Re: [PATCH v3 1/6] ACPI: scan: Extend acpi_walk_dep_device_list()

["Rafael J. Wysocki" <rafael@xxxxxxxxxx>]

On Mon, Mar 8, 2021 at 4:45 PM Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote:
>
> On Mon, Mar 8, 2021 at 2:57 PM Andy Shevchenko
> <andy.shevchenko@xxxxxxxxx> wrote:
> >
> > On Mon, Mar 08, 2021 at 02:36:27PM +0100, Rafael J. Wysocki wrote:
> > > On Sun, Mar 7, 2021 at 9:39 PM Andy Shevchenko
> > > <andy.shevchenko@xxxxxxxxx> wrote:
> > > > On Sun, Mar 7, 2021 at 3:36 PM Daniel Scally <djrscally@xxxxxxxxx> wrote:
> > > > > On 22/02/2021 13:34, Andy Shevchenko wrote:
> > > > > > On Mon, Feb 22, 2021 at 3:12 PM Daniel Scally <djrscally@xxxxxxxxx> wrote:
> > > > > >> The acpi_walk_dep_device_list() is not as generalisable as its name
> > > > > >> implies, serving only to decrement the dependency count for each
> > > > > >> dependent device of the input. Extend the function to instead accept
> > > > > >> a callback which can be applied to all the dependencies in acpi_dep_list.
> > > > > >> Replace all existing calls to the function with calls to a wrapper, passing
> > > > > >> a callback that applies the same dependency reduction.
> > > > > > The code looks okay to me, if it was the initial idea, feel free to add
> > > > > > Reviewed-by: Andy Shevchenko <andy.shevchenko@xxxxxxxxx>
> >
> > ...
> >
> > > > > >> +void acpi_dev_flag_dependency_met(acpi_handle handle)
> >
> > > > > > Since it's acpi_dev_* namespace, perhaps it should take struct acpi_device here?
> > > > >
> > > > > I can do this, but I avoided it because in most of the uses in the
> > > > > kernel currently there's no struct acpi_device, they're just passing
> > > > > ACPI_HANDLE(dev) instead, so I'd need to get the adev with
> > > > > ACPI_COMPANION() in each place. It didn't seem worth it...
> > >
> > > It may not even be possible sometimes, because that function may be
> > > called before creating all of the struct acpi_device objects (like in
> > > the case of deferred enumeration).
> > >
> > > > > but happy to
> > > > > do it if you'd prefer it that way?
> > > >
> > > > I see, let Rafael decide then. I'm not pushing here.
> > >
> > > Well, it's a matter of correctness.
> >
> > Looking at your above comment it is indeed. Thanks for clarification!
>
> Well, actually, the struct device for the object passed to this
> function should be there already, because otherwise it wouldn't make
> sense to update the list.  So my comment above is not really
> applicable to this particular device and the function could take a
> struct acpi_device pointer argument.  Sorry for the confusion.
>
> > But should we have acpi_dev_*() namespace for this function if it takes handle?
>
> It takes a device object handle.
>
> Anyway, as per the above, it can take a struct acpi_device pointer
> argument in which case the "acpi_dev_" prefix should be fine.
>
> > For time being nothing better than following comes to my mind:
> >
> > __acpi_dev_flag_dependency_met() => __acpi_flag_device_dependency_met()
> > acpi_dev_flag_dependency_met() => acpi_flag_device_dependency_met()
>
> The above said, the name is somewhat confusing overall IMV.
>
> Something like acpi_dev_clear_dependencies() might be better.
>
> So lets make it something like
>
> void acpi_dev_clear_dependencies(struct acpi_device *supplier);

To be precise, there are two functions in the patch,
acpi_dev_flag_dependency_met() which invokes
acpi_walk_dep_device_list() and __acpi_dev_flag_dependency_met()
invoked by the latter as a callback.

Above I was talking about the first one.

The callback should still take a struct acpi_dep_data pointer argument
and I would call it acpi_scan_clear_dep() or similar.