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> |
0a8a69dd RR |
26 | |
27 | #ifdef DEBUG | |
28 | /* For development, we want to crash whenever the ring is screwed. */ | |
9499f5e7 RR |
29 | #define BAD_RING(_vq, fmt, args...) \ |
30 | do { \ | |
31 | dev_err(&(_vq)->vq.vdev->dev, \ | |
32 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
33 | BUG(); \ | |
34 | } while (0) | |
c5f841f1 RR |
35 | /* Caller is supposed to guarantee no reentry. */ |
36 | #define START_USE(_vq) \ | |
37 | do { \ | |
38 | if ((_vq)->in_use) \ | |
9499f5e7 RR |
39 | panic("%s:in_use = %i\n", \ |
40 | (_vq)->vq.name, (_vq)->in_use); \ | |
c5f841f1 | 41 | (_vq)->in_use = __LINE__; \ |
9499f5e7 | 42 | } while (0) |
3a35ce7d | 43 | #define END_USE(_vq) \ |
97a545ab | 44 | do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0) |
0a8a69dd | 45 | #else |
9499f5e7 RR |
46 | #define BAD_RING(_vq, fmt, args...) \ |
47 | do { \ | |
48 | dev_err(&_vq->vq.vdev->dev, \ | |
49 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
50 | (_vq)->broken = true; \ | |
51 | } while (0) | |
0a8a69dd RR |
52 | #define START_USE(vq) |
53 | #define END_USE(vq) | |
54 | #endif | |
55 | ||
56 | struct vring_virtqueue | |
57 | { | |
58 | struct virtqueue vq; | |
59 | ||
60 | /* Actual memory layout for this queue */ | |
61 | struct vring vring; | |
62 | ||
7b21e34f RR |
63 | /* Can we use weak barriers? */ |
64 | bool weak_barriers; | |
65 | ||
0a8a69dd RR |
66 | /* Other side has made a mess, don't try any more. */ |
67 | bool broken; | |
68 | ||
9fa29b9d MM |
69 | /* Host supports indirect buffers */ |
70 | bool indirect; | |
71 | ||
a5c262c5 MT |
72 | /* Host publishes avail event idx */ |
73 | bool event; | |
74 | ||
0a8a69dd RR |
75 | /* Head of free buffer list. */ |
76 | unsigned int free_head; | |
77 | /* Number we've added since last sync. */ | |
78 | unsigned int num_added; | |
79 | ||
80 | /* Last used index we've seen. */ | |
1bc4953e | 81 | u16 last_used_idx; |
0a8a69dd RR |
82 | |
83 | /* How to notify other side. FIXME: commonalize hcalls! */ | |
46f9c2b9 | 84 | bool (*notify)(struct virtqueue *vq); |
0a8a69dd RR |
85 | |
86 | #ifdef DEBUG | |
87 | /* They're supposed to lock for us. */ | |
88 | unsigned int in_use; | |
e93300b1 RR |
89 | |
90 | /* Figure out if their kicks are too delayed. */ | |
91 | bool last_add_time_valid; | |
92 | ktime_t last_add_time; | |
0a8a69dd RR |
93 | #endif |
94 | ||
95 | /* Tokens for callbacks. */ | |
96 | void *data[]; | |
97 | }; | |
98 | ||
99 | #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) | |
100 | ||
13816c76 RR |
101 | static inline struct scatterlist *sg_next_chained(struct scatterlist *sg, |
102 | unsigned int *count) | |
103 | { | |
104 | return sg_next(sg); | |
105 | } | |
106 | ||
107 | static inline struct scatterlist *sg_next_arr(struct scatterlist *sg, | |
108 | unsigned int *count) | |
109 | { | |
110 | if (--(*count) == 0) | |
111 | return NULL; | |
112 | return sg + 1; | |
113 | } | |
114 | ||
9fa29b9d | 115 | /* Set up an indirect table of descriptors and add it to the queue. */ |
13816c76 RR |
116 | static inline int vring_add_indirect(struct vring_virtqueue *vq, |
117 | struct scatterlist *sgs[], | |
118 | struct scatterlist *(*next) | |
119 | (struct scatterlist *, unsigned int *), | |
120 | unsigned int total_sg, | |
121 | unsigned int total_out, | |
122 | unsigned int total_in, | |
123 | unsigned int out_sgs, | |
124 | unsigned int in_sgs, | |
125 | gfp_t gfp) | |
9fa29b9d MM |
126 | { |
127 | struct vring_desc *desc; | |
128 | unsigned head; | |
13816c76 RR |
129 | struct scatterlist *sg; |
130 | int i, n; | |
9fa29b9d | 131 | |
b92b1b89 WD |
132 | /* |
133 | * We require lowmem mappings for the descriptors because | |
134 | * otherwise virt_to_phys will give us bogus addresses in the | |
135 | * virtqueue. | |
136 | */ | |
137 | gfp &= ~(__GFP_HIGHMEM | __GFP_HIGH); | |
138 | ||
13816c76 | 139 | desc = kmalloc(total_sg * sizeof(struct vring_desc), gfp); |
9fa29b9d | 140 | if (!desc) |
