Skip to content
Open
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
173 changes: 95 additions & 78 deletions tools/rimage/README.md
Original file line number Diff line number Diff line change
@@ -1,15 +1,10 @@
# rimage

`rimage` is a DSP firmware image creation and signing tool targeting
the DSP on certain Intel System-on-Chip (SoC). This is used by
the [Sound Open Firmware (SOF)](https://github.com/thesofproject/sof)
to generate binary image files.
`rimage` is a DSP firmware image creation and signing tool targeting the DSP on certain Intel System-on-Chip (SoC). This is used by the [Sound Open Firmware (SOF)](https://github.com/thesofproject/sof) to generate binary image files.

## Building

Most SOF users never build `rimage` directly but as an ExternalProject
defined by CMake in SOF. This makes sure they always use an up-to-date
version of rimage and configuration files that have been fully tested.
Most SOF users never build `rimage` directly but as an ExternalProject defined by CMake in SOF. This makes sure they always use an up-to-date version of rimage and configuration files that have been fully tested.

If needed, `rimage` can be built manually with the usual CMake commands:

Expand All @@ -19,87 +14,109 @@ $ make -C build/ help # lists all targets
$ make -C build/
```

The `build/rimage` executable can then be copied to a directory in the
PATH. Zephyr users can run `west config rimage.path
/path/to/rimage/build/rimage`; Zephyr documentation and `west sign -h`
have more details.
The `build/rimage` executable can then be copied to a directory in the PATH. Zephyr users can run `west config rimage.path /path/to/rimage/build/rimage`; Zephyr documentation and `west sign -h` have more details.

## Testing tomlc99 changes with SOF Continuous Integration

This section is about leveraging SOF validation to test tomlc99 changes
_before_ submitting them to the tomlc99 repository.
This section is about leveraging SOF validation to test tomlc99 changes _before_ submitting them to the tomlc99 repository.

Nothing here is actually specific to SOF and tomlc99; you can apply the
same test logic to any submodule and parent on Github. In fact the same
logic applies to submodule alternatives. Github is the only requirement.
Nothing here is actually specific to SOF and tomlc99; you can apply the same test logic to any submodule and parent on Github. In fact the same logic applies to submodule alternatives. Github is the only requirement.

Comment on lines +23 to 24
Copy link

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Use consistent capitalization for the GitHub brand name (currently "Github" appears multiple times in this paragraph).

Copilot uses AI. Check for mistakes.
Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed, thanks!

### Get familiar with git submodules

This is unfortunately not optional for SOF and tomlc99.

For various reasons submodules seem to confuse many git users. Maybe
because the versions of the submodules are not directly visible in some
configuration file like with most alternatives? Either way, an
unfortunate prerequisite before doing any tomlc99 work is to get familiar
with git submodules in general. As submodules are built-in there are
many resources about them on the Internet. One possible starting point
is https://git-scm.com/book/en/v2/Git-Tools-Submodules but feel free
to use any other good tutorial instead. Make sure you actually practice
a tutorial; don't just read it. Practicing on a temporary and throw-away
copy of SOF + tomlc99 is a great idea.
For various reasons submodules seem to confuse many git users. Maybe because the versions of the submodules are not directly visible in some configuration file like with most alternatives? Either way, an unfortunate prerequisite before doing any tomlc99 work is to get familiar with git submodules in general. As submodules are built-in there are many resources about them on the Internet. One possible starting point is https://git-scm.com/book/en/v2/Git-Tools-Submodules but feel free to use any other good tutorial instead. Make sure you actually practice a tutorial; don't just read it. Practicing on a temporary and throw-away copy of SOF + tomlc99 is a great idea.

Obviously, you also need to be familiar with regular Github pull
requests.
Obviously, you also need to be familiar with regular Github pull requests.

### Run SOF tests on unmerged tomlc99 commits

First, push the tomlc99 commits you want to be tested to any branch of
your tomlc99 fork on Github. Do _not_ submit an tomlc99 pull request yet.

Note your tomlc99 fork must have been created using the actual "fork"
button on Github so Github is aware of the connection with the upstream
tomlc99 repo. In the top-left corner you should see `forked from
thesofproject/tomlc99` under the name of your fork. If not then search
the Internet for "re-attach detached github fork".

Then, **pretend** these tomlc99 commits have already been accepted and
merged (they have been neither) and submit to SOF a draft pull request
that updates the main SOF branch with your brand new tomlc99 commits to
test. The only SOF commit in this SOF TEST pull request is an SOF commit
that updates the tomlc99 pointer to the SHA of your last tomlc99
commit. If you're not sure how to do this then you must go back to the
previous section and practice submodules more.

Submit this SOF pull request as a Github _draft_ so reviewers are _not_
notified. Starting every pull request as a draft is always a good idea
but in this case this particular SOF pull request can be especially
confusing because it points at commits in a different repo and commits
