diff mbox series

[v13,1/3] x86/tdx: Add TDX Guest attestation interface driver

Message ID 20220909192708.1113126-2-sathyanarayanan.kuppuswamy@linux.intel.com (mailing list archive)
State New
Headers show
Series Add TDX Guest Attestation support | expand

Commit Message

Kuppuswamy Sathyanarayanan Sept. 9, 2022, 7:27 p.m. UTC
Attestation is used to verify the TDX guest trustworthiness to other
entities before provisioning secrets to the guest. For example, a key
server may request for attestation before releasing the encryption keys
to mount the encrypted rootfs or secondary drive.

During the TDX guest launch, the initial contents (including the
firmware image) and configuration of the guest are recorded by the
Intel TDX module in build time measurement register (MRTD). After TDX
guest is created, run-time measurement registers (RTMRs) can be used by
the guest software to extend the measurements. TDX supports 4 RTMR
registers, and TDG.MR.RTMR.EXTEND TDCALL is used to update the RTMR
registers securely. RTMRs are mainly used to record measurements
related to sections like the kernel image, command line parameters,
initrd, ACPI tables, firmware data, configuration firmware volume (CFV)
of TDVF, etc. For complete details, please refer to TDX Virtual
Firmware design specification, sec titled "TD Measurement".

At TDX guest runtime, the Intel TDX module reuses the Intel SGX
attestation infrastructure to provide support for attesting to these
measurements as described below.

The attestation process consists of two steps: TDREPORT generation and
Quote generation.

TDREPORT (TDREPORT_STRUCT) is a fixed-size data structure generated by
the TDX module which contains guest-specific information (such as build
and boot measurements), platform security version, and the MAC to
protect the integrity of the TDREPORT. The guest kernel uses
TDCALL[TDG.MR.REPORT] to get the TDREPORT from the TDX module. A
user-provided 64-Byte REPORTDATA is used as input and included in the
TDREPORT. Typically it can be some nonce provided by attestation
service so the TDREPORT can be verified uniquely. More details about
the TDREPORT can be found in Intel TDX Module specification, section
titled "TDG.MR.REPORT Leaf".

TDREPORT by design can only be verified on the local platform as the
MAC key is bound to the platform. To support remote verification of
the TDREPORT, TDX leverages Intel SGX Quote Enclave (QE) to verify
the TDREPORT locally and convert it to a remote verifiable Quote.

After getting the TDREPORT, the second step of the attestation process
is to send it to the QE to generate the Quote. TDX doesn't support SGX
inside the guest, so the QE can be deployed in the host, or in another
legacy VM with SGX support. QE checks the integrity of TDREPORT and if
it is valid, a certified quote signing key is used to sign the Quote.
How to send the TDREPORT to QE and receive the Quote is implementation
and deployment specific.

Implement a basic guest misc driver to allow userspace to get the
TDREPORT. After getting TDREPORT, the userspace attestation software
can choose whatever communication channel available (i.e. vsock or
hypercall) to send the TDREPORT to QE and receive the Quote.

Also note that explicit access permissions are not enforced in this
driver because the quote and measurements are not a secret. However
the access permissions of the device node can be used to set any
desired access policy. The udev default is usually root access
only.

Operations like getting TDREPORT or Quote generation involves sending
a blob of data as input and getting another blob of data as output. It
was considered to use a sysfs interface for this, but it doesn't fit
well into the standard sysfs model for configuring values. It would be
possible to do read/write on files, but it would need multiple file
descriptors, which would be somewhat messy. IOCTLs seems to be the best
fitting and simplest model for this use case. This is similar to AMD
SEV platform, which also uses IOCTL interface to support attestation.

Any distribution enabling TDX is also expected to need attestation. So
enable it by default with TDX guest support.

Reviewed-by: Tony Luck <tony.luck@intel.com>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Kai Huang <kai.huang@intel.com>
Acked-by: Wander Lairson Costa <wander@redhat.com>
Signed-off-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@linux.intel.com>
---

Changes since v12:
 * Added check to ensure reserved entries are set as 0.

Changes since v11:
 * Renamed DRIVER_NAME to TDX_GUEST_DEVICE and moved it to
   arch/x86/include/uapi/asm/tdx.h.
 * Fixed default error number in tdx_guest_ioctl().
 * Moved tdx_misc_dev definition out of tdx_guest_init() as
   per Greg's suggestion.
 * Reordered struct tdx_report_req to avoid holes and added
   required padding.

Changes since v10:
 * Replaced TD/TD Guest usage with TDX Guest or Guest.
 * Removed unnecessary comments.
 * Added more validation to user input in tdx_get_report().
 * Used u64_to_user_ptr when reading user u64 pointers.
 * Fixed commit log as per review comments.

Changes since v9:
 * Dropped the cover letter. Since this patch set only adds
   TDREPORT support, the commit log itself has all the required details.
 * Dropped the Quote support and event IRQ support as per Dave's
   review suggestion.
 * Dropped attest.c and moved its contents to tdx.c
 * Updated commit log and comments to reflect latest changes.

Changes since v8:
 * Please refer to https://lore.kernel.org/all/ \
   20220728034420.648314-1-sathyanarayanan.kuppuswamy@linux.intel.com/

 arch/x86/coco/tdx/tdx.c         | 115 ++++++++++++++++++++++++++++++++
 arch/x86/include/uapi/asm/tdx.h |  56 ++++++++++++++++
 2 files changed, 171 insertions(+)
 create mode 100644 arch/x86/include/uapi/asm/tdx.h

