boot: convert paging from assembly to Zig (take 2) #23
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Zig 0.12 lyckades åstadkomma ett stenras i Lappis, hehe, boot-asm rewrite'n funkar inte längre. |
|
Userland is now working with the following paging table. |
|
CI build requires
|
|
I'll fix the CI build and then merge after a short review. |
This code is MIT licensed and copied from the following repositories: * https://github.com/leecannon/zig-x86_64 * https://github.com/leecannon/zig-bitjuggle
Run equivalent of zigfmt to automate the update.
Note: used gdb to get latest hardcoded value of tss_addr xD
This was needed for debugging before the jump to long mode was fully functional.
This was used to switch from 16-bit segmented mode to 32-bit protected mode. As Lappis is now using Multiboot, the switch from 16-bit to 32-bit mode is done by the multiboot boot loader.
This was used for debug printing in 16-bit mode. As Lappis is now using multiboot, the boot loader handles the switch from 16-bit to 32-bit mode.
Run equivalent of zigfmt to automate the update.
Also, fix build for Zig version 0.11.0 And bump hardcoded tss64 address value.
Prior to this commit, set_up_page_tables was used to both initialize the page table tree mapping and to map kernel code and data segment pages. Split out the latter two into dedicated functions.
The the foo_32.zig files are compiled for 32-bit architecture, and functions are exported in the object file using the `export` keyword. To avoid re-exporting functions and facilitate re-use of declared constants, move paging constants to their own source file and include this source file where needed.
… (Cr0) This ties related logic together.
Also, bump hardcoded tss64 address value.
The absence of the nx bit when using the ld linker exposed a bug in the ld linker, as further described in [1]. In particular this bug lead to the OS being fully functional, but pages being loaded as RWX even though they were supposed to have the nx bit set. Very sneaky, and almost too good to be true, in the sense of "trusting trust" kind of bugs in compilers and linkers. In either case, future commits will create dedicated pages for userland code and data segments, and then make the kernel data segment NX again. Prior to this commit, `info tlb` gave the following: (qemu) info tlb 0000000000000000: 0000000000000000 --PDA--UW 0000000000200000: 0000000000200000 --PDA--UW 0000000000400000: 0000000000400000 --PDA--UW 0000000000600000: 0000000000600000 --PDA--UW 0000000000800000: 0000000000800000 --PDA--UW 0000000000a00000: 0000000000a00000 --PDA--UW 0000000000c00000: 0000000000c00000 --PDA--UW 0000000000e00000: 0000000000e00000 --PDA--UW 0000000001000000: 0000000001000000 --P----UW 0000000001200000: 0000000001200000 --P----UW 0000000001400000: 0000000001400000 --P----UW 0000000001600000: 0000000001600000 --P----UW 0000000001800000: 0000000001800000 --P----UW 0000000001a00000: 0000000001a00000 --P----UW 0000000001c00000: 0000000001c00000 --P----UW 0000000001e00000: 0000000001e00000 --P----UW 0000000002000000: 0000000002000000 --P----UW 0000000002200000: 0000000002200000 --P----UW 0000000002400000: 0000000002400000 --P----UW 0000000002600000: 0000000002600000 --P----UW 0000000002800000: 0000000002800000 --P----UW 0000000002a00000: 0000000002a00000 --P----UW 0000000002c00000: 0000000002c00000 --P----UW 0000000002e00000: 0000000002e00000 --P----UW 0000000003000000: 0000000003000000 --P----UW 0000000003200000: 0000000003200000 --P----UW 0000000003400000: 0000000003400000 --P----UW 0000000003600000: 0000000003600000 --P----UW 0000000003800000: 0000000003800000 --P----UW 0000000003a00000: 0000000003a00000 --P----UW 0000000003c00000: 0000000003c00000 --P----UW 0000000003e00000: 0000000003e00000 --P----UW 0000000004000000: 0000000004000000 --P----U- 0000000004200000: 0000000004200000 --P----UW 0000000004400000: 0000000004400000 --PDA--UW 0000000004600000: 0000000004600000 --P----U- 0000000004800000: 00000000fd000000 --PDA--UW 0000000004a00000: 00000000fd200000 --PDA--UW 0000000004c00000: 00000000fd400000 --PDA--UW And after the commit, `info tlb` gives the following: (qemu) info tlb 0000000000000000: 0000000000000000 --PDA--UW 0000000000200000: 0000000000200000 --PDA--UW 0000000000400000: 0000000000400000 --PDA--UW 0000000000600000: 0000000000600000 --PDA--UW 0000000000800000: 0000000000800000 --PDA--UW 0000000000a00000: 0000000000a00000 --PDA--UW 0000000000c00000: 0000000000c00000 --PDA--UW 0000000000e00000: 0000000000e00000 --PDA--UW 0000000001000000: 0000000001000000 --P----UW 0000000001200000: 0000000001200000 --P----UW 0000000001400000: 0000000001400000 --P----UW 0000000001600000: 0000000001600000 --P----UW 0000000001800000: 0000000001800000 --P----UW 0000000001a00000: 0000000001a00000 --P----UW 0000000001c00000: 0000000001c00000 --P----UW 