On Fri, 10 Jan 2003, Mark Studebaker wrote: > actually I meant alarms = chip | fail. > So the 'sensors' program would say "ALARM" if the read failed - > no changes to 'sensors' required. Cannot think of anything to argue there, for a single reader case. Maybe add retry in the i2c read part, in case chip inserts wait states the host does not accept. A retry from /proc is not effective for some seconds. Is there a reason why limits are read from chip in update? Seems to be the case with most drivers. If a measurement goes out of limits and returns chip's ALARM status is set and cleared on /proc update. With two or more unsynchronised readers, a reader may miss this update and transient failures. Maybe I could write a reader daemon to handle chip access and have libsensors talk with it. -- Ky?sti M?lkki kmalkki at cc.hut.fi
