sysutils: add dummy chroot program

.cargo/config.toml

@@ -1,2 +1,2 @@
 [alias]
-xtask = "run --manifest-path ./xtask/Cargo.toml --"
+xtask = "run --manifest-path ./xtask/Cargo.toml --release --"

.gitignore

@@ -1,3 +1,8 @@
 /target
 /toolchain
 /xtask.toml
+/qemu.toml
+/etc/boot/yboot.cfg
+/disk-*.img
+/tmp-*.txt
+/*.log

Cargo.toml

@@ -6,17 +6,18 @@ exclude = [
     "tool/abi-generator",
     "toolchain",
     "userspace/dynload-program",
-    "userspace/lib/ygglibc"
+    "userspace/lib/ygglibc",
+    "toolchain-c"
 ]
 members = [
     "xtask",
-    "kernel/tools/gentables",
     "kernel",
     "lib/abi",
     "lib/libyalloc",
     "lib/runtime",
     "lib/qemu",
-    "lib/unwind"
+    "lib/abi-serde",
+    "lib/libutil"
 ]
 
 [workspace.dependencies]
@@ -38,6 +39,7 @@ ahash = { version = "0.8.11", default-features = false, features = ["no-rng"] }
 
 # acpi
 acpi = { git = "https://github.com/alnyan/acpi.git", package = "acpi", branch = "acpi-system" }
+rsdp = { git = "https://github.com/alnyan/acpi.git", package = "rsdp", branch = "acpi-system" }
 aml = { git = "https://github.com/alnyan/acpi.git", branch = "acpi-system" }
 acpi-system = { git = "https://github.com/alnyan/acpi-system.git" }
 
@@ -46,12 +48,14 @@ yboot-proto.path = "boot/yboot-proto"
 
 # Local libs
 abi-lib.path = "lib/abi-lib"
+abi-serde.path = "lib/abi-serde"
 yggdrasil-abi.path = "lib/abi"
 abi-generator.path = "tool/abi-generator"
 
 # Kernel parts
 kernel-arch-interface.path = "kernel/arch/interface"
 kernel-arch-aarch64.path = "kernel/arch/aarch64"
+kernel-arch-riscv64.path = "kernel/arch/riscv64"
 kernel-arch-x86_64.path = "kernel/arch/x86_64"
 kernel-arch-i686.path = "kernel/arch/i686"
 kernel-arch-x86.path = "kernel/arch/x86"
@@ -62,7 +66,6 @@ libk-util.path = "kernel/libk/libk-util"
 libk-mm.path = "kernel/libk/libk-mm"
 libk-mm-interface.path = "kernel/libk/libk-mm/interface"
 libk-device.path = "kernel/libk/libk-device"
-memtables.path = "kernel/lib/memtables"
 vmalloc.path = "kernel/lib/vmalloc"
 device-api-macros.path = "kernel/lib/device-api/macros"
 device-tree.path = "kernel/lib/device-tree"
@@ -87,3 +90,15 @@ features = ["no_std_stream"]
 unexpected_cfgs = { level = "allow", check-cfg = ['cfg(rust_analyzer)'] }
 [workspace.lints.clippy]
 derivable_impls = { level = "allow" }
+
+[profile.dev]
+opt-level = 1
+# split-debuginfo = "packed"
+lto = "thin"
+panic = "abort"
+
+[profile.test]
+split-debuginfo = "none"
+
+[profile.dev.package."*"]
+opt-level = 3

LICENSE.txt

@@ -0,0 +1,9 @@
+MIT License
+
+Copyright (c) 2025 Mark Poliakov <mark@alnyan.me>
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the " Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next paragraph) shall be included in all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

README.md

@@ -8,22 +8,38 @@ Rust Unix-like operating system.
 Main features
 -------------
 
-* Architecture support: [aarch64](/kernel/src/arch/aarch64) and [x86_64](/kernel/src/arch/x86_64)
-* Kernel/userspace preemptive multithreading
-* Kernel-space multitasking with `async`/`await` runtime
-* Symmetric Multiprocessing
-* Unix-like virtual filesystem:
+* Architecture support:
+	* [aarch64](kernel/src/arch/aarch64)
+	* [x86_64](kernel/src/arch/x86_64)
+	* [i686](kernel/src/arch/i686) (Pentium Pro and later)
+* Core features:
+	* Kernel/userspace preemptive multithreading
+	* Kernel-space multitasking with `async`/`await` runtime
+	* Symmetric Multiprocessing
+	* Rust-style interfaces for most of the stuff like memory management, devices etc.
+* Filesystem features:
+	* Unix-like virtual filesystem:
 	files, directories, block/char devices, symlinks, mounts
-* [Kernel-user ABI](/lib/abi-def/yggdrasil.abi) generated from a rust-like description language
-* In-memory read-write filesystem for tar-based initrd
-* sysfs/devfs
-* Binary formats: ELF + `#!/...` shebangs
-* Rust-style interfaces for most of the stuff like memory management, devices etc.
-* PCI Express devices
-    * NVMe drive support (read/write, currently x86_64 only, due to lack of MSI-X support on aarch64).
-    * AHCI SATA drive support (read/write)
-    * xHCI USB host controller
-* USB HID keyboards
+	* In-memory read-write filesystem for tar-based initrd
+	* sysfs
+	* devfs
+	* ext2
+* Userspace features:
+	* [Kernel-user ABI](lib/abi-def/yggdrasil.abi) generated from a rust-like description language
+	* Sanitized system calls better suited for use in Rust
+	* Binary formats: ELF + `#!/...` shebangs
+	* C compatibility through a [custom Rust libc](userspace/lib/ygglibc)
+	* Userspace multithreading
+	* Synchronization primitives through futex-like interface
+	* Unix-like signals and exceptions
+	* [Dynamic loader](userspace/dyn-loader) for linking with shared libraries
+* Hardware features:
+	* PCI Express devices
+    	* NVMe drive support (read/write, currently x86_64 only, due to lack of MSI-X support on aarch64/i686).
+    	* AHCI SATA drive support (read/write)
+    	* xHCI USB host controller
+		* VirtIO Network + GPU framebuffer support
+	* USB HID keyboards
 
 aarch64-specific:
 
@@ -32,39 +48,34 @@ aarch64-specific:
 * ARM generic timer as system/monotonic timer
 * GICv2 IRQ controller
 
-x86_64-specific:
+x86-specific:
 
-* UEFI boot through [yboot](https://git.alnyan.me/yggdrasil/yboot)
-	(no plans for legacy boot)
+* Boot options:
+	* x86_64: UEFI [yboot](https://git.alnyan.me/yggdrasil/yboot)
+	* i686: multiboot/grub
 * I/O and Local APIC IRQ controllers
 * PS/2 keyboard
-* i8253-based timer (got some problems with HPET on
-	real hw, had to revert, lol)
+* HPET for x86_64
+* i8253-based timer for i686 or as a fallback timer
 * COM ports
 * ACPI, [work in progress](https://github.com/rust-osdev/acpi), mostly broken
-	on real hardware
+	on real hardware, so currently disabled
 	* ACPI shutdown
 	* PCI IRQ pin routing
 	* Events like power button, etc.
-* Fancy framebuffer console
-
-Userspace features:
-
-* Sanitized system calls better suited for Rust
-* Userspace threads
-* Synchronization primitives through futex-like interface
-* Unix-like signals and exceptions
 
 Building the OS
 ---------------
 
+**NOTE** This project uses `xtask`-style build system. To see help, use `cargo xtask --help`.
+
 Prerequisites:
 
 * Decent CPU and a sizable amount of RAM
-* ~20G of free disk space
+* ~40G of free disk space for a full build
 * Patience
 
-**NOTE** Full OS build requires you to build the `*-unknown-yggdrasil` 
+**NOTE** Full OS build requires you to build the `*-unknown-yggdrasil`
 Rust toolchain, which may take quite a while, so be prepared.
 
