Re: [RESEND] Fast TSC calibration fails with v3.14-rc1 and later

[joeyli <jlee@suse.com>]

於 四，2014-03-13 於 09:12 +0100，Thomas Gleixner 提到：
> On Thu, 13 Mar 2014, joeyli wrote:
> > 於 三，2014-03-12 於 20:59 -0700，H. Peter Anvin 提到：
> > > On 03/12/2014 08:55 PM, joeyli wrote:
> > > > 
> > > > So do not care "CMOS RTC Not Present", if TAD is present then we use it
> > > > instead of CMOS RTC in all kernel code? or we still can use CMOS RTC?
> > > > 
> > > 
> > > Why would we use *both*!?  How would that possibly make sense?
> > > 
> > > 	-hpa
> > > 
> > 
> > Yes, it does not make sense for using both.
> > 
> > I switched the code in get_rtc_time() set_rtc_time() to TAD when it
> > present, just make sure I'm on the right path.
> 
> No, you're not. get/set_rtc_time() is a complete trainwreck and as I
> said before it should move into the rtc subsystem.
> 
> There is no reason at all to keep that stuff in the arch specific
> code. It's there for historical reasons and that does not justify to
> add more mess to it.

Fully understand now. Thanks for your explanation.

> 
> So the right thing to do is:
> 
>  1) Add eventually missing functionality to the RTC subsystem  
> 
>  2) Move the arch specific cmos stuff to the rtc subsystem as proper
>     drivers 
> 
>  3) Add TAD there after #2 has been completed.
> 
> Any attempt to add TAD to arch/x86 is NACKed unconditionally.
> 
> Thanks,
> 
> 	tglx

Does there have any people already start the work of #1 and #2? 


Thanks
Joey Lee