Re: [PATCH v2 -next] powerpc: kernel/time.c - cleanup warnings

[Alexandre Belloni <alexandre.belloni@bootlin.com>]

On 24/03/2021 09:19:58+0100, Geert Uytterhoeven wrote:
> Hi Alexandre,
> 
> On Tue, Mar 23, 2021 at 11:18 PM Alexandre Belloni
> <alexandre.belloni@bootlin.com> wrote:
> > On 23/03/2021 05:12:57-0400, He Ying wrote:
> > > We found these warnings in arch/powerpc/kernel/time.c as follows:
> > > warning: symbol 'decrementer_max' was not declared. Should it be static?
> > > warning: symbol 'rtc_lock' was not declared. Should it be static?
> > > warning: symbol 'dtl_consumer' was not declared. Should it be static?
> > >
> > > Declare 'decrementer_max' and 'rtc_lock' in powerpc asm/time.h.
> > > Rename 'rtc_lock' in drviers/rtc/rtc-vr41xx.c to 'vr41xx_rtc_lock' to
> > > avoid the conflict with the variable in powerpc asm/time.h.
> > > Move 'dtl_consumer' definition behind "include <asm/dtl.h>" because it
> > > is declared there.
> > >
> > > Reported-by: Hulk Robot <hulkci@huawei.com>
> > > Signed-off-by: He Ying <heying24@huawei.com>
> > > ---
> > > v2:
> > > - Instead of including linux/mc146818rtc.h in powerpc kernel/time.c, declare
> > >   rtc_lock in powerpc asm/time.h.
> > >
> >
> > V1 was actually the correct thing to do. rtc_lock is there exactly
> > because chrp and maple are using mc146818 compatible RTCs. This is then
> > useful because then drivers/char/nvram.c is enabled. The proper fix
> > would be to scrap all of that and use rtc-cmos for those platforms as
> > this drives the RTC properly and exposes the NVRAM for the mc146818.
> >
> > Or at least, if there are no users for the char/nvram driver on those
> > two platforms, remove the spinlock and stop enabling CONFIG_NVRAM or
> > more likely rename the symbol as it seems to be abused by both chrp and
> > powermac.
> 
> IIRC, on CHRP LongTrail, NVRAM was inherited from CHRP's Mac ancestry,
> not from CHRP's PC ancestry, and thus NVRAM is not the one in the
> mc146818-compatible RTC.
> 
> http://users.telenet.be/geertu/Linux/PPC/DeviceTree.html confirms that,
> showing that nvram is a different device node than rtc.
> 

Yes, what I missed was the ifdefery in drivers/char/nvram.c that makes
it a completely different driver on both platforms. I tend to forget
about that as reading this driver is not a pleasant experience. I would
really like to get rid of the x86 part which would in turn allow to
remove the global rtc_lock spinlock on all architectures.

-- 
Alexandre Belloni, co-owner and COO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com