Re: [PATCH] x86/kvm, objtool: Avoid fastop ENDBR from being sealed

[Josh Poimboeuf]

On Thu, Aug 18, 2022 at 01:06:16PM +0200, Peter Zijlstra wrote:
> On Thu, Aug 18, 2022 at 03:38:45PM +0800, Pengfei Xu wrote:
> >   And this "Missing ENDBR: andw_ax_dx+0x0/0x10 [kvm]" issue was
> >   reproduced again after ran syzkaller 5min on ADL-P.
> 
> Ha, indeed :/
> 
> > > > +/*
> > > > + * Create a dummy function pointer reference to prevent objtool from marking
> > > > + * the function as needing to be "sealed" (i.e. ENDBR converted to NOP by
> > > > + * apply_ibt_endbr()).
> > > > + */
> > > > +#define IBT_NOSEAL(fname)				\
> > > > +	".pushsection .discard.endbr.noseal\n\t"	\
> > > > +	_ASM_PTR fname "\n\t"				\
> > > > +	".popsection\n\t"
> > > > +
> 
> objtool/check.c:validate_ibt():
> 
>                 if (!strncmp(sec->name, ".discard", 8)                  ||
> 		    ...)
> 		    continue;
> 
> So yeah, that'll just get ignored. So at the very least we need to
> special case this section.

Indeed.  I swear that patch worked last night.

I also found that not all fastops use __FOP_FUNC().

Anyway I'll post a new (official) pair of patches shortly.

-- 
Josh