Skip to contentSkip to conversation

ARM big-endian on current kernels for linux-3.8

PATCH New Ben Dooks 9 participants 77 messages Feb 8, 2013 – Feb 15, 2013

Subsystems (6)
lore
Share & tools
lore

Ben Dooks

author
#0 lore raw

I have been working on getting big-endian kernels working, mainly from little-endian boot envrionments. The following patch series is what I have been working on, mainly on the highbank and axp systems.

Feedback and testing is welcome.

I have published these on git.baserock.org

git://git.baserock.org/delta/linux baserock/arm/bjdooks/be8_v1
git://git.baserock.org/delta/linux baserock/arm/bjdooks/linux37-be8-1

The second is a backport of the v3.8-rc6 work due to problems with booting on the highbank which Rob is resolving at the moment.

author
#1 lore raw

PATCH 1/17 ARM: add CPU_BE8_BOOT_LE configuration

Add configuration for booting a BE8 system from a LE boot environment.

---
 arch/arm/mm/Kconfig |    7 +++++++
 1 file changed, 7 insertions(+)
arch/arm/mm/Kconfig +7 -0
index 3fd629d..70e2ad9 100644
@@ -688,6 +688,13 @@ config CPU_ENDIAN_BE8
 	help
 	  Support for the BE-8 (big-endian) mode on ARMv6 and ARMv7 processors.
 
+config CPU_BE8_BOOT_LE
+	bool "Boot BE system from LE loader"
+	depends on CPU_ENDIAN_BE8
+	help
+	  Support for running BE-8 (big-endian) from a boot environment that is
+	  running in LE mode.
+
 config CPU_ENDIAN_BE32
 	bool
 	depends on CPU_BIG_ENDIAN
Signature · 1 line
1.7.10.4
#2 lore raw

PATCH 1/17 ARM: add CPU_BE8_BOOT_LE configuration

Quoted 12 lines · Ben Dooks

Add configuration for booting a BE8 system from a LE boot environment.

---
 arch/arm/mm/Kconfig |    7 +++++++
 1 file changed, 7 insertions(+)
arch/arm/mm/Kconfig +0 -0
index 3fd629d..70e2ad9 100644

arch/arm/mm/Kconfig config CPU_ENDIAN_BE8 · line 697

Quoted 10 lines
arch/arm/mm/Kconfig +7 -0
@@ -688,6 +688,13 @@ config CPU_ENDIAN_BE8
 	help
 	  Support for the BE-8 (big-endian) mode on ARMv6 and ARMv7 processors.
 
+config CPU_BE8_BOOT_LE
+	bool "Boot BE system from LE loader"
+	depends on CPU_ENDIAN_BE8
+	help
+	  Support for running BE-8 (big-endian) from a boot environment that is
+	  running in LE mode.
+

This looks strange to have a patch that only adds a config symbol with no user. Could you move it later in the series, or fold it with, say, patch #4?

Quoted 12 lines
arch/arm/mm/Kconfig +0 -0
 config CPU_ENDIAN_BE32
 	bool
 	depends on CPU_BIG_ENDIAN
Signature · 7 lines
1.7.10.4


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
author
#3 lore raw

PATCH 1/17 ARM: add CPU_BE8_BOOT_LE configuration

Quoted 19 lines · Nicolas Pitre
Quoted 17 lines · Ben Dooks

Add configuration for booting a BE8 system from a LE boot environment.

---
 arch/arm/mm/Kconfig |    7 +++++++
 1 file changed, 7 insertions(+)
arch/arm/mm/Kconfig +0 -0
index 3fd629d..70e2ad9 100644
@@ -688,6 +688,13 @@ config CPU_ENDIAN_BE8
 	help
 	  Support for the BE-8 (big-endian) mode on ARMv6 and ARMv7 

processors.

Quoted 14 lines

arch/arm/mm/Kconfig config CPU_ENDIAN_BE8 · line 697

Quoted 8 lines
arch/arm/mm/Kconfig +5 -0
+config CPU_BE8_BOOT_LE
+	bool "Boot BE system from LE loader"
+	depends on CPU_ENDIAN_BE8
+	help
+	  Support for running BE-8 (big-endian) from a boot environment 

that is

arch/arm/mm/Kconfig +2 -0
+	  running in LE mode.
+

This looks strange to have a patch that only adds a config symbol with
no user. Could you move it later in the series, or fold it with, say,
patch #4?

Yes, will look at folding it in.

Signature · 1 line
Ben
author
#4 lore raw

PATCH 2/17 ARM: set BE8 if LE in head code

In reply to Ben Dooks

If we are booting in LE and compiled for BE8, then change the processor state in the head code. Since the instruction stream is always LE, we do not need to do anything special.

Also ensure that the secondary processors are started in the same mode.

---
 arch/arm/boot/compressed/head.S |    3 +++
 arch/arm/kernel/head.S          |    3 +++
 2 files changed, 6 insertions(+)
arch/arm/boot/compressed/head.S +3 -0
index fe4d9c3..39940a7 100644
@@ -134,6 +134,9 @@ start:
 		.word	_edata			@ zImage end address
  THUMB(		.thumb			)
 1:
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+		setend	be			@ go BE-8 if we booted LE / no-op if already BE-8
+#endif
 		mrs	r9, cpsr
 #ifdef CONFIG_ARM_VIRT_EXT
 		bl	__hyp_stub_install	@ get into SVC mode, reversibly
arch/arm/kernel/head.S +3 -0
index 486a15a..f1ab279 100644
@@ -333,6 +333,9 @@ ENTRY(secondary_startup)
 	 * the processor type - there is no need to check the machine type
 	 * as it has already been validated by the primary processor.
 	 */
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+	setend	be				@ if system starts LE, go to BE-8
+#endif
 #ifdef CONFIG_ARM_VIRT_EXT
 	bl	__hyp_stub_install_secondary
 #endif
Signature · 1 line
1.7.10.4
#5 lore raw

PATCH 2/17 ARM: set BE8 if LE in head code

Quoted 31 lines · Ben Dooks

If we are booting in LE and compiled for BE8, then change the processor state in the head code. Since the instruction stream is always LE, we do not need to do anything special.

Also ensure that the secondary processors are started in the same mode.

---
 arch/arm/boot/compressed/head.S |    3 +++
 arch/arm/kernel/head.S          |    3 +++
 2 files changed, 6 insertions(+)
arch/arm/boot/compressed/head.S +3 -0
index fe4d9c3..39940a7 100644
@@ -134,6 +134,9 @@ start:
 		.word	_edata			@ zImage end address
  THUMB(		.thumb			)
 1:
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+		setend	be			@ go BE-8 if we booted LE / no-op if already BE-8
+#endif
 		mrs	r9, cpsr
 #ifdef CONFIG_ARM_VIRT_EXT
 		bl	__hyp_stub_install	@ get into SVC mode, reversibly
arch/arm/kernel/head.S +0 -0
index 486a15a..f1ab279 100644

arch/arm/kernel/head.S ENTRY(secondary_startup) · line 341

Quoted 9 lines
arch/arm/kernel/head.S +3 -0
@@ -333,6 +333,9 @@ ENTRY(secondary_startup)
 	 * the processor type - there is no need to check the machine type
 	 * as it has already been validated by the primary processor.
 	 */
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+	setend	be				@ if system starts LE, go to BE-8
+#endif
 #ifdef CONFIG_ARM_VIRT_EXT
 	bl	__hyp_stub_install_secondary
 #endif

For consistency, you should add a setend on the primary CPU entry path in kernel/head.S and not rely on the decompressor. Some people don't use the compressed image.

Nicolas

author
#6 lore raw

PATCH 2/17 ARM: set BE8 if LE in head code

Quoted 43 lines · Nicolas Pitre
Quoted 41 lines · Ben Dooks

If we are booting in LE and compiled for BE8, then change the processor
state in the head code. Since the instruction stream is always LE, we
do not need to do anything special.

Also ensure that the secondary processors are started in the same mode.

---
 arch/arm/boot/compressed/head.S |    3 +++
 arch/arm/kernel/head.S          |    3 +++
 2 files changed, 6 insertions(+)
arch/arm/boot/compressed/head.S +3 -0
index fe4d9c3..39940a7 100644
@@ -134,6 +134,9 @@ start:
 		.word	_edata			@ zImage end address
  THUMB(		.thumb			)
 1:
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+		setend	be			@ go BE-8 if we booted LE / no-op if already BE-8
+#endif
 		mrs	r9, cpsr
 #ifdef CONFIG_ARM_VIRT_EXT
 		bl	__hyp_stub_install	@ get into SVC mode, reversibly
arch/arm/kernel/head.S +3 -0
index 486a15a..f1ab279 100644
@@ -333,6 +333,9 @@ ENTRY(secondary_startup)
 	 * the processor type - there is no need to check the machine type
 	 * as it has already been validated by the primary processor.
 	 */
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+	setend	be				@ if system starts LE, go to BE-8
+#endif
Quoted 8 lines

arch/arm/kernel/head.S ENTRY(secondary_startup) · line 341

Quoted 3 lines
arch/arm/kernel/head.S +0 -0
 #ifdef CONFIG_ARM_VIRT_EXT
 	bl	__hyp_stub_install_secondary
 #endif

For consistency, you should add a setend on the primary CPU entry path
in kernel/head.S and not rely on the decompressor. Some people don't use the compressed image.

Thank you for spotting this, will sort it out for the next round.

Signature · 1 line
Ben
#7 lore raw

PATCH 2/17 ARM: set BE8 if LE in head code

In reply to Ben Dooks
Quoted 17 lines · Ben Dooks

If we are booting in LE and compiled for BE8, then change the processor state in the head code. Since the instruction stream is always LE, we do not need to do anything special.

Also ensure that the secondary processors are started in the same mode.

---
 arch/arm/boot/compressed/head.S |    3 +++
 arch/arm/kernel/head.S          |    3 +++
 2 files changed, 6 insertions(+)
arch/arm/boot/compressed/head.S +0 -0
index fe4d9c3..39940a7 100644

arch/arm/boot/compressed/head.S start: · line 137

Quoted 4 lines
arch/arm/boot/compressed/head.S +1 -0
@@ -134,6 +134,9 @@ start:
 		.word	_edata			@ zImage end address
  THUMB(		.thumb			)
 1:
+#ifdef CONFIG_CPU_BE8_BOOT_LE

Can't you use CONFIG_CPU_BIG_ENDIAN here rather than a new config option?

Quoted 20 lines
arch/arm/boot/compressed/head.S +2 -0
+		setend	be			@ go BE-8 if we booted LE / no-op if already BE-8
+#endif
 		mrs	r9, cpsr
 #ifdef CONFIG_ARM_VIRT_EXT
 		bl	__hyp_stub_install	@ get into SVC mode, reversibly
arch/arm/kernel/head.S +3 -0
index 486a15a..f1ab279 100644
@@ -333,6 +333,9 @@ ENTRY(secondary_startup)
 	 * the processor type - there is no need to check the machine type
 	 * as it has already been validated by the primary processor.
 	 */
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+	setend	be				@ if system starts LE, go to BE-8
+#endif
 #ifdef CONFIG_ARM_VIRT_EXT
 	bl	__hyp_stub_install_secondary
 #endif
author
#8 lore raw

PATCH 2/17 ARM: set BE8 if LE in head code

Quoted 16 lines · Rob Herring
Quoted 15 lines · Ben Dooks

If we are booting in LE and compiled for BE8, then change the processor
state in the head code. Since the instruction stream is always LE, we
do not need to do anything special.

Also ensure that the secondary processors are started in the same mode.

---
 arch/arm/boot/compressed/head.S |    3 +++
 arch/arm/kernel/head.S          |    3 +++
 2 files changed, 6 insertions(+)
Quoted 13 lines

arch/arm/boot/compressed/head.S start: · line 137

Quoted 10 lines
arch/arm/boot/compressed/head.S +1 -0
index fe4d9c3..39940a7 100644
@@ -134,6 +134,9 @@ start:
 		.word	_edata			@ zImage end address
  THUMB(		.thumb			)
 1:
+#ifdef CONFIG_CPU_BE8_BOOT_LE

Can't you use CONFIG_CPU_BIG_ENDIAN here rather than a new config option?

For this, we can probably do this as it is a no-op, however other parts of the patch series do need this to allow it to boot from a little endian
booting environment.

Signature · 1 line
Ben
author
#9 lore raw

PATCH 2/17 ARM: set BE8 if LE in head code

In reply to Rob Herring
Quoted 13 lines · Rob Herring
Quoted 12 lines · Ben Dooks

If we are booting in LE and compiled for BE8, then change the processor state in the head code. Since the instruction stream is always LE, we do not need to do anything special.

Also ensure that the secondary processors are started in the same mode.

---
  arch/arm/boot/compressed/head.S |    3 +++
  arch/arm/kernel/head.S          |    3 +++
  2 files changed, 6 insertions(+)
Quoted 11 lines

arch/arm/boot/compressed/head.S start: · line 137

Quoted 9 lines
arch/arm/boot/compressed/head.S +1 -0
index fe4d9c3..39940a7 100644
@@ -134,6 +134,9 @@ start:
  		.word	_edata			@ zImage end address
   THUMB(		.thumb			)
  1:
+#ifdef CONFIG_CPU_BE8_BOOT_LE

Can't you use CONFIG_CPU_BIG_ENDIAN here rather than a new config option?

Actually, no as not all big endian CPUs will be BE8 (which should guarnatee us the 'setend be' to be available) as we could still build for BE32 systems where we would not want this.

I was considering doing the following:

ARM_BE8( setend be )

What do people think? It is a waste of 16 bytes if our system is already in BE8 mode, but is going to look cleaner than

ARM_BE8_LEBOOT( setend be )

Any thoughts?

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
author
#10 lore raw

PATCH 3/17 ARM: fix ARCH_IXP4xx usage of ARCH_SUPPORTS_BIG_ENDIAN

In reply to Ben Dooks

The Kconfig for arch/arm/mach-ixp4xx has a copy of the definition of ARCH_SUPPORTS_BIG_ENDIAN which defaults to yes. This means that if IXP4xx is selected and this symbol is selected elswhere then an warning is produced.

Clean the following error up by making the symbol be selected by the main ARCH_IXP4XX definition.

warning: (ARCH_xxx) selects ARCH_SUPPORTS_BIG_ENDIAN which has unmet direct dependencies (ARCH_IXP4XX)
warning: (ARCH_xxx) selects ARCH_SUPPORTS_BIG_ENDIAN which has unmet direct dependencies (ARCH_IXP4XX)

---
 arch/arm/Kconfig             |    1 +
 arch/arm/mach-ixp4xx/Kconfig |    4 ----
 2 files changed, 1 insertion(+), 4 deletions(-)
arch/arm/Kconfig +1 -0
index 67874b8..d3f6c4f 100644
@@ -517,6 +517,7 @@ config ARCH_IXP4XX
 	bool "IXP4xx-based"
 	depends on MMU
 	select ARCH_HAS_DMA_SET_COHERENT_MASK
+	select ARCH_SUPPORTS_BIG_ENDIAN
 	select ARCH_REQUIRE_GPIOLIB
 	select CLKSRC_MMIO
 	select CPU_XSCALE
arch/arm/mach-ixp4xx/Kconfig +0 -4
index 73a2d90..72de05f 100644
@@ -1,9 +1,5 @@
 if ARCH_IXP4XX
 
-config ARCH_SUPPORTS_BIG_ENDIAN
-	bool
-	default y
-
 menu "Intel IXP4xx Implementation Options"
 
 comment "IXP4xx Platforms"
Signature · 1 line
1.7.10.4
author
#11 lore raw

PATCH 4/17 ARM: Add ARCH_SUPPORTS_BIG_ENDIAN

In reply to Ben Dooks

Add ARCH_SUPPORTS_BIG_ENDIAN define to arch/arm/Kconfig. Since it used to live in arch/arm/mach-ixp4xx/Kconfig we move it up to the ARM one as the next best place for it.

It may be that it could be moved to a more generic location, with the other ARCH_SUPPORT functions.

