Timer on i.MX28 moving backwards

[u.kleine-koenig@pengutronix.de (Uwe Kleine-König)]

On Fri, Sep 07, 2018 at 09:21:05AM +0200, Uwe Kleine-K?nig wrote:
> On Thu, Sep 06, 2018 at 10:39:25PM +0100, Russell King - ARM Linux wrote:
> > On Thu, Sep 06, 2018 at 10:02:02PM +0200, Uwe Kleine-K?nig wrote:
> > > +cycle_t mxs_clocksource_mmio_readl_down(struct clocksource *c)
> > > +{
> > > +	cycle_t ret = ~(cycle_t)readl_relaxed(to_mmio_clksrc(c)->reg) & c->mask;
> > > +	trace_printk("time = %llx\n", ret);
> > 
> > I think it may be worth re-running your test with:
> > 
> > 	u32 one, two;
> > 
> > 	one = readl_relaxed(to_mmio_clksrc(c)->reg);
> > 	two = readl_relaxed(to_mmio_clksrc(c)->reg);
> > 	trace_printk("time=0x%08x:0x%08x\n", one, two);
> > 
> > 	return ~(cycle_t)two & c->mask;
> > 
> > and compare the values not only between each call but also between
> > each read of the register.  It could be a synchronisation issue when
> > reading the timer register, in which case you may find just one read
> > is incorrect.
> 
> Good idea, this is a part of a trace I captured with this implemented:
> 
>        mrp62439d-224   [000] d..h2..   790.936689: mxs_clocksource_mmio_readl_down: time = 0x948e3d1a:0x948e3d14
>        mrp62439d-224   [000] .......   790.936689: mxs_clocksource_mmio_readl_down: time = 0x948e4d26:0x948e4d20
>        mrp62439d-224   [000] .......   790.936904: mxs_clocksource_mmio_readl_down: time = 0x948e3690:0x948e368a
> 
> hmm, I'm unsure what to think now.

To clearify my (obviously too short) mail:

Judging form a trace over 43s the two consecutive reads always correctly
yield a decrementing counter as expected. But there is not always the
expected monotonicity between consecutive calls to
clocksource_mmio_readl_down.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-K?nig            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |