in 2.6.18-rc5
PATCH 1/1 in 2.6.18-rc5
Hi,
Some time ago I wrote about this bug in rc3.
I see it's still in 2.6.18-rc5 so here is what
I've found: with config like this:
CONFIG_SMP=yCONFIG_PREEMPT=yCONFIG_LOCKDEP=yCONFIG_DEBUG_LOCK_ALLOC=y
# CONFIG_PROVE_LOCKING is not set
spin_unlock_irqrestore() goes through lockdep
but spin_lock_irqrestore() doesn't.
I attach my proposal how to fix this.
Jarek P.
kernel/spinlock.c +1 -1
@@ -72,7 +72,7 @@* not re-enabled during lock-acquire (which the preempt-spin-ops do): */ #if !defined(CONFIG_PREEMPT ) || !defined(CONFIG_SMP ) || \ - defined(CONFIG_PROVE_LOCKING ) + defined(CONFIG_DEBUG_LOCK_ALLOC ) void __lockfunc _read_lock(rwlock_t *lock) {
On Fri, Sep 01, 2006 at 08:43:19AM +0200, Jarek Poplawski wrote:
...
Quoted 2 lines
spin_unlock_irqrestore()goes through lockdep
butspin_lock_irqrestore()doesn't.
but spin_lock_irqsave() doesn't!
Jarek P.
Quoted 2 lines · Jarek Poplawski
spin_unlock_irqrestore()goes through lockdep butspin_lock_irqrestore()doesn't.
good catch! Andrew: i think this is a must-have for 2.6.18.
Acked-by Ingo Molnar <[email protected]>
Ingo