[PATCH] fix free swap cache latency
commit6f5e6b9e69bf043074a0edabe3d271899c34eb79
authorHugh Dickins <hugh@veritas.com>
Fri, 17 Mar 2006 07:04:09 +0000 (16 23:04 -0800)
committerLinus Torvalds <torvalds@g5.osdl.org>
Fri, 17 Mar 2006 15:51:26 +0000 (17 07:51 -0800)
tree8833d7267d6653a2bf786b055f05640c9a64dc4d
parent7670f023aabd976c25862e4c6fb9f6d9d2758153
[PATCH] fix free swap cache latency

Lee Revell reported 28ms latency when process with lots of swapped memory
exits.

2.6.15 introduced a latency regression when unmapping: in accounting the
zap_work latency breaker, pte_none counted 1, pte_present PAGE_SIZE, but a
swap entry counted nothing at all.  We think of pages present as the slow
case, but Lee's trace shows that free_swap_and_cache's radix tree lookup
can make a lot of work - and we could have been doing it many thousands of
times without a latency break.

Move the zap_work update up to account swap entries like pages present.
This does account non-linear pte_file entries, and unmap_mapping_range
skipping over swap entries, by the same amount even though they're quick:
but neither of those cases deserves complicating the code (and they're
treated no worse than they were in 2.6.14).

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Nick Piggin <npiggin@suse.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
mm/memory.c