spi/fsl-espi: fix rx_buf in fsl_espi_cmd_trans()/fsl_espi_rw_trans()
authorValentin Longchamp <valentin.longchamp@keymile.com>
Fri, 16 May 2014 14:46:21 +0000 (16:46 +0200)
committerMark Brown <broonie@linaro.org>
Mon, 26 May 2014 15:56:45 +0000 (16:56 +0100)
By default for every espi transfer, the rx_buf is placed right after the
tx_buf. This can lead to a buffer overflow when the size of both the TX
and RX data cumulated is higher than the allocated 64K buffer for the
transfer (this is the case when sending for instance a read command and
reading 64K back, please see:
http://article.gmane.org/gmane.linux.drivers.mtd/53411 )

This gets fixed by always setting the RX buffer pointer at the begining
of the transfer buffer.

[The driver shouldn't be doing the copy in the first place and instead
sending directly from the supplied buffer but this is at least not worse
than what's there -- broonie]

Signed-off-by: Valentin Longchamp <valentin.longchamp@keymile.com>
Signed-off-by: Mark Brown <broonie@linaro.org>
drivers/spi/spi-fsl-espi.c

index 24096c8..f0d7662 100644 (file)
@@ -348,7 +348,7 @@ static void fsl_espi_cmd_trans(struct spi_message *m,
        }
 
        espi_trans->tx_buf = local_buf;
-       espi_trans->rx_buf = local_buf + espi_trans->n_tx;
+       espi_trans->rx_buf = local_buf;
        fsl_espi_do_trans(m, espi_trans);
 
        espi_trans->actual_length = espi_trans->len;
@@ -397,7 +397,7 @@ static void fsl_espi_rw_trans(struct spi_message *m,
                espi_trans->n_rx = trans_len;
                espi_trans->len = trans_len + n_tx;
                espi_trans->tx_buf = local_buf;
-               espi_trans->rx_buf = local_buf + n_tx;
+               espi_trans->rx_buf = local_buf;
                fsl_espi_do_trans(m, espi_trans);
 
                memcpy(rx_buf + pos, espi_trans->rx_buf + n_tx, trans_len);