ARM: PJ4: allow building in Thumb-2 mode

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Signed-off-by Ard Biesheuvel <[email protected]>

---
 arch/arm/kernel/Makefile |  1 +
 arch/arm/kernel/iwmmxt.S | 13 +++++++++++++
 2 files changed, 14 insertions(+)

arch/arm/kernel/Makefile +1 -0

index 4c5d1a5982dd..575a5e424b0a 100644
@@ -85,6 +85,7 @@ obj-$(CONFIG_CPU_PJ4B)		+= pj4-cp0.o
 obj-$(CONFIG_IWMMXT)		+= iwmmxt.o
 obj-$(CONFIG_PERF_EVENTS)	+= perf_regs.o
 obj-$(CONFIG_HW_PERF_EVENTS)	+= perf_event.o perf_event_cpu.o
+CFLAGS_pj4-cp0.o		:= -marm
 AFLAGS_iwmmxt.o			:= -Wa,-mcpu=iwmmxt
 obj-$(CONFIG_ARM_CPU_TOPOLOGY)  += topology.o
 

arch/arm/kernel/iwmmxt.S +13 -0

index ad58e565fe98..49fadbda8c63 100644
@@ -58,6 +58,7 @@
 #define MMX_SIZE		(0x98)
 
 	.text
+	.arm
 
 /*
  * Lazy switching of Concan coprocessor context

@@ -182,6 +183,8 @@ concan_load:
 	tmcr	wCon, r2
 	ret	lr
 
+ENDPROC(iwmmxt_task_enable)
+
 /*
  * Back up Concan regs to save area and disable access to them
  * (mainly for gdb or sleep mode usage)

@@ -232,6 +235,8 @@ ENTRY(iwmmxt_task_disable)
 1:	msr	cpsr_c, ip			@ restore interrupt mode
 	ldmfd	sp!, {r4, pc}
 
+ENDPROC(iwmmxt_task_disable)
+
 /*
  * Copy Concan state to given memory address
  *

@@ -268,6 +273,8 @@ ENTRY(iwmmxt_task_copy)
 	msr	cpsr_c, ip			@ restore interrupt mode
 	ret	r3
 
+ENDPROC(iwmmxt_task_copy)
+
 /*
  * Restore Concan state from given memory address
  *

@@ -304,6 +311,8 @@ ENTRY(iwmmxt_task_restore)
 	msr	cpsr_c, ip			@ restore interrupt mode
 	ret	r3
 
+ENDPROC(iwmmxt_task_restore)
+
 /*
  * Concan handling on task switch
  *

@@ -335,6 +344,8 @@ ENTRY(iwmmxt_task_switch)
 	mrc	p15, 0, r1, c2, c0, 0
 	sub	pc, lr, r1, lsr #32		@ cpwait and return
 
+ENDPROC(iwmmxt_task_switch)
+
 /*
  * Remove Concan ownership of given task
  *

@@ -353,6 +364,8 @@ ENTRY(iwmmxt_task_release)
 	msr	cpsr_c, r2			@ restore interrupts
 	ret	lr
 
+ENDPROC(iwmmxt_task_release)
+
 	.data
 concan_owner:
 	.word	0

Signature · 1 line

1.8.3.2

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Signed-off-by Ard Biesheuvel <[email protected]>

Quoted 9 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Quoted 1 line

Signed-off-by Ard Biesheuvel <[email protected]>

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Have you tested this on a machine that has IWMMXT enabled?

Quoted 12 lines · Arnd Bergmann

Quoted 10 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Signed-off-by Ard Biesheuvel <[email protected]>

Quoted 5 lines

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Have you tested this on a machine that has IWMMXT enabled?

No, I don't have access to such a machine, unfortunately.

Quoted 16 lines · On Monday 24 November 2014 16:34:47 Ard Biesheuvel wrote:

Quoted 15 lines · Arnd Bergmann

Quoted 10 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Signed-off-by Ard Biesheuvel <[email protected]>

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Quoted 4 lines

Quoted 2 lines

Have you tested this on a machine that has IWMMXT enabled?

