Re: [PATCH v2 0/7] Rename DTB overlay source files

[Andrew Davis]

On 11/14/22 10:06 AM, Andy Shevchenko wrote:
> On Sun, Nov 13, 2022 at 02:36:33PM +0200, Andy Shevchenko wrote:
> > On Sun, Nov 13, 2022 at 02:20:48PM +0200, Andy Shevchenko wrote:
> > > On Fri, Nov 11, 2022 at 03:05:20PM -0600, Andrew Davis wrote:
> > > > On 11/11/22 1:39 PM, Andy Shevchenko wrote:
> > > > > On Wed, Oct 26, 2022 at 09:11:21AM -0500, Rob Herring wrote:
> > > > > > On Mon, Oct 24, 2022 at 12:34 PM Andrew Davis <afd@xxxxxx> wrote:
> > > > > > > Hello all,
> > > > > > >
> > > > > > > This is a series based on my patch here[0]. As suggested by Rob
> > > > > > > I've resurrected Frank's patch and appended it to mine as a series.
> > > > > > >
> > > > > > > First patch here is my original patch, 3rd is Frank's patch but with
> > > > > > > the unittest changes pulled out into the 2nd patch. That was re-worked
> > > > > > > moving the source building macro into scripts/Makefile.lib.
> > > > > > >
> > > > > > > Patches 4, 5, and 6 are an attempt at renaming all the existing DTB
> > > > > > > overlays. Split out by platform so they could be taken by platform
> > > > > > > maintainers or if easier ACK'd here and taken all together.
> > > > > > >
> > > > > > > This should cover all the DTB overlays so we can remove the old .dts
> > > > > > > rule for overlays and make .dtso the only supported way, let me know
> > > > > > > if we want that this cycle and I can post that too.
> > > > > > >
> > > > > > > Thanks,
> > > > > > > Andrew
> > > > > > >
> > > > > > > Changes from v1[1]:
> > > > > > >    - Added patch to rename pi433 overlay.
> > > > > > >    - Cleaned wording on patch 4-6.
> > > > > > >    - Collected some ACKs
> > > > > > >
> > > > > > > [0] https://www.spinics.net/lists/kernel/msg4548509.html
> > > > > > > [1] https://www.spinics.net/lists/arm-kernel/msg1020165.html
> > > > > > >
> > > > > > > Andrew Davis (6):
> > > > > > >     kbuild: Allow DTB overlays to built from .dtso named source files
> > > > > > >     kbuild: Allow DTB overlays to built into .dtso.S files
> > > > > > >     arm64: dts: freescale: Rename DTB overlay source files from .dts to
> > > > > > >       .dtso
> > > > > > >     arm64: dts: renesas: Rename DTB overlay source files from .dts to
> > > > > > >       .dtso
> > > > > > >     arm64: dts: xilinx: Rename DTB overlay source files from .dts to .dtso
> > > > > > >     staging: pi433: overlay: Rename overlay source file from .dts to .dtso
> > > > > > >
> > > > > > > Frank Rowand (1):
> > > > > > >     of: overlay: rename overlay source files from .dts to .dtso
> > > > > >
> > > > > > I've applied patches 1-3 and 7. I'll send a PR for the branch to the
> > > > > > platform maintainers after a few days in linux-next.
> > > > >
> > > > > The patch
> > > > >
> > > > > commit 941214a512d8c80d47e720c17ec17e8539175e93
> > > > > Author: Andrew Davis <afd@xxxxxx>
> > > > > Date:   Mon Oct 24 12:34:29 2022 -0500
> > > > >
> > > > >       kbuild: Allow DTB overlays to built into .dtbo.S files
> > > > >
> > > > > broke the build reproducibility / no-op builds.
> > > > >
> > > > > Before:
> > > > >     2+ execution of `make` on non-changed tree did nothing
> > > > >
> > > > > Now:
> > > > >     Each run of `make` (even without a single bit changed) restarts vmlinux
> > > > >     rebuild.
> > > > >
> > > > > Please, revert or fix.
> > > >
> > > > I do not see this behavior. What config are you using?
> > > >
> > > > Not sure how this patch could be the root cause, it only adds
> > > > a build target/rule, but doesn't actually use it anywhere yet..
> > >
> > > For your reference I started with this one [1].
> > >
> > > When I bisected, I just answered with defaults on whatever `make` told me at
> > > the configuration stage.
> > >
> > > The actual `make` command I used:
> > >
> > > 	make O=/path/to/the/result W=1 C=1 CF=-D__CHECK_ENDIAN__ -j64
> > >
> > > But there is nothing that can affect the described issue.
> >
> > Actually, O= might affect which Makefile is used and how.
> > The C=, CF= are sparse flags, W= is just warning level.
>
> As far as I can tell right now it's the OF_UNITTEST on x86_64 that makes the
> above mentioned patch to be a culprit. Not sure if on ARM / ARM64 you can
> reproduce that. And it's really strange nobody reported this for a week+.
>
> Whatever, I'm open for the suggestions and material to test.

I think I found the issue, we forgot to add the new dtbo.S/o files
to the list of preserved intermediate targets, so Make was
removing them after build.

Sending the fix now.

Thanks,
Andrew

> > > [1]: https://p.defau.lt/?ZSOdGnNxF9v9AQtrfDo_KQ