Re: [PATCH for-rc] IB/hfi1: Insure pq is not left on waitlist

[Jason Gunthorpe <jgg@ziepe.ca>]

On Tue, Mar 17, 2020 at 12:05:10PM -0400, Dennis Dalessandro wrote:
> +static void flush_pq_iowait(struct hfi1_user_sdma_pkt_q *pq)
> +{
> +	unsigned long flags;
> +	seqlock_t *lock = pq->busy.lock;
> +
> +	if (!lock)
> +		return;
> +	write_seqlock_irqsave(lock, flags);
> +	if (!list_empty(&pq->busy.list)) {
> +		list_del_init(&pq->busy.list);
> +		pq->busy.lock = NULL;
> +	}
> +	write_sequnlock_irqrestore(lock, flags);

I'm trying to grasp how a seqlock is protecting a list_empty and
list_del_init, and this seems.. uh.. insane?

The only place that uses seqlock in infiniband is in hfi1

It only calls seqlock_init and write_seqlock

Never read_seqlock

So, this isn't a seqlock, it is a normal spinlock obfuscated as a
seqlock.

Please clean this mess too.

I don't know what to do with this patch, it might well be functionally
right, but everything about reading it screams wrong wrong wrong. I
don't want to send it to Linus in -rc like this.

At least add something to the commit message asking temporary
forgiveness for this madness.

Also s/insure/ensure/, right?

Jason