Nouveau Changes Queue Ahead Of Linux 4.19

[phoronix]

Phoronix: Nouveau Changes Queue Ahead Of Linux 4.19

Linux 4.19 is going to be another exciting kernel on the Direct Rendering Manager (DRM) front with a lot of good stuff included while hours ago we finally got a look at what's in store for the open-source NVIDIA "Nouveau" driver...

http://www.phoronix.com/scan.php?page=news_item&px=Nouveau-DRM-Linux-4.19

[Nepenthes]

Unfortunately, nothing regarding crashes on using `lspci` or graphical login on recent Optimus laptops (with GTX 10xx GPUs)

https://bugs.freedesktop.org/show_bug.cgi?id=101665

That's the root cause of the following bug, which basically makes Ubuntu 18.04 unusable with said laptops, as ubuntu devs made the choice to drop unsupported bbswitch module and use nouveau runtime power management instead:

https://bugs.launchpad.net/ubuntu/+source/nvidia-prime/+bug/1765363

It leaves you with a fast draining battery and overheating laptop if you stay with the default setting, disabled runtime PM, or an unusable laptop if you try to enable runtime PM.

Hopefully, someday this gets fixed, but this is a huge hardware support regression for Ubuntu, and even if they're not responsible for Ubuntu forcing the use of their driver, nouveau devs may be the only ones in a position to help with the issue.

[Lekensteyn]

Unfortunately, nothing regarding crashes on using `lspci` or graphical login on recent Optimus laptops (with GTX 10xx GPUs)

https://bugs.freedesktop.org/show_bug.cgi?id=101665

That's the root cause of the following bug, which basically makes Ubuntu 18.04 unusable with said laptops, as ubuntu devs made the choice to drop unsupported bbswitch module and use nouveau runtime power management instead

The runtime PM method is the prefered approach for newer hardware since that is also what Windows uses (i.e. better tested by the vendors). The old method used by bbswitch in fact caused memory corruption with certain Lenovo laptops.

Unfortunately there is still something unknown that causes issues when trying to restore power. When using bbswitch, usually you do not load a kernel module such as nouveau or nvidia. When using nouveau, it will try to power down the GPU by default. However when SDDM/Xorg starts, it will try to probe the /dev/dri/XXX device created by the nouveau driver which subsequently will try to runtime resume the device. Due to that 'unknown issue', it will result in a lock up.

Workarounds exist for that lockup, but it is a hack and might not work for all devices: https://bugzilla.kernel.org/show_bug.cgi?id=156341
If you have some time, feel free to look into it, I was unable to make further progress here.

[karolherbst]

The runtime PM method is the prefered approach for newer hardware since that is also what Windows uses (i.e. better tested by the vendors). The old method used by bbswitch in fact caused memory corruption with certain Lenovo laptops.

Unfortunately there is still something unknown that causes issues when trying to restore power. When using bbswitch, usually you do not load a kernel module such as nouveau or nvidia. When using nouveau, it will try to power down the GPU by default. However when SDDM/Xorg starts, it will try to probe the /dev/dri/XXX device created by the nouveau driver which subsequently will try to runtime resume the device. Due to that 'unknown issue', it will result in a lock up.

Workarounds exist for that lockup, but it is a hack and might not work for all devices: https://bugzilla.kernel.org/show_bug.cgi?id=156341
If you have some time, feel free to look into it, I was unable to make further progress here.

the cause is known actually, the pci subsystem fails to read from the pci device leading to the entire system going down (as nothing really checks for errors on the resume patch)