fix-mmc-busy-loop-on-bytes.patch

christer-mmc-byte-alignment.patch reduced the FIFO I/O granularity
from words to bytes. This also includes the decision when the FIFO
is empty or full.

However, we sometimes only want to transfer full words, in which
case do_pio_read/do_pio_write busy-loop until the FIFO has filled
up or drained enough.

In the case of do_pio_write, this can cause an endless loop if the
amount of data exceeds the FIFO size, because do_pio_write runs
before the transfer is initiated, so the FIFO never drains.

Signed-off-by: Werner Almesberger <werner@openmoko.org>

1 files changed, 8 insertions, 2 deletions

diff --git a/drivers/mmc/host/s3cmci.c b/drivers/mmc/host/s3cmci.c
index 2038c93cfd9..bf4df3a82f8 100644
--- a/drivers/mmc/host/s3cmci.c
+++ b/drivers/mmc/host/s3cmci.c
@@ -289,8 +289,11 @@ static void do_pio_read(struct s3cmci_host *host)
 		 * an even multiple of 4. */
 		if (fifo >= host->pio_bytes)
 			fifo = host->pio_bytes;
-		else
+		else {
 			fifo -= fifo & 3;
+			if (!fifo)
+				break;
+		}
 
 		host->pio_bytes -= fifo;
 		host->pio_count += fifo;
@@ -362,8 +365,11 @@ static void do_pio_write(struct s3cmci_host *host)
 		 * words, so round down to an even multiple of 4. */
 		if (fifo >= host->pio_bytes)
 			fifo = host->pio_bytes;
-		else
+		else {
 			fifo -= fifo & 3;
+			if (!fifo)
+				break;
+		}
 
 		host->pio_bytes -= fifo;
 		host->pio_count += fifo;