#include "ethernetif.h"

#include <stdbool.h>
#include <stdio.h>
#include <string.h>

#include "ch32fun.h"
#include "ch32v20xhw.h"
#include "lwip/etharp.h"
#include "lwip/snmp.h"
#include "systick.h"

#define IFNAME0 'e'
#define IFNAME1 'n'

typedef struct {
  volatile uint32_t head;  // producer idx: next free slot to write to
  volatile uint32_t tail;  // consumer idx: next slot to be txed
  volatile bool is_full;   // for N=1 size
} tx_queue_t;

struct ethernetif {
  ETH_DMADESCTypeDef* rx_desc_head;  // next desc to be filled by DMA
  ETH_DMADESCTypeDef* rx_desc_tail;  // next desc to be read by CPU
  tx_queue_t tx_q;
};

__attribute__((aligned(4))) ETH_DMADESCTypeDef g_dma_rx_descs[ETH_RX_BUF_COUNT];
__attribute__((aligned(4))) ETH_DMADESCTypeDef g_dma_tx_descs[ETH_TX_BUF_COUNT];
__attribute__((
    aligned(4))) uint8_t g_mac_rx_bufs[ETH_RX_BUF_COUNT * ETH_RX_BUF_SIZE];
__attribute__((
    aligned(4))) uint8_t g_mac_tx_bufs[ETH_TX_BUF_COUNT * ETH_TX_BUF_SIZE];

static struct ethernetif g_eth_state;
static volatile bool g_link_irq_flag = false;

static inline void tx_queue_init(tx_queue_t* q) {
  q->head = 0;
  q->tail = 0;
  q->is_full = false;
}

static inline bool tx_queue_is_empty(const tx_queue_t* q) {
  return !q->is_full && (q->head == q->tail);
}
static inline bool tx_queue_is_full(const tx_queue_t* q) { return q->is_full; }
static inline void tx_queue_produce(tx_queue_t* q) {
  q->head = (q->head + 1) % ETH_TX_BUF_COUNT;
  if (q->head == q->tail) {
    q->is_full = true;
  }
}
static inline void tx_queue_consume(tx_queue_t* q) {
  q->tail = (q->tail + 1) % ETH_TX_BUF_COUNT;
  q->is_full = false;
}

static void low_level_init(struct netif* netif);
static err_t low_level_output(struct netif* netif, struct pbuf* p);
static struct pbuf* low_level_input(struct netif* netif);
void phy_write_reg(uint8_t reg_add, uint16_t reg_val);
uint16_t phy_read_reg(uint8_t reg_add);

static void eth_get_mac_addr(uint8_t* mac) {
  // Mac is backwards.
  const uint8_t* macaddr_src = (const uint8_t*)(ROM_CFG_USERADR_ID + 5);
  for (int i = 0; i < 6; i++) {
    mac[i] = *(macaddr_src--);
  }
}

err_t ethernetif_init(struct netif* netif) {
#if LWIP_NETIF_HOSTNAME
  netif->hostname = "lwip-ch32";
#endif

  netif->state = &g_eth_state;
  netif->name[0] = IFNAME0;
  netif->name[1] = IFNAME1;

  netif->output = etharp_output;
  netif->linkoutput = low_level_output;

  MIB2_INIT_NETIF(netif, snmp_ifType_ethernet_csmacd, 10000000);  // 10Mbps

  netif->hwaddr_len = ETH_HWADDR_LEN;
  eth_get_mac_addr(netif->hwaddr);

  printf("MAC Address: %02X:%02X:%02X:%02X:%02X:%02X\n", netif->hwaddr[0],
         netif->hwaddr[1], netif->hwaddr[2], netif->hwaddr[3], netif->hwaddr[4],
         netif->hwaddr[5]);

  netif->mtu = 1500;
  netif->flags = NETIF_FLAG_BROADCAST | NETIF_FLAG_ETHARP;

  low_level_init(netif);

  return ERR_OK;
}

static void low_level_init(struct netif* netif) {
  struct ethernetif* ethernetif = netif->state;

  // clocks
  RCC->APB2PCENR |= RCC_APB2Periph_AFIO;
  RCC->CFGR0 |= RCC_ETHPRE;  // div 2
  EXTEN->EXTEN_CTR |= EXTEN_ETH_10M_EN;

  // reset mac rx and tx
  ETH10M->ECON1 = RB_ETH_ECON1_TXRST | RB_ETH_ECON1_RXRST;
  ETH10M->ECON1 = 0;

  // mac regs
  ETH10M->ERXFCON = RB_ETH_ERXFCON_BCEN | RB_ETH_ERXFCON_MCEN;
  ETH10M->MACON1 = RB_ETH_MACON1_MARXEN;
  ETH10M->MACON2 = PADCFG_AUTO_3 | RB_ETH_MACON2_TXCRCEN;
  ETH10M->MAMXFL = ETH_MAX_PACKET_SIZE;

  R8_ETH_MAADRL1 = netif->hwaddr[5];
  R8_ETH_MAADRL2 = netif->hwaddr[4];
  R8_ETH_MAADRL3 = netif->hwaddr[3];
  R8_ETH_MAADRL4 = netif->hwaddr[2];
  R8_ETH_MAADRL5 = netif->hwaddr[1];
  R8_ETH_MAADRL6 = netif->hwaddr[0];

  // PHY analog block
  ETH10M->ECON2 = RB_ETH_ECON2_DEFAULT;

  // init TX descriptors
  tx_queue_init(&ethernetif->tx_q);
  for (int i = 0; i < ETH_TX_BUF_COUNT; i++) {
    g_dma_tx_descs[i].Status = 0;
    g_dma_tx_descs[i].Buffer1Addr =
        (uint32_t)&g_mac_tx_bufs[i * ETH_TX_BUF_SIZE];
    g_dma_tx_descs[i].Buffer2NextDescAddr =
        (uint32_t)&g_dma_tx_descs[(i + 1) % ETH_TX_BUF_COUNT];
  }

  // init RX descriptors
  ethernetif->rx_desc_head = g_dma_rx_descs;
  ethernetif->rx_desc_tail = g_dma_rx_descs;
  for (int i = 0; i < ETH_RX_BUF_COUNT; i++) {
    g_dma_rx_descs[i].Status = ETH_DMARxDesc_OWN;
    g_dma_rx_descs[i].Buffer1Addr =
        (uint32_t)&g_mac_rx_bufs[i * ETH_RX_BUF_SIZE];
    g_dma_rx_descs[i].Buffer2NextDescAddr =
        (uint32_t)&g_dma_rx_descs[(i + 1) % ETH_RX_BUF_COUNT];
  }

  // set RX buffer start and enable receiver
  ETH10M->ERXST = ethernetif->rx_desc_head->Buffer1Addr;
  ETH10M->ECON1 = RB_ETH_ECON1_RXEN;

  phy_write_reg(PHY_BMCR, PHY_BMCR_RESET);
  Delay_Ms(200);

  phy_write_reg(PHY_BMCR, PHY_BMCR_FULL_DUPLEX);

  ETH10M->EIR = 0xFF;  // clear all interrupt flags
  ETH10M->EIE = RB_ETH_EIE_INTIE | RB_ETH_EIE_RXIE | RB_ETH_EIE_TXIE |
                RB_ETH_EIE_LINKIE | RB_ETH_EIE_TXERIE | RB_ETH_EIE_RXERIE |
                RB_ETH_EIE_R_EN50;

  NVIC_EnableIRQ(ETH_IRQn);
}

static void tx_start_if_possible(void) {
  // if TXRTS bit is set, MAC is busy sending a packet
  if (ETH10M->ECON1 & RB_ETH_ECON1_TXRTS) {
    return;
  }

  struct ethernetif* ethernetif = &g_eth_state;

  if (tx_queue_is_empty(&ethernetif->tx_q)) {
    return;
  }

  // get descriptor for the next packet to send
  uint32_t idx = ethernetif->tx_q.tail;
  ETH_DMADESCTypeDef* dma_desc = &g_dma_tx_descs[idx];

  uint16_t len = dma_desc->Status;

  // tell MAC which buffer to send
  ETH10M->ETXLN = len;
  ETH10M->ETXST = dma_desc->Buffer1Addr;
  // start tx
  ETH10M->ECON1 |= RB_ETH_ECON1_TXRTS;
}

static err_t low_level_output(struct netif* netif, struct pbuf* p) {
  struct ethernetif* ethernetif = netif->state;
  err_t errval = ERR_OK;

  if (tx_queue_is_full(&ethernetif->tx_q)) {
    // should this be ERR_BUF or ERR_MEM? does ERR_MEM re-queue the packet?
    // queue full, drop pkt
    errval = ERR_BUF;
    // errval = ERR_MEM;
  } else {
    uint32_t current_idx = ethernetif->tx_q.head;
    uint8_t* tx_buf_ptr = (uint8_t*)g_dma_tx_descs[current_idx].Buffer1Addr;
    uint32_t len = 0;

    for (struct pbuf* q = p; q != NULL; q = q->next) {
      memcpy(&tx_buf_ptr[len], q->payload, q->len);
      len += q->len;
    }

    g_dma_tx_descs[current_idx].Status = len;
    tx_queue_produce(&ethernetif->tx_q);
  }

  tx_start_if_possible();
  return errval;
}

static struct pbuf* low_level_input(struct netif* netif) {
  struct ethernetif* ethernetif = netif->state;
  struct pbuf* p = NULL;

  // if OWN bit is set, it's still owned by DMA and no packet rdy
  if (ethernetif->rx_desc_tail->Status & ETH_DMARxDesc_OWN) {
    return NULL;
  }

