and_save_block(rs, false); |
56e93d26 JQ |
2889 | /* no more pages to sent */ |
2890 | if (pages == 0) { | |
5c90308f | 2891 | done = 1; |
56e93d26 JQ |
2892 | break; |
2893 | } | |
23b28c3c | 2894 | rs->iterations++; |
070afca2 | 2895 | |
56e93d26 JQ |
2896 | /* we want to check in the 1st loop, just in case it was the 1st time |
2897 | and we had to sync the dirty bitmap. | |
2898 | qemu_get_clock_ns() is a bit expensive, so we only check each some | |
2899 | iterations | |
2900 | */ | |
2901 | if ((i & 63) == 0) { | |
2902 | uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000; | |
2903 | if (t1 > MAX_WAIT) { | |
55c4446b | 2904 | trace_ram_save_iterate_big_wait(t1, i); |
56e93d26 JQ |
2905 | break; |
2906 | } | |
2907 | } | |
2908 | i++; | |
2909 | } | |
ce25d337 | 2910 | flush_compressed_data(rs); |
56e93d26 JQ |
2911 | rcu_read_unlock(); |
2912 | ||
2913 | /* | |
2914 | * Must occur before EOS (or any QEMUFile operation) | |
2915 | * because of RDMA protocol. | |
2916 | */ | |
2917 | ram_control_after_iterate(f, RAM_CONTROL_ROUND); | |
2918 | ||
b2557345 | 2919 | out: |
56e93d26 | 2920 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
9360447d | 2921 | ram_counters.transferred += 8; |
56e93d26 JQ |
2922 | |
2923 | ret = qemu_file_get_error(f); | |
2924 | if (ret < 0) { | |
2925 | return ret; | |
2926 | } | |
2927 | ||
5c90308f | 2928 | return done; |
56e93d26 JQ |
2929 | } |
2930 | ||
3d0684b2 JQ |
2931 | /** |
2932 | * ram_save_complete: function called to send the remaining amount of ram | |
2933 | * | |
2934 | * Returns zero to indicate success | |
2935 | * | |
2936 | * Called with iothread lock | |
2937 | * | |
2938 | * @f: QEMUFile where to send the data | |
2939 | * @opaque: RAMState pointer | |
2940 | */ | |
56e93d26 JQ |
2941 | static int ram_save_complete(QEMUFile *f, void *opaque) |
2942 | { | |
53518d94 JQ |
2943 | RAMState **temp = opaque; |
2944 | RAMState *rs = *temp; | |
6f37bb8b | 2945 | |
56e93d26 JQ |
2946 | rcu_read_lock(); |
2947 | ||
5727309d | 2948 | if (!migration_in_postcopy()) { |
8d820d6f | 2949 | migration_bitmap_sync(rs); |
663e6c1d | 2950 | } |
56e93d26 JQ |
2951 | |
2952 | ram_control_before_iterate(f, RAM_CONTROL_FINISH); | |
2953 | ||
2954 | /* try transferring iterative blocks of memory */ | |
2955 | ||
2956 | /* flush all remaining blocks regardless of rate limiting */ | |
2957 | while (true) { | |
2958 | int pages; | |
2959 | ||
ce25d337 | 2960 | pages = ram_find_and_save_block(rs, !migration_in_colo_state()); |
56e93d26 JQ |
2961 | /* no more blocks to sent */ |
2962 | if (pages == 0) { | |
2963 | break; | |
2964 | } | |
2965 | } | |
2966 | ||
ce25d337 | 2967 | flush_compressed_data(rs); |
56e93d26 | 2968 | ram_control_after_iterate(f, RAM_CONTROL_FINISH); |
56e93d26 JQ |
2969 | |
2970 | rcu_read_unlock(); | |
d09a6fde | 2971 | |
56e93d26 JQ |
2972 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
2973 | ||
2974 | return 0; | |
2975 | } | |
2976 | ||
c31b098f | 2977 | static void ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size, |
47995026 VSO |
2978 | uint64_t *res_precopy_only, |
2979 | uint64_t *res_compatible, | |
2980 | uint64_t *res_postcopy_only) | |
56e93d26 | 2981 | { |
53518d94 JQ |
2982 | RAMState **temp = opaque; |
2983 | RAMState *rs = *temp; | |
56e93d26 JQ |
2984 | uint64_t remaining_size; |
2985 | ||
9edabd4d | 2986 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2987 | |
5727309d | 2988 | if (!migration_in_postcopy() && |
663e6c1d | 2989 | remaining_size < max_size) { |
56e93d26 JQ |
2990 | qemu_mutex_lock_iothread(); |
2991 | rcu_read_lock(); | |
8d820d6f | 2992 | migration_bitmap_sync(rs); |
56e93d26 JQ |
2993 | rcu_read_unlock(); |
2994 | qemu_mutex_unlock_iothread(); | |
9edabd4d | 2995 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 2996 | } |
c31b098f | 2997 | |
86e1167e VSO |
2998 | if (migrate_postcopy_ram()) { |
2999 | /* We can do postcopy, and all the data is postcopiable */ | |
47995026 | 3000 | *res_compatible += remaining_size; |
86e1167e | 3001 | } else { |
47995026 | 3002 | *res_precopy_only += remaining_size; |
86e1167e | 3003 | } |
56e93d26 JQ |
3004 | } |
3005 | ||
3006 | static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) | |
3007 | { | |
3008 | unsigned int xh_len; | |
3009 | int xh_flags; | |
063e760a | 3010 | uint8_t *loaded_data; |
