[PATCH 1/2] ARM: dcache: select DCACHE_WORD_ACCESS for big-endian CPUs

[PATCH 1/2] ARM: dcache: select DCACHE_WORD_ACCESS for big-endian CPUs

[Will Deacon]

With commit 11ec50caedb5 ("word-at-a-time: provide generic big-endian
zero_bytemask implementation"), the asm-generic word-at-a-time code now
provides a zero_bytemask implementation, allowing us to make use of
DCACHE_WORD_ACCESS on big-endian CPUs, providing our
load_unaligned_zeropad function is endianness-clean.

This patch reworks the load_unaligned_zeropad fixup code to work for
both big- and little-endian CPUs, then removes the !CPU_BIG_ENDIAN check
when selecting DCACHE_WORD_ACCESS.

Signed-off-by: Will Deacon <will.deacon@arm.com>
---
 arch/arm/Kconfig                      |  2 +-
 arch/arm/include/asm/word-at-a-time.h | 18 ++++++++++--------
 2 files changed, 11 insertions(+), 9 deletions(-)

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 88f726d36a40..38ccfd0adbf9 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -11,7 +11,7 @@ config ARM
 	select BUILDTIME_EXTABLE_SORT if MMU
 	select CLONE_BACKWARDS
 	select CPU_PM if (SUSPEND || CPU_IDLE)
-	select DCACHE_WORD_ACCESS if (CPU_V6 || CPU_V6K || CPU_V7) && !CPU_BIG_ENDIAN && MMU
+	select DCACHE_WORD_ACCESS if (CPU_V6 || CPU_V6K || CPU_V7) && MMU
 	select GENERIC_ATOMIC64 if (CPU_V7M || CPU_V6 || !CPU_32v6K || !AEABI)
 	select GENERIC_CLOCKEVENTS_BROADCAST if SMP
 	select GENERIC_IDLE_POLL_SETUP
diff --git a/arch/arm/include/asm/word-at-a-time.h b/arch/arm/include/asm/word-at-a-time.h
index 4d52f92967a6..a6d0a29861e7 100644
--- a/arch/arm/include/asm/word-at-a-time.h
+++ b/arch/arm/include/asm/word-at-a-time.h
@@ -48,10 +48,14 @@ static inline unsigned long find_zero(unsigned long mask)
 	return ret;
 }
 
-#ifdef CONFIG_DCACHE_WORD_ACCESS
-
 #define zero_bytemask(mask) (mask)
 
+#else	/* __ARMEB__ */
+#include <asm-generic/word-at-a-time.h>
+#endif
+
+#ifdef CONFIG_DCACHE_WORD_ACCESS
+
 /*
  * Load an unaligned word from kernel space.
  *
@@ -73,7 +77,11 @@ static inline unsigned long load_unaligned_zeropad(const void *addr)
 	"	bic	%2, %2, #0x3\n"
 	"	ldr	%0, [%2]\n"
 	"	lsl	%1, %1, #0x3\n"
+#ifndef __ARMEB__
 	"	lsr	%0, %0, %1\n"
+#else
+	"	lsl	%0, %0, %1\n"
+#endif
 	"	b	2b\n"
 	"	.popsection\n"
 	"	.pushsection __ex_table,\"a\"\n"
@@ -86,11 +94,5 @@ static inline unsigned long load_unaligned_zeropad(const void *addr)
 	return ret;
 }
 
-
 #endif	/* DCACHE_WORD_ACCESS */
-
-#else	/* __ARMEB__ */
-#include <asm-generic/word-at-a-time.h>
-#endif
-
 #endif /* __ASM_ARM_WORD_AT_A_TIME_H */
-- 
1.8.2.2

[PATCH 2/2] ARM: kconfig: select HAVE_EFFICIENT_UNALIGNED_ACCESS for CPUv6+ && MMU

[Will Deacon]

Modern ARM CPUs can perform efficient unaligned memory accesses in
hardware and this feature is relied up on by code such as the dcache
word-at-a-time name hashing.

This patch selects HAVE_EFFICIENT_UNALIGNED_ACCESS for these cores and
reworks the kconfig select logic for DCACHE_WORD_ACCESS to use the new
symbol.

Signed-off-by: Will Deacon <will.deacon@arm.com>
---
 arch/arm/Kconfig | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 38ccfd0adbf9..f8c64af7c768 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -11,7 +11,7 @@ config ARM
 	select BUILDTIME_EXTABLE_SORT if MMU
 	select CLONE_BACKWARDS
 	select CPU_PM if (SUSPEND || CPU_IDLE)
-	select DCACHE_WORD_ACCESS if (CPU_V6 || CPU_V6K || CPU_V7) && MMU
+	select DCACHE_WORD_ACCESS if HAVE_EFFICIENT_UNALIGNED_ACCESS
 	select GENERIC_ATOMIC64 if (CPU_V7M || CPU_V6 || !CPU_32v6K || !AEABI)
 	select GENERIC_CLOCKEVENTS_BROADCAST if SMP
 	select GENERIC_IDLE_POLL_SETUP
@@ -35,6 +35,7 @@ config ARM
 	select HAVE_DMA_ATTRS
 	select HAVE_DMA_CONTIGUOUS if MMU
 	select HAVE_DYNAMIC_FTRACE if (!XIP_KERNEL)
+	select HAVE_EFFICIENT_UNALIGNED_ACCESS if (CPU_V6 || CPU_V6K || CPU_V7) && MMU
 	select HAVE_FTRACE_MCOUNT_RECORD if (!XIP_KERNEL)
 	select HAVE_FUNCTION_GRAPH_TRACER if (!THUMB2_KERNEL)
 	select HAVE_FUNCTION_TRACER if (!XIP_KERNEL)
-- 
1.8.2.2

[PATCH 1/2] ARM: dcache: select DCACHE_WORD_ACCESS for big-endian CPUs

[Nicolas Pitre]

On Tue, 17 Dec 2013, Will Deacon wrote:

> With commit 11ec50caedb5 ("word-at-a-time: provide generic big-endian
> zero_bytemask implementation"), the asm-generic word-at-a-time code now
> provides a zero_bytemask implementation, allowing us to make use of
> DCACHE_WORD_ACCESS on big-endian CPUs, providing our
> load_unaligned_zeropad function is endianness-clean.
> 
> This patch reworks the load_unaligned_zeropad fixup code to work for
> both big- and little-endian CPUs, then removes the !CPU_BIG_ENDIAN check
> when selecting DCACHE_WORD_ACCESS.
> 
> Signed-off-by: Will Deacon <will.deacon@arm.com>

Small comment below.

