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> |
0a8a69dd RR |
27 | |
28 | #ifdef DEBUG | |
29 | /* For development, we want to crash whenever the ring is screwed. */ | |
9499f5e7 RR |
30 | #define BAD_RING(_vq, fmt, args...) \ |
31 | do { \ | |
32 | dev_err(&(_vq)->vq.vdev->dev, \ | |
33 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
34 | BUG(); \ | |
35 | } while (0) | |
c5f841f1 RR |
36 | /* Caller is supposed to guarantee no reentry. */ |
37 | #define START_USE(_vq) \ | |
38 | do { \ | |
39 | if ((_vq)->in_use) \ | |
9499f5e7 RR |
40 | panic("%s:in_use = %i\n", \ |
41 | (_vq)->vq.name, (_vq)->in_use); \ | |
c5f841f1 | 42 | (_vq)->in_use = __LINE__; \ |
9499f5e7 | 43 | } while (0) |
3a35ce7d | 44 | #define END_USE(_vq) \ |
97a545ab | 45 | do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0) |
0a8a69dd | 46 | #else |
9499f5e7 RR |
47 | #define BAD_RING(_vq, fmt, args...) \ |
48 | do { \ | |
49 | dev_err(&_vq->vq.vdev->dev, \ | |
50 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
51 | (_vq)->broken = true; \ | |
52 | } while (0) | |
0a8a69dd RR |
53 | #define START_USE(vq) |
54 | #define END_USE(vq) | |
55 | #endif | |
56 | ||
57 | struct vring_virtqueue | |
58 | { | |
59 | struct virtqueue vq; | |
60 | ||
61 | /* Actual memory layout for this queue */ | |
62 | struct vring vring; | |
63 | ||
7b21e34f RR |
64 | /* Can we use weak barriers? */ |
65 | bool weak_barriers; | |
66 | ||
0a8a69dd RR |
67 | /* Other side has made a mess, don't try any more. */ |
68 | bool broken; | |
69 | ||
9fa29b9d MM |
70 | /* Host supports indirect buffers */ |
71 | bool indirect; | |
72 | ||
a5c262c5 MT |
73 | /* Host publishes avail event idx */ |
74 | bool event; | |
75 | ||
0a8a69dd RR |
76 | /* Head of free buffer list. */ |
77 | unsigned int free_head; | |
78 | /* Number we've added since last sync. */ | |
79 | unsigned int num_added; | |
80 | ||
81 | /* Last used index we've seen. */ | |
1bc4953e | 82 | u16 last_used_idx; |
0a8a69dd RR |
83 | |
84 | /* How to notify other side. FIXME: commonalize hcalls! */ | |
46f9c2b9 | 85 | bool (*notify)(struct virtqueue *vq); |
0a8a69dd RR |
86 | |
87 | #ifdef DEBUG | |
88 | /* They're supposed to lock for us. */ | |
89 | unsigned int in_use; | |
e93300b1 RR |
90 | |
91 | /* Figure out if their kicks are too delayed. */ | |
92 | bool last_add_time_valid; | |
93 | ktime_t last_add_time; | |
0a8a69dd RR |
94 | #endif |
95 | ||
96 | /* Tokens for callbacks. */ | |
97 | void *data[]; | |
98 | }; | |
99 | ||
100 | #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) | |
101 | ||
9fa29b9d | 102 | /* Set up an indirect table of descriptors and add it to the queue. */ |
13816c76 RR |
103 | static inline int vring_add_indirect(struct vring_virtqueue *vq, |
104 | struct scatterlist *sgs[], | |
13816c76 | 105 | unsigned int total_sg, |
13816c76 RR |
106 | unsigned int out_sgs, |
107 | unsigned int in_sgs, | |
108 | gfp_t gfp) | |
9fa29b9d MM |
109 | { |
110 | struct vring_desc *desc; | |
111 | unsigned head; | |
13816c76 RR |
112 | struct scatterlist *sg; |
113 | int i, n; | |
9fa29b9d | 114 | |
b92b1b89 WD |
115 | /* |
116 | * We require lowmem mappings for the descriptors because | |
117 | * otherwise virt_to_phys will give us bogus addresses in the | |
118 | * virtqueue. | |
119 | */ | |
120 | gfp &= ~(__GFP_HIGHMEM | __GFP_HIGH); | |
121 | ||
13816c76 | 122 | desc = kmalloc(total_sg * sizeof(struct vring_desc), gfp); |
9fa29b9d | 123 | if (!desc) |
686d3637 | 124 | return -ENOMEM; |
9fa29b9d | 125 | |
13816c76 RR |
126 | /* Transfer entries from the sg lists into the indirect page */ |
127 | i = 0; | |
128 | for (n = 0; n < out_sgs; n++) { | |
eeebf9b1 | 129 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
13816c76 RR |
130 | desc[i].flags = VRING_DESC_F_NEXT; |
131 | desc[i].addr = sg_phys(sg); | |
132 | desc[i].len = sg->length; | |
133 | desc[i].next = i+1; | |
134 | i++; | |
135 | } | |
9fa29b9d | 136 | } |
13816c76 | 137 | for (; n < (out_sgs + in_sgs); n++) { |
