Re: [PATCH] net: core: Fix to store new mtu setting in netdevice.

From: Heiner Kallweit
Date: Wed Jan 02 2019 - 01:27:31 EST


On 02.01.2019 00:36, Andrew Lunn wrote:
>>> Is there a .ndo_change_mtu callback, which does not assign a new mtu itself?
>>>
>> So far all drivers have to do it themselves. But IMO this is more a workaround
>> for the core not doing it. It's something the core should do.
>> Now we can remove this from drivers.
>
> Hi Heiner
>
> I think somebody first needs to review all the ndo_change_mtu
> implementations and check that none do something funny like round to
> multiple of 2 or 4 to satisfy DMA restrictions, etc. If there is such
> a thing, we cannot easily move this into the core.
>
> Andrew
> .
>
Good point. I briefly grepped over all drivers and it's not that many
drivers not using the standard assignment dev->mtu = new_mtu.
Some are doing things like dev->mtu = max(new_mtu, xx) what could be
achieved easier with an appropriate max_mtu setting. But that's a
different story.
And I was under the impression that such things had been checked
because the patch had a "Reviewed-by" from Florian (although he
wasn't on cc).

Heiner