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:   Tue, 11 Jul 2023 09:15:40 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Masami Hiramatsu <mhiramat@kernel.org>
Cc:     Petr Pavlu <petr.pavlu@suse.com>, tglx@linutronix.de,
        mingo@redhat.com, bp@alien8.de, dave.hansen@linux.intel.com,
        hpa@zytor.com, samitolvanen@google.com, x86@kernel.org,
        linux-trace-kernel@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [RFC PATCH 2/2] x86/kprobes: Prohibit probing on compiler
 generated CFI checking code
Message-ID: <20230711071540.GC3062772@hirez.programming.kicks-ass.net>
References: <168899125356.80889.17967397360941194229.stgit@devnote2>
 <168899127520.80889.15418363018799407058.stgit@devnote2>
 <20230710161643.GB3040258@hirez.programming.kicks-ass.net>
 <20230711085837.fac80c964ea7667cb75bd6e5@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230711085837.fac80c964ea7667cb75bd6e5@kernel.org>
Precedence: bulk

On Tue, Jul 11, 2023 at 08:58:37AM +0900, Masami Hiramatsu wrote:

> > Hmm, so I was thinking something like the below, which also catches
> > things when we rewrite kCFI to FineIBT, except I don't think we care if
> > the FineIBT callsite gets re-written. FineIBT only relies on the __cfi_
> > symbol not getting poked at (which the previous patches should ensure).
> 
> Oh, is FineIBT different from kCFI? I thought those are same. But anyway
> for kCFI=y && FineIBT=n case, I think this code still needed.

Ah, I forgot to answer your other question; FineIBT is dynamically
patched since it relies on optional hardware features, only if the
hardware has IBT support can FineIBT work.

So yeah, your check is definitely needed. And I think it'll 'gracefully'
fail when FineIBT is patched in because the instructions aren't exactly
the same.

And since FineIBT has most of the magic in the __cfi_ prefix, which
isn't patched per the previous patch, I think we're good on the call-site.

> > So I tihnk I'm ok with the above, just adding the below for reference
> > (completely untested and everything).
> 
> I wonder the distance can be used outside of x86. CFI implementation
> depends on the arch?

Currently only arm64 and x86_64 have CFI, and while the particulars are a
little different, they do have a lot in common, including the reporting.
IIRC the arm64 variant is a little simpler because of fixed instruction
size. There is no worry someone will jump in the middle of an
instruction stream.