Re: [PATCH] binder: let ANDROID_BINDER_IPC_32BIT be selectable on 32bit ARM

[Greg KH]

On Tue, Aug 22, 2017 at 08:14:10PM -0700, John Stultz wrote:
> On Tue, Aug 22, 2017 at 8:01 PM, Jisheng Zhang <jszhang@xxxxxxxxxxx> wrote:
> > On Tue, 22 Aug 2017 19:57:04 -0700 John Stultz wrote:
> >
> >> On Tue, Aug 22, 2017 at 7:56 PM, John Stultz <john.stultz@xxxxxxxxxx> wrote:
> >> > On Tue, Aug 22, 2017 at 7:34 PM, Jisheng Zhang <jszhang@xxxxxxxxxxx> wrote:
> >> >> On Tue, 22 Aug 2017 18:51:08 -0700 Greg KH wrote:
> >> >>
> >> >>> On Tue, Aug 08, 2017 at 07:03:05PM +0800, Jisheng Zhang wrote:
> >> >>> > As noted in commit d0bdff0db809 ("staging: Fix build issues with new
> >> >>> > binder API"), we can add back the choice for 32bit ARM "once a 64bit
> >> >>> > __get_user_asm_* implementation is merged." Commit e38361d032f1 ("ARM:
> >> >>> > 8091/2: add get_user() support for 8 byte types") has added the
> >> >>> > support, so it's time to let ANDROID_BINDER_IPC_32BIT be selectable on
> >> >>> > 32bit ARM
> >> >>>
> >> >>> Ok, but:
> >> >>>
> >> >>> >
> >> >>> > Signed-off-by: Jisheng Zhang <jszhang@xxxxxxxxxxx>
> >> >>> > ---
> >> >>> >  drivers/android/Kconfig | 2 +-
> >> >>> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >> >>> >
> >> >>> > diff --git a/drivers/android/Kconfig b/drivers/android/Kconfig
> >> >>> > index 832e885349b1..aca5dc30b97b 100644
> >> >>> > --- a/drivers/android/Kconfig
> >> >>> > +++ b/drivers/android/Kconfig
> >> >>> > @@ -32,7 +32,7 @@ config ANDROID_BINDER_DEVICES
> >> >>> >       therefore logically separated from the other devices.
> >> >>> >
> >> >>> >  config ANDROID_BINDER_IPC_32BIT
> >> >>> > -   bool
> >> >>> > +   bool "Use old (Android 4.4 and earlier) 32-bit binder API"
> >> >>> >     depends on !64BIT && ANDROID_BINDER_IPC
> >> >>>
> >> >>> You don't actually change the depends line :(
> >> >>>
> >> >>> Please fix up, and test it, and then resend.
> >> >>
> >> >> IHOM, the dependency is correct: 64bit platforms don't support
> >> >> ANDROID_BINDER_IPC_32BIT. What do you think?
> >> >
> >> > I think this indicates the commit message is unclear.
> >> >
> >> > Part of it is that the config is inverted from the description. The
> >> > patch doesn't enable the 32bit legacy binder ABI on 32bit systems, it
> >> > just allows the option to be unselected, so that the 64bit ABI will be
> >> > used on 32bit systems.
> >> >
> >> > Conceptually I don't have an objection to the change (though maybe try
> >> > to rework the commit message), but I don't have anything to actually
> >> > test it on right now, so I'm hesitant to ack it.
> >>
> >> It might also be good to add some detail as to the motivation for this
> >> change? What benefit does it bring to 32bit platforms to use the newer
> >> 64bit ABI?
> >>
> >
> > To be honest, the motivation is just to add one more choice for 32bit
> > platform and let the code be tested under 32bit platform. Maybe we
> > could then remove ANDROID_BINDER_IPC_32BIT and the related code after
> > some time?
> 
> I'm mixed. It would be nice to deprecate the old 32bit ABI, but binder
> is a real Linux kernel interface now, so we don't break compatibility
> (at least if it affects anyone - which may be questionable here - not
> sure there's many upstream 32bit platforms concerned with running
> legacy Android builds).  But just adding the extra option just means
> there's yet another configuration to test and to keep working. So you
> may want to articulate the benefits better to make this worth the
> effort of doing a full transition.

Future versions of Android "might" be changing if they support, or not,
all of the different combinations of 32/64 bit apis.  So maybe we should
wait until a real solid direction with regards to this is published
before worrying about dropping anything...

thanks,

greg k-h