Re: [PULL] virtio

["Michael S. Tsirkin" <mst@redhat.com>]

On Tue, Nov 01, 2011 at 02:33:33PM +0200, Sasha Levin wrote:
> On Tue, 2011-11-01 at 13:45 +0200, Michael S. Tsirkin wrote:
> > On Tue, Nov 01, 2011 at 10:39:08AM +1030, Rusty Russell wrote:
> > >  * [new tag]         rusty@rustcorp.com.au-v3.1-7196-gac5be1e -> rusty@rustcorp.com.au-v3.1-7196-gac5be1e
> > > 
> > > The following changes since commit 839d8810747bbf39e0a5a7f223b67bffa7945f8d:
> > > 
> > >   Merge branch 'i2c-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jdelvare/staging (2011-10-30 15:54:59 -0700)
> > > 
> > > are available in the git repository at:
> > > 
> > >   git://github.com/rustyrussell/linux.git master
> > > 
> > > Alexey Kardashevskiy (1):
> > >       virtio-pci: Use PCI MMIO instead of PIO when available
> > 
> > I missed this one - wasn't Cc'd neither me, kvm or virtio mailing lists.
> > 
> > It's well known that mmio is much slower than pio on kvm, since
> > mmio needs to be emulated to get at the address.
> > So I'm expecting this will cause a performance regression.
> > IMO we should keep using PIO for VQ and interrupt status access
> > if PIO is available.
> > 
> > Another consideration is that in an attempt to pack data
> > densely in the PIO space the layout became messy.
> > It would be better to have common config space and
> > per-device config space in separate pages, possibly
> > with padding between them.
> > 
> > So I'd like a bit more discussion on this patch,
> > I'm concerned that if this is released in 3.2 as is we'll
> > have to support this forever. How about a revert for now?
> 
> Another thing, the patch tries to map BAR 2 and use it as the
> configuration space.
> 
> It's both not documented properly anywhere, and is not fully backwards
> compatible - we currently use BAR 2 as part of our MSIX handling in the
> kvm tool and I'm sure we're not the only ones to assume virtio-pci only
> uses BAR 0.
> 
> A proper solution would be for example a configuration in the PIO config
> space which points to the MMIO BAR to use instead.

I think it makes sense to put the configuration in PCI
configuration space, using vendor-specific capability.
This way we can reuse existing functionality for scanning
capability lists.

> Unless Michael pointed this patch out, it would have broken (at least)
> the kvm tool in a non obvious way which would require a rather long
> session of 'git bisect' to figure out whats wrong.
> 
> -- 
> 
> Sasha.