Malicious code execution in PCI expansion ROM
The malicious code in x86/x64 firmware can potentially reside in many places. One of them is in the PCI expansion ROM. In the past, the small amount of memory during PCI expansion ROM execution acted as a hindrance to this malicious code. The limited space for codes and data limited the possible tasks that could be carried out by such malicious codes. However, this article explains how a malicious PCI expansion ROM might exploit a little-known BIOS memory management interface to break through the memory "barrier," thus creating a potentially more complex threat. The discussion in this article is limited to PCI expansion ROM conforming to PCI firmware revision 3.1 specification. This article will not delve into EFI expansion ROM despite the fact that the subjects are very closely related. EFI expansion ROM will be presented later in a different article.
What is PCI expansion ROM anyway?
Before I delve into how malicious PCI expansion ROM code might use the memory management routine in the BIOS, it is important for you to know the basics of PCI expansion ROM. The next section will provide basic background information on PCI expansion ROM.
PCI device initialization
PCI Expansion ROM is a code, executed during BIOS initialization, that initializes certain types of PCI/PCI Express devices. (From this point on PCI Express will be abbreviated as PCIe.) The term "device" in this context refers to a logical—as opposed to physical—PCI/PCIe device. PCI specification covers this logical device detailed definition. PCIe devices are backward compatible to PCI in the logical level. Therefore, from programmer's point of view, PCIe devices are seen just like PCI devices. Of course there are several fundamental differences, but they are "extensions" of the basic PCI specification. For the purpose of this article, you can regard PCI and PCIe as interchangeable, except when I specifically state the matter to be PCI-specific or PCIe-specific.
A single physical PCI chip—let's say an Ethernet controller chip—can contain more than one logical PCI device. Each logical PCI device has a corresponding set of PCI configuration registers in the PCI configuration space. In the x86/x64 architecture, the PCI configuration space is defined as: 256 (100h) bytes of PCI configuration registers (per-logical PCI device) accessed through two 32-bit ports at I/O port CF8h-CFBh and CFCh-CFFh respectively. Port CF8h-CFBh is the "index" port, used to point to specific 32-bit PCI configuration register within the configuration space. Port CFCh-CFFh is the "data" port, used to write/read to/from PCI configuration register within the PCI configuration space.
Figure 1 shows the general layout of the header portion of the PCI configuration registers in a PCI device. This particular configuration is called PCI configuration Type 0. It is used for a PCI device which is not a PCI bridge. A PCI bridge connects two different PCI buses. PCI devices— such as an Ethernet card or video card—are not a PCI bridge. Therefore, I'm only focusing on PCI configuration Type 0.
Figure 1. Base Address Registers (BARs) in the PCI Configuration Space
PCI device registers (outside of the PCI configuration register) and memory (RAM and/or ROM) can be located anywhere within the CPU memory address space (or CPU I/O address space) because PCI devices contain programmable registers, called a Base Address Register (BAR). Figure 1 shows the location of the BARs in the header portion of the PCI configuration registers. A BAR controls where the PCI device registers (outside of the PCI configuration register) and where memory is mapped in the CPU memory, or I/O address space. The reason for this programmability is to eliminate address conflicts between devices, which plagues the old ISA and EISA devices.
Figure 1 also shows a very specific form of BAR at offset 30h, the Expansion ROM Base Address Register (XROMBAR). This particular register maps the PCI expansion ROM in a PCI device into the CPU memory or I/O address space. It's important that you take a note of this particular register because it's related directly to PCI expansion ROM execution. I'll talk more about this register in later sections of this article.
Some PCI devices also map their registers—the non-PCI configuration registers—into the I/O address-space, but as of today, that approach is not the preferred approach (except for the crucial PCI configuration registers). One of the reasons the PCI configuration registers are mapped to I/O space is to maintain backward compatibility, the backward compatibility is especially important for legacy operating system such as Windows XP and other older proprietary operating systems.
