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 | ||
00e6f3d9 MT |
102 | static struct vring_desc *alloc_indirect(struct virtqueue *_vq, |
103 | unsigned int total_sg, gfp_t gfp) | |
9fa29b9d MM |
104 | { |
105 | struct vring_desc *desc; | |
b25bd251 | 106 | unsigned int i; |
9fa29b9d | 107 | |
b92b1b89 WD |
108 | /* |
109 | * We require lowmem mappings for the descriptors because | |
110 | * otherwise virt_to_phys will give us bogus addresses in the | |
111 | * virtqueue. | |
112 | */ | |
113 | gfp &= ~(__GFP_HIGHMEM | __GFP_HIGH); | |
114 | ||
13816c76 | 115 | desc = kmalloc(total_sg * sizeof(struct vring_desc), gfp); |
9fa29b9d | 116 | if (!desc) |
b25bd251 | 117 | return NULL; |
9fa29b9d | 118 | |
b25bd251 | 119 | for (i = 0; i < total_sg; i++) |
00e6f3d9 | 120 | desc[i].next = cpu_to_virtio16(_vq->vdev, i + 1); |
b25bd251 | 121 | return desc; |
9fa29b9d MM |
122 | } |
123 | ||
13816c76 RR |
124 | static inline int virtqueue_add(struct virtqueue *_vq, |
125 | struct scatterlist *sgs[], | |
eeebf9b1 | 126 | unsigned int total_sg, |
13816c76 RR |
127 | unsigned int out_sgs, |
128 | unsigned int in_sgs, | |
129 | void *data, | |
130 | gfp_t gfp) | |
0a8a69dd RR |
131 | { |
132 | struct vring_virtqueue *vq = to_vvq(_vq); | |
13816c76 | 133 | struct scatterlist *sg; |
b25bd251 RR |
134 | struct vring_desc *desc; |
135 | unsigned int i, n, avail, descs_used, uninitialized_var(prev); | |
1fe9b6fe | 136 | int head; |
b25bd251 | 137 | bool indirect; |
0a8a69dd | 138 | |
9fa29b9d MM |
139 | START_USE(vq); |
140 | ||
0a8a69dd | 141 | BUG_ON(data == NULL); |
9fa29b9d | 142 | |
70670444 RR |
143 | if (unlikely(vq->broken)) { |
144 | END_USE(vq); | |
145 | return -EIO; | |
146 | } | |
147 | ||
e93300b1 RR |
148 | #ifdef DEBUG |
149 | { | |
150 | ktime_t now = ktime_get(); | |
151 | ||
152 | /* No kick or get, with .1 second between? Warn. */ | |
153 | if (vq->last_add_time_valid) | |
154 | WARN_ON(ktime_to_ms(ktime_sub(now, vq->last_add_time)) | |
155 | > 100); | |
156 | vq->last_add_time = now; | |
157 | vq->last_add_time_valid = true; | |
158 | } | |
159 | #endif | |
160 | ||
b25bd251 RR |
161 | BUG_ON(total_sg > vq->vring.num); |
162 | BUG_ON(total_sg == 0); | |
163 | ||
164 | head = vq->free_head; | |
165 | ||
9fa29b9d MM |
166 | /* If the host supports indirect descriptor tables, and we have multiple |
167 | * buffers, then go indirect. FIXME: tune this threshold */ | |
b25bd251 | 168 | if (vq->indirect && total_sg > 1 && vq->vq.num_free) |
00e6f3d9 | 169 | desc = alloc_indirect(_vq, total_sg, gfp); |
b25bd251 RR |
170 | else |
171 | desc = NULL; | |
172 | ||
173 | if (desc) { | |
174 | /* Use a single buffer which doesn't continue */ | |
00e6f3d9 MT |
175 | vq->vring.desc[head].flags = cpu_to_virtio16(_vq->vdev, VRING_DESC_F_INDIRECT); |
176 | vq->vring.desc[head].addr = cpu_to_virtio64(_vq->vdev, virt_to_phys(desc)); | |
b25bd251 RR |
177 | /* avoid kmemleak false positive (hidden by virt_to_phys) */ |
178 | kmemleak_ignore(desc); | |
00e6f3d9 | 179 | vq->vring.desc[head].len = cpu_to_virtio32(_vq->vdev, total_sg * sizeof(struct vring_desc)); |
b25bd251 RR |
180 | |
181 | /* Set up rest to use this indirect table. */ | |
182 | i = 0; | |
183 | descs_used = 1; | |
184 | indirect = true; | |
185 | } else { | |
186 | desc = vq->vring.desc; | |
187 | i = head; | |
188 | descs_used = total_sg; | |
189 | indirect = false; | |
9fa29b9d MM |
190 | } |
191 | ||
b25bd251 | 192 | if (vq->vq.num_free < descs_used) { |
0a8a69dd | 193 | pr_debug("Can't add buf len %i - avail = %i\n", |
b25bd251 | 194 | descs_used, vq->vq.num_free); |
44653eae RR |
195 | /* FIXME: for historical reasons, we force a notify here if |
196 | * there are outgoing parts to the buffer. Presumably the | |
197 | * host should service the ring ASAP. */ | |
13816c76 | 198 | if (out_sgs) |
44653eae | 199 | vq->notify(&vq->vq); |
