]>
Commit | Line | Data |
---|---|---|
eb59db53 DDAG |
1 | /* |
2 | * Postcopy migration for RAM | |
3 | * | |
4 | * Copyright 2013-2015 Red Hat, Inc. and/or its affiliates | |
5 | * | |
6 | * Authors: | |
7 | * Dave Gilbert <dgilbert@redhat.com> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU GPL, version 2 or later. | |
10 | * See the COPYING file in the top-level directory. | |
11 | * | |
12 | */ | |
13 | ||
14 | /* | |
15 | * Postcopy is a migration technique where the execution flips from the | |
16 | * source to the destination before all the data has been copied. | |
17 | */ | |
18 | ||
1393a485 | 19 | #include "qemu/osdep.h" |
eb59db53 DDAG |
20 | |
21 | #include "qemu-common.h" | |
22 | #include "migration/migration.h" | |
23 | #include "migration/postcopy-ram.h" | |
24 | #include "sysemu/sysemu.h" | |
371ff5a3 | 25 | #include "sysemu/balloon.h" |
eb59db53 DDAG |
26 | #include "qemu/error-report.h" |
27 | #include "trace.h" | |
28 | ||
e0b266f0 DDAG |
29 | /* Arbitrary limit on size of each discard command, |
30 | * keeps them around ~200 bytes | |
31 | */ | |
32 | #define MAX_DISCARDS_PER_COMMAND 12 | |
33 | ||
34 | struct PostcopyDiscardState { | |
35 | const char *ramblock_name; | |
36 | uint64_t offset; /* Bitmap entry for the 1st bit of this RAMBlock */ | |
37 | uint16_t cur_entry; | |
38 | /* | |
39 | * Start and length of a discard range (bytes) | |
40 | */ | |
41 | uint64_t start_list[MAX_DISCARDS_PER_COMMAND]; | |
42 | uint64_t length_list[MAX_DISCARDS_PER_COMMAND]; | |
43 | unsigned int nsentwords; | |
44 | unsigned int nsentcmds; | |
45 | }; | |
46 | ||
eb59db53 DDAG |
47 | /* Postcopy needs to detect accesses to pages that haven't yet been copied |
48 | * across, and efficiently map new pages in, the techniques for doing this | |
49 | * are target OS specific. | |
50 | */ | |
51 | #if defined(__linux__) | |
52 | ||
c4faeed2 | 53 | #include <poll.h> |
eb59db53 DDAG |
54 | #include <sys/mman.h> |
55 | #include <sys/ioctl.h> | |
56 | #include <sys/syscall.h> | |
eb59db53 DDAG |
57 | #include <asm/types.h> /* for __u64 */ |
58 | #endif | |
59 | ||
d8b9d771 MF |
60 | #if defined(__linux__) && defined(__NR_userfaultfd) && defined(CONFIG_EVENTFD) |
61 | #include <sys/eventfd.h> | |
eb59db53 DDAG |
62 | #include <linux/userfaultfd.h> |
63 | ||
64 | static bool ufd_version_check(int ufd) | |
65 | { | |
66 | struct uffdio_api api_struct; | |
67 | uint64_t ioctl_mask; | |
68 | ||
69 | api_struct.api = UFFD_API; | |
70 | api_struct.features = 0; | |
71 | if (ioctl(ufd, UFFDIO_API, &api_struct)) { | |
72 | error_report("postcopy_ram_supported_by_host: UFFDIO_API failed: %s", | |
73 | strerror(errno)); | |
74 | return false; | |
75 | } | |
76 | ||
77 | ioctl_mask = (__u64)1 << _UFFDIO_REGISTER | | |
78 | (__u64)1 << _UFFDIO_UNREGISTER; | |
79 | if ((api_struct.ioctls & ioctl_mask) != ioctl_mask) { | |
80 | error_report("Missing userfault features: %" PRIx64, | |
81 | (uint64_t)(~api_struct.ioctls & ioctl_mask)); | |
82 | return false; | |
83 | } | |
84 | ||
85 | return true; | |
86 | } | |
87 | ||
58b7c17e DDAG |
88 | /* |
89 | * Note: This has the side effect of munlock'ing all of RAM, that's | |
90 | * normally fine since if the postcopy succeeds it gets turned back on at the | |
91 | * end. | |
92 | */ | |
eb59db53 DDAG |
93 | bool postcopy_ram_supported_by_host(void) |
94 | { | |
95 | long pagesize = getpagesize(); | |
96 | int ufd = -1; | |
97 | bool ret = false; /* Error unless we change it */ | |
98 | void *testarea = NULL; | |
99 | struct uffdio_register reg_struct; | |
100 | struct uffdio_range range_struct; | |
101 | uint64_t feature_mask; | |
102 | ||
103 | if ((1ul << qemu_target_page_bits()) > pagesize) { | |
104 | error_report("Target page size bigger than host page size"); | |
105 | goto out; | |
106 | } | |
107 | ||
108 | ufd = syscall(__NR_userfaultfd, O_CLOEXEC); | |
109 | if (ufd == -1) { | |
110 | error_report("%s: userfaultfd not available: %s", __func__, | |
111 | strerror(errno)); | |
112 | goto out; | |
113 | } | |
114 | ||
115 | /* Version and features check */ | |
116 | if (!ufd_version_check(ufd)) { | |
117 | goto out; | |
118 | } | |
119 | ||
58b7c17e DDAG |
120 | /* |
121 | * userfault and mlock don't go together; we'll put it back later if | |
122 | * it was enabled. | |
123 | */ | |
124 | if (munlockall()) { | |
125 | error_report("%s: munlockall: %s", __func__, strerror(errno)); | |
126 | return -1; | |
127 | } | |
128 | ||
eb59db53 DDAG |
129 | /* |
130 | * We need to check that the ops we need are supported on anon memory | |
131 | * To do that we need to register a chunk and see the flags that | |
132 | * are returned. | |
133 | */ | |
134 | testarea = mmap(NULL, pagesize, PROT_READ | PROT_WRITE, MAP_PRIVATE | | |
135 | MAP_ANONYMOUS, -1, 0); | |
136 | if (testarea == MAP_FAILED) { | |
137 | error_report("%s: Failed to map test area: %s", __func__, | |
138 | strerror(errno)); | |
139 | goto out; | |
140 | } | |
141 | g_assert(((size_t)testarea & (pagesize-1)) == 0); | |
142 | ||
143 | reg_struct.range.start = (uintptr_t)testarea; | |
144 | reg_struct.range.len = pagesize; | |
145 | reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING; | |
146 | ||
147 | if (ioctl(ufd, UFFDIO_REGISTER, ®_struct)) { | |
148 | error_report("%s userfault register: %s", __func__, strerror(errno)); | |
149 | goto out; | |
150 | } | |
151 | ||
152 | range_struct.start = (uintptr_t)testarea; | |
153 | range_struct.len = pagesize; | |
154 | if (ioctl(ufd, UFFDIO_UNREGISTER, &range_struct)) { | |
155 | error_report("%s userfault unregister: %s", __func__, strerror(errno)); | |
156 | goto out; | |
157 | } | |
158 | ||
159 | feature_mask = (__u64)1 << _UFFDIO_WAKE | | |
160 | (__u64)1 << _UFFDIO_COPY | | |
161 | (__u64)1 << _UFFDIO_ZEROPAGE; | |
162 | if ((reg_struct.ioctls & feature_mask) != feature_mask) { | |
163 | error_report("Missing userfault map features: %" PRIx64, | |
164 | (uint64_t)(~reg_struct.ioctls & feature_mask)); | |
165 | goto out; | |
166 | } | |
167 | ||
168 | /* Success! */ | |
169 | ret = true; | |
170 | out: | |
171 | if (testarea) { | |
172 | munmap(testarea, pagesize); | |
173 | } | |
174 | if (ufd != -1) { | |
175 | close(ufd); | |
176 | } | |
177 | return ret; | |
178 | } | |
179 | ||
e0b266f0 DDAG |
180 | /** |
181 | * postcopy_ram_discard_range: Discard a range of memory. | |
182 | * We can assume that if we've been called postcopy_ram_hosttest returned true. | |
183 | * | |
184 | * @mis: Current incoming migration state. | |
185 | * @start, @length: range of memory to discard. | |
186 | * | |
187 | * returns: 0 on success. | |
188 | */ | |
189 | int postcopy_ram_discard_range(MigrationIncomingState *mis, uint8_t *start, | |
190 | size_t length) | |
191 | { | |
