Arch Linux

dcbdbis

Member

From: Aurora, Colorado

Registered: 2004-09-10

Posts: 247

[TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

//--------------------------
It's not fixed as it is a kernel issue. Link to article in post #5.

But there is a workaround in the meantime to at least get you back up and running.

//-------------------------------

I have an up to date Arch x86_64 machine. And an NVidia GTX-560TI. Like others, today's updates broke "X" fatally, when both the new 3.2 kernel came down with the new NVidia driver family (290).

I've done some tests that may help folks smarter than me figure out what is going wrong.

Test:

a)  Removed nvidia, nvidia-utils, and all related NVidia packages. Handled the libgl dep issue.
b)  Killed the xorg.conf file, removed the nvidia module from memory, and installed the xf86-video-vesa package.

A subsequent call to startx had the DE come right up (just in a lower resolution than a native driver @ 1920x1200).

c)  Went to the NVidia site and downloaded the 285.05.09 drivers and installed the same from a true cli (I personally don't use GDM's). Allowed the NVidia drivers to create it's own xorg.conf. Dealt with the libgl issue. I know this is not Arch orthodoxy, but remember I'm trying to restore my main work system to operational status until the nvidia driver/packages/whatever get sorted out..

d)  Used the nvidia-xconfig to create a new xorg.conf, modprobe'd the new nvidia module, then tried startx. (It's how I choose to start my DE's)

Same issue as with the 290.10 driver. The error of "screen found but no valid configuration found..." that's well documented elsewhere in the forums.

While I am a veteran Arch user, and with other distro's have built hundreds of custom kernels (and a few for Arch), that does not make me a dev. In my humble experience, this test limited the variables down to one. Now the only difference was the kernel rev of 3.2.

At issue is:

a)  Is the nvidia module needing to do something different with the way it interacts with the new 3.2 kernel, meaning that NVIdia itself is doing something that the 3.2 kernel doesn't appreciate?

or

b)  Does the kernel need an adjusted .config file before being rebuilt?

I am not smart enough to know the answer.

Right now, I've of course removed all traces of nvidia, installed libgl, and am using the vesa driver so I could at least access my DE of choice and operate (somewhat). I cannot work until the NVidia things gets sorted out as I use video editing software that requires the use if the cuda toolset.

I post this test only as a source of information that hopefully may point the dev(s) in perhaps more directions than just nvidia. In the end, it may be totally NVidia's issue (like the last time)....But when I plug in the old working drivers in, and still cannot get "X" to function.....Then I start thinking about kernel configuration/compatibility issues.

If anyone wants me to do some other tests I haven't thought of...I'm totally game for it, anything to be of help.....Just let me know what you want done and I'll do it and post the results.

Sincerely and respectfully,

Dave

Last edited by dcbdbis (2012-01-19 09:06:33)

shivik

Member

Registered: 2011-03-24

Posts: 8

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

Exactly the same problem.
Black screen when updated to 3.2.
Tried that: http://www.nvnews.net/vbulletin/showthr … ?p=2519164 - no go.
The moment I go back to 3.1 everything is working.
Tried linux-ck, linux-lqx - black screens,same behaviour
Xorg.0.log starts with:

[     3.964] 
X.Org X Server 1.11.3
Release Date: 2011-12-16
[     3.984] X Protocol Version 11, Revision 0
[     3.985] Build Operating System: Linux 3.1.5-1-ARCH x86_64 
[     3.985] Current Operating System: Linux archenemy 3.2.0-lqx #1 ZEN SMP PREEMPT Wed Jan 18 08:01:04 EET 2012 x86_64

and end with:

[    21.535] (EE) NVIDIA(0): WAIT: (E, 0, 0x827d)

Just noticed that X server's build is against 3.1.5 kernel if it matters at all...
Tried the nvidia-beta-ck with linux-ck - and again black screen.
I'm going to try amd_iommu=off and intel_iommu=off at the same time now and see what happens.
Just shooting random here.
Motherboard Gigabyte 890FXA-UD5, videocard Geforce GTS250

shivik

Member

Registered: 2011-03-24

Posts: 8

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

So, I can boot without a black screen now using a 3.2 kernel.
The option that does the trick for me is amd_iommu=off .
Maybe people with intel chipset should try intel_iommu=off .

dcbdbis

Member

From: Aurora, Colorado

Registered: 2004-09-10

Posts: 247

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

Bingo!

Adding the amd_iommu=off to my kernel boot command line in menu.lst........I was back up and running.  THANK YOU!

