Re: [PATCH] PCI/bwctrl: Fix NULL pointer deref on bus number exhaustion

["Ilpo Järvinen" <ilpo.jarvinen@linux.intel.com>]

[-- Attachment #1: Type: text/plain, Size: 1794 bytes --]

On Sun, 23 Mar 2025, Krzysztof Wilczyński wrote:

> Hello,
> 
> > The patch is working, no kernel crashes and shutting down after hotplugging the Thunderbolt 4 dock does not hang anymore, thanks!
> 
> Thank you for testing!  Appreciated!
> 
> > I still see messages in the kernel log that 'devices behind bridge are 
> > unusable' because 'no bus can be assigned to them', 'Hotplug bridge 
> > without secondary bus, ignoring', etc. These all refer to the 
> > Thunderbolt 4 bridge. Adding "pci=hpbussize=0x33" to the kernel 
> > doesn't make a difference. Adding 
> > "pci=realloc,asssign-busses,hpbussize=0x33" actually does 'fix' the 

Hi,

Hmm, it seems this reply got stuck into postponed messages so sending it 
only now.

There's a typo (3 x s) in the value, perhaps just for the email as it 
worked.

> > bus allocation failures (or at least I don't see any messages in the 
> > log anymore), but then amdgpu fails to initialize the IGP. 

That is very likely independent issue. I can try to look at it if it's 
some resource assignment related problem that occurs with pci=realloc. 
Please could you take a dmesg with this added into the kernel command 
line:

dyndbg="file drivers/pci/*.c +p"

I might also benefit from contents /proc/iomem if it turns out to be a 
resource assignment problem but we'll see.

-- 
 i.

> We can fix any outstanding issues or drop this patch.  I leave it to hot
> plug experts like Ilpo, Bjorn and Lukas to make the call here.

>
> > Anyway, all devices connected to the Thunderbolt 4 dock appear to work (USB, screens, ethernet) even despite these bus allocation failures, so I will just ignore them. 
> > 
> > Thanks again!
> 
> Would you be happy to provide your "Tested-by:" tag?
> 
> 	Krzysztof
>