56e93d26 | 3011 | |
56e93d26 JQ |
3012 | /* extract RLE header */ |
3013 | xh_flags = qemu_get_byte(f); | |
3014 | xh_len = qemu_get_be16(f); | |
3015 | ||
3016 | if (xh_flags != ENCODING_FLAG_XBZRLE) { | |
3017 | error_report("Failed to load XBZRLE page - wrong compression!"); | |
3018 | return -1; | |
3019 | } | |
3020 | ||
3021 | if (xh_len > TARGET_PAGE_SIZE) { | |
3022 | error_report("Failed to load XBZRLE page - len overflow!"); | |
3023 | return -1; | |
3024 | } | |
f265e0e4 | 3025 | loaded_data = XBZRLE.decoded_buf; |
56e93d26 | 3026 | /* load data and decode */ |
f265e0e4 | 3027 | /* it can change loaded_data to point to an internal buffer */ |
063e760a | 3028 | qemu_get_buffer_in_place(f, &loaded_data, xh_len); |
56e93d26 JQ |
3029 | |
3030 | /* decode RLE */ | |
063e760a | 3031 | if (xbzrle_decode_buffer(loaded_data, xh_len, host, |
56e93d26 JQ |
3032 | TARGET_PAGE_SIZE) == -1) { |
3033 | error_report("Failed to load XBZRLE page - decode error!"); | |
3034 | return -1; | |
3035 | } | |
3036 | ||
3037 | return 0; | |
3038 | } | |
3039 | ||
3d0684b2 JQ |
3040 | /** |
3041 | * ram_block_from_stream: read a RAMBlock id from the migration stream | |
3042 | * | |
3043 | * Must be called from within a rcu critical section. | |
3044 | * | |
56e93d26 | 3045 | * Returns a pointer from within the RCU-protected ram_list. |
a7180877 | 3046 | * |
3d0684b2 JQ |
3047 | * @f: QEMUFile where to read the data from |
3048 | * @flags: Page flags (mostly to see if it's a continuation of previous block) | |
a7180877 | 3049 | */ |
3d0684b2 | 3050 | static inline RAMBlock *ram_block_from_stream(QEMUFile *f, int flags) |
56e93d26 JQ |
3051 | { |
3052 | static RAMBlock *block = NULL; | |
3053 | char id[256]; | |
3054 | uint8_t len; | |
3055 | ||
3056 | if (flags & RAM_SAVE_FLAG_CONTINUE) { | |
4c4bad48 | 3057 | if (!block) { |
56e93d26 JQ |
3058 | error_report("Ack, bad migration stream!"); |
3059 | return NULL; | |
3060 | } | |
4c4bad48 | 3061 | return block; |
56e93d26 JQ |
3062 | } |
3063 | ||
3064 | len = qemu_get_byte(f); | |
3065 | qemu_get_buffer(f, (uint8_t *)id, len); | |
3066 | id[len] = 0; | |
3067 | ||
e3dd7493 | 3068 | block = qemu_ram_block_by_name(id); |
4c4bad48 HZ |
3069 | if (!block) { |
3070 | error_report("Can't find block %s", id); | |
3071 | return NULL; | |
56e93d26 JQ |
3072 | } |
3073 | ||
b895de50 CLG |
3074 | if (!qemu_ram_is_migratable(block)) { |
3075 | error_report("block %s should not be migrated !", id); | |
3076 | return NULL; | |
3077 | } | |
3078 | ||
4c4bad48 HZ |
3079 | return block; |
3080 | } | |
3081 | ||
3082 | static inline void *host_from_ram_block_offset(RAMBlock *block, | |
3083 | ram_addr_t offset) | |
3084 | { | |
3085 | if (!offset_in_ramblock(block, offset)) { | |
3086 | return NULL; | |
3087 | } | |
3088 | ||
3089 | return block->host + offset; | |
56e93d26 JQ |
3090 | } |
3091 | ||
3d0684b2 JQ |
3092 | /** |
3093 | * ram_handle_compressed: handle the zero page case | |
3094 | * | |
56e93d26 JQ |
3095 | * If a page (or a whole RDMA chunk) has been |
3096 | * determined to be zero, then zap it. | |
3d0684b2 JQ |
3097 | * |
3098 | * @host: host address for the zero page | |
3099 | * @ch: what the page is filled from. We only support zero | |
3100 | * @size: size of the zero page | |
56e93d26 JQ |
3101 | */ |
3102 | void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) | |
3103 | { | |
3104 | if (ch != 0 || !is_zero_range(host, size)) { | |
3105 | memset(host, ch, size); | |
3106 | } | |
3107 | } | |
3108 | ||
797ca154 XG |
3109 | /* return the size after decompression, or negative value on error */ |
3110 | static int | |
3111 | qemu_uncompress_data(z_stream *stream, uint8_t *dest, size_t dest_len, | |
3112 | const uint8_t *source, size_t source_len) | |
3113 | { | |
3114 | int err; | |
3115 | ||
3116 | err = inflateReset(stream); | |
3117 | if (err != Z_OK) { | |
3118 | return -1; | |
3119 | } | |
3120 | ||
3121 | stream->avail_in = source_len; | |
3122 | stream->next_in = (uint8_t *)source; | |
3123 | stream->avail_out = dest_len; | |
3124 | stream->next_out = dest; | |
3125 | ||
3126 | err = inflate(stream, Z_NO_FLUSH); | |
3127 | if (err != Z_STREAM_END) { | |
3128 | return -1; | |
3129 | } | |
3130 | ||
3131 | return stream->total_out; | |
3132 | } | |
3133 | ||
56e93d26 JQ |
