[SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

yaworski · 2018-08-05 19:14:37

michelesr wrote:

I've re-tested this after upgrading firmware and now bumblebee is able to successfully turn the card back on by loading nvidia modules and properly initialise the card.
I'm running bios version 1.3.0 and full up-to-date Arch system, but bbswitch still fails with the usual error (card refused to change power state).

I've also tried this and can confirm that the bbswitch still doesn't work.

michelesr wrote:

If I set the power control to auto for the NVIDIA card, then the nvidia module is not being loaded by xorg/lightdm/cinnamon.
This doesn't work for me. If I start Xorg it will load the driver even if the card power control is set to "auto". I'm running GNOME.

If I set it to auto on boot (before the xorg starts) then I also get the nvidia driver auto-loaded and it's in use. But if I don't change it on boot (it's "on" by default), then somehow the driver is not loaded in my case at all. I can then set the power control to auto and I can load and unload the driver freely.

Last edited by yaworski (2018-08-05 19:15:40)

IngeniousDox · 2018-08-06 14:47:17

I have opened a Feature request at the Bumblee github for a new method to unload the nvidia modules after the use of Optirun: https://github.com/Bumblebee-Project/Bu … issues/978

About the card being enabled on boot, I haven't had that issue thusfar. I'm using TLP. I'll recheck if this is still the case when I update my laptop to newer kernel this weekend.

yaworski · 2018-08-07 19:29:00

IngeniousDox, can you share your TLP configuration? I have some issues with mine (mostly default). When booting only on battery it seems it doesn't apply power control properly to all the PCI devices (they're still on "on", not "auto"). I have the tlp.service enabled, but I need to restart it after boot for it to properly apply the power control configuration.
I've noticed that today. I've rebooted the machine without AC being connected and even with NVIDIA card powered down I've had only 2-3 h of estimated time on the fully charged battery (powertop showed around 23 W of power consumption). After restarting the tlp, the power consumption dropped to 9 W on idle and the estimated time on battery rose to 6-7 hours.

IngeniousDox · 2018-08-07 20:28:27

I haven't changed anything about my configuration of TLP. I installed it, and I activated it.

I'm going to upgrade to newer bios and play with my laptop again this weekend. I want to have lowest battery usage on boot right away for everything. I have noticed a bit high usage. Powertop is reporting the fingerprint reader to be using 10W, but that seems excessive for a that part. So I'm going to have look into powermanagement more. I'll have to fiddle with turning off stuff, and setting low power to Touchscreen (I have the 4k screen), low power to intel gpu, and SSD etc. I'm not sure if this is the thread for it though, seems best to keep this about bbswitch issues, since that is the nice and visible as a thread then. So for Dell XPS 15 9570 power management, we should make a new thread, and figure out what the best options will be to get the lowest power consumption. I might make this forum thread while I'm fiddling again. I'm also thinking about making a page on Archwiki about the 9570 (first personal, and promote it to a real 9570 page if it looks good, unless someone beats me to it.)

Dox

ilpanich · 2018-08-11 10:43:11

Same issue for me on my just bought Dell XPS 15 9570.
Nvidia card won't turn on, tried several changes (also with a more expert friend that works with arch on his XPS 9560... using his confguration wouldn't help... maybe it is due to the slight difference of nvidia card) in kernel parameters, modules to load...

Currently using bumblebee 3.2.1-18, bbswitch 0.8-156, kernel 4.17.13.arch1-1, nvidia 396.45-6. Didn't have time to play a lot with power managers (My laptop has only one week of life ) and with bumblebee options.
Tried also some of the solution proposed in this thread without any success.

When I run

optirun glxspheres64

I receive the following errors:

[ 1748.253925] [ERROR]Cannot access secondary GPU - error: Could not enable discrete graphics card

[ 1748.253956] [ERROR]Aborting because fallback start is disabled.

Other (maybe) useful command outputs:

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

dmesg | grep bbswitch
[    2.654030] bbswitch: loading out-of-tree module taints kernel.
[    2.654309] bbswitch: version 0.8
[    2.654315] bbswitch: Found integrated VGA device 0000:00:02.0: \_SB_.PCI0.GFX0
[    2.654322] bbswitch: Found discrete VGA device 0000:01:00.0: \_SB_.PCI0.PEG0.PEGP
[    2.654474] bbswitch: detected an Optimus _DSM function
[    2.654593] bbswitch: disabling discrete graphics
[    2.698458] bbswitch: Succesfully loaded. Discrete card 0000:01:00.0 is off
[    4.683811]  mei soundcore i2c_i801 intel_lpss intel_soc_dts_iosf intel_pch_thermal typec hid shpchp battery int3403_thermal int340x_thermal_zone tpm_tis tpm_tis_core tpm rtc_cmos wmi dell_smo8800 evdev rng_core intel_hid mac_hid int3400_thermal sparse_keymap pcc_cpufreq acpi_thermal_rel ac sg crypto_user acpi_call(O) ip_tables x_tables ext4 crc32c_generic crc16 mbcache jbd2 fscrypto rtsx_pci_sdmmc serio_raw mmc_core atkbd libps2 ahci xhci_pci libahci xhci_hcd libata crc32c_intel usbcore rtsx_pci scsi_mod usb_common i8042 serio bbswitch(O) i915 intel_gtt i2c_algo_bit drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops drm agpgart
[ 1748.143524] bbswitch: enabling discrete graphics

EDIT: I forgot to mention that I have already flashed last BIOS version.

if anyone happen to solve his problems or can give some help I woul appreciate a lot.

Last edited by ilpanich (2018-08-11 13:27:30)

Kutoru · 2018-08-13 17:22:40

Don't run powertop --auto-tune.
If you use TLP you can auto-suspend the usb manually but leave the Runtime PM for Nvidia on. Reboot and login. Re-enable the Runtime PM so that the card can "power off". I can't seem to enable Runtime PM beforehand whether through TLP or a udev rule as it only works after I login in to my graphical session. Well, unless I switch RUNTIME_PM_ON_BAT to on, which then only the Nvidia card ends up having its Runtime PM enabled whereas the ones disabled from when it was auto can't seem to be turned off.

LazyLucretia · 2018-08-14 08:08:57

Hey, it seems that a new BIOS update just landed. Has anyone had chance to try this with new BIOS?

Kutoru · 2018-08-14 13:31:57

If you're referring to 1.3.0 then yes though it has been out for a while.

ilpanich · 2018-08-14 14:45:37

Tried some other solution proposed (adding other kernel parameters), but none working.
bbswitch always report card off, can't bring it on using optirun/primusrun (always report the error I reported in my previous post).
Also, dmesg report the following:

pci 0000:01:00.0: Refused to change power state, currently in D3

I have also updated nvidia drivers, kernel, bumblebee and bbswitch every time it was possible.

yaworski · 2018-08-21 14:02:14

I've found a way to prevent Xorg from loading nvidia module on boot. I've created a /etc/X11/xorg.conf.d/01-noautogpu.conf file with

Section "ServerFlags"
	Option "AutoAddGPU" "off"
EndSection

Then the iGPU needs to be added manually. I've created a /etc/X11/xorg.conf.d/20-intel.conf file with:

Section "Device"
	Identifier "Intel Graphics"
	Driver "intel"
EndSection

I've tested this multiple times and now the nvidia module is not loaded by the Xorg at all.

I don't use bbswitch. Instead I've created /etc/tmpfiles.d/nvidia_pm.conf file with:

w /sys/bus/pci/devices/0000:01:00.0/power/control - - - - auto

This switches on the kernel power management for the NVidia card. It depends on whether the nvidia module is loaded or not. The bumblebee loads the module, turning on the card when the optirun/primusrun is executed, but unfortunately it doesn't even try to unload it if bbswitch is not installed. But after manually unloading the module, the card turns off. The card is turned off after boot by default.

Last edited by yaworski (2018-08-21 14:05:52)

yaworski · 2018-08-21 15:07:31

I've forked bumblebee and created a dummy pm module, which doesn't really do anything than returning a state expected by the bumblebee.

Here's my modified source code: https://github.com/yawor/Bumblebee/tree/linuxpm. The name of the PM module is linuxpm. Can be set as PMMethod in bumblebee.conf.

The way bumblebee is written, this was the quickest way to force it to unload the nvidia module automatically. Please don't judge . I know it's basically a dirty hack, but it works with Linux PM as described in my previous post.

To build it, I've cloned bumblebee-git from AUR and replaced original bumblebee git URL in sources with:

"$_gitname::git://github.com/yawor/Bumblebee.git#branch=linuxpm"

and then just run makepkg and install resulting package.