686d3637 | 141 | return -ENOMEM; |
9fa29b9d | 142 | |
13816c76 RR |
143 | /* Transfer entries from the sg lists into the indirect page */ |
144 | i = 0; | |
145 | for (n = 0; n < out_sgs; n++) { | |
146 | for (sg = sgs[n]; sg; sg = next(sg, &total_out)) { | |
147 | desc[i].flags = VRING_DESC_F_NEXT; | |
148 | desc[i].addr = sg_phys(sg); | |
149 | desc[i].len = sg->length; | |
150 | desc[i].next = i+1; | |
151 | i++; | |
152 | } | |
9fa29b9d | 153 | } |
13816c76 RR |
154 | for (; n < (out_sgs + in_sgs); n++) { |
155 | for (sg = sgs[n]; sg; sg = next(sg, &total_in)) { | |
156 | desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
157 | desc[i].addr = sg_phys(sg); | |
158 | desc[i].len = sg->length; | |
159 | desc[i].next = i+1; | |
160 | i++; | |
161 | } | |
9fa29b9d | 162 | } |
13816c76 | 163 | BUG_ON(i != total_sg); |
9fa29b9d MM |
164 | |
165 | /* Last one doesn't continue. */ | |
166 | desc[i-1].flags &= ~VRING_DESC_F_NEXT; | |
167 | desc[i-1].next = 0; | |
168 | ||
169 | /* We're about to use a buffer */ | |
06ca287d | 170 | vq->vq.num_free--; |
9fa29b9d MM |
171 | |
172 | /* Use a single buffer which doesn't continue */ | |
173 | head = vq->free_head; | |
174 | vq->vring.desc[head].flags = VRING_DESC_F_INDIRECT; | |
175 | vq->vring.desc[head].addr = virt_to_phys(desc); | |
bb478d8b RR |
176 | /* kmemleak gives a false positive, as it's hidden by virt_to_phys */ |
177 | kmemleak_ignore(desc); | |
9fa29b9d MM |
178 | vq->vring.desc[head].len = i * sizeof(struct vring_desc); |
179 | ||
180 | /* Update free pointer */ | |
181 | vq->free_head = vq->vring.desc[head].next; | |
182 | ||
183 | return head; | |
184 | } | |
185 | ||
13816c76 RR |
186 | static inline int virtqueue_add(struct virtqueue *_vq, |
187 | struct scatterlist *sgs[], | |
188 | struct scatterlist *(*next) | |
189 | (struct scatterlist *, unsigned int *), | |
190 | unsigned int total_out, | |
191 | unsigned int total_in, | |
192 | unsigned int out_sgs, | |
193 | unsigned int in_sgs, | |
194 | void *data, | |
195 | gfp_t gfp) | |
0a8a69dd RR |
196 | { |
197 | struct vring_virtqueue *vq = to_vvq(_vq); | |
13816c76 RR |
198 | struct scatterlist *sg; |
199 | unsigned int i, n, avail, uninitialized_var(prev), total_sg; | |
1fe9b6fe | 200 | int head; |
0a8a69dd | 201 | |
9fa29b9d MM |
202 | START_USE(vq); |
203 | ||
0a8a69dd | 204 | BUG_ON(data == NULL); |
9fa29b9d | 205 | |
e93300b1 RR |
206 | #ifdef DEBUG |
207 | { | |
208 | ktime_t now = ktime_get(); | |
209 | ||
210 | /* No kick or get, with .1 second between? Warn. */ | |
211 | if (vq->last_add_time_valid) | |
212 | WARN_ON(ktime_to_ms(ktime_sub(now, vq->last_add_time)) | |
213 | > 100); | |
214 | vq->last_add_time = now; | |
215 | vq->last_add_time_valid = true; | |
216 | } | |
217 | #endif | |
218 | ||
13816c76 RR |
219 | total_sg = total_in + total_out; |
220 | ||
9fa29b9d MM |
221 | /* If the host supports indirect descriptor tables, and we have multiple |
222 | * buffers, then go indirect. FIXME: tune this threshold */ | |
13816c76 RR |
223 | if (vq->indirect && total_sg > 1 && vq->vq.num_free) { |
224 | head = vring_add_indirect(vq, sgs, next, total_sg, total_out, | |
225 | total_in, | |
226 | out_sgs, in_sgs, gfp); | |
1fe9b6fe | 227 | if (likely(head >= 0)) |
9fa29b9d MM |
228 | goto add_head; |
229 | } | |
230 | ||
13816c76 RR |
231 | BUG_ON(total_sg > vq->vring.num); |
232 | BUG_ON(total_sg == 0); | |
0a8a69dd | 233 | |
13816c76 | 234 | if (vq->vq.num_free < total_sg) { |
0a8a69dd | 235 | pr_debug("Can't add buf len %i - avail = %i\n", |
13816c76 | 236 | total_sg, vq->vq.num_free); |
44653eae RR |
237 | /* FIXME: for historical reasons, we force a notify here if |
238 | * there are outgoing parts to the buffer. Presumably the | |
239 | * host should service the ring ASAP. */ | |
13816c76 | 240 | if (out_sgs) |
44653eae | 241 | vq->notify(&vq->vq); |
0a8a69dd RR |
242 | END_USE(vq); |
243 | return -ENOSPC; | |
244 | } | |
245 | ||
246 | /* We're about to use some buffers from the free list. */ | |
