Message signal interrupts: Rule understanding

Mukesh Yadav mukesh.fkd at gmail.com
Tue Dec 27 12:49:51 EST 2011


Hi Guys,

Want to know some details about MSI queues.
Looked into many only online references for MSI, they points mainly to
introduction area.
I am looking for rule/decision criteria for distributing traffic on
different MSI queues.
Can we configure MSI queue at runtime for traffic distribution.

On my machine I can see below data in /proc/interrupts
 64:     218601      13334      25585       7505   PCI-MSI-edge      eth0-0
 65:    5717754    6491556     501052     729091   PCI-MSI-edge      eth0-1
 66:     844740    9897919         96        230   PCI-MSI-edge      eth0-2
 67:     222750     436800        403    1205846   PCI-MSI-edge      eth0-3
 68:        777    1502281     314536     100125   PCI-MSI-edge      eth0-4
 69:     482616     431247    2164970    1627540   PCI-MSI-edge      eth0-5
 70:     323501    1433873      81970      18359   PCI-MSI-edge      eth0-6
 71:      37298      35844       8271      18516   PCI-MSI-edge      eth0-7

When send UDP packets with IP X.X.X.100 interrupt happens in eth0-1 and
with IP X.X.X.101 interrupts happens in eth0-5.
Similary for Ip X.X.X.102 and X.X.X.104, interrupt happens on distinct MSI
queue...Seems like modules 4 operation here being criteria.
Can we have vlan based MSI queue rules?

When I set the core affinity,  with /proc/irq/<intrrpt no>/smp_affinity. It
changes on its own when I send burst of UPD packets with above IP as dest.
Is this expected?

Thanks
Mukesh
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20111227/292876e1/attachment.html 


More information about the Kernelnewbies mailing list