override self/exe

[ginge.git] / loader / emu.c

/*
 * GINGE - GINGE Is Not Gp2x Emulator
 * (C) notaz, 2010-2011,2016
 *
 * This work is licensed under the MAME license, see COPYING file for details.
 */
// a "gentle" reminder
#ifdef __ARM_EABI__
#error loader is meant to be OABI!
#endif
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <alloca.h>
#include <ctype.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <unistd.h>
#include <signal.h>
#include <asm/ucontext.h>
#include <errno.h>
#include <time.h>
#include <sched.h>
#include <sys/resource.h>
#include <sys/ioctl.h>
#include <sys/syscall.h>
#include <linux/soundcard.h>
#include <linux/fb.h>
#include <linux/futex.h>

#include "header.h"
#include "../common/host_fb.h"
#include "../common/cmn.h"
#include "syscalls.h"
#include "realfuncs.h"
#include "llibc.h"

#if (DBG & 2) && !(DBG & 4)
#define LOG_IO_UNK
#endif
#if (DBG & 4)
#define LOG_IO
#endif
//#define LOG_SEGV

#ifdef LOG_IO
#define iolog log_io
#else
#define iolog(...)
#endif

#ifdef LOG_IO_UNK
#define iolog_unh log_io
#else
#define iolog_unh(...)
#endif

#ifdef LOG_SEGV
#define segvlog g_printf
#else
#define segvlog(...)
#endif

#if defined(LOG_IO) || defined(LOG_IO_UNK)
#include "mmsp2-regs.h"
#endif

typedef unsigned long long u64;
typedef unsigned int   u32;
typedef unsigned short u16;
typedef unsigned char  u8;

#define THREAD_STACK_SIZE 0x200000

static int fb_sync_thread_paused;
static int fb_sync_thread_futex;

static struct {
  u32 dstctrl;
  u32 dstaddr;
  u32 dststride;
  u32 srcctrl;
  u32 srcaddr;          //
  u32 srcstride;
  u32 srcforcolor;
  u32 srcbackcolor;
  u32 patctrl;          //
  u32 patforcolor;
  u32 patbackcolor;
  u32 size;
  u32 ctrl;             //
  u32 run;
  u32 intc;
  u32 srcfifo;
} blitter;

#define SRCCTRL_INVIDEO         (1 << 8)
#define SRCCTRL_SRCENB          (1 << 7)
#define CTRL_TRANSPARENCYENB    (1 << 11)

static struct {
  // mmsp2
  u16 mlc_stl_cntl;
  union {
    u32 mlc_stl_adr; // mlcaddress for pollux
    struct {
      u16 mlc_stl_adrl;
      u16 mlc_stl_adrh;
    };
  };
  u16 mlc_stl_pallt_a;
  union {
    u16 mlc_stl_pallt_d[256*2];
    u32 mlc_stl_pallt_d32[256];
  };

  // pollux
  u32 mlccontrol;
  u16 mlcpalette[256];

  // state
  void *umem;
  u32 old_mlc_stl_adr;
  u32 btn_state; // as seen through /dev/GPIO: 0PVdVu YXBA RLSeSt 0Ri0Dn 0Le0Up
  struct {
    u32 width, height;
    u32 stride;
    u32 bpp;
    u32 dirty_pal:2;
  } v;
} mmsp2;
#define pollux mmsp2 // so that code doesn't look that weird
enum {
  DIRTY_PAL_MMSP2 = 1,
  DIRTY_PAL_POLLUX = 2,
};


#if defined(LOG_IO) || defined(LOG_IO_UNK)
static void log_io(const char *pfx, u32 a, u32 d, int size)
{
  const char *fmt, *reg = "";
  switch (size) {
  case  8: fmt = "%s %08x       %02x %s\n"; d &= 0xff; break;
  case 32: fmt = "%s %08x %08x %s\n";       break;
  default: fmt = "%s %08x     %04x %s\n";   d &= 0xffff; break;
  }

  if ((a & ~0xffff) == 0x7f000000)
    reg = regnames[a & 0xffff];

  g_printf(fmt, pfx, a, d, reg);
}
#endif

static void memset16(void *dst, u32 pattern, int count)
{
  u32 *dl;
  u16 *d;
  
  d = (u16 *)((long)dst & ~1);
  if ((long)d & 2) {
    *d++ = pattern;
    count--;
  }
  dl = (void *)d;
  pattern |= pattern << 16;

  while (count >= 2) {
    *dl++ = pattern;
    count -= 2;
  }
  if (count)
    *(u16 *)dl = pattern;
}

static void blt_tr(void *dst, void *src, u32 trc, int w)
{
  u16 *d = (u16 *)((long)dst & ~1);
  u16 *s = (u16 *)((long)src & ~1);

