[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <49113E31.8080106@cosmosbay.com>
Date: Wed, 05 Nov 2008 07:33:21 +0100
From: Eric Dumazet <dada1@...mosbay.com>
To: linux kernel <linux-kernel@...r.kernel.org>,
Linux Netdev List <netdev@...r.kernel.org>
Subject: [BUG] linux-2.6.28-rc3 regression: IRQ smp_affinities not respected
Hi all
One more problem it seems on 2.6.28-rc3
I wanted to get maximal throughput from my machine on a network bench with 3 Gigabit
links delivering 600.000 packets per second, so I tried to play with smp_affinity to
dedicate one CPU for each NIC.
It worked with 2.6.27, so there is a regression on this part.
Unfortunatly, git bisect is nearly impossible for me as after two steps
the resulting kernel doesnt event boot on this machine (same for oprofile
regression I mentioned earlier)
# grep eth1 /proc/interrupts
45: 20425 20418 20445 20441 20345 20349 20384 20397 PCI-MSI-edge eth1
# cat /proc/irq/45/smp_affinity
ff
# echo 1 >/proc/irq/45/smp_affinity
# cat /proc/irq/45/smp_affinity
01
# grep eth1 /proc/interrupts
45: 20928 20920 20943 20940 20845 20847 20887 20898 PCI-MSI-edge eth1
# grep eth1 /proc/interrupts
45: 21037 21030 21053 21049 20953 20956 20997 21007 PCI-MSI-edge eth1
# grep eth1 /proc/interrupts
45: 21141 21134 21156 21154 21057 21059 21101 21110 PCI-MSI-edge eth1
You can see interrupts keep being spreaded on all CPUS instead of CPU0 only.
That *kills* performance on high end routers and servers.
07:00.0 Ethernet controller: Broadcom Corporation NetXtreme II BCM5708S Gigabit Ethernet (rev 12)
Subsystem: Hewlett-Packard Company NC373i Integrated Multifunction Gigabit Server Adapter
Flags: bus master, 66Mhz, medium devsel, latency 64, IRQ 45
Memory at fa000000 (64-bit, non-prefetchable) [size=32M]
[virtual] Expansion ROM at d1000000 [disabled] [size=16K]
Capabilities: [40] PCI-X non-bridge device.
Capabilities: [48] Power Management version 2
Capabilities: [50] Vital Product Data
Capabilities: [58] Message Signalled Interrupts: 64bit+ Queue=0/0 Enable+
bnx2 driver
Thanks
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists