Re: r276200: EFI boot failure: kernel stops booting at pci0: <ACPI PCI bus> on pcib0

On Sun, 2014-12-28 at 20:57 +0100, O. Hartmann wrote:
> Am Fri, 26 Dec 2014 12:23:42 -0700
> Ian Lepore <ian_at_freebsd.org> schrieb:
> 
> > On Fri, 2014-12-26 at 08:48 -0800, Adrian Chadd wrote:
> > > On 26 December 2014 at 04:01, O. Hartmann <ohartman_at_zedat.fu-berlin.de> wrote:
> > > > Am Thu, 25 Dec 2014 11:40:47 -0800
> > > > Adrian Chadd <adrian_at_freebsd.org> schrieb:
> > > >
> > > >> Would you be able to narrow it down to a small range of commits?
> > > >> that'll make it easier to chase down. :)
> > > >>
> > > >> Thanks!
> > > >>
> > > >>
> > > >>
> > > >> -adrian
> > > >>
> > > >>
> > > >> On 25 December 2014 at 10:42, O. Hartmann <ohartman_at_zedat.fu-berlin.de> wrote:
> > > >> >
> > > >> > Since 23rd's update of CURRENT, the kernel fails to boot on systems that boot
> > > >> > via EFI. Systems with legacy booting seem not to be affected.
> > > >> >
> > > >> > I just ran today into the problem updating a notebook with a Intel Haswell Intel
> > > >> > i5-4200M CPU (Haswell) on a Lenovo ThinkPad E540, bboting via UEFI, CURRENT
> > > >> > r276200. The very same caode base is running on several other boxes which boot
> > > >> > via legacy method. The very same failure showed up at the lab on an older HP
> > > >> > Compaq 8300 system, based on H81 chipset equipted with an Ivy-Bridge CPU,
> > > >> > booting also via EFI. That box stops at the exact same spot as the notebook does.
> > > >> >
> > > >> > The systems in question, also the legacy booting systems (aka the oldstyle
> > > >> > loader boot method), load drm2, i915kms.
> > > >> >
> > > >> > Booting old kernel/modules (via "boot kernel.old"), at CURRENT r275896 is all
> > > >> > right.
> > > >> >
> > > >> > What is happening here?
> > > >> >
> > > >> > Merry christmas day,
> > > >> >
> > > >> > oh
> > > >
> > > >
> > > > I narrowed down the culprit commit to be between r276060 (works) and r276075 (works
> > > > not). To avoid interferences from rogue modules, I disabled all modules loaded by
> > > > the loader, including drm2 and i915kms, but the picture is always the same. I'm
> > > > sorry, I have some duties to perform, so intersecting further is possible later
> > > > only ... I performed the iterative search of the foul commit by "svn update -r
> > > > 276XXX" and then build kernel only via "make kernel" - this just for the record in
> > > > case some world-dependencies might have effects.
> > > 
> > > Hi!
> > > 
> > > Thanks for that. Would you please file a PR with the details and what
> > > you've done?
> > > 
> > > I hope you can narrow it down further. You've done a great job
> > > already, I just can't see any clear winner there for a commit to back
> > > out :(
> > 
> > r276064 looks like a candidate.  At least, it has 'efi' in the name. :)
> > 
> > -- Ian
> 
> Well, r276063 works, but r2766064 doesn't.
> 
> oh

cc'ing the author of r276064, since spotting the fact that 'efi' was
involved in that revision used up 100% of my expertise in efi stuff. :)

-- Ian