1 | /* |
2 | * Virtio Balloon Device |
3 | * |
4 | * Copyright IBM, Corp. 2008 |
5 | * Copyright (C) 2011 Red Hat, Inc. |
6 | * Copyright (C) 2011 Amit Shah <amit.shah@redhat.com> |
7 | * |
8 | * Authors: |
9 | * Anthony Liguori <aliguori@us.ibm.com> |
10 | * |
11 | * This work is licensed under the terms of the GNU GPL, version 2. See |
12 | * the COPYING file in the top-level directory. |
13 | * |
14 | */ |
15 | |
16 | #include "qemu/osdep.h" |
17 | #include "qemu/iov.h" |
18 | #include "qemu/module.h" |
19 | #include "qemu/timer.h" |
20 | #include "hw/virtio/virtio.h" |
21 | #include "hw/mem/pc-dimm.h" |
22 | #include "hw/qdev-properties.h" |
23 | #include "sysemu/balloon.h" |
24 | #include "hw/virtio/virtio-balloon.h" |
25 | #include "exec/address-spaces.h" |
26 | #include "qapi/error.h" |
27 | #include "qapi/qapi-events-misc.h" |
28 | #include "qapi/visitor.h" |
29 | #include "trace.h" |
30 | #include "qemu/error-report.h" |
31 | #include "migration/misc.h" |
32 | |
33 | #include "hw/virtio/virtio-bus.h" |
34 | #include "hw/virtio/virtio-access.h" |
35 | |
36 | #define BALLOON_PAGE_SIZE (1 << VIRTIO_BALLOON_PFN_SHIFT) |
37 | |
38 | typedef struct PartiallyBalloonedPage { |
39 | ram_addr_t base_gpa; |
40 | unsigned long *bitmap; |
41 | } PartiallyBalloonedPage; |
42 | |
43 | static void virtio_balloon_pbp_free(PartiallyBalloonedPage *pbp) |
44 | { |
45 | if (!pbp->bitmap) { |
46 | return; |
47 | } |
48 | g_free(pbp->bitmap); |
49 | pbp->bitmap = NULL; |
50 | } |
51 | |
52 | static void virtio_balloon_pbp_alloc(PartiallyBalloonedPage *pbp, |
53 | ram_addr_t base_gpa, |
54 | long subpages) |
55 | { |
56 | pbp->base_gpa = base_gpa; |
57 | pbp->bitmap = bitmap_new(subpages); |
58 | } |
59 | |
60 | static bool virtio_balloon_pbp_matches(PartiallyBalloonedPage *pbp, |
61 | ram_addr_t base_gpa) |
62 | { |
63 | return pbp->base_gpa == base_gpa; |
64 | } |
65 | |
66 | static void balloon_inflate_page(VirtIOBalloon *balloon, |
67 | MemoryRegion *mr, hwaddr mr_offset, |
68 | PartiallyBalloonedPage *pbp) |
69 | { |
70 | void *addr = memory_region_get_ram_ptr(mr) + mr_offset; |
71 | ram_addr_t rb_offset, rb_aligned_offset, base_gpa; |
72 | RAMBlock *rb; |
73 | size_t rb_page_size; |
74 | int subpages; |
75 | |
76 | /* XXX is there a better way to get to the RAMBlock than via a |
77 | * host address? */ |
78 | rb = qemu_ram_block_from_host(addr, false, &rb_offset); |
79 | rb_page_size = qemu_ram_pagesize(rb); |
80 | |
81 | if (rb_page_size == BALLOON_PAGE_SIZE) { |
82 | /* Easy case */ |
83 | |
84 | ram_block_discard_range(rb, rb_offset, rb_page_size); |
85 | /* We ignore errors from ram_block_discard_range(), because it |
86 | * has already reported them, and failing to discard a balloon |
87 | * page is not fatal */ |
88 | return; |
89 | } |
90 | |
91 | /* Hard case |
92 | * |
93 | * We've put a piece of a larger host page into the balloon - we |
94 | * need to keep track until we have a whole host page to |
95 | * discard |
96 | */ |
97 | warn_report_once( |
98 | "Balloon used with backing page size > 4kiB, this may not be reliable" ); |
99 | |
100 | rb_aligned_offset = QEMU_ALIGN_DOWN(rb_offset, rb_page_size); |
101 | subpages = rb_page_size / BALLOON_PAGE_SIZE; |
102 | base_gpa = memory_region_get_ram_addr(mr) + mr_offset - |
103 | (rb_offset - rb_aligned_offset); |
104 | |
105 | if (pbp->bitmap && !virtio_balloon_pbp_matches(pbp, base_gpa)) { |
106 | /* We've partially ballooned part of a host page, but now |
107 | * we're trying to balloon part of a different one. Too hard, |
108 | * give up on the old partial page */ |
109 | virtio_balloon_pbp_free(pbp); |
110 | } |
111 | |
112 | if (!pbp->bitmap) { |
113 | virtio_balloon_pbp_alloc(pbp, base_gpa, subpages); |
114 | } |
115 | |
116 | set_bit((rb_offset - rb_aligned_offset) / BALLOON_PAGE_SIZE, |
117 | pbp->bitmap); |
118 | |
119 | if (bitmap_full(pbp->bitmap, subpages)) { |
120 | /* We've accumulated a full host page, we can actually discard |
121 | * it now */ |
122 | |
123 | ram_block_discard_range(rb, rb_aligned_offset, rb_page_size); |
