bbswitch doesn't seem to work on kernel in recent kernel

[Elv13]

Hi,

I don't think I am experiencing https://github.com/Bumblebee-Project/bbswitch/issues/140 directly. However since Linux 4.8, bbswitch (with or without pcie_port_pm=off) doesn't seem to do "it's job". The power usage on 4.7 is ~9watts and ~20watts in 4.8+. I can enable and disable the card, but it doesn't affect the power usage at all. The fans still spin even with low CPU temp.

cat /proc/acpi/bbswitch                                                 
0000:01:00.0 OFF

I also tried echo 1 >/sys/bus/pci/devices/0000:01:00.0/remove but that also doesn't restore the power usage back to normal. I really, really don't care or want to NVIDIA GPU to be active. I don't use it and I need the longer battery life. There is nothing suspicious in dmesg. It seems to "everything is fine", but in reality nothing is done. Is there any reliable way to debug this (I am ok patching bbswitch or the kernel).

Device: 01:00.0 VGA compatible controller: NVIDIA Corporation GK107GLM [Quadro K1100M] (rev ff)

Kernel:

linux   /boot/vmlinuz-4.16.8 root=/dev/sda1 ro acpi_osi="!Windows 2013" acpi_osi=Linux acpi_backlight=vendor i915.allow_pc8=1 i915.enable_psr=1 i915.powersave=1 i915.enable_rc6=1 i915.enable_fbc=1 i915.lvds_downclock=1 i915.semaphores=1 modprobe.blacklist=nouveau pcie_port_pm=off

[ArchangeGabriel]

Hi,

I’ve also had suboptimal powersaving using recent kernels. I’m now using nouveau, with pcie_port_pm=off because else powersavings are not as good as expected. I suspect this is because of the forced acpi_osi… Have you tried using nouveau instead of bbswitch?

Also, note that i915.enable_rc6, i915.lvds_downclock and i915.semaphores as well as i915.powersave have been removed, so specifying them is useless.

[Elv13]

Have you tried using nouveau instead of bbswitch?

I don't want nouveau, I want the NVIDIA GPU to be turned off and use the Intel one exclusively. Also, I never had luck with nouveau. On my Mac turns the screen #00ff00 and then overheat and on this laptop it's a Quadro, so it's untested, laggy and crashy. I ran the NVIDIA propriatary driver for a while, but it eats the battery in no time. This GPU is good for CAD, but it's very power hungry.

What I am looking for is getting of the last pre-4.8 LTS kermel before it's discontinued. Also, 4.17 is supposed to have magical new scheduling capable of saving ~10% energy. This was my main motivation for trying again. I can't believe there is no way to turn off whatever causes this regression. Still, can't find any solution that works and we are not that many with this GPU so the pool of knowledge seems too small.

[ArchangeGabriel]

nouveau turns off the GPU. I use it mostly only for that (and the fact my HDMI/Thunderbolt 3 outputs are wired to it).

[Elv13]

Hi again,

Initially it didn't work, but with the latest git-master kernel the nouveau team fixed PM for my card, so at least it doesn't take 35w anymore.

From what I sent on nouveau IRC:

I tried with Kernel 4.4 and git-master with my script and with laptop-mode-tools. Totally idle with backlight off, 4.4 takes 6.4w and git-master+nouveau 12.0, almost twice as much. The test setup was running the kernel in single user mode, starting X using X :0 & sleep 5 && DISPLAY=:0 xterm -c 'powertop'. laptop-mode and my ACPI script have the same result because they do the same thing. I checked everything and the git-master kernel has all the PM features it can have, so the NVIDIA card is really the only possible cause for the doubled power usage

So to recap (fully idle single user mode tests):

type	power consumption
Linux 4.4 + bbswitch	6.4w
Linux 4.8-4.17 + broken bbswitch	~18w
Linux 4.4-17(?) + Nouveau	35.5w
Linux 4.18-rc3 + Nouveau + forced PM	12w

kernel args:

linux   /boot/vmlinuz-4.18.0-rc3 root=/dev/sda1 ro acpi_osi="!Windows 2013" acpi_osi=Linux acpi_backlight=vendor i915.allow_pc8=1 i915.enable_psr=1 i915.powersave=1 i915.enable_fbc=1 i915.lvds_downclock=1 nouveau.config=NvPmEnableGating=1 pcie_aspm=force pcie_port_pm=on pcie_aspm.policy=powersupersave

[adhefe]

Here, it seems that dkms doesn't goes fine because of some new kernel 4.16 features:

# dkms install -m bbswitch -v 0.8-6

DKMS make.log for bbswitch-0.8-6 for kernel 4.16.13-desktop-1omv (x86_64) qui jun 7 21:08:29 -03 2018 make -C /lib/modules/4.16.13-desktop-1omv/build M="$(pwd)" modules make[1]: Entering directory '/usr/src/linux-4.16.13-desktop-1omv' CC [M] /var/lib/dkms/bbswitch/0.8-6/build/bbswitch.o /var/lib/dkms/bbswitch/0.8-6/build/bbswitch.c: In function ‘acpi_call_dsm’: /var/lib/dkms/bbswitch/0.8-6/build/bbswitch.c:138:28: note: byref variable will be forcibly initialized struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL }; ^~~ /var/lib/dkms/bbswitch/0.8-6/build/bbswitch.c:109:29: note: byref variable will be forcibly initialized struct acpi_object_list input; ^~~~~ /var/lib/dkms/bbswitch/0.8-6/build/bbswitch.c:108:24: note: byref variable will be forcibly initialized struct acpi_buffer output = { ACPI_ALLOCATE_BUFFER, NULL }; ^~ /var/lib/dkms/bbswitch/0.8-6/build/bbswitch.c: In function ‘bbswitch_init’: /var/lib/dkms/bbswitch/0.8-6/build/bbswitch.c:398:28: note: byref variable will be forcibly initialized struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL }; ^~~ Building modules, stage 2. MODPOST 1 modules CC /var/lib/dkms/bbswitch/0.8-6/build/bbswitch.mod.o LD [M] /var/lib/dkms/bbswitch/0.8-6/build/bbswitch.ko ld: warning: Disabling --icf due to -r make[1]: Leaving directory '/usr/src/linux-4.16.13-desktop-1omv'