This document details the procedure for performing microcode updates on the AMD K8 processors. It also gives background information on the K8 microcode design and provides information on altering the microcode and loading the altered update for those who are interested in microcode hacking.
Source code is included for a simple Linux microcode update driver for those who want to update their K8's microcode without waiting for the motherboard vendor to add it to the BIOS. The latest microcode update blocks are included in the driver.
The information has been provided by Anonymous.
Modern x86 microprocessors from Intel and AMD contain a feature known as "microcode update", or as the vendors prefer to call it, "BIOS update". Essentially the processor can reconfigure parts of its own hardware to fix bugs ("errata") in the silicon that would normally require a recall.
This is done by loading a block of "patch data" created by the CPU vendor into the processor using special control registers. Microcode updates essentially override hardware features with sequences of the internal RISC-like micro-ops (uops) actually executed by the processor. They can also replace the implementations of microcoded instructions already handled by hard-wired sequences in an on-die microcode ROM.
AMD's U.S. Patent 6438664 ("Microcode patch device and method for patching microcode using match registers and patch routines") goes into substantial detail on this.
Typically microcode update blocks are stored in the BIOS flash ROM and loaded into the processor as the system boots. They can also be loaded by the operating system; for instance, Linux contains a microcode device driver for Intel chips.
AMD recently released a "BIOS fix" to motherboard makers to address Errata 109, in which REP MOVS instructions caused subsequent instructions to be skipped under specific pipeline conditions.
Previously it was not clear if and how AMD even supported microcode updates in the K8 family until this announcement. After analyzing a number of BIOS images, it appears that AMD has secretly used the microcode update facility on several occasions over the past few years, but obviously avoided publicly disclosing that it actually had bugs patchable in this manner.
Early K7 (Athlon) cores initially supported microcode updates as well, until ironically the microcode update mechanism itself was found to be broken and subsequently listed as an erratum.
The following sections describe the microcode update procedure, obtained by clean room reverse engineering various vendors' BIOS code. The actual microcode update blocks are embedded in the BIOS image; the most recent updates (created June 2004) have been included in the Linux driver source code attached to this description.
Microcode Update Procedure:
The update procedure expects the 64-bit virtual address of the update data, including the 64 byte header, to be in edx:eax:
edx = high 32 bits of 64-bit virtual address
eax = low 32 bits of 64-bit virtual address
ecx = 0xc0010020 (MSR to trigger update)
Execute wrmsr with these register values. If the address and update block data are valid, wrmsr completes successfully. Otherwise, a GP fault is taken.
The microcode does not appear to update MSR 0x8B with the new update signature as it does on Intel processors, despite the fact that some BIOS code that was analyzed does seem to check this field. It is possible the MSR is only updated under certain conditions, for instance when microcode is loaded before initializing the cache controller. Nonetheless, as we shall see below, the processor is clearly doing something internally when it claims to accept an update in this manner.
The update generally takes around 5500 clock cycles. This was tested on an Athlon 64 with CPUID 0x0F48. It was not tested on any other K8 cores, although the driver source code includes updates for CPUIDs 0x0F4A and 0x0F50.
Microcode Block Format:
The microcode block consists of a 64-byte header and an 896-byte data area. The processor consumes both the header and data area during an update. The header consists of various fields. The most important ones are:
- Offset 0: 32-bit word for update creation date (e.g. 0x20040602)
- Offset 12: 32-bit checksum: sum of all 32-bit words in the data area
- Offset 24: Low 8 bits of the cpuid (e.g. 0xf48 -> 0x48).
- Offset 28: 4 bytes: 0x01 0xaa 0xaa 0xaa (evidently a reference to (A)MD.)
The microcode blocks are typically padded out to 2048 bytes, just as the Intel format blocks are.
Surprisingly, the microcode itself is in no way encrypted as it is in Intel microcode updates; the raw data loaded into the microcode patch array is directly exposed. The repetitive structure of the data, bit patterns and fields characteristic of microcode indicate that apparently no encryption was performed.
