marcoguerri / broadcom-optionrom

Tool to overwrite OptionROM on Broadcom BCM575x NICs

broadcom-optionrom is a tool built based on the reverse engineering of Boardcom MS-DOS utility B57UDIAG.EXE, specifically the control paths which manipulate OptionROM in NVRAM. It can be used to write custom UEFI OptionROM to NVRAM of Broadcom BCM5751 NICs. It will likely work also for other devices belonging the BCM575x family, but so far it has been tested only on a BCM5751-RJ45 PCIe x1 1G Single-port Desktop Adapter.

Broadcom tooling

B57UDIAG.EXE is an old tool which can still be sourced from Broadcom website. Most of my work was based on top of version 15.6.1, 10/16/12, available as doc 12358473. The sha256 of the EXE file itself is 3be18d9bad27f20209a729e42fee8d28e03aacf2bbe571deea81b83d98c36686. Note that it did happen once that I had to restore NIC boot code through the tool running in engineering mode (-b57eng flag). The only version I could find that was capable of this type of recovery was 7.34, 6/17/04 (sha256sum 0eca4bd9152a7578b7fa867b0459da88a1d0726f74fe3d08f021ce5826ef09ef). I could not find a copy of this binary distributed directly by Broadom. I backed up the IDA databse file with all my custom annotations in the repository. I find reverse engineering annotations something very personal, so this database won't make much sense to anybody else beyond myself.

Usage

broadcom-optionrom supports two commands:

• check: performs NVRAM checks for UEFI OptionROM (e.g. directory checksum, OptionROM checksum, VPD checksum)
• write: writes custom OptionROM on a binary dump of NIC NVRAM

Example

First, the current content of NIC NRAM must be dumped via ethtool:

# ethtool -e <INTERFACE> raw on > NVRAM.bin

We can verify that all checks implemented by the tool do succeed:

$ go run main.go check -inNvramBinPath NVRAM.bin    
OK

write command can then be used to overwrite UEFI OptionROM:

$ go run main.go write -uefiOptionRomPath optionrom.v2.efi -inNvramBinPath NVRAM.bin -outNvramBinPath NVRAM.mod.bin
OK

We can check that the newly generated NVRAM image passes integrity checks:

$ go run main.go check -inNvramBinPath NVRAM.mod.bin
OK

The new NVRAM binary file can be written back to the device:

# ethtool -E <INTERFACE> magic 0x669955aa offset 0 length <FILE_LEN> < NVRAM.mod.bin

Testing on physical device

The system I used for testing consists of a PCEngines APU2D4 board. This is normally used to implement routers, firewalls, etc, but it’s a great setup to run firmware experiments, thanks to its support for Open System Firmware (coreboot), potentially with UEFI payload, and easy access to SPI flash with external programmers.

For OptionROM experiments, I built coreboot with UEFI pyload support (edk2-tianocore) through UefiPayloadPkg/UefiPayloadPkg.dsc. Support for enumerating PCIe bus and dispatching execution to UEFI OptionROMs was patched on top of upstream. When I run these experiments, Patrick Rudolph's patch had not been approved yet, but I see it's now merged, which is great news. Bus enumeration would normally be coreboot responsibility, but nothing prevents UEFI from scanning the bus again, identifying UEFI OptionROMs and executing them.

An example of full end to end test would be the following: Current OptionROM is executing correctly:

InstallProtocolInterface: 4006C0C1-FCB3-403E-996D-4A6C8724E06D CE8414F0
[Security] 3rd party image[0] can be loaded after EndOfDxe: PciRoot(0x0)/Pci(0x2,0x1)/Pci(0x0,0x0)/Offset(0x0,0x1FFF).
InstallProtocolInterface: 5B1B31A1-9562-11D2-8E3F-00A0C969723B CE842C40
Loading driver at 0x000CE7D3000 EntryPoint=0x000CE7D3F92 OptionROM.efi
InstallProtocolInterface: BC62157E-3E33-4FEC-9920-2D3B36D750DF CE841C98
ProtectUefiImageCommon - 0xCE842C40
  - 0x00000000CE7D3000 - 0x0000000000001E40
*** Nel mezzo del cammin di nostra vita
*** mi ritrovai per una selva oscura
*** che la diritta via era smarrita

We build a new OptionROM. From:

    DEBUG((DEBUG_INFO, "*** Nel mezzo del cammin di nostra vita\n"));
    DEBUG((DEBUG_INFO, "*** mi ritrovai per una selva oscura\n"));
    DEBUG((DEBUG_INFO, "*** che la diritta via era smarrita\n"));
    return EFI_SUCCESS;

to:

    DEBUG((DEBUG_INFO, "*** This is a new OptionROM binary \n"));
    return EFI_SUCCESS;

Through EDK tooling, we assemble the new UEFI OptionROM binary, with the device specific VendorID and DeviceID:

$ ./BaseTools/Source/C/bin/EfiRom -f 0x1677 -i 0x14e4 -e Build/OvmfX64/DEBUG_GCC5/X64/OptionROM/OptionROM/DEBUG/OptionROM.efi -o optionrom.efi

We dump the current NVRAM:

# ethtool -e enp1s0 raw on > NVRAM.bin

and overwrite OptionROM:

$ go run main.go write -uefiOptionRomPath optionrom.efi -inNvramBinPath NVRAM.bin -outNvramBinPath NVRAM.mod.bin
$ OK
$ go run main.go check -inNvramBinPath NVRAM.mod.bin
OK

We then write the NVRAM back to the device:

# ethtool -E enp1s0 magic 0x669955aa offset 0 length 131072 < NVRAM.mod.bin

and reboot. We see the OptionROM changes reflected in its execution at UEFI boot time:

InstallProtocolInterface: 4006C0C1-FCB3-403E-996D-4A6C8724E06D CE8414F0
[Security] 3rd party image[0] can be loaded after EndOfDxe: PciRoot(0x0)/Pci(0x2,0x1)/Pci(0x0,0x0)/Offset(0x0,0x1DFF).
InstallProtocolInterface: 5B1B31A1-9562-11D2-8E3F-00A0C969723B CE842C40
Loading driver at 0x000CE7D5000 EntryPoint=0x000CE7D5F92 OptionROM.efi
InstallProtocolInterface: BC62157E-3E33-4FEC-9920-2D3B36D750DF CE841C98
ProtectUefiImageCommon - 0xCE842C40
  - 0x00000000CE7D5000 - 0x0000000000001DC0
*** This is a new OptionROM binary 

Limitations

The tool has the following limitations, some of which are further analyzed in the corresponding post:

• It doesn't implement any type of NVRAM space look-up. It just searches for PXE entry in the directory and stores the new OptionROM starting from the same offset. It really doesn't care if in the process, anything else might be overwritten. This is obvioulsy super naive, and implementing a proper space look-up algorithm as described in the post, would not be difficult. I have gotten around to implementing this yet.
• It doesn't rewrite VendorID nor DeviceID in the OptionROM header based on the device we are working on. These two fields must be populated correctly at OptionROM build time.