blob: c70207478e572f2f855bcbb06481b5f9135f4527 [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/**
Rusty Russell13816c72013-03-20 15:37:09 +1030299 * virtqueue_add_sgs - expose buffers to other end
300 * @vq: the struct virtqueue we're talking about.
301 * @sgs: array of terminated scatterlists.
302 * @out_num: the number of scatterlists readable by other side
303 * @in_num: the number of scatterlists 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_sgs(struct virtqueue *_vq,
313 struct scatterlist *sgs[],
314 unsigned int out_sgs,
315 unsigned int in_sgs,
316 void *data,
317 gfp_t gfp)
318{
319 unsigned int i, total_out, total_in;
320
321 /* Count them first. */
322 for (i = total_out = total_in = 0; i < out_sgs; i++) {
323 struct scatterlist *sg;
324 for (sg = sgs[i]; sg; sg = sg_next(sg))
325 total_out++;
326 }
327 for (; i < out_sgs + in_sgs; i++) {
328 struct scatterlist *sg;
329 for (sg = sgs[i]; sg; sg = sg_next(sg))
330 total_in++;
331 }
332 return virtqueue_add(_vq, sgs, sg_next_chained,
333 total_out, total_in, out_sgs, in_sgs, data, gfp);
334}
335EXPORT_SYMBOL_GPL(virtqueue_add_sgs);
336
337/**
Rusty Russell282edb32013-03-20 15:44:26 +1030338 * virtqueue_add_outbuf - expose output buffers to other end
339 * @vq: the struct virtqueue we're talking about.
340 * @sgs: array of scatterlists (need not be terminated!)
341 * @num: the number of scatterlists readable by other side
342 * @data: the token identifying the buffer.
343 * @gfp: how to do memory allocations (if necessary).
344 *
345 * Caller must ensure we don't call this with other virtqueue operations
346 * at the same time (except where noted).
347 *
348 * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
349 */
350int virtqueue_add_outbuf(struct virtqueue *vq,
351 struct scatterlist sg[], unsigned int num,
352 void *data,
353 gfp_t gfp)
354{
355 return virtqueue_add(vq, &sg, sg_next_arr, num, 0, 1, 0, data, gfp);
356}
357EXPORT_SYMBOL_GPL(virtqueue_add_outbuf);
358
359/**
360 * virtqueue_add_inbuf - expose input buffers to other end
361 * @vq: the struct virtqueue we're talking about.
362 * @sgs: array of scatterlists (need not be terminated!)
363 * @num: the number of scatterlists writable by other side
364 * @data: the token identifying the buffer.
365 * @gfp: how to do memory allocations (if necessary).
366 *
367 * Caller must ensure we don't call this with other virtqueue operations
368 * at the same time (except where noted).
369 *
370 * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
371 */
372int virtqueue_add_inbuf(struct virtqueue *vq,
373 struct scatterlist sg[], unsigned int num,
374 void *data,
375 gfp_t gfp)
376{
377 return virtqueue_add(vq, &sg, sg_next_arr, 0, num, 0, 1, data, gfp);
378}
379EXPORT_SYMBOL_GPL(virtqueue_add_inbuf);
380
381/**
Rusty Russell41f03772012-01-12 15:44:43 +1030382 * virtqueue_kick_prepare - first half of split virtqueue_kick call.
Rusty Russell5dfc1762012-01-12 15:44:42 +1030383 * @vq: the struct virtqueue
384 *
Rusty Russell41f03772012-01-12 15:44:43 +1030385 * Instead of virtqueue_kick(), you can do:
386 * if (virtqueue_kick_prepare(vq))
387 * virtqueue_notify(vq);
Rusty Russell5dfc1762012-01-12 15:44:42 +1030388 *
Rusty Russell41f03772012-01-12 15:44:43 +1030389 * This is sometimes useful because the virtqueue_kick_prepare() needs
390 * to be serialized, but the actual virtqueue_notify() call does not.
