On Fri, Feb 17, 2023 at 08:13:01AM -0800, Suren Baghdasaryan wrote: > On Fri, Feb 17, 2023 at 2:21 AM Hyeonggon Yoo <42.hyeyoo@xxxxxxxxx> wrote: > > > > On Fri, Feb 17, 2023 at 11:15 AM Suren Baghdasaryan <surenb@xxxxxxxxxx> wrote: > > > > > > On Thu, Feb 16, 2023 at 11:43 AM Suren Baghdasaryan <surenb@xxxxxxxxxx> wrote: > > > > > > > > On Thu, Feb 16, 2023 at 7:44 AM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote: > > > > > > > > > > On Wed, Feb 15, 2023 at 09:17:41PM -0800, Suren Baghdasaryan wrote: > > > > > > When vma->anon_vma is not set, page fault handler will set it by either > > > > > > reusing anon_vma of an adjacent VMA if VMAs are compatible or by > > > > > > allocating a new one. find_mergeable_anon_vma() walks VMA tree to find > > > > > > a compatible adjacent VMA and that requires not only the faulting VMA > > > > > > to be stable but also the tree structure and other VMAs inside that tree. > > > > > > Therefore locking just the faulting VMA is not enough for this search. > > > > > > Fall back to taking mmap_lock when vma->anon_vma is not set. This > > > > > > situation happens only on the first page fault and should not affect > > > > > > overall performance. > > > > > > > > > > I think I asked this before, but don't remember getting an aswer. > > > > > Why do we defer setting anon_vma to the first fault? Why don't we > > > > > set it up at mmap time? > > > > > > > > Yeah, I remember that conversation Matthew and I could not find the > > > > definitive answer at the time. I'll look into that again or maybe > > > > someone can answer it here. > > > > > > After looking into it again I'm still under the impression that > > > vma->anon_vma is populated lazily (during the first page fault rather > > > than at mmap time) to avoid doing extra work for areas which are never > > > faulted. Though I might be missing some important detail here. > > > > I think this is because the kernel cannot merge VMAs that have > > different anon_vmas? > > > > Enabling lazy population of anon_vma could potentially increase the > > chances of merging VMAs. > > Hmm. Do you have a clear explanation why merging chances increase this > way? A couple of possibilities I can think of would be: > 1. If after mmap'ing a VMA and before faulting the first page into it > we often change something that affects anon_vma_compatible() decision, > like vm_policy; > 2. When mmap'ing VMAs we do not map them consecutively but the final > arrangement is actually contiguous. > > Don't think either of those cases would be very representative of a > usual case but maybe I'm wrong or there is another reason? Ok. I agree it does not represent common cases. Hmm then I wonder how it went from the initial approach of "allocate anon_vma objects only via fork()" [1] to "populate anon_vma at page faults". [2] [3] Maybe Hugh, Andrea or Andrew have opinions? [1] anon_vma RFC2, lore.kernel.org https://lore.kernel.org/lkml/20040311065254.GT30940@dualathlon.random [2] The status of object-based reverse mapping, LWN.net https://lwn.net/Articles/85908 [3] rmap 39 add anon_vma rmap https://gitlab.com/hyeyoo/linux-historical/-/commit/8aa3448cabdfca146aa3fd36e852d0209fb2276a > > > > > > > In the end rather than changing that logic I decided to skip > > > > vma->anon_vma==NULL cases because I measured them being less than > > > > 0.01% of all page faults, so ROI from changing that would be quite > > > > low. But I agree that the logic is weird and maybe we can improve > > > > that. I will have to review that again when I'm working on eliminating > > > > all these special cases we skip, like swap/userfaults/etc. > > > > -- > > To unsubscribe from this group and stop receiving emails from it, send an email to kernel-team+unsubscribe@xxxxxxxxxxx. > >