0a8a69dd RR |
200 | END_USE(vq); |
201 | return -ENOSPC; | |
202 | } | |
203 | ||
204 | /* We're about to use some buffers from the free list. */ | |
b25bd251 | 205 | vq->vq.num_free -= descs_used; |
13816c76 | 206 | |
13816c76 | 207 | for (n = 0; n < out_sgs; n++) { |
eeebf9b1 | 208 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
00e6f3d9 MT |
209 | desc[i].flags = cpu_to_virtio16(_vq->vdev, VRING_DESC_F_NEXT); |
210 | desc[i].addr = cpu_to_virtio64(_vq->vdev, sg_phys(sg)); | |
211 | desc[i].len = cpu_to_virtio32(_vq->vdev, sg->length); | |
13816c76 | 212 | prev = i; |
00e6f3d9 | 213 | i = virtio16_to_cpu(_vq->vdev, desc[i].next); |
13816c76 | 214 | } |
0a8a69dd | 215 | } |
13816c76 | 216 | for (; n < (out_sgs + in_sgs); n++) { |
eeebf9b1 | 217 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
00e6f3d9 MT |
218 | desc[i].flags = cpu_to_virtio16(_vq->vdev, VRING_DESC_F_NEXT | VRING_DESC_F_WRITE); |
219 | desc[i].addr = cpu_to_virtio64(_vq->vdev, sg_phys(sg)); | |
220 | desc[i].len = cpu_to_virtio32(_vq->vdev, sg->length); | |
13816c76 | 221 | prev = i; |
00e6f3d9 | 222 | i = virtio16_to_cpu(_vq->vdev, desc[i].next); |
13816c76 | 223 | } |
0a8a69dd RR |
224 | } |
225 | /* Last one doesn't continue. */ | |
00e6f3d9 | 226 | desc[prev].flags &= cpu_to_virtio16(_vq->vdev, ~VRING_DESC_F_NEXT); |
0a8a69dd RR |
227 | |
228 | /* Update free pointer */ | |
b25bd251 | 229 | if (indirect) |
00e6f3d9 | 230 | vq->free_head = virtio16_to_cpu(_vq->vdev, vq->vring.desc[head].next); |
b25bd251 RR |
231 | else |
232 | vq->free_head = i; | |
0a8a69dd RR |
233 | |
234 | /* Set token. */ | |
235 | vq->data[head] = data; | |
236 | ||
237 | /* Put entry in available array (but don't update avail->idx until they | |
3b720b8c | 238 | * do sync). */ |
00e6f3d9 MT |
239 | avail = virtio16_to_cpu(_vq->vdev, vq->vring.avail->idx) & (vq->vring.num - 1); |
240 | vq->vring.avail->ring[avail] = cpu_to_virtio16(_vq->vdev, head); | |
0a8a69dd | 241 | |
ee7cd898 RR |
242 | /* Descriptors and available array need to be set before we expose the |
243 | * new available array entries. */ | |
a9a0fef7 | 244 | virtio_wmb(vq->weak_barriers); |
00e6f3d9 | 245 | vq->vring.avail->idx = cpu_to_virtio16(_vq->vdev, virtio16_to_cpu(_vq->vdev, vq->vring.avail->idx) + 1); |
ee7cd898 RR |
246 | vq->num_added++; |
247 | ||
248 | /* This is very unlikely, but theoretically possible. Kick | |
249 | * just in case. */ | |
250 | if (unlikely(vq->num_added == (1 << 16) - 1)) | |
251 | virtqueue_kick(_vq); | |
252 | ||
0a8a69dd RR |
253 | pr_debug("Added buffer head %i to %p\n", head, vq); |
254 | END_USE(vq); | |
3c1b27d5 | 255 | |
98e8c6bc | 256 | return 0; |
0a8a69dd | 257 | } |
13816c76 | 258 | |
13816c76 RR |
259 | /** |
260 | * virtqueue_add_sgs - expose buffers to other end | |
261 | * @vq: the struct virtqueue we're talking about. | |
262 | * @sgs: array of terminated scatterlists. | |
263 | * @out_num: the number of scatterlists readable by other side | |
264 | * @in_num: the number of scatterlists which are writable (after readable ones) | |
265 | * @data: the token identifying the buffer. | |
266 | * @gfp: how to do memory allocations (if necessary). | |
267 | * | |
268 | * Caller must ensure we don't call this with other virtqueue operations | |
269 | * at the same time (except where noted). | |
270 | * | |
70670444 | 271 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
13816c76 RR |
272 | */ |
273 | int virtqueue_add_sgs(struct virtqueue *_vq, | |
274 | struct scatterlist *sgs[], | |
275 | unsigned int out_sgs, | |
276 | unsigned int in_sgs, | |
277 | void *data, | |
278 | gfp_t gfp) | |
279 | { | |
eeebf9b1 | 280 | unsigned int i, total_sg = 0; |
13816c76 RR |
281 | |
282 | /* Count them first. */ | |
eeebf9b1 | 283 | for (i = 0; i < out_sgs + in_sgs; i++) { |
13816c76 RR |