192 | trace_postcopy_ram_discard_range(start, length); | |
193 | if (madvise(start, length, MADV_DONTNEED)) { | |
194 | error_report("%s MADV_DONTNEED: %s", __func__, strerror(errno)); | |
195 | return -1; | |
196 | } | |
197 | ||
198 | return 0; | |
199 | } | |
200 | ||
1caddf8a DDAG |
201 | /* |
202 | * Setup an area of RAM so that it *can* be used for postcopy later; this | |
203 | * must be done right at the start prior to pre-copy. | |
204 | * opaque should be the MIS. | |
205 | */ | |
206 | static int init_range(const char *block_name, void *host_addr, | |
207 | ram_addr_t offset, ram_addr_t length, void *opaque) | |
208 | { | |
209 | MigrationIncomingState *mis = opaque; | |
210 | ||
211 | trace_postcopy_init_range(block_name, host_addr, offset, length); | |
212 | ||
213 | /* | |
214 | * We need the whole of RAM to be truly empty for postcopy, so things | |
215 | * like ROMs and any data tables built during init must be zero'd | |
216 | * - we're going to get the copy from the source anyway. | |
217 | * (Precopy will just overwrite this data, so doesn't need the discard) | |
218 | */ | |
219 | if (postcopy_ram_discard_range(mis, host_addr, length)) { | |
220 | return -1; | |
221 | } | |
222 | ||
223 | return 0; | |
224 | } | |
225 | ||
226 | /* | |
227 | * At the end of migration, undo the effects of init_range | |
228 | * opaque should be the MIS. | |
229 | */ | |
230 | static int cleanup_range(const char *block_name, void *host_addr, | |
231 | ram_addr_t offset, ram_addr_t length, void *opaque) | |
232 | { | |
233 | MigrationIncomingState *mis = opaque; | |
234 | struct uffdio_range range_struct; | |
235 | trace_postcopy_cleanup_range(block_name, host_addr, offset, length); | |
236 | ||
237 | /* | |
238 | * We turned off hugepage for the precopy stage with postcopy enabled | |
239 | * we can turn it back on now. | |
240 | */ | |
1d741439 | 241 | qemu_madvise(host_addr, length, QEMU_MADV_HUGEPAGE); |
1caddf8a DDAG |
242 | |
243 | /* | |
244 | * We can also turn off userfault now since we should have all the | |
245 | * pages. It can be useful to leave it on to debug postcopy | |
246 | * if you're not sure it's always getting every page. | |
247 | */ | |
248 | range_struct.start = (uintptr_t)host_addr; | |
249 | range_struct.len = length; | |
250 | ||
251 | if (ioctl(mis->userfault_fd, UFFDIO_UNREGISTER, &range_struct)) { | |
252 | error_report("%s: userfault unregister %s", __func__, strerror(errno)); | |
253 | ||
254 | return -1; | |
255 | } | |
256 | ||
257 | return 0; | |
258 | } | |
259 | ||
260 | /* | |
261 | * Initialise postcopy-ram, setting the RAM to a state where we can go into | |
262 | * postcopy later; must be called prior to any precopy. | |
263 | * called from arch_init's similarly named ram_postcopy_incoming_init | |
264 | */ | |
265 | int postcopy_ram_incoming_init(MigrationIncomingState *mis, size_t ram_pages) | |
266 | { | |
267 | if (qemu_ram_foreach_block(init_range, mis)) { | |
268 | return -1; | |
269 | } | |
270 | ||
271 | return 0; | |
272 | } | |
273 | ||
274 | /* | |
275 | * At the end of a migration where postcopy_ram_incoming_init was called. | |
276 | */ | |
277 | int postcopy_ram_incoming_cleanup(MigrationIncomingState *mis) | |
278 | { | |
c4faeed2 DDAG |
279 | trace_postcopy_ram_incoming_cleanup_entry(); |
280 | ||
281 | if (mis->have_fault_thread) { | |
282 | uint64_t tmp64; | |
283 | ||
