On Mon, Jul 07 2025 at 13:34, Arnd Bergmann wrote:

On Mon, Jul 7, 2025, at 08:57, Thomas Gleixner wrote:

quoted

On Tue, Jul 01 2025 at 10:58, Thomas Weißschuh wrote:

quoted

 
+#if defined(CONFIG_GENERIC_TIME_VSYSCALL) && defined(CONFIG_GENERIC_GETTIMEOFDAY) && \
+	defined(CONFIG_POSIX_AUX_CLOCKS)

CONFIG_GENERIC_GETTIMEOFDAY requires CONFIG_GENERIC_TIME_VSYSCALL, but
that's not expressed anywhere. This stuff has grown too many config
options over time:

GENERIC_TIME_VSYSCALL, HAVE_GENERIC_VDSO, GENERIC_VDSO_DATA_STORE

All architectures except SPARC64 use the generic code and the generic
VDSO data store implementation. That config maze wants to be
consolidated.

Would it help to replace the custom sparc64 vdso with the
thing that parisc does and call the fallback directly?

Definitely.

I doubt anyone still cares about the clock_gettime() performance
on sparc64, and removing it would remove all those special cases:

 arch/sparc/Kconfig                  |   1 -
 arch/sparc/include/asm/vvar.h       |  75 --------
 arch/sparc/kernel/Makefile          |   1 -
 arch/sparc/kernel/vdso.c            |  69 --------
 arch/sparc/vdso/Makefile            |   2 +-
 arch/sparc/vdso/vclock_gettime.c    | 340 +++---------------------------------
 arch/sparc/vdso/vdso.lds.S          |   2 -
 arch/sparc/vdso/vdso32/vdso32.lds.S |   3 +-
 arch/sparc/vdso/vma.c               | 265 +---------------------------
 9 files changed, 28 insertions(+), 730 deletions(-)

That's definitely an argument :)

(the added lines here also fix the missing clock_gettime64,
which was equally blocked on the sparc64 oddities)

I'm all for it.

Can you post a patch to that effect?

Thanks,

        tglx