Re: [PATCH v9] kernel/fork: beware of __put_task_struct calling context

[Wander Lairson Costa]

On Thu, Jun 1, 2023 at 3:14 PM Oleg Nesterov <oleg@xxxxxxxxxx> wrote:
>
> On 06/01, Wander Lairson Costa wrote:
> >
> > On Mon, May 29, 2023 at 9:23 AM Oleg Nesterov <oleg@xxxxxxxxxx> wrote:
> > >
> > > On 05/17, Wander Lairson Costa wrote:
> > > >
> > > > On Wed, May 17, 2023 at 12:26 PM Oleg Nesterov <oleg@xxxxxxxxxx> wrote:
> > > > >
> > > > > LGTM but we still need to understand the possible problems with CONFIG_PROVE_RAW_LOCK_NESTING ...
> > > > >
> > > > > Again, I'll try to investigate when I have time although I am not sure I can really help.
> > > > >
> > > > > Perhaps you too can try to do this ? ;)
> > > > >
> > > >
> > > > FWIW, I tested this patch with CONFIG_PROVE_LOCK_NESTING in RT and
> > > > stock kernels. No splat happened.
> > >
> > > Strange... FYI, I am running the kernel with this patch
> > >
> > >         diff --git a/kernel/sys.c b/kernel/sys.c
> > >         index 339fee3eff6a..3169cceddf3b 100644
> > >         --- a/kernel/sys.c
> > >         +++ b/kernel/sys.c
> > >         @@ -2412,6 +2412,17 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
> > >
> > >                 error = 0;
> > >                 switch (option) {
> > >         +       case 666: {
> > >         +               static DEFINE_SPINLOCK(l);
> > >         +               static DEFINE_RAW_SPINLOCK(r);
> > >         +
> > >         +               raw_spin_lock(&r);
> > >         +               spin_lock(&l);
> > >         +               spin_unlock(&l);
> > >         +               raw_spin_unlock(&r);
> > >         +
> > >         +               break;
> > >         +       }
> > >                 case PR_SET_PDEATHSIG:
> > >                         if (!valid_signal(arg2)) {
> > >                                 error = -EINVAL;
> > >
> > > applied (because I am too lazy to compile a module ;) and
> > >
> >
> > FWIW, I converted it to a module [1]
>
> where is [1] ?  not that I think this matters though...
>
> > >         # perl -e 'syscall 157,666'
> > >
> > > triggers the lockdep bug
> > >
> > >         =============================
> > >         [ BUG: Invalid wait context ]
> > >         6.4.0-rc2-00018-g4d6d4c7f541d-dirty #1176 Not tainted
> > >         -----------------------------
> > >         perl/35 is trying to lock:
> > >         ffffffff81c4cc18 (l){....}-{3:3}, at: __do_sys_prctl+0x21b/0x87b
> > >         other info that might help us debug this:
> > >         context-{5:5}
> > >         ...
> > >
> > > as expected.
> > >
> >
> > Yeah, I tried it here and I had the same results,
>
> OK,
>
> > but only in the RT kernel
>
> this again suggests that your testing was wrong or I am totally confused (quite
> possible, I know nothing about RT). I did the testing without CONFIG_PREEMPT_RT.
>

Hrm, could you please share your .config?

> > But running the reproducer for put_task_struct(), works fine.
>
> which reproducer ?
>

Only now I noticed I didn't add the reproducer to the commit message:

while true; do
    stress-ng --sched deadline --sched-period 1000000000
--sched-runtime 800000000 --sched-deadline 1000000000 --mmapfork 23 -t
20
done