[PATCH v3 1/2] x86: Define early read-mostly per-cpu macros

[PATCH v3 1/2] x86: Define early read-mostly per-cpu macros

[Vlad Zolotarov]

Some read-mostly per-cpu data may need to be declared or defined early,
so it can be initialized and accessed before per_cpu areas are
allocated.

Only the data that resides in the per_cpu areas should be read-mostly,
as there is little benefit in optimizing cache lines on initialization.

Signed-off-by: Vlad Zolotarov <vlad@scalemp.com>
Signed-off-by: Ido Yariv <ido@wizery.com>
Signed-off-by: Shai Fultheim <shai@scalemp.com>
---
 arch/x86/include/asm/percpu.h |   17 +++++++++++++++++
 1 files changed, 17 insertions(+), 0 deletions(-)

diff --git a/arch/x86/include/asm/percpu.h b/arch/x86/include/asm/percpu.h
index 7a11910..8b5c8e0 100644
--- a/arch/x86/include/asm/percpu.h
+++ b/arch/x86/include/asm/percpu.h
@@ -555,6 +555,12 @@ DECLARE_PER_CPU(unsigned long, this_cpu_off);
 				{ [0 ... NR_CPUS-1] = _initvalue };	\
 	__typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map
 
+#define DEFINE_EARLY_PER_CPU_READ_MOSTLY(_type, _name, _initvalue)	\
+	DEFINE_PER_CPU_READ_MOSTLY(_type, _name) = _initvalue;		\
+	__typeof__(_type) _name##_early_map[NR_CPUS] __initdata =	\
+				{ [0 ... NR_CPUS-1] = _initvalue };	\
+	__typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map
+
 #define EXPORT_EARLY_PER_CPU_SYMBOL(_name)			\
 	EXPORT_PER_CPU_SYMBOL(_name)
 
@@ -563,6 +569,11 @@ DECLARE_PER_CPU(unsigned long, this_cpu_off);
 	extern __typeof__(_type) *_name##_early_ptr;		\
 	extern __typeof__(_type)  _name##_early_map[]
 
+#define DECLARE_EARLY_PER_CPU_READ_MOSTLY(_type, _name)		\
+	DECLARE_PER_CPU_READ_MOSTLY(_type, _name);		\
+	extern __typeof__(_type) *_name##_early_ptr;		\
+	extern __typeof__(_type)  _name##_early_map[]
+
 #define	early_per_cpu_ptr(_name) (_name##_early_ptr)
 #define	early_per_cpu_map(_name, _idx) (_name##_early_map[_idx])
 #define	early_per_cpu(_name, _cpu) 				\
@@ -574,12 +585,18 @@ DECLARE_PER_CPU(unsigned long, this_cpu_off);
 #define	DEFINE_EARLY_PER_CPU(_type, _name, _initvalue)		\
 	DEFINE_PER_CPU(_type, _name) = _initvalue
 
+#define DEFINE_EARLY_PER_CPU_READ_MOSTLY(_type, _name, _initvalue)	\
+	DEFINE_PER_CPU_READ_MOSTLY(_type, _name) = _initvalue
+
 #define EXPORT_EARLY_PER_CPU_SYMBOL(_name)			\
 	EXPORT_PER_CPU_SYMBOL(_name)
 
 #define DECLARE_EARLY_PER_CPU(_type, _name)			\
 	DECLARE_PER_CPU(_type, _name)
 
+#define DECLARE_EARLY_PER_CPU_READ_MOSTLY(_type, _name)		\
+	DECLARE_PER_CPU_READ_MOSTLY(_type, _name)
+
 #define	early_per_cpu(_name, _cpu) per_cpu(_name, _cpu)
 #define	early_per_cpu_ptr(_name) NULL
 /* no early_per_cpu_map() */
-- 
1.7.5.4

Re: [PATCH v3 1/2] x86: Define early read-mostly per-cpu macros

[Ingo Molnar]

* Vlad Zolotarov <vlad@scalemp.com> wrote:

> Some read-mostly per-cpu data may need to be declared or defined early,
> so it can be initialized and accessed before per_cpu areas are
> allocated.
> 
> Only the data that resides in the per_cpu areas should be read-mostly,
> as there is little benefit in optimizing cache lines on initialization.
> 
> Signed-off-by: Vlad Zolotarov <vlad@scalemp.com>
> Signed-off-by: Ido Yariv <ido@wizery.com>
> Signed-off-by: Shai Fultheim <shai@scalemp.com>

That's not a valid signoff chain. If you send me a patch then 
you should be the last SOB - but as the author you should be the 
first SOB as well. I edited the two SOBs to be Acked-by's.

