On Mon, 07 Dec 2009, John David Anglin wrote:
> > is anybody able to reproduce it? configuration is attached.
With your config, I get the same error.
I have attached a potential work around. The basic idea is to force
'a' to a specific register to avoid the spill problem in reload.
It may be a bit risky to use 'r20' but there don't appear to be
any real calls in __raw_spin_is_locked.
Dave
--
J. David Anglin dave.anglin@xxxxxxxxxxxxxx
National Research Council of Canada (613) 990-0752 (FAX: 952-6602)
diff --git a/arch/parisc/include/asm/spinlock.h b/arch/parisc/include/asm/spinlock.h
index fae03e1..ce87e4b 100644
--- a/arch/parisc/include/asm/spinlock.h
+++ b/arch/parisc/include/asm/spinlock.h
@@ -18,10 +18,9 @@ static inline int __raw_spin_is_locked(raw_spinlock_t *x)
static inline void __raw_spin_lock_flags(raw_spinlock_t *x,
unsigned long flags)
{
- volatile unsigned int *a;
+ volatile register unsigned int *a asm ("r20") = __ldcw_align(x);
mb();
- a = __ldcw_align(x);
while (__ldcw(a) == 0)
while (*a == 0)
if (flags & PSW_SM_I) {
