]>
Commit | Line | Data |
---|---|---|
0a8a69dd RR |
1 | /* Virtio ring implementation. |
2 | * | |
3 | * Copyright 2007 Rusty Russell IBM Corporation | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
16 | * along with this program; if not, write to the Free Software | |
17 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
18 | */ | |
19 | #include <linux/virtio.h> | |
20 | #include <linux/virtio_ring.h> | |
e34f8725 | 21 | #include <linux/virtio_config.h> |
0a8a69dd | 22 | #include <linux/device.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
b5a2c4f1 | 24 | #include <linux/module.h> |
e93300b1 | 25 | #include <linux/hrtimer.h> |
6abb2dd9 | 26 | #include <linux/kmemleak.h> |
780bc790 | 27 | #include <linux/dma-mapping.h> |
78fe3987 | 28 | #include <xen/xen.h> |
0a8a69dd RR |
29 | |
30 | #ifdef DEBUG | |
31 | /* For development, we want to crash whenever the ring is screwed. */ | |
9499f5e7 RR |
32 | #define BAD_RING(_vq, fmt, args...) \ |
33 | do { \ | |
34 | dev_err(&(_vq)->vq.vdev->dev, \ | |
35 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
36 | BUG(); \ | |
37 | } while (0) | |
c5f841f1 RR |
38 | /* Caller is supposed to guarantee no reentry. */ |
39 | #define START_USE(_vq) \ | |
40 | do { \ | |
41 | if ((_vq)->in_use) \ | |
9499f5e7 RR |
42 | panic("%s:in_use = %i\n", \ |
43 | (_vq)->vq.name, (_vq)->in_use); \ | |
c5f841f1 | 44 | (_vq)->in_use = __LINE__; \ |
9499f5e7 | 45 | } while (0) |
3a35ce7d | 46 | #define END_USE(_vq) \ |
97a545ab | 47 | do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0) |
0a8a69dd | 48 | #else |
9499f5e7 RR |
49 | #define BAD_RING(_vq, fmt, args...) \ |
50 | do { \ | |
51 | dev_err(&_vq->vq.vdev->dev, \ | |
52 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
53 | (_vq)->broken = true; \ | |
54 | } while (0) | |
0a8a69dd RR |
55 | #define START_USE(vq) |
56 | #define END_USE(vq) | |
57 | #endif | |
58 | ||
780bc790 AL |
59 | struct vring_desc_state { |
60 | void *data; /* Data for callback. */ | |
61 | struct vring_desc *indir_desc; /* Indirect descriptor, if any. */ | |
62 | }; | |
63 | ||
43b4f721 | 64 | struct vring_virtqueue { |
0a8a69dd RR |
65 | struct virtqueue vq; |
66 | ||
67 | /* Actual memory layout for this queue */ | |
68 | struct vring vring; | |
69 | ||
7b21e34f RR |
70 | /* Can we use weak barriers? */ |
71 | bool weak_barriers; | |
72 | ||
0a8a69dd RR |
73 | /* Other side has made a mess, don't try any more. */ |
74 | bool broken; | |
75 | ||
9fa29b9d MM |
76 | /* Host supports indirect buffers */ |
77 | bool indirect; | |
78 | ||
a5c262c5 MT |
79 | /* Host publishes avail event idx */ |
80 | bool event; | |
81 | ||
0a8a69dd RR |
82 | /* Head of free buffer list. */ |
83 | unsigned int free_head; | |
84 | /* Number we've added since last sync. */ | |
85 | unsigned int num_added; | |
86 | ||
87 | /* Last used index we've seen. */ | |
1bc4953e | 88 | u16 last_used_idx; |
0a8a69dd | 89 | |
f277ec42 VS |
90 | /* Last written value to avail->flags */ |
91 | u16 avail_flags_shadow; | |
92 | ||
93 | /* Last written value to avail->idx in guest byte order */ | |
94 | u16 avail_idx_shadow; | |
95 | ||
0a8a69dd | 96 | /* How to notify other side. FIXME: commonalize hcalls! */ |
46f9c2b9 | 97 | bool (*notify)(struct virtqueue *vq); |
0a8a69dd | 98 | |
2a2d1382 AL |
99 | /* DMA, allocation, and size information */ |
100 | bool we_own_ring; | |
101 | size_t queue_size_in_bytes; | |
102 | dma_addr_t queue_dma_addr; | |
103 | ||
0a8a69dd RR |
104 | #ifdef DEBUG |
105 | /* They're supposed to lock for us. */ | |
106 | unsigned int in_use; | |
e93300b1 RR |
107 | |
108 | /* Figure out if their kicks are too delayed. */ | |
109 | bool last_add_time_valid; | |
110 | ktime_t last_add_time; | |
0a8a69dd RR |
111 | #endif |
112 | ||
780bc790 AL |
113 | /* Per-descriptor state. */ |
114 | struct vring_desc_state desc_state[]; | |
0a8a69dd RR |
115 | }; |
116 | ||
117 | #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) | |
118 | ||
d26c96c8 | 119 | /* |
1a937693 MT |
120 | * Modern virtio devices have feature bits to specify whether they need a |
121 | * quirk and bypass the IOMMU. If not there, just use the DMA API. | |
122 | * | |
123 | * If there, the interaction between virtio and DMA API is messy. | |
d26c96c8 AL |
124 | * |
125 | * On most systems with virtio, physical addresses match bus addresses, | |
126 | * and it doesn't particularly matter whether we use the DMA API. | |
127 | * | |
128 | * On some systems, including Xen and any system with a physical device | |
129 | * that speaks virtio behind a physical IOMMU, we must use the DMA API | |
130 | * for virtio DMA to work at all. | |
131 | * | |
132 | * On other systems, including SPARC and PPC64, virtio-pci devices are | |
133 | * enumerated as though they are behind an IOMMU, but the virtio host | |
134 | * ignores the IOMMU, so we must either pretend that the IOMMU isn't | |
135 | * there or somehow map everything as the identity. | |
136 | * | |
137 | * For the time being, we preserve historic behavior and bypass the DMA | |
138 | * API. | |
1a937693 MT |
139 | * |
140 | * TODO: install a per-device DMA ops structure that does the right thing | |
141 | * taking into account all the above quirks, and use the DMA API | |
142 | * unconditionally on data path. | |
d26c96c8 AL |
143 | */ |
144 | ||
145 | static bool vring_use_dma_api(struct virtio_device *vdev) | |
146 | { | |
1a937693 MT |
147 | if (!virtio_has_iommu_quirk(vdev)) |
148 | return true; | |
149 | ||
150 | /* Otherwise, we are left to guess. */ | |
78fe3987 AL |
151 | /* |
152 | * In theory, it's possible to have a buggy QEMU-supposed | |
153 | * emulated Q35 IOMMU and Xen enabled at the same time. On | |
154 | * such a configuration, virtio has never worked and will | |
155 | * not work without an even larger kludge. Instead, enable | |
156 | * the DMA API if we're a Xen guest, which at least allows | |
157 | * all of the sensible Xen configurations to work correctly. | |
158 | */ | |
159 | if (xen_domain()) | |
160 | return true; | |
161 | ||
d26c96c8 AL |
162 | return false; |
163 | } | |
164 | ||
780bc790 AL |
165 | /* |
166 | * The DMA ops on various arches are rather gnarly right now, and | |
167 | * making all of the arch DMA ops work on the vring device itself | |
