Home

Awesome

Analog Devices Blackfin Architecture Plugin

A Binary Ninja architecture plugin for the AD Blackfin architecture. It includes a standalone disassembler, and fairly robust lifting capabilities.

Disassembler

The disassembler used in this plugin (relagated to the disassembler/ subdirectory) can be used as a standalone decomposer/disassembler, and is not tied to the binaryninja API. It outputs tokenized structures representing each parsed machine code instruction, which can be parsed as desired by the disassembler frontend.

Disassembly
Yes, this is disassembly, not LLIL -- Blackfin uses arithmetic syntax for assembly

Instruction Support

Disassembly of all instructions defined in the Blackfin Processor Programming Reference (Rev 2.2) is implemented. Additionally, a set of pseudo_ instructions are implemented based on the libopcodes blackfin instruction parser, though these instructions do not appear in the above reference manual.

This disassembler supports only the Blackfin ISA, and not the Blackfin+ superset used on BF7xx processor models. In addition to several new instructions, the Blackfin+ ISA also features a number of additional registers, and support for and usage of 64 bit instructions. At a minimum, support for detecting 64 bit instructions is planned, which would at least allow for disassembly of Blackfin compatible instructions in a Blackfin+ image.

Lifter

HLIL
HLIL output example

The lifter is mostly complete, with all standard operations supported, but lacking in support of some of the more complex/unusual DSP instructions and operating modes, including operations which saturate rather than overflow, and things like add with reverse bit carry that are complicated to implement correctly in BNIL. Additionally, some vector operations have yet to be implemented.

Architecture Misc.

Building (Adapted from Vector35/arch-armv7)

Building the architecture plugin requires cmake 3.9 or above. You will also need the Binary Ninja API source.

First, set the BN_API_PATH environment variable to the path containing the Binary Ninja API source tree.

Run cmake. This can be done either from a separate build directory or from the source directory. If your app is installed in a non-default location, set BN_INSTALL_DIR in your cmake invocation, like cmake -DBN_INSTALL_DIR=/Applications/Binary\ Ninja\ DEV.app/'. Once that is complete, run make in the build directory to compile the plugin.

The plugin can be found in the root of the build directory as libarch_blackfin.so, libarch_blackfin.dylib or arch_blackfin.dll depending on your platform.

You can copy the plugin into the user plugins directory (you can locate this by using the "Open Plugin Folder" option in the Binary Ninja UI).

Build Example

acquire repositories

mkdir ~/repos/vector35
cd ~/repos/vector35
git clone git@github.com:Vector35/binaryninja-api.git
git clone git@github.com:git@github.com:Ethan-ks/arch-blackfin.git

environment variables

export BN_API_PATH=~/repos/vector35/binaryninja-api

cmake, make

cd arch-blackfin
mkdir build
cd build
cmake ..
make

Build Troubleshooting

example

CMake Error at CMakeLists.txt:8 (message):
  Provide path to Binary Ninja API source in BN_API_PATH
resolution:
ensure BN_API_PATH is in your environment

example

CMake Error at /Users/andrewl/repos/vector35/binaryninja-api/CMakeLists.txt:53 (message):
  Binary Ninja Core Not Found
resolution:
ensure BN_INSTALL_DIR is supplied at command line invocation of cmake
ensure some bad directory is not cached in CMakeCache.txt

example

cmake seems to ignore your setting of BN_INSTALL_DIR and other cmake variables
resolution:
rm CMakeCache.txt

example

undefined symbols at link time, like:
Undefined symbols for architecture x86_64:
  "_BNClearUserVariableValue", referenced from:
  BinaryNinja::Function::ClearUserVariableValue(BinaryNinja::Variable const&, unsigned long long) in libbinaryninjaapi.a(function.cpp.o)
resolution:
ensure that your api repo is on the same channel and at the same commit as the libbinaryninjacore you're linking against
eg: binaryninja is on dev update channel and is up-to-date and binaryninja-api repo is on branch dev with latest pulled

Known Limitations

Incomplete Lifting

Out-of-spec handling of 16 bit immediate loads

A common pattern observed in a number of sample binaries was the usage of a reg.h = imm16; reg.l = imm16 idiom for loading a 32 bit immediate into a full width register. In most cases, the high load immediately precedes the low load, and as such the lifter combines the two operations into a single 32 bit load, which improves the resulting IL. However, in some cases the two operations are split up, and the resulting IL becomes messy, and breaks Binary Ninja's Value Set Analysis (VSA).

In order to combat this, the lifter currently interprets a lone load high instruction as a 32 bit load to the full width register, assuming that the low bytes will be overwritten by a load low at some point before the register is read. This has been true in all cases observed during the development of this plugin, but does not follow the Blackfin spec and technically produces incorrect IL. So far, the risk of occasional incorrect IL is outweighed by the greatly improved IL in the majority of cases, but be warned.

No support for parallel execution

Blackfin ISA supports the parallel execution of up to three instructions at once. This is not implemented in the disassembler or lifter at present, and input machine code is treated as entirely sequential. As a consequence, under very specific circumstances the resulting disassembly may be misinterpreted -- for example, when one instruction in the parallel issued set sets a register, and the next in the set uses that same register as a source. In reality, because the instructions are being executed in parallel, the value of the register used in the latter operation will not have been updated at execution time, but the disassembly output will indicate that this is the case.