Did Your System Take A Dive With Linux 3.2?

[allquixotic]

80% of IOMMU hardware implementations today simply do not work. They are fatally flawed and don't work the way the drivers expect them to. The hardware manufacturers are either incapable or unwilling to fix the problem.

I was getting random GPU hangs even on my ATI Radeon system until I disabled VT-d in the BIOS.

IOMMU doesn't work unless you get very lucky and use the correct hardware.

[Kano]

I think U should at least disable intel_iommu by default. I see all xxx_iommu_default_on options are enabled, but this one really seems to affect several systems.

http://cateee.net/lkddb/web-lkddb/INTEL_IOMMU_DEFAULT_ON.html

[Aquous]

How is this a regression? DMAR has never worked on my Intel system, it's always spammed these messages and eventually locked up...

[Kano]

Most likely the day of the regression was when it was turned on by default. In U 3.0.0 the option was called differently and U had it unset:

# CONFIG_DMAR_DEFAULT_ON is not set

[xpander]

boots fine without workarounds
Host: xpander-desktop
Kernel: Linux 3.2.0-030200-generic x86_64
Distro: Linux Mint 12 Lisa
CPU: AMD Phenom(tm) II X4 955 Processor
4 cores/threads
3200.000 MHz
RAM: 7.79 GiB
GPU: GeForce GTX 560 Ti/PCI/SSE2
OpenGL 4.2.0 NVIDIA 290.10
GLSL 4.20 NVIDIA via Cg compiler

[Kano]

Did you notice that you dont have got an intel cpu? also for amd-v (the amd variant of vt-d for iommu) you need a 890fx or 9xx chipset which you did not mention.

[xpander]

have Gigabyte GA-970A-D3
but yes i noticed the "workaroud" was for intel cpu's
just saying that not everybody have the issues with that kernel because not everybody have intels

[AlbertP]

Intel Core 2 Duo here, GL960+ICH8-M chipset. Never had any problem like this on 3.1 or 3.2-rc7. But indeed, this chipset doesn't support VT-d.

[AnonymousCoward]

I think U should at least disable intel_iommu by default.

They already fixed that in V.

[drag]

Just a FYI for people reading this thread who may not understand what DMAR is for.

DMAR is 'DMA Remapping'.

It's a necessary feature for secure PCI pass through for virtualization. Without it a malicious person in control of a KVM or Xen guest can create interrupts to trigger a exploit to gain access to the host system.

All the IOMMU virtualization support for AMD supported DMAR from the beginning. For Intel older cpus/chipsets are vulnerable, but newer devices support DMAR, which I guess why people here are experiencing problems.

I have a AMD systems that supports IOMMU and I am not experiencing the issues, BTW. But I am running a Fedora 16 kernel.