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! */ | |
84 | void (*notify)(struct virtqueue *vq); | |
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); | |
176 | vq->vring.desc[head].len = i * sizeof(struct vring_desc); | |
177 | ||
178 | /* Update free pointer */ | |
179 | vq->free_head = vq->vring.desc[head].next; | |
180 | ||
181 | return head; | |
182 | } | |
183 | ||
13816c76 RR |
184 | static inline int virtqueue_add(struct virtqueue *_vq, |
185 | struct scatterlist *sgs[], | |
186 | struct scatterlist *(*next) | |
187 | (struct scatterlist *, unsigned int *), | |
188 | unsigned int total_out, | |
189 | unsigned int total_in, | |
190 | unsigned int out_sgs, | |
191 | unsigned int in_sgs, | |
192 | void *data, | |
193 | gfp_t gfp) | |
0a8a69dd RR |
194 | { |
195 | struct vring_virtqueue *vq = to_vvq(_vq); | |
13816c76 RR |
196 | struct scatterlist *sg; |
197 | unsigned int i, n, avail, uninitialized_var(prev), total_sg; | |
1fe9b6fe | 198 | int head; |
0a8a69dd | 199 | |
9fa29b9d MM |
200 | START_USE(vq); |
201 | ||
0a8a69dd | 202 | BUG_ON(data == NULL); |
9fa29b9d | 203 | |
e93300b1 RR |
204 | #ifdef DEBUG |
205 | { | |
206 | ktime_t now = ktime_get(); | |
207 | ||
208 | /* No kick or get, with .1 second between? Warn. */ | |
209 | if (vq->last_add_time_valid) | |
210 | WARN_ON(ktime_to_ms(ktime_sub(now, vq->last_add_time)) | |
211 | > 100); | |
212 | vq->last_add_time = now; | |
213 | vq->last_add_time_valid = true; | |
214 | } | |
215 | #endif | |
216 | ||
13816c76 RR |
217 | total_sg = total_in + total_out; |
218 | ||
9fa29b9d MM |
219 | /* If the host supports indirect descriptor tables, and we have multiple |
220 | * buffers, then go indirect. FIXME: tune this threshold */ | |
13816c76 RR |
221 | if (vq->indirect && total_sg > 1 && vq->vq.num_free) { |
222 | head = vring_add_indirect(vq, sgs, next, total_sg, total_out, | |
223 | total_in, | |
224 | out_sgs, in_sgs, gfp); | |
1fe9b6fe | 225 | if (likely(head >= 0)) |
9fa29b9d MM |
226 | goto add_head; |
227 | } | |
228 | ||
13816c76 RR |
229 | BUG_ON(total_sg > vq->vring.num); |
230 | BUG_ON(total_sg == 0); | |
0a8a69dd | 231 | |
13816c76 | 232 | if (vq->vq.num_free < total_sg) { |
0a8a69dd | 233 | pr_debug("Can't add buf len %i - avail = %i\n", |
13816c76 | 234 | total_sg, vq->vq.num_free); |
44653eae RR |
235 | /* FIXME: for historical reasons, we force a notify here if |
236 | * there are outgoing parts to the buffer. Presumably the | |
237 | * host should service the ring ASAP. */ | |
13816c76 | 238 | if (out_sgs) |
44653eae | 239 | vq->notify(&vq->vq); |
0a8a69dd RR |
240 | END_USE(vq); |
241 | return -ENOSPC; | |
242 | } | |
243 | ||
244 | /* We're about to use some buffers from the free list. */ | |
13816c76 RR |
245 | vq->vq.num_free -= total_sg; |
246 | ||
247 | head = i = vq->free_head; | |
248 | for (n = 0; n < out_sgs; n++) { | |
249 | for (sg = sgs[n]; sg; sg = next(sg, &total_out)) { | |
250 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT; | |
251 | vq->vring.desc[i].addr = sg_phys(sg); | |
252 | vq->vring.desc[i].len = sg->length; | |
253 | prev = i; | |
254 | i = vq->vring.desc[i].next; | |
255 | } | |
0a8a69dd | 256 | } |
13816c76 RR |
257 | for (; n < (out_sgs + in_sgs); n++) { |
258 | for (sg = sgs[n]; sg; sg = next(sg, &total_in)) { | |
259 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
260 | vq->vring.desc[i].addr = sg_phys(sg); | |
