>Is there any performance overhead in user mode cq polling (
>ibv_poll_cq  function ) compared to doing the polling from kernel mode?

Someone from Mellanox would need to confirm this, but I don't believe so.

The user space call basically goes directly to the provider library.  If you
compare the user space code with the kernel code (e.g. mthca_cq.c:mthca_poll_one
versus mlnx_uvp_cq.c:mthca_poll_one), they look identical.  And I don't think
there's any difference writing the CQ entries for a CQ that was allocated for a
user space client versus a kernel client.

- Sean

_______________________________________________
ofw mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/ofw

Reply via email to