3134 | static void *do_data_decompress(void *opaque) |
3135 | { | |
3136 | DecompressParam *param = opaque; | |
3137 | unsigned long pagesize; | |
33d151f4 | 3138 | uint8_t *des; |
34ab9e97 | 3139 | int len, ret; |
56e93d26 | 3140 | |
33d151f4 | 3141 | qemu_mutex_lock(¶m->mutex); |
90e56fb4 | 3142 | while (!param->quit) { |
33d151f4 LL |
3143 | if (param->des) { |
3144 | des = param->des; | |
3145 | len = param->len; | |
3146 | param->des = 0; | |
3147 | qemu_mutex_unlock(¶m->mutex); | |
3148 | ||
56e93d26 | 3149 | pagesize = TARGET_PAGE_SIZE; |
34ab9e97 XG |
3150 | |
3151 | ret = qemu_uncompress_data(¶m->stream, des, pagesize, | |
3152 | param->compbuf, len); | |
f548222c | 3153 | if (ret < 0 && migrate_get_current()->decompress_error_check) { |
34ab9e97 XG |
3154 | error_report("decompress data failed"); |
3155 | qemu_file_set_error(decomp_file, ret); | |
3156 | } | |
73a8912b | 3157 | |
33d151f4 LL |
3158 | qemu_mutex_lock(&decomp_done_lock); |
3159 | param->done = true; | |
3160 | qemu_cond_signal(&decomp_done_cond); | |
3161 | qemu_mutex_unlock(&decomp_done_lock); | |
3162 | ||
3163 | qemu_mutex_lock(¶m->mutex); | |
3164 | } else { | |
3165 | qemu_cond_wait(¶m->cond, ¶m->mutex); | |
3166 | } | |
56e93d26 | 3167 | } |
33d151f4 | 3168 | qemu_mutex_unlock(¶m->mutex); |
56e93d26 JQ |
3169 | |
3170 | return NULL; | |
3171 | } | |
3172 | ||
34ab9e97 | 3173 | static int wait_for_decompress_done(void) |
5533b2e9 LL |
3174 | { |
3175 | int idx, thread_count; | |
3176 | ||
3177 | if (!migrate_use_compression()) { | |
34ab9e97 | 3178 | return 0; |
5533b2e9 LL |
3179 | } |
3180 | ||
3181 | thread_count = migrate_decompress_threads(); | |
3182 | qemu_mutex_lock(&decomp_done_lock); | |
3183 | for (idx = 0; idx < thread_count; idx++) { | |
3184 | while (!decomp_param[idx].done) { | |
3185 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
3186 | } | |
3187 | } | |
3188 | qemu_mutex_unlock(&decomp_done_lock); | |
34ab9e97 | 3189 | return qemu_file_get_error(decomp_file); |
5533b2e9 LL |
3190 | } |
3191 | ||
f0afa331 | 3192 | static void compress_threads_load_cleanup(void) |
56e93d26 JQ |
3193 | { |
3194 | int i, thread_count; | |
3195 | ||
3416ab5b JQ |
3196 | if (!migrate_use_compression()) { |
3197 | return; | |
3198 | } | |
56e93d26 JQ |
3199 | thread_count = migrate_decompress_threads(); |
3200 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
3201 | /* |
3202 | * we use it as a indicator which shows if the thread is | |
3203 | * properly init'd or not | |
3204 | */ | |
3205 | if (!decomp_param[i].compbuf) { | |
3206 | break; | |
3207 | } | |
3208 | ||
56e93d26 | 3209 | qemu_mutex_lock(&decomp_param[i].mutex); |
90e56fb4 | 3210 | decomp_param[i].quit = true; |
56e93d26 JQ |
3211 | qemu_cond_signal(&decomp_param[i].cond); |
3212 | qemu_mutex_unlock(&decomp_param[i].mutex); | |
3213 | } | |
3214 | for (i = 0; i < thread_count; i++) { | |
797ca154 XG |
3215 | if (!decomp_param[i].compbuf) { |
3216 | break; | |
3217 | } | |
3218 | ||
56e93d26 JQ |
3219 | qemu_thread_join(decompress_threads + i); |
3220 | qemu_mutex_destroy(&decomp_param[i].mutex); | |
3221 | qemu_cond_destroy(&decomp_param[i].cond); | |
797ca154 | 3222 | inflateEnd(&decomp_param[i].stream); |
56e93d26 | 3223 | g_free(decomp_param[i].compbuf); |
797ca154 | 3224 | decomp_param[i].compbuf = NULL; |
56e93d26 JQ |
3225 | } |
3226 | g_free(decompress_threads); | |
3227 | g_free(decomp_param); | |
56e93d26 JQ |
3228 | decompress_threads = NULL; |
3229 | decomp_param = NULL; | |
34ab9e97 | 3230 | decomp_file = NULL; |
56e93d26 JQ |
3231 | } |
3232 | ||
34ab9e97 | 3233 | static int compress_threads_load_setup(QEMUFile *f) |
797ca154 XG |
3234 | { |
3235 | int i, thread_count; | |
3236 | ||
3237 | if (!migrate_use_compression()) { | |
3238 | return 0; | |
3239 | } | |
3240 | ||
3241 | thread_count = migrate_decompress_threads(); | |
3242 | decompress_threads = g_new0(QemuThread, thread_count); | |
3243 | decomp_param = g_new0(DecompressParam, thread_count); | |
3244 | qemu_mutex_init(&decomp_done_lock); | |
3245 | qemu_cond_init(&decomp_done_cond); | |
34ab9e97 | 3246 | decomp_file = f; |
797ca154 XG |
3247 | for (i = 0; i < thread_count; i++) { |
3248 | if (inflateInit(&decomp_param[i].stream) != Z_OK) { | |
3249 | goto exit; | |
3250 | } | |
3251 | ||
3252 | decomp_param[i].compbuf = g_malloc0(compressBound(TARGET_PAGE_SIZE)); | |
