Bug 101764

(In reply to Carlo Caione from comment #0)
> Created attachment 132629 [details]
> GP107_regression_oops_log
> 
> Hi,
> we have verified that commit eaa2c988033b ("drm/nouveau/gr/gp107: initial
> support") is actually breaking the GTX1050TI on an Acer laptop.
> It's not possible to reach Xorg anymore and the journal is filled with Oops
> as shown in the attachment.
> 
> Thanks,

in that commit initial acceleration support for GP107 GPUs were added which means there are more places to mess up. As a workaround you can boot with nouveau.modeset=0 or nouveau.noaccel=1.

The real issue is something else, but this should give you at least a working setup again.

> The real issue is something else, but this should give you at least 
> a working setup again.

Thank you, but since I have the hardware in my availability for the next week or so I really would like if possible to help to debug better this issue and maybe come up with a proper fix :)

Let me know if there is anything else I can provide that can help to debug this issue.

Thanks.

Jul 11 16:14:59 endless kernel: nouveau 0000:01:00.0: bus: MMIO read of 00000000 FAULT at 409800 [ TIMEOUT ]
Jul 11 16:15:01 endless kernel: ------------[ cut here ]------------
Jul 11 16:15:01 endless kernel: WARNING: CPU: 1 PID: 11134 at /usr/src/packages/BUILD/linux-4.11.0/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c:1499 gf100_gr_init_ctxctl_ext+0x3cc/0x7f0 [nouveau]
Jul 11 16:15:01 endless kernel: nouveau 0000:01:00.0: timeout

So it's having trouble loading the firmware. Please confirm that you have the latest firmware for gp107 (as there was a window of time when linux-firmware contained bad firmware for gp107 -- some time ago now already, but you could have a bad snapshot).

I believe I've seen this before though, also on a laptop. I see that it's using the fancy new thing instead of ACPI DSM for powering it on:

Jul 11 16:14:58 endless kernel: VGA switcheroo: detected Optimus DSM method \_SB_.PCI0.PEG0.PEGP handle
Jul 11 16:14:58 endless kernel: nouveau: detected PR support, will not use DSM

Can you try to boot in a way that forces the ACPI stuff to be used? I think pcie_pm_port=off is the thing. And nouveau.runpm=0 for good measure.

Er, pcie_port_pm=off

> So it's having trouble loading the firmware. Please confirm that you have the 
> latest firmware for gp107 (as there was a window of time when linux-firmware 
> contained bad firmware for gp107 -- some time ago now already, but you could 
> have a bad snapshot).
The firmware is the latest one shipped in the linux-firmware git repo

>>> nouveau.noaccel=1 (see attachment)
In this case the nouveau module is correct probed and I do not get any WARNING in the journal (even though a bit noisy on the PM part).

>>> only pcie_port_pm=off
The system hangs before reaching Xorg

>>> nouveau.runpm=0 (with or without pcie_port_pm=off) (see attachment)
Interestingly I can still reach Xorg but in the journal I have a lot of WARNINGs

Created attachment 132635 [details]
GP107_noaccel

Created attachment 132636 [details]
GP107_runpm

Looks to be a duplicate of https://bugs.freedesktop.org/show_bug.cgi?id=100228

Was going to mark as such, but wanted to hear the thoughts of imirkin or others more experienced with nouveau first.

In the log I have

nouveau 0000:01:00.0: bus: MMIO read of 00000000 FAULT at 10f200 [ IBUS ]

It is timing out on this line however, the same as bz#100228 which is also a GP107:

nouveau 0000:01:00.0: bus: MMIO read of 00000000 FAULT at 409800 [ TIMEOUT ]

*** This bug has been marked as a duplicate of bug 100228 ***