Re: Continuous acquisition latency and buffer overflow

From: Jonathan Wang <jtwang_at_MIT.EDU> · Date: Mon, 11 Oct 2004 13:44:59 -0400

I've seen some references to the callback capability from searches on 
Google, but can't find anything in the API for it now. Perhaps it was 
an old feature that was removed?

I think a good alternative that is supported by the current driver is 
to spawn a pthread to collect data and select() or poll() on the comedi 
file descriptor. Your thread then gets woken up whenever data is 
available.

Jonathan

On Oct 7, 2004, at 11:44 AM, Mike Reedy wrote:

> I have a continuous acquisiton application based on
> the mmap.c method and found the following useful.
>
> cmd->flags = TRIG_WAKE_EOS;
> This causes the result of "comedi_get_buffer_contents"
> to update scanwise rather than in virtual memory page
> size chunks (something like 4k byte chunks).  A
> scanwise callback would be nice but I don't know if it
> is possible.
>
> comedi_config /dev/comedi0 driver --read-buffer=4096
> Enlarge the read buffer as the default is quite small.
> Use any value appropriate for you application.  The
> application can sense buffer size with
> "comedi_get_buffer_size()".
>
> So long as you conusme the data and "markDataFree" at
> a  rate faster than is being acquired, the buffer is
> recycled indefinitely.
>
> Mike
>
>
> --- Eric Keller <eekeller_at_psu.edu> wrote:
>
>>
>> From: Jonathan Wang <jtwang_at_MIT.EDU>
>>
>> snip
>>
>>> Also, as I will be doing continuous acquisition for
>> a long time, I am
>>> worried about the mmap buffer overflowing. I see
>> that the demo program
>>> mmap.c in comedilib/demo simply increments pointers
>> in its mmap'ed
>>> buffer to read values. However, this seems
>> dangerous to me, since
>>> eventually the pointers will increment past 2^32
>> and the buffer will
>>> overflow. Could somebody clarify the way this
>> works?
>>
>> Jonathan,
>> I also worried about this, but you see that they
>> take the modulo (%) of the
>> counter with the buffer length, so that the index
>> into the buffer is always
>> within the range of the buffer.  It's obviously hard
>> to read, but that handles
>> the case when your read wraps around much more
>> simply that checking the bounds
>> of the index.
>>
>> As far as your control question, I also want to do
>> this.  My program does a lot
>> of processing, and I only want to be woken up at the
>> end of a scan.  There is
>> some mention of a comedi callback, but it appears
>> not to be documented except
>> possibly in the source.
>> Regards,
>> Eric Keller
>>
>>> Thanks,
>>> Jonathan
>>
>> _______________________________________________
>> comedi mailing list
>> comedi_at_comedi.org
>>
> https://cvs.comedi.org/cgi-bin/mailman/listinfo/comedi
>>
>
>
>
> 		
> _______________________________
> Do you Yahoo!?
> Declare Yourself - Register online to vote today!
> http://vote.yahoo.com
>
> _______________________________________________
> comedi mailing list
> comedi_at_comedi.org
> https://cvs.comedi.org/cgi-bin/mailman/listinfo/comedi