Re: [PATCH] ns16550: misc minor adjustments

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 12/11/15 12:10, Ian Campbell wrote:
> On Thu, 2015-11-12 at 04:52 -0700, Jan Beulich wrote:
>
>>>> @@ -530,7 +531,8 @@ static void ns16550_setup_preirq(struct
>>>>          /* Baud rate already set: read it out from the divisor
>>>> latch. */
>>>>          divisor  = ns_read_reg(uart, UART_DLL);
>>>>          divisor |= ns_read_reg(uart, UART_DLM) << 8;
>>>> -        uart->baud = uart->clock_hz / (divisor << 4);
>>>> +        if ( divisor )
>>>> +            uart->baud = uart->clock_hz / (divisor << 4);
>>> Will following code cope with uart->baud == BAUD_AUTO? Or should we
>>> pick a
>>> static fallback rate (115200?) and set the divisor appropriately?
>> The device won't work with it left as BAUD_AUTO. Setting a guessed
>> baud rate alone won't help, as we'd then also have to write it into the
>> respective registers. And I don't think anyway we should do any
>> guessing here - if the command line option was wrong (regardless of
>> whether due to using auto or just a wrong baud rate), communication
>> won't work. All I think we should really care about is to not crash, the
>> more in a way not diagnosable over serial console.
> BAUD_AUTO is the default in device-tree based situations, so it isn't
> necessarily the case that the command line is wrong. For the non-DT case I
> think it also ends up with an implicit BAUD_AUTO in some cases?
>
> That said I don't think it is worth going to the effort to distinguish
> implicit from explicit-auto configuration, and your argument about not
> writing to the registers certainly applies to the latter, so lets just
> worry about not crashing as you say.
>
> Is a printk worthwhile, due to the fact there might be a second console
> present?

Yes.  Even if there isn't a second serial console, the system might boot
and `xl dmesg` should contain a record.

~Andrew