On Thu, Apr 04, 2019 at 07:54:52AM +0100, Lee Jones wrote: > On Thu, 04 Apr 2019, Vaittinen, Matti wrote: > > > Actually, now that I thik of it the right way to do this would have > > been the function pointer in parent data as was done in original patch > > set. HW-colleagues tend to re-use HW blocks, and we like to re-use our > > drivers. If the next PMIC from ROHM uses same RTC block but does not > > provide watchdog - then it is cleanest solution to fall back to > > function pointer and leave it to NULL when there is no WDT or when WDT > > is unused. Another option is to export dummy function - which is not so > > nice. > > I think the converse is true. > > Pointers to functions outside of a subsystem API context are generally > horrible. It's much nicer to call a function which can be easily > stubbed out in a header file based on a Kconfig option. It's how most > kernel APIs work. I hate to admit but I see your point. This nicely solves any issues in syncronizing the startup for driver providing function pointer and for driver using it. > > Additional benefit from function pointer would have been that the > > function pointer can be only used by drivers which have acces to it. > > This exported function is globally visible. The WDT disable/enable is > > very specific procedure and it actually would be nicer design to not > > have it visible globally. It is not intended to be used by anything > > else besides the WDT and RTC here. > > Why would anything else try to use it? > > There are 1000's of exported functions in the kernel. If it's > properly namespaced a consumer would have to purposely call it, which > if they really wanted to, they could do anyway. I don't really see > your point. I could still argue on this. It _is_ less obvous that an exported function is not meant to be publicly used than it is for function pointers. But as you say, this is not a strong enough point to see the trouble in synchronizing the WDT/RTC startup. > > But I can't say there will be PMIC with same RTC and no WDT coming from > > ROHM. Also, I am not terribly excited about the option of changing this > > back to function-pointer as I already removed the pointer from parent > > data and this changed parent data is already adapted to all sub drivers > > - so this is all just babbling. Maybe it is just my huge ego shouting > > there - 'I was right, I must have the final say'. > > No, a call-back function would be a back-step. You are probably right. > Ego or no ego, you're wrong. =:-D I'd rephrase that as "It's not that I am wrong, but you are right." =) > > As a side note, I already did submit v12 with other styling fixes but > > which left the WDT function in MFD. If you still see the WDT functions > > should be exported from WDT - then please ignore the v12. I'll do v13 > > at the afternoon (my time, which is only a bit after noon your time I > > guess) which will export these functions from WDT. (Well, I had to try > > once more :D) > > Please keep the WDT code in the WDT driver. Create a little stub for > the cases where the WDT driver is not enabled - job done. Yes Sir. Br, Matti Vaittinen -- Matti Vaittinen, Linux device drivers ROHM Semiconductors, Finland SWDC Kiviharjunlenkki 1E 90220 OULU FINLAND ~~~ "I don't think so," said Rene Descartes. Just then he vanished ~~~
