preliminary x86_64 linux support

apps/hv/guest/vlbl/Makefile

@@ -0,0 +1,60 @@
+OUT ?= out
+
+entry-src := entry.S
+entry-obj := $(OUT)/entry.o
+loader-src := loader.c
+loader-obj := $(OUT)/loader.o
+virt-int-src := virt_int.S
+virt-int-obj := $(OUT)/virt_int.o
+virt-int-c-src := virt_int.c
+virt-int-c-obj := $(OUT)/virt_int_c.o
+
+ldscript := vlbl.lds
+target := $(OUT)/vlbl
+target-objs := $(entry-obj) $(loader-obj) $(virt-int-obj) $(virt-int-c-obj)
+target-elf := $(target).elf
+target-bin := $(target).bin
+target-disasm-16 := $(target).16.asm
+target-disasm-32 := $(target).32.asm
+
+CC ?= gcc
+AS ?= as
+LD ?= ld
+OBJCOPY ?= objcopy
+OBJDUMP ?= objdump
+
+all: $(OUT) $(target).bin disasm
+
+disasm: disasm16 disasm32
+
+disasm16:
+	$(OBJDUMP) -D -m i8086 -M intel $(target-elf) > $(target-disasm-16)
+
+disasm32:
+	$(OBJDUMP) -D -m i386 -M intel $(target-elf) > $(target-disasm-32)
+
+$(OUT):
+	mkdir -p $(OUT)
+
+$(entry-obj): $(entry-src)
+	$(AS) --32 -msyntax=intel -mnaked-reg $< -o $@
+
+$(loader-obj): $(loader-src)
+	$(CC) -m32 -fno-pie -fno-builtin -Os -c $< -o $@
+
+$(virt-int-obj): $(virt-int-src)
+	$(AS) --32 -msyntax=intel -mnaked-reg $< -o $@
+
+$(virt-int-c-obj): $(virt-int-c-src)
+	$(CC) -m16 -march=i386 -fno-pie -fno-builtin -Os -c $< -o $@
+
+$(target-elf): $(target-objs) $(ldscript)
+	$(LD) -T$(ldscript) $(target-objs) -o $@
+
+$(target-bin): $(target-elf)
+	$(OBJCOPY) $< --strip-all -O binary $@
+
+clean:
+	rm -rf $(OUT)
+
+.PHONY: all disasm clean

apps/hv/guest/vlbl/defs.h

@@ -0,0 +1,71 @@
+/* as we do not have 64-bit codes here, it's safe to assume that */
+typedef unsigned char uint8_t, *uint8_p;
+typedef unsigned short int uint16_t, *uint16_p;
+typedef unsigned long int uint32_t, *uint32_p;
+typedef signed char int8_t, *int8_p;
+typedef signed short int int16_t, *int16_p;
+typedef signed long int int32_t, *int32_p;
+
+#ifndef bool
+typedef uint8_t bool;
+# define true (1)
+# define false (0)
+#endif
+
+static inline void outb(uint8_t value, uint16_t port) {
+	asm volatile("outb %b0, %w1" : : "a"(value), "Nd"(port));
+}
+
+#define COM1 0x3f8
+static inline void putchar(char c) {
+    outb((uint8_t)c, COM1);
+}
+
+static inline void putsi(const char *s) {
+    while (*s != '\0') {
+        putchar(*(s++));
+    }
+}
+
+static inline void puts(const char *s) {
+    putsi(s);
+    putchar('\n');
+}
+
+static inline void putud(uint32_t num) {
+    static char buf[12];
+    int8_t cnt = 0;
+    while (num > 0) {
+        buf[cnt++] = '0' + num % 10;
+        num /= 10;
+    }
+
+    while (--cnt >= 0) {
+        putchar(buf[cnt]);
+    }
+}
+
+static inline void putux(uint32_t num, bool prefix, int8_t padding) {
+    static char buf[10];
+    const char *chars = "0123456789abcdef";
+    int8_t cnt = 0;
+    while (num > 0) {
+        buf[cnt++] = chars[num % 16];
+        num /= 16;
+    }
+
+    if (padding > cnt && padding <= 10) {
+        while (cnt < padding) {
+            buf[cnt++] = '0';
+        }
+    }
+
+    if (prefix) {
+        putchar('0');
+        putchar('x');
+    }
+
+    while (--cnt >= 0) {
+        putchar(buf[cnt]);
+    }
+}

