I was looking at common/soft_i2c.c i2c_init(). The first (only) thing it does is call send_reset(). In send_reset() the I2C_INIT macro gets run after the I2C_SCL() and I2C_SDA() macros which seems backwards to me (and somewhat goes against what's in the README). i would expect the port setup to happen first in case there's any weird requirements (clocks to be enabled, weird protection schemes, etc).

send_reset() only gets called from i2c_init() that I could see. Would there be any objections to a patch reversing the order of the calls ?