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 | ||
13816c76 RR |
102 | static inline struct scatterlist *sg_next_chained(struct scatterlist *sg, |
103 | unsigned int *count) | |
104 | { | |
105 | return sg_next(sg); | |
106 | } | |
107 | ||
108 | static inline struct scatterlist *sg_next_arr(struct scatterlist *sg, | |
109 | unsigned int *count) | |
110 | { | |
111 | if (--(*count) == 0) | |
112 | return NULL; | |
113 | return sg + 1; | |
114 | } | |
115 | ||
9fa29b9d | 116 | /* Set up an indirect table of descriptors and add it to the queue. */ |
13816c76 RR |
117 | static inline int vring_add_indirect(struct vring_virtqueue *vq, |
118 | struct scatterlist *sgs[], | |
119 | struct scatterlist *(*next) | |
120 | (struct scatterlist *, unsigned int *), | |
121 | unsigned int total_sg, | |
122 | unsigned int total_out, | |
123 | unsigned int total_in, | |
124 | unsigned int out_sgs, | |
125 | unsigned int in_sgs, | |
126 | gfp_t gfp) | |
9fa29b9d MM |
127 | { |
128 | struct vring_desc *desc; | |
129 | unsigned head; | |
13816c76 RR |
130 | struct scatterlist *sg; |
131 | int i, n; | |
9fa29b9d | 132 | |
b92b1b89 WD |
133 | /* |
134 | * We require lowmem mappings for the descriptors because | |
135 | * otherwise virt_to_phys will give us bogus addresses in the | |
136 | * virtqueue. | |
137 | */ | |
138 | gfp &= ~(__GFP_HIGHMEM | __GFP_HIGH); | |
139 | ||
13816c76 | 140 | desc = kmalloc(total_sg * sizeof(struct vring_desc), gfp); |
9fa29b9d | 141 | if (!desc) |
686d3637 | 142 | return -ENOMEM; |
9fa29b9d | 143 | |
13816c76 RR |
144 | /* Transfer entries from the sg lists into the indirect page */ |
145 | i = 0; | |
146 | for (n = 0; n < out_sgs; n++) { | |
147 | for (sg = sgs[n]; sg; sg = next(sg, &total_out)) { | |
148 | desc[i].flags = VRING_DESC_F_NEXT; | |
149 | desc[i].addr = sg_phys(sg); | |
150 | desc[i].len = sg->length; | |
151 | desc[i].next = i+1; | |
152 | i++; | |
153 | } | |
9fa29b9d | 154 | } |
13816c76 RR |
155 | for (; n < (out_sgs + in_sgs); n++) { |
156 | for (sg = sgs[n]; sg; sg = next(sg, &total_in)) { | |
157 | desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
158 | desc[i].addr = sg_phys(sg); | |
159 | desc[i].len = sg->length; | |
160 | desc[i].next = i+1; | |
161 | i++; | |
162 | } | |
9fa29b9d | 163 | } |
13816c76 | 164 | BUG_ON(i != total_sg); |
9fa29b9d MM |
165 | |
166 | /* Last one doesn't continue. */ | |
167 | desc[i-1].flags &= ~VRING_DESC_F_NEXT; | |
168 | desc[i-1].next = 0; | |
169 | ||
170 | /* We're about to use a buffer */ | |
06ca287d | 171 | vq->vq.num_free--; |
9fa29b9d MM |
172 | |
173 | /* Use a single buffer which doesn't continue */ | |
174 | head = vq->free_head; | |
175 | vq->vring.desc[head].flags = VRING_DESC_F_INDIRECT; | |
176 | vq->vring.desc[head].addr = virt_to_phys(desc); | |
bb478d8b RR |
177 | /* kmemleak gives a false positive, as it's hidden by virt_to_phys */ |
178 | kmemleak_ignore(desc); | |
9fa29b9d MM |
179 | vq->vring.desc[head].len = i * sizeof(struct vring_desc); |
180 | ||
181 | /* Update free pointer */ | |
182 | vq->free_head = vq->vring.desc[head].next; | |
183 | ||
184 | return head; | |
185 | } | |
186 | ||
13816c76 RR |
187 | static inline int virtqueue_add(struct virtqueue *_vq, |
188 | struct scatterlist *sgs[], | |
189 | struct scatterlist *(*next) | |
190 | (struct scatterlist *, unsigned int *), | |
191 | unsigned int total_out, | |
192 | unsigned int total_in, | |
193 | unsigned int out_sgs, | |
194 | unsigned int in_sgs, | |
195 | void *data, | |
196 | gfp_t gfp) | |
0a8a69dd RR |
197 | { |
198 | struct vring_virtqueue *vq = to_vvq(_vq); | |
13816c76 RR |
199 | struct scatterlist *sg; |
200 | unsigned int i, n, avail, uninitialized_var(prev), total_sg; | |