  // XXX: optimize
  for (; w > 0; d++, s++, w--)
    if (*s != trc)
      *d = *s;
}

#define dump_blitter() \
{ \
  u32 *r = &blitter.dstctrl; \
  int i; \
  for (i = 0; i < 4*4; i++, r++) { \
    g_printf("%08x ", *r); \
    if ((i & 3) == 3) \
      g_printf("\n"); \
  } \
}

static void *uppermem_lookup(u32 addr, u8 **mem_end)
{
  // XXX: maybe support mirroring?
  if ((addr & 0xfe000000) != 0x02000000)
    return NULL;

  *mem_end = (u8 *)mmsp2.umem + 0x02000000;
  return (u8 *)mmsp2.umem - 0x02000000 + addr;
}

static void blitter_do(void)
{
  u8 *dst, *dste, *src = NULL, *srce = NULL;
  int w, h, sstrd, dstrd;
  int to_screen = 0;
  u32 bpp, addr;

  w = blitter.size & 0x7ff;
  h = (blitter.size >> 16) & 0x7ff;
  sstrd = blitter.srcstride;
  dstrd = blitter.dststride;

  // XXX: need to confirm this..
  addr = (blitter.dstaddr & ~3) | ((blitter.dstctrl & 0x1f) >> 3);

  // use dst bpp.. How does it do blits with different src bpp?
  bpp = (blitter.dstctrl & 0x20) ? 16 : 8;

  // maybe the screen?
  if (((w == 320 && h == 240) || // blit whole screen
       (w * h >= 320*240/2)) &&  // ..or at least half of the area
       mmsp2.mlc_stl_adr <= addr && addr < mmsp2.mlc_stl_adr + 320*240*2)
    to_screen = 1;

  dst = uppermem_lookup(addr, &dste);

  // XXX: assume fill if no SRCENB, but it could be pattern blit..
  if (blitter.srcctrl & SRCCTRL_SRCENB) {
    if (!(blitter.srcctrl & SRCCTRL_INVIDEO))
      goto bad_blit;

    addr = (blitter.srcaddr & ~3) | ((blitter.srcctrl & 0x1f) >> 3);
    src = uppermem_lookup(addr, &srce);
    if (src == NULL)
      goto bad_blit;

    if (src + sstrd * h > srce) {
      err("blit %08x->%08x %dx%d did not fit src\n",
        blitter.srcaddr, blitter.dstaddr, w, h);
      h = (srce - src) / sstrd;
    }
  }

  if (dst == NULL)
    goto bad_blit;

  if (dst + dstrd * h > dste) {
    err("blit %08x->%08x %dx%d did not fit dst\n",
      blitter.srcaddr, blitter.dstaddr, w, h);
    h = (dste - dst) / dstrd;
  }

  if (src != NULL) {
    // copy
    if (bpp == 16 && (blitter.ctrl & CTRL_TRANSPARENCYENB)) {
      u32 trc = blitter.ctrl >> 16;
      for (; h > 0; h--, dst += dstrd, src += sstrd)
        blt_tr(dst, src, trc, w);
    }
    else {
      for (; h > 0; h--, dst += dstrd, src += sstrd)
        memcpy(dst, src, w * bpp / 8);
    }
  }
  else {
    // fill. Assume the pattern is cleared and bg color is used
    u32 bgc = blitter.patbackcolor & 0xffff;
    if (bpp == 16) {
      for (; h > 0; h--, dst += dstrd)
        memset16(dst, bgc, w);
    }
    else {
      for (; h > 0; h--, dst += dstrd)
        memset(dst, bgc, w); // bgc?
    }
  }

  if (to_screen) {
    fb_sync_thread_futex = 1;
    g_futex_raw(&fb_sync_thread_futex, FUTEX_WAKE, 1, NULL);
  }
  return;

bad_blit:
  err("blit %08x->%08x %dx%d translated to %p->%p\n",
    blitter.srcaddr, blitter.dstaddr, w, h, src, dst);
  dump_blitter();
}