Comments

Greg KH Sept. 9, 2022, 7:39 p.m. UTC | #1
On Fri, Sep 09, 2022 at 12:27:06PM -0700, Kuppuswamy Sathyanarayanan wrote:
> Attestation is used to verify the TDX guest trustworthiness to other
> entities before provisioning secrets to the guest. For example, a key
> server may request for attestation before releasing the encryption keys
> to mount the encrypted rootfs or secondary drive.
> 
> During the TDX guest launch, the initial contents (including the
> firmware image) and configuration of the guest are recorded by the
> Intel TDX module in build time measurement register (MRTD). After TDX
> guest is created, run-time measurement registers (RTMRs) can be used by
> the guest software to extend the measurements. TDX supports 4 RTMR
> registers, and TDG.MR.RTMR.EXTEND TDCALL is used to update the RTMR
> registers securely. RTMRs are mainly used to record measurements
> related to sections like the kernel image, command line parameters,
> initrd, ACPI tables, firmware data, configuration firmware volume (CFV)
> of TDVF, etc. For complete details, please refer to TDX Virtual
> Firmware design specification, sec titled "TD Measurement".
> 
> At TDX guest runtime, the Intel TDX module reuses the Intel SGX
> attestation infrastructure to provide support for attesting to these
> measurements as described below.
> 
> The attestation process consists of two steps: TDREPORT generation and
> Quote generation.
> 
> TDREPORT (TDREPORT_STRUCT) is a fixed-size data structure generated by
> the TDX module which contains guest-specific information (such as build
> and boot measurements), platform security version, and the MAC to
> protect the integrity of the TDREPORT. The guest kernel uses
> TDCALL[TDG.MR.REPORT] to get the TDREPORT from the TDX module. A
> user-provided 64-Byte REPORTDATA is used as input and included in the
> TDREPORT. Typically it can be some nonce provided by attestation
> service so the TDREPORT can be verified uniquely. More details about
> the TDREPORT can be found in Intel TDX Module specification, section
> titled "TDG.MR.REPORT Leaf".
> 
> TDREPORT by design can only be verified on the local platform as the
> MAC key is bound to the platform. To support remote verification of
> the TDREPORT, TDX leverages Intel SGX Quote Enclave (QE) to verify
> the TDREPORT locally and convert it to a remote verifiable Quote.
> 
> After getting the TDREPORT, the second step of the attestation process
> is to send it to the QE to generate the Quote. TDX doesn't support SGX
> inside the guest, so the QE can be deployed in the host, or in another
> legacy VM with SGX support. QE checks the integrity of TDREPORT and if
> it is valid, a certified quote signing key is used to sign the Quote.
> How to send the TDREPORT to QE and receive the Quote is implementation
> and deployment specific.
> 
> Implement a basic guest misc driver to allow userspace to get the
> TDREPORT. After getting TDREPORT, the userspace attestation software
> can choose whatever communication channel available (i.e. vsock or
> hypercall) to send the TDREPORT to QE and receive the Quote.
> 
> Also note that explicit access permissions are not enforced in this
> driver because the quote and measurements are not a secret. However
> the access permissions of the device node can be used to set any
> desired access policy. The udev default is usually root access
> only.
> 
> Operations like getting TDREPORT or Quote generation involves sending
> a blob of data as input and getting another blob of data as output. It
> was considered to use a sysfs interface for this, but it doesn't fit
> well into the standard sysfs model for configuring values. It would be
> possible to do read/write on files, but it would need multiple file
> descriptors, which would be somewhat messy. IOCTLs seems to be the best
> fitting and simplest model for this use case. This is similar to AMD
> SEV platform, which also uses IOCTL interface to support attestation.
> 
> Any distribution enabling TDX is also expected to need attestation. So
> enable it by default with TDX guest support.
> 
> Reviewed-by: Tony Luck <tony.luck@intel.com>
> Reviewed-by: Andi Kleen <ak@linux.intel.com>
> Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> Acked-by: Kai Huang <kai.huang@intel.com>
> Acked-by: Wander Lairson Costa <wander@redhat.com>
> Signed-off-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@linux.intel.com>
> ---
> 
> Changes since v12:
>  * Added check to ensure reserved entries are set as 0.
> 
> Changes since v11:
>  * Renamed DRIVER_NAME to TDX_GUEST_DEVICE and moved it to
>    arch/x86/include/uapi/asm/tdx.h.
>  * Fixed default error number in tdx_guest_ioctl().
>  * Moved tdx_misc_dev definition out of tdx_guest_init() as
>    per Greg's suggestion.
>  * Reordered struct tdx_report_req to avoid holes and added
>    required padding.
> 
> Changes since v10:
>  * Replaced TD/TD Guest usage with TDX Guest or Guest.
>  * Removed unnecessary comments.
>  * Added more validation to user input in tdx_get_report().
>  * Used u64_to_user_ptr when reading user u64 pointers.
>  * Fixed commit log as per review comments.
> 
> Changes since v9:
>  * Dropped the cover letter. Since this patch set only adds
>    TDREPORT support, the commit log itself has all the required details.
>  * Dropped the Quote support and event IRQ support as per Dave's
>    review suggestion.
>  * Dropped attest.c and moved its contents to tdx.c
>  * Updated commit log and comments to reflect latest changes.
> 
> Changes since v8:
>  * Please refer to https://lore.kernel.org/all/ \
>    20220728034420.648314-1-sathyanarayanan.kuppuswamy@linux.intel.com/
> 
>  arch/x86/coco/tdx/tdx.c         | 115 ++++++++++++++++++++++++++++++++
>  arch/x86/include/uapi/asm/tdx.h |  56 ++++++++++++++++
>  2 files changed, 171 insertions(+)
>  create mode 100644 arch/x86/include/uapi/asm/tdx.h
> 
> diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
> index 928dcf7a20d9..8b5c59110321 100644
> --- a/arch/x86/coco/tdx/tdx.c
> +++ b/arch/x86/coco/tdx/tdx.c
> @@ -5,16 +5,21 @@
>  #define pr_fmt(fmt)     "tdx: " fmt
>  
>  #include <linux/cpufeature.h>
> +#include <linux/miscdevice.h>
> +#include <linux/mm.h>
> +#include <linux/io.h>
>  #include <asm/coco.h>
>  #include <asm/tdx.h>
>  #include <asm/vmx.h>
>  #include <asm/insn.h>
>  #include <asm/insn-eval.h>
>  #include <asm/pgtable.h>
> +#include <uapi/asm/tdx.h>
>  
>  /* TDX module Call Leaf IDs */
>  #define TDX_GET_INFO			1
>  #define TDX_GET_VEINFO			3
> +#define TDX_GET_REPORT			4
>  #define TDX_ACCEPT_PAGE			6
>  
>  /* TDX hypercall Leaf IDs */
> @@ -775,3 +780,113 @@ void __init tdx_early_init(void)
>  
>  	pr_info("Guest detected\n");
>  }
> +
> +static long tdx_get_report(void __user *argp)
> +{
> +	u8 *reportdata, *tdreport;
> +	struct tdx_report_req req;
> +	u8 reserved[7] = {0};

Where is the magic "7" coming from?

> +	long ret;
> +
> +	if (copy_from_user(&req, argp, sizeof(req)))
> +		return -EFAULT;
> +
> +	/*
> +	 * Per TDX Module 1.0 specification, section titled
> +	 * "TDG.MR.REPORT", REPORTDATA length is fixed as
> +	 * TDX_REPORTDATA_LEN, TDREPORT length is fixed as
> +	 * TDX_REPORT_LEN, and TDREPORT subtype is fixed as
> +	 * 0. Also check for valid user pointers and make sure
> +	 * reserved entries values are zero.
> +	 */
> +	if (!req.reportdata || !req.tdreport || req.subtype ||
> +		req.rpd_len != TDX_REPORTDATA_LEN ||
> +		req.tdr_len != TDX_REPORT_LEN ||
> +		memcmp(req.reserved, reserved, 7))

Again, magic number?

thanks,

greg k-h
Dave Hansen Sept. 9, 2022, 7:41 p.m. UTC | #2
On 9/9/22 12:27, Kuppuswamy Sathyanarayanan wrote:
> +	u8 reserved[7] = {0};
...
> +	if (!req.reportdata || !req.tdreport || req.subtype ||
> +		req.rpd_len != TDX_REPORTDATA_LEN ||
> +		req.tdr_len != TDX_REPORT_LEN ||
> +		memcmp(req.reserved, reserved, 7))
> +		return -EINVAL;

Huh, so to look for 0's, you:

1. Declare an on-stack structure with a hard coded, magic numbered field
   that has to be zeroed.
2. memcmp() that structure
3. Feed memcmp() with another hard coded magic number

I've gotta ask: did you have any reservations writing this code?  Were
there any alarm bells going off saying that something might be wrong?

Using memcmp() itself is probably forgivable.  But, the two magic
numbers are pretty mortal sins in my book.  What's going to happen the
first moment someone wants to repurpose a reserved byte?  They're going
to do:

-	__u8 reserved[7];
+	__u8 my_new_byte;
+	__u8 reserved[6];

What's going to happen to the code you wrote?  Will it continue to work?
 Or will the memcmp() silently start doing crazy stuff as it overruns
the structure into garbage land?

What's wrong with:

	memchr_inv(&req.reserved, sizeof(req.reserved), 0)
Kuppuswamy Sathyanarayanan Sept. 9, 2022, 8:07 p.m. UTC | #3
On 9/9/22 12:41 PM, Dave Hansen wrote:
> On 9/9/22 12:27, Kuppuswamy Sathyanarayanan wrote:
>> +	u8 reserved[7] = {0};
> ...
>> +	if (!req.reportdata || !req.tdreport || req.subtype ||
>> +		req.rpd_len != TDX_REPORTDATA_LEN ||
>> +		req.tdr_len != TDX_REPORT_LEN ||
>> +		memcmp(req.reserved, reserved, 7))
>> +		return -EINVAL;
> 
> Huh, so to look for 0's, you:
> 
> 1. Declare an on-stack structure with a hard coded, magic numbered field
>    that has to be zeroed.
> 2. memcmp() that structure
> 3. Feed memcmp() with another hard coded magic number
> 
> I've gotta ask: did you have any reservations writing this code?  Were
> there any alarm bells going off saying that something might be wrong?
> 
> Using memcmp() itself is probably forgivable.  But, the two magic
> numbers are pretty mortal sins in my book.  What's going to happen the
> first moment someone wants to repurpose a reserved byte?  They're going
> to do:
> 
> -	__u8 reserved[7];
> +	__u8 my_new_byte;
> +	__u8 reserved[6];
> 
> What's going to happen to the code you wrote?  Will it continue to work?
>  Or will the memcmp() silently start doing crazy stuff as it overruns
> the structure into garbage land?
> 
> What's wrong with:
> 
> 	memchr_inv(&req.reserved, sizeof(req.reserved), 0)

I did not consider the hard coding issue. It is a mistake. Your suggestion
looks better. I will use it.
Dave Hansen Sept. 9, 2022, 8:54 p.m. UTC | #4
On 9/9/22 13:07, Sathyanarayanan Kuppuswamy wrote:
>> What's wrong with:
>>
>> 	memchr_inv(&req.reserved, sizeof(req.reserved), 0)
> I did not consider the hard coding issue. It is a mistake. Your suggestion
> looks better. I will use it.

BTW...  Please look at how memchr_inv() gets used in the kernel.  Don't
copy my nonsense too exactly.
Kirill A. Shutemov Sept. 12, 2022, 10:22 p.m. UTC | #5
On Fri, Sep 09, 2022 at 12:27:06PM -0700, Kuppuswamy Sathyanarayanan wrote:
> diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
> index 928dcf7a20d9..8b5c59110321 100644
> --- a/arch/x86/coco/tdx/tdx.c
> +++ b/arch/x86/coco/tdx/tdx.c
> @@ -5,16 +5,21 @@
>  #define pr_fmt(fmt)     "tdx: " fmt
>  
>  #include <linux/cpufeature.h>
> +#include <linux/miscdevice.h>
> +#include <linux/mm.h>
> +#include <linux/io.h>
>  #include <asm/coco.h>
>  #include <asm/tdx.h>
>  #include <asm/vmx.h>
>  #include <asm/insn.h>
>  #include <asm/insn-eval.h>
>  #include <asm/pgtable.h>
> +#include <uapi/asm/tdx.h>
>  
>  /* TDX module Call Leaf IDs */
>  #define TDX_GET_INFO			1
>  #define TDX_GET_VEINFO			3
> +#define TDX_GET_REPORT			4
>  #define TDX_ACCEPT_PAGE			6
>  
>  /* TDX hypercall Leaf IDs */
> @@ -775,3 +780,113 @@ void __init tdx_early_init(void)
>  
>  	pr_info("Guest detected\n");
>  }
> +
> +static long tdx_get_report(void __user *argp)
> +{
> +	u8 *reportdata, *tdreport;
> +	struct tdx_report_req req;
> +	u8 reserved[7] = {0};
> +	long ret;
> +
> +	if (copy_from_user(&req, argp, sizeof(req)))
> +		return -EFAULT;
> +
> +	/*
> +	 * Per TDX Module 1.0 specification, section titled
> +	 * "TDG.MR.REPORT", REPORTDATA length is fixed as
> +	 * TDX_REPORTDATA_LEN, TDREPORT length is fixed as
> +	 * TDX_REPORT_LEN, and TDREPORT subtype is fixed as
> +	 * 0. Also check for valid user pointers and make sure
> +	 * reserved entries values are zero.
> +	 */
> +	if (!req.reportdata || !req.tdreport || req.subtype ||
> +		req.rpd_len != TDX_REPORTDATA_LEN ||
> +		req.tdr_len != TDX_REPORT_LEN ||
> +		memcmp(req.reserved, reserved, 7))
> +		return -EINVAL;

Maybe make several checks instead of the monstrous one?

!req.reportdata and !req.tdreport look redundant. copy_from/to_user() will
catch them (and other bad address cases). And -EFAULT is more appropriate
in this case.

> +
> +	reportdata = kmalloc(req.rpd_len, GFP_KERNEL);
> +	if (!reportdata)
> +		return -ENOMEM;
> +
> +	tdreport = kzalloc(req.tdr_len, GFP_KERNEL);
> +	if (!tdreport) {
> +		kfree(reportdata);
> +		return -ENOMEM;
> +	}
> +
> +	if (copy_from_user(reportdata, u64_to_user_ptr(req.reportdata),
> +			   req.rpd_len)) {
> +		ret = -EFAULT;
> +		goto out;
> +	}
> +
> +	/*
> +	 * Generate TDREPORT using "TDG.MR.REPORT" TDCALL.
> +	 *
> +	 * Get the TDREPORT using REPORTDATA as input. Refer to
> +	 * section 22.3.3 TDG.MR.REPORT leaf in the TDX Module 1.0
> +	 * Specification for detailed information.
> +	 */
> +	ret = __tdx_module_call(TDX_GET_REPORT, virt_to_phys(tdreport),
> +				virt_to_phys(reportdata), req.subtype,
> +				0, NULL);
> +	if (ret) {
> +		ret = -EIO;

The spec says that it generate an error if invalid operand or busy. Maybe
translate the TDX error codes to errnos?