3253 | qemu_mutex_init(&decomp_param[i].mutex); | |
3254 | qemu_cond_init(&decomp_param[i].cond); | |
3255 | decomp_param[i].done = true; | |
3256 | decomp_param[i].quit = false; | |
3257 | qemu_thread_create(decompress_threads + i, "decompress", | |
3258 | do_data_decompress, decomp_param + i, | |
3259 | QEMU_THREAD_JOINABLE); | |
3260 | } | |
3261 | return 0; | |
3262 | exit: | |
3263 | compress_threads_load_cleanup(); | |
3264 | return -1; | |
3265 | } | |
3266 | ||
c1bc6626 | 3267 | static void decompress_data_with_multi_threads(QEMUFile *f, |
56e93d26 JQ |
3268 | void *host, int len) |
3269 | { | |
3270 | int idx, thread_count; | |
3271 | ||
3272 | thread_count = migrate_decompress_threads(); | |
73a8912b | 3273 | qemu_mutex_lock(&decomp_done_lock); |
56e93d26 JQ |
3274 | while (true) { |
3275 | for (idx = 0; idx < thread_count; idx++) { | |
73a8912b | 3276 | if (decomp_param[idx].done) { |
33d151f4 LL |
3277 | decomp_param[idx].done = false; |
3278 | qemu_mutex_lock(&decomp_param[idx].mutex); | |
c1bc6626 | 3279 | qemu_get_buffer(f, decomp_param[idx].compbuf, len); |
56e93d26 JQ |
3280 | decomp_param[idx].des = host; |
3281 | decomp_param[idx].len = len; | |
33d151f4 LL |
3282 | qemu_cond_signal(&decomp_param[idx].cond); |
3283 | qemu_mutex_unlock(&decomp_param[idx].mutex); | |
56e93d26 JQ |
3284 | break; |
3285 | } | |
3286 | } | |
3287 | if (idx < thread_count) { | |
3288 | break; | |
73a8912b LL |
3289 | } else { |
3290 | qemu_cond_wait(&decomp_done_cond, &decomp_done_lock); | |
56e93d26 JQ |
3291 | } |
3292 | } | |
73a8912b | 3293 | qemu_mutex_unlock(&decomp_done_lock); |
56e93d26 JQ |
3294 | } |
3295 | ||
f265e0e4 JQ |
3296 | /** |
3297 | * ram_load_setup: Setup RAM for migration incoming side | |
3298 | * | |
3299 | * Returns zero to indicate success and negative for error | |
3300 | * | |
3301 | * @f: QEMUFile where to receive the data | |
3302 | * @opaque: RAMState pointer | |
3303 | */ | |
3304 | static int ram_load_setup(QEMUFile *f, void *opaque) | |
3305 | { | |
34ab9e97 | 3306 | if (compress_threads_load_setup(f)) { |
797ca154 XG |
3307 | return -1; |
3308 | } | |
3309 | ||
f265e0e4 | 3310 | xbzrle_load_setup(); |
f9494614 | 3311 | ramblock_recv_map_init(); |
f265e0e4 JQ |
3312 | return 0; |
3313 | } | |
3314 | ||
3315 | static int ram_load_cleanup(void *opaque) | |
3316 | { | |
f9494614 | 3317 | RAMBlock *rb; |
f265e0e4 | 3318 | xbzrle_load_cleanup(); |
f0afa331 | 3319 | compress_threads_load_cleanup(); |
f9494614 | 3320 | |
b895de50 | 3321 | RAMBLOCK_FOREACH_MIGRATABLE(rb) { |
f9494614 AP |
3322 | g_free(rb->receivedmap); |
3323 | rb->receivedmap = NULL; | |
3324 | } | |
f265e0e4 JQ |
3325 | return 0; |
3326 | } | |
3327 | ||
3d0684b2 JQ |
3328 | /** |
3329 | * ram_postcopy_incoming_init: allocate postcopy data structures | |
3330 | * | |
3331 | * Returns 0 for success and negative if there was one error | |
3332 | * | |
3333 | * @mis: current migration incoming state | |
3334 | * | |
3335 | * Allocate data structures etc needed by incoming migration with | |
3336 | * postcopy-ram. postcopy-ram's similarly names | |
3337 | * postcopy_ram_incoming_init does the work. | |
1caddf8a DDAG |
3338 | */ |
3339 | int ram_postcopy_incoming_init(MigrationIncomingState *mis) | |
3340 | { | |
b8c48993 | 3341 | unsigned long ram_pages = last_ram_page(); |
1caddf8a DDAG |
3342 | |
3343 | return postcopy_ram_incoming_init(mis, ram_pages); | |
3344 | } | |
3345 | ||
3d0684b2 JQ |
3346 | /** |
3347 | * ram_load_postcopy: load a page in postcopy case | |
3348 | * | |
3349 | * Returns 0 for success or -errno in case of error | |
3350 | * | |
a7180877 DDAG |
3351 | * Called in postcopy mode by ram_load(). |
3352 | * rcu_read_lock is taken prior to this being called. | |
3d0684b2 JQ |
3353 | * |
3354 | * @f: QEMUFile where to send the data | |
a7180877 DDAG |
3355 | */ |
3356 | static int ram_load_postcopy(QEMUFile *f) | |
3357 | { | |
3358 | int flags = 0, ret = 0; | |
3359 | bool place_needed = false; | |
28abd200 | 3360 | bool matching_page_sizes = false; |
a7180877 DDAG |
3361 | MigrationIncomingState *mis = migration_incoming_get_current(); |
3362 | /* Temporary page that is later 'placed' */ | |
3363 | void *postcopy_host_page = postcopy_get_tmp_page(mis); | |
c53b7ddc | 3364 | void *last_host = NULL; |
