Re: [PATCH] PCI: Fix PCIe SBR dev/link wait error

[Lukas Wunner <lukas@wunner.de>]

On Mon, Dec 01, 2025 at 10:46:00PM +0800, guanghuifeng@linux.alibaba.com wrote:
> 在 2025/12/1 21:26, Lukas Wunner 写道:
> > On Mon, Dec 01, 2025 at 08:56:10PM +0800, guanghuifeng@linux.alibaba.com wrote:
> > > 2025/12/1 18:03, Lukas Wunner :
> > > > It seems highly unusual that the different functions of the same
> > > > physical device require different delays until they're accessible.
> > > >  I don't think we can accept such a sweeping change wholesale without
> > > > more details, so please share what the topology looks like (lspci -tv),
> > > > what devices are involved (lspci -vvv) and which device requires
> > > > extra wait time for some of its functions.
> 
> 1. Currently, there are significant differences in reset recovery times
> among some PCIe multifunction devices, especially when the functions and
> complexities vary greatly, including some devices used for testing purposes.

Which devices are we talking about exactly?  For internal test devices,
upstream changes to accomodate to quirks are usually frowned upon.
Because the expectation is that those changes can be kept in local
downstream kernel trees and products that actually ship to customers
do not exhibit those quirks.

> 2. Furthermore, similar implementations exist in various commercial devices
> (with vastly different functions), such as the multifunction devices
> (including VGA, audio functions, etc.) mentioned in the following link:
> 
> https://forums.developer.nvidia.com/t/nvidia-rtx-5090-not-detected-by-nvidia-smi-on-ubuntu-server-24-04/327409?page=2

I've looked at that forum discussion and it seems there's an issue with
Resizeable BAR support on certain Nvidia cards, but I'm not seeing
anything on that page relating to functions taking longer than
others to come out of reset.  So this doesn't seem to be related to
the present patch.

Thanks,

Lukas