Commit | Line | Data |
---|---|---|
fd534e9b | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
0a8a69dd RR |
2 | /* Virtio ring implementation. |
3 | * | |
4 | * Copyright 2007 Rusty Russell IBM Corporation | |
0a8a69dd RR |
5 | */ |
6 | #include <linux/virtio.h> | |
7 | #include <linux/virtio_ring.h> | |
e34f8725 | 8 | #include <linux/virtio_config.h> |
0a8a69dd | 9 | #include <linux/device.h> |
5a0e3ad6 | 10 | #include <linux/slab.h> |
b5a2c4f1 | 11 | #include <linux/module.h> |
e93300b1 | 12 | #include <linux/hrtimer.h> |
780bc790 | 13 | #include <linux/dma-mapping.h> |
f8ce7263 | 14 | #include <linux/spinlock.h> |
78fe3987 | 15 | #include <xen/xen.h> |
0a8a69dd RR |
16 | |
17 | #ifdef DEBUG | |
18 | /* For development, we want to crash whenever the ring is screwed. */ | |
9499f5e7 RR |
19 | #define BAD_RING(_vq, fmt, args...) \ |
20 | do { \ | |
21 | dev_err(&(_vq)->vq.vdev->dev, \ | |
22 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
23 | BUG(); \ | |
24 | } while (0) | |
c5f841f1 RR |
25 | /* Caller is supposed to guarantee no reentry. */ |
26 | #define START_USE(_vq) \ | |
27 | do { \ | |
28 | if ((_vq)->in_use) \ | |
9499f5e7 RR |
29 | panic("%s:in_use = %i\n", \ |
30 | (_vq)->vq.name, (_vq)->in_use); \ | |
c5f841f1 | 31 | (_vq)->in_use = __LINE__; \ |
9499f5e7 | 32 | } while (0) |
3a35ce7d | 33 | #define END_USE(_vq) \ |
97a545ab | 34 | do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0) |
4d6a105e TB |
35 | #define LAST_ADD_TIME_UPDATE(_vq) \ |
36 | do { \ | |
37 | ktime_t now = ktime_get(); \ | |
38 | \ | |
39 | /* No kick or get, with .1 second between? Warn. */ \ | |
40 | if ((_vq)->last_add_time_valid) \ | |
41 | WARN_ON(ktime_to_ms(ktime_sub(now, \ | |
42 | (_vq)->last_add_time)) > 100); \ | |
43 | (_vq)->last_add_time = now; \ | |
44 | (_vq)->last_add_time_valid = true; \ | |
45 | } while (0) | |
46 | #define LAST_ADD_TIME_CHECK(_vq) \ | |
47 | do { \ | |
48 | if ((_vq)->last_add_time_valid) { \ | |
49 | WARN_ON(ktime_to_ms(ktime_sub(ktime_get(), \ | |
50 | (_vq)->last_add_time)) > 100); \ | |
51 | } \ | |
52 | } while (0) | |
53 | #define LAST_ADD_TIME_INVALID(_vq) \ | |
54 | ((_vq)->last_add_time_valid = false) | |
0a8a69dd | 55 | #else |
9499f5e7 RR |
56 | #define BAD_RING(_vq, fmt, args...) \ |
57 | do { \ | |
58 | dev_err(&_vq->vq.vdev->dev, \ | |
59 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
60 | (_vq)->broken = true; \ | |
61 | } while (0) | |
0a8a69dd RR |
62 | #define START_USE(vq) |
63 | #define END_USE(vq) | |
4d6a105e TB |
64 | #define LAST_ADD_TIME_UPDATE(vq) |
65 | #define LAST_ADD_TIME_CHECK(vq) | |
66 | #define LAST_ADD_TIME_INVALID(vq) | |
0a8a69dd RR |
67 | #endif |
68 | ||
cbeedb72 | 69 | struct vring_desc_state_split { |
780bc790 AL |
70 | void *data; /* Data for callback. */ |
71 | struct vring_desc *indir_desc; /* Indirect descriptor, if any. */ | |
72 | }; | |
73 | ||
1ce9e605 TB |
74 | struct vring_desc_state_packed { |
75 | void *data; /* Data for callback. */ | |
76 | struct vring_packed_desc *indir_desc; /* Indirect descriptor, if any. */ | |
77 | u16 num; /* Descriptor list length. */ | |
1ce9e605 TB |
78 | u16 last; /* The last desc state in a list. */ |
79 | }; | |
80 | ||
1f28750f | 81 | struct vring_desc_extra { |
ef5c366f JW |
82 | dma_addr_t addr; /* Descriptor DMA addr. */ |
83 | u32 len; /* Descriptor length. */ | |
1ce9e605 | 84 | u16 flags; /* Descriptor flags. */ |
aeef9b47 | 85 | u16 next; /* The next desc state in a list. */ |
1ce9e605 TB |
86 | }; |
87 | ||
d76136e4 XZ |
88 | struct vring_virtqueue_split { |
89 | /* Actual memory layout for this queue. */ | |
90 | struct vring vring; | |
91 | ||
92 | /* Last written value to avail->flags */ | |
93 | u16 avail_flags_shadow; | |
94 | ||
95 | /* | |
96 | * Last written value to avail->idx in | |
97 | * guest byte order. | |
98 | */ | |
99 | u16 avail_idx_shadow; | |
100 | ||
101 | /* Per-descriptor state. */ | |
102 | struct vring_desc_state_split *desc_state; | |
103 | struct vring_desc_extra *desc_extra; | |
104 | ||
105 | /* DMA address and size information */ | |
106 | dma_addr_t queue_dma_addr; | |
107 | size_t queue_size_in_bytes; | |
108 | }; | |
109 | ||
110 | struct vring_virtqueue_packed { | |
111 | /* Actual memory layout for this queue. */ | |
112 | struct { | |
113 | unsigned int num; | |
114 | struct vring_packed_desc *desc; | |
115 | struct vring_packed_desc_event *driver; | |
116 | struct vring_packed_desc_event *device; | |
117 | } vring; | |
118 | ||
119 | /* Driver ring wrap counter. */ | |
120 | bool avail_wrap_counter; | |
121 | ||
122 | /* Avail used flags. */ | |
123 | u16 avail_used_flags; | |
124 | ||
125 | /* Index of the next avail descriptor. */ | |
126 | u16 next_avail_idx; | |
127 | ||
128 | /* | |
129 | * Last written value to driver->flags in | |
130 | * guest byte order. | |
131 | */ | |
132 | u16 event_flags_shadow; | |
133 | ||
134 | /* Per-descriptor state. */ | |
135 | struct vring_desc_state_packed *desc_state; | |
136 | struct vring_desc_extra *desc_extra; | |
137 | ||
138 | /* DMA address and size information */ | |
139 | dma_addr_t ring_dma_addr; | |
140 | dma_addr_t driver_event_dma_addr; | |
141 | dma_addr_t device_event_dma_addr; | |
142 | size_t ring_size_in_bytes; | |
143 | size_t event_size_in_bytes; | |
144 | }; | |
145 | ||
43b4f721 | 146 | struct vring_virtqueue { |
0a8a69dd RR |
147 | struct virtqueue vq; |
148 | ||
1ce9e605 TB |
149 | /* Is this a packed ring? */ |
150 | bool packed_ring; | |
151 | ||
fb3fba6b TB |
152 | /* Is DMA API used? */ |
153 | bool use_dma_api; | |
154 | ||
7b21e34f RR |
155 | /* Can we use weak barriers? */ |
156 | bool weak_barriers; | |
157 | ||
0a8a69dd RR |
158 | /* Other side has made a mess, don't try any more. */ |
159 | bool broken; | |
160 | ||
9fa29b9d MM |
161 | /* Host supports indirect buffers */ |
162 | bool indirect; | |
163 | ||
a5c262c5 MT |
164 | /* Host publishes avail event idx */ |
165 | bool event; | |
166 | ||
0a8a69dd RR |
167 | /* Head of free buffer list. */ |
168 | unsigned int free_head; | |
169 | /* Number we've added since last sync. */ | |
170 | unsigned int num_added; | |
171 | ||
a7722890 | 172 | /* Last used index we've seen. |
173 | * for split ring, it just contains last used index | |
174 | * for packed ring: | |
175 | * bits up to VRING_PACKED_EVENT_F_WRAP_CTR include the last used index. | |
176 | * bits from VRING_PACKED_EVENT_F_WRAP_CTR include the used wrap counter. | |
177 | */ | |
1bc4953e | 178 | u16 last_used_idx; |
0a8a69dd | 179 | |
8d622d21 MT |
180 | /* Hint for event idx: already triggered no need to disable. */ |
181 | bool event_triggered; | |
182 | ||
1ce9e605 TB |
183 | union { |
184 | /* Available for split ring */ | |
d76136e4 | 185 | struct vring_virtqueue_split split; |
e593bf97 | 186 | |
1ce9e605 | 187 | /* Available for packed ring */ |
d76136e4 | 188 | struct vring_virtqueue_packed packed; |
1ce9e605 | 189 | }; |
f277ec42 | 190 | |
0a8a69dd | 191 | /* How to notify other side. FIXME: commonalize hcalls! */ |
46f9c2b9 | 192 | bool (*notify)(struct virtqueue *vq); |
0a8a69dd | 193 | |
2a2d1382 AL |
194 | /* DMA, allocation, and size information */ |
195 | bool we_own_ring; | |
2a2d1382 | 196 | |
0a8a69dd RR |
197 | #ifdef DEBUG |
198 | /* They're supposed to lock for us. */ | |
199 | unsigned int in_use; | |
e93300b1 RR |
200 | |
201 | /* Figure out if their kicks are too delayed. */ | |
202 | bool last_add_time_valid; | |
203 | ktime_t last_add_time; | |
0a8a69dd | 204 | #endif |
0a8a69dd RR |
205 | }; |
206 | ||
07d9629d | 207 | static struct virtqueue *__vring_new_virtqueue(unsigned int index, |
cd4c812a | 208 | struct vring_virtqueue_split *vring_split, |
07d9629d XZ |
209 | struct virtio_device *vdev, |
210 | bool weak_barriers, | |
211 | bool context, | |
212 | bool (*notify)(struct virtqueue *), | |
213 | void (*callback)(struct virtqueue *), | |
214 | const char *name); | |
e6f633e5 TB |
215 | |
216 | /* | |
217 | * Helpers. | |
218 | */ | |
219 | ||
0a8a69dd RR |
220 | #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) |
221 | ||
35c51e09 | 222 | static inline bool virtqueue_use_indirect(struct vring_virtqueue *vq, |
2f18c2d1 TB |
223 | unsigned int total_sg) |
224 | { | |
2f18c2d1 TB |
225 | /* |
226 | * If the host supports indirect descriptor tables, and we have multiple | |
227 | * buffers, then go indirect. FIXME: tune this threshold | |
228 | */ | |
229 | return (vq->indirect && total_sg > 1 && vq->vq.num_free); | |
230 | } | |
231 | ||
d26c96c8 | 232 | /* |
1a937693 MT |
233 | * Modern virtio devices have feature bits to specify whether they need a |
234 | * quirk and bypass the IOMMU. If not there, just use the DMA API. | |
235 | * | |
236 | * If there, the interaction between virtio and DMA API is messy. | |
d26c96c8 AL |
237 | * |
238 | * On most systems with virtio, physical addresses match bus addresses, | |
239 | * and it doesn't particularly matter whether we use the DMA API. | |
240 | * | |
241 | * On some systems, including Xen and any system with a physical device | |
242 | * that speaks virtio behind a physical IOMMU, we must use the DMA API | |
243 | * for virtio DMA to work at all. | |
244 | * | |
245 | * On other systems, including SPARC and PPC64, virtio-pci devices are | |
246 | * enumerated as though they are behind an IOMMU, but the virtio host | |
247 | * ignores the IOMMU, so we must either pretend that the IOMMU isn't | |
248 | * there or somehow map everything as the identity. | |
249 | * | |
250 | * For the time being, we preserve historic behavior and bypass the DMA | |
251 | * API. | |
1a937693 MT |
252 | * |
253 | * TODO: install a per-device DMA ops structure that does the right thing | |
254 | * taking into account all the above quirks, and use the DMA API | |
255 | * unconditionally on data path. | |
d26c96c8 AL |
256 | */ |
257 | ||
258 | static bool vring_use_dma_api(struct virtio_device *vdev) | |
259 | { | |
24b6842a | 260 | if (!virtio_has_dma_quirk(vdev)) |
1a937693 MT |
261 | return true; |
262 | ||
263 | /* Otherwise, we are left to guess. */ | |
78fe3987 AL |
264 | /* |
265 | * In theory, it's possible to have a buggy QEMU-supposed | |
266 | * emulated Q35 IOMMU and Xen enabled at the same time. On | |
267 | * such a configuration, virtio has never worked and will | |
268 | * not work without an even larger kludge. Instead, enable | |
269 | * the DMA API if we're a Xen guest, which at least allows | |
270 | * all of the sensible Xen configurations to work correctly. | |
271 | */ | |
272 | if (xen_domain()) | |
273 | return true; | |
274 | ||
d26c96c8 AL |
275 | return false; |
276 | } | |
277 | ||
e6d6dd6c JR |
278 | size_t virtio_max_dma_size(struct virtio_device *vdev) |
279 | { | |
280 | size_t max_segment_size = SIZE_MAX; | |
281 | ||
282 | if (vring_use_dma_api(vdev)) | |
817fc978 | 283 | max_segment_size = dma_max_mapping_size(vdev->dev.parent); |
e6d6dd6c JR |
284 | |
285 | return max_segment_size; | |
286 | } | |
287 | EXPORT_SYMBOL_GPL(virtio_max_dma_size); | |
288 | ||
d79dca75 TB |
289 | static void *vring_alloc_queue(struct virtio_device *vdev, size_t size, |
290 | dma_addr_t *dma_handle, gfp_t flag) | |
291 | { | |
292 | if (vring_use_dma_api(vdev)) { | |
293 | return dma_alloc_coherent(vdev->dev.parent, size, | |
294 | dma_handle, flag); | |
295 | } else { | |
296 | void *queue = alloc_pages_exact(PAGE_ALIGN(size), flag); | |
297 | ||
298 | if (queue) { | |
299 | phys_addr_t phys_addr = virt_to_phys(queue); | |
300 | *dma_handle = (dma_addr_t)phys_addr; | |
301 | ||
302 | /* | |
303 | * Sanity check: make sure we dind't truncate | |
304 | * the address. The only arches I can find that | |
305 | * have 64-bit phys_addr_t but 32-bit dma_addr_t | |
306 | * are certain non-highmem MIPS and x86 | |
307 | * configurations, but these configurations | |
308 | * should never allocate physical pages above 32 | |
309 | * bits, so this is fine. Just in case, throw a | |
310 | * warning and abort if we end up with an | |
311 | * unrepresentable address. | |
312 | */ | |