---
 arch/arm/Kconfig |    6 ++++++
 1 file changed, 6 insertions(+)
arch/arm/Kconfig +6 -0
index d3f6c4f..8dfa634 100644
@@ -152,6 +152,12 @@ config ARCH_HAS_CPUFREQ
 	  and that the relevant menu configurations are displayed for
 	  it.
 
+config ARCH_SUPPORTS_BIG_ENDIAN
+	bool
+	help
+	  Internal node to specify the architecture can run in Big Endian
+	  mode.
+
 config GENERIC_HWEIGHT
 	bool
 	default y
Signature · 1 line
1.7.10.4
#12 lore raw

PATCH 4/17 ARM: Add ARCH_SUPPORTS_BIG_ENDIAN

Quoted 8 lines · Ben Dooks

Add ARCH_SUPPORTS_BIG_ENDIAN define to arch/arm/Kconfig. Since it used to live in arch/arm/mach-ixp4xx/Kconfig we move it up to the ARM one as the next best place for it.

It may be that it could be moved to a more generic location, with the other ARCH_SUPPORT functions.

So in patch 3, you remove this option (which removes it from the generated configs) and add it back in in the following patch. This makes this series non-bisectable. Please roll this into patch 3 and update the patch description accordingly.

I'd also suggest moving it to arch/arm/mm/Kconfig rather than arch/arm/Kconfig, just before its only user, CPU_BIG_ENDIAN.

author
#13 lore raw

PATCH 4/17 ARM: Add ARCH_SUPPORTS_BIG_ENDIAN

Quoted 19 lines · Russell King - ARM Linux
Quoted 10 lines · Ben Dooks

Add ARCH_SUPPORTS_BIG_ENDIAN define to arch/arm/Kconfig. Since it used
to live in arch/arm/mach-ixp4xx/Kconfig we move it up to the ARM one as the next best place for it.

It may be that it could be moved to a more generic location, with the
other ARCH_SUPPORT functions.

So in patch 3, you remove this option (which removes it from the generated
configs) and add it back in in the following patch. This makes this series
non-bisectable. Please roll this into patch 3 and update the patch description accordingly.

Quoted 2 lines

I'd also suggest moving it to arch/arm/mm/Kconfig rather than arch/arm/Kconfig, just before its only user, CPU_BIG_ENDIAN.

I was wondering if it should go elsewhere for other archs to use, but arch/arm/mmk/Kconfig seems reasonable. We always move it later.

Signature · 1 line
Ben
author
#14 lore raw

PATCH 5/17 ARM: fixup_pv_table bug when CPU_ENDIAN_BE8

In reply to Ben Dooks

The fixup_pv_table assumes that the instructions are in the same endian configuration as the data, but when the CPU is running in BE8 the instructions stay in little-endian format.

Make sure if CONFIG_CPU_ENDIAN_BE8 is set that we do all the alterations to the instructions taking in to account the LDR/STR will be swapping the data endian-ness.

Since the code is only modifying a byte, we avoid dual-swapping the data, and just change the bits we clear and ORR in.

Note, not tested against an actual bug, it looked wrong.

---
 arch/arm/kernel/head.S |    6 ++++++
 1 file changed, 6 insertions(+)
arch/arm/kernel/head.S +6 -0
index f1ab279..aafb151 100644
@@ -579,8 +579,14 @@ __fixup_a_pv_table:
 #else
 	b	2f
 1:	ldr	ip, [r7, r3]
+#ifdef CONFIG_CPU_ENDIAN_BE8
+	@ in BE8, we load data in BE, but instructions still in LE
+	bic	ip, ip, #0xff000000
+	orr	ip, ip, r6, lsl#24
+#else
 	bic	ip, ip, #0x000000ff
 	orr	ip, ip, r6	@ mask in offset bits 31-24
+#endif
 	str	ip, [r7, r3]
 2:	cmp	r4, r5
 	ldrcc	r7, [r4], #4	@ use branch for delay slot
Signature · 1 line
1.7.10.4
#15 lore raw

PATCH 5/17 ARM: fixup_pv_table bug when CPU_ENDIAN_BE8

Quoted 4 lines · Ben Dooks

The fixup_pv_table assumes that the instructions are in the same endian configuration as the data, but when the CPU is running in BE8 the instructions stay in little-endian format.

Quoted 8 lines

Make sure if CONFIG_CPU_ENDIAN_BE8 is set that we do all the alterations to the instructions taking in to account the LDR/STR will be swapping the data endian-ness.

Since the code is only modifying a byte, we avoid dual-swapping the data, and just change the bits we clear and ORR in.

Note, not tested against an actual bug, it looked wrong.

You are missing the CONFIG_THUMB2_KERNEL counterpart.

Quoted 34 lines
---
 arch/arm/kernel/head.S |    6 ++++++
 1 file changed, 6 insertions(+)
arch/arm/kernel/head.S +6 -0
index f1ab279..aafb151 100644
@@ -579,8 +579,14 @@ __fixup_a_pv_table:
 #else
 	b	2f
 1:	ldr	ip, [r7, r3]
+#ifdef CONFIG_CPU_ENDIAN_BE8
+	@ in BE8, we load data in BE, but instructions still in LE
+	bic	ip, ip, #0xff000000
+	orr	ip, ip, r6, lsl#24
+#else
 	bic	ip, ip, #0x000000ff
 	orr	ip, ip, r6	@ mask in offset bits 31-24
+#endif
 	str	ip, [r7, r3]
 2:	cmp	r4, r5
 	ldrcc	r7, [r4], #4	@ use branch for delay slot
Signature · 7 lines
1.7.10.4


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
author
#16 lore raw

PATCH 5/17 ARM: fixup_pv_table bug when CPU_ENDIAN_BE8

Quoted 10 lines · Nicolas Pitre
Quoted 8 lines · Ben Dooks

The fixup_pv_table assumes that the instructions are in the same endian configuration as the data, but when the CPU is running in BE8 the instructions stay in little-endian format.

Make sure if CONFIG_CPU_ENDIAN_BE8 is set that we do all the alterations to the instructions taking in to account the LDR/STR will be swapping the data endian-ness.

Quoted 6 lines
Quoted 4 lines

Since the code is only modifying a byte, we avoid dual-swapping the data, and just change the bits we clear and ORR in.

Note, not tested against an actual bug, it looked wrong.

You are missing the CONFIG_THUMB2_KERNEL counterpart.

Sorry, I will have a look into that, although I may need some help on that as I don't have any experience with CONFIG_THUMB2_KERNEL or a userland to test with.

Quoted 38 lines
Quoted 28 lines
---
  arch/arm/kernel/head.S |    6 ++++++
  1 file changed, 6 insertions(+)
arch/arm/kernel/head.S +6 -1
index f1ab279..aafb151 100644
@@ -579,8 +579,14 @@ __fixup_a_pv_table:
  #else
  	b	2f
  1:	ldr	ip, [r7, r3]
+#ifdef CONFIG_CPU_ENDIAN_BE8
+	@ in BE8, we load data in BE, but instructions still in LE
+	bic	ip, ip, #0xff000000
+	orr	ip, ip, r6, lsl#24
+#else
  	bic	ip, ip, #0x000000ff
  	orr	ip, ip, r6	@ mask in offset bits 31-24
+#endif
  	str	ip, [r7, r3]
  2:	cmp	r4, r5
  	ldrcc	r7, [r4], #4	@ use branch for delay slot
--

1.7.10.4

Quoted 5 lines

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
#17 lore raw

PATCH 5/17 ARM: fixup_pv_table bug when CPU_ENDIAN_BE8

Quoted 14 lines · Ben Dooks
Quoted 13 lines · Nicolas Pitre
Quoted 11 lines · Ben Dooks

The fixup_pv_table assumes that the instructions are in the same endian configuration as the data, but when the CPU is running in BE8 the instructions stay in little-endian format.

Make sure if CONFIG_CPU_ENDIAN_BE8 is set that we do all the alterations to the instructions taking in to account the LDR/STR will be swapping the data endian-ness.

Since the code is only modifying a byte, we avoid dual-swapping the data, and just change the bits we clear and ORR in.

Quoted 7 lines
Quoted 3 lines
Quoted 1 line

Note, not tested against an actual bug, it looked wrong.

You are missing the CONFIG_THUMB2_KERNEL counterpart.

Sorry, I will have a look into that, although I may need some help on that as I don't have any experience with CONFIG_THUMB2_KERNEL or a userland to test with.

You don't need to change your userland. The kernel may be compiled using the Thumb2 instruction set even if userland is traditional ARM mode.

Nicolas

author
#18 lore raw

PATCH 6/17 ARM: fixup head for atag verification

In reply to Ben Dooks

If we booted LE but running BE8, ensure we read ATAGs data from head code in the correct mode.

---
 arch/arm/kernel/head-common.S |    8 ++++++++
 1 file changed, 8 insertions(+)
arch/arm/kernel/head-common.S +8 -0
index 854bd22..4eb53b2 100644
@@ -21,6 +21,12 @@
 #define OF_DT_MAGIC 0xedfe0dd0 /* 0xd00dfeed in big-endian */
 #endif
 
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+#define BE8_LE(x...)	x
+#else
+#define BE8_LE(x...)
+#endif
+
 /*
  * Exception handling.  Something went wrong and we can't proceed.  We
  * ought to tell the user, but since we don't have any guarantee that
@@ -53,10 +59,12 @@ __vet_atags:
 	cmp	r5, r6
 	beq	2f
 #endif
+BE8_LE(	rev	r5, r5 )
 	cmp	r5, #ATAG_CORE_SIZE		@ is first tag ATAG_CORE?
 	cmpne	r5, #ATAG_CORE_SIZE_EMPTY
 	bne	1f
 	ldr	r5, [r2, #4]
+BE8_LE(	rev	r5, r5 )
 	ldr	r6, =ATAG_CORE
 	cmp	r5, r6
 	bne	1f
Signature · 1 line
1.7.10.4
#19 lore raw

PATCH 6/17 ARM: fixup head for atag verification

Quoted 4 lines · Ben Dooks

If we booted LE but running BE8, ensure we read ATAGs data from head code in the correct mode.

You must be working on some recent platform, right? Your cover letter mention highbank. Strange that you don't address the DT pointer validation as well.

Quoted 43 lines
---
 arch/arm/kernel/head-common.S |    8 ++++++++
 1 file changed, 8 insertions(+)
arch/arm/kernel/head-common.S +8 -0
index 854bd22..4eb53b2 100644
@@ -21,6 +21,12 @@
 #define OF_DT_MAGIC 0xedfe0dd0 /* 0xd00dfeed in big-endian */
 #endif
 
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+#define BE8_LE(x...)	x
+#else
+#define BE8_LE(x...)
+#endif
+
 /*
  * Exception handling.  Something went wrong and we can't proceed.  We
  * ought to tell the user, but since we don't have any guarantee that
@@ -53,10 +59,12 @@ __vet_atags:
 	cmp	r5, r6
 	beq	2f
 #endif
+BE8_LE(	rev	r5, r5 )
 	cmp	r5, #ATAG_CORE_SIZE		@ is first tag ATAG_CORE?
 	cmpne	r5, #ATAG_CORE_SIZE_EMPTY
 	bne	1f
 	ldr	r5, [r2, #4]
+BE8_LE(	rev	r5, r5 )
 	ldr	r6, =ATAG_CORE
 	cmp	r5, r6
 	bne	1f
Signature · 7 lines
1.7.10.4


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
author
#20 lore raw

PATCH 6/17 ARM: fixup head for atag verification

Quoted 6 lines · Nicolas Pitre
Quoted 4 lines · Ben Dooks

If we booted LE but running BE8, ensure we read ATAGs data from head code
in the correct mode.

Quoted 6 lines
Quoted 1 line

You must be working on some recent platform, right? Your cover letter
mention highbank. Strange that you don't address the DT pointer validation as well.

I am using highbank for my main tests, I believe the DT pointer is correct as the data is always guaranteed to be BE and the kernel is doing the correct thing to test for it.

I will re-check this, but it all looked fine when I went through it.

Signature · 1 line
Ben
author
#21 lore raw

PATCH 7/17 ARM: twd: data endian fix

In reply to Ben Dooks

Ensure the twd driver uses the correct calls to access the hardware to ensure that we do not end up with data in the wrong endian format.

---
 arch/arm/kernel/smp_twd.c |   24 ++++++++++++------------
 1 file changed, 12 insertions(+), 12 deletions(-)
arch/arm/kernel/smp_twd.c +12 -12
index 49f335d..3756f42 100644
@@ -46,7 +46,7 @@ static void twd_set_mode(enum clock_event_mode mode,
 	case CLOCK_EVT_MODE_PERIODIC:
 		ctrl = TWD_TIMER_CONTROL_ENABLE | TWD_TIMER_CONTROL_IT_ENABLE
 			| TWD_TIMER_CONTROL_PERIODIC;
-		__raw_writel(DIV_ROUND_CLOSEST(twd_timer_rate, HZ),
+		writel_relaxed(DIV_ROUND_CLOSEST(twd_timer_rate, HZ),
 			twd_base + TWD_TIMER_LOAD);
 		break;
 	case CLOCK_EVT_MODE_ONESHOT:
@@ -59,18 +59,18 @@ static void twd_set_mode(enum clock_event_mode mode,
 		ctrl = 0;
 	}
 
-	__raw_writel(ctrl, twd_base + TWD_TIMER_CONTROL);
+	writel_relaxed(ctrl, twd_base + TWD_TIMER_CONTROL);
 }
 
 static int twd_set_next_event(unsigned long evt,
 			struct clock_event_device *unused)
 {
-	unsigned long ctrl = __raw_readl(twd_base + TWD_TIMER_CONTROL);
+	unsigned long ctrl = readl_relaxed(twd_base + TWD_TIMER_CONTROL);
 
 	ctrl |= TWD_TIMER_CONTROL_ENABLE;
 
-	__raw_writel(evt, twd_base + TWD_TIMER_COUNTER);
-	__raw_writel(ctrl, twd_base + TWD_TIMER_CONTROL);
+	writel_relaxed(evt, twd_base + TWD_TIMER_COUNTER);
+	writel_relaxed(ctrl, twd_base + TWD_TIMER_CONTROL);
 
 	return 0;
 }
@@ -83,8 +83,8 @@ static int twd_set_next_event(unsigned long evt,
  */
 static int twd_timer_ack(void)
 {
-	if (__raw_readl(twd_base + TWD_TIMER_INTSTAT)) {
-		__raw_writel(1, twd_base + TWD_TIMER_INTSTAT);
+	if (readl_relaxed(twd_base + TWD_TIMER_INTSTAT)) {
+		writel_relaxed(1, twd_base + TWD_TIMER_INTSTAT);
 		return 1;
 	}
 
@@ -210,15 +210,15 @@ static void __cpuinit twd_calibrate_rate(void)
 		waitjiffies += 5;
 
 				 /* enable, no interrupt or reload */
-		__raw_writel(0x1, twd_base + TWD_TIMER_CONTROL);
+		writel_relaxed(0x1, twd_base + TWD_TIMER_CONTROL);
 
 				 /* maximum value */
-		__raw_writel(0xFFFFFFFFU, twd_base + TWD_TIMER_COUNTER);
+		writel_relaxed(0xFFFFFFFFU, twd_base + TWD_TIMER_COUNTER);
 
 		while (get_jiffies_64() < waitjiffies)
 			udelay(10);
 
-		count = __raw_readl(twd_base + TWD_TIMER_COUNTER);
+		count = readl_relaxed(twd_base + TWD_TIMER_COUNTER);
 
 		twd_timer_rate = (0xFFFFFFFFU - count) * (HZ / 5);
 
@@ -273,7 +273,7 @@ static int __cpuinit twd_timer_setup(struct clock_event_device *clk)
 	 * bother with the below.
 	 */
 	if (per_cpu(percpu_setup_called, cpu)) {
-		__raw_writel(0, twd_base + TWD_TIMER_CONTROL);
+		writel_relaxed(0, twd_base + TWD_TIMER_CONTROL);
 		clockevents_register_device(*__this_cpu_ptr(twd_evt));
 		enable_percpu_irq(clk->irq, 0);
 		return 0;
@@ -305,7 +305,7 @@ static int __cpuinit twd_timer_setup(struct clock_event_device *clk)
 	 * The following is done once per CPU the first time .setup() is
 	 * called.
 	 */
-	__raw_writel(0, twd_base + TWD_TIMER_CONTROL);
+	writel_relaxed(0, twd_base + TWD_TIMER_CONTROL);
 
 	clk->name = "local_timer";
 	clk->features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT |
Signature · 1 line
1.7.10.4
author
#22 lore raw

PATCH 8/17 ARM: smp_scu: data endian fixes

In reply to Ben Dooks

The smp_scu driver needs to use the relaxed readl/write accessors to avoid any issues with the endian mode the processor core is in.

