From: Arnaldo Carvalho de Melo <acme@redhat.com>
Date: Mon, 9 Feb 2009 19:04:06 +0000 (-0200)
Subject: ring_buffer: pahole struct ring_buffer
X-Git-Tag: v2.6.30-rc1~2^2~109^2
X-Git-Url: https://openfabrics.org/gitweb/?a=commitdiff_plain;h=00f62f614bb713027b9296068d1879fbca511eb7;p=~shefty%2Frdma-dev.git

ring_buffer: pahole struct ring_buffer

While fixing some bugs in pahole (built-in.o files were not being
processed due to relocation problems) I found out about these packable
structures:

$ pahole --packable kernel/trace/ring_buffer.o  | grep ring
ring_buffer	72	64	8
ring_buffer_per_cpu	112	104	8

If we take a look at the current layout of struct ring_buffer we can see
that we have two 4 bytes holes.

$ pahole -C ring_buffer kernel/trace/ring_buffer.o
struct ring_buffer {
	unsigned int               pages;           /*     0     4 */
	unsigned int               flags;           /*     4     4 */
	int                        cpus;            /*     8     4 */

	/* XXX 4 bytes hole, try to pack */

	cpumask_var_t              cpumask;         /*    16     8 */
	atomic_t                   record_disabled; /*    24     4 */

	/* XXX 4 bytes hole, try to pack */

	struct mutex               mutex;           /*    32    32 */
	/* --- cacheline 1 boundary (64 bytes) --- */
	struct ring_buffer_per_cpu * * buffers;     /*    64     8 */

	/* size: 72, cachelines: 2, members: 7 */
	/* sum members: 64, holes: 2, sum holes: 8 */
	/* last cacheline: 8 bytes */
};

So, if I ask pahole to reorganize it:

$ pahole -C ring_buffer --reorganize kernel/trace/ring_buffer.o

struct ring_buffer {
	unsigned int               pages;           /*     0     4 */
	unsigned int               flags;           /*     4     4 */
	int                        cpus;            /*     8     4 */
	atomic_t                   record_disabled; /*    12     4 */
	cpumask_var_t              cpumask;         /*    16     8 */
	struct mutex               mutex;           /*    24    32 */
	struct ring_buffer_per_cpu * * buffers;     /*    56     8 */
	/* --- cacheline 1 boundary (64 bytes) --- */

	/* size: 64, cachelines: 1, members: 7 */
};   /* saved 8 bytes and 1 cacheline! */

We get it using just one 64 bytes cacheline.

To see what it did:

$ pahole -C ring_buffer --reorganize --show_reorg_steps \
	kernel/trace/ring_buffer.o | grep \/
/* Moving 'record_disabled' from after 'cpumask' to after 'cpus' */

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---

diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 53ba3a6d16d..27ef3bf13ed 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -273,8 +273,8 @@ struct ring_buffer {
 	unsigned			pages;
 	unsigned			flags;
 	int				cpus;
-	cpumask_var_t			cpumask;
 	atomic_t			record_disabled;
+	cpumask_var_t			cpumask;
 
 	struct mutex			mutex;