Figure 2 shows a simplified mapping of PCI/PCIe device registers and memory into the CPU I/O address space and memory address space respectively. Remember, Figure 2 is a simplified version of the real-world device registers and memory mapping. In practice, you might find the mapping to be more complicated than that. Read the PCI/PCIe device datasheet, along with other related datasheets such as the CPU and chipset(s), to find out the exact mapping.
Figure 2. PCI/PCIe device registers and memory mapping to CPU/system address spaces
PCI device registers—non PCI configuration registers—and memory must occupy certain memory (or I/O address space) before they can be used. Therefore, the device must be initialized to known—exact—addresses within the CPU I/O or memory address space. The motherboard BIOS carries out this initialization task. Figure 2 shows a sample of the PCI/PCIe device register and memory mapping into the system I/O and memory address space after the initialization is completed.
Certain types of PCI devices require much more elaborate device-specific initialization. In these devices, mapping their device registers and memory to the CPU address space is only one of the initialization tasks that needs to be completed. Motherboard BIOS cannot carry out these device-specific initialization tasks because it doesn't have prior knowledge of all possible devices and their respective initialization procedures. Besides, doing the initialization in the motherboard BIOS is impossible due to the motherboard BIOS's size constraint. You would need as much space as an OS needs to do that, which is impractical. Therefore, the PCI expansion ROM emerges to fill the gap.
There are at least two types of PCI devices that require the additional device-specific initialization—and hence need PCI expansion ROM—i.e., devices used to boot an operating system and display devices. The first category includes storage controller chips, such as SCSI or RAID controllers, and network controllers, such as Ethernet controllers (for boot from LAN), while the second category includes video card chips.
PCI expansion ROM location
Now, let me turn to explaining where you can find the PCI expansion ROM. PCI expansion ROM is stored as a binary file in a flash ROM chip. There are two types of PCI expansion ROMs that can be defined by physical appearance depending on the location of the PCI device:
- The first one is PCI expansion ROM located in a PCI expansion card—in which case the binary file is placed in a flash ROM chip soldered or socketed into the PCI expansion card.
- The second one is PCI expansion ROM integrated—embedded—into the mainboard BIOS.
With the first case, the physical PCI chip to be initialized is soldered—or socketed —into the PCI expansion card, while in the second case, the PCI chip to be initialized is soldered into the motherboard. Figure 3 shows both of these arrangements.
Figure 3. PCI expansion ROM location illustration
Figure 3 shows an incomplete system that is missing motherboard RAM and many other motherboard components. That's because the emphasis is on showing you the possible locations of the PCI expansion ROM chip and the BIOS chip. The exact location of these chips depends on the particular motherboard and PCI/PCIe expansion card. Therefore, you should check the motherboard manual and the PCI/PCIe expansion card manual, or carry out physical inspection to make sure you have the correct location of these chips. If you're still unsure, you can look for flash ROM chip on the respective boards. Look for memory chips made by Winbond, SST, Atmel, or other flash ROM manufacturers, then cross reference the chip number on the web to make sure the position is correct. A BIOS chip usually has clear markings. However, PCI expansion ROM chips usually don't have marking, and in the case of PCIe expansion card, the information is sometimes soldered into the back of the card. Therefore, the latter case involves a bit of a guessing game.
PCI expansion ROM embedded in the motherboard BIOS vs. PCI Expansion ROM in PCI Expansion card
A PCI expansion ROM chip embedded in the motherboard BIOS and PCI expansion ROM residing in PCI expansion card differ logically from a programmer's perspective. The logical difference lies in the steps necessary for execution of a PCI expansion ROM and the role of the XROMBAR in the corresponding PCI device. The steps to execute PCI expansion ROM are:
- Motherboard BIOS copies the PCI Expansion ROM—from where the PCI expansion ROM is stored—to RAM in the expansion ROM area (within C_0000h – DFFFFh range).
- Motherboard BIOS jumps into offset 03h in the copied PCI expansion ROM to execute it.