Last edited by yaworski (2018-08-21 15:10:04)

LazyLucretia · 2018-08-22 09:25:41

yaworski wrote:

I've forked bumblebee and created a dummy pm module, which doesn't really do anything than returning a state expected by the bumblebee.
Here's my modified source code: https://github.com/yawor/Bumblebee/tree/linuxpm. The name of the PM module is linuxpm. Can be set as PMMethod in bumblebee.conf.
The way bumblebee is written, this was the quickest way to force it to unload the nvidia module automatically. Please don't judge . I know it's basically a dirty hack, but it works with Linux PM as described in my previous post.
To build it, I've cloned bumblebee-git from AUR and replaced original bumblebee git URL in sources with:
"$_gitname::git://github.com/yawor/Bumblebee.git#branch=linuxpm"
and then just run makepkg and install resulting package.

I couldn't quite understand how this works. In your previous message, you said "but unfortunately it doesn't even try to unload it if bbswitch is not installed. But after manually unloading the module, the card turns off". Does your fork handle this unloading issue? If it does, how do you even unload the module while it's still in use?

yaworski · 2018-08-22 15:10:09

Sorry if the description is not very clear. I'll try to describe this better.

The way bumblebee works is that it has a list of supported PM methods (two in the original source code). Each method defines if bumblebee needs to unload gGPU module or not. For example bbswitch method requires the module to be unloaded, but the gpu switcheroo doesn't (I don't know what it does but this is just something I've seen in the code). If no PM method is enabled (for example bbswitch is not installed in the system), then bumblebee loads the module, but it doesn't unload it after the optirun/primusrun exits.

What I did is I've created a third PM method, called linuxpm, which doesn't really do anything, but it tells bumblebee that it requires the module to be unloaded. It also returns status codes which bumblebee expects. That way bumblebee loads module when you run optirun, but unloads it when the program exits and the card is no longer used.

After the program exits, the module is no longer being used, so it can be unloaded. My first post shows how to prevent the nvidia module being loaded by the main Xorg server when it starts (this prevented the module from being unloaded as it was used by the Xorg).

This bumblebee modification is a dirty hack. I'll try to analyse the code deeper and maybe there will be a better solution without using this dummy PM method.

yaworski · 2018-08-22 23:15:31

Ok, I've rewritten my modification to Bumblebee. Instead of providing a dummy PM method like before, I've added a AlwaysUnloadKernelDriver configuration option, which should be added to driver-nvidia section in bumblebee.conf. It takes true or false. If not present or false, then Bumblebee works as before. If true, then Bumblebee tries to unload the driver even if there's no active PM method (for example no bbswitch installed in the system). This version is in forceunload branch in my fork. I've also created a pull request against the develop branch of the official Bumblebee git repo:
https://github.com/Bumblebee-Project/Bumblebee/pull/983
I've also referenced this thread in the PR.

Last edited by yaworski (2018-08-22 23:16:19)

IngeniousDox · 2018-08-23 07:25:34

I linked your PR in my feature request. What I requested looked like your linuxpm method. But your other configuration option could work aswell. Lets see what the bumblebee devs say about it.

ameed · 2018-08-28 06:34:37

You all seem to have this under control, but I want to add my own symptoms. Most of this is probably old news to the rest of you, but I want to document it centrally in case someone else shows up.

I'm on a Dell XPS 15 9570 with an NVIDIA card. I've installed nvidia, bumblebee, and bbswitch (not tlp or powertop), and this is what I get from various commands:

$ primusrun glxgears -info
primus: fatal: Bumblebee daemon reported: error: Could not enable discrete graphics card

(I also ran the above command earlier on the same boot, which explains why some things appear twice below)

$ systemctl status bumblebeed
● bumblebeed.service - Bumblebee C Daemon
   Loaded: loaded (/usr/lib/systemd/system/bumblebeed.service; enabled; vendor preset: disabled)
   Active: active (running) since Tue 2018-08-28 01:02:03 CDT; 22min ago
 Main PID: 430 (bumblebeed)
    Tasks: 1 (limit: 4915)
   Memory: 2.5M
   CGroup: /system.slice/bumblebeed.service
           └─430 /usr/bin/bumblebeed

Aug 28 01:02:03 minsky systemd[1]: Started Bumblebee C Daemon.
Aug 28 01:02:03 minsky bumblebeed[430]: [    2.920572] [INFO]/usr/bin/bumblebeed 3.2.1 started
Aug 28 01:04:38 minsky bumblebeed[430]: [  157.177495] [ERROR]Could not enable discrete graphics card
Aug 28 01:24:28 minsky bumblebeed[430]: [ 1347.678241] [ERROR]Could not enable discrete graphics card

$ dmesg | grep -A2 bbswitch
[    2.919222] bbswitch: loading out-of-tree module taints kernel.
[    2.919509] bbswitch: version 0.8
[    2.919512] bbswitch: Found integrated VGA device 0000:00:02.0: \_SB_.PCI0.GFX0
[    2.919516] bbswitch: Found discrete VGA device 0000:01:00.0: \_SB_.PCI0.PEG0.PEGP
[    2.919522] ACPI Warning: \_SB.PCI0.PEG0.PEGP._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20180531/nsarguments-66)
[    2.919606] bbswitch: detected an Optimus _DSM function
[    2.919615] pci 0000:01:00.0: enabling device (0006 -> 0007)
[    2.919659] bbswitch: Succesfully loaded. Discrete card 0000:01:00.0 is on
[    2.920628] bbswitch: disabling discrete graphics
[    3.055529] input: Video Bus as /devices/LNXSYSTM:00/LNXSYBUS:00/PNP0A08:00/LNXVIDEO:00/input/input12
[    3.056150] ACPI: Video Device [PEGP] (multi-head: no  rom: yes  post: no)
--
[  157.075949] bbswitch: enabling discrete graphics
[  157.117076] pci 0000:01:00.0: Refused to change power state, currently in D3
[  157.177416] pci 0000:01:00.0: Refused to change power state, currently in D3
--
[ 1347.678049] bbswitch: enabling discrete graphics
[ 1347.678065] pci 0000:01:00.0: Refused to change power state, currently in D3

# modprobe nvidia
modprobe: ERROR: could not insert 'nvidia': No such device

I've tried disabling power management by booting with pcie_port_pm=off, which didn't work. I haven't attempted to use manual module insertion and Linux PCI power management with bbswitch uninstalled (I can do that if you would find it helpful). I'm following Bumblebee-Project/Bumblebee#983; hopefully it will solve this problem.

Thanks for your work on this!

yaworski · 2018-08-28 07:12:28

@ameed at the moment no one was able to run bbswitch on this model. It doesn't properly turn off the card (it still draws power - at least in my case) and fails when trying to power it back on (as it thinks the card is powered off). So I think it's not recommended to even install bbswitch on this model as not only it doesn't work, but it also break things more.

Last edited by yaworski (2018-08-28 07:13:23)

ameed · 2018-08-28 07:22:00

@yaworski That's what I gathered. I was planning on leaving it there inert until your PR is pulled (hopefully); has there been any harm to having it there if you don't attempt to use the NVIDIA GPU? I skimmed the thread, and it seemed that the problem was with turning the dGPU on, not off--and not being able to use the dGPU is fine with me for now.

yaworski · 2018-08-28 09:54:18

@ameed, you need to test this for example with powertop. In my case bbswitch said that it turned off the card, but it really didn't. Powertop showed in the Device stats that NVIDIA was still powered at 100% and the power consumption on the battery was still high (15-19 W). With the Linux PM method, the power consumption goes down to below 10 W when the dGPU power is cut off.

ameed · 2018-08-29 12:13:52

@yaworski, I can confirm the behavior you're witnessing with bbswitch. Looks like I'll be using Linux power management.

ameed · 2018-08-29 19:26:57

I've made the changes that @yaworski did above, but I'm still seeing the NVIDIA card powered up in powertop.

Specifically, I created /etc/X11/xorg.conf.d/01-noautogpu.conf exactly as above, and 20-modesetting.conf as follows (since I use the modesetting driver for my Intel card):

Section "Device"
    Identifier "iGPU"
    Driver "modesetting"
    BusID "PCI:0:2:0"
EndSection

I also added the tmpfiles.d entry (though I had to use the path /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0/power/control), and confirmed that that file contains the word auto.

Still, powertop reports the NVIDIA card consuming 100% power. Any pointers? Thanks!

