Re: [RFC][PATCH 1/2] printk: lock console_sem before we unregister boot consoles

[Sergey Senozhatsky]

On (04/25/19 16:56), Sergey Senozhatsky wrote:
> > Great idea!
> > 
> > It would deserve a separate patch that moves the pr_info()
> > and removes the invalid comment.
> > 
> > Actually, the pr_info() would deserve a comment explaining
> > why it should be called before console_unlock().
> 
> Good. So I think I'll drop patch #1 from the series, add
> two more patches - invalid comment + pr_info() - and rework
> locking in patch #3 (this should take care of a race which
> patch #1 was intended to fix).

D'oh, I can't drop patch #1, we still need to factor out
__unregister_console(). Calling unregister_console(bcon)
under console_sem will deadlock us. So I'll just end up
having __registed_console() and __unregister_console(),
both should be called under console_sem:

register_console()
{
	console_lock()
	__register_console()
	{
		...
		for_each_console(bcon)
			if (bcon->flags & CON_BOOT)
				__unregister_console(bcon);
	}
	console_unlock()
}

	-ss