Bazel Notes

Both OpenTitan hardware and software is built with Bazel. While our Getting Started guides detail some of the Bazel commands that can be used to build both types of artifacts, below are detailed notes on:

  • how Bazel is configured for our project, and
  • brief examples of Bazel commands that are useful for:
    • querying,
    • building, and
    • running tests with Bazel.

OpenTitan Bazel Workspace

The rules for Bazel are described in a language called Starlark, which looks a lot like Python.

The $REPO_TOP directory is defined as a Bazel workspace by the //WORKSPACE file. BUILD files provide the information Bazel needs to build the targets in a directory. BUILD files also manage any subdirectories that don't have their own BUILD files.

OpenTitan uses .bzl files to specify custom rules to build artifacts that require specific attention like on-device test rules and project specific binaries.

WORKSPACE file

The WORKSPACE file controls external dependencies such that builds can be made reproducible and hermetic. Bazel loads specific external dependencies, such as various language toolchains. It uses them to build OpenTitan targets (like it does with bazel_embedded) or to satisfy dependencies (as it does with abseil). To produce increasingly stable releases the external dependencies loaded in WORKSPACE file attempts to fix a all external http_archives to a specific SHA. As we add more dependencies to the workspace, builds and tests will become less sensitive to external updates, and we will vastly simplify the Getting Started instructions.

BUILD files

Throughout the OpenTitan repository, BUILD files describe targets and dependencies in the same directory (and subdirectories that lack their own BUILD files). BUILD files are mostly hand-written. To maintain the invariant that hand-written files not be included in autogen directories, there are BUILD files that describe how to build and depend on auto-generated files in autogen subdirectories.

General Commands

  • Build everything (software and Verilator hardware):
    bazel build //...
    
  • Build and run all tests (on-host tests, and Verilator/FPGA on-device tests):
    bazel test --test_tag_filters=-broken,-dv //sw/...
    
    Note: this will take several hours.
  • Clean all build outputs and reclaim all disk and memory traces of a Bazel instance:
    bazel clean --expunge
    
    Note: you should rarely need to run this, see below for when this may be necessary.

Locating Build Artifacts

When Bazel builds a target, or the dependencies of a target, it places them in a Bazel-managed output directory. This can make them difficult to find manually, however, Bazel also provides a mechanism to query for built artifacts that is demonstrated below. Note, the outquery-* command shown below is a special command that is parsed via our bazelisk.sh script. Therefore, it cannot be run with a standalone bazel ... invocation.

opentitan_{rom,flash}_binary Artifacts

  • Query the locations of all Bazel-built artifacts for all OpenTitan devices for an opentitan_{rom,flash}_binary macro:
    ./bazelisk.sh outquery-all <target>
    
  • Query the locations of all Bazel-built artifacts for a specific OpenTitan device for an opentitan_{rom,flash}_binary macro:
    ./bazelisk.sh outquery-all <target>_<device>
    
    Note: <device> will be in {sim_dv, sim_verilator, fpga_cw310}.

See Building (and Testing) Software, device software can be built for multiple OpenTitan devices and memories, using OpenTitan-specific Bazel macros.

opentitan_functest Artifacts

As described Building (and Testing) Software, device software can be built for multiple OpenTitan devices and memories, using OpenTitan-specific Bazel macros. Since running tests on multiple OpenTitan devices (whether DV or Verilator simulation, or an FPGA) involves building several software images for multiple memories, we provide a Bazel macro for this. This macro is called opentitan_functest.

  • List all sh_test targets instantiated by a opentitan_functest, e.g. the UART smoketest:
    bazel query 'labels(tests, //sw/device/tests:uart_smoketest)'
    
  • Query the HW and SW dependencies of a specific opentitan_functest for the fpga_cw310 device, e.g. the UART smoketest:
    bazel query 'labels(data, //sw/device/tests:uart_smoketest_fpga_cw310)'
    
    or for any opentitan_functest target and <device> in {sim_dv, sim_verilator, fpga_cw310}
    bazel query 'labels(data, <target>_<device>)'
    
  • Query the software artifacts built for the opentitan_flash_binary that is a dependency of an opentitan_functest for the fpga_cw310 device, e.g. the UART smoketest:
    bazel query 'labels(srcs, //sw/device/tests:uart_smoketest_prog_fpga_cw310)'
    
    or for any opentitan_functest <target> and <device> in {sim_dv, sim_verilator, fpga_cw310}
    bazel query 'labels(srcs, <target>_prog_<device>)'
    
    Note: if an opentitan_functest target has the name foo, then the opentitan_flash_binary target that is instantiated by the opentitan_functest will be named foo_prog_<device>.

Building Software

  • To build OpenTitan software see here, or run
    bazel build <target>
    

Testing Software

On-Host Tests

  • To query-for/run on-host software tests see here.

On-Device Tests

All device software, regardless of the device, is built with Bazel. However, only Verilator simulation and FPGA device software tests can be run with Bazel.

