fix for kernelci errors?

fix for kernelci errors?

[Todd Kjos]

I'm looking through the android kernel build failures using the new UI
(still a little painful since it is missing some features compared to
the old UI, but getting better). I have questions on 3 of our error
cases:

1. android16-6.12 (gcc tools building *+allmodconfig):
/bin/sh: 1: gawk: not found
make[2]: *** [scripts/Makefile.vmlinux:47: modules.builtin.ranges] Error 127
make[2]: *** Deleting file 'modules.builtin.ranges'
make[1]: *** [/tmp/kci/linux/Makefile:1227: vmlinux] Error 2
make: *** [Makefile:242: __sub-make] Error 2

Is gawk missing on your build machines? How do we fix this?

2. android15-6.6 (clang tools building *+allmodconfig):
drivers/i3c/master/svc-i3c-master.c:1100:29: error: variable 'reg' is
uninitialized when used here [-Werror,-Wuninitialized]

It looks like the stable kernels only have 1 clang build configured
(x86_64_defconfig). Would it be possible to add an allmodconfig build
using clang so we can compare our allmodconfig failures with the
stable kernel?

3. android-mainline

Is there a reason that there are no allmodconfig builds of
mainline.master? It would help us to know if we introduced an issue if
we could see the equivalent upstream build. It seems like it would
also benefit upstream to see those issues early.

Re: fix for kernelci errors?

[Denys Fedoryshchenko]

On Tuesday, October 29, 2024 00:23 EET, Todd Kjos <tkjos@google.com> wrote:

> I'm looking through the android kernel build failures using the new UI
> (still a little painful since it is missing some features compared to
> the old UI, but getting better). I have questions on 3 of our error
> cases:
> 
> 1. android16-6.12 (gcc tools building *+allmodconfig):
> /bin/sh: 1: gawk: not found
> make[2]: *** [scripts/Makefile.vmlinux:47: modules.builtin.ranges] Error 127
> make[2]: *** Deleting file 'modules.builtin.ranges'
> make[1]: *** [/tmp/kci/linux/Makefile:1227: vmlinux] Error 2
> make: *** [Makefile:242: __sub-make] Error 2
> 
> Is gawk missing on your build machines? How do we fix this?
PR with fix is already in the testing, likely will be merged today and pushed into production
https://github.com/kernelci/kernelci-core/pull/2723

> 
> 2. android15-6.6 (clang tools building *+allmodconfig):
> drivers/i3c/master/svc-i3c-master.c:1100:29: error: variable 'reg' is
> uninitialized when used here [-Werror,-Wuninitialized]
> 
> It looks like the stable kernels only have 1 clang build configured
> (x86_64_defconfig). Would it be possible to add an allmodconfig build
> using clang so we can compare our allmodconfig failures with the
> stable kernel?
Do you mean stable-rc kernels?

> 
> 3. android-mainline
> 
> Is there a reason that there are no allmodconfig builds of
> mainline.master? It would help us to know if we introduced an issue if
> we could see the equivalent upstream build. It seems like it would
> also benefit upstream to see those issues early.
> 
allmodconfig is quite heavy build, and as legacy is still running, we are unable to use build capacity allocated for it, and Maestro-own build capacity is limited. If it is just mainline master - we can try to run it.

Re: fix for kernelci errors?

[Todd Kjos]

On Mon, Oct 28, 2024 at 9:10 PM Denys Fedoryshchenko
<nuclearcat@nuclearcat.com> wrote:
>
> On Tuesday, October 29, 2024 00:23 EET, Todd Kjos <tkjos@google.com> wrote:
>
> > I'm looking through the android kernel build failures using the new UI
> > (still a little painful since it is missing some features compared to
> > the old UI, but getting better). I have questions on 3 of our error
> > cases:
> >
> > 1. android16-6.12 (gcc tools building *+allmodconfig):
> > /bin/sh: 1: gawk: not found
> > make[2]: *** [scripts/Makefile.vmlinux:47: modules.builtin.ranges] Error 127
> > make[2]: *** Deleting file 'modules.builtin.ranges'
> > make[1]: *** [/tmp/kci/linux/Makefile:1227: vmlinux] Error 2
> > make: *** [Makefile:242: __sub-make] Error 2
> >
> > Is gawk missing on your build machines? How do we fix this?
> PR with fix is already in the testing, likely will be merged today and pushed into production
> https://github.com/kernelci/kernelci-core/pull/2723
>
> >
> > 2. android15-6.6 (clang tools building *+allmodconfig):
> > drivers/i3c/master/svc-i3c-master.c:1100:29: error: variable 'reg' is
> > uninitialized when used here [-Werror,-Wuninitialized]
> >
> > It looks like the stable kernels only have 1 clang build configured
> > (x86_64_defconfig). Would it be possible to add an allmodconfig build
> > using clang so we can compare our allmodconfig failures with the
> > stable kernel?
> Do you mean stable-rc kernels?

Yes, like "linux-6.6-y".
https://dashboard.kernelci.org/tree?origin=maestro&treeSearch=stable-rc&intervalInDays=14

>
> >
> > 3. android-mainline
> >
> > Is there a reason that there are no allmodconfig builds of
> > mainline.master? It would help us to know if we introduced an issue if
> > we could see the equivalent upstream build. It seems like it would
> > also benefit upstream to see those issues early.
> >
> allmodconfig is quite heavy build, and as legacy is still running, we are unable to use build capacity allocated for it, and Maestro-own build capacity is limited. If it is just mainline master - we can try to run it.

IMHO, allmodconfig is the best test to find build issues. Understood
that it is heavy. Yes, mainline master would be fine - and I think it
would be helpful for upstream rc validation too.