xref: /openbmc/linux/drivers/virtio/virtio_ring.c (revision 1bc4953ed44454c7f53d0b609445d1534981ee75)
10a8a69ddSRusty Russell /* Virtio ring implementation.
20a8a69ddSRusty Russell  *
30a8a69ddSRusty Russell  *  Copyright 2007 Rusty Russell IBM Corporation
40a8a69ddSRusty Russell  *
50a8a69ddSRusty Russell  *  This program is free software; you can redistribute it and/or modify
60a8a69ddSRusty Russell  *  it under the terms of the GNU General Public License as published by
70a8a69ddSRusty Russell  *  the Free Software Foundation; either version 2 of the License, or
80a8a69ddSRusty Russell  *  (at your option) any later version.
90a8a69ddSRusty Russell  *
100a8a69ddSRusty Russell  *  This program is distributed in the hope that it will be useful,
110a8a69ddSRusty Russell  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
120a8a69ddSRusty Russell  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
130a8a69ddSRusty Russell  *  GNU General Public License for more details.
140a8a69ddSRusty Russell  *
150a8a69ddSRusty Russell  *  You should have received a copy of the GNU General Public License
160a8a69ddSRusty Russell  *  along with this program; if not, write to the Free Software
170a8a69ddSRusty Russell  *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
180a8a69ddSRusty Russell  */
190a8a69ddSRusty Russell #include <linux/virtio.h>
200a8a69ddSRusty Russell #include <linux/virtio_ring.h>
210a8a69ddSRusty Russell #include <linux/device.h>
220a8a69ddSRusty Russell 
230a8a69ddSRusty Russell #ifdef DEBUG
240a8a69ddSRusty Russell /* For development, we want to crash whenever the ring is screwed. */
250a8a69ddSRusty Russell #define BAD_RING(vq, fmt...)			\
260a8a69ddSRusty Russell 	do { dev_err(&vq->vq.vdev->dev, fmt); BUG(); } while(0)
270a8a69ddSRusty Russell #define START_USE(vq) \
280a8a69ddSRusty Russell 	do { if ((vq)->in_use) panic("in_use = %i\n", (vq)->in_use); (vq)->in_use = __LINE__; mb(); } while(0)
290a8a69ddSRusty Russell #define END_USE(vq) \
300a8a69ddSRusty Russell 	do { BUG_ON(!(vq)->in_use); (vq)->in_use = 0; mb(); } while(0)
310a8a69ddSRusty Russell #else
320a8a69ddSRusty Russell #define BAD_RING(vq, fmt...)			\
330a8a69ddSRusty Russell 	do { dev_err(&vq->vq.vdev->dev, fmt); (vq)->broken = true; } while(0)
340a8a69ddSRusty Russell #define START_USE(vq)
350a8a69ddSRusty Russell #define END_USE(vq)
360a8a69ddSRusty Russell #endif
370a8a69ddSRusty Russell 
380a8a69ddSRusty Russell struct vring_virtqueue
390a8a69ddSRusty Russell {
400a8a69ddSRusty Russell 	struct virtqueue vq;
410a8a69ddSRusty Russell 
420a8a69ddSRusty Russell 	/* Actual memory layout for this queue */
430a8a69ddSRusty Russell 	struct vring vring;
440a8a69ddSRusty Russell 
450a8a69ddSRusty Russell 	/* Other side has made a mess, don't try any more. */
460a8a69ddSRusty Russell 	bool broken;
470a8a69ddSRusty Russell 
480a8a69ddSRusty Russell 	/* Number of free buffers */
490a8a69ddSRusty Russell 	unsigned int num_free;
500a8a69ddSRusty Russell 	/* Head of free buffer list. */
510a8a69ddSRusty Russell 	unsigned int free_head;
520a8a69ddSRusty Russell 	/* Number we've added since last sync. */
530a8a69ddSRusty Russell 	unsigned int num_added;
