[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20241004082029.GG18071@noisy.programming.kicks-ass.net>
Date: Fri, 4 Oct 2024 10:20:29 +0200
From: Peter Zijlstra <peterz@...radead.org>
To: Josh Poimboeuf <jpoimboe@...nel.org>
Cc: Ard Biesheuvel <ardb@...nel.org>, x86@...nel.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH] objtool: Detect non-relocated text references
On Fri, Oct 04, 2024 at 12:28:47AM -0700, Josh Poimboeuf wrote:
> On Fri, Oct 04, 2024 at 08:54:16AM +0200, Ard Biesheuvel wrote:
> > On Fri, 4 Oct 2024 at 02:31, Josh Poimboeuf <jpoimboe@...nel.org> wrote:
> > > However there are some "lea xxx(%rip)" cases which don't use relocations
> > > because they're referencing code in the same translation unit.
> >
> > input section
>
> "in the same translation unit and section" ?
>
> > > However if we enable -fpie, the compiler would start generating them and
> > > there would definitely be bugs in the IBT sealing.
> > >
> >
> > -fpie is guaranteed to break things, but even without it, Clang may
> > issue RIP-relative LEA instructions (or LLD when it performs
> > relaxations), so this is definitely worth addressing even if we don't
> > enable -fpie.
>
> I haven't seen this with Clang either. Also, objtool runs before the
> linker so LLD relaxations shouldn't matter.
LTO might have a few more cases, the input sections are bigger there.
But even there we run before the final link stage.
Powered by blists - more mailing lists