]>
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" |
7b1e1a22 | 36 | #include "ram.h" |
6666c96a | 37 | #include "migration.h" |
947701cc | 38 | #include "migration-stats.h" |
f2a8f0a6 | 39 | #include "migration/register.h" |
7b1e1a22 | 40 | #include "migration/misc.h" |
08a0aee1 | 41 | #include "qemu-file.h" |
be07b0ac | 42 | #include "postcopy-ram.h" |
53d37d36 | 43 | #include "page_cache.h" |
56e93d26 | 44 | #include "qemu/error-report.h" |
e688df6b | 45 | #include "qapi/error.h" |
ab7cbb0b | 46 | #include "qapi/qapi-types-migration.h" |
9af23989 | 47 | #include "qapi/qapi-events-migration.h" |
acac51ba | 48 | #include "qapi/qapi-commands-migration.h" |
8acabf69 | 49 | #include "qapi/qmp/qerror.h" |
56e93d26 | 50 | #include "trace.h" |
56e93d26 | 51 | #include "exec/ram_addr.h" |
f9494614 | 52 | #include "exec/target_page.h" |
56e93d26 | 53 | #include "qemu/rcu_queue.h" |
a91246c9 | 54 | #include "migration/colo.h" |
b0c3cf94 | 55 | #include "sysemu/cpu-throttle.h" |
edd090c7 | 56 | #include "savevm.h" |
b9ee2f7d | 57 | #include "qemu/iov.h" |
d32ca5ad | 58 | #include "multifd.h" |
278e2f55 | 59 | #include "sysemu/runstate.h" |
48408174 | 60 | #include "rdma.h" |
1f0776f1 | 61 | #include "options.h" |
acac51ba HH |
62 | #include "sysemu/dirtylimit.h" |
63 | #include "sysemu/kvm.h" | |
278e2f55 | 64 | |
e5fdf920 LS |
65 | #include "hw/boards.h" /* for machine_dump_guest_core() */ |
66 | ||
278e2f55 AG |
67 | #if defined(__linux__) |
68 | #include "qemu/userfaultfd.h" | |
69 | #endif /* defined(__linux__) */ | |
56e93d26 | 70 | |
56e93d26 JQ |
71 | /***********************************************************/ |
72 | /* ram save/restore */ | |
73 | ||
7b548761 JQ |
74 | /* |
75 | * RAM_SAVE_FLAG_ZERO used to be named RAM_SAVE_FLAG_COMPRESS, it | |
76 | * worked for pages that were filled with the same char. We switched | |
bb890ed5 | 77 | * it to only search for the zero value. And to avoid confusion with |
7b548761 | 78 | * RAM_SAVE_FLAG_COMPRESS_PAGE just rename it. |
0222111a FR |
79 | * |
80 | * RAM_SAVE_FLAG_FULL was obsoleted in 2009. | |
81 | * | |
82 | * RAM_SAVE_FLAG_COMPRESS_PAGE (0x100) was removed in QEMU 9.1. | |
7b548761 JQ |
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 | |
10cb3336 | 91 | /* 0x80 is reserved in rdma.h for RAM_SAVE_FLAG_HOOK */ |
294e5a40 | 92 | #define RAM_SAVE_FLAG_MULTIFD_FLUSH 0x200 |
7b548761 | 93 | /* We can't use any flag that is bigger than 0x200 */ |
56e93d26 | 94 | |
c2d5c4a7 FR |
95 | /* |
96 | * mapped-ram migration supports O_DIRECT, so we need to make sure the | |
97 | * userspace buffer, the IO operation size and the file offset are | |
98 | * aligned according to the underlying device's block size. The first | |
99 | * two are already aligned to page size, but we need to add padding to | |
100 | * the file to align the offset. We cannot read the block size | |
101 | * dynamically because the migration file can be moved between | |
102 | * different systems, so use 1M to cover most block sizes and to keep | |
103 | * the file offset aligned at page size as well. | |
104 | */ | |
105 | #define MAPPED_RAM_FILE_OFFSET_ALIGNMENT 0x100000 | |
106 | ||
2f6b8826 FR |
107 | /* |
108 | * When doing mapped-ram migration, this is the amount we read from | |
109 | * the pages region in the migration file at a time. | |
110 | */ | |
111 | #define MAPPED_RAM_LOAD_BUF_SIZE 0x100000 | |
112 | ||
9360447d JQ |
113 | XBZRLECacheStats xbzrle_counters; |
114 | ||
f1668764 PX |
115 | /* used by the search for pages to send */ |
116 | struct PageSearchStatus { | |
117 | /* The migration channel used for a specific host page */ | |
118 | QEMUFile *pss_channel; | |
ec6f3ab9 PX |
119 | /* Last block from where we have sent data */ |
120 | RAMBlock *last_sent_block; | |
f1668764 PX |
121 | /* Current block being searched */ |
122 | RAMBlock *block; | |
123 | /* Current page to search from */ | |
124 | unsigned long page; | |
125 | /* Set once we wrap around */ | |
126 | bool complete_round; | |
f1668764 PX |
127 | /* Whether we're sending a host page */ |
128 | bool host_page_sending; | |
129 | /* The start/end of current host page. Invalid if host_page_sending==false */ | |
130 | unsigned long host_page_start; | |
131 | unsigned long host_page_end; | |
132 | }; | |
133 | typedef struct PageSearchStatus PageSearchStatus; | |
134 | ||
56e93d26 JQ |
135 | /* struct contains XBZRLE cache and a static page |
136 | used by the compression */ | |
137 | static struct { | |
138 | /* buffer used for XBZRLE encoding */ | |
139 | uint8_t *encoded_buf; | |
140 | /* buffer for storing page content */ | |
141 | uint8_t *current_buf; | |
142 | /* Cache for XBZRLE, Protected by lock. */ | |
143 | PageCache *cache; | |
144 | QemuMutex lock; | |
c00e0928 JQ |
145 | /* it will store a page full of zeros */ |
146 | uint8_t *zero_target_page; | |
f265e0e4 JQ |
147 | /* buffer used for XBZRLE decoding */ |
148 | uint8_t *decoded_buf; | |
56e93d26 JQ |
149 | } XBZRLE; |
150 | ||
56e93d26 JQ |
151 | static void XBZRLE_cache_lock(void) |
152 | { | |
87dca0c9 | 153 | if (migrate_xbzrle()) { |
56e93d26 | 154 | qemu_mutex_lock(&XBZRLE.lock); |
f4c51a6b | 155 | } |
56e93d26 JQ |
156 | } |
157 | ||
158 | static void XBZRLE_cache_unlock(void) | |
159 | { | |
87dca0c9 | 160 | if (migrate_xbzrle()) { |
56e93d26 | 161 | qemu_mutex_unlock(&XBZRLE.lock); |
f4c51a6b | 162 | } |
56e93d26 JQ |
163 | } |
164 | ||
3d0684b2 JQ |
165 | /** |
166 | * xbzrle_cache_resize: resize the xbzrle cache | |
167 | * | |
cbde7be9 | 168 | * This function is called from migrate_params_apply in main |
3d0684b2 JQ |
169 | * thread, possibly while a migration is in progress. A running |
170 | * migration may be using the cache and might finish during this call, | |
171 | * hence changes to the cache are protected by XBZRLE.lock(). | |
172 | * | |
c9dede2d | 173 | * Returns 0 for success or -1 for error |
3d0684b2 JQ |
174 | * |
175 | * @new_size: new cache size | |
8acabf69 | 176 | * @errp: set *errp if the check failed, with reason |
56e93d26 | 177 | */ |
8b9407a0 | 178 | int xbzrle_cache_resize(uint64_t new_size, Error **errp) |
56e93d26 JQ |
179 | { |
180 | PageCache *new_cache; | |
c9dede2d | 181 | int64_t ret = 0; |
56e93d26 | 182 | |
8acabf69 JQ |
183 | /* Check for truncation */ |
184 | if (new_size != (size_t)new_size) { | |
185 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", | |
186 | "exceeding address space"); | |
187 | return -1; | |
188 | } | |
189 | ||
2a313e5c JQ |
190 | if (new_size == migrate_xbzrle_cache_size()) { |
191 | /* nothing to do */ | |
c9dede2d | 192 | return 0; |
2a313e5c JQ |
193 | } |
194 | ||
56e93d26 JQ |
195 | XBZRLE_cache_lock(); |
196 | ||
197 | if (XBZRLE.cache != NULL) { | |
80f8dfde | 198 | new_cache = cache_init(new_size, TARGET_PAGE_SIZE, errp); |
56e93d26 | 199 | if (!new_cache) { |
56e93d26 JQ |
200 | ret = -1; |
201 | goto out; | |
202 | } | |
203 | ||
204 | cache_fini(XBZRLE.cache); | |
205 | XBZRLE.cache = new_cache; | |
206 | } | |
56e93d26 JQ |
207 | out: |
208 | XBZRLE_cache_unlock(); | |
209 | return ret; | |
210 | } | |
211 | ||
20123ee1 PX |
212 | static bool postcopy_preempt_active(void) |
213 | { | |
214 | return migrate_postcopy_preempt() && migration_in_postcopy(); | |
215 | } | |
216 | ||
f161c88a | 217 | bool migrate_ram_is_ignored(RAMBlock *block) |
fbd162e6 YK |
218 | { |
219 | return !qemu_ram_is_migratable(block) || | |
b0182e53 SS |
220 | (migrate_ignore_shared() && qemu_ram_is_shared(block) |
221 | && qemu_ram_is_named_file(block)); | |
fbd162e6 YK |
222 | } |
223 | ||
343f632c DDAG |
224 | #undef RAMBLOCK_FOREACH |
225 | ||
fbd162e6 YK |
226 | int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque) |
227 | { | |
228 | RAMBlock *block; | |
229 | int ret = 0; | |
230 | ||
89ac5a1d DDAG |
231 | RCU_READ_LOCK_GUARD(); |
232 | ||
fbd162e6 YK |
233 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
234 | ret = func(block, opaque); | |
235 | if (ret) { | |
236 | break; | |
237 | } | |
238 | } | |
fbd162e6 YK |
239 | return ret; |
240 | } | |
241 | ||
f9494614 AP |
242 | static void ramblock_recv_map_init(void) |
243 | { | |
244 | RAMBlock *rb; | |
245 | ||
fbd162e6 | 246 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
f9494614 AP |
247 | assert(!rb->receivedmap); |
248 | rb->receivedmap = bitmap_new(rb->max_length >> qemu_target_page_bits()); | |
249 | } | |
250 | } | |
251 | ||
252 | int ramblock_recv_bitmap_test(RAMBlock *rb, void *host_addr) | |
253 | { | |
254 | return test_bit(ramblock_recv_bitmap_offset(host_addr, rb), | |
255 | rb->receivedmap); | |
256 | } | |
257 | ||
1cba9f6e DDAG |
258 | bool ramblock_recv_bitmap_test_byte_offset(RAMBlock *rb, uint64_t byte_offset) |
259 | { | |
260 | return test_bit(byte_offset >> TARGET_PAGE_BITS, rb->receivedmap); | |
261 | } | |
262 | ||
f9494614 AP |
263 | void ramblock_recv_bitmap_set(RAMBlock *rb, void *host_addr) |
264 | { | |
265 | set_bit_atomic(ramblock_recv_bitmap_offset(host_addr, rb), rb->receivedmap); | |
266 | } | |
267 | ||
268 | void ramblock_recv_bitmap_set_range(RAMBlock *rb, void *host_addr, | |
269 | size_t nr) | |
270 | { | |
271 | bitmap_set_atomic(rb->receivedmap, | |
272 | ramblock_recv_bitmap_offset(host_addr, rb), | |
273 | nr); | |
274 | } | |
275 | ||
5ef7e26b YL |
276 | void ramblock_recv_bitmap_set_offset(RAMBlock *rb, uint64_t byte_offset) |
277 | { | |
278 | set_bit_atomic(byte_offset >> TARGET_PAGE_BITS, rb->receivedmap); | |
279 | } | |
a335debb PX |
280 | #define RAMBLOCK_RECV_BITMAP_ENDING (0x0123456789abcdefULL) |
281 | ||
282 | /* | |
283 | * Format: bitmap_size (8 bytes) + whole_bitmap (N bytes). | |
284 | * | |
285 | * Returns >0 if success with sent bytes, or <0 if error. | |
286 | */ | |
287 | int64_t ramblock_recv_bitmap_send(QEMUFile *file, | |
288 | const char *block_name) | |
289 | { | |
290 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
291 | unsigned long *le_bitmap, nbits; | |
292 | uint64_t size; | |
293 | ||
294 | if (!block) { | |
295 | error_report("%s: invalid block name: %s", __func__, block_name); | |
296 | return -1; | |
297 | } | |
298 | ||
898ba906 | 299 | nbits = block->postcopy_length >> TARGET_PAGE_BITS; |
a335debb PX |
300 | |
301 | /* | |
302 | * Make sure the tmp bitmap buffer is big enough, e.g., on 32bit | |
303 | * machines we may need 4 more bytes for padding (see below | |
304 | * comment). So extend it a bit before hand. | |
305 | */ | |
306 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
307 | ||
308 | /* | |
309 | * Always use little endian when sending the bitmap. This is | |
310 | * required that when source and destination VMs are not using the | |
3a4452d8 | 311 | * same endianness. (Note: big endian won't work.) |
a335debb PX |
312 | */ |
313 | bitmap_to_le(le_bitmap, block->receivedmap, nbits); | |
314 | ||
315 | /* Size of the bitmap, in bytes */ | |
a725ef9f | 316 | size = DIV_ROUND_UP(nbits, 8); |
a335debb PX |
317 | |
318 | /* | |
319 | * size is always aligned to 8 bytes for 64bit machines, but it | |
320 | * may not be true for 32bit machines. We need this padding to | |
321 | * make sure the migration can survive even between 32bit and | |
322 | * 64bit machines. | |
323 | */ | |
324 | size = ROUND_UP(size, 8); | |
325 | ||
326 | qemu_put_be64(file, size); | |
327 | qemu_put_buffer(file, (const uint8_t *)le_bitmap, size); | |
be07a0ed | 328 | g_free(le_bitmap); |
a335debb PX |
329 | /* |
330 | * Mark as an end, in case the middle part is screwed up due to | |
3a4452d8 | 331 | * some "mysterious" reason. |
a335debb PX |
332 | */ |
333 | qemu_put_be64(file, RAMBLOCK_RECV_BITMAP_ENDING); | |
be07a0ed JQ |
334 | int ret = qemu_fflush(file); |
335 | if (ret) { | |
336 | return ret; | |
a335debb PX |
337 | } |
338 | ||
339 | return size + sizeof(size); | |
340 | } | |
341 | ||
ec481c6c JQ |
342 | /* |
343 | * An outstanding page request, on the source, having been received | |
344 | * and queued | |
345 | */ | |
346 | struct RAMSrcPageRequest { | |
347 | RAMBlock *rb; | |
348 | hwaddr offset; | |
349 | hwaddr len; | |
350 | ||
351 | QSIMPLEQ_ENTRY(RAMSrcPageRequest) next_req; | |
352 | }; | |
353 | ||
6f37bb8b JQ |
354 | /* State of RAM for migration */ |
355 | struct RAMState { | |
f1668764 PX |
356 | /* |
357 | * PageSearchStatus structures for the channels when send pages. | |
358 | * Protected by the bitmap_mutex. | |
359 | */ | |
360 | PageSearchStatus pss[RAM_CHANNEL_MAX]; | |
278e2f55 AG |
361 | /* UFFD file descriptor, used in 'write-tracking' migration */ |
362 | int uffdio_fd; | |
8d80e195 JQ |
363 | /* total ram size in bytes */ |
364 | uint64_t ram_bytes_total; | |
6f37bb8b JQ |
365 | /* Last block that we have visited searching for dirty pages */ |
366 | RAMBlock *last_seen_block; | |
269ace29 JQ |
367 | /* Last dirty target page we have sent */ |
368 | ram_addr_t last_page; | |
6f37bb8b JQ |
369 | /* last ram version we have seen */ |
370 | uint32_t last_version; | |
8d820d6f JQ |
371 | /* How many times we have dirty too many pages */ |
372 | int dirty_rate_high_cnt; | |
f664da80 JQ |
373 | /* these variables are used for bitmap sync */ |
374 | /* last time we did a full bitmap_sync */ | |
375 | int64_t time_last_bitmap_sync; | |
eac74159 | 376 | /* bytes transferred at start_time */ |
c4bdf0cf | 377 | uint64_t bytes_xfer_prev; |
a66cd90c | 378 | /* number of dirty pages since start_time */ |
68908ed6 | 379 | uint64_t num_dirty_pages_period; |
b5833fde JQ |
380 | /* xbzrle misses since the beginning of the period */ |
381 | uint64_t xbzrle_cache_miss_prev; | |
e460a4b1 WW |
382 | /* Amount of xbzrle pages since the beginning of the period */ |
383 | uint64_t xbzrle_pages_prev; | |
384 | /* Amount of xbzrle encoded bytes since the beginning of the period */ | |
385 | uint64_t xbzrle_bytes_prev; | |
f3095cc8 JQ |
386 | /* Are we really using XBZRLE (e.g., after the first round). */ |
387 | bool xbzrle_started; | |
05931ec5 JQ |
388 | /* Are we on the last stage of migration */ |
389 | bool last_stage; | |
76e03000 | 390 | |
be8b02ed XG |
391 | /* total handled target pages at the beginning of period */ |
392 | uint64_t target_page_count_prev; | |
393 | /* total handled target pages since start */ | |
394 | uint64_t target_page_count; | |
9360447d | 395 | /* number of dirty bits in the bitmap */ |
2dfaf12e | 396 | uint64_t migration_dirty_pages; |
f1668764 PX |
397 | /* |
398 | * Protects: | |
399 | * - dirty/clear bitmap | |
400 | * - migration_dirty_pages | |
401 | * - pss structures | |
402 | */ | |
108cfae0 | 403 | QemuMutex bitmap_mutex; |
68a098f3 JQ |
404 | /* The RAMBlock used in the last src_page_requests */ |
405 | RAMBlock *last_req_rb; | |
ec481c6c JQ |
406 | /* Queue of outstanding page requests from the destination */ |
407 | QemuMutex src_page_req_mutex; | |
b58deb34 | 408 | QSIMPLEQ_HEAD(, RAMSrcPageRequest) src_page_requests; |
1015ff54 PX |
409 | |
410 | /* | |
411 | * This is only used when postcopy is in recovery phase, to communicate | |
412 | * between the migration thread and the return path thread on dirty | |
413 | * bitmap synchronizations. This field is unused in other stages of | |
414 | * RAM migration. | |
415 | */ | |
416 | unsigned int postcopy_bmap_sync_requested; | |
6f37bb8b JQ |
417 | }; |
418 | typedef struct RAMState RAMState; | |
419 | ||
53518d94 | 420 | static RAMState *ram_state; |
6f37bb8b | 421 | |
bd227060 WW |
422 | static NotifierWithReturnList precopy_notifier_list; |
423 | ||
a1fe28df PX |
424 | /* Whether postcopy has queued requests? */ |
425 | static bool postcopy_has_request(RAMState *rs) | |
426 | { | |
427 | return !QSIMPLEQ_EMPTY_ATOMIC(&rs->src_page_requests); | |
428 | } | |
429 | ||
bd227060 WW |
430 | void precopy_infrastructure_init(void) |
431 | { | |
432 | notifier_with_return_list_init(&precopy_notifier_list); | |
433 | } | |
434 | ||
435 | void precopy_add_notifier(NotifierWithReturn *n) | |
436 | { | |
437 | notifier_with_return_list_add(&precopy_notifier_list, n); | |
438 | } | |
439 | ||
440 | void precopy_remove_notifier(NotifierWithReturn *n) | |
441 | { | |
442 | notifier_with_return_remove(n); | |
443 | } | |
444 | ||
445 | int precopy_notify(PrecopyNotifyReason reason, Error **errp) | |
446 | { | |
447 | PrecopyNotifyData pnd; | |
448 | pnd.reason = reason; | |
bd227060 | 449 | |
be19d836 | 450 | return notifier_with_return_list_notify(&precopy_notifier_list, &pnd, errp); |
bd227060 WW |
451 | } |
452 | ||
9edabd4d | 453 | uint64_t ram_bytes_remaining(void) |
2f4fde93 | 454 | { |
bae416e5 DDAG |
455 | return ram_state ? (ram_state->migration_dirty_pages * TARGET_PAGE_SIZE) : |
456 | 0; | |
2f4fde93 JQ |
457 | } |
458 | ||
26a26069 | 459 | void ram_transferred_add(uint64_t bytes) |
4c2d0f6d | 460 | { |
ae680668 | 461 | if (runstate_is_running()) { |
aff3f660 | 462 | stat64_add(&mig_stats.precopy_bytes, bytes); |
ae680668 | 463 | } else if (migration_in_postcopy()) { |
aff3f660 | 464 | stat64_add(&mig_stats.postcopy_bytes, bytes); |
ae680668 | 465 | } else { |
aff3f660 | 466 | stat64_add(&mig_stats.downtime_bytes, bytes); |
ae680668 | 467 | } |
4c2d0f6d DE |
468 | } |
469 | ||
4010ba38 JQ |
470 | struct MigrationOps { |
471 | int (*ram_save_target_page)(RAMState *rs, PageSearchStatus *pss); | |
472 | }; | |
473 | typedef struct MigrationOps MigrationOps; | |
474 | ||
475 | MigrationOps *migration_ops; | |
476 | ||
93589827 PX |
477 | static int ram_save_host_page_urgent(PageSearchStatus *pss); |
478 | ||
ebd88a49 PX |
479 | /* NOTE: page is the PFN not real ram_addr_t. */ |
480 | static void pss_init(PageSearchStatus *pss, RAMBlock *rb, ram_addr_t page) | |
481 | { | |
482 | pss->block = rb; | |
483 | pss->page = page; | |
484 | pss->complete_round = false; | |
485 | } | |
486 | ||
93589827 PX |
487 | /* |
488 | * Check whether two PSSs are actively sending the same page. Return true | |
489 | * if it is, false otherwise. | |
490 | */ | |
491 | static bool pss_overlap(PageSearchStatus *pss1, PageSearchStatus *pss2) | |
492 | { | |
493 | return pss1->host_page_sending && pss2->host_page_sending && | |
494 | (pss1->host_page_start == pss2->host_page_start); | |
495 | } | |
496 | ||
56e93d26 | 497 | /** |
3d0684b2 | 498 | * save_page_header: write page header to wire |
56e93d26 JQ |
499 | * |
500 | * If this is the 1st block, it also writes the block identification | |
501 | * | |
3d0684b2 | 502 | * Returns the number of bytes written |
56e93d26 | 503 | * |
ec6f3ab9 | 504 | * @pss: current PSS channel status |
56e93d26 JQ |
505 | * @block: block that contains the page we want to send |
506 | * @offset: offset inside the block for the page | |
507 | * in the lower bits, it contains flags | |
508 | */ | |
37502df3 LS |
509 | static size_t save_page_header(PageSearchStatus *pss, QEMUFile *f, |
510 | RAMBlock *block, ram_addr_t offset) | |
56e93d26 | 511 | { |
9f5f380b | 512 | size_t size, len; |
ec6f3ab9 | 513 | bool same_block = (block == pss->last_sent_block); |
56e93d26 | 514 | |
10661f11 | 515 | if (same_block) { |
24795694 JQ |
516 | offset |= RAM_SAVE_FLAG_CONTINUE; |
517 | } | |
2bf3aa85 | 518 | qemu_put_be64(f, offset); |
56e93d26 JQ |
519 | size = 8; |
520 | ||
10661f11 | 521 | if (!same_block) { |
9f5f380b | 522 | len = strlen(block->idstr); |
2bf3aa85 JQ |
523 | qemu_put_byte(f, len); |
524 | qemu_put_buffer(f, (uint8_t *)block->idstr, len); | |
9f5f380b | 525 | size += 1 + len; |
ec6f3ab9 | 526 | pss->last_sent_block = block; |
56e93d26 JQ |
527 | } |
528 | return size; | |
529 | } | |
530 | ||
3d0684b2 | 531 | /** |
179a8080 | 532 | * mig_throttle_guest_down: throttle down the guest |
3d0684b2 JQ |
533 | * |
534 | * Reduce amount of guest cpu execution to hopefully slow down memory | |
535 | * writes. If guest dirty memory rate is reduced below the rate at | |
536 | * which we can transfer pages to the destination then we should be | |
537 | * able to complete migration. Some workloads dirty memory way too | |
538 | * fast and will not effectively converge, even with auto-converge. | |
070afca2 | 539 | */ |
cbbf8182 KZ |
540 | static void mig_throttle_guest_down(uint64_t bytes_dirty_period, |
541 | uint64_t bytes_dirty_threshold) | |
070afca2 | 542 | { |
2a8ec380 | 543 | uint64_t pct_initial = migrate_cpu_throttle_initial(); |
9605c2ac | 544 | uint64_t pct_increment = migrate_cpu_throttle_increment(); |
873f674c | 545 | bool pct_tailslow = migrate_cpu_throttle_tailslow(); |
24155bd0 | 546 | int pct_max = migrate_max_cpu_throttle(); |
070afca2 | 547 | |
cbbf8182 KZ |
548 | uint64_t throttle_now = cpu_throttle_get_percentage(); |
549 | uint64_t cpu_now, cpu_ideal, throttle_inc; | |
550 | ||
070afca2 JH |
551 | /* We have not started throttling yet. Let's start it. */ |
552 | if (!cpu_throttle_active()) { | |
553 | cpu_throttle_set(pct_initial); | |
554 | } else { | |
555 | /* Throttling already on, just increase the rate */ | |
cbbf8182 KZ |
556 | if (!pct_tailslow) { |
557 | throttle_inc = pct_increment; | |
558 | } else { | |
559 | /* Compute the ideal CPU percentage used by Guest, which may | |
560 | * make the dirty rate match the dirty rate threshold. */ | |
561 | cpu_now = 100 - throttle_now; | |
562 | cpu_ideal = cpu_now * (bytes_dirty_threshold * 1.0 / | |
563 | bytes_dirty_period); | |
564 | throttle_inc = MIN(cpu_now - cpu_ideal, pct_increment); | |
565 | } | |
566 | cpu_throttle_set(MIN(throttle_now + throttle_inc, pct_max)); | |
070afca2 JH |
567 | } |
568 | } | |
569 | ||
91fe9a8d RL |
570 | void mig_throttle_counter_reset(void) |
571 | { | |
572 | RAMState *rs = ram_state; | |
573 | ||
574 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
575 | rs->num_dirty_pages_period = 0; | |
897fd8bd | 576 | rs->bytes_xfer_prev = migration_transferred_bytes(); |
91fe9a8d RL |
577 | } |
578 | ||
3d0684b2 JQ |
579 | /** |
580 | * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache | |
581 | * | |
582 | * @current_addr: address for the zero page | |
583 | * | |
584 | * Update the xbzrle cache to reflect a page that's been sent as all 0. | |
56e93d26 JQ |
585 | * The important thing is that a stale (not-yet-0'd) page be replaced |
586 | * by the new data. | |
587 | * As a bonus, if the page wasn't in the cache it gets added so that | |
3d0684b2 | 588 | * when a small write is made into the 0'd page it gets XBZRLE sent. |
56e93d26 | 589 | */ |
8f47d4ee | 590 | static void xbzrle_cache_zero_page(ram_addr_t current_addr) |
56e93d26 | 591 | { |
56e93d26 JQ |
592 | /* We don't care if this fails to allocate a new cache page |
593 | * as long as it updated an old one */ | |
c00e0928 | 594 | cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page, |
aff3f660 | 595 | stat64_get(&mig_stats.dirty_sync_count)); |
56e93d26 JQ |
596 | } |
597 | ||
598 | #define ENCODING_FLAG_XBZRLE 0x1 | |
599 | ||
600 | /** | |
601 | * save_xbzrle_page: compress and send current page | |
602 | * | |
603 | * Returns: 1 means that we wrote the page | |
604 | * 0 means that page is identical to the one already sent | |
605 | * -1 means that xbzrle would be longer than normal | |
606 | * | |
5a987738 | 607 | * @rs: current RAM state |
ec6f3ab9 | 608 | * @pss: current PSS channel |
3d0684b2 JQ |
609 | * @current_data: pointer to the address of the page contents |
610 | * @current_addr: addr of the page | |
56e93d26 JQ |
611 | * @block: block that contains the page we want to send |
612 | * @offset: offset inside the block for the page | |
56e93d26 | 613 | */ |
ec6f3ab9 | 614 | static int save_xbzrle_page(RAMState *rs, PageSearchStatus *pss, |
61717ea9 PX |
615 | uint8_t **current_data, ram_addr_t current_addr, |
616 | RAMBlock *block, ram_addr_t offset) | |
56e93d26 JQ |
617 | { |
618 | int encoded_len = 0, bytes_xbzrle; | |
619 | uint8_t *prev_cached_page; | |
ec6f3ab9 | 620 | QEMUFile *file = pss->pss_channel; |
aff3f660 | 621 | uint64_t generation = stat64_get(&mig_stats.dirty_sync_count); |
56e93d26 | 622 | |
536b5a4e | 623 | if (!cache_is_cached(XBZRLE.cache, current_addr, generation)) { |
9360447d | 624 | xbzrle_counters.cache_miss++; |
05931ec5 | 625 | if (!rs->last_stage) { |
56e93d26 | 626 | if (cache_insert(XBZRLE.cache, current_addr, *current_data, |
536b5a4e | 627 | generation) == -1) { |
56e93d26 JQ |
628 | return -1; |
629 | } else { | |
630 | /* update *current_data when the page has been | |
631 | inserted into cache */ | |
632 | *current_data = get_cached_data(XBZRLE.cache, current_addr); | |
633 | } | |
634 | } | |
635 | return -1; | |
636 | } | |
637 | ||
e460a4b1 WW |
638 | /* |
639 | * Reaching here means the page has hit the xbzrle cache, no matter what | |
640 | * encoding result it is (normal encoding, overflow or skipping the page), | |
3a4452d8 | 641 | * count the page as encoded. This is used to calculate the encoding rate. |
e460a4b1 WW |
642 | * |
643 | * Example: 2 pages (8KB) being encoded, first page encoding generates 2KB, | |
644 | * 2nd page turns out to be skipped (i.e. no new bytes written to the | |
645 | * page), the overall encoding rate will be 8KB / 2KB = 4, which has the | |
646 | * skipped page included. In this way, the encoding rate can tell if the | |
647 | * guest page is good for xbzrle encoding. | |
648 | */ | |
649 | xbzrle_counters.pages++; | |
56e93d26 JQ |
650 | prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); |
651 | ||
652 | /* save current buffer into memory */ | |
653 | memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); | |
654 | ||
655 | /* XBZRLE encoding (if there is no overflow) */ | |
7ba7db9f RH |
656 | encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, |
657 | TARGET_PAGE_SIZE, XBZRLE.encoded_buf, | |
658 | TARGET_PAGE_SIZE); | |
ca353803 WY |
659 | |
660 | /* | |
661 | * Update the cache contents, so that it corresponds to the data | |
662 | * sent, in all cases except where we skip the page. | |
663 | */ | |
05931ec5 | 664 | if (!rs->last_stage && encoded_len != 0) { |
ca353803 WY |
665 | memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); |
666 | /* | |
667 | * In the case where we couldn't compress, ensure that the caller | |
668 | * sends the data from the cache, since the guest might have | |
669 | * changed the RAM since we copied it. | |
670 | */ | |
671 | *current_data = prev_cached_page; | |
672 | } | |
673 | ||
56e93d26 | 674 | if (encoded_len == 0) { |
55c4446b | 675 | trace_save_xbzrle_page_skipping(); |
56e93d26 JQ |
676 | return 0; |
677 | } else if (encoded_len == -1) { | |
55c4446b | 678 | trace_save_xbzrle_page_overflow(); |
9360447d | 679 | xbzrle_counters.overflow++; |
e460a4b1 | 680 | xbzrle_counters.bytes += TARGET_PAGE_SIZE; |
56e93d26 JQ |
681 | return -1; |
682 | } | |
683 | ||
56e93d26 | 684 | /* Send XBZRLE based compressed page */ |
37502df3 | 685 | bytes_xbzrle = save_page_header(pss, pss->pss_channel, block, |
204b88b8 | 686 | offset | RAM_SAVE_FLAG_XBZRLE); |
61717ea9 PX |
687 | qemu_put_byte(file, ENCODING_FLAG_XBZRLE); |
688 | qemu_put_be16(file, encoded_len); | |
689 | qemu_put_buffer(file, XBZRLE.encoded_buf, encoded_len); | |
56e93d26 | 690 | bytes_xbzrle += encoded_len + 1 + 2; |
e460a4b1 | 691 | /* |
0222111a | 692 | * The xbzrle encoded bytes don't count the 8 byte header with |
e460a4b1 WW |
693 | * RAM_SAVE_FLAG_CONTINUE. |
694 | */ | |
695 | xbzrle_counters.bytes += bytes_xbzrle - 8; | |
4c2d0f6d | 696 | ram_transferred_add(bytes_xbzrle); |
56e93d26 JQ |
697 | |
698 | return 1; | |
699 | } | |
700 | ||
3d0684b2 | 701 | /** |
d9e474ea | 702 | * pss_find_next_dirty: find the next dirty page of current ramblock |
f3f491fc | 703 | * |
d9e474ea PX |
704 | * This function updates pss->page to point to the next dirty page index |
705 | * within the ramblock to migrate, or the end of ramblock when nothing | |
706 | * found. Note that when pss->host_page_sending==true it means we're | |
707 | * during sending a host page, so we won't look for dirty page that is | |
708 | * outside the host page boundary. | |
3d0684b2 | 709 | * |
d9e474ea | 710 | * @pss: the current page search status |
f3f491fc | 711 | */ |
d9e474ea | 712 | static void pss_find_next_dirty(PageSearchStatus *pss) |
56e93d26 | 713 | { |
d9e474ea | 714 | RAMBlock *rb = pss->block; |
6b6712ef JQ |
715 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; |
716 | unsigned long *bitmap = rb->bmap; | |
56e93d26 | 717 | |
f161c88a | 718 | if (migrate_ram_is_ignored(rb)) { |
d9e474ea PX |
719 | /* Points directly to the end, so we know no dirty page */ |
720 | pss->page = size; | |
721 | return; | |
722 | } | |
723 | ||
724 | /* | |
725 | * If during sending a host page, only look for dirty pages within the | |
726 | * current host page being send. | |
727 | */ | |
728 | if (pss->host_page_sending) { | |
729 | assert(pss->host_page_end); | |
730 | size = MIN(size, pss->host_page_end); | |
b895de50 CLG |
731 | } |
732 | ||
d9e474ea | 733 | pss->page = find_next_bit(bitmap, size, pss->page); |
56e93d26 JQ |
734 | } |
735 | ||
1230a25f | 736 | static void migration_clear_memory_region_dirty_bitmap(RAMBlock *rb, |
3143577d WW |
737 | unsigned long page) |
738 | { | |
739 | uint8_t shift; | |
740 | hwaddr size, start; | |
741 | ||
742 | if (!rb->clear_bmap || !clear_bmap_test_and_clear(rb, page)) { | |
743 | return; | |
744 | } | |
745 | ||
746 | shift = rb->clear_bmap_shift; | |
747 | /* | |
748 | * CLEAR_BITMAP_SHIFT_MIN should always guarantee this... this | |
749 | * can make things easier sometimes since then start address | |
750 | * of the small chunk will always be 64 pages aligned so the | |
751 | * bitmap will always be aligned to unsigned long. We should | |
752 | * even be able to remove this restriction but I'm simply | |
753 | * keeping it. | |
754 | */ | |
755 | assert(shift >= 6); | |
756 | ||
757 | size = 1ULL << (TARGET_PAGE_BITS + shift); | |
7648297d | 758 | start = QEMU_ALIGN_DOWN((ram_addr_t)page << TARGET_PAGE_BITS, size); |
3143577d WW |
759 | trace_migration_bitmap_clear_dirty(rb->idstr, start, size, page); |
760 | memory_region_clear_dirty_bitmap(rb->mr, start, size); | |
761 | } | |
762 | ||
763 | static void | |
1230a25f | 764 | migration_clear_memory_region_dirty_bitmap_range(RAMBlock *rb, |
3143577d WW |
765 | unsigned long start, |
766 | unsigned long npages) | |
767 | { | |
768 | unsigned long i, chunk_pages = 1UL << rb->clear_bmap_shift; | |
769 | unsigned long chunk_start = QEMU_ALIGN_DOWN(start, chunk_pages); | |
770 | unsigned long chunk_end = QEMU_ALIGN_UP(start + npages, chunk_pages); | |
771 | ||
772 | /* | |
773 | * Clear pages from start to start + npages - 1, so the end boundary is | |
774 | * exclusive. | |
775 | */ | |
776 | for (i = chunk_start; i < chunk_end; i += chunk_pages) { | |
1230a25f | 777 | migration_clear_memory_region_dirty_bitmap(rb, i); |
3143577d WW |
778 | } |
779 | } | |
780 | ||
a6a83cef RL |
781 | /* |
782 | * colo_bitmap_find_diry:find contiguous dirty pages from start | |
783 | * | |
784 | * Returns the page offset within memory region of the start of the contiguout | |
785 | * dirty page | |
786 | * | |
787 | * @rs: current RAM state | |
788 | * @rb: RAMBlock where to search for dirty pages | |
789 | * @start: page where we start the search | |
790 | * @num: the number of contiguous dirty pages | |
791 | */ | |
792 | static inline | |
793 | unsigned long colo_bitmap_find_dirty(RAMState *rs, RAMBlock *rb, | |
794 | unsigned long start, unsigned long *num) | |
795 | { | |
796 | unsigned long size = rb->used_length >> TARGET_PAGE_BITS; | |
797 | unsigned long *bitmap = rb->bmap; | |
798 | unsigned long first, next; | |
799 | ||
800 | *num = 0; | |
801 | ||
f161c88a | 802 | if (migrate_ram_is_ignored(rb)) { |
a6a83cef RL |
803 | return size; |
804 | } | |
805 | ||
806 | first = find_next_bit(bitmap, size, start); | |
807 | if (first >= size) { | |
808 | return first; | |
809 | } | |
810 | next = find_next_zero_bit(bitmap, size, first + 1); | |
811 | assert(next >= first); | |
812 | *num = next - first; | |
813 | return first; | |
814 | } | |
815 | ||
06b10688 | 816 | static inline bool migration_bitmap_clear_dirty(RAMState *rs, |
f20e2865 JQ |
817 | RAMBlock *rb, |
818 | unsigned long page) | |
a82d593b DDAG |
819 | { |
820 | bool ret; | |
a82d593b | 821 | |
002cad6b PX |
822 | /* |
823 | * Clear dirty bitmap if needed. This _must_ be called before we | |
824 | * send any of the page in the chunk because we need to make sure | |
825 | * we can capture further page content changes when we sync dirty | |
826 | * log the next time. So as long as we are going to send any of | |
827 | * the page in the chunk we clear the remote dirty bitmap for all. | |
828 | * Clearing it earlier won't be a problem, but too late will. | |
829 | */ | |
1230a25f | 830 | migration_clear_memory_region_dirty_bitmap(rb, page); |
002cad6b | 831 | |
6b6712ef | 832 | ret = test_and_clear_bit(page, rb->bmap); |
a82d593b | 833 | if (ret) { |
0d8ec885 | 834 | rs->migration_dirty_pages--; |
a82d593b | 835 | } |
386a907b | 836 | |
a82d593b DDAG |
837 | return ret; |
838 | } | |
839 | ||
be39b4cd DH |
840 | static void dirty_bitmap_clear_section(MemoryRegionSection *section, |
841 | void *opaque) | |
842 | { | |
843 | const hwaddr offset = section->offset_within_region; | |
844 | const hwaddr size = int128_get64(section->size); | |
845 | const unsigned long start = offset >> TARGET_PAGE_BITS; | |
846 | const unsigned long npages = size >> TARGET_PAGE_BITS; | |
847 | RAMBlock *rb = section->mr->ram_block; | |
848 | uint64_t *cleared_bits = opaque; | |
849 | ||
850 | /* | |
851 | * We don't grab ram_state->bitmap_mutex because we expect to run | |
852 | * only when starting migration or during postcopy recovery where | |
853 | * we don't have concurrent access. | |
854 | */ | |
855 | if (!migration_in_postcopy() && !migrate_background_snapshot()) { | |
856 | migration_clear_memory_region_dirty_bitmap_range(rb, start, npages); | |
857 | } | |
858 | *cleared_bits += bitmap_count_one_with_offset(rb->bmap, start, npages); | |
859 | bitmap_clear(rb->bmap, start, npages); | |
860 | } | |
861 | ||
862 | /* | |
863 | * Exclude all dirty pages from migration that fall into a discarded range as | |
864 | * managed by a RamDiscardManager responsible for the mapped memory region of | |
865 | * the RAMBlock. Clear the corresponding bits in the dirty bitmaps. | |
866 | * | |
867 | * Discarded pages ("logically unplugged") have undefined content and must | |
868 | * not get migrated, because even reading these pages for migration might | |
869 | * result in undesired behavior. | |
870 | * | |
871 | * Returns the number of cleared bits in the RAMBlock dirty bitmap. | |
872 | * | |
873 | * Note: The result is only stable while migrating (precopy/postcopy). | |
874 | */ | |
875 | static uint64_t ramblock_dirty_bitmap_clear_discarded_pages(RAMBlock *rb) | |
876 | { | |
877 | uint64_t cleared_bits = 0; | |
878 | ||
879 | if (rb->mr && rb->bmap && memory_region_has_ram_discard_manager(rb->mr)) { | |
880 | RamDiscardManager *rdm = memory_region_get_ram_discard_manager(rb->mr); | |
881 | MemoryRegionSection section = { | |
882 | .mr = rb->mr, | |
883 | .offset_within_region = 0, | |
884 | .size = int128_make64(qemu_ram_get_used_length(rb)), | |
885 | }; | |
886 | ||
887 | ram_discard_manager_replay_discarded(rdm, §ion, | |
888 | dirty_bitmap_clear_section, | |
889 | &cleared_bits); | |
890 | } | |
891 | return cleared_bits; | |
892 | } | |
893 | ||
9470c5e0 DH |
894 | /* |
895 | * Check if a host-page aligned page falls into a discarded range as managed by | |
896 | * a RamDiscardManager responsible for the mapped memory region of the RAMBlock. | |
897 | * | |
898 | * Note: The result is only stable while migrating (precopy/postcopy). | |
899 | */ | |
900 | bool ramblock_page_is_discarded(RAMBlock *rb, ram_addr_t start) | |
901 | { | |
902 | if (rb->mr && memory_region_has_ram_discard_manager(rb->mr)) { | |
903 | RamDiscardManager *rdm = memory_region_get_ram_discard_manager(rb->mr); | |
904 | MemoryRegionSection section = { | |
905 | .mr = rb->mr, | |
906 | .offset_within_region = start, | |
907 | .size = int128_make64(qemu_ram_pagesize(rb)), | |
908 | }; | |
909 | ||
910 | return !ram_discard_manager_is_populated(rdm, §ion); | |
911 | } | |
912 | return false; | |
913 | } | |
914 | ||
267691b6 | 915 | /* Called with RCU critical section */ |
7a3e9571 | 916 | static void ramblock_sync_dirty_bitmap(RAMState *rs, RAMBlock *rb) |
56e93d26 | 917 | { |
fb613580 KZ |
918 | uint64_t new_dirty_pages = |
919 | cpu_physical_memory_sync_dirty_bitmap(rb, 0, rb->used_length); | |
920 | ||
921 | rs->migration_dirty_pages += new_dirty_pages; | |
922 | rs->num_dirty_pages_period += new_dirty_pages; | |
56e93d26 JQ |
923 | } |
924 | ||
3d0684b2 JQ |
925 | /** |
926 | * ram_pagesize_summary: calculate all the pagesizes of a VM | |
927 | * | |
928 | * Returns a summary bitmap of the page sizes of all RAMBlocks | |
929 | * | |
930 | * For VMs with just normal pages this is equivalent to the host page | |
931 | * size. If it's got some huge pages then it's the OR of all the | |
932 | * different page sizes. | |
e8ca1db2 DDAG |
933 | */ |
934 | uint64_t ram_pagesize_summary(void) | |
935 | { | |
936 | RAMBlock *block; | |
937 | uint64_t summary = 0; | |
938 | ||
fbd162e6 | 939 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
e8ca1db2 DDAG |
940 | summary |= block->page_size; |
941 | } | |
942 | ||
943 | return summary; | |
944 | } | |
945 | ||
aecbfe9c XG |
946 | uint64_t ram_get_total_transferred_pages(void) |
947 | { | |
aff3f660 JQ |
948 | return stat64_get(&mig_stats.normal_pages) + |
949 | stat64_get(&mig_stats.zero_pages) + | |
0222111a | 950 | xbzrle_counters.pages; |
aecbfe9c XG |
951 | } |
952 | ||
b734035b XG |
953 | static void migration_update_rates(RAMState *rs, int64_t end_time) |
954 | { | |
be8b02ed | 955 | uint64_t page_count = rs->target_page_count - rs->target_page_count_prev; |
b734035b XG |
956 | |
957 | /* calculate period counters */ | |
aff3f660 | 958 | stat64_set(&mig_stats.dirty_pages_rate, |
72f8e587 JQ |
959 | rs->num_dirty_pages_period * 1000 / |
960 | (end_time - rs->time_last_bitmap_sync)); | |
b734035b | 961 | |
be8b02ed | 962 | if (!page_count) { |
b734035b XG |
963 | return; |
964 | } | |
965 | ||
87dca0c9 | 966 | if (migrate_xbzrle()) { |
e460a4b1 WW |
967 | double encoded_size, unencoded_size; |
968 | ||
b734035b | 969 | xbzrle_counters.cache_miss_rate = (double)(xbzrle_counters.cache_miss - |
be8b02ed | 970 | rs->xbzrle_cache_miss_prev) / page_count; |
b734035b | 971 | rs->xbzrle_cache_miss_prev = xbzrle_counters.cache_miss; |
e460a4b1 WW |
972 | unencoded_size = (xbzrle_counters.pages - rs->xbzrle_pages_prev) * |
973 | TARGET_PAGE_SIZE; | |
974 | encoded_size = xbzrle_counters.bytes - rs->xbzrle_bytes_prev; | |
92271402 | 975 | if (xbzrle_counters.pages == rs->xbzrle_pages_prev || !encoded_size) { |
e460a4b1 | 976 | xbzrle_counters.encoding_rate = 0; |
e460a4b1 WW |
977 | } else { |
978 | xbzrle_counters.encoding_rate = unencoded_size / encoded_size; | |
979 | } | |
980 | rs->xbzrle_pages_prev = xbzrle_counters.pages; | |
981 | rs->xbzrle_bytes_prev = xbzrle_counters.bytes; | |
b734035b XG |
982 | } |
983 | } | |
984 | ||
acac51ba HH |
985 | /* |
986 | * Enable dirty-limit to throttle down the guest | |
987 | */ | |
988 | static void migration_dirty_limit_guest(void) | |
989 | { | |
990 | /* | |
991 | * dirty page rate quota for all vCPUs fetched from | |
992 | * migration parameter 'vcpu_dirty_limit' | |
993 | */ | |
994 | static int64_t quota_dirtyrate; | |
995 | MigrationState *s = migrate_get_current(); | |
996 | ||
997 | /* | |
998 | * If dirty limit already enabled and migration parameter | |
999 | * vcpu-dirty-limit untouched. | |
1000 | */ | |
1001 | if (dirtylimit_in_service() && | |
1002 | quota_dirtyrate == s->parameters.vcpu_dirty_limit) { | |
1003 | return; | |
1004 | } | |
1005 | ||
1006 | quota_dirtyrate = s->parameters.vcpu_dirty_limit; | |
1007 | ||
1008 | /* | |
1009 | * Set all vCPU a quota dirtyrate, note that the second | |
1010 | * parameter will be ignored if setting all vCPU for the vm | |
1011 | */ | |
1012 | qmp_set_vcpu_dirty_limit(false, -1, quota_dirtyrate, NULL); | |
1013 | trace_migration_dirty_limit_guest(quota_dirtyrate); | |
1014 | } | |
1015 | ||
dc14a470 KZ |
1016 | static void migration_trigger_throttle(RAMState *rs) |
1017 | { | |
6499efdb | 1018 | uint64_t threshold = migrate_throttle_trigger_threshold(); |
23b7576d | 1019 | uint64_t bytes_xfer_period = |
897fd8bd | 1020 | migration_transferred_bytes() - rs->bytes_xfer_prev; |
dc14a470 KZ |
1021 | uint64_t bytes_dirty_period = rs->num_dirty_pages_period * TARGET_PAGE_SIZE; |
1022 | uint64_t bytes_dirty_threshold = bytes_xfer_period * threshold / 100; | |
1023 | ||
310ad562 HH |
1024 | /* |
1025 | * The following detection logic can be refined later. For now: | |
1026 | * Check to see if the ratio between dirtied bytes and the approx. | |
1027 | * amount of bytes that just got transferred since the last time | |
1028 | * we were in this routine reaches the threshold. If that happens | |
1029 | * twice, start or increase throttling. | |
1030 | */ | |
1031 | if ((bytes_dirty_period > bytes_dirty_threshold) && | |
1032 | (++rs->dirty_rate_high_cnt >= 2)) { | |
1033 | rs->dirty_rate_high_cnt = 0; | |
1034 | if (migrate_auto_converge()) { | |
dc14a470 | 1035 | trace_migration_throttle(); |
cbbf8182 KZ |
1036 | mig_throttle_guest_down(bytes_dirty_period, |
1037 | bytes_dirty_threshold); | |
acac51ba HH |
1038 | } else if (migrate_dirty_limit()) { |
1039 | migration_dirty_limit_guest(); | |
dc14a470 KZ |
1040 | } |
1041 | } | |
1042 | } | |
1043 | ||
1e493be5 | 1044 | static void migration_bitmap_sync(RAMState *rs, bool last_stage) |
56e93d26 JQ |
1045 | { |
1046 | RAMBlock *block; | |
56e93d26 | 1047 | int64_t end_time; |
56e93d26 | 1048 | |
aff3f660 | 1049 | stat64_add(&mig_stats.dirty_sync_count, 1); |
56e93d26 | 1050 | |
f664da80 JQ |
1051 | if (!rs->time_last_bitmap_sync) { |
1052 | rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56e93d26 JQ |
1053 | } |
1054 | ||
1055 | trace_migration_bitmap_sync_start(); | |
1e493be5 | 1056 | memory_global_dirty_log_sync(last_stage); |
56e93d26 | 1057 | |
62663f08 WG |
1058 | WITH_QEMU_LOCK_GUARD(&rs->bitmap_mutex) { |
1059 | WITH_RCU_READ_LOCK_GUARD() { | |
1060 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
1061 | ramblock_sync_dirty_bitmap(rs, block); | |
1062 | } | |
1063 | stat64_set(&mig_stats.dirty_bytes_last_sync, ram_bytes_remaining()); | |
89ac5a1d | 1064 | } |
56e93d26 | 1065 | } |
56e93d26 | 1066 | |
9458a9a1 | 1067 | memory_global_after_dirty_log_sync(); |
a66cd90c | 1068 | trace_migration_bitmap_sync_end(rs->num_dirty_pages_period); |
1ffb5dfd | 1069 | |
56e93d26 JQ |
1070 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1071 | ||
1072 | /* more than 1 second = 1000 millisecons */ | |
f664da80 | 1073 | if (end_time > rs->time_last_bitmap_sync + 1000) { |
dc14a470 | 1074 | migration_trigger_throttle(rs); |
070afca2 | 1075 | |
b734035b XG |
1076 | migration_update_rates(rs, end_time); |
1077 | ||
be8b02ed | 1078 | rs->target_page_count_prev = rs->target_page_count; |
d693c6f1 FF |
1079 | |
1080 | /* reset period counters */ | |
f664da80 | 1081 | rs->time_last_bitmap_sync = end_time; |
a66cd90c | 1082 | rs->num_dirty_pages_period = 0; |
897fd8bd | 1083 | rs->bytes_xfer_prev = migration_transferred_bytes(); |
56e93d26 | 1084 | } |
b890902c | 1085 | if (migrate_events()) { |
aff3f660 | 1086 | uint64_t generation = stat64_get(&mig_stats.dirty_sync_count); |
536b5a4e | 1087 | qapi_event_send_migration_pass(generation); |
4addcd4f | 1088 | } |
56e93d26 JQ |
1089 | } |
1090 | ||
1e493be5 | 1091 | static void migration_bitmap_sync_precopy(RAMState *rs, bool last_stage) |
bd227060 WW |
1092 | { |
1093 | Error *local_err = NULL; | |
1094 | ||
1095 | /* | |
1096 | * The current notifier usage is just an optimization to migration, so we | |
1097 | * don't stop the normal migration process in the error case. | |
1098 | */ | |
1099 | if (precopy_notify(PRECOPY_NOTIFY_BEFORE_BITMAP_SYNC, &local_err)) { | |
1100 | error_report_err(local_err); | |
b4a1733c | 1101 | local_err = NULL; |
bd227060 WW |
1102 | } |
1103 | ||
1e493be5 | 1104 | migration_bitmap_sync(rs, last_stage); |
bd227060 WW |
1105 | |
1106 | if (precopy_notify(PRECOPY_NOTIFY_AFTER_BITMAP_SYNC, &local_err)) { | |
1107 | error_report_err(local_err); | |
1108 | } | |
1109 | } | |
1110 | ||
a4dbaf8e | 1111 | void ram_release_page(const char *rbname, uint64_t offset) |
47fe16ff JQ |
1112 | { |
1113 | if (!migrate_release_ram() || !migration_in_postcopy()) { | |
1114 | return; | |
1115 | } | |
1116 | ||
1117 | ram_discard_range(rbname, offset, TARGET_PAGE_SIZE); | |
1118 | } | |
1119 | ||
56e93d26 | 1120 | /** |
3d0684b2 | 1121 | * save_zero_page: send the zero page to the stream |
56e93d26 | 1122 | * |
3d0684b2 | 1123 | * Returns the number of pages written. |
56e93d26 | 1124 | * |
ccc09db8 | 1125 | * @rs: current RAM state |
ec6f3ab9 | 1126 | * @pss: current PSS channel |