eeebf9b1 | 138 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
13816c76 RR |
139 | desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; |
140 | desc[i].addr = sg_phys(sg); | |
141 | desc[i].len = sg->length; | |
142 | desc[i].next = i+1; | |
143 | i++; | |
144 | } | |
9fa29b9d | 145 | } |
13816c76 | 146 | BUG_ON(i != total_sg); |
9fa29b9d MM |
147 | |
148 | /* Last one doesn't continue. */ | |
149 | desc[i-1].flags &= ~VRING_DESC_F_NEXT; | |
150 | desc[i-1].next = 0; | |
151 | ||
152 | /* We're about to use a buffer */ | |
06ca287d | 153 | vq->vq.num_free--; |
9fa29b9d MM |
154 | |
155 | /* Use a single buffer which doesn't continue */ | |
156 | head = vq->free_head; | |
157 | vq->vring.desc[head].flags = VRING_DESC_F_INDIRECT; | |
158 | vq->vring.desc[head].addr = virt_to_phys(desc); | |
bb478d8b RR |
159 | /* kmemleak gives a false positive, as it's hidden by virt_to_phys */ |
160 | kmemleak_ignore(desc); | |
9fa29b9d MM |
161 | vq->vring.desc[head].len = i * sizeof(struct vring_desc); |
162 | ||
163 | /* Update free pointer */ | |
164 | vq->free_head = vq->vring.desc[head].next; | |
165 | ||
166 | return head; | |
167 | } | |
168 | ||
13816c76 RR |
169 | static inline int virtqueue_add(struct virtqueue *_vq, |
170 | struct scatterlist *sgs[], | |
eeebf9b1 | 171 | unsigned int total_sg, |
13816c76 RR |
172 | unsigned int out_sgs, |
173 | unsigned int in_sgs, | |
174 | void *data, | |
175 | gfp_t gfp) | |
0a8a69dd RR |
176 | { |
177 | struct vring_virtqueue *vq = to_vvq(_vq); | |
13816c76 | 178 | struct scatterlist *sg; |
eeebf9b1 | 179 | unsigned int i, n, avail, uninitialized_var(prev); |
1fe9b6fe | 180 | int head; |
0a8a69dd | 181 | |
9fa29b9d MM |
182 | START_USE(vq); |
183 | ||
0a8a69dd | 184 | BUG_ON(data == NULL); |
9fa29b9d | 185 | |
70670444 RR |
186 | if (unlikely(vq->broken)) { |
187 | END_USE(vq); | |
188 | return -EIO; | |
189 | } | |
190 | ||
e93300b1 RR |
191 | #ifdef DEBUG |
192 | { | |
193 | ktime_t now = ktime_get(); | |
194 | ||
195 | /* No kick or get, with .1 second between? Warn. */ | |
196 | if (vq->last_add_time_valid) | |
197 | WARN_ON(ktime_to_ms(ktime_sub(now, vq->last_add_time)) | |
198 | > 100); | |
199 | vq->last_add_time = now; | |
200 | vq->last_add_time_valid = true; | |
201 | } | |
202 | #endif | |
203 | ||
9fa29b9d MM |
204 | /* If the host supports indirect descriptor tables, and we have multiple |
205 | * buffers, then go indirect. FIXME: tune this threshold */ | |
13816c76 | 206 | if (vq->indirect && total_sg > 1 && vq->vq.num_free) { |
eeebf9b1 | 207 | head = vring_add_indirect(vq, sgs, total_sg, |
13816c76 | 208 | out_sgs, in_sgs, gfp); |
1fe9b6fe | 209 | if (likely(head >= 0)) |
9fa29b9d MM |
210 | goto add_head; |
211 | } | |
212 | ||
13816c76 RR |
213 | BUG_ON(total_sg > vq->vring.num); |
214 | BUG_ON(total_sg == 0); | |
0a8a69dd | 215 | |
13816c76 | 216 | if (vq->vq.num_free < total_sg) { |
0a8a69dd | 217 | pr_debug("Can't add buf len %i - avail = %i\n", |
13816c76 | 218 | total_sg, vq->vq.num_free); |
44653eae RR |
219 | /* FIXME: for historical reasons, we force a notify here if |
220 | * there are outgoing parts to the buffer. Presumably the | |
221 | * host should service the ring ASAP. */ | |
13816c76 | 222 | if (out_sgs) |
44653eae | 223 | vq->notify(&vq->vq); |
0a8a69dd RR |
224 | END_USE(vq); |
225 | return -ENOSPC; | |
226 | } | |
227 | ||
228 | /* We're about to use some buffers from the free list. */ | |
13816c76 RR |
229 | vq->vq.num_free -= total_sg; |
230 | ||
231 | head = i = vq->free_head; | |
232 | for (n = 0; n < out_sgs; n++) { | |
eeebf9b1 | 233 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
13816c76 RR |
234 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT; |
235 | vq->vring.desc[i].addr = sg_phys(sg); | |
236 | vq->vring.desc[i].len = sg->length; | |
237 | prev = i; | |
238 | i = vq->vring.desc[i].next; | |
239 | } | |
0a8a69dd | 240 | } |
13816c76 | 241 | for (; n < (out_sgs + in_sgs); n++) { |
eeebf9b1 | 242 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
13816c76 RR |
