On Wed, Dec 07, 2011 at 04:02:45PM +0200, Sasha Levin wrote:

On Sun, 2011-12-04 at 20:23 +0200, Sasha Levin wrote:

[snip]

Rusty, Michael, does the below looks a reasonable optimization for you?

OK overall but a bit hard to say for sure as it looks pretty incomplete ...

should I send it as a patch?

What's the performance gain?

quoted

Something like the following patch:

diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index c7a2c20..3166ca0 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c

@@ -82,6 +82,7 @@ struct vring_virtqueue
 
 	/* Host supports indirect buffers */
 	bool indirect;
+	struct kmem_cache *indirect_cache;
 
 	/* Host publishes avail event idx */
 	bool event;

@@ -110,6 +111,9 @@ struct vring_virtqueue
 
 #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq)
 
+static unsigned int ind_alloc_thresh = 0;
+module_param(ind_alloc_thresh, uint, S_IRUGO);

0 will have no effect?

quoted

+
 /* Set up an indirect table of descriptors and add it to the queue. */
 static int vring_add_indirect(struct vring_virtqueue *vq,
 			      struct scatterlist sg[],

A global parameter is OK for testing but likely not what we want
in real life. This needs to be different per device.

quoted

@@ -121,7 +125,10 @@ static int vring_add_indirect(struct vring_virtqueue *vq,
 	unsigned head;
 	int i;
 
-	desc = kmalloc((out + in) * sizeof(struct vring_desc), gfp);
+	if ((out + in) <= ind_alloc_thresh)
+		desc = kmem_cache_alloc(vq->indirect_cache, gfp);
+	else
+		desc = kmalloc((out + in) * sizeof(struct vring_desc), gfp);
 	if (!desc)
 		return -ENOMEM;
 

free unaffected?

quoted

@@ -479,6 +486,9 @@ struct virtqueue *vring_new_virtqueue(unsigned int num,
 	vq->broken = false;
 	vq->last_used_idx = 0;
 	vq->num_added = 0;
+	if (ind_alloc_thresh)
+		vq->indirect_cache = KMEM_CACHE(vring_desc[ind_alloc_thresh], 0);

and need to cleanup too?

quoted

 	list_add_tail(&vq->vq.list, &vdev->vqs);
 #ifdef DEBUG
 	vq->in_use = false;

-- 

Sasha.