Re: 2.6.27-rc4: lots of 'in_atomic():1, irqs_disabled():0' with software-raid1

[Andrew Morton <akpm@linux-foundation.org>]

On Fri, 29 Aug 2008 17:36:08 +1000 Neil Brown <neilb@suse.de> wrote:

> On Thursday August 28, akpm@linux-foundation.org wrote:
> > 
> > I'm curious about how this happened.
> > 
> > afaict from reading the code, this:
> > 
> > 	rcu_read_lock();
> > 	might_sleep();
> > 	rcu_read_unlock();
> > 
> > will warn if CONFIG_PREEMPT_RCU=n but won't warn if
> > CONFIG_PREEMPT_RCU=y.  If correct, that's a nasty trap.
> > 
> > Is that what you did?
> 
> Uhm..
> $ grep PREEMPT /home/src/md-x86-64/.config
> CONFIG_PREEMPT_NONE=y
> # CONFIG_PREEMPT_VOLUNTARY is not set
> # CONFIG_PREEMPT is not set
> $ grep DEBUG_SPIN /home/src/md-x86-64/.config
> CONFIG_DEBUG_SPINLOCK=y
> CONFIG_DEBUG_SPINLOCK_SLEEP=y
> $ grep RCU /home/src/md-x86-64/.config
> CONFIG_CLASSIC_RCU=y
> # CONFIG_RCU_TORTURE_TEST is not set
> 
> This definitely doesn't report an error if I put "might_sleep" inside 
> rcu_read_lock.  I tried to figure out why but got lost.

CONFIG_DEBUG_SPINLOCK_SLEEP doesn't do much useful without
CONFIG_PREEMPT=y.  Perhaps we could improve the Kconfig situation
there.

The most fine Documentation/SubmitChecklist instructeth:

12: Has been tested with CONFIG_PREEMPT, CONFIG_DEBUG_PREEMPT,
    CONFIG_DEBUG_SLAB, CONFIG_DEBUG_PAGEALLOC, CONFIG_DEBUG_MUTEXES,
    CONFIG_DEBUG_SPINLOCK, CONFIG_DEBUG_SPINLOCK_SLEEP all simultaneously
    enabled.

(which might need updating for new goodies - it was written a while ago).