124 | /* We ignore errors from ram_block_discard_range(), because it |
125 | * has already reported them, and failing to discard a balloon |
126 | * page is not fatal */ |
127 | virtio_balloon_pbp_free(pbp); |
128 | } |
129 | } |
130 | |
131 | static void balloon_deflate_page(VirtIOBalloon *balloon, |
132 | MemoryRegion *mr, hwaddr mr_offset) |
133 | { |
134 | void *addr = memory_region_get_ram_ptr(mr) + mr_offset; |
135 | ram_addr_t rb_offset; |
136 | RAMBlock *rb; |
137 | size_t rb_page_size; |
138 | void *host_addr; |
139 | int ret; |
140 | |
141 | /* XXX is there a better way to get to the RAMBlock than via a |
142 | * host address? */ |
143 | rb = qemu_ram_block_from_host(addr, false, &rb_offset); |
144 | rb_page_size = qemu_ram_pagesize(rb); |
145 | |
146 | host_addr = (void *)((uintptr_t)addr & ~(rb_page_size - 1)); |
147 | |
148 | /* When a page is deflated, we hint the whole host page it lives |
149 | * on, since we can't do anything smaller */ |
150 | ret = qemu_madvise(host_addr, rb_page_size, QEMU_MADV_WILLNEED); |
151 | if (ret != 0) { |
152 | warn_report("Couldn't MADV_WILLNEED on balloon deflate: %s" , |
153 | strerror(errno)); |
154 | /* Otherwise ignore, failing to page hint shouldn't be fatal */ |
155 | } |
156 | } |
157 | |
158 | static const char *balloon_stat_names[] = { |
159 | [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in" , |
160 | [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out" , |
161 | [VIRTIO_BALLOON_S_MAJFLT] = "stat-major-faults" , |
162 | [VIRTIO_BALLOON_S_MINFLT] = "stat-minor-faults" , |
163 | [VIRTIO_BALLOON_S_MEMFREE] = "stat-free-memory" , |
164 | [VIRTIO_BALLOON_S_MEMTOT] = "stat-total-memory" , |
165 | [VIRTIO_BALLOON_S_AVAIL] = "stat-available-memory" , |
166 | [VIRTIO_BALLOON_S_CACHES] = "stat-disk-caches" , |
167 | [VIRTIO_BALLOON_S_HTLB_PGALLOC] = "stat-htlb-pgalloc" , |
168 | [VIRTIO_BALLOON_S_HTLB_PGFAIL] = "stat-htlb-pgfail" , |
169 | [VIRTIO_BALLOON_S_NR] = NULL |
170 | }; |
171 | |
172 | /* |
173 | * reset_stats - Mark all items in the stats array as unset |
174 | * |
175 | * This function needs to be called at device initialization and before |
176 | * updating to a set of newly-generated stats. This will ensure that no |
177 | * stale values stick around in case the guest reports a subset of the supported |
178 | * statistics. |
179 | */ |
180 | static inline void reset_stats(VirtIOBalloon *dev) |
181 | { |
182 | int i; |
183 | for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1); |
184 | } |
185 | |
186 | static bool balloon_stats_supported(const VirtIOBalloon *s) |
187 | { |
188 | VirtIODevice *vdev = VIRTIO_DEVICE(s); |
189 | return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ); |
190 | } |
191 | |
192 | static bool balloon_stats_enabled(const VirtIOBalloon *s) |
193 | { |
194 | return s->stats_poll_interval > 0; |
195 | } |
196 | |
197 | static void balloon_stats_destroy_timer(VirtIOBalloon *s) |
198 | { |
199 | if (balloon_stats_enabled(s)) { |
200 | timer_del(s->stats_timer); |
201 | timer_free(s->stats_timer); |
202 | s->stats_timer = NULL; |
203 | s->stats_poll_interval = 0; |
204 | } |
205 | } |
206 | |
207 | static void balloon_stats_change_timer(VirtIOBalloon *s, int64_t secs) |
208 | { |
209 | timer_mod(s->stats_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + secs * 1000); |
210 | } |
211 | |
212 | static void balloon_stats_poll_cb(void *opaque) |
213 | { |
214 | VirtIOBalloon *s = opaque; |
215 | VirtIODevice *vdev = VIRTIO_DEVICE(s); |
216 | |
217 | if (s->stats_vq_elem == NULL || !balloon_stats_supported(s)) { |
218 | /* re-schedule */ |
219 | balloon_stats_change_timer(s, s->stats_poll_interval); |
220 | return; |
221 | } |
222 | |
223 | virtqueue_push(s->svq, s->stats_vq_elem, s->stats_vq_offset); |
224 | virtio_notify(vdev, s->svq); |
225 | g_free(s->stats_vq_elem); |
226 | s->stats_vq_elem = NULL; |
227 | } |
228 | |
229 | static void balloon_stats_get_all(Object *obj, Visitor *v, const char *name, |
230 | void *opaque, Error **errp) |
231 | { |
232 | Error *err = NULL; |
