]>
Commit | Line | Data |
---|---|---|
967f97fa AL |
1 | /* |
2 | * Virtio Support | |
3 | * | |
4 | * Copyright IBM, Corp. 2007 | |
5 | * | |
6 | * Authors: | |
7 | * Anthony Liguori <aliguori@us.ibm.com> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
10 | * the COPYING file in the top-level directory. | |
11 | * | |
12 | */ | |
13 | ||
9b8bfe21 | 14 | #include "qemu/osdep.h" |
da34e65c | 15 | #include "qapi/error.h" |
4771d756 PB |
16 | #include "qemu-common.h" |
17 | #include "cpu.h" | |
64979a4d | 18 | #include "trace.h" |
fdfba1a2 | 19 | #include "exec/address-spaces.h" |
1de7afc9 | 20 | #include "qemu/error-report.h" |
0d09e41a | 21 | #include "hw/virtio/virtio.h" |
1de7afc9 | 22 | #include "qemu/atomic.h" |
0d09e41a | 23 | #include "hw/virtio/virtio-bus.h" |
6b321a3d | 24 | #include "migration/migration.h" |
cee3ca00 | 25 | #include "hw/virtio/virtio-access.h" |
8607f5c3 | 26 | #include "sysemu/dma.h" |
967f97fa | 27 | |
6ce69d1c PM |
28 | /* |
29 | * The alignment to use between consumer and producer parts of vring. | |
30 | * x86 pagesize again. This is the default, used by transports like PCI | |
31 | * which don't provide a means for the guest to tell the host the alignment. | |
32 | */ | |
f46f15bc AL |
33 | #define VIRTIO_PCI_VRING_ALIGN 4096 |
34 | ||
967f97fa AL |
35 | typedef struct VRingDesc |
36 | { | |
37 | uint64_t addr; | |
38 | uint32_t len; | |
39 | uint16_t flags; | |
40 | uint16_t next; | |
41 | } VRingDesc; | |
42 | ||
43 | typedef struct VRingAvail | |
44 | { | |
45 | uint16_t flags; | |
46 | uint16_t idx; | |
47 | uint16_t ring[0]; | |
48 | } VRingAvail; | |
49 | ||
50 | typedef struct VRingUsedElem | |
51 | { | |
52 | uint32_t id; | |
53 | uint32_t len; | |
54 | } VRingUsedElem; | |
55 | ||
56 | typedef struct VRingUsed | |
57 | { | |
58 | uint16_t flags; | |
59 | uint16_t idx; | |
60 | VRingUsedElem ring[0]; | |
61 | } VRingUsed; | |
62 | ||
c611c764 PB |
63 | typedef struct VRingMemoryRegionCaches { |
64 | struct rcu_head rcu; | |
65 | MemoryRegionCache desc; | |
66 | MemoryRegionCache avail; | |
67 | MemoryRegionCache used; | |
68 | } VRingMemoryRegionCaches; | |
69 | ||
967f97fa AL |
70 | typedef struct VRing |
71 | { | |
72 | unsigned int num; | |
46c5d082 | 73 | unsigned int num_default; |
6ce69d1c | 74 | unsigned int align; |
a8170e5e AK |
75 | hwaddr desc; |
76 | hwaddr avail; | |
77 | hwaddr used; | |
c611c764 | 78 | VRingMemoryRegionCaches *caches; |
967f97fa AL |
79 | } VRing; |
80 | ||
81 | struct VirtQueue | |
82 | { | |
83 | VRing vring; | |
be1fea9b VM |
84 | |
85 | /* Next head to pop */ | |
967f97fa | 86 | uint16_t last_avail_idx; |
b796fcd1 | 87 | |
be1fea9b VM |
88 | /* Last avail_idx read from VQ. */ |
89 | uint16_t shadow_avail_idx; | |
90 | ||
b796fcd1 VM |
91 | uint16_t used_idx; |
92 | ||
bcbabae8 MT |
93 | /* Last used index value we have signalled on */ |
94 | uint16_t signalled_used; | |
95 | ||
96 | /* Last used index value we have signalled on */ | |
97 | bool signalled_used_valid; | |
98 | ||
332fa82d SH |
99 | /* Notification enabled? */ |
100 | bool notification; | |
bcbabae8 | 101 | |
e78a2b42 JW |
102 | uint16_t queue_index; |
103 | ||
e66bcc40 | 104 | unsigned int inuse; |
bcbabae8 | 105 | |
7055e687 | 106 | uint16_t vector; |
bf1780b0 | 107 | VirtIOHandleOutput handle_output; |
07931698 | 108 | VirtIOHandleAIOOutput handle_aio_output; |
1cbdabe2 MT |
109 | VirtIODevice *vdev; |
110 | EventNotifier guest_notifier; | |
111 | EventNotifier host_notifier; | |
e0d686bf | 112 | QLIST_ENTRY(VirtQueue) node; |
967f97fa AL |
113 | }; |
114 | ||
c611c764 PB |
115 | static void virtio_free_region_cache(VRingMemoryRegionCaches *caches) |
116 | { | |
117 | if (!caches) { | |
118 | return; | |
119 | } | |
120 | ||
121 | address_space_cache_destroy(&caches->desc); | |
122 | address_space_cache_destroy(&caches->avail); | |
123 | address_space_cache_destroy(&caches->used); | |
124 | g_free(caches); | |
125 | } | |
126 | ||
127 | static void virtio_init_region_cache(VirtIODevice *vdev, int n) | |
128 | { | |
129 | VirtQueue *vq = &vdev->vq[n]; | |
130 | VRingMemoryRegionCaches *old = vq->vring.caches; | |
131 | VRingMemoryRegionCaches *new; | |
132 | hwaddr addr, size; | |
133 | int event_size; | |
134 | ||
135 | event_size = virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; | |
136 | ||
137 | addr = vq->vring.desc; | |
138 | if (!addr) { | |
139 | return; | |
140 | } | |
141 | new = g_new0(VRingMemoryRegionCaches, 1); | |
142 | size = virtio_queue_get_desc_size(vdev, n); | |
143 | address_space_cache_init(&new->desc, vdev->dma_as, | |
144 | addr, size, false); | |
145 | ||
146 | size = virtio_queue_get_used_size(vdev, n) + event_size; | |
147 | address_space_cache_init(&new->used, vdev->dma_as, | |
148 | vq->vring.used, size, true); | |
149 | ||
150 | size = virtio_queue_get_avail_size(vdev, n) + event_size; | |
151 | address_space_cache_init(&new->avail, vdev->dma_as, | |
152 | vq->vring.avail, size, false); | |
153 | ||
154 | atomic_rcu_set(&vq->vring.caches, new); | |
155 | if (old) { | |
156 | call_rcu(old, virtio_free_region_cache, rcu); | |
157 | } | |
158 | } | |
159 | ||
967f97fa | 160 | /* virt queue functions */ |
ab223c95 | 161 | void virtio_queue_update_rings(VirtIODevice *vdev, int n) |
967f97fa | 162 | { |
ab223c95 | 163 | VRing *vring = &vdev->vq[n].vring; |
53c25cea | 164 | |
ab223c95 CH |
165 | if (!vring->desc) { |
166 | /* not yet setup -> nothing to do */ | |
167 | return; | |
168 | } | |
169 | vring->avail = vring->desc + vring->num * sizeof(VRingDesc); | |
170 | vring->used = vring_align(vring->avail + | |
171 | offsetof(VRingAvail, ring[vring->num]), | |
172 | vring->align); | |
c611c764 | 173 | virtio_init_region_cache(vdev, n); |
967f97fa AL |
174 | } |
175 | ||
aa570d6f | 176 | static void vring_desc_read(VirtIODevice *vdev, VRingDesc *desc, |
5eba0404 | 177 | MemoryRegionCache *cache, int i) |
967f97fa | 178 | { |
5eba0404 PB |
179 | address_space_read_cached(cache, i * sizeof(VRingDesc), |
180 | desc, sizeof(VRingDesc)); | |
aa570d6f PB |
181 | virtio_tswap64s(vdev, &desc->addr); |
182 | virtio_tswap32s(vdev, &desc->len); | |
183 | virtio_tswap16s(vdev, &desc->flags); | |
184 | virtio_tswap16s(vdev, &desc->next); | |
967f97fa AL |
185 | } |
186 | ||
187 | static inline uint16_t vring_avail_flags(VirtQueue *vq) | |
188 | { | |
a8170e5e | 189 | hwaddr pa; |
967f97fa | 190 | pa = vq->vring.avail + offsetof(VRingAvail, flags); |
cee3ca00 | 191 | return virtio_lduw_phys(vq->vdev, pa); |
967f97fa AL |
192 | } |
193 | ||
194 | static inline uint16_t vring_avail_idx(VirtQueue *vq) | |
195 | { | |
a8170e5e | 196 | hwaddr pa; |
967f97fa | 197 | pa = vq->vring.avail + offsetof(VRingAvail, idx); |
be1fea9b VM |
198 | vq->shadow_avail_idx = virtio_lduw_phys(vq->vdev, pa); |
199 | return vq->shadow_avail_idx; | |
967f97fa AL |
200 | } |
201 | ||
202 | static inline uint16_t vring_avail_ring(VirtQueue *vq, int i) | |
203 | { | |
a8170e5e | 204 | hwaddr pa; |
967f97fa | 205 | pa = vq->vring.avail + offsetof(VRingAvail, ring[i]); |
cee3ca00 | 206 | return virtio_lduw_phys(vq->vdev, pa); |
967f97fa AL |
207 | } |
208 | ||
e9600c6c | 209 | static inline uint16_t vring_get_used_event(VirtQueue *vq) |
bcbabae8 MT |
210 | { |
211 | return vring_avail_ring(vq, vq->vring.num); | |
212 | } | |
213 | ||
1cdd2ee5 VM |
214 | static inline void vring_used_write(VirtQueue *vq, VRingUsedElem *uelem, |
215 | int i) | |
967f97fa | 216 | { |
a8170e5e | 217 | hwaddr pa; |
1cdd2ee5 VM |
218 | virtio_tswap32s(vq->vdev, &uelem->id); |
219 | virtio_tswap32s(vq->vdev, &uelem->len); | |
220 | pa = vq->vring.used + offsetof(VRingUsed, ring[i]); | |
8607f5c3 | 221 | address_space_write(vq->vdev->dma_as, pa, MEMTXATTRS_UNSPECIFIED, |
1cdd2ee5 | 222 | (void *)uelem, sizeof(VRingUsedElem)); |
967f97fa AL |
223 | } |
224 | ||
225 | static uint16_t vring_used_idx(VirtQueue *vq) | |
226 | { | |
a8170e5e | 227 | hwaddr pa; |
967f97fa | 228 | pa = vq->vring.used + offsetof(VRingUsed, idx); |
cee3ca00 | 229 | return virtio_lduw_phys(vq->vdev, pa); |
967f97fa AL |
230 | } |
231 | ||
bcbabae8 | 232 | static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val) |
967f97fa | 233 | { |
a8170e5e | 234 | hwaddr pa; |
967f97fa | 235 | pa = vq->vring.used + offsetof(VRingUsed, idx); |
cee3ca00 | 236 | virtio_stw_phys(vq->vdev, pa, val); |
b796fcd1 | 237 | vq->used_idx = val; |
967f97fa AL |
238 | } |
239 | ||
240 | static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask) | |
241 | { | |
cee3ca00 | 242 | VirtIODevice *vdev = vq->vdev; |
a8170e5e | 243 | hwaddr pa; |
967f97fa | 244 | pa = vq->vring.used + offsetof(VRingUsed, flags); |
cee3ca00 | 245 | virtio_stw_phys(vdev, pa, virtio_lduw_phys(vdev, pa) | mask); |
967f97fa AL |
246 | } |
247 | ||
248 | static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask) | |
249 | { | |
cee3ca00 | 250 | VirtIODevice *vdev = vq->vdev; |
a8170e5e | 251 | hwaddr pa; |
967f97fa | 252 | pa = vq->vring.used + offsetof(VRingUsed, flags); |
cee3ca00 | 253 | virtio_stw_phys(vdev, pa, virtio_lduw_phys(vdev, pa) & ~mask); |
967f97fa AL |
254 | } |
255 | ||
e9600c6c | 256 | static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val) |
bcbabae8 | 257 | { |
a8170e5e | 258 | hwaddr pa; |
332fa82d | 259 | if (!vq->notification) { |
bcbabae8 MT |
260 | return; |
261 | } | |
262 | pa = vq->vring.used + offsetof(VRingUsed, ring[vq->vring.num]); | |
cee3ca00 | 263 | virtio_stw_phys(vq->vdev, pa, val); |
bcbabae8 MT |
264 | } |
265 | ||
967f97fa AL |
266 | void virtio_queue_set_notification(VirtQueue *vq, int enable) |
267 | { | |
332fa82d | 268 | vq->notification = enable; |
95129d6f | 269 | if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) { |
e9600c6c | 270 | vring_set_avail_event(vq, vring_avail_idx(vq)); |
bcbabae8 | 271 | } else if (enable) { |
967f97fa | 272 | vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY); |
bcbabae8 | 273 | } else { |
967f97fa | 274 | vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY); |
bcbabae8 | 275 | } |
92045d80 MT |
276 | if (enable) { |
277 | /* Expose avail event/used flags before caller checks the avail idx. */ | |
278 | smp_mb(); | |
279 | } | |
967f97fa AL |
280 | } |
281 | ||
282 | int virtio_queue_ready(VirtQueue *vq) | |
283 | { | |
284 | return vq->vring.avail != 0; | |
285 | } | |
286 | ||
be1fea9b VM |
287 | /* Fetch avail_idx from VQ memory only when we really need to know if |
288 | * guest has added some buffers. */ | |
967f97fa AL |
289 | int virtio_queue_empty(VirtQueue *vq) |
290 | { | |
be1fea9b VM |
291 | if (vq->shadow_avail_idx != vq->last_avail_idx) { |
292 | return 0; | |
293 | } | |
294 | ||
967f97fa AL |
295 | return vring_avail_idx(vq) == vq->last_avail_idx; |
296 | } | |
297 | ||
ce317461 JW |
298 | static void virtqueue_unmap_sg(VirtQueue *vq, const VirtQueueElement *elem, |
299 | unsigned int len) | |
967f97fa | 300 | { |
8607f5c3 | 301 | AddressSpace *dma_as = vq->vdev->dma_as; |
967f97fa AL |
302 | unsigned int offset; |
303 | int i; | |
304 | ||
967f97fa AL |
305 | offset = 0; |
306 | for (i = 0; i < elem->in_num; i++) { | |
307 | size_t size = MIN(len - offset, elem->in_sg[i].iov_len); | |
308 | ||
8607f5c3 JW |
309 | dma_memory_unmap(dma_as, elem->in_sg[i].iov_base, |
310 | elem->in_sg[i].iov_len, | |
311 | DMA_DIRECTION_FROM_DEVICE, size); | |
967f97fa | 312 | |
0cea71a2 | 313 | offset += size; |
967f97fa AL |
314 | } |
315 | ||
26b258e1 | 316 | for (i = 0; i < elem->out_num; i++) |
8607f5c3 JW |
317 | dma_memory_unmap(dma_as, elem->out_sg[i].iov_base, |
318 | elem->out_sg[i].iov_len, | |
319 | DMA_DIRECTION_TO_DEVICE, | |
320 | elem->out_sg[i].iov_len); | |
ce317461 JW |
321 | } |
322 | ||
2640d2a5 SH |
323 | /* virtqueue_detach_element: |
324 | * @vq: The #VirtQueue | |
325 | * @elem: The #VirtQueueElement | |
326 | * @len: number of bytes written | |
327 | * | |
328 | * Detach the element from the virtqueue. This function is suitable for device | |
329 | * reset or other situations where a #VirtQueueElement is simply freed and will | |
330 | * not be pushed or discarded. | |
331 | */ | |
332 | void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem, | |
333 | unsigned int len) | |
334 | { | |
335 | vq->inuse--; | |
336 | virtqueue_unmap_sg(vq, elem, len); | |
337 | } | |
338 | ||
27e57efe | 339 | /* virtqueue_unpop: |
2640d2a5 SH |
340 | * @vq: The #VirtQueue |
341 | * @elem: The #VirtQueueElement | |
342 | * @len: number of bytes written | |
343 | * | |
344 | * Pretend the most recent element wasn't popped from the virtqueue. The next | |
345 | * call to virtqueue_pop() will refetch the element. | |
346 | */ | |
27e57efe LP |
347 | void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem, |
348 | unsigned int len) | |
29b9f5ef JW |
349 | { |
350 | vq->last_avail_idx--; | |
2640d2a5 | 351 | virtqueue_detach_element(vq, elem, len); |
29b9f5ef JW |
352 | } |
353 | ||
297a75e6 SH |
354 | /* virtqueue_rewind: |
355 | * @vq: The #VirtQueue | |
356 | * @num: Number of elements to push back | |
357 | * | |
358 | * Pretend that elements weren't popped from the virtqueue. The next | |
359 | * virtqueue_pop() will refetch the oldest element. | |
360 | * | |
27e57efe | 361 | * Use virtqueue_unpop() instead if you have a VirtQueueElement. |
297a75e6 SH |
362 | * |
363 | * Returns: true on success, false if @num is greater than the number of in use | |
364 | * elements. | |
365 | */ | |
366 | bool virtqueue_rewind(VirtQueue *vq, unsigned int num) | |
367 | { | |
368 | if (num > vq->inuse) { | |
369 | return false; | |
370 | } | |
371 | vq->last_avail_idx -= num; | |
372 | vq->inuse -= num; | |
373 | return true; | |
374 | } | |
375 | ||
ce317461 JW |
376 | void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem, |
377 | unsigned int len, unsigned int idx) | |
378 | { | |
1cdd2ee5 VM |
379 | VRingUsedElem uelem; |
380 | ||
ce317461 JW |
381 | trace_virtqueue_fill(vq, elem, len, idx); |
382 | ||
383 | virtqueue_unmap_sg(vq, elem, len); | |
26b258e1 | 384 | |
f5ed3663 SH |
385 | if (unlikely(vq->vdev->broken)) { |
386 | return; | |
387 | } | |
388 | ||
b796fcd1 | 389 | idx = (idx + vq->used_idx) % vq->vring.num; |
967f97fa | 390 | |
1cdd2ee5 VM |
391 | uelem.id = elem->index; |
392 | uelem.len = len; | |
393 | vring_used_write(vq, &uelem, idx); | |
967f97fa AL |
394 | } |
395 | ||
396 | void virtqueue_flush(VirtQueue *vq, unsigned int count) | |
397 | { | |
bcbabae8 | 398 | uint16_t old, new; |
f5ed3663 SH |
399 | |
400 | if (unlikely(vq->vdev->broken)) { | |
401 | vq->inuse -= count; | |
402 | return; | |
403 | } | |
404 | ||
967f97fa | 405 | /* Make sure buffer is written before we update index. */ |
b90d2f35 | 406 | smp_wmb(); |
64979a4d | 407 | trace_virtqueue_flush(vq, count); |
b796fcd1 | 408 | old = vq->used_idx; |
bcbabae8 MT |
409 | new = old + count; |
410 | vring_used_idx_set(vq, new); | |
967f97fa | 411 | vq->inuse -= count; |
bcbabae8 MT |
412 | if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old))) |
413 | vq->signalled_used_valid = false; | |
967f97fa AL |
414 | } |
415 | ||
416 | void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem, | |
417 | unsigned int len) | |
418 | { | |
419 | virtqueue_fill(vq, elem, len, 0); | |
420 | virtqueue_flush(vq, 1); | |
421 | } | |
422 | ||
423 | static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx) | |
424 | { | |
425 | uint16_t num_heads = vring_avail_idx(vq) - idx; | |
426 | ||
427 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
bb6834cf | 428 | if (num_heads > vq->vring.num) { |
4355c1ab | 429 | virtio_error(vq->vdev, "Guest moved used index from %u to %u", |
be1fea9b | 430 | idx, vq->shadow_avail_idx); |
4355c1ab | 431 | return -EINVAL; |
bb6834cf | 432 | } |
a821ce59 MT |
433 | /* On success, callers read a descriptor at vq->last_avail_idx. |
434 | * Make sure descriptor read does not bypass avail index read. */ | |
435 | if (num_heads) { | |
436 | smp_rmb(); | |
437 | } | |
967f97fa AL |
438 | |
439 | return num_heads; | |
440 | } | |
441 | ||
fb1131b6 SH |
442 | static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx, |
443 | unsigned int *head) | |
967f97fa | 444 | { |
967f97fa AL |
445 | /* Grab the next descriptor number they're advertising, and increment |
446 | * the index we've seen. */ | |
fb1131b6 | 447 | *head = vring_avail_ring(vq, idx % vq->vring.num); |
967f97fa AL |
448 | |
449 | /* If their number is silly, that's a fatal mistake. */ | |
fb1131b6 SH |
450 | if (*head >= vq->vring.num) { |
451 | virtio_error(vq->vdev, "Guest says index %u is available", *head); | |
452 | return false; | |
bb6834cf | 453 | } |
967f97fa | 454 | |
fb1131b6 | 455 | return true; |
967f97fa AL |
456 | } |
457 | ||
412e0e81 SH |
458 | enum { |
459 | VIRTQUEUE_READ_DESC_ERROR = -1, | |
460 | VIRTQUEUE_READ_DESC_DONE = 0, /* end of chain */ | |
461 | VIRTQUEUE_READ_DESC_MORE = 1, /* more buffers in chain */ | |
462 | }; | |
967f97fa | 463 | |
412e0e81 | 464 | static int virtqueue_read_next_desc(VirtIODevice *vdev, VRingDesc *desc, |
5eba0404 | 465 | MemoryRegionCache *desc_cache, unsigned int max, |
412e0e81 SH |
466 | unsigned int *next) |
467 | { | |
967f97fa | 468 | /* If this descriptor says it doesn't chain, we're done. */ |
aa570d6f | 469 | if (!(desc->flags & VRING_DESC_F_NEXT)) { |
412e0e81 | 470 | return VIRTQUEUE_READ_DESC_DONE; |
cee3ca00 | 471 | } |
967f97fa AL |
472 | |
473 | /* Check they're not leading us off end of descriptors. */ | |
412e0e81 | 474 | *next = desc->next; |
967f97fa | 475 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
b90d2f35 | 476 | smp_wmb(); |
967f97fa | 477 | |
412e0e81 SH |
478 | if (*next >= max) { |
479 | virtio_error(vdev, "Desc next is %u", *next); | |
480 | return VIRTQUEUE_READ_DESC_ERROR; | |
bb6834cf | 481 | } |
967f97fa | 482 | |
5eba0404 | 483 | vring_desc_read(vdev, desc, desc_cache, *next); |
412e0e81 | 484 | return VIRTQUEUE_READ_DESC_MORE; |
967f97fa AL |
485 | } |
486 | ||
0d8d7690 | 487 | void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes, |
e1f7b481 MT |
488 | unsigned int *out_bytes, |
489 | unsigned max_in_bytes, unsigned max_out_bytes) | |
967f97fa | 490 | { |
9796d0ac PB |
491 | VirtIODevice *vdev = vq->vdev; |
492 | unsigned int max, idx; | |
385ce95d | 493 | unsigned int total_bufs, in_total, out_total; |
5eba0404 PB |
494 | MemoryRegionCache vring_desc_cache; |
495 | MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID; | |
496 | int64_t len = 0; | |
412e0e81 | 497 | int rc; |
967f97fa AL |
498 | |
499 | idx = vq->last_avail_idx; | |
efeea6d0 | 500 | total_bufs = in_total = out_total = 0; |
9796d0ac PB |
501 | |
502 | max = vq->vring.num; | |
5eba0404 PB |
503 | len = address_space_cache_init(&vring_desc_cache, vdev->dma_as, |
504 | vq->vring.desc, max * sizeof(VRingDesc), | |
505 | false); | |
9796d0ac PB |
506 | if (len < max * sizeof(VRingDesc)) { |
507 | virtio_error(vdev, "Cannot map descriptor ring"); | |
508 | goto err; | |
509 | } | |
510 | ||
4355c1ab | 511 | while ((rc = virtqueue_num_heads(vq, idx)) > 0) { |
5eba0404 | 512 | MemoryRegionCache *desc_cache = &vring_desc_cache; |
9796d0ac | 513 | unsigned int num_bufs; |
aa570d6f | 514 | VRingDesc desc; |
b1c7c07f | 515 | unsigned int i; |
967f97fa | 516 | |
efeea6d0 | 517 | num_bufs = total_bufs; |
fb1131b6 SH |
518 | |
519 | if (!virtqueue_get_head(vq, idx++, &i)) { | |
520 | goto err; | |
521 | } | |
522 | ||
5eba0404 | 523 | vring_desc_read(vdev, &desc, desc_cache, i); |
efeea6d0 | 524 | |
aa570d6f PB |
525 | if (desc.flags & VRING_DESC_F_INDIRECT) { |
526 | if (desc.len % sizeof(VRingDesc)) { | |
d65abf85 SH |
527 | virtio_error(vdev, "Invalid size for indirect buffer table"); |
528 | goto err; | |
efeea6d0 MM |
529 | } |
530 | ||
531 | /* If we've got too many, that implies a descriptor loop. */ | |
532 | if (num_bufs >= max) { | |
d65abf85 SH |
533 | virtio_error(vdev, "Looped descriptor"); |
534 | goto err; | |
efeea6d0 MM |
535 | } |
536 | ||
537 | /* loop over the indirect descriptor table */ | |
5eba0404 PB |
538 | len = address_space_cache_init(&indirect_desc_cache, |
539 | vdev->dma_as, | |
540 | desc.addr, desc.len, false); | |
541 | desc_cache = &indirect_desc_cache; | |
9796d0ac PB |
542 | if (len < desc.len) { |
543 | virtio_error(vdev, "Cannot map indirect buffer"); | |
544 | goto err; | |
545 | } | |
546 | ||
aa570d6f | 547 | max = desc.len / sizeof(VRingDesc); |
1ae2757c | 548 | num_bufs = i = 0; |
5eba0404 | 549 | vring_desc_read(vdev, &desc, desc_cache, i); |
efeea6d0 MM |
550 | } |
551 | ||
967f97fa AL |
552 | do { |
553 | /* If we've got too many, that implies a descriptor loop. */ | |
5774cf98 | 554 | if (++num_bufs > max) { |
d65abf85 SH |
555 | virtio_error(vdev, "Looped descriptor"); |
556 | goto err; | |
bb6834cf | 557 | } |
967f97fa | 558 | |
aa570d6f PB |
559 | if (desc.flags & VRING_DESC_F_WRITE) { |
560 | in_total += desc.len; | |
967f97fa | 561 | } else { |
aa570d6f | 562 | out_total += desc.len; |
967f97fa | 563 | } |
e1f7b481 MT |
564 | if (in_total >= max_in_bytes && out_total >= max_out_bytes) { |
565 | goto done; | |
566 | } | |
412e0e81 | 567 | |
5eba0404 | 568 | rc = virtqueue_read_next_desc(vdev, &desc, desc_cache, max, &i); |
412e0e81 SH |
569 | } while (rc == VIRTQUEUE_READ_DESC_MORE); |
570 | ||
571 | if (rc == VIRTQUEUE_READ_DESC_ERROR) { | |
572 | goto err; | |
573 | } | |
efeea6d0 | 574 | |
5eba0404 PB |
575 | if (desc_cache == &indirect_desc_cache) { |
576 | address_space_cache_destroy(&indirect_desc_cache); | |
efeea6d0 | 577 | total_bufs++; |
9796d0ac PB |
578 | } else { |
579 | total_bufs = num_bufs; | |
580 | } | |
967f97fa | 581 | } |
4355c1ab SH |
582 | |
583 | if (rc < 0) { | |
584 | goto err; | |
585 | } | |
586 | ||
e1f7b481 | 587 | done: |
5eba0404 PB |
588 | address_space_cache_destroy(&indirect_desc_cache); |
589 | address_space_cache_destroy(&vring_desc_cache); | |
0d8d7690 AS |
590 | if (in_bytes) { |
591 | *in_bytes = in_total; | |
592 | } | |
593 | if (out_bytes) { | |
594 | *out_bytes = out_total; | |
595 | } | |
d65abf85 SH |
596 | return; |
597 | ||
598 | err: | |
599 | in_total = out_total = 0; | |
600 | goto done; | |
0d8d7690 | 601 | } |
967f97fa | 602 | |
0d8d7690 AS |
603 | int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes, |
604 | unsigned int out_bytes) | |
605 | { | |
606 | unsigned int in_total, out_total; | |
607 | ||
e1f7b481 MT |
608 | virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes); |
609 | return in_bytes <= in_total && out_bytes <= out_total; | |
967f97fa AL |
610 | } |
611 | ||
ec55da19 SH |
612 | static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg, |
613 | hwaddr *addr, struct iovec *iov, | |
3b3b0628 PB |
614 | unsigned int max_num_sg, bool is_write, |
615 | hwaddr pa, size_t sz) | |
616 | { | |
ec55da19 | 617 | bool ok = false; |
3b3b0628 PB |
618 | unsigned num_sg = *p_num_sg; |
619 | assert(num_sg <= max_num_sg); | |
620 | ||
1e7aed70 | 621 | if (!sz) { |
ec55da19 SH |
622 | virtio_error(vdev, "virtio: zero sized buffers are not allowed"); |
623 | goto out; | |
1e7aed70 PP |
624 | } |
625 | ||
3b3b0628 PB |
626 | while (sz) { |
627 | hwaddr len = sz; | |
628 | ||
629 | if (num_sg == max_num_sg) { | |
ec55da19 SH |
630 | virtio_error(vdev, "virtio: too many write descriptors in " |
631 | "indirect table"); | |
632 | goto out; | |
3b3b0628 PB |
633 | } |
634 | ||
8607f5c3 JW |
635 | iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len, |
636 | is_write ? | |
637 | DMA_DIRECTION_FROM_DEVICE : | |
638 | DMA_DIRECTION_TO_DEVICE); | |
973e7170 | 639 | if (!iov[num_sg].iov_base) { |
ec55da19 SH |
640 | virtio_error(vdev, "virtio: bogus descriptor or out of resources"); |
641 | goto out; | |
973e7170 PP |
642 | } |
643 | ||
3b3b0628 PB |
644 | iov[num_sg].iov_len = len; |
645 | addr[num_sg] = pa; | |
646 | ||
647 | sz -= len; | |
648 | pa += len; | |
649 | num_sg++; | |
650 | } | |
ec55da19 SH |
651 | ok = true; |
652 | ||
653 | out: | |
3b3b0628 | 654 | *p_num_sg = num_sg; |
ec55da19 SH |
655 | return ok; |
656 | } | |
657 | ||
658 | /* Only used by error code paths before we have a VirtQueueElement (therefore | |
659 | * virtqueue_unmap_sg() can't be used). Assumes buffers weren't written to | |
660 | * yet. | |
661 | */ | |
662 | static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num, | |
663 | struct iovec *iov) | |
664 | { | |
665 | unsigned int i; | |
666 | ||
667 | for (i = 0; i < out_num + in_num; i++) { | |
668 | int is_write = i >= out_num; | |
669 | ||
670 | cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0); | |
671 | iov++; | |
672 | } | |
3b3b0628 PB |
673 | } |
674 | ||
8607f5c3 JW |
675 | static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg, |
676 | hwaddr *addr, unsigned int *num_sg, | |
6bdc21c0 | 677 | int is_write) |
42fb2e07 KW |
678 | { |
679 | unsigned int i; | |
a8170e5e | 680 | hwaddr len; |
42fb2e07 | 681 | |
8059feee | 682 | for (i = 0; i < *num_sg; i++) { |
42fb2e07 | 683 | len = sg[i].iov_len; |
8607f5c3 JW |
684 | sg[i].iov_base = dma_memory_map(vdev->dma_as, |
685 | addr[i], &len, is_write ? | |
686 | DMA_DIRECTION_FROM_DEVICE : | |
687 | DMA_DIRECTION_TO_DEVICE); | |
8059feee | 688 | if (!sg[i].iov_base) { |
1a285899 | 689 | error_report("virtio: error trying to map MMIO memory"); |
42fb2e07 KW |
690 | exit(1); |
691 | } | |
3b3b0628 PB |
692 | if (len != sg[i].iov_len) { |
693 | error_report("virtio: unexpected memory split"); | |
8059feee MT |
694 | exit(1); |
695 | } | |
42fb2e07 KW |
696 | } |
697 | } | |
698 | ||
8607f5c3 | 699 | void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem) |
8059feee | 700 | { |
6bdc21c0 MT |
701 | virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, &elem->in_num, 1); |
702 | virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, &elem->out_num, 0); | |
3724650d PB |
703 | } |
704 | ||
bf91bd27 | 705 | static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num) |
3724650d PB |
706 | { |
707 | VirtQueueElement *elem; | |
708 | size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0])); | |
709 | size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]); | |
710 | size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]); | |
711 | size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0])); | |
712 | size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]); | |
713 | size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]); | |
714 | ||
715 | assert(sz >= sizeof(VirtQueueElement)); | |
716 | elem = g_malloc(out_sg_end); | |
717 | elem->out_num = out_num; | |
718 | elem->in_num = in_num; | |
719 | elem->in_addr = (void *)elem + in_addr_ofs; | |
720 | elem->out_addr = (void *)elem + out_addr_ofs; | |
721 | elem->in_sg = (void *)elem + in_sg_ofs; | |
722 | elem->out_sg = (void *)elem + out_sg_ofs; | |
723 | return elem; | |
8059feee MT |
724 | } |
725 | ||
51b19ebe | 726 | void *virtqueue_pop(VirtQueue *vq, size_t sz) |
967f97fa | 727 | { |
5774cf98 | 728 | unsigned int i, head, max; |
5eba0404 PB |
729 | MemoryRegionCache vring_desc_cache; |
730 | MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID; | |
731 | MemoryRegionCache *desc_cache; | |
732 | int64_t len; | |
cee3ca00 | 733 | VirtIODevice *vdev = vq->vdev; |
9796d0ac | 734 | VirtQueueElement *elem = NULL; |
3b3b0628 PB |
735 | unsigned out_num, in_num; |
736 | hwaddr addr[VIRTQUEUE_MAX_SIZE]; | |
737 | struct iovec iov[VIRTQUEUE_MAX_SIZE]; | |
aa570d6f | 738 | VRingDesc desc; |
412e0e81 | 739 | int rc; |
967f97fa | 740 | |
f5ed3663 SH |
741 | if (unlikely(vdev->broken)) { |
742 | return NULL; | |
743 | } | |
be1fea9b | 744 | if (virtio_queue_empty(vq)) { |
51b19ebe PB |
745 | return NULL; |
746 | } | |
be1fea9b VM |
747 | /* Needed after virtio_queue_empty(), see comment in |
748 | * virtqueue_num_heads(). */ | |
749 | smp_rmb(); | |
967f97fa AL |
750 | |
751 | /* When we start there are none of either input nor output. */ | |
3b3b0628 | 752 | out_num = in_num = 0; |
967f97fa | 753 | |
5774cf98 MM |
754 | max = vq->vring.num; |
755 | ||
afd9096e | 756 | if (vq->inuse >= vq->vring.num) { |
ec55da19 SH |
757 | virtio_error(vdev, "Virtqueue size exceeded"); |
758 | return NULL; | |
afd9096e SH |
759 | } |
760 | ||
fb1131b6 SH |
761 | if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) { |
762 | return NULL; | |
763 | } | |
764 | ||
95129d6f | 765 | if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) { |
e9600c6c | 766 | vring_set_avail_event(vq, vq->last_avail_idx); |
bcbabae8 | 767 | } |
efeea6d0 | 768 | |
fb1131b6 | 769 | i = head; |
9796d0ac | 770 | |
5eba0404 PB |
771 | len = address_space_cache_init(&vring_desc_cache, vdev->dma_as, |
772 | vq->vring.desc, max * sizeof(VRingDesc), | |
773 | false); | |
9796d0ac PB |
774 | if (len < max * sizeof(VRingDesc)) { |
775 | virtio_error(vdev, "Cannot map descriptor ring"); | |
776 | goto done; | |
777 | } | |
778 | ||
5eba0404 PB |
779 | desc_cache = &vring_desc_cache; |
780 | vring_desc_read(vdev, &desc, desc_cache, i); | |
aa570d6f PB |
781 | if (desc.flags & VRING_DESC_F_INDIRECT) { |
782 | if (desc.len % sizeof(VRingDesc)) { | |
ec55da19 | 783 | virtio_error(vdev, "Invalid size for indirect buffer table"); |
9796d0ac | 784 | goto done; |
efeea6d0 MM |
785 | } |
786 | ||
787 | /* loop over the indirect descriptor table */ | |
5eba0404 PB |
788 | len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as, |
789 | desc.addr, desc.len, false); | |
790 | desc_cache = &indirect_desc_cache; | |
9796d0ac PB |
791 | if (len < desc.len) { |
792 | virtio_error(vdev, "Cannot map indirect buffer"); | |
793 | goto done; | |
794 | } | |
795 | ||
aa570d6f | 796 | max = desc.len / sizeof(VRingDesc); |
efeea6d0 | 797 | i = 0; |
5eba0404 | 798 | vring_desc_read(vdev, &desc, desc_cache, i); |
efeea6d0 MM |
799 | } |
800 | ||
42fb2e07 | 801 | /* Collect all the descriptors */ |
967f97fa | 802 | do { |
ec55da19 SH |
803 | bool map_ok; |
804 | ||
aa570d6f | 805 | if (desc.flags & VRING_DESC_F_WRITE) { |
ec55da19 SH |
806 | map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num, |
807 | iov + out_num, | |
808 | VIRTQUEUE_MAX_SIZE - out_num, true, | |
809 | desc.addr, desc.len); | |
42fb2e07 | 810 | } else { |
3b3b0628 | 811 | if (in_num) { |
ec55da19 SH |
812 | virtio_error(vdev, "Incorrect order for descriptors"); |
813 | goto err_undo_map; | |
c8eac1cf | 814 | } |
ec55da19 SH |
815 | map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov, |
816 | VIRTQUEUE_MAX_SIZE, false, | |
817 | desc.addr, desc.len); | |
818 | } | |
819 | if (!map_ok) { | |
820 | goto err_undo_map; | |
42fb2e07 | 821 | } |
967f97fa | 822 | |
967f97fa | 823 | /* If we've got too many, that implies a descriptor loop. */ |
3b3b0628 | 824 | if ((in_num + out_num) > max) { |
ec55da19 SH |
825 | virtio_error(vdev, "Looped descriptor"); |
826 | goto err_undo_map; | |
bb6834cf | 827 | } |
412e0e81 | 828 | |
5eba0404 | 829 | rc = virtqueue_read_next_desc(vdev, &desc, desc_cache, max, &i); |
412e0e81 SH |
830 | } while (rc == VIRTQUEUE_READ_DESC_MORE); |
831 | ||
832 | if (rc == VIRTQUEUE_READ_DESC_ERROR) { | |
833 | goto err_undo_map; | |
834 | } | |
967f97fa | 835 | |
3b3b0628 PB |
836 | /* Now copy what we have collected and mapped */ |
837 | elem = virtqueue_alloc_element(sz, out_num, in_num); | |
967f97fa | 838 | elem->index = head; |
3b3b0628 PB |
839 | for (i = 0; i < out_num; i++) { |
840 | elem->out_addr[i] = addr[i]; | |
841 | elem->out_sg[i] = iov[i]; | |
842 | } | |
843 | for (i = 0; i < in_num; i++) { | |
844 | elem->in_addr[i] = addr[out_num + i]; | |
845 | elem->in_sg[i] = iov[out_num + i]; | |
846 | } | |
967f97fa AL |
847 | |
848 | vq->inuse++; | |
849 | ||
64979a4d | 850 | trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num); |
9796d0ac | 851 | done: |
5eba0404 PB |
852 | address_space_cache_destroy(&indirect_desc_cache); |
853 | address_space_cache_destroy(&vring_desc_cache); | |
9796d0ac | 854 | |
51b19ebe | 855 | return elem; |
ec55da19 SH |
856 | |
857 | err_undo_map: | |
858 | virtqueue_undo_map_desc(out_num, in_num, iov); | |
9796d0ac | 859 | goto done; |
967f97fa AL |
860 | } |
861 | ||
54e17709 YB |
862 | /* virtqueue_drop_all: |
863 | * @vq: The #VirtQueue | |
864 | * Drops all queued buffers and indicates them to the guest | |
865 | * as if they are done. Useful when buffers can not be | |
866 | * processed but must be returned to the guest. | |
867 | */ | |
868 | unsigned int virtqueue_drop_all(VirtQueue *vq) | |
869 | { | |
870 | unsigned int dropped = 0; | |
871 | VirtQueueElement elem = {}; | |
872 | VirtIODevice *vdev = vq->vdev; | |
873 | bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); | |
874 | ||
875 | if (unlikely(vdev->broken)) { | |
876 | return 0; | |
877 | } | |
878 | ||
879 | while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) { | |
880 | /* works similar to virtqueue_pop but does not map buffers | |
881 | * and does not allocate any memory */ | |
882 | smp_rmb(); | |
883 | if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) { | |
884 | break; | |
885 | } | |
886 | vq->inuse++; | |
887 | vq->last_avail_idx++; | |
888 | if (fEventIdx) { | |
889 | vring_set_avail_event(vq, vq->last_avail_idx); | |
890 | } | |
891 | /* immediately push the element, nothing to unmap | |
892 | * as both in_num and out_num are set to 0 */ | |
893 | virtqueue_push(vq, &elem, 0); | |
894 | dropped++; | |
895 | } | |
896 | ||
897 | return dropped; | |
898 | } | |
899 | ||
3724650d PB |
900 | /* Reading and writing a structure directly to QEMUFile is *awful*, but |
901 | * it is what QEMU has always done by mistake. We can change it sooner | |
902 | * or later by bumping the version number of the affected vm states. | |
903 | * In the meanwhile, since the in-memory layout of VirtQueueElement | |
904 | * has changed, we need to marshal to and from the layout that was | |
905 | * used before the change. | |
906 | */ | |
907 | typedef struct VirtQueueElementOld { | |
908 | unsigned int index; | |
909 | unsigned int out_num; | |
910 | unsigned int in_num; | |
911 | hwaddr in_addr[VIRTQUEUE_MAX_SIZE]; | |
912 | hwaddr out_addr[VIRTQUEUE_MAX_SIZE]; | |
913 | struct iovec in_sg[VIRTQUEUE_MAX_SIZE]; | |
914 | struct iovec out_sg[VIRTQUEUE_MAX_SIZE]; | |
915 | } VirtQueueElementOld; | |
916 | ||
8607f5c3 | 917 | void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz) |
ab281c17 | 918 | { |
3724650d PB |
919 | VirtQueueElement *elem; |
920 | VirtQueueElementOld data; | |
921 | int i; | |
922 | ||
923 | qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld)); | |
924 | ||
6bdc21c0 MT |
925 | /* TODO: teach all callers that this can fail, and return failure instead |
926 | * of asserting here. | |
927 | * When we do, we might be able to re-enable NDEBUG below. | |
928 | */ | |
929 | #ifdef NDEBUG | |
930 | #error building with NDEBUG is not supported | |
931 | #endif | |
932 | assert(ARRAY_SIZE(data.in_addr) >= data.in_num); | |
933 | assert(ARRAY_SIZE(data.out_addr) >= data.out_num); | |
934 | ||
3724650d PB |
935 | elem = virtqueue_alloc_element(sz, data.out_num, data.in_num); |
936 | elem->index = data.index; | |
937 | ||
938 | for (i = 0; i < elem->in_num; i++) { | |
939 | elem->in_addr[i] = data.in_addr[i]; | |
940 | } | |
941 | ||
942 | for (i = 0; i < elem->out_num; i++) { | |
943 | elem->out_addr[i] = data.out_addr[i]; | |
944 | } | |
945 | ||
946 | for (i = 0; i < elem->in_num; i++) { | |
947 | /* Base is overwritten by virtqueue_map. */ | |
948 | elem->in_sg[i].iov_base = 0; | |
949 | elem->in_sg[i].iov_len = data.in_sg[i].iov_len; | |
950 | } | |
951 | ||
