Re: [PATCH 2/8] clocksource: sun4i: Add clocksource and sched clock drivers

From: Baruch Siach <baruch@tkos.co.il>
To: Maxime Ripard <maxime.ripard@free-electrons.com>
Cc: John Stultz <john.stultz@linaro.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Emilio Lopez <emilio@elopez.com.ar>,
	linux-kernel@vger.kernel.org, linux-sunxi@googlegroups.com,
	sunny@allwinnertech.com, shuge@allwinnertech.com,
	kevin@allwinnertech.com, linux-arm-kernel@lists.infradead.org
Subject: Re: [PATCH 2/8] clocksource: sun4i: Add clocksource and sched clock drivers
Date: Thu, 27 Jun 2013 20:36:43 +0300	[thread overview]
Message-ID: <20130627173643.GQ4824@tarshish> (raw)
In-Reply-To: <20130627172144.GD4319@lukather>

Hi Maxime,

On Thu, Jun 27, 2013 at 07:21:44PM +0200, Maxime Ripard wrote:
> On Thu, Jun 27, 2013 at 12:46:49PM +0300, Baruch Siach wrote:
> > On Thu, Jun 27, 2013 at 11:35:58AM +0200, Maxime Ripard wrote:
> > > On Thu, Jun 27, 2013 at 09:02:34AM +0300, Baruch Siach wrote:
> > > > On Wed, Jun 26, 2013 at 11:16:55PM +0200, Maxime Ripard wrote:
> > > > > +static u32 sun4i_timer_sched_read(void)
> > > > 
> > > > You commit message mentions "64 bits free running counter", but this one only 
> > > > returns 32 bit.
> > > 
> > > Yeah, the callback setup by setup_sched_clock is supposed to be
> > > returning a u32, and clocksource_mmio_init only accepts up to 32 bits as
> > > well, so I'm only using the lower 32bits of this 64 bits counter.
> > > 
> > > I'll amend the commit log to state this.
> > 
> > But using 64 bit counter for sched_clock is much easier that using 32 bit one. 
> 
> Easier in what aspect? Both API looks similar.

You can just implement your own simple sched_clock() that just returns the 
current value of this 64 bit counter, and do away with all the tricky code in 
kernel/time/sched_clock.c (in tip.git) that is needed to make the 32 -> 64 
extension safe. This is not compatible with multi-platform kernel, though.

> > You can either wait for the rest of Stephen's patch set 
> > (http://thread.gmane.org/gmane.linux.ports.arm.msm/4092) before adding 64 
> > bit, or you can just make the trivial change to the now generic 
> > sched_clock code.
> 
> I'll wait for the Stephen's patches to be merged then, and add support
> for 64bits counters to clocksource_mmio_init.

Sound reasonable.

baruch

-- 
     http://baruch.siach.name/blog/                  ~. .~   Tk Open Systems
=}------------------------------------------------ooO--U--Ooo------------{=
   - baruch@tkos.co.il - tel: +972.2.679.5364, http://www.tkos.co.il -