a3b6ff6d | 3365 | bool all_zero = false; |
a7180877 DDAG |
3366 | |
3367 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
3368 | ram_addr_t addr; | |
3369 | void *host = NULL; | |
3370 | void *page_buffer = NULL; | |
3371 | void *place_source = NULL; | |
df9ff5e1 | 3372 | RAMBlock *block = NULL; |
a7180877 | 3373 | uint8_t ch; |
a7180877 DDAG |
3374 | |
3375 | addr = qemu_get_be64(f); | |
7a9ddfbf PX |
3376 | |
3377 | /* | |
3378 | * If qemu file error, we should stop here, and then "addr" | |
3379 | * may be invalid | |
3380 | */ | |
3381 | ret = qemu_file_get_error(f); | |
3382 | if (ret) { | |
3383 | break; | |
3384 | } | |
3385 | ||
a7180877 DDAG |
3386 | flags = addr & ~TARGET_PAGE_MASK; |
3387 | addr &= TARGET_PAGE_MASK; | |
3388 | ||
3389 | trace_ram_load_postcopy_loop((uint64_t)addr, flags); | |
3390 | place_needed = false; | |
bb890ed5 | 3391 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE)) { |
df9ff5e1 | 3392 | block = ram_block_from_stream(f, flags); |
4c4bad48 HZ |
3393 | |
3394 | host = host_from_ram_block_offset(block, addr); | |
a7180877 DDAG |
3395 | if (!host) { |
3396 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
3397 | ret = -EINVAL; | |
3398 | break; | |
3399 | } | |
28abd200 | 3400 | matching_page_sizes = block->page_size == TARGET_PAGE_SIZE; |
a7180877 | 3401 | /* |
28abd200 DDAG |
3402 | * Postcopy requires that we place whole host pages atomically; |
3403 | * these may be huge pages for RAMBlocks that are backed by | |
3404 | * hugetlbfs. | |
a7180877 DDAG |
3405 | * To make it atomic, the data is read into a temporary page |
3406 | * that's moved into place later. | |
3407 | * The migration protocol uses, possibly smaller, target-pages | |
3408 | * however the source ensures it always sends all the components | |
3409 | * of a host page in order. | |
3410 | */ | |
3411 | page_buffer = postcopy_host_page + | |
28abd200 | 3412 | ((uintptr_t)host & (block->page_size - 1)); |
a7180877 | 3413 | /* If all TP are zero then we can optimise the place */ |
28abd200 | 3414 | if (!((uintptr_t)host & (block->page_size - 1))) { |
a7180877 | 3415 | all_zero = true; |
c53b7ddc DDAG |
3416 | } else { |
3417 | /* not the 1st TP within the HP */ | |
3418 | if (host != (last_host + TARGET_PAGE_SIZE)) { | |
9af9e0fe | 3419 | error_report("Non-sequential target page %p/%p", |
c53b7ddc DDAG |
3420 | host, last_host); |
3421 | ret = -EINVAL; | |
3422 | break; | |
3423 | } | |
a7180877 DDAG |
3424 | } |
3425 | ||
c53b7ddc | 3426 | |
a7180877 DDAG |
3427 | /* |
3428 | * If it's the last part of a host page then we place the host | |
3429 | * page | |
3430 | */ | |
3431 | place_needed = (((uintptr_t)host + TARGET_PAGE_SIZE) & | |
28abd200 | 3432 | (block->page_size - 1)) == 0; |
a7180877 DDAG |
3433 | place_source = postcopy_host_page; |
3434 | } | |
c53b7ddc | 3435 | last_host = host; |
a7180877 DDAG |
3436 | |
3437 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { | |
bb890ed5 | 3438 | case RAM_SAVE_FLAG_ZERO: |
a7180877 DDAG |
3439 | ch = qemu_get_byte(f); |
3440 | memset(page_buffer, ch, TARGET_PAGE_SIZE); | |
3441 | if (ch) { | |
3442 | all_zero = false; | |
3443 | } | |
3444 | break; | |
3445 | ||
3446 | case RAM_SAVE_FLAG_PAGE: | |
3447 | all_zero = false; | |
3448 | if (!place_needed || !matching_page_sizes) { | |
3449 | qemu_get_buffer(f, page_buffer, TARGET_PAGE_SIZE); | |
3450 | } else { | |
3451 | /* Avoids the qemu_file copy during postcopy, which is | |
3452 | * going to do a copy later; can only do it when we | |
3453 | * do this read in one go (matching page sizes) | |
3454 | */ | |
3455 | qemu_get_buffer_in_place(f, (uint8_t **)&place_source, | |
3456 | TARGET_PAGE_SIZE); | |
3457 | } | |
3458 | break; | |
3459 | case RAM_SAVE_FLAG_EOS: | |
3460 | /* normal exit */ | |
3461 | break; | |
3462 | default: | |
3463 | error_report("Unknown combination of migration flags: %#x" | |
3464 | " (postcopy mode)", flags); | |
3465 | ret = -EINVAL; | |
7a9ddfbf PX |
3466 | break; |
3467 | } | |
3468 | ||
3469 | /* Detect for any possible file errors */ | |
3470 | if (!ret && qemu_file_get_error(f)) { | |
3471 | ret = qemu_file_get_error(f); | |
a7180877 DDAG |
3472 | } |
3473 | ||
7a9ddfbf | 3474 | if (!ret && place_needed) { |
a7180877 | 3475 | /* This gets called at the last target page in the host page */ |
df9ff5e1 DDAG |
