[c-nsp] 3750 high cpu from icmp

[Brian Turnbow]

Hello Everyone,
I have been working on a 3750 that has a high cpu usage and wanted to
ask for some help.
My first thought was tcam space , but that was ok and I don't see any
bad adjacencies or routes.

The switch has high interupt cpu levels and checking into it I have
found that it seems to be related to ICMP messages getting kicked to the
cpu.
sh plat port-asic stats drop 
Shows this increasing counter (a few seconds apart)
Supervisor TxQueue Drop Statistics
    Queue 11: 36618954
 Supervisor TxQueue Drop Statistics
    Queue 11: 36622889

And I have traced this queue down to icmp. The cpu controller shows high
icmp packets arriving to the cpu.(again a few seconds apart)

3750E-Jenner#sh controllers cpu-interface  | i icmp
icmp              1525306547 0          0          0          0
3750E-Jenner#sh controllers cpu-interface  | i icmp
icmp              1525456328 0          0          0          0


Tracing on the vlan I found alot of icmp redirects being bounced around
so I tried disabling redirects and the cpu usage went down dramatically
yet it is still high.

I was able to run a debug 
debug platform cpu-queues icmp-q
And see alot of these messages.
ICMP-Q:Dropped redirect disabled on L3 IF: Local Port Fwding L3If:Vlan82
L2If:FastEthernet1/0/11

It seems that with no redirects the packets gets sent to the cpu that
proceeds to drop the packet.


I tried to implement copp to see about limiting the messages sent to the
cpu , but it does not seem possible on the 3750.
Control-plane is there yet if I try to apply the service policy I get an
error message

QoS: policymap is supported on physical, VLAN, and ES interfaces only
Service Policy attachment failed
error: failed to install policy map control-plane-in

Besides redesigning to avoid icmp redirects anyone have any ideas?

Thanks in advance

Brian