Re: [PATCH] arm64: remove redundant FRAME_POINTER kconfig option

From: Will Deacon <will.deacon@arm.com>
To: Mark Rutland <mark.rutland@arm.com>
Cc: Yang Shi <yang.shi@linaro.org>,
	catalin.marinas@arm.com, linaro-kernel@lists.linaro.org,
	linux-kernel@vger.kernel.org,
	linux-arm-kernel@lists.infradead.org
Subject: Re: [PATCH] arm64: remove redundant FRAME_POINTER kconfig option
Date: Fri, 6 Nov 2015 15:42:58 +0000	[thread overview]
Message-ID: <20151106154257.GP6087@arm.com> (raw)
In-Reply-To: <20151106125002.GA8116@leverpostej>

On Fri, Nov 06, 2015 at 12:50:02PM +0000, Mark Rutland wrote:
> On Fri, Nov 06, 2015 at 12:30:09PM +0000, Will Deacon wrote:
> > On Wed, Nov 04, 2015 at 09:37:51AM -0800, Yang Shi wrote:
> > > FRAME_POINTER is defined in lib/Kconfig.debug, it is unnecessary to redefine
> > > it in arch/arm64/Kconfig.debug.
> > 
> > It might be worth noting that this adds a dependency on DEBUG_KERNEL
> > for building with frame pointers. I'm ok with that (it appears to be
> > enabled in defconfig and follows the vast majority of other archs) but
> > it is a change in behaviour.
> > 
> > With that:
> > 
> >   Acked-by: Will Deacon <will.deacon@arm.com>
> 
> The code in arch/arm64/kernel/stacktrace.c assumes we have frame
> pointers regardless of FRAME_POINTER. Depending on what the compiler
> decides to use x29 for, we could get some weird fake unwinding and/or
> dodgy memory accesses.
> 
> I think we should first audit the uses of frame pointers to ensure that
> they are guarded for !FRAME_POINTER.

Good point. The perf callchain code suffers from a similar issue.

Will