- PCI expansion ROM "returns" the execution back to the motherboard BIOS.
The first step above is different for PCI expansion ROM stored in flash ROM on the PCI expansion card and PCI expansion ROM embedded in the motherboard BIOS. In the first case, the motherboard BIOS copies the PCI expansion ROM from the flash ROM in the PCI expansion card to RAM by using the address stored in the XROMBAR of the PCI chip in the corresponding PCI device as the PCI expansion ROM address.
The motherboard BIOS initializes all of the BARs and XROMBARs in all PCI devices prior to starting to execute any PCI expansion ROM(s). The initialization effectively sets the location of the PCI Expansion ROM in the CPU I/O or memory address space. Therefore, the motherboard BIOS knows the address of the PCI expansion ROM in the CPU I/O or memory address space before copying it to the RAM. In this particular hardware configuration, the mechanism to copy the PCI expansion ROM is the same across all BIOS, regardless of the BIOS code base—whether it's AMI, Award-Phoenix, Insyde, or others.
However, PCI expansion ROM embedded in the motherboard BIOS is treated differently. Each BIOS code base— either from AMI, Award-Phoenix, Insyde, or others—has its own vendor-specific module that locates the PCI expansion ROM in the motherboard BIOS and then copies the PCI expansion ROM to RAM before the PCI expansion ROM executes. In this case, the format of the PCI expansion ROM in the motherboard BIOS is also vendor-dependent, each BIOS vendor has their own specific approach. However, most vendors store the PCI expansion ROM in a compressed format inside the motherboard BIOS to save space.
PCI and PCIe expansion ROM structure
Various specifications, including the PCI specifications, dictated the basic structure of both PCI and PCIe expansion ROM until roughly 2005. In 2005, the disparate specifications were consolidated into the PCI Firmware Specification Revision 3.0. This consolidated specification is, of course, backward compatible to the previous PCI specification revisions. The most recent version of the PCI Firmware Specification is revision 3.1, published in December 2010.
Present-day PCIe devices mostly still use PCI expansion ROM instead of EFI expansion ROM. However, some PCIe devices use a mixed PCI expansion ROM and EFI expansion ROM. A very small fraction of PCIe devices use EFI expansion ROM only. You will see very shortly how this is accomplished.
Let me start with the very basic requirement for a PCI expansion ROM. PCI expansion ROM can contain one or more "image(s"). Each "image" is a self-contained code and data that contains all of the necessary routines and data required to initialize the corresponding PCI/PCI chip. The purpose of having multiple "images" in one PCI expansion ROM is to support several CPU architectures, which have different instruction encoding. Therefore, despite the fact that the PCI expansion ROM contains several images, only one of them will be executed at runtime because only the first image that match the CPU architecture at that point would be executed by the platform firmware—the BIOS in the case of x86/x64.
Figure 4. Structure of PCI expansion ROM with Multiple Images
Figure 4 shows the structure of PCI expansion ROM with multiple images. Each image in the PCI expansion ROM must start with a 512-bytes boundary and must contain PCI expansion ROM header. As you will see later, a special field named Last Image Indicator (offset 15h) in the PCI data structure of the PCI expansion ROM indicates whether the image is the last image in the image list or not.
Each image in the PCI expansion ROM starts with the PCI expansion ROM header. Table 1 shows the format of the PCI expansion ROM header. The offset in Table 1 is calculates from the start of the image in the PCI expansion ROM.
Table 1. Generic PCI Expansion ROM Header Format
PCI Expansion ROM Header Field
PCI Firmware Specification has a specific rule for the format of the processor architecture unique data for x86/x64 architecture. Table 2 shows the PCI expansion ROM header format.
Table 2. x86/x64-specific PCI Expansion ROM Header Format
X86/x64 PCI Expansion ROM Header Field
The offset in Table 2 is calculated from the start of the PCI expansion ROM image. I previously explained that multi-image PCI expansion ROM contains images for different CPU architectures. A specific field, named Code Type (offset 14h) in the PCI Data Structure indicates the type of CPU architecture supported by the specific PCI expansion ROM image.
