I wonder if we could make some improvements to zapping pages to
reduce TLB flushes under PTL, and to single threaded pte updates
to reduce atomic operations.
This might require some changes to arch code, particularly the
last patch. I'd just like to see if I've missed something
fundamental with the mm or with pte/tlb behaviour.
Thanks,
Nick
Nicholas Piggin (4):
mm: munmap optimise single threaded page freeing
mm: zap_pte_range only flush under ptl if a dirty shared page was
unmapped
mm: zap_pte_range optimise fullmm handling for dirty shared pages
mm: optimise flushing and pte manipulation for single threaded access
include/asm-generic/tlb.h | 3 +++
mm/huge_memory.c | 4 ++--
mm/madvise.c | 4 ++--
mm/memory.c | 40 ++++++++++++++++++++++++++++++++-------
4 files changed, 40 insertions(+), 11 deletions(-)
--
2.17.0
