Use of memcpy in bpf programs

[Dave Thaler (dthaler)]

PR #3324 added helpers like bpf_memcpy.
However some web docs (e.g., stackoverflow thread, bpftrace discussion) recommend using __builtin_memcpy instead (supported by clang and gcc).

    // Copy memory safely using __builtin_memcpy
    __builtin_memcpy(&data.buf, src, sizeof(data.buf));

Though https://blogs.oracle.com/linux/post/intro-to-bcc-2 says it's a bcc thing.

Does the builtin work with clang and ebpf-for-windows?
Is there a difference in functionality?
Aren't the helper functions slower?
Are the helper functions really needed?

Also related to PR #4509

[Dave Thaler (dthaler)]

llvm/llvm-project#93700 says __builtin_memcpy only works if the size is a constant, but also says "Optimization level -O3 introduces memcpy calls with non-constexpr length."

[Dave Thaler (dthaler)]

Another reference on this topic is: foniod/redbpf#227

[Shankar Seal (shankarseal)]

Recommendation is to document __builtin_memcpy in the bpf_memcpy helper function.

[Dave Thaler (dthaler)]

Recommendation is to document __builtin_memcpy in the bpf_memcpy helper function.

Recommend adding to here: https://microsoft.github.io/ebpf-for-windows/bpf__helper__defs_8h.html#a40f52fbbeec82a25a1987641816bb5eb