[ofa-general] can an ib-bonding slave work independently?

Or Gerlitz ogerlitz at Voltaire.com
Tue Jul 21 22:24:47 PDT 2009


Liang Zhen wrote:
> When I dug into code, I found when creating bonding device,
> alloc_netdev(...ether_setup) will always set net_device::type to
> ARPHRD_ETHER,  that means if binding cmid on bonding device,
> rdma_copy_addr() will always set    rdma_dev_addr::dev_type to
> RDMA_NODE_RNIC, seems not correct to me... am I wrong?

Yes wrong, assuming you have access to Linux box with IB stack up and running you 
could trivially see this if setting ipoib/bonding and then running one of the 
librdmacm utils while instructing it to bind to a bond (I recommend you go and do that).
Now how it works? when the first ipoib device is added to the bond, bonding senses a mismatch between its ethertype to the bond one and calls bond_setup_by_slave(), this along with bonding automatically setting fail_over_mac to be BOND_FOM_ACTIVE for ipoib device, does the job.

Or.



More information about the general mailing list