Re: [Mjpeg-users] multiple dc10 cards and python module for mjpegtools

[Jean D. <kh...@li...>]

Hi Martin, Richard,

On Tue, 19 Aug 2008 21:41:21 +0200, Martin Samuelsson wrote:
> On Tue, 19 Aug 2008 12:58:16 +0200
> richard <tu...@xs...> wrote:
> 
> > Hello,
> > 
> > I'm using multiple (3 now) dc10plus cards for doing live video shows as
> 
> I've got a Buz, a 6 Eyes and a DC10+ in mine.
> 
> > So everything works fine 
> > except for the dc10plus cards
> > They are not always initialized correctly, so sometimes I have two
> > sometimes one and sometimes three cards working. 
> 
> I can imagine that. I have similar problems, but I'm a little surprised your system works as good as it does.
> 
> > It might have something to do with the order in which the cards are
> > initialized when all cards are found first an then initialized it seems
> > to work, but when a card is initialized before other cards are found
> > then there is an error.
> 
> That is most likely the problem, yes. If you dig at the really low
> levels in the initialization routines, you'll probably find out that
> the video encoders and decoders are initialized at least nine times
> each per card, due to some peculiarities concerning the interaction
> between i2c buses and i2c devices.

Not sure what you mean with "initialized". I fail to see how a given
decoder or encoder could be initialized more than once, given that the
address becomes busy once an i2c client is attached.

The main problem of the current code is when mixing different types of
zoran adapters. I don't expect that many problems when several adapters
of the same type (i.e. using the same encoder and decoder) are used.

> In short, the kernel should see each encoder and decoder once for every
> i2c bus on each of the cards, but as things are now, all of them are
> visible, without any buses differentiating them.
> 
> If this is the culprit, it has been identified, and is being worked on
> by Jean Delvare. I don't know what the time table looks like; there's
> quite a bit of fiddling involved.

Actually I'm done with the conversion now, and would welcome testers.
I'm attaching a (relatively large) patch against 2.6.27-rc3, which
converts the whole zoran driver set to the new i2c device binding model.
With this conversion, each zoran adapter will explicitly instantiate its
I2C encoder and decoder devices. No more probing and guessing.

I would like to head feedback from testers. In particular:
* I want to be sure that the patch doesn't break existing working
  setups.
* If the patch solves problems, I'd like to know. One thing for sure
  is that driver loading should be somewhat faster.

Note that the patch won't work with kernels older than 2.6.27-rc1.

> 
> You _might_ be helped by manually modprobing the encoder and decoder
> drivers before zr63060.
> 
> Follow-up question: Are you using the on-board composite output
> connector? My DC10+ stops after a number of frames when playing with
> lavplay -pC, stalling waiting for a buffer. I've reduced it to some
> kind of race condition, but I'm not sure where to go from there, or
> if indeed any other DC10+ users than me experience the problem.

I don't think I ever used my DC10+ output channels, sorry.

-- 
Jean Delvare