While testing adding/deleting large numbers of interfaces, I found
rt_run_flush() was the #1 cpu user in a kernel profile by far.
The below patch changes rt_run_flush() to only take each spinlock
protecting the rt_hash_table once instead of taking a spinlock for
every hash table bucket (and ending up taking the same small set
of locks over and over).
Deleting 256 interfaces on a 4-way SMP system with 16K buckets reduced
overall cpu-time more than 50% and reduced wall-time about 33%. I
suspect systems with large amounts of memory (and more buckets) will
see an even greater benefit.
Note there is a small change in that rt_free() is called while the
lock is held where before it was called without the lock held. I
don't think this should be an issue.
Signed-off-by: Dave Johnson <[EMAIL PROTECTED]>
===== net/ipv4/route.c 1.162 vs edited =====
--- 1.162/net/ipv4/route.c 2007-02-14 11:09:54 -05:00
+++ edited/net/ipv4/route.c 2007-05-04 17:53:33 -04:00
@@ -237,9 +237,17 @@
for (i = 0; i < RT_HASH_LOCK_SZ; i++) \
spin_lock_init(&rt_hash_locks[i]); \
}
+# define rt_hash_lock_for_each(lockaddr) \
+ for (lockaddr = rt_hash_lock_addr(RT_HASH_LOCK_SZ-1); lockaddr >=
rt_hash_locks; lockaddr--)
+# define rt_hash_lock_for_each_item(locknum, slot) \
+ for (slot = locknum-rt_hash_locks; slot <= rt_hash_mask; slot +=
RT_HASH_LOCK_SZ)
#else
# define rt_hash_lock_addr(slot) NULL
# define rt_hash_lock_init()
+# define rt_hash_lock_for_each(lockaddr) \
+ lockaddr = NULL;
+# define rt_hash_lock_for_each_item(locknum, slot) \
+ for (slot = rt_hash_mask; slot >= 0; slot--)
#endif
static struct rt_hash_bucket *rt_hash_table;
@@ -691,23 +699,26 @@
static void rt_run_flush(unsigned long dummy)
{
int i;
+ spinlock_t *lockaddr;
struct rtable *rth, *next;
rt_deadline = 0;
get_random_bytes(&rt_hash_rnd, 4);
- for (i = rt_hash_mask; i >= 0; i--) {
- spin_lock_bh(rt_hash_lock_addr(i));
- rth = rt_hash_table[i].chain;
- if (rth)
- rt_hash_table[i].chain = NULL;
- spin_unlock_bh(rt_hash_lock_addr(i));
-
- for (; rth; rth = next) {
- next = rth->u.dst.rt_next;
- rt_free(rth);
+ rt_hash_lock_for_each(lockaddr) {
+ spin_lock_bh(lockaddr);
+ rt_hash_lock_for_each_item(lockaddr, i) {
+ rth = rt_hash_table[i].chain;
+ if (rth) {
+ rt_hash_table[i].chain = NULL;
+ for (; rth; rth = next) {
+ next = rth->u.dst.rt_next;
+ rt_free(rth);
+ }
+ }
}
+ spin_unlock_bh(lockaddr);
}
}
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
