On Wed, Nov 24, 2021 at 02:29:38PM +0100, David Hildenbrand wrote: > On 24.11.21 14:28, Jason Gunthorpe wrote: > > On Wed, Nov 24, 2021 at 02:25:09PM +0100, David Hildenbrand wrote: > >> On 24.11.21 14:23, Jason Gunthorpe wrote: > >>> On Wed, Nov 24, 2021 at 09:57:32AM +0100, David Hildenbrand wrote: > >>> > >>>> Unfortunately it will only be a band aid AFAIU. I can rewrite my > >>>> reproducer fairly easily to pin the whole 2M range first, pin a second > >>>> time only a single page, and then unpin the 2M range, resulting in the > >>>> very same way to block THP. (I can block some THP less because I always > >>>> need the possibility to memlock 2M first, though). > >>> > >>> Oh! > >>> > >>> The issue is GUP always pins an entire compound, no matter how little > >>> the user requests. > >> > >> That's a different issue. I make sure to split the compound page before > >> pinning anything :) > > > > ?? Where is that done in GUP? > > It's done in my reproducer manually. Aren't there many ways for hostile unpriv userspace to cause memory fragmentation? You are picking on pinning here, but any approach that forces the kernel to make a kalloc on a THP subpage would do just as well. Arguably if we want to point to an issue here it is in MADV_FREE/etc that is the direct culprit in allowing userspace to break up THPs and then trigger fragmentation. If the objective is to prevent DOS of THP then MADV_FREE should conserve the THP and migrate the subpages to non-THP memory. FOLL_LONGTERM is not the issue here. Jason
