ppc/pnv: add a PnvICPState object
[qemu.git] / migration / postcopy-ram.c
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
19 #include "qemu/osdep.h"
20
21 #include "qemu-common.h"
22 #include "migration/migration.h"
23 #include "migration/postcopy-ram.h"
24 #include "sysemu/sysemu.h"
25 #include "sysemu/balloon.h"
26 #include "qemu/error-report.h"
27 #include "trace.h"
28
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
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
53 #include <poll.h>
54 #include <sys/ioctl.h>
55 #include <sys/syscall.h>
56 #include <asm/types.h> /* for __u64 */
57 #endif
58
59 #if defined(__linux__) && defined(__NR_userfaultfd) && defined(CONFIG_EVENTFD)
60 #include <sys/eventfd.h>
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 if (getpagesize() != ram_pagesize_summary()) {
85 bool have_hp = false;
86 /* We've got a huge page */
87 #ifdef UFFD_FEATURE_MISSING_HUGETLBFS
88 have_hp = api_struct.features & UFFD_FEATURE_MISSING_HUGETLBFS;
89 #endif
90 if (!have_hp) {
91 error_report("Userfault on this host does not support huge pages");
92 return false;
93 }
94 }
95 return true;
96 }
97
98 /* Callback from postcopy_ram_supported_by_host block iterator.
99 */
100 static int test_range_shared(const char *block_name, void *host_addr,
101 ram_addr_t offset, ram_addr_t length, void *opaque)
102 {
103 if (qemu_ram_is_shared(qemu_ram_block_by_name(block_name))) {
104 error_report("Postcopy on shared RAM (%s) is not yet supported",
105 block_name);
106 return 1;
107 }
108 return 0;
109 }
110
111 /*
112 * Note: This has the side effect of munlock'ing all of RAM, that's
113 * normally fine since if the postcopy succeeds it gets turned back on at the
114 * end.
115 */
116 bool postcopy_ram_supported_by_host(void)
117 {
118 long pagesize = getpagesize();
119 int ufd = -1;
120 bool ret = false; /* Error unless we change it */
121 void *testarea = NULL;
122 struct uffdio_register reg_struct;
123 struct uffdio_range range_struct;
124 uint64_t feature_mask;
125
126 if (qemu_target_page_size() > pagesize) {
127 error_report("Target page size bigger than host page size");
128 goto out;
129 }
130
131 ufd = syscall(__NR_userfaultfd, O_CLOEXEC);
132 if (ufd == -1) {
133 error_report("%s: userfaultfd not available: %s", __func__,
134 strerror(errno));
135 goto out;
136 }
137
138 /* Version and features check */
139 if (!ufd_version_check(ufd)) {
140 goto out;
141 }
142
143 /* We don't support postcopy with shared RAM yet */
144 if (qemu_ram_foreach_block(test_range_shared, NULL)) {
145 goto out;
146 }
147
148 /*
149 * userfault and mlock don't go together; we'll put it back later if
150 * it was enabled.
151 */
152 if (munlockall()) {
153 error_report("%s: munlockall: %s", __func__, strerror(errno));
154 return -1;
155 }
156
157 /*
158 * We need to check that the ops we need are supported on anon memory
159 * To do that we need to register a chunk and see the flags that
160 * are returned.
161 */
162 testarea = mmap(NULL, pagesize, PROT_READ | PROT_WRITE, MAP_PRIVATE |
163 MAP_ANONYMOUS, -1, 0);
164 if (testarea == MAP_FAILED) {
165 error_report("%s: Failed to map test area: %s", __func__,
166 strerror(errno));
167 goto out;
168 }
169 g_assert(((size_t)testarea & (pagesize-1)) == 0);
170
171 reg_struct.range.start = (uintptr_t)testarea;
172 reg_struct.range.len = pagesize;
173 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
174
175 if (ioctl(ufd, UFFDIO_REGISTER, &reg_struct)) {
176 error_report("%s userfault register: %s", __func__, strerror(errno));
177 goto out;
178 }
179
180 range_struct.start = (uintptr_t)testarea;
181 range_struct.len = pagesize;
182 if (ioctl(ufd, UFFDIO_UNREGISTER, &range_struct)) {
183 error_report("%s userfault unregister: %s", __func__, strerror(errno));
184 goto out;
185 }
186
187 feature_mask = (__u64)1 << _UFFDIO_WAKE |
188 (__u64)1 << _UFFDIO_COPY |
189 (__u64)1 << _UFFDIO_ZEROPAGE;
190 if ((reg_struct.ioctls & feature_mask) != feature_mask) {
191 error_report("Missing userfault map features: %" PRIx64,
192 (uint64_t)(~reg_struct.ioctls & feature_mask));
193 goto out;
194 }
195
196 /* Success! */
197 ret = true;
198 out:
199 if (testarea) {
200 munmap(testarea, pagesize);
201 }
202 if (ufd != -1) {
203 close(ufd);
204 }
205 return ret;
206 }
207
208 /*
209 * Setup an area of RAM so that it *can* be used for postcopy later; this
210 * must be done right at the start prior to pre-copy.
