blob: a78ad459cc855b58f20a4a343277893b7694a14a [file] [log] [blame]
Rusty Russell0a8a69d2007-10-22 11:03:40 +10001/* 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>
Rusty Russelle34f8722008-07-25 12:06:13 -050021#include <linux/virtio_config.h>
Rusty Russell0a8a69d2007-10-22 11:03:40 +100022#include <linux/device.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090023#include <linux/slab.h>
Paul Gortmakerb5a2c4f2011-07-03 16:20:30 -040024#include <linux/module.h>
Rusty Russelle93300b2012-01-12 15:44:43 +103025#include <linux/hrtimer.h>
Rusty Russell0a8a69d2007-10-22 11:03:40 +100026
27#ifdef DEBUG
28/* For development, we want to crash whenever the ring is screwed. */
Rusty Russell9499f5e2009-06-12 22:16:35 -060029#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)
Rusty Russellc5f841f2009-03-30 21:55:22 -060035/* Caller is supposed to guarantee no reentry. */
36#define START_USE(_vq) \
37 do { \
38 if ((_vq)->in_use) \
Rusty Russell9499f5e2009-06-12 22:16:35 -060039 panic("%s:in_use = %i\n", \
40 (_vq)->vq.name, (_vq)->in_use); \
Rusty Russellc5f841f2009-03-30 21:55:22 -060041 (_vq)->in_use = __LINE__; \
Rusty Russell9499f5e2009-06-12 22:16:35 -060042 } while (0)
Roel Kluin3a35ce72009-01-22 16:42:57 +010043#define END_USE(_vq) \
Rusty Russell97a545a2010-02-24 14:22:22 -060044 do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0)
Rusty Russell0a8a69d2007-10-22 11:03:40 +100045#else
Rusty Russell9499f5e2009-06-12 22:16:35 -060046#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)
Rusty Russell0a8a69d2007-10-22 11:03:40 +100052#define START_USE(vq)
53#define END_USE(vq)
54#endif
55
56struct vring_virtqueue
57{
58 struct virtqueue vq;
59
60 /* Actual memory layout for this queue */
61 struct vring vring;
62
Rusty Russell7b21e342012-01-12 15:44:42 +103063 /* Can we use weak barriers? */
64 bool weak_barriers;
65
Rusty Russell0a8a69d2007-10-22 11:03:40 +100066 /* Other side has made a mess, don't try any more. */
67 bool broken;
68
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +010069 /* Host supports indirect buffers */
70 bool indirect;
71
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +030072 /* Host publishes avail event idx */
73 bool event;
74
Rusty Russell0a8a69d2007-10-22 11:03:40 +100075 /* 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. */
Anthony Liguori1bc49532007-11-07 15:49:24 -060081 u16 last_used_idx;
Rusty Russell0a8a69d2007-10-22 11:03:40 +100082
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;
Rusty Russelle93300b2012-01-12 15:44:43 +103089
90 /* Figure out if their kicks are too delayed. */
91 bool last_add_time_valid;
92 ktime_t last_add_time;
Rusty Russell0a8a69d2007-10-22 11:03:40 +100093#endif
94
95 /* Tokens for callbacks. */
96 void *data[];
97};
98
99#define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq)
100
Rusty Russell13816c72013-03-20 15:37:09 +1030101static inline struct scatterlist *sg_next_chained(struct scatterlist *sg,
102 unsigned int *count)
103{
104 return sg_next(sg);
105}
106
107static 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
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100115/* Set up an indirect table of descriptors and add it to the queue. */
Rusty Russell13816c72013-03-20 15:37:09 +1030116static 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)
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100126{
127 struct vring_desc *desc;
128 unsigned head;
Rusty Russell13816c72013-03-20 15:37:09 +1030129 struct scatterlist *sg;
130 int i, n;
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100131
Will Deaconb92b1b82012-10-19 14:03:33 +0100132 /*
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
Rusty Russell13816c72013-03-20 15:37:09 +1030139 desc = kmalloc(total_sg * sizeof(struct vring_desc), gfp);
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100140 if (!desc)
Michael S. Tsirkin686d3632010-06-10 18:16:11 +0300141 return -ENOMEM;
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100142
Rusty Russell13816c72013-03-20 15:37:09 +1030143 /* 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 }
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100153 }
Rusty Russell13816c72013-03-20 15:37:09 +1030154 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 }
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100162 }
Rusty Russell13816c72013-03-20 15:37:09 +1030163 BUG_ON(i != total_sg);
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100164
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 */
Rusty Russell06ca2872012-10-16 23:56:14 +1030170 vq->vq.num_free--;
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100171
