On 16.02.22 20:04, Linus Torvalds wrote: > [ Added David Hildenbrand to the participants. David, see > > https://bugzilla.kernel.org/show_bug.cgi?id=215616 > > for details ] > Thanks for sharing. > On Wed, Feb 16, 2022 at 8:59 AM Oded Gabbay <oded.gabbay@xxxxxxxxx> wrote: >> >> All the details are in the bug, but the bottom line is that somehow, >> this patch causes corruption when the numa balancing feature is >> enabled AND we don't use process affinity AND we use GUP to pin pages >> so our accelerator can DMA to/from system memory. > > Hmm. I thought all the remaining issues were related to THP - and > David Hildenbrand had a series to fix those up. What I shared so far recently [1] was part 1 of my COW fixes to fix the COW security issues -- missed COW. This fixes 1) of [2]. Part 2 is around fixing "wrong COW" for FOLL_PIN a.k.a. memory corruption. That's essentially what PageAnonExclusive() will be all about, making sure that we don't lose synchronicity between GUP and mm due to a wrong COW. This will fix 3) of [2] Part 3 is converting O_DIRECT to use FOLL_PIN instead of FOLL_GET to similarly fix "wrong COW" for O_DIRECT. John is working on that. This will fix 2) of [2]. > > The fact that it also shows up with numa balancing is a bit > unfortunate, because I think that means that that patch series may not > have caught that case. > > That said - what does "we use GUP to pin pages" mean? Does it actually > use the pinning logic, or just regular old GUP? If it uses FOLL_PIN it might be handled by part 2, if it uses O_DIRECT magic it might be covered by part 3. If neither of both, more work might be needed to convert it to FOLL_PIN, as with the new COW logic we won't be able to have the same guarantees for FOLL_GET as we'll have for FOLL_PIN (which is a difference to our original plans to fix it all [3]). [1] https://lkml.kernel.org/r/20220126095557.32392-1-david@xxxxxxxxxx [2] https://lore.kernel.org/all/3ae33b08-d9ef-f846-56fb-645e3b9b4c66@xxxxxxxxxx/ [3] https://lore.kernel.org/all/20211217113049.23850-1-david@xxxxxxxxxx/T/#u -- Thanks, David / dhildenb
