Attacking Intel Trusted Execution Technology

Rafal Wojtczuk
rafal@invisiblethingslab.com

Joanna Rutkowska
joanna@invisiblethingslab.com

---===[ Invisible Things Lab ]===---

Abstract
In this paper we present the results of our research into security of the Intel Trusted
Execution Technology, part of the vProTM brand. We describe a practical attack that is
capable of bypassing the TXT's trusted boot process, a key building block for Intel's vision of
Trusted Computing. As part of the attack we also discuss practical attacks on SMM memory
in modern Intel systems.
keywords: Trusted Computing, Trusted Execution Technology, System Management Mode,
TXT, SMM, STM, BIOS, security, analysis, attacks.

1.

Introduction

and storing them in particular TPM registers. What

is extraordinary here is that TXT doesn't make any
assumptions about the state of the system before
loading the software, thus making it possible for a
user to ensure secure load of an OS or VMM, even
in a potentially compromised machine.

Trusted Computing is becoming a part of our lives,

whether we want it or not. These days almost every
new laptop comes with an on-board Trusted
Platform Module (TPM). Some of the Microsoft's
Palladium technologies made their way into Vista,
and Microsoft BitLocker is, without doubt, the most
successful, widely deployed product that is based
on the idea of Trusted Computing[8].

In other words, our system can be full of boot

sector viruses and BIOS rootkits, and god-knowswhat-else, and still TXT should allow to load a
clean VMM (or OS kernel) in a secure way, immune
to all those malware present in the system. This
TXT-supported load process is called Late Launch,
and is implemented via a special new CPU
instruction called SENTER. A good introduction to
Intel TXT architecture can be found in the David
Grawrock's book[3]. For a detailed technical
specification one should consult the Intel TXT
documentation[4].

On the hardware side, besides the famed TPM, we

also have had the LaGrande technology.
LaGrande, recently renamed Trusted Execution
Technology (TXT)[4], is Intel's response to the
Trusted Computing trend. TXT is currently part of
the vPro brand[5], and for about a year now
users can buy a vPro/TXT compatible hardware in
regular computer stores (the first one was the
DQ35J desktop board[7] that worked with certain
Core 2 Duo processors1).

We shall stress that TXT has not been designed to

provide runtime protection, e.g. against a buffer
overflow in a hypervisor code. TXT is supposed to
provide only the launch-time protection, i.e. ensure
that the code we load, at the moment of loading, is
what we really intended to load.

TXT is not an alternative to a TPM, in fact TXT

heavily relies on the TPM to provide basic services
like e.g. secure storage of measurements done by
the TXT. Also, Palladium, or whatever it is called
these days, is not a competition to TXT. Intel TXT
can provide building blocks to e.g. Vista Bitlocker,
arguably making it more secure than it is now.

It's worth mentioning AMD has its own version of a

late launch, implemented via an SKINIT instruction.
We haven't looked at the AMD technology
thoroughly yet, so we will refrain from commenting
on this any further.

The sole purpose of Intel TXT technology is to

provide a trusted way for loading and executing
system software, e.g. Operating System kernel or
Virtualization Machine Monitor (VMM). This is
achieved by performing software measurements

The late launch is a pretty amazing technology,

when we think about. It promises to effectively

1TXT requires support from both the CPU and the chipset

provide all the benefits of a computer restart

without actually restarting it.

TXT's late launch functionality. Tboot is also part of

the mainstream opensource Xen hypervisor[12].

It is hard to overemphasize the potential impact

that a technology such as TXT could have on
computer security. One can immediately see it
could provide basic building blocks that could be
later used
to eliminate all the system-level
persistent malware2 in other words we should be
able to easily build systems (VMMs or even
standard OSes) that would be immune to attacks
that try to compromise system binaries on disk, or
attack the system right from the bootloader or
BIOS. Combining this with VT-x and VT-d
technologies, system developers (for the first time,
at least as far as the "PC" platform is considered)
have gotten extremely strong tools into their hands
that should allow them to create really secure
VMMs and OSes

Normally tboot should ensure that when a correct,

i.e. unmodified, Xen hypervisor is loaded, and only
then, correct measurements will be loaded into
TPM registers. In practice this means that only a
chosen (trusted) version of the Xen hypervisor will
get access to certain secrets sealed in the TPM,
and/or will be able to positively authenticate itself to
some peer (e.g. system administrator's laptop)
using a special feature of a TPM called Remote
Attestation.
With our attack we show that by infecting an SMM
handler, we can modify at will the just-loaded (and
just-measured) VMM. In other words the attacks
completely bypass all the security functionality that
is supposed to be provided by the TXT for the
purpose of trusted boot. This is a direct result of the
SMM code surviving the late launch process in an
unmodified form whether compromised or not.

