On 08/07/2015 04:00 PM, Sebastian Andrzej Siewior wrote:
> The 8250-omap driver requires the DMA-engine driver to support the pause
> command in order to properly turn off programmed RX transfer before the
> driver stars manually reading from the FIFO.
> The lacking support of the requirement has been discovered recently. In
> order to stay safe here we disable support for RX-DMA as soon as we
> notice that it does not work. This should happen very early.
> If the user does not want to see this backtrace he can either disable
> DMA support (completely or RX-only) or backport the required patches for
> edma / omap-dma once they hit mainline.
>
> Cc: <sta...@vger.kernel.org>
> Signed-off-by: Sebastian Andrzej Siewior <bige...@linutronix.de>
> ---
> drivers/tty/serial/8250/8250_omap.c | 9 ++++++++-
> 1 file changed, 8 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/tty/serial/8250/8250_omap.c
> b/drivers/tty/serial/8250/8250_omap.c
> index 0340ee6ba970..07a11e0935e4 100644
> --- a/drivers/tty/serial/8250/8250_omap.c
> +++ b/drivers/tty/serial/8250/8250_omap.c
> @@ -112,6 +112,7 @@ struct omap8250_priv {
> struct work_struct qos_work;
> struct uart_8250_dma omap8250_dma;
> spinlock_t rx_dma_lock;
> + bool rx_dma_broken;
> };
>
> static u32 uart_read(struct uart_8250_port *up, u32 reg)
> @@ -761,6 +762,7 @@ static void omap_8250_rx_dma_flush(struct uart_8250_port
> *p)
> struct omap8250_priv *priv = p->port.private_data;
> struct uart_8250_dma *dma = p->dma;
> unsigned long flags;
> + int ret;
>
> spin_lock_irqsave(&priv->rx_dma_lock, flags);
>
> @@ -769,7 +771,9 @@ static void omap_8250_rx_dma_flush(struct uart_8250_port
> *p)
> return;
> }
>
> - dmaengine_pause(dma->rxchan);
> + ret = dmaengine_pause(dma->rxchan);
> + if (WARN_ON_ONCE(ret))
> + priv->rx_dma_broken = true;
No offense, Sebastian, but it boggles my mind that anyone could defend this
as solid api design. We're in the middle of an interrupt handler and the
slave dma driver is /just/ telling us now that it doesn't implement this
functionality?!!?
The dmaengine api has _so much_ setup and none of it contemplates telling the
consumer that critical functionality is missing?
Even dma_get_slave_caps() returns _true_ for cmd_pause support; ok, that
interface is pointless.
Rather than losing /critical data/ here, the interrupt handler should just
busy-wait until dmaengine_tx_status() returns DMA_COMPLETE for the rx_cookie.
Regards,
Peter Hurley
> spin_unlock_irqrestore(&priv->rx_dma_lock, flags);
>
> @@ -813,6 +817,9 @@ static int omap_8250_rx_dma(struct uart_8250_port *p,
> unsigned int iir)
> break;
> }
>
> + if (priv->rx_dma_broken)
> + return -EINVAL;
> +
> spin_lock_irqsave(&priv->rx_dma_lock, flags);
>
> if (dma->rx_running)
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