---
 arch/arm/kernel/smp_scu.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
arch/arm/kernel/smp_scu.c +7 -7
index 45eac87..0fcd0ce 100644
@@ -28,7 +28,7 @@
  */
 unsigned int __init scu_get_core_count(void __iomem *scu_base)
 {
-	unsigned int ncores = __raw_readl(scu_base + SCU_CONFIG);
+	unsigned int ncores = readl_relaxed(scu_base + SCU_CONFIG);
 	return (ncores & 0x03) + 1;
 }
 
@@ -42,19 +42,19 @@ void scu_enable(void __iomem *scu_base)
 #ifdef CONFIG_ARM_ERRATA_764369
 	/* Cortex-A9 only */
 	if ((read_cpuid(CPUID_ID) & 0xff0ffff0) == 0x410fc090) {
-		scu_ctrl = __raw_readl(scu_base + 0x30);
+		scu_ctrl = readl_relaxed(scu_base + 0x30);
 		if (!(scu_ctrl & 1))
-			__raw_writel(scu_ctrl | 0x1, scu_base + 0x30);
+			writel_relaxed(scu_ctrl | 0x1, scu_base + 0x30);
 	}
 #endif
 
-	scu_ctrl = __raw_readl(scu_base + SCU_CTRL);
+	scu_ctrl = readl_relaxed(scu_base + SCU_CTRL);
 	/* already enabled? */
 	if (scu_ctrl & 1)
 		return;
 
 	scu_ctrl |= 1;
-	__raw_writel(scu_ctrl, scu_base + SCU_CTRL);
+	writel_relaxed(scu_ctrl, scu_base + SCU_CTRL);
 
 	/*
 	 * Ensure that the data accessed by CPU0 before the SCU was
@@ -80,9 +80,9 @@ int scu_power_mode(void __iomem *scu_base, unsigned int mode)
 	if (mode > 3 || mode == 1 || cpu > 3)
 		return -EINVAL;
 
-	val = __raw_readb(scu_base + SCU_CPU_STATUS + cpu) & ~0x03;
+	val = readb_relaxed(scu_base + SCU_CPU_STATUS + cpu) & ~0x03;
 	val |= mode;
-	__raw_writeb(val, scu_base + SCU_CPU_STATUS + cpu);
+	writeb_relaxed(val, scu_base + SCU_CPU_STATUS + cpu);
 
 	return 0;
 }
Signature · 1 line
1.7.10.4
author
#23 lore raw

PATCH 9/17 ARM: add atag32_to_cpu() function

In reply to Ben Dooks

Add atag32_to_cpu() function to allow code manipulating ATAGs to deal with the case where the boot-loader was in little-endian and Linux is running big-endian.

---
 arch/arm/include/uapi/asm/setup.h |    9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)
arch/arm/include/uapi/asm/setup.h +7 -2
index 979ff40..d74a417 100644
@@ -175,13 +175,18 @@ struct tagtable {
 
 #define tag_member_present(tag,member)				\
 	((unsigned long)(&((struct tag *)0L)->member + 1)	\
-		<= (tag)->hdr.size * 4)
+		<= atag32_to_cpu((tag)->hdr.size) * 4)
 
-#define tag_next(t)	((struct tag *)((__u32 *)(t) + (t)->hdr.size))
+#define tag_next(t)	((struct tag *)((__u32 *)(t) + atag32_to_cpu((t)->hdr.size)))
 #define tag_size(type)	((sizeof(struct tag_header) + sizeof(struct type)) >> 2)
 
 #define for_each_tag(t,base)		\
 	for (t = base; t->hdr.size; t = tag_next(t))
 
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+#define atag32_to_cpu(x)	le32_to_cpu(x)
+#else
+#define atag32_to_cpu(x)	x
+#endif
 
 #endif /* _UAPI__ASMARM_SETUP_H */
Signature · 1 line
1.7.10.4
#24 lore raw

PATCH 9/17 ARM: add atag32_to_cpu() function

Quoted 5 lines · Ben Dooks

Add atag32_to_cpu() function to allow code manipulating ATAGs to deal with the case where the boot-loader was in little-endian and Linux is running big-endian.

Quoted 38 lines
---
 arch/arm/include/uapi/asm/setup.h |    9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)
arch/arm/include/uapi/asm/setup.h +7 -2
index 979ff40..d74a417 100644
@@ -175,13 +175,18 @@ struct tagtable {
 
 #define tag_member_present(tag,member)				\
 	((unsigned long)(&((struct tag *)0L)->member + 1)	\
-		<= (tag)->hdr.size * 4)
+		<= atag32_to_cpu((tag)->hdr.size) * 4)
 
-#define tag_next(t)	((struct tag *)((__u32 *)(t) + (t)->hdr.size))
+#define tag_next(t)	((struct tag *)((__u32 *)(t) + atag32_to_cpu((t)->hdr.size)))
 #define tag_size(type)	((sizeof(struct tag_header) + sizeof(struct type)) >> 2)
 
 #define for_each_tag(t,base)		\
 	for (t = base; t->hdr.size; t = tag_next(t))
 
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+#define atag32_to_cpu(x)	le32_to_cpu(x)
+#else
+#define atag32_to_cpu(x)	x
+#endif
 
 #endif /* _UAPI__ASMARM_SETUP_H */
Signature · 7 lines
1.7.10.4


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#25 lore raw

PATCH 9/17 ARM: add atag32_to_cpu() function

In reply to Ben Dooks
Quoted 3 lines · Ben Dooks

Add atag32_to_cpu() function to allow code manipulating ATAGs to deal with the case where the boot-loader was in little-endian and Linux is running big-endian.

...

arch/arm/include/uapi/asm/setup.h struct tagtable { · line 190

Quoted 5 lines
arch/arm/include/uapi/asm/setup.h +5 -0
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+#define atag32_to_cpu(x)	le32_to_cpu(x)
+#else
+#define atag32_to_cpu(x)	x
+#endif

Actually, with a LE kernel, le32_to_cpu() ends up being a no-op too. So I'm wondering if we should do this a different way...

Also, this will cause sparse to complain - le32_to_cpu() takes a le32 type not a u32 type.

This brings up a whole load of difficulties though: what format are the ATAGs on older BE hardware - I bet it's CPU-endian format there. So, I'm wondering if we should do this:

1. define all tags using a new __atagXX, etc types.
2. always use atagXX_to_cpu() to read these.
3. Implement:

#if defined(CONFIG_ATAG_LE)
typedef __le32 __atag32;
...
#define atag32_to_cpu(x) le32_to_cpu(x)
...
#elif defined(CONFIG_ATAG_BE)
typedef __be32 __atag32;
...
#define atag32_to_cpu(x) be32_to_cpu(x)
...
#elif defined(CONFIG_ATAG_NE)
typedef __u32 __atag32;
...
#define atag32_to_cpu(x) x
...
#endif

and select the appropriate definition. Obviously, this is a fundamental configuration just like the overall BE/LE configuration of the kernel.

author
#26 lore raw

PATCH 9/17 ARM: add atag32_to_cpu() function

Quoted 14 lines · Russell King - ARM Linux
Quoted 3 lines · Ben Dooks

Add atag32_to_cpu() function to allow code manipulating ATAGs to deal with the case where the boot-loader was in little-endian and Linux is running big-endian.

...

arch/arm/include/uapi/asm/setup.h struct tagtable { · line 190

Quoted 5 lines
arch/arm/include/uapi/asm/setup.h +5 -0
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+#define atag32_to_cpu(x)	le32_to_cpu(x)
+#else
+#define atag32_to_cpu(x)	x
+#endif

Actually, with a LE kernel, le32_to_cpu() ends up being a no-op too. So I'm wondering if we should do this a different way...

Quoted 2 lines

Also, this will cause sparse to complain - le32_to_cpu() takes a le32 type not a u32 type.

Yes, you're right on that.

Quoted 3 lines

This brings up a whole load of difficulties though: what format are the ATAGs on older BE hardware - I bet it's CPU-endian format there. So, I'm wondering if we should do this:

It isn't defined, and I have failed to find any information on it online. I have no idea if the IXP devices ran their boot-loader in big-endian or little-endian format?

Input welcome from anyone who can remember.

Quoted 21 lines

1. define all tags using a new __atagXX, etc types.
2. always use atagXX_to_cpu() to read these.
3. Implement:

#if defined(CONFIG_ATAG_LE)
typedef __le32 __atag32;
...
#define atag32_to_cpu(x) le32_to_cpu(x)
...
#elif defined(CONFIG_ATAG_BE)
typedef __be32 __atag32;
...
#define atag32_to_cpu(x) be32_to_cpu(x)
...
#elif defined(CONFIG_ATAG_NE)
typedef __u32 __atag32;
...
#define atag32_to_cpu(x) x
...
#endif

Quoted 2 lines

and select the appropriate definition. Obviously, this is a fundamental configuration just like the overall BE/LE configuration of the kernel.

Ok, should I sort out doing that for the next round of patches?

Should the ATAG_xx configurations go into arch/arm/Kconfig or is there somewhere better for them to go?

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
#27 lore raw

PATCH 9/17 ARM: add atag32_to_cpu() function

Quoted 3 lines · Ben Dooks

It isn't defined, and I have failed to find any information on it online. I have no idea if the IXP devices ran their boot-loader in big-endian or little-endian format?

It will be native endian, because that's the format which we've supported up to now.

Quoted 21 lines
Quoted 21 lines

1. define all tags using a new __atagXX, etc types.
2. always use atagXX_to_cpu() to read these.
3. Implement:

#if defined(CONFIG_ATAG_LE)
typedef __le32 __atag32;
...
#define atag32_to_cpu(x) le32_to_cpu(x)
...
#elif defined(CONFIG_ATAG_BE)
typedef __be32 __atag32;
...
#define atag32_to_cpu(x) be32_to_cpu(x)
...
#elif defined(CONFIG_ATAG_NE)
typedef __u32 __atag32;
...
#define atag32_to_cpu(x) x
...
#endif

Quoted 4 lines
Quoted 2 lines

and select the appropriate definition. Obviously, this is a fundamental configuration just like the overall BE/LE configuration of the kernel.

Ok, should I sort out doing that for the next round of patches?

Yes please.

Quoted 2 lines

Should the ATAG_xx configurations go into arch/arm/Kconfig or is there somewhere better for them to go?

arch/arm/Kconfig for the time being. I've debated about moving some config options to arch/arm/boot/Kconfig

#28 lore raw

PATCH 9/17 ARM: add atag32_to_cpu() function

In reply to Ben Dooks
Quoted 24 lines · Ben Dooks
Quoted 14 lines · Russell King - ARM Linux
Quoted 3 lines · Ben Dooks

Add atag32_to_cpu() function to allow code manipulating ATAGs to deal with the case where the boot-loader was in little-endian and Linux is running big-endian.

...

arch/arm/include/uapi/asm/setup.h struct tagtable { · line 190

Quoted 5 lines
arch/arm/include/uapi/asm/setup.h +5 -0
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+#define atag32_to_cpu(x)	le32_to_cpu(x)
+#else
+#define atag32_to_cpu(x)	x
+#endif

Actually, with a LE kernel, le32_to_cpu() ends up being a no-op too. So I'm wondering if we should do this a different way...

Quoted 2 lines

Also, this will cause sparse to complain - le32_to_cpu() takes a le32 type not a u32 type.

Yes, you're right on that.

Quoted 3 lines

This brings up a whole load of difficulties though: what format are the ATAGs on older BE hardware - I bet it's CPU-endian format there. So, I'm wondering if we should do this:

Quoted 5 lines

It isn't defined, and I have failed to find any information on it online. I have no idea if the IXP devices ran their boot-loader in big-endian or little-endian format?

Input welcome from anyone who can remember.

I run the ixp in be

and pass the atag in be

Best Regards,
J.

Quoted 38 lines
Quoted 22 lines

1. define all tags using a new __atagXX, etc types.
2. always use atagXX_to_cpu() to read these.
3. Implement:

#if defined(CONFIG_ATAG_LE)
typedef __le32 __atag32;
...
#define atag32_to_cpu(x) le32_to_cpu(x)
...
#elif defined(CONFIG_ATAG_BE)
typedef __be32 __atag32;
...
#define atag32_to_cpu(x) be32_to_cpu(x)
...
#elif defined(CONFIG_ATAG_NE)
typedef __u32 __atag32;
...
#define atag32_to_cpu(x) x
...
#endif

Quoted 2 lines

and select the appropriate definition. Obviously, this is a fundamental configuration just like the overall BE/LE configuration of the kernel.

Ok, should I sort out doing that for the next round of patches?

Should the ATAG_xx configurations go into arch/arm/Kconfig or is there somewhere better for them to go?

Signature · 7 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#29 lore raw

PATCH 9/17 ARM: add atag32_to_cpu() function

Quoted 3 lines · Jean-Christophe PLAGNIOL-VILLARD

I run the ixp in be

and pass the atag in be

So that would be "native endian" then.

author
#30 lore raw

PATCH 10/17 ARM: update atag-to-fdt code to be endian agnostic

In reply to Ben Dooks

The atag-to-fdt code can now use the conversion function we introduced previously to avoid it having to know about the endian-ness of the environment which booted the processor.

---
 arch/arm/boot/compressed/atags_to_fdt.c |   20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)
arch/arm/boot/compressed/atags_to_fdt.c +10 -10
index aabc02a..a64ec71 100644
@@ -108,9 +108,9 @@ int atags_to_fdt(void *atag_list, void *fdt, int total_space)
 	       return 0;
 
 	/* validate the ATAG */
-	if (atag->hdr.tag != ATAG_CORE ||
-	    (atag->hdr.size != tag_size(tag_core) &&
-	     atag->hdr.size != 2))
+	if (atag->hdr.tag != atag32_to_cpu(ATAG_CORE) ||
+	    (atag->hdr.size != atag32_to_cpu(tag_size(tag_core)) &&
+	     atag->hdr.size != atag32_to_cpu(2)))
 		return 1;
 
 	/* let's give it all the room it could need */
@@ -132,17 +132,17 @@ int atags_to_fdt(void *atag_list, void *fdt, int total_space)
 			else
 				setprop_string(fdt, "/chosen", "bootargs",
 					       atag->u.cmdline.cmdline);
-		} else if (atag->hdr.tag == ATAG_MEM) {
+		} else if (atag->hdr.tag == atag32_to_cpu(ATAG_MEM)) {
 			if (memcount >= sizeof(mem_reg_property)/4)
 				continue;
-			if (!atag->u.mem.size)
+			if (!atag32_to_cpu(atag->u.mem.size))
 				continue;
-			mem_reg_property[memcount++] = cpu_to_fdt32(atag->u.mem.start);
-			mem_reg_property[memcount++] = cpu_to_fdt32(atag->u.mem.size);
-		} else if (atag->hdr.tag == ATAG_INITRD2) {
+			mem_reg_property[memcount++] = cpu_to_fdt32(atag32_to_cpu(atag->u.mem.start));
+			mem_reg_property[memcount++] = cpu_to_fdt32(atag32_to_cpu(atag->u.mem.size));
+		} else if (atag->hdr.tag == atag32_to_cpu(ATAG_INITRD2)) {
 			uint32_t initrd_start, initrd_size;
-			initrd_start = atag->u.initrd.start;
-			initrd_size = atag->u.initrd.size;
+			initrd_start = atag32_to_cpu(atag->u.initrd.start);
+			initrd_size = atag32_to_cpu(____atag->u.initrd.size);
 			setprop_cell(fdt, "/chosen", "linux,initrd-start",
 					initrd_start);
 			setprop_cell(fdt, "/chosen", "linux,initrd-end",
Signature · 1 line
1.7.10.4
#31 lore raw

PATCH 10/17 ARM: update atag-to-fdt code to be endian agnostic

Quoted 5 lines · Ben Dooks

The atag-to-fdt code can now use the conversion function we introduced previously to avoid it having to know about the endian-ness of the environment which booted the processor.

