Hi,
This is a second version of two patches sent earlier:
http://marc.info/?t=138039800600001&r=1&w=2
http://marc.info/?t=138039800600002&r=1&w=2
The patch 1 is changed so that it won't touch the init/probe path. The
driver will still fail if CPU0 is not around, but that was not accepted
to be a bug.
The patch 2 is unmodified.
Test cases for EdgeRouter Lite:
1)
- Start pinging Octeon from a remote host with 1 second interval.
- Wait 10 seconds.
- Set /proc/irq/24/smp_affinity to 2.
- Wait 10 seconds.
- Quit ping.
- Expected result:
No packet loss reported.
- Unexpected/failed result:
Packet loss.
2)
- Boot Octeon with kernel parameter "octeon-ethernet.max_rx_cpus=1".
- Saturate the Octeon with external traffic (e.g. flood ping).
- Expected result:
Max 50% CPU time is consumed. Second core is available for
other tasks.
- Unexpected/failed result:
100% CPU time is consumed. System is unresponsive.
Aaro Koskinen (2):
staging: octeon-ethernet: allow to set IRQ smp_affinity freely
staging: octeon-ethernet: allow to use only 1 CPU for packet
processing
drivers/staging/octeon/ethernet-rx.c | 11 +++++------
1 file changed, 5 insertions(+), 6 deletions(-)
--
1.8.4.rc3
_______________________________________________
devel mailing list
[email protected]
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
