x86_64: more optimized versions of mem{cpy,set}

Signed-off-by: Anhad Singh <[email protected]>

Author: Andy-Python-Programmer

src/aero_kernel/src/arch/x86_64/mem.rs

@@ -0,0 +1,83 @@
+// Copyright (C) 2021-2023 The Aero Project Developers.
+//
+// This file is part of The Aero Project.
+//
+// Aero is free software: you can redistribute it and/or modify
+// it under the terms of the GNU General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// Aero is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU General Public License for more details.
+//
+// You should have received a copy of the GNU General Public License
+// along with Aero. If not, see <https://www.gnu.org/licenses/>.
+
+#[no_mangle]
+#[naked]
+unsafe extern "C" fn memcpy_movsq(dest: *mut u8, src: *const u8, len: usize) -> *mut u8 {
+    // Registers used:
+    //
+    // %rdi = argument 1, `dest`
+    // %rsi = argument 2, `src`
+    // %rdx = argument 3, `len`
+    asm!(
+        // Save the return value.
+        "mov rax, rdi",
+        // Copy in 8 byte chunks.
+        "mov rcx, rdx",
+        "shr rcx, 3",
+        "rep movsq",
+        // Copy the rest.
+        "mov rcx, rdx",
+        "and rcx, 0x7",
+        "rep movsb",
+        "ret",
+        options(noreturn)
+    );
+}
+
+#[no_mangle]
+#[naked]
+unsafe extern "C" fn memset_stosq(dest: *mut u8, byte: i32, len: usize) -> *mut u8 {
+    // Registers used:
+    //
+    // %rdi = argument 1, `dest`
+    // %rsi = argument 2, `byte`
+    // %rdx = argument 3, `len`
+    asm!(
+        // Save the return value.
+        "mov r11, rdi",
+        // Create an 8-byte copy of the pattern.
+        "mov rcx, rdx",
+        "movzx rax, sil",
+        "mov r10, 0x0101010101010101",
+        "mul r10",
+        "mov rdx, rcx",
+        // Copy in 8 byte chunks.
+        "shr rcx, 3",
+        "rep stosq",
+        // Copy the rest.
+        "mov rcx, rdx",
+        "and rcx, 0x7",
+        "rep stosb",
+        // Restore the return value.
+        "mov rax, r11",
+        "ret",
+        options(noreturn)
+    )
+}
+
+// FIXME(andypython): pick the best implementation for the current CPU using indirect functions.
+
+#[no_mangle]
+extern "C" fn memcpy(dest: *mut u8, src: *const u8, len: usize) -> *mut u8 {
+    unsafe { memcpy_movsq(dest, src, len) }
+}
+
+#[no_mangle]
+extern "C" fn memset(dest: *mut u8, byte: i32, len: usize) -> *mut u8 {
+    unsafe { memset_stosq(dest, byte, len) }
+}

src/aero_kernel/src/arch/x86_64/mod.rs

@@ -20,6 +20,7 @@ pub mod controlregs;
 pub mod gdt;
 pub mod interrupts;
 pub mod io;
+pub mod mem;
 pub mod signals;
 pub mod syscall;
 pub mod task;
@@ -32,7 +33,6 @@ use core::sync::atomic::Ordering;
 use crate::acpi::aml;
 use crate::{acpi, cmdline};
 
-use crate::mem;
 use crate::mem::paging;
 use crate::mem::paging::VirtAddr;
 
@@ -136,7 +136,7 @@ extern "C" fn arch_aero_main() -> ! {
     paging::init(memmap).unwrap();
     log::info!("loaded paging");
 
-    mem::alloc::init_heap();
+    crate::mem::alloc::init_heap();
     log::info!("loaded heap");
 
     // SMP initialization.

src/aero_kernel/src/arch/x86_64/user_copy.rs

@@ -46,10 +46,10 @@ unsafe extern "C" fn copy_to_from_user(
 ) -> bool {
     // Registers used:
     //
-    // %rax = argument 1, `dest`
+    // %rdi = argument 1, `dest`
     // %rsi = argument 2, `src`
     // %rdx = argument 3, `size`
-    // %rcx = argument 4, `fault_resume`
+    // %rcx = argument 4, `fault_resume` (copied to %r10)
     asm!(
         // Copy `fault_resume` out of %rcx because it will be utilized by `rep movsb` latter.
         "mov r10, rcx",