Hi @ldmstm! We appreciate you submitting your first issue for our open-source project. 🌟

Even though I'm a bot, I can assure you that the whole community is genuinely grateful for your time and effort. 🤖💙

Not clear what is the problem here, we indeed map the physical memory for all page tables, but we don't map the virtual address for the page table [CONFIG_XTENSA_MMU_PTEVADDR + 4MB], we just pin the l1 page for this area and the auto refill does the rest.

In the dup_tables routine it believes this entry you place at XTENSA_MMU_L1_POS(CONFIG_XTENSA_MMU_PTEVADDR) during kernel startup is a valid L1 page table entry to a different L2 page table, when in reality it's just itself.

When duplicating it, this causes memory to be mapped at CONFIG_XTENSA_MMU_PTEVADDR for other memory domains, when in reality we only want the mapping at the 4MB * user_asid offset

XTENSA_MMU_L1_POS

I think I got what you mean, it is confuse because we don't map the page itself until it is used, so if I understood correctly, we can do:

		if (is_pte_illegal(source_table[i]) || (i == XTENSA_MMU_L1_POS(XTENSA_MMU_PTEVADDR))) {
			dst_table[i] = XTENSA_MMU_PTE_ILLEGAL;
			continue;
		}

I can't see a security issue because it is mapped as using kernel ring and the whole page table (physical memory) is already mapped, but indeed we can save memory doing it :)

Looks perfect to me, a whole 4k saved per memory domain.

Thanks!