Rusty Russell5dfc1762012-01-12 15:44:42 +1030391 */
Rusty Russell41f03772012-01-12 15:44:43 +1030392bool virtqueue_kick_prepare(struct virtqueue *_vq)
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000393{
394 struct vring_virtqueue *vq = to_vvq(_vq);
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300395 u16 new, old;
Rusty Russell41f03772012-01-12 15:44:43 +1030396 bool needs_kick;
397
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000398 START_USE(vq);
Jason Wanga72caae2012-01-20 16:17:08 +0800399 /* We need to expose available array entries before checking avail
400 * event. */
Rusty Russella9a0fef2013-03-18 13:22:19 +1030401 virtio_mb(vq->weak_barriers);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000402
Rusty Russellee7cd892012-01-12 15:44:43 +1030403 old = vq->vring.avail->idx - vq->num_added;
404 new = vq->vring.avail->idx;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000405 vq->num_added = 0;
406
Rusty Russelle93300b2012-01-12 15:44:43 +1030407#ifdef DEBUG
408 if (vq->last_add_time_valid) {
409 WARN_ON(ktime_to_ms(ktime_sub(ktime_get(),
410 vq->last_add_time)) > 100);
411 }
412 vq->last_add_time_valid = false;
413#endif
414
Rusty Russell41f03772012-01-12 15:44:43 +1030415 if (vq->event) {
416 needs_kick = vring_need_event(vring_avail_event(&vq->vring),
417 new, old);
418 } else {
419 needs_kick = !(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY);
420 }
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000421 END_USE(vq);
Rusty Russell41f03772012-01-12 15:44:43 +1030422 return needs_kick;
423}
424EXPORT_SYMBOL_GPL(virtqueue_kick_prepare);
425
426/**
427 * virtqueue_notify - second half of split virtqueue_kick call.
428 * @vq: the struct virtqueue
429 *
430 * This does not need to be serialized.
431 */
432void virtqueue_notify(struct virtqueue *_vq)
433{
434 struct vring_virtqueue *vq = to_vvq(_vq);
435
436 /* Prod other side to tell it about changes. */
437 vq->notify(_vq);
438}
439EXPORT_SYMBOL_GPL(virtqueue_notify);
440
441/**
442 * virtqueue_kick - update after add_buf
443 * @vq: the struct virtqueue
444 *
Rusty Russellb3087e42013-05-20 12:15:44 +0930445 * After one or more virtqueue_add_* calls, invoke this to kick
Rusty Russell41f03772012-01-12 15:44:43 +1030446 * the other side.
447 *
448 * Caller must ensure we don't call this with other virtqueue
449 * operations at the same time (except where noted).
450 */
451void virtqueue_kick(struct virtqueue *vq)
452{
453 if (virtqueue_kick_prepare(vq))
454 virtqueue_notify(vq);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000455}
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300456EXPORT_SYMBOL_GPL(virtqueue_kick);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000457
458static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
459{
460 unsigned int i;
461
462 /* Clear data ptr. */
463 vq->data[head] = NULL;
464
465 /* Put back on free list: find end */
466 i = head;
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100467
468 /* Free the indirect table */
469 if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT)
470 kfree(phys_to_virt(vq->vring.desc[i].addr));
471
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000472 while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
473 i = vq->vring.desc[i].next;
Rusty Russell06ca2872012-10-16 23:56:14 +1030474 vq->vq.num_free++;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000475 }
476
477 vq->vring.desc[i].next = vq->free_head;
478 vq->free_head = head;
479 /* Plus final descriptor */
Rusty Russell06ca2872012-10-16 23:56:14 +1030480 vq->vq.num_free++;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000481}
482
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000483static inline bool more_used(const struct vring_virtqueue *vq)
484{
485 return vq->last_used_idx != vq->vring.used->idx;
486}
487
Rusty Russell5dfc1762012-01-12 15:44:42 +1030488/**
489 * virtqueue_get_buf - get the next used buffer
490 * @vq: the struct virtqueue we're talking about.
491 * @len: the length written into the buffer
492 *
493 * If the driver wrote data into the buffer, @len will be set to the
494 * amount written. This means you don't need to clear the buffer
495 * beforehand to ensure there's no data leakage in the case of short
496 * writes.
497 *
498 * Caller must ensure we don't call this with other virtqueue
499 * operations at the same time (except where noted).
500 *
501 * Returns NULL if there are no used buffers, or the "data" token
Rusty Russellb3087e42013-05-20 12:15:44 +0930502 * handed to virtqueue_add_*().
