Re: [PATCH 0/2] rcu: Transform kfree_rcu() into kvfree_rcu()

[Paul E. McKenney]

On Wed, Feb 07, 2018 at 08:57:00AM -0500, Steven Rostedt wrote:
> On Wed, 7 Feb 2018 00:31:04 -0800
> "Paul E. McKenney" <paulmck@xxxxxxxxxxxxxxxxxx> wrote:
> 
> > I see problems.  We would then have two different names for exactly the
> > same thing.
> > 
> > Seems like it would be a lot easier to simply document the existing
> > kfree_rcu() behavior, especially given that it apparently already works.
> > The really doesn't seem to me to be worth a name change.
> 
> Honestly, I don't believe this is an RCU sub-system decision. This is a
> memory management decision.

I couldn't agree more!

To that end, what are your thoughts on this patch?

https://lkml.kernel.org/r/1513895570-28640-1-git-send-email-rao.shoaib@xxxxxxxxxx

Advantages include the ability to optimize based on sl[aou]b state,
getting rid of the 4K offset hack in __is_kfree_rcu_offset(), better
cache localite, and, as you say, putting the naming responsibility
in the memory-management domain.

> If we have kmalloc(), vmalloc(), kfree(), vfree() and kvfree(), and we
> want kmalloc() to be freed with kfree(), and vmalloc() to be freed with
> vfree(), and for strange reasons, we don't know how the data was
> allocated we have kvfree(). That's an mm decision not an rcu one. We
> should have kfree_rcu(), vfree_rcu() and kvfree_rcu(), and honestly,
> they should not depend on kvfree() doing the same thing for everything.
> Each should call the corresponding member that they represent. Which
> would change this patch set.
> 
> Why? Too much coupling between RCU and MM. What if in the future
> something changes and kvfree() goes away or changes drastically. We
> would then have to go through all the users of RCU to change them too.
> 
> To me kvfree() is a special case and should not be used by RCU as a
> generic function. That would make RCU and MM much more coupled than
> necessary.

And that is one reason I am viewing the name-change patch with great
suspicion, especially given that there seems to be some controversy
among the memory-management folks as to exactly what the names should be.

							Thanx, Paul