Re: rtd520 broken in 0.7.65

On Mon, 1 Jul 2002, Dan Christian wrote:

> As of version 0.7.65 (and CVS), I'm getting "buffer overfow" errors on
> fast sampling.  If I drop back to 0.7.64, then it runs fine.
>
> Above about 50K sample/second, it consistantly fails at exactly 40960
> samples (with a 64K buffer).  If the input buffer is 32K, then it
> happens at 24576 samples.  At 128K, it happens at 73728 samples.
>
> With the 0.7.64 release, I can do 560K sample/sec without any problems.
> CPU load isn't a problem in either case.  The program is just reading
> and discarding the data.
>
> The diffs in rtd520.c between those versions are trivial (spelling
> fixes).
>
> Anybody have an idea what's broken?
>
> -Dan Christian
>

I probably broke it.  I made some changes to try and make reading/writing
to comedi's buffer more robust by going to a scheme that didn't make any
assumptions about the relative priorities of the reader and writer.
Actually, I wanted to go a little further than I did but I haven't yet...
I only made changes to the comedi_buf_put/get macros and the core comedi
read/write ioctls.  So it's possible that drivers that did their own
manipulation of the buffer might have gotten broken.  I could probably
convert your driver over to using comedi_buf_put() or
comedi_buf_put_array() if you like.

Frank

Received on 2002-07-01Z21:18:33