Any one using i2c-voodoo3 module on the 2.6 kernel?
Hello,
I've made some progress with the v3tv project. Using the recent 2.4 kernels, with various patches for v4l2, i2c-2.8.x, and sensors-2.8.x, I've got all of the v4l modules to attach to the i2c-voodoo3 module.
I've also got a v4l2 radio device working on the 2.4 kernel :)
Now I'd like to start looking at the 2.6 kernel. At this point it does not seem that the i2c-voodoo3 module is loading correctly. The module is totaly silent in the logs, so I can't offer any immediate list of symptoms, except to say I can find no reference to it in the sysfs and proc directories.
The client module probe functions never see the Voodoo3 adapter.
The call to pci_module_init returns zero. Replacing that with a call to pci_register_driver yeilds a one, but this may not be representative, since pci_register_driver always returns a non-zero value.
I'm going to add more printk's to the pci-driver.c file, and reboot when I have time for it.
I'd appreciate any advice on where to look first, and where in sysfs to look for evidence that the module loaded correctly.
Has anyone used the Banshee or Voodoo3 card with the i2c-voodoo3 module in the 2.6 kernel with success?
Thanks,
Perry
-----
Projects:
V3TV: http://www.gilfillan.org/v3tv/
VPX3224: http://www.gilfillan.org/vpx3224/
V3TV-radio: http://www.gilfillan.org/v3tv-v4l2/
snd-tvmixer: http://www.gilfillan.org/ALSA/
Quoted 5 lines
Now I'd like to start looking at the 2.6 kernel. At this point it does not seem that the i2c-voodoo3 module is loading correctly. The module is totaly silent in the logs, so I can't offer any immediate list of symptoms, except to say I can find no reference to it in the sysfs and proc directories.
It happens that I have an old Voodoo 3000 graphics card lying in a box somewhere, so I thought I would try and plug it, and see what happens.
I was able to load the i2c-voodoo3 driver (Linux 2.6.6-rc3) and it seems to work for me. I have the following line in the logs:
voodoo3 smbus 0000:01:00.0: Using Banshee/Voodoo3 I2C device at e8964000
After loading i2c-dev, "i2cdetect -l" sees two i2c busses on the card, and I can use i2cdetect, i2cdump etc. So it seems to work fine.
I suspect that you have another driver requesting the card, such has a framebuffer driver. The 2.4 driver doesn't request the device so no conflict happens. The 2.6 driver does request it, because it is considered Bad (TM) to use a device without requesting it. This probably means that the i2c-voodoo3 driver should me merged into the framebuffer driver, much like it was done for matroxfb and radeonfb. This also makes sense because the framebuffer driver can use the DDC channel (which is no more than I2C, renamed) to retrieve information about the monitor and adapt its output. But this also raises the problem that people may want to access the i2c busses without loading the full framebuffer driver...
Quoted 4 lines
The call to
pci_module_initreturns zero. Replacing that with a call topci_register_driveryeilds a one, but this may not be representative, sincepci_register_driveralways returns a non-zero value.
I'm not very qualified about PCI stuff, but this could confirm my suspicion above.
Quoted 2 lines
I'd appreciate any advice on where to look first, and where in sysfs to look for evidence that the module loaded correctly.
See what is done in proc/dump/i2cbusses.c (lm_sensors package). Looks like you should point your shell to /sys/class/i2c-dev. This suggests that you have actually loaded i2c-dev.
Quoted 2 lines
Has anyone used the Banshee or Voodoo3 card with the i2c-voodoo3 module in the 2.6 kernel with success?
Yes ;) although I didn't do anything useful with it after that.
Signature · 2 lines
Jean Delvare http://khali.linux-fr.org/
It's definitely a PCI registration conflict.
I have a modified driver that doesn't do PCI registration, and it works, and even more interesting, when I load the driver I get:
i2c-voodoo3.o version mds (mds)
i2c-voodoo3.o found device
tdfxfb 0000:01:05.0: Using Banshee/Voodoo3 I2C device at cb870000
I'll add a note on the 2.6 kernel info page about voodoo.
Quoted 55 lines · Jean Delvare
Quoted 5 lines
Now I'd like to start looking at the 2.6 kernel. At this point it does not seem that the i2c-voodoo3 module is loading correctly. The module is totaly silent in the logs, so I can't offer any immediate list of symptoms, except to say I can find no reference to it in the sysfs and proc directories.
