#include <aux/compiler.h>
#include <libk/std.h>
#include <libk/string.h>
#include <limine/requests.h>
#include <mm/pmm.h>
#include <sync/spin_lock.h>
#include <sys/mm.h>

#define AMD64_PG_PRESENT (1 << 0)
#define AMD64_PG_RW      (1 << 1)
#define AMD64_PG_USER    (1 << 2)

#define AMD64_PG_TABLE_ENTRIES_MAX 512

struct pg_index {
  uint16_t pml4, pml3, pml2, pml1;
} PACKED;

struct pd kernel_pd = {.lock = SPIN_LOCK_INIT};

static uintptr_t amd64_current_cr3 (void) {
  uintptr_t cr3;
  __asm__ volatile ("movq %%cr3, %0" : "=r"(cr3)::"memory");
  return cr3;
}

void amd64_load_kernel_cr3 (void) {
  __asm__ volatile ("movq %0, %%cr3" ::"r"(kernel_pd.cr3_paddr) : "memory");
}

static struct pg_index amd64_mm_page_index (uint64_t vaddr) {
  struct pg_index ret;

  ret.pml4 = ((vaddr >> 39) & 0x1FF);
  ret.pml3 = ((vaddr >> 30) & 0x1FF);
  ret.pml2 = ((vaddr >> 21) & 0x1FF);
  ret.pml1 = ((vaddr >> 12) & 0x1FF);

  return ret;
}

static uint64_t* amd64_mm_next_table (uint64_t* table, uint64_t entry_idx, bool alloc) {
  uint64_t entry = table[entry_idx];
  uint64_t paddr;

  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;

  if (entry & AMD64_PG_PRESENT)
    paddr = entry & ~0xFFFULL;
  else {
    if (!alloc)
      return NULL;

    paddr = pmm_alloc (1);

    if (paddr == 0)
      return NULL;

    memset ((void*)((uintptr_t)hhdm->offset + (uintptr_t)paddr), 0, PAGE_SIZE);
    table[entry_idx] = paddr | AMD64_PG_PRESENT | AMD64_PG_RW;
  }

  return (uint64_t*)((uintptr_t)hhdm->offset + (uintptr_t)paddr);
}

static uint64_t amd64_mm_resolve_flags (uint32_t generic) {
  uint64_t flags = 0;

  flags |= ((generic & MM_PG_PRESENT) ? AMD64_PG_PRESENT : 0);
  flags |= ((generic & MM_PG_RW) ? AMD64_PG_RW : 0);
  flags |= ((generic & MM_PG_USER) ? AMD64_PG_USER : 0);

  return flags;
}

static void amd64_reload_cr3 (void) {
  uint64_t cr3;
  __asm__ volatile ("movq %%cr3, %0; movq %0, %%cr3" : "=r"(cr3)::"memory");
}

void mm_map_page (struct pd* pd, uintptr_t paddr, uintptr_t vaddr, uint32_t flags) {
  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
  bool do_reload = false;

  if (flags & MM_PD_LOCK)
    spin_lock (&pd->lock);

  uint64_t amd64_flags = amd64_mm_resolve_flags (flags);

  uint64_t* pml4 = (uint64_t*)(pd->cr3_paddr + (uintptr_t)hhdm->offset);
  struct pg_index pg_index = amd64_mm_page_index (vaddr);

  uint64_t* pml3 = amd64_mm_next_table (pml4, pg_index.pml4, true);
  if (pml3 == NULL)
    goto done;

  uint64_t* pml2 = amd64_mm_next_table (pml3, pg_index.pml3, true);
  if (pml2 == NULL)
    goto done;

  uint64_t* pml1 = amd64_mm_next_table (pml2, pg_index.pml2, true);
  if (pml1 == NULL)
    goto done;

  uint64_t* pte = &pml1[pg_index.pml1];

  *pte = ((paddr & ~0xFFFULL) | (amd64_flags & 0x7ULL));
  do_reload = true;

done:
  if (do_reload)
    amd64_reload_cr3 ();

  if (flags & MM_PD_LOCK)
    spin_unlock (&pd->lock);
}

void mm_map_kernel_page (uintptr_t paddr, uintptr_t vaddr, uint32_t flags) {
  mm_map_page (&kernel_pd, paddr, vaddr, flags);
}

void mm_unmap_page (struct pd* pd, uintptr_t vaddr, uint32_t flags) {
  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
  bool do_reload = false;

  if (flags & MM_PD_LOCK)
    spin_lock (&pd->lock);

  uint64_t* pml4 = (uint64_t*)(pd->cr3_paddr + (uintptr_t)hhdm->offset);
  struct pg_index pg_index = amd64_mm_page_index (vaddr);

  uint64_t* pml3 = amd64_mm_next_table (pml4, pg_index.pml4, true);
  if (pml3 == NULL)
    goto done;

  uint64_t* pml2 = amd64_mm_next_table (pml3, pg_index.pml3, true);
  if (pml2 == NULL)
    goto done;

  uint64_t* pml1 = amd64_mm_next_table (pml2, pg_index.pml2, true);
  if (pml1 == NULL)
    goto done;

  uint64_t* pte = &pml1[pg_index.pml1];

  *pte &= ~AMD64_PG_PRESENT;
  do_reload = true;

done:
  if (do_reload)
    amd64_reload_cr3 ();

  if (flags & MM_PD_LOCK)
    spin_unlock (&pd->lock);
}

void mm_unmap_kernel_page (uintptr_t vaddr, uint32_t flags) {
  mm_unmap_page (&kernel_pd, vaddr, flags);
}

void mm_lock_kernel (void) { spin_lock (&kernel_pd.lock); }

void mm_unlock_kernel (void) { spin_unlock (&kernel_pd.lock); }

void mm_init (void) { kernel_pd.cr3_paddr = amd64_current_cr3 (); }