233 | VirtIOBalloon *s = opaque; |
234 | int i; |
235 | |
236 | visit_start_struct(v, name, NULL, 0, &err); |
237 | if (err) { |
238 | goto out; |
239 | } |
240 | visit_type_int(v, "last-update" , &s->stats_last_update, &err); |
241 | if (err) { |
242 | goto out_end; |
243 | } |
244 | |
245 | visit_start_struct(v, "stats" , NULL, 0, &err); |
246 | if (err) { |
247 | goto out_end; |
248 | } |
249 | for (i = 0; i < VIRTIO_BALLOON_S_NR; i++) { |
250 | visit_type_uint64(v, balloon_stat_names[i], &s->stats[i], &err); |
251 | if (err) { |
252 | goto out_nested; |
253 | } |
254 | } |
255 | visit_check_struct(v, &err); |
256 | out_nested: |
257 | visit_end_struct(v, NULL); |
258 | |
259 | if (!err) { |
260 | visit_check_struct(v, &err); |
261 | } |
262 | out_end: |
263 | visit_end_struct(v, NULL); |
264 | out: |
265 | error_propagate(errp, err); |
266 | } |
267 | |
268 | static void balloon_stats_get_poll_interval(Object *obj, Visitor *v, |
269 | const char *name, void *opaque, |
270 | Error **errp) |
271 | { |
272 | VirtIOBalloon *s = opaque; |
273 | visit_type_int(v, name, &s->stats_poll_interval, errp); |
274 | } |
275 | |
276 | static void balloon_stats_set_poll_interval(Object *obj, Visitor *v, |
277 | const char *name, void *opaque, |
278 | Error **errp) |
279 | { |
280 | VirtIOBalloon *s = opaque; |
281 | Error *local_err = NULL; |
282 | int64_t value; |
283 | |
284 | visit_type_int(v, name, &value, &local_err); |
285 | if (local_err) { |
286 | error_propagate(errp, local_err); |
287 | return; |
288 | } |
289 | |
290 | if (value < 0) { |
291 | error_setg(errp, "timer value must be greater than zero" ); |
292 | return; |
293 | } |
294 | |
295 | if (value > UINT32_MAX) { |
296 | error_setg(errp, "timer value is too big" ); |
297 | return; |
298 | } |
299 | |
300 | if (value == s->stats_poll_interval) { |
301 | return; |
302 | } |
303 | |
304 | if (value == 0) { |
305 | /* timer=0 disables the timer */ |
306 | balloon_stats_destroy_timer(s); |
307 | return; |
308 | } |
309 | |
310 | if (balloon_stats_enabled(s)) { |
311 | /* timer interval change */ |
312 | s->stats_poll_interval = value; |
313 | balloon_stats_change_timer(s, value); |
314 | return; |
315 | } |
316 | |
317 | /* create a new timer */ |
318 | g_assert(s->stats_timer == NULL); |
319 | s->stats_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, balloon_stats_poll_cb, s); |
320 | s->stats_poll_interval = value; |
321 | balloon_stats_change_timer(s, 0); |
322 | } |
323 | |
324 | static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq) |
325 | { |
326 | VirtIOBalloon *s = VIRTIO_BALLOON(vdev); |
327 | VirtQueueElement *elem; |
328 | MemoryRegionSection section; |
329 | |
330 | for (;;) { |
331 | PartiallyBalloonedPage pbp = {}; |
332 | size_t offset = 0; |
333 | uint32_t pfn; |
334 | |
335 | elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); |
336 | if (!elem) { |
337 | break; |
338 | } |
339 | |
340 | while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) { |
341 | unsigned int p = virtio_ldl_p(vdev, &pfn); |
342 | hwaddr pa; |
343 | |
344 | pa = (hwaddr) p << VIRTIO_BALLOON_PFN_SHIFT; |
345 | offset += 4; |
346 | |
347 | section = memory_region_find(get_system_memory(), pa, |
348 | BALLOON_PAGE_SIZE); |
349 | if (!section.mr) { |
350 | trace_virtio_balloon_bad_addr(pa); |
351 | continue; |
352 | } |
353 | if (!memory_region_is_ram(section.mr) || |
354 | memory_region_is_rom(section.mr) || |
355 | memory_region_is_romd(section.mr)) { |
356 | trace_virtio_balloon_bad_addr(pa); |
357 | memory_region_unref(section.mr); |
358 | continue; |
359 | } |
360 | |
361 | trace_virtio_balloon_handle_output(memory_region_name(section.mr), |
362 | pa); |
363 | if (!qemu_balloon_is_inhibited()) { |
364 | if (vq == s->ivq) { |
365 | balloon_inflate_page(s, section.mr, |
366 | section.offset_within_region, &pbp); |
367 | } else if (vq == s->dvq) { |
368 | balloon_deflate_page(s, section.mr, section.offset_within_region); |
369 | } else { |
370 | g_assert_not_reached(); |
371 | } |
372 | } |
373 | memory_region_unref(section.mr); |