1fe9b6fe | 201 | int head; |
0a8a69dd | 202 | |
9fa29b9d MM |
203 | START_USE(vq); |
204 | ||
0a8a69dd | 205 | BUG_ON(data == NULL); |
9fa29b9d | 206 | |
e93300b1 RR |
207 | #ifdef DEBUG |
208 | { | |
209 | ktime_t now = ktime_get(); | |
210 | ||
211 | /* No kick or get, with .1 second between? Warn. */ | |
212 | if (vq->last_add_time_valid) | |
213 | WARN_ON(ktime_to_ms(ktime_sub(now, vq->last_add_time)) | |
214 | > 100); | |
215 | vq->last_add_time = now; | |
216 | vq->last_add_time_valid = true; | |
217 | } | |
218 | #endif | |
219 | ||
13816c76 RR |
220 | total_sg = total_in + total_out; |
221 | ||
9fa29b9d MM |
222 | /* If the host supports indirect descriptor tables, and we have multiple |
223 | * buffers, then go indirect. FIXME: tune this threshold */ | |
13816c76 RR |
224 | if (vq->indirect && total_sg > 1 && vq->vq.num_free) { |
225 | head = vring_add_indirect(vq, sgs, next, total_sg, total_out, | |
226 | total_in, | |
227 | out_sgs, in_sgs, gfp); | |
1fe9b6fe | 228 | if (likely(head >= 0)) |
9fa29b9d MM |
229 | goto add_head; |
230 | } | |
231 | ||
13816c76 RR |
232 | BUG_ON(total_sg > vq->vring.num); |
233 | BUG_ON(total_sg == 0); | |
0a8a69dd | 234 | |
13816c76 | 235 | if (vq->vq.num_free < total_sg) { |
0a8a69dd | 236 | pr_debug("Can't add buf len %i - avail = %i\n", |
13816c76 | 237 | total_sg, vq->vq.num_free); |
44653eae RR |
238 | /* FIXME: for historical reasons, we force a notify here if |
239 | * there are outgoing parts to the buffer. Presumably the | |
240 | * host should service the ring ASAP. */ | |
13816c76 | 241 | if (out_sgs) |
44653eae | 242 | vq->notify(&vq->vq); |
0a8a69dd RR |
243 | END_USE(vq); |
244 | return -ENOSPC; | |
245 | } | |
246 | ||
247 | /* We're about to use some buffers from the free list. */ | |
13816c76 RR |
248 | vq->vq.num_free -= total_sg; |
249 | ||
250 | head = i = vq->free_head; | |
251 | for (n = 0; n < out_sgs; n++) { | |
252 | for (sg = sgs[n]; sg; sg = next(sg, &total_out)) { | |
253 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT; | |
254 | vq->vring.desc[i].addr = sg_phys(sg); | |
255 | vq->vring.desc[i].len = sg->length; | |
256 | prev = i; | |
257 | i = vq->vring.desc[i].next; | |
258 | } | |
0a8a69dd | 259 | } |
13816c76 RR |
260 | for (; n < (out_sgs + in_sgs); n++) { |
261 | for (sg = sgs[n]; sg; sg = next(sg, &total_in)) { | |
262 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
263 | vq->vring.desc[i].addr = sg_phys(sg); | |
264 | vq->vring.desc[i].len = sg->length; | |
265 | prev = i; | |
266 | i = vq->vring.desc[i].next; | |
267 | } | |
0a8a69dd RR |
268 | } |
269 | /* Last one doesn't continue. */ | |
270 | vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT; | |
271 | ||
272 | /* Update free pointer */ | |
273 | vq->free_head = i; | |
274 | ||
9fa29b9d | 275 | add_head: |
0a8a69dd RR |
276 | /* Set token. */ |
277 | vq->data[head] = data; | |
278 | ||
279 | /* Put entry in available array (but don't update avail->idx until they | |
3b720b8c | 280 | * do sync). */ |
ee7cd898 | 281 | avail = (vq->vring.avail->idx & (vq->vring.num-1)); |
0a8a69dd RR |
282 | vq->vring.avail->ring[avail] = head; |
283 | ||
ee7cd898 RR |
284 | /* Descriptors and available array need to be set before we expose the |
285 | * new available array entries. */ | |
a9a0fef7 | 286 | virtio_wmb(vq->weak_barriers); |
ee7cd898 RR |
287 | vq->vring.avail->idx++; |
288 | vq->num_added++; | |
289 | ||
290 | /* This is very unlikely, but theoretically possible. Kick | |
291 | * just in case. */ | |
292 | if (unlikely(vq->num_added == (1 << 16) - 1)) | |
293 | virtqueue_kick(_vq); | |
294 | ||
0a8a69dd RR |
295 | pr_debug("Added buffer head %i to %p\n", head, vq); |
296 | END_USE(vq); | |
