Re: [PATCH] net: usb: lan78xx: add weak dependency with micrel phy module

[Dragan Simic <dsimic@manjaro.org>]

Hello Greg, Jose and Luis,

On 2024-07-29 08:13, Greg KH wrote:
> On Mon, Jul 29, 2024 at 06:43:40AM +0200, Dragan Simic wrote:
>> On 2024-07-28 22:57, Andrew Lunn wrote:
>> > > In other words, this patch doesn't subtract anything.  Instead, it
>> > > just
>> > > adds a weakdep link between the lan78xx and micrel modules, so the
>> > > kernel
>> > > itself can report that dependency, which may actually result in one
>> > > more
>> > > PHY driver added to a generated initial ramdisk.
>> >
>> > So at the moment, does the initramfs contain all PHY modules? I guess
>> > it does, because you have no knowledge which are actually needed. And
>> > this does not help you in any way, as you said, it does not subtract
>> > anything.
>> 
>> Basically, an initial ramdisk shouldn't contain any PHY modules that
>> aren't automatically detected as needed on a particular system, for
>> which the initial ramdisk is built.  That's how selecting modules
>> while building the initial ramdisks works.  On the other hand, if it's
>> some initial ramdisk built by a Linux distribution and intended to
>> support multiple systems or boards, it may contain whatever the
>> distribution sees fit.
>> 
>> Having weakdeps defined actually does help here.  For example, a Linux
>> distribution mentioned above no longer needs to hand-craft the rules
>> for initial ramdisk generation for the PHY modules that should be put
>> into an initial ramdisk together with the lan78xx driver, if the Linux
>> distribution chooses to include the lax78xx driver.  Having weakdep(s)
>> defined makes the kernel do that instead.  Also, there's no point in
>> including every single PHY driver module, because not all of them are
>> needed for a particular selection of MAC drivers, which comes from the
>> intended purpose of the initial ramdisk built by a Linux distribution,
>> i.e. the target architecture, supported board category, etc.
>> 
>> Let's also keep in mind that including all PHY modules into an initial
>> ramdisk inevitably makes it larger, which often isn't an option for
>> resource-constrained embedded systems.
> 
> resource-constrained embedded systems know their dependancies and their
> hardware configurations, so I don't see how the weak-deps help at all
> here.
> 
> You are arguing two different things it seems, neither of which this
> change helps out at all with, so I will provide a:
> 
>   Nacked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> 
> here until it gets straightened out.

Quite frankly, all this makes me wonder why weakdeps were merged into
the mainline kernel [1] with no real consumers?  Perhaps this is good
time for Jose and Luis to chime in.

[1] 
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/include/linux/module.h?id=61842868de13aa7fd7391c626e889f4d6f1450bf