diff mbox

[09/11] sparc64: Convert U3copy_{from,to}_user to accurate exception reporting.

Message ID 20161025.110228.1744774836277392991.davem@davemloft.net
State Accepted
Delegated to: David Miller
Headers show

Commit Message

David Miller Oct. 25, 2016, 3:02 p.m. UTC 
Report the exact number of bytes which have not been successfully
copied when an exception occurs, using the running remaining length.

Signed-off-by: David S. Miller <davem@davemloft.net>
---
 arch/sparc/lib/U3copy_from_user.S |   8 +-
 arch/sparc/lib/U3copy_to_user.S   |   8 +-
 arch/sparc/lib/U3memcpy.S         | 227 ++++++++++++++++++++++++++------------
 3 files changed, 162 insertions(+), 81 deletions(-)

Comments

Sam Ravnborg Oct. 25, 2016, 6:39 p.m. UTC | #1 
Hi Dave.

One small comment below.
I did not try to build with these patches, but stumbled over this detail.

On Tue, Oct 25, 2016 at 11:02:28AM -0400, David Miller wrote:
> 
> Report the exact number of bytes which have not been successfully
> copied when an exception occurs, using the running remaining length.
> 
> Signed-off-by: David S. Miller <davem@davemloft.net>
> ---
>  arch/sparc/lib/U3copy_from_user.S |   8 +-
>  arch/sparc/lib/U3copy_to_user.S   |   8 +-
>  arch/sparc/lib/U3memcpy.S         | 227 ++++++++++++++++++++++++++------------
>  3 files changed, 162 insertions(+), 81 deletions(-)
> 
> diff --git a/arch/sparc/lib/U3copy_from_user.S b/arch/sparc/lib/U3copy_from_user.S
> index 1046e2b..db73010 100644
> --- a/arch/sparc/lib/U3copy_from_user.S
> +++ b/arch/sparc/lib/U3copy_from_user.S
> @@ -3,19 +3,19 @@
>   * Copyright (C) 1999, 2000, 2004 David S. Miller (davem@redhat.com)
>   */
>  
> -#define EX_LD(x)		\
> +#define EX_LD(x,y)		\
>  98:	x;			\
>  	.section __ex_table,"a";\
>  	.align 4;		\
> -	.word 98b, __retl_mone;	\
> +	.word 98b, y;		\
>  	.text;			\
>  	.align 4;
>  
> -#define EX_LD_FP(x)		\
> +#define EX_LD_FP(x,y)		\
>  98:	x;			\
>  	.section __ex_table,"a";\
>  	.align 4;		\
> -	.word 98b, __retl_mone_fp;\
> +	.word 98b, y##_fp;	\
>  	.text;			\
>  	.align 4;
>  
> diff --git a/arch/sparc/lib/U3copy_to_user.S b/arch/sparc/lib/U3copy_to_user.S
> index 032b0c5..c4ee858 100644
> --- a/arch/sparc/lib/U3copy_to_user.S
> +++ b/arch/sparc/lib/U3copy_to_user.S
> @@ -3,19 +3,19 @@
>   * Copyright (C) 1999, 2000, 2004 David S. Miller (davem@redhat.com)
>   */
>  
> -#define EX_ST(x)		\
> +#define EX_ST(x,y)		\
>  98:	x;			\
>  	.section __ex_table,"a";\
>  	.align 4;		\
> -	.word 98b, __retl_mone;	\
> +	.word 98b, y;		\
>  	.text;			\
>  	.align 4;
>  
> -#define EX_ST_FP(x)		\
> +#define EX_ST_FP(x,y)		\
>  98:	x;			\
>  	.section __ex_table,"a";\
>  	.align 4;		\
> -	.word 98b, __retl_mone_fp;\
> +	.word 98b, y##_fp;	\
>  	.text;			\
>  	.align 4;
>  
> diff --git a/arch/sparc/lib/U3memcpy.S b/arch/sparc/lib/U3memcpy.S
> index 491ee69..54f9870 100644
> --- a/arch/sparc/lib/U3memcpy.S
> +++ b/arch/sparc/lib/U3memcpy.S
> @@ -4,6 +4,7 @@
>   */
>  
>  #ifdef __KERNEL__
> +#include <linux/linkage.h>
>  #include <asm/visasm.h>
>  #include <asm/asi.h>
>  #define GLOBAL_SPARE	%g7
> @@ -22,21 +23,17 @@
>  #endif
>  
>  #ifndef EX_LD
> -#define EX_LD(x)	x
> +#define EX_LD(x,y)	x

