pluto/src/kernel/arch/x86/paging.zig

const std = @import("std");
const expectEqual = std.testing.expectEqual;
const expect = std.testing.expect;
const builtin = @import("builtin");
const panic = @import("../../panic.zig");
const arch = @import("arch.zig");
const isr = @import("isr.zig");
const MemProfile = @import("../../mem.zig").MemProfile;
const tty = @import("../../tty.zig");

/// An array of directory entries and page tables. Forms the first level of paging and covers the entire 4GB memory space.
const Directory = packed struct {
    /// The directory entries.
    entries: [ENTRIES_PER_DIRECTORY]DirectoryEntry,

    /// The tables allocated for the directory. This is ignored by the CPU.
    tables: [ENTRIES_PER_DIRECTORY]?*Table,
};

/// An array of table entries. Forms the second level of paging and covers a 4MB memory space.
const Table = packed struct {
    /// The table entries.
    entries: [ENTRIES_PER_TABLE]TableEntry,
};

/// All errors that can be thrown by paging functions.
const PagingError = error{
    /// Physical addresses are invalid (definition is up to the function).
    InvalidPhysAddresses,

    /// Virtual addresses are invalid (definition is up to the function).
    InvalidVirtAddresses,

    /// Physical and virtual addresses don't cover spaces of the same size.
    PhysicalVirtualMismatch,

    /// Physical addressses aren't aligned by page size.
    UnalignedPhysAddresses,

    /// Virtual addressses aren't aligned by page size.
    UnalignedVirtAddresses,
};

/// An entry within a directory. References a single page table.
/// Bit 0: Present. Set if present in physical memory.
///        When not set, all remaining 31 bits are ignored and available for use.
/// Bit 1: Writable. Set if writable.
/// Bit 2: User. Set if accessible by user mode.
/// Bit 3: Write through. Set if write-through caching is enabled.
/// Bit 4: Cache disabled. Set if caching is disabled for this table.
/// Bit 5: Accessed. Set by the CPU when the table is accessed. Not cleared by CPU.
/// Bit 6: Zero.
/// Bit 7: Page size. Set if this entry covers a single 4MB page rather than 1024 4KB pages.
/// Bit 8: Ignored.
/// Bits 9-11: Ignored and available for use by kernel.
/// Bits 12-31: The 4KB aligned physical address of the corresponding page table.
///             Must be 4MB aligned if the page size bit is set.
const DirectoryEntry = u32;

/// An entry within a page table. References a single page.
/// Bit 0: Present. Set if present in physical memory.
///        When not set, all remaining 31 bits are ignored and available for use.
/// Bit 1: Writable. Set if writable.
/// Bit 2: User. Set if accessible by user mode.
/// Bit 3: Write through. Set if write-through caching is enabled.
/// Bit 4: Cache disabled. Set if caching is disabled for this page.
/// Bit 5: Accessed. Set by the CPU when the page is accessed. Not cleared by CPU.
/// Bit 6: Dirty. Set by the CPU when the page has been written to. Not cleared by the CPU.
/// Bit 7: Zero.
/// Bit 8: Global. Set if the cached address for this page shouldn't be updated when cr3 is changed.
/// Bits 9-11: Ignored and available for use by the kernel.
/// Bits 12-31: The 4KB aligned physical address mapped to this page.
const TableEntry = u32;

/// Each directory has 1024 entries
const ENTRIES_PER_DIRECTORY: u32 = 1024;

/// Each table has 1024 entries
const ENTRIES_PER_TABLE: u32 = 1024;

/// The number of bytes in 4MB
const PAGE_SIZE_4MB: u32 = 0x400000;

/// The number of bytes in 4KB
const PAGE_SIZE_4KB: u32 = PAGE_SIZE_4MB / 1024;

/// There are 1024 entries per directory with each one covering 4KB
const PAGES_PER_DIR_ENTRY: u32 = 1024;

/// There are 1 million pages per directory
const PAGES_PER_DIR: u32 = ENTRIES_PER_DIRECTORY * PAGES_PER_DIR_ENTRY;

/// The bitmasks for the bits in a DirectoryEntry
const DENTRY_PRESENT: u32 = 0x1;
const DENTRY_WRITABLE: u32 = 0x2;
const DENTRY_USER: u32 = 0x4;
const DENTRY_WRITE_THROUGH: u32 = 0x8;
const DENTRY_CACHE_DISABLED: u32 = 0x10;
const DENTRY_ACCESSED: u32 = 0x20;
const DENTRY_ZERO: u32 = 0x40;
const DENTRY_4MB_PAGES: u32 = 0x80;
const DENTRY_IGNORED: u32 = 0x100;
const DENTRY_AVAILABLE: u32 = 0xE00;
const DENTRY_PAGE_ADDR: u32 = 0xFFFFF000;

