[openib-general] [PATCH] add HW specific data to libibverbs modify QP, SRQ response

[Roland Dreier]

    Roland> Hmm... it seems simpler to have userspace allocate the
    Roland> memory with mmap() before the resize_cq call, and then
    Roland> pass that new buffer into the resize_cq call.  That way
    Roland> you don't have a window where the kernel is putting
    Roland> completions into a buffer that userspace doesn't know
    Roland> about.

    Ralph> Perhaps. But this way, the code is the same for kernel and
    Ralph> user allocated queues.

I guess there is some benefit there.

    Ralph> Or the new kernel driver needs to handle the old way and
    Ralph> the new way.

Yeah.