Re: [PATCH 1/2] asm-generic/atomic: Prefer __always_inline for wrappers

[Mark Rutland]

On Fri, Nov 22, 2019 at 04:42:20PM +0100, Marco Elver wrote:
> Prefer __always_inline for atomic wrappers. When building for size
> (CC_OPTIMIZE_FOR_SIZE), some compilers appear to be less inclined to
> inline even relatively small static inline functions that are assumed to
> be inlinable such as atomic ops. This can cause problems, for example in
> UACCESS regions.

>From looking at the link below, the problem is tat objtool isn't happy
about non-whiteliested calls within UACCESS regions.

Is that a problem here? are the kasan/kcsan calls whitelisted?

> By using __always_inline, we let the real implementation and not the
> wrapper determine the final inlining preference.

That sounds reasonable to me, assuming that doesn't end up significantly
bloating the kernel text. What impact does this have on code size?

> This came up when addressing UACCESS warnings with CC_OPTIMIZE_FOR_SIZE
> in the KCSAN runtime:
> http://lkml.kernel.org/r/58708908-84a0-0a81-a836-ad97e33dbb62@xxxxxxxxxxxxx
> 
> Reported-by: Randy Dunlap <rdunlap@xxxxxxxxxxxxx>
> Signed-off-by: Marco Elver <elver@xxxxxxxxxx>
> ---
>  include/asm-generic/atomic-instrumented.h | 334 +++++++++++-----------
>  include/asm-generic/atomic-long.h         | 330 ++++++++++-----------
>  scripts/atomic/gen-atomic-instrumented.sh |   6 +-
>  scripts/atomic/gen-atomic-long.sh         |   2 +-
>  4 files changed, 336 insertions(+), 336 deletions(-)

Do we need to do similar for gen-atomic-fallback.sh and the fallbacks
defined in scripts/atomic/fallbacks/ ?

[...]

> diff --git a/scripts/atomic/gen-atomic-instrumented.sh b/scripts/atomic/gen-atomic-instrumented.sh
> index 8b8b2a6f8d68..68532d4f36ca 100755
> --- a/scripts/atomic/gen-atomic-instrumented.sh
> +++ b/scripts/atomic/gen-atomic-instrumented.sh
> @@ -84,7 +84,7 @@ gen_proto_order_variant()
>  	[ ! -z "${guard}" ] && printf "#if ${guard}\n"
>  
>  cat <<EOF
> -static inline ${ret}
> +static __always_inline ${ret}

We should add an include of <linux/compiler.h> to the preamble if we're
explicitly using __always_inline.

> diff --git a/scripts/atomic/gen-atomic-long.sh b/scripts/atomic/gen-atomic-long.sh
> index c240a7231b2e..4036d2dd22e9 100755
> --- a/scripts/atomic/gen-atomic-long.sh
> +++ b/scripts/atomic/gen-atomic-long.sh
> @@ -46,7 +46,7 @@ gen_proto_order_variant()
>  	local retstmt="$(gen_ret_stmt "${meta}")"
>  
>  cat <<EOF
> -static inline ${ret}
> +static __always_inline ${ret}

Likewise here

Thanks,
Mark.