Rusty Russell5dfc1762012-01-12 15:44:42 +1030503 */
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300504void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len)
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000505{
506 struct vring_virtqueue *vq = to_vvq(_vq);
507 void *ret;
508 unsigned int i;
Rusty Russell3b720b82012-01-12 15:44:43 +1030509 u16 last_used;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000510
511 START_USE(vq);
512
Rusty Russell5ef82752008-05-02 21:50:43 -0500513 if (unlikely(vq->broken)) {
514 END_USE(vq);
515 return NULL;
516 }
517
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000518 if (!more_used(vq)) {
519 pr_debug("No more buffers in queue\n");
520 END_USE(vq);
521 return NULL;
522 }
523
Michael S. Tsirkin2d61ba92009-10-25 15:28:53 +0200524 /* Only get used array entries after they have been exposed by host. */
Rusty Russella9a0fef2013-03-18 13:22:19 +1030525 virtio_rmb(vq->weak_barriers);
Michael S. Tsirkin2d61ba92009-10-25 15:28:53 +0200526
Rusty Russell3b720b82012-01-12 15:44:43 +1030527 last_used = (vq->last_used_idx & (vq->vring.num - 1));
528 i = vq->vring.used->ring[last_used].id;
529 *len = vq->vring.used->ring[last_used].len;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000530
531 if (unlikely(i >= vq->vring.num)) {
532 BAD_RING(vq, "id %u out of range\n", i);
533 return NULL;
534 }
535 if (unlikely(!vq->data[i])) {
536 BAD_RING(vq, "id %u is not a head!\n", i);
537 return NULL;
538 }
539
540 /* detach_buf clears data, so grab it now. */
541 ret = vq->data[i];
542 detach_buf(vq, i);
543 vq->last_used_idx++;
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300544 /* If we expect an interrupt for the next entry, tell host
545 * by writing event index and flush out the write before
546 * the read in the next get_buf call. */
547 if (!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) {
548 vring_used_event(&vq->vring) = vq->last_used_idx;
Rusty Russella9a0fef2013-03-18 13:22:19 +1030549 virtio_mb(vq->weak_barriers);
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300550 }
551
Rusty Russelle93300b2012-01-12 15:44:43 +1030552#ifdef DEBUG
553 vq->last_add_time_valid = false;
554#endif
555
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000556 END_USE(vq);
557 return ret;
558}
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300559EXPORT_SYMBOL_GPL(virtqueue_get_buf);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000560
Rusty Russell5dfc1762012-01-12 15:44:42 +1030561/**
562 * virtqueue_disable_cb - disable callbacks
563 * @vq: the struct virtqueue we're talking about.
564 *
565 * Note that this is not necessarily synchronous, hence unreliable and only
566 * useful as an optimization.
567 *
568 * Unlike other operations, this need not be serialized.
569 */
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300570void virtqueue_disable_cb(struct virtqueue *_vq)
Rusty Russell18445c42008-02-04 23:49:57 -0500571{
572 struct vring_virtqueue *vq = to_vvq(_vq);
573
Rusty Russell18445c42008-02-04 23:49:57 -0500574 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
Rusty Russell18445c42008-02-04 23:49:57 -0500575}
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300576EXPORT_SYMBOL_GPL(virtqueue_disable_cb);
Rusty Russell18445c42008-02-04 23:49:57 -0500577
Rusty Russell5dfc1762012-01-12 15:44:42 +1030578/**
579 * virtqueue_enable_cb - restart callbacks after disable_cb.
580 * @vq: the struct virtqueue we're talking about.
581 *
582 * This re-enables callbacks; it returns "false" if there are pending
583 * buffers in the queue, to detect a possible race between the driver
584 * checking for more work, and enabling callbacks.
585 *
586 * Caller must ensure we don't call this with other virtqueue
587 * operations at the same time (except where noted).
588 */
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300589bool virtqueue_enable_cb(struct virtqueue *_vq)
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000590{
591 struct vring_virtqueue *vq = to_vvq(_vq);
592
593 START_USE(vq);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000594
595 /* We optimistically turn back on interrupts, then check if there was
596 * more to do. */
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300597 /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to
598 * either clear the flags bit or point the event index at the next
599 * entry. Always do both to keep code simple. */
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000600 vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300601 vring_used_event(&vq->vring) = vq->last_used_idx;
Rusty Russella9a0fef2013-03-18 13:22:19 +1030602 virtio_mb(vq->weak_barriers);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000603 if (unlikely(more_used(vq))) {
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000604 END_USE(vq);
605 return false;
606 }
607
608 END_USE(vq);
609 return true;
610}
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300611EXPORT_SYMBOL_GPL(virtqueue_enable_cb);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000612
Rusty Russell5dfc1762012-01-12 15:44:42 +1030613/**
614 * virtqueue_enable_cb_delayed - restart callbacks after disable_cb.
615 * @vq: the struct virtqueue we're talking about.
