Re: comedi_poll for cb_pcidas64

On Thursday 07 April 2005 08:39 am, Jan Gukelberger wrote:
>
> ok, for the second question, I've just come across
> INSN_CONFIG_BLOCK_SIZE. So it should be possible to at least decrease
> the maximum time gap between acquisition and a sample appearing in the
> mmap'ed streaming buffer.
> Would setting BLOCK_SIZE to some extremely small value like 32 have any
> negative consequences? Probably yes, so what problems would I have to
> expect?

I think the smallest block size supported by the hardware is something like 
256 samples, as it only seemed to want to do a dma transfer when an entire 
"segment" of the board's fifo was full.  The negative consequences are 
that there is more cpu overhead (more interrupts) and the size of the 
board's fifo is reduced when you make the block size smaller (making fifo 
overruns slightly more probable).  Neither of these factors should be 
noticeable at the speeds you are taking data.

You could set the TRIG_WAKE_EOS command flag, which will cause the driver 
to turn off dma and do pio transfers at the end of each scan (even more 
cpu overhead, but should be okay for 1kHz scans).

>
> Anyway, this would be a hack as COMEDI_POLL is exactly what I'd really
> need.
>
> Frank: Do you have any PCI-DAS 60xx docs that indicate whether the
> driver could implement POLL'ing on this hardware?

You can get register level programming information from the MC website.  
You can also get a manual for the plx9080 pci chip on the web.  I don't 
think it is possible to get polling to do anything useful though.  As I 
recall, the samples just accumulate in the hardware fifo until a segment 
is full, then they are all dma'd over to computer memory together.  I 
couldn't figure out a way to make it initiate a dma transfer on fifo not 
empty.  Or if you are using TRIG_WAKE_EOS then you don't need to poll, 
unless you wanted to grab samples part-way through a channel scan.

-- 
Frank