Re: Regression: memory corruption on Atmel SAMA5D31

[Peter Rosin]

2022-07-28 at 10:39, Tudor.Ambarus@xxxxxxxxxxxxx wrote:
> On 7/28/22 10:45, Tudor.Ambarus@xxxxxxxxxxxxx wrote:
>> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
>>
>> On 7/13/22 19:01, Tudor.Ambarus@xxxxxxxxxxxxx wrote:
>>> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
>>>
>>> Hi, Peter,
>>>
>>> Thanks for the patience. I was still out of office last week,
>>> but now I have some news.
>>>
>>> On 6/27/22 19:53, Tudor.Ambarus@xxxxxxxxxxxxx wrote:
>>>> I think these are the last less invasive
>>>> changes that I try, I'll have to rewrite the logic anyway.
>>>
>>> I've chopped the driver to use virt-dma (check [1]). It's not clean, but
>>> it works and one can see how the logic is changed. Unfortunately the mem
>>> corruption is still present on high loads. Maybe it's a coherency problem.
>>> I need more time on it. Will get back to you.
>>>
>>> Cheers,
>>> ta
>>>
>>> [1] To github.com:ambarus/linux-0day.git
>>>    a7351e6f4c12..1557e0df0fd0  at-hdmac-virt-dma -> at-hdmac-virt-dma
>>
>> Hi, Peter,
>>
>> Does this [1] one line patch solve the mem corruption on your side?
>> Even if yes, there are still bugs in at-hdmac that can be squashed by
>> using virt-dma. I'd like to follow up with patches that integrate
>> virt-dma logic in at-hdmac.
>>
>> Cheers,
>> ta
>>
>> [1] https://lore.kernel.org/linux-mtd/20220728074014.145406-1-tudor.ambarus@xxxxxxxxxxxxx/T/#u
> 
> Hi, Peter,
> 
> Looks like I've already caught an oops in at-hdmac driver when not using virt-dma,
> see below. Would you please test with all the patches from [2] instead of just
> using the patch from [1]? I've run stress tests over night by using [2] and
> everything went fine on my side.
> 
> Cheers,
> ta
> 
> [2] To github.com:ambarus/linux-0day.git
>  * [new branch]                at-hdmac-virt-dma-2nd-iteration -> at-hdmac-virt-dma-2nd-iteration

Hi Tudor,

This last one feels very promising! It's been running for a few hours without
incidents, so even if it isn't fixed it's several magnitudes better.

I'll leave it running for the night. Fingers crossed...

Cheers,
Peter