Re: [PATCH v3] MIPS: Add a high resolution sched_clock() via cnt32_to_63().

[Wu Zhangjin <wuzhangjin@gmail.com>]

On Mon, 2009-11-23 at 09:44 +0100, Ingo Molnar wrote:
> * Wu Zhangjin <wuzhangjin@gmail.com> wrote:
> 
> > On Sun, 2009-11-22 at 22:03 +0300, Sergei Shtylyov wrote:
> > > Hello.
> > > 
> > > Wu Zhangjin wrote:
> > > 
> > > > From: Wu Zhangjin <wuzhangjin@gmail.com>
> > > >
> > > > (This v3 revision incorporates with the feedbacks from Ingo, Ralf and Sergei.)
> > > >
> > > > This patch adds a cnt32_to_63() and MIPS c0 count based sched_clock(),
> > > > which provides high resolution. and also, one new kernel option
> > > > (HR_SCHED_CLOCK) is added to enable/disable this sched_clock().
> > > >
> > > > Without it, the Ftrace for MIPS will give useless timestamp information.
> > > >
> > > > Because cnt32_to_63() needs to be called at least once per half period
> > > > to work properly, Differ from the old version, this v2 revision set up a
> > > > kernel timer to ensure the requirement of some MIPSs which have short c0
> > > > count period.
> > > >
> > > > Signed-off-by: Wu Zhangjin <wuzhangjin@gmail.com>
> > > > ---
> > > >  arch/mips/Kconfig           |   18 ++++++++++++++
> > > >  arch/mips/kernel/csrc-r4k.c |   55 +++++++++++++++++++++++++++++++++++++++++++
> > > >  2 files changed, 73 insertions(+), 0 deletions(-)
> > > >
> > > > diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
> > > > index b342197..5ea55f5 100644
> > > > --- a/arch/mips/Kconfig
> > > > +++ b/arch/mips/Kconfig
> > > > @@ -1952,6 +1952,24 @@ config NR_CPUS
> > > >  source "kernel/time/Kconfig"
> > > >  
> > > >  #
> > > > +# High Resolution sched_clock() Configuration
> > > > +#
> > > > +
> > > > +config HR_SCHED_CLOCK
> > > > +	bool "High Resolution sched_clock()"
> > > > +	depends on CSRC_R4K
> > > > +	default n
> > > > +	help
> > > > +	  This option enables the MIPS c0 count based high resolution
> > > > +	  sched_clock().
> > > > +
> > > > +	  If you need a ns precision timestamp, you are recommended to enable
> > > > +	  this option. For example, If you are using the Ftrace subsystem to do
> > > > +	  real time tracing, this option is needed.
> > > > +
> > > > +	  If unsure, disable it.
> > > > +
> > > > +#
> > > >  # Timer Interrupt Frequency Configuration
> > > >  #
> > > >  
> > > > diff --git a/arch/mips/kernel/csrc-r4k.c b/arch/mips/kernel/csrc-r4k.c
> > > > index e95a3cd..e287f0d 100644
> > > > --- a/arch/mips/kernel/csrc-r4k.c
> > > > +++ b/arch/mips/kernel/csrc-r4k.c
> > > > @@ -10,6 +10,57 @@
> > > >  
> > > >  #include <asm/time.h>
> > > >  
> > > > +#ifdef CONFIG_HR_SCHED_CLOCK
> > > > +#include <linux/cnt32_to_63.h>
> > > > +#include <linux/timer.h>
> > > > +
> > > > +/*
> > > > + * MIPS sched_clock implementation.
> > > > + *
> > > > + * because cnt32_to_63() needs to be called at least once per half period to
> > > > + * work properly, and some of the MIPS frequency is high, perhaps a kernel
> > > > + * timer is needed to be set up to ensure this requirement is always met.
> > > > + * Please refer to arch/arm/plat-orion/time.c and include/linux/cnt32_to_63.h
> > > > + */
> > > > +static unsigned long __read_mostly tclk2ns_scale;
> > > > +static unsigned long __read_mostly tclk2ns_scale_factor;
> > > > +
> > > > +unsigned long long notrace sched_clock(void)
> > > > +{
> > > > +	unsigned long long v = cnt32_to_63(read_c0_count());
> > > > +	return (v * tclk2ns_scale) >> tclk2ns_scale_factor;
> > > > +}
> > > > +
> > > > +static struct timer_list cnt32_to_63_keepwarm_timer;
> > > > +
> > > > +static void cnt32_to_63_keepwarm(unsigned long data)
> > > > +{
> > > > +	mod_timer(&cnt32_to_63_keepwarm_timer, round_jiffies(jiffies + data));
> > > > +	(void) sched_clock();
> > > > +}
> > > > +
> > > > +static void __init setup_sched_clock(struct clocksource *cs, unsigned long tclk)
> > > > +{
> > > > +	unsigned long long v;
> > > > +	unsigned long data;
> > > > +
> > > > +	v = cs->mult;
> > > > +	/*
> > > > +	 * We want an even value to automatically clear the top bit
> > > > +	 * returned by cnt32_to_63() without an additional run time
> > > > +	 * instruction. So if the LSB is 1 then round it up.
> > > > +	 */
> > > > +	if (v & 1)
> > > > +		v++;
> > > > +	tclk2ns_scale = v;
> > > > +	tclk2ns_scale_factor = cs->shift;
> > > > +
> > > > +	data = 0x80000000 / tclk * HZ;
> > > > +	setup_timer(&cnt32_to_63_keepwarm_timer, cnt32_to_63_keepwarm, data);
> > > > +	mod_timer(&cnt32_to_63_keepwarm_timer, round_jiffies(jiffies + data));
> > > > +}
> > > > +#endif	/* !CONFIG_HR_SCHED_CLOCK */
> > > > +
> > > >  static cycle_t c0_hpt_read(struct clocksource *cs)
> > > >  {
> > > >  	return read_c0_count();
> > > > @@ -32,6 +83,10 @@ int __init init_r4k_clocksource(void)
> > > >  
> > > >  	clocksource_set_clock(&clocksource_mips, mips_hpt_frequency);
> > > >  
> > > > +#ifdef CONFIG_HR_SCHED_CLOCK
> > > > +	setup_sched_clock(&clocksource_mips, mips_hpt_frequency);
> > > > +#endif
> > > > +
> > > >   
> > > 
> > >    According to CodingStyle #ifdef's in the code are frowned upon. I 
> > > suggest that you add an empty setup_sched_clock() definition under #else 
> > > above and have the call here without #ifdef. That's the preferred way of 
> > > doing such things...
> > > 
> > 
> > Done, thanks!
> > 
> > Wil resend it later.
> 
> The proper solution would be what i suggested in my previous mail: to 
> put it into a separate .c file. It's easy as the code has its own 
> Kconfig value already. The only #ifdef is out of sight in a .h file.
> 

Done, thanks!

Regards,
	Wu Zhangjin