Re: [PATCH 2/2] linux-yocto/3.4: add x32 configuration fragment and tuning hook

[Richard Purdie <richard.purdie@linuxfoundation.org>]

On Wed, 2012-09-12 at 08:58 -0400, Bruce Ashfield wrote:
> On Wed, Sep 12, 2012 at 8:25 AM, Richard Purdie
> <richard.purdie@linuxfoundation.org> wrote:
> > On Tue, 2012-09-11 at 09:07 -0700, Saul Wold wrote:
> >> On 09/11/2012 08:58 AM, Richard Purdie wrote:
> >> > On Tue, 2012-09-11 at 08:41 -0700, Saul Wold wrote:
> >> >> On 09/11/2012 08:39 AM, Bruce Ashfield wrote:
> >> >>> On 12-09-11 11:33 AM, Saul Wold wrote:
> >> >>>> On 09/11/2012 08:17 AM, Bruce Ashfield wrote:
> >> >>>>> When x32 is the tuning for a x86 MACHINE, the kernel should also have
> >> >>>>> CONFIG_X86_X32=y.
> >> >>>>>
> >> >>>>> This can be accomplished by adding the x32 configuraion fragment to the
> >> >>>>> KERNEL_FEATURES when x32 is the tuning for a given machine.
> >> >>>>>
> >> >>>>> cc: Saul Wold <sgw@linux.intel.com>
> >> >>>>> Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
> >> >>>>> ---
> >> >>>>> meta/recipes-kernel/linux/linux-yocto-rt_3.4.bb | 3 ++-
> >> >>>>> meta/recipes-kernel/linux/linux-yocto_3.4.bb | 3 ++-
> >> >>>>> 2 files changed, 4 insertions(+), 2 deletions(-)
> >> >>>>>
> >> >>>>> diff --git a/meta/recipes-kernel/linux/linux-yocto-rt_3.4.bb
> >> >>>>> b/meta/recipes-kernel/linux/linux-yocto-rt_3.4.bb
> >> >>>>> index 4fd3845..156fb93 100644
> >> >>>>> --- a/meta/recipes-kernel/linux/linux-yocto-rt_3.4.bb
> >> >>>>> +++ b/meta/recipes-kernel/linux/linux-yocto-rt_3.4.bb
> >> >>>>> @@ -10,7 +10,7 @@ KMETA = "meta"
> >> >>>>>
> >> >>>>> SRCREV_machine ?= "a35693b1287c0e50cdca33a1b95af0ff48b43cd0"
> >> >>>>> SRCREV_machine_qemuppc ?= "85a1190530cb5749f5f831670976b163438dc301"
> >> >>>>> -SRCREV_meta ?= "d9d5fc63d8b38705036e946ea77d971d95de11ad"
> >> >>>>> +SRCREV_meta ?= "e0374ce012e7e6fc8e5bb8b957addb0478950898"
> >> >>>>>
> >> >>>>> PR = "${INC_PR}.0"
> >> >>>>> PV = "${LINUX_VERSION}+git${SRCPV}"
> >> >>>>> @@ -27,3 +27,4 @@ KERNEL_FEATURES_append = " features/netfilter"
> >> >>>>> KERNEL_FEATURES_append = " features/taskstats"
> >> >>>>> KERNEL_FEATURES_append_qemux86 = " cfg/sound"
> >> >>>>> KERNEL_FEATURES_append_qemux86-64 = " cfg/sound"
> >> >>>>> +KERNEL_FEATURES_append_x32 = " cfg/x32"
> >> >>>>
> >> >>>> Scratch this bit and below, as I think I will use the other mechanism
> >> >>>> you talked about to go from a .conf file.
> >> >>>
> >> >>> Works for me. The meta change is staged and pushed out, I'll update this
> >> >>> patch to not have the KERNEL_FEATURES portion.
> >> >>>
> >> >> Thanks, see my other email to RP, since x32 is a feature that any x86-64
> >> >> machine might want to enable based on the DEFAULTTUNE it makes more
> >> >> sense to be in the machine config includes.
> >> >
> >> > No, it doesn't.
> >> >
> >> > What we need here is:
> >> >
> >> > -KERNEL_FEATURES_append = " features/taskstats"
> >> > +KERNEL_FEATURES_append = " features/taskstats ${@bb.utils.contains("TUNE_FEATURES", "mx32", " cfg/x32", "" ,d)}"
> >> >
> >> No, this would then only address the qemu machine, what about all the HW
> >> BSP that might want it, they would need to add this same line.  If I add
> >> the KERNEL_FEATURES_append to the arch-ia32.inc, conditional on mx32,
> >> then any x86-64 BSP can just enable that TUNE, isn't that the point of
> >> the machine config tuning?
> >
> > It is the point, however, the key part of this you're ignoring is that
> > the kernel fragment management only happens for linux-yocto. Only the
> > linux-yocto recipe supports the KERNEL_FEATURES mechanism.
> >
> > The arch-ia32.inc file and any machine config *cannot* depend on
> > linux-yocto.
> >
> > So this glue belongs in linux-yocto. I agree is suboptimal for people
> > not using it but such is life, there isn't any generic mechanism we can
> > place this into.
> 
> :)
> 
> linux-yocto it is, I've respun the patches to include the TUNE_FEATURES check
> and pushed them to poky-contrib zedd/x32:
> 
>  7aeb32 linux-yocto/3.4: add x32 configuration fragment
>  d34a34c linux-yocto*: append to KERNEL_FEATURES instead of assigning
> 
> I don't have a x32 build handy, but I assume Saul will take them for a spin.
> 
> If you want a full resend of the series, shout and I'll git send-email them out.

I've merged this, thanks. I'm assuming Saul will test the end result.

Cheers,

Richard