Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210127090747.364951-1-xie.he.0141@gmail.com> <20210128114659.2d81a85f@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
In-Reply-To: <20210128114659.2d81a85f@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
From:   Xie He <xie.he.0141@gmail.com>
Date:   Thu, 28 Jan 2021 14:06:58 -0800
Message-ID: <CAJht_EOSB-m--Ombr6wLMFq4mPy8UTpsBri2CPsaRTU-aks7Uw@mail.gmail.com>
Subject: Re: [PATCH net] net: hdlc_x25: Use qdisc to queue outgoing LAPB frames
To:     Jakub Kicinski <kuba@kernel.org>
Cc:     "David S. Miller" <davem@davemloft.net>,
        Linux X25 <linux-x25@vger.kernel.org>,
        Linux Kernel Network Developers <netdev@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Martin Schiller <ms@dev.tdt.de>,
        Krzysztof Halasa <khc@pm.waw.pl>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Jan 28, 2021 at 11:47 AM Jakub Kicinski <kuba@kernel.org> wrote:
>
> Noob question - could you point at or provide a quick guide to layering
> here? I take there is only one netdev, and something maintains an
> internal queue which is not stopped when HW driver stops the qdisc?

Yes, there is only one netdev. The LAPB module (net/lapb/) (which is
used as a library by the netdev driver - hdlc_x25.c) is maintaining an
internal queue which is not stopped when the HW driver stops the
qdisc.

The queue is "write_queue" in "struct lapb_cb" in
"include/net/lapb.h". The code that takes skbs out of the queue and
feeds them to lower layers for transmission is at the "lapb_kick"
function in "net/lapb/lapb_out.c".

The layering is like this:

Upper layer (Layer 3) (net/x25/ or net/packet/)

^
| L3 packets (with control info)
v

The netdev driver (hdlc_x25.c)

^
| L3 packets
v

The LAPB Module (net/lapb/)

^
| LAPB (L2) frames
v

The netdev driver (hdlc_x25.c)

^
| LAPB (L2) frames
| (also called HDLC frames in the context of the HDLC subsystem)
v

HDLC core (hdlc.c)

^
| HDLC frames
v

HDLC Hardware Driver

> Sounds like we're optimizing to prevent drops, and this was not
> reported from production, rather thru code inspection. Ergo I think
> net-next will be more appropriate here, unless Martin disagrees.

Yes, I have no problem in targeting net-next instead. Thanks!