Although I'm not entireley comfortable turning off the memory controller........but I'm sure that will be fixed shortly..

Again, Thank You very much!

Dave.......

The 3.2 kernel is more suspicious as the fixes come in!

I'm not going to edit the post as solved, because we've not solved it, we've band-aided it....just to get our workstations back. I'll mark it solved when whomever figures out who the culprit is, and squares it away. Be that the kernel folks, or NVidia.

swanson

Member

From: Sweden

Registered: 2011-02-05

Posts: 759

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

More info at Phoronix;
http://www.phoronix.com/scan.php?page=n … px=MTAzNzk

Seems to do with virtualization, which can be disabled in BIOS in some cases. I'm still at kernel 3.1.x, but following this.

dcbdbis

Member

From: Aurora, Colorado

Registered: 2004-09-10

Posts: 247

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

Great Article.....

I was leaning towards the kernel, and Phoronix nailed it on the head. I expect to see the kernel devs scrambling to fix the code and get out a patch quick.

IOMMU can affect virtualization. I have 6 VM's. Not critical at the moment.

But for now, the most important thing is I can get back to work, with minimal downtime today.

Thanks again for the posts.

Dave

s4ms3milia

Member

Registered: 2012-01-16

Posts: 24

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

I have a Geforce 560 and can't boot with the new kernel aswell.
Unfortunately neither intel_ nor amd_iommu=off fixes it for me. (got an amd cpu)

When GDM trys to load the screen goes blank and after some time I get this error:
gdm-binary[893]: WARINGING: GdmDisplay: display lasted 2.052566 seconds repeatedly.

I tried to uninstall nvidia and nvidia-utils but with nouveau I can't even get into a console, the screen goes blank. Had to chroot with a livecd, now I'm back at nvidia.

Please help me out on this....

swanson

Member

From: Sweden

Registered: 2011-02-05

Posts: 759

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

Downgrading or LTS kernel can be easy options, as a temporary measure. I'm on the CK-kernel and never upgraded from 3.1.x since it doesn't use BFQ anyway.

s4ms3milia

Member

Registered: 2012-01-16

Posts: 24

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

Sticking with the lts kernel sounds like a reasonable solution until this is fixed in the 3.2 Version. I guess I'll try that when I come back from work.

wulax

Member

Registered: 2006-02-20

Posts: 30

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

I also had this problem with my AMD + Nvidia system. Disabling IOMMU fixed it. Thanks!

Regarding virtualization, as far as I know most people don't use IOMMU even if their system has a functioning one. It's only useful for doing things like PCI-passthrough. Last time I checked neither VirtualBox nor VMware even supports it. Xen does though if you configure it right.

dcbdbis

Member

From: Aurora, Colorado

Registered: 2004-09-10

Posts: 247

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

Hello wulax, and thank you for the post.

I checked with the recent iterations of Virtualization. IOMMU is supported and used by VB, VM, and Xen....But as you mentioned only for PCI-Passthrough. Mostly used for giving the VM direct access to some proprietary PCI card, or the Graphics Card to assist with true 3D graphics instead of emulated 3D.

I have clients (not) on Arch but on Linux Mint to lower my admin hours, and their businesses can't afford down time from kernel.org's glitch in the 3.2.1 kernel. And two of them require NetFlix......which means their Windows VM's have to be at peak performance (IOMMU) or the NetFlix thing is choppy and glitchy.

My VM's however run perfectly on 3.2.1, but I have no need of 3D in them. They are simply "mirrors" of my clients machine so I can test new stuff in their enviornments before I deploy. There is no way to do this with my two NetFlix accounts, I do those purley on site and not remotely.

It does look like IOMMU (this is purly speculation) may come more into vogue to increase graphical performance as neither VMWare's nor VB's 3D support is really, really solid in all respects.

Your post did cause me to do a little more research on IOMMU, and I learned some things I could do better for my clients.

THANK YOU!

Dave

swanson

Member

From: Sweden

Registered: 2011-02-05

Posts: 759

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

LTS kernel just updated to 3.0.1.x, so it's not that far behind anymore. Seems even more as a good alternative.

Jack450

Member

From: Los Angeles

Registered: 2010-07-11

Posts: 58

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

I'm having issues getting nvidia to work with arch I've what you guys posted but my laptop still hangs when I get on it. I've tried latest and beta nvidia drivers, I've using LTS kernel and still nothing. Can anyone help me figure this problem out?

---

Welcome to the Internet!

