[PATCH v5] MIPS: Add a high resolution sched_clock() via cnt32_to_63().

[PATCH v5] MIPS: Add a high resolution sched_clock() via cnt32_to_63().

[Wu Zhangjin]

From: Wu Zhangjin <wuzhangjin@gmail.com>

(This v5 revision incorporates with the feedbacks from Ingo.)

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/include/asm/time.h     |   15 ++++++++++
 arch/mips/kernel/Makefile        |    1 +
 arch/mips/kernel/csrc-r4k-hres.c |   54 ++++++++++++++++++++++++++++++++++++++
 arch/mips/kernel/csrc-r4k.c      |    2 +
 5 files changed, 90 insertions(+), 0 deletions(-)
 create mode 100644 arch/mips/kernel/csrc-r4k-hres.c

diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
index da5d7fd..b3b6334 100644
--- a/arch/mips/Kconfig
+++ b/arch/mips/Kconfig
@@ -1953,6 +1953,24 @@ 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/include/asm/time.h b/arch/mips/include/asm/time.h
index df6a430..a697f7d 100644
--- a/arch/mips/include/asm/time.h
+++ b/arch/mips/include/asm/time.h
@@ -84,6 +84,21 @@ static inline int init_mips_clocksource(void)
 #endif
 }
 
+/*
+ * Setup the high resolution sched_clock()
+ */
+#ifdef CONFIG_HR_SCHED_CLOCK
+extern void setup_r4k_sched_clock(struct clocksource cs, unsigned int clock);
+#endif
+
+static inline void setup_hres_sched_clock(struct clocksource cs,
+		unsigned int clock)
+{
+#ifdef CONFIG_HR_SCHED_CLOCK
+	setup_r4k_sched_clock(cs, clock);
+#endif
+}
+
 extern void clocksource_set_clock(struct clocksource *cs, unsigned int clock);
 extern void clockevent_set_clock(struct clock_event_device *cd,
 		unsigned int clock);
diff --git a/arch/mips/kernel/Makefile b/arch/mips/kernel/Makefile
index 9326af5..a6e06c0 100644
--- a/arch/mips/kernel/Makefile
+++ b/arch/mips/kernel/Makefile
@@ -24,6 +24,7 @@ obj-$(CONFIG_CSRC_BCM1480)	+= csrc-bcm1480.o
 obj-$(CONFIG_CSRC_IOASIC)	+= csrc-ioasic.o
 obj-$(CONFIG_CSRC_POWERTV)	+= csrc-powertv.o
 obj-$(CONFIG_CSRC_R4K_LIB)	+= csrc-r4k.o
+obj-$(CONFIG_HR_SCHED_CLOCK)	+= csrc-r4k-hres.o
 obj-$(CONFIG_CSRC_SB1250)	+= csrc-sb1250.o
 obj-$(CONFIG_SYNC_R4K)		+= sync-r4k.o
 
diff --git a/arch/mips/kernel/csrc-r4k-hres.c b/arch/mips/kernel/csrc-r4k-hres.c
new file mode 100644
index 0000000..2fe8be7
--- /dev/null
+++ b/arch/mips/kernel/csrc-r4k-hres.c
@@ -0,0 +1,54 @@
+/*
+ * MIPS sched_clock implementation.
+ *
+ * Copyright (C) 2009 Lemote Inc.
+ * Author: Wu Zhangjin, wuzj@lemote.com
+ *
+ * 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
+ */
+
+#include <linux/clocksource.h>
+#include <linux/cnt32_to_63.h>
+#include <linux/init.h>
+#include <linux/timer.h>
+
+static unsigned long __read_mostly cycle2ns_scale;
+static unsigned long __read_mostly cycle2ns_scale_factor;
+
+unsigned long long notrace sched_clock(void)
+{
+	unsigned long long v = cnt32_to_63(read_c0_count());
+	return (v * cycle2ns_scale) >> cycle2ns_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));
+	sched_clock();
+}
+
+void setup_r4k_sched_clock(struct clocksource cs, unsigned int clock)
+{
+	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++;
+	cycle2ns_scale = v;
+	cycle2ns_scale_factor = cs.shift;
+
+	data = 0x80000000 / clock * HZ;
+	setup_timer(&cnt32_to_63_keepwarm_timer, cnt32_to_63_keepwarm, data);
+	mod_timer(&cnt32_to_63_keepwarm_timer, round_jiffies(jiffies + data));
+}
diff --git a/arch/mips/kernel/csrc-r4k.c b/arch/mips/kernel/csrc-r4k.c
index e95a3cd..3bd89bb 100644
--- a/arch/mips/kernel/csrc-r4k.c
+++ b/arch/mips/kernel/csrc-r4k.c
@@ -32,6 +32,8 @@ int __init init_r4k_clocksource(void)
 
 	clocksource_set_clock(&clocksource_mips, mips_hpt_frequency);
 
+	setup_hres_sched_clock(clocksource_mips, mips_hpt_frequency);
+
 	clocksource_register(&clocksource_mips);
 
 	return 0;
-- 
1.6.2.1

Re: [PATCH v5] MIPS: Add a high resolution sched_clock() via cnt32_to_63().

