Scalable Pagefaults
Subsystems (2)
We should continue the conversation from last year on the topic of page fault scalability. I presume that by the time of the conference Suren's current patches for per-VMA locks [1] [2] will be at least in Andrew's tree, even if not quite upstream yet. We will then be in a good place to discuss enhancements:
- File-backed VMAs
- UFFD
- Swap
- Improve performance for low-thread-count apps
- Full RCU handling of (some) page faults
Suren Baghdasaryan, Liam Howlett, Michel Lespinasse, Laurent Dufour, Peter Xu would all be good participants.
[1] https://lore.kernel.org/linux-mm/[email protected]/
[2] https://lwn.net/Articles/906852/
Quoted 11 lines · Matthew Wilcox
We should continue the conversation from last year on the topic of page fault scalability. I presume that by the time of the conference Suren's current patches for per-VMA locks [1] [2] will be at least in Andrew's tree, even if not quite upstream yet. We will then be in a good place to discuss enhancements:
- File-backed VMAs
- UFFD
- Swap
- Improve performance for low-thread-count apps
- Full RCU handling of (some) page faults
A few other enhancements that we should discuss:
- Adding protection of detached VMAs to mmap_lock page fault handler
- Using detached VMAs to stop stack expansion and allow mmap_write_lock to downgrade regardless of surrounding VMAs.
- Fast mmap_write_downgrade by using detached VMAs in other code paths.
Quoted 8 lines
Suren Baghdasaryan, Liam Howlett, Michel Lespinasse, Laurent Dufour, Peter Xu would all be good participants.
[1] https://lore.kernel.org/linux-mm/[email protected]/
[2] https://lwn.net/Articles/906852/