amd64: barely boot into kernel code

6 files changed, 490 insertions, 0 deletions

diff --git a/src/kernel/arch/amd64/32/32 b/src/kernel/arch/amd64/32/32
new file mode 100644
index 0000000..9bd97e6
--- /dev/null
+++ b/src/kernel/arch/amd64/32/32
@@ -0,0 +1,130 @@
+#include <kernel/arch/generic.h>
+#include <kernel/arch/amd64/gdt.h>
+#include <shared/mem.h>
+#include <stdbool.h>
+#include <stdint.h>
+
+extern char _isr_mini_stack;
+
+struct gdt_entry {
+	uint64_t limit_low  : 16;
+	uint64_t base_low   : 24;
+	uint64_t accessed   :  1; // set by the processor
+	                          // CODE        | DATA
+	uint64_t rw         :  1; // readable?   | writeable?
+	uint64_t conforming :  1; // conforming? | expands down?
+	uint64_t code       :  1; // 1           | 0
+
+	uint64_t codeordata :  1; // 1 for everything other than TSS and LDT
+	uint64_t ring       :  2;
+	uint64_t present    :  1; // always 1
+	uint64_t limit_high :  4;
+	uint64_t available  :  1; // ???
+	uint64_t long_mode  :  1;
+	uint64_t x32        :  1;
+	uint64_t gran       :  1; // 1 - 4kb, 0 - 1b
+	uint64_t base_high  :  8;
+} __attribute__((packed));
+
+struct tss_entry {
+	uint32_t _unused0;
+	uint32_t esp0; // kernel mode stack pointer
+	uint32_t ss0;  // kernel mode stack segment
+	uint8_t  _unused1[0x5c];
+} __attribute__((packed));
+
+struct lgdt_arg {
+	uint16_t limit;
+	uint32_t base;
+} __attribute__((packed));
+
+__attribute__((section(".shared")))
+static struct gdt_entry GDT[SEG_end];
+__attribute__((section(".shared")))
+static struct tss_entry TSS;
+static struct lgdt_arg lgdt_arg; // probably doesn't need to be global
+
+static void gdt_fillout(struct gdt_entry* entry, uint8_t ring, bool code);
+static void gdt_prepare(void);
+static void gdt_load(void);
+
+
+static void gdt_fillout(struct gdt_entry* entry, uint8_t ring, bool code) {
+	*entry = (struct gdt_entry) {
+		// set up the identity mapping
+		.limit_low  = 0xFFFF,
+		.limit_high = 0xF,
+		.gran       = 1, // 4KB * 0xFFFFF = (almost) 4GB
+		.base_low   = 0,
+		.base_high  = 0,
+
+		.ring       = ring,
+		.code       = code,
+
+		.accessed   = 0,
+		.rw         = 1,
+		.conforming = 0,
+		.codeordata = 1,
+		.present    = 1,
+		.long_mode  = 0, // ???
+		.available  = 1, // ???