211 * opaque should be the MIS.
212 */
213 static int init_range(const char *block_name, void *host_addr,
214 ram_addr_t offset, ram_addr_t length, void *opaque)
215 {
216 trace_postcopy_init_range(block_name, host_addr, offset, length);
217
218 /*
219 * We need the whole of RAM to be truly empty for postcopy, so things
220 * like ROMs and any data tables built during init must be zero'd
221 * - we're going to get the copy from the source anyway.
222 * (Precopy will just overwrite this data, so doesn't need the discard)
223 */
224 if (ram_discard_range(block_name, 0, length)) {
225 return -1;
226 }
227
228 return 0;
229 }
230
231 /*
232 * At the end of migration, undo the effects of init_range
233 * opaque should be the MIS.
234 */
235 static int cleanup_range(const char *block_name, void *host_addr,
236 ram_addr_t offset, ram_addr_t length, void *opaque)
237 {
238 MigrationIncomingState *mis = opaque;
239 struct uffdio_range range_struct;
240 trace_postcopy_cleanup_range(block_name, host_addr, offset, length);
241
242 /*
243 * We turned off hugepage for the precopy stage with postcopy enabled
244 * we can turn it back on now.
245 */
246 qemu_madvise(host_addr, length, QEMU_MADV_HUGEPAGE);
247
248 /*
249 * We can also turn off userfault now since we should have all the
250 * pages. It can be useful to leave it on to debug postcopy
251 * if you're not sure it's always getting every page.
252 */
253 range_struct.start = (uintptr_t)host_addr;
254 range_struct.len = length;
255
256 if (ioctl(mis->userfault_fd, UFFDIO_UNREGISTER, &range_struct)) {
257 error_report("%s: userfault unregister %s", __func__, strerror(errno));
258
259 return -1;
260 }
261
262 return 0;
263 }
264
265 /*
266 * Initialise postcopy-ram, setting the RAM to a state where we can go into
267 * postcopy later; must be called prior to any precopy.
268 * called from arch_init's similarly named ram_postcopy_incoming_init
269 */
270 int postcopy_ram_incoming_init(MigrationIncomingState *mis, size_t ram_pages)
271 {
272 if (qemu_ram_foreach_block(init_range, NULL)) {
273 return -1;
274 }
275
276 return 0;
277 }
278
279 /*
280 * At the end of a migration where postcopy_ram_incoming_init was called.
281 */
282 int postcopy_ram_incoming_cleanup(MigrationIncomingState *mis)
283 {
284 trace_postcopy_ram_incoming_cleanup_entry();
285
286 if (mis->have_fault_thread) {
287 uint64_t tmp64;
288
289 if (qemu_ram_foreach_block(cleanup_range, mis)) {
290 return -1;
291 }
292 /*
293 * Tell the fault_thread to exit, it's an eventfd that should
294 * currently be at 0, we're going to increment it to 1
295 */
296 tmp64 = 1;
297 if (write(mis->userfault_quit_fd, &tmp64, 8) == 8) {
298 trace_postcopy_ram_incoming_cleanup_join();
299 qemu_thread_join(&mis->fault_thread);
300 } else {
301 /* Not much we can do here, but may as well report it */
302 error_report("%s: incrementing userfault_quit_fd: %s", __func__,
303 strerror(errno));
304 }
305 trace_postcopy_ram_incoming_cleanup_closeuf();
306 close(mis->userfault_fd);
307 close(mis->userfault_quit_fd);
308 mis->have_fault_thread = false;
309 }
310
311 qemu_balloon_inhibit(false);
312
313 if (enable_mlock) {
314 if (os_mlock() < 0) {
315 error_report("mlock: %s", strerror(errno));
316 /*
317 * It doesn't feel right to fail at this point, we have a valid
318 * VM state.
