Sorry for the very late reply ! On 03/05/2019 13:30:12+0000, Dylan Howey wrote: > The 05/02/2019 22:55, Alexandre Belloni wrote: > > > * I don't think pcf2123_read_offset is working correctly. In the case of > > > a coarse offset the value is not being sign extended. So a negative > > > offset read will not be correct if the coarse bit is set (result would > > > be a positive number being returned if this is true). I need to look > > > into this some more. The fix would be to sign extend first, then if > > > coarse bit is set multiply the result by 2. > > > > > > > As the comment says, it is properly extended because after shifting, > > bits [6:0] become bit [7:1]. > > > But what about this: > > # echo 2170 > /sys/class/rtc/rtc0/offset > # cat /sys/class/rtc/rtc0/offset > 2170 > # echo -2170 > /sys/class/rtc/rtc0/offset > # cat /sys/class/rtc/rtc0/offset > 0 > # echo 4340 > /sys/class/rtc/rtc0/offset > # cat /sys/class/rtc/rtc0/offset > 4340 > # echo -4340 > /sys/class/rtc/rtc0/offset > # cat /sys/class/rtc/rtc0/offset > -2170 > > Negative offset reads seem to be off by a factor of 2. > The issue is not the sign extension, it is the rounding function that only works properly for positive numbers: reg = (s8)((offset + (OFFSET_STEP >> 1)) / OFFSET_STEP); I'm sending a patch. -- Alexandre Belloni, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