/// The bitmasks for the bits in a TableEntry
const TENTRY_PRESENT: u32 = 0x1;
const TENTRY_WRITABLE: u32 = 0x2;
const TENTRY_USER: u32 = 0x4;
const TENTRY_WRITE_THROUGH: u32 = 0x8;
const TENTRY_CACHE_DISABLED: u32 = 0x10;
const TENTRY_ACCESSED: u32 = 0x20;
const TENTRY_DIRTY: u32 = 0x40;
const TENTRY_ZERO: u32 = 0x80;
const TENTRY_GLOBAL: u32 = 0x100;
const TENTRY_AVAILABLE: u32 = 0xE00;
const TENTRY_PAGE_ADDR: u32 = 0xFFFFF000;

/// The kernel's virtual address offset. It's assigned in the init function and the virtToPhys test.
/// We can't just use KERNEL_ADDR_OFFSET since using externs in the virtToPhys test is broken in
/// release-safe. This is a workaround until that is fixed.
var ADDR_OFFSET: usize = undefined;
extern var KERNEL_ADDR_OFFSET: *u32;

///
/// Convert a virtual address to its physical counterpart by subtracting the kernel virtual offset from the virtual address.
///
/// Arguments:
///     IN virt: var - The virtual address to covert. Either an integer or pointer.
///
/// Return: @typeOf(virt)
///     The physical address.
///
inline fn virtToPhys(virt: var) @typeOf(virt) {
    const offset = ADDR_OFFSET;
    const T = @typeOf(virt);
    return switch (@typeId(T)) {
        .Pointer => @intToPtr(T, @ptrToInt(virt) - offset),
        .Int => virt - offset,
        else => @compileError("Only pointers and integers are supported"),
    };
}

///
/// Convert a virtual address to an index within an array of directory entries.
///
/// Arguments:
///     IN virt: usize - The virtual address to convert.
///
/// Return: usize
///     The index into an array of directory entries.
///
inline fn virtToDirEntryIdx(virt: usize) usize {
    return (virt / PAGE_SIZE_4MB) % ENTRIES_PER_DIRECTORY;
}

///
/// Convert a virtual address to an index within an array of table entries.
///
/// Arguments:
///     IN virt: usize - The virtual address to convert.
///
/// Return: usize
///     The index into an array of table entries.
///
inline fn virtToTableEntryIdx(virt: usize) usize {
    return (virt / PAGE_SIZE_4KB) % ENTRIES_PER_TABLE;
}

///
/// Map a page directory entry, setting the present, size, writable, write-through and physical address bits.
/// Clears the user and cache disabled bits. Entry should be zero'ed.
///
/// Arguments:
///     OUT dir: *Directory - The directory that this entry is in
///     IN virt_addr: usize - The start of the virtual space to map
///     IN virt_end: usize - The end of the virtual space to map
///     IN phys_addr: usize - The start of the physical space to map
///     IN phys_end: usize - The end of the physical space to map
///     IN allocator: *Allocator - The allocator to use to map any tables needed
///
/// Error: PagingError || std.mem.Allocator.Error
///     PagingError.InvalidPhysAddresses - The physical start address is greater than the end.
///     PagingError.InvalidVirtAddresses - The virtual start address is greater than the end or is larger than 4GB.
///     PagingError.PhysicalVirtualMismatch - The differences between the virtual addresses and the physical addresses aren't the same.
///     PagingError.UnalignedPhysAddresses - One or both of the physical addresses aren't page size aligned.
///     PagingError.UnalignedVirtAddresses - One or both of the virtual addresses aren't page size aligned.
///     std.mem.Allocator.Error.* - See std.mem.Allocator.alignedAlloc.
///
fn mapDirEntry(dir: *Directory, virt_start: usize, virt_end: usize, phys_start: usize, phys_end: usize, allocator: *std.mem.Allocator) (PagingError || std.mem.Allocator.Error)!void {
    if (phys_start > phys_end) {
        return PagingError.InvalidPhysAddresses;
    }
    if (virt_start > virt_end) {
        return PagingError.InvalidVirtAddresses;
    }
    if (phys_end - phys_start != virt_end - virt_start) {
        return PagingError.PhysicalVirtualMismatch;
    }
    if (!std.mem.isAligned(phys_start, PAGE_SIZE_4KB) or !std.mem.isAligned(phys_end, PAGE_SIZE_4KB)) {
        return PagingError.UnalignedPhysAddresses;
    }
    if (!std.mem.isAligned(virt_start, PAGE_SIZE_4KB) or !std.mem.isAligned(virt_end, PAGE_SIZE_4KB)) {
        return PagingError.UnalignedVirtAddresses;
    }

