Re: [RFC] kill sleep_on

[viro@parcelfarce.linux.theplanet.co.uk]

On Sat, Jan 17, 2004 at 07:28:21PM +0000, David Woodhouse wrote:
> On Sat, 2004-01-17 at 19:43 +0100, Manfred Spraul wrote:
> > Virtually all sleep_on / interruptible_sleep_on users are racy. Actually
> > there is only one safe case: if both wakeup and sleep happen under
> > lock_kernel.
> > Any objections against killing it entirely? Or what about marking it as
> > deprecated, as the first step towards killing it?
> > 
> > I'll follow with two patches that remove it from shaper and sunrpc/sched
> > - shaper is racy, rpciod_down is only safe if called with lock_kernel.
> 
> Deprecate it in 2.7.0 and add BUG_ON(BKL not held) to it. Not before
> time.

We need to remove racy uses anyway - that can't wait for 2.7.  And I really
wonder if there will be anything left after that - right now only reiserfs
uses look like something that might be not immediately broken.

AFAICS, _all_ uses of sleep_on() in drivers are broken in one way or another
and BKL won't help them.