Re: [PATCH] Revert "spi: omap2-mcspi: Set FIFO DMA trigger level to word length"

[David Lechner]

On 1/14/19 6:30 AM, Vignesh R wrote:
> On 14/01/19 2:19 AM, David Lechner wrote:
> > This reverts commit b682cffa3ac6d9d9e16e9b413c45caee3b391fab.
> >
> > That commit breaks displays using tinydrm drivers, such as ili9225.
> >
> > It causes corruption in the image that is displayed (it looks like only
> > 1/2 of the framebuffer data is sent, the other half of the display
> > remains blank.)
> >
> > The following error appears multiple times:
> >
> >      ili9225 spi1.0: EOW timed out
> >
> > Eventually, the system locks up without any additional errors.
> Oops, that's unfortunate. I see ili9225 is using bits_per_word = 16.
> I believe commit b682cffa3ac6d9d broke spi_transfers where
> bits_per_word anything other than 8 bits.
> I don't have ili9225 HW but was able emulate this condition using a SPI
> flash as slave on AM335x and successfully reproduced the issue. Could
> you test if this diff[1] helps in fixing the regression without needing
> to revert commit b682cffa3ac6d9d?
>
> If below diff does not help, then could you post logs of types of
> spi_transfers being queued by ili9225 driver? I guess you can dump them
> by enabling DEBUG option in tinydrm-helpers.c.
> Also which TI SoC are you using?
>
> Sorry for the trouble!
>
> [1]:
>
> Adjust maxburst size of DMA xfer according to the DMA trigger level and word length
> instead of just word length
>
>
> diff --git a/drivers/spi/spi-omap2-mcspi.c b/drivers/spi/spi-omap2-mcspi.c
> index 2fd8881fcd65..6b7edcff0e6b 100644
> --- a/drivers/spi/spi-omap2-mcspi.c
> +++ b/drivers/spi/spi-omap2-mcspi.c
> @@ -623,8 +623,8 @@ omap2_mcspi_txrx_dma(struct spi_device *spi, struct spi_transfer *xfer)
>          cfg.dst_addr = cs->phys + OMAP2_MCSPI_TX0;
>          cfg.src_addr_width = width;
>          cfg.dst_addr_width = width;
> -       cfg.src_maxburst = es;
> -       cfg.dst_maxburst = es;
> +       cfg.src_maxburst = width / es;
> +       cfg.dst_maxburst = width / es;

Just an observation... won't `width / es` always == 1 because width is always == es?


>   
>          rx = xfer->rx_buf;
>          tx = xfer->tx_buf;