Re: powerpc: userspace processes crash on abort() from libc.so.6

Hi Shivaprasad.

On 8/27/26 11:13 AM, Shivaprasad G Bhat wrote:
> 
> Hi All, Thomas,
> 
> On my PPC64 machine, I see the below systemd crash orginating from abort() during boot on top of the tree. Complete dmesg can be seen here [1]
>        abort (libc.so.6 + 0x29484)
>        __libc_message_impl (libc.so.6 + 0x9e00c)
>        __libc_fatal (libc.so.6 + 0x9e06c)
>        start_thread (libc.so.6 + 0xadffc)
>        __clone (libc.so.6 + 0x152d68)
>        __clone (libc.so.6 + 0x152d68)
>        create_thread.constprop.0 (libc.so.6 + 0xad9c8)
> 
> The git bisect pointed me to commit 3424d8c18a7da "(Merge tag 'core-entry-2026-08-17' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip)"
> 
> Please find the kernel config used to build the kernel at [2].
> 
> I have tried applying the pending fixes from Mukesh [3] and Ritesh [4] for the currently outstanding issues and still see the problem.
> 
> The system is running Fedora 42, with glibc-2.41-7.fc42.ppc64le. The
> complete list of toolchain and packages can be found here [5]
> 
> Thanks and Regards,
> Shivaprasad
> 
> References:
> 1 - https://gist.github.com/shivaprasadbhat/6f166374bab12b4f4c1cd4ac829e562c
> 2 - https://gist.github.com/shivaprasadbhat/aa499468e124015eec5932fb13e88e05
> 3 - https://lore.kernel.org/linuxppc-dev/[email protected]/#t
> 4 - https://lore.kernel.org/linuxppc-dev/[email protected]/
> 5 - https://gist.github.com/shivaprasadbhat/ae56ca849c3009e8642a6313dc465ea7
> 
> =========
> Backtrace from dmesg:
> [   22.039343][ T1833] systemd-coredump[1833]: Module libsystemd-shared-257.6-1.fc42.so from rpm systemd-257.6-1.fc42.ppc64le
> [   22.039378][ T1833] systemd-coredump[1833]: Stack trace of thread 1828:
> [   22.039378][ T1833] systemd-coredump[1833]: Stack trace of thread 1828:
> [   22.039401][ T1833] systemd-coredump[1833]: #0  0x00007fffa4c29484 abort (libc.so.6 + 0x29484)
> [   22.039401][ T1833] systemd-coredump[1833]: #0  0x00007fffa4c29484 abort (libc.so.6 + 0x29484)
> [   22.039429][ T1833] systemd-coredump[1833]: #1  0x00007fffa4c9e00c __libc_message_impl (libc.so.6 + 0x9e00c)
> [   22.039429][ T1833] systemd-coredump[1833]: #1  0x00007fffa4c9e00c __libc_message_impl (libc.so.6 + 0x9e00c)
> [   22.039461][ T1833] systemd-coredump[1833]: #2  0x00007fffa4c9e06c __libc_fatal (libc.so.6 + 0x9e06c)
> [   22.039461][ T1833] systemd-coredump[1833]: #2  0x00007fffa4c9e06c __libc_fatal (libc.so.6 + 0x9e06c)
> [   22.039491][ T1833] systemd-coredump[1833]: #3  0x00007fffa4cadffc start_thread (libc.so.6 + 0xadffc)
> [   22.039491][ T1833] systemd-coredump[1833]: #3  0x00007fffa4cadffc start_thread (libc.so.6 + 0xadffc)
> [   22.039521][ T1833] systemd-coredump[1833]: #4  0x00007fffa4d52d68 __clone (libc.so.6 + 0x152d68)
> [   22.039521][ T1833] systemd-coredump[1833]: #4  0x00007fffa4d52d68 __clone (libc.so.6 + 0x152d68)
> [   22.039551][ T1833] systemd-coredump[1833]: #5  0x00007fffa4d52d68 __clone (libc.so.6 + 0x152d68)
> [   22.039551][ T1833] systemd-coredump[1833]: #5  0x00007fffa4d52d68 __clone (libc.so.6 + 0x152d68)
> [   22.039580][ T1833] systemd-coredump[1833]: #6  0x00007fffa4cad9c8 create_thread.constprop.0 (libc.so.6 + 0xad9c8)
> [   22.039580][ T1833] systemd-coredump[1833]: #6  0x00007fffa4cad9c8 create_thread.constprop.0 (libc.so.6 + 0xad9c8)
> [   22.039614][ T1833] systemd-coredump[1833]: #7  0x00000000003af858 n/a (n/a + 0x0)
> [   22.039614][ T1833] systemd-coredump[1833]: #7  0x00000000003af858 n/a (n/a + 0x0)
> [   22.039639][ T1833] systemd-coredump[1833]: ELF object binary architecture: PowerPC64
> [   22.039639][ T1833] systemd-coredump[1833]: ELF object binary architecture: PowerPC64
> [   22.042109][    T1] systemd[1]: systemd-journald.service: Scheduled restart job, restart counter is at 1.

I was looking at merge patch and It is possible that TIF_SYSCALL_RET is still set when it is returned
and maybe causing a mismatch.

Does below help?
---

--- a/arch/powerpc/kernel/syscall.c
+++ b/arch/powerpc/kernel/syscall.c
@@ -19,8 +19,10 @@ notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
  	long ret;
  	syscall_fn f;
  
-	if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0)))
+	if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0))) {
+		clear_thread_flag(TIF_SYSCALL_RET);
  		return syscall_get_error(current, regs);
+	}
  
  	if (unlikely(test_and_clear_thread_flag(TIF_SYSCALL_RET)))
  		return syscall_get_error(current, regs);