On Saturday 25 March 2006 17:37, Duncan <1i5t5.duncan@...> wrote 
about '[gentoo-amd64]  Re: 2.6.16 and ndiswrapper':
> I only found another late in the cycle, when I upgraded to 8 gigs of
> memory from 1 gig.  .15 and stable releases thereof work.  .16 doesn't. 
> I haven't had time to trace that one further, however, and it's possible
> I just don't quite have the kernel configured correctly and .15 just
> happens to work anyway.  This issue has to do with SATA, actually, I
> think SCSI. With 8 gig of memory I'd normally configure IOMMU on but
> that doesn't work with .15 or .16.  With it off, .15 works, but .16
> fails when it tries to load the (libata based therefore SCSI based) SATA
> RAID, saying the SCSI device nodes don't exist!  My root is on RAID so
> this is early kernel, where it first tries to load
> SCSI-then-RAID-then-rootfs-read-only, so it's well before anything
> userspace side is running, so it /has/ to be a kernel issue.  There's a
> changelog entry saying they eliminated bounce-buffers for SCSI that I
> think is the problem, since bounce-buffers are >4 gig memory related,
> but as I said, I haven't traced it yet, so I can't say for sure.

I had something like this happening on my 4GB system.  With IOMMU off, I'd 
only have access to like 2G of memory.  With IOMMU on, I had various 
results, ranging from kernel panic before initrd is loaded to missing 
memory (small, like maybe the size of my PCI IO window?) to a completely 
working system with devices properly mapped to memory addresses beyond my 
physical RAM.  These various behaviors were controlled, oddly enough, by 
my BIOS settings.

I'm running a Tyan Dual-Opteron Dual-PCIe board... If you've got a similar 
board, I can share my BIOS and kernel settings and maybe you can resolve 
your IOMMU issues.

-- 
"If there's one thing we've established over the years,
it's that the vast majority of our users don't have the slightest
clue what's best for them in terms of package stability."
-- Gentoo Developer Ciaran McCreesh