> That one doesn't actually say what the point is either. It just claims > that there *is* a point that isn't actually explained or mentioned > further. I can't comment on Nick' lockless get_user_pages, but I can talk about the XIP part. As far as I understand it XIP is used with Linux in two kinds of systems. The very small (embedded ARM/MIPS) and the very big (s390). XIP support for the s390 use case was merged long ago but did not address the needs of the embedded use case. Not surprising given how different the systems are. 5 year old funky patches for the embedded use case have been the norm for hundreds of embedded designs, including millions of Japanese phones. When I asked Andrew for advice on how to proceed to get the funky patches merged he said, "We have an XIP subsystem, why don't you make that work?" When I looked in to the issue I found that the vm needed a little tweaking before we could leverage the filemap_xip.c work. Fortunately Nick was eager to help. The result of that effort was the VM_MIXEDMAP patch and then Nick's get_xip_page->get_xip_address patch. In a uniquely Linux situation we now have advocates of the smallest embedded systems and the big mainframe excited about the same patchset. The only problem is we are having problems sorting out how to deal with vm_normal_page() for the VM_MIXEDMAP case such that it works well for both ARM and s390. Hence the pte_special bit. Seems like the result is a reasonable compromise to me, I'm surprised Nicks patch is as clean as it is. Many of the other proposed solutions are very awkward for one or the other. - To unsubscribe from this list: send the line "unsubscribe linux-arch" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html