BTW, regarding busy case: do we want to protect against two parallel
TDX_GET_REPORT? What happens if we run the second TDX_GET_REPORT when the
first hasn't complete?

> +		goto out;
> +	}
> +
> +	if (copy_to_user(u64_to_user_ptr(req.tdreport), tdreport, req.tdr_len))
> +		ret = -EFAULT;
> +
> +out:
> +	kfree(reportdata);
> +	kfree(tdreport);
> +	return ret;
> +}
> +static long tdx_guest_ioctl(struct file *file, unsigned int cmd,
> +			    unsigned long arg)
> +{
> +	void __user *argp = (void __user *)arg;
> +	long ret = -ENOTTY;

Not a typewriter? Huh?

> +
> +	switch (cmd) {
> +	case TDX_CMD_GET_REPORT:
> +		ret = tdx_get_report(argp);
> +		break;
> +	default:
> +		pr_debug("cmd %d not supported\n", cmd);
> +		break;
> +	}
> +
> +	return ret;
> +}
Kuppuswamy Sathyanarayanan Sept. 12, 2022, 11 p.m. UTC | #6
On 9/12/22 3:22 PM, Kirill A . Shutemov wrote:
> On Fri, Sep 09, 2022 at 12:27:06PM -0700, Kuppuswamy Sathyanarayanan wrote:
>> diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
>> index 928dcf7a20d9..8b5c59110321 100644
>> --- a/arch/x86/coco/tdx/tdx.c
>> +++ b/arch/x86/coco/tdx/tdx.c
>> @@ -5,16 +5,21 @@
>>  #define pr_fmt(fmt)     "tdx: " fmt
>>  
>>  #include <linux/cpufeature.h>
>> +#include <linux/miscdevice.h>
>> +#include <linux/mm.h>
>> +#include <linux/io.h>
>>  #include <asm/coco.h>
>>  #include <asm/tdx.h>
>>  #include <asm/vmx.h>
>>  #include <asm/insn.h>
>>  #include <asm/insn-eval.h>
>>  #include <asm/pgtable.h>
>> +#include <uapi/asm/tdx.h>
>>  
>>  /* TDX module Call Leaf IDs */
>>  #define TDX_GET_INFO			1
>>  #define TDX_GET_VEINFO			3
>> +#define TDX_GET_REPORT			4
>>  #define TDX_ACCEPT_PAGE			6
>>  
>>  /* TDX hypercall Leaf IDs */
>> @@ -775,3 +780,113 @@ void __init tdx_early_init(void)
>>  
>>  	pr_info("Guest detected\n");
>>  }
>> +
>> +static long tdx_get_report(void __user *argp)
>> +{
>> +	u8 *reportdata, *tdreport;
>> +	struct tdx_report_req req;
>> +	u8 reserved[7] = {0};
>> +	long ret;
>> +
>> +	if (copy_from_user(&req, argp, sizeof(req)))
>> +		return -EFAULT;
>> +
>> +	/*
>> +	 * Per TDX Module 1.0 specification, section titled
>> +	 * "TDG.MR.REPORT", REPORTDATA length is fixed as
>> +	 * TDX_REPORTDATA_LEN, TDREPORT length is fixed as
>> +	 * TDX_REPORT_LEN, and TDREPORT subtype is fixed as
>> +	 * 0. Also check for valid user pointers and make sure
>> +	 * reserved entries values are zero.
>> +	 */
>> +	if (!req.reportdata || !req.tdreport || req.subtype ||
>> +		req.rpd_len != TDX_REPORTDATA_LEN ||
>> +		req.tdr_len != TDX_REPORT_LEN ||
>> +		memcmp(req.reserved, reserved, 7))
>> +		return -EINVAL;
> 
> Maybe make several checks instead of the monstrous one?

Agree. I will split it into two checks. One for spec related
checks and another for ABI validation.

> 
> !req.reportdata and !req.tdreport look redundant. copy_from/to_user() will
> catch them (and other bad address cases). And -EFAULT is more appropriate
> in this case.

We don't have to allocate kernel memory if we check it here. But I am not against
letting copy_from/to_user() handle it. I will remove the NULL check.

> 
>> +
>> +	reportdata = kmalloc(req.rpd_len, GFP_KERNEL);
>> +	if (!reportdata)
>> +		return -ENOMEM;
>> +
>> +	tdreport = kzalloc(req.tdr_len, GFP_KERNEL);
>> +	if (!tdreport) {
>> +		kfree(reportdata);
>> +		return -ENOMEM;
>> +	}
>> +
>> +	if (copy_from_user(reportdata, u64_to_user_ptr(req.reportdata),
>> +			   req.rpd_len)) {
>> +		ret = -EFAULT;
>> +		goto out;
>> +	}
>> +
>> +	/*
>> +	 * Generate TDREPORT using "TDG.MR.REPORT" TDCALL.
>> +	 *
>> +	 * Get the TDREPORT using REPORTDATA as input. Refer to
>> +	 * section 22.3.3 TDG.MR.REPORT leaf in the TDX Module 1.0
>> +	 * Specification for detailed information.
>> +	 */
>> +	ret = __tdx_module_call(TDX_GET_REPORT, virt_to_phys(tdreport),
>> +				virt_to_phys(reportdata), req.subtype,
>> +				0, NULL);
>> +	if (ret) {
>> +		ret = -EIO;
> 
> The spec says that it generate an error if invalid operand or busy. Maybe
> translate the TDX error codes to errnos?

User space has no need to know about the error code. In both error cases, if user
wants report, request has to re-submitted. So there is no use in translating
the error codes.

> 
> BTW, regarding busy case: do we want to protect against two parallel
> TDX_GET_REPORT? What happens if we run the second TDX_GET_REPORT when the
> first hasn't complete?

We don't have to protect against it here. It is a blocking call. So if user
makes a parallel request, we will wait for TDX Module to service them
in sequence.

> 
>> +		goto out;
>> +	}
>> +
>> +	if (copy_to_user(u64_to_user_ptr(req.tdreport), tdreport, req.tdr_len))
>> +		ret = -EFAULT;
>> +
>> +out:
>> +	kfree(reportdata);
>> +	kfree(tdreport);
>> +	return ret;
>> +}
>> +static long tdx_guest_ioctl(struct file *file, unsigned int cmd,
>> +			    unsigned long arg)
>> +{
>> +	void __user *argp = (void __user *)arg;
>> +	long ret = -ENOTTY;
> 
> Not a typewriter? Huh?

