On Wed, Nov 20, 2024 at 2:10 PM Shakeel Butt <shakeel.butt@xxxxxxxxx> wrote: > > On Tue, Nov 19, 2024 at 04:08:21PM -0800, Suren Baghdasaryan wrote: > > Back when per-vma locks were introduces, vm_lock was moved out of > > vm_area_struct in [1] because of the performance regression caused by > > false cacheline sharing. Recent investigation [2] revealed that the > > regressions is limited to a rather old Broadwell microarchitecture and > > even there it can be mitigated by disabling adjacent cacheline > > prefetching, see [3]. > > If 'struct vm_area_struct' is prone to performance issues due to > cacheline misalignments then we should do something about the > __randomize_layout tag for it. I imagine we can identify the fields > which might be performance critical to be on same cacheline or different > cacheline due to false sharing then we can divide the fields into > different cacheline groups and fields can be __randomize_layout within > the group. WDYT? I think that's a good idea since shuffling these fields around does affect performance. I can look into it once these changes get finalized and the layout gets more stable. >
