linux-kernel - Re: vmlinux.o: warning: objtool: stackleak_erase()+0x35: call to ftrace_likely

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 5 Apr 2022 10:03:35 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     Steven Rostedt <rostedt@...dmis.org>
Cc:     Kees Cook <keescook@...omium.org>,
        Josh Poimboeuf <jpoimboe@...hat.com>, kbuild-all@...ts.01.org,
        linux-kernel@...r.kernel.org, kernel test robot <lkp@...el.com>
Subject: Re: vmlinux.o: warning: objtool: stackleak_erase()+0x35: call to
 ftrace_likely_update() leaves .noinstr.text section

On Mon, Apr 04, 2022 at 03:25:02PM -0400, Steven Rostedt wrote:
> On Mon, 4 Apr 2022 11:27:47 -0700
> Kees Cook <keescook@...omium.org> wrote:
> 
> > I got CCed on this because of stackleak_erase() triggering, this, but it
> > appears to be an existing issue for other callers too. It looks like
> > __branch_check__() shouldn't be emitting ftrace_likely_update() calls
> > for noinstr functions...
> > 
> > I have no idea how a macro is meant to check for function attributes,
> > though... :(
> 
> Ignore it. It's the branch tracer enabled (that is, every "if ()" is being
> traced).
> 
> Although I still use the branch profiler, I do not believe anyone uses the
> branch tracer. The branch profiler updates an array of counters that tells
> when the branch was true or false, the branch tracer actually traces
> (records an event) for every branch in the system!

\o/

> I think I'll just send a patch to nuke the tracer. I'm sure Peter Zijlstra
> will be happy when I do that. But I still want the profiler, as I find that
> useful.

It'll explode the moment Lai's entry rework goes through. That'll make
us run C code before we switch to the kernel address space, so your
counters will not exist/be-mapped and *BOOM*.