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);
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, uninitialized_var(prev);
172 BUG_ON(data == NULL);
174 /* If the host supports indirect descriptor tables, and we have multiple
175 * buffers, then go indirect. FIXME: tune this threshold */
176 if (vq->indirect && (out + in) > 1 && vq->num_free) {
177 head = vring_add_indirect(vq, sg, out, in, gfp);
178 if (likely(head >= 0))
182 BUG_ON(out + in > vq->vring.num);
183 BUG_ON(out + in == 0);
185 if (vq->num_free < out + in) {
186 pr_debug("Can't add buf len %i - avail = %i\n",
187 out + in, vq->num_free);
188 /* FIXME: for historical reasons, we force a notify here if
189 * there are outgoing parts to the buffer. Presumably the
190 * host should service the ring ASAP. */
197 /* We're about to use some buffers from the free list. */
198 vq->num_free -= out + in;
200 head = vq->free_head;
201 for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) {
202 vq->vring.desc[i].flags = VRING_DESC_F_NEXT;
203 vq->vring.desc[i].addr = sg_phys(sg);
204 vq->vring.desc[i].len = sg->length;
208 for (; in; i = vq->vring.desc[i].next, in--) {
209 vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
210 vq->vring.desc[i].addr = sg_phys(sg);
211 vq->vring.desc[i].len = sg->length;
215 /* Last one doesn't continue. */
216 vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT;
218 /* Update free pointer */
223 vq->data[head] = data;
225 /* Put entry in available array (but don't update avail->idx until they
226 * do sync). FIXME: avoid modulus here? */
227 avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num;
228 vq->vring.avail->ring[avail] = head;
230 pr_debug("Added buffer head %i to %p\n", head, vq);
233 /* If we're indirect, we can fit many (assuming not OOM). */
235 return vq->num_free ? vq->vring.num : 0;
238 EXPORT_SYMBOL_GPL(virtqueue_add_buf_gfp);
240 void virtqueue_kick(struct virtqueue *_vq)
242 struct vring_virtqueue *vq = to_vvq(_vq);
244 /* Descriptors and available array need to be set before we expose the
245 * new available array entries. */
248 vq->vring.avail->idx += vq->num_added;
251 /* Need to update avail index before checking if we should notify */
254 if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY))
255 /* Prod other side to tell it about changes. */
260 EXPORT_SYMBOL_GPL(virtqueue_kick);
262 static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
266 /* Clear data ptr. */
267 vq->data[head] = NULL;
269 /* Put back on free list: find end */
272 /* Free the indirect table */
273 if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT)
274 kfree(phys_to_virt(vq->vring.desc[i].addr));
276 while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
277 i = vq->vring.desc[i].next;
281 vq->vring.desc[i].next = vq->free_head;
282 vq->free_head = head;
283 /* Plus final descriptor */
287 static inline bool more_used(const struct vring_virtqueue *vq)
289 return vq->last_used_idx != vq->vring.used->idx;
292 void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len)
294 struct vring_virtqueue *vq = to_vvq(_vq);
300 if (unlikely(vq->broken)) {
305 if (!more_used(vq)) {
306 pr_debug("No more buffers in queue\n");
311 /* Only get used array entries after they have been exposed by host. */
314 i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id;
315 *len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len;
317 if (unlikely(i >= vq->vring.num)) {
318 BAD_RING(vq, "id %u out of range\n", i);
321 if (unlikely(!vq->data[i])) {
322 BAD_RING(vq, "id %u is not a head!\n", i);
326 /* detach_buf clears data, so grab it now. */
333 EXPORT_SYMBOL_GPL(virtqueue_get_buf);
335 void virtqueue_disable_cb(struct virtqueue *_vq)
337 struct vring_virtqueue *vq = to_vvq(_vq);
339 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
341 EXPORT_SYMBOL_GPL(virtqueue_disable_cb);
343 bool virtqueue_enable_cb(struct virtqueue *_vq)
345 struct vring_virtqueue *vq = to_vvq(_vq);
349 /* We optimistically turn back on interrupts, then check if there was
351 vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
353 if (unlikely(more_used(vq))) {
361 EXPORT_SYMBOL_GPL(virtqueue_enable_cb);
363 void *virtqueue_detach_unused_buf(struct virtqueue *_vq)
365 struct vring_virtqueue *vq = to_vvq(_vq);
371 for (i = 0; i < vq->vring.num; i++) {
374 /* detach_buf clears data, so grab it now. */
380 /* That should have freed everything. */
381 BUG_ON(vq->num_free != vq->vring.num);
386 EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf);
388 irqreturn_t vring_interrupt(int irq, void *_vq)
390 struct vring_virtqueue *vq = to_vvq(_vq);
392 if (!more_used(vq)) {
393 pr_debug("virtqueue interrupt with no work for %p\n", vq);
397 if (unlikely(vq->broken))
400 pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
402 vq->vq.callback(&vq->vq);
406 EXPORT_SYMBOL_GPL(vring_interrupt);
408 struct virtqueue *vring_new_virtqueue(unsigned int num,
409 unsigned int vring_align,
410 struct virtio_device *vdev,
412 void (*notify)(struct virtqueue *),
413 void (*callback)(struct virtqueue *),
416 struct vring_virtqueue *vq;
419 /* We assume num is a power of 2. */
420 if (num & (num - 1)) {
421 dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num);
425 vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
429 vring_init(&vq->vring, num, pages, vring_align);
430 vq->vq.callback = callback;
435 vq->last_used_idx = 0;
437 list_add_tail(&vq->vq.list, &vdev->vqs);
442 vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC);
444 /* No callback? Tell other side not to bother us. */
446 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
448 /* Put everything in free lists. */
451 for (i = 0; i < num-1; i++) {
452 vq->vring.desc[i].next = i+1;
459 EXPORT_SYMBOL_GPL(vring_new_virtqueue);
461 void vring_del_virtqueue(struct virtqueue *vq)
466 EXPORT_SYMBOL_GPL(vring_del_virtqueue);
468 /* Manipulates transport-specific feature bits. */
469 void vring_transport_features(struct virtio_device *vdev)
473 for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) {
475 case VIRTIO_RING_F_INDIRECT_DESC:
478 /* We don't understand this bit. */
479 clear_bit(i, vdev->features);
483 EXPORT_SYMBOL_GPL(vring_transport_features);
485 MODULE_LICENSE("GPL");