(no subject)

[Sexton, Matt]

I applied the two patches to 2.6.13-1, but am still the same kind of
mapping problems.

My test [as I described in an earlier e-mail] has a PCI device DMA data
into a hugepage memory buffer, whose physical page addresses were
obtained via get_user_pages().  The user process, which allocated the
memory via mmap(), then checks the data.  It's often seeing the data in
the wrong page.

Have there been any relevant changes since 2.6.13-1?

Any suggestions for how I can provide more useful debugging information?

Thanks,
Matt