3c1b27d5 | 297 | |
98e8c6bc | 298 | return 0; |
0a8a69dd | 299 | } |
13816c76 | 300 | |
13816c76 RR |
301 | /** |
302 | * virtqueue_add_sgs - expose buffers to other end | |
303 | * @vq: the struct virtqueue we're talking about. | |
304 | * @sgs: array of terminated scatterlists. | |
305 | * @out_num: the number of scatterlists readable by other side | |
306 | * @in_num: the number of scatterlists which are writable (after readable ones) | |
307 | * @data: the token identifying the buffer. | |
308 | * @gfp: how to do memory allocations (if necessary). | |
309 | * | |
310 | * Caller must ensure we don't call this with other virtqueue operations | |
311 | * at the same time (except where noted). | |
312 | * | |
313 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM). | |
314 | */ | |
315 | int virtqueue_add_sgs(struct virtqueue *_vq, | |
316 | struct scatterlist *sgs[], | |
317 | unsigned int out_sgs, | |
318 | unsigned int in_sgs, | |
319 | void *data, | |
320 | gfp_t gfp) | |
321 | { | |
322 | unsigned int i, total_out, total_in; | |
323 | ||
324 | /* Count them first. */ | |
325 | for (i = total_out = total_in = 0; i < out_sgs; i++) { | |
326 | struct scatterlist *sg; | |
327 | for (sg = sgs[i]; sg; sg = sg_next(sg)) | |
328 | total_out++; | |
329 | } | |
330 | for (; i < out_sgs + in_sgs; i++) { | |
331 | struct scatterlist *sg; | |
332 | for (sg = sgs[i]; sg; sg = sg_next(sg)) | |
333 | total_in++; | |
334 | } | |
335 | return virtqueue_add(_vq, sgs, sg_next_chained, | |
336 | total_out, total_in, out_sgs, in_sgs, data, gfp); | |
337 | } | |
338 | EXPORT_SYMBOL_GPL(virtqueue_add_sgs); | |
339 | ||
282edb36 RR |
340 | /** |
341 | * virtqueue_add_outbuf - expose output buffers to other end | |
342 | * @vq: the struct virtqueue we're talking about. | |
343 | * @sgs: array of scatterlists (need not be terminated!) | |
344 | * @num: the number of scatterlists readable by other side | |
345 | * @data: the token identifying the buffer. | |
346 | * @gfp: how to do memory allocations (if necessary). | |
347 | * | |
348 | * Caller must ensure we don't call this with other virtqueue operations | |
349 | * at the same time (except where noted). | |
350 | * | |
351 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM). | |
352 | */ | |
353 | int virtqueue_add_outbuf(struct virtqueue *vq, | |
354 | struct scatterlist sg[], unsigned int num, | |
355 | void *data, | |
356 | gfp_t gfp) | |
357 | { | |
358 | return virtqueue_add(vq, &sg, sg_next_arr, num, 0, 1, 0, data, gfp); | |
359 | } | |
360 | EXPORT_SYMBOL_GPL(virtqueue_add_outbuf); | |
361 | ||
362 | /** | |
363 | * virtqueue_add_inbuf - expose input buffers to other end | |
364 | * @vq: the struct virtqueue we're talking about. | |
365 | * @sgs: array of scatterlists (need not be terminated!) | |
366 | * @num: the number of scatterlists writable by other side | |
367 | * @data: the token identifying the buffer. | |
368 | * @gfp: how to do memory allocations (if necessary). | |
369 | * | |
370 | * Caller must ensure we don't call this with other virtqueue operations | |
371 | * at the same time (except where noted). | |
372 | * | |
373 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM). | |
374 | */ | |
375 | int virtqueue_add_inbuf(struct virtqueue *vq, | |
376 | struct scatterlist sg[], unsigned int num, | |
377 | void *data, | |
378 | gfp_t gfp) | |
379 | { | |
380 | return virtqueue_add(vq, &sg, sg_next_arr, 0, num, 0, 1, data, gfp); | |
381 | } | |
382 | EXPORT_SYMBOL_GPL(virtqueue_add_inbuf); | |
383 | ||
5dfc1762 | 384 | /** |
41f0377f | 385 | * virtqueue_kick_prepare - first half of split virtqueue_kick call. |
5dfc1762 RR |
386 | * @vq: the struct virtqueue |
387 | * | |
