1 /* Virtio ring implementation.
3 * Copyright 2007 Rusty Russell IBM Corporation
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.
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.
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
19 #include <linux/virtio.h>
20 #include <linux/virtio_ring.h>
21 #include <linux/virtio_config.h>
22 #include <linux/device.h>
23 #include <linux/slab.h>
25 /* virtio guest is communicating with a virtual "device" that actually runs on
26 * a host processor. Memory barriers are used to control SMP effects. */
28 /* Where possible, use SMP barriers which are more lightweight than mandatory
29 * barriers, because mandatory barriers control MMIO effects on accesses
30 * through relaxed memory I/O windows (which virtio does not use). */
31 #define virtio_mb() smp_mb()
32 #define virtio_rmb() smp_rmb()
33 #define virtio_wmb() smp_wmb()
35 /* We must force memory ordering even if guest is UP since host could be
36 * running on another CPU, but SMP barriers are defined to barrier() in that
37 * configuration. So fall back to mandatory barriers instead. */
38 #define virtio_mb() mb()
39 #define virtio_rmb() rmb()
40 #define virtio_wmb() wmb()
44 /* For development, we want to crash whenever the ring is screwed. */
45 #define BAD_RING(_vq, fmt, args...) \
47 dev_err(&(_vq)->vq.vdev->dev, \
48 "%s:"fmt, (_vq)->vq.name, ##args); \
51 /* Caller is supposed to guarantee no reentry. */
52 #define START_USE(_vq) \
55 panic("%s:in_use = %i\n", \
56 (_vq)->vq.name, (_vq)->in_use); \
57 (_vq)->in_use = __LINE__; \
59 #define END_USE(_vq) \
60 do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0)
62 #define BAD_RING(_vq, fmt, args...) \
64 dev_err(&_vq->vq.vdev->dev, \
65 "%s:"fmt, (_vq)->vq.name, ##args); \
66 (_vq)->broken = true; \
72 struct vring_virtqueue
76 /* Actual memory layout for this queue */
79 /* Other side has made a mess, don't try any more. */
82 /* Host supports indirect buffers */
85 /* Number of free buffers */
86 unsigned int num_free
;
87 /* Head of free buffer list. */
88 unsigned int free_head
;
89 /* Number we've added since last sync. */
90 unsigned int num_added
;
92 /* Last used index we've seen. */
95 /* How to notify other side. FIXME: commonalize hcalls! */
96 void (*notify
)(struct virtqueue
*vq
);
99 /* They're supposed to lock for us. */
103 /* Tokens for callbacks. */
107 #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq)
109 /* Set up an indirect table of descriptors and add it to the queue. */
110 static int vring_add_indirect(struct vring_virtqueue
*vq
,
111 struct scatterlist sg
[],
116 struct vring_desc
*desc
;
120 desc
= kmalloc((out
+ in
) * sizeof(struct vring_desc
), gfp
);
122 return vq
->vring
.num
;
124 /* Transfer entries from the sg list into the indirect page */
125 for (i
= 0; i
< out
; i
++) {
126 desc
[i
].flags
= VRING_DESC_F_NEXT
;
127 desc
[i
].addr
= sg_phys(sg
);
128 desc
[i
].len
= sg
->length
;
132 for (; i
< (out
+ in
); i
++) {
133 desc
[i
].flags
= VRING_DESC_F_NEXT
|VRING_DESC_F_WRITE
;
134 desc
[i
].addr
= sg_phys(sg
);
135 desc
[i
].len
= sg
->length
;
140 /* Last one doesn't continue. */
141 desc
[i
-1].flags
&= ~VRING_DESC_F_NEXT
;
144 /* We're about to use a buffer */
147 /* Use a single buffer which doesn't continue */
148 head
= vq
->free_head
;
149 vq
->vring
.desc
[head
].flags
= VRING_DESC_F_INDIRECT
;
150 vq
->vring
.desc
[head
].addr
