Use huge L3 page table entries to offset map the entire physical memory?

Here is my config:

pub static BOOTLOADER_CONFIG: BootloaderConfig = {
    let mut config = BootloaderConfig::new_default();
    config.mappings.physical_memory = Some(Mapping::Dynamic);
    config
};

When I looped through every page table entry (L4, L3, L2, L1) I noticed that there were only huge pages for L2 entries (so only chunks of 2MiB), and no huge pages for L3 entries. Even with I run qemu with 4G of memory, there are still only many huge L2 entries and no huge L3 entries. Wouldn't it be better if huge L3 entries were used instead of huge L2 entries?

Start with BOOTLOADER_CONFIG and the Dynamic physical_memory mapping path, then trace how L4-L1 entries are created. Reproduce the 4G QEMU case and compare L3 and L2 huge-page handling; done means the feasibility and expected mapping behavior are established, with implementation scope clearly defined.