13816c76 RR |
247 | vq->vq.num_free -= total_sg; |
248 | ||
249 | head = i = vq->free_head; | |
250 | for (n = 0; n < out_sgs; n++) { | |
251 | for (sg = sgs[n]; sg; sg = next(sg, &total_out)) { | |
252 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT; | |
253 | vq->vring.desc[i].addr = sg_phys(sg); | |
254 | vq->vring.desc[i].len = sg->length; | |
255 | prev = i; | |
256 | i = vq->vring.desc[i].next; | |
257 | } | |
0a8a69dd | 258 | } |
13816c76 RR |
259 | for (; n < (out_sgs + in_sgs); n++) { |
260 | for (sg = sgs[n]; sg; sg = next(sg, &total_in)) { | |
261 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
262 | vq->vring.desc[i].addr = sg_phys(sg); | |
263 | vq->vring.desc[i].len = sg->length; | |
264 | prev = i; | |
265 | i = vq->vring.desc[i].next; | |
266 | } | |
0a8a69dd RR |
267 | } |
268 | /* Last one doesn't continue. */ | |
269 | vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT; | |
270 | ||
271 | /* Update free pointer */ | |
272 | vq->free_head = i; | |
273 | ||
9fa29b9d | 274 | add_head: |
0a8a69dd RR |
275 | /* Set token. */ |
276 | vq->data[head] = data; | |
277 | ||
278 | /* Put entry in available array (but don't update avail->idx until they | |
3b720b8c | 279 | * do sync). */ |
ee7cd898 | 280 | avail = (vq->vring.avail->idx & (vq->vring.num-1)); |
0a8a69dd RR |
281 | vq->vring.avail->ring[avail] = head; |
282 | ||
ee7cd898 RR |
283 | /* Descriptors and available array need to be set before we expose the |
284 | * new available array entries. */ | |
a9a0fef7 | 285 | virtio_wmb(vq->weak_barriers); |
ee7cd898 RR |
286 | vq->vring.avail->idx++; |
287 | vq->num_added++; | |
288 | ||
289 | /* This is very unlikely, but theoretically possible. Kick | |
290 | * just in case. */ | |
291 | if (unlikely(vq->num_added == (1 << 16) - 1)) | |
292 | virtqueue_kick(_vq); | |
293 | ||
0a8a69dd RR |
294 | pr_debug("Added buffer head %i to %p\n", head, vq); |
295 | END_USE(vq); | |
3c1b27d5 | 296 | |
98e8c6bc | 297 | return 0; |
0a8a69dd | 298 | } |
13816c76 | 299 | |
13816c76 RR |
300 | /** |
301 | * virtqueue_add_sgs - expose buffers to other end | |
302 | * @vq: the struct virtqueue we're talking about. | |
303 | * @sgs: array of terminated scatterlists. | |
304 | * @out_num: the number of scatterlists readable by other side | |
305 | * @in_num: the number of scatterlists which are writable (after readable ones) | |
306 | * @data: the token identifying the buffer. | |
307 | * @gfp: how to do memory allocations (if necessary). | |
308 | * | |
309 | * Caller must ensure we don't call this with other virtqueue operations | |
310 | * at the same time (except where noted). | |
311 | * | |
312 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM). | |
313 | */ | |
314 | int virtqueue_add_sgs(struct virtqueue *_vq, | |
315 | struct scatterlist *sgs[], | |
316 | unsigned int out_sgs, | |
317 | unsigned int in_sgs, | |
318 | void *data, | |
319 | gfp_t gfp) | |
320 | { | |
321 | unsigned int i, total_out, total_in; | |
322 | ||
323 | /* Count them first. */ | |
324 | for (i = total_out = total_in = 0; i < out_sgs; i++) { | |
325 | struct scatterlist *sg; | |
326 | for (sg = sgs[i]; sg; sg = sg_next(sg)) | |
327 | total_out++; | |
328 | } | |
329 | for (; i < out_sgs + in_sgs; i++) { | |
330 | struct scatterlist *sg; | |
331 | for (sg = sgs[i]; sg; sg = sg_next(sg)) | |
332 | total_in++; | |
333 | } | |
334 | return virtqueue_add(_vq, sgs, sg_next_chained, | |
335 | total_out, total_in, out_sgs, in_sgs, data, gfp); | |
336 | } | |
337 | EXPORT_SYMBOL_GPL(virtqueue_add_sgs); | |
338 | ||
282edb36 RR |
339 | /** |
340 | * virtqueue_add_outbuf - expose output buffers to other end | |
341 | * @vq: the struct virtqueue we're talking about. | |
342 | * @sgs: array of scatterlists (need not be terminated!) | |
343 | * @num: the number of scatterlists readable by other side | |
344 | * @data: the token identifying the buffer. | |
345 | * @gfp: how to do memory allocations (if necessary). | |