The above line will result in following assembelr code fragment:
.word 98b, ;           \
          ^
          empty
Maybe this is just fine - but it looks wrong.
Same goes for other uses of related macros.

	Sam
--
To unsubscribe from this list: send the line "unsubscribe sparclinux" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
David Miller Oct. 25, 2016, 7:07 p.m. UTC | #2 
From: Sam Ravnborg <sam@ravnborg.org>
Date: Tue, 25 Oct 2016 20:39:40 +0200

> One small comment below.
 ...
>> @@ -22,21 +23,17 @@
>>  #endif
>>  
>>  #ifndef EX_LD
>> -#define EX_LD(x)	x
>> +#define EX_LD(x,y)	x
> 
> The above line will result in following assembelr code fragment:
> .word 98b, ;           \
>           ^
>           empty
> Maybe this is just fine - but it looks wrong.
> Same goes for other uses of related macros.

It's a macro definition which just emits the first argument
when actually used.

Can you explain in more detail what you think happens?
--
To unsubscribe from this list: send the line "unsubscribe sparclinux" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Sam Ravnborg Oct. 25, 2016, 7:16 p.m. UTC | #3 
On Tue, Oct 25, 2016 at 03:07:56PM -0400, David Miller wrote:
> From: Sam Ravnborg <sam@ravnborg.org>
> Date: Tue, 25 Oct 2016 20:39:40 +0200
> 
> > One small comment below.
>  ...
> >> @@ -22,21 +23,17 @@
> >>  #endif
> >>  
> >>  #ifndef EX_LD
> >> -#define EX_LD(x)	x
> >> +#define EX_LD(x,y)	x
> > 
> > The above line will result in following assembelr code fragment:
> > .word 98b, ;           \
> >           ^
> >           empty
> > Maybe this is just fine - but it looks wrong.
> > Same goes for other uses of related macros.
> 
> It's a macro definition which just emits the first argument
> when actually used.
> 
> Can you explain in more detail what you think happens?

Thats simple - I am wasting your time.

I misread what was going on - and was too quick to reply.

	Sam
--
To unsubscribe from this list: send the line "unsubscribe sparclinux" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/arch/sparc/lib/U3copy_from_user.S b/arch/sparc/lib/U3copy_from_user.S
index 1046e2b..db73010 100644
--- a/arch/sparc/lib/U3copy_from_user.S
+++ b/arch/sparc/lib/U3copy_from_user.S
@@ -3,19 +3,19 @@ 
  * Copyright (C) 1999, 2000, 2004 David S. Miller (davem@redhat.com)
  */
 
-#define EX_LD(x)		\
+#define EX_LD(x,y)		\
 98:	x;			\
 	.section __ex_table,"a";\
 	.align 4;		\
-	.word 98b, __retl_mone;	\
+	.word 98b, y;		\
 	.text;			\
 	.align 4;
 
-#define EX_LD_FP(x)		\
+#define EX_LD_FP(x,y)		\
 98:	x;			\
 	.section __ex_table,"a";\
 	.align 4;		\
-	.word 98b, __retl_mone_fp;\
+	.word 98b, y##_fp;	\
 	.text;			\
 	.align 4;
 
diff --git a/arch/sparc/lib/U3copy_to_user.S b/arch/sparc/lib/U3copy_to_user.S
index 032b0c5..c4ee858 100644
--- a/arch/sparc/lib/U3copy_to_user.S
+++ b/arch/sparc/lib/U3copy_to_user.S
@@ -3,19 +3,19 @@ 
  * Copyright (C) 1999, 2000, 2004 David S. Miller (davem@redhat.com)
  */
 
