boot: convert paging from assembly to Zig #19

mewmew · 2022-05-23T17:45:02Z

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.

src/zasm.zig

karlek · 2022-05-24T19:56:10Z

I'll ask a ton of question over the coming days to learn zig and review your code :P

mewmew · 2022-05-24T20:52:54Z

I'll ask a ton of question over the coming days to learn zig and review your code :P

Gör så! :) Jag lär mig också så kul att se vad som känns skumt och ointuitivt.

src/zasm.zig

karlek · 2022-05-25T08:51:07Z

src/zasm.zig

+    const TargetType = @TypeOf(target);
+    const ReturnType = std.meta.Int(.unsigned, number_of_bits);
+
+    comptime {


Perform extra checks if we can during compile time? Otherwise, skip these checks if we call this function with runtime resolved values?

They are always run at compile time. And, we cannot call this function with runtime resolved values, they must be compile-time known. That is number_of_bits is marked as comptime in the function signature, The TargetType type is known for each call invocation at compile time, etc.

src/zasm.zig

karlek · 2022-05-25T08:55:34Z

src/zasm.zig

+// ~~~ [ zig-x86_64/src/instructions/tlb.zig ] ~~~
+
+/// Structure of a PCID. A PCID has to be <= 4096 for x86_64.
+pub const Pcid = packed struct {


Doesn't this seem overly boilerplate:y / overly verbose? Can't we just have a type synonym like:

PcidType = u12

I agree. This feels like unnecessary boiler plate. It would make sense if they added methods to Pcid.

We can do an alias like this:

const PcidType = u12;

However, I still propose that we keep the code as is, as it makes maintenance easier when we update the code against the x86_64 repo. Make minimal changes unless warranted. And in this case, the "make it slightly cleaner", vs. "simplify maintenance" does not seem worth it.

However, I can be persuaded of course :)

src/zasm.zig

karlek · 2022-05-25T09:07:45Z

src/zasm.zig

+
+    /// Creates a new physical address, throwing bits 52..64 away.
+    pub fn initTruncate(addr: u64) PhysAddr {
+        return PhysAddr{ .value = addr % TRUNCATE_CONST };


Interesting choice to use modulo (or does zig use remainder?) instead of bit mask AND which seems like the more common pattern.

I agree, this is a strange use of modulo. Had I written the code myself, I would also have preferred using a bit mask with AND.

src/zasm.zig

karlek · 2022-05-25T09:20:52Z

src/zasm.zig

+
+fn Msr(comptime register: u32) type {
+    return struct {
+        pub inline fn read() u64 {


Why explicitly make this method inline?

To avoid function prologue and epilogue of essentially single-line assembly functions.

In some cases it may be for performance, but the compiler should be able to figure those out.

However, sometimes using inline is essential for the correct handling of assembly code. For instance, if we wish to read or write the rsp value. Lets say we do this without using inline, then the function prologue and epilogue would trash the rsp value before we have time to read/write to it. So, in these cases, just inline the asm code into the caller function.

src/zasm.zig

karlek · 2022-05-25T09:25:08Z

src/zasm.zig

+    }
+};
+
+fn Msr(comptime register: u32) type {


Is this type private? Or how does exporting work in zig?

The Msr function is private, since it does not have the pub attribute.

As such, all invocations to Msr are done from within this source file. However, their return value (note: in Zig, types are values and can be passed as parameters or returned by functions) may be exported if stored in e.g. a field of an exported struct.

This is the case with the Efer struct, which is exported (i.e. has pub attribute):

/// The Extended Feature Enable Register. pub const Efer = packed struct { ... const REGISTER = Msr(0xC000_0080);

karlek · 2022-05-25T09:32:29Z

src/zasm.zig

+// ~~~ [ zig-x86_64/src/registers/model_specific.zig ] ~~~
+
+/// The Extended Feature Enable Register.
+pub const Efer = packed struct {


Sweet lord, this is nice haha

Packed structs?

karlek · 2022-05-25T11:40:45Z

Makefile

-# -O ReleaseFast
-bin/boot_zig.o: src/boot/boot.zig | bin
-	zig build-obj -target x86_64-freestanding-gnu -static -I./src/kernel -mno-red-zone -femit-bin=$@ $<
+PKGS=--pkg-begin zasm ./src/zasm.zig --pkg-end


Do we call these source files zasm? So they are then imported as @import("zasm")?

