Host / Target Compilation and Toolchains

[steven-bellock]

The libspdm build system currently employees DARCH and DTOOLCHAIN to specify the target machine architecture (x64, aarch64, etc), compiler executable (gcc, aarch64-linux-gnu-gcc), and which processor_bind.h file to #include. However, DTOOLCHAIN is slightly awkward in that, for example, when targeting aarch64 with GCC, DTOOLCHAIN=AARCH64_GCC is used rather than DTOOLCHAIN=GCC. There are additional target considerations such as baremetal and endianness. In light of that this thread is to discuss possible changes to how host, target, and toolchain are specified.

In particular CMake is able to detect the host architecture, so if the host and target are the same the user should not have to specify DARCH. We may also need another parameter to specify variants of an architecture's toolchain. For example aarch64 can encompass aarch64-none-elf, aarch64-none-linux-gnu, and aarch64_be-none-linux-gnu.

[LeeTroy]

We're working on porting libspdm to two platforms of our company, and which has integrated build environments, e.g. OpenBMC (Yocto-based) and Zephyr (West + CMake). The integrated build environment has the ability to feed cross compiler variables CMAKE_C_{COMPILER,AR,LINKER} into CMake. Predefining the toolchain executable makes porting harder without patching it.

[jyao1]

@LeeTroy , thanks for the info. Do you have any proposal?

[LeeTroy]

Right now, I have a very sketchy patch for it, I'll submit a PR when I get everything working.