41f0377f RR |
388 | * Instead of virtqueue_kick(), you can do: |
389 | * if (virtqueue_kick_prepare(vq)) | |
390 | * virtqueue_notify(vq); | |
5dfc1762 | 391 | * |
41f0377f RR |
392 | * This is sometimes useful because the virtqueue_kick_prepare() needs |
393 | * to be serialized, but the actual virtqueue_notify() call does not. | |
5dfc1762 | 394 | */ |
41f0377f | 395 | bool virtqueue_kick_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
396 | { |
397 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a5c262c5 | 398 | u16 new, old; |
41f0377f RR |
399 | bool needs_kick; |
400 | ||
0a8a69dd | 401 | START_USE(vq); |
a72caae2 JW |
402 | /* We need to expose available array entries before checking avail |
403 | * event. */ | |
a9a0fef7 | 404 | virtio_mb(vq->weak_barriers); |
0a8a69dd | 405 | |
ee7cd898 RR |
406 | old = vq->vring.avail->idx - vq->num_added; |
407 | new = vq->vring.avail->idx; | |
0a8a69dd RR |
408 | vq->num_added = 0; |
409 | ||
e93300b1 RR |
410 | #ifdef DEBUG |
411 | if (vq->last_add_time_valid) { | |
412 | WARN_ON(ktime_to_ms(ktime_sub(ktime_get(), | |
413 | vq->last_add_time)) > 100); | |
414 | } | |
415 | vq->last_add_time_valid = false; | |
416 | #endif | |
417 | ||
41f0377f RR |
418 | if (vq->event) { |
419 | needs_kick = vring_need_event(vring_avail_event(&vq->vring), | |
420 | new, old); | |
421 | } else { | |
422 | needs_kick = !(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY); | |
423 | } | |
0a8a69dd | 424 | END_USE(vq); |
41f0377f RR |
425 | return needs_kick; |
426 | } | |
427 | EXPORT_SYMBOL_GPL(virtqueue_kick_prepare); | |
428 | ||
429 | /** | |
430 | * virtqueue_notify - second half of split virtqueue_kick call. | |
431 | * @vq: the struct virtqueue | |
432 | * | |
433 | * This does not need to be serialized. | |
5b1bf7cb HG |
434 | * |
435 | * Returns false if host notify failed or queue is broken, otherwise true. | |
41f0377f | 436 | */ |
5b1bf7cb | 437 | bool virtqueue_notify(struct virtqueue *_vq) |
41f0377f RR |
438 | { |
439 | struct vring_virtqueue *vq = to_vvq(_vq); | |
440 | ||
5b1bf7cb HG |
441 | if (unlikely(vq->broken)) |
442 | return false; | |
443 | ||
41f0377f | 444 | /* Prod other side to tell it about changes. */ |
2342d6a6 | 445 | if (!vq->notify(_vq)) { |
5b1bf7cb HG |
446 | vq->broken = true; |
447 | return false; | |
448 | } | |
449 | return true; | |
41f0377f RR |
450 | } |
451 | EXPORT_SYMBOL_GPL(virtqueue_notify); | |
452 | ||
453 | /** | |
454 | * virtqueue_kick - update after add_buf | |
455 | * @vq: the struct virtqueue | |
456 | * | |
b3087e48 | 457 | * After one or more virtqueue_add_* calls, invoke this to kick |
41f0377f RR |
458 | * the other side. |
459 | * | |
460 | * Caller must ensure we don't call this with other virtqueue | |
461 | * operations at the same time (except where noted). | |
5b1bf7cb HG |
462 | * |
463 | * Returns false if kick failed, otherwise true. | |
41f0377f | 464 | */ |
5b1bf7cb | 465 | bool virtqueue_kick(struct virtqueue *vq) |
41f0377f RR |
466 | { |
467 | if (virtqueue_kick_prepare(vq)) | |
5b1bf7cb HG |
468 | return virtqueue_notify(vq); |
469 | return true; | |
0a8a69dd | 470 | } |
7c5e9ed0 | 471 | EXPORT_SYMBOL_GPL(virtqueue_kick); |
0a8a69dd RR |
472 | |
473 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head) | |
474 | { | |
475 | unsigned int i; | |
476 | ||
477 | /* Clear data ptr. */ | |
478 | vq->data[head] = NULL; | |
479 | ||
480 | /* Put back on free list: find end */ | |
481 | i = head; | |
9fa29b9d MM |
482 | |
483 | /* Free the indirect table */ | |
484 | if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT) | |
485 | kfree(phys_to_virt(vq->vring.desc[i].addr)); | |
486 | ||
0a8a69dd RR |
