Re: dcache shrink list corruption?

[Al Viro <viro@ZenIV.linux.org.uk>]

On Wed, Apr 30, 2014 at 04:14:14PM -0700, Linus Torvalds wrote:
> On Wed, Apr 30, 2014 at 4:04 PM, Linus Torvalds
> <torvalds@linux-foundation.org> wrote:
> >
> > Let me go talk to the paravirt people. Maybe they don't need this, and
> > I don't know exactly *how* they use that lock pointer after the unlock
> > in the "kick waiters" part. Maybe it's all good.
> 
> .. looking at current users (xen and kvm) it does in fact look all
> good. Yes, we "kick" possible waiters after the unlock, but the lock
> itself is not touched by that, it just uses the pointer to the lock as
> a way to figure out who to kick.
> 
> In fact, I kind of knew that, but had forgotten. We very much depend
> on spin_unlock being safe wrt immediate deleton already: the
> "completion" code very much depends on it. It does a "spin_unlock()"
> to release the completion, and it can get reused immediately (it might
> be on the stack, it might be in some data structure that gets freed).
> 
> So I'd suggest removing that whole RCU thing, because it should be
> safe to unlock something that can go away immediately. We'd have huge
> problems elsewhere if it wasn't safe.

OK, done and force-pushed.  Should propagate in a few...