On Fri, Oct 09, 2026 at 09:46:47PM -0700, Kees Cook wrote:
> On Fri, Oct 09, 2026 at 09:27:57PM -0700, Paul E. McKenney wrote:
> > On Fri, Oct 09, 2026 at 07:25:42PM -0700, Kees Cook wrote:
> > > On Fri, Oct 09, 2026 at 08:51:18AM -0700, Paul E. McKenney wrote:
> > > > On Thu, Oct 08, 2026 at 11:27:30PM -0700, Kees Cook wrote:
> > > > > Since commit 48ebe1cbbeb51 ("kbuild: compress the kernel with pigz if
> > > > > available"), cmd_gzip runs pigz through scripts/jobserver-exec, which
> > > > > takes the job slots make has free. GNU Make passes the jobserver pipe
> > > > > only to recipes it treats as recursive, but leaves --jobserver-auth in
> > > > > MAKEFLAGS for all of them. The make 4.3 in CentOS Stream 9 also hides
> > > > > the pipe in sub-makes (a backport of the fix for GNU Make bug 58232),
> > > > > so
> > > > > jobserver-exec cannot open it, runs pigz on one thread, and prints:
> > > > >
> > > > > WARNING: Unable to reopen jobserver read-side pipe:
> > > > > FileNotFoundError(2, 'No such file or directory')
> > > > >
> > > > > Mark the gzip recipes as recursive with "+", as commit ecab4115c44cc
> > > > > ("kbuild: mark `rustc` (and others) invocations as recursive") did for
> > > > > rustc. As with those, "make -n" now runs them.
> > > > >
> > > > > Tested ARCH=x86_64 and ARCH=arm64 defconfig with GNU Make 4.3 plus
> > > > > CentOS Stream 9's make-4.3-cloexec.patch, GCC 16.2.0, and -j64:
> > > > > without
> > > > > this change the warning appears and pigz runs with "-p 1"; with it
> > > > > there
> > > > > is no warning, and pigz runs with "-p 54" and "-p 64".
> > > >
> > > > Thank you, Kees!
> > > >
> > > > My initial test of this patch (using rcutorture) still showed the
> > > > failure.
> > >
> > > Oh! So the patch didn't fix it for you?
> > >
> > > > I am running GCC 11.5.0 and without any patches to "make". I take it
> > > > that I need to find and apply CentOS Stream 9's make-4.3-cloexec.patch?
> > > >
> > > > But if I use "make -j" instead of rcutorture, this patch does work fine.
> > > > Even with "make -j632".
> > >
> > > Without rcutorture was "make -jNN" broken for you without this patch?
> > >
> > > > I put rcutorture in the background using nohup and full redirection,
> > > > but "nohup make -j632 > /tmp/Make.out 2>&1 < /dev/null &" works just
> > > > fine. And running rcutorture in the foreground still fails.
> > > >
> > > > Very strange.
> > >
> > > O_O It only breaks ... when backgrounded? Oh my. Okay, I will try to
> > > figure that out.
> >
> > Apologies, let me present my results in a more organized fashion:
> >
> > foreground background
> > -next Fix -next Fix
> >
> > make -jN Bug Good Bug Good
> >
> > kvm.sh [1] Bug Bug Bug Bug
> >
> > [1] tools/testing/selftests/rcutorture/bin/kvm.sh --build-only --allcpus
> > --configs TREE02 --trust-make
> >
> > Here, -next is next-20261008, and Fix is with your patch on it.
> >
> > Just as mystifying to me, to be honest!
>
> Ah, I think I found it, but I'm not sure why kvm.sh tripped it but bare
> "make" didn't. I missed the non-literal forms:
>
> $(call if_changed,$(compress-y))
>
> I'll send a v2 with docs updated too.
Done for the day, but I will test tomorrow. Again, thank you!
Thanx, Paul
