runtime/type-registry: add new type-registry library for the type c…

…onfusion APIs

This is written in Rust like `memory-map` is, so I made the `CMakeLists.txt` very similar.
Since the library may be called concurrently, it uses a hash map which is non-trivial in C,
and we already have Rust libraries like `memory-map`, Rust seemed like a good choice here.

There are a few other things I think we want to do here,
such as ensuring a type doesn't change across the duration of the function call
(not just the beginning); this would be done with a read lock.
Also, we probably want to store the compartment number along with the type ID.
But I'll do these in follow-up PRs once we get the initial parts working.

runtime/CMakeLists.txt

@@ -4,3 +4,4 @@ project(runtime)
 add_subdirectory(libia2)
 add_subdirectory(partition-alloc)
 add_subdirectory(tracer)
+add_subdirectory(type-registry)

runtime/tracer/CMakeLists.txt

@@ -8,6 +8,7 @@ else()
 set(CARGO_TARGET_FLAG "")
 set(CARGO_ARCH_SUFFIX "")
 endif()
+
 add_custom_command(
   OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/${CARGO_ARCH_SUFFIX}/release/libmemory_map.so
   COMMAND CARGO_TARGET_DIR=${CMAKE_CURRENT_BINARY_DIR} cargo build ${CARGO_TARGET_FLAG} --manifest-path ${CMAKE_CURRENT_SOURCE_DIR}/memory-map/Cargo.toml --release

runtime/type-registry/CMakeLists.txt

@@ -0,0 +1,21 @@
+cmake_minimum_required(VERSION 3.12)
+project(type-registry)
+
+if (LIBIA2_AARCH64)
+set(CARGO_TARGET_FLAG "--target=aarch64-unknown-linux-gnu" "--config" "target.aarch64-unknown-linux-gnu.linker=\\\"aarch64-linux-gnu-gcc\\\"")
+set(CARGO_ARCH_SUFFIX "aarch64-unknown-linux-gnu")
+else()
+set(CARGO_TARGET_FLAG "")
+set(CARGO_ARCH_SUFFIX "")
+endif()
+
+add_custom_command(
+  OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/${CARGO_ARCH_SUFFIX}/release/libtype_registry.so
+  COMMAND CARGO_TARGET_DIR=${CMAKE_CURRENT_BINARY_DIR} cargo build ${CARGO_TARGET_FLAG} --manifest-path ${CMAKE_CURRENT_SOURCE_DIR}/Cargo.toml --release
+  DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/Cargo.toml ${CMAKE_CURRENT_SOURCE_DIR}/src/lib.rs
+)
+
+add_library(type-registry STATIC IMPORTED)
+set_property(TARGET type-registry PROPERTY IMPORTED_LOCATION ${CMAKE_CURRENT_BINARY_DIR}/${CARGO_ARCH_SUFFIX}/release/libtype_registry.so)
+add_custom_target(type-registry-tgt DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/${CARGO_ARCH_SUFFIX}/release/libtype_registry.so)
+add_dependencies(type-registry type-registry-tgt)

runtime/type-registry/Cargo.toml

@@ -0,0 +1,9 @@
+[package]
+name = "type-registry"
+version = "0.1.0"
+edition = "2021"
+
+[lib]
+crate-type = ["cdylib"]
+
+[dependencies]