Let us now describe how we have attacked this

new exciting technology

2.

Attacking Intel TXT

We describe the details of how to attack an SMM

handler in the next chapter.

TXT's key functionality, the late launch, is often

"advertised" as a way to load and start a piece of
trusted code (usually a VMM), no matter what is the
state of the system, at the moment just before
performing the launch. That is, however, not fully
correct. In fact there is one piece of system
software that should be trusted This system
software is a so called System Management Mode,
which we discuss in more detail in the next chapter.

Intel's remedy to malicious SMM handler is called

STM, which stands for SMM Transfer Monitor. The
purpose of STM is to sandbox the existing SMM
handler by virtualizing it using VT-x and VT-d
technologies. STM should be thought of as of a
peer hypervisor to the VMM that is being loaded
using late launch. STM is supposed to be
measured during the late launch process.

SMM, being the most privileged type of software

that ever executes on a CPU (see below), can
bypass security protections imposed by the late
launch process on a newly loaded VMM.
Unfortunately, the assumption that SMM can be
always trusted is incorrect, as we demonstrate here
with this research.

Unfortunately STM is, as of today, not available.

We discuss this in more detail in the last chapter of
this paper.

Indeed, one can imagine the following 2-stage

attack on the Intel TXT's late launch functionality:

System Management Mode is the most privileged

execution mode on x86/x86_64 architectures, even
more privileged than ring 0 mode and a hardware
hypervisor (VT/AMD-v), often referred to as "ring
-1".

1.

Infecting the system's SMM handler,

2.

Compromising the just-securely-loaded code

(in our example the Xen hypervisor) from
within the infected SMM handler.

3.

SMM Attacks

SMM aka "Ring -2"

One reason for considering this code to be so

privileged, is that an SMM code can access the
whole system memory, including the kernel and
hypervisor memory. Standard OS memory
protection mechanisms (Page Tables), as well as
hypervisor memory virtualization (Shadow Paging,
Nested Paging/EPT, IOMMU/VT-d), do not work
against the SMM code.

We have implemented a proof-of-concept code that

demonstrates the above attack scheme against the
Xen hypervisor loaded using tboot[6], the Intel's
open source implementation of trusted boot. Tboot
provides trusted boot for Linux and Xen using Intel

2 By a system-level malware we mean malware which is resident because of replacing/infecting crucial,privileged, nonvolatile code (e.g. a kernel binary, dll/sys
binary, or boot sector, or bios image)

Also, on Intel processors, an SMM code can

"preempt" even the VT-x hypervisor, whenever an
SMI interrupt is signaled. There exists a way3 for
the hypervisor to establish a special entity, called
STM, that would be able to intercept those SMI
interrupts, and we will discuss it later in this paper.

Previous SMM-related research

Although there have been several publications, in
the recent years, concerning the SMM-related
security issues ([2], [10], [1]), they were all
concerned about the security implications resulting
from an attacker gaining access to the SMM
memory, without however focusing on how to
bypass system SMM memory protection. That was
mostly because until recent years, it was
straightforward to gain a read-write access to the
SMM memory by only setting appropriate chipset
registers, so no special attack method was needed.

Consequently, it will be not much of an

exaggeration to name SMM a "ring -2".
Difficulties with finding flaws in SMM
A researcher willing to examine an SMM code for
potential security problems faces a non-trivial
problem the SMM memory is not accessible to
anyone except the SMM itself. Thus there is no
way for the attacker to get access to the actual
image of the SMM code. Without having access to
the (binary) code, one cannot, obviously, look for
potential flaws there. So, we came to a, somewhat
discouraging, conclusion that:

Today most modern systems take special steps in

order to protect the SMM memory from even a
read-only access from the OS, including the kernel
and/or the hypervisor.
Example #1: The Q35 remapping bug
During one of our presentations on Xen security at
the Black Hat conference in August 2008, we have
mentioned a bug we found in a DQ35JO
motherboard 4 BIOS that allowed us to e.g. bypass
the Xen hypervisor memory protection. A few
weeks later Intel has released an updated BIOS
that fixed the bug we exploited, and we have
published full details about the attack[9], together
with a proof-of-concept code[11].