= virt_to_phys(desc
);
151 vq
->vring
.desc
[head
].len
= i
* sizeof(struct vring_desc
);
153 /* Update free pointer */
154 vq
->free_head
= vq
->vring
.desc
[head
].next
;
159 int virtqueue_add_buf_gfp(struct virtqueue
*_vq
,
160 struct scatterlist sg
[],
166 struct vring_virtqueue
*vq
= to_vvq(_vq
);
167 unsigned int i
, avail
, head
, uninitialized_var(prev
);
171 BUG_ON(data
== NULL
);
173 /* If the host supports indirect descriptor tables, and we have multiple
174 * buffers, then go indirect. FIXME: tune this threshold */
175 if (vq
->indirect
&& (out
+ in
) > 1 && vq
->num_free
) {
176 head
= vring_add_indirect(vq
, sg
, out
, in
, gfp
);
177 if (head
!= vq
->vring
.num
)
181 BUG_ON(out
+ in
> vq
->vring
.num
);
182 BUG_ON(out
+ in
== 0);
184 if (vq
->num_free
< out
+ in
) {
185 pr_debug("Can't add buf len %i - avail = %i\n",
186 out
+ in
, vq
->num_free
);
187 /* FIXME: for historical reasons, we force a notify here if
188 * there are outgoing parts to the buffer. Presumably the
189 * host should service the ring ASAP. */
196 /* We're about to use some buffers from the free list. */
197 vq
->num_free
-= out
+ in
;
199 head
= vq
->free_head
;
200 for (i
= vq
->free_head
; out
; i
= vq
->vring
.desc
[i
].next
, out
--) {
201 vq
->vring
.desc
[i
].flags
= VRING_DESC_F_NEXT
;
202 vq
->vring
.desc
[i
].addr
= sg_phys(sg
);
203 vq
->vring
.desc
[i
].len
= sg
->length
;
207 for (; in
; i
= vq
->vring
.desc
[i
].next
, in
--) {
208 vq
->vring
.desc
[i
].flags
= VRING_DESC_F_NEXT
|VRING_DESC_F_WRITE
;
209 vq
->vring
.desc
[i
].addr
= sg_phys(sg
);
210 vq
->vring
.desc
[i
].len
= sg
->length
;
214 /* Last one doesn't continue. */
215 vq
->vring
.desc
[prev
].flags
&= ~VRING_DESC_F_NEXT
;
217 /* Update free pointer */
222 vq
->data
[head
] = data
;
224 /* Put entry in available array (but don't update avail->idx until they
225 * do sync). FIXME: avoid modulus here? */
226 avail
= (vq
->vring
.avail
->idx
+ vq
->num_added
++) % vq
->vring
.num
;
227 vq
->vring
.avail
->ring
[avail
] = head
;
229 pr_debug("Added buffer head %i to %p\n", head
, vq
);
232 /* If we're indirect, we can fit many (assuming not OOM). */
234 return vq
->num_free
? vq
->vring
.num
: 0;
237 EXPORT_SYMBOL_GPL(virtqueue_add_buf_gfp
);
239 void virtqueue_kick(struct virtqueue
*_vq
)
241 struct vring_virtqueue
*vq
= to_vvq(_vq
);
243 /* Descriptors and available array need to be set before we expose the
244 * new available array entries. */
247 vq
->vring
.avail
->idx
+= vq
->num_added
;
250 /* Need to update avail index before checking if we should notify */
253 if (!(vq
->vring
.used
->flags
& VRING_USED_F_NO_NOTIFY
))
254 /* Prod other side to tell it about changes. */
259 EXPORT_SYMBOL_GPL(virtqueue_kick
);
261 static void detach_buf(struct vring_virtqueue
*vq
, unsigned int head
)
265 /* Clear data ptr. */
266 vq
->data
[head
] = NULL
;
268 /* Put back on free list: find end */
271 /* Free the indirect table */
272 if (vq
->vring
.desc
[i
].flags
& VRING_DESC_F_INDIRECT
)
273 kfree(phys_to_virt(vq
->vring
.desc
[i
].addr
));
275 while (vq
->vring
.desc
[i
].flags
& VRING_DESC_F_NEXT
) {
276 i
= vq
->vring
.desc
[i
].next
;
280 vq
->vring
.desc
[i
].next
= vq
->free_head
;
281 vq
->free_head
= head
;
282 /* Plus final descriptor */
286 static inline bool more_used(const struct vring_virtqueue
*vq
)
288 return vq
->last_used_idx
!= vq
->vring
.used
->idx
;
291 void *virtqueue_get_buf(struct virtqueue