  // packet ready
  uint32_t len = (ethernetif->rx_desc_tail->Status & ETH_DMARxDesc_FL) >> 16;

  p = pbuf_alloc(PBUF_RAW, len, PBUF_POOL);
  if (p != NULL) {
    uint8_t* buffer = (uint8_t*)ethernetif->rx_desc_tail->Buffer1Addr;
    uint32_t offset = 0;
    for (struct pbuf* q = p; q != NULL; q = q->next) {
      memcpy(q->payload, buffer + offset, q->len);
      offset += q->len;
    }
    LINK_STATS_INC(link.recv);
  } else {
    LINK_STATS_INC(link.memerr);
    LINK_STATS_INC(link.drop);
  }

  // give buffer back to DMA
  ethernetif->rx_desc_tail->Status = ETH_DMARxDesc_OWN;
  // advance read pointer to the next descriptor in the ring
  ethernetif->rx_desc_tail =
      (ETH_DMADESCTypeDef*)ethernetif->rx_desc_tail->Buffer2NextDescAddr;

  return p;
}

void ethernetif_input(struct netif* netif) {
  struct pbuf* p;
  while ((p = low_level_input(netif)) != NULL) {
    if (netif->input(p, netif) != ERR_OK) {
      pbuf_free(p);
    }
  }
}

void ethernetif_link_poll(struct netif* netif) {
  if (!g_link_irq_flag) return;
  g_link_irq_flag = false;

  // supposedly, first read latches link status 2nd get cur val
  (void)phy_read_reg(PHY_BMSR);
  uint16_t bmsr = phy_read_reg(PHY_BMSR);

  if (bmsr & PHY_BMSR_LINK_STATUS) {
    if (!netif_is_link_up(netif)) {
      ETH10M->MACON2 |= RB_ETH_MACON2_FULDPX;
      netif_set_link_up(netif);
    }
  } else {
    if (netif_is_link_up(netif)) {
      netif_set_link_down(netif);
    }
  }
}

void ETH_IRQHandler(void) __attribute__((interrupt)) __attribute__((used));
void ETH_IRQHandler(void) {
  uint32_t flags = ETH10M->EIR;
  struct ethernetif* ethernetif = &g_eth_state;

  if (flags & RB_ETH_EIR_RXIF) {
    ETH10M->EIR = RB_ETH_EIR_RXIF;

    // descriptor should be owned by DMA
    if (ethernetif->rx_desc_head->Status & ETH_DMARxDesc_OWN) {
      ETH_DMADESCTypeDef* next_desc =
          (ETH_DMADESCTypeDef*)ethernetif->rx_desc_head->Buffer2NextDescAddr;

      // if next descriptor OWN bit is 0, ring is full and we must drop
      if (!(next_desc->Status & ETH_DMARxDesc_OWN)) {
        LINK_STATS_INC(link.drop);
      } else {
        // process and re-arm
        ethernetif->rx_desc_head->Status &= ~ETH_DMARxDesc_OWN;
        // write packet len into status field for CPU
        ethernetif->rx_desc_head->Status |=
            (ETH_DMARxDesc_FS | ETH_DMARxDesc_LS |
             (ETH10M->ERXLN << ETH_DMARxDesc_FrameLengthShift));
        // advance descripotor ptr
        ethernetif->rx_desc_head = next_desc;
        // re-arm receiver with new emtpy buf
        ETH10M->ERXST = (uint32_t)ethernetif->rx_desc_head->Buffer1Addr;
      }
    }
  }

  if (flags & RB_ETH_EIR_TXIF) {
    ETH10M->EIR = RB_ETH_EIR_TXIF;

    if (!tx_queue_is_empty(&ethernetif->tx_q)) {
      LINK_STATS_INC(link.xmit);
      tx_queue_consume(&ethernetif->tx_q);
    }

    tx_start_if_possible();
  }

  if (flags & RB_ETH_EIR_TXERIF) {
    ETH10M->EIR = RB_ETH_EIR_TXERIF;
    LINK_STATS_INC(link.err);

    if (!tx_queue_is_empty(&ethernetif->tx_q)) {
      tx_queue_consume(&ethernetif->tx_q);
    }
    tx_start_if_possible();
  }

  if (flags & RB_ETH_EIR_RXERIF) {
    ETH10M->EIR = RB_ETH_EIR_RXERIF;
    ETH10M->ECON1 |= RB_ETH_ECON1_RXEN;  // re-enable receiver
    LINK_STATS_INC(link.err);
  }

  if (flags & RB_ETH_EIR_LINKIF) {
    g_link_irq_flag = true;
    ETH10M->EIR = RB_ETH_EIR_LINKIF;
  }
}

void phy_write_reg(uint8_t reg_add, uint16_t reg_val) {
  R32_ETH_MIWR = (reg_add & RB_ETH_MIREGADR_MASK) | RB_ETH_MIWR_MIIWR |
                 (reg_val << RB_ETH_MIWR_DATA_SHIFT);
}

uint16_t phy_read_reg(uint8_t reg_add) {
  ETH10M->MIERGADR = reg_add;
  return ETH10M->MIRD;
}