No, I don't have access to such a machine, unfortunately.

Adding a few mvebu folks to Cc, maybe one of them can test your patch.

It's also possible that Olof or Kevin have a PJ4 machine with iwmmxt in their boot farms.

Arnd

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

Hi Arnd,

Quoted 20 lines · Arnd Bergmann

Quoted 19 lines · On Monday 24 November 2014 16:34:47 Ard Biesheuvel wrote:

Quoted 17 lines · Arnd Bergmann

Quoted 10 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Signed-off-by Ard Biesheuvel <[email protected]>

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Have you tested this on a machine that has IWMMXT enabled?

Quoted 3 lines

Quoted 1 line

No, I don't have access to such a machine, unfortunately.

Adding a few mvebu folks to Cc, maybe one of them can test your patch.

Actually even it is a feature of the PJ4 machine non of the mvebu machines currently use it. I only see this configuration enabled for the pxa family. So I think you would have more feedback with the pxa maintainers/owners.

Quoted 25 lines · Gregory CLEMENT

Hi Arnd,

Quoted 22 lines · Arnd Bergmann

Quoted 20 lines · On Monday 24 November 2014 16:34:47 Ard Biesheuvel wrote:

Quoted 10 lines · Arnd Bergmann

Quoted 9 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Quoted 7 lines

Quoted 1 line

Signed-off-by Ard Biesheuvel <[email protected]>

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Have you tested this on a machine that has IWMMXT enabled?

No, I don't have access to such a machine, unfortunately.

Quoted 6 lines

Quoted 1 line

Adding a few mvebu folks to Cc, maybe one of them can test your patch.

Actually even it is a feature of the PJ4 machine non of the mvebu machines currently use it. I only see this configuration enabled for the pxa family. So I think you would have more feedback with the pxa maintainers/owners.

Are you saying PJ4 should be dropped from 'default' here?

config IWMMXT

        bool "Enable iWMMXt support"
        depends on CPU_XSCALE || CPU_XSC3 || CPU_MOHAWK || CPU_PJ4 || CPU_PJ4B
        default y if PXA27x || PXA3xx || ARCH_MMP || CPU_PJ4 || CPU_PJ4B

Quoted 34 lines · Ard Biesheuvel

Quoted 31 lines · Gregory CLEMENT

Hi Arnd,

Quoted 17 lines · Arnd Bergmann

Quoted 16 lines · On Monday 24 November 2014 16:34:47 Ard Biesheuvel wrote:

Quoted 15 lines · Arnd Bergmann

Quoted 10 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Signed-off-by Ard Biesheuvel <[email protected]>

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Quoted 6 lines

Quoted 4 lines

Quoted 2 lines

Have you tested this on a machine that has IWMMXT enabled?

No, I don't have access to such a machine, unfortunately.

Adding a few mvebu folks to Cc, maybe one of them can test your patch.

Actually even it is a feature of the PJ4 machine non of the mvebu machines currently use it. I only see this configuration enabled for the pxa family. So I think you would have more feedback with the pxa maintainers/owners.

Quoted 1 line

Are you saying PJ4 should be dropped from 'default' here?

Not at all. I didn't find it in the configuration file, but I didn't realized it was because of this "default y".

Sorry for the noise.

Quoted 40 lines · Gregory CLEMENT

Quoted 27 lines · Ard Biesheuvel

Quoted 25 lines · Gregory CLEMENT

Hi Arnd,

Quoted 22 lines · Arnd Bergmann

Quoted 20 lines · On Monday 24 November 2014 16:34:47 Ard Biesheuvel wrote:

Quoted 10 lines · Arnd Bergmann

Quoted 9 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Quoted 7 lines

Quoted 1 line

Signed-off-by Ard Biesheuvel <[email protected]>

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Have you tested this on a machine that has IWMMXT enabled?

