On Wed, Sep 22, 2021 at 10:33 AM Peter Xu peterx@redhat.com wrote:
Hello, Axel,
On Wed, Sep 22, 2021 at 10:04:03AM -0700, Axel Rasmussen wrote:
Thanks for discussing the design Peter. I have some ideas which might make for a nicer v2; I'll massage the code a bit and see what I can come up with.
Sure thing. Note again that as I don't have a strong opinion on that, feel free to keep it. However if you provide v2, I'll read.
[off-topic below]
Another thing I probably have forgot but need your confirmation is, when you worked on uffd minor mode, did you explicitly disable thp, or is it allowed?
I gave a more detailed answer in the other thread, but: currently it is allowed, but this was a bug / oversight on my part. :) THP collapse can break the guarantees minor fault registration is trying to provide.
I think your approach of checking the VMA flags *in retract_page_tables specifically* is correct, and a similar thing should be done for minor registered VMAs too.
When I'm reworking the uffd-wp series, I noticed that commit e1e267c7928f ("khugepaged: skip collapse if uffd-wp detected", 2020-04-07) was actually awkward and not efficient, as we can simply lookup the vma flags for detecting uffd-wp enablement. I'm preparing a patch for it to do it by checking vmas (and that patch will also pave the way for file-backed).
Then I noticed we need similar thing for minor mode?
I think the answer is yes, but I didn't see any code that explicitly handled thp for minor mode, do you remember?
To be explicit, what if in mcontinue_atomic_pte() we get a shmem_getpage() call with a thp returned? Will minor mode break?
Ah so this I am not quite as sure about.
The issue I was describing in the other thread was more about THP collapse racing with UFFDIO_CONTINUE. E.g., collapsing after registration has happened, but before faults have been resolved.
But there's another scenario: what if the collapse happened well before registration happened? I *think* the existing code deals with THPs correctly in that case, but then again I don't think our selftest really covers this case, and it's not something I've tested in production either (to work around the other bug, we currently MADV_NOHUGEPAGE the area until after VM demand paging completes, and the UFFD registration is removed), so I am not super confident this is the case.
I plan to post the khugepaged patch soon and I plan to cover minor mode too there, but I'm not sure whether that's enough, as the thp can be there from the 1st day I think, but I could have missed something.
-- Peter Xu