[OE-core] [PATCH 1/2] kernel-yocto: ensure sccs variable is set when using KBUILD_DEFCONFIG

[Bruce Ashfield]

On 2017-11-29 1:24 PM, Saul Wold wrote:
> On Wed, 2017-11-29 at 13:13 -0500, Bruce Ashfield wrote:
>> On 11/29/2017 12:52 PM, Saul Wold wrote:
>>> On Wed, 2017-11-29 at 11:56 -0500, Bruce Ashfield wrote:
>>>> On 11/29/2017 11:30 AM, Saul Wold wrote:
>>>>> On Wed, 2017-11-29 at 09:23 -0500, Bruce Ashfield wrote:
>>>>>> On 11/28/2017 10:28 PM, Saul Wold wrote:
>>>>>>> When using KBUILD_DEFCONFIG, $sccs should be set to the
>>>>>>> $WORKDIR/defconfig
>>>>>>> regardless if it compares or is copied. Otherwise $sccs is
>>>>>>> not
>>>>>>> set
>>>>>>> and the
>>>>>>> defconfig is not found correctly.
>>>>>>
>>>>>> Actually, looking at this more today, and this morning in my
>>>>>> testing.
>>>>>> This shouldn't be necessary .. it doesn't hurt anything
>>>>>> (well,
>>>>>> actually
>>>>>> it could end up with two defconfigs in the variable, but that
>>>>>> also
>>>>>> should be ok).
>>>>>>
>>>>>
>>>>> Ok, I understand, it's in find_sccs() where if you have
>>>>> "defconfig"
>>>>> in
>>>>> the SRC_URI then with my change you could end up with two
>>>>> defconfigs.
>>>>>
>>>>> The problem I saw was if one just sets KBUILD_DEFCONFIG and
>>>>> does
>>>>> not
>>>>> set any config info on the SRC_URI then it's possible for $sccs
>>>>> to
>>>>> be
>>>>> empty, which was bad.
>>>>
>>>> I took a look at the conditions again, and I can't see that
>>>> path. But that doesn't mean it isn't there, is this a
>>>> configuration
>>>> that I can build and see myself ?
>>>>
>>>
>>> Set KBUILD_DEFCONFIG to some kernel defconfig, ensure SRC_URI does
>>> not
>>> have any defconfig, .scc or .cfg files.  Build once that will
>>> populate
>>> the $WORKDIR/defconfig with vai the else path of the first if and
>>> the
>>> cp path in the do_kernel_metadata() code.  Build a second time and
>>> the
>>> $WORKDIR/defconfig exists now and the first part of the if with the
>>> cmp
>>> will occur and the defconfig will not be found because $sccs does
>>> not
>>> get set in the original code when the cmp is equal.
>>
>> That's the ticket .. I didn't think of a 2nd pass through the
>> build.
>>
>> So with that extra assignment + a duplicate removal, it should
>> be safe for all the paths.
>>
>> Did you want to send a v2, or did you want me to make that tweak?
>>
> If you have the tweak in mind already, might be better for you to make
> it.
> 

One more question, since I'm still not able to trigger the issue
here.

When you are getting the failure mode, is the:

   bbwarn "defconfig detected in WORKDIR ..." being triggered ?

I actually think it would be better if I flip that warning around
and always overwrite what is in WORKDIR if someone is using the
KBUILD_DEFCONFIG functionality.

But for now, I will just remove the potential duplicate defconfig
and ponder that change in functionality more.

Bruce

> Thanks
>   Sau!
> 
>> Either way, I'll continue testing here.
>>
>> Bruce
>>
>>>
>>> Sau!
>>>
>>>>>
>>>>> Maybe a cleaning of multiple "defconfig" entries on $sccs is
>>>>> needed?
>>>>>
>>>>
>>>> Yah, regardless of my above statement, it certainly wouldn't hurt
>>>> and would be a good safeguard, since two defconfigs would trigger
>>>> a M x N configuration audit of options (where M and N could be
>>>> in the thousands) .. and hence, take a long time.
>>>>
>>>> Bruce
>>>>
>>>>> Sau!
>>>>>
>>>>>
>>>>>>>
>>>>>>> Part of
>>>>>>> [YOCTO #12162]
>>>>>>>
>>>>>>> Signed-off-by: Saul Wold <sgw at linux.intel.com>
>>>>>>> ---
>>>>>>>      meta/classes/kernel-yocto.bbclass | 2 +-
>>>>>>>      1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>>>
>>>>>>> diff --git a/meta/classes/kernel-yocto.bbclass
>>>>>>> b/meta/classes/kernel-yocto.bbclass
>>>>>>> index 1d447951c49..98ec78fb768 100644
>>>>>>> --- a/meta/classes/kernel-yocto.bbclass
>>>>>>> +++ b/meta/classes/kernel-yocto.bbclass
>>>>>>> @@ -110,8 +110,8 @@ do_kernel_metadata() {
>>>>>>>      				fi
>>>>>>>      			else
>>>>>>>      				cp -f
>>>>>>> ${S}/arch/${ARCH}/configs/${KBUILD_DEFCONFIG}
>>>>>>> ${WORKDIR}/defconfig
>>>>>>> -				sccs="${WORKDIR}/defconfig
>>>>>>> "
>>>>>>>      			fi
>>>>>>> +			sccs="${WORKDIR}/defconfig"
>>>>>>
>>>>>> The test that was protecting this assignment is:
>>>>>>
>>>>>>        if [ -f "${WORKDIR}/defconfig" ]; then
>>>>>>
>>>>>> and then:
>>>>>>
>>>>>>       cmp "${WORKDIR}/defconfig"
>>>>>> "${S}/arch/${ARCH}/configs/${KBUILD_DEFCONFIG}"
>>>>>>
>>>>>> The only way that a defconfig can be in ${WORKDIR}/defconfig
>>>>>> by
>>>>>> the time this runs, is if the fetcher puts it there. Which
>>>>>> means
>>>>>> it is on the SRC_URI and comes from the recipe writer's
>>>>>> layer.
>>>>>>
>>>>>> There is existing code that already picks this up and adds it
>>>>>> to the configuration queue:
>>>>>>
>>>>>>             sccs="$sccs ${@" ".join(find_sccs(d))}"
>>>>>>
>>>>>> So that defconfig, is already going to be picked up directly
>>>>>> from the SRC_URI.
>>>>>>
>>>>>> Bruce
>>>>>>
>>>>>>
>>>>>>>      		else
>>>>>>>      			bbfatal "A KBUILD_DEFCONFIG
>>>>>>> '${KBUILD_DEFCONFIG}' was specified, but not present in the
>>>>>>> source
>>>>>>> tree"
>>>>>>>      		fi
>>>>>>>
>>>>>>
>>>>>>
>>>>
>>>>
>>
>>