]>
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> |
780bc790 | 26 | #include <linux/dma-mapping.h> |
78fe3987 | 27 | #include <xen/xen.h> |
0a8a69dd RR |
28 | |
29 | #ifdef DEBUG | |
30 | /* For development, we want to crash whenever the ring is screwed. */ | |
9499f5e7 RR |
31 | #define BAD_RING(_vq, fmt, args...) \ |
32 | do { \ | |
33 | dev_err(&(_vq)->vq.vdev->dev, \ | |
34 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
35 | BUG(); \ | |
36 | } while (0) | |
c5f841f1 RR |
37 | /* Caller is supposed to guarantee no reentry. */ |
38 | #define START_USE(_vq) \ | |
39 | do { \ | |
40 | if ((_vq)->in_use) \ | |
9499f5e7 RR |
41 | panic("%s:in_use = %i\n", \ |
42 | (_vq)->vq.name, (_vq)->in_use); \ | |
c5f841f1 | 43 | (_vq)->in_use = __LINE__; \ |
9499f5e7 | 44 | } while (0) |
3a35ce7d | 45 | #define END_USE(_vq) \ |
97a545ab | 46 | do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0) |
0a8a69dd | 47 | #else |
9499f5e7 RR |
48 | #define BAD_RING(_vq, fmt, args...) \ |
49 | do { \ | |
50 | dev_err(&_vq->vq.vdev->dev, \ | |
51 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
52 | (_vq)->broken = true; \ | |
53 | } while (0) | |
0a8a69dd RR |
54 | #define START_USE(vq) |
55 | #define END_USE(vq) | |
56 | #endif | |
57 | ||
780bc790 AL |
58 | struct vring_desc_state { |
59 | void *data; /* Data for callback. */ | |
60 | struct vring_desc *indir_desc; /* Indirect descriptor, if any. */ | |
61 | }; | |
62 | ||
43b4f721 | 63 | struct vring_virtqueue { |
0a8a69dd RR |
64 | struct virtqueue vq; |
65 | ||
66 | /* Actual memory layout for this queue */ | |
67 | struct vring vring; | |
68 | ||
7b21e34f RR |
69 | /* Can we use weak barriers? */ |
70 | bool weak_barriers; | |
71 | ||
0a8a69dd RR |
72 | /* Other side has made a mess, don't try any more. */ |
73 | bool broken; | |
74 | ||
9fa29b9d MM |
75 | /* Host supports indirect buffers */ |
76 | bool indirect; | |
77 | ||
a5c262c5 MT |
78 | /* Host publishes avail event idx */ |
79 | bool event; | |
80 | ||
0a8a69dd RR |
81 | /* Head of free buffer list. */ |
82 | unsigned int free_head; | |
83 | /* Number we've added since last sync. */ | |
84 | unsigned int num_added; | |
85 | ||
86 | /* Last used index we've seen. */ | |
1bc4953e | 87 | u16 last_used_idx; |
0a8a69dd | 88 | |
f277ec42 VS |
89 | /* Last written value to avail->flags */ |
90 | u16 avail_flags_shadow; | |
91 | ||
92 | /* Last written value to avail->idx in guest byte order */ | |
93 | u16 avail_idx_shadow; | |
94 | ||
0a8a69dd | 95 | /* How to notify other side. FIXME: commonalize hcalls! */ |
46f9c2b9 | 96 | bool (*notify)(struct virtqueue *vq); |
0a8a69dd | 97 | |
2a2d1382 AL |
98 | /* DMA, allocation, and size information */ |
99 | bool we_own_ring; | |
100 | size_t queue_size_in_bytes; | |
101 | dma_addr_t queue_dma_addr; | |
102 | ||
0a8a69dd RR |
103 | #ifdef DEBUG |
104 | /* They're supposed to lock for us. */ | |
105 | unsigned int in_use; | |
e93300b1 RR |
106 | |
107 | /* Figure out if their kicks are too delayed. */ | |
108 | bool last_add_time_valid; | |
109 | ktime_t last_add_time; | |
0a8a69dd RR |
110 | #endif |
111 | ||
780bc790 AL |
112 | /* Per-descriptor state. */ |
113 | struct vring_desc_state desc_state[]; | |
0a8a69dd RR |
114 | }; |
115 | ||
116 | #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) | |
117 | ||
d26c96c8 | 118 | /* |
1a937693 MT |
119 | * Modern virtio devices have feature bits to specify whether they need a |
120 | * quirk and bypass the IOMMU. If not there, just use the DMA API. | |
121 | * | |
122 | * If there, the interaction between virtio and DMA API is messy. | |
d26c96c8 AL |
123 | * |
124 | * On most systems with virtio, physical addresses match bus addresses, | |
125 | * and it doesn't particularly matter whether we use the DMA API. | |
126 | * | |
127 | * On some systems, including Xen and any system with a physical device | |
128 | * that speaks virtio behind a physical IOMMU, we must use the DMA API | |
129 | * for virtio DMA to work at all. | |
130 | * | |
131 | * On other systems, including SPARC and PPC64, virtio-pci devices are | |
132 | * enumerated as though they are behind an IOMMU, but the virtio host | |
133 | * ignores the IOMMU, so we must either pretend that the IOMMU isn't | |
134 | * there or somehow map everything as the identity. | |
135 | * | |
136 | * For the time being, we preserve historic behavior and bypass the DMA | |
137 | * API. | |
1a937693 MT |
138 | * |
139 | * TODO: install a per-device DMA ops structure that does the right thing | |
140 | * taking into account all the above quirks, and use the DMA API | |
141 | * unconditionally on data path. | |
d26c96c8 AL |
142 | */ |
143 | ||
144 | static bool vring_use_dma_api(struct virtio_device *vdev) | |
145 | { | |
1a937693 MT |
146 | if (!virtio_has_iommu_quirk(vdev)) |
147 | return true; | |
148 | ||
149 | /* Otherwise, we are left to guess. */ | |
78fe3987 AL |
150 | /* |
151 | * In theory, it's possible to have a buggy QEMU-supposed | |
152 | * emulated Q35 IOMMU and Xen enabled at the same time. On | |
153 | * such a configuration, virtio has never worked and will | |
154 | * not work without an even larger kludge. Instead, enable | |
155 | * the DMA API if we're a Xen guest, which at least allows | |
156 | * all of the sensible Xen configurations to work correctly. | |
157 | */ | |
158 | if (xen_domain()) | |
159 | return true; | |
160 | ||
d26c96c8 AL |
161 | return false; |
162 | } | |
163 | ||
780bc790 AL |
164 | /* |
165 | * The DMA ops on various arches are rather gnarly right now, and | |
166 | * making all of the arch DMA ops work on the vring device itself | |
167 | * is a mess. For now, we use the parent device for DMA ops. | |
168 | */ | |
75bfa81b | 169 | static inline struct device *vring_dma_dev(const struct vring_virtqueue *vq) |
780bc790 AL |
170 | { |
171 | return vq->vq.vdev->dev.parent; | |
172 | } | |
173 | ||
174 | /* Map one sg entry. */ | |
175 | static dma_addr_t vring_map_one_sg(const struct vring_virtqueue *vq, | |
176 | struct scatterlist *sg, | |
177 | enum dma_data_direction direction) | |
178 | { | |
179 | if (!vring_use_dma_api(vq->vq.vdev)) | |
