i2c: lost sensors with 2.6.10(-mm1)

[khali@linux-fr.org (Jean Delvare)]

J.A. Magallon wrote:

> I have lost my sensors info with 2.6.10, in particular -mm1.
> They work fine with 2.6.9-mm1 (current state of the box, booted on
> 2.6.9 or 10, no other difference).
 > (...)
> I have noticed different contents in /sys:
> under 2.6.9:
> /sys/devices/platform/i2c-1:
> /sys/devices/platform/i2c-1/1-0290:
> /sys/devices/platform/i2c-1/1-0290/power:
> /sys/devices/platform/i2c-1/power:
> 
> under 2.6.10:
> /sys/devices/platform/i2c-1:
> /sys/devices/platform/i2c-1/power:
> 
> So some /sys nodes are missing !!!
> (the isa bus)

This basically means that the i2c client was not registered.

> Debug output from 2.6.10-mm1:
> (...)
> Jan  7 01:33:11 werewolf kernel: i2c-core: driver w83627hf registered.
> Jan  7 01:33:11 werewolf kernel: i2c_adapter i2c-1: found normal isa entry for adapter 9191, addr 0290

However, this suggests that the driver loaded properly and the base 
address was correctly read from Super-I/O space. This would mean that 
the problem happened later, in w83627hf_detect(). The most likely reason 
for this would be if the region request failed (unfortunately we have no 
message, not even debug, if this happens).

> Some ideas ?

Three things to try, in order:

1* Compare /proc/ioports in 2.6.9-mm1 and 2.6.10-mm1. I suspect that the 
0x290-0x297 range is held by some device in 2.6.10-mm1.

2* Try reverting this patch in 2.6.10-mm1:
http://www.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.10/2.6.10-mm1/broken-out/bk-i2c.patch
It does indeed include a change in the way the I/O region is requested. 
It should not make any difference, but maybe we are missing something 
and it actually does.

3* Try a vanilla 2.6.10 kernel and report how it is going.

Thanks,
-- 
Jean Delvare