On Wednesday, July 30, 2003, at 11:04 PM, Mark M. Hoffman wrote: > * Charles Lepple <clepple at ghz.cc> [2003-07-29 21:35:35 -0400]: >> I accidentally started up two copies of a sensord-for-2.6 kludge (just >> cats a bunch of /sys nodes, and jams the results into rrdtool). A >> little while after I started the second copy, I started seeing this in >> my kern.log: >> >> Jul 29 21:19:45 vole kernel: i2c_adapter i2c-0: SMBus collision! >> Jul 29 21:19:46 vole kernel: i2c_adapter i2c-0: : Sending abort. >> Jul 29 21:20:17 vole last message repeated 52 times > > What I2C bus driver? > > What sensors chip driver? my fault, that was an embarrassingly poor bug report. It's an AMD 76x bus controller (i2c-amd76x), and both a W83627HF and a W83782D chip controlled by the w83781d driver. This setup requires the crazy init=0 and force_subclients deal that other 760MP chipset users have reported in the trouble ticket area. Driver version is 2.7.0 (20021208). I haven't gotten around to making 2.8.0 work with the 2.6 kernel. >> removing the modules didn't help-- nothing showed up in >> /sys/bus/i2c/devices afterwards. >> >> ideas? > > I would like to try to reproduce this, but not until this weekend. > Would running 2-3 shell scripts simultaneously which cat the /sys > values to /dev/null be good enough? I would assume so. It'll probably take some random time skew between the scripts to trigger the problem, because I think there was a delay between the beginning of the script, and the beginning of the error messages. I was trying to figure out some power management and ntp interaction problems that required some extended uptime, so I didn't try and reproduce the problem after rebooting. Afterwards I happened to see some reports on LKML of interactions between ACPI and the sensor chips, but the ACPI implementation is so weak on this board (a Tyan S2460) that I doubt the ACPI portion of the BIOS even knows about the I2C bus. -- Charles Lepple <clepple at ghz.cc> http://www.ghz.cc/charles/
