Blackfin arch: fix bug - dmacopy test case fail on all platform

The cache code I added flushes 1 line too little if the start address is
not aligned to the cache size. Cache align the start address so that when
we straddle cache aligns, we get the right count.
Signed-off-by: NMike Frysinger <vapier.adi@gmail.com>
Signed-off-by: NBryan Wu <cooloney@kernel.org>

arch/blackfin/mach-common/cache.S

@@ -25,9 +25,13 @@
  */
 .macro do_flush flushins:req optflushins optnopins label
 
+	R2 = -L1_CACHE_BYTES;
+
+	/* start = (start & -L1_CACHE_BYTES) */
+	R0 = R0 & R2;
+
 	/* end = ((end - 1) & -L1_CACHE_BYTES) + L1_CACHE_BYTES; */
 	R1 += -1;
-	R2 = -L1_CACHE_BYTES;
 	R1 = R1 & R2;
 	R1 += L1_CACHE_BYTES;