> ---
>  arch/arm/Kconfig                      |  2 +-
>  arch/arm/include/asm/word-at-a-time.h | 18 ++++++++++--------
>  2 files changed, 11 insertions(+), 9 deletions(-)
> 
> diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
> index 88f726d36a40..38ccfd0adbf9 100644
> --- a/arch/arm/Kconfig
> +++ b/arch/arm/Kconfig
> @@ -11,7 +11,7 @@ config ARM
>  	select BUILDTIME_EXTABLE_SORT if MMU
>  	select CLONE_BACKWARDS
>  	select CPU_PM if (SUSPEND || CPU_IDLE)
> -	select DCACHE_WORD_ACCESS if (CPU_V6 || CPU_V6K || CPU_V7) && !CPU_BIG_ENDIAN && MMU
> +	select DCACHE_WORD_ACCESS if (CPU_V6 || CPU_V6K || CPU_V7) && MMU
>  	select GENERIC_ATOMIC64 if (CPU_V7M || CPU_V6 || !CPU_32v6K || !AEABI)
>  	select GENERIC_CLOCKEVENTS_BROADCAST if SMP
>  	select GENERIC_IDLE_POLL_SETUP
> diff --git a/arch/arm/include/asm/word-at-a-time.h b/arch/arm/include/asm/word-at-a-time.h
> index 4d52f92967a6..a6d0a29861e7 100644
> --- a/arch/arm/include/asm/word-at-a-time.h
> +++ b/arch/arm/include/asm/word-at-a-time.h
> @@ -48,10 +48,14 @@ static inline unsigned long find_zero(unsigned long mask)
>  	return ret;
>  }
>  
> -#ifdef CONFIG_DCACHE_WORD_ACCESS
> -
>  #define zero_bytemask(mask) (mask)
>  
> +#else	/* __ARMEB__ */
> +#include <asm-generic/word-at-a-time.h>
> +#endif
> +
> +#ifdef CONFIG_DCACHE_WORD_ACCESS
> +
>  /*
>   * Load an unaligned word from kernel space.
>   *
> @@ -73,7 +77,11 @@ static inline unsigned long load_unaligned_zeropad(const void *addr)
>  	"	bic	%2, %2, #0x3\n"
>  	"	ldr	%0, [%2]\n"
>  	"	lsl	%1, %1, #0x3\n"
> +#ifndef __ARMEB__
>  	"	lsr	%0, %0, %1\n"
> +#else
> +	"	lsl	%0, %0, %1\n"
> +#endif

You could get rid of the #ifdef by using:

		pull	%0, %0, %1

Except this is not a pure assembly file and can't rely on 
<asm/assembler.h>.  Bah.  In that case...

Acked-by: Nicolas Pitre <nico@linaro.org>


Nicolas

[PATCH 2/2] ARM: kconfig: select HAVE_EFFICIENT_UNALIGNED_ACCESS for CPUv6+ && MMU

[Nicolas Pitre]

On Tue, 17 Dec 2013, Will Deacon wrote:

> Modern ARM CPUs can perform efficient unaligned memory accesses in
> hardware and this feature is relied up on by code such as the dcache
> word-at-a-time name hashing.
> 
> This patch selects HAVE_EFFICIENT_UNALIGNED_ACCESS for these cores and
> reworks the kconfig select logic for DCACHE_WORD_ACCESS to use the new
> symbol.
> 
> Signed-off-by: Will Deacon <will.deacon@arm.com>

This symbol could be used in some more places too.  For example this 
would make the code a bit clearer:

diff --git a/arch/arm/include/asm/uaccess.h b/arch/arm/include/asm/uaccess.h
index 72abdc541f..12c3a5decc 100644
--- a/arch/arm/include/asm/uaccess.h
+++ b/arch/arm/include/asm/uaccess.h
@@ -19,7 +19,7 @@
 #include <asm/unified.h>
 #include <asm/compiler.h>
 
-#if __LINUX_ARM_ARCH__ < 6
+#ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
 #include <asm-generic/uaccess-unaligned.h>
 #else
 #define __get_user_unaligned __get_user


Nicolas

[PATCH 2/2] ARM: kconfig: select HAVE_EFFICIENT_UNALIGNED_ACCESS for CPUv6+ && MMU

[Will Deacon]

Hi Nicolas,

On Tue, Dec 17, 2013 at 08:34:18PM +0000, Nicolas Pitre wrote:
> On Tue, 17 Dec 2013, Will Deacon wrote:
> 
> > Modern ARM CPUs can perform efficient unaligned memory accesses in
> > hardware and this feature is relied up on by code such as the dcache
> > word-at-a-time name hashing.
> > 
> > This patch selects HAVE_EFFICIENT_UNALIGNED_ACCESS for these cores and
> > reworks the kconfig select logic for DCACHE_WORD_ACCESS to use the new
> > symbol.
> > 
> > Signed-off-by: Will Deacon <will.deacon@arm.com>
> 
> This symbol could be used in some more places too.  For example this 
> would make the code a bit clearer:

Well spotted.

