Re: [PATCH] compiler.h: Avoid the usage of __typeof_unqual__() when __GENKSYMS__ is defined

[Uros Bizjak]

On Wed, Apr 9, 2025 at 5:28 PM Borislav Petkov <bp@xxxxxxxxx> wrote:
>
> On Sun, Apr 06, 2025 at 05:36:13PM +0200, Uros Bizjak wrote:
> > On Fri, Apr 4, 2025 at 9:14 PM Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote:
> > >
> > > On Fri, Apr 4, 2025 at 11:37 PM Uros Bizjak <ubizjak@xxxxxxxxx> wrote:
> > > >
> > > > On Fri, Apr 4, 2025 at 4:06 PM Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote:
> > > >
> > > > > > > > Current version of genksyms doesn't know anything about __typeof_unqual__()
> > > > > > > > operator.  Avoid the usage of __typeof_unqual__() with genksyms to prevent
> > > > > > > > errors when symbols are versioned.
> > > > > > > >
> > > > > > > > There were no problems with gendwarfksyms.
> > > > > > > >
> > > > > > > > Signed-off-by: Uros Bizjak <ubizjak@xxxxxxxxx>
> > > > > > > > Fixes: ac053946f5c40 ("compiler.h: introduce TYPEOF_UNQUAL() macro")
> > > > > > > > Reported-by: Paul Menzel <pmenzel@xxxxxxxxxxxxx>
> > > > > > > > Closes: https://lore.kernel.org/lkml/81a25a60-de78-43fb-b56a-131151e1c035@xxxxxxxxxxxxx/
> > > > > > > > Cc: Sami Tolvanen <samitolvanen@xxxxxxxxxx>
> > > > > > > > Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> > > > > > > > ---
> > > > > > >
> > > > > > >
> > > > > > > Why don't you add it to the genksyms keyword table?
> > > > > >
> > > > > > It doesn't work, even if I patch it with an even more elaborate patch
> > > > > > (attached).
> > > > > >
> > > > > > I guess some more surgery will be needed, but for now a fallback works
> > > > > > as expected.
> > > > > >
> > > > > > Uros.
> > > > >
> > > > > The attached patch looks good to me.
> > > >
> > > > FAOD - do you refer to the submitted one for compiler.h or to the one
> > > > for scripts/genksyms/keywords.c? (The latter doesn't fix the warning,
> > > > though).
> > >
> > >
> > >
> > > You are still seeing the warnings because __typeof_unqual__
> > > is not only the issue.
> > >
> > > Hint:
> > >
> > > $ make -s KCFLAGS=-D__GENKSYMS__  arch/x86/kernel/setup_percpu.i
> > > $ grep  'this_cpu_off;'  arch/x86/kernel/setup_percpu.i
> >
> > I see.
> >
> > With my workaround, this_cpu_off is declared as:
> >
> > extern __attribute__((section(".data..percpu" "..hot.."
> > "this_cpu_off"))) __typeof__(unsigned long) this_cpu_off;
> >
> > while without workaround, the same variable is declared as:
> >
> > extern __seg_gs __attribute__((section(".data..percpu" "..hot.."
> > "this_cpu_off"))) __typeof__(unsigned long) this_cpu_off;
> >
> > It looks that genksyms should be extended to handle (or ignore)
> > __seg_gs/__seg_fs named address prefix. Somewhat surprising, because
> > genksyms can process:
> >
> > extern __attribute__((section(".data..percpu" "..hot.."
> > "const_current_task"))) __typeof__(struct task_struct * const
> > __seg_gs) const_current_task
> >
> > without problems.
> >
> > I'm sorry, but I'm not able to extend genksyms with a new keyword by myself...
>
> Well, we need a fix here because this fires a lot by now - triggers on my
> machines now too.
>
> So either take a fix or we'll need to revert until it is fixed properly.

The workaround is posted to the list. It should be committed to the
mainline until genksyms is fixed.

Uros.