284 | if (qemu_ram_foreach_block(cleanup_range, mis)) { | |
285 | return -1; | |
286 | } | |
287 | /* | |
288 | * Tell the fault_thread to exit, it's an eventfd that should | |
289 | * currently be at 0, we're going to increment it to 1 | |
290 | */ | |
291 | tmp64 = 1; | |
292 | if (write(mis->userfault_quit_fd, &tmp64, 8) == 8) { | |
293 | trace_postcopy_ram_incoming_cleanup_join(); | |
294 | qemu_thread_join(&mis->fault_thread); | |
295 | } else { | |
296 | /* Not much we can do here, but may as well report it */ | |
297 | error_report("%s: incrementing userfault_quit_fd: %s", __func__, | |
298 | strerror(errno)); | |
299 | } | |
300 | trace_postcopy_ram_incoming_cleanup_closeuf(); | |
301 | close(mis->userfault_fd); | |
302 | close(mis->userfault_quit_fd); | |
303 | mis->have_fault_thread = false; | |
1caddf8a DDAG |
304 | } |
305 | ||
371ff5a3 DDAG |
306 | qemu_balloon_inhibit(false); |
307 | ||
58b7c17e DDAG |
308 | if (enable_mlock) { |
309 | if (os_mlock() < 0) { | |
310 | error_report("mlock: %s", strerror(errno)); | |
311 | /* | |
312 | * It doesn't feel right to fail at this point, we have a valid | |
313 | * VM state. | |
314 | */ | |
315 | } | |
316 | } | |
317 | ||
c4faeed2 DDAG |
318 | postcopy_state_set(POSTCOPY_INCOMING_END); |
319 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
320 | ||
696ed9a9 DDAG |
321 | if (mis->postcopy_tmp_page) { |
322 | munmap(mis->postcopy_tmp_page, getpagesize()); | |
323 | mis->postcopy_tmp_page = NULL; | |
324 | } | |
c4faeed2 | 325 | trace_postcopy_ram_incoming_cleanup_exit(); |
1caddf8a DDAG |
326 | return 0; |
327 | } | |
328 | ||
f9527107 DDAG |
329 | /* |
330 | * Disable huge pages on an area | |
331 | */ | |
332 | static int nhp_range(const char *block_name, void *host_addr, | |
333 | ram_addr_t offset, ram_addr_t length, void *opaque) | |
334 | { | |
335 | trace_postcopy_nhp_range(block_name, host_addr, offset, length); | |
336 | ||
337 | /* | |
338 | * Before we do discards we need to ensure those discards really | |
339 | * do delete areas of the page, even if THP thinks a hugepage would | |
340 | * be a good idea, so force hugepages off. | |
341 | */ | |
1d741439 | 342 | qemu_madvise(host_addr, length, QEMU_MADV_NOHUGEPAGE); |
f9527107 DDAG |
343 | |
344 | return 0; | |
345 | } | |
346 | ||
347 | /* | |
348 | * Userfault requires us to mark RAM as NOHUGEPAGE prior to discard | |
349 | * however leaving it until after precopy means that most of the precopy | |
350 | * data is still THPd | |
351 | */ | |
352 | int postcopy_ram_prepare_discard(MigrationIncomingState *mis) | |
353 | { | |
354 | if (qemu_ram_foreach_block(nhp_range, mis)) { | |
355 | return -1; | |
356 | } | |
357 | ||
358 | postcopy_state_set(POSTCOPY_INCOMING_DISCARD); | |
359 | ||
360 | return 0; | |
361 | } | |
362 | ||
f0a227ad DDAG |
363 | /* |
364 | * Mark the given area of RAM as requiring notification to unwritten areas | |
365 | * Used as a callback on qemu_ram_foreach_block. | |
366 | * host_addr: Base of area to mark | |
367 | * offset: Offset in the whole ram arena | |
368 | * length: Length of the section | |
369 | * opaque: MigrationIncomingState pointer | |
370 | * Returns 0 on success | |
371 | */ | |
372 | static int ram_block_enable_notify(const char *block_name, void *host_addr, | |
373 | ram_addr_t offset, ram_addr_t length, | |
374 | void *opaque) | |
375 | { | |
376 | MigrationIncomingState *mis = opaque; | |
