serial: 8250: Fix __stop_tx() & DMA Tx restart races

Commit e8ffbb7 ("serial: 8250: use THRE & __stop_tx also with
DMA") changed __dma_tx_complete() to enable THRI that is cleared in
__stop_tx() once THRE is asserted as UART runs out bits to transmit. It
is possible, however, that more data arrives in between in which case
serial8250_tx_dma() resumes Tx. THRI is not supposed to be on during
DMA Tx because DMA is based on completion handler, therefore THRI must
be cleared unconditionally in serial8250_tx_dma().

When Tx is about to start, another race window exists with
serial8250_handle_irq() leading to a call into __stop_tx() while the
Tx has already been resumed:

__tx_complete():
  -> spin_lock(port->lock)
  -> dma->tx_running = 0
  -> serial8250_set_THRI()
  -> spin_unlock(port->lock)

uart_start():
				serial8250_handle_irq():
  -> spin_lock(port->lock)
  -> serial8250_tx_dma():
    -> dma->tx_running = 1
  -> spin_unlock(port->lock)
				  -> spin_lock(port->lock)
				  -> __stop_tx()

Close this race by checking !dma->tx_running before calling into
__stop_tx().

Fixes: e8ffbb7 ("serial: 8250: use THRE & __stop_tx also with DMA")
Signed-off-by: Ilpo Järvinen <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Greg Kroah-Hartman <[email protected]>

Author: ij-intel

drivers/tty/serial/8250/8250_dma.c

@@ -106,10 +106,10 @@ int serial8250_tx_dma(struct uart_8250_port *p)
 				   UART_XMIT_SIZE, DMA_TO_DEVICE);
 
 	dma_async_issue_pending(dma->txchan);
-	if (dma->tx_err) {
+	serial8250_clear_THRI(p);
+	if (dma->tx_err)
 		dma->tx_err = 0;
-		serial8250_clear_THRI(p);
-	}
+
 	return 0;
 err:
 	dma->tx_err = 1;

drivers/tty/serial/8250/8250_port.c

@@ -1949,7 +1949,7 @@ int serial8250_handle_irq(struct uart_port *port, unsigned int iir)
 	if ((status & UART_LSR_THRE) && (up->ier & UART_IER_THRI)) {
 		if (!up->dma || up->dma->tx_err)
 			serial8250_tx_chars(up);
-		else
+		else if (!up->dma->tx_running)
 			__stop_tx(up);
 	}