346 | * | |
347 | * Caller must ensure we don't call this with other virtqueue operations | |
348 | * at the same time (except where noted). | |
349 | * | |
350 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM). | |
351 | */ | |
352 | int virtqueue_add_outbuf(struct virtqueue *vq, | |
353 | struct scatterlist sg[], unsigned int num, | |
354 | void *data, | |
355 | gfp_t gfp) | |
356 | { | |
357 | return virtqueue_add(vq, &sg, sg_next_arr, num, 0, 1, 0, data, gfp); | |
358 | } | |
359 | EXPORT_SYMBOL_GPL(virtqueue_add_outbuf); | |
360 | ||
361 | /** | |
362 | * virtqueue_add_inbuf - expose input buffers to other end | |
363 | * @vq: the struct virtqueue we're talking about. | |
364 | * @sgs: array of scatterlists (need not be terminated!) | |
365 | * @num: the number of scatterlists writable by other side | |
366 | * @data: the token identifying the buffer. | |
367 | * @gfp: how to do memory allocations (if necessary). | |
368 | * | |
369 | * Caller must ensure we don't call this with other virtqueue operations | |
370 | * at the same time (except where noted). | |
371 | * | |
372 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM). | |
373 | */ | |
374 | int virtqueue_add_inbuf(struct virtqueue *vq, | |
375 | struct scatterlist sg[], unsigned int num, | |
376 | void *data, | |
377 | gfp_t gfp) | |
378 | { | |
379 | return virtqueue_add(vq, &sg, sg_next_arr, 0, num, 0, 1, data, gfp); | |
380 | } | |
381 | EXPORT_SYMBOL_GPL(virtqueue_add_inbuf); | |
382 | ||
5dfc1762 | 383 | /** |
41f0377f | 384 | * virtqueue_kick_prepare - first half of split virtqueue_kick call. |
5dfc1762 RR |
385 | * @vq: the struct virtqueue |
386 | * | |
41f0377f RR |
387 | * Instead of virtqueue_kick(), you can do: |
388 | * if (virtqueue_kick_prepare(vq)) | |
389 | * virtqueue_notify(vq); | |
5dfc1762 | 390 | * |
41f0377f RR |
391 | * This is sometimes useful because the virtqueue_kick_prepare() needs |
392 | * to be serialized, but the actual virtqueue_notify() call does not. | |
5dfc1762 | 393 | */ |
41f0377f | 394 | bool virtqueue_kick_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
395 | { |
396 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a5c262c5 | 397 | u16 new, old; |
41f0377f RR |
398 | bool needs_kick; |
399 | ||
0a8a69dd | 400 | START_USE(vq); |
a72caae2 JW |
401 | /* We need to expose available array entries before checking avail |
402 | * event. */ | |
a9a0fef7 | 403 | virtio_mb(vq->weak_barriers); |
0a8a69dd | 404 | |
ee7cd898 RR |
405 | old = vq->vring.avail->idx - vq->num_added; |
406 | new = vq->vring.avail->idx; | |
0a8a69dd RR |
407 | vq->num_added = 0; |
408 | ||
e93300b1 RR |
409 | #ifdef DEBUG |
410 | if (vq->last_add_time_valid) { | |
411 | WARN_ON(ktime_to_ms(ktime_sub(ktime_get(), | |
412 | vq->last_add_time)) > 100); | |
413 | } | |
414 | vq->last_add_time_valid = false; | |
415 | #endif | |
416 | ||
41f0377f RR |
417 | if (vq->event) { |
418 | needs_kick = vring_need_event(vring_avail_event(&vq->vring), | |
419 | new, old); | |
420 | } else { | |
421 | needs_kick = !(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY); | |
422 | } | |
0a8a69dd | 423 | END_USE(vq); |
41f0377f RR |
424 | return needs_kick; |
425 | } | |
426 | EXPORT_SYMBOL_GPL(virtqueue_kick_prepare); | |
427 | ||
428 | /** | |
429 | * virtqueue_notify - second half of split virtqueue_kick call. | |
430 | * @vq: the struct virtqueue | |
431 | * | |
432 | * This does not need to be serialized. | |
5b1bf7cb HG |
433 | * |
434 | * Returns false if host notify failed or queue is broken, otherwise true. | |
41f0377f | 435 | */ |
5b1bf7cb | 436 | bool virtqueue_notify(struct virtqueue *_vq) |
41f0377f RR |
437 | { |
438 | struct vring_virtqueue *vq = to_vvq(_vq); | |
439 | ||
5b1bf7cb HG |
440 | if (unlikely(vq->broken)) |
441 | return false; | |
442 | ||
41f0377f | 443 | /* Prod other side to tell it about changes. */ |
5b1bf7cb HG |
444 | if (vq->notify(_vq) < 0) { |
445 | vq->broken = true; | |
446 | return false; | |
447 | } | |