952 | for (i = 0; i < elem->out_num; i++) { | |
953 | /* Base is overwritten by virtqueue_map. */ | |
954 | elem->out_sg[i].iov_base = 0; | |
955 | elem->out_sg[i].iov_len = data.out_sg[i].iov_len; | |
956 | } | |
957 | ||
8607f5c3 | 958 | virtqueue_map(vdev, elem); |
ab281c17 PB |
959 | return elem; |
960 | } | |
961 | ||
962 | void qemu_put_virtqueue_element(QEMUFile *f, VirtQueueElement *elem) | |
963 | { | |
3724650d PB |
964 | VirtQueueElementOld data; |
965 | int i; | |
966 | ||
967 | memset(&data, 0, sizeof(data)); | |
968 | data.index = elem->index; | |
969 | data.in_num = elem->in_num; | |
970 | data.out_num = elem->out_num; | |
971 | ||
972 | for (i = 0; i < elem->in_num; i++) { | |
973 | data.in_addr[i] = elem->in_addr[i]; | |
974 | } | |
975 | ||
976 | for (i = 0; i < elem->out_num; i++) { | |
977 | data.out_addr[i] = elem->out_addr[i]; | |
978 | } | |
979 | ||
980 | for (i = 0; i < elem->in_num; i++) { | |
981 | /* Base is overwritten by virtqueue_map when loading. Do not | |
982 | * save it, as it would leak the QEMU address space layout. */ | |
983 | data.in_sg[i].iov_len = elem->in_sg[i].iov_len; | |
984 | } | |
985 | ||
986 | for (i = 0; i < elem->out_num; i++) { | |
987 | /* Do not save iov_base as above. */ | |
988 | data.out_sg[i].iov_len = elem->out_sg[i].iov_len; | |
989 | } | |
990 | qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld)); | |
ab281c17 PB |
991 | } |
992 | ||
967f97fa | 993 | /* virtio device */ |
7055e687 MT |
994 | static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector) |
995 | { | |
1c819449 FK |
996 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); |
997 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); | |
998 | ||
f5ed3663 SH |
999 | if (unlikely(vdev->broken)) { |
1000 | return; | |
1001 | } | |
1002 | ||
1c819449 FK |
1003 | if (k->notify) { |
1004 | k->notify(qbus->parent, vector); | |
7055e687 MT |
1005 | } |
1006 | } | |
967f97fa | 1007 | |
53c25cea | 1008 | void virtio_update_irq(VirtIODevice *vdev) |
967f97fa | 1009 | { |
7055e687 | 1010 | virtio_notify_vector(vdev, VIRTIO_NO_VECTOR); |
967f97fa AL |
1011 | } |
1012 | ||
0b352fd6 CH |
1013 | static int virtio_validate_features(VirtIODevice *vdev) |
1014 | { | |
1015 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); | |
1016 | ||
8607f5c3 JW |
1017 | if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) && |
1018 | !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) { | |
1019 | return -EFAULT; | |
1020 | } | |
1021 | ||
0b352fd6 CH |
1022 | if (k->validate_features) { |
1023 | return k->validate_features(vdev); | |
1024 | } else { | |
1025 | return 0; | |
1026 | } | |
1027 | } | |
1028 | ||
1029 | int virtio_set_status(VirtIODevice *vdev, uint8_t val) | |
4e1837f8 | 1030 | { |
181103cd | 1031 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); |
4e1837f8 SH |
1032 | trace_virtio_set_status(vdev, val); |
1033 | ||
95129d6f | 1034 | if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { |
0b352fd6 CH |
1035 | if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) && |
1036 | val & VIRTIO_CONFIG_S_FEATURES_OK) { | |
1037 | int ret = virtio_validate_features(vdev); | |
1038 | ||
1039 | if (ret) { | |
1040 | return ret; | |
1041 | } | |
1042 | } | |
1043 | } | |
181103cd FK |
1044 | if (k->set_status) { |
1045 | k->set_status(vdev, val); | |
4e1837f8 SH |
1046 | } |
1047 | vdev->status = val; | |
0b352fd6 | 1048 | return 0; |
4e1837f8 SH |
1049 | } |
1050 | ||
616a6552 GK |
1051 | bool target_words_bigendian(void); |
1052 | static enum virtio_device_endian virtio_default_endian(void) | |
1053 | { | |
1054 | if (target_words_bigendian()) { | |
1055 | return VIRTIO_DEVICE_ENDIAN_BIG; | |
1056 | } else { | |
1057 | return VIRTIO_DEVICE_ENDIAN_LITTLE; | |
1058 | } | |
1059 | } | |
1060 | ||
1061 | static enum virtio_device_endian virtio_current_cpu_endian(void) | |
1062 | { | |
1063 | CPUClass *cc = CPU_GET_CLASS(current_cpu); | |
1064 | ||
1065 | if (cc->virtio_is_big_endian(current_cpu)) { | |
1066 | return VIRTIO_DEVICE_ENDIAN_BIG; | |
1067 | } else { | |
1068 | return VIRTIO_DEVICE_ENDIAN_LITTLE; | |
1069 | } | |
1070 | } | |
1071 | ||
53c25cea | 1072 | void virtio_reset(void *opaque) |
967f97fa AL |
1073 | { |
1074 | VirtIODevice *vdev = opaque; | |
181103cd | 1075 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); |
967f97fa AL |
1076 | int i; |
1077 | ||
e0c472d8 | 1078 | virtio_set_status(vdev, 0); |
616a6552 GK |
1079 | if (current_cpu) { |
1080 | /* Guest initiated reset */ | |
1081 | vdev->device_endian = virtio_current_cpu_endian(); | |
1082 | } else { | |
1083 | /* System reset */ | |
1084 | vdev->device_endian = virtio_default_endian(); | |
1085 | } | |
e0c472d8 | 1086 | |
181103cd FK |
1087 | if (k->reset) { |
1088 | k->reset(vdev); | |
1089 | } | |
967f97fa | 1090 | |
f5ed3663 | 1091 | vdev->broken = false; |
704a76fc | 1092 | vdev->guest_features = 0; |
967f97fa AL |
1093 | vdev->queue_sel = 0; |
1094 | vdev->status = 0; | |
0687c37c | 1095 | atomic_set(&vdev->isr, 0); |
7055e687 MT |
1096 | vdev->config_vector = VIRTIO_NO_VECTOR; |
1097 | virtio_notify_vector(vdev, vdev->config_vector); | |
967f97fa | 1098 | |
87b3bd1c | 1099 | for(i = 0; i < VIRTIO_QUEUE_MAX; i++) { |
967f97fa AL |
1100 | vdev->vq[i].vring.desc = 0; |
1101 | vdev->vq[i].vring.avail = 0; | |
1102 | vdev->vq[i].vring.used = 0; | |
1103 | vdev->vq[i].last_avail_idx = 0; | |
be1fea9b | 1104 | vdev->vq[i].shadow_avail_idx = 0; |
b796fcd1 | 1105 | vdev->vq[i].used_idx = 0; |
e0d686bf | 1106 | virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR); |
bcbabae8 MT |
1107 | vdev->vq[i].signalled_used = 0; |
1108 | vdev->vq[i].signalled_used_valid = false; | |
332fa82d | 1109 | vdev->vq[i].notification = true; |
46c5d082 | 1110 | vdev->vq[i].vring.num = vdev->vq[i].vring.num_default; |
4b7f91ed | 1111 | vdev->vq[i].inuse = 0; |
967f97fa AL |
1112 | } |
1113 | } | |
1114 | ||
53c25cea | 1115 | uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr) |
967f97fa | 1116 | { |
181103cd | 1117 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); |
967f97fa AL |
1118 | uint8_t val; |
1119 | ||
5f5a1318 | 1120 | if (addr + sizeof(val) > vdev->config_len) { |
967f97fa | 1121 | return (uint32_t)-1; |
5f5a1318 JW |
1122 | } |
1123 | ||
1124 | k->get_config(vdev, vdev->config); | |
967f97fa | 1125 | |
06dbfc6f | 1126 | val = ldub_p(vdev->config + addr); |
967f97fa AL |
1127 | return val; |
1128 | } | |
1129 | ||
53c25cea | 1130 | uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr) |
967f97fa | 1131 | { |
181103cd | 1132 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); |
967f97fa AL |
1133 | uint16_t val; |
1134 | ||
5f5a1318 | 1135 | if (addr + sizeof(val) > vdev->config_len) { |
967f97fa | 1136 | return (uint32_t)-1; |
5f5a1318 JW |
1137 | } |
1138 | ||
1139 | k->get_config(vdev, vdev->config); | |
967f97fa | 1140 | |
06dbfc6f | 1141 | val = lduw_p(vdev->config + addr); |
967f97fa AL |
1142 | return val; |
1143 | } | |
1144 | ||
53c25cea | 1145 | uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr) |
967f97fa | 1146 | { |
181103cd | 1147 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); |
967f97fa AL |
1148 | uint32_t val; |
1149 | ||
5f5a1318 | 1150 | if (addr + sizeof(val) > vdev->config_len) { |
967f97fa | 1151 | return (uint32_t)-1; |
5f5a1318 JW |
1152 | } |
1153 | ||
1154 | k->get_config(vdev, vdev->config); | |
967f97fa | 1155 | |
06dbfc6f | 1156 | val = ldl_p(vdev->config + addr); |
967f97fa AL |
1157 | return val; |
1158 | } | |
1159 | ||
53c25cea | 1160 | void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data) |
967f97fa | 1161 | { |
181103cd | 1162 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); |
967f97fa AL |
1163 | uint8_t val = data; |
1164 | ||
5f5a1318 | 1165 | if (addr + sizeof(val) > vdev->config_len) { |
967f97fa | 1166 | return; |
5f5a1318 | 1167 | } |
967f97fa | 1168 | |
06dbfc6f | 1169 | stb_p(vdev->config + addr, val); |
967f97fa | 1170 | |
181103cd FK |
1171 | if (k->set_config) { |
1172 | k->set_config(vdev, vdev->config); | |
1173 | } | |
967f97fa AL |
1174 | } |
1175 | ||
53c25cea | 1176 | void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data) |
967f97fa | 1177 | { |
181103cd | 1178 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); |
967f97fa AL |
1179 | uint16_t val = data; |
1180 | ||
5f5a1318 | 1181 | if (addr + sizeof(val) > vdev->config_len) { |
967f97fa | 1182 | return; |
5f5a1318 | 1183 | } |
967f97fa | 1184 | |
06dbfc6f | 1185 | stw_p(vdev->config + addr, val); |
967f97fa | 1186 | |
181103cd FK |
1187 | if (k->set_config) { |
1188 | k->set_config(vdev, vdev->config); | |
1189 | } | |
967f97fa AL |
1190 | } |
1191 | ||
53c25cea | 1192 | void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data) |
967f97fa | 1193 | { |
181103cd | 1194 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); |
967f97fa AL |
1195 | uint32_t val = data; |
1196 | ||
5f5a1318 | 1197 | if (addr + sizeof(val) > vdev->config_len) { |
967f97fa | 1198 | return; |
5f5a1318 | 1199 | } |
967f97fa | 1200 | |
06dbfc6f | 1201 | stl_p(vdev->config + addr, val); |
967f97fa | 1202 | |
181103cd FK |
1203 | if (k->set_config) { |
1204 | k->set_config(vdev, vdev->config); | |
1205 | } | |
967f97fa AL |
1206 | } |
1207 | ||
adfb743c MT |
1208 | uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr) |
1209 | { | |
1210 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); | |
1211 | uint8_t val; | |
1212 | ||
1213 | if (addr + sizeof(val) > vdev->config_len) { | |
1214 | return (uint32_t)-1; | |
1215 | } | |
1216 | ||
1217 | k->get_config(vdev, vdev->config); | |
1218 | ||
1219 | val = ldub_p(vdev->config + addr); | |
1220 | return val; | |
1221 | } | |
1222 | ||
1223 | uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr) | |
1224 | { | |
1225 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); | |
1226 | uint16_t val; | |
1227 | ||
1228 | if (addr + sizeof(val) > vdev->config_len) { | |
1229 | return (uint32_t)-1; | |
1230 | } | |
1231 | ||
1232 | k->get_config(vdev, vdev->config); | |
1233 | ||
1234 | val = lduw_le_p(vdev->config + addr); | |
1235 | return val; | |
1236 | } | |
1237 | ||
1238 | uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr) | |
1239 | { | |
1240 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); | |
1241 | uint32_t val; | |
1242 | ||
1243 | if (addr + sizeof(val) > vdev->config_len) { | |
1244 | return (uint32_t)-1; | |
1245 | } | |
1246 | ||
1247 | k->get_config(vdev, vdev->config); | |
1248 | ||
1249 | val = ldl_le_p(vdev->config + addr); | |
1250 | return val; | |
1251 | } | |
1252 | ||
1253 | void virtio_config_modern_writeb(VirtIODevice *vdev, | |
1254 | uint32_t addr, uint32_t data) | |
1255 | { | |
1256 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); | |
1257 | uint8_t val = data; | |
1258 | ||
1259 | if (addr + sizeof(val) > vdev->config_len) { | |
1260 | return; | |
1261 | } | |
1262 | ||
1263 | stb_p(vdev->config + addr, val); | |
1264 | ||
1265 | if (k->set_config) { | |
1266 | k->set_config(vdev, vdev->config); | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | void virtio_config_modern_writew(VirtIODevice *vdev, | |
1271 | uint32_t addr, uint32_t data) | |
1272 | { | |
1273 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); | |
1274 | uint16_t val = data; | |
1275 | ||
1276 | if (addr + sizeof(val) > vdev->config_len) { | |
1277 | return; | |
1278 | } | |
1279 | ||
1280 | stw_le_p(vdev->config + addr, val); | |
1281 | ||
1282 | if (k->set_config) { | |
1283 | k->set_config(vdev, vdev->config); | |
1284 | } | |
1285 | } | |
1286 | ||
1287 | void virtio_config_modern_writel(VirtIODevice *vdev, | |
1288 | uint32_t addr, uint32_t data) | |
1289 | { | |
1290 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); | |
1291 | uint32_t val = data; | |
1292 | ||