377 | struct uffdio_register reg_struct; | |
378 | ||
379 | reg_struct.range.start = (uintptr_t)host_addr; | |
380 | reg_struct.range.len = length; | |
381 | reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING; | |
382 | ||
383 | /* Now tell our userfault_fd that it's responsible for this area */ | |
384 | if (ioctl(mis->userfault_fd, UFFDIO_REGISTER, ®_struct)) { | |
385 | error_report("%s userfault register: %s", __func__, strerror(errno)); | |
386 | return -1; | |
387 | } | |
388 | ||
389 | return 0; | |
390 | } | |
391 | ||
392 | /* | |
393 | * Handle faults detected by the USERFAULT markings | |
394 | */ | |
395 | static void *postcopy_ram_fault_thread(void *opaque) | |
396 | { | |
397 | MigrationIncomingState *mis = opaque; | |
c4faeed2 DDAG |
398 | struct uffd_msg msg; |
399 | int ret; | |
400 | size_t hostpagesize = getpagesize(); | |
401 | RAMBlock *rb = NULL; | |
402 | RAMBlock *last_rb = NULL; /* last RAMBlock we sent part of */ | |
f0a227ad | 403 | |
c4faeed2 | 404 | trace_postcopy_ram_fault_thread_entry(); |
f0a227ad | 405 | qemu_sem_post(&mis->fault_thread_sem); |
f0a227ad | 406 | |
c4faeed2 DDAG |
407 | while (true) { |
408 | ram_addr_t rb_offset; | |
c4faeed2 DDAG |
409 | struct pollfd pfd[2]; |
410 | ||
411 | /* | |
412 | * We're mainly waiting for the kernel to give us a faulting HVA, | |
413 | * however we can be told to quit via userfault_quit_fd which is | |
414 | * an eventfd | |
415 | */ | |
416 | pfd[0].fd = mis->userfault_fd; | |
417 | pfd[0].events = POLLIN; | |
418 | pfd[0].revents = 0; | |
419 | pfd[1].fd = mis->userfault_quit_fd; | |
420 | pfd[1].events = POLLIN; /* Waiting for eventfd to go positive */ | |
421 | pfd[1].revents = 0; | |
422 | ||
423 | if (poll(pfd, 2, -1 /* Wait forever */) == -1) { | |
424 | error_report("%s: userfault poll: %s", __func__, strerror(errno)); | |
425 | break; | |
426 | } | |
427 | ||
428 | if (pfd[1].revents) { | |
429 | trace_postcopy_ram_fault_thread_quit(); | |
430 | break; | |
431 | } | |
432 | ||
433 | ret = read(mis->userfault_fd, &msg, sizeof(msg)); | |
434 | if (ret != sizeof(msg)) { | |
435 | if (errno == EAGAIN) { | |
436 | /* | |
437 | * if a wake up happens on the other thread just after | |
438 | * the poll, there is nothing to read. | |
439 | */ | |
440 | continue; | |
441 | } | |
442 | if (ret < 0) { | |
443 | error_report("%s: Failed to read full userfault message: %s", | |
444 | __func__, strerror(errno)); | |
445 | break; | |
446 | } else { | |
447 | error_report("%s: Read %d bytes from userfaultfd expected %zd", | |
448 | __func__, ret, sizeof(msg)); | |
449 | break; /* Lost alignment, don't know what we'd read next */ | |
450 | } | |
451 | } | |
452 | if (msg.event != UFFD_EVENT_PAGEFAULT) { | |
453 | error_report("%s: Read unexpected event %ud from userfaultfd", | |
454 | __func__, msg.event); | |
455 | continue; /* It's not a page fault, shouldn't happen */ | |
456 | } | |
457 | ||
458 | rb = qemu_ram_block_from_host( | |
459 | (void *)(uintptr_t)msg.arg.pagefault.address, | |
f615f396 | 460 | true, &rb_offset); |
c4faeed2 DDAG |
461 | if (!rb) { |
462 | error_report("postcopy_ram_fault_thread: Fault outside guest: %" | |
463 | PRIx64, (uint64_t)msg.arg.pagefault.address); | |
464 | break; | |
465 | } | |
466 | ||
467 | rb_offset &= ~(hostpagesize - 1); | |
468 | trace_postcopy_ram_fault_thread_request(msg.arg.pagefault.address, | |