The PCI specification dictates that the PCI data structure must reside within the first 64KB of the PCI expansion ROM image. Now, let me proceed to show you the format of the PCI Data Structure that I have referenced several times above. Table 3 shows the format of the PCI Data Structure.
Table 3. PCI Data Structure
Signature, the "PCIR" string
Device List Pointer
PCI Data Structure Length
PCI Data Structure Revision
Revision Level of the Vendor's ROM
Last Image Indicator
Maximum Run-time Image Length
Pointer to Configuration Utility Code Header
Pointer to DMTF CLP Entry Point
Table 3 shows the code type field at offset 14h and last image indicator at offset 15h. Table 4 shows the meaning of each of this entry.
Table 4. PCI Data Structure Explanation
Intel x86, PC-AT compatible
Open Firmware standard for PCI
Hewlett-Packard PA RISC
Extensible Firmware Interface (EFI)Reserved
Now you have gone through three quite complex data structures, and you're wondering how all of them are interrelated. Let me put together what we have learned so far. Figure 5 shows the relationship between the data structures in a PCI expansion ROM explained above.
Figure 5. Relationships between the various PCI expansion ROM structures
Figure 5 shows the components of a multi-image PCI expansion ROM in detail. As you can see, the arrow in the PCI data structure field—in the PCI expansion ROM header—points to the PCI data structure within the first 64KB of the PCI expansion ROM image.
Support for multi-image PCI expansion ROM—which has existed for a very long time—opens the possibility of supporting a legacy BIOS system and the newer UEFI-based system in one multi-image PCI expansion ROM. As you can see in Table 4, code type for EFI and "ordinary" x86/x64 systems are different. Therefore, it is possible to create a multi-image PCI expansion ROM with one of the images consisting of "ordinary" x86 PCI expansion ROM and another consisting of EFI expansion ROM. Setting the correct code type field in the respective PCI expansion ROM image is enough to "signal" the platform firmware (BIOS or UEFI) to execute the right image. Refer to Table 4 for details of the code type field.
Memory limitation in PCI expansion ROM execution during Power-On Self-Test (POST)
In the previous sections I have covered PCI expansion ROM executed after all of the PCI device registers are initialized. This initialization is part of what is known as the Power-On Self-Test (POST) in the x86/x64 architecture.
X86/x64 boots up in 16-bit real mode. The machine briefly switched to big-real mode (16-bit with access to 4GB memory address space) several times, but most of the time the machine stays in 16-bit real mode. PCI expansion ROMs are not executed in place, i.e. not executed directly in ROM. Instead they have to be copied to RAM and executed there. In the old days, this probably didn't create complications because the size of the PCI expansion ROMs were quite small. The space allocated to execute all of the PCI expansion ROM was only 128KB, from C0000h to DFFFFh. Now this 128KB space becomes a limitation as PCI devices become more complex and require much more elaborate initialization.
Back in 1997, Intel Corp. and Phoenix Technologies had already foreseen size limitation as a very significant potential problem. Both companies created a specification to alleviate the problem. It was named POST Memory Manager (PMM) specification. This specification—in its version 1.01 incarnation—is still in use today. This specification allows PCI expansion ROMs to overcome the 128KB memory limitation and reserve more memory for their use. You have to be aware that even the 128KB space has to be shared between all of the PCI expansion ROMs in the system. It's virtually impossible for single PCI expansion ROM to occupy the entire 128 KB space. This is the root of the problem.
Now let's see how PMM might be used by the bad guys to give their malicious PCI expansion ROM more room to breathe. By using PMM services, placing a full-fledged complex routine in the PCI expansion ROM is no longer a dream. Malicious coder can, for example, place a rather complete file system infector (driver if you wish) in the PCI expansion ROM, or perhaps a cryptic system management mode code. The possibility is quite significant, but I'll leave a more specific discussion on this matter to be divulged in later sections of this article. The next section explains details of the PMM services.
