On Thu, Aug 28, 2003 at 01:19:55PM -0700, Brian Kuschak wrote: > Hello, > > I work for an networking storage company, supporting > linux for our custom embedded hardware. My company > makes extensive use of the I2C bus in our products. > Some of these product are chassis/blade based and as a > result many times a given I2C bus is multiplexed into > several bus segments to prevent address conflicts when > many I2C devices exist with the same address. Some > muxes are I2C based, some based on GPIO lines or > CPLDs, etc. > > In my experience this is a common hardware solution, > but it is difficult to fit into the existing i2c > driver implementation. > > I wanted to design a nice architecture to handle these > bus multiplexors in the most transparent and simple > way possible. My solution was to create a new i2c > adapter driver (i2c-virtual.c) which registers a new > i2c adapter/bus for each multiplexed bus segment. > This virtual adapter then makes calls to the real i2c > adapter to select the mux, perform the transfer, and > deselect again. Locking of the buses is handled. The > code also prevents "aliasing" of the same I2C device > onto multiple virtual I2C buses, putting all > non-multiplexed devices on the real bus, and only > multiplexed devices on the virtual busses. > > The result is that all the existing i2c client drivers > and sensor drivers work with this multiplexed hardware > without any changes, just by scanning the additional > buses. That sounds sane. > If anyone from this your group is interested in > reviewing or integrating this code into the official > sources, please let me know and I'll send a patch. I would be. Please send a patch against the latest 2.6 kernel and I'll work on getting it into the main kernel tree. > BTW, do you have a mailing list? I couldn't find one. You just sent your message to it :) thanks, greg k-h
