microfetch: move _start to binary crate, gate entry_rust with cfg(not(test))

This commit is contained in:
Amaan Qureshi 2026-03-27 18:04:19 -04:00
commit 2d4bb2e371
No known key found for this signature in database
2 changed files with 54 additions and 147 deletions

View file

@ -98,12 +98,15 @@ pub const unsafe extern "C" fn strlen(s: *const u8) -> usize {
/// Function pointer type for the main application entry point. /// Function pointer type for the main application entry point.
/// The function receives argc and argv and should return an exit code. /// The function receives argc and argv and should return an exit code.
#[cfg(not(test))]
pub type MainFn = unsafe extern "C" fn(i32, *const *const u8) -> i32; pub type MainFn = unsafe extern "C" fn(i32, *const *const u8) -> i32;
#[cfg(not(test))]
static mut MAIN_FN: Option<MainFn> = None; static mut MAIN_FN: Option<MainFn> = None;
/// Register the main function to be called from the entry point. /// Register the main function to be called from the entry point.
/// This must be called before the program starts (e.g., in a constructor). /// This must be called before the program starts (e.g., in a constructor).
#[cfg(not(test))]
pub fn register_main(main_fn: MainFn) { pub fn register_main(main_fn: MainFn) {
unsafe { unsafe {
MAIN_FN = Some(main_fn); MAIN_FN = Some(main_fn);
@ -125,6 +128,7 @@ pub fn register_main(main_fn: MainFn) {
/// ```rust,ignore /// ```rust,ignore
/// unsafe extern "C" fn main(argc: i32, argv: *const *const u8) -> i32` /// unsafe extern "C" fn main(argc: i32, argv: *const *const u8) -> i32`
/// ``` /// ```
#[cfg(not(test))]
#[unsafe(no_mangle)] #[unsafe(no_mangle)]
pub unsafe extern "C" fn entry_rust(stack: *const usize) -> i32 { pub unsafe extern "C" fn entry_rust(stack: *const usize) -> i32 {
// Read argc and argv from stack // Read argc and argv from stack
@ -141,155 +145,11 @@ pub unsafe extern "C" fn entry_rust(stack: *const usize) -> i32 {
// External main function that must be defined by the binary using this crate. // External main function that must be defined by the binary using this crate.
// Signature: `unsafe extern "C" fn main(argc: i32, argv: *const *const u8) -> // Signature: `unsafe extern "C" fn main(argc: i32, argv: *const *const u8) ->
// i32` // i32`
#[cfg(not(test))]
unsafe extern "C" { unsafe extern "C" {
fn main(argc: i32, argv: *const *const u8) -> i32; fn main(argc: i32, argv: *const *const u8) -> i32;
} }
#[cfg(target_arch = "x86_64")]
mod entry {
use core::arch::naked_asm;
/// Entry point that receives stack pointer directly from kernel.
/// On `x86_64` Linux at program start:
///
/// - `[rsp]` = argc
/// - `[rsp+8]` = argv[0]
/// - `[rsp+16]` = argv[1]
/// - ...
/// - `[rsp+8n]` = NULL
/// - `[rsp+8n+8]` = envp[0]
///
/// # Safety
///
/// This is a naked function with no prologue or epilogue. It directly
/// manipulates the stack pointer (`rsp`) and assumes it was called by the
/// kernel with a valid stack containing argc and argv. The function:
///
/// - Reads from `[rsp]` without validating the pointer
/// - Modifies `rsp` directly (16-byte alignment)
/// - Does not preserve any registers
/// - Does not return normally (exits via syscall)
///
/// This function MUST only be used as the program entry point (`_start`).
/// Calling it from any other context is undefined behavior. This has been
/// your safety notice. I WILL put UB in your Rust program.
#[unsafe(no_mangle)]
#[unsafe(naked)]
pub unsafe extern "C" fn _start() {
naked_asm!(
// Move stack pointer to first argument register
"mov rdi, rsp",
// Align stack to 16-byte boundary (System V AMD64 ABI requirement)
"and rsp, -16",
// Call into Rust code
"call {entry_rust}",
// Move return code to syscall argument
"mov rdi, rax",
// Exit syscall
"mov rax, 60", // SYS_exit
"syscall",
entry_rust = sym super::entry_rust,
);
}
}
#[cfg(target_arch = "aarch64")]
mod entry {
use core::arch::naked_asm;
/// Entry point that receives stack pointer directly from kernel.
/// On `aarch64` Linux at program start, the stack layout is identical
/// to x86_64:
///
/// - `[sp]` = argc
/// - `[sp+8]` = argv[0]
/// - ...
///
/// # Safety
///
/// This is a naked function with no prologue or epilogue. It directly
/// manipulates the stack pointer (`sp`) and assumes it was called by the
/// kernel with a valid stack containing argc and argv. The function:
///
/// - Reads from `[sp]` without validating the pointer
/// - Modifies `sp` directly (16-byte alignment)
/// - Does not preserve any registers
/// - Does not return normally (exits via SVC instruction)
///
/// This function MUST only be used as the program entry point (`_start`).
/// Calling it from any other context is undefined behavior.
#[unsafe(no_mangle)]
#[unsafe(naked)]
pub unsafe extern "C" fn _start() {
naked_asm!(
// Move stack pointer to first argument register
"mov x0, sp",
// Align stack to 16-byte boundary (AArch64 ABI requirement)
"mov x9, sp",
"and x9, x9, #-16",
"mov sp, x9",
// Call into Rust code
"bl {entry_rust}",
// Move return code to syscall argument
"mov x0, x0",
// Exit syscall
"mov x8, 93", // SYS_exit
"svc #0",
entry_rust = sym super::entry_rust,
);
}
}
#[cfg(target_arch = "riscv64")]
mod entry {
use core::arch::naked_asm;
/// Entry point that receives stack pointer directly from kernel.
/// On `riscv64` Linux at program start, the stack layout is identical
/// to x86_64:
///
/// - `[sp]` = argc
/// - `[sp+8]` = argv[0]
/// - ...
///
/// # Safety
///
/// This is a naked function with no prologue or epilogue. It directly
/// manipulates the stack pointer (`sp`) and assumes it was called by the
/// kernel with a valid stack containing argc and argv. The function:
///
/// - Reads from `[sp]` without validating the pointer
/// - Modifies `sp` directly (16-byte alignment)
/// - Does not preserve any registers
/// - Does not return normally (exits via ECALL instruction)
///
/// This function MUST only be used as the program entry point (`_start`).
/// Calling it from any other context is undefined behavior.
#[unsafe(no_mangle)]
#[unsafe(naked)]
pub unsafe extern "C" fn _start() {
naked_asm!(
// Move stack pointer to first argument register
"mv a0, sp",
// Align stack to 16-byte boundary (RISC-V ABI requirement)
"andi sp, sp, -16",
// Call into Rust code
"call {entry_rust}",
// Move return code to syscall argument
"mv a0, a0",
// Exit syscall
"li a7, 93", // SYS_exit
"ecall",
entry_rust = sym super::entry_rust,
);
}
}
// Re-export the entry point
#[cfg(target_arch = "x86_64")] pub use entry::_start;
#[cfg(target_arch = "aarch64")] pub use entry::_start;
#[cfg(target_arch = "riscv64")] pub use entry::_start;
/// Direct syscall to open a file /// Direct syscall to open a file
/// ///
/// # Returns /// # Returns