56e93d26 | 1127 | * @offset: offset inside the block for the page |
56e93d26 | 1128 | */ |
e8e4e7ac | 1129 | static int save_zero_page(RAMState *rs, PageSearchStatus *pss, |
61717ea9 | 1130 | ram_addr_t offset) |
56e93d26 | 1131 | { |
e8e4e7ac | 1132 | uint8_t *p = pss->block->host + offset; |
8697eb85 FR |
1133 | QEMUFile *file = pss->pss_channel; |
1134 | int len = 0; | |
56e93d26 | 1135 | |
5fdbb1df HX |
1136 | if (migrate_zero_page_detection() == ZERO_PAGE_DETECTION_NONE) { |
1137 | return 0; | |
1138 | } | |
1139 | ||
8697eb85 FR |
1140 | if (!buffer_is_zero(p, TARGET_PAGE_SIZE)) { |
1141 | return 0; | |
56e93d26 | 1142 | } |
ccc09db8 | 1143 | |
c2d5c4a7 FR |
1144 | stat64_add(&mig_stats.zero_pages, 1); |
1145 | ||
1146 | if (migrate_mapped_ram()) { | |
1147 | /* zero pages are not transferred with mapped-ram */ | |
f427d90b | 1148 | clear_bit_atomic(offset >> TARGET_PAGE_BITS, pss->block->file_bmap); |
c2d5c4a7 FR |
1149 | return 1; |
1150 | } | |
1151 | ||
e8e4e7ac | 1152 | len += save_page_header(pss, file, pss->block, offset | RAM_SAVE_FLAG_ZERO); |
8697eb85 FR |
1153 | qemu_put_byte(file, 0); |
1154 | len += 1; | |
e8e4e7ac | 1155 | ram_release_page(pss->block->idstr, offset); |
ccc09db8 FR |
1156 | ram_transferred_add(len); |
1157 | ||
1158 | /* | |
1159 | * Must let xbzrle know, otherwise a previous (now 0'd) cached | |
1160 | * page would be stale. | |
1161 | */ | |
1162 | if (rs->xbzrle_started) { | |
1163 | XBZRLE_cache_lock(); | |
e8e4e7ac | 1164 | xbzrle_cache_zero_page(pss->block->offset + offset); |
ccc09db8 FR |
1165 | XBZRLE_cache_unlock(); |
1166 | } | |
1167 | ||
8697eb85 | 1168 | return len; |
56e93d26 JQ |
1169 | } |
1170 | ||
059ff0fb XG |
1171 | /* |
1172 | * @pages: the number of pages written by the control path, | |
1173 | * < 0 - error | |
1174 | * > 0 - number of pages written | |
1175 | * | |
1176 | * Return true if the pages has been saved, otherwise false is returned. | |
1177 | */ | |
944853c2 | 1178 | static bool control_save_page(PageSearchStatus *pss, |
61717ea9 | 1179 | ram_addr_t offset, int *pages) |
059ff0fb | 1180 | { |
059ff0fb XG |
1181 | int ret; |
1182 | ||
944853c2 | 1183 | ret = rdma_control_save_page(pss->pss_channel, pss->block->offset, offset, |
e493008d | 1184 | TARGET_PAGE_SIZE); |
059ff0fb XG |
1185 | if (ret == RAM_SAVE_CONTROL_NOT_SUPP) { |
1186 | return false; | |
1187 | } | |
1188 | ||
059ff0fb | 1189 | if (ret == RAM_SAVE_CONTROL_DELAYED) { |
9c53d369 | 1190 | *pages = 1; |
059ff0fb XG |
1191 | return true; |
1192 | } | |
9c53d369 | 1193 | *pages = ret; |
059ff0fb XG |
1194 | return true; |
1195 | } | |
1196 | ||
65dacaa0 XG |
1197 | /* |
1198 | * directly send the page to the stream | |
1199 | * | |
1200 | * Returns the number of pages written. | |
1201 | * | |
ec6f3ab9 | 1202 | * @pss: current PSS channel |
65dacaa0 XG |
1203 | * @block: block that contains the page we want to send |
1204 | * @offset: offset inside the block for the page | |
1205 | * @buf: the page to be sent | |
1206 | * @async: send to page asyncly | |
1207 | */ | |
ec6f3ab9 | 1208 | static int save_normal_page(PageSearchStatus *pss, RAMBlock *block, |
61717ea9 | 1209 | ram_addr_t offset, uint8_t *buf, bool async) |
65dacaa0 | 1210 | { |
ec6f3ab9 PX |
1211 | QEMUFile *file = pss->pss_channel; |
1212 | ||
c2d5c4a7 FR |
1213 | if (migrate_mapped_ram()) { |
1214 | qemu_put_buffer_at(file, buf, TARGET_PAGE_SIZE, | |
1215 | block->pages_offset + offset); | |
1216 | set_bit(offset >> TARGET_PAGE_BITS, block->file_bmap); | |
65dacaa0 | 1217 | } else { |
c2d5c4a7 FR |
1218 | ram_transferred_add(save_page_header(pss, pss->pss_channel, block, |
1219 | offset | RAM_SAVE_FLAG_PAGE)); | |
1220 | if (async) { | |
1221 | qemu_put_buffer_async(file, buf, TARGET_PAGE_SIZE, | |
1222 | migrate_release_ram() && | |
1223 | migration_in_postcopy()); | |
1224 | } else { | |
1225 | qemu_put_buffer(file, buf, TARGET_PAGE_SIZE); | |
1226 | } | |
65dacaa0 | 1227 | } |
4c2d0f6d | 1228 | ram_transferred_add(TARGET_PAGE_SIZE); |
aff3f660 | 1229 | stat64_add(&mig_stats.normal_pages, 1); |
65dacaa0 XG |
1230 | return 1; |
1231 | } | |
1232 | ||
56e93d26 | 1233 | /** |
3d0684b2 | 1234 | * ram_save_page: send the given page to the stream |
56e93d26 | 1235 | * |
3d0684b2 | 1236 | * Returns the number of pages written. |
3fd3c4b3 DDAG |
1237 | * < 0 - error |
1238 | * >=0 - Number of pages written - this might legally be 0 | |
1239 | * if xbzrle noticed the page was the same. | |
56e93d26 | 1240 | * |
6f37bb8b | 1241 | * @rs: current RAM state |
56e93d26 JQ |
1242 | * @block: block that contains the page we want to send |
1243 | * @offset: offset inside the block for the page | |
56e93d26 | 1244 | */ |
05931ec5 | 1245 | static int ram_save_page(RAMState *rs, PageSearchStatus *pss) |
56e93d26 JQ |
1246 | { |
1247 | int pages = -1; | |
56e93d26 | 1248 | uint8_t *p; |
56e93d26 | 1249 | bool send_async = true; |
a08f6890 | 1250 | RAMBlock *block = pss->block; |
8bba004c | 1251 | ram_addr_t offset = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS; |
059ff0fb | 1252 | ram_addr_t current_addr = block->offset + offset; |
56e93d26 | 1253 | |
2f68e399 | 1254 | p = block->host + offset; |
1db9d8e5 | 1255 | trace_ram_save_page(block->idstr, (uint64_t)offset, p); |
56e93d26 | 1256 | |
56e93d26 | 1257 | XBZRLE_cache_lock(); |
f3095cc8 | 1258 | if (rs->xbzrle_started && !migration_in_postcopy()) { |
ec6f3ab9 | 1259 | pages = save_xbzrle_page(rs, pss, &p, current_addr, |
61717ea9 | 1260 | block, offset); |
05931ec5 | 1261 | if (!rs->last_stage) { |
059ff0fb XG |
1262 | /* Can't send this cached data async, since the cache page |
1263 | * might get updated before it gets to the wire | |
56e93d26 | 1264 | */ |
059ff0fb | 1265 | send_async = false; |
56e93d26 JQ |
1266 | } |
1267 | } | |
1268 | ||
1269 | /* XBZRLE overflow or normal page */ | |
1270 | if (pages == -1) { | |
ec6f3ab9 | 1271 | pages = save_normal_page(pss, block, offset, p, send_async); |
56e93d26 JQ |
1272 | } |
1273 | ||
1274 | XBZRLE_cache_unlock(); | |
1275 | ||
1276 | return pages; | |
1277 | } | |
1278 | ||
9346fa18 | 1279 | static int ram_save_multifd_page(RAMBlock *block, ram_addr_t offset) |
b9ee2f7d | 1280 | { |
d6556d17 | 1281 | if (!multifd_queue_page(block, offset)) { |
713f762a IR |
1282 | return -1; |
1283 | } | |
b9ee2f7d JQ |
1284 | |
1285 | return 1; | |
1286 | } | |
1287 | ||
3e81763e | 1288 | |
31e2ac74 JQ |
1289 | #define PAGE_ALL_CLEAN 0 |
1290 | #define PAGE_TRY_AGAIN 1 | |
1291 | #define PAGE_DIRTY_FOUND 2 | |
3d0684b2 JQ |
1292 | /** |
1293 | * find_dirty_block: find the next dirty page and update any state | |
1294 | * associated with the search process. | |
b9e60928 | 1295 | * |
31e2ac74 | 1296 | * Returns: |
294e5a40 | 1297 | * <0: An error happened |
31e2ac74 JQ |
1298 | * PAGE_ALL_CLEAN: no dirty page found, give up |
1299 | * PAGE_TRY_AGAIN: no dirty page found, retry for next block | |
1300 | * PAGE_DIRTY_FOUND: dirty page found | |
b9e60928 | 1301 | * |
6f37bb8b | 1302 | * @rs: current RAM state |
3d0684b2 JQ |
1303 | * @pss: data about the state of the current dirty page scan |
1304 | * @again: set to false if the search has scanned the whole of RAM | |
b9e60928 | 1305 | */ |
31e2ac74 | 1306 | static int find_dirty_block(RAMState *rs, PageSearchStatus *pss) |
b9e60928 | 1307 | { |
d9e474ea PX |
1308 | /* Update pss->page for the next dirty bit in ramblock */ |
1309 | pss_find_next_dirty(pss); | |
1310 | ||
6f37bb8b | 1311 | if (pss->complete_round && pss->block == rs->last_seen_block && |
a935e30f | 1312 | pss->page >= rs->last_page) { |
b9e60928 DDAG |
1313 | /* |
1314 | * We've been once around the RAM and haven't found anything. | |
1315 | * Give up. | |
1316 | */ | |
31e2ac74 | 1317 | return PAGE_ALL_CLEAN; |
b9e60928 | 1318 | } |
542147f4 DH |
1319 | if (!offset_in_ramblock(pss->block, |
1320 | ((ram_addr_t)pss->page) << TARGET_PAGE_BITS)) { | |
b9e60928 | 1321 | /* Didn't find anything in this RAM Block */ |
a935e30f | 1322 | pss->page = 0; |
b9e60928 DDAG |
1323 | pss->block = QLIST_NEXT_RCU(pss->block, next); |
1324 | if (!pss->block) { | |
d4f34485 | 1325 | if (migrate_multifd() && |
9d01778a FR |
1326 | (!migrate_multifd_flush_after_each_section() || |
1327 | migrate_mapped_ram())) { | |
294e5a40 | 1328 | QEMUFile *f = rs->pss[RAM_CHANNEL_PRECOPY].pss_channel; |
9346fa18 | 1329 | int ret = multifd_send_sync_main(); |
294e5a40 JQ |
1330 | if (ret < 0) { |
1331 | return ret; | |
1332 | } | |
9d01778a FR |
1333 | |
1334 | if (!migrate_mapped_ram()) { | |
1335 | qemu_put_be64(f, RAM_SAVE_FLAG_MULTIFD_FLUSH); | |
1336 | qemu_fflush(f); | |
1337 | } | |
294e5a40 | 1338 | } |
48df9d80 | 1339 | |
b9e60928 DDAG |
1340 | /* Hit the end of the list */ |
1341 | pss->block = QLIST_FIRST_RCU(&ram_list.blocks); | |
1342 | /* Flag that we've looped */ | |
1343 | pss->complete_round = true; | |
1a373522 | 1344 | /* After the first round, enable XBZRLE. */ |
87dca0c9 | 1345 | if (migrate_xbzrle()) { |
f3095cc8 | 1346 | rs->xbzrle_started = true; |
1a373522 | 1347 | } |
b9e60928 DDAG |
1348 | } |
1349 | /* Didn't find anything this time, but try again on the new block */ | |
31e2ac74 | 1350 | return PAGE_TRY_AGAIN; |
b9e60928 | 1351 | } else { |
31e2ac74 JQ |
1352 | /* We've found something */ |
1353 | return PAGE_DIRTY_FOUND; | |
b9e60928 DDAG |
1354 | } |
1355 | } | |
1356 | ||
3d0684b2 JQ |
1357 | /** |
1358 | * unqueue_page: gets a page of the queue | |
1359 | * | |
a82d593b | 1360 | * Helper for 'get_queued_page' - gets a page off the queue |
a82d593b | 1361 | * |
3d0684b2 JQ |
1362 | * Returns the block of the page (or NULL if none available) |
1363 | * | |
ec481c6c | 1364 | * @rs: current RAM state |
3d0684b2 | 1365 | * @offset: used to return the offset within the RAMBlock |
a82d593b | 1366 | */ |
f20e2865 | 1367 | static RAMBlock *unqueue_page(RAMState *rs, ram_addr_t *offset) |
a82d593b | 1368 | { |
a1fe28df | 1369 | struct RAMSrcPageRequest *entry; |
a82d593b DDAG |
1370 | RAMBlock *block = NULL; |
1371 | ||
a1fe28df | 1372 | if (!postcopy_has_request(rs)) { |
ae526e32 XG |
1373 | return NULL; |
1374 | } | |
1375 | ||
6e8a355d | 1376 | QEMU_LOCK_GUARD(&rs->src_page_req_mutex); |
a1fe28df PX |
1377 | |
1378 | /* | |
1379 | * This should _never_ change even after we take the lock, because no one | |
1380 | * should be taking anything off the request list other than us. | |
1381 | */ | |
1382 | assert(postcopy_has_request(rs)); | |
1383 | ||
1384 | entry = QSIMPLEQ_FIRST(&rs->src_page_requests); | |
1385 | block = entry->rb; | |
1386 | *offset = entry->offset; | |
1387 | ||
777f53c7 TH |
1388 | if (entry->len > TARGET_PAGE_SIZE) { |
1389 | entry->len -= TARGET_PAGE_SIZE; | |
1390 | entry->offset += TARGET_PAGE_SIZE; | |
a1fe28df PX |
1391 | } else { |
1392 | memory_region_unref(block->mr); | |
1393 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); | |
1394 | g_free(entry); | |
1395 | migration_consume_urgent_request(); | |
a82d593b | 1396 | } |
a82d593b DDAG |
1397 | |
1398 | return block; | |
1399 | } | |
1400 | ||
278e2f55 AG |
1401 | #if defined(__linux__) |
1402 | /** | |
1403 | * poll_fault_page: try to get next UFFD write fault page and, if pending fault | |
1404 | * is found, return RAM block pointer and page offset | |
1405 | * | |
1406 | * Returns pointer to the RAMBlock containing faulting page, | |
1407 | * NULL if no write faults are pending | |
1408 | * | |
1409 | * @rs: current RAM state | |
1410 | * @offset: page offset from the beginning of the block | |
1411 | */ | |
1412 | static RAMBlock *poll_fault_page(RAMState *rs, ram_addr_t *offset) | |
1413 | { | |
1414 | struct uffd_msg uffd_msg; | |
1415 | void *page_address; | |
82ea3e3b | 1416 | RAMBlock *block; |
278e2f55 AG |
1417 | int res; |
1418 | ||
1419 | if (!migrate_background_snapshot()) { | |
1420 | return NULL; | |
1421 | } | |
1422 | ||
1423 | res = uffd_read_events(rs->uffdio_fd, &uffd_msg, 1); | |
1424 | if (res <= 0) { | |
1425 | return NULL; | |
1426 | } | |
1427 | ||
1428 | page_address = (void *)(uintptr_t) uffd_msg.arg.pagefault.address; | |
82ea3e3b AG |
1429 | block = qemu_ram_block_from_host(page_address, false, offset); |
1430 | assert(block && (block->flags & RAM_UF_WRITEPROTECT) != 0); | |
1431 | return block; | |
278e2f55 AG |
1432 | } |
1433 | ||
1434 | /** | |
1435 | * ram_save_release_protection: release UFFD write protection after | |
1436 | * a range of pages has been saved | |
1437 | * | |
1438 | * @rs: current RAM state | |
1439 | * @pss: page-search-status structure | |
1440 | * @start_page: index of the first page in the range relative to pss->block | |
1441 | * | |
1442 | * Returns 0 on success, negative value in case of an error | |
1443 | */ | |
1444 | static int ram_save_release_protection(RAMState *rs, PageSearchStatus *pss, | |
1445 | unsigned long start_page) | |
1446 | { | |
1447 | int res = 0; | |
1448 | ||
1449 | /* Check if page is from UFFD-managed region. */ | |
1450 | if (pss->block->flags & RAM_UF_WRITEPROTECT) { | |
1451 | void *page_address = pss->block->host + (start_page << TARGET_PAGE_BITS); | |
258f5c98 | 1452 | uint64_t run_length = (pss->page - start_page) << TARGET_PAGE_BITS; |
278e2f55 AG |
1453 | |
1454 | /* Flush async buffers before un-protect. */ | |
61717ea9 | 1455 | qemu_fflush(pss->pss_channel); |
278e2f55 AG |
1456 | /* Un-protect memory range. */ |
1457 | res = uffd_change_protection(rs->uffdio_fd, page_address, run_length, | |
1458 | false, false); | |
1459 | } | |
1460 | ||
1461 | return res; | |
1462 | } | |
1463 | ||
1464 | /* ram_write_tracking_available: check if kernel supports required UFFD features | |
1465 | * | |
1466 | * Returns true if supports, false otherwise | |
1467 | */ | |
1468 | bool ram_write_tracking_available(void) | |
1469 | { | |
1470 | uint64_t uffd_features; | |
1471 | int res; | |
1472 | ||
1473 | res = uffd_query_features(&uffd_features); | |
1474 | return (res == 0 && | |
1475 | (uffd_features & UFFD_FEATURE_PAGEFAULT_FLAG_WP) != 0); | |
1476 | } | |
1477 | ||
1478 | /* ram_write_tracking_compatible: check if guest configuration is | |
1479 | * compatible with 'write-tracking' | |
1480 | * | |
1481 | * Returns true if compatible, false otherwise | |
1482 | */ | |
1483 | bool ram_write_tracking_compatible(void) | |
1484 | { | |
1485 | const uint64_t uffd_ioctls_mask = BIT(_UFFDIO_WRITEPROTECT); | |
1486 | int uffd_fd; | |
82ea3e3b | 1487 | RAMBlock *block; |
278e2f55 AG |
1488 | bool ret = false; |
1489 | ||
1490 | /* Open UFFD file descriptor */ | |
1491 | uffd_fd = uffd_create_fd(UFFD_FEATURE_PAGEFAULT_FLAG_WP, false); | |
1492 | if (uffd_fd < 0) { | |
1493 | return false; | |
1494 | } | |
1495 | ||
1496 | RCU_READ_LOCK_GUARD(); | |
1497 | ||
82ea3e3b | 1498 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
278e2f55 AG |
1499 | uint64_t uffd_ioctls; |
1500 | ||
1501 | /* Nothing to do with read-only and MMIO-writable regions */ | |
82ea3e3b | 1502 | if (block->mr->readonly || block->mr->rom_device) { |
278e2f55 AG |
1503 | continue; |
1504 | } | |
1505 | /* Try to register block memory via UFFD-IO to track writes */ | |
82ea3e3b | 1506 | if (uffd_register_memory(uffd_fd, block->host, block->max_length, |
278e2f55 AG |
1507 | UFFDIO_REGISTER_MODE_WP, &uffd_ioctls)) { |
1508 | goto out; | |
1509 | } | |
1510 | if ((uffd_ioctls & uffd_ioctls_mask) != uffd_ioctls_mask) { | |
1511 | goto out; | |
1512 | } | |
1513 | } | |
1514 | ret = true; | |
1515 | ||
1516 | out: | |
1517 | uffd_close_fd(uffd_fd); | |
1518 | return ret; | |
1519 | } | |
1520 | ||
f7b9dcfb DH |
1521 | static inline void populate_read_range(RAMBlock *block, ram_addr_t offset, |
1522 | ram_addr_t size) | |
1523 | { | |
5f19a449 DH |
1524 | const ram_addr_t end = offset + size; |
1525 | ||
f7b9dcfb DH |
1526 | /* |
1527 | * We read one byte of each page; this will preallocate page tables if | |
1528 | * required and populate the shared zeropage on MAP_PRIVATE anonymous memory | |
1529 | * where no page was populated yet. This might require adaption when | |
1530 | * supporting other mappings, like shmem. | |
1531 | */ | |
5f19a449 | 1532 | for (; offset < end; offset += block->page_size) { |
f7b9dcfb DH |
1533 | char tmp = *((char *)block->host + offset); |
1534 | ||
1535 | /* Don't optimize the read out */ | |
1536 | asm volatile("" : "+r" (tmp)); | |
1537 | } | |
1538 | } | |
1539 | ||
6fee3a1f DH |
1540 | static inline int populate_read_section(MemoryRegionSection *section, |
1541 | void *opaque) | |
1542 | { | |
1543 | const hwaddr size = int128_get64(section->size); | |
1544 | hwaddr offset = section->offset_within_region; | |
1545 | RAMBlock *block = section->mr->ram_block; | |
1546 | ||
1547 | populate_read_range(block, offset, size); | |
1548 | return 0; | |
1549 | } | |
1550 | ||
eeccb99c | 1551 | /* |
f7b9dcfb DH |
1552 | * ram_block_populate_read: preallocate page tables and populate pages in the |
1553 | * RAM block by reading a byte of each page. | |
eeccb99c AG |
1554 | * |
1555 | * Since it's solely used for userfault_fd WP feature, here we just | |
1556 | * hardcode page size to qemu_real_host_page_size. | |
1557 | * | |
82ea3e3b | 1558 | * @block: RAM block to populate |
eeccb99c | 1559 | */ |
6fee3a1f | 1560 | static void ram_block_populate_read(RAMBlock *rb) |
eeccb99c | 1561 | { |
6fee3a1f DH |
1562 | /* |
1563 | * Skip populating all pages that fall into a discarded range as managed by | |
1564 | * a RamDiscardManager responsible for the mapped memory region of the | |
1565 | * RAMBlock. Such discarded ("logically unplugged") parts of a RAMBlock | |
1566 | * must not get populated automatically. We don't have to track | |
1567 | * modifications via userfaultfd WP reliably, because these pages will | |
1568 | * not be part of the migration stream either way -- see | |
1569 | * ramblock_dirty_bitmap_exclude_discarded_pages(). | |
1570 | * | |
1571 | * Note: The result is only stable while migrating (precopy/postcopy). | |
1572 | */ | |
1573 | if (rb->mr && memory_region_has_ram_discard_manager(rb->mr)) { | |
1574 | RamDiscardManager *rdm = memory_region_get_ram_discard_manager(rb->mr); | |
1575 | MemoryRegionSection section = { | |
1576 | .mr = rb->mr, | |
1577 | .offset_within_region = 0, | |
1578 | .size = rb->mr->size, | |
1579 | }; | |
1580 | ||
1581 | ram_discard_manager_replay_populated(rdm, §ion, | |
1582 | populate_read_section, NULL); | |
1583 | } else { | |
1584 | populate_read_range(rb, 0, rb->used_length); | |
1585 | } | |
eeccb99c AG |
1586 | } |
1587 | ||
1588 | /* | |
1589 | * ram_write_tracking_prepare: prepare for UFFD-WP memory tracking | |
1590 | */ | |
1591 | void ram_write_tracking_prepare(void) | |
1592 | { | |
82ea3e3b | 1593 | RAMBlock *block; |
eeccb99c AG |
1594 | |
1595 | RCU_READ_LOCK_GUARD(); | |
1596 | ||
82ea3e3b | 1597 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
eeccb99c | 1598 | /* Nothing to do with read-only and MMIO-writable regions */ |
82ea3e3b | 1599 | if (block->mr->readonly || block->mr->rom_device) { |
eeccb99c AG |
1600 | continue; |
1601 | } | |
1602 | ||
1603 | /* | |
1604 | * Populate pages of the RAM block before enabling userfault_fd | |
1605 | * write protection. | |
1606 | * | |
1607 | * This stage is required since ioctl(UFFDIO_WRITEPROTECT) with | |
1608 | * UFFDIO_WRITEPROTECT_MODE_WP mode setting would silently skip | |
1609 | * pages with pte_none() entries in page table. | |
1610 | */ | |
f7b9dcfb | 1611 | ram_block_populate_read(block); |
eeccb99c AG |
1612 | } |
1613 | } | |
1614 | ||
e41c5770 DH |
1615 | static inline int uffd_protect_section(MemoryRegionSection *section, |
1616 | void *opaque) | |
1617 | { | |
1618 | const hwaddr size = int128_get64(section->size); | |
1619 | const hwaddr offset = section->offset_within_region; | |
1620 | RAMBlock *rb = section->mr->ram_block; | |
1621 | int uffd_fd = (uintptr_t)opaque; | |
1622 | ||
1623 | return uffd_change_protection(uffd_fd, rb->host + offset, size, true, | |
1624 | false); | |
1625 | } | |
1626 | ||
1627 | static int ram_block_uffd_protect(RAMBlock *rb, int uffd_fd) | |
1628 | { | |
1629 | assert(rb->flags & RAM_UF_WRITEPROTECT); | |
1630 | ||
1631 | /* See ram_block_populate_read() */ | |
1632 | if (rb->mr && memory_region_has_ram_discard_manager(rb->mr)) { | |
1633 | RamDiscardManager *rdm = memory_region_get_ram_discard_manager(rb->mr); | |
1634 | MemoryRegionSection section = { | |
1635 | .mr = rb->mr, | |
1636 | .offset_within_region = 0, | |
1637 | .size = rb->mr->size, | |
1638 | }; | |
1639 | ||
1640 | return ram_discard_manager_replay_populated(rdm, §ion, | |
1641 | uffd_protect_section, | |
1642 | (void *)(uintptr_t)uffd_fd); | |
1643 | } | |
1644 | return uffd_change_protection(uffd_fd, rb->host, | |
1645 | rb->used_length, true, false); | |
1646 | } | |
1647 | ||
278e2f55 AG |
1648 | /* |
1649 | * ram_write_tracking_start: start UFFD-WP memory tracking | |
1650 | * | |
1651 | * Returns 0 for success or negative value in case of error | |
1652 | */ | |
1653 | int ram_write_tracking_start(void) | |
1654 | { | |
1655 | int uffd_fd; | |
1656 | RAMState *rs = ram_state; | |
82ea3e3b | 1657 | RAMBlock *block; |
278e2f55 AG |
1658 | |
1659 | /* Open UFFD file descriptor */ | |
1660 | uffd_fd = uffd_create_fd(UFFD_FEATURE_PAGEFAULT_FLAG_WP, true); | |
1661 | if (uffd_fd < 0) { | |
1662 | return uffd_fd; | |
1663 | } | |
1664 | rs->uffdio_fd = uffd_fd; | |
1665 | ||
1666 | RCU_READ_LOCK_GUARD(); | |
1667 | ||
82ea3e3b | 1668 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
278e2f55 | 1669 | /* Nothing to do with read-only and MMIO-writable regions */ |
82ea3e3b | 1670 | if (block->mr->readonly || block->mr->rom_device) { |
278e2f55 AG |
1671 | continue; |
1672 | } | |
1673 | ||
1674 | /* Register block memory with UFFD to track writes */ | |
82ea3e3b AG |
1675 | if (uffd_register_memory(rs->uffdio_fd, block->host, |
1676 | block->max_length, UFFDIO_REGISTER_MODE_WP, NULL)) { | |
278e2f55 AG |
1677 | goto fail; |
1678 | } | |
72ef3a37 DH |
1679 | block->flags |= RAM_UF_WRITEPROTECT; |
1680 | memory_region_ref(block->mr); | |
1681 | ||
278e2f55 | 1682 | /* Apply UFFD write protection to the block memory range */ |
e41c5770 | 1683 | if (ram_block_uffd_protect(block, uffd_fd)) { |
278e2f55 AG |
1684 | goto fail; |
1685 | } | |
278e2f55 | 1686 | |
82ea3e3b AG |
1687 | trace_ram_write_tracking_ramblock_start(block->idstr, block->page_size, |
1688 | block->host, block->max_length); | |
278e2f55 AG |
1689 | } |
1690 | ||
1691 | return 0; | |
1692 | ||
1693 | fail: | |
1694 | error_report("ram_write_tracking_start() failed: restoring initial memory state"); | |
1695 | ||
82ea3e3b AG |
1696 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
1697 | if ((block->flags & RAM_UF_WRITEPROTECT) == 0) { | |
278e2f55 AG |
1698 | continue; |
1699 | } | |
82ea3e3b | 1700 | uffd_unregister_memory(rs->uffdio_fd, block->host, block->max_length); |
278e2f55 | 1701 | /* Cleanup flags and remove reference */ |
82ea3e3b AG |
1702 | block->flags &= ~RAM_UF_WRITEPROTECT; |
1703 | memory_region_unref(block->mr); | |
278e2f55 AG |
1704 | } |
1705 | ||
1706 | uffd_close_fd(uffd_fd); | |
1707 | rs->uffdio_fd = -1; | |
1708 | return -1; | |
1709 | } | |
1710 | ||
1711 | /** | |
1712 | * ram_write_tracking_stop: stop UFFD-WP memory tracking and remove protection | |
1713 | */ | |
1714 | void ram_write_tracking_stop(void) | |
1715 | { | |
1716 | RAMState *rs = ram_state; | |
82ea3e3b | 1717 | RAMBlock *block; |
278e2f55 AG |
1718 | |
1719 | RCU_READ_LOCK_GUARD(); | |
1720 | ||
82ea3e3b AG |
1721 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
1722 | if ((block->flags & RAM_UF_WRITEPROTECT) == 0) { | |
278e2f55 AG |
1723 | continue; |
1724 | } | |
82ea3e3b | 1725 | uffd_unregister_memory(rs->uffdio_fd, block->host, block->max_length); |
278e2f55 | 1726 | |
82ea3e3b AG |
1727 | trace_ram_write_tracking_ramblock_stop(block->idstr, block->page_size, |
1728 | block->host, block->max_length); | |
278e2f55 AG |
1729 | |
1730 | /* Cleanup flags and remove reference */ | |
82ea3e3b AG |
1731 | block->flags &= ~RAM_UF_WRITEPROTECT; |
1732 | memory_region_unref(block->mr); | |
278e2f55 AG |
1733 | } |
1734 | ||
1735 | /* Finally close UFFD file descriptor */ | |
1736 | uffd_close_fd(rs->uffdio_fd); | |
1737 | rs->uffdio_fd = -1; | |
1738 | } | |
1739 | ||
1740 | #else | |
1741 | /* No target OS support, stubs just fail or ignore */ | |
1742 | ||
1743 | static RAMBlock *poll_fault_page(RAMState *rs, ram_addr_t *offset) | |
1744 | { | |
1745 | (void) rs; | |
1746 | (void) offset; | |
1747 | ||
1748 | return NULL; | |
1749 | } | |
1750 | ||
1751 | static int ram_save_release_protection(RAMState *rs, PageSearchStatus *pss, | |
1752 | unsigned long start_page) | |
1753 | { | |
1754 | (void) rs; | |
1755 | (void) pss; | |
1756 | (void) start_page; | |
1757 | ||
1758 | return 0; | |
1759 | } | |
1760 | ||
1761 | bool ram_write_tracking_available(void) | |
1762 | { | |
1763 | return false; | |
1764 | } | |
1765 | ||
1766 | bool ram_write_tracking_compatible(void) | |
1767 | { | |
1768 | assert(0); | |
1769 | return false; | |
1770 | } | |
1771 | ||
1772 | int ram_write_tracking_start(void) | |
1773 | { | |
1774 | assert(0); | |
1775 | return -1; | |
1776 | } | |
1777 | ||
1778 | void ram_write_tracking_stop(void) | |
1779 | { | |
1780 | assert(0); | |
1781 | } | |
1782 | #endif /* defined(__linux__) */ | |
1783 | ||
3d0684b2 | 1784 | /** |
ff1543af | 1785 | * get_queued_page: unqueue a page from the postcopy requests |
3d0684b2 JQ |
1786 | * |
1787 | * Skips pages that are already sent (!dirty) | |
a82d593b | 1788 | * |
a5f7b1a6 | 1789 | * Returns true if a queued page is found |
a82d593b | 1790 | * |
6f37bb8b | 1791 | * @rs: current RAM state |
3d0684b2 | 1792 | * @pss: data about the state of the current dirty page scan |
a82d593b | 1793 | */ |
f20e2865 | 1794 | static bool get_queued_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b DDAG |
1795 | { |
1796 | RAMBlock *block; | |
1797 | ram_addr_t offset; | |
777f53c7 TH |
1798 | bool dirty; |
1799 | ||
1800 | do { | |
1801 | block = unqueue_page(rs, &offset); | |
1802 | /* | |
1803 | * We're sending this page, and since it's postcopy nothing else | |
1804 | * will dirty it, and we must make sure it doesn't get sent again | |
1805 | * even if this queue request was received after the background | |
1806 | * search already sent it. | |
1807 | */ | |
1808 | if (block) { | |
1809 | unsigned long page; | |
1810 | ||
1811 | page = offset >> TARGET_PAGE_BITS; | |
1812 | dirty = test_bit(page, block->bmap); | |
1813 | if (!dirty) { | |
1814 | trace_get_queued_page_not_dirty(block->idstr, (uint64_t)offset, | |
1815 | page); | |
1816 | } else { | |
1817 | trace_get_queued_page(block->idstr, (uint64_t)offset, page); | |
1818 | } | |
1819 | } | |
a82d593b | 1820 | |
777f53c7 | 1821 | } while (block && !dirty); |
a82d593b | 1822 | |
b062106d | 1823 | if (!block) { |
278e2f55 AG |
1824 | /* |
1825 | * Poll write faults too if background snapshot is enabled; that's | |
1826 | * when we have vcpus got blocked by the write protected pages. | |
1827 | */ | |
1828 | block = poll_fault_page(rs, &offset); | |
1829 | } | |
1830 | ||
a82d593b | 1831 | if (block) { |
a82d593b DDAG |
1832 | /* |
1833 | * We want the background search to continue from the queued page | |
1834 | * since the guest is likely to want other pages near to the page | |
1835 | * it just requested. | |
1836 | */ | |
1837 | pss->block = block; | |
a935e30f | 1838 | pss->page = offset >> TARGET_PAGE_BITS; |
422314e7 WY |
1839 | |
1840 | /* | |
1841 | * This unqueued page would break the "one round" check, even is | |
1842 | * really rare. | |
1843 | */ | |
1844 | pss->complete_round = false; | |
a82d593b DDAG |
1845 | } |
1846 | ||
1847 | return !!block; | |
1848 | } | |
1849 | ||
6c595cde | 1850 | /** |
5e58f968 JQ |
1851 | * migration_page_queue_free: drop any remaining pages in the ram |
1852 | * request queue | |
6c595cde | 1853 | * |
3d0684b2 JQ |
1854 | * It should be empty at the end anyway, but in error cases there may |
1855 | * be some left. in case that there is any page left, we drop it. | |
1856 | * | |
6c595cde | 1857 | */ |
83c13382 | 1858 | static void migration_page_queue_free(RAMState *rs) |
6c595cde | 1859 | { |
ec481c6c | 1860 | struct RAMSrcPageRequest *mspr, *next_mspr; |
6c595cde DDAG |
1861 | /* This queue generally should be empty - but in the case of a failed |
1862 | * migration might have some droppings in. | |
1863 | */ | |
89ac5a1d | 1864 | RCU_READ_LOCK_GUARD(); |
ec481c6c | 1865 | QSIMPLEQ_FOREACH_SAFE(mspr, &rs->src_page_requests, next_req, next_mspr) { |
6c595cde | 1866 | memory_region_unref(mspr->rb->mr); |
ec481c6c | 1867 | QSIMPLEQ_REMOVE_HEAD(&rs->src_page_requests, next_req); |
6c595cde DDAG |
1868 | g_free(mspr); |
1869 | } | |
6c595cde DDAG |
1870 | } |
1871 | ||
1872 | /** | |
3d0684b2 JQ |
1873 | * ram_save_queue_pages: queue the page for transmission |
1874 | * | |
1875 | * A request from postcopy destination for example. | |
1876 | * | |
1877 | * Returns zero on success or negative on error | |
1878 | * | |
3d0684b2 JQ |
1879 | * @rbname: Name of the RAMBLock of the request. NULL means the |
1880 | * same that last one. | |
1881 | * @start: starting address from the start of the RAMBlock | |
1882 | * @len: length (in bytes) to send | |
6c595cde | 1883 | */ |
7aa6070d PX |
1884 | int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len, |
1885 | Error **errp) | |
6c595cde DDAG |
1886 | { |
1887 | RAMBlock *ramblock; | |
53518d94 | 1888 | RAMState *rs = ram_state; |
6c595cde | 1889 | |
aff3f660 | 1890 | stat64_add(&mig_stats.postcopy_requests, 1); |
89ac5a1d DDAG |
1891 | RCU_READ_LOCK_GUARD(); |
1892 | ||
6c595cde DDAG |
1893 | if (!rbname) { |
1894 | /* Reuse last RAMBlock */ | |
68a098f3 | 1895 | ramblock = rs->last_req_rb; |
6c595cde DDAG |
1896 | |
1897 | if (!ramblock) { | |
1898 | /* | |
1899 | * Shouldn't happen, we can't reuse the last RAMBlock if | |
1900 | * it's the 1st request. | |
1901 | */ | |
7aa6070d | 1902 | error_setg(errp, "MIG_RP_MSG_REQ_PAGES has no previous block"); |
03acb4e9 | 1903 | return -1; |
6c595cde DDAG |
1904 | } |
1905 | } else { | |
1906 | ramblock = qemu_ram_block_by_name(rbname); | |
1907 | ||
1908 | if (!ramblock) { | |
1909 | /* We shouldn't be asked for a non-existent RAMBlock */ | |
7aa6070d | 1910 | error_setg(errp, "MIG_RP_MSG_REQ_PAGES has no block '%s'", rbname); |
03acb4e9 | 1911 | return -1; |
6c595cde | 1912 | } |
68a098f3 | 1913 | rs->last_req_rb = ramblock; |
6c595cde DDAG |
1914 | } |
1915 | trace_ram_save_queue_pages(ramblock->idstr, start, len); | |
542147f4 | 1916 | if (!offset_in_ramblock(ramblock, start + len - 1)) { |
7aa6070d PX |
1917 | error_setg(errp, "MIG_RP_MSG_REQ_PAGES request overrun, " |
1918 | "start=" RAM_ADDR_FMT " len=" | |
1919 | RAM_ADDR_FMT " blocklen=" RAM_ADDR_FMT, | |
1920 | start, len, ramblock->used_length); | |
03acb4e9 | 1921 | return -1; |
6c595cde DDAG |
1922 | } |
1923 | ||
93589827 PX |
1924 | /* |
1925 | * When with postcopy preempt, we send back the page directly in the | |
1926 | * rp-return thread. | |
1927 | */ | |
1928 | if (postcopy_preempt_active()) { | |
1929 | ram_addr_t page_start = start >> TARGET_PAGE_BITS; | |
1930 | size_t page_size = qemu_ram_pagesize(ramblock); | |
1931 | PageSearchStatus *pss = &ram_state->pss[RAM_CHANNEL_POSTCOPY]; | |
1932 | int ret = 0; | |
1933 | ||
1934 | qemu_mutex_lock(&rs->bitmap_mutex); | |
1935 | ||
1936 | pss_init(pss, ramblock, page_start); | |
1937 | /* | |
1938 | * Always use the preempt channel, and make sure it's there. It's | |
1939 | * safe to access without lock, because when rp-thread is running | |
1940 | * we should be the only one who operates on the qemufile | |
1941 | */ | |
1942 | pss->pss_channel = migrate_get_current()->postcopy_qemufile_src; | |
93589827 PX |
1943 | assert(pss->pss_channel); |
1944 | ||
1945 | /* | |
1946 | * It must be either one or multiple of host page size. Just | |
1947 | * assert; if something wrong we're mostly split brain anyway. | |
1948 | */ | |
1949 | assert(len % page_size == 0); | |
1950 | while (len) { | |
1951 | if (ram_save_host_page_urgent(pss)) { | |
7aa6070d PX |
1952 | error_setg(errp, "ram_save_host_page_urgent() failed: " |
1953 | "ramblock=%s, start_addr=0x"RAM_ADDR_FMT, | |
1954 | ramblock->idstr, start); | |
93589827 PX |
1955 | ret = -1; |
1956 | break; | |
1957 | } | |
1958 | /* | |
1959 | * NOTE: after ram_save_host_page_urgent() succeeded, pss->page | |
1960 | * will automatically be moved and point to the next host page | |
1961 | * we're going to send, so no need to update here. | |
1962 | * | |
1963 | * Normally QEMU never sends >1 host page in requests, so | |
1964 | * logically we don't even need that as the loop should only | |
1965 | * run once, but just to be consistent. | |
1966 | */ | |
1967 | len -= page_size; | |
1968 | }; | |
1969 | qemu_mutex_unlock(&rs->bitmap_mutex); | |
1970 | ||
1971 | return ret; | |
1972 | } | |
1973 | ||
ec481c6c | 1974 | struct RAMSrcPageRequest *new_entry = |
b21e2380 | 1975 | g_new0(struct RAMSrcPageRequest, 1); |
6c595cde DDAG |
1976 | new_entry->rb = ramblock; |
1977 | new_entry->offset = start; | |
1978 | new_entry->len = len; | |
1979 | ||
1980 | memory_region_ref(ramblock->mr); | |
ec481c6c JQ |
1981 | qemu_mutex_lock(&rs->src_page_req_mutex); |
1982 | QSIMPLEQ_INSERT_TAIL(&rs->src_page_requests, new_entry, next_req); | |
e03a34f8 | 1983 | migration_make_urgent_request(); |
ec481c6c | 1984 | qemu_mutex_unlock(&rs->src_page_req_mutex); |
6c595cde DDAG |
1985 | |
1986 | return 0; | |
6c595cde DDAG |
1987 | } |
1988 | ||
a82d593b | 1989 | /** |
4010ba38 | 1990 | * ram_save_target_page_legacy: save one target page |
a82d593b | 1991 | * |
3d0684b2 | 1992 | * Returns the number of pages written |
a82d593b | 1993 | * |
6f37bb8b | 1994 | * @rs: current RAM state |
3d0684b2 | 1995 | * @pss: data about the page we want to send |
a82d593b | 1996 | */ |
4010ba38 | 1997 | static int ram_save_target_page_legacy(RAMState *rs, PageSearchStatus *pss) |
a82d593b | 1998 | { |
8bba004c | 1999 | ram_addr_t offset = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS; |
a8ec91f9 XG |
2000 | int res; |
2001 | ||
944853c2 | 2002 | if (control_save_page(pss, offset, &res)) { |
a8ec91f9 XG |
2003 | return res; |
2004 | } | |
2005 | ||
e8e4e7ac | 2006 | if (save_zero_page(rs, pss, offset)) { |
8697eb85 | 2007 | return 1; |
d7400a34 XG |
2008 | } |
2009 | ||
9ae90f73 HX |
2010 | return ram_save_page(rs, pss); |
2011 | } | |
2012 | ||
2013 | /** | |
2014 | * ram_save_target_page_multifd: send one target page to multifd workers | |
2015 | * | |
2016 | * Returns 1 if the page was queued, -1 otherwise. | |
2017 | * | |
2018 | * @rs: current RAM state | |
2019 | * @pss: data about the page we want to send | |
2020 | */ | |
2021 | static int ram_save_target_page_multifd(RAMState *rs, PageSearchStatus *pss) | |
2022 | { | |
2023 | RAMBlock *block = pss->block; | |
2024 | ram_addr_t offset = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS; | |
2025 | ||
da3f56cb | 2026 | /* |
9ae90f73 HX |
2027 | * While using multifd live migration, we still need to handle zero |
2028 | * page checking on the migration main thread. | |
da3f56cb | 2029 | */ |
9ae90f73 HX |
2030 | if (migrate_zero_page_detection() == ZERO_PAGE_DETECTION_LEGACY) { |
2031 | if (save_zero_page(rs, pss, offset)) { | |
2032 | return 1; | |
2033 | } | |
a82d593b DDAG |
2034 | } |
2035 | ||
9ae90f73 | 2036 | return ram_save_multifd_page(block, offset); |
a82d593b DDAG |
2037 | } |
2038 | ||
d9e474ea PX |
2039 | /* Should be called before sending a host page */ |
2040 | static void pss_host_page_prepare(PageSearchStatus *pss) | |
2041 | { | |
2042 | /* How many guest pages are there in one host page? */ | |
2043 | size_t guest_pfns = qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
2044 | ||
2045 | pss->host_page_sending = true; | |
301d7ffe PX |
2046 | if (guest_pfns <= 1) { |
2047 | /* | |
2048 | * This covers both when guest psize == host psize, or when guest | |
2049 | * has larger psize than the host (guest_pfns==0). | |
2050 | * | |
2051 | * For the latter, we always send one whole guest page per | |
2052 | * iteration of the host page (example: an Alpha VM on x86 host | |
2053 | * will have guest psize 8K while host psize 4K). | |
2054 | */ | |
2055 | pss->host_page_start = pss->page; | |
2056 | pss->host_page_end = pss->page + 1; | |
2057 | } else { | |
2058 | /* | |
2059 | * The host page spans over multiple guest pages, we send them | |
2060 | * within the same host page iteration. | |
2061 | */ | |
2062 | pss->host_page_start = ROUND_DOWN(pss->page, guest_pfns); | |
2063 | pss->host_page_end = ROUND_UP(pss->page + 1, guest_pfns); | |
2064 | } | |
d9e474ea PX |
2065 | } |
2066 | ||
2067 | /* | |
2068 | * Whether the page pointed by PSS is within the host page being sent. | |
2069 | * Must be called after a previous pss_host_page_prepare(). | |
2070 | */ | |
2071 | static bool pss_within_range(PageSearchStatus *pss) | |
2072 | { | |
2073 | ram_addr_t ram_addr; | |
2074 | ||
2075 | assert(pss->host_page_sending); | |
2076 | ||
2077 | /* Over host-page boundary? */ | |
2078 | if (pss->page >= pss->host_page_end) { | |
2079 | return false; | |
2080 | } | |
2081 | ||
2082 | ram_addr = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS; | |
2083 | ||
2084 | return offset_in_ramblock(pss->block, ram_addr); | |
2085 | } | |
2086 | ||
2087 | static void pss_host_page_finish(PageSearchStatus *pss) | |
2088 | { | |
2089 | pss->host_page_sending = false; | |
2090 | /* This is not needed, but just to reset it */ | |
2091 | pss->host_page_start = pss->host_page_end = 0; | |
2092 | } | |
2093 | ||
93589827 PX |
2094 | /* |
2095 | * Send an urgent host page specified by `pss'. Need to be called with | |
2096 | * bitmap_mutex held. | |
2097 | * | |
2098 | * Returns 0 if save host page succeeded, false otherwise. | |
2099 | */ | |
2100 | static int ram_save_host_page_urgent(PageSearchStatus *pss) | |
2101 | { | |
2102 | bool page_dirty, sent = false; | |
2103 | RAMState *rs = ram_state; | |
2104 | int ret = 0; | |
2105 | ||
2106 | trace_postcopy_preempt_send_host_page(pss->block->idstr, pss->page); | |
2107 | pss_host_page_prepare(pss); | |
2108 | ||
2109 | /* | |
2110 | * If precopy is sending the same page, let it be done in precopy, or | |
2111 | * we could send the same page in two channels and none of them will | |
2112 | * receive the whole page. | |
2113 | */ | |
2114 | if (pss_overlap(pss, &ram_state->pss[RAM_CHANNEL_PRECOPY])) { | |
2115 | trace_postcopy_preempt_hit(pss->block->idstr, | |
2116 | pss->page << TARGET_PAGE_BITS); | |
2117 | return 0; | |
2118 | } | |
2119 | ||
2120 | do { | |
2121 | page_dirty = migration_bitmap_clear_dirty(rs, pss->block, pss->page); | |
2122 | ||
2123 | if (page_dirty) { | |
2124 | /* Be strict to return code; it must be 1, or what else? */ | |
4010ba38 | 2125 | if (migration_ops->ram_save_target_page(rs, pss) != 1) { |
93589827 PX |
2126 | error_report_once("%s: ram_save_target_page failed", __func__); |
2127 | ret = -1; | |
2128 | goto out; | |
2129 | } | |
2130 | sent = true; | |
2131 | } | |
2132 | pss_find_next_dirty(pss); | |
2133 | } while (pss_within_range(pss)); | |
2134 | out: | |
2135 | pss_host_page_finish(pss); | |
2136 | /* For urgent requests, flush immediately if sent */ | |
2137 | if (sent) { | |
2138 | qemu_fflush(pss->pss_channel); | |
2139 | } | |
2140 | return ret; | |
2141 | } | |
2142 | ||
a82d593b | 2143 | /** |
3d0684b2 | 2144 | * ram_save_host_page: save a whole host page |
a82d593b | 2145 | * |
3d0684b2 JQ |
2146 | * Starting at *offset send pages up to the end of the current host |
2147 | * page. It's valid for the initial offset to point into the middle of | |
2148 | * a host page in which case the remainder of the hostpage is sent. | |
2149 | * Only dirty target pages are sent. Note that the host page size may | |
2150 | * be a huge page for this block. | |
f3321554 | 2151 | * |
1eb3fc0a DDAG |
2152 | * The saving stops at the boundary of the used_length of the block |
2153 | * if the RAMBlock isn't a multiple of the host page size. | |
a82d593b | 2154 | * |
f3321554 PX |
2155 | * The caller must be with ram_state.bitmap_mutex held to call this |
2156 | * function. Note that this function can temporarily release the lock, but | |
2157 | * when the function is returned it'll make sure the lock is still held. | |
2158 | * | |
3d0684b2 JQ |
2159 | * Returns the number of pages written or negative on error |
2160 | * | |
6f37bb8b | 2161 | * @rs: current RAM state |
3d0684b2 | 2162 | * @pss: data about the page we want to send |
a82d593b | 2163 | */ |
05931ec5 | 2164 | static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss) |
a82d593b | 2165 | { |
f3321554 | 2166 | bool page_dirty, preempt_active = postcopy_preempt_active(); |
a82d593b | 2167 | int tmppages, pages = 0; |
a935e30f JQ |
2168 | size_t pagesize_bits = |
2169 | qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS; | |
278e2f55 AG |
2170 | unsigned long start_page = pss->page; |
2171 | int res; | |
4c011c37 | 2172 | |
f161c88a | 2173 | if (migrate_ram_is_ignored(pss->block)) { |
b895de50 CLG |
2174 | error_report("block %s should not be migrated !", pss->block->idstr); |
2175 | return 0; | |
2176 | } | |
2177 | ||
d9e474ea PX |
2178 | /* Update host page boundary information */ |
2179 | pss_host_page_prepare(pss); | |
2180 | ||
a82d593b | 2181 | do { |
f3321554 | 2182 | page_dirty = migration_bitmap_clear_dirty(rs, pss->block, pss->page); |
a82d593b | 2183 | |
f3321554 PX |
2184 | /* Check the pages is dirty and if it is send it */ |
2185 | if (page_dirty) { | |
ba1b7c81 | 2186 | /* |
f3321554 PX |
2187 | * Properly yield the lock only in postcopy preempt mode |
2188 | * because both migration thread and rp-return thread can | |
2189 | * operate on the bitmaps. | |
ba1b7c81 | 2190 | */ |
f3321554 PX |
2191 | if (preempt_active) { |
2192 | qemu_mutex_unlock(&rs->bitmap_mutex); | |
ba1b7c81 | 2193 | } |
4010ba38 | 2194 | tmppages = migration_ops->ram_save_target_page(rs, pss); |
f3321554 PX |
2195 | if (tmppages >= 0) { |
2196 | pages += tmppages; | |
2197 | /* | |
2198 | * Allow rate limiting to happen in the middle of huge pages if | |
2199 | * something is sent in the current iteration. | |
2200 | */ | |
2201 | if (pagesize_bits > 1 && tmppages > 0) { | |
2202 | migration_rate_limit(); | |
2203 | } | |
2204 | } | |
2205 | if (preempt_active) { | |
2206 | qemu_mutex_lock(&rs->bitmap_mutex); | |
2207 | } | |
2208 | } else { | |
2209 | tmppages = 0; | |
23feba90 | 2210 | } |
f3321554 PX |
2211 | |
2212 | if (tmppages < 0) { | |
d9e474ea | 2213 | pss_host_page_finish(pss); |
f3321554 PX |
2214 | return tmppages; |
2215 | } | |
2216 | ||
d9e474ea PX |
2217 | pss_find_next_dirty(pss); |
2218 | } while (pss_within_range(pss)); | |
2219 | ||
2220 | pss_host_page_finish(pss); | |
278e2f55 AG |
2221 | |
2222 | res = ram_save_release_protection(rs, pss, start_page); | |
2223 | return (res < 0 ? res : pages); | |
a82d593b | 2224 | } |
6c595cde | 2225 | |
56e93d26 | 2226 | /** |
3d0684b2 | 2227 | * ram_find_and_save_block: finds a dirty page and sends it to f |
56e93d26 JQ |
2228 | * |
2229 | * Called within an RCU critical section. | |
2230 | * | |
e8f3735f XG |
2231 | * Returns the number of pages written where zero means no dirty pages, |
2232 | * or negative on error | |
56e93d26 | 2233 | * |
6f37bb8b | 2234 | * @rs: current RAM state |
a82d593b DDAG |
2235 | * |
2236 | * On systems where host-page-size > target-page-size it will send all the | |
2237 | * pages in a host page that are dirty. | |
56e93d26 | 2238 | */ |
05931ec5 | 2239 | static int ram_find_and_save_block(RAMState *rs) |
56e93d26 | 2240 | { |
f1668764 | 2241 | PageSearchStatus *pss = &rs->pss[RAM_CHANNEL_PRECOPY]; |
56e93d26 | 2242 | int pages = 0; |
56e93d26 | 2243 | |
0827b9e9 | 2244 | /* No dirty page as there is zero RAM */ |
8d80e195 | 2245 | if (!rs->ram_bytes_total) { |
0827b9e9 AA |
2246 | return pages; |
2247 | } | |
2248 | ||
4934a5dd PX |
2249 | /* |
2250 | * Always keep last_seen_block/last_page valid during this procedure, | |
2251 | * because find_dirty_block() relies on these values (e.g., we compare | |
2252 | * last_seen_block with pss.block to see whether we searched all the | |
2253 | * ramblocks) to detect the completion of migration. Having NULL value | |
2254 | * of last_seen_block can conditionally cause below loop to run forever. | |
2255 | */ | |
2256 | if (!rs->last_seen_block) { | |
2257 | rs->last_seen_block = QLIST_FIRST_RCU(&ram_list.blocks); | |
2258 | rs->last_page = 0; | |
2259 | } | |
2260 | ||
f1668764 | 2261 | pss_init(pss, rs->last_seen_block, rs->last_page); |
b8fb8cb7 | 2262 | |
31e2ac74 | 2263 | while (true){ |
51efd36f | 2264 | if (!get_queued_page(rs, pss)) { |
b062106d | 2265 | /* priority queue empty, so just search for something dirty */ |