Quoted 55 lines
---
 arch/arm/boot/compressed/atags_to_fdt.c |   20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)
arch/arm/boot/compressed/atags_to_fdt.c +10 -10
index aabc02a..a64ec71 100644
@@ -108,9 +108,9 @@ int atags_to_fdt(void *atag_list, void *fdt, int total_space)
 	       return 0;
 
 	/* validate the ATAG */
-	if (atag->hdr.tag != ATAG_CORE ||
-	    (atag->hdr.size != tag_size(tag_core) &&
-	     atag->hdr.size != 2))
+	if (atag->hdr.tag != atag32_to_cpu(ATAG_CORE) ||
+	    (atag->hdr.size != atag32_to_cpu(tag_size(tag_core)) &&
+	     atag->hdr.size != atag32_to_cpu(2)))
 		return 1;
 
 	/* let's give it all the room it could need */
@@ -132,17 +132,17 @@ int atags_to_fdt(void *atag_list, void *fdt, int total_space)
 			else
 				setprop_string(fdt, "/chosen", "bootargs",
 					       atag->u.cmdline.cmdline);
-		} else if (atag->hdr.tag == ATAG_MEM) {
+		} else if (atag->hdr.tag == atag32_to_cpu(ATAG_MEM)) {
 			if (memcount >= sizeof(mem_reg_property)/4)
 				continue;
-			if (!atag->u.mem.size)
+			if (!atag32_to_cpu(atag->u.mem.size))
 				continue;
-			mem_reg_property[memcount++] = cpu_to_fdt32(atag->u.mem.start);
-			mem_reg_property[memcount++] = cpu_to_fdt32(atag->u.mem.size);
-		} else if (atag->hdr.tag == ATAG_INITRD2) {
+			mem_reg_property[memcount++] = cpu_to_fdt32(atag32_to_cpu(atag->u.mem.start));
+			mem_reg_property[memcount++] = cpu_to_fdt32(atag32_to_cpu(atag->u.mem.size));
+		} else if (atag->hdr.tag == atag32_to_cpu(ATAG_INITRD2)) {
 			uint32_t initrd_start, initrd_size;
-			initrd_start = atag->u.initrd.start;
-			initrd_size = atag->u.initrd.size;
+			initrd_start = atag32_to_cpu(atag->u.initrd.start);
+			initrd_size = atag32_to_cpu(____atag->u.initrd.size);
 			setprop_cell(fdt, "/chosen", "linux,initrd-start",
 					initrd_start);
 			setprop_cell(fdt, "/chosen", "linux,initrd-end",
Signature · 7 lines
1.7.10.4


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#32 lore raw

PATCH 10/17 ARM: update atag-to-fdt code to be endian agnostic

In reply to Ben Dooks

arch/arm/boot/compressed/atags_to_fdt.c int atags_to_fdt(void *atag_list, void *fdt, int total_space) · line 113

Quoted 6 lines · Ben Dooks
arch/arm/boot/compressed/atags_to_fdt.c +3 -3
-	if (atag->hdr.tag != ATAG_CORE ||
-	    (atag->hdr.size != tag_size(tag_core) &&
-	     atag->hdr.size != 2))
+	if (atag->hdr.tag != atag32_to_cpu(ATAG_CORE) ||
+	    (atag->hdr.size != atag32_to_cpu(tag_size(tag_core)) &&
+	     atag->hdr.size != atag32_to_cpu(2)))

This is wrong. You're saying that "ATAG_CORE" is in LE32 format. It isn't. It's in CPU endian format. If you want to do this kind of thing, you also need to define cpu_to_atag32() macros as well, otherwise this will cause sparse warnings.

arch/arm/boot/compressed/atags_to_fdt.c int atags_to_fdt(void *atag_list, void *fdt, int total_space) · line 145

Quoted 4 lines
arch/arm/boot/compressed/atags_to_fdt.c +2 -2
-			initrd_start = atag->u.initrd.start;
-			initrd_size = atag->u.initrd.size;
+			initrd_start = atag32_to_cpu(atag->u.initrd.start);
+			initrd_size = atag32_to_cpu(____atag->u.initrd.size);

Where did those four underscores come from? Has this been built?

author
#33 lore raw

PATCH 10/17 ARM: update atag-to-fdt code to be endian agnostic

Quoted 13 lines · Russell King - ARM Linux

arch/arm/boot/compressed/atags_to_fdt.c int atags_to_fdt(void *atag_list, void *fdt, int total_space) · line 113

Quoted 6 lines · Ben Dooks
arch/arm/boot/compressed/atags_to_fdt.c +3 -3
-	if (atag->hdr.tag != ATAG_CORE ||
-	    (atag->hdr.size != tag_size(tag_core)&&
-	     atag->hdr.size != 2))
+	if (atag->hdr.tag != atag32_to_cpu(ATAG_CORE) ||
+	    (atag->hdr.size != atag32_to_cpu(tag_size(tag_core))&&
+	     atag->hdr.size != atag32_to_cpu(2)))

This is wrong. You're saying that "ATAG_CORE" is in LE32 format. It isn't. It's in CPU endian format. If you want to do this kind of thing, you also need to define cpu_to_atag32() macros as well, otherwise this will cause sparse warnings.

Quoted 6 lines

arch/arm/boot/compressed/atags_to_fdt.c int atags_to_fdt(void *atag_list, void *fdt, int total_space) · line 145

Quoted 4 lines
arch/arm/boot/compressed/atags_to_fdt.c +2 -2
-			initrd_start = atag->u.initrd.start;
-			initrd_size = atag->u.initrd.size;
+			initrd_start = atag32_to_cpu(atag->u.initrd.start);
+			initrd_size = atag32_to_cpu(____atag->u.initrd.size);

Where did those four underscores come from? Has this been built?

I probably missed building this one. I've been mostly testing with DT based systems.

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
#34 lore raw

PATCH 10/17 ARM: update atag-to-fdt code to be endian agnostic

Quoted 21 lines · Ben Dooks
Quoted 13 lines · Russell King - ARM Linux

arch/arm/boot/compressed/atags_to_fdt.c int atags_to_fdt(void *atag_list, void *fdt, int total_space) · line 113

Quoted 6 lines · Ben Dooks
arch/arm/boot/compressed/atags_to_fdt.c +3 -3
-	if (atag->hdr.tag != ATAG_CORE ||
-	    (atag->hdr.size != tag_size(tag_core)&&
-	     atag->hdr.size != 2))
+	if (atag->hdr.tag != atag32_to_cpu(ATAG_CORE) ||
+	    (atag->hdr.size != atag32_to_cpu(tag_size(tag_core))&&
+	     atag->hdr.size != atag32_to_cpu(2)))

This is wrong. You're saying that "ATAG_CORE" is in LE32 format. It isn't. It's in CPU endian format. If you want to do this kind of thing, you also need to define cpu_to_atag32() macros as well, otherwise this will cause sparse warnings.

Quoted 6 lines

arch/arm/boot/compressed/atags_to_fdt.c int atags_to_fdt(void *atag_list, void *fdt, int total_space) · line 145

Quoted 4 lines
arch/arm/boot/compressed/atags_to_fdt.c +2 -2
-			initrd_start = atag->u.initrd.start;
-			initrd_size = atag->u.initrd.size;
+			initrd_start = atag32_to_cpu(atag->u.initrd.start);
+			initrd_size = atag32_to_cpu(____atag->u.initrd.size);

Where did those four underscores come from? Has this been built?

Quoted 2 lines

I probably missed building this one. I've been mostly testing with DT based systems.

Is this BE8 mode available on old systems? Or, will those BE8 capable old systems have BE userland compiled for them?

Where I want to get to is: do we need to support BE8 mode for ATAG based systems at all, given that most if not all the modern ones should be DT based now? Making the ATAG code BE8 aware is looking to be quite invasive for potentially no users at all.

Nicolas

author
#35 lore raw

PATCH 10/17 ARM: update atag-to-fdt code to be endian agnostic

Quoted 23 lines · Nicolas Pitre
Quoted 21 lines · Ben Dooks
Quoted 19 lines · Russell King - ARM Linux

arch/arm/boot/compressed/atags_to_fdt.c int atags_to_fdt(void *atag_list, void *fdt, int total_space) · line 113

Quoted 6 lines · Ben Dooks
arch/arm/boot/compressed/atags_to_fdt.c +3 -3
-	if (atag->hdr.tag != ATAG_CORE ||
-	    (atag->hdr.size != tag_size(tag_core)&&
-	     atag->hdr.size != 2))
+	if (atag->hdr.tag != atag32_to_cpu(ATAG_CORE) ||
+	    (atag->hdr.size != atag32_to_cpu(tag_size(tag_core))&&
+	     atag->hdr.size != atag32_to_cpu(2)))

This is wrong. You're saying that "ATAG_CORE" is in LE32 format. It isn't. It's in CPU endian format. If you want to do this kind of thing, you also need to define cpu_to_atag32() macros as well, otherwise this will cause sparse warnings.

arch/arm/boot/compressed/atags_to_fdt.c int atags_to_fdt(void *atag_list, void *fdt, int total_space) · line 145

Quoted 4 lines
arch/arm/boot/compressed/atags_to_fdt.c +2 -2
-			initrd_start = atag->u.initrd.start;
-			initrd_size = atag->u.initrd.size;
+			initrd_start = atag32_to_cpu(atag->u.initrd.start);
+			initrd_size = atag32_to_cpu(____atag->u.initrd.size);

Where did those four underscores come from? Has this been built?

Quoted 5 lines
Quoted 2 lines

I probably missed building this one. I've been mostly testing with DT based systems.

Is this BE8 mode available on old systems? Or, will those BE8 capable old systems have BE userland compiled for them?

BE8 is for ARMv6 and ARMv7 form of big endian. ARMv5 is BE32.

Quoted 4 lines

Where I want to get to is: do we need to support BE8 mode for ATAG based systems at all, given that most if not all the modern ones should be DT based now? Making the ATAG code BE8 aware is looking to be quite invasive for potentially no users at all.

The change is only useful when using a BE8 compiled kernel with a boot loader environment that is LE. Even if the ATAGs do not need fixing up, then we have a issue with uboot checking the zImage magic.

I have pushed the ATAGs issues to a new series as it is not necessary for the current work we are doing with highbank primarily.

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
#36 lore raw

PATCH 10/17 ARM: update atag-to-fdt code to be endian agnostic

Quoted 27 lines · Ben Dooks
Quoted 26 lines · Nicolas Pitre
Quoted 24 lines · Ben Dooks
Quoted 22 lines · Russell King - ARM Linux

arch/arm/boot/compressed/atags_to_fdt.c int atags_to_fdt(void *atag_list, void *fdt, int total_space) · line 113

Quoted 6 lines · Ben Dooks
arch/arm/boot/compressed/atags_to_fdt.c +3 -3
-	if (atag->hdr.tag != ATAG_CORE ||
-	    (atag->hdr.size != tag_size(tag_core)&&
-	     atag->hdr.size != 2))
+	if (atag->hdr.tag != atag32_to_cpu(ATAG_CORE) ||
+	    (atag->hdr.size != atag32_to_cpu(tag_size(tag_core))&&
+	     atag->hdr.size != atag32_to_cpu(2)))

This is wrong. You're saying that "ATAG_CORE" is in LE32 format. It isn't. It's in CPU endian format. If you want to do this kind of thing,
you also need to define cpu_to_atag32() macros as well, otherwise this will cause sparse warnings.

Quoted 6 lines
arch/arm/boot/compressed/atags_to_fdt.c +1 -2
-			initrd_start = atag->u.initrd.start;
-			initrd_size = atag->u.initrd.size;
+			initrd_start =

atag32_to_cpu(atag->u.initrd.start);

(quoted diff) +1 -0
+			initrd_size =

atag32_to_cpu(____atag->u.initrd.size);

Where did those four underscores come from? Has this been built?

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

I probably missed building this one. I've been mostly testing with DT based systems.

Is this BE8 mode available on old systems? Or, will those BE8 capable old systems have BE userland compiled for them?

BE8 is for ARMv6 and ARMv7 form of big endian. ARMv5 is BE32.

I think that by now all ARMv6+ targets should be DT enabled. So you shouldn't have to care about ATAGs at all.

Quoted 7 lines
Quoted 4 lines

Where I want to get to is: do we need to support BE8 mode for ATAG based systems at all, given that most if not all the modern ones should be DT based now? Making the ATAG code BE8 aware is looking to be quite invasive for potentially no users at all.

The change is only useful when using a BE8 compiled kernel with a boot loader environment that is LE.

That's what CONFIG_ARM_ATAG_DTB_COMPAT is for. So only that code would need to swap the ATAG data.

Quoted 2 lines

Even if the ATAGs do not need
fixing up, then we have a issue with uboot checking the zImage magic.

I gave you a patch for that.

Quoted 2 lines

I have pushed the ATAGs issues to a new series as it is not necessary for the current work we are doing with highbank primarily.

I don't think it is needed at all given that BE8 is available mainly on targets supporting DT, and is likely to be used on targets that do support only DT.

Nicolas

author
#37 lore raw

PATCH 10/17 ARM: update atag-to-fdt code to be endian agnostic

Quoted 54 lines · Nicolas Pitre
Quoted 27 lines · Ben Dooks
Quoted 26 lines · Nicolas Pitre
Quoted 24 lines · Ben Dooks
Quoted 22 lines · Russell King - ARM Linux

arch/arm/boot/compressed/atags_to_fdt.c int atags_to_fdt(void *atag_list, void *fdt, int total_space) · line 113

Quoted 6 lines · Ben Dooks
arch/arm/boot/compressed/atags_to_fdt.c +3 -3
-	if (atag->hdr.tag != ATAG_CORE ||
-	    (atag->hdr.size != tag_size(tag_core)&&
-	     atag->hdr.size != 2))
+	if (atag->hdr.tag != atag32_to_cpu(ATAG_CORE) ||
+	    (atag->hdr.size != atag32_to_cpu(tag_size(tag_core))&&
+	     atag->hdr.size != atag32_to_cpu(2)))

This is wrong. You're saying that "ATAG_CORE" is in LE32 format. It isn't. It's in CPU endian format. If you want to do this kind of thing,
you also need to define cpu_to_atag32() macros as well, otherwise this will cause sparse warnings.

Quoted 6 lines
arch/arm/boot/compressed/atags_to_fdt.c +1 -2
-			initrd_start = atag->u.initrd.start;
-			initrd_size = atag->u.initrd.size;
+			initrd_start =

atag32_to_cpu(atag->u.initrd.start);

(quoted diff) +1 -0
+			initrd_size =

atag32_to_cpu(____atag->u.initrd.size);

Where did those four underscores come from? Has this been built?

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

I probably missed building this one. I've been mostly testing with DT based systems.

Is this BE8 mode available on old systems? Or, will those BE8 capable old systems have BE userland compiled for them?

BE8 is for ARMv6 and ARMv7 form of big endian. ARMv5 is BE32.

I think that by now all ARMv6+ targets should be DT enabled. So you shouldn't have to care about ATAGs at all.

Quoted 7 lines
Quoted 4 lines

Where I want to get to is: do we need to support BE8 mode for ATAG based systems at all, given that most if not all the modern ones should be DT based now? Making the ATAG code BE8 aware is looking to be quite invasive for potentially no users at all.

The change is only useful when using a BE8 compiled kernel with a boot loader environment that is LE.

That's what CONFIG_ARM_ATAG_DTB_COMPAT is for. So only that code would need to swap the ATAG data.

Quoted 2 lines

Even if the ATAGs do not need
fixing up, then we have a issue with uboot checking the zImage magic.

Quoted 8 lines

I gave you a patch for that.

Quoted 2 lines

I have pushed the ATAGs issues to a new series as it is not necessary for the current work we are doing with highbank primarily.

