#include <string.h>
#include "r8c20.h"
#include "gloss.h"
#include "tty.h"

#define SPIHW	1

#define DAT1	p3.b0

#define DAT2	p3.b1

/* n/a */
#define	PING	p4.b4

#define DAT0	p3.b3
#define DO	p3.b3

#define DAT3	p3.b4
#define CS	p3.b4

#define CLK	p3.b5

#define PRESENT (!p4.b3)

#define CMD	p3.b7
#define DI	p3.b7

int old_present = 5;

static byte buffer[512];

static void pause()
{
  int i;
  for (i=0; i<5; i++)
    asm("");
}

/* Table for CRC-7 (polynomial x^7 + x^3 + 1) */
const byte crc7_syndrome_table[256] = {
  0x00, 0x09, 0x12, 0x1b, 0x24, 0x2d, 0x36, 0x3f,
  0x48, 0x41, 0x5a, 0x53, 0x6c, 0x65, 0x7e, 0x77,
  0x19, 0x10, 0x0b, 0x02, 0x3d, 0x34, 0x2f, 0x26,
  0x51, 0x58, 0x43, 0x4a, 0x75, 0x7c, 0x67, 0x6e,
  0x32, 0x3b, 0x20, 0x29, 0x16, 0x1f, 0x04, 0x0d,
  0x7a, 0x73, 0x68, 0x61, 0x5e, 0x57, 0x4c, 0x45,
  0x2b, 0x22, 0x39, 0x30, 0x0f, 0x06, 0x1d, 0x14,
  0x63, 0x6a, 0x71, 0x78, 0x47, 0x4e, 0x55, 0x5c,
  0x64, 0x6d, 0x76, 0x7f, 0x40, 0x49, 0x52, 0x5b,
  0x2c, 0x25, 0x3e, 0x37, 0x08, 0x01, 0x1a, 0x13,
  0x7d, 0x74, 0x6f, 0x66, 0x59, 0x50, 0x4b, 0x42,
  0x35, 0x3c, 0x27, 0x2e, 0x11, 0x18, 0x03, 0x0a,
  0x56, 0x5f, 0x44, 0x4d, 0x72, 0x7b, 0x60, 0x69,
  0x1e, 0x17, 0x0c, 0x05, 0x3a, 0x33, 0x28, 0x21,
  0x4f, 0x46, 0x5d, 0x54, 0x6b, 0x62, 0x79, 0x70,
  0x07, 0x0e, 0x15, 0x1c, 0x23, 0x2a, 0x31, 0x38,
  0x41, 0x48, 0x53, 0x5a, 0x65, 0x6c, 0x77, 0x7e,
  0x09, 0x00, 0x1b, 0x12, 0x2d, 0x24, 0x3f, 0x36,
  0x58, 0x51, 0x4a, 0x43, 0x7c, 0x75, 0x6e, 0x67,
  0x10, 0x19, 0x02, 0x0b, 0x34, 0x3d, 0x26, 0x2f,
  0x73, 0x7a, 0x61, 0x68, 0x57, 0x5e, 0x45, 0x4c,
  0x3b, 0x32, 0x29, 0x20, 0x1f, 0x16, 0x0d, 0x04,
  0x6a, 0x63, 0x78, 0x71, 0x4e, 0x47, 0x5c, 0x55,
  0x22, 0x2b, 0x30, 0x39, 0x06, 0x0f, 0x14, 0x1d,
  0x25, 0x2c, 0x37, 0x3e, 0x01, 0x08, 0x13, 0x1a,
  0x6d, 0x64, 0x7f, 0x76, 0x49, 0x40, 0x5b, 0x52,
  0x3c, 0x35, 0x2e, 0x27, 0x18, 0x11, 0x0a, 0x03,
  0x74, 0x7d, 0x66, 0x6f, 0x50, 0x59, 0x42, 0x4b,
  0x17, 0x1e, 0x05, 0x0c, 0x33, 0x3a, 0x21, 0x28,
  0x5f, 0x56, 0x4d, 0x44, 0x7b, 0x72, 0x69, 0x60,
  0x0e, 0x07, 0x1c, 0x15, 0x2a, 0x23, 0x38, 0x31,
  0x46, 0x4f, 0x54, 0x5d, 0x62, 0x6b, 0x70, 0x79
};

byte crc7(byte crc, byte data)
{
  return crc7_syndrome_table[(crc << 1) ^ data];
}