This isn't Daycare or the Justice System, here you're retarded until proven innocent.

dcbdbis

Member

From: Aurora, Colorado

Registered: 2004-09-10

Posts: 247

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

For Jack450,

I'll be happy to work with you through this....

I need: What flavor of Arch x32 or x64?

How current?

If you are using the lts kernel, then you need to use the nvidia-lts package with it or no love between the kernel and nvidia module.

a)  Ensure you do not have anything but the lts's api's, headers, and kernel installed. If you're missing a piece, I'd recommend installing it. If you've still got 3.2.1 stuff in your system, get rid of it.
b)  Kill your '/etc/X11/xorg.conf' file. (hint : sudo)
c)  Remove your existing nvidia kernel mod from memory

sudo modprobe -r nvidia

d)  pacman in the nvidia-lts stuff.
e)  Modprobe in the new module

 sudo modprobe nvidia (check my spelling, I don't know if the lts nvidia modules goes by 'nvidia' or 'nvidia-lts')

f)  From a cli do:

sudo nvidia-xconfig

g) startx, or reboot, whichever you prefer........
h)  And see if that doesn't give you back your GDM/DE.

And no this is not the only way to do things.....There may be easier ways, but being an old retired engineer...I do tend to be a bit retentive.....Others may have different suggestions just as worthy and perhaps easier. All I know is that, if your system was sitting on my desk, this is what I would do for starters.

Dave

PS: if this doesn't do it, as this is a laptop....then I need more info on your hardware. Please do an

lspci -v > HardwareInfo.txt

and PM me with that file, to conserve forum space and I'll then come up with a round two attempt if round one for some wierd reason doesn't solve your problem.

Last edited by dcbdbis (2012-01-20 04:56:56)

zyxnull

Member

Registered: 2009-12-09

Posts: 13

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

I have fixed the problem installing nvidia-lts and adding "amd_iommu=off" to the kernel, the only quirk I get is that GDM looks like fallback mode, but after that everything works fine

Jack450

Member

From: Los Angeles

Registered: 2010-07-11

Posts: 58

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

For Jack450,

I'll be happy to work with you through this....

I need: What flavor of Arch x32 or x64?

How current?

If you are using the lts kernel, then you need to use the nvidia-lts package with it or no love between the kernel and nvidia module.

a)  Ensure you do not have anything but the lts's api's, headers, and kernel installed. If you're missing a piece, I'd recommend installing it. If you've still got 3.2.1 stuff in your system, get rid of it.
b)  Kill your '/etc/X11/xorg.conf' file. (hint : sudo)
c)  Remove your existing nvidia kernel mod from memory

sudo modprobe -r nvidia

d)  pacman in the nvidia-lts stuff.
e)  Modprobe in the new module

 sudo modprobe nvidia (check my spelling, I don't know if the lts nvidia modules goes by 'nvidia' or 'nvidia-lts')

f)  From a cli do:

sudo nvidia-xconfig

g) startx, or reboot, whichever you prefer........
h)  And see if that doesn't give you back your GDM/DE.

And no this is not the only way to do things.....There may be easier ways, but being an old retired engineer...I do tend to be a bit retentive.....Others may have different suggestions just as worthy and perhaps easier. All I know is that, if your system was sitting on my desk, this is what I would do for starters.

Dave

PS: if this doesn't do it, as this is a laptop....then I need more info on your hardware. Please do an

lspci -v > HardwareInfo.txt

and PM me with that file, to conserve forum space and I'll then come up with a round two attempt if round one for some wierd reason doesn't solve your problem.

It was first x64 and I switched to x32 to see if it made any difference but it still kills the Xorg so I gave up. I tried LTS, different nvidia drivers plus nvidia-lts drives, intel_iommu=off didn't work. So I'm guessing I'm just out of luck and have to wait for an update.

---

Welcome to the Internet!

This isn't Daycare or the Justice System, here you're retarded until proven innocent.

dinsdale

Member

From: Riviere-du-loup

Registered: 2010-06-04

Posts: 15

Re: [TEMP-WORKAROUND] 290 NVIdia/3.2 kernel tests I've done

Any update on this issue? I have a Intel I7 (thinkpad T520) and ran into this bug. I disabled the VT options completelyh in the BIOS when this thread was active but now ran into a problem where a virtualbox instance of windows xp winxp would crash during the installation of updates. I fixed it by enabling the first VT option in the BIOS again - since virtualbox seems ok but enabling both still leads to a xorg freeze.

Last edited by dinsdale (2012-03-30 10:06:47)