[OE-core] Fix for the APIC hangs in qemux86-64

[Richard Purdie]

On Wed, 2017-12-13 at 09:07 -0500, Bruce Ashfield wrote:
> On 12/13/2017 09:05 AM, Richard Purdie wrote:
> > 
> > On Wed, 2017-12-13 at 08:38 -0500, Bruce Ashfield wrote:
> > > 
> > > On 12/13/2017 07:34 AM, Richard Purdie wrote:
> > > > 
> > > > 
> > > > Hi Bruce,
> > > > 
> > > > https://bugzilla.yoctoproject.org/show_bug.cgi?id=12301
> > > > 
> > > > We've root caused the hangs we're seeing in qemux86-64 on our
> > > > autobuilder workers. We've identified a commit on 4.15-rc3
> > > > which
> > > > fixes
> > > > the issue but backporting it to the autobuilder's host kernels
> > > > will
> > > > be
> > > > a pain.
> > > > 
> > > > As a workaround I'd like to enable x2apic on qemux86-64. Since
> > > > that
> > > > works via MSRs it avoids the problem.
> > > > 
> > > > Would it be a big issue to set CONFIG_X86_X2APIC=y for qemux86-
> > > > 64?
> > > Nope. No big deal at all. I've used the same config i the past
> > > when doing work with rt-KVM, so I know it well.
> > > 
> > > I'll prep a commit, and send it along as part of a rebased set
> > > of kernel patches.
> > Thanks, much appreciated.
> > 
> > Did we get to the bottom of the 4.12 selftest failures?
> > 
> Not yet no. I was distracted for a day, but need to learn how
> to run them myself. I was able to log in via serial in my normal
> boot flow, so that didn't get me anywhere.
> 
> Is the selftest HOWO in the yocto docs ? I'll google it up this
> morning.

No HOWTO but I did give you the command. In a poky checkout run:

oe-selftest -r runqemu.RunqemuTests.test_boot_deploy

which runs a single test that fails (setup DL_DIR/SSTATE_MIRRORS first
to speed it up a bit).

It may tell you to clear SANITY_TESTED_DISTROS, I think it auto adds
meta-selftest now. You may also need to set DISPLAY to something (e.g.
a dummy vncserver).

Not sure what is different about the serial qemu is using verses your
own tests...

> > We may also need to backport the defconfig change back to the older
> > releases so we don't get the hangs in those builds either...
> Not a problem either. I'll prepare several.> 

Thanks!

Cheers,

Richard
>