616 *
617 * This re-enables callbacks but hints to the other side to delay
618 * interrupts until most of the available buffers have been processed;
619 * it returns "false" if there are many pending buffers in the queue,
620 * to detect a possible race between the driver checking for more work,
621 * and enabling callbacks.
622 *
623 * Caller must ensure we don't call this with other virtqueue
624 * operations at the same time (except where noted).
625 */
Michael S. Tsirkin7ab358c2011-05-20 02:11:14 +0300626bool virtqueue_enable_cb_delayed(struct virtqueue *_vq)
627{
628 struct vring_virtqueue *vq = to_vvq(_vq);
629 u16 bufs;
630
631 START_USE(vq);
632
633 /* We optimistically turn back on interrupts, then check if there was
634 * more to do. */
635 /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to
636 * either clear the flags bit or point the event index at the next
637 * entry. Always do both to keep code simple. */
638 vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
639 /* TODO: tune this threshold */
640 bufs = (u16)(vq->vring.avail->idx - vq->last_used_idx) * 3 / 4;
641 vring_used_event(&vq->vring) = vq->last_used_idx + bufs;
Rusty Russella9a0fef2013-03-18 13:22:19 +1030642 virtio_mb(vq->weak_barriers);
Michael S. Tsirkin7ab358c2011-05-20 02:11:14 +0300643 if (unlikely((u16)(vq->vring.used->idx - vq->last_used_idx) > bufs)) {
644 END_USE(vq);
645 return false;
646 }
647
648 END_USE(vq);
649 return true;
650}
651EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed);
652
Rusty Russell5dfc1762012-01-12 15:44:42 +1030653/**
654 * virtqueue_detach_unused_buf - detach first unused buffer
655 * @vq: the struct virtqueue we're talking about.
656 *
Rusty Russellb3087e42013-05-20 12:15:44 +0930657 * Returns NULL or the "data" token handed to virtqueue_add_*().
Rusty Russell5dfc1762012-01-12 15:44:42 +1030658 * This is not valid on an active queue; it is useful only for device
659 * shutdown.
660 */
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300661void *virtqueue_detach_unused_buf(struct virtqueue *_vq)
Shirley Mac021eac2010-01-18 19:15:23 +0530662{
663 struct vring_virtqueue *vq = to_vvq(_vq);
664 unsigned int i;
665 void *buf;
666
667 START_USE(vq);
668
669 for (i = 0; i < vq->vring.num; i++) {
670 if (!vq->data[i])
671 continue;
672 /* detach_buf clears data, so grab it now. */
673 buf = vq->data[i];
674 detach_buf(vq, i);
Amit Shahb3258ff2011-03-16 19:12:10 +0530675 vq->vring.avail->idx--;
Shirley Mac021eac2010-01-18 19:15:23 +0530676 END_USE(vq);
677 return buf;
678 }
679 /* That should have freed everything. */
Rusty Russell06ca2872012-10-16 23:56:14 +1030680 BUG_ON(vq->vq.num_free != vq->vring.num);
Shirley Mac021eac2010-01-18 19:15:23 +0530681
682 END_USE(vq);
683 return NULL;
684}
Michael S. Tsirkin7c5e9ed2010-04-12 16:19:07 +0300685EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf);
Shirley Mac021eac2010-01-18 19:15:23 +0530686
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000687irqreturn_t vring_interrupt(int irq, void *_vq)
688{
689 struct vring_virtqueue *vq = to_vvq(_vq);
690
691 if (!more_used(vq)) {
692 pr_debug("virtqueue interrupt with no work for %p\n", vq);
693 return IRQ_NONE;
694 }
695
696 if (unlikely(vq->broken))
697 return IRQ_HANDLED;
698
699 pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
Rusty Russell18445c42008-02-04 23:49:57 -0500700 if (vq->vq.callback)
701 vq->vq.callback(&vq->vq);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000702
703 return IRQ_HANDLED;
704}
Rusty Russellc6fd4702008-02-04 23:50:05 -0500705EXPORT_SYMBOL_GPL(vring_interrupt);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000706
Jason Wang17bb6d42012-08-28 13:54:13 +0200707struct virtqueue *vring_new_virtqueue(unsigned int index,
708 unsigned int num,
Rusty Russell87c7d572008-12-30 09:26:03 -0600709 unsigned int vring_align,
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000710 struct virtio_device *vdev,
Rusty Russell7b21e342012-01-12 15:44:42 +1030711 bool weak_barriers,
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000712 void *pages,
713 void (*notify)(struct virtqueue *),
Rusty Russell9499f5e2009-06-12 22:16:35 -0600714 void (*callback)(struct virtqueue *),
