diff --git a/arch/blackfin/include/asm/dma.h b/arch/blackfin/include/asm/dma.h
index 2c09b1d50ec9..eedf3ca65ba2 100644
--- a/arch/blackfin/include/asm/dma.h
+++ b/arch/blackfin/include/asm/dma.h
@@ -276,6 +276,7 @@ static inline void clear_dma_irqstat(unsigned int channel)
 }
 
 void *dma_memcpy(void *dest, const void *src, size_t count);
+void *dma_memcpy_nocache(void *dest, const void *src, size_t count);
 void *safe_dma_memcpy(void *dest, const void *src, size_t count);
 void blackfin_dma_early_init(void);
 void early_dma_memcpy(void *dest, const void *src, size_t count);
diff --git a/arch/blackfin/kernel/bfin_dma_5xx.c b/arch/blackfin/kernel/bfin_dma_5xx.c
index 26403d1c9e65..1e485dfdc9f2 100644
--- a/arch/blackfin/kernel/bfin_dma_5xx.c
+++ b/arch/blackfin/kernel/bfin_dma_5xx.c
@@ -450,7 +450,6 @@ void *dma_memcpy(void *pdst, const void *psrc, size_t size)
 {
 	unsigned long dst = (unsigned long)pdst;
 	unsigned long src = (unsigned long)psrc;
-	size_t bulk, rest;
 
 	if (bfin_addr_dcacheable(src))
 		blackfin_dcache_flush_range(src, src + size);
@@ -458,6 +457,22 @@ void *dma_memcpy(void *pdst, const void *psrc, size_t size)
 	if (bfin_addr_dcacheable(dst))
 		blackfin_dcache_invalidate_range(dst, dst + size);
 
+	return dma_memcpy_nocache(pdst, psrc, size);
+}
+EXPORT_SYMBOL(dma_memcpy);
+
+/**
+ *	dma_memcpy_nocache - DMA memcpy under mutex lock
+ *	- No cache flush/invalidate
+ *
+ * Do not check arguments before starting the DMA memcpy.  Break the transfer
+ * up into two pieces.  The first transfer is in multiples of 64k and the
+ * second transfer is the piece smaller than 64k.
+ */
+void *dma_memcpy_nocache(void *pdst, const void *psrc, size_t size)
+{
+	size_t bulk, rest;
+
 	bulk = size & ~0xffff;
 	rest = size - bulk;
 	if (bulk)
@@ -465,7 +480,7 @@ void *dma_memcpy(void *pdst, const void *psrc, size_t size)
 	_dma_memcpy(pdst + bulk, psrc + bulk, rest);
 	return pdst;
 }
-EXPORT_SYMBOL(dma_memcpy);
+EXPORT_SYMBOL(dma_memcpy_nocache);
 
 /**
  *	safe_dma_memcpy - DMA memcpy w/argument checking
diff --git a/arch/blackfin/mach-common/pm.c b/arch/blackfin/mach-common/pm.c
index ea7f95f6bb4c..09c1fb410748 100644
--- a/arch/blackfin/mach-common/pm.c
+++ b/arch/blackfin/mach-common/pm.c
@@ -61,10 +61,11 @@ void bfin_pm_suspend_standby_enter(void)
 
 int bf53x_suspend_l1_mem(unsigned char *memptr)
 {
-	dma_memcpy(memptr, (const void *) L1_CODE_START, L1_CODE_LENGTH);
-	dma_memcpy(memptr + L1_CODE_LENGTH, (const void *) L1_DATA_A_START,
-			L1_DATA_A_LENGTH);
-	dma_memcpy(memptr + L1_CODE_LENGTH + L1_DATA_A_LENGTH,
+	dma_memcpy_nocache(memptr, (const void *) L1_CODE_START,
+			L1_CODE_LENGTH);
+	dma_memcpy_nocache(memptr + L1_CODE_LENGTH,
+			(const void *) L1_DATA_A_START, L1_DATA_A_LENGTH);
+	dma_memcpy_nocache(memptr + L1_CODE_LENGTH + L1_DATA_A_LENGTH,
 			(const void *) L1_DATA_B_START, L1_DATA_B_LENGTH);
 	memcpy(memptr + L1_CODE_LENGTH + L1_DATA_A_LENGTH +
 			L1_DATA_B_LENGTH, (const void *) L1_SCRATCH_START,
@@ -75,10 +76,10 @@ int bf53x_suspend_l1_mem(unsigned char *memptr)
 
 int bf53x_resume_l1_mem(unsigned char *memptr)
 {
-	dma_memcpy((void *) L1_CODE_START, memptr, L1_CODE_LENGTH);
-	dma_memcpy((void *) L1_DATA_A_START, memptr + L1_CODE_LENGTH,
+	dma_memcpy_nocache((void *) L1_CODE_START, memptr, L1_CODE_LENGTH);
+	dma_memcpy_nocache((void *) L1_DATA_A_START, memptr + L1_CODE_LENGTH,
 			L1_DATA_A_LENGTH);
-	dma_memcpy((void *) L1_DATA_B_START, memptr + L1_CODE_LENGTH +
+	dma_memcpy_nocache((void *) L1_DATA_B_START, memptr + L1_CODE_LENGTH +
 			L1_DATA_A_LENGTH, L1_DATA_B_LENGTH);
 	memcpy((void *) L1_SCRATCH_START, memptr + L1_CODE_LENGTH +
 			L1_DATA_A_LENGTH + L1_DATA_B_LENGTH, L1_SCRATCH_LENGTH);
@@ -167,7 +168,7 @@ int bfin_pm_suspend_mem_enter(void)
 	_disable_icplb();
 	bf53x_suspend_l1_mem(memptr);
 
-	do_hibernate(wakeup | vr_wakeup);	/* Goodbye */
+	do_hibernate(wakeup | vr_wakeup);	/* See you later! */
 
 	bf53x_resume_l1_mem(memptr);