[OE-core] [PATCH 3/3] powerpc: define TUNE_PKGARCH for powerpc/powerpc-nf

[Chris Larson]

On Thu, Mar 29, 2012 at 9:40 AM, Mark Hatle <mark.hatle at windriver.com> wrote:
> (This is going to be a bit rambling I'm afraid.. I'm working my way through
> investigation and a possible solution..  so be sure to look all the way to
> the end..)
>
>
> On 3/29/12 10:58 AM, McClintock Matthew-B29882 wrote:
>>
>> On Thu, Mar 29, 2012 at 10:38 AM, Mark Hatle<mark.hatle at windriver.com>
>>  wrote:
>>>
>>> On 3/28/12 11:54 PM, Chris Larson wrote:
>>>>
>>>>
>>>> On Wed, Mar 28, 2012 at 9:47 PM, McClintock Matthew-B29882
>>>> <B29882 at freescale.com>    wrote:
>>>>>
>>>>>
>>>>> On Tue, Mar 27, 2012 at 5:16 PM, Chris Larson<clarson at kergoth.com>
>>>>>  wrote:
>>>>>>>
>>>>>>>
>>>>>>> If you can explain why the override isn't overriding the default
>>>>>>> TUNE_PKGARCH (and it's intentional and not a bug), and we can
>>>>>>> consistently
>>>>>>> modify all of the elements... I'm happy to accept the changes to all
>>>>>>> of
>>>>>>> the
>>>>>>> tunings.
>>>>>>
>>>>>>
>>>>>>
>>>>>> See above. It's not an override. And plenty of files already specify
>>>>>> TUNE_PKGARCH_tune-<tune>, so I don't see how it'd be inconsistent to
>>>>>> do so for the defaults, personally.
>>>>>
>>>>>
>>>>>
>>>>> If no one else has complained so far it makes me believe they are not
>>>>> missing any TUNE_PKGARCH_tune-<tune>    then.
>>>>
>>>>
>>>>
>>>> I don't understand the point you're attempting to make here.
>>>
>>>
>>>
>>> Just an FYI -- I've not forgotten about this.. I'm going to look into
>>> what
>>> is currently implemented and figure out how to get it to be consistent..
>>> things have definitely changed since the initial implementation, and
>>> things
>>> are no longer consistent.
>>>
>>> Hopefully I'll have an update later today.
>>
>>
>> Just for some background:
>>
>> I added the TUNEPKG_ARCH = TUNE_PKGARCH_tune-$DEFAULTTUNE bit because
>> it seemed to make sense and it was missing from this list:
>>
>> meta/conf/bitbake.conf:TUNE_FEATURES ??=
>> "${TUNE_FEATURES_tune-${DEFAULTTUNE}}"
>> meta/conf/bitbake.conf:PACKAGE_EXTRA_ARCHS ??=
>> "${PACKAGE_EXTRA_ARCHS_tune-${DEFAULTTUNE}}"
>>
>> Also it seemed silly (at the time admittedly) to set TUNE_PKGARCH
>> based off a value in TUNE_FEATURES and using bb functions to do this
>> when the way I went with was simply setting a value. I did miss that
>> the default case no longer works properly unless you have a
>> TUNE_PKGARCH_tune-{powerpc,powerpc-nf,powerpc64} =
>> {powerpc,powerpc-nf,powerpc64} set in the default arch files.
>>
>> Anyways, I'm not stuck on one particular method at the time I was
>> trying to get ppce5500 multilib working properly.
>
>
> Ok, that makes a bit more sense.. What I see right now is:
>
> TUNE_PKGARCH ??= "${TUNE_PKGARCH_tune-${DEFAULTTUNE}}"
>
> (This was originally designed that TUNE_PKGARCH was defined based on the
> overrides... note, not only has the above affected that, but it's not
> consistent in the other architectures either.. so it was wrong, and is still
> likely wrong...)
>
> Arm defines it as:
>
> TUNE_PKGARCH = "${@d.getVar('ARMPKGARCH',
> True)}${ARMPKGSFX_THUMB}${ARMPKGSFX_DSP}${ARMPKGSFX_EABI}${ARMPKGSFX_ENDIAN}${ARMPKGSFX_FPU}"
>
> That way the ARMPKGARCH is the base of the configuration, while the DSP,
> ABI, Endian and FPU are added dynamically based on the feature flags (and
> associated settings)..
>
> IA defines is as:
> TUNE_PKGARCH ?= "${@bb.utils.contains("TUNE_FEATURES", "m32", "x86",
> "x86_64", d)}"
> TUNE_PKGARCH .= "${@bb.utils.contains("TUNE_FEATURES", "mx32", "_x32", "",
> d)}"
>
> And then some (c3, core2, i586) override that, but not all of them.. again
> this is likely wrong as well..
>
> MIPS defines it as:
>
> TUNE_PKGARCH ?= "${TUNE_ARCH}${MIPSPKGSFX_FPU}${MIPSPKGSFX_ABI}"
>
> TUNE_ARCH as:
> TUNE_ARCH = "mips${MIPSPKGSFX_BYTE}${MIPSPKGSFX_ENDIAN}
>
> and finally powerpc:
>
> There is NO TUNE_PKGARCH defined.. but there is an append:
>
> TUNE_PKGARCH_append = "${PPCPKGSFX_FPU}"
>
> And each of the tunings:
>
> tune-ppc603e.inc:TUNE_PKGARCH_tune-ppc603e = "ppc603e"
> tune-ppce300c2.inc:TUNE_PKGARCH_tune-ppce300c2 = "ppce300c2"
> tune-ppce500.inc:TUNE_PKGARCH_tune-ppce500 = "ppce500"
> tune-ppce500mc.inc:TUNE_PKGARCH_tune-ppce500mc = "ppce500mc"
> tune-ppce500v2.inc:TUNE_PKGARCH_tune-ppce500v2 = "ppce500v2"
> tune-ppce5500.inc:TUNE_PKGARCH_tune-ppce5500 = "ppce5500"
> tune-ppce5500.inc:TUNE_PKGARCH_tune-ppc64e5500 = "ppc64e5500"
>
>
> ....  so somehow we need to figure out a more consistent approach.
>  Personally I prefer the ARM/MIPS style approach.  Build up the format using
> a consistent set of features/arguments.

