[PATCH] A BTFIXUP'd fix for pte_read()

[PATCH] A BTFIXUP'd fix for pte_read()

[Art Haas]

Hi.

Here's my attempt at taking Bob Breuer's patch and adding in some of the
BTFIXUP magic that sparc32 needs. I'm running the current BK kernel with
this patch on my SS20, and things seem to work. The current kernel
complains during bootup and shutdown about accessing the system clock,
a problem I am guessing is related to changes elsewhere in the kernel.

Art Haas

=== arch/sparc/mm/srmmu.c 1.49 vs edited ==--- 1.49/arch/sparc/mm/srmmu.c	2005-01-20 23:02:11 -06:00
+++ edited/arch/sparc/mm/srmmu.c	2005-02-02 09:39:49 -06:00
@@ -2197,6 +2197,7 @@
 	BTFIXUPSET_CALL(free_pgd_fast, srmmu_free_pgd_fast, BTFIXUPCALL_NORM);
 	BTFIXUPSET_CALL(get_pgd_fast, srmmu_get_pgd_fast, BTFIXUPCALL_NORM);
 
+	BTFIXUPSET_HALF(pte_readi, SRMMU_NOREAD);
 	BTFIXUPSET_HALF(pte_writei, SRMMU_WRITE);
 	BTFIXUPSET_HALF(pte_dirtyi, SRMMU_DIRTY);
 	BTFIXUPSET_HALF(pte_youngi, SRMMU_REF);
=== include/asm-sparc/pgtable.h 1.24 vs edited ==--- 1.24/include/asm-sparc/pgtable.h	2004-12-22 03:36:07 -06:00
+++ edited/include/asm-sparc/pgtable.h	2005-02-02 09:51:07 -06:00
@@ -186,9 +186,16 @@
  * The following only work if pte_present() is true.
  * Undefined behaviour if not..
  */
+BTFIXUPDEF_HALF(pte_readi)
 BTFIXUPDEF_HALF(pte_writei)
 BTFIXUPDEF_HALF(pte_dirtyi)
 BTFIXUPDEF_HALF(pte_youngi)
+
+extern int pte_read(pte_t pte) __attribute_const__;
+extern __inline__ int pte_read(pte_t pte)
+{
+	return !(pte_val(pte) & BTFIXUP_HALF(pte_readi));
+}
 
 extern int pte_write(pte_t pte) __attribute_const__;
 extern __inline__ int pte_write(pte_t pte)
=== include/asm-sparc/pgtsrmmu.h 1.12 vs edited ==--- 1.12/include/asm-sparc/pgtsrmmu.h	2004-07-12 01:53:22 -05:00
+++ edited/include/asm-sparc/pgtsrmmu.h	2005-02-01 10:45:31 -06:00
@@ -73,6 +73,7 @@
 #define SRMMU_CACHE        0x80
 #define SRMMU_DIRTY        0x40
 #define SRMMU_REF          0x20
+#define SRMMU_NOREAD       0x10
 #define SRMMU_EXEC         0x08
 #define SRMMU_WRITE        0x04
 #define SRMMU_VALID        0x02 /* SRMMU_ET_PTE */
-- 
Man once surrendering his reason, has no remaining guard against absurdities
the most monstrous, and like a ship without rudder, is the sport of every wind.

-Thomas Jefferson to James Smith, 1822

Re: [PATCH] A BTFIXUP'd fix for pte_read()

[Tom 'spot' Callaway]

[-- Attachment #1: Type: text/plain, Size: 744 bytes --]

On Wed, 2005-02-02 at 14:04 -0600, Art Haas wrote:
> Hi.
> 
> Here's my attempt at taking Bob Breuer's patch and adding in some of the
> BTFIXUP magic that sparc32 needs. I'm running the current BK kernel with
> this patch on my SS20, and things seem to work. The current kernel
> complains during bootup and shutdown about accessing the system clock,
> a problem I am guessing is related to changes elsewhere in the kernel.

This patch only covers sun4m. The attached patch covers all sparc32.

