#include <aux/compiler.h>
#include <aux/elf.h>
#include <irq/irq.h>
#include <libk/align.h>
#include <libk/list.h>
#include <libk/rbtree.h>
#include <libk/std.h>
#include <libk/string.h>
#include <limine/requests.h>
#include <mm/liballoc.h>
#include <mm/pmm.h>
#include <proc/proc.h>
#include <proc/procgroup.h>
#include <proc/resource.h>
#include <rd/rd.h>
#include <sync/spin_lock.h>
#include <sys/debug.h>
#include <sys/mm.h>
#include <sys/proc.h>
#include <sys/sched.h>
#include <sys/smp.h>
#include <sys/spin.h>

#if defined(__x86_64__)
#include <amd64/intr_defs.h>
#endif

#define SCHED_REAP_FREQ 10

static struct rb_node_link* proc_tree = NULL;
static spin_lock_t proc_tree_lock = SPIN_LOCK_INIT;

static atomic_int sched_cycles = 0;

static bool proc_check_elf (uint8_t* elf) {
  if (!((elf[0] == 0x7F) && (elf[1] == 'E') && (elf[2] == 'L') && (elf[3] == 'F')))
    return false;
  return true;
}

struct elf_aux proc_load_segments (struct proc* proc, uint8_t* elf) {
  struct elf_aux aux;

  Elf64_Ehdr* ehdr = (Elf64_Ehdr*)elf;
  aux.entry = ehdr->e_entry;
  aux.phnum = ehdr->e_phnum;
  aux.phent = ehdr->e_phentsize;

  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;

  for (uint64_t segment = 0; segment < ehdr->e_phnum; segment++) {
    Elf64_Phdr* phdr =
        (Elf64_Phdr*)((uintptr_t)elf + ehdr->e_phoff + (ehdr->e_phentsize * segment));

    switch (phdr->p_type) {
    case PT_PHDR: {
      aux.phdr = (uint64_t)phdr->p_vaddr;
    } break;
    case PT_LOAD: {
      uintptr_t v_addr = align_down (phdr->p_vaddr, PAGE_SIZE);
      uintptr_t off = phdr->p_vaddr - v_addr;

      size_t blks = div_align_up (phdr->p_memsz + off, PAGE_SIZE);

      uint32_t pg_flags = MM_PG_USER | MM_PG_PRESENT;
      if (phdr->p_flags & PF_W)
        pg_flags |= MM_PG_RW;

      uintptr_t p_addr;
      procgroup_map (proc->procgroup, v_addr, blks, pg_flags, &p_addr);

      memset ((void*)((uintptr_t)hhdm->offset + p_addr), 0, blks * PAGE_SIZE);
      memcpy ((void*)((uintptr_t)hhdm->offset + p_addr + off),
              (void*)((uintptr_t)elf + phdr->p_offset), phdr->p_filesz);
    } break;
    case PT_TLS: {
#if defined(__x86_64__)
      size_t tls_align = phdr->p_align ? phdr->p_align : sizeof (uintptr_t);
      size_t tls_size = phdr->p_memsz;
      size_t tls_total_needed = tls_size + sizeof (uintptr_t);
      size_t blks = div_align_up (tls_total_needed, PAGE_SIZE);
      proc->procgroup->tls.tls_tmpl_pages = blks;
      proc->procgroup->tls.tls_tmpl_size = tls_size;
      proc->procgroup->tls.tls_tmpl_total_size = tls_total_needed;

      proc->procgroup->tls.tls_tmpl = malloc (blks * PAGE_SIZE);
      memset (proc->procgroup->tls.tls_tmpl, 0, blks * PAGE_SIZE);

      memcpy (proc->procgroup->tls.tls_tmpl, (void*)((uintptr_t)elf + phdr->p_offset),
              phdr->p_filesz);

      proc_init_tls (proc);
#endif
    } break;
    }
  }

  return aux;
}

struct proc* proc_spawn_rd (char* name) {
  struct rd_file* rd_file = rd_get_file (name);

  bool ok = proc_check_elf (rd_file->content);

  if (!ok)
    return NULL;

  return proc_from_elf (rd_file->content);
}

struct proc* proc_find_pid (int pid) {
  spin_lock_ctx_t ctxprtr;
  struct proc* proc = NULL;

  spin_lock (&proc_tree_lock, &ctxprtr);
  rbtree_find (struct proc, &proc_tree, pid, proc, proc_tree_link, pid);
  spin_unlock (&proc_tree_lock, &ctxprtr);

  return proc;
}

void proc_register (struct proc* proc, struct cpu* cpu1) {
  spin_lock_ctx_t ctxcpu, ctxprtr;

  proc->cpu = cpu1 != NULL ? cpu1 : cpu_find_lightest ();

  struct cpu* cpu = proc->cpu;

  spin_lock (&proc_tree_lock, &ctxprtr);
  spin_lock (&cpu->lock, &ctxcpu);

  rbtree_insert (struct proc, &proc_tree, &proc->proc_tree_link, proc_tree_link, pid);

  atomic_fetch_add (&cpu->proc_run_q_count, 1);
  list_append (cpu->proc_run_q, &proc->cpu_run_q_link);
  if (cpu->proc_current == NULL)
    cpu->proc_current = proc;

  spin_unlock (&proc_tree_lock, &ctxprtr);
  spin_unlock (&cpu->lock, &ctxcpu);
}

/* caller holds cpu->lock */
static struct proc* proc_find_sched (struct cpu* cpu) {
  if (!cpu->proc_run_q)
    return NULL;

  struct list_node_link *current, *start;

  if (cpu->proc_current)
    current = cpu->proc_current->cpu_run_q_link.next;
  else
    current = cpu->proc_run_q;

  if (!current)
    current = cpu->proc_run_q;

  start = current;

  do {
    struct proc* proc = list_entry (current, struct proc, cpu_run_q_link);

    if (atomic_load (&proc->state) == PROC_READY)
      return proc;

    current = current->next ? current->next : cpu->proc_run_q;
  } while (current != start);

  return NULL;
}

static void proc_reap (void) {
  struct proc* proc = NULL;
  struct list_node_link* reap_list = NULL;
  spin_lock_ctx_t ctxprtr;
  spin_lock_ctx_t ctxpr;

  spin_lock (&proc_tree_lock, &ctxprtr);

  struct rb_node_link* node;
  rbtree_first (&proc_tree, node);

  while (node) {
    struct rb_node_link* next;
    rbtree_next (node, next);
    proc = rbtree_entry (node, struct proc, proc_tree_link);

    if (atomic_load (&proc->state) == PROC_DEAD) {
      spin_lock (&proc->lock, &ctxpr);
      rbtree_delete (&proc_tree, &proc->proc_tree_link);
      list_append (reap_list, &proc->reap_link);
      spin_unlock (&proc->lock, &ctxpr);
    }

    node = next;
  }

  spin_unlock (&proc_tree_lock, &ctxprtr);

  struct list_node_link *reap_link, *reap_link_tmp;
  list_foreach (reap_list, reap_link, reap_link_tmp) {
    proc = list_entry (reap_link, struct proc, reap_link);

    list_remove (reap_list, &proc->reap_link);
    DEBUG ("cleanup PID %d\n", proc->pid);
    proc_cleanup (proc);
  }
}

void proc_sched (void) {
  spin_lock_ctx_t ctxcpu;

  int s_cycles = atomic_fetch_add (&sched_cycles, 1);

  if (s_cycles % SCHED_REAP_FREQ == 0)
    proc_reap ();

  struct proc* next = NULL;
  struct cpu* cpu = thiscpu;

  spin_lock (&cpu->lock, &ctxcpu);

  next = proc_find_sched (cpu);

  if (next) {
    cpu->proc_current = next;

    do_sched (next, &cpu->lock, &ctxcpu);
  } else {
    cpu->proc_current = NULL;
    spin_unlock (&cpu->lock, &ctxcpu);

    spin ();
  }
}

void proc_kill (struct proc* proc) {
  spin_lock_ctx_t ctxpr, ctxcpu;
  struct cpu* cpu = proc->cpu;

  spin_lock (&proc->lock, &ctxpr);
  atomic_store (&proc->state, PROC_DEAD);
  proc->cpu = NULL;
  spin_unlock (&proc->lock, &ctxpr);

  spin_lock (&cpu->lock, &ctxcpu);

  list_remove (cpu->proc_run_q, &proc->cpu_run_q_link);
  atomic_fetch_sub (&cpu->proc_run_q_count, 1);
  if (cpu->proc_current == proc)
    cpu->proc_current = NULL;

  spin_unlock (&cpu->lock, &ctxcpu);

  DEBUG ("killed PID %d\n", proc->pid);

  cpu_request_sched (cpu);
}

static void proc_irq_sched (void* arg, void* regs) {
  (void)arg;
  proc_sched ();
}

void proc_init (void) {
#if defined(__x86_64__)
  irq_attach (&proc_irq_sched, NULL, SCHED_PREEMPT_TIMER);
  irq_attach (&proc_irq_sched, NULL, CPU_REQUEST_SCHED);
#endif

  struct proc* spin_proc = proc_spawn_rd ("spin.exe");
  proc_register (spin_proc, thiscpu);

  struct proc* init = proc_spawn_rd ("init.exe");
  proc_register (init, NULL);

  spin_lock_ctx_t ctxcpu;
  spin_lock (&init->cpu->lock, &ctxcpu);
  do_sched (init, &init->cpu->lock, &ctxcpu);
}