On Fri, Oct 30, 2020 at 03:25:06PM -0500, Justin Forbes wrote:
> On Fri, Oct 30, 2020 at 3:05 PM Jiri Olsa <jolsa(a)redhat.com> wrote:
> >
> > On Fri, Oct 30, 2020 at 02:57:23PM -0500, Justin Forbes wrote:
> > > On Thu, Oct 22, 2020 at 11:54 AM Justin Forbes
<jmforbes(a)linuxtx.org> wrote:
> > > >
> > > > On Thu, Oct 22, 2020 at 11:01 AM Jiri Olsa <jolsa(a)redhat.com>
wrote:
> > > > >
> > > > > On Thu, Oct 22, 2020 at 10:36:01AM -0500, Justin Forbes wrote:
> > > > > > On Thu, Oct 22, 2020 at 4:39 AM Bastien Nocera
<bnocera(a)redhat.com> wrote:
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > ----- Original Message -----
> > > > > > > > On Wed, Oct 21, 2020 at 12:27:20PM -0400, Bastien
Nocera wrote:
> > > > > > > > > Hey,
> > > > > > > > >
> > > > > > > > > I'm trying to build test kernels on my
Fedora 33 machine
> > > > > > > > > and ran into:
> > > > > > > > >
https://bugzilla.redhat.com/show_bug.cgi?id=1890107
> > > > > > > > >
> > > > > > > > > Does anyone know a work-around for this? I
tried a "dnf downgrade",
> > > > > > > > > to no avail.
> > > > > > > >
> > > > > > > > if your problem is broken dwarf info then I
don't have any
> > > > > > > > workaround, if you can't build kernel because
of that, try
> > > > > > > > to disable CONFIG_DEBUG_INFO_BTF
> > > > > > >
> > > > > > > That was the one, thanks!
> > > > > >
> > > > > > Interestingly, this seems to be due to a bug introduced in
the
> > > > > > upstream kernel sometime late last week. We had a
successful build on
> > > > > > the 14th with CONFIG_DEBUG_INFO_BTF on, by Monday, this was
failing.
> > > > >
> > > > > for me it was introduced with recent gcc update to:
> > > > > $ gcc --version
> > > > > gcc (GCC) 10.2.1 20201005 (Red Hat 10.2.1-5)
> > > > >
> > > >
> > > > I thought that was the case, so I tried building an older kernel on
> > > > the newer gcc update and it worked.
> > > >
> > >
> > > This is definitely an upstream kernel regression. I can continue to
> > > build 5.9.x kernels with CONFIG_DEBUG_INFO_BTF on, but current linus
> > > tree including 5.10-rc1 fail on the same system.
> >
> > hi,
> > what gcc are you using? I just updated gcc to:
> >
> > [jolsa@dell-r440-01 ~]$ gcc --version
> > gcc (GCC) 10.2.1 20201016 (Red Hat 10.2.1-6)
> >
> > and I'm able to build linux 5.10-rc1 again with CONFIG_DEBUG_INFO_BTF=y
> >
>
> gcc --version
> gcc (GCC) 10.2.1 20201016 (Red Hat 10.2.1-6)
>
> LINK resolve_btfids
> FAILED unresolved symbol vfs_getattr
> make: *** [Makefile:1169: vmlinux] Error 255
> error: Bad exit status from /home/jmforbes/rpmbuild/tmp/rpm-tmp.7tWIJ4 (%build)
ugh sry.. I have local pahole workaround/fix and it sneaked in :-\
it's still broken
So, this clearly got triggered by a change in the upstream kernel, but
are you saying that this is in fact a pahole bug that something
upstream triggered? Where is the correct fix here? It would be nice
to get CONFIG_DEBUG_INFO_BTF back on, one way or another.
Thanks,
Justin