+		.x32        = 1,
+	};
+}
+
+static void gdt_prepare(void) {
+	GDT[SEG_null].present = 0;
+
+	gdt_fillout(&GDT[SEG_r0code], 0, true);
+	gdt_fillout(&GDT[SEG_r0data], 0, false);
+	gdt_fillout(&GDT[SEG_r3code], 3, true);
+	gdt_fillout(&GDT[SEG_r3data], 3, false);
+
+	// tss
+	memset(&TSS, 0, sizeof(TSS));
+	TSS.ss0 = SEG_r0data << 3; // kernel data segment
+	TSS.esp0 = (uintptr_t) &_isr_mini_stack;
+
+	GDT[SEG_TSS] = (struct gdt_entry) {
+		.limit_low  = sizeof(TSS),
+		.limit_high = sizeof(TSS) >> 16,
+		.gran       = 0,
+		.base_low   =  (uintptr_t) &TSS,
+		.base_high  = ((uintptr_t) &TSS) >> 24,
+
+		.accessed   = 1, // 1 for TSS
+		.rw         = 0, // 1 busy / 0 not busy
+		.conforming = 0, // 0 for TSS
+		.code       = 1, // 32bit
+		.codeordata = 0, // is a system entry
+		.ring       = 3,
+		.present    = 1,
+		.available  = 0, // 0 for TSS
+		.long_mode  = 0,
+		.x32        = 0, // idk
+	};
+}
+
+static void gdt_load(void) {
+	lgdt_arg.limit = sizeof(GDT) - 1;
+	lgdt_arg.base = (uintptr_t) &GDT;
+	asm("lgdt (%0)" 
+	    : : "r" (&lgdt_arg) : "memory");
+
+	asm("ltr %%ax"
+	    : : "a" (SEG_TSS << 3 | 3) : "memory");
+
+	// update all segment registers
+	gdt_farjump(SEG_r0code << 3);
+	asm("mov %0, %%ds;"
+	    "mov %0, %%ss;"
+	    "mov %0, %%es;"
+	    "mov %0, %%fs;"
+	    "mov %0, %%gs;"
+	    : : "r" (SEG_r0data << 3) : "memory");
+}
+
+void gdt_init(void) {
+	gdt_prepare();
+	gdt_load();
+}
diff --git a/src/kernel/arch/amd64/32/boot.s b/src/kernel/arch/amd64/32/boot.s
new file mode 100644
index 0000000..eb33c28
--- /dev/null
+++ b/src/kernel/arch/amd64/32/boot.s
@@ -0,0 +1,86 @@
+.section .text
+.global _start
+.type _start, @function
+_start:
+	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), %eax // PAE
+	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 // sysenter | long mode | NX
+	wrmsr
+
+	mov %cr0, %eax
+	or $0x80000000, %eax
+	mov %eax, %cr0
+
+	call gdt_init
+	lgdt (lgdt_arg)
+
+	pop %edi
+
+	// TODO import gdt.h for globals
+	mov $(2<<3), %eax
+	mov %eax, %ds // SEG_r0data
+	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
+
+// TODO not part of anything yet
+	call sysenter_setup
+	// TODO will fail
+	push %ebx // address of the Multiboot struct
+	call kmain_early
+
+.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
diff --git a/src/kernel/arch/amd64/32/farjump.s b/src/kernel/arch/amd64/32/farjump.s
new file mode 100644
index 0000000..2885d2b
--- /dev/null
+++ b/src/kernel/arch/amd64/32/farjump.s
@@ -0,0 +1,8 @@
+.section .text
+.global gdt_farjump
+.type gdt_farjump, @function
+gdt_farjump:
+	/* retf pops off the return address and code segment off the stack.
+	 * it turns out that in the i386 cdecl calling convention they're in
+	 * the correct place already. */
+	retf
diff --git a/src/kernel/arch/amd64/32/gdt.c b/src/kernel/arch/amd64/32/gdt.c
new file mode 100644
index 0000000..2229330
--- /dev/null
+++ b/src/kernel/arch/amd64/32/gdt.c
@@ -0,0 +1,118 @@
+#include <kernel/arch/amd64/32/gdt.h>
+#include <kernel/arch/generic.h>
+#include <stdbool.h>
+#include <stdint.h>
+
+extern char _isr_mini_stack;
+
+struct gdt_entry {
+	uint64_t limit_low  : 16;
+	uint64_t base_low   : 24;
+	uint64_t accessed   :  1; // set by the processor
+	                          // CODE        | DATA
+	uint64_t rw         :  1; // readable?   | writeable?
+	uint64_t conforming :  1; // conforming? | expands down?
+	uint64_t code       :  1; // 1           | 0
+
+	uint64_t codeordata :  1; // 1 for everything other than TSS and LDT
+	uint64_t ring       :  2;
+	uint64_t present    :  1; // always 1
+	uint64_t limit_high :  4;
+	uint64_t available  :  1; // ???
+	uint64_t long_mode  :  1;
+	uint64_t x32        :  1;
+	uint64_t gran       :  1; // 1 - 4kb, 0 - 1b
+	uint64_t base_high  :  8;
+} __attribute__((packed));
+
+struct tss_entry {
+	uint32_t reserved0;
+	uint64_t rsp[3];
+	uint64_t ist[8];
+	uint64_t reserved2;
+	uint16_t reserved3;
+	uint16_t iopb;
+} __attribute__((packed));
+
+struct lgdt_arg {
+	uint16_t limit;
+	uint32_t base;
+} __attribute__((packed));
+
+__attribute__((section(".shared")))
+struct gdt_entry GDT[SEG_end];
+__attribute__((section(".shared")))
+struct tss_entry TSS;
+
+struct lgdt_arg lgdt_arg;
+
+
+static void *memset32(void *s, int c, size_t n) {
+	uint8_t *s2 = s;
+	for (size_t i = 0; i < n; i++)
+		s2[i] = c;
+	return s;
+}
+
+
+static void gdt_fillout(struct gdt_entry* entry, uint8_t ring, bool code) {
+	*entry = (struct gdt_entry) {
+		// set up the identity mapping
+		.limit_low  = 0xFFFF,
+		.limit_high = 0xF,
+		.gran       = 1,
+		.base_low   = 0,
+		.base_high  = 0,
+
+		.ring       = ring,
+		.code       = code,
+
+		.accessed   = 0,
+		.rw         = 1,
+		.conforming = 0,
+		.codeordata = 1,
+		.present    = 1,
+		.long_mode  = 1,
+		.available  = 1,
+		.x32        = 0,
+	};
+}
+
+void gdt_init(void) {
+	GDT[SEG_null].present = 0;
+
+	gdt_fillout(&GDT[SEG_r0code], 0, true);
+	gdt_fillout(&GDT[SEG_r0data], 0, false);
+	gdt_fillout(&GDT[SEG_r3code], 3, true);
+	gdt_fillout(&GDT[SEG_r3data], 3, false);
+
+	lgdt_arg.limit = sizeof(GDT) - 1;
+	lgdt_arg.base = (uint64_t) &GDT;
+
+
+	memset32(&TSS, 0, sizeof(TSS));
+	for (int i = 0; i < 3; i++)
+		TSS.rsp[i] = (uint64_t) &_isr_mini_stack;
+	TSS.ist[1] = (uint64_t) &_isr_mini_stack;
+
+	uint64_t tss_addr = (uint64_t) &TSS;
+	GDT[SEG_TSS] = (struct gdt_entry) {
+		.limit_low  = sizeof(TSS),
+		.limit_high = sizeof(TSS) >> 16,
+		.gran       = 0,
+		.base_low   = tss_addr,
+		.base_high  = tss_addr >> 24,
+
+		.accessed   = 1,
+		.rw         = 0,
+		.conforming = 0,
+		.code       = 1,
+		.codeordata = 0,
+		.ring       = 0, // was 3 pre-port
+		.present    = 1,
+		.available  = 1,
+		.long_mode  = 0,
+		.x32        = 0,
+	};
+	*((uint64_t*)&GDT[SEG_TSS2]) = (tss_addr >> 32);
+}
diff --git a/src/kernel/arch/amd64/32/gdt.h b/src/kernel/arch/amd64/32/gdt.h
new file mode 100644
index 0000000..fbaf681
--- /dev/null
+++ b/src/kernel/arch/amd64/32/gdt.h
@@ -0,0 +1,17 @@
+#pragma once
+
+enum {
+	SEG_null,
+	// order dictated by SYSENTER
+	SEG_r0code,
+	SEG_r0data,
+	SEG_r3code,
+	SEG_r3data,
+	SEG_TSS,
+	SEG_TSS2,
+
+	SEG_end
+};
+
+void gdt_init(void);
+void gdt_farjump(int segment);
diff --git a/src/kernel/arch/amd64/32/paging.c b/src/kernel/arch/amd64/32/paging.c
new file mode 100644
index 0000000..975dd98
--- /dev/null
+++ b/src/kernel/arch/amd64/32/paging.c
@@ -0,0 +1,131 @@
+#include <stddef.h>
+#include <stdint.h>
+
+struct pml4e {
+	uint64_t present   : 1;
+	uint64_t writeable : 1;
