[PATCH 0/2] s390: Raise minimum clang version to 19.1.0

[PATCH 0/2] s390: Raise minimum clang version to 19.1.0

[Heiko Carstens]

In order to get rid of special inline assemblies for old clang versions,
raise the minimum clang version to 19.1.0 for s390.

This allows to make use of various inline assembly format flags for which
support has been added with llvm-project commit 9c75a981554d ("[SystemZ]
Implement A, O and R inline assembly format flags (#80685)").

Heiko Carstens (2):
  scripts/min-tool-version.sh: Raise minimum clang version to 19.1.0 for s390
  s390/fpu: Remove inline assembly variants for old clang versions

 arch/s390/include/asm/fpu-insn.h | 124 -------------------------------
 scripts/min-tool-version.sh      |   2 +-
 2 files changed, 1 insertion(+), 125 deletions(-)

-- 
2.45.2

[PATCH 1/2] scripts/min-tool-version.sh: Raise minimum clang version to 19.1.0 for s390

[Heiko Carstens]

Raise minimum clang version to 19.1.0 for s390 so that various inline
assembly format flags can be used. The missing format flags were
implemented with llvm-project commit 9c75a981554d ("[SystemZ] Implement A,
O and R inline assembly format flags (#80685)").

Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
---
 scripts/min-tool-version.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/scripts/min-tool-version.sh b/scripts/min-tool-version.sh
index 91c91201212c..2dc674a74624 100755
--- a/scripts/min-tool-version.sh
+++ b/scripts/min-tool-version.sh
@@ -25,7 +25,7 @@ gcc)
 	;;
 llvm)
 	if [ "$SRCARCH" = s390 ]; then
-		echo 15.0.0
+		echo 19.1.0
 	elif [ "$SRCARCH" = loongarch ]; then
 		echo 18.0.0
 	else
-- 
2.45.2

Re: [PATCH 1/2] scripts/min-tool-version.sh: Raise minimum clang version to 19.1.0 for s390

[Nathan Chancellor]

On Wed, Nov 13, 2024 at 04:40:12PM +0100, Heiko Carstens wrote:
> Raise minimum clang version to 19.1.0 for s390 so that various inline
> assembly format flags can be used. The missing format flags were
> implemented with llvm-project commit 9c75a981554d ("[SystemZ] Implement A,
> O and R inline assembly format flags (#80685)").
> 
> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>

Oof, this is quite new but now that kernel.org has LLVM binaries
available, I do not think this is an unreasonable ask, especially if it
makes your life easier with code maintenance.

Acked-by: Nathan Chancellor <nathan@kernel.org>

One question: Is it worth dropping the mention about CC=clang for
clang-18 and older in Documentation/kbuild/llvm.rst? Maybe it is better
to leave it around for a bit just in case people read the newer
documentation while working on an older kernel?

> ---
>  scripts/min-tool-version.sh | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/scripts/min-tool-version.sh b/scripts/min-tool-version.sh
> index 91c91201212c..2dc674a74624 100755
> --- a/scripts/min-tool-version.sh
> +++ b/scripts/min-tool-version.sh
> @@ -25,7 +25,7 @@ gcc)
>  	;;
>  llvm)
>  	if [ "$SRCARCH" = s390 ]; then
> -		echo 15.0.0
> +		echo 19.1.0
>  	elif [ "$SRCARCH" = loongarch ]; then
>  		echo 18.0.0
>  	else
> -- 
> 2.45.2
> 

Re: [PATCH 1/2] scripts/min-tool-version.sh: Raise minimum clang version to 19.1.0 for s390

[Heiko Carstens]

On Wed, Nov 13, 2024 at 11:21:09AM -0700, Nathan Chancellor wrote:
> On Wed, Nov 13, 2024 at 04:40:12PM +0100, Heiko Carstens wrote:
> > Raise minimum clang version to 19.1.0 for s390 so that various inline
> > assembly format flags can be used. The missing format flags were
> > implemented with llvm-project commit 9c75a981554d ("[SystemZ] Implement A,
> > O and R inline assembly format flags (#80685)").
> > 
> > Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
> 
> Oof, this is quite new but now that kernel.org has LLVM binaries
> available, I do not think this is an unreasonable ask, especially if it
> makes your life easier with code maintenance.

Thanks for your feedback! I read this as "is nearly unreasonable" :)

Even though you acked this, I guess I will drop this for now and
maybe address this a few clang versions later.

