How it works
The build pipeline, the TinyGo target, the runtime overlay and the C runtime.
your *.go ──(copy; pragmas → //export)──▶ tinygo build ──▶ go.ll ──IR fixups──▶ tinygo clang ──▶ go.o
gokm/linux (generated from BTF/DWARF) ───┘ + compiler-rt
runtime C (shim, threads, heap) + framework glue + generated wrappers + go.o ──kbuild──▶ name.ko
target.json: TinyGo target for kernel code (kernel code model, static relocations, soft-float, retpolines, threads scheduler, custom GC).- Runtime overlay (
internal/assets/_files/overlay): the platform layer, goroutines as kernel threads (internal/task), and the collector (runtime/gc_linuxkernel.go). - C runtime (
internal/assets/_files/kmod):threads.c(entry gate, thread records, parking, stop-the-world, futex, goroutines),heap.c(heap chunks, background collector),shim.c(module init/exit, services, IBT-safelongjmpforrecover()). - IR fixups (
internal/build/fixup.go):noredzone, return thunks, IBTendbr64, no unwind tables, direct (non-GOT) access to C globals. - compiler-rt builtins (soft-float, 128-bit math) compiled once with the kernel flags; only what a module needs is linked in.
- Frameworks (
_files/lib/<pkg>+_files/frameworks/<pkg>). internal/btf,internal/bindgen: BTF parser andgokm/linuxgenerator.- Stack analysis (
internal/build/stack.go): per-function frame sizes from LLVM, call graph from the IR, deepest paths from every entry point.