Re: [PATCH] ARM: kirkwood: add missing <linux/if_ether.h> for ETH_ALEN

Hi Andrew,

On Sat, Aug 07, 2021 at 04:17:44PM +0200, Andrew Lunn wrote:

quoted

What kernel is this? I've just tested with this exact commit as
base and it compiles just fine.

I'm not saying including the file is wrong, but it seems it isn't
needed in the upstream kernel and I don't know if it qualifies for
the stable queue therefore.

I would like to see a reproducer for mainline. Do you have a kernel
config which generates the problem.

I encountered the problem when building the 'kirkwood' target in
OpenWrt. I have now tried building vanilla, and the problem indeed
doesn't exist. After tracing the header includes with the precompiler
for some time I concluded that <linux/of_net.h> included in kirkwood.c
includes <linux/phy.h> which includes <linux/ethtool.h> which includes
<uapi/linux/ethtool.h> which includes <uapi/linux/if_ether.h> which
includes <linux/if_ether.h> which defined ETH_ALEN.

When building OpenWrt kernel which includes a backport of
"of: net: pass the dst buffer to of_get_mac_address()", this is not the
same as <linux/of_net.h> doesn't include <linux/phy.h> yet. This is
because we miss commit 0c65b2b90d13c1 ("net: of_get_phy_mode: Change
API to solve int/unit warnings") which has been in mainline for a long
time.

The change itself does seems reasonable, so if we can reproduce it, i
would be happy to merge it for stable.

Sorry for the noise caused, I'm not sure what the policy is in this
case, but certainly this is *not* a regression which should make it to
stable asap. The long and confusing chain of includes which lead to the
ETH_ALEN macro being defined in arch/arm/mach-mvebu/kirkwood.c is
certainly not ideal, and in case you still consider this patch worth
merging, I will post v2 with re-written commit description.