Re: [PATCH 1/2] scripts/min-tool-version.sh: Raise minimum clang version to 19.1.0 for s390

[Nathan Chancellor]

On Thu, Nov 14, 2024 at 06:05:24PM +0100, Heiko Carstens wrote:
> On Wed, Nov 13, 2024 at 11:21:09AM -0700, Nathan Chancellor wrote:
> > On Wed, Nov 13, 2024 at 04:40:12PM +0100, Heiko Carstens wrote:
> > > Raise minimum clang version to 19.1.0 for s390 so that various inline
> > > assembly format flags can be used. The missing format flags were
> > > implemented with llvm-project commit 9c75a981554d ("[SystemZ] Implement A,
> > > O and R inline assembly format flags (#80685)").
> > > 
> > > Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
> > 
> > Oof, this is quite new but now that kernel.org has LLVM binaries
> > available, I do not think this is an unreasonable ask, especially if it
> > makes your life easier with code maintenance.
> 
> Thanks for your feedback! I read this as "is nearly unreasonable" :)

Yeah, that is probably a fair way to put it :)

> Even though you acked this, I guess I will drop this for now and
> maybe address this a few clang versions later.

Yes, I think waiting until that LLVM change has had some time to
actually make it into the hands of users would be good if it is not too
much of a burden. However, if you start writing new kernel code that depends on
these format flags, it may be reasonable to revisit this series at that
point for ease of maintenance on your end.

Cheers,
Nathan

[PATCH 2/2] s390/fpu: Remove inline assembly variants for old clang versions

[Heiko Carstens]

With the minimal clang version raised to 19.1.0 there is no need to keep
the inline assemblies for old clang versions which do not support the O
and R inline assembly format flags.

Support for those flags was added with llvm-project commit 9c75a981554d
("[SystemZ] Implement A, O and R inline assembly format flags (#80685)").

Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
---
 arch/s390/include/asm/fpu-insn.h | 124 -------------------------------
 1 file changed, 124 deletions(-)

diff --git a/arch/s390/include/asm/fpu-insn.h b/arch/s390/include/asm/fpu-insn.h
index c1e2e521d9af..e57d80981f13 100644
--- a/arch/s390/include/asm/fpu-insn.h
+++ b/arch/s390/include/asm/fpu-insn.h
@@ -183,22 +183,6 @@ static __always_inline void fpu_vgfmg(u8 v1, u8 v2, u8 v3)
 		     : "memory");
 }
 
-#ifdef CONFIG_CC_IS_CLANG
-
-static __always_inline void fpu_vl(u8 v1, const void *vxr)
-{
-	instrument_read(vxr, sizeof(__vector128));
-	asm volatile("\n"
-		"	la	1,%[vxr]\n"
-		"	VL	%[v1],0,,1\n"
-		:
-		: [vxr] "R" (*(__vector128 *)vxr),
-		  [v1] "I" (v1)
-		: "memory", "1");
-}
-
-#else /* CONFIG_CC_IS_CLANG */
-
 static __always_inline void fpu_vl(u8 v1, const void *vxr)
 {
 	instrument_read(vxr, sizeof(__vector128));
@@ -209,8 +193,6 @@ static __always_inline void fpu_vl(u8 v1, const void *vxr)
 		     : "memory");
 }
 
-#endif /* CONFIG_CC_IS_CLANG */
-
 static __always_inline void fpu_vleib(u8 v, s16 val, u8 index)
 {
 	asm volatile("VLEIB	%[v],%[val],%[index]"
@@ -238,26 +220,6 @@ static __always_inline u64 fpu_vlgvf(u8 v, u16 index)
 	return val;
 }
 
-#ifdef CONFIG_CC_IS_CLANG
-
-static __always_inline void fpu_vll(u8 v1, u32 index, const void *vxr)
-{
-	unsigned int size;
-
-	size = min(index + 1, sizeof(__vector128));
-	instrument_read(vxr, size);
-	asm volatile("\n"
-		"	la	1,%[vxr]\n"
-		"	VLL	%[v1],%[index],0,1\n"
-		:
-		: [vxr] "R" (*(u8 *)vxr),
-		  [index] "d" (index),
-		  [v1] "I" (v1)
-		: "memory", "1");
-}
-
-#else /* CONFIG_CC_IS_CLANG */
-
 static __always_inline void fpu_vll(u8 v1, u32 index, const void *vxr)
 {
 	unsigned int size;
@@ -272,30 +234,6 @@ static __always_inline void fpu_vll(u8 v1, u32 index, const void *vxr)
 		     : "memory");
 }
 