243 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; |
244 | vq->vring.desc[i].addr = sg_phys(sg); | |
245 | vq->vring.desc[i].len = sg->length; | |
246 | prev = i; | |
247 | i = vq->vring.desc[i].next; | |
248 | } | |
0a8a69dd RR |
249 | } |
250 | /* Last one doesn't continue. */ | |
251 | vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT; | |
252 | ||
253 | /* Update free pointer */ | |
254 | vq->free_head = i; | |
255 | ||
9fa29b9d | 256 | add_head: |
0a8a69dd RR |
257 | /* Set token. */ |
258 | vq->data[head] = data; | |
259 | ||
260 | /* Put entry in available array (but don't update avail->idx until they | |
3b720b8c | 261 | * do sync). */ |
ee7cd898 | 262 | avail = (vq->vring.avail->idx & (vq->vring.num-1)); |
0a8a69dd RR |
263 | vq->vring.avail->ring[avail] = head; |
264 | ||
ee7cd898 RR |
265 | /* Descriptors and available array need to be set before we expose the |
266 | * new available array entries. */ | |
a9a0fef7 | 267 | virtio_wmb(vq->weak_barriers); |
ee7cd898 RR |
268 | vq->vring.avail->idx++; |
269 | vq->num_added++; | |
270 | ||
271 | /* This is very unlikely, but theoretically possible. Kick | |
272 | * just in case. */ | |
273 | if (unlikely(vq->num_added == (1 << 16) - 1)) | |
274 | virtqueue_kick(_vq); | |
275 | ||
0a8a69dd RR |
276 | pr_debug("Added buffer head %i to %p\n", head, vq); |
277 | END_USE(vq); | |
3c1b27d5 | 278 | |
98e8c6bc | 279 | return 0; |
0a8a69dd | 280 | } |
13816c76 | 281 | |
13816c76 RR |
282 | /** |
283 | * virtqueue_add_sgs - expose buffers to other end | |
284 | * @vq: the struct virtqueue we're talking about. | |
285 | * @sgs: array of terminated scatterlists. | |
286 | * @out_num: the number of scatterlists readable by other side | |
287 | * @in_num: the number of scatterlists which are writable (after readable ones) | |
288 | * @data: the token identifying the buffer. | |
289 | * @gfp: how to do memory allocations (if necessary). | |
290 | * | |
291 | * Caller must ensure we don't call this with other virtqueue operations | |
292 | * at the same time (except where noted). | |
293 | * | |
70670444 | 294 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
13816c76 RR |
295 | */ |
296 | int virtqueue_add_sgs(struct virtqueue *_vq, | |
297 | struct scatterlist *sgs[], | |
298 | unsigned int out_sgs, | |
299 | unsigned int in_sgs, | |
300 | void *data, | |
301 | gfp_t gfp) | |
302 | { | |
eeebf9b1 | 303 | unsigned int i, total_sg = 0; |
13816c76 RR |
304 | |
305 | /* Count them first. */ | |
eeebf9b1 | 306 | for (i = 0; i < out_sgs + in_sgs; i++) { |
13816c76 RR |
307 | struct scatterlist *sg; |
308 | for (sg = sgs[i]; sg; sg = sg_next(sg)) | |
eeebf9b1 | 309 | total_sg++; |
13816c76 | 310 | } |
eeebf9b1 | 311 | return virtqueue_add(_vq, sgs, total_sg, out_sgs, in_sgs, data, gfp); |
13816c76 RR |
312 | } |
313 | EXPORT_SYMBOL_GPL(virtqueue_add_sgs); | |
314 | ||
282edb36 RR |
315 | /** |
316 | * virtqueue_add_outbuf - expose output buffers to other end | |
317 | * @vq: the struct virtqueue we're talking about. | |
eeebf9b1 RR |
318 | * @sg: scatterlist (must be well-formed and terminated!) |
319 | * @num: the number of entries in @sg readable by other side | |
282edb36 RR |
320 | * @data: the token identifying the buffer. |
321 | * @gfp: how to do memory allocations (if necessary). | |
322 | * | |
323 | * Caller must ensure we don't call this with other virtqueue operations | |
324 | * at the same time (except where noted). | |
325 | * | |
70670444 | 326 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
282edb36 RR |
327 | */ |
328 | int virtqueue_add_outbuf(struct virtqueue *vq, | |
eeebf9b1 | 329 | struct scatterlist *sg, unsigned int num, |
282edb36 RR |
330 | void *data, |
331 | gfp_t gfp) | |
332 | { | |
eeebf9b1 | 333 | return virtqueue_add(vq, &sg, num, 1, 0, data, gfp); |
282edb36 RR |
334 | } |
335 | EXPORT_SYMBOL_GPL(virtqueue_add_outbuf); | |
336 | ||
337 | /** | |
338 | * virtqueue_add_inbuf - expose input buffers to other end | |
339 | * @vq: the struct virtqueue we're talking about. | |