It happens that I have an old Voodoo 3000 graphics card lying in a box somewhere, so I thought I would try and plug it, and see what happens.
I was able to load the i2c-voodoo3 driver (Linux 2.6.6-rc3) and it seems to work for me. I have the following line in the logs:
voodoo3 smbus 0000:01:00.0: Using Banshee/Voodoo3 I2C device at
e8964000After loading i2c-dev, "i2cdetect -l" sees two i2c busses on the card, and I can use i2cdetect, i2cdump etc. So it seems to work fine.
I suspect that you have another driver requesting the card, such has a framebuffer driver. The 2.4 driver doesn't request the device so no conflict happens. The 2.6 driver does request it, because it is considered Bad (TM) to use a device without requesting it. This probably means that the i2c-voodoo3 driver should me merged into the framebuffer driver, much like it was done for matroxfb and radeonfb. This also makes sense because the framebuffer driver can use the DDC channel (which is no more than I2C, renamed) to retrieve information about the monitor and adapt its output. But this also raises the problem that people may want to access the i2c busses without loading the full framebuffer driver...
Quoted 4 lines
The call to
pci_module_initreturns zero. Replacing that with a call topci_register_driveryeilds a one, but this may not be representative, sincepci_register_driveralways returns a non-zero value.I'm not very qualified about PCI stuff, but this could confirm my suspicion above.
Quoted 2 lines
I'd appreciate any advice on where to look first, and where in sysfs to look for evidence that the module loaded correctly.
See what is done in
proc/dump/i2cbusses.c(lm_sensorspackage). Looks like you should point your shell to/sys/class/i2c-dev. This suggests that you have actually loaded i2c-dev.Quoted 2 lines
Has anyone used the Banshee or Voodoo3 card with the i2c-voodoo3 module in the 2.6 kernel with success?
Yes ;) although I didn't do anything useful with it after that.
Quoted 8 lines · Perry Gilfillan
Quoted 1 line · Mark Studebaker
It's definitely a PCI registration conflict.
Then there are three drivers wanting access to the Voodoo3 PCI
registers: tdfxfb, i2c-voodoo3, and v3tv.
Quoted 9 lines
It's posible that v3tv and tdfxfb will touch the same registers and break each others functions. I've thought about merging them, but don't know that I've got the ability to do it right.
As Jean indicates that the matroxfb and radeonfb drivers have the DDC adapter merged, I take it merging i2c and posibly v3tv into tdfxfb might be acceptable?
I guess it's the only way to do it, right? Don't know what the v3tv people think.
Quoted 12 lines
Quoted 8 lines
I have a modified driver that doesn't do PCI registration, and it works, and even more interesting, when I load the driver I get:
i2c-voodoo3.o version mds (mds)
i2c-voodoo3.o found device
tdfxfb 0000:01:05.0: Using Banshee/Voodoo3 I2C device atcb870000I'll add a note on the 2.6 kernel info page about voodoo.
Did you do anything more than copy the
pci_for_each_devmacro?
correct. You can look at 2.4 kernel version of i2c-voodoo3 in lm_sensors to see. I just ported that over to 2.6 to test it.
Quoted 30 lines
Quoted 13 lines
Quoted 10 lines · Jean Delvare
I suspect that you have another driver requesting the card, such has a framebuffer driver. The 2.4 driver doesn't request the device so no conflict happens. The 2.6 driver does request it, because it is considered Bad (TM) to use a device without requesting it. This probably means that the i2c-voodoo3 driver should me merged into the framebuffer driver, much like it was done for matroxfb and radeonfb. This also makes sense because the framebuffer driver can use the DDC channel (which is no more than I2C, renamed) to retrieve information about the monitor and adapt its output. But this also raises the problem that people may want to access the i2c busses without loading the full framebuffer driver...
Thanks,
Perry
------
Gilfillan Family: http://www.gilfillan.org/Projects:
V3TV: http://www.gilfillan.org/v3tv/
VPX3224: http://www.gilfillan.org/vpx3224/
V3TV-radio: http://www.gilfillan.org/v3tv-v4l2/
snd-tvmixer: http://www.gilfillan.org/ALSA/