Yeah, we give the source file ./src/zasm.zig the package name zasm. Note, the source file could have been called hello_world.zig and we could still assign it to the package name zasm.

This is a getto way to handle packages, until the package manager is implemented in Zig (ziglang/zig#943).

karlek · 2022-05-25T11:40:57Z

Makefile

+		$<
+
+bin/%_elf64_zig.o: bin/%_32_zig.o
+	@objcopy --output-target elf64-x86-64 $< $@


Hack-hackli-hack

Maybe add a todo to remove this when we can solve this in zig?

Sure! You don't like hack-hackiilihack?? ;P

Done dc2d8cb.

karlek · 2022-05-25T11:42:15Z

src/boot/long_mode.asm

+	; elf64 format. The call from multiboot_start to init_long_mode is converted
+	; to call init_long_mode+4 after converting from elf32 to elf64. Thus we
+	; insert four nops.
+	times 4 nop


Hack-hackli-hack 🐓

:huge: 🐘 :in: :room:

Done ed18c7d.

karlek · 2022-05-25T11:42:28Z

src/boot/paging_32.zig

@@ -0,0 +1,31 @@
+const zasm = @import("zasm");


My guess was confirmed!

karlek · 2022-05-25T11:46:06Z

src/boot/paging_32.zig

+
+export fn enable_paging() void {
+    // Load P4 to CR3 register. (CPU uses this to access the P4 table).
+    const raw_p4_addr = @intCast(u64, @ptrToInt(&p4_table[0]));


Why is it &p4_table[0] and not just &p4_table?

The type is automatically assumed to be u64 from the @intCast?

Why is it &p4_table[0] and not just &p4_table?

&p4_table[0] results in a single-item pointer, whereas &p4_table results in a pointer to the array (This would be the same as taking the address of an array (&array has type *[4]int, whereas &array[0] has type *int) or a slice in Go, which gives a pointer to a reflect.SliceHeader struct).

In Zig there is also the notion of many-item pointers. If we have a slice and use the ptr field, we get a many-item pointer which essentially means a pointer to zero or more items of the slice element type. This is essentially what all pointers in C mean :P

Note: a single-item pointer and a multi-item pointer to the first element of a slice both have the same pointer address, they just have different types.

See example below :)

const std = @import("std"); pub fn main() anyerror!void { var array = [4]u8{ 1, 2, 3, 4 }; var slice = array[1..3]; foo(slice); bar(array); } pub fn foo(slice: []u8) void { std.debug.print("=== [ slice ] ===\n", .{}); std.debug.print("ptrs:\n", .{}); std.debug.print(" {any}\n", .{&slice[0]}); std.debug.print(" {any}\n", .{&slice}); std.debug.print(" {any}\n", .{slice.ptr}); std.debug.print("types:\n", .{}); std.debug.print(" {any}\n", .{@TypeOf(&slice[0])}); std.debug.print(" {any}\n", .{@TypeOf(&slice)}); std.debug.print(" {any}\n", .{@TypeOf(slice.ptr)}); std.debug.print("\n", .{}); } pub fn bar(array: [4]u8) void { std.debug.print("=== [ array ] ===\n", .{}); std.debug.print("ptrs:\n", .{}); std.debug.print(" {any}\n", .{&array[0]}); std.debug.print(" {any}\n", .{&array}); std.debug.print("types:\n", .{}); std.debug.print(" {any}\n", .{@TypeOf(&array[0])}); std.debug.print(" {any}\n", .{@TypeOf(&array)}); std.debug.print("\n", .{}); }

Output:

$ zig run bar.zig === [ slice ] === ptrs: u8@7ffc9d803c2b []u8@7ffc9d803c00 u8@7ffc9d803c2b types: *u8 *const []u8 [*]u8 === [ array ] === ptrs: u8@7ffc9d803c2a [4]u8@7ffc9d803c2a types: *const u8 *const [4]u8

The type is automatically assumed to be u64 from the @intcast?

