Re: [Bug 216859] New: PCI bridge to bus boot hang at enumeration

Re: [Bug 216859] New: PCI bridge to bus boot hang at enumeration

[Bjorn Helgaas]

[+cc sound folks]

On Wed, Dec 28, 2022 at 06:02:48AM -0600, Bjorn Helgaas wrote:
> On Wed, Dec 28, 2022 at 08:37:52AM +0000, bugzilla-daemon@kernel.org wrote:
> > https://bugzilla.kernel.org/show_bug.cgi?id=216859
> 
> >            Summary: PCI bridge to bus boot hang at enumeration
> >     Kernel Version: 6.1-rc1
> > ...
> 
> > With Kernel 6.1-rc1 the enumeration process stopped working for me,
> > see attachments.
> > 
> > The enumeration works fine with Kernel 6.0 and below.
> > 
> > Same problem still exists with v6.1. and v6.2.-rc1
> 
> Thank you very much for your report, Zeno!
> 
> v6.0 works, v6.1-rc1 fails.  Would you mind booting v6.1-rc1 with the
> "ignore_loglevel initcall_debug" kernel parameters and taking a photo
> when it hangs?
> 
> How did you conclude that the hang is related to a PCI bridge?  I see
> recent PCI messages in the photo, but it looks like the last message
> is from NFS, so I'm wondering if I'm missing some context.  The v6.0
> dmesg shows several other ntfs, fuse, JFS, etc messages before more
> PCI-related things.  Anyway, the "initcall_debug" might help us narrow
> it down a bit.

Thanks very much for the bisection (complete log at [1])!

The bisection claims the first bad commit is:

  833477fce7a1 ("Merge tag 'sound-6.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound")

with parents:

  7e6739b9336e ("Merge tag 'drm-next-2022-10-05' of git://anongit.freedesktop.org/drm/drm")
  86a4d29e7554 ("Merge tag 'asoc-v6.1' of https://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound into for-linus")

Both 7e6739b9336e and 86a4d29e7554 tested "good" during the bisection.

There is a minor conflict when merging 86a4d29e7554 into the upstream,
but I can't imagine that being resolved incorrectly.

Would you mind turning off CONFIG_SOUND in your .config and testing
833477fce7a1 again?  I'm a little skeptical that the hang would be
sound-related, but I guess it's a place to start.

Bjorn

[1] https://bugzilla.kernel.org/show_bug.cgi?id=216859#c35

Re: [Bug 216859] New: PCI bridge to bus boot hang at enumeration

[Zeno Davatz]

Dear Bjorn

On Thu, Jan 12, 2023 at 9:08 PM Bjorn Helgaas <helgaas@kernel.org> wrote:
>
> [+cc sound folks]
>
> On Wed, Dec 28, 2022 at 06:02:48AM -0600, Bjorn Helgaas wrote:
> > On Wed, Dec 28, 2022 at 08:37:52AM +0000, bugzilla-daemon@kernel.org wrote:
> > > https://bugzilla.kernel.org/show_bug.cgi?id=216859
> >
> > >            Summary: PCI bridge to bus boot hang at enumeration
> > >     Kernel Version: 6.1-rc1
> > > ...
> >
> > > With Kernel 6.1-rc1 the enumeration process stopped working for me,
> > > see attachments.
> > >
> > > The enumeration works fine with Kernel 6.0 and below.
> > >
> > > Same problem still exists with v6.1. and v6.2.-rc1
> >
> > Thank you very much for your report, Zeno!
> >
> > v6.0 works, v6.1-rc1 fails.  Would you mind booting v6.1-rc1 with the
> > "ignore_loglevel initcall_debug" kernel parameters and taking a photo
> > when it hangs?
> >
> > How did you conclude that the hang is related to a PCI bridge?  I see
> > recent PCI messages in the photo, but it looks like the last message
> > is from NFS, so I'm wondering if I'm missing some context.  The v6.0
> > dmesg shows several other ntfs, fuse, JFS, etc messages before more
> > PCI-related things.  Anyway, the "initcall_debug" might help us narrow
> > it down a bit.
>
> Thanks very much for the bisection (complete log at [1])!
>
> The bisection claims the first bad commit is:
>
>   833477fce7a1 ("Merge tag 'sound-6.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound")
>
> with parents:
>
>   7e6739b9336e ("Merge tag 'drm-next-2022-10-05' of git://anongit.freedesktop.org/drm/drm")
>   86a4d29e7554 ("Merge tag 'asoc-v6.1' of https://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound into for-linus")
>
> Both 7e6739b9336e and 86a4d29e7554 tested "good" during the bisection.
>
> There is a minor conflict when merging 86a4d29e7554 into the upstream,
> but I can't imagine that being resolved incorrectly.
>
> Would you mind turning off CONFIG_SOUND in your .config and testing
> 833477fce7a1 again?  I'm a little skeptical that the hang would be
> sound-related, but I guess it's a place to start.
>
> Bjorn
>
> [1] https://bugzilla.kernel.org/show_bug.cgi?id=216859#c35

