4.3 nouveau regression: nv04_fbcon_imageblit(?) deadlock

4.3 nouveau regression: nv04_fbcon_imageblit(?) deadlock

[Meelis Roos]

Hello, this is a report about a PC with 3 nvidia cards - PCI-E NV44, AGP 
NV34 and PCI NV05. Only NV44 has monitor attached currently.

4.2.0 (both packaged and self-compiled) worked fine, 4.3.0 and 4.4-rc2 
exhibit a problem: after the system has started up, I start getting 
messages like INFO: task kworker/0:2:652 blocked for more than 120 
seconds. where nv04_fbcon_imageblit and nouveau_pmops_runtime_suspend 
show up in trace. The trace keeps reappearing, also breaking lspci and 
clean reboot (they hang).

dmesg from 4.4-rc2, drm-debug=0xe
http://kodu.ut.ee/~mroos/download/dm-nouveau-4.4-rc2

dmesg from 4.2.0, drm.debug=0xe
http://kodu.ut.ee/~mroos/download/dm-nouveau-4.2

lspci -vvv
http://kodu.ut.ee/~mroos/download/lspci.nouveau

config:
http://kodu.ut.ee/~mroos/download/nouveau-config

Is there anything else I can provide?

-- 
Meelis Roos (mroos@linux.ee)

Re: 4.3 nouveau regression: nv04_fbcon_imageblit(?) deadlock

[Ilia Mirkin]

On Thu, Nov 26, 2015 at 12:10 PM, Meelis Roos <mroos@linux.ee> wrote:
> Hello, this is a report about a PC with 3 nvidia cards - PCI-E NV44, AGP
> NV34 and PCI NV05. Only NV44 has monitor attached currently.
>
> 4.2.0 (both packaged and self-compiled) worked fine, 4.3.0 and 4.4-rc2
> exhibit a problem: after the system has started up, I start getting
> messages like INFO: task kworker/0:2:652 blocked for more than 120
> seconds. where nv04_fbcon_imageblit and nouveau_pmops_runtime_suspend
> show up in trace. The trace keeps reappearing, also breaking lspci and
> clean reboot (they hang).
>
> dmesg from 4.4-rc2, drm-debug=0xe
> http://kodu.ut.ee/~mroos/download/dm-nouveau-4.4-rc2
>
> dmesg from 4.2.0, drm.debug=0xe
> http://kodu.ut.ee/~mroos/download/dm-nouveau-4.2
>
> lspci -vvv
> http://kodu.ut.ee/~mroos/download/lspci.nouveau
>
> config:
> http://kodu.ut.ee/~mroos/download/nouveau-config
>
> Is there anything else I can provide?

Runtime pm shouldn't be a thing for these GPUs... can you see if
booting with nouveau.runpm=0 fixes it? Sounds like we reintroduced
some problem for GPUs that don't have optimus-style acpi power off
hooks.

  -ilia

Re: 4.3 nouveau regression: nv04_fbcon_imageblit(?) deadlock

[Meelis Roos]

> > 4.2.0 (both packaged and self-compiled) worked fine, 4.3.0 and 4.4-rc2
> > exhibit a problem: after the system has started up, I start getting
> > messages like INFO: task kworker/0:2:652 blocked for more than 120
> > seconds. where nv04_fbcon_imageblit and nouveau_pmops_runtime_suspend
> > show up in trace. The trace keeps reappearing, also breaking lspci and
> > clean reboot (they hang).
> >
> > dmesg from 4.4-rc2, drm-debug=0xe
> > http://kodu.ut.ee/~mroos/download/dm-nouveau-4.4-rc2
> >
> > dmesg from 4.2.0, drm.debug=0xe
> > http://kodu.ut.ee/~mroos/download/dm-nouveau-4.2
> >
> > lspci -vvv
> > http://kodu.ut.ee/~mroos/download/lspci.nouveau
> >
> > config:
> > http://kodu.ut.ee/~mroos/download/nouveau-config
> >
> > Is there anything else I can provide?
> 
> Runtime pm shouldn't be a thing for these GPUs... can you see if
> booting with nouveau.runpm=0 fixes it? Sounds like we reintroduced
> some problem for GPUs that don't have optimus-style acpi power off
> hooks.

Yes, nouveau.runpm=0 makes the hang go away, and reboot and lspci also 
started working without hangs.

-- 
Meelis Roos (mroos@linux.ee)