It is the recommended return code for invalid IOCTL commands.

https://www.kernel.org/doc/html/latest/driver-api/ioctl.html

> 
>> +
>> +	switch (cmd) {
>> +	case TDX_CMD_GET_REPORT:
>> +		ret = tdx_get_report(argp);
>> +		break;
>> +	default:
>> +		pr_debug("cmd %d not supported\n", cmd);
>> +		break;
>> +	}
>> +
>> +	return ret;
>> +}
>
Huang, Kai Sept. 13, 2022, 1:25 a.m. UTC | #7
On Fri, 2022-09-09 at 12:27 -0700, Kuppuswamy Sathyanarayanan wrote:
> diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
> index 928dcf7a20d9..8b5c59110321 100644
> --- a/arch/x86/coco/tdx/tdx.c
> +++ b/arch/x86/coco/tdx/tdx.c
> @@ -5,16 +5,21 @@
>  #define pr_fmt(fmt)     "tdx: " fmt
>  
>  #include <linux/cpufeature.h>
> +#include <linux/miscdevice.h>
> +#include <linux/mm.h>
> +#include <linux/io.h>

Sorry perhaps I am missing something, but what is the reason to include
<linux/mm.h>?

<linux/io.h> is for virt_to_phys()?

And should we explicitly include <linux/uaccess.h> for copy_{from|to}_user(),
and include the header (<linux/string.h> ?) for memchr_inv()?
Kuppuswamy Sathyanarayanan Sept. 13, 2022, 2:44 a.m. UTC | #8
On 9/12/22 6:25 PM, Huang, Kai wrote:
>> diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
>> index 928dcf7a20d9..8b5c59110321 100644
>> --- a/arch/x86/coco/tdx/tdx.c
>> +++ b/arch/x86/coco/tdx/tdx.c
>> @@ -5,16 +5,21 @@
>>  #define pr_fmt(fmt)     "tdx: " fmt
>>  
>>  #include <linux/cpufeature.h>
>> +#include <linux/miscdevice.h>
>> +#include <linux/mm.h>
>> +#include <linux/io.h>
> Sorry perhaps I am missing something, but what is the reason to include
> <linux/mm.h>?

It is included for kmalloc/kfree, file related structs and copy_{from|to}_user().

> 
> <linux/io.h> is for virt_to_phys()?

Yes

> 
> And should we explicitly include <linux/uaccess.h> for copy_{from|to}_user(),

mm.h covers it. So I don't think we should explicitly include it.

> and include the header (<linux/string.h> ?) for memchr_inv()?

One of the previous headers includes linux/string.h (I am not sure which one).
So why include it explicitly?
Huang, Kai Sept. 13, 2022, 5:03 a.m. UTC | #9
On Mon, 2022-09-12 at 19:44 -0700, Sathyanarayanan Kuppuswamy wrote:
> 
> On 9/12/22 6:25 PM, Huang, Kai wrote:
> > > diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
> > > index 928dcf7a20d9..8b5c59110321 100644
> > > --- a/arch/x86/coco/tdx/tdx.c
> > > +++ b/arch/x86/coco/tdx/tdx.c
> > > @@ -5,16 +5,21 @@
> > >  #define pr_fmt(fmt)     "tdx: " fmt
> > >  
> > >  #include <linux/cpufeature.h>
> > > +#include <linux/miscdevice.h>
> > > +#include <linux/mm.h>
> > > +#include <linux/io.h>
> > Sorry perhaps I am missing something, but what is the reason to include
> > <linux/mm.h>?
> 
> It is included for kmalloc/kfree, file related structs and copy_{from|to}_user().
> 
> > 
> > <linux/io.h> is for virt_to_phys()?
> 
> Yes
> 
> > 
> > And should we explicitly include <linux/uaccess.h> for copy_{from|to}_user(),
> 
> mm.h covers it. So I don't think we should explicitly include it.
> 
> > and include the header (<linux/string.h> ?) for memchr_inv()?
> 
> One of the previous headers includes linux/string.h (I am not sure which one).
> So why include it explicitly?
> 

OK.
Dave Hansen Sept. 13, 2022, 9:01 a.m. UTC | #10
On 9/12/22 19:44, Sathyanarayanan Kuppuswamy wrote:
>> and include the header (<linux/string.h> ?) for memchr_inv()?
> One of the previous headers includes linux/string.h (I am not sure which one).
> So why include it explicitly?

Because it's a best practice.  What happens is that you ride along on
the coat tails of another #include, someone sees that include is no
longer used and removes it.  Then, your code is busted on some weird
.config.

*OR*, the header itself changes and doesn't #include the dependency you
need.

I guess you can go add this advice to Documentation/ if it's not there
already somewhere.
Kuppuswamy Sathyanarayanan Sept. 13, 2022, 3:21 p.m. UTC | #11
Hi,

On 9/13/22 2:01 AM, Dave Hansen wrote:
> On 9/12/22 19:44, Sathyanarayanan Kuppuswamy wrote:
>>> and include the header (<linux/string.h> ?) for memchr_inv()?
>> One of the previous headers includes linux/string.h (I am not sure which one).
>> So why include it explicitly?
> Because it's a best practice.  What happens is that you ride along on
> the coat tails of another #include, someone sees that include is no
> longer used and removes it.  Then, your code is busted on some weird
> .config.
> 
> *OR*, the header itself changes and doesn't #include the dependency you
> need.
> 
> I guess you can go add this advice to Documentation/ if it's not there
> already somewhere.

Ok. I will include it explicitly.
Dave Hansen Sept. 14, 2022, 11:36 a.m. UTC | #12
On 9/9/22 12:27, Kuppuswamy Sathyanarayanan wrote:
> 
>  arch/x86/coco/tdx/tdx.c         | 115 ++++++++++++++++++++++++++++++++
>  arch/x86/include/uapi/asm/tdx.h |  56 ++++++++++++++++
>  2 files changed, 171 insertions(+)
>  create mode 100644 arch/x86/include/uapi/asm/tdx.h

The SEV equivalent of this in in:

	drivers/virt/coco/sev-guest/sev-guest.c

right?

Why did you choose a different location?  Also, can you please study the
SEV implementation a bit?  It might help you find problems like the
ioctl() return code issue.  The SEV driver appears to have gotten that
right.
Kuppuswamy Sathyanarayanan Sept. 14, 2022, 3:36 p.m. UTC | #13
On 9/14/22 4:36 AM, Dave Hansen wrote:
> On 9/9/22 12:27, Kuppuswamy Sathyanarayanan wrote:
>>
>>  arch/x86/coco/tdx/tdx.c         | 115 ++++++++++++++++++++++++++++++++
>>  arch/x86/include/uapi/asm/tdx.h |  56 ++++++++++++++++
>>  2 files changed, 171 insertions(+)
>>  create mode 100644 arch/x86/include/uapi/asm/tdx.h
> 
> The SEV equivalent of this in in:
> 
> 	drivers/virt/coco/sev-guest/sev-guest.c
> 
> right?
> 
> Why did you choose a different location?  Also, can you please study the

When we initially submitted the attestation patches, virt/coco folder
was not created. I initially kept this driver in platform/x86/, but
later moved to arch/x86/coco based on the review comments in v4. There
was a discussion about the need for a new config and the location of
the driver. The outcome of that discussion is, since this is not a
traditional driver, but a basic TDX feature, we don't need a special
config and the code can be maintained in the arch/x86/coco folder.

https://lore.kernel.org/lkml/YmEfgn7fMcZ2oCnr@zn.tnic/

> SEV implementation a bit?  It might help you find problems like the
> ioctl() return code issue.  The SEV driver appears to have gotten that
> right.