168 | * is a mess. For now, we use the parent device for DMA ops. | |
169 | */ | |
75bfa81b | 170 | static inline struct device *vring_dma_dev(const struct vring_virtqueue *vq) |
780bc790 AL |
171 | { |
172 | return vq->vq.vdev->dev.parent; | |
173 | } | |
174 | ||
175 | /* Map one sg entry. */ | |
176 | static dma_addr_t vring_map_one_sg(const struct vring_virtqueue *vq, | |
177 | struct scatterlist *sg, | |
178 | enum dma_data_direction direction) | |
179 | { | |
180 | if (!vring_use_dma_api(vq->vq.vdev)) | |
181 | return (dma_addr_t)sg_phys(sg); | |
182 | ||
183 | /* | |
184 | * We can't use dma_map_sg, because we don't use scatterlists in | |
185 | * the way it expects (we don't guarantee that the scatterlist | |
186 | * will exist for the lifetime of the mapping). | |
187 | */ | |
188 | return dma_map_page(vring_dma_dev(vq), | |
189 | sg_page(sg), sg->offset, sg->length, | |
190 | direction); | |
191 | } | |
192 | ||
193 | static dma_addr_t vring_map_single(const struct vring_virtqueue *vq, | |
194 | void *cpu_addr, size_t size, | |
195 | enum dma_data_direction direction) | |
196 | { | |
197 | if (!vring_use_dma_api(vq->vq.vdev)) | |
198 | return (dma_addr_t)virt_to_phys(cpu_addr); | |
199 | ||
200 | return dma_map_single(vring_dma_dev(vq), | |
201 | cpu_addr, size, direction); | |
202 | } | |
203 | ||
204 | static void vring_unmap_one(const struct vring_virtqueue *vq, | |
205 | struct vring_desc *desc) | |
206 | { | |
207 | u16 flags; | |
208 | ||
209 | if (!vring_use_dma_api(vq->vq.vdev)) | |
210 | return; | |
211 | ||
212 | flags = virtio16_to_cpu(vq->vq.vdev, desc->flags); | |
213 | ||
214 | if (flags & VRING_DESC_F_INDIRECT) { | |
215 | dma_unmap_single(vring_dma_dev(vq), | |
216 | virtio64_to_cpu(vq->vq.vdev, desc->addr), | |
217 | virtio32_to_cpu(vq->vq.vdev, desc->len), | |
218 | (flags & VRING_DESC_F_WRITE) ? | |
219 | DMA_FROM_DEVICE : DMA_TO_DEVICE); | |
220 | } else { | |
221 | dma_unmap_page(vring_dma_dev(vq), | |
222 | virtio64_to_cpu(vq->vq.vdev, desc->addr), | |
223 | virtio32_to_cpu(vq->vq.vdev, desc->len), | |
224 | (flags & VRING_DESC_F_WRITE) ? | |
225 | DMA_FROM_DEVICE : DMA_TO_DEVICE); | |
226 | } | |
227 | } | |
228 | ||
229 | static int vring_mapping_error(const struct vring_virtqueue *vq, | |
230 | dma_addr_t addr) | |
231 | { | |
232 | if (!vring_use_dma_api(vq->vq.vdev)) | |
233 | return 0; | |
234 | ||
235 | return dma_mapping_error(vring_dma_dev(vq), addr); | |
236 | } | |
237 | ||
00e6f3d9 MT |
238 | static struct vring_desc *alloc_indirect(struct virtqueue *_vq, |
239 | unsigned int total_sg, gfp_t gfp) | |
9fa29b9d MM |
240 | { |
241 | struct vring_desc *desc; | |
b25bd251 | 242 | unsigned int i; |
9fa29b9d | 243 | |
b92b1b89 WD |
244 | /* |
245 | * We require lowmem mappings for the descriptors because | |
246 | * otherwise virt_to_phys will give us bogus addresses in the | |
247 | * virtqueue. | |
248 | */ | |
82107539 | 249 | gfp &= ~__GFP_HIGHMEM; |
b92b1b89 | 250 | |
13816c76 | 251 | desc = kmalloc(total_sg * sizeof(struct vring_desc), gfp); |
9fa29b9d | 252 | if (!desc) |
b25bd251 | 253 | return NULL; |
9fa29b9d | 254 | |
b25bd251 | 255 | for (i = 0; i < total_sg; i++) |
00e6f3d9 | 256 | desc[i].next = cpu_to_virtio16(_vq->vdev, i + 1); |
b25bd251 | 257 | return desc; |
9fa29b9d MM |
258 | } |
259 | ||
13816c76 RR |
260 | static inline int virtqueue_add(struct virtqueue *_vq, |
261 | struct scatterlist *sgs[], | |
eeebf9b1 | 262 | unsigned int total_sg, |
13816c76 RR |
263 | unsigned int out_sgs, |
264 | unsigned int in_sgs, | |
265 | void *data, | |
5a08b04f | 266 | void *ctx, |
13816c76 | 267 | gfp_t gfp) |
0a8a69dd RR |
268 | { |
269 | struct vring_virtqueue *vq = to_vvq(_vq); | |
13816c76 | 270 | struct scatterlist *sg; |
b25bd251 | 271 | struct vring_desc *desc; |
780bc790 | 272 | unsigned int i, n, avail, descs_used, uninitialized_var(prev), err_idx; |
1fe9b6fe | 273 | int head; |
b25bd251 | 274 | bool indirect; |
0a8a69dd | 275 | |
9fa29b9d MM |
276 | START_USE(vq); |
277 | ||
0a8a69dd | 278 | BUG_ON(data == NULL); |
5a08b04f | 279 | BUG_ON(ctx && vq->indirect); |
9fa29b9d | 280 | |
70670444 RR |
281 | if (unlikely(vq->broken)) { |
282 | END_USE(vq); | |
283 | return -EIO; | |
284 | } | |
285 | ||
e93300b1 RR |
286 | #ifdef DEBUG |
287 | { | |
288 | ktime_t now = ktime_get(); | |
289 | ||
290 | /* No kick or get, with .1 second between? Warn. */ | |
291 | if (vq->last_add_time_valid) | |
292 | WARN_ON(ktime_to_ms(ktime_sub(now, vq->last_add_time)) | |
293 | > 100); | |
294 | vq->last_add_time = now; | |
295 | vq->last_add_time_valid = true; | |
296 | } | |
297 | #endif | |
298 | ||
b25bd251 RR |
299 | BUG_ON(total_sg == 0); |
300 | ||
301 | head = vq->free_head; | |
302 | ||
9fa29b9d MM |
303 | /* If the host supports indirect descriptor tables, and we have multiple |
304 | * buffers, then go indirect. FIXME: tune this threshold */ | |
b25bd251 | 305 | if (vq->indirect && total_sg > 1 && vq->vq.num_free) |
00e6f3d9 | 306 | desc = alloc_indirect(_vq, total_sg, gfp); |
44ed8089 | 307 | else { |
b25bd251 | 308 | desc = NULL; |
44ed8089 RJ |
309 | WARN_ON_ONCE(total_sg > vq->vring.num && !vq->indirect); |
310 | } | |
b25bd251 RR |
311 | |
312 | if (desc) { | |
313 | /* Use a single buffer which doesn't continue */ | |
780bc790 | 314 | indirect = true; |
b25bd251 RR |
315 | /* Set up rest to use this indirect table. */ |
316 | i = 0; | |
317 | descs_used = 1; | |
b25bd251 | 318 | } else { |
780bc790 | 319 | indirect = false; |
b25bd251 RR |
320 | desc = vq->vring.desc; |
321 | i = head; | |
322 | descs_used = total_sg; | |
9fa29b9d MM |
323 | } |
324 | ||
b25bd251 | 325 | if (vq->vq.num_free < descs_used) { |
0a8a69dd | 326 | pr_debug("Can't add buf len %i - avail = %i\n", |
b25bd251 | 327 | descs_used, vq->vq.num_free); |
44653eae RR |
328 | /* FIXME: for historical reasons, we force a notify here if |
329 | * there are outgoing parts to the buffer. Presumably the | |
330 | * host should service the ring ASAP. */ | |
13816c76 | 331 | if (out_sgs) |
44653eae | 332 | vq->notify(&vq->vq); |
58625edf WY |
333 | if (indirect) |
334 | kfree(desc); | |
0a8a69dd RR |
335 | END_USE(vq); |
336 | return -ENOSPC; | |
337 | } | |
338 | ||
13816c76 | 339 | for (n = 0; n < out_sgs; n++) { |
