On Thu, Feb 14, 2019 at 01:12:31PM -0700, Jason Gunthorpe wrote: > On Thu, Feb 14, 2019 at 11:33:53AM -0800, Ira Weiny wrote: > > > > I think it had to do with double accounting pinned and mlocked pages > > > and thus delivering a lower than expected limit to userspace. > > > > > > vfio has this bug, RDMA does not. RDMA has a bug where it can > > > overallocate locked memory, vfio doesn't. > > > > Wouldn't vfio also be able to overallocate if the user had RDMA pinned pages? > > Yes > > > I think the problem is that if the user calls mlock on a large range then both > > vfio and RDMA could potentially overallocate even with this fix. This was your > > initial email to Daniel, I think... And Alex's concern. > > Here are the possibilities > - mlock and pin on the same pages - RDMA respects the limit, VFIO halfs it. > - mlock and pin on different pages - RDMA doubles the limit, VFIO > respects it > - VFIO and RDMA in the same process, the limit is halfed or doubled, depending. > > IHMO we should make VFIO & RDMA the same, and then decide what to do > about case #2. I'm not against that. Sorry if I came across that way. For this series I agree we should make it consistent. > > > > Really unclear how to fix this. The pinned/locked split with two > > > buckets may be the right way. > > > > Are you suggesting that we have 2 user limits? > > This is what RDMA has done since CL's patch. I don't understand? What is the other _user_ limit (other than RLIMIT_MEMLOCK)? > > It is very hard to fix as you need to track how many pages are mlocked > *AND* pinned. Understood. :-/ Ira > > Jason
