[PATCH] powerpc: Fix PowerPC 750CL CPU features

[PATCH] powerpc: Fix PowerPC 750CL CPU features

[Josh Boyer]

PowerPC 750CL has high BATs.  The patch below sets cpu_features to match that.
Without it, the original firmware mappings in the high BATs aren't cleared
which can cause "odd" address translations to occur.

Signed-off-by: Josh Boyer <jwboyer@linux.vnet.ibm.com>
---
 arch/powerpc/kernel/cputable.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

--- linux-2.6.orig/arch/powerpc/kernel/cputable.c
+++ linux-2.6/arch/powerpc/kernel/cputable.c
@@ -562,12 +562,12 @@ static struct cpu_spec cpu_specs[] = {
 		.pvr_mask		= 0xfffff0f0,
 		.pvr_value		= 0x00087010,
 		.cpu_name		= "750CL",
-		.cpu_features		= CPU_FTRS_750,
+		.cpu_features		= CPU_FTRS_750GX,
 		.cpu_user_features	= COMMON_USER | PPC_FEATURE_PPC_LE,
 		.icache_bsize		= 32,
 		.dcache_bsize		= 32,
 		.num_pmcs		= 4,
-		.cpu_setup		= __setup_cpu_750cx,
+		.cpu_setup		= __setup_cpu_750fx,
 		.platform		= "ppc750",
 	},
 	{	/* 745/755 */

Re: [PATCH] powerpc: Fix PowerPC 750CL CPU features

[Milton Miller]

On Thu Apr 12 22:49:06 EST 2007, Josh Boyer wrote:
> PowerPC 750CL has high BATs.  The patch below sets cpu_features to 
> match that.
> Without it, the original firmware mappings in the high BATs aren't 
> cleared
> which can cause "odd" address translations to occur.

Lets say what it does:  which continue to override the linux 
translations.

milton

Re: [PATCH] powerpc: Fix PowerPC 750CL CPU features

[Josh Boyer]

On Thu, 2007-04-12 at 08:49 -0500, Milton Miller wrote:
> On Thu Apr 12 22:49:06 EST 2007, Josh Boyer wrote:
> > PowerPC 750CL has high BATs.  The patch below sets cpu_features to 
> > match that.
> > Without it, the original firmware mappings in the high BATs aren't 
> > cleared
> > which can cause "odd" address translations to occur.
> 
> Lets say what it does:  which continue to override the linux 
> translations.

Sure, that's exactly what it does since without the HIGH_BATS feature
Linux doesn't know about them and therefore can't clear them.

josh

Re: [PATCH] powerpc: Fix PowerPC 750CL CPU features

[Segher Boessenkool]

> PowerPC 750CL has high BATs.  The patch below sets cpu_features to 
> match that.
> Without it, the original firmware mappings in the high BATs aren't 
> cleared
> which can cause "odd" address translations to occur.

> -		.cpu_features		= CPU_FTRS_750,
> +		.cpu_features		= CPU_FTRS_750GX,

The difference between these two is

#define CPU_FTRS_750GX  (CPU_FTR_DUAL_PLL_750FX | CPU_FTR_HAS_HIGH_BATS)
#define CPU_FTRS_750    (CPU_FTR_COMMON)

750GX excludes COMMON, which seems like a bug.  With
your change, 750CL uses DUAL_PLL_750FX; does it actually
have that feature?

Either way, it seems a good idea to create a CPU_FTRS_750CL,
the CL is an evolution of the CX, not the FX/GX (in name
at least, dunno about the gory details).

> -		.cpu_setup		= __setup_cpu_750cx,
> +		.cpu_setup		= __setup_cpu_750fx,

Same thing here.  Are you sure 750CL doesn't need
the NAP workaround?  [Is anyone sure the 750CX
actually needs it -- the comment in cpu_setup_6xx.S
doesn't instill much confidence.]