eeebf9b1 | 340 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
780bc790 AL |
341 | dma_addr_t addr = vring_map_one_sg(vq, sg, DMA_TO_DEVICE); |
342 | if (vring_mapping_error(vq, addr)) | |
343 | goto unmap_release; | |
344 | ||
00e6f3d9 | 345 | desc[i].flags = cpu_to_virtio16(_vq->vdev, VRING_DESC_F_NEXT); |
780bc790 | 346 | desc[i].addr = cpu_to_virtio64(_vq->vdev, addr); |
00e6f3d9 | 347 | desc[i].len = cpu_to_virtio32(_vq->vdev, sg->length); |
13816c76 | 348 | prev = i; |
00e6f3d9 | 349 | i = virtio16_to_cpu(_vq->vdev, desc[i].next); |
13816c76 | 350 | } |
0a8a69dd | 351 | } |
13816c76 | 352 | for (; n < (out_sgs + in_sgs); n++) { |
eeebf9b1 | 353 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
780bc790 AL |
354 | dma_addr_t addr = vring_map_one_sg(vq, sg, DMA_FROM_DEVICE); |
355 | if (vring_mapping_error(vq, addr)) | |
356 | goto unmap_release; | |
357 | ||
00e6f3d9 | 358 | desc[i].flags = cpu_to_virtio16(_vq->vdev, VRING_DESC_F_NEXT | VRING_DESC_F_WRITE); |
780bc790 | 359 | desc[i].addr = cpu_to_virtio64(_vq->vdev, addr); |
00e6f3d9 | 360 | desc[i].len = cpu_to_virtio32(_vq->vdev, sg->length); |
13816c76 | 361 | prev = i; |
00e6f3d9 | 362 | i = virtio16_to_cpu(_vq->vdev, desc[i].next); |
13816c76 | 363 | } |
0a8a69dd RR |
364 | } |
365 | /* Last one doesn't continue. */ | |
00e6f3d9 | 366 | desc[prev].flags &= cpu_to_virtio16(_vq->vdev, ~VRING_DESC_F_NEXT); |
0a8a69dd | 367 | |
780bc790 AL |
368 | if (indirect) { |
369 | /* Now that the indirect table is filled in, map it. */ | |
370 | dma_addr_t addr = vring_map_single( | |
371 | vq, desc, total_sg * sizeof(struct vring_desc), | |
372 | DMA_TO_DEVICE); | |
373 | if (vring_mapping_error(vq, addr)) | |
374 | goto unmap_release; | |
375 | ||
376 | vq->vring.desc[head].flags = cpu_to_virtio16(_vq->vdev, VRING_DESC_F_INDIRECT); | |
377 | vq->vring.desc[head].addr = cpu_to_virtio64(_vq->vdev, addr); | |
378 | ||
379 | vq->vring.desc[head].len = cpu_to_virtio32(_vq->vdev, total_sg * sizeof(struct vring_desc)); | |
380 | } | |
381 | ||
382 | /* We're using some buffers from the free list. */ | |
383 | vq->vq.num_free -= descs_used; | |
384 | ||
0a8a69dd | 385 | /* Update free pointer */ |
b25bd251 | 386 | if (indirect) |
00e6f3d9 | 387 | vq->free_head = virtio16_to_cpu(_vq->vdev, vq->vring.desc[head].next); |
b25bd251 RR |
388 | else |
389 | vq->free_head = i; | |
0a8a69dd | 390 | |
780bc790 AL |
391 | /* Store token and indirect buffer state. */ |
392 | vq->desc_state[head].data = data; | |
393 | if (indirect) | |
394 | vq->desc_state[head].indir_desc = desc; | |
87646a34 | 395 | else |
5a08b04f | 396 | vq->desc_state[head].indir_desc = ctx; |
0a8a69dd RR |
397 | |
398 | /* Put entry in available array (but don't update avail->idx until they | |
3b720b8c | 399 | * do sync). */ |
f277ec42 | 400 | avail = vq->avail_idx_shadow & (vq->vring.num - 1); |
00e6f3d9 | 401 | vq->vring.avail->ring[avail] = cpu_to_virtio16(_vq->vdev, head); |
0a8a69dd | 402 | |
ee7cd898 RR |
403 | /* Descriptors and available array need to be set before we expose the |
404 | * new available array entries. */ | |
a9a0fef7 | 405 | virtio_wmb(vq->weak_barriers); |
f277ec42 VS |
406 | vq->avail_idx_shadow++; |
407 | vq->vring.avail->idx = cpu_to_virtio16(_vq->vdev, vq->avail_idx_shadow); | |
ee7cd898 RR |
408 | vq->num_added++; |
409 | ||
5e05bf58 TH |
410 | pr_debug("Added buffer head %i to %p\n", head, vq); |
411 | END_USE(vq); | |
412 | ||
ee7cd898 RR |
413 | /* This is very unlikely, but theoretically possible. Kick |
414 | * just in case. */ | |
415 | if (unlikely(vq->num_added == (1 << 16) - 1)) | |
416 | virtqueue_kick(_vq); | |
417 | ||
98e8c6bc | 418 | return 0; |
780bc790 AL |
419 | |
420 | unmap_release: | |
421 | err_idx = i; | |
422 | i = head; | |
423 | ||
424 | for (n = 0; n < total_sg; n++) { | |
425 | if (i == err_idx) | |
426 | break; | |
427 | vring_unmap_one(vq, &desc[i]); | |
c60923cb | 428 | i = virtio16_to_cpu(_vq->vdev, vq->vring.desc[i].next); |
780bc790 AL |
429 | } |
430 | ||
780bc790 AL |
431 | if (indirect) |
432 | kfree(desc); | |
433 | ||
3cc36f6e | 434 | END_USE(vq); |
780bc790 | 435 | return -EIO; |
0a8a69dd | 436 | } |
13816c76 | 437 | |
13816c76 RR |
438 | /** |
439 | * virtqueue_add_sgs - expose buffers to other end | |
440 | * @vq: the struct virtqueue we're talking about. | |
441 | * @sgs: array of terminated scatterlists. | |
442 | * @out_num: the number of scatterlists readable by other side | |
443 | * @in_num: the number of scatterlists which are writable (after readable ones) | |
444 | * @data: the token identifying the buffer. | |
445 | * @gfp: how to do memory allocations (if necessary). | |
446 | * | |
447 | * Caller must ensure we don't call this with other virtqueue operations | |
448 | * at the same time (except where noted). | |
449 | * | |
70670444 | 450 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
13816c76 RR |
451 | */ |
452 | int virtqueue_add_sgs(struct virtqueue *_vq, | |
453 | struct scatterlist *sgs[], | |
454 | unsigned int out_sgs, | |
455 | unsigned int in_sgs, | |
456 | void *data, | |
457 | gfp_t gfp) | |
458 | { | |
eeebf9b1 | 459 | unsigned int i, total_sg = 0; |
13816c76 RR |
460 | |
461 | /* Count them first. */ | |
eeebf9b1 | 462 | for (i = 0; i < out_sgs + in_sgs; i++) { |
13816c76 RR |
463 | struct scatterlist *sg; |
464 | for (sg = sgs[i]; sg; sg = sg_next(sg)) | |
eeebf9b1 | 465 | total_sg++; |
13816c76 | 466 | } |
5a08b04f MT |
467 | return virtqueue_add(_vq, sgs, total_sg, out_sgs, in_sgs, |
468 | data, NULL, gfp); | |
13816c76 RR |
469 | } |
470 | EXPORT_SYMBOL_GPL(virtqueue_add_sgs); | |
471 | ||
282edb36 RR |
472 | /** |
473 | * virtqueue_add_outbuf - expose output buffers to other end | |
474 | * @vq: the struct virtqueue we're talking about. | |
eeebf9b1 RR |
475 | * @sg: scatterlist (must be well-formed and terminated!) |
476 | * @num: the number of entries in @sg readable by other side | |
282edb36 RR |
477 | * @data: the token identifying the buffer. |
478 | * @gfp: how to do memory allocations (if necessary). | |
479 | * | |
480 | * Caller must ensure we don't call this with other virtqueue operations | |
481 | * at the same time (except where noted). | |
482 | * | |