180 | return (dma_addr_t)sg_phys(sg); | |
181 | ||
182 | /* | |
183 | * We can't use dma_map_sg, because we don't use scatterlists in | |
184 | * the way it expects (we don't guarantee that the scatterlist | |
185 | * will exist for the lifetime of the mapping). | |
186 | */ | |
187 | return dma_map_page(vring_dma_dev(vq), | |
188 | sg_page(sg), sg->offset, sg->length, | |
189 | direction); | |
190 | } | |
191 | ||
192 | static dma_addr_t vring_map_single(const struct vring_virtqueue *vq, | |
193 | void *cpu_addr, size_t size, | |
194 | enum dma_data_direction direction) | |
195 | { | |
196 | if (!vring_use_dma_api(vq->vq.vdev)) | |
197 | return (dma_addr_t)virt_to_phys(cpu_addr); | |
198 | ||
199 | return dma_map_single(vring_dma_dev(vq), | |
200 | cpu_addr, size, direction); | |
201 | } | |
202 | ||
203 | static void vring_unmap_one(const struct vring_virtqueue *vq, | |
204 | struct vring_desc *desc) | |
205 | { | |
206 | u16 flags; | |
207 | ||
208 | if (!vring_use_dma_api(vq->vq.vdev)) | |
209 | return; | |
210 | ||
211 | flags = virtio16_to_cpu(vq->vq.vdev, desc->flags); | |
212 | ||
213 | if (flags & VRING_DESC_F_INDIRECT) { | |
214 | dma_unmap_single(vring_dma_dev(vq), | |
215 | virtio64_to_cpu(vq->vq.vdev, desc->addr), | |
216 | virtio32_to_cpu(vq->vq.vdev, desc->len), | |
217 | (flags & VRING_DESC_F_WRITE) ? | |
218 | DMA_FROM_DEVICE : DMA_TO_DEVICE); | |
219 | } else { | |
220 | dma_unmap_page(vring_dma_dev(vq), | |
221 | virtio64_to_cpu(vq->vq.vdev, desc->addr), | |
222 | virtio32_to_cpu(vq->vq.vdev, desc->len), | |
223 | (flags & VRING_DESC_F_WRITE) ? | |
224 | DMA_FROM_DEVICE : DMA_TO_DEVICE); | |
225 | } | |
226 | } | |
227 | ||
228 | static int vring_mapping_error(const struct vring_virtqueue *vq, | |
229 | dma_addr_t addr) | |
230 | { | |
231 | if (!vring_use_dma_api(vq->vq.vdev)) | |
232 | return 0; | |
233 | ||
234 | return dma_mapping_error(vring_dma_dev(vq), addr); | |
235 | } | |
236 | ||
00e6f3d9 MT |
237 | static struct vring_desc *alloc_indirect(struct virtqueue *_vq, |
238 | unsigned int total_sg, gfp_t gfp) | |
9fa29b9d MM |
239 | { |
240 | struct vring_desc *desc; | |
b25bd251 | 241 | unsigned int i; |
9fa29b9d | 242 | |
b92b1b89 WD |
243 | /* |
244 | * We require lowmem mappings for the descriptors because | |
245 | * otherwise virt_to_phys will give us bogus addresses in the | |
246 | * virtqueue. | |
247 | */ | |
82107539 | 248 | gfp &= ~__GFP_HIGHMEM; |
b92b1b89 | 249 | |
6da2ec56 | 250 | desc = kmalloc_array(total_sg, sizeof(struct vring_desc), gfp); |
9fa29b9d | 251 | if (!desc) |
b25bd251 | 252 | return NULL; |
9fa29b9d | 253 | |
b25bd251 | 254 | for (i = 0; i < total_sg; i++) |
00e6f3d9 | 255 | desc[i].next = cpu_to_virtio16(_vq->vdev, i + 1); |
b25bd251 | 256 | return desc; |
9fa29b9d MM |
257 | } |
258 | ||
13816c76 RR |
259 | static inline int virtqueue_add(struct virtqueue *_vq, |
260 | struct scatterlist *sgs[], | |
eeebf9b1 | 261 | unsigned int total_sg, |
13816c76 RR |
262 | unsigned int out_sgs, |
263 | unsigned int in_sgs, | |
264 | void *data, | |
5a08b04f | 265 | void *ctx, |
13816c76 | 266 | gfp_t gfp) |
0a8a69dd RR |
267 | { |
268 | struct vring_virtqueue *vq = to_vvq(_vq); | |
13816c76 | 269 | struct scatterlist *sg; |
b25bd251 | 270 | struct vring_desc *desc; |
780bc790 | 271 | unsigned int i, n, avail, descs_used, uninitialized_var(prev), err_idx; |
1fe9b6fe | 272 | int head; |
b25bd251 | 273 | bool indirect; |
0a8a69dd | 274 | |
9fa29b9d MM |
275 | START_USE(vq); |
276 | ||
0a8a69dd | 277 | BUG_ON(data == NULL); |
5a08b04f | 278 | BUG_ON(ctx && vq->indirect); |
9fa29b9d | 279 | |
70670444 RR |
280 | if (unlikely(vq->broken)) { |
281 | END_USE(vq); | |
282 | return -EIO; | |
283 | } | |
284 | ||
e93300b1 RR |
285 | #ifdef DEBUG |
286 | { | |
287 | ktime_t now = ktime_get(); | |
288 | ||
289 | /* No kick or get, with .1 second between? Warn. */ | |
290 | if (vq->last_add_time_valid) | |
291 | WARN_ON(ktime_to_ms(ktime_sub(now, vq->last_add_time)) | |
292 | > 100); | |
293 | vq->last_add_time = now; | |
294 | vq->last_add_time_valid = true; | |
295 | } | |
296 | #endif | |
297 | ||
b25bd251 RR |
298 | BUG_ON(total_sg == 0); |
299 | ||
300 | head = vq->free_head; | |
301 | ||
9fa29b9d MM |
302 | /* If the host supports indirect descriptor tables, and we have multiple |
303 | * buffers, then go indirect. FIXME: tune this threshold */ | |
b25bd251 | 304 | if (vq->indirect && total_sg > 1 && vq->vq.num_free) |
00e6f3d9 | 305 | desc = alloc_indirect(_vq, total_sg, gfp); |
44ed8089 | 306 | else { |
b25bd251 | 307 | desc = NULL; |
44ed8089 RJ |
308 | WARN_ON_ONCE(total_sg > vq->vring.num && !vq->indirect); |
309 | } | |
b25bd251 RR |
310 | |
311 | if (desc) { | |
312 | /* Use a single buffer which doesn't continue */ | |
780bc790 | 313 | indirect = true; |
b25bd251 RR |
314 | /* Set up rest to use this indirect table. */ |
315 | i = 0; | |
316 | descs_used = 1; | |
b25bd251 | 317 | } else { |
780bc790 | 318 | indirect = false; |
b25bd251 RR |
319 | desc = vq->vring.desc; |
320 | i = head; | |
321 | descs_used = total_sg; | |
9fa29b9d MM |
322 | } |
323 | ||
b25bd251 | 324 | if (vq->vq.num_free < descs_used) { |
0a8a69dd | 325 | pr_debug("Can't add buf len %i - avail = %i\n", |
b25bd251 | 326 | descs_used, vq->vq.num_free); |
44653eae RR |
327 | /* FIXME: for historical reasons, we force a notify here if |
328 | * there are outgoing parts to the buffer. Presumably the | |
329 | * host should service the ring ASAP. */ | |
13816c76 | 330 | if (out_sgs) |
44653eae | 331 | vq->notify(&vq->vq); |
58625edf WY |
332 | if (indirect) |
333 | kfree(desc); | |
0a8a69dd RR |
334 | END_USE(vq); |
335 | return -ENOSPC; | |
336 | } | |
337 | ||
13816c76 | 338 | for (n = 0; n < out_sgs; n++) { |
eeebf9b1 | 339 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
780bc790 AL |
340 | dma_addr_t addr = vring_map_one_sg(vq, sg, DMA_TO_DEVICE); |
341 | if (vring_mapping_error(vq, addr)) | |
342 | goto unmap_release; | |
343 | ||
00e6f3d9 | 344 | desc[i].flags = cpu_to_virtio16(_vq->vdev, VRING_DESC_F_NEXT); |
780bc790 | 345 | desc[i].addr = cpu_to_virtio64(_vq->vdev, addr); |
00e6f3d9 | 346 | desc[i].len = cpu_to_virtio32(_vq->vdev, sg->length); |