448 | return true; | |
41f0377f RR |
449 | } |
450 | EXPORT_SYMBOL_GPL(virtqueue_notify); | |
451 | ||
452 | /** | |
453 | * virtqueue_kick - update after add_buf | |
454 | * @vq: the struct virtqueue | |
455 | * | |
b3087e48 | 456 | * After one or more virtqueue_add_* calls, invoke this to kick |
41f0377f RR |
457 | * the other side. |
458 | * | |
459 | * Caller must ensure we don't call this with other virtqueue | |
460 | * operations at the same time (except where noted). | |
5b1bf7cb HG |
461 | * |
462 | * Returns false if kick failed, otherwise true. | |
41f0377f | 463 | */ |
5b1bf7cb | 464 | bool virtqueue_kick(struct virtqueue *vq) |
41f0377f RR |
465 | { |
466 | if (virtqueue_kick_prepare(vq)) | |
5b1bf7cb HG |
467 | return virtqueue_notify(vq); |
468 | return true; | |
0a8a69dd | 469 | } |
7c5e9ed0 | 470 | EXPORT_SYMBOL_GPL(virtqueue_kick); |
0a8a69dd RR |
471 | |
472 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head) | |
473 | { | |
474 | unsigned int i; | |
475 | ||
476 | /* Clear data ptr. */ | |
477 | vq->data[head] = NULL; | |
478 | ||
479 | /* Put back on free list: find end */ | |
480 | i = head; | |
9fa29b9d MM |
481 | |
482 | /* Free the indirect table */ | |
483 | if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT) | |
484 | kfree(phys_to_virt(vq->vring.desc[i].addr)); | |
485 | ||
0a8a69dd RR |
486 | while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) { |
487 | i = vq->vring.desc[i].next; | |
06ca287d | 488 | vq->vq.num_free++; |
0a8a69dd RR |
489 | } |
490 | ||
491 | vq->vring.desc[i].next = vq->free_head; | |
492 | vq->free_head = head; | |
493 | /* Plus final descriptor */ | |
06ca287d | 494 | vq->vq.num_free++; |
0a8a69dd RR |
495 | } |
496 | ||
0a8a69dd RR |
497 | static inline bool more_used(const struct vring_virtqueue *vq) |
498 | { | |
499 | return vq->last_used_idx != vq->vring.used->idx; | |
500 | } | |
501 | ||
5dfc1762 RR |
502 | /** |
503 | * virtqueue_get_buf - get the next used buffer | |
504 | * @vq: the struct virtqueue we're talking about. | |
505 | * @len: the length written into the buffer | |
506 | * | |
507 | * If the driver wrote data into the buffer, @len will be set to the | |
508 | * amount written. This means you don't need to clear the buffer | |
509 | * beforehand to ensure there's no data leakage in the case of short | |
510 | * writes. | |
511 | * | |
512 | * Caller must ensure we don't call this with other virtqueue | |
513 | * operations at the same time (except where noted). | |
514 | * | |
515 | * Returns NULL if there are no used buffers, or the "data" token | |
b3087e48 | 516 | * handed to virtqueue_add_*(). |
5dfc1762 | 517 | */ |
7c5e9ed0 | 518 | void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len) |
0a8a69dd RR |
519 | { |
520 | struct vring_virtqueue *vq = to_vvq(_vq); | |
521 | void *ret; | |
522 | unsigned int i; | |
3b720b8c | 523 | u16 last_used; |
0a8a69dd RR |
524 | |
525 | START_USE(vq); | |
526 | ||
5ef82752 RR |
527 | if (unlikely(vq->broken)) { |
528 | END_USE(vq); | |
529 | return NULL; | |
530 | } | |
531 | ||
0a8a69dd RR |
532 | if (!more_used(vq)) { |
533 | pr_debug("No more buffers in queue\n"); | |
534 | END_USE(vq); | |
535 | return NULL; | |
536 | } | |
537 | ||
2d61ba95 | 538 | /* Only get used array entries after they have been exposed by host. */ |
a9a0fef7 | 539 | virtio_rmb(vq->weak_barriers); |
2d61ba95 | 540 | |
3b720b8c RR |
541 | last_used = (vq->last_used_idx & (vq->vring.num - 1)); |
542 | i = vq->vring.used->ring[last_used].id; | |
543 | *len = vq->vring.used->ring[last_used].len; | |
0a8a69dd RR |
544 | |
545 | if (unlikely(i >= vq->vring.num)) { | |
546 | BAD_RING(vq, "id %u out of range\n", i); | |
547 | return NULL; | |
548 | } | |
549 | if (unlikely(!vq->data[i])) { | |
550 | BAD_RING(vq, "id %u is not a head!\n", i); | |
551 | return NULL; | |
552 | } | |
553 | ||
554 | /* detach_buf clears data, so grab it now. */ | |
555 | ret = vq->data[i]; | |
556 | detach_buf(vq, i); | |
