kernel/sys.c: get rid of expensive divides in groups_sort()
authorEric Dumazet <dada1@cosmosbay.com>
Wed, 6 Feb 2008 09:37:56 +0000 (01:37 -0800)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Wed, 6 Feb 2008 18:41:09 +0000 (10:41 -0800)
commit1bf47346d75790ebd2563d909d48046961c7ffd5
tree0f478764beb8dc4e0c71c5f3d6a657535579fe3a
parent6b2fb3c65844452bb9e8b449d50863d1b36c5dc0
kernel/sys.c: get rid of expensive divides in groups_sort()

groups_sort() can be quite long if user loads a large gid table.

This is because GROUP_AT(group_info, some_integer) uses an integer divide.
So having to do XXX thousand divides during one syscall can lead to very
high latencies.  (NGROUPS_MAX=65536)

In the past (25 Mar 2006), an analog problem was found in groups_search()
(commit d74beb9f33a5f16d2965f11b275e401f225c949d ) and at that time I
changed some variables to unsigned int.

I believe that a more generic fix is to make sure NGROUPS_PER_BLOCK is
unsigned.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/sched.h
kernel/sys.c