Booting without sound into the commit 833477fce7a1 did not help. Same hang.

Best
Zeno

Re: [REGRESSION] [Bug 216859] New: PCI bridge to bus boot hang at enumeration

[Bjorn Helgaas]

[+cc folks from 145eed48de27 and framebuffer folks, regression list]

On Thu, Jan 12, 2023 at 02:08:19PM -0600, Bjorn Helgaas wrote:
> On Wed, Dec 28, 2022 at 06:02:48AM -0600, Bjorn Helgaas wrote:
> > On Wed, Dec 28, 2022 at 08:37:52AM +0000, bugzilla-daemon@kernel.org wrote:
> > > https://bugzilla.kernel.org/show_bug.cgi?id=216859
> > 
> > >            Summary: PCI bridge to bus boot hang at enumeration
> > >     Kernel Version: 6.1-rc1
> > > ...
> > 
> > > With Kernel 6.1-rc1 the enumeration process stopped working for me,
> > > see attachments.
> > > 
> > > The enumeration works fine with Kernel 6.0 and below.
> > > 
> > > Same problem still exists with v6.1. and v6.2.-rc1

This is a regression between v6.0 and v6.1-rc1.  Console output during
boot freezes after nvidiafb deactivates the VGA console.

It was a lot of work for Zeno, but we finally isolated this console
hang to 145eed48de27 ("fbdev: Remove conflicting devices on PCI bus").

The system actually does continue to boot and is accessible via ssh, 
but the console appears hung, at least for output.  More details in
the bugzilla starting at
https://bugzilla.kernel.org/show_bug.cgi?id=216859#c47 .

Bjorn

Re: [REGRESSION] [Bug 216859] New: PCI bridge to bus boot hang at enumeration

[Bjorn Helgaas]

[+cc Geert]

On Thu, Jan 26, 2023 at 06:11:24AM -0600, Bjorn Helgaas wrote:
> On Thu, Jan 12, 2023 at 02:08:19PM -0600, Bjorn Helgaas wrote:
> > On Wed, Dec 28, 2022 at 06:02:48AM -0600, Bjorn Helgaas wrote:
> > > On Wed, Dec 28, 2022 at 08:37:52AM +0000, bugzilla-daemon@kernel.org wrote:
> > > > https://bugzilla.kernel.org/show_bug.cgi?id=216859
> > > 
> > > >            Summary: PCI bridge to bus boot hang at enumeration
> > > >     Kernel Version: 6.1-rc1
> > > > ...
> > > 
> > > > With Kernel 6.1-rc1 the enumeration process stopped working for me,
> > > > see attachments.
> > > > 
> > > > The enumeration works fine with Kernel 6.0 and below.
> > > > 
> > > > Same problem still exists with v6.1. and v6.2.-rc1
> 
> This is a regression between v6.0 and v6.1-rc1.  Console output during
> boot freezes after nvidiafb deactivates the VGA console.
> 
> It was a lot of work for Zeno, but we finally isolated this console
> hang to 145eed48de27 ("fbdev: Remove conflicting devices on PCI bus").
> 
> The system actually does continue to boot and is accessible via ssh, 
> but the console appears hung, at least for output.  More details in
> the bugzilla starting at
> https://bugzilla.kernel.org/show_bug.cgi?id=216859#c47 .

145eed48de27 ("fbdev: Remove conflicting devices on PCI bus") doesn't
say what the benefit is, or what would break if we reverted it.

Does anybody have any clues?  It would be nice to resolve this
regression before v6.2, which will probably be released 2/12 or 2/19.

Bjorn