Kernel: Don't load AP init code to 0xF000, but relocate it later

This cleans up the kernel executable as bootloaders don't have to
load AP init code straight to 0xF000, but it will be moved there once
kernel is doing the AP initialization.

kernel/arch/i686/Signal.S

@@ -1,4 +1,4 @@
-.section .userspace, "aw"
+.section .userspace, "ax"
 
 // stack contains
 //    return address

kernel/arch/i686/boot.S

@@ -235,26 +235,31 @@ system_halt:
 	jmp 1b
 
 
+#define AP_V2P(vaddr) ((vaddr) - ap_trampoline + 0xF000)
+
 .section .ap_init, "ax"
 
 .code16
 .global ap_trampoline
 ap_trampoline:
 	jmp 1f
+
 .align 8
 ap_stack_ptr:
 	.skip 4
-1:
+ap_stack_loaded:
-	cli; cld
+	.skip 1
-	ljmpl $0x00, $ap_cs_clear
-ap_cs_clear:
 
+1:	cli; cld
+	ljmpl $0x00, $AP_V2P(ap_cs_clear)
+
+ap_cs_clear:
 	# load ap gdt and enter protected mode
-	lgdt ap_gdtr
+	lgdt AP_V2P(ap_gdtr)
 	movl %cr0, %eax
 	orb $1, %al
 	movl %eax, %cr0
-	ljmpl $0x08, $ap_protected_mode
+	ljmpl $0x08, $AP_V2P(ap_protected_mode)
 
 .code32
 ap_protected_mode:
@@ -263,16 +268,15 @@ ap_protected_mode:
 	movw %ax, %ss
 	movw %ax, %es
 
-	movl ap_stack_ptr, %esp
+	movl AP_V2P(ap_stack_ptr), %esp
-	movb $1, V2P(g_ap_stack_loaded)
+	movb $1, AP_V2P(ap_stack_loaded)
 
-	call V2P(enable_sse)
+	leal V2P(enable_sse),        %ecx; call *%ecx
-
+	leal V2P(initialize_paging), %ecx; call *%ecx
-	call V2P(initialize_paging)
 
 	# load boot gdt and enter long mode
 	lgdt V2P(boot_gdtr)
-	ljmpl $0x08, $ap_flush_gdt
+	ljmpl $0x08, $AP_V2P(ap_flush_gdt)
 
 ap_flush_gdt:
 	# move stack pointer to higher half

kernel/arch/i686/linker.ld

@@ -4,13 +4,6 @@ KERNEL_OFFSET = 0xC0000000;
 
 SECTIONS
 {
-	. = 0xF000;
-	.ap_init ALIGN(4K) : AT(ADDR(.ap_init))
-	{
-		g_ap_init_addr = .;
-		*(.ap_init)
-	}
-
 	. = 0x00100000 + KERNEL_OFFSET;
 
 	g_kernel_start = .;
@@ -28,15 +21,20 @@ SECTIONS
 		g_userspace_end = .;
 		g_kernel_execute_end = .;
 	}
-	.rodata ALIGN(4K) : AT(ADDR(.rodata) - KERNEL_OFFSET)
+	.ap_init ALIGN(4K) : AT(ADDR(.ap_init))
 	{
-		*(.rodata.*)
+		g_ap_init_addr = .;
+		*(.ap_init)
 	}
 	.data ALIGN(4K) : AT(ADDR(.data) - KERNEL_OFFSET)
 	{
 		g_kernel_writable_start = .;
 		*(.data)
 	}
+	.rodata ALIGN(4K) : AT(ADDR(.rodata) - KERNEL_OFFSET)
+	{
+		*(.rodata.*)
+	}
 	.bss ALIGN(4K) : AT(ADDR(.bss) - KERNEL_OFFSET)
 	{
 		*(COMMON)

kernel/arch/x86_64/Signal.S

@@ -1,4 +1,4 @@
-.section .userspace, "aw"
+.section .userspace, "ax"
 
