On 02.10.23 19:33, David Hildenbrand wrote:
On 02.10.23 17:21, Peter Xu wrote:
On Mon, Oct 02, 2023 at 10:00:03AM +0200, David Hildenbrand wrote:
In case we cannot simply remap the page, the fallback sequence (from the cover letter) would be triggered.
- UFFDIO_COPY
- MADV_DONTNEED
So we would just handle the operation internally without a fallback.
Note that I think there will be a slight difference on whole remap atomicity, on what happens if the page is modified after UFFDIO_COPY but before DONTNEED.
If the page is writable (implies PAE), we can always move it. If it is R/O, it cannot change before we get a page fault and grab the PT lock (well, and page lock).
So I think something atomic can be implemented without too much issues.
UFFDIO_REMAP guarantees full atomicity when moving the page, IOW, threads can be updating the pages when ioctl(UFFDIO_REMAP), data won't get lost during movement, and it will generate a missing event after moved, with latest data showing up on dest.
If the page has to be copied, grab a reference and unmap it, then copy it and map it into the new process. Should be doable and handle all kinds of situations just fine.
Just throwing out ideas to get a less low-level interface.
[if one really wants to get notified when one cannot move without a copy, one could have a flag for such power users to control the behavior]
[of course, if someone would have a GUP-pin on such a page, the page exchange would be observable. Just have to documented the UFFDIO_MOVE semantics properly]