[ofa-general][PATCH] Re: mlx4: Completion EQ per cpu (MP support, Patch 10)

[Roland Dreier]

 > >CQ are not per port but netdevices are bounded to port (its correct that
 > >few of them can be bounded to the same port, eg with different PKEYs or
 > >VLAN tags), maybe it worth thinking on API that either let the ULP
 > >dictate to what CPU/core they want the EQ serving this CQ direct its
 > >interrupts or if the ULP doesn't care, let the driver allocate that in
 > >round robin fashion.
 > We've had some ehca code doing round robin scheme, which is an ehca
 > specific
 > policy.
 > Do you have any thoughts on the approach you want to pursue?

I would just like to see an approach that is fully thought through and
gives a way for applications/kernel drivers to choose a CQ vector based
on some information about what CPU it will go to.

If we want to add a way to allow a request for round-robin, that is
fine, but I don't think we want to change the default to round-robin,
unless someone can come up with a workload where it actually helps.

 > Will it be 2.6.26 or 2.6.27 instead?

Given that we always seem to start this discussion at the end of the
merge window, and then no one follows up, it may be never...

certainly not 2.6.26 at this point.

 - R.