13816c76 | 347 | prev = i; |
00e6f3d9 | 348 | i = virtio16_to_cpu(_vq->vdev, desc[i].next); |
13816c76 | 349 | } |
0a8a69dd | 350 | } |
13816c76 | 351 | for (; n < (out_sgs + in_sgs); n++) { |
eeebf9b1 | 352 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
780bc790 AL |
353 | dma_addr_t addr = vring_map_one_sg(vq, sg, DMA_FROM_DEVICE); |
354 | if (vring_mapping_error(vq, addr)) | |
355 | goto unmap_release; | |
356 | ||
00e6f3d9 | 357 | desc[i].flags = cpu_to_virtio16(_vq->vdev, VRING_DESC_F_NEXT | VRING_DESC_F_WRITE); |
780bc790 | 358 | desc[i].addr = cpu_to_virtio64(_vq->vdev, addr); |
00e6f3d9 | 359 | desc[i].len = cpu_to_virtio32(_vq->vdev, sg->length); |
13816c76 | 360 | prev = i; |
00e6f3d9 | 361 | i = virtio16_to_cpu(_vq->vdev, desc[i].next); |
13816c76 | 362 | } |
0a8a69dd RR |
363 | } |
364 | /* Last one doesn't continue. */ | |
00e6f3d9 | 365 | desc[prev].flags &= cpu_to_virtio16(_vq->vdev, ~VRING_DESC_F_NEXT); |
0a8a69dd | 366 | |
780bc790 AL |
367 | if (indirect) { |
368 | /* Now that the indirect table is filled in, map it. */ | |
369 | dma_addr_t addr = vring_map_single( | |
370 | vq, desc, total_sg * sizeof(struct vring_desc), | |
371 | DMA_TO_DEVICE); | |
372 | if (vring_mapping_error(vq, addr)) | |
373 | goto unmap_release; | |
374 | ||
375 | vq->vring.desc[head].flags = cpu_to_virtio16(_vq->vdev, VRING_DESC_F_INDIRECT); | |
376 | vq->vring.desc[head].addr = cpu_to_virtio64(_vq->vdev, addr); | |
377 | ||
378 | vq->vring.desc[head].len = cpu_to_virtio32(_vq->vdev, total_sg * sizeof(struct vring_desc)); | |
379 | } | |
380 | ||
381 | /* We're using some buffers from the free list. */ | |
382 | vq->vq.num_free -= descs_used; | |
383 | ||
0a8a69dd | 384 | /* Update free pointer */ |
b25bd251 | 385 | if (indirect) |
00e6f3d9 | 386 | vq->free_head = virtio16_to_cpu(_vq->vdev, vq->vring.desc[head].next); |
b25bd251 RR |
387 | else |
388 | vq->free_head = i; | |
0a8a69dd | 389 | |
780bc790 AL |
390 | /* Store token and indirect buffer state. */ |
391 | vq->desc_state[head].data = data; | |
392 | if (indirect) | |
393 | vq->desc_state[head].indir_desc = desc; | |
87646a34 | 394 | else |
5a08b04f | 395 | vq->desc_state[head].indir_desc = ctx; |
0a8a69dd RR |
396 | |
397 | /* Put entry in available array (but don't update avail->idx until they | |
3b720b8c | 398 | * do sync). */ |
f277ec42 | 399 | avail = vq->avail_idx_shadow & (vq->vring.num - 1); |
00e6f3d9 | 400 | vq->vring.avail->ring[avail] = cpu_to_virtio16(_vq->vdev, head); |
0a8a69dd | 401 | |
ee7cd898 RR |
402 | /* Descriptors and available array need to be set before we expose the |
403 | * new available array entries. */ | |
a9a0fef7 | 404 | virtio_wmb(vq->weak_barriers); |
f277ec42 VS |
405 | vq->avail_idx_shadow++; |
406 | vq->vring.avail->idx = cpu_to_virtio16(_vq->vdev, vq->avail_idx_shadow); | |
ee7cd898 RR |
407 | vq->num_added++; |
408 | ||
5e05bf58 TH |
409 | pr_debug("Added buffer head %i to %p\n", head, vq); |
410 | END_USE(vq); | |
411 | ||
ee7cd898 RR |
412 | /* This is very unlikely, but theoretically possible. Kick |
413 | * just in case. */ | |
414 | if (unlikely(vq->num_added == (1 << 16) - 1)) | |
415 | virtqueue_kick(_vq); | |
416 | ||
98e8c6bc | 417 | return 0; |
780bc790 AL |
418 | |
419 | unmap_release: | |
420 | err_idx = i; | |
421 | i = head; | |
422 | ||
423 | for (n = 0; n < total_sg; n++) { | |
424 | if (i == err_idx) | |
425 | break; | |
426 | vring_unmap_one(vq, &desc[i]); | |
c60923cb | 427 | i = virtio16_to_cpu(_vq->vdev, vq->vring.desc[i].next); |
780bc790 AL |
428 | } |
429 | ||
780bc790 AL |
430 | if (indirect) |
431 | kfree(desc); | |
432 | ||
3cc36f6e | 433 | END_USE(vq); |
780bc790 | 434 | return -EIO; |
0a8a69dd | 435 | } |
13816c76 | 436 | |
13816c76 RR |
437 | /** |
438 | * virtqueue_add_sgs - expose buffers to other end | |
439 | * @vq: the struct virtqueue we're talking about. | |
440 | * @sgs: array of terminated scatterlists. | |
441 | * @out_num: the number of scatterlists readable by other side | |
442 | * @in_num: the number of scatterlists which are writable (after readable ones) | |
443 | * @data: the token identifying the buffer. | |
444 | * @gfp: how to do memory allocations (if necessary). | |
445 | * | |
446 | * Caller must ensure we don't call this with other virtqueue operations | |
447 | * at the same time (except where noted). | |
448 | * | |
70670444 | 449 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
13816c76 RR |
450 | */ |
451 | int virtqueue_add_sgs(struct virtqueue *_vq, | |
452 | struct scatterlist *sgs[], | |
453 | unsigned int out_sgs, | |
454 | unsigned int in_sgs, | |
455 | void *data, | |
456 | gfp_t gfp) | |
457 | { | |
eeebf9b1 | 458 | unsigned int i, total_sg = 0; |
13816c76 RR |
459 | |
460 | /* Count them first. */ | |
eeebf9b1 | 461 | for (i = 0; i < out_sgs + in_sgs; i++) { |
13816c76 RR |
462 | struct scatterlist *sg; |
463 | for (sg = sgs[i]; sg; sg = sg_next(sg)) | |
eeebf9b1 | 464 | total_sg++; |
13816c76 | 465 | } |
5a08b04f MT |
466 | return virtqueue_add(_vq, sgs, total_sg, out_sgs, in_sgs, |
467 | data, NULL, gfp); | |
13816c76 RR |
468 | } |
469 | EXPORT_SYMBOL_GPL(virtqueue_add_sgs); | |
470 | ||
282edb36 RR |
471 | /** |
472 | * virtqueue_add_outbuf - expose output buffers to other end | |
473 | * @vq: the struct virtqueue we're talking about. | |
eeebf9b1 RR |
474 | * @sg: scatterlist (must be well-formed and terminated!) |
475 | * @num: the number of entries in @sg readable by other side | |
282edb36 RR |
476 | * @data: the token identifying the buffer. |
477 | * @gfp: how to do memory allocations (if necessary). | |
478 | * | |
479 | * Caller must ensure we don't call this with other virtqueue operations | |
480 | * at the same time (except where noted). | |
481 | * | |
70670444 | 482 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
282edb36 RR |
483 | */ |
484 | int virtqueue_add_outbuf(struct virtqueue *vq, | |
eeebf9b1 | 485 | struct scatterlist *sg, unsigned int num, |
282edb36 RR |
486 | void *data, |
487 | gfp_t gfp) | |
488 | { | |
5a08b04f | 489 | return virtqueue_add(vq, &sg, num, 1, 0, data, NULL, gfp); |
282edb36 RR |
490 | } |
491 | EXPORT_SYMBOL_GPL(virtqueue_add_outbuf); | |
