Early printk baudrate initialization DTS description

[Michal Simek]

On 06/30/2012 03:41 AM, David VomLehn (dvomlehn) wrote:
>> -----Original Message-----
>> From: devicetree-discuss [mailto:devicetree-discuss-
>> bounces+dvomlehn=cisco.com at lists.ozlabs.org] On Behalf Of Michal Simek
>> Sent: Tuesday, June 26, 2012 4:17 AM
>> To: devicetree-discuss at lists.ozlabs.org
>> Cc: John Williams
>> Subject: Early printk baudrate initialization DTS description
>>
>> Hi,
>>
>> we are trying to solve one problem with early printk initialization
>> on Microblaze uart16550. I have also checked arm bootup code and
>> the situation is the same there (NOTE: This IP is possible to use on
>> Zynq too).
>>
>> The problem is about baudrate initialization. Current code just expect
>> that early console is initialized by previous bootloader/firmware (in
>> our case u-boot).
>> But if there is not initialization and directly elf is loaded there is
>> no initialization
>> done and kernel bootlog is shown when regular serial driver is probed.
>>
>> I have create a solution which add baudrate initialization to
>> Microblaze early printk code.
>> And also two options how to load baudrate setting from device-tree.
>>
>> Code expects that early console is specified in chosen node.
>>
>> chosen {
>> 	linux,stdout-path = "/axi at 0/serial at 40400000";
>> } ;
>>
>> Code finds out which IP it is and based on that initialize early
> printk
>> driver.
>> Baudrate setting is counted from clock-frequency and current-speed
>> defined directly in uart node.
>
> The terminology is not quite clear to me. Many people use the term baud
> rate when they mean the more accurate term "data rate" and I'm pretty
> sure this is what you mean, but is this different than "current-speed"?

ok. Sorry not serial expert but please correct me.

I need to read required baudrate setting (9600,...,115200, etc) which is used
for communication(it should be numeric bits per second) and clock frequency.
And based on the code setups baud_divisor for the baud rate generator counter.

And based on ePAPR.. Specifies the current speed of a serial device in bits per second.
A boot program should set this property if it has initialized the serial device.

>
>>
>> RS232_Uart_1: serial at 40400000 {
>> 	clock-frequency =<50000000>;
>> 	compatible = "xlnx,axi-uart16550-1.01.a", "xlnx,xps-uart16550-
>> 2.00.a", "ns16550a";
>> 	current-speed =<115200>;
>> 	device_type = "serial";
>> 	reg =<  0x40400000 0x10000>;
>> 	reg-offset =<0x1000>;
>> 	reg-shift =<2>;
>> 	...irq, etc
>> } ;
>>
>> This was the first version which is working but using current-speed is
>> probably not the best solution.
>>
>> I have also create version where baudrate is directly specified in
>> chosen node. Clock frequency is still
>> loaded directly from the node.
>>
>> chosen {
>> 	linux,stdout-path = "/axi at 0/serial at 40400000,115200";
>> } ;
>>
>> What solution is more compatible with device-tree?
>
> Since nobody else has responded for a couple of days, I'll throw out the
> opinion that I prefer the first solution. The second solution requires
> you to do additional parsing and data validation which the device tree
> has already done for you in the first approach. It's also consistent
> with the ePAPR example in
> https://www.power.org/resources/downloads/Power_ePAPR_APPROVED_v1.0.pdf.

This is done in early phase which means that I don't believe that there is
any parsing done for the first approach too. It is simple value reading
with no checking.
And yes, current-speed should be used exactly for this purpose.

Thanks,
Michal




-- 
Michal Simek, Ing. (M.Eng)
w: www.monstr.eu p: +42-0-721842854
Maintainer of Linux kernel 2.6 Microblaze Linux - http://www.monstr.eu/fdt/
Microblaze U-BOOT custodian