Re: The "Out of IOMMU space" error and the "Please enable the IOMMU option" option

[Maurice Volaski <mvolaski@aecom.yu.edu>]

With 2.6.16.20 and iommu=memaper=4,

Jun 19 15:20:31 [kernel] [  103.697664] PCI-DMA: Disabling AGP.
Jun 19 15:20:31 [kernel] [  103.698226] PCI-DMA: using GART IOMMU.
Jun 19 15:20:31 [kernel] [  103.698341] PCI-DMA: Reserving 512MB of 
IOMMU area in the AGP aperture

I'm not sure why it tells me it's disabling AGP, but otherwise it looks good.

> On Thursday 15 June 2006 23:17, Maurice Volaski wrote:
> >  Occasionally,  we get errors like these:
> >
> >  Jun  9 00:56:35 [kernel] [18020.416620] PCI-DMA: Out of IOMMU space
> >  for 12288 bytes at device 0000:03:01.0
>
>
> >  and it looks like it might have something to do with these messages
> >  (after updating and setting BIOS as below):
> >
> >  Jun 15 16:38:54 [kernel] [    0.000000] Checking aperture...
> >  Jun 15 16:38:54 [kernel] [    0.000000] CPU 0: aperture @ 0 size 256 MB
> >  Jun 15 16:38:54 [kernel] [    0.000000] No AGP bridge found
> >  Jun 15 16:38:54 [kernel] [    0.000000] Your BIOS doesn't leave a
> >  aperture memory hole
> >  Jun 15 16:38:54 [kernel] [    0.000000] Please enable the IOMMU
> >  option in the BIOS setup
> >  Jun 15 16:38:54 [kernel] [    0.000000] This costs you 64 MB of RAM
> >  Jun 15 16:38:54 [kernel] [    0.000000] Mapping aperture over 65536
> >  KB of RAM @ 8000000
> >
> >  The box is a dual Opteron 250 with an Arima HDAMA Rev. D motherboard.
>
> Not directly - the kernel will create a fallback aperture if it can't
> find one provided by the BIOS. The default is 64MB, but it can be
> enlarged with iommu=memaper=4
>
> There was also a bug that might have caused this over longer uptimes.
> If you still see  it with latest 2.6.16.stable (19 or so).
>
> If it still happens with those two changes then something is likely
> leaking mappings or your have a extreme workload that needs
> a lot of mappings.
>
>
> >  First, am I correct to assume that I'm not getting the 256 MB?
>
> Yes you'll get 64MB fallback likely.
>
> >  Second, is the the case the BIOS is lying?
>
> Probably yes.
>
> -Andi


--

Maurice Volaski, mvolaski@aecom.yu.edu
Computing Support, Rose F. Kennedy Center
Albert Einstein College of Medicine of Yeshiva University
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/