31e2ac74 JQ |
2266 | int res = find_dirty_block(rs, pss); |
2267 | if (res != PAGE_DIRTY_FOUND) { | |
2268 | if (res == PAGE_ALL_CLEAN) { | |
51efd36f | 2269 | break; |
31e2ac74 JQ |
2270 | } else if (res == PAGE_TRY_AGAIN) { |
2271 | continue; | |
294e5a40 JQ |
2272 | } else if (res < 0) { |
2273 | pages = res; | |
2274 | break; | |
51efd36f JQ |
2275 | } |
2276 | } | |
56e93d26 | 2277 | } |
51efd36f | 2278 | pages = ram_save_host_page(rs, pss); |
31e2ac74 JQ |
2279 | if (pages) { |
2280 | break; | |
2281 | } | |
2282 | } | |
56e93d26 | 2283 | |
f1668764 PX |
2284 | rs->last_seen_block = pss->block; |
2285 | rs->last_page = pss->page; | |
56e93d26 JQ |
2286 | |
2287 | return pages; | |
2288 | } | |
2289 | ||
8008a272 | 2290 | static uint64_t ram_bytes_total_with_ignored(void) |
56e93d26 JQ |
2291 | { |
2292 | RAMBlock *block; | |
2293 | uint64_t total = 0; | |
2294 | ||
89ac5a1d DDAG |
2295 | RCU_READ_LOCK_GUARD(); |
2296 | ||
8008a272 JQ |
2297 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
2298 | total += block->used_length; | |
99e15582 | 2299 | } |
56e93d26 JQ |
2300 | return total; |
2301 | } | |
2302 | ||
fbd162e6 YK |
2303 | uint64_t ram_bytes_total(void) |
2304 | { | |
8008a272 JQ |
2305 | RAMBlock *block; |
2306 | uint64_t total = 0; | |
2307 | ||
2308 | RCU_READ_LOCK_GUARD(); | |
2309 | ||
2310 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
2311 | total += block->used_length; | |
2312 | } | |
2313 | return total; | |
fbd162e6 YK |
2314 | } |
2315 | ||
f265e0e4 | 2316 | static void xbzrle_load_setup(void) |
56e93d26 | 2317 | { |
f265e0e4 | 2318 | XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE); |
56e93d26 JQ |
2319 | } |
2320 | ||
f265e0e4 JQ |
2321 | static void xbzrle_load_cleanup(void) |
2322 | { | |
2323 | g_free(XBZRLE.decoded_buf); | |
2324 | XBZRLE.decoded_buf = NULL; | |
2325 | } | |
2326 | ||
7d7c96be PX |
2327 | static void ram_state_cleanup(RAMState **rsp) |
2328 | { | |
b9ccaf6d DDAG |
2329 | if (*rsp) { |
2330 | migration_page_queue_free(*rsp); | |
2331 | qemu_mutex_destroy(&(*rsp)->bitmap_mutex); | |
2332 | qemu_mutex_destroy(&(*rsp)->src_page_req_mutex); | |
2333 | g_free(*rsp); | |
2334 | *rsp = NULL; | |
2335 | } | |
7d7c96be PX |
2336 | } |
2337 | ||
84593a08 PX |
2338 | static void xbzrle_cleanup(void) |
2339 | { | |
2340 | XBZRLE_cache_lock(); | |
2341 | if (XBZRLE.cache) { | |
2342 | cache_fini(XBZRLE.cache); | |
2343 | g_free(XBZRLE.encoded_buf); | |
2344 | g_free(XBZRLE.current_buf); | |
2345 | g_free(XBZRLE.zero_target_page); | |
2346 | XBZRLE.cache = NULL; | |
2347 | XBZRLE.encoded_buf = NULL; | |
2348 | XBZRLE.current_buf = NULL; | |
2349 | XBZRLE.zero_target_page = NULL; | |
2350 | } | |
2351 | XBZRLE_cache_unlock(); | |
2352 | } | |
2353 | ||
92c20b2f CLG |
2354 | static void ram_bitmaps_destroy(void) |
2355 | { | |
2356 | RAMBlock *block; | |
2357 | ||
2358 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
2359 | g_free(block->clear_bmap); | |
2360 | block->clear_bmap = NULL; | |
2361 | g_free(block->bmap); | |
2362 | block->bmap = NULL; | |
2363 | g_free(block->file_bmap); | |
2364 | block->file_bmap = NULL; | |
2365 | } | |
2366 | } | |
2367 | ||
f265e0e4 | 2368 | static void ram_save_cleanup(void *opaque) |
56e93d26 | 2369 | { |
53518d94 | 2370 | RAMState **rsp = opaque; |
eb859c53 | 2371 | |
278e2f55 AG |
2372 | /* We don't use dirty log with background snapshots */ |
2373 | if (!migrate_background_snapshot()) { | |
a4a411fb | 2374 | /* caller have hold BQL or is in a bh, so there is |
278e2f55 AG |
2375 | * no writing race against the migration bitmap |
2376 | */ | |
63b41db4 HH |
2377 | if (global_dirty_tracking & GLOBAL_DIRTY_MIGRATION) { |
2378 | /* | |
2379 | * do not stop dirty log without starting it, since | |
2380 | * memory_global_dirty_log_stop will assert that | |
2381 | * memory_global_dirty_log_start/stop used in pairs | |
2382 | */ | |
2383 | memory_global_dirty_log_stop(GLOBAL_DIRTY_MIGRATION); | |
2384 | } | |
278e2f55 | 2385 | } |
6b6712ef | 2386 | |
92c20b2f | 2387 | ram_bitmaps_destroy(); |
56e93d26 | 2388 | |
84593a08 | 2389 | xbzrle_cleanup(); |
7d7c96be | 2390 | ram_state_cleanup(rsp); |
4010ba38 JQ |
2391 | g_free(migration_ops); |
2392 | migration_ops = NULL; | |
56e93d26 JQ |
2393 | } |
2394 | ||
6f37bb8b | 2395 | static void ram_state_reset(RAMState *rs) |
56e93d26 | 2396 | { |
ec6f3ab9 PX |
2397 | int i; |
2398 | ||
2399 | for (i = 0; i < RAM_CHANNEL_MAX; i++) { | |
2400 | rs->pss[i].last_sent_block = NULL; | |
2401 | } | |
2402 | ||
6f37bb8b | 2403 | rs->last_seen_block = NULL; |
269ace29 | 2404 | rs->last_page = 0; |
6f37bb8b | 2405 | rs->last_version = ram_list.version; |
f3095cc8 | 2406 | rs->xbzrle_started = false; |
56e93d26 JQ |
2407 | } |
2408 | ||
2409 | #define MAX_WAIT 50 /* ms, half buffered_file limit */ | |
2410 | ||
e0b266f0 DDAG |
2411 | /* **** functions for postcopy ***** */ |
2412 | ||
ced1c616 PB |
2413 | void ram_postcopy_migrated_memory_release(MigrationState *ms) |
2414 | { | |
2415 | struct RAMBlock *block; | |
ced1c616 | 2416 | |
fbd162e6 | 2417 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
6b6712ef JQ |
2418 | unsigned long *bitmap = block->bmap; |
2419 | unsigned long range = block->used_length >> TARGET_PAGE_BITS; | |
2420 | unsigned long run_start = find_next_zero_bit(bitmap, range, 0); | |
ced1c616 PB |
2421 | |
2422 | while (run_start < range) { | |
2423 | unsigned long run_end = find_next_bit(bitmap, range, run_start + 1); | |
8bba004c AR |
2424 | ram_discard_range(block->idstr, |
2425 | ((ram_addr_t)run_start) << TARGET_PAGE_BITS, | |
2426 | ((ram_addr_t)(run_end - run_start)) | |
2427 | << TARGET_PAGE_BITS); | |
ced1c616 PB |
2428 | run_start = find_next_zero_bit(bitmap, range, run_end + 1); |
2429 | } | |
2430 | } | |
2431 | } | |
2432 | ||
3d0684b2 JQ |
2433 | /** |
2434 | * postcopy_send_discard_bm_ram: discard a RAMBlock | |
2435 | * | |
e0b266f0 | 2436 | * Callback from postcopy_each_ram_send_discard for each RAMBlock |
3d0684b2 JQ |
2437 | * |
2438 | * @ms: current migration state | |
89dab31b | 2439 | * @block: RAMBlock to discard |
e0b266f0 | 2440 | */ |
9e7d1223 | 2441 | static void postcopy_send_discard_bm_ram(MigrationState *ms, RAMBlock *block) |
e0b266f0 | 2442 | { |
6b6712ef | 2443 | unsigned long end = block->used_length >> TARGET_PAGE_BITS; |
e0b266f0 | 2444 | unsigned long current; |
1e7cf8c3 | 2445 | unsigned long *bitmap = block->bmap; |
e0b266f0 | 2446 | |
6b6712ef | 2447 | for (current = 0; current < end; ) { |
1e7cf8c3 | 2448 | unsigned long one = find_next_bit(bitmap, end, current); |
33a5cb62 | 2449 | unsigned long zero, discard_length; |
e0b266f0 | 2450 | |
33a5cb62 WY |
2451 | if (one >= end) { |
2452 | break; | |
2453 | } | |
e0b266f0 | 2454 | |
1e7cf8c3 | 2455 | zero = find_next_zero_bit(bitmap, end, one + 1); |
33a5cb62 WY |
2456 | |
2457 | if (zero >= end) { | |
2458 | discard_length = end - one; | |
e0b266f0 | 2459 | } else { |
33a5cb62 WY |
2460 | discard_length = zero - one; |
2461 | } | |
810cf2bb | 2462 | postcopy_discard_send_range(ms, one, discard_length); |
33a5cb62 | 2463 | current = one + discard_length; |
e0b266f0 | 2464 | } |
e0b266f0 DDAG |
2465 | } |
2466 | ||
f30c2e5b PX |
2467 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, RAMBlock *block); |
2468 | ||
3d0684b2 JQ |
2469 | /** |
2470 | * postcopy_each_ram_send_discard: discard all RAMBlocks | |
2471 | * | |
e0b266f0 DDAG |
2472 | * Utility for the outgoing postcopy code. |
2473 | * Calls postcopy_send_discard_bm_ram for each RAMBlock | |
2474 | * passing it bitmap indexes and name. | |
e0b266f0 DDAG |
2475 | * (qemu_ram_foreach_block ends up passing unscaled lengths |
2476 | * which would mean postcopy code would have to deal with target page) | |
3d0684b2 JQ |
2477 | * |
2478 | * @ms: current migration state | |
e0b266f0 | 2479 | */ |
739fcc1b | 2480 | static void postcopy_each_ram_send_discard(MigrationState *ms) |
e0b266f0 DDAG |
2481 | { |
2482 | struct RAMBlock *block; | |
e0b266f0 | 2483 | |
fbd162e6 | 2484 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
810cf2bb | 2485 | postcopy_discard_send_init(ms, block->idstr); |
e0b266f0 | 2486 | |
f30c2e5b PX |
2487 | /* |
2488 | * Deal with TPS != HPS and huge pages. It discard any partially sent | |
2489 | * host-page size chunks, mark any partially dirty host-page size | |
2490 | * chunks as all dirty. In this case the host-page is the host-page | |
2491 | * for the particular RAMBlock, i.e. it might be a huge page. | |
2492 | */ | |
2493 | postcopy_chunk_hostpages_pass(ms, block); | |
2494 | ||
e0b266f0 DDAG |
2495 | /* |
2496 | * Postcopy sends chunks of bitmap over the wire, but it | |
2497 | * just needs indexes at this point, avoids it having | |
2498 | * target page specific code. | |
2499 | */ | |
739fcc1b | 2500 | postcopy_send_discard_bm_ram(ms, block); |
810cf2bb | 2501 | postcopy_discard_send_finish(ms); |
e0b266f0 | 2502 | } |
e0b266f0 DDAG |
2503 | } |
2504 | ||
3d0684b2 | 2505 | /** |
8324ef86 | 2506 | * postcopy_chunk_hostpages_pass: canonicalize bitmap in hostpages |
3d0684b2 JQ |
2507 | * |
2508 | * Helper for postcopy_chunk_hostpages; it's called twice to | |
2509 | * canonicalize the two bitmaps, that are similar, but one is | |
2510 | * inverted. | |
99e314eb | 2511 | * |
3d0684b2 JQ |
2512 | * Postcopy requires that all target pages in a hostpage are dirty or |
2513 | * clean, not a mix. This function canonicalizes the bitmaps. | |
99e314eb | 2514 | * |
3d0684b2 | 2515 | * @ms: current migration state |
3d0684b2 | 2516 | * @block: block that contains the page we want to canonicalize |
99e314eb | 2517 | */ |
1e7cf8c3 | 2518 | static void postcopy_chunk_hostpages_pass(MigrationState *ms, RAMBlock *block) |
99e314eb | 2519 | { |
53518d94 | 2520 | RAMState *rs = ram_state; |
6b6712ef | 2521 | unsigned long *bitmap = block->bmap; |
29c59172 | 2522 | unsigned int host_ratio = block->page_size / TARGET_PAGE_SIZE; |
6b6712ef | 2523 | unsigned long pages = block->used_length >> TARGET_PAGE_BITS; |
99e314eb DDAG |
2524 | unsigned long run_start; |
2525 | ||
29c59172 DDAG |
2526 | if (block->page_size == TARGET_PAGE_SIZE) { |
2527 | /* Easy case - TPS==HPS for a non-huge page RAMBlock */ | |
2528 | return; | |
2529 | } | |
2530 | ||
1e7cf8c3 WY |
2531 | /* Find a dirty page */ |
2532 | run_start = find_next_bit(bitmap, pages, 0); | |
99e314eb | 2533 | |
6b6712ef | 2534 | while (run_start < pages) { |
99e314eb DDAG |
2535 | |
2536 | /* | |
2537 | * If the start of this run of pages is in the middle of a host | |
2538 | * page, then we need to fixup this host page. | |
2539 | */ | |
9dec3cc3 | 2540 | if (QEMU_IS_ALIGNED(run_start, host_ratio)) { |
99e314eb | 2541 | /* Find the end of this run */ |
1e7cf8c3 | 2542 | run_start = find_next_zero_bit(bitmap, pages, run_start + 1); |
99e314eb DDAG |
2543 | /* |
2544 | * If the end isn't at the start of a host page, then the | |
2545 | * run doesn't finish at the end of a host page | |
2546 | * and we need to discard. | |
2547 | */ | |
99e314eb DDAG |
2548 | } |
2549 | ||
9dec3cc3 | 2550 | if (!QEMU_IS_ALIGNED(run_start, host_ratio)) { |
99e314eb | 2551 | unsigned long page; |
dad45ab2 WY |
2552 | unsigned long fixup_start_addr = QEMU_ALIGN_DOWN(run_start, |
2553 | host_ratio); | |
2554 | run_start = QEMU_ALIGN_UP(run_start, host_ratio); | |
99e314eb | 2555 | |
99e314eb DDAG |
2556 | /* Clean up the bitmap */ |
2557 | for (page = fixup_start_addr; | |
2558 | page < fixup_start_addr + host_ratio; page++) { | |
99e314eb DDAG |
2559 | /* |
2560 | * Remark them as dirty, updating the count for any pages | |
2561 | * that weren't previously dirty. | |
2562 | */ | |
0d8ec885 | 2563 | rs->migration_dirty_pages += !test_and_set_bit(page, bitmap); |
99e314eb DDAG |
2564 | } |
2565 | } | |
2566 | ||
1e7cf8c3 WY |
2567 | /* Find the next dirty page for the next iteration */ |
2568 | run_start = find_next_bit(bitmap, pages, run_start); | |
99e314eb DDAG |
2569 | } |
2570 | } | |
2571 | ||
3d0684b2 JQ |
2572 | /** |
2573 | * ram_postcopy_send_discard_bitmap: transmit the discard bitmap | |
2574 | * | |
e0b266f0 DDAG |
2575 | * Transmit the set of pages to be discarded after precopy to the target |
2576 | * these are pages that: | |
2577 | * a) Have been previously transmitted but are now dirty again | |
2578 | * b) Pages that have never been transmitted, this ensures that | |
2579 | * any pages on the destination that have been mapped by background | |
2580 | * tasks get discarded (transparent huge pages is the specific concern) | |
2581 | * Hopefully this is pretty sparse | |
3d0684b2 JQ |
2582 | * |
2583 | * @ms: current migration state | |
e0b266f0 | 2584 | */ |
739fcc1b | 2585 | void ram_postcopy_send_discard_bitmap(MigrationState *ms) |
e0b266f0 | 2586 | { |
53518d94 | 2587 | RAMState *rs = ram_state; |
e0b266f0 | 2588 | |
89ac5a1d | 2589 | RCU_READ_LOCK_GUARD(); |
e0b266f0 DDAG |
2590 | |
2591 | /* This should be our last sync, the src is now paused */ | |
1e493be5 | 2592 | migration_bitmap_sync(rs, false); |
e0b266f0 | 2593 | |
6b6712ef | 2594 | /* Easiest way to make sure we don't resume in the middle of a host-page */ |
ec6f3ab9 | 2595 | rs->pss[RAM_CHANNEL_PRECOPY].last_sent_block = NULL; |
6b6712ef | 2596 | rs->last_seen_block = NULL; |
6b6712ef | 2597 | rs->last_page = 0; |
e0b266f0 | 2598 | |
739fcc1b | 2599 | postcopy_each_ram_send_discard(ms); |
e0b266f0 | 2600 | |
739fcc1b | 2601 | trace_ram_postcopy_send_discard_bitmap(); |
e0b266f0 DDAG |
2602 | } |
2603 | ||
3d0684b2 JQ |
2604 | /** |
2605 | * ram_discard_range: discard dirtied pages at the beginning of postcopy | |
e0b266f0 | 2606 | * |
3d0684b2 | 2607 | * Returns zero on success |
e0b266f0 | 2608 | * |
36449157 JQ |
2609 | * @rbname: name of the RAMBlock of the request. NULL means the |
2610 | * same that last one. | |
3d0684b2 JQ |
2611 | * @start: RAMBlock starting page |
2612 | * @length: RAMBlock size | |
e0b266f0 | 2613 | */ |
aaa2064c | 2614 | int ram_discard_range(const char *rbname, uint64_t start, size_t length) |
e0b266f0 | 2615 | { |
36449157 | 2616 | trace_ram_discard_range(rbname, start, length); |
d3a5038c | 2617 | |
89ac5a1d | 2618 | RCU_READ_LOCK_GUARD(); |
36449157 | 2619 | RAMBlock *rb = qemu_ram_block_by_name(rbname); |
e0b266f0 DDAG |
2620 | |
2621 | if (!rb) { | |
36449157 | 2622 | error_report("ram_discard_range: Failed to find block '%s'", rbname); |
03acb4e9 | 2623 | return -1; |
e0b266f0 DDAG |
2624 | } |
2625 | ||
814bb08f PX |
2626 | /* |
2627 | * On source VM, we don't need to update the received bitmap since | |
2628 | * we don't even have one. | |
2629 | */ | |
2630 | if (rb->receivedmap) { | |
2631 | bitmap_clear(rb->receivedmap, start >> qemu_target_page_bits(), | |
2632 | length >> qemu_target_page_bits()); | |
2633 | } | |
2634 | ||
03acb4e9 | 2635 | return ram_block_discard_range(rb, start, length); |
e0b266f0 DDAG |
2636 | } |
2637 | ||
84593a08 PX |
2638 | /* |
2639 | * For every allocation, we will try not to crash the VM if the | |
2640 | * allocation failed. | |
2641 | */ | |
7bee8ba8 | 2642 | static bool xbzrle_init(Error **errp) |
84593a08 | 2643 | { |
87dca0c9 | 2644 | if (!migrate_xbzrle()) { |
7bee8ba8 | 2645 | return true; |
84593a08 PX |
2646 | } |
2647 | ||
2648 | XBZRLE_cache_lock(); | |
2649 | ||
2650 | XBZRLE.zero_target_page = g_try_malloc0(TARGET_PAGE_SIZE); | |
2651 | if (!XBZRLE.zero_target_page) { | |
7bee8ba8 | 2652 | error_setg(errp, "%s: Error allocating zero page", __func__); |
84593a08 PX |
2653 | goto err_out; |
2654 | } | |
2655 | ||
2656 | XBZRLE.cache = cache_init(migrate_xbzrle_cache_size(), | |
7bee8ba8 | 2657 | TARGET_PAGE_SIZE, errp); |
84593a08 | 2658 | if (!XBZRLE.cache) { |
84593a08 PX |
2659 | goto free_zero_page; |
2660 | } | |
2661 | ||
2662 | XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); | |
2663 | if (!XBZRLE.encoded_buf) { | |
7bee8ba8 | 2664 | error_setg(errp, "%s: Error allocating encoded_buf", __func__); |
84593a08 PX |
2665 | goto free_cache; |
2666 | } | |
2667 | ||
2668 | XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); | |
2669 | if (!XBZRLE.current_buf) { | |
7bee8ba8 | 2670 | error_setg(errp, "%s: Error allocating current_buf", __func__); |
84593a08 PX |
2671 | goto free_encoded_buf; |
2672 | } | |
2673 | ||
2674 | /* We are all good */ | |
2675 | XBZRLE_cache_unlock(); | |
7bee8ba8 | 2676 | return true; |
84593a08 PX |
2677 | |
2678 | free_encoded_buf: | |
2679 | g_free(XBZRLE.encoded_buf); | |
2680 | XBZRLE.encoded_buf = NULL; | |
2681 | free_cache: | |
2682 | cache_fini(XBZRLE.cache); | |
2683 | XBZRLE.cache = NULL; | |
2684 | free_zero_page: | |
2685 | g_free(XBZRLE.zero_target_page); | |
2686 | XBZRLE.zero_target_page = NULL; | |
2687 | err_out: | |
2688 | XBZRLE_cache_unlock(); | |
7bee8ba8 | 2689 | return false; |
84593a08 PX |
2690 | } |
2691 | ||
16ecd25a | 2692 | static bool ram_state_init(RAMState **rsp, Error **errp) |
56e93d26 | 2693 | { |
7d00ee6a PX |
2694 | *rsp = g_try_new0(RAMState, 1); |
2695 | ||
2696 | if (!*rsp) { | |
16ecd25a CLG |
2697 | error_setg(errp, "%s: Init ramstate fail", __func__); |
2698 | return false; | |
7d00ee6a | 2699 | } |
53518d94 JQ |
2700 | |
2701 | qemu_mutex_init(&(*rsp)->bitmap_mutex); | |
2702 | qemu_mutex_init(&(*rsp)->src_page_req_mutex); | |
2703 | QSIMPLEQ_INIT(&(*rsp)->src_page_requests); | |
8d80e195 | 2704 | (*rsp)->ram_bytes_total = ram_bytes_total(); |
56e93d26 | 2705 | |
7d00ee6a | 2706 | /* |
40c4d4a8 IR |
2707 | * Count the total number of pages used by ram blocks not including any |
2708 | * gaps due to alignment or unplugs. | |
03158519 | 2709 | * This must match with the initial values of dirty bitmap. |
7d00ee6a | 2710 | */ |
8d80e195 | 2711 | (*rsp)->migration_dirty_pages = (*rsp)->ram_bytes_total >> TARGET_PAGE_BITS; |
7d00ee6a PX |
2712 | ram_state_reset(*rsp); |
2713 | ||
16ecd25a | 2714 | return true; |
7d00ee6a PX |
2715 | } |
2716 | ||
d6eff5d7 | 2717 | static void ram_list_init_bitmaps(void) |
7d00ee6a | 2718 | { |
002cad6b | 2719 | MigrationState *ms = migrate_get_current(); |
d6eff5d7 PX |
2720 | RAMBlock *block; |
2721 | unsigned long pages; | |
002cad6b | 2722 | uint8_t shift; |
56e93d26 | 2723 | |
0827b9e9 AA |
2724 | /* Skip setting bitmap if there is no RAM */ |
2725 | if (ram_bytes_total()) { | |
002cad6b PX |
2726 | shift = ms->clear_bitmap_shift; |
2727 | if (shift > CLEAR_BITMAP_SHIFT_MAX) { | |
2728 | error_report("clear_bitmap_shift (%u) too big, using " | |
2729 | "max value (%u)", shift, CLEAR_BITMAP_SHIFT_MAX); | |
2730 | shift = CLEAR_BITMAP_SHIFT_MAX; | |
2731 | } else if (shift < CLEAR_BITMAP_SHIFT_MIN) { | |
2732 | error_report("clear_bitmap_shift (%u) too small, using " | |
2733 | "min value (%u)", shift, CLEAR_BITMAP_SHIFT_MIN); | |
2734 | shift = CLEAR_BITMAP_SHIFT_MIN; | |
2735 | } | |
2736 | ||
fbd162e6 | 2737 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
d6eff5d7 | 2738 | pages = block->max_length >> TARGET_PAGE_BITS; |
03158519 WY |
2739 | /* |
2740 | * The initial dirty bitmap for migration must be set with all | |
2741 | * ones to make sure we'll migrate every guest RAM page to | |
2742 | * destination. | |
40c4d4a8 IR |
2743 | * Here we set RAMBlock.bmap all to 1 because when rebegin a |
2744 | * new migration after a failed migration, ram_list. | |
2745 | * dirty_memory[DIRTY_MEMORY_MIGRATION] don't include the whole | |
2746 | * guest memory. | |
03158519 | 2747 | */ |
6b6712ef | 2748 | block->bmap = bitmap_new(pages); |
40c4d4a8 | 2749 | bitmap_set(block->bmap, 0, pages); |
c2d5c4a7 FR |
2750 | if (migrate_mapped_ram()) { |
2751 | block->file_bmap = bitmap_new(pages); | |
2752 | } | |
002cad6b PX |
2753 | block->clear_bmap_shift = shift; |
2754 | block->clear_bmap = bitmap_new(clear_bmap_size(pages, shift)); | |
0827b9e9 | 2755 | } |
f3f491fc | 2756 | } |
d6eff5d7 PX |
2757 | } |
2758 | ||
be39b4cd DH |
2759 | static void migration_bitmap_clear_discarded_pages(RAMState *rs) |
2760 | { | |
2761 | unsigned long pages; | |
2762 | RAMBlock *rb; | |
2763 | ||
2764 | RCU_READ_LOCK_GUARD(); | |
2765 | ||
2766 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { | |
2767 | pages = ramblock_dirty_bitmap_clear_discarded_pages(rb); | |
2768 | rs->migration_dirty_pages -= pages; | |
2769 | } | |
2770 | } | |
2771 | ||
030b56b2 | 2772 | static bool ram_init_bitmaps(RAMState *rs, Error **errp) |
d6eff5d7 | 2773 | { |
639ec3fb CLG |
2774 | bool ret = true; |
2775 | ||
d6eff5d7 | 2776 | qemu_mutex_lock_ramlist(); |
f3f491fc | 2777 | |
89ac5a1d DDAG |
2778 | WITH_RCU_READ_LOCK_GUARD() { |
2779 | ram_list_init_bitmaps(); | |
278e2f55 AG |
2780 | /* We don't use dirty log with background snapshots */ |
2781 | if (!migrate_background_snapshot()) { | |
030b56b2 | 2782 | ret = memory_global_dirty_log_start(GLOBAL_DIRTY_MIGRATION, errp); |
639ec3fb | 2783 | if (!ret) { |
639ec3fb CLG |
2784 | goto out_unlock; |
2785 | } | |
1e493be5 | 2786 | migration_bitmap_sync_precopy(rs, false); |
278e2f55 | 2787 | } |
89ac5a1d | 2788 | } |
639ec3fb | 2789 | out_unlock: |
56e93d26 | 2790 | qemu_mutex_unlock_ramlist(); |
be39b4cd | 2791 | |
639ec3fb CLG |
2792 | if (!ret) { |
2793 | ram_bitmaps_destroy(); | |
030b56b2 | 2794 | return false; |
639ec3fb CLG |
2795 | } |
2796 | ||
be39b4cd DH |
2797 | /* |
2798 | * After an eventual first bitmap sync, fixup the initial bitmap | |
2799 | * containing all 1s to exclude any discarded pages from migration. | |
2800 | */ | |
2801 | migration_bitmap_clear_discarded_pages(rs); | |
030b56b2 | 2802 | return true; |
d6eff5d7 PX |
2803 | } |
2804 | ||
030b56b2 | 2805 | static int ram_init_all(RAMState **rsp, Error **errp) |
d6eff5d7 | 2806 | { |
030b56b2 | 2807 | if (!ram_state_init(rsp, errp)) { |
d6eff5d7 PX |
2808 | return -1; |
2809 | } | |
2810 | ||
030b56b2 | 2811 | if (!xbzrle_init(errp)) { |
d6eff5d7 PX |
2812 | ram_state_cleanup(rsp); |
2813 | return -1; | |
2814 | } | |
2815 | ||
030b56b2 CLG |
2816 | if (!ram_init_bitmaps(*rsp, errp)) { |
2817 | return -1; | |
2818 | } | |
a91246c9 HZ |
2819 | |
2820 | return 0; | |
2821 | } | |
2822 | ||
08614f34 PX |
2823 | static void ram_state_resume_prepare(RAMState *rs, QEMUFile *out) |
2824 | { | |
2825 | RAMBlock *block; | |
2826 | uint64_t pages = 0; | |
2827 | ||
2828 | /* | |
2829 | * Postcopy is not using xbzrle/compression, so no need for that. | |
2830 | * Also, since source are already halted, we don't need to care | |
2831 | * about dirty page logging as well. | |
2832 | */ | |
2833 | ||
fbd162e6 | 2834 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
08614f34 PX |
2835 | pages += bitmap_count_one(block->bmap, |
2836 | block->used_length >> TARGET_PAGE_BITS); | |
2837 | } | |
2838 | ||
2839 | /* This may not be aligned with current bitmaps. Recalculate. */ | |
2840 | rs->migration_dirty_pages = pages; | |
2841 | ||
1a373522 | 2842 | ram_state_reset(rs); |
08614f34 PX |
2843 | |
2844 | /* Update RAMState cache of output QEMUFile */ | |
7f401b80 | 2845 | rs->pss[RAM_CHANNEL_PRECOPY].pss_channel = out; |
08614f34 PX |
2846 | |
2847 | trace_ram_state_resume_prepare(pages); | |
2848 | } | |
2849 | ||
6bcb05fc WW |
2850 | /* |
2851 | * This function clears bits of the free pages reported by the caller from the | |
2852 | * migration dirty bitmap. @addr is the host address corresponding to the | |
2853 | * start of the continuous guest free pages, and @len is the total bytes of | |
2854 | * those pages. | |
2855 | */ | |
2856 | void qemu_guest_free_page_hint(void *addr, size_t len) | |
2857 | { | |
2858 | RAMBlock *block; | |
2859 | ram_addr_t offset; | |
2860 | size_t used_len, start, npages; | |
6bcb05fc WW |
2861 | |
2862 | /* This function is currently expected to be used during live migration */ | |
7dcb3c87 | 2863 | if (!migration_is_setup_or_active()) { |
6bcb05fc WW |
2864 | return; |
2865 | } | |
2866 | ||
2867 | for (; len > 0; len -= used_len, addr += used_len) { | |
2868 | block = qemu_ram_block_from_host(addr, false, &offset); | |
2869 | if (unlikely(!block || offset >= block->used_length)) { | |
2870 | /* | |
2871 | * The implementation might not support RAMBlock resize during | |
2872 | * live migration, but it could happen in theory with future | |
2873 | * updates. So we add a check here to capture that case. | |
2874 | */ | |
2875 | error_report_once("%s unexpected error", __func__); | |
2876 | return; | |
2877 | } | |
2878 | ||
2879 | if (len <= block->used_length - offset) { | |
2880 | used_len = len; | |
2881 | } else { | |
2882 | used_len = block->used_length - offset; | |
2883 | } | |
2884 | ||
2885 | start = offset >> TARGET_PAGE_BITS; | |
2886 | npages = used_len >> TARGET_PAGE_BITS; | |
2887 | ||
2888 | qemu_mutex_lock(&ram_state->bitmap_mutex); | |
3143577d WW |
2889 | /* |
2890 | * The skipped free pages are equavalent to be sent from clear_bmap's | |
2891 | * perspective, so clear the bits from the memory region bitmap which | |
2892 | * are initially set. Otherwise those skipped pages will be sent in | |
2893 | * the next round after syncing from the memory region bitmap. | |
2894 | */ | |
1230a25f | 2895 | migration_clear_memory_region_dirty_bitmap_range(block, start, npages); |
6bcb05fc WW |
2896 | ram_state->migration_dirty_pages -= |
2897 | bitmap_count_one_with_offset(block->bmap, start, npages); | |
2898 | bitmap_clear(block->bmap, start, npages); | |
2899 | qemu_mutex_unlock(&ram_state->bitmap_mutex); | |
2900 | } | |
2901 | } | |
2902 | ||
c2d5c4a7 FR |
2903 | #define MAPPED_RAM_HDR_VERSION 1 |
2904 | struct MappedRamHeader { | |
2905 | uint32_t version; | |
2906 | /* | |
2907 | * The target's page size, so we know how many pages are in the | |
2908 | * bitmap. | |
2909 | */ | |
2910 | uint64_t page_size; | |
2911 | /* | |
2912 | * The offset in the migration file where the pages bitmap is | |
2913 | * stored. | |
2914 | */ | |
2915 | uint64_t bitmap_offset; | |
2916 | /* | |
2917 | * The offset in the migration file where the actual pages (data) | |
2918 | * are stored. | |
2919 | */ | |
2920 | uint64_t pages_offset; | |
2921 | } QEMU_PACKED; | |
2922 | typedef struct MappedRamHeader MappedRamHeader; | |
2923 | ||
2924 | static void mapped_ram_setup_ramblock(QEMUFile *file, RAMBlock *block) | |
2925 | { | |
2926 | g_autofree MappedRamHeader *header = NULL; | |
2927 | size_t header_size, bitmap_size; | |
2928 | long num_pages; | |
2929 | ||
2930 | header = g_new0(MappedRamHeader, 1); | |
2931 | header_size = sizeof(MappedRamHeader); | |
2932 | ||
2933 | num_pages = block->used_length >> TARGET_PAGE_BITS; | |
2934 | bitmap_size = BITS_TO_LONGS(num_pages) * sizeof(unsigned long); | |
2935 | ||
2936 | /* | |
2937 | * Save the file offsets of where the bitmap and the pages should | |
2938 | * go as they are written at the end of migration and during the | |
2939 | * iterative phase, respectively. | |
2940 | */ | |
2941 | block->bitmap_offset = qemu_get_offset(file) + header_size; | |
2942 | block->pages_offset = ROUND_UP(block->bitmap_offset + | |
2943 | bitmap_size, | |
2944 | MAPPED_RAM_FILE_OFFSET_ALIGNMENT); | |
2945 | ||
2946 | header->version = cpu_to_be32(MAPPED_RAM_HDR_VERSION); | |
2947 | header->page_size = cpu_to_be64(TARGET_PAGE_SIZE); | |
2948 | header->bitmap_offset = cpu_to_be64(block->bitmap_offset); | |
2949 | header->pages_offset = cpu_to_be64(block->pages_offset); | |
2950 | ||
2951 | qemu_put_buffer(file, (uint8_t *) header, header_size); | |
2952 | ||
2953 | /* prepare offset for next ramblock */ | |
2954 | qemu_set_offset(file, block->pages_offset + block->used_length, SEEK_SET); | |
2955 | } | |
2956 | ||
2f6b8826 FR |
2957 | static bool mapped_ram_read_header(QEMUFile *file, MappedRamHeader *header, |
2958 | Error **errp) | |
2959 | { | |
2960 | size_t ret, header_size = sizeof(MappedRamHeader); | |
2961 | ||
2962 | ret = qemu_get_buffer(file, (uint8_t *)header, header_size); | |
2963 | if (ret != header_size) { | |
2964 | error_setg(errp, "Could not read whole mapped-ram migration header " | |
2965 | "(expected %zd, got %zd bytes)", header_size, ret); | |
2966 | return false; | |
2967 | } | |
2968 | ||
2969 | /* migration stream is big-endian */ | |
2970 | header->version = be32_to_cpu(header->version); | |
2971 | ||
2972 | if (header->version > MAPPED_RAM_HDR_VERSION) { | |
2973 | error_setg(errp, "Migration mapped-ram capability version not " | |
2974 | "supported (expected <= %d, got %d)", MAPPED_RAM_HDR_VERSION, | |
2975 | header->version); | |
2976 | return false; | |
2977 | } | |
2978 | ||
2979 | header->page_size = be64_to_cpu(header->page_size); | |
2980 | header->bitmap_offset = be64_to_cpu(header->bitmap_offset); | |
2981 | header->pages_offset = be64_to_cpu(header->pages_offset); | |
2982 | ||
2983 | return true; | |
2984 | } | |
2985 | ||
3d0684b2 JQ |
2986 | /* |
2987 | * Each of ram_save_setup, ram_save_iterate and ram_save_complete has | |
a91246c9 HZ |
2988 | * long-running RCU critical section. When rcu-reclaims in the code |
2989 | * start to become numerous it will be necessary to reduce the | |
2990 | * granularity of these critical sections. | |
2991 | */ | |
2992 | ||
3d0684b2 JQ |
2993 | /** |
2994 | * ram_save_setup: Setup RAM for migration | |
2995 | * | |
2996 | * Returns zero to indicate success and negative for error | |
2997 | * | |
2998 | * @f: QEMUFile where to send the data | |
2999 | * @opaque: RAMState pointer | |
01c3ac68 | 3000 | * @errp: pointer to Error*, to store an error if it happens. |
3d0684b2 | 3001 | */ |
01c3ac68 | 3002 | static int ram_save_setup(QEMUFile *f, void *opaque, Error **errp) |
a91246c9 | 3003 | { |
53518d94 | 3004 | RAMState **rsp = opaque; |
a91246c9 | 3005 | RAMBlock *block; |
5d220369 | 3006 | int ret, max_hg_page_size; |
a91246c9 HZ |
3007 | |
3008 | /* migration has already setup the bitmap, reuse it. */ | |
3009 | if (!migration_in_colo_state()) { | |
030b56b2 | 3010 | if (ram_init_all(rsp, errp) != 0) { |
a91246c9 | 3011 | return -1; |
53518d94 | 3012 | } |
a91246c9 | 3013 | } |
7f401b80 | 3014 | (*rsp)->pss[RAM_CHANNEL_PRECOPY].pss_channel = f; |
a91246c9 | 3015 | |
5d220369 RH |
3016 | /* |
3017 | * ??? Mirrors the previous value of qemu_host_page_size, | |
3018 | * but is this really what was intended for the migration? | |
3019 | */ | |
3020 | max_hg_page_size = MAX(qemu_real_host_page_size(), TARGET_PAGE_SIZE); | |
3021 | ||
0e6ebd48 | 3022 | WITH_RCU_READ_LOCK_GUARD() { |
8008a272 JQ |
3023 | qemu_put_be64(f, ram_bytes_total_with_ignored() |
3024 | | RAM_SAVE_FLAG_MEM_SIZE); | |
56e93d26 | 3025 | |
0e6ebd48 DDAG |
3026 | RAMBLOCK_FOREACH_MIGRATABLE(block) { |
3027 | qemu_put_byte(f, strlen(block->idstr)); | |
3028 | qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); | |
3029 | qemu_put_be64(f, block->used_length); | |
5d220369 RH |
3030 | if (migrate_postcopy_ram() && |
3031 | block->page_size != max_hg_page_size) { | |
0e6ebd48 DDAG |
3032 | qemu_put_be64(f, block->page_size); |
3033 | } | |
3034 | if (migrate_ignore_shared()) { | |
3035 | qemu_put_be64(f, block->mr->addr); | |
3036 | } | |
c2d5c4a7 FR |
3037 | |
3038 | if (migrate_mapped_ram()) { | |
3039 | mapped_ram_setup_ramblock(f, block); | |
3040 | } | |
fbd162e6 | 3041 | } |
56e93d26 JQ |
3042 | } |
3043 | ||
b1b38387 | 3044 | ret = rdma_registration_start(f, RAM_CONTROL_SETUP); |
48408174 | 3045 | if (ret < 0) { |
01c3ac68 | 3046 | error_setg(errp, "%s: failed to start RDMA registration", __func__); |
48408174 | 3047 | qemu_file_set_error(f, ret); |
a2326705 | 3048 | return ret; |
48408174 | 3049 | } |
5f5b8858 | 3050 | |
b1b38387 | 3051 | ret = rdma_registration_stop(f, RAM_CONTROL_SETUP); |
5f5b8858 | 3052 | if (ret < 0) { |
01c3ac68 | 3053 | error_setg(errp, "%s: failed to stop RDMA registration", __func__); |
5f5b8858 | 3054 | qemu_file_set_error(f, ret); |
a2326705 | 3055 | return ret; |
5f5b8858 | 3056 | } |
56e93d26 | 3057 | |
4010ba38 | 3058 | migration_ops = g_malloc0(sizeof(MigrationOps)); |
9ae90f73 HX |
3059 | |
3060 | if (migrate_multifd()) { | |
3061 | migration_ops->ram_save_target_page = ram_save_target_page_multifd; | |
3062 | } else { | |
3063 | migration_ops->ram_save_target_page = ram_save_target_page_legacy; | |
3064 | } | |
930e239d | 3065 | |
195801d7 | 3066 | bql_unlock(); |
9346fa18 | 3067 | ret = multifd_send_sync_main(); |
195801d7 | 3068 | bql_lock(); |
33d70973 | 3069 | if (ret < 0) { |
01c3ac68 | 3070 | error_setg(errp, "%s: multifd synchronization failed", __func__); |
33d70973 LB |
3071 | return ret; |
3072 | } | |
3073 | ||
9d01778a FR |
3074 | if (migrate_multifd() && !migrate_multifd_flush_after_each_section() |
3075 | && !migrate_mapped_ram()) { | |
294e5a40 JQ |
3076 | qemu_put_be64(f, RAM_SAVE_FLAG_MULTIFD_FLUSH); |
3077 | } | |
3078 | ||
56e93d26 | 3079 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
76936bbc CLG |
3080 | ret = qemu_fflush(f); |
3081 | if (ret < 0) { | |
01c3ac68 | 3082 | error_setg_errno(errp, -ret, "%s failed", __func__); |
76936bbc CLG |
3083 | } |
3084 | return ret; | |
56e93d26 JQ |
3085 | } |
3086 | ||
c2d5c4a7 FR |
3087 | static void ram_save_file_bmap(QEMUFile *f) |
3088 | { | |
3089 | RAMBlock *block; | |
3090 | ||
3091 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
3092 | long num_pages = block->used_length >> TARGET_PAGE_BITS; | |
3093 | long bitmap_size = BITS_TO_LONGS(num_pages) * sizeof(unsigned long); | |
3094 | ||
3095 | qemu_put_buffer_at(f, (uint8_t *)block->file_bmap, bitmap_size, | |
3096 | block->bitmap_offset); | |
3097 | ram_transferred_add(bitmap_size); | |
f427d90b FR |
3098 | |
3099 | /* | |
3100 | * Free the bitmap here to catch any synchronization issues | |
3101 | * with multifd channels. No channels should be sending pages | |
3102 | * after we've written the bitmap to file. | |
3103 | */ | |
3104 | g_free(block->file_bmap); | |
3105 | block->file_bmap = NULL; | |
c2d5c4a7 FR |
3106 | } |
3107 | } | |
3108 | ||
c3cdf3fb | 3109 | void ramblock_set_file_bmap_atomic(RAMBlock *block, ram_addr_t offset, bool set) |
f427d90b | 3110 | { |
c3cdf3fb FR |
3111 | if (set) { |
3112 | set_bit_atomic(offset >> TARGET_PAGE_BITS, block->file_bmap); | |
3113 | } else { | |
3114 | clear_bit_atomic(offset >> TARGET_PAGE_BITS, block->file_bmap); | |
3115 | } | |
f427d90b FR |
3116 | } |
3117 | ||
3d0684b2 JQ |
3118 | /** |
3119 | * ram_save_iterate: iterative stage for migration | |
3120 | * | |
3121 | * Returns zero to indicate success and negative for error | |
3122 | * | |
3123 | * @f: QEMUFile where to send the data | |
3124 | * @opaque: RAMState pointer | |
3125 | */ | |
56e93d26 JQ |
3126 | static int ram_save_iterate(QEMUFile *f, void *opaque) |
3127 | { | |
53518d94 JQ |
3128 | RAMState **temp = opaque; |
3129 | RAMState *rs = *temp; | |
3d4095b2 | 3130 | int ret = 0; |
56e93d26 JQ |
3131 | int i; |
3132 | int64_t t0; | |
5c90308f | 3133 | int done = 0; |
56e93d26 | 3134 | |
63268c49 PX |
3135 | /* |
3136 | * We'll take this lock a little bit long, but it's okay for two reasons. | |
3137 | * Firstly, the only possible other thread to take it is who calls | |
3138 | * qemu_guest_free_page_hint(), which should be rare; secondly, see | |
3139 | * MAX_WAIT (if curious, further see commit 4508bd9ed8053ce) below, which | |
3140 | * guarantees that we'll at least released it in a regular basis. | |
3141 | */ | |
0983125b JQ |
3142 | WITH_QEMU_LOCK_GUARD(&rs->bitmap_mutex) { |
3143 | WITH_RCU_READ_LOCK_GUARD() { | |
3144 | if (ram_list.version != rs->last_version) { | |
3145 | ram_state_reset(rs); | |
3146 | } | |
56e93d26 | 3147 | |
0983125b JQ |
3148 | /* Read version before ram_list.blocks */ |
3149 | smp_rmb(); | |
56e93d26 | 3150 | |
0983125b JQ |
3151 | ret = rdma_registration_start(f, RAM_CONTROL_ROUND); |
3152 | if (ret < 0) { | |
3153 | qemu_file_set_error(f, ret); | |
3154 | goto out; | |
3155 | } | |
56e93d26 | 3156 | |
0983125b JQ |
3157 | t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
3158 | i = 0; | |
3159 | while ((ret = migration_rate_exceeded(f)) == 0 || | |
3160 | postcopy_has_request(rs)) { | |
3161 | int pages; | |
e03a34f8 | 3162 | |
0983125b JQ |
3163 | if (qemu_file_get_error(f)) { |
3164 | break; | |
3165 | } | |
e8f3735f | 3166 | |
0983125b JQ |
3167 | pages = ram_find_and_save_block(rs); |
3168 | /* no more pages to sent */ | |
3169 | if (pages == 0) { | |
3170 | done = 1; | |
3171 | break; | |
3172 | } | |
e8f3735f | 3173 | |
0983125b JQ |
3174 | if (pages < 0) { |
3175 | qemu_file_set_error(f, pages); | |
3176 | break; | |
3177 | } | |
89ac5a1d | 3178 | |
0983125b | 3179 | rs->target_page_count += pages; |
89ac5a1d | 3180 | |
0983125b JQ |
3181 | /* |
3182 | * we want to check in the 1st loop, just in case it was the 1st | |
3183 | * time and we had to sync the dirty bitmap. | |
3184 | * qemu_clock_get_ns() is a bit expensive, so we only check each | |
3185 | * some iterations | |
3186 | */ | |
3187 | if ((i & 63) == 0) { | |
3188 | uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / | |
3189 | 1000000; | |
3190 | if (t1 > MAX_WAIT) { | |
3191 | trace_ram_save_iterate_big_wait(t1, i); | |
3192 | break; | |
3193 | } | |
89ac5a1d | 3194 | } |
0983125b | 3195 | i++; |
89ac5a1d | 3196 | } |
56e93d26 | 3197 | } |
56e93d26 | 3198 | } |
56e93d26 JQ |
3199 | |
3200 | /* | |
3201 | * Must occur before EOS (or any QEMUFile operation) | |
3202 | * because of RDMA protocol. | |
3203 | */ | |
b1b38387 | 3204 | ret = rdma_registration_stop(f, RAM_CONTROL_ROUND); |
5f5b8858 JQ |
3205 | if (ret < 0) { |
3206 | qemu_file_set_error(f, ret); | |
3207 | } | |
56e93d26 | 3208 | |
b2557345 | 3209 | out: |
b69a0227 | 3210 | if (ret >= 0 |
7dcb3c87 | 3211 | && migration_is_setup_or_active()) { |
9d01778a FR |
3212 | if (migrate_multifd() && migrate_multifd_flush_after_each_section() && |
3213 | !migrate_mapped_ram()) { | |
9346fa18 | 3214 | ret = multifd_send_sync_main(); |
b05292c2 JQ |
3215 | if (ret < 0) { |
3216 | return ret; | |
3217 | } | |
33d70973 LB |
3218 | } |
3219 | ||
3d4095b2 | 3220 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
4c2d0f6d | 3221 | ram_transferred_add(8); |
be07a0ed | 3222 | ret = qemu_fflush(f); |
3d4095b2 | 3223 | } |
56e93d26 JQ |
3224 | if (ret < 0) { |
3225 | return ret; | |
3226 | } | |
3227 | ||
5c90308f | 3228 | return done; |
56e93d26 JQ |
3229 | } |
3230 | ||
3d0684b2 JQ |
3231 | /** |
3232 | * ram_save_complete: function called to send the remaining amount of ram | |
3233 | * | |
e8f3735f | 3234 | * Returns zero to indicate success or negative on error |
3d0684b2 | 3235 | * |
a4a411fb | 3236 | * Called with the BQL |
3d0684b2 JQ |
3237 | * |
3238 | * @f: QEMUFile where to send the data | |
3239 | * @opaque: RAMState pointer | |
3240 | */ | |
56e93d26 JQ |
3241 | static int ram_save_complete(QEMUFile *f, void *opaque) |
3242 | { | |
53518d94 JQ |
3243 | RAMState **temp = opaque; |
3244 | RAMState *rs = *temp; | |
e8f3735f | 3245 | int ret = 0; |
6f37bb8b | 3246 | |
05931ec5 JQ |
3247 | rs->last_stage = !migration_in_colo_state(); |
3248 | ||
89ac5a1d DDAG |
3249 | WITH_RCU_READ_LOCK_GUARD() { |
3250 | if (!migration_in_postcopy()) { | |
1e493be5 | 3251 | migration_bitmap_sync_precopy(rs, true); |
89ac5a1d | 3252 | } |
56e93d26 | 3253 | |
b1b38387 | 3254 | ret = rdma_registration_start(f, RAM_CONTROL_FINISH); |
48408174 JQ |
3255 | if (ret < 0) { |
3256 | qemu_file_set_error(f, ret); | |
a2326705 | 3257 | return ret; |
48408174 | 3258 | } |
56e93d26 | 3259 | |
89ac5a1d | 3260 | /* try transferring iterative blocks of memory */ |
56e93d26 | 3261 | |
89ac5a1d | 3262 | /* flush all remaining blocks regardless of rate limiting */ |
c13221b5 | 3263 | qemu_mutex_lock(&rs->bitmap_mutex); |
89ac5a1d DDAG |
3264 | while (true) { |
3265 | int pages; | |
56e93d26 | 3266 | |
05931ec5 | 3267 | pages = ram_find_and_save_block(rs); |
89ac5a1d DDAG |
3268 | /* no more blocks to sent */ |
3269 | if (pages == 0) { | |
3270 | break; | |
3271 | } | |
3272 | if (pages < 0) { | |
a2326705 PX |
3273 | qemu_mutex_unlock(&rs->bitmap_mutex); |
3274 | return pages; | |
89ac5a1d | 3275 | } |
e8f3735f | 3276 | } |
c13221b5 | 3277 | qemu_mutex_unlock(&rs->bitmap_mutex); |
56e93d26 | 3278 | |
a2326705 PX |
3279 | ret = rdma_registration_stop(f, RAM_CONTROL_FINISH); |
3280 | if (ret < 0) { | |
3281 | qemu_file_set_error(f, ret); | |
3282 | return ret; | |
5f5b8858 | 3283 | } |
89ac5a1d | 3284 | } |
d09a6fde | 3285 | |
9346fa18 | 3286 | ret = multifd_send_sync_main(); |
33d70973 LB |
3287 | if (ret < 0) { |
3288 | return ret; | |
3289 | } | |
3290 | ||
c2d5c4a7 FR |
3291 | if (migrate_mapped_ram()) { |
3292 | ram_save_file_bmap(f); | |
3293 | ||
3294 | if (qemu_file_get_error(f)) { | |
3295 | Error *local_err = NULL; | |
3296 | int err = qemu_file_get_error_obj(f, &local_err); | |
3297 | ||
3298 | error_reportf_err(local_err, "Failed to write bitmap to file: "); | |
3299 | return -err; | |
3300 | } | |
3301 | } | |
3302 | ||
9d01778a FR |
3303 | if (migrate_multifd() && !migrate_multifd_flush_after_each_section() && |
3304 | !migrate_mapped_ram()) { | |
294e5a40 JQ |
3305 | qemu_put_be64(f, RAM_SAVE_FLAG_MULTIFD_FLUSH); |
3306 | } | |
33d70973 | 3307 | qemu_put_be64(f, RAM_SAVE_FLAG_EOS); |
be07a0ed | 3308 | return qemu_fflush(f); |
56e93d26 JQ |
3309 | } |
3310 | ||
24beea4e JQ |
3311 | static void ram_state_pending_estimate(void *opaque, uint64_t *must_precopy, |
3312 | uint64_t *can_postcopy) | |
56e93d26 | 3313 | { |
53518d94 JQ |
3314 | RAMState **temp = opaque; |
3315 | RAMState *rs = *temp; | |
56e93d26 | 3316 | |
c8df4a7a | 3317 | uint64_t remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
56e93d26 | 3318 | |
c8df4a7a JQ |
3319 | if (migrate_postcopy_ram()) { |
3320 | /* We can do postcopy, and all the data is postcopiable */ | |
24beea4e | 3321 | *can_postcopy += remaining_size; |
c8df4a7a | 3322 | } else { |
24beea4e | 3323 | *must_precopy += remaining_size; |
c8df4a7a JQ |
3324 | } |
3325 | } | |
3326 | ||
24beea4e JQ |
3327 | static void ram_state_pending_exact(void *opaque, uint64_t *must_precopy, |
3328 | uint64_t *can_postcopy) | |
c8df4a7a JQ |
3329 | { |
3330 | RAMState **temp = opaque; | |
3331 | RAMState *rs = *temp; | |
b0504edd | 3332 | uint64_t remaining_size; |
c8df4a7a | 3333 | |
b0504edd | 3334 | if (!migration_in_postcopy()) { |
195801d7 | 3335 | bql_lock(); |
89ac5a1d | 3336 | WITH_RCU_READ_LOCK_GUARD() { |
1e493be5 | 3337 | migration_bitmap_sync_precopy(rs, false); |
89ac5a1d | 3338 | } |
195801d7 | 3339 | bql_unlock(); |
56e93d26 | 3340 | } |
c31b098f | 3341 | |
b0504edd PX |
3342 | remaining_size = rs->migration_dirty_pages * TARGET_PAGE_SIZE; |
3343 | ||
86e1167e VSO |
3344 | if (migrate_postcopy_ram()) { |
3345 | /* We can do postcopy, and all the data is postcopiable */ | |
24beea4e | 3346 | *can_postcopy += remaining_size; |
86e1167e | 3347 | } else { |
24beea4e | 3348 | *must_precopy += remaining_size; |
86e1167e | 3349 | } |
56e93d26 JQ |
3350 | } |
3351 | ||
3352 | static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) | |
3353 | { | |
3354 | unsigned int xh_len; | |
3355 | int xh_flags; | |
063e760a | 3356 | uint8_t *loaded_data; |
56e93d26 | 3357 | |
56e93d26 JQ |
3358 | /* extract RLE header */ |
3359 | xh_flags = qemu_get_byte(f); | |
3360 | xh_len = qemu_get_be16(f); | |
3361 | ||
3362 | if (xh_flags != ENCODING_FLAG_XBZRLE) { | |
3363 | error_report("Failed to load XBZRLE page - wrong compression!"); | |
3364 | return -1; | |
3365 | } | |
3366 | ||
3367 | if (xh_len > TARGET_PAGE_SIZE) { | |
3368 | error_report("Failed to load XBZRLE page - len overflow!"); | |
3369 | return -1; | |
3370 | } | |
f265e0e4 | 3371 | loaded_data = XBZRLE.decoded_buf; |
56e93d26 | 3372 | /* load data and decode */ |
f265e0e4 | 3373 | /* it can change loaded_data to point to an internal buffer */ |
063e760a | 3374 | qemu_get_buffer_in_place(f, &loaded_data, xh_len); |
56e93d26 JQ |
3375 | |
3376 | /* decode RLE */ | |
063e760a | 3377 | if (xbzrle_decode_buffer(loaded_data, xh_len, host, |
56e93d26 JQ |
3378 | TARGET_PAGE_SIZE) == -1) { |
3379 | error_report("Failed to load XBZRLE page - decode error!"); | |
3380 | return -1; | |
3381 | } | |
3382 | ||
3383 | return 0; | |
3384 | } | |
3385 | ||
3d0684b2 JQ |
3386 | /** |
3387 | * ram_block_from_stream: read a RAMBlock id from the migration stream | |
3388 | * | |
3389 | * Must be called from within a rcu critical section. | |
3390 | * | |
56e93d26 | 3391 | * Returns a pointer from within the RCU-protected ram_list. |
a7180877 | 3392 | * |
755e8d7c | 3393 | * @mis: the migration incoming state pointer |
3d0684b2 JQ |
3394 | * @f: QEMUFile where to read the data from |
3395 | * @flags: Page flags (mostly to see if it's a continuation of previous block) | |
c01b16ed | 3396 | * @channel: the channel we're using |
a7180877 | 3397 | */ |
755e8d7c | 3398 | static inline RAMBlock *ram_block_from_stream(MigrationIncomingState *mis, |
c01b16ed PX |
3399 | QEMUFile *f, int flags, |
3400 | int channel) | |
56e93d26 | 3401 | { |
c01b16ed | 3402 | RAMBlock *block = mis->last_recv_block[channel]; |
56e93d26 JQ |
3403 | char id[256]; |
3404 | uint8_t len; | |
