kernel oopses with 4GiB of RAM in QEMU

[hawkw]

running in QEMU with -machine accel=KVM -cpu host -m 4G gets us a kernel panic when adding the memory map to the page-frame allocator.

if i run with

$ cargo run-x64 --serial -- -m 4G

i get this oops:

[     ?.??????][i] ├mycelium_kernel:   PAddr(0x01500000)            FREE      3131269122 B
[     ?.??????][*] ├mycelium_kernel::allocator: adding to page allocator, region=Region {
                   │     base: PAddr(0x1500000),
                   │     size: 3131269122,
                   │     kind: FREE,
                   │ }
[     ?.??????][?] ├oops: Oops {
                   │     already_panicked: false,
                   │     already_faulted: false,
                   │     alloc: State {
                   │         allocating: 0,
                   │         deallocating: 1,
                   │         heap_size: 20336640,
                   │         allocated: 0,
                   │         min_size: 64,
                   │         bump_mode: false,
                   │         bump_allocated: 0,
                   │         bump_size: 1024,
                   │     },
                   │     situation: OopsSituation::Panic(
                   │         PanicInfo {
                   │             message: assertion failed: size <= core::i32::MAX as usize,
                   │             location: Location {
                   │                 file: "/home/eliza/Code/mycelium/hal-core/src/mem.rs",
                   │                 line: 103,
                   │                 col: 9,
                   │             },
                   │             can_unwind: true,
                   │             force_no_backtrace: false,
                   │         },
                   │     ),
                   │ }

which is this assertion in mem::Region::split_back:

mycelium/hal-core/src/mem.rs

Lines 102 to 103 in 993d1b9

	pub fn split_back(&mut self, size: usize) -> Option<Self> {
	assert!(size <= core::i32::MAX as usize);

this is because the Address::offset function tries to be clever and takes an i32 as the offset to ensure the offset is always in-bounds:

mycelium/hal-core/src/addr.rs

Lines 82 to 89 in 993d1b9

	fn offset(self, offset: i32) -> Self {
	if offset > 0 {
	self + offset as usize
	} else {
	let offset = -offset;
	self - offset as usize
	}
	}

so we have to make sure that the size of the region is < the max i32 to call that. this is silly. we should be able to handle memory regions bigger than 2,147,483,647 bytes (this one is 3,131,269,122 B).

we could either make Address::offset take an isize isntead, or have split_back not use Address::offset.