ROM Tests

  • Query for all ROM functional and E2E tests for FPGA:
    bazel query 'filter(".*_fpga_cw310", kind(".*test rule", //sw/device/silicon_creator/...))'
    
    and for Verilator:
    bazel query 'filter(".*_sim_verilator", kind(".*test rule", //sw/device/silicon_creator/...))'
    
  • Run all ROM functional and E2E tests on FPGA:
    bazel test --test_tag_filters=cw310 //sw/device/silicon_creator/...
    
    and for Verilator:
    bazel test --test_tag_filters=verilator //sw/device/silicon_creator/...
    
  • Run a single ROM functional or E2E test on FPGA and see the output in real time:
    bazel test \
      --define DISABLE_VERILATOR_BUILD=true \
      --test_tag_filters=cw310 \
      --test_output=streamed \
      //sw/device/silicon_creator/lib:boot_data_functest
    
    or, remove the define/filtering flags and just append the <device> name like:
    bazel test --test_output=streamed //sw/device/silicon_creator/lib:boot_data_functest_fpga_cw310
    
    and similarly for Verilator:
    bazel test --test_output=streamed //sw/device/silicon_creator/lib:boot_data_functest_sim_verilator
    

Chip-Level Tests

  • Query for all chip-level tests for FPGA:
    bazel query 'filter(".*_fpga_cw310", kind(".*test rule", //sw/device/tests/...))'
    
    and for Verilator:
    bazel query 'filter(".*_sim_verilator", kind(".*test rule", //sw/device/tests/...))'
    
  • Run all chip-level tests on FPGA:
    bazel test --define DISABLE_VERILATOR_BUILD=true --test_tag_filters=cw310 //sw/device/tests/...
    
    and for Verilator:
    bazel test --test_tag_filters=verilator //sw/device/tests/...
    
  • Run a single chip-level test on FPGA and see the output in real time:
    bazel test \
      --define DISABLE_VERILATOR_BUILD=true
      --test_tag_filters=cw310 \
      --test_output=streamed \
      //sw/device/tests:uart_smoketest
    
    or, remove the define/filtering flags and just append the <device> name like:
    bazel test --test_output=streamed //sw/device/tests:uart_smoketest_fpga_cw310
    
    and similarly for Verilator:
    bazel test --test_output=streamed //sw/device/tests:uart_smoketest_sim_verilator
    

Linting Software

There are several Bazel rules that enable running quality checks and fixers on code. The subsections below describe how to use them. All of the tools described below are run in CI on every pull request, so it is best to run them before committing code.

Linting C/C++ Code

The OpenTitan supported linter for C/C++ files is clang-format. It can be run with Bazel as shown below.

Run the following to check if all C/C++ code as been formatted correctly:

bazel test //quality:clang_format_check --test_output=streamed

and run the following to fix it, if it is not formatted correctly.

bazel run //quality:clang_format_fix

Linting Starlark

The OpenTitan supported linter for Bazel files is buildifier. It can be run with Bazel as shown below.

Run the following to check if all WORKSPACE, BUILD, and .bzl files have been formatted correctly:

bazel test //quality:buildifier_check --test_output=streamed

and run the following to fix them, if they are not formatted correctly.

bazel run //quality:buildifier_fix

Checking License Headers

Lastly, the OpenTitan supported linter for checking that every source code file contains a license header may be run with:

bazel run //quality:license_check --test_output=streamed

Building Hardware

Note, to run (software) tests on these OpenTitan hardware platforms does not require these Bazel commands be invoked before the test commands above. Bazel is aware of all dependency relationships, and knows what prerequisites to build to run a test.

  • Build FPGA bitstream with (test) ROM, see here.
  • Build FPGA bitstream with (production) ROM, see here.
  • Build Verilator simulation binary:
    bazel build //hw:verilator
    

Miscellaneous

Troubleshooting Builds

If you encounter an unexplained error building or running any bazel commands, you can issue a subsequent bazel clean command to erase any existing building directories to yield a clean build. Specifically, according to the Bazel documentation, issuing a

bazel clean

deletes all the output build directories, while running a

bazel clean --expunge

will wipe all disk and memory traces (i.e., any cached intermediate files) produced by Bazel. The latter sledgehammer is only intended to be used as a last resort when the existing configuration is seriously broken.

Create a .bazelrc File

Create a .bazelrc file in your home directory to simplify executing bazel commands. For example, you can use a .bazelrc to:

  • set up a disk cache, or
  • skip running tests on the CW310 FPGA if you don not have one.

A .bazelrc file that would accomplish this would look like:

# Make Bazel use a local directory as a remote cache.
build --disk_cache=~/bazel_cache

# Skip CW310 FPGA tests, since I do not have said FPGA.
test --test_tag_filters=-cw310

See the .bazelrc file in the OpenTitan repository for more examples. Additionally, for more information see the Bazel documentation.

