TDX 1.0 generates a #VE when accessing topology-related CPUID leafs
(0xB and 0x1F) and the X2APIC_APICID MSR. The kernel returns all
zeros on CPUID #VEs. In practice, this means that the kernel can only
boot with a plain topology. Any complications will cause problems.

The ENUM_TOPOLOGY feature allows the VMM to provide topology
information to the guest in a safe manner. Enabling the feature
eliminates topology-related #VEs: the TDX module virtualizes
accesses to the CPUID leafs and the MSR.

Enable ENUM_TOPOLOGY if it is available.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
---
 arch/x86/coco/tdx/tdx.c           | 20 ++++++++++++++++++++
 arch/x86/include/asm/shared/tdx.h |  3 +++
 2 files changed, 23 insertions(+)

diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
index d9ea82f8772d..291e45db8d54 100644
--- a/arch/x86/coco/tdx/tdx.c
+++ b/arch/x86/coco/tdx/tdx.c
@@ -280,6 +280,26 @@ static void tdx_setup(u64 *cc_mask)
 		else
 			tdx_panic(msg);
 	}
+
+	/*
+	 * TDX 1.0 generates a #VE when accessing topology-related CPUID leafs
+	 * (0xB and 0x1F) and the X2APIC_APICID MSR. The kernel returns all
+	 * zeros on CPUID #VEs. In practice, this means that the kernel can only
+	 * boot with a plain topology. Any complications will cause problems.
+	 *
+	 * The ENUM_TOPOLOGY feature allows the VMM to provide topology
+	 * information to the guest in a safe manner. Enabling the feature
+	 * eliminates topology-related #VEs: the TDX module virtualizes
+	 * accesses to the CPUID leafs and the MSR.
+	 *
+	 * Enable ENUM_TOPOLOGY if it is available.
+	 */
+	if ((features & TDX_FEATURES0_ENUM_TOPOLOGY) &&
+	    tdg_vm_rd(TDCS_TOPOLOGY_ENUM_CONFIGURED)) {
+		tdg_vm_wr(TDCS_TD_CTLS,
+			  TD_CTLS_ENUM_TOPOLOGY,
+			  TD_CTLS_ENUM_TOPOLOGY);
+	}
 }
 
 /*
diff --git a/arch/x86/include/asm/shared/tdx.h b/arch/x86/include/asm/shared/tdx.h
index 29a61c72e4dd..2964c506b241 100644
--- a/arch/x86/include/asm/shared/tdx.h
+++ b/arch/x86/include/asm/shared/tdx.h
@@ -27,15 +27,18 @@
 #define TDCS_CONFIG_FLAGS		0x1110000300000016
 #define TDCS_TD_CTLS			0x1110000300000017
 #define TDCS_NOTIFY_ENABLES		0x9100000000000010
+#define TDCS_TOPOLOGY_ENUM_CONFIGURED	0x9100000000000019
 
 /* TDCS_TDX_FEATURES0 bits */
 #define TDX_FEATURES0_PENDING_EPT_VIOLATION_V2	BIT_ULL(16)
+#define TDX_FEATURES0_ENUM_TOPOLOGY		BIT_ULL(20)
 
 /* TDCS_CONFIG_FLAGS bits */
 #define TDCS_CONFIG_FLEXIBLE_PENDING_VE	BIT_ULL(1)
 
 /* TDCS_TD_CTLS bits */
 #define TD_CTLS_PENDING_VE_DISABLE	BIT_ULL(0)
+#define TD_CTLS_ENUM_TOPOLOGY		BIT_ULL(1)
 
 /* TDX hypercall Leaf IDs */
 #define TDVMCALL_MAP_GPA		0x10001
-- 
2.43.0

On 3/9/24 1:02 PM, Kirill A. Shutemov wrote:
> TDX 1.0 generates a #VE when accessing topology-related CPUID leafs
> (0xB and 0x1F) and the X2APIC_APICID MSR. The kernel returns all
> zeros on CPUID #VEs. In practice, this means that the kernel can only
> boot with a plain topology. Any complications will cause problems.

Is this issue only for TDX 1.0? What about TDX > 1.0?

>
> The ENUM_TOPOLOGY feature allows the VMM to provide topology
> information to the guest in a safe manner. Enabling the feature
> eliminates topology-related #VEs: the TDX module virtualizes
> accesses to the CPUID leafs and the MSR.
>
> Enable ENUM_TOPOLOGY if it is available.

I cant find the ENUM_TOPOLOGY in ABI spec (https://cdrdv2.intel.com/v1/dl/getContent/795381).

Can you point me to the correct document?

