Warning dump on OMAP-L138 when g_zero module is removed

[Sekhar Nori]

Hi Alexandre,

On Sunday 06 August 2017 01:37 AM, Alexandre Bailon wrote:
> Hi Sekhar,
> On 07/21/2017 05:18 PM, Alexandre Bailon wrote:
>> Hi Sekhar,
>>
>> On 07/10/2017 01:00 PM, Sekhar Nori wrote:
>>> On Thursday 06 July 2017 10:43 PM, Alexandre Bailon wrote:
>>>> On 06/29/2017 03:50 PM, Sekhar Nori wrote:
>>>>> Hi Alexandre, Bin,
>>>>>
>>>>> With latest linux-next, I see a warning dump when I remove g_zero[1] on
>>>>> OMAP-L138 LCDK board. I am building the kernel with davinci_all_defconfig.
>>>>>
>>>>> It is not present in latest mainline because the warnings are introduced
>>>>> with CPPI DMA getting enabled in latest -next.
>>>>>
>>>>> Subsequent insertion of g_ether leads to a warning too[2], although the
>>>>> gadget seems to work (ping test).
>>>>>
>>>>> Since these are pretty annoying, it will be nice to get rid of them. I
>>>>> have not been able to debug any further. But if you have
>>>>> ideas/experiments to try, I can do that.
>>>
>>>> I got a lot of these warnings during development but it was only for the
>>>> host mode.
>>>> If I remember correctly, the cause was a race between the teardown
>>>> function and the interrupt handler.
>>>> The teardwon descriptor was pop from the queue by the interrupt handler,
>>>> preventing cppi41_tear_down_chan() to get it, which will result after
>>>> some retries to a couple of warnings.
>>>>
>>>> I will take a look to see if that is the same issue.
>> That is not the same issue. Still, for some reasons, we never get the
>> teardown descriptor.
>> Two possibilities:
>>  - Like the issue I had, the teardown is pop at the wrong place.
>>  - The teardown doesn't complete or the descriptor is not queued.
> The teardown descriptor is not queued. I have figured out the reason.
> The function cppi41_dma_channel_abort() enable the teardown by setting the bit corresponding to the endpoint number.
> The address of the register is USB_TDOWN but actually the address is wrong in the case of DA8xx.
> Using the right address (0x1c) fix the teardown warnings.
> I will work on a fix.
>> I think we should eliminate the first one before to work on the second one.
>> I think adding some logs to cppi41_pop_desc() could help to figure out
>> what is happening.
>>>
>>> Thanks! I also see similar warnings under fast ping traffic and g_ether
>>> inserted on LCDK. They dont come immediately though. Only after an hour
>>> or so under traffic.
> The fix should also fix this issue.
>> I would not have expected to have them during a ping but this probably
>> happens because some USB packet have timed out, which cause may a teardown.
>>>
>>> I can those provide logs too if its going to be helpful.
>> Currently, I think we should focus on the warnings that happens during
>> the rmmod as they are always reproducible.
>>
>> Thanks,
>> Alexandre
>>
> Note that the teardown is not the only thing that doesn't work correctly.
> I think same kind of issue should apply to the autoreq and the dma mode.
> I will work on a patch that should solve all of them.

Glad you found the issue. When you have a patch, I will be happy to test
it out.

Thanks,
Sekhar