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

[Sergey Senozhatsky]

Hi,

On (11/27/17 17:48), Byungchul Park wrote:
[..]
> > +				/* Owner will clear console_waiter on hand off */
> > +				while (READ_ONCE(console_waiter))
> > +					cpu_relax();
> > +
> > +				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_);
> 
> I'm afraid if it's ok even not to lock(or trylock) actually here. Is there
> any problem if you call console_trylock() instead of mutex_acquire() here?

console_trylock() will not work. console_trylock() implies that the
current printing process does up(), which a) opens a race with possible
console_lock()/console_trylock() from foreign CPUs, and b) additionally
wakes up a task from console_sem wait list [if there was one]. so chances
are some other CPU potentially can acquire the lock ahead of waiter,
forcing the waiter to continue spinning on console_sem. and the bad news
are a) it's spinning with local IRQs disabled and b) that another CPU,
which has acquired the console_sem, can schedule under console_sem.


anyway, we are not going to merge this patch. we already have discussed
that in V3 thread:

https://marc.info/?l=linux-kernel&m=151019815721161&w=2
https://marc.info/?l=linux-kernel&m=151020275921953&w=2
https://marc.info/?l=linux-kernel&m=151020404622181&w=2
https://marc.info/?l=linux-kernel&m=151020565222469&w=2


I took some parts of the Steven's patch set, tho, and backported them
to the current printk_kthread series.

	-ss