The type is converted to u64 while ensuring that the result type fits all the bits of the input value, otherwise it panics. In our case, this would either be a u64 to u64 conversion or an u32 to u64 conversion, so in both cases the source value is preserved.

ref: https://ziglang.org/documentation/master/#intCast

karlek · 2022-05-25T11:47:07Z

src/boot/paging_32.zig

+export fn enable_paging() void {
+    // Load P4 to CR3 register. (CPU uses this to access the P4 table).
+    const raw_p4_addr = @intCast(u64, @ptrToInt(&p4_table[0]));
+    var p4_addr = zasm.PhysAddr.initUnchecked(raw_p4_addr);


Why not const?

Why not const?

Done d85d018.

karlek · 2022-05-25T11:49:16Z

src/boot/paging_32.zig

+    var p4_addr = zasm.PhysAddr.initUnchecked(raw_p4_addr);
+    var cr3_value = zasm.Cr3.Contents{
+        .phys_frame = zasm.PhysFrame.fromStartAddressUnchecked(p4_addr),
+        .cr3_flags = zasm.Cr3Flags.fromU64(0),
+    };
+    zasm.Cr3.write(cr3_value);


Should we move this closer to the enable paging functionality at the end of the function for closer logical distance?

Done 5b2cf08.

karlek · 2022-05-25T11:50:26Z

src/boot/paging_32.zig

+    // Set the NX bit.
+    efer_value.no_execute_enable = true; // 1 << 11
+    efer_value.write();
+


suggestion: Add assert zasm.Efer.read().long_mode_active == true?

Done c0a85cb.

karlek · 2022-05-25T11:50:53Z

src/boot/paging_32.zig

+
+extern var p4_table: [4096]u8;
+
+export fn enable_paging() void {


Well this was readable hahaah

karlek · 2022-05-25T11:54:14Z

src/zasm.zig

+    /// - `FrameError::FrameNotPresent` if the entry doesn't have the `present` flag set.
+    /// - `FrameError::HugeFrame` if the entry has the `huge_page` flag set (for huge pages the
+    ///    `addr` function must be used)
+    pub fn getFrame(self: PageTableEntry) FrameError!PhysFrame {


Is FrameError the same as error{FrameNotPresent, HugeFrame}, or is it a new type?

FrameError is an alias for the error set error{FrameNotPresent, HugeFrame}, so they are one and the same and can be used interchangeably.

karlek · 2022-05-25T12:54:27Z

src/zasm.zig

+
+    /// Map the entry to the specified physical address
+    pub fn setAddr(self: *PageTableEntry, addr: PhysAddr) void {
+        //std.debug.assert(addr.isAligned(PageSize.Size4KiB.bytes()));


This assert seems reasonable, re-introduce it?

Commented out for now, since it works in 64-bit, but not 32-bit. And we are compiling paging_32.asm for 32-bit.

$ make clean; make run ./src/zasm.zig:734:64: error: expected type 'usize', found 'u64' std.debug.assert(addr.isAligned(PageSize.Size4KiB.bytes())); ^ ./src/zasm.zig:734:64: note: unsigned 32-bit int cannot represent all possible unsigned 64-bit values std.debug.assert(addr.isAligned(PageSize.Size4KiB.bytes())); ^ ./src/zasm.zig:623:38: error: expected type 'usize', found 'u64' return std.mem.isAligned(self.value, alignment); ^ ./src/zasm.zig:623:38: note: unsigned 32-bit int cannot represent all possible unsigned 64-bit values return std.mem.isAligned(self.value, alignment); ^ make: *** [Makefile:18: bin/paging_32_zig.o] Error 1 rm bin/boot_32_zig.o

Added TODO to re-enable assert for setAddr in ab5cc59.

ref: #19 (comment)

mewmew · 2022-05-26T19:44:56Z

src/boot/paging_32.zig

@@ -61,7 +61,10 @@ export fn map_kernel_code_segment() void {
        page_table_entry.setAddr(addr);
        // Set page table entry.
        const p2_index = page_num + p2_index_offset;
-        p2_table[p2_index] = page_table_entry.entry;
+        // TODO: uncomment when Zig 32-bit codegen bug is fixed; use work-around for now.