-#endif /* CONFIG_CC_IS_CLANG */
-
-#ifdef CONFIG_CC_IS_CLANG
-
-#define fpu_vlm(_v1, _v3, _vxrs)					\
-({									\
-	unsigned int size = ((_v3) - (_v1) + 1) * sizeof(__vector128);	\
-	struct {							\
-		__vector128 _v[(_v3) - (_v1) + 1];			\
-	} *_v = (void *)(_vxrs);					\
-									\
-	instrument_read(_v, size);					\
-	asm volatile("\n"						\
-		"	la	1,%[vxrs]\n"				\
-		"	VLM	%[v1],%[v3],0,1\n"			\
-		:							\
-		: [vxrs] "R" (*_v),					\
-		  [v1] "I" (_v1), [v3] "I" (_v3)			\
-		: "memory", "1");					\
-	(_v3) - (_v1) + 1;						\
-})
-
-#else /* CONFIG_CC_IS_CLANG */
-
 #define fpu_vlm(_v1, _v3, _vxrs)					\
 ({									\
 	unsigned int size = ((_v3) - (_v1) + 1) * sizeof(__vector128);	\
@@ -312,8 +250,6 @@ static __always_inline void fpu_vll(u8 v1, u32 index, const void *vxr)
 	(_v3) - (_v1) + 1;						\
 })
 
-#endif /* CONFIG_CC_IS_CLANG */
-
 static __always_inline void fpu_vlr(u8 v1, u8 v2)
 {
 	asm volatile("VLR	%[v1],%[v2]"
@@ -362,21 +298,6 @@ static __always_inline void fpu_vsrlb(u8 v1, u8 v2, u8 v3)
 		     : "memory");
 }
 
-#ifdef CONFIG_CC_IS_CLANG
-
-static __always_inline void fpu_vst(u8 v1, const void *vxr)
-{
-	instrument_write(vxr, sizeof(__vector128));
-	asm volatile("\n"
-		"	la	1,%[vxr]\n"
-		"	VST	%[v1],0,,1\n"
-		: [vxr] "=R" (*(__vector128 *)vxr)
-		: [v1] "I" (v1)
-		: "memory", "1");
-}
-
-#else /* CONFIG_CC_IS_CLANG */
-
 static __always_inline void fpu_vst(u8 v1, const void *vxr)
 {
 	instrument_write(vxr, sizeof(__vector128));
@@ -386,26 +307,6 @@ static __always_inline void fpu_vst(u8 v1, const void *vxr)
 		     : "memory");
 }
 
-#endif /* CONFIG_CC_IS_CLANG */
-
-#ifdef CONFIG_CC_IS_CLANG
-
-static __always_inline void fpu_vstl(u8 v1, u32 index, const void *vxr)
-{
-	unsigned int size;
-
-	size = min(index + 1, sizeof(__vector128));
-	instrument_write(vxr, size);
-	asm volatile("\n"
-		"	la	1,%[vxr]\n"
-		"	VSTL	%[v1],%[index],0,1\n"
-		: [vxr] "=R" (*(u8 *)vxr)
-		: [index] "d" (index), [v1] "I" (v1)
-		: "memory", "1");
-}
-
-#else /* CONFIG_CC_IS_CLANG */
-
 static __always_inline void fpu_vstl(u8 v1, u32 index, const void *vxr)
 {
 	unsigned int size;
@@ -418,29 +319,6 @@ static __always_inline void fpu_vstl(u8 v1, u32 index, const void *vxr)
 		     : "memory");
 }
 
-#endif /* CONFIG_CC_IS_CLANG */
-
-#ifdef CONFIG_CC_IS_CLANG
-
-#define fpu_vstm(_v1, _v3, _vxrs)					\
-({									\
-	unsigned int size = ((_v3) - (_v1) + 1) * sizeof(__vector128);	\
-	struct {							\
-		__vector128 _v[(_v3) - (_v1) + 1];			\
-	} *_v = (void *)(_vxrs);					\
-									\
-	instrument_write(_v, size);					\
-	asm volatile("\n"						\
-		"	la	1,%[vxrs]\n"				\
-		"	VSTM	%[v1],%[v3],0,1\n"			\
-		: [vxrs] "=R" (*_v)					\
-		: [v1] "I" (_v1), [v3] "I" (_v3)			\
-		: "memory", "1");					\
-	(_v3) - (_v1) + 1;						\
-})
-
-#else /* CONFIG_CC_IS_CLANG */
-
 #define fpu_vstm(_v1, _v3, _vxrs)					\
 ({									\
 	unsigned int size = ((_v3) - (_v1) + 1) * sizeof(__vector128);	\
@@ -456,8 +334,6 @@ static __always_inline void fpu_vstl(u8 v1, u32 index, const void *vxr)
 	(_v3) - (_v1) + 1;						\
 })
 