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
Rusty Russell13816c72013-03-20 15:37:09 +1030184static 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)
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000194{
195 struct vring_virtqueue *vq = to_vvq(_vq);
Rusty Russell13816c72013-03-20 15:37:09 +1030196 struct scatterlist *sg;
197 unsigned int i, n, avail, uninitialized_var(prev), total_sg;
Michael S. Tsirkin1fe9b6f2010-07-26 16:55:30 +0930198 int head;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000199
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100200 START_USE(vq);
201
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000202 BUG_ON(data == NULL);
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100203
Rusty Russelle93300b2012-01-12 15:44:43 +1030204#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
Rusty Russell13816c72013-03-20 15:37:09 +1030217 total_sg = total_in + total_out;
218
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100219 /* If the host supports indirect descriptor tables, and we have multiple
220 * buffers, then go indirect. FIXME: tune this threshold */
Rusty Russell13816c72013-03-20 15:37:09 +1030221 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);
Michael S. Tsirkin1fe9b6f2010-07-26 16:55:30 +0930225 if (likely(head >= 0))
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100226 goto add_head;
227 }
228
Rusty Russell13816c72013-03-20 15:37:09 +1030229 BUG_ON(total_sg > vq->vring.num);
230 BUG_ON(total_sg == 0);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000231
Rusty Russell13816c72013-03-20 15:37:09 +1030232 if (vq->vq.num_free < total_sg) {
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000233 pr_debug("Can't add buf len %i - avail = %i\n",
Rusty Russell13816c72013-03-20 15:37:09 +1030234 total_sg, vq->vq.num_free);
Rusty Russell44653ea2008-07-25 12:06:04 -0500235 /* 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. */
Rusty Russell13816c72013-03-20 15:37:09 +1030238 if (out_sgs)
Rusty Russell44653ea2008-07-25 12:06:04 -0500239 vq->notify(&vq->vq);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000240 END_USE(vq);
241 return -ENOSPC;
242 }
243
244 /* We're about to use some buffers from the free list. */
Rusty Russell13816c72013-03-20 15:37:09 +1030245 vq->vq.num_free -= total_sg;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000246
Rusty Russell13816c72013-03-20 15:37:09 +1030247 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 }
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000256 }
Rusty Russell13816c72013-03-20 15:37:09 +1030257 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 }
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000265 }
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
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100272add_head:
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000273 /* Set token. */
274 vq->data[head] = data;
275
276 /* Put entry in available array (but don't update avail->idx until they
Rusty Russell3b720b82012-01-12 15:44:43 +1030277 * do sync). */
Rusty Russellee7cd892012-01-12 15:44:43 +1030278 avail = (vq->vring.avail->idx & (vq->vring.num-1));
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000279 vq->vring.avail->ring[avail] = head;
280
Rusty Russellee7cd892012-01-12 15:44:43 +1030281 /* Descriptors and available array need to be set before we expose the
282 * new available array entries. */
Rusty Russella9a0fef2013-03-18 13:22:19 +1030283 virtio_wmb(vq->weak_barriers);
Rusty Russellee7cd892012-01-12 15:44:43 +1030284 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
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000292 pr_debug("Added buffer head %i to %p\n", head, vq);
293 END_USE(vq);
Rusty Russell3c1b27d2009-09-23 22:26:31 -0600294
Rusty Russell98e8c6b2012-10-16 23:56:15 +1030295 return 0;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000296}
Rusty Russell13816c72013-03-20 15:37:09 +1030297
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 */
312int 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}
Rusty Russellf96fde42012-01-12 15:44:42 +1030327EXPORT_SYMBOL_GPL(virtqueue_add_buf);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000328
Rusty Russell5dfc1762012-01-12 15:44:42 +1030329/**
Rusty Russell13816c72013-03-20 15:37:09 +1030330 * 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 */
343int 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}
366EXPORT_SYMBOL_GPL(virtqueue_add_sgs);
367
368/**
Rusty Russell41f03772012-01-12 15:44:43 +1030369 * virtqueue_kick_prepare - first half of split virtqueue_kick call.
Rusty Russell5dfc1762012-01-12 15:44:42 +1030370 * @vq: the struct virtqueue
371 *
Rusty Russell41f03772012-01-12 15:44:43 +1030372 * Instead of virtqueue_kick(), you can do:
373 * if (virtqueue_kick_prepare(vq))
374 * virtqueue_notify(vq);
Rusty Russell5dfc1762012-01-12 15:44:42 +1030375 *
Rusty Russell41f03772012-01-12 15:44:43 +1030376 * This is sometimes useful because the virtqueue_kick_prepare() needs
377 * to be serialized, but the actual virtqueue_notify() call does not.
Rusty Russell5dfc1762012-01-12 15:44:42 +1030378 */
Rusty Russell41f03772012-01-12 15:44:43 +1030379bool virtqueue_kick_prepare(struct virtqueue *_vq)
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000380{
381 struct vring_virtqueue *vq = to_vvq(_vq);
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300382 u16 new, old;
Rusty Russell41f03772012-01-12 15:44:43 +1030383 bool needs_kick;
384
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000385 START_USE(vq);
Jason Wanga72caae2012-01-20 16:17:08 +0800386 /* We need to expose available array entries before checking avail
387 * event. */
Rusty Russella9a0fef2013-03-18 13:22:19 +1030388 virtio_mb(vq->weak_barriers);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000389
Rusty Russellee7cd892012-01-12 15:44:43 +1030390 old = vq->vring.avail->idx - vq->num_added;
391 new = vq->vring.avail->idx;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000392 vq->num_added = 0;
393
Rusty Russelle93300b2012-01-12 15:44:43 +1030394#ifdef DEBUG
395 if (vq->last_add_time_valid) {
396 WARN_ON(ktime_to_ms(ktime_sub(ktime_get(),
397 vq->last_add_time)) > 100);
398 }
399 vq->last_add_time_valid = false;
400#endif
401
Rusty Russell41f03772012-01-12 15:44:43 +1030402 if (vq->event) {
403 needs_kick = vring_need_event(vring_avail_event(&vq->vring),
404 new, old);
405 } else {
406 needs_kick = !(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY);
407 }
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000408 END_USE(vq);
Rusty Russell41f03772012-01-12 15:44:43 +1030409 return needs_kick;
410}
411EXPORT_SYMBOL_GPL(virtqueue_kick_prepare);
412
413/**
414 * virtqueue_notify - second half of split virtqueue_kick call.
415 * @vq: the struct virtqueue
416 *
417 * This does not need to be serialized.
418 */
419void virtqueue_notify(struct virtqueue *_vq)
420{
421 struct vring_virtqueue *vq = to_vvq(_vq);
422
423 /* Prod other side to tell it about changes. */
424 vq->notify(_vq);
425}
426EXPORT_SYMBOL_GPL(virtqueue_notify);
427
428/**
429 * virtqueue_kick - update after add_buf
430 * @vq: the struct virtqueue
431 *
432 * After one or more virtqueue_add_buf calls, invoke this to kick
433 * the other side.