540a8a69ddSRusty Russell 
550a8a69ddSRusty Russell 	/* Last used index we've seen. */
56*1bc4953eSAnthony Liguori 	u16 last_used_idx;
570a8a69ddSRusty Russell 
580a8a69ddSRusty Russell 	/* How to notify other side. FIXME: commonalize hcalls! */
590a8a69ddSRusty Russell 	void (*notify)(struct virtqueue *vq);
600a8a69ddSRusty Russell 
610a8a69ddSRusty Russell #ifdef DEBUG
620a8a69ddSRusty Russell 	/* They're supposed to lock for us. */
630a8a69ddSRusty Russell 	unsigned int in_use;
640a8a69ddSRusty Russell #endif
650a8a69ddSRusty Russell 
660a8a69ddSRusty Russell 	/* Tokens for callbacks. */
670a8a69ddSRusty Russell 	void *data[];
680a8a69ddSRusty Russell };
690a8a69ddSRusty Russell 
700a8a69ddSRusty Russell #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq)
710a8a69ddSRusty Russell 
720a8a69ddSRusty Russell static int vring_add_buf(struct virtqueue *_vq,
730a8a69ddSRusty Russell 			 struct scatterlist sg[],
740a8a69ddSRusty Russell 			 unsigned int out,
750a8a69ddSRusty Russell 			 unsigned int in,
760a8a69ddSRusty Russell 			 void *data)
770a8a69ddSRusty Russell {
780a8a69ddSRusty Russell 	struct vring_virtqueue *vq = to_vvq(_vq);
790a8a69ddSRusty Russell 	unsigned int i, avail, head, uninitialized_var(prev);
800a8a69ddSRusty Russell 
810a8a69ddSRusty Russell 	BUG_ON(data == NULL);
820a8a69ddSRusty Russell 	BUG_ON(out + in > vq->vring.num);
830a8a69ddSRusty Russell 	BUG_ON(out + in == 0);
840a8a69ddSRusty Russell 
850a8a69ddSRusty Russell 	START_USE(vq);
860a8a69ddSRusty Russell 
870a8a69ddSRusty Russell 	if (vq->num_free < out + in) {
880a8a69ddSRusty Russell 		pr_debug("Can't add buf len %i - avail = %i\n",
890a8a69ddSRusty Russell 			 out + in, vq->num_free);
900a8a69ddSRusty Russell 		END_USE(vq);
910a8a69ddSRusty Russell 		return -ENOSPC;
920a8a69ddSRusty Russell 	}
930a8a69ddSRusty Russell 
940a8a69ddSRusty Russell 	/* We're about to use some buffers from the free list. */
950a8a69ddSRusty Russell 	vq->num_free -= out + in;
960a8a69ddSRusty Russell 
970a8a69ddSRusty Russell 	head = vq->free_head;
980a8a69ddSRusty Russell 	for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) {
990a8a69ddSRusty Russell 		vq->vring.desc[i].flags = VRING_DESC_F_NEXT;
1000a8a69ddSRusty Russell 		vq->vring.desc[i].addr = (page_to_pfn(sg_page(sg))<<PAGE_SHIFT)
1010a8a69ddSRusty Russell 			+ sg->offset;
1020a8a69ddSRusty Russell 		vq->vring.desc[i].len = sg->length;
1030a8a69ddSRusty Russell 		prev = i;
1040a8a69ddSRusty Russell 		sg++;
1050a8a69ddSRusty Russell 	}
1060a8a69ddSRusty Russell 	for (; in; i = vq->vring.desc[i].next, in--) {
1070a8a69ddSRusty Russell 		vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
1080a8a69ddSRusty Russell 		vq->vring.desc[i].addr = (page_to_pfn(sg_page(sg))<<PAGE_SHIFT)
1090a8a69ddSRusty Russell 			+ sg->offset;
1100a8a69ddSRusty Russell 		vq->vring.desc[i].len = sg->length;
1110a8a69ddSRusty Russell 		prev = i;
1120a8a69ddSRusty Russell 		sg++;
1130a8a69ddSRusty Russell 	}
1140a8a69ddSRusty Russell 	/* Last one doesn't continue. */
1150a8a69ddSRusty Russell 	vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT;
1160a8a69ddSRusty Russell 
1170a8a69ddSRusty Russell 	/* Update free pointer */