that are not merged yet. So you _really_ don't want to bother busy
reviewers (here's a secret: some of the reviewers don't like submodules
either). You can freely switch back and forth between draft and ready
status and should indeed switch to draft if you forgot at submission
time but you can never "un-notify" reviewers.

Github has very good support for submodules and will display your SOF
TEST pull request better than what the git command line can show. For
instance Github will list your tomlc99 changes directly in the SOF Pull
Request. So if something looks unexpected on Github then it means you
did something wrong. Stop immediately (except for switching to draft if
you forgot) and ask the closest git guru for help.

Search for "Submodule" in the build logs and make sure the last of your
new tomlc99 commits has been checked out.

Iterate and force-push your tomlc99 branch and your SOF TEST pull request
until all the SOF tests pass. Then you can submit your tomlc99 pull
request as usual. In the comments section of the tomlc99 pull request,
point at your test results on the SOF side to impress the tomlc99
reviewers and get your tomlc99 changes merged faster.

Finally, after your tomlc99 changes have been merged, you can if you want
submit one final SOF pull request that points to the final tomlc99
SHA. Or, if your tomlc99 change is not urgently needed, you can just wait
for someone else to do it later. If you do it, copy the tomlc99 git log
--oneline in the SOF commit message. Find some good (and less good)
commit message examples for submodule updates at
https://github.com/thesofproject/sof/commits/main/rimage
First, push the tomlc99 commits you want to be tested to any branch of your tomlc99 fork on Github. Do _not_ submit an tomlc99 pull request yet.
Copy link

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Use consistent capitalization for the GitHub brand name ("GitHub"), including in this sentence.

Copilot uses AI. Check for mistakes.
Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed, thanks!

Copy link

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Grammar: use "a tomlc99 pull request" (not "an tomlc99 pull request").

Copilot uses AI. Check for mistakes.
Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed grammar!


Note your tomlc99 fork must have been created using the actual "fork" button on Github so Github is aware of the connection with the upstream tomlc99 repo. In the top-left corner you should see `forked from thesofproject/tomlc99` under the name of your fork. If not then search the Internet for "re-attach detached github fork".

Then, **pretend** these tomlc99 commits have already been accepted and merged (they have been neither) and submit to SOF a draft pull request that updates the main SOF branch with your brand new tomlc99 commits to test. The only SOF commit in this SOF TEST pull request is an SOF commit that updates the tomlc99 pointer to the SHA of your last tomlc99 commit. If you're not sure how to do this then you must go back to the previous section and practice submodules more.

Submit this SOF pull request as a Github _draft_ so reviewers are _not_ notified. Starting every pull request as a draft is always a good idea but in this case this particular SOF pull request can be especially confusing because it points at commits in a different repo and commits that are not merged yet. So you _really_ don't want to bother busy reviewers (here's a secret: some of the reviewers don't like submodules either). You can freely switch back and forth between draft and ready status and should indeed switch to draft if you forgot at submission time but you can never "un-notify" reviewers.

Github has very good support for submodules and will display your SOF TEST pull request better than what the git command line can show. For instance Github will list your tomlc99 changes directly in the SOF Pull Request. So if something looks unexpected on Github then it means you did something wrong. Stop immediately (except for switching to draft if you forgot) and ask the closest git guru for help.
Comment on lines +37 to +43
Copy link

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Use consistent capitalization for the GitHub brand name in this paragraph ("Github" appears multiple times).

Copilot uses AI. Check for mistakes.
Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Capitalization fixed, thanks.


Search for "Submodule" in the build logs and make sure the last of your new tomlc99 commits has been checked out.

Iterate and force-push your tomlc99 branch and your SOF TEST pull request until all the SOF tests pass. Then you can submit your tomlc99 pull request as usual. In the comments section of the tomlc99 pull request, point at your test results on the SOF side to impress the tomlc99 reviewers and get your tomlc99 changes merged faster.

Finally, after your tomlc99 changes have been merged, you can if you want submit one final SOF pull request that points to the final tomlc99 SHA. Or, if your tomlc99 change is not urgently needed, you can just wait for someone else to do it later. If you do it, copy the tomlc99 git log --oneline in the SOF commit message. Find some good (and less good) commit message examples for submodule updates at https://github.com/thesofproject/sof/commits/main/rimage

## Deep Dive: ELF to Image Conversion and Manifest Structure

`rimage` is responsible for converting standard ELF libraries and executables into DSP-compatible firmware images. It parses input ELF files, extracts loadable sections, calculates cryptographic hashes and signatures, and encapsulates them with platform-specific manifests.

### The Build Process (ELF to Image)

```mermaid
graph TD
A[Command Line Args] --> B[Initialize ADSP Config]
C[Input ELF Files] --> D[Parse ELF Headers & Sections]
D --> E[Register Modules & Metadata]
B --> F[Allocate Manifest Memory]
E --> F
F --> G["Copy .text, .rodata, .bss to Image Buffer"]
G --> H["Hash Module Segments (SHA-256 / SHA-384)"]
H --> I["Finalize Manifest Headers (CSS/CSE/FW Desc)"]
Comment on lines +63 to +66
Copy link

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The build-flow diagram says ".bss" is copied into the image buffer, but the implementation only copies loadable text/rodata sections; BSS is represented as a segment size/VA and not copied from the ELF (see man_module_create() in src/manifest.c). Update the diagram/step text to avoid implying BSS bytes are copied into the output image.

Copilot uses AI. Check for mistakes.
Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Removed .bss from diagram, thanks.

I --> J[Sign Root Manifest with RSA Key]
J --> K[Write Output Files]
K --> L[.ri Signed Image]
K --> M[.uns Unsigned Module]
K --> N[.met Manifest Metadata]
```

1. **Configuration Initialization**: `rimage` parses its command-line arguments to find the ADSP machine configuration (e.g., Apollolake `apl`, Tigerlake `tgl`, Meteorlake `mtl`). This configuration defines the specific sizes, offsets, and versions the firmware image will require.
2. **Parsing ELF Files**: The `module_open()` and `module_parse_sections()` functions read the ELF headers (`.text`, `.rodata`, `.bss`). These sections are extracted to form the payload of the modules.
3. **Module Registration**: `rimage` reads either the `.module` metadata section directly from the embedded ELF, or a corresponding `TOML` configuration file, which holds instructions on module attributes (e.g., UUID, thread affinity, entry points).
4. **Manifest and Memory Allocation**: Space is pre-allocated for the firmware image. `man_init_image_xxx()` loads the ADSP template manifest into the start of the image buffer. Then, `man_copy_elf_sections()` copies `.text` and `.rodata` sections directly into the buffer, aligning properties sequentially up to `MAN_PAGE_SIZE` (4096 bytes).
5. **Cryptographic Hashing**: Using `hash_sha256()` or `hash_sha384()` (depending on the platform), `rimage` generates a cryptographic digest of each module's `.text` and `.rodata` block. These hash digests are stored in the respective `sof_man_module` entries of the manifest.
6. **Manifest Signing**: Once all modules and internal structures are mapped, the entire structure is finalized with converged security headers (`ri_css_xxx_hdr_create`, `ri_cse_create`). The root manifest signature is then signed (`ri_manifest_sign_...()`) with an RSA private key.
7. **Writing the Payload**: Finally, the newly encapsulated firmware is written to a `.ri` file (`man_write_fw_mod()`). `rimage` also outputs a stripped `.uns` (unsigned module payload) and `.met` (manifest metadata) file for debugging.

### Manifest Structure