Without having at least one bug in an SMM code

that could be exploited to read the SMM memory,
one cannot... search for bugs in the SMM memory!
That is indeed a nice example of a vicious circle
and partly explains why so little research has been
done in the area of SMM attacks (see the
discussion later).
One might be tempted to think that an easy shortcut to read the SMM code (and the whole BIOS),
would be to de-solder the SPI-flash chip from the
motherboard and read its memory using a special,
so called, programmer device. Unfortunately this
approach is not that straightforward as it might
seem. It turns out that BIOS code, as seen on the
flash chip, is usually heavily packed with custom
packing algorithms and its unpacking presents as
similar challenge as e.g. unpacking/deobfuscation
of modern malware, with the difference that code
emulation is extremely difficult to implement in the
process of unpacking the BIOS code. After all a
BIOS code is supposed to be executed in a very
unusual I/O environment, which is hard to emulate
properly.

The attack exploits the Intel chipset's memory

remapping (AKA memory reclaiming) feature and
allows to circumvent certain CPU- or chipsetimposed memory protection mechanisms. This
includes ability to bypass an SMM memory
protection, allowing the attacker to gain full access
to the SMM memory.
This bug, and a resulting attack, turned out to be
crucial with our further SMM research. It allowed us
to analyze the SMM binary code and find further
security problems there, which we describe below.
Example #2: VU#127284
On December 10th, 2008, we have reported the
discovery of a few new SMM bugs to Intel Product
Security Response Center. All those SMM bugs
result from a single design decision to implement
certain functionality in an unsafe way. This single
design decision has lead to some 40+ places in the
SMM handler5 , where each might potentially
introduce code execution vulnerability in SMM

Consequently we have taken a different approach,

that is described later. Before we proceed, let's take
a quick look at the research done by other authors
that involves SMM and security.

3 It is called "dual-monitor mode" by the Intel System Developers Manual.

4 Back then it was the most modern desktop motherboard from Intel available in shops.
5 We have looked at the DQ35JOE's motherboard's BIOS handler, with all the latest patches as of December 2008.

4.

mode. We have successfully exploited only two of

those implementation errors. We have seen no
point in trying to exploit others. The correct solution
to this problem should be based on redesigning the
current SMM handlers. One should not confuse this
design mistake in the SMM handler with the design
problem affecting the security of TXT, that we
discuss in the next chapter.

The TXT design problem

We have mentioned above that a remedy to

malicious SMM handlers is called an STM. We also
said that no STM, as of today, is unfortunately
available on the market, which yields our attack
applicable to all current systems. One aim of our
research, besides having fun and all, is to stimulate
developers to create an STM.

Those recent SMM bugs have still not been

patched by Intel. Intel confirmed 6 the issue in
"mobile, desktop, and server motherboards",
without providing any more details about which
exact models are vulnerable. We suspect it might
affect all recent Intel motherboards/BIOSes.

So, here comes the first question about STM: who

should be in charge of creating an STM? We have
been told by Intel engineers that STM should be
created by OEMs, or BIOS vendors. Apparently
STM should be part of a system BIOS, just like an
SMM is.

Intel estimates the firmware patches to be ready

before summer 2009. Intel requested that we
withhold the details about the SMM bugs until the
patches are available. We currently planning to
disclose them at the Black Hat USA 2009
conference, that takes place at the end of July
2009.

This rises a second question though: if we do not

trust OEMs to produce flawless SMM code(s), why
should we trust them to create a flawless STM?
After all, STM seems to us as something much
more complex than a typical SMM STM is a
hypervisor, a non-trivial hypervisor that should
provide memory, I/O and CPU virtualization to the
SMM handler. Moreover STM should work in
parallel with an existing VMM(s), like e.g. Xen.
There should be a way for the two hypervisors to
talk to each other, which, in turn, should require a
standardized inter-hypervisor protocol. Needless to
say, Intel does not provide any publicly available
documentation on how to write a working STM 7.

Intel told us that they have also notified CERT CC

about this problem, because they believed similar
SMM bugs might be present in other vendors'
BIOSes. CERT CC has assigned the following
tracking number to this issue: VU#127284.
Summary of the SMM issues
With our two distinct attacks on the recent SMM
handlers, we have shown that even the latest
systems don't correctly protect its most privileged
software layer, i.e. the System Management Mode.
In some aspects, an SMM code is even more
privileged than the hardware hypervisor, because it
has access to the whole system memory, including
the hypervisor memory, but not vice versa.

Intel claims8, however, that STM is not that difficult

to write and that Intel will provide detailed
specification on how to write one in the near
future 9. Intel argues that STM can be made more
secure than a typical SMM because it might not
need to be modified as often as SMM handlers are.
SMM needs to be "tuned" to each new
motherboard/system, while an STM should be fairly
generic. This could allow to have just a few mature
(and well audited) STMs in existence.

