[TroyMitchell911]

The CPU map has no explicit die level, so systems with multiple dies in one socket must flatten their clusters or describe each die as a separate socket.

Add optional dieN nodes between socket and cluster. A single-socket system may omit socket0. Existing descriptions with clusters directly below cpu-map or a socket remain valid, and die topology is independent of NUMA node association.

This is the companion binding proposal for the Linux RFC:
https://lore.kernel.org/all/20260920-riscv-die-topology-rfc-v1-0-071c0bf61d5f@linux.dev/

Kept as a draft for review alongside the kernel implementation.

Validation on the rebased branch:

• All 16 repository tests passed.
• Schema meta-validation, reference checks, and strict YAML lint passed.
• Nine targeted cases passed, covering explicit and omitted sockets, multiple sockets, legacy maps, SMT, and rejection of invalid die contents and mixed die/cluster children within a socket.
• CPU maps from both existing 32-CPU QEMU DTB fixtures validated successfully.

[krzk]

I do wonder why do you need the die information? I could not find the reason here or in https://lore.kernel.org/all/20260920-riscv-die-topology-rfc-v1-0-071c0bf61d5f@linux.dev/, except statement that die topology has to be flatten. OK, but why would that be a problem? Both here and in patchset you should explain what sort of problem are you solving.