* [PATCH] netpoll: fix NAPI polling race on SMP
@ 2005-01-11 22:30 Matt Mackall
0 siblings, 0 replies; only message in thread
From: Matt Mackall @ 2005-01-11 22:30 UTC (permalink / raw)
To: Andrew Morton; +Cc: netdev
This avoids a nasty NAPI race by checking that work was actually
scheduled for the CPU netpoll is running on and pulls the
NAPI-specific code out into a separate function.
Original idea from Jeff Moyer
Tested by Andrew Tridgell
Signed-off-by: Matt Mackall <mpm@selenic.com>
Index: netpoll/net/core/netpoll.c
===================================================================
--- netpoll.orig/net/core/netpoll.c 2005-01-10 14:34:49.058616769 -0800
+++ netpoll/net/core/netpoll.c 2005-01-10 14:52:49.799833988 -0800
@@ -65,27 +65,25 @@
return csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
}
-void netpoll_poll(struct netpoll *np)
+/*
+ * Check whether delayed processing was scheduled for our current CPU,
+ * and then manually invoke NAPI polling to pump data off the card.
+ *
+ * In cases where there is bi-directional communications, reading only
+ * one message at a time can lead to packets being dropped by the
+ * network adapter, forcing superfluous retries and possibly timeouts.
+ * Thus, we set our budget to greater than 1.
+ */
+static void poll_napi(struct netpoll *np)
{
- /*
- * In cases where there is bi-directional communications, reading
- * only one message at a time can lead to packets being dropped by
- * the network adapter, forcing superfluous retries and possibly
- * timeouts. Thus, we set our budget to a more reasonable value.
- */
int budget = 16;
unsigned long flags;
+ struct softnet_data *queue;
- if(!np->dev || !netif_running(np->dev) || !np->dev->poll_controller)
- return;
-
- /* Process pending work on NIC */
- np->dev->poll_controller(np->dev);
-
- /* If scheduling is stopped, tickle NAPI bits */
spin_lock_irqsave(&netpoll_poll_lock, flags);
- if (np->dev->poll &&
- test_bit(__LINK_STATE_RX_SCHED, &np->dev->state)) {
+ queue = &__get_cpu_var(softnet_data);
+ if (test_bit(__LINK_STATE_RX_SCHED, &np->dev->state) &&
+ !list_empty(&queue->poll_list)) {
np->dev->netpoll_rx |= NETPOLL_RX_DROP;
atomic_inc(&trapped);
@@ -95,6 +93,17 @@
np->dev->netpoll_rx &= ~NETPOLL_RX_DROP;
}
spin_unlock_irqrestore(&netpoll_poll_lock, flags);
+}
+
+void netpoll_poll(struct netpoll *np)
+{
+ if(!np->dev || !netif_running(np->dev) || !np->dev->poll_controller)
+ return;
+
+ /* Process pending work on NIC */
+ np->dev->poll_controller(np->dev);
+ if (np->dev->poll)
+ poll_napi(np);
zap_completion_queue();
}
--
Mathematics is the supreme nostalgia of our time.
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2005-01-11 22:30 UTC | newest]
Thread overview: (only message) (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2005-01-11 22:30 [PATCH] netpoll: fix NAPI polling race on SMP Matt Mackall
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).