Re: [PATCH 1/1] x86/kernel/rtc: add sanity check for RTC date and time

[Alexandre Belloni]

On 30/06/2021 06:25:44+0000, Zeh, Werner wrote:
> Hi Alexandre
> 
> > Hello,
> > 
> > On 24/06/2021 10:15:07+0200, Werner Zeh wrote:
> > > The timekeeper is synchronized with the CMOS RTC when it is initialized.
> > > If the RTC buffering is bad (not buffered at all, empty battery) the
> > > RTC registers can contain random data. In order to avoid date and time
> > > being completely rubbish check the sanity of the registers before
> > > calling mktime64. If the values are not valid, set tv_sec to 0 so that
> > > at least the starting time is valid.
> > >
> > > Signed-off-by: Werner Zeh <werner.zeh@xxxxxxxxxxx>
> > > ---
> > > [resent due to wrong lkml address]
> > > [added RTC maintainers to the recipients] This change introduces the
> > > same validity check that is already done in drivers/rtc/interface.c.
> > > If it is not done here, the timekeeper can be set up wrongly in the
> > > first run and won't be corrected once the RTC driver is started
> > > because the validity check in the RTC driver drops the time and date
> > > due to invalid entries.
> > >
> > >  arch/x86/kernel/rtc.c | 12 +++++++++++-
> > >  1 file changed, 11 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/arch/x86/kernel/rtc.c b/arch/x86/kernel/rtc.c index
> > > 586f718b8e95..f4af7b18c6c0 100644
> > > --- a/arch/x86/kernel/rtc.c
> > > +++ b/arch/x86/kernel/rtc.c
> > > @@ -9,6 +9,7 @@
> > >  #include <linux/export.h>
> > >  #include <linux/pnp.h>
> > >  #include <linux/of.h>
> > > +#include <linux/rtc.h>
> > >
> > >  #include <asm/vsyscall.h>
> > >  #include <asm/x86_init.h>
> > > @@ -64,6 +65,7 @@ void mach_get_cmos_time(struct timespec64 *now)
> > {
> > >  	unsigned int status, year, mon, day, hour, min, sec, century = 0;
> > >  	unsigned long flags;
> > > +	struct rtc_time tm = {0};
> > >
> > >  	/*
> > >  	 * If pm_trace abused the RTC as storage, set the timespec to 0, @@
> > > -118,7 +120,15 @@ void mach_get_cmos_time(struct timespec64 *now)
> > >  	} else
> > >  		year += CMOS_YEARS_OFFS;
> > >
> > > -	now->tv_sec = mktime64(year, mon, day, hour, min, sec);
> > > +	tm.tm_sec = sec;
> > > +	tm.tm_min = min;
> > > +	tm.tm_hour = hour;
> > > +	tm.tm_mday = day;
> > > +	tm.tm_mon = mon;
> > > +	tm.tm_year = year;
> > > +	now->tv_sec = 0;
> > > +	if (rtc_valid_tm(&tm) == 0)
> > 
> > Doesn't that make the x86 architecture depend on CONFIG_RTC_LIB?
> > 
> CONFIG_RTC_LIB is already default enabled for x86, see arch/x86/Kconfig.
> Do you have any other dependencies in mind I have overseen?
> 

Nope, everything is fine, it would be better if we could get rid of
mach_get_cmos_time but I don't have any clue as to why this is
necessary.


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