3476 | void *place_dest = host + TARGET_PAGE_SIZE - block->page_size; |
3477 | ||
a7180877 | 3478 | if (all_zero) { |
df9ff5e1 | 3479 | ret = postcopy_place_page_zero(mis, place_dest, |
8be4620b | 3480 | block); |
a7180877 | 3481 | } else { |
df9ff5e1 | 3482 | ret = postcopy_place_page(mis, place_dest, |
8be4620b | 3483 | place_source, block); |
a7180877 DDAG |
3484 | } |
3485 | } | |
a7180877 DDAG |
3486 | } |
3487 | ||
3488 | return ret; | |
3489 | } | |
3490 | ||
acab30b8 DHB |
3491 | static bool postcopy_is_advised(void) |
3492 | { | |
3493 | PostcopyState ps = postcopy_state_get(); | |
3494 | return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END; | |
3495 | } | |
3496 | ||
3497 | static bool postcopy_is_running(void) | |
3498 | { | |
3499 | PostcopyState ps = postcopy_state_get(); | |
3500 | return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; | |
3501 | } | |
3502 | ||
56e93d26 JQ |
3503 | static int ram_load(QEMUFile *f, void *opaque, int version_id) |
3504 | { | |
edc60127 | 3505 | int flags = 0, ret = 0, invalid_flags = 0; |
56e93d26 JQ |
3506 | static uint64_t seq_iter; |
3507 | int len = 0; | |
a7180877 DDAG |
3508 | /* |
3509 | * If system is running in postcopy mode, page inserts to host memory must | |
3510 | * be atomic | |
3511 | */ | |
acab30b8 | 3512 | bool postcopy_running = postcopy_is_running(); |
ef08fb38 | 3513 | /* ADVISE is earlier, it shows the source has the postcopy capability on */ |
acab30b8 | 3514 | bool postcopy_advised = postcopy_is_advised(); |
56e93d26 JQ |
3515 | |
3516 | seq_iter++; | |
3517 | ||
3518 | if (version_id != 4) { | |
3519 | ret = -EINVAL; | |
3520 | } | |
3521 | ||
edc60127 JQ |
3522 | if (!migrate_use_compression()) { |
3523 | invalid_flags |= RAM_SAVE_FLAG_COMPRESS_PAGE; | |
3524 | } | |
56e93d26 JQ |
3525 | /* This RCU critical section can be very long running. |
3526 | * When RCU reclaims in the code start to become numerous, | |
3527 | * it will be necessary to reduce the granularity of this | |
3528 | * critical section. | |
3529 | */ | |
3530 | rcu_read_lock(); | |
a7180877 DDAG |
3531 | |
3532 | if (postcopy_running) { | |
3533 | ret = ram_load_postcopy(f); | |
3534 | } | |
3535 | ||
3536 | while (!postcopy_running && !ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
56e93d26 | 3537 | ram_addr_t addr, total_ram_bytes; |
a776aa15 | 3538 | void *host = NULL; |
56e93d26 JQ |
3539 | uint8_t ch; |
3540 | ||
3541 | addr = qemu_get_be64(f); | |
3542 | flags = addr & ~TARGET_PAGE_MASK; | |
3543 | addr &= TARGET_PAGE_MASK; | |
3544 | ||
edc60127 JQ |
3545 | if (flags & invalid_flags) { |
3546 | if (flags & invalid_flags & RAM_SAVE_FLAG_COMPRESS_PAGE) { | |
3547 | error_report("Received an unexpected compressed page"); | |
3548 | } | |
3549 | ||
3550 | ret = -EINVAL; | |
3551 | break; | |
3552 | } | |
3553 | ||
bb890ed5 | 3554 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
a776aa15 | 3555 | RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) { |
4c4bad48 HZ |
3556 | RAMBlock *block = ram_block_from_stream(f, flags); |
3557 | ||
3558 | host = host_from_ram_block_offset(block, addr); | |
a776aa15 DDAG |
3559 | if (!host) { |
3560 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
3561 | ret = -EINVAL; | |
3562 | break; | |
3563 | } | |
f9494614 | 3564 | ramblock_recv_bitmap_set(block, host); |
1db9d8e5 | 3565 | trace_ram_load_loop(block->idstr, (uint64_t)addr, flags, host); |
a776aa15 DDAG |
3566 | } |
3567 | ||
56e93d26 JQ |
3568 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { |
3569 | case RAM_SAVE_FLAG_MEM_SIZE: | |
3570 | /* Synchronize RAM block list */ | |
3571 | total_ram_bytes = addr; | |
3572 | while (!ret && total_ram_bytes) { | |
3573 | RAMBlock *block; | |
56e93d26 JQ |
3574 | char id[256]; |
3575 | ram_addr_t length; | |
3576 | ||
3577 | len = qemu_get_byte(f); | |
3578 | qemu_get_buffer(f, (uint8_t *)id, len); | |
3579 | id[len] = 0; | |
3580 | length = qemu_get_be64(f); | |
3581 | ||
e3dd7493 | 3582 | block = qemu_ram_block_by_name(id); |
b895de50 CLG |
3583 | if (block && !qemu_ram_is_migratable(block)) { |
3584 | error_report("block %s should not be migrated !", id); | |
3585 | ret = -EINVAL; | |
3586 | } else if (block) { | |
e3dd7493 DDAG |
3587 | if (length != block->used_length) { |
3588 | Error *local_err = NULL; | |
56e93d26 | 3589 | |
fa53a0e5 | 3590 | ret = qemu_ram_resize(block, length, |
e3dd7493 DDAG |
3591 | &local_err); |
3592 | if (local_err) { | |
3593 | error_report_err(local_err); | |
56e93d26 | 3594 | } |
56e93d26 | 3595 | } |
ef08fb38 DDAG |
3596 | /* For postcopy we need to check hugepage sizes match */ |
3597 | if (postcopy_advised && | |
3598 | block->page_size != qemu_host_page_size) { | |
3599 | uint64_t remote_page_size = qemu_get_be64(f); | |
3600 | if (remote_page_size != block->page_size) { | |
3601 | error_report("Mismatched RAM page size %s " | |
3602 | "(local) %zd != %" PRId64, | |
3603 | id, block->page_size, | |
3604 | remote_page_size); | |
3605 | ret = -EINVAL; | |
3606 | } | |
3607 | } | |
e3dd7493 DDAG |
3608 | ram_control_load_hook(f, RAM_CONTROL_BLOCK_REG, |
3609 | block->idstr); | |
3610 | } else { | |
56e93d26 JQ |
3611 | error_report("Unknown ramblock \"%s\", cannot " |
3612 | "accept migration", id); | |
3613 | ret = -EINVAL; | |
3614 | } | |
3615 | ||
3616 | total_ram_bytes -= length; | |
3617 | } | |
3618 | break; | |
a776aa15 | 3619 | |
bb890ed5 | 3620 | case RAM_SAVE_FLAG_ZERO: |
56e93d26 JQ |
3621 | ch = qemu_get_byte(f); |
3622 | ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); | |
3623 | break; | |
a776aa15 | 3624 | |
56e93d26 | 3625 | case RAM_SAVE_FLAG_PAGE: |
56e93d26 JQ |
3626 | qemu_get_buffer(f, host, TARGET_PAGE_SIZE); |
3627 | break; | |
56e93d26 | 3628 | |
a776aa15 | 3629 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
56e93d26 JQ |
3630 | len = qemu_get_be32(f); |
3631 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
3632 | error_report("Invalid compressed data length: %d", len); | |
3633 | ret = -EINVAL; | |
3634 | break; | |
3635 | } | |
c1bc6626 | 3636 | decompress_data_with_multi_threads(f, host, len); |
56e93d26 | 3637 | break; |
a776aa15 | 3638 | |
56e93d26 | 3639 | case RAM_SAVE_FLAG_XBZRLE: |
56e93d26 JQ |
3640 | if (load_xbzrle(f, addr, host) < 0) { |
3641 | error_report("Failed to decompress XBZRLE page at " | |
3642 | RAM_ADDR_FMT, addr); | |
3643 | ret = -EINVAL; | |
3644 | break; | |
3645 | } | |
3646 | break; | |
3647 | case RAM_SAVE_FLAG_EOS: | |
3648 | /* normal exit */ | |
3649 | break; | |
3650 | default: | |
3651 | if (flags & RAM_SAVE_FLAG_HOOK) { | |
632e3a5c | 3652 | ram_control_load_hook(f, RAM_CONTROL_HOOK, NULL); |
56e93d26 JQ |
3653 | } else { |
3654 | error_report("Unknown combination of migration flags: %#x", | |
3655 | flags); | |
3656 | ret = -EINVAL; | |
3657 | } | |
3658 | } | |
3659 | if (!ret) { | |
3660 | ret = qemu_file_get_error(f); | |
3661 | } | |
3662 | } | |
3663 | ||
34ab9e97 | 3664 | ret |= wait_for_decompress_done(); |
56e93d26 | 3665 | rcu_read_unlock(); |
55c4446b | 3666 | trace_ram_load_complete(ret, seq_iter); |
56e93d26 JQ |
3667 | return ret; |
3668 | } | |
3669 | ||
c6467627 VSO |
3670 | static bool ram_has_postcopy(void *opaque) |
3671 | { | |
3672 | return migrate_postcopy_ram(); | |
3673 | } | |
3674 | ||
edd090c7 PX |
3675 | /* Sync all the dirty bitmap with destination VM. */ |
3676 | static int ram_dirty_bitmap_sync_all(MigrationState *s, RAMState *rs) | |
3677 | { | |
3678 | RAMBlock *block; | |
3679 | QEMUFile *file = s->to_dst_file; | |
3680 | int ramblock_count = 0; | |
3681 | ||
3682 | trace_ram_dirty_bitmap_sync_start(); | |
3683 | ||
ff0769a4 | 3684 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
edd090c7 PX |
3685 | qemu_savevm_send_recv_bitmap(file, block->idstr); |
3686 | trace_ram_dirty_bitmap_request(block->idstr); | |
3687 | ramblock_count++; | |
3688 | } | |
3689 | ||
3690 | trace_ram_dirty_bitmap_sync_wait(); | |
3691 | ||
3692 | /* Wait until all the ramblocks' dirty bitmap synced */ | |
3693 | while (ramblock_count--) { | |
3694 | qemu_sem_wait(&s->rp_state.rp_sem); | |
3695 | } | |
3696 | ||
3697 | trace_ram_dirty_bitmap_sync_complete(); | |
3698 | ||
3699 | return 0; | |
3700 | } | |
3701 | ||
3702 | static void ram_dirty_bitmap_reload_notify(MigrationState *s) | |
3703 | { | |
3704 | qemu_sem_post(&s->rp_state.rp_sem); | |
3705 | } | |
3706 | ||
a335debb PX |
3707 | /* |
3708 | * Read the received bitmap, revert it as the initial dirty bitmap. | |
3709 | * This is only used when the postcopy migration is paused but wants | |
3710 | * to resume from a middle point. | |
3711 | */ | |
3712 | int ram_dirty_bitmap_reload(MigrationState *s, RAMBlock *block) | |
3713 | { | |
3714 | int ret = -EINVAL; | |
3715 | QEMUFile *file = s->rp_state.from_dst_file; | |
3716 | unsigned long *le_bitmap, nbits = block->used_length >> TARGET_PAGE_BITS; | |
3717 | uint64_t local_size = nbits / 8; | |
3718 | uint64_t size, end_mark; | |
3719 | ||
3720 | trace_ram_dirty_bitmap_reload_begin(block->idstr); | |
3721 | ||
3722 | if (s->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
3723 | error_report("%s: incorrect state %s", __func__, | |
3724 | MigrationStatus_str(s->state)); | |
3725 | return -EINVAL; | |
3726 | } | |
3727 | ||
3728 | /* | |
3729 | * Note: see comments in ramblock_recv_bitmap_send() on why we | |
3730 | * need the endianess convertion, and the paddings. | |
3731 | */ | |
3732 | local_size = ROUND_UP(local_size, 8); | |
3733 | ||
3734 | /* Add paddings */ | |
3735 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
3736 | ||
3737 | size = qemu_get_be64(file); | |
3738 | ||
3739 | /* The size of the bitmap should match with our ramblock */ | |
3740 | if (size != local_size) { | |
3741 | error_report("%s: ramblock '%s' bitmap size mismatch " | |
3742 | "(0x%"PRIx64" != 0x%"PRIx64")", __func__, | |
3743 | block->idstr, size, local_size); | |
3744 | ret = -EINVAL; | |
3745 | goto out; | |
3746 | } | |
3747 | ||
3748 | size = qemu_get_buffer(file, (uint8_t *)le_bitmap, local_size); | |
3749 | end_mark = qemu_get_be64(file); | |
3750 | ||
3751 | ret = qemu_file_get_error(file); | |
3752 | if (ret || size != local_size) { | |
3753 | error_report("%s: read bitmap failed for ramblock '%s': %d" | |
3754 | " (size 0x%"PRIx64", got: 0x%"PRIx64")", | |
3755 | __func__, block->idstr, ret, local_size, size); | |
3756 | ret = -EIO; | |
3757 | goto out; | |
3758 | } | |
3759 | ||
3760 | if (end_mark != RAMBLOCK_RECV_BITMAP_ENDING) { | |
3761 | error_report("%s: ramblock '%s' end mark incorrect: 0x%"PRIu64, | |
3762 | __func__, block->idstr, end_mark); | |
3763 | ret = -EINVAL; | |
3764 | goto out; | |
3765 | } | |
3766 | ||
3767 | /* | |
3768 | * Endianess convertion. We are during postcopy (though paused). | |
3769 | * The dirty bitmap won't change. We can directly modify it. | |
3770 | */ | |
3771 | bitmap_from_le(block->bmap, le_bitmap, nbits); | |
3772 | ||
3773 | /* | |
3774 | * What we received is "received bitmap". Revert it as the initial | |
3775 | * dirty bitmap for this ramblock. | |
3776 | */ | |
3777 | bitmap_complement(block->bmap, block->bmap, nbits); | |
3778 | ||
3779 | trace_ram_dirty_bitmap_reload_complete(block->idstr); | |
3780 | ||
edd090c7 PX |
3781 | /* |
3782 | * We succeeded to sync bitmap for current ramblock. If this is | |
3783 | * the last one to sync, we need to notify the main send thread. | |
3784 | */ | |
3785 | ram_dirty_bitmap_reload_notify(s); | |
3786 | ||
a335debb PX |
3787 | ret = 0; |
3788 | out: | |
bf269906 | 3789 | g_free(le_bitmap); |
a335debb PX |
3790 | return ret; |
3791 | } | |
3792 | ||
edd090c7 PX |
3793 | static int ram_resume_prepare(MigrationState *s, void *opaque) |
3794 | { | |
3795 | RAMState *rs = *(RAMState **)opaque; | |
08614f34 | 3796 | int ret; |
edd090c7 | 3797 | |
08614f34 PX |
3798 | ret = ram_dirty_bitmap_sync_all(s, rs); |
3799 | if (ret) { | |
3800 | return ret; | |
3801 | } | |
3802 | ||
3803 | ram_state_resume_prepare(rs, s->to_dst_file); | |
3804 | ||
3805 | return 0; | |
edd090c7 PX |
3806 | } |
3807 | ||
56e93d26 | 3808 | static SaveVMHandlers savevm_ram_handlers = { |
9907e842 | 3809 | .save_setup = ram_save_setup, |
56e93d26 | 3810 | .save_live_iterate = ram_save_iterate, |
763c906b | 3811 | .save_live_complete_postcopy = ram_save_complete, |
a3e06c3d | 3812 | .save_live_complete_precopy = ram_save_complete, |
c6467627 | 3813 | .has_postcopy = ram_has_postcopy, |
56e93d26 JQ |
3814 | .save_live_pending = ram_save_pending, |
3815 | .load_state = ram_load, | |
f265e0e4 JQ |
3816 | .save_cleanup = ram_save_cleanup, |
3817 | .load_setup = ram_load_setup, | |
3818 | .load_cleanup = ram_load_cleanup, | |
edd090c7 | 3819 | .resume_prepare = ram_resume_prepare, |
56e93d26 JQ |
3820 | }; |
3821 | ||
3822 | void ram_mig_init(void) | |
3823 | { | |
3824 | qemu_mutex_init(&XBZRLE.lock); | |
6f37bb8b | 3825 | register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, &ram_state); |
56e93d26 | 3826 | } |