On Wed, Nov 30, 2022 at 7:00 PM Pavan Kondeti <quic_pkondeti@xxxxxxxxxxx> wrote: > > On Wed, Nov 30, 2022 at 03:17:39PM -0800, Andrew Morton wrote: > > > > On Wed, 9 Nov 2022 10:48:36 +0530 Pavankumar Kondeti <quic_pkondeti@xxxxxxxxxxx> wrote: > > > > > When MADV_PAGEOUT is called on a private file mapping VMA region, > > > we bail out early if the process is neither owner nor write capable > > > of the file. However, this VMA may have both private/shared clean > > > pages and private dirty pages. The opportunity of paging out the > > > private dirty pages (Anon pages) is missed. Fix this by caching > > > the file access check and use it later along with PageAnon() during > > > page walk. > > > > > > We observe ~10% improvement in zram usage, thus leaving more available > > > memory on a 4GB RAM system running Android. > > > > > > > Could we please have some reviewer input on this? > > > > Thanks. > > > > Thanks Andrew for the reminder. Fyi, this patch has been included in Android > Generic Kernel Image (5.10 and 5.15 kernels) as we have seen good savings on > Android. It would make a difference on a low memory android devices. > > Suren/Minchan, > > Can you please do the needful? Yeah, I saw this patch before and it makes sense to me. When discussing it with Minchan one concern was that if the vma has no private dirty pages then we will be wasting cpu cycles scanning it. However I guess it's the choice of the userspace process to call madvise() on such mappings and risk wasting some cycles... > > Thanks, > Pavan