284 | struct scatterlist *sg; |
285 | for (sg = sgs[i]; sg; sg = sg_next(sg)) | |
eeebf9b1 | 286 | total_sg++; |
13816c76 | 287 | } |
eeebf9b1 | 288 | return virtqueue_add(_vq, sgs, total_sg, out_sgs, in_sgs, data, gfp); |
13816c76 RR |
289 | } |
290 | EXPORT_SYMBOL_GPL(virtqueue_add_sgs); | |
291 | ||
282edb36 RR |
292 | /** |
293 | * virtqueue_add_outbuf - expose output buffers to other end | |
294 | * @vq: the struct virtqueue we're talking about. | |
eeebf9b1 RR |
295 | * @sg: scatterlist (must be well-formed and terminated!) |
296 | * @num: the number of entries in @sg readable by other side | |
282edb36 RR |
297 | * @data: the token identifying the buffer. |
298 | * @gfp: how to do memory allocations (if necessary). | |
299 | * | |
300 | * Caller must ensure we don't call this with other virtqueue operations | |
301 | * at the same time (except where noted). | |
302 | * | |
70670444 | 303 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
282edb36 RR |
304 | */ |
305 | int virtqueue_add_outbuf(struct virtqueue *vq, | |
eeebf9b1 | 306 | struct scatterlist *sg, unsigned int num, |
282edb36 RR |
307 | void *data, |
308 | gfp_t gfp) | |
309 | { | |
eeebf9b1 | 310 | return virtqueue_add(vq, &sg, num, 1, 0, data, gfp); |
282edb36 RR |
311 | } |
312 | EXPORT_SYMBOL_GPL(virtqueue_add_outbuf); | |
313 | ||
314 | /** | |
315 | * virtqueue_add_inbuf - expose input buffers to other end | |
316 | * @vq: the struct virtqueue we're talking about. | |
eeebf9b1 RR |
317 | * @sg: scatterlist (must be well-formed and terminated!) |
318 | * @num: the number of entries in @sg writable by other side | |
282edb36 RR |
319 | * @data: the token identifying the buffer. |
320 | * @gfp: how to do memory allocations (if necessary). | |
321 | * | |
322 | * Caller must ensure we don't call this with other virtqueue operations | |
323 | * at the same time (except where noted). | |
324 | * | |
70670444 | 325 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
282edb36 RR |
326 | */ |
327 | int virtqueue_add_inbuf(struct virtqueue *vq, | |
eeebf9b1 | 328 | struct scatterlist *sg, unsigned int num, |
282edb36 RR |
329 | void *data, |
330 | gfp_t gfp) | |
331 | { | |
eeebf9b1 | 332 | return virtqueue_add(vq, &sg, num, 0, 1, data, gfp); |
282edb36 RR |
333 | } |
334 | EXPORT_SYMBOL_GPL(virtqueue_add_inbuf); | |
335 | ||
5dfc1762 | 336 | /** |
41f0377f | 337 | * virtqueue_kick_prepare - first half of split virtqueue_kick call. |
5dfc1762 RR |
338 | * @vq: the struct virtqueue |
339 | * | |
41f0377f RR |
340 | * Instead of virtqueue_kick(), you can do: |
341 | * if (virtqueue_kick_prepare(vq)) | |
342 | * virtqueue_notify(vq); | |
5dfc1762 | 343 | * |
41f0377f RR |
344 | * This is sometimes useful because the virtqueue_kick_prepare() needs |
345 | * to be serialized, but the actual virtqueue_notify() call does not. | |
5dfc1762 | 346 | */ |
41f0377f | 347 | bool virtqueue_kick_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
348 | { |
349 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a5c262c5 | 350 | u16 new, old; |
41f0377f RR |
351 | bool needs_kick; |
352 | ||
0a8a69dd | 353 | START_USE(vq); |
a72caae2 JW |
354 | /* We need to expose available array entries before checking avail |
355 | * event. */ | |
a9a0fef7 | 356 | virtio_mb(vq->weak_barriers); |
0a8a69dd | 357 | |
00e6f3d9 MT |
358 | old = virtio16_to_cpu(_vq->vdev, vq->vring.avail->idx) - vq->num_added; |
359 | new = virtio16_to_cpu(_vq->vdev, vq->vring.avail->idx); | |
0a8a69dd RR |
360 | vq->num_added = 0; |
361 | ||
e93300b1 RR |
362 | #ifdef DEBUG |
363 | if (vq->last_add_time_valid) { | |
364 | WARN_ON(ktime_to_ms(ktime_sub(ktime_get(), | |
365 | vq->last_add_time)) > 100); | |
366 | } | |
367 | vq->last_add_time_valid = false; | |
368 | #endif | |
369 | ||
41f0377f | 370 | if (vq->event) { |
