linux 5.3 breaks building glibc for riscv64

[Matthias Klose]

seen in
https://launchpad.net/ubuntu/+source/cross-toolchain-base-ports/35ubuntu1/+build/17748279

/<<PKGBUILDDIR>>/glibc-2.30/debian/include/linux/sched.h: Assembler messages:
/<<PKGBUILDDIR>>/glibc-2.30/debian/include/linux/sched.h:39: Error: unrecognized opcode `struct clone_args{'
/<<PKGBUILDDIR>>/glibc-2.30/debian/include/linux/sched.h:40: Error: unrecognized opcode `__aligned_u64 flags'
/<<PKGBUILDDIR>>/glibc-2.30/debian/include/linux/sched.h:41: Error: unrecognized opcode `__aligned_u64 pidfd'
/<<PKGBUILDDIR>>/glibc-2.30/debian/include/linux/sched.h:42: Error: unrecognized opcode `__aligned_u64 child_tid'
/<<PKGBUILDDIR>>/glibc-2.30/debian/include/linux/sched.h:43: Error: unrecognized opcode `__aligned_u64 parent_tid'
/<<PKGBUILDDIR>>/glibc-2.30/debian/include/linux/sched.h:44: Error: unrecognized opcode `__aligned_u64 exit_signal'
/<<PKGBUILDDIR>>/glibc-2.30/debian/include/linux/sched.h:45: Error: unrecognized opcode `__aligned_u64 stack'
/<<PKGBUILDDIR>>/glibc-2.30/debian/include/linux/sched.h:46: Error: unrecognized opcode `__aligned_u64 stack_size'
/<<PKGBUILDDIR>>/glibc-2.30/debian/include/linux/sched.h:47: Error: unrecognized opcode `__aligned_u64 tls'
/<<PKGBUILDDIR>>/glibc-2.30/debian/include/linux/sched.h:48: Error: junk at end of line, first unrecognized character is `}'
make[4]: *** [/<<PKGBUILDDIR>>/glibc-2.30/build-tree/riscv64-libc/sysd-rules:57: /<<PKGBUILDDIR>>/glibc-2.30/build-tree/riscv64-libc/posix/vfork.os] Error 1

[Seth Forshee (sforshee)]

Hmm, perhaps clone_args needs to be within a #ifndef __ASSEMBLY__ block ...

[Dimitri John Ledkov (xnox)]

but it doesn't break on any other arch.

[Seth Forshee (sforshee)]

On Wed, Sep 11, 2019 at 03:25:06PM -0000, Dimitri John Ledkov wrote:
> but it doesn't break on any other arch.

Yes because risc-v is the only arch which includes that file into an
assembly source file.

I'm building a test build which adds the __ASSEMBLY__ guards around the
struct for testing, should be ready soon.

[Seth Forshee (sforshee)]

Here's the test build:

https://people.canonical.com/~sforshee/lp1843458/

Please let me know whether or not this fixes the issue.

[Matthias Klose (doko)]

can't test, because the provided linux-source package is empty.

[Seth Forshee (sforshee)]

Sorry, forgot that we generate an empty linux source package for test builds. I replaced that with another build with that "feature" turned off, in the same location.

[Matthias Klose (doko)]

this fixes the issue. Please point to a commit, or a patch fixing the issue.

[Seth Forshee (sforshee)]

The patch is in the same directory as the build. I've applied it for eoan, it will be in the next upload. Will also send the patch upstream.

https://git.launchpad.net/~ubuntu-kernel/ubuntu/+source/linux/+git/eoan/commit/?h=master-next&id=9b4605005a307039e66de2db416ed826399e5103

[Christian Brauner (cbrauner)]

Fwiw, Seth sent the patch upstream and it's in mainline and backported already:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=61129dd29f7962f278b618a2a3e8fdb986a66dc8