2017-03-08 03:08:29

by David Daney

Subject: [PATCH] module: set .init_array alignment to 8

The proper idiom for aligning linker sections in modules is different
than for built-in sections. ". = ALIGN();" followed by a forced
output address of 0 does nothing, as forcing the address changes the
value of ".".

Use output section alignment specifier instead.

Fixes: 9ddf82521c86 ("kernel: add support for .init_array.* constructors")
Signed-off-by: David Daney <[email protected]>
---

I noticed this when doing the __jump_table thing. Doesn't seem to
break a defconfig build, but otherwise untested.

scripts/module-common.lds | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/scripts/module-common.lds b/scripts/module-common.lds
index 9b6e246..d61b9e8 100644
--- a/scripts/module-common.lds
+++ b/scripts/module-common.lds
@@ -20,8 +20,7 @@ SECTIONS {
__kcrctab_unused_gpl 0 : { *(SORT(___kcrctab_unused_gpl+*)) }
__kcrctab_gpl_future 0 : { *(SORT(___kcrctab_gpl_future+*)) }

- . = ALIGN(8);
- .init_array 0 : { *(SORT(.init_array.*)) *(.init_array) }
+ .init_array 0 : ALIGN(8) { *(SORT(.init_array.*)) *(.init_array) }

__jump_table 0 : ALIGN(8) { KEEP(*(__jump_table)) }
}
--
2.9.3

2017-03-08 01:36:44

by David Daney

Subject: Re: [PATCH] module: set .init_array alignment to 8

It looks like Andrey's e-mail changed. Replying with what I hope is a
valid e-mail for Andrey.

On 03/07/2017 05:31 PM, David Daney wrote:
> The proper idiom for aligning linker sections in modules is different
> than for built-in sections. ". = ALIGN();" followed by a forced
> output address of 0 does nothing, as forcing the address changes the
> value of ".".
>
> Use output section alignment specifier instead.
>
> Fixes: 9ddf82521c86 ("kernel: add support for .init_array.* constructors")
> Signed-off-by: David Daney <[email protected]>
> ---
>
> I noticed this when doing the __jump_table thing. Doesn't seem to
> break a defconfig build, but otherwise untested.
>
> scripts/module-common.lds | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/scripts/module-common.lds b/scripts/module-common.lds
> index 9b6e246..d61b9e8 100644
> --- a/scripts/module-common.lds
> +++ b/scripts/module-common.lds
> @@ -20,8 +20,7 @@ SECTIONS {
> __kcrctab_unused_gpl 0 : { *(SORT(___kcrctab_unused_gpl+*)) }
> __kcrctab_gpl_future 0 : { *(SORT(___kcrctab_gpl_future+*)) }
>
> - . = ALIGN(8);
> - .init_array 0 : { *(SORT(.init_array.*)) *(.init_array) }
> + .init_array 0 : ALIGN(8) { *(SORT(.init_array.*)) *(.init_array) }
>
> __jump_table 0 : ALIGN(8) { KEEP(*(__jump_table)) }
> }
>

2017-03-09 14:26:22

by Andrey Ryabinin

Subject: Re: [PATCH] module: set .init_array alignment to 8

On 03/08/2017 04:31 AM, David Daney wrote:
> The proper idiom for aligning linker sections in modules is different
> than for built-in sections. ". = ALIGN();" followed by a forced
> output address of 0 does nothing, as forcing the address changes the
> value of ".".
>
> Use output section alignment specifier instead.
>
> Fixes: 9ddf82521c86 ("kernel: add support for .init_array.* constructors")
> Signed-off-by: David Daney <[email protected]>

Reviewed-by: Andrey Ryabinin <[email protected]>

> ---
>
> I noticed this when doing the __jump_table thing. Doesn't seem to
> break a defconfig build, but otherwise untested.
>
> scripts/module-common.lds | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/scripts/module-common.lds b/scripts/module-common.lds
> index 9b6e246..d61b9e8 100644
> --- a/scripts/module-common.lds
> +++ b/scripts/module-common.lds
> @@ -20,8 +20,7 @@ SECTIONS {
> __kcrctab_unused_gpl 0 : { *(SORT(___kcrctab_unused_gpl+*)) }
> __kcrctab_gpl_future 0 : { *(SORT(___kcrctab_gpl_future+*)) }
>
> - . = ALIGN(8);
> - .init_array 0 : { *(SORT(.init_array.*)) *(.init_array) }
> + .init_array 0 : ALIGN(8) { *(SORT(.init_array.*)) *(.init_array) }
>
> __jump_table 0 : ALIGN(8) { KEEP(*(__jump_table)) }
> }
>

2017-03-13 18:46:20

by Jessica Yu

Subject: Re: [PATCH] module: set .init_array alignment to 8

+++ David Daney [07/03/17 17:31 -0800]:
>The proper idiom for aligning linker sections in modules is different
>than for built-in sections. ". = ALIGN();" followed by a forced
>output address of 0 does nothing, as forcing the address changes the
>value of ".".
>
>Use output section alignment specifier instead.
>
>Fixes: 9ddf82521c86 ("kernel: add support for .init_array.* constructors")
>Signed-off-by: David Daney <[email protected]>

Good catch, thanks. I've applied this to modules-next.

Jessica

>I noticed this when doing the __jump_table thing. Doesn't seem to
>break a defconfig build, but otherwise untested.
>
> scripts/module-common.lds | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
>diff --git a/scripts/module-common.lds b/scripts/module-common.lds
>index 9b6e246..d61b9e8 100644
>--- a/scripts/module-common.lds
>+++ b/scripts/module-common.lds
>@@ -20,8 +20,7 @@ SECTIONS {
> __kcrctab_unused_gpl 0 : { *(SORT(___kcrctab_unused_gpl+*)) }
> __kcrctab_gpl_future 0 : { *(SORT(___kcrctab_gpl_future+*)) }
>
>- . = ALIGN(8);
>- .init_array 0 : { *(SORT(.init_array.*)) *(.init_array) }
>+ .init_array 0 : ALIGN(8) { *(SORT(.init_array.*)) *(.init_array) }
>
> __jump_table 0 : ALIGN(8) { KEEP(*(__jump_table)) }
> }
>--
>2.9.3
>