[Arch] optirun does not work, no apparent message error

[SALVOuankeBALDO]

Hello everybody,

I've just installed bumblebee on Arch Linux to make use of my nvidia graphics card, that is:

lspci | grep 3D
01:00.0 3D controller: NVIDIA Corporation GF117M [GeForce 610M/710M/810M/820M / GT 620M/625M/630M/720M] (rev ff)

I've installed bumblebee by following the guide on the Arch Wiki, https://wiki.archlinux.org/index.php/Bumblebee. I downloaded the nvidia driver and downloaded the necessary tools for 32-bit applications (as my machine is 64-bit).

After installation, installation of bbswitch and acpid and reboot I tried to test glxgears using optirun. Very verbose output:

optirun -vv glxgears --info
[ 2241.303014] [DEBUG]Reading file: /etc/bumblebee/bumblebee.conf
[ 2241.303506] [INFO]Configured driver: nvidia
[ 2241.303755] [DEBUG]optirun version 3.2.1 starting...
[ 2241.303776] [DEBUG]Active configuration:
[ 2241.303783] [DEBUG] bumblebeed config file: /etc/bumblebee/bumblebee.conf
[ 2241.303790] [DEBUG] X display: :8
[ 2241.303795] [DEBUG] LD_LIBRARY_PATH: /usr/lib/nvidia:/usr/lib32/nvidia
[ 2241.303802] [DEBUG] Socket path: /var/run/bumblebee.socket
[ 2241.303808] [DEBUG] Accel/display bridge: auto
[ 2241.303814] [DEBUG] VGL Compression: proxy
[ 2241.303824] [DEBUG] VGLrun extra options: 
[ 2241.303840] [DEBUG] Primus LD Path: /usr/lib/primus:/usr/lib32/primus
[ 2241.303907] [DEBUG]Using auto-detected bridge virtualgl
[ 2242.654193] [INFO]Response: Yes. X is active.

[ 2242.654226] [INFO]Running application using virtualgl.
[ 2242.654423] [DEBUG]Process vglrun started, PID 1842.
[ 2242.718384] [DEBUG]SIGCHILD received, but wait failed with No child processes
[ 2242.718419] [DEBUG]Socket closed.
[ 2242.718436] [DEBUG]Killing all remaining processes.

and optirun exits.
dmesg output:

[ 2241.489411] bbswitch: enabling discrete graphics
[ 2242.091758] nvidia-nvlink: Nvlink Core is being initialized, major device number 243
[ 2242.091779] NVRM: loading NVIDIA UNIX x86_64 Kernel Module  367.35  Mon Jul 11 23:14:21 PDT 2016
[ 2242.110868] vgaarb: this pci device is not a vga device
[ 2242.112750] ACPI Warning: \_SB.PCI0.PEG0.PEGP._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20160108/nsarguments-95)
[ 2242.112795] ACPI Warning: \_SB.PCI0.PEG0.PEGP._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20160108/nsarguments-95)
[ 2242.112842] ACPI Warning: \_SB.PCI0.PEG0.PEGP._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20160108/nsarguments-95)
[ 2242.112863] ACPI Warning: \_SB.PCI0.PEG0.PEGP._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20160108/nsarguments-95)
[ 2242.112883] ACPI Warning: \_SB.PCI0.PEG0.PEGP._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20160108/nsarguments-95)
[ 2242.112928] ACPI Warning: \_SB.PCI0.PEG0.PEGP._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20160108/nsarguments-95)
[ 2242.112948] ACPI Warning: \_SB.PCI0.PEG0.PEGP._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20160108/nsarguments-95)
[ 2242.127054] ACPI Warning: \_SB.PCI0.PEG0.PEGP._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20160108/nsarguments-95)
[ 2242.316760] vgaarb: this pci device is not a vga device
[ 2242.405092] nvidia-modeset: Loading NVIDIA Kernel Mode Setting Driver for UNIX platforms  367.35  Mon Jul 11 22:43:59 PDT 2016
[ 2242.882110] glxgears[1842]: segfault at 10 ip 00007f6c8af3f7ad sp 00007ffda33184b0 error 4 in libGLX_nvidia.so.367.35[7f6c8ae82000+c8000]
[ 2242.965999] ACPI Warning: \_SB.PCI0.PEG0.PEGP._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20160108/nsarguments-95)
[ 2242.970250] nvidia-modeset: Unloading
[ 2242.982429] nvidia-nvlink: Unregistered the Nvlink Core, major device number 243
[ 2242.996664] bbswitch: disabling discrete graphics
[ 2242.996676] ACPI Warning: \_SB.PCI0.PEG0.PEGP._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20160108/nsarguments-95)
[ 2243.008738] pci 0000:01:00.0: Refused to change power state, currently in D0

Content of /var/log/Xorg.8.log: http://pastebin.com/uWMQN7UP (NVIDIA start appearing in the middle)

I've also tried using primusrun, and a strange thing happens: primusrun glxgears works (also confirmed by cat-ting bbswitch when glxgears is running, and bbswitch says ON while running, and goes to OFF when I end glxgears). Strange thing is that, when I close glxgears, the following error appears in console:

primus: warning: dropping a frame to avoid deadlock                                                                                                                                             
primus: warning: timeout waiting for display worker                                                                                                                                             
XIO:  fatal IO error 11 (Resource temporarily unavailable) on X server ":0"                                                                                                                     
      after 34 requests (34 known processed) with 0 events remaining. 

glxgears does not terminate, and I have to take back control manually, through CTRL+C.

I've tried googling around, but having no obvious error message is hampering my research. Anybody can help?

