LinuxLists.cc -

On Monday 17 November 2008, Bryan Wu wrote:
> From: Mike Frysinger <[email protected]>
>
> Signed-off-by: Mike Frysinger <[email protected]>
> Signed-off-by: Bryan Wu <[email protected]>

Acked-by: David Brownell <[email protected]>

assuming: this patch grows a real patch comment
instead of only having the long $SUBJECT ...

> ---
> drivers/spi/spi_bfin5xx.c | 47 ++++++++++++++++++++------------------------
> 1 files changed, 21 insertions(+), 26 deletions(-)
>
> diff --git a/drivers/spi/spi_bfin5xx.c b/drivers/spi/spi_bfin5xx.c
> index 642c402..df424a2 100644
> --- a/drivers/spi/spi_bfin5xx.c
> +++ b/drivers/spi/spi_bfin5xx.c
> @@ -755,18 +755,19 @@ static void pump_transfers(unsigned long data)
> if (!full_duplex && drv_data->cur_chip->enable_dma
> && drv_data->len > 6) {
>
> + unsigned long dma_start_addr;
> +
> disable_dma(drv_data->dma_channel);
> clear_dma_irqstat(drv_data->dma_channel);
> bfin_spi_disable(drv_data);
>
> /* config dma channel */
> dev_dbg(&drv_data->pdev->dev, "doing dma transfer\n");
> + set_dma_x_count(drv_data->dma_channel, drv_data->len);
> if (width == CFG_SPI_WORDSIZE16) {
> - set_dma_x_count(drv_data->dma_channel, drv_data->len);
> set_dma_x_modify(drv_data->dma_channel, 2);
> dma_width = WDSIZE_16;
> } else {
> - set_dma_x_count(drv_data->dma_channel, drv_data->len);
> set_dma_x_modify(drv_data->dma_channel, 1);
> dma_width = WDSIZE_8;
> }
> @@ -801,6 +802,7 @@ static void pump_transfers(unsigned long data)
> }
>
> /* In dma mode, rx or tx must be NULL in one transfer */
> + dma_config = (RESTART | dma_width | DI_EN);
> if (drv_data->rx != NULL) {
> /* set transfer mode, and enable SPI */
> dev_dbg(&drv_data->pdev->dev, "doing DMA in.\n");
> @@ -814,19 +816,9 @@ static void pump_transfers(unsigned long data)
> /* clear tx reg soformer data is not shifted out */
> write_TDBR(drv_data, 0xFFFF);
>
> - set_dma_x_count(drv_data->dma_channel, drv_data->len);
> -
> - /* start dma */
> - dma_enable_irq(drv_data->dma_channel);
> - dma_config = (WNR | RESTART | dma_width | DI_EN);
> - set_dma_config(drv_data->dma_channel, dma_config);
> - set_dma_start_addr(drv_data->dma_channel,
> - (unsigned long)drv_data->rx);
> - enable_dma(drv_data->dma_channel);
> -
> - /* start SPI transfer */
> - write_CTRL(drv_data,
> - (cr | CFG_SPI_DMAREAD | BIT_CTL_ENABLE));
> + dma_config |= WNR;
> + dma_start_addr = (unsigned long)drv_data->rx;
> + cr |= CFG_SPI_DMAREAD;
>
> } else if (drv_data->tx != NULL) {
> dev_dbg(&drv_data->pdev->dev, "doing DMA out.\n");
> @@ -837,18 +829,21 @@ static void pump_transfers(unsigned long data)
> (unsigned long) (drv_data->tx +
> drv_data->len_in_bytes));
>
> - /* start dma */
> - dma_enable_irq(drv_data->dma_channel);
> - dma_config = (RESTART | dma_width | DI_EN);
> - set_dma_config(drv_data->dma_channel, dma_config);
> - set_dma_start_addr(drv_data->dma_channel,
> - (unsigned long)drv_data->tx);
> - enable_dma(drv_data->dma_channel);
> + dma_start_addr = (unsigned long)drv_data->tx;
> + cr |= CFG_SPI_DMAWRITE;
> +
> + } else
> + BUG();
> +
> + /* start dma */
> + dma_enable_irq(drv_data->dma_channel);
> + set_dma_config(drv_data->dma_channel, dma_config);
> + set_dma_start_addr(drv_data->dma_channel, dma_start_addr);
> + enable_dma(drv_data->dma_channel);
> +
> + /* start SPI transfer */
> + write_CTRL(drv_data, (cr | BIT_CTL_ENABLE));
>
> - /* start SPI transfer */
> - write_CTRL(drv_data,
> - (cr | CFG_SPI_DMAWRITE | BIT_CTL_ENABLE));
> - }
> } else {
> /* IO mode write then read */
> dev_dbg(&drv_data->pdev->dev, "doing IO transfer\n");
> --
> 1.5.6.3
>
>

2008-11-20 20:53:47

by David Brownell

[permalink] [raw]

On Thursday 20 November 2008, Mike Frysinger wrote:
> On Thu, Nov 20, 2008 at 16:47, David Brownell wrote:

> > Basically, dma_map_single(), dma_unmap_single() ... and
> > remember that the caller may have done the mappings for
> > you already.
>
> these arent required to provide coherent memory right ? if that's the
> case, i can take a look at getting things updated.

Right. If caller provides dma-coherent memory to you, they
must also have provided the DMA addresses the controller
driver should be using ... so you won't need dma mapping
calls on those paths.

> > It'd be nice if one of patches snuck in a comment on that
> > point: "Full duplex only works for non-DMA transfers."
> > Same rationale: you may know this hardware inside out,
> > but the next person won't.
>
> well, hopefully if they dont know they wont be touching the driver ;).
> i'll add a comment in this code chunk. thanks for the feedback.

The way it usually works is someone observes a problem and
then starts experimenting on relevant code. So they learn
a bit while debugging ... and code authors/maintainers need
to keep that learning curve from being too steep. ;)

- Dave