Skip to contentSkip to conversation

CAS DFP support

PATCH In review Jake Moilanen 6 participants 12 messages Nov 22, 2006 – Nov 23, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 CAS DFP support

When booting a Power6 in compat mode, the DFP bit must be set in the client architecture call, or else firmware will downgrade the compliance to P5 compat mode, thus removing VMX and DFP instructions.

--- 2.6/arch/powerpc/kernel/prom_init.c~ 2006-11-13 15:01:13.000000000
-0600
+++ 2.6/arch/powerpc/kernel/prom_init.c 2006-11-21 18:38:27.000000000
-0600
@@ -627,6 +627,7 @@ static void __init early_cmdline_parse(v
/* Option vector 3: processor options supported */
#define OV3_FP 0x80 /* floating point */
#define OV3_VMX 0x40 /* VMX/Altivec */
+#define OV3_DFP 0x20 /* decimal floating point */

/* Option vector 5: PAPR/OF options supported */
#define OV5_LPAR 0x80 /* logical partitioning supported */
@@ -674,7 +675,7 @@ static unsigned char ibm_architecture_ve
/* option vector 3: processor options supported */
3 - 2, /* length */
0, /* don't ignore, don't halt */
- OV3_FP | OV3_VMX,
+ OV3_FP | OV3_VMX | OV3_DFP,

/* option vector 4: IBM PAPR implementation */
2 - 2, /* length */

#1 lore raw
Quoted 5 lines · Jake Moilanen

When booting a Power6 in compat mode, the DFP bit must be set in the client architecture call, or else firmware will downgrade the compliance to P5 compat mode, thus removing VMX and DFP instructions.

For DFP do we also need a cpu_user_features, that propagates out to user space?

- k

Quoted 28 lines

--- 2.6/arch/powerpc/kernel/prom_init.c~ 2006-11-13 15:01:13.000000000
-0600
+++ 2.6/arch/powerpc/kernel/prom_init.c 2006-11-21 18:38:27.000000000
-0600

(quoted diff) +2 -1
@@ -627,6 +627,7 @@ static void __init early_cmdline_parse(v
 /* Option vector 3: processor options supported */
 #define OV3_FP			0x80	/* floating point */
 #define OV3_VMX			0x40	/* VMX/Altivec */
+#define OV3_DFP			0x20	/* decimal floating point */
 
 /* Option vector 5: PAPR/OF options supported */
 #define OV5_LPAR		0x80	/* logical partitioning supported */
@@ -674,7 +675,7 @@ static unsigned char ibm_architecture_ve
 	/* option vector 3: processor options supported */
 	3 - 2,				/* length */
 	0,				/* don't ignore, don't halt */
-	OV3_FP | OV3_VMX,
+	OV3_FP | OV3_VMX | OV3_DFP,
 
 	/* option vector 4: IBM PAPR implementation */
 	2 - 2,				/* length */

_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://ozlabs.org/mailman/listinfo/linuxppc-dev

#2 lore raw
In reply to Jake Moilanen
Quoted 11 lines · Jake Moilanen

When booting a Power6 in compat mode, the DFP bit must be set in the client architecture call, or else firmware will downgrade the compliance to P5 compat mode, thus removing VMX and DFP instructions.

--- 2.6/arch/powerpc/kernel/prom_init.c~ 2006-11-13 15:01:13.000000000
-0600
+++ 2.6/arch/powerpc/kernel/prom_init.c 2006-11-21 18:38:27.000000000
-0600

Review static void __init early_cmdline_parse(v · line 630

Quoted 4 lines
(quoted diff) +1 -0
@@ -627,6 +627,7 @@ static void __init early_cmdline_parse(v
 /* Option vector 3: processor options supported */
 #define OV3_FP			0x80	/* floating point */
 #define OV3_VMX			0x40	/* VMX/Altivec */
+#define OV3_DFP			0x20	/* decimal floating point */

Can we make the flag OV3_DECFP to distinguish it from Double FP.

