Re: [virtio-dev] Re: VIRTIO - compatibility with different virtualization solutions

[Wei Liu <wei.liu2@citrix.com>]

On Fri, Feb 21, 2014 at 11:24:14AM +1030, Rusty Russell wrote:
> Daniel Kiper <daniel.kiper@oracle.com> writes:
> > Hey,
> >
> > On Thu, Feb 20, 2014 at 06:18:00PM +1030, Rusty Russell wrote:
> >> Ian Campbell <Ian.Campbell@citrix.com> writes:
> >> > On Wed, 2014-02-19 at 10:56 +1030, Rusty Russell wrote:
> >> >> For platforms using EPT, I don't think you want anything but guest
> >> >> addresses, do you?
> >> >
> >> > No, the arguments for preventing unfettered access by backends to
> >> > frontend RAM applies to EPT as well.
> >>
> >> I can see how you'd parse my sentence that way, I think, but the two
> >> are orthogonal.
> >>
> >> AFAICT your grant-table access restrictions are page granularity, though
> >> you don't use page-aligned data (eg. in xen-netfront).  This level of
> >> access control is possible using the virtio ring too, but noone has
> >> implemented such a thing AFAIK.
> >
> > Could you say in short how it should be done? DMA API is an option but
> > if there is a simpler mechanism available in VIRTIO itself we will be
> > happy to use it in Xen.
> 
> OK, this challenged me to think harder.
> 
> The queue itself is effectively a grant table (as long as you don't give
> the backend write access to it).  The available ring tells you where the
> buffers are and whether they are readable or writable.  The used ring
> tells you when they're used.
> 
> However, performance would suck due to no caching: you'd end up doing a
> map and unmap on every packet.  I'm assuming Xen currently avoids that
> somehow?  Seems likely...
> 

If you're talking about Xen drivers in Linux kernel...

At least for Xen network backend in mainline Linux, it uses copy instead
of map.  Zoltan Kiss @ Citrix is working on a mapping network backend.
He uses batch unmap to avoid performance penalty.

Wei.

> On the other hand, if we wanted a more Xen-like setup, it would looke
> like this:
> 
> 1) Abstract away the "physical addresses" to "handles" in the standard,
>    and allow some platform-specific mapping setup and teardown.
> 
> 2) In Linux, implement a virtio DMA ops which handles the grant table
>    stuff for Xen (returning grant table ids + offset or something?),
>    noop for others.  This would be a runtime thing.
> 
> 3) In Linux, change the drivers to use this API.
> 
> Now, Xen will not be able to use vhost to accelerate, but it doesn't now
> anyway.
> 
> Am I missing anything?
> 
> Cheers,
> Rusty.