linux-next/Documentation/video4linux/zr36120.txt

Driver for Trust Computer Products Framegrabber, version 0.6.1
------ --- ----- -------- -------- ------------  ------- - - -

- ZORAN ------------------------------------------------------
 Author: Pauline Middelink <middelin@polyware.nl> 
   Date: 18 September 1999
Version: 0.6.1

- Description ------------------------------------------------

Video4Linux compatible driver for an unknown brand framegrabber
(Sold in the Netherlands by TRUST Computer Products) and various
other zoran zr36120 based framegrabbers.

The card contains a ZR36120 Multimedia PCI Interface and a Philips
SAA7110 Onechip Frontend videodecoder. There is also an DSP of
which I have forgotten the number, since i will never get that thing
to work without specs from the vendor itself.

The SAA711x are capable of processing 6 different video inputs,
CVBS1..6 and Y1+C1, Y2+C2, Y3+C3. All in 50/60Hz, NTSC, PAL or
SECAM and delivering a YUV datastream.  On my card the input
'CVBS-0' corresponds to channel CVBS2 and 'S-Video' to Y2+C2.

I have some reports of other cards working with the mentioned
chip sets. For a list of other working cards please have a look
at the cards named in the tvcards struct in the beginning of
zr36120.c

After some testing, I discovered that the carddesigner messed up
on the I2C interface. The Zoran chip includes 2 lines SDA and SCL
which (s)he connected reversely. So we have to clock on the SDA
and r/w data on the SCL pin. Life is fun... Each cardtype now has
a bit which signifies if you have a card with the same deficiency.

Oh, for the completeness of this story I must mention that my
card delivers the VSYNC pulse of the SAA chip to GIRQ1, not
GIRQ0 as some other cards have. This is also incorporated in
the driver be clearing/setting the 'useirq1' bit in the tvcard
description.

Another problems of continuous capturing data with a Zoran chip
is something nasty inside the chip. It effectively halves the
fps we ought to get... Here is the scenario: capturing frames
to memory is done in the so-called snapshot mode. In this mode
the Zoran stops after capturing a frame worth of data and wait
till the application set GRAB bit to indicate readiness for the
next frame. After detecting a set bit, the chip neatly waits
till the start of a frame, captures it and it goes back to off.
Smart ppl will notice the problem here. Its the waiting on the
_next_ frame each time we set the GRAB bit... Oh well, 12,5 fps
is still plenty fast for me.
-- update 28/7/1999 --
Don't believe a word I just said... Proof is the output
of `streamer -t 300 -r 25 -f avi15 -o /dev/null`
	++--+-+-+-+-+-+-+-+-+-+-+-+-s+-+-+-+-+-+-+-+-+-+-+- 25/25
	+-s+-+-+-+-+-+-+-+-+-+-+-+-+-s+-+-+-+-+-+-+-+-+-+-+- 25/25
	+-s+-+-+-+-+-+-+-+-+-+-+-+-+-s+-+-+-+-+-+-+-+-+-+-+- 25/25
	+-s+-+-+-+-+-+-+-+-+-+-+-+-+-s+-+-+-+-+-+-+-+-+-+-+- 25/25
	+-s+-+-+-+-+-+-+-+-+-+-+-+-+-s+-+-+-+-+-+-+-+-+-+-+- 25/25
	+-s+-+-+-+-+-+-+-+-+-+-+-+-+-s+-+-+-+-+-+-+-+-+-+-+- 25/25
	+-s+-+-+-+-+-+-+-+-+-+-+-+-+-s+-+-+-+-+-+-+-+-+-+-+- 25/25
	+-s+-+-+-+-+-+-+-+-+-+-+-+-+-s+-+-+-+-+-+-+-+-+-+-+- 25/25
	+-s+-+-+-+-+-+-+-+-+-+-+-+-+-s+-+-+-+-+-+-+-+-+-+-+- 25/25
	+-s+-+-+-+-+-+-+-+-+-+-+-+-+-s+-+-+-+-+-+-+-+-+-+-+- 25/25
	+-s+-+-+-+-+-+-+-+-+-+-+-+-+-s+-+-+-+-+-+-+-+-+-+-+- 25/25
	+-s+-+-+-+-+-+-+-+-+-+-+-+-+-s+-+-+-+-+-+-+-+-+-+-+-
	syncer: done
	writer: done
(note the /dev/null is prudent here, my system is not able to
 grab /and/ write 25 fps to a file... gifts welcome :) )
The technical reasoning follows: The zoran completed the last
frame, the VSYNC goes low, and GRAB is cleared. The interrupt
routine starts to work since its VSYNC driven, and again
activates the GRAB bit. A few ms later the VSYNC (re-)rises and
the zoran starts to work on a new and freshly broadcasted frame....