-#define EX_ST(x)		\
+#define EX_ST(x,y)		\
 98:	x;			\
 	.section __ex_table,"a";\
 	.align 4;		\
-	.word 98b, __retl_mone;	\
+	.word 98b, y;		\
 	.text;			\
 	.align 4;
 
-#define EX_ST_FP(x)		\
+#define EX_ST_FP(x,y)		\
 98:	x;			\
 	.section __ex_table,"a";\
 	.align 4;		\
-	.word 98b, __retl_mone_fp;\
+	.word 98b, y##_fp;	\
 	.text;			\
 	.align 4;
 
diff --git a/arch/sparc/lib/U3memcpy.S b/arch/sparc/lib/U3memcpy.S
index 491ee69..54f9870 100644
--- a/arch/sparc/lib/U3memcpy.S
+++ b/arch/sparc/lib/U3memcpy.S
@@ -4,6 +4,7 @@ 
  */
 
 #ifdef __KERNEL__
+#include <linux/linkage.h>
 #include <asm/visasm.h>
 #include <asm/asi.h>
 #define GLOBAL_SPARE	%g7
@@ -22,21 +23,17 @@ 
 #endif
 
 #ifndef EX_LD
-#define EX_LD(x)	x
+#define EX_LD(x,y)	x
 #endif
 #ifndef EX_LD_FP
-#define EX_LD_FP(x)	x
+#define EX_LD_FP(x,y)	x
 #endif
 
 #ifndef EX_ST
-#define EX_ST(x)	x
+#define EX_ST(x,y)	x
 #endif
 #ifndef EX_ST_FP
-#define EX_ST_FP(x)	x
-#endif
-
-#ifndef EX_RETVAL
-#define EX_RETVAL(x)	x
+#define EX_ST_FP(x,y)	x
 #endif
 
 #ifndef LOAD
@@ -77,6 +74,87 @@ 
 	 */
 
 	.text
+#ifndef EX_RETVAL
+#define EX_RETVAL(x)	x
+__restore_fp:
+	VISExitHalf
+	retl
+	 nop
+ENTRY(U3_retl_o2_plus_g2_plus_g1_plus_1_fp)
+	add	%g1, 1, %g1
+	add	%g2, %g1, %g2
+	ba,pt	%xcc, __restore_fp
+	 add	%o2, %g2, %o0
+ENDPROC(U3_retl_o2_plus_g2_plus_g1_plus_1_fp)
+ENTRY(U3_retl_o2_plus_g2_fp)
+	ba,pt	%xcc, __restore_fp
+	 add	%o2, %g2, %o0
+ENDPROC(U3_retl_o2_plus_g2_fp)
+ENTRY(U3_retl_o2_plus_g2_plus_8_fp)
+	add	%g2, 8, %g2
+	ba,pt	%xcc, __restore_fp
+	 add	%o2, %g2, %o0
+ENDPROC(U3_retl_o2_plus_g2_plus_8_fp)
+ENTRY(U3_retl_o2)
+	retl
+	 mov	%o2, %o0
+ENDPROC(U3_retl_o2)
+ENTRY(U3_retl_o2_plus_1)
+	retl
+	 add	%o2, 1, %o0
+ENDPROC(U3_retl_o2_plus_1)
+ENTRY(U3_retl_o2_plus_4)
+	retl
+	 add	%o2, 4, %o0
+ENDPROC(U3_retl_o2_plus_4)
+ENTRY(U3_retl_o2_plus_8)
+	retl
+	 add	%o2, 8, %o0
+ENDPROC(U3_retl_o2_plus_8)
+ENTRY(U3_retl_o2_plus_g1_plus_1)
+	add	%g1, 1, %g1
+	retl
+	 add	%o2, %g1, %o0
+ENDPROC(U3_retl_o2_plus_g1_plus_1)
+ENTRY(U3_retl_o2_fp)
+	ba,pt	%xcc, __restore_fp
+	 mov	%o2, %o0
+ENDPROC(U3_retl_o2_fp)
+ENTRY(U3_retl_o2_plus_o3_sll_6_plus_0x80_fp)
+	sll	%o3, 6, %o3
+	add	%o3, 0x80, %o3
+	ba,pt	%xcc, __restore_fp
+	 add	%o2, %o3, %o0
+ENDPROC(U3_retl_o2_plus_o3_sll_6_plus_0x80_fp)
+ENTRY(U3_retl_o2_plus_o3_sll_6_plus_0x40_fp)
+	sll	%o3, 6, %o3
+	add	%o3, 0x40, %o3
+	ba,pt	%xcc, __restore_fp
+	 add	%o2, %o3, %o0
+ENDPROC(U3_retl_o2_plus_o3_sll_6_plus_0x40_fp)
+ENTRY(U3_retl_o2_plus_GS_plus_0x10)
+	add	GLOBAL_SPARE, 0x10, GLOBAL_SPARE
+	retl
+	 add	%o2, GLOBAL_SPARE, %o0
+ENDPROC(U3_retl_o2_plus_GS_plus_0x10)
+ENTRY(U3_retl_o2_plus_GS_plus_0x08)
+	add	GLOBAL_SPARE, 0x08, GLOBAL_SPARE
+	retl
+	 add	%o2, GLOBAL_SPARE, %o0
+ENDPROC(U3_retl_o2_plus_GS_plus_0x08)
+ENTRY(U3_retl_o2_and_7_plus_GS)
+	and	%o2, 7, %o2
+	retl
+	 add	%o2, GLOBAL_SPARE, %o2
+ENDPROC(U3_retl_o2_and_7_plus_GS)
+ENTRY(U3_retl_o2_and_7_plus_GS_plus_8)
+	add	GLOBAL_SPARE, 8, GLOBAL_SPARE
+	and	%o2, 7, %o2
+	retl
+	 add	%o2, GLOBAL_SPARE, %o2
+ENDPROC(U3_retl_o2_and_7_plus_GS_plus_8)
+#endif
+
 	.align		64
 