View file

@ -3,12 +3,59 @@
extern crate alloc; extern crate alloc;
use core::panic::PanicInfo; use core::{arch::naked_asm, panic::PanicInfo};
use microfetch_alloc::BumpAllocator; use microfetch_alloc::BumpAllocator;
// Re-export libc replacement functions from asm crate // Re-export libc replacement functions from asm crate
pub use microfetch_asm::{memcpy, memset, strlen}; pub use microfetch_asm::{memcpy, memset, strlen};
use microfetch_asm::{sys_exit, sys_write}; use microfetch_asm::{entry_rust, sys_exit, sys_write};
#[cfg(target_arch = "x86_64")]
#[unsafe(no_mangle)]
#[unsafe(naked)]
unsafe extern "C" fn _start() {
naked_asm!(
"mov rdi, rsp",
"and rsp, -16",
"call {entry_rust}",
"mov rdi, rax",
"mov rax, 60",
"syscall",
entry_rust = sym entry_rust,
);
}
#[cfg(target_arch = "aarch64")]
#[unsafe(no_mangle)]
#[unsafe(naked)]
unsafe extern "C" fn _start() {
naked_asm!(
"mov x0, sp",
"mov x9, sp",
"and x9, x9, #-16",
"mov sp, x9",
"bl {entry_rust}",
"mov x0, x0",
"mov x8, 93",
"svc #0",
entry_rust = sym entry_rust,
);
}
#[cfg(target_arch = "riscv64")]
#[unsafe(no_mangle)]
#[unsafe(naked)]
unsafe extern "C" fn _start() {
naked_asm!(
"mv a0, sp",
"andi sp, sp, -16",
"call {entry_rust}",
"mv a0, a0",
"li a7, 93",
"ecall",
entry_rust = sym entry_rust,
);
}
// Global allocator // Global allocator
#[global_allocator] #[global_allocator]