/* https://github.com/cirosantilli/x86-bare-metal-examples#smp */

/* Must be a multiple of 0x1000. */
.equ STARTUP_CODE_ADDRESS, 0x1000
.equ SPINLOCK_ADDRESS, 0x2000

#include "common.h"
BEGIN
    STAGE2
    CLEAR
    /* TODO do we need 32-bit mode? I think yes because the APIC register
     * FEE00300 is too high for 16-bit?
     */
    PROTECTED_MODE
    IDT_SETUP_48_ISRS
    REMAP_PIC_32
    PIT_GENERATE_FREQUENCY
    /* Each tick is 20us. */
    PIT_SET_FREQ 50000
    sti

    /* Setup the code that will be run
     * on the other processors when they start up.
     * Should be somewhere into the first 1Mb,
     * as processors start in real mode.
     */
    cld
    mov $init_len, %ecx
    mov $init, %esi
    mov $STARTUP_CODE_ADDRESS, %edi
    rep movsb

    /* Setup the value that threads will modify for us. */
    movb $0, SPINLOCK_ADDRESS

    /* Move data into the lower ICR register:
     * this should start the other processors.
     * - Destination Shorthand = 11 = all except self
     * - Trigger Mode = ?
     * - Level = ?
     * - Delivery Status = 0 = Idle
     * - Destination Mode = ? = Does not matter since shorthand used
     * - Delivery Mode = 110 = Startup
     * - Vector = ? = does it matter for SIPI?
     */

    /* Load address of ICR low dword into ESI. */
    mov PIC_ICR_ADDRESS, %esi
    /* Load ICR encoding for broadcast INIT IPI to all APs. */
    mov $0x000C4500, %eax
    /* Broadcast INIT IPI to all APs */
    mov %eax, (%esi)
    /* 10-millisecond delay loop. */
    PIT_SLEEP_TICKS $500
    /* Load ICR encoding for broadcast SIPI IP to all APs.
     * The low byte of this is the vector which encodes the staring address for the processors!
     * This address is multiplied by 0x1000: processors start at CS = vector * 0x100 and IP = 0.
     */
    mov $0x000C4600 + STARTUP_CODE_ADDRESS / 0x1000, %eax
    /* Broadcast SIPI IPI to all APs. */
    mov %eax, (%esi)
    /* 200-microsecond delay loop. */
    PIT_SLEEP_TICKS $10
    /* Broadcast second SIPI IPI to all APs */
    mov %eax, (%esi)

    /* TODO improve this spinlock. */
not_started:
    cmpb $1, SPINLOCK_ADDRESS
    jne not_started

    /* This only happens if another thread starts and changes the spinlock.
     * So if we see the message, SMP is working!
     * /
    VGA_PRINT_STRING $message

    /* Testing if it is possible in 16-bit real mode. */
    /*PRINT_STRING $message*/
    hlt
message:
    .asciz "SMP started"
IDT_48_ENTRIES
PIT_SLEEP_TICKS_GLOBALS
interrupt_handler:
    cmp PIT_ISR_NUMBER, 4(%esp)
    jne not_pit
    PIT_SLEEP_TICKS_HANDLER_UPDATE
not_pit:
    ret

/* Code that will run on the second, third,
 * etc. processors (Application Processors),
 */
.code16
init:
    xor %ax, %ax
    mov %ax, %ds
    movb $1, SPINLOCK_ADDRESS
    /* TODO mandatory?
     * - is lock prefix enough?
     * - is caching even on? I not because of CR0.CD and CR0.NW
     */
    wbinvd
    hlt
.equ init_len, . - init