 // stack contains
 //    return address

kernel/arch/x86_64/boot.S

@@ -108,9 +108,6 @@ g_ap_startup_done:
 .global g_ap_running_count
 g_ap_running_count:
 	.byte 0
-.global g_ap_stack_loaded
-g_ap_stack_loaded:
-	.byte 0
 
 .section .text
 
@@ -194,7 +191,6 @@ _start:
 
 	call check_requirements
 	call enable_sse
-
 	call initialize_paging
 
 	# flush gdt and jump to 64 bit
@@ -237,26 +233,31 @@ system_halt:
 	jmp 1b
 
 
+#define AP_V2P(vaddr) ((vaddr) - ap_trampoline + 0xF000)
+
 .section .ap_init, "ax"
 
 .code16
 .global ap_trampoline
 ap_trampoline:
 	jmp 1f
+
 .align 8
 ap_stack_ptr:
 	.skip 4
-1:
+ap_stack_loaded:
-	cli; cld
+	.skip 1
-	ljmpl $0x00, $ap_cs_clear
-ap_cs_clear:
 
+1:	cli; cld
+	ljmpl $0x00, $AP_V2P(ap_cs_clear)
+
+ap_cs_clear:
 	# load ap gdt and enter protected mode
-	lgdt ap_gdtr
+	lgdt AP_V2P(ap_gdtr)
 	movl %cr0, %eax
 	orb $1, %al
 	movl %eax, %cr0
-	ljmpl $0x08, $ap_protected_mode
+	ljmpl $0x08, $AP_V2P(ap_protected_mode)
 
 .code32
 ap_protected_mode:
@@ -265,16 +266,15 @@ ap_protected_mode:
 	movw %ax, %ss
 	movw %ax, %es
 
-	movl ap_stack_ptr, %esp
+	movl AP_V2P(ap_stack_ptr), %esp
-	movb $1, V2P(g_ap_stack_loaded)
+	movb $1, AP_V2P(ap_stack_loaded)
 
-	call V2P(enable_sse)
+	leal V2P(enable_sse),        %ecx; call *%ecx
-
+	leal V2P(initialize_paging), %ecx; call *%ecx
-	call V2P(initialize_paging)
 
 	# load boot gdt and enter long mode
 	lgdt V2P(boot_gdtr)
-	ljmpl $0x08, $ap_long_mode
+	ljmpl $0x08, $AP_V2P(ap_long_mode)
 
 .code64
 ap_long_mode:
@@ -282,22 +282,20 @@ ap_long_mode:
 	movl %esp, %esp
 	addq $KERNEL_OFFSET, %rsp
 
-	# jump to higher half
-	movabsq $ap_higher_half, %rcx
-	jmp *%rcx
-
-ap_higher_half:
 	# clear rbp for stacktrace
 	xorq %rbp, %rbp
 
+	xorb %al, %al
 1:	pause
-	cmpb $0, g_ap_startup_done
+	cmpb %al, g_ap_startup_done
 	jz 1b
 
 	lock incb g_ap_running_count
 
-	call ap_main
+	# jump to ap_main in higher half
-	jmp system_halt
+	movabsq $ap_main, %rcx
+	call *%rcx
+	jmp V2P(system_halt)
 
 ap_gdt:
 	.quad 0x0000000000000000 # null descriptor

kernel/arch/x86_64/linker.ld

@@ -4,13 +4,6 @@ KERNEL_OFFSET = 0xFFFFFFFF80000000;
 
 SECTIONS
 {
-	. = 0xF000;
-	.ap_init ALIGN(4K) : AT(ADDR(.ap_init))
-	{
-		g_ap_init_addr = .;
-		*(.ap_init)
-	}
-
 	. = 0x00100000 + KERNEL_OFFSET;
 
 	g_kernel_start = .;
@@ -28,15 +21,20 @@ SECTIONS
 		g_userspace_end = .;
 		g_kernel_execute_end = .;
 	}
-	.rodata ALIGN(4K) : AT(ADDR(.rodata) - KERNEL_OFFSET)
+	.ap_init ALIGN(4K) : AT(ADDR(.ap_init))
 	{
-		*(.rodata.*)
+		g_ap_init_addr = .;
+		*(.ap_init)
 	}
 	.data ALIGN(4K) : AT(ADDR(.data) - KERNEL_OFFSET)
 	{
 		g_kernel_writable_start = .;
 		*(.data)
 	}
+	.rodata ALIGN(4K) : AT(ADDR(.rodata) - KERNEL_OFFSET)
+	{
+		*(.rodata.*)
+	}
 	.bss ALIGN(4K) : AT(ADDR(.bss) - KERNEL_OFFSET)
 	{
 		*(COMMON)

kernel/kernel/APIC.cpp

@@ -31,7 +31,6 @@ extern uint8_t g_ap_init_addr[];
 
 extern volatile uint8_t g_ap_startup_done[];
 extern volatile uint8_t g_ap_running_count[];
-extern volatile uint8_t g_ap_stack_loaded[];
 
 namespace Kernel
 {
@@ -261,8 +260,6 @@ namespace Kernel
 					__builtin_ia32_pause();
 			};
 
-		const size_t ap_init_page = reinterpret_cast<vaddr_t>(g_ap_init_addr) / PAGE_SIZE;
-
 		dprintln("System has {} processors", m_processors.size());
 
 		uint8_t bsp_id = Kernel::Processor::current_id().as_u32();
@@ -275,6 +272,11 @@ namespace Kernel
 			return;
 		}
 
+		constexpr paddr_t ap_init_paddr = 0xF000;
+		PageTable::with_fast_page(ap_init_paddr, [&] {
+			memcpy(PageTable::fast_page_as_ptr(), g_ap_init_addr, PAGE_SIZE);
+		});
+
 		for (auto& processor : m_processors)
 		{
 			if (processor.apic_id == bsp_id)
@@ -289,10 +291,10 @@ namespace Kernel
 			dprintln("Trying to enable processor (lapic id {})", processor.apic_id);
 
 			auto& proc = Kernel::Processor::create(ProcessorID(processor.apic_id));
-			PageTable::with_fast_page((paddr_t)g_ap_init_addr, [&] {
+			PageTable::with_fast_page(ap_init_paddr, [&] {
-				PageTable::fast_page_as_sized<uint32_t>(2) = V2P(proc.stack_top());
+				PageTable::fast_page_as_sized<uint32_t>(2) = kmalloc_paddr_of(proc.stack_top()).value();
+				PageTable::fast_page_as_sized<uint8_t>(13) = 0;
 			});
-			*g_ap_stack_loaded = 0;
 
 			write_to_local_apic(LAPIC_ERROR_REG, 0x00);
 
@@ -323,18 +325,20 @@ namespace Kernel
 					| ICR_LO_level_assert
 					| ICR_LO_trigger_mode_edge
 					| ICR_LO_destination_shorthand_none
-					| ap_init_page
+					| (ap_init_paddr / PAGE_SIZE)
 					, 200
 				);
 			}
 
 			// give processor upto 100 * 100 us + 200 us to boot
-			for (int i = 0; i < 100; i++)
+			PageTable::with_fast_page(ap_init_paddr, [&] {
-			{
+				for (int i = 0; i < 100; i++)
-				if (__atomic_load_n(&g_ap_stack_loaded[0], __ATOMIC_SEQ_CST))
+				{
-					break;
+					if (__atomic_load_n(&PageTable::fast_page_as_sized<uint8_t>(13), __ATOMIC_SEQ_CST))
-				udelay(100);
+						break;
-			}
+					udelay(100);
+				}
+			});
 		}
 
 		__atomic_store_n(&g_ap_startup_done[0], 1, __ATOMIC_SEQ_CST);