00e6f3d9 | 371 | needs_kick = vring_need_event(virtio16_to_cpu(_vq->vdev, vring_avail_event(&vq->vring)), |
41f0377f RR |
372 | new, old); |
373 | } else { | |
00e6f3d9 | 374 | needs_kick = !(vq->vring.used->flags & cpu_to_virtio16(_vq->vdev, VRING_USED_F_NO_NOTIFY)); |
41f0377f | 375 | } |
0a8a69dd | 376 | END_USE(vq); |
41f0377f RR |
377 | return needs_kick; |
378 | } | |
379 | EXPORT_SYMBOL_GPL(virtqueue_kick_prepare); | |
380 | ||
381 | /** | |
382 | * virtqueue_notify - second half of split virtqueue_kick call. | |
383 | * @vq: the struct virtqueue | |
384 | * | |
385 | * This does not need to be serialized. | |
5b1bf7cb HG |
386 | * |
387 | * Returns false if host notify failed or queue is broken, otherwise true. | |
41f0377f | 388 | */ |
5b1bf7cb | 389 | bool virtqueue_notify(struct virtqueue *_vq) |
41f0377f RR |
390 | { |
391 | struct vring_virtqueue *vq = to_vvq(_vq); | |
392 | ||
5b1bf7cb HG |
393 | if (unlikely(vq->broken)) |
394 | return false; | |
395 | ||
41f0377f | 396 | /* Prod other side to tell it about changes. */ |
2342d6a6 | 397 | if (!vq->notify(_vq)) { |
5b1bf7cb HG |
398 | vq->broken = true; |
399 | return false; | |
400 | } | |
401 | return true; | |
41f0377f RR |
402 | } |
403 | EXPORT_SYMBOL_GPL(virtqueue_notify); | |
404 | ||
405 | /** | |
406 | * virtqueue_kick - update after add_buf | |
407 | * @vq: the struct virtqueue | |
408 | * | |
b3087e48 | 409 | * After one or more virtqueue_add_* calls, invoke this to kick |
41f0377f RR |
410 | * the other side. |
411 | * | |
412 | * Caller must ensure we don't call this with other virtqueue | |
413 | * operations at the same time (except where noted). | |
5b1bf7cb HG |
414 | * |
415 | * Returns false if kick failed, otherwise true. | |
41f0377f | 416 | */ |
5b1bf7cb | 417 | bool virtqueue_kick(struct virtqueue *vq) |
41f0377f RR |
418 | { |
419 | if (virtqueue_kick_prepare(vq)) | |
5b1bf7cb HG |
420 | return virtqueue_notify(vq); |
421 | return true; | |
0a8a69dd | 422 | } |
7c5e9ed0 | 423 | EXPORT_SYMBOL_GPL(virtqueue_kick); |
0a8a69dd RR |
424 | |
425 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head) | |
426 | { | |
427 | unsigned int i; | |
428 | ||
429 | /* Clear data ptr. */ | |
430 | vq->data[head] = NULL; | |
431 | ||
432 | /* Put back on free list: find end */ | |
433 | i = head; | |
9fa29b9d MM |
434 | |
435 | /* Free the indirect table */ | |
00e6f3d9 MT |
436 | if (vq->vring.desc[i].flags & cpu_to_virtio16(vq->vq.vdev, VRING_DESC_F_INDIRECT)) |
437 | kfree(phys_to_virt(virtio64_to_cpu(vq->vq.vdev, vq->vring.desc[i].addr))); | |
9fa29b9d | 438 | |
00e6f3d9 MT |
439 | while (vq->vring.desc[i].flags & cpu_to_virtio16(vq->vq.vdev, VRING_DESC_F_NEXT)) { |
440 | i = virtio16_to_cpu(vq->vq.vdev, vq->vring.desc[i].next); | |
06ca287d | 441 | vq->vq.num_free++; |
0a8a69dd RR |
442 | } |
443 | ||
00e6f3d9 | 444 | vq->vring.desc[i].next = cpu_to_virtio16(vq->vq.vdev, vq->free_head); |
0a8a69dd RR |
445 | vq->free_head = head; |
446 | /* Plus final descriptor */ | |
06ca287d | 447 | vq->vq.num_free++; |
0a8a69dd RR |
448 | } |
449 | ||
0a8a69dd RR |
450 | static inline bool more_used(const struct vring_virtqueue *vq) |
451 | { | |
00e6f3d9 | 452 | return vq->last_used_idx != virtio16_to_cpu(vq->vq.vdev, vq->vring.used->idx); |
0a8a69dd RR |
453 | } |
454 | ||
5dfc1762 RR |
455 | /** |
456 | * virtqueue_get_buf - get the next used buffer | |
457 | * @vq: the struct virtqueue we're talking about. | |
458 | * @len: the length written into the buffer | |
459 | * | |
460 | * If the driver wrote data into the buffer, @len will be set to the | |
461 | * amount written. This means you don't need to clear the buffer | |
462 | * beforehand to ensure there's no data leakage in the case of short | |
463 | * writes. | |
464 | * | |
465 | * Caller must ensure we don't call this with other virtqueue | |
466 | * operations at the same time (except where noted). | |
