[RFC] serial/8250: Adjusting FIFO parameters for LPC32xx

Hi,

On 27/05/12 04:32, Arnd Bergmann wrote:
>> this is how the modified FIFO is handled in the repositories at
>> git.lpclinux.com. Is there a better way for doing this (without ifdef)?  Looks
>> like registering additional types (like PORT_16550A) isn't encouraged.  Maybe
>> extending of_serial.c? The latter currently doesn't handle .fifosize and
>> .tx_loadsz, though.
>>
>> Any suggestions appreciated.
> 
> We occasionally add new types, and I think that would be ok here too, if it's
> actually a new type. It definitely doesn't look like a 16550.
> 
> I wonder if it's in fact a PORT_16654, or if we can at least pretend it's one.

Thanks for the hint!

I did it via the following modification in of_serial.c:


Works for now, will need to test more thoroughly with LPC32xx (note that
the manual explicitly mentions "downwards compatible with the INS16Cx50"
and "Register locations conform to ‘550 industry standard" but nothing
about an EFR register which 16654 is configured for in 8250.c).

The patch should be useful in of_serial anyway (even without the LPC32xx
case).

Wondering if it should rather be .compatible = "st16654"?
Will prepare a patch with adjustment of Documentation/.../of-serial.txt when
we agree on sth.

Thanks,

Roland

On Sunday 27 May 2012, Roland Stigge wrote:
> I did it via the following modification in of_serial.c:
> 
> --- a/drivers/tty/serial/of_serial.c
> +++ b/drivers/tty/serial/of_serial.c
> @@ -179,6 +179,7 @@ static struct of_device_id __devinitdata of_platform_serial_table[] = {
>         { .compatible = "ns16450",  .data = (void *)PORT_16450, },
>         { .compatible = "ns16550a", .data = (void *)PORT_16550A, },
>         { .compatible = "ns16550",  .data = (void *)PORT_16550, },
> +       { .compatible = "ns16654",  .data = (void *)PORT_16654, },
>         { .compatible = "ns16750",  .data = (void *)PORT_16750, },
>         { .compatible = "ns16850",  .data = (void *)PORT_16850, },
>         { .compatible = "nvidia,tegra20-uart", .data = (void *)PORT_TEGRA, },
> 
> Works for now, will need to test more thoroughly with LPC32xx (note that
> the manual explicitly mentions "downwards compatible with the INS16Cx50"
> and "Register locations conform to ‘550 industry standard" but nothing
> about an EFR register which 16654 is configured for in 8250.c).
> 
> The patch should be useful in of_serial anyway (even without the LPC32xx
> case).
> 
> Wondering if it should rather be .compatible = "st16654"?
> Will prepare a patch with adjustment of Documentation/.../of-serial.txt when
> we agree on sth.

I think st16654 would be better than ns16654 here, yes.

Actually some of the other entries are wrong, too. The ns8250 entry should
be i8250, and I guess the 16750 and 16850 were also not made by national.

	Arnd

Hi,

On 27/05/12 12:23, Arnd Bergmann wrote:
>> --- a/drivers/tty/serial/of_serial.c
>> +++ b/drivers/tty/serial/of_serial.c
>> @@ -179,6 +179,7 @@ static struct of_device_id __devinitdata of_platform_serial_table[] = {
>>         { .compatible = "ns16450",  .data = (void *)PORT_16450, },
>>         { .compatible = "ns16550a", .data = (void *)PORT_16550A, },
>>         { .compatible = "ns16550",  .data = (void *)PORT_16550, },
>> +       { .compatible = "ns16654",  .data = (void *)PORT_16654, },
>>         { .compatible = "ns16750",  .data = (void *)PORT_16750, },
>>         { .compatible = "ns16850",  .data = (void *)PORT_16850, },
>>         { .compatible = "nvidia,tegra20-uart", .data = (void *)PORT_TEGRA, },
>>
>> Wondering if it should rather be .compatible = "st16654"?
>> Will prepare a patch with adjustment of Documentation/.../of-serial.txt when
>> we agree on sth.
> 
> I think st16654 would be better than ns16654 here, yes.

OK, will provide a patch accordingly.

> Actually some of the other entries are wrong, too. The ns8250 entry should
> be i8250, and I guess the 16750 and 16850 were also not made by national.

But maybe we should keep the existing compatible strings because
existing boards/bootloaders are providing them already?

Roland

On Sunday 27 May 2012, Roland Stigge wrote:
> > Actually some of the other entries are wrong, too. The ns8250 entry should
> > be i8250, and I guess the 16750 and 16850 were also not made by national.
> 
> But maybe we should keep the existing compatible strings because
> existing boards/bootloaders are providing them already?

Right, we should only add to the list. Any entry that's already used potentially
needs to stay in there as you say.

	Arnd