Re: Removal of printk safe buffers delays NMI context printk

[John Ogness]

Hi Nick,

On 2021-11-05, Nicholas Piggin <npiggin@xxxxxxxxx> wrote:
> It seems printk from NMI context is now delayed indefinitely and
> there is no printk_safe_flush equivalent (or I can't see one) to
> allow a NMI buffer to be flushed by a different CPU.

NMI flushing is triggered using irq work (for the same CPU). This should
not have changed recently. Are you reporting a new issue?

> This causes hard lockup watchdog messages to not get shown on the
> console. I can call printk from a different CPU and that seems to
> flush the stuck CPU's NMI buffer immediately.

Perhaps we should be triggering the irq work on multiple CPUs if from
NMI context?

> What's the best way to expose this? Can we have something like tihs?
>
> void printk_flush(void)
> {
> 	preempt_disable();
> 	if (console_trylock_spinning())
> 		console_unlock();
> 	preempt_enable();
>         wake_up_klogd();
> }

We are planning on implementing a pr_flush() that will do something
similar. But I am wondering how you are planning on triggering a CPU to
call that function.

John Ogness