| Message ID | 20190107162320.15530-4-geert+renesas@glider.be (mailing list archive) |
|---|---|
| State | Accepted |
| Delegated to: | Geert Uytterhoeven |
| Headers | show |
| Series | serial: sh-sci: Fix fallback to PIO on DMA failure | expand |
On Mon, Jan 07, 2019 at 05:23:19PM +0100, Geert Uytterhoeven wrote: > When submitting a DMA request fails in sci_dma_rx_complete(), the driver > tries to fall back to PIO, but that does not work: no more data will be > received, or the kernel will even crash. > > Fix this similar as in (but not identical to) sci_submit_rx(): > - On SCIF, PIO cannot take over if any DMA transactions are pending, > hence they must be terminated first. > - All active cookies must be invalidated, else rx_timer_fn() may > trigger a NULL pointer dereference. > - Restarting the port is not needed, as it is already running, but > serial port interrupts must be directed back from the DMA engine to > the CPU. > > Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
> On January 7, 2019 at 5:23 PM Geert Uytterhoeven <geert+renesas@glider.be> wrote: > > > When submitting a DMA request fails in sci_dma_rx_complete(), the driver > tries to fall back to PIO, but that does not work: no more data will be > received, or the kernel will even crash. > > Fix this similar as in (but not identical to) sci_submit_rx(): > - On SCIF, PIO cannot take over if any DMA transactions are pending, > hence they must be terminated first. > - All active cookies must be invalidated, else rx_timer_fn() may > trigger a NULL pointer dereference. > - Restarting the port is not needed, as it is already running, but > serial port interrupts must be directed back from the DMA engine to > the CPU. > > Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> > --- > v5: > - Call new helpers sci_dma_rx_chan_invalidate() and > sci_dma_rx_reenable_irq() instead of open-coding, > > v4: > - No changes, > > v3: > - Add missing definition of "u16 scr", > - Move call to dmaengine_terminate_async() inside the spinlock, for > symmetry with sci_submit_rx(), > - Split in multiple patches. > --- > drivers/tty/serial/sh-sci.c | 5 +++-- > 1 file changed, 3 insertions(+), 2 deletions(-) > > diff --git a/drivers/tty/serial/sh-sci.c b/drivers/tty/serial/sh-sci.c > index 4d814c30c4189b56..891833315698b953 100644 > --- a/drivers/tty/serial/sh-sci.c > +++ b/drivers/tty/serial/sh-sci.c > @@ -1337,8 +1337,9 @@ static void sci_dma_rx_complete(void *arg) > dev_warn(port->dev, "Failed submitting Rx DMA descriptor\n"); > /* Switch to PIO */ > spin_lock_irqsave(&port->lock, flags); > - s->chan_rx = NULL; > - sci_start_rx(port); > + dmaengine_terminate_async(chan); > + sci_dma_rx_chan_invalidate(s); > + sci_dma_rx_reenable_irq(s); > spin_unlock_irqrestore(&port->lock, flags); > } > > -- > 2.17.1 > Reviewed-by: Ulrich Hecht <uli+renesas@fpond.eu> CU Uli
diff --git a/drivers/tty/serial/sh-sci.c b/drivers/tty/serial/sh-sci.c index 4d814c30c4189b56..891833315698b953 100644 --- a/drivers/tty/serial/sh-sci.c +++ b/drivers/tty/serial/sh-sci.c @@ -1337,8 +1337,9 @@ static void sci_dma_rx_complete(void *arg) dev_warn(port->dev, "Failed submitting Rx DMA descriptor\n"); /* Switch to PIO */ spin_lock_irqsave(&port->lock, flags); - s->chan_rx = NULL; - sci_start_rx(port); + dmaengine_terminate_async(chan); + sci_dma_rx_chan_invalidate(s); + sci_dma_rx_reenable_irq(s); spin_unlock_irqrestore(&port->lock, flags); }
When submitting a DMA request fails in sci_dma_rx_complete(), the driver tries to fall back to PIO, but that does not work: no more data will be received, or the kernel will even crash. Fix this similar as in (but not identical to) sci_submit_rx(): - On SCIF, PIO cannot take over if any DMA transactions are pending, hence they must be terminated first. - All active cookies must be invalidated, else rx_timer_fn() may trigger a NULL pointer dereference. - Restarting the port is not needed, as it is already running, but serial port interrupts must be directed back from the DMA engine to the CPU. Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> --- v5: - Call new helpers sci_dma_rx_chan_invalidate() and sci_dma_rx_reenable_irq() instead of open-coding, v4: - No changes, v3: - Add missing definition of "u16 scr", - Move call to dmaengine_terminate_async() inside the spinlock, for symmetry with sci_submit_rx(), - Split in multiple patches. --- drivers/tty/serial/sh-sci.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)