Dell XPS 15 9560: System crashes when turning GPU on/off repeatedly

[skoehler]

On my brand new Dell XPS 9560, the bbswitch kernel module loads and repeatedly turning the GPU on and off lets the system hang. I don't know why and I can't see any kernel output, unfortunately. I'm using kernel 4.9.7.

My Distro is Gentoo. When booting with systemd, my system would actually hang before I could reach the login prompt. (Probably because the bumblebee daemon tries to disable the GPU). When booting with OpenRC (very serial booting process), I could reach the graphical login primpt, but turning GPU on and off repeatedly would result in a system freeze and crash. I just an echo ON/OFF > /proc/acpi/bbswitch.

It might be that the new 10 series GPU come with new ACPI tables for turning on/off the GPU. On a Dell XPS 15 9550 (960M GPU) everything was working fine.

[ArchangeGabriel]

Those issues are quite known/expected I think. Did you take a look at #140?

[skoehler]

Something else is at work here. I added the suggest workaround pcie_port_pm=off to my grub config and it didn't help. The system still freezes during bootup, probably when bumblebeed is loaded.

[skoehler]

Update: starting X will lock up the machine if the Nvidia device is OFF. So bbswitch itself seems to work, but X won't start. Neither nouveau or nvidia kernel module is loaded. Does X try to talk to the PCI device, somehow?

[ArchangeGabriel]

Sorry, answered in a hurry. I have two laptop at the moment:

– Dell XPS 9530, which works OK with Bumblebee, not sure with bbswitch alone, should test that.
– HP zBook Studio G3, on which I use bbswitch alone (need to try Bumblebee), but don’t load it at boot since it then freezes my system. So that looks like your case (but it’s not a Pascal GPU).

Both are affected by #140 btw.

I didn’t have the time until now to look into those issues, but I’ll do so next week. I did some tests with nouveau too, but don’t remember the results, and a new kernel requires new tests. ;)

I’ll keep you updated, and we’ll see with @Lekensteyn what can be done then.

[skoehler]

Something loads the nvidia kernel module when I start sddm. I blame X for that, yet the Xorg logs mention nothing. That seems to be the reason why my laptop hangs when starting X with the dGPU turned off.

From bug #140 I gather that support for PCIe PM is in the works and that will eventually replace the DSM method?

[Lekensteyn]

The hang issue is possibly not related to pcie_port_pm, but https://bugzilla.kernel.org/show_bug.cgi?id=156341 (Bumblebee-Project/Bumblebee#764). Try the acpi_osi workaround listed there.

[ArchangeGabriel]

@Lekensteyn OK that’s definitively it for me (my machine is already listed by someone else). Looks like I’ve been missing a lot of fun lately (my comment review queue for Bumblebee is at 1098 right now…). I’ve added myself to the CC list.

[Lekensteyn]

For me Github still shows 143 unread notifications for Bumblebee (none for bbswitch), but there are lots of open issues that still need a response (luckily users are helping each other). Hmm

[bluca]

I think I've given answers to all Ubuntu/Debian related issues, if I missed some please do tag me

[stefansedich]

I am also seeing a hang on startup too on my Precision 5200 running Ubuntu with 4.9.8, if I do no install the module and manually load it everything is fine, but if I install it and reboot my machine hangs before getting to the login prompt with some ACPI related errors.

acpi_osi=! acpi_osi=Windows 2009 got me booting but things like backlight did not work, for now I am just running a make load whenever I reboot as that is working without installing the module and hopefully there is a fix I have not tried yet :)!

[zeraien]

I also have the startup issue on Dell XPS 9560, and it seems to be mainly related to X, as it does not seem to affect Wayland.

[stefansedich]

I can confirm this @zeraien I just installed 3.22 and moved over to it today using wayland, I can boot fine with bbswitch installed although my system does lock up a short time after booting.

[zeraien]

I also have issues with freezing while in Wayland, but it seems to be related to the launching of older X-based applications (ex: steam, wine), which probably start the X machinery and this probably triggers the nvidia driver/bbswitch and thus you get a crash.

One way I avoided a crash with X, is by forcing bbswitch to turn the dGPU ON, before logging into X. So it seems related to the card being OFF and then X starts and crashes.

I'm on Fedora 25 (kernel 4.9.8), bumblebee 3.2.1, Nvidia 1050, Dell XPS 9560

[stefansedich]

If I just manually load bbswitch sudo make load on each boot everything is working fine for me without any crashes, Only seems to be an issue for me when installed, I might just run a script to load it this way after boot.

[pronobis]

@stefansedich can you explain what exactly you're doing with make load? I'm on XPS 15 9560 and having the freeze with bbswitch off. Indeed acpi_osi=! acpi_osi=Windows 2009 helps, but that disables touchpad as discussed above.