3405 | ||
3406 | if (flags & RAM_SAVE_FLAG_CONTINUE) { | |
4c4bad48 | 3407 | if (!block) { |
56e93d26 JQ |
3408 | error_report("Ack, bad migration stream!"); |
3409 | return NULL; | |
3410 | } | |
4c4bad48 | 3411 | return block; |
56e93d26 JQ |
3412 | } |
3413 | ||
3414 | len = qemu_get_byte(f); | |
3415 | qemu_get_buffer(f, (uint8_t *)id, len); | |
3416 | id[len] = 0; | |
3417 | ||
e3dd7493 | 3418 | block = qemu_ram_block_by_name(id); |
4c4bad48 HZ |
3419 | if (!block) { |
3420 | error_report("Can't find block %s", id); | |
3421 | return NULL; | |
56e93d26 JQ |
3422 | } |
3423 | ||
f161c88a | 3424 | if (migrate_ram_is_ignored(block)) { |
b895de50 CLG |
3425 | error_report("block %s should not be migrated !", id); |
3426 | return NULL; | |
3427 | } | |
3428 | ||
c01b16ed | 3429 | mis->last_recv_block[channel] = block; |
755e8d7c | 3430 | |
4c4bad48 HZ |
3431 | return block; |
3432 | } | |
3433 | ||
3434 | static inline void *host_from_ram_block_offset(RAMBlock *block, | |
3435 | ram_addr_t offset) | |
3436 | { | |
3437 | if (!offset_in_ramblock(block, offset)) { | |
3438 | return NULL; | |
3439 | } | |
3440 | ||
3441 | return block->host + offset; | |
56e93d26 JQ |
3442 | } |
3443 | ||
6a23f639 DH |
3444 | static void *host_page_from_ram_block_offset(RAMBlock *block, |
3445 | ram_addr_t offset) | |
3446 | { | |
3447 | /* Note: Explicitly no check against offset_in_ramblock(). */ | |
3448 | return (void *)QEMU_ALIGN_DOWN((uintptr_t)(block->host + offset), | |
3449 | block->page_size); | |
3450 | } | |
3451 | ||
3452 | static ram_addr_t host_page_offset_from_ram_block_offset(RAMBlock *block, | |
3453 | ram_addr_t offset) | |
3454 | { | |
3455 | return ((uintptr_t)block->host + offset) & (block->page_size - 1); | |
3456 | } | |
3457 | ||
871cfc54 LS |
3458 | void colo_record_bitmap(RAMBlock *block, ram_addr_t *normal, uint32_t pages) |
3459 | { | |
3460 | qemu_mutex_lock(&ram_state->bitmap_mutex); | |
3461 | for (int i = 0; i < pages; i++) { | |
3462 | ram_addr_t offset = normal[i]; | |
3463 | ram_state->migration_dirty_pages += !test_and_set_bit( | |
3464 | offset >> TARGET_PAGE_BITS, | |
3465 | block->bmap); | |
3466 | } | |
3467 | qemu_mutex_unlock(&ram_state->bitmap_mutex); | |
3468 | } | |
3469 | ||
13af18f2 | 3470 | static inline void *colo_cache_from_block_offset(RAMBlock *block, |
8af66371 | 3471 | ram_addr_t offset, bool record_bitmap) |
13af18f2 ZC |
3472 | { |
3473 | if (!offset_in_ramblock(block, offset)) { | |
3474 | return NULL; | |
3475 | } | |
3476 | if (!block->colo_cache) { | |
3477 | error_report("%s: colo_cache is NULL in block :%s", | |
3478 | __func__, block->idstr); | |
3479 | return NULL; | |
3480 | } | |
7d9acafa ZC |
3481 | |
3482 | /* | |
3483 | * During colo checkpoint, we need bitmap of these migrated pages. | |
3484 | * It help us to decide which pages in ram cache should be flushed | |
3485 | * into VM's RAM later. | |
3486 | */ | |
871cfc54 LS |
3487 | if (record_bitmap) { |
3488 | colo_record_bitmap(block, &offset, 1); | |
7d9acafa | 3489 | } |
13af18f2 ZC |
3490 | return block->colo_cache + offset; |
3491 | } | |
3492 | ||
3d0684b2 | 3493 | /** |
7091dabe | 3494 | * ram_handle_zero: handle the zero page case |
3d0684b2 | 3495 | * |
56e93d26 JQ |
3496 | * If a page (or a whole RDMA chunk) has been |
3497 | * determined to be zero, then zap it. | |
3d0684b2 JQ |
3498 | * |
3499 | * @host: host address for the zero page | |
3500 | * @ch: what the page is filled from. We only support zero | |
3501 | * @size: size of the zero page | |
56e93d26 | 3502 | */ |
7091dabe | 3503 | void ram_handle_zero(void *host, uint64_t size) |
56e93d26 | 3504 | { |
7091dabe JQ |
3505 | if (!buffer_is_zero(host, size)) { |
3506 | memset(host, 0, size); | |
56e93d26 JQ |
3507 | } |
3508 | } | |
3509 | ||
b70cb3b4 RL |
3510 | static void colo_init_ram_state(void) |
3511 | { | |
16ecd25a CLG |
3512 | Error *local_err = NULL; |
3513 | ||
3514 | if (!ram_state_init(&ram_state, &local_err)) { | |
3515 | error_report_err(local_err); | |
3516 | } | |
b70cb3b4 RL |
3517 | } |
3518 | ||
13af18f2 ZC |
3519 | /* |
3520 | * colo cache: this is for secondary VM, we cache the whole | |
3521 | * memory of the secondary VM, it is need to hold the global lock | |
3522 | * to call this helper. | |
3523 | */ | |
3524 | int colo_init_ram_cache(void) | |
3525 | { | |
3526 | RAMBlock *block; | |
3527 | ||
44901b5a PB |
3528 | WITH_RCU_READ_LOCK_GUARD() { |
3529 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3530 | block->colo_cache = qemu_anon_ram_alloc(block->used_length, | |
8dbe22c6 | 3531 | NULL, false, false); |
44901b5a PB |
3532 | if (!block->colo_cache) { |
3533 | error_report("%s: Can't alloc memory for COLO cache of block %s," | |
3534 | "size 0x" RAM_ADDR_FMT, __func__, block->idstr, | |
3535 | block->used_length); | |
3536 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3537 | if (block->colo_cache) { | |
3538 | qemu_anon_ram_free(block->colo_cache, block->used_length); | |
3539 | block->colo_cache = NULL; | |
3540 | } | |
89ac5a1d | 3541 | } |
44901b5a | 3542 | return -errno; |
89ac5a1d | 3543 | } |
e5fdf920 LS |
3544 | if (!machine_dump_guest_core(current_machine)) { |
3545 | qemu_madvise(block->colo_cache, block->used_length, | |
3546 | QEMU_MADV_DONTDUMP); | |
3547 | } | |
13af18f2 | 3548 | } |
13af18f2 | 3549 | } |
44901b5a | 3550 | |
7d9acafa ZC |
3551 | /* |
3552 | * Record the dirty pages that sent by PVM, we use this dirty bitmap together | |
3553 | * with to decide which page in cache should be flushed into SVM's RAM. Here | |
3554 | * we use the same name 'ram_bitmap' as for migration. | |
3555 | */ | |
3556 | if (ram_bytes_total()) { | |
fbd162e6 | 3557 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
7d9acafa | 3558 | unsigned long pages = block->max_length >> TARGET_PAGE_BITS; |
7d9acafa | 3559 | block->bmap = bitmap_new(pages); |
7d9acafa ZC |
3560 | } |
3561 | } | |
7d9acafa | 3562 | |
b70cb3b4 | 3563 | colo_init_ram_state(); |
13af18f2 | 3564 | return 0; |
13af18f2 ZC |
3565 | } |
3566 | ||
0393031a HZ |
3567 | /* TODO: duplicated with ram_init_bitmaps */ |
3568 | void colo_incoming_start_dirty_log(void) | |
3569 | { | |
3570 | RAMBlock *block = NULL; | |
639ec3fb CLG |
3571 | Error *local_err = NULL; |
3572 | ||
0393031a | 3573 | /* For memory_global_dirty_log_start below. */ |
195801d7 | 3574 | bql_lock(); |
0393031a HZ |
3575 | qemu_mutex_lock_ramlist(); |
3576 | ||
1e493be5 | 3577 | memory_global_dirty_log_sync(false); |
0393031a HZ |
3578 | WITH_RCU_READ_LOCK_GUARD() { |
3579 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3580 | ramblock_sync_dirty_bitmap(ram_state, block); | |
3581 | /* Discard this dirty bitmap record */ | |
3582 | bitmap_zero(block->bmap, block->max_length >> TARGET_PAGE_BITS); | |
3583 | } | |
639ec3fb CLG |
3584 | if (!memory_global_dirty_log_start(GLOBAL_DIRTY_MIGRATION, |
3585 | &local_err)) { | |
3586 | error_report_err(local_err); | |
3587 | } | |
0393031a HZ |
3588 | } |
3589 | ram_state->migration_dirty_pages = 0; | |
3590 | qemu_mutex_unlock_ramlist(); | |
195801d7 | 3591 | bql_unlock(); |
0393031a HZ |
3592 | } |
3593 | ||
13af18f2 ZC |
3594 | /* It is need to hold the global lock to call this helper */ |
3595 | void colo_release_ram_cache(void) | |
3596 | { | |
3597 | RAMBlock *block; | |
3598 | ||
63b41db4 | 3599 | memory_global_dirty_log_stop(GLOBAL_DIRTY_MIGRATION); |
fbd162e6 | 3600 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
7d9acafa ZC |
3601 | g_free(block->bmap); |
3602 | block->bmap = NULL; | |
3603 | } | |
3604 | ||
89ac5a1d DDAG |
3605 | WITH_RCU_READ_LOCK_GUARD() { |
3606 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3607 | if (block->colo_cache) { | |
3608 | qemu_anon_ram_free(block->colo_cache, block->used_length); | |
3609 | block->colo_cache = NULL; | |
3610 | } | |
13af18f2 ZC |
3611 | } |
3612 | } | |
0393031a | 3613 | ram_state_cleanup(&ram_state); |
13af18f2 ZC |
3614 | } |
3615 | ||
f265e0e4 JQ |
3616 | /** |
3617 | * ram_load_setup: Setup RAM for migration incoming side | |
3618 | * | |
3619 | * Returns zero to indicate success and negative for error | |
3620 | * | |
3621 | * @f: QEMUFile where to receive the data | |
3622 | * @opaque: RAMState pointer | |
e4fa064d | 3623 | * @errp: pointer to Error*, to store an error if it happens. |
f265e0e4 | 3624 | */ |
e4fa064d | 3625 | static int ram_load_setup(QEMUFile *f, void *opaque, Error **errp) |
f265e0e4 JQ |
3626 | { |
3627 | xbzrle_load_setup(); | |
f9494614 | 3628 | ramblock_recv_map_init(); |
13af18f2 | 3629 | |
f265e0e4 JQ |
3630 | return 0; |
3631 | } | |
3632 | ||
3633 | static int ram_load_cleanup(void *opaque) | |
3634 | { | |
f9494614 | 3635 | RAMBlock *rb; |
56eb90af | 3636 | |
fbd162e6 | 3637 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
bd108a44 | 3638 | qemu_ram_block_writeback(rb); |
56eb90af JH |
3639 | } |
3640 | ||
f265e0e4 | 3641 | xbzrle_load_cleanup(); |
f9494614 | 3642 | |
fbd162e6 | 3643 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
f9494614 AP |
3644 | g_free(rb->receivedmap); |
3645 | rb->receivedmap = NULL; | |
3646 | } | |
13af18f2 | 3647 | |
f265e0e4 JQ |
3648 | return 0; |
3649 | } | |
3650 | ||
3d0684b2 JQ |
3651 | /** |
3652 | * ram_postcopy_incoming_init: allocate postcopy data structures | |
3653 | * | |
3654 | * Returns 0 for success and negative if there was one error | |
3655 | * | |
3656 | * @mis: current migration incoming state | |
3657 | * | |
3658 | * Allocate data structures etc needed by incoming migration with | |
3659 | * postcopy-ram. postcopy-ram's similarly names | |
3660 | * postcopy_ram_incoming_init does the work. | |
1caddf8a DDAG |
3661 | */ |
3662 | int ram_postcopy_incoming_init(MigrationIncomingState *mis) | |
3663 | { | |
c136180c | 3664 | return postcopy_ram_incoming_init(mis); |
1caddf8a DDAG |
3665 | } |
3666 | ||
3d0684b2 JQ |
3667 | /** |
3668 | * ram_load_postcopy: load a page in postcopy case | |
3669 | * | |
3670 | * Returns 0 for success or -errno in case of error | |
3671 | * | |
a7180877 DDAG |
3672 | * Called in postcopy mode by ram_load(). |
3673 | * rcu_read_lock is taken prior to this being called. | |
3d0684b2 JQ |
3674 | * |
3675 | * @f: QEMUFile where to send the data | |
36f62f11 | 3676 | * @channel: the channel to use for loading |
a7180877 | 3677 | */ |
36f62f11 | 3678 | int ram_load_postcopy(QEMUFile *f, int channel) |
a7180877 DDAG |
3679 | { |
3680 | int flags = 0, ret = 0; | |
3681 | bool place_needed = false; | |
1aa83678 | 3682 | bool matches_target_page_size = false; |
a7180877 | 3683 | MigrationIncomingState *mis = migration_incoming_get_current(); |
36f62f11 | 3684 | PostcopyTmpPage *tmp_page = &mis->postcopy_tmp_pages[channel]; |
a7180877 DDAG |
3685 | |
3686 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { | |
3687 | ram_addr_t addr; | |
a7180877 DDAG |
3688 | void *page_buffer = NULL; |
3689 | void *place_source = NULL; | |
df9ff5e1 | 3690 | RAMBlock *block = NULL; |
a7180877 | 3691 | uint8_t ch; |
a7180877 DDAG |
3692 | |
3693 | addr = qemu_get_be64(f); | |
7a9ddfbf PX |
3694 | |
3695 | /* | |
3696 | * If qemu file error, we should stop here, and then "addr" | |
3697 | * may be invalid | |
3698 | */ | |
3699 | ret = qemu_file_get_error(f); | |
3700 | if (ret) { | |
3701 | break; | |
3702 | } | |
3703 | ||
a7180877 DDAG |
3704 | flags = addr & ~TARGET_PAGE_MASK; |
3705 | addr &= TARGET_PAGE_MASK; | |
3706 | ||
36f62f11 | 3707 | trace_ram_load_postcopy_loop(channel, (uint64_t)addr, flags); |
0222111a | 3708 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE)) { |
c01b16ed | 3709 | block = ram_block_from_stream(mis, f, flags, channel); |
6a23f639 DH |
3710 | if (!block) { |
3711 | ret = -EINVAL; | |
3712 | break; | |
3713 | } | |
4c4bad48 | 3714 | |
898ba906 DH |
3715 | /* |
3716 | * Relying on used_length is racy and can result in false positives. | |
3717 | * We might place pages beyond used_length in case RAM was shrunk | |
3718 | * while in postcopy, which is fine - trying to place via | |
3719 | * UFFDIO_COPY/UFFDIO_ZEROPAGE will never segfault. | |
3720 | */ | |
3721 | if (!block->host || addr >= block->postcopy_length) { | |
a7180877 DDAG |
3722 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); |
3723 | ret = -EINVAL; | |
3724 | break; | |
3725 | } | |
77dadc3f | 3726 | tmp_page->target_pages++; |
1aa83678 | 3727 | matches_target_page_size = block->page_size == TARGET_PAGE_SIZE; |
a7180877 | 3728 | /* |
28abd200 DDAG |
3729 | * Postcopy requires that we place whole host pages atomically; |
3730 | * these may be huge pages for RAMBlocks that are backed by | |
3731 | * hugetlbfs. | |
a7180877 DDAG |
3732 | * To make it atomic, the data is read into a temporary page |
3733 | * that's moved into place later. | |
3734 | * The migration protocol uses, possibly smaller, target-pages | |
3735 | * however the source ensures it always sends all the components | |
91ba442f | 3736 | * of a host page in one chunk. |
a7180877 | 3737 | */ |
77dadc3f | 3738 | page_buffer = tmp_page->tmp_huge_page + |
6a23f639 DH |
3739 | host_page_offset_from_ram_block_offset(block, addr); |
3740 | /* If all TP are zero then we can optimise the place */ | |
77dadc3f PX |
3741 | if (tmp_page->target_pages == 1) { |
3742 | tmp_page->host_addr = | |
3743 | host_page_from_ram_block_offset(block, addr); | |
3744 | } else if (tmp_page->host_addr != | |
3745 | host_page_from_ram_block_offset(block, addr)) { | |
c53b7ddc | 3746 | /* not the 1st TP within the HP */ |
36f62f11 | 3747 | error_report("Non-same host page detected on channel %d: " |
cfc7dc8a PX |
3748 | "Target host page %p, received host page %p " |
3749 | "(rb %s offset 0x"RAM_ADDR_FMT" target_pages %d)", | |
36f62f11 | 3750 | channel, tmp_page->host_addr, |
cfc7dc8a PX |
3751 | host_page_from_ram_block_offset(block, addr), |
3752 | block->idstr, addr, tmp_page->target_pages); | |
6a23f639 DH |
3753 | ret = -EINVAL; |
3754 | break; | |
a7180877 DDAG |
3755 | } |
3756 | ||
3757 | /* | |
3758 | * If it's the last part of a host page then we place the host | |
3759 | * page | |
3760 | */ | |
77dadc3f PX |
3761 | if (tmp_page->target_pages == |
3762 | (block->page_size / TARGET_PAGE_SIZE)) { | |
4cbb3c63 | 3763 | place_needed = true; |
4cbb3c63 | 3764 | } |
77dadc3f | 3765 | place_source = tmp_page->tmp_huge_page; |
a7180877 DDAG |
3766 | } |
3767 | ||
3768 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { | |
bb890ed5 | 3769 | case RAM_SAVE_FLAG_ZERO: |
a7180877 | 3770 | ch = qemu_get_byte(f); |
413d64fe JQ |
3771 | if (ch != 0) { |
3772 | error_report("Found a zero page with value %d", ch); | |
3773 | ret = -EINVAL; | |
3774 | break; | |
3775 | } | |
2e36bc1b WY |
3776 | /* |
3777 | * Can skip to set page_buffer when | |
3778 | * this is a zero page and (block->page_size == TARGET_PAGE_SIZE). | |
3779 | */ | |
413d64fe | 3780 | if (!matches_target_page_size) { |
2e36bc1b WY |
3781 | memset(page_buffer, ch, TARGET_PAGE_SIZE); |
3782 | } | |
a7180877 DDAG |
3783 | break; |
3784 | ||
3785 | case RAM_SAVE_FLAG_PAGE: | |
77dadc3f | 3786 | tmp_page->all_zero = false; |
1aa83678 PX |
3787 | if (!matches_target_page_size) { |
3788 | /* For huge pages, we always use temporary buffer */ | |
a7180877 DDAG |
3789 | qemu_get_buffer(f, page_buffer, TARGET_PAGE_SIZE); |
3790 | } else { | |
1aa83678 PX |
3791 | /* |
3792 | * For small pages that matches target page size, we | |
3793 | * avoid the qemu_file copy. Instead we directly use | |
3794 | * the buffer of QEMUFile to place the page. Note: we | |
3795 | * cannot do any QEMUFile operation before using that | |
3796 | * buffer to make sure the buffer is valid when | |
3797 | * placing the page. | |
a7180877 DDAG |
3798 | */ |
3799 | qemu_get_buffer_in_place(f, (uint8_t **)&place_source, | |
3800 | TARGET_PAGE_SIZE); | |
3801 | } | |
3802 | break; | |
294e5a40 JQ |
3803 | case RAM_SAVE_FLAG_MULTIFD_FLUSH: |
3804 | multifd_recv_sync_main(); | |
3805 | break; | |
a7180877 DDAG |
3806 | case RAM_SAVE_FLAG_EOS: |
3807 | /* normal exit */ | |
d4f34485 JQ |
3808 | if (migrate_multifd() && |
3809 | migrate_multifd_flush_after_each_section()) { | |
b05292c2 JQ |
3810 | multifd_recv_sync_main(); |
3811 | } | |
a7180877 DDAG |
3812 | break; |
3813 | default: | |
29fccade | 3814 | error_report("Unknown combination of migration flags: 0x%x" |
a7180877 DDAG |
3815 | " (postcopy mode)", flags); |
3816 | ret = -EINVAL; | |
7a9ddfbf PX |
3817 | break; |
3818 | } | |
3819 | ||
3820 | /* Detect for any possible file errors */ | |
3821 | if (!ret && qemu_file_get_error(f)) { | |
3822 | ret = qemu_file_get_error(f); | |
a7180877 DDAG |
3823 | } |
3824 | ||
7a9ddfbf | 3825 | if (!ret && place_needed) { |
77dadc3f PX |
3826 | if (tmp_page->all_zero) { |
3827 | ret = postcopy_place_page_zero(mis, tmp_page->host_addr, block); | |
a7180877 | 3828 | } else { |
77dadc3f PX |
3829 | ret = postcopy_place_page(mis, tmp_page->host_addr, |
3830 | place_source, block); | |
a7180877 | 3831 | } |
ddf35bdf | 3832 | place_needed = false; |
77dadc3f | 3833 | postcopy_temp_page_reset(tmp_page); |
a7180877 | 3834 | } |
a7180877 DDAG |
3835 | } |
3836 | ||
3837 | return ret; | |
3838 | } | |
3839 | ||
acab30b8 DHB |
3840 | static bool postcopy_is_running(void) |
3841 | { | |
3842 | PostcopyState ps = postcopy_state_get(); | |
3843 | return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; | |
3844 | } | |
3845 | ||
e6f4aa18 ZC |
3846 | /* |
3847 | * Flush content of RAM cache into SVM's memory. | |
3848 | * Only flush the pages that be dirtied by PVM or SVM or both. | |
3849 | */ | |
24fa16f8 | 3850 | void colo_flush_ram_cache(void) |
e6f4aa18 ZC |
3851 | { |
3852 | RAMBlock *block = NULL; | |
3853 | void *dst_host; | |
3854 | void *src_host; | |
3855 | unsigned long offset = 0; | |
3856 | ||
1e493be5 | 3857 | memory_global_dirty_log_sync(false); |
9d638407 | 3858 | qemu_mutex_lock(&ram_state->bitmap_mutex); |
89ac5a1d DDAG |
3859 | WITH_RCU_READ_LOCK_GUARD() { |
3860 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { | |
3861 | ramblock_sync_dirty_bitmap(ram_state, block); | |
3862 | } | |
d1955d22 | 3863 | } |
d1955d22 | 3864 | |
e6f4aa18 | 3865 | trace_colo_flush_ram_cache_begin(ram_state->migration_dirty_pages); |
89ac5a1d DDAG |
3866 | WITH_RCU_READ_LOCK_GUARD() { |
3867 | block = QLIST_FIRST_RCU(&ram_list.blocks); | |
e6f4aa18 | 3868 | |
89ac5a1d | 3869 | while (block) { |
a6a83cef | 3870 | unsigned long num = 0; |
e6f4aa18 | 3871 | |
a6a83cef | 3872 | offset = colo_bitmap_find_dirty(ram_state, block, offset, &num); |
542147f4 DH |
3873 | if (!offset_in_ramblock(block, |
3874 | ((ram_addr_t)offset) << TARGET_PAGE_BITS)) { | |
89ac5a1d | 3875 | offset = 0; |
a6a83cef | 3876 | num = 0; |
89ac5a1d DDAG |
3877 | block = QLIST_NEXT_RCU(block, next); |
3878 | } else { | |
a6a83cef RL |
3879 | unsigned long i = 0; |
3880 | ||
3881 | for (i = 0; i < num; i++) { | |
3882 | migration_bitmap_clear_dirty(ram_state, block, offset + i); | |
3883 | } | |
8bba004c AR |
3884 | dst_host = block->host |
3885 | + (((ram_addr_t)offset) << TARGET_PAGE_BITS); | |
3886 | src_host = block->colo_cache | |
3887 | + (((ram_addr_t)offset) << TARGET_PAGE_BITS); | |
a6a83cef RL |
3888 | memcpy(dst_host, src_host, TARGET_PAGE_SIZE * num); |
3889 | offset += num; | |
89ac5a1d | 3890 | } |
e6f4aa18 ZC |
3891 | } |
3892 | } | |
9d638407 | 3893 | qemu_mutex_unlock(&ram_state->bitmap_mutex); |
e6f4aa18 ZC |
3894 | trace_colo_flush_ram_cache_end(); |
3895 | } | |
3896 | ||
a49d15a3 FR |
3897 | static size_t ram_load_multifd_pages(void *host_addr, size_t size, |
3898 | uint64_t offset) | |
3899 | { | |
3900 | MultiFDRecvData *data = multifd_get_recv_data(); | |
3901 | ||
3902 | data->opaque = host_addr; | |
3903 | data->file_offset = offset; | |
3904 | data->size = size; | |
3905 | ||
3906 | if (!multifd_recv()) { | |
3907 | return 0; | |
3908 | } | |
3909 | ||
3910 | return size; | |
3911 | } | |
3912 | ||
2f6b8826 FR |
3913 | static bool read_ramblock_mapped_ram(QEMUFile *f, RAMBlock *block, |
3914 | long num_pages, unsigned long *bitmap, | |
3915 | Error **errp) | |
3916 | { | |
3917 | ERRP_GUARD(); | |
3918 | unsigned long set_bit_idx, clear_bit_idx; | |
3919 | ram_addr_t offset; | |
3920 | void *host; | |
3921 | size_t read, unread, size; | |
3922 | ||
3923 | for (set_bit_idx = find_first_bit(bitmap, num_pages); | |
3924 | set_bit_idx < num_pages; | |
3925 | set_bit_idx = find_next_bit(bitmap, num_pages, clear_bit_idx + 1)) { | |
3926 | ||
3927 | clear_bit_idx = find_next_zero_bit(bitmap, num_pages, set_bit_idx + 1); | |
3928 | ||
3929 | unread = TARGET_PAGE_SIZE * (clear_bit_idx - set_bit_idx); | |
3930 | offset = set_bit_idx << TARGET_PAGE_BITS; | |
3931 | ||
3932 | while (unread > 0) { | |
3933 | host = host_from_ram_block_offset(block, offset); | |
3934 | if (!host) { | |
3935 | error_setg(errp, "page outside of ramblock %s range", | |
3936 | block->idstr); | |
3937 | return false; | |
3938 | } | |
3939 | ||
3940 | size = MIN(unread, MAPPED_RAM_LOAD_BUF_SIZE); | |
3941 | ||
a49d15a3 FR |
3942 | if (migrate_multifd()) { |
3943 | read = ram_load_multifd_pages(host, size, | |
3944 | block->pages_offset + offset); | |
3945 | } else { | |
3946 | read = qemu_get_buffer_at(f, host, size, | |
3947 | block->pages_offset + offset); | |
3948 | } | |
3949 | ||
2f6b8826 FR |
3950 | if (!read) { |
3951 | goto err; | |
3952 | } | |
3953 | offset += read; | |
3954 | unread -= read; | |
3955 | } | |
3956 | } | |
3957 | ||
3958 | return true; | |
3959 | ||
3960 | err: | |
3961 | qemu_file_get_error_obj(f, errp); | |
3962 | error_prepend(errp, "(%s) failed to read page " RAM_ADDR_FMT | |
3963 | "from file offset %" PRIx64 ": ", block->idstr, offset, | |
3964 | block->pages_offset + offset); | |
3965 | return false; | |
3966 | } | |
3967 | ||
3968 | static void parse_ramblock_mapped_ram(QEMUFile *f, RAMBlock *block, | |
3969 | ram_addr_t length, Error **errp) | |
3970 | { | |
3971 | g_autofree unsigned long *bitmap = NULL; | |
3972 | MappedRamHeader header; | |
3973 | size_t bitmap_size; | |
3974 | long num_pages; | |
3975 | ||
3976 | if (!mapped_ram_read_header(f, &header, errp)) { | |
3977 | return; | |
3978 | } | |
3979 | ||
3980 | block->pages_offset = header.pages_offset; | |
3981 | ||
3982 | /* | |
3983 | * Check the alignment of the file region that contains pages. We | |
3984 | * don't enforce MAPPED_RAM_FILE_OFFSET_ALIGNMENT to allow that | |
3985 | * value to change in the future. Do only a sanity check with page | |
