Re: [PATCH 1/2] objtool: Optimize re-writing jump_label

[Josh Poimboeuf]

On Fri, Oct 08, 2021 at 12:03:31PM +0200, Peter Zijlstra wrote:
> On Thu, Oct 07, 2021 at 11:55:50PM -0700, Josh Poimboeuf wrote:
> > On Thu, Oct 07, 2021 at 11:22:12PM +0200, Peter Zijlstra wrote:
> > > There's no point to re-write the jump_label NOP when it's already a NOP.
> > > 
> > > Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> > > ---
> > >  tools/objtool/check.c |    2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > --- a/tools/objtool/check.c
> > > +++ b/tools/objtool/check.c
> > > @@ -1397,7 +1397,7 @@ static int handle_jump_alt(struct objtoo
> > >  		return -1;
> > >  	}
> > >  
> > > -	if (special_alt->key_addend & 2) {
> > > +	if ((special_alt->key_addend & 2) && orig_insn->type != INSN_NOP) {
> > >  		struct reloc *reloc = insn_reloc(file, orig_insn);
> > >  
> > >  		if (reloc) {
> > 
> > While you're at it, a comment would be very helpful for that whole
> > clause.
> 
> Like so?
> 
> --- a/tools/objtool/check.c
> +++ b/tools/objtool/check.c
> @@ -1397,7 +1397,16 @@ static int handle_jump_alt(struct objtoo
>  		return -1;
>  	}
>  
> -	if (special_alt->key_addend & 2) {
> +	/*
> +	 * When, for whatever reason, the jump-label site cannot emit a right
> +	 * sized NOP, then it can use Bit-1 of the struct static_key pointer to
> +	 * indicate this instruction should be NOP'ed by objtool.
> +	 *
> +	 * Also see arch/x86/include/asm/jump_label.h:arch_static_branch(),
> +	 * where we leave the assembler to pick between jmp.d8 and jmp.d32
> +	 * based on destination offset.
> +	 */
> +	if ((special_alt->key_addend & 2) && orig_insn->type != INSN_NOP) {
>  		struct reloc *reloc = insn_reloc(file, orig_insn);
>  
>  		if (reloc) {
> 

Much better!  Can you also mention why it's checking INSN_NOP?  The "run
objtool twice" case is far from obvious.

-- 
Josh