70670444 | 483 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
282edb36 RR |
484 | */ |
485 | int virtqueue_add_outbuf(struct virtqueue *vq, | |
eeebf9b1 | 486 | struct scatterlist *sg, unsigned int num, |
282edb36 RR |
487 | void *data, |
488 | gfp_t gfp) | |
489 | { | |
5a08b04f | 490 | return virtqueue_add(vq, &sg, num, 1, 0, data, NULL, gfp); |
282edb36 RR |
491 | } |
492 | EXPORT_SYMBOL_GPL(virtqueue_add_outbuf); | |
493 | ||
494 | /** | |
495 | * virtqueue_add_inbuf - expose input buffers to other end | |
496 | * @vq: the struct virtqueue we're talking about. | |
eeebf9b1 RR |
497 | * @sg: scatterlist (must be well-formed and terminated!) |
498 | * @num: the number of entries in @sg writable by other side | |
282edb36 RR |
499 | * @data: the token identifying the buffer. |
500 | * @gfp: how to do memory allocations (if necessary). | |
501 | * | |
502 | * Caller must ensure we don't call this with other virtqueue operations | |
503 | * at the same time (except where noted). | |
504 | * | |
70670444 | 505 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
282edb36 RR |
506 | */ |
507 | int virtqueue_add_inbuf(struct virtqueue *vq, | |
eeebf9b1 | 508 | struct scatterlist *sg, unsigned int num, |
282edb36 RR |
509 | void *data, |
510 | gfp_t gfp) | |
511 | { | |
5a08b04f | 512 | return virtqueue_add(vq, &sg, num, 0, 1, data, NULL, gfp); |
282edb36 RR |
513 | } |
514 | EXPORT_SYMBOL_GPL(virtqueue_add_inbuf); | |
515 | ||
5a08b04f MT |
516 | /** |
517 | * virtqueue_add_inbuf_ctx - expose input buffers to other end | |
518 | * @vq: the struct virtqueue we're talking about. | |
519 | * @sg: scatterlist (must be well-formed and terminated!) | |
520 | * @num: the number of entries in @sg writable by other side | |
521 | * @data: the token identifying the buffer. | |
522 | * @ctx: extra context for the token | |
523 | * @gfp: how to do memory allocations (if necessary). | |
524 | * | |
525 | * Caller must ensure we don't call this with other virtqueue operations | |
526 | * at the same time (except where noted). | |
527 | * | |
528 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). | |
529 | */ | |
530 | int virtqueue_add_inbuf_ctx(struct virtqueue *vq, | |
531 | struct scatterlist *sg, unsigned int num, | |
532 | void *data, | |
533 | void *ctx, | |
534 | gfp_t gfp) | |
535 | { | |
536 | return virtqueue_add(vq, &sg, num, 0, 1, data, ctx, gfp); | |
537 | } | |
538 | EXPORT_SYMBOL_GPL(virtqueue_add_inbuf_ctx); | |
539 | ||
5dfc1762 | 540 | /** |
41f0377f | 541 | * virtqueue_kick_prepare - first half of split virtqueue_kick call. |
5dfc1762 RR |
542 | * @vq: the struct virtqueue |
543 | * | |
41f0377f RR |
544 | * Instead of virtqueue_kick(), you can do: |
545 | * if (virtqueue_kick_prepare(vq)) | |
546 | * virtqueue_notify(vq); | |
5dfc1762 | 547 | * |
41f0377f RR |
548 | * This is sometimes useful because the virtqueue_kick_prepare() needs |
549 | * to be serialized, but the actual virtqueue_notify() call does not. | |
5dfc1762 | 550 | */ |
41f0377f | 551 | bool virtqueue_kick_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
552 | { |
553 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a5c262c5 | 554 | u16 new, old; |
41f0377f RR |
555 | bool needs_kick; |
556 | ||
0a8a69dd | 557 | START_USE(vq); |
a72caae2 JW |
558 | /* We need to expose available array entries before checking avail |
559 | * event. */ | |
a9a0fef7 | 560 | virtio_mb(vq->weak_barriers); |
0a8a69dd | 561 | |
f277ec42 VS |
562 | old = vq->avail_idx_shadow - vq->num_added; |
563 | new = vq->avail_idx_shadow; | |
0a8a69dd RR |
564 | vq->num_added = 0; |
565 | ||
e93300b1 RR |
566 | #ifdef DEBUG |
567 | if (vq->last_add_time_valid) { | |
568 | WARN_ON(ktime_to_ms(ktime_sub(ktime_get(), | |
569 | vq->last_add_time)) > 100); | |
570 | } | |
571 | vq->last_add_time_valid = false; | |
572 | #endif | |
573 | ||
41f0377f | 574 | if (vq->event) { |
00e6f3d9 | 575 | needs_kick = vring_need_event(virtio16_to_cpu(_vq->vdev, vring_avail_event(&vq->vring)), |
41f0377f RR |
576 | new, old); |
577 | } else { | |
00e6f3d9 | 578 | needs_kick = !(vq->vring.used->flags & cpu_to_virtio16(_vq->vdev, VRING_USED_F_NO_NOTIFY)); |
41f0377f | 579 | } |
0a8a69dd | 580 | END_USE(vq); |
41f0377f RR |
581 | return needs_kick; |
582 | } | |
583 | EXPORT_SYMBOL_GPL(virtqueue_kick_prepare); | |
584 | ||
585 | /** | |
586 | * virtqueue_notify - second half of split virtqueue_kick call. | |
587 | * @vq: the struct virtqueue | |
588 | * | |
589 | * This does not need to be serialized. | |
5b1bf7cb HG |
590 | * |
591 | * Returns false if host notify failed or queue is broken, otherwise true. | |
41f0377f | 592 | */ |
5b1bf7cb | 593 | bool virtqueue_notify(struct virtqueue *_vq) |
41f0377f RR |
594 | { |
595 | struct vring_virtqueue *vq = to_vvq(_vq); | |
596 | ||
5b1bf7cb HG |
597 | if (unlikely(vq->broken)) |
598 | return false; | |
599 | ||
41f0377f | 600 | /* Prod other side to tell it about changes. */ |
2342d6a6 | 601 | if (!vq->notify(_vq)) { |
5b1bf7cb HG |
602 | vq->broken = true; |
603 | return false; | |
604 | } | |
605 | return true; | |
41f0377f RR |
606 | } |
607 | EXPORT_SYMBOL_GPL(virtqueue_notify); | |
608 | ||
609 | /** | |
610 | * virtqueue_kick - update after add_buf | |
611 | * @vq: the struct virtqueue | |
612 | * | |
b3087e48 | 613 | * After one or more virtqueue_add_* calls, invoke this to kick |
41f0377f RR |
614 | * the other side. |
615 | * | |
616 | * Caller must ensure we don't call this with other virtqueue | |
617 | * operations at the same time (except where noted). | |
5b1bf7cb HG |
618 | * |
619 | * Returns false if kick failed, otherwise true. | |
41f0377f | 620 | */ |
5b1bf7cb | 621 | bool virtqueue_kick(struct virtqueue *vq) |
41f0377f RR |
622 | { |
623 | if (virtqueue_kick_prepare(vq)) | |
5b1bf7cb HG |
624 | return virtqueue_notify(vq); |
625 | return true; | |
0a8a69dd | 626 | } |
7c5e9ed0 | 627 | EXPORT_SYMBOL_GPL(virtqueue_kick); |
0a8a69dd | 628 | |
5a08b04f MT |
629 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head, |
630 | void **ctx) | |
0a8a69dd | 631 | { |
780bc790 | 632 | unsigned int i, j; |
c60923cb | 633 | __virtio16 nextflag = cpu_to_virtio16(vq->vq.vdev, VRING_DESC_F_NEXT); |
0a8a69dd RR |
634 | |
635 | /* Clear data ptr. */ | |