/* Table for CRC-16 (polynomial x^16 + x^12 + x^5 + 1) */
const word crc16_syndrome_table[256] = {
  0x0000, 0x1021, 0x2042, 0x3063, 0x4084, 0x50a5, 0x60c6, 0x70e7,
  0x8108, 0x9129, 0xa14a, 0xb16b, 0xc18c, 0xd1ad, 0xe1ce, 0xf1ef,
  0x1231, 0x0210, 0x3273, 0x2252, 0x52b5, 0x4294, 0x72f7, 0x62d6,
  0x9339, 0x8318, 0xb37b, 0xa35a, 0xd3bd, 0xc39c, 0xf3ff, 0xe3de,
  0x2462, 0x3443, 0x0420, 0x1401, 0x64e6, 0x74c7, 0x44a4, 0x5485,
  0xa56a, 0xb54b, 0x8528, 0x9509, 0xe5ee, 0xf5cf, 0xc5ac, 0xd58d,
  0x3653, 0x2672, 0x1611, 0x0630, 0x76d7, 0x66f6, 0x5695, 0x46b4,
  0xb75b, 0xa77a, 0x9719, 0x8738, 0xf7df, 0xe7fe, 0xd79d, 0xc7bc,
  0x48c4, 0x58e5, 0x6886, 0x78a7, 0x0840, 0x1861, 0x2802, 0x3823,
  0xc9cc, 0xd9ed, 0xe98e, 0xf9af, 0x8948, 0x9969, 0xa90a, 0xb92b,
  0x5af5, 0x4ad4, 0x7ab7, 0x6a96, 0x1a71, 0x0a50, 0x3a33, 0x2a12,
  0xdbfd, 0xcbdc, 0xfbbf, 0xeb9e, 0x9b79, 0x8b58, 0xbb3b, 0xab1a,
  0x6ca6, 0x7c87, 0x4ce4, 0x5cc5, 0x2c22, 0x3c03, 0x0c60, 0x1c41,
  0xedae, 0xfd8f, 0xcdec, 0xddcd, 0xad2a, 0xbd0b, 0x8d68, 0x9d49,
  0x7e97, 0x6eb6, 0x5ed5, 0x4ef4, 0x3e13, 0x2e32, 0x1e51, 0x0e70,
  0xff9f, 0xefbe, 0xdfdd, 0xcffc, 0xbf1b, 0xaf3a, 0x9f59, 0x8f78,
  0x9188, 0x81a9, 0xb1ca, 0xa1eb, 0xd10c, 0xc12d, 0xf14e, 0xe16f,
  0x1080, 0x00a1, 0x30c2, 0x20e3, 0x5004, 0x4025, 0x7046, 0x6067,
  0x83b9, 0x9398, 0xa3fb, 0xb3da, 0xc33d, 0xd31c, 0xe37f, 0xf35e,
  0x02b1, 0x1290, 0x22f3, 0x32d2, 0x4235, 0x5214, 0x6277, 0x7256,
  0xb5ea, 0xa5cb, 0x95a8, 0x8589, 0xf56e, 0xe54f, 0xd52c, 0xc50d,
  0x34e2, 0x24c3, 0x14a0, 0x0481, 0x7466, 0x6447, 0x5424, 0x4405,
  0xa7db, 0xb7fa, 0x8799, 0x97b8, 0xe75f, 0xf77e, 0xc71d, 0xd73c,
  0x26d3, 0x36f2, 0x0691, 0x16b0, 0x6657, 0x7676, 0x4615, 0x5634,
  0xd94c, 0xc96d, 0xf90e, 0xe92f, 0x99c8, 0x89e9, 0xb98a, 0xa9ab,
  0x5844, 0x4865, 0x7806, 0x6827, 0x18c0, 0x08e1, 0x3882, 0x28a3,
  0xcb7d, 0xdb5c, 0xeb3f, 0xfb1e, 0x8bf9, 0x9bd8, 0xabbb, 0xbb9a,
  0x4a75, 0x5a54, 0x6a37, 0x7a16, 0x0af1, 0x1ad0, 0x2ab3, 0x3a92,
  0xfd2e, 0xed0f, 0xdd6c, 0xcd4d, 0xbdaa, 0xad8b, 0x9de8, 0x8dc9,
  0x7c26, 0x6c07, 0x5c64, 0x4c45, 0x3ca2, 0x2c83, 0x1ce0, 0x0cc1,
  0xef1f, 0xff3e, 0xcf5d, 0xdf7c, 0xaf9b, 0xbfba, 0x8fd9, 0x9ff8,
  0x6e17, 0x7e36, 0x4e55, 0x5e74, 0x2e93, 0x3eb2, 0x0ed1, 0x1ef0
};

word crc16(word crc, const byte data)
{
  return (crc << 8) ^ crc16_syndrome_table[((crc >> 8) ^ data) & 0xff];
}


static void
spi_init ()
{
  pd3.b = 0x00;

  pd3.b0 = 1; /* DAT1 */
  pd3.b1 = 1; /* DAT2 */
  pd3.b3 = 0; /* DO */
  pd3.b4 = 1; /* CS */
  pd3.b5 = 1; /* CLK */
  pd3.b7 = 1; /* DI */

  CLK = 0;
  CS = 1;
  DI = 1;

#if SPIHW
  pmr.b = 0x10; /* tx/rx */

  sser.b = 0x00; /* disable tx/rx */
  sscrh.b = 0x21; /* f/128 (156KHz), master, continuous */
  sscrl.b = 0x20; /* SOL high */
  sscrl.b = 0x20; /* SOL high */
  ssmr.b = 0x68; /* L when stops, change on even edges */
  sssr.b = 0; /* clear status */
  ssmr2.b = 0x40; /* three-wire, cmos, cs=port, SSCK */
  sser.b = 0x18; /* enable tx/rx */
#endif
}

static void
spi_slow ()
{
  sscrh.b = 0x21; /* f/128 (156KHz), master, continuous */
}

static void
spi_fast ()
{
  sscrh.b = 0x26; /* f/4 (5MHz), master, continuous */
}

spi_cs (int asserted)
{
  while (!sssr.tend)
    ;
  CS = asserted ? 0 : 1;
}

/* It's important that we send 0xff when reading back data, else the
   card gets confused.  Perhaps it thinks a new command has
   started.  */
#define spi_read() spi_data(0xff)

static byte
spi_data (byte c)
{
  byte rv = 0;
#if SPIHW
  sstdr = c;
  while (!sssr.rdrf)
    ;
  rv = ssrdr;
#else
  int i;
  //cprintf("\033[34m[%b:", c);
  for (i=0; i<8; i++)
    {
      DI = (c & 0x80) ? 1 : 0;
      c <<= 1;
      pause();
      CLK = 1;
      rv = (rv << 1) | DO;
      pause();
      CLK = 0;
    }
  //cprintf("%b]\033[0m", rv);
#endif
  return rv;
}