487 | while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) { |
488 | i = vq->vring.desc[i].next; | |
06ca287d | 489 | vq->vq.num_free++; |
0a8a69dd RR |
490 | } |
491 | ||
492 | vq->vring.desc[i].next = vq->free_head; | |
493 | vq->free_head = head; | |
494 | /* Plus final descriptor */ | |
06ca287d | 495 | vq->vq.num_free++; |
0a8a69dd RR |
496 | } |
497 | ||
0a8a69dd RR |
498 | static inline bool more_used(const struct vring_virtqueue *vq) |
499 | { | |
500 | return vq->last_used_idx != vq->vring.used->idx; | |
501 | } | |
502 | ||
5dfc1762 RR |
503 | /** |
504 | * virtqueue_get_buf - get the next used buffer | |
505 | * @vq: the struct virtqueue we're talking about. | |
506 | * @len: the length written into the buffer | |
507 | * | |
508 | * If the driver wrote data into the buffer, @len will be set to the | |
509 | * amount written. This means you don't need to clear the buffer | |
510 | * beforehand to ensure there's no data leakage in the case of short | |
511 | * writes. | |
512 | * | |
513 | * Caller must ensure we don't call this with other virtqueue | |
514 | * operations at the same time (except where noted). | |
515 | * | |
516 | * Returns NULL if there are no used buffers, or the "data" token | |
b3087e48 | 517 | * handed to virtqueue_add_*(). |
5dfc1762 | 518 | */ |
7c5e9ed0 | 519 | void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len) |
0a8a69dd RR |
520 | { |
521 | struct vring_virtqueue *vq = to_vvq(_vq); | |
522 | void *ret; | |
523 | unsigned int i; | |
3b720b8c | 524 | u16 last_used; |
0a8a69dd RR |
525 | |
526 | START_USE(vq); | |
527 | ||
5ef82752 RR |
528 | if (unlikely(vq->broken)) { |
529 | END_USE(vq); | |
530 | return NULL; | |
531 | } | |
532 | ||
0a8a69dd RR |
533 | if (!more_used(vq)) { |
534 | pr_debug("No more buffers in queue\n"); | |
535 | END_USE(vq); | |
536 | return NULL; | |
537 | } | |
538 | ||
2d61ba95 | 539 | /* Only get used array entries after they have been exposed by host. */ |
a9a0fef7 | 540 | virtio_rmb(vq->weak_barriers); |
2d61ba95 | 541 | |
3b720b8c RR |
542 | last_used = (vq->last_used_idx & (vq->vring.num - 1)); |
543 | i = vq->vring.used->ring[last_used].id; | |
544 | *len = vq->vring.used->ring[last_used].len; | |
0a8a69dd RR |
545 | |
546 | if (unlikely(i >= vq->vring.num)) { | |
547 | BAD_RING(vq, "id %u out of range\n", i); | |
548 | return NULL; | |
549 | } | |
550 | if (unlikely(!vq->data[i])) { | |
551 | BAD_RING(vq, "id %u is not a head!\n", i); | |
552 | return NULL; | |
553 | } | |
554 | ||
555 | /* detach_buf clears data, so grab it now. */ | |
556 | ret = vq->data[i]; | |
557 | detach_buf(vq, i); | |
558 | vq->last_used_idx++; | |
a5c262c5 MT |
559 | /* If we expect an interrupt for the next entry, tell host |
560 | * by writing event index and flush out the write before | |
561 | * the read in the next get_buf call. */ | |
562 | if (!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) { | |
563 | vring_used_event(&vq->vring) = vq->last_used_idx; | |
a9a0fef7 | 564 | virtio_mb(vq->weak_barriers); |
a5c262c5 MT |
565 | } |
566 | ||
e93300b1 RR |
567 | #ifdef DEBUG |
568 | vq->last_add_time_valid = false; | |
569 | #endif | |
570 | ||
0a8a69dd RR |
571 | END_USE(vq); |
572 | return ret; | |
573 | } | |
7c5e9ed0 | 574 | EXPORT_SYMBOL_GPL(virtqueue_get_buf); |
0a8a69dd | 575 | |
5dfc1762 RR |
576 | /** |
577 | * virtqueue_disable_cb - disable callbacks | |
578 | * @vq: the struct virtqueue we're talking about. | |
579 | * | |
580 | * Note that this is not necessarily synchronous, hence unreliable and only | |
581 | * useful as an optimization. | |
582 | * | |
583 | * Unlike other operations, this need not be serialized. | |
584 | */ | |
7c5e9ed0 | 585 | void virtqueue_disable_cb(struct virtqueue *_vq) |
