Re: [PATCH v9 11/14] x86/cpu/keylocker: Check Register File Data Sampling mitigation

[Pawan Gupta]

On Thu, Mar 28, 2024 at 06:53:43PM -0700, Chang S. Bae wrote:
> The Register File Data Sampling vulnerability may allow malicious
> userspace programs to infer stale kernel register data, potentially
> exposing sensitive key values, including AES keys.
> 
> To address this vulnerability, a microcode update needs to be applied to
> the CPU, which modifies the VERW instruction to flush the affected CPU
> buffers.
> 
> The kernel already has a facility to flush CPU buffers before returning
> to userspace, which is indicated by the X86_FEATURE_CLEAR_CPU_BUF flag.
> 
> Ensure the mitigation before enabling Key Locker. Do not enable the
> feature on CPUs affected by the vulnerability but lacks mitigation.
> 
> Signed-off-by: Chang S. Bae <chang.seok.bae@xxxxxxxxx>
> Cc: Dave Hansen <dave.hansen@xxxxxxxxx>
> Cc: Pawan Gupta <pawan.kumar.gupta@xxxxxxxxxxxxxxx>
> ---
> Change from v8:
> * Add as a new patch.
> 
> Note that the code change follows the mitigation guidance [1]:
>   "Software loading Key Locker keys using LOADIWKEY should execute a VERW
>    to clear registers before transitioning to untrusted code to prevent
>    later software from inferring the loaded key."
> 
> [1] https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/advisory-guidance/register-file-data-sampling.html
> ---
>  arch/x86/kernel/keylocker.c | 17 +++++++++++++++++
>  1 file changed, 17 insertions(+)
> 
> diff --git a/arch/x86/kernel/keylocker.c b/arch/x86/kernel/keylocker.c
> index d4f3aa65ea8a..6e805c4da76d 100644
> --- a/arch/x86/kernel/keylocker.c
> +++ b/arch/x86/kernel/keylocker.c
> @@ -135,12 +135,29 @@ static bool __init have_gds_mitigation(void)
>  	return false;
>  }
>  
> +/*
> + * IA32_ARCH_CAPABILITIES MSR is retrieved during the setting of
> + * X86_BUG_RFDS. Ensure that the mitigation is applied to flush CPU
> + * buffers by checking the flag.
> + */
> +static bool __init have_rfds_mitigation(void)
> +{
> +	if (boot_cpu_has(X86_FEATURE_CLEAR_CPU_BUF))
> +		return true;

X86_FEATURE_CLEAR_CPU_BUF is also set by other VERW based mitigations
like MDS. The feature flag does not guarantee that the microcode
required to mitigate RFDS is loaded.

A more robust check would be:

	if (rfds_mitigation == RFDS_MITIGATION_VERW)
		return true;

And it would be apt to move this function to arch/x86/kernel/cpu/bugs.c

> +
> +	pr_warn("x86/keylocker: Susceptible to the RFDS vulnerability.\n");
> +	return false;
> +}
> +
>  /* Check if Key Locker is secure enough to be used. */
>  static bool __init secure_keylocker(void)
>  {
>  	if (boot_cpu_has_bug(X86_BUG_GDS) && !have_gds_mitigation())
>  		return false;
>  
> +	if (boot_cpu_has_bug(X86_BUG_RFDS) && !have_rfds_mitigation())
> +		return false;
> +
>  	return true;
>  }