Quoted 18 lines
(quoted diff) +1 -1
 /* Option vector 5: PAPR/OF options supported */
 #define OV5_LPAR		0x80	/* logical partitioning supported */
@@ -674,7 +675,7 @@ static unsigned char ibm_architecture_ve
 	/* option vector 3: processor options supported */
 	3 - 2,				/* length */
 	0,				/* don't ignore, don't halt */
-	OV3_FP | OV3_VMX,
+	OV3_FP | OV3_VMX | OV3_DFP,
 
 	/* option vector 4: IBM PAPR implementation */
 	2 - 2,				/* length */

_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://ozlabs.org/mailman/listinfo/linuxppc-dev

#3 lore raw
Quoted 9 lines · Kumar Gala
Quoted 7 lines · Jake Moilanen

When booting a Power6 in compat mode, the DFP bit must be set in the client architecture call, or else firmware will downgrade the compliance
to P5 compat mode, thus removing VMX and DFP instructions.

Quoted 11 lines

Review static void __init early_cmdline_parse(v · line 630

Quoted 9 lines

--- 2.6/arch/powerpc/kernel/prom_init.c~ 2006-11-13 15:01:13.000000000
-0600
+++ 2.6/arch/powerpc/kernel/prom_init.c 2006-11-21 18:38:27.000000000
-0600

(quoted diff) +1 -0
@@ -627,6 +627,7 @@ static void __init early_cmdline_parse(v
 /* Option vector 3: processor options supported */
 #define OV3_FP			0x80	/* floating point */
 #define OV3_VMX			0x40	/* VMX/Altivec */
+#define OV3_DFP			0x20	/* decimal floating point */

Can we make the flag OV3_DECFP to distinguish it from Double FP.

Yup...

--- 2.6/arch/powerpc/kernel/prom_init.c~ 2006-11-13 15:01:13.000000000
-0600
+++ 2.6/arch/powerpc/kernel/prom_init.c 2006-11-21 18:38:27.000000000
-0600
@@ -627,6 +627,7 @@ static void __init early_cmdline_parse(v
/* Option vector 3: processor options supported */
#define OV3_FP 0x80 /* floating point */
#define OV3_VMX 0x40 /* VMX/Altivec */
+#define OV3_DECFP 0x20 /* decimal floating point */

/* Option vector 5: PAPR/OF options supported */
#define OV5_LPAR 0x80 /* logical partitioning supported */
@@ -674,7 +675,7 @@ static unsigned char ibm_architecture_ve
/* option vector 3: processor options supported */
3 - 2, /* length */
0, /* don't ignore, don't halt */
- OV3_FP | OV3_VMX,
+ OV3_FP | OV3_VMX | OV3_DECFP,

/* option vector 4: IBM PAPR implementation */
2 - 2, /* length */

#4 lore raw
In reply to Kumar Gala
Quoted 9 lines · Kumar Gala
Quoted 7 lines · Jake Moilanen

When booting a Power6 in compat mode, the DFP bit must be set in the client architecture call, or else firmware will downgrade the compliance
to P5 compat mode, thus removing VMX and DFP instructions.

Quoted 11 lines

Review static void __init early_cmdline_parse(v · line 630

Quoted 9 lines

--- 2.6/arch/powerpc/kernel/prom_init.c~ 2006-11-13 15:01:13.000000000
-0600
+++ 2.6/arch/powerpc/kernel/prom_init.c 2006-11-21 18:38:27.000000000
-0600

(quoted diff) +1 -0
@@ -627,6 +627,7 @@ static void __init early_cmdline_parse(v
 /* Option vector 3: processor options supported */
 #define OV3_FP			0x80	/* floating point */
 #define OV3_VMX			0x40	/* VMX/Altivec */
+#define OV3_DFP			0x20	/* decimal floating point */

Can we make the flag OV3_DECFP to distinguish it from Double FP.

One thing OV3_DFP has going for it is that is matches the nomenclature used by GCC, binutils. I'm not sure what you mean by Double FP, unless you're referring to the double wide FP unit on the Blue Gene/L 440 processor, but they don't refer to it as Double FP.

Peter

#5 lore raw
Quoted 24 lines · Peter Bergner
Quoted 9 lines · Kumar Gala
Quoted 7 lines · Jake Moilanen

When booting a Power6 in compat mode, the DFP bit must be set in the client architecture call, or else firmware will downgrade the compliance
to P5 compat mode, thus removing VMX and DFP instructions.

