[BUG REPORT] 260.19.36 64-bit Linux driver conflicts with AMD 890FX IOMMU

[jdb2]

I'm running Kubuntu 10.10 Maverick with kernel version "2.6.35-26-generic #46-Ubuntu SMP Sun Jan 30 06:59:07 UTC 2011 x86_64 GNU/Linux". I'm using an ASUS Crosshair IV Extreme motherboard with 2 EVGA GTX 580s and when I enable the IOMMU in the BIOS I get this error in Xorg.0.log :

[ 61.697] (EE) NVIDIA(0): Failed to initialize the NVIDIA graphics device PCI:1:0:0. 
[ 61.697] (EE) NVIDIA(0): Please check your system's kernel log for additional error
[ 61.697] (EE) NVIDIA(0): messages and refer to Chapter 8: Common Problems in the
[ 61.697] (EE) NVIDIA(0): README for additional information.
[ 61.697] (EE) NVIDIA(0): Failed to initialize one NVIDIA graphics device!
[ 61.697] (WW) NVIDIA(0): Failed to initialize SLI configuration. Reason: One GPU
[ 61.697] (WW) NVIDIA(0): failed to initialize; Only one GPU will be used for this X
[ 61.697] (WW) NVIDIA(0): screen.
[ 73.868] (EE) NVIDIA(0): Failed to initialize the NVIDIA graphics device PCI:1:0:0. 
[ 73.868] (EE) NVIDIA(0): Please check your system's kernel log for additional error
[ 73.868] (EE) NVIDIA(0): messages and refer to Chapter 8: Common Problems in the
[ 73.868] (EE) NVIDIA(0): README for additional information.
[ 73.868] (EE) NVIDIA(0): Failed to initialize the NVIDIA graphics device!
[ 73.868] (II) UnloadModule: "nvidia"
[ 73.868] (II) UnloadModule: "wfb"
[ 73.868] (II) UnloadModule: "fb"
[ 73.868] (EE) Screen(s) found, but none have a usable configuration.
[ 73.868] 
Fatal server error:
[ 73.868] no screens found
[ 73.868] 
Please consult the The X.Org Foundation support 
 at http://wiki.x.org
 for help. 
[ 73.868] Please also check the log file at "/var/log/Xorg.0.log" for additional information.
[ 73.868] 
[ 73.889] ddxSigGiveUp: Closing log

Checking dmesg reveals the problem :

[ 14.886270] AMD-Vi: Event logged [IO_PAGE_FAULT device=01:00.0 domain=0x0018 address=0x0000000214cc6200 flags=0x0010]
[ 22.895142] NVRM: RmInitAdapter failed! (0x27:0x38:1046)
[ 22.895148] NVRM: rm_init_adapter(0) failed

Disabling the IOMMU solves the problem completely.

Hopefully Nvidia will address this soon.

jdb2

[AireTamStrm]

Is this on a virtualized environment?
If so, what Hypervisor are you running?

If not, IOMMU has close to no meaning (it isn't used even by LLVM, only VT-x/AMD-V). Why do you need the option enabled?

[jdb2]

Originally Posted by AireTamStrm

Is this on a virtualized environment?
If so, what Hypervisor are you running?

No, it's on the bare hardware.

If not, IOMMU has close to no meaning (it isn't used even by LLVM, only VT-x/AMD-V). Why do you need the option enabled?

I run VMWare and other virtual machines and I do kernel development. Plus, just because this feature isn't widely used doesn't mean that one has to turn it off because of broken Nvidia drivers. The drivers are still broken nonetheless.

jdb2

[AireTamStrm]

The point isn't that it is not widely used, its that it is questionable that you're even using the feature. (You don't really need it when running a pair of 580s in SLI, that's for sure) I don't think VMWare is using it either, unless you have some bleeding edge version, ESX/ESXi 4 or above, or some other enterprise version.

I have a TYAN S7025 (Dual Processor/NUMA) board with Intel's IOMMU support, and there are no issues with the 260.19.* drivers.

I'm not saying that your issue isn't a bug or that you should not be doing something you are. I'm saying you should make your use case more clear.

[Ilendir]

same here, without IOMMU linux bootes normally.

[zappa86]

I'm having an issue with nvidia drivers / linux which I also resolved by disabling iommu. I run some virtualboxes on my computer and its my understanding that without IOMMU the performance degrades.

I'm using nvidia drivers: 295.20

When I was using linux 3.1.5 I could have IOMMU on in my bios. However, once I upgraded to linux 3.2.6 I had to disable it inorder to get everything working. This may be a linux 3.2.x regression, or linux 3.1.x just worked with broken nvidia drivers. I'm not sure. Is anyone else running on Linux 3.2.x with IOMMU enabled?

[sandipt]

Internally NVIDIA filed bug 935405 to track this issue `RmInitAdapter fails with intel_iommu=on`

[babetz]

Originally Posted by sandipt

Internally NVIDIA filed bug 935405 to track this issue `RmInitAdapter fails with intel_iommu=on`

The same thing happen with amd_iommu, with kernel 3.2 iommu is activeted by default so I had this same issue as I reported on this thread.

[Deanjo]

Same thing here on a Sabertooth 990FX board and running the 295.20 drivers on a 3.2.7 kernel.

[   12.539606] AMD-Vi: Event logged [IO_PAGE_FAULT device=01:00.0 domain=0x0017 address=0x000000021f37a200 flags=0x0010]
[   12.539622] AMD-Vi: Event logged [IO_PAGE_FAULT device=01:00.0 domain=0x0017 address=0x000000e02e32d000 flags=0x0010]
[   12.539632] AMD-Vi: Event logged [IO_PAGE_FAULT device=01:00.0 domain=0x0017 address=0x00000088e6ff2000 flags=0x0010]
[   12.539641] AMD-Vi: Event logged [IO_PAGE_FAULT device=01:00.0 domain=0x0017 address=0x0000003aa1f2b000 flags=0x0010]
[   12.539649] AMD-Vi: Event logged [IO_PAGE_FAULT device=01:00.0 domain=0x0017 address=0x00000088e6ff2040 flags=0x0010]
[   12.539657] AMD-Vi: Event logged [IO_PAGE_FAULT device=01:00.0 domain=0x0017 address=0x0000003aa1f2b040 flags=0x0010]
[   20.551296] NVRM: RmInitAdapter failed! (0x27:0x38:1190)
[   20.551304] NVRM: rm_init_adapter(0) failed

[marisn]

Just jumping to "me too".

Dell Latitude E6500 w 3.2.9 running Gentoo ~AMD64 with 295.20

When I enable intel_iommu, system goes just fine till the X start moment. Screen goes blank and in a few seconds also backlight goes off. Hard reboot + intel_iommu=off required to start normally.

[Deanjo]

Originally Posted by sandipt

Internally NVIDIA filed bug 935405 to track this issue `RmInitAdapter fails with intel_iommu=on`

As mentioned this also occurs with amd_iommu (in fact that is what the original report was on). It would really be nice to see this one resolved soon as it is holding our companies HCL approval.

[Deanjo]

Do the 295.33 (long-lived branch release) for Linux x86/x86_64 release address this issue?