Current git fails to boot with VT-D enabled due to flood of DMAR errors

Current git fails to boot with VT-D enabled due to flood of DMAR errors

[Robert Hancock]

On an Asus P7P55D Pro-based system (Lynnfield CPU), with Intel VT-d 
enabled in the BIOS, on 2.6.30.5 (the previous version I tested) I get 
these lines in dmesg:

DMAR:Host address width 36
DMAR:DRHD (flags: 0x00000001)base: 0x00000000fed90000
DMAR:RMRR base: 0x00000000000e4000 end: 0x00000000000e7fff
DMAR:RMRR base: 0x00000000bf7ec000 end: 0x00000000bf7fffff
Not all IO-APIC's listed under remapping hardware

In current -git, I get these lines as well, but then I get this:

DRHD: handling fault status reg ffffffff
DMAR:[DMA Read] Request device [ff:1f.7] fault addr fffffffffffff000
DMAR:[fault reason 255] Unknown

The last two lines repeat in a continuous flood and the system doesn't 
boot. This happens even if I pass "intel_iommu=off" (which seems like 
it's already the default).

Could be a kernel bug or a BIOS bug, but in the latter case we should 
probably deal with it better. The registers all reading FF seems like 
maybe the registers are disabled or something..

Re: Current git fails to boot with VT-D enabled due to flood of DMAR errors

[David Woodhouse]

On Sat, 2009-09-19 at 22:32 -0600, Robert Hancock wrote:
> On an Asus P7P55D Pro-based system (Lynnfield CPU), with Intel VT-d 
> enabled in the BIOS, on 2.6.30.5 (the previous version I tested) I get 
> these lines in dmesg:
> 
> DMAR:Host address width 36
> DMAR:DRHD (flags: 0x00000001)base: 0x00000000fed90000
> DMAR:RMRR base: 0x00000000000e4000 end: 0x00000000000e7fff
> DMAR:RMRR base: 0x00000000bf7ec000 end: 0x00000000bf7fffff
> Not all IO-APIC's listed under remapping hardware
> 
> In current -git, I get these lines as well, but then I get this:
> 
> DRHD: handling fault status reg ffffffff
> DMAR:[DMA Read] Request device [ff:1f.7] fault addr fffffffffffff000
> DMAR:[fault reason 255] Unknown
> 
> The last two lines repeat in a continuous flood and the system doesn't 
> boot. This happens even if I pass "intel_iommu=off" (which seems like 
> it's already the default).
> 
> Could be a kernel bug or a BIOS bug, but in the latter case we should 
> probably deal with it better. The registers all reading FF seems like 
> maybe the registers are disabled or something..

This ought to be 'fixed' in linux-next by this commit:
http://git.infradead.org/iommu-2.6.git/commitdiff/0815565ad

Please try linux-next or the iommu tree from
git://git.infradead.org/iommu-2.6.git

-- 
David Woodhouse                            Open Source Technology Centre
David.Woodhouse@intel.com                              Intel Corporation

Re: Current git fails to boot with VT-D enabled due to flood of DMAR errors

[Robert Hancock]

On 09/19/2009 10:41 PM, David Woodhouse wrote:
> On Sat, 2009-09-19 at 22:32 -0600, Robert Hancock wrote:
>> On an Asus P7P55D Pro-based system (Lynnfield CPU), with Intel VT-d
>> enabled in the BIOS, on 2.6.30.5 (the previous version I tested) I get
>> these lines in dmesg:
>>
>> DMAR:Host address width 36
>> DMAR:DRHD (flags: 0x00000001)base: 0x00000000fed90000
>> DMAR:RMRR base: 0x00000000000e4000 end: 0x00000000000e7fff
>> DMAR:RMRR base: 0x00000000bf7ec000 end: 0x00000000bf7fffff
>> Not all IO-APIC's listed under remapping hardware
>>
>> In current -git, I get these lines as well, but then I get this:
>>
>> DRHD: handling fault status reg ffffffff
>> DMAR:[DMA Read] Request device [ff:1f.7] fault addr fffffffffffff000
>> DMAR:[fault reason 255] Unknown
>>
>> The last two lines repeat in a continuous flood and the system doesn't
>> boot. This happens even if I pass "intel_iommu=off" (which seems like
>> it's already the default).
>>
>> Could be a kernel bug or a BIOS bug, but in the latter case we should
>> probably deal with it better. The registers all reading FF seems like
>> maybe the registers are disabled or something..
>
> This ought to be 'fixed' in linux-next by this commit:
> http://git.infradead.org/iommu-2.6.git/commitdiff/0815565ad
>
> Please try linux-next or the iommu tree from
> git://git.infradead.org/iommu-2.6.git

Yeah, I just came across that patch. Is that going to get pushed for 
2.6.32? It seems like something caused us to start tripping over this by 
default when we didn't before. (Passing intel_iommu=on with 2.6.30.5 
does cause the same problem.)

I'll try and report the problem to Asus, obviously something's borked 
with VT-d in their BIOS currently..

Re: Current git fails to boot with VT-D enabled due to flood of DMAR errors

[David Woodhouse]

On Sat, 2009-09-19 at 22:46 -0600, Robert Hancock wrote:
> 
> Yeah, I just came across that patch. Is that going to get pushed for 
> 2.6.32? It seems like something caused us to start tripping over this
> by default when we didn't before. (Passing intel_iommu=on with
> 2.6.30.5 does cause the same problem.)

Yes, I'll push it in the next day or so. Just waiting for another
last-minute fix to be tested by someone other than me, and then I'll ask
Linus to pull.

> I'll try and report the problem to Asus, obviously something's borked 
> with VT-d in their BIOS currently..

Good luck with that. Their normal response is to say "it boots Windows
so it must be perfect". Despite the fact that Windows doesn't actually
use VT-d. Idiots.

-- 
dwmw2