Blackfin: rewrite cache handling functions

Take the cache flush functions from the kernel as they use hardware loops in
order to get optimal performance.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>
diff --git a/cpu/blackfin/cache.S b/cpu/blackfin/cache.S
index 9facadf..6ed655a 100644
--- a/cpu/blackfin/cache.S
+++ b/cpu/blackfin/cache.S
@@ -1,5 +1,10 @@
-/* cache.S - low level cache handling routines
- * Copyright (C) 2003-2007 Analog Devices Inc.
+/*
+ * Blackfin cache control code
+ *
+ * Copyright 2003-2008 Analog Devices Inc.
+ *
+ * Enter bugs at http://blackfin.uclinux.org/
+ *
  * Licensed under the GPL-2 or later.
  */
 
@@ -8,54 +13,75 @@
 #include <asm/blackfin.h>
 
 .text
-.align 2
-ENTRY(_blackfin_icache_flush_range)
-	R2 = -32;
-	R2 = R0 & R2;
-	P0 = R2;
-	P1 = R1;
-	CSYNC;
+/* Since all L1 caches work the same way, we use the same method for flushing
+ * them.  Only the actual flush instruction differs.  We write this in asm as
+ * GCC can be hard to coax into writing nice hardware loops.
+ *
+ * Also, we assume the following register setup:
+ * R0 = start address
+ * R1 = end address
+ */
+.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;
+	R1 = R1 & R2;
+	R1 += L1_CACHE_BYTES;
+
+	/* count = (end - start) >> L1_CACHE_SHIFT */
+	R2 = R1 - R0;
+	R2 >>= L1_CACHE_SHIFT;
+	P1 = R2;
+
+.ifnb \label
+\label :
+.endif
+	P0 = R0;
+	LSETUP (1f, 2f) LC1 = P1;
 1:
-	IFLUSH[P0++];
-	CC = P0 < P1(iu);
-	IF CC JUMP 1b(bp);
-	IFLUSH[P0];
-	SSYNC;
+.ifnb \optflushins
+	\optflushins [P0];
+.endif
+#if ANOMALY_05000443
+.ifb \optnopins
+2:
+.endif
+	\flushins [P0++];
+.ifnb \optnopins
+2:	\optnopins;
+.endif
+#else
+2:	\flushins [P0++];
+#endif
+
 	RTS;
+.endm
+
+/* Invalidate all instruction cache lines assocoiated with this memory area */
+ENTRY(_blackfin_icache_flush_range)
+	do_flush IFLUSH, , nop
 ENDPROC(_blackfin_icache_flush_range)
 
-ENTRY(_blackfin_dcache_flush_range)
-	R2 = -32;
-	R2 = R0 & R2;
-	P0 = R2;
-	P1 = R1;
-	CSYNC;
-1:
-	FLUSH[P0++];
-	CC = P0 < P1(iu);
-	IF CC JUMP 1b(bp);
-	FLUSH[P0];
-	SSYNC;
-	RTS;
-ENDPROC(_blackfin_dcache_flush_range)
+/* Flush all cache lines assocoiated with this area of memory. */
+ENTRY(_blackfin_icache_dcache_flush_range)
+	do_flush FLUSH, IFLUSH
+ENDPROC(_blackfin_icache_dcache_flush_range)
 
+/* Throw away all D-cached data in specified region without any obligation to
+ * write them back.  Since the Blackfin ISA does not have an "invalidate"
+ * instruction, we use flush/invalidate.  Perhaps as a speed optimization we
+ * could bang on the DTEST MMRs ...
+ */
 ENTRY(_blackfin_dcache_flush_invalidate_range)
-	R2 = -32;
-	R2 = R0 & R2;
-	P0 = R2;
-	P1 = R1;
-	CSYNC;
-1:
-	FLUSHINV[P0++];
-	CC = P0 < P1(iu);
-	IF CC JUMP 1b(bp);
-
-	/*
-	 * If the data crosses a cache line, then we'll be pointing to
-	 * the last cache line, but won't have flushed/invalidated it yet, so do
-	 * one more.
-	 */
-	FLUSHINV[P0];
-	SSYNC;
-	RTS;
+	do_flush FLUSHINV
 ENDPROC(_blackfin_dcache_flush_invalidate_range)
+
+/* Flush all data cache lines assocoiated with this memory area */
+ENTRY(_blackfin_dcache_flush_range)
+	do_flush FLUSH, , , .Ldfr
+ENDPROC(_blackfin_dcache_flush_range)
diff --git a/include/asm-blackfin/blackfin_local.h b/include/asm-blackfin/blackfin_local.h
index 90259ba..4e15834 100644
--- a/include/asm-blackfin/blackfin_local.h
+++ b/include/asm-blackfin/blackfin_local.h
@@ -43,6 +43,9 @@
 #define SCLK_TO_MSEC(sclk) ((MSEC_PER_SEC * ((sclk) / USEC_PER_MSEC)) / (BFIN_SCLK / USEC_PER_MSEC))
 #define MSEC_TO_SCLK(msec) ((((BFIN_SCLK / USEC_PER_MSEC) * (msec)) / MSEC_PER_SEC) * USEC_PER_MSEC)
 
+#define L1_CACHE_SHIFT 5
+#define L1_CACHE_BYTES (1 << L1_CACHE_SHIFT)
+
 #include <asm/linkage.h>
 
 #ifndef __ASSEMBLY__
@@ -60,6 +63,7 @@
 
 extern void blackfin_icache_flush_range(const void *, const void *);
 extern void blackfin_dcache_flush_range(const void *, const void *);
+extern void blackfin_icache_dcache_flush_range(const void *, const void *);
 extern void blackfin_dcache_flush_invalidate_range(const void *, const void *);
 
 /* Use DMA to move data from on chip to external memory.  While this is
diff --git a/lib_blackfin/cache.c b/lib_blackfin/cache.c
index 870c5bf..1557864 100644
--- a/lib_blackfin/cache.c
+++ b/lib_blackfin/cache.c
@@ -15,15 +15,25 @@
 
 void flush_cache(unsigned long addr, unsigned long size)
 {
+	void *start_addr, *end_addr;
+	int istatus, dstatus;
+
 	/* no need to flush stuff in on chip memory (L1/L2/etc...) */
 	if (addr >= 0xE0000000)
 		return;
 
-	if (icache_status())
-		blackfin_icache_flush_range((void *)addr, (void *)(addr + size));
+	start_addr = (void *)addr;
+	end_addr = (void *)(addr + size);
+	istatus = icache_status();
+	dstatus = dcache_status();
 
-	if (dcache_status())
-		blackfin_dcache_flush_range((void *)addr, (void *)(addr + size));
+	if (istatus) {
+		if (dstatus)
+			blackfin_icache_dcache_flush_range(start_addr, end_addr);
+		else
+			blackfin_icache_flush_range(start_addr, end_addr);
+	} else if (dstatus)
+		blackfin_dcache_flush_range(start_addr, end_addr);
 }
 
 void icache_enable(void)