492 | ||
493 | /** | |
494 | * virtqueue_add_inbuf - expose input buffers to other end | |
495 | * @vq: the struct virtqueue we're talking about. | |
eeebf9b1 RR |
496 | * @sg: scatterlist (must be well-formed and terminated!) |
497 | * @num: the number of entries in @sg writable by other side | |
282edb36 RR |
498 | * @data: the token identifying the buffer. |
499 | * @gfp: how to do memory allocations (if necessary). | |
500 | * | |
501 | * Caller must ensure we don't call this with other virtqueue operations | |
502 | * at the same time (except where noted). | |
503 | * | |
70670444 | 504 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
282edb36 RR |
505 | */ |
506 | int virtqueue_add_inbuf(struct virtqueue *vq, | |
eeebf9b1 | 507 | struct scatterlist *sg, unsigned int num, |
282edb36 RR |
508 | void *data, |
509 | gfp_t gfp) | |
510 | { | |
5a08b04f | 511 | return virtqueue_add(vq, &sg, num, 0, 1, data, NULL, gfp); |
282edb36 RR |
512 | } |
513 | EXPORT_SYMBOL_GPL(virtqueue_add_inbuf); | |
514 | ||
5a08b04f MT |
515 | /** |
516 | * virtqueue_add_inbuf_ctx - expose input buffers to other end | |
517 | * @vq: the struct virtqueue we're talking about. | |
518 | * @sg: scatterlist (must be well-formed and terminated!) | |
519 | * @num: the number of entries in @sg writable by other side | |
520 | * @data: the token identifying the buffer. | |
521 | * @ctx: extra context for the token | |
522 | * @gfp: how to do memory allocations (if necessary). | |
523 | * | |
524 | * Caller must ensure we don't call this with other virtqueue operations | |
525 | * at the same time (except where noted). | |
526 | * | |
527 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). | |
528 | */ | |
529 | int virtqueue_add_inbuf_ctx(struct virtqueue *vq, | |
530 | struct scatterlist *sg, unsigned int num, | |
531 | void *data, | |
532 | void *ctx, | |
533 | gfp_t gfp) | |
534 | { | |
535 | return virtqueue_add(vq, &sg, num, 0, 1, data, ctx, gfp); | |
536 | } | |
537 | EXPORT_SYMBOL_GPL(virtqueue_add_inbuf_ctx); | |
538 | ||
5dfc1762 | 539 | /** |
41f0377f | 540 | * virtqueue_kick_prepare - first half of split virtqueue_kick call. |
5dfc1762 RR |
541 | * @vq: the struct virtqueue |
542 | * | |
41f0377f RR |
543 | * Instead of virtqueue_kick(), you can do: |
544 | * if (virtqueue_kick_prepare(vq)) | |
545 | * virtqueue_notify(vq); | |
5dfc1762 | 546 | * |
41f0377f RR |
547 | * This is sometimes useful because the virtqueue_kick_prepare() needs |
548 | * to be serialized, but the actual virtqueue_notify() call does not. | |
5dfc1762 | 549 | */ |
41f0377f | 550 | bool virtqueue_kick_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
551 | { |
552 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a5c262c5 | 553 | u16 new, old; |
41f0377f RR |
554 | bool needs_kick; |
555 | ||
0a8a69dd | 556 | START_USE(vq); |
a72caae2 JW |
557 | /* We need to expose available array entries before checking avail |
558 | * event. */ | |
a9a0fef7 | 559 | virtio_mb(vq->weak_barriers); |
0a8a69dd | 560 | |
f277ec42 VS |
561 | old = vq->avail_idx_shadow - vq->num_added; |
562 | new = vq->avail_idx_shadow; | |
0a8a69dd RR |
563 | vq->num_added = 0; |
564 | ||
e93300b1 RR |
565 | #ifdef DEBUG |
566 | if (vq->last_add_time_valid) { | |
567 | WARN_ON(ktime_to_ms(ktime_sub(ktime_get(), | |
568 | vq->last_add_time)) > 100); | |
569 | } | |
570 | vq->last_add_time_valid = false; | |
571 | #endif | |
572 | ||
41f0377f | 573 | if (vq->event) { |
00e6f3d9 | 574 | needs_kick = vring_need_event(virtio16_to_cpu(_vq->vdev, vring_avail_event(&vq->vring)), |
41f0377f RR |
575 | new, old); |
576 | } else { | |
00e6f3d9 | 577 | needs_kick = !(vq->vring.used->flags & cpu_to_virtio16(_vq->vdev, VRING_USED_F_NO_NOTIFY)); |
41f0377f | 578 | } |
0a8a69dd | 579 | END_USE(vq); |
41f0377f RR |
580 | return needs_kick; |
581 | } | |
582 | EXPORT_SYMBOL_GPL(virtqueue_kick_prepare); | |
583 | ||
584 | /** | |
585 | * virtqueue_notify - second half of split virtqueue_kick call. | |
586 | * @vq: the struct virtqueue | |
587 | * | |
588 | * This does not need to be serialized. | |
5b1bf7cb HG |
589 | * |
590 | * Returns false if host notify failed or queue is broken, otherwise true. | |
41f0377f | 591 | */ |
5b1bf7cb | 592 | bool virtqueue_notify(struct virtqueue *_vq) |
41f0377f RR |
593 | { |
594 | struct vring_virtqueue *vq = to_vvq(_vq); | |
595 | ||
5b1bf7cb HG |
596 | if (unlikely(vq->broken)) |
597 | return false; | |
598 | ||
41f0377f | 599 | /* Prod other side to tell it about changes. */ |
2342d6a6 | 600 | if (!vq->notify(_vq)) { |
5b1bf7cb HG |
601 | vq->broken = true; |
602 | return false; | |
603 | } | |
604 | return true; | |
41f0377f RR |
605 | } |
606 | EXPORT_SYMBOL_GPL(virtqueue_notify); | |
607 | ||
608 | /** | |
609 | * virtqueue_kick - update after add_buf | |
610 | * @vq: the struct virtqueue | |
611 | * | |
b3087e48 | 612 | * After one or more virtqueue_add_* calls, invoke this to kick |
41f0377f RR |
613 | * the other side. |
614 | * | |
615 | * Caller must ensure we don't call this with other virtqueue | |
616 | * operations at the same time (except where noted). | |
5b1bf7cb HG |
617 | * |
618 | * Returns false if kick failed, otherwise true. | |
41f0377f | 619 | */ |
5b1bf7cb | 620 | bool virtqueue_kick(struct virtqueue *vq) |
41f0377f RR |
621 | { |
622 | if (virtqueue_kick_prepare(vq)) | |
5b1bf7cb HG |
623 | return virtqueue_notify(vq); |
624 | return true; | |
0a8a69dd | 625 | } |
7c5e9ed0 | 626 | EXPORT_SYMBOL_GPL(virtqueue_kick); |
0a8a69dd | 627 | |
5a08b04f MT |
628 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head, |
629 | void **ctx) | |
0a8a69dd | 630 | { |
780bc790 | 631 | unsigned int i, j; |
c60923cb | 632 | __virtio16 nextflag = cpu_to_virtio16(vq->vq.vdev, VRING_DESC_F_NEXT); |
0a8a69dd RR |
633 | |
634 | /* Clear data ptr. */ | |
780bc790 | 635 | vq->desc_state[head].data = NULL; |
0a8a69dd | 636 | |
780bc790 | 637 | /* Put back on free list: unmap first-level descriptors and find end */ |
0a8a69dd | 638 | i = head; |
9fa29b9d | 639 | |
780bc790 AL |
640 | while (vq->vring.desc[i].flags & nextflag) { |
641 | vring_unmap_one(vq, &vq->vring.desc[i]); | |