Site-specific .bazelrc-site options

We use the term “compute site” to refer to a particular development host, or more broadly a collection of compute nodes, that is operated by an organization with its own unique compute requirements or IT policies. The experience of working in such an environment may be different than using an off-the-shelf Linux distribution, and so team members working on some compute sites may require the use of slightly different Bazel options.

In addition to each user's $HOME/.bazelrc file and the project-level configurations in $REPO_TOP/.bazelrc, there is the option to add site-specific configuration options, by adding them to the file $REPO_TOP/.bazelrc-site.

The .bazelrc-site file can be useful for enforcing site-specific policies such as:

  • Default locations for build artifacts (e.g., new defaults for --output_user_root or perhaps --output_base options, outside the $HOME filesystem).
  • Site-specific, non-standard paths for build-tools, libraries, or package configuration data.

Putting Bazel options in this file adds another layer of configuration to help groups of individuals working on a common compute site share the same site-specific default options.

At a more fine-grained level, individual users can still place options in $HOME/.bazelrc, potentially overriding the options used in .bazelrc-site and $REPO_TOP/.bazelrc. However the “$HOME/.bazelrc” file is not specific to OpenTitan, which could create confusion for users working on a number of projects.

The policies and paths for each contributing site could vary greatly. So in order to avoid accidental conflicts between sites, the .bazelrc-site file is explicitly not included in the git repository and OpenTitan CI scripts will reject any commit that includes a .bazelrc-site file.

Disk Cache

Bazel can use a directory on the file system as a remote cache. This is useful for sharing build artifacts across multiple git worktrees or multiple workspaces of the same project, such as multiple checkouts.

Use the --disk_cache=<filename> to specify a cache directory. For example, running

bazel build //... --disk_cache=~/.cache/bazel-disk-cache

will cache all built artifacts. Alternatively add the following to $HOME/.bazelrc to avoid having automatically use the disk cache on every Bazel invocation, as shown above.

Note that Bazel does not perform any garbage collection on the disk cache. To clean out the disk cache, you can set a cron job to periodically delete all files that have not been accessed for a certain amount of time. For example add the following line with the path to your disk cache to your crontab (using crontab -e) to delete all files that were last accessed over 60 days ago.

0 0 * * 0 /usr/bin/find /path/to/disk/cache -type f -atime +60 -delete

For more documentation on Bazel disk caches see the official documentation.

Excluding Verilator Simulation Binary Builds

Many device software targets depend on the Verilator simulation binary, The Verilator simulation binary is slow to build. To avoid building it, use the --define DISABLE_VERILATOR_BUILD=true build option. For example, to build the UART smoke test artifacts but not the Verilator simulation binary run

bazel build --define DISABLE_VERILATOR_BUILD=true //sw/device/tests:uart_smoketest

Displaying Test Outputs in Real Time

By default, Bazel does not write test outputs to STDOUT. To see a test's output in real time, use the --test_output=streamed flag, like:

bazel test --test_output=streamed //sw/device/tests:uart_smoketest_fpga_cw310

Filtering Broken Tests

Some tests are marked known to be broken (and are in the process of being triaged). To prevent running these tests when running a block of tests, use the test_tag_filters=-broken flag. For example, to run all chip-level tests except the broken ones on FPGA:

bazel test --test_tag_filters=cw310,-broken //sw/device/tests/...

Using Bazel with Git Worktrees

Bazel was not optimized for the git worktree workflow, but using worktrees can help with branch management and provides the advantage of being able to run multiple Bazel jobs simultaneously. Here are some tips that can improve the developer experience when using worktrees.

  1. Follow the instructions above to enable the disk cache. Bazel uses the workspace's path when caching actions. Because each worktree is a separate workspace at a different path, different worktrees cannot share an action cache. They can, however, share a disk cache, which helps avoid rebuilding the same artifacts across different worktrees. Note that the repository cache is shared by default across all workspaces, so no additional configuration is needed there.
  2. Before deleting a worktree, be sure to run bazel clean --expunge to remove Bazel's generated files. Otherwise, files from old worktrees can accumulate in your output user root (located at ~/.cache/bazel/_bazel_${USER}/ by default).

Commonly Encountered Issues

Cannot find ld

One issue encountered while using bazel is the following error message when attempting to build:

  = note: collect2: fatal error: cannot find 'ld'
          compilation terminated.

The reason this occurs is related to these issues:

Specifically, when the rustc compiler is invoked, it uses the LLVM linker that is not managed by the bazel flow. This means bazel cannot ensure it is installed at a specific location, and instead just uses whatever is available on the host machine. The issue above points out that rustc expects the linker to be located in the same directory as gcc, so if on the host machine this statement is untrue, there can be build issues.

To resolve this problem:

  1. Install the LLVM linker with, e.g., sudo apt install lld.
  2. Symlink lld to where gcc is installed.

This workaround will be needed until the rust issue is resolved.