1293 | if (addr + sizeof(val) > vdev->config_len) { | |
1294 | return; | |
1295 | } | |
1296 | ||
1297 | stl_le_p(vdev->config + addr, val); | |
1298 | ||
1299 | if (k->set_config) { | |
1300 | k->set_config(vdev, vdev->config); | |
1301 | } | |
1302 | } | |
1303 | ||
a8170e5e | 1304 | void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr) |
967f97fa | 1305 | { |
ab223c95 CH |
1306 | vdev->vq[n].vring.desc = addr; |
1307 | virtio_queue_update_rings(vdev, n); | |
53c25cea PB |
1308 | } |
1309 | ||
a8170e5e | 1310 | hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n) |
53c25cea | 1311 | { |
ab223c95 CH |
1312 | return vdev->vq[n].vring.desc; |
1313 | } | |
1314 | ||
1315 | void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc, | |
1316 | hwaddr avail, hwaddr used) | |
1317 | { | |
1318 | vdev->vq[n].vring.desc = desc; | |
1319 | vdev->vq[n].vring.avail = avail; | |
1320 | vdev->vq[n].vring.used = used; | |
c611c764 | 1321 | virtio_init_region_cache(vdev, n); |
53c25cea PB |
1322 | } |
1323 | ||
e63c0ba1 PM |
1324 | void virtio_queue_set_num(VirtIODevice *vdev, int n, int num) |
1325 | { | |
f6049f44 PM |
1326 | /* Don't allow guest to flip queue between existent and |
1327 | * nonexistent states, or to set it to an invalid size. | |
1328 | */ | |
1329 | if (!!num != !!vdev->vq[n].vring.num || | |
1330 | num > VIRTQUEUE_MAX_SIZE || | |
1331 | num < 0) { | |
1332 | return; | |
e63c0ba1 | 1333 | } |
f6049f44 | 1334 | vdev->vq[n].vring.num = num; |
e63c0ba1 PM |
1335 | } |
1336 | ||
e0d686bf JW |
1337 | VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector) |
1338 | { | |
1339 | return QLIST_FIRST(&vdev->vector_queues[vector]); | |
1340 | } | |
1341 | ||
1342 | VirtQueue *virtio_vector_next_queue(VirtQueue *vq) | |
1343 | { | |
1344 | return QLIST_NEXT(vq, node); | |
1345 | } | |
1346 | ||
53c25cea PB |
1347 | int virtio_queue_get_num(VirtIODevice *vdev, int n) |
1348 | { | |
1349 | return vdev->vq[n].vring.num; | |
1350 | } | |
967f97fa | 1351 | |
8c797e75 MT |
1352 | int virtio_queue_get_max_num(VirtIODevice *vdev, int n) |
1353 | { | |
1354 | return vdev->vq[n].vring.num_default; | |
1355 | } | |
1356 | ||
8ad176aa JW |
1357 | int virtio_get_num_queues(VirtIODevice *vdev) |
1358 | { | |
1359 | int i; | |
1360 | ||
87b3bd1c | 1361 | for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { |
8ad176aa JW |
1362 | if (!virtio_queue_get_num(vdev, i)) { |
1363 | break; | |
1364 | } | |
1365 | } | |
1366 | ||
1367 | return i; | |
1368 | } | |
1369 | ||
6ce69d1c PM |
1370 | void virtio_queue_set_align(VirtIODevice *vdev, int n, int align) |
1371 | { | |
1372 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); | |
1373 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); | |
1374 | ||
ab223c95 | 1375 | /* virtio-1 compliant devices cannot change the alignment */ |
95129d6f | 1376 | if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { |
ab223c95 CH |
1377 | error_report("tried to modify queue alignment for virtio-1 device"); |
1378 | return; | |
1379 | } | |
6ce69d1c PM |
1380 | /* Check that the transport told us it was going to do this |
1381 | * (so a buggy transport will immediately assert rather than | |
1382 | * silently failing to migrate this state) | |
1383 | */ | |
1384 | assert(k->has_variable_vring_alignment); | |
1385 | ||
1386 | vdev->vq[n].vring.align = align; | |
ab223c95 | 1387 | virtio_queue_update_rings(vdev, n); |
6ce69d1c PM |
1388 | } |
1389 | ||
07931698 | 1390 | static bool virtio_queue_notify_aio_vq(VirtQueue *vq) |
344dc16f MT |
1391 | { |
1392 | if (vq->vring.desc && vq->handle_aio_output) { | |
1393 | VirtIODevice *vdev = vq->vdev; | |
1394 | ||
1395 | trace_virtio_queue_notify(vdev, vq - vdev->vq, vq); | |
07931698 | 1396 | return vq->handle_aio_output(vdev, vq); |
344dc16f | 1397 | } |
07931698 FZ |
1398 | |
1399 | return false; | |
344dc16f MT |
1400 | } |
1401 | ||
2b2cbcad | 1402 | static void virtio_queue_notify_vq(VirtQueue *vq) |
25db9ebe | 1403 | { |
9e0f5b81 | 1404 | if (vq->vring.desc && vq->handle_output) { |
25db9ebe | 1405 | VirtIODevice *vdev = vq->vdev; |
9e0f5b81 | 1406 | |
f5ed3663 SH |
1407 | if (unlikely(vdev->broken)) { |
1408 | return; | |
1409 | } | |
1410 | ||
25db9ebe SH |
1411 | trace_virtio_queue_notify(vdev, vq - vdev->vq, vq); |
1412 | vq->handle_output(vdev, vq); | |
1413 | } | |
1414 | } | |
1415 | ||
53c25cea PB |
1416 | void virtio_queue_notify(VirtIODevice *vdev, int n) |
1417 | { | |
7157e2e2 | 1418 | virtio_queue_notify_vq(&vdev->vq[n]); |
967f97fa AL |
1419 | } |
1420 | ||
7055e687 MT |
1421 | uint16_t virtio_queue_vector(VirtIODevice *vdev, int n) |
1422 | { | |
87b3bd1c | 1423 | return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector : |
7055e687 MT |
1424 | VIRTIO_NO_VECTOR; |
1425 | } | |
1426 | ||
1427 | void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector) | |
1428 | { | |
e0d686bf JW |
1429 | VirtQueue *vq = &vdev->vq[n]; |
1430 | ||
87b3bd1c | 1431 | if (n < VIRTIO_QUEUE_MAX) { |
e0d686bf JW |
1432 | if (vdev->vector_queues && |
1433 | vdev->vq[n].vector != VIRTIO_NO_VECTOR) { | |
1434 | QLIST_REMOVE(vq, node); | |
1435 | } | |
7055e687 | 1436 | vdev->vq[n].vector = vector; |
e0d686bf JW |
1437 | if (vdev->vector_queues && |
1438 | vector != VIRTIO_NO_VECTOR) { | |
1439 | QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node); | |
1440 | } | |
1441 | } | |
7055e687 MT |
1442 | } |
1443 | ||
f1ac6a55 PB |
1444 | VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size, |
1445 | VirtIOHandleOutput handle_output) | |
967f97fa AL |
1446 | { |
1447 | int i; | |
1448 | ||
87b3bd1c | 1449 | for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { |
967f97fa AL |
1450 | if (vdev->vq[i].vring.num == 0) |
1451 | break; | |
1452 | } | |
1453 | ||
87b3bd1c | 1454 | if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE) |
967f97fa AL |
1455 | abort(); |
1456 | ||
1457 | vdev->vq[i].vring.num = queue_size; | |
46c5d082 | 1458 | vdev->vq[i].vring.num_default = queue_size; |
6ce69d1c | 1459 | vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN; |
967f97fa | 1460 | vdev->vq[i].handle_output = handle_output; |
344dc16f | 1461 | vdev->vq[i].handle_aio_output = NULL; |
967f97fa AL |
1462 | |
1463 | return &vdev->vq[i]; | |
1464 | } | |
1465 | ||
f23fd811 JW |
1466 | void virtio_del_queue(VirtIODevice *vdev, int n) |
1467 | { | |
87b3bd1c | 1468 | if (n < 0 || n >= VIRTIO_QUEUE_MAX) { |
f23fd811 JW |
1469 | abort(); |
1470 | } | |
1471 | ||
1472 | vdev->vq[n].vring.num = 0; | |
46c5d082 | 1473 | vdev->vq[n].vring.num_default = 0; |
f23fd811 JW |
1474 | } |
1475 | ||
0687c37c PB |
1476 | static void virtio_set_isr(VirtIODevice *vdev, int value) |
1477 | { | |
1478 | uint8_t old = atomic_read(&vdev->isr); | |
1479 | ||
1480 | /* Do not write ISR if it does not change, so that its cacheline remains | |
1481 | * shared in the common case where the guest does not read it. | |
1482 | */ | |
1483 | if ((old & value) != value) { | |
1484 | atomic_or(&vdev->isr, value); | |
1485 | } | |
1486 | } | |
1487 | ||
c25d97c4 | 1488 | static bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq) |
bcbabae8 MT |
1489 | { |
1490 | uint16_t old, new; | |
1491 | bool v; | |
a281ebc1 MT |
1492 | /* We need to expose used array entries before checking used event. */ |
1493 | smp_mb(); | |
97b83deb | 1494 | /* Always notify when queue is empty (when feature acknowledge) */ |
95129d6f | 1495 | if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) && |
be1fea9b | 1496 | !vq->inuse && virtio_queue_empty(vq)) { |
bcbabae8 MT |
1497 | return true; |
1498 | } | |
1499 | ||
95129d6f | 1500 | if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) { |
bcbabae8 MT |
1501 | return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT); |
1502 | } | |
1503 | ||
1504 | v = vq->signalled_used_valid; | |
1505 | vq->signalled_used_valid = true; | |
1506 | old = vq->signalled_used; | |
b796fcd1 | 1507 | new = vq->signalled_used = vq->used_idx; |
e9600c6c | 1508 | return !v || vring_need_event(vring_get_used_event(vq), new, old); |
bcbabae8 MT |
1509 | } |
1510 | ||
83d768b5 PB |
1511 | void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq) |
1512 | { | |
1513 | if (!virtio_should_notify(vdev, vq)) { | |
1514 | return; | |
1515 | } | |
1516 | ||
1517 | trace_virtio_notify_irqfd(vdev, vq); | |
1518 | ||
1519 | /* | |
1520 | * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but | |
1521 | * windows drivers included in virtio-win 1.8.0 (circa 2015) are | |
1522 | * incorrectly polling this bit during crashdump and hibernation | |
1523 | * in MSI mode, causing a hang if this bit is never updated. | |
1524 | * Recent releases of Windows do not really shut down, but rather | |
1525 | * log out and hibernate to make the next startup faster. Hence, | |
1526 | * this manifested as a more serious hang during shutdown with | |
1527 | * | |
1528 | * Next driver release from 2016 fixed this problem, so working around it | |
1529 | * is not a must, but it's easy to do so let's do it here. | |
1530 | * | |
1531 | * Note: it's safe to update ISR from any thread as it was switched | |
1532 | * to an atomic operation. | |
1533 | */ | |
1534 | virtio_set_isr(vq->vdev, 0x1); | |
1535 | event_notifier_set(&vq->guest_notifier); | |
1536 | } | |
1537 | ||
bcbabae8 MT |
1538 | void virtio_notify(VirtIODevice *vdev, VirtQueue *vq) |
1539 | { | |
adb3feda | 1540 | if (!virtio_should_notify(vdev, vq)) { |
967f97fa | 1541 | return; |
bcbabae8 | 1542 | } |
967f97fa | 1543 | |
64979a4d | 1544 | trace_virtio_notify(vdev, vq); |
0687c37c | 1545 | virtio_set_isr(vq->vdev, 0x1); |
7055e687 | 1546 | virtio_notify_vector(vdev, vq->vector); |
967f97fa AL |
1547 | } |
1548 | ||
1549 | void virtio_notify_config(VirtIODevice *vdev) | |
1550 | { | |
7625162c AL |
1551 | if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) |
1552 | return; | |
1553 | ||
0687c37c | 1554 | virtio_set_isr(vdev, 0x3); |
b8f05908 | 1555 | vdev->generation++; |
7055e687 | 1556 | virtio_notify_vector(vdev, vdev->config_vector); |
967f97fa AL |
1557 | } |
1558 | ||
616a6552 GK |
1559 | static bool virtio_device_endian_needed(void *opaque) |
1560 | { | |
1561 | VirtIODevice *vdev = opaque; | |
1562 | ||
1563 | assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN); | |
95129d6f | 1564 | if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { |
3c185597 CH |
1565 | return vdev->device_endian != virtio_default_endian(); |
1566 | } | |
1567 | /* Devices conforming to VIRTIO 1.0 or later are always LE. */ | |
1568 | return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE; | |
616a6552 GK |
1569 | } |
1570 | ||
019a3edb GH |
1571 | static bool virtio_64bit_features_needed(void *opaque) |
1572 | { | |
1573 | VirtIODevice *vdev = opaque; | |
1574 | ||
1575 | return (vdev->host_features >> 32) != 0; | |
1576 | } | |
1577 | ||
74aae7b2 JW |
1578 | static bool virtio_virtqueue_needed(void *opaque) |
1579 | { | |
1580 | VirtIODevice *vdev = opaque; | |
1581 | ||
1582 | return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1); | |
1583 | } | |
1584 | ||
46c5d082 CH |
1585 | static bool virtio_ringsize_needed(void *opaque) |
1586 | { | |
1587 | VirtIODevice *vdev = opaque; | |
1588 | int i; | |
1589 | ||
1590 | for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { | |
1591 | if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) { | |
1592 | return true; | |
1593 | } | |
1594 | } | |
1595 | return false; | |
1596 | } | |
1597 | ||
a6df8adf JW |
1598 | static bool virtio_extra_state_needed(void *opaque) |
1599 | { | |
1600 | VirtIODevice *vdev = opaque; | |
1601 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); | |
1602 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); | |
1603 | ||
1604 | return k->has_extra_state && | |
1605 | k->has_extra_state(qbus->parent); | |
