diff --git a/arch/blackfin/include/asm/system.h b/arch/blackfin/include/asm/system.h
index 19e2c7c..44bd0cc 100644
--- a/arch/blackfin/include/asm/system.h
+++ b/arch/blackfin/include/asm/system.h
@@ -19,11 +19,11 @@
  * Force strict CPU ordering.
  */
 #define nop()  __asm__ __volatile__ ("nop;\n\t" : : )
-#define mb()   __asm__ __volatile__ (""   : : : "memory")
-#define rmb()  __asm__ __volatile__ (""   : : : "memory")
-#define wmb()  __asm__ __volatile__ (""   : : : "memory")
-#define set_mb(var, value) do { (void) xchg(&var, value); } while (0)
-#define read_barrier_depends() 		do { } while(0)
+#define smp_mb()  mb()
+#define smp_rmb() rmb()
+#define smp_wmb() wmb()
+#define set_mb(var, value) do { var = value; mb(); } while (0)
+#define smp_read_barrier_depends()	read_barrier_depends()
 
 #ifdef CONFIG_SMP
 asmlinkage unsigned long __raw_xchg_1_asm(volatile void *ptr, unsigned long value);
@@ -37,16 +37,16 @@
 					unsigned long new, unsigned long old);
 
 #ifdef __ARCH_SYNC_CORE_DCACHE
-# define smp_mb()	do { barrier(); smp_check_barrier(); smp_mark_barrier(); } while (0)
-# define smp_rmb()	do { barrier(); smp_check_barrier(); } while (0)
-# define smp_wmb()	do { barrier(); smp_mark_barrier(); } while (0)
-#define smp_read_barrier_depends()	do { barrier(); smp_check_barrier(); } while (0)
-
+/* Force Core data cache coherence */
+# define mb()	do { barrier(); smp_check_barrier(); smp_mark_barrier(); } while (0)
+# define rmb()	do { barrier(); smp_check_barrier(); } while (0)
+# define wmb()	do { barrier(); smp_mark_barrier(); } while (0)
+# define read_barrier_depends()	do { barrier(); smp_check_barrier(); } while (0)
 #else
-# define smp_mb()	barrier()
-# define smp_rmb()	barrier()
-# define smp_wmb()	barrier()
-#define smp_read_barrier_depends()	barrier()
+# define mb()	barrier()
+# define rmb()	barrier()
+# define wmb()	barrier()
+# define read_barrier_depends()	do { } while (0)
 #endif
 
 static inline unsigned long __xchg(unsigned long x, volatile void *ptr,
@@ -99,10 +99,10 @@
 
 #else /* !CONFIG_SMP */
 
-#define smp_mb()	barrier()
-#define smp_rmb()	barrier()
-#define smp_wmb()	barrier()
-#define smp_read_barrier_depends()	do { } while(0)
+#define mb()	barrier()
+#define rmb()	barrier()
+#define wmb()	barrier()
+#define read_barrier_depends()	do { } while (0)
 
 struct __xchg_dummy {
 	unsigned long a[100];
diff --git a/arch/blackfin/kernel/gptimers.c b/arch/blackfin/kernel/gptimers.c
index cdbe075..8b81dc0 100644
--- a/arch/blackfin/kernel/gptimers.c
+++ b/arch/blackfin/kernel/gptimers.c
@@ -268,7 +268,7 @@
 	_disable_gptimers(mask);
 	for (i = 0; i < MAX_BLACKFIN_GPTIMERS; ++i)
 		if (mask & (1 << i))
-			group_regs[BFIN_TIMER_OCTET(i)]->status |= trun_mask[i];
+			group_regs[BFIN_TIMER_OCTET(i)]->status = trun_mask[i];
 	SSYNC();
 }
 EXPORT_SYMBOL(disable_gptimers);
diff --git a/arch/blackfin/kernel/time-ts.c b/arch/blackfin/kernel/time-ts.c
index 8c9a43d..cdb4beb 100644
--- a/arch/blackfin/kernel/time-ts.c
+++ b/arch/blackfin/kernel/time-ts.c
@@ -206,8 +206,14 @@
 {
 	struct clock_event_device *evt = dev_id;
 	smp_mb();
-	evt->event_handler(evt);
+	/*
+	 * We want to ACK before we handle so that we can handle smaller timer
+	 * intervals.  This way if the timer expires again while we're handling
+	 * things, we're more likely to see that 2nd int rather than swallowing
+	 * it by ACKing the int at the end of this handler.
+	 */
 	bfin_gptmr0_ack();
+	evt->event_handler(evt);
 	return IRQ_HANDLED;
 }
 
diff --git a/arch/blackfin/mach-common/smp.c b/arch/blackfin/mach-common/smp.c
index 6e17a26..8bce5ed 100644
--- a/arch/blackfin/mach-common/smp.c
+++ b/arch/blackfin/mach-common/smp.c
@@ -109,10 +109,23 @@
 	struct blackfin_flush_data *fdata = info;
 
 	/* Invalidate the memory holding the bounds of the flushed region. */
-	invalidate_dcache_range((unsigned long)fdata,
-		(unsigned long)fdata + sizeof(*fdata));
+	blackfin_dcache_invalidate_range((unsigned long)fdata,
+					 (unsigned long)fdata + sizeof(*fdata));
 
-	flush_icache_range(fdata->start, fdata->end);
+	/* Make sure all write buffers in the data side of the core
+	 * are flushed before trying to invalidate the icache.  This
+	 * needs to be after the data flush and before the icache
+	 * flush so that the SSYNC does the right thing in preventing
+	 * the instruction prefetcher from hitting things in cached
+	 * memory at the wrong time -- it runs much further ahead than
+	 * the pipeline.
+	 */
+	SSYNC();
+
+	/* ipi_flaush_icache is invoked by generic flush_icache_range,
+	 * so call blackfin arch icache flush directly here.
+	 */
+	blackfin_icache_flush_range(fdata->start, fdata->end);
 }
 
 static void ipi_call_function(unsigned int cpu, struct ipi_message *msg)