Ok.
Dave Hansen Sept. 14, 2022, 4:12 p.m. UTC | #14
On 9/14/22 08:36, Sathyanarayanan Kuppuswamy wrote:
> When we initially submitted the attestation patches, virt/coco folder
> was not created. I initially kept this driver in platform/x86/, but
> later moved to arch/x86/coco based on the review comments in v4. There
> was a discussion about the need for a new config and the location of
> the driver. The outcome of that discussion is, since this is not a
> traditional driver, but a basic TDX feature, we don't need a special
> config and the code can be maintained in the arch/x86/coco folder.

Could you please include the following in this set somewhere:

"The code to do the SEV analog of this TDX functionality is in
___insert_path_here____.   This code is different from that because
______reason______ so it is instead placed in ____other_path____."

?
Kuppuswamy Sathyanarayanan Sept. 14, 2022, 4:25 p.m. UTC | #15
On 9/14/22 9:12 AM, Dave Hansen wrote:
> On 9/14/22 08:36, Sathyanarayanan Kuppuswamy wrote:
>> When we initially submitted the attestation patches, virt/coco folder
>> was not created. I initially kept this driver in platform/x86/, but
>> later moved to arch/x86/coco based on the review comments in v4. There
>> was a discussion about the need for a new config and the location of
>> the driver. The outcome of that discussion is, since this is not a
>> traditional driver, but a basic TDX feature, we don't need a special
>> config and the code can be maintained in the arch/x86/coco folder.
> 
> Could you please include the following in this set somewhere:
> 
> "The code to do the SEV analog of this TDX functionality is in
> ___insert_path_here____.   This code is different from that because
> ______reason______ so it is instead placed in ____other_path____."
> 
> ?

Ok. I will include it in the cover letter.
Kuppuswamy Sathyanarayanan Sept. 15, 2022, 12:30 a.m. UTC | #16
On 9/14/22 9:12 AM, Dave Hansen wrote:
> On 9/14/22 08:36, Sathyanarayanan Kuppuswamy wrote:
>> When we initially submitted the attestation patches, virt/coco folder
>> was not created. I initially kept this driver in platform/x86/, but
>> later moved to arch/x86/coco based on the review comments in v4. There
>> was a discussion about the need for a new config and the location of
>> the driver. The outcome of that discussion is, since this is not a
>> traditional driver, but a basic TDX feature, we don't need a special
>> config and the code can be maintained in the arch/x86/coco folder.
> 
> Could you please include the following in this set somewhere:
> 
> "The code to do the SEV analog of this TDX functionality is in
> ___insert_path_here____.   This code is different from that because
> ______reason______ so it is instead placed in ____other_path____."
> 
> ?

I have also included info about why we don't use a separate config
option for it.

The code for the SEV equivalent of this TDX attestation functionality
can be found in drivers/virt/coco/sev-guest/. It is implemented as a
platform module driver, and it can be enabled using the CONFIG_SEV_GUEST
config option. However, in the case of TDX, it is implemented as a
built-in driver in the arch/x86/coco/tdx/tdx.c because of the following
reasons:

1. Attestation is expected to be needed by all distributions that support
   TDX. Therefore, using a separate configuration option is not necessary.
   With TDX support, it can be enabled by default, and a built-in driver
   model will work better in this use case.
2. Since it is not a conventional device driver and the code is very simple,
   creating an individual driver for it may be an overkill.
Greg KH Sept. 15, 2022, 11:07 a.m. UTC | #17
On Wed, Sep 14, 2022 at 05:30:45PM -0700, Sathyanarayanan Kuppuswamy wrote:
> I have also included info about why we don't use a separate config
> option for it.
> 
> The code for the SEV equivalent of this TDX attestation functionality
> can be found in drivers/virt/coco/sev-guest/. It is implemented as a
> platform module driver, and it can be enabled using the CONFIG_SEV_GUEST
> config option. However, in the case of TDX, it is implemented as a
> built-in driver in the arch/x86/coco/tdx/tdx.c because of the following
> reasons:
> 
> 1. Attestation is expected to be needed by all distributions that support
>    TDX. Therefore, using a separate configuration option is not necessary.
>    With TDX support, it can be enabled by default, and a built-in driver
>    model will work better in this use case.

No, that's not valid.  Distros want to enable everything, but only load
stuff that is only present.  You don't allow this for this code, which
isn't ok.

> 2. Since it is not a conventional device driver and the code is very simple,
>    creating an individual driver for it may be an overkill.

"simple" is not the issue, again, this should be a "normal" driver that
autoloads when the hardware is present and not load when the hardware is
not present.  This is not "special" to avoid the normal functionality of
all other drivers.

So again, no, this is not ok, fix this properly, don't be lazy.

thanks,

greg k-h
Greg KH Sept. 15, 2022, 11:09 a.m. UTC | #18
On Fri, Sep 09, 2022 at 12:27:06PM -0700, Kuppuswamy Sathyanarayanan wrote:
> +static int __init tdx_guest_init(void)
> +{
> +	int ret;
> +
> +	if (!cpu_feature_enabled(X86_FEATURE_TDX_GUEST))
> +		return -EIO;
> +
> +	ret = misc_register(&tdx_misc_dev);
> +	if (ret) {
> +		pr_err("misc device registration failed\n");
> +		return ret;
> +	}
> +
> +	return 0;
> +}
> +device_initcall(tdx_guest_init)

As mentioned elsewhere, make this a normal module_init() format and only
load the module if the hardware is present.  Don't just always be
built/loaded, that's not ok.

thanks,

greg k-h
Kuppuswamy Sathyanarayanan Sept. 15, 2022, 3:22 p.m. UTC | #19
Hi,

On 9/15/22 4:09 AM, Greg Kroah-Hartman wrote:
> On Fri, Sep 09, 2022 at 12:27:06PM -0700, Kuppuswamy Sathyanarayanan wrote:
>> +static int __init tdx_guest_init(void)
>> +{
>> +	int ret;
>> +
>> +	if (!cpu_feature_enabled(X86_FEATURE_TDX_GUEST))
>> +		return -EIO;
>> +
>> +	ret = misc_register(&tdx_misc_dev);
>> +	if (ret) {
>> +		pr_err("misc device registration failed\n");
>> +		return ret;
>> +	}
>> +
>> +	return 0;
>> +}
>> +device_initcall(tdx_guest_init)
> 
> As mentioned elsewhere, make this a normal module_init() format and only
> load the module if the hardware is present.  Don't just always be

This feature needs to be enabled by default for all valid TDX guests.