-#endif /* CONFIG_CC_IS_CLANG */
-
 static __always_inline void fpu_vupllf(u8 v1, u8 v2)
 {
 	asm volatile("VUPLLF	%[v1],%[v2]"
-- 
2.45.2

Re: [PATCH 2/2] s390/fpu: Remove inline assembly variants for old clang versions

[Nathan Chancellor]

On Wed, Nov 13, 2024 at 04:40:13PM +0100, Heiko Carstens wrote:
> With the minimal clang version raised to 19.1.0 there is no need to keep
> the inline assemblies for old clang versions which do not support the O
> and R inline assembly format flags.
> 
> Support for those flags was added with llvm-project commit 9c75a981554d
> ("[SystemZ] Implement A, O and R inline assembly format flags (#80685)").
> 
> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>

Reviewed-by: Nathan Chancellor <nathan@kernel.org>

I will trust that it compiles :)

> ---
>  arch/s390/include/asm/fpu-insn.h | 124 -------------------------------
>  1 file changed, 124 deletions(-)
> 
> diff --git a/arch/s390/include/asm/fpu-insn.h b/arch/s390/include/asm/fpu-insn.h
> index c1e2e521d9af..e57d80981f13 100644
> --- a/arch/s390/include/asm/fpu-insn.h
> +++ b/arch/s390/include/asm/fpu-insn.h
> @@ -183,22 +183,6 @@ static __always_inline void fpu_vgfmg(u8 v1, u8 v2, u8 v3)
>  		     : "memory");
>  }
>  
> -#ifdef CONFIG_CC_IS_CLANG
> -
> -static __always_inline void fpu_vl(u8 v1, const void *vxr)
> -{
> -	instrument_read(vxr, sizeof(__vector128));
> -	asm volatile("\n"
> -		"	la	1,%[vxr]\n"
> -		"	VL	%[v1],0,,1\n"
> -		:
> -		: [vxr] "R" (*(__vector128 *)vxr),
> -		  [v1] "I" (v1)
> -		: "memory", "1");
> -}
> -
> -#else /* CONFIG_CC_IS_CLANG */
> -
>  static __always_inline void fpu_vl(u8 v1, const void *vxr)
>  {
>  	instrument_read(vxr, sizeof(__vector128));
> @@ -209,8 +193,6 @@ static __always_inline void fpu_vl(u8 v1, const void *vxr)
>  		     : "memory");
>  }
>  
> -#endif /* CONFIG_CC_IS_CLANG */
> -
>  static __always_inline void fpu_vleib(u8 v, s16 val, u8 index)
>  {
>  	asm volatile("VLEIB	%[v],%[val],%[index]"
> @@ -238,26 +220,6 @@ static __always_inline u64 fpu_vlgvf(u8 v, u16 index)
>  	return val;
>  }
>  
> -#ifdef CONFIG_CC_IS_CLANG
> -
> -static __always_inline void fpu_vll(u8 v1, u32 index, const void *vxr)
> -{
> -	unsigned int size;
> -
> -	size = min(index + 1, sizeof(__vector128));
> -	instrument_read(vxr, size);
> -	asm volatile("\n"
> -		"	la	1,%[vxr]\n"
> -		"	VLL	%[v1],%[index],0,1\n"
> -		:
> -		: [vxr] "R" (*(u8 *)vxr),
> -		  [index] "d" (index),
> -		  [v1] "I" (v1)
> -		: "memory", "1");
> -}
> -
> -#else /* CONFIG_CC_IS_CLANG */
> -
>  static __always_inline void fpu_vll(u8 v1, u32 index, const void *vxr)
>  {
>  	unsigned int size;
> @@ -272,30 +234,6 @@ static __always_inline void fpu_vll(u8 v1, u32 index, const void *vxr)
>  		     : "memory");
>  }
>  
> -#endif /* CONFIG_CC_IS_CLANG */
> -
> -#ifdef CONFIG_CC_IS_CLANG
> -
> -#define fpu_vlm(_v1, _v3, _vxrs)					\
> -({									\
> -	unsigned int size = ((_v3) - (_v1) + 1) * sizeof(__vector128);	\
> -	struct {							\
> -		__vector128 _v[(_v3) - (_v1) + 1];			\
> -	} *_v = (void *)(_vxrs);					\
> -									\
> -	instrument_read(_v, size);					\
