Re: [PATCH] atomic{64}_t: Explicitly specify data storage length and alignment

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Jul 09, 2018 at 03:47:41PM +0300, Alexey Brodkin wrote:
> diff --git a/include/asm-generic/atomic64.h b/include/asm-generic/atomic64.h
> index 8d28eb010d0d..b94b749b5952 100644
> --- a/include/asm-generic/atomic64.h
> +++ b/include/asm-generic/atomic64.h
> @@ -13,7 +13,7 @@
>  #define _ASM_GENERIC_ATOMIC64_H
>  
>  typedef struct {
> -	long long counter;
> +	u64 __aligned(8) counter;
>  } atomic64_t;

The type is wrong, atomic is signed, the alignment also really doesn't
matter, generic atomic64 is utter crap.

>  #define ATOMIC64_INIT(i)	{ (i) }
> diff --git a/include/linux/types.h b/include/linux/types.h
> index 9834e90aa010..e2f631782621 100644
> --- a/include/linux/types.h
> +++ b/include/linux/types.h
> @@ -174,12 +174,12 @@ typedef phys_addr_t resource_size_t;
>  typedef unsigned long irq_hw_number_t;
>  
>  typedef struct {
> -	int counter;
> +	u32 __aligned(4) counter;
>  } atomic_t;

u32 is wrong, the atomic type is signed.

Also, if an architecture doesn't properly align its native machine word
size but requires alignment for atomics it's a broken architecture.

>  
>  #ifdef CONFIG_64BIT
>  typedef struct {
> -	long counter;
> +	u64 __aligned(8) counter;
>  } atomic64_t;
>  #endif
>  

Similar for this one, on 64bit archs that support atomics the native
64bit types (long included) had better already imply this alignment.