 Steps:
@@ -94,21 +105,30 @@ Steps:
 	```
 
 2. Run `cargo xtask toolchain` to fetch, build and link the toolchain
+
+	**NOTE** if toolchain fetch fails for some reason, try cloning directly
+	from `https://git.alnyan.me/yggdrasil/yggdrasil-rust.git` with appropriate
+	branch.
+
 3. Run `cargo xtask` to build the OS.
 
-Once the OS has been built, you can run it in QEMU by executing 
+Once the OS has been built, you can run it in QEMU by executing
 `cargo xtask qemu`. For more `xtask` commands, see `cargo xtask --help`.
 
 General plans (in no particular order)
 --------------------------------------
 
-* Better unification of architecture code
-* `async` for VFS (?)
-* Code cleanup, I've been doing quite a lazy job at that lately...
+1. Get it running on more real hardware
+2. Get a full LLVM build to work
+3. Get rustc to work
+4. Get self-hosted
+5. Run doom (?)
+
+In addition to eternal code cleanup, I've been doing quite a lazy job at that lately...
 
 Navigation
 ----------
 
-* Kernel: [`/kernel`](/kernel)
-* Userspace: [`/userspace`](/userspace)
-* ABI definitions: [`/lib/abi-def`](/lib/abi-def)
+* Kernel: [`kernel`](kernel)
+* Userspace: [`userspace`](userspace)
+* ABI definitions: [`lib/abi-def`](lib/abi-def)

boot/yboot-proto/src/lib.rs

@@ -1,3 +1,4 @@
+#![allow(unused)]
 #![no_std]
 
 use bytemuck::{Pod, Zeroable};

boot/yboot-proto/src/v1.rs

@@ -4,6 +4,9 @@ use bytemuck::{Pod, Zeroable};
 
 use crate::{AvailableRegion, IterableMemoryMap, LoadProtocolHeader};
 
+pub const PIXEL_R8G8B8A8: u32 = 1;
+pub const PIXEL_B8G8R8A8: u32 = 2;
+
 #[derive(Clone, Copy, Pod, Zeroable)]
 #[repr(C)]
 pub struct LoadProtocol {
@@ -12,6 +15,9 @@ pub struct LoadProtocol {
 
     pub memory_map: MemoryMap,
 
+    pub cmdline: u64,
+    pub cmdline_len: u64,
+
     pub rsdp_address: u64,
     pub initrd_address: u64,
     pub initrd_size: u64,
@@ -46,6 +52,8 @@ pub struct FramebufferOption {
     pub res_stride: u64,
     pub res_address: u64,
     pub res_size: u64,
+    pub res_format: u32,
+    pub _0: u32,
 }
 
 impl AvailableRegion for AvailableMemoryRegion {

boot/yboot/src/config.rs

@@ -0,0 +1,39 @@
+use uefi::{
+    proto::media::file::{Directory, File, FileAttribute, FileMode},
+    CStr16, Result, Status,
+};
+
+#[derive(Debug)]
+pub struct Config {
+    pub cmdline: [u8; 4096],
+    pub cmdline_len: usize,
+}
+
+impl Default for Config {
+    fn default() -> Self {
+        Self {
+            cmdline: [0; 4096],
+            cmdline_len: 0,
+        }
+    }
+}
+
+impl Config {
+    pub fn load(root: &mut Directory, path: &CStr16) -> Result<Self> {
+        let file = match root.open(path, FileMode::Read, FileAttribute::empty()) {
+            Ok(file) => file,
+            Err(error) => {
+                root.reset_entry_readout().ok();
+                log::warn!("Couldn't open {path:?}: {error:?}");
+                return Ok(Self::default());
+            }
+        };
+        root.reset_entry_readout().ok();
+        let mut this = Self::default();
+        let mut file = file.into_regular_file().ok_or(Status::INVALID_PARAMETER)?;
+
+        this.cmdline_len = file.read(&mut this.cmdline)?;
+
+        Ok(this)
+    }
+}

boot/yboot/src/elf.rs

@@ -1,7 +1,8 @@
 use core::mem::size_of;
 
 use bytemuck::Zeroable;
-use log::{debug, error, info};
+use log::{error, info};
+use types::{Rela, SHT_RELA};
 // TODO use 'elf' crate
 use uefi::{
     prelude::BootServices,
@@ -15,6 +16,7 @@ use crate::elf::types::{PT_LOAD, SHF_ALLOC, SHF_WRITE, SHT_PROGBITS};
 
 use self::types::{Ehdr, Phdr, Shdr};
 
+#[allow(unused)]
 mod types {
     use bytemuck::{Pod, Zeroable};
 
@@ -23,14 +25,18 @@ mod types {
     pub type Half = u16;
     pub type Word = u32;
     pub type XWord = u64;
+    pub type SXWord = i64;
 
     pub const PT_LOAD: Word = 1;
 
     pub const SHT_PROGBITS: Word = 1;
+    pub const SHT_RELA: Word = 4;
 
     pub const SHF_WRITE: XWord = 1 << 0;
     pub const SHF_ALLOC: XWord = 1 << 1;
 
+    pub const R_X86_64_RELATIVE: u32 = 8;
+
     #[derive(Clone, Copy, Zeroable, Pod)]
     #[repr(C)]
     pub struct Ehdr {
@@ -77,6 +83,20 @@ mod types {
         pub memsz: XWord,
         pub align: XWord,
     }
+
+    #[derive(Clone, Copy, Zeroable, Pod)]
+    #[repr(C)]
+    pub struct Rela {
+        pub offset: Addr,
+        pub info: XWord,
+        pub addend: SXWord,
+    }
+
+    impl Rela {
+        pub fn r_type(&self) -> u32 {
+            self.info as u32
+        }
+    }
 }
 
 // Maximum address this loader can map in the target kernel
@@ -92,6 +112,8 @@ pub struct LoadedObject {
     pub image_start: u64,
     pub image_end: u64,
 
+    pub load_address: u64,
+
     pub entry: u64,
 
     pub protocol_struct_paddr: u64,
@@ -105,6 +127,12 @@ struct LocatedProtocol {
     size: usize,
 }
 
+struct RelaSection {
+    offset: u64,
+    entry_count: usize,
+    entry_size: usize,
+}
+
 trait ReadExact {
     fn read_exact(&mut self, buf: &mut [u8]) -> Result<(), Error>;
 }
@@ -119,6 +147,23 @@ impl ReadExact for RegularFile {
     }
 }
 
+impl RelaSection {
+    pub fn from_shdr(shdr: &Shdr) -> Option<Self> {
+        if shdr.type_ != SHT_RELA {
+            return None;
+        }
+
+        let entry_size = shdr.entsize as usize;
+        let entry_count = shdr.size as usize / entry_size;
+
+        Some(Self {
+            offset: shdr.offset,
+            entry_size,
+            entry_count,
+        })
+    }
+}
+
 impl Object {
     pub fn open<D: File>(root: &mut D, path: &CStr16) -> Result<Self, Error> {
         let file = root.open(path, FileMode::Read, FileAttribute::empty())?;
@@ -140,11 +185,11 @@ impl Object {
             return Err(Error::new(Status::LOAD_ERROR, ()));
         }
 
-        // Check that the entry point is set
-        if ehdr.entry == 0 {
-            error!("Image does not have a valid entry point");
-            return Err(Error::new(Status::LOAD_ERROR, ()));
-        }
+        // // Check that the entry point is set
+        // if ehdr.entry == 0 {
+        //     error!("Image does not have a valid entry point");
+        //     return Err(Error::new(Status::LOAD_ERROR, ()));
+        // }
 
         Ok(Self { file, ehdr })
     }
@@ -170,11 +215,6 @@ impl Object {
         self.file
             .read_exact(bytemuck::bytes_of_mut(&mut proto_data))?;
 
-        info!(
-            "Kernel is virtually mapped at {:#x}",
-            proto_data.kernel_virt_offset
-        );
-
         // 2. Find the kernel's range and check that the loaded physical addresses are actually
         //    usable from UEFI
         let mut image_start = u64::MAX;
@@ -209,16 +249,22 @@ impl Object {
         assert_eq!(image_start & 0xFFF, 0);
         assert_eq!(image_end & 0xFFF, 0);
 
-        info!("Image start: {:#x}, end: {:#x}", image_start, image_end);
-        // Reserve the kernel memory
-        let reserved_addr = bs
+        // Allocate memory to load the kernel into
+        let kernel_load_address = bs
             .allocate_pages(
-                AllocateType::Address(image_start),
+                AllocateType::MaxAddress(0xFFFFFFFF),
                 MemoryType::LOADER_DATA,
                 (image_end - image_start) as usize / 0x1000,
             )
-            .expect("Could not allocate memory for kernel image");
-        assert_eq!(reserved_addr, image_start);
+            .expect("Could not allocate memory for the kernel");
+
+        // Print info
+        info!("Image start: {:#x}, end: {:#x}", image_start, image_end);
+        info!(
+            "Kernel virtual offset: {:#x}",
+            proto_data.kernel_virt_offset
+        );
+        info!("Kernel load address: {kernel_load_address:#x}");
 
         // 3. Load the segments
         for i in 0..self.ehdr.phnum {
@@ -228,53 +274,80 @@ impl Object {
                 continue;
             }
 
+            let segment_load_base = phdr.paddr + kernel_load_address;
+
             info!(
-                "Load segment {}: {:#x?}",
-                i,
-                phdr.paddr..phdr.paddr + phdr.memsz
+                "[{i}] Load {:#x?}",
+                segment_load_base..segment_load_base + phdr.memsz
             );
 
             if phdr.filesz > 0 {
-                // The section has load data
-                let dst = unsafe {
-                    core::slice::from_raw_parts_mut(phdr.paddr as *mut u8, phdr.filesz as usize)
+                let dst_slice = unsafe {
+                    core::slice::from_raw_parts_mut(
+                        segment_load_base as *mut u8,
+                        phdr.filesz as usize,
+                    )
                 };
 
-                debug!(
-                    "Load {:#x?} from ELF offset {:#x}",
-                    phdr.paddr..phdr.paddr + phdr.filesz,
-                    phdr.offset
-                );
                 self.file.set_position(phdr.offset)?;
-                self.file.read_exact(dst)?;
+                self.file.read_exact(dst_slice)?;
             }
 
             if phdr.memsz > phdr.filesz {
-                let dst = unsafe {
+                let dst_slice = unsafe {
                     core::slice::from_raw_parts_mut(
-                        (phdr.paddr + phdr.filesz) as *mut u8,
+                        (segment_load_base + phdr.filesz) as *mut u8,
                         (phdr.memsz - phdr.filesz) as usize,
                     )
                 };
 
-                debug!(
-                    "Zero data {:#x?}",
-                    phdr.paddr + phdr.filesz..phdr.paddr + phdr.memsz
-                );
+                dst_slice.fill(0);
+            }
+        }
 
-                dst.fill(0);
+        // 4. Perform kernel relocation
+        let mut rela_section = None;
+        for i in 0..self.ehdr.shnum as usize {
+            let shdr = self.read_shdr(i)?;
+            if let Some(rela) = RelaSection::from_shdr(&shdr) {
+                rela_section = Some(rela);
+                break;
+            }
+        }
+
+        if let Some(rela_section) = rela_section {
+            info!("Relocating kernel: {image_start:#x} -> {kernel_load_address:#x}");
+            info!("({} relocations)", rela_section.entry_count);
+
+            let b = (kernel_load_address + proto_data.kernel_virt_offset) as i64;
+
+            for i in 0..rela_section.entry_count {
+                let mut rela = Rela::zeroed();
+                self.file
+                    .set_position(rela_section.offset + (i * rela_section.entry_size) as u64)?;
+                self.file.read_exact(bytemuck::bytes_of_mut(&mut rela))?;
+
+                match rela.r_type() {
+                    types::R_X86_64_RELATIVE => {
+                        let qword = (rela.offset + kernel_load_address) as *mut i64;
+                        let value = rela.addend + b;
+                        unsafe { qword.write_volatile(value) };
+                    }
+                    other => todo!("Unsupported relocation type: {other}"),
+                }
             }
         }
 
         // Now that the image is in memory, protocol structure can be written in the further steps
-        let protocol_struct_paddr = (loc_proto.address as u64) - proto_data.kernel_virt_offset;
+        let protocol_struct_paddr = loc_proto.address as u64 + kernel_load_address; // (loc_proto.address as u64) - proto_data.kernel_virt_offset;
         let protocol_version = proto_data.header.version;
 
-        let entry = self.ehdr.entry;
+        let entry = self.ehdr.entry + kernel_load_address;
 
         Ok(LoadedObject {
             image_start,
             image_end,
+            load_address: kernel_load_address,
             entry,
             protocol_struct_paddr,
             protocol_version,

boot/yboot/src/initrd.rs

@@ -25,7 +25,7 @@ pub fn load_somewhere(
 
     let file_info: &FileInfo = file.get_info(&mut info_buffer).unwrap();
     let size = file_info.file_size();
-    let page_count = (size + 0xFFF) / 0x1000;
+    let page_count = size.div_ceil(0x1000);
 
     let base = bs.allocate_pages(
         AllocateType::MaxAddress(MAXIMUM_ADDRESS),

boot/yboot/src/main.rs

@@ -2,6 +2,7 @@
 #![no_std]
 #![no_main]
 
+pub mod config;
 pub mod elf;
 pub mod initrd;
 pub mod mem;
@@ -9,13 +10,16 @@ pub mod protocol_ext;
 
 use core::{arch::asm, mem::size_of, ops::Deref};
 
+use config::Config;
 use elf::Object;
 use log::{debug, error, info};
 use uefi::{
     prelude::*,
     proto::{
-        console::gop::GraphicsOutput, device_path::DevicePath, loaded_image::LoadedImage,
-        media::fs::SimpleFileSystem,
+        console::gop::{self, GraphicsOutput, PixelFormat},
+        device_path::DevicePath,
+        loaded_image::LoadedImage,
+        media::{file::Directory, fs::SimpleFileSystem},
     },
     table::{
         boot::{AllocateType, MemoryType, ScopedProtocol},
@@ -24,34 +28,49 @@ use uefi::{
     Error,
 };
 use yboot_proto::{
-    v1::{AvailableMemoryRegion, FramebufferOption},
+    v1::{self, AvailableMemoryRegion, FramebufferOption},
     LoadProtocolV1, LOADER_MAGIC,
 };
 
-use crate::{mem::MemoryDescriptorExt, protocol_ext::GraphicsOutputExt};
+use crate::mem::MemoryDescriptorExt;
+
+fn mode_score(mode: &gop::Mode) -> usize {
+    let (w, h) = mode.info().resolution();
+    let mut size_score = w * h;
+    if w > 1920 || h > 1080 {
+        // Don't pick too large sizes
+        size_score = 0;
+    }
+    size_score
+}
 
 fn setup_framebuffer(bs: &BootServices, fb: &mut FramebufferOption) -> Result<(), Error> {
     let gop_handle = bs.get_handle_for_protocol::<GraphicsOutput>()?;
     let mut gop = bs.open_protocol_exclusive::<GraphicsOutput>(gop_handle)?;
 
-    // Find the requested mode
-    let mode = gop.match_mode(fb.req_width, fb.req_height).ok_or_else(|| {
-        error!(
-            "Requested mode is not supported: {}x{}",
-            fb.req_width, fb.req_height
-        );
+    let mode = gop.modes().max_by_key(mode_score).ok_or_else(|| {
+        error!("No mode found");
         Error::new(Status::INVALID_PARAMETER, ())
     })?;
 
     gop.set_mode(&mode)?;
 
+    let (res_width, res_height) = mode.info().resolution();
+
     let mut result = gop.frame_buffer();
 
-    fb.res_width = fb.req_width;
-    fb.res_height = fb.req_height;
+    let format = match mode.info().pixel_format() {
+        PixelFormat::Bgr => v1::PIXEL_B8G8R8A8,
+        PixelFormat::Rgb => v1::PIXEL_R8G8B8A8,
+        _ => 0,
+    };
+
+    fb.res_width = res_width as _;
+    fb.res_height = res_height as _;
     fb.res_address = result.as_mut_ptr() as _;
     fb.res_stride = mode.info().stride() as u64 * 4;
     fb.res_size = result.size() as _;
+    fb.res_format = format;
 
     info!(
         "Framebuffer: {}x{} @ {:#x}",
@@ -74,7 +93,7 @@ fn locate_rsdp(st: &SystemTable<Boot>) -> Option<u64> {
 fn boot_partition(
     image: Handle,
     bs: &BootServices,
-) -> Result<ScopedProtocol<SimpleFileSystem>, Error> {
+) -> Result<ScopedProtocol<'_, SimpleFileSystem>, Error> {
     let loaded_image = bs.open_protocol_exclusive::<LoadedImage>(image)?;
     let device_handle = loaded_image.device();
 
@@ -85,27 +104,45 @@ fn boot_partition(
     bs.open_protocol_exclusive::<SimpleFileSystem>(fs_handle)
 }
 
+fn open_root(image: Handle, bs: &BootServices) -> Result<Directory, Error> {
+    let mut boot_partition = boot_partition(image, bs)?;
+    boot_partition.open_volume()
+}
+
 fn load_kernel<'a>(
-    ih: Handle,
+    config: &Config,
+    root: &mut Directory,
     st: &SystemTable<Boot>,
-) -> Result<(u64, u64, &'a mut LoadProtocolV1), Error> {
+) -> Result<(u64, u64, u64, &'a mut LoadProtocolV1), Error> {
     let bs = st.boot_services();
 
-    let mut fs = boot_partition(ih, bs)?;
-    let mut root = fs.open_volume()?;
-
-    let mut kernel_obj = Object::open(&mut root, cstr16!("kernel.elf"))?;
+    let mut kernel_obj = Object::open(root, cstr16!("kernel.elf"))?;
     let loaded_obj = kernel_obj.load(bs)?;
 
     debug!("Loaded object: {:#x?}", loaded_obj);
 
     // Load initrd
-    let (initrd_start, initrd_size) = initrd::load_somewhere(bs, &mut root, cstr16!("initrd.img"))?;
+    let (initrd_start, initrd_size) = initrd::load_somewhere(bs, root, cstr16!("initrd.img"))?;
     debug!(
         "Loaded initrd: {:#x?}",
         initrd_start..initrd_start + initrd_size
     );
 
+    // Load cmdline
+    let cmdline = if config.cmdline_len != 0 {
+        let address = bs.allocate_pages(AllocateType::AnyPages, MemoryType::LOADER_DATA, 1)?;
+
+        let dst =
+            unsafe { core::slice::from_raw_parts_mut(address as *mut u8, config.cmdline_len) };
+        dst.copy_from_slice(&config.cmdline[..config.cmdline_len]);
+
+        debug!("Cmdline at {:#x?}", address);
+
+        address
+    } else {
+        0
+    };
+
     // Other versions are not existent yet
     assert_eq!(loaded_obj.protocol_version, 1);
     let proto_data = unsafe { &mut *(loaded_obj.protocol_struct_paddr as *mut LoadProtocolV1) };
@@ -116,6 +153,9 @@ fn load_kernel<'a>(
     })?;
     info!("RSDP at {:#x}", rsdp);
 
+    proto_data.cmdline = cmdline;
+    proto_data.cmdline_len = config.cmdline_len as _;
+
     proto_data.rsdp_address = rsdp;
     proto_data.initrd_address = initrd_start;
     proto_data.initrd_size = initrd_size;
@@ -143,13 +183,14 @@ fn load_kernel<'a>(
 
     let entry = loaded_obj.entry + proto_data.kernel_virt_offset;
 
-    Ok((entry, mmap_memory, proto_data))
+    Ok((entry, loaded_obj.load_address, mmap_memory, proto_data))
 }
 
 unsafe fn map_and_enter_kernel(
     st: SystemTable<Boot>,
     proto_data: &mut LoadProtocolV1,
     mmap_memory: u64,
+    load_base: u64,
     entry: u64,
 ) -> ! {
     let (_, mmap) = st.exit_boot_services();
@@ -176,16 +217,43 @@ unsafe fn map_and_enter_kernel(
     let cr3 = mem::map_image();
     asm!("cli; wbinvd; mov {0}, %cr3", in(reg) cr3, options(att_syntax));
 
-    asm!("jmp *{0}", in(reg) entry, in("eax") LOADER_MAGIC, options(noreturn, att_syntax));
+    asm!("jmp *{0}", in(reg) entry, in("eax") LOADER_MAGIC, in("ecx") load_base, options(noreturn, att_syntax));
 }
 
 #[entry]
 fn efi_main(image_handle: Handle, mut system_table: SystemTable<Boot>) -> Status {
-    uefi_services::init(&mut system_table).unwrap();
+    if uefi_services::init(&mut system_table).is_err() {
+        return Status::LOAD_ERROR;
+    }
 
-    let (entry, mmap_memory, proto_data) = load_kernel(image_handle, &system_table).unwrap();
+    let bs = system_table.boot_services();
+
+    let mut root = match open_root(image_handle, bs) {
+        Ok(root) => root,
+        Err(error) => {
+            error!("Could not open boot partition root: {error:?}");
+            return Status::LOAD_ERROR;
+        }
+    };
+
+    let config = match Config::load(&mut root, cstr16!("yboot.cfg")) {
+        Ok(config) => config,
+        Err(error) => {
+            error!("Malformed yboot.cfg: {error:?}");
+            return Status::LOAD_ERROR;
+        }
+    };
+
+    let (entry, load_base, mmap_memory, proto_data) =
+        match load_kernel(&config, &mut root, &system_table) {
+            Ok(e) => e,
+            Err(error) => {
+                error!("Failed to load the kernel/initrd: {error:?}");
+                return Status::LOAD_ERROR;
+            }
+        };
 
     unsafe {
-        map_and_enter_kernel(system_table, proto_data, mmap_memory, entry);
+        map_and_enter_kernel(system_table, proto_data, mmap_memory, load_base, entry);
     }
 }

boot/yboot/src/mem.rs

@@ -66,7 +66,7 @@ impl MemoryDescriptorExt for MemoryDescriptor {
     }
 }
 
-pub fn memory_map(bs: &BootServices) -> Result<MemoryMap, Error> {
+pub fn memory_map(bs: &BootServices) -> Result<MemoryMap<'_>, Error> {
     bs.memory_map(unsafe { &mut MMAP_BUFFER.data })
 }
 

doc/raspi4b.txt

@@ -0,0 +1,98 @@
+**NOTE** I haven't yet tested direct boot through Raspberry's
+         proprietary bootloader.
+
+Booting Yggdrasil on Raspberry Pi 4B with u-boot:
+
+1. Clone u-boot sources and build with rpi_4_defconfig
+
+**NOTE** I assume you have all the bootloader parts in the boot partition already.
+         If not, clone raspberry fw repo and copy the following files to the boot partition:
+            * bootcode.bin
+            * start4.elf
+            * all the .dtb files (a bcm2711-rpi-4-b.dtb should be enough though)
+
+2. Copy u-boot.bin into the Pi's boot partition and edit the config.txt:
+enable_uart=1
+arm64_bit=1
+kernel=u-boot.bin
+
+3. Compile the OS with `cargo xtask --arch=aarch64 --board=raspi4b --release`
+4. Copy the following files into some directory:
+    * target/aarch64-unknown-raspi4b/release/kernel.bin
+    * userspace/target/aarch64-unknown-yggdrasil/release/initrd.tar
+5. cd into that directory and start a TFTP server of your choice. I used `uftpd`.
+
+6. Connect an ethernet and serial to the Pi and run the following commands in u-boot shell:
+
+### If using DHCP
+$ dhcp
+### If not using DHCP
+$ env set ipaddr <RASPBERRY-IP-ADDR>
+$ env set fdt_addr_r 0x11000000
+$ env set initrd_addr_r 0x04000000
+$ tftpboot ${initrd_addr_r} <BUILD-MACHINE-IP-ADDR>:initrd.tar
+$ tftpboot ${loadaddr} <BUILD-MACHINE-IP-ADDR>:kernel.bin
+$ load mmc 0:1 ${fdt_addr_r} bcm2711-rpi-4-b.dtb
+$ fdt addr ${fdt_addr_r}
+$ fdt resize
+$ fdt memory 0x0 0x3C000000
+$ booti ${loadaddr} ${initrd_addr_r}:<SIZE-PRINTED-WHEN-LOADING-INITRD> ${fdt_addr_r}
+
+###### Assuming BUILD-MACHINE-IP-ADDR is 13.0.0.1 and RASPBERRY-IP-ADDR is 13.0.0.2, here's
+###### a quick command for a development boot
+###### (FIXME when initrd gets larger than 64MiB)
+
+env set ipaddr 13.0.0.2; env set fdt_addr_r 0x11000000; env set initrd_addr_r 0x04000000; tftpboot ${initrd_addr_r} 13.0.0.1:initrd.tar; tftpboot ${loadaddr} 13.0.0.1:kernel.bin; load mmc 0:1 ${fdt_addr_r} bcm2711-rpi-4-b.dtb; fdt addr ${fdt_addr_r}; fdt resize; fdt memory 0x0 0x3C000000; booti ${loadaddr} ${initrd_addr_r}:67108864 ${fdt_addr_r}
+
+dhcp;
+env set initrd_addr_r 0x20000000; tftpboot ${initrd_addr_r} 192.168.88.10:initrd.img; tftpboot ${loadaddr} 192.168.88.10:yggdrasil-kernel.bin; load mmc 1:1 ${fdt_addr_r} bcm2711-rpi-4-b.dtb; fdt addr ${fdt_addr_r}; fdt resize; fdt memory 0x0 0x3C000000; booti ${loadaddr} ${initrd_addr_r}:0x4000000 ${fdt_addr_r}
+
+
+Missing drivers:
+
+No driver for Some("hvs@7e400000") ("brcm,bcm2711-hvs")
+No driver for Some("i2c@7e804000") ("brcm,bcm2711-i2c")
+  also "brcm,bcm2835-i2c"
+No driver for Some("usb@7e980000") ("brcm,bcm2835-usb")
+No driver for Some("local_intc@40000000") ("brcm,bcm2836-l1-intc")
+: avs-monitor@7d5d2000: probed
+No driver for Some("thermal") ("brcm,bcm2711-thermal")
+No driver for Some("dma@7e007000") ("brcm,bcm2835-dma")
+No driver for Some("watchdog@7e100000") ("brcm,bcm2835-pm")
+  also "brcm,bcm2835-pm-wdt"
+No driver for Some("rng@7e104000") ("brcm,bcm2711-rng200")
+No driver for Some("pixelvalve@7e206000") ("brcm,bcm2711-pixelvalve0")
+No driver for Some("pixelvalve@7e207000") ("brcm,bcm2711-pixelvalve1")
+No driver for Some("pixelvalve@7e20a000") ("brcm,bcm2711-pixelvalve2")
+No driver for Some("pwm@7e20c800") ("brcm,bcm2835-pwm")
+No driver for Some("pixelvalve@7e216000") ("brcm,bcm2711-pixelvalve4")
+No driver for Some("clock@7ef00000") ("brcm,brcm2711-dvp")
+No driver for Some("interrupt-controller@7ef00100") ("brcm,bcm2711-l2-intc")
+  also "brcm,l2-intc"
+No driver for Some("hdmi@7ef00700") ("brcm,bcm2711-hdmi0")
+No driver for Some("i2c@7ef04500") ("brcm,bcm2711-hdmi-i2c")
+No driver for Some("hdmi@7ef05700") ("brcm,bcm2711-hdmi1")
+No driver for Some("i2c@7ef09500") ("brcm,bcm2711-hdmi-i2c")
+No driver for Some("firmware") ("raspberrypi,bcm2835-firmware")
+  also "simple-mfd"
+No driver for Some("clocks") ("raspberrypi,firmware-clocks")
+No driver for Some("gpio") ("raspberrypi,firmware-gpio")
+No driver for Some("reset") ("raspberrypi,firmware-reset")
+No driver for Some("power") ("raspberrypi,bcm2835-power")
+No driver for Some("mailbox@7e00b840") ("brcm,bcm2835-vchiq")
+No driver for Some("phy") ("usb-nop-xceiv")
+No driver for Some("gpu") ("brcm,bcm2711-vc5")
+No driver for Some("mmc@7e340000") ("brcm,bcm2711-emmc2")
+No driver for Some("arm-pmu") ("arm,cortex-a72-pmu")
+  also "arm,armv8-pmuv3"
+No driver for Some("cpu@0") ("arm,cortex-a72")
+No driver for Some("cpu@1") ("arm,cortex-a72")
+No driver for Some("cpu@2") ("arm,cortex-a72")
+No driver for Some("cpu@3") ("arm,cortex-a72")
+No driver for Some("pcie@7d500000") ("brcm,bcm2711-pcie")
+No driver for Some("ethernet@7d580000") ("brcm,bcm2711-genet-v5")
+No driver for Some("mdio@e14") ("brcm,genet-mdio-v5")
+No driver for Some("leds") ("gpio-leds")
+No driver for Some("wifi-pwrseq") ("mmc-pwrseq-simple")
+No driver for Some("sd_io_1v8_reg") ("regulator-gpio")
+No driver for Some("sd_vcc_reg") ("regulator-fixed")

doc/visionfive2.txt

@@ -0,0 +1,130 @@
+Booting Yggdrasil OS on Starfive VisionFive 2 RISC-V board:
+
+* TODO: proper format for initrd image
+
+Prerequisites:
+
+* OpenSBI + u-boot (you can use the regular debian installation from Starfive)
+* yggdrasil-kernel.bin
+* initrd.img
+
+Steps:
+
+1. Copy yggdrasil-kernel.bin and initrd.img into some directory and start a TFTP server there
+2. Connect to VF2's serial port, ethernet and enter u-boot
+3. Run the following commands in u-boot:
+
+### If using DHCP
+$ dhcp
+### If not using DHCP
+$ env set ipaddr <VF2-IP-ADDR>
+$ env set initrd_addr_r 0x70000000
+### [Optional] set some kernel cmdline params
+$ env set bootargs "debug.serial-level=info"
+$ tftpboot ${initrd_addr_r} <BUILD-MACHINE-IP-ADDR>:initrd.img
+$ tftpboot ${loadaddr} <BUILD-MACHINE-IP-ADDR>:yggdrasil-kernel.bin
+$ load mmc 1:3 ${fdt_addr_r} dtbs/6.6.20-starfive/starfive/${fdtfile}
+$ fdt resize
+$ booti ${loadaddr} ${initrd_addr_r}:<initrd-size> ${fdt_addr_r}
+
+###### Assuming BUILD-MACHINE-IP-ADDR is 13.0.0.1 and VF2-IP-ADDR is 13.0.0.2, here's
+###### a quick command for a development boot
+###### (FIXME when initrd gets larger than 64MiB)
+
+env set ipaddr 13.0.0.2; env set initrd_addr_r 0x70000000; tftpboot ${initrd_addr_r} 13.0.0.1:initrd.img; tftpboot ${loadaddr} 13.0.0.1:yggdrasil-kernel.bin; load mmc 1:3 ${fdt_addr_r} dtbs/6.6.20-starfive/starfive/${fdtfile}; fdt resize; booti ${loadaddr} ${initrd_addr_r}:60000000 ${fdt_addr_r}
+
+#### For DHCP boot with BUILD-MACHINE-IP-ADDR 192.168.88.10
+
+dhcp
+dhcp; env set initrd_addr_r 0x70000000; tftpboot ${initrd_addr_r} 192.168.88.10:initrd.img; tftpboot ${loadaddr} 192.168.88.10:yggdrasil-kernel.bin; load mmc 1:3 ${fdt_addr_r} dtbs/6.6.20-starfive/starfive/${fdtfile}; fdt resize; booti ${loadaddr} ${initrd_addr_r}:60000000 ${fdt_addr_r}
+
+
+Missing drivers:
+
+Clock/reset/pin:
+    No driver for Some("pinctrl@17020000") ("starfive,jh7110-aon-pinctrl")
+    No driver for Some("clock-controller@19810000") ("starfive,jh7110-ispcrg")
+    No driver for Some("clock-controller@295c0000") ("starfive,jh7110-voutcrg")
+
+Power/reg/GPIO:
+    No driver for Some("opp-table-0") ("operating-points-v2")
+    No driver for Some("pmic@36") ("x-powers,axp15060")
+    No driver for Some("power-controller@17030000") ("starfive,jh7110-pmu")
+    No driver for Some("leds") ("gpio-leds")
+    No driver for Some("gpio-restart") ("gpio-restart")
+
+Serial:
+    No driver for Some("i2c@10030000") ("snps,designware-i2c")
+    No driver for Some("i2c@10050000") ("snps,designware-i2c")
+    No driver for Some("i2c@12050000") ("snps,designware-i2c")
+    No driver for Some("i2c@12060000") ("snps,designware-i2c")
+    No driver for Some("spi@10060000") ("arm,pl022")
+      also "arm,primecell"
+
+Bus:
+    No driver for Some("usb@10100000") ("starfive,jh7110-usb")
+    No driver for Some("usb@0") ("cdns,usb3")
+    No driver for Some("phy@10200000") ("starfive,jh7110-usb-phy")
+    No driver for Some("phy@10210000") ("starfive,jh7110-pcie-phy")
+    No driver for Some("phy@10220000") ("starfive,jh7110-pcie-phy")
+    No driver for Some("pcie@940000000") ("starfive,jh7110-pcie")
+    No driver for Some("pcie@9c0000000") ("starfive,jh7110-pcie")
+
+Interrupt:
+    No driver for Some("interrupt-controller") ("riscv,cpu-intc")
+    No driver for Some("timer@2000000") ("starfive,jh7110-clint")
+      also "sifive,clint0"
+
+Display/GPU subsystem:
+    No driver for Some("display-subsystem") ("starfive,jh7110-display")
+      also "verisilicon,display-subsystem"
+    No driver for Some("dsi-output") ("starfive,jh7110-display-encoder")
+      also "verisilicon,dsi-encoder"
+    No driver for Some("jpu@13090000") ("starfive,jpu")
+    No driver for Some("vpu_dec@130a0000") ("starfive,vdec")
+    No driver for Some("vpu_enc@130b0000") ("starfive,venc")
+    No driver for Some("gpu@18000000") ("img-gpu")
+    No driver for Some("vin_sysctl@19800000") ("starfive,jh7110-vin")
+    No driver for Some("phy@19820000") ("starfive,jh7110-dphy-rx")
+    No driver for Some("dc8200@29400000") ("starfive,jh7110-dc8200")
+      also "verisilicon,dc8200"
+    No driver for Some("hdmi@29590000") ("starfive,jh7110-hdmi")
+      also "inno,hdmi"
+    No driver for Some("mipi@295d0000") ("starfive,jh7110-mipi_dsi")
+      also "cdns,dsi"
+    No driver for Some("mipi-dphy@295e0000") ("starfive,jh7110-mipi-dphy-tx")
+      also "m31,mipi-dphy-tx"
+
+Misc:
+    No driver for Some("mailbox_client") ("starfive,mailbox-test")
+    No driver for Some("cache-controller@2010000") ("starfive,jh7110-ccache")
+      also "sifive,ccache0"
+      also "cache"
+    No driver for Some("pwm@120d0000") ("starfive,jh7110-pwm")
+      also "opencores,pwm-v1"
+    No driver for Some("temperature-sensor@120e0000") ("starfive,jh7110-temp")
+    No driver for Some("timer@13050000") ("starfive,jh7110-timer")
+    No driver for Some("mailbox@13060000") ("starfive,mail_box")
+    No driver for Some("watchdog@13070000") ("starfive,jh7110-wdt")
+    No driver for Some("crypto@16000000") ("starfive,jh7110-crypto")
+    No driver for Some("rng@1600c000") ("starfive,jh7110-trng")
+    No driver for Some("mdio") ("snps,dwmac-mdio")
+    No driver for Some("mdio") ("snps,dwmac-mdio")
+    No driver for Some("dma-controller@16050000") ("starfive,jh7110-axi-dma")
+    No driver for Some("dma-controller@16008000") ("arm,pl080")
+      also "arm,primecell"
+    No driver for Some("rtc@17040000") ("starfive,jh7110-rtc")
+    No driver for Some("e24@6e210000") ("starfive,e24")
+    No driver for Some("linux,cma") ("shared-dma-pool")
+
+Storage:
+    No driver for Some("spi@13010000") ("starfive,jh7110-qspi")
+      also "cdns,qspi-nor"
+    No driver for Some("flash@0") ("jedec,spi-nor")
+    No driver for Some("partitions") ("fixed-partitions")
+    No driver for Some("mmc@16010000") ("starfive,jh7110-mmc")
+    No driver for Some("mmc@16020000") ("starfive,jh7110-mmc")
+
+Audio:
+    No driver for Some("pwmdac@100b0000") ("starfive,jh7110-pwmdac")
+    No driver for Some("i2s@120b0000") ("starfive,jh7110-i2stx0")

etc/aarch64-unknown-none.json

@@ -1,9 +1,9 @@
 {
-  "is-builtin": false,
   "arch": "aarch64",
   "os": "none",
+  "abi": "softfloat",
   "llvm-target": "aarch64-unknown-none",
-  "data-layout": "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32",
+  "data-layout": "e-m:e-p270:32:32-p271:32:32-p272:64:64-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32",
   "max-atomic-width": 128,
   "target-pointer-width": "64",
   "features": "+v8a,+strict-align,-neon,-fp-armv8",
@@ -13,6 +13,7 @@
   "panic-strategy": "abort",
   "dynamic-linking": true,
   "relocation-model": "pic",
+  "position-independent-executables": true,
   "eh-frame-header": false,
 
   "linker": "rust-lld",

etc/aarch64-unknown-qemu.ld

@@ -1,53 +0,0 @@
-ENTRY(__aarch64_entry);
-
-KERNEL_PHYS_BASE = 0x40080000;
-KERNEL_VIRT_OFFSET = 0xFFFFFF8000000000;
-
-SECTIONS {
-    . = KERNEL_PHYS_BASE;
-    PROVIDE(__kernel_start = . + KERNEL_VIRT_OFFSET);
-
-    .text.entry : {
-        *(.text.entry)
-    }
-
-    . = ALIGN(16);
-    . = . + KERNEL_VIRT_OFFSET;
-
-    .text : AT(. - KERNEL_VIRT_OFFSET) {
-        KEEP(*(.text.vectors));
-        *(.text*)
-    }
-
-    . = ALIGN(4K);
-    .rodata : AT(. - KERNEL_VIRT_OFFSET) {
-        *(.eh_frame*)
-        . = ALIGN(16);
-        PROVIDE(__dt_probes_start = .);
-        KEEP(*(.dt_probes));
-        PROVIDE(__dt_probes_end = .);
-        *(.rodata*)
-    }
-
-    . = ALIGN(4K);
-    .data.tables : AT (. - KERNEL_VIRT_OFFSET) {
-        KEEP(*(.data.tables))
-    }
-
-    . = ALIGN(4K);
-    .data : AT(. - KERNEL_VIRT_OFFSET) {
-        *(.data*)
-        *(.got*)
-    }
-
-    . = ALIGN(4K);
-    PROVIDE(__bss_start_phys = . - KERNEL_VIRT_OFFSET);
-    .bss : AT(. - KERNEL_VIRT_OFFSET) {
-        *(COMMON)
-        *(.bss*)
-    }
-    . = ALIGN(4K);
-    PROVIDE(__bss_end_phys = . - KERNEL_VIRT_OFFSET);
-
-    PROVIDE(__kernel_end = .);
-};

etc/dtb/aarch64-qemu-virt.dts

@@ -0,0 +1,433 @@
+/dts-v1/;
+
+/ {
+	interrupt-parent = <0x8005>;
+	dma-coherent;
+	model = "linux,dummy-virt";
+	#size-cells = <0x02>;
+	#address-cells = <0x02>;
+	compatible = "linux,dummy-virt";
+
+	psci {
+		migrate = <0xc4000005>;
+		cpu_on = <0xc4000003>;
+		cpu_off = <0x84000002>;
+		cpu_suspend = <0xc4000001>;
+		method = "smc";
+		compatible = "arm,psci-1.0", "arm,psci-0.2", "arm,psci";
+	};
+
+	memory@40000000 {
+		reg = <0x00 0x40000000 0x00 0x20000000>;
+		device_type = "memory";
+	};
+
+	platform-bus@c000000 {
+		interrupt-parent = <0x8005>;
+		ranges = <0x00 0x00 0xc000000 0x2000000>;
+		#address-cells = <0x01>;
+		#size-cells = <0x01>;
+		compatible = "qemu,platform", "simple-bus";
+	};
+
+	fw-cfg@9020000 {
+		dma-coherent;
+		reg = <0x00 0x9020000 0x00 0x18>;
+		compatible = "qemu,fw-cfg-mmio";
+	};
+
+	virtio_mmio@a000000 {
+		dma-coherent;
+		interrupts = <0x00 0x10 0x01>;
+		reg = <0x00 0xa000000 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a000200 {
+		dma-coherent;
+		interrupts = <0x00 0x11 0x01>;
+		reg = <0x00 0xa000200 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a000400 {
+		dma-coherent;
+		interrupts = <0x00 0x12 0x01>;
+		reg = <0x00 0xa000400 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a000600 {
+		dma-coherent;
+		interrupts = <0x00 0x13 0x01>;
+		reg = <0x00 0xa000600 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a000800 {
+		dma-coherent;
+		interrupts = <0x00 0x14 0x01>;
+		reg = <0x00 0xa000800 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a000a00 {
+		dma-coherent;
+		interrupts = <0x00 0x15 0x01>;
+		reg = <0x00 0xa000a00 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a000c00 {
+		dma-coherent;
+		interrupts = <0x00 0x16 0x01>;
+		reg = <0x00 0xa000c00 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a000e00 {
+		dma-coherent;
+		interrupts = <0x00 0x17 0x01>;
+		reg = <0x00 0xa000e00 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a001000 {
+		dma-coherent;
+		interrupts = <0x00 0x18 0x01>;
+		reg = <0x00 0xa001000 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a001200 {
+		dma-coherent;
+		interrupts = <0x00 0x19 0x01>;
+		reg = <0x00 0xa001200 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a001400 {
+		dma-coherent;
+		interrupts = <0x00 0x1a 0x01>;
+		reg = <0x00 0xa001400 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a001600 {
+		dma-coherent;
+		interrupts = <0x00 0x1b 0x01>;
+		reg = <0x00 0xa001600 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a001800 {
+		dma-coherent;
+		interrupts = <0x00 0x1c 0x01>;
+		reg = <0x00 0xa001800 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a001a00 {
+		dma-coherent;
+		interrupts = <0x00 0x1d 0x01>;
+		reg = <0x00 0xa001a00 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a001c00 {
+		dma-coherent;
+		interrupts = <0x00 0x1e 0x01>;
+		reg = <0x00 0xa001c00 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a001e00 {
+		dma-coherent;
+		interrupts = <0x00 0x1f 0x01>;
+		reg = <0x00 0xa001e00 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a002000 {
+		dma-coherent;
+		interrupts = <0x00 0x20 0x01>;
+		reg = <0x00 0xa002000 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a002200 {
+		dma-coherent;
+		interrupts = <0x00 0x21 0x01>;
+		reg = <0x00 0xa002200 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a002400 {
+		dma-coherent;
+		interrupts = <0x00 0x22 0x01>;
+		reg = <0x00 0xa002400 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a002600 {
+		dma-coherent;
+		interrupts = <0x00 0x23 0x01>;
+		reg = <0x00 0xa002600 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a002800 {
+		dma-coherent;
+		interrupts = <0x00 0x24 0x01>;
+		reg = <0x00 0xa002800 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a002a00 {
+		dma-coherent;
+		interrupts = <0x00 0x25 0x01>;
+		reg = <0x00 0xa002a00 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a002c00 {
+		dma-coherent;
+		interrupts = <0x00 0x26 0x01>;
+		reg = <0x00 0xa002c00 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a002e00 {
+		dma-coherent;
+		interrupts = <0x00 0x27 0x01>;
+		reg = <0x00 0xa002e00 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a003000 {
+		dma-coherent;
+		interrupts = <0x00 0x28 0x01>;
+		reg = <0x00 0xa003000 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a003200 {
+		dma-coherent;
+		interrupts = <0x00 0x29 0x01>;
+		reg = <0x00 0xa003200 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a003400 {
+		dma-coherent;
+		interrupts = <0x00 0x2a 0x01>;
+		reg = <0x00 0xa003400 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a003600 {
+		dma-coherent;
+		interrupts = <0x00 0x2b 0x01>;
+		reg = <0x00 0xa003600 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a003800 {
+		dma-coherent;
+		interrupts = <0x00 0x2c 0x01>;
+		reg = <0x00 0xa003800 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a003a00 {
+		dma-coherent;
+		interrupts = <0x00 0x2d 0x01>;
+		reg = <0x00 0xa003a00 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a003c00 {
+		dma-coherent;
+		interrupts = <0x00 0x2e 0x01>;
+		reg = <0x00 0xa003c00 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	virtio_mmio@a003e00 {
+		dma-coherent;
+		interrupts = <0x00 0x2f 0x01>;
+		reg = <0x00 0xa003e00 0x00 0x200>;
+		compatible = "virtio,mmio";
+	};
+
+	gpio-keys {
+		compatible = "gpio-keys";
+
+		poweroff {
+			gpios = <0x8007 0x03 0x00>;
+			linux,code = <0x74>;
+			label = "GPIO Key Poweroff";
+		};
+	};
+
+	pl061@9030000 {
+		phandle = <0x8007>;
+		clock-names = "apb_pclk";
+		clocks = <0x8000>;
+		interrupts = <0x00 0x07 0x04>;
+		gpio-controller;
+		#gpio-cells = <0x02>;
+		compatible = "arm,pl061", "arm,primecell";
+		reg = <0x00 0x9030000 0x00 0x1000>;
+	};
+
+	pcie@10000000 {
+		interrupt-map-mask = <0x1800 0x00 0x00 0x07>;
+		interrupt-map = <0x00 0x00 0x00 0x01 0x8005 0x00 0x00 0x00 0x03 0x04 0x00 0x00 0x00 0x02 0x8005 0x00 0x00 0x00 0x04 0x04 0x00 0x00 0x00 0x03 0x8005 0x00 0x00 0x00 0x05 0x04 0x00 0x00 0x00 0x04 0x8005 0x00 0x00 0x00 0x06 0x04 0x800 0x00 0x00 0x01 0x8005 0x00 0x00 0x00 0x04 0x04 0x800 0x00 0x00 0x02 0x8005 0x00 0x00 0x00 0x05 0x04 0x800 0x00 0x00 0x03 0x8005 0x00 0x00 0x00 0x06 0x04 0x800 0x00 0x00 0x04 0x8005 0x00 0x00 0x00 0x03 0x04 0x1000 0x00 0x00 0x01 0x8005 0x00 0x00 0x00 0x05 0x04 0x1000 0x00 0x00 0x02 0x8005 0x00 0x00 0x00 0x06 0x04 0x1000 0x00 0x00 0x03 0x8005 0x00 0x00 0x00 0x03 0x04 0x1000 0x00 0x00 0x04 0x8005 0x00 0x00 0x00 0x04 0x04 0x1800 0x00 0x00 0x01 0x8005 0x00 0x00 0x00 0x06 0x04 0x1800 0x00 0x00 0x02 0x8005 0x00 0x00 0x00 0x03 0x04 0x1800 0x00 0x00 0x03 0x8005 0x00 0x00 0x00 0x04 0x04 0x1800 0x00 0x00 0x04 0x8005 0x00 0x00 0x00 0x05 0x04>;
+		#interrupt-cells = <0x01>;
+		ranges = <0x1000000 0x00 0x00 0x00 0x3eff0000 0x00 0x10000 0x2000000 0x00 0x10000000 0x00 0x10000000 0x00 0x2eff0000 0x3000000 0x80 0x00 0x80 0x00 0x80 0x00>;
+		reg = <0x40 0x10000000 0x00 0x10000000>;
+		msi-map = <0x00 0x8006 0x00 0x10000>;
+		dma-coherent;
+		bus-range = <0x00 0xff>;
+		linux,pci-domain = <0x00>;
+		#size-cells = <0x02>;
+		#address-cells = <0x03>;
+		device_type = "pci";
+		compatible = "pci-host-ecam-generic";
+	};
+
+	pl031@9010000 {
+		clock-names = "apb_pclk";
+		clocks = <0x8000>;
+		interrupts = <0x00 0x02 0x04>;
+		reg = <0x00 0x9010000 0x00 0x1000>;
+		compatible = "arm,pl031", "arm,primecell";
+	};
+
+	pl011@9000000 {
+		clock-names = "uartclk", "apb_pclk";
+		clocks = <0x8000 0x8000>;
+		interrupts = <0x00 0x01 0x04>;
+		reg = <0x00 0x9000000 0x00 0x1000>;
+		compatible = "arm,pl011", "arm,primecell";
+	};
+
+	pmu {
+		interrupts = <0x01 0x07 0xf04>;
