[v2] Old platforms: bring out your dead

[Michael Ellerman]

Arnd Bergmann <arnd at kernel.org> writes:
> On Sun, Jan 17, 2021 at 11:33 AM Michael Ellerman <mpe at ellerman.id.au> wrote:
>> Arnd Bergmann <arnd at kernel.org> writes:
>> > On Fri, Jan 15, 2021 at 9:37 PM John Paul Adrian Glaubitz
>> > <glaubitz at physik.fu-berlin.de> wrote:
>> >>
>> >> Hello Arnd!
>> >>
>> >> > * No objection to removing arch/powerpc/platforms/cell that I
>> >> >   had mentioned I plan to do.
>> >>
>> >> Does this affect the capability to run Linux on the PS3?
>> >>
>> >> If yes, it would be great if it could stay as the PS3 is a rather
>> >> widely used platform although you certainly won't find any PS3
>> >> users on the LKML.
>> >
>> > No, as I wrote in the initial email, I'm planning to move the things
>> > (like spufs) that are shared with PS3 into arch/powerpc/platforms/ps3
>> > and remove the parts that are only used on the IBM blades.
>>
>> As I said a while back, I'm not convinced that's a good idea.
>>
>> The only way I have of testing cell is using a QS22, I don't have a PS3
>> capable of running Linux these days. I worry that if I can't test cell
>> at all then the PS3 support will bit rot.
>
> Fair enough. I must have missed your previous reply and expected
> that your QS22 had stopped being operational years ago and that
> there were already zero machines getting kernel updates.

No worries, we all get lots of email :)

So far that QS22 is still going OK with no signs of imminent failure.

> Are you aware of any other users?

No.

I can think of one spufs patch ~year ago that came from someone I assume
is a real Cell user, but otherwise it's pretty much only Geoff that ever
sends Cell/PS3 patches.

>> I know Geoff tests PS3, but that seems sporadic, I don't think he tests
>> linux-next every day.
>>
>> I also don't think the cell blade support is really causing much in the
>> way of maintenance overhead. The thing that's causing work is spufs, and
>> that would remain either way.
>>
>> I'd be happy to drop any QS20/21 code we have, but I'm not convinced
>> dropping QS22 is a good trade off.
>
> Right, I agree that there is little to gain from dropping QS20/21, the
> only files that I see this would impact are
>
> arch/powerpc/platforms/cell/spider-pci.c
> arch/powerpc/platforms/cell/spider-pic.c
> drivers/net/ethernet/toshiba/spider_net*
>
> Dropping all of native (as opposed to PS3 hypervisor based) Cell support
> would be a useful cleanup I think, but not as long as you still use it.

I would still be happy to drop QS20/21, it's still a few thousand lines
of code, and hasn't been tested for years.

I think it might also let us clean up some of our IO_WORKAROUNDS stuff.

But I agree it's much less of an obvious maintenance win than dropping
all of the native code.

cheers