Re: [PATCH net-next 1/3] ethtool: add tx aggregation parameters

From: Gal Pressman <gal@nvidia.com>
To: Jakub Kicinski <kuba@kernel.org>, Daniele Palmas <dnlplm@gmail.com>
Cc: "David Miller" <davem@davemloft.net>,
	"Paolo Abeni" <pabeni@redhat.com>,
	"Eric Dumazet" <edumazet@google.com>,
	"Subash Abhinov Kasiviswanathan" <quic_subashab@quicinc.com>,
	"Sean Tranchetti" <quic_stranche@quicinc.com>,
	"Jonathan Corbet" <corbet@lwn.net>, "Bjørn Mork" <bjorn@mork.no>,
	"Greg Kroah-Hartman" <gregkh@linuxfoundation.org>,
	netdev@vger.kernel.org
Subject: Re: [PATCH net-next 1/3] ethtool: add tx aggregation parameters
Date: Tue, 15 Nov 2022 12:59:47 +0200	[thread overview]
Message-ID: <6d41e6da-aaa4-6569-d027-896e25711c86@nvidia.com> (raw)
In-Reply-To: <20221114164238.209f3a9d@kernel.org>

On 15/11/2022 02:42, Jakub Kicinski wrote:
> On Mon, 14 Nov 2022 11:06:19 +0100 Daniele Palmas wrote:
>>> Isn't this the same as TX copybreak? TX
>>> copybreak for multiple packets?  
>> I tried looking at how tx copybreak works to understand your comment,
>> but I could not find any useful document. Probably my fault, but can
>> you please point me to something I can read?
> FWIW it's not exactly copy break, as it applies to all packets.
> But there is indeed an extra copy.
>
> Daniele's explanation is pretty solid, USB devices very often try 
> to pack multiple packets into a single URB for better perf. IIUC Linux
> drivers implement the feature on the Rx side, but fall short of doing
> the same on the Tx side, because there's no API to control it.
>
> I've seen the same thing in the WiFi USB devices I worked on in my
> youth :)

Thanks for the explanation.
How would this apply to a pci netdev driver?