> diff --git a/arch/arm/include/asm/uaccess.h b/arch/arm/include/asm/uaccess.h
> index 72abdc541f..12c3a5decc 100644
> --- a/arch/arm/include/asm/uaccess.h
> +++ b/arch/arm/include/asm/uaccess.h
> @@ -19,7 +19,7 @@
>  #include <asm/unified.h>
>  #include <asm/compiler.h>
>  
> -#if __LINUX_ARM_ARCH__ < 6
> +#ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
>  #include <asm-generic/uaccess-unaligned.h>
>  #else
>  #define __get_user_unaligned __get_user

The original patch is now queued, so we can add this hunk as a follow-up
cleanup patch. I've got you as the author, but can I add your S-o-B please?

Cheers,

Will

[PATCH 2/2] ARM: kconfig: select HAVE_EFFICIENT_UNALIGNED_ACCESS for CPUv6+ && MMU

[Nicolas Pitre]

On Mon, 6 Jan 2014, Will Deacon wrote:

> Hi Nicolas,
> 
> On Tue, Dec 17, 2013 at 08:34:18PM +0000, Nicolas Pitre wrote:
> > On Tue, 17 Dec 2013, Will Deacon wrote:
> > 
> > > Modern ARM CPUs can perform efficient unaligned memory accesses in
> > > hardware and this feature is relied up on by code such as the dcache
> > > word-at-a-time name hashing.
> > > 
> > > This patch selects HAVE_EFFICIENT_UNALIGNED_ACCESS for these cores and
> > > reworks the kconfig select logic for DCACHE_WORD_ACCESS to use the new
> > > symbol.
> > > 
> > > Signed-off-by: Will Deacon <will.deacon@arm.com>
> > 
> > This symbol could be used in some more places too.  For example this 
> > would make the code a bit clearer:
> 
> Well spotted.
> 
> > diff --git a/arch/arm/include/asm/uaccess.h b/arch/arm/include/asm/uaccess.h
> > index 72abdc541f..12c3a5decc 100644
> > --- a/arch/arm/include/asm/uaccess.h
> > +++ b/arch/arm/include/asm/uaccess.h
> > @@ -19,7 +19,7 @@
> >  #include <asm/unified.h>
> >  #include <asm/compiler.h>
> >  
> > -#if __LINUX_ARM_ARCH__ < 6
> > +#ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
> >  #include <asm-generic/uaccess-unaligned.h>
> >  #else
> >  #define __get_user_unaligned __get_user
> 
> The original patch is now queued, so we can add this hunk as a follow-up
> cleanup patch. I've got you as the author, but can I add your S-o-B please?

Signed-off-by: Nicolas Pitre <nico@linaro.org>



> 
> Cheers,
> 
> Will
> 

[PATCH 1/2] ARM: dcache: select DCACHE_WORD_ACCESS for big-endian CPUs

[Ben Dooks]

On 17/12/13 12:16, Will Deacon wrote:
> With commit 11ec50caedb5 ("word-at-a-time: provide generic big-endian
> zero_bytemask implementation"), the asm-generic word-at-a-time code now
> provides a zero_bytemask implementation, allowing us to make use of
> DCACHE_WORD_ACCESS on big-endian CPUs, providing our
> load_unaligned_zeropad function is endianness-clean.
>
> This patch reworks the load_unaligned_zeropad fixup code to work for
> both big- and little-endian CPUs, then removes the !CPU_BIG_ENDIAN check
> when selecting DCACHE_WORD_ACCESS.


> @@ -73,7 +77,11 @@ static inline unsigned long load_unaligned_zeropad(const void *addr)
>   	"	bic	%2, %2, #0x3\n"
>   	"	ldr	%0, [%2]\n"
>   	"	lsl	%1, %1, #0x3\n"
> +#ifndef __ARMEB__
>   	"	lsr	%0, %0, %1\n"
> +#else
> +	"	lsl	%0, %0, %1\n"
> +#endif

IIRC we have ARM_BE() and ARM_LE() helpers for this


-- 
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius