netdev - Re: [RCU PATCH 00/14] Remove qdisc lock around ingress Qdisc

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAM_iQpUd=KDVG-6hxQCeHKp+LRx6Yi5jSUo2MEivJiar+QETGg@mail.gmail.com>
Date:	Wed, 12 Mar 2014 11:25:29 -0700
From:	Cong Wang <xiyou.wangcong@...il.com>
To:	John Fastabend <john.fastabend@...il.com>
Cc:	Jamal Hadi Salim <jhs@...atatu.com>,
	Linux Kernel Network Developers <netdev@...r.kernel.org>,
	David Miller <davem@...emloft.net>
Subject: Re: [RCU PATCH 00/14] Remove qdisc lock around ingress Qdisc

On Mon, Mar 10, 2014 at 10:03 AM, John Fastabend
<john.fastabend@...il.com> wrote:
> This series drops the qdisc lock that is currently protecting the
> ingress qdisc. This can be done after the tcf filters are made
> lockless and the statistic accounting is safe to run without locks.
>
> To do this the classifiers are converted to use RCU. This requires
> updating each classifier individually to handle the new copy/update
> requirement and also to update the core list traversals. This is
> done in patches 2-11. This also makes the assumption that updates
> to the tables are infrequent in comparison to the packet per second
> being classified. On a 10Gbps running near line rate we can easily
> produce 12+ million packets per second so IMO this is a reasonable
> assumption. And the updates are serialized by RTNL.
>
> In order to have working statistics patch 13 converts the bstats,
> which do accounting for bytes and packets, into percpu variables
> and the u64_stats_update_{begin|end} infrastructure is used to
> maintain consistent statistics. Because these statistics are also
> used by the estimators those function calls had to be udpated as
> well. So that I didn't have to modify all qdiscs at this time many
> of which don't have an easy path to make lockless the percpu
> statistics are only used when the TCQ_F_LLQDISC flag is set. Its
> worth noting that in the mq and mqprio case sub-qdisc's are already
> mapped 1:1 with TX queues which tend to be equal to the number
> of CPUs in the system so its not clear that removing locking in
> these cases would provide any benefit. Most likely a new qdisc
> written from scratch would be needed to implement a mq-htb or
> mq-tbf qdisc.
>
> As for some history I wrote what was basically these patches some
> time ago and then got stalled working on other things. Cong Wang
> made a proposal to remove the locking around the ingress qdisc
> which then kicked me to get these patches working again.
>
> I have started to test this series and do no see any immediate
> splats or issues. I will continue doing some more testing for the
> rest of the week before submitting without the RCU tag but any
> feedback would be good. If someone has a better idea for the
> percpu union of gnet_stats_basic_* in struct Qdisc or finds it
> paticularly ugly that would be good to know.

I assume you mean RFC, otherwise RCU tag doesn't make
any sense to me.

And what about sch->qstats? You seem still don't touch it
in this patchset, and it is you who pointed out qstats.drop++
is not atomic on all architecture to me. :)
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html