commit
2576c28e3f623ed401db7e6197241865328620ef upstream.
- arch_spin_lock/unlock were lacking the ACQUIRE/RELEASE barriers
Since ARCv2 only provides load/load, store/store and all/all, we need
the full barrier
- LLOCK/SCOND based atomics, bitops, cmpxchg, which return modified
values were lacking the explicit smp barriers.
- Non LLOCK/SCOND varaints don't need the explicit barriers since that
is implicity provided by the spin locks used to implement the
critical section (the spin lock barriers in turn are also fixed in
this commit as explained above
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
{ \
unsigned int temp; \
\
+ /* \
+ * Explicit full memory barrier needed before/after as \
+ * LLOCK/SCOND thmeselves don't provide any such semantics \
+ */ \
+ smp_mb(); \
+ \
__asm__ __volatile__( \
"1: llock %0, [%1] \n" \
" " #asm_op " %0, %0, %2 \n" \
: "r"(&v->counter), "ir"(i) \
: "cc"); \
\
+ smp_mb(); \
+ \
return temp; \
}
unsigned long flags; \
unsigned long temp; \
\
+ /* \
+ * spin lock/unlock provides the needed smp_mb() before/after \
+ */ \
atomic_ops_lock(flags); \
temp = v->counter; \
temp c_op i; \
#define __atomic_add_unless(v, a, u) \
({ \
int c, old; \
+ \
+ /* \
+ * Explicit full memory barrier needed before/after as \
+ * LLOCK/SCOND thmeselves don't provide any such semantics \
+ */ \
+ smp_mb(); \
+ \
c = atomic_read(v); \
while (c != (u) && (old = atomic_cmpxchg((v), c, c + (a))) != c)\
c = old; \
+ \
+ smp_mb(); \
+ \
c; \
})
if (__builtin_constant_p(nr))
nr &= 0x1f;
+ /*
+ * Explicit full memory barrier needed before/after as
+ * LLOCK/SCOND themselves don't provide any such semantics
+ */
+ smp_mb();
+
__asm__ __volatile__(
"1: llock %0, [%2] \n"
" bset %1, %0, %3 \n"
: "r"(m), "ir"(nr)
: "cc");
+ smp_mb();
+
return (old & (1 << nr)) != 0;
}
if (__builtin_constant_p(nr))
nr &= 0x1f;
+ smp_mb();
+
__asm__ __volatile__(
"1: llock %0, [%2] \n"
" bclr %1, %0, %3 \n"
: "r"(m), "ir"(nr)
: "cc");
+ smp_mb();
+
return (old & (1 << nr)) != 0;
}
if (__builtin_constant_p(nr))
nr &= 0x1f;
+ smp_mb();
+
__asm__ __volatile__(
"1: llock %0, [%2] \n"
" bxor %1, %0, %3 \n"
: "r"(m), "ir"(nr)
: "cc");
+ smp_mb();
+
return (old & (1 << nr)) != 0;
}
if (__builtin_constant_p(nr))
nr &= 0x1f;
+ /*
+ * spin lock/unlock provide the needed smp_mb() before/after
+ */
bitops_lock(flags);
old = *m;
#define __ASM_ARC_CMPXCHG_H
#include <linux/types.h>
+
+#include <asm/barrier.h>
#include <asm/smp.h>
#ifdef CONFIG_ARC_HAS_LLSC
{
unsigned long prev;
+ /*
+ * Explicit full memory barrier needed before/after as
+ * LLOCK/SCOND thmeselves don't provide any such semantics
+ */
+ smp_mb();
+
__asm__ __volatile__(
"1: llock %0, [%1] \n"
" brne %0, %2, 2f \n"
"r"(new) /* can't be "ir". scond can't take limm for "b" */
: "cc");
+ smp_mb();
+
return prev;
}
int prev;
volatile unsigned long *p = ptr;
+ /*
+ * spin lock/unlock provide the needed smp_mb() before/after
+ */
atomic_ops_lock(flags);
prev = *p;
if (prev == expected)
switch (size) {
case 4:
+ smp_mb();
+
__asm__ __volatile__(
" ex %0, [%1] \n"
: "+r"(val)
: "r"(ptr)
: "memory");
+ smp_mb();
+
return val;
}
return __xchg_bad_pointer();
{
unsigned int tmp = __ARCH_SPIN_LOCK_LOCKED__;
+ /*
+ * This smp_mb() is technically superfluous, we only need the one
+ * after the lock for providing the ACQUIRE semantics.
+ * However doing the "right" thing was regressing hackbench
+ * so keeping this, pending further investigation
+ */
+ smp_mb();
+
__asm__ __volatile__(
"1: ex %0, [%1] \n"
" breq %0, %2, 1b \n"
: "+&r" (tmp)
: "r"(&(lock->slock)), "ir"(__ARCH_SPIN_LOCK_LOCKED__)
: "memory");
+
+ /*
+ * ACQUIRE barrier to ensure load/store after taking the lock
+ * don't "bleed-up" out of the critical section (leak-in is allowed)
+ * http://www.spinics.net/lists/kernel/msg2010409.html
+ *
+ * ARCv2 only has load-load, store-store and all-all barrier
+ * thus need the full all-all barrier
+ */
+ smp_mb();
}
static inline int arch_spin_trylock(arch_spinlock_t *lock)
{
unsigned int tmp = __ARCH_SPIN_LOCK_LOCKED__;
+ smp_mb();
+
__asm__ __volatile__(
"1: ex %0, [%1] \n"
: "+r" (tmp)
: "r"(&(lock->slock))
: "memory");
+ smp_mb();
+
return (tmp == __ARCH_SPIN_LOCK_UNLOCKED__);
}
{
unsigned int tmp = __ARCH_SPIN_LOCK_UNLOCKED__;
+ /*
+ * RELEASE barrier: given the instructions avail on ARCv2, full barrier
+ * is the only option
+ */
+ smp_mb();
+
__asm__ __volatile__(
" ex %0, [%1] \n"
: "+r" (tmp)
: "r"(&(lock->slock))
: "memory");
+ /*
+ * superfluous, but keeping for now - see pairing version in
+ * arch_spin_lock above
+ */
smp_mb();
}