apps/hv/guest/vlbl/entry.S

@@ -0,0 +1,128 @@
+# it's 7c00 here
+.section .text
+.code16
+.global entry
+entry:
+    cli
+    cld
+
+    # 初始化段寄存器
+    xor     ax, ax
+    mov     ds, ax
+    mov     es, ax
+    mov     ss, ax
+
+    # 进入保护模式
+    lgdt    [prot_gdt_desc]
+    mov     eax, cr0
+    or      eax, 0x1
+    mov     cr0, eax
+
+    ljmp    0x8, entry32
+
+.code32
+.global entry32
+entry32:
+    # 初始化段选择子到数据段
+    mov     ax, 0x10
+    mov     ds, ax
+    mov     es, ax
+    mov     ss, ax
+    mov     fs, ax
+    mov     gs, ax
+
+    # 初始化临时栈
+    mov     ebp, 0x7b00
+    mov     esp, 0x7b00
+
+    # 加载内核
+    push    0x00100000 # 32-bit 加载地址
+    push    0x0001e000 # 栈顶地址
+    push    0x00010000 # 16-bit 加载地址
+    push    0x70200000 # 镜像地址
+    call    load_kernel
+
+    # 跳转到16-bit保护模式
+    lgdt    [prot_16_gdt_desc]
+    ljmp    0x8, entry_prot_16
+
+
+.code16
+.global entry_prot_16
+entry_prot_16:
+    # 再次初始化段选择子到数据段
+    mov     ax, 0x10
+    mov     ds, ax
+    mov     es, ax
+    mov     ss, ax
+    mov     fs, ax
+    mov     gs, ax
+
+    # 加载中断向量表
+    lidt    [real_idt_dest]
+
+    # 关闭保护模式
+    mov     eax, cr0
+    or      eax, 0x1
+    xor     eax, 0x1
+    mov     cr0, eax
+
+    # 跳转到实模式
+    ljmp    0x0, entry_last_jump
+
+.code16
+.global entry_last_jump
+entry_last_jump:
+    # 初始化段寄存器和栈，准备启动
+    mov     ax, 0x1000
+    mov     ds, ax
+    mov     es, ax
+    mov     fs, ax
+    mov     gs, ax
+    mov     ss, ax
+    mov     sp, 0xe000
+
+    # 清理寄存器
+    xor     eax, eax
+    xor     ebx, ebx
+    xor     ecx, ecx
+    xor     edx, edx
+    xor     ebp, ebp
+    xor     esi, esi
+    xor     edi, edi
+
+    # 跳转到linux内核入口
+    ljmp    0x1020, 0
+
+    # 以下代码不应该被执行，使用一条无效的in指令返回
+    in      ax, 0x11
+    jmp $
+
+.global end_of_code
+end_of_code:
+
+# 32位GDT
+.balign 16
+.global prot_gdt
+prot_gdt:
+    .quad 0x0000000000000000    # 0x00: null
+    .quad 0x00cf9b000000ffff    # 0x08: code segment (base=0, limit=0xfffff, type=32bit code exec/read, DPL=0, 4k)
+    .quad 0x00cf93000000ffff    # 0x10: data segment (base=0, limit=0xfffff, type=32bit data read/write, DPL=0, 4k)
+
+.global prot_gdt_desc
+prot_gdt_desc:
+    .short  prot_gdt_desc - prot_gdt - 1    # limit
+    .long   prot_gdt                        # base
+
+# 16位GDT
+.balign 16
+.global prot_16_gdt
+prot_16_gdt:
+    .quad 0x0000000000000000    # 0x00: null
+    .quad 0x000f9b000000ffff    # 0x08: code segment (base=0, limit=0xfffff, type=16bit code exec/read, DPL=0, byte)
+    .quad 0x000f93000000ffff    # 0x10: data segment (base=0, limit=0xfffff, type=16bit data read/write, DPL=0, byte)
+
+.global prot_16_gdt_desc
+prot_16_gdt_desc:
+    .short  prot_16_gdt_desc - prot_16_gdt - 1    # limit
+    .long   prot_16_gdt                        # base

apps/hv/guest/vlbl/loader.c

@@ -0,0 +1,92 @@
+#include "defs.h"
+
+#pragma pack(push, 1)
+typedef struct _kernel_header {
+    uint8_t _01f0;
+    uint8_t setup_sects;
+    uint16_t root_flags;
+    uint32_t syssize;
+    uint16_t ramsize;
+    uint16_t vid_mode;
+    uint16_t root_dev;
+    uint16_t boot_flag;
+    uint16_t jump;
+    uint32_t header;
+    uint16_t version;
+    uint32_t realmode_swtch;
+    uint16_t start_sys_seg;
+    uint16_t kernel_version;
+    uint8_t type_of_loader;
+    uint8_t loadflags;
+    uint16_t setup_move_size;
+    uint32_t code32_start;
+    uint32_t ramdisk_image;
+    uint32_t ramdisk_size;
+    uint32_t bootsect_kludge;
+    uint16_t heap_end_ptr;
+    uint8_t ext_loader_ver;
+    uint8_t ext_loader_type;
+    uint32_t cmd_line_ptr;
+    uint32_t initrd_addr_max;
+    uint32_t kernel_alignment;
+    uint8_t relocatable_kernel;
+    uint8_t min_alignment;
+    uint16_t xloadflags;
+    uint32_t cmdline_size;
+    uint32_t hardware_subarch;
+    uint32_t hardware_subarch_data_l;
+    uint32_t hardware_subarch_data_h;
+    uint32_t payload_offset;
+    uint32_t payload_length;
+    uint32_t setup_data_l;
+    uint32_t setup_data_h;
+    uint32_t pref_address_l;
+    uint32_t pref_address_h;
+    uint32_t init_size;
+    uint32_t handover_offset;
+    uint32_t kernel_info_offset;
+} kernel_header, *kernel_header_ptr;
+#pragma pack(pop)
+
+void cpy4(void *dst, const void *src, uint32_t size) {
+    const uint32_t * ptr_src = src;
+    uint32_p ptr_dst = dst;
+    for (int i = 0; i < size; i += 4) {
+        *ptr_dst = *ptr_src;
+        ptr_src++;
+        ptr_dst++;
+    }
+}
+
+const char cmd[256] = "console=uart8250,io,0x3f8,115200n8 debug root=/dev/vda\0";
+
+int load_kernel(void *kernel_image, void *loc_real, void *stack_end, void *loc_prot) {
+    kernel_header_ptr orig_header = kernel_image + 0x1f0;
+
+    uint32_t kernel_lower_size = ((orig_header->setup_sects ? orig_header->setup_sects : 4) + 1) * 512;
+    uint32_t kernel_upper_size = orig_header->syssize * 16;
+    void *prot = kernel_image + kernel_lower_size;
+
+    cpy4(loc_real, kernel_image, kernel_lower_size);
+    cpy4(loc_prot, prot, kernel_upper_size);
+
+    void *cmd_base = stack_end;
+    void *setup_data_base = cmd_base + 256;
+
+    cpy4(cmd_base, cmd, 256);
+
+    kernel_header_ptr header = loc_real + 0x1f0;
+
+    header->vid_mode = 0xffff;
+    header->type_of_loader = 0xff;
+    header->loadflags = (header->loadflags & 0x1f) | 0x80;
+    header->code32_start = (uint32_t)(loc_prot);
+    header->ramdisk_image = 0;
+    header->ramdisk_size = 0;
+    header->heap_end_ptr = (uint16_t)(stack_end - loc_real - 0x200);
+    header->cmd_line_ptr = (uint32_t)cmd_base;
+    header->setup_data_l = 0;
+    header->setup_data_h = 0;
+
+    return 0;
+}