Dynamically adding a PCI subfunction

[Christian König]

Am 09.10.2017 um 17:57 schrieb Ilia Mirkin:
> On Mon, Oct 9, 2017 at 11:45 AM, Christian König
> <ckoenig.leichtzumerken at gmail.com> wrote:
>> Am 09.10.2017 um 16:41 schrieb Ilia Mirkin:
>>> Hello,
>>>
>>> As a bit of background, all NVIDIA GPUs since GT215 have an audio
>>> subfunction for HDMI(/DP) audio to be sent to the sink. This generally
>>> works.
>>>
>>> However some, especially laptop, devices come up with that function
>>> disabled. We have a quirk to enable it when coming back from runpm,
>>> but that doesn't help the init case. Basically we have to write a bit
>>> to the PCI config space:
>>>
>>>
>>> https://github.com/torvalds/linux/blob/v4.12/drivers/gpu/drm/nouveau/nouveau_drm.c#L783
>>>
>>> (MMIO 0x88000 is an alias for the PCI config space)
>>>
>>> This works for runtime pm resume, since the device was originally
>>> there and we just have to make sure the underlying device agrees with
>>> it, but when it's missing on boot, we have to convince linux that it
>>> exists, bind a driver, etc.
>>>
>>> What's the best way of going about doing that?
>>
>> Sounds similar to my work about resizing BARs.
>>
>> I would just try to enable the device and then trigger a rescan of the BUS
>> where it is attached (similar to echo 1 >
>> /sys/bus/pci/devices/$pci_id_of_your_bus/rescan).
>>
>> There is certainly a function in the PCI subsystem you just need to call for
>> that.
> I'm not exactly familiar with the pci subsystem. I know that hotplug
> is a thing, at least in theory. I figured it'd be something related to
> that, since this is in essence a random device appearing on the bus
> after the initial scan. Is a bus rescan safe on a fully initialized
> system?
>
> Either way, some specific functions to call or look at would be great.

Well I would just try to use pci_rescan_bus(), as far as I know that 
should be save even when devices are already configured.

Christian.

>
> Cheers,
>
>    -ilia