467 | * | |
468 | * Returns NULL if there are no used buffers, or the "data" token | |
b3087e48 | 469 | * handed to virtqueue_add_*(). |
5dfc1762 | 470 | */ |
7c5e9ed0 | 471 | void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len) |
0a8a69dd RR |
472 | { |
473 | struct vring_virtqueue *vq = to_vvq(_vq); | |
474 | void *ret; | |
475 | unsigned int i; | |
3b720b8c | 476 | u16 last_used; |
0a8a69dd RR |
477 | |
478 | START_USE(vq); | |
479 | ||
5ef82752 RR |
480 | if (unlikely(vq->broken)) { |
481 | END_USE(vq); | |
482 | return NULL; | |
483 | } | |
484 | ||
0a8a69dd RR |
485 | if (!more_used(vq)) { |
486 | pr_debug("No more buffers in queue\n"); | |
487 | END_USE(vq); | |
488 | return NULL; | |
489 | } | |
490 | ||
2d61ba95 | 491 | /* Only get used array entries after they have been exposed by host. */ |
a9a0fef7 | 492 | virtio_rmb(vq->weak_barriers); |
2d61ba95 | 493 | |
3b720b8c | 494 | last_used = (vq->last_used_idx & (vq->vring.num - 1)); |
00e6f3d9 MT |
495 | i = virtio32_to_cpu(_vq->vdev, vq->vring.used->ring[last_used].id); |
496 | *len = virtio32_to_cpu(_vq->vdev, vq->vring.used->ring[last_used].len); | |
0a8a69dd RR |
497 | |
498 | if (unlikely(i >= vq->vring.num)) { | |
499 | BAD_RING(vq, "id %u out of range\n", i); | |
500 | return NULL; | |
501 | } | |
502 | if (unlikely(!vq->data[i])) { | |
503 | BAD_RING(vq, "id %u is not a head!\n", i); | |
504 | return NULL; | |
505 | } | |
506 | ||
507 | /* detach_buf clears data, so grab it now. */ | |
508 | ret = vq->data[i]; | |
509 | detach_buf(vq, i); | |
510 | vq->last_used_idx++; | |
a5c262c5 MT |
511 | /* If we expect an interrupt for the next entry, tell host |
512 | * by writing event index and flush out the write before | |
513 | * the read in the next get_buf call. */ | |
00e6f3d9 MT |
514 | if (!(vq->vring.avail->flags & cpu_to_virtio16(_vq->vdev, VRING_AVAIL_F_NO_INTERRUPT))) { |
515 | vring_used_event(&vq->vring) = cpu_to_virtio16(_vq->vdev, vq->last_used_idx); | |
a9a0fef7 | 516 | virtio_mb(vq->weak_barriers); |
a5c262c5 MT |
517 | } |
518 | ||
e93300b1 RR |
519 | #ifdef DEBUG |
520 | vq->last_add_time_valid = false; | |
521 | #endif | |
522 | ||
0a8a69dd RR |
523 | END_USE(vq); |
524 | return ret; | |
525 | } | |
7c5e9ed0 | 526 | EXPORT_SYMBOL_GPL(virtqueue_get_buf); |
0a8a69dd | 527 | |
5dfc1762 RR |
528 | /** |
529 | * virtqueue_disable_cb - disable callbacks | |
530 | * @vq: the struct virtqueue we're talking about. | |
531 | * | |
532 | * Note that this is not necessarily synchronous, hence unreliable and only | |
533 | * useful as an optimization. | |
534 | * | |
535 | * Unlike other operations, this need not be serialized. | |
536 | */ | |
7c5e9ed0 | 537 | void virtqueue_disable_cb(struct virtqueue *_vq) |
18445c4d RR |
538 | { |
539 | struct vring_virtqueue *vq = to_vvq(_vq); | |
540 | ||
00e6f3d9 | 541 | vq->vring.avail->flags |= cpu_to_virtio16(_vq->vdev, VRING_AVAIL_F_NO_INTERRUPT); |
18445c4d | 542 | } |
7c5e9ed0 | 543 | EXPORT_SYMBOL_GPL(virtqueue_disable_cb); |
18445c4d | 544 | |
5dfc1762 | 545 | /** |
cc229884 | 546 | * virtqueue_enable_cb_prepare - restart callbacks after disable_cb |
5dfc1762 RR |
547 | * @vq: the struct virtqueue we're talking about. |
548 | * | |
cc229884 MT |
549 | * This re-enables callbacks; it returns current queue state |
550 | * in an opaque unsigned value. This value should be later tested by | |
551 | * virtqueue_poll, to detect a possible race between the driver checking for | |
552 | * more work, and enabling callbacks. | |
5dfc1762 RR |
553 | * |
554 | * Caller must ensure we don't call this with other virtqueue | |
555 | * operations at the same time (except where noted). | |
556 | */ | |
cc229884 | 557 | unsigned virtqueue_enable_cb_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
558 | { |
559 | struct vring_virtqueue *vq = to_vvq(_vq); | |