Thanks,

	Ingo

Re: [PATCH v3 1/2] x86: Define early read-mostly per-cpu macros

[Vlad Zolotarov]

On Monday, May 21, 2012 10:46:26 AM Ingo Molnar wrote:
> * Vlad Zolotarov <vlad@scalemp.com> wrote:
> > Some read-mostly per-cpu data may need to be declared or defined early,
> > so it can be initialized and accessed before per_cpu areas are
> > allocated.
> > 
> > Only the data that resides in the per_cpu areas should be read-mostly,
> > as there is little benefit in optimizing cache lines on initialization.
> > 
> > Signed-off-by: Vlad Zolotarov <vlad@scalemp.com>
> > Signed-off-by: Ido Yariv <ido@wizery.com>
> > Signed-off-by: Shai Fultheim <shai@scalemp.com>
> 
> That's not a valid signoff chain. If you send me a patch then
> you should be the last SOB - but as the author you should be the
> first SOB as well. I edited the two SOBs to be Acked-by's.

Thanks, Ingo. I understand. However the "history" of this patch is not as 
simple as u represent. ;) 

It was originally written by Ido and I just posted it (then I put my SOB at 
the end of SOB list). But after u noted that it didn't compile on !SMP i 
reworked it but yet Ido did at least half of the coding, so I put my SOB at 
the first place (as I was the last one writing the code) and then Ido's as he 
actually wrote the code in this patch as well and I do wanted to give him a 
credit.

Do u still think changing his (Ido's) SOB to Acked-by was the right thing to 
do?

Thanks,
vlad

> 
> Thanks,
> 
> 	Ingo

Re: [PATCH v3 1/2] x86: Define early read-mostly per-cpu macros

[Ingo Molnar]

* Vlad Zolotarov <vlad@scalemp.com> wrote:

> On Monday, May 21, 2012 10:46:26 AM Ingo Molnar wrote:
> > * Vlad Zolotarov <vlad@scalemp.com> wrote:
> > > Some read-mostly per-cpu data may need to be declared or defined early,
> > > so it can be initialized and accessed before per_cpu areas are
> > > allocated.
> > > 
> > > Only the data that resides in the per_cpu areas should be read-mostly,
> > > as there is little benefit in optimizing cache lines on initialization.
> > > 
> > > Signed-off-by: Vlad Zolotarov <vlad@scalemp.com>
> > > Signed-off-by: Ido Yariv <ido@wizery.com>
> > > Signed-off-by: Shai Fultheim <shai@scalemp.com>
> > 
> > That's not a valid signoff chain. If you send me a patch then
> > you should be the last SOB - but as the author you should be the
> > first SOB as well. I edited the two SOBs to be Acked-by's.
> 
> Thanks, Ingo. I understand. However the "history" of this 
> patch is not as simple as u represent. ;)
> 
> It was originally written by Ido and I just posted it (then I 
> put my SOB at the end of SOB list). But after u noted that it 
> didn't compile on !SMP i reworked it but yet Ido did at least 
> half of the coding, so I put my SOB at the first place (as I 
> was the last one writing the code) and then Ido's as he 
> actually wrote the code in this patch as well and I do wanted 
> to give him a credit.

The proper way to do that is to add a "Originally-From: " tag, 
or to keep him as the author (and first SOB) and desccribe your 
own changes as comments between the SOBs:

 Signed-off-by: Original Author
 [ cleaned up the code a bit ]
 Signed-off-by: Second Author
 [ restructured it ]
 Signed-off-by: You

> Do u still think changing his (Ido's) SOB to Acked-by was the 
> right thing to do?

If they agree with it then you can send it to me that way - or 
the above way, or you can add any extra authors in a copyright 
notice - but the signoff chain should be a chain, with the 
primary author as the first SOB and the person sending me the 
patch being the last SOB.

Thanks,

	Ingo

Re: [PATCH v3 1/2] x86: Define early read-mostly per-cpu macros

[Vlad Zolotarov]

> > 
> > Thanks, Ingo. I understand. However the "history" of this
> > patch is not as simple as u represent. ;)
> > 
> > It was originally written by Ido and I just posted it (then I
> > put my SOB at the end of SOB list). But after u noted that it
> > didn't compile on !SMP i reworked it but yet Ido did at least
> > half of the coding, so I put my SOB at the first place (as I
> > was the last one writing the code) and then Ido's as he
> > actually wrote the code in this patch as well and I do wanted
> > to give him a credit.
> 
> The proper way to do that is to add a "Originally-From: " tag,
> or to keep him as the author (and first SOB) and desccribe your
> own changes as comments between the SOBs:
> 
>  Signed-off-by: Original Author
>  [ cleaned up the code a bit ]
>  Signed-off-by: Second Author
>  [ restructured it ]
>  Signed-off-by: You
> 
> > Do u still think changing his (Ido's) SOB to Acked-by was the
> > right thing to do?
> 
> If they agree with it then you can send it to me that way - or
> the above way, or you can add any extra authors in a copyright
> notice - but the signoff chain should be a chain, with the
> primary author as the first SOB and the person sending me the
> patch being the last SOB.

Got it. I think the "above" way would describe the reality in the most exact 
manner and would give the credit in the most exact manner.

I'd like to respin the series if it's still possible. 
Pls., let me know.

thanks,
vlad

> 
> Thanks,
> 
> 	Ingo