Re: [PATCH v3 3/6] fuse: {io-uring} Use bitmaps to track registered queues

[Joanne Koong <joannelkoong@gmail.com>]

> @@ -983,6 +1031,10 @@ static void fuse_uring_do_register(struct fuse_ring_ent *ent,
>         struct fuse_ring *ring = queue->ring;
>         struct fuse_conn *fc = ring->fc;
>         struct fuse_iqueue *fiq = &fc->iq;
> +       int node = cpu_to_node(queue->qid);

Am I reading the correct version of the libfuse implementation in [1]?
As I understand it, the libfuse implementation sets queue->qid
sequentially no matter the user-specified cpu-set [2], but the qid is
used as the cpu id, and we get the numa node from that. My
understanding is that in practice, sequential CPU numbering often
follows NUMA topology (eg NUMA node0 cpus: 0-15, 32-47; NUMA node1
cpus: 16-31), so it seems like this has a high chance of being all on
the same numa node? Or am I missing something here?

Thanks,
Joanne

[1] https://github.com/bsbernd/libfuse/commit/afcc654c8760f0c95f9e042fc92f05b1b9c3df13
[2] https://github.com/bsbernd/libfuse/blob/afcc654c8760f0c95f9e042fc92f05b1b9c3df13/lib/fuse_uring.c#L830


> +
> +       if (WARN_ON_ONCE(node >= ring->nr_numa_nodes))
> +               node = 0;
>
>         fuse_uring_prepare_cancel(cmd, issue_flags, ent);
>
> @@ -991,6 +1043,9 @@ static void fuse_uring_do_register(struct fuse_ring_ent *ent,
>         fuse_uring_ent_avail(ent, queue);
>         spin_unlock(&queue->lock);
>
> +       cpumask_set_cpu(queue->qid, ring->registered_q_mask);
> +       cpumask_set_cpu(queue->qid, ring->numa_registered_q_mask[node]);
> +
>         if (!ring->ready) {
>                 bool ready = is_ring_ready(ring, queue->qid);
>