For pointers I used the specs of both chips. Below are the URLs:
  http://www.zoran.com/ftp/download/devices/pci/ZR36120/36120data.pdf
  http://www-us.semiconductor.philips.com/acrobat/datasheets/SAA_7110_A_1.pdf
Some alternatives for the Philips SAA 7110 datasheet are:
  http://www.datasheetcatalog.com/datasheets_pdf/S/A/A/7/SAA7110.shtml
  http://www.datasheetarchive.com/search.php?search=SAA7110&sType=part

The documentation has very little on absolute numbers or timings
needed for the various modes/resolutions, but there are other
programs you can borrow those from.

------ Install --------------------------------------------
Read the file called TODO.  Note its long list of limitations.

Build a kernel with VIDEO4LINUX enabled. Activate the
BT848 driver; we need this because we have need for the
other modules (i2c and videodev) it enables.

To install this software, extract it into a suitable directory.
Examine the makefile and change anything you don't like.  Type "make".

After making the modules check if you have the much needed
/dev/video devices. If not, execute the following 4 lines:
	mknod /dev/video c 81 0
	mknod /dev/video1 c 81 1
	mknod /dev/video2 c 81 2
	mknod /dev/video3 c 81 3
	mknod /dev/video4 c 81 4

After making/checking the devices do:
	modprobe i2c
	modprobe videodev
	modprobe saa7110	(optional)
	modprobe saa7111	(optional)
	modprobe tuner		(optional)
	insmod zoran cardtype=<n>

<n> is the cardtype of the card you have. The cardnumber can
be found in the source of zr36120. Look for tvcards. If your
card is not there, please try if any other card gives some
response, and mail me if you got a working tvcard addition. 

PS. <TVCard editors behold!)
    Dont forget to set video_input to the number of inputs
    you defined in the video_mux part of the tvcard definition.
    Its a common error to add a channel but not incrementing
    video_input and getting angry with me/v4l/linux/linus :(

You are now ready to test the framegrabber with your favorite
video4linux compatible tool

------ Application ----------------------------------------

This device works with all Video4Linux compatible applications,
given the limitations in the TODO file.

------ API ------------------------------------------------

This uses the V4L interface as of kernel release 2.1.116, and in
fact has not been tested on any lower version.  There are a couple
of minor differences due to the fact that the amount of data returned
with each frame varies, and no doubt there are discrepancies due to my
misunderstanding of the API.  I intend to convert this driver to the
new V4L2 API when it has stabilized more.

------ Current state --------------------------------------

The driver is capable of overlaying a video image in screen, and
even capable of grabbing frames. It uses the BIGPHYSAREA patch
to allocate lots of large memory blocks when tis patch is
found in the kernel, but it doesn't need it.
The consequence is that, when loading the driver as a module,
the module may tell you it's out of memory, but 'free' says
otherwise. The reason is simple; the modules wants its memory
contiguous, not fragmented, and after a long uptime there
probably isn't a fragment of memory large enough...

The driver uses a double buffering scheme, which should really
be an n-way buffer, depending on the size of allocated framebuffer
and the requested grab-size/format.
This current version also fixes a dead-lock situation during irq
time, which really, really froze my system... :)

Good luck.
  Pauline