[PATCHv2] serial: 8250: Fix thread unsafe __dma_tx_complete function

[PATCHv2] serial: 8250: Fix thread unsafe __dma_tx_complete function

[Loic Poulain]

__dma_tx_complete is not protected against concurrent
call of serial8250_tx_dma. it can lead to circular tail
index corruption or parallel call of serial_tx_dma on the
same data portion.

This patch fixes this issue by holding the port lock.

Signed-off-by: Loic Poulain <loic.poulain@intel.com>
---
 v2: extend lock to protect xmit tail (cf doc)
     and tx_running. tx_running and tail have to
     be updated atomically.
     update commit message.

 drivers/tty/serial/8250/8250_dma.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/tty/serial/8250/8250_dma.c b/drivers/tty/serial/8250/8250_dma.c
index 7046769..ab9096d 100644
--- a/drivers/tty/serial/8250/8250_dma.c
+++ b/drivers/tty/serial/8250/8250_dma.c
@@ -20,12 +20,15 @@ static void __dma_tx_complete(void *param)
 	struct uart_8250_port	*p = param;
 	struct uart_8250_dma	*dma = p->dma;
 	struct circ_buf		*xmit = &p->port.state->xmit;
-
-	dma->tx_running = 0;
+	unsigned long	flags;
 
 	dma_sync_single_for_cpu(dma->txchan->device->dev, dma->tx_addr,
 				UART_XMIT_SIZE, DMA_TO_DEVICE);
 
+	spin_lock_irqsave(&p->port.lock, flags);
+
+	dma->tx_running = 0;
+
 	xmit->tail += dma->tx_size;
 	xmit->tail &= UART_XMIT_SIZE - 1;
 	p->port.icount.tx += dma->tx_size;
@@ -35,6 +38,8 @@ static void __dma_tx_complete(void *param)
 
 	if (!uart_circ_empty(xmit) && !uart_tx_stopped(&p->port))
 		serial8250_tx_dma(p);
+
+	spin_unlock_irqrestore(&p->port.lock, flags);
 }
 
 static void __dma_rx_complete(void *param)
-- 
1.8.3.2

Re: [PATCHv2] serial: 8250: Fix thread unsafe __dma_tx_complete function

[Heikki Krogerus]

On Thu, Apr 24, 2014 at 11:34:48AM +0200, Loic Poulain wrote:
> __dma_tx_complete is not protected against concurrent
> call of serial8250_tx_dma. it can lead to circular tail
> index corruption or parallel call of serial_tx_dma on the
> same data portion.
> 
> This patch fixes this issue by holding the port lock.
> 
> Signed-off-by: Loic Poulain <loic.poulain@intel.com>

I think this should also go to stable tree. It applies on top of >=v3.9.

Reviewed-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>

Thanks!

-- 
heikki