I don't think it is needed at all given that BE8 is available mainly on targets supporting DT, and is likely to be used on targets that do support only DT.

Some devices we have still only have ATAG booting in u-boot, so it i possible that it may have to stick around to allow the fdt to be updated with any extra command line stuff...

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
author
#38 lore raw

PATCH 11/17 ARM: fixup atags to be endian agnostic

In reply to Ben Dooks

Use the atag32_to_cpu() function to allow the main atag handling code to be agnostic of any endian configuration.

---
 arch/arm/kernel/atags_parse.c |   51 +++++++++++++++++++++--------------------
 arch/arm/kernel/atags_proc.c  |    6 ++---
 2 files changed, 29 insertions(+), 28 deletions(-)
arch/arm/kernel/atags_parse.c +26 -25
index 14512e6..cffcbc2 100644
@@ -43,19 +43,19 @@ static struct {
 	struct tag_mem32  mem;
 	struct tag_header hdr3;
 } default_tags __initdata = {
-	{ tag_size(tag_core), ATAG_CORE },
-	{ 1, PAGE_SIZE, 0xff },
-	{ tag_size(tag_mem32), ATAG_MEM },
-	{ MEM_SIZE },
-	{ 0, ATAG_NONE }
+	{ atag32_to_cpu(tag_size(tag_core)), atag32_to_cpu(ATAG_CORE) },
+	{ atag32_to_cpu(1), atag32_to_cpu(PAGE_SIZE), atag32_to_cpu(0xff) },
+	{ atag32_to_cpu(tag_size(tag_mem32)), atag32_to_cpu(ATAG_MEM) },
+	{ atag32_to_cpu(MEM_SIZE) },
+	{ atag32_to_cpu(0), atag32_to_cpu(ATAG_NONE) }
 };
 
 static int __init parse_tag_core(const struct tag *tag)
 {
-	if (tag->hdr.size > 2) {
-		if ((tag->u.core.flags & 1) == 0)
+	if (atag32_to_cpu(tag->hdr.size) > 2) {
+		if ((atag32_to_cpu(tag->u.core.flags) & 1) == 0)
 			root_mountflags &= ~MS_RDONLY;
-		ROOT_DEV = old_decode_dev(tag->u.core.rootdev);
+		ROOT_DEV = old_decode_dev(atag32_to_cpu(tag->u.core.rootdev));
 	}
 	return 0;
 }
@@ -64,7 +64,7 @@ __tagtable(ATAG_CORE, parse_tag_core);
 
 static int __init parse_tag_mem32(const struct tag *tag)
 {
-	return arm_add_memory(tag->u.mem.start, tag->u.mem.size);
+	return arm_add_memory(atag32_to_cpu(tag->u.mem.start), atag32_to_cpu(tag->u.mem.size));
 }
 
 __tagtable(ATAG_MEM, parse_tag_mem32);
@@ -91,13 +91,14 @@ __tagtable(ATAG_VIDEOTEXT, parse_tag_videotext);
 static int __init parse_tag_ramdisk(const struct tag *tag)
 {
 	extern int rd_size, rd_image_start, rd_prompt, rd_doload;
+	unsigned int flags = atag32_to_cpu(tag->u.ramdisk.flags);
 
-	rd_image_start = tag->u.ramdisk.start;
-	rd_doload = (tag->u.ramdisk.flags & 1) == 0;
-	rd_prompt = (tag->u.ramdisk.flags & 2) == 0;
+	rd_image_start = atag32_to_cpu(tag->u.ramdisk.start);
+	rd_doload = (flags & 1) == 0;
+	rd_prompt = (flags & 2) == 0;
 
-	if (tag->u.ramdisk.size)
-		rd_size = tag->u.ramdisk.size;
+	if (atag32_to_cpu(tag->u.ramdisk.size))
+		rd_size = atag32_to_cpu(tag->u.ramdisk.size);
 
 	return 0;
 }
@@ -107,8 +108,8 @@ __tagtable(ATAG_RAMDISK, parse_tag_ramdisk);
 
 static int __init parse_tag_serialnr(const struct tag *tag)
 {
-	system_serial_low = tag->u.serialnr.low;
-	system_serial_high = tag->u.serialnr.high;
+	system_serial_low = atag32_to_cpu(tag->u.serialnr.low);
+	system_serial_high = atag32_to_cpu(tag->u.serialnr.high);
 	return 0;
 }
 
@@ -116,7 +117,7 @@ __tagtable(ATAG_SERIAL, parse_tag_serialnr);
 
 static int __init parse_tag_revision(const struct tag *tag)
 {
-	system_rev = tag->u.revision.rev;
+	system_rev = atag32_to_cpu(tag->u.revision.rev);
 	return 0;
 }
 
@@ -150,7 +151,7 @@ static int __init parse_tag(const struct tag *tag)
 	struct tagtable *t;
 
 	for (t = &__tagtable_begin; t < &__tagtable_end; t++)
-		if (tag->hdr.tag == t->tag) {
+		if (atag32_to_cpu(tag->hdr.tag) == t->tag) {
 			t->parse(tag);
 			break;
 		}
@@ -164,17 +165,17 @@ static int __init parse_tag(const struct tag *tag)
  */
 static void __init parse_tags(const struct tag *t)
 {
-	for (; t->hdr.size; t = tag_next(t))
+	for_each_tag(t, t)
 		if (!parse_tag(t))
 			printk(KERN_WARNING
 				"Ignoring unrecognised tag 0x%08x\n",
-				t->hdr.tag);
+				atag32_to_cpu(t->hdr.tag));
 }
 
 static void __init squash_mem_tags(struct tag *tag)
 {
-	for (; tag->hdr.size; tag = tag_next(tag))
-		if (tag->hdr.tag == ATAG_MEM)
+	for_each_tag(tag, tag)
+		if (tag->hdr.tag == atag32_to_cpu(ATAG_MEM))
 			tag->hdr.tag = ATAG_NONE;
 }
 
@@ -213,10 +214,10 @@ struct machine_desc * __init setup_machine_tags(phys_addr_t __atags_pointer,
 	 * If we have the old style parameters, convert them to
 	 * a tag list.
 	 */
-	if (tags->hdr.tag != ATAG_CORE)
+	if (tags->hdr.tag != atag32_to_cpu(ATAG_CORE))
 		convert_to_tag_list(tags);
 #endif
-	if (tags->hdr.tag != ATAG_CORE) {
+	if (tags->hdr.tag != atag32_to_cpu(ATAG_CORE)) {
 		early_print("Warning: Neither atags nor dtb found\n");
 		tags = (struct tag *)&default_tags;
 	}
@@ -224,7 +225,7 @@ struct machine_desc * __init setup_machine_tags(phys_addr_t __atags_pointer,
 	if (mdesc->fixup)
 		mdesc->fixup(tags, &from, &meminfo);
 
-	if (tags->hdr.tag == ATAG_CORE) {
+	if (tags->hdr.tag == atag32_to_cpu(ATAG_CORE)) {
 		if (meminfo.nr_banks != 0)
 			squash_mem_tags(tags);
 		save_atags(tags);
arch/arm/kernel/atags_proc.c +3 -3
index 42a1a14..50e7e9c 100644
@@ -46,18 +46,18 @@ static int __init init_atags_procfs(void)
 	struct buffer *b;
 	size_t size;
 
-	if (tag->hdr.tag != ATAG_CORE) {
+	if (tag->hdr.tag != atag32_to_cpu(ATAG_CORE)) {
 		printk(KERN_INFO "No ATAGs?");
 		return -EINVAL;
 	}
 
-	for (; tag->hdr.size; tag = tag_next(tag))
+	for_each_tag(tag, tag)
 		;
 
 	/* include the terminating ATAG_NONE */
 	size = (char *)tag - atags_copy + sizeof(struct tag_header);
 
-	WARN_ON(tag->hdr.tag != ATAG_NONE);
+	WARN_ON(tag->hdr.tag != atag32_to_cpu(ATAG_NONE));
 
 	b = kmalloc(sizeof(*b) + size, GFP_KERNEL);
 	if (!b)
Signature · 1 line
1.7.10.4
#39 lore raw

PATCH 11/17 ARM: fixup atags to be endian agnostic

Quoted 4 lines · Ben Dooks

Use the atag32_to_cpu() function to allow the main atag handling code to be agnostic of any endian configuration.

This is looking rather error prone, especially since there is no sparse attribute to flag dereferences without the proper accessor.

Wouldn't it be simpler if you just determined the actual size of the ATAG block and byte swapped it all?

Quoted 174 lines
---
 arch/arm/kernel/atags_parse.c |   51 +++++++++++++++++++++--------------------
 arch/arm/kernel/atags_proc.c  |    6 ++---
 2 files changed, 29 insertions(+), 28 deletions(-)
arch/arm/kernel/atags_parse.c +26 -25
index 14512e6..cffcbc2 100644
@@ -43,19 +43,19 @@ static struct {
 	struct tag_mem32  mem;
 	struct tag_header hdr3;
 } default_tags __initdata = {
-	{ tag_size(tag_core), ATAG_CORE },
-	{ 1, PAGE_SIZE, 0xff },
-	{ tag_size(tag_mem32), ATAG_MEM },
-	{ MEM_SIZE },
-	{ 0, ATAG_NONE }
+	{ atag32_to_cpu(tag_size(tag_core)), atag32_to_cpu(ATAG_CORE) },
+	{ atag32_to_cpu(1), atag32_to_cpu(PAGE_SIZE), atag32_to_cpu(0xff) },
+	{ atag32_to_cpu(tag_size(tag_mem32)), atag32_to_cpu(ATAG_MEM) },
+	{ atag32_to_cpu(MEM_SIZE) },
+	{ atag32_to_cpu(0), atag32_to_cpu(ATAG_NONE) }
 };
 
 static int __init parse_tag_core(const struct tag *tag)
 {
-	if (tag->hdr.size > 2) {
-		if ((tag->u.core.flags & 1) == 0)
+	if (atag32_to_cpu(tag->hdr.size) > 2) {
+		if ((atag32_to_cpu(tag->u.core.flags) & 1) == 0)
 			root_mountflags &= ~MS_RDONLY;
-		ROOT_DEV = old_decode_dev(tag->u.core.rootdev);
+		ROOT_DEV = old_decode_dev(atag32_to_cpu(tag->u.core.rootdev));
 	}
 	return 0;
 }
@@ -64,7 +64,7 @@ __tagtable(ATAG_CORE, parse_tag_core);
 
 static int __init parse_tag_mem32(const struct tag *tag)
 {
-	return arm_add_memory(tag->u.mem.start, tag->u.mem.size);
+	return arm_add_memory(atag32_to_cpu(tag->u.mem.start), atag32_to_cpu(tag->u.mem.size));
 }
 
 __tagtable(ATAG_MEM, parse_tag_mem32);
@@ -91,13 +91,14 @@ __tagtable(ATAG_VIDEOTEXT, parse_tag_videotext);
 static int __init parse_tag_ramdisk(const struct tag *tag)
 {
 	extern int rd_size, rd_image_start, rd_prompt, rd_doload;
+	unsigned int flags = atag32_to_cpu(tag->u.ramdisk.flags);
 
-	rd_image_start = tag->u.ramdisk.start;
-	rd_doload = (tag->u.ramdisk.flags & 1) == 0;
-	rd_prompt = (tag->u.ramdisk.flags & 2) == 0;
+	rd_image_start = atag32_to_cpu(tag->u.ramdisk.start);
+	rd_doload = (flags & 1) == 0;
+	rd_prompt = (flags & 2) == 0;
 
-	if (tag->u.ramdisk.size)
-		rd_size = tag->u.ramdisk.size;
+	if (atag32_to_cpu(tag->u.ramdisk.size))
+		rd_size = atag32_to_cpu(tag->u.ramdisk.size);
 
 	return 0;
 }
@@ -107,8 +108,8 @@ __tagtable(ATAG_RAMDISK, parse_tag_ramdisk);
 
 static int __init parse_tag_serialnr(const struct tag *tag)
 {
-	system_serial_low = tag->u.serialnr.low;
-	system_serial_high = tag->u.serialnr.high;
+	system_serial_low = atag32_to_cpu(tag->u.serialnr.low);
+	system_serial_high = atag32_to_cpu(tag->u.serialnr.high);
 	return 0;
 }
 
@@ -116,7 +117,7 @@ __tagtable(ATAG_SERIAL, parse_tag_serialnr);
 
 static int __init parse_tag_revision(const struct tag *tag)
 {
-	system_rev = tag->u.revision.rev;
+	system_rev = atag32_to_cpu(tag->u.revision.rev);
 	return 0;
 }
 
@@ -150,7 +151,7 @@ static int __init parse_tag(const struct tag *tag)
 	struct tagtable *t;
 
 	for (t = &__tagtable_begin; t < &__tagtable_end; t++)
-		if (tag->hdr.tag == t->tag) {
+		if (atag32_to_cpu(tag->hdr.tag) == t->tag) {
 			t->parse(tag);
 			break;
 		}
@@ -164,17 +165,17 @@ static int __init parse_tag(const struct tag *tag)
  */
 static void __init parse_tags(const struct tag *t)
 {
-	for (; t->hdr.size; t = tag_next(t))
+	for_each_tag(t, t)
 		if (!parse_tag(t))
 			printk(KERN_WARNING
 				"Ignoring unrecognised tag 0x%08x\n",
-				t->hdr.tag);
+				atag32_to_cpu(t->hdr.tag));
 }
 
 static void __init squash_mem_tags(struct tag *tag)
 {
-	for (; tag->hdr.size; tag = tag_next(tag))
-		if (tag->hdr.tag == ATAG_MEM)
+	for_each_tag(tag, tag)
+		if (tag->hdr.tag == atag32_to_cpu(ATAG_MEM))
 			tag->hdr.tag = ATAG_NONE;
 }
 
@@ -213,10 +214,10 @@ struct machine_desc * __init setup_machine_tags(phys_addr_t __atags_pointer,
 	 * If we have the old style parameters, convert them to
 	 * a tag list.
 	 */
-	if (tags->hdr.tag != ATAG_CORE)
+	if (tags->hdr.tag != atag32_to_cpu(ATAG_CORE))
 		convert_to_tag_list(tags);
 #endif
-	if (tags->hdr.tag != ATAG_CORE) {
+	if (tags->hdr.tag != atag32_to_cpu(ATAG_CORE)) {
 		early_print("Warning: Neither atags nor dtb found\n");
 		tags = (struct tag *)&default_tags;
 	}
@@ -224,7 +225,7 @@ struct machine_desc * __init setup_machine_tags(phys_addr_t __atags_pointer,
 	if (mdesc->fixup)
 		mdesc->fixup(tags, &from, &meminfo);
 
-	if (tags->hdr.tag == ATAG_CORE) {
+	if (tags->hdr.tag == atag32_to_cpu(ATAG_CORE)) {
 		if (meminfo.nr_banks != 0)
 			squash_mem_tags(tags);
 		save_atags(tags);
arch/arm/kernel/atags_proc.c +3 -3
index 42a1a14..50e7e9c 100644
@@ -46,18 +46,18 @@ static int __init init_atags_procfs(void)
 	struct buffer *b;
 	size_t size;
 
-	if (tag->hdr.tag != ATAG_CORE) {
+	if (tag->hdr.tag != atag32_to_cpu(ATAG_CORE)) {
 		printk(KERN_INFO "No ATAGs?");
 		return -EINVAL;
 	}
 
-	for (; tag->hdr.size; tag = tag_next(tag))
+	for_each_tag(tag, tag)
 		;
 
 	/* include the terminating ATAG_NONE */
 	size = (char *)tag - atags_copy + sizeof(struct tag_header);
 
-	WARN_ON(tag->hdr.tag != ATAG_NONE);
+	WARN_ON(tag->hdr.tag != atag32_to_cpu(ATAG_NONE));
 
 	b = kmalloc(sizeof(*b) + size, GFP_KERNEL);
 	if (!b)
Signature · 7 lines
1.7.10.4


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#40 lore raw

PATCH 11/17 ARM: fixup atags to be endian agnostic

In reply to Ben Dooks

arch/arm/kernel/atags_parse.c static struct { · line 50

Quoted 5 lines · Ben Dooks
arch/arm/kernel/atags_parse.c +5 -0
+	{ atag32_to_cpu(tag_size(tag_core)), atag32_to_cpu(ATAG_CORE) },
+	{ atag32_to_cpu(1), atag32_to_cpu(PAGE_SIZE), atag32_to_cpu(0xff) },
+	{ atag32_to_cpu(tag_size(tag_mem32)), atag32_to_cpu(ATAG_MEM) },
+	{ atag32_to_cpu(MEM_SIZE) },
+	{ atag32_to_cpu(0), atag32_to_cpu(ATAG_NONE) }

Again, this needs to be cpu_to_atag32().

