netdev - Re: iproute2: tc deletion freezes whole server

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: <CANxWus9vSe=WtggXveB+YW_29fD8_qb-7A1pCgMUHz7SFfKhTA@mail.gmail.com>
Date:   Wed, 15 Apr 2020 17:01:01 +0200
From:   Václav Zindulka <vaclav.zindulka@...pnet.cz>
To:     Cong Wang <xiyou.wangcong@...il.com>
Cc:     Linux Kernel Network Developers <netdev@...r.kernel.org>
Subject: Re: iproute2: tc deletion freezes whole server

On Mon, Apr 13, 2020 at 7:29 PM Cong Wang <xiyou.wangcong@...il.com> wrote:
>
> On Sun, Apr 12, 2020 at 1:18 PM Václav Zindulka
> <vaclav.zindulka@...pnet.cz> wrote:
> >
> > On Wed, Apr 8, 2020 at 10:18 PM Cong Wang <xiyou.wangcong@...il.com> wrote:
> > >
> > > Hi, Václav
> >
> > Hello Cong,
> >
> > > Sorry for the delay.
> >
> > No problem, I'm actually glad you are diagnosing the problem further.
> > I didn't have much time until today and late yesterday to apply
> > patches and to test it.
> >
> > > The problem is actually more complicated than I thought, although it
> > > needs more work, below is the first pile of patches I have for you to
> > > test:
> > >
> > > https://github.com/congwang/linux/commits/qdisc_reset
> > >
> > > It is based on the latest net-next branch. Please let me know the result.
> >
> > I have applied all the patches in your four commits to my custom 5.4.6
> > kernel source. There was no change in the amount of fq_codel_reset
> > calls. Tested on ifb, RJ45 and SFP+ interfaces.
>
> It is true my patches do not reduce the number of fq_codel_reset() calls,
> they are intended to reduce the CPU time spent in each fq_codel_reset().
>
> Can you measure this? Note, you do not have to add your own printk()
> any more, because my patches add a few tracepoints, especially for
> qdisc_reset(). So you can obtain the time by checking the timestamps
> of these trace events. Of course, you can also use perf trace like you
> did before.

Sorry for delayed responses. We were moving to a new house so I didn't
have much time to test it. I've measured your pile of patches applied
vs unpatched kernel. Result is a little bit better, but it is only
about 1s faster. Results are here. Do you need any additional reports
or measurements of other interfaces?
https://github.com/zvalcav/tc-kernel/tree/master/20200415 I've
recompiled the kernel without printk which had some overhead too.