780bc790 | 636 | vq->desc_state[head].data = NULL; |
0a8a69dd | 637 | |
780bc790 | 638 | /* Put back on free list: unmap first-level descriptors and find end */ |
0a8a69dd | 639 | i = head; |
9fa29b9d | 640 | |
780bc790 AL |
641 | while (vq->vring.desc[i].flags & nextflag) { |
642 | vring_unmap_one(vq, &vq->vring.desc[i]); | |
00e6f3d9 | 643 | i = virtio16_to_cpu(vq->vq.vdev, vq->vring.desc[i].next); |
06ca287d | 644 | vq->vq.num_free++; |
0a8a69dd RR |
645 | } |
646 | ||
780bc790 | 647 | vring_unmap_one(vq, &vq->vring.desc[i]); |
00e6f3d9 | 648 | vq->vring.desc[i].next = cpu_to_virtio16(vq->vq.vdev, vq->free_head); |
0a8a69dd | 649 | vq->free_head = head; |
780bc790 | 650 | |
0a8a69dd | 651 | /* Plus final descriptor */ |
06ca287d | 652 | vq->vq.num_free++; |
780bc790 | 653 | |
5a08b04f | 654 | if (vq->indirect) { |
780bc790 | 655 | struct vring_desc *indir_desc = vq->desc_state[head].indir_desc; |
5a08b04f MT |
656 | u32 len; |
657 | ||
658 | /* Free the indirect table, if any, now that it's unmapped. */ | |
659 | if (!indir_desc) | |
660 | return; | |
661 | ||
662 | len = virtio32_to_cpu(vq->vq.vdev, vq->vring.desc[head].len); | |
780bc790 AL |
663 | |
664 | BUG_ON(!(vq->vring.desc[head].flags & | |
665 | cpu_to_virtio16(vq->vq.vdev, VRING_DESC_F_INDIRECT))); | |
666 | BUG_ON(len == 0 || len % sizeof(struct vring_desc)); | |
667 | ||
668 | for (j = 0; j < len / sizeof(struct vring_desc); j++) | |
669 | vring_unmap_one(vq, &indir_desc[j]); | |
670 | ||
5a08b04f | 671 | kfree(indir_desc); |
780bc790 | 672 | vq->desc_state[head].indir_desc = NULL; |
5a08b04f MT |
673 | } else if (ctx) { |
674 | *ctx = vq->desc_state[head].indir_desc; | |
780bc790 | 675 | } |
0a8a69dd RR |
676 | } |
677 | ||
0a8a69dd RR |
678 | static inline bool more_used(const struct vring_virtqueue *vq) |
679 | { | |
00e6f3d9 | 680 | return vq->last_used_idx != virtio16_to_cpu(vq->vq.vdev, vq->vring.used->idx); |
0a8a69dd RR |
681 | } |
682 | ||
5dfc1762 RR |
683 | /** |
684 | * virtqueue_get_buf - get the next used buffer | |
685 | * @vq: the struct virtqueue we're talking about. | |
686 | * @len: the length written into the buffer | |
687 | * | |
0c7eaf59 | 688 | * If the device wrote data into the buffer, @len will be set to the |
5dfc1762 RR |
689 | * amount written. This means you don't need to clear the buffer |
690 | * beforehand to ensure there's no data leakage in the case of short | |
691 | * writes. | |
692 | * | |
693 | * Caller must ensure we don't call this with other virtqueue | |
694 | * operations at the same time (except where noted). | |
695 | * | |
696 | * Returns NULL if there are no used buffers, or the "data" token | |
b3087e48 | 697 | * handed to virtqueue_add_*(). |
5dfc1762 | 698 | */ |
5a08b04f MT |
699 | void *virtqueue_get_buf_ctx(struct virtqueue *_vq, unsigned int *len, |
700 | void **ctx) | |
0a8a69dd RR |
701 | { |
702 | struct vring_virtqueue *vq = to_vvq(_vq); | |
703 | void *ret; | |
704 | unsigned int i; | |
3b720b8c | 705 | u16 last_used; |
0a8a69dd RR |
706 | |
707 | START_USE(vq); | |
708 | ||
5ef82752 RR |
709 | if (unlikely(vq->broken)) { |
710 | END_USE(vq); | |
711 | return NULL; | |
712 | } | |
713 | ||
0a8a69dd RR |
714 | if (!more_used(vq)) { |
715 | pr_debug("No more buffers in queue\n"); | |
716 | END_USE(vq); | |
717 | return NULL; | |
718 | } | |
719 | ||
2d61ba95 | 720 | /* Only get used array entries after they have been exposed by host. */ |
a9a0fef7 | 721 | virtio_rmb(vq->weak_barriers); |
2d61ba95 | 722 | |
3b720b8c | 723 | last_used = (vq->last_used_idx & (vq->vring.num - 1)); |
00e6f3d9 MT |
724 | i = virtio32_to_cpu(_vq->vdev, vq->vring.used->ring[last_used].id); |
725 | *len = virtio32_to_cpu(_vq->vdev, vq->vring.used->ring[last_used].len); | |
0a8a69dd RR |
726 | |
727 | if (unlikely(i >= vq->vring.num)) { | |
728 | BAD_RING(vq, "id %u out of range\n", i); | |
729 | return NULL; | |
730 | } | |
780bc790 | 731 | if (unlikely(!vq->desc_state[i].data)) { |
0a8a69dd RR |
732 | BAD_RING(vq, "id %u is not a head!\n", i); |
733 | return NULL; | |
734 | } | |
735 | ||
736 | /* detach_buf clears data, so grab it now. */ | |
780bc790 | 737 | ret = vq->desc_state[i].data; |
5a08b04f | 738 | detach_buf(vq, i, ctx); |
0a8a69dd | 739 | vq->last_used_idx++; |
a5c262c5 MT |
740 | /* If we expect an interrupt for the next entry, tell host |
741 | * by writing event index and flush out the write before | |
742 | * the read in the next get_buf call. */ | |
788e5b3a MT |
743 | if (!(vq->avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT)) |
744 | virtio_store_mb(vq->weak_barriers, | |
745 | &vring_used_event(&vq->vring), | |
746 | cpu_to_virtio16(_vq->vdev, vq->last_used_idx)); | |
a5c262c5 | 747 | |
e93300b1 RR |
748 | #ifdef DEBUG |
749 | vq->last_add_time_valid = false; | |
750 | #endif | |
751 | ||
0a8a69dd RR |
752 | END_USE(vq); |
753 | return ret; | |
754 | } | |
5a08b04f | 755 | EXPORT_SYMBOL_GPL(virtqueue_get_buf_ctx); |
0a8a69dd | 756 | |
5a08b04f MT |
757 | void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len) |
758 | { | |
759 | return virtqueue_get_buf_ctx(_vq, len, NULL); | |
760 | } | |
761 | EXPORT_SYMBOL_GPL(virtqueue_get_buf); | |
5dfc1762 RR |
762 | /** |
763 | * virtqueue_disable_cb - disable callbacks | |
764 | * @vq: the struct virtqueue we're talking about. | |
765 | * | |
766 | * Note that this is not necessarily synchronous, hence unreliable and only | |
767 | * useful as an optimization. | |
768 | * | |
769 | * Unlike other operations, this need not be serialized. | |
770 | */ | |
7c5e9ed0 | 771 | void virtqueue_disable_cb(struct virtqueue *_vq) |
18445c4d RR |
772 | { |
773 | struct vring_virtqueue *vq = to_vvq(_vq); | |
774 | ||
f277ec42 VS |
775 | if (!(vq->avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT)) { |
776 | vq->avail_flags_shadow |= VRING_AVAIL_F_NO_INTERRUPT; | |
0ea1e4a6 LP |
777 | if (!vq->event) |
778 | vq->vring.avail->flags = cpu_to_virtio16(_vq->vdev, vq->avail_flags_shadow); | |
f277ec42 VS |
779 | } |
780 | ||
18445c4d | 781 | } |
7c5e9ed0 | 782 | EXPORT_SYMBOL_GPL(virtqueue_disable_cb); |
18445c4d | 783 | |
5dfc1762 | 784 | /** |
cc229884 | 785 | * virtqueue_enable_cb_prepare - restart callbacks after disable_cb |
5dfc1762 RR |
786 | * @vq: the struct virtqueue we're talking about. |
