kernel: set up the GDT in assembly

This is just for simplicity's sake.

I think I could even omit the `movw $TSS, (GdtTss + 2)` and have the linker
fill that out as a relocation, but that would probably be more complex overall.

1 files changed, 114 insertions, 0 deletions

diff --git a/src/kernel/arch/amd64/32/boot.S b/src/kernel/arch/amd64/32/boot.S
new file mode 100644
index 0000000..3f5798f
--- /dev/null
+++ b/src/kernel/arch/amd64/32/boot.S
@@ -0,0 +1,114 @@
+.section .text
+.global _start
+.type _start, @function
+_start:
+	cli
+	mov $_stack_top, %esp
+	push %ebx // save the address of the multiboot struct
+
+	mov $0x80000000, %eax // check CPUID extended functions
+	cpuid
+	cmp $0x80000001, %eax
+	jb panic_early
+
+	mov $0x80000001, %eax
+	cpuid
+	test $(1<<29), %edx // check long mode support
+	jz panic_early
+
+	mov %cr4, %eax
+	or $(1<<5 | 1<<9 | 1<<10), %eax // PAE | SSE | SSE
+	mov %eax, %cr4
+
+	call pml4_identity_init
+	mov $pml4_identity, %eax
+	mov %eax, %cr3
+
+	mov $0xC0000080, %ecx // EFER MSR
+	rdmsr
+	or $(1 | 1<<8 | 1<<11), %eax // syscall/ret | long mode | NX
+	wrmsr
+
+	mov %cr0, %eax
+	or $0x80000002, %eax // enable paging, coprocessor monitoring
+	and $(~4), %eax // disable coprocessor emulation
+	mov %eax, %cr0
+
+	/* fill out the TSS's address */
+	movw $TSS, (GdtTss + 2)
+	lgdt (GdtPointer)
+
+	pop %edi
+
+	mov $(2 << 3), %eax // SEG_r0data
+	mov %eax, %ds
+	mov %eax, %ss
+	mov %eax, %es
+	mov %eax, %fs
+	mov %eax, %gs
+
+	ljmp $(1 << 3), $boot64 // SEG_r0code
+
+panic_early:
+	// output a vga Fuck
+	movl $0x4F754F46, 0xB872A
+	movl $0x4F6B4F63, 0xB872E
+	jmp cpu_halt
+
+.global cpu_shutdown
+.type cpu_shutdown, @function
+cpu_shutdown:
+/* This quits QEMU. While I couldn't find this officially documented anywhere,
+ * it is used by QEMU in tests/tcg/i386/system/boot.S (as of commit 40d6ee), so
+ * I assume that this is safe-ish to use */
+	mov $0x604, %edx
+	mov $0x2000, %eax
+	outw %ax, %dx
+
+.global cpu_halt
+.type cpu_halt, @function
+cpu_halt:
+	cli
+1:	hlt
+	jmp 1b
+
+
+.global cpu_pause
+.type cpu_pause, @function
+cpu_pause:
+	sti
+	hlt
+	cli
+	ret
+
+.section .shared
+.global GDT
+.align 8
+GDT:
+#define GdtLimit(p)   (p & 0xFFFF) | ((p >> 16) << 48) /* doesn't check if p fits in 20 bits */
+#define GdtAccessed   (1<<40)
+#define GdtReadWrite  (1<<41)
+#define GdtCode       (1<<43)
+#define GdtCodeOrData (1<<44)
+#define GdtRing(r)    (r<<45)
+#define GdtPresent    (1<<47)
+#define GdtAvailable  (1<<52)
+#define GdtLongMode   (1<<53)
+#define GdtPageGran   (1<<55) /* limit is in pages */
+#define GdtCommon     (GdtLimit(0xFFFFF) | GdtPageGran | GdtAccessed | \
+                       GdtReadWrite | GdtCodeOrData | GdtPresent | GdtAvailable)
+.8byte 0                                              /* [0] = SEG_null */
+.8byte GdtCommon | GdtRing(0) | GdtLongMode | GdtCode /* [1] = SEG_r0code */
+.8byte GdtCommon | GdtRing(0) | GdtLongMode           /* [2] = SEG_r0data */
+.8byte GdtCommon | GdtRing(3)               | GdtCode /* [3] = SEG_r3code32 */
+.8byte GdtCommon | GdtRing(3) | GdtLongMode           /* [4] = SEG_r3data */
+.8byte GdtCommon | GdtRing(3) | GdtLongMode | GdtCode /* [5] = SEG_r3code */
+GdtTss:
+.8byte GdtLimit(104) | (9<<40) | GdtPresent | GdtAvailable /* [6] = SEG_TSS */
+.8byte 0                                                   /* [7] = SEG_TSS2 */
+
+.section .data
+.global GdtPointer
+GdtPointer:
+.2byte 63  /* size of the GDT - 1 */
+.4byte GDT /* address of the GDT */