3DFX Voodoo3 3500 TV

2.4 kernel

Think you have a defective card?

Digikey         Panasonic        Desc
PCE3062CT-ND    ECE-V1CA100SR    CAP 10UF 16V VS ELECT SMD
PCE3182CT-ND    ECE-V1CA101WP    CAP 100UF 16V ELECT VS SMD

Adding a fan to the card

My changes to the v4l_voodoo_irq function may not be as effective as I first thought.  On a couple of occasions now the machine has locked up hard.  I'm still working on porting the voodoo-dump module to a printk format that can be called when the irq function dies.  You would probably need to have a serial console configured to see the output.  The kgdb page linked above has some notes on setting this up.

voodoo.c & voodoo.h

voodoo-i2c.c

voodoo-v4l.c

I've seen the v4l_voodoo_irq function lock up twice now with the current busy loop in place.     Rather than the machine freezing up completly, the irq is now released, and the Voodoo3 VMI interupt is disabled.  On the first occasion, I unloaded and reloaded all the v3tv related modules, then went back to viewing without incident.  The second time, I did not unload the modules, and the machine froze up when I launched motv.

voodoo-audiochip.h

voodoo-v4l.c

The most significant change is that I now use a constant sized video input field, calculated when the tv standard is selected.  This removes the need for the non-standard ioctl for the vpx3224, DECODER_SET_WINDOW.  Since it seems none of the other pixel decoders use the decimation feature, I've decided not to go forward with it.  The voodoo3_update_overlay_setup function already has the ability to configure the Voodoo3 to do all scaling.

voodoo.c

voodoo.h

voodoo-overlay.c

voodoo-v4l.c

voodoo-audiochip.h ( linux/drivers/media/video/audiochip.h )

vpx3224.c

Let's call this update "V3TV-Light!"

I've stripped the I2C adapter code from the v3tv module, making it that much lighter, and utilizing the kernel module i2c-voodoo3.o.  For the moment, I've set up a compile time option to include or exclude the i2c adapter code.  As distributed, the v3tv/Makefile is set to exclude the i2c adapter.  Uncomment the INCLUDE_I2C variable if you want to build with the v3tv i2c adapter?  I've also set the other EXCLUDE options to not build the various voodoo-xxx modules now available in the kernel to make building quicker.

I've also expaned the patch against the msp3400 module to change the audio input source to the rca jacks on the pod when you select Composite or S-VGA input.  It's in the kernel-diffs directory as msp3400.i2c.diff.

I went through and made all of the printk statements write "v3tv:" rather than "voodoo:" or nothing at all in some cases.  This is a prelude to posibly renaming all of the functions "v3tv_xxx_yyy()"

voodoo.h

voodoo.c

Significant changes related to removing the i2c adapter code.  The adapter is not really gone, just made optional at compile time.  The only thing left here related to i2c are two calls to the new function voodoo3_i2c_setup that either enables or disables the i2c interface.

Removed some other cruft related to older kernels.

voodoo-i2c.c

Everything above the function voodoo3_get_adapter is excluded if INCLUDE_I2C is not defined.

The new function voodoo3_get_adapter finds the i2c-voodoo3.o i2c bus.

The procedures in the new function voodoo3_i2c_setup is dependent on if INCLUDE_I2C is defined when compiled.  If INCLUDE_I2C is defined it does what it always did to add or delete the i2c and ddc busses.  Otherwise it installs a dummy i2c driver.

The procedures in the voodoo3_call_i2c_client are also dependent on if INCLUDE_I2C is defined when compiled.  If INCLUDE_I2C is defined it does what it always did to call the clients. Otherwise it calls them in a similar manner, using the i2c-voodoo3 i2c_adapter it got earlier.

voodoo-v4l.c

Moved the v4l_voodoo_irq function here, with it's attendant global variables and the new function v4l_irq_printk. 

Added the new function v4l_set_palette to examine the dstFormat register and set picture_settings->palette and depth accordingly.

Added a procedure to v4l_voodoo_open that sets the tuner type, only if INCLUDE_I2C is defined when compiled.

Added a call to v4l_set_palette in v4l_voodoo_open so that palette and depth will be correct durring subsequent calls to DECODER_SET_PALETTE and voodoo3_capture_frame will have correct values.  This is the intent, but action is yet to be taken.

Optimized the VIDIOCCAPTURE ioctl somewhat.

voodoo-capture.c

Reworked the busy loop in the voodoo3_MakeRoom function use a while loop.

Added a call to the new function v4l_set_palette an effort to set the overlay palette correctly.  This may be unneeded, and is not fully implemented yet.  I'm thinking it wil be needed when frame grabbing is done.

Moved v4l_voodoo_irq function to voodoo-v4l.c.  I we worked the busy loop to use a while loop, and hopefully not lock the machine down hard.  Yet to see it in action.

voodoo-overlay.c

voodoo.h

voodoo-capture.c

voodoo-overlay.c

voodoo-tuner.h

voodoo-v4l.c

It is possible to use the kernel versiones of tvmixer and tuner.  They need to be patched with the diffs in here.  To get them to compile, apply the patches, and add the vpx3224 module as well.

voodoo-audiochip.h

voodoo-i2c.c

voodoo-mixer.c

voodoo-msp3400.c

There are two significant changes that have taken place.  The first is no longer being dependent on i2c_get_client.  I looked at i2c-compat.h again, and something clicked when I took a momnet to understand the i2c_clients_command function.  I had seen some thing similar in the voodoo source, and sure enough, we have voodoo_call_i2c_client in voodoo-i2c.c.  Removing the i2c_get_client calls means we don't have access to the client structure, so the printk's that accompanied them were no longer relevant.  This reduced each function in voodoo-func.c that called i2c_get_client to just 2 or four calls to voodoo_call_i2c_client.  I opted to move the calls to voodoo_call_i2c_client to the functions that call into voodoo-func.c.  I added a printk to voodoo_call_i2c_client that will identify any i2c client that is not available.

