]>
Commit | Line | Data |
---|---|---|
56e93d26 JQ |
1 | /* |
2 | * QEMU System Emulator | |
3 | * | |
4 | * Copyright (c) 2003-2008 Fabrice Bellard | |
76cc7b58 JQ |
5 | * Copyright (c) 2011-2015 Red Hat Inc |
6 | * | |
7 | * Authors: | |
8 | * Juan Quintela <quintela@redhat.com> | |
56e93d26 JQ |
9 | * |
10 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
11 | * of this software and associated documentation files (the "Software"), to deal | |
12 | * in the Software without restriction, including without limitation the rights | |
13 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
14 | * copies of the Software, and to permit persons to whom the Software is | |
15 | * furnished to do so, subject to the following conditions: | |
16 | * | |
17 | * The above copyright notice and this permission notice shall be included in | |
18 | * all copies or substantial portions of the Software. | |
19 | * | |
20 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
21 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
22 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
23 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
24 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
25 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
26 | * THE SOFTWARE. | |
27 | */ | |
e688df6b | 28 | |
1393a485 | 29 | #include "qemu/osdep.h" |
f348b6d1 | 30 | #include "qemu/cutils.h" |
56e93d26 JQ |
31 | #include "qemu/bitops.h" |
32 | #include "qemu/bitmap.h" | |
b85ea5fa | 33 | #include "qemu/madvise.h" |
7205c9ec | 34 | #include "qemu/main-loop.h" |
709e3fe8 | 35 | #include "xbzrle.h" |
b5ca3368 | 36 | #include "ram-compress.h" |
7b1e1a22 | 37 | #include "ram.h" |
6666c96a | 38 | #include "migration.h" |
947701cc | 39 | #include "migration-stats.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" |
ab7cbb0b | 47 | #include "qapi/qapi-types-migration.h" |
9af23989 | 48 | #include "qapi/qapi-events-migration.h" |
acac51ba | 49 | #include "qapi/qapi-commands-migration.h" |
8acabf69 | 50 | #include "qapi/qmp/qerror.h" |
56e93d26 | 51 | #include "trace.h" |
56e93d26 | 52 | #include "exec/ram_addr.h" |
f9494614 | 53 | #include "exec/target_page.h" |
56e93d26 | 54 | #include "qemu/rcu_queue.h" |
a91246c9 | 55 | #include "migration/colo.h" |
53d37d36 | 56 | #include "block.h" |
b0c3cf94 | 57 | #include "sysemu/cpu-throttle.h" |
edd090c7 | 58 | #include "savevm.h" |
b9ee2f7d | 59 | #include "qemu/iov.h" |
d32ca5ad | 60 | #include "multifd.h" |
278e2f55 | 61 | #include "sysemu/runstate.h" |
1f0776f1 | 62 | #include "options.h" |
acac51ba HH |
63 | #include "sysemu/dirtylimit.h" |
64 | #include "sysemu/kvm.h" | |
278e2f55 | 65 | |
e5fdf920 LS |
66 | #include "hw/boards.h" /* for machine_dump_guest_core() */ |
67 | ||
278e2f55 AG |
68 | #if defined(__linux__) |
69 | #include "qemu/userfaultfd.h" | |
70 | #endif /* defined(__linux__) */ | |
56e93d26 | 71 | |
56e93d26 JQ |
72 | /***********************************************************/ |
73 | /* ram save/restore */ | |
74 | ||
7b548761 JQ |
75 | /* |
76 | * RAM_SAVE_FLAG_ZERO used to be named RAM_SAVE_FLAG_COMPRESS, it | |
77 | * worked for pages that were filled with the same char. We switched | |
bb890ed5 | 78 | * it to only search for the zero value. And to avoid confusion with |
7b548761 | 79 | * RAM_SAVE_FLAG_COMPRESS_PAGE just rename it. |
bb890ed5 | 80 | */ |
7b548761 JQ |
81 | /* |
82 | * RAM_SAVE_FLAG_FULL was obsoleted in 2009, it can be reused now | |
83 | */ | |
84 | #define RAM_SAVE_FLAG_FULL 0x01 | |
bb890ed5 | 85 | #define RAM_SAVE_FLAG_ZERO 0x02 |
56e93d26 JQ |
86 | #define RAM_SAVE_FLAG_MEM_SIZE 0x04 |
87 | #define RAM_SAVE_FLAG_PAGE 0x08 | |
88 | #define RAM_SAVE_FLAG_EOS 0x10 | |
89 | #define RAM_SAVE_FLAG_CONTINUE 0x20 | |
90 | #define RAM_SAVE_FLAG_XBZRLE 0x40 | |
7b548761 | 91 | /* 0x80 is reserved in qemu-file.h for RAM_SAVE_FLAG_HOOK */ |
56e93d26 | 92 | #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 |
294e5a40 | 93 | #define RAM_SAVE_FLAG_MULTIFD_FLUSH 0x200 |
7b548761 | 94 | /* We can't use any flag that is bigger than 0x200 */ |
56e93d26 | 95 | |
9360447d JQ |
96 | XBZRLECacheStats xbzrle_counters; |
97 | ||
f1668764 PX |
98 | /* used by the search for pages to send */ |
99 | struct PageSearchStatus { | |
100 | /* The migration channel used for a specific host page */ | |
101 | QEMUFile *pss_channel; | |
ec6f3ab9 PX |
102 | /* Last block from where we have sent data */ |
103 | RAMBlock *last_sent_block; | |
f1668764 PX |
104 | /* Current block being searched */ |
105 | RAMBlock *block; | |
106 | /* Current page to search from */ | |
107 | unsigned long page; | |
108 | /* Set once we wrap around */ | |
109 | bool complete_round; | |
f1668764 PX |
110 | /* Whether we're sending a host page */ |
111 | bool host_page_sending; | |
112 | /* The start/end of current host page. Invalid if host_page_sending==false */ | |
113 | unsigned long host_page_start; | |
114 | unsigned long host_page_end; | |
115 | }; | |
116 | typedef struct PageSearchStatus PageSearchStatus; | |
117 | ||
56e93d26 JQ |
118 | /* struct contains XBZRLE cache and a static page |
119 | used by the compression */ | |
120 | static struct { | |
121 | /* buffer used for XBZRLE encoding */ | |
122 | uint8_t *encoded_buf; | |
123 | /* buffer for storing page content */ | |
124 | uint8_t *current_buf; | |
125 | /* Cache for XBZRLE, Protected by lock. */ | |
126 | PageCache *cache; | |
127 | QemuMutex lock; | |
c00e0928 JQ |
128 | /* it will store a page full of zeros */ |
129 | uint8_t *zero_target_page; | |
f265e0e4 JQ |
130 | /* buffer used for XBZRLE decoding */ |
131 | uint8_t *decoded_buf; | |
56e93d26 JQ |
132 | } XBZRLE; |
133 | ||
56e93d26 JQ |
134 | static void XBZRLE_cache_lock(void) |
135 | { | |
87dca0c9 | 136 | if (migrate_xbzrle()) { |
56e93d26 | 137 | qemu_mutex_lock(&XBZRLE.lock); |
f4c51a6b | 138 | } |
56e93d26 JQ |
139 | } |
140 | ||
141 | static void XBZRLE_cache_unlock(void) | |
142 | { | |
87dca0c9 | 143 | if (migrate_xbzrle()) { |
56e93d26 | 144 | qemu_mutex_unlock(&XBZRLE.lock); |
f4c51a6b | 145 | } |
56e93d26 JQ |
146 | } |
147 | ||
3d0684b2 JQ |
148 | /** |
149 | * xbzrle_cache_resize: resize the xbzrle cache | |
150 | * | |
cbde7be9 | 151 | * This function is called from migrate_params_apply in main |
3d0684b2 JQ |
152 | * thread, possibly while a migration is in progress. A running |
153 | * migration may be using the cache and might finish during this call, | |
154 | * hence changes to the cache are protected by XBZRLE.lock(). | |
155 | * | |
c9dede2d | 156 | * Returns 0 for success or -1 for error |
3d0684b2 JQ |
157 | * |
158 | * @new_size: new cache size | |
8acabf69 | 159 | * @errp: set *errp if the check failed, with reason |
56e93d26 | 160 | */ |
8b9407a0 | 161 | int xbzrle_cache_resize(uint64_t new_size, Error **errp) |
56e93d26 JQ |
162 | { |
163 | PageCache *new_cache; | |
c9dede2d | 164 | int64_t ret = 0; |
56e93d26 | 165 | |
8acabf69 JQ |
166 | /* Check for truncation */ |
167 | if (new_size != (size_t)new_size) { | |
168 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", | |
169 | "exceeding address space"); | |
170 | return -1; | |
171 | } | |
172 | ||
2a313e5c JQ |
173 | if (new_size == migrate_xbzrle_cache_size()) { |
174 | /* nothing to do */ | |
c9dede2d | 175 | return 0; |
2a313e5c JQ |
176 | } |
177 | ||
56e93d26 JQ |
178 | XBZRLE_cache_lock(); |
179 | ||
180 | if (XBZRLE.cache != NULL) { | |
80f8dfde | 181 | new_cache = cache_init(new_size, TARGET_PAGE_SIZE, errp); |
56e93d26 | 182 | if (!new_cache) { |
56e93d26 JQ |
183 | ret = -1; |
184 | goto out; | |
185 | } | |
186 | ||
187 | cache_fini(XBZRLE.cache); | |
188 | XBZRLE.cache = new_cache; | |
189 | } | |
56e93d26 JQ |
190 | out: |
191 | XBZRLE_cache_unlock(); | |
192 | return ret; | |
193 | } | |
194 | ||
20123ee1 PX |
195 | static bool postcopy_preempt_active(void) |
196 | { | |
197 | return migrate_postcopy_preempt() && migration_in_postcopy(); | |
198 | } | |
199 | ||
f161c88a | 200 | bool migrate_ram_is_ignored(RAMBlock *block) |
fbd162e6 YK |
201 | { |
202 | return !qemu_ram_is_migratable(block) || | |
b0182e53 SS |
203 | (migrate_ignore_shared() && qemu_ram_is_shared(block) |
204 | && qemu_ram_is_named_file(block)); | |
fbd162e6 YK |
205 | } |
206 | ||
343f632c DDAG |
207 | #undef RAMBLOCK_FOREACH |
208 | ||
fbd162e6 YK |
209 | int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque) |
210 | { | |
211 | RAMBlock *block; | |
212 | int ret = 0; | |
213 | ||
89ac5a1d DDAG |
214 | RCU_READ_LOCK_GUARD(); |
215 | ||
fbd162e6 YK |
216 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
217 | ret = func(block, opaque); | |
218 | if (ret) { | |
219 | break; | |
220 | } | |
221 | } | |
fbd162e6 YK |
222 | return ret; |
223 | } | |
224 | ||
f9494614 AP |
225 | static void ramblock_recv_map_init(void) |
226 | { | |
227 | RAMBlock *rb; | |
228 | ||
fbd162e6 | 229 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
f9494614 AP |
230 | assert(!rb->receivedmap); |
231 | rb->receivedmap = bitmap_new(rb->max_length >> qemu_target_page_bits()); | |
232 | } | |
233 | } | |
234 | ||
235 | int ramblock_recv_bitmap_test(RAMBlock *rb, void *host_addr) | |
236 | { | |
237 | return test_bit(ramblock_recv_bitmap_offset(host_addr, rb), | |
238 | rb->receivedmap); | |
239 | } | |
240 | ||
1cba9f6e DDAG |
241 | bool ramblock_recv_bitmap_test_byte_offset(RAMBlock *rb, uint64_t byte_offset) |
242 | { | |
243 | return test_bit(byte_offset >> TARGET_PAGE_BITS, rb->receivedmap); | |
244 | } | |
245 | ||
f9494614 AP |
246 | void ramblock_recv_bitmap_set(RAMBlock *rb, void *host_addr) |
247 | { | |
248 | set_bit_atomic(ramblock_recv_bitmap_offset(host_addr, rb), rb->receivedmap); | |
249 | } | |
250 | ||
251 | void ramblock_recv_bitmap_set_range(RAMBlock *rb, void *host_addr, | |
252 | size_t nr) | |
253 | { | |
254 | bitmap_set_atomic(rb->receivedmap, | |
255 | ramblock_recv_bitmap_offset(host_addr, rb), | |
256 | nr); | |
257 | } | |
258 | ||
a335debb PX |
259 | #define RAMBLOCK_RECV_BITMAP_ENDING (0x0123456789abcdefULL) |
260 | ||
261 | /* | |
262 | * Format: bitmap_size (8 bytes) + whole_bitmap (N bytes). | |
263 | * | |
264 | * Returns >0 if success with sent bytes, or <0 if error. | |
265 | */ | |
266 | int64_t ramblock_recv_bitmap_send(QEMUFile *file, | |
267 | const char *block_name) | |
268 | { | |
269 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
270 | unsigned long *le_bitmap, nbits; | |
271 | uint64_t size; | |
272 | ||
273 | if (!block) { | |
274 | error_report("%s: invalid block name: %s", __func__, block_name); | |
275 | return -1; | |
276 | } | |
277 | ||
898ba906 | 278 | nbits = block->postcopy_length >> TARGET_PAGE_BITS; |
a335debb PX |
279 | |
280 | /* | |
281 | * Make sure the tmp bitmap buffer is big enough, e.g., on 32bit | |
282 | * machines we may need 4 more bytes for padding (see below | |
283 | * comment). So extend it a bit before hand. | |
284 | */ | |
285 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
286 | ||
287 | /* | |
288 | * Always use little endian when sending the bitmap. This is | |
289 | * required that when source and destination VMs are not using the | |
3a4452d8 | 290 | * same endianness. (Note: big endian won't work.) |
a335debb PX |
291 | */ |
292 | bitmap_to_le(le_bitmap, block->receivedmap, nbits); | |
293 | ||
294 | /* Size of the bitmap, in bytes */ | |
a725ef9f | 295 | size = DIV_ROUND_UP(nbits, 8); |
a335debb PX |
296 | |
297 | /* | |
298 | * size is always aligned to 8 bytes for 64bit machines, but it | |
299 | * may not be true for 32bit machines. We need this padding to | |
300 | * make sure the migration can survive even between 32bit and | |
301 | * 64bit machines. | |
302 | */ | |
303 | size = ROUND_UP(size, 8); | |
304 | ||
305 | qemu_put_be64(file, size); | |
306 | qemu_put_buffer(file, (const uint8_t *)le_bitmap, size); | |
307 | /* | |
308 | * Mark as an end, in case the middle part is screwed up due to | |
3a4452d8 | 309 | * some "mysterious" reason. |
a335debb PX |
310 | */ |
311 | qemu_put_be64(file, RAMBLOCK_RECV_BITMAP_ENDING); | |
312 | qemu_fflush(file); | |
313 | ||
bf269906 | 314 | g_free(le_bitmap); |
a335debb PX |
315 | |
316 | if (qemu_file_get_error(file)) { | |
317 | return qemu_file_get_error(file); | |
318 | } | |
319 | ||
320 | return size + sizeof(size); | |
321 | } | |
322 | ||
ec481c6c JQ |
323 | /* |
324 | * An outstanding page request, on the source, having been received | |
325 | * and queued | |
326 | */ | |
327 | struct RAMSrcPageRequest { | |
328 | RAMBlock *rb; | |
329 | hwaddr offset; | |
330 | hwaddr len; | |
331 | ||
332 | QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; | |
333 | }; | |
334 | ||
6f37bb8b JQ |
335 | /* State of RAM for migration */ |
336 | struct RAMState { | |
f1668764 PX |
337 | /* |
338 | * PageSearchStatus structures for the channels when send pages. | |
339 | * Protected by the bitmap_mutex. | |
340 | */ | |
341 | PageSearchStatus pss[RAM_CHANNEL_MAX]; | |
278e2f55 AG |
342 | /* UFFD file descriptor, used in 'write-tracking' migration */ |
343 | int uffdio_fd; | |
8d80e195 JQ |
344 | /* total ram size in bytes */ |
345 | uint64_t ram_bytes_total; | |
6f37bb8b JQ |
346 | /* Last block that we have visited searching for dirty pages */ |
347 | RAMBlock *last_seen_block; | |
269ace29 JQ |
348 | /* Last dirty target page we have sent */ |
349 | ram_addr_t last_page; | |
6f37bb8b JQ |
350 | /* last ram version we have seen */ |
351 | uint32_t last_version; | |
8d820d6f JQ |
352 | /* How many times we have dirty too many pages */ |
353 | int dirty_rate_high_cnt; | |
f664da80 JQ |
354 | /* these variables are used for bitmap sync */ |
355 | /* last time we did a full bitmap_sync */ | |
356 | int64_t time_last_bitmap_sync; | |
eac74159 | 357 | /* bytes transferred at start_time */ |
c4bdf0cf | 358 | uint64_t bytes_xfer_prev; |
a66cd90c | 359 | /* number of dirty pages since start_time */ |
68908ed6 | 360 | uint64_t num_dirty_pages_period; |
b5833fde JQ |
361 | /* xbzrle misses since the beginning of the period */ |
362 | uint64_t xbzrle_cache_miss_prev; | |
e460a4b1 WW |
363 | /* Amount of xbzrle pages since the beginning of the period */ |
364 | uint64_t xbzrle_pages_prev; | |
365 | /* Amount of xbzrle encoded bytes since the beginning of the period */ | |
366 | uint64_t xbzrle_bytes_prev; | |
f3095cc8 JQ |
367 | /* Are we really using XBZRLE (e.g., after the first round). */ |
368 | bool xbzrle_started; | |
05931ec5 JQ |
369 | /* Are we on the last stage of migration */ |
370 | bool last_stage; | |
76e03000 XG |
371 | /* compression statistics since the beginning of the period */ |
372 | /* amount of count that no free thread to compress data */ | |
373 | uint64_t compress_thread_busy_prev; | |
374 | /* amount bytes after compression */ | |
375 | uint64_t compressed_size_prev; | |
376 | /* amount of compressed pages */ | |
377 | uint64_t compress_pages_prev; | |
378 | ||
be8b02ed XG |
379 | /* total handled target pages at the beginning of period */ |
380 | uint64_t target_page_count_prev; | |
381 | /* total handled target pages since start */ | |
382 | uint64_t target_page_count; | |
9360447d | 383 | /* number of dirty bits in the bitmap */ |
2dfaf12e | 384 | uint64_t migration_dirty_pages; |
f1668764 PX |
385 | /* |
386 | * Protects: | |
387 | * - dirty/clear bitmap | |
388 | * - migration_dirty_pages | |
389 | * - pss structures | |
390 | */ | |
108cfae0 | 391 | QemuMutex bitmap_mutex; |
68a098f3 JQ |
392 | /* The RAMBlock used in the last src_page_requests */ |
393 | RAMBlock *last_req_rb; | |
ec481c6c JQ |
394 | /* Queue of outstanding page requests from the destination */ |
395 | QemuMutex src_page_req_mutex; | |
b58deb34 | 396 | QSIMPLEQ_HEAD(, RAMSrcPageRequest) src_page_requests; |
6f37bb8b JQ |
397 | }; |
398 | typedef struct RAMState RAMState; | |
399 | ||
53518d94 | 400 | static RAMState *ram_state; |
6f37bb8b | 401 | |
bd227060 WW |
402 | static NotifierWithReturnList precopy_notifier_list; |
403 | ||
a1fe28df PX |
404 | /* Whether postcopy has queued requests? */ |
405 | static bool postcopy_has_request(RAMState *rs) | |
406 | { | |
407 | return !QSIMPLEQ_EMPTY_ATOMIC(&rs->src_page_requests); | |
408 | } | |
409 | ||
bd227060 WW |
410 | void precopy_infrastructure_init(void) |
411 | { | |
412 | notifier_with_return_list_init(&precopy_notifier_list); | |
413 | } | |
414 | ||
415 | void precopy_add_notifier(NotifierWithReturn *n) | |
416 | { | |
417 | notifier_with_return_list_add(&precopy_notifier_list, n); | |
418 | } | |
419 | ||
420 | void precopy_remove_notifier(NotifierWithReturn *n) | |
421 | { | |
422 | notifier_with_return_remove(n); | |
423 | } | |
424 | ||
425 | int precopy_notify(PrecopyNotifyReason reason, Error **errp) | |
426 | { | |
427 | PrecopyNotifyData pnd; | |
428 | pnd.reason = reason; | |
429 | pnd.errp = errp; | |
430 | ||
431 | return notifier_with_return_list_notify(&precopy_notifier_list, &pnd); | |
432 | } | |
433 | ||
9edabd4d | 434 | uint64_t ram_bytes_remaining(void) |
2f4fde93 | 435 | { |
bae416e5 DDAG |
436 | return ram_state ? (ram_state->migration_dirty_pages * TARGET_PAGE_SIZE) : |
437 | 0; | |
2f4fde93 JQ |
438 | } |
439 | ||
26a26069 | 440 | void ram_transferred_add(uint64_t bytes) |
4c2d0f6d | 441 | { |
ae680668 | 442 | if (runstate_is_running()) { |
aff3f660 | 443 | stat64_add(&mig_stats.precopy_bytes, bytes); |
ae680668 | 444 | } else if (migration_in_postcopy()) { |
aff3f660 | 445 | stat64_add(&mig_stats.postcopy_bytes, bytes); |
ae680668 | 446 | } else { |
aff3f660 | 447 | stat64_add(&mig_stats.downtime_bytes, bytes); |
ae680668 | 448 | } |
aff3f660 | 449 | stat64_add(&mig_stats.transferred, bytes); |
4c2d0f6d DE |
450 | } |
451 | ||
4010ba38 JQ |
452 | struct MigrationOps { |
453 | int (*ram_save_target_page)(RAMState *rs, PageSearchStatus *pss); | |
454 | }; | |
455 | typedef struct MigrationOps MigrationOps; | |
456 | ||
457 | MigrationOps *migration_ops; | |
458 | ||
93589827 PX |
459 | static int ram_save_host_page_urgent(PageSearchStatus *pss); |
460 | ||
ebd88a49 PX |
461 | /* NOTE: page is the PFN not real ram_addr_t. */ |
462 | static void pss_init(PageSearchStatus *pss, RAMBlock *rb, ram_addr_t page) | |
463 | { | |
464 | pss->block = rb; | |
465 | pss->page = page; | |
466 | pss->complete_round = false; | |
467 | } | |
468 | ||
93589827 PX |
469 | /* |
470 | * Check whether two PSSs are actively sending the same page. Return true | |
471 | * if it is, false otherwise. | |
472 | */ | |
473 | static bool pss_overlap(PageSearchStatus *pss1, PageSearchStatus *pss2) | |
474 | { | |
475 | return pss1->host_page_sending && pss2->host_page_sending && | |
476 | (pss1->host_page_start == pss2->host_page_start); | |
477 | } | |
478 | ||
56e93d26 | 479 | /** |
3d0684b2 | 480 | * save_page_header: write page header to wire |
56e93d26 JQ |
481 | * |
482 | * If this is the 1st block, it also writes the block identification | |
483 | * | |
3d0684b2 | 484 | * Returns the number of bytes written |
56e93d26 | 485 | * |
ec6f3ab9 | 486 | * @pss: current PSS channel status |
56e93d26 JQ |
487 | * @block: block that contains the page we want to send |
488 | * @offset: offset inside the block for the page | |
489 | * in the lower bits, it contains flags | |
490 | */ | |
37502df3 LS |
491 | static size_t save_page_header(PageSearchStatus *pss, QEMUFile *f, |
492 | RAMBlock *block, ram_addr_t offset) | |
56e93d26 | 493 | { |
9f5f380b | 494 | size_t size, len; |
ec6f3ab9 | 495 | bool same_block = (block == pss->last_sent_block); |
56e93d26 | 496 | |
10661f11 | 497 | if (same_block) { |
24795694 JQ |
498 | offset |= RAM_SAVE_FLAG_CONTINUE; |
499 | } | |
2bf3aa85 | 500 | qemu_put_be64(f, offset); |
56e93d26 JQ |
501 | size = 8; |
502 | ||
10661f11 | 503 | if (!same_block) { |
9f5f380b | 504 | len = strlen(block->idstr); |
2bf3aa85 JQ |
505 | qemu_put_byte(f, len); |
506 | qemu_put_buffer(f, (uint8_t *)block->idstr, len); | |
9f5f380b | 507 | size += 1 + len; |
ec6f3ab9 | 508 | pss->last_sent_block = block; |
56e93d26 JQ |
509 | } |
510 | return size; | |
511 | } | |
512 | ||
3d0684b2 | 513 | /** |
179a8080 | 514 | * mig_throttle_guest_down: throttle down the guest |
3d0684b2 JQ |
515 | * |
516 | * Reduce amount of guest cpu execution to hopefully slow down memory | |
517 | * writes. If guest dirty memory rate is reduced below the rate at | |
518 | * which we can transfer pages to the destination then we should be | |
519 | * able to complete migration. Some workloads dirty memory way too | |
520 | * fast and will not effectively converge, even with auto-converge. | |
070afca2 | 521 | */ |
cbbf8182 KZ |
522 | static void mig_throttle_guest_down(uint64_t bytes_dirty_period, |
523 | uint64_t bytes_dirty_threshold) | |
070afca2 | 524 | { |
2a8ec380 | 525 | uint64_t pct_initial = migrate_cpu_throttle_initial(); |
9605c2ac | 526 | uint64_t pct_increment = migrate_cpu_throttle_increment(); |
873f674c | 527 | bool pct_tailslow = migrate_cpu_throttle_tailslow(); |
24155bd0 | 528 | int pct_max = migrate_max_cpu_throttle(); |
070afca2 | 529 | |
cbbf8182 KZ |
530 | uint64_t throttle_now = cpu_throttle_get_percentage(); |
531 | uint64_t cpu_now, cpu_ideal, throttle_inc; | |
532 | ||
070afca2 JH |
533 | /* We have not started throttling yet. Let's start it. */ |
534 | if (!cpu_throttle_active()) { | |
535 | cpu_throttle_set(pct_initial); | |
536 | } else { | |
537 | /* Throttling already on, just increase the rate */ | |
cbbf8182 KZ |
538 | if (!pct_tailslow) { |
539 | throttle_inc = pct_increment; | |
540 | } else { | |
541 | /* Compute the ideal CPU percentage used by Guest, which may | |
542 | * make the dirty rate match the dirty rate threshold. */ | |
543 | cpu_now = 100 - throttle_now; | |
