memory_map_kernel.md

Memory Map

All kernel threads share the same memory map. For the user space processes memory maps, see memory_map_process.

Kernel 32-bit

Special pages mapped to fixed high locations

VA start	VA end	alias	mapped from	Permissions
FFFF F000	FFFF FFFF	TRAMPOLINE	char trampoline[]	R, X
FFFF E000	FFFF EFFF
FFFF D000	FFFF DFFF	proc 0 kernel stack	kalloc()	R, W
FFFF C000	FFFF CFFF	(empty guard)
FFFF B000	FFFF BFFF	proc 1 kernel stack	kalloc()	R, W
FFFF A000	FFFF AFFF	(empty guard)
FFFF 9000	FFFF 9FFF	proc 2 kernel stack	kalloc()	R, W
		... to proc max

RAM (bare metal)

On bare-metal qemu the RAM starts at 0x8000.0000:

VA start	VA end	alias	mapped from	Permissions
end_of_text	87FF FFFF	~128 MB RAM	VA = PA	R, W
8000 0000	end_of_text	kernel binary	VA = PA	R, X

RAM (SBI)

On SBI enabled qemu the usable RAM starts at 0x8020.0000:

VA start	VA end	alias	mapped from	Permissions
end_of_text	87FF FFFF	~128 MB RAM	VA = PA	R, W
8020 0000	end_of_text	kernel binary	VA = PA	R, X

Memory Mapped IO Devices

VA start	VA end	alias	mapped from	Permissions
1000 2000	7FFF FFFF	unused
1000 1000	1000 1FFF	VIRTIO disk	VA = PA	R, W
1000 0000	1000 0FFF	UART0	VA = PA	R, W
		unused
0C00 0000	0FFF FFFF	PLIC	VA = PA	R, W
		unused
0200 0000	0200 FFFF	CLINT	VA = PA	R, W
		unused
0010 1000	0010 1FFF	RTC	VA = PA	R, W
0010 0000	0010 0FFF	VIRTIO test	VA = PA	R, W

• kalloc(): One page provided by kalloc() mapped.
• VA = PA: the virtual address matches the physical address

Kernel 64-bit

Same as 32-bit, except that virtual memory goes up to 0x40.0000.0000 so all special pages are moved up. sv39 supports a max virtual memory address of 2^39 which is 0x80.0000.0000, but VIMIX does not use the highest bit to avoid bit extending addresses.

---

Overview: kernel

Boot: boot_process | init_overview

Subsystems: interrupts | devices | file_system | memory_management processes | scheduling | syscalls