If TDX support is enabled and the guest is a valid TDX guest, the
"X86 FEATURE TDX GUEST" feature flag will be set. So looking for
"if(!cpu feature enabled(X86 FEATURE TDX GUEST))" will ensure that
the interface is only created in a valid TDX guest.

Even if we make it into a separate driver and use module init(), we'll
have to use the same "if(!cpu feature enabled(X86 FEATURE TDX GUEST))"
check to create and load the device. This approach was used in earlier
versions of this driver. We later changed it to initcall because it
appeared to be a roundabout approach.

Let me know if you still suggest to use module_init() model.

Following is the sample implementation with module_init() and this code
will be compiled with CONFIG_INTEL_TDX_GUEST=y.

+static struct platform_driver tdx_attest_driver = {
+	.probe		= tdx_attest_probe,
+	.remove		= tdx_attest_remove,
+	.driver		= {
+		.name	= DRIVER_NAME,
+	},
+};
+
+static int __init tdx_attest_init(void)
+{
+	int ret;
+
+	/* Make sure we are in a valid TDX platform */
+	if (!cpu_feature_enabled(X86_FEATURE_TDX_GUEST))
+		return -EIO;
+
+	ret = platform_driver_register(&tdx_attest_driver);
+	if (ret) {
+		pr_err("failed to register driver, err=%d\n", ret);
+		return ret;
+	}
+
+	pdev = platform_device_register_simple(DRIVER_NAME, -1, NULL, 0);
+	if (IS_ERR(pdev)) {
+		ret = PTR_ERR(pdev);
+		pr_err("failed to allocate device, err=%d\n", ret);
+		platform_driver_unregister(&tdx_attest_driver);
+		return ret;
+	}
+
+	return 0;
+}
+
+static void __exit tdx_attest_exit(void)
+{
+	platform_device_unregister(pdev);
+	platform_driver_unregister(&tdx_attest_driver);
+}
+
+module_init(tdx_attest_init);
+module_exit(tdx_attest_exit);

> built/loaded, that's not ok.



> 
> thanks,
> 
> greg k-h
Greg KH Sept. 16, 2022, 8:12 a.m. UTC | #20
On Thu, Sep 15, 2022 at 08:22:37AM -0700, Sathyanarayanan Kuppuswamy wrote:
> Hi,
> 
> On 9/15/22 4:09 AM, Greg Kroah-Hartman wrote:
> > On Fri, Sep 09, 2022 at 12:27:06PM -0700, Kuppuswamy Sathyanarayanan wrote:
> >> +static int __init tdx_guest_init(void)
> >> +{
> >> +	int ret;
> >> +
> >> +	if (!cpu_feature_enabled(X86_FEATURE_TDX_GUEST))
> >> +		return -EIO;
> >> +
> >> +	ret = misc_register(&tdx_misc_dev);
> >> +	if (ret) {
> >> +		pr_err("misc device registration failed\n");
> >> +		return ret;
> >> +	}
> >> +
> >> +	return 0;
> >> +}
> >> +device_initcall(tdx_guest_init)
> > 
> > As mentioned elsewhere, make this a normal module_init() format and only
> > load the module if the hardware is present.  Don't just always be
> 
> This feature needs to be enabled by default for all valid TDX guests.

Why?  What is so needed by userspace to require this brand new char
device node just to use TDX?

> If TDX support is enabled and the guest is a valid TDX guest, the
> "X86 FEATURE TDX GUEST" feature flag will be set. So looking for
> "if(!cpu feature enabled(X86 FEATURE TDX GUEST))" will ensure that
> the interface is only created in a valid TDX guest.

Yes, but that's not the point.  We don't just "build all drivers into
the kernel and only bind to hardware we actually have".  That's not how
Linux works, sorry.

> Even if we make it into a separate driver and use module init(), we'll
> have to use the same "if(!cpu feature enabled(X86 FEATURE TDX GUEST))"
> check to create and load the device. This approach was used in earlier
> versions of this driver. We later changed it to initcall because it
> appeared to be a roundabout approach.

Sorry, no, do it properly, have it be autoloaded only on the systems
that have the cpu feature.

> Let me know if you still suggest to use module_init() model.

Yes, it is a requirement.  Do you want every driver to try to copy what
you are doing here?

> Following is the sample implementation with module_init() and this code
> will be compiled with CONFIG_INTEL_TDX_GUEST=y.
> 
> +static struct platform_driver tdx_attest_driver = {
> +	.probe		= tdx_attest_probe,
> +	.remove		= tdx_attest_remove,
> +	.driver		= {
> +		.name	= DRIVER_NAME,
> +	},
> +};
> +
> +static int __init tdx_attest_init(void)
> +{
> +	int ret;
> +
> +	/* Make sure we are in a valid TDX platform */
> +	if (!cpu_feature_enabled(X86_FEATURE_TDX_GUEST))
> +		return -EIO;
> +
> +	ret = platform_driver_register(&tdx_attest_driver);
> +	if (ret) {
> +		pr_err("failed to register driver, err=%d\n", ret);
> +		return ret;
> +	}
> +
> +	pdev = platform_device_register_simple(DRIVER_NAME, -1, NULL, 0);
> +	if (IS_ERR(pdev)) {
> +		ret = PTR_ERR(pdev);
> +		pr_err("failed to allocate device, err=%d\n", ret);
> +		platform_driver_unregister(&tdx_attest_driver);
> +		return ret;
> +	}
> +
> +	return 0;
> +}
> +
> +static void __exit tdx_attest_exit(void)
> +{
> +	platform_device_unregister(pdev);
> +	platform_driver_unregister(&tdx_attest_driver);
> +}
> +
> +module_init(tdx_attest_init);
> +module_exit(tdx_attest_exit);

Sorry, no, this too is not ok as you are not telling userspace if it
needs to load your driver or not automatically.  Please do this
properly.

