Re: [Qemu-devel] [PULL v3 4/7] rdma: core logic

[Michael R. Hines]

Thanks. Comments inline....

On 04/17/2013 05:16 AM, Paolo Bonzini wrote:

> +/*
> + * Create protection domain and completion queues
> + */
> +static int qemu_rdma_alloc_pd_cq(RDMAContext *rdma)
> +{
> +    /* allocate pd */
> +    rdma->pd = ibv_alloc_pd(rdma->verbs);
> +    if (!rdma->pd) {
> +        fprintf(stderr, "failed to allocate protection domain\n");
> +        return -1;
> +    }
> +
> +    /* create completion channel */
> +    rdma->comp_channel = ibv_create_comp_channel(rdma->verbs);
> +    if (!rdma->comp_channel) {
> +        fprintf(stderr, "failed to allocate completion channel\n");
> +        goto err_alloc_pd_cq;
> +    }
> +
> +    qemu_set_nonblock(rdma->comp_channel->fd);
> Again showing my lack of IB-fu, do you have to test for POLLOUT
> somewhere too?  Or does it just work?


There is no output to that file descriptor - it's strictly used for read()
in order to know when events occur.

I've already incorporated the yield() coroutine patch you requested,
so you'll see it in my next posting.

> > @@ -121,7 +125,6 @@ void process_incoming_migration(QEMUFile *f)
> >       Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
> >       int fd = qemu_get_fd(f);
> >   
> > -    assert(fd != -1);
> >       qemu_set_nonblock(fd);
> So in the end you are setting the fd to non-blocking. :)
>
> Please add a get_fd implementation to QEMUFile that returns
> rdma->comp_channel->fd, and leave the assertion in place.  The fd will
> be set twice to nonblocking, but that's ok.

Cool. =)