[PATCH] x86: a P4 is a P6 not an i486

[PATCH] x86: a P4 is a P6 not an i486

[Hugh Dickins]

P4 has been coming out as CPU_FAMILY=4 instead of 6: fix MPENTIUM4 typo.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
---

 arch/x86/Kconfig.cpu |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

--- 2.6.25-rc3-git/arch/x86/Kconfig.cpu	2008-02-27 07:57:20.000000000 +0000
+++ linux/arch/x86/Kconfig.cpu	2008-03-04 13:50:39.000000000 +0000
@@ -388,7 +388,7 @@ config X86_OOSTORE
 #
 config X86_P6_NOP
 	def_bool y
-	depends on (X86_64 || !X86_GENERIC) && (M686 || MPENTIUMII || MPENTIUMIII || MPENTIUMM || MCORE2 || PENTIUM4)
+	depends on (X86_64 || !X86_GENERIC) && (M686 || MPENTIUMII || MPENTIUMIII || MPENTIUMM || MCORE2 || MPENTIUM4)
 
 config X86_TSC
 	def_bool y

Re: [PATCH] x86: a P4 is a P6 not an i486

[Ingo Molnar]

* Hugh Dickins <hugh@veritas.com> wrote:

> P4 has been coming out as CPU_FAMILY=4 instead of 6: fix MPENTIUM4 
> typo.

thanks Hugh, applied.

	Ingo

Re: [PATCH] x86: a P4 is a P6 not an i486

[H. Peter Anvin]

Ingo Molnar wrote:
> * Hugh Dickins <hugh@veritas.com> wrote:
> 
>> P4 has been coming out as CPU_FAMILY=4 instead of 6: fix MPENTIUM4 
>> typo.
> 
> thanks Hugh, applied.
> 
> 	Ingo

The patch is correct, obviously; the comment is not: the real comment 
should be:

Enable the use of P6 NOPs on Pentium 4

The CPU_FAMILY field is really the minimum CPU family that can boot the 
kernel (that's why there is no 5; we don't have any i586 family 
dependencies not encoded by CPUID bits.)

	-hpa

Re: [PATCH] x86: a P4 is a P6 not an i486

[Ingo Molnar]

* H. Peter Anvin <hpa@zytor.com> wrote:

>>> P4 has been coming out as CPU_FAMILY=4 instead of 6: fix MPENTIUM4 
>>> typo.
>>
>> thanks Hugh, applied.
>
> The patch is correct, obviously; the comment is not: the real comment 
> should be:
>
> Enable the use of P6 NOPs on Pentium 4
>
> The CPU_FAMILY field is really the minimum CPU family that can boot 
> the kernel (that's why there is no 5; we don't have any i586 family 
> dependencies not encoded by CPUID bits.)

yeah, i've updated the commit. (find it below)

	Ingo

----------->
Subject: x86: enable the use of P6 NOPs on Pentium 4
From: Hugh Dickins <hugh@veritas.com>
Date: Tue, 4 Mar 2008 19:33:24 +0000 (GMT)

fix MPENTIUM4 typo - it resulted in P4s using less optimal NOPs.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 arch/x86/Kconfig.cpu |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Index: linux-x86.q/arch/x86/Kconfig.cpu
===================================================================
--- linux-x86.q.orig/arch/x86/Kconfig.cpu
+++ linux-x86.q/arch/x86/Kconfig.cpu
@@ -388,7 +388,7 @@ config X86_OOSTORE
 #
 config X86_P6_NOP
 	def_bool y
-	depends on (X86_64 || !X86_GENERIC) && (M686 || MPENTIUMII || MPENTIUMIII || MPENTIUMM || MCORE2 || PENTIUM4)
+	depends on (X86_64 || !X86_GENERIC) && (M686 || MPENTIUMII || MPENTIUMIII || MPENTIUMM || MCORE2 || MPENTIUM4)
 
 config X86_TSC
 	def_bool y

P6 NOPs again: MPSC?

