Received-SPF: pass (google.com: domain of linux-kernel+bounces-53024-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Date: Mon, 5 Feb 2024 17:35:50 +0100
From: Herve Codina <herve.codina@bootlin.com>
To: Paolo Abeni <pabeni@redhat.com>
Cc: Vadim Fedorenko <vadim.fedorenko@linux.dev>, "David S. Miller"
 <davem@davemloft.net>, Eric Dumazet <edumazet@google.com>, Jakub Kicinski
 <kuba@kernel.org>, linux-kernel@vger.kernel.org, netdev@vger.kernel.org,
 linuxppc-dev@lists.ozlabs.org, Andrew Lunn <andrew@lunn.ch>, Mark Brown
 <broonie@kernel.org>, Christophe Leroy <christophe.leroy@csgroup.eu>,
 Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Subject: Re: [PATCH v2 1/6] net: wan: Add support for QMC HDLC
Message-ID: <20240205173550.2a6b8f20@bootlin.com>
In-Reply-To: <d174364c12366b1f5eeb616cba078f6682d629f5.camel@redhat.com>
References: <20240130084035.115086-1-herve.codina@bootlin.com>
	<20240130084035.115086-2-herve.codina@bootlin.com>
	<b1968b5c7e88edd448d5f55b57dfa40257b2b06c.camel@redhat.com>
	<20240205152208.73535549@bootlin.com>
	<d174364c12366b1f5eeb616cba078f6682d629f5.camel@redhat.com>
Organization: Bootlin
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

On Mon, 05 Feb 2024 16:49:33 +0100
Paolo Abeni <pabeni@redhat.com> wrote:

[...
> > > 
> > > In general is quite bad that the existing infra does not allow
> > > leveraging NAPI. Have you considered expanding the QMC to accomodate
> > > such user?  
> > 
> > I cannot mask/unmask the 'end of transfer' interrupt.
> > Indeed, other streams use this interrupt among them audio streams and so
> > masking the interrupt for HDLC data will also mask the interrupt for audio
> > data.  
> 
> Uhm... I fear the above makes the available options list empty :(
> 
> > At the HDLC driver level, the best I can to is to store a queue of complete
> > HDLC skbs (queue filled on interrupts) and send them to the network stack
> > when the napi poll() is called.
> > 
> > I am not sure that this kind of queue (additional level between always
> > enabled interrupts and the network stack) makes sense.
> > 
> > Do you have any opinion about this additional queue management for NAPI
> > support?  
> 
> With such idea in place, what HDLC-level data will be accessed by the
> napi context? The RX interrupts will remain unmasked after the
> interrupt and before the napi poll right? That would be
> problematic/could cause drop if the ingress pkt/interrupt rate will be
> higher that what the napi could process - and that in turn could bring
> back old bad livelock times :(

Indeed.
So the best thing to do is to keep this driver without NAPI support.

Best regards,
Hervé