cc229884 | 560 | u16 last_used_idx; |
0a8a69dd RR |
561 | |
562 | START_USE(vq); | |
0a8a69dd RR |
563 | |
564 | /* We optimistically turn back on interrupts, then check if there was | |
565 | * more to do. */ | |
a5c262c5 MT |
566 | /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to |
567 | * either clear the flags bit or point the event index at the next | |
568 | * entry. Always do both to keep code simple. */ | |
00e6f3d9 MT |
569 | vq->vring.avail->flags &= cpu_to_virtio16(_vq->vdev, ~VRING_AVAIL_F_NO_INTERRUPT); |
570 | vring_used_event(&vq->vring) = cpu_to_virtio16(_vq->vdev, last_used_idx = vq->last_used_idx); | |
cc229884 MT |
571 | END_USE(vq); |
572 | return last_used_idx; | |
573 | } | |
574 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_prepare); | |
575 | ||
576 | /** | |
577 | * virtqueue_poll - query pending used buffers | |
578 | * @vq: the struct virtqueue we're talking about. | |
579 | * @last_used_idx: virtqueue state (from call to virtqueue_enable_cb_prepare). | |
580 | * | |
581 | * Returns "true" if there are pending used buffers in the queue. | |
582 | * | |
583 | * This does not need to be serialized. | |
584 | */ | |
585 | bool virtqueue_poll(struct virtqueue *_vq, unsigned last_used_idx) | |
586 | { | |
587 | struct vring_virtqueue *vq = to_vvq(_vq); | |
588 | ||
a9a0fef7 | 589 | virtio_mb(vq->weak_barriers); |
00e6f3d9 | 590 | return (u16)last_used_idx != virtio16_to_cpu(_vq->vdev, vq->vring.used->idx); |
cc229884 MT |
591 | } |
592 | EXPORT_SYMBOL_GPL(virtqueue_poll); | |
0a8a69dd | 593 | |
cc229884 MT |
594 | /** |
595 | * virtqueue_enable_cb - restart callbacks after disable_cb. | |
596 | * @vq: the struct virtqueue we're talking about. | |
597 | * | |
598 | * This re-enables callbacks; it returns "false" if there are pending | |
599 | * buffers in the queue, to detect a possible race between the driver | |
600 | * checking for more work, and enabling callbacks. | |
601 | * | |
602 | * Caller must ensure we don't call this with other virtqueue | |
603 | * operations at the same time (except where noted). | |
604 | */ | |
605 | bool virtqueue_enable_cb(struct virtqueue *_vq) | |
606 | { | |
607 | unsigned last_used_idx = virtqueue_enable_cb_prepare(_vq); | |
608 | return !virtqueue_poll(_vq, last_used_idx); | |
0a8a69dd | 609 | } |
7c5e9ed0 | 610 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb); |
0a8a69dd | 611 | |
5dfc1762 RR |
612 | /** |
613 | * virtqueue_enable_cb_delayed - restart callbacks after disable_cb. | |
614 | * @vq: the struct virtqueue we're talking about. | |
615 | * | |
616 | * This re-enables callbacks but hints to the other side to delay | |
617 | * interrupts until most of the available buffers have been processed; | |
618 | * it returns "false" if there are many pending buffers in the queue, | |
619 | * to detect a possible race between the driver checking for more work, | |
620 | * and enabling callbacks. | |
621 | * | |
622 | * Caller must ensure we don't call this with other virtqueue | |
623 | * operations at the same time (except where noted). | |
624 | */ | |
7ab358c2 MT |
625 | bool virtqueue_enable_cb_delayed(struct virtqueue *_vq) |
626 | { | |
627 | struct vring_virtqueue *vq = to_vvq(_vq); | |
628 | u16 bufs; | |
629 | ||
630 | START_USE(vq); | |
631 | ||
632 | /* We optimistically turn back on interrupts, then check if there was | |
633 | * more to do. */ | |
634 | /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to | |
635 | * either clear the flags bit or point the event index at the next | |
636 | * entry. Always do both to keep code simple. */ | |
00e6f3d9 | 637 | vq->vring.avail->flags &= cpu_to_virtio16(_vq->vdev, ~VRING_AVAIL_F_NO_INTERRUPT); |
7ab358c2 | 638 | /* TODO: tune this threshold */ |
00e6f3d9 MT |
639 | bufs = (u16)(virtio16_to_cpu(_vq->vdev, vq->vring.avail->idx) - vq->last_used_idx) * 3 / 4; |