> -	asm volatile("\n"						\
> -		"	la	1,%[vxrs]\n"				\
> -		"	VLM	%[v1],%[v3],0,1\n"			\
> -		:							\
> -		: [vxrs] "R" (*_v),					\
> -		  [v1] "I" (_v1), [v3] "I" (_v3)			\
> -		: "memory", "1");					\
> -	(_v3) - (_v1) + 1;						\
> -})
> -
> -#else /* CONFIG_CC_IS_CLANG */
> -
>  #define fpu_vlm(_v1, _v3, _vxrs)					\
>  ({									\
>  	unsigned int size = ((_v3) - (_v1) + 1) * sizeof(__vector128);	\
> @@ -312,8 +250,6 @@ static __always_inline void fpu_vll(u8 v1, u32 index, const void *vxr)
>  	(_v3) - (_v1) + 1;						\
>  })
>  
> -#endif /* CONFIG_CC_IS_CLANG */
> -
>  static __always_inline void fpu_vlr(u8 v1, u8 v2)
>  {
>  	asm volatile("VLR	%[v1],%[v2]"
> @@ -362,21 +298,6 @@ static __always_inline void fpu_vsrlb(u8 v1, u8 v2, u8 v3)
>  		     : "memory");
>  }
>  
> -#ifdef CONFIG_CC_IS_CLANG
> -
> -static __always_inline void fpu_vst(u8 v1, const void *vxr)
> -{
> -	instrument_write(vxr, sizeof(__vector128));
> -	asm volatile("\n"
> -		"	la	1,%[vxr]\n"
> -		"	VST	%[v1],0,,1\n"
> -		: [vxr] "=R" (*(__vector128 *)vxr)
> -		: [v1] "I" (v1)
> -		: "memory", "1");
> -}
> -
> -#else /* CONFIG_CC_IS_CLANG */
> -
>  static __always_inline void fpu_vst(u8 v1, const void *vxr)
>  {
>  	instrument_write(vxr, sizeof(__vector128));
> @@ -386,26 +307,6 @@ static __always_inline void fpu_vst(u8 v1, const void *vxr)
>  		     : "memory");
>  }
>  
> -#endif /* CONFIG_CC_IS_CLANG */
> -
> -#ifdef CONFIG_CC_IS_CLANG
> -
> -static __always_inline void fpu_vstl(u8 v1, u32 index, const void *vxr)
> -{
> -	unsigned int size;
> -
> -	size = min(index + 1, sizeof(__vector128));
> -	instrument_write(vxr, size);
> -	asm volatile("\n"
> -		"	la	1,%[vxr]\n"
> -		"	VSTL	%[v1],%[index],0,1\n"
> -		: [vxr] "=R" (*(u8 *)vxr)
> -		: [index] "d" (index), [v1] "I" (v1)
> -		: "memory", "1");
> -}
> -
> -#else /* CONFIG_CC_IS_CLANG */
> -
>  static __always_inline void fpu_vstl(u8 v1, u32 index, const void *vxr)
>  {
>  	unsigned int size;
> @@ -418,29 +319,6 @@ static __always_inline void fpu_vstl(u8 v1, u32 index, const void *vxr)
>  		     : "memory");
>  }
>  
> -#endif /* CONFIG_CC_IS_CLANG */
> -
> -#ifdef CONFIG_CC_IS_CLANG
> -
> -#define fpu_vstm(_v1, _v3, _vxrs)					\
> -({									\
> -	unsigned int size = ((_v3) - (_v1) + 1) * sizeof(__vector128);	\
> -	struct {							\
> -		__vector128 _v[(_v3) - (_v1) + 1];			\
> -	} *_v = (void *)(_vxrs);					\
> -									\
> -	instrument_write(_v, size);					\
> -	asm volatile("\n"						\
> -		"	la	1,%[vxrs]\n"				\
> -		"	VSTM	%[v1],%[v3],0,1\n"			\
> -		: [vxrs] "=R" (*_v)					\
> -		: [v1] "I" (_v1), [v3] "I" (_v3)			\
> -		: "memory", "1");					\
> -	(_v3) - (_v1) + 1;						\
> -})
> -
> -#else /* CONFIG_CC_IS_CLANG */
> -
>  #define fpu_vstm(_v1, _v3, _vxrs)					\
>  ({									\
>  	unsigned int size = ((_v3) - (_v1) + 1) * sizeof(__vector128);	\
> @@ -456,8 +334,6 @@ static __always_inline void fpu_vstl(u8 v1, u32 index, const void *vxr)
>  	(_v3) - (_v1) + 1;						\
>  })
>  
> -#endif /* CONFIG_CC_IS_CLANG */
> -
>  static __always_inline void fpu_vupllf(u8 v1, u8 v2)
>  {
>  	asm volatile("VUPLLF	%[v1],%[v2]"
> -- 
> 2.45.2
>