    const entry = virt_start / PAGE_SIZE_4MB;
    if (entry >= ENTRIES_PER_DIRECTORY)
        return PagingError.InvalidVirtAddresses;
    var dir_entry = &dir.entries[entry];
    dir_entry.* |= DENTRY_PRESENT;
    dir_entry.* |= DENTRY_WRITABLE;
    dir_entry.* &= ~u32(DENTRY_USER);
    dir_entry.* |= DENTRY_WRITE_THROUGH;
    dir_entry.* &= ~u32(DENTRY_CACHE_DISABLED);
    dir_entry.* &= ~u32(DENTRY_4MB_PAGES);

    // Only create a new table if one hasn't already been created for this dir entry.
    // Prevents us from overriding previous mappings.
    var table: *Table = undefined;
    if (dir.tables[entry]) |tbl| {
        table = tbl;
    } else {
        // Create a table and put the physical address in the dir entry
        table = &(try allocator.alignedAlloc(Table, @truncate(u29, PAGE_SIZE_4KB), 1))[0];
        @memset(@ptrCast([*]u8, table), 0, @sizeOf(Table));
        const table_phys_addr = @ptrToInt(virtToPhys(table));
        dir_entry.* |= @intCast(u32, DENTRY_PAGE_ADDR & table_phys_addr);
        dir.tables[entry] = table;
    }

    // Map the table entries within the requested space
    var virt = virt_start;
    var phys = phys_start;
    var tentry = virtToTableEntryIdx(virt);
    while (virt < virt_end) : ({
        virt += PAGE_SIZE_4KB;
        phys += PAGE_SIZE_4KB;
        tentry += 1;
    }) {
        try mapTableEntry(&table.entries[tentry], phys);
    }
}

///
/// Map a table entry by setting its bits to the appropriate values.
/// Sets the entry to be present, writable, kernel access, write through, cache enabled, non-global and the page address bits.
///
/// Arguments:
///     OUT entry: *align(1) TableEntry - The entry to map. 1 byte aligned.
///     IN phys_addr: usize - The physical address to map the table entry to.
///
/// Error: PagingError
///     PagingError.UnalignedPhysAddresses - If the physical address isn't page size aligned.
///
fn mapTableEntry(entry: *align(1) TableEntry, phys_addr: usize) PagingError!void {
    if (!std.mem.isAligned(phys_addr, PAGE_SIZE_4KB)) {
        return PagingError.UnalignedPhysAddresses;
    }
    entry.* |= TENTRY_PRESENT;
    entry.* |= TENTRY_WRITABLE;
    entry.* &= ~u32(TENTRY_USER);
    entry.* |= TENTRY_WRITE_THROUGH;
    entry.* &= ~u32(TENTRY_CACHE_DISABLED);
    entry.* &= ~u32(TENTRY_GLOBAL);
    entry.* |= TENTRY_PAGE_ADDR & @intCast(u32, phys_addr);
}

///
/// Map a page directory. The addresses passed must be page size aligned and be the same distance apart.
///
/// Arguments:
///     OUT entry: *Directory - The directory to map
///     IN phys_start: usize - The physical address at which to start mapping
///     IN phys_end: usize - The physical address at which to stop mapping
///     IN virt_start: usize - The virtual address at which to start mapping
///     IN virt_end: usize - The virtual address at which to stop mapping
///     IN allocator: *Allocator - The allocator to use to map any tables needed
///
/// Error: std.mem.allocator.Error || PagingError
///     * - See mapDirEntry.
///
fn mapDir(dir: *Directory, phys_start: usize, phys_end: usize, virt_start: usize, virt_end: usize, allocator: *std.mem.Allocator) (std.mem.Allocator.Error || PagingError)!void {
    var virt_addr = virt_start;
    var phys_addr = phys_start;
    var page = virt_addr / PAGE_SIZE_4KB;
    var entry_idx = virt_addr / PAGE_SIZE_4MB;
    while (entry_idx < ENTRIES_PER_DIRECTORY and virt_addr < virt_end) : ({
        phys_addr += PAGE_SIZE_4MB;
        virt_addr += PAGE_SIZE_4MB;
        entry_idx += 1;
    }) {
        try mapDirEntry(dir, virt_addr, std.math.min(virt_end, virt_addr + PAGE_SIZE_4MB), phys_addr, std.math.min(phys_end, phys_addr + PAGE_SIZE_4MB), allocator);
    }
}