POST Memory Manager (PMM) comes to the rescue
The code in the PCI expansion ROM must check for the presence of the PMM services before it can use it. A structure, named PMM structure signals the presence of the PMM services. PMM structure is located in the system BIOS address space on a paragraph (16-byte) boundary between addresses E0000h and FFFF0h—or equivalently, between segment E000h and FFFFh. The presence of the PMM structure within that address range indicates PMM services are present and could be called. Table 5 shows the contents of the PMM structure.
Table 5. PMM Structure
A PCI expansion ROM follows this procedure to locate and access PMM Services (functions):
- Search for the four-byte "$PMM" string on paragraph (16-byte) boundaries starting at E0000h, and ending, if not found, at FFFF0h.
- Verify that the PMM Structure data is valid by performing a checksum. The checksum is calculated by doing a byte-wise sum of the entire PMM Structure and comparing this sum with zero. If the checksum is not zero, then the PMM Structure data is not valid and the EntryPoint field of the PMM structure should not be called.
- It is optional to inspect the StructureRevision field to determine the appropriate structure map. The StructureRevision field changes if previously reserved fields in the PMM Structure are redefined to be valid fields.
- Make calls (far calls) to the EntryPoint field in the PMM Structure to allocate and free memory as desired.
Calls to PMM services (functions) are defined as if called from the C programming language. The PMM interface uses the standard C (16-bit) large model calling convention. In particular, return values of the unsigned long type are returned in the DX:AX register pair. PCI expansion ROMs call the PMM Services by executing a far call to the address specified in the EntryPoint field in the PMM structure. PMM services return to the PCI expansion ROM routine via a far return. Values returned to the PCI expansion ROM routine are placed in the DX:AX register pair.
BIOS that provide support for PMM follows this rule: When control is passed to an option ROM—including PCI expansion ROM—from a BIOS that supports PMM, the processor will be in big real mode, and Gate A20 will be disabled (segment wrap turned off). This allows access to extended memory blocks (memory above the 1MB limit) using real mode addressing. In big real mode, you can gain access to memory above 1MB by using a 32-bit extended index register (EDI, ESI, etc.) and setting the segment register to 0000h. The code snippet in Listing 1 below shows an example of reading data from the extended memory block to AX register.
Listing 1. Accessing extended memory block sample code
; Assume here that dx:ax contains the 32-bit address of an already allocated buffer.
; Clear the ES segment register.
; Put the DX:AX 32-bit buffer address into EBX.
mov bx, dx ; Get the upper word.
shl ebx, 16 ; Shift it to upper EBX.
mov bx, ax ; Get the lower word.
; read the first two bytes of the extended memory buffer to ax.
mov ax, es:[ebx]; ES:EBX is used as the memory pointer.
Because the control is passed from the BIOS to the PCI expansion ROM in big real mode, code in the PCI expansion ROM should not mess with the state of the CPU. Code in the PCI expansion ROM must not modify the state of Gate A20, put the processor in protected mode, or call BIOS Interrupt 15h, functions 87h or 89h. Any of these actions would alter the big real mode of the processor and could lead to a catastrophic system failure. This requirement puts a constraint on the type of code that you can place in the PCI expansion ROM. If you wish to use a C compiler to make PCI expansion ROM code, you have to make sure the compiler supports emitting code in big real mode form (remember the old DOS32 days?) or just plain real mode code. This is probably amusing, but compilers used during the days of the first Doom game come to the rescue in this situation.
Now that you know the entry point to use PMM functions, the time has come to find out what functions are offered by the PMM entry point. But before that, I will show you the generic form of the function. Listing 2 shows the function prototype. The entryPoint in Listing 2 is the entry point defined in the EntryPoint field in the PMM structure shown in Table 5.
