[build fail] v6.11-rc2 from "ARM: 9404/1: arm32: enable HAVE_LD_DEAD_CODE_DATA_ELIMINATION"
liuyuntao (F)
liuyuntao12 at huawei.com
Thu Aug 8 01:31:31 PDT 2024
On 2024/8/7 23:41, Arnd Bergmann wrote:
> On Wed, Aug 7, 2024, at 17:36, Harith George wrote:
>> On 07-08-2024 20:52, liuyuntao (F) wrote:
>>> Thanks, I reproduce the link error with toolchain
>>> gcc version 9.3.0
>>> GNU ld (GNU Binutils) 2.33.1
>>>
>>> with same gcc version, just upgrading ld version to 2.36.1, it does not
>>> segfault and build completes. there should be bugs in low version of ld,
>>> and the ".reloc .text, R_ARM_NONE, ." triggers that.
>>>
>> Thanks for confirming.
>>
>> I guess we need to add something like
>> #if !CONFIG_CC_IS_GCC || CONFIG_LD_VERSION >= 23600
>> around the entry-armv.S changes and maybe select
>> HAVE_LD_DEAD_CODE_DATA_ELIMINATION in arch/arm/Kconfig only if the same
>> conditions are met ??
>
> I think it makes most sense to have a minimum LD
> version as a dependency for HAVE_LD_DEAD_CODE_DATA_ELIMINATION.
> Are you sure that 2.36 is the first one that works, and it's
> not just 2.33 specifically that is broken?
>
> If so, we could use
>
> --- a/arch/arm/Kconfig
> +++ b/arch/arm/Kconfig
> @@ -117,7 +117,7 @@ config ARM
> select HAVE_KERNEL_XZ
> select HAVE_KPROBES if !XIP_KERNEL && !CPU_ENDIAN_BE32 && !CPU_V7M && !CPU_32v3
> select HAVE_KRETPROBES if HAVE_KPROBES
> - select HAVE_LD_DEAD_CODE_DATA_ELIMINATION
> + select HAVE_LD_DEAD_CODE_DATA_ELIMINATION if (LD_VERSION >= 23600 || LD_IS_LLD)
> select HAVE_MOD_ARCH_SPECIFIC
> select HAVE_NMI
> select HAVE_OPTPROBES if !THUMB2_KERNEL
>
>
> binutils only takes a few seconds to build from source, so
> you could just try all version from 2.25 (the oldest supported)
> to 2.36) to see which ones work.
After testing, all version of ld with version earlier than 2.36 has a
build issue.
as Harith mentioned:
> The select alone may not be enough. Wont the changes in arch/arm/kernel/entry-armv.S still result in LD Segfaults even if the HAVE_LD_DEAD_CODE_DATA_ELIMINATION flag is not set in .config for older toolchains?
I think we could eliminate the impact of ".reloc .text, R_ARM_NONE, ."
when CONFIG_LD_DEAD_CODE_DATA_ELIMINATION is not enabled.
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 54b2bb817a7f..173159e93c99 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -117,7 +117,7 @@ config ARM
select HAVE_KERNEL_XZ
select HAVE_KPROBES if !XIP_KERNEL && !CPU_ENDIAN_BE32 && !CPU_V7M
select HAVE_KRETPROBES if HAVE_KPROBES
- select HAVE_LD_DEAD_CODE_DATA_ELIMINATION
+ select HAVE_LD_DEAD_CODE_DATA_ELIMINATION if (LD_VERSION >=
23600 || LD_IS_LLD)
select HAVE_MOD_ARCH_SPECIFIC
select HAVE_NMI
select HAVE_OPTPROBES if !THUMB2_KERNEL
diff --git a/arch/arm/kernel/entry-armv.S b/arch/arm/kernel/entry-armv.S
index f01d23a220e6..cd443faf8645 100644
--- a/arch/arm/kernel/entry-armv.S
+++ b/arch/arm/kernel/entry-armv.S
@@ -29,6 +29,12 @@
#include "entry-header.S"
#include <asm/probes.h>
+#ifdef CONFIG_HAVE_LD_DEAD_CODE_DATA_ELIMINATION
+#define RELOC_TEXT_NONE (.reloc .text, R_ARM_NONE, .)
+#else
+#define RELOC_TEXT_NONE
+#endif
+
/*
* Interrupt handling.
*/
@@ -1065,7 +1071,7 @@ vector_addrexcptn:
.globl vector_fiq
.section .vectors, "ax", %progbits
- .reloc .text, R_ARM_NONE, .
+ RELOC_TEXT_NONE
W(b) vector_rst
W(b) vector_und
ARM( .reloc ., R_ARM_LDR_PC_G0, .L__vector_swi )
@@ -1079,7 +1085,7 @@ THUMB( .reloc ., R_ARM_THM_PC12,
.L__vector_swi )
#ifdef CONFIG_HARDEN_BRANCH_HISTORY
.section .vectors.bhb.loop8, "ax", %progbits
- .reloc .text, R_ARM_NONE, .
+ RELOC_TEXT_NONE
W(b) vector_rst
W(b) vector_bhb_loop8_und
ARM( .reloc ., R_ARM_LDR_PC_G0, .L__vector_bhb_loop8_swi )
@@ -1092,7 +1098,7 @@ THUMB( .reloc ., R_ARM_THM_PC12,
.L__vector_bhb_loop8_swi )
W(b) vector_bhb_loop8_fiq
.section .vectors.bhb.bpiall, "ax", %progbits
- .reloc .text, R_ARM_NONE, .
+ RELOC_TEXT_NONE
W(b) vector_rst
W(b) vector_bhb_bpiall_und
ARM( .reloc ., R_ARM_LDR_PC_G0, .L__vector_bhb_bpiall_swi )
More information about the linux-arm-kernel
mailing list