319 */
320 }
321 }
322
323 postcopy_state_set(POSTCOPY_INCOMING_END);
324 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
325
326 if (mis->postcopy_tmp_page) {
327 munmap(mis->postcopy_tmp_page, mis->largest_page_size);
328 mis->postcopy_tmp_page = NULL;
329 }
330 if (mis->postcopy_tmp_zero_page) {
331 munmap(mis->postcopy_tmp_zero_page, mis->largest_page_size);
332 mis->postcopy_tmp_zero_page = NULL;
333 }
334 trace_postcopy_ram_incoming_cleanup_exit();
335 return 0;
336 }
337
338 /*
339 * Disable huge pages on an area
340 */
341 static int nhp_range(const char *block_name, void *host_addr,
342 ram_addr_t offset, ram_addr_t length, void *opaque)
343 {
344 trace_postcopy_nhp_range(block_name, host_addr, offset, length);
345
346 /*
347 * Before we do discards we need to ensure those discards really
348 * do delete areas of the page, even if THP thinks a hugepage would
349 * be a good idea, so force hugepages off.
350 */
351 qemu_madvise(host_addr, length, QEMU_MADV_NOHUGEPAGE);
352
353 return 0;
354 }
355
356 /*
357 * Userfault requires us to mark RAM as NOHUGEPAGE prior to discard
358 * however leaving it until after precopy means that most of the precopy
359 * data is still THPd
360 */
361 int postcopy_ram_prepare_discard(MigrationIncomingState *mis)
362 {
363 if (qemu_ram_foreach_block(nhp_range, mis)) {
364 return -1;
365 }
366
367 postcopy_state_set(POSTCOPY_INCOMING_DISCARD);
368
369 return 0;
370 }
371
372 /*
373 * Mark the given area of RAM as requiring notification to unwritten areas
374 * Used as a callback on qemu_ram_foreach_block.
375 * host_addr: Base of area to mark
376 * offset: Offset in the whole ram arena
377 * length: Length of the section
378 * opaque: MigrationIncomingState pointer
379 * Returns 0 on success
380 */
381 static int ram_block_enable_notify(const char *block_name, void *host_addr,
382 ram_addr_t offset, ram_addr_t length,
383 void *opaque)
384 {
385 MigrationIncomingState *mis = opaque;
386 struct uffdio_register reg_struct;
387
388 reg_struct.range.start = (uintptr_t)host_addr;
389 reg_struct.range.len = length;
390 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
391
392 /* Now tell our userfault_fd that it's responsible for this area */
393 if (ioctl(mis->userfault_fd, UFFDIO_REGISTER, &reg_struct)) {
394 error_report("%s userfault register: %s", __func__, strerror(errno));
395 return -1;
396 }
397 if (!(reg_struct.ioctls & ((__u64)1 << _UFFDIO_COPY))) {
398 error_report("%s userfault: Region doesn't support COPY", __func__);
399 return -1;
400 }
401
402 return 0;
403 }
404
405 /*
406 * Handle faults detected by the USERFAULT markings
407 */
408 static void *postcopy_ram_fault_thread(void *opaque)
409 {
410 MigrationIncomingState *mis = opaque;
411 struct uffd_msg msg;
412 int ret;
413 RAMBlock *rb = NULL;
414 RAMBlock *last_rb = NULL; /* last RAMBlock we sent part of */
415
416 trace_postcopy_ram_fault_thread_entry();
417 qemu_sem_post(&mis->fault_thread_sem);
418
419 while (true) {
420 ram_addr_t rb_offset;
421 struct pollfd pfd[2];
422
423 /*
424 * We're mainly waiting for the kernel to give us a faulting HVA,
425 * however we can be told to quit via userfault_quit_fd which is
426 * an eventfd
427 */
428 pfd[0].fd = mis->userfault_fd;
429 pfd[0].events = POLLIN;
430 pfd[0].revents = 0;
431 pfd[1].fd = mis->userfault_quit_fd;
432 pfd[1].events = POLLIN; /* Waiting for eventfd to go positive */
433 pfd[1].revents = 0;
434
435 if (poll(pfd, 2, -1 /* Wait forever */) == -1) {
436 error_report("%s: userfault poll: %s", __func__, strerror(errno));
437 break;
438 }
439
440 if (pfd[1].revents) {
441 trace_postcopy_ram_fault_thread_quit();
442 break;
443 }
444
445 ret = read(mis->userfault_fd, &msg, sizeof(msg));
446 if (ret != sizeof(msg)) {
447 if (errno == EAGAIN) {
448 /*
449 * if a wake up happens on the other thread just after
450 * the poll, there is nothing to read.
451 */
452 continue;
453 }
454 if (ret < 0) {
455 error_report("%s: Failed to read full userfault message: %s",
456 __func__, strerror(errno));
457 break;
458 } else {
459 error_report("%s: Read %d bytes from userfaultfd expected %zd",
460 __func__, ret, sizeof(msg));
461 break; /* Lost alignment, don't know what we'd read next */
462 }
463 }
464 if (msg.event != UFFD_EVENT_PAGEFAULT) {
465 error_report("%s: Read unexpected event %ud from userfaultfd",
466 __func__, msg.event);
467 continue; /* It's not a page fault, shouldn't happen */
468 }
469
470 rb = qemu_ram_block_from_host(
471 (void *)(uintptr_t)msg.arg.pagefault.address,
472 true, &rb_offset);
473 if (!rb) {
474 error_report("postcopy_ram_fault_thread: Fault outside guest: %"
475 PRIx64, (uint64_t)msg.arg.pagefault.address);
476 break;
477 }
478
479 rb_offset &= ~(qemu_ram_pagesize(rb) - 1);
480 trace_postcopy_ram_fault_thread_request(msg.arg.pagefault.address,
481 qemu_ram_get_idstr(rb),
482 rb_offset);
483
484 /*
485 * Send the request to the source - we want to request one
486 * of our host page sizes (which is >= TPS)
487 */
488 if (rb != last_rb) {
489 last_rb = rb;
490 migrate_send_rp_req_pages(mis, qemu_ram_get_idstr(rb),
491 rb_offset, qemu_ram_pagesize(rb));
492 } else {
493 /* Save some space */
494 migrate_send_rp_req_pages(mis, NULL,
495 rb_offset, qemu_ram_pagesize(rb));
496 }
497 }
498 trace_postcopy_ram_fault_thread_exit();
499 return NULL;
500 }
501
502 int postcopy_ram_enable_notify(MigrationIncomingState *mis)
503 {
504 /* Open the fd for the kernel to give us userfaults */
505 mis->userfault_fd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
506 if (mis->userfault_fd == -1) {
507 error_report("%s: Failed to open userfault fd: %s", __func__,
508 strerror(errno));
509 return -1;
510 }
511
512 /*
513 * Although the host check already tested the API, we need to
514 * do the check again as an ABI handshake on the new fd.
515 */
516 if (!ufd_version_check(mis->userfault_fd)) {
517 return -1;
518 }
519
520 /* Now an eventfd we use to tell the fault-thread to quit */
521 mis->userfault_quit_fd = eventfd(0, EFD_CLOEXEC);
522 if (mis->userfault_quit_fd == -1) {
523 error_report("%s: Opening userfault_quit_fd: %s", __func__,
524 strerror(errno));
525 close(mis->userfault_fd);
526 return -1;
527 }
528
529 qemu_sem_init(&mis->fault_thread_sem, 0);
530 qemu_thread_create(&mis->fault_thread, "postcopy/fault",
531 postcopy_ram_fault_thread, mis, QEMU_THREAD_JOINABLE);
532 qemu_sem_wait(&mis->fault_thread_sem);
533 qemu_sem_destroy(&mis->fault_thread_sem);
534 mis->have_fault_thread = true;
535
536 /* Mark so that we get notified of accesses to unwritten areas */
537 if (qemu_ram_foreach_block(ram_block_enable_notify, mis)) {
538 return -1;
539 }
540
541 /*
542 * Ballooning can mark pages as absent while we're postcopying
543 * that would cause false userfaults.