Pay attention to what you're converting from and to. Consider them totally separate types that just happen to be integers, and ensure that you match the types in your functions.

author
#41 lore raw

PATCH 12/17 ARM: fix magic for bootloader in BE8

In reply to Ben Dooks

If the boot-loader is running little endian and the kernel is compiled big endian then it will not verify the magic number of the front of the kernel.

Note, do we need to also reverse the two values after the magic, or are they used later on in the kernel decompressor?

---
 arch/arm/boot/compressed/head.S |    4 ++++
 1 file changed, 4 insertions(+)
arch/arm/boot/compressed/head.S +4 -0
index 39940a7..b6c7ec8 100644
@@ -129,7 +129,11 @@ start:
  THUMB(		adr	r12, BSYM(1f)	)
  THUMB(		bx	r12		)
 
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+		.word	0x18286f01
+#else
 		.word	0x016f2818		@ Magic numbers to help the loader
+#endif
 		.word	start			@ absolute load/run zImage address
 		.word	_edata			@ zImage end address
  THUMB(		.thumb			)
Signature · 1 line
1.7.10.4
#42 lore raw

PATCH 12/17 ARM: fix magic for bootloader in BE8

Quoted 6 lines · Ben Dooks

If the boot-loader is running little endian and the kernel is compiled big endian then it will not verify the magic number of the front of the kernel.

Note, do we need to also reverse the two values after the magic, or are they used later on in the kernel decompressor?

Those value instances are not used by the code. So for consistency they would have to be reversed, yes. Given those are variable and determined by the linker, I don't know how you could manage it.

Quoted 30 lines
---
 arch/arm/boot/compressed/head.S |    4 ++++
 1 file changed, 4 insertions(+)
arch/arm/boot/compressed/head.S +4 -0
index 39940a7..b6c7ec8 100644
@@ -129,7 +129,11 @@ start:
  THUMB(		adr	r12, BSYM(1f)	)
  THUMB(		bx	r12		)
 
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+		.word	0x18286f01
+#else
 		.word	0x016f2818		@ Magic numbers to help the loader
+#endif
 		.word	start			@ absolute load/run zImage address
 		.word	_edata			@ zImage end address
  THUMB(		.thumb			)
Signature · 7 lines
1.7.10.4


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#43 lore raw

PATCH 12/17 ARM: fix magic for bootloader in BE8

Quoted 6 lines · Nicolas Pitre
Quoted 4 lines · Ben Dooks

If the boot-loader is running little endian and the kernel is compiled big endian then it will not verify the magic number of the front of the kernel.

Quoted 6 lines
Quoted 2 lines

Note, do we need to also reverse the two values after the magic, or are they used later on in the kernel decompressor?

Those value instances are not used by the code. So for consistency they would have to be reversed, yes. Given those are variable and determined by the linker, I don't know how you could manage it.

Which is why I wish we never had them, they've got in the way before and they'll continue to do so. I've seen boot loaders which do precisely the wrong thing with these too, and had to fight them for doing so.

#44 lore raw

PATCH 12/17 ARM: fix magic for bootloader in BE8

Quoted 10 lines · Russell King - ARM Linux
Quoted 9 lines · Nicolas Pitre
Quoted 6 lines · Ben Dooks

If the boot-loader is running little endian and the kernel is compiled big endian then it will not verify the magic number of the front of the kernel.

Note, do we need to also reverse the two values after the magic, or are they used later on in the kernel decompressor?

Quoted 8 lines
Quoted 3 lines

Those value instances are not used by the code. So for consistency they would have to be reversed, yes. Given those are variable and determined by the linker, I don't know how you could manage it.

Which is why I wish we never had them, they've got in the way before and they'll continue to do so. I've seen boot loaders which do precisely the wrong thing with these too, and had to fight them for doing so.

in barebox we detect the type of file and detect we boot a zImage so be able to detect we need to boot a be zImage is really appreciate

so we do not ask user how to boot but handle it automaticly

Best Regards,
J.

author
#45 lore raw

PATCH 12/17 ARM: fix magic for bootloader in BE8

Quoted 19 lines · Jean-Christophe PLAGNIOL-VILLARD
Quoted 10 lines · Russell King - ARM Linux
Quoted 9 lines · Nicolas Pitre
Quoted 6 lines · Ben Dooks

If the boot-loader is running little endian and the kernel is compiled big endian then it will not verify the magic number of the front of the kernel.

Note, do we need to also reverse the two values after the magic, or are they used later on in the kernel decompressor?

Quoted 8 lines
Quoted 3 lines

Those value instances are not used by the code. So for consistency they would have to be reversed, yes. Given those are variable and determined by the linker, I don't know how you could manage it.

Which is why I wish we never had them, they've got in the way before and they'll continue to do so. I've seen boot loaders which do precisely the wrong thing with these too, and had to fight them for doing so.

Quoted 7 lines

in barebox we detect the type of file and detect we boot a zImage so be able to detect we need to boot a be zImage is really appreciate

so we do not ask user how to boot but handle it automaticly

Best Regards,
J.

IIRC, IXP is BE32, not BE8 so this shouldn't affect it.

I'm removing this from the base patch-set anyway and considering making it conditional on the ATAGS endian-ness.

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
#46 lore raw

PATCH 12/17 ARM: fix magic for bootloader in BE8

Quoted 23 lines · Ben Dooks
Quoted 22 lines · Jean-Christophe PLAGNIOL-VILLARD
Quoted 10 lines · Russell King - ARM Linux
Quoted 9 lines · Nicolas Pitre
Quoted 6 lines · Ben Dooks

If the boot-loader is running little endian and the kernel is compiled big endian then it will not verify the magic number of the front of the kernel.

Note, do we need to also reverse the two values after the magic, or are they used later on in the kernel decompressor?

Quoted 8 lines
Quoted 3 lines

Those value instances are not used by the code. So for consistency they would have to be reversed, yes. Given those are variable and determined by the linker, I don't know how you could manage it.

Which is why I wish we never had them, they've got in the way before and they'll continue to do so. I've seen boot loaders which do precisely the wrong thing with these too, and had to fight them for doing so.

in barebox we detect the type of file and detect we boot a zImage so be able to detect we need to boot a be zImage is really appreciate

Quoted 6 lines
Quoted 4 lines

so we do not ask user how to boot but handle it automaticly

Best Regards,
J.

IIRC, IXP is BE32, not BE8 so this shouldn't affect it.

yes IXP is BE32

Quoted 3 lines

I'm removing this from the base patch-set anyway and considering making it conditional on the ATAGS endian-ness.

we need to be able to detect it that's all

Best Regards,
J.

Quoted 4 lines
Signature · 3 lines
-- 
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
author
#47 lore raw

PATCH 12/17 ARM: fix magic for bootloader in BE8

In reply to Nicolas Pitre
Quoted 6 lines · Nicolas Pitre
Quoted 4 lines · Ben Dooks

If the boot-loader is running little endian and the kernel is compiled big endian then it will not verify the magic number of the front of the kernel.

Quoted 6 lines
Quoted 2 lines

Note, do we need to also reverse the two values after the magic, or are they used later on in the kernel decompressor?

Those value instances are not used by the code. So for consistency they would have to be reversed, yes. Given those are variable and determined by the linker, I don't know how you could manage it.

By breaking it down into parts I think...

I think this patch should be re-worked depending on the outcome of re-doing the whole ATAG endian configuration so that these are in the same endian-ness as we expect the ATAGs to be in.

Quoted 35 lines
Quoted 25 lines
---
  arch/arm/boot/compressed/head.S |    4 ++++
  1 file changed, 4 insertions(+)
arch/arm/boot/compressed/head.S +4 -1
index 39940a7..b6c7ec8 100644
@@ -129,7 +129,11 @@ start:
   THUMB(		adr	r12, BSYM(1f)	)
   THUMB(		bx	r12		)
 
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+		.word	0x18286f01
+#else
  		.word	0x016f2818		@ Magic numbers to help the loader
+#endif
  		.word	start			@ absolute load/run zImage address
  		.word	_edata			@ zImage end address
   THUMB(		.thumb			)
--

1.7.10.4

Quoted 5 lines

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
#48 lore raw

PATCH 12/17 ARM: fix magic for bootloader in BE8

Quoted 14 lines · Ben Dooks
Quoted 6 lines · Nicolas Pitre
Quoted 4 lines · Ben Dooks

If the boot-loader is running little endian and the kernel is compiled big endian then it will not verify the magic number of the front of the kernel.

Quoted 6 lines
Quoted 2 lines

Note, do we need to also reverse the two values after the magic, or are they used later on in the kernel decompressor?

Those value instances are not used by the code. So for consistency they would have to be reversed, yes. Given those are variable and determined by the linker, I don't know how you could manage it.

Quoted 5 lines

By breaking it down into parts I think...

I think this patch should be re-worked depending on the outcome of re-doing the whole ATAG endian configuration so that these are in the same endian-ness as we expect the ATAGs to be in.

I don't think this is related to ATAGS at all.

Those values are making a header for bootloaders to use in identifying an ARM Linux zImage binary. If you want to boot from a LE environment, the header must look like if this is a LE kernel.

Nicolas

author
#49 lore raw

PATCH 13/17 ARM: pl01x debug code endian fix

In reply to Ben Dooks

The PL01X debug code needs to take into account which endian mode the processor is running in. If it is big-endian, ensure the data is swapped appropriately.

Note, we could do this slightly more efficiently if we have an macro to do the necessary swap for the bits used by test.

---
 arch/arm/include/asm/hardware/debug-pl01x.S |    8 ++++++++
 1 file changed, 8 insertions(+)
arch/arm/include/asm/hardware/debug-pl01x.S +8 -0
index f9fd083..6f3a003 100644
@@ -12,18 +12,26 @@
 */
 #include <linux/amba/serial.h>
 
+#ifdef __ARMEB__
+#define ARM_BE(x...) x
+#else
+#define ARM_BE(x..)
+#endif
+
 		.macro	senduart,rd,rx
 		strb	\rd, [\rx, #UART01x_DR]
 		.endm
 
 		.macro	waituart,rd,rx
 1001:		ldr	\rd, [\rx, #UART01x_FR]
+  ARM_BE(	rev	\rd, \rd )
 		tst	\rd, #UART01x_FR_TXFF
 		bne	1001b
 		.endm
 
 		.macro	busyuart,rd,rx
 1001:		ldr	\rd, [\rx, #UART01x_FR]
+  ARM_BE(	rev	\rd, \rd )
 		tst	\rd, #UART01x_FR_BUSY
 		bne	1001b
 		.endm
Signature · 1 line
1.7.10.4
#50 lore raw

PATCH 13/17 ARM: pl01x debug code endian fix

arch/arm/include/asm/hardware/debug-pl01x.S line 18

Quoted 4 lines · Ben Dooks
arch/arm/include/asm/hardware/debug-pl01x.S +4 -0
+#ifdef __ARMEB__
+#define ARM_BE(x...) x
+#else
+#define ARM_BE(x..)

Buggy.

Quoted 1 line

+#endif

Rather than littering the code with stuff like this, define it in one central place please, and include it where necessary.

author
#51 lore raw

PATCH 13/17 ARM: pl01x debug code endian fix

Quoted 8 lines · Russell King - ARM Linux

arch/arm/include/asm/hardware/debug-pl01x.S line 18

Quoted 4 lines · Ben Dooks
arch/arm/include/asm/hardware/debug-pl01x.S +4 -0
+#ifdef __ARMEB__
+#define ARM_BE(x...) x
+#else
+#define ARM_BE(x..)

Buggy.

Quoted 4 lines
Quoted 1 line

+#endif

Rather than littering the code with stuff like this, define it in one central place please, and include it where necessary.

I'll add this to arch/arm/include/asm/assembler.h

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
author
#52 lore raw

PATCH 14/17 highbank: enable big-endian

In reply to Ben Dooks

Apart from a xgmac driver issue, the highbank seems to work correctly in big-endian mode. Allow the selection of big-endian in the system.

---
 arch/arm/mach-highbank/Kconfig |    1 +
 1 file changed, 1 insertion(+)
arch/arm/mach-highbank/Kconfig +1 -0
index 551c97e..9373571 100644
@@ -13,3 +13,4 @@ config ARCH_HIGHBANK
 	select HAVE_SMP
 	select SPARSE_IRQ
 	select USE_OF
+	select ARCH_SUPPORTS_BIG_ENDIAN
Signature · 1 line
1.7.10.4
#53 lore raw

PATCH 14/17 highbank: enable big-endian

Quoted 13 lines · Ben Dooks

Apart from a xgmac driver issue, the highbank seems to work correctly in big-endian mode. Allow the selection of big-endian in the system.

---
 arch/arm/mach-highbank/Kconfig |    1 +
 1 file changed, 1 insertion(+)
arch/arm/mach-highbank/Kconfig +0 -0
index 551c97e..9373571 100644

arch/arm/mach-highbank/Kconfig config ARCH_HIGHBANK · line 16

Quoted 4 lines
arch/arm/mach-highbank/Kconfig +1 -0
@@ -13,3 +13,4 @@ config ARCH_HIGHBANK
 	select HAVE_SMP
 	select SPARSE_IRQ
 	select USE_OF
+	select ARCH_SUPPORTS_BIG_ENDIAN

Considering that there is nothing highbank specific to enable BE support and all the support is within the A9 core (as BE support was never an intended feature), I wonder if this is really the right place for this option. All this really says is BE has been tested on this platform and presumably it would work on other platforms ignoring raw vs. relaxed accessors.

In any case:

Rob

author
#54 lore raw

PATCH 14/17 highbank: enable big-endian

Quoted 22 lines · Rob Herring
Quoted 15 lines · Ben Dooks

Apart from a xgmac driver issue, the highbank seems to work correctly in
big-endian mode. Allow the selection of big-endian in the system.

---
 arch/arm/mach-highbank/Kconfig |    1 +
 1 file changed, 1 insertion(+)
arch/arm/mach-highbank/Kconfig +0 -0
index 551c97e..9373571 100644

arch/arm/mach-highbank/Kconfig config ARCH_HIGHBANK · line 16

Quoted 4 lines
arch/arm/mach-highbank/Kconfig +1 -0
@@ -13,3 +13,4 @@ config ARCH_HIGHBANK
 	select HAVE_SMP
 	select SPARSE_IRQ
 	select USE_OF
+	select ARCH_SUPPORTS_BIG_ENDIAN

Considering that there is nothing highbank specific to enable BE

Quoted 8 lines

support
and all the support is within the A9 core (as BE support was never an intended feature), I wonder if this is really the right place for this
option. All this really says is BE has been tested on this platform and
presumably it would work on other platforms ignoring raw vs. relaxed accessors.

From reading previous ARM documentation, I believe that the big endian
is not a requirement for ARMv7 and can be compiled out of the core if the implementor does not want it. However, we've yet to find any of the cores out there with it removed.

As such, we still have to test on Freescale and OMAP devices.

If anyone else objects to this, or feels it would be better moved to the V7 cpu selection, or removed entirely then let me know.

Quoted 5 lines

In any case:

Rob

author
#55 lore raw

PATCH 15/17 xgmac: fix printing of hardware version

In reply to Ben Dooks

The current driver attempts to print netdev_info() before registering the network device and allowing the name to be set. Change this print to be after the network deviec has been registered.

