Re: [PATCH 4/4] cpuset,mm: use rwlock to protect task->mempolicyand mems_allowed

From: Nick Piggin
Date: Thu Mar 04 2010 - 11:34:27 EST

On Thu, Mar 04, 2010 at 03:58:24PM +0100, Peter Zijlstra wrote:
> On Thu, 2010-03-04 at 14:30 +1100, Nick Piggin wrote:
> >
> > Thanks for working on this. However, rwlocks are pretty nasty to use
> > when you have short critical sections and hot read-side (they're twice
> > as heavy as even spinlocks in that case).
> Should we add a warning for them?

Yes I think it could be useful.

Most people agree rwlock is *almost* always the wrong thing to do. Or at
least, they can easily be used wrongly because they seem like a great
idea for read-mostly data.

> There really rarely is a good case for using rwlock_t, for as you say
> they're a pain and often more expensive than a spinlock_t, and if
> possible RCU has the best performance.

Yep. Not to mention they starve writers (and don't FIFO like spinlocks).

Between normal spinlocks, RCU, percpu, and seqlocks, there's not much
room for rwlocks. Even tasklist lock should be RCUable if the effort is
put into it.

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at