1180a8a69ddSRusty Russell 	vq->free_head = i;
1190a8a69ddSRusty Russell 
1200a8a69ddSRusty Russell 	/* Set token. */
1210a8a69ddSRusty Russell 	vq->data[head] = data;
1220a8a69ddSRusty Russell 
1230a8a69ddSRusty Russell 	/* Put entry in available array (but don't update avail->idx until they
1240a8a69ddSRusty Russell 	 * do sync).  FIXME: avoid modulus here? */
1250a8a69ddSRusty Russell 	avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num;
1260a8a69ddSRusty Russell 	vq->vring.avail->ring[avail] = head;
1270a8a69ddSRusty Russell 
1280a8a69ddSRusty Russell 	pr_debug("Added buffer head %i to %p\n", head, vq);
1290a8a69ddSRusty Russell 	END_USE(vq);
1300a8a69ddSRusty Russell 	return 0;
1310a8a69ddSRusty Russell }
1320a8a69ddSRusty Russell 
1330a8a69ddSRusty Russell static void vring_kick(struct virtqueue *_vq)
1340a8a69ddSRusty Russell {
1350a8a69ddSRusty Russell 	struct vring_virtqueue *vq = to_vvq(_vq);
1360a8a69ddSRusty Russell 	START_USE(vq);
1370a8a69ddSRusty Russell 	/* Descriptors and available array need to be set before we expose the
1380a8a69ddSRusty Russell 	 * new available array entries. */
1390a8a69ddSRusty Russell 	wmb();
1400a8a69ddSRusty Russell 
1410a8a69ddSRusty Russell 	vq->vring.avail->idx += vq->num_added;
1420a8a69ddSRusty Russell 	vq->num_added = 0;
1430a8a69ddSRusty Russell 
1440a8a69ddSRusty Russell 	/* Need to update avail index before checking if we should notify */
1450a8a69ddSRusty Russell 	mb();
1460a8a69ddSRusty Russell 
1470a8a69ddSRusty Russell 	if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY))
1480a8a69ddSRusty Russell 		/* Prod other side to tell it about changes. */
1490a8a69ddSRusty Russell 		vq->notify(&vq->vq);
1500a8a69ddSRusty Russell 
1510a8a69ddSRusty Russell 	END_USE(vq);
1520a8a69ddSRusty Russell }
1530a8a69ddSRusty Russell 
1540a8a69ddSRusty Russell static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
1550a8a69ddSRusty Russell {
1560a8a69ddSRusty Russell 	unsigned int i;
1570a8a69ddSRusty Russell 
1580a8a69ddSRusty Russell 	/* Clear data ptr. */
1590a8a69ddSRusty Russell 	vq->data[head] = NULL;
1600a8a69ddSRusty Russell 
1610a8a69ddSRusty Russell 	/* Put back on free list: find end */
1620a8a69ddSRusty Russell 	i = head;
1630a8a69ddSRusty Russell 	while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
1640a8a69ddSRusty Russell 		i = vq->vring.desc[i].next;
1650a8a69ddSRusty Russell 		vq->num_free++;
1660a8a69ddSRusty Russell 	}
1670a8a69ddSRusty Russell 
1680a8a69ddSRusty Russell 	vq->vring.desc[i].next = vq->free_head;
1690a8a69ddSRusty Russell 	vq->free_head = head;
1700a8a69ddSRusty Russell 	/* Plus final descriptor */
1710a8a69ddSRusty Russell 	vq->num_free++;
1720a8a69ddSRusty Russell }
1730a8a69ddSRusty Russell 
1740a8a69ddSRusty Russell /* FIXME: We need to tell other side about removal, to synchronize. */
1750a8a69ddSRusty Russell static void vring_shutdown(struct virtqueue *_vq)
1760a8a69ddSRusty Russell {
1770a8a69ddSRusty Russell 	struct vring_virtqueue *vq = to_vvq(_vq);
1780a8a69ddSRusty Russell 	unsigned int i;
1790a8a69ddSRusty Russell 
1800a8a69ddSRusty Russell 	for (i = 0; i < vq->vring.num; i++)
1810a8a69ddSRusty Russell 		detach_buf(vq, i);
