Skip to content

Latest commit

 

History

History
65 lines (42 loc) · 4.04 KB

memory_map_process.md

File metadata and controls

65 lines (42 loc) · 4.04 KB

Memory Map of a Process

Each user space process has its own memory map. See memory_map_kernel for the kernel memory map.

User Space Processes 32-bit

Processes inherit the memory map of their parent process in fork. See memory map of the init code / first process (see init_userspace) below. Note that all the init code does is call execv for the init ELF file. So it does not need a stack, but is still set up like any other process with one.

VA start VA end alias mapped from Permissions
FFFF F000 FFFF FFFF TRAMPOLINE char trampoline[] R, X
FFFF E000 FFFF EFFF TRAPFRAME proc->trapframe R, W
FFFF 0000 FFFF 0FFF USER_STACK_HIGH kalloc() R, W
0000 1000 0000 1FFF init code kalloc() & g_initcode memcopy R, W, X, U
0000 0000 0000 0FFF NOT MAPPED!
  • USER_STACK_HIGH could be just below the TRAPFRAME, but is in fact even lower to make the address more recognizable during debugging.
  • The lowest page is not mapped to allow NULL pointer exceptions on access both for read and write accesses.

execv will load the new binary to the locations read from the ELF file. These will be code, data and uninitialized variables in bss starting at location USER_TEXT_START. It's set in the Makefiles to export the same value to the kernel and the user space linker script. Afterwards the memory map looks like this:

VA start VA end alias mapped from Permissions
FFFF F000 FFFF FFFF TRAMPOLINE char trampoline[] R, X
FFFF E000 FFFF EFFF TRAPFRAME proc->trapframe R, W
FFFF 0000 FFFF 0FFF USER_STACK_HIGH kalloc() R, W
heap_begin heap_end heap set via sbrk(), grows up R, W, U
... App size bss kalloc() from ELF
... ... data kalloc() & memcopy from ELF from ELF
0000 1000 text size code kalloc() & memcopy from ELF from ELF
0000 0000 0000 0FFF NOT MAPPED!
  • heap_begin / heap_end are members of struct process (processes).

Stack growing

The stack can grow dynamically: if the app tries to access an invalid page between its stack pointer and its stack, the kernel will grow the stack. This is detected in the user_mode_interrupt_handler() which calls proc_grow_stack().

During scheduling before switching to the process it is checked based on the stack pointer and the current stack size if pages can be freed again. If so proc_shrink_stack() gets called.

Note: The kernel stack is still limited to one page.

User Space 64-bit

Same as 32-bit, except that virtual memory goes up to 0x40.0000.0000.

Overview: kernel

Boot: boot_process | init_overview

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