[s0600204]

I can report the same problem. Also running Arch 64-bit, but I don't have the 32-bit libraries. Problem started about a week ago, after updating my system.

I notice that optirun glxspheres64 doesn't work, but optirun -b primus glxspheres64 does.

As the default bridge is auto which defaults to virtualgl, I'm assuming that something's recently changed/broken/regressed there.

Looking over pacmans logs, I notice that the packages mesa-libgl, nvidia, and nvidia-utils were recently updated (as were others, but these seemed most relevant)

Current version of virtualgl is 2.5-2, and has a build date of March 18th.

Current version of mesa-libgl is 12.0.1-2, and has a build date of July 24th. Was until earlier today running version 12.0.1-1 (build date: July 9th) and optirun didn't work by default then either. Before that, version 11.2.2-1 (build date: 10th May) and optirun did work.

@SALVOuankeBALDO, if I may make a suggestion, try using the -b primus argument of optirun (as I have above). If it works, you should be able to set optirun to default to this by opening /etc/bumblebee/bumblebee.conf with elevated privileges and changing the line Bridge=auto to Bridge=primus. Hopefully that works until a better solution comes along.

[SALVOuankeBALDO]

As I've written, I'm aware of the fact that primus is working. It gives me a strange error when I close glxgears and freezes, forcing me to CTRL+C out of the error. I'm still debugging that.

Still, thanks for the suggestion.

[s0600204]

I would then suggest to check if any other programs require you ctrl-c to end the program when run through primusrun. From my own experimentation, I note that glxgears and glxheads do, but glxspheres64, teapot, 0ad/pyrogenesis and supertuxkart do not.

I don't know if you've tried it, but I notice that prefixing primusrun (or optirun) with vblank_mode=X, where X is either 0 or 1, means you don't have to ctrl-c out of glxgears. For some reason. Of course, this does mean that you lose v-sync, so it's not the best solution.

Returning back to optirun not working with virtualgl as the bridge, after some mucking around, it suddenly occurred to me that I should check the exit code. One short echo $? later, and it turns out that it fails each time with code 139 (0x0213). A little looking online reveals this means an "Out of Bounds", or segmentation fault.

Turning to gdb (ie. optirun gdb glxgears) and voila:

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff54607ad in ?? () from /usr/lib/libGLX_nvidia.so.0

Looking at the reported bugs on arch's package-tracking site (which I probably should have checked sooner) turns up this: https://bugs.archlinux.org/task/50067

Anyhow, I've downgraded the nvidia (and nvidia-utils) package locally and optirun is working with virtualgl again.

Doesn't seem to have fixed your problem, though. Good luck, and please keep us appraised.

[SALVOuankeBALDO]

Thanks for the suggestion. I tried as you did an echo $?, obtaining 130 as output (just after running glxgears with virtualgl). A gdb after (I did not know the tool as I'm a newbie, but it's always nice to have antoher tool!) I had exactly the same error as you:

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff54607ad in ?? () from /usr/lib/libGLX_nvidia.so.0

So I guess it's a problem of compatibility between bumblebee and the new nvidia driver. I'll try downgrading, as running with primus crashes a Steam game that I'm interested in, so I think the workaround is not too much effective.

[SALVOuankeBALDO]

After bumbebee update, optirun is working. Many thanks to the developers.

[Avinash-Bhat]

@SALVOuankeBALDO in which version of bumblebee is this fixed?

[Lekensteyn]

@Avinash-Bhat Try the git version, bracnh "develop". It should really be pushed out, but ...

[SALVOuankeBALDO]

@Avinash-Bhat bumblebee 3.2.1-12, I downloaded it from Arch's repositories through pacman. I'm in no "developer" or "test" repository, just the standard ones.

[Avinash-Bhat]

@SALVOuankeBALDO it didn't work for me OOTB, but with some help from the comment by @s0600204 on setting Bridge=primus it worked for me.

[SALVOuankeBALDO]

Oh, shit.

Sorry, I tested optirun glxgears convinced that I had Bridge=virtualgl. Turns out I had modified it to primus, that was working. I switched back to virtualgl and returned in the same situation.

So I guess I have to reopen this. Sorry all, I don't know where I have my head in these days.

[MrBollie]

While the primus workaround works with glxgears, I have more complex applications, that even crash with that. Same segfaults, though the processes get stuck and need some kill -9.

Still, that workaround doesn't look healthy at all:
optirun -b primus glxgears leads to this:

Process 5166 (glxgears) of user 1000 dumped core.

Stack trace of thread 5166:
#0 0x00007fa51c5dc7ad n/a (libGLX_nvidia.so.0)
#1 0x00007fa51c5a50b8 n/a (libGLX_nvidia.so.0)
#2 0x00007fa51c5982b9 glXGetConfig (libGLX_nvidia.so.0)
#3 0x00007fa51eae7666 n/a (libvglfaker.so)
#4 0x00007fa51eae8abd n/a (libvglfaker.so)
#5 0x00007fa51eabf59a glXChooseVisual (libvglfaker.so)
#6 0x00000000004036bf n/a (glxgears)
#7 0x0000000000401a07 n/a (glxgears)
#8 0x00007fa51de6a291 __libc_start_main (libc.so.6)
#9 0x0000000000402349 n/a (glxgears)

I also wonder if this issue is limited to ARCH only.

[MrBollie]

Please have a look at the latest comments on https://bugs.archlinux.org/task/50067
At least on my system, the latest nvidia beta driver seems to fix the issue. However, having bumblebeed started with bbswitch on lets X hang when it starts (with automatic graphic cards detection). Not sure, if that's a bumblebee issue.