crypto: disable preemption while benchmarking RAID5 xor checksumming
authorJim Kukunas <james.t.kukunas@linux.intel.com>
Tue, 22 May 2012 03:54:04 +0000 (13:54 +1000)
committerNeilBrown <neilb@suse.de>
Tue, 22 May 2012 03:54:04 +0000 (13:54 +1000)
With CONFIG_PREEMPT=y, we need to disable preemption while benchmarking
RAID5 xor checksumming to ensure we're actually measuring what we think
we're measuring.

Signed-off-by: Jim Kukunas <james.t.kukunas@linux.intel.com>
Signed-off-by: NeilBrown <neilb@suse.de>
crypto/xor.c

index 87884435bddb15b01abdc52cb801557452cc28f5..84daa1111dad7531089aaae0bf512b2b86a39dfa 100644 (file)
@@ -21,6 +21,7 @@
 #include <linux/gfp.h>
 #include <linux/raid/xor.h>
 #include <linux/jiffies.h>
+#include <linux/preempt.h>
 #include <asm/xor.h>
 
 /* The xor routines to use.  */
@@ -69,6 +70,8 @@ do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
        tmpl->next = template_list;
        template_list = tmpl;
 
+       preempt_disable();
+
        /*
         * Count the number of XORs done during a whole jiffy, and use
         * this to calculate the speed of checksumming.  We use a 2-page
@@ -91,6 +94,8 @@ do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
                        max = count;
        }
 
+       preempt_enable();
+
        speed = max * (HZ * BENCH_SIZE / 1024);
        tmpl->speed = speed;