Re: [PATCH v2] ARM: add __always_inline to functions called from __get_user_check()

[Nick Desaulniers]

On Tue, Oct 1, 2019 at 1:37 AM Masahiro Yamada
<yamada.masahiro@xxxxxxxxxxxxx> wrote:
>
> KernelCI reports that bcm2835_defconfig is no longer booting since
> commit ac7c3e4ff401 ("compiler: enable CONFIG_OPTIMIZE_INLINING
> forcibly") (https://lkml.org/lkml/2019/9/26/825).
>
> I also received a regression report from Nicolas Saenz Julienne
> (https://lkml.org/lkml/2019/9/27/263).
>
> This problem has cropped up on bcm2835_defconfig because it enables
> CONFIG_CC_OPTIMIZE_FOR_SIZE. The compiler tends to prefer not inlining
> functions with -Os. I was able to reproduce it with other boards and
> defconfig files by manually enabling CONFIG_CC_OPTIMIZE_FOR_SIZE.
>
> The __get_user_check() specifically uses r0, r1, r2 registers.
> So, uaccess_save_and_enable() and uaccess_restore() must be inlined.
> Otherwise, those register assignments would be entirely dropped,
> according to my analysis of the disassembly.
>
> Prior to commit 9012d011660e ("compiler: allow all arches to enable
> CONFIG_OPTIMIZE_INLINING"), the 'inline' marker was always enough for
> inlining functions, except on x86.
>
> Since that commit, all architectures can enable CONFIG_OPTIMIZE_INLINING.
> So, __always_inline is now the only guaranteed way of forcible inlining.

No, the C preprocessor is the only guaranteed way of inlining.  I
preferred v1; if you're going to <strikethrough>play with
fire</strikethrough>write assembly, don't get burned.

>
> I also added __always_inline to 4 functions in the call-graph from the
> __get_user_check() macro.
>
> Fixes: 9012d011660e ("compiler: allow all arches to enable CONFIG_OPTIMIZE_INLINING")
> Reported-by: "kernelci.org bot" <bot@xxxxxxxxxxxx>
> Reported-by: Nicolas Saenz Julienne <nsaenzjulienne@xxxxxxx>
> Signed-off-by: Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx>
> ---
>
> Changes in v2:
>   - Use __always_inline instead of changing the function call places
>      (per Russell King)
>   - The previous submission is: https://lore.kernel.org/patchwork/patch/1132459/
>
>  arch/arm/include/asm/domain.h  | 8 ++++----
>  arch/arm/include/asm/uaccess.h | 4 ++--
>  2 files changed, 6 insertions(+), 6 deletions(-)
>
> diff --git a/arch/arm/include/asm/domain.h b/arch/arm/include/asm/domain.h
> index 567dbede4785..f1d0a7807cd0 100644
> --- a/arch/arm/include/asm/domain.h
> +++ b/arch/arm/include/asm/domain.h
> @@ -82,7 +82,7 @@
>  #ifndef __ASSEMBLY__
>
>  #ifdef CONFIG_CPU_CP15_MMU
> -static inline unsigned int get_domain(void)
> +static __always_inline unsigned int get_domain(void)
>  {
>         unsigned int domain;
>
> @@ -94,7 +94,7 @@ static inline unsigned int get_domain(void)
>         return domain;
>  }
>
> -static inline void set_domain(unsigned val)
> +static __always_inline void set_domain(unsigned int val)
>  {
>         asm volatile(
>         "mcr    p15, 0, %0, c3, c0      @ set domain"
> @@ -102,12 +102,12 @@ static inline void set_domain(unsigned val)
>         isb();
>  }
>  #else
> -static inline unsigned int get_domain(void)
> +static __always_inline unsigned int get_domain(void)
>  {
>         return 0;
>  }
>
> -static inline void set_domain(unsigned val)
> +static __always_inline void set_domain(unsigned int val)
>  {
>  }
>  #endif
> diff --git a/arch/arm/include/asm/uaccess.h b/arch/arm/include/asm/uaccess.h
> index 303248e5b990..98c6b91be4a8 100644
> --- a/arch/arm/include/asm/uaccess.h
> +++ b/arch/arm/include/asm/uaccess.h
> @@ -22,7 +22,7 @@
>   * perform such accesses (eg, via list poison values) which could then
>   * be exploited for priviledge escalation.
>   */
> -static inline unsigned int uaccess_save_and_enable(void)
> +static __always_inline unsigned int uaccess_save_and_enable(void)
>  {
>  #ifdef CONFIG_CPU_SW_DOMAIN_PAN
>         unsigned int old_domain = get_domain();
> @@ -37,7 +37,7 @@ static inline unsigned int uaccess_save_and_enable(void)
>  #endif
>  }
>
> -static inline void uaccess_restore(unsigned int flags)
> +static __always_inline void uaccess_restore(unsigned int flags)
>  {
>  #ifdef CONFIG_CPU_SW_DOMAIN_PAN
>         /* Restore the user access mask */
> --
> 2.17.1
>


-- 
Thanks,
~Nick Desaulniers