Memory-management short topics: page-table sharing and working sets

Posted Jan 9, 2023 19:31 UTC (Mon) by quotemstr (subscriber, #45331)
Parent article: Memory-management short topics: page-table sharing and working sets

> A process must opt into the COW behavior with a new prctl() command (PR_SET_COW_PTE)

If there's no user-visible behavior change, why make processes opt in?

Memory-management short topics: page-table sharing and working sets

Posted Jan 9, 2023 19:42 UTC (Mon) by mb (subscriber, #50428) [Link] (1 responses)

>why make processes opt in?

There's a whole paragraph in the article about the reasons for why this is opt-in.

Memory-management short topics: page-table sharing and working sets

Posted Jan 9, 2023 19:54 UTC (Mon) by quotemstr (subscriber, #45331) [Link]

Yes, I noticed that. :-) The question stands though: in most other cases, performance optimizations don't have to be individually enabled. (Is there a prctl for folios or EAS?) I understand that the patch currently regresses performance in some use cases, but that's just a signal to me that it's not yet ready. Ideally, the system would tune itself. If it can't --- because the kernel doesn't have enough information to know what a process is going to do --- then the prctl or personality flag or whatever should provide the missing information to the kernel at a high level, not act as a low level toggle for an implementation detail of the mm subsystem.