Re: [PATCH v4] printk: Add console owner and waiter logic to load balance console writes

[Petr Mladek]

Hello,

thanks a lot for help. I am sorry for the late response. I wanted to
handle this mail with a clean head.

On Tue 2017-11-28 10:42:29, Byungchul Park wrote:
> On Fri, Nov 24, 2017 at 04:58:16PM +0100, Petr Mladek wrote:
> > @@ -1797,13 +1797,6 @@ asmlinkage int vprintk_emit(int facility, int level,
> >  				spin_release(&console_owner_dep_map, 1, _THIS_IP_);
> >  				printk_safe_exit_irqrestore(flags);
> >  
> > -				/*
> > -				 * The owner passed the console lock to us.
> > -				 * Since we did not spin on console lock, annotate
> > -				 * this as a trylock. Otherwise lockdep will
> > -				 * complain.
> > -				 */
> > -				mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
> 
> Hello Petr,
> 
> IMHO, it would get unbalanced if you only remove this mutex_acquire().
> 
> >  				console_unlock();
> >  				printk_safe_enter_irqsave(flags);
> >  			}
> > @@ -2334,10 +2327,10 @@ void console_unlock(void)
> >  		/* The waiter is now free to continue */
> >  		spin_release(&console_owner_dep_map, 1, _THIS_IP_);
> >  		/*
> > -		 * Hand off console_lock to waiter. The waiter will perform
> > -		 * the up(). After this, the waiter is the console_lock owner.
> > +		 * Hand off console_lock to waiter. After this, the waiter
> > +		 * is the console_lock owner.
> >  		 */
> > -		mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
> 
> IMHO, this release() should be moved to somewhere properly.
> 
> > +		lock_commit_crosslock((struct lockdep_map *)&console_lock_dep_map);
> >  		printk_safe_exit_irqrestore(flags);
> >  		/* Note, if waiter is set, logbuf_lock is not held */
> >  		return;
> 
> However, now that cross-release was introduces, lockdep can be applied
> to semaphore operations. Actually, I have a plan to do that. I think it
> would be better to make semaphore tracked with lockdep and remove all
> these manual acquire() and release() here. What do you think about it?

IMHO, it would be great to add lockdep annotations into semaphore
operations.

Well, I am not sure if this would be enough in this case. I think
that the locking dependency in this Steven's patch is special.
The semaphore is passed from one owner to another one without
unlocking. Both sides wait for each other using a busy loop.

The busy loop/waiting is activated only when the current owner
is not sleeping to avoid softlockup. I think that it is
a kind of conditional cross-release or something even
more special.

Sigh, I wish I was able to clean my head even more to be
able to think about this.

Best Regards,
Petr