18445c4d RR |
586 | { |
587 | struct vring_virtqueue *vq = to_vvq(_vq); | |
588 | ||
18445c4d | 589 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; |
18445c4d | 590 | } |
7c5e9ed0 | 591 | EXPORT_SYMBOL_GPL(virtqueue_disable_cb); |
18445c4d | 592 | |
5dfc1762 | 593 | /** |
cc229884 | 594 | * virtqueue_enable_cb_prepare - restart callbacks after disable_cb |
5dfc1762 RR |
595 | * @vq: the struct virtqueue we're talking about. |
596 | * | |
cc229884 MT |
597 | * This re-enables callbacks; it returns current queue state |
598 | * in an opaque unsigned value. This value should be later tested by | |
599 | * virtqueue_poll, to detect a possible race between the driver checking for | |
600 | * more work, and enabling callbacks. | |
5dfc1762 RR |
601 | * |
602 | * Caller must ensure we don't call this with other virtqueue | |
603 | * operations at the same time (except where noted). | |
604 | */ | |
cc229884 | 605 | unsigned virtqueue_enable_cb_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
606 | { |
607 | struct vring_virtqueue *vq = to_vvq(_vq); | |
cc229884 | 608 | u16 last_used_idx; |
0a8a69dd RR |
609 | |
610 | START_USE(vq); | |
0a8a69dd RR |
611 | |
612 | /* We optimistically turn back on interrupts, then check if there was | |
613 | * more to do. */ | |
a5c262c5 MT |
614 | /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to |
615 | * either clear the flags bit or point the event index at the next | |
616 | * entry. Always do both to keep code simple. */ | |
0a8a69dd | 617 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; |
cc229884 MT |
618 | vring_used_event(&vq->vring) = last_used_idx = vq->last_used_idx; |
619 | END_USE(vq); | |
620 | return last_used_idx; | |
621 | } | |
622 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_prepare); | |
623 | ||
624 | /** | |
625 | * virtqueue_poll - query pending used buffers | |
626 | * @vq: the struct virtqueue we're talking about. | |
627 | * @last_used_idx: virtqueue state (from call to virtqueue_enable_cb_prepare). | |
628 | * | |
629 | * Returns "true" if there are pending used buffers in the queue. | |
630 | * | |
631 | * This does not need to be serialized. | |
632 | */ | |
633 | bool virtqueue_poll(struct virtqueue *_vq, unsigned last_used_idx) | |
634 | { | |
635 | struct vring_virtqueue *vq = to_vvq(_vq); | |
636 | ||
a9a0fef7 | 637 | virtio_mb(vq->weak_barriers); |
cc229884 MT |
638 | return (u16)last_used_idx != vq->vring.used->idx; |
639 | } | |
640 | EXPORT_SYMBOL_GPL(virtqueue_poll); | |
0a8a69dd | 641 | |
cc229884 MT |
642 | /** |
643 | * virtqueue_enable_cb - restart callbacks after disable_cb. | |
644 | * @vq: the struct virtqueue we're talking about. | |
645 | * | |
646 | * This re-enables callbacks; it returns "false" if there are pending | |
647 | * buffers in the queue, to detect a possible race between the driver | |
648 | * checking for more work, and enabling callbacks. | |
649 | * | |
650 | * Caller must ensure we don't call this with other virtqueue | |
651 | * operations at the same time (except where noted). | |
652 | */ | |
653 | bool virtqueue_enable_cb(struct virtqueue *_vq) | |
654 | { | |
655 | unsigned last_used_idx = virtqueue_enable_cb_prepare(_vq); | |
656 | return !virtqueue_poll(_vq, last_used_idx); | |
0a8a69dd | 657 | } |
7c5e9ed0 | 658 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb); |
0a8a69dd | 659 | |
5dfc1762 RR |
660 | /** |
661 | * virtqueue_enable_cb_delayed - restart callbacks after disable_cb. | |
662 | * @vq: the struct virtqueue we're talking about. | |
663 | * | |
664 | * This re-enables callbacks but hints to the other side to delay | |
665 | * interrupts until most of the available buffers have been processed; | |
