Re: [PATCH printk v4 17/27] printk: nbcon: Use nbcon consoles in console_flush_all()

[John Ogness <john.ogness@linutronix.de>]

On 2024-04-11, Petr Mladek <pmladek@suse.com> wrote:
> console_flush_all() is called from console_unlock().
> It might be called in atomic context. But the current
> owner might be theoretically scheduled out.

Nice catch. This problem was introduced in this series after you pointed
out that return value of nbcon_legacy_emit_next_record() did not match
the semantics of console_emit_next_record().

> I see 3 possible solutions:
>
>   1. Enforce that nbcon context can be acquired only with preemtion
>      disabled.

Not acceptable.

>   2. Enforce that nbcon context can be acquired only with
>      interrupts. It would prevent deadlock when some future
>      code interrupt flush in NBCON_PRIO_EMERGENCY context.
>      And then a potential nested console_flush_all() won't be
>      able to takeover the interrupted NBCON_PRIO_CONTEXT
>      and there will be no progress.

Not acceptable.

>   3. console_flush_all() should ignore nbcon console when
>      it is not able to get the context, aka no progress.

This was the previous implementation, but as you point out, it is an
issue that console_flush_all() is no longer reliable.

I will continue this topic by responding to your follow-up message.

John