search

Bumblebee-Project / bbswitch

Disable discrete graphics (currently nvidia only)
GNU General Public License v2.0
488 stars 78 forks source link

Dedicated GPU stays on during suspend on Acer E5-574G #132

Open jacobmischka opened 8 years ago

jacobmischka commented 8 years ago

As requested in #115 I am creating a new issue.

When suspending, bbswitch powers on the dedicated GPU as it is intended to, but the card continues to draw power during suspend, draining the battery very rapidly and producing heat. The card is a GTX 940M.

As a workaround, I'm using systemd hooks to rmmod and modprobe bbswitch on suspend and resume. This causes suspend to work properly and not drain battery, but means I have to reboot in order for the GPU to work again. I am currently using the stable version of bumblebee, I am setting no acpi_osi parameter, and am running Linux kernel 4.6.

You mentioned bios v1.14, I can update if you think it will make a difference, but I probably won't otherwise.

If there is any other information I can provide I will be happy to do so.

acpidump: https://gist.github.com/jacobmischka/e7f3d1da3039a0f207635d235c10f5a4

Lekensteyn commented 8 years ago

Updating the BIOS will likely not make a difference. What will likely work is the bbswitch_dev module from the acpi-pr3 branch combined with the four patches from https://git.kernel.org/cgit/linux/kernel/git/helgaas/pci.git/commit/?h=pci/pm&id=8b71f5652eeac561acf883da01ab4810f763ee42 applied on top of your current kernel (I tested it on top of v4.6).

bbswitch_dev (with dev standing for "PCI device", but "in development" is also a possible explanation) will eventually be merged into bbswitch itself. For now you have to activate it manually:

make modname=bbswitch_dev
sudo insmod bbswitch_dev
# replace 13d9 by your device ID from lspci -nnd10de:
sudo tee /sys/bus/pci/drivers/bbswitch_dev/new_id <<<"10de 13d9"

Once loaded it should show:

bbswitch_dev: disabling discrete graphics
bbswitch_dev 0000:01:00.0: power state changed by ACPI to D3cold
pcieport 0000:00:01.0: PME# enabled
pcieport 0000:00:01.0: power state changed by ACPI to D3cold

If the last two messages are missing, then either the four pci/pm patches were not applied to the kernel or (not in your case) the BIOS is older than 2015.

jacobmischka commented 8 years ago

Thanks, I've been meaning to try this but I haven't found the time lately to patch and compile a kernel.

I just tried that kernel.org link and it seems broken, the commit hash no longer seems to exist. Sorry again for the delay, thank you for your help.

Lekensteyn commented 8 years ago

New branch is based on v4.7-rc1, see https://git.kernel.org/cgit/linux/kernel/git/helgaas/pci.git/log/?h=pci/pm

jacobmischka commented 8 years ago

Oh perfect that makes things much easier, I'll install it now and report back soon

Edit: Oh sorry I misread, I thought it was merged into 4.7rc1. I won't be able to do it right now in that case but I will soon. Thanks again.

Lekensteyn commented 8 years ago

Have you found time to test nouveau? I'd like to prepare the second revision of the patchset for nouveau, see https://github.com/Bumblebee-Project/bbswitch/issues/78#issuecomment-223072012 for something based on v4.6.

If possible, can you try nouveau on v4.6 (with no patches) and nouveau (with the linked patchset)?

jacobmischka commented 8 years ago

Sure, I'll also try the patch mentioned before with nvidia today as well finally.

Nouveau without the patches behaved as expected, high power usage. Compiling the patches now, will report back.

jacobmischka commented 8 years ago

I'm sorry, I was wrong. Nouveau without the patches doesn't work as expected. I've never been able to get xrandr to recognize my nvidia gpu to use with PRIME, and I'm still unable to.

I assumed it was working correctly because the power usage went up, but that's just because I hadn't reenabled power management for the PCI device yet since disabling bbswitch. I have no xorg.d configs messing with my graphics config to my knowledge, xrandr just can't see it. So I can't confirm that the card actually works, but nouveau is loaded and power usage is low.

I'm not sure if that means it's working or not. If you have any further ideas to try to get xrandr to recognize my card I'll be happy to try them.

I'll try bbswitch_dev now. I'm gonna use the same kernel I just compiled to test nouveau, hopefully the 4 nouveau-specific commits don't affect it.

jacobmischka commented 8 years ago

Sorry, I seem to be having problems enabling bbswitch_dev. It seems to be installed correctly, but /sys/bus/pci/drivers/bbswitch_dev/ doesn't exist, and bumblebee is complaining that it can't find bbswitch. I tried changing PMMethod=bbswitch to bbswitch_dev, but that didn't seem to work.

Any ideas?

jacobmischka commented 8 years ago

Okay, I'm not sure what I was doing wrong before but I'm able to correctly load the module now. However, I do not see the two pcieport messages.

journalctl -r:

Jul 16 01:33:20 mischka-laptop kernel: bbswitch_dev 0000:01:00.0: Refused to change power state, currently in D0
Jul 16 01:33:20 mischka-laptop kernel: bbswitch_dev: disabling discrete graphics

I'm pretty certain I have the patch properly applied. Is there a way to check that you know of?

Lekensteyn commented 8 years ago

Can you post the output of lspci -nn -s01:? If you happen to have an audio device there that is not suspended, the pcie root port will also not suspend. It is also important to note that this module will only save power when the appropriate pci/pm patches are applied (possibly in Linux 4.8 now).

Also, do you need to use Bumblebee for acceleration or would you just like to save power?

jacobmischka commented 8 years ago

Sorry for the delay. My output of lspci -nn s01: is:

01:00.0 3D controller [0302]: NVIDIA Corporation GM108M [GeForce 940M] [10de:1347] (rev a2)

I almost never use my external gpu, saving power is my primary concern, though I'd like to be able to use the gpu occasionally.

Unfortunately, it looks like 4.8 rc2 isn't automatically disabling the card either. I would prefer to use Nouveau if it'll correctly suspend the card, is there something special I'm supposed to do to get it to work?

Edit: I missed the bumblebee.conf that was blacklisting nouveau, power saving is working correctly with 4.8 rc2 for me. I still can't get xrandr to recognize my gpu so I can use PRIME, but that's a different problem.

Thank you for all of your help, sorry that my responses were so delayed.

jacobmischka commented 8 years ago

Oh. Do you not need to set up the provider offload sink in xrandr for PRIME with DRI3?

After reading either badly written docs—or docs that suggest that, I just tried using PRIME and it seems to work. I get terrible performance with nouveau, but I suppose that's more or less to be expected. Looks like everything is working now, thanks again.

Lekensteyn commented 8 years ago

DRI3 does not require the xrandr --setprovideroffloadsink stuff. Thanks for testing, no problem for the delay, I got busy too ;)

BernardoGO commented 8 years ago

Seems like 4.8rc8 successfully suspend the GPU using the stock bbswitch. Sometimes the GPU does not work after resuming but it does work mostly.

Hopefully it may be solved in future releases.