[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAM_iQpUmPv8BjFLyS9faR+OCfeW5L6=qyLem5MhgXxk=J2iiyw@mail.gmail.com>
Date: Sun, 19 Sep 2021 16:38:30 -0700
From: Cong Wang <xiyou.wangcong@...il.com>
To: Jakub Kicinski <kuba@...nel.org>
Cc: David Miller <davem@...emloft.net>,
Linux Kernel Network Developers <netdev@...r.kernel.org>,
Jamal Hadi Salim <jhs@...atatu.com>,
Jiri Pirko <jiri@...nulli.us>,
Eric Dumazet <edumazet@...gle.com>,
Matthew Massey <matthewmassey@...com>,
Dave Taht <dave.taht@...il.com>
Subject: Re: [PATCH net-next 1/3] net: sched: update default qdisc visibility
after Tx queue cnt changes
On Fri, Sep 17, 2021 at 6:09 AM Jakub Kicinski <kuba@...nel.org> wrote:
>
> On Thu, 16 Sep 2021 22:46:42 -0700 Cong Wang wrote:
> > On Wed, Sep 15, 2021 at 12:36 PM Jakub Kicinski <kuba@...nel.org> wrote:
> > > On Wed, 15 Sep 2021 09:31:08 -0700 Cong Wang wrote:
> > > > Don't we need to flip the device with dev_deactivate()+dev_activate()?
> > > > It looks like the only thing this function resets is qdisc itself, and only
> > > > partially.
> > >
> > > We're only making the qdiscs visible, there should be
> > > no datapath-visible change.
> >
> > Isn't every qdisc under mq visible to datapath?
> >
> > Packets can be pending in qdisc's, and qdisc's can be scheduled
> > in TX softirq, so essentially we need to flip the device like other
>
> By visible I mean tc qdisc dump shows them. I'm adding/removing
> the qdiscs to netdev->qdisc_hash. That's only used by control
> paths to dump or find qdiscs by handle.
Oh, I see, I missed this difference. If datapath can't see this change
immediately, we don't need to flip it.
Thanks.
Powered by blists - more mailing lists