On Mon, May 14, 2018 at 02:49:01PM -0700, Andrew Morton wrote: > On Mon, 14 May 2018 20:28:01 +0300 Boaz Harrosh <boazh@xxxxxxxxxx> wrote: > > In this project we utilize a per-core server thread so everything > > is kept local. If we use the regular zap_ptes() API All CPU's > > are scheduled for the unmap, though in our case we know that we > > have only used a single core. The regular zap_ptes adds a very big > > latency on every operation and mostly kills the concurrency of the > > over all system. Because it imposes a serialization between all cores > > I'd have thought that in this situation, only the local CPU's bit is > set in the vma's mm_cpumask() and the remote invalidations are not > performed. Is that a misunderstanding, or is all that stuff not working > correctly? I think you misunderstand Boaz's architecture. He has one thread per CPU, so every bit will be set in the mm's (not vma's) mm_cpumask.