///
/// Called when a page fault occurs.
///
/// Arguments:
///     IN state: *arch.InterruptContext - The CPU's state when the fault occured.
///
fn pageFault(state: *arch.InterruptContext) void {
    @panic("Page fault");
}

///
/// Initialise x86 paging, overwriting any previous paging set up.
///
/// Arguments:
///     IN mem_profile: *const MemProfile - The memory profile of the system and kernel
///     IN allocator: *std.mem.Allocator - The allocator to use
///
pub fn init(mem_profile: *const MemProfile, allocator: *std.mem.Allocator) void {
    ADDR_OFFSET = @ptrToInt(&KERNEL_ADDR_OFFSET);
    // Calculate start and end of mapping
    const v_start = std.mem.alignBackward(@ptrToInt(mem_profile.vaddr_start), PAGE_SIZE_4KB);
    const v_end = std.mem.alignForward(@ptrToInt(mem_profile.vaddr_end) + mem_profile.fixed_alloc_size, PAGE_SIZE_4KB);
    const p_start = std.mem.alignBackward(@ptrToInt(mem_profile.physaddr_start), PAGE_SIZE_4KB);
    const p_end = std.mem.alignForward(@ptrToInt(mem_profile.physaddr_end) + mem_profile.fixed_alloc_size, PAGE_SIZE_4KB);

    var tmp = allocator.alignedAlloc(Directory, @truncate(u29, PAGE_SIZE_4KB), 1) catch panic.panicFmt(@errorReturnTrace(), "Failed to allocate page directory");
    var kernel_directory = @ptrCast(*Directory, tmp.ptr);
    @memset(@ptrCast([*]u8, kernel_directory), 0, @sizeOf(Directory));

    // Map in kernel
    mapDir(kernel_directory, p_start, p_end, v_start, v_end, allocator) catch panic.panicFmt(@errorReturnTrace(), "Failed to map kernel directory");
    const tty_addr = tty.getVideoBufferAddress();
    // If the previous mappping space didn't cover the tty buffer, do so now
    if (v_start > tty_addr or v_end <= tty_addr) {
        const tty_phys = virtToPhys(tty_addr);
        const tty_buff_size = 32 * 1024;
        mapDir(kernel_directory, tty_phys, tty_phys + tty_buff_size, tty_addr, tty_addr + tty_buff_size, allocator) catch panic.panicFmt(@errorReturnTrace(), "Failed to map vga buffer in kernel directory");
    }

    const dir_physaddr = @ptrToInt(virtToPhys(kernel_directory));
    asm volatile ("mov %[addr], %%cr3"
        :
        : [addr] "{eax}" (dir_physaddr)
    );
    isr.registerIsr(14, pageFault) catch panic.panicFmt(@errorReturnTrace(), "Failed to register page fault ISR");
}

fn checkDirEntry(entry: DirectoryEntry, virt_start: usize, virt_end: usize, phys_start: usize, table: *Table) void {
    expect(entry & DENTRY_PRESENT != 0);
    expect(entry & DENTRY_WRITABLE != 0);
    expectEqual(entry & DENTRY_USER, 0);
    expect(entry & DENTRY_WRITE_THROUGH != 0);
    expectEqual(entry & DENTRY_CACHE_DISABLED, 0);
    expectEqual(entry & DENTRY_4MB_PAGES, 0);
    expectEqual(entry & DENTRY_ZERO, 0);

    var tentry_idx = virtToTableEntryIdx(virt_start);
    var tentry_idx_end = virtToTableEntryIdx(virt_end);
    var phys = phys_start;
    while (tentry_idx < tentry_idx_end) : ({
        tentry_idx += 1;
        phys += PAGE_SIZE_4KB;
    }) {
        const tentry = table.entries[tentry_idx];
        checkTableEntry(tentry, phys);
    }
}

