Re: [PATCH 1/4] nvme-tcp: per-controller I/O workqueues

[Christoph Hellwig <hch@lst.de>]

On Wed, Jul 03, 2024 at 03:50:18PM +0200, Hannes Reinecke wrote:
> Implement per-controller I/O workqueues to reduce workqueue contention
> during I/O.

Gee, I can see that you are implementing it.  But given that workqueues
are scalable contexts I'd really like to see where you hit limits.
Maybe it's totally expected, maybe you found a bug in the workqueues,
or maybe you just did because you need some of this later.

You'll need to tell the reviewers and also people finding the inevitable
bugs later on why you are doing this.

> +		queue_work_on(queue->io_cpu, queue->ctrl->io_wq, &queue->io_work);

.. and avoid allthe overly long lines.  Maybe by adding a helper to
de-duplicate this code?

> +		queue_work_on(queue->io_cpu, queue->ctrl->io_wq, &queue->io_work);

> +		queue_work_on(queue->io_cpu, queue->ctrl->io_wq, &queue->io_work);

> +	queue_work_on(queue->io_cpu, queue->ctrl->io_wq, &queue->io_work);

> +		queue_work_on(queue->io_cpu, queue->ctrl->io_wq, &queue->io_work);