[PATCH v4 21/22] pdcp: add thread safe processing

[Stephen Hemminger]

On Sat, 27 May 2023 02:32:13 +0530
Anoob Joseph <anoobj at marvell.com> wrote:

> +	/* Lock to protect concurrent updates */
> +	rte_rwlock_t rwl;
>  	/** Bitmap memory region

Reader/writer locks are significantly more expensive than simple
spin lock. Even uncontended access requires additional cache update.
Unless there is case with a long region where lock is held,
just use a spin lock.

See some Paul McKenney's locking talks (LCA video is good) about
lock overheads in Linux. Since DPDK has same set of instructions
to use; the same issues apply to userspace.