On 03/10/2019 16:49:06+0200, Lukasz Majewski wrote: > Hi Alexandre, > I'm rather thinking about following use cases: > > I. Adjusting time: > > 1. I start with time < 01.01.2099 > > 2. I issue ioctl to set the time to e.g. 2100 > > - When driver receives such request I setup century bits > > - and also perform in kernel driver time correction (and store > corrected time in RTC) > > 3. Subsequent reads from rtc use century bits to provide the time > (after year 2100). Century bits are set, so the correction may be > performed if needed. > > > II. The system is started at year 2098 and is supposed to run for e.g. 3 > years: > > 1. The time is read from the rtc - the "passing" of centuries need to > be detected. > > From the documentation [1] (point 4.5): > > "The two century bits, CB1 and CB0, are bits 7 and 6, respectively, in > the Month / Century register at address 06h. Together, they comprise a > 2 - bit counter which increments at the turn of each century. CB1 is > the most significant bit." > > If those bits increment when we pass century boundaries, we can detect > this fact and correct time when ioctl is issued. > No, you can't because you simply don't know if you still need to correct the time or if you already did it the last time the system was started. Example: Date is set to 2100-02-28, some time pass, the rtc now thinks it is 2100-02-29. You correct it to 2100-03-01, fine. Now, date is set to 2100-02-28, the system is shutdown, some time pass, it starts on 2100-03-02, the rtc thinks 2100-03-01 you can't correct it because you can't know whether a day has been missed. > > The only useful range for an RTC is its fully contiguous range. > > Does the automatic increment of century bits count to "contiguous > range" ? > No, because of the leap day issue. -- Alexandre Belloni, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
