Re: [PATCHv8 00/11] Linear Address Masking enabling

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Sep 05, 2022 at 07:47:08PM +0300, Kirill A. Shutemov wrote:

> Fair enough. How about this?
> 
> diff --git a/arch/x86/include/asm/uaccess.h b/arch/x86/include/asm/uaccess.h
> index 803241dfc473..1a03c65a9c0f 100644
> --- a/arch/x86/include/asm/uaccess.h
> +++ b/arch/x86/include/asm/uaccess.h
> @@ -22,6 +22,8 @@ static inline bool pagefault_disabled(void);
>  #endif
>  
>  #ifdef CONFIG_X86_64
> +DECLARE_STATIC_KEY_FALSE(tagged_addr_key);
> +
>  /*
>   * Mask out tag bits from the address.
>   *
> @@ -30,8 +32,10 @@ static inline bool pagefault_disabled(void);
>   */
>  #define untagged_addr(mm, addr)	({					\
>  	u64 __addr = (__force u64)(addr);				\
> -	s64 sign = (s64)__addr >> 63;					\
> -	__addr &= (mm)->context.untag_mask | sign;			\
> +	if (static_branch_unlikely(&tagged_addr_key)) {			\
> +		s64 sign = (s64)__addr >> 63;				\
> +		__addr &= (mm)->context.untag_mask | sign;		\
> +	}								\
>  	(__force __typeof__(addr))__addr;				\
>  })
>  
> diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
> index 337f80a0862f..63194bf43c9a 100644
> --- a/arch/x86/kernel/process_64.c
> +++ b/arch/x86/kernel/process_64.c
> @@ -742,6 +742,9 @@ static long prctl_map_vdso(const struct vdso_image *image, unsigned long addr)
>  }
>  #endif
>  
> +DEFINE_STATIC_KEY_FALSE(tagged_addr_key);

So here you use the: false-unlikely scenario which seems suboptimal in
this case, I was thinking the false-likely case would generate better
code (see the comment in linux/jump_label.h).

> +EXPORT_SYMBOL_GPL(tagged_addr_key);
> +
>  static void enable_lam_func(void *mm)
>  {
>  	struct mm_struct *loaded_mm = this_cpu_read(cpu_tlbstate.loaded_mm);
> @@ -813,6 +816,7 @@ static int prctl_enable_tagged_addr(struct mm_struct *mm, unsigned long nr_bits)
>  	}
>  
>  	on_each_cpu_mask(mm_cpumask(mm), enable_lam_func, mm, true);
> +	static_branch_enable(&tagged_addr_key);
>  out:
>  	mutex_unlock(&mm->context.lock);
>  	mmap_write_unlock(mm);

Aside from the one nit above, this looks about right.