On 19.05.21 18:09, Frieder Schrempf wrote:
> On 06.05.21 10:32, Frieder Schrempf wrote:
>> On 06.05.21 03:04, Peng Fan (OSS) wrote:
>>> From: Peng Fan <peng.fan@xxxxxxx>
>>>
>>>
>>> V2:
>>> - Add R-b/A-b tag
>>> - Merge V1 patch 13 to V2 patch 6
>>> - Drop V1 patch 15
>>> - Merge V1 patch 16 to V2 patch 5 and add comments in patch 5 to explain
>>> details
>>> - Add explaination in patch 8 for "why the resets are not defined"
>>>
>>> This patchset is a pick up Lucas's gpcv2 work for i.MX8MM and several
>>> minor changes from me to make it could work with i.MX BLK-CTL driver.
>>>
>>> Thanks for Lucas's work and suggestion, Frieder Schrempf for collecting
>>> all the patches, Jacky Bai on help debug issues.
>>
>> I tested this series together with the BLK CTL patches by using the GPU and the display stack. Everything looks good to me.
>>
>> Tested-by: Frieder Schrempf <frieder.schrempf@xxxxxxxxxx>
>
> So after some more testing on different hardware I stumbled upon the problem that USB autosuspend doesn't work properly anymore.
>
> I have an onboard LTE module that is connected to OTG2 on the i.MX8MM. When using the mainline TF-A (that enables USB power-domains by default) and removing the power-domain control from the kernel, the device comes up after a few seconds and is enumerated on the bus.
>
> Now, when I let the kernel control the power-domains, the device comes up at boot, but isn't enumerated on the USB bus. As soon as I disable autosuspend for the port, it comes up.
>
> Is this something that needs to be fixed on the USB driver side or is something to be considered for the GPCv2 driver?
So I think this is something that needs to be covered on the USB driver side. I would expect that a device appearing on the bus should resume the autosuspended bus, but I don't really know much about USB so there might be other things I miss. For now I will disable autosuspend in this case.
A different, probably more severe problem is that I was still able to reliably run into lockups with suspend/resume and the GPU. I thought I had tested this before as it was one of the things that already failed with the previous implementation, but I must have missed something as it still fails with kernel v5.12.1 + v2 of the GPC patches.
This is how I run into the lockup:
echo mem > /sys/power/state # Sleep
# Wake up again
glmark2-es2-drm # Use the GPU
# Device locks up
Peng, is this something you can reproduce?