374 | } |
375 | |
376 | virtqueue_push(vq, elem, offset); |
377 | virtio_notify(vdev, vq); |
378 | g_free(elem); |
379 | virtio_balloon_pbp_free(&pbp); |
380 | } |
381 | } |
382 | |
383 | static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq) |
384 | { |
385 | VirtIOBalloon *s = VIRTIO_BALLOON(vdev); |
386 | VirtQueueElement *elem; |
387 | VirtIOBalloonStat stat; |
388 | size_t offset = 0; |
389 | qemu_timeval tv; |
390 | |
391 | elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); |
392 | if (!elem) { |
393 | goto out; |
394 | } |
395 | |
396 | if (s->stats_vq_elem != NULL) { |
397 | /* This should never happen if the driver follows the spec. */ |
398 | virtqueue_push(vq, s->stats_vq_elem, 0); |
399 | virtio_notify(vdev, vq); |
400 | g_free(s->stats_vq_elem); |
401 | } |
402 | |
403 | s->stats_vq_elem = elem; |
404 | |
405 | /* Initialize the stats to get rid of any stale values. This is only |
406 | * needed to handle the case where a guest supports fewer stats than it |
407 | * used to (ie. it has booted into an old kernel). |
408 | */ |
409 | reset_stats(s); |
410 | |
411 | while (iov_to_buf(elem->out_sg, elem->out_num, offset, &stat, sizeof(stat)) |
412 | == sizeof(stat)) { |
413 | uint16_t tag = virtio_tswap16(vdev, stat.tag); |
414 | uint64_t val = virtio_tswap64(vdev, stat.val); |
415 | |
416 | offset += sizeof(stat); |
417 | if (tag < VIRTIO_BALLOON_S_NR) |
418 | s->stats[tag] = val; |
419 | } |
420 | s->stats_vq_offset = offset; |
421 | |
422 | if (qemu_gettimeofday(&tv) < 0) { |
423 | warn_report("%s: failed to get time of day" , __func__); |
424 | goto out; |
425 | } |
426 | |
427 | s->stats_last_update = tv.tv_sec; |
428 | |
429 | out: |
430 | if (balloon_stats_enabled(s)) { |
431 | balloon_stats_change_timer(s, s->stats_poll_interval); |
432 | } |
433 | } |
434 | |
435 | static void virtio_balloon_handle_free_page_vq(VirtIODevice *vdev, |
436 | VirtQueue *vq) |
437 | { |
438 | VirtIOBalloon *s = VIRTIO_BALLOON(vdev); |
439 | qemu_bh_schedule(s->free_page_bh); |
440 | } |
441 | |
442 | static bool get_free_page_hints(VirtIOBalloon *dev) |
443 | { |
444 | VirtQueueElement *elem; |
445 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
446 | VirtQueue *vq = dev->free_page_vq; |
447 | bool ret = true; |
448 | |
449 | while (dev->block_iothread) { |
450 | qemu_cond_wait(&dev->free_page_cond, &dev->free_page_lock); |
451 | } |
452 | |
453 | elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); |
454 | if (!elem) { |
455 | return false; |
456 | } |
457 | |
458 | if (elem->out_num) { |
459 | uint32_t id; |
460 | size_t size = iov_to_buf(elem->out_sg, elem->out_num, 0, |
461 | &id, sizeof(id)); |
462 | |
463 | virtio_tswap32s(vdev, &id); |
464 | if (unlikely(size != sizeof(id))) { |
465 | virtio_error(vdev, "received an incorrect cmd id" ); |
466 | ret = false; |
467 | goto out; |
468 | } |
469 | if (id == dev->free_page_report_cmd_id) { |
470 | dev->free_page_report_status = FREE_PAGE_REPORT_S_START; |
471 | } else { |
472 | /* |
473 | * Stop the optimization only when it has started. This |
474 | * avoids a stale stop sign for the previous command. |
475 | */ |
476 | if (dev->free_page_report_status == FREE_PAGE_REPORT_S_START) { |
477 | dev->free_page_report_status = FREE_PAGE_REPORT_S_STOP; |
478 | } |
479 | } |
480 | } |
481 | |
482 | if (elem->in_num) { |
483 | if (dev->free_page_report_status == FREE_PAGE_REPORT_S_START) { |
484 | qemu_guest_free_page_hint(elem->in_sg[0].iov_base, |
485 | elem->in_sg[0].iov_len); |
486 | } |
487 | } |
488 | |
489 | out: |
490 | virtqueue_push(vq, elem, 1); |
491 | g_free(elem); |
492 | return ret; |
493 | } |
494 | |
495 | static void virtio_ballloon_get_free_page_hints(void *opaque) |
496 | { |
497 | VirtIOBalloon *dev = opaque; |
498 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
499 | VirtQueue *vq = dev->free_page_vq; |
500 | bool continue_to_get_hints; |
501 | |
502 | do { |
503 | qemu_mutex_lock(&dev->free_page_lock); |
504 | virtio_queue_set_notification(vq, 0); |
