In the fifth [part](https://0xax.gitbook.io/linux-insides/summary/booting/linux-bootstrap-5) of the series `Linux kernel booting process` we learned about what the kernel does in its earliest stage. In the next step the kernel will initialize different things like `initrd` mounting, lockdep initialization, and many many other things, before we can see how the kernel runs the first init process.
In my view, memory management is one of the most complex parts of the Linux kernel and in system programming in general. This is why we need to get acquainted with paging, before we proceed with the kernel initialization stuff.
`Paging` is a mechanism that translates a linear memory address to a physical address. If you have read the previous parts of this book, you may remember that we saw segmentation in real mode when physical addresses are calculated by shifting a segment register by four and adding an offset. We also saw segmentation in protected mode, where we used the descriptor tables and base addresses from descriptors with offsets to calculate the physical addresses. Now we will see paging in 64-bit mode.
> Paging provides a mechanism for implementing a conventional demand-paged, virtual-memory system where sections of a program’s execution environment are mapped into physical memory as needed.
So... In this post I will try to explain the theory behind paging. Of course it will be closely related to the `x86_64` version of the Linux kernel, but we will not go into too much details (at least in this post).
We already saw where those bits were set in [arch/x86/boot/compressed/head_64.S](https://github.com/torvalds/linux/blob/16f73eb02d7e1765ccab3d2018e0bd98eb93d973/arch/x86/boot/compressed/head_64.S):
Paging divides the linear address space into fixed-size pages. Pages can be mapped into the physical address space or external storage. This fixed size is `4096` bytes for the `x86_64` Linux kernel. To perform the translation from linear address to physical address, special structures are used. Every structure is `4096` bytes and contains `512` entries (this only for `PAE` and `IA32_EFER.LME` modes). Paging structures are hierarchical and the Linux kernel uses 4 level of paging in the `x86_64` architecture. The CPU uses a part of linear addresses to identify the entry in another paging structure which is at the lower level, physical memory region (`page frame`) or physical address in this region (`page offset`). The address of the top level paging structure located in the `cr3` register. We have already seen this in [arch/x86/boot/compressed/head_64.S](https://github.com/torvalds/linux/blob/16f73eb02d7e1765ccab3d2018e0bd98eb93d973/arch/x86/boot/compressed/head_64.S):
We build the page table structures and put the address of the top-level structure in the `cr3` register. Here `cr3` is used to store the address of the top-level structure, the `PML4` or `Page Global Directory` as it is called in the Linux kernel. `cr3` is 64-bit register and has the following structure:
* Bits 4 : 3 - PWT or Page-Level Writethrough and PCD or Page-level cache disable indicate. These bits control the way the page or Page Table is handled by the hardware cache;
* 64-bit linear address is split into some parts. Only low 48 bits are significant, it means that `2^48` or 256 TBytes of linear-address space may be accessed at any given time.
*`47:39` bits of the given linear address store an index into the paging structure level-4, `38:30` bits store index into the paging structure level-3, `29:21` bits store an index into the paging structure level-2, `20:12` bits store an index into the paging structure level-1 and `11:0` bits provide the offset into the physical page in byte.
Every access to a linear address is either a supervisor-mode access or a user-mode access. This access is determined by the `CPL` (current privilege level). If `CPL < 3` it is a supervisor mode access level, otherwise it is a user mode access level. For example, the top level page table entry contains access bits and has the following structure (See [arch/x86/include/asm/pgtable_types.h](https://github.com/torvalds/linux/blob/16f73eb02d7e1765ccab3d2018e0bd98eb93d973/arch/x86/include/asm/pgtable_types.h) for the bit offset definitions):
After you've compiled and installed the Linux kernel, you can see the `System.map` file which stores the virtual addresses of the functions that are used by the kernel. For example:
We can see `0xffffffff81efe497` here. I doubt you really have that much RAM installed. But anyway, `start_kernel` and `x86_64_start_kernel` will be executed. The address space in `x86_64` is `2^64` wide, but it's too large, that's why a smaller address space is used, only 48-bits wide. So we have a situation where the physical address space is limited to 48 bits, but addressing still performs with 64 bit pointers. How is this problem solved? Look at this diagram:
This solution is `sign extension`. Here we can see that the lower 48 bits of a virtual address can be used for addressing. Bits `63:48` can be either only zeroes or only ones. Note that the virtual address space is split into 2 parts:
Userspace occupies the lower part of the virtual address space, from `0x000000000000000` to `0x00007fffffffffff` and kernel space occupies the highest part from `0xffff8000000000` to `0xffffffffffffffff`. Note that bits `63:48` is 0 for userspace and 1 for kernel space. All addresses which are in kernel space and in userspace or in other words which higher `63:48` bits are zeroes or ones are called `canonical` addresses. There is a `non-canonical` area between these memory regions. Together these two memory regions (kernel space and user space) are exactly `2^48` bits wide. We can find the virtual memory map with 4 level page tables in the [Documentation/x86/x86_64/mm.txt](https://github.com/torvalds/linux/blob/16f73eb02d7e1765ccab3d2018e0bd98eb93d973/Documentation/x86/x86_64/mm.txt):
We can see here the memory map for user space, kernel space and the non-canonical area in-between them. The user space memory map is simple. Let's take a closer look at the kernel space. We can see that it starts from the guard hole which is reserved for the hypervisor. We can find the definition of this guard hole in [arch/x86/include/asm/page_64_types.h](https://github.com/torvalds/linux/blob/16f73eb02d7e1765ccab3d2018e0bd98eb93d973/arch/x86/include/asm/page_64_types.h):
Previously this guard hole and `__PAGE_OFFSET` was from `0xffff800000000000` to `0xffff87ffffffffff` to prevent access to non-canonical area, but was later extended by 3 bits for the hypervisor.
Next is the lowest usable address in kernel space - `ffff880000000000`. This virtual memory region is for direct mapping of all the physical memory. After the memory space which maps all the physical addresses, the guard hole. It needs to be between the direct mapping of all the physical memory and the vmalloc area. After the virtual memory map for the first terabyte and the unused hole after it, we can see the `kasan` shadow memory. It was added by [commit](https://github.com/torvalds/linux/commit/ef7f0d6a6ca8c9e4b27d78895af86c2fbfaeedb2) and provides the kernel address sanitizer. After the next unused hole we can see the `esp` fixup stacks (we will talk about it in other parts of this book) and the start of the kernel text mapping from the physical address - `0`. We can find the definition of this address in the same file as the `__PAGE_OFFSET`:
Usually kernel's `.text` starts here with the `CONFIG_PHYSICAL_START` offset. We have seen it in the post about [ELF64](https://github.com/0xAX/linux-insides/blob/master/Theory/ELF.md):
Here I check `vmlinux` with `CONFIG_PHYSICAL_START` is `0x1000000`. So we have the start point of the kernel `.text` - `0xffffffff80000000` and offset - `0x1000000`, the resulted virtual address will be `0xffffffff80000000 + 1000000 = 0xffffffff81000000`.
We've seen how virtual memory map in the kernel is laid out and how a virtual address is translated into a physical one. Let's take the following address as example:
It's the end of this short part about paging theory. Of course this post doesn't cover every detail of paging, but soon we'll see in practice how the Linux kernel builds paging structures and works with them.
**Please note that English is not my first language and I am really sorry for any inconvenience. If you've found any mistakes please send me PR to [linux-insides](https://github.com/0xAX/linux-insides).**
