Google Summer Of Code 2021- Support for CPU and Platform Profiles



Hi all,

This blog post is a brief summary about the work I did on the summer of 2021 with Rizin on adding support for CPU and platform profiles.


Title of the project

Support for CPU and platform profiles


Rizin previously relied upon manually writing code for adding a new CPU or an IO port. This implementation was unfit as the vast and growing ecology of hardware components such as CPUs and SoCs regularly implement a part of architecture with custom instructions, registers, and address configurations with trivial differences, making it infeasible to maintain all of them inside Rizin.

Providing a level of abstraction in handling this entropy in embedded systems by adding support for editable CPU and platform profile was the goal of this project. This also made adding and maintaining these ports easier with less interaction with Rizin’s core. This project also added more flexibility in having a way of importing existing hardware data description documents so that reverse engineering on particular chipsets is easier. Now, it’s also easier to memory map the peripheral accesses and registers to provide a better reverse engineering experience. Currently, this is being extended to benefit Rizin in terms of compatibility and the end users in terms of user experience.

Getting ready

The microtask I had to do before the actual project began was to make an SVD parser plugin.

SVD files are files containing information about a device’s peripherals, MMIO registers and other particulars. They are usually made by the manufacturer. This plugin would load the data from SVD file to Rizin and add them as flags (labels) and comments, simplifying the reverse engineering experience.

Here’s how the SVD file of the STM32F030 microcontroller begins:

<?xml version="1.0" encoding="utf-8" standalone="no"?>
<device schemaVersion="1.1"
  <!--Bus Interface Properties-->
  <!--Cortex-M0 is byte addressable-->

This is a very big document with lots of information about peripherals, registers and much more:

  <description>Data register</description>

The plugin parses all the registers' name, size, base address and its offset to add a flag or a label and parses its description to add a comment at its offset.

That was probably the first moderately large piece of software I had ever written from scratch. It was a bit challenging at the beginning but eventually I got the hang of and got it into a state where I could parse some of the register’s description and offsets and add it as flags right before receiving the e-mail saying that I got accepted:

Summer time

After getting accepted, the first thing I did was to remove the existing implementation of RzSyscallPorts - the module which took care of the architecture and CPU specefic system registers.

ioports.c housed RzSyscallPorts, which were hardcoded definitions of system registers which were iteratively searched and served using rz_syscall_get_io() and other APIs. These hardcoded values are moved to SDB in this Pull Request.

A port using RzSyscallPorts looked like this:

RzSyscallPort sysport_avr[] = {
    { 0x3e, "SPH: Stack higher bits SP8-SP10" },
    { 0x3d, "SPL: Stack lower bits SP0-SP7" },

This Pull Request moved all of those existing definitions to SDB, which follows a format like this:

SPH.comment=Stack higher bits SP8 SP10

Here, I made two new modules: RzSysregsDB and RzSysregItem to make this happen. RzSysregsDB just housed a hashtable which paired the address of the port and an RzSysregItem which contained the comment, type and all the other information related it. This Pull Request also introduced rz_sysreg_get() to which you can pass on the type (mmio/reg) and the offset to get the corresnponding port. This is used where these ports are displayed as comments from disasm.c.

After that, I jumped straight into the first phase and the crux of the project: CPU profiles.

The CPU profiles, basically the files containing information about the CPUs were stored in SDB files at librz/asm/cpus following a naming convention arch-cpu are loaded up atrz_arch_profiles_init() at the beginning. Then, it’s parsed and stored into various data structures inside RzArchProfile, where RzArchTarget houses the name of the CPU and architecture and a pointer to RzArchProfile (RzArchTarget is currently under RzAnalysis). Information about the IO registers and Extended IO registers were put inside a hashtable and the other data in normal ut64 and character array variables for easy and fast access.

typedef struct rz_arch_profile_t {
  ut64 rom_size;
  ut64 ram_size;
  HtUP /* <ut64 , char *> */ *registers_mmio;
  HtUP /* <ut64 , char *> */ *registers_extended;
} RzArchProfile;

typedef struct rz_arch_target_t {
  char *cpu;
  char *arch;
  RzArchProfile *profile;
} RzArchTarget;

The information is integrated during the analysis loop. During analysis (aa), rz_arch_profile_add_flag_every_io() is called, which parses the two hashtables and adds the information as flags. For that, two new flagspaces were realized: registers.mmio and registers.extended for the corresponding ones.

You can see the CPU specific registers being added as flags here. This is when loaded up with an AVR firmware and CPU set to ATmega168. You can see the registers from the CPU profiles added as flags (labels) right near the offsets: cpu-profiles-in-action

Since we already have size of the ROM and its starting address on the CPU profile, it makes good sense to add its range as a section (iS command). To implement that, I introduced rz_analysis_add_io_registers_map() which takes care of the job when during the analysis loop.