[Ingo Molnar]

* Wu Zhangjin <wuzhangjin@gmail.com> wrote:

> From: Wu Zhangjin <wuzhangjin@gmail.com>
> 
> (This v5 revision incorporates with the feedbacks from Ingo.)
> 
> 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/include/asm/time.h     |   15 ++++++++++
>  arch/mips/kernel/Makefile        |    1 +
>  arch/mips/kernel/csrc-r4k-hres.c |   54 ++++++++++++++++++++++++++++++++++++++
>  arch/mips/kernel/csrc-r4k.c      |    2 +
>  5 files changed, 90 insertions(+), 0 deletions(-)
>  create mode 100644 arch/mips/kernel/csrc-r4k-hres.c

Looks good!

Reviewed-by: Ingo Molnar <mingo@elte.hu>

Thanks,

	Ingo

Re: [PATCH v5] MIPS: Add a high resolution sched_clock() via cnt32_to_63().

[Sergei Shtylyov]

Hello.

Wu Zhangjin wrote:

> From: Wu Zhangjin <wuzhangjin@gmail.com>

> (This v5 revision incorporates with the feedbacks from Ingo.)

> 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>

> diff --git a/arch/mips/kernel/csrc-r4k-hres.c b/arch/mips/kernel/csrc-r4k-hres.c
> new file mode 100644
> index 0000000..2fe8be7
> --- /dev/null
> +++ b/arch/mips/kernel/csrc-r4k-hres.c

    I don't think this is really good name for this file (one might think 
that this is another implementation of clocksource instead of some 
sched_clock() code tied to this particular clocksource), and I don't think 
we indeed needed to separate that thing into a file of its own, i.e. I'm 
against Ingo's suggestion in this case.

WBR, Sergei

Re: [PATCH v5] MIPS: Add a high resolution sched_clock() via cnt32_to_63().

[Ingo Molnar]

* Sergei Shtylyov <sshtylyov@ru.mvista.com> wrote:

> Hello.
> 
> Wu Zhangjin wrote:
> 
> >From: Wu Zhangjin <wuzhangjin@gmail.com>
> 
> >(This v5 revision incorporates with the feedbacks from Ingo.)
> 
> >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>
> 
> >diff --git a/arch/mips/kernel/csrc-r4k-hres.c b/arch/mips/kernel/csrc-r4k-hres.c
> >new file mode 100644
> >index 0000000..2fe8be7
> >--- /dev/null
> >+++ b/arch/mips/kernel/csrc-r4k-hres.c
> 
>    I don't think this is really good name for this file (one might 
> think that this is another implementation of clocksource instead of 
> some sched_clock() code tied to this particular clocksource), and I 
> don't think we indeed needed to separate that thing into a file of its 
> own, i.e. I'm against Ingo's suggestion in this case.

Well this patch is clearly cleaner than the previous ones - we prefer 
not to contaminate .c files with large #ifdef blocks if it can be 
avoided. (and here it can be avoided easily) YMMV.

Thanks,

	Ingo

Re: [PATCH v5] MIPS: Add a high resolution sched_clock() via cnt32_to_63().

[Sergei Shtylyov]

Hello, I wrote:

>> From: Wu Zhangjin <wuzhangjin@gmail.com>

>> (This v5 revision incorporates with the feedbacks from Ingo.)

>> 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>

>> diff --git a/arch/mips/kernel/csrc-r4k-hres.c 
>> b/arch/mips/kernel/csrc-r4k-hres.c
>> new file mode 100644
>> index 0000000..2fe8be7
>> --- /dev/null
>> +++ b/arch/mips/kernel/csrc-r4k-hres.c
> 
> 
>    I don't think this is really good name for this file (one might think 
> that this is another implementation of clocksource instead of some 
> sched_clock() code tied to this particular clocksource), and I don't 

    Seriously, if this file have to live a life of its own, name it like 
sched-r4k.c but not the way you named it -- this is not another clocksource 
module...

WBR, Sergei

Re: [PATCH v5] MIPS: Add a high resolution sched_clock() via cnt32_to_63().

[Wu Zhangjin]

On Thu, 2009-11-26 at 21:15 +0300, Sergei Shtylyov wrote:
> Hello, I wrote:
> 
> >> From: Wu Zhangjin <wuzhangjin@gmail.com>
> 
> >> (This v5 revision incorporates with the feedbacks from Ingo.)
> 
> >> 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>
> 
> >> diff --git a/arch/mips/kernel/csrc-r4k-hres.c 
> >> b/arch/mips/kernel/csrc-r4k-hres.c
> >> new file mode 100644
> >> index 0000000..2fe8be7
> >> --- /dev/null
> >> +++ b/arch/mips/kernel/csrc-r4k-hres.c
> > 
> > 
> >    I don't think this is really good name for this file (one might think 
> > that this is another implementation of clocksource instead of some 
> > sched_clock() code tied to this particular clocksource), and I don't 
> 
>     Seriously, if this file have to live a life of its own, name it like 
> sched-r4k.c but not the way you named it -- this is not another clocksource 
> module...
> 
> WBR, Sergei

okay, later.

Thanks,
	Wu Zhangjin

Re: [PATCH v5] MIPS: Add a high resolution sched_clock() via cnt32_to_63().

[Wu Zhangjin]

On Thu, 2009-11-26 at 21:15 +0300, Sergei Shtylyov wrote:
> Hello, I wrote:
> 
> >> From: Wu Zhangjin <wuzhangjin@gmail.com>
> 
> >> (This v5 revision incorporates with the feedbacks from Ingo.)
> 
> >> 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>
> 
> >> diff --git a/arch/mips/kernel/csrc-r4k-hres.c 
> >> b/arch/mips/kernel/csrc-r4k-hres.c
> >> new file mode 100644
> >> index 0000000..2fe8be7
> >> --- /dev/null
> >> +++ b/arch/mips/kernel/csrc-r4k-hres.c
> > 
> > 
> >    I don't think this is really good name for this file (one might think 
> > that this is another implementation of clocksource instead of some 
> > sched_clock() code tied to this particular clocksource), and I don't 
> 
>     Seriously, if this file have to live a life of its own, name it like 
> sched-r4k.c but not the way you named it -- this is not another clocksource 
> module...
> 

Hello, Sergei Shtylyov, I will use hres_sched_clock.c instead of
sched-r4k.c, is it okay?

Hi, Ralf, which one will you apply? If hres_sched_clock.c is okay, I
will resend it asap.
 
Thakns & Regards,
	Wu Zhangjin

Re: [PATCH v5] MIPS: Add a high resolution sched_clock() via cnt32_to_63().

[David Daney]

On 11/26/2009 07:05 PM, Wu Zhangjin wrote:
> On Thu, 2009-11-26 at 21:15 +0300, Sergei Shtylyov wrote:
[...]
>>>
>>>     I don't think this is really good name for this file (one might think
>>> that this is another implementation of clocksource instead of some
>>> sched_clock() code tied to this particular clocksource), and I don't
>>>        
>>      Seriously, if this file have to live a life of its own, name it like
>> sched-r4k.c but not the way you named it -- this is not another clocksource
>> module...
>>
>>      
> Hello, Sergei Shtylyov, I will use hres_sched_clock.c instead of
> sched-r4k.c, is it okay?
>
> Hi, Ralf, which one will you apply? If hres_sched_clock.c is okay, I
> will resend it asap.
>    

Like Sergei, I think putting this in a seperate file is a bad idea.  
This sched_clock is just a slightly different view of the csrc-r4k.c  
put it in there.

Octeon has its own clock source (cavium-octeon/scrc-octeon.c) and 
doesn't use csrc-r4k, so if you put it in a seperate file, there will 
have to be makefile hackery to keep this thing out of an octeon kernel.

When and if you add this to csrc-r4k, I will submit the patch (already 
written) that adds a high resolution sched_clock() to csrc-octeon.c.

Thanks,
David Daney

Re: [PATCH v5] MIPS: Add a high resolution sched_clock() via cnt32_to_63().

[Wu Zhangjin]

Hi, Ralf

This patch is not applicable, please ignore it!

Thanks!
	Wu Zhangjin

On Thu, 2009-11-26 at 20:45 -0800, David Daney wrote:
> On 11/26/2009 07:05 PM, Wu Zhangjin wrote:
> > On Thu, 2009-11-26 at 21:15 +0300, Sergei Shtylyov wrote:
> [...]
> >>>
> >>>     I don't think this is really good name for this file (one might think
> >>> that this is another implementation of clocksource instead of some
> >>> sched_clock() code tied to this particular clocksource), and I don't
> >>>        
> >>      Seriously, if this file have to live a life of its own, name it like
> >> sched-r4k.c but not the way you named it -- this is not another clocksource
> >> module...
> >>
> >>      
> > Hello, Sergei Shtylyov, I will use hres_sched_clock.c instead of
> > sched-r4k.c, is it okay?
> >
> > Hi, Ralf, which one will you apply? If hres_sched_clock.c is okay, I
> > will resend it asap.
> >    
> 
> Like Sergei, I think putting this in a seperate file is a bad idea.  
> This sched_clock is just a slightly different view of the csrc-r4k.c  
> put it in there.
> 
> Octeon has its own clock source (cavium-octeon/scrc-octeon.c) and 
> doesn't use csrc-r4k, so if you put it in a seperate file, there will 
> have to be makefile hackery to keep this thing out of an octeon kernel.
> 
> When and if you add this to csrc-r4k, I will submit the patch (already 
> written) that adds a high resolution sched_clock() to csrc-octeon.c.
> 
> Thanks,
> David Daney
> 
> 
> 
>