Changed CMP handling and added simple JMP

I have no idea what I am doing rn

Cargo.lock

@@ -13,6 +13,34 @@ dependencies = [
  "version_check",
 ]
 
+[[package]]
+name = "allocator-api2"
+version = "0.2.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0942ffc6dcaadf03badf6e6a2d0228460359d5e34b57ccdc720b7382dfbd5ec5"
+
+[[package]]
+name = "ariadne"
+version = "0.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "72fe02fc62033df9ba41cba57ee19acf5e742511a140c7dbc3a873e19a19a1bd"
+dependencies = [
+ "unicode-width",
+ "yansi",
+]
+
+[[package]]
+name = "beef"
+version = "0.5.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3a8241f3ebb85c056b509d4327ad0358fbbba6ffb340bf388f26350aeda225b1"
+
+[[package]]
+name = "bytemuck"
+version = "1.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "17febce684fd15d89027105661fec94afb475cb995fbc59d2865198446ba2eea"
+
 [[package]]
 name = "cfg-if"
 version = "1.0.0"
@@ -20,8 +48,10 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "baf1de4339761588bc0619e3cbc0120ee582ebb74b53b4efbf79117bd2da40fd"
 
 [[package]]
-name = "compiler"
-version = "0.1.0"
+name = "fnv"
+version = "1.0.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3f9eec918d3f24069decb9af1554cad7c880e2da24a9afd88aca000531ab82c1"
 
 [[package]]
 name = "hashbrown"
@@ -33,30 +63,169 @@ dependencies = [
 ]
 
 [[package]]
-name = "hbvm"
-version = "0.1.0"
+name = "hashbrown"
+version = "0.14.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2c6201b9ff9fd90a5a3bac2e56a830d0caa509576f0e503818ee82c181b3437a"
 dependencies = [
- "hashbrown",
- "log",
+ "ahash",
+ "allocator-api2",
 ]
 
 [[package]]
-name = "log"
-version = "0.4.17"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "abb12e687cfb44aa40f41fc3978ef76448f9b6038cad6aef4259d3c095a2382e"
+name = "hbasm"
+version = "0.1.0"
 dependencies = [
- "cfg-if",
+ "ariadne",
+ "bytemuck",
+ "hashbrown 0.14.0",
+ "hbbytecode",
+ "lasso",
+ "literify",
+ "logos",
+ "paste",
+]
+
+[[package]]
+name = "hbbytecode"
+version = "0.1.0"
+
+[[package]]
+name = "hbvm"
+version = "0.1.0"
+dependencies = [
+ "hbbytecode",
+]
+
+[[package]]
+name = "lasso"
+version = "0.7.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4644821e1c3d7a560fe13d842d13f587c07348a1a05d3a797152d41c90c56df2"
+dependencies = [
+ "ahash",
+ "hashbrown 0.13.2",
+]
+
+[[package]]
+name = "literify"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "54e4d365df794ed78b4ce1061886f82eae7afa8e3a98ce4c4b0bfd0c777b1175"
+dependencies = [
+ "litrs",
+ "proc-macro2",
+ "quote",
+]
+
+[[package]]
+name = "litrs"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4f17c3668f3cc1132437cdadc93dab05e52d592f06948d3f64828430c36e4a70"
+dependencies = [
+ "proc-macro2",
+]
+
+[[package]]
+name = "logos"
+version = "0.13.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c000ca4d908ff18ac99b93a062cb8958d331c3220719c52e77cb19cc6ac5d2c1"
+dependencies = [
+ "logos-derive",
+]
+
+[[package]]
+name = "logos-codegen"
+version = "0.13.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dc487311295e0002e452025d6b580b77bb17286de87b57138f3b5db711cded68"
+dependencies = [
+ "beef",
+ "fnv",
+ "proc-macro2",
+ "quote",
+ "regex-syntax",
+ "syn",
+]
+
+[[package]]
+name = "logos-derive"
+version = "0.13.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dbfc0d229f1f42d790440136d941afd806bc9e949e2bcb8faa813b0f00d1267e"
+dependencies = [
+ "logos-codegen",
 ]
 
 [[package]]
 name = "once_cell"
-version = "1.17.1"
+version = "1.18.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b7e5500299e16ebb147ae15a00a942af264cf3688f47923b8fc2cd5858f23ad3"
+checksum = "dd8b5dd2ae5ed71462c540258bedcb51965123ad7e7ccf4b9a8cafaa4a63576d"
+
+[[package]]
+name = "paste"
+version = "1.0.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "de3145af08024dea9fa9914f381a17b8fc6034dfb00f3a84013f7ff43f29ed4c"
+
+[[package]]
+name = "proc-macro2"
+version = "1.0.66"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "18fb31db3f9bddb2ea821cde30a9f70117e3f119938b5ee630b7403aa6e2ead9"
+dependencies = [
+ "unicode-ident",
+]
+
+[[package]]
+name = "quote"
+version = "1.0.33"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5267fca4496028628a95160fc423a33e8b2e6af8a5302579e322e4b520293cae"
+dependencies = [
+ "proc-macro2",
+]
+
+[[package]]
+name = "regex-syntax"
+version = "0.6.29"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f162c6dd7b008981e4d40210aca20b4bd0f9b60ca9271061b07f78537722f2e1"
+
+[[package]]
+name = "syn"
+version = "2.0.29"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c324c494eba9d92503e6f1ef2e6df781e78f6a7705a0202d9801b198807d518a"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "unicode-ident",
+]
+
+[[package]]
+name = "unicode-ident"
+version = "1.0.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "301abaae475aa91687eb82514b328ab47a211a533026cb25fc3e519b86adfc3c"
+
+[[package]]
+name = "unicode-width"
+version = "0.1.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c0edd1e5b14653f783770bce4a4dabb4a5108a5370a5f5d8cfe8710c361f6c8b"
 
 [[package]]
 name = "version_check"
 version = "0.9.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "49874b5167b65d7193b8aba1567f5c7d93d001cafc34600cee003eda787e483f"
+
+[[package]]
+name = "yansi"
+version = "0.5.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "09041cd90cf85f7f8b2df60c646f853b7f535ce68f85244eb6731cf89fa498ec"

Cargo.toml

@@ -1,2 +1,3 @@
 [workspace]
-members = ["hbvm", "compiler"]
+resolver = "2"
+members  = ["hbasm", "hbbytecode", "hbvm"]

DESIGN_DOC.md

@@ -1,28 +0,0 @@
-# Math operations
-```
-MATH_OP
-    Add
-    Sub
-    Mul
-    Div
-    Mod
-```
-```
-MATH_TYPE
-    Unsigned
-    Signed
-    FloatingPoint
-```
-
-```
-MATH_OP_SIDES
-    Register Constant
-    Register Register
-    Constant Constant
-    Constant Register
-```
-`[MATH_OP] [MATH_OP_SIDES] [MATH_TYPE] [IMM_LHS] [IMM_RHS] [REG]`
-
-
-
-

assets/example.asm

@@ -1,4 +0,0 @@
-load 0 a0        ;; 05 00 A0
-load 10 a1       ;; 05 10 A1
-add a0 1 a0      ;; 01 A0 01 A0
-jump_neq a0 a1 0 ;; a1 A0 A1 0

assets/example.bytes

@@ -1,4 +0,0 @@
-load 10 A1
-load 0 A0
-add A0 1
-jump_less_than A0 A1 0

c-abi.md

@@ -0,0 +1,29 @@
+# C ABI (proposal)
+
+## C datatypes
+| C Type      | Description              |     Size (B) |
+|:------------|:-------------------------|-------------:|
+| char        | Character / byte         |            8 |
+| short       | Short integer            |           16 |
+| int         | Integer                  |           32 |
+| long        | Long integer             |           64 |
+| long long   | Long long integer        |           64 |
+| T*          | Pointer                  |           64 |
+| float       | Single-precision float   |           32 |
+| double      | Double-precision float   |           64 |
+| long double | Extended-precision float | **Bikeshed** |
+
+## Registers
+| Register | ABI Name | Description    | Saver  |
+|:---------|:---------|:---------------|:-------|
+| `r0`     | —        | Zero register  | N/A    |
+| `r1`     | `ra`     | Return address | Caller |
+| `r2`     | `sp`     | Stack pointer  | Callee |
+| `r3`     | `tp`     | Thread pointer | N/A    |
+
+**TODO:** Parameters
+
+**TODO:** Saved
+
+**TODO:** Temp
+

compiler/Cargo.toml

@@ -1,8 +0,0 @@
-[package]
-name = "compiler"
-version = "0.1.0"
-edition = "2021"
-
-# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
-
-[dependencies]

compiler/src/main.rs

@@ -1,5 +0,0 @@
-fn main() {
-    let prog = "load 1, A0
-    jump 0";
-    println!("Hello, world!");
-}

hbasm/Cargo.toml

@@ -0,0 +1,22 @@
+[package]
+name = "hbasm"
+version = "0.1.0"
+edition = "2021"
+
+[dependencies]
+ariadne    = "0.3"
+bytemuck   = "1.13"
+hashbrown  = "0.14"
+hbbytecode = { path = "../hbbytecode" }
+literify   = "0.1"
+paste      = "1.0"
+
+[dependencies.lasso]
+version = "0.7"
+default-features = false
+features = ["no-std"]
+
+[dependencies.logos]
+version = "0.13"
+default-features = false
+features = ["export_derive"]

hbasm/assets/add.hbasm

@@ -0,0 +1,12 @@
+-- Add two numbers
+-- A + B = C
+
+-- r1 A
+li r1, 2
+-- r2 Result
+li r2, 0
+-- B = 4
+addi r2, r1, 4
+
+-- terminate execution
+tx

hbasm/assets/celsius_converter.hbasm

@@ -0,0 +1,16 @@
+-- r1 will be the temp in fahrenheit
+-- r2 temp in celsius 
+-- r3/r4/r5 will be used by constants
+
+-- (f - 32) * 5 / 9
+
+li r1, 100
+
+li r3, 32
+li r4, 5
+li r5, 9
+
+sub r2, r1, r3
+mul r2, r2, r4
+dir r2, r0, r2, r5
+tx

hbasm/assets/ecall.hbasm

@@ -0,0 +1,14 @@
+li r255, 0
+ecall
+
+li r255, 1
+li r254, 1
+li r253, 100
+ecall
+
+li r255, 2
+li r254, 0
+li r253, 0
+ecall
+
+tx

hbasm/assets/inf_loop.hbasm

@@ -0,0 +1,2 @@
+L:
+ jal r0, r0, L

hbasm/assets/memory.hbasm

@@ -0,0 +1,4 @@
+li r20, 1010
+st r20, r24, 0, 1
+addi r24, r0, 10
+tx

hbasm/assets/serial_driver.hbasm

@@ -0,0 +1,18 @@
+jmp r0, start
+start:
+    jmp r0, init_serial_port
+
+-- Uses r20 to set the port
+init_serial_port:
+    add r20, r30, r10
+    li r20, 00
+
+   -- outb(PORT + 1, 0x00);    // Disable all interrupts
+   -- outb(PORT + 3, 0x80);    // Enable DLAB (set baud rate divisor)
+   -- outb(PORT + 0, 0x03);    // Set divisor to 3 (lo byte) 38400 baud
+   -- outb(PORT + 1, 0x00);    //                  (hi byte)
+   -- outb(PORT + 3, 0x03);    // 8 bits, no parity, one stop bit
+   -- outb(PORT + 2, 0xC7);    // Enable FIFO, clear them, with 14-byte threshold
+   -- outb(PORT + 4, 0x0B);    // IRQs enabled, RTS/DSR set
+   -- outb(PORT + 4, 0x1E);    // Set in loopback mode, test the serial chip
+   -- outb(PORT + 0, 0xAE);    // Test serial chip (send byte 0xAE and check if serial returns same byte)

hbasm/src/lib.rs

@@ -0,0 +1,104 @@
+//! Holey Bytes Assembler
+//!
+//! Some people claim:
+//! > Write programs to handle text streams, because that is a universal interface.
+//!
+//! We at AbleCorp believe that nice programatic API is nicer than piping some text
+//! into a program. It's less error-prone and faster.
+//!
+//! So this crate contains both assembleer with API for programs and a text assembler
+//! for humans to write
+
+#![no_std]
+
+extern crate alloc;
+
+mod macros;
+
+use {
+    alloc::{vec, vec::Vec},
+    hashbrown::HashSet,
+};
+
+/// Assembler
+///
+/// - Opcode-generic, instruction-type-specific methods are named `i_param_<type>`
+///     - You likely won't need to use them, but they are here, just in case :)
+/// - Instruction-specific methods are named `i_<instruction>`
+pub struct Assembler {
+    pub buf: Vec<u8>,
+    pub sub: HashSet<usize>,
+}
+
+impl Default for Assembler {
+    fn default() -> Self {
+        Self {
+            buf: vec![0; 4],
+            sub: Default::default(),
+        }
+    }
+}
+
+hbbytecode::invoke_with_def!(macros::text::gen_text);
+
+impl Assembler {
+    hbbytecode::invoke_with_def!(macros::asm::impl_asm);
+
+    /// Append 12 zeroes (UN) at the end and add magic to the begining
+    ///
+    /// # HoleyBytes lore
+    ///
+    /// In reference HBVM implementation checks are done in
+    /// a separate phase before execution.
+    ///
+    /// This way execution will be much faster as they have to
+    /// be done only once.
+    ///
+    /// There was an issue. You cannot statically check register values and
+    /// `JAL` instruction could hop at the end of program to some byte, which
+    /// will be interpreted as some valid opcode and VM in attempt to decode
+    /// the instruction performed out-of-bounds read which leads to undefined behaviour.
+    ///
+    /// Several options were considered to overcome this, but inserting some data at
+    /// program's end which when executed would lead to undesired behaviour, though
+    /// not undefined behaviour.
+    ///
+    /// Newly created `UN` (as UNreachable) was chosen as
+    /// - It was a good idea to add some equivalent to `ud2` anyways
+    /// - It was chosen to be zero
+    /// - What if you somehow reached that code, it will appropriately bail :)
+    /// - (yes, originally `NOP` was considered)
+    ///
+    /// Why 12 bytes? That's the size of largest instruction parameter part.
+    pub fn finalise(&mut self) {
+        self.buf.extend([0; 12]);
+        self.buf[0..4].copy_from_slice(&0xAB1E0B_u32.to_le_bytes());
+    }
+}
+
+/// Immediate value
+///
+/// # Implementor notice
+/// It should insert exactly 8 bytes, otherwise output will be malformed.
+/// This is not checked in any way
+pub trait Imm {
+    /// Insert immediate value
+    fn insert(&self, asm: &mut Assembler);
+}
+
+/// Implement immediate values
+macro_rules! impl_imm_le_bytes {
+    ($($ty:ty),* $(,)?) => {
+        $(
+            impl Imm for $ty {
+                #[inline(always)]
+                fn insert(&self, asm: &mut Assembler) {
+                    // Convert to little-endian bytes, insert.
+                    asm.buf.extend(self.to_le_bytes());
+                }
+            }
+        )*
+    };
+}
+
+impl_imm_le_bytes!(u64, i64, f64);

hbasm/src/macros/asm.rs

@@ -0,0 +1,89 @@
+//! Macros to generate [`crate::Assembler`]
+
+/// Incremental token-tree muncher to implement specific instruction
+/// functions based on generic function for instruction type
+macro_rules! impl_asm_opcodes {
+    ( // End case
+        $generic:ident
+        ($($param_i:ident: $param_ty:ty),*)
+        => []
+    ) => {};
+
+    (
+        $generic:ident
+        ($($param_i:ident: $param_ty:ty),*)
+        => [$opcode:ident, $($rest:tt)*]
+    ) => {
+        // Instruction-specific function
+        paste::paste! {
+            #[inline(always)]
+            pub fn [<i_ $opcode:lower>](&mut self, $($param_i: $param_ty),*) {
+                self.$generic(hbbytecode::opcode::$opcode, $($param_i),*)
+            }
+        }
+
+        // And recurse!
+        macros::asm::impl_asm_opcodes!(
+            $generic($($param_i: $param_ty),*)
+            => [$($rest)*]
+        );
+    };
+}
+
+/// Numeric value insert
+macro_rules! impl_asm_insert {
+    // Immediate - this is trait-based,
+    // the insertion is delegated to its implementation
+    ($self:expr, $id:ident, I) => {
+        Imm::insert(&$id, $self)
+    };
+
+    // Length - cannot be more than 2048
+    ($self:expr, $id:ident, L) => {{
+        assert!($id <= 2048);
+        $self.buf.extend($id.to_le_bytes())
+    }};
+
+    // Other numbers, just insert their bytes, little endian
+    ($self:expr, $id:ident, $_:ident) => {
+        $self.buf.extend($id.to_le_bytes())
+    };
+}
+
+/// Implement assembler
+macro_rules! impl_asm {
+    (
+        $(
+            $ityn:ident
+            ($($param_i:ident: $param_ty:ident),* $(,)?)
+            => [$($opcode:ident),* $(,)?],
+        )*
+    ) => {
+        paste::paste! {
+            $(
+                // Opcode-generic functions specific for instruction types
+                pub fn [<i_param_ $ityn>](&mut self, opcode: u8, $($param_i: macros::asm::ident_map_ty!($param_ty)),*) {
+                    self.buf.push(opcode);
+                    $(macros::asm::impl_asm_insert!(self, $param_i, $param_ty);)*
+                }
+
+                // Generate opcode-specific functions calling the opcode-generic ones
+                macros::asm::impl_asm_opcodes!(
+                    [<i_param_ $ityn>]($($param_i: macros::asm::ident_map_ty!($param_ty)),*)
+                    => [$($opcode,)*]
+                );
+            )*
+        }
+    };
+}
+
+/// Map operand type to Rust type
+#[rustfmt::skip]
+macro_rules! ident_map_ty {
+    (R)         => { u8 };       // Register is just u8
+    (I)         => { impl Imm }; // Immediate is anything implementing the trait
+    (L)         => { u16 };      // Copy count
+    ($id:ident) => { $id };      // Anything else → identity map
+}
+
+pub(crate) use {ident_map_ty, impl_asm, impl_asm_insert, impl_asm_opcodes};

hbasm/src/macros/mod.rs

@@ -0,0 +1,6 @@
+//! And here the land of macros begin.
+//!
+//! They do not bite, really. Have you seen what Yandros is writing?
+
+pub mod asm;
+pub mod text;

hbasm/src/macros/text.rs

@@ -0,0 +1,293 @@
+//! Macros to generate text-code assembler at [`crate::text`]
+// Refering in module which generates a module to that module — is that even legal? :D
+
+/// Generate text code based assembler
+macro_rules! gen_text {
+    (
+        $(
+            $ityn:ident
+            ($($param_i:ident: $param_ty:ident),* $(,)?)
+            => [$($opcode:ident),* $(,)?],
+        )*
+    ) => {
+        /// # Text assembler
+        /// Text assembler generated simply calls methods in the [`crate::Assembler`] type.
+        /// 
+        /// # Syntax
+        /// ```text
+        /// instruction op1, op2, …
+        /// …
+        /// ```
+        /// - Opcode names are lowercase
+        /// - Registers are prefixed with `r` followed by number
+        /// - Operands are separated by `,`
+        /// - Instructions are separated by either line feed or `;` (αυτό δεν είναι ερωτηματικό!)
+        /// - Labels are defined by their names followed by colon `label:`
+        /// - Labels are referenced simply by their names
+        /// - Immediates are numbers, can be negative, floats are not yet supported
+        pub mod text {
+            use {
+                crate::{
+                    Assembler,
+                    macros::text::*,
+                },
+                hashbrown::HashMap,
+                lasso::{Key, Rodeo, Spur},
+                logos::{Lexer, Logos, Span},
+            };
+
+            paste::paste!(literify::literify! {
+                /// Assembly token
+                #[derive(Clone, Copy, Debug, PartialEq, Eq, Logos)]
+                #[logos(extras = Rodeo)]
+                #[logos(skip r"[ \t\t]+")]
+                #[logos(skip r"-- .*")]
+                pub enum Token {
+                    $($(#[token(~([<$opcode:lower>]), |_| hbbytecode::opcode::[<$opcode:upper>])])*)*
+                    Opcode(u8),
+
+                    #[regex("[0-9]+", |l| l.slice().parse().ok())]
+                    #[regex(
+                        "-[0-9]+",
+                        |lexer| {
+                            Some(u64::from_ne_bytes(lexer.slice().parse::<i64>().ok()?.to_ne_bytes()))
+                        },
+                    )] Integer(u64),
+
+                    #[regex(
+                        "r[0-9]+",
+                        |lexer| match lexer.slice()[1..].parse() {
+                            Ok(n) => Some(n),
+                            _ => None
+                        },
+                    )] Register(u8),
+
+                    #[regex(
+                        r"\p{XID_Start}\p{XID_Continue}*:",
+                        |lexer| lexer.extras.get_or_intern(&lexer.slice()[..lexer.slice().len() - 1]),
+                    )] Label(Spur),
+
+                    #[regex(
+                        r"\p{XID_Start}\p{XID_Continue}*",
+                        |lexer| lexer.extras.get_or_intern(lexer.slice()),
+                    )] Symbol(Spur),
+
+                    #[token("\n")]
+                    #[token(";")] ISep,
+                    #[token(",")] PSep,
+                }
+            });
+
+            /// Type of error
+            #[derive(Copy, Clone, Debug, PartialEq, Eq)]
+            pub enum ErrorKind {
+                UnexpectedToken,
+                InvalidToken,
+                UnexpectedEnd,
+                InvalidSymbol,
+            }
+
+            /// Text assembly error
+            #[derive(Clone, Debug, PartialEq, Eq)]
+            pub struct Error {
+                pub kind: ErrorKind,
+                pub span: Span,
+            }
+
+            /// Parse code and insert instructions
+            pub fn assemble(asm: &mut Assembler, code: &str) -> Result<(), Error> {
+                pub struct TextAsm<'a> {
+                    asm: &'a mut Assembler,
+                    lexer: Lexer<'a, Token>,
+                    symloc: HashMap<Spur, usize>,
+                }
+
+                impl<'a> TextAsm<'a> {
+                    fn next(&mut self) -> Result<Token, ErrorKind> {
+                        match self.lexer.next() {
+                            Some(Ok(t))   => Ok(t),
+                            Some(Err(())) => Err(ErrorKind::InvalidToken),
+                            None          => Err(ErrorKind::UnexpectedEnd),
+                        }
+                    }
+
+                    #[inline(always)]
+                    fn run(&mut self) -> Result<(), ErrorKind> {
+                        loop {
+                            match self.lexer.next() {
+                                // Got an opcode
+                                Some(Ok(Token::Opcode(op))) => {
+                                    match op {
+                                        // Special-cased
+                                        hbbytecode::opcode::BRC => {
+                                            param_extract_itm!(
+                                                self,
+                                                p0: R,
+                                                p1: R,
+                                                p2: u8
+                                            );
+
+                                            self.asm.i_param_bbb(op, p0, p1, p2);
+                                        },
+
+                                        // Take all the opcodes and match them to their corresponding functions
+                                        $(
+                                            #[allow(unreachable_patterns)]
+                                            $(hbbytecode::opcode::$opcode)|* => paste::paste!({
+                                                param_extract_itm!(self, $($param_i: $param_ty),*);
+                                                self.asm.[<i_param_ $ityn>](op, $($param_i),*);
+                                            }),
+                                        )*
+
+                                        // Already matched in Logos, should not be able to obtain
+                                        // invalid opcode.
+                                        _ => unreachable!(),
+                                    }
+                                }
+                                // Insert label to table
+                                Some(Ok(Token::Label(lbl))) => {
+                                    self.symloc.insert(lbl, self.asm.buf.len());
+                                }
+                                // Instruction separator (LF, ;)
+                                Some(Ok(Token::ISep)) => (),
+                                Some(Ok(_))           => return Err(ErrorKind::UnexpectedToken),
+                                Some(Err(()))         => return Err(ErrorKind::InvalidToken),
+                                None                  => return Ok(()),
+                            }
+                        }
+                    }
+                }
+
+                let mut asm = TextAsm {
+                    asm,
+                    lexer: Token::lexer(code),
+                    symloc: HashMap::default(),
+                };
+
+                asm.run()
+                    .map_err(|kind| Error { kind, span: asm.lexer.span() })?;
+
+                // Walk table and substitute labels
+                // for their addresses
+                for &loc in &asm.asm.sub {
+                    // Extract indices from the code and get addresses from table
+                    let val = asm.symloc
+                        .get(
+                            &Spur::try_from_usize(bytemuck::pod_read_unaligned::<u64>(
+                                &asm.asm.buf[loc..loc + core::mem::size_of::<u64>()]) as _
+                            ).unwrap()
+                        )
+                        .ok_or(Error { kind: ErrorKind::InvalidSymbol, span: 0..0 })?
+                        .to_le_bytes();
+
+                    // New address
+                    asm.asm.buf[loc..]
+                        .iter_mut()
+                        .zip(val)
+                        .for_each(|(dst, src)| *dst = src);
+                }
+
+                Ok(())
+            }
+
+            // Fun fact: this is a little hack
+            // It may slow the things a little bit down, but
+            // it made the macro to be made pretty nice.
+            // 
+            // If you have any idea how to get rid of this,
+            // contributions are welcome :)
+            // I *likely* won't try anymore.
+            enum InternalImm {
+                Const(u64),
+                Named(Spur),
+            }
+
+            impl $crate::Imm for InternalImm {
+                #[inline]
+                fn insert(&self, asm: &mut Assembler) {
+                    match self {
+                        // Constant immediate, just put it in
+                        Self::Const(a) => a.insert(asm),
+                        // Label
+                        Self::Named(a) => {
+                            // Insert to the sub table that substitution will be
+                            // requested
+                            asm.sub.insert(asm.buf.len());
+                            // Insert value from interner in place
+                            asm.buf.extend((a.into_usize() as u64).to_le_bytes());
+                        },
+                    }
+                }
+            }
+        }
+    };
+}
+
+/// Extract item by pattern, otherwise return [`ErrorKind::UnexpectedToken`]
+macro_rules! extract_pat {
+    ($self:expr, $pat:pat) => {
+        let $pat = $self.next()?
+            else { return Err(ErrorKind::UnexpectedToken) };
+    };
+}
+
+/// Generate extract macro
+macro_rules! gen_extract {
+    // Integer types have same body
+    ($($int:ident),* $(,)?) => {
+        /// Extract operand from code
+        macro_rules! extract {
+            // Register (require prefixing with r)
+            ($self:expr, R, $id:ident) => {
+                extract_pat!($self, Token::Register($id));
+            };
+
+            ($self:expr, L, $id:ident) => {
+                extract_pat!($self, Token::Integer($id));
+                if $id > 2048 {
+                    return Err(ErrorKind::InvalidToken);
+                }
+
+                let $id = u16::try_from($id).unwrap();
+            };
+
+            // Immediate
+            ($self:expr, I, $id:ident) => {
+                let $id = match $self.next()? {
+                    // Either straight up integer
+                    Token::Integer(a) => InternalImm::Const(a),
+                    // …or a label
+                    Token::Symbol(a) => InternalImm::Named(a),
+                    _ => return Err(ErrorKind::UnexpectedToken),
+                };
+            };
+
+            // Get $int, if not fitting, the token is claimed invalid
+            $(($self:expr, $int, $id:ident) => {
+                extract_pat!($self, Token::Integer($id));
+                let $id = $int::try_from($id).map_err(|_| ErrorKind::InvalidToken)?;
+            });*;
+        }
+    };
+}
+
+gen_extract!(u8, u16, u32);
+
+/// Parameter extract incremental token-tree muncher
+/// 
+/// What else would it mean?
+macro_rules! param_extract_itm {
+    ($self:expr, $($id:ident: $ty:ident)? $(, $($tt:tt)*)?) => {
+        // Extract pattern
+        $(extract!($self, $ty, $id);)?
+        $(
+            // Require operand separator
+            extract_pat!($self, Token::PSep);
+            // And go to the next (recursive)
+            // …munch munch… yummy token trees.
+            param_extract_itm!($self, $($tt)*);
+        )?
+    };
+}
+
+pub(crate) use {extract, extract_pat, gen_text, param_extract_itm};

hbasm/src/main.rs

@@ -0,0 +1,56 @@
+use std::io::Write;
+
+use hbasm::Assembler;
+
+use {
+    ariadne::{ColorGenerator, Label, Report, ReportKind, Source},
+    std::{
+        error::Error,
+        io::{stdin, Read},
+    },
+};
+
+fn main() -> Result<(), Box<dyn Error>> {
+    let mut code = String::new();
+    stdin().read_to_string(&mut code)?;
+
+    let mut assembler = Assembler::default();
+    if let Err(e) = hbasm::text::assemble(&mut assembler, &code) {
+        let mut colors = ColorGenerator::new();
+
+        let e_code = match e.kind {
+            hbasm::text::ErrorKind::UnexpectedToken => 1,
+            hbasm::text::ErrorKind::InvalidToken => 2,
+            hbasm::text::ErrorKind::UnexpectedEnd => 3,
+            hbasm::text::ErrorKind::InvalidSymbol => 4,
+        };
+        let message = match e.kind {
+            hbasm::text::ErrorKind::UnexpectedToken => "This token is not expected!",
+            hbasm::text::ErrorKind::InvalidToken => "The token is not valid!",
+            hbasm::text::ErrorKind::UnexpectedEnd => {
+                "The assembler reached the end of input unexpectedly!"
+            }
+            hbasm::text::ErrorKind::InvalidSymbol => {
+                "This referenced symbol doesn't have a corresponding label!"
+            }
+        };
+        let a = colors.next();
+
+        Report::build(ReportKind::Error, "engine_internal", e.span.clone().start)
+            .with_code(e_code)
+            .with_message(format!("{:?}", e.kind))
+            .with_label(
+                Label::new(("engine_internal", e.span))
+                    .with_message(message)
+                    .with_color(a),
+            )
+            .finish()
+            .eprint(("engine_internal", Source::from(&code)))
+            .unwrap();
+    } else {
+        assembler.finalise();
+        std::io::stdout().lock().write_all(&assembler.buf).unwrap();
+    }
+
+    Ok(())
+}

hbbytecode/Cargo.toml

@@ -0,0 +1,6 @@
+[package]
+name = "hbbytecode"
+version = "0.1.0"
+edition = "2021"
+
+[dependencies]

hbbytecode/hbbytecode.h

@@ -0,0 +1,68 @@
+/* HoleyBytes Bytecode representation in C
+ * Requires C23 compiler or better
+ *
+ * Uses MSVC pack pragma extension,
+ * proved to work with Clang and GNU® GCC™.
+ */
+
+#pragma once
+#include <assert.h>
+#include <limits.h>
+#include <stdint.h>
+
+static_assert(CHAR_BIT == 8, "Cursed architectures are not supported");
+
+enum hbbc_Opcode: uint8_t {
+    hbbc_Op_UN    , hbbc_Op_TX   , hbbc_Op_NOP   , hbbc_Op_ADD   , hbbc_Op_SUB   , hbbc_Op_MUL   ,
+    hbbc_Op_AND   , hbbc_Op_OR   , hbbc_Op_XOR   , hbbc_Op_SL    , hbbc_Op_SR    , hbbc_Op_SRS   ,
+    hbbc_Op_CMP   , hbbc_Op_CMPU , hbbc_Op_DIR   , hbbc_Op_NEG   , hbbc_Op_NOT   , hbbc_Op_ADDI  , 
+    hbbc_Op_MULI  , hbbc_Op_ANDI , hbbc_Op_ORI   , hbbc_Op_XORI  , hbbc_Op_SLI   , hbbc_Op_SRI   ,
+    hbbc_Op_SRSI  , hbbc_Op_CMPI , hbbc_Op_CMPUI , hbbc_Op_CP    , hbbc_Op_SWA   , hbbc_Op_LI    ,
+    hbbc_Op_LD    , hbbc_Op_ST   , hbbc_Op_BMC   , hbbc_Op_BRC   , hbbc_Op_JMP   , hbbc_Op_JAL   ,
+    hbbc_Op_JEQ   , hbbc_Op_JNE  , hbbc_Op_JLT   , hbbc_Op_JGT   , hbbc_Op_JLTU  , hbbc_Op_JGTU  ,
+    hbbc_Op_ECALL , hbbc_Op_ADDF , hbbc_Op_SUBF  , hbbc_Op_MULF  , hbbc_Op_DIRF  , hbbc_Op_FMAF ,
+    hbbc_Op_NEGF  , hbbc_Op_ITF  , hbbc_Op_FTI   , hbbc_Op_ADDFI , hbbc_Op_MULFI ,
+} typedef hbbc_Opcode;
+
+static_assert(sizeof(hbbc_Opcode) == 1);
+
+#pragma pack(push, 1)
+struct hbbc_ParamBBBB
+    { uint8_t _0; uint8_t _1; uint8_t _2; uint8_t _3; }
+    typedef hbbc_ParamBBBB;
+    static_assert(sizeof(hbbc_ParamBBBB) == 32 / 8);
+
+struct hbbc_ParamBBB
+    { uint8_t _0; uint8_t _1; uint8_t _2; }
+    typedef hbbc_ParamBBB;
+    static_assert(sizeof(hbbc_ParamBBB) == 24 / 8);
+
+struct hbbc_ParamBBDH
+    { uint8_t _0; uint8_t _1; uint64_t _2; uint16_t _3; }
+    typedef hbbc_ParamBBDH;
+    static_assert(sizeof(hbbc_ParamBBDH) == 96 / 8);
+
+struct hbbc_ParamBBD
+    { uint8_t _0; uint8_t _1; uint64_t _2; }
+    typedef hbbc_ParamBBD;
+    static_assert(sizeof(hbbc_ParamBBD) == 80 / 8);
+
+struct hbbc_ParamBBW
+    { uint8_t _0; uint8_t _1; uint32_t _2; }
+    typedef hbbc_ParamBBW;
+    static_assert(sizeof(hbbc_ParamBBW) == 48 / 8);
+
+struct hbbc_ParamBB
+    { uint8_t _0; uint8_t _1; }
+    typedef hbbc_ParamBB;
+    static_assert(sizeof(hbbc_ParamBB) == 16 / 8);
+
+struct hbbc_ParamBD
+    { uint8_t _0; uint64_t _1; }
+    typedef hbbc_ParamBD;
+    static_assert(sizeof(hbbc_ParamBD) == 72 / 8);
+
+typedef uint64_t hbbc_ParamD;
+    static_assert(sizeof(hbbc_ParamD) == 64 / 8);
+
+#pragma pack(pop)

hbbytecode/src/gen_valider.rs

@@ -0,0 +1,170 @@
+//! Generate HoleyBytes code validator
+
+macro_rules! gen_valider {
+    (
+        $(
+            $ityn:ident
+            ($($param_i:ident: $param_ty:ident),* $(,)?)
+            => [$($opcode:ident),* $(,)?],
+        )*
+    ) => {
+        #[allow(unreachable_code)]
+        pub mod valider {
+            //! Validate if program is sound to execute
+
+            /// Program validation error kind
+            #[derive(Clone, Copy, Debug, PartialEq, Eq)]
+            pub enum ErrorKind {
+                /// Unknown opcode
+                InvalidInstruction,
+                /// VM doesn't implement this valid opcode
+                Unimplemented,
+                /// Attempted to copy over register boundary
+                RegisterArrayOverflow,
+                /// Program is not validly terminated
+                InvalidEnd,
+                /// Program misses magic
+                MissingMagic
+            }
+
+            /// Error
+            #[derive(Clone, Copy, Debug, PartialEq, Eq)]
+            pub struct Error {
+                /// Kind
+                pub kind:  ErrorKind,
+                /// Location in bytecode
+                pub index: usize,
+            }
+
+            /// Perform bytecode validation. If it passes, the program should be
+            /// sound to execute.
+            pub fn validate(mut program: &[u8]) -> Result<(), Error> {
+                // Validate magic
+                if program.get(0..4) != Some(&0xAB1E0B_u32.to_le_bytes()) {
+                    return Err(Error {
+                        kind: ErrorKind::MissingMagic,
+                        index: 0,
+                    });
+                }
+
+                // Program has to end with 12 zeroes, if there is less than
+                // 12 bytes, program is invalid.
+                if program.len() < 12 {
+                    return Err(Error {
+                        kind:  ErrorKind::InvalidEnd,
+                        index: 0,
+                    });
+                }
+
+                // Verify that program ends with 12 zeroes
+                for (index, item) in program.iter().enumerate().skip(program.len() - 12) {
+                    if *item != 0 {
+                        return Err(Error {
+                            kind: ErrorKind::InvalidEnd,
+                            index,
+                        });
+                    }
+                }
+
+                let start = program;
+                program = &program[4..];
+                loop {
+                    use crate::opcode::*;
+                    program = match program {
+                        // End of program
+                        [] => return Ok(()),
+
+                        // Memory load/store cannot go out-of-bounds register array
+                        //         B   B  D1 D2 D3 D4 D5 D6 D7 D8    H1      H2
+                        [LD..=ST, reg, _, _, _, _, _, _, _, _, _, count_0, count_1, ..]
+                            if usize::from(*reg) * 8
+                                + usize::from(u16::from_le_bytes([*count_0, *count_1]))
+                                > 2048 =>
+                        {
+                            return Err(Error {
+                                kind:  ErrorKind::RegisterArrayOverflow,
+                                index: (program.as_ptr() as usize) - (start.as_ptr() as usize),
+                            });
+                        }
+
+                        // Block register copy cannot go out-of-bounds register array
+                        [BRC, src, dst, count, ..]
+                            if src.checked_add(*count).is_none()
+                                || dst.checked_add(*count).is_none() =>
+                        {
+                            return Err(Error {
+                                kind:  ErrorKind::RegisterArrayOverflow,
+                                index: (program.as_ptr() as usize) - (start.as_ptr() as usize),
+                            });
+                        }
+
+                        $(
+                            $crate::gen_valider::inst_chk!(
+                                rest, $ityn, $($opcode),*
+                            )
+                        )|* => rest,
+
+                        // The plebs
+                        _ => {
+                            return Err(Error {
+                                kind:  ErrorKind::InvalidInstruction,
+                                index: (program.as_ptr() as usize) - (start.as_ptr() as usize),
+                            })
+                        }
+                    }
+                }
+            }
+        }
+    };
+}
+
+/// Generate instruction check pattern
+macro_rules! inst_chk {
+    // Sadly this has hardcoded instruction types,
+    // as I cannot generate parts of patterns+
+
+    ($rest:ident, bbbb, $($opcode:ident),*) => {
+        //             B  B  B  B
+        [$($opcode)|*, _, _, _, _, $rest @ ..]
+    };
+
+    ($rest:ident, bbb, $($opcode:ident),*) => {
+        //             B  B  B
+        [$($opcode)|*, _, _, _, $rest @ ..]
+    };
+
+    ($rest:ident, bbdh, $($opcode:ident),*) => {
+        //             B  B  D1 D2 D3 D4 D5 D6 D7 D8 H1 H2
+        [$($opcode)|*, _, _, _, _, _, _, _, _, _, _, _, _, $rest @ ..]
+    };
+
+    ($rest:ident, bbd, $($opcode:ident),*) => {
+        //             B  B  D1 D2 D3 D4 D5 D6 D7 D8
+        [$($opcode)|*, _, _, _, _, _, _, _, _, _, _, $rest @ ..]
+    };
+
+    ($rest:ident, bbw, $($opcode:ident),*) => {
+        //             B  B  W1 W2 W3 W4
+        [$($opcode)|*, _, _, _, _, _, _, $rest @ ..]
+    };
+
+    ($rest:ident, bb, $($opcode:ident),*) => {
+        //             B  B
+        [$($opcode)|*, _, _, $rest @ ..]
+    };
+
+    ($rest:ident, bd, $($opcode:ident),*) => {
+        //             B  D1 D2 D3 D4 D5 D6 D7 D8
+        [$($opcode)|*, _, _, _, _, _, _, _, _, _, $rest @ ..]
+    };
+
+    ($rest:ident, n, $($opcode:ident),*) => {
+        [$($opcode)|*, $rest @ ..]
+    };
+
+    ($_0:ident, $($_1:ident),*) => {
+        compile_error!("Invalid instruction type");
+    }
+}
+
+pub(crate) use {gen_valider, inst_chk};

hbbytecode/src/lib.rs

@@ -0,0 +1,163 @@
+#![no_std]
+
+mod gen_valider;
+
+macro_rules! constmod {
+    ($vis:vis $mname:ident($repr:ty) {
+        $(#![doc = $mdoc:literal])?
+        $($cname:ident = $val:expr $(,$doc:literal)?;)*
+    }) => {
+        $(#[doc = $mdoc])?
+        $vis mod $mname {
+            $(
+                $(#[doc = $doc])?
+                pub const $cname: $repr = $val;
+            )*
+        }
+    };
+}
+
+#[allow(rustdoc::invalid_rust_codeblocks)]
+/// Invoke macro with bytecode definition
+/// # Input syntax
+/// ```no_run
+/// macro!(
+///     INSTRUCTION_TYPE(p0: TYPE, p1: TYPE, …)
+///         => [INSTRUCTION_A, INSTRUCTION_B, …],
+///     …
+/// );
+/// ```
+/// - Instruction type determines opcode-generic, instruction-type-specific
+///   function. Name: `i_param_INSTRUCTION_TYPE`
+/// - Per-instructions there will be generated opcode-specific functions calling the generic ones
+/// - Operand types
+///     - R: Register (u8)
+///     - I: Immediate
+///     - L: Memory load / store size (u16)
+///     - Other types are identity-mapped
+///
+/// # BRC special-case
+/// BRC's 3rd operand is plain byte, not a register. Encoding is the same, but for some cases it may matter.
+///
+/// Please, if you distinguish in your API between byte and register, special case this one.
+///
+/// Sorry for that :(
+#[macro_export]
+macro_rules! invoke_with_def {
+    ($macro:path) => {
+        $macro!(
+            bbbb(p0: R, p1: R, p2: R, p3: R)
+                => [DIR, DIRF, FMAF],
+            bbb(p0: R, p1: R, p2: R)
+                => [ADD, SUB, MUL, AND, OR, XOR, SL, SR, SRS, CMP, CMPU, BRC, ADDF, SUBF, MULF],
+            bbdh(p0: R, p1: R, p2: I, p3: L)
+                => [LD, ST],
+            bbd(p0: R, p1: R, p2: I)
+                => [ADDI, MULI, ANDI, ORI, XORI, CMPI, CMPUI, BMC, JAL, JEQ, JNE, JLT, JGT, JLTU,
+                    JGTU, ADDFI, MULFI],
+            bbw(p0: R, p1: R, p2: u32)
+                => [SLI, SRI, SRSI],
+            bb(p0: R, p1: R)
+                => [NEG, NOT, CP, SWA, NEGF, ITF, FTI],
+            bd(p0: R, p1: I)
+                => [LI],
+            n()
+                => [UN, TX, NOP, ECALL],
+        );
+    };
+}
+
+invoke_with_def!(gen_valider::gen_valider);
+
+constmod!(pub opcode(u8) {
+    //! Opcode constant module
+
+    UN  = 0, "N; Raises a trap";
+    TX  = 1, "N; Terminate execution";
+    NOP = 2, "N; Do nothing";
+
+    ADD  = 3,  "BBB; #0 ← #1 + #2";
+    SUB  = 4,  "BBB; #0 ← #1 - #2";
+    MUL  = 5,  "BBB; #0 ← #1 × #2";
+    AND  = 6,  "BBB; #0 ← #1 & #2";
+    OR   = 7,  "BBB; #0 ← #1 | #2";
+    XOR  = 8,  "BBB; #0 ← #1 ^ #2";
+    SL   = 9,  "BBB; #0 ← #1 « #2";
+    SR   = 10,  "BBB; #0 ← #1 » #2";
+    SRS  = 11,  "BBB; #0 ← #1 » #2 (signed)";
+    CMP  = 12, "BBB; #0 ← #1 <=> #2";
+    CMPU = 13, "BBB; #0 ← #1 <=> #2 (unsigned)";
+    DIR  = 14, "BBBB; #0 ← #2 / #3, #1 ← #2 % #3";
+    NEG  = 15, "BB; #0 ← -#1";
+    NOT  = 16, "BB; #0 ← !#1";
+
+    ADDI  = 17, "BBD; #0 ← #1 + imm #2";
+    MULI  = 18, "BBD; #0 ← #1 × imm #2";
+    ANDI  = 19, "BBD; #0 ← #1 & imm #2";
+    ORI   = 20, "BBD; #0 ← #1 | imm #2";
+    XORI  = 21, "BBD; #0 ← #1 ^ imm #2";
+    SLI   = 22, "BBW; #0 ← #1 « imm #2";
+    SRI   = 23, "BBW; #0 ← #1 » imm #2";
+    SRSI  = 24, "BBW; #0 ← #1 » imm #2 (signed)";
+    CMPI  = 25, "BBD; #0 ← #1 <=> imm #2";
+    CMPUI = 26, "BBD; #0 ← #1 <=> imm #2 (unsigned)";
+
+    CP  = 27, "BB; Copy #0 ← #1";
+    SWA = 28, "BB; Swap #0 and #1";
+    LI  = 29, "BD; #0 ← imm #1";
+    LD  = 30, "BBDB; #0 ← [#1 + imm #3], imm #4 bytes, overflowing";
+    ST  = 31, "BBDB; [#1 + imm #3] ← #0, imm #4 bytes, overflowing";
+    BMC = 32, "BBD; [#0] ← [#1], imm #2 bytes";
+    BRC = 33, "BBB; #0 ← #1, imm #2 registers";
+
+    JMP   = 34, "D; Unconditional, non-linking absolute jump";
+    JAL   = 35, "BD;  Copy PC to #0 and unconditional jump [#1 + imm #2]";
+    JEQ   = 36, "BBD; if #0 = #1 → jump imm #2";
+    JNE   = 37, "BBD; if #0 ≠ #1 → jump imm #2";
+    JLT   = 38, "BBD; if #0 < #1 → jump imm #2";
+    JGT   = 39, "BBD; if #0 > #1 → jump imm #2";
+    JLTU  = 40, "BBD; if #0 < #1 → jump imm #2 (unsigned)";
+    JGTU  = 41, "BBD; if #0 > #1 → jump imm #2 (unsigned)";
+    ECALL = 42, "N; Issue system call";
+
+    ADDF = 43, "BBB; #0 ← #1 +. #2";
+    SUBF = 44, "BBB; #0 ← #1 -. #2";
+    MULF = 45, "BBB; #0 ← #1 +. #2";
+    DIRF = 46, "BBBB; #0 ← #2 / #3, #1 ← #2 % #3";
+    FMAF = 47, "BBBB; #0 ← (#1 * #2) + #3";
+    NEGF = 48, "BB; #0 ← -#1";
+    ITF  = 49, "BB; #0 ← #1 as float";
+    FTI  = 50, "BB; #0 ← #1 as int";
+
+    ADDFI = 51, "BBD; #0 ← #1 +. imm #2";
+    MULFI = 52, "BBD; #0 ← #1 *. imm #2";
+});
+
+#[repr(packed)]
+pub struct ParamBBBB(pub u8, pub u8, pub u8, pub u8);
+#[repr(packed)]
+pub struct ParamBBB(pub u8, pub u8, pub u8);
+#[repr(packed)]
+pub struct ParamBBDH(pub u8, pub u8, pub u64, pub u16);
+#[repr(packed)]
+pub struct ParamBBD(pub u8, pub u8, pub u64);
+#[repr(packed)]
+pub struct ParamBBW(pub u8, pub u8, pub u32);
+#[repr(packed)]
+pub struct ParamBB(pub u8, pub u8);
+#[repr(packed)]
+pub struct ParamBD(pub u8, pub u64);
+
+/// # Safety
+/// Has to be valid to be decoded from bytecode.
+pub unsafe trait ProgramVal {}
+unsafe impl ProgramVal for ParamBBBB {}
+unsafe impl ProgramVal for ParamBBB {}
+unsafe impl ProgramVal for ParamBBDH {}
+unsafe impl ProgramVal for ParamBBD {}
+unsafe impl ProgramVal for ParamBBW {}
+unsafe impl ProgramVal for ParamBB {}
+unsafe impl ProgramVal for ParamBD {}
+unsafe impl ProgramVal for u64 {}
+unsafe impl ProgramVal for u8 {} // Opcode
+unsafe impl ProgramVal for () {}

hbvm/Cargo.toml

@@ -3,8 +3,13 @@ name = "hbvm"
 version = "0.1.0"
 edition = "2021"
 
-# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+[profile.release]
+lto = true
+
+[features]
+default = ["alloc"]
+alloc   = []
+nightly = []
 