*_vq
, unsigned int *len
)
293 struct vring_virtqueue
*vq
= to_vvq(_vq
);
299 if (unlikely(vq
->broken
)) {
304 if (!more_used(vq
)) {
305 pr_debug("No more buffers in queue\n");
310 /* Only get used array entries after they have been exposed by host. */
313 i
= vq
->vring
.used
->ring
[vq
->last_used_idx
%vq
->vring
.num
].id
;
314 *len
= vq
->vring
.used
->ring
[vq
->last_used_idx
%vq
->vring
.num
].len
;
316 if (unlikely(i
>= vq
->vring
.num
)) {
317 BAD_RING(vq
, "id %u out of range\n", i
);
320 if (unlikely(!vq
->data
[i
])) {
321 BAD_RING(vq
, "id %u is not a head!\n", i
);
325 /* detach_buf clears data, so grab it now. */
332 EXPORT_SYMBOL_GPL(virtqueue_get_buf
);
334 void virtqueue_disable_cb(struct virtqueue
*_vq
)
336 struct vring_virtqueue
*vq
= to_vvq(_vq
);
338 vq
->vring
.avail
->flags
|= VRING_AVAIL_F_NO_INTERRUPT
;
340 EXPORT_SYMBOL_GPL(virtqueue_disable_cb
);
342 bool virtqueue_enable_cb(struct virtqueue
*_vq
)
344 struct vring_virtqueue
*vq
= to_vvq(_vq
);
348 /* We optimistically turn back on interrupts, then check if there was
350 vq
->vring
.avail
->flags
&= ~VRING_AVAIL_F_NO_INTERRUPT
;
352 if (unlikely(more_used(vq
))) {
360 EXPORT_SYMBOL_GPL(virtqueue_enable_cb
);
362 void *virtqueue_detach_unused_buf(struct virtqueue
*_vq
)
364 struct vring_virtqueue
*vq
= to_vvq(_vq
);
370 for (i
= 0; i
< vq
->vring
.num
; i
++) {
373 /* detach_buf clears data, so grab it now. */
379 /* That should have freed everything. */
380 BUG_ON(vq
->num_free
!= vq
->vring
.num
);
385 EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf
);
387 irqreturn_t
vring_interrupt(int irq
, void *_vq
)
389 struct vring_virtqueue
*vq
= to_vvq(_vq
);
391 if (!more_used(vq
)) {
392 pr_debug("virtqueue interrupt with no work for %p\n", vq
);
396 if (unlikely(vq
->broken
))
399 pr_debug("virtqueue callback for %p (%p)\n", vq
, vq
->vq
.callback
);
401 vq
->vq
.callback(&vq
->vq
);
405 EXPORT_SYMBOL_GPL(vring_interrupt
);
407 struct virtqueue
*vring_new_virtqueue(unsigned int num
,
408 unsigned int vring_align
,
409 struct virtio_device
*vdev
,
411 void (*notify
)(struct virtqueue
*),
412 void (*callback
)(struct virtqueue
*),
415 struct vring_virtqueue
*vq
;
418 /* We assume num is a power of 2. */
419 if (num
& (num
- 1)) {
420 dev_warn(&vdev
->dev
, "Bad virtqueue length %u\n", num
);
424 vq
= kmalloc(sizeof(*vq
) + sizeof(void *)*num
, GFP_KERNEL
);
428 vring_init(&vq
->vring
, num
, pages
, vring_align
);
429 vq
->vq
.callback
= callback
;
434 vq
->last_used_idx
= 0;
436 list_add_tail(&vq
->vq
.list
, &vdev
->vqs
);
441 vq
->indirect
= virtio_has_feature(vdev
, VIRTIO_RING_F_INDIRECT_DESC
);
443 /* No callback? Tell other side not to bother us. */
445 vq
->vring
.avail
->flags
|= VRING_AVAIL_F_NO_INTERRUPT
;
447 /* Put everything in free lists. */
450 for (i
= 0; i
< num
-1; i
++) {
451 vq
->vring
.desc
[i
].next
= i
+1;
458 EXPORT_SYMBOL_GPL(vring_new_virtqueue
);
460 void vring_del_virtqueue(struct virtqueue
*vq
)
465 EXPORT_SYMBOL_GPL(vring_del_virtqueue
);
467 /* Manipulates transport-specific feature bits. */
468 void vring_transport_features(struct virtio_device
*vdev
)
472 for (i
= VIRTIO_TRANSPORT_F_START
; i
< VIRTIO_TRANSPORT_F_END
; i
++) {
474 case VIRTIO_RING_F_INDIRECT_DESC
:
477 /* We don't understand this bit. */
478 clear_bit(i
, vdev
->features
);
482 EXPORT_SYMBOL_GPL(vring_transport_features
);
484 MODULE_LICENSE("GPL");