spi: sirf: avoid uninitialized-use warning

24778be20 "spi: convert drivers to use bits_per_word_mask" removed the "default" statement in the spi_sirfsoc_setup_transfer switch, causing a new warning:

drivers/spi/spi-sirf.c: In function 'spi_sirfsoc_setup_transfer':
arch/arm/include/asm/io.h:90:2: warning: 'rxfifo_ctrl' may be used uninitialized in this function [-Wmaybe-uninitialized]
asm volatile("str %1, %0"
^
drivers/spi/spi-sirf.c:364:19: note: 'rxfifo_ctrl' was declared here

  u32 txfifo_ctrl, rxfifo_ctrl;
                   ^

The compiler has correctly identified that this case may happen, but since we know that things are horribly broken if bits_per_word is ever something other than the values we tested, calling BUG() is an appropriate action and tells the compiler that execution will not continue afterwards.

---
I keep using the wrong address for Mark even though I know it doesn't work any more. Sorry for the duplicate mail.

index 265a5c6..fc20bcf 100644
@@ -409,6 +409,8 @@ spi_sirfsoc_setup_transfer(struct spi_device *spi, struct spi_transfer *t)
 		rxfifo_ctrl = SIRFSOC_SPI_FIFO_THD(SIRFSOC_SPI_FIFO_SIZE / 2) |
 					SIRFSOC_SPI_FIFO_WIDTH_DWORD;
 		break;
+	default:
+		BUG();
 	}
 
 	if (!(spi->mode & SPI_CS_HIGH))

I guess this is fine, although BUG() seems a little scary, even though it is a bug.

In some of the other drivers, I converted:

if () {
} else if () {
} else {
error
}

... to:

if () {
} else {
}

... the equivalent of which would be to just add a "default:" to, say, the 8-bpw case in the switch. But I guess I don't feel too strongly either way.