Re: [PATCH 19/19] [INCOMPLETE] ARM: make return_address available for ARM_UNWIND

From: Dave Martin <dave.martin@linaro.org>
To: Keun-O Park <kpark3469@gmail.com>
Cc: Arnd Bergmann <arnd@arndb.de>,
	linux-arm-kernel@lists.infradead.org,
	Steven Rostedt <rostedt@goodmis.org>,
	sahara <keun-o.park@windriver.com>,
	Russell King <linux@arm.linux.org.uk>,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH 19/19] [INCOMPLETE] ARM: make return_address available for ARM_UNWIND
Date: Mon, 28 Jan 2013 12:50:23 +0000	[thread overview]
Message-ID: <20130128125023.GA2027@linaro.org> (raw)
In-Reply-To: <CA+KhAHYJus7+LsVhNa4RcxfpeTBtV=_vf2YDACtoMujo5c6tXA@mail.gmail.com>

On Mon, Jan 28, 2013 at 11:33:11AM +0900, Keun-O Park wrote:
> Hello guys,
> 
> Could you please review the patch of fixing bug first of returning
> wrong address when using frame pointer?
> I am wondering if the first patch is not delivered to the mailing.

I posted a similar patch to alkml a couple of months ago, but I got
no response and it looks like I forgot about it.

http://lists.infradead.org/pipermail/linux-arm-kernel/2012-November/129381.html

[...]

> 
> ~~~~~~~~~~~~~~~~~~~~~snip~~~~~~~~~~~~~~~~~~~~~~~~~
> From 3a60b536d22a2043d735c890a9aac9e7cb72de8f Mon Sep 17 00:00:00 2001
> From: sahara <keun-o.park@windriver.com>
> Date: Thu, 3 Jan 2013 17:12:37 +0900
> Subject: [PATCH 1/2] arm: fix returning wrong CALLER_ADDRx
> 
> This makes return_address return correct value for ftrace feature.
> unwind_frame does not update frame->lr but frame->pc for backtrace.
> And, the initialization for data.addr was missing so that wrong value
> returned when unwind_frame failed.
> 
> Signed-off-by: sahara <keun-o.park@windriver.com>
> ---
>  arch/arm/kernel/return_address.c |    5 +++--
>  1 files changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm/kernel/return_address.c b/arch/arm/kernel/return_address.c
> index 8085417..fafedd8 100644
> --- a/arch/arm/kernel/return_address.c
> +++ b/arch/arm/kernel/return_address.c
> @@ -26,7 +26,7 @@ static int save_return_addr(struct stackframe *frame, void *d)
>         struct return_address_data *data = d;
> 
>         if (!data->level) {
> -               data->addr = (void *)frame->lr;
> +               data->addr = (void *)frame->pc;
> 
>                 return 1;
>         } else {
> @@ -41,7 +41,8 @@ void *return_address(unsigned int level)
>         struct stackframe frame;
>         register unsigned long current_sp asm ("sp");
> 
> -       data.level = level + 1;
> +       data.level = level + 2;
> +       data.addr = NULL;

Can you explain why this is needed?  I think I concluded it wasn't
necessary, but you may be right -- I think if walk_stackframe()
fails to unwind the next frame just after data.level reaches zero,
then data.addr can remain unset and return_address() may return
uninitialised garbage.

Initialising data.addr to NULL before we start seems a good way
to avoid that.

Cheers
---Dave