Segher

Re: [PATCH] powerpc: Fix PowerPC 750CL CPU features

[Kumar Gala]

On Apr 12, 2007, at 11:41 AM, Segher Boessenkool wrote:

>> PowerPC 750CL has high BATs.  The patch below sets cpu_features to
>> match that.
>> Without it, the original firmware mappings in the high BATs aren't
>> cleared
>> which can cause "odd" address translations to occur.
>
>> -		.cpu_features		= CPU_FTRS_750,
>> +		.cpu_features		= CPU_FTRS_750GX,
>
> The difference between these two is
>
> #define CPU_FTRS_750GX  (CPU_FTR_DUAL_PLL_750FX |  
> CPU_FTR_HAS_HIGH_BATS)
> #define CPU_FTRS_750    (CPU_FTR_COMMON)
>
> 750GX excludes COMMON, which seems like a bug.  With
> your change, 750CL uses DUAL_PLL_750FX; does it actually
> have that feature?

I agree the 750GX should include COMMON, we probably don't notice  
since its never in a SMP system.

- k

Re: [PATCH] powerpc: Fix PowerPC 750CL CPU features

[Josh Boyer]

On Thu, 2007-04-12 at 12:06 -0500, Kumar Gala wrote:
> On Apr 12, 2007, at 11:41 AM, Segher Boessenkool wrote:
> 
> >> PowerPC 750CL has high BATs.  The patch below sets cpu_features to
> >> match that.
> >> Without it, the original firmware mappings in the high BATs aren't
> >> cleared
> >> which can cause "odd" address translations to occur.
> >
> >> -		.cpu_features		= CPU_FTRS_750,
> >> +		.cpu_features		= CPU_FTRS_750GX,
> >
> > The difference between these two is
> >
> > #define CPU_FTRS_750GX  (CPU_FTR_DUAL_PLL_750FX |  
> > CPU_FTR_HAS_HIGH_BATS)
> > #define CPU_FTRS_750    (CPU_FTR_COMMON)
> >
> > 750GX excludes COMMON, which seems like a bug.  With
> > your change, 750CL uses DUAL_PLL_750FX; does it actually
> > have that feature?
> 
> I agree the 750GX should include COMMON, we probably don't notice  
> since its never in a SMP system.

I'll send a separate patch to fix that.

josh

Re: [PATCH] powerpc: Fix PowerPC 750CL CPU features

[Josh Boyer]

On Thu, 2007-04-12 at 18:41 +0200, Segher Boessenkool wrote:
> > PowerPC 750CL has high BATs.  The patch below sets cpu_features to 
> > match that.
> > Without it, the original firmware mappings in the high BATs aren't 
> > cleared
> > which can cause "odd" address translations to occur.
> 
> > -		.cpu_features		= CPU_FTRS_750,
> > +		.cpu_features		= CPU_FTRS_750GX,
> 
> The difference between these two is
> 
> #define CPU_FTRS_750GX  (CPU_FTR_DUAL_PLL_750FX | CPU_FTR_HAS_HIGH_BATS)
> #define CPU_FTRS_750    (CPU_FTR_COMMON)
> 
> 750GX excludes COMMON, which seems like a bug.  With

CPU_FTR_COMMON is actually a nop in this case anyway.  It should
probably be listed for "completeness", but it's defined to be 0 so it
doesn't really do anything.

> your change, 750CL uses DUAL_PLL_750FX; does it actually
> have that feature?

No, it doesn't, I missed that in the GX feature list.  Good catch.

> 
> Either way, it seems a good idea to create a CPU_FTRS_750CL,
> the CL is an evolution of the CX, not the FX/GX (in name
> at least, dunno about the gory details).

I can do that, or I can just do:

CPU_FTRS_750 | CPU_FTR_HAS_HIGH_BATS

> 
> > -		.cpu_setup		= __setup_cpu_750cx,
> > +		.cpu_setup		= __setup_cpu_750fx,
> 
> Same thing here.  Are you sure 750CL doesn't need
> the NAP workaround?  [Is anyone sure the 750CX
> actually needs it -- the comment in cpu_setup_6xx.S
> doesn't instill much confidence.]