Fix the following issue:

calxedaxgmac fff50000.ethernet (unregistered net_device): h/w version is 0x1012

---
 drivers/net/ethernet/calxeda/xgmac.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/net/ethernet/calxeda/xgmac.c +3 -3
index f7f0290..f91d9b2 100644
@@ -1715,9 +1715,6 @@ static int xgmac_probe(struct platform_device *pdev)
 		goto err_io;
 	}
 
-	uid = readl(priv->base + XGMAC_VERSION);
-	netdev_info(ndev, "h/w version is 0x%x\n", uid);
-
 	writel(0, priv->base + XGMAC_DMA_INTR_ENA);
 	ndev->irq = platform_get_irq(pdev, 0);
 	if (ndev->irq == -ENXIO) {
@@ -1771,6 +1768,9 @@ static int xgmac_probe(struct platform_device *pdev)
 	if (ret)
 		goto err_reg;
 
+	uid = readl(priv->base + XGMAC_VERSION);
+	netdev_info(ndev, "h/w version is 0x%x\n", uid);
+
 	return 0;
 
 err_reg:
Signature · 1 line
1.7.10.4
#56 lore raw

PATCH 15/17 xgmac: fix printing of hardware version

Quoted 3 lines · Ben Dooks

The current driver attempts to print netdev_info() before registering the network device and allowing the name to be set. Change this print to be after the network deviec has been registered.

typo: device

author
#57 lore raw

PATCH 15/17 xgmac: fix printing of hardware version

Quoted 8 lines · Russell King - ARM Linux
Quoted 5 lines · Ben Dooks

The current driver attempts to print netdev_info() before registering the
network device and allowing the name to be set. Change this print to be
after the network deviec has been registered.

typo: device

Thanks. I'll send this patch to the netdev list, forgot to remove it from
here.

Signature · 1 line
Ben
#58 lore raw

PATCH 15/17 xgmac: fix printing of hardware version

In reply to Ben Dooks
Quoted 10 lines · Ben Dooks

The current driver attempts to print netdev_info() before registering the network device and allowing the name to be set. Change this print to be after the network deviec has been registered.

Fix the following issue:

calxedaxgmac fff50000.ethernet (unregistered net_device): h/w version is 0x1012

---

Please post to netdev list.

Quoted 28 lines
 drivers/net/ethernet/calxeda/xgmac.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/net/ethernet/calxeda/xgmac.c +3 -3
index f7f0290..f91d9b2 100644
@@ -1715,9 +1715,6 @@ static int xgmac_probe(struct platform_device *pdev)
 		goto err_io;
 	}
 
-	uid = readl(priv->base + XGMAC_VERSION);
-	netdev_info(ndev, "h/w version is 0x%x\n", uid);
-
 	writel(0, priv->base + XGMAC_DMA_INTR_ENA);
 	ndev->irq = platform_get_irq(pdev, 0);
 	if (ndev->irq == -ENXIO) {
@@ -1771,6 +1768,9 @@ static int xgmac_probe(struct platform_device *pdev)
 	if (ret)
 		goto err_reg;
 
+	uid = readl(priv->base + XGMAC_VERSION);
+	netdev_info(ndev, "h/w version is 0x%x\n", uid);
+
 	return 0;
 
 err_reg:
author
#59 lore raw

PATCH 16/17 xgmac: replace __raw with relaxed IO

In reply to Ben Dooks

The xgmac driver is using __raw read and write functions, where it should be using readl_relaxed and writel_relaxed to avoid any issues with which endian mode the CPU is running in.

---
 drivers/net/ethernet/calxeda/xgmac.c |   12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
drivers/net/ethernet/calxeda/xgmac.c +6 -6
index f91d9b2..08a6f9e 100644
@@ -1202,7 +1202,7 @@ static int xgmac_poll(struct napi_struct *napi, int budget)
 
 	if (work_done < budget) {
 		napi_complete(napi);
-		__raw_writel(DMA_INTR_DEFAULT_MASK, priv->base + XGMAC_DMA_INTR_ENA);
+		writel_relaxed(DMA_INTR_DEFAULT_MASK, priv->base + XGMAC_DMA_INTR_ENA);
 	}
 	return work_done;
 }
@@ -1347,7 +1347,7 @@ static irqreturn_t xgmac_pmt_interrupt(int irq, void *dev_id)
 	struct xgmac_priv *priv = netdev_priv(dev);
 	void __iomem *ioaddr = priv->base;
 
-	intr_status = __raw_readl(ioaddr + XGMAC_INT_STAT);
+	intr_status = readl_relaxed(ioaddr + XGMAC_INT_STAT);
 	if (intr_status & XGMAC_INT_STAT_PMT) {
 		netdev_dbg(priv->dev, "received Magic frame\n");
 		/* clear the PMT bits 5 and 6 by reading the PMT */
@@ -1365,9 +1365,9 @@ static irqreturn_t xgmac_interrupt(int irq, void *dev_id)
 	struct xgmac_extra_stats *x = &priv->xstats;
 
 	/* read the status register (CSR5) */
-	intr_status = __raw_readl(priv->base + XGMAC_DMA_STATUS);
-	intr_status &= __raw_readl(priv->base + XGMAC_DMA_INTR_ENA);
-	__raw_writel(intr_status, priv->base + XGMAC_DMA_STATUS);
+	intr_status = readl_relaxed(priv->base + XGMAC_DMA_STATUS);
+	intr_status &= readl_relaxed(priv->base + XGMAC_DMA_INTR_ENA);
+	writel_relaxed(intr_status, priv->base + XGMAC_DMA_STATUS);
 
 	/* It displays the DMA process states (CSR5 register) */
 	/* ABNORMAL interrupts */
@@ -1403,7 +1403,7 @@ static irqreturn_t xgmac_interrupt(int irq, void *dev_id)
 
 	/* TX/RX NORMAL interrupts */
 	if (intr_status & (DMA_STATUS_RI | DMA_STATUS_TU | DMA_STATUS_TI)) {
-		__raw_writel(DMA_INTR_ABNORMAL, priv->base + XGMAC_DMA_INTR_ENA);
+		writel_relaxed(DMA_INTR_ABNORMAL, priv->base + XGMAC_DMA_INTR_ENA);
 		napi_schedule(&priv->napi);
 	}
 
Signature · 1 line
1.7.10.4
#60 lore raw

PATCH 16/17 xgmac: replace __raw with relaxed IO

Quoted 8 lines · Ben Dooks

The xgmac driver is using __raw read and write functions, where it should be using readl_relaxed and writel_relaxed to avoid any issues with which endian mode the CPU is running in.

---
 drivers/net/ethernet/calxeda/xgmac.c |   12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

This will break on arches which don't have the _relaxed variants. David Miller will reject restricting this to build on ARM only. See this:

http://www.spinics.net/lists/netdev/msg215457.html

Rob

Quoted 46 lines
drivers/net/ethernet/calxeda/xgmac.c +6 -6
index f91d9b2..08a6f9e 100644
@@ -1202,7 +1202,7 @@ static int xgmac_poll(struct napi_struct *napi, int budget)
 
 	if (work_done < budget) {
 		napi_complete(napi);
-		__raw_writel(DMA_INTR_DEFAULT_MASK, priv->base + XGMAC_DMA_INTR_ENA);
+		writel_relaxed(DMA_INTR_DEFAULT_MASK, priv->base + XGMAC_DMA_INTR_ENA);
 	}
 	return work_done;
 }
@@ -1347,7 +1347,7 @@ static irqreturn_t xgmac_pmt_interrupt(int irq, void *dev_id)
 	struct xgmac_priv *priv = netdev_priv(dev);
 	void __iomem *ioaddr = priv->base;
 
-	intr_status = __raw_readl(ioaddr + XGMAC_INT_STAT);
+	intr_status = readl_relaxed(ioaddr + XGMAC_INT_STAT);
 	if (intr_status & XGMAC_INT_STAT_PMT) {
 		netdev_dbg(priv->dev, "received Magic frame\n");
 		/* clear the PMT bits 5 and 6 by reading the PMT */
@@ -1365,9 +1365,9 @@ static irqreturn_t xgmac_interrupt(int irq, void *dev_id)
 	struct xgmac_extra_stats *x = &priv->xstats;
 
 	/* read the status register (CSR5) */
-	intr_status = __raw_readl(priv->base + XGMAC_DMA_STATUS);
-	intr_status &= __raw_readl(priv->base + XGMAC_DMA_INTR_ENA);
-	__raw_writel(intr_status, priv->base + XGMAC_DMA_STATUS);
+	intr_status = readl_relaxed(priv->base + XGMAC_DMA_STATUS);
+	intr_status &= readl_relaxed(priv->base + XGMAC_DMA_INTR_ENA);
+	writel_relaxed(intr_status, priv->base + XGMAC_DMA_STATUS);
 
 	/* It displays the DMA process states (CSR5 register) */
 	/* ABNORMAL interrupts */
@@ -1403,7 +1403,7 @@ static irqreturn_t xgmac_interrupt(int irq, void *dev_id)
 
 	/* TX/RX NORMAL interrupts */
 	if (intr_status & (DMA_STATUS_RI | DMA_STATUS_TU | DMA_STATUS_TI)) {
-		__raw_writel(DMA_INTR_ABNORMAL, priv->base + XGMAC_DMA_INTR_ENA);
+		writel_relaxed(DMA_INTR_ABNORMAL, priv->base + XGMAC_DMA_INTR_ENA);
 		napi_schedule(&priv->napi);
 	}
 
author
#61 lore raw

PATCH 16/17 xgmac: replace __raw with relaxed IO

Quoted 10 lines · Rob Herring
Quoted 8 lines · Ben Dooks

The xgmac driver is using __raw read and write functions, where it should be using readl_relaxed and writel_relaxed to avoid any issues with which endian mode the CPU is running in.

---
 drivers/net/ethernet/calxeda/xgmac.c |   12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
Quoted 5 lines

This will break on arches which don't have the _relaxed variants. David
Miller will reject restricting this to build on ARM only. See this:

http://www.spinics.net/lists/netdev/msg215457.html

I was considering changing it anyway to keep the __raw functions and simply swapping the endian-ness of the data when necessary as we don't need about three rev instructions in there where one will do.

Will sort out looking at this when I am back in the office on Monday.

Signature · 1 line
Ben
author
#62 lore raw

PATCH 17/17 mvebu: support running big-endian

In reply to Ben Dooks

Add indication we can run these cores in BE mode, and ensure that the secondary CPU is set to big-endian mode in the initialisation code as the initial code runs little-endian.

---
 arch/arm/mach-mvebu/Kconfig   |    1 +
 arch/arm/mach-mvebu/headsmp.S |    4 ++++
 2 files changed, 5 insertions(+)
arch/arm/mach-mvebu/Kconfig +1 -0
index 440b13e..2afa026 100644
@@ -1,5 +1,6 @@
 config ARCH_MVEBU
 	bool "Marvell SOCs with Device Tree support" if ARCH_MULTI_V7
+	select ARCH_SUPPORTS_BIG_ENDIAN
 	select CLKSRC_MMIO
 	select COMMON_CLK
 	select GENERIC_CLOCKEVENTS
arch/arm/mach-mvebu/headsmp.S +4 -0
index a06e0ed..8b09f8d 100644
@@ -36,6 +36,10 @@
  */
 ENTRY(armada_xp_secondary_startup)
 
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+	setend	be
+#endif
+
 	/* Read CPU id */
 	mrc     p15, 0, r1, c0, c0, 5
 	and     r1, r1, #0xF
Signature · 1 line
1.7.10.4
#63 lore raw
In reply to Ben Dooks

Dear Ben Dooks,

Quoted 3 lines · Ben Dooks

I have been working on getting big-endian kernels working, mainly from little-endian boot envrionments. The following patch series is what I have been working on, mainly on the highbank and axp systems.

Nice!

Quoted 1 line

Feedback and testing is welcome.

I've tested your patch set on Armada XP. The kernel boots fine, but it fails to bring up the secondary CPUs:

Booting CPU 1
CPU1: failed to come online
Booting CPU 2
CPU2: failed to come online
Booting CPU 3
CPU3: failed to come online
Brought up 1 CPUs
SMP: Total of 1 processors activated (1332.01 BogoMIPS).

The other problem is that I've not been able to build a working armeb userspace (I tried only very quickly with a Crosstool-NG toolchain I built for armeb). Do you have a known working armeb toolchain around?

Thanks,

Thomas

Signature · 4 lines
Thomas Petazzoni, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com
author
#64 lore raw
Quoted 7 lines · Thomas Petazzoni

Dear Ben Dooks,

Quoted 3 lines · Ben Dooks

I have been working on getting big-endian kernels working, mainly from little-endian boot envrionments. The following patch series is what I have been working on, mainly on the highbank and axp systems.

Quoted 6 lines

Nice!

Quoted 1 line

Feedback and testing is welcome.

I've tested your patch set on Armada XP. The kernel boots fine, but it fails to bring up the secondary CPUs:

Yes, missed updating the coherency_ll.S patch.

Attached new patch

Signature · 9 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
-------------- next part --------------
A non-text attachment was scrubbed...
Name: mvebu-be2.patch
Type: text/x-patch
Size: 1777 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20130212/01874b7e/attachment.bin>
#65 lore raw

Dear Ben Dooks,

Quoted 8 lines · Ben Dooks
Quoted 4 lines
Quoted 1 line

Feedback and testing is welcome.

I've tested your patch set on Armada XP. The kernel boots fine, but it fails to bring up the secondary CPUs:

Yes, missed updating the coherency_ll.S patch.

Attached new patch

Thanks, I'll test this one tomorrow. Any pointer for a known-working armeb toolchain?

Thanks,

Thomas

Signature · 4 lines
Thomas Petazzoni, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com
author
#66 lore raw
Quoted 9 lines · Thomas Petazzoni

Dear Ben Dooks,

Quoted 5 lines · Ben Dooks
Quoted 4 lines
Quoted 1 line

Feedback and testing is welcome.

I've tested your patch set on Armada XP. The kernel boots fine, but it fails to bring up the secondary CPUs:

Quoted 6 lines
Quoted 3 lines

Yes, missed updating the coherency_ll.S patch.

Attached new patch

Thanks, I'll test this one tomorrow. Any pointer for a known-working armeb toolchain?

I've been building with gcc 4.6.3

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
#67 lore raw
In reply to Ben Dooks

Dear Ben Dooks,

Quoted 6 lines · Ben Dooks
Quoted 2 lines

I've tested your patch set on Armada XP. The kernel boots fine, but it fails to bring up the secondary CPUs:

Yes, missed updating the coherency_ll.S patch.

Attached new patch

Thanks. Your patch did not build, because ARM_BE() isn't defined anywhere in coherency_ll.S. I've fixed this by using an ifdef CONFIG_CPU_BE8_BOOT_LE (updated patch attached), and now the secondary CPUs boot fine.

That said, I still haven't fixed my userspace, so I could only verify that the kernel boots, not more.

Thanks!

Thomas

Signature · 11 lines
Thomas Petazzoni, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-mvebu-support-running-big-endian.patch
Type: text/x-patch
Size: 2201 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20130213/995addba/attachment.bin>
#68 lore raw
In reply to Ben Dooks
Quoted 3 lines · Ben Dooks

I have been working on getting big-endian kernels working, mainly from little-endian boot envrionments. The following patch series is what I have been working on, mainly on the highbank and axp systems.

What is missing from this is the justification about why we need this additional pain, given that all the supporting userspaces today are all LE based.

Sure, I know that telcos have an endless love of big endian and don't understand anything else, but getting this working on non-telco socs seems to be a little odd.

author
#69 lore raw
Quoted 5 lines · Russell King - ARM Linux
Quoted 3 lines · Ben Dooks

I have been working on getting big-endian kernels working, mainly from little-endian boot envrionments. The following patch series is what I have been working on, mainly on the highbank and axp systems.

Quoted 7 lines

What is missing from this is the justification about why we need this additional pain, given that all the supporting userspaces today are all LE based.

Sure, I know that telcos have an endless love of big endian and don't understand anything else, but getting this working on non-telco socs seems to be a little odd.

Our problem is we have some code that works on a lot of big endian data but is not easy to re-build to work on ARM little endian. The current solution is to change to running the system big endian.

Unfortunately we cannot just run user-space big endian as the MMU is fetched in the same endian mode as the processor's data.

