v4l - some comments about interrupt driven operation

[Peter Pregler]

Hi all,

I have thrown away (literally speaking) the old interrupt driven
v4l-code I wrote and started to implement a scheme based on polling. The
reasons are:

- Despite quite some effort I could not get a stable interrupt
  scheme. The driver constantly misses interrupts which in my scheme
  would stall capturing which is not acceptable. I see no 
  easy way to debug the interrupt-code (in interface_pp.c/
  interface_pp_low.c) without quite some help from
  the linux parport-people (timeouts, etc.) and quite some
  more information about the cpia. I simply don't have the
  time to do that.
- I can see no real performance penalty in a polling scheme. Loading an
  image from the camera can take up to 100ms and cannot be done
  at interrupt time. Compared to the time needed and loaded needed
  to read a picture polling if done intelligent is negliable.
- My code needed a rewrite anyway.
- Polling does seem to work stable with the current module.c
  code. So there is no reason it should not work stable if 
  done in a kernel-task.

If you want to do further work on interrupt-driven operation contanct me.
I guess I can give you quite some information now about that. But as I
said I do not think it is worth the effort. And my new implementation will
prove that anyway. :) Sorry for promising too much, but I guess if one does
such a thing for the first time one does follow the wrong paths. :(

My plans are to get a stable version as soon as possible (mmap/read, some
ioctls to set size cif/qcif, adjust color parameters, compression). After
that I will add sif/qsif sizes because I need that for ISABEL, which is
why I write the code anyway (http://isabel.dit.upm.es/).

Greetings, Peter


-------------------------------
Email: Peter.Pregler@risc.uni-linz.ac.at
WWW:   http://www.risc.uni-linz.ac.at/people/ppregler
-----------------------------------------------------------------------------
To unsubscribe from this mailinglist, send the line "unsubscribe vision-webcam" in the
body of a message to "majordomo@errors.no".