+		compatible = "arm,armv8-pmuv3";
+	};
+
+	intc@8000000 {
+		phandle = <0x8005>;
+		interrupts = <0x01 0x09 0x04>;
+		reg = <0x00 0x8000000 0x00 0x10000 0x00 0x8010000 0x00 0x10000 0x00 0x8030000 0x00 0x10000 0x00 0x8040000 0x00 0x10000>;
+		compatible = "arm,cortex-a15-gic";
+		ranges;
+		#size-cells = <0x02>;
+		#address-cells = <0x02>;
+		interrupt-controller;
+		#interrupt-cells = <0x03>;
+
+		v2m@8020000 {
+			phandle = <0x8006>;
+			reg = <0x00 0x8020000 0x00 0x1000>;
+			msi-controller;
+			compatible = "arm,gic-v2m-frame";
+		};
+	};
+
+	flash@0 {
+		bank-width = <0x04>;
+		reg = <0x00 0x00 0x00 0x4000000 0x00 0x4000000 0x00 0x4000000>;
+		compatible = "cfi-flash";
+	};
+
+	cpus {
+		#size-cells = <0x00>;
+		#address-cells = <0x01>;
+
+		cpu-map {
+
+			socket0 {
+
+				cluster0 {
+
+					core0 {
+						cpu = <0x8004>;
+					};
+
+					core1 {
+						cpu = <0x8003>;
+					};
+
+					core2 {
+						cpu = <0x8002>;
+					};
+
+					core3 {
+						cpu = <0x8001>;
+					};
+				};
+			};
+		};
+
+		cpu@0 {
+			phandle = <0x8004>;
+			reg = <0x00>;
+			enable-method = "psci";
+			compatible = "arm,cortex-a57";
+			device_type = "cpu";
+		};
+
+		cpu@1 {
+			phandle = <0x8003>;
+			reg = <0x01>;
+			enable-method = "psci";
+			compatible = "arm,cortex-a57";
+			device_type = "cpu";
+		};
+
+		cpu@2 {
+			phandle = <0x8002>;
+			reg = <0x02>;
+			enable-method = "psci";
+			compatible = "arm,cortex-a57";
+			device_type = "cpu";
+		};
+
+		cpu@3 {
+			phandle = <0x8001>;
+			reg = <0x03>;
+			enable-method = "psci";
+			compatible = "arm,cortex-a57";
+			device_type = "cpu";
+		};
+	};
+
+	timer {
+		interrupts = <0x01 0x0d 0xf04 0x01 0x0e 0xf04 0x01 0x0b 0xf04 0x01 0x0a 0xf04 0x01 0x0c 0xf04>;
+		always-on;
+		compatible = "arm,armv8-timer", "arm,armv7-timer";
+	};
+
+	apb-pclk {
+		phandle = <0x8000>;
+		clock-output-names = "clk24mhz";
+		clock-frequency = <0x16e3600>;
+		#clock-cells = <0x00>;
+		compatible = "fixed-clock";
+	};
+
+	aliases {
+		serial0 = "/pl011@9000000";
+	};
+
+	chosen {
+		linux,initrd-end = <0x00 0x49fd4600>;
+		linux,initrd-start = <0x00 0x48000000>;
+		stdout-path = "/pl011@9000000";
+		rng-seed = <0xf119f64b 0xacade219 0xaefd1e87 0x5fb37f65 0xc770054a 0xd779b25f 0x1ba6d6e9 0x8121c19d>;
+		kaslr-seed = <0x1f500308 0xbb36e27a>;
+	};
+};

etc/dtb/riscv64-qemu-virt.dts

@@ -0,0 +1,217 @@
+/dts-v1/;
+
+/ {
+	#address-cells = <0x02>;
+	#size-cells = <0x02>;
+	compatible = "riscv-virtio";
+	model = "riscv-virtio,qemu";
+
+	poweroff {
+		value = <0x5555>;
+		offset = <0x00>;
+		regmap = <0x04>;
+		compatible = "syscon-poweroff";
+	};
+
+	reboot {
+		value = <0x7777>;
+		offset = <0x00>;
+		regmap = <0x04>;
+		compatible = "syscon-reboot";
+	};
+
+	platform-bus@4000000 {
+		interrupt-parent = <0x03>;
+		ranges = <0x00 0x00 0x4000000 0x2000000>;
+		#address-cells = <0x01>;
+		#size-cells = <0x01>;
+		compatible = "qemu,platform", "simple-bus";
+	};
+
+	memory@80000000 {
+		device_type = "memory";
+		reg = <0x00 0x80000000 0x00 0x40000000>;
+	};
+
+	cpus {
+		#address-cells = <0x01>;
+		#size-cells = <0x00>;
+		timebase-frequency = <0x989680>;
+
+		cpu@0 {
+			phandle = <0x01>;
+			device_type = "cpu";
+			reg = <0x00>;
+			status = "okay";
+			compatible = "riscv";
+			riscv,cbop-block-size = <0x40>;
+			riscv,cboz-block-size = <0x40>;
+			riscv,cbom-block-size = <0x40>;
+			riscv,isa-extensions = "i", "m", "a", "f", "d", "c", "h", "zic64b", "zicbom", "zicbop", "zicboz", "ziccamoa", "ziccif", "zicclsm", "ziccrse", "zicntr", "zicsr", "zifencei", "zihintntl", "zihintpause", "zihpm", "zmmul", "za64rs", "zaamo", "zalrsc", "zawrs", "zfa", "zca", "zcd", "zba", "zbb", "zbc", "zbs", "shcounterenw", "shgatpa", "shtvala", "shvsatpa", "shvstvala", "shvstvecd", "ssccptr", "sscounterenw", "sstc", "sstvala", "sstvecd", "ssu64xl", "svadu", "svvptc";
+			riscv,isa-base = "rv64i";
+			riscv,isa = "rv64imafdch_zic64b_zicbom_zicbop_zicboz_ziccamoa_ziccif_zicclsm_ziccrse_zicntr_zicsr_zifencei_zihintntl_zihintpause_zihpm_zmmul_za64rs_zaamo_zalrsc_zawrs_zfa_zca_zcd_zba_zbb_zbc_zbs_shcounterenw_shgatpa_shtvala_shvsatpa_shvstvala_shvstvecd_ssccptr_sscounterenw_sstc_sstvala_sstvecd_ssu64xl_svadu_svvptc";
+			mmu-type = "riscv,sv57";
+
+			interrupt-controller {
+				#interrupt-cells = <0x01>;
+				interrupt-controller;
+				compatible = "riscv,cpu-intc";
+				phandle = <0x02>;
+			};
+		};
+
+		cpu-map {
+
+			cluster0 {
+
+				core0 {
+					cpu = <0x01>;
+				};
+			};
+		};
+	};
+
+	pmu {
+		riscv,event-to-mhpmcounters = <0x01 0x01 0x7fff9 0x02 0x02 0x7fffc 0x10019 0x10019 0x7fff8 0x1001b 0x1001b 0x7fff8 0x10021 0x10021 0x7fff8>;
+		compatible = "riscv,pmu";
+	};
+
+	fw-cfg@10100000 {
+		dma-coherent;
+		reg = <0x00 0x10100000 0x00 0x18>;
+		compatible = "qemu,fw-cfg-mmio";
+	};
+
+	flash@20000000 {
+		bank-width = <0x04>;
+		reg = <0x00 0x20000000 0x00 0x2000000 0x00 0x22000000 0x00 0x2000000>;
+		compatible = "cfi-flash";
+	};
+
+	aliases {
+		serial0 = "/soc/serial@10000000";
+	};
+
+	chosen {
+		linux,initrd-end = <0x00 0xa2b4f200>;
+		linux,initrd-start = <0x00 0xa0200000>;
+		stdout-path = "/soc/serial@10000000";
+		rng-seed = <0xa7074b10 0xf3373c0c 0x94a3a9a0 0xa2442477 0x817e30af 0x6460a6d7 0xbcaa71c4 0xb75dd35>;
+	};
+
+	soc {
+		#address-cells = <0x02>;
+		#size-cells = <0x02>;
+		compatible = "simple-bus";
+		ranges;
+
+		rtc@101000 {
+			interrupts = <0x0b>;
+			interrupt-parent = <0x03>;
+			reg = <0x00 0x101000 0x00 0x1000>;
+			compatible = "google,goldfish-rtc";
+		};
+
+		serial@10000000 {
+			interrupts = <0x0a>;
+			interrupt-parent = <0x03>;
+			clock-frequency = "", "8@";
+			reg = <0x00 0x10000000 0x00 0x100>;
+			compatible = "ns16550a";
+		};
+
+		test@100000 {
+			phandle = <0x04>;
+			reg = <0x00 0x100000 0x00 0x1000>;
+			compatible = "sifive,test1", "sifive,test0", "syscon";
+		};
+
+		virtio_mmio@10008000 {
+			interrupts = <0x08>;
+			interrupt-parent = <0x03>;
+			reg = <0x00 0x10008000 0x00 0x1000>;
+			compatible = "virtio,mmio";
+		};
+
+		virtio_mmio@10007000 {
+			interrupts = <0x07>;
+			interrupt-parent = <0x03>;
+			reg = <0x00 0x10007000 0x00 0x1000>;
+			compatible = "virtio,mmio";
+		};
+
+		virtio_mmio@10006000 {
+			interrupts = <0x06>;
+			interrupt-parent = <0x03>;
+			reg = <0x00 0x10006000 0x00 0x1000>;
+			compatible = "virtio,mmio";
+		};
+
+		virtio_mmio@10005000 {
+			interrupts = <0x05>;
+			interrupt-parent = <0x03>;
+			reg = <0x00 0x10005000 0x00 0x1000>;
+			compatible = "virtio,mmio";
+		};
+
+		virtio_mmio@10004000 {
+			interrupts = <0x04>;
+			interrupt-parent = <0x03>;
+			reg = <0x00 0x10004000 0x00 0x1000>;
+			compatible = "virtio,mmio";
+		};
+
+		virtio_mmio@10003000 {
+			interrupts = <0x03>;
+			interrupt-parent = <0x03>;
+			reg = <0x00 0x10003000 0x00 0x1000>;
+			compatible = "virtio,mmio";
+		};
+
+		virtio_mmio@10002000 {
+			interrupts = <0x02>;
+			interrupt-parent = <0x03>;
+			reg = <0x00 0x10002000 0x00 0x1000>;
+			compatible = "virtio,mmio";
+		};
+
+		virtio_mmio@10001000 {
+			interrupts = <0x01>;
+			interrupt-parent = <0x03>;
+			reg = <0x00 0x10001000 0x00 0x1000>;
+			compatible = "virtio,mmio";
+		};
+
+		plic@c000000 {
+			phandle = <0x03>;
+			riscv,ndev = <0x5f>;
+			reg = <0x00 0xc000000 0x00 0x600000>;
+			interrupts-extended = <0x02 0x0b 0x02 0x09>;
+			interrupt-controller;
+			compatible = "sifive,plic-1.0.0", "riscv,plic0";
+			#address-cells = <0x00>;
+			#interrupt-cells = <0x01>;
+		};
+
+		clint@2000000 {
+			interrupts-extended = <0x02 0x03 0x02 0x07>;
+			reg = <0x00 0x2000000 0x00 0x10000>;
+			compatible = "sifive,clint0", "riscv,clint0";
+		};
+
+		pci@30000000 {
+			interrupt-map-mask = <0x1800 0x00 0x00 0x07>;
+			interrupt-map = <0x00 0x00 0x00 0x01 0x03 0x20 0x00 0x00 0x00 0x02 0x03 0x21 0x00 0x00 0x00 0x03 0x03 0x22 0x00 0x00 0x00 0x04 0x03 0x23 0x800 0x00 0x00 0x01 0x03 0x21 0x800 0x00 0x00 0x02 0x03 0x22 0x800 0x00 0x00 0x03 0x03 0x23 0x800 0x00 0x00 0x04 0x03 0x20 0x1000 0x00 0x00 0x01 0x03 0x22 0x1000 0x00 0x00 0x02 0x03 0x23 0x1000 0x00 0x00 0x03 0x03 0x20 0x1000 0x00 0x00 0x04 0x03 0x21 0x1800 0x00 0x00 0x01 0x03 0x23 0x1800 0x00 0x00 0x02 0x03 0x20 0x1800 0x00 0x00 0x03 0x03 0x21 0x1800 0x00 0x00 0x04 0x03 0x22>;
+			ranges = <0x1000000 0x00 0x00 0x00 0x3000000 0x00 0x10000 0x2000000 0x00 0x40000000 0x00 0x40000000 0x00 0x40000000 0x3000000 0x04 0x00 0x04 0x00 0x04 0x00>;
+			reg = <0x00 0x30000000 0x00 0x10000000>;
+			dma-coherent;
+			bus-range = <0x00 0xff>;
+			linux,pci-domain = <0x00>;
+			device_type = "pci";
+			compatible = "pci-host-ecam-generic";
+			#size-cells = <0x02>;
+			#interrupt-cells = <0x01>;
+			#address-cells = <0x03>;
+		};
+	};
+};

etc/i686-unknown-none.json

@@ -1,29 +0,0 @@
-{
-  "is-builtin": false,
-  "arch": "x86",
-  "cpu": "pentium4",
-  "os": "none",
-  "llvm-target": "i686-unknown-linux-gnu",
-  "data-layout": "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-i128:128-f64:32:64-f80:32-n8:16:32-S128",
-  "max-atomic-width": 64,
-  "target-pointer-width": "32",
-  "features": "-avx,-sse,+soft-float",
-
-  "executables": true,
-  "stack-probes": {
-    "kind": "inline"
-  },
-  "dynamic-linking": true,
-  "panic-strategy": "abort",
-  "relocation-model": "pic",
-
-  "has-thread-local": false,
-
-  "supported-split-debuginfo": [
-    "packed",
-    "unpacked",
-    "off"
-  ],
-  "linker": "rust-lld",
-  "linker-flavor": "ld.lld"
-}

etc/i686-unknown-none.ld

@@ -1,53 +0,0 @@
-ENTRY(__i686_entry);
-
-KERNEL_PHYS_BASE = 0x100000;
-KERNEL_VIRT_OFFSET = 0xC0000000;
-
-SECTIONS {
-    . = KERNEL_PHYS_BASE;
-    PROVIDE(__kernel_start = . + KERNEL_VIRT_OFFSET);
-
-    .text.entry : {
-        KEEP(*(.multiboot))
-        *(.text.entry)
-    }
-
-    . = ALIGN(16);
-    . = . + KERNEL_VIRT_OFFSET;
-
-    .text : AT(. - KERNEL_VIRT_OFFSET) {
-        *(.text*)
-    }
-
-    .export.text : AT(. - KERNEL_VIRT_OFFSET) {
-        KEEP(*(.export.text*))
-    }
-
-    . = ALIGN(4K);
-    .rodata : AT(. - KERNEL_VIRT_OFFSET) {
-        *(.eh_frame*)
-        *(.rodata*)
-    }
-
-    . = ALIGN(4K);
-    .data.tables : AT (. - KERNEL_VIRT_OFFSET) {
-        KEEP(*(.data.tables))
-    }
-
-    .data : AT(. - KERNEL_VIRT_OFFSET) {
-        KEEP(*(.data.yboot))
-        *(.data*)
-        *(.got*)
-    }
-
-    . = ALIGN(4K);
-    PROVIDE(__bss_start_phys = . - KERNEL_VIRT_OFFSET);
-    .bss : AT(. - KERNEL_VIRT_OFFSET) {
-        *(COMMON)
-        *(.bss*)
-    }
-    . = ALIGN(4K);
-    PROVIDE(__bss_end_phys = . - KERNEL_VIRT_OFFSET);
-
-    PROVIDE(__kernel_end = .);
-};

etc/ld/aarch64-unknown-none.ld

@@ -0,0 +1,75 @@
+ENTRY(__aarch64_entry);
+
+SECTIONS {
+    . = 0x0;
+
+    PROVIDE(__kernel_start = .);
+
+    .text : {
+        *(.text.entry)
+        *(.text*)
+    }
+
+    . = ALIGN(4K);
+
+    .rodata : {
+        *(.rodata*)
+        *(.eh_frame*)
+
+        . = ALIGN(16);
+        PROVIDE(__init_array_start = .);
+        KEEP(*(.init_array*));
+        PROVIDE(__init_array_end = .);
+    }
+
+    . = ALIGN(4K);
+
+    .rela : {
+        PROVIDE(__rela_start = .);
+        *(.rela*)
+        PROVIDE(__rela_end = .);
+    }
+
+    .got : {
+        *(.got*)
+    }
+
+    .dynamic : {
+        *(.dynamic)
+    }
+
+    . = ALIGN(4K);
+
+    .data : {
+        *(.data*)
+    }
+
+    . = ALIGN(4K);
+    PROVIDE(__bss_start = .);
+
+    .bss : {
+        *(COMMON)
+        *(.bss*)
+    }
+
+    . = ALIGN(4K);
+    PROVIDE(__bss_end = .);
+    PROVIDE(__kernel_end = .);
+    PROVIDE(__kernel_size = __kernel_end - __kernel_start);
+
+    .dynsym : {
+        *(.dynsym)
+    }
+
+    .gnu.hash : {
+        *(.gnu.hash)
+    }
+
+    .hash : {
+        *(.hash)
+    }
+
+    .dynstr : {
+        *(.dynstr)
+    }
+}

etc/ld/riscv64-unknown-none.ld

@@ -0,0 +1,75 @@
+ENTRY(__riscv64_entry);
+
+SECTIONS {
+    . = 0x0;
+
+    PROVIDE(__kernel_start = .);
+
+    .text : {
+        *(.text.entry)
+        *(.text*)
+    }
+
+    . = ALIGN(4K);
+
+    .rodata : {
+        *(.rodata*)
+        *(.eh_frame*)
+
+        . = ALIGN(16);
+        PROVIDE(__init_array_start = .);
+        KEEP(*(.init_array*));
+        PROVIDE(__init_array_end = .);
+    }
+
+    . = ALIGN(4K);
+
+    .rela : {
+        PROVIDE(__rela_start = .);
+        *(.rela*)
+        PROVIDE(__rela_end = .);
+    }
+
+    .got : {
+        *(.got*)
+    }
+
+    .dynamic : {
+        *(.dynamic)
+    }
+
+    . = ALIGN(4K);
+
+    .data : {
+        *(.data*)
+    }
+
+    . = ALIGN(4K);
+    PROVIDE(__bss_start = .);
+
+    .bss : {
+        *(COMMON)
+        *(.bss*)
+    }
+
+    . = ALIGN(4K);
+    PROVIDE(__bss_end = .);
+    PROVIDE(__kernel_end = .);
+    PROVIDE(__kernel_size = __kernel_end - __kernel_start);
+
+    .dynsym : {
+        *(.dynsym)
+    }
+
+    .gnu.hash : {
+        *(.gnu.hash)
+    }
+
+    .hash : {
+        *(.hash)
+    }
+
+    .dynstr : {
+        *(.dynstr)
+    }
+}

etc/ld/x86_64-unknown-none.ld

@@ -0,0 +1,73 @@
+ENTRY(__x86_64_entry);
+
+SECTIONS {
+    . = 0x0;
+
+    PROVIDE(__kernel_start = .);
+
+    .text : {
+        *(.text.entry)
+        *(.text*)
+    }
+
+    . = ALIGN(4K);
+
+    .rodata : {
+        *(.rodata*)
+        *(.eh_frame*)
+
+        . = ALIGN(16);
+        PROVIDE(__init_array_start = .);
+        KEEP(*(.init_array*));
+        PROVIDE(__init_array_end = .);
+    }
+
+    . = ALIGN(4K);
+
+    .rela : {
+        PROVIDE(__rela_start = .);
+        *(.rela*)
+        PROVIDE(__rela_end = .);
+    }
+
+    .dynamic : {
+        *(.dynamic)
+    }
+
+    . = ALIGN(4K);
+
+    .data : {
+        KEEP(*(.data.yboot))
+        *(.data*)
+        *(.got*)
+    }
+
+    . = ALIGN(4K);
+    PROVIDE(__bss_start = .);
+
+    .bss : {
+        *(COMMON)
+        *(.bss*)
+    }
+
+    . = ALIGN(4K);
+    PROVIDE(__bss_end = .);
+    PROVIDE(__kernel_end = .);
+    PROVIDE(__kernel_size = __kernel_end - __kernel_start);
+
+    .dynsym : {
+        *(.dynsym)
+    }
+
+    .gnu.hash : {
+        *(.gnu.hash)
+    }
+
+    .hash : {
+        *(.hash)
+    }
+
+    .dynstr : {
+        *(.dynstr)
+    }
+}

etc/riscv64-unknown-none.json

@@ -0,0 +1,27 @@
+{
+  "arch": "riscv64",
+  "os": "none",
+  "abi": "softfloat",
+  "cpu": "generic-rv64",
+  "llvm-target": "riscv64",
+  "data-layout": "e-m:e-p:64:64-i64:64-i128:128-n32:64-S128",
+  "max-atomic-width": 64,
+  "target-pointer-width": "64",
+  "features": "+m,+a,+c",
+
+  "disable-redzone": true,
+  "executables": true,
+  "panic-strategy": "abort",
+  "dynamic-linking": true,
+  "relocation-model": "pic",
+  "position-independent-executables": true,
+  "code-model": "medium",
+  "eh-frame-header": false,
+
+  "crt-objects-fallback": "false",
+  "emit-debug-gdb-scripts": false,
+  "llvm-abiname": "lp64",
+
+  "linker": "rust-lld",
+  "linker-flavor": "ld.lld"
+}

etc/x86_64-unknown-none.json

@@ -1,19 +1,21 @@
 {
-  "is-builtin": false,
   "arch": "x86_64",
   "cpu": "x86-64",
   "os": "none",
+  "abi": "softfloat",
+  "rustc-abi": "x86-softfloat",
   "llvm-target": "x86_64-unknown-linux-gnu",
   "data-layout": "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128",
   "max-atomic-width": 64,
   "target-pointer-width": "64",
-  "features": "-avx,-sse,+soft-float",
+  "features": "-avx,-sse,-avx2,+soft-float",
 
   "disable-redzone": true,
   "executables": true,
   "panic-strategy": "abort",
   "dynamic-linking": true,
   "relocation-model": "pic",
+  "position-independent-executables": true,
 
   "has-thread-local": false,
 

etc/x86_64-unknown-none.ld

@@ -1,52 +0,0 @@
-ENTRY(__x86_64_entry);
-
-KERNEL_PHYS_BASE = 0x200000;
-KERNEL_VIRT_OFFSET = 0xFFFFFF8000000000;
-
-SECTIONS {
-    . = KERNEL_PHYS_BASE;
-    PROVIDE(__kernel_start = . + KERNEL_VIRT_OFFSET);
-
-    .text.entry : {
-        *(.text.entry)
-    }
-
-    . = ALIGN(16);
-    . = . + KERNEL_VIRT_OFFSET;
-
-    .text : AT(. - KERNEL_VIRT_OFFSET) {
-        *(.text*)
-    }
-
-    .export.text : AT(. - KERNEL_VIRT_OFFSET) {
-        KEEP(*(.export.text*))
-    }
-
-    . = ALIGN(4K);
-    .rodata : AT(. - KERNEL_VIRT_OFFSET) {
-        *(.eh_frame*)
-        *(.rodata*)
-    }
-
-    . = ALIGN(4K);
-    .data.tables : AT (. - KERNEL_VIRT_OFFSET) {
-        KEEP(*(.data.tables))
-    }
-
-    .data : AT(. - KERNEL_VIRT_OFFSET) {
-        KEEP(*(.data.yboot))
-        *(.data*)
-        *(.got*)
-    }
-
-    . = ALIGN(4K);
-    PROVIDE(__bss_start_phys = . - KERNEL_VIRT_OFFSET);
-    .bss : AT(. - KERNEL_VIRT_OFFSET) {
-        *(COMMON)
-        *(.bss*)
-    }
-    . = ALIGN(4K);
-    PROVIDE(__bss_end_phys = . - KERNEL_VIRT_OFFSET);
-
-    PROVIDE(__kernel_end = .);
-};

kernel/Cargo.toml

@@ -7,19 +7,18 @@ authors = ["Mark Poliakov <mark@alnyan.me>"]
 
 [dependencies]
 abi-lib.workspace = true
+abi-serde.workspace = true
 yggdrasil-abi.workspace = true
 kernel-arch-interface.workspace = true
 libk.workspace = true
 libk-util.workspace = true
 libk-mm.workspace = true
-libk-device.workspace = true
 elf.workspace = true
 chrono.workspace = true
 
 device-api = { workspace = true, features = ["derive"] }
 device-api-macros.workspace = true
 
-memtables.workspace = true
 vmalloc.workspace = true
 kernel-arch.workspace = true
 
@@ -29,13 +28,17 @@ ygg_driver_usb = { path = "driver/bus/usb" }
 ygg_driver_net_core = { path = "driver/net/core" }
 ygg_driver_net_loopback = { path = "driver/net/loopback" }
 ygg_driver_virtio_net = { path = "driver/virtio/net", features = ["pci"] }
+ygg_driver_virtio_gpu = { path = "driver/virtio/gpu", features = ["pci"] }
+ygg_driver_virtio_blk = { path = "driver/virtio/blk", features = ["pci"] }
+ygg_driver_nvme = { path = "driver/block/nvme" }
 ygg_driver_ahci = { path = "driver/block/ahci" }
-ygg_driver_usb_xhci = { path = "driver/usb/xhci" }
 ygg_driver_input = { path = "driver/input" }
+ygg_driver_usb_xhci.path = "driver/usb/xhci"
+ygg_driver_net_rtl81xx.path = "driver/net/rtl81xx"
 
-kernel-fs = { path = "driver/fs/kernel-fs" }
 memfs = { path = "driver/fs/memfs" }
 ext2 = { path = "driver/fs/ext2" }
+ygg_driver_fat32.path = "driver/fs/fat32"
 
 log.workspace = true
 bitflags.workspace = true
@@ -45,6 +48,7 @@ bytemuck.workspace = true
 futures-util.workspace = true
 crossbeam-queue.workspace = true
 async-trait.workspace = true
+cfg-if.workspace = true
 
 git-version = "0.3.9"
 
@@ -52,21 +56,26 @@ git-version = "0.3.9"
 aarch64-cpu.workspace = true
 device-tree.workspace = true
 kernel-arch-aarch64.workspace = true
+ygg_driver_bsp_arm.path = "driver/bsp/arm"
+ygg_driver_bsp_bcm283x.path = "driver/bsp/bcm283x"
+
+[target.'cfg(target_arch = "riscv64")'.dependencies]
+device-tree.workspace = true
+kernel-arch-riscv64.workspace = true
+ygg_driver_bsp_arm.path = "driver/bsp/arm"  # PrimeCell components
+ygg_driver_bsp_riscv.path = "driver/bsp/riscv"
+ygg_driver_net_stmmac.path = "driver/net/stmmac"
+ygg_driver_bsp_jh7110.path = "driver/bsp/jh7110"
 
 [target.'cfg(target_arch = "x86_64")'.dependencies]
 yboot-proto.workspace = true
 kernel-arch-x86_64.workspace = true
 kernel-arch-x86.workspace = true
 
-ygg_driver_nvme = { path = "driver/block/nvme" }
+ygg_driver_acpi.path = "driver/acpi"
+ygg_driver_net_igbe.path = "driver/net/igbe"
 
 acpi.workspace = true
-aml.workspace = true
-acpi-system.workspace = true
-
-[target.'cfg(target_arch = "x86")'.dependencies]
-kernel-arch-i686.workspace = true
-kernel-arch-x86.workspace = true
 
 [build-dependencies]
 abi-generator.workspace = true
@@ -78,9 +87,16 @@ prettyplease = "0.2.15"
 aarch64-cpu.workspace = true
 device-tree.workspace = true
 kernel-arch-x86_64.workspace = true
-kernel-arch-i686.workspace = true
 kernel-arch-x86.workspace = true
 kernel-arch-aarch64.workspace = true
+kernel-arch-riscv64.workspace = true
+
+ygg_driver_acpi.path = "driver/acpi"
+ygg_driver_bsp_arm.path = "driver/bsp/arm"
+ygg_driver_bsp_riscv.path = "driver/bsp/riscv"
+ygg_driver_net_stmmac.path = "driver/net/stmmac"
+ygg_driver_bsp_bcm283x.path = "driver/bsp/bcm283x"
+ygg_driver_bsp_jh7110.path = "driver/bsp/jh7110"
 
 [features]
 default = ["fb_console"]

kernel/arch/Cargo.toml

@@ -3,21 +3,19 @@ name = "kernel-arch"
 version = "0.1.0"
 edition = "2021"
 
-# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
-
 [target.'cfg(all(target_os = "none", target_arch = "x86_64"))'.dependencies]
-kernel-arch-x86_64 = { path = "x86_64" }
+kernel-arch-x86_64.path = "x86_64"
 
 [target.'cfg(all(target_os = "none", target_arch = "aarch64"))'.dependencies]
-kernel-arch-aarch64 = { path = "aarch64" }
+kernel-arch-aarch64.path = "aarch64"
 
-[target.'cfg(all(target_os = "none", target_arch = "x86"))'.dependencies]
-kernel-arch-i686 = { path = "i686" }
+[target.'cfg(all(target_os = "none", target_arch = "riscv64"))'.dependencies]
+kernel-arch-riscv64.path = "riscv64"
 
 [target.'cfg(not(target_os = "none"))'.dependencies]
-kernel-arch-hosted = { path = "hosted" }
+kernel-arch-hosted.path = "hosted"
 
 [dependencies]
-kernel-arch-interface = { path = "interface" }
+kernel-arch-interface.path = "interface"
 
 cfg-if.workspace = true

kernel/arch/aarch64/Cargo.toml

@@ -7,7 +7,6 @@ edition = "2021"
 yggdrasil-abi.workspace = true
 kernel-arch-interface.workspace = true
 libk-mm-interface.workspace = true
-memtables.workspace = true
 device-api = { workspace = true, features = ["derive"] }
 
 bitflags.workspace = true
@@ -17,3 +16,6 @@ tock-registers.workspace = true
 
 [build-dependencies]
 cc = "1.0"
+
+[lints]
+workspace = true

kernel/arch/aarch64/build.rs

@@ -5,7 +5,7 @@ fn build_fp_context_obj() {
 
     let out_dir = env::var("OUT_DIR").unwrap();
 
-    println!("cargo:rerun-if-changed={}", FP_CONTEXT_S);
+    println!("cargo:rerun-if-changed={FP_CONTEXT_S}");
 
     cc::Build::new()
         .out_dir(&out_dir)

kernel/arch/aarch64/src/context.S

@@ -23,6 +23,8 @@
 .endm
 
 .macro LOAD_TASK_STATE
+    dsb ishst
+
     // x19 == tpidr_el0, x20 = ttbr0_el1
     ldp x19, x20, [sp, #16 * 6]
     msr tpidr_el0, x19
@@ -36,6 +38,12 @@
     ldp x29, x30, [sp, #16 * 5]
 
     add sp, sp, #{context_size}
+
+    isb sy
+    tlbi vmalle1is
+    ic iallu
+    dsb ish
+    isb sy
 .endm
 
 __aarch64_task_enter_kernel:
@@ -87,7 +95,7 @@ __aarch64_task_enter_user:
 
     mov lr, xzr
 
-    dmb ish
+    dsb ish
     isb sy
 
     eret

kernel/arch/aarch64/src/context.rs

@@ -208,10 +208,12 @@ impl<K: KernelTableManager, PA: PhysicalMemoryAllocator<Address = PhysicalAddres
         stack.push(mdscr_el1);
         stack.push(context.stack_pointer);
 
+        let ttbr0 = context.address_space | (context.asid << 48) | 1;
+
         setup_common_context(
             &mut stack,
             __aarch64_task_enter_user as _,
-            context.address_space,
+            ttbr0,
             context.thread_pointer as _,
         );
 

kernel/arch/aarch64/src/lib.rs

@@ -1,16 +1,19 @@
 #![no_std]
-#![feature(naked_functions, trait_upcasting)]
+#![feature(decl_macro)]
 #![allow(clippy::new_without_default)]
 
 extern crate alloc;
 
 use core::sync::atomic::{AtomicUsize, Ordering};
 
-use aarch64_cpu::registers::{DAIF, MPIDR_EL1, TPIDR_EL1};
-use alloc::{boxed::Box, vec::Vec};
-use device_api::interrupt::{LocalInterruptController, MessageInterruptController};
+use aarch64_cpu::{
+    asm::barrier,
+    registers::{DAIF, MPIDR_EL1, TPIDR_EL1},
+};
+use alloc::{boxed::Box, sync::Arc, vec::Vec};
+use device_api::interrupt::LocalInterruptController;
 use kernel_arch_interface::{
-    cpu::{CpuImpl, IpiQueue},
+    cpu::{CpuData, CpuImpl, IpiQueue},
     guard::IrqGuard,
     task::Scheduler,
     util::OneTimeInit,
@@ -29,20 +32,21 @@ pub struct ArchitectureImpl;
 pub trait GicInterface: LocalInterruptController {}
 
 pub struct PerCpuData {
-    pub gic: OneTimeInit<&'static dyn GicInterface>,
+    pub gic: OneTimeInit<Arc<dyn GicInterface>>,
 }
 
+impl CpuData for PerCpuData {}
+
 static IPI_QUEUES: OneTimeInit<Vec<IpiQueue<ArchitectureImpl>>> = OneTimeInit::new();
 pub static CPU_COUNT: AtomicUsize = AtomicUsize::new(1);
 
-#[naked]
+#[unsafe(naked)]
 extern "C" fn idle_task(_: usize) -> ! {
-    unsafe {
-        core::arch::naked_asm!("1: nop; b 1b");
-    }
+    core::arch::naked_asm!("1: nop; b 1b");
 }
 
 impl ArchitectureImpl {
+    #[inline]
     pub fn local_cpu_data() -> Option<&'static mut PerCpuData> {
         unsafe { (Self::local_cpu() as *mut PerCpuData).as_mut() }
     }
@@ -63,6 +67,7 @@ impl Architecture for ArchitectureImpl {
         DAIF.read(DAIF::I) != 0
     }
 
+    #[inline(never)]
     unsafe fn set_interrupt_mask(mask: bool) -> bool {
         let old = Self::interrupt_mask();
         if mask {
@@ -120,13 +125,7 @@ impl Architecture for ArchitectureImpl {
     }
 
     fn local_interrupt_controller() -> Option<&'static dyn LocalInterruptController> {
-        let local = Self::local_cpu_data()?;
-        let intc = *local.gic.try_get()?;
-        Some(intc)
-    }
-
-    fn message_interrupt_controller() -> &'static dyn MessageInterruptController {
-        todo!()
+        None
     }
 
     fn cpu_available_features<S: Scheduler>(_cpu: &CpuImpl<Self, S>) -> Option<&Self::CpuFeatures> {
@@ -136,4 +135,30 @@ impl Architecture for ArchitectureImpl {
     fn cpu_enabled_features<S: Scheduler>(_cpu: &CpuImpl<Self, S>) -> Option<&Self::CpuFeatures> {
         None
     }
+
+    // Cache/barrier operation
+    fn load_barrier() {
+        barrier::dmb(barrier::ISHLD);
+    }
+
+    fn store_barrier() {
+        barrier::dmb(barrier::ISHST);
+    }
+
+    fn memory_barrier() {
+        barrier::dsb(barrier::SY);
+    }
+
+    fn flush_virtual_range(range: core::ops::Range<usize>) {
+        // TODO cache line assumed to be 64 bytes
+        const CLSIZE: usize = 64;
+        let start = range.start & !(CLSIZE - 1);
+        let end = (range.end + (CLSIZE - 1)) & !(CLSIZE - 1);
+
+        for line in (start..end).step_by(CLSIZE) {
+            unsafe {
+                core::arch::asm!("dc ivac, {address}", address = in(reg) line);
+            }
+        }
+    }
 }

kernel/arch/aarch64/src/mem/fixed.rs

@@ -0,0 +1,124 @@
+use core::ops::Range;
+
+use aarch64_cpu::registers::{TTBR0_EL1, TTBR1_EL1};
+use kernel_arch_interface::{mem::DeviceMemoryAttributes, KERNEL_VIRT_OFFSET};
+use libk_mm_interface::{
+    address::PhysicalAddress,
+    device::{DevicePageManager, DevicePageTableLevel},
+    table::EntryLevel,
+};
+
+use crate::mem::{
+    auto_lower_address,
+    table::{PageAttributes, PageEntry, PageTable, L1, L2, L3},
+    tlb_flush_range_va,
+};
+
+pub const IDENTITY_SIZE_L1: usize = 8;
+pub const DEVICE_L1: usize = IDENTITY_SIZE_L1;
+pub const DEVICE_MAPPING_L3_COUNT: usize = 32;
+pub const DEVICE_MAPPING_OFFSET: usize = KERNEL_VIRT_OFFSET + (DEVICE_L1 << L1::SHIFT);
+
+#[repr(transparent)]
+pub struct L2DeviceMemory(pub PageTable<L2>);
+#[repr(transparent)]
+pub struct L3DeviceMemory(pub [PageTable<L3>; DEVICE_MAPPING_L3_COUNT]);
+
+static mut KERNEL_L1: PageTable<L1> = PageTable::zeroed();
+pub(super) static mut DEVICE_MEMORY: DevicePageManager<L3DeviceMemory, L2DeviceMemory> =
+    DevicePageManager::new(
+        L3DeviceMemory([PageTable::zeroed(); DEVICE_MAPPING_L3_COUNT]),
+        L2DeviceMemory(PageTable::zeroed()),
+    );
+
+impl DevicePageTableLevel for L2DeviceMemory {
+    type Level = L2;
+    const VIRTUAL_BASE: usize = DEVICE_MAPPING_OFFSET;
+    const INDEX_RANGE: Range<usize> = DEVICE_MAPPING_L3_COUNT..512;
+
+    fn map_page(
+        &mut self,
+        index: usize,
+        physical: PhysicalAddress,
+        attrs: &DeviceMemoryAttributes,
+    ) {
+        let _ = attrs;
+        self.0[index] = PageEntry::device_block(physical);
+    }
+
+    fn unmap_page(&mut self, index: usize) {
+        self.0[index] = PageEntry::INVALID;
+    }
+
+    fn is_mapped(&self, index: usize) -> bool {
+        self.0[index].is_present()
+    }
+
+    fn flush_range(range: Range<usize>) {
+        let start = range.start * L2::SIZE + Self::VIRTUAL_BASE;
+        let size = (range.end - range.start) * L2::SIZE;
+        tlb_flush_range_va(start, size);
+    }
+}
+
+impl DevicePageTableLevel for L3DeviceMemory {
+    type Level = L3;
+    const VIRTUAL_BASE: usize = DEVICE_MAPPING_OFFSET;
+    const INDEX_RANGE: Range<usize> = 0..512 * DEVICE_MAPPING_L3_COUNT;
+
+    fn map_page(
+        &mut self,
+        index: usize,
+        physical: PhysicalAddress,
+        attrs: &DeviceMemoryAttributes,
+    ) {
+        let _ = attrs;
+        let l2i = index / 512;
+        let l3i = index % 512;
+        self.0[l2i][l3i] = PageEntry::device_page(physical);
+    }
+
+    fn unmap_page(&mut self, index: usize) {
+        let l2i = index / 512;
+        let l3i = index % 512;
+        self.0[l2i][l3i] = PageEntry::INVALID;
+    }
+
+    fn is_mapped(&self, index: usize) -> bool {
+        let l2i = index / 512;
+        let l3i = index % 512;
+        self.0[l2i][l3i].is_present()
+    }
+
+    fn flush_range(range: Range<usize>) {
+        let start = range.start * L3::SIZE + Self::VIRTUAL_BASE;
+        let size = (range.end - range.start) * L3::SIZE;
+        tlb_flush_range_va(start, size);
+    }
+}
+
+pub unsafe fn setup() {
+    // 0..IDENTITY_SIZE_L1 -> lower RAM region
+    for i in 0..IDENTITY_SIZE_L1 {
+        let phys = PhysicalAddress::from_usize(i << L1::SHIFT);
+        KERNEL_L1[i] = PageEntry::normal_block(phys, PageAttributes::empty());
+    }
+
+    // DEVICE_L1 -> Device L2 table
+    //               0..DEVICE_MAPPING_L3_COUNT -> Device L3 tables -> Device L3 pages
+    //               ..512                      -> Device L2 pages
+    for i in 0..DEVICE_MAPPING_L3_COUNT {
+        let phys =
+            PhysicalAddress::from_usize(auto_lower_address(&raw const DEVICE_MEMORY.normal.0[i]));
+        DEVICE_MEMORY.large.0[i] = PageEntry::table(phys, PageAttributes::empty());
+    }
+    let phys = PhysicalAddress::from_usize(auto_lower_address(&raw const DEVICE_MEMORY.large.0));
+    KERNEL_L1[DEVICE_L1] = PageEntry::table(phys, PageAttributes::empty());
+}
+
+pub unsafe fn load() {
+    let ttbr_physical = auto_lower_address(&raw const KERNEL_L1) as u64;
+
+    TTBR0_EL1.set_baddr(ttbr_physical);
+    TTBR1_EL1.set_baddr(ttbr_physical);
+}

kernel/arch/aarch64/src/mem/intrinsics.rs

@@ -0,0 +1,141 @@
+use core::sync::atomic::{self, Ordering};
+
+use aarch64_cpu::{
+    asm::barrier,
+    registers::{PAR_EL1, SCTLR_EL1},
+};
+use libk_mm_interface::table::{EntryLevel, EntryLevelExt};
+use tock_registers::interfaces::{ReadWriteable, Readable};
+
+use crate::mem::table::L3;
+
+/// Enables data cache.
+///
+/// # Safety
+///
+/// Manipulates low-level machine state, use with care.
+pub unsafe fn enable_dcache() {
+    barrier::dsb(barrier::ISHST);
+    barrier::isb(barrier::SY);
+
+    SCTLR_EL1.modify(SCTLR_EL1::C::Cacheable);
+
+    barrier::dsb(barrier::ISH);
+    barrier::isb(barrier::SY);
+}
+
+/// Enables instruction cache.
+///
+/// # Safety
+///
+/// Manipulates low-level machine state, use with care.
+pub unsafe fn enable_icache() {
+    barrier::isb(barrier::SY);
+
+    SCTLR_EL1.modify(SCTLR_EL1::I::Cacheable);
+
+    barrier::dsb(barrier::ISH);
+    barrier::isb(barrier::SY);
+}
+
+/// Disables instruction cache.
+///
+/// # Safety
+///
+/// Manipulates low-level machine state, use with care. Might break some instructions.
+pub unsafe fn disable_icache() {
+    barrier::isb(barrier::SY);
+
+    ic_iallu();
+    SCTLR_EL1.modify(SCTLR_EL1::I::NonCacheable);
+
+    barrier::dsb(barrier::ISH);
+    barrier::isb(barrier::SY);
+}
+
+#[inline]
+pub fn tlb_flush_asid(asid: u8) {
+    barrier::dsb(barrier::ISHST);
+    let value = (asid as u64) << 48;
+    unsafe {
+        core::arch::asm!("tlbi aside1, {value}", value = in(reg) value);
+    }
+    barrier::dsb(barrier::ISH);
+    barrier::isb(barrier::SY);
+}
+
+#[inline]
+pub fn tlb_flush_all() {
+    barrier::dsb(barrier::ISHST);
+    unsafe {
+        core::arch::asm!("tlbi vmalle1is");
+    }
+    barrier::dsb(barrier::ISH);
+    barrier::isb(barrier::SY);
+}
+
+#[inline]
+pub fn tlb_flush_vaae1(page: usize) {
+    barrier::dsb(barrier::ISHST);
+    let argument = page >> 12;
+    unsafe {
+        core::arch::asm!("tlbi vaae1, {argument}", argument = in(reg) argument);
+    }
+    barrier::dsb(barrier::ISH);
+    barrier::isb(barrier::SY);
+}
+
+#[inline]
+pub fn tlb_flush_range_va(base: usize, size: usize) {
+    let end = (base + size).page_align_up::<L3>();
+    let base = base.page_align_down::<L3>();
+    let count = (end - base).page_count::<L3>();
+    for i in 0..count {
+        tlb_flush_vaae1(base + i * L3::SIZE);
+    }
+}
+
+pub fn at_s1e0r(input: usize) -> Option<u64> {
+    barrier::dsb(barrier::ISHST);
+    unsafe {
+        core::arch::asm!("at s1e0r, {address}", address = in(reg) input);
+    }
+    barrier::dsb(barrier::ISH);
+    barrier::isb(barrier::SY);
+    if PAR_EL1.matches_all(PAR_EL1::F::TranslationSuccessfull) {
+        Some(PAR_EL1.read(PAR_EL1::PA))
+    } else {
+        None
+    }
+}
+
+pub fn at_s1e1r(input: usize) -> Option<u64> {
+    barrier::dsb(barrier::ISHST);
+    unsafe {
+        core::arch::asm!("at s1e1r, {address}", address = in(reg) input);
+    }
+    barrier::dsb(barrier::ISH);
+    barrier::isb(barrier::SY);
+    if PAR_EL1.matches_all(PAR_EL1::F::TranslationSuccessfull) {
+        Some(PAR_EL1.read(PAR_EL1::PA))
+    } else {
+        None
+    }
+}
+
+pub fn ic_iallu() {
+    atomic::compiler_fence(Ordering::SeqCst);
+    barrier::dsb(barrier::ISH);
+    barrier::isb(barrier::SY);
+    unsafe {
+        core::arch::asm!("ic iallu");
+    }
+    barrier::isb(barrier::SY);
+}
+
+pub fn dc_cvac(input: usize) {
+    barrier::dsb(barrier::ISHST);
+    unsafe {
+        core::arch::asm!("dc cvac, {address}", address = in(reg) input);
+    }
+}

kernel/arch/aarch64/src/mem/mod.rs

@@ -1,99 +1,49 @@
-use core::{
-    alloc::Layout,
-    ops::{Deref, DerefMut},
-    ptr::addr_of,
-    sync::atomic::AtomicUsize,
-    sync::atomic::Ordering,
+#![allow(clippy::missing_safety_doc)]
+use aarch64_cpu::{
+    asm::barrier,
+    registers::{MAIR_EL1, SCTLR_EL1, TCR_EL1},
 };
-
-use aarch64_cpu::registers::{TTBR0_EL1, TTBR1_EL1};
 use kernel_arch_interface::{
     mem::{DeviceMemoryAttributes, KernelTableManager, RawDeviceMemoryMapping},
-    sync::split_spinlock,
+    sync::IrqSafeSpinlock,
     KERNEL_VIRT_OFFSET,
 };
-use libk_mm_interface::{
-    address::PhysicalAddress,
-    table::{page_index, EntryLevel, EntryLevelExt},
-};
-use memtables::aarch64::{FixedTables, KERNEL_L3_COUNT};
-use static_assertions::const_assert_eq;
-use tock_registers::interfaces::Writeable;
+use libk_mm_interface::{address::PhysicalAddress, table::EntryLevel};
+use tock_registers::interfaces::{ReadWriteable, Writeable};
 use yggdrasil_abi::error::Error;
 
-use self::table::{PageAttributes, PageEntry, PageTable, L1, L2, L3};
+pub use intrinsics::*;
 
+use crate::{mem::table::L1, ArchitectureImpl};
+
+pub mod fixed;
+pub mod intrinsics;
 pub mod process;
 pub mod table;
 
 #[derive(Debug)]
 pub struct KernelTableManagerImpl;
 
-// TODO eliminate this requirement by using precomputed indices
-const MAPPING_OFFSET: usize = KERNEL_VIRT_OFFSET;
-const KERNEL_PHYS_BASE: usize = 0x40080000;
-
-// Precomputed mappings
-const KERNEL_L1_INDEX: usize = page_index::<L1>(KERNEL_VIRT_OFFSET + KERNEL_PHYS_BASE);
-const KERNEL_START_L2_INDEX: usize = page_index::<L2>(KERNEL_VIRT_OFFSET + KERNEL_PHYS_BASE);
-const KERNEL_END_L2_INDEX: usize = KERNEL_START_L2_INDEX + KERNEL_L3_COUNT;
-
-// Must not be zero, should be at 4MiB
-const_assert_eq!(KERNEL_START_L2_INDEX, 0);
-// From static mapping
-const_assert_eq!(KERNEL_L1_INDEX, 1);
-
-// Runtime mappings
-// 2MiB max
-const EARLY_MAPPING_L2I: usize = KERNEL_END_L2_INDEX + 1;
-// 1GiB max
-const DEVICE_MAPPING_L1I: usize = KERNEL_L1_INDEX + 2;
-const DEVICE_MAPPING_L3_COUNT: usize = 4;
-// 16GiB max
-const RAM_MAPPING_START_L1I: usize = KERNEL_L1_INDEX + 3;
-pub const RAM_MAPPING_L1_COUNT: usize = 16;
-
-// 2MiB for early mappings
-const EARLY_MAPPING_OFFSET: usize =
-    MAPPING_OFFSET | (KERNEL_L1_INDEX * L1::SIZE) | (EARLY_MAPPING_L2I * L2::SIZE);
-static mut EARLY_MAPPING_L3: PageTable<L3> = PageTable::zeroed();
-// 1GiB for device MMIO mapping
-const DEVICE_MAPPING_OFFSET: usize = MAPPING_OFFSET | (DEVICE_MAPPING_L1I * L1::SIZE);
-static mut DEVICE_MAPPING_L2: PageTable<L2> = PageTable::zeroed();
-static mut DEVICE_MAPPING_L3S: [PageTable<L3>; DEVICE_MAPPING_L3_COUNT] =
-    [PageTable::zeroed(); DEVICE_MAPPING_L3_COUNT];
-// 16GiB for RAM mapping
-pub const RAM_MAPPING_OFFSET: usize = MAPPING_OFFSET | (RAM_MAPPING_START_L1I * L1::SIZE);
-pub static MEMORY_LIMIT: AtomicUsize = AtomicUsize::new(0);
-
-split_spinlock! {
-    use crate::ArchitectureImpl;
-    use crate::mem::FixedTables;
-    use libk_mm_interface::KernelImageObject;
-
-    #[link_section = ".data.tables"]
-    static KERNEL_TABLES<lock: ArchitectureImpl>: KernelImageObject<FixedTables> =
-        unsafe { KernelImageObject::new(FixedTables::zeroed()) };
-}
+static KERNEL_MEMORY_LOCK: IrqSafeSpinlock<ArchitectureImpl, ()> = IrqSafeSpinlock::new(());
 
 impl KernelTableManager for KernelTableManagerImpl {
     fn virtualize(address: u64) -> usize {
         let address = address as usize;
-        if address < MEMORY_LIMIT.load(Ordering::Acquire) {
-            address + RAM_MAPPING_OFFSET
+        if address < fixed::IDENTITY_SIZE_L1 * L1::SIZE {
+            address + KERNEL_VIRT_OFFSET
         } else {
-            panic!("Invalid physical address: {:#x}", address);
+            panic!("Invalid physical address: {address:#x}");
         }
     }
 
     fn physicalize(address: usize) -> u64 {
-        if address < RAM_MAPPING_OFFSET
-            || address - RAM_MAPPING_OFFSET >= MEMORY_LIMIT.load(Ordering::Acquire)
+        if address < KERNEL_VIRT_OFFSET
+            || address - KERNEL_VIRT_OFFSET >= fixed::IDENTITY_SIZE_L1 * L1::SIZE
         {
-            panic!("Not a virtualized physical address: {:#x}", address);
+            panic!("Invalid virtual (-> physical) address {address:#x}");
         }
 
-        (address - RAM_MAPPING_OFFSET) as _
+        (address - KERNEL_VIRT_OFFSET) as u64
     }
 
     unsafe fn map_device_pages(
@@ -101,297 +51,96 @@ impl KernelTableManager for KernelTableManagerImpl {
         count: usize,
         attrs: DeviceMemoryAttributes,
     ) -> Result<RawDeviceMemoryMapping<Self>, Error> {
-        map_device_memory(PhysicalAddress::from_u64(base), count, attrs)
+        let _lock = KERNEL_MEMORY_LOCK.lock();
+        #[allow(static_mut_refs)]
+        fixed::DEVICE_MEMORY.map_device_pages(PhysicalAddress::from_u64(base), count, attrs)
     }
 
     unsafe fn unmap_device_pages(mapping: &RawDeviceMemoryMapping<Self>) {
-        unmap_device_memory(mapping)
-    }
-}
-
-/// Memory mapping which may be used for performing early kernel initialization
-pub struct EarlyMapping<'a, T: ?Sized> {
-    value: &'a mut T,
-    page_count: usize,
-}
-
-impl<'a, T: Sized> EarlyMapping<'a, T> {
-    /// # Safety
-    ///
-    /// `physical` address provided must be a valid non-NULL address actually containing `T`.
-    pub unsafe fn map_slice(
-        physical: PhysicalAddress,
-        len: usize,
-    ) -> Result<EarlyMapping<'a, [T]>, Error> {
-        let layout = Layout::array::<T>(len).unwrap();
-        let aligned = physical.page_align_down::<L3>();
-        let offset = physical.page_offset::<L3>();
-        let page_count = (offset + layout.size()).div_ceil(L3::SIZE);
-
-        let virt = map_early_pages(aligned, page_count)?;
-        let value = core::slice::from_raw_parts_mut((virt + offset) as *mut T, len);
-
-        Ok(EarlyMapping { value, page_count })
-    }
-}
-
-impl<T: ?Sized> Deref for EarlyMapping<'_, T> {
-    type Target = T;
-
-    fn deref(&self) -> &Self::Target {
-        self.value
-    }
-}
-
-impl<T: ?Sized> DerefMut for EarlyMapping<'_, T> {
-    fn deref_mut(&mut self) -> &mut Self::Target {
-        self.value
-    }
-}
-
-impl<T: ?Sized> Drop for EarlyMapping<'_, T> {
-    fn drop(&mut self) {
-        let address = (self.value as *mut T).addr() & !(L3::SIZE - 1);
-
-        for i in 0..self.page_count {
-            let page = address + i * L3::SIZE;
-
-            unsafe {
-                unmap_early_page(page);
-            }
-        }
-    }
-}
-
-fn kernel_table_flags() -> PageAttributes {
-    PageAttributes::TABLE
-        | PageAttributes::ACCESS
-        | PageAttributes::SH_INNER
-        | PageAttributes::PAGE_ATTR_NORMAL
-        | PageAttributes::PRESENT