Looks like this is a bug in ld, not Zig. The ELF32 object file produced by Zig contains the correct assembly. And the ELF64 object file produced by objcopy also contains the correct assembly. But the assembly code produced by ld is trashed, such that the high 32-bits of page_table_entry when stored in p2_table are trashed (overwritten by the low 32-bits). Furthermore, using the LLVM linker ld.lld produces correct assembly. So, we can conclude that there is a bug in ld that we manage to tickle in just the right way.

CORRECT ASM: map_kernel_stack (ELF32 object produced by Zig):

16b: 89 14 c5 04 00 00 00 mov DWORD PTR [eax*8+0x4],edx 172: 89 0c c5 00 00 00 00 mov DWORD PTR [eax*8+0x0],ecx

CORRECT ASM: map_kernel_stack (ELF64 object produced by objcopy):

16b: 89 14 c5 04 00 00 00 mov DWORD PTR [rax*8+0x4],edx 172: 89 0c c5 00 00 00 00 mov DWORD PTR [rax*8+0x0],ecx

INCORRECT ASM: map_kernel_stack (ELF64 executable produced by ld):

10016b: 89 14 c5 00 20 10 00 mov DWORD PTR [rax*8+0x102000],edx ; INCORRECT: should be `mov DWORD PTR [rax*8+0x102004],edx` 100172: 89 0c c5 00 20 10 00 mov DWORD PTR [rax*8+0x102000],ecx

CORRECT ASM: map_kernel_stack (ELF64 executable produced by ld.lld (LLVM's linker)):

10016b: 89 14 c5 04 20 10 00 mov DWORD PTR [rax*8+0x102004],edx 100172: 89 0c c5 00 20 10 00 mov DWORD PTR [rax*8+0x102000],ecx

This code is MIT licensed and copied from the following repositories: * https://github.com/leecannon/zig-x86_64 * https://github.com/leecannon/zig-bitjuggle

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.

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.

ref: #19 (comment)

… (Cr0) This ties related logic together.

Turns out there is a codegen bug in Zig 0.9.1 when generating 32-bit code that operate on 64-bit arrays. In particular, `foo[index] = bar`, where `foo` is an `[512]u64` array, causes the high 32-bit of `bar` to be discarded. The assembly code actually attempts to set the high 32-bit of bar, but uses the incorrect offset, and thus overwrites the value with the low 32-bit part of bar. This issue was identified as the NX bit was _not_ set in the p2_table, even though the Zig code contained a `page_table_entry.no_execute = true` statement. Gotta love compiler bugs that makes mapped pages w+x :) Tribute to Ken on Trusting Trust <3 Prior to this fix, `info tlb` gave: (qemu) info tlb // kernel code 0000000000000000: 0000000000000000 --PDA--UW // kernel data: 0000000000200000: 0000000000200000 --PDA--UW ... 0000000003e00000: 0000000003e00000 --P----UW // kernel stack (guard) 0000000004000000: 0000000004000000 --P----U- // kernel stack 0000000004200000: 0000000004200000 --P----UW 0000000004400000: 0000000004400000 --PDA--UW // kernel stack (guard) 0000000004600000: 0000000004600000 --P----U- // frame buffer: 0000000004800000: 00000000fd000000 --PDA--UW 0000000004a00000: 00000000fd200000 --PDA--UW 0000000004c00000: 00000000fd400000 --PDA--UW After this commit, `info tlb` gives: (qemu) info tlb // kernel code 0000000000000000: 0000000000000000 --PDA--UW // kernel data: 0000000000200000: 0000000000200000 X-PDA--UW ... 0000000003e00000: 0000000003e00000 X-P----UW // kernel stack (guard) 0000000004000000: 0000000004000000 X-P----U- // kernel stack 0000000004200000: 0000000004200000 X-P----UW 0000000004400000: 0000000004400000 X-PDA--UW // kernel stack (guard) 0000000004600000: 0000000004600000 X-P----U- // frame buffer: 0000000004800000: 00000000fd000000 X-PDA--UW 0000000004a00000: 00000000fd200000 X-PDA--UW 0000000004c00000: 00000000fd400000 X-PDA--UW For the morbidly curious, this is the incorrect assembly as generated by Zig 0.9.1 32-bit codegen: 0x1006a5 <map_kernel_stack+341> mov ecx, DWORD PTR [rbp-0x8] ; 0x4200087 0x1006a8 <map_kernel_stack+344> mov edx, DWORD PTR [rbp-0x4] ; 0x80000000 0x1006ab <map_kernel_stack+347> mov DWORD PTR [rax*8+0x112000], edx ; BUG! writes high 32-bit part of page_table_entry (including NX bit) to incorrect offset. 0x1006b2 <map_kernel_stack+354> mov DWORD PTR [rax*8+0x112000], ecx ; overwrites previous write with low 32-bit part of page_table_entry (overwrites NX bit).