434 *
435 * Caller must ensure we don't call this with other virtqueue
436 * operations at the same time (except where noted).
437 */
438void virtqueue_kick(struct virtqueue *vq)
439{
440 if (virtqueue_kick_prepare(vq))
441 virtqueue_notify(vq);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000442}
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300443EXPORT_SYMBOL_GPL(virtqueue_kick);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000444
445static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
446{
447 unsigned int i;
448
449 /* Clear data ptr. */
450 vq->data[head] = NULL;
451
452 /* Put back on free list: find end */
453 i = head;
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100454
455 /* Free the indirect table */
456 if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT)
457 kfree(phys_to_virt(vq->vring.desc[i].addr));
458
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000459 while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
460 i = vq->vring.desc[i].next;
Rusty Russell06ca2872012-10-16 23:56:14 +1030461 vq->vq.num_free++;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000462 }
463
464 vq->vring.desc[i].next = vq->free_head;
465 vq->free_head = head;
466 /* Plus final descriptor */
Rusty Russell06ca2872012-10-16 23:56:14 +1030467 vq->vq.num_free++;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000468}
469
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000470static inline bool more_used(const struct vring_virtqueue *vq)
471{
472 return vq->last_used_idx != vq->vring.used->idx;
473}
474
Rusty Russell5dfc1762012-01-12 15:44:42 +1030475/**
476 * virtqueue_get_buf - get the next used buffer
477 * @vq: the struct virtqueue we're talking about.
478 * @len: the length written into the buffer
479 *
480 * If the driver wrote data into the buffer, @len will be set to the
481 * amount written. This means you don't need to clear the buffer
482 * beforehand to ensure there's no data leakage in the case of short
483 * writes.
484 *
485 * Caller must ensure we don't call this with other virtqueue
486 * operations at the same time (except where noted).
487 *
488 * Returns NULL if there are no used buffers, or the "data" token
Rusty Russellf96fde42012-01-12 15:44:42 +1030489 * handed to virtqueue_add_buf().
Rusty Russell5dfc1762012-01-12 15:44:42 +1030490 */
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300491void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len)
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000492{
493 struct vring_virtqueue *vq = to_vvq(_vq);
494 void *ret;
495 unsigned int i;
Rusty Russell3b720b82012-01-12 15:44:43 +1030496 u16 last_used;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000497
498 START_USE(vq);
499
Rusty Russell5ef82752008-05-02 21:50:43 -0500500 if (unlikely(vq->broken)) {
501 END_USE(vq);
502 return NULL;
503 }
504
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000505 if (!more_used(vq)) {
506 pr_debug("No more buffers in queue\n");
507 END_USE(vq);
508 return NULL;
509 }
510
Michael S. Tsirkin2d61ba92009-10-25 15:28:53 +0200511 /* Only get used array entries after they have been exposed by host. */
Rusty Russella9a0fef2013-03-18 13:22:19 +1030512 virtio_rmb(vq->weak_barriers);
Michael S. Tsirkin2d61ba92009-10-25 15:28:53 +0200513
Rusty Russell3b720b82012-01-12 15:44:43 +1030514 last_used = (vq->last_used_idx & (vq->vring.num - 1));
515 i = vq->vring.used->ring[last_used].id;
516 *len = vq->vring.used->ring[last_used].len;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000517
518 if (unlikely(i >= vq->vring.num)) {
519 BAD_RING(vq, "id %u out of range\n", i);
520 return NULL;
521 }
522 if (unlikely(!vq->data[i])) {
523 BAD_RING(vq, "id %u is not a head!\n", i);
524 return NULL;
525 }
526
527 /* detach_buf clears data, so grab it now. */
528 ret = vq->data[i];
529 detach_buf(vq, i);
530 vq->last_used_idx++;
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300531 /* If we expect an interrupt for the next entry, tell host
532 * by writing event index and flush out the write before
533 * the read in the next get_buf call. */
534 if (!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) {
535 vring_used_event(&vq->vring) = vq->last_used_idx;
Rusty Russella9a0fef2013-03-18 13:22:19 +1030536 virtio_mb(vq->weak_barriers);
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300537 }
538
Rusty Russelle93300b2012-01-12 15:44:43 +1030539#ifdef DEBUG
540 vq->last_add_time_valid = false;
541#endif
542
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000543 END_USE(vq);
544 return ret;
545}
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300546EXPORT_SYMBOL_GPL(virtqueue_get_buf);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000547
Rusty Russell5dfc1762012-01-12 15:44:42 +1030548/**
549 * virtqueue_disable_cb - disable callbacks
550 * @vq: the struct virtqueue we're talking about.