00e6f3d9 | 642 | i = virtio16_to_cpu(vq->vq.vdev, vq->vring.desc[i].next); |
06ca287d | 643 | vq->vq.num_free++; |
0a8a69dd RR |
644 | } |
645 | ||
780bc790 | 646 | vring_unmap_one(vq, &vq->vring.desc[i]); |
00e6f3d9 | 647 | vq->vring.desc[i].next = cpu_to_virtio16(vq->vq.vdev, vq->free_head); |
0a8a69dd | 648 | vq->free_head = head; |
780bc790 | 649 | |
0a8a69dd | 650 | /* Plus final descriptor */ |
06ca287d | 651 | vq->vq.num_free++; |
780bc790 | 652 | |
5a08b04f | 653 | if (vq->indirect) { |
780bc790 | 654 | struct vring_desc *indir_desc = vq->desc_state[head].indir_desc; |
5a08b04f MT |
655 | u32 len; |
656 | ||
657 | /* Free the indirect table, if any, now that it's unmapped. */ | |
658 | if (!indir_desc) | |
659 | return; | |
660 | ||
661 | len = virtio32_to_cpu(vq->vq.vdev, vq->vring.desc[head].len); | |
780bc790 AL |
662 | |
663 | BUG_ON(!(vq->vring.desc[head].flags & | |
664 | cpu_to_virtio16(vq->vq.vdev, VRING_DESC_F_INDIRECT))); | |
665 | BUG_ON(len == 0 || len % sizeof(struct vring_desc)); | |
666 | ||
667 | for (j = 0; j < len / sizeof(struct vring_desc); j++) | |
668 | vring_unmap_one(vq, &indir_desc[j]); | |
669 | ||
5a08b04f | 670 | kfree(indir_desc); |
780bc790 | 671 | vq->desc_state[head].indir_desc = NULL; |
5a08b04f MT |
672 | } else if (ctx) { |
673 | *ctx = vq->desc_state[head].indir_desc; | |
780bc790 | 674 | } |
0a8a69dd RR |
675 | } |
676 | ||
0a8a69dd RR |
677 | static inline bool more_used(const struct vring_virtqueue *vq) |
678 | { | |
00e6f3d9 | 679 | return vq->last_used_idx != virtio16_to_cpu(vq->vq.vdev, vq->vring.used->idx); |
0a8a69dd RR |
680 | } |
681 | ||
5dfc1762 RR |
682 | /** |
683 | * virtqueue_get_buf - get the next used buffer | |
684 | * @vq: the struct virtqueue we're talking about. | |
685 | * @len: the length written into the buffer | |
686 | * | |
0c7eaf59 | 687 | * If the device wrote data into the buffer, @len will be set to the |
5dfc1762 RR |
688 | * amount written. This means you don't need to clear the buffer |
689 | * beforehand to ensure there's no data leakage in the case of short | |
690 | * writes. | |
691 | * | |
692 | * Caller must ensure we don't call this with other virtqueue | |
693 | * operations at the same time (except where noted). | |
694 | * | |
695 | * Returns NULL if there are no used buffers, or the "data" token | |
b3087e48 | 696 | * handed to virtqueue_add_*(). |
5dfc1762 | 697 | */ |
5a08b04f MT |
698 | void *virtqueue_get_buf_ctx(struct virtqueue *_vq, unsigned int *len, |
699 | void **ctx) | |
0a8a69dd RR |
700 | { |
701 | struct vring_virtqueue *vq = to_vvq(_vq); | |
702 | void *ret; | |
703 | unsigned int i; | |
3b720b8c | 704 | u16 last_used; |
0a8a69dd RR |
705 | |
706 | START_USE(vq); | |
707 | ||
5ef82752 RR |
708 | if (unlikely(vq->broken)) { |
709 | END_USE(vq); | |
710 | return NULL; | |
711 | } | |
712 | ||
0a8a69dd RR |
713 | if (!more_used(vq)) { |
714 | pr_debug("No more buffers in queue\n"); | |
715 | END_USE(vq); | |
716 | return NULL; | |
717 | } | |
718 | ||
2d61ba95 | 719 | /* Only get used array entries after they have been exposed by host. */ |
a9a0fef7 | 720 | virtio_rmb(vq->weak_barriers); |
2d61ba95 | 721 | |
3b720b8c | 722 | last_used = (vq->last_used_idx & (vq->vring.num - 1)); |
00e6f3d9 MT |
723 | i = virtio32_to_cpu(_vq->vdev, vq->vring.used->ring[last_used].id); |
724 | *len = virtio32_to_cpu(_vq->vdev, vq->vring.used->ring[last_used].len); | |
0a8a69dd RR |
725 | |
726 | if (unlikely(i >= vq->vring.num)) { | |
727 | BAD_RING(vq, "id %u out of range\n", i); | |
728 | return NULL; | |
729 | } | |
780bc790 | 730 | if (unlikely(!vq->desc_state[i].data)) { |
0a8a69dd RR |
731 | BAD_RING(vq, "id %u is not a head!\n", i); |
732 | return NULL; | |
733 | } | |
734 | ||
735 | /* detach_buf clears data, so grab it now. */ | |
780bc790 | 736 | ret = vq->desc_state[i].data; |
5a08b04f | 737 | detach_buf(vq, i, ctx); |
0a8a69dd | 738 | vq->last_used_idx++; |
a5c262c5 MT |
739 | /* If we expect an interrupt for the next entry, tell host |
740 | * by writing event index and flush out the write before | |
741 | * the read in the next get_buf call. */ | |
788e5b3a MT |
742 | if (!(vq->avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT)) |
743 | virtio_store_mb(vq->weak_barriers, | |
744 | &vring_used_event(&vq->vring), | |
745 | cpu_to_virtio16(_vq->vdev, vq->last_used_idx)); | |
a5c262c5 | 746 | |
e93300b1 RR |
747 | #ifdef DEBUG |
748 | vq->last_add_time_valid = false; | |
749 | #endif | |
750 | ||
0a8a69dd RR |
751 | END_USE(vq); |
752 | return ret; | |
753 | } | |
5a08b04f | 754 | EXPORT_SYMBOL_GPL(virtqueue_get_buf_ctx); |
0a8a69dd | 755 | |
5a08b04f MT |
756 | void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len) |
757 | { | |
758 | return virtqueue_get_buf_ctx(_vq, len, NULL); | |
759 | } | |
760 | EXPORT_SYMBOL_GPL(virtqueue_get_buf); | |
5dfc1762 RR |
761 | /** |
762 | * virtqueue_disable_cb - disable callbacks | |
763 | * @vq: the struct virtqueue we're talking about. | |
764 | * | |
765 | * Note that this is not necessarily synchronous, hence unreliable and only | |
766 | * useful as an optimization. | |
767 | * | |
768 | * Unlike other operations, this need not be serialized. | |
769 | */ | |
7c5e9ed0 | 770 | void virtqueue_disable_cb(struct virtqueue *_vq) |
18445c4d RR |
771 | { |
772 | struct vring_virtqueue *vq = to_vvq(_vq); | |
773 | ||
f277ec42 VS |
774 | if (!(vq->avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT)) { |
775 | vq->avail_flags_shadow |= VRING_AVAIL_F_NO_INTERRUPT; | |
0ea1e4a6 LP |
776 | if (!vq->event) |
777 | vq->vring.avail->flags = cpu_to_virtio16(_vq->vdev, vq->avail_flags_shadow); | |
f277ec42 VS |
778 | } |
779 | ||
18445c4d | 780 | } |
7c5e9ed0 | 781 | EXPORT_SYMBOL_GPL(virtqueue_disable_cb); |
18445c4d | 782 | |
5dfc1762 | 783 | /** |
cc229884 | 784 | * virtqueue_enable_cb_prepare - restart callbacks after disable_cb |
5dfc1762 RR |
785 | * @vq: the struct virtqueue we're talking about. |
786 | * | |
cc229884 MT |
787 | * This re-enables callbacks; it returns current queue state |
788 | * in an opaque unsigned value. This value should be later tested by | |