Listing 2. PMM functions entrypoint
unsigned long (*entryPoint)(
unsigned int functionNumber, // this is where you specify which function to call
… // other function parameters, depending on the function being called
You invoke the functions provided by PMM by using their function number, i.e. function number is set to 0 for function 0, etc. There are three functions offered by the PMM entry point:
- Function 0: pmmAllocate(). This function allocates memory either in the conventional memory (within the first 1MB) or in extended memory (above the first 1MB). There is an optional "handle" in the parameter to call this function. The "handle" is used to identify the memory block that's allocated. It's important only if you want to access that same memory block in your PCI expansion ROM long after the memory block has been allocated.
- Function 1: pmmFind(). This function searches for the memory block by using the "handle" that's used in a pmmAllocate() call previously, and returns a pointer to (the address of) the allocated memory block.
- Function 2: pmmDeallocate(). This function frees the memory block previously allocated in the call to pmmAllocate() function.
Now that the PMM functions are clear, I'm going to move to the details of those interface functions.
Function 0: pmmAllocate()
The pmmAllocate() function attempts to allocate a memory block of the specified type and size, and returns the address of the memory block to the caller. The address returned by the pmmAllocate() function is a "far" address, i.e. 32-bit physical address of the allocated memory block. The memory block is a contiguous array of paragraphs (16-bytes) whose size is specified by the length parameter. The contents of the allocated memory block are undefined and must be initialized by the PCI expansion ROM. Listing 3 shows the pmmAllocate() function prototype.
Listing 3. pmmAllocate() function prototype
unsigned long (*entryPoint)(
unsigned int function, // 0 for pmmAllocate
unsigned long length, // in paragraphs (multiple of 16-bytes)
unsigned long handle, // handle to assign to memory block
unsigned int flags // bit flags specifying options
The entryPoint identifier in Listing 3 is of type function pointer, and is a "far" address. It must be assigned the value of the EntryPoint field in the PMM structure before it can be called. The meaning of the parameters in the pmmAllocate() function as follows:
- function, 0 for pmmAllocate. Invalid values for the function parameter (0x0003…0xFFFF) cause an error value of 0xFFFFFFFF to be returned, signaling that the function is not supported.
- length, The size of the requested memory block in paragraphs, or if the length is 0x00000000, no memory is allocated and the value returned is the size of the largest memory block available for the memory type specified in the flags parameter (below). The alignment bit in the flags register is ignored when calculating the largest memory block available.
- handle, identifier to be associated with the allocated memory block, specified by the routine in the PCI expansion ROM. A handle of 0xFFFFFFFF indicates that no identifier should be associated with the block. Such a memory block is known as an "anonymous" memory block and cannot be found using the pmmFind function (see below). If a specified handle for a requested memory block is already used in a currently allocated memory block, the error value of 0x00000000 is returned.
- flags, a "bitmap" used by the PCI expansion ROM routine to designate options regarding memory allocation. Table 6 shows the possible values for the flag.
Table 6. pmmAllocate() flag
1 = Requesting conventional memory block (0 to 1MB).
2 = Requesting extended memory block (1MB to 4GB).
3 = Requesting either a conventional or an extended memory block, whichever is available.2Alignment0..10 = No alignment.
1 = Use alignment from the length parameter.3Attribute0..10 = Temporary memory use during POST
1 = Permanent memory use4..15Reserved0Reserved for future expansion, must all be zero.
The most interesting flag in the pmmAllocate() function is the MemoryType flag and Attribute. If you request an extended memory block in your call to pmmAllocate() and set the attribute to permanent, the memory allocated would behave like SMM memory which is not accessible to the OS and reside above the 1MB limit. This is probably a boon for malware creators, but for the defender, he or she wants to make sure he can scan those areas for malicious code.
Function 1: pmmFind()
The pmmFind() function returns the address of the memory block associated with the specified handle. Recall that the handle is specified as handle parameter in the call to the pmmAllocate() function previously. The entryPoint identifier is of type function pointer, and is a far address. It must be assigned the value of the EntryPoint field in the PMM structure before it can be called. The return value is of type unsigned long (32-bit), and represents the 32-bit physical memory address of the memory block that is associated with the handle. A return value of 0x00000000 indicates that the handle does not correspond to a currently allocated memory block.