1606 | } | |
1607 | ||
791b1daf SH |
1608 | static bool virtio_broken_needed(void *opaque) |
1609 | { | |
1610 | VirtIODevice *vdev = opaque; | |
1611 | ||
1612 | return vdev->broken; | |
1613 | } | |
1614 | ||
50e5ae4d | 1615 | static const VMStateDescription vmstate_virtqueue = { |
74aae7b2 | 1616 | .name = "virtqueue_state", |
50e5ae4d DDAG |
1617 | .version_id = 1, |
1618 | .minimum_version_id = 1, | |
1619 | .fields = (VMStateField[]) { | |
1620 | VMSTATE_UINT64(vring.avail, struct VirtQueue), | |
1621 | VMSTATE_UINT64(vring.used, struct VirtQueue), | |
1622 | VMSTATE_END_OF_LIST() | |
1623 | } | |
74aae7b2 JW |
1624 | }; |
1625 | ||
1626 | static const VMStateDescription vmstate_virtio_virtqueues = { | |
1627 | .name = "virtio/virtqueues", | |
1628 | .version_id = 1, | |
1629 | .minimum_version_id = 1, | |
1630 | .needed = &virtio_virtqueue_needed, | |
1631 | .fields = (VMStateField[]) { | |
3e996cc5 DDAG |
1632 | VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice, |
1633 | VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue), | |
74aae7b2 JW |
1634 | VMSTATE_END_OF_LIST() |
1635 | } | |
1636 | }; | |
1637 | ||
50e5ae4d | 1638 | static const VMStateDescription vmstate_ringsize = { |
46c5d082 | 1639 | .name = "ringsize_state", |
50e5ae4d DDAG |
1640 | .version_id = 1, |
1641 | .minimum_version_id = 1, | |
1642 | .fields = (VMStateField[]) { | |
1643 | VMSTATE_UINT32(vring.num_default, struct VirtQueue), | |
1644 | VMSTATE_END_OF_LIST() | |
1645 | } | |
46c5d082 CH |
1646 | }; |
1647 | ||
1648 | static const VMStateDescription vmstate_virtio_ringsize = { | |
1649 | .name = "virtio/ringsize", | |
1650 | .version_id = 1, | |
1651 | .minimum_version_id = 1, | |
1652 | .needed = &virtio_ringsize_needed, | |
1653 | .fields = (VMStateField[]) { | |
3e996cc5 DDAG |
1654 | VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice, |
1655 | VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue), | |
46c5d082 CH |
1656 | VMSTATE_END_OF_LIST() |
1657 | } | |
1658 | }; | |
1659 | ||
2c21ee76 JD |
1660 | static int get_extra_state(QEMUFile *f, void *pv, size_t size, |
1661 | VMStateField *field) | |
a6df8adf JW |
1662 | { |
1663 | VirtIODevice *vdev = pv; | |
1664 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); | |
1665 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); | |
1666 | ||
1667 | if (!k->load_extra_state) { | |
1668 | return -1; | |
1669 | } else { | |
1670 | return k->load_extra_state(qbus->parent, f); | |
1671 | } | |
1672 | } | |
1673 | ||
2c21ee76 JD |
1674 | static int put_extra_state(QEMUFile *f, void *pv, size_t size, |
1675 | VMStateField *field, QJSON *vmdesc) | |
a6df8adf JW |
1676 | { |
1677 | VirtIODevice *vdev = pv; | |
1678 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); | |
1679 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); | |
1680 | ||
1681 | k->save_extra_state(qbus->parent, f); | |
2c21ee76 | 1682 | return 0; |
a6df8adf JW |
1683 | } |
1684 | ||
1685 | static const VMStateInfo vmstate_info_extra_state = { | |
1686 | .name = "virtqueue_extra_state", | |
1687 | .get = get_extra_state, | |
1688 | .put = put_extra_state, | |
1689 | }; | |
1690 | ||
1691 | static const VMStateDescription vmstate_virtio_extra_state = { | |
1692 | .name = "virtio/extra_state", | |
1693 | .version_id = 1, | |
1694 | .minimum_version_id = 1, | |
1695 | .needed = &virtio_extra_state_needed, | |
1696 | .fields = (VMStateField[]) { | |
1697 | { | |
1698 | .name = "extra_state", | |
1699 | .version_id = 0, | |
1700 | .field_exists = NULL, | |
1701 | .size = 0, | |
1702 | .info = &vmstate_info_extra_state, | |
1703 | .flags = VMS_SINGLE, | |
1704 | .offset = 0, | |
1705 | }, | |
1706 | VMSTATE_END_OF_LIST() | |
1707 | } | |
1708 | }; | |
1709 | ||
616a6552 GK |
1710 | static const VMStateDescription vmstate_virtio_device_endian = { |
1711 | .name = "virtio/device_endian", | |
1712 | .version_id = 1, | |
1713 | .minimum_version_id = 1, | |
5cd8cada | 1714 | .needed = &virtio_device_endian_needed, |
616a6552 GK |
1715 | .fields = (VMStateField[]) { |
1716 | VMSTATE_UINT8(device_endian, VirtIODevice), | |
1717 | VMSTATE_END_OF_LIST() | |
1718 | } | |
1719 | }; | |
1720 | ||
019a3edb GH |
1721 | static const VMStateDescription vmstate_virtio_64bit_features = { |
1722 | .name = "virtio/64bit_features", | |
1723 | .version_id = 1, | |
1724 | .minimum_version_id = 1, | |
5cd8cada | 1725 | .needed = &virtio_64bit_features_needed, |
019a3edb GH |
1726 | .fields = (VMStateField[]) { |
1727 | VMSTATE_UINT64(guest_features, VirtIODevice), | |
1728 | VMSTATE_END_OF_LIST() | |
1729 | } | |
1730 | }; | |
1731 | ||
791b1daf SH |
1732 | static const VMStateDescription vmstate_virtio_broken = { |
1733 | .name = "virtio/broken", | |
1734 | .version_id = 1, | |
1735 | .minimum_version_id = 1, | |
1736 | .needed = &virtio_broken_needed, | |
1737 | .fields = (VMStateField[]) { | |
1738 | VMSTATE_BOOL(broken, VirtIODevice), | |
1739 | VMSTATE_END_OF_LIST() | |
1740 | } | |
1741 | }; | |
1742 | ||
6b321a3d GK |
1743 | static const VMStateDescription vmstate_virtio = { |
1744 | .name = "virtio", | |
1745 | .version_id = 1, | |
1746 | .minimum_version_id = 1, | |
1747 | .minimum_version_id_old = 1, | |
1748 | .fields = (VMStateField[]) { | |
1749 | VMSTATE_END_OF_LIST() | |
616a6552 | 1750 | }, |
5cd8cada JQ |
1751 | .subsections = (const VMStateDescription*[]) { |
1752 | &vmstate_virtio_device_endian, | |
1753 | &vmstate_virtio_64bit_features, | |
74aae7b2 | 1754 | &vmstate_virtio_virtqueues, |
46c5d082 | 1755 | &vmstate_virtio_ringsize, |
791b1daf | 1756 | &vmstate_virtio_broken, |
a6df8adf | 1757 | &vmstate_virtio_extra_state, |
5cd8cada | 1758 | NULL |
6b321a3d GK |
1759 | } |
1760 | }; | |
1761 | ||
967f97fa AL |
1762 | void virtio_save(VirtIODevice *vdev, QEMUFile *f) |
1763 | { | |
1c819449 FK |
1764 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); |
1765 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); | |
1b5fc0de | 1766 | VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev); |
019a3edb | 1767 | uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff); |
967f97fa AL |
1768 | int i; |
1769 | ||
1c819449 FK |
1770 | if (k->save_config) { |
1771 | k->save_config(qbus->parent, f); | |
1772 | } | |
967f97fa | 1773 | |
967f97fa AL |
1774 | qemu_put_8s(f, &vdev->status); |
1775 | qemu_put_8s(f, &vdev->isr); | |
1776 | qemu_put_be16s(f, &vdev->queue_sel); | |
019a3edb | 1777 | qemu_put_be32s(f, &guest_features_lo); |
967f97fa AL |
1778 | qemu_put_be32(f, vdev->config_len); |
1779 | qemu_put_buffer(f, vdev->config, vdev->config_len); | |
1780 | ||
87b3bd1c | 1781 | for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { |
967f97fa AL |
1782 | if (vdev->vq[i].vring.num == 0) |
1783 | break; | |
1784 | } | |
1785 | ||
1786 | qemu_put_be32(f, i); | |
1787 | ||
87b3bd1c | 1788 | for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { |
967f97fa AL |
1789 | if (vdev->vq[i].vring.num == 0) |
1790 | break; | |
1791 | ||
1792 | qemu_put_be32(f, vdev->vq[i].vring.num); | |
6ce69d1c PM |
1793 | if (k->has_variable_vring_alignment) { |
1794 | qemu_put_be32(f, vdev->vq[i].vring.align); | |
1795 | } | |
ab223c95 CH |
1796 | /* XXX virtio-1 devices */ |
1797 | qemu_put_be64(f, vdev->vq[i].vring.desc); | |
967f97fa | 1798 | qemu_put_be16s(f, &vdev->vq[i].last_avail_idx); |
1c819449 FK |
1799 | if (k->save_queue) { |
1800 | k->save_queue(qbus->parent, i, f); | |
1801 | } | |
967f97fa | 1802 | } |
1b5fc0de GK |
1803 | |
1804 | if (vdc->save != NULL) { | |
1805 | vdc->save(vdev, f); | |
1806 | } | |
6b321a3d | 1807 | |
ea43e259 DDAG |
1808 | if (vdc->vmsd) { |
1809 | vmstate_save_state(f, vdc->vmsd, vdev, NULL); | |
1810 | } | |
1811 | ||
6b321a3d | 1812 | /* Subsections */ |
8118f095 | 1813 | vmstate_save_state(f, &vmstate_virtio, vdev, NULL); |
967f97fa AL |
1814 | } |
1815 | ||
1a665855 | 1816 | /* A wrapper for use as a VMState .put function */ |
2c21ee76 JD |
1817 | static int virtio_device_put(QEMUFile *f, void *opaque, size_t size, |
1818 | VMStateField *field, QJSON *vmdesc) | |
1a665855 HP |
1819 | { |
1820 | virtio_save(VIRTIO_DEVICE(opaque), f); | |
2c21ee76 JD |
1821 | |
1822 | return 0; | |
1a665855 HP |
1823 | } |
1824 | ||
1825 | /* A wrapper for use as a VMState .get function */ | |
2c21ee76 JD |
1826 | static int virtio_device_get(QEMUFile *f, void *opaque, size_t size, |
1827 | VMStateField *field) | |
1a665855 HP |
1828 | { |
1829 | VirtIODevice *vdev = VIRTIO_DEVICE(opaque); | |
1830 | DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev)); | |
1831 | ||
1832 | return virtio_load(vdev, f, dc->vmsd->version_id); | |
1833 | } | |
1834 | ||
1835 | const VMStateInfo virtio_vmstate_info = { | |
1836 | .name = "virtio", | |
1837 | .get = virtio_device_get, | |
1838 | .put = virtio_device_put, | |
1839 | }; | |
1840 | ||
6c0196d7 | 1841 | static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val) |
ad0c9332 | 1842 | { |
181103cd | 1843 | VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); |
6b8f1020 | 1844 | bool bad = (val & ~(vdev->host_features)) != 0; |
ad0c9332 | 1845 | |
6b8f1020 | 1846 | val &= vdev->host_features; |
181103cd FK |
1847 | if (k->set_features) { |
1848 | k->set_features(vdev, val); | |
ad0c9332 PB |
1849 | } |
1850 | vdev->guest_features = val; | |
1851 | return bad ? -1 : 0; | |
1852 | } | |
1853 | ||
6c0196d7 CH |
1854 | int virtio_set_features(VirtIODevice *vdev, uint64_t val) |
1855 | { | |
1856 | /* | |
1857 | * The driver must not attempt to set features after feature negotiation | |
1858 | * has finished. | |
1859 | */ | |
1860 | if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) { | |
1861 | return -EINVAL; | |
1862 | } | |
1863 | return virtio_set_features_nocheck(vdev, val); | |
1864 | } | |
1865 | ||
1b5fc0de | 1866 | int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id) |
967f97fa | 1867 | { |
cc459952 | 1868 | int i, ret; |
a890a2f9 | 1869 | int32_t config_len; |
cc459952 | 1870 | uint32_t num; |
6d74ca5a | 1871 | uint32_t features; |
1c819449 FK |
1872 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); |
1873 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); | |
1b5fc0de | 1874 | VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev); |
967f97fa | 1875 | |
616a6552 GK |
1876 | /* |
1877 | * We poison the endianness to ensure it does not get used before | |
1878 | * subsections have been loaded. | |
1879 | */ | |
1880 | vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN; | |
1881 | ||
1c819449 FK |
1882 | if (k->load_config) { |
1883 | ret = k->load_config(qbus->parent, f); | |
ff24bd58 MT |
1884 | if (ret) |
1885 | return ret; | |
1886 | } | |
967f97fa | 1887 | |
967f97fa AL |
1888 | qemu_get_8s(f, &vdev->status); |
1889 | qemu_get_8s(f, &vdev->isr); | |
1890 | qemu_get_be16s(f, &vdev->queue_sel); | |
87b3bd1c | 1891 | if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) { |
4b53c2c7 MR |
1892 | return -1; |
1893 | } | |
6d74ca5a | 1894 | qemu_get_be32s(f, &features); |
ad0c9332 | 1895 | |
62cee1a2 MT |
1896 | /* |
1897 | * Temporarily set guest_features low bits - needed by | |
1898 | * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS | |
1899 | * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ. | |
1900 | * | |
1901 | * Note: devices should always test host features in future - don't create | |
1902 | * new dependencies like this. | |
1903 | */ | |
1904 | vdev->guest_features = features; | |
1905 | ||
a890a2f9 | 1906 | config_len = qemu_get_be32(f); |
2f5732e9 DDAG |
1907 | |
1908 | /* | |
1909 | * There are cases where the incoming config can be bigger or smaller | |
1910 | * than what we have; so load what we have space for, and skip | |
1911 | * any excess that's in the stream. | |
1912 | */ | |
1913 | qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len)); | |