+	uint64_t user      : 1;
+	uint64_t writethru : 1;
+
+	uint64_t uncached  : 1;
+	uint64_t accessed  : 1;
+	uint64_t _unused1  : 1;
+	uint64_t reserved  : 1; // always 0
+
+	uint64_t _unused2  : 3;
+	uint64_t _unused3  : 1; // HLAT thing
+
+	uint64_t address   : 40;
+
+	uint64_t _unused4  : 11;
+	uint64_t noexec    : 1;
+} __attribute__((packed));
+
+struct pdpte { // page directory pointer table entry, 1gb page | 512 * pde
+	uint64_t present   : 1;
+	uint64_t writeable : 1;
+	uint64_t user      : 1;
+	uint64_t writethru : 1;
+
+	uint64_t uncached  : 1;
+	uint64_t accessed  : 1;
+	uint64_t _unused1  : 1;
+	uint64_t large     : 1; // 1gb page
+
+	uint64_t _unused2  : 3;
+	uint64_t _unused3  : 1; // HLAT
+
+	uint64_t address   : 40;
+
+	uint64_t _unused4  : 11;
+	uint64_t noexec    : 1;
+} __attribute__((packed));
+
+struct pde { // page directory entry, 2mb page | 512 * pte
+	uint64_t present   : 1;
+	uint64_t writeable : 1;
+	uint64_t user      : 1;
+	uint64_t writethru : 1;
+
+	uint64_t uncached  : 1;
+	uint64_t accessed  : 1;
+	uint64_t dirty     : 1; // only if large
+	uint64_t large     : 1; // 2mb
+
+	uint64_t global    : 1; // only if large ; TODO enable CR4.PGE
+	uint64_t _unused2  : 2;
+	uint64_t _unused3  : 1; // HLAT
+
+	uint64_t address   : 40; // highest bit - PAT
+
+	uint64_t _unused4  : 7;
+	uint64_t pke       : 4;
+	uint64_t noexec    : 1;
+} __attribute__((packed));
+
+struct pte { // page table entry, 4kb page
+	uint64_t present   : 1;
+	uint64_t writeable : 1;
+	uint64_t user      : 1;
+	uint64_t writethru : 1;
+
+	uint64_t uncached  : 1;
+	uint64_t accessed  : 1;
+	uint64_t dirty     : 1;
+	uint64_t pat       : 1;
+
+	uint64_t global    : 1; // TODO enable CR4.PGE
+	uint64_t _unused2  : 2;
+	uint64_t _unused3  : 1; // HLAT
+
+	uint64_t address   : 40;
+
+	uint64_t _unused4  : 7;
+	uint64_t pke       : 4;
+	uint64_t noexec    : 1;
+} __attribute__((packed));
+
+__attribute__((aligned(4096)))
+struct pml4e pml4_identity[512];
+
+__attribute__((aligned(4096)))
+struct pdpte pdpte_low[512]; // 0-512gb
+
+__attribute__((aligned(4096)))
+struct pde pde_low[512]; // 0-1gb
+
+
+static void *memset32(void *s, int c, size_t n) {
+	uint8_t *s2 = s;
+	for (size_t i = 0; i < n; i++)
+		s2[i] = c;
+	return s;
+}
+
+
+void pml4_identity_init(void) {
+	memset32(pml4_identity, 0, sizeof pml4_identity);
+	memset32(pdpte_low, 0, sizeof pdpte_low);
+	memset32(pde_low, 0, sizeof pde_low);
+
+	pml4_identity[0] = (struct pml4e) {
+		.present = 1,
+		.writeable = 1,
+		.address = ((uintptr_t)pdpte_low) >> 12,
+	};
+
+	pdpte_low[0] = (struct pdpte) {
+		.present = 1,
+		.writeable = 1,
+		.address = ((uintptr_t)pde_low) >> 12,
+	};
+
+	for (int i = 0; i < 512; i++) {
+		pde_low[i] = (struct pde) {
+			.present = 1,
+			.writeable = 1,
+			.large = 1,
+			.address = (i * 2 * 1024 * 1024) >> 12,
+		};
+	}
+}