Quoted 13 lines

Review static void __init early_cmdline_parse(v · line 630

Quoted 11 lines

--- 2.6/arch/powerpc/kernel/prom_init.c~ 2006-11-13
15:01:13.000000000
-0600
+++ 2.6/arch/powerpc/kernel/prom_init.c 2006-11-21
18:38:27.000000000
-0600

(quoted diff) +1 -0
@@ -627,6 +627,7 @@ static void __init early_cmdline_parse(v
 /* Option vector 3: processor options supported */
 #define OV3_FP			0x80	/* floating point */
 #define OV3_VMX			0x40	/* VMX/Altivec */
+#define OV3_DFP			0x20	/* decimal floating point */

Can we make the flag OV3_DECFP to distinguish it from Double FP.

Quoted 4 lines

One thing OV3_DFP has going for it is that is matches the nomenclature used by GCC, binutils. I'm not sure what you mean by Double FP, unless you're referring to the double wide FP unit on the Blue Gene/L 440 processor, but they don't refer to it as Double FP.

I'm referring to plain old double precision floating point. The e500 core from Freescale has version that provide embedded single precision floating point (EFP_SINGLE) and versions that provide both embedded double precision floating point (EFP_DOUBLE). I just want use to be clear the DFP could mean either double or decimal as far as I'm concerned.

Also, I think the 602 (may have never seen the light of day) only implemented single precision floating point.

- k

#6 lore raw
Quoted 32 lines · Kumar Gala
Quoted 24 lines · Peter Bergner
Quoted 9 lines · Kumar Gala
Quoted 7 lines · Jake Moilanen

When booting a Power6 in compat mode, the DFP bit must be set in the client architecture call, or else firmware will downgrade the compliance
to P5 compat mode, thus removing VMX and DFP instructions.

Quoted 13 lines

Review static void __init early_cmdline_parse(v · line 630

Quoted 11 lines

--- 2.6/arch/powerpc/kernel/prom_init.c~ 2006-11-13
15:01:13.000000000
-0600
+++ 2.6/arch/powerpc/kernel/prom_init.c 2006-11-21
18:38:27.000000000
-0600

(quoted diff) +1 -0
@@ -627,6 +627,7 @@ static void __init early_cmdline_parse(v
 /* Option vector 3: processor options supported */
 #define OV3_FP			0x80	/* floating point */
 #define OV3_VMX			0x40	/* VMX/Altivec */
+#define OV3_DFP			0x20	/* decimal floating point */

Can we make the flag OV3_DECFP to distinguish it from Double FP.

Quoted 5 lines

One thing OV3_DFP has going for it is that is matches the nomenclature used by GCC, binutils. I'm not sure what you mean by Double FP, unless
you're referring to the double wide FP unit on the Blue Gene/L 440 processor, but they don't refer to it as Double FP.

Quoted 6 lines

I'm referring to plain old double precision floating point. The e500 core from Freescale has version that provide embedded single precision floating point (EFP_SINGLE) and versions that provide both embedded double precision floating point (EFP_DOUBLE). I just want use to be clear the DFP could mean either double or decimal as far as I'm concerned.

the comment on the #define "/* decimal floating point */" should be sufficient for that clarification tho, right?

Quoted 9 lines

Also, I think the 602 (may have never seen the light of day) only implemented single precision floating point.

- k
_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://ozlabs.org/mailman/listinfo/linuxppc-dev

#7 lore raw
Quoted 40 lines · Will Schmidt
Quoted 32 lines · Kumar Gala
Quoted 24 lines · Peter Bergner
Quoted 9 lines · Kumar Gala
Quoted 7 lines · Jake Moilanen

When booting a Power6 in compat mode, the DFP bit must be set in the client architecture call, or else firmware will downgrade the compliance
to P5 compat mode, thus removing VMX and DFP instructions.