557 | vq->last_used_idx++; | |
a5c262c5 MT |
558 | /* If we expect an interrupt for the next entry, tell host |
559 | * by writing event index and flush out the write before | |
560 | * the read in the next get_buf call. */ | |
561 | if (!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) { | |
562 | vring_used_event(&vq->vring) = vq->last_used_idx; | |
a9a0fef7 | 563 | virtio_mb(vq->weak_barriers); |
a5c262c5 MT |
564 | } |
565 | ||
e93300b1 RR |
566 | #ifdef DEBUG |
567 | vq->last_add_time_valid = false; | |
568 | #endif | |
569 | ||
0a8a69dd RR |
570 | END_USE(vq); |
571 | return ret; | |
572 | } | |
7c5e9ed0 | 573 | EXPORT_SYMBOL_GPL(virtqueue_get_buf); |
0a8a69dd | 574 | |
5dfc1762 RR |
575 | /** |
576 | * virtqueue_disable_cb - disable callbacks | |
577 | * @vq: the struct virtqueue we're talking about. | |
578 | * | |
579 | * Note that this is not necessarily synchronous, hence unreliable and only | |
580 | * useful as an optimization. | |
581 | * | |
582 | * Unlike other operations, this need not be serialized. | |
583 | */ | |
7c5e9ed0 | 584 | void virtqueue_disable_cb(struct virtqueue *_vq) |
18445c4d RR |
585 | { |
586 | struct vring_virtqueue *vq = to_vvq(_vq); | |
587 | ||
18445c4d | 588 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; |
18445c4d | 589 | } |
7c5e9ed0 | 590 | EXPORT_SYMBOL_GPL(virtqueue_disable_cb); |
18445c4d | 591 | |
5dfc1762 | 592 | /** |
cc229884 | 593 | * virtqueue_enable_cb_prepare - restart callbacks after disable_cb |
5dfc1762 RR |
594 | * @vq: the struct virtqueue we're talking about. |
595 | * | |
cc229884 MT |
596 | * This re-enables callbacks; it returns current queue state |
597 | * in an opaque unsigned value. This value should be later tested by | |
598 | * virtqueue_poll, to detect a possible race between the driver checking for | |
599 | * more work, and enabling callbacks. | |
5dfc1762 RR |
600 | * |
601 | * Caller must ensure we don't call this with other virtqueue | |
602 | * operations at the same time (except where noted). | |
603 | */ | |
cc229884 | 604 | unsigned virtqueue_enable_cb_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
605 | { |
606 | struct vring_virtqueue *vq = to_vvq(_vq); | |
cc229884 | 607 | u16 last_used_idx; |
0a8a69dd RR |
608 | |
609 | START_USE(vq); | |
0a8a69dd RR |
610 | |
611 | /* We optimistically turn back on interrupts, then check if there was | |
612 | * more to do. */ | |
a5c262c5 MT |
613 | /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to |
614 | * either clear the flags bit or point the event index at the next | |
615 | * entry. Always do both to keep code simple. */ | |
0a8a69dd | 616 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; |
cc229884 MT |
617 | vring_used_event(&vq->vring) = last_used_idx = vq->last_used_idx; |
618 | END_USE(vq); | |
619 | return last_used_idx; | |
620 | } | |
621 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_prepare); | |
622 | ||
623 | /** | |
624 | * virtqueue_poll - query pending used buffers | |
625 | * @vq: the struct virtqueue we're talking about. | |
626 | * @last_used_idx: virtqueue state (from call to virtqueue_enable_cb_prepare). | |
627 | * | |
628 | * Returns "true" if there are pending used buffers in the queue. | |
629 | * | |
630 | * This does not need to be serialized. | |
631 | */ | |
632 | bool virtqueue_poll(struct virtqueue *_vq, unsigned last_used_idx) | |
633 | { | |
634 | struct vring_virtqueue *vq = to_vvq(_vq); | |
635 | ||
a9a0fef7 | 636 | virtio_mb(vq->weak_barriers); |
cc229884 MT |
637 | return (u16)last_used_idx != vq->vring.used->idx; |
638 | } | |
639 | EXPORT_SYMBOL_GPL(virtqueue_poll); | |
0a8a69dd | 640 | |
cc229884 MT |
641 | /** |
642 | * virtqueue_enable_cb - restart callbacks after disable_cb. | |
643 | * @vq: the struct virtqueue we're talking about. | |
644 | * | |
645 | * This re-enables callbacks; it returns "false" if there are pending | |
646 | * buffers in the queue, to detect a possible race between the driver | |