-}
-
-fn ram_block_flags() -> PageAttributes {
-    // TODO UXN, PXN
-    PageAttributes::BLOCK
-        | PageAttributes::ACCESS
-        | PageAttributes::SH_INNER
-        | PageAttributes::PAGE_ATTR_NORMAL
-        | PageAttributes::PRESENT
-}
-
-// Early mappings
-unsafe fn map_early_pages(physical: PhysicalAddress, count: usize) -> Result<usize, Error> {
-    for l3i in 0..512 {
-        let mut taken = false;
-        for i in 0..count {
-            if EARLY_MAPPING_L3[i + l3i].is_present() {
-                taken = true;
-                break;
-            }
-        }
-
-        if taken {
-            continue;
-        }
-
-        for i in 0..count {
-            let page = physical.add(i * L3::SIZE);
-            // TODO NX, NC
-            EARLY_MAPPING_L3[i + l3i] = PageEntry::normal_page(page, PageAttributes::empty());
-        }
-
-        return Ok(EARLY_MAPPING_OFFSET + l3i * L3::SIZE);
-    }
-
-    Err(Error::OutOfMemory)
-}
-
-unsafe fn unmap_early_page(address: usize) {
-    if !(EARLY_MAPPING_OFFSET..EARLY_MAPPING_OFFSET + L2::SIZE).contains(&address) {
-        panic!("Tried to unmap invalid early mapping: {:#x}", address);
-    }
-
-    let l3i = (address - EARLY_MAPPING_OFFSET).page_index::<L3>();
-
-    assert!(EARLY_MAPPING_L3[l3i].is_present());
-    EARLY_MAPPING_L3[l3i] = PageEntry::INVALID;
-
-    // TODO invalidate tlb
-}
-
-/// # Safety
-///
-/// Only meant to be used by the architecture initialization functions.
-pub unsafe fn map_ram_l1(index: usize) {
-    if index >= RAM_MAPPING_L1_COUNT {
-        todo!()
-    }
-    let mut tables = KERNEL_TABLES.lock();
-    assert_eq!(tables.l1.data[index + RAM_MAPPING_START_L1I], 0);
-
-    tables.l1.data[index + RAM_MAPPING_START_L1I] =
-        ((index * L1::SIZE) as u64) | ram_block_flags().bits();
-}
-
-// Device mappings
-unsafe fn map_device_memory_l3(
-    base: PhysicalAddress,
-    count: usize,
-    _attrs: DeviceMemoryAttributes,
-) -> Result<usize, Error> {
-    // TODO don't map pages if already mapped
-
-    'l0: for i in 0..DEVICE_MAPPING_L3_COUNT * 512 {
-        for j in 0..count {
-            let l2i = (i + j) / 512;
-            let l3i = (i + j) % 512;
-
-            if DEVICE_MAPPING_L3S[l2i][l3i].is_present() {
-                continue 'l0;
-            }
-        }
-
-        for j in 0..count {
-            let l2i = (i + j) / 512;
-            let l3i = (i + j) % 512;
-
-            // TODO NX, NC
-            DEVICE_MAPPING_L3S[l2i][l3i] = PageEntry::device_page(base.add(j * L3::SIZE));
-        }
-
-        return Ok(DEVICE_MAPPING_OFFSET + i * L3::SIZE);
-    }
-
-    Err(Error::OutOfMemory)
-}
-
-unsafe fn map_device_memory_l2(
-    base: PhysicalAddress,
-    count: usize,
-    _attrs: DeviceMemoryAttributes,
-) -> Result<usize, Error> {
-    'l0: for i in DEVICE_MAPPING_L3_COUNT..512 {
-        for j in 0..count {
-            if DEVICE_MAPPING_L2[i + j].is_present() {
-                continue 'l0;
-            }
-        }
-
-        for j in 0..count {
-            DEVICE_MAPPING_L2[i + j] = PageEntry::<L2>::device_block(base.add(j * L2::SIZE));
-        }
-
-        // log::debug!(
-        //     "map l2s: base={:#x}, count={} -> {:#x}",
-        //     base,
-        //     count,
-        //     DEVICE_MAPPING_OFFSET + i * L2::SIZE
-        // );
-        return Ok(DEVICE_MAPPING_OFFSET + i * L2::SIZE);
-    }
-
-    Err(Error::OutOfMemory)
-}
-
-pub(crate) unsafe fn map_device_memory(
-    base: PhysicalAddress,
-    size: usize,
-    attrs: DeviceMemoryAttributes,
-) -> Result<RawDeviceMemoryMapping<KernelTableManagerImpl>, Error> {
-    let l3_aligned = base.page_align_down::<L3>();
-    let l3_offset = base.page_offset::<L3>();
-    let page_count = (l3_offset + size).page_count::<L3>();
-
-    if page_count > 256 {
-        // Large mapping, use L2 mapping instead
-        let l2_aligned = base.page_align_down::<L2>();
-        let l2_offset = base.page_offset::<L2>();
-        let page_count = (l2_offset + size).page_count::<L2>();
-
-        let base_address = map_device_memory_l2(l2_aligned, page_count, attrs)?;
-        let address = base_address + l2_offset;
-
-        Ok(RawDeviceMemoryMapping::from_raw_parts(
-            address,
-            base_address,
-            page_count,
-            L2::SIZE,
-        ))
-    } else {
-        // Just map the pages directly
-        let base_address = map_device_memory_l3(l3_aligned, page_count, attrs)?;
-        let address = base_address + l3_offset;
-
-        Ok(RawDeviceMemoryMapping::from_raw_parts(
-            address,
-            base_address,
-            page_count,
-            L3::SIZE,
-        ))
-    }
-}
-
-pub(crate) unsafe fn unmap_device_memory(map: &RawDeviceMemoryMapping<KernelTableManagerImpl>) {
-    match map.page_size {
-        L3::SIZE => {
-            for i in 0..map.page_count {
-                let page = map.base_address + i * L3::SIZE;
-                let l2i = page.page_index::<L2>();
-                let l3i = page.page_index::<L3>();
-                assert!(DEVICE_MAPPING_L3S[l2i][l3i].is_present());
-                DEVICE_MAPPING_L3S[l2i][l3i] = PageEntry::INVALID;
-
-                tlb_flush_vaae1(page);
-            }
-        }
-        L2::SIZE => todo!(),
-        _ => unimplemented!(),
+        let _lock = KERNEL_MEMORY_LOCK.lock();
+        #[allow(static_mut_refs)]
+        fixed::DEVICE_MEMORY.unmap_device_pages(mapping);
     }
 }
 
 #[inline]
-pub fn tlb_flush_vaae1(mut page: usize) {
-    page >>= 12;
-    unsafe {
-        core::arch::asm!("tlbi vaae1, {page}", page = in(reg) page);
+pub fn auto_lower_address<T>(ptr: *const T) -> usize {
+    let address = ptr.addr();
+    if address < KERNEL_VIRT_OFFSET {
+        address
+    } else {
+        address - KERNEL_VIRT_OFFSET
     }
 }
 
-/// (BSP-early init) loads precomputed kernel mapping tables for the kernel to jump to "higher-half"
-///
-/// # Safety
-///
-/// Unsafe, must only be called by BSP during its early init while still in "lower-half"
-pub unsafe fn load_fixed_tables() {
-    let ttbr0 = KERNEL_TABLES.lock().l1.data.as_ptr().addr() as u64;
-    TTBR0_EL1.set(ttbr0);
-    TTBR1_EL1.set(ttbr0);
-}
-
-/// Sets up additional translation tables for kernel usage
-///
-/// # Safety
-///
-/// Unsafe, must only be called by BSP during its early init, must already be in "higher-half"
-pub unsafe fn init_fixed_tables() {
-    // TODO this could be built in compile-time too?
-    let mut tables = KERNEL_TABLES.lock();
-    let early_mapping_l3_phys = addr_of!(EARLY_MAPPING_L3) as usize - KERNEL_VIRT_OFFSET;
-    let device_mapping_l2_phys = addr_of!(DEVICE_MAPPING_L2) as usize - KERNEL_VIRT_OFFSET;
-
-    for i in 0..DEVICE_MAPPING_L3_COUNT {
-        let device_mapping_l3_phys = PhysicalAddress::from_usize(
-            &DEVICE_MAPPING_L3S[i] as *const _ as usize - KERNEL_VIRT_OFFSET,
-        );
-        DEVICE_MAPPING_L2[i] = PageEntry::table(device_mapping_l3_phys, PageAttributes::empty());
+#[inline]
+pub fn auto_upper_address<T>(ptr: *const T) -> usize {
+    let address = ptr.addr();
+    if address < KERNEL_VIRT_OFFSET {
+        address + KERNEL_VIRT_OFFSET
+    } else {
+        address
     }
-
-    assert_eq!(tables.l2.data[EARLY_MAPPING_L2I], 0);
-    tables.l2.data[EARLY_MAPPING_L2I] =
-        (early_mapping_l3_phys as u64) | kernel_table_flags().bits();
-
-    assert_eq!(tables.l1.data[DEVICE_MAPPING_L1I], 0);
-    tables.l1.data[DEVICE_MAPPING_L1I] =
-        (device_mapping_l2_phys as u64) | kernel_table_flags().bits();
+}
+
+fn setup_memory_attributes() {
+    // TODO: Figure out why WriteBack_NonTransient_ReadWriteAlloc doesn't work on Pi 4B
+
+    MAIR_EL1.write(
+        //// Attribute 0 -- normal memory
+        MAIR_EL1::Attr0_Normal_Inner::WriteBack_NonTransient +
+         MAIR_EL1::Attr0_Normal_Outer::WriteBack_NonTransient +
+         //// Attribute 1 -- normal non-cacheable memory
+         MAIR_EL1::Attr0_Normal_Inner::NonCacheable +
+         MAIR_EL1::Attr0_Normal_Outer::NonCacheable +
+         //// Attribute 2 -- device memory
+         MAIR_EL1::Attr1_Device::nonGathering_nonReordering_EarlyWriteAck,
+    );
+}
+
+unsafe fn enable_mmu() {
+    TCR_EL1.write(
+        TCR_EL1::AS::ASID8Bits +
+         TCR_EL1::A1::TTBR0 +
+         TCR_EL1::HD::CLEAR +
+         // General
+         TCR_EL1::IPS::Bits_48 +
+         // TTBR0
+         TCR_EL1::TG0::KiB_4 + TCR_EL1::T0SZ.val(25) + TCR_EL1::SH0::Inner +
+         // TTBR1
+         TCR_EL1::TG1::KiB_4 + TCR_EL1::T1SZ.val(25) + TCR_EL1::SH1::Inner,
+    );
+
+    barrier::dsb(barrier::ISHST);
+    barrier::isb(barrier::SY);
+
+    SCTLR_EL1.modify(
+        SCTLR_EL1::E0E::LittleEndian
+            + SCTLR_EL1::EE::LittleEndian
+            + SCTLR_EL1::WXN::Disable
+            + SCTLR_EL1::SA0::Enable
+            + SCTLR_EL1::SA::Enable
+            + SCTLR_EL1::A::Enable
+            + SCTLR_EL1::I::NonCacheable
+            + SCTLR_EL1::C::NonCacheable,
+    );
+
+    barrier::dsb(barrier::ISH);
+    barrier::isb(barrier::SY);
+
+    // Enable translation
+    SCTLR_EL1.modify(SCTLR_EL1::M::Enable);
+
+    // Enable caches
+    enable_icache();
+    enable_dcache();
+}
+
+pub unsafe fn init_lower(bsp: bool) {
+    setup_memory_attributes();
+    if bsp {
+        fixed::setup();
+    }
+    fixed::load();
+    enable_mmu();
 }

kernel/arch/aarch64/src/mem/process.rs

@@ -7,7 +7,7 @@ use core::{
 use libk_mm_interface::{
     address::{AsPhysicalAddress, PhysicalAddress},
     pointer::PhysicalRefMut,
-    process::ProcessAddressSpaceManager,
+    process::{PageAttributeUpdate, ProcessAddressSpaceManager},
     table::{
         EntryLevel, EntryLevelDrop, EntryLevelExt, MapAttributes, NextPageTable, TableAllocator,
     },
@@ -17,8 +17,9 @@ use yggdrasil_abi::error::Error;
 use crate::{mem::table::PageEntry, KernelTableManagerImpl};
 
 use super::{
-    table::{PageTable, L1, L2, L3},
-    tlb_flush_vaae1,
+    dc_cvac, ic_iallu,
+    table::{PageAttributes, PageTable, L1, L2, L3},
+    tlb_flush_asid, tlb_flush_vaae1,
 };
 
 /// AArch64 implementation of a process address space table
@@ -49,6 +50,8 @@ impl<TA: TableAllocator> ProcessAddressSpaceManager<TA> for ProcessAddressSpaceI
             l1[i] = PageEntry::INVALID;
         }
 
+        tlb_flush_asid(asid);
+
         Ok(Self {
             l1,
             asid,
@@ -68,17 +71,29 @@ impl<TA: TableAllocator> ProcessAddressSpaceManager<TA> for ProcessAddressSpaceI
     ) -> Result<(), Error> {
         self.write_l3_entry(
             address,
-            PageEntry::normal_page(physical, flags.into()),
+            PageEntry::normal_page(
+                physical,
+                PageAttributes::from(flags) | PageAttributes::NON_GLOBAL,
+            ),
             false,
         )
     }
 
-    unsafe fn unmap_page(&mut self, address: usize) -> Result<PhysicalAddress, Error> {
+    unsafe fn update_page_attributes(
+        &mut self,
+        address: usize,
+        update: &PageAttributeUpdate,
+    ) -> Result<(), Error> {
+        self.update_l3_entry(address, |entry| entry.update(update))
+    }
+
+    unsafe fn unmap_page(&mut self, address: usize) -> Result<(PhysicalAddress, bool), Error> {
         self.pop_l3_entry(address)
     }
 
-    fn as_address_with_asid(&self) -> u64 {
-        unsafe { u64::from(self.l1.as_physical_address()) | ((self.asid as u64) << 48) }
+    fn as_address_with_asid(&self) -> (u64, u64) {
+        let physical = unsafe { u64::from(self.l1.as_physical_address()) };
+        (physical, self.asid as u64)
     }
 
     unsafe fn clear(&mut self) {
@@ -107,12 +122,38 @@ impl<TA: TableAllocator> ProcessAddressSpaceImpl<TA> {
         }
 
         l3[l3i] = entry;
+        dc_cvac((&raw const l3[l3i]).addr());
         tlb_flush_vaae1(virt);
 
         Ok(())
     }
 
-    fn pop_l3_entry(&mut self, virt: usize) -> Result<PhysicalAddress, Error> {
+    fn update_l3_entry<F: FnOnce(&mut PageEntry<L3>) -> Result<(), Error>>(
+        &mut self,
+        virt: usize,
+        mapper: F,
+    ) -> Result<(), Error> {
+        let l1i = virt.page_index::<L1>();
+        let l2i = virt.page_index::<L2>();
+        let l3i = virt.page_index::<L3>();
+
+        let mut l2 = self.l1.get_mut(l1i).ok_or(Error::DoesNotExist)?;
+        let mut l3 = l2.get_mut(l2i).ok_or(Error::DoesNotExist)?;
+
+        let entry = &mut l3[l3i];
+        if !entry.is_present() {
+            return Err(Error::DoesNotExist);
+        }
+
+        mapper(entry)?;
+        ic_iallu();
+        dc_cvac((&raw const l3[l3i]).addr());
+        tlb_flush_vaae1(virt);
+
+        Ok(())
+    }
+
+    fn pop_l3_entry(&mut self, virt: usize) -> Result<(PhysicalAddress, bool), Error> {
         let l1i = virt.page_index::<L1>();
         let l2i = virt.page_index::<L2>();
         let l3i = virt.page_index::<L3>();
@@ -121,12 +162,16 @@ impl<TA: TableAllocator> ProcessAddressSpaceImpl<TA> {
         let mut l2 = self.l1.get_mut(l1i).ok_or(Error::DoesNotExist)?;
         let mut l3 = l2.get_mut(l2i).ok_or(Error::DoesNotExist)?;
 
-        let page = l3[l3i].as_page().ok_or(Error::DoesNotExist)?;
+        let entry = l3[l3i];
+        let page = entry.as_page().ok_or(Error::DoesNotExist)?;
+        let dirty = entry.is_dirty();
 
         l3[l3i] = PageEntry::INVALID;
+        ic_iallu();
+        dc_cvac((&raw const l3[l3i]).addr());
         tlb_flush_vaae1(virt);
 
-        Ok(page)
+        Ok((page, dirty))
     }
 
     fn read_l3_entry(&self, virt: usize) -> Option<(PhysicalAddress, MapAttributes)> {

kernel/arch/aarch64/src/mem/table.rs

@@ -1,12 +1,15 @@
 use core::{
+    fmt,
     marker::PhantomData,
     ops::{Index, IndexMut, Range},
 };
 
 use bitflags::bitflags;
+use kernel_arch_interface::KERNEL_VIRT_OFFSET;
 use libk_mm_interface::{
     address::{AsPhysicalAddress, PhysicalAddress},
     pointer::{PhysicalRef, PhysicalRefMut},
+    process::PageAttributeUpdate,
     table::{
         EntryLevel, EntryLevelDrop, MapAttributes, NextPageTable, NonTerminalEntryLevel,
         TableAllocator,
@@ -16,6 +19,8 @@ use yggdrasil_abi::error::Error;
 
 use crate::KernelTableManagerImpl;
 
+use super::dc_cvac;
+
 bitflags! {
     #[derive(Clone, Copy, PartialEq, Eq)]
     pub struct PageAttributes: u64 {
@@ -37,10 +42,12 @@ bitflags! {
         const SH_INNER = 3 << 8;
 
         const PAGE_ATTR_NORMAL = 0 << 2;
-        const PAGE_ATTR_DEVICE = 1 << 2;
+        const PAGE_ATTR_NORMAL_NC = 1 << 2;
+        const PAGE_ATTR_DEVICE = 2 << 2;
 
         const NON_GLOBAL = 1 << 11;
 
+        const DIRTY = 1 << 51;
         const PXN = 1 << 53;
         const UXN = 1 << 54;
     }
@@ -62,6 +69,13 @@ pub struct L2;
 #[derive(Clone, Copy)]
 pub struct L3;
 
+#[derive(Debug, Clone, Copy)]
+pub enum EntryType {
+    Table(PhysicalAddress),
+    Page(PhysicalAddress),
+    Invalid,
+}
+
 impl NonTerminalEntryLevel for L1 {
     type NextLevel = L2;
 }
@@ -101,11 +115,55 @@ impl<L: EntryLevel> PageTable<L> {
 
         Ok(table)
     }
+
+    /// Creates a reference to [PageTable] from a physical address.
+    ///
+    /// # Safety
+    ///
+    /// The function takes in a raw physical address.
+    pub unsafe fn from_physical(
+        physical: PhysicalAddress,
+    ) -> Option<PhysicalRefMut<'static, Self, KernelTableManagerImpl>> {
+        if physical.into_usize() >= KERNEL_VIRT_OFFSET {
+            // Looks fishy
+            return None;
+        }
+        if !physical.is_aligned_for::<L3>() {
+            return None;
+        }
+
+        let inner = PhysicalRefMut::map(physical);
+        Some(inner)
+    }
 }
 
 impl<L: EntryLevel> PageEntry<L> {
+    const ATTR_MASK: u64 = 0xFFF | (0xFFFF << 48);
     pub const INVALID: Self = Self(0, PhantomData);
 
+    pub fn update(&mut self, update: &PageAttributeUpdate) -> Result<(), Error> {
+        let mut attrs = PageAttributes::from_bits_retain(self.0);
+        if let Some(write) = update.user_write {
+            // Make writeable/non-writeable
+            if write {
+                attrs &= !PageAttributes::AP_ACCESS_MASK;
+                attrs |= PageAttributes::AP_BOTH_READWRITE;
+            } else {
+                todo!();
+            }
+        }
+        if let Some(dirty) = update.dirty {
+            if dirty {
+                attrs |= PageAttributes::DIRTY;
+            } else {
+                attrs &= !PageAttributes::DIRTY;
+            }
+        }
+        self.0 &= !Self::ATTR_MASK;
+        self.0 |= attrs.bits() & Self::ATTR_MASK;
+        Ok(())
+    }
+
     pub const fn is_present(self) -> bool {
         self.0 & PageAttributes::PRESENT.bits() != 0
     }
@@ -115,6 +173,12 @@ impl<L: EntryLevel> PageEntry<L> {
     }
 }
 
+impl<L: NonTerminalEntryLevel> PageTable<L> {
+    pub fn walk(&self, index: usize) -> EntryType {
+        self[index].classify()
+    }
+}
+
 impl<L: NonTerminalEntryLevel + 'static> NextPageTable for PageTable<L> {
     type NextLevel = PageTable<L::NextLevel>;
     type TableRef = PhysicalRef<'static, PageTable<L::NextLevel>, KernelTableManagerImpl>;
@@ -146,6 +210,7 @@ impl<L: NonTerminalEntryLevel + 'static> NextPageTable for PageTable<L> {
                 unsafe { table.as_physical_address() },
                 PageAttributes::empty(),
             );
+            dc_cvac((&raw const self[index]).addr());
             Ok(table)
         }
     }
@@ -184,6 +249,7 @@ where
             }
 
             self[index] = PageEntry::INVALID;
+            dc_cvac((&raw const self[index]).addr());
         }
     }
 }
@@ -202,7 +268,7 @@ impl<L: NonTerminalEntryLevel> PageEntry<L> {
                 | (PageAttributes::BLOCK
                     | PageAttributes::PRESENT
                     | PageAttributes::ACCESS
-                    | PageAttributes::SH_INNER
+                    | PageAttributes::SH_OUTER
                     | PageAttributes::PAGE_ATTR_NORMAL
                     | attrs)
                     .bits(),
@@ -231,11 +297,21 @@ impl<L: NonTerminalEntryLevel> PageEntry<L> {
         if self.0 & PageAttributes::PRESENT.bits() != 0
             && self.0 & PageAttributes::BLOCK.bits() == 0
         {
-            Some(PhysicalAddress::from_u64(self.0 & !0xFFF))
+            Some(PhysicalAddress::from_u64(self.0 & !Self::ATTR_MASK))
         } else {
             None
         }
     }
+
+    pub fn classify(self) -> EntryType {
+        if !self.is_present() {
+            EntryType::Invalid
+        } else if let Some(table) = self.as_table() {
+            EntryType::Table(table)
+        } else {
+            EntryType::Page(PhysicalAddress::from_u64(self.0 & !Self::ATTR_MASK))
+        }
+    }
 }
 
 impl PageEntry<L3> {
@@ -260,18 +336,20 @@ impl PageEntry<L3> {
                     | PageAttributes::PRESENT
                     | PageAttributes::ACCESS
                     | PageAttributes::SH_OUTER
-                    | PageAttributes::PAGE_ATTR_DEVICE
-                    | PageAttributes::UXN
-                    | PageAttributes::PXN)
+                    | PageAttributes::PAGE_ATTR_DEVICE)
                     .bits(),
             PhantomData,
         )
     }
 
+    pub fn is_dirty(&self) -> bool {
+        self.0 & PageAttributes::DIRTY.bits() != 0
+    }
+
     pub fn as_page(&self) -> Option<PhysicalAddress> {
         let mask = (PageAttributes::PRESENT | PageAttributes::PAGE).bits();
         if self.0 & mask == mask {
-            Some(PhysicalAddress::from_u64(self.0 & !0xFFF))
+            Some(PhysicalAddress::from_u64(self.0 & !Self::ATTR_MASK))
         } else {
             None
         }
@@ -307,6 +385,10 @@ impl From<MapAttributes> for PageAttributes {
             out |= PageAttributes::AP_KERNEL_READONLY;
         }
 
+        if value.contains(MapAttributes::DIRTY) {
+            out |= PageAttributes::DIRTY;
+        }
+
         if value.contains(MapAttributes::NON_GLOBAL) {
             out |= PageAttributes::NON_GLOBAL;
         }
@@ -329,6 +411,10 @@ impl From<PageAttributes> for MapAttributes {
             _ => unreachable!(),
         };
 
+        if value.contains(PageAttributes::DIRTY) {
+            out |= MapAttributes::DIRTY;
+        }
+
         if value.contains(PageAttributes::NON_GLOBAL) {
             out |= MapAttributes::NON_GLOBAL;
         }
@@ -336,3 +422,13 @@ impl From<PageAttributes> for MapAttributes {
         out
     }
 }
+
+impl fmt::Display for EntryType {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            Self::Table(address) => write!(f, "table @ {address:#x}"),
+            Self::Page(address) => write!(f, "page @ {address:#x}"),
+            Self::Invalid => f.write_str("<invalid>"),
+        }
+    }
+}

kernel/arch/hosted/Cargo.toml

@@ -7,3 +7,4 @@ edition = "2021"
 kernel-arch-interface.workspace = true
 yggdrasil-abi.workspace = true
 libk-mm-interface.workspace = true
+device-api.workspace = true

kernel/arch/hosted/src/lib.rs

@@ -1,11 +1,13 @@
-#![feature(never_type)]
+#![feature(never_type, allocator_api, slice_ptr_get)]
 use std::{
+    alloc::{Allocator, Global, Layout},
     marker::PhantomData,
     sync::atomic::{AtomicBool, Ordering},
 };
 
+use device_api::dma::{DmaAllocation, DmaAllocator};
 use kernel_arch_interface::{
-    cpu::IpiQueue,
+    cpu::{CpuData, IpiQueue},
     mem::{
         DeviceMemoryAttributes, KernelTableManager, PhysicalMemoryAllocator, RawDeviceMemoryMapping,
     },
@@ -36,17 +38,21 @@ pub struct TaskContextImpl<K: KernelTableManager, PA: PhysicalMemoryAllocator>(
 
 static DUMMY_INTERRUPT_MASK: AtomicBool = AtomicBool::new(true);
 
+pub struct DummyCpuData;
+
+impl CpuData for DummyCpuData {}
+
 impl Architecture for ArchitectureImpl {
-    type PerCpuData = ();
+    type PerCpuData = DummyCpuData;
     type CpuFeatures = ();
     type BreakpointType = u8;
     const BREAKPOINT_VALUE: Self::BreakpointType = 0x00;
 
-    fn local_cpu() -> *mut Self::PerCpuData {
+    fn local_cpu() -> *mut () {
         unimplemented!()
     }
 
-    unsafe fn set_local_cpu(_cpu: *mut Self::PerCpuData) {
+    unsafe fn set_local_cpu(_cpu: *mut ()) {
         unimplemented!()
     }
 
@@ -101,6 +107,14 @@ impl Architecture for ArchitectureImpl {
     fn ipi_queue(_cpu_id: u32) -> Option<&'static IpiQueue<Self>> {
         None
     }
+
+    fn load_barrier() {}
+
+    fn store_barrier() {}
+
+    fn memory_barrier() {}
+
+    fn flush_virtual_range(_range: std::ops::Range<usize>) {}
 }
 
 impl KernelTableManager for KernelTableManagerImpl {
@@ -146,7 +160,7 @@ impl<TA: TableAllocator> ProcessAddressSpaceManager<TA> for ProcessAddressSpaceI
         unimplemented!()
     }
 
-    unsafe fn unmap_page(&mut self, _address: usize) -> Result<PhysicalAddress, Error> {
+    unsafe fn unmap_page(&mut self, _address: usize) -> Result<(PhysicalAddress, bool), Error> {
         unimplemented!()
     }
 
@@ -154,7 +168,7 @@ impl<TA: TableAllocator> ProcessAddressSpaceManager<TA> for ProcessAddressSpaceI
         unimplemented!()
     }
 
-    fn as_address_with_asid(&self) -> u64 {
+    fn as_address_with_asid(&self) -> (u64, u64) {
         unimplemented!()
     }
 }
@@ -198,3 +212,19 @@ impl<K: KernelTableManager, PA: PhysicalMemoryAllocator> TaskContext<K, PA>
 extern "Rust" fn __signal_process_group(_group_id: ProcessGroupId, _signal: Signal) {
     unimplemented!()
 }
+
+pub struct HostedDmaAllocator;
+
+impl DmaAllocator for HostedDmaAllocator {
+    fn allocate(&self, layout: Layout) -> Result<DmaAllocation, Error> {
+        let ptr = Global.allocate(layout.align_to(0x1000).unwrap()).unwrap();
+        let base = ptr.as_non_null_ptr();
+        let addr: usize = base.addr().into();
+        Ok(DmaAllocation {
+            host_virtual: base.cast(),
+            host_physical: addr as _,
+            page_count: layout.size().div_ceil(0x1000),
+            bus_address: addr as _,
+        })
+    }
+}

kernel/arch/i686/src/context.S

@@ -1,116 +0,0 @@
-// vi: set ft=asm :
-
-.macro SAVE_TASK_STATE
-    push %edi
-    push %esi
-    push %ebp
-    push %ebx
-.endm
-
-.macro LOAD_TASK_STATE
-    pop %ebx
-    pop %ebp
-    pop %esi
-    pop %edi
-.endm
-
-.section .text
-
-.global __i686_task_enter_kernel
-.global __i686_task_enter_user
-.global __i686_task_enter_from_fork
-.global __i686_switch_task
-.global __i686_enter_task
-.global __i686_switch_and_drop
-
-__i686_task_enter_kernel:
-    // %esp + 4: argument
-    // %esp + 0: entry
-    xor %ecx, %ecx
-    xchg (%esp), %ecx
-
-    // Enable IRQ in EFLAGS
-    pushfl
-    pop %edx
-    or $(1 << 9), %edx
-
-    // Setup iret
-    push %edx       // eflags
-    pushl $0x08     // cs
-    push %ecx       // eip
-
-    iret
-
-__i686_task_enter_user:
-    pop %edx // User %esp
-    pop %ecx // entry
-    pop %eax // flags
-
-    // Setup iret
-
-    // %ss:%esp
-    pushl $0x23
-    push %edx
-
-    // %eflags
-    push %eax
-
-    // %cs:%eip
-    pushl $0x1B
-    push %ecx
-
-    mov $0x23, %bx
-    mov %bx, %ds
-    mov %bx, %es
-    mov %bx, %fs
-
-    iret
-
-__i686_task_enter_from_fork:
-    jmp .
-
-__i686_switch_task:
-    // %esp + 0: return
-    // %esp + 4: destination
-    // %esp + 8: source
-    mov 4(%esp), %eax
-    mov 8(%esp), %ecx
-
-    SAVE_TASK_STATE
-
-    // Store stack to "from" context
-    mov %esp, (%ecx)
-
-    // Load stack from "to" context
-    mov (%eax), %esp
-
-    LOAD_TASK_STATE
-
-    ret
-
-__i686_enter_task:
-    // %esp + 0: return
-    // %esp + 4: destination
-
-    // Switch to destination stack
-    mov 4(%esp), %eax
-    mov (%eax), %esp
-
-    LOAD_TASK_STATE
-
-    ret
-
-__i686_switch_and_drop:
-    // %esp + 0: return
-    // %esp + 4: destination
-    // %esp + 8: thread to drop
-    mov 8(%esp), %ecx
-    mov 4(%esp), %eax
-    // Switch to stack
-    mov (%eax), %esp
-
-    LOAD_TASK_STATE
-
-    // TODO actually drop the thread
-
-    ret

kernel/arch/i686/src/context.rs

@@ -1,462 +0,0 @@
-use core::{arch::global_asm, cell::UnsafeCell, marker::PhantomData};
-
-use kernel_arch_interface::{
-    mem::{KernelTableManager, PhysicalMemoryAllocator},
-    task::{StackBuilder, TaskContext, TaskFrame, UserContextInfo},
-};
-use kernel_arch_x86::registers::{FpuContext, CR3};
-use libk_mm_interface::address::{AsPhysicalAddress, PhysicalAddress};
-use tock_registers::interfaces::Writeable;
-use yggdrasil_abi::{arch::SavedFrame, error::Error};
-
-use crate::{
-    gdt::{self, TSS},
-    mem::KERNEL_TABLES,
-};
-
-#[allow(unused)]
-#[repr(C)]
-pub struct ExceptionFrame {
-    pub eax: u32,
-    pub ecx: u32,
-    pub edx: u32,
-    pub ebx: u32,
-    pub ebp: u32,
-    pub esi: u32,
-    pub edi: u32,
-
-    pub exc_number: u32,
-    pub exc_code: u32,
-
-    pub eip: u32,
-    pub cs: u32,
-    pub eflags: u32,
-    pub esp: u32,
-    pub ss: u32,
-}
-
-#[allow(unused)]
-#[derive(Debug)]
-#[repr(C)]
-pub struct SyscallFrame {
-    pub eax: usize,
-    // ebx, ecx, edx, esi, edi, ebp
-    pub args: [usize; 6],
-
-    pub eip: u32,
-    pub cs: u32,
-    pub eflags: u32,
-    pub esp: u32,
-    pub ss: u32,
-}
-
-#[allow(unused)]
-#[repr(C)]
-pub struct InterruptFrame {
-    pub eax: u32,
-    pub ecx: u32,
-    pub edx: u32,
-    pub ebx: u32,
-    pub ebp: u32,
-    pub esi: u32,
-    pub edi: u32,
-
-    pub irq_number: u32,
-
-    pub eip: u32,
-    pub cs: u32,
-    pub eflags: u32,
-    esp: u32,
-    ss: u32,
-}
-
-#[repr(C, align(0x10))]
-struct Inner {
-    // 0x00
-    sp: usize,
-
-    gs_base: usize,
-}
-
-#[allow(dead_code)]
-pub struct TaskContextImpl<
-    K: KernelTableManager,
-    PA: PhysicalMemoryAllocator<Address = PhysicalAddress>,
-> {
-    inner: UnsafeCell<Inner>,
-    fpu_context: Option<UnsafeCell<FpuContext>>,
-    stack_base_phys: PhysicalAddress,
-    stack_size: usize,
-
-    cr3: u32,
-    tss_esp0: u32,
-
-    _pd: PhantomData<(K, PA)>,
-}
-
-impl<K: KernelTableManager, PA: PhysicalMemoryAllocator<Address = PhysicalAddress>>
-    TaskContextImpl<K, PA>
-{
-    unsafe fn store_state(&self) {
-        if let Some(fpu) = self.fpu_context.as_ref() {
-            FpuContext::store(fpu.get());
-        }
-    }
-
-    unsafe fn load_state(&self) {
-        if let Some(fpu) = self.fpu_context.as_ref() {
-            FpuContext::restore(fpu.get());
-        }
-        gdt::set_gs_base((*self.inner.get()).gs_base);
-        TSS.esp0 = self.tss_esp0;
-        CR3.set(self.cr3 as _);
-    }
-}
-
-impl<K: KernelTableManager, PA: PhysicalMemoryAllocator<Address = PhysicalAddress>>
-    TaskContext<K, PA> for TaskContextImpl<K, PA>
-{
-    const SIGNAL_STACK_EXTRA_ALIGN: usize = 0;
-    const USER_STACK_EXTRA_ALIGN: usize = 0;
-
-    fn user(context: UserContextInfo) -> Result<Self, Error> {
-        const USER_TASK_PAGES: usize = 16;
-
-        let stack_base_phys = PA::allocate_contiguous_pages(USER_TASK_PAGES)?;
-        let stack_base = stack_base_phys.raw_virtualize::<K>();
-
-        let mut stack = StackBuilder::new(stack_base, USER_TASK_PAGES * 0x1000);
-        let mut flags = 0x200;
-
-        if context.single_step {
-            flags |= 1 << 8;
-        }
-
-        stack.push(flags);
-        stack.push(context.entry as _);
-        stack.push(context.stack_pointer);
-
-        setup_common_context(&mut stack, __i686_task_enter_user as _);
-
-        let sp = stack.build();
-        let esp0 = stack_base + USER_TASK_PAGES * 0x1000;
-        let fpu_context = FpuContext::new(true);
-
-        Ok(Self {
-            inner: UnsafeCell::new(Inner {
-                sp,
-                gs_base: context.thread_pointer,
-            }),
-            fpu_context: Some(UnsafeCell::new(fpu_context)),
-            stack_base_phys,
-            stack_size: USER_TASK_PAGES * 0x1000,
-
-            tss_esp0: esp0 as _,
-            cr3: context.address_space.try_into().unwrap(),
-
-            _pd: PhantomData,
-        })
-    }
-
-    fn kernel(
-        entry: extern "C" fn(usize) -> !,
-        arg: usize,
-    ) -> Result<Self, yggdrasil_abi::error::Error> {
-        const KERNEL_TASK_PAGES: usize = 32;
-
-        let stack_base_phys = PA::allocate_contiguous_pages(KERNEL_TASK_PAGES)?;
-        let stack_base = stack_base_phys.raw_virtualize::<K>();
-
-        let mut stack = StackBuilder::new(stack_base, KERNEL_TASK_PAGES * 0x1000);
-
-        // Entry and argument
-        stack.push(arg);
-        stack.push(entry as _);
-
-        // XXX
-        setup_common_context(&mut stack, __i686_task_enter_kernel as _);
-
-        let sp = stack.build();
-        let cr3 = unsafe {
-            KERNEL_TABLES
-                .lock()
-                .as_physical_address()
-                .try_into_u32()
-                .unwrap()
-        };
-
-        // TODO stack is leaked
-
-        Ok(Self {
-            inner: UnsafeCell::new(Inner { sp, gs_base: 0 }),
-            fpu_context: None,
-            stack_base_phys,
-            stack_size: KERNEL_TASK_PAGES * 0x1000,
-
-            tss_esp0: 0,
-            cr3,
-
-            _pd: PhantomData,
-        })
-    }
-
-    unsafe fn switch(&self, from: &Self) {
-        if core::ptr::addr_eq(self, from) {
-            return;
-        }
-
-        from.store_state();
-        self.load_state();
-
-        __i686_switch_task(self.inner.get(), from.inner.get());
-    }
-
-    unsafe fn enter(&self) -> ! {
-        self.load_state();
-        __i686_enter_task(self.inner.get())
-    }
-
-    unsafe fn switch_and_drop(&self, thread: *const ()) {
-        self.load_state();
-        __i686_switch_and_drop(self.inner.get(), thread);
-    }
-
-    fn set_thread_pointer(&self, tp: usize) {
-        unsafe { (*self.inner.get()).gs_base = tp };
-        gdt::set_gs_base(tp);
-    }
-
-    fn align_stack_for_entry(sp: usize) -> usize {
-        (sp & !0xF) - 12
-    }
-}
-
-fn setup_common_context(builder: &mut StackBuilder, entry: usize) {
-    builder.push(entry);
-
-    builder.push(0); // %edi
-    builder.push(0); // %esi
-    builder.push(0); // %ebp
-    builder.push(0); // %ebx
-}
-
-extern "C" {
-    fn __i686_task_enter_kernel();
-    fn __i686_task_enter_user();
-    fn __i686_task_enter_from_fork();
-    fn __i686_enter_task(to: *mut Inner) -> !;
-    fn __i686_switch_task(to: *mut Inner, from: *mut Inner);
-    fn __i686_switch_and_drop(to: *mut Inner, from: *const ());
-}
-
-impl TaskFrame for SyscallFrame {
-    fn store(&self) -> SavedFrame {
-        SavedFrame {
-            eax: self.eax as _,
-            ecx: self.args[1] as _,
-            edx: self.args[2] as _,
-            ebx: self.args[0] as _,
-            ebp: self.args[5] as _,
-            esi: self.args[3] as _,
-            edi: self.args[4] as _,
-
-            user_ip: self.eip,
-            user_sp: self.esp,
-            eflags: self.eflags,
-        }
-    }
-
-    fn restore(&mut self, saved: &SavedFrame) {
-        self.eax = saved.eax as _;
-        self.args[0] = saved.ebx as _;
-        self.args[1] = saved.ecx as _;
-        self.args[2] = saved.edx as _;
-        self.args[3] = saved.esi as _;
-        self.args[4] = saved.edi as _;
-        self.args[5] = saved.ebp as _;
-
-        self.eip = saved.user_ip;
-        self.esp = saved.user_sp;
-        self.eflags = saved.eflags;
-    }
-
-    fn user_sp(&self) -> usize {
-        todo!()
-    }
-
-    fn user_ip(&self) -> usize {
-        todo!()
-    }
-
-    fn argument(&self) -> u64 {
-        self.args[0] as _
-    }
-
-    fn set_user_sp(&mut self, value: usize) {
-        self.esp = value as _;
-    }
-
-    fn set_user_ip(&mut self, value: usize) {
-        self.eip = value as _;
-    }
-
-    fn set_argument(&mut self, value: u64) {
-        // TODO implement ABI for passing 64-bit values via EAX/EDX
-        if value & (1 << 63) != 0 {
-            assert_eq!(value & 0xFFFFFFFF00000000, 0xFFFFFFFF00000000);
-        }
-
-        self.eax = value as usize;
-    }
-
-    fn set_single_step(&mut self, _step: bool) {
-        todo!()
-    }
-
-    fn set_return_value(&mut self, value: u64) {
-        // TODO implement ABI for returning 64-bit values via EAX/EDX
-        if value & (1 << 63) != 0 {
-            assert_eq!(value & 0xFFFFFFFF00000000, 0xFFFFFFFF00000000);
-        }
-
-        self.eax = value as usize;
-    }
-}
-
-impl TaskFrame for InterruptFrame {
-    fn store(&self) -> SavedFrame {
-        SavedFrame {
-            eax: self.eax,
-            ecx: self.ecx,
-            edx: self.edx,
-            ebx: self.ebx,
-            ebp: self.ebp,
-            esi: self.esi,
-            edi: self.edi,
-
-            user_ip: self.eip,
-            user_sp: self.esp,
-            eflags: self.eflags,
-        }
-    }
-
-    fn restore(&mut self, _saved: &SavedFrame) {
-        todo!()
-    }
-
-    fn user_sp(&self) -> usize {
-        todo!()
-    }
-
-    fn user_ip(&self) -> usize {
-        todo!()
-    }
-
-    fn argument(&self) -> u64 {
-        todo!()
-    }
-
-    fn set_user_sp(&mut self, value: usize) {
-        self.esp = value as u32;
-    }
-
-    fn set_user_ip(&mut self, value: usize) {
-        self.eip = value as u32;
-    }
-
-    fn set_argument(&mut self, value: u64) {
-        // TODO implement ABI for returning 64-bit values via EAX/EDX
-        if value & (1 << 63) != 0 {
-            assert_eq!(value & 0xFFFFFFFF00000000, 0xFFFFFFFF00000000);
-        }
-
-        self.eax = value as u32;
-    }
-
-    fn set_single_step(&mut self, step: bool) {
-        if step {
-            self.eflags |= 1 << 8;
-        } else {
-            self.eflags &= !(1 << 8);
-        }
-    }
-
-    fn set_return_value(&mut self, value: u64) {
-        // TODO implement ABI for returning 64-bit values via EAX/EDX
-        if value & (1 << 63) != 0 {
-            assert_eq!(value & 0xFFFFFFFF00000000, 0xFFFFFFFF00000000);
-        }
-
-        self.eax = value as u32;
-    }
-}
-
-impl TaskFrame for ExceptionFrame {
-    fn store(&self) -> SavedFrame {
-        SavedFrame {
-            eax: self.eax,
-            ecx: self.ecx,
-            edx: self.edx,
-            ebx: self.ebx,
-            ebp: self.ebp,
-            esi: self.esi,
-            edi: self.edi,
-
-            user_ip: self.eip,
-            user_sp: self.esp,
-            eflags: self.eflags,
-        }
-    }
-
-    fn restore(&mut self, _saved: &SavedFrame) {
-        todo!()
-    }
-
-    fn user_sp(&self) -> usize {
-        todo!()
-    }
-
-    fn user_ip(&self) -> usize {
-        self.eip as _
-    }
-
-    fn argument(&self) -> u64 {
-        todo!()
-    }
-
-    fn set_user_sp(&mut self, value: usize) {
-        self.esp = value as u32;
-    }
-
-    fn set_user_ip(&mut self, value: usize) {
-        self.eip = value as u32;
-    }
-
-    fn set_argument(&mut self, value: u64) {
-        // TODO implement ABI for returning 64-bit values via EAX/EDX
-        if value & (1 << 63) != 0 {
-            assert_eq!(value & 0xFFFFFFFF00000000, 0xFFFFFFFF00000000);
-        }
-
-        self.eax = value as u32;
-    }
-
-    fn set_single_step(&mut self, step: bool) {
-        if step {
-            self.eflags |= 1 << 8;
-        } else {
-            self.eflags &= !(1 << 8);
-        }
-    }
-
-    fn set_return_value(&mut self, value: u64) {
-        // TODO implement ABI for returning 64-bit values via EAX/EDX
-        if value & (1 << 63) != 0 {
-            assert_eq!(value & 0xFFFFFFFF00000000, 0xFFFFFFFF00000000);
-        }
-
-        self.eax = value as u32;
-    }
-}
-
-global_asm!(include_str!("context.S"), options(att_syntax));

kernel/arch/i686/src/gdt.rs

@@ -1,108 +0,0 @@
-use core::{cell::UnsafeCell, ptr::addr_of_mut};
-
-use kernel_arch_interface::guard::IrqGuard;
-pub use kernel_arch_x86::gdt::{Entry, Pointer};
-
-use crate::ArchitectureImpl;
-
-#[allow(dead_code)]
-#[repr(C, packed)]
-pub struct Tss {
-    prev_tss: u32,
-    pub esp0: u32,
-    pub ss0: u16,
-    _res0: u16,
-    esp1: u32,
-    ss1: u16,
-    _res1: u16,
-    esp2: u32,
-    ss2: u16,
-    _res2: u16,
-    cr3: u32,
-    eip: u32,
-    eflags: u32,
-    eax: u32,
-    ecx: u32,
-    edx: u32,
-    ebx: u32,
-    esp: u32,
-    ebp: u32,
-    esi: u32,
-    edi: u32,
-    es: u32,
-    cs: u32,
-    ss: u32,
-    ds: u32,
-    fs: u32,
-    gs: u32,
-    ldt: u32,
-    trap: u16,
-    iomap_base: u16,
-}
-
-impl Tss {
-    const NULL: Self = Self {
-        prev_tss: 0,
-        esp0: 0,
-        ss0: 0x10,
-        _res0: 0,
-        esp1: 0,
-        ss1: 0,
-        _res1: 0,
-        esp2: 0,
-        ss2: 0,
-        _res2: 0,
-        cr3: 0,
-        eip: 0,
-        eflags: 0,
-        eax: 0,
-        ecx: 0,
-        edx: 0,
-        ebx: 0,
-        esp: 0,
-        ebp: 0,
-        esi: 0,
-        edi: 0,
-        es: 0,
-        cs: 0,
-        ss: 0,
-        ds: 0,
-        fs: 0,
-        gs: 0,
-        ldt: 0,
-        trap: 0,
-        iomap_base: 0,
-    };
-}
-
-pub static mut TSS: Tss = Tss::NULL;
-pub static mut GDT: UnsafeCell<[Entry; 7]> = UnsafeCell::new([
-    Entry::NULL,       // 0x00
-    Entry::RING0_CS32, // 0x08
-    Entry::RING0_DS32, // 0x10
-    Entry::RING3_CS32, // 0x1B
-    Entry::RING3_DS32, // 0x23
-    Entry::NULL,       // 0x28, TSS
-    Entry::RING3_GS32, // 0x33, Task GS
-]);
-
-pub fn create_gdt() -> (&'static [Entry], &'static Tss) {
-    // Won't be deallocated, so leaks are not a concern
-    let tss = unsafe { &mut *addr_of_mut!(TSS) };
-    tss.ss0 = 0x10;
-    let tss_addr = (tss as *mut Tss).addr();
-    #[allow(static_mut_refs)]
-    let gdt = unsafe { GDT.get_mut() };
-    gdt[5] = Entry::tss(tss_addr as u32, (size_of::<Tss>() - 1) as u32);
-
-    (gdt, tss)
-}
-
-pub fn set_gs_base(gs_base: usize) {
-    let _guard = IrqGuard::<ArchitectureImpl>::acquire();
-    unsafe {
-        #[allow(static_mut_refs)]
-        GDT.get_mut()[6].set_base(gs_base);
-        core::arch::asm!("mov $0x33, %ax; mov %ax, %gs", out("ax") _, options(att_syntax, nostack));
-    }
-}

kernel/arch/i686/src/lib.rs

@@ -1,140 +0,0 @@
-#![feature(never_type, naked_functions, trace_macros)]
-#![no_std]
-
-extern crate alloc;
-
-use core::ptr::null_mut;
-
-use alloc::vec::Vec;
-use device_api::interrupt::{LocalInterruptController, MessageInterruptController};
-use kernel_arch_interface::{
-    cpu::{CpuImpl, IpiQueue},
-    task::Scheduler,
-    Architecture,
-};
-
-pub mod context;
-pub mod gdt;
-pub mod mem;
-
-pub use context::TaskContextImpl;
-use kernel_arch_x86::cpuid::CpuFeatures;
-pub use mem::{KernelTableManagerImpl, ProcessAddressSpaceImpl};
-
-pub struct ArchitectureImpl;
-
-#[repr(C)]
-pub struct PerCpuData {
-    pub available_features: CpuFeatures,
-    pub enabled_features: CpuFeatures,
-}
-
-static mut CPU: *mut () = null_mut();
-
-#[naked]
-extern "C" fn idle_task(_: usize) -> ! {
-    unsafe {
-        core::arch::naked_asm!(
-            r#"
-        1:
-            nop
-            jmp 1b
-        "#,
-            options(att_syntax)
-        );
-    }
-}
-
-impl Architecture for ArchitectureImpl {
-    type PerCpuData = PerCpuData;
-    type CpuFeatures = CpuFeatures;
-    type BreakpointType = u8;
-
-    const BREAKPOINT_VALUE: Self::BreakpointType = 0xCC;
-
-    unsafe fn init_local_cpu<S: Scheduler + 'static>(id: Option<u32>, data: Self::PerCpuData) {
-        use alloc::boxed::Box;
-
-        let cpu = Box::leak(Box::new(CpuImpl::<Self, S>::new(
-            id.expect("x86_64 required manual CPU ID set"),
-            data,
-        )));
-
-        cpu.set_local();
-    }
-
-    unsafe fn set_interrupt_mask(mask: bool) -> bool {
-        let old = Self::interrupt_mask();
-        if mask {
-            core::arch::asm!("cli");
-        } else {
-            core::arch::asm!("sti");
-        }
-        old
-    }
-
-    fn interrupt_mask() -> bool {
-        let mut flags: u32;
-        unsafe {
-            core::arch::asm!("pushfl; pop {0:e}", out(reg) flags, options(att_syntax));
-        }
-        // If IF is zero, interrupts are disabled (masked)
-        flags & (1 << 9) == 0
-    }
-
-    fn wait_for_interrupt() {
-        unsafe {
-            core::arch::asm!("hlt");
-        }
-    }
-
-    unsafe fn init_ipi_queues(_queues: Vec<IpiQueue<Self>>) {}
-
-    fn local_cpu() -> *mut () {
-        unsafe { CPU }
-    }
-
-    fn cpu_index<S: Scheduler + 'static>() -> u32 {
-        0
-    }
-
-    unsafe fn set_local_cpu(cpu: *mut ()) {
-        CPU = cpu;
-    }
-
-    fn cpu_count() -> usize {
-        1
-    }
-
-    fn message_interrupt_controller() -> &'static dyn MessageInterruptController {
-        unimplemented!()
-    }
-
-    fn local_interrupt_controller() -> Option<&'static dyn LocalInterruptController> {
-        None
-    }
-
-    fn ipi_queue(_cpu_id: u32) -> Option<&'static IpiQueue<Self>> {
-        None
-    }
-
-    fn idle_task() -> extern "C" fn(usize) -> ! {
-        idle_task
-    }
-
-    fn halt() -> ! {
-        loop {
-            unsafe {
-                core::arch::asm!("cli; hlt");
-            }
-        }
-    }
-
-    fn cpu_available_features<S: Scheduler>(cpu: &CpuImpl<Self, S>) -> Option<&Self::CpuFeatures> {
-        Some(&cpu.available_features)
-    }
-
-    fn cpu_enabled_features<S: Scheduler>(cpu: &CpuImpl<Self, S>) -> Option<&Self::CpuFeatures> {
-        Some(&cpu.enabled_features)
-    }
-}

kernel/arch/i686/src/mem/fixed.rs

@@ -1,141 +0,0 @@
-use kernel_arch_interface::{sync::IrqSafeSpinlock, KERNEL_VIRT_OFFSET};
-use libk_mm_interface::{address::PhysicalAddress, table::EntryLevel, KernelImageObject};
-use yggdrasil_abi::error::Error;
-
-use crate::{
-    mem::{flush_tlb_entry, table::PageAttributes},
-    ArchitectureImpl,
-};
-
-use super::{
-    table::{PageEntry, PageTable, L0, L3},
-    KERNEL_TABLES,
-};
-
-pub const KERNEL_SPLIT_L0: usize = KERNEL_VIRT_OFFSET >> 22;
-pub const DYNAMIC_MAP_COUNT: usize = 64;
-pub const FIXED_MAP_COUNT: usize = 1024 - (KERNEL_SPLIT_L0 + DYNAMIC_MAP_COUNT);
-pub const DYNAMIC_MAP_OFFSET: usize = (KERNEL_SPLIT_L0 + FIXED_MAP_COUNT) << L0::SHIFT;
-pub const MAX_FIXED_PHYSICAL: PhysicalAddress =
-    PhysicalAddress::from_u64((FIXED_MAP_COUNT as u64) << 22);
-
-#[repr(C)]
-pub struct FixedTables {
-    pub l0: KernelL0,
-    pub dynamic: IrqSafeSpinlock<ArchitectureImpl, KernelDynamic>,
-}
-
-#[repr(C, align(0x1000))]
-pub struct KernelL0 {
-    pub lower: [PageEntry<L0>; KERNEL_SPLIT_L0],
-    pub kernel: [PageEntry<L0>; FIXED_MAP_COUNT],
-    pub dynamic: [PageEntry<L0>; DYNAMIC_MAP_COUNT],
-}
-
-#[allow(unused)]
-pub struct KernelDynamic {
-    pub l3s: [KernelImageObject<PageTable<L3>>; DYNAMIC_MAP_COUNT],
-    free: usize,
-}
-
-impl FixedTables {
-    pub const fn zeroed() -> Self {
-        Self {
-            l0: KernelL0::zeroed(),
-            dynamic: IrqSafeSpinlock::new(KernelDynamic::zeroed()),
-        }
-    }
-
-    pub fn virtualize(&mut self, address: PhysicalAddress) -> usize {
-        if address < MAX_FIXED_PHYSICAL {
-            // It's a fixed address
-            address.into_u64() as usize + KERNEL_VIRT_OFFSET
-        } else {
-            todo!()
-        }
-    }
-
-    pub fn physicalize(&mut self, address: usize) -> Option<PhysicalAddress> {
-        if address < KERNEL_VIRT_OFFSET {
-            return None;
-        }
-
-        if address < KERNEL_VIRT_OFFSET + MAX_FIXED_PHYSICAL.into_u64() as usize {
-            // It's a fixed address
-            Some(PhysicalAddress::from_usize(address - KERNEL_VIRT_OFFSET))
-        } else {
-            todo!()
-        }
-    }
-
-    pub fn map_dynamic_memory(&mut self, base: u64, page_count: usize) -> Result<usize, Error> {
-        self.dynamic.lock().map(base, page_count)
-    }
-}
-
-impl KernelL0 {
-    pub const fn zeroed() -> Self {
-        Self {
-            lower: [PageEntry::INVALID; KERNEL_SPLIT_L0],
-            kernel: [PageEntry::INVALID; FIXED_MAP_COUNT],
-            dynamic: [PageEntry::INVALID; DYNAMIC_MAP_COUNT],
-        }
-    }
-}
-
-impl KernelDynamic {
-    pub const fn zeroed() -> Self {
-        Self {
-            l3s: [const { unsafe { KernelImageObject::new(PageTable::zeroed()) } };
-                DYNAMIC_MAP_COUNT],
-            free: DYNAMIC_MAP_COUNT * 1024,
-        }
-    }
-
-    fn map(&mut self, base: u64, page_count: usize) -> Result<usize, Error> {
-        if page_count > self.free {
-            return Err(Error::OutOfMemory);
-        }
-
-        'l0: for i in 0..DYNAMIC_MAP_COUNT * 1024 - page_count {
-            for j in 0..page_count {
-                let entry = self.entry(i + j);
-                if entry.is_present() {
-                    continue 'l0;
-                }
-            }
-
-            self.free -= page_count;
-            for j in 0..page_count {
-                let address = PhysicalAddress::from_u64(base + ((j as u64) << L3::SHIFT));
-                *self.entry_mut(i + j) = PageEntry::page(address, PageAttributes::WRITABLE);
-                unsafe {
-                    flush_tlb_entry(DYNAMIC_MAP_OFFSET + ((i + j) << L3::SHIFT));
-                }
-            }
-            let addr = DYNAMIC_MAP_OFFSET + (i << L3::SHIFT);
-            return Ok(addr);
-        }
-
-        Err(Error::OutOfMemory)
-    }
-
-    fn entry(&self, index: usize) -> &PageEntry<L3> {
-        &self.l3s[index / 1024][index % 1024]
-    }
-
-    fn entry_mut(&mut self, index: usize) -> &mut PageEntry<L3> {
-        &mut self.l3s[index / 1024][index % 1024]
-    }
-}
-
-pub fn clone_kernel_tables(dst: &mut PageTable<L0>) {
-    let tables = KERNEL_TABLES.lock();
-    for (i, entry) in tables.l0.kernel.iter().enumerate() {
-        dst[i + KERNEL_SPLIT_L0] = *entry;
-    }
-
-    for (i, entry) in tables.l0.dynamic.iter().enumerate() {
-        dst[i + KERNEL_SPLIT_L0 + FIXED_MAP_COUNT] = *entry;
-    }
-}

kernel/arch/i686/src/mem/mod.rs

@@ -1,125 +0,0 @@
-use fixed::FixedTables;
-use kernel_arch_interface::{
-    mem::{DeviceMemoryAttributes, KernelTableManager, RawDeviceMemoryMapping},
-    sync::split_spinlock,
-    KERNEL_VIRT_OFFSET,
-};
-use libk_mm_interface::{
-    address::{AsPhysicalAddress, PhysicalAddress},
-    table::{page_count, EntryLevel},
-};
-use table::{PageAttributes, PageEntry, L0, L3};
-use yggdrasil_abi::error::Error;
-
-pub mod fixed;
-pub mod process;
-pub mod table;
-
-pub use process::ProcessAddressSpaceImpl;
-
-#[derive(Debug)]
-pub struct KernelTableManagerImpl;
-
-split_spinlock! {
-    use libk_mm_interface::KernelImageObject;
-    use crate::mem::FixedTables;
-    use crate::ArchitectureImpl;
-
-    #[link_section = ".data.tables"]
-    static KERNEL_TABLES<lock: ArchitectureImpl>: KernelImageObject<FixedTables> = unsafe {
-        KernelImageObject::new(FixedTables::zeroed())
-    };
-}
-
-impl KernelTableManager for KernelTableManagerImpl {
-    unsafe fn map_device_pages(
-        base: u64,
-        count: usize,
-        _attrs: DeviceMemoryAttributes,
-    ) -> Result<RawDeviceMemoryMapping<Self>, Error> {
-        // TODO page align up
-        let end = base + count as u64;
-        assert_eq!(base & 0xFFF, 0);
-        if end < fixed::MAX_FIXED_PHYSICAL.into_u64() {
-            // 1:1
-            let address = Self::virtualize(base);
-            Ok(RawDeviceMemoryMapping::from_raw_parts(
-                address, address, 0, 0,
-            ))
-        } else {
-            assert_eq!(base & 0xFFF, 0);
-            log::info!("map_device_pages({:#x}, {})", base, count);
-            let page_count = page_count::<L3>(count);
-            let virt = KERNEL_TABLES.lock().map_dynamic_memory(base, page_count)?;
-
-            Ok(RawDeviceMemoryMapping::from_raw_parts(
-                virt, virt, page_count, 0,
-            ))
-        }
-    }
-
-    unsafe fn unmap_device_pages(_mapping: &RawDeviceMemoryMapping<Self>) {
-        todo!()
-    }
-
-    fn virtualize(phys: u64) -> usize {
-        KERNEL_TABLES
-            .lock()
-            .virtualize(PhysicalAddress::from_u64(phys))
-    }
-
-    fn physicalize(virt: usize) -> u64 {
-        KERNEL_TABLES
-            .lock()
-            .physicalize(virt)
-            .expect("Invalid virtual address")
-            .into_u64()
-    }
-
-    unsafe fn unmap_physical_address(virt: usize) {
-        if virt < KERNEL_VIRT_OFFSET {
-            panic!("Invalid 'virtualized' address: {:#x}", virt);
-        }
-        let virt = virt - KERNEL_VIRT_OFFSET;
-        if virt >= fixed::FIXED_MAP_COUNT << L0::SHIFT {
-            todo!()
-        }
-    }
-}
-
-/// Sets up fixed MMU translation tables.
-///
-/// # Safety
-///
-/// Only meant to be called once during early OS init.
-pub unsafe fn init_fixed_tables() {
-    let mut tables = KERNEL_TABLES.lock();
-
-    // Unmap lower stuff
-    for (i, entry) in tables.l0.lower.iter_mut().enumerate() {
-        *entry = PageEntry::INVALID;
-        flush_tlb_entry(i << 22);
-    }
-
-    // Map the rest of fixed translation
-    for (i, entry) in tables.l0.kernel.iter_mut().enumerate() {
-        let virt = KERNEL_VIRT_OFFSET + (i << L0::SHIFT);
-        let phys = (i << L0::SHIFT) as u32;
-        *entry = PageEntry::block(PhysicalAddress::from_u32(phys), PageAttributes::WRITABLE);
-        flush_tlb_entry(virt);
-    }
-
-    let dynamic_len = tables.l0.dynamic.len();
-    for i in 0..dynamic_len {
-        let phys = tables.dynamic.lock().l3s[i].as_physical_address();
-        tables.l0.dynamic[i] = PageEntry::table(phys, PageAttributes::WRITABLE);
-    }
-}
-
-/// # Safety
-///
-/// `address` must be page-aligned.
-#[inline]
-pub unsafe fn flush_tlb_entry(address: usize) {
-    core::arch::asm!("invlpg ({0})", in(reg) address, options(att_syntax));
-}

kernel/arch/i686/src/mem/process.rs

@@ -1,137 +0,0 @@
-use core::marker::PhantomData;
-
-use kernel_arch_interface::KERNEL_VIRT_OFFSET;
-use libk_mm_interface::{
-    address::{AsPhysicalAddress, PhysicalAddress},
-    pointer::PhysicalRefMut,
-    process::ProcessAddressSpaceManager,
-    table::{
-        EntryLevel, EntryLevelDrop, EntryLevelExt, MapAttributes, NextPageTable, TableAllocator,
-    },
-};
-use yggdrasil_abi::error::Error;
-
-use crate::{mem::flush_tlb_entry, KernelTableManagerImpl};
-
-use super::{
-    fixed::{clone_kernel_tables, KERNEL_SPLIT_L0},
-    table::{PageEntry, PageTable, L0, L3},
-};
-
-#[repr(C)]
-pub struct ProcessAddressSpaceImpl<TA: TableAllocator> {
-    l0: PhysicalRefMut<'static, PageTable<L0>, KernelTableManagerImpl>,
-    _alloc: PhantomData<TA>,
-}
-
-impl<TA: TableAllocator> ProcessAddressSpaceManager<TA> for ProcessAddressSpaceImpl<TA> {
-    const UPPER_LIMIT_PFN: usize = KERNEL_VIRT_OFFSET >> L3::SHIFT;
-    const LOWER_LIMIT_PFN: usize = 32;
-
-    fn new() -> Result<Self, Error> {
-        let mut l0 = unsafe {
-            PhysicalRefMut::<'static, PageTable<L0>, KernelTableManagerImpl>::map(
-                TA::allocate_page_table()?,
-            )
-        };
-
-        for i in 0..1024 {
-            l0[i] = PageEntry::INVALID;
-        }
-
-        clone_kernel_tables(&mut l0);
-
-        Ok(Self {
-            l0,
-            _alloc: PhantomData,
-        })
-    }
-
-    unsafe fn clear(&mut self) {
-        self.l0.drop_range::<TA>(0..KERNEL_SPLIT_L0);
-    }
-
-    fn translate(&self, address: usize) -> Result<(PhysicalAddress, MapAttributes), Error> {
-        self.read_l3_entry(address).ok_or(Error::DoesNotExist)
-    }
-
-    unsafe fn map_page(
-        &mut self,
-        address: usize,
-        physical: PhysicalAddress,
-        flags: MapAttributes,
-    ) -> Result<(), Error> {
-        self.write_l3_entry(address, PageEntry::page(physical, flags.into()), false)
-    }
-
-    unsafe fn unmap_page(&mut self, address: usize) -> Result<PhysicalAddress, Error> {
-        self.pop_l3_entry(address)
-    }
-
-    fn as_address_with_asid(&self) -> u64 {
-        unsafe { self.l0.as_physical_address().into_u64() }
-    }
-}
-
-impl<TA: TableAllocator> ProcessAddressSpaceImpl<TA> {
-    // Write a single 4KiB entry
-    fn write_l3_entry(
-        &mut self,
-        virt: usize,
-        entry: PageEntry<L3>,
-        overwrite: bool,
-    ) -> Result<(), Error> {
-        let l0i = virt.page_index::<L0>();
-        let l3i = virt.page_index::<L3>();
-
-        let mut l3 = self.l0.get_mut_or_alloc::<TA>(l0i)?;
-
-        if l3[l3i].is_present() && !overwrite {
-            todo!();
-        }
-
-        l3[l3i] = entry;
-        unsafe {
-            flush_tlb_entry(virt);
-        }
-
-        Ok(())
-    }
-
-    fn pop_l3_entry(&mut self, virt: usize) -> Result<PhysicalAddress, Error> {
-        let l0i = virt.page_index::<L0>();
-        let l3i = virt.page_index::<L3>();
-
-        let mut l3 = self.l0.get_mut(l0i).ok_or(Error::DoesNotExist)?;
-        let page = l3[l3i].as_page().ok_or(Error::DoesNotExist)?;
-
-        l3[l3i] = PageEntry::INVALID;
-        unsafe {
-            flush_tlb_entry(virt);
-        }
-
-        Ok(page)
-    }
-
-    fn read_l3_entry(&self, virt: usize) -> Option<(PhysicalAddress, MapAttributes)> {
-        let l0i = virt.page_index::<L0>();
-        let l3i = virt.page_index::<L3>();
-
-        let l3 = self.l0.get(l0i)?;
-        let page = l3[l3i].as_page()?;
-
-        Some((page.add(virt & 0xFFF), l3[l3i].attributes().into()))
-    }
-}
-
-impl<TA: TableAllocator> Drop for ProcessAddressSpaceImpl<TA> {
-    fn drop(&mut self) {
-        // SAFETY: with safe usage of the ProcessAddressSpaceImpl, clearing and dropping
-        // is safe, no one refers to the memory
-        unsafe {
-            self.clear();
-            let l0_phys = self.l0.as_physical_address();
-            TA::free_page_table(l0_phys);
-        }
-    }
-}

kernel/arch/i686/src/mem/table.rs

@@ -1,256 +0,0 @@
-use core::{
-    marker::PhantomData,
-    ops::{Index, IndexMut, Range},
-};
-
-use bitflags::bitflags;
-use libk_mm_interface::{
-    address::{AsPhysicalAddress, PhysicalAddress},
-    pointer::{PhysicalRef, PhysicalRefMut},
-    table::{
-        EntryLevel, EntryLevelDrop, MapAttributes, NextPageTable, NonTerminalEntryLevel,
-        TableAllocator,
-    },
-};
-use yggdrasil_abi::error::Error;
-
-use crate::KernelTableManagerImpl;
-
-bitflags! {
-    /// Describes how each page table entry is mapped
-    pub struct PageAttributes: u32 {
-        /// When set, the mapping is considered valid and pointing somewhere
-        const PRESENT = 1 << 0;
-        /// For tables, allows writes to further translation levels, for pages/blocks, allows
-        /// writes to the region covered by the entry
-        const WRITABLE = 1 << 1;
-        /// When set for L2 entries, the mapping specifies a 2MiB page instead of a page table
-        /// reference
-        const BLOCK = 1 << 7;
-        /// For tables, allows user access to further translation levels, for pages/blocks, allows
-        /// user access to the region covered by the entry
-        const USER = 1 << 2;
-    }
-}
-// TODO stuff for PAE?
-
-#[derive(Debug, Clone, Copy)]
-pub struct L3;
-#[derive(Debug, Clone, Copy)]
-pub struct L0;
-
-#[derive(Clone, Copy, Debug)]
-pub struct PageEntry<L: EntryLevel>(u32, PhantomData<L>);
-
-#[derive(Clone, Copy, Debug)]
-#[repr(C, align(0x1000))]
-pub struct PageTable<L: EntryLevel> {
-    data: [PageEntry<L>; 1024],
-}
-
-impl EntryLevel for L3 {
-    const SHIFT: usize = 12;
-}
-
-impl EntryLevel for L0 {
-    const SHIFT: usize = 22;
-}
-
-impl NonTerminalEntryLevel for L0 {
-    type NextLevel = L3;
-}
-
-impl PageEntry<L3> {
-    pub fn page(address: PhysicalAddress, attrs: PageAttributes) -> Self {
-        Self(
-            address.try_into_u32().unwrap() | (PageAttributes::PRESENT | attrs).bits(),
-            PhantomData,
-        )
-    }
-
-    pub fn as_page(&self) -> Option<PhysicalAddress> {
-        if self.0 & PageAttributes::PRESENT.bits() != 0 {
-            Some(PhysicalAddress::from_u32(self.0 & !0xFFF))
-        } else {
-            None
-        }
-    }
-}
-
-impl PageEntry<L0> {
-    pub fn block(address: PhysicalAddress, attrs: PageAttributes) -> Self {
-        Self(
-            address.try_into_u32().unwrap()
-                | (PageAttributes::PRESENT | PageAttributes::BLOCK | attrs).bits(),
-            PhantomData,
-        )
-    }
-
-    pub fn table(address: PhysicalAddress, attrs: PageAttributes) -> Self {
-        Self(
-            address.try_into_u32().unwrap() | (PageAttributes::PRESENT | attrs).bits(),
-            PhantomData,
-        )
-    }
-
-    pub fn as_table(&self) -> Option<PhysicalAddress> {
-        if self.0 & PageAttributes::PRESENT.bits() != 0
-            && self.0 & PageAttributes::BLOCK.bits() == 0
-        {
-            Some(PhysicalAddress::from_u32(self.0 & !0xFFF))
-        } else {
-            None
-        }
-    }
-}
-
-impl<L: EntryLevel> PageEntry<L> {
-    pub const INVALID: Self = Self(0, PhantomData);
-
-    pub fn is_present(&self) -> bool {
-        self.0 & (1 << 0) != 0
-    }
-
-    pub fn attributes(&self) -> PageAttributes {
-        PageAttributes::from_bits_retain(self.0)
-    }
-}
-
-impl<L: EntryLevel> PageTable<L> {
-    pub const fn zeroed() -> Self {
-        Self {
-            data: [PageEntry::INVALID; 1024],
-        }
-    }
-
-    pub fn new_zeroed<'a, TA: TableAllocator>(
-    ) -> Result<PhysicalRefMut<'a, Self, KernelTableManagerImpl>, Error> {
-        let physical = TA::allocate_page_table()?;
-        let mut table =
-            unsafe { PhysicalRefMut::<'a, Self, KernelTableManagerImpl>::map(physical) };
-
-        for i in 0..1024 {
-            table[i] = PageEntry::INVALID;
-        }
-
-        Ok(table)
-    }
-
-    /// Recursively clears and deallocates the translation table.
-    ///
-    /// # Safety
-    ///
-    /// The caller must ensure the table is no longer in use and is not referenced anymore.
-    pub unsafe fn free<TA: TableAllocator>(this: PhysicalRefMut<Self, KernelTableManagerImpl>) {
-        let physical = this.as_physical_address();
-        TA::free_page_table(physical);
-    }
-}
-
-impl NextPageTable for PageTable<L0> {
-    type NextLevel = PageTable<L3>;
-    type TableRef = PhysicalRef<'static, Self::NextLevel, KernelTableManagerImpl>;
-    type TableRefMut = PhysicalRefMut<'static, Self::NextLevel, KernelTableManagerImpl>;
-
-    fn get(&self, index: usize) -> Option<Self::TableRef> {
-        self[index]
-            .as_table()
-            .map(|addr| unsafe { PhysicalRef::map(addr) })
-    }
-
-    fn get_mut(&mut self, index: usize) -> Option<Self::TableRefMut> {
-        self[index]
-            .as_table()
-            .map(|addr| unsafe { PhysicalRefMut::map(addr) })
-    }
-
-    fn get_mut_or_alloc<TA: TableAllocator>(
-        &mut self,
-        index: usize,
-    ) -> Result<Self::TableRefMut, Error> {
-        let entry = self[index];
-
-        if let Some(table) = entry.as_table() {
-            Ok(unsafe { PhysicalRefMut::map(table) })
-        } else {
-            let table = PageTable::new_zeroed::<TA>()?;
-            self[index] = PageEntry::<L0>::table(
-                unsafe { table.as_physical_address() },
-                PageAttributes::WRITABLE | PageAttributes::USER,
-            );
-            Ok(table)
-        }
-    }
-}
-
-impl<L: EntryLevel> Index<usize> for PageTable<L> {
-    type Output = PageEntry<L>;
-
-    fn index(&self, index: usize) -> &Self::Output {
-        &self.data[index]
-    }
-}
-
-impl<L: EntryLevel> IndexMut<usize> for PageTable<L> {
-    fn index_mut(&mut self, index: usize) -> &mut Self::Output {
-        &mut self.data[index]
-    }
-}
-
-impl EntryLevelDrop for PageTable<L3> {
-    const FULL_RANGE: Range<usize> = 0..1024;
-
-    unsafe fn drop_range<TA: TableAllocator>(&mut self, _range: Range<usize>) {}
-}
-
-impl EntryLevelDrop for PageTable<L0> {
-    const FULL_RANGE: Range<usize> = 0..1024;
-
-    unsafe fn drop_range<TA: TableAllocator>(&mut self, range: Range<usize>) {
-        for index in range {
-            let entry = self[index];
-
-            if let Some(table) = entry.as_table() {
-                let mut table_ref: PhysicalRefMut<PageTable<L3>, KernelTableManagerImpl> =
-                    PhysicalRefMut::map(table);
-
-                table_ref.drop_all::<TA>();
-
-                TA::free_page_table(table);
-            } else if entry.is_present() {
-                // Memory must've been cleared beforehand, so no non-table entries must be present
-                panic!(
-                    "Expected a table containing only tables, got table[{}] = {:#x?}",
-                    index, entry.0
-                );
-            }
-
-            self[index] = PageEntry::INVALID;
-        }
-    }
-}
-
-impl From<MapAttributes> for PageAttributes {
-    fn from(value: MapAttributes) -> Self {
-        let mut res = PageAttributes::WRITABLE;
-        if value.intersects(MapAttributes::USER_READ | MapAttributes::USER_WRITE) {
-            res |= PageAttributes::USER;
-        }
-        res
-    }
-}
-
-impl From<PageAttributes> for MapAttributes {
-    fn from(value: PageAttributes) -> Self {
-        let mut res = MapAttributes::empty();
-        if value.contains(PageAttributes::USER) {
-            res |= MapAttributes::USER_READ;
-            if value.contains(PageAttributes::WRITABLE) {
-                res |= MapAttributes::USER_WRITE;
-            }
-        }
-        // TODO ???
-        res |= MapAttributes::NON_GLOBAL;
-        res
-    }
-}

kernel/arch/interface/src/cpu.rs

@@ -30,6 +30,18 @@ pub struct IpiQueue<A: Architecture> {
     data: IrqSafeSpinlock<A, Option<IpiMessage>>,
 }
 
+pub trait CpuData {
+    fn is_bootstrap(&self, id: u32) -> bool {
+        // On most architectures
+        id == 0
+    }
+
+    fn queue_index(&self, id: u32) -> usize {
+        // On most architectures
+        id as usize
+    }
+}
+
 pub trait CpuFeatureSet {
     fn iter(&self) -> impl Iterator<Item = &'static str>;
 }
@@ -50,6 +62,14 @@ impl<A: Architecture, S: Scheduler + 'static> CpuImpl<A, S> {
         unsafe { A::init_ipi_queues(queues) }
     }
 
+    pub fn is_bootstrap(&self) -> bool {
+        self.inner.is_bootstrap(self.id)
+    }
+
+    pub fn queue_index(&self) -> usize {
+        self.inner.queue_index(self.id)
+    }
+
     pub fn set_current_thread_id(&mut self, id: Option<S::ThreadId>) {
         self.current_thread_id = id;
     }

kernel/arch/interface/src/lib.rs

@@ -2,13 +2,18 @@
 #![feature(step_trait, const_trait_impl, never_type, decl_macro)]
 #![allow(clippy::new_without_default)]
 
+use core::ops::Range;
+
 use alloc::vec::Vec;
-use cpu::{CpuFeatureSet, CpuImpl, IpiQueue};
-use device_api::interrupt::{LocalInterruptController, MessageInterruptController};
+use cpu::{CpuData, CpuFeatureSet, CpuImpl, IpiQueue};
+use device_api::interrupt::LocalInterruptController;
 use task::Scheduler;
 
 extern crate alloc;
 
+#[macro_use]
+pub mod macros;
+
 pub mod cpu;
 pub mod guard;
 pub mod mem;
@@ -16,14 +21,13 @@ pub mod sync;
 pub mod task;
 pub mod util;
 
-#[cfg(any(target_pointer_width = "32", rust_analyzer))]
-pub const KERNEL_VIRT_OFFSET: usize = 0xC0000000;
-
-#[cfg(any(target_pointer_width = "64", rust_analyzer))]
+#[cfg(any(target_arch = "aarch64", target_arch = "x86_64", rust_analyzer))]
 pub const KERNEL_VIRT_OFFSET: usize = 0xFFFFFF8000000000;
+#[cfg(any(target_arch = "riscv64", rust_analyzer))]
+pub const KERNEL_VIRT_OFFSET: usize = 0xFFFFFFF000000000;
 
 pub trait Architecture: Sized + 'static {
-    type PerCpuData;
+    type PerCpuData: CpuData;
     type CpuFeatures: CpuFeatureSet;
 
     type BreakpointType;
@@ -65,11 +69,7 @@ pub trait Architecture: Sized + 'static {
 
     // Architectural devices
     fn local_interrupt_controller() -> Option<&'static dyn LocalInterruptController> {
-        unimplemented!()
-    }
-
-    fn message_interrupt_controller() -> &'static dyn MessageInterruptController {
-        unimplemented!()
+        None
     }
 
     #[allow(unused)]
@@ -80,4 +80,16 @@ pub trait Architecture: Sized + 'static {
     fn cpu_enabled_features<S: Scheduler>(cpu: &CpuImpl<Self, S>) -> Option<&Self::CpuFeatures> {
         None
     }
+
+    // Cache/barrier operation
+
+    fn load_barrier();
+    fn store_barrier();
+    fn memory_barrier() {
+        Self::store_barrier();
+        Self::load_barrier();
+    }
+
+    /// Flushes/invalidates a range of virtual memory from the CPU's data cache.
+    fn flush_virtual_range(range: Range<usize>);
 }

kernel/arch/interface/src/macros.rs

@@ -0,0 +1,92 @@
+/// Helper macro to implement "split" locks. This may be needed when a very specific storage
+/// layout for the locked type is required.
+// pub macro split_spinlock(
+// ) {
+#[macro_export]
+macro_rules! split_spinlock {
+    (
+        $(use $use:path;)*
+
+        $(#[$meta:meta])*
+        static $name:ident: $ty:ty = $init:expr;
+    ) => {
+        pub use $name::$name;
+
+        #[allow(non_snake_case)]
+        pub mod $name {
+            $(use $use;)*
+
+            use core::cell::UnsafeCell;
+            use core::marker::PhantomData;
+            use core::sync::atomic::{AtomicU32, Ordering};
+
+            #[repr(transparent)]
+            pub struct __Wrapper {
+                inner: UnsafeCell<$ty>
+            }
+
+            $(#[$meta])*
+            pub static $name: __Wrapper = __Wrapper {
+                inner: UnsafeCell::new($init)
+            };
+
+            static __LOCK: AtomicU32 = AtomicU32::new(0);
+            pub struct __Guard($crate::guard::IrqGuard<ArchitectureImpl>);
+            pub struct __UnsafeGuard($crate::guard::IrqGuard<ArchitectureImpl>);
+
+            impl __Wrapper {
+                #[inline(never)]
+                pub fn lock(&self) -> __Guard {
+                    let irq = $crate::guard::IrqGuard::acquire();
+
+                    while __LOCK.compare_exchange(0, 1, Ordering::Acquire, Ordering::Relaxed).is_err() {
+                        core::hint::spin_loop();
+                    }
+
+                    __Guard(irq)
+                }
+
+                #[inline(never)]
+                pub unsafe fn grab(&self) -> __UnsafeGuard {
+                    let irq = $crate::guard::IrqGuard::acquire();
+                    __UnsafeGuard(irq)
+                }
+            }
+
+            unsafe impl Sync for __Wrapper {}
+
+            impl core::ops::Deref for __Guard {
+                type Target = $ty;
+
+                fn deref(&self) -> &Self::Target {
+                    unsafe { &*$name.inner.get() }
+                }
+            }
+
+            impl core::ops::DerefMut for __Guard {
+                fn deref_mut(&mut self) -> &mut Self::Target {
+                    unsafe { &mut *$name.inner.get() }
+                }
+            }
+
+            impl core::ops::Deref for __UnsafeGuard {
+                type Target = $ty;
+
+                fn deref(&self) -> &Self::Target {
+                    unsafe { &*$name.inner.get() }
+                }
+            }
+            impl core::ops::DerefMut for __UnsafeGuard {
+                fn deref_mut(&mut self) -> &mut Self::Target {
+                    unsafe { &mut *$name.inner.get() }
+                }
+            }
+
+            impl Drop for __Guard {
+                fn drop(&mut self) {
+                    __LOCK.store(0, Ordering::Release)
+                }
+            }
+        }
+    };
+}

kernel/arch/interface/src/mem/mod.rs

@@ -2,9 +2,6 @@ use core::{fmt, marker::PhantomData, mem::size_of, ptr::NonNull};
 
 use yggdrasil_abi::error::Error;
 
-pub mod address;
-pub mod table;
-
 pub trait PhysicalMemoryAllocator {
     type Address;
 
@@ -35,6 +32,8 @@ pub struct DeviceMemoryAttributes {
 /// Describes a single device memory mapping
 #[derive(Debug)]
 pub struct RawDeviceMemoryMapping<A: KernelTableManager> {
+    /// Physical base address of the object
+    pub physical_base: u64,
     /// Virtual address of the mapped object
     pub address: usize,
     /// Base address of the mapping start
@@ -98,7 +97,8 @@ impl<A: KernelTableManager> RawDeviceMemoryMapping<A> {
         address
     }
 
-    pub fn into_raw_parts(self) -> (usize, usize, usize, usize) {
+    pub fn into_raw_parts(self) -> (u64, usize, usize, usize, usize) {
+        let physical_base = self.physical_base;
         let address = self.address;
         let base_address = self.base_address;
         let page_count = self.page_count;
@@ -106,7 +106,7 @@ impl<A: KernelTableManager> RawDeviceMemoryMapping<A> {
 
         core::mem::forget(self);
 
-        (address, base_address, page_count, page_size)
+        (physical_base, address, base_address, page_count, page_size)
     }
 
     /// # Safety
@@ -114,12 +114,14 @@ impl<A: KernelTableManager> RawDeviceMemoryMapping<A> {
     /// Preconditions: all the fields must come from a [RawDeviceMemoryMapping::into_raw_parts]
     /// call.
     pub unsafe fn from_raw_parts(
+        physical_base: u64,
         address: usize,
         base_address: usize,
         page_count: usize,
         page_size: usize,
     ) -> Self {
         Self {
+            physical_base,
             address,
             base_address,
             page_count,

kernel/arch/interface/src/sync.rs

@@ -42,7 +42,7 @@ impl<A: Architecture, T> Spinlock<A, T> {
         }
     }
 
-    pub fn lock(&self) -> SpinlockGuard<A, T> {
+    pub fn lock(&self) -> SpinlockGuard<'_, A, T> {
         // Loop until the lock can be acquired
         if LOCK_HACK.load(Ordering::Acquire) {
             return SpinlockGuard { lock: self };
@@ -103,7 +103,7 @@ impl<A: Architecture, T> IrqSafeSpinlock<A, T> {
     }
 
     /// Attempts to acquire a lock. IRQs will be disabled until the lock is released.
-    pub fn lock(&self) -> IrqSafeSpinlockGuard<A, T> {
+    pub fn lock(&self) -> IrqSafeSpinlockGuard<'_, A, T> {
         // Disable IRQs to avoid IRQ handler trying to acquire the same lock
         let irq_guard = IrqGuard::acquire();
 
@@ -154,67 +154,6 @@ impl<A: Architecture, T> DerefMut for IrqSafeSpinlockGuard<'_, A, T> {
     }
 }
 
-/// Helper macro to implement "split" locks. This may be needed when a very specific storage
-/// layout for the locked type is required.
-pub macro split_spinlock(
-    $(use $use:path;)*
-
-    $(#[$meta:meta])*
-    static $name:ident<$lock:ident: $arch:ty>: $ty:ty = $init:expr;
-) {
-    pub use $name::$name;
-
-    #[allow(non_snake_case)]
-    pub mod $name {
-        $(use $use;)*
-
-        use core::cell::UnsafeCell;
-        use core::marker::PhantomData;
-        use core::sync::atomic::{AtomicBool, Ordering};
-
-        #[repr(transparent)]
-        pub struct __Wrapper(UnsafeCell<$ty>);
-
-        $(#[$meta])*
-        pub static $name: __Wrapper = __Wrapper(UnsafeCell::new($init));
-
-        static __LOCK: AtomicBool = AtomicBool::new(false);
-        pub struct __Guard($crate::guard::IrqGuard<$arch>);
-
-        impl __Wrapper {
-            pub fn $lock(&self) -> __Guard {
-                let irq = $crate::guard::IrqGuard::acquire();
-                while __LOCK.compare_exchange(false, true, Ordering::Acquire, Ordering::Relaxed).is_err() {
-                    core::hint::spin_loop();
-                }
-                __Guard(irq)
-            }
-        }
-
-        unsafe impl Sync for __Wrapper {}
-
-        impl core::ops::Deref for __Guard {
-            type Target = $ty;
-
-            fn deref(&self) -> &Self::Target {
-                unsafe { &*$name.0.get() }
-            }
-        }
-
-        impl core::ops::DerefMut for __Guard {
-            fn deref_mut(&mut self) -> &mut Self::Target {
-                unsafe { &mut *$name.0.get() }
-            }
-        }
-
-        impl Drop for __Guard {
-            fn drop(&mut self) {
-                __LOCK.store(false, Ordering::Release)
-            }
-        }
-    }
-}
-
 static LOCK_HACK: AtomicBool = AtomicBool::new(false);
 
 /// "Hacks" all the locks in the kernel to make them function as "NULL"-locks instead of spinlocks.

kernel/arch/interface/src/task.rs

@@ -83,6 +83,7 @@ pub struct UserContextInfo {
     pub stack_pointer: usize,
     pub thread_pointer: usize,
     pub address_space: u64,
+    pub asid: u64,
     pub single_step: bool,
 }
 

kernel/arch/interface/src/util.rs

@@ -45,31 +45,80 @@ impl<T> OneTimeInit<T> {
     }
 
     pub fn try_init_with_opt<F: FnOnce() -> Result<T, Error>>(&self, f: F) -> Result<&T, Error> {
-        if self
-            .state
-            .compare_exchange(
-                Self::STATE_UNINITIALIZED,
-                Self::STATE_INITIALIZING,
-                Ordering::Release,
-                Ordering::Relaxed,
-            )
-            .is_err()
-        {
+        if !self.try_begin_init() {
             // Already initialized
             return Err(Error::AlreadyExists);
         }
 
-        let value = match f() {
-            Ok(val) => val,
-            Err(err) => {
-                self.state
-                    .store(Self::STATE_UNINITIALIZED, Ordering::Release);
-                return Err(err);
+        match f() {
+            Ok(value) => {
+                let value = unsafe { (*self.value.get()).write(value) };
+                self.finish_init();
+                Ok(value)
             }
-        };
+            Err(error) => {
+                self.fail_init();
+                Err(error)
+            }
+        }
+    }
 
-        let value = unsafe { (*self.value.get()).write(value) };
+    pub fn or_init_with<F: FnOnce() -> T>(&self, f: F) -> &T {
+        if !self.try_begin_init() {
+            return self.wait_for_init();
+        }
+        let value = unsafe { (*self.value.get()).write(f()) };
+        self.finish_init();
+        value
+    }
 
+    pub fn or_init_with_opt<F: FnOnce() -> Option<T>>(&self, f: F) -> Option<&T> {
+        if !self.try_begin_init() {
+            return Some(self.wait_for_init());
+        }
+        match f() {
+            Some(value) => {
+                let value = unsafe { (*self.value.get()).write(value) };
+                self.finish_init();
+                Some(value)
+            }
+            None => {
+                self.fail_init();
+                None
+            }
+        }
+    }
+
+    pub fn or_try_init_with<F: FnOnce() -> Result<T, Error>>(&self, f: F) -> Result<&T, Error> {
+        if !self.try_begin_init() {
+            return Ok(self.wait_for_init());
+        }
+        match f() {
+            Ok(value) => {
+                let value = unsafe { (*self.value.get()).write(value) };
+                self.finish_init();
+                Ok(value)
+            }
+            Err(error) => {
+                // Init failed
+                self.fail_init();
+                Err(error)
+            }
+        }
+    }
+
+    fn try_begin_init(&self) -> bool {
+        self.state
+            .compare_exchange(
+                Self::STATE_UNINITIALIZED,
+                Self::STATE_INITIALIZING,
+                Ordering::Acquire,
+                Ordering::Relaxed,
+            )
+            .is_ok()
+    }
+
+    fn finish_init(&self) {
         self.state
             .compare_exchange(
                 Self::STATE_INITIALIZING,
@@ -78,37 +127,38 @@ impl<T> OneTimeInit<T> {
                 Ordering::Relaxed,
             )
             .unwrap();
-
-        Ok(value)
     }
 
-    pub fn try_init_with<F: FnOnce() -> T>(&self, f: F) -> Option<&T> {
-        if self
-            .state
+    fn fail_init(&self) {
+        self.state
             .compare_exchange(
-                Self::STATE_UNINITIALIZED,
                 Self::STATE_INITIALIZING,
+                Self::STATE_UNINITIALIZED,
                 Ordering::Release,
                 Ordering::Relaxed,
             )
-            .is_err()
-        {
+            .unwrap();
+    }
+
+    fn wait_for_init(&self) -> &T {
+        while self.state.load(Ordering::Acquire) != Self::STATE_INITIALIZED {
+            core::hint::spin_loop();
+        }
+
+        unsafe { (*self.value.get()).assume_init_ref() }
+    }
+
+    pub fn try_init_with<F: FnOnce() -> T>(&self, f: F) -> Result<&T, Error> {
+        if !self.try_begin_init() {
             // Already initialized
-            return None;
+            return Err(Error::AlreadyExists);
         }
 
         let value = unsafe { (*self.value.get()).write(f()) };
 
-        self.state
-            .compare_exchange(
-                Self::STATE_INITIALIZING,
-                Self::STATE_INITIALIZED,
-                Ordering::Release,
-                Ordering::Relaxed,
-            )
-            .unwrap();
+        self.finish_init();
 
-        Some(value)
+        Ok(value)
     }
 
     /// Sets the underlying value of the [OneTimeInit]. If already initialized, panics.

kernel/arch/riscv64/Cargo.toml

@@ -1,16 +1,19 @@
 [package]
-name = "kernel-arch-i686"
+name = "kernel-arch-riscv64"
 version = "0.1.0"
-edition = "2021"
+edition = "2024"
 
 [dependencies]
 yggdrasil-abi.workspace = true
 kernel-arch-interface.workspace = true
 libk-mm-interface.workspace = true
 device-api = { workspace = true, features = ["derive"] }
-kernel-arch-x86.workspace = true
 
+tock-registers.workspace = true
 bitflags.workspace = true
 static_assertions.workspace = true
-tock-registers.workspace = true
 log.workspace = true
+cfg-if.workspace = true
+
+[lints]
+workspace = true

kernel/arch/riscv64/src/context.S

@@ -0,0 +1,128 @@
+// vi:ft=asm:
+
+.section .text
+
+.macro SAVE_TASK_STATE
+    addi sp, sp, -{context_size}
+
+    sd ra,   0 * 8(sp)
+    sd gp,   1 * 8(sp)
+    sd s11,  2 * 8(sp)
+    sd s10,  3 * 8(sp)
+    sd s9,   4 * 8(sp)
+    sd s8,   5 * 8(sp)
+    sd s7,   6 * 8(sp)
+    sd s6,   7 * 8(sp)
+    sd s5,   8 * 8(sp)
+    sd s4,   9 * 8(sp)
+    sd s3,  10 * 8(sp)
+    sd s2,  11 * 8(sp)
+    sd s1,  12 * 8(sp)
+    sd s0,  13 * 8(sp)
+.endm
+
+.macro LOAD_TASK_STATE
+    ld ra,   0 * 8(sp)
+    ld gp,   1 * 8(sp)
+    ld s11,  2 * 8(sp)
+    ld s10,  3 * 8(sp)
+    ld s9,   4 * 8(sp)
+    ld s8,   5 * 8(sp)
+    ld s7,   6 * 8(sp)
+    ld s6,   7 * 8(sp)
+    ld s5,   8 * 8(sp)
+    ld s4,   9 * 8(sp)
+    ld s3,  10 * 8(sp)
+    ld s2,  11 * 8(sp)
+    ld s1,  12 * 8(sp)
+    ld s0,  13 * 8(sp)
+
+    addi sp, sp, {context_size}
+.endm
+
+.option push
+.option norvc
+
+.global __rv64_task_enter_kernel
+.global __rv64_task_enter_user
+.global __rv64_switch_task
+.global __rv64_switch_task_and_drop
+.global __rv64_enter_task
+
+// Context switching
+.type __rv64_enter_task, @function
+__rv64_enter_task:
+    // a0 - task ctx
+    ld sp, (a0)
+    LOAD_TASK_STATE
+    ret
+.size __rv64_enter_task, . - __rv64_enter_task
+
+.type __rv64_switch_task, @function
+__rv64_switch_task:
+    // a0 - destination task ctx
+    // a1 - source task ctx
+    SAVE_TASK_STATE
+    sd sp, (a1)
+    ld sp, (a0)
+    LOAD_TASK_STATE
+    ret
+.size __rv64_switch_task, . - __rv64_switch_task
+
+.type __rv64_switch_task_and_drop, @function
+__rv64_switch_task_and_drop:
+    // a0 - destination task ctx
+    // a1 - thread struct to drop
+    ld sp, (a0)
+
+    mv a0, a1
+    call __arch_drop_thread
+
+    LOAD_TASK_STATE
+    ret
+.size __rv64_switch_task_and_drop, . - __rv64_switch_task_and_drop
+
+// Entry functions
+.type __rv64_task_enter_kernel, @function
+__rv64_task_enter_kernel:
+    ld a0, (sp)     // argument
+    ld ra, 8(sp)    // entry
+    addi sp, sp, 16
+
+    // Set SPIE to enable interrupts
+    // Set SPP = 1 to indicate a return to S-mode
+    csrr t0, sstatus
+    ori t0, t0, (1 << 5)
+    ori t0, t0, (1 << 8)
+
+    csrw sstatus, t0
+    csrw sepc, ra
+    csrw sscratch, zero
+
+    sret
+.size __rv64_task_enter_kernel, . - __rv64_task_enter_kernel
+
+.type __rv64_task_enter_user, @function
+__rv64_task_enter_user:
+    csrw sscratch, tp
+
+    ld a0, 0 * 8(sp)    // argument
+    ld ra, 1 * 8(sp)    // entry
+    ld tp, 2 * 8(sp)    // thread pointer
+    ld sp, 3 * 8(sp)    // user stack
+
+    // Set SPIE to enable interrupts
+    // Set SPP = 0 to indicate a return to U-mode
+    li t1, (1 << 8)
+    not t1, t1
+
+    csrr t0, sstatus
+    ori t0, t0, (1 << 5)
+    and t0, t0, t1
+    csrw sstatus, t0
+    csrw sepc, ra
+
+    sret
+.size __rv64_task_enter_user, . - __rv64_task_enter_user
+
+.option pop

kernel/arch/riscv64/src/context.rs

@@ -0,0 +1,221 @@
+use core::{arch::global_asm, cell::UnsafeCell, marker::PhantomData};
+
+use kernel_arch_interface::{
+    mem::{KernelTableManager, PhysicalMemoryAllocator},
+    task::{StackBuilder, TaskContext, UserContextInfo},
+    Architecture,
+};
+use libk_mm_interface::address::PhysicalAddress;
+use tock_registers::{
+    interfaces::{Readable, Writeable},
+    registers::InMemoryRegister,
+};
+use yggdrasil_abi::error::Error;
+
+use crate::{
+    mem::{self},
+    registers::SATP,
+    ArchitectureImpl, PerCpuData,
+};
+
+pub const CONTEXT_SIZE: usize = 14 * size_of::<usize>();
+
+#[repr(C, align(0x10))]
+struct TaskContextInner {
+    // 0x00
+    sp: usize,
+
+    satp: InMemoryRegister<u64, SATP::Register>,
+}
+
+pub struct TaskContextImpl<
+    K: KernelTableManager,
+    PA: PhysicalMemoryAllocator<Address = PhysicalAddress>,
+> {
+    inner: UnsafeCell<TaskContextInner>,
+    // fp_context: UnsafeCell<FpContext>,
+    stack_base_phys: PhysicalAddress,
+    stack_top: usize,
+    stack_size: usize,
+
+    _pd: PhantomData<(K, PA)>,
+}
+
+impl<K: KernelTableManager, PA: PhysicalMemoryAllocator<Address = PhysicalAddress>>
+    TaskContextImpl<K, PA>
+{
+    unsafe fn load_state(&self) {
+        // TODO load new SATP value
+        let inner = unsafe { &*self.inner.get() };
+        let cpu = unsafe { &mut *ArchitectureImpl::local_cpu().cast::<PerCpuData>() };
+
+        // Copy new SATP
+        let satp = inner.satp.get();
+        let asid = inner.satp.read(SATP::ASID);
+        if satp != SATP.get() {
+            mem::tlb_flush_asid(asid as usize);
+            SATP.set(satp);
+        }
+        cpu.smode_sp = self.stack_top;
+    }
+
+    unsafe fn store_state(&self) {}
+}
+
+impl<K: KernelTableManager, PA: PhysicalMemoryAllocator<Address = PhysicalAddress>>
+    TaskContext<K, PA> for TaskContextImpl<K, PA>
+{
+    const USER_STACK_EXTRA_ALIGN: usize = 8;
+    const SIGNAL_STACK_EXTRA_ALIGN: usize = 0;
+
+    fn user(context: UserContextInfo) -> Result<Self, Error> {
+        const USER_TASK_PAGES: usize = 16;
+        let stack_base_phys = PA::allocate_contiguous_pages(USER_TASK_PAGES)?;
+        let stack_base = stack_base_phys.raw_virtualize::<K>();
+
+        let mut stack = StackBuilder::new(stack_base, USER_TASK_PAGES * 0x1000);
+
+        log::debug!(
+            "Set up user task: pc={:#x}, sp={:#x}, tp={:#x}",
+            context.entry,
+            context.stack_pointer,
+            context.thread_pointer
+        );
+        stack.push(context.stack_pointer);
+        stack.push(context.thread_pointer);
+        stack.push(context.entry);
+        stack.push(context.argument);
+
+        setup_common_context(&mut stack, __rv64_task_enter_user as _);
+
+        let sp = stack.build();
+        let satp = InMemoryRegister::new(0);
+        satp.write(
+            SATP::MODE::Sv39
+                + SATP::ASID.val(context.asid)
+                + SATP::PPN.val(context.address_space >> 12),
+        );
+
+        Ok(Self {
+            inner: UnsafeCell::new(TaskContextInner { sp, satp }),
+            // fp_context: UnsafeCell::new(FpContext::new()),
+            stack_base_phys,
+            stack_top: stack_base + USER_TASK_PAGES * 0x1000,
+            stack_size: USER_TASK_PAGES * 0x1000,
+
+            _pd: PhantomData,
+        })
+    }
+
+    fn kernel(entry: extern "C" fn(usize) -> !, arg: usize) -> Result<Self, Error> {
+        const KERNEL_TASK_PAGES: usize = 8;
+        let stack_base_phys = PA::allocate_contiguous_pages(KERNEL_TASK_PAGES)?;
+        let stack_base = stack_base_phys.raw_virtualize::<K>();
+
+        let mut stack = StackBuilder::new(stack_base, KERNEL_TASK_PAGES * 0x1000);
+
+        // Entry and argument
+        stack.push(entry as _);
+        stack.push(arg);
+
+        setup_common_context(&mut stack, __rv64_task_enter_kernel as _);
+
+        let sp = stack.build();
+
+        // TODO stack is leaked
+        let satp = InMemoryRegister::new(0);
+        let kernel_table_phys = mem::fixed::table_physical_address().into_u64();
+        satp.write(SATP::MODE::Sv39 + SATP::ASID.val(0) + SATP::PPN.val(kernel_table_phys >> 12));
+
+        Ok(Self {
+            inner: UnsafeCell::new(TaskContextInner { sp, satp }),
+            // fp_context: UnsafeCell::new(FpContext::new()),
+            stack_base_phys,
+            stack_top: 0,
+            stack_size: KERNEL_TASK_PAGES * 0x1000,
+
+            _pd: PhantomData,
+        })
+    }
+
+    fn set_thread_pointer(&self, tp: usize) {
+        let _ = tp;
+        todo!()
+    }
+
+    fn align_stack_for_entry(sp: usize) -> usize {
+        sp
+    }
+
+    unsafe fn enter(&self) -> ! {
+        unsafe {
+            self.load_state();
+            __rv64_enter_task(self.inner.get())
+        }
+    }
+
+    unsafe fn switch(&self, from: &Self) {
+        if core::ptr::addr_eq(self, from) {
+            return;
+        }
+
+        unsafe {
+            from.store_state();
+            self.load_state();
+            __rv64_switch_task(self.inner.get(), from.inner.get())
+        }
+    }
+
+    unsafe fn switch_and_drop(&self, thread: *const ()) {
+        unsafe {
+            self.load_state();
+            __rv64_switch_task_and_drop(self.inner.get(), thread)
+        }
+    }
+}
+
+impl<K: KernelTableManager, PA: PhysicalMemoryAllocator<Address = PhysicalAddress>> Drop
+    for TaskContextImpl<K, PA>
+{
+    fn drop(&mut self) {
+        assert_eq!(self.stack_size % 0x1000, 0);
+
+        for offset in (0..self.stack_size).step_by(0x1000) {
+            unsafe {
+                PA::free_page(self.stack_base_phys.add(offset));
+            }
+        }
+    }
+}
+
+fn setup_common_context(builder: &mut StackBuilder, entry: usize) {
+    builder.push(0); // x8/s0/fp
+    builder.push(0); // x9/s1
+    builder.push(0); // x18/s2
+    builder.push(0); // x19/s3
+    builder.push(0); // x20/s4
+    builder.push(0); // x21/s5
+    builder.push(0); // x22/s6
+    builder.push(0); // x23/s7
+    builder.push(0); // x24/s8
+    builder.push(0); // x25/s9
+    builder.push(0); // x26/s10
+    builder.push(0); // x27/s11
+    builder.push(0); // x4/gp
+    builder.push(entry); // x1/ra return address
+}
+
+unsafe extern "C" {
+    fn __rv64_enter_task(to: *mut TaskContextInner) -> !;
+    fn __rv64_switch_task(to: *mut TaskContextInner, from: *mut TaskContextInner);
+    fn __rv64_switch_task_and_drop(to: *mut TaskContextInner, thread: *const ()) -> !;
+    fn __rv64_task_enter_kernel();
+    fn __rv64_task_enter_user();
+    // fn __rv64_fp_store_context(to: *mut c_void);
+    // fn __rv64_fp_restore_context(from: *const c_void);
+}
+
+global_asm!(
+    include_str!("context.S"),
+    context_size = const CONTEXT_SIZE,
+);

kernel/arch/riscv64/src/intrinsics.rs

@@ -0,0 +1,6 @@
+#[inline]
+pub fn rdtime() -> u64 {
+    let mut output: u64;
+    unsafe { core::arch::asm!("rdtime {0}", out(reg) output) };
+    output
+}

kernel/arch/riscv64/src/lib.rs

@@ -0,0 +1,190 @@
+#![feature(decl_macro)]
+#![no_std]
+
+extern crate alloc;
+
+use core::{
+    ops::Range,
+    sync::atomic::{AtomicUsize, Ordering},
+};
+
+use alloc::{boxed::Box, collections::btree_map::BTreeMap, vec::Vec};
+use device_api::interrupt::LocalInterruptController;
+use kernel_arch_interface::{
+    cpu::{CpuData, CpuImpl, IpiQueue},
+    sync::IrqSafeSpinlock,
+    task::Scheduler,
+    util::OneTimeInit,
+    Architecture,
+};
+use tock_registers::interfaces::{ReadWriteable, Readable};
+
+use registers::SSTATUS;
+
+pub mod mem;
+pub use mem::{process::ProcessAddressSpaceImpl, KernelTableManagerImpl};
+pub mod context;
+pub use context::TaskContextImpl;
+pub mod intrinsics;
+pub mod registers;
+pub mod sbi;
+
+pub struct ArchitectureImpl;
+
+#[repr(C)]
+pub struct PerCpuData {
+    // Used in assembly
+    pub tmp_t0: usize,   // 0x00
+    pub umode_sp: usize, // 0x08
+    pub smode_sp: usize, // 0x10
+
+    // Used elsewhere
+    pub bootstrap: bool,
+    pub queue_index: usize,
+}
+
+pub static CPU_COUNT: AtomicUsize = AtomicUsize::new(1);
+pub static mut BOOT_HART_ID: u64 = 0;
+static IPI_QUEUES: OneTimeInit<Vec<IpiQueue<ArchitectureImpl>>> = OneTimeInit::new();
+static HART_TO_QUEUE: IrqSafeSpinlock<ArchitectureImpl, BTreeMap<u32, usize>> =
+    IrqSafeSpinlock::new(BTreeMap::new());
+
+impl CpuData for PerCpuData {
+    fn is_bootstrap(&self, id: u32) -> bool {
+        let _ = id;
+        self.bootstrap
+    }
+
+    fn queue_index(&self, id: u32) -> usize {
+        let _ = id;
+        self.queue_index
+    }
+}
+
+/// Returns the ID of the bootstrap HART
+pub fn boot_hart_id() -> u64 {
+    unsafe { BOOT_HART_ID }
+}
+
+#[unsafe(naked)]
+extern "C" fn idle_task(_: usize) -> ! {
+    core::arch::naked_asm!("1: nop; j 1b");
+}
+
+impl ArchitectureImpl {
+    pub fn for_each_hart<F: FnMut(u32, usize, &IpiQueue<ArchitectureImpl>)>(mut f: F) {
+        let map = HART_TO_QUEUE.lock();
+        map.iter().for_each(|(&hart_id, &queue_index)| {
+            let queue = &IPI_QUEUES.get()[queue_index];
+            f(hart_id, queue_index, queue);
+        });
+    }
+}
+
+impl Architecture for ArchitectureImpl {
+    type PerCpuData = PerCpuData;
+    type CpuFeatures = ();
+    type BreakpointType = u32;
+
+    const BREAKPOINT_VALUE: Self::BreakpointType = 0;
+
+    fn halt() -> ! {
+        loop {
+            unsafe { Self::set_interrupt_mask(true) };
+            Self::wait_for_interrupt();
+        }
+    }
+
+    unsafe fn set_local_cpu(cpu: *mut ()) {
+        unsafe { core::arch::asm!("mv tp, {0}", in(reg) cpu) };
+    }
+
+    #[inline]
+    fn local_cpu() -> *mut () {
+        let value: u64;
+        unsafe { core::arch::asm!("mv {0}, tp", out(reg) value) };
+        value as _
+    }
+
+    unsafe fn init_local_cpu<S: Scheduler + 'static>(id: Option<u32>, data: Self::PerCpuData) {
+        let id = id.expect("riscv64 requires an explicit HART ID in its per-processor struct");
+        let queue_index = data.queue_index;
+        HART_TO_QUEUE.lock().insert(id, queue_index);
+        let cpu = Box::leak(Box::new(CpuImpl::<Self, S>::new(id, data)));
+        unsafe { cpu.set_local() };
+    }
+
+    unsafe fn init_ipi_queues(queues: Vec<IpiQueue<Self>>) {
+        IPI_QUEUES.init(queues);
+    }
+
+    fn ipi_queue(cpu_id: u32) -> Option<&'static IpiQueue<Self>> {
+        let queue_index = *HART_TO_QUEUE.lock().get(&cpu_id)?;
+        IPI_QUEUES.try_get().and_then(|q| q.get(queue_index))
+    }
+
+    #[inline]
+    unsafe fn set_interrupt_mask(mask: bool) -> bool {
+        let old = Self::interrupt_mask();
+        if mask {
+            SSTATUS.modify(SSTATUS::SIE::CLEAR);
+        } else {
+            SSTATUS.modify(SSTATUS::SIE::SET);
+        }
+        old
+    }
+
+    #[inline]
+    fn interrupt_mask() -> bool {
+        SSTATUS.matches_all(SSTATUS::SIE::CLEAR)
+    }
+
+    fn wait_for_interrupt() {
+        unsafe {
+            core::arch::asm!("wfi");
+        }
+    }
+
+    fn cpu_count() -> usize {
+        CPU_COUNT.load(Ordering::Acquire)
+    }
+
+    fn cpu_index<S: Scheduler + 'static>() -> u32 {
+        CpuImpl::<Self, S>::local().id()
+    }
+
+    fn cpu_enabled_features<S: Scheduler>(cpu: &CpuImpl<Self, S>) -> Option<&Self::CpuFeatures> {
+        let _ = cpu;
+        todo!()
+    }
+
+    fn cpu_available_features<S: Scheduler>(cpu: &CpuImpl<Self, S>) -> Option<&Self::CpuFeatures> {
+        let _ = cpu;
+        todo!()
+    }
+
+    fn local_interrupt_controller() -> Option<&'static dyn LocalInterruptController> {
+        todo!()
+    }
+
+    fn idle_task() -> extern "C" fn(usize) -> ! {
+        idle_task
+    }
+
+    // Cache/barrier operation
+    fn load_barrier() {
+        unsafe { core::arch::asm!("fence r, w") };
+    }
+
+    fn store_barrier() {
+        unsafe { core::arch::asm!("fence w, r") };
+    }
+
+    fn memory_barrier() {
+        unsafe { core::arch::asm!("fence rw, rw") };
+    }
+
+    fn flush_virtual_range(_range: Range<usize>) {
+        // TODO
+    }
+}

kernel/arch/riscv64/src/mem/fixed.rs

@@ -0,0 +1,32 @@
+use kernel_arch_interface::sync::IrqSafeSpinlock;
+use libk_mm_interface::{address::PhysicalAddress, table::EntryLevel};
+
+use crate::{
+    mem::{
+        auto_lower_address,
+        table::{PageEntry, PageTable, L1},
+        KERNEL_VIRT_OFFSET,
+    },
+    ArchitectureImpl,
+};
+
+pub const IDENTITY_SIZE_L1: usize = 64;
+
+pub(super) static mut KERNEL_L1: PageTable<L1> = const {
+    let mut table = PageTable::zeroed();
+
+    let mut index = 0;
+    while index < IDENTITY_SIZE_L1 {
+        let entry = PageEntry::identity_block(PhysicalAddress::from_usize(index << L1::SHIFT));
+        table.entries[index] = entry;
+        table.entries[index + ((KERNEL_VIRT_OFFSET >> L1::SHIFT) & 0x1FF)] = entry;
+        index += 1;
+    }
+
+    table
+};
+pub(super) static LOCK: IrqSafeSpinlock<ArchitectureImpl, ()> = IrqSafeSpinlock::new(());
+
+pub fn table_physical_address() -> PhysicalAddress {
+    PhysicalAddress::from_usize(auto_lower_address(&raw const KERNEL_L1))
+}

kernel/arch/riscv64/src/mem/intrinsics.rs

@@ -0,0 +1,51 @@
+use libk_mm_interface::table::{EntryLevel, EntryLevelExt};
+
+use crate::mem::table::L3;
+
+pub fn tlb_flush_global_full() {
+    tlb_flush_full();
+    // TODO send TLB shootdown IPI to other harts
+}
+
+pub fn tlb_flush_global_va(va: usize) {
+    tlb_flush_va(va);
+    // TODO send TLB shootdown IPI to other harts
+}
+
+pub fn tlb_flush_range_va(start: usize, size: usize) {
+    let end = (start + size).page_align_up::<L3>();
+    let start = start.page_align_down::<L3>();
+
+    for page in (start..end).step_by(L3::SIZE) {
+        tlb_flush_va(page);
+    }
+}
+
+pub fn tlb_flush_range_va_asid(asid: usize, start: usize, size: usize) {
+    let end = (start + size).page_align_up::<L3>();
+    let start = start.page_align_down::<L3>();
+
+    for page in (start..end).step_by(L3::SIZE) {
+        tlb_flush_va_asid(page, asid);
+    }
+}
+
+#[inline]
+pub fn tlb_flush_full() {
+    unsafe { core::arch::asm!("sfence.vma") };
+}
+
+#[inline]
+pub fn tlb_flush_va(va: usize) {
+    unsafe { core::arch::asm!("sfence.vma {0}, zero", in(reg) va) };
+}
+
+#[inline]
+pub fn tlb_flush_asid(asid: usize) {
+    unsafe { core::arch::asm!("sfence.vma zero, {0}", in(reg) asid) };
+}
+
+#[inline]
+pub fn tlb_flush_va_asid(va: usize, asid: usize) {
+    unsafe { core::arch::asm!("sfence.vma {0}, {1}", in(reg) va, in(reg) asid) };
+}

kernel/arch/riscv64/src/mem/mod.rs

@@ -0,0 +1,105 @@
+use kernel_arch_interface::mem::{
+    DeviceMemoryAttributes, KernelTableManager, RawDeviceMemoryMapping,
+};
+use libk_mm_interface::{
+    address::PhysicalAddress,
+    table::{page_index, EntryLevel, EntryLevelExt},
+};
+use tock_registers::interfaces::Writeable;
+use yggdrasil_abi::error::Error;
+
+use crate::{
+    mem::table::{PageTable, L1, L3},
+    registers::SATP,
+};
+
+pub use intrinsics::*;
+
+pub mod fixed;
+pub mod intrinsics;
+pub mod process;
+pub mod table;
+
+pub const KERNEL_VIRT_OFFSET: usize = kernel_arch_interface::KERNEL_VIRT_OFFSET;
+
+/// Any VAs above this one are sign-extended
+pub const USER_BOUNDARY: usize = 0x40_00000000;
+
+#[derive(Debug)]
+pub struct KernelTableManagerImpl;
+
+impl KernelTableManager for KernelTableManagerImpl {
+    fn virtualize(address: u64) -> usize {
+        let address = address as usize;
+        if address < fixed::IDENTITY_SIZE_L1 * L1::SIZE {
+            address + KERNEL_VIRT_OFFSET
+        } else {
+            panic!("Invalid physical address: {address:#x}")
+        }
+    }
+
+    fn physicalize(address: usize) -> u64 {
+        if address < KERNEL_VIRT_OFFSET
+            || address - KERNEL_VIRT_OFFSET >= fixed::IDENTITY_SIZE_L1 * L1::SIZE
+        {
+            panic!("Invalid virtualized address: {address:#x}");
+        }
+        (address - KERNEL_VIRT_OFFSET) as u64
+    }
+
+    unsafe fn map_device_pages(
+        base: u64,
+        count: usize,
+        attrs: DeviceMemoryAttributes,
+    ) -> Result<RawDeviceMemoryMapping<Self>, Error> {
+        let _ = attrs;
+        let _lock = fixed::LOCK.lock();
+        let base = PhysicalAddress::from_u64(base);
+        let l3_aligned_base = base.page_align_down::<L3>();
+        let l3_aligned_end = base.add(count).page_align_up::<L3>();
+        let l3_offset = base - l3_aligned_base;
+        let l3_page_count = (l3_aligned_end - l3_aligned_base).page_count::<L3>();
+        let l3_aligned_virt = l3_aligned_base.add(KERNEL_VIRT_OFFSET).into_usize();
+
+        Ok(unsafe {
+            RawDeviceMemoryMapping::from_raw_parts(
+                l3_aligned_base.into_u64(),
+                l3_aligned_virt + l3_offset,
+                l3_aligned_virt,
+                l3_page_count,
+                L3::SIZE,
+            )
+        })
+    }
+
+    unsafe fn unmap_device_pages(mapping: &RawDeviceMemoryMapping<Self>) {
+        let _ = mapping;
+    }
+}
+
+pub fn auto_lower_address<T>(x: *const T) -> usize {
+    let x = x.addr();
+    if x >= KERNEL_VIRT_OFFSET {
+        x - KERNEL_VIRT_OFFSET
+    } else {
+        x
+    }
+}
+
+/// Enables the memory translation.
+///
+/// # Safety
+///
+/// Only meant to be called once per each HART during their early init.
+pub unsafe fn enable_mmu() {
+    let l1_phys = auto_lower_address(&raw const fixed::KERNEL_L1) as u64;
+    tlb_flush_full();
+    SATP.write(SATP::PPN.val(l1_phys >> 12) + SATP::MODE::Sv39);
+}
+
+pub fn clone_kernel_tables(dst: &mut PageTable<L1>) {
+    let _lock = fixed::LOCK.lock();
+    for l1i in page_index::<L1>(USER_BOUNDARY)..512 {
+        dst[l1i] = unsafe { fixed::KERNEL_L1[l1i] };
+    }
+}

kernel/arch/riscv64/src/mem/process.rs

@@ -0,0 +1,238 @@
+use core::{
+    marker::PhantomData,
+    sync::atomic::{AtomicU16, Ordering},
+};
+
+use libk_mm_interface::{
+    address::{AsPhysicalAddress, PhysicalAddress},
+    pointer::PhysicalRefMut,
+    process::{PageAttributeUpdate, ProcessAddressSpaceManager},
+    table::{
+        EntryLevel, EntryLevelDrop, EntryLevelExt, MapAttributes, NextPageTable, TableAllocator,
+    },
+};
+use yggdrasil_abi::error::Error;
+
+use crate::mem::{
+    clone_kernel_tables,
+    table::{PageAttributes, PageEntry},
+};
+
+use super::{
+    table::{DroppableRange, PageTable, L1, L2, L3},
+    KernelTableManagerImpl, USER_BOUNDARY,
+};
+
+pub struct ProcessAddressSpaceImpl<TA: TableAllocator> {
+    l1: PhysicalRefMut<'static, PageTable<L1>, KernelTableManagerImpl>,
+    asid: u16,
+    _pd: PhantomData<TA>,
+}
+
+impl<TA: TableAllocator> ProcessAddressSpaceManager<TA> for ProcessAddressSpaceImpl<TA> {
+    const LOWER_LIMIT_PFN: usize = 8;
+    const UPPER_LIMIT_PFN: usize = (16 << 30) / L3::SIZE;
+
+    fn new() -> Result<Self, Error> {
+        static LAST_ASID: AtomicU16 = AtomicU16::new(1);
+
+        let mut l1 = unsafe {
+            PhysicalRefMut::<'static, PageTable<L1>, KernelTableManagerImpl>::map(
+                TA::allocate_page_table()?,
+            )
+        };
+
+        for i in 0..512 {
+            l1[i] = PageEntry::INVALID;
+        }
+        // Copy the kernel mappings
+        clone_kernel_tables(&mut l1);
+
+        let asid = LAST_ASID.fetch_add(1, Ordering::AcqRel);
+
+        Ok(Self {
+            l1,
+            asid,
+            _pd: PhantomData,
+        })
+    }
+
+    unsafe fn map_page(
+        &mut self,
+        address: usize,
+        physical: PhysicalAddress,
+        flags: MapAttributes,
+    ) -> Result<(), Error> {
+        self.write_l3_entry(
+            address,
+            PageEntry::page(physical, to_page_attributes(flags)),
+            false,
+        )
+        .unwrap();
+        Ok(())
+    }
+
+    unsafe fn update_page_attributes(
+        &mut self,
+        address: usize,
+        update: &PageAttributeUpdate,
+    ) -> Result<(), Error> {
+        self.update_l3_entry(address, |entry| entry.update(update))
+    }
+
+    unsafe fn unmap_page(&mut self, address: usize) -> Result<(PhysicalAddress, bool), Error> {
+        self.pop_l3_entry(address)
+    }
+
+    fn translate(&self, address: usize) -> Result<(PhysicalAddress, MapAttributes), Error> {
+        self.read_l3_entry(address).ok_or(Error::DoesNotExist)
+    }
+
+    fn as_address_with_asid(&self) -> (u64, u64) {
+        let physical = unsafe { self.l1.as_physical_address() }.into_u64();
+        (physical, self.asid as u64)
+    }
+
+    unsafe fn clear(&mut self) {
+        unsafe { self.l1.drop_range::<TA>(L1::DROPPABLE_RANGE) };
+    }
+}
+
+impl<TA: TableAllocator> ProcessAddressSpaceImpl<TA> {
+    // Write a single 4KiB entry
+    fn write_l3_entry(
+        &mut self,
+        virt: usize,
+        entry: PageEntry<L3>,
+        overwrite: bool,
+    ) -> Result<(), Error> {
+        if virt >= USER_BOUNDARY {
+            log::warn!("Tried to map a userspace page to a non-userspace virtual region");
+            return Err(Error::InvalidArgument);
+        }
+
+        let l1i = virt.page_index::<L1>();
+        let l2i = virt.page_index::<L2>();
+        let l3i = virt.page_index::<L3>();
+
+        let mut l2 = self.l1.get_mut_or_alloc::<TA>(l1i)?;
+        let mut l3 = l2.get_mut_or_alloc::<TA>(l2i)?;
+
+        if l3[l3i].is_present() && !overwrite {
+            todo!();
+        }
+
+        l3[l3i] = entry;
+        super::tlb_flush_va_asid(virt, self.asid as usize);
+
+        Ok(())
+    }
+
+    fn update_l3_entry<F: FnOnce(&mut PageEntry<L3>) -> Result<(), Error>>(
+        &mut self,
+        virt: usize,
+        mapper: F,
+    ) -> Result<(), Error> {
+        let l1i = virt.page_index::<L1>();
+        let l2i = virt.page_index::<L2>();
+        let l3i = virt.page_index::<L3>();
+
+        // TODO somehow drop tables if they're known to be empty?
+        let mut l2 = self.l1.get_mut(l1i).ok_or(Error::DoesNotExist)?;
+        let mut l3 = l2.get_mut(l2i).ok_or(Error::DoesNotExist)?;
+
+        let entry = &mut l3[l3i];
+        if !entry.is_present() {
+            return Err(Error::DoesNotExist);
+        }
+        mapper(entry)?;
+        super::tlb_flush_va_asid(virt, self.asid as usize);
+
+        Ok(())
+    }
+
+    fn pop_l3_entry(&mut self, virt: usize) -> Result<(PhysicalAddress, bool), Error> {
+        let l1i = virt.page_index::<L1>();
+        let l2i = virt.page_index::<L2>();
+        let l3i = virt.page_index::<L3>();
+
+        // TODO somehow drop tables if they're known to be empty?
+        let mut l2 = self.l1.get_mut(l1i).ok_or(Error::DoesNotExist)?;
+        let mut l3 = l2.get_mut(l2i).ok_or(Error::DoesNotExist)?;
+
+        let entry = l3[l3i];
+        let page = entry.as_page().ok_or(Error::DoesNotExist)?;
+        let dirty = entry.is_dirty();
+
+        l3[l3i] = PageEntry::INVALID;
+        super::tlb_flush_va_asid(virt, self.asid as usize);
+
+        Ok((page, dirty))
+    }
+
+    fn read_l3_entry(&self, virt: usize) -> Option<(PhysicalAddress, MapAttributes)> {
+        if virt >= USER_BOUNDARY {
+            log::warn!("Tried read an userspace page to a non-userspace virtual region");
+            return None;
+        }
+
+        let l1i = virt.page_index::<L1>();
+        let l2i = virt.page_index::<L2>();
+        let l3i = virt.page_index::<L3>();
+
+        let l2 = self.l1.get(l1i)?;
+        let l3 = l2.get(l2i)?;
+
+        let page = l3[l3i].as_page()?;
+
+        Some((
+            page.add(virt & 0xFFF),
+            to_map_attributes(l3[l3i].attributes()),
+        ))
+    }
+}
+
+impl<TA: TableAllocator> Drop for ProcessAddressSpaceImpl<TA> {
+    fn drop(&mut self) {
+        // SAFETY: with safe usage of the ProcessAddressSpaceImpl, clearing and dropping
+        // is safe, no one refers to the memory
+        unsafe {
+            self.clear();
+            let l1_phys = self.l1.as_physical_address();
+            TA::free_page_table(l1_phys);
+            super::tlb_flush_asid(self.asid as usize);
+        }
+    }
+}
+
+fn to_page_attributes(src: MapAttributes) -> PageAttributes {
+    let mut result = PageAttributes::R | PageAttributes::X;
+    if src.contains(MapAttributes::USER_WRITE) {
+        result |= PageAttributes::W;
+    }
+    if src.intersects(MapAttributes::USER_READ | MapAttributes::USER_WRITE) {
+        result |= PageAttributes::U;
+    }
+    if src.contains(MapAttributes::DIRTY) {
+        result |= PageAttributes::SW_DIRTY;
+    }
+    result
+}
+
+fn to_map_attributes(src: PageAttributes) -> MapAttributes {
+    let mut result = MapAttributes::NON_GLOBAL;
+
+    if src.contains(PageAttributes::U) {
+        result |= MapAttributes::USER_READ;
+
+        if src.contains(PageAttributes::W) {
+            result |= MapAttributes::USER_WRITE;
+        }
+    }
+
+    if src.contains(PageAttributes::SW_DIRTY) {
+        result |= MapAttributes::DIRTY;
+    }
+
+    result
+}

kernel/arch/riscv64/src/mem/table.rs

@@ -0,0 +1,343 @@
+use core::{
+    fmt,
+    marker::PhantomData,
+    ops::{Index, IndexMut, Range},
+};
+
+use bitflags::bitflags;
+use libk_mm_interface::{
+    address::{AsPhysicalAddress, PhysicalAddress},
+    pointer::{PhysicalRef, PhysicalRefMut},
+    process::PageAttributeUpdate,
+    table::{
+        page_index, EntryLevel, EntryLevelDrop, NextPageTable, NonTerminalEntryLevel,
+        TableAllocator,
+    },
+};
+use yggdrasil_abi::error::Error;
+
+use super::{KernelTableManagerImpl, USER_BOUNDARY};
+
+bitflags! {
+    #[derive(Clone, Copy, PartialEq, Eq, Debug)]
+    pub struct PageAttributes: u64 {
+        const N = 1 << 63;
+        /// Software-tracked dirty bit (RSW[0])
+        const SW_DIRTY = 1 << 9;
+        /// Dirty bit
+        const D = 1 << 7;
+        /// Access bit
+        const A = 1 << 6;
+        /// Global mapping bit, implies all lower levels are also global
+        const G = 1 << 5;
+        /// U-mode access permission
+        const U = 1 << 4;
+        /// Execute permission
+        const X = 1 << 3;
+        /// Write permission
+        const W = 1 << 2;
+        /// Read-permission
+        const R = 1 << 1;
+        /// Valid bit
+        const V = 1 << 0;
+    }
+
+    // X W R    Meaning
+    // 0 0 0    Pointer to next level of page table
+    // 0 0 1    Read-only page
+    // 0 1 0    ---
+    // 0 1 1    Read-write page
+    // 1 0 0    Execute only
+    // 1 0 1    Read-execute page
+    // 1 1 0    ---
+    // 1 1 1    Read-write-execute page
+}
+
+/// L3 - entry is 4KiB
+#[derive(Debug, Clone, Copy)]
+pub struct L3;
+/// L2 - entry is 2MiB
+#[derive(Debug, Clone, Copy)]
+pub struct L2;
+/// L1 - entry is 1GiB
+#[derive(Debug, Clone, Copy)]
+pub struct L1;
+
+impl EntryLevel for L3 {
+    const SHIFT: usize = 12;
+}
+
+impl EntryLevel for L2 {
+    const SHIFT: usize = 21;
+}
+
+impl EntryLevel for L1 {
+    const SHIFT: usize = 30;
+}
+
+#[repr(C, align(0x1000))]
+pub struct PageTable<L: EntryLevel> {
+    pub(crate) entries: [PageEntry<L>; 512],
+}
+
+#[derive(Clone, Copy, Debug, PartialEq)]
+pub struct PageEntry<L: EntryLevel>(pub u64, PhantomData<L>);
+
+pub(super) trait DroppableRange {
+    const DROPPABLE_RANGE: Range<usize>;
+}
+
+impl DroppableRange for L1 {
+    const DROPPABLE_RANGE: Range<usize> = 0..page_index::<L1>(USER_BOUNDARY);
+}
+
+impl DroppableRange for L2 {
+    const DROPPABLE_RANGE: Range<usize> = 0..512;
+}
+
+impl NonTerminalEntryLevel for L1 {
+    type NextLevel = L2;
+}
+impl NonTerminalEntryLevel for L2 {
+    type NextLevel = L3;
+}
+
+impl<L: EntryLevel> PageTable<L> {
+    pub const fn zeroed() -> Self {
+        Self {
+            entries: [PageEntry::INVALID; 512],
+        }
+    }
+
+    pub fn new_zeroed<'a, TA: TableAllocator>(
+    ) -> Result<PhysicalRefMut<'a, PageTable<L>, KernelTableManagerImpl>, Error> {
+        let physical = TA::allocate_page_table()?;
+        let mut table =
+            unsafe { PhysicalRefMut::<'a, Self, KernelTableManagerImpl>::map(physical) };
+
+        for i in 0..512 {
+            table[i] = PageEntry::INVALID;
+        }
+
+        Ok(table)
+    }
+}
+
+impl<L: EntryLevel> PageEntry<L> {
+    // Upper + lower 10 bits
+    const ATTR_MASK: u64 = 0xFFC00000000003FF;
+    pub const INVALID: Self = Self(0, PhantomData);
+
+    /// Constructs a [PageEntry] from its raw representation.
+    ///
+    /// # Safety
+    ///
+    /// The caller must ensure `value` is actually a "valid" PTE.
+    pub const unsafe fn from_raw(value: u64) -> Self {
+        Self(value, PhantomData)
+    }
+
+    pub const fn is_present(&self) -> bool {
+        self.0 & PageAttributes::V.bits() != 0
+    }
+
+    pub fn update(&mut self, update: &PageAttributeUpdate) -> Result<(), Error> {
+        let mut attrs = self.attributes();
+        if let Some(write) = update.user_write {
+            attrs.set(PageAttributes::W, write);
+        }
+        if let Some(dirty) = update.dirty {
+            attrs.set(PageAttributes::SW_DIRTY, dirty);
+        }
+        self.0 &= !Self::ATTR_MASK;
+        self.0 |= attrs.bits() & Self::ATTR_MASK;
+        Ok(())
+    }
+
+    pub const fn is_dirty(&self) -> bool {
+        self.0 & PageAttributes::SW_DIRTY.bits() != 0
+    }
+
+    pub fn attributes(self) -> PageAttributes {
+        PageAttributes::from_bits_retain(self.0)
+    }
+}
+
+impl<L: NonTerminalEntryLevel + DroppableRange> EntryLevelDrop for PageTable<L>
+where
+    PageTable<L::NextLevel>: EntryLevelDrop,
+{
+    const FULL_RANGE: Range<usize> = L::DROPPABLE_RANGE;
+
+    unsafe fn drop_range<TA: TableAllocator>(&mut self, range: Range<usize>) {
+        for index in range {
+            let entry = self[index];
+
+            if let Some(table) = entry.as_table() {
+                unsafe {
+                    let mut table_ref: PhysicalRefMut<
+                        PageTable<L::NextLevel>,
+                        KernelTableManagerImpl,
+                    > = PhysicalRefMut::map(table);
+
+                    table_ref.drop_all::<TA>();
+
+                    TA::free_page_table(table);
+                }
+            } else if entry.is_present() {
+                // Memory must've been cleared beforehand, so no non-table entries must be present
+                panic!(
+                    "Expected a table containing only tables, got table[{}] = {:#x?}",
+                    index, entry.0
+                );
+            }
+
+            self[index] = PageEntry::INVALID;
+            // dc_cvac((&raw const self[index]).addr());
+        }
+    }
+}
+
+impl EntryLevelDrop for PageTable<L3> {
+    const FULL_RANGE: Range<usize> = 0..512;
+
+    // Do nothing
+    unsafe fn drop_range<TA: TableAllocator>(&mut self, _range: Range<usize>) {}
+}
+
+impl<L: NonTerminalEntryLevel + 'static> NextPageTable for PageTable<L> {
+    type NextLevel = PageTable<L::NextLevel>;
+    type TableRef = PhysicalRef<'static, PageTable<L::NextLevel>, KernelTableManagerImpl>;
+    type TableRefMut = PhysicalRefMut<'static, PageTable<L::NextLevel>, KernelTableManagerImpl>;
+
+    fn get(&self, index: usize) -> Option<Self::TableRef> {
+        let table = self[index].as_table()?;
+        Some(unsafe { PhysicalRef::map(table) })
+    }
+
+    fn get_mut(&mut self, index: usize) -> Option<Self::TableRefMut> {
+        let table = self[index].as_table()?;
+        Some(unsafe { PhysicalRefMut::map(table) })
+    }
+
+    fn get_mut_or_alloc<TA: TableAllocator>(
+        &mut self,
+        index: usize,
+    ) -> Result<Self::TableRefMut, Error> {
+        if let Some(table) = self[index].as_table() {
+            Ok(unsafe { PhysicalRefMut::map(table) })
+        } else {
+            let table = PageTable::new_zeroed::<TA>()?;
+            self[index] = PageEntry::<L>::table(
+                unsafe { table.as_physical_address() },
+                PageAttributes::empty(),
+            );
+            // dc_cvac((&raw const self[index]).addr());
+            Ok(table)
+        }
+    }
+}
+
+impl<L: NonTerminalEntryLevel> PageEntry<L> {
+    pub const fn identity_block(address: PhysicalAddress) -> Self {
+        Self(
+            (address.into_u64() >> 2)
+                | PageAttributes::R.bits()
+                | PageAttributes::W.bits()
+                | PageAttributes::X.bits()
+                | PageAttributes::V.bits()
+                | PageAttributes::D.bits()
+                | PageAttributes::A.bits(),
+            PhantomData,
+        )
+    }
+
+    pub fn block(address: PhysicalAddress, attrs: PageAttributes) -> Self {
+        // TODO validate address alignment
+        Self(
+            (address.into_u64() >> 2)
+                | (PageAttributes::R
+                    | PageAttributes::A
+                    | PageAttributes::D
+                    | PageAttributes::V
+                    | attrs)
+                    .bits(),
+            PhantomData,
+        )
+    }
+
+    pub fn table(address: PhysicalAddress, mut attrs: PageAttributes) -> Self {
+        attrs.remove(PageAttributes::R | PageAttributes::W | PageAttributes::X);
+        Self(
+            (address.into_u64() >> 2) | (PageAttributes::V | attrs).bits(),
+            PhantomData,
+        )
+    }
+
+    pub fn as_table(&self) -> Option<PhysicalAddress> {
+        (self.0
+            & (PageAttributes::R | PageAttributes::W | PageAttributes::X | PageAttributes::V)
+                .bits()
+            == PageAttributes::V.bits())
+        .then_some((self.0 & !Self::ATTR_MASK) << 2)
+        .map(PhysicalAddress::from_u64)
+    }
+}
+
+impl PageEntry<L3> {
+    pub fn page(address: PhysicalAddress, attrs: PageAttributes) -> Self {
+        Self(
+            (address.into_u64() >> 2)
+                | (PageAttributes::R
+                    | PageAttributes::A
+                    | PageAttributes::D
+                    | PageAttributes::V
+                    | attrs)
+                    .bits(),
+            PhantomData,
+        )
+    }
+
+    pub fn as_page(&self) -> Option<PhysicalAddress> {
+        (self.0 & PageAttributes::V.bits() != 0)
+            .then_some((self.0 & !Self::ATTR_MASK) << 2)
+            .map(PhysicalAddress::from_u64)
+    }
+}
+
+impl<L: EntryLevel> Index<usize> for PageTable<L> {
+    type Output = PageEntry<L>;
+
+    fn index(&self, index: usize) -> &Self::Output {
+        &self.entries[index]
+    }
+}
+
+impl<L: EntryLevel> IndexMut<usize> for PageTable<L> {
+    fn index_mut(&mut self, index: usize) -> &mut Self::Output {
+        &mut self.entries[index]
+    }
+}
+
+impl fmt::Display for PageAttributes {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        use fmt::Write;
+
+        macro_rules! bit {
+            ($self:ident, $field:expr, $letter:literal) => {
+                if $self.contains($field) {
+                    f.write_char($letter)
+                } else {
+                    f.write_char('-')
+                }
+            };
+        }
+
+        bit!(self, Self::R, 'r')?;
+        bit!(self, Self::W, 'w')?;
+        bit!(self, Self::X, 'x')?;
+        bit!(self, Self::U, 'u')?;
+
+        Ok(())
+    }
+}

kernel/arch/riscv64/src/registers.rs

@@ -0,0 +1,221 @@
+macro impl_csr_read($struct:ident, $repr:ty, $reg:ident, $register:ty) {
+    impl tock_registers::interfaces::Readable for $struct {
+        type T = $repr;
+        type R = $register;
+
+        #[inline]
+        fn get(&self) -> $repr {
+            let mut value: $repr;
+            unsafe {
+                core::arch::asm!(concat!("csrr {0}, ", stringify!($reg)), out(reg) value);
+            }
+            value
+        }
+    }
+}
+macro impl_csr_write($struct:ident, $repr:ty, $reg:ident, $register:ty) {
+    impl tock_registers::interfaces::Writeable for $struct {
+        type T = $repr;
+        type R = $register;
+
+        #[inline]
+        fn set(&self, value: $repr) {
+            unsafe {
+                core::arch::asm!(concat!("csrw ", stringify!($reg), ", {0}"), in(reg) value);
+            }
+        }
+    }
+}
+
+pub mod satp {
+    use tock_registers::register_bitfields;
+
+    use super::{impl_csr_read, impl_csr_write};
+
+    register_bitfields!(
+        u64,
+        pub SATP [
+            PPN OFFSET(0) NUMBITS(44) [],
+            ASID OFFSET(44) NUMBITS(16) [],
+            MODE OFFSET(60) NUMBITS(4) [
+                Bare = 0,
+                Sv39 = 8,
+                Sv48 = 9,
+                Sv57 = 10,
+                Sv64 = 11,
+            ],
+        ]
+    );
+
+    pub struct Reg;
+
+    impl_csr_read!(Reg, u64, satp, SATP::Register);
+    impl_csr_write!(Reg, u64, satp, SATP::Register);
+
+    pub const SATP: Reg = Reg;
+}
+
+pub mod stvec {
+    use tock_registers::{interfaces::ReadWriteable, register_bitfields};
+
+    use super::{impl_csr_read, impl_csr_write};
+
+    register_bitfields!(
+        u64,
+        pub STVEC [
+            MODE OFFSET(0) NUMBITS(2) [
+                Direct = 0,
+                Vectored = 1
+            ],
+            BASE OFFSET(2) NUMBITS(62) [],
+        ]
+    );
+
+    pub struct Reg;
+
+    impl_csr_read!(Reg, u64, stvec, STVEC::Register);
+    impl_csr_write!(Reg, u64, stvec, STVEC::Register);
+
+    impl Reg {
+        pub fn set_base(&self, base: usize) {
+            debug_assert_eq!(base & 0xF, 0);
+            let mask = match base & 63 != 0 {
+                false => 0,
+                true => 0x3 << 62,
+            };
+            self.modify(STVEC::BASE.val(((base as u64) >> 2) | mask));
+        }
+    }
+
+    pub const STVEC: Reg = Reg;
+}
+
+pub mod scause {
+    use tock_registers::register_bitfields;
+
+    use super::{impl_csr_read, impl_csr_write};
+
+    register_bitfields!(
+        u64,
+        pub SCAUSE [
+            CODE OFFSET(0) NUMBITS(63) [],
+            INTERRUPT OFFSET(63) NUMBITS(1) [],
+        ]
+    );
+
+    pub struct Reg;
+
+    impl_csr_read!(Reg, u64, scause, SCAUSE::Register);
+    impl_csr_write!(Reg, u64, scause, SCAUSE::Register);
+
+    pub const SCAUSE: Reg = Reg;
+}
+
+pub mod stval {
+    use super::{impl_csr_read, impl_csr_write};
+
+    pub struct Reg;
+
+    impl_csr_read!(Reg, u64, stval, ());
+    impl_csr_write!(Reg, u64, stval, ());
+
+    pub const STVAL: Reg = Reg;
+}
+
+pub mod sepc {
+    use super::{impl_csr_read, impl_csr_write};
+
+    pub struct Reg;
+
+    impl_csr_read!(Reg, u64, sepc, ());
+    impl_csr_write!(Reg, u64, sepc, ());
+
+    pub const SEPC: Reg = Reg;
+}
+
+pub mod sstatus {
+    use tock_registers::register_bitfields;
+
+    use super::{impl_csr_read, impl_csr_write};
+
+    register_bitfields!(
+        u64,
+        pub SSTATUS [
+            SUM OFFSET(18) NUMBITS(1) [],
+            SPP OFFSET(8) NUMBITS(1) [],
+            SIE OFFSET(1) NUMBITS(1) [],
+        ]
+    );
+
+    pub struct Reg;
+
+    impl_csr_read!(Reg, u64, sstatus, SSTATUS::Register);
+    impl_csr_write!(Reg, u64, sstatus, SSTATUS::Register);
+
+    pub const SSTATUS: Reg = Reg;
+}
+
+pub mod sscratch {
+    use super::{impl_csr_read, impl_csr_write};
+
+    pub struct Reg;
+
+    impl_csr_read!(Reg, u64, sscratch, ());
