openwrtv4/target/linux/layerscape/patches-4.4/7021-net-Make-the-netdev-watchdog-aware-of-hardware-multi.patch

49 lines
2 KiB
Diff
Raw Normal View History

From 6aa1eca98ca44f515e10d8058d0ff6db3c8a3c11 Mon Sep 17 00:00:00 2001
From: Bogdan Hamciuc <bogdan.hamciuc@freescale.com>
Date: Fri, 28 Sep 2012 17:04:33 +0300
Subject: [PATCH 21/70] net: Make the netdev watchdog aware of hardware
multiqueue devices
If the netdev declares the NETIF_F_HW_ACCEL_MQ (accelerated multiqueue)
capability, tell the watchdog to consider the per-netdev trans_start
field rather than any individual multiqueue's timestamp. That is
justified by the fact that queues only go in and out of congestion
in groups, not individually, as far as the net device is concerned.
Change-Id: I07a6693bf1f0bb1e9396c5e232452223a511ecc1
Signed-off-by: Bogdan Hamciuc <bogdan.hamciuc@freescale.com>
Reviewed-on: http://git.am.freescale.net:8181/1033
Reviewed-by: Fleming Andrew-AFLEMING <AFLEMING@freescale.com>
Tested-by: Fleming Andrew-AFLEMING <AFLEMING@freescale.com>
---
net/sched/sch_generic.c | 17 ++++++++++++++++-
1 file changed, 16 insertions(+), 1 deletion(-)
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -290,8 +290,23 @@ static void dev_watchdog(unsigned long a
txq = netdev_get_tx_queue(dev, i);
/*
* old device drivers set dev->trans_start
+ *
+ * (Actually, not only "old" devices, but also
+ * those which perform queue management in a
+ * separate hw accelerator. So even though the
+ * net device itself is single-queued, it makes
+ * sense (and is safe, too) to use kernel's
+ * multiqueue interface, specifically to avoid
+ * unnecessary device locking in SMP systems.
+ * In this case, we ought to consider not an
+ * individual txq's timestamp as a congestion
+ * indicator, but the "old" per-netdev field.)
*/
- trans_start = txq->trans_start ? : dev->trans_start;
+ if (dev->features & NETIF_F_HW_ACCEL_MQ)
+ trans_start = dev->trans_start;
+ else
+ trans_start = txq->trans_start ? :
+ dev->trans_start;
if (netif_xmit_stopped(txq) &&
time_after(jiffies, (trans_start +
dev->watchdog_timeo))) {