#include <amd64/apic.h>
#include <amd64/gdt.h>
#include <amd64/intr.h>
#include <amd64/intr_defs.h>
#include <amd64/mm.h>
#include <amd64/msr-index.h>
#include <amd64/msr.h>
#include <amd64/sse.h>
#include <fs/vfs.h>
#include <libk/std.h>
#include <libk/string.h>
#include <limine/requests.h>
#include <proc/proc.h>
#include <proc/reschedule.h>
#include <sync/spin_lock.h>
#include <sys/debug.h>
#include <sys/intr.h>
#include <sys/sched.h>
#include <sys/smp.h>
#include <sys/syscall.h>

/// The CPUs
static struct cpu cpus[CPUS_MAX];

static atomic_int cpu_counter;
static atomic_int cpu_id_counter = 0;

/// Allocate a CPU structure
struct cpu* cpu_make(uint64_t lapic_id, uint64_t acpi_id) {
  int id = atomic_fetch_add(&cpu_id_counter, 1);
  struct cpu* cpu = &cpus[id];

  memset(cpu, 0, sizeof(*cpu));
  cpu->lock = SPIN_LOCK_INIT;
  cpu->id = id;
  cpu->acpi_id = acpi_id;
  cpu->lapic_id = lapic_id;

  wrmsr(MSR_GS_BASE, (uint64_t)cpu);

  return cpu;
}

struct cpu* cpu_get(void) {
  struct cpu* ptr = (struct cpu*)rdmsr(MSR_GS_BASE);
  return ptr;
}

void cpu_request_sched(struct cpu* cpu, bool user) {
  if (cpu == thiscpu) {
    proc_sched(user);
    return;
  }

  lapic_ipi(cpu->lapic_id, INTR_CPU_REQUEST_SCHED);
}

struct cpu* cpu_find_lightest(void) {
  uint64_t fc;
  struct limine_mp_response* mp = limine_mp_request.response;

  int start = thiscpu->id;

  struct cpu* best_cpu = &cpus[start];

  spin_lock(&best_cpu->lock, &fc);
  int best_load = best_cpu->proc_run_q_count;
  spin_unlock(&best_cpu->lock, fc);

  for (int i = 1; i < (int)mp->cpu_count; i++) {
    int idx = (start + i) % mp->cpu_count;

    struct cpu* cpu = &cpus[idx];

    spin_lock(&cpu->lock, &fc);
    int l = cpu->proc_run_q_count;
    spin_unlock(&cpu->lock, fc);

    if (l < best_load) {
      best_load = l;
      best_cpu = cpu;
    }

    if (best_load == 0)
      break;
  }

  return best_cpu;
}

/// Bootstrap code for non-BSP CPUs
static void smp_bootstrap(struct limine_mp_info* mp_info) {
  uint64_t fc;

  load_kernel_cr3();

  struct cpu* cpu = cpu_make(mp_info->lapic_id, mp_info->processor_id);

  gdt_init(cpu);
  idt_load();
  syscall_init();
  sse_enable();

  lapic_init(1000);

  DEBUG("CPU %u is online!\n", thiscpu->id);

  cpu->kproc = kproc_create();

  atomic_fetch_sub(&cpu_counter, 1);

  intr_enable();

  struct reschedule_ctx rctx;
  memset(&rctx, 0, sizeof(rctx));

  struct proc* spin_proc = proc_from_file(thiscpu->kproc, "sys", "/spin", &rctx);
  proc_register(spin_proc, thiscpu, &rctx);

  spin_lock(&spin_proc->cpu->lock, &fc);
  do_sched(spin_proc, &spin_proc->cpu->lock);

  for (;;)
    ;
}

/// Initialize SMP subsystem for AMD64. Start AP CPUs
void smp_init(void) {
  struct limine_mp_response* mp = limine_mp_request.response;

  cpu_counter = mp->cpu_count - 1;

  for (size_t i = 0; i < mp->cpu_count; i++) {
    if (mp->cpus[i]->processor_id != thiscpu->acpi_id) {
      mp->cpus[i]->goto_address = &smp_bootstrap;
    }
  }

  DEBUG("Waiting for other CPUs:\n");

  while (atomic_load(&cpu_counter) > 0)
    ;

  DEBUG("All CPUs are up!\n");
}