666 | * it returns "false" if there are many pending buffers in the queue, | |
667 | * to detect a possible race between the driver checking for more work, | |
668 | * and enabling callbacks. | |
669 | * | |
670 | * Caller must ensure we don't call this with other virtqueue | |
671 | * operations at the same time (except where noted). | |
672 | */ | |
7ab358c2 MT |
673 | bool virtqueue_enable_cb_delayed(struct virtqueue *_vq) |
674 | { | |
675 | struct vring_virtqueue *vq = to_vvq(_vq); | |
676 | u16 bufs; | |
677 | ||
678 | START_USE(vq); | |
679 | ||
680 | /* We optimistically turn back on interrupts, then check if there was | |
681 | * more to do. */ | |
682 | /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to | |
683 | * either clear the flags bit or point the event index at the next | |
684 | * entry. Always do both to keep code simple. */ | |
685 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
686 | /* TODO: tune this threshold */ | |
687 | bufs = (u16)(vq->vring.avail->idx - vq->last_used_idx) * 3 / 4; | |
688 | vring_used_event(&vq->vring) = vq->last_used_idx + bufs; | |
a9a0fef7 | 689 | virtio_mb(vq->weak_barriers); |
7ab358c2 MT |
690 | if (unlikely((u16)(vq->vring.used->idx - vq->last_used_idx) > bufs)) { |
691 | END_USE(vq); | |
692 | return false; | |
693 | } | |
694 | ||
695 | END_USE(vq); | |
696 | return true; | |
697 | } | |
698 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed); | |
699 | ||
5dfc1762 RR |
700 | /** |
701 | * virtqueue_detach_unused_buf - detach first unused buffer | |
702 | * @vq: the struct virtqueue we're talking about. | |
703 | * | |
b3087e48 | 704 | * Returns NULL or the "data" token handed to virtqueue_add_*(). |
5dfc1762 RR |
705 | * This is not valid on an active queue; it is useful only for device |
706 | * shutdown. | |
707 | */ | |
7c5e9ed0 | 708 | void *virtqueue_detach_unused_buf(struct virtqueue *_vq) |
c021eac4 SM |
709 | { |
710 | struct vring_virtqueue *vq = to_vvq(_vq); | |
711 | unsigned int i; | |
712 | void *buf; | |
713 | ||
714 | START_USE(vq); | |
715 | ||
716 | for (i = 0; i < vq->vring.num; i++) { | |
717 | if (!vq->data[i]) | |
718 | continue; | |
719 | /* detach_buf clears data, so grab it now. */ | |
720 | buf = vq->data[i]; | |
721 | detach_buf(vq, i); | |
b3258ff1 | 722 | vq->vring.avail->idx--; |
c021eac4 SM |
723 | END_USE(vq); |
724 | return buf; | |
725 | } | |
726 | /* That should have freed everything. */ | |
06ca287d | 727 | BUG_ON(vq->vq.num_free != vq->vring.num); |
c021eac4 SM |
728 | |
729 | END_USE(vq); | |
730 | return NULL; | |
731 | } | |
7c5e9ed0 | 732 | EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf); |
c021eac4 | 733 | |
0a8a69dd RR |
734 | irqreturn_t vring_interrupt(int irq, void *_vq) |
735 | { | |
736 | struct vring_virtqueue *vq = to_vvq(_vq); | |
737 | ||
738 | if (!more_used(vq)) { | |
739 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
740 | return IRQ_NONE; | |
741 | } | |
742 | ||
743 | if (unlikely(vq->broken)) | |
744 | return IRQ_HANDLED; | |
745 | ||
746 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); | |
18445c4d RR |
747 | if (vq->vq.callback) |
748 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
749 | |
750 | return IRQ_HANDLED; | |
751 | } | |
c6fd4701 | 752 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd | 753 | |
17bb6d40 JW |
754 | struct virtqueue *vring_new_virtqueue(unsigned int index, |
755 | unsigned int num, | |
87c7d57c | 756 | unsigned int vring_align, |
0a8a69dd | 757 | struct virtio_device *vdev, |
7b21e34f | 758 | bool weak_barriers, |
0a8a69dd | 759 | void *pages, |
46f9c2b9 | 760 | bool (*notify)(struct virtqueue *), |
9499f5e7 RR |
