Date: Wed, 8 Mar 2017 15:29:59 -0600
From: Josh Poimboeuf <jpoimboe@redhat.com>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>, Pavel Machek <pavel@ucw.cz>,
        kernel list <linux-kernel@vger.kernel.org>,
        Ingo Molnar <mingo@kernel.org>, Andrew Lutomirski <luto@kernel.org>,
        Borislav Petkov <bp@alien8.de>, Brian Gerst <brgerst@gmail.com>,
        Denys Vlasenko <dvlasenk@redhat.com>, Peter Anvin <hpa@zytor.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Steven Rostedt <rostedt@goodmis.org>
Subject: Re: v4.10: kernel stack frame pointer .. has bad value (null)
Message-ID: <20170308212959.ctb2ktqkg5l6r2xj@treble>
References: <20170225050439.7dplheb6nyne4nkm@treble>
 <20170302234514.3qcqdozibcltkdai@treble>
 <20170306163807.GA20689@amd>
 <20170307173821.yknj5htr7plgdwxv@treble>
 <CA+55aFw+WVjwPBPKpnKcZUkqX=mYWv7A6M44uJpmWuzGU0zpqw@mail.gmail.com>
 <CALCETrWHdra-Po0rNrHn-cV2G=AM2pWR45rRf4t2RZfQR8Lqjg@mail.gmail.com>
 <20170307182855.262ezbon2pm67qfd@treble>
 <CA+55aFyjEuRLN6PMZZqi6k0rA6=AsTTZeN37eyyJM5ig-mnM4w@mail.gmail.com>
 <20170308173703.2h57rsltma3smbcm@treble>
 <CA+55aFw8ZVHtTDT8jCCtwP=oZFVKnDZVa3bMFdzt6+sGqFNeKw@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <CA+55aFw8ZVHtTDT8jCCtwP=oZFVKnDZVa3bMFdzt6+sGqFNeKw@mail.gmail.com>
User-Agent: Mutt/1.6.0.1 (2016-04-01)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1852
Lines: 48

[adding Steven Rostedt to CC as an FYI]

On Wed, Mar 08, 2017 at 10:25:01AM -0800, Linus Torvalds wrote:
> On Wed, Mar 8, 2017 at 9:37 AM, Josh Poimboeuf <jpoimboe@redhat.com> wrote:
> > - CONFIG_FUNCTION_GRAPH_TRACER sets it on x86-32 because of a gcc bug
> >   where the stack gets aligned before the mcount call.  This issue
> >   should be mostly obsolete as most modern compilers now have -mfentry.
> >   We could make it dependent on CC_USING_FENTRY.
> 
> Yeah. At some point we might even upgrade the compiler requirements to
> no longer accept the mcount model.

The plot slightly thickens...

So I was mistaken about this problem not existing with newer versions of
gcc, because the x86-32 ftrace code doesn't use -mfentry.  It still
relies on mcount.  So CONFIG_FUNCTION_GRAPH_TRACER will still need
-maccumulate-outgoing-args for *all* versions of gcc on x86-32.

(Of course, that situation would improve if ftrace on x86-32 were ported
to use -mfentry.)

Also, since -Os tells gcc to ignore -maccumulate-outgoing-args, this
means that CONFIG_FUNCTION_GRAPH_TRACER with mcount needs a dependency
on CONFIG_CC_OPTIMIZE_FOR_PERFORMANCE.

I suspect these issues also affect x86-64 with gcc 4.4.x and 4.5.x,
which corresponds to the window after the funky DRAP prologue was
introduced but before -mfentry was introduced.

In summary, here are the changes I'm looking at:

- set -maccumulate-outgoing-args if CONFIG_FUNCTION_GRAPH_TRACER &&
  !CC_USING_ENTRY
  (for both 32- and 64-bit)

- somehow make CONFIG_FUNCTION_GRAPH_TRACER depend on either
  CC_USING_FENTRY or CONFIG_CC_OPTIMIZE_FOR_PERFORMANCE
  (for both 32- and 64-bit)

  (not sure how to do that -- maybe just fail the build in the
  graph tracer + mcount + '-Os' case)

- set -maccumulate-outgoing-args if CONFIG_JUMP_LABEL && gcc < 4.5.2
  (for both 32-bit and 64-bit)

-- 
Josh