544 | cpu_ideal = cpu_now * (bytes_dirty_threshold * 1.0 / | |
545 | bytes_dirty_period); | |
546 | throttle_inc = MIN(cpu_now - cpu_ideal, pct_increment); | |
547 | } | |
548 | cpu_throttle_set(MIN(throttle_now + throttle_inc, pct_max)); | |
070afca2 JH |
549 | } |
550 | } | |
551 | ||
91fe9a8d RL |
552 | void mig_throttle_counter_reset(void) |
553 | { | |
554 | RAMState *rs = ram_state; | |
555 | ||
556 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
557 | rs->num_dirty_pages_period = 0; | |
aff3f660 | 558 | rs->bytes_xfer_prev = stat64_get(&mig_stats.transferred); |
91fe9a8d RL |
559 | } |
560 | ||
3d0684b2 JQ |
561 | /** |
562 | * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache | |
563 | * | |
6f37bb8b | 564 | * @rs: current RAM state |
3d0684b2 JQ |
565 | * @current_addr: address for the zero page |
566 | * | |
567 | * Update the xbzrle cache to reflect a page that's been sent as all 0. | |
56e93d26 JQ |
568 | * The important thing is that a stale (not-yet-0'd) page be replaced |
569 | * by the new data. | |
570 | * As a bonus, if the page wasn't in the cache it gets added so that | |
3d0684b2 | 571 | * when a small write is made into the 0'd page it gets XBZRLE sent. |
56e93d26 | 572 | */ |
6f37bb8b | 573 | static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr) |
56e93d26 | 574 | { |
56e93d26 JQ |
575 | /* We don't care if this fails to allocate a new cache page |
576 | * as long as it updated an old one */ | |
c00e0928 | 577 | cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page, |
aff3f660 | 578 | stat64_get(&mig_stats.dirty_sync_count)); |
56e93d26 JQ |
579 | } |
580 | ||
581 | #define ENCODING_FLAG_XBZRLE 0x1 | |
582 | ||
583 | /** | |
584 | * save_xbzrle_page: compress and send current page | |
585 | * | |
586 | * Returns: 1 means that we wrote the page | |
587 | * 0 means that page is identical to the one already sent | |
588 | * -1 means that xbzrle would be longer than normal | |
589 | * | |
5a987738 | 590 | * @rs: current RAM state |
ec6f3ab9 | 591 | * @pss: current PSS channel |
3d0684b2 JQ |
592 | * @current_data: pointer to the address of the page contents |
593 | * @current_addr: addr of the page | |
56e93d26 JQ |
594 | * @block: block that contains the page we want to send |
595 | * @offset: offset inside the block for the page | |
56e93d26 | 596 | */ |
ec6f3ab9 | 597 | static int save_xbzrle_page(RAMState *rs, PageSearchStatus *pss, |
61717ea9 PX |
598 | uint8_t **current_data, ram_addr_t current_addr, |
599 | RAMBlock *block, ram_addr_t offset) | |
56e93d26 JQ |
600 | { |
601 | int encoded_len = 0, bytes_xbzrle; | |
602 | uint8_t *prev_cached_page; | |
ec6f3ab9 | 603 | QEMUFile *file = pss->pss_channel; |
aff3f660 | 604 | uint64_t generation = stat64_get(&mig_stats.dirty_sync_count); |
56e93d26 | 605 | |
536b5a4e | 606 | if (!cache_is_cached(XBZRLE.cache, current_addr, generation)) { |
9360447d | 607 | xbzrle_counters.cache_miss++; |
05931ec5 | 608 | if (!rs->last_stage) { |
56e93d26 | 609 | if (cache_insert(XBZRLE.cache, current_addr, *current_data, |
536b5a4e | 610 | generation) == -1) { |
56e93d26 JQ |
611 | return -1; |
612 | } else { | |
613 | /* update *current_data when the page has been | |
614 | inserted into cache */ | |
615 | *current_data = get_cached_data(XBZRLE.cache, current_addr); | |
616 | } | |
617 | } | |
618 | return -1; | |
619 | } | |
620 | ||
e460a4b1 WW |
621 | /* |
622 | * Reaching here means the page has hit the xbzrle cache, no matter what | |
623 | * encoding result it is (normal encoding, overflow or skipping the page), | |
3a4452d8 | 624 | * count the page as encoded. This is used to calculate the encoding rate. |
e460a4b1 WW |
625 | * |
626 | * Example: 2 pages (8KB) being encoded, first page encoding generates 2KB, | |
627 | * 2nd page turns out to be skipped (i.e. no new bytes written to the | |
628 | * page), the overall encoding rate will be 8KB / 2KB = 4, which has the | |
629 | * skipped page included. In this way, the encoding rate can tell if the | |
630 | * guest page is good for xbzrle encoding. | |
631 | */ | |
632 | xbzrle_counters.pages++; | |
56e93d26 JQ |
633 | prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); |
634 | ||
635 | /* save current buffer into memory */ | |
636 | memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); | |
637 | ||
638 | /* XBZRLE encoding (if there is no overflow) */ | |
7ba7db9f RH |
639 | encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, |
640 | TARGET_PAGE_SIZE, XBZRLE.encoded_buf, | |
641 | TARGET_PAGE_SIZE); | |
ca353803 WY |
642 | |
643 | /* | |
644 | * Update the cache contents, so that it corresponds to the data | |
645 | * sent, in all cases except where we skip the page. | |
646 | */ | |
05931ec5 | 647 | if (!rs->last_stage && encoded_len != 0) { |
ca353803 WY |
648 | memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); |
649 | /* | |
650 | * In the case where we couldn't compress, ensure that the caller | |
651 | * sends the data from the cache, since the guest might have | |
652 | * changed the RAM since we copied it. | |
653 | */ | |
654 | *current_data = prev_cached_page; | |
655 | } | |
656 | ||
56e93d26 | 657 | if (encoded_len == 0) { |
55c4446b | 658 | trace_save_xbzrle_page_skipping(); |
56e93d26 JQ |
659 | return 0; |
660 | } else if (encoded_len == -1) { | |
55c4446b | 661 | trace_save_xbzrle_page_overflow(); |
9360447d | 662 | xbzrle_counters.overflow++; |
e460a4b1 | 663 | xbzrle_counters.bytes += TARGET_PAGE_SIZE; |
56e93d26 JQ |
664 | return -1; |
665 | } | |
666 | ||
56e93d26 | 667 | /* Send XBZRLE based compressed page */ |
37502df3 | 668 | bytes_xbzrle = save_page_header(pss, pss->pss_channel, block, |
204b88b8 | 669 | offset | RAM_SAVE_FLAG_XBZRLE); |
61717ea9 PX |
670 | qemu_put_byte(file, ENCODING_FLAG_XBZRLE); |
671 | qemu_put_be16(file, encoded_len); | |
672 | qemu_put_buffer(file, XBZRLE.encoded_buf, encoded_len); | |
56e93d26 | 673 | bytes_xbzrle += encoded_len + 1 + 2; |
e460a4b1 WW |
674 | /* |
675 | * Like compressed_size (please see update_compress_thread_counts), | |
676 | * the xbzrle encoded bytes don't count the 8 byte header with | |
677 | * RAM_SAVE_FLAG_CONTINUE. | |
678 | */ | |
679 | xbzrle_counters.bytes += bytes_xbzrle - 8; | |
4c2d0f6d | 680 | ram_transferred_add(bytes_xbzrle); |
56e93d26 JQ |
681 | |
682 | return 1; | |
683 | } | |
684 | ||
3d0684b2 | 685 | /** |
d9e474ea | 686 | * pss_find_next_dirty: find the next dirty page of current ramblock |
f3f491fc | 687 | * |
d9e474ea PX |
688 | * This function updates pss->page to point to the next dirty page index |
689 | * within the ramblock to migrate, or the end of ramblock when nothing | |
690 | * found. Note that when pss->host_page_sending==true it means we're | |
691 | * during sending a host page, so we won't look for dirty page that is | |
692 | * outside the host page boundary. | |
3d0684b2 | 693 | * |
d9e474ea | 694 | * @pss: the current page search status |
f3f491fc | 695 | */ |
d9e474ea | 696 | static void pss_find_next_dirty(PageSearchStatus *pss) |
56e93d26 | 697 | { |
d9e474ea | 698 | RAMBlock *rb = pss->block; |
6b6712ef JQ |
699 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; |
700 | unsigned long *bitmap = rb->bmap; | |
56e93d26 | 701 | |
f161c88a | 702 | if (migrate_ram_is_ignored(rb)) { |
d9e474ea PX |
703 | /* Points directly to the end, so we know no dirty page */ |
704 | pss->page = size; | |
705 | return; | |
706 | } | |
707 | ||
708 | /* | |
709 | * If during sending a host page, only look for dirty pages within the | |
710 | * current host page being send. | |
711 | */ | |
712 | if (pss->host_page_sending) { | |
713 | assert(pss->host_page_end); | |
714 | size = MIN(size, pss->host_page_end); | |
b895de50 CLG |
715 | } |
716 | ||
d9e474ea | 717 | pss->page = find_next_bit(bitmap, size, pss->page); |
56e93d26 JQ |
718 | } |
719 | ||
1230a25f | 720 | static void migration_clear_memory_region_dirty_bitmap(RAMBlock *rb, |
3143577d WW |
721 | unsigned long page) |
722 | { | |
723 | uint8_t shift; | |
724 | hwaddr size, start; | |
725 | ||
726 | if (!rb->clear_bmap || !clear_bmap_test_and_clear(rb, page)) { | |
727 | return; | |
728 | } | |
729 | ||
730 | shift = rb->clear_bmap_shift; | |
731 | /* | |
732 | * CLEAR_BITMAP_SHIFT_MIN should always guarantee this... this | |
733 | * can make things easier sometimes since then start address | |
734 | * of the small chunk will always be 64 pages aligned so the | |
735 | * bitmap will always be aligned to unsigned long. We should | |
736 | * even be able to remove this restriction but I'm simply | |
737 | * keeping it. | |
738 | */ | |
739 | assert(shift >= 6); | |
740 | ||
741 | size = 1ULL << (TARGET_PAGE_BITS + shift); | |
7648297d | 742 | start = QEMU_ALIGN_DOWN((ram_addr_t)page << TARGET_PAGE_BITS, size); |
3143577d WW |
743 | trace_migration_bitmap_clear_dirty(rb->idstr, start, size, page); |
744 | memory_region_clear_dirty_bitmap(rb->mr, start, size); | |
745 | } | |
746 | ||
747 | static void | |
1230a25f | 748 | migration_clear_memory_region_dirty_bitmap_range(RAMBlock *rb, |
3143577d WW |
749 | unsigned long start, |
750 | unsigned long npages) | |
751 | { | |
752 | unsigned long i, chunk_pages = 1UL << rb->clear_bmap_shift; | |
753 | unsigned long chunk_start = QEMU_ALIGN_DOWN(start, chunk_pages); | |
754 | unsigned long chunk_end = QEMU_ALIGN_UP(start + npages, chunk_pages); | |
755 | ||
756 | /* | |
757 | * Clear pages from start to start + npages - 1, so the end boundary is | |
758 | * exclusive. | |
759 | */ | |
760 | for (i = chunk_start; i < chunk_end; i += chunk_pages) { | |
1230a25f | 761 | migration_clear_memory_region_dirty_bitmap(rb, i); |
3143577d WW |
762 | } |
763 | } | |
764 | ||
a6a83cef RL |
765 | /* |
766 | * colo_bitmap_find_diry:find contiguous dirty pages from start | |
767 | * | |
768 | * Returns the page offset within memory region of the start of the contiguout | |
769 | * dirty page | |
770 | * | |
771 | * @rs: current RAM state | |
772 | * @rb: RAMBlock where to search for dirty pages | |
773 | * @start: page where we start the search | |
774 | * @num: the number of contiguous dirty pages | |
775 | */ | |
776 | static inline | |
777 | unsigned long colo_bitmap_find_dirty(RAMState *rs, RAMBlock *rb, | |
778 | unsigned long start, unsigned long *num) | |
779 | { | |
780 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; | |
781 | unsigned long *bitmap = rb->bmap; | |
782 | unsigned long first, next; | |
783 | ||
784 | *num = 0; | |
785 | ||
f161c88a | 786 | if (migrate_ram_is_ignored(rb)) { |
a6a83cef RL |
787 | return size; |
788 | } | |
789 | ||
790 | first = find_next_bit(bitmap, size, start); | |
791 | if (first >= size) { | |
792 | return first; | |
793 | } | |
794 | next = find_next_zero_bit(bitmap, size, first + 1); | |
795 | assert(next >= first); | |
796 | *num = next - first; | |
797 | return first; | |
798 | } | |
799 | ||
06b10688 | 800 | static inline bool migration_bitmap_clear_dirty(RAMState *rs, |
f20e2865 JQ |
801 | RAMBlock *rb, |
802 | unsigned long page) | |
a82d593b DDAG |
803 | { |
804 | bool ret; | |
a82d593b | 805 | |
002cad6b PX |
806 | /* |
807 | * Clear dirty bitmap if needed. This _must_ be called before we | |
808 | * send any of the page in the chunk because we need to make sure | |
809 | * we can capture further page content changes when we sync dirty | |
810 | * log the next time. So as long as we are going to send any of | |
811 | * the page in the chunk we clear the remote dirty bitmap for all. | |
812 | * Clearing it earlier won't be a problem, but too late will. | |
813 | */ | |
1230a25f | 814 | migration_clear_memory_region_dirty_bitmap(rb, page); |
002cad6b | 815 | |
6b6712ef | 816 | ret = test_and_clear_bit(page, rb->bmap); |
a82d593b | 817 | if (ret) { |
0d8ec885 | 818 | rs->migration_dirty_pages--; |
a82d593b | 819 | } |
386a907b | 820 | |
a82d593b DDAG |
821 | return ret; |
822 | } | |
823 | ||
be39b4cd DH |
824 | static void dirty_bitmap_clear_section(MemoryRegionSection *section, |
825 | void *opaque) | |
826 | { | |
827 | const hwaddr offset = section->offset_within_region; | |
828 | const hwaddr size = int128_get64(section->size); | |
829 | const unsigned long start = offset >> TARGET_PAGE_BITS; | |
830 | const unsigned long npages = size >> TARGET_PAGE_BITS; | |
831 | RAMBlock *rb = section->mr->ram_block; | |
832 | uint64_t *cleared_bits = opaque; | |
833 | ||
834 | /* | |
835 | * We don't grab ram_state->bitmap_mutex because we expect to run | |
836 | * only when starting migration or during postcopy recovery where | |
837 | * we don't have concurrent access. | |
838 | */ | |
839 | if (!migration_in_postcopy() && !migrate_background_snapshot()) { | |
840 | migration_clear_memory_region_dirty_bitmap_range(rb, start, npages); | |
841 | } | |
842 | *cleared_bits += bitmap_count_one_with_offset(rb->bmap, start, npages); | |
843 | bitmap_clear(rb->bmap, start, npages); | |
844 | } | |
845 | ||
846 | /* | |
847 | * Exclude all dirty pages from migration that fall into a discarded range as | |
848 | * managed by a RamDiscardManager responsible for the mapped memory region of | |
849 | * the RAMBlock. Clear the corresponding bits in the dirty bitmaps. | |
850 | * | |
851 | * Discarded pages ("logically unplugged") have undefined content and must | |
852 | * not get migrated, because even reading these pages for migration might | |
853 | * result in undesired behavior. | |
854 | * | |
855 | * Returns the number of cleared bits in the RAMBlock dirty bitmap. | |
856 | * | |
857 | * Note: The result is only stable while migrating (precopy/postcopy). | |
858 | */ | |
859 | static uint64_t ramblock_dirty_bitmap_clear_discarded_pages(RAMBlock *rb) | |
860 | { | |
861 | uint64_t cleared_bits = 0; | |
862 | ||
863 | if (rb->mr && rb->bmap && memory_region_has_ram_discard_manager(rb->mr)) { | |
864 | RamDiscardManager *rdm = memory_region_get_ram_discard_manager(rb->mr); | |
865 | MemoryRegionSection section = { | |
866 | .mr = rb->mr, | |
867 | .offset_within_region = 0, | |
868 | .size = int128_make64(qemu_ram_get_used_length(rb)), | |
869 | }; | |
870 | ||
871 | ram_discard_manager_replay_discarded(rdm, §ion, | |
872 | dirty_bitmap_clear_section, | |
873 | &cleared_bits); | |
874 | } | |
875 | return cleared_bits; | |
876 | } | |
877 | ||
9470c5e0 DH |
878 | /* |
879 | * Check if a host-page aligned page falls into a discarded range as managed by | |
880 | * a RamDiscardManager responsible for the mapped memory region of the RAMBlock. | |
881 | * | |
882 | * Note: The result is only stable while migrating (precopy/postcopy). | |
883 | */ | |
884 | bool ramblock_page_is_discarded(RAMBlock *rb, ram_addr_t start) | |
885 | { | |
886 | if (rb->mr && memory_region_has_ram_discard_manager(rb->mr)) { | |
887 | RamDiscardManager *rdm = memory_region_get_ram_discard_manager(rb->mr); | |
888 | MemoryRegionSection section = { | |
889 | .mr = rb->mr, | |
890 | .offset_within_region = start, | |
891 | .size = int128_make64(qemu_ram_pagesize(rb)), | |
892 | }; | |
893 | ||
894 | return !ram_discard_manager_is_populated(rdm, §ion); | |
895 | } | |
896 | return false; | |
897 | } | |
898 | ||
267691b6 | 899 | /* Called with RCU critical section */ |
7a3e9571 | 900 | static void ramblock_sync_dirty_bitmap(RAMState *rs, RAMBlock *rb) |
56e93d26 | 901 | { |
fb613580 KZ |
902 | uint64_t new_dirty_pages = |
903 | cpu_physical_memory_sync_dirty_bitmap(rb, 0, rb->used_length); | |
904 | ||
905 | rs->migration_dirty_pages += new_dirty_pages; | |
906 | rs->num_dirty_pages_period += new_dirty_pages; | |
56e93d26 JQ |
907 | } |
908 | ||
3d0684b2 JQ |
909 | /** |
910 | * ram_pagesize_summary: calculate all the pagesizes of a VM | |
911 | * | |
912 | * Returns a summary bitmap of the page sizes of all RAMBlocks | |
913 | * | |
914 | * For VMs with just normal pages this is equivalent to the host page | |
915 | * size. If it's got some huge pages then it's the OR of all the | |
916 | * different page sizes. | |
e8ca1db2 DDAG |
917 | */ |
918 | uint64_t ram_pagesize_summary(void) | |
919 | { | |
920 | RAMBlock *block; | |
921 | uint64_t summary = 0; | |
922 | ||
fbd162e6 | 923 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
e8ca1db2 DDAG |
924 | summary |= block->page_size; |
925 | } | |
926 | ||
927 | return summary; | |
928 | } | |
929 | ||
aecbfe9c XG |
930 | uint64_t ram_get_total_transferred_pages(void) |
931 | { | |
aff3f660 JQ |
932 | return stat64_get(&mig_stats.normal_pages) + |
933 | stat64_get(&mig_stats.zero_pages) + | |
23b7576d | 934 | compression_counters.pages + xbzrle_counters.pages; |
aecbfe9c XG |
935 | } |
936 | ||
b734035b XG |
937 | static void migration_update_rates(RAMState *rs, int64_t end_time) |
938 | { | |
be8b02ed | 939 | uint64_t page_count = rs->target_page_count - rs->target_page_count_prev; |
76e03000 | 940 | double compressed_size; |
b734035b XG |
941 | |
942 | /* calculate period counters */ | |
aff3f660 | 943 | stat64_set(&mig_stats.dirty_pages_rate, |
72f8e587 JQ |
944 | rs->num_dirty_pages_period * 1000 / |
945 | (end_time - rs->time_last_bitmap_sync)); | |
b734035b | 946 | |
be8b02ed | 947 | if (!page_count) { |
b734035b XG |
948 | return; |
949 | } | |
950 | ||
87dca0c9 | 951 | if (migrate_xbzrle()) { |
e460a4b1 WW |
952 | double encoded_size, unencoded_size; |
953 | ||
b734035b | 954 | xbzrle_counters.cache_miss_rate = (double)(xbzrle_counters.cache_miss - |
be8b02ed | 955 | rs->xbzrle_cache_miss_prev) / page_count; |
b734035b | 956 | rs->xbzrle_cache_miss_prev = xbzrle_counters.cache_miss; |
e460a4b1 WW |
957 | unencoded_size = (xbzrle_counters.pages - rs->xbzrle_pages_prev) * |
958 | TARGET_PAGE_SIZE; | |
959 | encoded_size = xbzrle_counters.bytes - rs->xbzrle_bytes_prev; | |
92271402 | 960 | if (xbzrle_counters.pages == rs->xbzrle_pages_prev || !encoded_size) { |
e460a4b1 | 961 | xbzrle_counters.encoding_rate = 0; |
e460a4b1 WW |
962 | } else { |
963 | xbzrle_counters.encoding_rate = unencoded_size / encoded_size; | |
964 | } | |
965 | rs->xbzrle_pages_prev = xbzrle_counters.pages; | |
966 | rs->xbzrle_bytes_prev = xbzrle_counters.bytes; | |
b734035b | 967 | } |
76e03000 | 968 | |
a7a94d14 | 969 | if (migrate_compress()) { |
76e03000 XG |
970 | compression_counters.busy_rate = (double)(compression_counters.busy - |
971 | rs->compress_thread_busy_prev) / page_count; | |
972 | rs->compress_thread_busy_prev = compression_counters.busy; | |
973 | ||
974 | compressed_size = compression_counters.compressed_size - | |
975 | rs->compressed_size_prev; | |
976 | if (compressed_size) { | |
977 | double uncompressed_size = (compression_counters.pages - | |
978 | rs->compress_pages_prev) * TARGET_PAGE_SIZE; | |
979 | ||
980 | /* Compression-Ratio = Uncompressed-size / Compressed-size */ | |
981 | compression_counters.compression_rate = | |
982 | uncompressed_size / compressed_size; | |
983 | ||
984 | rs->compress_pages_prev = compression_counters.pages; | |
985 | rs->compressed_size_prev = compression_counters.compressed_size; | |
986 | } | |
987 | } | |
b734035b XG |
988 | } |
989 | ||
acac51ba HH |
990 | /* |
991 | * Enable dirty-limit to throttle down the guest | |
992 | */ | |
993 | static void migration_dirty_limit_guest(void) | |
994 | { | |
995 | /* | |
996 | * dirty page rate quota for all vCPUs fetched from | |
997 | * migration parameter 'vcpu_dirty_limit' | |
998 | */ | |
999 | static int64_t quota_dirtyrate; | |
1000 | MigrationState *s = migrate_get_current(); | |
1001 | ||
1002 | /* | |
1003 | * If dirty limit already enabled and migration parameter | |
1004 | * vcpu-dirty-limit untouched. | |
1005 | */ | |
1006 | if (dirtylimit_in_service() && | |
1007 | quota_dirtyrate == s->parameters.vcpu_dirty_limit) { | |
1008 | return; | |
1009 | } | |
1010 | ||
1011 | quota_dirtyrate = s->parameters.vcpu_dirty_limit; | |
1012 | ||
1013 | /* | |
1014 | * Set all vCPU a quota dirtyrate, note that the second | |
1015 | * parameter will be ignored if setting all vCPU for the vm | |
1016 | */ | |
1017 | qmp_set_vcpu_dirty_limit(false, -1, quota_dirtyrate, NULL); | |
1018 | trace_migration_dirty_limit_guest(quota_dirtyrate); | |
1019 | } | |
1020 | ||
dc14a470 KZ |
1021 | static void migration_trigger_throttle(RAMState *rs) |
1022 | { | |
6499efdb | 1023 | uint64_t threshold = migrate_throttle_trigger_threshold(); |
23b7576d | 1024 | uint64_t bytes_xfer_period = |
aff3f660 | 1025 | stat64_get(&mig_stats.transferred) - rs->bytes_xfer_prev; |
dc14a470 KZ |
1026 | uint64_t bytes_dirty_period = rs->num_dirty_pages_period * TARGET_PAGE_SIZE; |
1027 | uint64_t bytes_dirty_threshold = bytes_xfer_period * threshold / 100; | |
1028 | ||
1029 | /* During block migration the auto-converge logic incorrectly detects | |
1030 | * that ram migration makes no progress. Avoid this by disabling the | |
1031 | * throttling logic during the bulk phase of block migration. */ | |
bb9993c6 HH |
1032 | if (blk_mig_bulk_active()) { |
1033 | return; | |
1034 | } | |
1035 | ||
310ad562 HH |
1036 | /* |
1037 | * The following detection logic can be refined later. For now: | |
1038 | * Check to see if the ratio between dirtied bytes and the approx. | |
1039 | * amount of bytes that just got transferred since the last time | |
1040 | * we were in this routine reaches the threshold. If that happens | |
1041 | * twice, start or increase throttling. | |
1042 | */ | |
1043 | if ((bytes_dirty_period > bytes_dirty_threshold) && | |
1044 | (++rs->dirty_rate_high_cnt >= 2)) { | |
1045 | rs->dirty_rate_high_cnt = 0; | |
1046 | if (migrate_auto_converge()) { | |
dc14a470 | 1047 | trace_migration_throttle(); |
cbbf8182 KZ |
1048 | mig_throttle_guest_down(bytes_dirty_period, |
1049 | bytes_dirty_threshold); | |
acac51ba HH |
1050 | } else if (migrate_dirty_limit()) { |
1051 | migration_dirty_limit_guest(); | |
dc14a470 KZ |
1052 | } |
1053 | } | |
1054 | } | |
1055 | ||
1e493be5 | 1056 | static void migration_bitmap_sync(RAMState *rs, bool last_stage) |
56e93d26 JQ |
1057 | { |
1058 | RAMBlock *block; | |
56e93d26 | 1059 | int64_t end_time; |
56e93d26 | 1060 | |
aff3f660 | 1061 | stat64_add(&mig_stats.dirty_sync_count, 1); |
56e93d26 | 1062 | |
f664da80 JQ |
1063 | if (!rs->time_last_bitmap_sync) { |
1064 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56e93d26 JQ |
1065 | } |
1066 | ||
1067 | trace_migration_bitmap_sync_start(); | |
1e493be5 | 1068 | memory_global_dirty_log_sync(last_stage); |
56e93d26 | 1069 | |
108cfae0 | 1070 | qemu_mutex_lock(&rs->bitmap_mutex); |
89ac5a1d DDAG |
1071 | WITH_RCU_READ_LOCK_GUARD() { |
1072 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
1073 | ramblock_sync_dirty_bitmap(rs, block); | |
1074 | } | |
aff3f660 | 1075 | stat64_set(&mig_stats.dirty_bytes_last_sync, ram_bytes_remaining()); |
56e93d26 | 1076 | } |
108cfae0 | 1077 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 1078 | |
9458a9a1 | 1079 | memory_global_after_dirty_log_sync(); |
a66cd90c | 1080 | trace_migration_bitmap_sync_end(rs->num_dirty_pages_period); |
1ffb5dfd | 1081 | |
56e93d26 JQ |
1082 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1083 | ||
1084 | /* more than 1 second = 1000 millisecons */ | |
f664da80 | 1085 | if (end_time > rs->time_last_bitmap_sync + 1000) { |
dc14a470 | 1086 | migration_trigger_throttle(rs); |
070afca2 | 1087 | |
b734035b XG |
1088 | migration_update_rates(rs, end_time); |
1089 | ||
be8b02ed | 1090 | rs->target_page_count_prev = rs->target_page_count; |
d693c6f1 FF |
1091 | |
1092 | /* reset period counters */ | |
f664da80 | 1093 | rs->time_last_bitmap_sync = end_time; |
a66cd90c | 1094 | rs->num_dirty_pages_period = 0; |
aff3f660 | 1095 | rs->bytes_xfer_prev = stat64_get(&mig_stats.transferred); |
56e93d26 | 1096 | } |
b890902c | 1097 | if (migrate_events()) { |
aff3f660 | 1098 | uint64_t generation = stat64_get(&mig_stats.dirty_sync_count); |
536b5a4e | 1099 | qapi_event_send_migration_pass(generation); |
4addcd4f | 1100 | } |
56e93d26 JQ |
1101 | } |
1102 | ||
1e493be5 | 1103 | static void migration_bitmap_sync_precopy(RAMState *rs, bool last_stage) |
bd227060 WW |
1104 | { |
1105 | Error *local_err = NULL; | |
1106 | ||
1107 | /* | |
1108 | * The current notifier usage is just an optimization to migration, so we | |
1109 | * don't stop the normal migration process in the error case. | |
1110 | */ | |
1111 | if (precopy_notify(PRECOPY_NOTIFY_BEFORE_BITMAP_SYNC, &local_err)) { | |
1112 | error_report_err(local_err); | |
b4a1733c | 1113 | local_err = NULL; |
bd227060 WW |
1114 | } |
1115 | ||
1e493be5 | 1116 | migration_bitmap_sync(rs, last_stage); |
bd227060 WW |
1117 | |
1118 | if (precopy_notify(PRECOPY_NOTIFY_AFTER_BITMAP_SYNC, &local_err)) { | |
1119 | error_report_err(local_err); | |
1120 | } | |
1121 | } | |
1122 | ||
a4dbaf8e | 1123 | void ram_release_page(const char *rbname, uint64_t offset) |
47fe16ff JQ |
1124 | { |
1125 | if (!migrate_release_ram() || !migration_in_postcopy()) { | |
1126 | return; | |
1127 | } | |
1128 | ||
1129 | ram_discard_range(rbname, offset, TARGET_PAGE_SIZE); | |
1130 | } | |
1131 | ||
6c97ec5f XG |
1132 | /** |
1133 | * save_zero_page_to_file: send the zero page to the file | |
1134 | * | |
1135 | * Returns the size of data written to the file, 0 means the page is not | |
1136 | * a zero page | |
1137 | * | |
ec6f3ab9 | 1138 | * @pss: current PSS channel |
6c97ec5f XG |
1139 | * @block: block that contains the page we want to send |
1140 | * @offset: offset inside the block for the page | |
1141 | */ | |
37502df3 | 1142 | static int save_zero_page_to_file(PageSearchStatus *pss, QEMUFile *file, |
6c97ec5f XG |
1143 | RAMBlock *block, ram_addr_t offset) |
1144 | { | |
1145 | uint8_t *p = block->host + offset; | |
1146 | int len = 0; | |
1147 | ||
bad452a7 | 1148 | if (buffer_is_zero(p, TARGET_PAGE_SIZE)) { |
37502df3 | 1149 | len += save_page_header(pss, file, block, offset | RAM_SAVE_FLAG_ZERO); |
6c97ec5f XG |
1150 | qemu_put_byte(file, 0); |
1151 | len += 1; | |
47fe16ff | 1152 | ram_release_page(block->idstr, offset); |
6c97ec5f XG |
1153 | } |
1154 | return len; | |
1155 | } | |
1156 | ||
56e93d26 | 1157 | /** |
3d0684b2 | 1158 | * save_zero_page: send the zero page to the stream |
56e93d26 | 1159 | * |
3d0684b2 | 1160 | * Returns the number of pages written. |
56e93d26 | 1161 | * |
ec6f3ab9 | 1162 | * @pss: current PSS channel |
56e93d26 JQ |
1163 | * @block: block that contains the page we want to send |
1164 | * @offset: offset inside the block for the page | |
56e93d26 | 1165 | */ |
37502df3 | 1166 | static int save_zero_page(PageSearchStatus *pss, QEMUFile *f, RAMBlock *block, |
61717ea9 | 1167 | ram_addr_t offset) |
56e93d26 | 1168 | { |
37502df3 | 1169 | int len = save_zero_page_to_file(pss, f, block, offset); |
56e93d26 | 1170 | |
6c97ec5f | 1171 | if (len) { |
aff3f660 | 1172 | stat64_add(&mig_stats.zero_pages, 1); |
4c2d0f6d | 1173 | ram_transferred_add(len); |
6c97ec5f | 1174 | return 1; |
56e93d26 | 1175 | } |
6c97ec5f | 1176 | return -1; |
56e93d26 JQ |
1177 | } |
1178 | ||
059ff0fb XG |
1179 | /* |
1180 | * @pages: the number of pages written by the control path, | |
1181 | * < 0 - error | |
1182 | * > 0 - number of pages written | |
1183 | * | |
1184 | * Return true if the pages has been saved, otherwise false is returned. | |
1185 | */ | |
61717ea9 PX |
1186 | static bool control_save_page(PageSearchStatus *pss, RAMBlock *block, |
1187 | ram_addr_t offset, int *pages) | |
059ff0fb | 1188 | { |
059ff0fb XG |
1189 | int ret; |
1190 | ||
61717ea9 | 1191 | ret = ram_control_save_page(pss->pss_channel, block->offset, offset, |
9c53d369 | 1192 | TARGET_PAGE_SIZE); |
059ff0fb XG |
1193 | if (ret == RAM_SAVE_CONTROL_NOT_SUPP) { |
1194 | return false; | |
1195 | } | |
1196 | ||
059ff0fb | 1197 | if (ret == RAM_SAVE_CONTROL_DELAYED) { |
9c53d369 | 1198 | *pages = 1; |
059ff0fb XG |
1199 | return true; |
1200 | } | |
9c53d369 | 1201 | *pages = ret; |
059ff0fb XG |
1202 | return true; |
1203 | } | |
1204 | ||
65dacaa0 XG |
1205 | /* |
1206 | * directly send the page to the stream | |
1207 | * | |
1208 | * Returns the number of pages written. | |
1209 | * | |
ec6f3ab9 | 1210 | * @pss: current PSS channel |
65dacaa0 XG |
1211 | * @block: block that contains the page we want to send |
1212 | * @offset: offset inside the block for the page | |
1213 | * @buf: the page to be sent | |
1214 | * @async: send to page asyncly | |
1215 | */ | |
ec6f3ab9 | 1216 | static int save_normal_page(PageSearchStatus *pss, RAMBlock *block, |
61717ea9 | 1217 | ram_addr_t offset, uint8_t *buf, bool async) |
65dacaa0 | 1218 | { |
ec6f3ab9 PX |
1219 | QEMUFile *file = pss->pss_channel; |
1220 | ||
37502df3 | 1221 | ram_transferred_add(save_page_header(pss, pss->pss_channel, block, |
4c2d0f6d | 1222 | offset | RAM_SAVE_FLAG_PAGE)); |
65dacaa0 | 1223 | if (async) { |
61717ea9 | 1224 | qemu_put_buffer_async(file, buf, TARGET_PAGE_SIZE, |
f912ec5b | 1225 | migrate_release_ram() && |
65dacaa0 XG |
1226 | migration_in_postcopy()); |
1227 | } else { | |
61717ea9 | 1228 | qemu_put_buffer(file, buf, TARGET_PAGE_SIZE); |
65dacaa0 | 1229 | } |
4c2d0f6d | 1230 | ram_transferred_add(TARGET_PAGE_SIZE); |
aff3f660 | 1231 | stat64_add(&mig_stats.normal_pages, 1); |
65dacaa0 XG |
1232 | return 1; |
1233 | } | |
1234 | ||
56e93d26 | 1235 | /** |
3d0684b2 | 1236 | * ram_save_page: send the given page to the stream |
56e93d26 | 1237 | * |
3d0684b2 | 1238 | * Returns the number of pages written. |
3fd3c4b3 DDAG |
1239 | * < 0 - error |
1240 | * >=0 - Number of pages written - this might legally be 0 | |
1241 | * if xbzrle noticed the page was the same. | |
56e93d26 | 1242 | * |
6f37bb8b | 1243 | * @rs: current RAM state |
56e93d26 JQ |
1244 | * @block: block that contains the page we want to send |
1245 | * @offset: offset inside the block for the page | |
56e93d26 | 1246 | */ |
05931ec5 | 1247 | static int ram_save_page(RAMState *rs, PageSearchStatus *pss) |
56e93d26 JQ |
1248 | { |
1249 | int pages = -1; | |
56e93d26 | 1250 | uint8_t *p; |
56e93d26 | 1251 | bool send_async = true; |
a08f6890 | 1252 | RAMBlock *block = pss->block; |
8bba004c | 1253 | ram_addr_t offset = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS; |
059ff0fb | 1254 | ram_addr_t current_addr = block->offset + offset; |
56e93d26 | 1255 | |
2f68e399 | 1256 | p = block->host + offset; |
1db9d8e5 | 1257 | trace_ram_save_page(block->idstr, (uint64_t)offset, p); |
56e93d26 | 1258 | |
56e93d26 | 1259 | XBZRLE_cache_lock(); |
f3095cc8 | 1260 | if (rs->xbzrle_started && !migration_in_postcopy()) { |
ec6f3ab9 | 1261 | pages = save_xbzrle_page(rs, pss, &p, current_addr, |
61717ea9 | 1262 | block, offset); |
05931ec5 | 1263 | if (!rs->last_stage) { |
059ff0fb XG |
1264 | /* Can't send this cached data async, since the cache page |
1265 | * might get updated before it gets to the wire | |
56e93d26 | 1266 | */ |
059ff0fb | 1267 | send_async = false; |
56e93d26 JQ |
1268 | } |
1269 | } | |
1270 | ||
1271 | /* XBZRLE overflow or normal page */ | |
1272 | if (pages == -1) { | |
ec6f3ab9 | 1273 | pages = save_normal_page(pss, block, offset, p, send_async); |
56e93d26 JQ |
1274 | } |
1275 | ||
1276 | XBZRLE_cache_unlock(); | |
1277 | ||
1278 | return pages; | |
1279 | } | |
1280 | ||
61717ea9 | 1281 | static int ram_save_multifd_page(QEMUFile *file, RAMBlock *block, |
b9ee2f7d JQ |
1282 | ram_addr_t offset) |
1283 | { | |
61717ea9 | 1284 | if (multifd_queue_page(file, block, offset) < 0) { |
713f762a IR |
1285 | return -1; |
1286 | } | |
aff3f660 | 1287 | stat64_add(&mig_stats.normal_pages, 1); |
b9ee2f7d JQ |
1288 | |
1289 | return 1; | |
1290 | } | |
1291 | ||
5e5fdcff XG |
1292 | static void |
1293 | update_compress_thread_counts(const CompressParam *param, int bytes_xmit) | |
1294 | { | |
4c2d0f6d | 1295 | ram_transferred_add(bytes_xmit); |
76e03000 | 1296 | |
97274a87 | 1297 | if (param->result == RES_ZEROPAGE) { |
aff3f660 | 1298 | stat64_add(&mig_stats.zero_pages, 1); |
76e03000 | 1299 | return; |
5e5fdcff | 1300 | } |
76e03000 XG |
1301 | |
1302 | /* 8 means a header with RAM_SAVE_FLAG_CONTINUE. */ | |
1303 | compression_counters.compressed_size += bytes_xmit - 8; | |
1304 | compression_counters.pages++; | |
56e93d26 JQ |
1305 | } |
1306 | ||
32b05495 XG |
1307 | static bool save_page_use_compression(RAMState *rs); |
1308 | ||
3e81763e | 1309 | static int send_queued_data(CompressParam *param) |
56e93d26 | 1310 | { |
3e81763e | 1311 | PageSearchStatus *pss = &ram_state->pss[RAM_CHANNEL_PRECOPY]; |
eaa238ab | 1312 | MigrationState *ms = migrate_get_current(); |
3e81763e LS |
1313 | QEMUFile *file = ms->to_dst_file; |
1314 | int len = 0; | |
1315 | ||
1316 | RAMBlock *block = param->block; | |
1317 | ram_addr_t offset = param->offset; | |
1318 | ||
1319 | if (param->result == RES_NONE) { | |
1320 | return 0; | |
1321 | } | |
1322 | ||
1323 | assert(block == pss->last_sent_block); | |
1324 | ||
1325 | if (param->result == RES_ZEROPAGE) { | |
4024cc85 | 1326 | assert(qemu_file_buffer_empty(param->file)); |
3e81763e LS |
1327 | len += save_page_header(pss, file, block, offset | RAM_SAVE_FLAG_ZERO); |
1328 | qemu_put_byte(file, 0); | |
1329 | len += 1; | |
1330 | ram_release_page(block->idstr, offset); | |
1331 | } else if (param->result == RES_COMPRESS) { | |
4024cc85 | 1332 | assert(!qemu_file_buffer_empty(param->file)); |
3e81763e LS |
1333 | len += save_page_header(pss, file, block, |
1334 | offset | RAM_SAVE_FLAG_COMPRESS_PAGE); | |
1335 | len += qemu_put_qemu_file(file, param->file); | |
1336 | } else { | |
1337 | abort(); | |
1338 | } | |
1339 | ||
680628d2 LS |
1340 | update_compress_thread_counts(param, len); |
1341 | ||
3e81763e LS |
1342 | return len; |
1343 | } | |
1344 | ||
ef4f5f5d LS |
1345 | static void ram_flush_compressed_data(RAMState *rs) |
1346 | { | |
1347 | if (!save_page_use_compression(rs)) { | |
1348 | return; | |
1349 | } | |
1350 | ||
1351 | flush_compressed_data(send_queued_data); | |
1352 | } | |
1353 | ||
31e2ac74 JQ |
1354 | #define PAGE_ALL_CLEAN 0 |
1355 | #define PAGE_TRY_AGAIN 1 | |
1356 | #define PAGE_DIRTY_FOUND 2 | |
3d0684b2 JQ |
1357 | /** |
1358 | * find_dirty_block: find the next dirty page and update any state | |
1359 | * associated with the search process. | |
b9e60928 | 1360 | * |
31e2ac74 | 1361 | * Returns: |
294e5a40 | 1362 | * <0: An error happened |
31e2ac74 JQ |
1363 | * PAGE_ALL_CLEAN: no dirty page found, give up |
1364 | * PAGE_TRY_AGAIN: no dirty page found, retry for next block | |
1365 | * PAGE_DIRTY_FOUND: dirty page found | |
b9e60928 | 1366 | * |
6f37bb8b | 1367 | * @rs: current RAM state |
3d0684b2 JQ |
1368 | * @pss: data about the state of the current dirty page scan |
1369 | * @again: set to false if the search has scanned the whole of RAM | |
b9e60928 | 1370 | */ |
31e2ac74 | 1371 | static int find_dirty_block(RAMState *rs, PageSearchStatus *pss) |
b9e60928 | 1372 | { |
d9e474ea PX |
1373 | /* Update pss->page for the next dirty bit in ramblock */ |
1374 | pss_find_next_dirty(pss); | |
1375 | ||
6f37bb8b | 1376 | if (pss->complete_round && pss->block == rs->last_seen_block && |
a935e30f | 1377 | pss->page >= rs->last_page) { |
b9e60928 DDAG |
1378 | /* |
1379 | * We've been once around the RAM and haven't found anything. | |
1380 | * Give up. | |
1381 | */ | |
31e2ac74 | 1382 | return PAGE_ALL_CLEAN; |
b9e60928 | 1383 | } |
542147f4 DH |
1384 | if (!offset_in_ramblock(pss->block, |
1385 | ((ram_addr_t)pss->page) << TARGET_PAGE_BITS)) { | |
b9e60928 | 1386 | /* Didn't find anything in this RAM Block */ |
a935e30f | 1387 | pss->page = 0; |
b9e60928 DDAG |
1388 | pss->block = QLIST_NEXT_RCU(pss->block, next); |
1389 | if (!pss->block) { | |
294e5a40 JQ |
1390 | if (!migrate_multifd_flush_after_each_section()) { |
1391 | QEMUFile *f = rs->pss[RAM_CHANNEL_PRECOPY].pss_channel; | |
1392 | int ret = multifd_send_sync_main(f); | |
1393 | if (ret < 0) { | |
1394 | return ret; | |
1395 | } | |
1396 | qemu_put_be64(f, RAM_SAVE_FLAG_MULTIFD_FLUSH); | |
1397 | qemu_fflush(f); | |
1398 | } | |
48df9d80 XG |
1399 | /* |
1400 | * If memory migration starts over, we will meet a dirtied page | |
1401 | * which may still exists in compression threads's ring, so we | |
1402 | * should flush the compressed data to make sure the new page | |
1403 | * is not overwritten by the old one in the destination. | |
1404 | * | |
1405 | * Also If xbzrle is on, stop using the data compression at this | |
1406 | * point. In theory, xbzrle can do better than compression. | |
1407 | */ | |
ef4f5f5d | 1408 | ram_flush_compressed_data(rs); |
48df9d80 | 1409 | |
b9e60928 DDAG |
1410 | /* Hit the end of the list */ |
1411 | pss->block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1412 | /* Flag that we've looped */ | |
1413 | pss->complete_round = true; | |
1a373522 | 1414 | /* After the first round, enable XBZRLE. */ |
87dca0c9 | 1415 | if (migrate_xbzrle()) { |
f3095cc8 | 1416 | rs->xbzrle_started = true; |
1a373522 | 1417 | } |
b9e60928 DDAG |
1418 | } |
1419 | /* Didn't find anything this time, but try again on the new block */ | |
31e2ac74 | 1420 | return PAGE_TRY_AGAIN; |
b9e60928 | 1421 | } else { |
31e2ac74 JQ |
1422 | /* We've found something */ |
1423 | return PAGE_DIRTY_FOUND; | |
b9e60928 DDAG |
1424 | } |
1425 | } | |
1426 | ||
3d0684b2 JQ |
1427 | /** |
1428 | * unqueue_page: gets a page of the queue | |
1429 | * | |
a82d593b | 1430 | * Helper for 'get_queued_page' - gets a page off the queue |
a82d593b | 1431 | * |
3d0684b2 JQ |
1432 | * Returns the block of the page (or NULL if none available) |
1433 | * | |
ec481c6c | 1434 | * @rs: current RAM state |
3d0684b2 | 1435 | * @offset: used to return the offset within the RAMBlock |
a82d593b | 1436 | */ |
f20e2865 | 1437 | static RAMBlock *unqueue_page(RAMState *rs, ram_addr_t *offset) |
a82d593b | 1438 | { |
a1fe28df | 1439 | struct RAMSrcPageRequest *entry; |
a82d593b DDAG |
1440 | RAMBlock *block = NULL; |
1441 | ||
a1fe28df | 1442 | if (!postcopy_has_request(rs)) { |
ae526e32 XG |
1443 | return NULL; |
1444 | } | |
1445 | ||
6e8a355d | 1446 | QEMU_LOCK_GUARD(&rs->src_page_req_mutex); |
a1fe28df PX |
1447 | |
1448 | /* | |
1449 | * This should _never_ change even after we take the lock, because no one | |
1450 | * should be taking anything off the request list other than us. | |
1451 | */ | |
1452 | assert(postcopy_has_request(rs)); | |
1453 | ||
1454 | entry = QSIMPLEQ_FIRST(&rs->src_page_requests); | |
1455 | block = entry->rb; | |
1456 | *offset = entry->offset; | |
1457 | ||
777f53c7 TH |
1458 | if (entry->len > TARGET_PAGE_SIZE) { |
1459 | entry->len -= TARGET_PAGE_SIZE; | |
1460 | entry->offset += TARGET_PAGE_SIZE; | |
a1fe28df PX |
1461 | } else { |
1462 | memory_region_unref(block->mr); | |
1463 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); | |
1464 | g_free(entry); | |
1465 | migration_consume_urgent_request(); | |
a82d593b | 1466 | } |
a82d593b DDAG |
1467 | |
1468 | return block; | |
1469 | } | |
1470 | ||
278e2f55 AG |
1471 | #if defined(__linux__) |
1472 | /** | |
1473 | * poll_fault_page: try to get next UFFD write fault page and, if pending fault | |
1474 | * is found, return RAM block pointer and page offset | |
1475 | * | |
1476 | * Returns pointer to the RAMBlock containing faulting page, | |
1477 | * NULL if no write faults are pending | |
1478 | * | |
1479 | * @rs: current RAM state | |
1480 | * @offset: page offset from the beginning of the block | |
1481 | */ | |
1482 | static RAMBlock *poll_fault_page(RAMState *rs, ram_addr_t *offset) | |
1483 | { | |
1484 | struct uffd_msg uffd_msg; | |
1485 | void *page_address; | |
82ea3e3b | 1486 | RAMBlock *block; |
278e2f55 AG |
1487 | int res; |
1488 | ||
1489 | if (!migrate_background_snapshot()) { | |
1490 | return NULL; | |
1491 | } | |
1492 | ||
1493 | res = uffd_read_events(rs->uffdio_fd, &uffd_msg, 1); | |
1494 | if (res <= 0) { | |
1495 | return NULL; | |
1496 | } | |
1497 | ||
1498 | page_address = (void *)(uintptr_t) uffd_msg.arg.pagefault.address; | |
82ea3e3b AG |
1499 | block = qemu_ram_block_from_host(page_address, false, offset); |
1500 | assert(block && (block->flags & RAM_UF_WRITEPROTECT) != 0); | |
1501 | return block; | |
278e2f55 AG |
1502 | } |
1503 | ||
1504 | /** | |
1505 | * ram_save_release_protection: release UFFD write protection after | |
1506 | * a range of pages has been saved | |
1507 | * | |
1508 | * @rs: current RAM state | |
1509 | * @pss: page-search-status structure | |
1510 | * @start_page: index of the first page in the range relative to pss->block | |
1511 | * | |
1512 | * Returns 0 on success, negative value in case of an error | |
1513 | */ | |
1514 | static int ram_save_release_protection(RAMState *rs, PageSearchStatus *pss, | |
1515 | unsigned long start_page) | |
1516 | { | |
1517 | int res = 0; | |
1518 | ||
1519 | /* Check if page is from UFFD-managed region. */ | |
1520 | if (pss->block->flags & RAM_UF_WRITEPROTECT) { | |
1521 | void *page_address = pss->block->host + (start_page << TARGET_PAGE_BITS); | |
258f5c98 | 1522 | uint64_t run_length = (pss->page - start_page) << TARGET_PAGE_BITS; |
278e2f55 AG |
1523 | |
1524 | /* Flush async buffers before un-protect. */ | |
61717ea9 | 1525 | qemu_fflush(pss->pss_channel); |
278e2f55 AG |
1526 | /* Un-protect memory range. */ |
1527 | res = uffd_change_protection(rs->uffdio_fd, page_address, run_length, | |
1528 | false, false); | |
1529 | } | |
1530 | ||
1531 | return res; | |
1532 | } | |
1533 | ||
1534 | /* ram_write_tracking_available: check if kernel supports required UFFD features | |
1535 | * | |
1536 | * Returns true if supports, false otherwise | |
1537 | */ | |
1538 | bool ram_write_tracking_available(void) | |
1539 | { | |
1540 | uint64_t uffd_features; | |
1541 | int res; | |
1542 | ||
1543 | res = uffd_query_features(&uffd_features); | |
1544 | return (res == 0 && | |
1545 | (uffd_features & UFFD_FEATURE_PAGEFAULT_FLAG_WP) != 0); | |
1546 | } | |
1547 | ||
1548 | /* ram_write_tracking_compatible: check if guest configuration is | |
1549 | * compatible with 'write-tracking' | |
1550 | * | |
1551 | * Returns true if compatible, false otherwise | |
1552 | */ | |
1553 | bool ram_write_tracking_compatible(void) | |
1554 | { | |
1555 | const uint64_t uffd_ioctls_mask = BIT(_UFFDIO_WRITEPROTECT); | |
1556 | int uffd_fd; | |
82ea3e3b | 1557 | RAMBlock *block; |
278e2f55 AG |
1558 | bool ret = false; |
1559 | ||
1560 | /* Open UFFD file descriptor */ | |
1561 | uffd_fd = uffd_create_fd(UFFD_FEATURE_PAGEFAULT_FLAG_WP, false); | |
1562 | if (uffd_fd < 0) { | |
1563 | return false; | |
1564 | } | |
1565 | ||
1566 | RCU_READ_LOCK_GUARD(); | |
1567 | ||
82ea3e3b | 1568 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
278e2f55 AG |
1569 | uint64_t uffd_ioctls; |
1570 | ||
1571 | /* Nothing to do with read-only and MMIO-writable regions */ | |
82ea3e3b | 1572 | if (block->mr->readonly || block->mr->rom_device) { |
278e2f55 AG |
1573 | continue; |
1574 | } | |
1575 | /* Try to register block memory via UFFD-IO to track writes */ | |
82ea3e3b | 1576 | if (uffd_register_memory(uffd_fd, block->host, block->max_length, |
278e2f55 AG |
1577 | UFFDIO_REGISTER_MODE_WP, &uffd_ioctls)) { |
1578 | goto out; | |
1579 | } | |
1580 | if ((uffd_ioctls & uffd_ioctls_mask) != uffd_ioctls_mask) { | |
1581 | goto out; | |
1582 | } | |
1583 | } | |
1584 | ret = true; | |
1585 | ||
1586 | out: | |
1587 | uffd_close_fd(uffd_fd); | |
1588 | return ret; | |
1589 | } | |
1590 | ||
f7b9dcfb DH |
1591 | static inline void populate_read_range(RAMBlock *block, ram_addr_t offset, |
1592 | ram_addr_t size) | |
1593 | { | |
5f19a449 DH |
1594 | const ram_addr_t end = offset + size; |
1595 | ||
f7b9dcfb DH |
1596 | /* |
1597 | * We read one byte of each page; this will preallocate page tables if | |
1598 | * required and populate the shared zeropage on MAP_PRIVATE anonymous memory | |
1599 | * where no page was populated yet. This might require adaption when | |
1600 | * supporting other mappings, like shmem. | |
1601 | */ | |
5f19a449 | 1602 | for (; offset < end; offset += block->page_size) { |
f7b9dcfb DH |
1603 | char tmp = *((char *)block->host + offset); |
1604 | ||
1605 | /* Don't optimize the read out */ | |
1606 | asm volatile("" : "+r" (tmp)); | |
1607 | } | |
1608 | } | |
1609 | ||
6fee3a1f DH |
1610 | static inline int populate_read_section(MemoryRegionSection *section, |
1611 | void *opaque) | |
1612 | { | |
1613 | const hwaddr size = int128_get64(section->size); | |
1614 | hwaddr offset = section->offset_within_region; | |
1615 | RAMBlock *block = section->mr->ram_block; | |
1616 | ||
1617 | populate_read_range(block, offset, size); | |
1618 | return 0; | |
1619 | } | |
1620 | ||
eeccb99c | 1621 | /* |
f7b9dcfb DH |
1622 | * ram_block_populate_read: preallocate page tables and populate pages in the |
1623 | * RAM block by reading a byte of each page. | |
eeccb99c AG |
1624 | * |
1625 | * Since it's solely used for userfault_fd WP feature, here we just | |
1626 | * hardcode page size to qemu_real_host_page_size. | |
1627 | * | |
82ea3e3b | 1628 | * @block: RAM block to populate |
eeccb99c | 1629 | */ |
6fee3a1f | 1630 | static void ram_block_populate_read(RAMBlock *rb) |
eeccb99c | 1631 | { |
6fee3a1f DH |
1632 | /* |
1633 | * Skip populating all pages that fall into a discarded range as managed by | |
1634 | * a RamDiscardManager responsible for the mapped memory region of the | |
1635 | * RAMBlock. Such discarded ("logically unplugged") parts of a RAMBlock | |
1636 | * must not get populated automatically. We don't have to track | |
1637 | * modifications via userfaultfd WP reliably, because these pages will | |
1638 | * not be part of the migration stream either way -- see | |
1639 | * ramblock_dirty_bitmap_exclude_discarded_pages(). | |
1640 | * | |
1641 | * Note: The result is only stable while migrating (precopy/postcopy). | |
1642 | */ | |
1643 | if (rb->mr && memory_region_has_ram_discard_manager(rb->mr)) { | |
1644 | RamDiscardManager *rdm = memory_region_get_ram_discard_manager(rb->mr); | |
1645 | MemoryRegionSection section = { | |
1646 | .mr = rb->mr, | |
1647 | .offset_within_region = 0, | |
1648 | .size = rb->mr->size, | |
1649 | }; | |
1650 | ||
1651 | ram_discard_manager_replay_populated(rdm, §ion, | |
1652 | populate_read_section, NULL); | |
1653 | } else { | |
1654 | populate_read_range(rb, 0, rb->used_length); | |
1655 | } | |
eeccb99c AG |
1656 | } |
1657 | ||
1658 | /* | |
1659 | * ram_write_tracking_prepare: prepare for UFFD-WP memory tracking | |
1660 | */ | |
1661 | void ram_write_tracking_prepare(void) | |
1662 | { | |
82ea3e3b | 1663 | RAMBlock *block; |
eeccb99c AG |
1664 | |
1665 | RCU_READ_LOCK_GUARD(); | |
1666 | ||
82ea3e3b | 1667 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
eeccb99c | 1668 | /* Nothing to do with read-only and MMIO-writable regions */ |
82ea3e3b | 1669 | if (block->mr->readonly || block->mr->rom_device) { |
eeccb99c AG |
1670 | continue; |
1671 | } | |
1672 | ||
1673 | /* | |
1674 | * Populate pages of the RAM block before enabling userfault_fd | |
1675 | * write protection. | |
1676 | * | |
1677 | * This stage is required since ioctl(UFFDIO_WRITEPROTECT) with | |
1678 | * UFFDIO_WRITEPROTECT_MODE_WP mode setting would silently skip | |
1679 | * pages with pte_none() entries in page table. | |
1680 | */ | |
f7b9dcfb | 1681 | ram_block_populate_read(block); |
eeccb99c AG |
1682 | } |
1683 | } | |
1684 | ||
e41c5770 DH |
1685 | static inline int uffd_protect_section(MemoryRegionSection *section, |
1686 | void *opaque) | |
1687 | { | |
1688 | const hwaddr size = int128_get64(section->size); | |
1689 | const hwaddr offset = section->offset_within_region; | |
1690 | RAMBlock *rb = section->mr->ram_block; | |
1691 | int uffd_fd = (uintptr_t)opaque; | |
1692 | ||
1693 | return uffd_change_protection(uffd_fd, rb->host + offset, size, true, | |
1694 | false); | |
1695 | } | |
1696 | ||
1697 | static int ram_block_uffd_protect(RAMBlock *rb, int uffd_fd) | |
1698 | { | |
1699 | assert(rb->flags & RAM_UF_WRITEPROTECT); | |
1700 | ||
1701 | /* See ram_block_populate_read() */ | |
1702 | if (rb->mr && memory_region_has_ram_discard_manager(rb->mr)) { | |
1703 | RamDiscardManager *rdm = memory_region_get_ram_discard_manager(rb->mr); | |
1704 | MemoryRegionSection section = { | |
1705 | .mr = rb->mr, | |
1706 | .offset_within_region = 0, | |
1707 | .size = rb->mr->size, | |
1708 | }; | |
1709 | ||
1710 | return ram_discard_manager_replay_populated(rdm, §ion, | |
1711 | uffd_protect_section, | |
1712 | (void *)(uintptr_t)uffd_fd); | |
1713 | } | |
1714 | return uffd_change_protection(uffd_fd, rb->host, | |
1715 | rb->used_length, true, false); | |
1716 | } | |
1717 | ||
278e2f55 AG |
1718 | /* |
1719 | * ram_write_tracking_start: start UFFD-WP memory tracking | |
1720 | * | |
1721 | * Returns 0 for success or negative value in case of error | |
1722 | */ | |
1723 | int ram_write_tracking_start(void) | |
1724 | { | |
1725 | int uffd_fd; | |
1726 | RAMState *rs = ram_state; | |
82ea3e3b | 1727 | RAMBlock *block; |
278e2f55 AG |
1728 | |
1729 | /* Open UFFD file descriptor */ | |
1730 | uffd_fd = uffd_create_fd(UFFD_FEATURE_PAGEFAULT_FLAG_WP, true); | |
1731 | if (uffd_fd < 0) { | |
1732 | return uffd_fd; | |
1733 | } | |
1734 | rs->uffdio_fd = uffd_fd; | |
1735 | ||
1736 | RCU_READ_LOCK_GUARD(); | |
1737 | ||
82ea3e3b | 1738 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
278e2f55 | 1739 | /* Nothing to do with read-only and MMIO-writable regions */ |
82ea3e3b | 1740 | if (block->mr->readonly || block->mr->rom_device) { |
278e2f55 AG |
1741 | continue; |
1742 | } | |
1743 | ||
1744 | /* Register block memory with UFFD to track writes */ | |
82ea3e3b AG |
1745 | if (uffd_register_memory(rs->uffdio_fd, block->host, |
1746 | block->max_length, UFFDIO_REGISTER_MODE_WP, NULL)) { | |
278e2f55 AG |
1747 | goto fail; |
1748 | } | |
72ef3a37 DH |
1749 | block->flags |= RAM_UF_WRITEPROTECT; |
1750 | memory_region_ref(block->mr); | |
1751 | ||
278e2f55 | 1752 | /* Apply UFFD write protection to the block memory range */ |
e41c5770 | 1753 | if (ram_block_uffd_protect(block, uffd_fd)) { |
278e2f55 AG |
1754 | goto fail; |
1755 | } | |
278e2f55 | 1756 | |
82ea3e3b AG |
1757 | trace_ram_write_tracking_ramblock_start(block->idstr, block->page_size, |
1758 | block->host, block->max_length); | |
278e2f55 AG |
1759 | } |
1760 | ||
1761 | return 0; | |
1762 | ||
1763 | fail: | |
1764 | error_report("ram_write_tracking_start() failed: restoring initial memory state"); | |
1765 | ||
82ea3e3b AG |
1766 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
1767 | if ((block->flags & RAM_UF_WRITEPROTECT) == 0) { | |
278e2f55 AG |
1768 | continue; |
1769 | } | |
82ea3e3b | 1770 | uffd_unregister_memory(rs->uffdio_fd, block->host, block->max_length); |
278e2f55 | 1771 | /* Cleanup flags and remove reference */ |
82ea3e3b AG |
1772 | block->flags &= ~RAM_UF_WRITEPROTECT; |
1773 | memory_region_unref(block->mr); | |
278e2f55 AG |
1774 | } |
1775 | ||
1776 | uffd_close_fd(uffd_fd); | |
1777 | rs->uffdio_fd = -1; | |
1778 | return -1; | |
1779 | } | |
1780 | ||
1781 | /** | |
1782 | * ram_write_tracking_stop: stop UFFD-WP memory tracking and remove protection | |
1783 | */ | |
1784 | void ram_write_tracking_stop(void) | |
1785 | { | |
1786 | RAMState *rs = ram_state; | |
82ea3e3b | 1787 | RAMBlock *block; |
278e2f55 AG |
1788 | |
1789 | RCU_READ_LOCK_GUARD(); | |
1790 | ||
82ea3e3b AG |
1791 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
1792 | if ((block->flags & RAM_UF_WRITEPROTECT) == 0) { | |
278e2f55 AG |
1793 | continue; |
1794 | } | |
82ea3e3b | 1795 | uffd_unregister_memory(rs->uffdio_fd, block->host, block->max_length); |
278e2f55 | 1796 | |
82ea3e3b AG |
1797 | trace_ram_write_tracking_ramblock_stop(block->idstr, block->page_size, |
1798 | block->host, block->max_length); | |
278e2f55 AG |
1799 | |
1800 | /* Cleanup flags and remove reference */ | |
82ea3e3b AG |
1801 | block->flags &= ~RAM_UF_WRITEPROTECT; |
1802 | memory_region_unref(block->mr); | |
278e2f55 AG |
1803 | } |
1804 | ||
1805 | /* Finally close UFFD file descriptor */ | |
1806 | uffd_close_fd(rs->uffdio_fd); | |
1807 | rs->uffdio_fd = -1; | |
1808 | } | |
1809 | ||
1810 | #else | |
1811 | /* No target OS support, stubs just fail or ignore */ | |
1812 | ||
1813 | static RAMBlock *poll_fault_page(RAMState *rs, ram_addr_t *offset) | |
1814 | { | |
1815 | (void) rs; | |
1816 | (void) offset; | |
1817 | ||
1818 | return NULL; | |
1819 | } | |
1820 | ||
1821 | static int ram_save_release_protection(RAMState *rs, PageSearchStatus *pss, | |
1822 | unsigned long start_page) | |
1823 | { | |
1824 | (void) rs; | |
1825 | (void) pss; | |
1826 | (void) start_page; | |
1827 | ||
1828 | return 0; | |
1829 | } | |
1830 | ||
1831 | bool ram_write_tracking_available(void) | |
1832 | { | |
1833 | return false; | |
1834 | } | |
1835 | ||
1836 | bool ram_write_tracking_compatible(void) | |
1837 | { | |
1838 | assert(0); | |
1839 | return false; | |
1840 | } | |
1841 | ||
1842 | int ram_write_tracking_start(void) | |
1843 | { | |
1844 | assert(0); | |
1845 | return -1; | |
1846 | } | |
1847 | ||
1848 | void ram_write_tracking_stop(void) | |
1849 | { | |
1850 | assert(0); | |
1851 | } | |
1852 | #endif /* defined(__linux__) */ | |
1853 | ||
3d0684b2 | 1854 | /** |
ff1543af | 1855 | * get_queued_page: unqueue a page from the postcopy requests |
3d0684b2 JQ |
1856 | * |
1857 | * Skips pages that are already sent (!dirty) | |
a82d593b | 1858 | * |
a5f7b1a6 | 1859 | * Returns true if a queued page is found |
a82d593b | 1860 | * |
6f37bb8b | 1861 | * @rs: current RAM state |
3d0684b2 | 1862 | * @pss: data about the state of the current dirty page scan |
a82d593b | 1863 | */ |
f20e2865 | 1864 | static bool get_queued_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b DDAG |
1865 | { |
1866 | RAMBlock *block; | |
1867 | ram_addr_t offset; | |
777f53c7 TH |
1868 | bool dirty; |
1869 | ||
1870 | do { | |
1871 | block = unqueue_page(rs, &offset); | |
1872 | /* | |
1873 | * We're sending this page, and since it's postcopy nothing else | |
1874 | * will dirty it, and we must make sure it doesn't get sent again | |
1875 | * even if this queue request was received after the background | |
1876 | * search already sent it. | |
1877 | */ | |
1878 | if (block) { | |
1879 | unsigned long page; | |
1880 | ||
1881 | page = offset >> TARGET_PAGE_BITS; | |
1882 | dirty = test_bit(page, block->bmap); | |
1883 | if (!dirty) { | |
1884 | trace_get_queued_page_not_dirty(block->idstr, (uint64_t)offset, | |
1885 | page); | |
1886 | } else { | |
1887 | trace_get_queued_page(block->idstr, (uint64_t)offset, page); | |
1888 | } | |
1889 | } | |
a82d593b | 1890 | |
777f53c7 | 1891 | } while (block && !dirty); |
a82d593b | 1892 | |
b062106d | 1893 | if (!block) { |
278e2f55 AG |
1894 | /* |
1895 | * Poll write faults too if background snapshot is enabled; that's | |
1896 | * when we have vcpus got blocked by the write protected pages. | |
1897 | */ | |
1898 | block = poll_fault_page(rs, &offset); | |
1899 | } | |
1900 | ||
a82d593b | 1901 | if (block) { |
a82d593b DDAG |
1902 | /* |
1903 | * We want the background search to continue from the queued page | |
1904 | * since the guest is likely to want other pages near to the page | |
1905 | * it just requested. | |
1906 | */ | |
1907 | pss->block = block; | |
a935e30f | 1908 | pss->page = offset >> TARGET_PAGE_BITS; |
422314e7 WY |
1909 | |
1910 | /* | |
1911 | * This unqueued page would break the "one round" check, even is | |
1912 | * really rare. | |
1913 | */ | |
1914 | pss->complete_round = false; | |
a82d593b DDAG |
1915 | } |
1916 | ||
1917 | return !!block; | |
1918 | } | |
1919 | ||
6c595cde | 1920 | /** |
5e58f968 JQ |
1921 | * migration_page_queue_free: drop any remaining pages in the ram |
1922 | * request queue | |
6c595cde | 1923 | * |
3d0684b2 JQ |
1924 | * It should be empty at the end anyway, but in error cases there may |
1925 | * be some left. in case that there is any page left, we drop it. | |
1926 | * | |
6c595cde | 1927 | */ |
83c13382 | 1928 | static void migration_page_queue_free(RAMState *rs) |
6c595cde | 1929 | { |
ec481c6c | 1930 | struct RAMSrcPageRequest *mspr, *next_mspr; |
6c595cde DDAG |
1931 | /* This queue generally should be empty - but in the case of a failed |
1932 | * migration might have some droppings in. | |
1933 | */ | |
89ac5a1d | 1934 | RCU_READ_LOCK_GUARD(); |
ec481c6c | 1935 | QSIMPLEQ_FOREACH_SAFE(mspr, &rs->src_page_requests, next_req, next_mspr) { |
6c595cde | 1936 | memory_region_unref(mspr->rb->mr); |
ec481c6c | 1937 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
6c595cde DDAG |
1938 | g_free(mspr); |
1939 | } | |
6c595cde DDAG |
1940 | } |
1941 | ||
1942 | /** | |
3d0684b2 JQ |
1943 | * ram_save_queue_pages: queue the page for transmission |
1944 | * | |
1945 | * A request from postcopy destination for example. | |
1946 | * | |
1947 | * Returns zero on success or negative on error | |
1948 | * | |
3d0684b2 JQ |
1949 | * @rbname: Name of the RAMBLock of the request. NULL means the |
1950 | * same that last one. | |
1951 | * @start: starting address from the start of the RAMBlock | |
1952 | * @len: length (in bytes) to send | |
6c595cde | 1953 | */ |
96506894 | 1954 | int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len) |
6c595cde DDAG |
1955 | { |
1956 | RAMBlock *ramblock; | |
53518d94 | 1957 | RAMState *rs = ram_state; |
6c595cde | 1958 | |
aff3f660 | 1959 | stat64_add(&mig_stats.postcopy_requests, 1); |
89ac5a1d DDAG |
1960 | RCU_READ_LOCK_GUARD(); |
1961 | ||
6c595cde DDAG |
1962 | if (!rbname) { |
1963 | /* Reuse last RAMBlock */ | |
68a098f3 | 1964 | ramblock = rs->last_req_rb; |
6c595cde DDAG |
1965 | |
1966 | if (!ramblock) { | |
1967 | /* | |
1968 | * Shouldn't happen, we can't reuse the last RAMBlock if | |
1969 | * it's the 1st request. | |
1970 | */ | |
1971 | error_report("ram_save_queue_pages no previous block"); | |
03acb4e9 | 1972 | return -1; |
6c595cde DDAG |
1973 | } |
1974 | } else { | |
1975 | ramblock = qemu_ram_block_by_name(rbname); | |
1976 | ||
1977 | if (!ramblock) { | |
1978 | /* We shouldn't be asked for a non-existent RAMBlock */ | |
1979 | error_report("ram_save_queue_pages no block '%s'", rbname); | |
03acb4e9 | 1980 | return -1; |
6c595cde | 1981 | } |
68a098f3 | 1982 | rs->last_req_rb = ramblock; |
6c595cde DDAG |
1983 | } |
1984 | trace_ram_save_queue_pages(ramblock->idstr, start, len); | |
542147f4 | 1985 | if (!offset_in_ramblock(ramblock, start + len - 1)) { |
9458ad6b JQ |
1986 | error_report("%s request overrun start=" RAM_ADDR_FMT " len=" |
1987 | RAM_ADDR_FMT " blocklen=" RAM_ADDR_FMT, | |
6c595cde | 1988 | __func__, start, len, ramblock->used_length); |
03acb4e9 | 1989 | return -1; |
6c595cde DDAG |
1990 | } |
1991 | ||
93589827 PX |
1992 | /* |
1993 | * When with postcopy preempt, we send back the page directly in the | |
1994 | * rp-return thread. | |
1995 | */ | |
1996 | if (postcopy_preempt_active()) { | |
1997 | ram_addr_t page_start = start >> TARGET_PAGE_BITS; | |
1998 | size_t page_size = qemu_ram_pagesize(ramblock); | |
1999 | PageSearchStatus *pss = &ram_state->pss[RAM_CHANNEL_POSTCOPY]; | |
2000 | int ret = 0; | |
2001 | ||
2002 | qemu_mutex_lock(&rs->bitmap_mutex); | |
2003 | ||
2004 | pss_init(pss, ramblock, page_start); | |
2005 | /* | |
2006 | * Always use the preempt channel, and make sure it's there. It's | |
2007 | * safe to access without lock, because when rp-thread is running | |
2008 | * we should be the only one who operates on the qemufile | |
2009 | */ | |
2010 | pss->pss_channel = migrate_get_current()->postcopy_qemufile_src; | |
93589827 PX |
2011 | assert(pss->pss_channel); |
2012 | ||
2013 | /* | |
2014 | * It must be either one or multiple of host page size. Just | |
2015 | * assert; if something wrong we're mostly split brain anyway. | |
2016 | */ | |
2017 | assert(len % page_size == 0); | |
2018 | while (len) { | |
2019 | if (ram_save_host_page_urgent(pss)) { | |
2020 | error_report("%s: ram_save_host_page_urgent() failed: " | |
2021 | "ramblock=%s, start_addr=0x"RAM_ADDR_FMT, | |
2022 | __func__, ramblock->idstr, start); | |
2023 | ret = -1; | |
2024 | break; | |
2025 | } | |
2026 | /* | |
2027 | * NOTE: after ram_save_host_page_urgent() succeeded, pss->page | |
2028 | * will automatically be moved and point to the next host page | |
2029 | * we're going to send, so no need to update here. | |
2030 | * | |
2031 | * Normally QEMU never sends >1 host page in requests, so | |
2032 | * logically we don't even need that as the loop should only | |
2033 | * run once, but just to be consistent. | |
2034 | */ | |
2035 | len -= page_size; | |
2036 | }; | |
2037 | qemu_mutex_unlock(&rs->bitmap_mutex); | |
2038 | ||
2039 | return ret; | |
2040 | } | |
2041 | ||
ec481c6c | 2042 | struct RAMSrcPageRequest *new_entry = |
b21e2380 | 2043 | g_new0(struct RAMSrcPageRequest, 1); |
6c595cde DDAG |
2044 | new_entry->rb = ramblock; |
2045 | new_entry->offset = start; | |
2046 | new_entry->len = len; | |
2047 | ||
2048 | memory_region_ref(ramblock->mr); | |
ec481c6c JQ |
2049 | qemu_mutex_lock(&rs->src_page_req_mutex); |
2050 | QSIMPLEQ_INSERT_TAIL(&rs->src_page_requests, new_entry, next_req); | |
e03a34f8 | 2051 | migration_make_urgent_request(); |
ec481c6c | 2052 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
6c595cde DDAG |
2053 | |
2054 | return 0; | |
6c595cde DDAG |
2055 | } |
2056 | ||
d7400a34 XG |
2057 | static bool save_page_use_compression(RAMState *rs) |
2058 | { | |
a7a94d14 | 2059 | if (!migrate_compress()) { |
d7400a34 XG |
2060 | return false; |
2061 | } | |
2062 | ||
2063 | /* | |
1a373522 DH |
2064 | * If xbzrle is enabled (e.g., after first round of migration), stop |
2065 | * using the data compression. In theory, xbzrle can do better than | |
2066 | * compression. | |
d7400a34 | 2067 | */ |
f3095cc8 | 2068 | if (rs->xbzrle_started) { |
1a373522 | 2069 | return false; |
d7400a34 XG |
2070 | } |
2071 | ||
1a373522 | 2072 | return true; |
d7400a34 XG |
2073 | } |
2074 | ||
5e5fdcff XG |
2075 | /* |
2076 | * try to compress the page before posting it out, return true if the page | |
2077 | * has been properly handled by compression, otherwise needs other | |
2078 | * paths to handle it | |
2079 | */ | |
ec6f3ab9 PX |
2080 | static bool save_compress_page(RAMState *rs, PageSearchStatus *pss, |
2081 | RAMBlock *block, ram_addr_t offset) | |
5e5fdcff XG |
2082 | { |
2083 | if (!save_page_use_compression(rs)) { | |
2084 | return false; | |
2085 | } | |
2086 | ||
2087 | /* | |
2088 | * When starting the process of a new block, the first page of | |
2089 | * the block should be sent out before other pages in the same | |
2090 | * block, and all the pages in last block should have been sent | |
2091 | * out, keeping this order is important, because the 'cont' flag | |
2092 | * is used to avoid resending the block name. | |
2093 | * | |
2094 | * We post the fist page as normal page as compression will take | |
2095 | * much CPU resource. | |
2096 | */ | |
ec6f3ab9 | 2097 | if (block != pss->last_sent_block) { |
ef4f5f5d | 2098 | ram_flush_compressed_data(rs); |
5e5fdcff XG |
2099 | return false; |
2100 | } | |
2101 | ||
ef4f5f5d | 2102 | if (compress_page_with_multi_thread(block, offset, send_queued_data) > 0) { |
5e5fdcff XG |
2103 | return true; |
2104 | } | |
2105 | ||
76e03000 | 2106 | compression_counters.busy++; |
5e5fdcff XG |
2107 | return false; |
2108 | } | |
2109 | ||
a82d593b | 2110 | /** |
4010ba38 | 2111 | * ram_save_target_page_legacy: save one target page |
a82d593b | 2112 | * |
3d0684b2 | 2113 | * Returns the number of pages written |
a82d593b | 2114 | * |
6f37bb8b | 2115 | * @rs: current RAM state |
3d0684b2 | 2116 | * @pss: data about the page we want to send |
a82d593b | 2117 | */ |
4010ba38 | 2118 | static int ram_save_target_page_legacy(RAMState *rs, PageSearchStatus *pss) |
a82d593b | 2119 | { |
a8ec91f9 | 2120 | RAMBlock *block = pss->block; |
8bba004c | 2121 | ram_addr_t offset = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS; |
a8ec91f9 XG |
2122 | int res; |
2123 | ||
61717ea9 | 2124 | if (control_save_page(pss, block, offset, &res)) { |
a8ec91f9 XG |
2125 | return res; |
2126 | } | |
2127 | ||
ec6f3ab9 | 2128 | if (save_compress_page(rs, pss, block, offset)) { |
5e5fdcff | 2129 | return 1; |
d7400a34 XG |
2130 | } |
2131 | ||
37502df3 | 2132 | res = save_zero_page(pss, pss->pss_channel, block, offset); |
d7400a34 XG |
2133 | if (res > 0) { |
2134 | /* Must let xbzrle know, otherwise a previous (now 0'd) cached | |
2135 | * page would be stale | |
2136 | */ | |
f3095cc8 | 2137 | if (rs->xbzrle_started) { |
d7400a34 XG |
2138 | XBZRLE_cache_lock(); |
2139 | xbzrle_cache_zero_page(rs, block->offset + offset); | |
2140 | XBZRLE_cache_unlock(); | |
2141 | } | |
d7400a34 XG |
2142 | return res; |
2143 | } | |
2144 | ||
da3f56cb | 2145 | /* |
6f39c90b PX |
2146 | * Do not use multifd in postcopy as one whole host page should be |
2147 | * placed. Meanwhile postcopy requires atomic update of pages, so even | |
2148 | * if host page size == guest page size the dest guest during run may | |