1820a8a69ddSRusty Russell }
1830a8a69ddSRusty Russell 
1840a8a69ddSRusty Russell static inline bool more_used(const struct vring_virtqueue *vq)
1850a8a69ddSRusty Russell {
1860a8a69ddSRusty Russell 	return vq->last_used_idx != vq->vring.used->idx;
1870a8a69ddSRusty Russell }
1880a8a69ddSRusty Russell 
1890a8a69ddSRusty Russell static void *vring_get_buf(struct virtqueue *_vq, unsigned int *len)
1900a8a69ddSRusty Russell {
1910a8a69ddSRusty Russell 	struct vring_virtqueue *vq = to_vvq(_vq);
1920a8a69ddSRusty Russell 	void *ret;
1930a8a69ddSRusty Russell 	unsigned int i;
1940a8a69ddSRusty Russell 
1950a8a69ddSRusty Russell 	START_USE(vq);
1960a8a69ddSRusty Russell 
1970a8a69ddSRusty Russell 	if (!more_used(vq)) {
1980a8a69ddSRusty Russell 		pr_debug("No more buffers in queue\n");
1990a8a69ddSRusty Russell 		END_USE(vq);
2000a8a69ddSRusty Russell 		return NULL;
2010a8a69ddSRusty Russell 	}
2020a8a69ddSRusty Russell 
2030a8a69ddSRusty Russell 	i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id;
2040a8a69ddSRusty Russell 	*len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len;
2050a8a69ddSRusty Russell 
2060a8a69ddSRusty Russell 	if (unlikely(i >= vq->vring.num)) {
2070a8a69ddSRusty Russell 		BAD_RING(vq, "id %u out of range\n", i);
2080a8a69ddSRusty Russell 		return NULL;
2090a8a69ddSRusty Russell 	}
2100a8a69ddSRusty Russell 	if (unlikely(!vq->data[i])) {
2110a8a69ddSRusty Russell 		BAD_RING(vq, "id %u is not a head!\n", i);
2120a8a69ddSRusty Russell 		return NULL;
2130a8a69ddSRusty Russell 	}
2140a8a69ddSRusty Russell 
2150a8a69ddSRusty Russell 	/* detach_buf clears data, so grab it now. */
2160a8a69ddSRusty Russell 	ret = vq->data[i];
2170a8a69ddSRusty Russell 	detach_buf(vq, i);
2180a8a69ddSRusty Russell 	vq->last_used_idx++;
2190a8a69ddSRusty Russell 	END_USE(vq);
2200a8a69ddSRusty Russell 	return ret;
2210a8a69ddSRusty Russell }
2220a8a69ddSRusty Russell 
2230a8a69ddSRusty Russell static bool vring_restart(struct virtqueue *_vq)
2240a8a69ddSRusty Russell {
2250a8a69ddSRusty Russell 	struct vring_virtqueue *vq = to_vvq(_vq);
2260a8a69ddSRusty Russell 
2270a8a69ddSRusty Russell 	START_USE(vq);
2280a8a69ddSRusty Russell 	BUG_ON(!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT));
2290a8a69ddSRusty Russell 
2300a8a69ddSRusty Russell 	/* We optimistically turn back on interrupts, then check if there was
2310a8a69ddSRusty Russell 	 * more to do. */
2320a8a69ddSRusty Russell 	vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
2330a8a69ddSRusty Russell 	mb();
2340a8a69ddSRusty Russell 	if (unlikely(more_used(vq))) {
2350a8a69ddSRusty Russell 		vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
2360a8a69ddSRusty Russell 		END_USE(vq);
2370a8a69ddSRusty Russell 		return false;
2380a8a69ddSRusty Russell 	}
2390a8a69ddSRusty Russell 
2400a8a69ddSRusty Russell 	END_USE(vq);
2410a8a69ddSRusty Russell 	return true;
2420a8a69ddSRusty Russell }
2430a8a69ddSRusty Russell 
2440a8a69ddSRusty Russell irqreturn_t vring_interrupt(int irq, void *_vq)
2450a8a69ddSRusty Russell {
2460a8a69ddSRusty Russell 	struct vring_virtqueue *vq = to_vvq(_vq);
2470a8a69ddSRusty Russell 