This should be fine.

I'll send out a newer version, and fixup the description with the
comment Milton made.

josh

Re: [PATCH] powerpc: Fix PowerPC 750CL CPU features

[Segher Boessenkool]

>> The difference between these two is
>>
>> #define CPU_FTRS_750GX  (CPU_FTR_DUAL_PLL_750FX | 
>> CPU_FTR_HAS_HIGH_BATS)
>> #define CPU_FTRS_750    (CPU_FTR_COMMON)
>>
>> 750GX excludes COMMON, which seems like a bug.  With
>
> CPU_FTR_COMMON is actually a nop in this case anyway.  It should
> probably be listed for "completeness", but it's defined to be 0 so it
> doesn't really do anything.

Well you *can* have SMP 750 systems.  Not that Linux can
handle them but...  ;-)

>> Either way, it seems a good idea to create a CPU_FTRS_750CL,
>> the CL is an evolution of the CX, not the FX/GX (in name
>> at least, dunno about the gory details).
>
> I can do that, or I can just do:
>
> CPU_FTRS_750 | CPU_FTR_HAS_HIGH_BATS

Yeah, you can #define CPU_FTRS_750CL that way too.
Most current defines should be like that if you ask me,
it's a whole lot of copy'n'waste right now.

>>> -		.cpu_setup		= __setup_cpu_750cx,
>>> +		.cpu_setup		= __setup_cpu_750fx,
>>
>> Same thing here.  Are you sure 750CL doesn't need
>> the NAP workaround?  [Is anyone sure the 750CX
>> actually needs it -- the comment in cpu_setup_6xx.S
>> doesn't instill much confidence.]
>
> This should be fine.

Okay.  The name isn't fine though.


Segher

Re: [PATCH] powerpc: Fix PowerPC 750CL CPU features

[Benjamin Herrenschmidt]

> The difference between these two is
> 
> #define CPU_FTRS_750GX  (CPU_FTR_DUAL_PLL_750FX | CPU_FTR_HAS_HIGH_BATS)
> #define CPU_FTRS_750    (CPU_FTR_COMMON)
> 
> 750GX excludes COMMON, which seems like a bug.  With
> your change, 750CL uses DUAL_PLL_750FX; does it actually
> have that feature?

Looks indeed like that lack of COMMON in the GX entry is a bug...

Josh, care to fix that too while at it ? :-)

> Either way, it seems a good idea to create a CPU_FTRS_750CL,
> the CL is an evolution of the CX, not the FX/GX (in name
> at least, dunno about the gory details).

Agreed.

> > -		.cpu_setup		= __setup_cpu_750cx,
> > +		.cpu_setup		= __setup_cpu_750fx,
> 
> Same thing here.  Are you sure 750CL doesn't need
> the NAP workaround?  [Is anyone sure the 750CX
> actually needs it -- the comment in cpu_setup_6xx.S
> doesn't instill much confidence.]

Agreed too.

Ben.

Re: [PATCH] powerpc: Fix PowerPC 750CL CPU features

[Josh Boyer]

On Fri, Apr 13, 2007 at 12:28:30PM +1000, Benjamin Herrenschmidt wrote:
> > The difference between these two is
> > 
> > #define CPU_FTRS_750GX  (CPU_FTR_DUAL_PLL_750FX | CPU_FTR_HAS_HIGH_BATS)
> > #define CPU_FTRS_750    (CPU_FTR_COMMON)
> > 
> > 750GX excludes COMMON, which seems like a bug.  With
> > your change, 750CL uses DUAL_PLL_750FX; does it actually
> > have that feature?
> 
> Looks indeed like that lack of COMMON in the GX entry is a bug...
> 
> Josh, care to fix that too while at it ? :-)

I did.  See patch version 2. :)

josh