313 | if (WARN_ON_ONCE(*dma_handle != phys_addr)) { | |
314 | free_pages_exact(queue, PAGE_ALIGN(size)); | |
315 | return NULL; | |
316 | } | |
317 | } | |
318 | return queue; | |
319 | } | |
320 | } | |
321 | ||
322 | static void vring_free_queue(struct virtio_device *vdev, size_t size, | |
323 | void *queue, dma_addr_t dma_handle) | |
324 | { | |
325 | if (vring_use_dma_api(vdev)) | |
326 | dma_free_coherent(vdev->dev.parent, size, queue, dma_handle); | |
327 | else | |
328 | free_pages_exact(queue, PAGE_ALIGN(size)); | |
329 | } | |
330 | ||
780bc790 AL |
331 | /* |
332 | * The DMA ops on various arches are rather gnarly right now, and | |
333 | * making all of the arch DMA ops work on the vring device itself | |
334 | * is a mess. For now, we use the parent device for DMA ops. | |
335 | */ | |
75bfa81b | 336 | static inline struct device *vring_dma_dev(const struct vring_virtqueue *vq) |
780bc790 AL |
337 | { |
338 | return vq->vq.vdev->dev.parent; | |
339 | } | |
340 | ||
341 | /* Map one sg entry. */ | |
342 | static dma_addr_t vring_map_one_sg(const struct vring_virtqueue *vq, | |
343 | struct scatterlist *sg, | |
344 | enum dma_data_direction direction) | |
345 | { | |
fb3fba6b | 346 | if (!vq->use_dma_api) |
780bc790 AL |
347 | return (dma_addr_t)sg_phys(sg); |
348 | ||
349 | /* | |
350 | * We can't use dma_map_sg, because we don't use scatterlists in | |
351 | * the way it expects (we don't guarantee that the scatterlist | |
352 | * will exist for the lifetime of the mapping). | |
353 | */ | |
354 | return dma_map_page(vring_dma_dev(vq), | |
355 | sg_page(sg), sg->offset, sg->length, | |
356 | direction); | |
357 | } | |
358 | ||
359 | static dma_addr_t vring_map_single(const struct vring_virtqueue *vq, | |
360 | void *cpu_addr, size_t size, | |
361 | enum dma_data_direction direction) | |
362 | { | |
fb3fba6b | 363 | if (!vq->use_dma_api) |
780bc790 AL |
364 | return (dma_addr_t)virt_to_phys(cpu_addr); |
365 | ||
366 | return dma_map_single(vring_dma_dev(vq), | |
367 | cpu_addr, size, direction); | |
368 | } | |
369 | ||
e6f633e5 TB |
370 | static int vring_mapping_error(const struct vring_virtqueue *vq, |
371 | dma_addr_t addr) | |
372 | { | |
fb3fba6b | 373 | if (!vq->use_dma_api) |
e6f633e5 TB |
374 | return 0; |
375 | ||
376 | return dma_mapping_error(vring_dma_dev(vq), addr); | |
377 | } | |
378 | ||
3a897128 XZ |
379 | static void virtqueue_init(struct vring_virtqueue *vq, u32 num) |
380 | { | |
381 | vq->vq.num_free = num; | |
382 | ||
383 | if (vq->packed_ring) | |
384 | vq->last_used_idx = 0 | (1 << VRING_PACKED_EVENT_F_WRAP_CTR); | |
385 | else | |
386 | vq->last_used_idx = 0; | |
387 | ||
388 | vq->event_triggered = false; | |
389 | vq->num_added = 0; | |
390 | ||
391 | #ifdef DEBUG | |
392 | vq->in_use = false; | |
393 | vq->last_add_time_valid = false; | |
394 | #endif | |
395 | } | |
396 | ||
e6f633e5 TB |
397 | |
398 | /* | |
399 | * Split ring specific functions - *_split(). | |
400 | */ | |
401 | ||
72b5e895 JW |
402 | static void vring_unmap_one_split_indirect(const struct vring_virtqueue *vq, |
403 | struct vring_desc *desc) | |
780bc790 AL |
404 | { |
405 | u16 flags; | |
406 | ||
fb3fba6b | 407 | if (!vq->use_dma_api) |
780bc790 AL |
408 | return; |
409 | ||
410 | flags = virtio16_to_cpu(vq->vq.vdev, desc->flags); | |
411 | ||
b4282ebc XZ |
412 | dma_unmap_page(vring_dma_dev(vq), |
413 | virtio64_to_cpu(vq->vq.vdev, desc->addr), | |
414 | virtio32_to_cpu(vq->vq.vdev, desc->len), | |
415 | (flags & VRING_DESC_F_WRITE) ? | |
416 | DMA_FROM_DEVICE : DMA_TO_DEVICE); | |
780bc790 AL |
417 | } |
418 | ||
72b5e895 JW |
419 | static unsigned int vring_unmap_one_split(const struct vring_virtqueue *vq, |
420 | unsigned int i) | |
421 | { | |
422 | struct vring_desc_extra *extra = vq->split.desc_extra; | |
423 | u16 flags; | |
424 | ||
425 | if (!vq->use_dma_api) | |
426 | goto out; | |
427 | ||
428 | flags = extra[i].flags; | |
429 | ||
430 | if (flags & VRING_DESC_F_INDIRECT) { | |
431 | dma_unmap_single(vring_dma_dev(vq), | |
432 | extra[i].addr, | |
433 | extra[i].len, | |
434 | (flags & VRING_DESC_F_WRITE) ? | |
435 | DMA_FROM_DEVICE : DMA_TO_DEVICE); | |
436 | } else { | |
437 | dma_unmap_page(vring_dma_dev(vq), | |
438 | extra[i].addr, | |
439 | extra[i].len, | |
440 | (flags & VRING_DESC_F_WRITE) ? | |
441 | DMA_FROM_DEVICE : DMA_TO_DEVICE); | |
442 | } | |
443 | ||
444 | out: | |
445 | return extra[i].next; | |
446 | } | |
447 | ||
138fd251 TB |
448 | static struct vring_desc *alloc_indirect_split(struct virtqueue *_vq, |
449 | unsigned int total_sg, | |
450 | gfp_t gfp) | |
9fa29b9d MM |
451 | { |
452 | struct vring_desc *desc; | |
b25bd251 | 453 | unsigned int i; |
9fa29b9d | 454 | |
b92b1b89 WD |
455 | /* |
456 | * We require lowmem mappings for the descriptors because | |
457 | * otherwise virt_to_phys will give us bogus addresses in the | |
458 | * virtqueue. | |
459 | */ | |
82107539 | 460 | gfp &= ~__GFP_HIGHMEM; |
b92b1b89 | 461 | |
6da2ec56 | 462 | desc = kmalloc_array(total_sg, sizeof(struct vring_desc), gfp); |
9fa29b9d | 463 | if (!desc) |
b25bd251 | 464 | return NULL; |
9fa29b9d | 465 | |
b25bd251 | 466 | for (i = 0; i < total_sg; i++) |
00e6f3d9 | 467 | desc[i].next = cpu_to_virtio16(_vq->vdev, i + 1); |
b25bd251 | 468 | return desc; |
9fa29b9d MM |
469 | } |
470 | ||
fe4c3862 JW |
471 | static inline unsigned int virtqueue_add_desc_split(struct virtqueue *vq, |
472 | struct vring_desc *desc, | |
473 | unsigned int i, | |
474 | dma_addr_t addr, | |
475 | unsigned int len, | |
72b5e895 JW |
476 | u16 flags, |
477 | bool indirect) | |
fe4c3862 | 478 | { |
72b5e895 JW |
479 | struct vring_virtqueue *vring = to_vvq(vq); |
480 | struct vring_desc_extra *extra = vring->split.desc_extra; | |
481 | u16 next; | |
482 | ||
fe4c3862 JW |
483 | desc[i].flags = cpu_to_virtio16(vq->vdev, flags); |
484 | desc[i].addr = cpu_to_virtio64(vq->vdev, addr); | |
485 | desc[i].len = cpu_to_virtio32(vq->vdev, len); | |
486 | ||
72b5e895 JW |
487 | if (!indirect) { |
488 | next = extra[i].next; | |
489 | desc[i].next = cpu_to_virtio16(vq->vdev, next); | |
490 | ||
491 | extra[i].addr = addr; | |
492 | extra[i].len = len; | |
493 | extra[i].flags = flags; | |
494 | } else | |
495 | next = virtio16_to_cpu(vq->vdev, desc[i].next); | |
496 | ||
497 | return next; | |
fe4c3862 JW |
498 | } |
499 | ||
138fd251 TB |
500 | static inline int virtqueue_add_split(struct virtqueue *_vq, |
501 | struct scatterlist *sgs[], | |
502 | unsigned int total_sg, | |
503 | unsigned int out_sgs, | |
504 | unsigned int in_sgs, | |
505 | void *data, | |
506 | void *ctx, | |
507 | gfp_t gfp) | |
0a8a69dd RR |
508 | { |
509 | struct vring_virtqueue *vq = to_vvq(_vq); | |
13816c76 | 510 | struct scatterlist *sg; |
b25bd251 | 511 | struct vring_desc *desc; |
3f649ab7 | 512 | unsigned int i, n, avail, descs_used, prev, err_idx; |
1fe9b6fe | 513 | int head; |
b25bd251 | 514 | bool indirect; |
0a8a69dd | 515 | |
9fa29b9d MM |
516 | START_USE(vq); |
517 | ||
0a8a69dd | 518 | BUG_ON(data == NULL); |
5a08b04f | 519 | BUG_ON(ctx && vq->indirect); |
9fa29b9d | 520 | |
70670444 RR |
521 | if (unlikely(vq->broken)) { |
522 | END_USE(vq); | |
523 | return -EIO; | |
524 | } | |
525 | ||
4d6a105e | 526 | LAST_ADD_TIME_UPDATE(vq); |
e93300b1 | 527 | |
b25bd251 RR |
528 | BUG_ON(total_sg == 0); |
529 | ||
530 | head = vq->free_head; | |
531 | ||
35c51e09 | 532 | if (virtqueue_use_indirect(vq, total_sg)) |
138fd251 | 533 | desc = alloc_indirect_split(_vq, total_sg, gfp); |
44ed8089 | 534 | else { |
b25bd251 | 535 | desc = NULL; |
e593bf97 | 536 | WARN_ON_ONCE(total_sg > vq->split.vring.num && !vq->indirect); |
44ed8089 | 537 | } |
b25bd251 RR |
538 | |
539 | if (desc) { | |
540 | /* Use a single buffer which doesn't continue */ | |
780bc790 | 541 | indirect = true; |
b25bd251 RR |
542 | /* Set up rest to use this indirect table. */ |
543 | i = 0; | |
544 | descs_used = 1; | |
b25bd251 | 545 | } else { |
780bc790 | 546 | indirect = false; |
e593bf97 | 547 | desc = vq->split.vring.desc; |
b25bd251 RR |
548 | i = head; |
549 | descs_used = total_sg; | |
9fa29b9d MM |
550 | } |
551 | ||
b4b4ff73 | 552 | if (unlikely(vq->vq.num_free < descs_used)) { |
0a8a69dd | 553 | pr_debug("Can't add buf len %i - avail = %i\n", |
b25bd251 | 554 | descs_used, vq->vq.num_free); |
44653eae RR |
555 | /* FIXME: for historical reasons, we force a notify here if |
556 | * there are outgoing parts to the buffer. Presumably the | |
557 | * host should service the ring ASAP. */ | |
13816c76 | 558 | if (out_sgs) |
44653eae | 559 | vq->notify(&vq->vq); |
58625edf WY |
560 | if (indirect) |
561 | kfree(desc); | |
0a8a69dd RR |
562 | END_USE(vq); |
563 | return -ENOSPC; | |
564 | } | |
565 | ||
13816c76 | 566 | for (n = 0; n < out_sgs; n++) { |
eeebf9b1 | 567 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
780bc790 AL |
568 | dma_addr_t addr = vring_map_one_sg(vq, sg, DMA_TO_DEVICE); |
569 | if (vring_mapping_error(vq, addr)) | |
570 | goto unmap_release; | |
571 | ||
13816c76 | 572 | prev = i; |
72b5e895 JW |
573 | /* Note that we trust indirect descriptor |
574 | * table since it use stream DMA mapping. | |
575 | */ | |
fe4c3862 | 576 | i = virtqueue_add_desc_split(_vq, desc, i, addr, sg->length, |
72b5e895 JW |
577 | VRING_DESC_F_NEXT, |
578 | indirect); | |
13816c76 | 579 | } |
0a8a69dd | 580 | } |
13816c76 | 581 | for (; n < (out_sgs + in_sgs); n++) { |
eeebf9b1 | 582 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
780bc790 AL |
583 | dma_addr_t addr = vring_map_one_sg(vq, sg, DMA_FROM_DEVICE); |
584 | if (vring_mapping_error(vq, addr)) | |
585 | goto unmap_release; | |
586 | ||
13816c76 | 587 | prev = i; |
72b5e895 JW |
588 | /* Note that we trust indirect descriptor |
589 | * table since it use stream DMA mapping. | |
590 | */ | |
fe4c3862 JW |
591 | i = virtqueue_add_desc_split(_vq, desc, i, addr, |
592 | sg->length, | |
593 | VRING_DESC_F_NEXT | | |
72b5e895 JW |
594 | VRING_DESC_F_WRITE, |
595 | indirect); | |
13816c76 | 596 | } |
0a8a69dd RR |
597 | } |
598 | /* Last one doesn't continue. */ | |
00e6f3d9 | 599 | desc[prev].flags &= cpu_to_virtio16(_vq->vdev, ~VRING_DESC_F_NEXT); |
72b5e895 | 600 | if (!indirect && vq->use_dma_api) |
890d3356 | 601 | vq->split.desc_extra[prev & (vq->split.vring.num - 1)].flags &= |
72b5e895 | 602 | ~VRING_DESC_F_NEXT; |
0a8a69dd | 603 | |
780bc790 AL |
604 | if (indirect) { |
605 | /* Now that the indirect table is filled in, map it. */ | |
606 | dma_addr_t addr = vring_map_single( | |
607 | vq, desc, total_sg * sizeof(struct vring_desc), | |
608 | DMA_TO_DEVICE); | |
609 | if (vring_mapping_error(vq, addr)) | |
610 | goto unmap_release; | |
611 | ||
fe4c3862 JW |
612 | virtqueue_add_desc_split(_vq, vq->split.vring.desc, |
613 | head, addr, | |
614 | total_sg * sizeof(struct vring_desc), | |
72b5e895 JW |
615 | VRING_DESC_F_INDIRECT, |
616 | false); | |
780bc790 AL |
617 | } |
618 | ||
619 | /* We're using some buffers from the free list. */ | |
620 | vq->vq.num_free -= descs_used; | |
621 | ||
0a8a69dd | 622 | /* Update free pointer */ |
b25bd251 | 623 | if (indirect) |
72b5e895 | 624 | vq->free_head = vq->split.desc_extra[head].next; |
b25bd251 RR |
625 | else |
626 | vq->free_head = i; | |
0a8a69dd | 627 | |
780bc790 | 628 | /* Store token and indirect buffer state. */ |
cbeedb72 | 629 | vq->split.desc_state[head].data = data; |
780bc790 | 630 | if (indirect) |
cbeedb72 | 631 | vq->split.desc_state[head].indir_desc = desc; |
87646a34 | 632 | else |
cbeedb72 | 633 | vq->split.desc_state[head].indir_desc = ctx; |
0a8a69dd RR |
634 | |
635 | /* Put entry in available array (but don't update avail->idx until they | |
3b720b8c | 636 | * do sync). */ |
e593bf97 TB |
637 | avail = vq->split.avail_idx_shadow & (vq->split.vring.num - 1); |
638 | vq->split.vring.avail->ring[avail] = cpu_to_virtio16(_vq->vdev, head); | |
0a8a69dd | 639 | |
ee7cd898 RR |
640 | /* Descriptors and available array need to be set before we expose the |
641 | * new available array entries. */ | |
a9a0fef7 | 642 | virtio_wmb(vq->weak_barriers); |
