Re: [Xen-devel] [OSSTEST PATCH 2/2] RFC: enable CONFIG_LIVEPATCH in all relevant arm64 builds [and 1 more messages]

[Julien Grall <julien.grall@xxxxxxx>]

Hi Ian,

On 05/06/2019 12:31, Ian Jackson wrote:
> Julien Grall writes ("Re: [OSSTEST PATCH 2/2] RFC: enable CONFIG_LIVEPATCH in all 
> relevant arm64 builds"):
> > On 05/06/2019 12:13, Ian Jackson wrote:
> > > Switching on CONFIG_LIVEPATCH for the affected tests will, hopefully,
> > > make this problem "go away" again.  This is clearly a bodge.  But it
> > > is better than simply force pushing: if we can get it to boot, we will
> > > be able to run the other tests.
> > How about disabling the test on rochester?
> I could invent a new hostflag
>    xen-4.11-arm64-seems-to-boot-ok
> but that seems ridiculous.
>
> And, we presumably want to actually test other things on rochester!
The problem is I have no way to tell whether a given Xen will boot on Thunder-X 
(or any other Arm platform we may buy) at any point the future.
It would be preferable if we have a way to say "test FOO can't run on 
rochester{0,1}".
> Jan Beulich writes ("Re: [OSSTEST PATCH 2/2] RFC: enable CONFIG_LIVEPATCH in all 
> relevant arm64 builds"):
> > On 05.06.19 at 13:13, <ian.jackson@xxxxxxxxxxxxx> wrote:
> > > --- a/mfi-common
> > > +++ b/mfi-common
> > > @@ -373,7 +373,8 @@ create_build_jobs () {
> > >           fi
> > >       fi
> > >   
> > > -    if branch_wants_livepatch; then
> > > +    if branch_wants_livepatch ||
> > > +          [ $arch = arm64 -a "$xenbranch" = xen-4.11-testing ]; then
> > >          livepatch_runvars='enable_livepatch=true'
> > >       fi
> > Isn't this overly restrictive, i.e. wouldn't this better be done uniformly
> > for all branches?
> No.  Because the bug is a random failure, I don't want to permute all
> the other branches and maybe cause them to experience it.
>
> I still think it would be better to fix this in the Xen code.
I am afraid that's not going to be possible. It requires a full rework of the 
boot and memory code. At the moment, I have 100 patches written, and I would 
expect much more to come.
> Would it be possible to make whether to do "wrong thing A" (which does
> not boot on rochster) or "wrong thing B" (which boots on rochester but
> not on some other unknown machine(s)) a config or boot-time option ?
That's not an option for me. There are no promise that Xen will not crash in the 
future on Thunder-X (or any new platform we have) without addressing all the 
problem.
This is an annoying problem, but the only thing we can do is limiting the 
testing on Thunder-X for the time being.
This is not the first time we face Arm Arm violation in Xen. We already had 
similar problem on Arm32 with the Set/Way (remember guest crashing time to time).
I hope such issues will show how critical it is to work actively making Xen more 
compliant rather than piling up more work on top of unstable foundation.
Cheers,

--
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel