A spi transfer with zero length is not invalid.  For example, such
transfer (len == 0 && delay_usecs != 0) can be used to achieve delay
before first CLK edge after chipselect assertion.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
---
Though some discussion about zero length transfer were raised by this
patch last time, I think there were no explicit objection to this
patch itself.

diff --git a/drivers/spi/atmel_spi.c b/drivers/spi/atmel_spi.c
index 293b7ca..5dff5e0 100644
--- a/drivers/spi/atmel_spi.c
+++ b/drivers/spi/atmel_spi.c
@@ -606,7 +606,7 @@ static int atmel_spi_transfer(struct spi_device *spi, struct spi_message *msg)
 		return -ESHUTDOWN;
 
 	list_for_each_entry(xfer, &msg->transfers, transfer_list) {
-		if (!(xfer->tx_buf || xfer->rx_buf)) {
+		if (!(xfer->tx_buf || xfer->rx_buf) && xfer->len) {
 			dev_dbg(&spi->dev, "missing rx or tx buf\n");
 			return -EINVAL;
 		}