1914 | ||
1915 | while (config_len > vdev->config_len) { | |
1916 | qemu_get_byte(f); | |
1917 | config_len--; | |
a890a2f9 | 1918 | } |
967f97fa AL |
1919 | |
1920 | num = qemu_get_be32(f); | |
1921 | ||
87b3bd1c | 1922 | if (num > VIRTIO_QUEUE_MAX) { |
8a1be662 | 1923 | error_report("Invalid number of virtqueues: 0x%x", num); |
cc459952 MT |
1924 | return -1; |
1925 | } | |
1926 | ||
967f97fa AL |
1927 | for (i = 0; i < num; i++) { |
1928 | vdev->vq[i].vring.num = qemu_get_be32(f); | |
6ce69d1c PM |
1929 | if (k->has_variable_vring_alignment) { |
1930 | vdev->vq[i].vring.align = qemu_get_be32(f); | |
1931 | } | |
ab223c95 | 1932 | vdev->vq[i].vring.desc = qemu_get_be64(f); |
967f97fa | 1933 | qemu_get_be16s(f, &vdev->vq[i].last_avail_idx); |
bcbabae8 | 1934 | vdev->vq[i].signalled_used_valid = false; |
332fa82d | 1935 | vdev->vq[i].notification = true; |
967f97fa | 1936 | |
ab223c95 CH |
1937 | if (vdev->vq[i].vring.desc) { |
1938 | /* XXX virtio-1 devices */ | |
1939 | virtio_queue_update_rings(vdev, i); | |
1abeb5a6 MT |
1940 | } else if (vdev->vq[i].last_avail_idx) { |
1941 | error_report("VQ %d address 0x0 " | |
6daf194d | 1942 | "inconsistent with Host index 0x%x", |
1abeb5a6 MT |
1943 | i, vdev->vq[i].last_avail_idx); |
1944 | return -1; | |
8275e2f6 | 1945 | } |
1c819449 FK |
1946 | if (k->load_queue) { |
1947 | ret = k->load_queue(qbus->parent, i, f); | |
ff24bd58 MT |
1948 | if (ret) |
1949 | return ret; | |
7055e687 | 1950 | } |
967f97fa AL |
1951 | } |
1952 | ||
7055e687 | 1953 | virtio_notify_vector(vdev, VIRTIO_NO_VECTOR); |
1b5fc0de GK |
1954 | |
1955 | if (vdc->load != NULL) { | |
6b321a3d GK |
1956 | ret = vdc->load(vdev, f, version_id); |
1957 | if (ret) { | |
1958 | return ret; | |
1959 | } | |
1b5fc0de GK |
1960 | } |
1961 | ||
ea43e259 DDAG |
1962 | if (vdc->vmsd) { |
1963 | ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id); | |
1964 | if (ret) { | |
1965 | return ret; | |
1966 | } | |
1967 | } | |
1968 | ||
616a6552 GK |
1969 | /* Subsections */ |
1970 | ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1); | |
1971 | if (ret) { | |
1972 | return ret; | |
1973 | } | |
1974 | ||
1975 | if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) { | |
1976 | vdev->device_endian = virtio_default_endian(); | |
1977 | } | |
1978 | ||
019a3edb GH |
1979 | if (virtio_64bit_features_needed(vdev)) { |
1980 | /* | |
1981 | * Subsection load filled vdev->guest_features. Run them | |
1982 | * through virtio_set_features to sanity-check them against | |
1983 | * host_features. | |
1984 | */ | |
1985 | uint64_t features64 = vdev->guest_features; | |
6c0196d7 | 1986 | if (virtio_set_features_nocheck(vdev, features64) < 0) { |
019a3edb GH |
1987 | error_report("Features 0x%" PRIx64 " unsupported. " |
1988 | "Allowed features: 0x%" PRIx64, | |
1989 | features64, vdev->host_features); | |
1990 | return -1; | |
1991 | } | |
1992 | } else { | |
6c0196d7 | 1993 | if (virtio_set_features_nocheck(vdev, features) < 0) { |
019a3edb GH |
1994 | error_report("Features 0x%x unsupported. " |
1995 | "Allowed features: 0x%" PRIx64, | |
1996 | features, vdev->host_features); | |
1997 | return -1; | |
1998 | } | |
1999 | } | |
2000 | ||
616a6552 | 2001 | for (i = 0; i < num; i++) { |
ab223c95 | 2002 | if (vdev->vq[i].vring.desc) { |
616a6552 GK |
2003 | uint16_t nheads; |
2004 | nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx; | |
2005 | /* Check it isn't doing strange things with descriptor numbers. */ | |
2006 | if (nheads > vdev->vq[i].vring.num) { | |
2007 | error_report("VQ %d size 0x%x Guest index 0x%x " | |
2008 | "inconsistent with Host index 0x%x: delta 0x%x", | |
2009 | i, vdev->vq[i].vring.num, | |
2010 | vring_avail_idx(&vdev->vq[i]), | |
2011 | vdev->vq[i].last_avail_idx, nheads); | |
2012 | return -1; | |
2013 | } | |
b796fcd1 | 2014 | vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]); |
be1fea9b | 2015 | vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]); |
bccdef6b SH |
2016 | |
2017 | /* | |
2018 | * Some devices migrate VirtQueueElements that have been popped | |
2019 | * from the avail ring but not yet returned to the used ring. | |
e66bcc40 HP |
2020 | * Since max ring size < UINT16_MAX it's safe to use modulo |
2021 | * UINT16_MAX + 1 subtraction. | |
bccdef6b | 2022 | */ |
e66bcc40 HP |
2023 | vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx - |
2024 | vdev->vq[i].used_idx); | |
bccdef6b SH |
2025 | if (vdev->vq[i].inuse > vdev->vq[i].vring.num) { |
2026 | error_report("VQ %d size 0x%x < last_avail_idx 0x%x - " | |
2027 | "used_idx 0x%x", | |
2028 | i, vdev->vq[i].vring.num, | |
2029 | vdev->vq[i].last_avail_idx, | |
2030 | vdev->vq[i].used_idx); | |
2031 | return -1; | |
2032 | } | |
616a6552 GK |
2033 | } |
2034 | } | |
2035 | ||
2036 | return 0; | |
967f97fa AL |
2037 | } |
2038 | ||
6a1a8cc7 | 2039 | void virtio_cleanup(VirtIODevice *vdev) |
b946a153 | 2040 | { |
85cf2a8d | 2041 | qemu_del_vm_change_state_handler(vdev->vmstate); |
8e05db92 FK |
2042 | } |
2043 | ||
1dfb4dd9 | 2044 | static void virtio_vmstate_change(void *opaque, int running, RunState state) |
85cf2a8d MT |
2045 | { |
2046 | VirtIODevice *vdev = opaque; | |
1c819449 FK |
2047 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); |
2048 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); | |
85cf2a8d | 2049 | bool backend_run = running && (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK); |
9e8e8c48 | 2050 | vdev->vm_running = running; |
85cf2a8d MT |
2051 | |
2052 | if (backend_run) { | |
2053 | virtio_set_status(vdev, vdev->status); | |
2054 | } | |
2055 | ||
1c819449 FK |
2056 | if (k->vmstate_change) { |
2057 | k->vmstate_change(qbus->parent, backend_run); | |
85cf2a8d MT |
2058 | } |
2059 | ||
2060 | if (!backend_run) { | |
2061 | virtio_set_status(vdev, vdev->status); | |
2062 | } | |
2063 | } | |
2064 | ||
c8075caf GA |
2065 | void virtio_instance_init_common(Object *proxy_obj, void *data, |
2066 | size_t vdev_size, const char *vdev_name) | |
2067 | { | |
2068 | DeviceState *vdev = data; | |
2069 | ||
2070 | object_initialize(vdev, vdev_size, vdev_name); | |
2071 | object_property_add_child(proxy_obj, "virtio-backend", OBJECT(vdev), NULL); | |
2072 | object_unref(OBJECT(vdev)); | |
2073 | qdev_alias_all_properties(vdev, proxy_obj); | |
2074 | } | |
2075 | ||
8e05db92 FK |
2076 | void virtio_init(VirtIODevice *vdev, const char *name, |
2077 | uint16_t device_id, size_t config_size) | |
967f97fa | 2078 | { |
e0d686bf JW |
2079 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); |
2080 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); | |
b8193adb | 2081 | int i; |
e0d686bf JW |
2082 | int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0; |
2083 | ||
2084 | if (nvectors) { | |
2085 | vdev->vector_queues = | |
2086 | g_malloc0(sizeof(*vdev->vector_queues) * nvectors); | |
2087 | } | |
2088 | ||
53c25cea | 2089 | vdev->device_id = device_id; |
967f97fa | 2090 | vdev->status = 0; |
0687c37c | 2091 | atomic_set(&vdev->isr, 0); |
967f97fa | 2092 | vdev->queue_sel = 0; |
7055e687 | 2093 | vdev->config_vector = VIRTIO_NO_VECTOR; |
87b3bd1c | 2094 | vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_QUEUE_MAX); |
1354869c | 2095 | vdev->vm_running = runstate_is_running(); |
f5ed3663 | 2096 | vdev->broken = false; |
87b3bd1c | 2097 | for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { |
b8193adb | 2098 | vdev->vq[i].vector = VIRTIO_NO_VECTOR; |
1cbdabe2 | 2099 | vdev->vq[i].vdev = vdev; |
e78a2b42 | 2100 | vdev->vq[i].queue_index = i; |
1cbdabe2 | 2101 | } |
967f97fa | 2102 | |
967f97fa AL |
2103 | vdev->name = name; |
2104 | vdev->config_len = config_size; | |
8e05db92 | 2105 | if (vdev->config_len) { |
7267c094 | 2106 | vdev->config = g_malloc0(config_size); |
8e05db92 | 2107 | } else { |
967f97fa | 2108 | vdev->config = NULL; |
8e05db92 FK |
2109 | } |
2110 | vdev->vmstate = qemu_add_vm_change_state_handler(virtio_vmstate_change, | |
2111 | vdev); | |
616a6552 | 2112 | vdev->device_endian = virtio_default_endian(); |
5669655a | 2113 | vdev->use_guest_notifier_mask = true; |
8e05db92 | 2114 | } |
967f97fa | 2115 | |
a8170e5e | 2116 | hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n) |
1cbdabe2 MT |
2117 | { |
2118 | return vdev->vq[n].vring.desc; | |
2119 | } | |
2120 | ||
a8170e5e | 2121 | hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n) |
1cbdabe2 MT |
2122 | { |
2123 | return vdev->vq[n].vring.avail; | |
2124 | } | |
2125 | ||
a8170e5e | 2126 | hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n) |
1cbdabe2 MT |
2127 | { |
2128 | return vdev->vq[n].vring.used; | |
2129 | } | |
2130 | ||
a8170e5e | 2131 | hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n) |
1cbdabe2 MT |
2132 | { |
2133 | return sizeof(VRingDesc) * vdev->vq[n].vring.num; | |
2134 | } | |
2135 | ||
a8170e5e | 2136 | hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n) |
1cbdabe2 MT |
2137 | { |
2138 | return offsetof(VRingAvail, ring) + | |
50764fc8 | 2139 | sizeof(uint16_t) * vdev->vq[n].vring.num; |
1cbdabe2 MT |
2140 | } |
2141 | ||
a8170e5e | 2142 | hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n) |
1cbdabe2 MT |
2143 | { |
2144 | return offsetof(VRingUsed, ring) + | |
2145 | sizeof(VRingUsedElem) * vdev->vq[n].vring.num; | |
2146 | } | |
2147 | ||
1cbdabe2 MT |
2148 | uint16_t virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n) |
2149 | { | |
2150 | return vdev->vq[n].last_avail_idx; | |
2151 | } | |
2152 | ||
2153 | void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n, uint16_t idx) | |
2154 | { | |
2155 | vdev->vq[n].last_avail_idx = idx; | |
be1fea9b | 2156 | vdev->vq[n].shadow_avail_idx = idx; |
1cbdabe2 MT |
2157 | } |
2158 | ||
312d3b35 YB |
2159 | void virtio_queue_update_used_idx(VirtIODevice *vdev, int n) |
2160 | { | |
2161 | vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]); | |
2162 | } | |
2163 | ||
6793dfd1 SH |
2164 | void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n) |
2165 | { | |
2166 | vdev->vq[n].signalled_used_valid = false; | |
2167 | } | |
2168 | ||
1cbdabe2 MT |
2169 | VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n) |
2170 | { | |
2171 | return vdev->vq + n; | |
2172 | } | |
2173 | ||
e78a2b42 JW |
2174 | uint16_t virtio_get_queue_index(VirtQueue *vq) |
2175 | { | |
2176 | return vq->queue_index; | |
2177 | } | |
2178 | ||
15b2bd18 PB |
2179 | static void virtio_queue_guest_notifier_read(EventNotifier *n) |
2180 | { | |
2181 | VirtQueue *vq = container_of(n, VirtQueue, guest_notifier); | |
2182 | if (event_notifier_test_and_clear(n)) { | |
83d768b5 | 2183 | virtio_notify_vector(vq->vdev, vq->vector); |
15b2bd18 PB |
2184 | } |
2185 | } | |
2186 | ||
2187 | void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign, | |
2188 | bool with_irqfd) | |
2189 | { | |
2190 | if (assign && !with_irqfd) { | |
d6da1e9e | 2191 | event_notifier_set_handler(&vq->guest_notifier, |
15b2bd18 PB |
2192 | virtio_queue_guest_notifier_read); |
2193 | } else { | |
d6da1e9e | 2194 | event_notifier_set_handler(&vq->guest_notifier, NULL); |
15b2bd18 PB |
2195 | } |
2196 | if (!assign) { | |
2197 | /* Test and clear notifier before closing it, | |
2198 | * in case poll callback didn't have time to run. */ | |
2199 | virtio_queue_guest_notifier_read(&vq->guest_notifier); | |
2200 | } | |
2201 | } | |
2202 | ||
1cbdabe2 MT |
2203 | EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq) |
2204 | { | |
2205 | return &vq->guest_notifier; | |
2206 | } | |
b1f416aa | 2207 | |
344dc16f | 2208 | static void virtio_queue_host_notifier_aio_read(EventNotifier *n) |
b1f416aa PB |
2209 | { |
2210 | VirtQueue *vq = container_of(n, VirtQueue, host_notifier); | |
2211 | if (event_notifier_test_and_clear(n)) { | |
344dc16f | 2212 | virtio_queue_notify_aio_vq(vq); |
b1f416aa PB |
2213 | } |
2214 | } | |
2215 | ||
