Skip to contentSkip to conversation

omap_uwire byteswap bugfix

PATCH In review David Brownell 2 participants 4 messages May 3, 2006 – May 4, 2006 SPI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 omap_uwire byteswap bugfix

I mentioned a while back that my original omap_uwire code had some bugs that came from a bogus optimization ... here's a patch fixing them. (Please merge...)

The consequence of that bogus optimization is that some SPI protocol drivers (layered on top of this driver) would get data in strange byte orders. Hence for example an ads7846 issue that Juha and Imre noticed, but were trying to fix at the wrong level. (So in a moment I'll post a separate patch for that driver.)

Other drivers layered on top of omap_uwire may also need to revisit how they've handled byte order.

- Dave

Get rid of broken optimization in MicroWire driver: don't try to morph consecutive single-byte operations into one (faster) two-byte operation. This resolves some byteswap problems. (And consequently allows fixing some bugs in at least the OMAP version of the ads7846 driver...)

Also, reject LSB-first device modes; this controller doesn't support them.

drivers/spi/omap_uwire.c +19 -16
Index: osk/drivers/spi/omap_uwire.c
===================================================================
@@ -196,6 +196,9 @@ static int uwire_txrx(struct spi_device 
 	if (t->tx_buf && t->rx_buf)
 		return -EPERM;
 
+	if (!bits)
+		bits = 8;
+
 	w = spi->chip_select << 10;
 	w |= CS_CMD;
 
@@ -206,18 +209,15 @@ static int uwire_txrx(struct spi_device 
 
 		/* write one or two bytes at a time */
 		while (len >= 1) {
-			/* tx is msb-aligned */
+			/* tx bit 15 is first sent; we byteswap multibyte words
+			 * (msb-first) on the way out from memory.
+			 */
 			val = *buf++;
-			if (len > 1 && (!bits || bits > 8)) {
-				if (!bits)
-					bits = 16;
+			if (bits > 8) {
 				bytes = 2;
 				val |= *buf++ << 8;
-			} else {
-				if (!bits || bits > 8)
-					bits = 8;
+			} else
 				bytes = 1;
-			}
 			val <<= 16 - bits;
 
 #ifdef	VERBOSE
@@ -253,15 +253,10 @@ static int uwire_txrx(struct spi_device 
 
 		/* read one or two bytes at a time */
 		while (len) {
-			if (len > 1 && (!bits || bits > 8)) {
-				if (!bits)
-					bits = 16;
+			if (bits > 8) {
 				bytes = 2;
-			} else {
-				if (!bits || bits > 8)
-					bits = 8;
+			} else
 				bytes = 1;
-			}
 
 			/* start read */
 			val = START | w | (bits << 0);
@@ -275,7 +270,9 @@ static int uwire_txrx(struct spi_device 
 						RDRB, 0))
 				goto eio;
 
-			/* rx is lsb-aligned */
+			/* rx bit 0 is last received; multibyte words will
+			 * be properly byteswapped on the way to memory.
+			 */
 			val = uwire_read_reg(UWIRE_RDR);
 			val &= (1 << bits) - 1;
 			*buf++ = (u8) val;
@@ -320,6 +317,12 @@ static int uwire_setup(struct spi_device
 		goto done;
 	}
 
+	if (spi->mode & SPI_LSB_FIRST) {
+		pr_debug("%s: lsb first?\n", spi->dev.bus_id);
+		status = -EINVAL;
+		goto done;
+	}
+
 	/* mode 0..3, clock inverted separately;
 	 * standard nCS signaling;
 	 * don't treat DI=high as "not ready"
#1 lore raw
Quoted 11 lines · David Brownell

Also, reject LSB-first device modes; this controller doesn't support them.