[Hugh Dickins]

I've now noticed that the machine I call MPENTIUM4 for 32-bit kernels
is called MPSC for 64-bit kernels, and in that case it still doesn't
get the P6 NOPs which I'm guessing it ought to (yeah, I'm so excited
about the speedup I can expect from P6 NOPs, can we perhaps have a
little silvery "Using Genuine P6 NOPs" sticker to put on the box? ;)

I'm suspecting that the patch below is actually wrong, and that it's
really the "(X86_64 || !X86_GENERIC) &&" which should be changed;
but very unsure of my ground and what's right for CPU_GENERIC -
the 32/64 heritage of x86/Kconfig.cpu rather confuses me.

Over to you to do the right thing, I'm hoping, please, Peter!

Thanks,
Hugh

--- 2.6.25-rc5/arch/x86/Kconfig.cpu	2008-03-05 11:27:23.000000000 +0000
+++ linux/arch/x86/Kconfig.cpu	2008-03-10 18:48:10.000000000 +0000
@@ -388,7 +388,7 @@ config X86_OOSTORE
 #
 config X86_P6_NOP
 	def_bool y
-	depends on (X86_64 || !X86_GENERIC) && (M686 || MPENTIUMII || MPENTIUMIII || MPENTIUMM || MCORE2 || MPENTIUM4)
+	depends on (X86_64 || !X86_GENERIC) && (M686 || MPENTIUMII || MPENTIUMIII || MPENTIUMM || MCORE2 || MPENTIUM4 || MPSC)
 
 config X86_TSC
 	def_bool y

Re: P6 NOPs again: MPSC?

[H. Peter Anvin]

Hugh Dickins wrote:
> I've now noticed that the machine I call MPENTIUM4 for 32-bit kernels
> is called MPSC for 64-bit kernels, and in that case it still doesn't
> get the P6 NOPs which I'm guessing it ought to (yeah, I'm so excited
> about the speedup I can expect from P6 NOPs, can we perhaps have a
> little silvery "Using Genuine P6 NOPs" sticker to put on the box? ;)
> 
> I'm suspecting that the patch below is actually wrong, and that it's
> really the "(X86_64 || !X86_GENERIC) &&" which should be changed;
> but very unsure of my ground and what's right for CPU_GENERIC -
> the 32/64 heritage of x86/Kconfig.cpu rather confuses me.
> 
> Over to you to do the right thing, I'm hoping, please, Peter!
> 

X86_64 && !X86_GENERIC comes from not wanting to be a compatibility 
issue when compiling for generic CPUs.  There are some 32-bit otherwise 
i686-compatible chips (from VIA and Transmeta) which don't have these NOPs.

	-hpa

Re: P6 NOPs again: MPSC?

[Hugh Dickins]

On Mon, 10 Mar 2008, H. Peter Anvin wrote:
> Hugh Dickins wrote:
> > 
> > I'm suspecting that the patch below is actually wrong, and that it's
> > really the "(X86_64 || !X86_GENERIC) &&" which should be changed;
> > but very unsure of my ground and what's right for CPU_GENERIC -
> > the 32/64 heritage of x86/Kconfig.cpu rather confuses me.
> 
> X86_64 && !X86_GENERIC comes from not wanting to be a compatibility issue when
> compiling for generic CPUs.  There are some 32-bit otherwise i686-compatible
> chips (from VIA and Transmeta) which don't have these NOPs.

Yup, I realized that (with || in place of &&): I meant that perhaps
it was supposed to say X86_64 || (!X86_GENERIC && (M686 || ....))
since quite a lot of other options there have "|| X86_64" to cover
all the 64-bit possibilities at once i.e. is it only MPSC that
needs to be added, or both MPSC and CPU_GENERIC?

(But I've lost my way around here.  A good example of the 32/64
confusion is how one has to remember that X86_GENERIC is peculiar
to 32-bit, and CPU_GENERIC peculiar to 64-bit.  And CPU_GENERIC is
understandably an alternative to the specific 64-bit models, but
X86_GENERIC is strangely in addition to the specific 32-bit models.)

Hugh

Re: P6 NOPs again: MPSC?

