gdt: make add_entry const

[josephlr]

This requires making the push method const as well, which means we need
the const_panic feature (not too much of a risk as this feature is
failly well along).

I also moved the pointer calculation (for GDT and IDT) into their own
(currently private) methods. Eventually these methods can be public, but
that's only useful if they can be const, which would require a breaking change.

The caclulation for the GDT pointer was also updated to use:

self.next_free * size_of::<u64>() - 1

instead of

self.table.len() * size_of::<u64>() - 1

Signed-off-by: Joe Richey [email protected]

[phil-opp]

Looks all reasonable to me, thanks!

[phil-opp]

Eventually these methods can be public, but
that's only useful if they can be const, which would require a breaking change.

Could you clarify which change would be breaking?

[josephlr]

Could you clarify which change would be breaking?

To make GlobalDescriptorTable::pointer() a const fn, DescriptorTablePointer.base would need to have a pointer type instead of a u64 type, and that field is public.

My current idea for a later PR:

• Change DescriptorTablePointer.base to type VirtAddr
• Make it so VirtAddr::from_ptr could be const fn (have some new from_ptr_unchecked method).

[phil-opp]

Sounds good to me! However, I think we can make the VirtAddr::new check a const fn too: https://play.rust-lang.org/?version=nightly&mode=debug&edition=2018&gist=87076d1f107a5a36970175cbb1bc9fa9

[phil-opp]

Ideally, I would want to make the bit_field crate functions const at some point, but right now not all of its features are const compatible (e.g. being generic over RangeBounds).