Re: SME/32-bit regression

[Boris Ostrovsky]

On 09/06/2017 05:26 AM, Borislav Petkov wrote:
> On Tue, Sep 05, 2017 at 11:45:07PM -0400, Boris Ostrovsky wrote:
>> It appears there is a regression for 32-bit kernels due to SME changes.
>>
>> I bisected my particular problem
> It being? Doesn't boot, splats?

Xen guest crashes very early, before a splat can can be generated.

>
>> (Xen PV guest) to
>> 21729f81ce8ae76a6995681d40e16f7ce8075db4 but I also saw pmd_clear_bad()
>> errors on baremetal. This seems to be caused by sme_me_mask being an
>> unsigned long as opposed to phys_addr_t (the actual problem is that
>> __PHYSICAL_MASK is truncated). When I declare it as u64 and drop unsigned
>> long cast in __sme_set()/__sme_clr() the problem goes way. (This presumably
>> won't work for non-PAE which I haven't tried).
> Right, so I think we should do this because those macros should not have
> any effect on !CONFIG_AMD_MEM_ENCRYPT setups.

This won't help though if kernel is built with SME support.

-boris

>
> ---
> diff --git a/include/linux/mem_encrypt.h b/include/linux/mem_encrypt.h
> index 1255f09f5e42..823eec6ba951 100644
> --- a/include/linux/mem_encrypt.h
> +++ b/include/linux/mem_encrypt.h
> @@ -35,6 +35,7 @@ static inline unsigned long sme_get_me_mask(void)
>  	return sme_me_mask;
>  }
>  
> +#ifdef CONFIG_AMD_MEM_ENCRYPT
>  /*
>   * The __sme_set() and __sme_clr() macros are useful for adding or removing
>   * the encryption mask from a value (e.g. when dealing with pagetable
> @@ -42,6 +43,10 @@ static inline unsigned long sme_get_me_mask(void)
>   */
>  #define __sme_set(x)		((unsigned long)(x) | sme_me_mask)
>  #define __sme_clr(x)		((unsigned long)(x) & ~sme_me_mask)
> +#else
> +#define __sme_set(x)		(x)
> +#define __sme_clr(x)		(x)
> +#endif
>  
>  #endif	/* __ASSEMBLY__ */
>  
>
>
>