789 | * virtqueue_poll, to detect a possible race between the driver checking for | |
790 | * more work, and enabling callbacks. | |
5dfc1762 RR |
791 | * |
792 | * Caller must ensure we don't call this with other virtqueue | |
793 | * operations at the same time (except where noted). | |
794 | */ | |
cc229884 | 795 | unsigned virtqueue_enable_cb_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
796 | { |
797 | struct vring_virtqueue *vq = to_vvq(_vq); | |
cc229884 | 798 | u16 last_used_idx; |
0a8a69dd RR |
799 | |
800 | START_USE(vq); | |
0a8a69dd RR |
801 | |
802 | /* We optimistically turn back on interrupts, then check if there was | |
803 | * more to do. */ | |
a5c262c5 MT |
804 | /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to |
805 | * either clear the flags bit or point the event index at the next | |
806 | * entry. Always do both to keep code simple. */ | |
f277ec42 VS |
807 | if (vq->avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT) { |
808 | vq->avail_flags_shadow &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
0ea1e4a6 LP |
809 | if (!vq->event) |
810 | vq->vring.avail->flags = cpu_to_virtio16(_vq->vdev, vq->avail_flags_shadow); | |
f277ec42 | 811 | } |
00e6f3d9 | 812 | vring_used_event(&vq->vring) = cpu_to_virtio16(_vq->vdev, last_used_idx = vq->last_used_idx); |
cc229884 MT |
813 | END_USE(vq); |
814 | return last_used_idx; | |
815 | } | |
816 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_prepare); | |
817 | ||
818 | /** | |
819 | * virtqueue_poll - query pending used buffers | |
820 | * @vq: the struct virtqueue we're talking about. | |
821 | * @last_used_idx: virtqueue state (from call to virtqueue_enable_cb_prepare). | |
822 | * | |
823 | * Returns "true" if there are pending used buffers in the queue. | |
824 | * | |
825 | * This does not need to be serialized. | |
826 | */ | |
827 | bool virtqueue_poll(struct virtqueue *_vq, unsigned last_used_idx) | |
828 | { | |
829 | struct vring_virtqueue *vq = to_vvq(_vq); | |
830 | ||
a9a0fef7 | 831 | virtio_mb(vq->weak_barriers); |
00e6f3d9 | 832 | return (u16)last_used_idx != virtio16_to_cpu(_vq->vdev, vq->vring.used->idx); |
cc229884 MT |
833 | } |
834 | EXPORT_SYMBOL_GPL(virtqueue_poll); | |
0a8a69dd | 835 | |
cc229884 MT |
836 | /** |
837 | * virtqueue_enable_cb - restart callbacks after disable_cb. | |
838 | * @vq: the struct virtqueue we're talking about. | |
839 | * | |
840 | * This re-enables callbacks; it returns "false" if there are pending | |
841 | * buffers in the queue, to detect a possible race between the driver | |
842 | * checking for more work, and enabling callbacks. | |
843 | * | |
844 | * Caller must ensure we don't call this with other virtqueue | |
845 | * operations at the same time (except where noted). | |
846 | */ | |
847 | bool virtqueue_enable_cb(struct virtqueue *_vq) | |
848 | { | |
849 | unsigned last_used_idx = virtqueue_enable_cb_prepare(_vq); | |
850 | return !virtqueue_poll(_vq, last_used_idx); | |
0a8a69dd | 851 | } |
7c5e9ed0 | 852 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb); |
0a8a69dd | 853 | |
5dfc1762 RR |
854 | /** |
855 | * virtqueue_enable_cb_delayed - restart callbacks after disable_cb. | |
856 | * @vq: the struct virtqueue we're talking about. | |
857 | * | |
858 | * This re-enables callbacks but hints to the other side to delay | |
859 | * interrupts until most of the available buffers have been processed; | |
860 | * it returns "false" if there are many pending buffers in the queue, | |
861 | * to detect a possible race between the driver checking for more work, | |
862 | * and enabling callbacks. | |
863 | * | |
864 | * Caller must ensure we don't call this with other virtqueue | |
865 | * operations at the same time (except where noted). | |
866 | */ | |
7ab358c2 MT |
867 | bool virtqueue_enable_cb_delayed(struct virtqueue *_vq) |
868 | { | |
869 | struct vring_virtqueue *vq = to_vvq(_vq); | |
870 | u16 bufs; | |
871 | ||
872 | START_USE(vq); | |
873 | ||
874 | /* We optimistically turn back on interrupts, then check if there was | |
875 | * more to do. */ | |
876 | /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to | |
877 | * either clear the flags bit or point the event index at the next | |
0ea1e4a6 | 878 | * entry. Always update the event index to keep code simple. */ |
f277ec42 VS |
879 | if (vq->avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT) { |
880 | vq->avail_flags_shadow &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
0ea1e4a6 LP |
881 | if (!vq->event) |
882 | vq->vring.avail->flags = cpu_to_virtio16(_vq->vdev, vq->avail_flags_shadow); | |
f277ec42 | 883 | } |
7ab358c2 | 884 | /* TODO: tune this threshold */ |
f277ec42 | 885 | bufs = (u16)(vq->avail_idx_shadow - vq->last_used_idx) * 3 / 4; |
788e5b3a MT |
886 | |
887 | virtio_store_mb(vq->weak_barriers, | |
888 | &vring_used_event(&vq->vring), | |
889 | cpu_to_virtio16(_vq->vdev, vq->last_used_idx + bufs)); | |
890 | ||
00e6f3d9 | 891 | if (unlikely((u16)(virtio16_to_cpu(_vq->vdev, vq->vring.used->idx) - vq->last_used_idx) > bufs)) { |
7ab358c2 MT |
892 | END_USE(vq); |
893 | return false; | |
894 | } | |
895 | ||
896 | END_USE(vq); | |
897 | return true; | |
898 | } | |
899 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed); | |
900 | ||
5dfc1762 RR |
901 | /** |
902 | * virtqueue_detach_unused_buf - detach first unused buffer | |
903 | * @vq: the struct virtqueue we're talking about. | |
904 | * | |
b3087e48 | 905 | * Returns NULL or the "data" token handed to virtqueue_add_*(). |
5dfc1762 RR |
906 | * This is not valid on an active queue; it is useful only for device |
907 | * shutdown. | |
908 | */ | |
7c5e9ed0 | 909 | void *virtqueue_detach_unused_buf(struct virtqueue *_vq) |
c021eac4 SM |
910 | { |
911 | struct vring_virtqueue *vq = to_vvq(_vq); | |
912 | unsigned int i; | |
913 | void *buf; | |
914 | ||
915 | START_USE(vq); | |
916 | ||
917 | for (i = 0; i < vq->vring.num; i++) { | |
780bc790 | 918 | if (!vq->desc_state[i].data) |
c021eac4 SM |
919 | continue; |
920 | /* detach_buf clears data, so grab it now. */ | |
780bc790 | 921 | buf = vq->desc_state[i].data; |
5a08b04f | 922 | detach_buf(vq, i, NULL); |
f277ec42 VS |
