Re: [rtc-linux] Re: [PATCH 4/9 RESEND] RFC: timekeeping: introduceflag systime_was_set

[John Stultz]

On 06/14/2013 11:05 AM, Alexander Holler wrote:
> Am 14.06.2013 19:41, schrieb John Stultz:
> > On 06/14/2013 09:52 AM, Alexander Holler wrote:
> > > In order to let an RTC set the time at boot without the problem that a
> > > second RTC overwrites it, the flag systime_was_set is introduced.
> > >
> > > systime_was_set will be true, if a persistent clock sets the time at
> > > boot,
> > > or if do_settimeofday() is called (e.g. by the RTC subsystem or
> > > userspace).
> > >
> > > Signed-off-by: Alexander Holler <holler@xxxxxxxxxxxxx>
> > > ---
> > >   include/linux/time.h      |  6 ++++++
> > >   kernel/time/timekeeping.c | 10 +++++++++-
> > >   2 files changed, 15 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/include/linux/time.h b/include/linux/time.h
> > > index d5d229b..888280f 100644
> > > --- a/include/linux/time.h
> > > +++ b/include/linux/time.h
> > > @@ -129,6 +129,12 @@ extern int update_persistent_clock(struct
> > > timespec now);
> > >   void timekeeping_init(void);
> > >   extern int timekeeping_suspended;
> > > +/*
> > > + * Will be true if the system time was set at least once by
> > > + * a persistent clock, RTC or userspace.
> > > + */
> > > +extern bool systime_was_set;
> > > +
> >
> > Probably should make this static to timekeeping.c and create an accessor
> > function so you don't have to export locking rules on this.
> >
> >
> > >   unsigned long get_seconds(void);
> > >   struct timespec current_kernel_time(void);
> > >   struct timespec __current_kernel_time(void); /* does not take
> > > xtime_lock */
> > > diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
> > > index baeeb5c..07d8531 100644
> > > --- a/kernel/time/timekeeping.c
> > > +++ b/kernel/time/timekeeping.c
> > > @@ -37,6 +37,9 @@ int __read_mostly timekeeping_suspended;
> > >   /* Flag for if there is a persistent clock on this platform */
> > >   bool __read_mostly persistent_clock_exist = false;
> > > +/* Flag for if the system time was set at least once */
> > > +bool __read_mostly systime_was_set;
> > > +
> > Probably should also move this to be part of the timekeeper structure
> > (since it will be protected by the timekeeper lock.
>
> I wanted to avoid locks for this silly flag at all. It is only set 
> once at boot (and resume) and set to 0 at suspend. And I don't see any 
> possible race condition which could make a lock necessary. Therefor 
> I've decided to not use a lock or atomic_* in order to skip any delay 
> in setting the time.

Even so, having random flag variables with special rules being exported 
out is likely to cause eventual trouble (someone will mis-use or 
overload some meaning on it).

So at least providing a accessor function for non-timekeeping.c uses 
would be good.


> Of course, I might be wrong and there might be a use case where 
> multiple things do set the system time concurrently and nothing else 
> did set system time before, but I found that extremly unlikely.

Yea, the condition check and the action won't be both be done under a 
lock, so its likely going to be racy anyway.


> > >   static inline void tk_normalize_xtime(struct timekeeper *tk)
> > >   {
> > >       while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
> > > @@ -498,6 +501,9 @@ int do_settimeofday(const struct timespec *tv)
> > >       raw_spin_lock_irqsave(&timekeeper_lock, flags);
> > >       write_seqcount_begin(&timekeeper_seq);
> > > +    systime_was_set = true;
> > > +
> > > +
> > >       timekeeping_forward_now(tk);
> > >       xt = tk_xtime(tk);
> >
> > Might also want to add the flag to inject_offset as well, since that
> > could be used to set the time.
>
> I wasn't sure about that because I had only a quick look at 
> inject_offset() and had the impression it's only able to inject a 
> relative small offset (so not usable at boot). And, as written 
> sometimes before, I haven't had a deep look at suspend/resume, which 
> might be the only place where it is really used to set the clock when
> systime_was_set is false.

Not via suspend/resume, since those modify the boottime to account for 
the sleep time that has past.

I'm thinking via adjtimex ADJ_SETOFFSET (which is relatively new, and 
not widely used). See the do_adjtimex() path in timekeeping.c


thanks
-john

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/