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