923 | vq->avail_idx_shadow--; |
924 | vq->vring.avail->idx = cpu_to_virtio16(_vq->vdev, vq->avail_idx_shadow); | |
c021eac4 SM |
925 | END_USE(vq); |
926 | return buf; | |
927 | } | |
928 | /* That should have freed everything. */ | |
06ca287d | 929 | BUG_ON(vq->vq.num_free != vq->vring.num); |
c021eac4 SM |
930 | |
931 | END_USE(vq); | |
932 | return NULL; | |
933 | } | |
7c5e9ed0 | 934 | EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf); |
c021eac4 | 935 | |
0a8a69dd RR |
936 | irqreturn_t vring_interrupt(int irq, void *_vq) |
937 | { | |
938 | struct vring_virtqueue *vq = to_vvq(_vq); | |
939 | ||
940 | if (!more_used(vq)) { | |
941 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
942 | return IRQ_NONE; | |
943 | } | |
944 | ||
945 | if (unlikely(vq->broken)) | |
946 | return IRQ_HANDLED; | |
947 | ||
948 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); | |
18445c4d RR |
949 | if (vq->vq.callback) |
950 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
951 | |
952 | return IRQ_HANDLED; | |
953 | } | |
c6fd4701 | 954 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd | 955 | |
2a2d1382 AL |
956 | struct virtqueue *__vring_new_virtqueue(unsigned int index, |
957 | struct vring vring, | |
958 | struct virtio_device *vdev, | |
959 | bool weak_barriers, | |
f94682dd | 960 | bool context, |
2a2d1382 AL |
961 | bool (*notify)(struct virtqueue *), |
962 | void (*callback)(struct virtqueue *), | |
963 | const char *name) | |
0a8a69dd | 964 | { |
0a8a69dd | 965 | unsigned int i; |
2a2d1382 | 966 | struct vring_virtqueue *vq; |
0a8a69dd | 967 | |
2a2d1382 | 968 | vq = kmalloc(sizeof(*vq) + vring.num * sizeof(struct vring_desc_state), |
780bc790 | 969 | GFP_KERNEL); |
0a8a69dd RR |
970 | if (!vq) |
971 | return NULL; | |
972 | ||
2a2d1382 | 973 | vq->vring = vring; |
0a8a69dd RR |
974 | vq->vq.callback = callback; |
975 | vq->vq.vdev = vdev; | |
9499f5e7 | 976 | vq->vq.name = name; |
2a2d1382 | 977 | vq->vq.num_free = vring.num; |
06ca287d | 978 | vq->vq.index = index; |
2a2d1382 AL |
979 | vq->we_own_ring = false; |
980 | vq->queue_dma_addr = 0; | |
981 | vq->queue_size_in_bytes = 0; | |
0a8a69dd | 982 | vq->notify = notify; |
7b21e34f | 983 | vq->weak_barriers = weak_barriers; |
0a8a69dd RR |
984 | vq->broken = false; |
985 | vq->last_used_idx = 0; | |
f277ec42 VS |
986 | vq->avail_flags_shadow = 0; |
987 | vq->avail_idx_shadow = 0; | |
0a8a69dd | 988 | vq->num_added = 0; |
9499f5e7 | 989 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0a8a69dd RR |
990 | #ifdef DEBUG |
991 | vq->in_use = false; | |
e93300b1 | 992 | vq->last_add_time_valid = false; |
0a8a69dd RR |
993 | #endif |
994 | ||
5a08b04f MT |
995 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC) && |
996 | !context; | |
a5c262c5 | 997 | vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); |
9fa29b9d | 998 | |
0a8a69dd | 999 | /* No callback? Tell other side not to bother us. */ |
f277ec42 VS |
1000 | if (!callback) { |
1001 | vq->avail_flags_shadow |= VRING_AVAIL_F_NO_INTERRUPT; | |
0ea1e4a6 LP |
1002 | if (!vq->event) |
1003 | vq->vring.avail->flags = cpu_to_virtio16(vdev, vq->avail_flags_shadow); | |
f277ec42 | 1004 | } |
0a8a69dd RR |
1005 | |
1006 | /* Put everything in free lists. */ | |
0a8a69dd | 1007 | vq->free_head = 0; |
2a2d1382 | 1008 | for (i = 0; i < vring.num-1; i++) |
00e6f3d9 | 1009 | vq->vring.desc[i].next = cpu_to_virtio16(vdev, i + 1); |
2a2d1382 | 1010 | memset(vq->desc_state, 0, vring.num * sizeof(struct vring_desc_state)); |
0a8a69dd RR |
1011 | |
1012 | return &vq->vq; | |
1013 | } | |
2a2d1382 AL |
1014 | EXPORT_SYMBOL_GPL(__vring_new_virtqueue); |
1015 | ||
1016 | static void *vring_alloc_queue(struct virtio_device *vdev, size_t size, | |
1017 | dma_addr_t *dma_handle, gfp_t flag) | |
1018 | { | |
1019 | if (vring_use_dma_api(vdev)) { | |
1020 | return dma_alloc_coherent(vdev->dev.parent, size, | |
1021 | dma_handle, flag); | |
1022 | } else { | |
1023 | void *queue = alloc_pages_exact(PAGE_ALIGN(size), flag); | |
1024 | if (queue) { | |
1025 | phys_addr_t phys_addr = virt_to_phys(queue); | |
1026 | *dma_handle = (dma_addr_t)phys_addr; | |
1027 | ||
1028 | /* | |
1029 | * Sanity check: make sure we dind't truncate | |
1030 | * the address. The only arches I can find that | |
1031 | * have 64-bit phys_addr_t but 32-bit dma_addr_t | |
1032 | * are certain non-highmem MIPS and x86 | |
1033 | * configurations, but these configurations | |
1034 | * should never allocate physical pages above 32 | |
1035 | * bits, so this is fine. Just in case, throw a | |
1036 | * warning and abort if we end up with an | |
1037 | * unrepresentable address. | |
1038 | */ | |
1039 | if (WARN_ON_ONCE(*dma_handle != phys_addr)) { | |
1040 | free_pages_exact(queue, PAGE_ALIGN(size)); | |
1041 | return NULL; | |
1042 | } | |
1043 | } | |
1044 | return queue; | |
1045 | } | |
1046 | } | |
1047 | ||
1048 | static void vring_free_queue(struct virtio_device *vdev, size_t size, | |
1049 | void *queue, dma_addr_t dma_handle) | |
1050 | { | |
1051 | if (vring_use_dma_api(vdev)) { | |
1052 | dma_free_coherent(vdev->dev.parent, size, queue, dma_handle); | |
1053 | } else { | |
1054 | free_pages_exact(queue, PAGE_ALIGN(size)); | |
1055 | } | |
1056 | } | |
1057 | ||
1058 | struct virtqueue *vring_create_virtqueue( | |
1059 | unsigned int index, | |
1060 | unsigned int num, | |
1061 | unsigned int vring_align, | |
1062 | struct virtio_device *vdev, | |
1063 | bool weak_barriers, | |
1064 | bool may_reduce_num, | |
f94682dd | 1065 | bool context, |
2a2d1382 AL |
1066 | bool (*notify)(struct virtqueue *), |
1067 | void (*callback)(struct virtqueue *), | |
1068 | const char *name) | |
1069 | { | |
1070 | struct virtqueue *vq; | |
e00f7bd2 | 1071 | void *queue = NULL; |
2a2d1382 AL |
1072 | dma_addr_t dma_addr; |
1073 | size_t queue_size_in_bytes; | |
1074 | struct vring vring; | |
1075 | ||
1076 | /* We assume num is a power of 2. */ | |
1077 | if (num & (num - 1)) { | |
1078 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
1079 | return NULL; | |
1080 | } | |
1081 | ||
1082 | /* TODO: allocate each queue chunk individually */ | |
1083 | for (; num && vring_size(num, vring_align) > PAGE_SIZE; num /= 2) { | |
