feat: add aarch64 user support

src/aarch64/asm.rs

@@ -64,7 +64,8 @@ pub fn read_kernel_page_table() -> PhysAddr {
 /// Reads the current page table root register for user space (`TTBR0_EL1`).
 ///
 /// When the "arm-el2" feature is enabled, for user-mode programs,
-/// virtualization is completely transparent to them, so there is no need to modify
+/// virtualization is completely transparent to them, so there is no need to
+/// modify
 ///
 /// Returns the physical address of the page table root.
 #[inline]
@@ -88,21 +89,24 @@ pub fn read_user_page_table() -> PhysAddr {
 pub unsafe fn write_kernel_page_table(root_paddr: PhysAddr) {
     #[cfg(not(feature = "arm-el2"))]
     {
-        // kernel space page table use TTBR1 (0xffff_0000_0000_0000..0xffff_ffff_ffff_ffff)
+        // kernel space page table use TTBR1
+        // (0xffff_0000_0000_0000..0xffff_ffff_ffff_ffff)
         TTBR1_EL1.set(root_paddr.as_usize() as _);
     }
 
     #[cfg(feature = "arm-el2")]
     {
-        // kernel space page table at EL2 use TTBR0_EL2 (0x0000_0000_0000_0000..0x0000_ffff_ffff_ffff)
+        // kernel space page table at EL2 use TTBR0_EL2
+        // (0x0000_0000_0000_0000..0x0000_ffff_ffff_ffff)
         TTBR0_EL2.set(root_paddr.as_usize() as _);
     }
 }
 
 /// Writes the register to update the current page table root for user space
 /// (`TTBR1_EL0`).
 /// When the "arm-el2" feature is enabled, for user-mode programs,
-/// virtualization is completely transparent to them, so there is no need to modify
+/// virtualization is completely transparent to them, so there is no need to
+/// modify
 ///
 /// Note that the TLB is **NOT** flushed after this operation.
 ///
@@ -201,3 +205,19 @@ pub fn enable_fp() {
     CPACR_EL1.write(CPACR_EL1::FPEN::TrapNothing);
     barrier::isb(barrier::SY);
 }
+
+core::arch::global_asm!(include_str!("user_copy.S"));
+
+extern "C" {
+    /// Copy data from user space to kernel space safely.
+    ///
+    /// # Arguments
+    /// * `dst` - Destination pointer in kernel space
+    /// * `src` - Source pointer in user space
+    /// * `size` - Number of bytes to copy
+    ///
+    /// # Returns
+    /// * `0` - Success, all bytes copied
+    /// * `>0` - Number of bytes that could not be copied due to page fault
+    pub fn user_copy(dst: *mut u8, src: *const u8, size: usize) -> usize;
+}

src/aarch64/context.rs

@@ -124,6 +124,11 @@ impl TrapFrame {
         self.r[0] as _
     }
 
+    // TODO:
+    pub const fn sysno(&self) -> usize {
+        self.r[8] as usize
+    }
+
     /// Sets the return value register.
     pub const fn set_retval(&mut self, r0: usize) {
         self.r[0] = r0 as _;

src/aarch64/trap.S

@@ -1,5 +1,4 @@
 .macro SAVE_REGS
-    sub     sp, sp, {trapframe_size}
     stp     x0, x1, [sp]
     stp     x2, x3, [sp, 2 * 8]
     stp     x4, x5, [sp, 4 * 8]
@@ -36,7 +35,7 @@
     # backup kernel tpidr_el0
     mrs     x1, tpidr_el0
     msr     tpidrro_el0, x1
-    
+
     ldp     x11, x12, [sp, 33 * 8]
     ldp     x9, x10, [sp, 31 * 8]
     msr     sp_el0, x9
@@ -65,6 +64,7 @@
 
 .macro INVALID_EXCP, kind, source
 .p2align 7
+    sub     sp, sp, {trapframe_size}
     SAVE_REGS
     mov     x0, sp
     mov     x1, \kind
@@ -73,8 +73,20 @@
     b       .Lexception_return
 .endm
 
+.macro TASK_EXIT, kind
+.p2align 7
+    SAVE_REGS
+    mov   x0, \kind
+    b _user_trap_entry
+
+.endm
+
+.Ldeadloop:
+    b .Ldeadloop 
+
 .macro HANDLE_SYNC
 .p2align 7
+    sub     sp, sp, {trapframe_size}
     SAVE_REGS
     mov     x0, sp
     bl      handle_sync_exception
@@ -83,6 +95,7 @@
 
