[PATCH] usb: gadget: ether: split start/stop from init/halt

[Marek Vasut]

On 1/20/23 18:26, Niel Fourie wrote:
> Hi Marek

Hi,

> On 18/12/2022 02:51, Marek Vasut wrote:
>> On 12/16/22 17:35, lusus at denx.de wrote:
>>> Hi Marek,
>>
>> Hi,
>>
>> [...]
>>
>>>> Should "[PATCH v2] net: eth-uclass: revalidate priv after stop() in
>>>> eth_halt()" be dropped ? It seems this patch fully replaces it.
>>>
>>> That old patch only exists for in case there were showstopper issues 
>>> with the new patch which I missed, or as a stopgap if major changes 
>>> were needed first. I did my best to test the new patch on the 
>>> hardware I have, but if any other gadget drivers were to misbehave, I 
>>> would not know about it, for example. But if you are happy with this 
>>> new patch, that old patch could gladly be dropped.
>>
>> The new patch is by far preferable. If you can sort out the 
>> network_started comment, I would like to pick this via usb/next.
> 
> I did some more testing and I found that keeping network_started as it 
> is works best, so I did not change it for v2 of this patch.

This doesn't answer the question whether netconsole connection gets 
broken or not by this change, it think the netconsole connection would 
get broken and that's not good.

> Unfortunately during local testing (on imx8mp), I uncovered an issue in 
> the dwc3 gadget driver causing it to hang. I have a workaround, but I am 
> still looking into fixing the root cause, and that would have to be 
> fixed in a separate patch series.

You can send the workaround as [RFC][PATCH] so it wouldn't get included 
accidentally and possibly CC Peng Fan from NXP.

> Similar issues might also still crop up with other gadget drivers.

[...]