261 | vq->vring.desc[i].len = sg->length; | |
262 | prev = i; | |
263 | i = vq->vring.desc[i].next; | |
264 | } | |
0a8a69dd RR |
265 | } |
266 | /* Last one doesn't continue. */ | |
267 | vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT; | |
268 | ||
269 | /* Update free pointer */ | |
270 | vq->free_head = i; | |
271 | ||
9fa29b9d | 272 | add_head: |
0a8a69dd RR |
273 | /* Set token. */ |
274 | vq->data[head] = data; | |
275 | ||
276 | /* Put entry in available array (but don't update avail->idx until they | |
3b720b8c | 277 | * do sync). */ |
ee7cd898 | 278 | avail = (vq->vring.avail->idx & (vq->vring.num-1)); |
0a8a69dd RR |
279 | vq->vring.avail->ring[avail] = head; |
280 | ||
ee7cd898 RR |
281 | /* Descriptors and available array need to be set before we expose the |
282 | * new available array entries. */ | |
a9a0fef7 | 283 | virtio_wmb(vq->weak_barriers); |
ee7cd898 RR |
284 | vq->vring.avail->idx++; |
285 | vq->num_added++; | |
286 | ||
287 | /* This is very unlikely, but theoretically possible. Kick | |
288 | * just in case. */ | |
289 | if (unlikely(vq->num_added == (1 << 16) - 1)) | |
290 | virtqueue_kick(_vq); | |
291 | ||
0a8a69dd RR |
292 | pr_debug("Added buffer head %i to %p\n", head, vq); |
293 | END_USE(vq); | |
3c1b27d5 | 294 | |
98e8c6bc | 295 | return 0; |
0a8a69dd | 296 | } |
13816c76 RR |
297 | |
298 | /** | |
299 | * virtqueue_add_buf - expose buffer to other end | |
300 | * @vq: the struct virtqueue we're talking about. | |
301 | * @sg: the description of the buffer(s). | |
302 | * @out_num: the number of sg readable by other side | |
303 | * @in_num: the number of sg which are writable (after readable ones) | |
304 | * @data: the token identifying the buffer. | |
305 | * @gfp: how to do memory allocations (if necessary). | |
306 | * | |
307 | * Caller must ensure we don't call this with other virtqueue operations | |
308 | * at the same time (except where noted). | |
309 | * | |
310 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM). | |
311 | */ | |
312 | int virtqueue_add_buf(struct virtqueue *_vq, | |
313 | struct scatterlist sg[], | |
314 | unsigned int out, | |
315 | unsigned int in, | |
316 | void *data, | |
317 | gfp_t gfp) | |
318 | { | |
319 | struct scatterlist *sgs[2]; | |
320 | ||
321 | sgs[0] = sg; | |
322 | sgs[1] = sg + out; | |
323 | ||
324 | return virtqueue_add(_vq, sgs, sg_next_arr, | |
325 | out, in, out ? 1 : 0, in ? 1 : 0, data, gfp); | |
326 | } | |
f96fde41 | 327 | EXPORT_SYMBOL_GPL(virtqueue_add_buf); |
0a8a69dd | 328 | |
13816c76 RR |
329 | /** |
330 | * virtqueue_add_sgs - expose buffers to other end | |
331 | * @vq: the struct virtqueue we're talking about. | |
332 | * @sgs: array of terminated scatterlists. | |
333 | * @out_num: the number of scatterlists readable by other side | |
334 | * @in_num: the number of scatterlists which are writable (after readable ones) | |
335 | * @data: the token identifying the buffer. | |
336 | * @gfp: how to do memory allocations (if necessary). | |
337 | * | |
338 | * Caller must ensure we don't call this with other virtqueue operations | |
339 | * at the same time (except where noted). | |
340 | * | |
341 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM). | |
342 | */ | |
343 | int virtqueue_add_sgs(struct virtqueue *_vq, | |
344 | struct scatterlist *sgs[], | |
345 | unsigned int out_sgs, | |
346 | unsigned int in_sgs, | |
347 | void *data, | |
