Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
[PATCH] Fix up per-cpu page batch sizes
The code to clamp batch sizes to 2^n - 1 went missing and an extra
check got added, which must have been a hunk of the "higer order pcp
batch refills" work sneaking in.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
authored by
Nick Piggin
and committed by
Linus Torvalds
0ceaacc9
9a405257
+8
-8
+8
-8
mm/page_alloc.c
+8
-8
mm/page_alloc.c
···
1772
1772
batch = 1;
1773
1773
1774
1774
/*
1775
-
* We will be trying to allcoate bigger chunks of contiguous
1776
-
* memory of the order of fls(batch). This should result in
1777
-
* better cache coloring.
1775
+
* Clamp the batch to a 2^n - 1 value. Having a power
1776
+
* of 2 value was found to be more likely to have
1777
+
* suboptimal cache aliasing properties in some cases.
1778
1778
*
1779
-
* A sanity check also to ensure that batch is still in limits.
1779
+
* For example if 2 tasks are alternately allocating
1780
+
* batches of pages, one task can end up with a lot
1781
+
* of pages of one half of the possible page colors
1782
+
* and the other with pages of the other colors.
1780
1783
*/
1781
-
batch = (1 << fls(batch + batch/2));
1782
-
1783
-
if (fls(batch) >= (PAGE_SHIFT + MAX_ORDER - 2))
1784
-
batch = PAGE_SHIFT + ((MAX_ORDER - 1 - PAGE_SHIFT)/2);
1784
+
batch = (1 << (fls(batch + batch/2)-1)) - 1;
1785
1785
1786
1786
return batch;
1787
1787
}