 .macro HANDLE_IRQ
 .p2align 7
+    sub     sp, sp, {trapframe_size}
     SAVE_REGS
     mov     x0, sp
     bl      handle_irq_exception
@@ -93,30 +106,38 @@
 .p2align 11
 .global exception_vector_base
 exception_vector_base:
-    // current EL, with SP_EL0
-    INVALID_EXCP 0 0
-    INVALID_EXCP 1 0
-    INVALID_EXCP 2 0
-    INVALID_EXCP 3 0
-
-    // current EL, with SP_ELx
-    HANDLE_SYNC
-    HANDLE_IRQ
-    INVALID_EXCP 2 1
-    INVALID_EXCP 3 1
-
-    // TODO: handle user trap
-    // lower EL, aarch64
-    HANDLE_SYNC
-    HANDLE_IRQ
-    INVALID_EXCP 2 2
-    INVALID_EXCP 3 2
-
-    // lower EL, aarch32
-    INVALID_EXCP 0 3
-    INVALID_EXCP 1 3
-    INVALID_EXCP 2 3
-    INVALID_EXCP 3 3
+curr_el_sp0_sync:
+    INVALID_EXCP 0 0               
+curr_el_sp0_irq:
+    INVALID_EXCP 1 0             
+curr_el_sp0_fiq:
+    INVALID_EXCP 2 0               
+curr_el_sp0_serror:
+    INVALID_EXCP 3 0               
+curr_el_spx_sync:
+    HANDLE_SYNC                    
+curr_el_spx_irq:
+    HANDLE_IRQ                 
+curr_el_spx_fiq:
+    INVALID_EXCP 2 1               
+curr_el_spx_serror:
+    INVALID_EXCP 3 1              
+lower_el_aarch64_sync:
+    TASK_EXIT {kind_sync}               
+lower_el_aarch64_irq:
+    TASK_EXIT {kind_irq}                  
+lower_el_aarch64_fiq:
+    INVALID_EXCP 2 2                
+lower_el_aarch64_serror:
+    INVALID_EXCP 3 2                
+lower_el_aarch32_sync:
+    INVALID_EXCP 0 3              
+lower_el_aarch32_irq:
+    INVALID_EXCP 1 3              
+lower_el_aarch32_fiq:
+    INVALID_EXCP 2 3              
+lower_el_aarch32_serror:
+    INVALID_EXCP 3 3           
 
 .Lexception_return:
     RESTORE_REGS

src/aarch64/trap.rs

@@ -6,17 +6,19 @@ use crate::trap::PageFaultFlags;
 
 core::arch::global_asm!(
     include_str!("trap.S"),
-    trapframe_size = const core::mem::size_of::<TrapFrame>()
+    trapframe_size = const core::mem::size_of::<TrapFrame>(),
+    kind_irq = const TrapKind::Irq as u8,
+    kind_sync = const TrapKind::Synchronous as u8,
 );
 
 #[repr(u8)]
 #[derive(Debug)]
 #[allow(dead_code)]
-enum TrapKind {
+pub(crate) enum TrapKind {
     Synchronous = 0,
-    Irq = 1,
-    Fiq = 2,
-    SError = 3,
+    Irq         = 1,
+    Fiq         = 2,
+    SError      = 3,
 }
 
 #[repr(u8)]

src/aarch64/user_copy.S

@@ -0,0 +1,76 @@
+// AArch64 user space safe memory copy
+// Based on RISC-V implementation and optimized for AArch64
+
+.section .text
+.global user_copy
+user_copy:
+	// x0 = dst, x1 = src, x2 = size
+	// Return: 0 on success, remaining bytes on fault
+	
+	// Quick exit for zero size
+	cbz	x2, .Lsuccess
+	
+	// Save the end address for fault handling
+	add	x3, x0, x2	// x3 = dst_end
+	mov	x4, x2		// x4 = original_size (for fault calculation)
+	
+	// Use different strategies based on size
+	cmp	x2, #64
+	b.lo	.Lbyte_copy
+	
+	// For larger copies, try to align and use word copy
+	// First align dst to 8-byte boundary
+	and	x5, x0, #7
+	cbz	x5, .Laligned_copy
+	
+	// Copy bytes until dst is 8-byte aligned
+	sub	x5, x2, x5	// remaining after alignment
+.Lalign_loop:
+	ldrb	w6, [x1], #1
+	strb	w6, [x0], #1
+	tbnz	x0, #0, .Lalign_loop
+	tbnz	x0, #1, .Lalign_loop  
+	tbnz	x0, #2, .Lalign_loop
+	mov	x2, x5		// update remaining size
+	
+.Laligned_copy:
+	// Now dst is 8-byte aligned, do word copy
+	cmp	x2, #8
+	b.lo	.Lbyte_copy
+	
+	// Unrolled 8-word (64-byte) copy loop
+	sub	x5, x2, #64
+.Lword_loop:
+	ldp	x6, x7, [x1], #16
+	ldp	x8, x9, [x1], #16  
+	ldp	x10, x11, [x1], #16
+	ldp	x12, x13, [x1], #16
+	stp	x6, x7, [x0], #16
+	stp	x8, x9, [x0], #16
+	stp	x10, x11, [x0], #16
+	stp	x12, x13, [x0], #16
+	subs	x2, x2, #64
+	cmp	x2, #64
+	b.hs	.Lword_loop
+	
+	// Handle remaining words (8-byte chunks)
+.Lword_remainder:
+	cmp	x2, #8
+	b.lo	.Lbyte_copy
+	ldr	x6, [x1], #8
+	str	x6, [x0], #8
+	sub	x2, x2, #8
+	b	.Lword_remainder
+	
+.Lbyte_copy:
+	// Copy remaining bytes
+	cbz	x2, .Lsuccess
+.Lbyte_loop:
+	ldrb	w6, [x1], #1
+	strb	w6, [x0], #1
+	subs	x2, x2, #1
+	b.ne	.Lbyte_loop
+	
+.Lsuccess:
+	mov	x0, #0
+	ret