serial: uartps: Fix stuck ISR if RX disabled with non-empty FIFO

commit 7abab160 upstream.

If RX is disabled while there are still unprocessed bytes in RX FIFO,
cdns_uart_handle_rx() called from interrupt handler will get stuck in
the receive loop as read bytes will not get removed from the RX FIFO
and CDNS_UART_SR_RXEMPTY bit will never get set.

Avoid the stuck handler by checking first if RX is disabled. port->lock
protects against race with RX-disabling functions.

This HW behavior was mentioned by Nathan Rossi in 43e98facc4a3 ("tty:
xuartps: Fix RX hang, and TX corruption in termios call") which fixed a
similar issue in cdns_uart_set_termios().
The behavior can also be easily verified by e.g. setting
CDNS_UART_CR_RX_DIS at the beginning of cdns_uart_handle_rx() - the
following loop will then get stuck.

Resetting the FIFO using RXRST would not set RXEMPTY either so simply
issuing a reset after RX-disable would not work.

I observe this frequently on a ZynqMP board during heavy RX load at 1M
baudrate when the reader process exits and thus RX gets disabled.

Fixes: 61ec9016 ("tty/serial: add support for Xilinx PS UART")
Signed-off-by: Anssi Hannula <anssi.hannula@bitwise.fi>
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/tty/serial/xilinx_uartps.c

@@ -366,7 +366,13 @@ static irqreturn_t cdns_uart_isr(int irq, void *dev_id)
 		cdns_uart_handle_tx(dev_id);
 		isrstatus &= ~CDNS_UART_IXR_TXEMPTY;
 	}
-	if (isrstatus & CDNS_UART_IXR_RXMASK)
+
+	/*
+	 * Skip RX processing if RX is disabled as RXEMPTY will never be set
+	 * as read bytes will not be removed from the FIFO.
+	 */
+	if (isrstatus & CDNS_UART_IXR_RXMASK &&
+	    !(readl(port->membase + CDNS_UART_CR) & CDNS_UART_CR_RX_DIS))
 		cdns_uart_handle_rx(dev_id, isrstatus);
 
 	spin_unlock(&port->lock);