Re: [PATCH v2 1/2] net: dsa: ensure linearized SKBs in case of tail taggers

From: Lino Sanfilippo <LinoSanfilippo@gmx.de>
To: Vladimir Oltean <olteanv@gmail.com>
Cc: Andrew Lunn <andrew@lunn.ch>,
	Florian Fainelli <f.fainelli@gmail.com>,
	woojung.huh@microchip.com, UNGLinuxDriver@microchip.com,
	vivien.didelot@gmail.com, davem@davemloft.net, kuba@kernel.org,
	netdev@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2 1/2] net: dsa: ensure linearized SKBs in case of tail taggers
Date: Fri, 23 Jul 2021 15:41:21 +0200	[thread overview]
Message-ID: <e5438617-8dc5-b9c8-7176-91075ca1fc36@gmx.de> (raw)
In-Reply-To: <20210723122213.fvhudwyk36u7pw52@skbuf>

Hi Vladimir,

On 23.07.21 at 14:22, Vladimir Oltean wrote:
> On Fri, Jul 23, 2021 at 09:47:39AM +0200, Lino Sanfilippo wrote:
>> since I got a message that the patches have already been applied to netdev/net.git.
>> How should I proceed if I want to send a new version of the series? Just ignore the
>> merge to netdev and send the patches nevertheless?
>
> Since the git history is immutable you need to work with what is already
> in the current net/master branch. What do you want to change, just
> address the feedback I gave? If that is all, just don't bother, I intend
> to look at adding a framework through which the DSA master can declare
> what features it supports in conjunction with specific DSA tagging protocols.
> That is material for net-next, and Dave took your patch at the last
> minute for the "net" pull request towards Linus' tree. If you send
> another patch on "net" in that area now, we'd have to wait for another
> week or two until "net" will be merged again into "net-next". Not sure
> if it's worth it. The only thing that was of concern to me is that you
> assign the DSA interface's slave->vlan_features = master->vlan_features.
> So even though you clear the NETIF_F_SG feature for the DSA slave
> interface, VLAN uppers on top of DSA interfaces will still have NETIF_F_SG.
> However, those skbs will be linearized during the dev_queue_xmit call
> done by the 8021q driver towards DSA, so in the end, the way in which
> you restructured the code may not be cosmetically ideal, but also
> appears to not be functionally problematic.
> Anyway, your patch will probably conflict with the stable trees (the
> tag_ops->needed_tailroom was introduced very recently), so we will have
> another chance to fix it up when Greg sends the email that the patch
> failed to apply.
>

Yes, I just wanted to address your feedback concerning the feature assignment
in a new patch version. But as you explained this is not needed and would make
things just unnecessary complicated. So lets wait and see if there are any
conflicts with Gregs stable tree. Thanks for the explanation.

Best Regards,
Lino