[PATCH 03/16] tty: serial: 8250_core: read only RX if there is something in the FIFO

[Peter Hurley]

Hi Nicolas,

Thanks for the report.

On 02/09/2015 08:34 AM, Nicolas Schichan wrote:
> On 09/10/2014 09:29 PM, Sebastian Andrzej Siewior wrote:
>> The serial8250_do_startup() function unconditionally clears the
>> interrupts and for that it reads from the RX-FIFO without checking if
>> there is a byte in the FIFO or not. This works fine on OMAP4+ HW like
>> AM335x or DRA7.
>> OMAP3630 ES1.1 (which means probably all OMAP3 and earlier) does not like
>> this:
> 
> Hello,
> 
> Sorry to wake up an old thread, but I'm affraid that this patch causes
> problems on Marvell 88f6282 (Kirkwood).
> 
> When a caracter is received on the UART while the kernel is printing
> the boot messages, as soon as the kernel configures the UART for
> receiving (after root filesystem mount), it gets stuck printing the
> following message repeatedly:
> 
> serial8250: too much work for irq29
> 
> Once stuck, the reception of another character allows the boot process
> to finish.
> 
> From what I can gather, when we hit that, the UART_IIR_NO_INT is 0 (so the
> interrupt is raised), but the UART_LSR_DR bit is 0 as well so the UART_RX
> register is never read to clear the interrupt.

The "too much work" message means serial8250_handle_irq() is returning 0,
ie., not handled. Which in turn means IIR indicates no interrupt is pending
(UART_IIR_NO_INT == 1).

Can you log the register values for LSR and IIR at both patch locations
in serial8250_do_startup()?

(I can get you a debug patch, if necessary. Let me know)

Regards,
Peter Hurley

> We are using the second UART multiplexed on mpps 15 and 16.
> 
> Reverting this particular patch fixes the issue.
> 
> We are seing the problem on a 3.18 kernel.