[RFC 1/2] AT91: Support SAM9260 and SAM9G20-based boards in the same kernel image

[linux@arm.linux.org.uk (Russell King - ARM Linux)]

On Tue, Jan 18, 2011 at 06:18:14PM +0100, Gilles Chanteperdrix wrote:
> Russell King - ARM Linux wrote:
> > On Tue, Jan 18, 2011 at 04:00:30PM +0100, Jean-Christophe PLAGNIOL-VILLARD wrote:
> >> this idea is to do not have CONFIG_ARCH_AT91SAM9G20 anymore only 9260
> >>
> >> and detect it
> >>
> >> but the man issue is the CLOCK_TICK_RATE which is different between both of
> >> them except if we use a common one for those soc or the sam9/11 we could not
> >> put them in the same kernel
> > 
> > If you switch to clocksource/clockevents, then I think CLOCK_TICK_RATE
> > is irrelevant as time advances according to the interval measured by
> > the previous and current clocksource read, rather than 1/HZ intervals.
> > 
> > However, I'm never happy to say "just set CLOCK_TICK_RATE to some random
> > value that's a multiple of HZ" because I can't convince myself that these
> > don't have any effect when using clocksources.  The list of symbols which
> > depend on CLOCK_TICK_RATE are:
> > 
> > ACTHZ
> > LATCH
> > TICK_NSEC
> > TICK_USEC_TO_NSEC
> > LOW_RES_NSEC
> > MONOTONIC_RES_NSEC
> > NSEC_PER_JIFFY
> > KTIME_LOW_RES
> > 
> > and if you grep for those outside of arch/, you find them being used in
> > a fair amount of code under kernel/, as well as the odd driver here and
> > there.
> 
> at91rm9200_time.c even seems to use LATCH, even though the clockevent
> frequency is not explicitly set to CLOCK_TICK_RATE.

Well, it seems I'm not entirely right - some things only advance one
tick:

void tick_handle_periodic(struct clock_event_device *dev)
{
        int cpu = smp_processor_id();
        ktime_t next;

        tick_periodic(cpu);

...
static void tick_periodic(int cpu)
{
        if (tick_do_timer_cpu == cpu) {
                write_seqlock(&xtime_lock);

                /* Keep track of the next tick event */
                tick_next_period = ktime_add(tick_next_period, tick_period);

                do_timer(1);
                write_sequnlock(&xtime_lock);
        }

        update_process_times(user_mode(get_irq_regs()));
        profile_tick(CPU_PROFILING);
}

void do_timer(unsigned long ticks)
{
        jiffies_64 += ticks;
        update_wall_time();
        calc_global_load(ticks);
}

but update_wall_time() will advance by the delta between the last
clocksource read and the current clocksource read:

void update_wall_time(void)
{
        clock = timekeeper.clock;

#ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
        offset = timekeeper.cycle_interval;
#else
        offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
#endif

and it then uses 'offset' to adjust current time.