 [dependencies]
-log = "*"
-hashbrown = "0.13.2"
+hbbytecode.path = "../hbbytecode"

hbvm/fuzz/.gitignore

@@ -0,0 +1,5 @@
+target
+artifacts
+corpus
+coverage
+Cargo.lock

hbvm/fuzz/Cargo.toml

@@ -0,0 +1,30 @@
+[package]
+name = "hbvm-fuzz"
+version = "0.0.0"
+publish = false
+edition = "2021"
+
+[package.metadata]
+cargo-fuzz = true
+
+[dependencies]
+libfuzzer-sys = "0.4"
+
+[dependencies.hbvm]
+path = ".."
+
+[dependencies.hbbytecode]
+path = "../../hbbytecode"
+
+# Prevent this from interfering with workspaces
+[workspace]
+members = ["."]
+
+[profile.release]
+debug = 1
+
+[[bin]]
+name = "vm"
+path = "fuzz_targets/vm.rs"
+test = false
+doc = false

hbvm/fuzz/fuzz_targets/vm.rs

@@ -0,0 +1,85 @@
+#![no_main]
+
+use {
+    hbbytecode::valider::validate,
+    hbvm::{
+        mem::{
+            softpaging::{
+                paging::{PageTable, Permission},
+                HandlePageFault, PageSize, SoftPagedMem,
+            },
+            Address, MemoryAccessReason,
+        },
+        Vm,
+    },
+    libfuzzer_sys::fuzz_target,
+};
+
+fuzz_target!(|data: &[u8]| {
+    if validate(data).is_ok() {
+        let mut vm = unsafe {
+            Vm::<_, 16384>::new(
+                SoftPagedMem::<_, true> {
+                    pf_handler: TestTrapHandler,
+                    program:    data,
+                    root_pt:    Box::into_raw(Default::default()),
+                    icache:     Default::default(),
+                },
+                Address::new(4),
+            )
+        };
+
+        // Alloc and map some memory
+        let pages = [
+            alloc_and_map(&mut vm.memory, 0),
+            alloc_and_map(&mut vm.memory, 4096),
+        ];
+
+        // Run VM
+        let _ = vm.run();
+
+        // Unmap and dealloc the memory
+        for (i, page) in pages.into_iter().enumerate() {
+            unmap_and_dealloc(&mut vm.memory, page, i as u64 * 4096);
+        }
+
+        let _ = unsafe { Box::from_raw(vm.memory.root_pt) };
+    }
+});
+
+fn alloc_and_map(memory: &mut SoftPagedMem<TestTrapHandler>, at: u64) -> *mut u8 {
+    let ptr = Box::into_raw(Box::<Page>::default()).cast();
+    unsafe {
+        memory
+            .map(ptr, Address::new(at), Permission::Write, PageSize::Size4K)
+            .unwrap()
+    };
+    ptr
+}
+
+fn unmap_and_dealloc(memory: &mut SoftPagedMem<TestTrapHandler>, ptr: *mut u8, from: u64) {
+    memory.unmap(Address::new(from)).unwrap();
+    let _ = unsafe { Box::from_raw(ptr.cast::<Page>()) };
+}
+
+#[repr(align(4096))]
+struct Page([u8; 4096]);
+impl Default for Page {
+    fn default() -> Self {
+        unsafe { std::mem::MaybeUninit::zeroed().assume_init() }
+    }
+}
+
+struct TestTrapHandler;
+impl HandlePageFault for TestTrapHandler {
+    fn page_fault(
+        &mut self,
+        _: MemoryAccessReason,
+        _: &mut PageTable,
+        _: Address,
+        _: PageSize,
+        _: *mut u8,
+    ) -> bool {
+        false
+    }
+}

hbvm/src/bmc.rs

@@ -0,0 +1,135 @@
+//! Block memory copier state machine
+
+use {
+    super::{mem::MemoryAccessReason, Memory, VmRunError},
+    crate::mem::Address,
+    core::{mem::MaybeUninit, task::Poll},
+};
+
+/// Buffer size (defaults to 4 KiB, a smallest page size on most platforms)
+const BUF_SIZE: usize = 4096;
+
+/// Buffer of possibly uninitialised bytes, aligned to [`BUF_SIZE`]
+#[repr(align(4096))]
+struct AlignedBuf([MaybeUninit<u8>; BUF_SIZE]);
+
+/// State for block memory copy
+pub struct BlockCopier {
+    /// Source address
+    src:       Address,
+    /// Destination address
+    dst:       Address,
+    /// How many buffer sizes to copy?
+    n_buffers: usize,
+    /// …and what remainds after?
+    rem:       usize,
+}
+
+impl BlockCopier {
+    /// Construct a new one
+    #[inline]
+    pub fn new(src: Address, dst: Address, count: usize) -> Self {
+        Self {
+            src,
+            dst,
+            n_buffers: count / BUF_SIZE,
+            rem: count % BUF_SIZE,
+        }
+    }
+
+    /// Copy one block
+    ///
+    /// # Safety
+    /// - Same as for [`Memory::load`] and [`Memory::store`]
+    pub unsafe fn poll(&mut self, memory: &mut impl Memory) -> Poll<Result<(), BlkCopyError>> {
+        // Safety: Assuming uninit of array of MaybeUninit is sound
+        let mut buf = AlignedBuf(MaybeUninit::uninit().assume_init());
+
+        // We have at least one buffer size to copy
+        if self.n_buffers != 0 {
+            if let Err(e) = act(
+                memory,
+                self.src,
+                self.dst,
+                buf.0.as_mut_ptr().cast(),
+                BUF_SIZE,
+            ) {
+                return Poll::Ready(Err(e));
+            }
+
+            // Bump source and destination address
+            self.src += BUF_SIZE;
+            self.dst += BUF_SIZE;
+
+            self.n_buffers -= 1;
+
+            return if self.n_buffers + self.rem == 0 {
+                // If there is nothing left, we are done
+                Poll::Ready(Ok(()))
+            } else {
+                // Otherwise let's advice to run it again
+                Poll::Pending
+            };
+        }
+
+        if self.rem != 0 {
+            if let Err(e) = act(
+                memory,
+                self.src,
+                self.dst,
+                buf.0.as_mut_ptr().cast(),
+                self.rem,
+            ) {
+                return Poll::Ready(Err(e));
+            }
+        }
+
+        Poll::Ready(Ok(()))
+    }
+}
+
+/// Load to buffer and store from buffer
+#[inline]
+unsafe fn act(
+    memory: &mut impl Memory,
+    src: Address,
+    dst: Address,
+    buf: *mut u8,
+    count: usize,
+) -> Result<(), BlkCopyError> {
+    // Load to buffer
+    memory
+        .load(src, buf, count)
+        .map_err(|super::mem::LoadError(addr)| BlkCopyError {
+            access_reason: MemoryAccessReason::Load,
+            addr,
+        })?;
+
+    // Store from buffer
+    memory
+        .store(dst, buf, count)
+        .map_err(|super::mem::StoreError(addr)| BlkCopyError {
+            access_reason: MemoryAccessReason::Store,
+            addr,
+        })?;
+
+    Ok(())
+}
+
+/// Error occured when copying a block of memory
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub struct BlkCopyError {
+    /// Kind of access
+    access_reason: MemoryAccessReason,
+    /// VM Address
+    addr: Address,
+}
+
+impl From<BlkCopyError> for VmRunError {
+    fn from(value: BlkCopyError) -> Self {
+        match value.access_reason {
+            MemoryAccessReason::Load => Self::LoadAccessEx(value.addr),
+            MemoryAccessReason::Store => Self::StoreAccessEx(value.addr),
+        }
+    }
+}

hbvm/src/bytecode/mod.rs

@@ -1,2 +0,0 @@
-pub mod ops;
-pub mod types;

hbvm/src/bytecode/ops.rs

@@ -1,68 +0,0 @@
-#[repr(u8)]
-pub enum Operations {
-    NOP = 0,
-
-    ADD = 1,
-    SUB = 2,
-    MUL = 3,
-    DIV = 4,
-    MOD = 5,
-
-    AND = 6,
-    OR = 7,
-    XOR = 8,
-    NOT = 9,
-
-    // LOADs a memory address/constant into a register
-    LOAD = 15,
-    // STOREs a register/constant into a memory address
-    STORE = 16,
-
-    MapPage = 17,
-    UnmapPage = 18,
-
-    // SHIFT LEFT 16 A0
-    Shift = 20,
-
-    JUMP = 100,
-    JumpCond = 101,
-    RET = 103,
-
-    EnviromentCall = 255,
-}
-
-pub enum PageMapTypes {
-    // Have the host make a new VMPage
-    VMPage = 0,
-    // Ask the host to map a RealPage into memory
-    RealPage = 1,
-}
-
-pub enum MathOpSubTypes {
-    Unsigned = 0,
-    Signed = 1,
-    FloatingPoint = 2,
-}
-
-pub enum MathOpSides {
-    RegisterConstant = 0,
-    RegisterRegister = 1,
-    ConstantConstant = 2,
-    ConstantRegister = 3,
-}
-
-pub enum RWSubTypes {
-    AddrToReg = 0,
-    RegToAddr,
-    ConstToReg,
-    ConstToAddr,
-}
-
-pub enum JumpConditionals {
-    Equal = 0,
-    NotEqual = 1,
-    LessThan = 2,
-    LessThanOrEqualTo = 3,
-    GreaterThan = 4,
-    GreaterThanOrEqualTo = 5,
-}

hbvm/src/bytecode/types.rs

@@ -1,9 +0,0 @@
-
-pub const CONST_U8: u8 = 0x00;
-pub const CONST_I8: i8 = 0x01;
-
-pub const CONST_U64: u8 = 0x02;
-pub const CONST_I64: u8 = 0x03;
-pub const CONST_F64: u8 = 0x04;
-
-pub const ADDRESS: u8 = 0x05;

hbvm/src/engine/call_stack.rs

@@ -1,6 +0,0 @@
-use alloc::vec::Vec;
-
-pub type CallStack = Vec<FnCall>;
-pub struct FnCall {
-    pub ret: usize,
-}

hbvm/src/engine/config.rs

@@ -1,13 +0,0 @@
-pub struct EngineConfig {
-    pub call_stack_depth: usize,
-    pub quantum: u32,
-}
-
-impl EngineConfig {
-    pub fn default() -> Self {
-        Self {
-            call_stack_depth: 32,
-            quantum: 0,
-        }
-    }
-}

hbvm/src/engine/enviroment_calls.rs

@@ -1,3 +0,0 @@
-use super::Engine;
-
-pub type EnviromentCall = fn(&mut Engine) -> Result<&mut Engine, u64>;

hbvm/src/engine/mod.rs

@@ -1,100 +0,0 @@
-pub mod call_stack;
-pub mod config;
-pub mod enviroment_calls;
-pub mod regs;
-#[cfg(test)]
-pub mod tests;
-
-use {
-    self::call_stack::CallStack,
-    crate::{memory, HaltStatus, RuntimeErrors},
-    alloc::vec::Vec,
-    config::EngineConfig,
-    log::trace,
-    regs::Registers,
-};
-// pub const PAGE_SIZE: usize = 8192;
-
-pub struct RealPage {
-    pub ptr: *mut u8,
-}
-
-#[derive(Debug, Clone, Copy)]
-pub struct VMPage {
-    pub data: [u8; 8192],
-}
-impl VMPage {
-    pub fn new() -> Self {
-        Self {
-            data: [0; 4096 * 2],
-        }
-    }
-}
-
-pub enum Page {
-    VMPage(VMPage),
-    RealPage(RealPage),
-}
-impl Page {
-    pub fn data(&self) -> [u8; 4096 * 2] {
-        match self {
-            Page::VMPage(vmpage) => vmpage.data,
-            Page::RealPage(_) => {
-                unimplemented!("Memmapped hw page not yet supported")
-            }
-        }
-    }
-}
-
-pub fn empty_enviroment_call(engine: &mut Engine) -> Result<&mut Engine, u64> {
-    trace!("Registers {:?}", engine.registers);
-    Err(0)
-}
-
-pub struct Engine {
-    pub index:     usize,
-    pub program:   Vec<u8>,
-    pub registers: Registers,
-    pub config:    EngineConfig,
-
-    /// BUG: This DOES NOT account for overflowing
-    pub last_timer_count: u32,
-    pub timer_callback: Option<fn() -> u32>,
-    pub memory: memory::Memory,
-    pub enviroment_call_table: [Option<EnviromentCall>; 256],
-    pub call_stack: CallStack,
-}
-use crate::engine::enviroment_calls::EnviromentCall;
-impl Engine {
-    pub fn set_timer_callback(&mut self, func: fn() -> u32) {
-        self.timer_callback = Some(func);
-    }
-    pub fn set_register(&mut self, register: u8, value: u64) {}
-}
-
-impl Engine {
-    pub fn new(program: Vec<u8>) -> Self {
-        let mut mem = memory::Memory::new();
-        for (addr, byte) in program.clone().into_iter().enumerate() {
-            let _ = mem.set_addr8(addr as u64, byte);
-        }
-        trace!("{:?}", mem.read_addr8(0));
-        let ecall_table: [Option<EnviromentCall>; 256] = [None; 256];
-        Self {
-            index: 0,
-            program,
-            registers: Registers::new(),
-            config: EngineConfig::default(),
-            last_timer_count: 0,
-            timer_callback: None,
-            enviroment_call_table: ecall_table,
-            memory: mem,
-            call_stack: Vec::new(),
-        }
-    }
-
-    pub fn dump(&self) {}
-    pub fn run(&mut self) -> Result<HaltStatus, RuntimeErrors> {
-        Ok(HaltStatus::Halted)
-    }
-}

hbvm/src/engine/regs.rs

@@ -1,32 +0,0 @@
-#[rustfmt::skip]
-#[derive(Debug, Clone, Copy)]
-pub struct Registers {
-    pub a0: u64, pub b0: u64, pub c0: u64, pub d0: u64, pub e0: u64, pub f0: u64,
-    pub a1: u64, pub b1: u64, pub c1: u64, pub d1: u64, pub e1: u64, pub f1: u64,
-    pub a2: u64, pub b2: u64, pub c2: u64, pub d2: u64, pub e2: u64, pub f2: u64,
-    pub a3: u64, pub b3: u64, pub c3: u64, pub d3: u64, pub e3: u64, pub f3: u64,
-    pub a4: u64, pub b4: u64, pub c4: u64, pub d4: u64, pub e4: u64, pub f4: u64,
-    pub a5: u64, pub b5: u64, pub c5: u64, pub d5: u64, pub e5: u64, pub f5: u64,
-    pub a6: u64, pub b6: u64, pub c6: u64, pub d6: u64, pub e6: u64, pub f6: u64,
-    pub a7: u64, pub b7: u64, pub c7: u64, pub d7: u64, pub e7: u64, pub f7: u64,
-    pub a8: u64, pub b8: u64, pub c8: u64, pub d8: u64, pub e8: u64, pub f8: u64,
-    pub a9: u64, pub b9: u64, pub c9: u64, pub d9: u64, pub e9: u64, pub f9: u64,
-}
-
-impl Registers {
-    #[rustfmt::skip]
-    pub fn new() -> Self{
-        Self {
-            a0: 0, b0: 0, c0: 0, d0: 0, e0: 0, f0: 0,
-            a1: 0, b1: 0, c1: 0, d1: 0, e1: 0, f1: 0,
-            a2: 0, b2: 0, c2: 0, d2: 0, e2: 0, f2: 0,
-            a3: 0, b3: 0, c3: 0, d3: 0, e3: 0, f3: 0,
-            a4: 0, b4: 0, c4: 0, d4: 0, e4: 0, f4: 0,
-            a5: 0, b5: 0, c5: 0, d5: 0, e5: 0, f5: 0,
-            a6: 0, b6: 0, c6: 0, d6: 0, e6: 0, f6: 0,
-            a7: 0, b7: 0, c7: 0, d7: 0, e7: 0, f7: 0,
-            a8: 0, b8: 0, c8: 0, d8: 0, e8: 0, f8: 0,
-            a9: 0, b9: 0, c9: 0, d9: 0, e9: 0, f9: 0,
-        }
-    }
-}

hbvm/src/engine/tests/mod.rs

@@ -1,125 +0,0 @@
-use {
-    super::Engine,
-    crate::{HaltStatus, RuntimeErrors},
-    alloc::vec,
-    RuntimeErrors::*,
-};
-
-#[test]
-fn invalid_program() {
-    let prog = vec![1, 0];
-    let mut eng = Engine::new(prog);
-    let ret = eng.run();
-    assert_eq!(ret, Err(InvalidOpcodePair(1, 0)));
-}
-
-#[test]
-fn empty_program() {
-    let prog = vec![];
-    let mut eng = Engine::new(prog);
-    let ret = eng.run();
-    assert_eq!(ret, Ok(HaltStatus::Halted));
-}
-
-#[test]
-fn max_quantum_reached() {
-    let prog = vec![0, 0, 0, 0];
-    let mut eng = Engine::new(prog);
-    eng.set_timer_callback(|| {
-        return 1;
-    });
-    eng.config.quantum = 1;
-    let ret = eng.run();
-    assert_eq!(ret, Ok(HaltStatus::Running));
-}
-
-#[test]
-fn jump_out_of_bounds() {
-    use crate::bytecode::ops::Operations::JUMP;
-    let prog = vec![JUMP as u8, 0, 0, 0, 0, 0, 0, 1, 0];
-    let mut eng = Engine::new(prog);
-    let ret = eng.run();
-    assert_eq!(ret, Err(InvalidJumpAddress(256)));
-}
-
-#[test]
-fn invalid_system_call() {
-    let prog = vec![255, 0];
-    let mut eng = Engine::new(prog);
-    let ret = eng.run();
-    assert_eq!(ret, Err(InvalidSystemCall(0)));
-}
-
-#[test]
-fn add_u8() {
-    use crate::bytecode::ops::{MathOpSides::ConstantConstant, Operations::ADD};
-
-    let prog = vec![ADD as u8, ConstantConstant as u8, 100, 98, 0xA0];
-    let mut eng = Engine::new(prog);
-    let _ = eng.run();
-    assert_eq!(eng.registers.a0, 2);
-}
-
-#[test]
-fn sub_u8() {
-    use crate::bytecode::ops::Operations::SUB;
-
-    let prog = vec![SUB as u8];
-    let mut eng = Engine::new(prog);
-    let _ = eng.run();
-    assert_eq!(eng.registers.a0, 1);
-}
-#[test]
-fn mul_u8() {
-    use crate::bytecode::ops::{MathOpSides::ConstantConstant, Operations::MUL};
-
-    let prog = vec![MUL as u8, ConstantConstant as u8, 1, 2, 0xA0];
-    let mut eng = Engine::new(prog);
-    let _ = eng.run();
-    assert_eq!(eng.registers.a0, 2);
-}
-
-#[test]
-fn div_u8() {
-    use crate::bytecode::ops::Operations::DIV;
-
-    let prog = vec![DIV as u8];
-    let mut eng = Engine::new(prog);
-    let _ = eng.run();
-    assert_eq!(eng.registers.a0, 2);
-}
-
-#[test]
-fn set_register() {
-    let prog = alloc::vec![];
-    let mut eng = Engine::new(prog);
-    eng.set_register(0xA0, 1);
-    assert_eq!(eng.registers.a0, 1);
-}
-
-#[test]
-fn load_u8() {
-    use crate::bytecode::ops::{Operations::LOAD, RWSubTypes::AddrToReg};
-
-    let prog = vec![LOAD as u8, AddrToReg as u8, 0, 0, 0, 0, 0, 0, 1, 0, 0xA0];
-    let mut eng = Engine::new(prog);
-    let ret = eng.memory.set_addr8(256, 1);
-    assert_eq!(ret, Ok(()));
-    let _ = eng.run();
-    assert_eq!(eng.registers.a0, 1);
-}
-#[test]
-fn set_memory_8() {
-    let prog = vec![];
-    let mut eng = Engine::new(prog);
-    let ret = eng.memory.set_addr8(256, 1);
-    assert_eq!(ret, Ok(()));
-}
-
-#[test]
-fn set_memory_64() {
-    let prog = vec![];
-    let mut eng = Engine::new(prog);
-    let ret = eng.memory.set_addr64(256, 1);
-    assert_eq!(ret, Ok(()));
-}

hbvm/src/lib.rs

@@ -1,23 +1,108 @@
+//! HoleyBytes Virtual Machine
+//!
+//! # Alloc feature
+//! - Enabled by default
+//! - Provides mapping / unmapping, as well as [`Default`] and [`Drop`]
+//!   implementations for soft-paged memory implementation
+
+// # General safety notice:
+// - Validation has to assure there is 256 registers (r0 - r255)
+// - Instructions have to be valid as specified (values and sizes)
+// - Mapped pages should be at least 4 KiB
+
 #![no_std]
+#![cfg_attr(feature = "nightly", feature(fn_align))]
+#![warn(missing_docs)]
+
+use mem::{Memory, Address};
+
+#[cfg(feature = "alloc")]
 extern crate alloc;
 
-pub mod bytecode;
-pub mod engine;
-pub mod memory;
+pub mod mem;
+pub mod value;
 
-#[derive(Debug, PartialEq)]
-pub enum RuntimeErrors {
-    InvalidOpcodePair(u8, u8),
-    RegisterTooSmall,
-    HostError(u64),
-    PageNotMapped(u64),
-    InvalidJumpAddress(u64),
-    InvalidSystemCall(u8),
+mod bmc;
+mod vmrun;
+mod utils;
+
+use {bmc::BlockCopier, value::Value};
+
+/// HoleyBytes Virtual Machine
+pub struct Vm<Mem, const TIMER_QUOTIENT: usize> {
+    /// Holds 256 registers
+    ///
+    /// Writing to register 0 is considered undefined behaviour
+    /// in terms of HoleyBytes program execution
+    pub registers: [Value; 256],
+
+    /// Memory implementation
+    pub memory: Mem,
+
+    /// Program counter
+    pub pc: Address,
+
+    /// Program timer
+    timer: usize,
+
+    /// Saved block copier
+    copier: Option<BlockCopier>,
 }
 
-// If you solve the halting problem feel free to remove this
-#[derive(PartialEq, Debug)]
-pub enum HaltStatus {
-    Halted,
-    Running,
+impl<Mem, const TIMER_QUOTIENT: usize> Vm<Mem, TIMER_QUOTIENT>
+where
+    Mem: Memory,
+{
+    /// Create a new VM with program and trap handler
+    ///
+    /// # Safety
+    /// Program code has to be validated
+    pub unsafe fn new(memory: Mem, entry: Address) -> Self {
+        Self {
+            registers: [Value::from(0_u64); 256],
+            memory,
+            pc: entry,
+            timer: 0,
+            copier: None,
+        }
+    }
+}
+
+/// Virtual machine halt error
+#[derive(Copy, Clone, Debug, PartialEq, Eq)]
+#[repr(u8)]
+pub enum VmRunError {
+    /// Tried to execute invalid instruction
+    InvalidOpcode(u8),
+
+    /// Unhandled load access exception
+    LoadAccessEx(Address),
+
+    /// Unhandled instruction load access exception
+    ProgramFetchLoadEx(Address),
+
+    /// Unhandled store access exception
+    StoreAccessEx(Address),
+
+    /// Register out-of-bounds access
+    RegOutOfBounds,
+
+    /// Address out-of-bounds
+    AddrOutOfBounds,
+
+    /// Reached unreachable code
+    Unreachable,
+}
+
+/// Virtual machine halt ok
+#[derive(Copy, Clone, Debug, PartialEq, Eq)]
+pub enum VmRunOk {
+    /// Program has eached its end
+    End,
+
+    /// Program was interrupted by a timer
+    Timer,
+
+    /// Environment call
+    Ecall,
 }

hbvm/src/main.rs

@@ -1,31 +1,83 @@
-use hbvm::{
-    bytecode::ops::{Operations::*, RWSubTypes::*},
-    engine::Engine,
-    RuntimeErrors,
+use hbvm::mem::Address;
+
+use {
+    hbbytecode::valider::validate,
+    hbvm::{
+        mem::{
+            softpaging::{paging::PageTable, HandlePageFault, PageSize, SoftPagedMem},
+            MemoryAccessReason,
+        },
+        Vm,
+    },
+    std::io::{stdin, Read},
 };
 
-fn main() -> Result<(), RuntimeErrors> {
-    // TODO: Grab program from cmdline
-    #[rustfmt::skip]
-        let prog: Vec<u8> = vec![
-            NOP as u8,            
-            JUMP as u8, 0, 0, 0, 0, 0, 0, 0, 0,
-        ];
+fn main() -> Result<(), Box<dyn std::error::Error>> {
+    let mut prog = vec![];
+    stdin().read_to_end(&mut prog)?;
 
-    let mut eng = Engine::new(prog);
-    // eng.set_timer_callback(time);
-    eng.enviroment_call_table[10] = Some(print_fn);
-    eng.run()?;
-    eng.dump();
-    println!("{:#?}", eng.registers);
+    if let Err(e) = validate(&prog) {
+        eprintln!("Program validation error: {e:?}");
+        return Ok(());
+    } else {
+        unsafe {
+            let mut vm = Vm::<_, 0>::new(
+                SoftPagedMem::<_, true> {
+                    pf_handler: TestTrapHandler,
+                    program:    &prog,
+                    root_pt:    Box::into_raw(Default::default()),
+                    icache:     Default::default(),
+                },
+                Address::new(4),
+            );
+            let data = {
+                let ptr = std::alloc::alloc_zeroed(std::alloc::Layout::from_size_align_unchecked(
+                    4096, 4096,
+                ));
+                if ptr.is_null() {
+                    panic!("Alloc error tbhl");
+                }
+                ptr
+            };
 
+            vm.memory
+                .map(
+                    data,
+                    Address::new(8192),
+                    hbvm::mem::softpaging::paging::Permission::Write,
+                    PageSize::Size4K,
+                )
+                .unwrap();
+
+            println!("Program interrupt: {:?}", vm.run());
+            println!("{:?}", vm.registers);
+
+            std::alloc::dealloc(
+                data,
+                std::alloc::Layout::from_size_align_unchecked(4096, 4096),
+            );
+            vm.memory.unmap(Address::new(8192)).unwrap();
+            let _ = Box::from_raw(vm.memory.root_pt);
+        }
+    }
     Ok(())
 }
 
 pub fn time() -> u32 {
     9
 }
-pub fn print_fn(engine: &mut Engine) -> Result<&mut Engine, u64> {
-    println!("hello");
-    Ok(engine)
+
+#[derive(Default)]
+struct TestTrapHandler;
+impl HandlePageFault for TestTrapHandler {
+    fn page_fault(
+        &mut self,
+        _: MemoryAccessReason,
+        _: &mut PageTable,
+        _: Address,
+        _: PageSize,
+        _: *mut u8,
+    ) -> bool {
+        false
+    }
 }

hbvm/src/mem/addr.rs

@@ -0,0 +1,110 @@
+//! Virtual(?) memory address
+
+use core::{fmt::Debug, ops};
+
+use crate::utils::impl_display;
+
+/// Memory address
+#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
+pub struct Address(u64);
+impl Address {
+    /// A null address
+    pub const NULL: Self = Self(0);
+
+    /// Saturating integer addition. Computes self + rhs, saturating at the numeric bounds instead of overflowing.
+    #[inline]
+    pub fn saturating_add<T: AddressOp>(self, rhs: T) -> Self {
+        Self(self.0.saturating_add(rhs.cast_u64()))
+    }
+
+    /// Saturating integer subtraction. Computes self - rhs, saturating at the numeric bounds instead of overflowing.
+    #[inline]
+    pub fn saturating_sub<T: AddressOp>(self, rhs: T) -> Self {
+        Self(self.0.saturating_sub(rhs.cast_u64()))
+    }
+
+    /// Cast or if smaller, truncate to [`usize`]
+    pub fn truncate_usize(self) -> usize {
+        self.0 as _
+    }
+
+    /// Get inner value
+    #[inline(always)]
+    pub fn get(self) -> u64 {
+        self.0
+    }
+
+    /// Construct new address
+    #[inline(always)]
+    pub fn new(val: u64) -> Self {
+        Self(val)
+    }
+
+    /// Do something with inner value
+    #[inline(always)]
+    pub fn map(self, f: impl Fn(u64) -> u64) -> Self {
+        Self(f(self.0))
+    }
+}
+
+impl_display!(for Address =>
+    |Address(a)| "{a:0x}"
+);
+
+impl<T: AddressOp> ops::Add<T> for Address {
+    type Output = Self;
+
+    #[inline]
+    fn add(self, rhs: T) -> Self::Output {
+        Self(self.0.wrapping_add(rhs.cast_u64()))
+    }
+}
+
+impl<T: AddressOp> ops::Sub<T> for Address {
+    type Output = Self;
+
+    #[inline]
+    fn sub(self, rhs: T) -> Self::Output {
+        Self(self.0.wrapping_sub(rhs.cast_u64()))
+    }
+}
+
+impl<T: AddressOp> ops::AddAssign<T> for Address {
+    fn add_assign(&mut self, rhs: T) {
+        self.0 = self.0.wrapping_add(rhs.cast_u64())
+    }
+}
+
+impl<T: AddressOp> ops::SubAssign<T> for Address {
+    fn sub_assign(&mut self, rhs: T) {
+        self.0 = self.0.wrapping_sub(rhs.cast_u64())
+    }
+}
+
+impl From<Address> for u64 {
+    #[inline(always)]
+    fn from(value: Address) -> Self {
+        value.0
+    }
+}
+
+impl Debug for Address {
+    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        write!(f, "[{:0x}]", self.0)
+    }
+}
+
+/// Can perform address operations with
+pub trait AddressOp {
+    /// Cast to u64, truncating or extending
+    fn cast_u64(self) -> u64;
+}
+
+macro_rules! impl_address_ops(($($ty:ty),* $(,)?) => {
+    $(impl AddressOp for $ty {
+        #[inline(always)]
+        fn cast_u64(self) -> u64 { self as _ }
+    })*
+});
+
+impl_address_ops!(u8, u16, u32, u64, usize);

hbvm/src/mem/mod.rs

@@ -0,0 +1,86 @@
+//! Memory implementations
+
+pub mod softpaging;
+
+mod addr;
+
+pub use addr::Address;
+
+use {crate::utils::impl_display, hbbytecode::ProgramVal};
+
+/// Load-store memory access
+pub trait Memory {
+    /// Load data from memory on address
+    ///
+    /// # Safety
+    /// - Shall not overrun the buffer
+    unsafe fn load(
+        &mut self,
+        addr: Address,
+        target: *mut u8,
+        count: usize,
+    ) -> Result<(), LoadError>;
+
+    /// Store data to memory on address
+    ///
+    /// # Safety
+    /// - Shall not overrun the buffer
+    unsafe fn store(
+        &mut self,
+        addr: Address,
+        source: *const u8,
+        count: usize,
+    ) -> Result<(), StoreError>;
+
+    /// Read from program memory to execute
+    ///
+    /// # Safety
+    /// - Data read have to be valid
+    unsafe fn prog_read<T: ProgramVal>(&mut self, addr: Address) -> Option<T>;
+
+    /// Read from program memory to exectue
+    ///
+    /// # Safety
+    /// - You have to be really sure that these bytes are there, understand?
+    unsafe fn prog_read_unchecked<T: ProgramVal>(&mut self, addr: Address) -> T;
+}
+
+/// Unhandled load access trap
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub struct LoadError(pub Address);
+impl_display!(for LoadError =>
+    |LoadError(a)| "Load access error at address {a}",
+);
+
+/// Unhandled store access trap
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub struct StoreError(pub Address);
+impl_display!(for StoreError =>
+    |StoreError(a)| "Load access error at address {a}",
+);
+
+/// Reason to access memory
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum MemoryAccessReason {
+    /// Memory was accessed for load (read)
+    Load,
+    /// Memory was accessed for store (write)
+    Store,
+}
+
+impl_display!(for MemoryAccessReason => match {
+    Self::Load  => const "Load";
+    Self::Store => const "Store";
+});
+
+impl From<LoadError> for crate::VmRunError {
+    fn from(value: LoadError) -> Self {
+        Self::LoadAccessEx(value.0)
+    }
+}
+
+impl From<StoreError> for crate::VmRunError {
+    fn from(value: StoreError) -> Self {
+        Self::StoreAccessEx(value.0)
+    }
+}

hbvm/src/mem/softpaging/icache.rs

@@ -0,0 +1,109 @@
+//! Program instruction cache
+
+use crate::mem::Address;
+
+use {
+    super::{lookup::AddrPageLookuper, paging::PageTable, PageSize},
+    core::{
+        mem::{size_of, MaybeUninit},
+        ptr::{copy_nonoverlapping, NonNull},
+    },
+};
+
+/// Instruction cache
+#[derive(Clone, Debug)]
+pub struct ICache {
+    /// Current page address base
+    base: Address,
+    /// Curent page pointer
+    data: Option<NonNull<u8>>,
+    /// Current page size
+    size: PageSize,
+    /// Address mask
+    mask: u64,
+}
+
+impl Default for ICache {
+    fn default() -> Self {
+        Self {
+            base: Address::NULL,
+            data: Default::default(),
+            size: PageSize::Size4K,
+            mask: Default::default(),
+        }
+    }
+}
+
+impl ICache {
+    /// Fetch instruction from cache
+    ///
+    /// # Safety
+    /// `T` should be valid to read from instruction memory
+    pub(super) unsafe fn fetch<T>(
+        &mut self,
+        addr: Address,
+        root_pt: *const PageTable,
+    ) -> Option<T> {
+        let mut ret = MaybeUninit::<T>::uninit();
+
+        let pbase = self
+            .data
+            .or_else(|| self.fetch_page(self.base + self.size, root_pt))?;
+
+        // Get address base
+        let base = addr.map(|x| x & self.mask);
+
+        // Base not matching, fetch anew
+        if base != self.base {
+            self.fetch_page(base, root_pt)?;
+        };
+
+        let offset = addr.get() & !self.mask;
+        let requ_size = size_of::<T>();
+
+        // Page overflow
+        let rem = (offset as usize)
+            .saturating_add(requ_size)
+            .saturating_sub(self.size as _);
+        let first_copy = requ_size.saturating_sub(rem);
+
+        // Copy non-overflowing part
+        copy_nonoverlapping(pbase.as_ptr(), ret.as_mut_ptr().cast::<u8>(), first_copy);
+
+        // Copy overflow
+        if rem != 0 {
+            let pbase = self.fetch_page(self.base + self.size, root_pt)?;
+
+            // Unlikely, unsupported scenario
+            if rem > self.size as _ {
+                return None;
+            }
+
+            copy_nonoverlapping(
+                pbase.as_ptr(),
+                ret.as_mut_ptr().cast::<u8>().add(first_copy),
+                rem,
+            );
+        }
+
+        Some(ret.assume_init())
+    }
+
+    /// Fetch a page
+    unsafe fn fetch_page(&mut self, addr: Address, pt: *const PageTable) -> Option<NonNull<u8>> {
+        let res = AddrPageLookuper::new(addr, 0, pt).next()?.ok()?;
+        if !super::perm_check::executable(res.perm) {
+            return None;
+        }
+
+        (self.size, self.mask) = match res.size {
+            4096 => (PageSize::Size4K, !((1 << 8) - 1)),
+            2097152 => (PageSize::Size2M, !((1 << (8 * 2)) - 1)),
+            1073741824 => (PageSize::Size1G, !((1 << (8 * 3)) - 1)),
+            _ => return None,
+        };
+        self.data = Some(NonNull::new(res.ptr)?);
+        self.base = addr.map(|x| x & self.mask);
+        self.data
+    }
+}

hbvm/src/mem/softpaging/lookup.rs

@@ -0,0 +1,126 @@
+//! Address lookup
+
+use crate::mem::addr::Address;
+
+use super::{
+    addr_extract_index,
+    paging::{PageTable, Permission},
+    PageSize,
+};
+
+/// Good result from address split
+pub struct AddrPageLookupOk {
+    /// Virtual address
+    pub vaddr: Address,
+
+    /// Pointer to the start for perform operation
+    pub ptr: *mut u8,
+
+    /// Size to the end of page / end of desired size
+    pub size: usize,
+
+    /// Page permission
+    pub perm: Permission,
+}
+
+/// Errornous address split result
+pub struct AddrPageLookupError {
+    /// Address of failure
+    pub addr: Address,
+
+    /// Requested page size
+    pub size: PageSize,
+}
+
+/// Address splitter into pages
+pub struct AddrPageLookuper {
+    /// Current address
+    addr: Address,
+
+    /// Size left
+    size: usize,
+
+    /// Page table
+    pagetable: *const PageTable,
+}
+
+impl AddrPageLookuper {
+    /// Create a new page lookuper
+    #[inline]
+    pub const fn new(addr: Address, size: usize, pagetable: *const PageTable) -> Self {
+        Self {
+            addr,
+            size,
+            pagetable,
+        }
+    }
+
+    /// Bump address by size X
+    pub fn bump(&mut self, page_size: PageSize) {
+        self.addr += page_size;
+        self.size = self.size.saturating_sub(page_size as _);
+    }
+}
+
+impl Iterator for AddrPageLookuper {
+    type Item = Result<AddrPageLookupOk, AddrPageLookupError>;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        // The end, everything is fine
+        if self.size == 0 {
+            return None;
+        }
+
+        let (base, perm, size, offset) = 'a: {
+            let mut current_pt = self.pagetable;
+
+            // Walk the page table
+            for lvl in (0..5).rev() {
+                // Get an entry
+                unsafe {
+                    let entry = (*current_pt)
+                        .table
+                        .get_unchecked(addr_extract_index(self.addr, lvl));
+
+                    let ptr = entry.ptr();
+                    match entry.permission() {
+                        // No page → page fault
+                        Permission::Empty => {
+                            return Some(Err(AddrPageLookupError {
+                                addr: self.addr,
+                                size: PageSize::from_lvl(lvl)?,
+                            }))
+                        }
+
+                        // Node → proceed waking
+                        Permission::Node => current_pt = ptr as _,
+
+                        // Leaf → return relevant data
+                        perm => {
+                            break 'a (
+                                // Pointer in host memory
+                                ptr as *mut u8,
+                                perm,
+                                PageSize::from_lvl(lvl)?,
+                                // In-page offset
+                                addr_extract_index(self.addr, lvl),
+                            );
+                        }
+                    }
+                }
+            }
+            return None; // Reached the end (should not happen)
+        };
+
+        // Get available byte count in the selected page with offset
+        let avail = (size as usize).saturating_sub(offset).clamp(0, self.size);
+        self.bump(size);
+
+        Some(Ok(AddrPageLookupOk {
+            vaddr: self.addr,
+            ptr: unsafe { base.add(offset) }, // Return pointer to the start of region
+            size: avail,
+            perm,
+        }))
+    }
+}

hbvm/src/mem/softpaging/mapping.rs

@@ -0,0 +1,166 @@
+//! Automatic memory mapping
+
+use crate::{mem::addr::Address, utils::impl_display};
+
+use {
+    super::{
+        addr_extract_index,
+        paging::{PageTable, Permission, PtEntry, PtPointedData},
+        PageSize, SoftPagedMem,
+    },
+    alloc::boxed::Box,
+};
+
+impl<'p, A, const OUT_PROG_EXEC: bool> SoftPagedMem<'p, A, OUT_PROG_EXEC> {
+    /// Maps host's memory into VM's memory
+    ///
+    /// # Safety
+    /// - Your faith in the gods of UB
+    ///     - Addr-san claims it's fine but who knows is she isn't lying :ferrisSus:
+    ///     - Alright, Miri-sama is also fine with this, who knows why
+    pub unsafe fn map(
+        &mut self,
+        host: *mut u8,
+        target: Address,
+        perm: Permission,
+        pagesize: PageSize,
+    ) -> Result<(), MapError> {
+        let mut current_pt = self.root_pt;
+
+        // Decide on what level depth are we going
+        let lookup_depth = match pagesize {
+            PageSize::Size4K => 0,
+            PageSize::Size2M => 1,
+            PageSize::Size1G => 2,
+        };
+
+        // Walk pagetable levels
+        for lvl in (lookup_depth + 1..5).rev() {
+            let entry = (*current_pt)
+                .table
+                .get_unchecked_mut(addr_extract_index(target, lvl));
+
+            let ptr = entry.ptr();
+            match entry.permission() {
+                // Still not on target and already seeing empty entry?
+                // No worries! Let's create one (allocates).
+                Permission::Empty => {
+                    // Increase children count
+                    (*current_pt).childen += 1;
+
+                    let table = Box::into_raw(Box::new(PtPointedData {
+                        pt: PageTable::default(),
+                    }));
+
+                    core::ptr::write(entry, PtEntry::new(table, Permission::Node));
+                    current_pt = table as _;
+                }
+                // Continue walking
+                Permission::Node => current_pt = ptr as _,
+
+                // There is some entry on place of node
+                _ => return Err(MapError::PageOnNode),
+            }
+        }
+
+        let node = (*current_pt)
+            .table
+            .get_unchecked_mut(addr_extract_index(target, lookup_depth));
+
+        // Check if node is not mapped
+        if node.permission() != Permission::Empty {
+            return Err(MapError::AlreadyMapped);
+        }
+
+        // Write entry
+        (*current_pt).childen += 1;
+        core::ptr::write(node, PtEntry::new(host.cast(), perm));
+
+        Ok(())
+    }
+
+    /// Unmaps pages from VM's memory
+    ///
+    /// If errors, it only means there is no entry to unmap and in most cases
+    /// just should be ignored.
+    pub fn unmap(&mut self, addr: Address) -> Result<(), NothingToUnmap> {
+        let mut current_pt = self.root_pt;
+        let mut page_tables = [core::ptr::null_mut(); 5];
+
+        // Walk page table in reverse
+        for lvl in (0..5).rev() {
+            let entry = unsafe {
+                (*current_pt)
+                    .table
+                    .get_unchecked_mut(addr_extract_index(addr, lvl))
+            };
+
+            let ptr = entry.ptr();
+            match entry.permission() {
+                // Nothing is there, throw an error, not critical!
+                Permission::Empty => return Err(NothingToUnmap),
+                // Node – Save to visited pagetables and continue walking
+                Permission::Node => {
+                    page_tables[lvl as usize] = entry;
+                    current_pt = ptr as _
+                }
+                // Page entry – zero it out!
+                // Zero page entry is completely valid entry with
+                // empty permission - no UB here!
+                _ => unsafe {
+                    core::ptr::write_bytes(entry, 0, 1);
+                    break;
+                },
+            }
+        }
+
+        // Now walk in order visited page tables
+        for entry in page_tables.into_iter() {
+            // Level not visited, skip.
+            if entry.is_null() {
+                continue;
+            }
+
+            unsafe {
+                let children = &mut (*(*entry).ptr()).pt.childen;
+                *children -= 1; // Decrease children count
+
+                // If there are no children, deallocate.
+                if *children == 0 {
+                    let _ = Box::from_raw((*entry).ptr() as *mut PageTable);
+
+                    // Zero visited entry
+                    core::ptr::write_bytes(entry, 0, 1);
+                } else {
+                    break;
+                }
+            }
+        }
+
+        Ok(())
+    }
+}
+
+/// Error mapping
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum MapError {
+    /// Entry was already mapped
+    AlreadyMapped,
+    /// When walking a page entry was
+    /// encounterd.
+    PageOnNode,
+}
+
+impl_display!(for MapError => match {
+    Self::AlreadyMapped => "There is already a page mapped on specified address";
+    Self::PageOnNode    => "There was a page mapped on the way instead of node";
+});
+
+/// There was no entry in page table to unmap
+///
+/// No worry, don't panic, nothing bad has happened,
+/// but if you are 120% sure there should be something,
+/// double-check your addresses.
+#[derive(Clone, Copy, Debug)]
+pub struct NothingToUnmap;
+impl_display!(for NothingToUnmap => "There is no entry to unmap");

hbvm/src/mem/softpaging/mod.rs

@@ -0,0 +1,296 @@
+//! Platform independent, software paged memory implementation
+
+pub mod icache;
+pub mod lookup;
+pub mod paging;
+
+#[cfg(feature = "alloc")]
+pub mod mapping;
+
+use {
+    super::{addr::Address, LoadError, Memory, MemoryAccessReason, StoreError},
+    core::mem::size_of,
+    icache::ICache,
+    lookup::{AddrPageLookupError, AddrPageLookupOk, AddrPageLookuper},
+    paging::{PageTable, Permission},
+};
+
+/// HoleyBytes software paged memory
+///
+/// - `OUT_PROG_EXEC`: set to `false` to disable executing program
+///   not contained in initially provided program, even the pages
+///   are executable
+#[derive(Clone, Debug)]
+pub struct SoftPagedMem<'p, PfH, const OUT_PROG_EXEC: bool = true> {
+    /// Root page table
+    pub root_pt:    *mut PageTable,
+    /// Page fault handler
+    pub pf_handler: PfH,
+    /// Program memory segment
+    pub program:    &'p [u8],
+    /// Program instruction cache
+    pub icache:     ICache,
+}
+
+impl<'p, PfH: HandlePageFault, const OUT_PROG_EXEC: bool> Memory
+    for SoftPagedMem<'p, PfH, OUT_PROG_EXEC>
+{
+    /// Load value from an address
+    ///
+    /// # Safety
+    /// Applies same conditions as for [`core::ptr::copy_nonoverlapping`]
+    unsafe fn load(
+        &mut self,
+        addr: Address,
+        target: *mut u8,
+        count: usize,
+    ) -> Result<(), LoadError> {
+        self.memory_access(
+            MemoryAccessReason::Load,
+            addr,
+            target,
+            count,
+            perm_check::readable,
+            |src, dst, count| core::ptr::copy_nonoverlapping(src, dst, count),
+        )
+        .map_err(LoadError)
+    }
+
+    /// Store value to an address
+    ///
+    /// # Safety
+    /// Applies same conditions as for [`core::ptr::copy_nonoverlapping`]
+    unsafe fn store(
+        &mut self,
+        addr: Address,
+        source: *const u8,
+        count: usize,
+    ) -> Result<(), StoreError> {
+        self.memory_access(
+            MemoryAccessReason::Store,
+            addr,
+            source.cast_mut(),
+            count,
+            perm_check::writable,
+            |dst, src, count| core::ptr::copy_nonoverlapping(src, dst, count),
+        )
+        .map_err(StoreError)
+    }
+
+    #[inline(always)]
+    unsafe fn prog_read<T>(&mut self, addr: Address) -> Option<T> {
+        if OUT_PROG_EXEC && addr.truncate_usize() > self.program.len() {
+            return self.icache.fetch::<T>(addr, self.root_pt);
+        }
+
+        let addr = addr.truncate_usize();
+        self.program
+            .get(addr..addr + size_of::<T>())
+            .map(|x| x.as_ptr().cast::<T>().read())
+    }
+
+    #[inline(always)]
+    unsafe fn prog_read_unchecked<T>(&mut self, addr: Address) -> T {
+        if OUT_PROG_EXEC && addr.truncate_usize() > self.program.len() {
+            return self
+                .icache
+                .fetch::<T>(addr, self.root_pt)
+                .unwrap_or_else(|| core::mem::zeroed());
+        }
+
+        self.program
+            .as_ptr()
+            .add(addr.truncate_usize())
+            .cast::<T>()
+            .read()
+    }
+}
+
+impl<'p, PfH: HandlePageFault, const OUT_PROG_EXEC: bool> SoftPagedMem<'p, PfH, OUT_PROG_EXEC> {
+    // Everyone behold, the holy function, the god of HBVM memory accesses!
+
+    /// Split address to pages, check their permissions and feed pointers with offset
+    /// to a specified function.
+    ///
+    /// If page is not found, execute page fault trap handler.
+    #[allow(clippy::too_many_arguments)] // Silence peasant
+    fn memory_access(
+        &mut self,
+        reason: MemoryAccessReason,
+        src: Address,
+        mut dst: *mut u8,
+        len: usize,
+        permission_check: fn(Permission) -> bool,
+        action: fn(*mut u8, *mut u8, usize),
+    ) -> Result<(), Address> {
+        // Memory load from program section
+        let (src, len) = if src.truncate_usize() < self.program.len() as _ {
+            // Allow only loads
+            if reason != MemoryAccessReason::Load {
+                return Err(src);
+            }
+
+            // Determine how much data to copy from here
+            let to_copy = len.clamp(0, self.program.len().saturating_sub(src.truncate_usize()));
+
+            // Perform action
+            action(
+                unsafe { self.program.as_ptr().add(src.truncate_usize()).cast_mut() },
+                dst,
+                to_copy,
+            );
+
+            // Return shifted from what we've already copied
+            (
+                src.saturating_add(to_copy as u64),
+                len.saturating_sub(to_copy),
+            )
+        } else {
+            (src, len) // Nothing weird!
+        };
+
+        // Nothing to copy? Don't bother doing anything, bail.
+        if len == 0 {
+            return Ok(());
+        }
+
+        // Create new splitter
+        let mut pspl = AddrPageLookuper::new(src, len, self.root_pt);
+        loop {
+            match pspl.next() {
+                // Page is found
+                Some(Ok(AddrPageLookupOk {
+                    vaddr,
+                    ptr,
+                    size,
+                    perm,
+                })) => {
+                    if !permission_check(perm) {
+                        return Err(vaddr);
+                    }
+
+                    // Perform specified memory action and bump destination pointer
+                    action(ptr, dst, size);
+                    dst = unsafe { dst.add(size) };
+                }
+                // No page found
+                Some(Err(AddrPageLookupError { addr, size })) => {
+                    // Attempt to execute page fault handler
+                    if self.pf_handler.page_fault(
+                        reason,
+                        unsafe { &mut *self.root_pt },
+                        addr,
+                        size,
+                        dst,
+                    ) {
+                        // Shift the splitter address
+                        pspl.bump(size);
+
+                        // Bump dst pointer
+                        dst = unsafe { dst.add(size as _) };
+                    } else {
+                        return Err(addr); // Unhandleable, VM will yield.
+                    }
+                }
+                // No remaining pages, we are done!
+                None => return Ok(()),
+            }
+        }
+    }
+}
+
+/// Extract index in page table on specified level
+///
+/// The level shall not be larger than 4, otherwise
+/// the output of the function is unspecified (yes, it can also panic :)
+pub fn addr_extract_index(addr: Address, lvl: u8) -> usize {
+    debug_assert!(lvl <= 4);
+    let addr = addr.get();
+    usize::try_from((addr >> (lvl * 8 + 12)) & ((1 << 8) - 1)).expect("?conradluget a better CPU")
+}
+
+/// Page size
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum PageSize {
+    /// 4 KiB page (on level 0)
+    Size4K = 4096,
+
+    /// 2 MiB page (on level 1)
+    Size2M = 1024 * 1024 * 2,
+
+    /// 1 GiB page (on level 2)
+    Size1G = 1024 * 1024 * 1024,
+}
+
+impl PageSize {
+    /// Convert page table level to size of page
+    const fn from_lvl(lvl: u8) -> Option<Self> {
+        match lvl {
+            0 => Some(PageSize::Size4K),
+            1 => Some(PageSize::Size2M),
+            2 => Some(PageSize::Size1G),
+            _ => None,
+        }
+    }
+}
+
+impl core::ops::Add<PageSize> for Address {
+    type Output = Self;
+
+    #[inline(always)]
+    fn add(self, rhs: PageSize) -> Self::Output {
+        self + (rhs as u64)
+    }
+}
+
+impl core::ops::AddAssign<PageSize> for Address {
+    #[inline(always)]
+    fn add_assign(&mut self, rhs: PageSize) {
+        *self = Self::new(self.get().wrapping_add(rhs as u64));
+    }
+}
+
+/// Permisison checks
+pub mod perm_check {
+    use super::paging::Permission;
+
+    /// Page is readable
+    #[inline(always)]
+    pub const fn readable(perm: Permission) -> bool {
+        matches!(
+            perm,
+            Permission::Readonly | Permission::Write | Permission::Exec
+        )
+    }
+
+    /// Page is writable
+    #[inline(always)]
+    pub const fn writable(perm: Permission) -> bool {
+        matches!(perm, Permission::Write)
+    }
+
+    /// Page is executable
+    #[inline(always)]
+    pub const fn executable(perm: Permission) -> bool {
+        matches!(perm, Permission::Exec)
+    }
+}
+
+/// Handle VM traps
+pub trait HandlePageFault {
+    /// Handle page fault
+    ///
+    /// Return true if handling was sucessful,
+    /// otherwise the program will be interrupted and will
+    /// yield an error.
+    fn page_fault(
+        &mut self,
+        reason: MemoryAccessReason,
+        pagetable: &mut PageTable,
+        vaddr: Address,
+        size: PageSize,
+        dataptr: *mut u8,
+    ) -> bool
+    where
+        Self: Sized;
+}