787 | * | |
cc229884 MT |
788 | * This re-enables callbacks; it returns current queue state |
789 | * in an opaque unsigned value. This value should be later tested by | |
790 | * virtqueue_poll, to detect a possible race between the driver checking for | |
791 | * more work, and enabling callbacks. | |
5dfc1762 RR |
792 | * |
793 | * Caller must ensure we don't call this with other virtqueue | |
794 | * operations at the same time (except where noted). | |
795 | */ | |
cc229884 | 796 | unsigned virtqueue_enable_cb_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
797 | { |
798 | struct vring_virtqueue *vq = to_vvq(_vq); | |
cc229884 | 799 | u16 last_used_idx; |
0a8a69dd RR |
800 | |
801 | START_USE(vq); | |
0a8a69dd RR |
802 | |
803 | /* We optimistically turn back on interrupts, then check if there was | |
804 | * more to do. */ | |
a5c262c5 MT |
805 | /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to |
806 | * either clear the flags bit or point the event index at the next | |
807 | * entry. Always do both to keep code simple. */ | |
f277ec42 VS |
808 | if (vq->avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT) { |
809 | vq->avail_flags_shadow &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
0ea1e4a6 LP |
810 | if (!vq->event) |
811 | vq->vring.avail->flags = cpu_to_virtio16(_vq->vdev, vq->avail_flags_shadow); | |
f277ec42 | 812 | } |
00e6f3d9 | 813 | vring_used_event(&vq->vring) = cpu_to_virtio16(_vq->vdev, last_used_idx = vq->last_used_idx); |
cc229884 MT |
814 | END_USE(vq); |
815 | return last_used_idx; | |
816 | } | |
817 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_prepare); | |
818 | ||
819 | /** | |
820 | * virtqueue_poll - query pending used buffers | |
821 | * @vq: the struct virtqueue we're talking about. | |
822 | * @last_used_idx: virtqueue state (from call to virtqueue_enable_cb_prepare). | |
823 | * | |
824 | * Returns "true" if there are pending used buffers in the queue. | |
825 | * | |
826 | * This does not need to be serialized. | |
827 | */ | |
828 | bool virtqueue_poll(struct virtqueue *_vq, unsigned last_used_idx) | |
829 | { | |
830 | struct vring_virtqueue *vq = to_vvq(_vq); | |
831 | ||
a9a0fef7 | 832 | virtio_mb(vq->weak_barriers); |
00e6f3d9 | 833 | return (u16)last_used_idx != virtio16_to_cpu(_vq->vdev, vq->vring.used->idx); |
cc229884 MT |
834 | } |
835 | EXPORT_SYMBOL_GPL(virtqueue_poll); | |
0a8a69dd | 836 | |
cc229884 MT |
837 | /** |
838 | * virtqueue_enable_cb - restart callbacks after disable_cb. | |
839 | * @vq: the struct virtqueue we're talking about. | |
840 | * | |
841 | * This re-enables callbacks; it returns "false" if there are pending | |
842 | * buffers in the queue, to detect a possible race between the driver | |
843 | * checking for more work, and enabling callbacks. | |
844 | * | |
845 | * Caller must ensure we don't call this with other virtqueue | |
846 | * operations at the same time (except where noted). | |
847 | */ | |
848 | bool virtqueue_enable_cb(struct virtqueue *_vq) | |
849 | { | |
850 | unsigned last_used_idx = virtqueue_enable_cb_prepare(_vq); | |
851 | return !virtqueue_poll(_vq, last_used_idx); | |
0a8a69dd | 852 | } |
7c5e9ed0 | 853 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb); |
0a8a69dd | 854 | |
5dfc1762 RR |
855 | /** |
856 | * virtqueue_enable_cb_delayed - restart callbacks after disable_cb. | |
857 | * @vq: the struct virtqueue we're talking about. | |
858 | * | |
859 | * This re-enables callbacks but hints to the other side to delay | |
860 | * interrupts until most of the available buffers have been processed; | |
861 | * it returns "false" if there are many pending buffers in the queue, | |
862 | * to detect a possible race between the driver checking for more work, | |
863 | * and enabling callbacks. | |
864 | * | |
865 | * Caller must ensure we don't call this with other virtqueue | |
866 | * operations at the same time (except where noted). | |
867 | */ | |
7ab358c2 MT |
868 | bool virtqueue_enable_cb_delayed(struct virtqueue *_vq) |
869 | { | |
870 | struct vring_virtqueue *vq = to_vvq(_vq); | |
871 | u16 bufs; | |
872 | ||
873 | START_USE(vq); | |
874 | ||
875 | /* We optimistically turn back on interrupts, then check if there was | |
876 | * more to do. */ | |
877 | /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to | |
878 | * either clear the flags bit or point the event index at the next | |
0ea1e4a6 | 879 | * entry. Always update the event index to keep code simple. */ |
f277ec42 VS |
880 | if (vq->avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT) { |
881 | vq->avail_flags_shadow &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
0ea1e4a6 LP |
882 | if (!vq->event) |
883 | vq->vring.avail->flags = cpu_to_virtio16(_vq->vdev, vq->avail_flags_shadow); | |
f277ec42 | 884 | } |
7ab358c2 | 885 | /* TODO: tune this threshold */ |
f277ec42 | 886 | bufs = (u16)(vq->avail_idx_shadow - vq->last_used_idx) * 3 / 4; |
788e5b3a MT |
887 | |
888 | virtio_store_mb(vq->weak_barriers, | |
889 | &vring_used_event(&vq->vring), | |
890 | cpu_to_virtio16(_vq->vdev, vq->last_used_idx + bufs)); | |
891 | ||
00e6f3d9 | 892 | if (unlikely((u16)(virtio16_to_cpu(_vq->vdev, vq->vring.used->idx) - vq->last_used_idx) > bufs)) { |
7ab358c2 MT |
893 | END_USE(vq); |
894 | return false; | |
895 | } | |
896 | ||
897 | END_USE(vq); | |
898 | return true; | |
899 | } | |
900 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed); | |
901 | ||
5dfc1762 RR |
902 | /** |
903 | * virtqueue_detach_unused_buf - detach first unused buffer | |
904 | * @vq: the struct virtqueue we're talking about. | |
905 | * | |
b3087e48 | 906 | * Returns NULL or the "data" token handed to virtqueue_add_*(). |
5dfc1762 RR |
907 | * This is not valid on an active queue; it is useful only for device |
908 | * shutdown. | |
909 | */ | |
7c5e9ed0 | 910 | void *virtqueue_detach_unused_buf(struct virtqueue *_vq) |
c021eac4 SM |
911 | { |
912 | struct vring_virtqueue *vq = to_vvq(_vq); | |
913 | unsigned int i; | |
914 | void *buf; | |
915 | ||
916 | START_USE(vq); | |
917 | ||
918 | for (i = 0; i < vq->vring.num; i++) { | |
780bc790 | 919 | if (!vq->desc_state[i].data) |
c021eac4 SM |
920 | continue; |