No, I don't have access to such a machine, unfortunately.

Quoted 8 lines

Quoted 6 lines

Quoted 1 line

Adding a few mvebu folks to Cc, maybe one of them can test your patch.

Actually even it is a feature of the PJ4 machine non of the mvebu machines currently use it. I only see this configuration enabled for the pxa family. So I think you would have more feedback with the pxa maintainers/owners.

Are you saying PJ4 should be dropped from 'default' here?

Not at all. I didn't find it in the configuration file, but I didn't realized it was because of this "default y".

Quoted 2 lines

Sorry for the noise.

Ah, ok.

In that case, could you perhaps also test this patch on a PJ4 machine with multi_v7_defconfig but with CONFIG_THUMB2_KERNEL enabled?

Olof has already done so successfully, but just to be sure.

Thanks,
Ard.

Quoted 42 lines · Ard Biesheuvel

Quoted 40 lines · Gregory CLEMENT

Quoted 27 lines · Ard Biesheuvel

Quoted 25 lines · Gregory CLEMENT

Hi Arnd,

Quoted 22 lines · Arnd Bergmann

Quoted 20 lines · On Monday 24 November 2014 16:34:47 Ard Biesheuvel wrote:

Quoted 10 lines · Arnd Bergmann

Quoted 9 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Quoted 7 lines

Quoted 1 line

Signed-off-by Ard Biesheuvel <[email protected]>

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Have you tested this on a machine that has IWMMXT enabled?

No, I don't have access to such a machine, unfortunately.

Quoted 8 lines

Quoted 6 lines

Quoted 1 line

Adding a few mvebu folks to Cc, maybe one of them can test your patch.

Actually even it is a feature of the PJ4 machine non of the mvebu machines currently use it. I only see this configuration enabled for the pxa family. So I think you would have more feedback with the pxa maintainers/owners.

Are you saying PJ4 should be dropped from 'default' here?

Not at all. I didn't find it in the configuration file, but I didn't realized it was because of this "default y".

Quoted 7 lines

Quoted 2 lines

Sorry for the noise.

Ah, ok.

In that case, could you perhaps also test this patch on a PJ4 machine with multi_v7_defconfig but with CONFIG_THUMB2_KERNEL enabled?

Actually as explained in the commit "ARM: 8040/1: pj4: properly detect existence of the iWMMXt co-processor", the Armada 370 and Armada XP have a PJ4B CPU without the iWMMXt extension. The only boards I have using CPus belonging to the PJ4 family use these SoCs so I wasn't able to fully tested it. At least I can say that these SoC are now buildable in Thumb mode with this patch and I didn't see any regression while booting the system compiled in Thumb2.

Quoted 16 lines · On Monday 24 November 2014 16:34:47 Ard Biesheuvel wrote:

Quoted 15 lines · Arnd Bergmann

Quoted 10 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Signed-off-by Ard Biesheuvel <[email protected]>

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Quoted 4 lines

Quoted 2 lines

Have you tested this on a machine that has IWMMXT enabled?

No, I don't have access to such a machine, unfortunately.

Adding a few mvebu folks to Cc, maybe one of them can test your patch. It's also possible that Olof or Kevin have a PJ4 machine with iwmmxt in their boot farms.

Quoted 17 lines · Arnd Bergmann

Quoted 16 lines · On Monday 24 November 2014 16:34:47 Ard Biesheuvel wrote:

Quoted 15 lines · Arnd Bergmann

Quoted 10 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Signed-off-by Ard Biesheuvel <[email protected]>

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Quoted 8 lines

Quoted 4 lines

Quoted 2 lines

Have you tested this on a machine that has IWMMXT enabled?

No, I don't have access to such a machine, unfortunately.

Adding a few mvebu folks to Cc, maybe one of them can test your patch. It's also possible that Olof or Kevin have a PJ4 machine with iwmmxt in their boot farms.