469 | qemu_ram_get_idstr(rb), | |
470 | rb_offset); | |
471 | ||
472 | /* | |
473 | * Send the request to the source - we want to request one | |
474 | * of our host page sizes (which is >= TPS) | |
475 | */ | |
476 | if (rb != last_rb) { | |
477 | last_rb = rb; | |
478 | migrate_send_rp_req_pages(mis, qemu_ram_get_idstr(rb), | |
479 | rb_offset, hostpagesize); | |
480 | } else { | |
481 | /* Save some space */ | |
482 | migrate_send_rp_req_pages(mis, NULL, | |
483 | rb_offset, hostpagesize); | |
484 | } | |
485 | } | |
486 | trace_postcopy_ram_fault_thread_exit(); | |
f0a227ad DDAG |
487 | return NULL; |
488 | } | |
489 | ||
490 | int postcopy_ram_enable_notify(MigrationIncomingState *mis) | |
491 | { | |
c4faeed2 DDAG |
492 | /* Open the fd for the kernel to give us userfaults */ |
493 | mis->userfault_fd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK); | |
494 | if (mis->userfault_fd == -1) { | |
495 | error_report("%s: Failed to open userfault fd: %s", __func__, | |
496 | strerror(errno)); | |
497 | return -1; | |
498 | } | |
499 | ||
500 | /* | |
501 | * Although the host check already tested the API, we need to | |
502 | * do the check again as an ABI handshake on the new fd. | |
503 | */ | |
504 | if (!ufd_version_check(mis->userfault_fd)) { | |
505 | return -1; | |
506 | } | |
507 | ||
508 | /* Now an eventfd we use to tell the fault-thread to quit */ | |
509 | mis->userfault_quit_fd = eventfd(0, EFD_CLOEXEC); | |
510 | if (mis->userfault_quit_fd == -1) { | |
511 | error_report("%s: Opening userfault_quit_fd: %s", __func__, | |
512 | strerror(errno)); | |
513 | close(mis->userfault_fd); | |
514 | return -1; | |
515 | } | |
516 | ||
f0a227ad DDAG |
517 | qemu_sem_init(&mis->fault_thread_sem, 0); |
518 | qemu_thread_create(&mis->fault_thread, "postcopy/fault", | |
519 | postcopy_ram_fault_thread, mis, QEMU_THREAD_JOINABLE); | |
520 | qemu_sem_wait(&mis->fault_thread_sem); | |
521 | qemu_sem_destroy(&mis->fault_thread_sem); | |
c4faeed2 | 522 | mis->have_fault_thread = true; |
f0a227ad DDAG |
523 | |
524 | /* Mark so that we get notified of accesses to unwritten areas */ | |
525 | if (qemu_ram_foreach_block(ram_block_enable_notify, mis)) { | |
526 | return -1; | |
527 | } | |
528 | ||
371ff5a3 DDAG |
529 | /* |
530 | * Ballooning can mark pages as absent while we're postcopying | |
531 | * that would cause false userfaults. | |
532 | */ | |
533 | qemu_balloon_inhibit(true); | |
534 | ||
c4faeed2 DDAG |
535 | trace_postcopy_ram_enable_notify(); |
536 | ||
f0a227ad DDAG |
537 | return 0; |
538 | } | |
539 | ||
696ed9a9 DDAG |
540 | /* |
541 | * Place a host page (from) at (host) atomically | |
542 | * returns 0 on success | |
543 | */ | |
544 | int postcopy_place_page(MigrationIncomingState *mis, void *host, void *from) | |
545 | { | |
546 | struct uffdio_copy copy_struct; | |
547 | ||
548 | copy_struct.dst = (uint64_t)(uintptr_t)host; | |
549 | copy_struct.src = (uint64_t)(uintptr_t)from; | |
550 | copy_struct.len = getpagesize(); | |
551 | copy_struct.mode = 0; | |
552 | ||
553 | /* copy also acks to the kernel waking the stalled thread up | |
554 | * TODO: We can inhibit that ack and only do it if it was requested | |
555 | * which would be slightly cheaper, but we'd have to be careful | |
556 | * of the order of updating our page state. | |
557 | */ | |
558 | if (ioctl(mis->userfault_fd, UFFDIO_COPY, ©_struct)) { | |