640 | vring_used_event(&vq->vring) = cpu_to_virtio16(_vq->vdev, vq->last_used_idx + bufs); | |
a9a0fef7 | 641 | virtio_mb(vq->weak_barriers); |
00e6f3d9 | 642 | if (unlikely((u16)(virtio16_to_cpu(_vq->vdev, vq->vring.used->idx) - vq->last_used_idx) > bufs)) { |
7ab358c2 MT |
643 | END_USE(vq); |
644 | return false; | |
645 | } | |
646 | ||
647 | END_USE(vq); | |
648 | return true; | |
649 | } | |
650 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed); | |
651 | ||
5dfc1762 RR |
652 | /** |
653 | * virtqueue_detach_unused_buf - detach first unused buffer | |
654 | * @vq: the struct virtqueue we're talking about. | |
655 | * | |
b3087e48 | 656 | * Returns NULL or the "data" token handed to virtqueue_add_*(). |
5dfc1762 RR |
657 | * This is not valid on an active queue; it is useful only for device |
658 | * shutdown. | |
659 | */ | |
7c5e9ed0 | 660 | void *virtqueue_detach_unused_buf(struct virtqueue *_vq) |
c021eac4 SM |
661 | { |
662 | struct vring_virtqueue *vq = to_vvq(_vq); | |
663 | unsigned int i; | |
664 | void *buf; | |
665 | ||
666 | START_USE(vq); | |
667 | ||
668 | for (i = 0; i < vq->vring.num; i++) { | |
669 | if (!vq->data[i]) | |
670 | continue; | |
671 | /* detach_buf clears data, so grab it now. */ | |
672 | buf = vq->data[i]; | |
673 | detach_buf(vq, i); | |
00e6f3d9 | 674 | vq->vring.avail->idx = cpu_to_virtio16(_vq->vdev, virtio16_to_cpu(_vq->vdev, vq->vring.avail->idx) - 1); |
c021eac4 SM |
675 | END_USE(vq); |
676 | return buf; | |
677 | } | |
678 | /* That should have freed everything. */ | |
06ca287d | 679 | BUG_ON(vq->vq.num_free != vq->vring.num); |
c021eac4 SM |
680 | |
681 | END_USE(vq); | |
682 | return NULL; | |
683 | } | |
7c5e9ed0 | 684 | EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf); |
c021eac4 | 685 | |
0a8a69dd RR |
686 | irqreturn_t vring_interrupt(int irq, void *_vq) |
687 | { | |
688 | struct vring_virtqueue *vq = to_vvq(_vq); | |
689 | ||
690 | if (!more_used(vq)) { | |
691 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
692 | return IRQ_NONE; | |
693 | } | |
694 | ||
695 | if (unlikely(vq->broken)) | |
696 | return IRQ_HANDLED; | |
697 | ||
698 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); | |
18445c4d RR |
699 | if (vq->vq.callback) |
700 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
701 | |
702 | return IRQ_HANDLED; | |
703 | } | |
c6fd4701 | 704 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd | 705 | |
17bb6d40 JW |
706 | struct virtqueue *vring_new_virtqueue(unsigned int index, |
707 | unsigned int num, | |
87c7d57c | 708 | unsigned int vring_align, |
0a8a69dd | 709 | struct virtio_device *vdev, |
7b21e34f | 710 | bool weak_barriers, |
0a8a69dd | 711 | void *pages, |
46f9c2b9 | 712 | bool (*notify)(struct virtqueue *), |
9499f5e7 RR |
713 | void (*callback)(struct virtqueue *), |
714 | const char *name) | |
0a8a69dd RR |
715 | { |
716 | struct vring_virtqueue *vq; | |
717 | unsigned int i; | |
718 | ||
42b36cc0 RR |
719 | /* We assume num is a power of 2. */ |
720 | if (num & (num - 1)) { | |
721 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
722 | return NULL; | |
723 | } | |
724 | ||
0a8a69dd RR |
725 | vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL); |
726 | if (!vq) | |
727 | return NULL; | |
728 | ||
87c7d57c | 729 | vring_init(&vq->vring, num, pages, vring_align); |
0a8a69dd RR |
730 | vq->vq.callback = callback; |
731 | vq->vq.vdev = vdev; | |
9499f5e7 | 732 | vq->vq.name = name; |
06ca287d RR |
733 | vq->vq.num_free = num; |
734 | vq->vq.index = index; | |
0a8a69dd | 735 | vq->notify = notify; |
7b21e34f | 736 | vq->weak_barriers = weak_barriers; |
0a8a69dd RR |
737 | vq->broken = false; |
738 | vq->last_used_idx = 0; | |
739 | vq->num_added = 0; | |
9499f5e7 | 740 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0a8a69dd RR |
