Re: [net-next PATCH 13/14] net: phy: qcom: deatch qca83xx PHY driver from at803x

[Christian Marangi <ansuelsmth@gmail.com>]

On Wed, Nov 29, 2023 at 11:20:25AM +0000, Russell King (Oracle) wrote:
> On Wed, Nov 29, 2023 at 11:37:56AM +0100, Christian Marangi wrote:
> > On Wed, Nov 29, 2023 at 09:53:00AM +0000, Russell King (Oracle) wrote:
> > > On Wed, Nov 29, 2023 at 03:12:18AM +0100, Christian Marangi wrote:
> > > > diff --git a/drivers/net/phy/qcom/Makefile b/drivers/net/phy/qcom/Makefile
> > > > index 6a68da8aaa7b..43e4d14df8ea 100644
> > > > --- a/drivers/net/phy/qcom/Makefile
> > > > +++ b/drivers/net/phy/qcom/Makefile
> > > > @@ -1,2 +1,3 @@
> > > >  # SPDX-License-Identifier: GPL-2.0
> > > > -obj-$(CONFIG_AT803X_PHY)	+= at803x.o
> > > > +obj-$(CONFIG_AT803X_PHY)	+= at803x.o common.o
> > > > +obj-$(CONFIG_QCA83XX_PHY)	+= qca83xx.o common.o
> > > 
> > > These PHY drivers can be built as modules. You will end up with several
> > > modules - at803x.ko, qca83xx.ko and common.ko. You don't mark any
> > > functions in common.c as exported, no module license, no author, no
> > > description. common.ko is way too generic a name as well.
> > > 
> > > Please think about this more and test building these drivers as a
> > > module.
> > >
> > 
> > Had some fear about this...
> > 
> > What would be the preferred way for this?
> > 
> > Having a .ko that EXPORT symbol or making the PHY driver .ko to compile
> > the common.o in it?
> 
> I think the former, otherwise we end up with common.o duplicated in
> each module, which becomes unnecessary bloat. This is how the Broadcom
> stuff (which also has a "library") does it.
> 
> > Honestly I would like the second option since I would prefer not to
> > create a .ko with shared function and EXPORT lots of symbols. On SoC it's
> > expected to have only one of the PHY (at max 2 when the qca807x PHY will
> > be implemented, with the at808x also present) so the size increase is
> > minimal.
> > 
> > (just to be more clear, talking about this makefile implementation)
> > 
> > at803x-objs			+= common.o
> > obj-$(CONFIG_AT803X_PHY)	+= at803x.o
> > qca83xx-objs			+= common.o
> > obj-$(CONFIG_QCA83XX_PHY)	+= qca83xx.o
> > qca808x-objs			+= common.o
> > obj-$(CONFIG_QCA808X_PHY)	+= qca808x.o
> 
> That won't work - the -objs needs to list the corresponding .o file
> as well, and it needs to be a different name (you can't do this:
> 
> qca808x-objs			+= common.o qca808x.o
> 
> it has to be something like:
> 
> qca808x-phy-objs		+= common.o qca808x.o
> obj-$(CONFIG_QCA808X_PHY)	+= qca808x-phy.o
> 
> However, I don't like this because it means each module ends up with
> a copy of common.o in it.
> 

Ok will do the .ko library approach np.

> > For name of common.c, is qcom_ethphy_common.c a better name?
> 
> or qcom-phy-lib.c which follows what we have for Broadcom.
>

Think we have to add some reference that it's about ethernet somewhere.

There are already lots of qcom-phy .ko outside net/ 

qcom-ethphy-lib.c ?

-- 
	Ansuel