On Tue, Aug 18, 2015 at 01:53:27PM +0200, Sebastian Hesselbarth wrote:

On 08/18/2015 01:38 PM, Antoine Tenart wrote:

quoted

+	val = berlin_pwm_readl(pwm, pwm_dev->hwpwm, BERLIN_PWM_CONTROL);
+	val &= ~BERLIN_PWM_PRESCALE_MASK;
+	val |= prescale;
+	berlin_pwm_writel(val, pwm, pwm_dev->hwpwm, BERLIN_PWM_CONTROL);
+
+	berlin_pwm_writel(duty, pwm, pwm_dev->hwpwm, BERLIN_PWM_DUTY);
+	berlin_pwm_writel(period, pwm, pwm_dev->hwpwm, BERLIN_PWM_TCNT);

The reason why I usually tend to _not_ use _relaxed() in low-performance
setup code is that you'll have to think about reordering issues when
using _relaxed ones.

If that is your concern, then you should read Documentation/memory-barriers.txt,
specifically the section on "ACQUIRES VS I/O ACCESSES".  Using the non-
relaxed accessors doesn't save you in every circumstance.

The question here is: Is it _guaranteed_ that above writel_relaxed()
will be issued _before_ actually releasing the spin_lock?

There is no guarantee on all hardware that writel() will be seen by the
hardware before other accesses within a separate spinlocked region
using the same lock.

-- 
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.