eeebf9b1 RR |
340 | * @sg: scatterlist (must be well-formed and terminated!) |
341 | * @num: the number of entries in @sg writable by other side | |
282edb36 RR |
342 | * @data: the token identifying the buffer. |
343 | * @gfp: how to do memory allocations (if necessary). | |
344 | * | |
345 | * Caller must ensure we don't call this with other virtqueue operations | |
346 | * at the same time (except where noted). | |
347 | * | |
70670444 | 348 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
282edb36 RR |
349 | */ |
350 | int virtqueue_add_inbuf(struct virtqueue *vq, | |
eeebf9b1 | 351 | struct scatterlist *sg, unsigned int num, |
282edb36 RR |
352 | void *data, |
353 | gfp_t gfp) | |
354 | { | |
eeebf9b1 | 355 | return virtqueue_add(vq, &sg, num, 0, 1, data, gfp); |
282edb36 RR |
356 | } |
357 | EXPORT_SYMBOL_GPL(virtqueue_add_inbuf); | |
358 | ||
5dfc1762 | 359 | /** |
41f0377f | 360 | * virtqueue_kick_prepare - first half of split virtqueue_kick call. |
5dfc1762 RR |
361 | * @vq: the struct virtqueue |
362 | * | |
41f0377f RR |
363 | * Instead of virtqueue_kick(), you can do: |
364 | * if (virtqueue_kick_prepare(vq)) | |
365 | * virtqueue_notify(vq); | |
5dfc1762 | 366 | * |
41f0377f RR |
367 | * This is sometimes useful because the virtqueue_kick_prepare() needs |
368 | * to be serialized, but the actual virtqueue_notify() call does not. | |
5dfc1762 | 369 | */ |
41f0377f | 370 | bool virtqueue_kick_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
371 | { |
372 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a5c262c5 | 373 | u16 new, old; |
41f0377f RR |
374 | bool needs_kick; |
375 | ||
0a8a69dd | 376 | START_USE(vq); |
a72caae2 JW |
377 | /* We need to expose available array entries before checking avail |
378 | * event. */ | |
a9a0fef7 | 379 | virtio_mb(vq->weak_barriers); |
0a8a69dd | 380 | |
ee7cd898 RR |
381 | old = vq->vring.avail->idx - vq->num_added; |
382 | new = vq->vring.avail->idx; | |
0a8a69dd RR |
383 | vq->num_added = 0; |
384 | ||
e93300b1 RR |
385 | #ifdef DEBUG |
386 | if (vq->last_add_time_valid) { | |
387 | WARN_ON(ktime_to_ms(ktime_sub(ktime_get(), | |
388 | vq->last_add_time)) > 100); | |
389 | } | |
390 | vq->last_add_time_valid = false; | |
391 | #endif | |
392 | ||
41f0377f RR |
393 | if (vq->event) { |
394 | needs_kick = vring_need_event(vring_avail_event(&vq->vring), | |
395 | new, old); | |
396 | } else { | |
397 | needs_kick = !(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY); | |
398 | } | |
0a8a69dd | 399 | END_USE(vq); |
41f0377f RR |
400 | return needs_kick; |
401 | } | |
402 | EXPORT_SYMBOL_GPL(virtqueue_kick_prepare); | |
403 | ||
404 | /** | |
405 | * virtqueue_notify - second half of split virtqueue_kick call. | |
406 | * @vq: the struct virtqueue | |
407 | * | |
408 | * This does not need to be serialized. | |
5b1bf7cb HG |
409 | * |
410 | * Returns false if host notify failed or queue is broken, otherwise true. | |
41f0377f | 411 | */ |
5b1bf7cb | 412 | bool virtqueue_notify(struct virtqueue *_vq) |
41f0377f RR |
413 | { |
414 | struct vring_virtqueue *vq = to_vvq(_vq); | |
415 | ||
5b1bf7cb HG |
416 | if (unlikely(vq->broken)) |
417 | return false; | |
418 | ||
41f0377f | 419 | /* Prod other side to tell it about changes. */ |
2342d6a6 | 420 | if (!vq->notify(_vq)) { |
5b1bf7cb HG |
421 | vq->broken = true; |
422 | return false; | |
423 | } | |
424 | return true; | |
41f0377f RR |
425 | } |
426 | EXPORT_SYMBOL_GPL(virtqueue_notify); | |
427 | ||
428 | /** | |
429 | * virtqueue_kick - update after add_buf | |
430 | * @vq: the struct virtqueue | |
431 | * | |
b3087e48 | 432 | * After one or more virtqueue_add_* calls, invoke this to kick |
41f0377f RR |
433 | * the other side. |
434 | * | |
435 | * Caller must ensure we don't call this with other virtqueue | |
436 | * operations at the same time (except where noted). | |
5b1bf7cb HG |
437 | * |
438 | * Returns false if kick failed, otherwise true. | |
41f0377f | 439 | */ |
5b1bf7cb | 440 | bool virtqueue_kick(struct virtqueue *vq) |