e593bf97 TB |
643 | vq->split.avail_idx_shadow++; |
644 | vq->split.vring.avail->idx = cpu_to_virtio16(_vq->vdev, | |
645 | vq->split.avail_idx_shadow); | |
ee7cd898 RR |
646 | vq->num_added++; |
647 | ||
5e05bf58 TH |
648 | pr_debug("Added buffer head %i to %p\n", head, vq); |
649 | END_USE(vq); | |
650 | ||
ee7cd898 RR |
651 | /* This is very unlikely, but theoretically possible. Kick |
652 | * just in case. */ | |
653 | if (unlikely(vq->num_added == (1 << 16) - 1)) | |
654 | virtqueue_kick(_vq); | |
655 | ||
98e8c6bc | 656 | return 0; |
780bc790 AL |
657 | |
658 | unmap_release: | |
659 | err_idx = i; | |
cf8f1696 ML |
660 | |
661 | if (indirect) | |
662 | i = 0; | |
663 | else | |
664 | i = head; | |
780bc790 AL |
665 | |
666 | for (n = 0; n < total_sg; n++) { | |
667 | if (i == err_idx) | |
668 | break; | |
72b5e895 JW |
669 | if (indirect) { |
670 | vring_unmap_one_split_indirect(vq, &desc[i]); | |
671 | i = virtio16_to_cpu(_vq->vdev, desc[i].next); | |
672 | } else | |
673 | i = vring_unmap_one_split(vq, i); | |
780bc790 AL |
674 | } |
675 | ||
780bc790 AL |
676 | if (indirect) |
677 | kfree(desc); | |
678 | ||
3cc36f6e | 679 | END_USE(vq); |
f7728002 | 680 | return -ENOMEM; |
0a8a69dd | 681 | } |
13816c76 | 682 | |
138fd251 | 683 | static bool virtqueue_kick_prepare_split(struct virtqueue *_vq) |
0a8a69dd RR |
684 | { |
685 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a5c262c5 | 686 | u16 new, old; |
41f0377f RR |
687 | bool needs_kick; |
688 | ||
0a8a69dd | 689 | START_USE(vq); |
a72caae2 JW |
690 | /* We need to expose available array entries before checking avail |
691 | * event. */ | |
a9a0fef7 | 692 | virtio_mb(vq->weak_barriers); |
0a8a69dd | 693 | |
e593bf97 TB |
694 | old = vq->split.avail_idx_shadow - vq->num_added; |
695 | new = vq->split.avail_idx_shadow; | |
0a8a69dd RR |
696 | vq->num_added = 0; |
697 | ||
4d6a105e TB |
698 | LAST_ADD_TIME_CHECK(vq); |
699 | LAST_ADD_TIME_INVALID(vq); | |
e93300b1 | 700 | |
41f0377f | 701 | if (vq->event) { |
e593bf97 TB |
702 | needs_kick = vring_need_event(virtio16_to_cpu(_vq->vdev, |
703 | vring_avail_event(&vq->split.vring)), | |
41f0377f RR |
704 | new, old); |
705 | } else { | |
e593bf97 TB |
706 | needs_kick = !(vq->split.vring.used->flags & |
707 | cpu_to_virtio16(_vq->vdev, | |
708 | VRING_USED_F_NO_NOTIFY)); | |
41f0377f | 709 | } |
0a8a69dd | 710 | END_USE(vq); |
41f0377f RR |
711 | return needs_kick; |
712 | } | |
138fd251 | 713 | |
138fd251 TB |
714 | static void detach_buf_split(struct vring_virtqueue *vq, unsigned int head, |
715 | void **ctx) | |
0a8a69dd | 716 | { |
780bc790 | 717 | unsigned int i, j; |
c60923cb | 718 | __virtio16 nextflag = cpu_to_virtio16(vq->vq.vdev, VRING_DESC_F_NEXT); |
0a8a69dd RR |
719 | |
720 | /* Clear data ptr. */ | |
cbeedb72 | 721 | vq->split.desc_state[head].data = NULL; |
0a8a69dd | 722 | |
780bc790 | 723 | /* Put back on free list: unmap first-level descriptors and find end */ |
0a8a69dd | 724 | i = head; |
9fa29b9d | 725 | |
e593bf97 | 726 | while (vq->split.vring.desc[i].flags & nextflag) { |
72b5e895 JW |
727 | vring_unmap_one_split(vq, i); |
728 | i = vq->split.desc_extra[i].next; | |
06ca287d | 729 | vq->vq.num_free++; |
0a8a69dd RR |
730 | } |
731 | ||
72b5e895 JW |
732 | vring_unmap_one_split(vq, i); |
733 | vq->split.desc_extra[i].next = vq->free_head; | |
0a8a69dd | 734 | vq->free_head = head; |
780bc790 | 735 | |
0a8a69dd | 736 | /* Plus final descriptor */ |
06ca287d | 737 | vq->vq.num_free++; |
780bc790 | 738 | |
5a08b04f | 739 | if (vq->indirect) { |
cbeedb72 TB |
740 | struct vring_desc *indir_desc = |
741 | vq->split.desc_state[head].indir_desc; | |
5a08b04f MT |
742 | u32 len; |
743 | ||
744 | /* Free the indirect table, if any, now that it's unmapped. */ | |
745 | if (!indir_desc) | |
746 | return; | |
747 | ||
72b5e895 | 748 | len = vq->split.desc_extra[head].len; |
780bc790 | 749 | |
72b5e895 JW |
750 | BUG_ON(!(vq->split.desc_extra[head].flags & |
751 | VRING_DESC_F_INDIRECT)); | |
780bc790 AL |
752 | BUG_ON(len == 0 || len % sizeof(struct vring_desc)); |
753 | ||
754 | for (j = 0; j < len / sizeof(struct vring_desc); j++) | |
72b5e895 | 755 | vring_unmap_one_split_indirect(vq, &indir_desc[j]); |
780bc790 | 756 | |
5a08b04f | 757 | kfree(indir_desc); |
cbeedb72 | 758 | vq->split.desc_state[head].indir_desc = NULL; |
5a08b04f | 759 | } else if (ctx) { |
cbeedb72 | 760 | *ctx = vq->split.desc_state[head].indir_desc; |
780bc790 | 761 | } |
0a8a69dd RR |
762 | } |
763 | ||
138fd251 | 764 | static inline bool more_used_split(const struct vring_virtqueue *vq) |
0a8a69dd | 765 | { |
e593bf97 TB |
766 | return vq->last_used_idx != virtio16_to_cpu(vq->vq.vdev, |
767 | vq->split.vring.used->idx); | |
0a8a69dd RR |
768 | } |
769 | ||
138fd251 TB |
770 | static void *virtqueue_get_buf_ctx_split(struct virtqueue *_vq, |
771 | unsigned int *len, | |
772 | void **ctx) | |
0a8a69dd RR |
773 | { |
774 | struct vring_virtqueue *vq = to_vvq(_vq); | |
775 | void *ret; | |
776 | unsigned int i; | |
3b720b8c | 777 | u16 last_used; |
0a8a69dd RR |
778 | |
779 | START_USE(vq); | |
780 | ||
5ef82752 RR |
781 | if (unlikely(vq->broken)) { |
782 | END_USE(vq); | |
783 | return NULL; | |
784 | } | |
785 | ||
138fd251 | 786 | if (!more_used_split(vq)) { |
0a8a69dd RR |
787 | pr_debug("No more buffers in queue\n"); |
788 | END_USE(vq); | |
789 | return NULL; | |
790 | } | |
791 | ||
2d61ba95 | 792 | /* Only get used array entries after they have been exposed by host. */ |
a9a0fef7 | 793 | virtio_rmb(vq->weak_barriers); |
2d61ba95 | 794 | |
e593bf97 TB |
795 | last_used = (vq->last_used_idx & (vq->split.vring.num - 1)); |
796 | i = virtio32_to_cpu(_vq->vdev, | |
797 | vq->split.vring.used->ring[last_used].id); | |
798 | *len = virtio32_to_cpu(_vq->vdev, | |
799 | vq->split.vring.used->ring[last_used].len); | |
0a8a69dd | 800 | |
e593bf97 | 801 | if (unlikely(i >= vq->split.vring.num)) { |
0a8a69dd RR |
802 | BAD_RING(vq, "id %u out of range\n", i); |
803 | return NULL; | |
804 | } | |
cbeedb72 | 805 | if (unlikely(!vq->split.desc_state[i].data)) { |
0a8a69dd RR |
806 | BAD_RING(vq, "id %u is not a head!\n", i); |
807 | return NULL; | |
808 | } | |
809 | ||
138fd251 | 810 | /* detach_buf_split clears data, so grab it now. */ |
cbeedb72 | 811 | ret = vq->split.desc_state[i].data; |
138fd251 | 812 | detach_buf_split(vq, i, ctx); |
0a8a69dd | 813 | vq->last_used_idx++; |
a5c262c5 MT |
814 | /* If we expect an interrupt for the next entry, tell host |
815 | * by writing event index and flush out the write before | |
816 | * the read in the next get_buf call. */ | |
e593bf97 | 817 | if (!(vq->split.avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT)) |
788e5b3a | 818 | virtio_store_mb(vq->weak_barriers, |
e593bf97 | 819 | &vring_used_event(&vq->split.vring), |
788e5b3a | 820 | cpu_to_virtio16(_vq->vdev, vq->last_used_idx)); |
a5c262c5 | 821 | |
4d6a105e | 822 | LAST_ADD_TIME_INVALID(vq); |
e93300b1 | 823 | |
0a8a69dd RR |
824 | END_USE(vq); |
825 | return ret; | |
826 | } | |
138fd251 | 827 | |
138fd251 | 828 | static void virtqueue_disable_cb_split(struct virtqueue *_vq) |
18445c4d RR |
829 | { |
830 | struct vring_virtqueue *vq = to_vvq(_vq); | |
831 | ||
e593bf97 TB |
832 | if (!(vq->split.avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT)) { |
833 | vq->split.avail_flags_shadow |= VRING_AVAIL_F_NO_INTERRUPT; | |
8d622d21 MT |
834 | if (vq->event) |
835 | /* TODO: this is a hack. Figure out a cleaner value to write. */ | |
836 | vring_used_event(&vq->split.vring) = 0x0; | |
837 | else | |
e593bf97 TB |
838 | vq->split.vring.avail->flags = |
839 | cpu_to_virtio16(_vq->vdev, | |
840 | vq->split.avail_flags_shadow); | |
f277ec42 | 841 | } |
18445c4d RR |
842 | } |
843 | ||
31532340 | 844 | static unsigned int virtqueue_enable_cb_prepare_split(struct virtqueue *_vq) |
0a8a69dd RR |
845 | { |
846 | struct vring_virtqueue *vq = to_vvq(_vq); | |
cc229884 | 847 | u16 last_used_idx; |
0a8a69dd RR |
848 | |
849 | START_USE(vq); | |
0a8a69dd RR |
850 | |
851 | /* We optimistically turn back on interrupts, then check if there was | |
852 | * more to do. */ | |
a5c262c5 MT |
853 | /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to |
854 | * either clear the flags bit or point the event index at the next | |
855 | * entry. Always do both to keep code simple. */ | |
e593bf97 TB |
856 | if (vq->split.avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT) { |
857 | vq->split.avail_flags_shadow &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
0ea1e4a6 | 858 | if (!vq->event) |
e593bf97 TB |
859 | vq->split.vring.avail->flags = |
860 | cpu_to_virtio16(_vq->vdev, | |
861 | vq->split.avail_flags_shadow); | |
f277ec42 | 862 | } |
e593bf97 TB |
863 | vring_used_event(&vq->split.vring) = cpu_to_virtio16(_vq->vdev, |
864 | last_used_idx = vq->last_used_idx); | |
cc229884 MT |
865 | END_USE(vq); |
866 | return last_used_idx; | |
867 | } | |
138fd251 | 868 | |
31532340 | 869 | static bool virtqueue_poll_split(struct virtqueue *_vq, unsigned int last_used_idx) |
138fd251 TB |
870 | { |
871 | struct vring_virtqueue *vq = to_vvq(_vq); | |
872 | ||
873 | return (u16)last_used_idx != virtio16_to_cpu(_vq->vdev, | |
e593bf97 | 874 | vq->split.vring.used->idx); |
138fd251 TB |
875 | } |
876 | ||
138fd251 | 877 | static bool virtqueue_enable_cb_delayed_split(struct virtqueue *_vq) |
7ab358c2 MT |
878 | { |
879 | struct vring_virtqueue *vq = to_vvq(_vq); | |
880 | u16 bufs; | |
881 | ||
882 | START_USE(vq); | |
883 | ||
884 | /* We optimistically turn back on interrupts, then check if there was | |
885 | * more to do. */ | |
886 | /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to | |
887 | * either clear the flags bit or point the event index at the next | |
0ea1e4a6 | 888 | * entry. Always update the event index to keep code simple. */ |
e593bf97 TB |
889 | if (vq->split.avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT) { |
890 | vq->split.avail_flags_shadow &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
0ea1e4a6 | 891 | if (!vq->event) |
e593bf97 TB |
892 | vq->split.vring.avail->flags = |
893 | cpu_to_virtio16(_vq->vdev, | |
894 | vq->split.avail_flags_shadow); | |
f277ec42 | 895 | } |
7ab358c2 | 896 | /* TODO: tune this threshold */ |
e593bf97 | 897 | bufs = (u16)(vq->split.avail_idx_shadow - vq->last_used_idx) * 3 / 4; |
788e5b3a MT |
898 | |
899 | virtio_store_mb(vq->weak_barriers, | |
e593bf97 | 900 | &vring_used_event(&vq->split.vring), |
788e5b3a MT |
901 | cpu_to_virtio16(_vq->vdev, vq->last_used_idx + bufs)); |
902 | ||
e593bf97 TB |
903 | if (unlikely((u16)(virtio16_to_cpu(_vq->vdev, vq->split.vring.used->idx) |
904 | - vq->last_used_idx) > bufs)) { | |
7ab358c2 MT |
905 | END_USE(vq); |
906 | return false; | |
907 | } | |
908 | ||
909 | END_USE(vq); | |
910 | return true; | |
911 | } | |
7ab358c2 | 912 | |
138fd251 | 913 | static void *virtqueue_detach_unused_buf_split(struct virtqueue *_vq) |
c021eac4 SM |
914 | { |
915 | struct vring_virtqueue *vq = to_vvq(_vq); | |
916 | unsigned int i; | |
917 | void *buf; | |
918 | ||
919 | START_USE(vq); | |
920 | ||
e593bf97 | 921 | for (i = 0; i < vq->split.vring.num; i++) { |
cbeedb72 | 922 | if (!vq->split.desc_state[i].data) |
c021eac4 | 923 | continue; |
138fd251 | 924 | /* detach_buf_split clears data, so grab it now. */ |
cbeedb72 | 925 | buf = vq->split.desc_state[i].data; |
138fd251 | 926 | detach_buf_split(vq, i, NULL); |
e593bf97 TB |
927 | vq->split.avail_idx_shadow--; |
928 | vq->split.vring.avail->idx = cpu_to_virtio16(_vq->vdev, | |
929 | vq->split.avail_idx_shadow); | |
c021eac4 SM |
930 | END_USE(vq); |
931 | return buf; | |
932 | } | |
933 | /* That should have freed everything. */ | |
e593bf97 | 934 | BUG_ON(vq->vq.num_free != vq->split.vring.num); |
c021eac4 SM |
935 | |
936 | END_USE(vq); | |
937 | return NULL; | |
938 | } | |
138fd251 | 939 | |
d79dca75 TB |
940 | static struct virtqueue *vring_create_virtqueue_split( |
941 | unsigned int index, | |
942 | unsigned int num, | |
943 | unsigned int vring_align, | |
944 | struct virtio_device *vdev, | |
945 | bool weak_barriers, | |
946 | bool may_reduce_num, | |
947 | bool context, | |
948 | bool (*notify)(struct virtqueue *), | |
949 | void (*callback)(struct virtqueue *), | |
950 | const char *name) | |
951 | { | |
cd4c812a | 952 | struct vring_virtqueue_split vring_split = {}; |
d79dca75 TB |
953 | struct virtqueue *vq; |
954 | void *queue = NULL; | |
955 | dma_addr_t dma_addr; | |
956 | size_t queue_size_in_bytes; | |
957 | struct vring vring; | |
958 | ||
959 | /* We assume num is a power of 2. */ | |
960 | if (num & (num - 1)) { | |
961 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
962 | return NULL; | |
963 | } | |
964 | ||
965 | /* TODO: allocate each queue chunk individually */ | |
966 | for (; num && vring_size(num, vring_align) > PAGE_SIZE; num /= 2) { | |
967 | queue = vring_alloc_queue(vdev, vring_size(num, vring_align), | |
968 | &dma_addr, | |
c7cc29aa | 969 | GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO); |
d79dca75 TB |
970 | if (queue) |
971 | break; | |
cf94db21 CH |
972 | if (!may_reduce_num) |
973 | return NULL; | |
d79dca75 TB |
974 | } |
975 | ||
976 | if (!num) | |
977 | return NULL; | |
978 | ||
979 | if (!queue) { | |
980 | /* Try to get a single page. You are my only hope! */ | |
981 | queue = vring_alloc_queue(vdev, vring_size(num, vring_align), | |
982 | &dma_addr, GFP_KERNEL|__GFP_ZERO); | |
983 | } | |
984 | if (!queue) | |
985 | return NULL; | |
986 | ||
987 | queue_size_in_bytes = vring_size(num, vring_align); | |
cd4c812a | 988 | vring_init(&vring_split.vring, num, queue, vring_align); |
d79dca75 | 989 | |
cd4c812a XZ |
990 | vq = __vring_new_virtqueue(index, &vring_split, vdev, weak_barriers, |
991 | context, notify, callback, name); | |
d79dca75 TB |
992 | if (!vq) { |
993 | vring_free_queue(vdev, queue_size_in_bytes, queue, | |
994 | dma_addr); | |
995 | return NULL; | |
996 | } | |
997 | ||
998 | to_vvq(vq)->split.queue_dma_addr = dma_addr; | |
999 | to_vvq(vq)->split.queue_size_in_bytes = queue_size_in_bytes; | |
1000 | to_vvq(vq)->we_own_ring = true; | |
1001 | ||
1002 | return vq; | |
1003 | } | |
1004 | ||
e6f633e5 | 1005 | |
1ce9e605 TB |
1006 | /* |
1007 | * Packed ring specific functions - *_packed(). | |
1008 | */ | |
a7722890 | 1009 | static inline bool packed_used_wrap_counter(u16 last_used_idx) |
1010 | { | |
1011 | return !!(last_used_idx & (1 << VRING_PACKED_EVENT_F_WRAP_CTR)); | |
1012 | } | |
1013 | ||
1014 | static inline u16 packed_last_used(u16 last_used_idx) | |
1015 | { | |
1016 | return last_used_idx & ~(-(1 << VRING_PACKED_EVENT_F_WRAP_CTR)); | |
1017 | } | |
1ce9e605 | 1018 | |
d80dc15b XZ |
1019 | static void vring_unmap_extra_packed(const struct vring_virtqueue *vq, |
1020 | struct vring_desc_extra *extra) | |
1ce9e605 TB |
1021 | { |
1022 | u16 flags; | |
1023 | ||
1024 | if (!vq->use_dma_api) | |
1025 | return; | |
1026 | ||
d80dc15b | 1027 | flags = extra->flags; |
1ce9e605 TB |
1028 | |
1029 | if (flags & VRING_DESC_F_INDIRECT) { | |
1030 | dma_unmap_single(vring_dma_dev(vq), | |
d80dc15b | 1031 | extra->addr, extra->len, |
1ce9e605 TB |
1032 | (flags & VRING_DESC_F_WRITE) ? |
1033 | DMA_FROM_DEVICE : DMA_TO_DEVICE); | |
1034 | } else { | |
1035 | dma_unmap_page(vring_dma_dev(vq), | |
d80dc15b | 1036 | extra->addr, extra->len, |
1ce9e605 TB |
1037 | (flags & VRING_DESC_F_WRITE) ? |
1038 | DMA_FROM_DEVICE : DMA_TO_DEVICE); | |
1039 | } | |
1040 | } | |
1041 | ||
1042 | static void vring_unmap_desc_packed(const struct vring_virtqueue *vq, | |
1043 | struct vring_packed_desc *desc) | |
1044 | { | |
1045 | u16 flags; | |
1046 | ||
1047 | if (!vq->use_dma_api) | |
1048 | return; | |
1049 | ||
1050 | flags = le16_to_cpu(desc->flags); | |
1051 | ||
920379a4 XZ |
1052 | dma_unmap_page(vring_dma_dev(vq), |
1053 | le64_to_cpu(desc->addr), | |
1054 | le32_to_cpu(desc->len), | |
1055 | (flags & VRING_DESC_F_WRITE) ? | |
1056 | DMA_FROM_DEVICE : DMA_TO_DEVICE); | |
1ce9e605 TB |
1057 | } |
1058 | ||
1059 | static struct vring_packed_desc *alloc_indirect_packed(unsigned int total_sg, | |
1060 | gfp_t gfp) | |
1061 | { | |
1062 | struct vring_packed_desc *desc; | |
1063 | ||
1064 | /* | |
1065 | * We require lowmem mappings for the descriptors because | |
1066 | * otherwise virt_to_phys will give us bogus addresses in the | |
1067 | * virtqueue. | |
1068 | */ | |
1069 | gfp &= ~__GFP_HIGHMEM; | |
1070 | ||
1071 | desc = kmalloc_array(total_sg, sizeof(struct vring_packed_desc), gfp); | |
1072 | ||
1073 | return desc; | |
1074 | } | |
1075 | ||
1076 | static int virtqueue_add_indirect_packed(struct vring_virtqueue *vq, | |
8d7670f3 XZ |
1077 | struct scatterlist *sgs[], |
1078 | unsigned int total_sg, | |
1079 | unsigned int out_sgs, | |
1080 | unsigned int in_sgs, | |
1081 | void *data, | |
1082 | gfp_t gfp) | |
1ce9e605 TB |
1083 | { |
1084 | struct vring_packed_desc *desc; | |
1085 | struct scatterlist *sg; | |
1086 | unsigned int i, n, err_idx; | |
1087 | u16 head, id; | |
1088 | dma_addr_t addr; | |
1089 | ||
1090 | head = vq->packed.next_avail_idx; | |
1091 | desc = alloc_indirect_packed(total_sg, gfp); | |
fc6d70f4 XZ |
1092 | if (!desc) |
1093 | return -ENOMEM; | |
1ce9e605 TB |
1094 | |
1095 | if (unlikely(vq->vq.num_free < 1)) { | |
1096 | pr_debug("Can't add buf len 1 - avail = 0\n"); | |
df0bfe75 | 1097 | kfree(desc); |
1ce9e605 TB |
1098 | END_USE(vq); |
1099 | return -ENOSPC; | |
1100 | } | |
1101 | ||
1102 | i = 0; | |
1103 | id = vq->free_head; | |
1104 | BUG_ON(id == vq->packed.vring.num); | |
1105 | ||
1106 | for (n = 0; n < out_sgs + in_sgs; n++) { | |
1107 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { | |
1108 | addr = vring_map_one_sg(vq, sg, n < out_sgs ? | |
1109 | DMA_TO_DEVICE : DMA_FROM_DEVICE); | |
1110 | if (vring_mapping_error(vq, addr)) | |
1111 | goto unmap_release; | |
1112 | ||
1113 | desc[i].flags = cpu_to_le16(n < out_sgs ? | |
1114 | 0 : VRING_DESC_F_WRITE); | |
1115 | desc[i].addr = cpu_to_le64(addr); | |
1116 | desc[i].len = cpu_to_le32(sg->length); | |
1117 | i++; | |
1118 | } | |
1119 | } | |
1120 | ||
1121 | /* Now that the indirect table is filled in, map it. */ | |
1122 | addr = vring_map_single(vq, desc, | |
1123 | total_sg * sizeof(struct vring_packed_desc), | |
1124 | DMA_TO_DEVICE); | |
1125 | if (vring_mapping_error(vq, addr)) | |
1126 | goto unmap_release; | |
1127 | ||
1128 | vq->packed.vring.desc[head].addr = cpu_to_le64(addr); | |
1129 | vq->packed.vring.desc[head].len = cpu_to_le32(total_sg * | |
1130 | sizeof(struct vring_packed_desc)); | |
1131 | vq->packed.vring.desc[head].id = cpu_to_le16(id); | |
1132 | ||
1133 | if (vq->use_dma_api) { | |
1134 | vq->packed.desc_extra[id].addr = addr; | |
1135 | vq->packed.desc_extra[id].len = total_sg * | |
1136 | sizeof(struct vring_packed_desc); | |
1137 | vq->packed.desc_extra[id].flags = VRING_DESC_F_INDIRECT | | |
1138 | vq->packed.avail_used_flags; | |
1139 | } | |
1140 | ||
1141 | /* | |
1142 | * A driver MUST NOT make the first descriptor in the list | |
1143 | * available before all subsequent descriptors comprising | |
1144 | * the list are made available. | |
1145 | */ | |
1146 | virtio_wmb(vq->weak_barriers); | |
1147 | vq->packed.vring.desc[head].flags = cpu_to_le16(VRING_DESC_F_INDIRECT | | |
1148 | vq->packed.avail_used_flags); | |
1149 | ||
1150 | /* We're using some buffers from the free list. */ | |
1151 | vq->vq.num_free -= 1; | |
1152 | ||
1153 | /* Update free pointer */ | |
1154 | n = head + 1; | |
1155 | if (n >= vq->packed.vring.num) { | |
1156 | n = 0; | |
1157 | vq->packed.avail_wrap_counter ^= 1; | |
1158 | vq->packed.avail_used_flags ^= | |
1159 | 1 << VRING_PACKED_DESC_F_AVAIL | | |
1160 | 1 << VRING_PACKED_DESC_F_USED; | |
1161 | } | |
1162 | vq->packed.next_avail_idx = n; | |
aeef9b47 | 1163 | vq->free_head = vq->packed.desc_extra[id].next; |
1ce9e605 TB |
1164 | |
1165 | /* Store token and indirect buffer state. */ | |
1166 | vq->packed.desc_state[id].num = 1; | |
1167 | vq->packed.desc_state[id].data = data; | |
1168 | vq->packed.desc_state[id].indir_desc = desc; | |
1169 | vq->packed.desc_state[id].last = id; | |
1170 | ||
1171 | vq->num_added += 1; | |
1172 | ||
1173 | pr_debug("Added buffer head %i to %p\n", head, vq); | |
1174 | END_USE(vq); | |
1175 | ||
1176 | return 0; | |
1177 | ||
1178 | unmap_release: | |
1179 | err_idx = i; | |
1180 | ||
1181 | for (i = 0; i < err_idx; i++) | |
1182 | vring_unmap_desc_packed(vq, &desc[i]); | |
1183 | ||
1184 | kfree(desc); | |
1185 | ||
1186 | END_USE(vq); | |
f7728002 | 1187 | return -ENOMEM; |
1ce9e605 TB |
1188 | } |
1189 | ||
1190 | static inline int virtqueue_add_packed(struct virtqueue *_vq, | |
1191 | struct scatterlist *sgs[], | |
1192 | unsigned int total_sg, | |
1193 | unsigned int out_sgs, | |
1194 | unsigned int in_sgs, | |
1195 | void *data, | |
1196 | void *ctx, | |
1197 | gfp_t gfp) | |
1198 | { | |
1199 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1200 | struct vring_packed_desc *desc; | |
1201 | struct scatterlist *sg; | |
1202 | unsigned int i, n, c, descs_used, err_idx; | |
3f649ab7 KC |
1203 | __le16 head_flags, flags; |
1204 | u16 head, id, prev, curr, avail_used_flags; | |
fc6d70f4 | 1205 | int err; |
1ce9e605 TB |
1206 | |
1207 | START_USE(vq); | |
1208 | ||
1209 | BUG_ON(data == NULL); | |
1210 | BUG_ON(ctx && vq->indirect); | |
1211 | ||
1212 | if (unlikely(vq->broken)) { | |
1213 | END_USE(vq); | |
1214 | return -EIO; | |
1215 | } | |
1216 | ||
1217 | LAST_ADD_TIME_UPDATE(vq); | |
1218 | ||
1219 | BUG_ON(total_sg == 0); | |
1220 | ||
35c51e09 | 1221 | if (virtqueue_use_indirect(vq, total_sg)) { |
fc6d70f4 XZ |
1222 | err = virtqueue_add_indirect_packed(vq, sgs, total_sg, out_sgs, |
1223 | in_sgs, data, gfp); | |
1861ba62 MT |
1224 | if (err != -ENOMEM) { |
1225 | END_USE(vq); | |
fc6d70f4 | 1226 | return err; |
1861ba62 | 1227 | } |
fc6d70f4 XZ |
1228 | |
1229 | /* fall back on direct */ | |
1230 | } | |
1ce9e605 TB |
1231 | |
1232 | head = vq->packed.next_avail_idx; | |
1233 | avail_used_flags = vq->packed.avail_used_flags; | |
1234 | ||
1235 | WARN_ON_ONCE(total_sg > vq->packed.vring.num && !vq->indirect); | |
1236 | ||
1237 | desc = vq->packed.vring.desc; | |
1238 | i = head; | |
1239 | descs_used = total_sg; | |
1240 | ||
1241 | if (unlikely(vq->vq.num_free < descs_used)) { | |
1242 | pr_debug("Can't add buf len %i - avail = %i\n", | |
1243 | descs_used, vq->vq.num_free); | |
1244 | END_USE(vq); | |
1245 | return -ENOSPC; | |
1246 | } | |
1247 | ||
1248 | id = vq->free_head; | |
1249 | BUG_ON(id == vq->packed.vring.num); | |
1250 | ||
1251 | curr = id; | |
1252 | c = 0; | |
1253 | for (n = 0; n < out_sgs + in_sgs; n++) { | |
1254 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { | |
1255 | dma_addr_t addr = vring_map_one_sg(vq, sg, n < out_sgs ? | |
1256 | DMA_TO_DEVICE : DMA_FROM_DEVICE); | |
1257 | if (vring_mapping_error(vq, addr)) | |
1258 | goto unmap_release; | |
1259 | ||
1260 | flags = cpu_to_le16(vq->packed.avail_used_flags | | |
1261 | (++c == total_sg ? 0 : VRING_DESC_F_NEXT) | | |
1262 | (n < out_sgs ? 0 : VRING_DESC_F_WRITE)); | |
1263 | if (i == head) | |
1264 | head_flags = flags; | |
1265 | else | |
1266 | desc[i].flags = flags; | |
1267 | ||
1268 | desc[i].addr = cpu_to_le64(addr); | |
1269 | desc[i].len = cpu_to_le32(sg->length); | |
1270 | desc[i].id = cpu_to_le16(id); | |
1271 | ||
1272 | if (unlikely(vq->use_dma_api)) { | |
1273 | vq->packed.desc_extra[curr].addr = addr; | |
1274 | vq->packed.desc_extra[curr].len = sg->length; | |
1275 | vq->packed.desc_extra[curr].flags = | |
1276 | le16_to_cpu(flags); | |
1277 | } | |
1278 | prev = curr; | |
aeef9b47 | 1279 | curr = vq->packed.desc_extra[curr].next; |
1ce9e605 TB |
1280 | |
1281 | if ((unlikely(++i >= vq->packed.vring.num))) { | |
1282 | i = 0; | |
1283 | vq->packed.avail_used_flags ^= | |
1284 | 1 << VRING_PACKED_DESC_F_AVAIL | | |
1285 | 1 << VRING_PACKED_DESC_F_USED; | |
1286 | } | |
1287 | } | |
1288 | } | |
1289 | ||
1290 | if (i < head) | |
1291 | vq->packed.avail_wrap_counter ^= 1; | |
1292 | ||
