Re: [PATCH RFC 03/13] vhost: batching fetches

[Jason Wang <jasowang@redhat.com>]

On 2020/6/7 下午9:57, Michael S. Tsirkin wrote:
> On Fri, Jun 05, 2020 at 11:40:17AM +0800, Jason Wang wrote:
>> On 2020/6/4 下午4:59, Michael S. Tsirkin wrote:
>>> On Wed, Jun 03, 2020 at 03:27:39PM +0800, Jason Wang wrote:
>>>> On 2020/6/2 下午9:06, Michael S. Tsirkin wrote:
>>>>> With this patch applied, new and old code perform identically.
>>>>>
>>>>> Lots of extra optimizations are now possible, e.g.
>>>>> we can fetch multiple heads with copy_from/to_user now.
>>>>> We can get rid of maintaining the log array.  Etc etc.
>>>>>
>>>>> Signed-off-by: Michael S. Tsirkin<mst@redhat.com>
>>>>> Signed-off-by: Eugenio Pérez<eperezma@redhat.com>
>>>>> Link:https://lore.kernel.org/r/20200401183118.8334-4-eperezma@redhat.com
>>>>> Signed-off-by: Michael S. Tsirkin<mst@redhat.com>
>>>>> ---
>>>>>     drivers/vhost/test.c  |  2 +-
>>>>>     drivers/vhost/vhost.c | 47 ++++++++++++++++++++++++++++++++++++++-----
>>>>>     drivers/vhost/vhost.h |  5 ++++-
>>>>>     3 files changed, 47 insertions(+), 7 deletions(-)
>>>>>
>>>>> diff --git a/drivers/vhost/test.c b/drivers/vhost/test.c
>>>>> index 9a3a09005e03..02806d6f84ef 100644
>>>>> --- a/drivers/vhost/test.c
>>>>> +++ b/drivers/vhost/test.c
>>>>> @@ -119,7 +119,7 @@ static int vhost_test_open(struct inode *inode, struct file *f)
>>>>>     	dev = &n->dev;
>>>>>     	vqs[VHOST_TEST_VQ] = &n->vqs[VHOST_TEST_VQ];
>>>>>     	n->vqs[VHOST_TEST_VQ].handle_kick = handle_vq_kick;
>>>>> -	vhost_dev_init(dev, vqs, VHOST_TEST_VQ_MAX, UIO_MAXIOV,
>>>>> +	vhost_dev_init(dev, vqs, VHOST_TEST_VQ_MAX, UIO_MAXIOV + 64,
>>>>>     		       VHOST_TEST_PKT_WEIGHT, VHOST_TEST_WEIGHT, NULL);
>>>>>     	f->private_data = n;
>>>>> diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
>>>>> index 8f9a07282625..aca2a5b0d078 100644
>>>>> --- a/drivers/vhost/vhost.c
>>>>> +++ b/drivers/vhost/vhost.c
>>>>> @@ -299,6 +299,7 @@ static void vhost_vq_reset(struct vhost_dev *dev,
>>>>>     {
>>>>>     	vq->num = 1;
>>>>>     	vq->ndescs = 0;
>>>>> +	vq->first_desc = 0;
>>>>>     	vq->desc = NULL;
>>>>>     	vq->avail = NULL;
>>>>>     	vq->used = NULL;
>>>>> @@ -367,6 +368,11 @@ static int vhost_worker(void *data)
>>>>>     	return 0;
>>>>>     }
>>>>> +static int vhost_vq_num_batch_descs(struct vhost_virtqueue *vq)
>>>>> +{
>>>>> +	return vq->max_descs - UIO_MAXIOV;
>>>>> +}
>>>> 1 descriptor does not mean 1 iov, e.g userspace may pass several 1 byte
>>>> length memory regions for us to translate.
>>>>
>>> Yes but I don't see the relevance. This tells us how many descriptors to
>>> batch, not how many IOVs.
>> Yes, but questions are:
>>
>> - this introduce another obstacle to support more than 1K queue size
>> - if we support 1K queue size, does it mean we need to cache 1K descriptors,
>> which seems a large stress on the cache
>>
>> Thanks
>>
>>
> Still don't understand the relevance. We support up to 1K descriptors
> per buffer just for IOV since we always did. This adds 64 more
> descriptors - is that a big deal?


If I understanding correctly, for net, the code tries to batch 
descriptors for at last one packet.

If we allow 1K queue size then we allow a packet that consists of 1K 
descriptors. Then we need to cache 1K descriptors.

Thanks