715 const char *name)
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000716{
717 struct vring_virtqueue *vq;
718 unsigned int i;
719
Rusty Russell42b36cc2007-11-12 13:39:18 +1100720 /* We assume num is a power of 2. */
721 if (num & (num - 1)) {
722 dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num);
723 return NULL;
724 }
725
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000726 vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
727 if (!vq)
728 return NULL;
729
Rusty Russell87c7d572008-12-30 09:26:03 -0600730 vring_init(&vq->vring, num, pages, vring_align);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000731 vq->vq.callback = callback;
732 vq->vq.vdev = vdev;
Rusty Russell9499f5e2009-06-12 22:16:35 -0600733 vq->vq.name = name;
Rusty Russell06ca2872012-10-16 23:56:14 +1030734 vq->vq.num_free = num;
735 vq->vq.index = index;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000736 vq->notify = notify;
Rusty Russell7b21e342012-01-12 15:44:42 +1030737 vq->weak_barriers = weak_barriers;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000738 vq->broken = false;
739 vq->last_used_idx = 0;
740 vq->num_added = 0;
Rusty Russell9499f5e2009-06-12 22:16:35 -0600741 list_add_tail(&vq->vq.list, &vdev->vqs);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000742#ifdef DEBUG
743 vq->in_use = false;
Rusty Russelle93300b2012-01-12 15:44:43 +1030744 vq->last_add_time_valid = false;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000745#endif
746
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100747 vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC);
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300748 vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100749
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000750 /* No callback? Tell other side not to bother us. */
751 if (!callback)
752 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
753
754 /* Put everything in free lists. */
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000755 vq->free_head = 0;
Amit Shah3b870622010-02-12 10:32:14 +0530756 for (i = 0; i < num-1; i++) {
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000757 vq->vring.desc[i].next = i+1;
Amit Shah3b870622010-02-12 10:32:14 +0530758 vq->data[i] = NULL;
759 }
760 vq->data[i] = NULL;
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000761
762 return &vq->vq;
763}
Rusty Russellc6fd4702008-02-04 23:50:05 -0500764EXPORT_SYMBOL_GPL(vring_new_virtqueue);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000765
766void vring_del_virtqueue(struct virtqueue *vq)
767{
Rusty Russell9499f5e2009-06-12 22:16:35 -0600768 list_del(&vq->list);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000769 kfree(to_vvq(vq));
770}
Rusty Russellc6fd4702008-02-04 23:50:05 -0500771EXPORT_SYMBOL_GPL(vring_del_virtqueue);
Rusty Russell0a8a69d2007-10-22 11:03:40 +1000772
Rusty Russelle34f8722008-07-25 12:06:13 -0500773/* Manipulates transport-specific feature bits. */
774void vring_transport_features(struct virtio_device *vdev)
775{
776 unsigned int i;
777
778 for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) {
779 switch (i) {
Mark McLoughlin9fa29b9d2009-05-11 18:11:45 +0100780 case VIRTIO_RING_F_INDIRECT_DESC:
781 break;
Michael S. Tsirkina5c262c2011-05-20 02:10:44 +0300782 case VIRTIO_RING_F_EVENT_IDX:
783 break;
Rusty Russelle34f8722008-07-25 12:06:13 -0500784 default:
785 /* We don't understand this bit. */
786 clear_bit(i, vdev->features);
787 }
788 }
789}
790EXPORT_SYMBOL_GPL(vring_transport_features);
791
Rusty Russell5dfc1762012-01-12 15:44:42 +1030792/**
793 * virtqueue_get_vring_size - return the size of the virtqueue's vring
794 * @vq: the struct virtqueue containing the vring of interest.
795 *
796 * Returns the size of the vring. This is mainly used for boasting to
797 * userspace. Unlike other operations, this need not be serialized.
798 */
Rick Jones8f9f4662011-10-19 08:10:59 +0000799unsigned int virtqueue_get_vring_size(struct virtqueue *_vq)
800{
801
802 struct vring_virtqueue *vq = to_vvq(_vq);
803
804 return vq->vring.num;
805}
806EXPORT_SYMBOL_GPL(virtqueue_get_vring_size);
807
Rusty Russellc6fd4702008-02-04 23:50:05 -0500808MODULE_LICENSE("GPL");