Re: [PATCH 0/3] ibmvscsis driver rewrite

[Bart Van Assche <bvanassche@acm.org>]

On Thu, Feb 10, 2011 at 7:41 PM, Nicholas A. Bellinger
<nab@linux-iscsi.org> wrote:
>
> On Thu, 2011-02-10 at 13:48 +0100, Hannes Reinecke wrote:
> >
> > On 02/10/2011 01:21 PM, FUJITA Tomonori wrote:
> > > Finally, I have the working driver. We are ready to remove the old
> > > ibmvstgt driver.
> > >
> > Hmm. Wouldn't it be possible to use it as a template for a
> > virtio-srp target?
> > From what I've seen we 'just' need to modify it to use virtio
> > instead for pSeries rdma callbacks.
> > And then we would have a virtio-srp target onto which any kvm guests
> > could connect to.

One should keep in mind that there exist two kinds of SRP drivers for Linux:
- Those based on synchronous data transfer calls - calls that only
return after the entire transfer has finished. The ibmvscsi, ibmvscsis
drivers and the libsrp library fall in this category.
- Those based on asynchronous data transfers - calls that return
immediately and where a callback function is invoked to report
completion. ib_srp and ib_srpt fall into this category (ib_srp is
upstream, ib_srpt not yet).

As far as I know the virtio interface uses callbacks and hence is an
asynchronous interface ? If so, the InfiniBand drivers might be a
better starting point.

Bart.