2149 | * still see partially copied pages which is data corruption. | |
da3f56cb | 2150 | */ |
51b07548 | 2151 | if (migrate_multifd() && !migration_in_postcopy()) { |
61717ea9 | 2152 | return ram_save_multifd_page(pss->pss_channel, block, offset); |
a82d593b DDAG |
2153 | } |
2154 | ||
05931ec5 | 2155 | return ram_save_page(rs, pss); |
a82d593b DDAG |
2156 | } |
2157 | ||
d9e474ea PX |
2158 | /* Should be called before sending a host page */ |
2159 | static void pss_host_page_prepare(PageSearchStatus *pss) | |
2160 | { | |
2161 | /* How many guest pages are there in one host page? */ | |
2162 | size_t guest_pfns = qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
2163 | ||
2164 | pss->host_page_sending = true; | |
301d7ffe PX |
2165 | if (guest_pfns <= 1) { |
2166 | /* | |
2167 | * This covers both when guest psize == host psize, or when guest | |
2168 | * has larger psize than the host (guest_pfns==0). | |
2169 | * | |
2170 | * For the latter, we always send one whole guest page per | |
2171 | * iteration of the host page (example: an Alpha VM on x86 host | |
2172 | * will have guest psize 8K while host psize 4K). | |
2173 | */ | |
2174 | pss->host_page_start = pss->page; | |
2175 | pss->host_page_end = pss->page + 1; | |
2176 | } else { | |
2177 | /* | |
2178 | * The host page spans over multiple guest pages, we send them | |
2179 | * within the same host page iteration. | |
2180 | */ | |
2181 | pss->host_page_start = ROUND_DOWN(pss->page, guest_pfns); | |
2182 | pss->host_page_end = ROUND_UP(pss->page + 1, guest_pfns); | |
2183 | } | |
d9e474ea PX |
2184 | } |
2185 | ||
2186 | /* | |
2187 | * Whether the page pointed by PSS is within the host page being sent. | |
2188 | * Must be called after a previous pss_host_page_prepare(). | |
2189 | */ | |
2190 | static bool pss_within_range(PageSearchStatus *pss) | |
2191 | { | |
2192 | ram_addr_t ram_addr; | |
2193 | ||
2194 | assert(pss->host_page_sending); | |
2195 | ||
2196 | /* Over host-page boundary? */ | |
2197 | if (pss->page >= pss->host_page_end) { | |
2198 | return false; | |
2199 | } | |
2200 | ||
2201 | ram_addr = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS; | |
2202 | ||
2203 | return offset_in_ramblock(pss->block, ram_addr); | |
2204 | } | |
2205 | ||
2206 | static void pss_host_page_finish(PageSearchStatus *pss) | |
2207 | { | |
2208 | pss->host_page_sending = false; | |
2209 | /* This is not needed, but just to reset it */ | |
2210 | pss->host_page_start = pss->host_page_end = 0; | |
2211 | } | |
2212 | ||
93589827 PX |
2213 | /* |
2214 | * Send an urgent host page specified by `pss'. Need to be called with | |
2215 | * bitmap_mutex held. | |
2216 | * | |
2217 | * Returns 0 if save host page succeeded, false otherwise. | |
2218 | */ | |
2219 | static int ram_save_host_page_urgent(PageSearchStatus *pss) | |
2220 | { | |
2221 | bool page_dirty, sent = false; | |
2222 | RAMState *rs = ram_state; | |
2223 | int ret = 0; | |
2224 | ||
2225 | trace_postcopy_preempt_send_host_page(pss->block->idstr, pss->page); | |
2226 | pss_host_page_prepare(pss); | |
2227 | ||
2228 | /* | |
2229 | * If precopy is sending the same page, let it be done in precopy, or | |
2230 | * we could send the same page in two channels and none of them will | |
2231 | * receive the whole page. | |
2232 | */ | |
2233 | if (pss_overlap(pss, &ram_state->pss[RAM_CHANNEL_PRECOPY])) { | |
2234 | trace_postcopy_preempt_hit(pss->block->idstr, | |
2235 | pss->page << TARGET_PAGE_BITS); | |
2236 | return 0; | |
2237 | } | |
2238 | ||
2239 | do { | |
2240 | page_dirty = migration_bitmap_clear_dirty(rs, pss->block, pss->page); | |
2241 | ||
2242 | if (page_dirty) { | |
2243 | /* Be strict to return code; it must be 1, or what else? */ | |
4010ba38 | 2244 | if (migration_ops->ram_save_target_page(rs, pss) != 1) { |
93589827 PX |
2245 | error_report_once("%s: ram_save_target_page failed", __func__); |
2246 | ret = -1; | |
2247 | goto out; | |
2248 | } | |
2249 | sent = true; | |
2250 | } | |
2251 | pss_find_next_dirty(pss); | |
2252 | } while (pss_within_range(pss)); | |
2253 | out: | |
2254 | pss_host_page_finish(pss); | |
2255 | /* For urgent requests, flush immediately if sent */ | |
2256 | if (sent) { | |
2257 | qemu_fflush(pss->pss_channel); | |
2258 | } | |
2259 | return ret; | |
2260 | } | |
2261 | ||
a82d593b | 2262 | /** |
3d0684b2 | 2263 | * ram_save_host_page: save a whole host page |
a82d593b | 2264 | * |
3d0684b2 JQ |
2265 | * Starting at *offset send pages up to the end of the current host |
2266 | * page. It's valid for the initial offset to point into the middle of | |
2267 | * a host page in which case the remainder of the hostpage is sent. | |
2268 | * Only dirty target pages are sent. Note that the host page size may | |
2269 | * be a huge page for this block. | |
f3321554 | 2270 | * |
1eb3fc0a DDAG |
2271 | * The saving stops at the boundary of the used_length of the block |
2272 | * if the RAMBlock isn't a multiple of the host page size. | |
a82d593b | 2273 | * |
f3321554 PX |
2274 | * The caller must be with ram_state.bitmap_mutex held to call this |
2275 | * function. Note that this function can temporarily release the lock, but | |
2276 | * when the function is returned it'll make sure the lock is still held. | |
2277 | * | |
3d0684b2 JQ |
2278 | * Returns the number of pages written or negative on error |
2279 | * | |
6f37bb8b | 2280 | * @rs: current RAM state |
3d0684b2 | 2281 | * @pss: data about the page we want to send |
a82d593b | 2282 | */ |
05931ec5 | 2283 | static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b | 2284 | { |
f3321554 | 2285 | bool page_dirty, preempt_active = postcopy_preempt_active(); |
a82d593b | 2286 | int tmppages, pages = 0; |
a935e30f JQ |
2287 | size_t pagesize_bits = |
2288 | qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
278e2f55 AG |
2289 | unsigned long start_page = pss->page; |
2290 | int res; | |
4c011c37 | 2291 | |
f161c88a | 2292 | if (migrate_ram_is_ignored(pss->block)) { |
b895de50 CLG |
2293 | error_report("block %s should not be migrated !", pss->block->idstr); |
2294 | return 0; | |
2295 | } | |
2296 | ||
d9e474ea PX |
2297 | /* Update host page boundary information */ |
2298 | pss_host_page_prepare(pss); | |
2299 | ||
a82d593b | 2300 | do { |
f3321554 | 2301 | page_dirty = migration_bitmap_clear_dirty(rs, pss->block, pss->page); |
a82d593b | 2302 | |
f3321554 PX |
2303 | /* Check the pages is dirty and if it is send it */ |
2304 | if (page_dirty) { | |
ba1b7c81 | 2305 | /* |
f3321554 PX |
2306 | * Properly yield the lock only in postcopy preempt mode |
2307 | * because both migration thread and rp-return thread can | |
2308 | * operate on the bitmaps. | |
ba1b7c81 | 2309 | */ |
f3321554 PX |
2310 | if (preempt_active) { |
2311 | qemu_mutex_unlock(&rs->bitmap_mutex); | |
ba1b7c81 | 2312 | } |
4010ba38 | 2313 | tmppages = migration_ops->ram_save_target_page(rs, pss); |
f3321554 PX |
2314 | if (tmppages >= 0) { |
2315 | pages += tmppages; | |
2316 | /* | |
2317 | * Allow rate limiting to happen in the middle of huge pages if | |
2318 | * something is sent in the current iteration. | |
2319 | */ | |
2320 | if (pagesize_bits > 1 && tmppages > 0) { | |
2321 | migration_rate_limit(); | |
2322 | } | |
2323 | } | |
2324 | if (preempt_active) { | |
2325 | qemu_mutex_lock(&rs->bitmap_mutex); | |
2326 | } | |
2327 | } else { | |
2328 | tmppages = 0; | |
23feba90 | 2329 | } |
f3321554 PX |
2330 | |
2331 | if (tmppages < 0) { | |
d9e474ea | 2332 | pss_host_page_finish(pss); |
f3321554 PX |
2333 | return tmppages; |
2334 | } | |
2335 | ||
d9e474ea PX |
2336 | pss_find_next_dirty(pss); |
2337 | } while (pss_within_range(pss)); | |
2338 | ||
2339 | pss_host_page_finish(pss); | |
278e2f55 AG |
2340 | |
2341 | res = ram_save_release_protection(rs, pss, start_page); | |
2342 | return (res < 0 ? res : pages); | |
a82d593b | 2343 | } |
6c595cde | 2344 | |
56e93d26 | 2345 | /** |
3d0684b2 | 2346 | * ram_find_and_save_block: finds a dirty page and sends it to f |
56e93d26 JQ |
2347 | * |
2348 | * Called within an RCU critical section. | |
2349 | * | |
e8f3735f XG |
2350 | * Returns the number of pages written where zero means no dirty pages, |
2351 | * or negative on error | |
56e93d26 | 2352 | * |
6f37bb8b | 2353 | * @rs: current RAM state |
a82d593b DDAG |
2354 | * |
2355 | * On systems where host-page-size > target-page-size it will send all the | |
2356 | * pages in a host page that are dirty. | |
56e93d26 | 2357 | */ |
05931ec5 | 2358 | static int ram_find_and_save_block(RAMState *rs) |
56e93d26 | 2359 | { |
f1668764 | 2360 | PageSearchStatus *pss = &rs->pss[RAM_CHANNEL_PRECOPY]; |
56e93d26 | 2361 | int pages = 0; |
56e93d26 | 2362 | |
0827b9e9 | 2363 | /* No dirty page as there is zero RAM */ |
8d80e195 | 2364 | if (!rs->ram_bytes_total) { |
0827b9e9 AA |
2365 | return pages; |
2366 | } | |
2367 | ||
4934a5dd PX |
2368 | /* |
2369 | * Always keep last_seen_block/last_page valid during this procedure, | |
2370 | * because find_dirty_block() relies on these values (e.g., we compare | |
2371 | * last_seen_block with pss.block to see whether we searched all the | |
2372 | * ramblocks) to detect the completion of migration. Having NULL value | |
2373 | * of last_seen_block can conditionally cause below loop to run forever. | |
2374 | */ | |
2375 | if (!rs->last_seen_block) { | |
2376 | rs->last_seen_block = QLIST_FIRST_RCU(&ram_list.blocks); | |
2377 | rs->last_page = 0; | |
2378 | } | |
2379 | ||
f1668764 | 2380 | pss_init(pss, rs->last_seen_block, rs->last_page); |
b8fb8cb7 | 2381 | |
31e2ac74 | 2382 | while (true){ |
51efd36f | 2383 | if (!get_queued_page(rs, pss)) { |
b062106d | 2384 | /* priority queue empty, so just search for something dirty */ |
31e2ac74 JQ |
2385 | int res = find_dirty_block(rs, pss); |
2386 | if (res != PAGE_DIRTY_FOUND) { | |
2387 | if (res == PAGE_ALL_CLEAN) { | |
51efd36f | 2388 | break; |
31e2ac74 JQ |
2389 | } else if (res == PAGE_TRY_AGAIN) { |
2390 | continue; | |
294e5a40 JQ |
2391 | } else if (res < 0) { |
2392 | pages = res; | |
2393 | break; | |
51efd36f JQ |
2394 | } |
2395 | } | |
56e93d26 | 2396 | } |
51efd36f | 2397 | pages = ram_save_host_page(rs, pss); |
31e2ac74 JQ |
2398 | if (pages) { |
2399 | break; | |
2400 | } | |
2401 | } | |
56e93d26 | 2402 | |
f1668764 PX |
2403 | rs->last_seen_block = pss->block; |
2404 | rs->last_page = pss->page; | |
56e93d26 JQ |
2405 | |
2406 | return pages; | |
2407 | } | |
2408 | ||
8008a272 | 2409 | static uint64_t ram_bytes_total_with_ignored(void) |
56e93d26 JQ |
2410 | { |
2411 | RAMBlock *block; | |
2412 | uint64_t total = 0; | |
2413 | ||
89ac5a1d DDAG |
2414 | RCU_READ_LOCK_GUARD(); |
2415 | ||
8008a272 JQ |
2416 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
2417 | total += block->used_length; | |
99e15582 | 2418 | } |
56e93d26 JQ |
2419 | return total; |
2420 | } | |
2421 | ||
fbd162e6 YK |
2422 | uint64_t ram_bytes_total(void) |
2423 | { | |
8008a272 JQ |
2424 | RAMBlock *block; |
2425 | uint64_t total = 0; | |
2426 | ||
2427 | RCU_READ_LOCK_GUARD(); | |
2428 | ||
2429 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
2430 | total += block->used_length; | |
2431 | } | |
2432 | return total; | |
fbd162e6 YK |
2433 | } |
2434 | ||
f265e0e4 | 2435 | static void xbzrle_load_setup(void) |
56e93d26 | 2436 | { |
f265e0e4 | 2437 | XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE); |
56e93d26 JQ |
2438 | } |
2439 | ||
f265e0e4 JQ |
2440 | static void xbzrle_load_cleanup(void) |
2441 | { | |
2442 | g_free(XBZRLE.decoded_buf); | |
2443 | XBZRLE.decoded_buf = NULL; | |
2444 | } | |
2445 | ||
7d7c96be PX |
2446 | static void ram_state_cleanup(RAMState **rsp) |
2447 | { | |
b9ccaf6d DDAG |
2448 | if (*rsp) { |
2449 | migration_page_queue_free(*rsp); | |
2450 | qemu_mutex_destroy(&(*rsp)->bitmap_mutex); | |
2451 | qemu_mutex_destroy(&(*rsp)->src_page_req_mutex); | |
2452 | g_free(*rsp); | |
2453 | *rsp = NULL; | |
2454 | } | |
7d7c96be PX |
2455 | } |
2456 | ||
84593a08 PX |
2457 | static void xbzrle_cleanup(void) |
2458 | { | |
2459 | XBZRLE_cache_lock(); | |
2460 | if (XBZRLE.cache) { | |
2461 | cache_fini(XBZRLE.cache); | |
2462 | g_free(XBZRLE.encoded_buf); | |
2463 | g_free(XBZRLE.current_buf); | |
2464 | g_free(XBZRLE.zero_target_page); | |
2465 | XBZRLE.cache = NULL; | |
2466 | XBZRLE.encoded_buf = NULL; | |
2467 | XBZRLE.current_buf = NULL; | |
2468 | XBZRLE.zero_target_page = NULL; | |
2469 | } | |
2470 | XBZRLE_cache_unlock(); | |
2471 | } | |
2472 | ||
f265e0e4 | 2473 | static void ram_save_cleanup(void *opaque) |
56e93d26 | 2474 | { |
53518d94 | 2475 | RAMState **rsp = opaque; |
6b6712ef | 2476 | RAMBlock *block; |
eb859c53 | 2477 | |
278e2f55 AG |
2478 | /* We don't use dirty log with background snapshots */ |
2479 | if (!migrate_background_snapshot()) { | |
2480 | /* caller have hold iothread lock or is in a bh, so there is | |
2481 | * no writing race against the migration bitmap | |
2482 | */ | |
63b41db4 HH |
2483 | if (global_dirty_tracking & GLOBAL_DIRTY_MIGRATION) { |
2484 | /* | |
2485 | * do not stop dirty log without starting it, since | |
2486 | * memory_global_dirty_log_stop will assert that | |
2487 | * memory_global_dirty_log_start/stop used in pairs | |
2488 | */ | |
2489 | memory_global_dirty_log_stop(GLOBAL_DIRTY_MIGRATION); | |
2490 | } | |
278e2f55 | 2491 | } |
6b6712ef | 2492 | |
fbd162e6 | 2493 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
002cad6b PX |
2494 | g_free(block->clear_bmap); |
2495 | block->clear_bmap = NULL; | |
6b6712ef JQ |
2496 | g_free(block->bmap); |
2497 | block->bmap = NULL; | |
56e93d26 JQ |
2498 | } |
2499 | ||
84593a08 | 2500 | xbzrle_cleanup(); |
f0afa331 | 2501 | compress_threads_save_cleanup(); |
7d7c96be | 2502 | ram_state_cleanup(rsp); |
4010ba38 JQ |
2503 | g_free(migration_ops); |
2504 | migration_ops = NULL; | |
56e93d26 JQ |
2505 | } |
2506 | ||
6f37bb8b | 2507 | static void ram_state_reset(RAMState *rs) |
56e93d26 | 2508 | { |
ec6f3ab9 PX |
2509 | int i; |
2510 | ||
2511 | for (i = 0; i < RAM_CHANNEL_MAX; i++) { | |
2512 | rs->pss[i].last_sent_block = NULL; | |
2513 | } | |
2514 | ||
6f37bb8b | 2515 | rs->last_seen_block = NULL; |
269ace29 | 2516 | rs->last_page = 0; |
6f37bb8b | 2517 | rs->last_version = ram_list.version; |
f3095cc8 | 2518 | rs->xbzrle_started = false; |
56e93d26 JQ |
2519 | } |
2520 | ||
2521 | #define MAX_WAIT 50 /* ms, half buffered_file limit */ | |
2522 | ||
e0b266f0 DDAG |
2523 | /* **** functions for postcopy ***** */ |
2524 | ||
ced1c616 PB |
2525 | void ram_postcopy_migrated_memory_release(MigrationState *ms) |
2526 | { | |
2527 | struct RAMBlock *block; | |
ced1c616 | 2528 | |
fbd162e6 | 2529 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
6b6712ef JQ |
2530 | unsigned long *bitmap = block->bmap; |
2531 | unsigned long range = block->used_length >> TARGET_PAGE_BITS; | |
2532 | unsigned long run_start = find_next_zero_bit(bitmap, range, 0); | |
ced1c616 PB |
2533 | |
2534 | while (run_start < range) { | |
2535 | unsigned long run_end = find_next_bit(bitmap, range, run_start + 1); | |
8bba004c AR |
2536 | ram_discard_range(block->idstr, |
2537 | ((ram_addr_t)run_start) << TARGET_PAGE_BITS, | |
2538 | ((ram_addr_t)(run_end - run_start)) | |
2539 | << TARGET_PAGE_BITS); | |
ced1c616 PB |
2540 | run_start = find_next_zero_bit(bitmap, range, run_end + 1); |
2541 | } | |
2542 | } | |
2543 | } | |
2544 | ||
3d0684b2 JQ |
2545 | /** |
2546 | * postcopy_send_discard_bm_ram: discard a RAMBlock | |
2547 | * | |
e0b266f0 | 2548 | * Callback from postcopy_each_ram_send_discard for each RAMBlock |
3d0684b2 JQ |
2549 | * |
2550 | * @ms: current migration state | |
89dab31b | 2551 | * @block: RAMBlock to discard |
e0b266f0 | 2552 | */ |
9e7d1223 | 2553 | static void postcopy_send_discard_bm_ram(MigrationState *ms, RAMBlock *block) |
e0b266f0 | 2554 | { |
6b6712ef | 2555 | unsigned long end = block->used_length >> TARGET_PAGE_BITS; |
e0b266f0 | 2556 | unsigned long current; |
1e7cf8c3 | 2557 | unsigned long *bitmap = block->bmap; |
e0b266f0 | 2558 | |
6b6712ef | 2559 | for (current = 0; current < end; ) { |
1e7cf8c3 | 2560 | unsigned long one = find_next_bit(bitmap, end, current); |
33a5cb62 | 2561 | unsigned long zero, discard_length; |
e0b266f0 | 2562 | |
33a5cb62 WY |
2563 | if (one >= end) { |
2564 | break; | |
2565 | } | |
e0b266f0 | 2566 | |
1e7cf8c3 | 2567 | zero = find_next_zero_bit(bitmap, end, one + 1); |
33a5cb62 WY |
2568 | |
2569 | if (zero >= end) { | |
2570 | discard_length = end - one; | |
e0b266f0 | 2571 | } else { |
33a5cb62 WY |
2572 | discard_length = zero - one; |
2573 | } | |
810cf2bb | 2574 | postcopy_discard_send_range(ms, one, discard_length); |
33a5cb62 | 2575 | current = one + discard_length; |
e0b266f0 | 2576 | } |
e0b266f0 DDAG |
2577 | } |
2578 | ||
f30c2e5b PX |
2579 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, RAMBlock *block); |
2580 | ||
3d0684b2 JQ |
2581 | /** |
2582 | * postcopy_each_ram_send_discard: discard all RAMBlocks | |
2583 | * | |
e0b266f0 DDAG |
2584 | * Utility for the outgoing postcopy code. |
2585 | * Calls postcopy_send_discard_bm_ram for each RAMBlock | |
2586 | * passing it bitmap indexes and name. | |
e0b266f0 DDAG |
2587 | * (qemu_ram_foreach_block ends up passing unscaled lengths |
2588 | * which would mean postcopy code would have to deal with target page) | |
3d0684b2 JQ |
2589 | * |
2590 | * @ms: current migration state | |
e0b266f0 | 2591 | */ |
739fcc1b | 2592 | static void postcopy_each_ram_send_discard(MigrationState *ms) |
e0b266f0 DDAG |
2593 | { |
2594 | struct RAMBlock *block; | |
e0b266f0 | 2595 | |
fbd162e6 | 2596 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
810cf2bb | 2597 | postcopy_discard_send_init(ms, block->idstr); |
e0b266f0 | 2598 | |
f30c2e5b PX |
2599 | /* |
2600 | * Deal with TPS != HPS and huge pages. It discard any partially sent | |
2601 | * host-page size chunks, mark any partially dirty host-page size | |
2602 | * chunks as all dirty. In this case the host-page is the host-page | |
2603 | * for the particular RAMBlock, i.e. it might be a huge page. | |
2604 | */ | |
2605 | postcopy_chunk_hostpages_pass(ms, block); | |
2606 | ||
e0b266f0 DDAG |
2607 | /* |
2608 | * Postcopy sends chunks of bitmap over the wire, but it | |
2609 | * just needs indexes at this point, avoids it having | |
2610 | * target page specific code. | |
2611 | */ | |
739fcc1b | 2612 | postcopy_send_discard_bm_ram(ms, block); |
810cf2bb | 2613 | postcopy_discard_send_finish(ms); |
e0b266f0 | 2614 | } |
e0b266f0 DDAG |
2615 | } |
2616 | ||
3d0684b2 | 2617 | /** |
8324ef86 | 2618 | * postcopy_chunk_hostpages_pass: canonicalize bitmap in hostpages |
3d0684b2 JQ |
2619 | * |
2620 | * Helper for postcopy_chunk_hostpages; it's called twice to | |
2621 | * canonicalize the two bitmaps, that are similar, but one is | |
2622 | * inverted. | |
99e314eb | 2623 | * |
3d0684b2 JQ |
2624 | * Postcopy requires that all target pages in a hostpage are dirty or |
2625 | * clean, not a mix. This function canonicalizes the bitmaps. | |
99e314eb | 2626 | * |
3d0684b2 | 2627 | * @ms: current migration state |
3d0684b2 | 2628 | * @block: block that contains the page we want to canonicalize |
99e314eb | 2629 | */ |
1e7cf8c3 | 2630 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, RAMBlock *block) |
99e314eb | 2631 | { |
53518d94 | 2632 | RAMState *rs = ram_state; |
6b6712ef | 2633 | unsigned long *bitmap = block->bmap; |
29c59172 | 2634 | unsigned int host_ratio = block->page_size / TARGET_PAGE_SIZE; |
6b6712ef | 2635 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
99e314eb DDAG |
2636 | unsigned long run_start; |
2637 | ||
29c59172 DDAG |
2638 | if (block->page_size == TARGET_PAGE_SIZE) { |
2639 | /* Easy case - TPS==HPS for a non-huge page RAMBlock */ | |
2640 | return; | |
2641 | } | |
2642 | ||
1e7cf8c3 WY |
2643 | /* Find a dirty page */ |
2644 | run_start = find_next_bit(bitmap, pages, 0); | |
99e314eb | 2645 | |
6b6712ef | 2646 | while (run_start < pages) { |
99e314eb DDAG |
2647 | |
2648 | /* | |
2649 | * If the start of this run of pages is in the middle of a host | |
2650 | * page, then we need to fixup this host page. | |
2651 | */ | |
9dec3cc3 | 2652 | if (QEMU_IS_ALIGNED(run_start, host_ratio)) { |
99e314eb | 2653 | /* Find the end of this run */ |
1e7cf8c3 | 2654 | run_start = find_next_zero_bit(bitmap, pages, run_start + 1); |
99e314eb DDAG |
2655 | /* |
2656 | * If the end isn't at the start of a host page, then the | |
2657 | * run doesn't finish at the end of a host page | |
2658 | * and we need to discard. | |
2659 | */ | |
99e314eb DDAG |
2660 | } |
2661 | ||
9dec3cc3 | 2662 | if (!QEMU_IS_ALIGNED(run_start, host_ratio)) { |
99e314eb | 2663 | unsigned long page; |
dad45ab2 WY |
2664 | unsigned long fixup_start_addr = QEMU_ALIGN_DOWN(run_start, |
2665 | host_ratio); | |
2666 | run_start = QEMU_ALIGN_UP(run_start, host_ratio); | |
99e314eb | 2667 | |
99e314eb DDAG |
2668 | /* Clean up the bitmap */ |
2669 | for (page = fixup_start_addr; | |
2670 | page < fixup_start_addr + host_ratio; page++) { | |
99e314eb DDAG |
2671 | /* |
2672 | * Remark them as dirty, updating the count for any pages | |
2673 | * that weren't previously dirty. | |
2674 | */ | |
0d8ec885 | 2675 | rs->migration_dirty_pages += !test_and_set_bit(page, bitmap); |
99e314eb DDAG |
2676 | } |
2677 | } | |
2678 | ||
1e7cf8c3 WY |
2679 | /* Find the next dirty page for the next iteration */ |
2680 | run_start = find_next_bit(bitmap, pages, run_start); | |
99e314eb DDAG |
2681 | } |
2682 | } | |
2683 | ||
3d0684b2 JQ |
2684 | /** |
2685 | * ram_postcopy_send_discard_bitmap: transmit the discard bitmap | |
2686 | * | |
e0b266f0 DDAG |