 	/* The cheetah's flexible spine, oversized liver, enlarged heart,
@@ -126,8 +204,8 @@  FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 	 and		%g2, 0x38, %g2
 
 1:	subcc		%g1, 0x1, %g1
-	EX_LD_FP(LOAD(ldub, %o1 + 0x00, %o3))
-	EX_ST_FP(STORE(stb, %o3, %o1 + GLOBAL_SPARE))
+	EX_LD_FP(LOAD(ldub, %o1 + 0x00, %o3), U3_retl_o2_plus_g2_plus_g1_plus_1)
+	EX_ST_FP(STORE(stb, %o3, %o1 + GLOBAL_SPARE), U3_retl_o2_plus_g2_plus_g1_plus_1)
 	bgu,pt		%XCC, 1b
 	 add		%o1, 0x1, %o1
 
@@ -138,20 +216,20 @@  FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 	be,pt		%icc, 3f
 	 alignaddr	%o1, %g0, %o1
 
-	EX_LD_FP(LOAD(ldd, %o1, %f4))
-1:	EX_LD_FP(LOAD(ldd, %o1 + 0x8, %f6))
+	EX_LD_FP(LOAD(ldd, %o1, %f4), U3_retl_o2_plus_g2)
+1:	EX_LD_FP(LOAD(ldd, %o1 + 0x8, %f6), U3_retl_o2_plus_g2)
 	add		%o1, 0x8, %o1
 	subcc		%g2, 0x8, %g2
 	faligndata	%f4, %f6, %f0
-	EX_ST_FP(STORE(std, %f0, %o0))
+	EX_ST_FP(STORE(std, %f0, %o0), U3_retl_o2_plus_g2_plus_8)
 	be,pn		%icc, 3f
 	 add		%o0, 0x8, %o0
 
-	EX_LD_FP(LOAD(ldd, %o1 + 0x8, %f4))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x8, %f4), U3_retl_o2_plus_g2)
 	add		%o1, 0x8, %o1
 	subcc		%g2, 0x8, %g2
 	faligndata	%f6, %f4, %f2
-	EX_ST_FP(STORE(std, %f2, %o0))
+	EX_ST_FP(STORE(std, %f2, %o0), U3_retl_o2_plus_g2_plus_8)
 	bne,pt		%icc, 1b
 	 add		%o0, 0x8, %o0
 