1293 | /* We're using some buffers from the free list. */ | |
1294 | vq->vq.num_free -= descs_used; | |
1295 | ||
1296 | /* Update free pointer */ | |
1297 | vq->packed.next_avail_idx = i; | |
1298 | vq->free_head = curr; | |
1299 | ||
1300 | /* Store token. */ | |
1301 | vq->packed.desc_state[id].num = descs_used; | |
1302 | vq->packed.desc_state[id].data = data; | |
1303 | vq->packed.desc_state[id].indir_desc = ctx; | |
1304 | vq->packed.desc_state[id].last = prev; | |
1305 | ||
1306 | /* | |
1307 | * A driver MUST NOT make the first descriptor in the list | |
1308 | * available before all subsequent descriptors comprising | |
1309 | * the list are made available. | |
1310 | */ | |
1311 | virtio_wmb(vq->weak_barriers); | |
1312 | vq->packed.vring.desc[head].flags = head_flags; | |
1313 | vq->num_added += descs_used; | |
1314 | ||
1315 | pr_debug("Added buffer head %i to %p\n", head, vq); | |
1316 | END_USE(vq); | |
1317 | ||
1318 | return 0; | |
1319 | ||
1320 | unmap_release: | |
1321 | err_idx = i; | |
1322 | i = head; | |
44593865 | 1323 | curr = vq->free_head; |
1ce9e605 TB |
1324 | |
1325 | vq->packed.avail_used_flags = avail_used_flags; | |
1326 | ||
1327 | for (n = 0; n < total_sg; n++) { | |
1328 | if (i == err_idx) | |
1329 | break; | |
d80dc15b | 1330 | vring_unmap_extra_packed(vq, &vq->packed.desc_extra[curr]); |
44593865 | 1331 | curr = vq->packed.desc_extra[curr].next; |
1ce9e605 TB |
1332 | i++; |
1333 | if (i >= vq->packed.vring.num) | |
1334 | i = 0; | |
1335 | } | |
1336 | ||
1337 | END_USE(vq); | |
1338 | return -EIO; | |
1339 | } | |
1340 | ||
1341 | static bool virtqueue_kick_prepare_packed(struct virtqueue *_vq) | |
1342 | { | |
1343 | struct vring_virtqueue *vq = to_vvq(_vq); | |
f51f9826 | 1344 | u16 new, old, off_wrap, flags, wrap_counter, event_idx; |
1ce9e605 TB |
1345 | bool needs_kick; |
1346 | union { | |
1347 | struct { | |
1348 | __le16 off_wrap; | |
1349 | __le16 flags; | |
1350 | }; | |
1351 | u32 u32; | |
1352 | } snapshot; | |
1353 | ||
1354 | START_USE(vq); | |
1355 | ||
1356 | /* | |
1357 | * We need to expose the new flags value before checking notification | |
1358 | * suppressions. | |
1359 | */ | |
1360 | virtio_mb(vq->weak_barriers); | |
1361 | ||
f51f9826 TB |
1362 | old = vq->packed.next_avail_idx - vq->num_added; |
1363 | new = vq->packed.next_avail_idx; | |
1ce9e605 TB |
1364 | vq->num_added = 0; |
1365 | ||
1366 | snapshot.u32 = *(u32 *)vq->packed.vring.device; | |
1367 | flags = le16_to_cpu(snapshot.flags); | |
1368 | ||
1369 | LAST_ADD_TIME_CHECK(vq); | |
1370 | LAST_ADD_TIME_INVALID(vq); | |
1371 | ||
f51f9826 TB |
1372 | if (flags != VRING_PACKED_EVENT_FLAG_DESC) { |
1373 | needs_kick = (flags != VRING_PACKED_EVENT_FLAG_DISABLE); | |
1374 | goto out; | |
1375 | } | |
1376 | ||
1377 | off_wrap = le16_to_cpu(snapshot.off_wrap); | |
1378 | ||
1379 | wrap_counter = off_wrap >> VRING_PACKED_EVENT_F_WRAP_CTR; | |
1380 | event_idx = off_wrap & ~(1 << VRING_PACKED_EVENT_F_WRAP_CTR); | |
1381 | if (wrap_counter != vq->packed.avail_wrap_counter) | |
1382 | event_idx -= vq->packed.vring.num; | |
1383 | ||
1384 | needs_kick = vring_need_event(event_idx, new, old); | |
1385 | out: | |
1ce9e605 TB |
1386 | END_USE(vq); |
1387 | return needs_kick; | |
1388 | } | |
1389 | ||
1390 | static void detach_buf_packed(struct vring_virtqueue *vq, | |
1391 | unsigned int id, void **ctx) | |
1392 | { | |
1393 | struct vring_desc_state_packed *state = NULL; | |
1394 | struct vring_packed_desc *desc; | |
1395 | unsigned int i, curr; | |
1396 | ||
1397 | state = &vq->packed.desc_state[id]; | |
1398 | ||
1399 | /* Clear data ptr. */ | |
1400 | state->data = NULL; | |
1401 | ||
aeef9b47 | 1402 | vq->packed.desc_extra[state->last].next = vq->free_head; |
1ce9e605 TB |
1403 | vq->free_head = id; |
1404 | vq->vq.num_free += state->num; | |
1405 | ||
1406 | if (unlikely(vq->use_dma_api)) { | |
1407 | curr = id; | |
1408 | for (i = 0; i < state->num; i++) { | |
d80dc15b XZ |
1409 | vring_unmap_extra_packed(vq, |
1410 | &vq->packed.desc_extra[curr]); | |
aeef9b47 | 1411 | curr = vq->packed.desc_extra[curr].next; |
1ce9e605 TB |
1412 | } |
1413 | } | |
1414 | ||
1415 | if (vq->indirect) { | |
1416 | u32 len; | |
1417 | ||
1418 | /* Free the indirect table, if any, now that it's unmapped. */ | |
1419 | desc = state->indir_desc; | |
1420 | if (!desc) | |
1421 | return; | |
1422 | ||
1423 | if (vq->use_dma_api) { | |
1424 | len = vq->packed.desc_extra[id].len; | |
1425 | for (i = 0; i < len / sizeof(struct vring_packed_desc); | |
1426 | i++) | |
1427 | vring_unmap_desc_packed(vq, &desc[i]); | |
1428 | } | |
1429 | kfree(desc); | |
1430 | state->indir_desc = NULL; | |
1431 | } else if (ctx) { | |
1432 | *ctx = state->indir_desc; | |
1433 | } | |
1434 | } | |
1435 | ||
1436 | static inline bool is_used_desc_packed(const struct vring_virtqueue *vq, | |
1437 | u16 idx, bool used_wrap_counter) | |
1438 | { | |
1439 | bool avail, used; | |
1440 | u16 flags; | |
1441 | ||
1442 | flags = le16_to_cpu(vq->packed.vring.desc[idx].flags); | |
1443 | avail = !!(flags & (1 << VRING_PACKED_DESC_F_AVAIL)); | |
1444 | used = !!(flags & (1 << VRING_PACKED_DESC_F_USED)); | |
1445 | ||
1446 | return avail == used && used == used_wrap_counter; | |
1447 | } | |
1448 | ||
1449 | static inline bool more_used_packed(const struct vring_virtqueue *vq) | |
1450 | { | |
a7722890 | 1451 | u16 last_used; |
1452 | u16 last_used_idx; | |
1453 | bool used_wrap_counter; | |
1454 | ||
1455 | last_used_idx = READ_ONCE(vq->last_used_idx); | |
1456 | last_used = packed_last_used(last_used_idx); | |
1457 | used_wrap_counter = packed_used_wrap_counter(last_used_idx); | |
1458 | return is_used_desc_packed(vq, last_used, used_wrap_counter); | |
1ce9e605 TB |
1459 | } |
1460 | ||
1461 | static void *virtqueue_get_buf_ctx_packed(struct virtqueue *_vq, | |
1462 | unsigned int *len, | |
1463 | void **ctx) | |
1464 | { | |
1465 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a7722890 | 1466 | u16 last_used, id, last_used_idx; |
1467 | bool used_wrap_counter; | |
1ce9e605 TB |
1468 | void *ret; |
1469 | ||
1470 | START_USE(vq); | |
1471 | ||
1472 | if (unlikely(vq->broken)) { | |
1473 | END_USE(vq); | |
1474 | return NULL; | |
1475 | } | |
1476 | ||
1477 | if (!more_used_packed(vq)) { | |
1478 | pr_debug("No more buffers in queue\n"); | |
1479 | END_USE(vq); | |
1480 | return NULL; | |
1481 | } | |
1482 | ||
1483 | /* Only get used elements after they have been exposed by host. */ | |
1484 | virtio_rmb(vq->weak_barriers); | |
1485 | ||
a7722890 | 1486 | last_used_idx = READ_ONCE(vq->last_used_idx); |
1487 | used_wrap_counter = packed_used_wrap_counter(last_used_idx); | |
1488 | last_used = packed_last_used(last_used_idx); | |
1ce9e605 TB |
1489 | id = le16_to_cpu(vq->packed.vring.desc[last_used].id); |
1490 | *len = le32_to_cpu(vq->packed.vring.desc[last_used].len); | |
1491 | ||
1492 | if (unlikely(id >= vq->packed.vring.num)) { | |
1493 | BAD_RING(vq, "id %u out of range\n", id); | |
1494 | return NULL; | |
1495 | } | |
1496 | if (unlikely(!vq->packed.desc_state[id].data)) { | |
1497 | BAD_RING(vq, "id %u is not a head!\n", id); | |
1498 | return NULL; | |
1499 | } | |
1500 | ||
1501 | /* detach_buf_packed clears data, so grab it now. */ | |
1502 | ret = vq->packed.desc_state[id].data; | |
1503 | detach_buf_packed(vq, id, ctx); | |
1504 | ||
a7722890 | 1505 | last_used += vq->packed.desc_state[id].num; |
1506 | if (unlikely(last_used >= vq->packed.vring.num)) { | |
1507 | last_used -= vq->packed.vring.num; | |
1508 | used_wrap_counter ^= 1; | |
1ce9e605 TB |
1509 | } |
1510 | ||
a7722890 | 1511 | last_used = (last_used | (used_wrap_counter << VRING_PACKED_EVENT_F_WRAP_CTR)); |
1512 | WRITE_ONCE(vq->last_used_idx, last_used); | |
1513 | ||
f51f9826 TB |
1514 | /* |
1515 | * If we expect an interrupt for the next entry, tell host | |
1516 | * by writing event index and flush out the write before | |
1517 | * the read in the next get_buf call. | |
1518 | */ | |
1519 | if (vq->packed.event_flags_shadow == VRING_PACKED_EVENT_FLAG_DESC) | |
1520 | virtio_store_mb(vq->weak_barriers, | |
1521 | &vq->packed.vring.driver->off_wrap, | |
a7722890 | 1522 | cpu_to_le16(vq->last_used_idx)); |
f51f9826 | 1523 | |
1ce9e605 TB |
1524 | LAST_ADD_TIME_INVALID(vq); |
1525 | ||
1526 | END_USE(vq); | |
1527 | return ret; | |
1528 | } | |
1529 | ||
1530 | static void virtqueue_disable_cb_packed(struct virtqueue *_vq) | |
1531 | { | |
1532 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1533 | ||
1534 | if (vq->packed.event_flags_shadow != VRING_PACKED_EVENT_FLAG_DISABLE) { | |
1535 | vq->packed.event_flags_shadow = VRING_PACKED_EVENT_FLAG_DISABLE; | |
1536 | vq->packed.vring.driver->flags = | |
1537 | cpu_to_le16(vq->packed.event_flags_shadow); | |
1538 | } | |
1539 | } | |
1540 | ||
31532340 | 1541 | static unsigned int virtqueue_enable_cb_prepare_packed(struct virtqueue *_vq) |
1ce9e605 TB |
1542 | { |
1543 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1544 | ||
1545 | START_USE(vq); | |
1546 | ||
1547 | /* | |
1548 | * We optimistically turn back on interrupts, then check if there was | |
1549 | * more to do. | |
1550 | */ | |
1551 | ||
f51f9826 TB |
1552 | if (vq->event) { |
1553 | vq->packed.vring.driver->off_wrap = | |
a7722890 | 1554 | cpu_to_le16(vq->last_used_idx); |
f51f9826 TB |
1555 | /* |
1556 | * We need to update event offset and event wrap | |
1557 | * counter first before updating event flags. | |
1558 | */ | |
1559 | virtio_wmb(vq->weak_barriers); | |
1560 | } | |
1561 | ||
1ce9e605 | 1562 | if (vq->packed.event_flags_shadow == VRING_PACKED_EVENT_FLAG_DISABLE) { |
f51f9826 TB |
1563 | vq->packed.event_flags_shadow = vq->event ? |
1564 | VRING_PACKED_EVENT_FLAG_DESC : | |
1565 | VRING_PACKED_EVENT_FLAG_ENABLE; | |
1ce9e605 TB |
1566 | vq->packed.vring.driver->flags = |
1567 | cpu_to_le16(vq->packed.event_flags_shadow); | |
1568 | } | |
1569 | ||
1570 | END_USE(vq); | |
a7722890 | 1571 | return vq->last_used_idx; |
1ce9e605 TB |
1572 | } |
1573 | ||
1574 | static bool virtqueue_poll_packed(struct virtqueue *_vq, u16 off_wrap) | |
1575 | { | |
1576 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1577 | bool wrap_counter; | |
1578 | u16 used_idx; | |
1579 | ||
1580 | wrap_counter = off_wrap >> VRING_PACKED_EVENT_F_WRAP_CTR; | |
1581 | used_idx = off_wrap & ~(1 << VRING_PACKED_EVENT_F_WRAP_CTR); | |
1582 | ||
1583 | return is_used_desc_packed(vq, used_idx, wrap_counter); | |
1584 | } | |
1585 | ||
1586 | static bool virtqueue_enable_cb_delayed_packed(struct virtqueue *_vq) | |
1587 | { | |
1588 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a7722890 | 1589 | u16 used_idx, wrap_counter, last_used_idx; |
f51f9826 | 1590 | u16 bufs; |
1ce9e605 TB |
1591 | |
1592 | START_USE(vq); | |
1593 | ||
1594 | /* | |
1595 | * We optimistically turn back on interrupts, then check if there was | |
1596 | * more to do. | |
1597 | */ | |
1598 | ||
f51f9826 TB |
1599 | if (vq->event) { |
1600 | /* TODO: tune this threshold */ | |
1601 | bufs = (vq->packed.vring.num - vq->vq.num_free) * 3 / 4; | |
a7722890 | 1602 | last_used_idx = READ_ONCE(vq->last_used_idx); |
1603 | wrap_counter = packed_used_wrap_counter(last_used_idx); | |
f51f9826 | 1604 | |
a7722890 | 1605 | used_idx = packed_last_used(last_used_idx) + bufs; |
f51f9826 TB |
1606 | if (used_idx >= vq->packed.vring.num) { |
1607 | used_idx -= vq->packed.vring.num; | |
1608 | wrap_counter ^= 1; | |
1609 | } | |
1610 | ||
1611 | vq->packed.vring.driver->off_wrap = cpu_to_le16(used_idx | | |
1612 | (wrap_counter << VRING_PACKED_EVENT_F_WRAP_CTR)); | |
1613 | ||
1614 | /* | |
1615 | * We need to update event offset and event wrap | |
1616 | * counter first before updating event flags. | |
1617 | */ | |
1618 | virtio_wmb(vq->weak_barriers); | |
f51f9826 | 1619 | } |
1ce9e605 TB |
1620 | |
1621 | if (vq->packed.event_flags_shadow == VRING_PACKED_EVENT_FLAG_DISABLE) { | |
f51f9826 TB |
1622 | vq->packed.event_flags_shadow = vq->event ? |
1623 | VRING_PACKED_EVENT_FLAG_DESC : | |
1624 | VRING_PACKED_EVENT_FLAG_ENABLE; | |
