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. Jean Delvare wrote: >>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... > > >>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 not very qualified about PCI stuff, but this could confirm my > suspicion above. > > >>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. > > >>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. >
