On Fri, Sep 29, 2006 at 02:31:16AM +0300, Muli Ben-Yehuda wrote:On Thu, Sep 28, 2006 at 07:03:28PM -0400, Jeff Garzik wrote:
hmmmm. What kernels did you test?mainline as of today + several unrelated Calgary patches I'll post
shortly + your PCI domains patch + my Calgary patch. I'll test with
iommu=off next.
That should narrow down the problems. A problem with aic94xx sorta sounds like something unrelated.Not necessarily - Calgary is an isolating IOMMU, meaning that if we
set up a mapping for aic94xx in the wrong IO space due to a Calgary
bug, aic94xx will fall over and die. Usually however this happens a
lot sooner. Also, we have code in Calgary to detect when an errant DMA
happens and it hasn't triggered in this case.
Ok, turns out it's neither a PCI domains nor Calgary issue, since I
can reproduce it on mainline with iommu=off. Must be an aic94xx
issue, I'll send the details to linux-scsi in a bit.