1ce9e605 TB |
1625 | vq->packed.vring.driver->flags = |
1626 | cpu_to_le16(vq->packed.event_flags_shadow); | |
1627 | } | |
1628 | ||
1629 | /* | |
1630 | * We need to update event suppression structure first | |
1631 | * before re-checking for more used buffers. | |
1632 | */ | |
1633 | virtio_mb(vq->weak_barriers); | |
1634 | ||
a7722890 | 1635 | last_used_idx = READ_ONCE(vq->last_used_idx); |
1636 | wrap_counter = packed_used_wrap_counter(last_used_idx); | |
1637 | used_idx = packed_last_used(last_used_idx); | |
1638 | if (is_used_desc_packed(vq, used_idx, wrap_counter)) { | |
1ce9e605 TB |
1639 | END_USE(vq); |
1640 | return false; | |
1641 | } | |
1642 | ||
1643 | END_USE(vq); | |
1644 | return true; | |
1645 | } | |
1646 | ||
1647 | static void *virtqueue_detach_unused_buf_packed(struct virtqueue *_vq) | |
1648 | { | |
1649 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1650 | unsigned int i; | |
1651 | void *buf; | |
1652 | ||
1653 | START_USE(vq); | |
1654 | ||
1655 | for (i = 0; i < vq->packed.vring.num; i++) { | |
1656 | if (!vq->packed.desc_state[i].data) | |
1657 | continue; | |
1658 | /* detach_buf clears data, so grab it now. */ | |
1659 | buf = vq->packed.desc_state[i].data; | |
1660 | detach_buf_packed(vq, i, NULL); | |
1661 | END_USE(vq); | |
1662 | return buf; | |
1663 | } | |
1664 | /* That should have freed everything. */ | |
1665 | BUG_ON(vq->vq.num_free != vq->packed.vring.num); | |
1666 | ||
1667 | END_USE(vq); | |
1668 | return NULL; | |
1669 | } | |
1670 | ||
96ef18a2 | 1671 | static struct vring_desc_extra *vring_alloc_desc_extra(unsigned int num) |
5a222421 JW |
1672 | { |
1673 | struct vring_desc_extra *desc_extra; | |
1674 | unsigned int i; | |
1675 | ||
1676 | desc_extra = kmalloc_array(num, sizeof(struct vring_desc_extra), | |
1677 | GFP_KERNEL); | |
1678 | if (!desc_extra) | |
1679 | return NULL; | |
1680 | ||
1681 | memset(desc_extra, 0, num * sizeof(struct vring_desc_extra)); | |
1682 | ||
1683 | for (i = 0; i < num - 1; i++) | |
1684 | desc_extra[i].next = i + 1; | |
1685 | ||
1686 | return desc_extra; | |
1687 | } | |
1688 | ||
1ce9e605 TB |
1689 | static struct virtqueue *vring_create_virtqueue_packed( |
1690 | unsigned int index, | |
1691 | unsigned int num, | |
1692 | unsigned int vring_align, | |
1693 | struct virtio_device *vdev, | |
1694 | bool weak_barriers, | |
1695 | bool may_reduce_num, | |
1696 | bool context, | |
1697 | bool (*notify)(struct virtqueue *), | |
1698 | void (*callback)(struct virtqueue *), | |
1699 | const char *name) | |
1700 | { | |
1701 | struct vring_virtqueue *vq; | |
1702 | struct vring_packed_desc *ring; | |
1703 | struct vring_packed_desc_event *driver, *device; | |
1704 | dma_addr_t ring_dma_addr, driver_event_dma_addr, device_event_dma_addr; | |
1705 | size_t ring_size_in_bytes, event_size_in_bytes; | |
1ce9e605 TB |
1706 | |
1707 | ring_size_in_bytes = num * sizeof(struct vring_packed_desc); | |
1708 | ||
1709 | ring = vring_alloc_queue(vdev, ring_size_in_bytes, | |
1710 | &ring_dma_addr, | |
1711 | GFP_KERNEL|__GFP_NOWARN|__GFP_ZERO); | |
1712 | if (!ring) | |
1713 | goto err_ring; | |
1714 | ||
1715 | event_size_in_bytes = sizeof(struct vring_packed_desc_event); | |
1716 | ||
1717 | driver = vring_alloc_queue(vdev, event_size_in_bytes, | |
1718 | &driver_event_dma_addr, | |
1719 | GFP_KERNEL|__GFP_NOWARN|__GFP_ZERO); | |
1720 | if (!driver) | |
1721 | goto err_driver; | |
1722 | ||
1723 | device = vring_alloc_queue(vdev, event_size_in_bytes, | |
1724 | &device_event_dma_addr, | |
1725 | GFP_KERNEL|__GFP_NOWARN|__GFP_ZERO); | |
1726 | if (!device) | |
1727 | goto err_device; | |
1728 | ||
1729 | vq = kmalloc(sizeof(*vq), GFP_KERNEL); | |
1730 | if (!vq) | |
1731 | goto err_vq; | |
1732 | ||
1733 | vq->vq.callback = callback; | |
1734 | vq->vq.vdev = vdev; | |
1735 | vq->vq.name = name; | |
1ce9e605 TB |
1736 | vq->vq.index = index; |
1737 | vq->we_own_ring = true; | |
1738 | vq->notify = notify; | |
1739 | vq->weak_barriers = weak_barriers; | |
c346dae4 | 1740 | #ifdef CONFIG_VIRTIO_HARDEN_NOTIFICATION |
8b4ec69d | 1741 | vq->broken = true; |
c346dae4 JW |
1742 | #else |
1743 | vq->broken = false; | |
1744 | #endif | |
1ce9e605 TB |
1745 | vq->packed_ring = true; |
1746 | vq->use_dma_api = vring_use_dma_api(vdev); | |
1ce9e605 TB |
1747 | |
1748 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC) && | |
1749 | !context; | |
1750 | vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); | |
1751 | ||
45383fb0 TB |
1752 | if (virtio_has_feature(vdev, VIRTIO_F_ORDER_PLATFORM)) |
1753 | vq->weak_barriers = false; | |
1754 | ||
1ce9e605 TB |
1755 | vq->packed.ring_dma_addr = ring_dma_addr; |
1756 | vq->packed.driver_event_dma_addr = driver_event_dma_addr; | |
1757 | vq->packed.device_event_dma_addr = device_event_dma_addr; | |
1758 | ||
1759 | vq->packed.ring_size_in_bytes = ring_size_in_bytes; | |
1760 | vq->packed.event_size_in_bytes = event_size_in_bytes; | |
1761 | ||
1762 | vq->packed.vring.num = num; | |
1763 | vq->packed.vring.desc = ring; | |
1764 | vq->packed.vring.driver = driver; | |
1765 | vq->packed.vring.device = device; | |
1766 | ||
1767 | vq->packed.next_avail_idx = 0; | |
1768 | vq->packed.avail_wrap_counter = 1; | |
1ce9e605 TB |
1769 | vq->packed.event_flags_shadow = 0; |
1770 | vq->packed.avail_used_flags = 1 << VRING_PACKED_DESC_F_AVAIL; | |
1771 | ||
1772 | vq->packed.desc_state = kmalloc_array(num, | |
1773 | sizeof(struct vring_desc_state_packed), | |
1774 | GFP_KERNEL); | |
1775 | if (!vq->packed.desc_state) | |
1776 | goto err_desc_state; | |
1777 | ||
1778 | memset(vq->packed.desc_state, 0, | |
1779 | num * sizeof(struct vring_desc_state_packed)); | |
1780 | ||
1781 | /* Put everything in free lists. */ | |
1782 | vq->free_head = 0; | |
1ce9e605 | 1783 | |
96ef18a2 | 1784 | vq->packed.desc_extra = vring_alloc_desc_extra(num); |
1ce9e605 TB |
1785 | if (!vq->packed.desc_extra) |
1786 | goto err_desc_extra; | |
1787 | ||
1ce9e605 TB |
1788 | /* No callback? Tell other side not to bother us. */ |
1789 | if (!callback) { | |
1790 | vq->packed.event_flags_shadow = VRING_PACKED_EVENT_FLAG_DISABLE; | |
1791 | vq->packed.vring.driver->flags = | |
1792 | cpu_to_le16(vq->packed.event_flags_shadow); | |
1793 | } | |
1794 | ||
3a897128 XZ |
1795 | virtqueue_init(vq, num); |
1796 | ||
0e566c8f | 1797 | spin_lock(&vdev->vqs_list_lock); |
e152d8af | 1798 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0e566c8f | 1799 | spin_unlock(&vdev->vqs_list_lock); |
1ce9e605 TB |
1800 | return &vq->vq; |
1801 | ||
1802 | err_desc_extra: | |
1803 | kfree(vq->packed.desc_state); | |
1804 | err_desc_state: | |
1805 | kfree(vq); | |
1806 | err_vq: | |
ae93d8ea | 1807 | vring_free_queue(vdev, event_size_in_bytes, device, device_event_dma_addr); |
1ce9e605 | 1808 | err_device: |
ae93d8ea | 1809 | vring_free_queue(vdev, event_size_in_bytes, driver, driver_event_dma_addr); |
1ce9e605 TB |
1810 | err_driver: |
1811 | vring_free_queue(vdev, ring_size_in_bytes, ring, ring_dma_addr); | |
1812 | err_ring: | |
1813 | return NULL; | |
1814 | } | |
1815 | ||
1816 | ||
e6f633e5 TB |
1817 | /* |
1818 | * Generic functions and exported symbols. | |
1819 | */ | |
1820 | ||
1821 | static inline int virtqueue_add(struct virtqueue *_vq, | |
1822 | struct scatterlist *sgs[], | |
1823 | unsigned int total_sg, | |
1824 | unsigned int out_sgs, | |
1825 | unsigned int in_sgs, | |
1826 | void *data, | |
1827 | void *ctx, | |
1828 | gfp_t gfp) | |
1829 | { | |
1ce9e605 TB |
1830 | struct vring_virtqueue *vq = to_vvq(_vq); |
1831 | ||
1832 | return vq->packed_ring ? virtqueue_add_packed(_vq, sgs, total_sg, | |
1833 | out_sgs, in_sgs, data, ctx, gfp) : | |
1834 | virtqueue_add_split(_vq, sgs, total_sg, | |
1835 | out_sgs, in_sgs, data, ctx, gfp); | |
e6f633e5 TB |
1836 | } |
1837 | ||
1838 | /** | |
1839 | * virtqueue_add_sgs - expose buffers to other end | |
a5581206 | 1840 | * @_vq: the struct virtqueue we're talking about. |
e6f633e5 | 1841 | * @sgs: array of terminated scatterlists. |
a5581206 JB |
1842 | * @out_sgs: the number of scatterlists readable by other side |
1843 | * @in_sgs: the number of scatterlists which are writable (after readable ones) | |
e6f633e5 TB |
1844 | * @data: the token identifying the buffer. |
1845 | * @gfp: how to do memory allocations (if necessary). | |
1846 | * | |
1847 | * Caller must ensure we don't call this with other virtqueue operations | |
1848 | * at the same time (except where noted). | |
1849 | * | |
1850 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). | |
1851 | */ | |
1852 | int virtqueue_add_sgs(struct virtqueue *_vq, | |
1853 | struct scatterlist *sgs[], | |
1854 | unsigned int out_sgs, | |
1855 | unsigned int in_sgs, | |
1856 | void *data, | |
1857 | gfp_t gfp) | |
1858 | { | |
1859 | unsigned int i, total_sg = 0; | |
1860 | ||
1861 | /* Count them first. */ | |
1862 | for (i = 0; i < out_sgs + in_sgs; i++) { | |
1863 | struct scatterlist *sg; | |
1864 | ||
1865 | for (sg = sgs[i]; sg; sg = sg_next(sg)) | |
1866 | total_sg++; | |
1867 | } | |
1868 | return virtqueue_add(_vq, sgs, total_sg, out_sgs, in_sgs, | |
1869 | data, NULL, gfp); | |
1870 | } | |
1871 | EXPORT_SYMBOL_GPL(virtqueue_add_sgs); | |
1872 | ||
1873 | /** | |
1874 | * virtqueue_add_outbuf - expose output buffers to other end | |
1875 | * @vq: the struct virtqueue we're talking about. | |
1876 | * @sg: scatterlist (must be well-formed and terminated!) | |
1877 | * @num: the number of entries in @sg readable by other side | |
1878 | * @data: the token identifying the buffer. | |
1879 | * @gfp: how to do memory allocations (if necessary). | |
1880 | * | |
1881 | * Caller must ensure we don't call this with other virtqueue operations | |
1882 | * at the same time (except where noted). | |
1883 | * | |
1884 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). | |
1885 | */ | |
1886 | int virtqueue_add_outbuf(struct virtqueue *vq, | |
1887 | struct scatterlist *sg, unsigned int num, | |
1888 | void *data, | |
1889 | gfp_t gfp) | |
1890 | { | |
1891 | return virtqueue_add(vq, &sg, num, 1, 0, data, NULL, gfp); | |
1892 | } | |
1893 | EXPORT_SYMBOL_GPL(virtqueue_add_outbuf); | |
1894 | ||
1895 | /** | |
1896 | * virtqueue_add_inbuf - expose input buffers to other end | |
1897 | * @vq: the struct virtqueue we're talking about. | |
1898 | * @sg: scatterlist (must be well-formed and terminated!) | |
1899 | * @num: the number of entries in @sg writable by other side | |
1900 | * @data: the token identifying the buffer. | |
1901 | * @gfp: how to do memory allocations (if necessary). | |
1902 | * | |
1903 | * Caller must ensure we don't call this with other virtqueue operations | |
1904 | * at the same time (except where noted). | |
1905 | * | |
1906 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). | |
1907 | */ | |
1908 | int virtqueue_add_inbuf(struct virtqueue *vq, | |
1909 | struct scatterlist *sg, unsigned int num, | |
1910 | void *data, | |
1911 | gfp_t gfp) | |
1912 | { | |
1913 | return virtqueue_add(vq, &sg, num, 0, 1, data, NULL, gfp); | |
1914 | } | |
1915 | EXPORT_SYMBOL_GPL(virtqueue_add_inbuf); | |
1916 | ||
1917 | /** | |
1918 | * virtqueue_add_inbuf_ctx - expose input buffers to other end | |
1919 | * @vq: the struct virtqueue we're talking about. | |
1920 | * @sg: scatterlist (must be well-formed and terminated!) | |
1921 | * @num: the number of entries in @sg writable by other side | |
1922 | * @data: the token identifying the buffer. | |
1923 | * @ctx: extra context for the token | |
1924 | * @gfp: how to do memory allocations (if necessary). | |
1925 | * | |
1926 | * Caller must ensure we don't call this with other virtqueue operations | |
1927 | * at the same time (except where noted). | |
1928 | * | |
1929 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). | |
1930 | */ | |
1931 | int virtqueue_add_inbuf_ctx(struct virtqueue *vq, | |
1932 | struct scatterlist *sg, unsigned int num, | |
1933 | void *data, | |
1934 | void *ctx, | |
1935 | gfp_t gfp) | |
1936 | { | |
1937 | return virtqueue_add(vq, &sg, num, 0, 1, data, ctx, gfp); | |