551 *
552 * Note that this is not necessarily synchronous, hence unreliable and only
553 * useful as an optimization.
554 *
555 * Unlike other operations, this need not be serialized.
556 */
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300557void virtqueue_disable_cb(struct virtqueue *_vq)
Rusty Russell18445c42008-02-04 23:49:57 -0500558{
559 struct vring_virtqueue *vq = to_vvq(_vq);
560
Rusty Russell18445c42008-02-04 23:49:57 -0500561 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
Rusty Russell18445c42008-02-04 23:49:57 -0500562}
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300563EXPORT_SYMBOL_GPL(virtqueue_disable_cb);
Rusty Russell18445c42008-02-04 23:49:57 -0500564
Rusty Russell5dfc1762012-01-12 15:44:42 +1030565/**
566 * virtqueue_enable_cb - restart callbacks after disable_cb.
567 * @vq: the struct virtqueue we're talking about.
568 *
569 * This re-enables callbacks; it returns "false" if there are pending
570 * buffers in the queue, to detect a possible race between the driver
571 * checking for more work, and enabling callbacks.
572 *
573 * Caller must ensure we don't call this with other virtqueue
574 * operations at the same time (except where noted).
575 */
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300576bool virtqueue_enable_cb(struct virtqueue *_vq)
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000577{
578 struct vring_virtqueue *vq = to_vvq(_vq);
579
580 START_USE(vq);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000581
582 /* We optimistically turn back on interrupts, then check if there was
583 * more to do. */
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300584 /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to
585 * either clear the flags bit or point the event index at the next
586 * entry. Always do both to keep code simple. */
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000587 vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300588 vring_used_event(&vq->vring) = vq->last_used_idx;
Rusty Russella9a0fef2013-03-18 13:22:19 +1030589 virtio_mb(vq->weak_barriers);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000590 if (unlikely(more_used(vq))) {
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000591 END_USE(vq);
592 return false;
593 }
594
595 END_USE(vq);
596 return true;
597}
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300598EXPORT_SYMBOL_GPL(virtqueue_enable_cb);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000599
Rusty Russell5dfc1762012-01-12 15:44:42 +1030600/**
601 * virtqueue_enable_cb_delayed - restart callbacks after disable_cb.
602 * @vq: the struct virtqueue we're talking about.
603 *
604 * This re-enables callbacks but hints to the other side to delay
605 * interrupts until most of the available buffers have been processed;
606 * it returns "false" if there are many pending buffers in the queue,
607 * to detect a possible race between the driver checking for more work,
608 * and enabling callbacks.
609 *
610 * Caller must ensure we don't call this with other virtqueue
611 * operations at the same time (except where noted).
