Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 13 Apr 2023 12:30:48 -0700
From:   Josh Poimboeuf <jpoimboe@kernel.org>
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     x86@kernel.org, linux-kernel@vger.kernel.org,
        Miroslav Benes <mbenes@suse.cz>
Subject: Re: [PATCH 3/3] objtool: Generate ORC data for __pfx code
Message-ID: <20230413193048.biwxfseynqyc4qkx@treble>
References: <cover.1681331135.git.jpoimboe@kernel.org>
 <bc3344e51f3e87102f1301a0be0f72a7689ea4a4.1681331135.git.jpoimboe@kernel.org>
 <20230413112426.GM4253@hirez.programming.kicks-ass.net>
 <20230413152933.cxhmocvbdlucvizx@treble>
 <20230413192449.GQ4253@hirez.programming.kicks-ass.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <20230413192449.GQ4253@hirez.programming.kicks-ass.net>
Precedence: bulk

On Thu, Apr 13, 2023 at 09:24:49PM +0200, Peter Zijlstra wrote:
> On Thu, Apr 13, 2023 at 08:29:33AM -0700, Josh Poimboeuf wrote:
> > On Thu, Apr 13, 2023 at 01:24:26PM +0200, Peter Zijlstra wrote:
> > > > +	if (!insn->cfi) {
> > > > +		/*
> > > > +		 * This can happen if stack validation isn't enabled or the
> > > > +		 * function is annotated with STACK_FRAME_NON_STANDARD.
> > > > +		 */
> > > > +		return 0;
> > > > +	}
> > > > +
> > > > +	/* Propagate insn->cfi to the prefix code */
> > > > +	cfi = cfi_hash_find_or_add(insn->cfi);
> > > > +	for (; prev != insn; prev = next_insn_same_sec(file, prev))
> > > > +		prev->cfi = cfi;
> > > > +
> > > >  	return 0;
> > > >  }
> > > 
> > > FWIW, this makes the whole thing hard rely on the prefix being single
> > > byte NOPs -- which they are, but perhaps we should assert this?
> > 
> > Couldn't they be any stack-invariant instructions?
> 
> Hmm, I was thikning that since we don't know the size of the
> instructions being written, we need CFI for all offsets. But perhaps,
> since we do a left-match on IP, only one entry at the __pfx+0 location
> would work?

Right, while in objtool (almost) every insn has insn->cfi, the actual
ORC entries only get created at the boundaries of change.

-- 
Josh