Re: [PATCH v6 0/5] usb: xhci: Add support for Renesas USB controllers

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

On 31.1.2020 10.40, Vinod Koul wrote:
> Hi Mathias,
> 
> On 30-01-20, 19:07, Mathias Nyman wrote:
>> On 25.1.2020 7.32, Vinod Koul wrote:
>>>>>>>>
>>>>>>>> On Mon, Jan 13, 2020 at 12:42 AM Vinod Koul <vkoul@kernel.org> wrote:
>>>>>>>>>
>>>>>>>>> This series add support for Renesas USB controllers uPD720201 and uPD720202.
>>>>>>>>> These require firmware to be loaded and in case devices have ROM those can
>>>>>>>>> also be programmed if empty. If ROM is programmed, it runs from ROM as well.
>>>>>>>>>
>>>>>>>>> This includes two patches from Christian which supported these controllers
>>>>>>>>> w/o ROM and later my patches for ROM support and multiple firmware versions,
>>>>>>>>> debugfs hook for rom erase and export of xhci-pci functions.
>>>>>>>>>

...

>>
>> There are a few other opens regarding this series. Mostly because I'm not (yet)
>> familiar with all the details, so I'll just just list them here.
>>
>> - Is it really enough to add the Renesas driver to Makefile before xhci-pci
>>    driver to make sure it gets matched and probed based on vendor/device id
>>    before xhci-pci driver is matched and probed based on pci class?
>>    What if the Renesas driver is a module and xhci-pci compiled in?
> 
> The driver loading rules are based on level and makefile order. So
> renesas will always be loaded before xhci driver. This is required since
> xhci claims all devices, so we need to make sure it loads before.
> 
> I think we should make it inbuilt driver rather than a module. xhci
> driver is only inbuilt.
> 
> If there is a better way to handle this, am all for it.
> 
>> - Previously probe didn't return before hcd's were added and everything set up.
>>    Now with request_firmware_nowait() probe returns early successfully, and the
>>    old xhci_pci_probe() which sets up everything is called later by the request
>>    firmware callback. So there could be whole new set of races possible.
>>    For example pci remove can be called mid firmware loading, or when the old
>>    xhci_pci_probe is still setting up things.
> 
> I think this is a valid concern. Let me think about how to solve this..
> maybe add a flag in remove which ensure remove doesnt run untill the
> probe/firmware callback is completed.

How about initiating async firmware loading before probe is called by using
DECLARE_PCI_FIXUP_*() hooks?

Probe would then just check if there is a valid running firmware, if not just
defer probe until later (return -EPROBE_DEFER).

No need for a separate Renesas xhci driver, no worries about which driver
claims the device first, no races because of probe returning successfully
too early.

Can we check the device for a valid running firmware without disrupting a
ongoing firmware write?

-Mathias