/*
 * Copyright (C) 2018-2022 Intel Corporation.
 *
 * SPDX-License-Identifier: BSD-3-Clause
 *
 * This is entry for AP startup and BSP S3 wakeup
 *
 * When system jump to trampoline_start16, the CPU is in x86 real
 * mode with no stack setup. CS:IP points to trampoline_start16.
 *
 * The CPU will be changed to long mode finally with temperay
 * page table and gdt in this file. Then jump to different C main
 * entry according to whether it's AP startup or BSP S3 resume.
 * The real page table and gdt will be setup in C main entry.
 */

/* NOTE:
 *
 * MISRA C requires that all unsigned constants should have the suffix 'U'
 * (e.g. 0xffU), but the assembler may not accept such C-style constants. For
 * example, binutils 2.26 fails to compile assembly in that case. To work this
 * around, all unsigned constants must be explicitly spells out in assembly
 * with a comment tracking the original expression from which the magic
 * number is calculated. As an example:
 *
 *    /* 0x00000668 =
 *     *    (CR4_DE | CR4_PAE | CR4_MCE | CR4_OSFXSR | CR4_OSXMMEXCPT) *\/
 *    movl    $0x00000668, %eax
 *
 * Make sure that these numbers are updated accordingly if the definition of
 * the macros involved are changed.
 */

    .extern     init_secondary_pcpu

    .section     .trampoline_reset,"ax"

   .align   4
   .code16
   .global      trampoline_start16
   .org 0
trampoline_start16:

    /* Disable local interrupts */
    cli

    /*
     * There are two paths system could come here:
     *   - AP startup
     *     Silicon will set AP to real mode and setup CS:IP before
     *     jmp to trampoline_start16. And the IP is always 0 for sure.
     *   - BSP wakeup from S3
     *     Some bootloader (like ABL) doesn't guarante IP is set to
     *     zero before jump to trampoline_start16 after system resume
     *     from S3.
     *
     * To make trampoline code could work with all these cases, a far
     * jump is issued here as fixup. It will update the CS:IP according
     * to where the trampoline code is located.
     *
     * Here, we issue a far jump with "JMP ptr16:16" format (please refer
     * sdm vol2A - JMP instruction description). The jump target is set
     * to trampoline_fixup_target_addr. From trampoline_fixup_target_addr,
     * The CS has same value for both AP startup and BSP wakeup from S3.
     *
     * Because the limitation of real mode (can't access ip register
     * directly. So can't setup the trampoline_fixup_ip and
     * trampoline_fixup_cs), we have to update the trampoline_fixup_ip
     * and trampoline_fixup_cs when we preparing the trampoline code.
     *
     * Refer to preparing_trampoline() for fixup CS:IP setup
     */
    .byte 0xea	/* Opcode of "JMP ptr16:16" */
    .global trampoline_fixup_ip
trampoline_fixup_ip:
    .word   0	/* "EIP is intruction following JUMP instruction" */
    .global trampoline_fixup_cs
trampoline_fixup_cs:
    .word   0	/* CS */

    .global trampoline_fixup_target
trampoline_fixup_target:
    mov %cs, %ax
    mov %ax, %ds

    /* Set DE, PAE, MCE and OS support bits in CR4 */

    /* 0x00000668 =
     *    (CR4_DE | CR4_PAE | CR4_MCE | CR4_OSFXSR | CR4_OSXMMEXCPT) */
    movl    $0x00000668, %eax
    mov     %eax, %cr4

    /* Set CR3 to PML4 table address */

    movl    $cpu_boot_page_tables_ptr, %ebx
    mov     (%ebx), %eax
    mov     %eax, %cr3

    /* Set LME bit in EFER */

    /* 0xc0000080 = MSR_IA32_EFER */
    movl    $0xc0000080, %ecx
    rdmsr
    /* 0x00000100 = MSR_IA32_EFER_LME_BIT */
    orl     $0x00000100, %eax
    wrmsr

    /* 0xc0000080 = MSR_IA32_EFER */
    movl    $0xc0000080, %ecx
    rdmsr
    /* 0x00000800 = MSR_IA32_EFER_NXE_BIT */
    orl     $0x00000800, %eax
    wrmsr

    /* Enable paging, protection, numeric error and co-processor
       monitoring in CR0 to enter long mode */

    mov     %cr0, %ebx
    /* 0x80000023 = (CR0_PG | CR0_PE | CR0_MP | CR0_NE) */
    orl     $0x80000023, %ebx
    mov     %ebx, %cr0

    /* Load temportary GDT pointer value */
    lgdt    (trampoline_gdt_ptr - trampoline_start16)

    /* Perform a long jump based to start executing in 64-bit mode */

    movl    $trampoline_start64_fixup, %ebx
    ljmpl   *(%ebx)

    .align 8
    .global trampoline_start64_fixup
trampoline_start64_fixup:
    .long   trampoline_start64
    /* 0x0008 = HOST_GDT_RING0_CODE_SEL */
    .word   0x0008

    .code64
trampoline_start64:

    /* Set up all other data segment registers */
    /* 0x0010 = HOST_GDT_RING0_DATA_SEL */
    movl    $0x0010, %eax
    mov     %eax, %ss
    mov     %eax, %ds
    mov     %eax, %es
    mov     %eax, %fs
    mov     %eax, %gs

    movq    secondary_cpu_stack(%rip), %rsp

    /* Jump to C entry */
    movq    main_entry(%rip), %rax
    jmp     *%rax


/* main entry */
    .align  8
    .global main_entry
main_entry:
    .quad   init_secondary_pcpu /* default entry is AP start entry */

    .global secondary_cpu_stack
secondary_cpu_stack:
    .quad   0

/* GDT table */
    .align  4
trampoline_gdt:
    .quad   0x0000000000000000
    .quad   0x00af9b000000ffff
    .quad   0x00cf93000000ffff
trampoline_gdt_end:

/* GDT pointer */
    .align  2
    .global trampoline_gdt_ptr
trampoline_gdt_ptr:
    .short  (trampoline_gdt_end - trampoline_gdt) - 1
    .quad   trampoline_gdt

/* PML4, PDPT, and PD tables initialized to map first 4 GBytes of memory */
    .align  4
    .global cpu_boot_page_tables_ptr
cpu_boot_page_tables_ptr:
    .long cpu_boot_page_tables_start

    /*0x1000 = PAGE_SIZE*/
    .align  0x1000
    .global cpu_boot_page_tables_start
cpu_boot_page_tables_start:
    /* 0x3 = (PAGE_PRESENT | PAGE_RW) */
    .quad   trampoline_pdpt_addr + 0x3
    /*0x1000 = PAGE_SIZE*/
    .align  0x1000
    .global trampoline_pdpt_addr
trampoline_pdpt_addr:
    address = 0
    .rept   4
    /* 0x3 = (PAGE_PRESENT | PAGE_RW) */
    .quad   trampoline_pdt_addr + address + 0x3
    /*0x1000 = PAGE_SIZE*/
    address = address + 0x1000
    .endr
    /*0x1000 = PAGE_SIZE*/
    .align  0x1000
trampoline_pdt_addr:
    address = 0
    .rept  2048
    /* 0x83 = (PAGE_PSE | PAGE_PRESENT | PAGE_RW) */
    .quad  address + 0x83
    address = address + 0x200000
    .endr

    .end