Skip to contentSkip to conversation

module.bbclass: add HOSTCC for out-of-tree modules

PATCH In review Vyacheslav Yurkov 2 participants 7 messages Oct 1, 2020 1 message not in the archive Kbuild

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 module.bbclass: add HOSTCC for out-of-tree modules

From: Vyacheslav Yurkov <[email protected]>

Module build environment should be aware of native C and C++ compiler's environment otherwise kernel Makefile might silently fail for some checks. A particular example is CONFIG_STACK_VALIDATION when CONFIG_UNWINDER_ORC is used, Makefile tries unsuccessfully locate libelf, even though it's already available in recipe sysroot.

---
 meta/classes/module.bbclass | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
meta/classes/module.bbclass +1 -1
index c0dfa35061..8a691922c7 100644
@@ -1,6 +1,6 @@
 inherit module-base kernel-module-split pkgconfig
 
-EXTRA_OEMAKE += "KERNEL_SRC=${STAGING_KERNEL_DIR}"
+EXTRA_OEMAKE += "KERNEL_SRC=${STAGING_KERNEL_DIR} HOSTCC="${BUILD_CC} ${BUILD_CFLAGS} ${BUILD_LDFLAGS}" HOSTCPP="${BUILD_CPP}""
 
 MODULES_INSTALL_TARGET ?= "modules_install"
 MODULES_MODULE_SYMVERS_LOCATION ?= ""
Signature · 1 line
2.25.1
#1 lore raw
Quoted 19 lines · Vyacheslav Yurkov

From: Vyacheslav Yurkov <[email protected]>

Module build environment should be aware of native C and C++ compiler's environment otherwise kernel Makefile might silently fail for some checks. A particular example is CONFIG_STACK_VALIDATION when CONFIG_UNWINDER_ORC is used, Makefile tries unsuccessfully locate libelf, even though it's already available in recipe sysroot.

---
 meta/classes/module.bbclass | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
meta/classes/module.bbclass +0 -0
index c0dfa35061..8a691922c7 100644

meta/classes/module.bbclass line 4

Quoted 5 lines
meta/classes/module.bbclass +1 -1
@@ -1,6 +1,6 @@
 inherit module-base kernel-module-split pkgconfig
 
-EXTRA_OEMAKE += "KERNEL_SRC=${STAGING_KERNEL_DIR}"
+EXTRA_OEMAKE += "KERNEL_SRC=${STAGING_KERNEL_DIR} HOSTCC="${BUILD_CC} ${BUILD_CFLAGS} ${BUILD_LDFLAGS}" HOSTCPP="${BUILD_CPP}""
 

We already ran into this with the main kernel flags, and we don't want to duplicate them here.

Why aren't what we have in kernel.bbclass working in your scenario ?

Bruce

Quoted 8 lines
meta/classes/module.bbclass +0 -1
 MODULES_INSTALL_TARGET ?= "modules_install"
 MODULES_MODULE_SYMVERS_LOCATION ?= ""
--

2.25.1

Signature · 3 lines
- Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end
- "Use the force Harry" - Gandalf, Star Trek II
#2 lore raw
Quoted 7 lines · Bruce Ashfield

We already ran into this with the main kernel flags, and we don't want to duplicate them here.

Why aren't what we have in kernel.bbclass working in your scenario ?

Bruce

Because kernel.bbclass is not pulled in into the environment of out-of-tree module recipe. Should it be? This can easily be checked when you do 'bitbake hello-mod -c devshell' and examine $EXTRA_OEMAKE variable.

Vyacheslav

#3 lore raw
Quoted 5 lines · Vyacheslav Yurkov
Quoted 3 lines · Bruce Ashfield

We already ran into this with the main kernel flags, and we don't want to duplicate them here.

Quoted 8 lines
Quoted 4 lines

Why aren't what we have in kernel.bbclass working in your scenario ?

Bruce

Because kernel.bbclass is not pulled in into the environment of out-of-tree module recipe. Should it be?
This can easily be checked when you do 'bitbake hello-mod -c devshell' and examine $EXTRA_OEMAKE variable.

Yes, it should be.

We build out of tree modules in each autobuilder test run, so it is functional for the most part. I'd be interested to see the reproducing kernel module and build steps, so they can be built into the tests to catch this.

But we absolutely do not want to duplicate those flags into this class.

So with a reproducer, we can perhaps shuffle things around and make sure that both this new case and the old ones continue to work.

Bruce

Quoted 2 lines

Vyacheslav

Signature · 3 lines
- Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end
- "Use the force Harry" - Gandalf, Star Trek II
#4 lore raw
Quoted 20 lines · Bruce Ashfield
Quoted 4 lines · Vyacheslav Yurkov
Quoted 3 lines · Bruce Ashfield

We already ran into this with the main kernel flags, and we don't want to duplicate them here.

Quoted 8 lines
Quoted 4 lines

Why aren't what we have in kernel.bbclass working in your scenario ?

Bruce

Because kernel.bbclass is not pulled in into the environment of out-of-tree module recipe. Should it be?
This can easily be checked when you do 'bitbake hello-mod -c devshell' and examine $EXTRA_OEMAKE variable.

Yes, it should be.

We build out of tree modules in each autobuilder test run, so it is functional for the most part. I'd be interested to see the reproducing kernel module and build steps, so they can be built into the tests to catch this.

Quoted 7 lines

But we absolutely do not want to duplicate those flags into this class.

So with a reproducer, we can perhaps shuffle things around and make sure that both this new case and the old ones continue to work.

Bruce

That's interesting to know. It has been easy to reproduce in 4.x kernel, but with 5.x in master it fails silently. So here's what you do with 5.x and meta-skeleton/recipes-kernel/hello-mod/hello-mod_0.1.bb module

1. Set machine to any x86, e.g. qemux86-64
2. Enable CONFIG_UNWINDER_ORC in the kernel configuration
3. Modify kernel Makefile as follows:
Vyacheslav, [01.10.20 17:01]
index 8b1d5b9017a5..e35da9f9bb75 100644

Makefile +0 -0
@@ -1058,7 +1058,9 @@ ifdef CONFIG_STACK_VALIDATION

      echo "int main() {}" | $(HOSTCC) -xc -o /dev/null $(HOST_LIBELF_LIBS) -,1,0) ifeq ($(has_libelf),1)
     objtool_target := tools/objtool FORCE
+    $(info "Validation enabled")
   else
+    $(info "Check failed, disabling stack validation") SKIP_STACK_VALIDATION := 1
     export SKIP_STACK_VALIDATION
   endif

4. If HOSTCC is set properly, then you should see "Validation enabled" in log.do_compile of hello-mod recipe.

Any hints where autobuilder tests are? Perhaps I could take a look and find a way to cover this...

Vyacheslav

#5 lore raw
Quoted 52 lines · Vyacheslav Yurkov
Quoted 20 lines · Bruce Ashfield
Quoted 4 lines · Vyacheslav Yurkov
Quoted 3 lines · Bruce Ashfield

We already ran into this with the main kernel flags, and we don't want to duplicate them here.

Quoted 8 lines
Quoted 4 lines

Why aren't what we have in kernel.bbclass working in your scenario ?

Bruce

Because kernel.bbclass is not pulled in into the environment of out-of-tree module recipe. Should it be?
This can easily be checked when you do 'bitbake hello-mod -c devshell' and examine $EXTRA_OEMAKE variable.

Yes, it should be.

We build out of tree modules in each autobuilder test run, so it is functional for the most part. I'd be interested to see the reproducing kernel module and build steps, so they can be built into the tests to catch this.

Quoted 7 lines

But we absolutely do not want to duplicate those flags into this class.

So with a reproducer, we can perhaps shuffle things around and make sure that both this new case and the old ones continue to work.

Bruce

That's interesting to know. It has been easy to reproduce in 4.x kernel, but with 5.x in master it fails silently. So here's what you do with 5.x and meta-skeleton/recipes-kernel/hello-mod/hello-mod_0.1.bb module

1. Set machine to any x86, e.g. qemux86-64
2. Enable CONFIG_UNWINDER_ORC in the kernel configuration
3. Modify kernel Makefile as follows:
Vyacheslav, [01.10.20 17:01]
index 8b1d5b9017a5..e35da9f9bb75 100644

Makefile +2 -0
@@ -1058,7 +1058,9 @@ ifdef CONFIG_STACK_VALIDATION
       echo "int main() {}" | $(HOSTCC) -xc -o /dev/null
 $(HOST_LIBELF_LIBS) -,1,0)
    ifeq ($(has_libelf),1)
      objtool_target := tools/objtool FORCE
+    $(info "Validation enabled")
    else
+    $(info "Check failed, disabling stack validation")
      SKIP_STACK_VALIDATION := 1
      export SKIP_STACK_VALIDATION

endif

Quoted 3 lines

4. If HOSTCC is set properly, then you should see "Validation enabled" in log.do_compile of hello-mod recipe.

Ah yes.

As in my previous email, those warnings have been around for a bit and we've just never had a reason to fix them, and in other scenarios, I've just been explicitly enabling/adding the variables where I need them. Since (for instance), we can't always assume that HOSTCC is available on-target, the architecture support, or HOSTCC is the same as CC, etc, so we want to cover the combinations properly.

The warnings aren't only seen on module build, just a simple 'make scripts prepare' will show them, so that's enough of a test to see if they are or aren't picking up the compiler. I wouldn't make these a failure, just a warning, as they currently are.

Perhaps this is the motivation to do that fixup :D

Bruce

Quoted 4 lines

Any hints where autobuilder tests are? Perhaps I could take a look and find a way to cover this...

Vyacheslav

Signature · 3 lines
- Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end
- "Use the force Harry" - Gandalf, Star Trek II

A message that is not in the archive (sent off-list or not yet received): [email protected]

#7 lore raw
Quoted 16 lines · Bruce Ashfield via lists.openembedded.org
Quoted 5 lines · Vyacheslav Yurkov
Quoted 3 lines · Bruce Ashfield

We already ran into this with the main kernel flags, and we don't want to duplicate them here.

Quoted 8 lines
Quoted 4 lines

Why aren't what we have in kernel.bbclass working in your scenario ?

Bruce

Because kernel.bbclass is not pulled in into the environment of out-of-tree module recipe. Should it be?
This can easily be checked when you do 'bitbake hello-mod -c devshell' and examine $EXTRA_OEMAKE variable.

Quoted 6 lines

Yes, it should be.

We build out of tree modules in each autobuilder test run, so it is functional for the most part. I'd be interested to see the reproducing kernel module and build steps, so they can be built into the tests to catch this.

And by that, I mean more than the warnings about ORC, etc. Since those have been popping up for some time.

I'm not saying we shouldn't be fixing those, I'm just wondering if this is an actual build failure, versus the ones we know about.

(and either way, it doesn't change the way we fix it, just that we can test it better).

Bruce

Quoted 20 lines

But we absolutely do not want to duplicate those flags into this class.

So with a reproducer, we can perhaps shuffle things around and make sure that both this new case and the old ones continue to work.

Bruce

Quoted 2 lines

Vyacheslav

--
- Thou shalt not follow the NULL pointer, for chaos and madness await thee at its end
- "Use the force Harry" - Gandalf, Star Trek II

Signature · 3 lines
- Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end
- "Use the force Harry" - Gandalf, Star Trek II