544 */
545 qemu_balloon_inhibit(true);
546
547 trace_postcopy_ram_enable_notify();
548
549 return 0;
550 }
551
552 /*
553 * Place a host page (from) at (host) atomically
554 * returns 0 on success
555 */
556 int postcopy_place_page(MigrationIncomingState *mis, void *host, void *from,
557 size_t pagesize)
558 {
559 struct uffdio_copy copy_struct;
560
561 copy_struct.dst = (uint64_t)(uintptr_t)host;
562 copy_struct.src = (uint64_t)(uintptr_t)from;
563 copy_struct.len = pagesize;
564 copy_struct.mode = 0;
565
566 /* copy also acks to the kernel waking the stalled thread up
567 * TODO: We can inhibit that ack and only do it if it was requested
568 * which would be slightly cheaper, but we'd have to be careful
569 * of the order of updating our page state.
570 */
571 if (ioctl(mis->userfault_fd, UFFDIO_COPY, &copy_struct)) {
572 int e = errno;
573 error_report("%s: %s copy host: %p from: %p (size: %zd)",
574 __func__, strerror(e), host, from, pagesize);
575
576 return -e;
577 }
578
579 trace_postcopy_place_page(host);
580 return 0;
581 }
582
583 /*
584 * Place a zero page at (host) atomically
585 * returns 0 on success
586 */
587 int postcopy_place_page_zero(MigrationIncomingState *mis, void *host,
588 size_t pagesize)
589 {
590 trace_postcopy_place_page_zero(host);
591
592 if (pagesize == getpagesize()) {
593 struct uffdio_zeropage zero_struct;
594 zero_struct.range.start = (uint64_t)(uintptr_t)host;
595 zero_struct.range.len = getpagesize();
596 zero_struct.mode = 0;
597
598 if (ioctl(mis->userfault_fd, UFFDIO_ZEROPAGE, &zero_struct)) {
599 int e = errno;
600 error_report("%s: %s zero host: %p",
601 __func__, strerror(e), host);
602
603 return -e;
604 }
605 } else {
606 /* The kernel can't use UFFDIO_ZEROPAGE for hugepages */
607 if (!mis->postcopy_tmp_zero_page) {
608 mis->postcopy_tmp_zero_page = mmap(NULL, mis->largest_page_size,
609 PROT_READ | PROT_WRITE,
610 MAP_PRIVATE | MAP_ANONYMOUS,
611 -1, 0);
612 if (mis->postcopy_tmp_zero_page == MAP_FAILED) {
613 int e = errno;
614 mis->postcopy_tmp_zero_page = NULL;
615 error_report("%s: %s mapping large zero page",
616 __func__, strerror(e));
617 return -e;
618 }
619 memset(mis->postcopy_tmp_zero_page, '\0', mis->largest_page_size);
620 }
621 return postcopy_place_page(mis, host, mis->postcopy_tmp_zero_page,
622 pagesize);
623 }
624
625 return 0;
626 }
627
628 /*
629 * Returns a target page of memory that can be mapped at a later point in time
630 * using postcopy_place_page
631 * The same address is used repeatedly, postcopy_place_page just takes the
632 * backing page away.
