Re: Question about atomic FIFO/LIFO queues

[Ruslan Nikolaev <nruslan_devel%yahoo.com@localhost>]

Hi Taylor,

Thanks for you response! Do I understand correctly that you basically 
already have some lockless FIFO queue (in subr_pcq.c), is it correct?

Please let me know if there is any interest for our FIFO queue anywhere 
else in the codebase, not necessarily networking.

Ruslan

On 9/27/20 2:58 PM, Taylor R Campbell wrote:
> > Date: Fri, 25 Sep 2020 17:58:41 -0400
> > From: Ruslan Nikolaev <nruslan_devel%yahoo.com@localhost>
> >
> > I came across this page
> > https://wiki.netbsd.org/projects/project/atomic_fifo_lifo_queues/
> >
> > Recently, we have designed a high-performant (and what is also very
> > important -- truly lock-free -- not just "lockless") and memory
> > efficient FIFO queue based on ring buffers that use FAA (fetch-and-add)
> > and traditional M&S queue as an outer layer (for unbounded queues).
>
> Cool!
>
> I don't actually know what the author of that project description had
> in mind, and he's no longer around, so we should probably just take
> that page down unless someone else remembers and it's still relevant.
> Parts of the network stack use pcq(9) for inter-CPU packet queueing,
> although in new code (and incrementally for old code) we generally try
> to minimize inter-CPU communication except to distribute load.
>
> https://man.NetBSD.org/pcq.9
> https://nxr.NetBSD.org/xref/src/sys/kern/subr_pcq.c