2480a8a69ddSRusty Russell 	if (!more_used(vq)) {
2490a8a69ddSRusty Russell 		pr_debug("virtqueue interrupt with no work for %p\n", vq);
2500a8a69ddSRusty Russell 		return IRQ_NONE;
2510a8a69ddSRusty Russell 	}
2520a8a69ddSRusty Russell 
2530a8a69ddSRusty Russell 	if (unlikely(vq->broken))
2540a8a69ddSRusty Russell 		return IRQ_HANDLED;
2550a8a69ddSRusty Russell 
2560a8a69ddSRusty Russell 	pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
2570a8a69ddSRusty Russell 	if (vq->vq.callback && !vq->vq.callback(&vq->vq))
2580a8a69ddSRusty Russell 		vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
2590a8a69ddSRusty Russell 
2600a8a69ddSRusty Russell 	return IRQ_HANDLED;
2610a8a69ddSRusty Russell }
2620a8a69ddSRusty Russell 
2630a8a69ddSRusty Russell static struct virtqueue_ops vring_vq_ops = {
2640a8a69ddSRusty Russell 	.add_buf = vring_add_buf,
2650a8a69ddSRusty Russell 	.get_buf = vring_get_buf,
2660a8a69ddSRusty Russell 	.kick = vring_kick,
2670a8a69ddSRusty Russell 	.restart = vring_restart,
2680a8a69ddSRusty Russell 	.shutdown = vring_shutdown,
2690a8a69ddSRusty Russell };
2700a8a69ddSRusty Russell 
2710a8a69ddSRusty Russell struct virtqueue *vring_new_virtqueue(unsigned int num,
2720a8a69ddSRusty Russell 				      struct virtio_device *vdev,
2730a8a69ddSRusty Russell 				      void *pages,
2740a8a69ddSRusty Russell 				      void (*notify)(struct virtqueue *),
2750a8a69ddSRusty Russell 				      bool (*callback)(struct virtqueue *))
2760a8a69ddSRusty Russell {
2770a8a69ddSRusty Russell 	struct vring_virtqueue *vq;
2780a8a69ddSRusty Russell 	unsigned int i;
2790a8a69ddSRusty Russell 
2800a8a69ddSRusty Russell 	vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
2810a8a69ddSRusty Russell 	if (!vq)
2820a8a69ddSRusty Russell 		return NULL;
2830a8a69ddSRusty Russell 
2840a8a69ddSRusty Russell 	vring_init(&vq->vring, num, pages);
2850a8a69ddSRusty Russell 	vq->vq.callback = callback;
2860a8a69ddSRusty Russell 	vq->vq.vdev = vdev;
2870a8a69ddSRusty Russell 	vq->vq.vq_ops = &vring_vq_ops;
2880a8a69ddSRusty Russell 	vq->notify = notify;
2890a8a69ddSRusty Russell 	vq->broken = false;
2900a8a69ddSRusty Russell 	vq->last_used_idx = 0;
2910a8a69ddSRusty Russell 	vq->num_added = 0;
2920a8a69ddSRusty Russell #ifdef DEBUG
2930a8a69ddSRusty Russell 	vq->in_use = false;
2940a8a69ddSRusty Russell #endif
2950a8a69ddSRusty Russell 
2960a8a69ddSRusty Russell 	/* No callback?  Tell other side not to bother us. */
2970a8a69ddSRusty Russell 	if (!callback)
2980a8a69ddSRusty Russell 		vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
2990a8a69ddSRusty Russell 
3000a8a69ddSRusty Russell 	/* Put everything in free lists. */
3010a8a69ddSRusty Russell 	vq->num_free = num;
3020a8a69ddSRusty Russell 	vq->free_head = 0;
3030a8a69ddSRusty Russell 	for (i = 0; i < num-1; i++)
3040a8a69ddSRusty Russell 		vq->vring.desc[i].next = i+1;
3050a8a69ddSRusty Russell 
3060a8a69ddSRusty Russell 	return &vq->vq;
3070a8a69ddSRusty Russell }
3080a8a69ddSRusty Russell 
3090a8a69ddSRusty Russell void vring_del_virtqueue(struct virtqueue *vq)
3100a8a69ddSRusty Russell {
3110a8a69ddSRusty Russell 	kfree(to_vvq(vq));
3120a8a69ddSRusty Russell }
3130a8a69ddSRusty Russell 
314