That seems reasonable to me. I personally don't care much about which
approach is used as long as we're consistent about it :)

> The one thing I'm still struggling with though is what is the final purpose
> of these variables.  I may be able to make a better determination as to the
> right values if I truly understood the difference.  Below is my attempt at
> understanding this... (let me know if this is correct or if you have
> corrections)
>
> DEFAULTTUNE - The default tuning for a given machine or build
> [multilib build uses overrides to change the DEFAULTTUNE setting], tune
> values are defined by the conf/machine/include/* items
>
> TUNE_FEATURES_tune-<tune> - Should not be manually modified (i.e. not in
> local.conf, only specified by a tune file), but inherited via the
> DEFAULTTUNE value -- this defines the set of tuning variables that a given
> tuning uses to setup cflags, pkgarch, abi, etc..
>
> TUNE_CCARGS - Setup the cflags based on the TUNE_FEATURES settings.  These
> should be additive when defined using "+=".  All items in this list should
> be dynamic! i.e. ${@bb.utils.contains("TUNE_FEATURES", "feature", "cflag",
> "!cflag", d)}
>
> AVAILTUNES - This is a list of all available tunes within the machine
> configuration.  This does not mean that all available tunes are compatible
> with that machine configuration, or with each other.  The DEFAULTTUNE must
> be in this list.  Each tuning file can add to this list using "+=".
>
> PACKAGE_EXTRA_ARCHS_tune-<tune> - List of all compatible package arches to a
> given tune.
>
> TUNE_ARCH - GNU canonical arch for the architecture.  I.e. arm, armeb, mips,
> mips64, etc.
>
> TUNE_PKGARCH - The package architecture listed for the purpose of package
> feeds.  These are the names used by the PACKAGE_EXTRA_ARCHS for
> compatibility reasons.
>
> TARGET_FPU - The FPU setting for a given architecture, hard (generate
> floating point instructions), soft (generate internal gcc calls), "other"
> architecture specific floating point.  This is synchronized with the
> compiler and other toolchain items.  This should be dynamically configured
> in the same way the TUNE_CCARGS is.
>
> BASE_LIB_tune-<tune> - The "/lib" location for a specific ABI.

That's my understanding of those as well, if someone disagrees, it'd
be very interesting to know that. Further, it'd be nice to document
these somewhere.. either documentation.conf, though they're intended
for the bsp maintainers not the average user, or somewhere else
appropriate.

> So how about this for a possible solution:
>
> bitbake.conf:
> Keep - TUNE_FEATURES ??= "${TUNE_FEATURES_tune-${DEFAULTTUNE}}"
>
> Remove - TUNE_PKGARCH
>
> In the tuning files, specifically the architectural tuning files:
>
> Require a default setting of TUNE_PKGARCH.  The default setting MIGHT be
> specific to an arch (as in the ARM case above), or could be
> TUNE_PKGARCH = ${TUNE_PKGARCH_tune-${DEFAULTTUNE}} for something like
> PowerPC where it may be easier to specific the exact package arch.
>
> Each architecture file needs a small README to indicate how someone is to
> add a new tuning to that architecture.
>
> ----
>
> Does the above seem reasonable?  If so I'll try to get started on that...

That seems reasonable to me, for what that's worth. I could see
simplifying the way the powerpc stuff works in the future, but at
least we'd have something functional for now as well. It keeps primary
responsibility for TUNE_PKGARCH in arch-powerpc*.inc, which is as it
should be -- the arch knows best how to manage it. I'm curious to hear
Matthew's perspective on this.
-- 
Christopher Larson