fn checkTableEntry(entry: TableEntry, page_phys: usize) void {
    expect(entry & TENTRY_PRESENT != 0);
    expect(entry & TENTRY_WRITABLE != 0);
    expectEqual(entry & TENTRY_USER, 0);
    expect(entry & TENTRY_WRITE_THROUGH != 0);
    expectEqual(entry & TENTRY_CACHE_DISABLED, 0);
    expectEqual(entry & TENTRY_ZERO, 0);
    expectEqual(entry & TENTRY_GLOBAL, 0);
    expectEqual(entry & TENTRY_PAGE_ADDR, @intCast(u32, page_phys));
}

test "virtToPhys" {
    ADDR_OFFSET = 0xC0000000;
    const offset: usize = ADDR_OFFSET;
    expectEqual(virtToPhys(offset + 0), 0);
    expectEqual(virtToPhys(offset + 123), 123);
    expectEqual(virtToPhys(@intToPtr(*usize, offset + 123)), @intToPtr(*usize, 123));
}

test "virtToDirEntryIdx" {
    expectEqual(virtToDirEntryIdx(0), 0);
    expectEqual(virtToDirEntryIdx(123), 0);
    expectEqual(virtToDirEntryIdx(PAGE_SIZE_4MB - 1), 0);
    expectEqual(virtToDirEntryIdx(PAGE_SIZE_4MB), 1);
    expectEqual(virtToDirEntryIdx(PAGE_SIZE_4MB + 1), 1);
    expectEqual(virtToDirEntryIdx(PAGE_SIZE_4MB * 2), 2);
    expectEqual(virtToDirEntryIdx(PAGE_SIZE_4MB * (ENTRIES_PER_DIRECTORY - 1)), ENTRIES_PER_DIRECTORY - 1);
}

test "virtToTableEntryIdx" {
    expectEqual(virtToTableEntryIdx(0), 0);
    expectEqual(virtToTableEntryIdx(123), 0);
    expectEqual(virtToTableEntryIdx(PAGE_SIZE_4KB - 1), 0);
    expectEqual(virtToTableEntryIdx(PAGE_SIZE_4KB), 1);
    expectEqual(virtToTableEntryIdx(PAGE_SIZE_4KB + 1), 1);
    expectEqual(virtToTableEntryIdx(PAGE_SIZE_4KB * 2), 2);
    expectEqual(virtToTableEntryIdx(PAGE_SIZE_4KB * (ENTRIES_PER_TABLE - 1)), ENTRIES_PER_TABLE - 1);
    expectEqual(virtToTableEntryIdx(PAGE_SIZE_4KB * (ENTRIES_PER_TABLE)), 0);
}

test "mapDirEntry" {
    var allocator = std.heap.direct_allocator;
    var dir: Directory = Directory{ .entries = [_]DirectoryEntry{0} ** ENTRIES_PER_DIRECTORY, .tables = [_]?*Table{null} ** ENTRIES_PER_DIRECTORY };
    var phys: usize = 0 * PAGE_SIZE_4MB;
    const phys_end: usize = phys + PAGE_SIZE_4MB;
    const virt: usize = 1 * PAGE_SIZE_4MB;
    const virt_end: usize = virt + PAGE_SIZE_4MB;
    try mapDirEntry(&dir, virt, virt_end, phys, phys_end, allocator);

    const entry_idx = virtToDirEntryIdx(virt);
    const entry = dir.entries[entry_idx];
    const table = dir.tables[entry_idx] orelse unreachable;
    checkDirEntry(entry, virt, virt_end, phys, table);
}

test "mapDir" {
    var allocator = std.heap.direct_allocator;
    var dir = Directory{ .entries = [_]DirectoryEntry{0} ** ENTRIES_PER_DIRECTORY, .tables = [_]?*Table{null} ** ENTRIES_PER_DIRECTORY };
    const phys_start: usize = PAGE_SIZE_4MB * 2;
    const virt_start: usize = PAGE_SIZE_4MB * 4;
    const phys_end: usize = PAGE_SIZE_4MB * 4;
    const virt_end: usize = PAGE_SIZE_4MB * 6;
    mapDir(&dir, phys_start, phys_end, virt_start, virt_end, allocator) catch unreachable;

    var virt = virt_start;
    var phys = phys_start;
    while (virt < virt_end) : ({
        virt += PAGE_SIZE_4MB;
        phys += PAGE_SIZE_4MB;
    }) {
        const entry_idx = virtToDirEntryIdx(virt);
        const entry = dir.entries[entry_idx];
        const table = dir.tables[entry_idx] orelse unreachable;
        checkDirEntry(entry, virt, virt + PAGE_SIZE_4MB, phys, table);
    }
}