diff mbox

ARM: asm/unified.h: guard the IT macros against redefinition when using LTO

Message ID alpine.LFD.2.20.1507241739280.1806@knanqh.ubzr (mailing list archive)
State New, archived
Headers show

Commit Message

Nicolas Pitre July 24, 2015, 9:42 p.m. UTC
With LTO, the intermediate representation for each .c file contains those
global assembly macro definitions. When the lot is merged together during
the final link, we get duplicate definition errors:

[...]
/tmp/cc48fUMp.s:42352: Error: Macro `it' was already defined
/tmp/cc48fUMp.s:42354: Error: Macro `itt' was already defined
/tmp/cc48fUMp.s:42356: Error: Macro `ite' was already defined
[...]

Guard those against redefinitions.

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

Comments

Dave Martin July 27, 2015, 11:16 a.m. UTC | #1
On Fri, Jul 24, 2015 at 05:42:15PM -0400, Nicolas Pitre wrote:
> 
> With LTO, the intermediate representation for each .c file contains those
> global assembly macro definitions. When the lot is merged together during
> the final link, we get duplicate definition errors:
> 
> [...]
> /tmp/cc48fUMp.s:42352: Error: Macro `it' was already defined
> /tmp/cc48fUMp.s:42354: Error: Macro `itt' was already defined
> /tmp/cc48fUMp.s:42356: Error: Macro `ite' was already defined
> [...]
> 
> Guard those against redefinitions.
> 
> Signed-off-by: Nicolas Pitre <nico@linaro.org>

Reviewed-by: Dave Martin <Dave.Martin@arm.com>

I have a vague memory of hitting this before, but LTO had other problems
at the time...

Another scary thing with LTO is asms that use directives like
.arch_extension, .fpu etc. that will stay active for the rest of the LTO
unit of nothing is done about it.

I can't remember whether that was resolved somehow.

Cheers
---Dave
 
> diff --git a/arch/arm/include/asm/unified.h b/arch/arm/include/asm/unified.h
> index a91ae49961..ee257a3de7 100644
> --- a/arch/arm/include/asm/unified.h
> +++ b/arch/arm/include/asm/unified.h
> @@ -103,6 +103,8 @@
>  	.endm
>  #else	/* !__ASSEMBLY__ */
>  __asm__(
> +/* the .L_it_macros_defined guards against redefinition in the LTO case */
> +"	.ifndef	.L_it_macros_defined\n"
>  "	.macro	it, cond\n"
>  "	.endm\n"
>  "	.macro	itt, cond\n"
> @@ -132,7 +134,9 @@ __asm__(
>  "	.macro	iteet, cond\n"
>  "	.endm\n"
>  "	.macro	iteee, cond\n"
> -"	.endm\n");
> +"	.endm\n"
> +"	.set	.L_it_macros_defined,1\n"
> +"	.endif");
>  #endif	/* __ASSEMBLY__ */
>  
>  #endif	/* CONFIG_ARM_ASM_UNIFIED */
> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Russell King - ARM Linux July 27, 2015, 11:21 a.m. UTC | #2
On Mon, Jul 27, 2015 at 12:16:15PM +0100, Dave Martin wrote:
> Another scary thing with LTO is asms that use directives like
> .arch_extension, .fpu etc. that will stay active for the rest of the LTO
> unit of nothing is done about it.

That would be a problem, especially for things like the neon support,
as we specifically want stuff like that to be built separately.  The
same probably applies where we want to build with different architecture
options.

So, maybe the answer is to build most of the kernel with LTO enabled, but
those bits which require explicit -march= or -mfpu= options need to remain
separate non-LTO-able objects.
diff mbox

Patch

diff --git a/arch/arm/include/asm/unified.h b/arch/arm/include/asm/unified.h
index a91ae49961..ee257a3de7 100644
--- a/arch/arm/include/asm/unified.h
+++ b/arch/arm/include/asm/unified.h
@@ -103,6 +103,8 @@ 
 	.endm
 #else	/* !__ASSEMBLY__ */
 __asm__(
+/* the .L_it_macros_defined guards against redefinition in the LTO case */
+"	.ifndef	.L_it_macros_defined\n"
 "	.macro	it, cond\n"
 "	.endm\n"
 "	.macro	itt, cond\n"
@@ -132,7 +134,9 @@  __asm__(
 "	.macro	iteet, cond\n"
 "	.endm\n"
 "	.macro	iteee, cond\n"
-"	.endm\n");
+"	.endm\n"
+"	.set	.L_it_macros_defined,1\n"
+"	.endif");
 #endif	/* __ASSEMBLY__ */
 
 #endif	/* CONFIG_ARM_ASM_UNIFIED */