On Wed, Feb 05, 2025 at 09:00:07PM +0106, John Ogness wrote:

On 2025-02-05, "Paul E. McKenney" [off-list ref] wrote:

quoted

quoted

This is caused by RCU falling behind a callback-flooding kthread that
invokes call_rcu() in a semi-tight loop.  Setting rcutree.kthread_prio=40
avoids the splat, but still gets the shutdown-time hang.  Retrying with
the default rcutree.kthread_prio=2 failed to reproduce the splat, but
it did reproduce the shutdown-time hang.

OK, maybe printk buffers are not being flushed?  A 100-millisecond sleep
at the end of of rcu_torture_cleanup() got all of rcutorture's output
flushed, but lost the subsequent shutdown-time console traffic.  The
pr_flush(HZ/10,1) seems more sensible, but this is private to printk().

I would like to log the shutdown-time console traffic because RCU can
sometimes break things on that path.

pr_flush() was changed to private because there were no users. It would
not be a problem to make it available. Adding a pr_flush() to
rcu_torture_cleanup() would be an appropriate workaround for now (more
on this at the end).

Ah, got it.

quoted

There is a call to kmsg_dump(KMSG_DUMP_SHUTDOWN) in kernel_power_off()
that appears to be intended to dump out the printk() buffers,

It only dumps the buffers to the registered kmsg_dumpers. It is not
responsible for flushing console backlogs.

That would explain its not doing much for me.  ;-)

quoted

but it
does not seem to do so in kernels built with CONFIG_PREEMPT_RT=y.
Does there need to be a pr_flush() call prior to the call to
migrate_to_reboot_cpu()?  Or maybe even to do_kernel_power_off_prepare()
or kernel_shutdown_prepare()?

With CONFIG_PREEMPT_RT=y, legacy consoles only print via a dedicated
kthread. Without a pr_flush() somewhere, there is basically no chance
that they will get backlogs flushed because noone is waitig for them.

The new console API (NBCON) provides support for "atomic consoles",
which _do_ flush by transitioning to synchronous printing during
shutdown/reboot. Unfortunately we still don't have any NBCON atomic
console implemented in the kernel. The 8250 UART will be our first
driver, most likely available in 6.15. (With the current PREEMPT_RT
patch applied, the 8250 NBCON atomic driver is used.)

Since only CONFIG_PREEMPT_RT=y has this issue, I am not sure if we want
to sprinkle pr_flush() calls on all sleepable shutdown/reboot paths,
although that is certainly one way to handle it. For your case, adding a
pr_flush() to rcu_torture_cleanup() and making pr_flush() non-private
would be an easy solution to avoid your problem.

Would it make sense to put a call to pr_flush() at the beginning of the
kernel_power_off() function?  I suspect that would take care of much
more than just rcutorture.

Or maybe after the pr_emerg("Power down\n"), since that is normally
the last thing that shows up on rcutorture's console logs.

I will give that a shot and see what happens.  ;-)

							Thanx, Paul