U.S. Patent 6438664 describes the most probable structure of this data; the bit patterns in the update blocks show the outline of the uop triads and control fields known to exist in K8 microcode. Further analysis of the microcode format is in progress.
Even more surprising is the total lack of strong authentication that the update block has not been damaged or altered. The processor's sole means of validating an update is to take the sum of all 32-bit words in the 896 byte update block and compare it to the 32-bit checksum at offset 12; this verification is done by microcode already stored in the microcode ROM.
Modifying random bits within the update block was tested, regenerating a correct checksum, and loading the block into the processor. In many cases the processor accepts the block with no visible effects; other cases cause a spontaneous reboot.
Most alarming is the way in which certain bit modifications cause the processor to perform very bizarrely, for instance raising segfaults and performing incorrect computations on certain microcoded instructions.
The processor also apparently does not check the header to see if the loaded update matches its exact model and stepping; it is possible to load updates intended for an Opteron onto an Athlon 64 CPU, although this will crash the machine or cause bizarre behavior.
Depending on which data block bits are modified, loading an invalid update apparently causes an internal fault and the CPU spontaneously reboots.
The ability to fundamentally alter instruction decoding and execution on AMD K8 processors is sure to interest hardware hackers everywhere.
Unfortunately, it is not clear if this has much practical use. The updates are structured to patch specific microcode lines, and there are a very limited number of patch slots available (around 64 if the patented technique was actually implemented as described). Adding useful new instructions to the ISA is therefore unlikely; at best we could enable a previously undefined opcode to execute a few lines of uops and return. The primary purpose of microcode patching is to modify or disable defective functionality, rather than add new features.
Interestingly, this does have serious implications for system security. If one is able to get root access on a machine even once, it is hypothetically possible to install a microcode update specifically to help compromise security from userspace at a later time. Such an update could be flashed into the BIOS to make it persistent across reboots.
For instance, by patching the appropriate microcode lines, it may be possible to catch an opcode that would normally be illegal, and instead handle it by tricking the TLB into thinking we're in kernel mode when in fact the attacker has only compromised a userspace process. From there, the attacker could control the entire machine, all without altering a single bit of "software".
Imagine the fiasco that would ensue if a system were compromised by altering the CPU itself. This would be the hardware equivalent of Ken Thompson's legendary self-replicating compiler (http://www.acm.org/classics/sep95). A few years ago, Intel had to answer to public scrutiny over the exploitability of their own microcode update feature; their solution was security through obscurity and layers of encryption and authentication (see http://www.eetimes.com/news/97/963news/hole.html). Evidently AMD was not as wise by assuming their microcode was uncrackable.
There may also be a hidden danger to altering K8 microcode without complete information. It is possible (though very unlikely) that the microcode could electrically reconfigure signal routing in a fashion similar to FPGAs, for instance to cut off defective logic and reroute signals to redundant arrays. This approach has been used in the past and the AMD patents even suggest it.
If this were the case, there is a very remote chance the CPU itself could be permanently damaged, for instance, by tri-stating pass
transistors into a high current draw state or adjusting the K8's voltage and frequency scaling controls out of spec. This is not meant to discourage potential hackers.
Nonetheless, it is suspected that with sufficient analysis or maybe a bit of inside information, one could do some interesting things with microcode hacking.
At the very least, the information here should be useful for adding AMD support to the Linux microcode update driver, which already supports Intel's update mechanism.
Proof of Concept:
* AMD K8 (Athlon 64 / Opteron) Microcode Update Driver
* This code has been tested on a 64-bit Linux 2.6 kernel
* running on an Athlon 64. It has not been tested on
* other K8 cores. It may or may not work on 32-bit
* Linux kernels (but it should).
* This program is free software; it is licensed under
* the GNU General Public License, version 2.
* Version: 2004 Jul 20