[PATCH] x86/lam: Disable ADDRESS_MASKING in most cases

[PATCH] x86/lam: Disable ADDRESS_MASKING in most cases

[Pawan Gupta]

Intel feature Linear Address Masking (LAM) has a weakness related to
transient execution as described in the SLAM paper[1]. Unless Linear
Address Space Separation (LASS) is enabled this weakness may be
exploitable.

Until kernel adds support for LASS[2], only allow LAM for COMPILE_TEST,
or when speculation mitigations have been disabled at compile time,
otherwise keep LAM disabled.

[1] SLAM: https://download.vusec.net/papers/slam_sp24.pdf
[2] LASS: https://lore.kernel.org/lkml/20230609183632.48706-1-alexander.shishkin@linux.intel.com/

Cc: stable@vger.kernel.org # v6.4+
Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
---
 arch/x86/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 1566748f16c4..794517df8068 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -2270,6 +2270,7 @@ config RANDOMIZE_MEMORY_PHYSICAL_PADDING
 config ADDRESS_MASKING
 	bool "Linear Address Masking support"
 	depends on X86_64
+	depends on COMPILE_TEST || !SPECULATION_MITIGATIONS # wait for LASS
 	help
 	  Linear Address Masking (LAM) modifies the checking that is applied
 	  to 64-bit linear addresses, allowing software to use of the

base-commit: 0dd3ee31125508cd67f7e7172247f05b7fd1753a
-- 
2.34.1

Re: [PATCH] x86/lam: Disable ADDRESS_MASKING in most cases

[kirill.shutemov]

On Thu, Jan 18, 2024 at 06:35:47PM -0800, Pawan Gupta wrote:
> Intel feature Linear Address Masking (LAM) has a weakness related to
> transient execution as described in the SLAM paper[1]. Unless Linear
> Address Space Separation (LASS) is enabled this weakness may be
> exploitable.
> 
> Until kernel adds support for LASS[2], only allow LAM for COMPILE_TEST,
> or when speculation mitigations have been disabled at compile time,
> otherwise keep LAM disabled.
> 

It worth adding that there's no HW on market that supports LAM, so nobody
affected by the issue.

> [1] SLAM: https://download.vusec.net/papers/slam_sp24.pdf
> [2] LASS: https://lore.kernel.org/lkml/20230609183632.48706-1-alexander.shishkin@linux.intel.com/
> 
> Cc: stable@vger.kernel.org # v6.4+
> Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>

Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>

-- 
  Kiryl Shutsemau / Kirill A. Shutemov

Re: [PATCH] x86/lam: Disable ADDRESS_MASKING in most cases

[Pawan Gupta]

On Fri, Jan 19, 2024 at 07:58:28AM +0300, kirill.shutemov@linux.intel.com wrote:
> On Thu, Jan 18, 2024 at 06:35:47PM -0800, Pawan Gupta wrote:
> > Intel feature Linear Address Masking (LAM) has a weakness related to
> > transient execution as described in the SLAM paper[1]. Unless Linear
> > Address Space Separation (LASS) is enabled this weakness may be
> > exploitable.
> > 
> > Until kernel adds support for LASS[2], only allow LAM for COMPILE_TEST,
> > or when speculation mitigations have been disabled at compile time,
> > otherwise keep LAM disabled.
> > 
> 
> It worth adding that there's no HW on market that supports LAM, so nobody
> affected by the issue.

Right.

> > [1] SLAM: https://download.vusec.net/papers/slam_sp24.pdf
> > [2] LASS: https://lore.kernel.org/lkml/20230609183632.48706-1-alexander.shishkin@linux.intel.com/
> > 
> > Cc: stable@vger.kernel.org # v6.4+
> > Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
> 
> Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>

Thanks.

Re: [PATCH] x86/lam: Disable ADDRESS_MASKING in most cases

[Pawan Gupta]

On Fri, Jan 19, 2024 at 09:48:14AM -0800, Sohil Mehta wrote:
> Hi Pawan,
> 
> On 1/18/2024 6:35 PM, Pawan Gupta wrote:
> >  arch/x86/Kconfig | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> > index 1566748f16c4..794517df8068 100644
> > --- a/arch/x86/Kconfig
> > +++ b/arch/x86/Kconfig
> > @@ -2270,6 +2270,7 @@ config RANDOMIZE_MEMORY_PHYSICAL_PADDING
> >  config ADDRESS_MASKING
> >  	bool "Linear Address Masking support"
> >  	depends on X86_64
> > +	depends on COMPILE_TEST || !SPECULATION_MITIGATIONS # wait for LASS
> 
> I was wondering if the COMPILE_TEST dependency here is a bit redundant.
> 
> Having ADDRESS_MASKING depend on just !SPECULATION_MITIGATIONS might be
> enough to get the LAM code compile tested through various configurations.
>
> I don't have a strong preference here. Mainly looking to understand the
> reasoning. Other than that the patch looks fine to me.

The goal is to compile test it whenever possible. As
SPECULATION_MITIGATIONS are ON by default, it wont get tested for most
configuration.

Re: [PATCH] x86/lam: Disable ADDRESS_MASKING in most cases

[Sohil Mehta]

Hi Pawan,

On 1/18/2024 6:35 PM, Pawan Gupta wrote:
>  arch/x86/Kconfig | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> index 1566748f16c4..794517df8068 100644
> --- a/arch/x86/Kconfig
> +++ b/arch/x86/Kconfig
> @@ -2270,6 +2270,7 @@ config RANDOMIZE_MEMORY_PHYSICAL_PADDING
>  config ADDRESS_MASKING
>  	bool "Linear Address Masking support"
>  	depends on X86_64
> +	depends on COMPILE_TEST || !SPECULATION_MITIGATIONS # wait for LASS

I was wondering if the COMPILE_TEST dependency here is a bit redundant.

Having ADDRESS_MASKING depend on just !SPECULATION_MITIGATIONS might be
enough to get the LAM code compile tested through various configurations.

I don't have a strong preference here. Mainly looking to understand the
reasoning. Other than that the patch looks fine to me.

Sohil

>  	help
>  	  Linear Address Masking (LAM) modifies the checking that is applied
>  	  to 64-bit linear addresses, allowing software to use of the

Re: [PATCH] x86/lam: Disable ADDRESS_MASKING in most cases

[Pawan Gupta]

On Fri, Jan 19, 2024 at 03:36:22PM -0800, Sohil Mehta wrote:
> On 1/19/2024 8:21 AM, Pawan Gupta wrote:
> 
> > The goal is to compile test it whenever possible. As
> > SPECULATION_MITIGATIONS are ON by default, it wont get tested for most
> > configuration.
> 
> Sure, that would help increase the compile coverage. I wasn't sure how
> much of a difference it would make. Anyways, the restriction is expected
> to be short lived until LASS is available.

True.

> Please feel free to add:
> Reviewed-by: Sohil Mehta <sohil.mehta@intel.com>

Thanks

Re: [PATCH] x86/lam: Disable ADDRESS_MASKING in most cases

[Sohil Mehta]

On 1/19/2024 8:21 AM, Pawan Gupta wrote:

> The goal is to compile test it whenever possible. As
> SPECULATION_MITIGATIONS are ON by default, it wont get tested for most
> configuration.

Sure, that would help increase the compile coverage. I wasn't sure how
much of a difference it would make. Anyways, the restriction is expected
to be short lived until LASS is available.

Please feel free to add:
Reviewed-by: Sohil Mehta <sohil.mehta@intel.com>