41f0377f RR |
441 | { |
442 | if (virtqueue_kick_prepare(vq)) | |
5b1bf7cb HG |
443 | return virtqueue_notify(vq); |
444 | return true; | |
0a8a69dd | 445 | } |
7c5e9ed0 | 446 | EXPORT_SYMBOL_GPL(virtqueue_kick); |
0a8a69dd RR |
447 | |
448 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head) | |
449 | { | |
450 | unsigned int i; | |
451 | ||
452 | /* Clear data ptr. */ | |
453 | vq->data[head] = NULL; | |
454 | ||
455 | /* Put back on free list: find end */ | |
456 | i = head; | |
9fa29b9d MM |
457 | |
458 | /* Free the indirect table */ | |
459 | if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT) | |
460 | kfree(phys_to_virt(vq->vring.desc[i].addr)); | |
461 | ||
0a8a69dd RR |
462 | while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) { |
463 | i = vq->vring.desc[i].next; | |
06ca287d | 464 | vq->vq.num_free++; |
0a8a69dd RR |
465 | } |
466 | ||
467 | vq->vring.desc[i].next = vq->free_head; | |
468 | vq->free_head = head; | |
469 | /* Plus final descriptor */ | |
06ca287d | 470 | vq->vq.num_free++; |
0a8a69dd RR |
471 | } |
472 | ||
0a8a69dd RR |
473 | static inline bool more_used(const struct vring_virtqueue *vq) |
474 | { | |
475 | return vq->last_used_idx != vq->vring.used->idx; | |
476 | } | |
477 | ||
5dfc1762 RR |
478 | /** |
479 | * virtqueue_get_buf - get the next used buffer | |
480 | * @vq: the struct virtqueue we're talking about. | |
481 | * @len: the length written into the buffer | |
482 | * | |
483 | * If the driver wrote data into the buffer, @len will be set to the | |
484 | * amount written. This means you don't need to clear the buffer | |
485 | * beforehand to ensure there's no data leakage in the case of short | |
486 | * writes. | |
487 | * | |
488 | * Caller must ensure we don't call this with other virtqueue | |
489 | * operations at the same time (except where noted). | |
490 | * | |
491 | * Returns NULL if there are no used buffers, or the "data" token | |
b3087e48 | 492 | * handed to virtqueue_add_*(). |
5dfc1762 | 493 | */ |
7c5e9ed0 | 494 | void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len) |
0a8a69dd RR |
495 | { |
496 | struct vring_virtqueue *vq = to_vvq(_vq); | |
497 | void *ret; | |
498 | unsigned int i; | |
3b720b8c | 499 | u16 last_used; |
0a8a69dd RR |
500 | |
501 | START_USE(vq); | |
502 | ||
5ef82752 RR |
503 | if (unlikely(vq->broken)) { |
504 | END_USE(vq); | |
505 | return NULL; | |
506 | } | |
507 | ||
0a8a69dd RR |
508 | if (!more_used(vq)) { |
509 | pr_debug("No more buffers in queue\n"); | |
510 | END_USE(vq); | |
511 | return NULL; | |
512 | } | |
513 | ||
2d61ba95 | 514 | /* Only get used array entries after they have been exposed by host. */ |
a9a0fef7 | 515 | virtio_rmb(vq->weak_barriers); |
2d61ba95 | 516 | |
3b720b8c RR |
517 | last_used = (vq->last_used_idx & (vq->vring.num - 1)); |
518 | i = vq->vring.used->ring[last_used].id; | |
519 | *len = vq->vring.used->ring[last_used].len; | |
0a8a69dd RR |
520 | |
521 | if (unlikely(i >= vq->vring.num)) { | |
522 | BAD_RING(vq, "id %u out of range\n", i); | |
523 | return NULL; | |
524 | } | |
525 | if (unlikely(!vq->data[i])) { | |
526 | BAD_RING(vq, "id %u is not a head!\n", i); | |
527 | return NULL; | |
528 | } | |
529 | ||
530 | /* detach_buf clears data, so grab it now. */ | |
531 | ret = vq->data[i]; | |
532 | detach_buf(vq, i); | |
533 | vq->last_used_idx++; | |
a5c262c5 MT |
534 | /* If we expect an interrupt for the next entry, tell host |
535 | * by writing event index and flush out the write before | |
536 | * the read in the next get_buf call. */ | |
537 | if (!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) { | |
538 | vring_used_event(&vq->vring) = vq->last_used_idx; | |
a9a0fef7 | 539 | virtio_mb(vq->weak_barriers); |
a5c262c5 MT |
540 | } |
541 | ||
e93300b1 RR |
542 | #ifdef DEBUG |
543 | vq->last_add_time_valid = false; | |
544 | #endif | |
545 | ||
0a8a69dd RR |
546 | END_USE(vq); |
547 | return ret; | |
548 | } | |
7c5e9ed0 | 549 | EXPORT_SYMBOL_GPL(virtqueue_get_buf); |
0a8a69dd | 550 | |
5dfc1762 RR |
551 | /** |
552 | * virtqueue_disable_cb - disable callbacks | |
553 | * @vq: the struct virtqueue we're talking about. | |
554 | * | |
555 | * Note that this is not necessarily synchronous, hence unreliable and only | |
556 | * useful as an optimization. | |
557 | * | |
558 | * Unlike other operations, this need not be serialized. | |
559 | */ | |
7c5e9ed0 | 560 | void virtqueue_disable_cb(struct virtqueue *_vq) |
18445c4d RR |
561 | { |
562 | struct vring_virtqueue *vq = to_vvq(_vq); | |
563 | ||
18445c4d | 564 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; |
18445c4d | 565 | } |
7c5e9ed0 | 566 | EXPORT_SYMBOL_GPL(virtqueue_disable_cb); |
18445c4d | 567 | |
5dfc1762 | 568 | /** |
cc229884 | 569 | * virtqueue_enable_cb_prepare - restart callbacks after disable_cb |
5dfc1762 RR |
570 | * @vq: the struct virtqueue we're talking about. |
571 | * | |
cc229884 MT |
572 | * This re-enables callbacks; it returns current queue state |
573 | * in an opaque unsigned value. This value should be later tested by | |
574 | * virtqueue_poll, to detect a possible race between the driver checking for | |
575 | * more work, and enabling callbacks. | |
5dfc1762 RR |
576 | * |
577 | * Caller must ensure we don't call this with other virtqueue | |
578 | * operations at the same time (except where noted). | |
579 | */ | |
cc229884 | 580 | unsigned virtqueue_enable_cb_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
581 | { |
582 | struct vring_virtqueue *vq = to_vvq(_vq); | |
cc229884 | 583 | u16 last_used_idx; |
0a8a69dd RR |
584 | |
585 | START_USE(vq); | |
0a8a69dd RR |
586 | |
587 | /* We optimistically turn back on interrupts, then check if there was | |
588 | * more to do. */ | |
a5c262c5 MT |
589 | /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to |
590 | * either clear the flags bit or point the event index at the next | |
591 | * entry. Always do both to keep code simple. */ | |
0a8a69dd | 592 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; |
cc229884 MT |
593 | vring_used_event(&vq->vring) = last_used_idx = vq->last_used_idx; |
594 | END_USE(vq); | |
595 | return last_used_idx; | |
596 | } | |
597 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_prepare); | |
598 | ||
599 | /** | |
600 | * virtqueue_poll - query pending used buffers | |
601 | * @vq: the struct virtqueue we're talking about. | |
602 | * @last_used_idx: virtqueue state (from call to virtqueue_enable_cb_prepare). | |
603 | * | |
604 | * Returns "true" if there are pending used buffers in the queue. | |
605 | * | |
606 | * This does not need to be serialized. | |
607 | */ | |
608 | bool virtqueue_poll(struct virtqueue *_vq, unsigned last_used_idx) | |
609 | { | |
610 | struct vring_virtqueue *vq = to_vvq(_vq); | |
611 | ||
a9a0fef7 | 612 | virtio_mb(vq->weak_barriers); |
cc229884 MT |
613 | return (u16)last_used_idx != vq->vring.used->idx; |
614 | } | |
615 | EXPORT_SYMBOL_GPL(virtqueue_poll); | |
0a8a69dd | 616 | |
cc229884 MT |
617 | /** |
618 | * virtqueue_enable_cb - restart callbacks after disable_cb. | |
619 | * @vq: the struct virtqueue we're talking about. | |
620 | * | |
621 | * This re-enables callbacks; it returns "false" if there are pending | |
622 | * buffers in the queue, to detect a possible race between the driver | |
623 | * checking for more work, and enabling callbacks. | |
624 | * | |
625 | * Caller must ensure we don't call this with other virtqueue | |
626 | * operations at the same time (except where noted). | |
627 | */ | |
628 | bool virtqueue_enable_cb(struct virtqueue *_vq) | |
629 | { | |
630 | unsigned last_used_idx = virtqueue_enable_cb_prepare(_vq); | |
631 | return !virtqueue_poll(_vq, last_used_idx); | |
0a8a69dd | 632 | } |
7c5e9ed0 | 633 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb); |
0a8a69dd | 634 | |
5dfc1762 RR |
635 | /** |
636 | * virtqueue_enable_cb_delayed - restart callbacks after disable_cb. | |
637 | * @vq: the struct virtqueue we're talking about. | |
638 | * | |
639 | * This re-enables callbacks but hints to the other side to delay | |
640 | * interrupts until most of the available buffers have been processed; | |
641 | * it returns "false" if there are many pending buffers in the queue, | |
642 | * to detect a possible race between the driver checking for more work, | |
643 | * and enabling callbacks. | |
644 | * | |
645 | * Caller must ensure we don't call this with other virtqueue | |
646 | * operations at the same time (except where noted). | |
647 | */ | |
7ab358c2 MT |
648 | bool virtqueue_enable_cb_delayed(struct virtqueue *_vq) |
649 | { | |
650 | struct vring_virtqueue *vq = to_vvq(_vq); | |
651 | u16 bufs; | |
652 | ||
653 | START_USE(vq); | |
654 | ||
655 | /* We optimistically turn back on interrupts, then check if there was | |
656 | * more to do. */ | |
657 | /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to | |
658 | * either clear the flags bit or point the event index at the next | |
659 | * entry. Always do both to keep code simple. */ | |
660 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
661 | /* TODO: tune this threshold */ | |
662 | bufs = (u16)(vq->vring.avail->idx - vq->last_used_idx) * 3 / 4; | |
663 | vring_used_event(&vq->vring) = vq->last_used_idx + bufs; | |
a9a0fef7 | 664 | virtio_mb(vq->weak_barriers); |
7ab358c2 MT |
665 | if (unlikely((u16)(vq->vring.used->idx - vq->last_used_idx) > bufs)) { |
666 | END_USE(vq); | |
667 | return false; | |
668 | } | |
669 | ||
670 | END_USE(vq); | |
671 | return true; | |
672 | } | |
673 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed); | |
674 | ||
5dfc1762 RR |
675 | /** |
676 | * virtqueue_detach_unused_buf - detach first unused buffer | |
677 | * @vq: the struct virtqueue we're talking about. | |
678 | * | |
b3087e48 | 679 | * Returns NULL or the "data" token handed to virtqueue_add_*(). |
5dfc1762 RR |
680 | * This is not valid on an active queue; it is useful only for device |
681 | * shutdown. | |
682 | */ | |
7c5e9ed0 | 683 | void *virtqueue_detach_unused_buf(struct virtqueue *_vq) |
c021eac4 SM |
684 | { |
685 | struct vring_virtqueue *vq = to_vvq(_vq); | |
686 | unsigned int i; | |
687 | void *buf; | |
688 | ||
689 | START_USE(vq); | |
690 | ||
691 | for (i = 0; i < vq->vring.num; i++) { | |
692 | if (!vq->data[i]) | |
693 | continue; | |
694 | /* detach_buf clears data, so grab it now. */ | |
695 | buf = vq->data[i]; | |
696 | detach_buf(vq, i); | |
b3258ff1 | 697 | vq->vring.avail->idx--; |
c021eac4 SM |
698 | END_USE(vq); |
699 | return buf; | |
700 | } | |
701 | /* That should have freed everything. */ | |
06ca287d | 702 | BUG_ON(vq->vq.num_free != vq->vring.num); |
c021eac4 SM |
703 | |
704 | END_USE(vq); | |
705 | return NULL; | |
706 | } | |
7c5e9ed0 | 707 | EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf); |
c021eac4 | 708 | |
0a8a69dd RR |
709 | irqreturn_t vring_interrupt(int irq, void *_vq) |
710 | { | |
711 | struct vring_virtqueue *vq = to_vvq(_vq); | |
712 | ||
713 | if (!more_used(vq)) { | |
714 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
715 | return IRQ_NONE; | |
716 | } | |
717 | ||
718 | if (unlikely(vq->broken)) | |
719 | return IRQ_HANDLED; | |
720 | ||
721 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); | |
18445c4d RR |
722 | if (vq->vq.callback) |
723 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
724 | |
725 | return IRQ_HANDLED; | |
726 | } | |
c6fd4701 | 727 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd | 728 | |
17bb6d40 JW |
729 | struct virtqueue *vring_new_virtqueue(unsigned int index, |
730 | unsigned int num, | |
87c7d57c | 731 | unsigned int vring_align, |
0a8a69dd | 732 | struct virtio_device *vdev, |
7b21e34f | 733 | bool weak_barriers, |
0a8a69dd | 734 | void *pages, |
46f9c2b9 | 735 | bool (*notify)(struct virtqueue *), |
9499f5e7 RR |
736 | void (*callback)(struct virtqueue *), |
737 | const char *name) | |
0a8a69dd RR |
738 | { |
739 | struct vring_virtqueue *vq; | |
740 | unsigned int i; | |
741 | ||
42b36cc0 RR |
742 | /* We assume num is a power of 2. */ |
