memory-trim: Fix timestamp overflow warning right after reboot.

If OVN is started less than 30 seconds after system boot, it logs:

  |memory_trim|WARN|Detected last active timestamp overflow

The 'now < trim_timeout_ms' is not for a timestamp overflow, but
for the later subtraction.  'now < last_active_ms' is enough to
check for the overflow.

Technically, we shouldn't need to check that now > trim_timeout_ms
before subtraction, because the result should be a signed integer,
but it's cleaner this way.

Fixes: 12ddb1c ("lflow-cache: Automatically trim cache when it becomes inactive.")
Signed-off-by: Ilya Maximets <[email protected]>
Signed-off-by: Dumitru Ceara <[email protected]>
(cherry picked from commit 11cd04a)

controller/lflow-cache.c

@@ -332,21 +332,15 @@ lflow_cache_run(struct lflow_cache *lc)
     }
 
     long long int now = time_msec();
-    if (now < lc->last_active_ms || now < lc->trim_timeout_ms) {
+    if (now < lc->last_active_ms) {
         VLOG_WARN_RL(&rl, "Detected cache last active timestamp overflow");
         lc->recently_active = false;
         lflow_cache_trim__(lc, true);
         return;
     }
 
-    if (now < lc->trim_timeout_ms) {
-        VLOG_WARN_RL(&rl, "Detected very large trim timeout: "
-                     "now %lld ms timeout %"PRIu32" ms",
-                     now, lc->trim_timeout_ms);
-        return;
-    }
-
-    if (now - lc->trim_timeout_ms >= lc->last_active_ms) {
+    if (now > lc->trim_timeout_ms
+        && now - lc->trim_timeout_ms >= lc->last_active_ms) {
         VLOG_INFO_RL(&rl, "Detected cache inactivity "
                     "(last active %lld ms ago): trimming cache",
                     now - lc->last_active_ms);