Re: [PATCH RFC] serial: 8250: Handle invalid UART state

[Peter Hurley <peter@hurleysoftware.com>]

On 08/19/2015 04:12 PM, california.l.sullivan@intel.com wrote:
> From: California Sullivan <california.l.sullivan@intel.com>
> 
> The debug UART on the Bay Trail is buggy and will sometimes enter a
> state where there is a Character Timeout interrupt, but the Data
> Ready bit in the Line Status Register is not set, despite there
> being data available in the FIFO. It stays in this state until the
> Receive Buffer is read. Because the 8250 driver does not read the
> Receive Buffer without the Data Ready bit being set, the UART is
> never read once we reach this point, and thus the Character Timeout
> interrupt is never cleared. This makes the driver spin in a loop
> causing multiple "too much work for irq" errors and eventually
> locking up entirely.

LSR doesn't reflect the actual state of the fifo?
So kgdb (CONSOLE_POLL) won't work on these platforms either, right?

Does this happen with only a single byte in the fifo or is it any
number of bytes under the fifo trigger?


> This patch handles the invalid state by setting the Data Ready bit
> in the 'status' variable when the invalid state occurs. This allows
> the receive buffer to be read, which clears the interrupt
> identification register in the UART and brings us back to a correct
> state.
> 
> After this patch was submitted for internal comment, a similar bug on
> the HSUARTs of the Bay Trail and Braswell platforms was pointed out.
> On those UARTs, the invalid state mentioned previously is reached for
> some amount of time, cauing the same "too much work for irq" messages,
> but not permanently locking up the UART. This patch has been confirmed
> to solve that issue as well.
> 
> We considered solving this by adding a new UART_BUG_ define and
> detecting the buggy UART by either identifying the processor or by
> adding a new PNP device ID to firmware. However, this solution
> would be more complex and have zero performance benefits, as the
> ISR would require a similar 'if' condition to detect and handle the
> bug.
> 
> Our main concern with this fix is whether it having a Character
> Timeout with no Data Ready is an invalid state for all UARTs or
> just some. If other UARTs have a Character Timeout without
> immediately flipping the Data Ready bit in the Line Status Register
> for a good reason, setting the Data Ready bit in the 'status'
> variable could have unintended consequences.

I think there is every likelihood of spurious RX timeout interrupts
tripping this patch, sorry.

Unfortunately, I think UART_BUG_ is the only viable possibility.
Or perhaps fixing the port type as PORT_8250 (thus disabling the fifos).

Regards,
Peter Hurley


> Signed-off-by: California Sullivan <california.l.sullivan@intel.com>
> ---
>  drivers/tty/serial/8250/8250_core.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/drivers/tty/serial/8250/8250_core.c b/drivers/tty/serial/8250/8250_core.c
> index ea1a8da..078b118 100644
> --- a/drivers/tty/serial/8250/8250_core.c
> +++ b/drivers/tty/serial/8250/8250_core.c
> @@ -1604,6 +1604,14 @@ int serial8250_handle_irq(struct uart_port *port, unsigned int iir)
>  
>  	DEBUG_INTR("status = %x...", status);
>  
> +	/*
> +	 * Workaround for when there is a character timeout interrupt
> +	 * but the data ready bit is not set in the Line Status Register.
> +	 */
> +	if ((iir & UART_IIR_RX_TIMEOUT) &&
> +	    !(status & (UART_LSR_DR | UART_LSR_BI)))
> +		status |= UART_LSR_DR;
> +
>  	if (status & (UART_LSR_DR | UART_LSR_BI)) {
>  		if (up->dma)
>  			dma_err = up->dma->rx_dma(up, iir);
>