From mboxrd@z Thu Jan  1 00:00:00 1970
From: Hans de Goede <hdegoede@redhat.com>
Date: Tue, 5 Apr 2016 10:44:42 +0200
Subject: [U-Boot] [PATCH 1/2] arm: Replace
 v7_maint_dcache_all(ARMV7_DCACHE_CLEAN_INVAL_ALL) with asm code
In-Reply-To: <5703784E.2030300@redhat.com>
References: <1459794709-21601-1-git-send-email-hdegoede@redhat.com>
 <20160404235956.GS23166@bill-the-cat> <5703784E.2030300@redhat.com>
Message-ID: <57037AFA.3010806@redhat.com>
List-Id: <u-boot.lists.denx.de>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: u-boot@lists.denx.de

Hi,

On 05-04-16 10:33, Hans de Goede wrote:
> Hi,
>
> On 05-04-16 01:59, Tom Rini wrote:
>> On Mon, Apr 04, 2016 at 08:31:48PM +0200, Hans de Goede wrote:
>>
>>> v7_maint_dcache_all() does not work reliable when build with gcc6,
>>> see: https://bugzilla.redhat.com/show_bug.cgi?id=1318788
>>
>> So, I see on the bug you want to know if U-Boot is OK with this patch
>> series being the fix.  At the high level, yes, borrowing code from the
>> Linux Kernel is a good clean-up and I'd like to see this series could
>> clean up things a little more and borrow from cache-v7.S when we could.
>>
>> But I'd also push back on the toolchain team.  Are they happy saying
>> "that code is just too fragile, it's probably relying on undefined
>> behavior, investigation concluded" ?
>
> You should be able to reproduce the problems we're seeing on sunxi
> yourself. Add a "noinline" to "v7_clean_inval_dcache_level_setway"
> and then boot on say a lime2 you should see a data abort
> after "Starting kernel" instead of, well, the kernel starting.
>
> Given that just adding a noinline already breaks the code with
> gcc-5 the "too fragile" thing was my own conclusion really.
>
> I'll ask for some more info from the toolchain team in the bug
> (note you're welcome to join the discussion in bugzilla yourself
>   creating an account if you don't have one only requires an email
>   address).

Answer from Jakub Jelinek who has been helping me from the toolchain side
with this bug sofar:

"That would need to answer somebody familiar with the ARM cache flushing
instructions.  All I can say is that I haven't found any obvious errors on the
toolchain side when compiling the code.  Ask somebody from Linaro or ARM?"

Regards,

Hans