505 | continue_to_get_hints = get_free_page_hints(dev); |
506 | qemu_mutex_unlock(&dev->free_page_lock); |
507 | virtio_notify(vdev, vq); |
508 | /* |
509 | * Start to poll the vq once the reporting started. Otherwise, continue |
510 | * only when there are entries on the vq, which need to be given back. |
511 | */ |
512 | } while (continue_to_get_hints || |
513 | dev->free_page_report_status == FREE_PAGE_REPORT_S_START); |
514 | virtio_queue_set_notification(vq, 1); |
515 | } |
516 | |
517 | static bool virtio_balloon_free_page_support(void *opaque) |
518 | { |
519 | VirtIOBalloon *s = opaque; |
520 | VirtIODevice *vdev = VIRTIO_DEVICE(s); |
521 | |
522 | return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT); |
523 | } |
524 | |
525 | static void virtio_balloon_free_page_start(VirtIOBalloon *s) |
526 | { |
527 | VirtIODevice *vdev = VIRTIO_DEVICE(s); |
528 | |
529 | /* For the stop and copy phase, we don't need to start the optimization */ |
530 | if (!vdev->vm_running) { |
531 | return; |
532 | } |
533 | |
534 | if (s->free_page_report_cmd_id == UINT_MAX) { |
535 | s->free_page_report_cmd_id = |
536 | VIRTIO_BALLOON_FREE_PAGE_REPORT_CMD_ID_MIN; |
537 | } else { |
538 | s->free_page_report_cmd_id++; |
539 | } |
540 | |
541 | s->free_page_report_status = FREE_PAGE_REPORT_S_REQUESTED; |
542 | virtio_notify_config(vdev); |
543 | } |
544 | |
545 | static void virtio_balloon_free_page_stop(VirtIOBalloon *s) |
546 | { |
547 | VirtIODevice *vdev = VIRTIO_DEVICE(s); |
548 | |
549 | if (s->free_page_report_status != FREE_PAGE_REPORT_S_STOP) { |
550 | /* |
551 | * The lock also guarantees us that the |
552 | * virtio_ballloon_get_free_page_hints exits after the |
553 | * free_page_report_status is set to S_STOP. |
554 | */ |
555 | qemu_mutex_lock(&s->free_page_lock); |
556 | /* |
557 | * The guest hasn't done the reporting, so host sends a notification |
558 | * to the guest to actively stop the reporting. |
559 | */ |
560 | s->free_page_report_status = FREE_PAGE_REPORT_S_STOP; |
561 | qemu_mutex_unlock(&s->free_page_lock); |
562 | virtio_notify_config(vdev); |
563 | } |
564 | } |
565 | |
566 | static void virtio_balloon_free_page_done(VirtIOBalloon *s) |
567 | { |
568 | VirtIODevice *vdev = VIRTIO_DEVICE(s); |
569 | |
570 | s->free_page_report_status = FREE_PAGE_REPORT_S_DONE; |
571 | virtio_notify_config(vdev); |
572 | } |
573 | |
574 | static int |
575 | virtio_balloon_free_page_report_notify(NotifierWithReturn *n, void *data) |
576 | { |
577 | VirtIOBalloon *dev = container_of(n, VirtIOBalloon, |
578 | free_page_report_notify); |
579 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
580 | PrecopyNotifyData *pnd = data; |
581 | |
582 | if (!virtio_balloon_free_page_support(dev)) { |
583 | /* |
584 | * This is an optimization provided to migration, so just return 0 to |
585 | * have the normal migration process not affected when this feature is |
586 | * not supported. |
587 | */ |
588 | return 0; |
589 | } |
590 | |
591 | switch (pnd->reason) { |
592 | case PRECOPY_NOTIFY_SETUP: |
593 | precopy_enable_free_page_optimization(); |
594 | break; |
595 | case PRECOPY_NOTIFY_COMPLETE: |
596 | case PRECOPY_NOTIFY_CLEANUP: |
597 | case PRECOPY_NOTIFY_BEFORE_BITMAP_SYNC: |
598 | virtio_balloon_free_page_stop(dev); |
599 | break; |
600 | case PRECOPY_NOTIFY_AFTER_BITMAP_SYNC: |
601 | if (vdev->vm_running) { |
602 | virtio_balloon_free_page_start(dev); |
603 | } else { |
604 | virtio_balloon_free_page_done(dev); |
605 | } |
606 | break; |
607 | default: |
608 | virtio_error(vdev, "%s: %d reason unknown" , __func__, pnd->reason); |
609 | } |
610 | |
611 | return 0; |
612 | } |
613 | |
614 | static size_t virtio_balloon_config_size(VirtIOBalloon *s) |
615 | { |
616 | uint64_t features = s->host_features; |
617 | |
618 | if (s->qemu_4_0_config_size) { |
619 | return sizeof(struct virtio_balloon_config); |
620 | } |
621 | if (virtio_has_feature(features, VIRTIO_BALLOON_F_PAGE_POISON)) { |
622 | return sizeof(struct virtio_balloon_config); |
623 | } |
624 | if (virtio_has_feature(features, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { |
625 | return offsetof(struct virtio_balloon_config, poison_val); |
626 | } |
627 | return offsetof(struct virtio_balloon_config, free_page_report_cmd_id); |
628 | } |
629 | |
630 | static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data) |
631 | { |
632 | VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); |
633 | struct virtio_balloon_config config = {}; |
634 | |
635 | config.num_pages = cpu_to_le32(dev->num_pages); |
636 | config.actual = cpu_to_le32(dev->actual); |
637 | |
638 | if (dev->free_page_report_status == FREE_PAGE_REPORT_S_REQUESTED) { |
639 | config.free_page_report_cmd_id = |
640 | cpu_to_le32(dev->free_page_report_cmd_id); |
641 | } else if (dev->free_page_report_status == FREE_PAGE_REPORT_S_STOP) { |
642 | config.free_page_report_cmd_id = |
643 | cpu_to_le32(VIRTIO_BALLOON_CMD_ID_STOP); |
644 | } else if (dev->free_page_report_status == FREE_PAGE_REPORT_S_DONE) { |
645 | config.free_page_report_cmd_id = |
646 | cpu_to_le32(VIRTIO_BALLOON_CMD_ID_DONE); |
647 | } |
648 | |
649 | trace_virtio_balloon_get_config(config.num_pages, config.actual); |
650 | memcpy(config_data, &config, virtio_balloon_config_size(dev)); |
651 | } |
652 | |
653 | static int build_dimm_list(Object *obj, void *opaque) |
654 | { |
655 | GSList **list = opaque; |
656 | |
657 | if (object_dynamic_cast(obj, TYPE_PC_DIMM)) { |
658 | DeviceState *dev = DEVICE(obj); |
659 | if (dev->realized) { /* only realized DIMMs matter */ |
660 | *list = g_slist_prepend(*list, dev); |
661 | } |
662 | } |
663 | |
664 | object_child_foreach(obj, build_dimm_list, opaque); |
665 | return 0; |
666 | } |
667 | |
668 | static ram_addr_t get_current_ram_size(void) |
669 | { |
670 | GSList *list = NULL, *item; |
671 | ram_addr_t size = ram_size; |
672 | |
673 | build_dimm_list(qdev_get_machine(), &list); |
674 | for (item = list; item; item = g_slist_next(item)) { |
675 | Object *obj = OBJECT(item->data); |
676 | if (!strcmp(object_get_typename(obj), TYPE_PC_DIMM)) { |
677 | size += object_property_get_int(obj, PC_DIMM_SIZE_PROP, |
678 | &error_abort); |
679 | } |
680 | } |
681 | g_slist_free(list); |
682 | |
683 | return size; |
684 | } |
685 | |
686 | static void virtio_balloon_set_config(VirtIODevice *vdev, |
687 | const uint8_t *config_data) |
688 | { |
689 | VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); |
690 | struct virtio_balloon_config config; |
691 | uint32_t oldactual = dev->actual; |
692 | ram_addr_t vm_ram_size = get_current_ram_size(); |
693 | |
694 | memcpy(&config, config_data, virtio_balloon_config_size(dev)); |
695 | dev->actual = le32_to_cpu(config.actual); |
696 | if (dev->actual != oldactual) { |
697 | qapi_event_send_balloon_change(vm_ram_size - |
698 | ((ram_addr_t) dev->actual << VIRTIO_BALLOON_PFN_SHIFT)); |
699 | } |
700 | trace_virtio_balloon_set_config(dev->actual, oldactual); |
701 | } |
702 | |
703 | static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f, |
704 | Error **errp) |
705 | { |
706 | VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); |
707 | f |= dev->host_features; |
708 | virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ); |
709 | |
710 | return f; |
711 | } |
712 | |
713 | static void virtio_balloon_stat(void *opaque, BalloonInfo *info) |
714 | { |
715 | VirtIOBalloon *dev = opaque; |
716 | info->actual = get_current_ram_size() - ((uint64_t) dev->actual << |
717 | VIRTIO_BALLOON_PFN_SHIFT); |
718 | } |
719 | |
720 | static void virtio_balloon_to_target(void *opaque, ram_addr_t target) |
721 | { |
722 | VirtIOBalloon *dev = VIRTIO_BALLOON(opaque); |
723 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
724 | ram_addr_t vm_ram_size = get_current_ram_size(); |
725 | |
726 | if (target > vm_ram_size) { |
727 | target = vm_ram_size; |
728 | } |
729 | if (target) { |
730 | dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT; |
731 | virtio_notify_config(vdev); |
732 | } |
733 | trace_virtio_balloon_to_target(target, dev->num_pages); |
734 | } |
735 | |
736 | static int virtio_balloon_post_load_device(void *opaque, int version_id) |
