Re: nouveau on G5 – trouble anyone?
Hello Peter!
Thanks for your answer.
I am currently installing Gentoo Linux in a chroot from the Debian
installation. X.org is actually running, and I switched to a VT which
surprisingly works stable in every way. To check it, I just stopped
X.org by running "systemctl stop kdm" in one of the consoles. So it was
running, but not interfering with the system.
dmesg shows the line:
nouveau E[ DRM] GPU lockup - switching to software fbcon
And then the "FUN" thing: after stopping kdm, I now got these messages
(after about 2 minutes):
nouveau E[Xorg[838]] failed to idle channel 0xcccc0001 [Xorg[838]]
These messages repeat every ~15 seconds and ONLY THEN the system was not
usable anymore. I could still switch consoles, and type in commands, but
they were heavily delayed.
I was then turning off kdm for the next boot, using "systemctl disable
kdm" (delayed, but working). So here comes FUN #2: this solved the delay
problem. System was usable again. I am now continuing to build Gentoo
(without rebooting). For now. Hope this will not cause trouble.
I think I will build a custom kernel later on. Also, I will play with
Gentoo – once I get it up and running, I shall play with current
nouveau, mesa and kernels. Maybe there are some fixes there already? Or
I can apply the 2013 patches mentioned somewhere…
Thanks again,
Andreas aka Linux User #330250
Am 2016-01-04 um 22:50 schrieb Peter Saisanas:
> Hi Andreas,
>
> The default ppc64 kernel as supplied by Debian has been configured
> with 64KB kernel pagesizes. At least valid since the last time i have
> checked.
> I have been building my own kernels for a while, haven't used a
> default Debian kernel for a while.
> I believe the change to the default Debian kernel pagesize config for
> ppc64 kernel builds was made around jessie!
>
> The nouveau driver currently doesn't work with anything other than 4KB
> kernel pagesize.
> This isn't much of a surprise since x86 and amd64 both use 4KB
> pagesize which is more or less the target market (also ppc32 used 4KB
> kernel pagesize).
>
> To correct this, unfortunately you need to modify the kernel config
> and build your own custom kernel.
> Otherwise you may use kernel deb packages that i have prepared
> earlier, otherwise you can build your own kernel.
>
> Also, for many of the the nVidia GPU's commonly found in PMAC 11.2,
> the nouveau driver currently defaults to MSI interrupts.
> I have tested the equivalent GPUs (NV47 with x86 VGA BIOS for instance
> in an amd64 box) and the MSI interrupts seem to work fine with the
> particular GPU running the nouveau driver.
> So it seems that the issue is that currently the nVidia GPU's running
> the nouveau driver on ppc64 arch, you must disable MSI interrupts.
> There are various methods of doing this, however i tend to pass module
> parameters instead of passing options on the kernel command line. Each
> to their own i guess.
>
> For both of these issues, IIRC it doesn't seem to affect the nouveau
> fb console, however once you attempt to get xorg up and running, you
> will soon get a gpu lockup... forcing you to reboot.
>
> Lastly, kernel versions between 3.18.16 & 4.3.0-rc6 had issues
> extracting the DCB block from the OpenFirmware fcode ROM used in the
> oem nVidia GPU's sourced from Apple. There was a workaround for this,
> however if you use a suitable kernel version, the workaround doesn't
> need to be applied.
>
>
> Best Regards,
> Peter
>
>
Reply to: