Blackfin: pull io funcs from linux

Some common code uses more of the io.h funcs than we currently provide, so
pull in all of the ones from the linux kernel.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>
diff --git a/lib_blackfin/Makefile b/lib_blackfin/Makefile
index cbf47f0..eebb131 100644
--- a/lib_blackfin/Makefile
+++ b/lib_blackfin/Makefile
@@ -31,10 +31,12 @@
 
 LIB	= $(obj)lib$(ARCH).a
 
+SOBJS-y	+= ins.o
 SOBJS-y	+= memcmp.o
 SOBJS-y	+= memcpy.o
 SOBJS-y	+= memmove.o
 SOBJS-y	+= memset.o
+SOBJS-y	+= outs.o
 
 COBJS-y	+= board.o
 COBJS-y	+= boot.o
diff --git a/lib_blackfin/ins.S b/lib_blackfin/ins.S
new file mode 100644
index 0000000..4519596
--- /dev/null
+++ b/lib_blackfin/ins.S
@@ -0,0 +1,117 @@
+/*
+ * arch/blackfin/lib/ins.S - ins{bwl} using hardware loops
+ *
+ * Copyright 2004-2008 Analog Devices Inc.
+ * Copyright (C) 2005 Bas Vermeulen, BuyWays BV <bas@buyways.nl>
+ * Licensed under the GPL-2 or later.
+ */
+
+#include <asm/blackfin.h>
+
+.align 2
+
+#ifdef CONFIG_IPIPE
+# define DO_CLI \
+	[--sp] = rets; \
+	[--sp] = (P5:0); \
+	sp += -12; \
+	call ___ipipe_disable_root_irqs_hw; \
+	sp += 12; \
+	(P5:0) = [sp++];
+# define CLI_INNER_NOP
+#else
+# define DO_CLI cli R3;
+# define CLI_INNER_NOP nop; nop; nop;
+#endif
+
+#ifdef CONFIG_IPIPE
+# define DO_STI \
+	sp += -12; \
+	call ___ipipe_enable_root_irqs_hw; \
+	sp += 12; \
+2:	rets = [sp++];
+#else
+# define DO_STI 2: sti R3;
+#endif
+
+#ifdef CONFIG_BFIN_INS_LOWOVERHEAD
+# define CLI_OUTER DO_CLI;
+# define STI_OUTER DO_STI;
+# define CLI_INNER 1:
+# if ANOMALY_05000416
+#  define STI_INNER nop; 2: nop;
+# else
+#  define STI_INNER 2:
+# endif
+#else
+# define CLI_OUTER
+# define STI_OUTER
+# define CLI_INNER 1: DO_CLI; CLI_INNER_NOP;
+# define STI_INNER DO_STI;
+#endif
+
+/*
+ * Reads on the Blackfin are speculative. In Blackfin terms, this means they
+ * can be interrupted at any time (even after they have been issued on to the
+ * external bus), and re-issued after the interrupt occurs.
+ *
+ * If a FIFO is sitting on the end of the read, it will see two reads,
+ * when the core only sees one. The FIFO receives the read which is cancelled,
+ * and not delivered to the core.
+ *
+ * To solve this, interrupts are turned off before reads occur to I/O space.
+ * There are 3 versions of all these functions
+ *  - turns interrupts off every read (higher overhead, but lower latency)
+ *  - turns interrupts off every loop (low overhead, but longer latency)
+ *  - DMA version, which do not suffer from this issue. DMA versions have
+ *      different name (prefixed by dma_ ), and are located in
+ *      ../kernel/bfin_dma_5xx.c
+ * Using the dma related functions are recommended for transfering large
+ * buffers in/out of FIFOs.
+ */
+
+#define COMMON_INS(func, ops) \
+ENTRY(_ins##func) \
+	P0 = R0;	/* P0 = port */ \
+	CLI_OUTER;	/* 3 instructions before first read access */ \
+	P1 = R1;	/* P1 = address */ \
+	P2 = R2;	/* P2 = count */ \
+	SSYNC; \
+ \
+	LSETUP(1f, 2f) LC0 = P2; \
+	CLI_INNER; \
+	ops; \
+	STI_INNER; \
+ \
+	STI_OUTER; \
+	RTS; \
+ENDPROC(_ins##func)
+
+COMMON_INS(l, \
+	R0 = [P0]; \
+	[P1++] = R0; \
+)
+
+COMMON_INS(w, \
+	R0 = W[P0]; \
+	W[P1++] = R0; \
+)
+
+COMMON_INS(w_8, \
+	R0 = W[P0]; \
+	B[P1++] = R0; \
+	R0 = R0 >> 8; \
+	B[P1++] = R0; \
+)
+
+COMMON_INS(b, \
+	R0 = B[P0]; \
+	B[P1++] = R0; \
+)
+
+COMMON_INS(l_16, \
+	R0 = [P0]; \
+	W[P1++] = R0; \
+	R0 = R0 >> 16; \
+	W[P1++] = R0; \
+)
diff --git a/lib_blackfin/outs.S b/lib_blackfin/outs.S
new file mode 100644
index 0000000..90c6033
--- /dev/null
+++ b/lib_blackfin/outs.S
@@ -0,0 +1,60 @@
+/*
+ * Implementation of outs{bwl} for BlackFin processors using zero overhead loops.
+ *
+ * Copyright 2005-2009 Analog Devices Inc.
+ *                2005 BuyWays BV
+ *                      Bas Vermeulen <bas@buyways.nl>
+ *
+ * Licensed under the GPL-2.
+ */
+
+#include <asm/linkage.h>
+
+.align 2
+
+ENTRY(_outsl)
+	P0 = R0;	/* P0 = port */
+	P1 = R1;	/* P1 = address */
+	P2 = R2;	/* P2 = count */
+
+	LSETUP( .Llong_loop_s, .Llong_loop_e) LC0 = P2;
+.Llong_loop_s: R0 = [P1++];
+.Llong_loop_e: [P0] = R0;
+	RTS;
+ENDPROC(_outsl)
+
+ENTRY(_outsw)
+	P0 = R0;	/* P0 = port */
+	P1 = R1;	/* P1 = address */
+	P2 = R2;	/* P2 = count */
+
+	LSETUP( .Lword_loop_s, .Lword_loop_e) LC0 = P2;
+.Lword_loop_s: R0 = W[P1++];
+.Lword_loop_e: W[P0] = R0;
+	RTS;
+ENDPROC(_outsw)
+
+ENTRY(_outsb)
+	P0 = R0;	/* P0 = port */
+	P1 = R1;	/* P1 = address */
+	P2 = R2;	/* P2 = count */
+
+	LSETUP( .Lbyte_loop_s, .Lbyte_loop_e) LC0 = P2;
+.Lbyte_loop_s: R0 = B[P1++];
+.Lbyte_loop_e: B[P0] = R0;
+	RTS;
+ENDPROC(_outsb)
+
+ENTRY(_outsw_8)
+	P0 = R0;	/* P0 = port */
+	P1 = R1;	/* P1 = address */
+	P2 = R2;	/* P2 = count */
+
+	LSETUP( .Lword8_loop_s, .Lword8_loop_e) LC0 = P2;
+.Lword8_loop_s: R1 = B[P1++];
+		R0 = B[P1++];
+		R0 = R0 << 8;
+		R0 = R0 + R1;
+.Lword8_loop_e: W[P0] = R0;
+	RTS;
+ENDPROC(_outsw_8)