Listing 4. pmmFind() function prototype
unsigned long (*entryPoint)(
unsigned int function, // 1 for pmmFind
unsigned long handle // handle assigned to memory block
The meaning of the parameters in the pmmFind() function are as follows:
- function, 1 for pmmFind(). Invalid values for the function parameter (0x0003…0xFFFF) cause an error value of 0xFFFFFFFF to be returned, signaling that the function is not supported.
- handle, An identifier specified by the PCI expansion ROM that was assigned to a memory block when the pmmAllocate() function was called. If called with an anonymous handle (0xFFFFFFFF) or a currently undefined handle, a value of 0x00000000 will be returned indicating that no associated memory block was found.
This function is not very important with respect to creating malicious PCI expansion ROM. However, it could be beneficial if the need to "tag" certain allocated PMM memory arises.
Function 2: pmmDeallocate()
This function is the counterpart of the pmmAllocate() function. The pmmDeallocate() function frees the specified memory block that was previously allocated by pmmAllocate(). Memory blocks are not cleared to all zeros by the PMM before being deallocated. Memory below 1MB is cleared by the BIOS before OS boot, but deallocated extended memory blocks in particular will not explicitly be cleared. Any code that calls the PMM should not rely on the contents of deallocated memory blocks.
Listing 5. pmmDeallocate() function prototype
unsigned long (*entryPoint)(
unsigned int function, // 2 for pmmDeallocate
unsigned long buffer // value returned by pmmAllocate
The meaning of the parameters in the pmmDeallocate() function are as follows:
- function, 2 for pmmDeallocate. Invalid values for the function parameter (0x0003…0xFFFF) cause an error value of 0xFFFFFFFF to be returned, signaling that the function is not supported.
- buffer, The 32-bit physical address of the memory block. This is the value that was returned by the call to the pmmAllocate() function previously.
This function is important if you want to deallocate memory that's previously allocated with pmmAllocate() call. However, if you allocated a permanent memory in your call to pmmAllocate()—see the attribute flag in the pmmAllocate() explanation—you probably don't need this function.
At this point all of the functions needed to allocate and work with a larger memory pool are clear. The 128KB limitation no longer exists. However, you might want to obfuscate the presence of your code in the PCI expansion ROM by going several steps further. The next section explains what we could do to accomplish that goal.
The role of executable compressors
You may have your arsenal ready to fight the memory limitations in your PCI expansion ROM code, but it's not enough. Most PCI expansion card flash ROMs are very limited in terms of capacity; the largest at his point is probably 128KB. The same is true with BIOS flash ROM chip, the size of BIOS flash ROM chip maxed out at 1MB (there might be some exotic boards out there with 2MB). The way to overcome this limitation is to compress your PCI expansion ROM code (and data), so that you can get the most out of those limited flash ROM chips.
A quick survey of available executable file compressors shows that UPX is probably the best for PCI expansion ROM code compression because it supports compressing 16-bit x86 code. The entry point for the decompressor routine in a UPX-compressed file resides at the beginning of the compressed binary. Therefore, it's easy for you to integrate UPX-compressed binary into your code. The compression ratio is also quite good. However, because UPX focuses on executable code compression, its data compression capability is not as good as data-focused compressors out there. Nonetheless, 16-bit decompressor is quite hard to come by these days.
Now let me show you the possible layout of a UPX-compressed PCI expansion ROM. Figure 6 shows the layout of PCI expansion ROM containing a UPX-compressed module.
Figure 6. PCI Expansion ROM containing UPX-compressed module
Figure 6 doesn't show that you should minimize the distance between the PCI ROM header and PCI data structure. However, you should be sure to minimize that distance because you want to maximize the space for the UPX-compressed module. The "prologue" section contains the routine to pass execution into the UPX-compressed module. Theoretically you could innovate at this point and do whatever you like, just make sure you don't create a code that overwrites important part of the PCI Expansion ROM, and make sure the final execution of the PCI expansion ROM returns to the motherboard BIOS with a "far" return.