a7c8215e SH |
2216 | static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n) |
2217 | { | |
2218 | VirtQueue *vq = container_of(n, VirtQueue, host_notifier); | |
2219 | ||
2220 | virtio_queue_set_notification(vq, 0); | |
2221 | } | |
2222 | ||
0062ea0f SH |
2223 | static bool virtio_queue_host_notifier_aio_poll(void *opaque) |
2224 | { | |
2225 | EventNotifier *n = opaque; | |
2226 | VirtQueue *vq = container_of(n, VirtQueue, host_notifier); | |
07931698 | 2227 | bool progress; |
0062ea0f SH |
2228 | |
2229 | if (virtio_queue_empty(vq)) { | |
2230 | return false; | |
2231 | } | |
2232 | ||
07931698 | 2233 | progress = virtio_queue_notify_aio_vq(vq); |
1448c133 SH |
2234 | |
2235 | /* In case the handler function re-enabled notifications */ | |
2236 | virtio_queue_set_notification(vq, 0); | |
07931698 | 2237 | return progress; |
0062ea0f SH |
2238 | } |
2239 | ||
a7c8215e SH |
2240 | static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n) |
2241 | { | |
2242 | VirtQueue *vq = container_of(n, VirtQueue, host_notifier); | |
2243 | ||
2244 | /* Caller polls once more after this to catch requests that race with us */ | |
2245 | virtio_queue_set_notification(vq, 1); | |
2246 | } | |
2247 | ||
a1afb606 | 2248 | void virtio_queue_aio_set_host_notifier_handler(VirtQueue *vq, AioContext *ctx, |
07931698 | 2249 | VirtIOHandleAIOOutput handle_output) |
a1afb606 | 2250 | { |
a378b49a PB |
2251 | if (handle_output) { |
2252 | vq->handle_aio_output = handle_output; | |
a1afb606 | 2253 | aio_set_event_notifier(ctx, &vq->host_notifier, true, |
0062ea0f SH |
2254 | virtio_queue_host_notifier_aio_read, |
2255 | virtio_queue_host_notifier_aio_poll); | |
a7c8215e SH |
2256 | aio_set_event_notifier_poll(ctx, &vq->host_notifier, |
2257 | virtio_queue_host_notifier_aio_poll_begin, | |
2258 | virtio_queue_host_notifier_aio_poll_end); | |
a1afb606 | 2259 | } else { |
f6a51c84 | 2260 | aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL, NULL); |
a1afb606 PB |
2261 | /* Test and clear notifier before after disabling event, |
2262 | * in case poll callback didn't have time to run. */ | |
344dc16f | 2263 | virtio_queue_host_notifier_aio_read(&vq->host_notifier); |
a378b49a | 2264 | vq->handle_aio_output = NULL; |
344dc16f MT |
2265 | } |
2266 | } | |
2267 | ||
fa283a4a | 2268 | void virtio_queue_host_notifier_read(EventNotifier *n) |
344dc16f MT |
2269 | { |
2270 | VirtQueue *vq = container_of(n, VirtQueue, host_notifier); | |
2271 | if (event_notifier_test_and_clear(n)) { | |
2272 | virtio_queue_notify_vq(vq); | |
a1afb606 PB |
2273 | } |
2274 | } | |
2275 | ||
1cbdabe2 MT |
2276 | EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq) |
2277 | { | |
2278 | return &vq->host_notifier; | |
2279 | } | |
8e05db92 | 2280 | |
1034e9cf FK |
2281 | void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name) |
2282 | { | |
9e288406 | 2283 | g_free(vdev->bus_name); |
80e0090a | 2284 | vdev->bus_name = g_strdup(bus_name); |
1034e9cf FK |
2285 | } |
2286 | ||
f5ed3663 SH |
2287 | void GCC_FMT_ATTR(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...) |
2288 | { | |
2289 | va_list ap; | |
2290 | ||
2291 | va_start(ap, fmt); | |
2292 | error_vreport(fmt, ap); | |
2293 | va_end(ap); | |
2294 | ||
2295 | vdev->broken = true; | |
2296 | ||
2297 | if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { | |
2298 | virtio_set_status(vdev, vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET); | |
2299 | virtio_notify_config(vdev); | |
2300 | } | |
2301 | } | |
2302 | ||
c611c764 PB |
2303 | static void virtio_memory_listener_commit(MemoryListener *listener) |
2304 | { | |
2305 | VirtIODevice *vdev = container_of(listener, VirtIODevice, listener); | |
2306 | int i; | |
2307 | ||
2308 | for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { | |
2309 | if (vdev->vq[i].vring.num == 0) { | |
2310 | break; | |
2311 | } | |
2312 | virtio_init_region_cache(vdev, i); | |
2313 | } | |
2314 | } | |
2315 | ||
1d244b42 AF |
2316 | static void virtio_device_realize(DeviceState *dev, Error **errp) |
2317 | { | |
2318 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); | |
2319 | VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev); | |
2320 | Error *err = NULL; | |
2321 | ||
ea43e259 DDAG |
2322 | /* Devices should either use vmsd or the load/save methods */ |
2323 | assert(!vdc->vmsd || !vdc->load); | |
2324 | ||
1d244b42 AF |
2325 | if (vdc->realize != NULL) { |
2326 | vdc->realize(dev, &err); | |
2327 | if (err != NULL) { | |
2328 | error_propagate(errp, err); | |
2329 | return; | |
2330 | } | |
8e05db92 | 2331 | } |
e8398045 JW |
2332 | |
2333 | virtio_bus_device_plugged(vdev, &err); | |
2334 | if (err != NULL) { | |
2335 | error_propagate(errp, err); | |
2336 | return; | |
2337 | } | |
c611c764 PB |
2338 | |
2339 | vdev->listener.commit = virtio_memory_listener_commit; | |
2340 | memory_listener_register(&vdev->listener, vdev->dma_as); | |
8e05db92 FK |
2341 | } |
2342 | ||
1d244b42 | 2343 | static void virtio_device_unrealize(DeviceState *dev, Error **errp) |
1034e9cf | 2344 | { |
1d244b42 | 2345 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
306ec6c3 AF |
2346 | VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev); |
2347 | Error *err = NULL; | |
1d244b42 | 2348 | |
83d07047 PB |
2349 | virtio_bus_device_unplugged(vdev); |
2350 | ||
306ec6c3 AF |
2351 | if (vdc->unrealize != NULL) { |
2352 | vdc->unrealize(dev, &err); | |
2353 | if (err != NULL) { | |
2354 | error_propagate(errp, err); | |
2355 | return; | |
2356 | } | |
5e96f5d2 | 2357 | } |
1d244b42 | 2358 | |
9e288406 MA |
2359 | g_free(vdev->bus_name); |
2360 | vdev->bus_name = NULL; | |
1034e9cf FK |
2361 | } |
2362 | ||
c611c764 PB |
2363 | static void virtio_device_free_virtqueues(VirtIODevice *vdev) |
2364 | { | |
2365 | int i; | |
2366 | if (!vdev->vq) { | |
2367 | return; | |
2368 | } | |
2369 | ||
2370 | for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { | |
2371 | VRingMemoryRegionCaches *caches; | |
2372 | if (vdev->vq[i].vring.num == 0) { | |
2373 | break; | |
2374 | } | |
2375 | caches = atomic_read(&vdev->vq[i].vring.caches); | |
2376 | atomic_set(&vdev->vq[i].vring.caches, NULL); | |
2377 | virtio_free_region_cache(caches); | |
2378 | } | |
2379 | g_free(vdev->vq); | |
2380 | } | |
2381 | ||
2382 | static void virtio_device_instance_finalize(Object *obj) | |
2383 | { | |
2384 | VirtIODevice *vdev = VIRTIO_DEVICE(obj); | |
2385 | ||
2386 | memory_listener_unregister(&vdev->listener); | |
2387 | virtio_device_free_virtqueues(vdev); | |
2388 | ||
2389 | g_free(vdev->config); | |
2390 | g_free(vdev->vector_queues); | |
2391 | } | |
2392 | ||
6b8f1020 CH |
2393 | static Property virtio_properties[] = { |
2394 | DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features), | |
2395 | DEFINE_PROP_END_OF_LIST(), | |
2396 | }; | |
2397 | ||
ff4c07df PB |
2398 | static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev) |
2399 | { | |
2400 | VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev))); | |
ff4c07df PB |
2401 | int n, r, err; |
2402 | ||
2403 | for (n = 0; n < VIRTIO_QUEUE_MAX; n++) { | |
fa283a4a | 2404 | VirtQueue *vq = &vdev->vq[n]; |
ff4c07df PB |
2405 | if (!virtio_queue_get_num(vdev, n)) { |
2406 | continue; | |
2407 | } | |
ed08a2a0 | 2408 | r = virtio_bus_set_host_notifier(qbus, n, true); |
ff4c07df PB |
2409 | if (r < 0) { |
2410 | err = r; | |
2411 | goto assign_error; | |
2412 | } | |
d6da1e9e | 2413 | event_notifier_set_handler(&vq->host_notifier, |
fa283a4a | 2414 | virtio_queue_host_notifier_read); |
6019f3b9 PB |
2415 | } |
2416 | ||
2417 | for (n = 0; n < VIRTIO_QUEUE_MAX; n++) { | |
2418 | /* Kick right away to begin processing requests already in vring */ | |
2419 | VirtQueue *vq = &vdev->vq[n]; | |
2420 | if (!vq->vring.num) { | |
2421 | continue; | |
2422 | } | |
2423 | event_notifier_set(&vq->host_notifier); | |
ff4c07df PB |
2424 | } |
2425 | return 0; | |
2426 | ||
2427 | assign_error: | |
2428 | while (--n >= 0) { | |
fa283a4a | 2429 | VirtQueue *vq = &vdev->vq[n]; |
ff4c07df PB |
2430 | if (!virtio_queue_get_num(vdev, n)) { |
2431 | continue; | |
2432 | } | |
2433 | ||
d6da1e9e | 2434 | event_notifier_set_handler(&vq->host_notifier, NULL); |
ed08a2a0 | 2435 | r = virtio_bus_set_host_notifier(qbus, n, false); |
ff4c07df PB |
2436 | assert(r >= 0); |
2437 | } | |
2438 | return err; | |
2439 | } | |
2440 | ||
2441 | int virtio_device_start_ioeventfd(VirtIODevice *vdev) | |
2442 | { | |
2443 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); | |
2444 | VirtioBusState *vbus = VIRTIO_BUS(qbus); | |
2445 | ||
2446 | return virtio_bus_start_ioeventfd(vbus); | |
2447 | } | |
2448 | ||
2449 | static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev) | |
2450 | { | |
2451 | VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev))); | |
ff4c07df PB |
2452 | int n, r; |
2453 | ||
2454 | for (n = 0; n < VIRTIO_QUEUE_MAX; n++) { | |
fa283a4a PB |
2455 | VirtQueue *vq = &vdev->vq[n]; |
2456 | ||
ff4c07df PB |
2457 | if (!virtio_queue_get_num(vdev, n)) { |
2458 | continue; | |
2459 | } | |
d6da1e9e | 2460 | event_notifier_set_handler(&vq->host_notifier, NULL); |
ed08a2a0 | 2461 | r = virtio_bus_set_host_notifier(qbus, n, false); |
ff4c07df PB |
2462 | assert(r >= 0); |
2463 | } | |
2464 | } | |
2465 | ||
2466 | void virtio_device_stop_ioeventfd(VirtIODevice *vdev) | |
2467 | { | |
2468 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); | |
2469 | VirtioBusState *vbus = VIRTIO_BUS(qbus); | |
2470 | ||
2471 | virtio_bus_stop_ioeventfd(vbus); | |
2472 | } | |
2473 | ||
310837de PB |
2474 | int virtio_device_grab_ioeventfd(VirtIODevice *vdev) |
2475 | { | |
2476 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); | |
2477 | VirtioBusState *vbus = VIRTIO_BUS(qbus); | |
2478 | ||
2479 | return virtio_bus_grab_ioeventfd(vbus); | |
2480 | } | |
2481 | ||
2482 | void virtio_device_release_ioeventfd(VirtIODevice *vdev) | |
2483 | { | |
2484 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); | |
2485 | VirtioBusState *vbus = VIRTIO_BUS(qbus); | |
2486 | ||
2487 | virtio_bus_release_ioeventfd(vbus); | |
2488 | } | |
2489 | ||
8e05db92 FK |
2490 | static void virtio_device_class_init(ObjectClass *klass, void *data) |
2491 | { | |
2492 | /* Set the default value here. */ | |
ff4c07df | 2493 | VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); |
8e05db92 | 2494 | DeviceClass *dc = DEVICE_CLASS(klass); |
1d244b42 AF |
2495 | |
2496 | dc->realize = virtio_device_realize; | |
2497 | dc->unrealize = virtio_device_unrealize; | |
8e05db92 | 2498 | dc->bus_type = TYPE_VIRTIO_BUS; |
6b8f1020 | 2499 | dc->props = virtio_properties; |
ff4c07df PB |
2500 | vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl; |
2501 | vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl; | |
9b706dbb MT |
2502 | |
2503 | vdc->legacy_features |= VIRTIO_LEGACY_FEATURES; | |
8e05db92 FK |
2504 | } |
2505 | ||
8e93cef1 PB |
2506 | bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev) |
2507 | { | |
2508 | BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); | |
2509 | VirtioBusState *vbus = VIRTIO_BUS(qbus); | |
2510 | ||
2511 | return virtio_bus_ioeventfd_enabled(vbus); | |
2512 | } | |
2513 | ||
8e05db92 FK |
2514 | static const TypeInfo virtio_device_info = { |
2515 | .name = TYPE_VIRTIO_DEVICE, | |
2516 | .parent = TYPE_DEVICE, | |
2517 | .instance_size = sizeof(VirtIODevice), | |
2518 | .class_init = virtio_device_class_init, | |
c611c764 | 2519 | .instance_finalize = virtio_device_instance_finalize, |
8e05db92 FK |
2520 | .abstract = true, |
2521 | .class_size = sizeof(VirtioDeviceClass), | |
2522 | }; | |
2523 | ||
2524 | static void virtio_register_types(void) | |
2525 | { | |
2526 | type_register_static(&virtio_device_info); | |
2527 | } | |
2528 | ||
2529 | type_init(virtio_register_types) |