@@ -161,25 +239,25 @@  FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 	LOAD(prefetch, %o1 + 0x080, #one_read)
 	LOAD(prefetch, %o1 + 0x0c0, #one_read)
 	LOAD(prefetch, %o1 + 0x100, #one_read)
-	EX_LD_FP(LOAD(ldd, %o1 + 0x000, %f0))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x000, %f0), U3_retl_o2)
 	LOAD(prefetch, %o1 + 0x140, #one_read)
-	EX_LD_FP(LOAD(ldd, %o1 + 0x008, %f2))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x008, %f2), U3_retl_o2)
 	LOAD(prefetch, %o1 + 0x180, #one_read)
-	EX_LD_FP(LOAD(ldd, %o1 + 0x010, %f4))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x010, %f4), U3_retl_o2)
 	LOAD(prefetch, %o1 + 0x1c0, #one_read)
 	faligndata	%f0, %f2, %f16
-	EX_LD_FP(LOAD(ldd, %o1 + 0x018, %f6))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x018, %f6), U3_retl_o2)
 	faligndata	%f2, %f4, %f18
-	EX_LD_FP(LOAD(ldd, %o1 + 0x020, %f8))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x020, %f8), U3_retl_o2)
 	faligndata	%f4, %f6, %f20
-	EX_LD_FP(LOAD(ldd, %o1 + 0x028, %f10))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x028, %f10), U3_retl_o2)
 	faligndata	%f6, %f8, %f22
 
-	EX_LD_FP(LOAD(ldd, %o1 + 0x030, %f12))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x030, %f12), U3_retl_o2)
 	faligndata	%f8, %f10, %f24
-	EX_LD_FP(LOAD(ldd, %o1 + 0x038, %f14))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x038, %f14), U3_retl_o2)
 	faligndata	%f10, %f12, %f26
-	EX_LD_FP(LOAD(ldd, %o1 + 0x040, %f0))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x040, %f0), U3_retl_o2)
 
 	subcc		GLOBAL_SPARE, 0x80, GLOBAL_SPARE
 	add		%o1, 0x40, %o1
@@ -190,26 +268,26 @@  FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 
 	.align		64
 1:
-	EX_LD_FP(LOAD(ldd, %o1 + 0x008, %f2))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x008, %f2), U3_retl_o2_plus_o3_sll_6_plus_0x80)
 	faligndata	%f12, %f14, %f28
-	EX_LD_FP(LOAD(ldd, %o1 + 0x010, %f4))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x010, %f4), U3_retl_o2_plus_o3_sll_6_plus_0x80)
 	faligndata	%f14, %f0, %f30
-	EX_ST_FP(STORE_BLK(%f16, %o0))
-	EX_LD_FP(LOAD(ldd, %o1 + 0x018, %f6))
+	EX_ST_FP(STORE_BLK(%f16, %o0), U3_retl_o2_plus_o3_sll_6_plus_0x80)
+	EX_LD_FP(LOAD(ldd, %o1 + 0x018, %f6), U3_retl_o2_plus_o3_sll_6_plus_0x40)
 	faligndata	%f0, %f2, %f16
 	add		%o0, 0x40, %o0
 
-	EX_LD_FP(LOAD(ldd, %o1 + 0x020, %f8))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x020, %f8), U3_retl_o2_plus_o3_sll_6_plus_0x40)
 	faligndata	%f2, %f4, %f18
-	EX_LD_FP(LOAD(ldd, %o1 + 0x028, %f10))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x028, %f10), U3_retl_o2_plus_o3_sll_6_plus_0x40)
 	faligndata	%f4, %f6, %f20
-	EX_LD_FP(LOAD(ldd, %o1 + 0x030, %f12))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x030, %f12), U3_retl_o2_plus_o3_sll_6_plus_0x40)
 	subcc		%o3, 0x01, %o3
 	faligndata	%f6, %f8, %f22
-	EX_LD_FP(LOAD(ldd, %o1 + 0x038, %f14))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x038, %f14), U3_retl_o2_plus_o3_sll_6_plus_0x80)
 
