Red Hat Bugzilla – Bug 831136
kernel 3.4 and possible bug with passing PCI devices in D3
Last modified: 2012-08-03 17:39:39 EDT
Description of problem:
With latest kernel 3.4.0-1.fc17.x86_64 I'm unabel to use optirun any long
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1. start system with kernel 3.4.0-1.fc17.x86_64
2. configure dkms, bbswitch, bumblebee and virtualgl
3. run command "sudo optirun glxgears"
[ERROR]Cannot access secondary GPU - error: [XORG] (EE) [drm] failed to open device
[ERROR]Aborting because fallback start is disabled.
X window with glxgears runnign and good performances due to using nouveau kernel module
Booting with kernel 3.3.7-1 it is ok.
I'm going to attach Xorg.8.log for both kernels.
Searching through forums it seems there could be a general bug in kernel 3.4 related to passing PCI devices in D3 state and that this is the one affecting me.
Created attachment 591152 [details]
Xorg.8.log when successfully run optirun command with kernel 3.3.7
Created attachment 591153 [details]
Xorg.8.log when unsuccessfully run optirun command with kernel 3.4.0
Created attachment 591154 [details]
My system is an Asus U36SD laptop with nVidia GeForce GT 520M discrete gpu.
$ lsmod|egrep “nouve|i915″
i915 472032 3
nouveau 799637 0
mxm_wmi 12823 1 nouveau
wmi 18697 3 mxm_wmi,nouveau,asus_wmi
video 18935 2 i915,nouveau
i2c_algo_bit 13257 2 i915,nouveau
drm_kms_helper 40411 2 i915,nouveau
ttm 79760 1 nouveau
drm 244591 6 ttm,i915,drm_kms_helper,nouveau
i2c_core 38028 7 drm,i915,i2c_i801,drm_kms_helper,i2c_algo_bit,nouveau,videodev
Same situation here.
01:00.0 VGA compatible controller: nVidia Corporation GF106 [GeForce GT 555M SDDR3] (rev ff)
Fedora 17 32bit.
Kernel 3.3.7-1 works OK.
Apart from inability to use nvidia adapter for particular tasks (windows vm with remote-viewer, wine, ecc) this decreases my battery life by about 90 minutes as I'm unable to offline it ....
Problem still persists with kernel 3.4.2-4.fc17.x86_64 got from updates channel
Possible bug related:
Or perhaps this one?
But it should be fixed in final 3.4, so I don't know...
Trying to debug with bumblebee guys it seems I could have a problem with kernel 3.4 referring to this kernel bug:
It continues also with 3.4.3-1.fc17.x86_64
It is the same with kernel-3.4.4-5.fc17.x86_64
Anyone looking at this? I'm available for further debug if needed..
I'm yet forced to boot in 3.3.7 kernel...
(In reply to comment #11)
> It is the same with kernel-3.4.4-5.fc17.x86_64
> Anyone looking at this? I'm available for further debug if needed..
> I'm yet forced to boot in 3.3.7 kernel...
You might want to pursue this issue with the upstream kernel community. At the moment, nobody is looking into this.
Hoping that perhaps someone being at the same time both an upstream kernel and fedora kernel maintainer could have time to go deeper inside.
It seems actually that Ben Skeggs at
points to another upstream kernel bug I already posted:
But that bug should be corrected in kernel-3.4.4-5.fc17.x86_64 as for its changelog
* Tue Jul 03 2012 Josh Boyer <email@example.com>
- Allow ACPI backlight to still work if _DOS isn't present (rhbz 834910)
Actually in my case it doesn't seem so..
Finally 3.5.0-2.fc17.x86_64 is the first kernel after kernel-3.3.7-1.fc17.x86_64 where I can use my Nvidia and run command "sudo optirun glxgears" without errors.
Discrete card correctly disabled after optirun exits.
Thanks (possibly due to upstream changes in 3.5...)
Put closed upstream as motivation... donna if it is correct.
Feel free to change ...