Re: [External] Re: [PATCH 2/2] sched: mark PRINTK_DEFERRED_CONTEXT_MASK in __schedule()

[Peter Zijlstra <peterz@infradead.org>]

On Tue, Sep 29, 2020 at 04:27:51PM +0200, Petr Mladek wrote:

> Upstreaming the console handling will be the next big step. I am sure
> that there will be long discussion about it. But there might be
> few things that would help removing printk_deferred().
> 
> 1. Messages will be printed on consoles by dedicated kthreads. It will
>    be safe context. No deadlocks.

This, that's what I remember. With sane consoles having a
->write_atomic() callback which is called in-line.

Current -RT has a single kthread that's flushing the consoles.

> 2. The registration and unregistration of consoles should not longer
>    be handled by console_lock (semaphore). It should be possible to
>    call most consoles without a sleeping lock. It should remove all
>    these deadlocks between printk() and scheduler().

I'm confused, who cares about registation? That only happens once at
boot, right?

>    There might be problems with some consoles. For example, tty would
>    most likely still need a sleeping lock because it is using the console
>    semaphore also internally.

But per 1) above, that's done from a kthread, so nobody cares.

> 3. We will try harder to get the messages out immediately during
>    panic().

As long as you guarantee to first write everything to consoles with
->write_atomic() before attempting to flush others that should be fine.

> It would take some time until the above reaches upstream. But it seems
> to be the right way to go.

Yep.

> Finally, the deadlock happens "only" when someone is waiting on
> console_lock() in parallel. Otherwise, the waitqueue for the semaphore
> is empty and scheduler is not called.

There's more deadlocks. In fact the one described earlier in this
thread isn't the one you mention.

> It means that there is quite a big change to see the WARN(). It might
> be even bigger than with printk_deferred() because WARN() in scheduler
> means that the scheduler is big troubles. Nobody guarantees that
> the deferred messages will get handled later.

I only care about ->write_atomic() :-) Anything else is a
best-effort-loss anyway.