diff mbox

mmc: mvsdio: Work around broken TX DMA

Message ID 1415830208-1888-1-git-send-email-andrew@lunn.ch (mailing list archive)
State New, archived
Headers show

Commit Message

Andrew Lunn Nov. 12, 2014, 10:10 p.m. UTC
In order to use the mvsdio driver for sdio, it has been necessary to
use a module parameter to disable DMA so to force PIO is used. It is
then possible to use wireless LAN devices like mwifiex found on
topkick and mirabox. However, accessing an MMC SD card does work with
DMA.

Investigation has shown that MMC block device accesses are always
aligned to 64 byte boundaries, where as transfers from mwifiex are
rarely more than word aligned. It has also been determined that card
to host transfers work with DMA for SDIO devices, but host to card
transfers with DMA have problems.

This patch extends the current checks for buffers which are not word
aligned or multiple of words. All host to card transfers which are not
64 byte aligned are now also performed via PIO. This should not affect
the performance of SD cards, but allow sdio devices to work out of the
box, and they are likely to be more efficient since DMA will be used
for card to host transfers.

Tested on mirabox for wifi via mwifiex
Tested on 370 RD for file systems on an SD card.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
---
 drivers/mmc/host/mvsdio.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

Comments

Ulf Hansson Nov. 19, 2014, 10:31 a.m. UTC | #1
On 12 November 2014 23:10, Andrew Lunn <andrew@lunn.ch> wrote:
> In order to use the mvsdio driver for sdio, it has been necessary to
> use a module parameter to disable DMA so to force PIO is used. It is
> then possible to use wireless LAN devices like mwifiex found on
> topkick and mirabox. However, accessing an MMC SD card does work with
> DMA.
>
> Investigation has shown that MMC block device accesses are always
> aligned to 64 byte boundaries, where as transfers from mwifiex are
> rarely more than word aligned. It has also been determined that card
> to host transfers work with DMA for SDIO devices, but host to card
> transfers with DMA have problems.
>
> This patch extends the current checks for buffers which are not word
> aligned or multiple of words. All host to card transfers which are not
> 64 byte aligned are now also performed via PIO. This should not affect
> the performance of SD cards, but allow sdio devices to work out of the
> box, and they are likely to be more efficient since DMA will be used
> for card to host transfers.
>
> Tested on mirabox for wifi via mwifiex
> Tested on 370 RD for file systems on an SD card.
>
> Signed-off-by: Andrew Lunn <andrew@lunn.ch>

Thanks! Applied for next.

Kind regards
Uffe

> ---
>  drivers/mmc/host/mvsdio.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/mmc/host/mvsdio.c b/drivers/mmc/host/mvsdio.c
> index 6b4c5ad3b393..4f8618f4522d 100644
> --- a/drivers/mmc/host/mvsdio.c
> +++ b/drivers/mmc/host/mvsdio.c
> @@ -111,10 +111,15 @@ static int mvsd_setup_data(struct mvsd_host *host, struct mmc_data *data)
>         mvsd_write(MVSD_BLK_COUNT, data->blocks);
>         mvsd_write(MVSD_BLK_SIZE, data->blksz);
>
> -       if (nodma || (data->blksz | data->sg->offset) & 3) {
> +       if (nodma || (data->blksz | data->sg->offset) & 3 ||
> +           ((!(data->flags & MMC_DATA_READ) && data->sg->offset & 0x3f))) {
>                 /*
>                  * We cannot do DMA on a buffer which offset or size
>                  * is not aligned on a 4-byte boundary.
> +                *
> +                * It also appears the host to card DMA can corrupt
> +                * data when the buffer is not aligned on a 64 byte
> +                * boundary.
>                  */
>                 host->pio_size = data->blocks * data->blksz;
>                 host->pio_ptr = sg_virt(data->sg);
> --
> 2.1.3
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/drivers/mmc/host/mvsdio.c b/drivers/mmc/host/mvsdio.c
index 6b4c5ad3b393..4f8618f4522d 100644
--- a/drivers/mmc/host/mvsdio.c
+++ b/drivers/mmc/host/mvsdio.c
@@ -111,10 +111,15 @@  static int mvsd_setup_data(struct mvsd_host *host, struct mmc_data *data)
 	mvsd_write(MVSD_BLK_COUNT, data->blocks);
 	mvsd_write(MVSD_BLK_SIZE, data->blksz);
 
-	if (nodma || (data->blksz | data->sg->offset) & 3) {
+	if (nodma || (data->blksz | data->sg->offset) & 3 ||
+	    ((!(data->flags & MMC_DATA_READ) && data->sg->offset & 0x3f))) {
 		/*
 		 * We cannot do DMA on a buffer which offset or size
 		 * is not aligned on a 4-byte boundary.
+		 *
+		 * It also appears the host to card DMA can corrupt
+		 * data when the buffer is not aligned on a 64 byte
+		 * boundary.
 		 */
 		host->pio_size = data->blocks * data->blksz;
 		host->pio_ptr = sg_virt(data->sg);