Hi,
Here is version 2 of fix to armv7-m build failure in sigreturn_codes.S. It is based on Dave's suggestion on [1]. Basically it set arch to arm4t explicitly if CONFIG_CPU_THUMBONLY is set. That enables both arm and thumb opcodes and code merged with the rest of image.
Version 1 [2] used conditional compilation.
Fix was tested linux-next with efm32_defconfig build (along with few other fixes) rmk-next BE/LE arndale build/boot and LTP rt_sigaction0? tests run
Dave, I've added your name with Suggested-by tag, please let me know if it is not OK with you, I'll remove it then.
Uwe, is it possible for you to test that this fix runs on efm32? Sorry, for multiple requests.
To address concern about fragility of proposed solution I looked binutils bfd/elf32-arm.c
https://sourceware.org/git/?p=binutils.git%3Ba=blob%3Bf=bfd/elf32-arm.c%3Bh=...
Attributes merge deals with couple things:
Tag_CPU_arch: tag_cpu_arch_combine function deals with it and from its tables it seems that v4t is compatible with any latter version and resulting value will come from latter version. I.e v4t and v7 (v7m) would merge fine.
Tag_CPU_arch_profile: since in case of '.arch armv4t' profile attribute is not generated it is merged fine with any other profile. Unlike in case of '.arch armv7a' and '.arch armv7m' profile values would be 'Application' and 'Microcontroller' and those conflict.
Above logic seems to be universal, so other linkers may follow it too. So it seems it is good to use '.arch armv4t' with armv7m code.
Thanks, Victor
[1] http://lists.infradead.org/pipermail/linux-arm-kernel/2013-November/210631.h... [2] http://lists.infradead.org/pipermail/linux-arm-kernel/2013-November/210393.h...
Victor Kamensky (1): ARM: signal: fix armv7-m build issue in sigreturn_codes.S
arch/arm/kernel/sigreturn_codes.S | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
In case of armv7-m architecture arm instructions are not allowed. For this architecture CONFIG_CPU_THUMBONLY is set. Let's explicitly set minimal architecture that allows both required thumb and arm opcodes. It is OK to do, since file as used as array of code snippets, which is indexed by signal.c code.
Suggested-by: Dave Martin Dave.Martin@arm.com Signed-off-by: Victor Kamensky victor.kamensky@linaro.org --- arch/arm/kernel/sigreturn_codes.S | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/arch/arm/kernel/sigreturn_codes.S b/arch/arm/kernel/sigreturn_codes.S index 3c5d0f2..081a041 100644 --- a/arch/arm/kernel/sigreturn_codes.S +++ b/arch/arm/kernel/sigreturn_codes.S @@ -30,12 +30,12 @@ * snippets. */
-#if __LINUX_ARM_ARCH__ <= 4 +#if (__LINUX_ARM_ARCH__ <= 4) || defined(CONFIG_CPU_THUMBONLY) /* * Note we manually set minimally required arch that supports - * required thumb opcodes for early arch versions. It is OK - * for this file to be used in combination with other - * lower arch variants, since these code snippets are only + * required thumb and arm opcodes for early arch versions or + * thumb only CPU. It is OK for this file to be used in combination + * with other arch variants, since these code snippets are only * used as input data. */ .arch armv4t
On Tue, Nov 12, 2013 at 10:57:46PM -0800, Victor Kamensky wrote:
Hi,
Here is version 2 of fix to armv7-m build failure in sigreturn_codes.S. It is based on Dave's suggestion on [1]. Basically it set arch to arm4t explicitly if CONFIG_CPU_THUMBONLY is set. That enables both arm and thumb opcodes and code merged with the rest of image.
Version 1 [2] used conditional compilation.
Fix was tested linux-next with efm32_defconfig build (along with few other fixes) rmk-next BE/LE arndale build/boot and LTP rt_sigaction0? tests run
Dave, I've added your name with Suggested-by tag, please let me know if it is not OK with you, I'll remove it then.
Uwe, is it possible for you to test that this fix runs on efm32? Sorry, for multiple requests.
To address concern about fragility of proposed solution I looked binutils bfd/elf32-arm.c
https://sourceware.org/git/?p=binutils.git%3Ba=blob%3Bf=bfd/elf32-arm.c%3Bh=...
Attributes merge deals with couple things:
Tag_CPU_arch: tag_cpu_arch_combine function deals with it and from its tables it seems that v4t is compatible with any latter version and resulting value will come from latter version. I.e v4t and v7 (v7m) would merge fine.
Tag_CPU_arch_profile: since in case of '.arch armv4t' profile attribute is not generated it is merged fine with any other profile. Unlike in case of '.arch armv7a' and '.arch armv7m' profile values would be 'Application' and 'Microcontroller' and those conflict.
Above logic seems to be universal, so other linkers may follow it too. So it seems it is good to use '.arch armv4t' with armv7m code.
[Apologies if you already saw a reply to this. I was writing one previously, but I think I never finished it. Anyway, here it is...]
After thinking about this, I'm not sure that this logic is really sound.
By allowing v7M objects to be linked together with ARM objects, ld creates objects with nonsensical attributes:
File Attributes Tag_CPU_name: "7-M" Tag_CPU_arch: v7 Tag_CPU_arch_profile: Microcontroller Tag_ARM_ISA_use: Yes Tag_THUMB_ISA_use: Thumb-2
i.e., a object cannot really be compatible with the microcontroller profile and also contain ARM instructions. v7-M is absolutely not a superset of v4T.
So, I think we are getting lucky here, and there's no real reason why a different linker (or future versions of GNU ld) should allow this.
The alternative is to use #ifdefs, and replace the ARM instructions and associates directives with suitable .space directives or nops in the ARM case.
This brings us back closer to your original suggestion, I guess.
Cheers ---Dave
Hi Dave,
Thank you for looking into this.
On 15 November 2013 11:03, Dave Martin Dave.Martin@arm.com wrote:
On Tue, Nov 12, 2013 at 10:57:46PM -0800, Victor Kamensky wrote:
Hi,
Here is version 2 of fix to armv7-m build failure in sigreturn_codes.S. It is based on Dave's suggestion on [1]. Basically it set arch to arm4t explicitly if CONFIG_CPU_THUMBONLY is set. That enables both arm and thumb opcodes and code merged with the rest of image.
Version 1 [2] used conditional compilation.
Fix was tested linux-next with efm32_defconfig build (along with few other fixes) rmk-next BE/LE arndale build/boot and LTP rt_sigaction0? tests run
Dave, I've added your name with Suggested-by tag, please let me know if it is not OK with you, I'll remove it then.
Uwe, is it possible for you to test that this fix runs on efm32? Sorry, for multiple requests.
To address concern about fragility of proposed solution I looked binutils bfd/elf32-arm.c
https://sourceware.org/git/?p=binutils.git%3Ba=blob%3Bf=bfd/elf32-arm.c%3Bh=...
Attributes merge deals with couple things:
Tag_CPU_arch: tag_cpu_arch_combine function deals with it and from its tables it seems that v4t is compatible with any latter version and resulting value will come from latter version. I.e v4t and v7 (v7m) would merge fine.
Tag_CPU_arch_profile: since in case of '.arch armv4t' profile attribute is not generated it is merged fine with any other profile. Unlike in case of '.arch armv7a' and '.arch armv7m' profile values would be 'Application' and 'Microcontroller' and those conflict.
Above logic seems to be universal, so other linkers may follow it too. So it seems it is good to use '.arch armv4t' with armv7m code.
[Apologies if you already saw a reply to this. I was writing one previously, but I think I never finished it. Anyway, here it is...]
After thinking about this, I'm not sure that this logic is really sound.
By allowing v7M objects to be linked together with ARM objects, ld creates objects with nonsensical attributes:
File Attributes Tag_CPU_name: "7-M" Tag_CPU_arch: v7 Tag_CPU_arch_profile: Microcontroller Tag_ARM_ISA_use: Yes Tag_THUMB_ISA_use: Thumb-2
i.e., a object cannot really be compatible with the microcontroller profile and also contain ARM instructions. v7-M is absolutely not a superset of v4T.
