On Sun, Jun 22, 2014 at 01:24:48PM +0300, Michael S. Tsirkin wrote:
> On Fri, Jun 20, 2014 at 11:29:40PM +0800, Ming Lei wrote:
> > @@ -24,8 +26,8 @@ static struct workqueue_struct *virtblk_wq;
> > struct virtio_blk
> > {
> > struct virtio_device *vdev;
> > - struct virtqueue *vq;
> > - spinlock_t vq_lock;
> > + struct virtqueue *vq[MAX_NUM_VQ];
> > + spinlock_t vq_lock[MAX_NUM_VQ];
>
> array of struct {
> *vq;
> spinlock_t lock;
> }
> would use more memory but would get us better locality.
> It might even make sense to add padding to avoid
> cacheline sharing between two unrelated VQs.
> Want to try?
It's still false sharing because the queue objects share cachelines.
To operate without contention they have to be physically separated
from each other like so:
struct vq {
struct virtqueue *q;
spinlock_t lock;
} ____cacheline_aligned_in_smp;
struct some_other_struct {
....
struct vq vq[MAX_NUM_VQ];
....
};
This keeps locality to objects within a queue, but separates each
queue onto it's own cacheline....
Cheers,
Dave.
--
Dave Chinner
david@xxxxxxxxxxxxx
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
