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