Luca Tettamanti wrote: > Il Sun, Jan 04, 2009 at 08:14:25PM +0100, Hans de Goede ha scritto: >> Luca Tettamanti wrote: >>> 2 has been taken care of >> Nice! One small mistake though, you forgot to do the value *= 100 when >> its a temp sensor in the show_limitX functions. > > Ops, fixed. > >>> will do 3. >> Ok. > > It was simpler than I thought: I implemented per-attribute caching since > updating all the values may take a long time (here unconnected fans take > about 0.5s); the interval between reading is 1s (due to the high latency > of the fan sensor it probably makes little sense to lower this value). > Erm, your current code is not correct (as in will not work under all circumstances). Your current code initializes last_updated to 0 for all new sensors, and you do not use a valid flag (as the other hwmon drivers do). Notice that jiffies can wrap! And time_after handles this, so to make this easy lets say that the jiffies counter is only 8 bits, then a counter value of 128 is not after 0, but rather before. So if your driver gets loaded (or is not used before) the jiffies counter reaching over half its range, then the current time will be seen as before the 0 init value of last_updated, so no update will be done. This is why all the other hwmon drivers have a combo of a valid flag and a last_updated counter, and first check the flag, and only when that is set (which happens after the first caching, when last_updated also gets set to something sane) then check the last_updated counter. Regards, Hans