Index: osk/drivers/spi/omap_uwire.c
===================================================================

drivers/spi/omap_uwire.c +0 -0
@@ -320,6 +317,12 @@ static int uwire_setup(struct spi_device
 		goto done;
 	}

drivers/spi/omap_uwire.c static int uwire_setup(struct spi_device · line 328

Quoted 10 lines
drivers/spi/omap_uwire.c +6 -0
 
+	if (spi->mode & SPI_LSB_FIRST) {
+		pr_debug("%s: lsb first?\n", spi->dev.bus_id);
+		status = -EINVAL;
+		goto done;
+	}
+
 	/* mode 0..3, clock inverted separately;
 	 * standard nCS signaling;
 	 * don't treat DI=high as "not ready"

This breaks the build, SPI_LSB_FIRST isn't defined anywhere..

drivers/spi/omap_uwire.c: In function `uwire_setup':
drivers/spi/omap_uwire.c:320: error: `SPI_LSB_FIRST' undeclared (first use in this function) drivers/spi/omap_uwire.c:320: error: (Each undeclared identifier is reported only once drivers/spi/omap_uwire.c:320: error: for each function it appears in.) make[2]: *** [drivers/spi/omap_uwire.o] Error 1
make[1]: *** [drivers/spi] Error 2
make: *** [drivers] Error 2

Looks like this is missing..

http://www.kernel.org/pub/linux/kernel/people/gregkh/gregkh-2.6/gregkh-01-driver/spi-devices-can-require-lsb-first-encodings.patch

which isn't in Linus's tree either yet however.

#2 lore raw
Quoted 14 lines · Paul Mundt

drivers/spi/omap_uwire.c static int uwire_setup(struct spi_device · line 328

Quoted 9 lines
drivers/spi/omap_uwire.c +6 -0
+	if (spi->mode & SPI_LSB_FIRST) {
+		pr_debug("%s: lsb first?\n", spi->dev.bus_id);
+		status = -EINVAL;
+		goto done;
+	}
+
 	/* mode 0..3, clock inverted separately;
 	 * standard nCS signaling;
 	 * don't treat DI=high as "not ready"

This breaks the build, SPI_LSB_FIRST isn't defined anywhere..
...
Looks like this is missing..

Quoted 3 lines

http://www.kernel.org/pub/linux/kernel/people/gregkh/gregkh-2.6/gregkh-01-driver/spi-devices-can-require-lsb-first-encodings.patch

which isn't in Linus's tree either yet however.

Whoops, yes. My bad, sorry; it was in _my_ tree and I overlooked how it got there. I thought maybe Juha had merged it.

Simplest fix would be to merge that patch from Greg's tree.

- Dave

#3 lore raw
Quoted 6 lines · David Brownell
Quoted 4 lines · Paul Mundt

This breaks the build, SPI_LSB_FIRST isn't defined anywhere..
...
Looks like this is missing..

Quoted 8 lines
Quoted 3 lines

http://www.kernel.org/pub/linux/kernel/people/gregkh/gregkh-2.6/gregkh-01-driver/spi-devices-can-require-lsb-first-encodings.patch

which isn't in Linus's tree either yet however.

Whoops, yes. My bad, sorry; it was in _my_ tree and I overlooked how it got there. I thought maybe Juha had merged it.

Simplest fix would be to merge that patch from Greg's tree.

I take that back. Simplest fix is to remove that code fragment, since an equivalent one -- in the spi_bitbang infrastructure -- is found in that nyet-merged patch.

- Dave

Build fix ... the LSB_FIRST option isn't yet merged, and in any case this driver doesn't need it since the LSB_FIRST patch handles it in the spi_bitbang infrastructure used by omap_uwire.

drivers/spi/omap_uwire.c +0 -6
Index: osk/drivers/spi/omap_uwire.c
===================================================================
@@ -317,12 +317,6 @@ static int uwire_setup(struct spi_device
 		goto done;
 	}
 
-	if (spi->mode & SPI_LSB_FIRST) {
-		pr_debug("%s: lsb first?\n", spi->dev.bus_id);
-		status = -EINVAL;
-		goto done;
-	}
-
 	/* mode 0..3, clock inverted separately;
 	 * standard nCS signaling;
 	 * don't treat DI=high as "not ready"