Quoted 13 lines

Review static void __init early_cmdline_parse(v · line 630

Quoted 11 lines

--- 2.6/arch/powerpc/kernel/prom_init.c~ 2006-11-13
15:01:13.000000000
-0600
+++ 2.6/arch/powerpc/kernel/prom_init.c 2006-11-21
18:38:27.000000000
-0600

(quoted diff) +1 -0
@@ -627,6 +627,7 @@ static void __init early_cmdline_parse(v
 /* Option vector 3: processor options supported */
 #define OV3_FP			0x80	/* floating point */
 #define OV3_VMX			0x40	/* VMX/Altivec */
+#define OV3_DFP			0x20	/* decimal floating point */

Can we make the flag OV3_DECFP to distinguish it from Double FP.

Quoted 5 lines

One thing OV3_DFP has going for it is that is matches the nomenclature used by GCC, binutils. I'm not sure what you mean by Double FP, unless
you're referring to the double wide FP unit on the Blue Gene/L 440 processor, but they don't refer to it as Double FP.

Quoted 6 lines

I'm referring to plain old double precision floating point. The e500 core from Freescale has version that provide embedded single precision floating point (EFP_SINGLE) and versions that provide both embedded double precision floating point (EFP_DOUBLE). I just want use to be clear the DFP could mean either double or decimal as far as I'm concerned.

Quoted 2 lines

the comment on the #define "/* decimal floating point */" should be sufficient for that clarification tho, right?

For this one case it is, but I'm sure we are going to end up with some other defines for related to Decimal FP and I'd like them to be clear.

Quoted 8 lines
Quoted 8 lines

Also, I think the 602 (may have never seen the light of day) only implemented single precision floating point.

- k
_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://ozlabs.org/mailman/listinfo/linuxppc-dev

#8 lore raw
In reply to Kumar Gala
Quoted 1 line

Also, I think the 602 (may have never seen the light of day)

I never saw any 602 in the wild, and no one I ever asked has.

Quoted 1 line

only implemented single precision floating point.

Actually, it implements double precision just fine (although all double precision arithmetic operations trap and have to be emulated). Loads and stores work, the FPRs are 64 bit.

Anyway, off-topic, and this CPU doesn't actually exist anyway.

Segher

#9 lore raw
In reply to Kumar Gala
Quoted 1 line · Kumar Gala

Can we make the flag OV3_DECFP to distinguish it from Double FP.

I don't think it's worth it; the OV3_DFP name is defined and used only in prom_init.c, and only in the context of the code for doing the ibm,client-architecture-support call. Anyway, the comment says what it's about, so I don't think there's any potential for confusion.

Paul.

Acked-by on v1 1/1
#10 lore raw
In reply to Jake Moilanen
Quoted 5 lines · Jake Moilanen

When booting a Power6 in compat mode, the DFP bit must be set in the client architecture call, or else firmware will downgrade the compliance to P5 compat mode, thus removing VMX and DFP instructions.

Quoted 28 lines

--- 2.6/arch/powerpc/kernel/prom_init.c~ 2006-11-13 15:01:13.000000000
-0600
+++ 2.6/arch/powerpc/kernel/prom_init.c 2006-11-21 18:38:27.000000000
-0600

(quoted diff) +2 -1
@@ -627,6 +627,7 @@ static void __init early_cmdline_parse(v
 /* Option vector 3: processor options supported */
 #define OV3_FP			0x80	/* floating point */
 #define OV3_VMX			0x40	/* VMX/Altivec */
+#define OV3_DFP			0x20	/* decimal floating point */
 
 /* Option vector 5: PAPR/OF options supported */
 #define OV5_LPAR		0x80	/* logical partitioning supported */
@@ -674,7 +675,7 @@ static unsigned char ibm_architecture_ve
 	/* option vector 3: processor options supported */
 	3 - 2,				/* length */
 	0,				/* don't ignore, don't halt */
-	OV3_FP | OV3_VMX,
+	OV3_FP | OV3_VMX | OV3_DFP,
 
 	/* option vector 4: IBM PAPR implementation */
 	2 - 2,				/* length */

_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://ozlabs.org/mailman/listinfo/linuxppc-dev

#11 lore raw
In reply to Jake Moilanen
Quoted 3 lines · Jake Moilanen

When booting a Power6 in compat mode, the DFP bit must be set in the client architecture call, or else firmware will downgrade the compliance to P5 compat mode, thus removing VMX and DFP instructions.

This change is already in the powerpc.git tree; it was part of my "Distinguish POWER6 partition modes and tell userspace" patch.

Paul.