+    impl_csr_write!(Reg, u64, sscratch, ());
+
+    pub const SSCRATCH: Reg = Reg;
+}
+
+pub mod sip {
+    use tock_registers::register_bitfields;
+
+    use super::{impl_csr_read, impl_csr_write};
+
+    register_bitfields!(
+        u64,
+        pub SIP [
+            SSIP OFFSET(1) NUMBITS(1) [],
+            STIP OFFSET(5) NUMBITS(1) [],
+            SEIP OFFSET(9) NUMBITS(1) [],
+        ]
+    );
+
+    pub struct Reg;
+
+    impl_csr_read!(Reg, u64, sip, SIP::Register);
+    impl_csr_write!(Reg, u64, sip, SIP::Register);
+
+    pub const SIP: Reg = Reg;
+}
+
+pub mod sie {
+    use tock_registers::register_bitfields;
+
+    use super::{impl_csr_read, impl_csr_write};
+
+    register_bitfields!(
+        u64,
+        pub SIE [
+            SSIE OFFSET(1) NUMBITS(1) [],
+            STIE OFFSET(5) NUMBITS(1) [],
+            SEIE OFFSET(9) NUMBITS(1) [],
+        ]
+    );
+
+    pub struct Reg;
+
+    impl_csr_read!(Reg, u64, sie, SIE::Register);
+    impl_csr_write!(Reg, u64, sie, SIE::Register);
+
+    pub const SIE: Reg = Reg;
+}
+
+pub use satp::SATP;
+pub use scause::SCAUSE;
+pub use sepc::SEPC;
+pub use sie::SIE;
+pub use sip::SIP;
+pub use sscratch::SSCRATCH;
+pub use sstatus::SSTATUS;
+pub use stval::STVAL;
+pub use stvec::STVEC;

kernel/arch/riscv64/src/sbi.rs

@@ -0,0 +1,110 @@
+use yggdrasil_abi::{error::Error, primitive_enum};
+
+const EXT_HSM: u64 = 0x48534D;
+const EXT_TIME: u64 = 0x54494D45;
+const EXT_DBCN: u64 = 0x4442434E;
+const EXT_SPI: u64 = 0x735049;
+
+primitive_enum! {
+    pub enum Status: i64 {
+        Failed = -1,
+        NotSupported = -2,
+        InvalidParam = -3,
+        Denied = -4,
+        InvalidAddress = -5,
+        AlreadyAvailable = -6,
+        AlreadyStarted = -7,
+        AlreadyStopped = -8,
+        NoShmem = -9,
+        InvalidState = -10,
+        BadRange = -11,
+        Timeout = -12,
+        Io = -13,
+    }
+}
+
+primitive_enum! {
+    pub enum HartState: u64 {
+        Started = 0,
+        Stopped = 1,
+        StartPending = 2,
+        StopPending = 3,
+        Suspended = 4,
+        SuspendPending = 5,
+        ResumePending = 6,
+    }
+}
+
+pub enum SbiError {
+    Status(Status),
+    Other(i64),
+}
+
+impl From<i64> for SbiError {
+    #[inline]
+    fn from(value: i64) -> Self {
+        match Status::try_from(value) {
+            Ok(value) => Self::Status(value),
+            Err(_) => Self::Other(value),
+        }
+    }
+}
+
+#[allow(clippy::too_many_arguments)]
+#[inline(always)]
+unsafe fn sbi_do_call(
+    extension: u64,
+    function: u64,
+    mut a0: u64,
+    mut a1: u64,
+    a2: u64,
+    a3: u64,
+    a4: u64,
+    a5: u64,
+) -> Result<u64, SbiError> {
+    unsafe {
+        core::arch::asm!(
+            "ecall",
+            inlateout("a0") a0,
+            inlateout("a1") a1,
+            in("a2") a2,
+            in("a3") a3,
+            in("a4") a4,
+            in("a5") a5,
+            in("a6") function,
+            in("a7") extension,
+        );
+    }
+    let a0 = a0 as i64;
+    if a0 == 0 {
+        Ok(a1)
+    } else {
+        Err(a0.into())
+    }
+}
+
+pub fn sbi_hart_start(hart_id: u64, start_addr: u64, opaque: u64) -> Result<(), Error> {
+    match unsafe { sbi_do_call(EXT_HSM, 0x00, hart_id, start_addr, opaque, 0, 0, 0) } {
+        Ok(_) => Ok(()),
+        Err(SbiError::Status(Status::AlreadyAvailable)) => Err(Error::AlreadyExists),
+        Err(SbiError::Status(Status::InvalidParam)) => Err(Error::DoesNotExist),
+        Err(SbiError::Status(Status::InvalidAddress)) => Err(Error::InvalidArgument),
+        Err(_) => Err(Error::InvalidOperation),
+    }
+}
+
+pub fn sbi_send_ipi(hart_mask: u64, hart_mask_base: u64) -> Result<(), Error> {
+    match unsafe { sbi_do_call(EXT_SPI, 0x00, hart_mask, hart_mask_base, 0, 0, 0, 0) } {
+        Ok(_) => Ok(()),
+        Err(SbiError::Status(Status::InvalidParam)) => Err(Error::DoesNotExist),
+        Err(_) => Err(Error::InvalidOperation),
+    }
+}
+
+pub fn sbi_debug_console_write_byte(byte: u8) {
+    unsafe { sbi_do_call(EXT_DBCN, 0x02, byte as u64, 0, 0, 0, 0, 0) }.ok();
+}
+
+pub fn sbi_set_timer(next_event: u64) {
+    unsafe { sbi_do_call(EXT_TIME, 0x00, next_event, 0, 0, 0, 0, 0) }.ok();
+}

kernel/arch/src/lib.rs

@@ -26,8 +26,8 @@ cfg_if! {
         extern crate kernel_arch_aarch64 as imp;
     } else if #[cfg(target_arch = "x86_64")] {
         extern crate kernel_arch_x86_64 as imp;
-    } else if #[cfg(target_arch = "x86")] {
-        extern crate kernel_arch_i686 as imp;
+    } else if #[cfg(target_arch = "riscv64")] {
+        extern crate kernel_arch_riscv64 as imp;
     } else {
         compile_error!("Unsupported architecture");
     }

kernel/arch/x86/src/cpuid.rs

@@ -95,10 +95,17 @@ cpuid_features! {
     ]
 }
 
+cpuid_features! {
+    pub ExtEdxFeatures: u32 [
+        PDPE1GB: 26
+    ]
+}
+
 #[derive(Clone, Copy, Debug)]
 pub struct CpuFeatures {
     pub ecx: EcxFeatures,
     pub edx: EdxFeatures,
+    pub ext_edx: ExtEdxFeatures,
 }
 
 impl CpuFeatures {
@@ -106,6 +113,7 @@ impl CpuFeatures {
         Self {
             ecx: EcxFeatures::empty(),
             edx: EdxFeatures::empty(),
+            ext_edx: ExtEdxFeatures::empty(),
         }
     }
 
@@ -120,6 +128,7 @@ impl CpuFeatures {
             Err(Self {
                 ecx: features.ecx & !self.ecx,
                 edx: features.edx & !self.edx,
+                ext_edx: features.ext_edx & !self.ext_edx,
             })
         }
     }
@@ -132,6 +141,7 @@ impl BitAnd<CpuFeatures> for CpuFeatures {
         Self {
             ecx: self.ecx & rhs.ecx,
             edx: self.edx & rhs.edx,
+            ext_edx: self.ext_edx & rhs.ext_edx,
         }
     }
 }
@@ -143,6 +153,7 @@ impl BitOr<CpuFeatures> for CpuFeatures {
         Self {
             ecx: self.ecx | rhs.ecx,
             edx: self.edx | rhs.edx,
+            ext_edx: self.ext_edx | rhs.ext_edx,
         }
     }
 }
@@ -151,8 +162,9 @@ impl CpuFeatureSet for CpuFeatures {
     fn iter(&self) -> impl Iterator<Item = &'static str> {
         let ecx = self.ecx.iter().map(|e| e.as_str());
         let edx = self.edx.iter().map(|e| e.as_str());
+        let ext_edx = self.ext_edx.iter().map(|e| e.as_str());
 
-        core::iter::chain(ecx, edx)
+        core::iter::chain(core::iter::chain(ecx, edx), ext_edx)
     }
 }
 
@@ -190,20 +202,26 @@ unsafe fn raw_cpuid(eax: u32, result: &mut [u32]) {
     );
 }
 
-fn cpuid_features() -> (EcxFeatures, EdxFeatures) {
+fn cpuid_features() -> (EcxFeatures, EdxFeatures, ExtEdxFeatures) {
     let mut raw = [0; 3];
 
     unsafe {
         raw_cpuid(0x1, &mut raw);
     }
 
-    (
-        EcxFeatures::from_bits_truncate(raw[2]),
-        EdxFeatures::from_bits_truncate(raw[1]),
-    )
+    let ecx = EcxFeatures::from_bits_truncate(raw[2]);
+    let edx = EdxFeatures::from_bits_truncate(raw[1]);
+
+    unsafe {
+        raw_cpuid(0x80000001, &mut raw);
+    }
+
+    let ext_edx = ExtEdxFeatures::from_bits_truncate(raw[1]);
+
+    (ecx, edx, ext_edx)
 }
 
-fn enable_features(ecx: EcxFeatures, edx: EdxFeatures) {
+fn enable_features(ecx: EcxFeatures, edx: EdxFeatures, _ext_edx: ExtEdxFeatures) {
     if ecx.contains(EcxFeatures::XSAVE) {
         CR4.modify(CR4::OSXSAVE::SET);
     }
@@ -230,13 +248,16 @@ fn enable_features(ecx: EcxFeatures, edx: EdxFeatures) {
     if ecx.contains(EcxFeatures::PCID) {
         CR4.modify(CR4::PCIDE::SET);
     }
+    if edx.contains(EdxFeatures::PSE) {
+        CR4.modify(CR4::PSE::SET);
+    }
 
     CR0.modify(CR0::TS::CLEAR);
 }
 
 fn read_features() -> CpuFeatures {
-    let (ecx, edx) = cpuid_features();
-    CpuFeatures { ecx, edx }
+    let (ecx, edx, ext_edx) = cpuid_features();
+    CpuFeatures { ecx, edx, ext_edx }
 }
 
 pub fn setup_features(
@@ -250,7 +271,7 @@ pub fn setup_features(
         return (have_features, Err(missing_features));
     }
 
-    enable_features(will_features.ecx, will_features.edx);
+    enable_features(will_features.ecx, will_features.edx, will_features.ext_edx);
 
     (have_features, Ok(will_features))
 }

kernel/arch/x86/src/gdt.rs

@@ -1,5 +1,5 @@
 #[allow(dead_code)]
-#[repr(packed)]
+#[repr(C, packed)]
 pub struct Entry {
     pub limit_lo: u16,
     pub base_lo: u16,
@@ -10,7 +10,7 @@ pub struct Entry {
 }
 
 #[allow(dead_code)]
-#[repr(packed)]
+#[repr(C, packed)]
 pub struct Pointer {
     pub limit: u16,
     pub offset: usize,
@@ -121,7 +121,7 @@ mod imp {
     use super::{Entry, Pointer};
 
     #[allow(dead_code)]
-    #[repr(packed)]
+    #[repr(C, packed)]
     pub struct Tss {
         _0: u32,
         rsp0: u64,

kernel/arch/x86/src/intrinsics.rs

@@ -16,6 +16,10 @@ pub trait IoPortAccess<T> {
     fn read(&self) -> T;
     /// Writes a value to the port
     fn write(&self, value: T);
+
+    fn modify<F: FnOnce(T) -> T>(&self, f: F) {
+        self.write(f(self.read()));
+    }
 }
 
 impl<T> IoPort<T> {

kernel/arch/x86/src/lib.rs

@@ -6,4 +6,10 @@ extern crate alloc;
 
 pub mod cpuid;
 pub mod gdt;
+pub mod intrinsics;
 pub mod registers;
+
+#[cfg(any(target_arch = "x86_64", rust_analyzer))]
+pub const ISA_IRQ_OFFSET: u32 = 1024;
+#[cfg(any(target_arch = "x86", rust_analyzer))]
+pub const ISA_IRQ_OFFSET: u32 = 0;

kernel/arch/x86/src/registers.rs

@@ -146,7 +146,7 @@ mod cr2 {
 }
 
 mod cr3 {
-    use tock_registers::{interfaces::ReadWriteable, register_bitfields};
+    use tock_registers::{interfaces::Writeable, register_bitfields};
 
     register_bitfields! {
         usize,
@@ -164,7 +164,7 @@ mod cr3 {
     impl Reg {
         pub fn set_address(&self, address: usize) {
             assert_eq!(address & 0xFFF, 0);
-            self.modify(CR3::ADDR.val(address >> 12))
+            self.write(CR3::ADDR.val(address >> 12))
         }
     }
 

kernel/arch/x86_64/Cargo.toml

@@ -7,7 +7,6 @@ edition = "2021"
 yggdrasil-abi.workspace = true
 kernel-arch-interface.workspace = true
 libk-mm-interface.workspace = true
-memtables.workspace = true
 device-api = { workspace = true, features = ["derive"] }
 kernel-arch-x86.workspace = true
 

kernel/arch/x86_64/src/context.rs

@@ -5,11 +5,14 @@ use kernel_arch_interface::{
     task::{ForkFrame, StackBuilder, TaskContext, TaskFrame, UserContextInfo},
 };
 use kernel_arch_x86::registers::{FpuContext, CR3, MSR_IA32_FS_BASE};
-use libk_mm_interface::address::{AsPhysicalAddress, PhysicalAddress};
+use libk_mm_interface::address::PhysicalAddress;
 use tock_registers::interfaces::Writeable;
 use yggdrasil_abi::{arch::SavedFrame, error::Error};
 
-use crate::{mem::KERNEL_TABLES, ArchitectureImpl};
+use crate::{
+    mem::{auto_lower_address, fixed},
+    ArchitectureImpl,
+};
 
 /// Frame saved onto the stack when taking an IRQ
 #[derive(Debug)]
@@ -431,7 +434,7 @@ impl<K: KernelTableManager, PA: PhysicalMemoryAllocator<Address = PhysicalAddres
     fn kernel(entry: extern "C" fn(usize) -> !, arg: usize) -> Result<Self, Error> {
         const KERNEL_TASK_PAGES: usize = 32;
 
-        let cr3: usize = unsafe { KERNEL_TABLES.lock().as_physical_address() }.into();
+        let cr3: usize = auto_lower_address(&raw const fixed::KERNEL_PML4); // unsafe { KERNEL_TABLES.lock().as_physical_address() }.into();
 
         let stack_base_phys = PA::allocate_contiguous_pages(KERNEL_TASK_PAGES)?;
         let stack_base = stack_base_phys.raw_virtualize::<K>();
@@ -536,7 +539,7 @@ impl<K: KernelTableManager, PA: PhysicalMemoryAllocator<Address = PhysicalAddres
     for TaskContextImpl<K, PA>
 {
     fn drop(&mut self) {
-        log::info!("Drop Context {:#p}", self);
+        log::trace!("Drop Context {:#p}", self);
         assert_eq!(self.stack_size % 0x1000, 0);
 
         for offset in (0..self.stack_size).step_by(0x1000) {

kernel/arch/x86_64/src/lib.rs

@@ -1,18 +1,17 @@
 #![no_std]
 #![allow(clippy::new_without_default)]
-#![feature(naked_functions, trait_upcasting)]
 
 extern crate alloc;
 
 use core::{
-    ops::DerefMut,
+    ops::{DerefMut, Range},
     sync::atomic::{AtomicUsize, Ordering},
 };
 
-use alloc::vec::Vec;
+use alloc::{sync::Arc, vec::Vec};
 use device_api::interrupt::{LocalInterruptController, MessageInterruptController};
 use kernel_arch_interface::{
-    cpu::{CpuImpl, IpiQueue},
+    cpu::{CpuData, CpuImpl, IpiQueue},
     task::Scheduler,
     util::OneTimeInit,
     Architecture,
@@ -52,32 +51,33 @@ pub struct PerCpuData {
     // 0x10, used in assembly
     pub tmp_address: usize,
 
-    pub local_apic: &'static dyn LocalApicInterface,
+    pub local_apic: Arc<dyn LocalApicInterface>,
     pub available_features: CpuFeatures,
     pub enabled_features: CpuFeatures,
 }
 
+impl CpuData for PerCpuData {}
+
 impl PerCpuData {
-    pub fn local_apic(&self) -> &'static dyn LocalApicInterface {
-        self.local_apic
+    #[inline]
+    pub fn local_apic(&self) -> &dyn LocalApicInterface {
+        self.local_apic.as_ref()
     }
 }
 
 static IPI_QUEUES: OneTimeInit<Vec<IpiQueue<ArchitectureImpl>>> = OneTimeInit::new();
 pub static CPU_COUNT: AtomicUsize = AtomicUsize::new(1);
 
-#[naked]
+#[unsafe(naked)]
 extern "C" fn idle_task(_: usize) -> ! {
-    unsafe {
-        core::arch::naked_asm!(
-            r#"
+    core::arch::naked_asm!(
+        r#"
             1:
                 nop
                 jmp 1b
             "#,
-            options(att_syntax)
-        );
-    }
+        options(att_syntax)
+    );
 }
 
 impl ArchitectureImpl {
@@ -182,12 +182,7 @@ impl Architecture for ArchitectureImpl {
 
     fn local_interrupt_controller() -> Option<&'static dyn LocalInterruptController> {
         let cpu = Self::local_cpu_data()?;
-        Some(cpu.local_apic)
-    }
-
-    fn message_interrupt_controller() -> &'static dyn MessageInterruptController {
-        let local = Self::local_cpu_data().unwrap();
-        local.local_apic
+        Some(cpu.local_apic.as_ref())
     }
 
     fn cpu_enabled_features<S: Scheduler>(cpu: &CpuImpl<Self, S>) -> Option<&Self::CpuFeatures> {
@@ -197,4 +192,30 @@ impl Architecture for ArchitectureImpl {
     fn cpu_available_features<S: Scheduler>(cpu: &CpuImpl<Self, S>) -> Option<&Self::CpuFeatures> {
         Some(&cpu.available_features)
     }
+
+    // Cache/barrier
+
+    fn load_barrier() {
+        unsafe { core::arch::x86_64::_mm_lfence() };
+    }
+
+    fn store_barrier() {
+        unsafe { core::arch::x86_64::_mm_sfence() };
+    }
+
+    fn memory_barrier() {
+        unsafe { core::arch::x86_64::_mm_mfence() };
+    }
+
+    fn flush_virtual_range(range: Range<usize>) {
+        // TODO I assume 64-byte cache line on all CPUs
+        // TODO clflush instruction may not be available, test for it
+        const CLSIZE: usize = 64;
+        let start = range.start & !(CLSIZE - 1);
+        let end = (range.end + (CLSIZE - 1)) & !(CLSIZE - 1);
+
+        for line in (start..end).step_by(CLSIZE) {
+            unsafe { core::arch::x86_64::_mm_clflush(line as _) };
+        }
+    }
 }

kernel/arch/x86_64/src/mem/fixed.rs

@@ -0,0 +1,131 @@
+use core::ops::Range;
+
+use kernel_arch_interface::{mem::DeviceMemoryAttributes, sync::IrqSafeSpinlock, Architecture};
+use kernel_arch_x86::registers::CR3;
+use libk_mm_interface::{
+    address::PhysicalAddress,
+    device::{DevicePageManager, DevicePageTableLevel},
+    table::{page_index, EntryLevel},
+};
+
+use crate::{
+    mem::{
+        auto_lower_address,
+        table::{PageAttributes, PageEntry, PageTable, L0, L1, L2, L3},
+    },
+    ArchitectureImpl, KERNEL_VIRT_OFFSET,
+};
+
+pub const IDENTITY_SIZE_L1: usize = 64;
+pub const KERNEL_L0I: usize = page_index::<L0>(KERNEL_VIRT_OFFSET);
+
+pub const DEVICE_L1: usize = IDENTITY_SIZE_L1;
+pub const DEVICE_MAPPING_L3_COUNT: usize = 32;
+pub const DEVICE_MAPPING_OFFSET: usize = KERNEL_VIRT_OFFSET + (DEVICE_L1 << L1::SHIFT);
+
+pub static LOCK: IrqSafeSpinlock<ArchitectureImpl, ()> = IrqSafeSpinlock::new(());
+pub static mut KERNEL_PDPT: PageTable<L1> = PageTable::zeroed();
+pub static mut KERNEL_PML4: PageTable<L0> = PageTable::zeroed();
+
+pub(super) static mut DEVICE_MEMORY: DevicePageManager<L3DeviceMemory, L2DeviceMemory> =
+    DevicePageManager::new(
+        L3DeviceMemory([PageTable::zeroed(); DEVICE_MAPPING_L3_COUNT]),
+        L2DeviceMemory(PageTable::zeroed()),
+    );
+
+#[repr(transparent)]
+pub struct L2DeviceMemory(pub PageTable<L2>);
+#[repr(transparent)]
+pub struct L3DeviceMemory(pub [PageTable<L3>; DEVICE_MAPPING_L3_COUNT]);
+
+impl DevicePageTableLevel for L2DeviceMemory {
+    type Level = L2;
+    const VIRTUAL_BASE: usize = DEVICE_MAPPING_OFFSET;
+    const INDEX_RANGE: Range<usize> = DEVICE_MAPPING_L3_COUNT..512;
+
+    fn map_page(
+        &mut self,
+        index: usize,
+        physical: PhysicalAddress,
+        attrs: &DeviceMemoryAttributes,
+    ) {
+        let _ = attrs;
+        self.0[index] = PageEntry::<L2>::block(physical, PageAttributes::WRITABLE);
+    }
+
+    fn unmap_page(&mut self, index: usize) {
+        self.0[index - DEVICE_MAPPING_L3_COUNT] = PageEntry::INVALID;
+    }
+
+    fn is_mapped(&self, index: usize) -> bool {
+        self.0[index - DEVICE_MAPPING_L3_COUNT].is_present()
+    }
+
+    fn flush_range(range: Range<usize>) {
+        let _ = range;
+    }
+}
+
+impl DevicePageTableLevel for L3DeviceMemory {
+    type Level = L3;
+    const VIRTUAL_BASE: usize = DEVICE_MAPPING_OFFSET;
+    const INDEX_RANGE: Range<usize> = 0..512 * DEVICE_MAPPING_L3_COUNT;
+
+    fn map_page(
+        &mut self,
+        index: usize,
+        physical: PhysicalAddress,
+        attrs: &DeviceMemoryAttributes,
+    ) {
+        let _ = attrs;
+        let l2i = index / 512;
+        let l3i = index % 512;
+        self.0[l2i][l3i] = PageEntry::page(physical, PageAttributes::WRITABLE);
+    }
+
+    fn unmap_page(&mut self, index: usize) {
+        let l2i = index / 512;
+        let l3i = index % 512;
+        self.0[l2i][l3i] = PageEntry::INVALID;
+    }
+
+    fn is_mapped(&self, index: usize) -> bool {
+        let l2i = index / 512;
+        let l3i = index % 512;
+        self.0[l2i][l3i].is_present()
+    }
+
+    fn flush_range(range: Range<usize>) {
+        let _ = range;
+    }
+}
+
+pub(super) unsafe fn setup(have_1gib_pages: bool) {
+    let phys = PhysicalAddress::from_usize(auto_lower_address(&raw const KERNEL_PDPT));
+    KERNEL_PML4[KERNEL_L0I] = PageEntry::table(phys, PageAttributes::WRITABLE);
+
+    if have_1gib_pages {
+        for i in 0..IDENTITY_SIZE_L1 {
+            let phys = PhysicalAddress::from_usize(i * L1::SIZE);
+            KERNEL_PDPT[i] = PageEntry::<L1>::block(phys, PageAttributes::WRITABLE);
+        }
+    } else {
+        // TODO
+        ArchitectureImpl::halt();
+    }
+
+    // DEVICE_L1 -> Device L2 table
+    //               0..DEVICE_MAPPING_L3_COUNT -> Device L3 tables -> Device L3 pages
+    //               ..512                      -> Device L2 pages
+    for i in 0..DEVICE_MAPPING_L3_COUNT {
+        let phys =
+            PhysicalAddress::from_usize(auto_lower_address(&raw const DEVICE_MEMORY.normal.0[i]));
+        DEVICE_MEMORY.large.0[i] = PageEntry::table(phys, PageAttributes::WRITABLE);
+    }
+    let phys = PhysicalAddress::from_usize(auto_lower_address(&raw const DEVICE_MEMORY.large.0));
+    KERNEL_PDPT[DEVICE_L1] = PageEntry::table(phys, PageAttributes::WRITABLE);
+}
+
+pub(super) unsafe fn load() {
+    CR3.set_address(auto_lower_address(&raw const KERNEL_PML4));
+}

kernel/arch/x86_64/src/mem/mod.rs

@@ -1,100 +1,37 @@
-use core::{
-    alloc::Layout,
-    ops::{Deref, DerefMut},
-    ptr::addr_of,
-    sync::atomic::{AtomicUsize, Ordering},
+use kernel_arch_interface::mem::{
+    DeviceMemoryAttributes, KernelTableManager, RawDeviceMemoryMapping,
 };
-
-use kernel_arch_interface::{
-    mem::{DeviceMemoryAttributes, KernelTableManager, RawDeviceMemoryMapping},
-    sync::split_spinlock,
-};
-use kernel_arch_x86::registers::CR3;
-use libk_mm_interface::{
-    address::PhysicalAddress,
-    table::{page_index, EntryLevel, EntryLevelExt},
-};
-use memtables::x86_64::FixedTables;
-use static_assertions::{const_assert_eq, const_assert_ne};
+use libk_mm_interface::{address::PhysicalAddress, table::EntryLevel};
 use yggdrasil_abi::error::Error;
 
 use crate::KERNEL_VIRT_OFFSET;
 
-use self::table::{PageAttributes, PageEntry, PageTable, L0, L1, L2, L3};
+use self::table::{PageTable, L0, L1};
 
+pub mod fixed;
 pub mod process;
 pub mod table;
 
 #[derive(Debug)]
 pub struct KernelTableManagerImpl;
 
-const CANONICAL_ADDRESS_MASK: usize = 0xFFFF000000000000;
-const KERNEL_PHYS_BASE: usize = 0x200000;
-
-// Mapped at compile time
-const KERNEL_MAPPING_BASE: usize = KERNEL_VIRT_OFFSET + KERNEL_PHYS_BASE;
-const KERNEL_L0_INDEX: usize = page_index::<L0>(KERNEL_MAPPING_BASE);
-const KERNEL_L1_INDEX: usize = page_index::<L1>(KERNEL_MAPPING_BASE);
-const KERNEL_START_L2_INDEX: usize = page_index::<L2>(KERNEL_MAPPING_BASE);
-
-// Must not be zero, should be at 4MiB
-const_assert_ne!(KERNEL_START_L2_INDEX, 0);
-// From static mapping
-const_assert_eq!(KERNEL_L0_INDEX, 511);
-const_assert_eq!(KERNEL_L1_INDEX, 0);
-
-// Mapped at boot
-const EARLY_MAPPING_L2I: usize = KERNEL_START_L2_INDEX - 1;
-const DEVICE_MAPPING_L1I: usize = KERNEL_L1_INDEX + 2;
-const RAM_MAPPING_L0I: usize = KERNEL_L0_INDEX - 1;
-
-const DEVICE_MAPPING_L3_COUNT: usize = 4;
-
-split_spinlock! {
-    use crate::ArchitectureImpl;
-    use crate::mem::FixedTables;
-    use libk_mm_interface::KernelImageObject;
-
-    #[link_section = ".data.tables"]
-    static KERNEL_TABLES<lock: ArchitectureImpl>: KernelImageObject<FixedTables> =
-        unsafe { KernelImageObject::new(FixedTables::zeroed()) };
-}
-
-// 2MiB for early mappings
-const EARLY_MAPPING_OFFSET: usize = CANONICAL_ADDRESS_MASK
-    | (KERNEL_L0_INDEX * L0::SIZE)
-    | (KERNEL_L1_INDEX * L1::SIZE)
-    | (EARLY_MAPPING_L2I * L2::SIZE);
-static mut EARLY_MAPPING_L3: PageTable<L3> = PageTable::zeroed();
-// 1GiB for device MMIO mapping
-const DEVICE_MAPPING_OFFSET: usize =
-    CANONICAL_ADDRESS_MASK | (KERNEL_L0_INDEX * L0::SIZE) | (DEVICE_MAPPING_L1I * L1::SIZE);
-static mut DEVICE_MAPPING_L2: PageTable<L2> = PageTable::zeroed();
-static mut DEVICE_MAPPING_L3S: [PageTable<L3>; DEVICE_MAPPING_L3_COUNT] =
-    [PageTable::zeroed(); DEVICE_MAPPING_L3_COUNT];
-// 512GiB for whole RAM mapping
-pub const RAM_MAPPING_OFFSET: usize = CANONICAL_ADDRESS_MASK | (RAM_MAPPING_L0I * L0::SIZE);
-pub static MEMORY_LIMIT: AtomicUsize = AtomicUsize::new(0);
-pub static mut RAM_MAPPING_L1: PageTable<L1> = PageTable::zeroed();
-
 impl KernelTableManager for KernelTableManagerImpl {
     fn virtualize(address: u64) -> usize {
         let address = address as usize;
-        if address < MEMORY_LIMIT.load(Ordering::Acquire) {
-            address + RAM_MAPPING_OFFSET
+        if address < fixed::IDENTITY_SIZE_L1 * L1::SIZE {
+            address + KERNEL_VIRT_OFFSET
         } else {
-            panic!("Invalid physical address: {:#x}", address);
+            panic!("Invalid physical address: {address:#x}");
         }
     }
 
     fn physicalize(address: usize) -> u64 {
-        if address < RAM_MAPPING_OFFSET
-            || address - RAM_MAPPING_OFFSET >= MEMORY_LIMIT.load(Ordering::Acquire)
+        if address < KERNEL_VIRT_OFFSET
+            || address - KERNEL_VIRT_OFFSET >= fixed::IDENTITY_SIZE_L1 * L1::SIZE
         {
-            panic!("Not a virtualized physical address: {:#x}", address);
+            panic!("Invalid virtualized address: {address:#x}");
         }
-
-        (address - RAM_MAPPING_OFFSET) as _
+        (address - KERNEL_VIRT_OFFSET) as u64
     }
 
     unsafe fn map_device_pages(
@@ -102,238 +39,30 @@ impl KernelTableManager for KernelTableManagerImpl {
         count: usize,
         attrs: DeviceMemoryAttributes,
     ) -> Result<RawDeviceMemoryMapping<Self>, Error> {
-        map_device_memory(PhysicalAddress::from_u64(base), count, attrs)
+        let _lock = fixed::LOCK.lock();
+        #[allow(static_mut_refs)]
+        fixed::DEVICE_MEMORY.map_device_pages(PhysicalAddress::from_u64(base), count, attrs)
     }
 
     unsafe fn unmap_device_pages(mapping: &RawDeviceMemoryMapping<Self>) {
-        unmap_device_memory(mapping)
-    }
-}
-
-// Early mappings
-unsafe fn map_early_pages(physical: PhysicalAddress, count: usize) -> Result<usize, Error> {
-    for l3i in 0..512 {
-        let mut taken = false;
-        for i in 0..count {
-            if EARLY_MAPPING_L3[i + l3i].is_present() {
-                taken = true;
-                break;
-            }
-        }
-
-        if taken {
-            continue;
-        }
-
-        for i in 0..count {
-            // TODO NX, NC
-            EARLY_MAPPING_L3[i + l3i] =
-                PageEntry::page(physical.add(i * L3::SIZE), PageAttributes::WRITABLE);
-        }
-
-        return Ok(EARLY_MAPPING_OFFSET + l3i * L3::SIZE);
-    }
-
-    Err(Error::OutOfMemory)
-}
-
-unsafe fn unmap_early_page(address: usize) {
-    if !(EARLY_MAPPING_OFFSET..EARLY_MAPPING_OFFSET + L2::SIZE).contains(&address) {
-        panic!("Tried to unmap invalid early mapping: {:#x}", address);
-    }
-
-    let l3i = (address - EARLY_MAPPING_OFFSET).page_index::<L3>();
-
-    assert!(EARLY_MAPPING_L3[l3i].is_present());
-    EARLY_MAPPING_L3[l3i] = PageEntry::INVALID;
-}
-
-// Device mappings
-unsafe fn map_device_memory_l3(
-    base: PhysicalAddress,
-    count: usize,
-    _attrs: DeviceMemoryAttributes,
-) -> Result<usize, Error> {
-    // TODO don't map pages if already mapped
-
-    'l0: for i in 0..DEVICE_MAPPING_L3_COUNT * 512 {
-        for j in 0..count {
-            let l2i = (i + j) / 512;
-            let l3i = (i + j) % 512;
-
-            if DEVICE_MAPPING_L3S[l2i][l3i].is_present() {
-                continue 'l0;
-            }
-        }
-
-        for j in 0..count {
-            let l2i = (i + j) / 512;
-            let l3i = (i + j) % 512;
-
-            // TODO NX, NC
-            DEVICE_MAPPING_L3S[l2i][l3i] =
-                PageEntry::page(base.add(j * L3::SIZE), PageAttributes::WRITABLE);
-        }
-
-        return Ok(DEVICE_MAPPING_OFFSET + i * L3::SIZE);
-    }
-
-    Err(Error::OutOfMemory)
-}
-
-unsafe fn map_device_memory_l2(
-    base: PhysicalAddress,
-    count: usize,
-    _attrs: DeviceMemoryAttributes,
-) -> Result<usize, Error> {
-    'l0: for i in DEVICE_MAPPING_L3_COUNT..512 {
-        for j in 0..count {
-            if DEVICE_MAPPING_L2[i + j].is_present() {
-                continue 'l0;
-            }
-        }
-
-        for j in 0..count {
-            DEVICE_MAPPING_L2[i + j] =
-                PageEntry::<L2>::block(base.add(j * L2::SIZE), PageAttributes::WRITABLE);
-        }
-
-        return Ok(DEVICE_MAPPING_OFFSET + i * L2::SIZE);
-    }
-
-    Err(Error::OutOfMemory)
-}
-
-unsafe fn map_device_memory(
-    base: PhysicalAddress,
-    size: usize,
-    attrs: DeviceMemoryAttributes,
-) -> Result<RawDeviceMemoryMapping<KernelTableManagerImpl>, Error> {
-    let l3_aligned = base.page_align_down::<L3>();
-    let l3_offset = base.page_offset::<L3>();
-    let page_count = (l3_offset + size).page_count::<L3>();
-
-    if page_count > 256 {
-        // Large mapping, use L2 mapping instead
-        let l2_aligned = base.page_align_down::<L2>();
-        let l2_offset = base.page_offset::<L2>();
-        let page_count = (l2_offset + size).page_count::<L2>();
-
-        let base_address = map_device_memory_l2(l2_aligned, page_count, attrs)?;
-        let address = base_address + l2_offset;
-
-        Ok(RawDeviceMemoryMapping::from_raw_parts(
-            address,
-            base_address,
-            page_count,
-            L2::SIZE,
-        ))
-    } else {
-        // Just map the pages directly
-        let base_address = map_device_memory_l3(l3_aligned, page_count, attrs)?;
-        let address = base_address + l3_offset;
-
-        Ok(RawDeviceMemoryMapping::from_raw_parts(
-            address,
-            base_address,
-            page_count,
-            L3::SIZE,
-        ))
-    }
-}
-
-unsafe fn unmap_device_memory(map: &RawDeviceMemoryMapping<KernelTableManagerImpl>) {
-    match map.page_size {
-        L3::SIZE => {
-            for i in 0..map.page_count {
-                let page = map.base_address + i * L3::SIZE;
-                let l2i = page.page_index::<L2>();
-                let l3i = page.page_index::<L3>();
-                assert!(DEVICE_MAPPING_L3S[l2i][l3i].is_present());
-                DEVICE_MAPPING_L3S[l2i][l3i] = PageEntry::INVALID;
-                flush_tlb_entry(page);
-            }
-        }
-        L2::SIZE => todo!(),
-        _ => unimplemented!(),
-    }
-}
-
-/// Memory mapping which may be used for performing early kernel initialization
-pub struct EarlyMapping<'a, T: ?Sized> {
-    value: &'a mut T,
-    page_count: usize,
-}
-
-impl<'a, T: Sized> EarlyMapping<'a, T> {
-    /// # Safety
-    ///
-    /// `physical` address provided must be a valid non-NULL address actually containing `T`.
-    pub unsafe fn map(physical: PhysicalAddress) -> Result<EarlyMapping<'a, T>, Error> {
-        let layout = Layout::new::<T>();
-        let aligned = physical.page_align_down::<L3>();
-        let offset = physical.page_offset::<L3>();
-        let page_count = (offset + layout.size()).div_ceil(L3::SIZE);
-
-        let virt = map_early_pages(aligned, page_count)?;
-        let value = &mut *((virt + offset) as *mut T);
-
-        Ok(EarlyMapping { value, page_count })
-    }
-
-    /// # Safety
-    ///
-    /// `physical` address provided must be a valid non-NULL address actually containing a `T`
-    /// slice of given `len`.
-    pub unsafe fn map_slice(
-        physical: PhysicalAddress,
-        len: usize,
-    ) -> Result<EarlyMapping<'a, [T]>, Error> {
-        let layout = Layout::array::<T>(len).unwrap();
-        let aligned = physical.page_align_down::<L3>();
-        let offset = physical.page_offset::<L3>();
-        let page_count = (offset + layout.size()).div_ceil(L3::SIZE);
-
-        let virt = map_early_pages(aligned, page_count)?;
-        let value = core::slice::from_raw_parts_mut((virt + offset) as *mut T, len);
-
-        Ok(EarlyMapping { value, page_count })
-    }
-}
-
-impl<T: ?Sized> Deref for EarlyMapping<'_, T> {
-    type Target = T;
-
-    fn deref(&self) -> &Self::Target {
-        self.value
-    }
-}
-
-impl<T: ?Sized> DerefMut for EarlyMapping<'_, T> {
-    fn deref_mut(&mut self) -> &mut Self::Target {
-        self.value
-    }
-}
-
-impl<T: ?Sized> Drop for EarlyMapping<'_, T> {
-    fn drop(&mut self) {
-        let address = (self.value as *mut T).addr() & !(L3::SIZE - 1);
-
-        for i in 0..self.page_count {
-            let page = address + i * L3::SIZE;
-
-            unsafe {
-                unmap_early_page(page);
-            }
-        }
+        let _lock = fixed::LOCK.lock();
+        #[allow(static_mut_refs)]
+        fixed::DEVICE_MEMORY.unmap_device_pages(mapping);
     }
 }
 
 pub fn clone_kernel_tables(dst: &mut PageTable<L0>) {
-    let tables = KERNEL_TABLES.lock();
     unsafe {
-        dst[KERNEL_L0_INDEX] = PageEntry::from_raw(tables.l0.data[KERNEL_L0_INDEX]);
-        dst[RAM_MAPPING_L0I] = PageEntry::from_raw(tables.l0.data[RAM_MAPPING_L0I]);
+        dst[fixed::KERNEL_L0I] = fixed::KERNEL_PML4[fixed::KERNEL_L0I];
+    }
+}
+
+pub fn auto_lower_address<T>(pointer: *const T) -> usize {
+    let address = pointer.addr();
+    if address < KERNEL_VIRT_OFFSET {
+        address
+    } else {
+        address - KERNEL_VIRT_OFFSET
     }
 }
 
@@ -353,35 +82,12 @@ pub fn clone_kernel_tables(dst: &mut PageTable<L0>) {
 /// # Safety
 ///
 /// Unsafe, must only be called by BSP during its early init, must already be in "higher-half"
-pub unsafe fn init_fixed_tables() {
-    let mut tables = KERNEL_TABLES.lock();
-    // TODO this could be built in compile-time too?
-    let early_mapping_l3_phys = addr_of!(EARLY_MAPPING_L3) as usize - KERNEL_VIRT_OFFSET;
-    let device_mapping_l2_phys = addr_of!(DEVICE_MAPPING_L2) as usize - KERNEL_VIRT_OFFSET;
-    let ram_mapping_l1_phys = addr_of!(RAM_MAPPING_L1) as usize - KERNEL_VIRT_OFFSET;
-
-    for i in 0..DEVICE_MAPPING_L3_COUNT {
-        let device_mapping_l3_phys = PhysicalAddress::from_usize(
-            &DEVICE_MAPPING_L3S[i] as *const _ as usize - KERNEL_VIRT_OFFSET,
-        );
-        DEVICE_MAPPING_L2[i] = PageEntry::table(device_mapping_l3_phys, PageAttributes::WRITABLE);
+#[inline(never)]
+pub unsafe fn init_fixed_tables(have_1gib_pages: bool, bsp: bool) {
+    fixed::setup(have_1gib_pages);
+    if bsp {
+        fixed::load();
     }
-
-    assert_eq!(tables.kernel_l2.data[EARLY_MAPPING_L2I], 0);
-    tables.kernel_l2.data[EARLY_MAPPING_L2I] = (early_mapping_l3_phys as u64)
-        | (PageAttributes::WRITABLE | PageAttributes::PRESENT).bits();
-
-    assert_eq!(tables.kernel_l1.data[DEVICE_MAPPING_L1I], 0);
-    tables.kernel_l1.data[DEVICE_MAPPING_L1I] = (device_mapping_l2_phys as u64)
-        | (PageAttributes::WRITABLE | PageAttributes::PRESENT).bits();
-
-    assert_eq!(tables.l0.data[RAM_MAPPING_L0I], 0);
-    tables.l0.data[RAM_MAPPING_L0I] =
-        (ram_mapping_l1_phys as u64) | (PageAttributes::WRITABLE | PageAttributes::PRESENT).bits();
-
-    // TODO ENABLE EFER.NXE
-    let cr3 = (&raw const tables.l0).addr() - KERNEL_VIRT_OFFSET;
-    CR3.set_address(cr3);
 }
 
 /// # Safety

kernel/arch/x86_64/src/mem/process.rs

@@ -4,7 +4,7 @@ use core::marker::PhantomData;
 use libk_mm_interface::{
     address::{AsPhysicalAddress, PhysicalAddress},
     pointer::PhysicalRefMut,
-    process::ProcessAddressSpaceManager,
+    process::{PageAttributeUpdate, ProcessAddressSpaceManager},
     table::{
         EntryLevel, EntryLevelDrop, EntryLevelExt, MapAttributes, NextPageTable, TableAllocator,
     },
@@ -61,7 +61,15 @@ impl<TA: TableAllocator> ProcessAddressSpaceManager<TA> for ProcessAddressSpaceI
         self.write_l3_entry(address, PageEntry::page(physical, flags.into()), false)
     }
 
-    unsafe fn unmap_page(&mut self, address: usize) -> Result<PhysicalAddress, Error> {
+    unsafe fn update_page_attributes(
+        &mut self,
+        address: usize,
+        update: &PageAttributeUpdate,
+    ) -> Result<(), Error> {
+        self.update_l3_entry(address, |entry| entry.update(update))
+    }
+
+    unsafe fn unmap_page(&mut self, address: usize) -> Result<(PhysicalAddress, bool), Error> {
         self.pop_l3_entry(address)
     }
 
@@ -71,9 +79,9 @@ impl<TA: TableAllocator> ProcessAddressSpaceManager<TA> for ProcessAddressSpaceI
             .ok_or(Error::InvalidMemoryOperation)
     }
 
-    fn as_address_with_asid(&self) -> u64 {
+    fn as_address_with_asid(&self) -> (u64, u64) {
         // TODO x86-64 PCID/ASID?
-        unsafe { self.l0.as_physical_address().into_u64() }
+        (unsafe { self.l0.as_physical_address().into_u64() }, 0)
     }
 
     unsafe fn clear(&mut self) {
@@ -111,7 +119,33 @@ impl<TA: TableAllocator> ProcessAddressSpaceImpl<TA> {
         Ok(())
     }
 
-    fn pop_l3_entry(&mut self, virt: usize) -> Result<PhysicalAddress, Error> {
+    fn update_l3_entry<F: FnOnce(&mut PageEntry<L3>) -> Result<(), Error>>(
+        &mut self,
+        virt: usize,
+        mapper: F,
+    ) -> Result<(), Error> {
+        let l0i = virt.page_index::<L0>();
+        let l1i = virt.page_index::<L1>();
+        let l2i = virt.page_index::<L2>();
+        let l3i = virt.page_index::<L3>();
+
+        let mut l1 = self.l0.get_mut(l0i).ok_or(Error::DoesNotExist)?;
+        let mut l2 = l1.get_mut(l1i).ok_or(Error::DoesNotExist)?;
+        let mut l3 = l2.get_mut(l2i).ok_or(Error::DoesNotExist)?;
+
+        let entry = &mut l3[l3i];
+        if !entry.is_present() {
+            return Err(Error::DoesNotExist);
+        }
+        mapper(entry)?;
+        unsafe {
+            flush_tlb_entry(virt);
+        }
+
+        Ok(())
+    }
+
+    fn pop_l3_entry(&mut self, virt: usize) -> Result<(PhysicalAddress, bool), Error> {
         let l0i = virt.page_index::<L0>();
         let l1i = virt.page_index::<L1>();
         let l2i = virt.page_index::<L2>();
@@ -123,13 +157,14 @@ impl<TA: TableAllocator> ProcessAddressSpaceImpl<TA> {
         let mut l3 = l2.get_mut(l2i).ok_or(Error::DoesNotExist)?;
 
         let page = l3[l3i].as_page().ok_or(Error::DoesNotExist)?;
+        let dirty = l3[l3i].is_dirty();
 
         l3[l3i] = PageEntry::INVALID;
         unsafe {
             flush_tlb_entry(virt);
         }
 
-        Ok(page)
+        Ok((page, dirty))
     }
 
     fn read_l3_entry(&self, virt: usize) -> Option<(PhysicalAddress, MapAttributes)> {

kernel/arch/x86_64/src/mem/table.rs

@@ -8,6 +8,7 @@ use bitflags::bitflags;
 use libk_mm_interface::{
     address::{AsPhysicalAddress, PhysicalAddress},
     pointer::{PhysicalRef, PhysicalRefMut},
+    process::PageAttributeUpdate,
     table::{
         EntryLevel, EntryLevelDrop, MapAttributes, NextPageTable, NonTerminalEntryLevel,
         TableAllocator,
@@ -31,6 +32,8 @@ bitflags! {
         /// For tables, allows user access to further translation levels, for pages/blocks, allows
         /// user access to the region covered by the entry
         const USER = 1 << 2;
+        /// If set, the page has been written to
+        const DIRTY = 1 << 6;
     }
 }
 
@@ -98,11 +101,15 @@ impl PageEntry<L3> {
     /// not
     pub fn as_page(self) -> Option<PhysicalAddress> {
         if self.0 & PageAttributes::PRESENT.bits() != 0 {
-            Some(PhysicalAddress::from_u64(self.0 & !0xFFF))
+            Some(PhysicalAddress::from_u64(self.0 & !Self::ATTR_MASK))
         } else {
             None
         }
     }
+
+    pub fn is_dirty(&self) -> bool {
+        self.0 & PageAttributes::DIRTY.bits() != 0
+    }
 }
 
 impl PageEntry<L2> {
@@ -145,7 +152,7 @@ impl<L: NonTerminalEntryLevel> PageEntry<L> {
         if self.0 & PageAttributes::PRESENT.bits() != 0
             && self.0 & PageAttributes::BLOCK.bits() == 0
         {
-            Some(PhysicalAddress::from_u64(self.0 & !0xFFF))
+            Some(PhysicalAddress::from_u64(self.0 & !Self::ATTR_MASK))
         } else {
             None
         }
@@ -158,6 +165,8 @@ impl<L: NonTerminalEntryLevel> PageEntry<L> {
 }
 
 impl<L: EntryLevel> PageEntry<L> {
+    const ATTR_MASK: u64 = 0xFFF | (1 << 63);
+
     /// An entry that is not mapped
     pub const INVALID: Self = Self(0, PhantomData);
 
@@ -179,6 +188,21 @@ impl<L: EntryLevel> PageEntry<L> {
     pub fn is_present(&self) -> bool {
         self.0 & PageAttributes::PRESENT.bits() != 0
     }
+
+    pub fn update(&mut self, update: &PageAttributeUpdate) -> Result<(), Error> {
+        let mut attrs = PageAttributes::from_bits_retain(self.0);
+        if let Some(write) = update.user_write {
+            if write {
+                attrs |= PageAttributes::WRITABLE;
+            } else {
+                attrs &= !PageAttributes::WRITABLE;
+            }
+        }
+        // Dirty is ignored, it's hardware-managed
+        self.0 &= !Self::ATTR_MASK;
+        self.0 |= attrs.bits() & Self::ATTR_MASK;
+        Ok(())
+    }
 }
 
 impl<L: EntryLevel> PageTable<L> {

kernel/build.rs

@@ -15,7 +15,7 @@ fn build_x86_64() {
     const DEFAULT_8086_AS: &str = "nasm";
     const AP_BOOTSTRAP_S: &str = "src/arch/x86_64/boot/ap_boot.S";
 
-    println!("cargo:rerun-if-changed={}", AP_BOOTSTRAP_S);
+    println!("cargo:rerun-if-changed={AP_BOOTSTRAP_S}");
 
     let out_dir = env::var("OUT_DIR").unwrap();
     let assembler = env::var("AS8086").unwrap_or(DEFAULT_8086_AS.to_owned());
@@ -35,7 +35,7 @@ fn build_x86_64() {
 
     if !output.status.success() {
         io::stderr().write_all(&output.stderr).ok();
-        panic!("{}: could not assemble {}", assembler, AP_BOOTSTRAP_S);
+        panic!("{assembler}: could not assemble {AP_BOOTSTRAP_S}");
     }
 }
 
@@ -92,6 +92,7 @@ fn main() {
         "x86" => (),
         "x86_64" => build_x86_64(),
         "aarch64" => (),
-        _ => panic!("Unknown target arch: {:?}", arch),
+        "riscv64" => (),
+        _ => panic!("Unknown target arch: {arch:?}"),
     }
 }

kernel/driver/acpi/Cargo.toml

@@ -0,0 +1,18 @@
+[package]
+name = "ygg_driver_acpi"
+version = "0.1.0"
+edition = "2024"
+
+[dependencies]
+libk-util.workspace = true
+libk-mm.workspace = true
+libk.workspace = true
+device-api.workspace = true
+kernel-arch-x86.path = "../../arch/x86"
+
+acpi.workspace = true
+rsdp.workspace = true
+aml.workspace = true
+acpi-system.workspace = true