737 | { |
738 | VirtIOBalloon *s = VIRTIO_BALLOON(opaque); |
739 | |
740 | if (balloon_stats_enabled(s)) { |
741 | balloon_stats_change_timer(s, s->stats_poll_interval); |
742 | } |
743 | return 0; |
744 | } |
745 | |
746 | static const VMStateDescription vmstate_virtio_balloon_free_page_report = { |
747 | .name = "virtio-balloon-device/free-page-report" , |
748 | .version_id = 1, |
749 | .minimum_version_id = 1, |
750 | .needed = virtio_balloon_free_page_support, |
751 | .fields = (VMStateField[]) { |
752 | VMSTATE_UINT32(free_page_report_cmd_id, VirtIOBalloon), |
753 | VMSTATE_UINT32(free_page_report_status, VirtIOBalloon), |
754 | VMSTATE_END_OF_LIST() |
755 | } |
756 | }; |
757 | |
758 | static const VMStateDescription vmstate_virtio_balloon_device = { |
759 | .name = "virtio-balloon-device" , |
760 | .version_id = 1, |
761 | .minimum_version_id = 1, |
762 | .post_load = virtio_balloon_post_load_device, |
763 | .fields = (VMStateField[]) { |
764 | VMSTATE_UINT32(num_pages, VirtIOBalloon), |
765 | VMSTATE_UINT32(actual, VirtIOBalloon), |
766 | VMSTATE_END_OF_LIST() |
767 | }, |
768 | .subsections = (const VMStateDescription * []) { |
769 | &vmstate_virtio_balloon_free_page_report, |
770 | NULL |
771 | } |
772 | }; |
773 | |
774 | static void virtio_balloon_device_realize(DeviceState *dev, Error **errp) |
775 | { |
776 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
777 | VirtIOBalloon *s = VIRTIO_BALLOON(dev); |
778 | int ret; |
779 | |
780 | virtio_init(vdev, "virtio-balloon" , VIRTIO_ID_BALLOON, |
781 | virtio_balloon_config_size(s)); |
782 | |
783 | ret = qemu_add_balloon_handler(virtio_balloon_to_target, |
784 | virtio_balloon_stat, s); |
785 | |
786 | if (ret < 0) { |
787 | error_setg(errp, "Only one balloon device is supported" ); |
788 | virtio_cleanup(vdev); |
789 | return; |
790 | } |
791 | |
792 | s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output); |
793 | s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output); |
794 | s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats); |
795 | |
796 | if (virtio_has_feature(s->host_features, |
797 | VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { |
798 | s->free_page_vq = virtio_add_queue(vdev, VIRTQUEUE_MAX_SIZE, |
799 | virtio_balloon_handle_free_page_vq); |
800 | s->free_page_report_status = FREE_PAGE_REPORT_S_STOP; |
801 | s->free_page_report_cmd_id = |
802 | VIRTIO_BALLOON_FREE_PAGE_REPORT_CMD_ID_MIN; |
803 | s->free_page_report_notify.notify = |
804 | virtio_balloon_free_page_report_notify; |
805 | precopy_add_notifier(&s->free_page_report_notify); |
806 | if (s->iothread) { |
807 | object_ref(OBJECT(s->iothread)); |
808 | s->free_page_bh = aio_bh_new(iothread_get_aio_context(s->iothread), |
809 | virtio_ballloon_get_free_page_hints, s); |
810 | qemu_mutex_init(&s->free_page_lock); |
811 | qemu_cond_init(&s->free_page_cond); |
812 | s->block_iothread = false; |
813 | } else { |
814 | /* Simply disable this feature if the iothread wasn't created. */ |
815 | s->host_features &= ~(1 << VIRTIO_BALLOON_F_FREE_PAGE_HINT); |
816 | virtio_error(vdev, "iothread is missing" ); |
817 | } |
818 | } |
819 | reset_stats(s); |
820 | } |
821 | |
822 | static void virtio_balloon_device_unrealize(DeviceState *dev, Error **errp) |
823 | { |
824 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
825 | VirtIOBalloon *s = VIRTIO_BALLOON(dev); |
826 | |
827 | if (virtio_balloon_free_page_support(s)) { |
828 | qemu_bh_delete(s->free_page_bh); |
829 | virtio_balloon_free_page_stop(s); |
830 | precopy_remove_notifier(&s->free_page_report_notify); |
831 | } |
832 | balloon_stats_destroy_timer(s); |
833 | qemu_remove_balloon_handler(s); |
834 | virtio_cleanup(vdev); |
835 | } |
836 | |
837 | static void virtio_balloon_device_reset(VirtIODevice *vdev) |
838 | { |
839 | VirtIOBalloon *s = VIRTIO_BALLOON(vdev); |
840 | |
841 | if (virtio_balloon_free_page_support(s)) { |
842 | virtio_balloon_free_page_stop(s); |
843 | } |
844 | |
845 | if (s->stats_vq_elem != NULL) { |
