[U-Boot] [RFC] 0/4] Add Pl330 DMA support

[Dinh Nguyen]

On 11/01/2016 03:09 PM, Marek Vasut wrote:
> On 11/01/2016 04:43 PM, Dinh Nguyen wrote:
>>
>>
>> On 10/29/2016 02:59 PM, Marek Vasut wrote:
>>> On 10/18/2016 05:50 PM, Dinh Nguyen wrote:
>>>> On Sun, Oct 16, 2016 at 10:59 AM, Marek Vasut <marex at denx.de> wrote:
>>>>> On 10/15/2016 12:10 AM, Dinh Nguyen wrote:
>>>>>>
>>>>>>
>>>>>> On 10/10/2016 10:52 AM, Dinh Nguyen wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>> This patchset adds the PL330 DMA driver to U-Boot. The SoCFPGA platform makes
>>>>>>> use of the PL330 to transfer zereos to the SDRAM which enables the ECC bit.
>>>>>>>
>>>>>>> This is only an RFC because after rebasing to v2016.11-rc1, I had to revert
>>>>>>> the following patch "a78cd8613204 ARM: Rework and correct barrier definitions"
>>>>>>> in order for this patch set to work correctly. With the above patch applied,
>>>>>>> after the SDRAM scrubbing is complete, the SPL is failing to find the MMC
>>>>>>> device.
>>>>>>>
>>>>>>
>>>>>> I kinda figured out why patch "a78cd8613204 ARM: Rework and correct
>>>>>> barrier definitions" is causing SPL to fail after the SDRAM scrub
>>>>>> operation. It appears that before the above mentioned patch, the dmb()
>>>>>> operation is defined as this:
>>>>>>
>>>>>> #define dmb()          __asm__ __volatile__ ("" : : : "memory")
>>>>>>
>>>>>> after the patch, the dmb is now:
>>>>>>
>>>>>> #define DMB    asm volatile ("dmb sy" : : : "memory")
>>>>>>
>>>>>> So the "dmb sy" as described in the ARM documentation is a "Full system
>>>>>> DMB operation". For whatever reason, this change is causing the SPL
>>>>>> error after the SDRAM scrub operation.
>>>>>
>>>>> Thanks for looking into this, can you poke into it some more ?
>>>>>
>>>>
>>>> Of course..
>>>
>>> The patch is causing a failure even in normal operation -- if you boot
>>> the socfpga board and type "reset", it won't reset with the patch.
>>>
>>
>> Can I ask what board are you seeing this on? I'm seeing that on my Atlas
>> board, it does not reset properly(even with the patch reverted). But
>> "reset" is working fine on my devkit, with and without the patch.
> 
> SoCKit and MCVEVK.
> 
> I tried bisecting it, it worked on 2016.07 and doesn't on 2016.09 , and
> I arrived at this patch. Maybe you can double-check this ?
>

Okay, I can check it on the SoCKit.

Dinh