2687 | * Transmit the set of pages to be discarded after precopy to the target |
2688 | * these are pages that: | |
2689 | * a) Have been previously transmitted but are now dirty again | |
2690 | * b) Pages that have never been transmitted, this ensures that | |
2691 | * any pages on the destination that have been mapped by background | |
2692 | * tasks get discarded (transparent huge pages is the specific concern) | |
2693 | * Hopefully this is pretty sparse | |
3d0684b2 JQ |
2694 | * |
2695 | * @ms: current migration state | |
e0b266f0 | 2696 | */ |
739fcc1b | 2697 | void ram_postcopy_send_discard_bitmap(MigrationState *ms) |
e0b266f0 | 2698 | { |
53518d94 | 2699 | RAMState *rs = ram_state; |
e0b266f0 | 2700 | |
89ac5a1d | 2701 | RCU_READ_LOCK_GUARD(); |
e0b266f0 DDAG |
2702 | |
2703 | /* This should be our last sync, the src is now paused */ | |
1e493be5 | 2704 | migration_bitmap_sync(rs, false); |
e0b266f0 | 2705 | |
6b6712ef | 2706 | /* Easiest way to make sure we don't resume in the middle of a host-page */ |
ec6f3ab9 | 2707 | rs->pss[RAM_CHANNEL_PRECOPY].last_sent_block = NULL; |
6b6712ef | 2708 | rs->last_seen_block = NULL; |
6b6712ef | 2709 | rs->last_page = 0; |
e0b266f0 | 2710 | |
739fcc1b | 2711 | postcopy_each_ram_send_discard(ms); |
e0b266f0 | 2712 | |
739fcc1b | 2713 | trace_ram_postcopy_send_discard_bitmap(); |
e0b266f0 DDAG |
2714 | } |
2715 | ||
3d0684b2 JQ |
2716 | /** |
2717 | * ram_discard_range: discard dirtied pages at the beginning of postcopy | |
e0b266f0 | 2718 | * |
3d0684b2 | 2719 | * Returns zero on success |
e0b266f0 | 2720 | * |
36449157 JQ |
2721 | * @rbname: name of the RAMBlock of the request. NULL means the |
2722 | * same that last one. | |
3d0684b2 JQ |
2723 | * @start: RAMBlock starting page |
2724 | * @length: RAMBlock size | |
e0b266f0 | 2725 | */ |
aaa2064c | 2726 | int ram_discard_range(const char *rbname, uint64_t start, size_t length) |
e0b266f0 | 2727 | { |
36449157 | 2728 | trace_ram_discard_range(rbname, start, length); |
d3a5038c | 2729 | |
89ac5a1d | 2730 | RCU_READ_LOCK_GUARD(); |
36449157 | 2731 | RAMBlock *rb = qemu_ram_block_by_name(rbname); |
e0b266f0 DDAG |
2732 | |
2733 | if (!rb) { | |
36449157 | 2734 | error_report("ram_discard_range: Failed to find block '%s'", rbname); |
03acb4e9 | 2735 | return -1; |
e0b266f0 DDAG |
2736 | } |
2737 | ||
814bb08f PX |
2738 | /* |
2739 | * On source VM, we don't need to update the received bitmap since | |
2740 | * we don't even have one. | |
2741 | */ | |
2742 | if (rb->receivedmap) { | |
2743 | bitmap_clear(rb->receivedmap, start >> qemu_target_page_bits(), | |
2744 | length >> qemu_target_page_bits()); | |
2745 | } | |
2746 | ||
03acb4e9 | 2747 | return ram_block_discard_range(rb, start, length); |
e0b266f0 DDAG |
2748 | } |
2749 | ||
84593a08 PX |
2750 | /* |
2751 | * For every allocation, we will try not to crash the VM if the | |
2752 | * allocation failed. | |
2753 | */ | |
2754 | static int xbzrle_init(void) | |
2755 | { | |
2756 | Error *local_err = NULL; | |
2757 | ||
87dca0c9 | 2758 | if (!migrate_xbzrle()) { |
84593a08 PX |
2759 | return 0; |
2760 | } | |
2761 | ||
2762 | XBZRLE_cache_lock(); | |
2763 | ||
2764 | XBZRLE.zero_target_page = g_try_malloc0(TARGET_PAGE_SIZE); | |
2765 | if (!XBZRLE.zero_target_page) { | |
2766 | error_report("%s: Error allocating zero page", __func__); | |
2767 | goto err_out; | |
2768 | } | |
2769 | ||
2770 | XBZRLE.cache = cache_init(migrate_xbzrle_cache_size(), | |
2771 | TARGET_PAGE_SIZE, &local_err); | |
2772 | if (!XBZRLE.cache) { | |
2773 | error_report_err(local_err); | |
2774 | goto free_zero_page; | |
2775 | } | |
2776 | ||
2777 | XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); | |
2778 | if (!XBZRLE.encoded_buf) { | |
2779 | error_report("%s: Error allocating encoded_buf", __func__); | |
2780 | goto free_cache; | |
2781 | } | |
2782 | ||
2783 | XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); | |
2784 | if (!XBZRLE.current_buf) { | |
2785 | error_report("%s: Error allocating current_buf", __func__); | |
2786 | goto free_encoded_buf; | |
2787 | } | |
2788 | ||
2789 | /* We are all good */ | |
2790 | XBZRLE_cache_unlock(); | |
2791 | return 0; | |
2792 | ||
2793 | free_encoded_buf: | |
2794 | g_free(XBZRLE.encoded_buf); | |
2795 | XBZRLE.encoded_buf = NULL; | |
2796 | free_cache: | |
2797 | cache_fini(XBZRLE.cache); | |
2798 | XBZRLE.cache = NULL; | |
2799 | free_zero_page: | |
2800 | g_free(XBZRLE.zero_target_page); | |
2801 | XBZRLE.zero_target_page = NULL; | |
2802 | err_out: | |
2803 | XBZRLE_cache_unlock(); | |
2804 | return -ENOMEM; | |
2805 | } | |
2806 | ||
53518d94 | 2807 | static int ram_state_init(RAMState **rsp) |
56e93d26 | 2808 | { |
7d00ee6a PX |
2809 | *rsp = g_try_new0(RAMState, 1); |
2810 | ||
2811 | if (!*rsp) { | |
2812 | error_report("%s: Init ramstate fail", __func__); | |
2813 | return -1; | |
2814 | } | |
53518d94 JQ |
2815 | |
2816 | qemu_mutex_init(&(*rsp)->bitmap_mutex); | |
2817 | qemu_mutex_init(&(*rsp)->src_page_req_mutex); | |
2818 | QSIMPLEQ_INIT(&(*rsp)->src_page_requests); | |
8d80e195 | 2819 | (*rsp)->ram_bytes_total = ram_bytes_total(); |
56e93d26 | 2820 | |
7d00ee6a | 2821 | /* |
40c4d4a8 IR |
2822 | * Count the total number of pages used by ram blocks not including any |
2823 | * gaps due to alignment or unplugs. | |
03158519 | 2824 | * This must match with the initial values of dirty bitmap. |
7d00ee6a | 2825 | */ |
8d80e195 | 2826 | (*rsp)->migration_dirty_pages = (*rsp)->ram_bytes_total >> TARGET_PAGE_BITS; |
7d00ee6a PX |
2827 | ram_state_reset(*rsp); |
2828 | ||
2829 | return 0; | |
2830 | } | |
2831 | ||
d6eff5d7 | 2832 | static void ram_list_init_bitmaps(void) |
7d00ee6a | 2833 | { |
002cad6b | 2834 | MigrationState *ms = migrate_get_current(); |
d6eff5d7 PX |
2835 | RAMBlock *block; |
2836 | unsigned long pages; | |
002cad6b | 2837 | uint8_t shift; |
56e93d26 | 2838 | |
0827b9e9 AA |
2839 | /* Skip setting bitmap if there is no RAM */ |
2840 | if (ram_bytes_total()) { | |
002cad6b PX |
2841 | shift = ms->clear_bitmap_shift; |
2842 | if (shift > CLEAR_BITMAP_SHIFT_MAX) { | |
2843 | error_report("clear_bitmap_shift (%u) too big, using " | |
2844 | "max value (%u)", shift, CLEAR_BITMAP_SHIFT_MAX); | |
2845 | shift = CLEAR_BITMAP_SHIFT_MAX; | |
2846 | } else if (shift < CLEAR_BITMAP_SHIFT_MIN) { | |
2847 | error_report("clear_bitmap_shift (%u) too small, using " | |
2848 | "min value (%u)", shift, CLEAR_BITMAP_SHIFT_MIN); | |
2849 | shift = CLEAR_BITMAP_SHIFT_MIN; | |
2850 | } | |
2851 | ||
fbd162e6 | 2852 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
d6eff5d7 | 2853 | pages = block->max_length >> TARGET_PAGE_BITS; |
03158519 WY |
2854 | /* |
2855 | * The initial dirty bitmap for migration must be set with all | |
2856 | * ones to make sure we'll migrate every guest RAM page to | |
2857 | * destination. | |
40c4d4a8 IR |
2858 | * Here we set RAMBlock.bmap all to 1 because when rebegin a |
2859 | * new migration after a failed migration, ram_list. | |
2860 | * dirty_memory[DIRTY_MEMORY_MIGRATION] don't include the whole | |
2861 | * guest memory. | |
03158519 | 2862 | */ |
6b6712ef | 2863 | block->bmap = bitmap_new(pages); |
40c4d4a8 | 2864 | bitmap_set(block->bmap, 0, pages); |
002cad6b PX |
2865 | block->clear_bmap_shift = shift; |
2866 | block->clear_bmap = bitmap_new(clear_bmap_size(pages, shift)); | |
0827b9e9 | 2867 | } |
f3f491fc | 2868 | } |
d6eff5d7 PX |
2869 | } |
2870 | ||
be39b4cd DH |
2871 | static void migration_bitmap_clear_discarded_pages(RAMState *rs) |
2872 | { | |
2873 | unsigned long pages; | |
2874 | RAMBlock *rb; | |
2875 | ||
2876 | RCU_READ_LOCK_GUARD(); | |
2877 | ||
2878 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { | |
2879 | pages = ramblock_dirty_bitmap_clear_discarded_pages(rb); | |
2880 | rs->migration_dirty_pages -= pages; | |
2881 | } | |
2882 | } | |
2883 | ||
d6eff5d7 PX |
2884 | static void ram_init_bitmaps(RAMState *rs) |
2885 | { | |
2886 | /* For memory_global_dirty_log_start below. */ | |
2887 | qemu_mutex_lock_iothread(); | |
2888 | qemu_mutex_lock_ramlist(); | |
f3f491fc | 2889 | |
89ac5a1d DDAG |
2890 | WITH_RCU_READ_LOCK_GUARD() { |
2891 | ram_list_init_bitmaps(); | |
278e2f55 AG |
2892 | /* We don't use dirty log with background snapshots */ |
2893 | if (!migrate_background_snapshot()) { | |
63b41db4 | 2894 | memory_global_dirty_log_start(GLOBAL_DIRTY_MIGRATION); |
1e493be5 | 2895 | migration_bitmap_sync_precopy(rs, false); |
278e2f55 | 2896 | } |
89ac5a1d | 2897 | } |
56e93d26 | 2898 | qemu_mutex_unlock_ramlist(); |
49877834 | 2899 | qemu_mutex_unlock_iothread(); |
be39b4cd DH |
2900 | |
2901 | /* | |
2902 | * After an eventual first bitmap sync, fixup the initial bitmap | |
2903 | * containing all 1s to exclude any discarded pages from migration. | |
2904 | */ | |
2905 | migration_bitmap_clear_discarded_pages(rs); | |
d6eff5d7 PX |
2906 | } |
2907 | ||
2908 | static int ram_init_all(RAMState **rsp) | |
2909 | { | |
2910 | if (ram_state_init(rsp)) { | |
2911 | return -1; | |
2912 | } | |
2913 | ||
2914 | if (xbzrle_init()) { | |
2915 | ram_state_cleanup(rsp); | |
2916 | return -1; | |
2917 | } | |
2918 | ||
2919 | ram_init_bitmaps(*rsp); | |
a91246c9 HZ |
2920 | |
2921 | return 0; | |
2922 | } | |
2923 | ||
08614f34 PX |
2924 | static void ram_state_resume_prepare(RAMState *rs, QEMUFile *out) |
2925 | { | |
2926 | RAMBlock *block; | |
2927 | uint64_t pages = 0; | |
2928 | ||
2929 | /* | |
2930 | * Postcopy is not using xbzrle/compression, so no need for that. | |
2931 | * Also, since source are already halted, we don't need to care | |
2932 | * about dirty page logging as well. | |
2933 | */ | |
2934 | ||
fbd162e6 | 2935 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
08614f34 PX |
2936 | pages += bitmap_count_one(block->bmap, |
2937 | block->used_length >> TARGET_PAGE_BITS); | |
2938 | } | |
2939 | ||
2940 | /* This may not be aligned with current bitmaps. Recalculate. */ | |
2941 | rs->migration_dirty_pages = pages; | |
2942 | ||
1a373522 | 2943 | ram_state_reset(rs); |
08614f34 PX |
2944 | |
2945 | /* Update RAMState cache of output QEMUFile */ | |
7f401b80 | 2946 | rs->pss[RAM_CHANNEL_PRECOPY].pss_channel = out; |
08614f34 PX |
2947 | |
2948 | trace_ram_state_resume_prepare(pages); | |
2949 | } | |
2950 | ||
6bcb05fc WW |
2951 | /* |
2952 | * This function clears bits of the free pages reported by the caller from the | |
2953 | * migration dirty bitmap. @addr is the host address corresponding to the | |
2954 | * start of the continuous guest free pages, and @len is the total bytes of | |
2955 | * those pages. | |
2956 | */ | |
2957 | void qemu_guest_free_page_hint(void *addr, size_t len) | |
2958 | { | |
2959 | RAMBlock *block; | |
2960 | ram_addr_t offset; | |
2961 | size_t used_len, start, npages; | |
2962 | MigrationState *s = migrate_get_current(); | |
2963 | ||
2964 | /* This function is currently expected to be used during live migration */ | |
2965 | if (!migration_is_setup_or_active(s->state)) { | |
2966 | return; | |
2967 | } | |
2968 | ||
2969 | for (; len > 0; len -= used_len, addr += used_len) { | |
2970 | block = qemu_ram_block_from_host(addr, false, &offset); | |
2971 | if (unlikely(!block || offset >= block->used_length)) { | |
2972 | /* | |
2973 | * The implementation might not support RAMBlock resize during | |
2974 | * live migration, but it could happen in theory with future | |
2975 | * updates. So we add a check here to capture that case. | |
2976 | */ | |
2977 | error_report_once("%s unexpected error", __func__); | |
2978 | return; | |
2979 | } | |
2980 | ||
2981 | if (len <= block->used_length - offset) { | |
2982 | used_len = len; | |
2983 | } else { | |
2984 | used_len = block->used_length - offset; | |
2985 | } | |
2986 | ||
2987 | start = offset >> TARGET_PAGE_BITS; | |
2988 | npages = used_len >> TARGET_PAGE_BITS; | |
2989 | ||
2990 | qemu_mutex_lock(&ram_state->bitmap_mutex); | |
3143577d WW |
2991 | /* |
2992 | * The skipped free pages are equavalent to be sent from clear_bmap's | |
2993 | * perspective, so clear the bits from the memory region bitmap which | |
2994 | * are initially set. Otherwise those skipped pages will be sent in | |
2995 | * the next round after syncing from the memory region bitmap. | |
2996 | */ | |
1230a25f | 2997 | migration_clear_memory_region_dirty_bitmap_range(block, start, npages); |
6bcb05fc WW |
2998 | ram_state->migration_dirty_pages -= |
2999 | bitmap_count_one_with_offset(block->bmap, start, npages); | |
3000 | bitmap_clear(block->bmap, start, npages); | |
3001 | qemu_mutex_unlock(&ram_state->bitmap_mutex); | |
3002 | } | |
3003 | } | |
3004 | ||
3d0684b2 JQ |
3005 | /* |
3006 | * Each of ram_save_setup, ram_save_iterate and ram_save_complete has | |
a91246c9 HZ |
3007 | * long-running RCU critical section. When rcu-reclaims in the code |
3008 | * start to become numerous it will be necessary to reduce the | |
3009 | * granularity of these critical sections. | |
3010 | */ | |
3011 | ||
3d0684b2 JQ |
3012 | /** |
3013 | * ram_save_setup: Setup RAM for migration | |
3014 | * | |
3015 | * Returns zero to indicate success and negative for error | |
3016 | * | |
3017 | * @f: QEMUFile where to send the data | |
3018 | * @opaque: RAMState pointer | |
3019 | */ | |
a91246c9 HZ |
3020 | static int ram_save_setup(QEMUFile *f, void *opaque) |
3021 | { | |
53518d94 | 3022 | RAMState **rsp = opaque; |
a91246c9 | 3023 | RAMBlock *block; |
33d70973 | 3024 | int ret; |
a91246c9 | 3025 | |
dcaf446e XG |
3026 | if (compress_threads_save_setup()) { |
3027 | return -1; | |
3028 | } | |
3029 | ||
a91246c9 HZ |
3030 | /* migration has already setup the bitmap, reuse it. */ |
3031 | if (!migration_in_colo_state()) { | |
7d00ee6a | 3032 | if (ram_init_all(rsp) != 0) { |
dcaf446e | 3033 | compress_threads_save_cleanup(); |
a91246c9 | 3034 | return -1; |
53518d94 | 3035 | } |
a91246c9 | 3036 | } |
7f401b80 | 3037 | (*rsp)->pss[RAM_CHANNEL_PRECOPY].pss_channel = f; |
a91246c9 | 3038 | |
0e6ebd48 | 3039 | WITH_RCU_READ_LOCK_GUARD() { |
8008a272 JQ |
3040 | qemu_put_be64(f, ram_bytes_total_with_ignored() |
3041 | | RAM_SAVE_FLAG_MEM_SIZE); | |
56e93d26 | 3042 | |
0e6ebd48 DDAG |
3043 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
3044 | qemu_put_byte(f, strlen(block->idstr)); | |
3045 | qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); | |
3046 | qemu_put_be64(f, block->used_length); | |
3047 | if (migrate_postcopy_ram() && block->page_size != | |
3048 | qemu_host_page_size) { | |
3049 | qemu_put_be64(f, block->page_size); | |
3050 | } | |
3051 | if (migrate_ignore_shared()) { | |
3052 | qemu_put_be64(f, block->mr->addr); | |
3053 | } | |
fbd162e6 | 3054 | } |
56e93d26 JQ |
3055 | } |
3056 | ||
56e93d26 JQ |
3057 | ram_control_before_iterate(f, RAM_CONTROL_SETUP); |
3058 | ram_control_after_iterate(f, RAM_CONTROL_SETUP); | |
3059 | ||
4010ba38 JQ |
3060 | migration_ops = g_malloc0(sizeof(MigrationOps)); |
3061 | migration_ops->ram_save_target_page = ram_save_target_page_legacy; | |
8ebb6ecc | 3062 | ret = multifd_send_sync_main(f); |
33d70973 LB |
3063 | if (ret < 0) { |
3064 | return ret; | |
3065 | } | |
3066 | ||
294e5a40 JQ |
3067 | if (!migrate_multifd_flush_after_each_section()) { |
3068 | qemu_put_be64(f, RAM_SAVE_FLAG_MULTIFD_FLUSH); | |
3069 | } | |
3070 | ||
56e93d26 | 3071 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
35374cbd | 3072 | qemu_fflush(f); |
56e93d26 JQ |
3073 | |
3074 | return 0; | |
3075 | } | |
3076 | ||
3d0684b2 JQ |
3077 | /** |
3078 | * ram_save_iterate: iterative stage for migration | |
3079 | * | |
3080 | * Returns zero to indicate success and negative for error | |
3081 | * | |
3082 | * @f: QEMUFile where to send the data | |
3083 | * @opaque: RAMState pointer | |
3084 | */ | |
56e93d26 JQ |
3085 | static int ram_save_iterate(QEMUFile *f, void *opaque) |
3086 | { | |
53518d94 JQ |
3087 | RAMState **temp = opaque; |
3088 | RAMState *rs = *temp; | |
3d4095b2 | 3089 | int ret = 0; |
56e93d26 JQ |
3090 | int i; |
3091 | int64_t t0; | |
5c90308f | 3092 | int done = 0; |
56e93d26 | 3093 | |
b2557345 PL |
3094 | if (blk_mig_bulk_active()) { |
3095 | /* Avoid transferring ram during bulk phase of block migration as | |
3096 | * the bulk phase will usually take a long time and transferring | |
3097 | * ram updates during that time is pointless. */ | |
3098 | goto out; | |
3099 | } | |
3100 | ||
63268c49 PX |
3101 | /* |
3102 | * We'll take this lock a little bit long, but it's okay for two reasons. | |
3103 | * Firstly, the only possible other thread to take it is who calls | |
3104 | * qemu_guest_free_page_hint(), which should be rare; secondly, see | |
3105 | * MAX_WAIT (if curious, further see commit 4508bd9ed8053ce) below, which | |
3106 | * guarantees that we'll at least released it in a regular basis. | |
3107 | */ | |
3108 | qemu_mutex_lock(&rs->bitmap_mutex); | |
89ac5a1d DDAG |
3109 | WITH_RCU_READ_LOCK_GUARD() { |
3110 | if (ram_list.version != rs->last_version) { | |
3111 | ram_state_reset(rs); | |
3112 | } | |
56e93d26 | 3113 | |
89ac5a1d DDAG |
3114 | /* Read version before ram_list.blocks */ |
3115 | smp_rmb(); | |
56e93d26 | 3116 | |
89ac5a1d | 3117 | ram_control_before_iterate(f, RAM_CONTROL_ROUND); |
56e93d26 | 3118 | |
89ac5a1d DDAG |
3119 | t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
3120 | i = 0; | |
e1fde0e0 | 3121 | while ((ret = migration_rate_exceeded(f)) == 0 || |
a1fe28df | 3122 | postcopy_has_request(rs)) { |
89ac5a1d | 3123 | int pages; |
e03a34f8 | 3124 | |
89ac5a1d DDAG |
3125 | if (qemu_file_get_error(f)) { |
3126 | break; | |
3127 | } | |
e8f3735f | 3128 | |
05931ec5 | 3129 | pages = ram_find_and_save_block(rs); |
89ac5a1d DDAG |
3130 | /* no more pages to sent */ |
3131 | if (pages == 0) { | |
3132 | done = 1; | |
3133 | break; | |
3134 | } | |
e8f3735f | 3135 | |
89ac5a1d DDAG |
3136 | if (pages < 0) { |
3137 | qemu_file_set_error(f, pages); | |
56e93d26 JQ |
3138 | break; |
3139 | } | |
89ac5a1d DDAG |
3140 | |
3141 | rs->target_page_count += pages; | |
3142 | ||
644acf99 WY |
3143 | /* |
3144 | * During postcopy, it is necessary to make sure one whole host | |
3145 | * page is sent in one chunk. | |
3146 | */ | |
3147 | if (migrate_postcopy_ram()) { | |
ef4f5f5d | 3148 | ram_flush_compressed_data(rs); |
644acf99 WY |
3149 | } |
3150 | ||
89ac5a1d DDAG |
3151 | /* |
3152 | * we want to check in the 1st loop, just in case it was the 1st | |
3153 | * time and we had to sync the dirty bitmap. | |
3154 | * qemu_clock_get_ns() is a bit expensive, so we only check each | |
3155 | * some iterations | |
3156 | */ | |
3157 | if ((i & 63) == 0) { | |
3158 | uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / | |
3159 | 1000000; | |
3160 | if (t1 > MAX_WAIT) { | |
3161 | trace_ram_save_iterate_big_wait(t1, i); | |
3162 | break; | |
3163 | } | |
3164 | } | |
3165 | i++; | |
56e93d26 | 3166 | } |
56e93d26 | 3167 | } |
63268c49 | 3168 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 JQ |
3169 | |
3170 | /* | |
3171 | * Must occur before EOS (or any QEMUFile operation) | |
3172 | * because of RDMA protocol. | |
3173 | */ | |
3174 | ram_control_after_iterate(f, RAM_CONTROL_ROUND); | |
3175 | ||
b2557345 | 3176 | out: |
b69a0227 JQ |
3177 | if (ret >= 0 |
3178 | && migration_is_setup_or_active(migrate_get_current()->state)) { | |
b05292c2 JQ |
3179 | if (migrate_multifd_flush_after_each_section()) { |
3180 | ret = multifd_send_sync_main(rs->pss[RAM_CHANNEL_PRECOPY].pss_channel); | |
3181 | if (ret < 0) { | |
3182 | return ret; | |
3183 | } | |
33d70973 LB |
3184 | } |
3185 | ||
3d4095b2 JQ |
3186 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
3187 | qemu_fflush(f); | |
4c2d0f6d | 3188 | ram_transferred_add(8); |
56e93d26 | 3189 | |
3d4095b2 JQ |
3190 | ret = qemu_file_get_error(f); |
3191 | } | |
56e93d26 JQ |
3192 | if (ret < 0) { |
3193 | return ret; | |
3194 | } | |
3195 | ||
5c90308f | 3196 | return done; |
56e93d26 JQ |
3197 | } |
3198 | ||
3d0684b2 JQ |
3199 | /** |
3200 | * ram_save_complete: function called to send the remaining amount of ram | |
3201 | * | |
e8f3735f | 3202 | * Returns zero to indicate success or negative on error |
3d0684b2 JQ |
3203 | * |
3204 | * Called with iothread lock | |
3205 | * | |
3206 | * @f: QEMUFile where to send the data | |
3207 | * @opaque: RAMState pointer | |
3208 | */ | |
56e93d26 JQ |
3209 | static int ram_save_complete(QEMUFile *f, void *opaque) |
3210 | { | |
53518d94 JQ |
3211 | RAMState **temp = opaque; |
3212 | RAMState *rs = *temp; | |
e8f3735f | 3213 | int ret = 0; |
6f37bb8b | 3214 | |
05931ec5 JQ |
3215 | rs->last_stage = !migration_in_colo_state(); |
3216 | ||
89ac5a1d DDAG |
3217 | WITH_RCU_READ_LOCK_GUARD() { |
3218 | if (!migration_in_postcopy()) { | |
1e493be5 | 3219 | migration_bitmap_sync_precopy(rs, true); |
89ac5a1d | 3220 | } |
56e93d26 | 3221 | |
89ac5a1d | 3222 | ram_control_before_iterate(f, RAM_CONTROL_FINISH); |
56e93d26 | 3223 | |
89ac5a1d | 3224 | /* try transferring iterative blocks of memory */ |
56e93d26 | 3225 | |
89ac5a1d | 3226 | /* flush all remaining blocks regardless of rate limiting */ |
c13221b5 | 3227 | qemu_mutex_lock(&rs->bitmap_mutex); |
89ac5a1d DDAG |
3228 | while (true) { |
3229 | int pages; | |
56e93d26 | 3230 | |
05931ec5 | 3231 | pages = ram_find_and_save_block(rs); |
89ac5a1d DDAG |
3232 | /* no more blocks to sent */ |
3233 | if (pages == 0) { | |
3234 | break; | |
3235 | } | |
3236 | if (pages < 0) { | |
3237 | ret = pages; | |
3238 | break; | |
3239 | } | |
e8f3735f | 3240 | } |
c13221b5 | 3241 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 3242 | |
ef4f5f5d | 3243 | ram_flush_compressed_data(rs); |
89ac5a1d DDAG |
3244 | ram_control_after_iterate(f, RAM_CONTROL_FINISH); |
3245 | } | |
