On 10/08/2017 18:56, Richard W.M. Jones wrote:
> Since switching to blk-mq as the default in commit 5c279bd9e406
> ("scsi: default to scsi-mq"), virtio-scsi LUNs consume about 10x as
> much kernel memory.
>
> qemu currently allocates a fixed 128 entry virtqueue. can_queue
> currently is set to 1024. But with indirect descriptors, each command
> in the queue takes 1 virtqueue entry, so the number of commands which
> can be queued is equal to the length of the virtqueue.
>
> Note I intend to send a patch to qemu to allow the virtqueue size to
> be configured from the qemu command line.
>
> Thanks Paolo Bonzini, Christoph Hellwig.
>
> Signed-off-by: Richard W.M. Jones <rjo...@redhat.com>
> ---
> drivers/scsi/virtio_scsi.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/scsi/virtio_scsi.c b/drivers/scsi/virtio_scsi.c
> index 9be211d68b15..7c28e8d4955a 100644
> --- a/drivers/scsi/virtio_scsi.c
> +++ b/drivers/scsi/virtio_scsi.c
> @@ -818,7 +818,6 @@ static struct scsi_host_template
> virtscsi_host_template_single = {
> .eh_timed_out = virtscsi_eh_timed_out,
> .slave_alloc = virtscsi_device_alloc,
>
> - .can_queue = 1024,
> .dma_boundary = UINT_MAX,
> .use_clustering = ENABLE_CLUSTERING,
> .target_alloc = virtscsi_target_alloc,
> @@ -839,7 +838,6 @@ static struct scsi_host_template
> virtscsi_host_template_multi = {
> .eh_timed_out = virtscsi_eh_timed_out,
> .slave_alloc = virtscsi_device_alloc,
>
> - .can_queue = 1024,
> .dma_boundary = UINT_MAX,
> .use_clustering = ENABLE_CLUSTERING,
> .target_alloc = virtscsi_target_alloc,
> @@ -972,6 +970,8 @@ static int virtscsi_probe(struct virtio_device *vdev)
> if (err)
> goto virtscsi_init_failed;
>
> + shost->can_queue = virtqueue_get_vring_size(vscsi->req_vqs[0].vq);
> +
> cmd_per_lun = virtscsi_config_get(vdev, cmd_per_lun) ?: 1;
> shost->cmd_per_lun = min_t(u32, cmd_per_lun, shost->can_queue);
> shost->max_sectors = virtscsi_config_get(vdev, max_sectors) ?: 0xFFFF;
>
Acked-by: Paolo Bonzini <pbonz...@redhat.com>