921 | /* detach_buf clears data, so grab it now. */ | |
780bc790 | 922 | buf = vq->desc_state[i].data; |
5a08b04f | 923 | detach_buf(vq, i, NULL); |
f277ec42 VS |
924 | vq->avail_idx_shadow--; |
925 | vq->vring.avail->idx = cpu_to_virtio16(_vq->vdev, vq->avail_idx_shadow); | |
c021eac4 SM |
926 | END_USE(vq); |
927 | return buf; | |
928 | } | |
929 | /* That should have freed everything. */ | |
06ca287d | 930 | BUG_ON(vq->vq.num_free != vq->vring.num); |
c021eac4 SM |
931 | |
932 | END_USE(vq); | |
933 | return NULL; | |
934 | } | |
7c5e9ed0 | 935 | EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf); |
c021eac4 | 936 | |
0a8a69dd RR |
937 | irqreturn_t vring_interrupt(int irq, void *_vq) |
938 | { | |
939 | struct vring_virtqueue *vq = to_vvq(_vq); | |
940 | ||
941 | if (!more_used(vq)) { | |
942 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
943 | return IRQ_NONE; | |
944 | } | |
945 | ||
946 | if (unlikely(vq->broken)) | |
947 | return IRQ_HANDLED; | |
948 | ||
949 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); | |
18445c4d RR |
950 | if (vq->vq.callback) |
951 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
952 | |
953 | return IRQ_HANDLED; | |
954 | } | |
c6fd4701 | 955 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd | 956 | |
2a2d1382 AL |
957 | struct virtqueue *__vring_new_virtqueue(unsigned int index, |
958 | struct vring vring, | |
959 | struct virtio_device *vdev, | |
960 | bool weak_barriers, | |
f94682dd | 961 | bool context, |
2a2d1382 AL |
962 | bool (*notify)(struct virtqueue *), |
963 | void (*callback)(struct virtqueue *), | |
964 | const char *name) | |
0a8a69dd | 965 | { |
0a8a69dd | 966 | unsigned int i; |
2a2d1382 | 967 | struct vring_virtqueue *vq; |
0a8a69dd | 968 | |
2a2d1382 | 969 | vq = kmalloc(sizeof(*vq) + vring.num * sizeof(struct vring_desc_state), |
780bc790 | 970 | GFP_KERNEL); |
0a8a69dd RR |
971 | if (!vq) |
972 | return NULL; | |
973 | ||
2a2d1382 | 974 | vq->vring = vring; |
0a8a69dd RR |
975 | vq->vq.callback = callback; |
976 | vq->vq.vdev = vdev; | |
9499f5e7 | 977 | vq->vq.name = name; |
2a2d1382 | 978 | vq->vq.num_free = vring.num; |
06ca287d | 979 | vq->vq.index = index; |
2a2d1382 AL |
980 | vq->we_own_ring = false; |
981 | vq->queue_dma_addr = 0; | |
982 | vq->queue_size_in_bytes = 0; | |
0a8a69dd | 983 | vq->notify = notify; |
7b21e34f | 984 | vq->weak_barriers = weak_barriers; |
0a8a69dd RR |
985 | vq->broken = false; |
986 | vq->last_used_idx = 0; | |
f277ec42 VS |
987 | vq->avail_flags_shadow = 0; |
988 | vq->avail_idx_shadow = 0; | |
0a8a69dd | 989 | vq->num_added = 0; |
9499f5e7 | 990 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0a8a69dd RR |
991 | #ifdef DEBUG |
992 | vq->in_use = false; | |
e93300b1 | 993 | vq->last_add_time_valid = false; |
0a8a69dd RR |
994 | #endif |
995 | ||
5a08b04f MT |
996 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC) && |
997 | !context; | |
a5c262c5 | 998 | vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); |
9fa29b9d | 999 | |
0a8a69dd | 1000 | /* No callback? Tell other side not to bother us. */ |
f277ec42 VS |
1001 | if (!callback) { |
1002 | vq->avail_flags_shadow |= VRING_AVAIL_F_NO_INTERRUPT; | |
0ea1e4a6 LP |
1003 | if (!vq->event) |
1004 | vq->vring.avail->flags = cpu_to_virtio16(vdev, vq->avail_flags_shadow); | |
f277ec42 | 1005 | } |
0a8a69dd RR |
1006 | |
1007 | /* Put everything in free lists. */ | |
0a8a69dd | 1008 | vq->free_head = 0; |
2a2d1382 | 1009 | for (i = 0; i < vring.num-1; i++) |
00e6f3d9 | 1010 | vq->vring.desc[i].next = cpu_to_virtio16(vdev, i + 1); |
2a2d1382 | 1011 | memset(vq->desc_state, 0, vring.num * sizeof(struct vring_desc_state)); |
0a8a69dd RR |
1012 | |
1013 | return &vq->vq; | |
1014 | } | |
2a2d1382 AL |
1015 | EXPORT_SYMBOL_GPL(__vring_new_virtqueue); |
1016 | ||
1017 | static void *vring_alloc_queue(struct virtio_device *vdev, size_t size, | |
1018 | dma_addr_t *dma_handle, gfp_t flag) | |
1019 | { | |
1020 | if (vring_use_dma_api(vdev)) { | |
1021 | return dma_alloc_coherent(vdev->dev.parent, size, | |
1022 | dma_handle, flag); | |
1023 | } else { | |
1024 | void *queue = alloc_pages_exact(PAGE_ALIGN(size), flag); | |
1025 | if (queue) { | |
1026 | phys_addr_t phys_addr = virt_to_phys(queue); | |
1027 | *dma_handle = (dma_addr_t)phys_addr; | |
1028 | ||
1029 | /* | |
1030 | * Sanity check: make sure we dind't truncate | |
1031 | * the address. The only arches I can find that | |
1032 | * have 64-bit phys_addr_t but 32-bit dma_addr_t | |
1033 | * are certain non-highmem MIPS and x86 | |
1034 | * configurations, but these configurations | |
1035 | * should never allocate physical pages above 32 | |
1036 | * bits, so this is fine. Just in case, throw a | |
1037 | * warning and abort if we end up with an | |
1038 | * unrepresentable address. | |
1039 | */ | |
1040 | if (WARN_ON_ONCE(*dma_handle != phys_addr)) { | |
1041 | free_pages_exact(queue, PAGE_ALIGN(size)); | |
1042 | return NULL; | |
1043 | } | |
1044 | } | |
1045 | return queue; | |
1046 | } | |
1047 | } | |
1048 | ||
1049 | static void vring_free_queue(struct virtio_device *vdev, size_t size, | |
1050 | void *queue, dma_addr_t dma_handle) | |
1051 | { | |
1052 | if (vring_use_dma_api(vdev)) { | |
1053 | dma_free_coherent(vdev->dev.parent, size, queue, dma_handle); | |
1054 | } else { | |
1055 | free_pages_exact(queue, PAGE_ALIGN(size)); | |
1056 | } | |
1057 | } | |
1058 | ||
1059 | struct virtqueue *vring_create_virtqueue( | |
1060 | unsigned int index, | |
1061 | unsigned int num, | |
1062 | unsigned int vring_align, | |
1063 | struct virtio_device *vdev, | |
1064 | bool weak_barriers, | |
1065 | bool may_reduce_num, | |
f94682dd | 1066 | bool context, |
2a2d1382 AL |
1067 | bool (*notify)(struct virtqueue *), |
1068 | void (*callback)(struct virtqueue *), | |
1069 | const char *name) | |
1070 | { | |
1071 | struct virtqueue *vq; | |
e00f7bd2 | 1072 | void *queue = NULL; |
2a2d1382 AL |
1073 | dma_addr_t dma_addr; |
1074 | size_t queue_size_in_bytes; | |
1075 | struct vring vring; | |
1076 | ||
1077 | /* We assume num is a power of 2. */ | |
1078 | if (num & (num - 1)) { | |
1079 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
1080 | return NULL; | |
1081 | } | |
1082 | ||
1083 | /* TODO: allocate each queue chunk individually */ | |
