The use of 64 bit time introduces an expensive 64 bit
division operation. Since the time lapse being calculated
in osc_cache_too_much will never be more than seventy years
we can cast the time lapse to an long and perform a normal
32 bit divison operation instead.

Signed-off-by: James Simmons <uja.o...@yahoo.com>
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-8835
Reviewed-on: https://review.whamcloud.com/23814
Reviewed-by: Andreas Dilger <andreas.dil...@intel.com>
Reviewed-by: Dmitry Eremin <dmitry.ere...@intel.com>
Signed-off-by: James Simmons <jsimm...@infradead.org>
---
 drivers/staging/lustre/lustre/osc/osc_page.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/drivers/staging/lustre/lustre/osc/osc_page.c 
b/drivers/staging/lustre/lustre/osc/osc_page.c
index 0461408..ab9d0d7 100644
--- a/drivers/staging/lustre/lustre/osc/osc_page.c
+++ b/drivers/staging/lustre/lustre/osc/osc_page.c
@@ -370,12 +370,17 @@ static int osc_cache_too_much(struct client_obd *cli)
                        return lru_shrink_min(cli);
        } else {
                time64_t duration = ktime_get_real_seconds();
+               long timediff;
 
                /* knock out pages by duration of no IO activity */
                duration -= cli->cl_lru_last_used;
-               duration >>= 6; /* approximately 1 minute */
-               if (duration > 0 &&
-                   pages >= div64_s64((s64)budget, duration))
+               /*
+                * The difference shouldn't be more than 70 years
+                * so we can safely case to a long. Round to
+                * approximately 1 minute.
+                */
+               timediff = (long)(duration >> 6);
+               if (timediff > 0 && pages >= budget / timediff)
                        return lru_shrink_min(cli);
        }
        return 0;
-- 
1.8.3.1

_______________________________________________
devel mailing list
de...@linuxdriverproject.org
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel

Reply via email to