612 */
Michael S. Tsirkin7ab358c2011-05-20 02:11:14 +0300613bool virtqueue_enable_cb_delayed(struct virtqueue *_vq)
614{
615 struct vring_virtqueue *vq = to_vvq(_vq);
616 u16 bufs;
617
618 START_USE(vq);
619
620 /* We optimistically turn back on interrupts, then check if there was
621 * more to do. */
622 /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to
623 * either clear the flags bit or point the event index at the next
624 * entry. Always do both to keep code simple. */
625 vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
626 /* TODO: tune this threshold */
627 bufs = (u16)(vq->vring.avail->idx - vq->last_used_idx) * 3 / 4;
628 vring_used_event(&vq->vring) = vq->last_used_idx + bufs;
Rusty Russella9a0fef2013-03-18 13:22:19 +1030629 virtio_mb(vq->weak_barriers);
Michael S. Tsirkin7ab358c2011-05-20 02:11:14 +0300630 if (unlikely((u16)(vq->vring.used->idx - vq->last_used_idx) > bufs)) {
631 END_USE(vq);
632 return false;
633 }
634
635 END_USE(vq);
636 return true;
637}
638EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed);
639
Rusty Russell5dfc1762012-01-12 15:44:42 +1030640/**
641 * virtqueue_detach_unused_buf - detach first unused buffer
642 * @vq: the struct virtqueue we're talking about.
643 *
Rusty Russellf96fde42012-01-12 15:44:42 +1030644 * Returns NULL or the "data" token handed to virtqueue_add_buf().
Rusty Russell5dfc1762012-01-12 15:44:42 +1030645 * This is not valid on an active queue; it is useful only for device
646 * shutdown.
647 */
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300648void *virtqueue_detach_unused_buf(struct virtqueue *_vq)
Shirley Mac021eac2010-01-18 19:15:23 +0530649{
650 struct vring_virtqueue *vq = to_vvq(_vq);
651 unsigned int i;
652 void *buf;
653
654 START_USE(vq);
655
656 for (i = 0; i < vq->vring.num; i++) {
657 if (!vq->data[i])
658 continue;
659 /* detach_buf clears data, so grab it now. */
660 buf = vq->data[i];
661 detach_buf(vq, i);
Amit Shahb3258ff2011-03-16 19:12:10 +0530662 vq->vring.avail->idx--;
Shirley Mac021eac2010-01-18 19:15:23 +0530663 END_USE(vq);
664 return buf;
665 }
666 /* That should have freed everything. */
Rusty Russell06ca2872012-10-16 23:56:14 +1030667 BUG_ON(vq->vq.num_free != vq->vring.num);
Shirley Mac021eac2010-01-18 19:15:23 +0530668
669 END_USE(vq);
670 return NULL;
671}
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300672EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf);
Shirley Mac021eac2010-01-18 19:15:23 +0530673
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000674irqreturn_t vring_interrupt(int irq, void *_vq)
675{
676 struct vring_virtqueue *vq = to_vvq(_vq);
677
678 if (!more_used(vq)) {
679 pr_debug("virtqueue interrupt with no work for %p\n", vq);
680 return IRQ_NONE;
681 }
682
683 if (unlikely(vq->broken))
684 return IRQ_HANDLED;
685
686 pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
Rusty Russell18445c42008-02-04 23:49:57 -0500687 if (vq->vq.callback)
688 vq->vq.callback(&vq->vq);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000689
690 return IRQ_HANDLED;
691}
Rusty Russellc6fd4702008-02-04 23:50:05 -0500692EXPORT_SYMBOL_GPL(vring_interrupt);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000693
Jason Wang17bb6d42012-08-28 13:54:13 +0200694struct virtqueue *vring_new_virtqueue(unsigned int index,
695 unsigned int num,
Rusty Russell87c7d572008-12-30 09:26:03 -0600696 unsigned int vring_align,
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000697 struct virtio_device *vdev,
Rusty Russell7b21e342012-01-12 15:44:42 +1030698 bool weak_barriers,
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000699 void *pages,
700 void (*notify)(struct virtqueue *),
Rusty Russell9499f5e2009-06-12 22:16:35 -0600701 void (*callback)(struct virtqueue *),
702 const char *name)
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000703{
704 struct vring_virtqueue *vq;
705 unsigned int i;
706
Rusty Russell42b36cc2007-11-12 13:39:18 +1100707 /* We assume num is a power of 2. */
708 if (num & (num - 1)) {