3986 | * size alignment. | |
3987 | */ | |
3988 | if (!QEMU_IS_ALIGNED(block->pages_offset, TARGET_PAGE_SIZE)) { | |
3989 | error_setg(errp, | |
3990 | "Error reading ramblock %s pages, region has bad alignment", | |
3991 | block->idstr); | |
3992 | return; | |
3993 | } | |
3994 | ||
3995 | num_pages = length / header.page_size; | |
3996 | bitmap_size = BITS_TO_LONGS(num_pages) * sizeof(unsigned long); | |
3997 | ||
3998 | bitmap = g_malloc0(bitmap_size); | |
3999 | if (qemu_get_buffer_at(f, (uint8_t *)bitmap, bitmap_size, | |
4000 | header.bitmap_offset) != bitmap_size) { | |
4001 | error_setg(errp, "Error reading dirty bitmap"); | |
4002 | return; | |
4003 | } | |
4004 | ||
4005 | if (!read_ramblock_mapped_ram(f, block, num_pages, bitmap, errp)) { | |
4006 | return; | |
4007 | } | |
4008 | ||
4009 | /* Skip pages array */ | |
4010 | qemu_set_offset(f, block->pages_offset + length, SEEK_SET); | |
4011 | ||
4012 | return; | |
4013 | } | |
4014 | ||
2f5ced5b NB |
4015 | static int parse_ramblock(QEMUFile *f, RAMBlock *block, ram_addr_t length) |
4016 | { | |
4017 | int ret = 0; | |
4018 | /* ADVISE is earlier, it shows the source has the postcopy capability on */ | |
4019 | bool postcopy_advised = migration_incoming_postcopy_advised(); | |
5d220369 | 4020 | int max_hg_page_size; |
2f6b8826 | 4021 | Error *local_err = NULL; |
2f5ced5b NB |
4022 | |
4023 | assert(block); | |
4024 | ||
2f6b8826 FR |
4025 | if (migrate_mapped_ram()) { |
4026 | parse_ramblock_mapped_ram(f, block, length, &local_err); | |
4027 | if (local_err) { | |
4028 | error_report_err(local_err); | |
4029 | return -EINVAL; | |
4030 | } | |
4031 | return 0; | |
4032 | } | |
4033 | ||
2f5ced5b NB |
4034 | if (!qemu_ram_is_migratable(block)) { |
4035 | error_report("block %s should not be migrated !", block->idstr); | |
4036 | return -EINVAL; | |
4037 | } | |
4038 | ||
4039 | if (length != block->used_length) { | |
2f5ced5b NB |
4040 | ret = qemu_ram_resize(block, length, &local_err); |
4041 | if (local_err) { | |
4042 | error_report_err(local_err); | |
2c36076a | 4043 | return ret; |
2f5ced5b NB |
4044 | } |
4045 | } | |
5d220369 RH |
4046 | |
4047 | /* | |
4048 | * ??? Mirrors the previous value of qemu_host_page_size, | |
4049 | * but is this really what was intended for the migration? | |
4050 | */ | |
4051 | max_hg_page_size = MAX(qemu_real_host_page_size(), TARGET_PAGE_SIZE); | |
4052 | ||
2f5ced5b NB |
4053 | /* For postcopy we need to check hugepage sizes match */ |
4054 | if (postcopy_advised && migrate_postcopy_ram() && | |
5d220369 | 4055 | block->page_size != max_hg_page_size) { |
2f5ced5b NB |
4056 | uint64_t remote_page_size = qemu_get_be64(f); |
4057 | if (remote_page_size != block->page_size) { | |
4058 | error_report("Mismatched RAM page size %s " | |
4059 | "(local) %zd != %" PRId64, block->idstr, | |
4060 | block->page_size, remote_page_size); | |
2c36076a | 4061 | return -EINVAL; |
2f5ced5b NB |
4062 | } |
4063 | } | |
4064 | if (migrate_ignore_shared()) { | |
4065 | hwaddr addr = qemu_get_be64(f); | |
4066 | if (migrate_ram_is_ignored(block) && | |
4067 | block->mr->addr != addr) { | |
4068 | error_report("Mismatched GPAs for block %s " | |
4069 | "%" PRId64 "!= %" PRId64, block->idstr, | |
4070 | (uint64_t)addr, (uint64_t)block->mr->addr); | |
2c36076a | 4071 | return -EINVAL; |
2f5ced5b NB |
4072 | } |
4073 | } | |
4074 | ret = rdma_block_notification_handle(f, block->idstr); | |
4075 | if (ret < 0) { | |
4076 | qemu_file_set_error(f, ret); | |
4077 | } | |
4078 | ||
4079 | return ret; | |
4080 | } | |
4081 | ||
4082 | static int parse_ramblocks(QEMUFile *f, ram_addr_t total_ram_bytes) | |
4083 | { | |
4084 | int ret = 0; | |
4085 | ||
4086 | /* Synchronize RAM block list */ | |
4087 | while (!ret && total_ram_bytes) { | |
4088 | RAMBlock *block; | |
4089 | char id[256]; | |
4090 | ram_addr_t length; | |
4091 | int len = qemu_get_byte(f); | |
4092 | ||
4093 | qemu_get_buffer(f, (uint8_t *)id, len); | |
4094 | id[len] = 0; | |
4095 | length = qemu_get_be64(f); | |
4096 | ||
4097 | block = qemu_ram_block_by_name(id); | |
4098 | if (block) { | |
4099 | ret = parse_ramblock(f, block, length); | |
4100 | } else { | |
4101 | error_report("Unknown ramblock \"%s\", cannot accept " | |
4102 | "migration", id); | |
4103 | ret = -EINVAL; | |
4104 | } | |
4105 | total_ram_bytes -= length; | |
4106 | } | |
4107 | ||
4108 | return ret; | |
4109 | } | |
4110 | ||
10da4a36 WY |
4111 | /** |
4112 | * ram_load_precopy: load pages in precopy case | |
4113 | * | |
4114 | * Returns 0 for success or -errno in case of error | |
4115 | * | |
4116 | * Called in precopy mode by ram_load(). | |
4117 | * rcu_read_lock is taken prior to this being called. | |
4118 | * | |
4119 | * @f: QEMUFile where to send the data | |
4120 | */ | |
4121 | static int ram_load_precopy(QEMUFile *f) | |
56e93d26 | 4122 | { |
755e8d7c | 4123 | MigrationIncomingState *mis = migration_incoming_get_current(); |
0222111a | 4124 | int flags = 0, ret = 0, invalid_flags = 0, i = 0; |
a7180877 | 4125 | |
9d01778a FR |
4126 | if (migrate_mapped_ram()) { |
4127 | invalid_flags |= (RAM_SAVE_FLAG_HOOK | RAM_SAVE_FLAG_MULTIFD_FLUSH | | |
4128 | RAM_SAVE_FLAG_PAGE | RAM_SAVE_FLAG_XBZRLE | | |
4129 | RAM_SAVE_FLAG_ZERO); | |
4130 | } | |
4131 | ||
10da4a36 | 4132 | while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { |
2f5ced5b | 4133 | ram_addr_t addr; |
0393031a | 4134 | void *host = NULL, *host_bak = NULL; |
56e93d26 JQ |
4135 | uint8_t ch; |
4136 | ||
e65cec5e YK |
4137 | /* |
4138 | * Yield periodically to let main loop run, but an iteration of | |
4139 | * the main loop is expensive, so do it each some iterations | |
4140 | */ | |
4141 | if ((i & 32767) == 0 && qemu_in_coroutine()) { | |
4142 | aio_co_schedule(qemu_get_current_aio_context(), | |
4143 | qemu_coroutine_self()); | |
4144 | qemu_coroutine_yield(); | |
4145 | } | |
4146 | i++; | |
4147 | ||
56e93d26 | 4148 | addr = qemu_get_be64(f); |
12ab1e4f MD |
4149 | ret = qemu_file_get_error(f); |
4150 | if (ret) { | |
4151 | error_report("Getting RAM address failed"); | |
4152 | break; | |
4153 | } | |
4154 | ||
56e93d26 JQ |
4155 | flags = addr & ~TARGET_PAGE_MASK; |
4156 | addr &= TARGET_PAGE_MASK; | |
4157 | ||
edc60127 | 4158 | if (flags & invalid_flags) { |
9d01778a FR |
4159 | error_report("Unexpected RAM flags: %d", flags & invalid_flags); |
4160 | ||
edc60127 JQ |
4161 | ret = -EINVAL; |
4162 | break; | |
4163 | } | |
4164 | ||
bb890ed5 | 4165 | if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE | |
0222111a | 4166 | RAM_SAVE_FLAG_XBZRLE)) { |
c01b16ed PX |
4167 | RAMBlock *block = ram_block_from_stream(mis, f, flags, |
4168 | RAM_CHANNEL_PRECOPY); | |
4c4bad48 | 4169 | |
0393031a | 4170 | host = host_from_ram_block_offset(block, addr); |
13af18f2 | 4171 | /* |
0393031a HZ |
4172 | * After going into COLO stage, we should not load the page |
4173 | * into SVM's memory directly, we put them into colo_cache firstly. | |
4174 | * NOTE: We need to keep a copy of SVM's ram in colo_cache. | |
4175 | * Previously, we copied all these memory in preparing stage of COLO | |
4176 | * while we need to stop VM, which is a time-consuming process. | |
4177 | * Here we optimize it by a trick, back-up every page while in | |
4178 | * migration process while COLO is enabled, though it affects the | |
4179 | * speed of the migration, but it obviously reduce the downtime of | |
4180 | * back-up all SVM'S memory in COLO preparing stage. | |
13af18f2 | 4181 | */ |
0393031a HZ |
4182 | if (migration_incoming_colo_enabled()) { |
4183 | if (migration_incoming_in_colo_state()) { | |
4184 | /* In COLO stage, put all pages into cache temporarily */ | |
8af66371 | 4185 | host = colo_cache_from_block_offset(block, addr, true); |
0393031a HZ |
4186 | } else { |
4187 | /* | |
4188 | * In migration stage but before COLO stage, | |
4189 | * Put all pages into both cache and SVM's memory. | |
4190 | */ | |
8af66371 | 4191 | host_bak = colo_cache_from_block_offset(block, addr, false); |
0393031a | 4192 | } |
13af18f2 | 4193 | } |
a776aa15 DDAG |
4194 | if (!host) { |
4195 | error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); | |
4196 | ret = -EINVAL; | |
4197 | break; | |
4198 | } | |
13af18f2 ZC |
4199 | if (!migration_incoming_in_colo_state()) { |
4200 | ramblock_recv_bitmap_set(block, host); | |
4201 | } | |
4202 | ||
1db9d8e5 | 4203 | trace_ram_load_loop(block->idstr, (uint64_t)addr, flags, host); |
a776aa15 DDAG |
4204 | } |
4205 | ||
56e93d26 JQ |
4206 | switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { |
4207 | case RAM_SAVE_FLAG_MEM_SIZE: | |
2f5ced5b | 4208 | ret = parse_ramblocks(f, addr); |
1a6e217c PX |
4209 | /* |
4210 | * For mapped-ram migration (to a file) using multifd, we sync | |
4211 | * once and for all here to make sure all tasks we queued to | |
4212 | * multifd threads are completed, so that all the ramblocks | |
4213 | * (including all the guest memory pages within) are fully | |
4214 | * loaded after this sync returns. | |
4215 | */ | |
9d01778a FR |
4216 | if (migrate_mapped_ram()) { |
4217 | multifd_recv_sync_main(); | |
4218 | } | |
56e93d26 | 4219 | break; |
a776aa15 | 4220 | |
bb890ed5 | 4221 | case RAM_SAVE_FLAG_ZERO: |
56e93d26 | 4222 | ch = qemu_get_byte(f); |
413d64fe JQ |
4223 | if (ch != 0) { |
4224 | error_report("Found a zero page with value %d", ch); | |
4225 | ret = -EINVAL; | |
4226 | break; | |
4227 | } | |
7091dabe | 4228 | ram_handle_zero(host, TARGET_PAGE_SIZE); |
56e93d26 | 4229 | break; |
a776aa15 | 4230 | |
56e93d26 | 4231 | case RAM_SAVE_FLAG_PAGE: |
56e93d26 JQ |
4232 | qemu_get_buffer(f, host, TARGET_PAGE_SIZE); |
4233 | break; | |
56e93d26 | 4234 | |
56e93d26 | 4235 | case RAM_SAVE_FLAG_XBZRLE: |
56e93d26 JQ |
4236 | if (load_xbzrle(f, addr, host) < 0) { |
4237 | error_report("Failed to decompress XBZRLE page at " | |
4238 | RAM_ADDR_FMT, addr); | |
4239 | ret = -EINVAL; | |
4240 | break; | |
4241 | } | |
4242 | break; | |
294e5a40 JQ |
4243 | case RAM_SAVE_FLAG_MULTIFD_FLUSH: |
4244 | multifd_recv_sync_main(); | |
4245 | break; | |
56e93d26 JQ |
4246 | case RAM_SAVE_FLAG_EOS: |
4247 | /* normal exit */ | |
d4f34485 | 4248 | if (migrate_multifd() && |
9d01778a FR |
4249 | migrate_multifd_flush_after_each_section() && |
4250 | /* | |
4251 | * Mapped-ram migration flushes once and for all after | |
4252 | * parsing ramblocks. Always ignore EOS for it. | |
4253 | */ | |
4254 | !migrate_mapped_ram()) { | |
b05292c2 JQ |
4255 | multifd_recv_sync_main(); |
4256 | } | |
56e93d26 | 4257 | break; |
5f1e7540 | 4258 | case RAM_SAVE_FLAG_HOOK: |
b1b38387 | 4259 | ret = rdma_registration_handle(f); |
f6d6c089 JQ |
4260 | if (ret < 0) { |
4261 | qemu_file_set_error(f, ret); | |
4262 | } | |
5f1e7540 | 4263 | break; |
56e93d26 | 4264 | default: |
5f1e7540 JQ |
4265 | error_report("Unknown combination of migration flags: 0x%x", flags); |
4266 | ret = -EINVAL; | |
56e93d26 JQ |
4267 | } |
4268 | if (!ret) { | |
4269 | ret = qemu_file_get_error(f); | |
4270 | } | |
0393031a HZ |
4271 | if (!ret && host_bak) { |
4272 | memcpy(host_bak, host, TARGET_PAGE_SIZE); | |
4273 | } | |
56e93d26 JQ |
4274 | } |
4275 | ||
10da4a36 WY |
4276 | return ret; |
4277 | } | |
4278 | ||
4279 | static int ram_load(QEMUFile *f, void *opaque, int version_id) | |
4280 | { | |
4281 | int ret = 0; | |
4282 | static uint64_t seq_iter; | |
4283 | /* | |
4284 | * If system is running in postcopy mode, page inserts to host memory must | |
4285 | * be atomic | |
4286 | */ | |
4287 | bool postcopy_running = postcopy_is_running(); | |
4288 | ||
4289 | seq_iter++; | |
4290 | ||
4291 | if (version_id != 4) { | |
4292 | return -EINVAL; | |
4293 | } | |
4294 | ||
4295 | /* | |
4296 | * This RCU critical section can be very long running. | |
4297 | * When RCU reclaims in the code start to become numerous, | |
4298 | * it will be necessary to reduce the granularity of this | |
4299 | * critical section. | |
4300 | */ | |
89ac5a1d DDAG |
4301 | WITH_RCU_READ_LOCK_GUARD() { |
4302 | if (postcopy_running) { | |
36f62f11 PX |
4303 | /* |
4304 | * Note! Here RAM_CHANNEL_PRECOPY is the precopy channel of | |
4305 | * postcopy migration, we have another RAM_CHANNEL_POSTCOPY to | |
4306 | * service fast page faults. | |
4307 | */ | |
4308 | ret = ram_load_postcopy(f, RAM_CHANNEL_PRECOPY); | |
89ac5a1d DDAG |
4309 | } else { |
4310 | ret = ram_load_precopy(f); | |
4311 | } | |
10da4a36 | 4312 | } |
55c4446b | 4313 | trace_ram_load_complete(ret, seq_iter); |
e6f4aa18 | 4314 | |
56e93d26 JQ |
4315 | return ret; |
4316 | } | |
4317 | ||
c6467627 VSO |
4318 | static bool ram_has_postcopy(void *opaque) |
4319 | { | |
469dd51b | 4320 | RAMBlock *rb; |
fbd162e6 | 4321 | RAMBLOCK_FOREACH_NOT_IGNORED(rb) { |
469dd51b JH |
4322 | if (ramblock_is_pmem(rb)) { |
4323 | info_report("Block: %s, host: %p is a nvdimm memory, postcopy" | |
4324 | "is not supported now!", rb->idstr, rb->host); | |
4325 | return false; | |
4326 | } | |
4327 | } | |
4328 | ||
c6467627 VSO |
4329 | return migrate_postcopy_ram(); |
4330 | } | |
4331 | ||
edd090c7 PX |
4332 | /* Sync all the dirty bitmap with destination VM. */ |
4333 | static int ram_dirty_bitmap_sync_all(MigrationState *s, RAMState *rs) | |
4334 | { | |
4335 | RAMBlock *block; | |
4336 | QEMUFile *file = s->to_dst_file; | |
edd090c7 PX |
4337 | |
4338 | trace_ram_dirty_bitmap_sync_start(); | |
4339 | ||
1015ff54 | 4340 | qatomic_set(&rs->postcopy_bmap_sync_requested, 0); |
fbd162e6 | 4341 | RAMBLOCK_FOREACH_NOT_IGNORED(block) { |
edd090c7 PX |
4342 | qemu_savevm_send_recv_bitmap(file, block->idstr); |
4343 | trace_ram_dirty_bitmap_request(block->idstr); | |
1015ff54 | 4344 | qatomic_inc(&rs->postcopy_bmap_sync_requested); |
edd090c7 PX |
4345 | } |
4346 | ||
4347 | trace_ram_dirty_bitmap_sync_wait(); | |
4348 | ||
4349 | /* Wait until all the ramblocks' dirty bitmap synced */ | |
1015ff54 | 4350 | while (qatomic_read(&rs->postcopy_bmap_sync_requested)) { |
f8c543e8 PX |
4351 | if (migration_rp_wait(s)) { |
4352 | return -1; | |
4353 | } | |
edd090c7 PX |
4354 | } |
4355 | ||
4356 | trace_ram_dirty_bitmap_sync_complete(); | |
4357 | ||
4358 | return 0; | |
4359 | } | |
4360 | ||
a335debb PX |
4361 | /* |
4362 | * Read the received bitmap, revert it as the initial dirty bitmap. | |
4363 | * This is only used when the postcopy migration is paused but wants | |
4364 | * to resume from a middle point. | |
88577f32 PX |
4365 | * |
4366 | * Returns true if succeeded, false for errors. | |
a335debb | 4367 | */ |
88577f32 | 4368 | bool ram_dirty_bitmap_reload(MigrationState *s, RAMBlock *block, Error **errp) |
a335debb | 4369 | { |
43044ac0 | 4370 | /* from_dst_file is always valid because we're within rp_thread */ |
a335debb | 4371 | QEMUFile *file = s->rp_state.from_dst_file; |
1a36e4c9 PMD |
4372 | g_autofree unsigned long *le_bitmap = NULL; |
4373 | unsigned long nbits = block->used_length >> TARGET_PAGE_BITS; | |
a725ef9f | 4374 | uint64_t local_size = DIV_ROUND_UP(nbits, 8); |
a335debb | 4375 | uint64_t size, end_mark; |
1015ff54 | 4376 | RAMState *rs = ram_state; |
a335debb PX |
4377 | |
4378 | trace_ram_dirty_bitmap_reload_begin(block->idstr); | |
4379 | ||
4380 | if (s->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
7aa6070d PX |
4381 | error_setg(errp, "Reload bitmap in incorrect state %s", |
4382 | MigrationStatus_str(s->state)); | |
88577f32 | 4383 | return false; |
a335debb PX |
4384 | } |
4385 | ||
4386 | /* | |
4387 | * Note: see comments in ramblock_recv_bitmap_send() on why we | |
3a4452d8 | 4388 | * need the endianness conversion, and the paddings. |
a335debb PX |
4389 | */ |
4390 | local_size = ROUND_UP(local_size, 8); | |
4391 | ||
4392 | /* Add paddings */ | |
4393 | le_bitmap = bitmap_new(nbits + BITS_PER_LONG); | |
4394 | ||
4395 | size = qemu_get_be64(file); | |
4396 | ||
4397 | /* The size of the bitmap should match with our ramblock */ | |
4398 | if (size != local_size) { | |
7aa6070d PX |
4399 | error_setg(errp, "ramblock '%s' bitmap size mismatch (0x%"PRIx64 |
4400 | " != 0x%"PRIx64")", block->idstr, size, local_size); | |
88577f32 | 4401 | return false; |
a335debb PX |
4402 | } |
4403 | ||
4404 | size = qemu_get_buffer(file, (uint8_t *)le_bitmap, local_size); | |
4405 | end_mark = qemu_get_be64(file); | |
4406 | ||
88577f32 PX |
4407 | if (qemu_file_get_error(file) || size != local_size) { |
4408 | error_setg(errp, "read bitmap failed for ramblock '%s': " | |
4409 | "(size 0x%"PRIx64", got: 0x%"PRIx64")", | |
4410 | block->idstr, local_size, size); | |
4411 | return false; | |
a335debb PX |
4412 | } |
4413 | ||
4414 | if (end_mark != RAMBLOCK_RECV_BITMAP_ENDING) { | |
7aa6070d PX |
4415 | error_setg(errp, "ramblock '%s' end mark incorrect: 0x%"PRIx64, |
4416 | block->idstr, end_mark); | |
88577f32 | 4417 | return false; |
a335debb PX |
4418 | } |
4419 | ||
4420 | /* | |
3a4452d8 | 4421 | * Endianness conversion. We are during postcopy (though paused). |
a335debb PX |
4422 | * The dirty bitmap won't change. We can directly modify it. |
4423 | */ | |
4424 | bitmap_from_le(block->bmap, le_bitmap, nbits); | |
4425 | ||
4426 | /* | |
4427 | * What we received is "received bitmap". Revert it as the initial | |
4428 | * dirty bitmap for this ramblock. | |
4429 | */ | |
4430 | bitmap_complement(block->bmap, block->bmap, nbits); | |
4431 | ||
be39b4cd DH |
4432 | /* Clear dirty bits of discarded ranges that we don't want to migrate. */ |
4433 | ramblock_dirty_bitmap_clear_discarded_pages(block); | |
4434 | ||
4435 | /* We'll recalculate migration_dirty_pages in ram_state_resume_prepare(). */ | |
a335debb PX |
4436 | trace_ram_dirty_bitmap_reload_complete(block->idstr); |
4437 | ||
1015ff54 PX |
4438 | qatomic_dec(&rs->postcopy_bmap_sync_requested); |
4439 | ||
edd090c7 | 4440 | /* |
5e79a4bf PX |
4441 | * We succeeded to sync bitmap for current ramblock. Always kick the |
4442 | * migration thread to check whether all requested bitmaps are | |
4443 | * reloaded. NOTE: it's racy to only kick when requested==0, because | |
4444 | * we don't know whether the migration thread may still be increasing | |
4445 | * it. | |
edd090c7 | 4446 | */ |
5e79a4bf | 4447 | migration_rp_kick(s); |
edd090c7 | 4448 | |
88577f32 | 4449 | return true; |
a335debb PX |
4450 | } |
4451 | ||
edd090c7 PX |
4452 | static int ram_resume_prepare(MigrationState *s, void *opaque) |
4453 | { | |
4454 | RAMState *rs = *(RAMState **)opaque; | |
08614f34 | 4455 | int ret; |
edd090c7 | 4456 | |
08614f34 PX |
4457 | ret = ram_dirty_bitmap_sync_all(s, rs); |
4458 | if (ret) { | |
4459 | return ret; | |
4460 | } | |
4461 | ||
4462 | ram_state_resume_prepare(rs, s->to_dst_file); | |
4463 | ||
4464 | return 0; | |
edd090c7 PX |
4465 | } |
4466 | ||
36f62f11 PX |
4467 | void postcopy_preempt_shutdown_file(MigrationState *s) |
4468 | { | |
4469 | qemu_put_be64(s->postcopy_qemufile_src, RAM_SAVE_FLAG_EOS); | |
4470 | qemu_fflush(s->postcopy_qemufile_src); | |
4471 | } | |
4472 | ||
56e93d26 | 4473 | static SaveVMHandlers savevm_ram_handlers = { |
9907e842 | 4474 | .save_setup = ram_save_setup, |
56e93d26 | 4475 | .save_live_iterate = ram_save_iterate, |
763c906b | 4476 | .save_live_complete_postcopy = ram_save_complete, |
a3e06c3d | 4477 | .save_live_complete_precopy = ram_save_complete, |
c6467627 | 4478 | .has_postcopy = ram_has_postcopy, |
c8df4a7a JQ |
4479 | .state_pending_exact = ram_state_pending_exact, |
4480 | .state_pending_estimate = ram_state_pending_estimate, | |
56e93d26 | 4481 | .load_state = ram_load, |
f265e0e4 JQ |
4482 | .save_cleanup = ram_save_cleanup, |
4483 | .load_setup = ram_load_setup, | |
4484 | .load_cleanup = ram_load_cleanup, | |
edd090c7 | 4485 | .resume_prepare = ram_resume_prepare, |
56e93d26 JQ |
4486 | }; |
4487 | ||
c7c0e724 DH |
4488 | static void ram_mig_ram_block_resized(RAMBlockNotifier *n, void *host, |
4489 | size_t old_size, size_t new_size) | |
4490 | { | |
cc61c703 | 4491 | PostcopyState ps = postcopy_state_get(); |
c7c0e724 DH |
4492 | ram_addr_t offset; |
4493 | RAMBlock *rb = qemu_ram_block_from_host(host, false, &offset); | |
4494 | Error *err = NULL; | |
4495 | ||
f75ed59f DF |
4496 | if (!rb) { |
4497 | error_report("RAM block not found"); | |
4498 | return; | |
4499 | } | |
4500 | ||
f161c88a | 4501 | if (migrate_ram_is_ignored(rb)) { |
c7c0e724 DH |
4502 | return; |
4503 | } | |
4504 | ||
4505 | if (!migration_is_idle()) { | |
4506 | /* | |
4507 | * Precopy code on the source cannot deal with the size of RAM blocks | |
4508 | * changing at random points in time - especially after sending the | |
4509 | * RAM block sizes in the migration stream, they must no longer change. | |
4510 | * Abort and indicate a proper reason. | |
4511 | */ | |
4512 | error_setg(&err, "RAM block '%s' resized during precopy.", rb->idstr); | |
458fecca | 4513 | migration_cancel(err); |
c7c0e724 | 4514 | error_free(err); |
c7c0e724 | 4515 | } |
cc61c703 DH |
4516 | |
4517 | switch (ps) { | |
4518 | case POSTCOPY_INCOMING_ADVISE: | |
4519 | /* | |
4520 | * Update what ram_postcopy_incoming_init()->init_range() does at the | |
4521 | * time postcopy was advised. Syncing RAM blocks with the source will | |
4522 | * result in RAM resizes. | |
4523 | */ | |
4524 | if (old_size < new_size) { | |
4525 | if (ram_discard_range(rb->idstr, old_size, new_size - old_size)) { | |
4526 | error_report("RAM block '%s' discard of resized RAM failed", | |
4527 | rb->idstr); | |
4528 | } | |
4529 | } | |
898ba906 | 4530 | rb->postcopy_length = new_size; |
cc61c703 DH |
4531 | break; |
4532 | case POSTCOPY_INCOMING_NONE: | |
4533 | case POSTCOPY_INCOMING_RUNNING: | |
4534 | case POSTCOPY_INCOMING_END: | |
4535 | /* | |
4536 | * Once our guest is running, postcopy does no longer care about | |
4537 | * resizes. When growing, the new memory was not available on the | |
4538 | * source, no handler needed. | |
4539 | */ | |
4540 | break; | |
4541 | default: | |
4542 | error_report("RAM block '%s' resized during postcopy state: %d", | |
4543 | rb->idstr, ps); | |
4544 | exit(-1); | |
4545 | } | |
c7c0e724 DH |
4546 | } |
4547 | ||
4548 | static RAMBlockNotifier ram_mig_ram_notifier = { | |
4549 | .ram_block_resized = ram_mig_ram_block_resized, | |
4550 | }; | |
4551 | ||
56e93d26 JQ |
4552 | void ram_mig_init(void) |
4553 | { | |
4554 | qemu_mutex_init(&XBZRLE.lock); | |
ce62df53 | 4555 | register_savevm_live("ram", 0, 4, &savevm_ram_handlers, &ram_state); |
c7c0e724 | 4556 | ram_block_notifier_add(&ram_mig_ram_notifier); |
56e93d26 | 4557 | } |