Re: detect auto baud rate on can bus in linux driver.

[Wolfgang Grandegger <wg@grandegger.com>]

On Tue, 29 Apr 2014 15:33:55 +0200, Kurt Van Dijck
<dev.kurt@vandijck-laurijssen.be> wrote:
> On Tue, Apr 29, 2014 at 12:57:05PM +0530, Ratheendran wrote:
>> Thanks John , Kurt Van Dijck for your valuable support.
>> 
>> I am going ahead with the shell script & iproute2 supported tool
testing
>> a set
>> of fixed baud rate from table.
>> 
>> My observations with the existing imx6 flexcan hardware running linux
>> 3.0.x.
>> 
>> Two imx6 board ,both are enabled for FLEXCAN  communication.
>> CAN dev bring up command with baud rate.
>> command -> ip link set can0 up type can bitrate 125000

For bus error reporting the option "berr-reporting on" needs to
be added. Otherwise only CAN error state changes will be send.

>> one board run cansend while other end candump--e command

It's good practice to use:

 # candump -td -e any,0:0,#FFFFFFFF

> 
> The 'autobaud' node must do
> 	$ ip link set canX type can listen-only on
> 
>> 
>> 
>> Case 1 both imx6 operate on same CAN baudrate
>>   cansend can0 -i0x100 11 22 33 44
>>   on candump terminal I receive  <0x100> [4] 0b 16 21 2c
> 
> Do you use the latest tools from gitorious?
> The syntax of cansend and the output of candump look different.
> 
> [...]
>>   Case 2: imx6 CAN on different baud rate. 
>>  received packets( I am not able to interpret the below packet)
>>  <0x004> [8] 00 04 00 00 00 00 00 00 
> 
> Well, this is not ok.

Ratheendran, what kernel version are you using? Your flexcan driver might
have (old) issues.

> I'd expect an error or status frame, but I don't see it in your candump
> output.

Me too, that's weird.

> I do notice that the received frame has 8 bytes, and only 1 bit set.
> So, if I suppose that the CAN_ERR_FLAG has been set,
> I see (with include/linux/can/error.h aside)
> <0x004> Controller problem, info in data[1]
> 
> data[1] 04: Reached warning level for RX errors.
> 
> I'd expect something like that.
> Wolfgang Grandegger may explain more, as he did author these.

Sounds reasonable, also because the rx error counter did reach 128.

> Next step is to make a script that does all that, and that makes candump
> stop after a limited time (few seconds).
> I would suggest to skip a baudrate if you receive any error frame, so
the
> interpretation may be not necessary.

Wolfgang.