How Kernel‑Space Memory Profiling Works with Code Tagging (memprofiling)

This piece introduces a kernel‑space memory profiler (memprofiling) created by Kent Overstree and Suren Baghdasaryan, which visualizes which modules, files, functions and line numbers are responsible for kernel memory allocations.

The profiler relies on a code tagging library that records allocation callers in linear arrays, making iteration straightforward.

On top of the basic code tag, an alloc_tag structure adds a counter for the number of bytes allocated and the number of allocation calls at each code location. When the same location is hit again, the counter is incremented.

The profiler replaces each allocation function with a hook that records the tag at allocation time.

The key macro is DEFINE_ALLOC_TAG, which defines a static struct alloc_tag _alloc_tag placed in a dedicated linker section via __section(ALLOC_TAG_SECTION_NAME). Because the tags reside in a static section, their addresses are known at compile time.

To locate a specific tag, the profiler subtracts the start address of the ALLOC_TAG_SECTION_NAME section from the tag’s address, yielding its offset within the linear tag array without needing runtime debug information.

Overall, the memory‑allocation profiling implementation leverages compile‑time preprocessing and special linker‑section tricks to keep CPU overhead low, making the profiler suitable for both debugging and production deployment.