RFC: Page size on PPC/PPC64 builders

Josh Boyer jwboyer at gmail.com
Tue Mar 4 18:47:42 UTC 2008


On Tue, 04 Mar 2008 17:51:22 +0000
David Woodhouse <dwmw2 at infradead.org> wrote:

> On Tue, 2008-03-04 at 09:59 +0100, Jim Meyering wrote:
> > David Woodhouse <dwmw2 at infradead.org> wrote:
> > > I'm concerned that switching back to 4KiB pages is just papering over
> > > real bugs to make life easier for PPC folks. I suspect that what I
> > > should _actually_ do is keep it 64KiB, brazen it out and LART people who
> > > just exclude ppc builds without actually looking at the problem for
> > > themselves. But I'm lazy too... maybe we should switch the x86_64
> > > builders to 64KiB instead? :)
> > >
> > > Opinions?
> > 
> > I'm surprised there's so much debate.
> > Isn't it obvious?  A service that finds generic bugs -- and ones
> > that would likely be much harder to diagnose in any other context.
> > People will complain no matter what you do, so take the high road:
> > 
> > Brazen it out with a LART :-)
> 
> The problem is that there is a risk that it'll make the less
> conscientious packagers just think 'oh, building for ppc is painful' and
> exclude that architecture. There's enough of that already, with some
> people even sticking to that line even after the same bug bites them on
> x86_64 too.
> 
> That's not _such_ an issue because there are relatively few such
> packagers, and we have the rule that all architecture exclusions _must_
> be filed in bugzilla and we can keep track of them through the
> ExcludeArch tracker bugs.
> 
> My _real_ concern is that continuing to use 64KiB pages is likely to
> increase the motivation for people to let PPC builds fail _without_
> aborting the main build on x86/x86_64, so the laziest of packagers don't
> even have to _look_ when their build fails. And then a lot of the
> benefit is lost (or we just start pushing even more of the generic bugs
> onto the arch team and making it really hard for them to keep in sync
> with the development tree properly).
> 
> Yes, that would be extremely misguided, but I think it's dangerously
> likely to happen and I don't want to make it _more_ likely or accelerate
> it.

My biggest concern is that it doesn't match Fedora, which makes it
harder for people who _do_ care to actually reproduce the issue and
debug the problem.  Local mock builds on a Fedora PPC system would use
4KiB pages and would likely build/test just fine.  Reproducibility is
fairly important IMHO.

If we have a ppc64 machine running a RHEL kernel for people to log into
to do mock builds, etc my concerns are lessened slight.  But I really
think we should be eating our own dog food here.

josh




More information about the fedora-devel-list mailing list