743 | if (num & (num - 1)) { | |
744 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
745 | return NULL; | |
746 | } | |
747 | ||
0a8a69dd RR |
748 | vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL); |
749 | if (!vq) | |
750 | return NULL; | |
751 | ||
87c7d57c | 752 | vring_init(&vq->vring, num, pages, vring_align); |
0a8a69dd RR |
753 | vq->vq.callback = callback; |
754 | vq->vq.vdev = vdev; | |
9499f5e7 | 755 | vq->vq.name = name; |
06ca287d RR |
756 | vq->vq.num_free = num; |
757 | vq->vq.index = index; | |
0a8a69dd | 758 | vq->notify = notify; |
7b21e34f | 759 | vq->weak_barriers = weak_barriers; |
0a8a69dd RR |
760 | vq->broken = false; |
761 | vq->last_used_idx = 0; | |
762 | vq->num_added = 0; | |
9499f5e7 | 763 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0a8a69dd RR |
764 | #ifdef DEBUG |
765 | vq->in_use = false; | |
e93300b1 | 766 | vq->last_add_time_valid = false; |
0a8a69dd RR |
767 | #endif |
768 | ||
9fa29b9d | 769 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC); |
a5c262c5 | 770 | vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); |
9fa29b9d | 771 | |
0a8a69dd RR |
772 | /* No callback? Tell other side not to bother us. */ |
773 | if (!callback) | |
774 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; | |
775 | ||
776 | /* Put everything in free lists. */ | |
0a8a69dd | 777 | vq->free_head = 0; |
3b870624 | 778 | for (i = 0; i < num-1; i++) { |
0a8a69dd | 779 | vq->vring.desc[i].next = i+1; |
3b870624 AS |
780 | vq->data[i] = NULL; |
781 | } | |
782 | vq->data[i] = NULL; | |
0a8a69dd RR |
783 | |
784 | return &vq->vq; | |
785 | } | |
c6fd4701 | 786 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd RR |
787 | |
788 | void vring_del_virtqueue(struct virtqueue *vq) | |
789 | { | |
9499f5e7 | 790 | list_del(&vq->list); |
0a8a69dd RR |
791 | kfree(to_vvq(vq)); |
792 | } | |
c6fd4701 | 793 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 794 | |
e34f8725 RR |
795 | /* Manipulates transport-specific feature bits. */ |
796 | void vring_transport_features(struct virtio_device *vdev) | |
797 | { | |
798 | unsigned int i; | |
799 | ||
800 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
801 | switch (i) { | |
9fa29b9d MM |
802 | case VIRTIO_RING_F_INDIRECT_DESC: |
803 | break; | |
a5c262c5 MT |
804 | case VIRTIO_RING_F_EVENT_IDX: |
805 | break; | |
e34f8725 RR |
806 | default: |
807 | /* We don't understand this bit. */ | |
808 | clear_bit(i, vdev->features); | |
809 | } | |
810 | } | |
811 | } | |
812 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
813 | ||
5dfc1762 RR |
814 | /** |
815 | * virtqueue_get_vring_size - return the size of the virtqueue's vring | |
816 | * @vq: the struct virtqueue containing the vring of interest. | |
817 | * | |
818 | * Returns the size of the vring. This is mainly used for boasting to | |
819 | * userspace. Unlike other operations, this need not be serialized. | |
820 | */ | |
8f9f4668 RJ |
821 | unsigned int virtqueue_get_vring_size(struct virtqueue *_vq) |
822 | { | |
823 | ||
824 | struct vring_virtqueue *vq = to_vvq(_vq); | |
825 | ||
826 | return vq->vring.num; | |
827 | } | |
828 | EXPORT_SYMBOL_GPL(virtqueue_get_vring_size); | |
829 | ||
b3b32c94 HG |
830 | bool virtqueue_is_broken(struct virtqueue *_vq) |
831 | { | |
832 | struct vring_virtqueue *vq = to_vvq(_vq); | |
833 | ||
834 | return vq->broken; | |
835 | } | |
836 | EXPORT_SYMBOL_GPL(virtqueue_is_broken); | |
837 | ||
e2dcdfe9 RR |
838 | /* |
839 | * This should prevent the device from being used, allowing drivers to | |
840 | * recover. You may need to grab appropriate locks to flush. | |
841 | */ | |
842 | void virtio_break_device(struct virtio_device *dev) | |
843 | { | |
844 | struct virtqueue *_vq; | |
845 | ||
846 | list_for_each_entry(_vq, &dev->vqs, list) { | |
847 | struct vring_virtqueue *vq = to_vvq(_vq); | |
848 | vq->broken = true; | |
849 | } | |
850 | } | |
851 | EXPORT_SYMBOL_GPL(virtio_break_device); | |
852 | ||
c6fd4701 | 853 | MODULE_LICENSE("GPL"); |