[H. Peter Anvin]

Hugh Dickins wrote:
> 
> Yup, I realized that (with || in place of &&): I meant that perhaps
> it was supposed to say X86_64 || (!X86_GENERIC && (M686 || ....))
> since quite a lot of other options there have "|| X86_64" to cover
> all the 64-bit possibilities at once i.e. is it only MPSC that
> needs to be added, or both MPSC and CPU_GENERIC?
> 

No.  The left-hand clause (X86_64 || !X86_GENERIC) indicates when using 
these NOPs are *permitted*; the right-hand clause indicates when doing 
so is *desirable*.  The latter isn't, for example, for K8, even on 64 bits.

	-hpa

Re: P6 NOPs again: MPSC?

[Ingo Molnar]

* Hugh Dickins <hugh@veritas.com> wrote:

> (But I've lost my way around here.  A good example of the 32/64 
> confusion is how one has to remember that X86_GENERIC is peculiar to 
> 32-bit, and CPU_GENERIC peculiar to 64-bit.  And CPU_GENERIC is 
> understandably an alternative to the specific 64-bit models, but 
> X86_GENERIC is strangely in addition to the specific 32-bit models.)

yes - this is one of the arbitrary deviations between 32-bit and 64-bit. 
Patches to bring the concepts and names in sync are welcome :-)

	Ingo

Re: P6 NOPs again: MPSC?

[Hugh Dickins]

On Tue, 11 Mar 2008, Ingo Molnar wrote:
> * Hugh Dickins <hugh@veritas.com> wrote:
> 
> > (But I've lost my way around here.  A good example of the 32/64 
> > confusion is how one has to remember that X86_GENERIC is peculiar to 
> > 32-bit, and CPU_GENERIC peculiar to 64-bit.  And CPU_GENERIC is 
> > understandably an alternative to the specific 64-bit models, but 
> > X86_GENERIC is strangely in addition to the specific 32-bit models.)
> 
> yes - this is one of the arbitrary deviations between 32-bit and 64-bit. 
> Patches to bring the concepts and names in sync are welcome :-)

:-)  I durst not!

Hugh

Re: P6 NOPs again: MPSC?

[Hugh Dickins]

On Tue, 11 Mar 2008, H. Peter Anvin wrote:
> Hugh Dickins wrote:
> > 
> > Yup, I realized that (with || in place of &&): I meant that perhaps
> > it was supposed to say X86_64 || (!X86_GENERIC && (M686 || ....))
> > since quite a lot of other options there have "|| X86_64" to cover
> > all the 64-bit possibilities at once i.e. is it only MPSC that
> > needs to be added, or both MPSC and CPU_GENERIC?
> 
> No.  The left-hand clause (X86_64 || !X86_GENERIC) indicates when using these
> NOPs are *permitted*; the right-hand clause indicates when doing so is
> *desirable*.  The latter isn't, for example, for K8, even on 64 bits.

Okay, so if we're to exclude CPU_GENERIC and MK8, I think that
makes my original patch the right one: here it is with a signoff.


I've now noticed that the machine I call MPENTIUM4 for 32-bit kernels
is called MPSC for 64-bit kernels, and in that case it still doesn't
get the P6 NOPs it ought to.  hpa explains that MK8 should still be
excluded, so it's just a matter of including MPSC along with MPENTIUM4.

Signed-off-by: Hugh Dickins <hugh@veritas.com>

--- 2.6.25-rc5/arch/x86/Kconfig.cpu	2008-03-05 11:27:23.000000000 +0000
+++ linux/arch/x86/Kconfig.cpu	2008-03-10 18:48:10.000000000 +0000
@@ -388,7 +388,7 @@ config X86_OOSTORE
 #
 config X86_P6_NOP
 	def_bool y
-	depends on (X86_64 || !X86_GENERIC) && (M686 || MPENTIUMII || MPENTIUMIII || MPENTIUMM || MCORE2 || MPENTIUM4)
+	depends on (X86_64 || !X86_GENERIC) && (M686 || MPENTIUMII || MPENTIUMIII || MPENTIUMM || MCORE2 || MPENTIUM4 || MPSC)
 
 config X86_TSC
 	def_bool y