741 | #ifdef DEBUG |
742 | vq->in_use = false; | |
e93300b1 | 743 | vq->last_add_time_valid = false; |
0a8a69dd RR |
744 | #endif |
745 | ||
9fa29b9d | 746 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC); |
a5c262c5 | 747 | vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); |
9fa29b9d | 748 | |
0a8a69dd RR |
749 | /* No callback? Tell other side not to bother us. */ |
750 | if (!callback) | |
00e6f3d9 | 751 | vq->vring.avail->flags |= cpu_to_virtio16(vdev, VRING_AVAIL_F_NO_INTERRUPT); |
0a8a69dd RR |
752 | |
753 | /* Put everything in free lists. */ | |
0a8a69dd | 754 | vq->free_head = 0; |
3b870624 | 755 | for (i = 0; i < num-1; i++) { |
00e6f3d9 | 756 | vq->vring.desc[i].next = cpu_to_virtio16(vdev, i + 1); |
3b870624 AS |
757 | vq->data[i] = NULL; |
758 | } | |
759 | vq->data[i] = NULL; | |
0a8a69dd RR |
760 | |
761 | return &vq->vq; | |
762 | } | |
c6fd4701 | 763 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd RR |
764 | |
765 | void vring_del_virtqueue(struct virtqueue *vq) | |
766 | { | |
9499f5e7 | 767 | list_del(&vq->list); |
0a8a69dd RR |
768 | kfree(to_vvq(vq)); |
769 | } | |
c6fd4701 | 770 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 771 | |
e34f8725 RR |
772 | /* Manipulates transport-specific feature bits. */ |
773 | void vring_transport_features(struct virtio_device *vdev) | |
774 | { | |
775 | unsigned int i; | |
776 | ||
777 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
778 | switch (i) { | |
9fa29b9d MM |
779 | case VIRTIO_RING_F_INDIRECT_DESC: |
780 | break; | |
a5c262c5 MT |
781 | case VIRTIO_RING_F_EVENT_IDX: |
782 | break; | |
747ae34a MT |
783 | case VIRTIO_F_VERSION_1: |
784 | break; | |
e34f8725 RR |
785 | default: |
786 | /* We don't understand this bit. */ | |
e16e12be | 787 | __virtio_clear_bit(vdev, i); |
e34f8725 RR |
788 | } |
789 | } | |
790 | } | |
791 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
792 | ||
5dfc1762 RR |
793 | /** |
794 | * virtqueue_get_vring_size - return the size of the virtqueue's vring | |
795 | * @vq: the struct virtqueue containing the vring of interest. | |
796 | * | |
797 | * Returns the size of the vring. This is mainly used for boasting to | |
798 | * userspace. Unlike other operations, this need not be serialized. | |
799 | */ | |
8f9f4668 RJ |
800 | unsigned int virtqueue_get_vring_size(struct virtqueue *_vq) |
801 | { | |
802 | ||
803 | struct vring_virtqueue *vq = to_vvq(_vq); | |
804 | ||
805 | return vq->vring.num; | |
806 | } | |
807 | EXPORT_SYMBOL_GPL(virtqueue_get_vring_size); | |
808 | ||
b3b32c94 HG |
809 | bool virtqueue_is_broken(struct virtqueue *_vq) |
810 | { | |
811 | struct vring_virtqueue *vq = to_vvq(_vq); | |
812 | ||
813 | return vq->broken; | |
814 | } | |
815 | EXPORT_SYMBOL_GPL(virtqueue_is_broken); | |
816 | ||
e2dcdfe9 RR |
817 | /* |
818 | * This should prevent the device from being used, allowing drivers to | |
819 | * recover. You may need to grab appropriate locks to flush. | |
820 | */ | |
821 | void virtio_break_device(struct virtio_device *dev) | |
822 | { | |
823 | struct virtqueue *_vq; | |
824 | ||
825 | list_for_each_entry(_vq, &dev->vqs, list) { | |
826 | struct vring_virtqueue *vq = to_vvq(_vq); | |
827 | vq->broken = true; | |
828 | } | |
829 | } | |
830 | EXPORT_SYMBOL_GPL(virtio_break_device); | |
831 | ||
89062652 CH |
832 | void *virtqueue_get_avail(struct virtqueue *_vq) |
833 | { | |
834 | struct vring_virtqueue *vq = to_vvq(_vq); | |
835 | ||
836 | return vq->vring.avail; | |
837 | } | |
838 | EXPORT_SYMBOL_GPL(virtqueue_get_avail); | |
839 | ||
840 | void *virtqueue_get_used(struct virtqueue *_vq) | |
841 | { | |
842 | struct vring_virtqueue *vq = to_vvq(_vq); | |
843 | ||
844 | return vq->vring.used; | |
845 | } | |
846 | EXPORT_SYMBOL_GPL(virtqueue_get_used); | |
847 | ||
c6fd4701 | 848 | MODULE_LICENSE("GPL"); |