1938 | } | |
1939 | EXPORT_SYMBOL_GPL(virtqueue_add_inbuf_ctx); | |
1940 | ||
1941 | /** | |
1942 | * virtqueue_kick_prepare - first half of split virtqueue_kick call. | |
a5581206 | 1943 | * @_vq: the struct virtqueue |
e6f633e5 TB |
1944 | * |
1945 | * Instead of virtqueue_kick(), you can do: | |
1946 | * if (virtqueue_kick_prepare(vq)) | |
1947 | * virtqueue_notify(vq); | |
1948 | * | |
1949 | * This is sometimes useful because the virtqueue_kick_prepare() needs | |
1950 | * to be serialized, but the actual virtqueue_notify() call does not. | |
1951 | */ | |
1952 | bool virtqueue_kick_prepare(struct virtqueue *_vq) | |
1953 | { | |
1ce9e605 TB |
1954 | struct vring_virtqueue *vq = to_vvq(_vq); |
1955 | ||
1956 | return vq->packed_ring ? virtqueue_kick_prepare_packed(_vq) : | |
1957 | virtqueue_kick_prepare_split(_vq); | |
e6f633e5 TB |
1958 | } |
1959 | EXPORT_SYMBOL_GPL(virtqueue_kick_prepare); | |
1960 | ||
1961 | /** | |
1962 | * virtqueue_notify - second half of split virtqueue_kick call. | |
a5581206 | 1963 | * @_vq: the struct virtqueue |
e6f633e5 TB |
1964 | * |
1965 | * This does not need to be serialized. | |
1966 | * | |
1967 | * Returns false if host notify failed or queue is broken, otherwise true. | |
1968 | */ | |
1969 | bool virtqueue_notify(struct virtqueue *_vq) | |
1970 | { | |
1971 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1972 | ||
1973 | if (unlikely(vq->broken)) | |
1974 | return false; | |
1975 | ||
1976 | /* Prod other side to tell it about changes. */ | |
1977 | if (!vq->notify(_vq)) { | |
1978 | vq->broken = true; | |
1979 | return false; | |
1980 | } | |
1981 | return true; | |
1982 | } | |
1983 | EXPORT_SYMBOL_GPL(virtqueue_notify); | |
1984 | ||
1985 | /** | |
1986 | * virtqueue_kick - update after add_buf | |
1987 | * @vq: the struct virtqueue | |
1988 | * | |
1989 | * After one or more virtqueue_add_* calls, invoke this to kick | |
1990 | * the other side. | |
1991 | * | |
1992 | * Caller must ensure we don't call this with other virtqueue | |
1993 | * operations at the same time (except where noted). | |
1994 | * | |
1995 | * Returns false if kick failed, otherwise true. | |
1996 | */ | |
1997 | bool virtqueue_kick(struct virtqueue *vq) | |
1998 | { | |
1999 | if (virtqueue_kick_prepare(vq)) | |
2000 | return virtqueue_notify(vq); | |
2001 | return true; | |
2002 | } | |
2003 | EXPORT_SYMBOL_GPL(virtqueue_kick); | |
2004 | ||
2005 | /** | |
31c11db6 | 2006 | * virtqueue_get_buf_ctx - get the next used buffer |
a5581206 | 2007 | * @_vq: the struct virtqueue we're talking about. |
e6f633e5 | 2008 | * @len: the length written into the buffer |
a5581206 | 2009 | * @ctx: extra context for the token |
e6f633e5 TB |
2010 | * |
2011 | * If the device wrote data into the buffer, @len will be set to the | |
2012 | * amount written. This means you don't need to clear the buffer | |
2013 | * beforehand to ensure there's no data leakage in the case of short | |
2014 | * writes. | |
2015 | * | |
2016 | * Caller must ensure we don't call this with other virtqueue | |
2017 | * operations at the same time (except where noted). | |
2018 | * | |
2019 | * Returns NULL if there are no used buffers, or the "data" token | |
2020 | * handed to virtqueue_add_*(). | |
2021 | */ | |
2022 | void *virtqueue_get_buf_ctx(struct virtqueue *_vq, unsigned int *len, | |
2023 | void **ctx) | |
2024 | { | |
1ce9e605 TB |
2025 | struct vring_virtqueue *vq = to_vvq(_vq); |
2026 | ||
2027 | return vq->packed_ring ? virtqueue_get_buf_ctx_packed(_vq, len, ctx) : | |
2028 | virtqueue_get_buf_ctx_split(_vq, len, ctx); | |
e6f633e5 TB |
2029 | } |
2030 | EXPORT_SYMBOL_GPL(virtqueue_get_buf_ctx); | |
2031 | ||
2032 | void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len) | |
2033 | { | |
2034 | return virtqueue_get_buf_ctx(_vq, len, NULL); | |
2035 | } | |
2036 | EXPORT_SYMBOL_GPL(virtqueue_get_buf); | |
e6f633e5 TB |
2037 | /** |
2038 | * virtqueue_disable_cb - disable callbacks | |
a5581206 | 2039 | * @_vq: the struct virtqueue we're talking about. |
e6f633e5 TB |
2040 | * |
2041 | * Note that this is not necessarily synchronous, hence unreliable and only | |
2042 | * useful as an optimization. | |
2043 | * | |
2044 | * Unlike other operations, this need not be serialized. | |
2045 | */ | |
2046 | void virtqueue_disable_cb(struct virtqueue *_vq) | |
2047 | { | |
1ce9e605 TB |
2048 | struct vring_virtqueue *vq = to_vvq(_vq); |
2049 | ||
8d622d21 MT |
2050 | /* If device triggered an event already it won't trigger one again: |
2051 | * no need to disable. | |
2052 | */ | |
2053 | if (vq->event_triggered) | |
2054 | return; | |
2055 | ||
1ce9e605 TB |
2056 | if (vq->packed_ring) |
2057 | virtqueue_disable_cb_packed(_vq); | |
2058 | else | |
2059 | virtqueue_disable_cb_split(_vq); | |
e6f633e5 TB |
2060 | } |
2061 | EXPORT_SYMBOL_GPL(virtqueue_disable_cb); | |
2062 | ||
2063 | /** | |
2064 | * virtqueue_enable_cb_prepare - restart callbacks after disable_cb | |
a5581206 | 2065 | * @_vq: the struct virtqueue we're talking about. |
e6f633e5 TB |
2066 | * |
2067 | * This re-enables callbacks; it returns current queue state | |
2068 | * in an opaque unsigned value. This value should be later tested by | |
2069 | * virtqueue_poll, to detect a possible race between the driver checking for | |
2070 | * more work, and enabling callbacks. | |
2071 | * | |
2072 | * Caller must ensure we don't call this with other virtqueue | |
2073 | * operations at the same time (except where noted). | |
2074 | */ | |
31532340 | 2075 | unsigned int virtqueue_enable_cb_prepare(struct virtqueue *_vq) |
e6f633e5 | 2076 | { |
1ce9e605 TB |
2077 | struct vring_virtqueue *vq = to_vvq(_vq); |
2078 | ||
8d622d21 MT |
2079 | if (vq->event_triggered) |
2080 | vq->event_triggered = false; | |
2081 | ||
1ce9e605 TB |
2082 | return vq->packed_ring ? virtqueue_enable_cb_prepare_packed(_vq) : |
2083 | virtqueue_enable_cb_prepare_split(_vq); | |
e6f633e5 TB |
2084 | } |
2085 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_prepare); | |
2086 | ||
2087 | /** | |
2088 | * virtqueue_poll - query pending used buffers | |
a5581206 | 2089 | * @_vq: the struct virtqueue we're talking about. |
e6f633e5 TB |
2090 | * @last_used_idx: virtqueue state (from call to virtqueue_enable_cb_prepare). |
2091 | * | |
2092 | * Returns "true" if there are pending used buffers in the queue. | |
2093 | * | |
2094 | * This does not need to be serialized. | |
2095 | */ | |
31532340 | 2096 | bool virtqueue_poll(struct virtqueue *_vq, unsigned int last_used_idx) |
e6f633e5 TB |
2097 | { |
2098 | struct vring_virtqueue *vq = to_vvq(_vq); | |
2099 | ||
481a0d74 MW |
2100 | if (unlikely(vq->broken)) |
2101 | return false; | |
2102 | ||
e6f633e5 | 2103 | virtio_mb(vq->weak_barriers); |
1ce9e605 TB |
2104 | return vq->packed_ring ? virtqueue_poll_packed(_vq, last_used_idx) : |
2105 | virtqueue_poll_split(_vq, last_used_idx); | |
e6f633e5 TB |
2106 | } |
2107 | EXPORT_SYMBOL_GPL(virtqueue_poll); | |
2108 | ||
2109 | /** | |
2110 | * virtqueue_enable_cb - restart callbacks after disable_cb. | |
a5581206 | 2111 | * @_vq: the struct virtqueue we're talking about. |
e6f633e5 TB |
2112 | * |
2113 | * This re-enables callbacks; it returns "false" if there are pending | |
2114 | * buffers in the queue, to detect a possible race between the driver | |
2115 | * checking for more work, and enabling callbacks. | |
2116 | * | |
2117 | * Caller must ensure we don't call this with other virtqueue | |
2118 | * operations at the same time (except where noted). | |
2119 | */ | |
2120 | bool virtqueue_enable_cb(struct virtqueue *_vq) | |
2121 | { | |
31532340 | 2122 | unsigned int last_used_idx = virtqueue_enable_cb_prepare(_vq); |
e6f633e5 TB |
2123 | |
2124 | return !virtqueue_poll(_vq, last_used_idx); | |
2125 | } | |
2126 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb); | |
2127 | ||
2128 | /** | |
2129 | * virtqueue_enable_cb_delayed - restart callbacks after disable_cb. | |
a5581206 | 2130 | * @_vq: the struct virtqueue we're talking about. |
e6f633e5 TB |
2131 | * |
2132 | * This re-enables callbacks but hints to the other side to delay | |
2133 | * interrupts until most of the available buffers have been processed; | |
2134 | * it returns "false" if there are many pending buffers in the queue, | |
2135 | * to detect a possible race between the driver checking for more work, | |
2136 | * and enabling callbacks. | |
2137 | * | |
2138 | * Caller must ensure we don't call this with other virtqueue | |
2139 | * operations at the same time (except where noted). | |
2140 | */ | |
2141 | bool virtqueue_enable_cb_delayed(struct virtqueue *_vq) | |
2142 | { | |
1ce9e605 TB |
2143 | struct vring_virtqueue *vq = to_vvq(_vq); |
2144 | ||
8d622d21 MT |
2145 | if (vq->event_triggered) |
2146 | vq->event_triggered = false; | |
2147 | ||
1ce9e605 TB |
2148 | return vq->packed_ring ? virtqueue_enable_cb_delayed_packed(_vq) : |
2149 | virtqueue_enable_cb_delayed_split(_vq); | |
e6f633e5 TB |
2150 | } |
2151 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed); | |
2152 | ||
138fd251 TB |
2153 | /** |
2154 | * virtqueue_detach_unused_buf - detach first unused buffer | |
a5581206 | 2155 | * @_vq: the struct virtqueue we're talking about. |
138fd251 TB |
2156 | * |
2157 | * Returns NULL or the "data" token handed to virtqueue_add_*(). | |
a62eecb3 XZ |
2158 | * This is not valid on an active queue; it is useful for device |
2159 | * shutdown or the reset queue. | |
138fd251 TB |
2160 | */ |
2161 | void *virtqueue_detach_unused_buf(struct virtqueue *_vq) | |
2162 | { | |
1ce9e605 TB |
2163 | struct vring_virtqueue *vq = to_vvq(_vq); |
2164 | ||
2165 | return vq->packed_ring ? virtqueue_detach_unused_buf_packed(_vq) : | |
2166 | virtqueue_detach_unused_buf_split(_vq); | |
138fd251 | 2167 | } |
7c5e9ed0 | 2168 | EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf); |
c021eac4 | 2169 | |
138fd251 TB |
2170 | static inline bool more_used(const struct vring_virtqueue *vq) |
2171 | { | |
1ce9e605 | 2172 | return vq->packed_ring ? more_used_packed(vq) : more_used_split(vq); |
138fd251 TB |
2173 | } |
2174 | ||
0a8a69dd RR |
2175 | irqreturn_t vring_interrupt(int irq, void *_vq) |
2176 | { | |
2177 | struct vring_virtqueue *vq = to_vvq(_vq); | |
2178 | ||
2179 | if (!more_used(vq)) { | |
2180 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
2181 | return IRQ_NONE; | |
2182 | } | |
2183 | ||
8b4ec69d | 2184 | if (unlikely(vq->broken)) { |
c346dae4 | 2185 | #ifdef CONFIG_VIRTIO_HARDEN_NOTIFICATION |
8b4ec69d JW |
2186 | dev_warn_once(&vq->vq.vdev->dev, |
2187 | "virtio vring IRQ raised before DRIVER_OK"); | |
2188 | return IRQ_NONE; | |
c346dae4 JW |
2189 | #else |
2190 | return IRQ_HANDLED; | |
2191 | #endif | |
8b4ec69d | 2192 | } |
0a8a69dd | 2193 | |
8d622d21 MT |
2194 | /* Just a hint for performance: so it's ok that this can be racy! */ |
2195 | if (vq->event) | |
2196 | vq->event_triggered = true; | |
2197 | ||
0a8a69dd | 2198 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); |
18445c4d RR |
2199 | if (vq->vq.callback) |
2200 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
2201 | |
2202 | return IRQ_HANDLED; | |
2203 | } | |
c6fd4701 | 2204 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd | 2205 | |
1ce9e605 | 2206 | /* Only available for split ring */ |
07d9629d | 2207 | static struct virtqueue *__vring_new_virtqueue(unsigned int index, |
cd4c812a | 2208 | struct vring_virtqueue_split *vring_split, |
07d9629d XZ |
2209 | struct virtio_device *vdev, |
2210 | bool weak_barriers, | |
2211 | bool context, | |
2212 | bool (*notify)(struct virtqueue *), | |
2213 | void (*callback)(struct virtqueue *), | |
2214 | const char *name) | |
0a8a69dd | 2215 | { |
2a2d1382 | 2216 | struct vring_virtqueue *vq; |
0a8a69dd | 2217 | |
1ce9e605 TB |
2218 | if (virtio_has_feature(vdev, VIRTIO_F_RING_PACKED)) |
2219 | return NULL; | |
2220 | ||
cbeedb72 | 2221 | vq = kmalloc(sizeof(*vq), GFP_KERNEL); |
0a8a69dd RR |
2222 | if (!vq) |
2223 | return NULL; | |
2224 | ||
