On Wed, Aug 04, 2021 at 11:17:33AM +0100, Valentin Schneider wrote:
>
> +Cc Scott, Sebastian
>
> On 03/08/21 16:43, Paul E. McKenney wrote:
> > On Tue, Aug 03, 2021 at 11:54:36PM +0100, Valentin Schneider wrote:
> >> Running RCU torture with CONFIG_PREEMPT_RT under v5.13-rt1 triggers:
> >>
> >> [ 10.821700] DEBUG_LOCKS_WARN_ON(this_cpu_read(softirq_ctrl.cnt))
> >> [ 10.821716] WARNING: CPU: 5 PID: 137 at kernel/softirq.c:173 __local_bh_disable_ip (kernel/softirq.c:173 (discriminator 31))
> >> [ 10.821739] Modules linked in:
> >> [ 10.821749] CPU: 5 PID: 137 Comm: rcu_torture_rea Not tainted 5.13.0-rt1-00005-g08bbda29766a #129
> >> [ 10.821759] Hardware name: ARM Juno development board (r0) (DT)
> >> [ 10.821765] pstate: 60000005 (nZCv daif -PAN -UAO -TCO BTYPE=--)
> >> [ 10.821938] Call trace:
> >> [ 10.821941] __local_bh_disable_ip (kernel/softirq.c:173 (discriminator 31))
> >> [ 10.821950] rcutorture_one_extend (./include/linux/rcupdate.h:274 ./include/linux/rcupdate.h:737 kernel/rcu/rcutorture.c:1443)
> >> [ 10.821960] rcu_torture_one_read (kernel/rcu/rcutorture.c:1590 kernel/rcu/rcutorture.c:1638)
> >> [ 10.821968] rcu_torture_reader (kernel/rcu/rcutorture.c:1730)
> >> [ 10.821976] kthread (kernel/kthread.c:321)
> >> [ 10.821986] ret_from_fork (arch/arm64/kernel/entry.S:1005)
> >> [ 10.821997] irq event stamp: 478635
> >> [ 10.822001] hardirqs last enabled at (478635): _raw_spin_unlock_irq (./arch/arm64/include/asm/irqflags.h:35 ./include/linux/spinlock_api_smp.h:168 kernel/locking/spinlock.c:202)
> >> [ 10.822016] hardirqs last disabled at (478634): __schedule (kernel/sched/core.c:5154 (discriminator 1))
> >> [ 10.822029] softirqs last enabled at (478626): __local_bh_enable_ip (./arch/arm64/include/asm/irqflags.h:85 kernel/softirq.c:262)
> >> [ 10.822040] softirqs last disabled at (478622): rcutorture_one_extend (./include/linux/bottom_half.h:19 kernel/rcu/rcutorture.c:1441)
> >>
> >> Per this warning, softirqs cannot be disabled in a non-preemptible region
> >> under CONFIG_PREEMPT_RT. Adjust RCU torture accordingly.
> >>
> >> Signed-off-by: Valentin Schneider <valentin.schneider@xxxxxxx>
> >> ---
> >> kernel/rcu/rcutorture.c | 2 ++
> >> 1 file changed, 2 insertions(+)
> >>
> >> diff --git a/kernel/rcu/rcutorture.c b/kernel/rcu/rcutorture.c
> >> index 6096a7d14342..680f66b65f14 100644
> >> --- a/kernel/rcu/rcutorture.c
> >> +++ b/kernel/rcu/rcutorture.c
> >> @@ -1537,6 +1537,8 @@ rcutorture_extend_mask(int oldmask, struct torture_random_state *trsp)
> >> * them on non-RT.
> >> */
> >> if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
> >
> > This depends on some rcutorture patches in -rt that are not yet in
> > -rcu. Would -rt be a good place for this one, or are those patches
> > now ready for -rcu?
> >
>
> Right, this goes along with
>
> 72d6f4f680bf ("rcutorture: Avoid problematic critical section nesting on RT")
>
> (from v5.13-rt1), which seems to apply cleanly on top of the current
> mainline. So if we want this to go straight into -rcu, the above needs to
> go along with it.
Ah, this one needs to have its changes conditioned on PREEMPT_RT.
The situations that this patch excludes really can happen in mainline,
so they still need to be tested in mainline.
It all comes back now. ;-)
Thanx, Paul
> Thomas et al, any objections?
>
> > Thanx, Paul
> >
> >> + /* Can't disable bh in atomic context under PREEMPT_RT */
> >> + mask &= ~(RCUTORTURE_RDR_ATOM_BH | RCUTORTURE_RDR_ATOM_RBH);
> >> /*
> >> * Can't release the outermost rcu lock in an irq disabled
> >> * section without preemption also being disabled, if irqs
> >> --
> >> 2.25.1
> >>
