[openib-general] openib/MPI + bigphysarea data corruption problem

[Roland Dreier]

    Matt> We've been looking into how openib does the mapping,
    Matt> pinning, and freeing of the user space buffer, but we're
    Matt> getting a little lost between get_user_pages(), put_page(),
    Matt> and how it all relates to the page cache.

    Matt> Can someone please comment on whether there's someway for
    Matt> openib to accept this unusual memory?

The userspace verbs code calls get_user_pages() to get a list of page
structs for the user memory being registered.  I'm not familiar with
how the bigphysarea patch works, so I'm not sure if get_user_pages()
will work on memory allocated this way.

By the way, what MPI library are you using?  I was not aware that an
MPI binding for the openib userspace verbs was available yet.

 - R.