Re: [PATCH v3 4/4] platform/x86: intel_tdx_attest: Add TDX Guest attestation interface driver

From: Sathyanarayanan Kuppuswamy  <sathyanarayanan.kuppuswamy@linux.intel.com>
To: Isaku Yamahata <isaku.yamahata@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
	Dave Hansen <dave.hansen@linux.intel.com>,
	x86@kernel.org, Hans de Goede <hdegoede@redhat.com>,
	Mark Gross <mgross@linux.intel.com>,
	"H . Peter Anvin" <hpa@zytor.com>,
	"Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>,
	Tony Luck <tony.luck@intel.com>, Andi Kleen <ak@linux.intel.com>,
	linux-kernel@vger.kernel.org,
	platform-driver-x86@vger.kernel.org
Subject: Re: [PATCH v3 4/4] platform/x86: intel_tdx_attest: Add TDX Guest attestation interface driver
Date: Thu, 21 Apr 2022 07:44:19 -0700	[thread overview]
Message-ID: <70436892-0b4f-30c5-051b-2c86f7387101@linux.intel.com> (raw)
In-Reply-To: <20220421070444.GB1423762@private.email.ne.jp>

On 4/21/22 12:04 AM, Isaku Yamahata wrote:
> On Tue, Apr 19, 2022 at 06:26:43PM -0700,
> Sathyanarayanan Kuppuswamy <sathyanarayanan.kuppuswamy@linux.intel.com> wrote:
> 
>> On 4/19/22 6:20 PM, Isaku Yamahata wrote:
>>> If timeout occurs, the state of adev->tdquote_buf is unknown.  It's not safe
>>> to continue to using adev->tdquote_buf.  VMM would continue to processing
>>> getquote request with this buffer.  What if TDX_CMD_GEN_QUOTE is issued again,
>>> and tdquote_buf is re-used?
>>
>> This part is not clearly discussed in the specification. May be spec
>> should define some reasonable timeout and teardown details.
>>
>> Regarding not using this buffer again, what happens if we de-allocate
>> it on timeout and the host still updates it?
> 
> Until GET_QUOTE_IN_FLIGHT is cleared, the shared page is owned by VMM, TD
> attestation driver shouldn't reuse/free the pages.
> 
> In the case of this driver, I think of two options
> - don't timeout. wait for interrupt to arrive and check the shared GPA state.
> - allow timeout. When the next request comes, check the shared GPA state.
>    If it's still GET_QUOTE_IN_FLIGHT, return EBUSY.

Out of the above two options, I think option 1 is better. It is simpler
to serialize them using mutex compared to checking the shared buffer of
previous request.

> 
> It's possible for VMM to keep the shared GPA forever maliciously(DoS) or
> unintentionally due to bug.  TD can't do much about it.

I will add a note about it in commit log.

-- 
Sathyanarayanan Kuppuswamy
Linux Kernel Developer