[NET]: rt_check_expire() can take a long time, add a cond_resched()
authorEric Dumazet <dada1@cosmosbay.com>
Thu, 15 Nov 2007 00:14:05 +0000 (16:14 -0800)
committerDavid S. Miller <davem@davemloft.net>
Thu, 15 Nov 2007 00:14:05 +0000 (16:14 -0800)
On commit 39c90ece7565f5c47110c2fa77409d7a9478bd5b:

[IPV4]: Convert rt_check_expire() from softirq processing to workqueue.

we converted rt_check_expire() from softirq to workqueue, allowing the
function to perform all work it was supposed to do.

When the IP route cache is big, rt_check_expire() can take a long time
to run.  (default settings : 20% of the hash table is scanned at each
invocation)

Adding cond_resched() helps giving cpu to higher priority tasks if
necessary.

Using a "if (need_resched())" test before calling "cond_resched();" is
necessary to avoid spending too much time doing the resched check.
(My tests gave a time reduction from 88 ms to 25 ms per
rt_check_expire() run on my i686 test machine)

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/route.c

index 45651834e1e2193bcc537b18e04946dc36e194d6..1bff9ed349ff5a9e5d14ba1f09888e304bd88950 100644 (file)
@@ -578,6 +578,9 @@ static void rt_check_expire(struct work_struct *work)
                i = (i + 1) & rt_hash_mask;
                rthp = &rt_hash_table[i].chain;
 
+               if (need_resched())
+                       cond_resched();
+
                if (*rthp == NULL)
                        continue;
                spin_lock_bh(rt_hash_lock_addr(i));