… Zig

…VM linker) ref: 1f0b42f#diff-9c89cba7a3cf9ef6f6ca79d1b4f36385748dc642e307d24cf6874445e258cca2

ref: b91bf47

In Zig 0.11.0, the package management system was rewritten, and as such the support for --begin-pkg was removed. To simplify the story of package management now, simply include zasm.zig as a file (rather than placing it in a separate Git repository). Since Zig doesn't allow local import of files outside of the root directory of the Zig file that is using the import, move zasm.zig to the boot/ directory.

Run equivalent of zigfmt to automate the update.

mewmew · 2024-03-22T01:58:51Z

Nu är less-asm branchen re-base'ad på main och bygger med Zig 0.11.0.

Fick lägga till ett till "hack" för att få compilen att funka, då vi kör statisk länkning så var det en runtime funktion till Zig som inte följde med när vi länka vår kernel:

lappis/Makefile

Line 15 in 839dcb9

# NOTE: use `-fcompiler-rt` to include __zig_probe_stack symbol when linking

För tillfället så bygger den som den ska och kan köra för att visa doggie : )

ref: #19 (comment)

mewmew · 2024-04-15T23:37:52Z

This PR has been superseded by #23.

ref: #19 (comment)

mewmew force-pushed the less-asm branch from 6f5bcbc to ddbef75 Compare May 23, 2022 17:46

karlek reviewed May 23, 2022

View reviewed changes

src/zasm.zig Outdated Show resolved Hide resolved

karlek reviewed May 24, 2022

View reviewed changes

karlek reviewed May 25, 2022

View reviewed changes

src/zasm.zig Outdated Show resolved Hide resolved

karlek reviewed May 25, 2022

View reviewed changes

mewmew added a commit that referenced this pull request May 25, 2022

makefile: add TODO to remove objcopy hack

dc2d8cb

ref: #19 (comment)

mewmew added a commit that referenced this pull request May 25, 2022

boot: add TODO to remove 4xNOP hack in init_long_mode

ed18c7d

ref: #19 (comment)

mewmew added a commit that referenced this pull request May 25, 2022

boot/paging: use const instead of var where applicable

d85d018

ref: #19 (comment)

mewmew commented May 26, 2022

View reviewed changes

mewmew force-pushed the less-asm branch 2 times, most recently from cdf0142 to d229b12 Compare May 26, 2022 20:33

mewmew added 13 commits March 22, 2024 01:52

zasm: add util functions for control and model specific registers

015591c

This code is MIT licensed and copied from the following repositories: * https://github.com/leecannon/zig-x86_64 * https://github.com/leecannon/zig-bitjuggle

boot: rewrite assembly version of enable_paging in Zig

e9f7d3a

boot: remove unused print32.asm

347c696

This was needed for debugging before the jump to long mode was fully functional.

boot: remove unused protected_mode.asm

d95e853

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.

boot: remove unused print.asm

e326c98

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.

zasm: add PageTableEntry and PageTableFlags from x86_64 repo

348030c

boot: ensure that no MMX instructions are used in 32-bit boot code

7ee811a

boot: rewrite assembly version of map_frame_buffer in Zig

eddf808

boot: rewrite assembly version of map_kernel_stack in Zig

467fedc

boot: use dedicated functions to map kernel code and data segment pages

5918f24

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.