```mermaid
graph TD
A[FW Image Manifest]
A --> B[1. CSS Header]
A --> C[2. CSE Header]
A --> D[3. Firmware Descriptor]
A --> E[4. Module Entries Array]
A --> F[5. ADSP Extension Metadata]
Comment on lines +86 to +91
Copy link

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The manifest-structure diagram/text lists the CSS header before the CSE partition directory, but for v1.8+/v2.5 manifests the CSE partition dir header is at offset 0 and the CSS header follows (see struct fw_image_manifest_v1_8/_v2_5 in src/include/rimage/manifest.h). Consider reordering the diagram or clarifying that ordering differs by manifest version.

Copilot uses AI. Check for mistakes.
Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added a clarification note for v1.8+/v2.5 manifest permutations!


B --> B1("RSA Signatures")
B --> B2("Public Key & Modulus")

C --> C1("Security Partitions Directory")

D --> D1("Versions & Build IDs")
D --> D2("num_module_entries")

E --> E1[Module 0]
E --> E2[Module N]
E1 --> E1a("UUID, Entry Point")
E1 --> E1b("Segment Relative Offsets")
E1 --> E1c("SHA Hash of Sub-Segments")

F --> F1("Proprietary Components")
F --> F2("Hardware Specific Hashes")
```

The firmware image begins with a structured Manifest. Different platforms use altered sizes and variants (e.g., v1.5, v1.8, v2.5, ace_v1.5), but they typically follow the same overarching hierarchy:

1. **CSS Header** (`struct css_header`): Cryptographic Signature Structure. Defines the fundamental size, modulus, public key, and RSA signatures used to authenticate the firmware image on the CSME/DSP side.
2. **CSE Header** (`struct cse_header`): Converged Security Engine Header. A directory mapping security partitions to the ADSP firmware metadata.
3. **Firmware Descriptor** (`struct sof_man_fw_desc`):
Comment on lines +113 to +115
Copy link

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The README references struct css_header and struct cse_header, but the code defines versioned/typed structs (e.g., struct css_header_v1_8/_v2_5 and struct CsePartitionDirHeader/_v2_5). Update these names (or describe them generically) to keep the documentation aligned with the actual types.

Suggested change
1. **CSS Header** (`struct css_header`): Cryptographic Signature Structure. Defines the fundamental size, modulus, public key, and RSA signatures used to authenticate the firmware image on the CSME/DSP side.
2. **CSE Header** (`struct cse_header`): Converged Security Engine Header. A directory mapping security partitions to the ADSP firmware metadata.
3. **Firmware Descriptor** (`struct sof_man_fw_desc`):
1. **CSS Header** (CSS header struct, e.g. `struct css_header_v1_8`, `struct css_header_v2_5`): Cryptographic Signature Structure. Defines the fundamental size, modulus, public key, and RSA signatures used to authenticate the firmware image on the CSME/DSP side.
2. **CSE Header** (CSE partition directory header struct, e.g. `struct CsePartitionDirHeader`, `struct CsePartitionDirHeader_v2_5`): Converged Security Engine Header. A directory mapping security partitions to the ADSP firmware metadata.
3. **Firmware Descriptor** (`struct sof_man_fw_desc` and versioned variants):

Copilot uses AI. Check for mistakes.
Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed, thanks!

- `header`: Holds the basic firmware versions, build IDs, `preload_page_count`, and the `num_module_entries`.
4. **Module Entries** (`struct sof_man_module`): An array corresponding to each module contained in the image.
- `struct_id`: Has the literal value `"$AME"`.
- `name`, `uuid`, `entry_point`, `affinity`.
- `segment`: Details on the relative position `file_offset` and memory offset `v_base_addr` of `.text`, `.rodata`, and `.bss` partitions.
- `hash`: SHA-256 or SHA-384 digest of the active segment.
5. **ADSP Extension Metadata** (e.g., `sof_man_adsp_meta_file_ext_v2_5`): Additional proprietary component descriptors attached after the base manifest, frequently hashed recursively for verification integrity.
Loading