Skip to content

Latest commit

 

History

History
62 lines (44 loc) · 3.05 KB

Documentation.md

File metadata and controls

62 lines (44 loc) · 3.05 KB
Raw

Documentation

Memory Management

  • 4K pages are used
  • Cores' boot time stack pointer is placed at 4M, 6M, 8M, 10M

RPi4 Physical Memory Layout

0xFC000000 and above ( > 3G960M) devices

0x40000000 - 0xFC000000 (1G ~ 3G960M) free

0x38400000 - 0x40000000 (948M ~ 1G) VC used

0x00000000 - 0x38400000 (0 ~ 948M) free

The kernel is loaded at address 0x00080000 (512K), currently the image is ~40K, avoid using the memory region < 1M should be good now.

Physical Memory Dynamic Allocation

get_free_page() allocates physical frames from 1G upwards to 3G960M

Virtual Memory Design

To setup virtual memory, start by mapping the first 2M of memory identically and map all 4G of physical RAM to virtual address starting from 0xffff000000000000, turn on MMU then branch to high address (must relocate the of the high address).

Kernel Virtual Memory Layout

map property PA start PA end VA start VA end flags
identity 0x0 0x1000000(16M) 0x0 0x1000000(16M) normal-nc
high address first part 0x0 0x3B400000(948M) 0xffff000000000000 0xffff00003B400000 normal-nc
high address vc 0x3B400000 0x40000000(1G) 0xffff00003B400000 0xffff000040000000 no map
high address second part 0x40000000 0xFC000000(3G960M) 0xffff000040000000 0xffff0000FC000000 normal-nc
high address device 0xFC000000 0x100000000(4G) 0xffff0000FC000000 0xffff000100000000 device

Process Memory Management

copy_process allocates a page for the process's task_struct and kernel stack, then copy_virt_memory is called which loops with allocate_user_page which allocates pages used for the page tables and user code/data.

Relocation

Because of PC-relative addressing, KERNEL_START macro should still work after moving to high address, but some help is needed when we load the high address.

User space

Because we don't have storage support, the only way for user programs to work is to link it inside the kernel image, but this also causes the ksyms to include symbols in user space. Later on it might be possible to compile kernel and user code separately, stick them together raw, and load it as the "kernel image".

Tracing

Ksyms

Short for kernel symbols, used by tracing, main purpose is to be able to translate addresses to symbol names in runtime.

Ksyms notes

  • _symbols: section name for the symbol area
  • src/symbol_area.S: assembly file generated after the first pass, containing formated nm output, to be linked in in the second pass
  • ksyms: the symbol which code accesses, it is the start of the symbol area

Ksyms implementation

Concerns (nasty stuff)

  • syms_values.symbol_area_file has to be under the src/ directory
  • Is bss_end section close to the limit of the adr instruction at the start of the booting process?
  • There are two copies of output formating code for MiniUART and PL011UART
  • In sysregs.h, some values use the BIT(x) macro, some don't

Tools used on development machine

  • python3, for symbol generation
  • nm (both native and cross compiling version should work)