[RFC PATCH v1 2/2] usb: host: add xhci-exynos to support Exynos SOCs

[Mathias Nyman]

On 1.12.2022 11.01, Arnd Bergmann wrote:
> On Thu, Dec 1, 2022, at 09:06, Greg Kroah-Hartman wrote:
>> On Thu, Dec 01, 2022 at 11:13:31AM +0900, Daehwan Jung wrote:
>>> This driver works with xhci platform driver. It needs to override
>>> functions of xhci_plat_hc_driver. Wakelocks are used for sleep/wakeup
>>> scenario of system.
>>
>> So this means that no other platform xhci driver can be supported in the
>> same system at the same time.
>>
>> Which kind of makes sense as that's not anything a normal system would
>> have, BUT it feels very odd.  This whole idea of "override the platform
>> driver" feels fragile, why not make these just real platform drivers and
>> have the xhci platform code be a library that the other ones can use?
>> That way you have more control overall, right?

Agree that overriding the generic platform driver xhci_hc_platform_driver
from this exynos driver is odd.

But I don't understand how this works.
Where are the hcds created and added when this xhci-exonys driver binds to
the device? all this driver does in probe is the overriding?

Am I missing something here?

> 
> Agreed, having another layer here (hcd -> xhci -> xhcd_platform ->
> xhcd_exynos) would fit perfectly well into how other SoC specific
> drivers are abstracted. This could potentially also help reduce
> the amount of code duplication between other soc specific variants
> (mtk, tegra, mvebu, ...) that are all platform drivers but don't
> share code with xhci-plat.c.
> 
> Alternatively, it seems that all of the xhci-exynos support could
> just be part of the generic xhci-platform driver: as far as I can
> tell, none of the added code is exynos specific at all, instead it
> is a generic xhci that is using the wakeup_source framework.

Sounds reasonable as well, and if some exynos specific code is needed
then just create a xhci_plat_priv struct for exynos and pass it in
of_device_id data like other vendors that use the generic
xhci-platform driver do.

-Mathias