Re: [openib-general] ipoib lockdep warning

[Roland Dreier <rdreier@cisco.com>]

 > > So, ugh... maybe the best thing to do is change lib/idr.c to use
 > > spin_lock_irqsave() internally?
 > 
 > I dunno, it seems to have had _irq() locking in the past?  From the
 > comment at the top:
 > 
 >  * Modified by George Anzinger to reuse immediately and to use
 >  * find bit instructions.  Also removed _irq on spinlocks.

Well, _irq would be no good, because we might want to call idr stuff
with interrupts disabled.  But making idr internally _irqsave seems
like the right fix to me.

I think the real issue here is that the sa_query.c stuff wants to use
the idr mechanism to assign "query ids", and other modules want to be
able to start queries from any context.  So if idr uses bare spin_lock
internally, then sa_query.c has no choice but to wrap all idr calls
inside spin_lock_irqsave and do all allocation with GFP_ATOMIC, which
doesn't seem very nice.

 - R.