Basic issues like this shouldn't be showing up in v13 of a patch series.
Please take the time and start over and go and get a lot of internal
review before sending anything out again.

thanks,

greg k-h
diff mbox series

Patch

diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
index 928dcf7a20d9..8b5c59110321 100644
--- a/arch/x86/coco/tdx/tdx.c
+++ b/arch/x86/coco/tdx/tdx.c
@@ -5,16 +5,21 @@ 
 #define pr_fmt(fmt)     "tdx: " fmt
 
 #include <linux/cpufeature.h>
+#include <linux/miscdevice.h>
+#include <linux/mm.h>
+#include <linux/io.h>
 #include <asm/coco.h>
 #include <asm/tdx.h>
 #include <asm/vmx.h>
 #include <asm/insn.h>
 #include <asm/insn-eval.h>
 #include <asm/pgtable.h>
+#include <uapi/asm/tdx.h>
 
 /* TDX module Call Leaf IDs */
 #define TDX_GET_INFO			1
 #define TDX_GET_VEINFO			3
+#define TDX_GET_REPORT			4
 #define TDX_ACCEPT_PAGE			6
 
 /* TDX hypercall Leaf IDs */
@@ -775,3 +780,113 @@  void __init tdx_early_init(void)
 
 	pr_info("Guest detected\n");
 }
+
+static long tdx_get_report(void __user *argp)
+{
+	u8 *reportdata, *tdreport;
+	struct tdx_report_req req;
+	u8 reserved[7] = {0};
+	long ret;
+
+	if (copy_from_user(&req, argp, sizeof(req)))
+		return -EFAULT;
+
+	/*
+	 * Per TDX Module 1.0 specification, section titled
+	 * "TDG.MR.REPORT", REPORTDATA length is fixed as
+	 * TDX_REPORTDATA_LEN, TDREPORT length is fixed as
+	 * TDX_REPORT_LEN, and TDREPORT subtype is fixed as
+	 * 0. Also check for valid user pointers and make sure
+	 * reserved entries values are zero.
+	 */
+	if (!req.reportdata || !req.tdreport || req.subtype ||
+		req.rpd_len != TDX_REPORTDATA_LEN ||
+		req.tdr_len != TDX_REPORT_LEN ||
+		memcmp(req.reserved, reserved, 7))
+		return -EINVAL;
+
+	reportdata = kmalloc(req.rpd_len, GFP_KERNEL);
+	if (!reportdata)
+		return -ENOMEM;
+
+	tdreport = kzalloc(req.tdr_len, GFP_KERNEL);
+	if (!tdreport) {
+		kfree(reportdata);
+		return -ENOMEM;
+	}
+
+	if (copy_from_user(reportdata, u64_to_user_ptr(req.reportdata),
+			   req.rpd_len)) {
+		ret = -EFAULT;
+		goto out;
+	}
+
+	/*
+	 * Generate TDREPORT using "TDG.MR.REPORT" TDCALL.
+	 *
+	 * Get the TDREPORT using REPORTDATA as input. Refer to
+	 * section 22.3.3 TDG.MR.REPORT leaf in the TDX Module 1.0
+	 * Specification for detailed information.
+	 */
+	ret = __tdx_module_call(TDX_GET_REPORT, virt_to_phys(tdreport),
+				virt_to_phys(reportdata), req.subtype,
+				0, NULL);
+	if (ret) {
+		ret = -EIO;
+		goto out;
+	}
+
+	if (copy_to_user(u64_to_user_ptr(req.tdreport), tdreport, req.tdr_len))
+		ret = -EFAULT;
+
+out:
+	kfree(reportdata);
+	kfree(tdreport);
+	return ret;
+}
+static long tdx_guest_ioctl(struct file *file, unsigned int cmd,
+			    unsigned long arg)
+{
+	void __user *argp = (void __user *)arg;
+	long ret = -ENOTTY;
+
+	switch (cmd) {
+	case TDX_CMD_GET_REPORT:
+		ret = tdx_get_report(argp);
+		break;
+	default:
+		pr_debug("cmd %d not supported\n", cmd);
+		break;
+	}
+
+	return ret;
+}
+
+static const struct file_operations tdx_guest_fops = {
+	.owner		= THIS_MODULE,
+	.unlocked_ioctl	= tdx_guest_ioctl,
+	.llseek		= no_llseek,
+};
+
+static struct miscdevice tdx_misc_dev = {
+	.name           = TDX_GUEST_DEVICE,
+	.minor          = MISC_DYNAMIC_MINOR,
+	.fops           = &tdx_guest_fops,
+};
+
+static int __init tdx_guest_init(void)
+{
+	int ret;
+
+	if (!cpu_feature_enabled(X86_FEATURE_TDX_GUEST))
+		return -EIO;
+
+	ret = misc_register(&tdx_misc_dev);
+	if (ret) {
+		pr_err("misc device registration failed\n");
+		return ret;
+	}
+
+	return 0;
+}
+device_initcall(tdx_guest_init)
diff --git a/arch/x86/include/uapi/asm/tdx.h b/arch/x86/include/uapi/asm/tdx.h
new file mode 100644
index 000000000000..687c86c9e3fb
--- /dev/null
+++ b/arch/x86/include/uapi/asm/tdx.h
@@ -0,0 +1,56 @@ 
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+#ifndef _UAPI_ASM_X86_TDX_H
+#define _UAPI_ASM_X86_TDX_H
+
+#include <linux/types.h>
+#include <linux/ioctl.h>
+
+#define TDX_GUEST_DEVICE		"tdx-guest"
+
+/* Length of the REPORTDATA used in TDG.MR.REPORT TDCALL */
+#define TDX_REPORTDATA_LEN              64
+
+/* Length of TDREPORT used in TDG.MR.REPORT TDCALL */
+#define TDX_REPORT_LEN                  1024
+
+/**
+ * struct tdx_report_req: Get TDREPORT using REPORTDATA as input.
+ *
+ * @reportdata     : User-defined REPORTDATA to be included into
+ *                   TDREPORT. Typically it can be some nonce
+ *                   provided by attestation service, so the
+ *                   generated TDREPORT can be uniquely verified.
+ * @tdreport       : TDREPORT output from TDCALL[TDG.MR.REPORT].
+ * @rpd_len        : Length of the REPORTDATA (fixed as 64 bytes by
+ *                   the TDX Module specification, but parameter is
+ *                   added to handle future extension).
+ * @tdr_len        : Length of the TDREPORT (fixed as 1024 bytes by
+ *                   the TDX Module specification, but a parameter
+ *                   is added to accommodate future extension).
+ * @subtype        : Subtype of TDREPORT (fixed as 0 by TDX Module
+ *                   specification, but added a parameter to handle
+ *                   future extension).
+ * @reserved       : Reserved entries to handle future requirements.
+ *                   Default acceptable value is 0.
+ *
+ * Used in TDX_CMD_GET_REPORT IOCTL request.
+ */
+struct tdx_report_req {
+	__u64 reportdata;
+	__u64 tdreport;
+	__u32 rpd_len;
+	__u32 tdr_len;
+	__u8  subtype;
+	__u8 reserved[7];
+};
+
+/*
+ * TDX_CMD_GET_REPORT - Get TDREPORT using TDCALL[TDG.MR.REPORT]
+ *
+ * Return 0 on success, -EIO on TDCALL execution failure, and
+ * standard errno on other general error cases.
+ *
+ */
+#define TDX_CMD_GET_REPORT		_IOWR('T', 0x01, __u64)
+
+#endif /* _UAPI_ASM_X86_TDX_H */