1084 | queue = vring_alloc_queue(vdev, vring_size(num, vring_align), | |
1085 | &dma_addr, | |
1086 | GFP_KERNEL|__GFP_NOWARN|__GFP_ZERO); | |
1087 | if (queue) | |
1088 | break; | |
1089 | } | |
1090 | ||
1091 | if (!num) | |
1092 | return NULL; | |
1093 | ||
1094 | if (!queue) { | |
1095 | /* Try to get a single page. You are my only hope! */ | |
1096 | queue = vring_alloc_queue(vdev, vring_size(num, vring_align), | |
1097 | &dma_addr, GFP_KERNEL|__GFP_ZERO); | |
1098 | } | |
1099 | if (!queue) | |
1100 | return NULL; | |
1101 | ||
1102 | queue_size_in_bytes = vring_size(num, vring_align); | |
1103 | vring_init(&vring, num, queue, vring_align); | |
1104 | ||
f94682dd | 1105 | vq = __vring_new_virtqueue(index, vring, vdev, weak_barriers, context, |
2a2d1382 AL |
1106 | notify, callback, name); |
1107 | if (!vq) { | |
1108 | vring_free_queue(vdev, queue_size_in_bytes, queue, | |
1109 | dma_addr); | |
1110 | return NULL; | |
1111 | } | |
1112 | ||
1113 | to_vvq(vq)->queue_dma_addr = dma_addr; | |
1114 | to_vvq(vq)->queue_size_in_bytes = queue_size_in_bytes; | |
1115 | to_vvq(vq)->we_own_ring = true; | |
1116 | ||
1117 | return vq; | |
1118 | } | |
1119 | EXPORT_SYMBOL_GPL(vring_create_virtqueue); | |
1120 | ||
1121 | struct virtqueue *vring_new_virtqueue(unsigned int index, | |
1122 | unsigned int num, | |
1123 | unsigned int vring_align, | |
1124 | struct virtio_device *vdev, | |
1125 | bool weak_barriers, | |
f94682dd | 1126 | bool context, |
2a2d1382 AL |
1127 | void *pages, |
1128 | bool (*notify)(struct virtqueue *vq), | |
1129 | void (*callback)(struct virtqueue *vq), | |
1130 | const char *name) | |
1131 | { | |
1132 | struct vring vring; | |
1133 | vring_init(&vring, num, pages, vring_align); | |
f94682dd | 1134 | return __vring_new_virtqueue(index, vring, vdev, weak_barriers, context, |
2a2d1382 AL |
1135 | notify, callback, name); |
1136 | } | |
c6fd4701 | 1137 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd | 1138 | |
2a2d1382 | 1139 | void vring_del_virtqueue(struct virtqueue *_vq) |
0a8a69dd | 1140 | { |
2a2d1382 AL |
1141 | struct vring_virtqueue *vq = to_vvq(_vq); |
1142 | ||
1143 | if (vq->we_own_ring) { | |
1144 | vring_free_queue(vq->vq.vdev, vq->queue_size_in_bytes, | |
1145 | vq->vring.desc, vq->queue_dma_addr); | |
1146 | } | |
1147 | list_del(&_vq->list); | |
1148 | kfree(vq); | |
0a8a69dd | 1149 | } |
c6fd4701 | 1150 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 1151 | |
e34f8725 RR |
1152 | /* Manipulates transport-specific feature bits. */ |
1153 | void vring_transport_features(struct virtio_device *vdev) | |
1154 | { | |
1155 | unsigned int i; | |
1156 | ||
1157 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
1158 | switch (i) { | |
9fa29b9d MM |
1159 | case VIRTIO_RING_F_INDIRECT_DESC: |
1160 | break; | |
a5c262c5 MT |
1161 | case VIRTIO_RING_F_EVENT_IDX: |
1162 | break; | |
747ae34a MT |
1163 | case VIRTIO_F_VERSION_1: |
1164 | break; | |
1a937693 MT |
1165 | case VIRTIO_F_IOMMU_PLATFORM: |
1166 | break; | |
e34f8725 RR |
1167 | default: |
1168 | /* We don't understand this bit. */ | |
e16e12be | 1169 | __virtio_clear_bit(vdev, i); |
e34f8725 RR |
1170 | } |
1171 | } | |
1172 | } | |
1173 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
1174 | ||
5dfc1762 RR |
1175 | /** |
1176 | * virtqueue_get_vring_size - return the size of the virtqueue's vring | |
1177 | * @vq: the struct virtqueue containing the vring of interest. | |
1178 | * | |
1179 | * Returns the size of the vring. This is mainly used for boasting to | |
1180 | * userspace. Unlike other operations, this need not be serialized. | |
1181 | */ | |
8f9f4668 RJ |
1182 | unsigned int virtqueue_get_vring_size(struct virtqueue *_vq) |
1183 | { | |
1184 | ||
1185 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1186 | ||
1187 | return vq->vring.num; | |
1188 | } | |
1189 | EXPORT_SYMBOL_GPL(virtqueue_get_vring_size); | |
1190 | ||
b3b32c94 HG |
1191 | bool virtqueue_is_broken(struct virtqueue *_vq) |
1192 | { | |
1193 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1194 | ||
1195 | return vq->broken; | |
1196 | } | |
1197 | EXPORT_SYMBOL_GPL(virtqueue_is_broken); | |
1198 | ||
e2dcdfe9 RR |
1199 | /* |
1200 | * This should prevent the device from being used, allowing drivers to | |
1201 | * recover. You may need to grab appropriate locks to flush. | |
1202 | */ | |
1203 | void virtio_break_device(struct virtio_device *dev) | |
1204 | { | |
1205 | struct virtqueue *_vq; | |
1206 | ||
1207 | list_for_each_entry(_vq, &dev->vqs, list) { | |
1208 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1209 | vq->broken = true; | |
1210 | } | |
1211 | } | |
1212 | EXPORT_SYMBOL_GPL(virtio_break_device); | |
1213 | ||
2a2d1382 | 1214 | dma_addr_t virtqueue_get_desc_addr(struct virtqueue *_vq) |
89062652 CH |
1215 | { |
1216 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1217 | ||
2a2d1382 AL |
1218 | BUG_ON(!vq->we_own_ring); |
1219 | ||
1220 | return vq->queue_dma_addr; | |
89062652 | 1221 | } |
2a2d1382 | 1222 | EXPORT_SYMBOL_GPL(virtqueue_get_desc_addr); |
89062652 | 1223 | |
2a2d1382 | 1224 | dma_addr_t virtqueue_get_avail_addr(struct virtqueue *_vq) |
89062652 CH |
1225 | { |
1226 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1227 | ||
2a2d1382 AL |
1228 | BUG_ON(!vq->we_own_ring); |
1229 | ||
1230 | return vq->queue_dma_addr + | |
1231 | ((char *)vq->vring.avail - (char *)vq->vring.desc); | |
1232 | } | |
1233 | EXPORT_SYMBOL_GPL(virtqueue_get_avail_addr); | |
1234 | ||
1235 | dma_addr_t virtqueue_get_used_addr(struct virtqueue *_vq) | |
1236 | { | |
1237 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1238 | ||
1239 | BUG_ON(!vq->we_own_ring); | |
1240 | ||
1241 | return vq->queue_dma_addr + | |
1242 | ((char *)vq->vring.used - (char *)vq->vring.desc); | |
1243 | } | |
1244 | EXPORT_SYMBOL_GPL(virtqueue_get_used_addr); | |
1245 | ||
1246 | const struct vring *virtqueue_get_vring(struct virtqueue *vq) | |
1247 | { | |
1248 | return &to_vvq(vq)->vring; | |
89062652 | 1249 | } |
2a2d1382 | 1250 | EXPORT_SYMBOL_GPL(virtqueue_get_vring); |
89062652 | 1251 | |
c6fd4701 | 1252 | MODULE_LICENSE("GPL"); |