Move some code from the trampoline to global_asm!()

src/acpi/madt/arch/x86.rs

@@ -82,8 +82,8 @@ pub(super) fn init(madt: Madt) {
                 let idt_ptr = crate::arch::idt::allocate_and_init_idt(cpu_id);
 
                 let args = KernelArgsAp {
+                    stack_end: stack_end as *mut u8,
                     cpu_id,
-                    page_table: page_table_physaddr,
                     pcr_ptr,
                     idt_ptr,
                 };
@@ -91,16 +91,13 @@ pub(super) fn init(madt: Madt) {
                 let ap_ready = (TRAMPOLINE + 8) as *mut u64;
                 let ap_args_ptr = unsafe { ap_ready.add(1) };
                 let ap_page_table = unsafe { ap_ready.add(2) };
-                let ap_stack_end = unsafe { ap_ready.add(3) };
-                let ap_code = unsafe { ap_ready.add(4) };
+                let ap_code = unsafe { ap_ready.add(3) };
 
                 // Set the ap_ready to 0, volatile
                 unsafe {
                     ap_ready.write(0);
                     ap_args_ptr.write(&args as *const _ as u64);
                     ap_page_table.write(page_table_physaddr as u64);
-                    ap_stack_end.write(stack_end as u64);
-                    #[expect(clippy::fn_to_numeric_cast)]
                     ap_code.write(kstart_ap as u64);
 
                     // TODO: Is this necessary (this fence)?
@@ -127,7 +124,6 @@ pub(super) fn init(madt: Madt) {
 
                 // Send START IPI
                 {
-                    //Start at 0x0800:0000 => 0x8000. Hopefully the bootloader code is still there
                     let ap_segment = (TRAMPOLINE >> 12) & 0xFF;
                     let mut icr = 0x4600 | ap_segment as u64;
 

src/arch/x86_shared/start.rs

@@ -6,6 +6,7 @@ use core::{
     arch::global_asm,
     cell::SyncUnsafeCell,
     hint,
+    mem::offset_of,
     sync::atomic::{AtomicBool, Ordering},
 };
 
@@ -13,13 +14,8 @@ use core::{
 use crate::acpi;
 
 use crate::{
-    allocator,
-    cpu_set::LogicalCpuId,
-    device,
-    devices::graphical_debug,
-    gdt, idt, interrupt,
-    paging::{self, PhysicalAddress, RmmA, RmmArch, TableKind},
-    startup::KernelArgs,
+    allocator, cpu_set::LogicalCpuId, device, devices::graphical_debug, gdt, idt, interrupt,
+    paging, startup::KernelArgs,
 };
 
 /// Test of zero values in BSS.
@@ -171,14 +167,51 @@ unsafe extern "C" fn start(args_ptr: *const KernelArgs, stack_end: usize) -> ! {
 }
 
 pub struct KernelArgsAp {
+    pub stack_end: *mut u8,
     pub cpu_id: LogicalCpuId,
-    pub page_table: usize,
     pub pcr_ptr: *mut gdt::ProcessorControlRegion,
     pub idt_ptr: *mut idt::Idt,
 }
 
+// FIXME use extern "custom"
+unsafe extern "C" {
+    pub fn kstart_ap();
+}
+
+#[cfg(target_arch = "x86")]
+global_asm!("
+    .globl kstart_ap
+    kstart_ap:
+        mov esp, dword ptr [edi + {args_stack}]
+        mov [esp + 4], edi
+        mov [esp + 8], esp
+
+        jmp {start_ap}
+    ",
+    args_stack = const offset_of!(KernelArgsAp, stack_end),
+    start_ap = sym start_ap,
+);
+
+#[cfg(target_arch = "x86_64")]
+global_asm!("
+    .globl kstart_ap
+    kstart_ap:
+        // Note: The System V ABI requires the stack to be aligned to 16 bytes
+        // before the call instruction. As we jump rather than call it has to
+        // be offset by 8 bytes. Additionally reserve a bit more space at the
+        // end of the stack to ensure that the start function returns to
+        // address 0.
+        mov rax, qword ptr [rdi + {args_stack}]
+        lea rsp, [rax - 24]
+
+        jmp {start_ap}
+    ",
+    args_stack = const offset_of!(KernelArgsAp, stack_end),
+    start_ap = sym start_ap,
+);
+
 /// Entry to rust for an AP
-pub unsafe extern "C" fn kstart_ap(args_ptr: *const KernelArgsAp) -> ! {
+unsafe extern "C" fn start_ap(args_ptr: *const KernelArgsAp) -> ! {
     unsafe {
         let cpu_id = {
             let args = &*args_ptr;
@@ -190,7 +223,6 @@ pub unsafe extern "C" fn kstart_ap(args_ptr: *const KernelArgsAp) -> ! {
             idt::install_idt(args.idt_ptr);
 
             // Initialize paging
-            RmmA::set_table(TableKind::Kernel, PhysicalAddress::new(args.page_table));
             paging::init();
 
             #[cfg(all(target_arch = "x86_64", feature = "profiling"))]

src/asm/x86/trampoline.asm

@@ -11,7 +11,6 @@ trampoline:
     .ready: dq 0
     .args_ptr: dq 0
     .page_table: dq 0
-    .stack_end: dq 0
     .code: dq 0
 
 startup_ap:
@@ -68,19 +67,11 @@ protected_mode_ap:
     mov gs, eax
     mov ss, eax
 
-    mov eax, [trampoline.stack_end]
-    lea esp, [eax - 256]
-
-    mov eax, [trampoline.args_ptr]
-    push eax
+    mov edi, [trampoline.args_ptr]
 
     mov eax, [trampoline.code]
     mov dword [trampoline.ready], 1
-    call eax
-.halt:
-    cli
-    hlt
-    jmp .halt
+    jmp eax
 
 struc GDTEntry
     .limitl resw 1

src/asm/x86_64/trampoline.asm

@@ -11,7 +11,6 @@ trampoline:
     .ready: dq 0
     .args_ptr: dq 0
     .page_table: dq 0
-    .stack_end: dq 0
     .code: dq 0
 
 startup_ap:
@@ -75,9 +74,6 @@ long_mode_ap:
     mov gs, rax
     mov ss, rax
 
-    mov rcx, [trampoline.stack_end]
-    lea rsp, [rcx - 256]
-
     mov rdi, [trampoline.args_ptr]
 
     mov rax, [trampoline.code]