 	faligndata	%f8, %f10, %f24
-	EX_LD_FP(LOAD(ldd, %o1 + 0x040, %f0))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x040, %f0), U3_retl_o2_plus_o3_sll_6_plus_0x80)
 	LOAD(prefetch, %o1 + 0x1c0, #one_read)
 	faligndata	%f10, %f12, %f26
 	bg,pt		%XCC, 1b
@@ -217,29 +295,29 @@  FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 
 	/* Finally we copy the last full 64-byte block. */
 2:
-	EX_LD_FP(LOAD(ldd, %o1 + 0x008, %f2))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x008, %f2), U3_retl_o2_plus_o3_sll_6_plus_0x80)
 	faligndata	%f12, %f14, %f28
-	EX_LD_FP(LOAD(ldd, %o1 + 0x010, %f4))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x010, %f4), U3_retl_o2_plus_o3_sll_6_plus_0x80)
 	faligndata	%f14, %f0, %f30
-	EX_ST_FP(STORE_BLK(%f16, %o0))
-	EX_LD_FP(LOAD(ldd, %o1 + 0x018, %f6))
+	EX_ST_FP(STORE_BLK(%f16, %o0), U3_retl_o2_plus_o3_sll_6_plus_0x80)
+	EX_LD_FP(LOAD(ldd, %o1 + 0x018, %f6), U3_retl_o2_plus_o3_sll_6_plus_0x40)
 	faligndata	%f0, %f2, %f16
-	EX_LD_FP(LOAD(ldd, %o1 + 0x020, %f8))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x020, %f8), U3_retl_o2_plus_o3_sll_6_plus_0x40)
 	faligndata	%f2, %f4, %f18
-	EX_LD_FP(LOAD(ldd, %o1 + 0x028, %f10))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x028, %f10), U3_retl_o2_plus_o3_sll_6_plus_0x40)
 	faligndata	%f4, %f6, %f20
-	EX_LD_FP(LOAD(ldd, %o1 + 0x030, %f12))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x030, %f12), U3_retl_o2_plus_o3_sll_6_plus_0x40)
 	faligndata	%f6, %f8, %f22
-	EX_LD_FP(LOAD(ldd, %o1 + 0x038, %f14))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x038, %f14), U3_retl_o2_plus_o3_sll_6_plus_0x40)
 	faligndata	%f8, %f10, %f24
 	cmp		%g1, 0
 	be,pt		%XCC, 1f
 	 add		%o0, 0x40, %o0
-	EX_LD_FP(LOAD(ldd, %o1 + 0x040, %f0))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x040, %f0), U3_retl_o2_plus_o3_sll_6_plus_0x40)
 1:	faligndata	%f10, %f12, %f26
 	faligndata	%f12, %f14, %f28
 	faligndata	%f14, %f0, %f30
-	EX_ST_FP(STORE_BLK(%f16, %o0))
+	EX_ST_FP(STORE_BLK(%f16, %o0), U3_retl_o2_plus_o3_sll_6_plus_0x40)
 	add		%o0, 0x40, %o0
 	add		%o1, 0x40, %o1
 	membar		#Sync
@@ -259,20 +337,20 @@  FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 
 	sub		%o2, %g2, %o2
 	be,a,pt		%XCC, 1f
-	 EX_LD_FP(LOAD(ldd, %o1 + 0x00, %f0))
+	 EX_LD_FP(LOAD(ldd, %o1 + 0x00, %f0), U3_retl_o2_plus_g2)
 
