Attempt to work-around possible time warp

A theory - we see gettimeofday() going backwards on some
systems, try and correct for that by just returning 0
for time passed.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

diff --git a/time.c b/time.c
index c6a6287819e83b601ac0584db0b716e4dd6acec4..4fbc98bbc57f8b781d95c7a8362ce8b8ab348700 100644 (file)
--- a/time.c
+++ b/time.c
@@ -8,7 +8,7 @@ static unsigned long ns_granularity;
 
 unsigned long utime_since(struct timeval *s, struct timeval *e)
 {
-       long sec, usec;
+       long sec, usec, ret;
 
        sec = e->tv_sec - s->tv_sec;
        usec = e->tv_usec - s->tv_usec;
@@ -18,8 +18,15 @@ unsigned long utime_since(struct timeval *s, struct timeval *e)
        }
 
        sec *= 1000000UL;
+       ret = sec + usec;
 
-       return sec + usec;
+       /*
+        * time warp bug on some kernels?
+        */
+       if (ret < 0)
+               ret = 0;
+
+       return ret;
 }
 
 unsigned long utime_since_now(struct timeval *s)
@@ -32,7 +39,7 @@ unsigned long utime_since_now(struct timeval *s)
 
 unsigned long mtime_since(struct timeval *s, struct timeval *e)
 {
-       long sec, usec;
+       long sec, usec, ret;
 
        sec = e->tv_sec - s->tv_sec;
        usec = e->tv_usec - s->tv_usec;
@@ -43,8 +50,15 @@ unsigned long mtime_since(struct timeval *s, struct timeval *e)
 
        sec *= 1000UL;
        usec /= 1000UL;
+       ret = sec + usec;
+
+       /*
+        * time warp bug on some kernels?
+        */
+       if (ret < 0)
+               ret = 0;
 
-       return sec + usec;
+       return ret;
 }
 
 unsigned long mtime_since_now(struct timeval *s)