a working x86-64 loader

am/arch/x86_64-qemu.mk

@@ -14,7 +14,7 @@ image:
 
 run:
 	@( echo -n $(mainargs); ) | dd if=/dev/stdin of=$(BINARY) bs=512 count=1 seek=1 conv=notrunc status=none
-	@qemu-system-x86_64 -serial none -machine accel=kvm:tcg -smp "$(smp)" -drive format=raw,file=$(BINARY) -nographic
+	@qemu-system-x86_64 -serial stdio -machine accel=kvm:tcg -smp "$(smp)" -drive format=raw,file=$(BINARY)
 
 debug:
 	@( echo -n $(mainargs); ) | dd if=/dev/stdin of=$(BINARY) bs=512 count=1 seek=1 conv=notrunc status=none

am/include/x86.h

@@ -266,7 +266,7 @@ static inline void set_cr0(uint32_t cr0) {
 }
 
 
-/*
+#ifndef __x86_64__
 static inline void set_idt(GateDesc *idt, int size) {
   volatile static uint16_t data[3];
   data[0] = size - 1;
@@ -286,7 +286,7 @@ static inline void set_gdt(SegDesc *gdt, int size) {
 static inline void set_tr(int selector) {
   asm volatile ("ltr %0" : : "r"((uint16_t)selector));
 }
-*/
+#endif
 
 static inline uint32_t get_cr2() {
   volatile uint32_t val;

am/src/x86_64/qemu/start.S

@@ -1,51 +1,63 @@
-#include <x86.h>
-
-#define KERNEL_BASE 0x00100000
-
 .code32
 
-.globl _start
+#define PML4_ADDR_TO_ENTRY_INDEX(addr) (((addr) >> 39) & 0x1FF)
+#define PDPT_ADDR_TO_ENTRY_INDEX(addr) (((addr) >> 30) & 0x1FF)
+#define TWO_MEGABYTES_SHIFT 21
+#define TWO_MEGABYTES (1 << TWO_MEGABYTES_SHIFT)
+#define KERNEL_BASE 0x00100000
+
+.section .bss
+// bootstrap page table
+.comm pml4, 0x1000, 0x1000
+.comm low_pdpt, 0x1000, 0x1000
+.comm high_pdpt, 0x1000, 0x1000
+.comm low_page_directory_table, 0x1000, 0x1000
+.comm high_page_directory_table, 0x1000, 0x1000
+
+.section .text
+.global _start
+.type _start, @function
 _start:
-  movl $0x1000, %edi
-  movl %edi, %cr3
+  movl $low_pdpt + 3, %eax
+  movl %eax, pml4 + (PML4_ADDR_TO_ENTRY_INDEX(KERNEL_BASE) * 8)
+  movl $high_pdpt + 3, %eax
+  movl %eax, pml4 + (PML4_ADDR_TO_ENTRY_INDEX(KERNEL_BASE) * 8)
+  movl $low_page_directory_table + 3, %eax
+  movl %eax, low_pdpt + (PDPT_ADDR_TO_ENTRY_INDEX(KERNEL_BASE) * 8)
+  movl $high_page_directory_table + 3, %eax
+  movl %eax, high_pdpt + (PDPT_ADDR_TO_ENTRY_INDEX(KERNEL_BASE) * 8)
+  mov $0, %ecx
+  movl $_end, %esi
+  shrl $21, %esi
+  addl $1, %esi
+page_directory_table_loop:
+  movl $200000, %eax
+  mul %ecx
+  or $0x83, %eax
+  movl %eax, low_page_directory_table(, %ecx, 8)
+  movl %eax, high_page_directory_table(, %ecx, 8)
 
-  movl $0x2003, (%edi)
-  addl $0x1000, %edi
-  movl $0x3003, (%edi)
-  addl $0x1000, %edi
-  movl $0x4003, (%edi)
-  addl $0x1000, %edi
+  inc %ecx
+  cmp %esi, %ecx
+  jne page_directory_table_loop  // if not equal redo loop
 
-  movl $0x3, %ebx
-  movl $512, %ecx
-.fill_pt:
-  movl %ebx, (%edi)
-  addl $4096, %ebx
-  addl $8, %edi
-  loop .fill_pt
-  
 .align 128
-  movl %cr0, %eax
-  orl $(CR0_PE | CR0_PG), %eax
-  movl %eax, %cr4 // enable CR4 PE and PG
-
+  movl $pml4, %eax
+  movl %eax, %cr3     // %cr3 = PML4 base
+  movl $0x20, %eax
+  movl %eax, %cr4     // %cr4.PAE = 1
   movl $0xc0000080, %ecx
   rdmsr
-  orl $256, %eax
-  wrmsr // enable long mode
+  or $0x100, %eax
+  wrmsr // %msr.LME = 1.
+  movl $0x80000011, %eax
+  movl %eax, %cr0    // %cr0.PG = 1; %cr0.PE = 1;
+  lgdt gdt_ptr + KERNEL_BASE // set a gdt
+  ljmp $8, $_start64 + KERNEL_BASE
+  cli
+  hlt
 
-  movl %cr0, %eax
-  orl $(CR0_PE | CR0_PG), %eax
-  movl %eax, %cr0 // eanble CR0 PE and PG
-
-  lgdt gdtr64
-  ljmp $8, $entry64
-
-.align 128
-entry64:
-  jmp .
-
-gdtr64:
+gdt_ptr:
   .word gdt64_end - gdt64_begin - 1;
   .quad gdt64_begin + KERNEL_BASE
 
@@ -57,3 +69,22 @@ gdt64_begin:
   .long 0x00000000 # 2: Data, R/W, Expand Down
   .long 0x00009000
 gdt64_end:
+
+.code64
+
+.global _start64
+.type _start64, @function
+_start64:
+  movw $0, %ax
+  movw %ax, %ds
+  movw %ax, %es
+  movw %ax, %fs
+  movw %ax, %gs
+  movw %ax, %ss
+
+  movq $0x4000, %rsp
+  movq $0x7e00, %rdi
+  call _start_c
+
+  cli
+  hlt

am/src/x86_64/qemu/trm.c

@@ -5,12 +5,15 @@ _Area _heap = {}; // the heap memory defined in AM spec
 
 int main(const char *args);
 
-void _start64(char *args) {
+void _start_c(char *args) {
+  // TODO: build full page table
   int ret = main(args);
   _halt(ret);
 }
 
 void _putc(char ch) {
+  #define COM1 0x3f8
+  outb(COM1, ch);
 }
 
 void _halt(int code) {
@@ -19,4 +22,4 @@ void _halt(int code) {
 
 void some_code() {
   
-}
+}

tests/hello/hello.c

@@ -1,3 +1,20 @@
-int main() {
-  while (1);
+#include <am.h>
+#include <klib.h>
+
+void putd(int n) {
+  _putc("0123456789"[n]);
+}
+
+int main(const char *args) {
+  printf("args = \"");
+  while (*args) {
+    _putc(*args++);
+  }
+  printf("\"\n");
+
+  printf("sizeof(long) = ");
+  putd(sizeof(long));
+  printf("\n");
+  while (1);
+  return 0;
 }