348 | gfp_t gfp) | |
349 | { | |
350 | unsigned int i, total_out, total_in; | |
351 | ||
352 | /* Count them first. */ | |
353 | for (i = total_out = total_in = 0; i < out_sgs; i++) { | |
354 | struct scatterlist *sg; | |
355 | for (sg = sgs[i]; sg; sg = sg_next(sg)) | |
356 | total_out++; | |
357 | } | |
358 | for (; i < out_sgs + in_sgs; i++) { | |
359 | struct scatterlist *sg; | |
360 | for (sg = sgs[i]; sg; sg = sg_next(sg)) | |
361 | total_in++; | |
362 | } | |
363 | return virtqueue_add(_vq, sgs, sg_next_chained, | |
364 | total_out, total_in, out_sgs, in_sgs, data, gfp); | |
365 | } | |
366 | EXPORT_SYMBOL_GPL(virtqueue_add_sgs); | |
367 | ||
282edb36 RR |
368 | /** |
369 | * virtqueue_add_outbuf - expose output buffers to other end | |
370 | * @vq: the struct virtqueue we're talking about. | |
371 | * @sgs: array of scatterlists (need not be terminated!) | |
372 | * @num: the number of scatterlists readable by other side | |
373 | * @data: the token identifying the buffer. | |
374 | * @gfp: how to do memory allocations (if necessary). | |
375 | * | |
376 | * Caller must ensure we don't call this with other virtqueue operations | |
377 | * at the same time (except where noted). | |
378 | * | |
379 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM). | |
380 | */ | |
381 | int virtqueue_add_outbuf(struct virtqueue *vq, | |
382 | struct scatterlist sg[], unsigned int num, | |
383 | void *data, | |
384 | gfp_t gfp) | |
385 | { | |
386 | return virtqueue_add(vq, &sg, sg_next_arr, num, 0, 1, 0, data, gfp); | |
387 | } | |
388 | EXPORT_SYMBOL_GPL(virtqueue_add_outbuf); | |
389 | ||
390 | /** | |
391 | * virtqueue_add_inbuf - expose input buffers to other end | |
392 | * @vq: the struct virtqueue we're talking about. | |
393 | * @sgs: array of scatterlists (need not be terminated!) | |
394 | * @num: the number of scatterlists writable by other side | |
395 | * @data: the token identifying the buffer. | |
396 | * @gfp: how to do memory allocations (if necessary). | |
397 | * | |
398 | * Caller must ensure we don't call this with other virtqueue operations | |
399 | * at the same time (except where noted). | |
400 | * | |
401 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM). | |
402 | */ | |
403 | int virtqueue_add_inbuf(struct virtqueue *vq, | |
404 | struct scatterlist sg[], unsigned int num, | |
405 | void *data, | |
406 | gfp_t gfp) | |
407 | { | |
408 | return virtqueue_add(vq, &sg, sg_next_arr, 0, num, 0, 1, data, gfp); | |
409 | } | |
410 | EXPORT_SYMBOL_GPL(virtqueue_add_inbuf); | |
411 | ||
5dfc1762 | 412 | /** |
41f0377f | 413 | * virtqueue_kick_prepare - first half of split virtqueue_kick call. |
5dfc1762 RR |
414 | * @vq: the struct virtqueue |
415 | * | |
41f0377f RR |
416 | * Instead of virtqueue_kick(), you can do: |
417 | * if (virtqueue_kick_prepare(vq)) | |
418 | * virtqueue_notify(vq); | |
5dfc1762 | 419 | * |
41f0377f RR |
420 | * This is sometimes useful because the virtqueue_kick_prepare() needs |
421 | * to be serialized, but the actual virtqueue_notify() call does not. | |
5dfc1762 | 422 | */ |
41f0377f | 423 | bool virtqueue_kick_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
424 | { |
425 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a5c262c5 | 426 | u16 new, old; |
41f0377f RR |
427 | bool needs_kick; |
428 | ||
0a8a69dd | 429 | START_USE(vq); |
a72caae2 JW |
430 | /* We need to expose available array entries before checking avail |
