Re: [Qemu-devel] [PATCH 0/3] ARM: three easy patches for coverity-reported issues

[Alexander Graf]

On 02/18/2014 01:37 PM, Peter Maydell wrote:
> On 18 February 2014 12:17, Alexander Graf <address@hidden> wrote:
> > On 02/18/2014 12:22 PM, Peter Maydell wrote:
> > > My criteria for ARM in the past has typically been "there's
> > > a new release every three months, anything that got past
> > > the release testing process for release N is sufficiently
> > > non-critical it can just go into release N+1".
> > Unfortunately this doesn't work for distributions. Distros
> > need to maintain a stable branch for the lifetime of a release
> > to ensure that we're reasonably regression free.
> >
> > If you indicate that this doesn't apply to ARM it basically means you admit
> > that ARM systems are not yet ready for "stable" use by customers when they
> > want to use KVM. At least at the point when we agree that customers do want
> > to run on a stable base for virtualization on ARM we need a working -stable
> > system for critical fixes.
> I agree in general that ARM support needs to move from
> its traditional "this is just a dev tool" situation to
> a broader level of support/stability guarantees for KVM.
> (We're not yet guaranteeing cross-version migration,
> for another example there.)

Yup. I think it's reasonably to not declare ARM a "stable" target at the 
current point in time. But I think we agree that we want to change that 
- timeframe wise probably around the release after 2.0 at which point 
hopefully PCI and virtio 1.0 have settled.

> However again we run into the definition of "what's a
> critical fix?". I think if distros need a stable branch
> then they need to be prepared to do the work of sorting
> through what counts as a critical fix that needs to be
> ported to that branch. (For instance, which boards and
> targets do they care about?)

I think this is up for discussion. If I had to declare anything, I 
wouldn't consider anything but the virt machine as supported for example 
- similar to how x86 only really considers the pc machine type stable.

> For instance patch 3 only applies to the integrator
> board, and I don't consider the guest-to-host border
> to be a security boundary for most of our legacy board
> models: there's just too much unaudited device code for
> that to be trustable.

Yes, I fully agree. Traditionally what I've done is to reply to patches 
that I consider stable material and nag the maintainer about CCing it. 
After a while people got so afraid of my emails that they started doing 
the CC themselves :). But in case of the integrator board I personally 
wouldn't bother ;).


Alex