Re: Linux 2.6.38-rc1 doesn't boot

[Shaohua Li <shaohua.li@intel.com>]

On Wed, 2011-01-19 at 17:09 +0800, Ingo Molnar wrote:
> * H. Peter Anvin <h.peter.anvin@intel.com> wrote:
> 
> > On 01/19/2011 12:12 AM, Shaohua Li wrote:
> > > On Wed, 2011-01-19 at 15:49 +0800, Markus Trippelsdorf wrote:
> > >> On 2011.01.19 at 08:39 +0100, Markus Trippelsdorf wrote:
> > >>> On 2011.01.18 at 15:54 -0800, Linus Torvalds wrote:
> > >>>>
> > >>>> And as usual, report any regressions to the lists and the appropriate
> > >>>> authorities.
> > >>>
> > >>> Unfortunately 2.6.38-rc1 doesn't even boot on my machine (amd64).
> > >>> This is caused by 86b1e8dd83cbb0f:
> > >>>     x86: Make relocatable kernel work with new binutils
> > >>>
> > >>> Reverting the commit solves the problem.
> > >>
> > >> I'm running the latest binutils:
> > >> GNU ld (Linux/GNU Binutils) 2.21.51.0.5.20110104
> > > Hmm, reproduce it here with binutils-2.21.51.0.6-20110118
> > > but not with GNU ld (GNU Binutils for Ubuntu) 2.20.51-system.20100908
> > > I got this in system.map: ffffffff03514880 D jiffies_64, which looks
> > > wrong.
> > > looks binutils changed something again.
> > > Have no idea, CC Lu Hongjiu.
> > > 
> > 
> > Either way... the whole jiffies vs jiffies_64 thing is kind of
> > ridiculous.  We should be able to do it in a completely
> > architecture-generic way by either making it a union(!) (with "jiffies"
> > and "jiffies_64" presumably would be #defines, or we do a global replace
> > across the tree), moving the variable declaration itself to a .S file
> > (which would only have data components and therefore would be
> > arch-generic) or doing something like the attached (untested since it is
> > 1 am here) patch.
> > 
> > This should let us get rid of the hacks in *all* the architectures, not
> > just x86.
> 
> Ok - until it's resolved i'll queue up a revert - a known build failure is preferred 
> to a boot regression.
it's not a build failure, without it, my i386 kernel can't boot
actually.

Thanks,
Shaohua