d09a6fde | 3246 | |
33d70973 LB |
3247 | if (ret < 0) { |
3248 | return ret; | |
3d4095b2 | 3249 | } |
56e93d26 | 3250 | |
7f401b80 | 3251 | ret = multifd_send_sync_main(rs->pss[RAM_CHANNEL_PRECOPY].pss_channel); |
33d70973 LB |
3252 | if (ret < 0) { |
3253 | return ret; | |
3254 | } | |
3255 | ||
294e5a40 JQ |
3256 | if (!migrate_multifd_flush_after_each_section()) { |
3257 | qemu_put_be64(f, RAM_SAVE_FLAG_MULTIFD_FLUSH); | |
3258 | } | |
33d70973 LB |
3259 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
3260 | qemu_fflush(f); | |
3261 | ||
3262 | return 0; | |
56e93d26 JQ |
3263 | } |
3264 | ||
24beea4e JQ |
3265 | static void ram_state_pending_estimate(void *opaque, uint64_t *must_precopy, |
3266 | uint64_t *can_postcopy) | |
56e93d26 | 3267 | { |
53518d94 JQ |
3268 | RAMState **temp = opaque; |
3269 | RAMState *rs = *temp; | |
56e93d26 | 3270 | |
c8df4a7a | 3271 | uint64_t remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 3272 | |
c8df4a7a JQ |
3273 | if (migrate_postcopy_ram()) { |
3274 | /* We can do postcopy, and all the data is postcopiable */ | |
24beea4e | 3275 | *can_postcopy += remaining_size; |
c8df4a7a | 3276 | } else { |
24beea4e | 3277 | *must_precopy += remaining_size; |
c8df4a7a JQ |
3278 | } |
3279 | } | |
3280 | ||
24beea4e JQ |
3281 | static void ram_state_pending_exact(void *opaque, uint64_t *must_precopy, |
3282 | uint64_t *can_postcopy) | |
c8df4a7a | 3283 | { |
28ef5339 | 3284 | MigrationState *s = migrate_get_current(); |
c8df4a7a JQ |
3285 | RAMState **temp = opaque; |
3286 | RAMState *rs = *temp; | |
3287 | ||
3288 | uint64_t remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; | |
3289 | ||
28ef5339 | 3290 | if (!migration_in_postcopy() && remaining_size < s->threshold_size) { |
56e93d26 | 3291 | qemu_mutex_lock_iothread(); |
89ac5a1d | 3292 | WITH_RCU_READ_LOCK_GUARD() { |
1e493be5 | 3293 | migration_bitmap_sync_precopy(rs, false); |
89ac5a1d | 3294 | } |
56e93d26 | 3295 | qemu_mutex_unlock_iothread(); |
9edabd4d | 3296 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 3297 | } |
c31b098f | 3298 | |
86e1167e VSO |
3299 | if (migrate_postcopy_ram()) { |
3300 | /* We can do postcopy, and all the data is postcopiable */ | |
24beea4e | 3301 | *can_postcopy += remaining_size; |
86e1167e | 3302 | } else { |
24beea4e | 3303 | *must_precopy += remaining_size; |
86e1167e | 3304 | } |
56e93d26 JQ |
3305 | } |
3306 | ||
3307 | static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) | |
3308 | { | |
3309 | unsigned int xh_len; | |
3310 | int xh_flags; | |
063e760a | 3311 | uint8_t *loaded_data; |
56e93d26 | 3312 | |
56e93d26 JQ |
3313 | /* extract RLE header */ |
3314 | xh_flags = qemu_get_byte(f); | |
3315 | xh_len = qemu_get_be16(f); | |
3316 | ||
3317 | if (xh_flags != ENCODING_FLAG_XBZRLE) { | |
3318 | error_report("Failed to load XBZRLE page - wrong compression!"); | |
3319 | return -1; | |
3320 | } | |
3321 | ||
3322 | if (xh_len > TARGET_PAGE_SIZE) { | |
3323 | error_report("Failed to load XBZRLE page - len overflow!"); | |
3324 | return -1; | |
3325 | } | |
f265e0e4 | 3326 | loaded_data = XBZRLE.decoded_buf; |
56e93d26 | 3327 | /* load data and decode */ |
f265e0e4 | 3328 | /* it can change loaded_data to point to an internal buffer */ |
063e760a | 3329 | qemu_get_buffer_in_place(f, &loaded_data, xh_len); |
56e93d26 JQ |
3330 | |
3331 | /* decode RLE */ | |
063e760a | 3332 | if (xbzrle_decode_buffer(loaded_data, xh_len, host, |
56e93d26 JQ |
3333 | TARGET_PAGE_SIZE) == -1) { |
3334 | error_report("Failed to load XBZRLE page - decode error!"); | |
3335 | return -1; | |
3336 | } | |
3337 | ||
3338 | return 0; | |
3339 | } | |
3340 | ||
3d0684b2 JQ |
3341 | /** |
3342 | * ram_block_from_stream: read a RAMBlock id from the migration stream | |
3343 | * | |
3344 | * Must be called from within a rcu critical section. | |
3345 | * | |
56e93d26 | 3346 | * Returns a pointer from within the RCU-protected ram_list. |
a7180877 | 3347 | * |
755e8d7c | 3348 | * @mis: the migration incoming state pointer |
3d0684b2 JQ |
3349 | * @f: QEMUFile where to read the data from |
3350 | * @flags: Page flags (mostly to see if it's a continuation of previous block) | |
c01b16ed | 3351 | * @channel: the channel we're using |
a7180877 | 3352 | */ |
755e8d7c | 3353 | static inline RAMBlock *ram_block_from_stream(MigrationIncomingState *mis, |
c01b16ed PX |
3354 | QEMUFile *f, int flags, |
3355 | int channel) | |
56e93d26 | 3356 | { |
c01b16ed | 3357 | RAMBlock *block = mis->last_recv_block[channel]; |
56e93d26 JQ |
3358 | char id[256]; |
3359 | uint8_t len; | |
3360 | ||
3361 | if (flags & RAM_SAVE_FLAG_CONTINUE) { | |
4c4bad48 | 3362 | if (!block) { |
56e93d26 JQ |
3363 | error_report("Ack, bad migration stream!"); |
3364 | return NULL; | |
3365 | } | |
4c4bad48 | 3366 | return block; |
56e93d26 JQ |
3367 | } |
3368 | ||
3369 | len = qemu_get_byte(f); | |
3370 | qemu_get_buffer(f, (uint8_t *)id, len); | |
3371 | id[len] = 0; | |
3372 | ||
e3dd7493 | 3373 | block = qemu_ram_block_by_name(id); |
4c4bad48 HZ |
3374 | if (!block) { |
3375 | error_report("Can't find block %s", id); | |
3376 | return NULL; | |
56e93d26 JQ |
3377 | } |
3378 | ||
f161c88a | 3379 | if (migrate_ram_is_ignored(block)) { |
b895de50 CLG |
3380 | error_report("block %s should not be migrated !", id); |
3381 | return NULL; | |
3382 | } | |
3383 | ||
c01b16ed | 3384 | mis->last_recv_block[channel] = block; |
755e8d7c | 3385 | |
4c4bad48 HZ |
3386 | return block; |
3387 | } | |
3388 | ||
3389 | static inline void *host_from_ram_block_offset(RAMBlock *block, | |
3390 | ram_addr_t offset) | |
3391 | { | |
3392 | if (!offset_in_ramblock(block, offset)) { | |
3393 | return NULL; | |
3394 | } | |
3395 | ||
3396 | return block->host + offset; | |
56e93d26 JQ |
3397 | } |
3398 | ||
6a23f639 DH |
3399 | static void *host_page_from_ram_block_offset(RAMBlock *block, |
3400 | ram_addr_t offset) | |
3401 | { | |
3402 | /* Note: Explicitly no check against offset_in_ramblock(). */ | |
3403 | return (void *)QEMU_ALIGN_DOWN((uintptr_t)(block->host + offset), | |
3404 | block->page_size); | |
3405 | } | |
3406 | ||
3407 | static ram_addr_t host_page_offset_from_ram_block_offset(RAMBlock *block, | |
3408 | ram_addr_t offset) | |
3409 | { | |
3410 | return ((uintptr_t)block->host + offset) & (block->page_size - 1); | |
3411 | } | |
3412 | ||
871cfc54 LS |
3413 | void colo_record_bitmap(RAMBlock *block, ram_addr_t *normal, uint32_t pages) |
3414 | { | |
3415 | qemu_mutex_lock(&ram_state->bitmap_mutex); | |
3416 | for (int i = 0; i < pages; i++) { | |
3417 | ram_addr_t offset = normal[i]; | |
3418 | ram_state->migration_dirty_pages += !test_and_set_bit( | |
3419 | offset >> TARGET_PAGE_BITS, | |
3420 | block->bmap); | |
3421 | } | |
3422 | qemu_mutex_unlock(&ram_state->bitmap_mutex); | |
3423 | } | |
3424 | ||
13af18f2 | 3425 | static inline void *colo_cache_from_block_offset(RAMBlock *block, |
8af66371 | 3426 | ram_addr_t offset, bool record_bitmap) |
13af18f2 ZC |
3427 | { |
3428 | if (!offset_in_ramblock(block, offset)) { | |
3429 | return NULL; | |
3430 | } | |
3431 | if (!block->colo_cache) { | |
3432 | error_report("%s: colo_cache is NULL in block :%s", | |
3433 | __func__, block->idstr); | |
3434 | return NULL; | |
3435 | } | |
7d9acafa ZC |
3436 | |
3437 | /* | |
3438 | * During colo checkpoint, we need bitmap of these migrated pages. | |
3439 | * It help us to decide which pages in ram cache should be flushed | |
3440 | * into VM's RAM later. | |
3441 | */ | |
871cfc54 LS |
3442 | if (record_bitmap) { |
3443 | colo_record_bitmap(block, &offset, 1); | |
7d9acafa | 3444 | } |
13af18f2 ZC |
3445 | return block->colo_cache + offset; |
3446 | } | |
3447 | ||
3d0684b2 JQ |
3448 | /** |
3449 | * ram_handle_compressed: handle the zero page case | |
3450 | * | |
56e93d26 JQ |
3451 | * If a page (or a whole RDMA chunk) has been |
3452 | * determined to be zero, then zap it. | |
3d0684b2 JQ |
3453 | * |
3454 | * @host: host address for the zero page | |
3455 | * @ch: what the page is filled from. We only support zero | |
3456 | * @size: size of the zero page | |
56e93d26 JQ |
3457 | */ |
3458 | void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) | |
3459 | { | |
bad452a7 | 3460 | if (ch != 0 || !buffer_is_zero(host, size)) { |
56e93d26 JQ |
3461 | memset(host, ch, size); |
3462 | } | |
3463 | } | |
3464 | ||
b70cb3b4 RL |
3465 | static void colo_init_ram_state(void) |
3466 | { | |
3467 | ram_state_init(&ram_state); | |
b70cb3b4 RL |
3468 | } |
3469 | ||
13af18f2 ZC |
3470 | /* |
3471 | * colo cache: this is for secondary VM, we cache the whole | |
3472 | * memory of the secondary VM, it is need to hold the global lock | |
3473 | * to call this helper. | |
3474 | */ | |
3475 | int colo_init_ram_cache(void) | |
3476 | { | |
3477 | RAMBlock *block; | |
3478 | ||
44901b5a PB |
3479 | WITH_RCU_READ_LOCK_GUARD() { |
3480 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3481 | block->colo_cache = qemu_anon_ram_alloc(block->used_length, | |
8dbe22c6 | 3482 | NULL, false, false); |
44901b5a PB |
3483 | if (!block->colo_cache) { |
3484 | error_report("%s: Can't alloc memory for COLO cache of block %s," | |
3485 | "size 0x" RAM_ADDR_FMT, __func__, block->idstr, | |
3486 | block->used_length); | |
3487 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3488 | if (block->colo_cache) { | |
3489 | qemu_anon_ram_free(block->colo_cache, block->used_length); | |
3490 | block->colo_cache = NULL; | |
3491 | } | |
89ac5a1d | 3492 | } |
44901b5a | 3493 | return -errno; |
89ac5a1d | 3494 | } |
e5fdf920 LS |
3495 | if (!machine_dump_guest_core(current_machine)) { |
3496 | qemu_madvise(block->colo_cache, block->used_length, | |
3497 | QEMU_MADV_DONTDUMP); | |
3498 | } | |
13af18f2 | 3499 | } |
13af18f2 | 3500 | } |
44901b5a | 3501 | |
7d9acafa ZC |
3502 | /* |
3503 | * Record the dirty pages that sent by PVM, we use this dirty bitmap together | |
3504 | * with to decide which page in cache should be flushed into SVM's RAM. Here | |
3505 | * we use the same name 'ram_bitmap' as for migration. | |
3506 | */ | |
3507 | if (ram_bytes_total()) { | |
fbd162e6 | 3508 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
7d9acafa | 3509 | unsigned long pages = block->max_length >> TARGET_PAGE_BITS; |
7d9acafa | 3510 | block->bmap = bitmap_new(pages); |
7d9acafa ZC |
3511 | } |
3512 | } | |
7d9acafa | 3513 | |
b70cb3b4 | 3514 | colo_init_ram_state(); |
13af18f2 | 3515 | return 0; |
13af18f2 ZC |
3516 | } |
3517 | ||
0393031a HZ |
3518 | /* TODO: duplicated with ram_init_bitmaps */ |
3519 | void colo_incoming_start_dirty_log(void) | |
3520 | { | |
3521 | RAMBlock *block = NULL; | |
3522 | /* For memory_global_dirty_log_start below. */ | |
3523 | qemu_mutex_lock_iothread(); | |
3524 | qemu_mutex_lock_ramlist(); | |
3525 | ||
1e493be5 | 3526 | memory_global_dirty_log_sync(false); |
0393031a HZ |
3527 | WITH_RCU_READ_LOCK_GUARD() { |
3528 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3529 | ramblock_sync_dirty_bitmap(ram_state, block); | |
3530 | /* Discard this dirty bitmap record */ | |
3531 | bitmap_zero(block->bmap, block->max_length >> TARGET_PAGE_BITS); | |
3532 | } | |
63b41db4 | 3533 | memory_global_dirty_log_start(GLOBAL_DIRTY_MIGRATION); |
0393031a HZ |
3534 | } |
3535 | ram_state->migration_dirty_pages = 0; | |
3536 | qemu_mutex_unlock_ramlist(); | |
3537 | qemu_mutex_unlock_iothread(); | |
3538 | } | |
3539 | ||
13af18f2 ZC |
3540 | /* It is need to hold the global lock to call this helper */ |
3541 | void colo_release_ram_cache(void) | |
3542 | { | |
3543 | RAMBlock *block; | |
3544 | ||
63b41db4 | 3545 | memory_global_dirty_log_stop(GLOBAL_DIRTY_MIGRATION); |
fbd162e6 | 3546 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
7d9acafa ZC |
3547 | g_free(block->bmap); |
3548 | block->bmap = NULL; | |
3549 | } | |
3550 | ||
89ac5a1d DDAG |
3551 | WITH_RCU_READ_LOCK_GUARD() { |
3552 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3553 | if (block->colo_cache) { | |
3554 | qemu_anon_ram_free(block->colo_cache, block->used_length); | |
3555 | block->colo_cache = NULL; | |
3556 | } | |
13af18f2 ZC |
3557 | } |
3558 | } | |
0393031a | 3559 | ram_state_cleanup(&ram_state); |
13af18f2 ZC |
3560 | } |
3561 | ||
f265e0e4 JQ |
3562 | /** |
3563 | * ram_load_setup: Setup RAM for migration incoming side | |
3564 | * | |
3565 | * Returns zero to indicate success and negative for error | |
3566 | * | |
3567 | * @f: QEMUFile where to receive the data | |
3568 | * @opaque: RAMState pointer | |
3569 | */ | |
3570 | static int ram_load_setup(QEMUFile *f, void *opaque) | |
3571 | { | |
3572 | xbzrle_load_setup(); | |
f9494614 | 3573 | ramblock_recv_map_init(); |
13af18f2 | 3574 | |
f265e0e4 JQ |
3575 | return 0; |
3576 | } | |
3577 | ||
3578 | static int ram_load_cleanup(void *opaque) | |
3579 | { | |
f9494614 | 3580 | RAMBlock *rb; |
56eb90af | 3581 | |
fbd162e6 | 3582 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
bd108a44 | 3583 | qemu_ram_block_writeback(rb); |
56eb90af JH |
3584 | } |
3585 | ||
f265e0e4 | 3586 | xbzrle_load_cleanup(); |
f9494614 | 3587 | |
fbd162e6 | 3588 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
f9494614 AP |
3589 | g_free(rb->receivedmap); |
3590 | rb->receivedmap = NULL; | |
3591 | } | |
13af18f2 | 3592 | |
f265e0e4 JQ |
3593 | return 0; |
3594 | } | |
3595 | ||
3d0684b2 JQ |
3596 | /** |
3597 | * ram_postcopy_incoming_init: allocate postcopy data structures | |
3598 | * | |
3599 | * Returns 0 for success and negative if there was one error | |
3600 | * | |
3601 | * @mis: current migration incoming state | |
3602 | * | |
3603 | * Allocate data structures etc needed by incoming migration with | |
3604 | * postcopy-ram. postcopy-ram's similarly names | |
3605 | * postcopy_ram_incoming_init does the work. | |
1caddf8a DDAG |
3606 | */ |
3607 | int ram_postcopy_incoming_init(MigrationIncomingState *mis) | |
3608 | { | |
c136180c | 3609 | return postcopy_ram_incoming_init(mis); |
1caddf8a DDAG |
3610 | } |
3611 | ||
3d0684b2 JQ |
3612 | /** |
3613 | * ram_load_postcopy: load a page in postcopy case | |
3614 | * | |
3615 | * Returns 0 for success or -errno in case of error | |
3616 | * | |
a7180877 DDAG |
3617 | * Called in postcopy mode by ram_load(). |
3618 | * rcu_read_lock is taken prior to this being called. | |
3d0684b2 JQ |
3619 | * |
3620 | * @f: QEMUFile where to send the data | |
36f62f11 | 3621 | * @channel: the channel to use for loading |
a7180877 | 3622 | */ |
36f62f11 | 3623 | int ram_load_postcopy(QEMUFile *f, int channel) |
a7180877 DDAG |
3624 | { |
3625 | int flags = 0, ret = 0; | |
3626 | bool place_needed = false; | |
1aa83678 | 3627 | bool matches_target_page_size = false; |
a7180877 | 3628 | MigrationIncomingState *mis = migration_incoming_get_current(); |
36f62f11 | 3629 | PostcopyTmpPage *tmp_page = &mis->postcopy_tmp_pages[channel]; |
a7180877 DDAG |
3630 | |
3631 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
3632 | ram_addr_t addr; | |
a7180877 DDAG |
3633 | void *page_buffer = NULL; |
3634 | void *place_source = NULL; | |
df9ff5e1 | 3635 | RAMBlock *block = NULL; |
a7180877 | 3636 | uint8_t ch; |
644acf99 | 3637 | int len; |
a7180877 DDAG |
3638 | |
3639 | addr = qemu_get_be64(f); | |
7a9ddfbf PX |
3640 | |
3641 | /* | |
3642 | * If qemu file error, we should stop here, and then "addr" | |
3643 | * may be invalid | |
3644 | */ | |
3645 | ret = qemu_file_get_error(f); | |
3646 | if (ret) { | |
3647 | break; | |
3648 | } | |
3649 | ||
a7180877 DDAG |
3650 | flags = addr & ~TARGET_PAGE_MASK; |
3651 | addr &= TARGET_PAGE_MASK; | |
3652 | ||
36f62f11 | 3653 | trace_ram_load_postcopy_loop(channel, (uint64_t)addr, flags); |
644acf99 WY |
3654 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
3655 | RAM_SAVE_FLAG_COMPRESS_PAGE)) { | |
c01b16ed | 3656 | block = ram_block_from_stream(mis, f, flags, channel); |
6a23f639 DH |
3657 | if (!block) { |
3658 | ret = -EINVAL; | |
3659 | break; | |
3660 | } | |
4c4bad48 | 3661 | |
898ba906 DH |
3662 | /* |
3663 | * Relying on used_length is racy and can result in false positives. | |
3664 | * We might place pages beyond used_length in case RAM was shrunk | |
3665 | * while in postcopy, which is fine - trying to place via | |
3666 | * UFFDIO_COPY/UFFDIO_ZEROPAGE will never segfault. | |
3667 | */ | |
3668 | if (!block->host || addr >= block->postcopy_length) { | |
a7180877 DDAG |
3669 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); |
3670 | ret = -EINVAL; | |
3671 | break; | |
3672 | } | |
77dadc3f | 3673 | tmp_page->target_pages++; |
1aa83678 | 3674 | matches_target_page_size = block->page_size == TARGET_PAGE_SIZE; |
a7180877 | 3675 | /* |
28abd200 DDAG |
3676 | * Postcopy requires that we place whole host pages atomically; |
3677 | * these may be huge pages for RAMBlocks that are backed by | |
3678 | * hugetlbfs. | |
a7180877 DDAG |
3679 | * To make it atomic, the data is read into a temporary page |
3680 | * that's moved into place later. | |
3681 | * The migration protocol uses, possibly smaller, target-pages | |
3682 | * however the source ensures it always sends all the components | |
91ba442f | 3683 | * of a host page in one chunk. |
a7180877 | 3684 | */ |
77dadc3f | 3685 | page_buffer = tmp_page->tmp_huge_page + |
6a23f639 DH |
3686 | host_page_offset_from_ram_block_offset(block, addr); |
3687 | /* If all TP are zero then we can optimise the place */ | |
77dadc3f PX |
3688 | if (tmp_page->target_pages == 1) { |
3689 | tmp_page->host_addr = | |
3690 | host_page_from_ram_block_offset(block, addr); | |
3691 | } else if (tmp_page->host_addr != | |
3692 | host_page_from_ram_block_offset(block, addr)) { | |
c53b7ddc | 3693 | /* not the 1st TP within the HP */ |
36f62f11 | 3694 | error_report("Non-same host page detected on channel %d: " |
cfc7dc8a PX |
3695 | "Target host page %p, received host page %p " |
3696 | "(rb %s offset 0x"RAM_ADDR_FMT" target_pages %d)", | |
36f62f11 | 3697 | channel, tmp_page->host_addr, |
cfc7dc8a PX |
3698 | host_page_from_ram_block_offset(block, addr), |
3699 | block->idstr, addr, tmp_page->target_pages); | |
6a23f639 DH |
3700 | ret = -EINVAL; |
3701 | break; | |
a7180877 DDAG |
3702 | } |
3703 | ||
3704 | /* | |
3705 | * If it's the last part of a host page then we place the host | |
3706 | * page | |
3707 | */ | |
77dadc3f PX |
3708 | if (tmp_page->target_pages == |
3709 | (block->page_size / TARGET_PAGE_SIZE)) { | |
4cbb3c63 | 3710 | place_needed = true; |
4cbb3c63 | 3711 | } |
77dadc3f | 3712 | place_source = tmp_page->tmp_huge_page; |
a7180877 DDAG |
3713 | } |
3714 | ||
3715 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { | |
bb890ed5 | 3716 | case RAM_SAVE_FLAG_ZERO: |
a7180877 | 3717 | ch = qemu_get_byte(f); |
2e36bc1b WY |
3718 | /* |
3719 | * Can skip to set page_buffer when | |
3720 | * this is a zero page and (block->page_size == TARGET_PAGE_SIZE). | |
3721 | */ | |
3722 | if (ch || !matches_target_page_size) { | |
3723 | memset(page_buffer, ch, TARGET_PAGE_SIZE); | |
3724 | } | |
a7180877 | 3725 | if (ch) { |
77dadc3f | 3726 | tmp_page->all_zero = false; |
a7180877 DDAG |
3727 | } |
3728 | break; | |
3729 | ||
3730 | case RAM_SAVE_FLAG_PAGE: | |
77dadc3f | 3731 | tmp_page->all_zero = false; |
1aa83678 PX |
3732 | if (!matches_target_page_size) { |
3733 | /* For huge pages, we always use temporary buffer */ | |
a7180877 DDAG |
3734 | qemu_get_buffer(f, page_buffer, TARGET_PAGE_SIZE); |
3735 | } else { | |
1aa83678 PX |
3736 | /* |
3737 | * For small pages that matches target page size, we | |
3738 | * avoid the qemu_file copy. Instead we directly use | |
3739 | * the buffer of QEMUFile to place the page. Note: we | |
3740 | * cannot do any QEMUFile operation before using that | |
3741 | * buffer to make sure the buffer is valid when | |
3742 | * placing the page. | |
a7180877 DDAG |
3743 | */ |
3744 | qemu_get_buffer_in_place(f, (uint8_t **)&place_source, | |
3745 | TARGET_PAGE_SIZE); | |
3746 | } | |
3747 | break; | |
644acf99 | 3748 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
77dadc3f | 3749 | tmp_page->all_zero = false; |
644acf99 WY |
3750 | len = qemu_get_be32(f); |
3751 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
3752 | error_report("Invalid compressed data length: %d", len); | |
3753 | ret = -EINVAL; | |
3754 | break; | |
3755 | } | |
3756 | decompress_data_with_multi_threads(f, page_buffer, len); | |
3757 | break; | |
294e5a40 JQ |
3758 | case RAM_SAVE_FLAG_MULTIFD_FLUSH: |
3759 | multifd_recv_sync_main(); | |
3760 | break; | |
a7180877 DDAG |
3761 | case RAM_SAVE_FLAG_EOS: |
3762 | /* normal exit */ | |
b05292c2 JQ |
3763 | if (migrate_multifd_flush_after_each_section()) { |
3764 | multifd_recv_sync_main(); | |
3765 | } | |
a7180877 DDAG |
3766 | break; |
3767 | default: | |
29fccade | 3768 | error_report("Unknown combination of migration flags: 0x%x" |
a7180877 DDAG |
3769 | " (postcopy mode)", flags); |
3770 | ret = -EINVAL; | |
7a9ddfbf PX |
3771 | break; |
3772 | } | |
3773 | ||
644acf99 WY |
3774 | /* Got the whole host page, wait for decompress before placing. */ |
3775 | if (place_needed) { | |
3776 | ret |= wait_for_decompress_done(); | |
3777 | } | |
3778 | ||
7a9ddfbf PX |
3779 | /* Detect for any possible file errors */ |
3780 | if (!ret && qemu_file_get_error(f)) { | |
3781 | ret = qemu_file_get_error(f); | |
a7180877 DDAG |
3782 | } |
3783 | ||
7a9ddfbf | 3784 | if (!ret && place_needed) { |
77dadc3f PX |
3785 | if (tmp_page->all_zero) { |
3786 | ret = postcopy_place_page_zero(mis, tmp_page->host_addr, block); | |
a7180877 | 3787 | } else { |
77dadc3f PX |
3788 | ret = postcopy_place_page(mis, tmp_page->host_addr, |
3789 | place_source, block); | |
a7180877 | 3790 | } |