>
> Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> ---
>  arch/x86/coco/tdx/tdx.c           | 20 ++++++++++++++++++++
>  arch/x86/include/asm/shared/tdx.h |  3 +++
>  2 files changed, 23 insertions(+)
>
> diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
> index d9ea82f8772d..291e45db8d54 100644
> --- a/arch/x86/coco/tdx/tdx.c
> +++ b/arch/x86/coco/tdx/tdx.c
> @@ -280,6 +280,26 @@ static void tdx_setup(u64 *cc_mask)
>  		else
>  			tdx_panic(msg);
>  	}
> +
> +	/*
> +	 * TDX 1.0 generates a #VE when accessing topology-related CPUID leafs
> +	 * (0xB and 0x1F) and the X2APIC_APICID MSR. The kernel returns all
> +	 * zeros on CPUID #VEs. In practice, this means that the kernel can only
> +	 * boot with a plain topology. Any complications will cause problems.
> +	 *
> +	 * The ENUM_TOPOLOGY feature allows the VMM to provide topology
> +	 * information to the guest in a safe manner. Enabling the feature
> +	 * eliminates topology-related #VEs: the TDX module virtualizes
> +	 * accesses to the CPUID leafs and the MSR.
> +	 *
> +	 * Enable ENUM_TOPOLOGY if it is available.
> +	 */

Why are we overriding it in guest? Why can't this by done by TDX Module
or QEMU during initialization?

> +	if ((features & TDX_FEATURES0_ENUM_TOPOLOGY) &&
> +	    tdg_vm_rd(TDCS_TOPOLOGY_ENUM_CONFIGURED)) {
> +		tdg_vm_wr(TDCS_TD_CTLS,
> +			  TD_CTLS_ENUM_TOPOLOGY,
> +			  TD_CTLS_ENUM_TOPOLOGY);
> +	}
>  }
>  
>  /*
> diff --git a/arch/x86/include/asm/shared/tdx.h b/arch/x86/include/asm/shared/tdx.h
> index 29a61c72e4dd..2964c506b241 100644
> --- a/arch/x86/include/asm/shared/tdx.h
> +++ b/arch/x86/include/asm/shared/tdx.h
> @@ -27,15 +27,18 @@
>  #define TDCS_CONFIG_FLAGS		0x1110000300000016
>  #define TDCS_TD_CTLS			0x1110000300000017
>  #define TDCS_NOTIFY_ENABLES		0x9100000000000010
> +#define TDCS_TOPOLOGY_ENUM_CONFIGURED	0x9100000000000019
>  
>  /* TDCS_TDX_FEATURES0 bits */
>  #define TDX_FEATURES0_PENDING_EPT_VIOLATION_V2	BIT_ULL(16)
> +#define TDX_FEATURES0_ENUM_TOPOLOGY		BIT_ULL(20)
>  
>  /* TDCS_CONFIG_FLAGS bits */
>  #define TDCS_CONFIG_FLEXIBLE_PENDING_VE	BIT_ULL(1)
>  
>  /* TDCS_TD_CTLS bits */
>  #define TD_CTLS_PENDING_VE_DISABLE	BIT_ULL(0)
> +#define TD_CTLS_ENUM_TOPOLOGY		BIT_ULL(1)
>  
>  /* TDX hypercall Leaf IDs */
>  #define TDVMCALL_MAP_GPA		0x10001

-- 
Sathyanarayanan Kuppuswamy
Linux Kernel Developer

On Sat, Mar 09, 2024 at 07:56:11PM -0800, Kuppuswamy Sathyanarayanan wrote:
> 
> On 3/9/24 1:02 PM, Kirill A. Shutemov wrote:
> > TDX 1.0 generates a #VE when accessing topology-related CPUID leafs
> > (0xB and 0x1F) and the X2APIC_APICID MSR. The kernel returns all
> > zeros on CPUID #VEs. In practice, this means that the kernel can only
> > boot with a plain topology. Any complications will cause problems.
> 
> Is this issue only for TDX 1.0? What about TDX > 1.0?

TDX 1.0 defines the baseline for TDX. Any change in behaviour will be
gated by explicit opt-in. In this case, it is ENUM_TOPOLOGY.

> > The ENUM_TOPOLOGY feature allows the VMM to provide topology
> > information to the guest in a safe manner. Enabling the feature
> > eliminates topology-related #VEs: the TDX module virtualizes
> > accesses to the CPUID leafs and the MSR.
> >
> > Enable ENUM_TOPOLOGY if it is available.
> 
> I cant find the ENUM_TOPOLOGY in ABI spec (https://cdrdv2.intel.com/v1/dl/getContent/795381).
> 
> Can you point me to the correct document?

I guess JSON dump is out of sync with the PDF doc. See this:

https://cdrdv2.intel.com/v1/dl/getContent/733579

-- 
  Kiryl Shutsemau / Kirill A. Shutemov