On Wed, Jan 25, 2023 at 10:06 AM Yang Shi <shy828301@xxxxxxxxx> wrote: > > On Tue, Jan 24, 2023 at 5:58 PM Zach O'Keefe <zokeefe@xxxxxxxxxx> wrote: > > > > During collapse, in a few places we check to see if a given small page > > has any unaccounted references. If the refcount on the page doesn't > > match our expectations, it must be there is an unknown user concurrently > > interested in the page, and so it's not safe to move the contents > > elsewhere. However, the unaccounted pins are likely an ephemeral state. > > > > In such a situation, make MADV_COLLAPSE set EAGAIN errno, indicating that > > collapse may succeed on retry. > > The page may be DMA pinned (for example, pin_user_pages()), it is not > worth retrying for such pages. But it may also not be worth optimizing > for this case at this point. > > So the patch looks good to me. Reviewed-by: Yang Shi <shy828301@xxxxxxxxx> Thanks as always, Yang, and good point about DMA pinning. As you mentioned, I don't know if it's worth considering that too much right now, as it's unlikely these two uses (MADV_COLLAPSE and DMA pining) would be used together. We can revisit if necessary later if it's an issue, but for now, I think it's a win that MADV_COLLAPSE (+ a bounded userspace retry loop based off erno) is more likely to succeed.