709 dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num);
710 return NULL;
711 }
712
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000713 vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
714 if (!vq)
715 return NULL;
716
Rusty Russell87c7d572008-12-30 09:26:03 -0600717 vring_init(&vq->vring, num, pages, vring_align);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000718 vq->vq.callback = callback;
719 vq->vq.vdev = vdev;
Rusty Russell9499f5e2009-06-12 22:16:35 -0600720 vq->vq.name = name;
Rusty Russell06ca2872012-10-16 23:56:14 +1030721 vq->vq.num_free = num;
722 vq->vq.index = index;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000723 vq->notify = notify;
Rusty Russell7b21e342012-01-12 15:44:42 +1030724 vq->weak_barriers = weak_barriers;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000725 vq->broken = false;
726 vq->last_used_idx = 0;
727 vq->num_added = 0;
Rusty Russell9499f5e2009-06-12 22:16:35 -0600728 list_add_tail(&vq->vq.list, &vdev->vqs);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000729#ifdef DEBUG
730 vq->in_use = false;
Rusty Russelle93300b2012-01-12 15:44:43 +1030731 vq->last_add_time_valid = false;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000732#endif
733
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100734 vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC);
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300735 vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100736
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000737 /* No callback? Tell other side not to bother us. */
738 if (!callback)
739 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
740
741 /* Put everything in free lists. */
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000742 vq->free_head = 0;
Amit Shah3b870622010-02-12 10:32:14 +0530743 for (i = 0; i < num-1; i++) {
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000744 vq->vring.desc[i].next = i+1;
Amit Shah3b870622010-02-12 10:32:14 +0530745 vq->data[i] = NULL;
746 }
747 vq->data[i] = NULL;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000748
749 return &vq->vq;
750}
Rusty Russellc6fd4702008-02-04 23:50:05 -0500751EXPORT_SYMBOL_GPL(vring_new_virtqueue);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000752
753void vring_del_virtqueue(struct virtqueue *vq)
754{
Rusty Russell9499f5e2009-06-12 22:16:35 -0600755 list_del(&vq->list);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000756 kfree(to_vvq(vq));
757}
Rusty Russellc6fd4702008-02-04 23:50:05 -0500758EXPORT_SYMBOL_GPL(vring_del_virtqueue);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000759
Rusty Russelle34f8722008-07-25 12:06:13 -0500760/* Manipulates transport-specific feature bits. */
761void vring_transport_features(struct virtio_device *vdev)
762{
763 unsigned int i;
764
765 for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) {
766 switch (i) {
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100767 case VIRTIO_RING_F_INDIRECT_DESC:
768 break;
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300769 case VIRTIO_RING_F_EVENT_IDX:
770 break;
Rusty Russelle34f8722008-07-25 12:06:13 -0500771 default:
772 /* We don't understand this bit. */
773 clear_bit(i, vdev->features);
774 }
775 }
776}
777EXPORT_SYMBOL_GPL(vring_transport_features);
778
Rusty Russell5dfc1762012-01-12 15:44:42 +1030779/**
780 * virtqueue_get_vring_size - return the size of the virtqueue's vring
781 * @vq: the struct virtqueue containing the vring of interest.
782 *
783 * Returns the size of the vring. This is mainly used for boasting to
784 * userspace. Unlike other operations, this need not be serialized.
785 */
Rick Jones8f9f4662011-10-19 08:10:59 +0000786unsigned int virtqueue_get_vring_size(struct virtqueue *_vq)
787{
788
789 struct vring_virtqueue *vq = to_vvq(_vq);
790
791 return vq->vring.num;
792}
793EXPORT_SYMBOL_GPL(virtqueue_get_vring_size);
794
Rusty Russellc6fd4702008-02-04 23:50:05 -0500795MODULE_LICENSE("GPL");