Re: [RFC 00/13] [net-next] drivers/net/Space.c cleanup

[Arnd Bergmann]

On Mon, May 17, 2021 at 4:38 PM Paul Gortmaker
<paul.gortmaker@xxxxxxxxxxxxx> wrote:
> > From: Arnd Bergmann <arnd@xxxxxxxx>
>
> I'd be willing to do a "Phase 2" of 930d52c012b8 ISA net delete;  I'm
> not sure the bounce through stable on the way out does much other than
> muddy the git history.  I'd be tempted to just propose the individual
> deletes and see where that goes....

I think the main benefit of going through drivers/staging would be that
it could be done somewhat more aggressively by moving more of the
presumed-unused hardware out at once, and waiting for media
coverage to wake up any remaining users. Then again, reverting
a removal isn't that different from reverting the move to staging.

If we do it one driver at a time, a good start might be the ones that
don't already have support for probing multiple devices in a loadable
module (ni65,smc9194, cs89x0-isa, cops, ltpc), which would let us
kill off drivers/net/Space.c by always using the module_init() logic.
These are all non-ISAPnP devices, and a second step could be
to remove support for all non-PNP devices (lance, wd80x3,
corkscrew and maybe the ISA WAN devices).

Unless we can agree to remove CONFIG_ISA completely, it's probably
best to leave ISA support for those drivers that also work on PCI, EISA
or m68k hardware, as those have the best chance to still get tested.
3c509 is probably the best maintained, given that it's the only network
driver that got converted to isa_register_driver() from the linux-2.0
style static probing. 3c515 has the distinction of being the only
remaining 100mbit one, and ne2k is obviously the one that one can
still find on ebay or the parent's attic.

       Arnd