Signed-off-by: Tom 'spot' Callaway <tcallawa@redhat.com>

~spot
---
Tom "spot" Callaway <tcallawa(a)redhat*com> LCA, RHCE 
Red Hat Sales Engineer || Aurora Linux Project Leader

"If you are going through hell, keep going."
-- Sir Winston Churchill

[-- Attachment #2: linux-2.6.10-sparc-pte_read.patch --]
[-- Type: text/x-patch, Size: 2365 bytes --]

--- linux-2.6.10/arch/sparc/mm/srmmu.c.BAD	2005-01-30 13:41:29.238542766 -0500
+++ linux-2.6.10/arch/sparc/mm/srmmu.c	2005-01-30 13:45:57.034831542 -0500
@@ -2197,6 +2197,7 @@
 	BTFIXUPSET_CALL(free_pgd_fast, srmmu_free_pgd_fast, BTFIXUPCALL_NORM);
 	BTFIXUPSET_CALL(get_pgd_fast, srmmu_get_pgd_fast, BTFIXUPCALL_NORM);
 
+	BTFIXUPSET_HALF(pte_readi, SRMMU_NOREAD);
 	BTFIXUPSET_HALF(pte_writei, SRMMU_WRITE);
 	BTFIXUPSET_HALF(pte_dirtyi, SRMMU_DIRTY);
 	BTFIXUPSET_HALF(pte_youngi, SRMMU_REF);
--- linux-2.6.10/arch/sparc/mm/sun4c.c.BAD	2005-01-30 13:47:47.488040102 -0500
+++ linux-2.6.10/arch/sparc/mm/sun4c.c	2005-01-30 13:50:42.850380974 -0500
@@ -2225,6 +2225,7 @@
 	BTFIXUPSET_CALL(free_pgd_fast, sun4c_free_pgd_fast, BTFIXUPCALL_NORM);
 	BTFIXUPSET_CALL(get_pgd_fast, sun4c_get_pgd_fast, BTFIXUPCALL_NORM);
 
+	BTFIXUPSET_HALF(pte_readi, _SUN4C_PAGE_READ);
 	BTFIXUPSET_HALF(pte_writei, _SUN4C_PAGE_WRITE);
 	BTFIXUPSET_HALF(pte_dirtyi, _SUN4C_PAGE_MODIFIED);
 	BTFIXUPSET_HALF(pte_youngi, _SUN4C_PAGE_ACCESSED);
--- linux-2.6.10/include/asm-sparc/pgtsrmmu.h.BAD	2005-01-30 12:37:27.924511262 -0500
+++ linux-2.6.10/include/asm-sparc/pgtsrmmu.h	2005-01-30 12:37:35.554351350 -0500
@@ -73,6 +73,7 @@
 #define SRMMU_CACHE        0x80
 #define SRMMU_DIRTY        0x40
 #define SRMMU_REF          0x20
+#define SRMMU_NOREAD       0x10
 #define SRMMU_EXEC         0x08
 #define SRMMU_WRITE        0x04
 #define SRMMU_VALID        0x02 /* SRMMU_ET_PTE */
--- linux-2.6.10/include/asm-sparc/pgtable.h.BAD	2005-01-30 14:19:42.633893678 -0500
+++ linux-2.6.10/include/asm-sparc/pgtable.h	2005-01-30 14:22:17.764310270 -0500
@@ -186,10 +186,31 @@
  * The following only work if pte_present() is true.
  * Undefined behaviour if not..
  */
+BTFIXUPDEF_HALF(pte_readi)
 BTFIXUPDEF_HALF(pte_writei)
 BTFIXUPDEF_HALF(pte_dirtyi)
 BTFIXUPDEF_HALF(pte_youngi)
 
+extern int pte_read(pte_t pte) __attribute_const__;
+extern __inline__ int pte_read(pte_t pte)
+{
+	switch (sparc_cpu_model){
+	case sun4:
+	case sun4c:
+		return pte_val(pte) & BTFIXUP_HALF(pte_readi);
+	case sun4d:
+	case sun4e:
+	case sun4m:
+		return !(pte_val(pte) & BTFIXUP_HALF(pte_readi));
+	/* pacify gcc warnings */
+	case sun4u:
+	case sun_unknown:
+	case ap1000:
+	default:
+		return 0;
+	}
+}
+
 extern int pte_write(pte_t pte) __attribute_const__;
 extern __inline__ int pte_write(pte_t pte)
 {

