========================= QEMU capabilities testing ========================= Purpose ======= Test data in this directory is used: - to excercise the capability parsing code in ``qemucapabilitiestest`` - provides "real" capabilities data for test suites such as ``domaincapstest`` ``qemuxmlconftest``, and others - provides the required data to validate the QMP commands used by libvirt against qemu's QMP schema Naming ====== Files in this directory have the following naming scheme:: caps_$QEMUVERSION_$ARCHITECTURE.$SUFFIX or:: caps_$QEMUVERSION_$ARCHITECTURE+$VARIANT.$SUFFIX ``$QEMUVERSION`` Numeric representation of the qemu version, e.g.: ``7.0.0`` ``$ARCHITECTURE`` Architecture string such as ``x86_64``, ``aarch64``, etc. ``$SUFFIX`` ``.replies`` for the dump of the QMP communication used to probe qemu. ``.xml`` for the generated capability dump ``$VARIANT`` The variant name is an optional arbitrary string, not containing any dot. A variant is an additional named version of capabilities for given version and architecture tuple. This allows for testing special cases which e.g. depend on a specific host platform or operating system feature, which differs from the main tests. Note that in the test code the variant name is an empty string or includes the '+' sign for ease of use. Known test variants ------------------- ``+hvf`` Variant of the test data using the Apple OSX Hypervisor Framework acceleration for qemu. Usage in tests ============== Test suites such as ``qemucapabilitiestest`` or ``domaincapstest`` pick up the test data automatically once the corresponding ``.xml`` or ``.replies`` file is present in ``tests/qemucapabilitiesdata``. Other test suites such as ``qemuxmlconftest`` provide macros which invoke test cases using this data such as ``DO_TEST_CAPS_LATEST``. Capturing QEMU capabilities =========================== QEMU capabilities are captured by running the ``qemucapsprobe`` on the QEMU binary on given architecture and then capturing the output. Since virtualization acceleration is also probed it's required to run it on real hardware. The capabilities dumps contain also host-specific information such as the exact CPU definition of the machine where it was ran on, thus they can differ significantly when run on other machines. Probing QEMU ------------ Run the ``qemucapsprobe`` tool:: $ LIBVIRT_BUILDDIR/tests/qemucapsprobe /path/to/qemu > output.replies The tool spawns the qemu binary and performs probing as if libvirt would do that. The QMP conversation between qemu and libvirt is dumped to stdout. User running the probe must be able to access the virtualization accelerator (e.g. have proper permissions on ``/dev/kvm``) Generating the output files --------------------------- Place the captured output ``.replies`` file into this directory and run:: $ VIR_TEST_REGENERATE_OUTPUT=1 ninja test This runs the test-suite instructing it to update and/or generate all new data the test would normally expect. Manual modifications the ``.replies`` file ========================================== In certain cases it's impractical or impossible to re-generate the ``.replies`` file on a code change causing a change to the actual QMP query process. In such case a careful manual modification of the ``.replies`` is tolerated. To aid such modification the tool ``scripts/qemu-replies-tool.py`` can be used. The tool validates and updates the numbering of the entries in the QMP dump in case something was modified. The tool also allows programatic modification of the ``.replies`` file. Fake test data dumps for certain architectures ============================================== For some architectures it was impossible or impractical to fetch real capability dumps. To ensure coverate of certain cases the dumps were collected from corresponding binaries running on a different architecture. Capabilities dumps for the following architectures are usually produced on real hardware: - x86_64 - aarch64 - ppc64 - s390x In most other cases, x86_64 will be used as the host architecture. A fake caps dump can be usually spotted by absence of KVM support.