tipc: ensure that link congestion and wakeup use same criteria
authorJon Paul Maloy <jon.maloy@ericsson.com>
Tue, 16 Aug 2016 15:53:51 +0000 (11:53 -0400)
committerDavid S. Miller <davem@davemloft.net>
Fri, 19 Aug 2016 04:14:37 +0000 (21:14 -0700)
When a link is attempted woken up after congestion, it uses a different,
more generous criteria than when it was originally declared congested.
This has the effect that the link, and the sending process, sometimes
will be woken up unnecessarily, just to immediately return to congestion
when it turns out there is not not enough space in its send queue to
host the pending message. This is a waste of CPU cycles.

We now change the function link_prepare_wakeup() to use exactly the same
criteria as tipc_link_xmit(). However, since we are now excluding the
window limit from the wakeup calculation, and the current backlog limit
for the lowest level is too small to house even a single maximum-size
message, we have to expand this limit. We do this by evaluating an
alternative, minimum value during the setting of the importance limits.

Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/tipc/link.c

index 877d94f..2c6e1b9 100644 (file)
@@ -807,7 +807,7 @@ void link_prepare_wakeup(struct tipc_link *l)
 
        skb_queue_walk_safe(&l->wakeupq, skb, tmp) {
                imp = TIPC_SKB_CB(skb)->chain_imp;
-               lim = l->window + l->backlog[imp].limit;
+               lim = l->backlog[imp].limit;
                pnd[imp] += TIPC_SKB_CB(skb)->chain_sz;
                if ((pnd[imp] + l->backlog[imp].len) >= lim)
                        break;
@@ -873,9 +873,11 @@ int tipc_link_xmit(struct tipc_link *l, struct sk_buff_head *list,
        struct sk_buff *skb, *_skb, *bskb;
 
        /* Match msg importance against this and all higher backlog limits: */
-       for (i = imp; i <= TIPC_SYSTEM_IMPORTANCE; i++) {
-               if (unlikely(l->backlog[i].len >= l->backlog[i].limit))
-                       return link_schedule_user(l, list);
+       if (!skb_queue_empty(backlogq)) {
+               for (i = imp; i <= TIPC_SYSTEM_IMPORTANCE; i++) {
+                       if (unlikely(l->backlog[i].len >= l->backlog[i].limit))
+                               return link_schedule_user(l, list);
+               }
        }
        if (unlikely(msg_size(hdr) > mtu)) {
                skb_queue_purge(list);
@@ -1692,10 +1694,10 @@ void tipc_link_set_queue_limits(struct tipc_link *l, u32 win)
        int max_bulk = TIPC_MAX_PUBLICATIONS / (l->mtu / ITEM_SIZE);
 
        l->window = win;
-       l->backlog[TIPC_LOW_IMPORTANCE].limit      = win / 2;
-       l->backlog[TIPC_MEDIUM_IMPORTANCE].limit   = win;
-       l->backlog[TIPC_HIGH_IMPORTANCE].limit     = win / 2 * 3;
-       l->backlog[TIPC_CRITICAL_IMPORTANCE].limit = win * 2;
+       l->backlog[TIPC_LOW_IMPORTANCE].limit      = max_t(u16, 50, win);
+       l->backlog[TIPC_MEDIUM_IMPORTANCE].limit   = max_t(u16, 100, win * 2);
+       l->backlog[TIPC_HIGH_IMPORTANCE].limit     = max_t(u16, 150, win * 3);
+       l->backlog[TIPC_CRITICAL_IMPORTANCE].limit = max_t(u16, 200, win * 4);
        l->backlog[TIPC_SYSTEM_IMPORTANCE].limit   = max_bulk;
 }