Re: [PATCH net-next 1/2] net: dsa: mv88e6xxx: Provide dummy implementations for trunk setters

On Fri, Jan 15, 2021 at 15:48, Andrew Lunn [off-list ref] wrote:

On Fri, Jan 15, 2021 at 04:36:49PM +0200, Vladimir Oltean wrote:

quoted

On Fri, Jan 15, 2021 at 03:30:30PM +0100, Andrew Lunn wrote:

quoted

On Fri, Jan 15, 2021 at 11:58:33AM +0100, Tobias Waldekranz wrote:

quoted

Support for Global 2 registers is build-time optional.

I was never particularly happy about that. Maybe we should revisit
what features we loose when global 2 is dropped, and see if it still
makes sense to have it as optional?

Marvell switch newbie here, what do you mean "global 2 is dropped"?

I was not aware detect() actually enforced it when needed. It used to
be, you could leave it out, and you would just get reduced
functionality for devices which had global2, but the code was not
compiled in.

At the beginning of the life of this driver, i guess it was maybe
25%/75% without/with global2, so it might of made sense to reduce the
binary size. But today the driver is much bigger with lots of other
things which those early chips don't have, SERDES for example. And
that ratio has dramatically reduced, there are very few devices
without those registers. This is why i think we can make our lives
easier and make global2 always compiled in.

Hear, hear!

I took a quick look at the (stripped) object sizes (ppc32):

# du -ab
6116    ./global1_vtu.o
5904    ./devlink.o
11500   ./port.o
9640    ./global2.o
3016    ./phy.o
5368    ./global1.o
51784   ./chip.o
9892    ./serdes.o
5140    ./global1_atu.o
1916    ./global2_avb.o
2248    ./global2_scratch.o
948     ./port_hidden.o
1828    ./smi.o
119396  .

So, roughly, you save 10%/13k. That hardly justifies the complexity IMO.

Andrew, do you want to do this? If not, I can look into it.