[REGRESSION 6.14] Some PCI device BARs inacessible

[REGRESSION 6.14] Some PCI device BARs inacessible

[Michał Pecio]

Hi,

This is a heads up that an apparent PCI regression has been reported
and mistakenly assigned to USB in the kernel bugzilla:

https://bugzilla.kernel.org/show_bug.cgi?id=220016

Visible symptom is missing USB devices, but the whole controller fails
to probe, apparently due to devm_request_mem_region() returning NULL,
see drivers/usb/core/hcd-pci.c and usb_hcd_pci_probe().

Same systems also show a similar failure with some AHCI controller.
It seems specific to particular ASUS AMD motherboards.

Somebody found that disabling CONFIG_PCI_REALLOC_ENABLE_AUTO helps.

That's all I know, reporters can be reached via bugzilla.

Regards,
Michal

Re: [REGRESSION 6.14] Some PCI device BARs inacessible

[Bjorn Helgaas]

[+cc Joe (reporter), Ilpo]
[+bcc Steven (another reporter, bcc since email hasn't been used publicly recently]

On Fri, Apr 18, 2025 at 10:29:26AM +0200, Michał Pecio wrote:
> This is a heads up that an apparent PCI regression has been reported
> and mistakenly assigned to USB in the kernel bugzilla:
> 
> https://bugzilla.kernel.org/show_bug.cgi?id=220016
> 
> Visible symptom is missing USB devices, but the whole controller fails
> to probe, apparently due to devm_request_mem_region() returning NULL,
> see drivers/usb/core/hcd-pci.c and usb_hcd_pci_probe().
> 
> Same systems also show a similar failure with some AHCI controller.
> It seems specific to particular ASUS AMD motherboards.
> 
> Somebody found that disabling CONFIG_PCI_REALLOC_ENABLE_AUTO helps.
> 
> That's all I know, reporters can be reached via bugzilla.

#regzbot link: https://bugzilla.kernel.org/show_bug.cgi?id=220016

Re: [REGRESSION 6.14] Some PCI device BARs inacessible

[Bjorn Helgaas]

[+cc Ilpo, +bcc bugzilla reporters and commenters]

On Fri, Apr 18, 2025 at 10:29:26AM +0200, Michał Pecio wrote:
> Hi,
> 
> This is a heads up that an apparent PCI regression has been reported
> and mistakenly assigned to USB in the kernel bugzilla:
> 
> https://bugzilla.kernel.org/show_bug.cgi?id=220016
> 
> Visible symptom is missing USB devices, but the whole controller fails
> to probe, apparently due to devm_request_mem_region() returning NULL,
> see drivers/usb/core/hcd-pci.c and usb_hcd_pci_probe().
> 
> Same systems also show a similar failure with some AHCI controller.
> It seems specific to particular ASUS AMD motherboards.
> 
> Somebody found that disabling CONFIG_PCI_REALLOC_ENABLE_AUTO helps.
> 
> That's all I know, reporters can be reached via bugzilla.

I see some activity in the bugzilla but no resolution.  Ilpo has done
a lot of work in this area since v6.14, though, so I wouldn't be
surprised at all if this has been fixed. 

Would anybody be interested in testing v6.19 or v7.0-rc to see if this
problem still exists?

Bjorn

Re: [REGRESSION 6.14] Some PCI device BARs inacessible

[Ilpo Järvinen]

[-- Attachment #1: Type: text/plain, Size: 1533 bytes --]

On Mon, 16 Mar 2026, Bjorn Helgaas wrote:

> [+cc Ilpo, +bcc bugzilla reporters and commenters]
> 
> On Fri, Apr 18, 2025 at 10:29:26AM +0200, Michał Pecio wrote:
> > Hi,
> > 
> > This is a heads up that an apparent PCI regression has been reported
> > and mistakenly assigned to USB in the kernel bugzilla:
> > 
> > https://bugzilla.kernel.org/show_bug.cgi?id=220016
> > 
> > Visible symptom is missing USB devices, but the whole controller fails
> > to probe, apparently due to devm_request_mem_region() returning NULL,
> > see drivers/usb/core/hcd-pci.c and usb_hcd_pci_probe().
> > 
> > Same systems also show a similar failure with some AHCI controller.
> > It seems specific to particular ASUS AMD motherboards.
> > 
> > Somebody found that disabling CONFIG_PCI_REALLOC_ENABLE_AUTO helps.
> > 
> > That's all I know, reporters can be reached via bugzilla.
> 
> I see some activity in the bugzilla but no resolution.  Ilpo has done
> a lot of work in this area since v6.14, though, so I wouldn't be
> surprised at all if this has been fixed. 
> 
> Would anybody be interested in testing v6.19 or v7.0-rc to see if this
> problem still exists?

My plan was to look at this ticket again anyway (or has been for like a 
few months now but there has been higher prio items get in the way of 
looking the ticket up).

(It went to lower prio bin once it was determined it was not a real
regression but fc enabling pci=realloc. Despite that, I'd still want to 
have it fixed eventually.)

-- 
 i.