633 * Returns: Pointer to allocated page
634 *
635 */
636 void *postcopy_get_tmp_page(MigrationIncomingState *mis)
637 {
638 if (!mis->postcopy_tmp_page) {
639 mis->postcopy_tmp_page = mmap(NULL, mis->largest_page_size,
640 PROT_READ | PROT_WRITE, MAP_PRIVATE |
641 MAP_ANONYMOUS, -1, 0);
642 if (mis->postcopy_tmp_page == MAP_FAILED) {
643 mis->postcopy_tmp_page = NULL;
644 error_report("%s: %s", __func__, strerror(errno));
645 return NULL;
646 }
647 }
648
649 return mis->postcopy_tmp_page;
650 }
651
652 #else
653 /* No target OS support, stubs just fail */
654 bool postcopy_ram_supported_by_host(void)
655 {
656 error_report("%s: No OS support", __func__);
657 return false;
658 }
659
660 int postcopy_ram_incoming_init(MigrationIncomingState *mis, size_t ram_pages)
661 {
662 error_report("postcopy_ram_incoming_init: No OS support");
663 return -1;
664 }
665
666 int postcopy_ram_incoming_cleanup(MigrationIncomingState *mis)
667 {
668 assert(0);
669 return -1;
670 }
671
672 int postcopy_ram_prepare_discard(MigrationIncomingState *mis)
673 {
674 assert(0);
675 return -1;
676 }
677
678 int postcopy_ram_enable_notify(MigrationIncomingState *mis)
679 {
680 assert(0);
681 return -1;
682 }
683
684 int postcopy_place_page(MigrationIncomingState *mis, void *host, void *from,
685 size_t pagesize)
686 {
687 assert(0);
688 return -1;
689 }
690
691 int postcopy_place_page_zero(MigrationIncomingState *mis, void *host,
692 size_t pagesize)
693 {
694 assert(0);
695 return -1;
696 }
697
698 void *postcopy_get_tmp_page(MigrationIncomingState *mis)
699 {
700 assert(0);
701 return NULL;
702 }
703
704 #endif
705
706 /* ------------------------------------------------------------------------- */
707
708 /**
709 * postcopy_discard_send_init: Called at the start of each RAMBlock before
710 * asking to discard individual ranges.
711 *
712 * @ms: The current migration state.
713 * @offset: the bitmap offset of the named RAMBlock in the migration
714 * bitmap.
715 * @name: RAMBlock that discards will operate on.
716 *
717 * returns: a new PDS.
718 */
719 PostcopyDiscardState *postcopy_discard_send_init(MigrationState *ms,
720 unsigned long offset,
721 const char *name)
722 {
723 PostcopyDiscardState *res = g_malloc0(sizeof(PostcopyDiscardState));
724
725 if (res) {
726 res->ramblock_name = name;
727 res->offset = offset;
728 }
729
730 return res;
731 }
732
733 /**
734 * postcopy_discard_send_range: Called by the bitmap code for each chunk to
735 * discard. May send a discard message, may just leave it queued to
736 * be sent later.
737 *
738 * @ms: Current migration state.
739 * @pds: Structure initialised by postcopy_discard_send_init().
740 * @start,@length: a range of pages in the migration bitmap in the
741 * RAM block passed to postcopy_discard_send_init() (length=1 is one page)
742 */
743 void postcopy_discard_send_range(MigrationState *ms, PostcopyDiscardState *pds,
744 unsigned long start, unsigned long length)
745 {
746 size_t tp_size = qemu_target_page_size();
747 /* Convert to byte offsets within the RAM block */
748 pds->start_list[pds->cur_entry] = (start - pds->offset) * tp_size;
749 pds->length_list[pds->cur_entry] = length * tp_size;
750 trace_postcopy_discard_send_range(pds->ramblock_name, start, length);
751 pds->cur_entry++;
752 pds->nsentwords++;
753
754 if (pds->cur_entry == MAX_DISCARDS_PER_COMMAND) {
755 /* Full set, ship it! */
756 qemu_savevm_send_postcopy_ram_discard(ms->to_dst_file,
757 pds->ramblock_name,
758 pds->cur_entry,
759 pds->start_list,
760 pds->length_list);
761 pds->nsentcmds++;
762 pds->cur_entry = 0;
763 }
764 }
765
766 /**
767 * postcopy_discard_send_finish: Called at the end of each RAMBlock by the
768 * bitmap code. Sends any outstanding discard messages, frees the PDS
769 *
770 * @ms: Current migration state.
771 * @pds: Structure initialised by postcopy_discard_send_init().
772 */
773 void postcopy_discard_send_finish(MigrationState *ms, PostcopyDiscardState *pds)
774 {
775 /* Anything unsent? */
776 if (pds->cur_entry) {
777 qemu_savevm_send_postcopy_ram_discard(ms->to_dst_file,
778 pds->ramblock_name,
779 pds->cur_entry,
780 pds->start_list,
781 pds->length_list);
782 pds->nsentcmds++;
783 }
784
785 trace_postcopy_discard_send_finish(pds->ramblock_name, pds->nsentwords,
786 pds->nsentcmds);
787
788 g_free(pds);
789 }