Re: [PATCH] objtool: Support Clang RAX DRAP sequence

[Josh Poimboeuf]

On Tue, Mar 17, 2026 at 09:18:25AM +0100, Peter Zijlstra wrote:
> On Mon, Mar 16, 2026 at 05:47:56PM -0700, Josh Poimboeuf wrote:
> > Recent Clang can use RAX as a temporary register for the DRAP stack
> > alignment sequence.  Add support for that.
> > 
> > Fixes the following warning:
> > 
> >   vmlinux.o: error: objtool: vmw_host_printf+0xd: unknown CFA base reg 0
> > 
> > Reported-by: Arnd Bergmann <arnd@xxxxxxxx>
> > Closes: https://lore.kernel.org/cefefdd1-7b82-406d-8ff4-e4b167e45ee6@xxxxxxxxxxxxxxxx
> > Signed-off-by: Josh Poimboeuf <jpoimboe@xxxxxxxxxx>
> > ---
> >  arch/x86/include/asm/orc_types.h       | 1 +
> >  arch/x86/kernel/unwind_orc.c           | 8 ++++++++
> >  tools/arch/x86/include/asm/orc_types.h | 1 +
> >  tools/objtool/arch/x86/decode.c        | 3 +++
> >  tools/objtool/arch/x86/orc.c           | 5 +++++
> >  5 files changed, 18 insertions(+)
> > 
> > diff --git a/arch/x86/include/asm/orc_types.h b/arch/x86/include/asm/orc_types.h
> > index e0125afa53fb..b3cc7970fa54 100644
> > --- a/arch/x86/include/asm/orc_types.h
> > +++ b/arch/x86/include/asm/orc_types.h
> > @@ -37,6 +37,7 @@
> >  #define ORC_REG_R13			7
> >  #define ORC_REG_BP_INDIRECT		8
> >  #define ORC_REG_SP_INDIRECT		9
> > +#define ORC_REG_AX			10
> >  #define ORC_REG_MAX			15
> 
> I'm 'annoyed' at the placement of that register. I know its not
> important, but I can't silence my OCD saying that AX should come before
> DX.
> 
> Also, SP comes before BP and both before DI.
> 
> Something like so perhaps?

That bothered me too, but I was trying to keep the diffstat down.  But
sure, why not...

What about the switch statements?  Should we reorder them too while at
it?  If so, starting to feel like a separate patch.

-- 
Josh