#if 0
void
spi_read_block (byte *buf, int cnt)
{
  /* We're running at f1/4 - 32 clocks per byte. */
  sstdr = 0xff;
  cnt --;
  while (1)
    {
      while (!sssr.tdre)
	;
      sstdr = 0xff;
      /* most of the transfer happens here */
      while (!sssr.rdrf)
	;
      *buf++ = ssrdr;
      if (--cnt == 0)
	break;
    }
  PING = 1;
  while (!sssr.rdrf)
    ;
  PING = 0;
  *buf++ = ssrdr;
}
#endif

void
wait_for_card_present ()
{
  int i;
  int present;
  CS = 1;
  while (1)
    {
      present = PRESENT;
      if (present == old_present)
	continue;
      old_present = present;
      if (present)
	{
	  cprintf("Card present\n");
	  break;
	}
      else
	cprintf("Card not present\n");
    }
  wait_ms (2);
  for (i=0; i<10; i++)
    spi_read ();
}

void
wait_for_card_not_present ()
{
  int present;
  while (1)
    {
      present = PRESENT;
      if (present == old_present)
	continue;
      old_present = present;
      if (present)
	cprintf("Card present\n");
      else
	{
	  cprintf("Card not present\n");
	  break;
	}
    }
}

static void
sdcmd (byte cmd, byte a0, byte a1, byte a2, byte a3)
{
  byte crc = 0;
  spi_data (0xff);
  spi_cs (1);
  /* This byte gives the card a few "me" clocks to synchronize things.  */
  spi_data (0xff);
  spi_data (0x40 | cmd);	crc = crc7(crc, 0x40|cmd);
  spi_data (a0);		crc = crc7(crc, a0);
  spi_data (a1);		crc = crc7(crc, a1);
  spi_data (a2);		crc = crc7(crc, a2);
  spi_data (a3);		crc = crc7(crc, a3);
  crc = (crc << 1) | 1;
  spi_data (crc);
}

static byte
sdr1 (int leave_open, byte *buf, int blen)
{
  int i;
  byte r1;
  for (i=0; i<80; i++)
    {
      r1 = spi_read ();
      if ((r1 & 0x80) == 0)
	break;
#if 0
      tty_putc('!');
      wait_ms (1);
#endif
    }
  if (buf && blen)
    {
      byte rdy, crc1, crc2;
      for (i=0; i<80; i++)
	{
	  rdy = spi_read ();
	  if (rdy == 0xfe)
	    break;
#if 0
	  tty_putc('?');
	  wait_ms (1);
#endif
	}
      while (blen--)
	*buf++ = spi_read ();
      crc1 = spi_read ();
      crc2 = spi_read ();
    }
  if (!leave_open)
    {
      spi_cs (0);
      /* Give the card a few more clocks to finalize the command
	 internally.  */
      spi_data (0xff);
    }
  return r1;
}

static byte
sdr7 (byte *r7)
{
  byte r1;
  int i;

  *r7++ = r1 = sdr1 (1, 0, 0);
  for (i=0; i<4; i++)
    *r7++ = spi_read ();
  return r1;
}

static byte
r1cmd (int cmd, byte a0, byte a1, byte a2, byte a3)
{
  sdcmd (cmd, a0, a1, a2, a3);
  return sdr1(0, 0, 0);
}

static void
sdacmd (byte cmd, byte a0, byte a1, byte a2, byte a3)
{
  byte crc = 0;
  byte r1;

  r1 = r1cmd (55, 0, 0, 0, 0);
  cprintf("acmd %d r1 %b\n", cmd, r1);
  spi_cs (1);
  crc = 0;
  spi_data (0xff);
  spi_data (0x40 | cmd);	crc = crc7(crc, 0x40|cmd);
  spi_data (a0);		crc = crc7(crc, a0);
  spi_data (a1);		crc = crc7(crc, a1);
  spi_data (a2);		crc = crc7(crc, a2);
  spi_data (a3);		crc = crc7(crc, a3);
  crc = (crc << 1) | 1;
  spi_data (crc);
}

static void
spi_wait_for (byte b)
{
  int i;
  for (i=100; i; i--)
    {
      if (spi_read () == b)
	return;
    }
}

sd_read_csd ()
{
  int i;
  byte csd[16];
  sdcmd (9, 0, 0, 0, 0);
  sdr1 (0, csd, 16);
  cprintf("CSD: ");
  for (i=0; i<16; i++)
    cprintf(" %b", csd[i]);
  cprintf("  ");
  for (i=0; i<16; i++)
    if (isgraph (csd[i]))
      tty_putc (csd[i]);
    else
      tty_putc('.');
  cprintf("\n");
}

sd_read_cid ()
{
  int i;
  byte csd[16];
  sdcmd (10, 0, 0, 0, 0);
  sdr1 (0, csd, 16);
  cprintf("CID: ");
  for (i=0; i<16; i++)
    cprintf(" %b", csd[i]);
  cprintf("  ");
  for (i=0; i<16; i++)
    if (isgraph (csd[i]))
      tty_putc (csd[i]);
    else
      tty_putc('.');
  cprintf("\n");
}

sd_read_scr ()
{
  int i;
  byte scr[8];
  sdacmd (51, 0, 0, 0, 0);
  sdr1 (0, scr, 8);
  cprintf("SCR: ");
  for (i=0; i<8; i++)
    cprintf(" %b", scr[i]);
  cprintf("  ");
  for (i=0; i<8; i++)
    if (isgraph (scr[i]))
      tty_putc (scr[i]);
    else
      tty_putc('.');
  cprintf("\n");
}

main()
{
  int i;
  byte r1, r7[5];

  tty_init ();

  pd4.b = 0x10;
  PING = 0;

  spi_init();

  while (1)
    {
      DAT1 = 0;
      wait_for_card_present ();

      spi_slow ();

      sdcmd (0, 0, 0, 0, 0);
      r1 = sdr1(0, 0, 0);
      cprintf("CMD0 Response: %b\n", r1);

      spi_fast ();

      /* The '1' means "2.7-3.3v" which is what we run at.  */
      sdcmd (8, 0, 0, 1, 0xaa);
      sdr7(r7);
      cprintf("CMD8 Response: %b %b %b %b %b\n", r7[0], r7[1], r7[2], r7[3], r7[4]);

      sdcmd (58, 0, 0, 0, 0);
      sdr7 (r7);
      cprintf("CMD58 Response: %b %b %b %b %b\n", r7[0], r7[1], r7[2], r7[3], r7[4]);

      for (i=0; i<100; i++)
	{
	  sdcmd (1, 0, 0, 0, 0);
	  r1 = sdr1(0, 0, 0);
	  if (r1 == 0x00)
	    break;
	  tty_putc('!');
	  wait_ms (100);
	}
      cprintf("CMD1 Response: %B\n", r1);

      sd_read_csd ();
      sd_read_cid ();
      sd_read_scr ();

      sdcmd (16, 0, 0, 2, 0);
      r1 = sdr1 (0, 0, 0);
      cprintf("CMD16 Response: %B\n", r1);

      sdcmd (17, 0, 0, 0, 0);
      r1 = sdr1 (1, 0, 0);
      cprintf("CMD17 Response: %B\n", r1);
      if (r1 == 0)
	{
	  byte d, c1, c2;
	  word crc = 0;
	  spi_wait_for (0xfe);
	  DAT1 = 1;
	  spi_read_block (buffer, 512);
	  DAT1 = 0;
	  for (i=0; i<512; i++)
	    {
	      d = buffer[i];
	      crc = crc16 (crc, d);
	      if (i % 32 == 0)
		cprintf("%w", i);
	      cprintf(" %b", d);
	      if (i % 32 == 31)
		cprintf("\n");
	    }
	  c1 = spi_read ();
	  c2 = spi_read ();
	  cprintf(" - crc: %b %b vs %w\n", c1, c2, crc);
	  spi_cs (1);
	}

      wait_for_card_not_present ();
    }
}