431 | * event. */ | |
a9a0fef7 | 432 | virtio_mb(vq->weak_barriers); |
0a8a69dd | 433 | |
ee7cd898 RR |
434 | old = vq->vring.avail->idx - vq->num_added; |
435 | new = vq->vring.avail->idx; | |
0a8a69dd RR |
436 | vq->num_added = 0; |
437 | ||
e93300b1 RR |
438 | #ifdef DEBUG |
439 | if (vq->last_add_time_valid) { | |
440 | WARN_ON(ktime_to_ms(ktime_sub(ktime_get(), | |
441 | vq->last_add_time)) > 100); | |
442 | } | |
443 | vq->last_add_time_valid = false; | |
444 | #endif | |
445 | ||
41f0377f RR |
446 | if (vq->event) { |
447 | needs_kick = vring_need_event(vring_avail_event(&vq->vring), | |
448 | new, old); | |
449 | } else { | |
450 | needs_kick = !(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY); | |
451 | } | |
0a8a69dd | 452 | END_USE(vq); |
41f0377f RR |
453 | return needs_kick; |
454 | } | |
455 | EXPORT_SYMBOL_GPL(virtqueue_kick_prepare); | |
456 | ||
457 | /** | |
458 | * virtqueue_notify - second half of split virtqueue_kick call. | |
459 | * @vq: the struct virtqueue | |
460 | * | |
461 | * This does not need to be serialized. | |
462 | */ | |
463 | void virtqueue_notify(struct virtqueue *_vq) | |
464 | { | |
465 | struct vring_virtqueue *vq = to_vvq(_vq); | |
466 | ||
467 | /* Prod other side to tell it about changes. */ | |
468 | vq->notify(_vq); | |
469 | } | |
470 | EXPORT_SYMBOL_GPL(virtqueue_notify); | |
471 | ||
472 | /** | |
473 | * virtqueue_kick - update after add_buf | |
474 | * @vq: the struct virtqueue | |
475 | * | |
476 | * After one or more virtqueue_add_buf calls, invoke this to kick | |
477 | * the other side. | |
478 | * | |
479 | * Caller must ensure we don't call this with other virtqueue | |
480 | * operations at the same time (except where noted). | |
481 | */ | |
482 | void virtqueue_kick(struct virtqueue *vq) | |
483 | { | |
484 | if (virtqueue_kick_prepare(vq)) | |
485 | virtqueue_notify(vq); | |
0a8a69dd | 486 | } |
7c5e9ed0 | 487 | EXPORT_SYMBOL_GPL(virtqueue_kick); |
0a8a69dd RR |
488 | |
489 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head) | |
490 | { | |
491 | unsigned int i; | |
492 | ||
493 | /* Clear data ptr. */ | |
494 | vq->data[head] = NULL; | |
495 | ||
496 | /* Put back on free list: find end */ | |
497 | i = head; | |
9fa29b9d MM |
498 | |
499 | /* Free the indirect table */ | |
500 | if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT) | |
501 | kfree(phys_to_virt(vq->vring.desc[i].addr)); | |
502 | ||
0a8a69dd RR |
503 | while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) { |
504 | i = vq->vring.desc[i].next; | |
06ca287d | 505 | vq->vq.num_free++; |
0a8a69dd RR |
506 | } |
507 | ||
508 | vq->vring.desc[i].next = vq->free_head; | |
509 | vq->free_head = head; | |
510 | /* Plus final descriptor */ | |
06ca287d | 511 | vq->vq.num_free++; |
0a8a69dd RR |
512 | } |
513 | ||
0a8a69dd RR |
514 | static inline bool more_used(const struct vring_virtqueue *vq) |
515 | { | |
516 | return vq->last_used_idx != vq->vring.used->idx; | |
517 | } | |
518 | ||
5dfc1762 RR |
519 | /** |
520 | * virtqueue_get_buf - get the next used buffer | |
521 | * @vq: the struct virtqueue we're talking about. | |
522 | * @len: the length written into the buffer | |
523 | * | |
524 | * If the driver wrote data into the buffer, @len will be set to the | |
525 | * amount written. This means you don't need to clear the buffer | |
526 | * beforehand to ensure there's no data leakage in the case of short | |
527 | * writes. | |
528 | * | |
529 | * Caller must ensure we don't call this with other virtqueue | |
530 | * operations at the same time (except where noted). | |
531 | * | |
532 | * Returns NULL if there are no used buffers, or the "data" token | |
f96fde41 | 533 | * handed to virtqueue_add_buf(). |
5dfc1762 | 534 | */ |
7c5e9ed0 | 535 | void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len) |
0a8a69dd RR |
536 | { |
537 | struct vring_virtqueue *vq = to_vvq(_vq); | |
538 | void *ret; | |
539 | unsigned int i; | |
3b720b8c | 540 | u16 last_used; |
0a8a69dd RR |
541 | |
542 | START_USE(vq); | |
543 | ||
5ef82752 RR |
544 | if (unlikely(vq->broken)) { |
545 | END_USE(vq); | |
546 | return NULL; | |
547 | } | |
548 | ||
0a8a69dd RR |
549 | if (!more_used(vq)) { |
550 | pr_debug("No more buffers in queue\n"); | |
551 | END_USE(vq); | |
552 | return NULL; | |
553 | } | |
554 | ||
2d61ba95 | 555 | /* Only get used array entries after they have been exposed by host. */ |
a9a0fef7 | 556 | virtio_rmb(vq->weak_barriers); |
2d61ba95 | 557 | |
3b720b8c RR |
558 | last_used = (vq->last_used_idx & (vq->vring.num - 1)); |
559 | i = vq->vring.used->ring[last_used].id; | |
560 | *len = vq->vring.used->ring[last_used].len; | |
0a8a69dd RR |
561 | |
562 | if (unlikely(i >= vq->vring.num)) { | |
563 | BAD_RING(vq, "id %u out of range\n", i); | |
564 | return NULL; | |
565 | } | |
566 | if (unlikely(!vq->data[i])) { | |
567 | BAD_RING(vq, "id %u is not a head!\n", i); | |
568 | return NULL; | |
569 | } | |
570 | ||
571 | /* detach_buf clears data, so grab it now. */ | |
572 | ret = vq->data[i]; | |
573 | detach_buf(vq, i); | |
574 | vq->last_used_idx++; | |
a5c262c5 MT |
575 | /* If we expect an interrupt for the next entry, tell host |
576 | * by writing event index and flush out the write before | |
577 | * the read in the next get_buf call. */ | |
578 | if (!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) { | |
579 | vring_used_event(&vq->vring) = vq->last_used_idx; | |
a9a0fef7 | 580 | virtio_mb(vq->weak_barriers); |
a5c262c5 MT |
581 | } |
582 | ||
e93300b1 RR |
583 | #ifdef DEBUG |
584 | vq->last_add_time_valid = false; | |
585 | #endif | |
586 | ||
0a8a69dd RR |
587 | END_USE(vq); |
588 | return ret; | |
589 | } | |
7c5e9ed0 | 590 | EXPORT_SYMBOL_GPL(virtqueue_get_buf); |
0a8a69dd | 591 | |
5dfc1762 RR |
592 | /** |
593 | * virtqueue_disable_cb - disable callbacks | |
594 | * @vq: the struct virtqueue we're talking about. | |
595 | * | |
596 | * Note that this is not necessarily synchronous, hence unreliable and only | |
597 | * useful as an optimization. | |
598 | * | |
599 | * Unlike other operations, this need not be serialized. | |
600 | */ | |
7c5e9ed0 | 601 | void virtqueue_disable_cb(struct virtqueue *_vq) |
18445c4d RR |
602 | { |
603 | struct vring_virtqueue *vq = to_vvq(_vq); | |
604 | ||
18445c4d | 605 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; |
18445c4d | 606 | } |
7c5e9ed0 | 607 | EXPORT_SYMBOL_GPL(virtqueue_disable_cb); |
18445c4d | 608 | |
5dfc1762 | 609 | /** |
c23b1ece | 610 | * virtqueue_enable_cb_prepare - restart callbacks after disable_cb |
5dfc1762 RR |
611 | * @vq: the struct virtqueue we're talking about. |
612 | * | |
c23b1ece MT |
613 | * This re-enables callbacks; it returns current queue state |
614 | * in an opaque unsigned value. This value should be later tested by | |
615 | * virtqueue_poll, to detect a possible race between the driver checking for | |
616 | * more work, and enabling callbacks. | |
5dfc1762 RR |
617 | * |
618 | * Caller must ensure we don't call this with other virtqueue | |
619 | * operations at the same time (except where noted). | |
620 | */ | |
c23b1ece | 621 | unsigned virtqueue_enable_cb_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
622 | { |
623 | struct vring_virtqueue *vq = to_vvq(_vq); | |
c23b1ece | 624 | u16 last_used_idx; |
0a8a69dd RR |
625 | |
626 | START_USE(vq); | |
0a8a69dd RR |
627 | |
628 | /* We optimistically turn back on interrupts, then check if there was | |
629 | * more to do. */ | |
a5c262c5 MT |
630 | /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to |
631 | * either clear the flags bit or point the event index at the next | |
632 | * entry. Always do both to keep code simple. */ | |
0a8a69dd | 633 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; |
c23b1ece MT |
634 | vring_used_event(&vq->vring) = last_used_idx = vq->last_used_idx; |
635 | END_USE(vq); | |
636 | return last_used_idx; | |
637 | } | |
638 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_prepare); | |
639 | ||
640 | /** | |
641 | * virtqueue_poll - query pending used buffers | |
642 | * @vq: the struct virtqueue we're talking about. | |
643 | * @last_used_idx: virtqueue state (from call to virtqueue_enable_cb_prepare). | |
644 | * | |
645 | * Returns "true" if there are pending used buffers in the queue. | |
646 | * | |
647 | * This does not need to be serialized. | |
648 | */ | |
649 | bool virtqueue_poll(struct virtqueue *_vq, unsigned last_used_idx) | |
650 | { | |
651 | struct vring_virtqueue *vq = to_vvq(_vq); | |
652 | ||
a9a0fef7 | 653 | virtio_mb(vq->weak_barriers); |
c23b1ece MT |
654 | return (u16)last_used_idx != vq->vring.used->idx; |
655 | } | |
656 | EXPORT_SYMBOL_GPL(virtqueue_poll); | |
0a8a69dd | 657 | |
c23b1ece MT |
658 | /** |
659 | * virtqueue_enable_cb - restart callbacks after disable_cb. | |
660 | * @vq: the struct virtqueue we're talking about. | |
661 | * | |
662 | * This re-enables callbacks; it returns "false" if there are pending | |
663 | * buffers in the queue, to detect a possible race between the driver | |
664 | * checking for more work, and enabling callbacks. | |
665 | * | |
666 | * Caller must ensure we don't call this with other virtqueue | |
667 | * operations at the same time (except where noted). | |
668 | */ | |
669 | bool virtqueue_enable_cb(struct virtqueue *_vq) | |
670 | { | |
671 | unsigned last_used_idx = virtqueue_enable_cb_prepare(_vq); | |
672 | return !virtqueue_poll(_vq, last_used_idx); | |
0a8a69dd | 673 | } |
7c5e9ed0 | 674 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb); |
0a8a69dd | 675 | |
5dfc1762 RR |
676 | /** |
677 | * virtqueue_enable_cb_delayed - restart callbacks after disable_cb. | |
678 | * @vq: the struct virtqueue we're talking about. | |
679 | * | |
680 | * This re-enables callbacks but hints to the other side to delay | |
681 | * interrupts until most of the available buffers have been processed; | |
682 | * it returns "false" if there are many pending buffers in the queue, | |
683 | * to detect a possible race between the driver checking for more work, | |
684 | * and enabling callbacks. | |
685 | * | |
686 | * Caller must ensure we don't call this with other virtqueue | |
687 | * operations at the same time (except where noted). | |
688 | */ | |
7ab358c2 MT |
689 | bool virtqueue_enable_cb_delayed(struct virtqueue *_vq) |
690 | { | |
691 | struct vring_virtqueue *vq = to_vvq(_vq); | |
692 | u16 bufs; | |
693 | ||
694 | START_USE(vq); | |
695 | ||
696 | /* We optimistically turn back on interrupts, then check if there was | |
697 | * more to do. */ | |
698 | /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to | |
699 | * either clear the flags bit or point the event index at the next | |
700 | * entry. Always do both to keep code simple. */ | |
701 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
702 | /* TODO: tune this threshold */ | |
703 | bufs = (u16)(vq->vring.avail->idx - vq->last_used_idx) * 3 / 4; | |
704 | vring_used_event(&vq->vring) = vq->last_used_idx + bufs; | |
a9a0fef7 | 705 | virtio_mb(vq->weak_barriers); |
7ab358c2 MT |
706 | if (unlikely((u16)(vq->vring.used->idx - vq->last_used_idx) > bufs)) { |
707 | END_USE(vq); | |
708 | return false; | |
709 | } | |
710 | ||
711 | END_USE(vq); | |
712 | return true; | |
713 | } | |
714 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed); | |
715 | ||
5dfc1762 RR |
716 | /** |
717 | * virtqueue_detach_unused_buf - detach first unused buffer | |
718 | * @vq: the struct virtqueue we're talking about. | |
719 | * | |
f96fde41 | 720 | * Returns NULL or the "data" token handed to virtqueue_add_buf(). |
5dfc1762 RR |
721 | * This is not valid on an active queue; it is useful only for device |
722 | * shutdown. | |
723 | */ | |
7c5e9ed0 | 724 | void *virtqueue_detach_unused_buf(struct virtqueue *_vq) |
c021eac4 SM |
725 | { |
726 | struct vring_virtqueue *vq = to_vvq(_vq); | |
727 | unsigned int i; | |
728 | void *buf; | |
729 | ||
730 | START_USE(vq); | |
731 | ||
732 | for (i = 0; i < vq->vring.num; i++) { | |
733 | if (!vq->data[i]) | |
734 | continue; | |
735 | /* detach_buf clears data, so grab it now. */ | |
736 | buf = vq->data[i]; | |
737 | detach_buf(vq, i); | |
b3258ff1 | 738 | vq->vring.avail->idx--; |
c021eac4 SM |
739 | END_USE(vq); |
740 | return buf; | |
741 | } | |
742 | /* That should have freed everything. */ | |
06ca287d | 743 | BUG_ON(vq->vq.num_free != vq->vring.num); |
c021eac4 SM |
744 | |
745 | END_USE(vq); | |
746 | return NULL; | |
747 | } | |
7c5e9ed0 | 748 | EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf); |
c021eac4 | 749 | |
0a8a69dd RR |
750 | irqreturn_t vring_interrupt(int irq, void *_vq) |
751 | { | |
752 | struct vring_virtqueue *vq = to_vvq(_vq); | |
753 | ||
754 | if (!more_used(vq)) { | |
755 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
756 | return IRQ_NONE; | |
757 | } | |
758 | ||
759 | if (unlikely(vq->broken)) | |
760 | return IRQ_HANDLED; | |
761 | ||
762 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); | |
18445c4d RR |
763 | if (vq->vq.callback) |
764 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
765 | |
766 | return IRQ_HANDLED; | |
767 | } | |
c6fd4701 | 768 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd | 769 | |
17bb6d40 JW |
770 | struct virtqueue *vring_new_virtqueue(unsigned int index, |
771 | unsigned int num, | |
87c7d57c | 772 | unsigned int vring_align, |
0a8a69dd | 773 | struct virtio_device *vdev, |
7b21e34f | 774 | bool weak_barriers, |
0a8a69dd RR |
775 | void *pages, |
776 | void (*notify)(struct virtqueue *), | |
9499f5e7 RR |
777 | void (*callback)(struct virtqueue *), |
778 | const char *name) | |
0a8a69dd RR |
779 | { |
780 | struct vring_virtqueue *vq; | |
781 | unsigned int i; | |
782 | ||
42b36cc0 RR |
783 | /* We assume num is a power of 2. */ |
784 | if (num & (num - 1)) { | |
785 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
786 | return NULL; | |
787 | } | |
788 | ||
0a8a69dd RR |
789 | vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL); |
790 | if (!vq) | |
791 | return NULL; | |
792 | ||
87c7d57c | 793 | vring_init(&vq->vring, num, pages, vring_align); |
0a8a69dd RR |
794 | vq->vq.callback = callback; |
795 | vq->vq.vdev = vdev; | |
9499f5e7 | 796 | vq->vq.name = name; |
06ca287d RR |
797 | vq->vq.num_free = num; |
798 | vq->vq.index = index; | |
0a8a69dd | 799 | vq->notify = notify; |
7b21e34f | 800 | vq->weak_barriers = weak_barriers; |
0a8a69dd RR |
801 | vq->broken = false; |
802 | vq->last_used_idx = 0; | |
803 | vq->num_added = 0; | |
9499f5e7 | 804 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0a8a69dd RR |
805 | #ifdef DEBUG |
806 | vq->in_use = false; | |
e93300b1 | 807 | vq->last_add_time_valid = false; |
0a8a69dd RR |
808 | #endif |
809 | ||
9fa29b9d | 810 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC); |
a5c262c5 | 811 | vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); |
9fa29b9d | 812 | |
0a8a69dd RR |
813 | /* No callback? Tell other side not to bother us. */ |
814 | if (!callback) | |
815 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; | |
816 | ||
817 | /* Put everything in free lists. */ | |
0a8a69dd | 818 | vq->free_head = 0; |
3b870624 | 819 | for (i = 0; i < num-1; i++) { |
0a8a69dd | 820 | vq->vring.desc[i].next = i+1; |
3b870624 AS |
821 | vq->data[i] = NULL; |
822 | } | |
823 | vq->data[i] = NULL; | |
0a8a69dd RR |
824 | |
825 | return &vq->vq; | |
826 | } | |
c6fd4701 | 827 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd RR |
828 | |
829 | void vring_del_virtqueue(struct virtqueue *vq) | |
830 | { | |
9499f5e7 | 831 | list_del(&vq->list); |
0a8a69dd RR |
832 | kfree(to_vvq(vq)); |
833 | } | |
c6fd4701 | 834 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 835 | |
e34f8725 RR |
836 | /* Manipulates transport-specific feature bits. */ |
837 | void vring_transport_features(struct virtio_device *vdev) | |
838 | { | |
839 | unsigned int i; | |
840 | ||
841 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
842 | switch (i) { | |
9fa29b9d MM |
843 | case VIRTIO_RING_F_INDIRECT_DESC: |
844 | break; | |
a5c262c5 MT |
845 | case VIRTIO_RING_F_EVENT_IDX: |
846 | break; | |
e34f8725 RR |
847 | default: |
848 | /* We don't understand this bit. */ | |
849 | clear_bit(i, vdev->features); | |
850 | } | |
851 | } | |
852 | } | |
853 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
854 | ||
5dfc1762 RR |
855 | /** |
856 | * virtqueue_get_vring_size - return the size of the virtqueue's vring | |
857 | * @vq: the struct virtqueue containing the vring of interest. | |
858 | * | |
859 | * Returns the size of the vring. This is mainly used for boasting to | |
860 | * userspace. Unlike other operations, this need not be serialized. | |
861 | */ | |
8f9f4668 RJ |
862 | unsigned int virtqueue_get_vring_size(struct virtqueue *_vq) |
863 | { | |
864 | ||
865 | struct vring_virtqueue *vq = to_vvq(_vq); | |
866 | ||
867 | return vq->vring.num; | |
868 | } | |
869 | EXPORT_SYMBOL_GPL(virtqueue_get_vring_size); | |
870 | ||
c6fd4701 | 871 | MODULE_LICENSE("GPL"); |