New warning in linus/master

[Thomas Weißschuh]

On Wed, Apr 22, 2026 at 04:06:45PM +0800, Ricardo Ribalda wrote:
> On Wed, 22 Apr 2026 at 15:32, Thomas Weißschuh
> <thomas.weissschuh at linutronix.de> wrote:
> >
> > On Wed, Apr 22, 2026 at 03:00:11PM +0800, Ricardo Ribalda wrote:
> > > On Wed, 22 Apr 2026 at 13:57, Thomas Weißschuh
> > > <thomas.weissschuh at linutronix.de> wrote:
> > > > On Wed, Apr 22, 2026 at 11:51:45AM +0800, Ricardo Ribalda wrote:
> > > > > Media-CI has found a couple of new warnings in the latest kernel
> > > > > version for aarch64 and powerpc. They get fixed with this patch and
> > > > > before moving I wanted to know if this was under your radar.
> > > >
> > > > Thanks for the report. I was not aware of these so far.
> > > >
> > > > > diff --git a/arch/arm64/kernel/vdso/Makefile b/arch/arm64/kernel/vdso/Makefile
> > > > > index 7dec05dd33b7..65914842fae0 100644
> > > > > --- a/arch/arm64/kernel/vdso/Makefile
> > > > > +++ b/arch/arm64/kernel/vdso/Makefile
> > > > > @@ -50,7 +50,7 @@ CFLAGS_vgettimeofday.o = $(CC_FLAGS_ADD_VDSO)
> > > > >  CFLAGS_vgetrandom.o = $(CC_FLAGS_ADD_VDSO)
> > > > >
> > > > >  ifneq ($(c-gettimeofday-y),)
> > > > > -  CFLAGS_vgettimeofday.o += -include $(c-gettimeofday-y)
> > > > > +  CFLAGS_vgettimeofday.o += -include $(c-gettimeofday-y)
> > > > > -Wno-maybe-uninitialized
> > > > >  endif
> > > >
> > > > (...)
> > > >
> > > > I'd like to know exactly what is going on before suppressing the warning.
> > > > It is a non-standard warning, only enabled by *some* of the vDSO builds
> > > > for some reason.
> > > >
> > > > > https://gitlab.freedesktop.org/linux-media/users/ribalda/-/pipelines/1649144/test_report?job_name=cross-gcc
> > > >
> > > > While I was able to download a configuration from this job and also use the
> > > > same container image, I can not reproduce the issue. Is the configuration the
> > > > full one or only the template?
> > > >
> > > > Could you provide full reproduction steps?
> > >
> > > You can try repro with:
> > >
> > > work/linux $ podman run -v .:/workdir/ --rm -it
> > > registry.freedesktop.org/linux-media/media-ci/build:latest
> > > $ cd /workdir
> > > $ CROSS_COMPILE=aarch64-linux-gnu- ARCH=arm64 make allyesconfig
> > > $ CROSS_COMPILE=aarch64-linux-gnu- ARCH=arm64 make
> > > arch/arm64/kernel/vdso/vgettimeofday.o
> > >
> > > The gcc version discrepancy is because I the error was due to old gcc
> > > version and I was playing around with that... but it fails in both
> > > gcc14 and gcc15
> >
> > Ack.
> >
> > > You can try with debian testing with
> > > work/linux$ podman run -v .:/workdir/ --rm -it debian:testing
> > > $ apt-get update
> > > $ apt-get install gcc-aarch64-linux-gnu build-essential flex bison libssl-dev bc
> > > $ cd /workdir
> > > $ CROSS_COMPILE=aarch64-linux-gnu- ARCH=arm64 make allyesconfig
> > > $ CROSS_COMPILE=aarch64-linux-gnu- ARCH=arm64 make
> > > arch/arm64/kernel/vdso/vgettimeofday.o
> >
> > Both reproducers do *not* reproduce the issue for me.
> > (It is more or less exactly what I tried before)
> 
> 
> Ok, I think I found what is going on. media-ci was forcing
> KCFLAGS=-Wmaybe-uninitialized
> 
> https://gitlab.freedesktop.org/linux-media/media-ci/-/blob/main/test-build.sh?ref_type=heads#L29
> 
> And something has changed in the kernel in the last version that
> triggers a (hopefully) false positive.
> 
> can you try with:
> 
> CROSS_COMPILE=aarch64-linux-gnu- ARCH=arm64 make
> KCFLAGS=-Wmaybe-uninitialized  allyesconfig
> CROSS_COMPILE=aarch64-linux-gnu- ARCH=arm64 make
> KCFLAGS=-Wmaybe-uninitialized  prepare

That works!

The warning was introduced in ed78b7b2c5ae ("vdso/gettimeofday: Add a helper to
read the sequence lock of a time namespace aware clock").

It is *not* a false-positive, specifically:

do {
	if (vdso_read_begin_timens(vc, &seq)) {
		vd = __arch_get_vdso_u_timens_data(vd);
		vc = &vd->aux_clock_data[idx];
		/* Re-read from the real time data page */
		continue;

This 'continue' jumps to the end of the loop body and if the real vDSO datapage
by chance has a sequence counter of '1' will exit the loop, leaving 'sec' and
'nsec' uninitialized. My believe was that the 'continue' would jump to the
*beginning* of the loop body, which is cleary wrong. Previously there was an
inner while loop() which would make the 'continue' effectively start at the
beginning of the outer do-while loop.

	}

	/* Auxclock disabled? */
	if (vc->clock_mode == VDSO_CLOCKMODE_NONE)
		return false;

	if (!vdso_get_timestamp(vd, vc, VDSO_BASE_AUX, &sec, &ns))
		return false;

} while (vdso_read_retry(vc, seq));

vdso_set_timespec(ts, sec, ns);


I'm working on a fix.


Thomas