If we ignore the ATAG issue, then most of the support is currently in the kernel to do this and there's not a lot of changes. I think that some of the boot changes could be dealt with by a post-process of the zImage (such as the zImage magic) which would make this series less intrusive to the kernel.

I will send a new series in a day or so and see what people think.

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
#70 lore raw
Quoted 2 lines · Ben Dooks

If we ignore the ATAG issue, then most of the support is currently in the kernel to do this and there's not a lot of changes.

You could make CPU_BE8_BOOT_LE depend on !ATAG.

Quoted 3 lines

I think that some of the boot changes could be dealt with by a post-process of the zImage (such as the zImage magic) which would make this series less intrusive to the kernel.

No, please don't do that. Especially if the kernel binary _expects_ to be booted from a LE environment. Otherwise we might end up with kernels that don't boot because there are environment expectation mismatch, etc.

Here's what could be done for the zImage header:

arch/arm/boot/compressed/head.S +3 -3
index 4d5f4192ee..2077ebb734 100644
@@ -129,9 +129,9 @@ start:
  THUMB(		adr	r12, BSYM(1f)	)
  THUMB(		bx	r12		)
 
-		.word	0x016f2818		@ Magic numbers to help the loader
-		.word	start			@ absolute load/run zImage address
-		.word	_edata			@ zImage end address
+		.word	_magic_sig	@ Signature to help the loader (0x016f2818)
+		.word	_magic_start	@ zImage load/run address (0 if relocatable)
+		.word	_magic_end	@ zImage end address
  THUMB(		.thumb			)
 1:
 		mrs	r9, cpsr
arch/arm/boot/compressed/Makefile +4 -1
index a517153a13..52539f46c1 100644
@@ -200,8 +200,11 @@ CFLAGS_font.o := -Dstatic=
 $(obj)/font.c: $(FONTC)
 	$(call cmd,shipped)
 
-$(obj)/vmlinux.lds: $(obj)/vmlinux.lds.in arch/arm/boot/Makefile $(KCONFIG_CONFIG)
+$(obj)/vmlinux.lds.S: $(obj)/vmlinux.lds.in arch/arm/boot/Makefile $(KCONFIG_CONFIG)
 	@sed "$(SEDFLAGS)" < $< > $@
 
+$(obj)/vmlinux.lds: $(obj)/vmlinux.lds.S
+	$(call if_changed,cpp_lds_S)
+
 $(obj)/hyp-stub.S: $(srctree)/arch/$(SRCARCH)/kernel/hyp-stub.S
 	$(call cmd,shipped)
arch/arm/boot/compressed/vmlinux.lds.in +14 -0
index 4919f2ac8b..567250fb7d 100644
@@ -7,6 +7,16 @@
  * it under the terms of the GNU General Public License version 2 as
  * published by the Free Software Foundation.
  */
+
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+#define ZIMAGE_MAGIC(x) ( (((x) >> 24) & 0x000000ff) | \
+			  (((x) >>  8) & 0x0000ff00) | \
+			  (((x) <<  8) & 0x00ff0000) | \
+			  (((x) << 24) & 0xff000000) )
+#else
+#define ZIMAGE_MAGIC(x) (x)
+#endif
+
 OUTPUT_ARCH(arm)
 ENTRY(_start)
 SECTIONS
@@ -57,6 +67,10 @@ SECTIONS
   .pad			: { BYTE(0); . = ALIGN(8); }
   _edata = .;
 
+  _magic_sig = ZIMAGE_MAGIC(0x016f2818);
+  _magic_start = ZIMAGE_MAGIC(_start);
+  _magic_end = ZIMAGE_MAGIC(_edata);
+
   . = BSS_START;
   __bss_start = .;
   .bss			: { *(.bss) }
author
#71 lore raw
Quoted 6 lines · Nicolas Pitre
Quoted 2 lines · Ben Dooks

If we ignore the ATAG issue, then most of the support is currently in the kernel to do this and there's not a lot of changes.

You could make CPU_BE8_BOOT_LE depend on !ATAG.

That would be one consideration. For my next series I've removed the CPU_BE8_BOOT_LE and added the 'setend be' as conditional on running BE8 as this is such as small code addition it seems to make sense to do it that way.

I will put together a new series that deals with the endian-ness of the bootloader.

Quoted 10 lines
Quoted 3 lines

I think that some of the boot changes could be dealt with by a post-process of the zImage (such as the zImage magic) which would make this series less intrusive to the kernel.

No, please don't do that. Especially if the kernel binary _expects_ to be booted from a LE environment. Otherwise we might end up with kernels that don't boot because there are environment expectation mismatch, etc.

Here's what could be done for the zImage header:

I will keep this in mind, it looks like a pretty good solution to the whole issue.

Quoted 66 lines
arch/arm/boot/compressed/head.S +3 -3
index 4d5f4192ee..2077ebb734 100644
@@ -129,9 +129,9 @@ start:
   THUMB(		adr	r12, BSYM(1f)	)
   THUMB(		bx	r12		)
 
-		.word	0x016f2818		@ Magic numbers to help the loader
-		.word	start			@ absolute load/run zImage address
-		.word	_edata			@ zImage end address
+		.word	_magic_sig	@ Signature to help the loader (0x016f2818)
+		.word	_magic_start	@ zImage load/run address (0 if relocatable)
+		.word	_magic_end	@ zImage end address
   THUMB(		.thumb			)
  1:
  		mrs	r9, cpsr
arch/arm/boot/compressed/Makefile +4 -1
index a517153a13..52539f46c1 100644
@@ -200,8 +200,11 @@ CFLAGS_font.o := -Dstatic=
  $(obj)/font.c: $(FONTC)
  	$(call cmd,shipped)
 
-$(obj)/vmlinux.lds: $(obj)/vmlinux.lds.in arch/arm/boot/Makefile $(KCONFIG_CONFIG)
+$(obj)/vmlinux.lds.S: $(obj)/vmlinux.lds.in arch/arm/boot/Makefile $(KCONFIG_CONFIG)
  	@sed "$(SEDFLAGS)"<  $<  >  $@
 
+$(obj)/vmlinux.lds: $(obj)/vmlinux.lds.S
+	$(call if_changed,cpp_lds_S)
+
  $(obj)/hyp-stub.S: $(srctree)/arch/$(SRCARCH)/kernel/hyp-stub.S
  	$(call cmd,shipped)
arch/arm/boot/compressed/vmlinux.lds.in +14 -0
index 4919f2ac8b..567250fb7d 100644
@@ -7,6 +7,16 @@
   * it under the terms of the GNU General Public License version 2 as
   * published by the Free Software Foundation.
   */
+
+#ifdef CONFIG_CPU_BE8_BOOT_LE
+#define ZIMAGE_MAGIC(x) ( (((x)>>  24)&  0x000000ff) | \
+			  (((x)>>   8)&  0x0000ff00) | \
+			  (((x)<<   8)&  0x00ff0000) | \
+			  (((x)<<  24)&  0xff000000) )
+#else
+#define ZIMAGE_MAGIC(x) (x)
+#endif
+
  OUTPUT_ARCH(arm)
  ENTRY(_start)
  SECTIONS
@@ -57,6 +67,10 @@ SECTIONS
    .pad			: { BYTE(0); . = ALIGN(8); }
    _edata = .;
 
+  _magic_sig = ZIMAGE_MAGIC(0x016f2818);
+  _magic_start = ZIMAGE_MAGIC(_start);
+  _magic_end = ZIMAGE_MAGIC(_edata);
+
    . = BSS_START;
    __bss_start = .;
    .bss			: { *(.bss) }
Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
#72 lore raw
In reply to Ben Dooks

Ben Dooks a ?crit :

Quoted 13 lines · Russell King - ARM Linux
Quoted 11 lines · Ben Dooks
Quoted 3 lines

I have been working on getting big-endian kernels working, mainly from little-endian boot envrionments. The following patch series is what I have been working on, mainly on the highbank and axp systems.

What is missing from this is the justification about why we need this additional pain, given that all the supporting userspaces today are all LE based.

Sure, I know that telcos have an endless love of big endian and don't understand anything else, but getting this working on non-telco socs seems to be a little odd.

Quoted 7 lines

Our problem is we have some code that works on a lot of big endian data but is not easy to re-build to work on ARM little endian. The current solution is to change to running the system big endian.

Unfortunately we cannot just run user-space big endian as the MMU is fetched in the same endian mode as the processor's data.

Which architecture are you using ?

On armv7 the mmu endian mode is selected by SCTLR.EE [1], but userspace can change it's endian with setend instruction [2].

As the endian state is stored in cpsr, it should be preserved across context switch.

Matthieu

[1]
Exception Endianness. This bit defines the value of the CPSR.E bit on entry to an exception vector,
including reset. The possible values of this bit are:
0 Little-endian.
1 Big-endian.
This bit value also defines the endianness of the translation table data for translation table lookups.
In an implementation that includes the Security Extensions, this bit is Banked between the Secure
and Non-secure copies of the register.

[2]
The ARM and Thumb instruction sets both include an instruction to manipulate ENDIANSTATE:
SETEND BE Sets ENDIANSTATE to 1, for big-endian operation.
SETEND LE Sets ENDIANSTATE to 0, for little-endian operation.

author
#73 lore raw
Quoted 21 lines · Matthieu CASTET

Ben Dooks a ?crit :

Quoted 13 lines · Russell King - ARM Linux
Quoted 11 lines · Ben Dooks
Quoted 3 lines

I have been working on getting big-endian kernels working, mainly from little-endian boot envrionments. The following patch series is what I have been working on, mainly on the highbank and axp systems.

What is missing from this is the justification about why we need this additional pain, given that all the supporting userspaces today are all LE based.

Sure, I know that telcos have an endless love of big endian and don't understand anything else, but getting this working on non-telco socs seems to be a little odd.

Quoted 7 lines

Our problem is we have some code that works on a lot of big endian data but is not easy to re-build to work on ARM little endian. The current solution is to change to running the system big endian.

Unfortunately we cannot just run user-space big endian as the MMU is fetched in the same endian mode as the processor's data.

Quoted 6 lines

Which architecture are you using ?

On armv7 the mmu endian mode is selected by SCTLR.EE [1], but userspace can change it's endian with setend instruction [2].

As the endian state is stored in cpsr, it should be preserved across context switch.

Ah yes, I got confused as there is also an endian mode for the page tables that needs to match the endian mode that the kernel is doing loads/stores in.

However if we had BE userland with LE kernel then things are going to get difficult with a BE<>LE syscall issue.

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
#74 lore raw
Quoted 27 lines · Ben Dooks
Quoted 26 lines · Matthieu CASTET

Ben Dooks a ?crit :

Quoted 13 lines · Russell King - ARM Linux
Quoted 11 lines · Ben Dooks
Quoted 3 lines

I have been working on getting big-endian kernels working, mainly from little-endian boot envrionments. The following patch series is what I have been working on, mainly on the highbank and axp systems.

What is missing from this is the justification about why we need this additional pain, given that all the supporting userspaces today are all LE based.

Sure, I know that telcos have an endless love of big endian and don't understand anything else, but getting this working on non-telco socs seems to be a little odd.

Quoted 7 lines

Our problem is we have some code that works on a lot of big endian data but is not easy to re-build to work on ARM little endian. The current solution is to change to running the system big endian.

Unfortunately we cannot just run user-space big endian as the MMU is fetched in the same endian mode as the processor's data.

Which architecture are you using ?

On armv7 the mmu endian mode is selected by SCTLR.EE [1], but userspace can change it's endian with setend instruction [2].

Quoted 8 lines
Quoted 1 line

As the endian state is stored in cpsr, it should be preserved across context switch.

Ah yes, I got confused as there is also an endian mode for the page tables that needs to match the endian mode that the kernel is doing loads/stores in.

However if we had BE userland with LE kernel then things are going to get difficult with a BE<>LE syscall issue.

I had an (insane) idea once but no time to pursue. You can enable the 'compat' layer for a 32-bit ARM kernel and define all the compat_* types to be the same as the native ones. The compat layer has several handlers for syscalls which pretty much do the conversion between compat and native structures. The compat structures are read/written from/to user using get_user/put_user on each member. You then need change the ARM get_user/put_user code to test a new TIF_BE flag and do a 'rev' on the data. At this point the native kernel structures would have the correct little endianness.

I reckon the above would cover 70-80% of the syscalls. You need to chase other syscalls and update the binfmt_elf.c to cope with BE ELF files. Probably there are other issues as well.

Signature · 1 line
Catalin
author
#75 lore raw
Quoted 47 lines · Catalin Marinas
Quoted 27 lines · Ben Dooks
Quoted 26 lines · Matthieu CASTET

Ben Dooks a ?crit :

Quoted 13 lines · Russell King - ARM Linux
Quoted 11 lines · Ben Dooks
Quoted 3 lines

I have been working on getting big-endian kernels working, mainly from little-endian boot envrionments. The following patch series is what I have been working on, mainly on the highbank and axp systems.

What is missing from this is the justification about why we need this additional pain, given that all the supporting userspaces today are all LE based.

Sure, I know that telcos have an endless love of big endian and don't understand anything else, but getting this working on non-telco socs seems to be a little odd.

Quoted 7 lines

Our problem is we have some code that works on a lot of big endian data but is not easy to re-build to work on ARM little endian. The current solution is to change to running the system big endian.

Unfortunately we cannot just run user-space big endian as the MMU is fetched in the same endian mode as the processor's data.

Which architecture are you using ?

On armv7 the mmu endian mode is selected by SCTLR.EE [1], but userspace can change it's endian with setend instruction [2].

Quoted 8 lines
Quoted 1 line

As the endian state is stored in cpsr, it should be preserved across context switch.

Ah yes, I got confused as there is also an endian mode for the page tables that needs to match the endian mode that the kernel is doing loads/stores in.

However if we had BE userland with LE kernel then things are going to get difficult with a BE<>LE syscall issue.

I had an (insane) idea once but no time to pursue. You can enable the 'compat' layer for a 32-bit ARM kernel and define all the compat_* types to be the same as the native ones. The compat layer has several handlers for syscalls which pretty much do the conversion between compat and native structures. The compat structures are read/written from/to user using get_user/put_user on each member. You then need change the ARM get_user/put_user code to test a new TIF_BE flag and do a 'rev' on the data. At this point the native kernel structures would have the correct little endianness.

Quoted 3 lines

I reckon the above would cover 70-80% of the syscalls. You need to chase other syscalls and update the binfmt_elf.c to cope with BE ELF files. Probably there are other issues as well.

Yes, it is far more work than just running the entire system BE, but could be done. I however only have a budget of a beer, or maybe two beers if pushing it.

I might have a go at this if I have some spare time

Signature · 2 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius
#76 lore raw
In reply to Catalin Marinas
Quoted 11 lines · On Wednesday 13 February 2013 15:52:20 Catalin Marinas wrote:

I had an (insane) idea once but no time to pursue. You can enable the 'compat' layer for a 32-bit ARM kernel and define all the compat_* types to be the same as the native ones. The compat layer has several handlers for syscalls which pretty much do the conversion between compat and native structures. The compat structures are read/written from/to user using get_user/put_user on each member. You then need change the ARM get_user/put_user code to test a new TIF_BE flag and do a 'rev' on the data. At this point the native kernel structures would have the correct little endianness.

Quoted 3 lines

I reckon the above would cover 70-80% of the syscalls. You need to chase other syscalls and update the binfmt_elf.c to cope with BE ELF files. Probably there are other issues as well.

ioctl will be the biggest one by far. You would have to add compat handlers for every device driver and a lot of other things like network protocols.

Doable in theory, but also much more work than the 32 bit emulation on 64 bit kernels, and that was something that took a lot of work to get right.

However, qemu-user already has a mostly syscall emulation layer including ioctl that covers most of the common stuff, certainly enough to run most applications. The missing piece there is switching to native other-endian execution instead of interpreting the instructions. You might need to run the task in a separate process though and use ptrace to trap all system calls.

The part you cannot solve this way is IPC: any data that is shared between processes needs to match, and the only sane solution for that is probably to run a separate container for all the big-endian processes, and not let that talk to any of the little-endian tasks using things like AF_UNIX sockets or shared mmap.

Arnd