761 | void (*callback)(struct virtqueue *), |
762 | const char *name) | |
0a8a69dd RR |
763 | { |
764 | struct vring_virtqueue *vq; | |
765 | unsigned int i; | |
766 | ||
42b36cc0 RR |
767 | /* We assume num is a power of 2. */ |
768 | if (num & (num - 1)) { | |
769 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
770 | return NULL; | |
771 | } | |
772 | ||
0a8a69dd RR |
773 | vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL); |
774 | if (!vq) | |
775 | return NULL; | |
776 | ||
87c7d57c | 777 | vring_init(&vq->vring, num, pages, vring_align); |
0a8a69dd RR |
778 | vq->vq.callback = callback; |
779 | vq->vq.vdev = vdev; | |
9499f5e7 | 780 | vq->vq.name = name; |
06ca287d RR |
781 | vq->vq.num_free = num; |
782 | vq->vq.index = index; | |
0a8a69dd | 783 | vq->notify = notify; |
7b21e34f | 784 | vq->weak_barriers = weak_barriers; |
0a8a69dd RR |
785 | vq->broken = false; |
786 | vq->last_used_idx = 0; | |
787 | vq->num_added = 0; | |
9499f5e7 | 788 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0a8a69dd RR |
789 | #ifdef DEBUG |
790 | vq->in_use = false; | |
e93300b1 | 791 | vq->last_add_time_valid = false; |
0a8a69dd RR |
792 | #endif |
793 | ||
9fa29b9d | 794 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC); |
a5c262c5 | 795 | vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); |
9fa29b9d | 796 | |
0a8a69dd RR |
797 | /* No callback? Tell other side not to bother us. */ |
798 | if (!callback) | |
799 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; | |
800 | ||
801 | /* Put everything in free lists. */ | |
0a8a69dd | 802 | vq->free_head = 0; |
3b870624 | 803 | for (i = 0; i < num-1; i++) { |
0a8a69dd | 804 | vq->vring.desc[i].next = i+1; |
3b870624 AS |
805 | vq->data[i] = NULL; |
806 | } | |
807 | vq->data[i] = NULL; | |
0a8a69dd RR |
808 | |
809 | return &vq->vq; | |
810 | } | |
c6fd4701 | 811 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd RR |
812 | |
813 | void vring_del_virtqueue(struct virtqueue *vq) | |
814 | { | |
9499f5e7 | 815 | list_del(&vq->list); |
0a8a69dd RR |
816 | kfree(to_vvq(vq)); |
817 | } | |
c6fd4701 | 818 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 819 | |
e34f8725 RR |
820 | /* Manipulates transport-specific feature bits. */ |
821 | void vring_transport_features(struct virtio_device *vdev) | |
822 | { | |
823 | unsigned int i; | |
824 | ||
825 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
826 | switch (i) { | |
9fa29b9d MM |
827 | case VIRTIO_RING_F_INDIRECT_DESC: |
828 | break; | |
a5c262c5 MT |
829 | case VIRTIO_RING_F_EVENT_IDX: |
830 | break; | |
e34f8725 RR |
831 | default: |
832 | /* We don't understand this bit. */ | |
833 | clear_bit(i, vdev->features); | |
834 | } | |
835 | } | |
836 | } | |
837 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
838 | ||
5dfc1762 RR |
839 | /** |
840 | * virtqueue_get_vring_size - return the size of the virtqueue's vring | |
841 | * @vq: the struct virtqueue containing the vring of interest. | |
842 | * | |
843 | * Returns the size of the vring. This is mainly used for boasting to | |
844 | * userspace. Unlike other operations, this need not be serialized. | |
845 | */ | |
8f9f4668 RJ |
846 | unsigned int virtqueue_get_vring_size(struct virtqueue *_vq) |
847 | { | |
848 | ||
849 | struct vring_virtqueue *vq = to_vvq(_vq); | |
850 | ||
851 | return vq->vring.num; | |
852 | } | |
853 | EXPORT_SYMBOL_GPL(virtqueue_get_vring_size); | |
854 | ||
b3b32c94 HG |
855 | bool virtqueue_is_broken(struct virtqueue *_vq) |
856 | { | |
857 | struct vring_virtqueue *vq = to_vvq(_vq); | |
858 | ||
859 | return vq->broken; | |
860 | } | |
861 | EXPORT_SYMBOL_GPL(virtqueue_is_broken); | |
862 | ||
c6fd4701 | 863 | MODULE_LICENSE("GPL"); |