boot: rewrite assembly version of set_up_page_tables in Zig

35654ea

makefile: add TODO to remove objcopy hack

2b20130

ref: #19 (comment)

mewmew added 11 commits March 22, 2024 01:56

boot: add TODO to remove 4xNOP hack in init_long_mode

2e9446f

ref: #19 (comment)

boot/paging: use const instead of var where applicable

b5a2839

ref: #19 (comment)

boot/paging: move load of P4 table (Cr3) closer to enable paging code…

11fb174

… (Cr0) This ties related logic together.

boot/paging: assert that long mode active is set after enable_paging

fa6d76c

zasm: add TODO to re-enable assert in setAddr

bfedda3

boot/paging: update TODO for 32-bit codegen bug, the bug is in ld not…

98aae7f

… Zig

boot/paging: revert work-around for bug in ld; use ld.lld instead (LL…

8b000e5

…VM linker) ref: 1f0b42f#diff-9c89cba7a3cf9ef6f6ca79d1b4f36385748dc642e307d24cf6874445e258cca2

notes: add logbook entry for ld bug

0e3a862

ref: b91bf47

ci: add lld dependency for ld.lld linker

3d6346f

mewmew force-pushed the less-asm branch from d229b12 to 839dcb9 Compare March 22, 2024 01:55

mewmew added 2 commits March 22, 2024 02:55

boot: update Zig syntax to version 0.11.0

d57450c

Run equivalent of zigfmt to automate the update.

make: fix build for Zig version 0.11.0

839dcb9

mewmew requested a review from karlek March 22, 2024 01:59

mewmew added a commit that referenced this pull request Apr 15, 2024

makefile: add TODO to remove objcopy hack

33dc2aa

ref: #19 (comment)

mewmew added a commit that referenced this pull request Apr 15, 2024

boot: add TODO to remove 4xNOP hack in init_long_mode

e9335d3

ref: #19 (comment)

mewmew added a commit that referenced this pull request Apr 15, 2024

boot/paging: use const instead of var where applicable

b26ddb8

ref: #19 (comment)

mewmew mentioned this pull request Apr 15, 2024

boot: convert paging from assembly to Zig (take 2) #23

Merged

mewmew closed this Apr 15, 2024

karlek pushed a commit that referenced this pull request Apr 29, 2024

makefile: add TODO to remove objcopy hack

da51fe0

ref: #19 (comment)

karlek pushed a commit that referenced this pull request Apr 29, 2024

boot: add TODO to remove 4xNOP hack in init_long_mode

1c8db56

ref: #19 (comment)

karlek pushed a commit that referenced this pull request Apr 29, 2024

boot/paging: use const instead of var where applicable

2ccf788

ref: #19 (comment)

karlek pushed a commit that referenced this pull request May 3, 2024

makefile: add TODO to remove objcopy hack

f13bb0a

ref: #19 (comment)

karlek pushed a commit that referenced this pull request May 3, 2024

boot: add TODO to remove 4xNOP hack in init_long_mode

35d34ee

ref: #19 (comment)

karlek pushed a commit that referenced this pull request May 3, 2024

boot/paging: use const instead of var where applicable

f0155eb

ref: #19 (comment)

This pull request was closed.


		extern var p4_table: [4096]u8;

		export fn enable_paging() void {

boot: convert paging from assembly to Zig #19

boot: convert paging from assembly to Zig #19

Conversation

mewmew commented May 23, 2022

karlek commented May 24, 2022

mewmew commented May 24, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mewmew May 25, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mewmew May 25, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mewmew May 25, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mewmew May 25, 2022 • edited Loading

Choose a reason for hiding this comment

mewmew May 26, 2022 • edited Loading

Choose a reason for hiding this comment

mewmew commented Mar 22, 2024

mewmew commented Apr 15, 2024

mewmew commented May 24, 2022 •

edited

Loading

mewmew May 25, 2022 •

edited

Loading

mewmew May 25, 2022 •

edited

Loading

mewmew May 25, 2022 •

edited

Loading

mewmew May 25, 2022 •

edited

Loading

mewmew May 26, 2022 •

edited

Loading