[REGRESSION] Arndale Octa panics when booting 3.14-rc1

[Tushar Behera]

On 4 March 2014 17:33, Tomasz Figa <tomasz.figa at gmail.com> wrote:
> On 04.03.2014 13:01, Tushar Behera wrote:
>>
>> On 4 March 2014 16:09, Javi Merino <javi.merino at arm.com> wrote:
>>>
>>> On Tue, Mar 04, 2014 at 10:30:19AM +0000, Sylwester Nawrocki wrote:
>>>>
>>>> On 04/03/14 11:16, Javi Merino wrote:
>>>>>
>>>>> Yes, with [1] applied I don't get a kernel panic but the kernel fails
>>>>> to boot later on with an Imprecise external abort.  Removing the mdma
>>>>> nodes from the dts gets rid of that.  I guess what's missing is what
>>>>> you said: clocks for the mdma devices.
>>>>
>>>>
>>>> Is removing mdm0 node enough to fix the boot failure, or both have to be
>>>> removed ?
>>>
>>>
>>> Actually, you it's only mdma1.  Just removing the mdma1 node from the
>>> dt fixes the imprecise external abort.
>>>
>>
>> MDMA1 can support both secure and non-secure AXI transactions, the
>> actual behaviour is controlled by trustzone software. It may be the
>> case that MDMA1 is configured to be used in secure mode only, hence
>> accessing it in non-secure mode is causing the oops.
>>
>> Right now, the only solution looks like disabling this node in
>> Arndale-Octa dts file.
>>
>> --- a/arch/arm/boot/dts/exynos5420-arndale-octa.dts
>> +++ b/arch/arm/boot/dts/exynos5420-arndale-octa.dts
>> @@ -354,4 +354,16 @@
>>                  samsung,i2s-controller = <&i2s0>;
>>                  samsung,audio-codec = <&i2s_stub>;
>>          };
>> +
>> +       amba {
>> +               mdma1: mdma at 11C10000 {
>> +                       /*
>> +                        * MDMA1 can support both secure and non-secure
>> +                        * AXI transactions. When this is enabled in the
>> kernel
>> +                        * for boards that run in secure mode, we are
>> getting
>> +                        * imprecise external aborts causing the kernel to
>> oops.
>> +                        */
>> +                       status = "disabled";
>> +               };
>> +       };
>>
>> If it works, I will submit this patch.
>
>
> On Exynos 4 SoCs there were two instances of MDMA1, one secure and one
> non-secure. Isn't it the case for Exynos5420 as well? If yes, the common
> DTSI could be changed to always use the non-secure one.
>

AFAICS, there is only one MDMA1 controller in Exynos5420.

> Best regards,
> Tomasz



-- 
Tushar Behera