Re: [PATCH] usb: acpi: fix boot hang due to early incorrect 'tunneled' USB3 device links

[Mathias Nyman <mathias.nyman@linux.intel.com>]

On 23.10.2024 21.07, Harry Wentland wrote:
> 
> 
> On 2024-10-23 12:50, Mario Limonciello wrote:
>> On 10/23/2024 07:12, Mathias Nyman wrote:
>>> On 22.10.2024 16.22, Mika Westerberg wrote:
>>>> The test case would be something like this:
>>>>
>>>> 1. Add "thunderbolt.host_reset=0" in the kernel command line.
>>>> 2. Boot with USB4 device connected (and so that it has USB 3.x device
>>>>      connected such as USB 3 memory stick).
>>>> 3. Since there is no device link Thunderbolt/USB4 can runtime suspend
>>>> freely.
>>>> 4. Once that happens the tunnels are gone, including the USB 3.x tunnel
>>>>      so the xHCI might see disconnect here.
>>>>
>>>> Also on resume path it will not be recreating the tunnel before xHCI
>>>> because there is no device link. I can try this on my side too if you
>>>> like.
>>>>
>>>
>>> You are right, I was able to reproduce it.
>>>
>>> Device link won't be created if BIOS created the tunnel, thunderbolt driver
>>> probes after this usb device is created, and "thunderbolt.host_reset=0" is set.
>>>
>>> Turning the device link "stateless" could possible help.
>>> It removes driver presence dependency but keeps correct suspend/resume and
>>> shutdown ordering.
>>> It should also help with boot hang/probe issues seen on the AMD platforms.
>>>
>>> Mario, Harry, does the following work for you?
>>
>> I didn't repro Harry's problem, but I did try on two OEM systems (Rembrandt and Phoenix based) with this change in place on a 6.12-rc4 base and didn't notice anything worse happening.
> 
> Yeah, the following diff works for me.
> 
> If you create a patch feel free to add my
> Tested-by: Harry Wentland <harry.wentland@amd.com>
> 
> Harry

Thanks for testing, I'll turn this into a proper v2 patch

-Mathias