On Tue 2017-06-20 14:24:00, Alexandre Belloni wrote: > On 20/06/2017 at 14:10:11 +0200, Pavel Machek wrote: > > On Tue 2017-06-20 12:03:48, Alexandre Belloni wrote: > > > On 20/06/2017 at 11:35:08 +0200, Benjamin Gaignard wrote: > > > > rtc_time_to_tm() and rtc_tm_to_time() are deprecated because they > > > > rely on 32bits variables and that will make rtc break in y2038/2016. > > > > > > Please don't, because this hide the fact that the hardware will not > > > handle dates in y2038 anyway and as pointed by Russell a few month ago, > > > rtc_time_to_tm will be able to catch it but the 64 bit version will > > > silently ignore it. > > > > Reference? Because rtc on PCs stores date in binary coded decimal, so > > it is likely to break in 2100, not 2038... > > I'm not saying it should be done but clearly, that is not the correct > thing to do for RTCs that are using a single 32 bits register to store > the time. > You give one example, I can give you three: armada38x, at91sam9, > at32ap700x and that just in the beginning of the series. I wanted reference to Russell's mail. > And yes, on PC, they will break in 2100, other in 2106, some in 2070. > I've delayed the tree wide patching until I manage to finish reworking > the infrastructure needed to handle the limits of the RTCs. Kernel does not know if it has working RTC or not, so it should use 64bit variables. Merging this -- does not hide anything. Yes, some drivers still may have problems -- so we fix the drivers. You said: > > > rtc_time_to_tm will be able to catch it but the 64 bit will > > > silently ignore it. If that's the case, we may need 64bit version to be smarter.. but we'll still need 64bit variables in the kernel. Pavel -- (english) http://www.livejournal.com/~pavelmachek (cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
Attachment:
signature.asc
Description: Digital signature