So, I think we are getting lucky here, and there's no real reason why a different linker (or future versions of GNU ld) should allow this.
The alternative is to use #ifdefs, and replace the ARM instructions and associates directives with suitable .space directives or nops in the ARM case.
This brings us back closer to your original suggestion, I guess.
I am OK with this. What are the next steps then?
Should we continue review of [1]? If you could take a look at would be great. Or should I resubmit the same patches as V3? [1] was already tested on armv7m by Uwe.
[1] http://lists.infradead.org/pipermail/linux-arm-kernel/2013-November/210393.h...
Thanks, Victor
Cheers ---Dave
On Fri, Nov 15, 2013 at 11:24:17AM -0800, Victor Kamensky wrote:
Hi Dave,
Thank you for looking into this.
On 15 November 2013 11:03, Dave Martin Dave.Martin@arm.com wrote:
On Tue, Nov 12, 2013 at 10:57:46PM -0800, Victor Kamensky wrote:
Hi,
Here is version 2 of fix to armv7-m build failure in sigreturn_codes.S. It is based on Dave's suggestion on [1]. Basically it set arch to arm4t explicitly if CONFIG_CPU_THUMBONLY is set. That enables both arm and thumb opcodes and code merged with the rest of image.
Version 1 [2] used conditional compilation.
Fix was tested linux-next with efm32_defconfig build (along with few other fixes) rmk-next BE/LE arndale build/boot and LTP rt_sigaction0? tests run
Dave, I've added your name with Suggested-by tag, please let me know if it is not OK with you, I'll remove it then.
Uwe, is it possible for you to test that this fix runs on efm32? Sorry, for multiple requests.
To address concern about fragility of proposed solution I looked binutils bfd/elf32-arm.c
https://sourceware.org/git/?p=binutils.git%3Ba=blob%3Bf=bfd/elf32-arm.c%3Bh=...
Attributes merge deals with couple things:
Tag_CPU_arch: tag_cpu_arch_combine function deals with it and from its tables it seems that v4t is compatible with any latter version and resulting value will come from latter version. I.e v4t and v7 (v7m) would merge fine.
Tag_CPU_arch_profile: since in case of '.arch armv4t' profile attribute is not generated it is merged fine with any other profile. Unlike in case of '.arch armv7a' and '.arch armv7m' profile values would be 'Application' and 'Microcontroller' and those conflict.
Above logic seems to be universal, so other linkers may follow it too. So it seems it is good to use '.arch armv4t' with armv7m code.
[Apologies if you already saw a reply to this. I was writing one previously, but I think I never finished it. Anyway, here it is...]
After thinking about this, I'm not sure that this logic is really sound.
By allowing v7M objects to be linked together with ARM objects, ld creates objects with nonsensical attributes:
File Attributes Tag_CPU_name: "7-M" Tag_CPU_arch: v7 Tag_CPU_arch_profile: Microcontroller Tag_ARM_ISA_use: Yes Tag_THUMB_ISA_use: Thumb-2
i.e., a object cannot really be compatible with the microcontroller profile and also contain ARM instructions. v7-M is absolutely not a superset of v4T.
So, I think we are getting lucky here, and there's no real reason why a different linker (or future versions of GNU ld) should allow this.
The alternative is to use #ifdefs, and replace the ARM instructions and associates directives with suitable .space directives or nops in the ARM case.
This brings us back closer to your original suggestion, I guess.
I am OK with this. What are the next steps then?
Should we continue review of [1]? If you could take a look at would be great. Or should I resubmit the same patches as V3? [1] was already tested on armv7m by Uwe.
[1] http://lists.infradead.org/pipermail/linux-arm-kernel/2013-November/210393.h...
It looks like we'll probably need to go back to something like that. I've commented on that thread again with a few new thoughts.
Cheers ---Dave
linaro-kernel@lists.linaro.org
-
Dave Martin -
Victor Kamensky