Re: [PATCH] A BTFIXUP'd fix for pte_read()

[William Lee Irwin III]

On Wed, 2005-02-02 at 14:04 -0600, Art Haas wrote:
>> Here's my attempt at taking Bob Breuer's patch and adding in some of the
>> BTFIXUP magic that sparc32 needs. I'm running the current BK kernel with
>> this patch on my SS20, and things seem to work. The current kernel
>> complains during bootup and shutdown about accessing the system clock,
>> a problem I am guessing is related to changes elsewhere in the kernel.

On Wed, Feb 02, 2005 at 03:54:53PM -0600, Tom 'spot' Callaway wrote:
> This patch only covers sun4m. The attached patch covers all sparc32.
> Signed-off-by: Tom 'spot' Callaway <tcallawa@redhat.com>

Doing a quick doublecheck to be sure this actually applies before
shipping upstream. Sorry about the delay, folks.


-- wli

Re: [PATCH] A BTFIXUP'd fix for pte_read()

[David S. Miller]

On Wed, 2 Feb 2005 14:40:19 -0800
William Lee Irwin III <wli@holomorphy.com> wrote:

> On Wed, 2005-02-02 at 14:04 -0600, Art Haas wrote:
> >> Here's my attempt at taking Bob Breuer's patch and adding in some of the
> >> BTFIXUP magic that sparc32 needs. I'm running the current BK kernel with
> >> this patch on my SS20, and things seem to work. The current kernel
> >> complains during bootup and shutdown about accessing the system clock,
> >> a problem I am guessing is related to changes elsewhere in the kernel.
> 
> On Wed, Feb 02, 2005 at 03:54:53PM -0600, Tom 'spot' Callaway wrote:
> > This patch only covers sun4m. The attached patch covers all sparc32.
> > Signed-off-by: Tom 'spot' Callaway <tcallawa@redhat.com>
> 
> Doing a quick doublecheck to be sure this actually applies before
> shipping upstream. Sorry about the delay, folks.

I disagree with Tom's patch in that it puts this huge switch()
statement inline.  The whole point of BTFIXUP_HIGH() is that it
resolves the call to a single instruction which can be easily
patched at runtime.

If it's going to expand to multiple BTFIXUP_HIGH() calls and a
switch statement, just make it a normal BTFIXUP() to a function
and place the implementation in sun4c.c and sun4m.c

Re: [PATCH] A BTFIXUP'd fix for pte_read()

[David S. Miller]

On Wed, 2 Feb 2005 16:55:41 -0800
William Lee Irwin III <wli@holomorphy.com> wrote:

> On Wed, 2 Feb 2005 14:40:19 -0800 William Lee Irwin III <wli@holomorphy.com> wrote:
>
> On Wed, Feb 02, 2005 at 03:32:21PM -0800, David S. Miller wrote:
> > I disagree with Tom's patch in that it puts this huge switch()
> > statement inline.  The whole point of BTFIXUP_HIGH() is that it
> > resolves the call to a single instruction which can be easily
> > patched at runtime.
> > If it's going to expand to multiple BTFIXUP_HIGH() calls and a
> > switch statement, just make it a normal BTFIXUP() to a function
> > and place the implementation in sun4c.c and sun4m.c
> 
> I'm on the road and people are antsy for something that builds, so
> I'm rather tempted to take it for expedience and properly BTFIXUP-ize
> it after 2.6.11, which appears rather imminent, since I basically
> won't push anything I've written myself but not tested.

No objection to using it as a short-term fix.

Re: [PATCH] A BTFIXUP'd fix for pte_read()

[William Lee Irwin III]

On Wed, 2 Feb 2005 14:40:19 -0800 William Lee Irwin III <wli@holomorphy.com> wrote:
>> Doing a quick doublecheck to be sure this actually applies before
>> shipping upstream. Sorry about the delay, folks.

