Bug 15831

Summary:	mesa fails to build when rm_work is enabled
Product:	[Build System, Metadata & Runtime] OE-Core	Reporter:	Khem Raj <raj.khem>
Component:	graphics	Assignee:	Khem Raj <raj.khem>
Status:	RESOLVED WORKSFORME	QA Contact:
Severity:	normal
Priority:	Medium+	CC:	dbaryshkov, meta.mr.watcher, meta.watcher, randy.macleod, ross.burton
Version:	unspecified
Target Milestone:	5.3 M1
Hardware:	x86
OS:	Multiple
Whiteboard:
OS type for building Yocto:	---	Type of Regression:	---
Verified:		Documentation change:	No (bug/feature does not impact docs)

Description Khem Raj 2025-04-14 20:09:23 UTC

mesa 25 upgrade is newly building clc pieces for which it use mesa_clc tool to do the conversion to SPIRV backend and utilizes clang to do the work underneath. Clang argument are passed on commandline after -- separator.

This works ok when rm_work is not enabled however when INHERIT:append = " rm_work" in local.conf and use MACHINE = "qemux86-64"

Then build mesa

bitbake -ccleansstate mesa-native mesa
bitbake mesa

It ends up in errors e.g.

/mnt/b/yoe/master/build/tmp/work/core2-64-yoe-linux/mesa/25.0.2/recipe-sysroot-native/usr/bin/mesa_clc ../mesa-25.0.2/src/intel/shaders/libintel_shaders.h ../mesa-25.0.2/src/intel/shaders/generate.cl ../mesa-25.0.2/src/intel/shaders/generate_draws.cl ../mesa-25.0.2/src/intel/shaders/generate_draws_iris.cl ../mesa-25.0.2/src/intel/shaders/query_copy.cl ../mesa-25.0.2/src/intel/shaders/util.cl -o src/intel/shaders/intel_gfx90_shaders.spv -- -DNDEBUG=1 -DGFX_VERx10=90 -I/mnt/b/yoe/master/build/tmp/work/core2-64-yoe-linux/mesa/25.0.2/mesa-25.0.2/src/intel/shaders/. -I/mnt/b/yoe/master/build/tmp/work/core2-64-yoe-linux/mesa/25.0.2/mesa-25.0.2/src/compiler/libcl -I/mnt/b/yoe/master/build/tmp/work/core2-64-yoe-linux/mesa/25.0.2/mesa-25.0.2/src -I/mnt/b/yoe/master/build/tmp/work/core2-64-yoe-linux/mesa/25.0.2/mesa-25.0.2/src/intel -I/mnt/b/yoe/master/build/tmp/work/core2-64-yoe-linux/mesa/25.0.2/build/src/intel -I/mnt/b/yoe/master/build/tmp/work/core2-64-yoe-linux/mesa/25.0.2/mesa-25.0.2/src/intel/genxml
<built-in>:1:10: fatal error: 'opencl-c-base.h' file not found
Error executing LLVM compilation action.

Comment 1 Khem Raj 2025-04-14 20:34:19 UTC

I manually appended --isysroot=/mnt/b/yoe/master/build/tmp/work/core2-64-yoe-linux/mesa/25.0│
│.2/recipe-sysroot

this gets into next error

Couldn't create Clang invocation.

Comment 2 Ross Burton 2025-04-17 15:03:52 UTC

Between this and the meta-clang breakage our llvm/clang recipes are pretty broken.

I'm close to advocating we revert the llvm changes in core... Khem, Dmitry: can you work out how to resolve this?

Comment 3 Dmitry Baryshkov 2025-04-18 01:03:03 UTC

With the following I can successfully build nodistro + mesa doing the build package-by-package, so that rm_work can kick in inbetween.

- current github/master
- https://lore.kernel.org/openembedded-core/20250418-mesa-25-fixes-v2-0-e110ccfe291f@oss.qualcomm.com/
- https://github.com/kraj/meta-clang/pull/1085

I definitely don't think that reverting llvm / mesa in acceptable solution: this means shipping a very outdated and not supported component.

Comment 4 Khem Raj 2025-04-18 01:22:14 UTC

(In reply to Ross Burton from comment #2)
> Between this and the meta-clang breakage our llvm/clang recipes are pretty
> broken.
> 
> I'm close to advocating we revert the llvm changes in core... Khem, Dmitry:
> can you work out how to resolve this?

I have pushed a new branch to contrib tree which contains a clang merge into core.

https://web.git.yoctoproject.org/poky-contrib/log/?h=kraj/clang-merge-2

 Please look into it and lets make it work. I can build mesa with this and also full featured clang compiler. I will start adapting meta-clang to remove clang and make adjustments as needed to work with clang coming from core.

I think this will straighten the path forward.

Comment 5 Khem Raj 2025-04-18 01:52:47 UTC

(In reply to Dmitry Baryshkov from comment #3)
> With the following I can successfully build nodistro + mesa doing the build
> package-by-package, so that rm_work can kick in inbetween.
> 
> - current github/master
> -
> https://lore.kernel.org/openembedded-core/20250418-mesa-25-fixes-v2-0-
> e110ccfe291f@oss.qualcomm.com/

I have staged these patches into kraj/clang-merge-2 branch

> - https://github.com/kraj/meta-clang/pull/1085
> 

I think this wont be needed.

> I definitely don't think that reverting llvm / mesa in acceptable solution:
> this means shipping a very outdated and not supported component.

yes, I am hopeful that we don't have to take this step.

Comment 6 Randy MacLeod 2025-07-17 18:49:02 UTC

From IRC, #yocto

Randy: khem: RP thinks that this issue has been resolved, do you agree?
Khem:  I think yes,I have not seen this failures on CI systems for some weeks now