hbvm/src/mem/softpaging/paging.rs

@@ -0,0 +1,86 @@
+//! Page table and associated structures implementation
+
+use core::{fmt::Debug, mem::MaybeUninit};
+
+/// Page entry permission
+#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)]
+#[repr(u8)]
+pub enum Permission {
+    /// No page present
+    #[default]
+    Empty,
+    /// Points to another pagetable
+    Node,
+    /// Page is read only
+    Readonly,
+    /// Page is readable and writable
+    Write,
+    /// Page is readable and executable
+    Exec,
+}
+
+/// Page table entry
+#[derive(Clone, Copy, Default, PartialEq, Eq)]
+pub struct PtEntry(u64);
+impl PtEntry {
+    /// Create new
+    ///
+    /// # Safety
+    /// - `ptr` has to point to valid data and shall not be deallocated
+    ///    troughout the entry lifetime
+    #[inline]
+    pub unsafe fn new(ptr: *mut PtPointedData, permission: Permission) -> Self {
+        Self(ptr as u64 | permission as u64)
+    }
+
+    /// Get permission
+    #[inline]
+    pub fn permission(&self) -> Permission {
+        unsafe { core::mem::transmute(self.0 as u8 & 0b111) }
+    }
+
+    /// Get pointer to the data (leaf) or next page table (node)
+    #[inline]
+    pub fn ptr(&self) -> *mut PtPointedData {
+        (self.0 & !((1 << 12) - 1)) as _
+    }
+}
+
+impl Debug for PtEntry {
+    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        f.debug_struct("PtEntry")
+            .field("ptr", &self.ptr())
+            .field("permission", &self.permission())
+            .finish()
+    }
+}
+
+/// Page table
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+#[repr(align(4096))]
+pub struct PageTable {
+    /// How much entries are in use
+    pub childen: u8,
+    /// Entries
+    pub table:   [PtEntry; 256],
+}
+
+impl Default for PageTable {
+    fn default() -> Self {
+        // SAFETY: It's fine, zeroed page table entry is valid (= empty)
+        Self {
+            childen: 0,
+            table:   unsafe { MaybeUninit::zeroed().assume_init() },
+        }
+    }
+}
+
+/// Data page table entry can possibly point to
+#[derive(Clone, Copy)]
+#[repr(C, align(4096))]
+pub union PtPointedData {
+    /// Node - next page table
+    pub pt:   PageTable,
+    /// Leaf
+    pub page: u8,
+}

hbvm/src/memory.rs

@@ -1,70 +0,0 @@
-use crate::engine::VMPage;
-
-use {
-    crate::{engine::Page, RuntimeErrors},
-    alloc::vec::Vec,
-    hashbrown::HashMap,
-    log::trace,
-};
-
-pub struct Memory {
-    inner: HashMap<u64, Page>,
-}
-
-impl Memory {
-    pub fn new() -> Self {
-        Self {
-            inner: HashMap::new(),
-        }
-        //
-    }
-
-    pub fn map_vec(&mut self, address: u64, vec: Vec<u8>) {
-        panic!("Mapping vectors into pages is not supported yet");
-    }
-}
-
-impl Memory {
-    pub fn read_addr8(&mut self, address: u64) -> Result<u8, RuntimeErrors> {
-        let (page, offset) = addr_to_page(address);
-        trace!("page {} offset {}", page, offset);
-        match self.inner.get(&page) {
-            Some(page) => {
-                let val = page.data()[offset as usize];
-                trace!("Value {}", val);
-                Ok(val)
-            }
-            None => {
-                trace!("page not mapped");
-                Err(RuntimeErrors::PageNotMapped(page))
-            }
-        }
-    }
-    pub fn read_addr64(&mut self, address: u64) -> u64 {
-        unimplemented!()
-    }
-
-    pub fn set_addr8(&mut self, address: u64, value: u8) -> Result<(), RuntimeErrors> {
-        let (page, offset) = addr_to_page(address);
-        let ret: Option<(&u64, &mut Page)> = self.inner.get_key_value_mut(&page);
-        match ret {
-            Some((_, page)) => {
-                page.data()[offset as usize] = value;
-            }
-            None => {
-                let mut pg = VMPage::new();
-                pg.data[offset as usize] = value;
-                self.inner.insert(page, Page::VMPage(pg));
-                trace!("Mapped page {}", page);
-            }
-        }
-        Ok(())
-    }
-    pub fn set_addr64(&mut self, address: u64, value: u64) -> Result<(), RuntimeErrors> {
-        unimplemented!()
-    }
-}
-
-fn addr_to_page(addr: u64) -> (u64, u64) {
-    (addr / 8192, addr % 8192)
-}

hbvm/src/utils.rs

@@ -0,0 +1,53 @@
+macro_rules! impl_display {
+    (for $ty:ty => $(|$selfty:pat_param|)? $fmt:literal $(, $($param:expr),+)? $(,)?) => {
+        impl ::core::fmt::Display for $ty {
+            fn fmt(&self, f: &mut ::core::fmt::Formatter<'_>) -> ::core::fmt::Result {
+                $(let $selfty = self;)?
+                write!(f, $fmt, $($param),*)
+            }
+        }
+    };
+
+    (for $ty:ty => $str:literal) => {
+        impl ::core::fmt::Display for $ty {
+            fn fmt(&self, f: &mut ::core::fmt::Formatter<'_>) -> ::core::fmt::Result {
+                f.write_str($str)
+            }
+        }
+    };
+
+    (for $ty:ty => match {$(
+        $bind:pat => $($const:ident)? $fmt:literal $(,$($params:tt)*)?;
+    )*}) => {
+        impl ::core::fmt::Display for $ty {
+            fn fmt(&self, f: &mut ::core::fmt::Formatter<'_>) -> ::core::fmt::Result {
+                match self {
+                    $(
+                        $bind => $crate::utils::internal::impl_display_match_fragment!($($const,)? f, $fmt $(, $($params)*)?)
+                    ),*
+                }
+            }
+        }
+    }
+}
+
+#[doc(hidden)]
+pub(crate) mod internal {
+    macro_rules! impl_display_match_fragment {
+        (const, $f:expr, $lit:literal) => {
+            $f.write_str($lit)
+        };
+
+        ($f:expr, $fmt:literal $(, $($params:tt)*)?) => {
+            write!($f, $fmt, $($($params)*)?)
+        };
+    }
+
+    pub(crate) use impl_display_match_fragment;
+}
+
+macro_rules! static_assert_eq(($l:expr, $r:expr $(,)?) => {
+    const _: [(); ($l != $r) as usize] = [];
+});
+
+pub(crate) use {impl_display, static_assert_eq};

hbvm/src/value.rs

@@ -0,0 +1,80 @@
+//! HoleyBytes register value definition
+
+/// Define [`Value`] union
+///
+/// # Safety
+/// Union variants have to be sound to byte-reinterpretate
+/// between each other. Otherwise the behaviour is undefined.
+macro_rules! value_def {
+    ($($ty:ident),* $(,)?) => {
+        /// HBVM register value
+        #[derive(Copy, Clone)]
+        #[repr(packed)]
+        pub union Value {
+            $(
+                #[doc = concat!(stringify!($ty), " type")]
+                pub $ty: $ty
+            ),*
+        }
+
+
+        $(
+            impl From<$ty> for Value {
+                #[inline]
+                fn from(value: $ty) -> Self {
+                    Self { $ty: value }
+                }
+            }
+
+            crate::utils::static_assert_eq!(
+                core::mem::size_of::<$ty>(),
+                core::mem::size_of::<Value>(),
+            );
+
+            impl private::Sealed for $ty {}
+            unsafe impl ValueVariant for $ty {}
+        )*
+    };
+}
+
+impl Value {
+    /// Byte reinterpret value to target variant
+    #[inline]
+    pub fn cast<V: ValueVariant>(self) -> V {
+        /// Evil.
+        ///
+        /// Transmute cannot be performed with generic type
+        /// as size is unknown, so union is used.
+        ///
+        /// # Safety
+        /// If [`ValueVariant`] implemented correctly, it's fine :)
+        ///
+        /// :ferrisClueless:
+        union Transmute<Variant: ValueVariant> {
+            /// Self
+            src:     Value,
+            /// Target variant
+            variant: Variant,
+        }
+
+        unsafe { Transmute { src: self }.variant }
+    }
+}
+
+/// # Safety
+/// - N/A, not to be implemented manually
+pub unsafe trait ValueVariant: private::Sealed + Copy + Into<Value> {}
+
+mod private {
+    pub trait Sealed {}
+}
+
+value_def!(u64, i64, f64);
+crate::utils::static_assert_eq!(core::mem::size_of::<Value>(), 8);
+
+impl core::fmt::Debug for Value {
+    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        // Print formatted as hexadecimal, unsigned integer
+        write!(f, "{:x}", self.cast::<u64>())
+    }
+}

hbvm/src/vmrun.rs

@@ -0,0 +1,409 @@
+//! Welcome to the land of The Great Dispatch Loop
+//!
+//! Have fun
+
+use crate::mem::Address;
+
+use {
+    super::{
+        bmc::BlockCopier,
+        mem::Memory,
+        value::{Value, ValueVariant},
+        Vm, VmRunError, VmRunOk,
+    },
+    core::{cmp::Ordering, mem::size_of, ops},
+    hbbytecode::{
+        ParamBB, ParamBBB, ParamBBBB, ParamBBD, ParamBBDH, ParamBBW, ParamBD, ProgramVal,
+    },
+};
+
+impl<Mem, const TIMER_QUOTIENT: usize> Vm<Mem, TIMER_QUOTIENT>
+where
+    Mem: Memory,
+{
+    /// Execute program
+    ///
+    /// Program can return [`VmRunError`] if a trap handling failed
+    #[cfg_attr(feature = "nightly", repr(align(4096)))]
+    pub fn run(&mut self) -> Result<VmRunOk, VmRunError> {
+        use hbbytecode::opcode::*;
+        loop {
+            // Big match
+            //
+            // Contribution guide:
+            // - Zero register shall never be overwitten. It's value has to always be 0.
+            //     - Prefer `Self::read_reg` and `Self::write_reg` functions
+            // - Extract parameters using `param!` macro
+            // - Prioritise speed over code size
+            //     - Memory is cheap, CPUs not that much
+            // - Do not heap allocate at any cost
+            //     - Yes, user-provided trap handler may allocate,
+            //       but that is not our »fault«.
+            // - Unsafe is kinda must, but be sure you have validated everything
+            //     - Your contributions have to pass sanitizers and Miri
+            // - Strictly follow the spec
+            //     - The spec does not specify how you perform actions, in what order,
+            //       just that the observable effects have to be performed in order and
+            //       correctly.
+            // - Yes, we assume you run 64 bit CPU. Else ?conradluget a better CPU
+            //   sorry 8 bit fans, HBVM won't run on your Speccy :(
+            unsafe {
+                match self
+                    .memory
+                    .prog_read::<u8>(self.pc as _)
+                    .ok_or(VmRunError::ProgramFetchLoadEx(self.pc as _))?
+                {
+                    UN => {
+                        self.decode::<()>();
+                        return Err(VmRunError::Unreachable);
+                    }
+                    TX => {
+                        self.decode::<()>();
+                        return Ok(VmRunOk::End);
+                    }
+                    NOP => self.decode::<()>(),
+                    ADD => self.binary_op(u64::wrapping_add),
+                    SUB => self.binary_op(u64::wrapping_sub),
+                    MUL => self.binary_op(u64::wrapping_mul),
+                    AND => self.binary_op::<u64>(ops::BitAnd::bitand),
+                    OR => self.binary_op::<u64>(ops::BitOr::bitor),
+                    XOR => self.binary_op::<u64>(ops::BitXor::bitxor),
+                    SL => self.binary_op(|l, r| u64::wrapping_shl(l, r as u32)),
+                    SR => self.binary_op(|l, r| u64::wrapping_shr(l, r as u32)),
+                    SRS => self.binary_op(|l, r| i64::wrapping_shl(l, r as u32)),
+                    CMP => {
+                        // Compare a0 <=> a1
+                        // < →  0
+                        // > →  1
+                        // = →  2
+
+                        let ParamBBB(tg, a0, a1) = self.decode();
+                        self.write_reg(
+                            tg,
+                            self.read_reg(a0)
+                                .cast::<i64>()
+                                .cmp(&self.read_reg(a1).cast::<i64>())
+                                as i64
+                                + 1,
+                        );
+                    }
+                    CMPU => {
+                        // Unsigned comparsion
+                        let ParamBBB(tg, a0, a1) = self.decode();
+                        self.write_reg(
+                            tg,
+                            self.read_reg(a0)
+                                .cast::<u64>()
+                                .cmp(&self.read_reg(a1).cast::<u64>())
+                                as i64
+                                + 1,
+                        );
+                    }
+                    NOT => {
+                        // Logical negation
+                        let ParamBB(tg, a0) = self.decode();
+                        self.write_reg(tg, !self.read_reg(a0).cast::<u64>());
+                    }
+                    NEG => {
+                        // Bitwise negation
+                        let ParamBB(tg, a0) = self.decode();
+                        self.write_reg(
+                            tg,
+                            match self.read_reg(a0).cast::<u64>() {
+                                0 => 1_u64,
+                                _ => 0,
+                            },
+                        );
+                    }
+                    DIR => {
+                        // Fused Division-Remainder
+                        let ParamBBBB(dt, rt, a0, a1) = self.decode();
+                        let a0 = self.read_reg(a0).cast::<u64>();
+                        let a1 = self.read_reg(a1).cast::<u64>();
+                        self.write_reg(dt, a0.checked_div(a1).unwrap_or(u64::MAX));
+                        self.write_reg(rt, a0.checked_rem(a1).unwrap_or(u64::MAX));
+                    }
+                    ADDI => self.binary_op_imm(u64::wrapping_add),
+                    MULI => self.binary_op_imm(u64::wrapping_sub),
+                    ANDI => self.binary_op_imm::<u64>(ops::BitAnd::bitand),
+                    ORI => self.binary_op_imm::<u64>(ops::BitOr::bitor),
+                    XORI => self.binary_op_imm::<u64>(ops::BitXor::bitxor),
+                    SLI => self.binary_op_ims(u64::wrapping_shl),
+                    SRI => self.binary_op_ims(u64::wrapping_shr),
+                    SRSI => self.binary_op_ims(i64::wrapping_shr),
+                    CMPI => {
+                        let ParamBBD(tg, a0, imm) = self.decode();
+                        self.write_reg(
+                            tg,
+                            self.read_reg(a0)
+                                .cast::<i64>()
+                                .cmp(&Value::from(imm).cast::<i64>())
+                                as i64,
+                        );
+                    }
+                    CMPUI => {
+                        let ParamBBD(tg, a0, imm) = self.decode();
+                        self.write_reg(tg, self.read_reg(a0).cast::<u64>().cmp(&imm) as i64);
+                    }
+                    CP => {
+                        let ParamBB(tg, a0) = self.decode();
+                        self.write_reg(tg, self.read_reg(a0));
+                    }
+                    SWA => {
+                        // Swap registers
+                        let ParamBB(r0, r1) = self.decode();
+                        match (r0, r1) {
+                            (0, 0) => (),
+                            (dst, 0) | (0, dst) => self.write_reg(dst, 0_u64),
+                            (r0, r1) => {
+                                core::ptr::swap(
+                                    self.registers.get_unchecked_mut(usize::from(r0)),
+                                    self.registers.get_unchecked_mut(usize::from(r1)),
+                                );
+                            }
+                        }
+                    }
+                    LI => {
+                        let ParamBD(tg, imm) = self.decode();
+                        self.write_reg(tg, imm);
+                    }
+                    LD => {
+                        // Load. If loading more than register size, continue on adjecent registers
+                        let ParamBBDH(dst, base, off, count) = self.decode();
+                        let n: u8 = match dst {
+                            0 => 1,
+                            _ => 0,
+                        };
+
+                        self.memory.load(
+                            self.ldst_addr_uber(dst, base, off, count, n)?,
+                            self.registers
+                                .as_mut_ptr()
+                                .add(usize::from(dst) + usize::from(n))
+                                .cast(),
+                            usize::from(count).saturating_sub(n.into()),
+                        )?;
+                    }
+                    ST => {
+                        // Store. Same rules apply as to LD
+                        let ParamBBDH(dst, base, off, count) = self.decode();
+                        self.memory.store(
+                            self.ldst_addr_uber(dst, base, off, count, 0)?,
+                            self.registers.as_ptr().add(usize::from(dst)).cast(),
+                            count.into(),
+                        )?;
+                    }
+                    BMC => {
+                        // Block memory copy
+                        match if let Some(copier) = &mut self.copier {
+                            // There is some copier, poll.
+                            copier.poll(&mut self.memory)
+                        } else {
+                            // There is none, make one!
+                            let ParamBBD(src, dst, count) = self.decode();
+
+                            // So we are still on BMC on next cycle
+                            self.pc -= size_of::<ParamBBD>() + 1;
+
+                            self.copier = Some(BlockCopier::new(
+                                Address::new(self.read_reg(src).cast()),
+                                Address::new(self.read_reg(dst).cast()),
+                                count as _,
+                            ));
+
+                            self.copier
+                                .as_mut()
+                                .unwrap_unchecked() // SAFETY: We just assigned there
+                                .poll(&mut self.memory)
+                        } {
+                            // We are done, shift program counter
+                            core::task::Poll::Ready(Ok(())) => {
+                                self.copier = None;
+                                self.pc += size_of::<ParamBBD>() + 1;
+                            }
+                            // Error, shift program counter (for consistency)
+                            // and yield error
+                            core::task::Poll::Ready(Err(e)) => {
+                                self.pc += size_of::<ParamBBD>() + 1;
+                                return Err(e.into());
+                            }
+                            // Not done yet, proceed to next cycle
+                            core::task::Poll::Pending => (),
+                        }
+                    }
+                    BRC => {
+                        // Block register copy
+                        let ParamBBB(src, dst, count) = self.decode();
+                        if src.checked_add(count).is_none() || dst.checked_add(count).is_none() {
+                            return Err(VmRunError::RegOutOfBounds);
+                        }
+
+                        core::ptr::copy(
+                            self.registers.get_unchecked(usize::from(src)),
+                            self.registers.get_unchecked_mut(usize::from(dst)),
+                            usize::from(count),
+                        );
+                    }
+                    JAL => {
+                        // Jump and link. Save PC after this instruction to
+                        // specified register and jump to reg + offset.
+                        let ParamBBD(save, reg, offset) = self.decode();
+                        self.write_reg(save, self.pc.get());
+                        self.pc =
+                            Address::new(self.read_reg(reg).cast::<u64>().saturating_add(offset));
+                    }
+                    // Conditional jumps, jump only to immediates
+                    JEQ => self.cond_jmp::<u64>(Ordering::Equal),
+                    JNE => {
+                        let ParamBBD(a0, a1, jt) = self.decode();
+                        if self.read_reg(a0).cast::<u64>() != self.read_reg(a1).cast::<u64>() {
+                            self.pc = Address::new(jt);
+                        }
+                    }
+                    JLT => self.cond_jmp::<u64>(Ordering::Less),
+                    JGT => self.cond_jmp::<u64>(Ordering::Greater),
+                    JLTU => self.cond_jmp::<i64>(Ordering::Less),
+                    JGTU => self.cond_jmp::<i64>(Ordering::Greater),
+                    ECALL => {
+                        self.decode::<()>();
+
+                        // So we don't get timer interrupt after ECALL
+                        if TIMER_QUOTIENT != 0 {
+                            self.timer = self.timer.wrapping_add(1);
+                        }
+                        return Ok(VmRunOk::Ecall);
+                    }
+                    ADDF => self.binary_op::<f64>(ops::Add::add),
+                    SUBF => self.binary_op::<f64>(ops::Sub::sub),
+                    MULF => self.binary_op::<f64>(ops::Mul::mul),
+                    DIRF => {
+                        let ParamBBBB(dt, rt, a0, a1) = self.decode();
+                        let a0 = self.read_reg(a0).cast::<f64>();
+                        let a1 = self.read_reg(a1).cast::<f64>();
+                        self.write_reg(dt, a0 / a1);
+                        self.write_reg(rt, a0 % a1);
+                    }
+                    FMAF => {
+                        let ParamBBBB(dt, a0, a1, a2) = self.decode();
+                        self.write_reg(
+                            dt,
+                            self.read_reg(a0).cast::<f64>() * self.read_reg(a1).cast::<f64>()
+                                + self.read_reg(a2).cast::<f64>(),
+                        );
+                    }
+                    NEGF => {
+                        let ParamBB(dt, a0) = self.decode();
+                        self.write_reg(dt, -self.read_reg(a0).cast::<f64>());
+                    }
+                    ITF => {
+                        let ParamBB(dt, a0) = self.decode();
+                        self.write_reg(dt, self.read_reg(a0).cast::<i64>() as f64);
+                    }
+                    FTI => {
+                        let ParamBB(dt, a0) = self.decode();
+                        self.write_reg(dt, self.read_reg(a0).cast::<f64>() as i64);
+                    }
+                    ADDFI => self.binary_op_imm::<f64>(ops::Add::add),
+                    MULFI => self.binary_op_imm::<f64>(ops::Mul::mul),
+                    op => return Err(VmRunError::InvalidOpcode(op)),
+                }
+            }
+
+            if TIMER_QUOTIENT != 0 {
+                self.timer = self.timer.wrapping_add(1);
+                if self.timer % TIMER_QUOTIENT == 0 {
+                    return Ok(VmRunOk::Timer);
+                }
+            }
+        }
+    }
+
+    /// Decode instruction operands
+    #[inline(always)]
+    unsafe fn decode<T: ProgramVal>(&mut self) -> T {
+        let pc1 = self.pc + 1_u64;
+        let data = self.memory.prog_read_unchecked::<T>(pc1 as _);
+        self.pc += 1 + size_of::<T>();
+        data
+    }
+
+    /// Perform binary operating over two registers
+    #[inline(always)]
+    unsafe fn binary_op<T: ValueVariant>(&mut self, op: impl Fn(T, T) -> T) {
+        let ParamBBB(tg, a0, a1) = self.decode();
+        self.write_reg(
+            tg,
+            op(self.read_reg(a0).cast::<T>(), self.read_reg(a1).cast::<T>()),
+        );
+    }
+
+    /// Perform binary operation over register and immediate
+    #[inline(always)]
+    unsafe fn binary_op_imm<T: ValueVariant>(&mut self, op: impl Fn(T, T) -> T) {
+        let ParamBBD(tg, reg, imm) = self.decode();
+        self.write_reg(
+            tg,
+            op(self.read_reg(reg).cast::<T>(), Value::from(imm).cast::<T>()),
+        );
+    }
+
+    /// Perform binary operation over register and shift immediate
+    #[inline(always)]
+    unsafe fn binary_op_ims<T: ValueVariant>(&mut self, op: impl Fn(T, u32) -> T) {
+        let ParamBBW(tg, reg, imm) = self.decode();
+        self.write_reg(tg, op(self.read_reg(reg).cast::<T>(), imm));
+    }
+
+    /// Jump at `#3` if ordering on `#0 <=> #1` is equal to expected
+    #[inline(always)]
+    unsafe fn cond_jmp<T: ValueVariant + Ord>(&mut self, expected: Ordering) {
+        let ParamBBD(a0, a1, ja) = self.decode();
+        if self
+            .read_reg(a0)
+            .cast::<T>()
+            .cmp(&self.read_reg(a1).cast::<T>())
+            == expected
+        {
+            self.pc = Address::new(ja);
+        }
+    }
+
+    /// Read register
+    #[inline(always)]
+    unsafe fn read_reg(&self, n: u8) -> Value {
+        *self.registers.get_unchecked(n as usize)
+    }
+
+    /// Write a register.
+    /// Writing to register 0 is no-op.
+    #[inline(always)]
+    unsafe fn write_reg(&mut self, n: u8, value: impl Into<Value>) {
+        if n != 0 {
+            *self.registers.get_unchecked_mut(n as usize) = value.into();
+        }
+    }
+
+    /// Load / Store Address check-computation überfunction
+    #[inline(always)]
+    unsafe fn ldst_addr_uber(
+        &self,
+        dst: u8,
+        base: u8,
+        offset: u64,
+        size: u16,
+        adder: u8,
+    ) -> Result<Address, VmRunError> {
+        let reg = dst.checked_add(adder).ok_or(VmRunError::RegOutOfBounds)?;
+
+        if usize::from(reg) * 8 + usize::from(size) > 2048 {
+            Err(VmRunError::RegOutOfBounds)
+        } else {
+            self.read_reg(base)
+                .cast::<u64>()
+                .checked_add(offset)
+                .and_then(|x| x.checked_add(adder.into()))
+                .ok_or(VmRunError::AddrOutOfBounds)
+                .map(Address::new)
+        }
+    }
+}

rustfmt.toml

@@ -1,3 +1,4 @@
 hex_literal_case = "Upper"
 imports_granularity = "One"
-struct_field_align_threshold = 5
+struct_field_align_threshold = 8
+enum_discrim_align_threshold = 8

spec.md

@@ -0,0 +1,317 @@
+# HoleyBytes ISA Specification
+
+# Bytecode format
+- Holey Bytes program should start with following magic: `[0xAB, 0x1E, 0x0B]`
+- All numbers are encoded little-endian
+- There is 256 registers, they are represented by a byte
+- Immediate values are 64 bit
+- Program is by spec required to be terminated with 12 zero bytes
+
+### Instruction encoding
+- Instruction parameters are packed (no alignment)
+- [opcode, …parameters…]
+
+### Instruction parameter types
+- B = Byte
+- D = Doubleword (64 bits)
+- H = Halfword (16 bits)
+
+| Name | Size    |
+|:----:|:--------|
+| BBBB | 32 bits |
+| BBB  | 24 bits |
+| BBDH | 96 bits |
+| BBD  | 80 bits |
+| BBW  | 48 bits |
+|  BB  | 16 bits |
+|  BD  | 72 bits |
+|  D   | 64 bits |
+|  N   | 0  bits |
+
+# Instructions
+- `#n`: register in parameter *n*
+- `imm #n`: for immediate in parameter *n*
+- `P ← V`: Set register P to value V
+- `[x]`: Address x
+
+## Program execution control
+- N type
+
+| Opcode | Name |            Action             |
+|:------:|:----:|:-----------------------------:|
+|   0    |  UN  | Trigger unreachable code trap |
+|   1    |  TX  |      Terminate execution      |
+|   2    | NOP  |          Do nothing           |
+
+## Integer binary ops.
+- BBB type
+- `#0 ← #1 <op> #2`
+
+| Opcode | Name |         Action          |
+|:------:|:----:|:-----------------------:|
+|   3    | ADD  |    Wrapping addition    |
+|   4    | SUB  |  Wrapping subtraction   |
+|   5    | MUL  | Wrapping multiplication |
+|   6    | AND  |         Bitand          |
+|   7    |  OR  |          Bitor          |
+|   8    | XOR  |         Bitxor          |
+|   9    |  SL  | Unsigned left bitshift  |
+|   10   |  SR  | Unsigned right bitshift |
+|   11   | SRS  |  Signed right bitshift  |
+
+### Comparsion
+| Opcode | Name |       Action        |
+|:------:|:----:|:-------------------:|
+|   12   | CMP  |  Signed comparsion  |
+|   13   | CMPU | Unsigned comparsion |
+
+#### Comparsion table
+| #1 *op* #2 | Result |
+|:----------:|:------:|
+|     <      |   0    |
+|     =      |   1    |
+|     >      |   2    |
+
+### Division-remainder
+- Type BBBB
+- In case of `#3` is zero, the resulting value is all-ones
+- `#0 ← #2 ÷ #3`
+- `#1 ← #2 % #3`
+
+| Opcode | Name |             Action              |
+|:------:|:----:|:-------------------------------:|
+|   14   | DIR  | Divide and remainder combinated |
+
+### Negations
+- Type BB
+- `#0 ← #1 <op> #2`
+
+| Opcode | Name |      Action      |
+|:------:|:----:|:----------------:|
+|   15   | NEG  |   Bit negation   |
+|   16   | NOT  | Logical negation |
+
+## Integer immediate binary ops.
+- Type BBD
+- `#0 ← #1 <op> imm #2`
+
+| Opcode | Name |        Action        |
+|:------:|:----:|:--------------------:|
+|   17   | ADDI |  Wrapping addition   |
+|   18   | MULI | Wrapping subtraction |
+|   19   | ANDI |        Bitand        |
+|   20   | ORI  |        Bitor         |
+|   21   | XORI |        Bitxor        |
+
+### Bitshifts
+- Type BBW
+| Opcode | Name |         Action          |
+|:------:|:----:|:-----------------------:|
+|   22   | SLI  | Unsigned left bitshift  |
+|   23   | SRI  | Unsigned right bitshift |
+|   24   | SRSI |  Signed right bitshift  |
+
+### Comparsion
+- Comparsion is the same as when RRR type
+
+| Opcode | Name  |       Action        |
+|:------:|:-----:|:-------------------:|
+|   25   | CMPI  |  Signed comparsion  |
+|   26   | CMPUI | Unsigned comparsion |
+
+## Register value set / copy
+
+### Copy
+- Type BB
+- `#0 ← #1`
+
+| Opcode | Name | Action |
+|:------:|:----:|:------:|
+|   27   |  CP  |  Copy  |
+
+### Swap
+- Type BB
+- Swap #0 and #1
+- Zero register rules:
+    - Both: no-op
+    - One: Copy zero to the non-zero register
+
+| Opcode | Name | Action |
+|:------:|:----:|:------:|
+|   28   | SWA  |  Swap  |
+
+### Load immediate
+- Type BD
+- `#0 ← #1`
+
+| Opcode | Name |     Action     |
+|:------:|:----:|:--------------:|
+|   29   |  LI  | Load immediate |
+
+## Memory operations
+- Type BBDH
+- If loaded/store value exceeds one register size, continue accessing following registers
+
+### Load / Store
+| Opcode | Name |                 Action                  |
+|:------:|:----:|:---------------------------------------:|
+|   30   |  LD  | `#0 ← [#1 + imm #3], copy imm #4 bytes` |
+|   31   |  ST  | `[#1 + imm #3] ← #0, copy imm #4 bytes` |
+
+## Block copy
+- Block copy source and target can overlap
+
+### Memory copy
+- Type BBD
+
+| Opcode | Name |              Action              |
+|:------:|:----:|:--------------------------------:|
+|   32   | BMC  | `[#1] ← [#0], copy imm #2 bytes` |
+
+### Register copy
+- Type BBB
+- Copy a block a register to another location (again, overflowing to following registers)
+
+| Opcode | Name |              Action              |
+|:------:|:----:|:--------------------------------:|
+|   33   | BRC  | `#1 ← #0, copy imm #2 registers` |
+
+## Control flow
+
+### Unconditional jump
+- Type D
+| Opcode | Name |             Action              |
+|:------:|:----:|:-------------------------------:|
+|   34   | JMP  | Unconditional, non-linking jump |
+
+### Unconditional linking jump
+- Type BBD
+
+| Opcode | Name |                       Action                       |
+|:------:|:----:|:--------------------------------------------------:|
+|   35   | JAL  | Save PC past JAL to `#0` and jump at `#1 + imm #2` |
+
+### Conditional jumps
+- Type BBD
+- Jump at `imm #2` if `#0 <op> #1`
+
+| Opcode | Name |  Comparsion  |
+|:------:|:----:|:------------:|
+|   36   | JEQ  |      =       |
+|   37   | JNE  |      ≠       |
+|   38   | JLT  |  < (signed)  |
+|   39   | JGT  |  > (signed)  |
+|   40   | JLTU | < (unsigned) |
+|   41   | JGTU | > (unsigned) |
+
+### Environment call
+- Type N
+
+| Opcode | Name  |                Action                 |
+|:------:|:-----:|:-------------------------------------:|
+|   42   | ECALL | Cause an trap to the host environment |
+
+## Floating point operations
+- Type BBB
+- `#0 ← #1 <op> #2`
+
+| Opcode | Name |     Action     |
+|:------:|:----:|:--------------:|
+|   43   | ADDF |    Addition    |
+|   44   | SUBF |  Subtraction   |
+|   45   | MULF | Multiplication |
+
+### Division-remainder
+- Type BBBB
+
+| Opcode | Name |          Action           |
+|:------:|:----:|:-------------------------:|
+|   46   | DIRF | Same as for integer `DIR` |
+
+### Fused Multiply-Add
+- Type BBBB
+
+| Opcode | Name |        Action         |
+|:------:|:----:|:---------------------:|
+|   47   | FMAF | `#0 ← (#1 * #2) + #3` |
+
+### Negation
+- Type BB
+| Opcode | Name |   Action   |
+|:------:|:----:|:----------:|
+|   48   | NEGF | `#0 ← -#1` |
+
+### Conversion
+- Type BB
+- Signed
+- `#0 ← #1 as _`
+
+| Opcode | Name |    Action    |
+|:------:|:----:|:------------:|
+|   49   | ITF  | Int to Float |
+|   50   | FTI  | Float to Int |
+
+## Floating point immediate operations
+- Type BBD
+- `#0 ← #1 <op> imm #2`
+
+| Opcode | Name  |     Action     |
+|:------:|:-----:|:--------------:|
+|   51   | ADDFI |    Addition    |
+|   52   | MULFI | Multiplication |
+
+# Registers
+- There is 255 registers + one zero register (with index 0)
+- Reading from zero register yields zero
+- Writing to zero register is a no-op
+
+# Memory
+- Addresses are 64 bit
+- Program should be in the same address space as all other data
+- Memory implementation is arbitrary
+    - Address `0x0` may or may not be valid. Count with compilers
+      considering it invalid!
+- In case of accessing invalid address:
+    - Program shall trap (LoadAccessEx, StoreAccessEx) with parameter of accessed address
+    - Value of register when trapped is undefined
+
+## Recommendations
+- If paging used:
+    - Leave first page invalid
+    - Pages should be at least 4 KiB
+
+# Program execution
+- The way of program execution is implementation defined
+- The execution is arbitrary, as long all effects are obervable
+    in the way as program was executed literally, in order.
+
+# Program validation
+- Invalid program should cause runtime error:
+    - The form of error is arbitrary. Can be a trap or an interpreter-specified error
+    - It shall not be handleable from within the program
+- Executing invalid opcode should trap
+- Program can be validaded either before execution or when executing
+
+# Traps
+Program should at least implement these traps:
+- Environment call
+- Invalid instruction exception
+- Load address exception
+- Store address exception
+- Unreachable instruction
+
+and executing environment should be able to get information about them,
+like the opcode of invalid instruction or attempted address to load/store.
+Details about these are left as an implementation detail.
+
+# Assembly
+HoleyBytes assembly format is not defined, this is just a weak description
+of `hbasm` syntax.
+
+- Opcode names correspond to specified opcode names, lowercase (`nop`)
+- Parameters are separated by comma (`addi r0, r0, 1`)
+- Instructions are separated by either line feed or semicolon
+- Registers are represented by `r` followed by the number (`r10`)
+- Labels are defined by label name followed with colon (`loop:`)
+- Labels are references simply by their name (`print`)
+- Immediates are entered plainly. Negative numbers supported.