559 | int e = errno; | |
560 | error_report("%s: %s copy host: %p from: %p", | |
561 | __func__, strerror(e), host, from); | |
562 | ||
563 | return -e; | |
564 | } | |
565 | ||
566 | trace_postcopy_place_page(host); | |
567 | return 0; | |
568 | } | |
569 | ||
570 | /* | |
571 | * Place a zero page at (host) atomically | |
572 | * returns 0 on success | |
573 | */ | |
574 | int postcopy_place_page_zero(MigrationIncomingState *mis, void *host) | |
575 | { | |
576 | struct uffdio_zeropage zero_struct; | |
577 | ||
578 | zero_struct.range.start = (uint64_t)(uintptr_t)host; | |
579 | zero_struct.range.len = getpagesize(); | |
580 | zero_struct.mode = 0; | |
581 | ||
582 | if (ioctl(mis->userfault_fd, UFFDIO_ZEROPAGE, &zero_struct)) { | |
583 | int e = errno; | |
584 | error_report("%s: %s zero host: %p", | |
585 | __func__, strerror(e), host); | |
586 | ||
587 | return -e; | |
588 | } | |
589 | ||
590 | trace_postcopy_place_page_zero(host); | |
591 | return 0; | |
592 | } | |
593 | ||
594 | /* | |
595 | * Returns a target page of memory that can be mapped at a later point in time | |
596 | * using postcopy_place_page | |
597 | * The same address is used repeatedly, postcopy_place_page just takes the | |
598 | * backing page away. | |
599 | * Returns: Pointer to allocated page | |
600 | * | |
601 | */ | |
602 | void *postcopy_get_tmp_page(MigrationIncomingState *mis) | |
603 | { | |
604 | if (!mis->postcopy_tmp_page) { | |
605 | mis->postcopy_tmp_page = mmap(NULL, getpagesize(), | |
606 | PROT_READ | PROT_WRITE, MAP_PRIVATE | | |
607 | MAP_ANONYMOUS, -1, 0); | |
608 | if (!mis->postcopy_tmp_page) { | |
609 | error_report("%s: %s", __func__, strerror(errno)); | |
610 | return NULL; | |
611 | } | |
612 | } | |
613 | ||
614 | return mis->postcopy_tmp_page; | |
615 | } | |
616 | ||
eb59db53 DDAG |
617 | #else |
618 | /* No target OS support, stubs just fail */ | |
619 | bool postcopy_ram_supported_by_host(void) | |
620 | { | |
621 | error_report("%s: No OS support", __func__); | |
622 | return false; | |
623 | } | |
624 | ||
1caddf8a DDAG |
625 | int postcopy_ram_incoming_init(MigrationIncomingState *mis, size_t ram_pages) |
626 | { | |
627 | error_report("postcopy_ram_incoming_init: No OS support"); | |
628 | return -1; | |
629 | } | |
630 | ||
631 | int postcopy_ram_incoming_cleanup(MigrationIncomingState *mis) | |
632 | { | |
633 | assert(0); | |
634 | return -1; | |
635 | } | |
636 | ||
e0b266f0 DDAG |
637 | int postcopy_ram_discard_range(MigrationIncomingState *mis, uint8_t *start, |
638 | size_t length) | |
639 | { | |
640 | assert(0); | |
1caddf8a | 641 | return -1; |
e0b266f0 | 642 | } |
f0a227ad | 643 | |
f9527107 DDAG |
644 | int postcopy_ram_prepare_discard(MigrationIncomingState *mis) |
645 | { | |
646 | assert(0); | |
647 | return -1; | |
648 | } | |
649 | ||
f0a227ad DDAG |
650 | int postcopy_ram_enable_notify(MigrationIncomingState *mis) |
651 | { | |
652 | assert(0); | |
653 | return -1; | |
654 | } | |
696ed9a9 DDAG |
655 | |
656 | int postcopy_place_page(MigrationIncomingState *mis, void *host, void *from) | |
657 | { | |
658 | assert(0); | |
659 | return -1; | |
660 | } | |
661 | ||
662 | int postcopy_place_page_zero(MigrationIncomingState *mis, void *host) | |
663 | { | |
664 | assert(0); | |
665 | return -1; | |
666 | } | |
667 | ||
668 | void *postcopy_get_tmp_page(MigrationIncomingState *mis) | |