1084 | for (; num && vring_size(num, vring_align) > PAGE_SIZE; num /= 2) { | |
1085 | queue = vring_alloc_queue(vdev, vring_size(num, vring_align), | |
1086 | &dma_addr, | |
1087 | GFP_KERNEL|__GFP_NOWARN|__GFP_ZERO); | |
1088 | if (queue) | |
1089 | break; | |
1090 | } | |
1091 | ||
1092 | if (!num) | |
1093 | return NULL; | |
1094 | ||
1095 | if (!queue) { | |
1096 | /* Try to get a single page. You are my only hope! */ | |
1097 | queue = vring_alloc_queue(vdev, vring_size(num, vring_align), | |
1098 | &dma_addr, GFP_KERNEL|__GFP_ZERO); | |
1099 | } | |
1100 | if (!queue) | |
1101 | return NULL; | |
1102 | ||
1103 | queue_size_in_bytes = vring_size(num, vring_align); | |
1104 | vring_init(&vring, num, queue, vring_align); | |
1105 | ||
f94682dd | 1106 | vq = __vring_new_virtqueue(index, vring, vdev, weak_barriers, context, |
2a2d1382 AL |
1107 | notify, callback, name); |
1108 | if (!vq) { | |
1109 | vring_free_queue(vdev, queue_size_in_bytes, queue, | |
1110 | dma_addr); | |
1111 | return NULL; | |
1112 | } | |
1113 | ||
1114 | to_vvq(vq)->queue_dma_addr = dma_addr; | |
1115 | to_vvq(vq)->queue_size_in_bytes = queue_size_in_bytes; | |
1116 | to_vvq(vq)->we_own_ring = true; | |
1117 | ||
1118 | return vq; | |
1119 | } | |
1120 | EXPORT_SYMBOL_GPL(vring_create_virtqueue); | |
1121 | ||
1122 | struct virtqueue *vring_new_virtqueue(unsigned int index, | |
1123 | unsigned int num, | |
1124 | unsigned int vring_align, | |
1125 | struct virtio_device *vdev, | |
1126 | bool weak_barriers, | |
f94682dd | 1127 | bool context, |
2a2d1382 AL |
1128 | void *pages, |
1129 | bool (*notify)(struct virtqueue *vq), | |
1130 | void (*callback)(struct virtqueue *vq), | |
1131 | const char *name) | |
1132 | { | |
1133 | struct vring vring; | |
1134 | vring_init(&vring, num, pages, vring_align); | |
f94682dd | 1135 | return __vring_new_virtqueue(index, vring, vdev, weak_barriers, context, |
2a2d1382 AL |
1136 | notify, callback, name); |
1137 | } | |
c6fd4701 | 1138 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd | 1139 | |
2a2d1382 | 1140 | void vring_del_virtqueue(struct virtqueue *_vq) |
0a8a69dd | 1141 | { |
2a2d1382 AL |
1142 | struct vring_virtqueue *vq = to_vvq(_vq); |
1143 | ||
1144 | if (vq->we_own_ring) { | |
1145 | vring_free_queue(vq->vq.vdev, vq->queue_size_in_bytes, | |
1146 | vq->vring.desc, vq->queue_dma_addr); | |
1147 | } | |
1148 | list_del(&_vq->list); | |
1149 | kfree(vq); | |
0a8a69dd | 1150 | } |
c6fd4701 | 1151 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 1152 | |
e34f8725 RR |
1153 | /* Manipulates transport-specific feature bits. */ |
1154 | void vring_transport_features(struct virtio_device *vdev) | |
1155 | { | |
1156 | unsigned int i; | |
1157 | ||
1158 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
1159 | switch (i) { | |
9fa29b9d MM |
1160 | case VIRTIO_RING_F_INDIRECT_DESC: |
1161 | break; | |
a5c262c5 MT |
1162 | case VIRTIO_RING_F_EVENT_IDX: |
1163 | break; | |
747ae34a MT |
1164 | case VIRTIO_F_VERSION_1: |
1165 | break; | |
1a937693 MT |
1166 | case VIRTIO_F_IOMMU_PLATFORM: |
1167 | break; | |
e34f8725 RR |
1168 | default: |
1169 | /* We don't understand this bit. */ | |
e16e12be | 1170 | __virtio_clear_bit(vdev, i); |
e34f8725 RR |
1171 | } |
1172 | } | |
1173 | } | |
1174 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
1175 | ||
5dfc1762 RR |
1176 | /** |
1177 | * virtqueue_get_vring_size - return the size of the virtqueue's vring | |
1178 | * @vq: the struct virtqueue containing the vring of interest. | |
1179 | * | |
1180 | * Returns the size of the vring. This is mainly used for boasting to | |
1181 | * userspace. Unlike other operations, this need not be serialized. | |
1182 | */ | |
8f9f4668 RJ |
1183 | unsigned int virtqueue_get_vring_size(struct virtqueue *_vq) |
1184 | { | |
1185 | ||
1186 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1187 | ||
1188 | return vq->vring.num; | |
1189 | } | |
1190 | EXPORT_SYMBOL_GPL(virtqueue_get_vring_size); | |
1191 | ||
b3b32c94 HG |
1192 | bool virtqueue_is_broken(struct virtqueue *_vq) |
1193 | { | |
1194 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1195 | ||
1196 | return vq->broken; | |
1197 | } | |
1198 | EXPORT_SYMBOL_GPL(virtqueue_is_broken); | |
1199 | ||
e2dcdfe9 RR |
1200 | /* |
1201 | * This should prevent the device from being used, allowing drivers to | |
1202 | * recover. You may need to grab appropriate locks to flush. | |
1203 | */ | |
1204 | void virtio_break_device(struct virtio_device *dev) | |
1205 | { | |
1206 | struct virtqueue *_vq; | |
1207 | ||
1208 | list_for_each_entry(_vq, &dev->vqs, list) { | |
1209 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1210 | vq->broken = true; | |
1211 | } | |
1212 | } | |
1213 | EXPORT_SYMBOL_GPL(virtio_break_device); | |
1214 | ||
2a2d1382 | 1215 | dma_addr_t virtqueue_get_desc_addr(struct virtqueue *_vq) |
89062652 CH |
1216 | { |
1217 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1218 | ||
2a2d1382 AL |
1219 | BUG_ON(!vq->we_own_ring); |
1220 | ||
1221 | return vq->queue_dma_addr; | |
89062652 | 1222 | } |
2a2d1382 | 1223 | EXPORT_SYMBOL_GPL(virtqueue_get_desc_addr); |
89062652 | 1224 | |
2a2d1382 | 1225 | dma_addr_t virtqueue_get_avail_addr(struct virtqueue *_vq) |
89062652 CH |
1226 | { |
1227 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1228 | ||
2a2d1382 AL |
1229 | BUG_ON(!vq->we_own_ring); |
1230 | ||
1231 | return vq->queue_dma_addr + | |
1232 | ((char *)vq->vring.avail - (char *)vq->vring.desc); | |
1233 | } | |
1234 | EXPORT_SYMBOL_GPL(virtqueue_get_avail_addr); | |
1235 | ||
1236 | dma_addr_t virtqueue_get_used_addr(struct virtqueue *_vq) | |
1237 | { | |
1238 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1239 | ||
1240 | BUG_ON(!vq->we_own_ring); | |
1241 | ||
1242 | return vq->queue_dma_addr + | |
1243 | ((char *)vq->vring.used - (char *)vq->vring.desc); | |
1244 | } | |
1245 | EXPORT_SYMBOL_GPL(virtqueue_get_used_addr); | |
1246 | ||
1247 | const struct vring *virtqueue_get_vring(struct virtqueue *vq) | |
1248 | { | |
1249 | return &to_vvq(vq)->vring; | |
89062652 | 1250 | } |
2a2d1382 | 1251 | EXPORT_SYMBOL_GPL(virtqueue_get_vring); |
89062652 | 1252 | |
c6fd4701 | 1253 | MODULE_LICENSE("GPL"); |