Hi,
ThisMayWork wrote: My kernel is now able to manage physical memory (using a bitmap), I enabled paging and I can also map virtual to physical addresses. I identity-mapped the first 4MB so I can use the video buffer and to prevent anything below 1MB from going crazy and I also mapped my kernel to 3GB. For that, I used two page tables (which are allocated using my physical memory manager). My question is, what do I do with the rest of the virtual memory addresses? Do I identity map them as well?
Typically the first thing you do is split virtual address spaces into zones - maybe one huge area for processes (e.g. "user space" from 0x00000000 to 0xBFFFFFFFF) and another huge area for the kernel itself (e.g. "kernel space" from 0xC0000000 to 0xFFFFFFFF).
Then those zones are split into "sub-zones". More specifically; "user space" is split into sub-zones by the process (e.g. into ".text", ".rodata", ".data" and ".bss" by the linker script, and then the remainder used for shared libraries and/or garbage collection and/or malloc/free and/or whatever the process feels like) where how user-space is split up into sub-zones is none of the kernel's concern (kernel only provides functionality that allows a process to do whatever it wants); and kernel space is split into sub-zones by the kernel (e.g. into ".text", ".rodata", ".data" and ".bss" by the kernel's linker script, and then the remainder used for kernel modules and/or garbage collection and/or malloc/free and/or whatever the kernel feels like).
For an example, for my micro-kernels I always use one part of kernel-space (e.g. from 0xFFC00000 to 0xFFFFFFFF) for "recursive mapping" page tables (to make virtual memory management easier), and another (relatively large) part of kernel-space (e.g. from 0xC0000000 to 0xDFFFFFFF) for message buffers (where this "message buffer area" is managed explicitly by my kernel's message handling code).
ThisMayWork wrote: Until I get to the stage of a heap allocator, or even plan the memory layout, what is the "typical" approach to handle the "free" pages?
Typically, if you have no use for an area of the virtual address space then you leave that area marked as "not present" in page tables, etc; so that if software accesses something it shouldn't it causes a page fault, which increase the chance of detecting bugs (e.g. dodgy pointers, etc).
Cheers,
Brendan
|
) and overwrites part of the memory it is not supposed to. If the pages are identity-mapped, will it be easier for me to debug? Thank you for your answers!