$ rizin test/bins/firmware/arduino_avr.bin
[0x00000158]> aaa
[x] Analyze all flags starting with sym. and entry0 (aa)
[x] Analyze function calls (aac)
[0x00000158]> iS~.rom
 8   0x00001fff  0x4000 0x00001fff  0x4000 -r-x .rom      NULL  

It was time to set my sails jolly for the next phase.

In this patch series, I introduced RzArchPlatformTarget and RzArchPlatformItem for platform profiles.

The platform profiles stored as SDB files in librz/asm/platforms following a file naming convention similar to a triple target arch-cpu-platform, are firstly loaded up at rz_arch_platform_init(). Subsequently, it’s parsed and stored in a hashtable inside RzArchPlatformTarget which is a pair of ut64 address and a RzArchPlatformItem. RzArchPlatformItem is a struct/module which houses the name and the comment (if it exists) of the corresponding port. The names are added as flags and comments as comments (CCu command) in rz_arch_platform_add_flags_comments().

typedef struct rz_platform_item_t {
  char *name;
  char *comment;
} RzArchPlatformItem;

typedef struct rz_platform_target_t {
  HtUP /* <ut64 , RzArchPlatformItem> */ *platforms;
} RzArchPlatformTarget;

Platform Profiles also follow a format similar to the CPU profiles that you saw earlier. Here’s an excerpt BCM 2835’s platform profile:

AUX_MU_IER_REG.comment=Mini UART Interrupt Enable

AUX_MU_IIR_REG.comment=Mini UART Interrupt Identify

A new configuration variable asm.platform was also added to choose the platform profile. This will let the user choose the name of the profile they want to load and Rizin will load the profile based upon the CPU and the architecture that the user have previously set. For that, I added a new variable platforms to RzAsmPlugin which will hold the list of all supported platforms of that architecture.

Italian Trulli

After that, I spend a lot of time writing unit and integration tests for both CPU and platform profiles. A lot of time went to debugging and fixing bugs and other memory leaks. Thanks to Coverity Scan, wargio and other Continuous Integration tests, it was very easy to spot and fix them with their help!

In the previous Pull Request, I had only added support for one platform profile - the one for BCM2835, which one of the Raspberry Pi runs on.

And over here, more profiles: BCM2711 and OMAP 3430 were added along with the x86 IO ports. Since the code for the module was already in, it was just as simple as getting the data in the right format, putting it on the corresponding directory and adding it to its :)


Folks at had made wonderful a tool named uefi_r2 which can be used to analyze UEFI modules. I ported that to Rizin.

This tool works by analyzing the firmware using Rizin’s RzAnalysis utilities and then by checking out with the analyzed functions, strings and all (For example, while searching for the UEFI guilds). Here, the tool is a Python package and all the interaction with rizin is done through rz-pipe’s Python module.

You can get plenty of information with this. For example, all the protocol GUIDs, list of boot services and its addresses can be obtained just by passing the path to the UEFI image to get_summary().

>>> from rzuefi.uefi_analyzer import UefiAnalyzer

>>> summary = UefiAnalyzer.get_summary(image_path)

>>> summary['bs_list']
[{'name': 'LocateProtocol', 'address': 142068}, {'name': 'LocateProtocol', 'address': 142598}, {'name': 'LocateProtocol', 'address': 134087}, {'name': 'ReinstallProtocolInterface', 'address': 164582},

Finally, I added some tests: couple of simple pytests to make sure that this is working, a small CI and I put it over at my GitHub profile.

Overall, this was not particularly challenging but it was indeed very informative. UEFI is insanely complex!

Alright, that’s done. It was time to get back on improving the SVD plugin again.

As I said, SVD files are basically XML files. Still, it wasn’t so easy to parse them, at least with yxml due to its small size and compact design. We decided not to use any other XML parses since yxml is already used inside Rizin. Thanks to xvilka and his comments, after a couple of days of painful debugging, I was able to parse the SVD files from STM and add the necessary information as flags and comments!

Here’s a quick demo!

Actually, the plan to end the second phase was to create a plugin inside rz-lang- to create a sort of wrapper above the C structs with PyObject and its friends so that you can interact with it from Python. Since, a solid module for architecture (RzArch) hasn’t been implemented yet, we switched plans and worked on improving the SVD plugin and other things. Still, I hope I can create the plugin when it’s time. That’s just one of the things I’m staying for at Rizin :)


I would like to thank my mentors xvilka and wargio for their guidance. I was regularly in touch with them and they were constantly trying make sure that everything was going smooth.

I have gained amazing insights and feel very grateful to have worked and learned from one of the smartest, kindest and knowledgeable people I’ve ever known. Huge respect!

Also kudos to all the folks at #Rizin-dev, #gsoc-2021 and the other channels where my queries were cleared.

I’m forever indebted to the community for this amazing experience.

Best regards,