OdinEidolon · 2018-09-01 21:17:11

I have a 9550 and I am trying to get nvidia/bumblebee/bbswitch to work on a 9570 of a friend. I thought the setup was similar, but it turns out it is not!
If no video drivers are installed X (gnome-shell) starts with the modesetting driver, as it should, but if I install `nvidia` X starts with the nvidia card (according to the output of `lsof /dev/nvidia*`, the nvidia card sits at 100% and cannot be turned off). If I use the xorg configuration files listed above by @ameed, the computer hangs when Xorg starts, with no error nor any possibility of getting to a terminal. Any hint?

The configuration is quite a bit simpler with my 9550. X (KDE Plasma) starts with the Intel card by default, with no configuration needed, even if `nvidia` is installed. Bumblebee and bbswitch work out of the box with no additional configuration required. The word `nvidia` never appears in the Xorg logs.

Last edited by OdinEidolon (2018-09-01 21:26:14)

michelesr · 2018-09-04 16:49:50

A dirty hack to disable the graphic card before Xorg starts is:
# echo 1 > /sys/bus/pci/devices/0000:01:00.0/remove
$ lspci | grep -i nvidia
This will hide the PCIe bus and prevent the nvidia module to be loaded.
To restore you can force a rescan:
# echo 1 > /sys/bus/pci/rescan
$ lspci | grep -i nvidia
1:00.0 3D controller: NVIDIA Corporation GP107M [GeForce GTX 1050 Ti Mobile] (rev a1)
For some reason the rescan is able to restore the Nvidia card bus only if AC is plugged.

This is because the PCIe device needs to have power control set to 'on', and not 'auto', otherwise it will get suspended and the rescan won't work:

sudo tee /sys/bus/pci/devices/0000:00:01.0/power/control <<<on

In my case, TLP was switching the control to 'auto' on battery. I got rid of it.

Also, I customized the nvidia-xrun script to deal with this: https://github.com/michelesr/nvidia-xru … vidia-xrun, and created a systemd service to remove it automatically on boot:

$ cat /etc/systemd/system/nvidia-disable.service                                                                  
[Unit]
Description="Disable Nvidia PCI bus"

[Service]
ExecStart=/bin/bash -c 'echo 1 > /sys/bus/pci/devices/0000:01:00.0/remove'

[Install]
WantedBy=multi-user.target

$ sudo systemctl enable nvidia-disable.service

Use this if you can't prevent the nvidia module to be loaded.

OdinEidolon · 2018-09-04 19:43:46

Just a heads up, we found that gnome-shell was loading the `nvidia` driver (by issuing `lsof /dev/nvidia*`) no matter what. I do not know why, but other DEs, such as KDE Plasma, seem to to be unaffected by this. We switched to Plasma for the time being. The `nvidia` module seems to be not loaded at boot now.

Last edited by OdinEidolon (2018-09-04 19:44:22)

OdinEidolon · 2018-09-10 13:53:50

Is there any bbswitch bugreport to track this on? Or is the 'primus not working / card not shutting down' issue completely separate from the 'bbswitch not working' issue?
If not, I can try to open one.

Arch Linux

#51 2018-08-05 19:14:37

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#52 2018-08-06 14:47:17

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#53 2018-08-07 19:29:00

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#54 2018-08-07 20:28:27

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#55 2018-08-11 10:43:11

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#56 2018-08-13 17:22:40

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#57 2018-08-14 08:08:57

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#58 2018-08-14 13:31:57

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#59 2018-08-14 14:45:37

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#60 2018-08-21 14:02:14

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#61 2018-08-21 15:07:31

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#62 2018-08-22 09:25:41

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#63 2018-08-22 15:10:09

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#64 2018-08-22 23:15:31

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#65 2018-08-23 07:25:34

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#66 2018-08-28 06:34:37

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#67 2018-08-28 07:12:28

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#68 2018-08-28 07:22:00

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#69 2018-08-28 09:54:18

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#70 2018-08-29 12:13:52

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#71 2018-08-29 19:26:57

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#72 2018-09-01 21:17:11

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#73 2018-09-04 16:49:50

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#74 2018-09-04 19:43:46

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

#75 2018-09-10 13:53:50

Re: [SOLVED] Dell XPS 9570 bbswitch not working, Nvdia won't power off/on

Board footer