spi: move bitbang txrx utility functions to private header

A number of files in drivers/spi fail checkincludes.pl due to the double
include of <linux/spi/spi_bitbang.h>.

The first include is needed to get the struct spi_bitbang definition and
the spi_bitbang_* function prototypes.

The second include happens after defining EXPAND_BITBANG_TXRX to get the
inlined bitbang_txrx_* utility functions.

The <linux/spi/spi_bitbang.h> header is also included by a number of other
spi drivers, as well as some arch/ code, in order to use struct spi_bitbang
and the associated functions.

To fix the double include, and remove any potential confusion about it, move
the inlined bitbang_txrx_* functions to a new private header in drivers/spi
and also remove the need to define EXPAND_BITBANG_TXRX.

Signed-off-by: H Hartley Sweeten <hsweeten@visionengravers.com>
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
diff --git a/drivers/spi/spi_bitbang_txrx.h b/drivers/spi/spi_bitbang_txrx.h
new file mode 100644
index 0000000..fc033bb
--- /dev/null
+++ b/drivers/spi/spi_bitbang_txrx.h
@@ -0,0 +1,93 @@
+/*
+ * Mix this utility code with some glue code to get one of several types of
+ * simple SPI master driver.  Two do polled word-at-a-time I/O:
+ *
+ *   -	GPIO/parport bitbangers.  Provide chipselect() and txrx_word[](),
+ *	expanding the per-word routines from the inline templates below.
+ *
+ *   -	Drivers for controllers resembling bare shift registers.  Provide
+ *	chipselect() and txrx_word[](), with custom setup()/cleanup() methods
+ *	that use your controller's clock and chipselect registers.
+ *
+ * Some hardware works well with requests at spi_transfer scope:
+ *
+ *   -	Drivers leveraging smarter hardware, with fifos or DMA; or for half
+ *	duplex (MicroWire) controllers.  Provide chipselect() and txrx_bufs(),
+ *	and custom setup()/cleanup() methods.
+ */
+
+/*
+ * The code that knows what GPIO pins do what should have declared four
+ * functions, ideally as inlines, before including this header:
+ *
+ *  void setsck(struct spi_device *, int is_on);
+ *  void setmosi(struct spi_device *, int is_on);
+ *  int getmiso(struct spi_device *);
+ *  void spidelay(unsigned);
+ *
+ * setsck()'s is_on parameter is a zero/nonzero boolean.
+ *
+ * setmosi()'s is_on parameter is a zero/nonzero boolean.
+ *
+ * getmiso() is required to return 0 or 1 only. Any other value is invalid
+ * and will result in improper operation.
+ *
+ * A non-inlined routine would call bitbang_txrx_*() routines.  The
+ * main loop could easily compile down to a handful of instructions,
+ * especially if the delay is a NOP (to run at peak speed).
+ *
+ * Since this is software, the timings may not be exactly what your board's
+ * chips need ... there may be several reasons you'd need to tweak timings
+ * in these routines, not just make to make it faster or slower to match a
+ * particular CPU clock rate.
+ */
+
+static inline u32
+bitbang_txrx_be_cpha0(struct spi_device *spi,
+		unsigned nsecs, unsigned cpol,
+		u32 word, u8 bits)
+{
+	/* if (cpol == 0) this is SPI_MODE_0; else this is SPI_MODE_2 */
+
+	/* clock starts at inactive polarity */
+	for (word <<= (32 - bits); likely(bits); bits--) {
+
+		/* setup MSB (to slave) on trailing edge */
+		setmosi(spi, word & (1 << 31));
+		spidelay(nsecs);	/* T(setup) */
+
+		setsck(spi, !cpol);
+		spidelay(nsecs);
+
+		/* sample MSB (from slave) on leading edge */
+		word <<= 1;
+		word |= getmiso(spi);
+		setsck(spi, cpol);
+	}
+	return word;
+}
+
+static inline u32
+bitbang_txrx_be_cpha1(struct spi_device *spi,
+		unsigned nsecs, unsigned cpol,
+		u32 word, u8 bits)
+{
+	/* if (cpol == 0) this is SPI_MODE_1; else this is SPI_MODE_3 */
+
+	/* clock starts at inactive polarity */
+	for (word <<= (32 - bits); likely(bits); bits--) {
+
+		/* setup MSB (to slave) on leading edge */
+		setsck(spi, !cpol);
+		setmosi(spi, word & (1 << 31));
+		spidelay(nsecs); /* T(setup) */
+
+		setsck(spi, cpol);
+		spidelay(nsecs);
+
+		/* sample MSB (from slave) on trailing edge */
+		word <<= 1;
+		word |= getmiso(spi);
+	}
+	return word;
+}
diff --git a/drivers/spi/spi_butterfly.c b/drivers/spi/spi_butterfly.c
index c218486..8b52812 100644
--- a/drivers/spi/spi_butterfly.c
+++ b/drivers/spi/spi_butterfly.c
@@ -149,8 +149,7 @@
 #define	spidelay(X)	do{}while(0)
 //#define	spidelay	ndelay
 
-#define	EXPAND_BITBANG_TXRX
-#include <linux/spi/spi_bitbang.h>
+#include "spi_bitbang_txrx.h"
 
 static u32
 butterfly_txrx_word_mode0(struct spi_device *spi,
diff --git a/drivers/spi/spi_gpio.c b/drivers/spi/spi_gpio.c
index 26bd03e..7edbd58 100644
--- a/drivers/spi/spi_gpio.c
+++ b/drivers/spi/spi_gpio.c
@@ -127,8 +127,7 @@
  */
 #define spidelay(nsecs)	do {} while (0)
 
-#define	EXPAND_BITBANG_TXRX
-#include <linux/spi/spi_bitbang.h>
+#include "spi_bitbang_txrx.h"
 
 /*
  * These functions can leverage inline expansion of GPIO calls to shrink
diff --git a/drivers/spi/spi_lm70llp.c b/drivers/spi/spi_lm70llp.c
index 568c781..86fb7b5 100644
--- a/drivers/spi/spi_lm70llp.c
+++ b/drivers/spi/spi_lm70llp.c
@@ -174,8 +174,7 @@
 }
 /*--------------------------------------------------------------------*/
 
-#define EXPAND_BITBANG_TXRX 1
-#include <linux/spi/spi_bitbang.h>
+#include "spi_bitbang_txrx.h"
 
 static void lm70_chipselect(struct spi_device *spi, int value)
 {
diff --git a/drivers/spi/spi_s3c24xx_gpio.c b/drivers/spi/spi_s3c24xx_gpio.c
index bbf9371..8979a75 100644
--- a/drivers/spi/spi_s3c24xx_gpio.c
+++ b/drivers/spi/spi_s3c24xx_gpio.c
@@ -58,8 +58,7 @@
 
 #define spidelay(x) ndelay(x)
 
-#define	EXPAND_BITBANG_TXRX
-#include <linux/spi/spi_bitbang.h>
+#include "spi_bitbang_txrx.h"
 
 
 static u32 s3c2410_spigpio_txrx_mode0(struct spi_device *spi,
diff --git a/drivers/spi/spi_sh_sci.c b/drivers/spi/spi_sh_sci.c
index a65c12f..a511be7 100644
--- a/drivers/spi/spi_sh_sci.c
+++ b/drivers/spi/spi_sh_sci.c
@@ -78,8 +78,7 @@
 
 #define spidelay(x) ndelay(x)
 
-#define EXPAND_BITBANG_TXRX
-#include <linux/spi/spi_bitbang.h>
+#include "spi_bitbang_txrx.h"
 
 static u32 sh_sci_spi_txrx_mode0(struct spi_device *spi,
 				      unsigned nsecs, u32 word, u8 bits)