0000000001e00000: 0000000001e00000 --P----UW 0000000002000000: 0000000002000000 --P----UW 0000000002200000: 0000000002200000 --P----UW 0000000002400000: 0000000002400000 --P----UW 0000000002600000: 0000000002600000 --P----UW 0000000002800000: 0000000002800000 --P----UW 0000000002a00000: 0000000002a00000 --P----UW 0000000002c00000: 0000000002c00000 --P----UW 0000000002e00000: 0000000002e00000 --P----UW 0000000003000000: 0000000003000000 --P----UW 0000000003200000: 0000000003200000 --P----UW 0000000003400000: 0000000003400000 --P----UW 0000000003600000: 0000000003600000 --P----UW 0000000003800000: 0000000003800000 --P----UW 0000000003a00000: 0000000003a00000 --P----UW 0000000003c00000: 0000000003c00000 --P----UW 0000000003e00000: 0000000003e00000 --P----UW 0000000004000000: 0000000004000000 X-P----U- 0000000004200000: 0000000004200000 X-P----UW 0000000004400000: 0000000004400000 X-PDA--UW 0000000004600000: 0000000004600000 X-P----U- 0000000004800000: 00000000fd000000 X-PDA--UW 0000000004a00000: 00000000fd200000 X-PDA--UW 0000000004c00000: 00000000fd400000 X-PDA--UW [1]: 9cf1a8a#r883015557
Refactor of paging_32.zig now allows for virtual addresses of up-to 4GB. Now implementing a state-of-the-art ASCII compatible virtual address scheme for next-gen debugging performance. Specifically map the "bootstrap" page to start the kernel. In time we'll probably zero out this segment and page-out the null page. Move the non-temporary stack position calculation to zig.
Since we are using a very naive virtual paging algorithm which basically maps virtual to physical 1-to-1, we use a _lot_ of memory. A clear improvement would be to track how many pages we actually use instead of calculating the base address from the page index. That way we can reduce the memory footprint to probably less than 1G.
Before, the ctrl-c would cancel the script. This made gdb interrupts impossible.
mewmew
commented
Apr 30, 2024
debug.sh
Outdated
| @@ -22,7 +22,7 @@ kitty --class qemu-starter \ | |||
| -drive media=disk,index=0,file=bin/zipfs.img,format=raw,if=ide \ | |||
| -cdrom bin/kernel.iso & | |||
|
|
|||
| gdb \ | |||
| kitty --class lappis-serial-raw -e gdb \ | |||
There was a problem hiding this comment.
Fet quality of life!
Enable Supervisor Memory Execute Protection (SMEP) and lay down pipes for a future when we can also enable Supervisor Memory Access Protection (SMAP).
- remove nested fish for tailing `serial.raw` - move sleep after qemu to give focus to gdb - give class names similar style (`lappis-` prefix)
We parse and handle the ELF file as long as possible in kernel memory before calling `userland_malloc` and `userland_memcpy`. It quickly gets murky when mixing kernel & userland objects.
Lots of things: - `swapgs` to swap between KernelGSBase and GSBase. Basically, save arbitrary offset in `gs` segment. We use this to point to `tss64` for quickly and easily accessing a scratch space for .rsp0 (kernel) and .rsp2 (userland). - syscall calling conventions are wonky when maintaining C ABI. Since rcx contains the return address of userland, the 4th argument is instead passed in r10. Also, r11 contains the RFLAGS (masked by MSR_FMASK), we use this to turn off interrupts when calling syscalls. That's why we no longer need `cli` and `sti`. - MSR_STAR is wonky. It's upper bits denotes which selectors to update with on `sysret`. But, they use different offsets so it's very confusing: ``` // syscall offsets CS.Selector ← IA32_STAR[47:32] AND FFFCH (* Operating system provides CS; RPL forced to 0 *) SS.Selector ← IA32_STAR[47:32] + 8; // sysret offsets CS.Selector ← IA32_STAR[63:48]+16; SS.Selector ← (IA32_STAR[63:48]+8) OR 3; ``` This makes the `IA32_STAR[63:48]` point to the kernel data segment, because when the offsets resolve they actually point to the userland code segment instead. Why it's offset 8 and 16 instead of 0 and 8 like for CPL 0 is beyond me. - Could probably use some more documentation about the machine specific registers and their magic values. However, there's a lot of refactoring left before syscalls work with SMAP! (Like copying userland buffers into the kernel heap).
Also, changed the immediate value to 1 to _not_ catch the sys_print (since SMAP would crash us).
Basically: ``` ptr = kmalloc(SIZE_OF_USERLAND_OBJ); userland_memcpy(ptr, userland_obj, SIZE_OF_USERLAND_OBJ); ```
Also fix missing register clobbering of `rsi`.
Dear god it actually works!
|
I'll merge this now, huge success! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The main focus of this PR is to make it easier to work with paging.
For this reason, the kernel code and data segments are treated separately by dedicated functions, and each function has a start index (e.g.
P2_KERNEL_STACK_FIRST_INDEX) to make it easier to rearrange the relative order of page mappings in the future.This PR supersedes #19.