I have a Cubox with Dove, which does implement iWMMXt v2, based on boot messages.

I however do not have any kind of userspace programs that make use of it to test with.

Quoted 30 lines · Olof Johansson

Quoted 17 lines · Arnd Bergmann

Quoted 16 lines · On Monday 24 November 2014 16:34:47 Ard Biesheuvel wrote:

Quoted 15 lines · Arnd Bergmann

Quoted 10 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Signed-off-by Ard Biesheuvel <[email protected]>

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Quoted 8 lines

Quoted 4 lines

Quoted 2 lines

Have you tested this on a machine that has IWMMXT enabled?

No, I don't have access to such a machine, unfortunately.

Adding a few mvebu folks to Cc, maybe one of them can test your patch. It's also possible that Olof or Kevin have a PJ4 machine with iwmmxt in their boot farms.

I have a Cubox with Dove, which does implement iWMMXt v2, based on boot messages.

Quoted 3 lines

I however do not have any kind of userspace programs that make use of it to test with.

Well, I think the idea is to make sure it doesn't explode when running multi_v7_defconfig in thumb2 mode. That requires a machine that has the actual coprocessor, but as far as I can figure out from the code, this should not require any actual floating point use.

Quoted 31 lines · Ard Biesheuvel

Quoted 30 lines · Olof Johansson

Quoted 17 lines · Arnd Bergmann

Quoted 16 lines · On Monday 24 November 2014 16:34:47 Ard Biesheuvel wrote:

Quoted 15 lines · Arnd Bergmann

Quoted 10 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Signed-off-by Ard Biesheuvel <[email protected]>

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Quoted 8 lines

Quoted 4 lines

Quoted 2 lines

Have you tested this on a machine that has IWMMXT enabled?

No, I don't have access to such a machine, unfortunately.

Adding a few mvebu folks to Cc, maybe one of them can test your patch. It's also possible that Olof or Kevin have a PJ4 machine with iwmmxt in their boot farms.

I have a Cubox with Dove, which does implement iWMMXt v2, based on boot messages.

Quoted 8 lines

Quoted 3 lines

I however do not have any kind of userspace programs that make use of it to test with.

Well, I think the idea is to make sure it doesn't explode when running multi_v7_defconfig in thumb2 mode. That requires a machine that has the actual coprocessor, but as far as I can figure out from the code, this should not require any actual floating point use.

Well, that's easy to confirm, and I've done so now.

Not sure I'd claim that's worth a tested-by though, but whatever:

Tested-by Olof Johansson <[email protected]>

Quoted 17 lines · Arnd Bergmann

Quoted 16 lines · On Monday 24 November 2014 16:34:47 Ard Biesheuvel wrote:

Quoted 15 lines · Arnd Bergmann

Quoted 10 lines · On Monday 24 November 2014 12:56:40 Ard Biesheuvel wrote:

Two files that get included when building the multi_v7_defconfig target fail to build when selecting THUMB2_KERNEL for this configuration.

In both cases, we can just build the file as ARM code, as none of its symbols are exported to modules, so there are no interworking concerns. In the iwmmxt.S case, add ENDPROC() declarations so the symbols are annotated as functions, resulting in the linker to emit the appropriate mode switches.

Signed-off-by Ard Biesheuvel <[email protected]>

Ah, very nice. I tried this before, but my version didn't actually work, presumably because I didn't know about the ENDPROC() stuff.

Quoted 8 lines

Quoted 4 lines

Quoted 2 lines

Have you tested this on a machine that has IWMMXT enabled?

No, I don't have access to such a machine, unfortunately.

Adding a few mvebu folks to Cc, maybe one of them can test your patch. It's also possible that Olof or Kevin have a PJ4 machine with iwmmxt in their boot farms.

I have a Cubox with Dove, which does implement iWMMXt v2, based on boot messages.

I however do not have any kind of userspace programs that make use of it to test with.

From the looks of it I wouldn't see any problem... but nothing beats