647 | * checking for more work, and enabling callbacks. | |
648 | * | |
649 | * Caller must ensure we don't call this with other virtqueue | |
650 | * operations at the same time (except where noted). | |
651 | */ | |
652 | bool virtqueue_enable_cb(struct virtqueue *_vq) | |
653 | { | |
654 | unsigned last_used_idx = virtqueue_enable_cb_prepare(_vq); | |
655 | return !virtqueue_poll(_vq, last_used_idx); | |
0a8a69dd | 656 | } |
7c5e9ed0 | 657 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb); |
0a8a69dd | 658 | |
5dfc1762 RR |
659 | /** |
660 | * virtqueue_enable_cb_delayed - restart callbacks after disable_cb. | |
661 | * @vq: the struct virtqueue we're talking about. | |
662 | * | |
663 | * This re-enables callbacks but hints to the other side to delay | |
664 | * interrupts until most of the available buffers have been processed; | |
665 | * it returns "false" if there are many pending buffers in the queue, | |
666 | * to detect a possible race between the driver checking for more work, | |
667 | * and enabling callbacks. | |
668 | * | |
669 | * Caller must ensure we don't call this with other virtqueue | |
670 | * operations at the same time (except where noted). | |
671 | */ | |
7ab358c2 MT |
672 | bool virtqueue_enable_cb_delayed(struct virtqueue *_vq) |
673 | { | |
674 | struct vring_virtqueue *vq = to_vvq(_vq); | |
675 | u16 bufs; | |
676 | ||
677 | START_USE(vq); | |
678 | ||
679 | /* We optimistically turn back on interrupts, then check if there was | |
680 | * more to do. */ | |
681 | /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to | |
682 | * either clear the flags bit or point the event index at the next | |
683 | * entry. Always do both to keep code simple. */ | |
684 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
685 | /* TODO: tune this threshold */ | |
686 | bufs = (u16)(vq->vring.avail->idx - vq->last_used_idx) * 3 / 4; | |
687 | vring_used_event(&vq->vring) = vq->last_used_idx + bufs; | |
a9a0fef7 | 688 | virtio_mb(vq->weak_barriers); |
7ab358c2 MT |
689 | if (unlikely((u16)(vq->vring.used->idx - vq->last_used_idx) > bufs)) { |
690 | END_USE(vq); | |
691 | return false; | |
692 | } | |
693 | ||
694 | END_USE(vq); | |
695 | return true; | |
696 | } | |
697 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed); | |
698 | ||
5dfc1762 RR |
699 | /** |
700 | * virtqueue_detach_unused_buf - detach first unused buffer | |
701 | * @vq: the struct virtqueue we're talking about. | |
702 | * | |
b3087e48 | 703 | * Returns NULL or the "data" token handed to virtqueue_add_*(). |
5dfc1762 RR |
704 | * This is not valid on an active queue; it is useful only for device |
705 | * shutdown. | |
706 | */ | |
7c5e9ed0 | 707 | void *virtqueue_detach_unused_buf(struct virtqueue *_vq) |
c021eac4 SM |
708 | { |
709 | struct vring_virtqueue *vq = to_vvq(_vq); | |
710 | unsigned int i; | |
711 | void *buf; | |
712 | ||
713 | START_USE(vq); | |
714 | ||
715 | for (i = 0; i < vq->vring.num; i++) { | |
716 | if (!vq->data[i]) | |
717 | continue; | |
718 | /* detach_buf clears data, so grab it now. */ | |
719 | buf = vq->data[i]; | |
720 | detach_buf(vq, i); | |
b3258ff1 | 721 | vq->vring.avail->idx--; |
c021eac4 SM |
722 | END_USE(vq); |
723 | return buf; | |
724 | } | |
725 | /* That should have freed everything. */ | |
06ca287d | 726 | BUG_ON(vq->vq.num_free != vq->vring.num); |
c021eac4 SM |
727 | |
728 | END_USE(vq); | |
729 | return NULL; | |
730 | } | |
7c5e9ed0 | 731 | EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf); |
c021eac4 | 732 | |
0a8a69dd RR |
733 | irqreturn_t vring_interrupt(int irq, void *_vq) |
734 | { | |
735 | struct vring_virtqueue *vq = to_vvq(_vq); | |
736 | ||
737 | if (!more_used(vq)) { | |
738 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
739 | return IRQ_NONE; | |
740 | } | |
741 | ||
742 | if (unlikely(vq->broken)) | |
743 | return IRQ_HANDLED; | |
744 | ||
745 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); | |
18445c4d RR |
746 | if (vq->vq.callback) |