-1:	EX_LD_FP(LOAD(ldd, %o1 + 0x08, %f2))
+1:	EX_LD_FP(LOAD(ldd, %o1 + 0x08, %f2), U3_retl_o2_plus_g2)
 	add		%o1, 0x8, %o1
 	subcc		%g2, 0x8, %g2
 	faligndata	%f0, %f2, %f8
-	EX_ST_FP(STORE(std, %f8, %o0))
+	EX_ST_FP(STORE(std, %f8, %o0), U3_retl_o2_plus_g2_plus_8)
 	be,pn		%XCC, 2f
 	 add		%o0, 0x8, %o0
-	EX_LD_FP(LOAD(ldd, %o1 + 0x08, %f0))
+	EX_LD_FP(LOAD(ldd, %o1 + 0x08, %f0), U3_retl_o2_plus_g2)
 	add		%o1, 0x8, %o1
 	subcc		%g2, 0x8, %g2
 	faligndata	%f2, %f0, %f8
-	EX_ST_FP(STORE(std, %f8, %o0))
+	EX_ST_FP(STORE(std, %f8, %o0), U3_retl_o2_plus_g2_plus_8)
 	bne,pn		%XCC, 1b
 	 add		%o0, 0x8, %o0
 
@@ -292,30 +370,33 @@  FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 	 andcc		%o2, 0x8, %g0
 	be,pt		%icc, 1f
 	 nop
-	EX_LD(LOAD(ldx, %o1, %o5))
-	EX_ST(STORE(stx, %o5, %o1 + %o3))
+	EX_LD(LOAD(ldx, %o1, %o5), U3_retl_o2)
+	EX_ST(STORE(stx, %o5, %o1 + %o3), U3_retl_o2)
 	add		%o1, 0x8, %o1
+	sub		%o2, 8, %o2
 
 1:	andcc		%o2, 0x4, %g0
 	be,pt		%icc, 1f
 	 nop
-	EX_LD(LOAD(lduw, %o1, %o5))
-	EX_ST(STORE(stw, %o5, %o1 + %o3))
+	EX_LD(LOAD(lduw, %o1, %o5), U3_retl_o2)
+	EX_ST(STORE(stw, %o5, %o1 + %o3), U3_retl_o2)
 	add		%o1, 0x4, %o1
+	sub		%o2, 4, %o2
 
 1:	andcc		%o2, 0x2, %g0
 	be,pt		%icc, 1f
 	 nop
-	EX_LD(LOAD(lduh, %o1, %o5))
-	EX_ST(STORE(sth, %o5, %o1 + %o3))
+	EX_LD(LOAD(lduh, %o1, %o5), U3_retl_o2)
+	EX_ST(STORE(sth, %o5, %o1 + %o3), U3_retl_o2)
 	add		%o1, 0x2, %o1
+	sub		%o2, 2, %o2
 
 1:	andcc		%o2, 0x1, %g0
 	be,pt		%icc, 85f
 	 nop
-	EX_LD(LOAD(ldub, %o1, %o5))
+	EX_LD(LOAD(ldub, %o1, %o5), U3_retl_o2)
 	ba,pt		%xcc, 85f
-	 EX_ST(STORE(stb, %o5, %o1 + %o3))
+	 EX_ST(STORE(stb, %o5, %o1 + %o3), U3_retl_o2)
 