On Wed, Feb 02, 2005 at 03:32:21PM -0800, David S. Miller wrote:
> I disagree with Tom's patch in that it puts this huge switch()
> statement inline.  The whole point of BTFIXUP_HIGH() is that it
> resolves the call to a single instruction which can be easily
> patched at runtime.
> If it's going to expand to multiple BTFIXUP_HIGH() calls and a
> switch statement, just make it a normal BTFIXUP() to a function
> and place the implementation in sun4c.c and sun4m.c

I'm on the road and people are antsy for something that builds, so
I'm rather tempted to take it for expedience and properly BTFIXUP-ize
it after 2.6.11, which appears rather imminent, since I basically
won't push anything I've written myself but not tested. What I have in
terms of remote access I've lost for want of time. There's also
something of a reluctance to resynch before 2.6.11 out of disgust for
certain changes and/or the manner in which they were merged.

Maybe I can cook something better up before my return flight while at
the airport, though I'd be glad to see spot or ahaas get done before me.


-- wli

Re: [PATCH] A BTFIXUP'd fix for pte_read()

[Tom 'spot' Callaway]

[-- Attachment #1: Type: text/plain, Size: 738 bytes --]

On Wed, 2005-02-02 at 15:32 -0800, David S. Miller wrote:

> I disagree with Tom's patch in that it puts this huge switch()
> statement inline.  The whole point of BTFIXUP_HIGH() is that it
> resolves the call to a single instruction which can be easily
> patched at runtime.
> 
> If it's going to expand to multiple BTFIXUP_HIGH() calls and a
> switch statement, just make it a normal BTFIXUP() to a function
> and place the implementation in sun4c.c and sun4m.c

How about this instead?

Signed-off-by: Tom 'spot' Callaway <tcallawa@redhat.com>

~spot
---
Tom "spot" Callaway <tcallawa(a)redhat*com> LCA, RHCE 
Red Hat Sales Engineer || Aurora Linux Project Leader

"If you are going through hell, keep going."
-- Sir Winston Churchill

[-- Attachment #2: linux-2.6.10-sparc-pte_read2.patch --]
[-- Type: text/x-patch, Size: 2721 bytes --]

--- linux-2.6.10/arch/sparc/mm/srmmu.c.BAD	2005-02-02 23:04:32.308346046 -0500
+++ linux-2.6.10/arch/sparc/mm/srmmu.c	2005-02-02 23:05:14.682904134 -0500
@@ -160,6 +160,9 @@
 static inline int srmmu_pte_present(pte_t pte)
 { return ((pte_val(pte) & SRMMU_ET_MASK) == SRMMU_ET_PTE); }
 
+static inline int srmmu_pte_read(pte_t pte)
+{ return !(pte_val(pte) & SRMMU_NOREAD); }
+
 static inline void srmmu_pte_clear(pte_t *ptep)
 { srmmu_set_pte(ptep, __pte(0)); }
 
@@ -2167,6 +2170,7 @@
 
 	BTFIXUPSET_CALL(pte_present, srmmu_pte_present, BTFIXUPCALL_NORM);
 	BTFIXUPSET_CALL(pte_clear, srmmu_pte_clear, BTFIXUPCALL_SWAPO0G0);
+	BTFIXUPSET_CALL(pte_read, srmmu_pte_read, BTFIXUPCALL_NORM);
 
 	BTFIXUPSET_CALL(pmd_bad, srmmu_pmd_bad, BTFIXUPCALL_NORM);
 	BTFIXUPSET_CALL(pmd_present, srmmu_pmd_present, BTFIXUPCALL_NORM);
--- linux-2.6.10/arch/sparc/mm/sun4c.c.BAD	2005-02-02 23:04:19.581280854 -0500
+++ linux-2.6.10/arch/sparc/mm/sun4c.c	2005-02-02 23:04:08.354987510 -0500
@@ -1745,6 +1745,11 @@
 }
 static void sun4c_pte_clear(pte_t *ptep)	{ *ptep = __pte(0); }
 
+static int sun4c_pte_read(pte_t pte)
+{
+        return (pte_val(pte) & _SUN4C_PAGE_READ);
+}
+
 static int sun4c_pmd_bad(pmd_t pmd)
 {
 	return (((pmd_val(pmd) & ~PAGE_MASK) != PGD_TABLE) ||
@@ -2199,6 +2204,7 @@
 
