Those are interesting suggestions. One well-known problem is that you can't easily pass arguments to non-modular drivers. In any case, the arguments for scanning (ignore_range, etc.) are only supported in the lm_sensors drivers using i2c-proc.[ch]. They aren't in the media/video drivers. If you set an ignore parameter at kernel boot (see doc/chips/MODPARMS for syntax) then all the sensors drivers should see them, which may be what you want to keep the w83781d module from scanning the matrox bus. Each driver controls how much of the bus it scans (through normal_i2c and normal_i2c_range). It looks like at least some of the media/video driver do set these. You may wish to see if the drivers you are using are setting these correctly. I looked at the matrox timeout, it is 100 jiffies (= 1 sec). It could be less. On the via686 PWM, there's nothing about it in my Via docs. Either it's undocumented, it's a newer version of the chip, it uses GPIO, or there's a separate PWM chip. Marek Michalkiewicz wrote: > On Sat, Jun 21, 2003 at 12:36:26PM -0400, Mark D. Studebaker wrote: > >>If you have all the chip drivers compiled in, >>each one will scan each i2c bus as it is registered. >>We do not recommend compiling-in all the chip drivers, keep them as modules. > > > I have a completely non-modular kernel (CONFIG_MODULES=n) - I find it > easier that way (compiling the kernel on the faster box, then copying > just a single kernel image file to the slower one). Another advantage > of this is that the recent ptrace/kmod exploit didn't work here ;) > > Anyway, I don't have all the chip drivers compiled in, just the ones > I have in my machines (W83781D and VIA686A), which don't even have > to be accessed through I2C (ISA/PCI access works fine). Perhaps there > should be a way for an I2C bus driver like i2c-matroxfb to exclude > itself from these automatic scans for all possible chips? After all, > we know exactly what chips are on the Matrox I2C buses. > > The good old G400 does not run very hot - but some of the new hot > (literally) graphics cards could really have temp. and fan sensors ;) > > Could at least part of this I2C bus scanning be moved to user space > in the future? There could be a config file to determine which > I2C buses should be scanned for which chips, to avoid conflicts. > If auto-probing on kernel startup is disabled, there should be no > problem with even all chip drivers compiled in (RAM is cheap...). > I imagine that chips could be registered/unregistered with some > ioctl calls on an open I2C device file descriptor. > > One unrelated question - I have MSI MS-6368L v 5.0 motherboards > (with VIA VT82C686B), and the manual (bottom of page 2-17) says: > > "CPU Fan supports the fan control. You can install the PC Alert > utility that will automatically control the CPU Fan speed according > to the actual CPU temperature." > > This suggests that the VIA chip has a fan PWM output, but I could find > no mention of this in the lm_sensors docs. Do you have any more info > on this? Or is this the typical "VIA won't release docs" problem? > > Thanks, > Marek >