ddf35bdf | 3791 | place_needed = false; |
77dadc3f | 3792 | postcopy_temp_page_reset(tmp_page); |
a7180877 | 3793 | } |
a7180877 DDAG |
3794 | } |
3795 | ||
3796 | return ret; | |
3797 | } | |
3798 | ||
acab30b8 DHB |
3799 | static bool postcopy_is_running(void) |
3800 | { | |
3801 | PostcopyState ps = postcopy_state_get(); | |
3802 | return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; | |
3803 | } | |
3804 | ||
e6f4aa18 ZC |
3805 | /* |
3806 | * Flush content of RAM cache into SVM's memory. | |
3807 | * Only flush the pages that be dirtied by PVM or SVM or both. | |
3808 | */ | |
24fa16f8 | 3809 | void colo_flush_ram_cache(void) |
e6f4aa18 ZC |
3810 | { |
3811 | RAMBlock *block = NULL; | |
3812 | void *dst_host; | |
3813 | void *src_host; | |
3814 | unsigned long offset = 0; | |
3815 | ||
1e493be5 | 3816 | memory_global_dirty_log_sync(false); |
9d638407 | 3817 | qemu_mutex_lock(&ram_state->bitmap_mutex); |
89ac5a1d DDAG |
3818 | WITH_RCU_READ_LOCK_GUARD() { |
3819 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3820 | ramblock_sync_dirty_bitmap(ram_state, block); | |
3821 | } | |
d1955d22 | 3822 | } |
d1955d22 | 3823 | |
e6f4aa18 | 3824 | trace_colo_flush_ram_cache_begin(ram_state->migration_dirty_pages); |
89ac5a1d DDAG |
3825 | WITH_RCU_READ_LOCK_GUARD() { |
3826 | block = QLIST_FIRST_RCU(&ram_list.blocks); | |
e6f4aa18 | 3827 | |
89ac5a1d | 3828 | while (block) { |
a6a83cef | 3829 | unsigned long num = 0; |
e6f4aa18 | 3830 | |
a6a83cef | 3831 | offset = colo_bitmap_find_dirty(ram_state, block, offset, &num); |
542147f4 DH |
3832 | if (!offset_in_ramblock(block, |
3833 | ((ram_addr_t)offset) << TARGET_PAGE_BITS)) { | |
89ac5a1d | 3834 | offset = 0; |
a6a83cef | 3835 | num = 0; |
89ac5a1d DDAG |
3836 | block = QLIST_NEXT_RCU(block, next); |
3837 | } else { | |
a6a83cef RL |
3838 | unsigned long i = 0; |
3839 | ||
3840 | for (i = 0; i < num; i++) { | |
3841 | migration_bitmap_clear_dirty(ram_state, block, offset + i); | |
3842 | } | |
8bba004c AR |
3843 | dst_host = block->host |
3844 | + (((ram_addr_t)offset) << TARGET_PAGE_BITS); | |
3845 | src_host = block->colo_cache | |
3846 | + (((ram_addr_t)offset) << TARGET_PAGE_BITS); | |
a6a83cef RL |
3847 | memcpy(dst_host, src_host, TARGET_PAGE_SIZE * num); |
3848 | offset += num; | |
89ac5a1d | 3849 | } |
e6f4aa18 ZC |
3850 | } |
3851 | } | |
9d638407 | 3852 | qemu_mutex_unlock(&ram_state->bitmap_mutex); |
e6f4aa18 ZC |
3853 | trace_colo_flush_ram_cache_end(); |
3854 | } | |
3855 | ||
10da4a36 WY |
3856 | /** |
3857 | * ram_load_precopy: load pages in precopy case | |
3858 | * | |
3859 | * Returns 0 for success or -errno in case of error | |
3860 | * | |
3861 | * Called in precopy mode by ram_load(). | |
3862 | * rcu_read_lock is taken prior to this being called. | |
3863 | * | |
3864 | * @f: QEMUFile where to send the data | |
3865 | */ | |
3866 | static int ram_load_precopy(QEMUFile *f) | |
56e93d26 | 3867 | { |
755e8d7c | 3868 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e65cec5e | 3869 | int flags = 0, ret = 0, invalid_flags = 0, len = 0, i = 0; |
ef08fb38 | 3870 | /* ADVISE is earlier, it shows the source has the postcopy capability on */ |
80fe315c | 3871 | bool postcopy_advised = migration_incoming_postcopy_advised(); |
a7a94d14 | 3872 | if (!migrate_compress()) { |
edc60127 JQ |
3873 | invalid_flags |= RAM_SAVE_FLAG_COMPRESS_PAGE; |
3874 | } | |
a7180877 | 3875 | |
10da4a36 | 3876 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { |
56e93d26 | 3877 | ram_addr_t addr, total_ram_bytes; |
0393031a | 3878 | void *host = NULL, *host_bak = NULL; |
56e93d26 JQ |
3879 | uint8_t ch; |
3880 | ||
e65cec5e YK |
3881 | /* |
3882 | * Yield periodically to let main loop run, but an iteration of | |
3883 | * the main loop is expensive, so do it each some iterations | |
3884 | */ | |
3885 | if ((i & 32767) == 0 && qemu_in_coroutine()) { | |
3886 | aio_co_schedule(qemu_get_current_aio_context(), | |
3887 | qemu_coroutine_self()); | |
3888 | qemu_coroutine_yield(); | |
3889 | } | |
3890 | i++; | |
3891 | ||
56e93d26 JQ |
3892 | addr = qemu_get_be64(f); |
3893 | flags = addr & ~TARGET_PAGE_MASK; | |
3894 | addr &= TARGET_PAGE_MASK; | |
3895 | ||
edc60127 JQ |
3896 | if (flags & invalid_flags) { |
3897 | if (flags & invalid_flags & RAM_SAVE_FLAG_COMPRESS_PAGE) { | |
3898 | error_report("Received an unexpected compressed page"); | |
3899 | } | |
3900 | ||
3901 | ret = -EINVAL; | |
3902 | break; | |
3903 | } | |
3904 | ||
bb890ed5 | 3905 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
a776aa15 | 3906 | RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) { |
c01b16ed PX |
3907 | RAMBlock *block = ram_block_from_stream(mis, f, flags, |
3908 | RAM_CHANNEL_PRECOPY); | |
4c4bad48 | 3909 | |
0393031a | 3910 | host = host_from_ram_block_offset(block, addr); |
13af18f2 | 3911 | /* |
0393031a HZ |
3912 | * After going into COLO stage, we should not load the page |
3913 | * into SVM's memory directly, we put them into colo_cache firstly. | |
3914 | * NOTE: We need to keep a copy of SVM's ram in colo_cache. | |
3915 | * Previously, we copied all these memory in preparing stage of COLO | |
3916 | * while we need to stop VM, which is a time-consuming process. | |
3917 | * Here we optimize it by a trick, back-up every page while in | |
3918 | * migration process while COLO is enabled, though it affects the | |
3919 | * speed of the migration, but it obviously reduce the downtime of | |
3920 | * back-up all SVM'S memory in COLO preparing stage. | |
13af18f2 | 3921 | */ |
0393031a HZ |
3922 | if (migration_incoming_colo_enabled()) { |
3923 | if (migration_incoming_in_colo_state()) { | |
3924 | /* In COLO stage, put all pages into cache temporarily */ | |
8af66371 | 3925 | host = colo_cache_from_block_offset(block, addr, true); |
0393031a HZ |
3926 | } else { |
3927 | /* | |
3928 | * In migration stage but before COLO stage, | |
3929 | * Put all pages into both cache and SVM's memory. | |
3930 | */ | |
8af66371 | 3931 | host_bak = colo_cache_from_block_offset(block, addr, false); |
0393031a | 3932 | } |
13af18f2 | 3933 | } |
a776aa15 DDAG |
3934 | if (!host) { |
3935 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
3936 | ret = -EINVAL; | |
3937 | break; | |
3938 | } | |
13af18f2 ZC |
3939 | if (!migration_incoming_in_colo_state()) { |
3940 | ramblock_recv_bitmap_set(block, host); | |
3941 | } | |
3942 | ||
1db9d8e5 | 3943 | trace_ram_load_loop(block->idstr, (uint64_t)addr, flags, host); |
a776aa15 DDAG |
3944 | } |
3945 | ||
56e93d26 JQ |
3946 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { |
3947 | case RAM_SAVE_FLAG_MEM_SIZE: | |
3948 | /* Synchronize RAM block list */ | |
3949 | total_ram_bytes = addr; | |
3950 | while (!ret && total_ram_bytes) { | |
3951 | RAMBlock *block; | |
56e93d26 JQ |
3952 | char id[256]; |
3953 | ram_addr_t length; | |
3954 | ||
3955 | len = qemu_get_byte(f); | |
3956 | qemu_get_buffer(f, (uint8_t *)id, len); | |
3957 | id[len] = 0; | |
3958 | length = qemu_get_be64(f); | |
3959 | ||
e3dd7493 | 3960 | block = qemu_ram_block_by_name(id); |
b895de50 CLG |
3961 | if (block && !qemu_ram_is_migratable(block)) { |
3962 | error_report("block %s should not be migrated !", id); | |
3963 | ret = -EINVAL; | |
3964 | } else if (block) { | |
e3dd7493 DDAG |
3965 | if (length != block->used_length) { |
3966 | Error *local_err = NULL; | |
56e93d26 | 3967 | |
fa53a0e5 | 3968 | ret = qemu_ram_resize(block, length, |
e3dd7493 DDAG |
3969 | &local_err); |
3970 | if (local_err) { | |
3971 | error_report_err(local_err); | |
56e93d26 | 3972 | } |
56e93d26 | 3973 | } |
ef08fb38 | 3974 | /* For postcopy we need to check hugepage sizes match */ |
e846b746 | 3975 | if (postcopy_advised && migrate_postcopy_ram() && |
ef08fb38 DDAG |
3976 | block->page_size != qemu_host_page_size) { |
3977 | uint64_t remote_page_size = qemu_get_be64(f); | |
3978 | if (remote_page_size != block->page_size) { | |
3979 | error_report("Mismatched RAM page size %s " | |
3980 | "(local) %zd != %" PRId64, | |
3981 | id, block->page_size, | |
3982 | remote_page_size); | |
3983 | ret = -EINVAL; | |
3984 | } | |
3985 | } | |
fbd162e6 | 3986 | if (migrate_ignore_shared()) { |
7f3de3f0 | 3987 | hwaddr addr2 = qemu_get_be64(f); |
f161c88a | 3988 | if (migrate_ram_is_ignored(block) && |
7f3de3f0 | 3989 | block->mr->addr != addr2) { |
fbd162e6 YK |
3990 | error_report("Mismatched GPAs for block %s " |
3991 | "%" PRId64 "!= %" PRId64, | |
7f3de3f0 | 3992 | id, (uint64_t)addr2, |
fbd162e6 YK |
3993 | (uint64_t)block->mr->addr); |
3994 | ret = -EINVAL; | |
3995 | } | |
3996 | } | |
e3dd7493 DDAG |
3997 | ram_control_load_hook(f, RAM_CONTROL_BLOCK_REG, |
3998 | block->idstr); | |
3999 | } else { | |
56e93d26 JQ |
4000 | error_report("Unknown ramblock \"%s\", cannot " |
4001 | "accept migration", id); | |
4002 | ret = -EINVAL; | |
4003 | } | |
4004 | ||
4005 | total_ram_bytes -= length; | |
4006 | } | |
4007 | break; | |
a776aa15 | 4008 | |
bb890ed5 | 4009 | case RAM_SAVE_FLAG_ZERO: |
56e93d26 JQ |
4010 | ch = qemu_get_byte(f); |
4011 | ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); | |
4012 | break; | |
a776aa15 | 4013 | |
56e93d26 | 4014 | case RAM_SAVE_FLAG_PAGE: |
56e93d26 JQ |
4015 | qemu_get_buffer(f, host, TARGET_PAGE_SIZE); |
4016 | break; | |
56e93d26 | 4017 | |
a776aa15 | 4018 | case RAM_SAVE_FLAG_COMPRESS_PAGE: |
56e93d26 JQ |
4019 | len = qemu_get_be32(f); |
4020 | if (len < 0 || len > compressBound(TARGET_PAGE_SIZE)) { | |
4021 | error_report("Invalid compressed data length: %d", len); | |
4022 | ret = -EINVAL; | |
4023 | break; | |
4024 | } | |
c1bc6626 | 4025 | decompress_data_with_multi_threads(f, host, len); |
56e93d26 | 4026 | break; |
a776aa15 | 4027 | |
56e93d26 | 4028 | case RAM_SAVE_FLAG_XBZRLE: |
56e93d26 JQ |
4029 | if (load_xbzrle(f, addr, host) < 0) { |
4030 | error_report("Failed to decompress XBZRLE page at " | |
4031 | RAM_ADDR_FMT, addr); | |
4032 | ret = -EINVAL; | |
4033 | break; | |
4034 | } | |
4035 | break; | |
294e5a40 JQ |
4036 | case RAM_SAVE_FLAG_MULTIFD_FLUSH: |
4037 | multifd_recv_sync_main(); | |
4038 | break; | |
56e93d26 JQ |
4039 | case RAM_SAVE_FLAG_EOS: |
4040 | /* normal exit */ | |
b05292c2 JQ |
4041 | if (migrate_multifd_flush_after_each_section()) { |
4042 | multifd_recv_sync_main(); | |
4043 | } | |
56e93d26 | 4044 | break; |
5f1e7540 JQ |
4045 | case RAM_SAVE_FLAG_HOOK: |
4046 | ram_control_load_hook(f, RAM_CONTROL_HOOK, NULL); | |
4047 | break; | |
56e93d26 | 4048 | default: |
5f1e7540 JQ |
4049 | error_report("Unknown combination of migration flags: 0x%x", flags); |
4050 | ret = -EINVAL; | |
56e93d26 JQ |
4051 | } |
4052 | if (!ret) { | |
4053 | ret = qemu_file_get_error(f); | |
4054 | } | |
0393031a HZ |
4055 | if (!ret && host_bak) { |
4056 | memcpy(host_bak, host, TARGET_PAGE_SIZE); | |
4057 | } | |
56e93d26 JQ |
4058 | } |
4059 | ||
ca1a6b70 | 4060 | ret |= wait_for_decompress_done(); |
10da4a36 WY |
4061 | return ret; |
4062 | } | |
4063 | ||
4064 | static int ram_load(QEMUFile *f, void *opaque, int version_id) | |
4065 | { | |
4066 | int ret = 0; | |
4067 | static uint64_t seq_iter; | |
4068 | /* | |
4069 | * If system is running in postcopy mode, page inserts to host memory must | |
4070 | * be atomic | |
4071 | */ | |
4072 | bool postcopy_running = postcopy_is_running(); | |
4073 | ||
4074 | seq_iter++; | |
4075 | ||
4076 | if (version_id != 4) { | |
4077 | return -EINVAL; | |
4078 | } | |
4079 | ||
4080 | /* | |
4081 | * This RCU critical section can be very long running. | |
4082 | * When RCU reclaims in the code start to become numerous, | |
4083 | * it will be necessary to reduce the granularity of this | |
4084 | * critical section. | |
4085 | */ | |
89ac5a1d DDAG |
4086 | WITH_RCU_READ_LOCK_GUARD() { |
4087 | if (postcopy_running) { | |
36f62f11 PX |
4088 | /* |
4089 | * Note! Here RAM_CHANNEL_PRECOPY is the precopy channel of | |
4090 | * postcopy migration, we have another RAM_CHANNEL_POSTCOPY to | |
4091 | * service fast page faults. | |
4092 | */ | |
4093 | ret = ram_load_postcopy(f, RAM_CHANNEL_PRECOPY); | |
89ac5a1d DDAG |
4094 | } else { |
4095 | ret = ram_load_precopy(f); | |
4096 | } | |
10da4a36 | 4097 | } |
55c4446b | 4098 | trace_ram_load_complete(ret, seq_iter); |
e6f4aa18 | 4099 | |
56e93d26 JQ |
4100 | return ret; |
4101 | } | |
4102 | ||
c6467627 VSO |
4103 | static bool ram_has_postcopy(void *opaque) |
4104 | { | |
469dd51b | 4105 | RAMBlock *rb; |
fbd162e6 | 4106 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
469dd51b JH |
4107 | if (ramblock_is_pmem(rb)) { |
4108 | info_report("Block: %s, host: %p is a nvdimm memory, postcopy" | |
4109 | "is not supported now!", rb->idstr, rb->host); | |
4110 | return false; | |
4111 | } | |
4112 | } | |
4113 | ||
c6467627 VSO |
4114 | return migrate_postcopy_ram(); |
4115 | } | |
4116 | ||
edd090c7 PX |
4117 | /* Sync all the dirty bitmap with destination VM. */ |
4118 | static int ram_dirty_bitmap_sync_all(MigrationState *s, RAMState *rs) | |
4119 | { | |
4120 | RAMBlock *block; | |
4121 | QEMUFile *file = s->to_dst_file; | |
4122 | int ramblock_count = 0; | |
4123 | ||
4124 | trace_ram_dirty_bitmap_sync_start(); | |
4125 | ||
fbd162e6 | 4126 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
edd090c7 PX |
4127 | qemu_savevm_send_recv_bitmap(file, block->idstr); |
4128 | trace_ram_dirty_bitmap_request(block->idstr); | |
4129 | ramblock_count++; | |
4130 | } | |
4131 | ||
4132 | trace_ram_dirty_bitmap_sync_wait(); | |
4133 | ||
4134 | /* Wait until all the ramblocks' dirty bitmap synced */ | |
4135 | while (ramblock_count--) { | |
4136 | qemu_sem_wait(&s->rp_state.rp_sem); | |
4137 | } | |
4138 | ||
4139 | trace_ram_dirty_bitmap_sync_complete(); | |
4140 | ||
4141 | return 0; | |
4142 | } | |
4143 | ||
4144 | static void ram_dirty_bitmap_reload_notify(MigrationState *s) | |
4145 | { | |
4146 | qemu_sem_post(&s->rp_state.rp_sem); | |
4147 | } | |
4148 | ||
a335debb PX |
4149 | /* |
4150 | * Read the received bitmap, revert it as the initial dirty bitmap. | |
4151 | * This is only used when the postcopy migration is paused but wants | |
4152 | * to resume from a middle point. | |
4153 | */ | |
4154 | int ram_dirty_bitmap_reload(MigrationState *s, RAMBlock *block) | |
4155 | { | |
4156 | int ret = -EINVAL; | |
43044ac0 | 4157 | /* from_dst_file is always valid because we're within rp_thread */ |
a335debb PX |
4158 | QEMUFile *file = s->rp_state.from_dst_file; |
4159 | unsigned long *le_bitmap, nbits = block->used_length >> TARGET_PAGE_BITS; | |
a725ef9f | 4160 | uint64_t local_size = DIV_ROUND_UP(nbits, 8); |
a335debb PX |
4161 | uint64_t size, end_mark; |
4162 | ||
4163 | trace_ram_dirty_bitmap_reload_begin(block->idstr); | |
4164 | ||
4165 | if (s->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
4166 | error_report("%s: incorrect state %s", __func__, | |
4167 | MigrationStatus_str(s->state)); | |
4168 | return -EINVAL; | |
4169 | } | |
4170 | ||
4171 | /* | |
4172 | * Note: see comments in ramblock_recv_bitmap_send() on why we | |
3a4452d8 | 4173 | * need the endianness conversion, and the paddings. |
a335debb PX |
4174 | */ |
4175 | local_size = ROUND_UP(local_size, 8); | |
4176 | ||
4177 | /* Add paddings */ | |
4178 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
4179 | ||
4180 | size = qemu_get_be64(file); | |
4181 | ||
4182 | /* The size of the bitmap should match with our ramblock */ | |
4183 | if (size != local_size) { | |
4184 | error_report("%s: ramblock '%s' bitmap size mismatch " | |
4185 | "(0x%"PRIx64" != 0x%"PRIx64")", __func__, | |
4186 | block->idstr, size, local_size); | |
4187 | ret = -EINVAL; | |
4188 | goto out; | |
4189 | } | |
4190 | ||
4191 | size = qemu_get_buffer(file, (uint8_t *)le_bitmap, local_size); | |
4192 | end_mark = qemu_get_be64(file); | |
4193 | ||
4194 | ret = qemu_file_get_error(file); | |
4195 | if (ret || size != local_size) { | |
4196 | error_report("%s: read bitmap failed for ramblock '%s': %d" | |
4197 | " (size 0x%"PRIx64", got: 0x%"PRIx64")", | |
4198 | __func__, block->idstr, ret, local_size, size); | |
4199 | ret = -EIO; | |
4200 | goto out; | |
4201 | } | |
4202 | ||
4203 | if (end_mark != RAMBLOCK_RECV_BITMAP_ENDING) { | |
af3bbbe9 | 4204 | error_report("%s: ramblock '%s' end mark incorrect: 0x%"PRIx64, |
a335debb PX |
4205 | __func__, block->idstr, end_mark); |
4206 | ret = -EINVAL; | |
4207 | goto out; | |
4208 | } | |
4209 | ||
4210 | /* | |
3a4452d8 | 4211 | * Endianness conversion. We are during postcopy (though paused). |
a335debb PX |
4212 | * The dirty bitmap won't change. We can directly modify it. |
4213 | */ | |
4214 | bitmap_from_le(block->bmap, le_bitmap, nbits); | |
4215 | ||
4216 | /* | |
4217 | * What we received is "received bitmap". Revert it as the initial | |
4218 | * dirty bitmap for this ramblock. | |
4219 | */ | |
4220 | bitmap_complement(block->bmap, block->bmap, nbits); | |
4221 | ||
be39b4cd DH |
4222 | /* Clear dirty bits of discarded ranges that we don't want to migrate. */ |
4223 | ramblock_dirty_bitmap_clear_discarded_pages(block); | |
4224 | ||
4225 | /* We'll recalculate migration_dirty_pages in ram_state_resume_prepare(). */ | |
a335debb PX |
4226 | trace_ram_dirty_bitmap_reload_complete(block->idstr); |
4227 | ||
edd090c7 PX |
4228 | /* |
4229 | * We succeeded to sync bitmap for current ramblock. If this is | |
4230 | * the last one to sync, we need to notify the main send thread. | |
4231 | */ | |
4232 | ram_dirty_bitmap_reload_notify(s); | |
4233 | ||
a335debb PX |
4234 | ret = 0; |
4235 | out: | |
bf269906 | 4236 | g_free(le_bitmap); |
a335debb PX |
4237 | return ret; |
4238 | } | |
4239 | ||
edd090c7 PX |
4240 | static int ram_resume_prepare(MigrationState *s, void *opaque) |
4241 | { | |
4242 | RAMState *rs = *(RAMState **)opaque; | |
08614f34 | 4243 | int ret; |
edd090c7 | 4244 | |
08614f34 PX |
4245 | ret = ram_dirty_bitmap_sync_all(s, rs); |
4246 | if (ret) { | |
4247 | return ret; | |
4248 | } | |
4249 | ||
4250 | ram_state_resume_prepare(rs, s->to_dst_file); | |
4251 | ||
4252 | return 0; | |
edd090c7 PX |
4253 | } |
4254 | ||
36f62f11 PX |
4255 | void postcopy_preempt_shutdown_file(MigrationState *s) |
4256 | { | |
4257 | qemu_put_be64(s->postcopy_qemufile_src, RAM_SAVE_FLAG_EOS); | |
4258 | qemu_fflush(s->postcopy_qemufile_src); | |
4259 | } | |
4260 | ||
56e93d26 | 4261 | static SaveVMHandlers savevm_ram_handlers = { |
9907e842 | 4262 | .save_setup = ram_save_setup, |
56e93d26 | 4263 | .save_live_iterate = ram_save_iterate, |
763c906b | 4264 | .save_live_complete_postcopy = ram_save_complete, |
a3e06c3d | 4265 | .save_live_complete_precopy = ram_save_complete, |
c6467627 | 4266 | .has_postcopy = ram_has_postcopy, |
c8df4a7a JQ |
4267 | .state_pending_exact = ram_state_pending_exact, |
4268 | .state_pending_estimate = ram_state_pending_estimate, | |
56e93d26 | 4269 | .load_state = ram_load, |
f265e0e4 JQ |
4270 | .save_cleanup = ram_save_cleanup, |
4271 | .load_setup = ram_load_setup, | |
4272 | .load_cleanup = ram_load_cleanup, | |
edd090c7 | 4273 | .resume_prepare = ram_resume_prepare, |
56e93d26 JQ |
4274 | }; |
4275 | ||
c7c0e724 DH |
4276 | static void ram_mig_ram_block_resized(RAMBlockNotifier *n, void *host, |
4277 | size_t old_size, size_t new_size) | |
4278 | { | |
cc61c703 | 4279 | PostcopyState ps = postcopy_state_get(); |
c7c0e724 DH |
4280 | ram_addr_t offset; |
4281 | RAMBlock *rb = qemu_ram_block_from_host(host, false, &offset); | |
4282 | Error *err = NULL; | |
4283 | ||
f161c88a | 4284 | if (migrate_ram_is_ignored(rb)) { |
c7c0e724 DH |
4285 | return; |
4286 | } | |
4287 | ||
4288 | if (!migration_is_idle()) { | |
4289 | /* | |
4290 | * Precopy code on the source cannot deal with the size of RAM blocks | |
4291 | * changing at random points in time - especially after sending the | |
4292 | * RAM block sizes in the migration stream, they must no longer change. | |
4293 | * Abort and indicate a proper reason. | |
4294 | */ | |
4295 | error_setg(&err, "RAM block '%s' resized during precopy.", rb->idstr); | |
458fecca | 4296 | migration_cancel(err); |
c7c0e724 | 4297 | error_free(err); |
c7c0e724 | 4298 | } |
cc61c703 DH |
4299 | |
4300 | switch (ps) { | |
4301 | case POSTCOPY_INCOMING_ADVISE: | |
4302 | /* | |
4303 | * Update what ram_postcopy_incoming_init()->init_range() does at the | |
4304 | * time postcopy was advised. Syncing RAM blocks with the source will | |
4305 | * result in RAM resizes. | |
4306 | */ | |
4307 | if (old_size < new_size) { | |
4308 | if (ram_discard_range(rb->idstr, old_size, new_size - old_size)) { | |
4309 | error_report("RAM block '%s' discard of resized RAM failed", | |
4310 | rb->idstr); | |
4311 | } | |
4312 | } | |
898ba906 | 4313 | rb->postcopy_length = new_size; |
cc61c703 DH |
4314 | break; |
4315 | case POSTCOPY_INCOMING_NONE: | |
4316 | case POSTCOPY_INCOMING_RUNNING: | |
4317 | case POSTCOPY_INCOMING_END: | |
4318 | /* | |
4319 | * Once our guest is running, postcopy does no longer care about | |
4320 | * resizes. When growing, the new memory was not available on the | |
4321 | * source, no handler needed. | |
4322 | */ | |
4323 | break; | |
4324 | default: | |
4325 | error_report("RAM block '%s' resized during postcopy state: %d", | |
4326 | rb->idstr, ps); | |
4327 | exit(-1); | |
4328 | } | |
c7c0e724 DH |
4329 | } |
4330 | ||
4331 | static RAMBlockNotifier ram_mig_ram_notifier = { | |
4332 | .ram_block_resized = ram_mig_ram_block_resized, | |
4333 | }; | |
4334 | ||
56e93d26 JQ |
4335 | void ram_mig_init(void) |
4336 | { | |
4337 | qemu_mutex_init(&XBZRLE.lock); | |
ce62df53 | 4338 | register_savevm_live("ram", 0, 4, &savevm_ram_handlers, &ram_state); |
c7c0e724 | 4339 | ram_block_notifier_add(&ram_mig_ram_notifier); |
56e93d26 | 4340 | } |