 	BTFIXUPSET_CALL(pte_present, sun4c_pte_present, BTFIXUPCALL_NORM);
 	BTFIXUPSET_CALL(pte_clear, sun4c_pte_clear, BTFIXUPCALL_STG0O0);
+	BTFIXUPSET_CALL(pte_read, sun4c_pte_read, BTFIXUPCALL_NORM);
 
 	BTFIXUPSET_CALL(pmd_bad, sun4c_pmd_bad, BTFIXUPCALL_NORM);
 	BTFIXUPSET_CALL(pmd_present, sun4c_pmd_present, BTFIXUPCALL_NORM);
--- linux-2.6.10/include/asm-sparc/pgtable.h.BAD	2005-02-02 23:07:05.633037150 -0500
+++ linux-2.6.10/include/asm-sparc/pgtable.h	2005-02-02 23:08:32.685803118 -0500
@@ -150,6 +150,7 @@
 BTFIXUPDEF_SETHI(none_mask)
 BTFIXUPDEF_CALL_CONST(int, pte_present, pte_t)
 BTFIXUPDEF_CALL(void, pte_clear, pte_t *)
+BTFIXUPDEF_CALL(int, pte_read, pte_t)
 
 extern __inline__ int pte_none(pte_t pte)
 {
@@ -158,6 +159,7 @@
 
 #define pte_present(pte) BTFIXUP_CALL(pte_present)(pte)
 #define pte_clear(pte) BTFIXUP_CALL(pte_clear)(pte)
+#define pte_read(pte) BTFIXUP_CALL(pte_read)(pte)
 
 BTFIXUPDEF_CALL_CONST(int, pmd_bad, pmd_t)
 BTFIXUPDEF_CALL_CONST(int, pmd_present, pmd_t)
--- linux-2.6.10/include/asm-sparc/pgtsrmmu.h.BAD	2005-02-02 23:08:51.855888822 -0500
+++ linux-2.6.10/include/asm-sparc/pgtsrmmu.h	2005-02-02 22:46:44.398692990 -0500
@@ -73,6 +73,7 @@
 #define SRMMU_CACHE        0x80
 #define SRMMU_DIRTY        0x40
 #define SRMMU_REF          0x20
+#define SRMMU_NOREAD       0x10
 #define SRMMU_EXEC         0x08
 #define SRMMU_WRITE        0x04
 #define SRMMU_VALID        0x02 /* SRMMU_ET_PTE */

Re: [PATCH] A BTFIXUP'd fix for pte_read()

[David S. Miller]

On Wed, 02 Feb 2005 22:22:24 -0600
"Tom 'spot' Callaway" <tcallawa@redhat.com> wrote:

> On Wed, 2005-02-02 at 15:32 -0800, David S. Miller wrote:
> 
> > I disagree with Tom's patch in that it puts this huge switch()
> > statement inline.  The whole point of BTFIXUP_HIGH() is that it
> > resolves the call to a single instruction which can be easily
> > patched at runtime.
> > 
> > If it's going to expand to multiple BTFIXUP_HIGH() calls and a
> > switch statement, just make it a normal BTFIXUP() to a function
> > and place the implementation in sun4c.c and sun4m.c
> 
> How about this instead?

Perfect Tom.

Re: [PATCH] A BTFIXUP'd fix for pte_read()

[William Lee Irwin III]

On Wed, 2005-02-02 at 15:32 -0800, David S. Miller wrote:
>> I disagree with Tom's patch in that it puts this huge switch()
>> statement inline.  The whole point of BTFIXUP_HIGH() is that it
>> resolves the call to a single instruction which can be easily
>> patched at runtime.
>> If it's going to expand to multiple BTFIXUP_HIGH() calls and a
>> switch statement, just make it a normal BTFIXUP() to a function
>> and place the implementation in sun4c.c and sun4m.c

On Wed, Feb 02, 2005 at 10:22:24PM -0600, Tom 'spot' Callaway wrote:
> How about this instead?
> Signed-off-by: Tom 'spot' Callaway <tcallawa@redhat.com>

I'll make sure this gets into 2.6.11-final.


-- wli