From today's conf call, I was asked to post info on the iommu issue I'm
hitting on ppc64 systems. This bug shows up as data corruption when I
run stressful mvapich2 tests that force lots of dma mappings in the ppc
iommu code. The problem happens on kernels with the iommu size != host
page size. This is the default config for rh5.1 at least. Workaround
is to make the host page size 4KB.
Here is my original thread on this:
http://lkml.org/lkml/2007/12/20/368
Included in that thread is a proposed kernel fix from Ben that I tested ok.
Email if you have questions.
Steve.
_______________________________________________
ewg mailing list
ewg@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/ewg