Re: [Qemu-devel] [virtio-dev] Re: [PATCH v13 1/2] virtio-crypto: Add virtio crypto device specification

[Michael S. Tsirkin]

On Thu, Nov 10, 2016 at 02:25:44AM +0000, Gonglei (Arei) wrote:
> 
> >
> > Subject: [virtio-dev] Re: [Qemu-devel] [PATCH v13 1/2] virtio-crypto: Add 
> > virtio
> > crypto device specification
> > 
> > On Wed, Nov 09, 2016 at 01:11:20AM +0000, Gonglei (Arei) wrote:
> > > Nope, Actually I kept those description here is because I wanted to 
> > > represent
> > each packet
> > > Intuitionally, otherwise I don't know how to explain them only occupy one
> > entry in desc table
> > > by indirect table method.
> > 
> > whether to use the indirect method should not be up to device.
> > I don't see where does the spec require indirect method,
> > ANY_LAYOUT is implied so spec should not dictate that IMO.
> > 
> Yes, I reread the virtio spec. And
> 
> 2.4.4 says: 
>  " The framing of messages with descriptors is independent of the contents of 
> the buffers."
> 
> 2.4.4.2 says:
>  "The driver SHOULD NOT use an excessive number of descriptors to describe a 
> buffer."
> 
> For virtio-crypto, I think I can just add a NOTE to remind people using 
> indirect method 
> if they want to get better performance in production environment, can I? 
> 
> 
> Thanks,
> -Gonglei

I'm not sure why frankly. Is there a reason crypto is special here?

> > > So I changed the code completely as Stefan's suggestion and
> > > revised the spec a little.
> > >
> > > This just is a representative of the realization so that the people can 
> > > easily
> > understand what
> > > the virtio crypto request's components. It isn't completely same with the
> > code.
> > > For virtio-scsi device, the struct virtio_scsi_req_cmd also used this way 
> > > IIUR.
> > >
> > > Thanks,
> > > -Gonglei
> > 
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: address@hidden
> > For additional commands, e-mail: address@hidden