On 26/01/2016 at 06:56:40 -0800, Joshua Clayton wrote : > Ping? > > On Monday, January 04, 2016 10:31:18 AM Joshua Clayton wrote: > > Alexandre, > > this is the second version of apatch set to be able to adjust the > > clock on a pcf2123, but with a big addition. I am also proposing > > a couple of new rtc api functions and a new sysfs file. > > > > Patches 1-5 are very much v2 patches that address All of the > > concerns you had with v1. It is my hope to get them merged without > > too much more trouble. > > > The rtc _needs_ patches 1-5 in order to use the clock offset > adjustment at all (current upstream code resets it to zero during > device probe, so it doesn't survive a reboot) > Should I split out patches 1-5 to make them easier to review? >
The merge widows just closed so I've started to collect a few patches since monday. This is a big series and I didn't have time to review everything yet. > > Patches 6-8 are attemping to add a new function to the rtc class > > to adjust the clock rate. I hope I am going about this the right way, > > But here goes. > > > I am hoping to have some discussion about 6, 7 and 8. > Particularly whether parts per billion is the right unit (I think it is) > and whether "offset" is the best name for the attribute (not as sure). > Yeah, My plan was to take 1 to 5 and leave out 6-8 for discussion, I'm not yet sure what the correct api should look like. > > A number of rtc devices, such as the NXP pcf2123 include a facility > > to adjust the clock in order to compensate for temperature or a > > crystal, capacitor, etc, that results in the rtc clock not running > > at exactly 32.768 kHz. > > > > This patchset adds kernel and sysfs hooks to access that ability. > > > > One datasheet suggests it might be adjusted based on input from > > a temperature sensor. I could also potentially see it being set > > as part of ntp calibration. > > > > Data sheets I have seen refer to this as a clock offset, and measure it > > in parts per million (ppm), however they often reference ppm to 2 digits > > of precision, which makes integer ppm less than ideal. I use parts per > > billion, which more than covers the precision needed and works nicely > > within 32 bits > > > > The name "offset" came from the pcf-2123 datasheet and is used by at > > least some other data sheets. I would be happy to use a different term > > if someone else comes up with something more concise. > > > > Changes since v1: > > - Use the BIT() macro for all register bits > > - Remove unneeded range checks from read/write functions > > - Merge patch 3 (replace magic numbers with defines) into patch 2 > > - Add a proper commit message to patch 5, now patch 4 > > - Fix a function alignment bug. > > - Move OSC_HAS_STOPPED check into pcf2123_rtc_read_time, and get rid of > > pcf2123_time_valid() > > - Drop patches refactoring pcf2123 sysfs. > > - Add rtc interface and rtc sysfs file for clock offset > > > > Joshua Clayton (8): > > rtc-pcf2123: define registers and bit macros > > rtc-pcf2123: clean up reads from the chip > > rtc-pcf2123: clean up writes to the rtc chip > > rtc-pcf2123: refactor chip reset into a function > > rtc-pcf2123: avoid resetting the clock if possible > > rtc: Add functions to set and read clock offset > > rtc: implement a sysfs interface for clock offset > > rtc-pcf2123: implement read_offset and set_offset > > > > drivers/rtc/interface.c | 57 ++++++++++ > > drivers/rtc/rtc-pcf2123.c | 271 > > +++++++++++++++++++++++++++++++++------------- > > drivers/rtc/rtc-sysfs.c | 29 +++++ > > include/linux/rtc.h | 4 + > > 4 files changed, 284 insertions(+), 77 deletions(-) > > > > > Thanks again, > Joshua Clayton -- Alexandre Belloni, Free Electrons Embedded Linux, Kernel and Android engineering http://free-electrons.com