runtime/type-registry/src/lib.rs

@@ -0,0 +1,176 @@
+use std::collections::HashMap;
+use std::fmt;
+use std::fmt::Debug;
+use std::fmt::Display;
+use std::fmt::Formatter;
+use std::sync::LazyLock;
+use std::sync::RwLock;
+
+/// A pointer's address, used purely as an address for comparisons.
+#[derive(PartialEq, Eq, Hash)]
+#[repr(transparent)]
+pub struct PtrAddr(*const ());
+
+/// SAFETY: [`PtrAddr`] is used purely as an address.
+unsafe impl Send for PtrAddr {}
+
+/// SAFETY: [`PtrAddr`] is used purely as an address.
+unsafe impl Sync for PtrAddr {}
+
+impl Display for PtrAddr {
+    fn fmt(&self, f: &mut Formatter) -> fmt::Result {
+        write!(f, "{:?}", self.0)
+    }
+}
+
+impl Debug for PtrAddr {
+    fn fmt(&self, f: &mut Formatter) -> fmt::Result {
+        write!(f, "{}", self)
+    }
+}
+
+/// A unique ID for a type.
+///
+/// This can be anything, as long as it's unique for a type in a program.
+#[derive(Clone, Copy, PartialEq, Eq, Hash)]
+#[repr(transparent)]
+pub struct TypeId(usize);
+
+impl Display for TypeId {
+    fn fmt(&self, f: &mut Formatter) -> fmt::Result {
+        write!(f, "Type{}", self.0)
+    }
+}
+
+impl Debug for TypeId {
+    fn fmt(&self, f: &mut Formatter) -> fmt::Result {
+        write!(f, "{}", self)
+    }
+}
+
+static GLOBAL_TYPE_REGISTRY: LazyLock<TypeRegistry> = LazyLock::new(TypeRegistry::default);
+
+/// See [`TypeRegistry::construct`].
+#[no_mangle]
+pub extern "C" fn ia2_type_registry_construct(ptr: PtrAddr, type_id: TypeId) {
+    GLOBAL_TYPE_REGISTRY.construct(ptr, type_id);
+}
+
+/// See [`TypeRegistry::destruct`].
+#[no_mangle]
+pub extern "C" fn ia2_type_registry_destruct(ptr: PtrAddr, expected_type_id: TypeId) {
+    GLOBAL_TYPE_REGISTRY.destruct(ptr, expected_type_id);
+}
+
+/// See [`TypeRegistry::check`].
+#[no_mangle]
+pub extern "C" fn ia2_type_registry_check(ptr: PtrAddr, expected_type_id: TypeId) {
+    GLOBAL_TYPE_REGISTRY.check(ptr, expected_type_id);
+}
+
+#[derive(Default)]
+pub struct TypeRegistry {
+    map: RwLock<HashMap<PtrAddr, TypeId>>,
+}
+
+impl TypeRegistry {
+    /// Called after an object is constructed to register it and its type.
+    ///
+    /// Panics if `ptr` is already constructed.
+    #[track_caller]
+    pub fn construct(&self, ptr: PtrAddr, type_id: TypeId) {
+        let guard = &mut *self.map.write().unwrap();
+        let prev_type_id = guard.insert(ptr, type_id);
+        assert_eq!(prev_type_id, None);
+    }
+
+    /// Called after an object is destructured to unregister it and its type.
+    ///
+    /// Panics if `ptr` has a different type.
+    #[track_caller]
+    pub fn destruct(&self, ptr: PtrAddr, expected_type_id: TypeId) {
+        let guard = &mut *self.map.write().unwrap();
+        let type_id = guard.remove(&ptr);
+        assert_eq!(type_id, Some(expected_type_id));
+    }
+
+    /// Called to check a pointer has the expected type.
+    ///
+    /// Panics if `ptr` is not registered or has a different type.
+    #[track_caller]
+    pub fn check(&self, ptr: PtrAddr, expected_type_id: TypeId) {
+        let guard = &*self.map.read().unwrap();
+        let type_id = guard.get(&ptr).copied();
+        assert_eq!(type_id, Some(expected_type_id));
+    }
+}
+
+impl Drop for TypeRegistry {
+    fn drop(&mut self) {
+        match self.map.get_mut() {
+            Ok(map) => {
+                if !map.is_empty() {
+                    eprintln!("warning: leak: {map:?}");
+                }
+            }
+            Err(e) => {
+                eprintln!("poison: {e}")
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn ptr(addr: usize) -> PtrAddr {
+        PtrAddr(addr as *const ())
+    }
+
+    #[test]
+    fn normal() {
+        let registry = TypeRegistry::default();
+        registry.construct(ptr(0xA), TypeId(1));
+        registry.check(ptr(0xA), TypeId(1));
+        registry.construct(ptr(0xB), TypeId(2));
+        registry.check(ptr(0xA), TypeId(1));
+        registry.check(ptr(0xB), TypeId(2));
+        registry.destruct(ptr(0xA), TypeId(1));
+        registry.check(ptr(0xB), TypeId(2));
+        registry.destruct(ptr(0xB), TypeId(2));
+    }
+
+    #[test]
+    #[should_panic]
+    fn destruct_non_existent_ptr() {
+        let registry = TypeRegistry::default();
+        registry.destruct(ptr(0xA), TypeId(1));
+    }
+
+    #[test]
+    #[should_panic]
+    fn check_non_existent_ptr() {
+        let registry = TypeRegistry::default();
+        registry.check(ptr(0xA), TypeId(1));
+    }
+
+    #[test]
+    #[should_panic]
+    fn check_freed_ptr() {
+        let registry = TypeRegistry::default();
+        registry.construct(ptr(0xA), TypeId(1));
+        registry.check(ptr(0xA), TypeId(1));
+        registry.destruct(ptr(0xA), TypeId(1));
+        registry.check(ptr(0xA), TypeId(1));
+    }
+
+    #[test]
+    #[should_panic]
+    fn construct_existing_ptr() {
+        let registry = TypeRegistry::default();
+        registry.construct(ptr(0xA), TypeId(1));
+        registry.check(ptr(0xA), TypeId(1));
+        registry.construct(ptr(0xA), TypeId(1));
+    }
+}