+
+log.workspace = true

kernel/driver/acpi/src/aml_handler.rs

@@ -0,0 +1,131 @@
+use core::time::Duration;
+
+use crate::AcpiHandlerImpl;
+
+impl aml::Handler for AcpiHandlerImpl {
+    fn read_io_u8(&self, port: u16) -> u8 {
+        <Self as acpi_system::Handler>::io_read_u8(port)
+    }
+
+    fn read_io_u16(&self, port: u16) -> u16 {
+        <Self as acpi_system::Handler>::io_read_u16(port)
+    }
+
+    fn read_io_u32(&self, port: u16) -> u32 {
+        <Self as acpi_system::Handler>::io_read_u32(port)
+    }
+
+    fn write_io_u8(&self, port: u16, value: u8) {
+        <Self as acpi_system::Handler>::io_write_u8(port, value)
+    }
+
+    fn write_io_u16(&self, port: u16, value: u16) {
+        <Self as acpi_system::Handler>::io_write_u16(port, value)
+    }
+
+    fn write_io_u32(&self, port: u16, value: u32) {
+        <Self as acpi_system::Handler>::io_write_u32(port, value)
+    }
+
+    fn read_u8(&self, address: usize) -> u8 {
+        <Self as acpi_system::Handler>::mem_read_u8(address as u64)
+    }
+
+    fn read_u16(&self, address: usize) -> u16 {
+        <Self as acpi_system::Handler>::mem_read_u16(address as u64)
+    }
+
+    fn read_u32(&self, address: usize) -> u32 {
+        <Self as acpi_system::Handler>::mem_read_u32(address as u64)
+    }
+
+    fn read_u64(&self, address: usize) -> u64 {
+        <Self as acpi_system::Handler>::mem_read_u64(address as u64)
+    }
+
+    fn write_u8(&self, address: usize, value: u8) {
+        <Self as acpi_system::Handler>::mem_write_u8(address as u64, value)
+    }
+
+    fn write_u16(&self, address: usize, value: u16) {
+        <Self as acpi_system::Handler>::mem_write_u16(address as u64, value)
+    }
+
+    fn write_u32(&self, address: usize, value: u32) {
+        <Self as acpi_system::Handler>::mem_write_u32(address as u64, value)
+    }
+
+    fn write_u64(&self, address: usize, value: u64) {
+        <Self as acpi_system::Handler>::mem_write_u64(address as u64, value)
+    }
+
+    fn read_pci_u8(&self, _segment: u16, _bus: u8, _device: u8, _function: u8, _offset: u16) -> u8 {
+        0xFF
+    }
+
+    fn read_pci_u16(
+        &self,
+        _segment: u16,
+        _bus: u8,
+        _device: u8,
+        _function: u8,
+        _offset: u16,
+    ) -> u16 {
+        0xFFFF
+    }
+
+    fn read_pci_u32(
+        &self,
+        _segment: u16,
+        _bus: u8,
+        _device: u8,
+        _function: u8,
+        _offset: u16,
+    ) -> u32 {
+        0xFFFFFFFF
+    }
+
+    fn write_pci_u8(
+        &self,
+        _segment: u16,
+        _bus: u8,
+        _device: u8,
+        _function: u8,
+        _offset: u16,
+        _value: u8,
+    ) {
+    }
+
+    fn write_pci_u16(
+        &self,
+        _segment: u16,
+        _bus: u8,
+        _device: u8,
+        _function: u8,
+        _offset: u16,
+        _value: u16,
+    ) {
+    }
+
+    fn write_pci_u32(
+        &self,
+        _segment: u16,
+        _bus: u8,
+        _device: u8,
+        _function: u8,
+        _offset: u16,
+        _value: u32,
+    ) {
+    }
+
+    fn read_ec_u8(&self, _address: u64) -> u8 {
+        0x00
+    }
+
+    fn write_ec_u8(&self, _address: u64, _value: u8) {}
+
+    fn sleep(&self, _duration: Duration) {
+        todo!()
+        // util::polling_sleep(duration).unwrap();
+    }
+}

kernel/driver/acpi/src/handler.rs

@@ -0,0 +1,171 @@
+use core::{ptr::NonNull, time::Duration};
+
+use acpi::PhysicalMapping;
+use acpi_system::AcpiSystemError;
+use alloc::sync::Arc;
+use device_api::{
+    device::Device,
+    interrupt::{InterruptHandler, Irq, IrqVector},
+};
+use kernel_arch_x86::{intrinsics, ISA_IRQ_OFFSET};
+use libk::device::external_interrupt_controller;
+use libk_mm::{
+    address::{PhysicalAddress, Virtualize},
+    pointer::PhysicalRef,
+};
+
+use crate::{
+    mem::{read_memory, write_memory},
+    ACPI_SYSTEM,
+};
+
+#[derive(Clone, Copy)]
+#[doc(hidden)]
+pub struct AcpiHandlerImpl;
+
+struct SciHandler;
+
+impl acpi_system::Handler for AcpiHandlerImpl {
+    type MappedSlice = PhysicalRef<'static, [u8]>;
+
+    unsafe fn map_slice(address: u64, length: u64) -> Self::MappedSlice {
+        unsafe {
+            PhysicalRef::map_slice(
+                PhysicalAddress::from_u64(address),
+                length.try_into().unwrap(),
+            )
+        }
+    }
+
+    fn io_read_u8(port: u16) -> u8 {
+        let value = unsafe { intrinsics::inb(port) };
+        log::trace!("io_read_u8 {:#x} <- {:#x}", port, value);
+        value
+    }
+
+    fn io_read_u16(port: u16) -> u16 {
+        let value = unsafe { intrinsics::inw(port) };
+        log::trace!("io_read_u16 {:#x} <- {:#x}", port, value);
+        value
+    }
+
+    fn io_read_u32(port: u16) -> u32 {
+        let value = unsafe { intrinsics::inl(port) };
+        log::trace!("io_read_u32 {:#x} <- {:#x}", port, value);
+        value
+    }
+
+    fn io_write_u8(port: u16, value: u8) {
+        log::trace!("io_write_u8 {:#x}, {:#x}", port, value);
+        unsafe { intrinsics::outb(port, value) }
+    }
+
+    fn io_write_u16(port: u16, value: u16) {
+        log::trace!("io_write_u16 {:#x}, {:#x}", port, value);
+        unsafe { intrinsics::outw(port, value) }
+    }
+
+    fn io_write_u32(port: u16, value: u32) {
+        log::trace!("io_write_u32 {:#x}, {:#x}", port, value);
+        unsafe { intrinsics::outl(port, value) }
+    }
+
+    fn mem_read_u8(address: u64) -> u8 {
+        let value = unsafe { read_memory(PhysicalAddress::from_u64(address)) };
+        log::trace!("mem_read_u8 {:#x} -> {:#x}", address, value);
+        value
+    }
+
+    fn mem_read_u16(address: u64) -> u16 {
+        let value = unsafe { read_memory(PhysicalAddress::from_u64(address)) };
+        log::trace!("mem_read_u16 {:#x} -> {:#x}", address, value);
+        value
+    }
+
+    fn mem_read_u32(address: u64) -> u32 {
+        let value = unsafe { read_memory(PhysicalAddress::from_u64(address)) };
+        log::trace!("mem_read_u32 {:#x} -> {:#x}", address, value);
+        value
+    }
+
+    fn mem_read_u64(address: u64) -> u64 {
+        let value = unsafe { read_memory(PhysicalAddress::from_u64(address)) };
+        log::trace!("mem_read_u64 {:#x} -> {:#x}", address, value);
+        value
+    }
+
+    fn mem_write_u8(address: u64, value: u8) {
+        log::trace!("mem_write_u8 {:#x}, {:#x}", address, value);
+        unsafe { write_memory(PhysicalAddress::from_u64(address), value) }
+    }
+
+    fn mem_write_u16(address: u64, value: u16) {
+        log::trace!("mem_write_u16 {:#x}, {:#x}", address, value);
+        unsafe { write_memory(PhysicalAddress::from_u64(address), value) }
+    }
+
+    fn mem_write_u32(address: u64, value: u32) {
+        log::trace!("mem_write_u32 {:#x}, {:#x}", address, value);
+        unsafe { write_memory(PhysicalAddress::from_u64(address), value) }
+    }
+
+    fn mem_write_u64(address: u64, value: u64) {
+        log::trace!("mem_write_u64 {:#x}, {:#x}", address, value);
+        unsafe { write_memory(PhysicalAddress::from_u64(address), value) }
+    }
+
+    fn install_interrupt_handler(irq: u32) -> Result<(), AcpiSystemError> {
+        log::info!("Installing ACPI SCI handler at IRQ #{}", irq);
+
+        let intc = external_interrupt_controller().expect("No external intc");
+        let handler = Arc::new(SciHandler);
+        let irq = Irq::External(irq + ISA_IRQ_OFFSET);
+
+        intc.register_irq(irq, Default::default(), handler).unwrap();
+        intc.enable_irq(irq).unwrap();
+
+        Ok(())
+    }
+
+    fn stall(_duration: Duration) {
+        // TODO polling_sleep is not yet implemented properly
+        todo!()
+        // util::polling_sleep(duration).ok();
+    }
+}
+
+impl rsdp::handler::AcpiHandler for AcpiHandlerImpl {
+    unsafe fn map_physical_region<T>(
+        &self,
+        physical_address: usize,
+        size: usize,
+    ) -> PhysicalMapping<Self, T> {
+        unsafe {
+            PhysicalMapping::new(
+                physical_address,
+                NonNull::new_unchecked(
+                    PhysicalAddress::from_usize(physical_address).virtualize() as *mut T
+                ),
+                size,
+                size,
+                *self,
+            )
+        }
+    }
+
+    fn unmap_physical_region<T>(_region: &acpi::PhysicalMapping<Self, T>) {}
+}
+
+impl InterruptHandler for SciHandler {
+    fn handle_irq(self: Arc<Self>, _vector: IrqVector) -> bool {
+        log::trace!("ACPI SCI received");
+        ACPI_SYSTEM.get().lock().handle_sci();
+        true
+    }
+}
+
+impl Device for SciHandler {
+    fn display_name(&self) -> &str {
+        "ACPI SCI handler"
+    }
+}

kernel/driver/acpi/src/lib.rs

@@ -0,0 +1,89 @@
+#![feature(allocator_api)]
+#![no_std]
+
+use acpi::AcpiTables;
+use acpi_system::{AcpiInterruptMethod, AcpiSystem};
+use alloc::boxed::Box;
+use libk::error::Error;
+use libk_util::{sync::IrqSafeSpinlock, OneTimeInit};
+
+extern crate alloc;
+
+pub mod mem;
+pub use mem::AcpiAllocator;
+pub mod handler;
+pub use handler::AcpiHandlerImpl;
+pub mod aml_handler;
+
+pub use acpi_system::{
+    EventAction, FixedEvent, InterruptPolarity, InterruptTrigger, IrqDescriptor, PciPin,
+};
+
+static ACPI_SYSTEM: OneTimeInit<IrqSafeSpinlock<AcpiSystem<AcpiHandlerImpl>>> = OneTimeInit::new();
+
+pub fn add_event_handler<F: Fn(&AcpiSystem<AcpiHandlerImpl>) -> EventAction + 'static>(
+    event: &FixedEvent,
+    handler: F,
+) -> Result<(), Error> {
+    ACPI_SYSTEM
+        .get()
+        .lock()
+        .enable_fixed_event(event, Box::new(handler))
+        .map_err(|_| Error::InvalidArgument)
+}
+
+pub fn get_pci_route(
+    aml_path: &str,
+    device: u16,
+    function: u16,
+    pin: PciPin,
+) -> Option<IrqDescriptor> {
+    ACPI_SYSTEM
+        .get()
+        .lock()
+        .pci_route(aml_path, device, function, pin)
+        .ok()
+}
+
+/// Initializes ACPI management
+pub fn switch_to_acpi(tables: &'static AcpiTables<AcpiHandlerImpl>) -> Result<(), Error> {
+    // NOTE mostly broken for real HW
+    let mut system = AcpiSystem::new(tables, Box::new(AcpiHandlerImpl)).unwrap();
+
+    system.initialize(AcpiInterruptMethod::Apic).unwrap();
+
+    // system
+    //     .enable_fixed_event(
+    //         &FixedEvent::POWER_BUTTON,
+    //         Box::new(|_| {
+    //             log::info!("Power button was pressed");
+
+    //             // TODO the correct way would be to
+    //             // 1. Nicely ask all the processes to quit
+    //             // 2. Wait for some time
+    //             // 3. Kill the remaining ones
+    //             // 4. Halt other cores
+    //             // 5. Sync filesystem
+    //             // 6. Do something with the devices
+    //             // 7. Actually enter the S5 state
+
+    //             unsafe {
+    //                 PLATFORM
+    //                     .send_ipi(IpiDeliveryTarget::OtherCpus, IpiMessage::Shutdown)
+    //                     .unwrap();
+    //             }
+
+    //             SHUTDOWN_FENCE.signal();
+    //             SHUTDOWN_FENCE.wait_all(CPU_COUNT.load(Ordering::Acquire));
+
+    //             log::info!("CPUs are parked, can shutdown now");
+
+    //             EventAction::EnterSleepState(AcpiSleepState::S5)
+    //         }),
+    //     )
+    //     .unwrap();
+
+    ACPI_SYSTEM.init(IrqSafeSpinlock::new(system));
+
+    Ok(())
+}

kernel/driver/acpi/src/mem.rs

@@ -0,0 +1,70 @@
+//! ACPI memory IO and management functions
+
+use core::{
+    alloc::{AllocError, Allocator, GlobalAlloc, Layout},
+    ptr::NonNull,
+};
+
+use libk_mm::{address::PhysicalAddress, device::DeviceMemoryMapping, heap::GLOBAL_HEAP};
+
+#[derive(Clone, Copy)]
+#[doc(hidden)]
+pub struct AcpiAllocator;
+
+unsafe impl Allocator for AcpiAllocator {
+    fn allocate(&self, layout: Layout) -> Result<NonNull<[u8]>, AllocError> {
+        let ptr = unsafe { GLOBAL_HEAP.alloc(layout) };
+        log::trace!("ACPI alloc: {:?} -> {:p}", layout, ptr);
+
+        if ptr.is_null() {
+            Err(AllocError)
+        } else {
+            unsafe {
+                Ok(NonNull::slice_from_raw_parts(
+                    NonNull::new_unchecked(ptr),
+                    layout.size(),
+                ))
+            }
+        }
+    }
+
+    unsafe fn deallocate(&self, ptr: NonNull<u8>, layout: Layout) {
+        log::trace!("ACPI dealloc: {:?}, {:?}", ptr, layout);
+        unsafe { GLOBAL_HEAP.dealloc(ptr.as_ptr(), layout) };
+    }
+}
+
+// TODO don't map memory as device if not necessary
+/// # Safety
+///
+/// Allows direct reads from physical memory, unsafe
+pub unsafe fn read_memory<T>(address: PhysicalAddress) -> T {
+    let io =
+        unsafe { DeviceMemoryMapping::map(address, size_of::<T>(), Default::default()).unwrap() };
+    let address = io.address();
+
+    unsafe {
+        if address % align_of::<T>() == 0 {
+            (address as *const T).read_volatile()
+        } else {
+            (address as *const T).read_unaligned()
+        }
+    }
+}
+
+/// # Safety
+///
+/// Allows direct writes to physical memory, unsafe
+pub unsafe fn write_memory<T>(address: PhysicalAddress, value: T) {
+    let io =
+        unsafe { DeviceMemoryMapping::map(address, size_of::<T>(), Default::default()).unwrap() };
+    let address = io.address();
+
+    unsafe {
+        if address % align_of::<T>() == 0 {
+            (address as *mut T).write_volatile(value)
+        } else {
+            (address as *mut T).write_unaligned(value)
+        }
+    }
+}

kernel/driver/block/ahci/src/command.rs

@@ -1,12 +1,10 @@
 use core::mem::{size_of, MaybeUninit};
 
-use libk_mm::{
-    address::{AsPhysicalAddress, PhysicalAddress},
-    PageBox, PageSlice,
-};
+use device_api::dma::DmaAllocator;
+use libk::dma::{BusAddress, DmaBuffer, DmaSliceMut};
 use tock_registers::register_structs;
 
-use crate::{data::AtaString, error::AhciError, MAX_PRD_SIZE, SECTOR_SIZE};
+use crate::{data::AtaString, error::AhciError, MAX_PRD_SIZE};
 
 #[derive(Clone, Copy, PartialEq, Eq, Debug)]
 #[repr(u8)]
@@ -22,7 +20,7 @@ pub trait AtaCommand {
 
     fn lba(&self) -> u64;
     fn sector_count(&self) -> usize;
-    fn buffer(&self) -> Option<(PhysicalAddress, usize)>;
+    fn buffer(&self) -> Option<(BusAddress, usize)>;
     unsafe fn into_response(self) -> Self::Response;
 
     fn prd_count(&self) -> usize {
@@ -64,44 +62,41 @@ register_structs! {
 }
 
 pub struct AtaIdentify {
-    buffer: PageBox<MaybeUninit<AtaIdentifyResponse>>,
+    buffer: DmaBuffer<MaybeUninit<AtaIdentifyResponse>>,
 }
 
 pub struct AtaReadDmaEx {
     lba: u64,
     sector_count: usize,
-    buffer_base: PhysicalAddress,
+    buffer_base: BusAddress,
     buffer_size: usize,
 }
 
 impl AtaIdentify {
-    pub fn create() -> Result<Self, AhciError> {
-        PageBox::new_uninit()
+    pub fn create(dma: &dyn DmaAllocator) -> Result<Self, AhciError> {
+        DmaBuffer::new_uninit(dma)
             .map(Self::with_data)
             .map_err(AhciError::MemoryError)
     }
 
-    pub fn with_data(buffer: PageBox<MaybeUninit<AtaIdentifyResponse>>) -> Self {
+    pub fn with_data(buffer: DmaBuffer<MaybeUninit<AtaIdentifyResponse>>) -> Self {
         Self { buffer }
     }
 }
 
 impl AtaReadDmaEx {
-    pub fn new(lba: u64, sector_count: usize, buffer: &PageSlice<MaybeUninit<u8>>) -> Self {
-        assert_eq!(buffer.len() % SECTOR_SIZE, 0);
-        assert_ne!(buffer.len(), 0);
-
+    pub fn new(lba: u64, sector_count: usize, buffer: DmaSliceMut<MaybeUninit<u8>>) -> Self {
         Self {
             lba,
             sector_count,
-            buffer_base: unsafe { buffer.as_physical_address() },
+            buffer_base: buffer.bus_address(),
             buffer_size: buffer.len(),
         }
     }
 }
 
 impl AtaCommand for AtaIdentify {
-    type Response = PageBox<AtaIdentifyResponse>;
+    type Response = DmaBuffer<AtaIdentifyResponse>;
 
     const COMMAND_ID: AtaCommandId = AtaCommandId::Identify;
 
@@ -113,14 +108,14 @@ impl AtaCommand for AtaIdentify {
         0
     }
 
-    fn buffer(&self) -> Option<(PhysicalAddress, usize)> {
-        let base = unsafe { self.buffer.as_physical_address() };
+    fn buffer(&self) -> Option<(BusAddress, usize)> {
+        let base = self.buffer.bus_address();
         let size = size_of::<AtaIdentifyResponse>();
         Some((base, size))
     }
 
     unsafe fn into_response(self) -> Self::Response {
-        self.buffer.assume_init()
+        DmaBuffer::assume_init(self.buffer)
     }
 }
 
@@ -137,7 +132,7 @@ impl AtaCommand for AtaReadDmaEx {
         self.sector_count
     }
 
-    fn buffer(&self) -> Option<(PhysicalAddress, usize)> {
+    fn buffer(&self) -> Option<(BusAddress, usize)> {
         Some((self.buffer_base, self.buffer_size))
     }
 

kernel/driver/block/ahci/src/data.rs

@@ -2,7 +2,7 @@ use core::mem::size_of;
 
 use alloc::string::String;
 use bytemuck::{Pod, Zeroable};
-use libk_mm::address::PhysicalAddress;
+use libk::dma::BusAddress;
 use libk_util::{ConstAssert, IsTrue};
 use static_assertions::const_assert_eq;
 
@@ -166,7 +166,7 @@ impl CommandTable {
             }
             assert_eq!(prd, command.prd_count());
 
-            self.prdt[prd..].fill_with(|| PhysicalRegionDescriptor::zeroed());
+            self.prdt[prd..].fill_with(PhysicalRegionDescriptor::zeroed);
         }
 
         Ok(())
@@ -174,7 +174,7 @@ impl CommandTable {
 }
 
 impl CommandListEntry {
-    pub fn new(command_table_entry: PhysicalAddress, prd_count: usize) -> Result<Self, AhciError> {
+    pub fn new(command_table_entry: BusAddress, prd_count: usize) -> Result<Self, AhciError> {
         if prd_count > 0xFFFF {
             todo!()
         }
@@ -183,7 +183,7 @@ impl CommandListEntry {
             attr: (size_of::<RegisterHostToDeviceFis>() / size_of::<u32>()) as _,
             prdtl: prd_count as _,
             prdbc: 0,
-            ctba: command_table_entry.into(),
+            ctba: command_table_entry.into_u64(),
             _0: [0; 4],
         })
     }
@@ -201,18 +201,14 @@ unsafe impl Zeroable for CommandTable {
 }
 
 impl PhysicalRegionDescriptor {
-    pub fn new(
-        address: PhysicalAddress,
-        byte_count: usize,
-        is_last: bool,
-    ) -> Result<Self, AhciError> {
+    pub fn new(address: BusAddress, byte_count: usize, is_last: bool) -> Result<Self, AhciError> {
         if byte_count > MAX_PRD_SIZE {
             return Err(AhciError::RegionTooLarge);
         }
 
         let dbc_mask = (is_last as u32) << 31;
         Ok(Self {
-            buffer_address: address.into(),
+            buffer_address: address.into_u64(),
             _0: 0,
             dbc: ((byte_count as u32 - 1) << 1) | 1 | dbc_mask,
         })

kernel/driver/block/ahci/src/lib.rs

@@ -4,29 +4,27 @@
 
 extern crate alloc;
 
-use alloc::{boxed::Box, format, vec, vec::Vec};
+use alloc::{format, sync::Arc, vec::Vec};
 use bytemuck::Zeroable;
 use data::ReceivedFis;
 use device_api::{
-    interrupt::{InterruptAffinity, InterruptHandler},
-    Device,
+    device::{Device, DeviceInitContext},
+    dma::DmaAllocator,
+    interrupt::{InterruptAffinity, InterruptHandler, IrqVector},
 };
 use error::AhciError;
-use kernel_fs::devfs;
-use libk::{
-    task::runtime,
-    vfs::block::{probe_partitions, NgBlockDeviceWrapper},
-};
-use libk_mm::{address::AsPhysicalAddress, device::DeviceMemoryIo, PageBox};
+use libk::{device::manager::probe_partitions, dma::DmaBuffer, fs::devfs, task::runtime};
+use libk_mm::device::DeviceMemoryIo;
 use libk_util::{sync::IrqSafeSpinlock, OneTimeInit};
 use port::AhciPort;
 use regs::{PortRegs, Regs};
 use tock_registers::interfaces::{ReadWriteable, Readable, Writeable};
 use ygg_driver_pci::{
     device::{PciDeviceInfo, PreferredInterruptMode},
+    macros::pci_driver,
     PciCommandRegister, PciConfigurationSpace,
 };
-use yggdrasil_abi::error::Error;
+use yggdrasil_abi::{error::Error, io::FileMode};
 
 use crate::regs::{Version, CAP, GHC, SSTS};
 
@@ -43,8 +41,9 @@ const MAX_DRIVES: usize = (b'z' - b'a') as usize;
 
 pub struct AhciController {
     regs: IrqSafeSpinlock<DeviceMemoryIo<'static, Regs>>,
-    ports: OneTimeInit<Vec<&'static AhciPort>>,
-    received_fis_buffers: OneTimeInit<[Option<PageBox<ReceivedFis>>; 16]>,
+    dma: Arc<dyn DmaAllocator>,
+    ports: OneTimeInit<Vec<Arc<AhciPort>>>,
+    received_fis_buffers: OneTimeInit<[Option<DmaBuffer<ReceivedFis>>; 16]>,
 
     version: Version,
     max_port_count: usize,
@@ -53,7 +52,7 @@ pub struct AhciController {
 }
 
 impl AhciController {
-    async fn late_init(&'static self) -> Result<(), AhciError> {
+    async fn late_init(self: Arc<Self>) -> Result<(), AhciError> {
         log::info!("Initializing AHCI SATA Controller {:?}", self.version);
 
         let regs = self.regs.lock();
@@ -70,7 +69,7 @@ impl AhciController {
 
         let pi = regs.PI.get();
 
-        let mut ports = vec![];
+        let mut ports = Vec::new();
 
         drop(regs);
 
@@ -84,8 +83,9 @@ impl AhciController {
             let regs = self.regs.lock();
             let port = &regs.PORTS[i];
 
-            let buffer = PageBox::new(ReceivedFis::zeroed()).map_err(AhciError::MemoryError)?;
-            port.set_received_fis_address_64(unsafe { buffer.as_physical_address() });
+            let buffer = DmaBuffer::new(&*self.dma, ReceivedFis::zeroed())
+                .map_err(AhciError::MemoryError)?;
+            port.set_received_fis_address_64(buffer.bus_address());
             *fis_buffer_slot = Some(buffer);
         }
 
@@ -117,7 +117,7 @@ impl AhciController {
 
             drop(regs);
 
-            let port = match AhciPort::create(port, self, i) {
+            let port = match AhciPort::create(port, self.clone(), i) {
                 Ok(port) => port,
                 Err(error) => {
                     log::warn!("Port {} init error: {:?}", i, error);
@@ -134,13 +134,13 @@ impl AhciController {
         self.regs.lock().GHC.modify(GHC::IE::SET);
 
         // Setup the detected ports
-        for (i, &port) in ports.iter().enumerate() {
+        for (i, port) in ports.iter().enumerate() {
             log::info!("Init port {}", i);
-            port.init().await?;
+            port.init_inner().await?;
         }
 
         // Dump info about the drives
-        for (i, &port) in ports.iter().enumerate() {
+        for (i, port) in ports.iter().enumerate() {
             let info = port.info().unwrap();
             log::info!(
                 "Port {}: model={:?}, serial={:?}, lba_count={}",
@@ -151,25 +151,8 @@ impl AhciController {
             );
         }
 
-        {
-            let mut lock = SATA_DRIVES.lock();
-            for &port in ports.iter() {
-                let n = lock.len();
-                if n >= MAX_DRIVES {
-                    todo!("Too many drives, ran out of letters");
-                }
-                let n = n as u8;
-                lock.push(port);
-
-                let name = format!("sd{}", (n + b'a') as char);
-
-                let blk = NgBlockDeviceWrapper::new(port);
-                devfs::add_named_block_device(blk, name.clone()).ok();
-                probe_partitions(blk, move |index, partition| {
-                    devfs::add_block_device_partition(name.clone(), index, partition)
-                })
-                .ok();
-            }
+        for port in ports.iter() {
+            register_sata_drive(port.clone(), true);
         }
 
         log::debug!("All ports initialized");
@@ -179,7 +162,7 @@ impl AhciController {
 }
 
 impl InterruptHandler for AhciController {
-    fn handle_irq(&self, _vector: Option<usize>) -> bool {
+    fn handle_irq(self: Arc<Self>, _vector: IrqVector) -> bool {
         let regs = self.regs.lock();
 
         let is = regs.IS.get();
@@ -188,7 +171,7 @@ impl InterruptHandler for AhciController {
                 // Clear global interrupt status
                 regs.IS.set(u32::MAX);
 
-                for &port in ports {
+                for port in ports {
                     if is & (1 << port.index) != 0 {
                         port.handle_pending_interrupts();
                     }
@@ -201,58 +184,106 @@ impl InterruptHandler for AhciController {
 }
 
 impl Device for AhciController {
-    unsafe fn init(&'static self) -> Result<(), Error> {
+    unsafe fn init(self: Arc<Self>, _cx: DeviceInitContext) -> Result<(), Error> {
         // Do the init in background
         runtime::spawn(self.late_init())?;
         Ok(())
     }
 
-    fn display_name(&self) -> &'static str {
-        "AHCI SATA Controller"
+    fn display_name(&self) -> &str {
+        "AHCI Controller"
     }
 }
 
-static SATA_DRIVES: IrqSafeSpinlock<Vec<&'static AhciPort>> = IrqSafeSpinlock::new(Vec::new());
+pub fn register_sata_drive(drive: Arc<AhciPort>, probe: bool) {
+    let index = {
+        let mut drives = SATA_DRIVES.lock();
+        let index = drives.len();
+        if index >= MAX_DRIVES {
+            log::error!("Cannot add a SATA drive: too many of them");
+            return;
+        }
+        drives.push(drive.clone());
+        index
+    };
+    let letter = (index as u8 + b'a') as char;
 
-pub fn probe(info: &PciDeviceInfo) -> Result<&'static dyn Device, Error> {
-    let bar5 = info.config_space.bar(5).ok_or(Error::InvalidOperation)?;
-    let bar5 = bar5.as_memory().ok_or(Error::InvalidOperation)?;
+    let name = format!("sd{letter}");
+    log::info!("Register SATA drive: {name}");
 
-    let mut cmd = PciCommandRegister::from_bits_retain(info.config_space.command());
-    cmd &= !(PciCommandRegister::DISABLE_INTERRUPTS | PciCommandRegister::ENABLE_IO);
-    cmd |= PciCommandRegister::ENABLE_MEMORY | PciCommandRegister::BUS_MASTER;
-    info.config_space.set_command(cmd.bits());
+    devfs::add_named_block_device(drive.clone(), name.clone(), FileMode::new(0o600)).ok();
 
-    info.init_interrupts(PreferredInterruptMode::Msi)?;
-
-    // // TODO support regular PCI interrupts (ACPI dependency)
-    // let Some(mut msi) = info.config_space.capability::<MsiCapability>() else {
-    //     log::warn!("Ignoring AHCI: does not support MSI (and the OS doesn't yet support PCI IRQ)");
-    //     return Err(Error::InvalidOperation);
-    // };
-
-    // Map the registers
-    let regs = unsafe { DeviceMemoryIo::<Regs>::map(bar5, Default::default()) }?;
-
-    let version = Version::try_from(regs.VS.get())?;
-    let ahci_only = regs.CAP.matches_all(CAP::SAM::SET);
-    let max_port_count = regs.CAP.read(CAP::NP) as usize;
-    let has_64_bit = regs.CAP.matches_all(CAP::S64A::SET);
-
-    // TODO extract Number of Command Slots
-
-    let ahci = Box::leak(Box::new(AhciController {
-        regs: IrqSafeSpinlock::new(regs),
-        ports: OneTimeInit::new(),
-        received_fis_buffers: OneTimeInit::new(),
-        version,
-        max_port_count,
-        ahci_only,
-        has_64_bit,
-    }));
-
-    // TODO use multiple vectors if capable
-    info.map_interrupt(InterruptAffinity::Any, ahci)?;
-
-    Ok(ahci)
+    if probe {
+        runtime::spawn(async move {
+            let name = name;
+            log::info!("Probing partitions for {name}");
+            probe_partitions(drive, |index, partition| {
+                let partition_name = format!("{name}{}", index + 1);
+                devfs::add_named_block_device(
+                    Arc::new(partition),
+                    partition_name,
+                    FileMode::new(0o600),
+                )
+                .ok();
+            })
+            .await
+            .ok();
+        })
+        .ok();
+    }
+}
+
+static SATA_DRIVES: IrqSafeSpinlock<Vec<Arc<AhciPort>>> = IrqSafeSpinlock::new(Vec::new());
+
+pci_driver! {
+    matches: [class (0x01:0x06:0x01)],
+    driver: {
+        fn driver_name(&self) -> &str {
+            "ahci"
+        }
+
+        fn probe(&self, info: &PciDeviceInfo, dma: &Arc<dyn DmaAllocator>) -> Result<Arc<dyn Device>, Error> {
+            let bar5 = info.config_space.bar(5).ok_or(Error::InvalidOperation)?;
+            let bar5 = bar5.as_memory().ok_or(Error::InvalidOperation)?;
+
+            let mut cmd = PciCommandRegister::from_bits_retain(info.config_space.command());
+            cmd &= !(PciCommandRegister::DISABLE_INTERRUPTS | PciCommandRegister::ENABLE_IO);
+            cmd |= PciCommandRegister::ENABLE_MEMORY | PciCommandRegister::BUS_MASTER;
+            info.config_space.set_command(cmd.bits());
+
+            info.init_interrupts(PreferredInterruptMode::Msi(true))?;
+
+            // // TODO support regular PCI interrupts (ACPI dependency)
+            // let Some(mut msi) = info.config_space.capability::<MsiCapability>() else {
+            //     log::warn!("Ignoring AHCI: does not support MSI (and the OS doesn't yet support PCI IRQ)");
+            //     return Err(Error::InvalidOperation);
+            // };
+
+            // Map the registers
+            let regs = unsafe { DeviceMemoryIo::<Regs>::map(bar5, Default::default()) }?;
+
+            let version = Version::try_from(regs.VS.get())?;
+            let ahci_only = regs.CAP.matches_all(CAP::SAM::SET);
+            let max_port_count = regs.CAP.read(CAP::NP) as usize;
+            let has_64_bit = regs.CAP.matches_all(CAP::S64A::SET);
+
+            // TODO extract Number of Command Slots
+
+            let ahci = Arc::new(AhciController {
+                regs: IrqSafeSpinlock::new(regs),
+                dma: dma.clone(),
+                ports: OneTimeInit::new(),
+                received_fis_buffers: OneTimeInit::new(),
+                version,
+                max_port_count,
+                ahci_only,
+                has_64_bit,
+            });
+
+            // TODO use multiple vectors if capable
+            info.map_interrupt(InterruptAffinity::Any, ahci.clone())?;
+
+            Ok(ahci)
+        }
+    }
 }

kernel/driver/block/ahci/src/port.rs

@@ -5,12 +5,20 @@ use core::{
     task::{Context, Poll},
 };
 
-use alloc::{boxed::Box, string::String};
+use alloc::{boxed::Box, string::String, sync::Arc};
 use async_trait::async_trait;
 use bytemuck::Zeroable;
+use device_api::{device::Device, dma::DmaAllocator};
 use futures_util::task::AtomicWaker;
-use libk::vfs::block::NgBlockDevice;
-use libk_mm::{address::AsPhysicalAddress, device::DeviceMemoryIo, PageBox, PageSlice};
+use libk::{
+    device::block::BlockDevice,
+    dma::{DmaBuffer, DmaSlice, DmaSliceMut},
+    error::Error,
+};
+use libk_mm::{
+    address::PhysicalAddress, device::DeviceMemoryIo, table::MapAttributes, OnDemandPage,
+    PageProvider, VirtualPage,
+};
 use libk_util::{sync::IrqSafeSpinlock, waker::QueueWaker, OneTimeInit};
 use tock_registers::interfaces::{Readable, Writeable};
 
@@ -31,8 +39,8 @@ struct PortInner {
     regs: DeviceMemoryIo<'static, PortRegs>,
 
     #[allow(unused)]
-    received_fis: PageBox<ReceivedFis>,
-    command_list: PageBox<[CommandListEntry]>,
+    received_fis: DmaBuffer<ReceivedFis>,
+    command_list: DmaBuffer<[CommandListEntry]>,
 }
 
 pub struct PortInfo {
@@ -44,7 +52,7 @@ pub struct PortInfo {
 #[allow(unused)]
 pub struct AhciPort {
     inner: IrqSafeSpinlock<PortInner>,
-    ahci: &'static AhciController,
+    ahci: Arc<AhciController>,
     ty: PortType,
     pub(crate) index: usize,
     info: OneTimeInit<PortInfo>,
@@ -84,18 +92,16 @@ impl Drop for SubmittedCommand<'_> {
 impl PortInner {
     fn submit_command<C: AtaCommand>(
         &mut self,
+        dma: &dyn DmaAllocator,
         index: usize,
         command: &C,
     ) -> Result<(), AhciError> {
         let list_entry = &mut self.command_list[index];
         let mut table_entry =
-            PageBox::new(CommandTable::zeroed()).map_err(AhciError::MemoryError)?;
+            DmaBuffer::new(dma, CommandTable::zeroed()).map_err(AhciError::MemoryError)?;
 
         table_entry.setup_command(command)?;
-        *list_entry = CommandListEntry::new(
-            unsafe { table_entry.as_physical_address() },
-            command.prd_count(),
-        )?;
+        *list_entry = CommandListEntry::new(table_entry.bus_address(), command.prd_count())?;
 
         // Sync before send
         // XXX do this properly
@@ -120,9 +126,9 @@ impl PortInner {
 impl AhciPort {
     pub fn create(
         regs: DeviceMemoryIo<'static, PortRegs>,
-        ahci: &'static AhciController,
+        ahci: Arc<AhciController>,
         index: usize,
-    ) -> Result<&'static Self, AhciError> {
+    ) -> Result<Arc<Self>, AhciError> {
         log::debug!("Initialize port {}", index);
         regs.stop()?;
 
@@ -131,12 +137,14 @@ impl AhciPort {
             return Err(AhciError::DeviceError);
         }
 
-        let received_fis = PageBox::new(ReceivedFis::zeroed()).map_err(AhciError::MemoryError)?;
-        let command_list = PageBox::new_slice(CommandListEntry::zeroed(), COMMAND_LIST_LENGTH)
-            .map_err(AhciError::MemoryError)?;
+        let received_fis =
+            DmaBuffer::new(&*ahci.dma, ReceivedFis::zeroed()).map_err(AhciError::MemoryError)?;
+        let command_list =
+            DmaBuffer::new_slice(&*ahci.dma, CommandListEntry::zeroed(), COMMAND_LIST_LENGTH)
+                .map_err(AhciError::MemoryError)?;
 
-        regs.set_received_fis_address_64(unsafe { received_fis.as_physical_address() });
-        regs.set_command_list_address_64(unsafe { command_list.as_physical_address() });
+        regs.set_received_fis_address_64(received_fis.bus_address());
+        regs.set_command_list_address_64(command_list.bus_address());
 
         regs.IE.write(
             IE::DPE::SET
@@ -160,7 +168,7 @@ impl AhciPort {
         let command_available = QueueWaker::new();
         let command_allocation = IrqSafeSpinlock::new(0);
 
-        Ok(Box::leak(Box::new(Self {
+        let port = Arc::new(Self {
             inner: IrqSafeSpinlock::new(inner),
             ty: PortType::Sata,
             info: OneTimeInit::new(),
@@ -170,11 +178,15 @@ impl AhciPort {
             command_completion,
             command_allocation,
             command_available,
-        })))
+        });
+
+        Ok(port)
     }
 
-    pub async fn init(&'static self) -> Result<(), AhciError> {
-        let identify = self.perform_command(AtaIdentify::create()?).await?;
+    pub async fn init_inner(&self) -> Result<(), AhciError> {
+        let identify = self
+            .perform_command(AtaIdentify::create(&*self.ahci.dma)?)
+            .await?;
 
         let model = identify.model_number.to_string();
         let serial = identify.serial_number.to_string();
@@ -223,13 +235,17 @@ impl AhciPort {
         .await
     }
 
-    async fn submit<C: AtaCommand>(&self, command: &C) -> Result<SubmittedCommand, AhciError> {
+    async fn submit<C: AtaCommand>(&self, command: &C) -> Result<SubmittedCommand<'_>, AhciError> {
         if command.prd_count() > 2 {
             log::warn!("TODO: AHCI doesn't like 3+ PRD transfers");
             return Err(AhciError::RegionTooLarge);
         }
         let index = self.allocate_command().await;
-        if let Err(error) = self.inner.lock().submit_command(index, command) {
+        if let Err(error) = self
+            .inner
+            .lock()
+            .submit_command(&*self.ahci.dma, index, command)
+        {
             self.free_command(index);
             return Err(error);
         }
@@ -293,32 +309,48 @@ impl AhciPort {
 }
 
 #[async_trait]
-impl NgBlockDevice for AhciPort {
-    type Error = AhciError;
-
-    async fn read(
-        &self,
-        lba: u64,
-        buffer: &mut PageSlice<MaybeUninit<u8>>,
-    ) -> Result<(), AhciError> {
-        if buffer.len() % SECTOR_SIZE != 0 {
-            return Err(AhciError::InvalidBufferSize(buffer.len()));
-        }
-
-        let command = AtaReadDmaEx::new(lba, buffer.len() / SECTOR_SIZE, buffer);
-        self.submit(&command).await?.wait_for_completion().await
+impl BlockDevice for AhciPort {
+    fn allocate_buffer(&self, size: usize) -> Result<DmaBuffer<[MaybeUninit<u8>]>, Error> {
+        DmaBuffer::new_uninit_slice(&*self.ahci.dma, size)
     }
 
-    async fn write(&self, _lba: u64, _buffer: &PageSlice<u8>) -> Result<(), AhciError> {
-        // TODO AtaDmaWriteEx
-        Err(AhciError::FeatureNotImplemented)
+    async fn read_aligned(
+        &self,
+        position: u64,
+        buffer: DmaSliceMut<'_, MaybeUninit<u8>>,
+    ) -> Result<(), Error> {
+        if buffer.len() % SECTOR_SIZE != 0 {
+            log::warn!("ahci: misaligned buffer size: {}", buffer.len());
+            return Err(Error::InvalidOperation);
+        }
+        if position % SECTOR_SIZE as u64 != 0 {
+            log::warn!("ahci: misaligned read");
+            return Err(Error::InvalidOperation);
+        }
+
+        let lba = position / SECTOR_SIZE as u64;
+        let lba_count = buffer.len() / SECTOR_SIZE;
+        if lba + lba_count as u64 >= self.block_count() {
+            log::warn!("ahci: read crosses medium end");
+            return Err(Error::InvalidOperation);
+        }
+
+        let command = AtaReadDmaEx::new(lba, lba_count, buffer);
+        self.submit(&command).await?.wait_for_completion().await?;
+
+        Ok(())
+    }
+
+    async fn write_aligned(&self, _position: u64, _buffer: DmaSlice<'_, u8>) -> Result<(), Error> {
+        // TODO AtaWriteDmaEx
+        Err(Error::NotImplemented)
     }
 
     fn block_size(&self) -> usize {
         SECTOR_SIZE
     }
 
-    fn block_count(&self) -> usize {
+    fn block_count(&self) -> u64 {
         self.info().as_ref().map(|i| i.lba_count).unwrap() as _
     }
 
@@ -326,3 +358,37 @@ impl NgBlockDevice for AhciPort {
         (MAX_PRD_SIZE * 2) / SECTOR_SIZE
     }
 }
+
+impl Device for AhciPort {
+    fn display_name(&self) -> &str {
+        "AHCI SATA Drive"
+    }
+}
+
+impl PageProvider for AhciPort {
+    fn ondemand_fetch(&self, _opaque: u64) -> Result<OnDemandPage, Error> {
+        unimplemented!()
+    }
+
+    fn get_page(&self, _offset: u64) -> Result<VirtualPage, Error> {
+        unimplemented!()
+    }
+
+    fn release_page(
+        &self,
+        _offset: u64,
+        _phys: PhysicalAddress,
+        _dirty: bool,
+    ) -> Result<(), Error> {
+        unimplemented!()
+    }
+
+    fn clone_page(
+        &self,
+        _offset: u64,
+        _src_phys: PhysicalAddress,
+        _src_attrs: MapAttributes,
+    ) -> Result<PhysicalAddress, Error> {
+        unimplemented!()
+    }
+}

kernel/driver/block/ahci/src/regs.rs

@@ -1,4 +1,4 @@
-use libk_mm::address::PhysicalAddress;
+use libk::dma::BusAddress;
 use tock_registers::{
     interfaces::{ReadWriteable, Readable, Writeable},
     register_bitfields, register_structs,
@@ -141,14 +141,14 @@ impl PortRegs {
         Ok(())
     }
 
-    pub fn set_received_fis_address_64(&self, address: PhysicalAddress) {
-        let address: u64 = address.into();
+    pub fn set_received_fis_address_64(&self, address: BusAddress) {
+        let address: u64 = address.into_u64();
         self.FB.set(address as u32);
         self.FBU.set((address >> 32) as u32);
     }
 
-    pub fn set_command_list_address_64(&self, address: PhysicalAddress) {
-        let address: u64 = address.into();
+    pub fn set_command_list_address_64(&self, address: BusAddress) {
+        let address: u64 = address.into_u64();
         self.CLB.set(address as u32);
         self.CLBU.set((address >> 32) as u32);
     }

kernel/driver/block/nvme/Cargo.toml

@@ -10,6 +10,7 @@ libk-util.workspace = true
 libk-mm.workspace = true
 libk.workspace = true
 device-api = { workspace = true, features = ["derive"] }
+kernel-arch.workspace = true
 
 ygg_driver_pci = { path = "../../bus/pci" }
 kernel-fs = { path = "../../fs/kernel-fs" }

kernel/driver/block/nvme/src/command.rs

@@ -2,7 +2,7 @@
 
 use core::fmt::{self, Write};
 
-use libk_mm::address::PhysicalAddress;
+use libk::dma::BusAddress;
 use tock_registers::{interfaces::Readable, register_structs, registers::ReadOnly, UIntLike};
 
 use crate::queue::PhysicalRegionPage;
@@ -74,7 +74,7 @@ pub struct CreateIoCompletionQueue {
     pub id: u32,
     pub size: usize,
     pub vector: u32,
-    pub data: PhysicalAddress,
+    pub data: BusAddress,
 }
 
 #[derive(Clone, Copy, Debug)]
@@ -82,7 +82,7 @@ pub struct CreateIoSubmissionQueue {
     pub id: u32,
     pub cq_id: u32,
     pub size: usize,
-    pub data: PhysicalAddress,
+    pub data: BusAddress,
 }
 
 // Replies