C code in PCI expansion ROMs
It is now clear how increase PCI expansion ROM to meet your needs. However, you're not really comfortable using assembly language, and you want to use C as your language to develop PCI expansion ROM. What should you do? Unfortunately, it's (almost) impossible to use only C language—you still need some assembly language here and there. However, it's possible to use only C language for the UPX- compressed module that you see in Figure 6.
Now, let's explore what options that you have if you want to use a C compiler. I've been toying with using GCC to develop an incomplete "bootloader" in a PCI expansion ROM for a long time. You can see the complete explanation at: https://sites.google.com/site/pinczakko/low-cost-embedded-x86-teaching-tool-2. However, the code in that link switches the machine to 32-bit protected mode because it was designed as a sort of "precursor" to a small operating system. Moreover, it resides in an Initial Program Loader (IPL) device (a storage controller card), i.e. a device that the BIOS regards as having the capability to boot into an OS akin to a hard drive. Therefore, it could hook into interrupt 19h to boot the system, and switching to 32-bit protected mode is just fine in that scenario. If you only want to use the "large" memory space available from the PMM functions, you should take another approach.
It's close to impossible to use the mainline GCC that runs in Linux or Unix to develop your C compiled PCI expansion ROM module. There are other better routes though. In this case, you could opt to use OpenWatcom (www.openwatcom.org) or DJGPP (http://www.delorie.com/djgpp/ ). I will leave the rest of the task to another article, but you could also try it out yourself. There are probably other compilers out there that can emit 16-bit real mode or 16-bit big real mode code. It's wise to find the one that best suits your needs.
Potential malicious payload in PCI expansion ROM
After all the explanation above you might ask: What are the benefits of a larger memory footprint for a malware creator? Let's see:
- The larger memory footprint enables a malware creator to place (at least) a simple file system infector inside the PCI expansion ROM (a compressed one). During PCI expansion ROM execution, the compressed file system infector could have the memory it requires through memory allocation with the PMM functions, provided that the BIOS implemented PMM—which is most likely the case in the last 3 to 5 years.
- Another issue is that a malware creator might abuse the presence of the "permanent" memory allocated for PCI expansion ROM through the pmmAllocate() function by using the permanent memory flag during the call to pmmAllocate().
- Additionally, a rogue but simple network "interceptor" code might be possible given the jump in the memory footprint, and if the interceptor hides in the "permanent" memory, it could be troublesome.
This article has provided the building blocks required to carry out the task of creating a malicious PCI expansion ROM that is not limited by the scarce memory footprint during POST. This is, of course, a double-edged sword, but it's important to create awareness of the possibility in the computer security community,. Otherwise, most of us would be left blind in the wake of such an advanced type of attack.
Malicious PCI expansion ROM detection
I have yet to see a real-world malicious PCI expansion ROM. However, it should be noted that this kind of malware is not directly accessible to processes running in the host operating system. How would you detect such malware? Unfortunately, this task is the same as scanning most of the physical memory address space to make sure there's no malicious PCI expansion ROM in the "permanent" memory section of the RAM (memory allocated by pmmAllocate() with the permanent flag set to 1). One of the heuristics that comes to mind is whether the routine in the PCI expansion ROM contains function(s) that accesses the operating system. Well, you could think of other heuristics as well.
Now, let me talk a bit about the very near future. Windows 8, in one or more of its incarnations, will only support signed PCI expansion ROM or EFI expansion ROM. I have yet to see how this will play out in the real implementation, but that is a step forward. However, you should be aware that it would take years to migrate to Windows 8 and the installed base with present day PCI expansion ROM is far larger. I'll leave this matter for another article.
To summarize: breaking through the very small memory footprint during POST is possible through the use of the PMM functions.