How kernel page table get initialized?

0
votes

I am following Gorman's virtual memory management book.

There is a section about kernel table page initialization which is said to be divided into two phases, bootstrapping and finalizing.

Here is what it says about the bootstrapping phase.

The assembler function startup_32() is responsible for enabling the paging unit in arch/i386/kernel/head.S. While all normal kernel code in vmlinuz is compiled with the base address at PAGE_OFFSET + 1MiB, the kernel is actually loaded beginning at the first megabyte (0x00100000) of memory. The first megabyte is used by some devices for communication with the BIOS and is skipped. The bootstrap code in this file treats 1MiB as its base address by subtracting __PAGE_OFFSET from any address until the paging unit is enabled. Therefore before the paging unit is enabled, a page table mapping has to be established that translates the 8MiB of physical memory to the virtual address PAGE_OFFSET.

  1. Why we want to subtract __PAGE_OFFEST? For what purpose?

  2. Why we have to do subtracting before the paging unit is enabled? Isn't that we always use subtracting for mapping kernel virtual address to physical memory address?

  3. Why it is 8MB?

Thanks,

1
linuxmemory-managementkernel

1 Answers

4
votes

Since x86 code isn't generally position-independent, if it's compiled to execute at address X (__PAGE_OFFSET + 1MB) but loaded at address Y (1MB), all addresses inside of it need to be decremented by Y-X (__PAGE_OFFSET + 1MB - 1MB = __PAGE_OFFSET) for it to work.

For example, if there's an instruction to read a byte of memory from the beginning of the kernel, __PAGE_OFFSET + 1MB, the address is reduced by __PAGE_OFFSET and the actual read location becomes 1MB, exactly where the kernel starts in the memory.

When page translation is finally enabled, __PAGE_OFFSET can and, I believe, is effectively subtracted by the page translation mechanism by mapping a range of virtual addresses to a range of physical addresses that are smaller by __PAGE_OFFSET (that is, physical=virtual-__PAGE_OFFSET per the page tables).

Unless there's some additional kernel relocation involved, 8MB is likely just the size of the mapping range, sufficient to map the entire kernel.