I then looked at the functions in voodoo-func.c that called the voodoo3_conf function.  They were one liners, or a single conditional to enable or disable a feature.  I chose to move these to the calling functions as well.

That only left voodoo3_print_status, which is only used by vdebug.c.  I've chosen not to get involved with the vdebug.c, and related files.  They will not be updated, and will definitly be broken.  If write access to the registers is needed, I'll add a proc/sys/dev/sensors directory like I've started to do for the vpx3224 module.

Inlcuding i2c-compat.h also provides i2c_get_clientdata, which is only in the 260 kernel.  this eliminates some preprocessor conditionals.

voodoo.c

voodoo-i2c.c

voodoo-overlay.c

voodoo-v4l.c

voodoo-msp3400.c

vpx322x.c

vpx322x-i2c.c

I've spent the last couple of weeks working on a new vpx3224 module.  It combines the best features of the vpx322x we all know and love, and the structure and methods of the vpx3220 module found in the 2.6.0 kernel, plus a few features from other places.

Many changes have been made since christmas... Where to begin?

The file video_decoder.diff needs to be applied.  This one liner removes the dependency between v3tv and vpx322x.

Added i2c-compat.h to eliminate some of the conditional compile statements.

voodoo.c

voodoo-func.c

voodoo-mixer.c

voodoo-msp3400.c

voodoo-overlay.c

voodoo-v4l.c

vpx3224-i2c.c

vpx3224.c

My sisters computer went Fitz, so with the stores closed, I've sacrificed my computer with the STB TV-PCI card, yielding a supposedly functioning power supply. Turns out the PS fan has failed, so I'll be buying two PS before the year is out. The up shot of this is that I can't do any further testing of tv-out for now. I'd gotten as far as the static patterns changing when I tinkered with the settings in /proc/sys/dev/sensors/bt869*/

Following Frank's guide again, I've added/modified as appropriate, and commented out those new sections that are as yet not implemented. The module compiles and loads clean. Testing shall have to wait for v3tv.o to be ready. Some parts of the attach function, tvmixer_adapters, that are commented out should be implemented sooner rather than later. It appears to me that the detach function, tvmixer_clients, actually does more than just detach, and evaluates-performs some commands. I haven't yet grokked how this distinction is made. Maybe I'm reading more into it than really exists.

Thank you Lars Munch! I made only a few trivial changes. Removed the deprecated 'default:' label at the end of the tuner_command() case structure. Added decoder statements to tuner_attach(). They are commented out until I understand the correct usage. Changed the i2c_drive struct from a coma separated list to a 'label: value,' structure. Gave the i2c_driver driver structure a more explicit name. Gave the i2c_client structure a name. Added MODULE_LICENSE("GPL"); to silence modprobe.

Again, thanks go to to Lars. I made similar trivial changes as with the tuner.

voodoo.c : no significant change
voodoo-i2c.c : Removed voodoo3_inc/dec in
voodoo-func.c : This is all high level stuff.
voodoo-util.c : PCI regs and Memory management
voodoo-overlay.c : No i2c struct or funct defined.
voodoo-capture.c : No i2c struct or funct defined.
voodoo-v4l.c : No i2c struct or funct defined.

Everything builds clean, except for 4 warnings that have been there for the longest time. I'll comment out the relevent sections, as they are all 'defined but not used' messages.

I'm happy to say that it WORKS!! Just as it always has! xawtv still blanks when the mouse cursor moves between various windows. It is so good to have sensors and TV at the same time!

Both of these modules compile and install with out modification. I did add MODULE_LICENSE("GPL"); to them, to silence the complaint by modprobe about a tainted kernel.

I wrote a few small scripts that grep /proc/voodoo-dump for the relevant sections. They can be found in the scripts directory. I put them in ~/bin on my system. I'll add one to make it simple to change the bt869 proc entries as well.

I'm going to try bt869.c from lm_sensors 2.8.1. The practical approach is to include bt869 in the kernel config. There is some discussion in the archives that voodoo-i2c will conflict with lm_sensors. We shall see.

It seems I can't do any thing to test the bt869 module with out i2c-voodoo from lm_sensors, or the v3tv module, which links the voodoo-i2c.o object file. I'll have to patch the kernel with lm_sensors before I can further test with i2c-voodoo3.

I've made some changes to voodoo-bt869.c to bring it in line with I2C 2.8.1. I'll also test with this module and lm_sensors i2c-voodoo3.

Following Frank's guide, I've added a few sections, and modified some things. I think I've covered all the bases. The module compiles cleanly and inserts without compliant. If there were an easy way to test it...

Some of the other modules I've looked at as references use the struct i2c_client and client_template. I saw no simple way or any immediate reason to try to implement this feature at this time.

voodoo-i2c.c no longer refrences voodoo.c functions voodoo3_{inc|dec}. The module v3tv is no longer having its use count updated!