846 | virtqueue_unpop(s->svq, s->stats_vq_elem, 0); |
847 | g_free(s->stats_vq_elem); |
848 | s->stats_vq_elem = NULL; |
849 | } |
850 | } |
851 | |
852 | static void virtio_balloon_set_status(VirtIODevice *vdev, uint8_t status) |
853 | { |
854 | VirtIOBalloon *s = VIRTIO_BALLOON(vdev); |
855 | |
856 | if (!s->stats_vq_elem && vdev->vm_running && |
857 | (status & VIRTIO_CONFIG_S_DRIVER_OK) && virtqueue_rewind(s->svq, 1)) { |
858 | /* poll stats queue for the element we have discarded when the VM |
859 | * was stopped */ |
860 | virtio_balloon_receive_stats(vdev, s->svq); |
861 | } |
862 | |
863 | if (virtio_balloon_free_page_support(s)) { |
864 | /* |
865 | * The VM is woken up and the iothread was blocked, so signal it to |
866 | * continue. |
867 | */ |
868 | if (vdev->vm_running && s->block_iothread) { |
869 | qemu_mutex_lock(&s->free_page_lock); |
870 | s->block_iothread = false; |
871 | qemu_cond_signal(&s->free_page_cond); |
872 | qemu_mutex_unlock(&s->free_page_lock); |
873 | } |
874 | |
875 | /* The VM is stopped, block the iothread. */ |
876 | if (!vdev->vm_running) { |
877 | qemu_mutex_lock(&s->free_page_lock); |
878 | s->block_iothread = true; |
879 | qemu_mutex_unlock(&s->free_page_lock); |
880 | } |
881 | } |
882 | } |
883 | |
884 | static void virtio_balloon_instance_init(Object *obj) |
885 | { |
886 | VirtIOBalloon *s = VIRTIO_BALLOON(obj); |
887 | |
888 | object_property_add(obj, "guest-stats" , "guest statistics" , |
889 | balloon_stats_get_all, NULL, NULL, s, NULL); |
890 | |
891 | object_property_add(obj, "guest-stats-polling-interval" , "int" , |
892 | balloon_stats_get_poll_interval, |
893 | balloon_stats_set_poll_interval, |
894 | NULL, s, NULL); |
895 | } |
896 | |
897 | static const VMStateDescription vmstate_virtio_balloon = { |
898 | .name = "virtio-balloon" , |
899 | .minimum_version_id = 1, |
900 | .version_id = 1, |
901 | .fields = (VMStateField[]) { |
902 | VMSTATE_VIRTIO_DEVICE, |
903 | VMSTATE_END_OF_LIST() |
904 | }, |
905 | }; |
906 | |
907 | static Property virtio_balloon_properties[] = { |
908 | DEFINE_PROP_BIT("deflate-on-oom" , VirtIOBalloon, host_features, |
909 | VIRTIO_BALLOON_F_DEFLATE_ON_OOM, false), |
910 | DEFINE_PROP_BIT("free-page-hint" , VirtIOBalloon, host_features, |
911 | VIRTIO_BALLOON_F_FREE_PAGE_HINT, false), |
912 | /* QEMU 4.0 accidentally changed the config size even when free-page-hint |
913 | * is disabled, resulting in QEMU 3.1 migration incompatibility. This |
914 | * property retains this quirk for QEMU 4.1 machine types. |
915 | */ |
916 | DEFINE_PROP_BOOL("qemu-4-0-config-size" , VirtIOBalloon, |
917 | qemu_4_0_config_size, false), |
918 | DEFINE_PROP_LINK("iothread" , VirtIOBalloon, iothread, TYPE_IOTHREAD, |
919 | IOThread *), |
920 | DEFINE_PROP_END_OF_LIST(), |
921 | }; |
922 | |
923 | static void virtio_balloon_class_init(ObjectClass *klass, void *data) |
924 | { |
925 | DeviceClass *dc = DEVICE_CLASS(klass); |
926 | VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); |
927 | |
928 | dc->props = virtio_balloon_properties; |
929 | dc->vmsd = &vmstate_virtio_balloon; |
930 | set_bit(DEVICE_CATEGORY_MISC, dc->categories); |
931 | vdc->realize = virtio_balloon_device_realize; |
932 | vdc->unrealize = virtio_balloon_device_unrealize; |
933 | vdc->reset = virtio_balloon_device_reset; |
934 | vdc->get_config = virtio_balloon_get_config; |
935 | vdc->set_config = virtio_balloon_set_config; |
936 | vdc->get_features = virtio_balloon_get_features; |
937 | vdc->set_status = virtio_balloon_set_status; |
938 | vdc->vmsd = &vmstate_virtio_balloon_device; |
939 | } |
940 | |
941 | static const TypeInfo virtio_balloon_info = { |
942 | .name = TYPE_VIRTIO_BALLOON, |
943 | .parent = TYPE_VIRTIO_DEVICE, |
944 | .instance_size = sizeof(VirtIOBalloon), |
945 | .instance_init = virtio_balloon_instance_init, |
946 | .class_init = virtio_balloon_class_init, |
947 | }; |
948 | |
949 | static void virtio_register_types(void) |
950 | { |
951 | type_register_static(&virtio_balloon_info); |
952 | } |
953 | |
954 | type_init(virtio_register_types) |
955 | |