1ce9e605 | 2225 | vq->packed_ring = false; |
0a8a69dd RR |
2226 | vq->vq.callback = callback; |
2227 | vq->vq.vdev = vdev; | |
9499f5e7 | 2228 | vq->vq.name = name; |
06ca287d | 2229 | vq->vq.index = index; |
2a2d1382 | 2230 | vq->we_own_ring = false; |
0a8a69dd | 2231 | vq->notify = notify; |
7b21e34f | 2232 | vq->weak_barriers = weak_barriers; |
c346dae4 | 2233 | #ifdef CONFIG_VIRTIO_HARDEN_NOTIFICATION |
8b4ec69d | 2234 | vq->broken = true; |
c346dae4 JW |
2235 | #else |
2236 | vq->broken = false; | |
2237 | #endif | |
fb3fba6b | 2238 | vq->use_dma_api = vring_use_dma_api(vdev); |
0a8a69dd | 2239 | |
5a08b04f MT |
2240 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC) && |
2241 | !context; | |
a5c262c5 | 2242 | vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); |
9fa29b9d | 2243 | |
45383fb0 TB |
2244 | if (virtio_has_feature(vdev, VIRTIO_F_ORDER_PLATFORM)) |
2245 | vq->weak_barriers = false; | |
2246 | ||
d79dca75 TB |
2247 | vq->split.queue_dma_addr = 0; |
2248 | vq->split.queue_size_in_bytes = 0; | |
2249 | ||
cd4c812a | 2250 | vq->split.vring = vring_split->vring; |
e593bf97 TB |
2251 | vq->split.avail_flags_shadow = 0; |
2252 | vq->split.avail_idx_shadow = 0; | |
2253 | ||
0a8a69dd | 2254 | /* No callback? Tell other side not to bother us. */ |
f277ec42 | 2255 | if (!callback) { |
e593bf97 | 2256 | vq->split.avail_flags_shadow |= VRING_AVAIL_F_NO_INTERRUPT; |
0ea1e4a6 | 2257 | if (!vq->event) |
e593bf97 TB |
2258 | vq->split.vring.avail->flags = cpu_to_virtio16(vdev, |
2259 | vq->split.avail_flags_shadow); | |
f277ec42 | 2260 | } |
0a8a69dd | 2261 | |
cd4c812a | 2262 | vq->split.desc_state = kmalloc_array(vring_split->vring.num, |
cbeedb72 | 2263 | sizeof(struct vring_desc_state_split), GFP_KERNEL); |
5bc72234 JW |
2264 | if (!vq->split.desc_state) |
2265 | goto err_state; | |
cbeedb72 | 2266 | |
cd4c812a | 2267 | vq->split.desc_extra = vring_alloc_desc_extra(vring_split->vring.num); |
72b5e895 JW |
2268 | if (!vq->split.desc_extra) |
2269 | goto err_extra; | |
2270 | ||
0a8a69dd | 2271 | /* Put everything in free lists. */ |
0a8a69dd | 2272 | vq->free_head = 0; |
cd4c812a | 2273 | memset(vq->split.desc_state, 0, vring_split->vring.num * |
cbeedb72 | 2274 | sizeof(struct vring_desc_state_split)); |
0a8a69dd | 2275 | |
cd4c812a | 2276 | virtqueue_init(vq, vring_split->vring.num); |
3a897128 | 2277 | |
0e566c8f | 2278 | spin_lock(&vdev->vqs_list_lock); |
e152d8af | 2279 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0e566c8f | 2280 | spin_unlock(&vdev->vqs_list_lock); |
0a8a69dd | 2281 | return &vq->vq; |
5bc72234 | 2282 | |
72b5e895 JW |
2283 | err_extra: |
2284 | kfree(vq->split.desc_state); | |
5bc72234 JW |
2285 | err_state: |
2286 | kfree(vq); | |
2287 | return NULL; | |
0a8a69dd | 2288 | } |
2a2d1382 | 2289 | |
2a2d1382 AL |
2290 | struct virtqueue *vring_create_virtqueue( |
2291 | unsigned int index, | |
2292 | unsigned int num, | |
2293 | unsigned int vring_align, | |
2294 | struct virtio_device *vdev, | |
2295 | bool weak_barriers, | |
2296 | bool may_reduce_num, | |
f94682dd | 2297 | bool context, |
2a2d1382 AL |
2298 | bool (*notify)(struct virtqueue *), |
2299 | void (*callback)(struct virtqueue *), | |
2300 | const char *name) | |
2301 | { | |
1ce9e605 TB |
2302 | |
2303 | if (virtio_has_feature(vdev, VIRTIO_F_RING_PACKED)) | |
2304 | return vring_create_virtqueue_packed(index, num, vring_align, | |
2305 | vdev, weak_barriers, may_reduce_num, | |
2306 | context, notify, callback, name); | |
2307 | ||
d79dca75 TB |
2308 | return vring_create_virtqueue_split(index, num, vring_align, |
2309 | vdev, weak_barriers, may_reduce_num, | |
2310 | context, notify, callback, name); | |
2a2d1382 AL |
2311 | } |
2312 | EXPORT_SYMBOL_GPL(vring_create_virtqueue); | |
2313 | ||
1ce9e605 | 2314 | /* Only available for split ring */ |
2a2d1382 AL |
2315 | struct virtqueue *vring_new_virtqueue(unsigned int index, |
2316 | unsigned int num, | |
2317 | unsigned int vring_align, | |
2318 | struct virtio_device *vdev, | |
2319 | bool weak_barriers, | |
f94682dd | 2320 | bool context, |
2a2d1382 AL |
2321 | void *pages, |
2322 | bool (*notify)(struct virtqueue *vq), | |
2323 | void (*callback)(struct virtqueue *vq), | |
2324 | const char *name) | |
2325 | { | |
cd4c812a | 2326 | struct vring_virtqueue_split vring_split = {}; |
1ce9e605 TB |
2327 | |
2328 | if (virtio_has_feature(vdev, VIRTIO_F_RING_PACKED)) | |
2329 | return NULL; | |
2330 | ||
cd4c812a XZ |
2331 | vring_init(&vring_split.vring, num, pages, vring_align); |
2332 | return __vring_new_virtqueue(index, &vring_split, vdev, weak_barriers, | |
2333 | context, notify, callback, name); | |
2a2d1382 | 2334 | } |
c6fd4701 | 2335 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd | 2336 | |
3ea19e32 | 2337 | static void vring_free(struct virtqueue *_vq) |
0a8a69dd | 2338 | { |
2a2d1382 AL |
2339 | struct vring_virtqueue *vq = to_vvq(_vq); |
2340 | ||
2341 | if (vq->we_own_ring) { | |
1ce9e605 TB |
2342 | if (vq->packed_ring) { |
2343 | vring_free_queue(vq->vq.vdev, | |
2344 | vq->packed.ring_size_in_bytes, | |
2345 | vq->packed.vring.desc, | |
2346 | vq->packed.ring_dma_addr); | |
2347 | ||
2348 | vring_free_queue(vq->vq.vdev, | |
2349 | vq->packed.event_size_in_bytes, | |
2350 | vq->packed.vring.driver, | |
2351 | vq->packed.driver_event_dma_addr); | |
2352 | ||
2353 | vring_free_queue(vq->vq.vdev, | |
2354 | vq->packed.event_size_in_bytes, | |
2355 | vq->packed.vring.device, | |
2356 | vq->packed.device_event_dma_addr); | |
2357 | ||
2358 | kfree(vq->packed.desc_state); | |
2359 | kfree(vq->packed.desc_extra); | |
2360 | } else { | |
2361 | vring_free_queue(vq->vq.vdev, | |
2362 | vq->split.queue_size_in_bytes, | |
2363 | vq->split.vring.desc, | |
2364 | vq->split.queue_dma_addr); | |
1ce9e605 | 2365 | } |
2a2d1382 | 2366 | } |
72b5e895 | 2367 | if (!vq->packed_ring) { |
f13f09a1 | 2368 | kfree(vq->split.desc_state); |
72b5e895 JW |
2369 | kfree(vq->split.desc_extra); |
2370 | } | |
3ea19e32 XZ |
2371 | } |
2372 | ||
2373 | void vring_del_virtqueue(struct virtqueue *_vq) | |
2374 | { | |
2375 | struct vring_virtqueue *vq = to_vvq(_vq); | |
2376 | ||
2377 | spin_lock(&vq->vq.vdev->vqs_list_lock); | |
2378 | list_del(&_vq->list); | |
2379 | spin_unlock(&vq->vq.vdev->vqs_list_lock); | |
2380 | ||
2381 | vring_free(_vq); | |
2382 | ||
2a2d1382 | 2383 | kfree(vq); |
0a8a69dd | 2384 | } |
c6fd4701 | 2385 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 2386 | |
e34f8725 RR |
2387 | /* Manipulates transport-specific feature bits. */ |
2388 | void vring_transport_features(struct virtio_device *vdev) | |
2389 | { | |
2390 | unsigned int i; | |
2391 | ||
2392 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
2393 | switch (i) { | |
9fa29b9d MM |
2394 | case VIRTIO_RING_F_INDIRECT_DESC: |
2395 | break; | |
a5c262c5 MT |
2396 | case VIRTIO_RING_F_EVENT_IDX: |
2397 | break; | |
747ae34a MT |
2398 | case VIRTIO_F_VERSION_1: |
2399 | break; | |
321bd212 | 2400 | case VIRTIO_F_ACCESS_PLATFORM: |
1a937693 | 2401 | break; |
f959a128 TB |
2402 | case VIRTIO_F_RING_PACKED: |
2403 | break; | |
45383fb0 TB |
2404 | case VIRTIO_F_ORDER_PLATFORM: |
2405 | break; | |
e34f8725 RR |
2406 | default: |
2407 | /* We don't understand this bit. */ | |
e16e12be | 2408 | __virtio_clear_bit(vdev, i); |
e34f8725 RR |
2409 | } |
2410 | } | |
2411 | } | |
2412 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
2413 | ||
5dfc1762 RR |
2414 | /** |
2415 | * virtqueue_get_vring_size - return the size of the virtqueue's vring | |
a5581206 | 2416 | * @_vq: the struct virtqueue containing the vring of interest. |
5dfc1762 RR |
2417 | * |
2418 | * Returns the size of the vring. This is mainly used for boasting to | |
2419 | * userspace. Unlike other operations, this need not be serialized. | |
2420 | */ | |
8f9f4668 RJ |
2421 | unsigned int virtqueue_get_vring_size(struct virtqueue *_vq) |
2422 | { | |
2423 | ||
2424 | struct vring_virtqueue *vq = to_vvq(_vq); | |
2425 | ||
1ce9e605 | 2426 | return vq->packed_ring ? vq->packed.vring.num : vq->split.vring.num; |
8f9f4668 RJ |
2427 | } |
2428 | EXPORT_SYMBOL_GPL(virtqueue_get_vring_size); | |
2429 | ||
b3b32c94 HG |
2430 | bool virtqueue_is_broken(struct virtqueue *_vq) |
2431 | { | |
2432 | struct vring_virtqueue *vq = to_vvq(_vq); | |
2433 | ||
60f07798 | 2434 | return READ_ONCE(vq->broken); |
b3b32c94 HG |
2435 | } |
2436 | EXPORT_SYMBOL_GPL(virtqueue_is_broken); | |
2437 | ||
e2dcdfe9 RR |
2438 | /* |
2439 | * This should prevent the device from being used, allowing drivers to | |
2440 | * recover. You may need to grab appropriate locks to flush. | |
2441 | */ | |
2442 | void virtio_break_device(struct virtio_device *dev) | |
2443 | { | |
2444 | struct virtqueue *_vq; | |
2445 | ||
0e566c8f | 2446 | spin_lock(&dev->vqs_list_lock); |
e2dcdfe9 RR |
2447 | list_for_each_entry(_vq, &dev->vqs, list) { |
2448 | struct vring_virtqueue *vq = to_vvq(_vq); | |
60f07798 PP |
2449 | |
2450 | /* Pairs with READ_ONCE() in virtqueue_is_broken(). */ | |
2451 | WRITE_ONCE(vq->broken, true); | |
e2dcdfe9 | 2452 | } |
0e566c8f | 2453 | spin_unlock(&dev->vqs_list_lock); |
e2dcdfe9 RR |
2454 | } |
2455 | EXPORT_SYMBOL_GPL(virtio_break_device); | |
2456 | ||
be83f04d JW |
2457 | /* |
2458 | * This should allow the device to be used by the driver. You may | |
2459 | * need to grab appropriate locks to flush the write to | |
2460 | * vq->broken. This should only be used in some specific case e.g | |
2461 | * (probing and restoring). This function should only be called by the | |
2462 | * core, not directly by the driver. | |
2463 | */ | |
2464 | void __virtio_unbreak_device(struct virtio_device *dev) | |
2465 | { | |
2466 | struct virtqueue *_vq; | |
2467 | ||
2468 | spin_lock(&dev->vqs_list_lock); | |
2469 | list_for_each_entry(_vq, &dev->vqs, list) { | |
2470 | struct vring_virtqueue *vq = to_vvq(_vq); | |
2471 | ||
2472 | /* Pairs with READ_ONCE() in virtqueue_is_broken(). */ | |
2473 | WRITE_ONCE(vq->broken, false); | |
2474 | } | |
2475 | spin_unlock(&dev->vqs_list_lock); | |
2476 | } | |
2477 | EXPORT_SYMBOL_GPL(__virtio_unbreak_device); | |
2478 | ||
2a2d1382 | 2479 | dma_addr_t virtqueue_get_desc_addr(struct virtqueue *_vq) |
89062652 CH |
2480 | { |
2481 | struct vring_virtqueue *vq = to_vvq(_vq); | |
2482 | ||
2a2d1382 AL |
2483 | BUG_ON(!vq->we_own_ring); |
2484 | ||
1ce9e605 TB |
2485 | if (vq->packed_ring) |
2486 | return vq->packed.ring_dma_addr; | |
2487 | ||
d79dca75 | 2488 | return vq->split.queue_dma_addr; |
89062652 | 2489 | } |
2a2d1382 | 2490 | EXPORT_SYMBOL_GPL(virtqueue_get_desc_addr); |
89062652 | 2491 | |
2a2d1382 | 2492 | dma_addr_t virtqueue_get_avail_addr(struct virtqueue *_vq) |
89062652 CH |
2493 | { |
2494 | struct vring_virtqueue *vq = to_vvq(_vq); | |
2495 | ||
2a2d1382 AL |
2496 | BUG_ON(!vq->we_own_ring); |
2497 | ||
1ce9e605 TB |
2498 | if (vq->packed_ring) |
2499 | return vq->packed.driver_event_dma_addr; | |
2500 | ||
d79dca75 | 2501 | return vq->split.queue_dma_addr + |
e593bf97 | 2502 | ((char *)vq->split.vring.avail - (char *)vq->split.vring.desc); |
2a2d1382 AL |
2503 | } |
2504 | EXPORT_SYMBOL_GPL(virtqueue_get_avail_addr); | |
2505 | ||
2506 | dma_addr_t virtqueue_get_used_addr(struct virtqueue *_vq) | |
2507 | { | |
2508 | struct vring_virtqueue *vq = to_vvq(_vq); | |
2509 | ||
2510 | BUG_ON(!vq->we_own_ring); | |
2511 | ||
1ce9e605 TB |
2512 | if (vq->packed_ring) |
2513 | return vq->packed.device_event_dma_addr; | |
2514 | ||
d79dca75 | 2515 | return vq->split.queue_dma_addr + |
e593bf97 | 2516 | ((char *)vq->split.vring.used - (char *)vq->split.vring.desc); |
2a2d1382 AL |
2517 | } |
2518 | EXPORT_SYMBOL_GPL(virtqueue_get_used_addr); | |
2519 | ||
1ce9e605 | 2520 | /* Only available for split ring */ |
2a2d1382 AL |
2521 | const struct vring *virtqueue_get_vring(struct virtqueue *vq) |
2522 | { | |
e593bf97 | 2523 | return &to_vvq(vq)->split.vring; |
89062652 | 2524 | } |
2a2d1382 | 2525 | EXPORT_SYMBOL_GPL(virtqueue_get_vring); |
89062652 | 2526 | |
c6fd4701 | 2527 | MODULE_LICENSE("GPL"); |