Re: [PATCH v6 00/10] acpi, clocksource: add GTDT driver and GTDT support in arm_arch_timer

From: Rafael J. Wysocki
Date: Thu Jul 07 2016 - 07:58:52 EST


On Thursday, July 07, 2016 07:12:38 PM Hanjun Guo wrote:
> On 2016/7/6 8:00, Rafael J. Wysocki wrote:
> > On Tue, Jul 5, 2016 at 4:18 PM, Graeme Gregory <gg@xxxxxxxxxxxxxxx> wrote:
> >> On Mon, Jul 04, 2016 at 02:53:20PM +0200, Rafael J. Wysocki wrote:
> >>> On Fri, Jul 1, 2016 at 11:04 PM, Rafael J. Wysocki <rjw@xxxxxxxxxxxxx> wrote:
> >>>> On Friday, July 01, 2016 04:23:40 PM Will Deacon wrote:
> >>>>> On Thu, Jun 30, 2016 at 09:48:02PM +0800, Hanjun Guo wrote:
> >>>>>> On 2016/6/30 21:27, Rafael J. Wysocki wrote:
> >>>>>>> On Thursday, June 30, 2016 10:10:02 AM Hanjun Guo wrote:
> >>>>>>>> GTDT is part of ACPI spec, drivers/acpi/ is for driver code of
> >>>>>>>> ACPI spec, I think it can stay in drivers/acpi/ from this point
> >>>>>>>> of view, am I right?
> >>>>>>>
> >>>>>>> The question is not "Can it?", but "Does it need to?".
> >>>>>>>
> >>>>>>> It is in the spec, but still there's only one architecture needing it.
> >>>>>>>
> >>>>>>> There is no way to test it on any other architecture and no reason to build it
> >>>>>>> for any other architecture, so why does it need to be located in drivers/acpi/ ?
> >>>>>>
> >>>>>> I'm fine to move it to other places such as arch/arm64/kernel/, but I
> >>>>>> would like to ask ARM64 maintainer's suggestion for this.
> >>>>>>
> >>>>>> Will, Catalin, what's your opinion on this?
> >>>>>
> >>>>> We don't have any device-tree code for the architected timer under
> >>>>> arch/arm64, so I don't see why we should need anything for ACPI either.
> >>>>
> >>>> And I don't see a reason for the GTDT code to be there in drivers/acpi/.
> >>>>
> >>>> What gives?
> >>>
> >>> Well, since there are things like acpi_lpss in there, my position here
> >>> is kind of weak. :-)
> >>>
> >>> That said I'm not particularly happy with having them in
> >>> drivers/acpi/, so I definitely won't object against attempts to moving
> >>> them somewhere else.
> >>>
> >>>> Maybe it should go to the same place as the analogus DT code, then?
> >>>
> >>> I'm mostly concerned about how (and by whom) that code is going to be
> >>> maintained going forward, though. I also think it should be made
> >>> clear that it is ARM64-only.
> >>>
> >>
> >> So is this a documentation issue in which case Fu Wei can add that to
> >> the file to explain its limited to ARM64. Or we could even rename the
> >> file acpi_arm64_gtdt.c
> >>
> >> It seems a pity as the comment on this series were minors to block
> >> things on a filename/location.
> >
> > Let me repeat what I said above:
> >
> > I'm mostly concerned about how (and by whom) that code is going to be
> > maintained going forward.
> >
> > This is not about documentation, it is about responsibility.
> >
> > Honestly, I don't think I'm the right maintainer to apply the patch
> > introducing this code and then handle bug reports regarding it and so
> > on. That has to be done by somebody else.
>
> I'm working on ACPI for years and upstreamed the ARM64 ACPI core
> support (with lots of people's help), I'm willing to maintain the ARM64
> ACPI code under drivers/acpi/ if no objections.

OK

Can the ARM64-specific code go under drivers/acpi/arm64/ then, for clarity?

> >
> > That's one thing.
> >
> > Another one is the question I asked a few messages ago: Why having the
> > GTDT code in drivers/acpi/ is actually useful to anyone? It
> > definitely would not be useful to me as the maintainer of
> > drivers/acpi/, but maybe it would be useful to somebody for a specific
> > practical reason. Or is it just "let's put this into drivers/acpi/
> > for the lack of a better place"?
>
> Having GTDT code in drivers/acpi/ is useful as it is code that is used
> by two different subsystems, clocksource and watchdog,and where people
> look by default for utility ACPI code.
>
> If the mostly concerned thing (maintainer ship) is settled down, the
> second question would be easily solved.

Fair enough.

Thanks,
Rafael