problem with dwmmc ? panic's on bad filesystem...

[Michal Meloun]

On 12.03.2021 13:18, Bjoern A. Zeeb wrote:
> On 12 Mar 2021, at 12:01, Søren Schmidt wrote:
> 
>> On 12 Mar 2021, at 11.09, Emmanuel Vadot <manu at bidouilliste.com> wrote:
>>>
>>> On Fri, 12 Mar 2021 09:47:40 +0100
>>> Søren Schmidt <soren.schmidt at gmail.com> wrote:
>>>
>>>> On 11 Mar 2021, at 23.30, Jesper Schmitz Mouridsen <jsm at FreeBSD.org> 
>>>> wrote:
>>>>>
>>>>>
>>>>> On 11.03.2021 21.29, Søren Schmidt wrote:
>>>>>> Hi
>>>>>>
>>>>>> I have had this panic several times on rk3399 based systems when 
>>>>>> the filesystem has gone bad (and just needs a fsck).
>>>>>> It doesn?t happen on Marvell a3720, so its not arm64 specific, 
>>>>>> more likely dwmmc specific or maybe the dma on the rk3399 ?
>>>>>>
>>>>>> 13-stable of today, but problem has been around for a long time..
>>>>>>
>>>>> <snip>
>>>>>
>>>>> Check under known issues https://wiki.freebsd.org/arm/RockChip.
>>>>>
>>>>> Should be fixed in 
>>>>> https://cgit.freebsd.org/src/commit/?id=8727c174b0fe44766bb7ea765dac6d5f82818103 
>>>>>
>>>>>
>>>>> But perhaps it is not included in stable/13 or it is not totally 
>>>>> fixed?
>>>>
>>>> That fix is indeed in stable13, so it looks like its not really 
>>>> fixed, or this is another problem...
>>>>
>>>> -- 
>>>> Søren Schmidt
>>>> sos at deepcore.dk / sos at freebsd.org
>>>> "So much code to hack, so little time"
>>>
>>> Seems like it's the same issue.
>>> Your hitting EFBIG from bus_dmamap_load.
>>> Does that happens everytime ?
>>
>> Yes it does.
>>
>>> Does that happens if you disable journaling or softupdate ?
>>
>> Journalling seems to be what triggers this error.
>>
>> The last (error) request contains 0 segs and a non %blocksize length 
>> if that helps.
> 
> This sounded familiar;  I am not sure what happened to the comments from 
>   this:  https://reviews.freebsd.org/D26735
> 
> 
It was reincarnated to 
https://cgit.freebsd.org/src/commit/sys/arm64/arm64/busdma_bounce.c?id=f635cef2a4206fb4e077412c493f218a6a26a3daSøren,
is the problem reproducible also on current ?
If yes are you able to dump VA and size of buffer passed to last call to 
bus_dmamap_load?

Michal