669 | { | |
670 | assert(0); | |
671 | return NULL; | |
672 | } | |
673 | ||
eb59db53 DDAG |
674 | #endif |
675 | ||
e0b266f0 DDAG |
676 | /* ------------------------------------------------------------------------- */ |
677 | ||
678 | /** | |
679 | * postcopy_discard_send_init: Called at the start of each RAMBlock before | |
680 | * asking to discard individual ranges. | |
681 | * | |
682 | * @ms: The current migration state. | |
683 | * @offset: the bitmap offset of the named RAMBlock in the migration | |
684 | * bitmap. | |
685 | * @name: RAMBlock that discards will operate on. | |
686 | * | |
687 | * returns: a new PDS. | |
688 | */ | |
689 | PostcopyDiscardState *postcopy_discard_send_init(MigrationState *ms, | |
690 | unsigned long offset, | |
691 | const char *name) | |
692 | { | |
693 | PostcopyDiscardState *res = g_malloc0(sizeof(PostcopyDiscardState)); | |
694 | ||
695 | if (res) { | |
696 | res->ramblock_name = name; | |
697 | res->offset = offset; | |
698 | } | |
699 | ||
700 | return res; | |
701 | } | |
702 | ||
703 | /** | |
704 | * postcopy_discard_send_range: Called by the bitmap code for each chunk to | |
705 | * discard. May send a discard message, may just leave it queued to | |
706 | * be sent later. | |
707 | * | |
708 | * @ms: Current migration state. | |
709 | * @pds: Structure initialised by postcopy_discard_send_init(). | |
710 | * @start,@length: a range of pages in the migration bitmap in the | |
711 | * RAM block passed to postcopy_discard_send_init() (length=1 is one page) | |
712 | */ | |
713 | void postcopy_discard_send_range(MigrationState *ms, PostcopyDiscardState *pds, | |
714 | unsigned long start, unsigned long length) | |
715 | { | |
716 | size_t tp_bits = qemu_target_page_bits(); | |
717 | /* Convert to byte offsets within the RAM block */ | |
718 | pds->start_list[pds->cur_entry] = (start - pds->offset) << tp_bits; | |
719 | pds->length_list[pds->cur_entry] = length << tp_bits; | |
720 | trace_postcopy_discard_send_range(pds->ramblock_name, start, length); | |
721 | pds->cur_entry++; | |
722 | pds->nsentwords++; | |
723 | ||
724 | if (pds->cur_entry == MAX_DISCARDS_PER_COMMAND) { | |
725 | /* Full set, ship it! */ | |
89a02a9f HZ |
726 | qemu_savevm_send_postcopy_ram_discard(ms->to_dst_file, |
727 | pds->ramblock_name, | |
e0b266f0 DDAG |
728 | pds->cur_entry, |
729 | pds->start_list, | |
730 | pds->length_list); | |
731 | pds->nsentcmds++; | |
732 | pds->cur_entry = 0; | |
733 | } | |
734 | } | |
735 | ||
736 | /** | |
737 | * postcopy_discard_send_finish: Called at the end of each RAMBlock by the | |
738 | * bitmap code. Sends any outstanding discard messages, frees the PDS | |
739 | * | |
740 | * @ms: Current migration state. | |
741 | * @pds: Structure initialised by postcopy_discard_send_init(). | |
742 | */ | |
743 | void postcopy_discard_send_finish(MigrationState *ms, PostcopyDiscardState *pds) | |
744 | { | |
745 | /* Anything unsent? */ | |
746 | if (pds->cur_entry) { | |
89a02a9f HZ |
747 | qemu_savevm_send_postcopy_ram_discard(ms->to_dst_file, |
748 | pds->ramblock_name, | |
e0b266f0 DDAG |
749 | pds->cur_entry, |
750 | pds->start_list, | |
751 | pds->length_list); | |
752 | pds->nsentcmds++; | |
753 | } | |
754 | ||
755 | trace_postcopy_discard_send_finish(pds->ramblock_name, pds->nsentwords, | |
756 | pds->nsentcmds); | |
757 | ||
758 | g_free(pds); | |
759 | } |