We should note, however, that our SMM attacks

still do not allow to e.g. re-flash the BIOS. Today,
most BIOSes are well protected against re-flashing
with unsigned images. Also our attacks do not
affect the Intel AMT technology, which is
independent from the main processor and does not
rely on SMM security.

The third question comes to mind however: if

indeed STMs were so easy to write, why would
Intel still hadn't created one? After all, TXT-capable
hardware has been available in shops for over a
year now, and also the Intel's tboot project is more
than a year old.

6 Private communication between ITL (we) and Intel.

7 In fact the term STM is not defined in any of the Intel documents available on intel.com website. The term and its purpose is only introduced in the, already
mentioned, David Grawrock's book[3].
8 Private communication between ITL (we) and Intel.
9 Intel claims the specification is already available for select vendors under NDA.

Intel's counter-argument10 is that there has not

been enough "market demand" for an STM as of
yet, and consequently OEMs have not been
interested in developing an STM.

[2]

Loic Duflot. Security Issues Related to Pentium

System Management Mode. Presented at
CanSecWest 2006, Vancouver, Canada, 2006.
[3] David Grawrock. The Intel Safer Computing
Initiative: Building Blocks for Trusted Computing
(Computer System Design). Intel Press, 2006.
[4] Intel Corp. Intel Trusted Execution Technology.
http://www.intel.com/technology/security/,
[5] Intel Corp. Intel vPro Technology. http://
www.intel.com/technology/vpro/index.htm,
[6] Intel Corp. Trusted Boot (tboot). http://
sourceforge.net/projects/tboot, 2007.
[7] Intel Corp. Intel Desktop Board DQ35JO. http://
www.intel.com/products/desktop/motherboards/
dq35jo/dq35jo-overview.htm, 2008.
[8] Joanna Rutkowska. Why do I miss Microsoft
BitLocker? http://
theinvisiblethings.blogspot.com/2009/01/why-doi-miss-microsoft-bitlocker.html, 2009.
[9] Joanna Rutkowska and Rafal Wojtczuk. Detecting
& Preventing the Xen Hypervisor Subversions.
Presented at Black Hat USA, Las Vegas, NV,
USA, 2008.
[10] Sherri Sparks and Shawn Embleton. SMM
Rootkits: A New Breed of OS Independent
Malware. Presented at Black Hat USA, Las
Vegas, NV, USA, 2008.
[11] Rafal Wojtczuk, Joanna Rutkowska, and
Alexander Tereshkin. Xen 0wning Trilogy: code
and demos. http://invisiblethingslab.com/
resources/bh08/, 2008.
[12] Xen Hypervisor. http://xen.org/

We cannot agree with this counter-argument,

however. First, we shall not forget that Intel is
also an OEM/BIOS vendor it does sell
motherboards and does make BIOSes for them.
Second, when it comes to security, one should not
use "market demand" as an excuse. If we followed
this line of reasoning, we might very well never
started using the snprintf() function;)
Consequently, we should ask if that was indeed a
good idea to design TXT in such a way that it
requires additional, probably quite complex, entity
called STM to function correctly? Maybe it was a
mistake to allow TXT to function without STM? After
all the whole point about TXT was to get rid of the
Static Root of Trust Measurement, in favor of the
Dynamic Trust Measurement. Assuming even that
STM itself will not contain any flaws it is unclear
to us whether integration of STM and TXT can be,
and will be, done securely; we cannot say more on
the topic until we can evaluate an actual STM
implementation.

5.

Summary

We have described a successful attack against

Intel Trusted Execution technology. We have also
implemented a working proof-of-concept exploit
that works against Xen loaded via tboot Intel's
implementation of TXT-based trusted boot.
Probably equally interesting as the TXT itself is
another aspect of our research the novel
research into SMM attacks. SMM attacks have
more implications than just attacking TXT SMM
compromises can also be used for creating
advanced rootkits, backdoors and trojans. SMM
bugs can also be used to compromise hypervisor
memory in virtualized systems, even such
advanced as Xen.
The details of our new SMM attacks will be made
available once Intel patches its firmware, most
likely we will present them at the Black Hat USA
conference in summer 2009. We will also make the
code of our TXT exploit available.

References
[1]

BSDaemon, Coideloko, and D0Nand0N. System

Management Mode Hack: Using SMM for "Other
Purposes". In Phrack Magazine, Vol 0x0c, Issue
0x41, 2008.

10 Private communication between ITL (we) and Intel.

http://invisiblethingslab.com