Re: NOHZ tick-stop error: local softirq work is pending, handler #08!!! on Dell XPS 13 9360

[Frederic Weisbecker <frederic@kernel.org>]

Le Tue, Feb 11, 2025 at 12:57:33PM +0100, Paul Menzel a écrit :
> Dear Frederic,
> 
> 
> Thank you for your reply.
> 
> 
> Am 10.02.25 um 14:26 schrieb Frederic Weisbecker:
> > Le Mon, Feb 10, 2025 at 12:59:42PM +0100, Paul Menzel a écrit :
> 
> > > Am 10.02.25 um 12:45 schrieb Michał Pecio:
> > > 
> > > > > > > > > > > On Dell XPS 13 9360/0596KF, BIOS 2.21.0 06/02/2022, with Linux
> > > > > > > > > > > 6.9-rc2+
> > > > 
> > > > > Just for the record, I am still seeing this with 6.14.0-rc1
> > > > 
> > > > Is this a regression? If so, which versions were not affected?
> > > 
> > > Unfortunately, I do not know. Right now, my logs go back until September
> > > 2024.
> > > 
> > >      Sep 22 13:08:04 abreu kernel: Linux version 6.11.0-07273-g1e7530883cd2 (build@bohemianrhapsody.molgen.mpg.de) (gcc (Debian 14.2.0-5) 14.2.0, GNU ld (GNU Binutils for Debian) 2.43.1) #12 SMP PREEMPT_DYNAMIC Sun Sep 22 09:57:36 CEST 2024
> > > 
> > > > How hard to reproduce? Wasn't it during resume from hibernation?
> > > 
> > > It’s not easy to reproduce, and I believe it’s not related with resuming
> > > from hibernation (which I do not use) or ACPI S3 suspend. I think, I can
> > > force it more, when having the USB-C adapter with only the network cable
> > > plugged into it, and then running `sudo powertop --auto-tune`. But sometimes
> > > it seems unrelated.
> > > 
> > > > IRQ isuses may be a red herring, this code here is a busy wait under
> > > > spinlock. There are a few of those, they cause various problems.
> > > > 
> > > >                   if (xhci_handshake(&xhci->op_regs->status,
> > > >                                 STS_RESTORE, 0, 100 * 1000)) {
> > > >                           xhci_warn(xhci, "WARN: xHC restore state timeout\n");
> > > > 			spin_unlock_irq(&xhci->lock);
> > > >                           return -ETIMEDOUT;
> > > >                   }
> > > > 
> > > > This thing timing out may be close to the root cause of everything.
> > > 
> > > Interesting. Hopefully the USB folks have an idea.
> > 
> > Handler #08 is NET_RX. So something raised the NET_RX on some non-appropriate
> > place, perhaps...
> > 
> > Can I ask you one more trace dump?
> > 
> > I need:
> > 
> > echo 1 > /sys/kernel/tracing/events/irq/softirq_raise/enable
> > echo 1 > /sys/kernel/tracing/options/stacktrace
> > 
> > Unfortunately this will also involve a small patch:
> > 
> > diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
> > index fa058510af9c..accd2eb8c927 100644
> > --- a/kernel/time/tick-sched.c
> > +++ b/kernel/time/tick-sched.c
> > @@ -1159,6 +1159,9 @@ static bool report_idle_softirq(void)
> >   	if (local_bh_blocked())
> >   		return false;
> > +	trace_printk("STOP\n");
> > +	trace_dump_stack(0);
> > +	tracing_off();
> >   	pr_warn("NOHZ tick-stop error: local softirq work is pending, handler #%02x!!!\n",
> >   		pending);
> >   	ratelimit++;
> 
> Thank you for your help. I applied the patch on top of 6.14-rc2, and was
> able to reproduce the issue. Please find the Linux messages attached, and
> the trace can be downloaded [1].

So here is the offender:

 => __raise_softirq_irqoff
 => __napi_schedule
 => rtl8152_runtime_resume.isra.0
 => rtl8152_resume
 => usb_resume_interface.isra.0
 => usb_resume_both
 => __rpm_callback
 => rpm_callback
 => rpm_resume
 => __pm_runtime_resume
 => usb_autoresume_device
 => usb_remote_wakeup
 => hub_event
 => process_one_work
 => worker_thread
 => kthread
 => ret_from_fork
 => ret_from_fork_asm

It is calling napi_schedule() from a non-interrupt. And since
____napi_schedule() assumes to be called from an interrupt, it
raises the softirq accordingly without waking up ksoftirqd.

Can you try the following fix (untested, sorry...) ?

diff --git a/drivers/net/usb/r8152.c b/drivers/net/usb/r8152.c
index 468c73974046..8f6ea4e7685c 100644
--- a/drivers/net/usb/r8152.c
+++ b/drivers/net/usb/r8152.c
@@ -8537,8 +8537,11 @@ static int rtl8152_runtime_resume(struct r8152 *tp)
 		clear_bit(SELECTIVE_SUSPEND, &tp->flags);
 		smp_mb__after_atomic();
 
-		if (!list_empty(&tp->rx_done))
+		if (!list_empty(&tp->rx_done)) {
+			local_bh_disable();
 			napi_schedule(&tp->napi);
+			local_bh_enable();
+		}
 
 		usb_submit_urb(tp->intr_urb, GFP_NOIO);
 	} else {
diff --git a/include/linux/lockdep.h b/include/linux/lockdep.h
index 67964dc4db95..1bd730b881f0 100644
--- a/include/linux/lockdep.h
+++ b/include/linux/lockdep.h
@@ -619,6 +619,17 @@ do {									\
 		     (!in_softirq() || in_irq() || in_nmi()));		\
 } while (0)
 
+/*
+ * Assert to be either in hardirq or in serving softirq or with
+ * softirqs disabled. Verifies a safe context to queue a softirq
+ * with __raise_softirq_irqoff().
+ */
+#define lockdep_assert_in_interrupt()				\
+do {								\
+	WARN_ON_ONCE(__lockdep_enabled && !in_interrupt());	\
+} while (0)
+
+
 extern void lockdep_assert_in_softirq_func(void);
 
 #else
@@ -634,6 +645,7 @@ extern void lockdep_assert_in_softirq_func(void);
 # define lockdep_assert_preemption_enabled() do { } while (0)
 # define lockdep_assert_preemption_disabled() do { } while (0)
 # define lockdep_assert_in_softirq() do { } while (0)
+# define lockdep_assert_in_interrupt() do { } while (0)
 # define lockdep_assert_in_softirq_func() do { } while (0)
 #endif
 
diff --git a/net/core/dev.c b/net/core/dev.c
index c0021cbd28fc..80e415ccf2c8 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4666,6 +4666,7 @@ static inline void ____napi_schedule(struct softnet_data *sd,
 	struct task_struct *thread;
 
 	lockdep_assert_irqs_disabled();
+	lockdep_assert_in_interrupt();
 
 	if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
 		/* Paired with smp_mb__before_atomic() in