On Tue, 2015-10-13 at 09:17 +0100, Luca Boccassi wrote: > On Oct 12, 2015 12:09, "Andreas Beckmann" <[email protected]> wrote: > > > > Thanks for finding some of the missing pieces :-) > > > > I assume the module nowadays gets loaded by udev ... is there any > way to > > get some more verbose log out of udev to see what hardware it found > and > > which modules (and with which options) it tries to load for that ... > > I am not sure. I'll try having "debug" in the kernel cmdline and see > if I get extra info. > > > Test with the driver package from jessie, not backports. > > Ok, I'll downgrade again before doing further tests.
[CC'ed 801598 so that relevant info appears there too] Downgraded and booted with "debug", and systemd got A LOT more verbose :-) But it was very useful, as digging through the log I figured out what is loading the module. As you suspected, it's udev. More precisely, it's this systemd unit at boot: /lib/systemd/system/systemd-udev-trigger.service And even more precisely, this udevadm call by that unit: udevadm trigger --type=devices --action=add If I remove that command from the service and reboot, the nvidia module is still loaded and the devices are still created but with 660 root:video permissions/ownership (so Gnome oopses as expected). So, how are we going to deal with this? Kind regards, Luca Boccassi