 	.align		64
 70: /* 16 < len <= 64 */
@@ -326,26 +407,26 @@  FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 	andn		%o2, 0xf, GLOBAL_SPARE
 	and		%o2, 0xf, %o2
 1:	subcc		GLOBAL_SPARE, 0x10, GLOBAL_SPARE
-	EX_LD(LOAD(ldx, %o1 + 0x00, %o5))
-	EX_LD(LOAD(ldx, %o1 + 0x08, %g1))
-	EX_ST(STORE(stx, %o5, %o1 + %o3))
+	EX_LD(LOAD(ldx, %o1 + 0x00, %o5), U3_retl_o2_plus_GS_plus_0x10)
+	EX_LD(LOAD(ldx, %o1 + 0x08, %g1), U3_retl_o2_plus_GS_plus_0x10)
+	EX_ST(STORE(stx, %o5, %o1 + %o3), U3_retl_o2_plus_GS_plus_0x10)
 	add		%o1, 0x8, %o1
-	EX_ST(STORE(stx, %g1, %o1 + %o3))
+	EX_ST(STORE(stx, %g1, %o1 + %o3), U3_retl_o2_plus_GS_plus_0x08)
 	bgu,pt		%XCC, 1b
 	 add		%o1, 0x8, %o1
 73:	andcc		%o2, 0x8, %g0
 	be,pt		%XCC, 1f
 	 nop
 	sub		%o2, 0x8, %o2
-	EX_LD(LOAD(ldx, %o1, %o5))
-	EX_ST(STORE(stx, %o5, %o1 + %o3))
+	EX_LD(LOAD(ldx, %o1, %o5), U3_retl_o2_plus_8)
+	EX_ST(STORE(stx, %o5, %o1 + %o3), U3_retl_o2_plus_8)
 	add		%o1, 0x8, %o1
 1:	andcc		%o2, 0x4, %g0
 	be,pt		%XCC, 1f
 	 nop
 	sub		%o2, 0x4, %o2
-	EX_LD(LOAD(lduw, %o1, %o5))
-	EX_ST(STORE(stw, %o5, %o1 + %o3))
+	EX_LD(LOAD(lduw, %o1, %o5), U3_retl_o2_plus_4)
+	EX_ST(STORE(stw, %o5, %o1 + %o3), U3_retl_o2_plus_4)
 	add		%o1, 0x4, %o1
 1:	cmp		%o2, 0
 	be,pt		%XCC, 85f
@@ -361,8 +442,8 @@  FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 	sub		%o2, %g1, %o2
 
 1:	subcc		%g1, 1, %g1
-	EX_LD(LOAD(ldub, %o1, %o5))
-	EX_ST(STORE(stb, %o5, %o1 + %o3))
+	EX_LD(LOAD(ldub, %o1, %o5), U3_retl_o2_plus_g1_plus_1)
+	EX_ST(STORE(stb, %o5, %o1 + %o3), U3_retl_o2_plus_g1_plus_1)
 	bgu,pt		%icc, 1b
 	 add		%o1, 1, %o1
 
@@ -378,16 +459,16 @@  FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 
 8:	mov		64, %o3
 	andn		%o1, 0x7, %o1
-	EX_LD(LOAD(ldx, %o1, %g2))
+	EX_LD(LOAD(ldx, %o1, %g2), U3_retl_o2)
 	sub		%o3, %g1, %o3
 	andn		%o2, 0x7, GLOBAL_SPARE
 	sllx		%g2, %g1, %g2
-1:	EX_LD(LOAD(ldx, %o1 + 0x8, %g3))
+1:	EX_LD(LOAD(ldx, %o1 + 0x8, %g3), U3_retl_o2_and_7_plus_GS)
 	subcc		GLOBAL_SPARE, 0x8, GLOBAL_SPARE
 	add		%o1, 0x8, %o1
 	srlx		%g3, %o3, %o5
 	or		%o5, %g2, %o5
-	EX_ST(STORE(stx, %o5, %o0))
+	EX_ST(STORE(stx, %o5, %o0), U3_retl_o2_and_7_plus_GS_plus_8)
 	add		%o0, 0x8, %o0
 	bgu,pt		%icc, 1b
 	 sllx		%g3, %g1, %g2
@@ -407,8 +488,8 @@  FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 
 1:
 	subcc		%o2, 4, %o2
-	EX_LD(LOAD(lduw, %o1, %g1))
-	EX_ST(STORE(stw, %g1, %o1 + %o3))
+	EX_LD(LOAD(lduw, %o1, %g1), U3_retl_o2_plus_4)
+	EX_ST(STORE(stw, %g1, %o1 + %o3), U3_retl_o2_plus_4)
 	bgu,pt		%XCC, 1b
 	 add		%o1, 4, %o1
 
@@ -418,8 +499,8 @@  FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 	.align		32
 90:
 	subcc		%o2, 1, %o2
-	EX_LD(LOAD(ldub, %o1, %g1))
-	EX_ST(STORE(stb, %g1, %o1 + %o3))
+	EX_LD(LOAD(ldub, %o1, %g1), U3_retl_o2_plus_1)
+	EX_ST(STORE(stb, %g1, %o1 + %o3), U3_retl_o2_plus_1)
 	bgu,pt		%XCC, 90b
 	 add		%o1, 1, %o1
 	retl