747 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
748 | |
749 | return IRQ_HANDLED; | |
750 | } | |
c6fd4701 | 751 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd | 752 | |
17bb6d40 JW |
753 | struct virtqueue *vring_new_virtqueue(unsigned int index, |
754 | unsigned int num, | |
87c7d57c | 755 | unsigned int vring_align, |
0a8a69dd | 756 | struct virtio_device *vdev, |
7b21e34f | 757 | bool weak_barriers, |
0a8a69dd | 758 | void *pages, |
46f9c2b9 | 759 | bool (*notify)(struct virtqueue *), |
9499f5e7 RR |
760 | void (*callback)(struct virtqueue *), |
761 | const char *name) | |
0a8a69dd RR |
762 | { |
763 | struct vring_virtqueue *vq; | |
764 | unsigned int i; | |
765 | ||
42b36cc0 RR |
766 | /* We assume num is a power of 2. */ |
767 | if (num & (num - 1)) { | |
768 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
769 | return NULL; | |
770 | } | |
771 | ||
0a8a69dd RR |
772 | vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL); |
773 | if (!vq) | |
774 | return NULL; | |
775 | ||
87c7d57c | 776 | vring_init(&vq->vring, num, pages, vring_align); |
0a8a69dd RR |
777 | vq->vq.callback = callback; |
778 | vq->vq.vdev = vdev; | |
9499f5e7 | 779 | vq->vq.name = name; |
06ca287d RR |
780 | vq->vq.num_free = num; |
781 | vq->vq.index = index; | |
0a8a69dd | 782 | vq->notify = notify; |
7b21e34f | 783 | vq->weak_barriers = weak_barriers; |
0a8a69dd RR |
784 | vq->broken = false; |
785 | vq->last_used_idx = 0; | |
786 | vq->num_added = 0; | |
9499f5e7 | 787 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0a8a69dd RR |
788 | #ifdef DEBUG |
789 | vq->in_use = false; | |
e93300b1 | 790 | vq->last_add_time_valid = false; |
0a8a69dd RR |
791 | #endif |
792 | ||
9fa29b9d | 793 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC); |
a5c262c5 | 794 | vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); |
9fa29b9d | 795 | |
0a8a69dd RR |
796 | /* No callback? Tell other side not to bother us. */ |
797 | if (!callback) | |
798 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; | |
799 | ||
800 | /* Put everything in free lists. */ | |
0a8a69dd | 801 | vq->free_head = 0; |
3b870624 | 802 | for (i = 0; i < num-1; i++) { |
0a8a69dd | 803 | vq->vring.desc[i].next = i+1; |
3b870624 AS |
804 | vq->data[i] = NULL; |
805 | } | |
806 | vq->data[i] = NULL; | |
0a8a69dd RR |
807 | |
808 | return &vq->vq; | |
809 | } | |
c6fd4701 | 810 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd RR |
811 | |
812 | void vring_del_virtqueue(struct virtqueue *vq) | |
813 | { | |
9499f5e7 | 814 | list_del(&vq->list); |
0a8a69dd RR |
815 | kfree(to_vvq(vq)); |
816 | } | |
c6fd4701 | 817 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 818 | |
e34f8725 RR |
819 | /* Manipulates transport-specific feature bits. */ |
820 | void vring_transport_features(struct virtio_device *vdev) | |
821 | { | |
822 | unsigned int i; | |
823 | ||
824 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
825 | switch (i) { | |
9fa29b9d MM |
826 | case VIRTIO_RING_F_INDIRECT_DESC: |
827 | break; | |
a5c262c5 MT |
828 | case VIRTIO_RING_F_EVENT_IDX: |
829 | break; | |
e34f8725 RR |
830 | default: |
831 | /* We don't understand this bit. */ | |
832 | clear_bit(i, vdev->features); | |
833 | } | |
834 | } | |
835 | } | |
836 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
837 | ||
5dfc1762 RR |
838 | /** |
839 | * virtqueue_get_vring_size - return the size of the virtqueue's vring | |
840 | * @vq: the struct virtqueue containing the vring of interest. | |
841 | * | |
842 | * Returns the size of the vring. This is mainly used for boasting to | |
843 | * userspace. Unlike other operations, this need not be serialized. | |
844 | */ | |
8f9f4668 RJ |
845 | unsigned int virtqueue_get_vring_size(struct virtqueue *_vq) |
846 | { | |
847 | ||
848 | struct vring_virtqueue *vq = to_vvq(_vq); | |
849 | ||
850 | return vq->vring.num; | |
851 | } | |
852 | EXPORT_SYMBOL_GPL(virtqueue_get_vring_size); | |
853 | ||
c6fd4701 | 854 | MODULE_LICENSE("GPL"); |