// FIXME: pass real dimensions to blitters
static void mlc_flip(void *src, int bpp, int stride)
{
  static int old_bpp;

  // only pass pal to host if it's dirty
  if (bpp <= 8 && mmsp2.v.dirty_pal) {
    if (mmsp2.v.dirty_pal == DIRTY_PAL_MMSP2)
      host_video_update_pal32(mmsp2.mlc_stl_pallt_d32);
    else
      host_video_update_pal16(mmsp2.mlcpalette);
    mmsp2.v.dirty_pal = 0;
  }

  if (bpp != old_bpp) {
    host_video_change_bpp(bpp);
    old_bpp = bpp;
  }

  switch (bpp) {
  case  4:
    host_video_blit4(src, 320, 240, stride);
    break;

  case  8:
    host_video_blit8(src, 320, 240, stride);
    break;

  case 16:
    host_video_blit16(src, 320, 240, stride);
    break;

  case 24:
    // TODO
    break;
  }
}

static void *fb_sync_thread(void *arg)
{
  unsigned long sigmask[2] = { ~0ul, ~0ul };
  struct timespec ts = { 0, 0 };
  int invalid_fb_addr = 1;
  int manual_refresh = 0;
  int frame_counter = 0;
  int wait_ret;

  // this thread can't run any signal handlers since the
  // app's stack/tls stuff will never be set up here
  sigmask[0] &= ~(1ul << (SIGSEGV - 1));
  g_rt_sigprocmask_raw(SIG_SETMASK, sigmask, NULL, sizeof(sigmask));

  //ret = setpriority(PRIO_PROCESS, 0, -1);
  //log("setpriority %d\n", ret);

  // tell the main thread we're done init
  fb_sync_thread_futex = 0;
  g_futex_raw(&fb_sync_thread_futex, FUTEX_WAKE, 1, NULL);

  while (1) {
    u8 *gp2x_fb, *gp2x_fb_end;

    wait_ret = g_futex_raw(&fb_sync_thread_futex, FUTEX_WAIT, 0, &ts);

    // this is supposed to be done atomically, but to make life
    // easier ignore it for now, race impact is low anyway
    fb_sync_thread_futex = 0;

    if (wait_ret != 0 && wait_ret != -EWOULDBLOCK
        && wait_ret != -ETIMEDOUT)
    {
      err("fb_thread: futex error: %d\n", wait_ret);
      sleep(1);
      goto check_keys;
    }
    if (fb_sync_thread_paused) {
      ts.tv_nsec = 100000000;
      goto check_keys;
    }

    if (wait_ret == 0) {
      ts.tv_nsec = 50000000;
      manual_refresh++;
      if (manual_refresh == 2)
        dbg("fb_thread: switch to manual refresh\n");
    } else {
      ts.tv_nsec = 16666667;
      if (manual_refresh > 1)
        dbg("fb_thread: switch to auto refresh\n");
      manual_refresh = 0;
    }

    gp2x_fb = uppermem_lookup(mmsp2.mlc_stl_adr, &gp2x_fb_end);
    if (gp2x_fb == NULL || gp2x_fb + 320*240 * mmsp2.v.bpp / 8 > gp2x_fb_end) {
      if (!invalid_fb_addr) {
        err("fb_thread: %08x is out of range\n", mmsp2.mlc_stl_adr);
        invalid_fb_addr = 1;
      }
      continue;
    }

    invalid_fb_addr = 0;
    mlc_flip(gp2x_fb, mmsp2.v.bpp, mmsp2.v.stride);

    frame_counter++;
    if (frame_counter & 0x0f)
      continue;

check_keys:
    // this is to check for kill key, in case main thread hung
    // or something else went wrong.
    pollux.btn_state = host_read_btns();
  }
}

static void fb_thread_pause(void)
{
  fb_sync_thread_paused = 1;
  // wait until it finishes last refresh
  // that it might be doing now
  usleep(10000);
}

static void fb_thread_resume(void)
{
  fb_sync_thread_paused = 0;
}

static u32 xread32_io_cmn(u32 a, u32 *handled)
{
  u32 d = 0;

  *handled = 1;
  switch (a) {
  // Wiz stuff
  case 0x402c: // MLCVSTRIDE0
  case 0x4060: // MLCVSTRIDE1
    d = pollux.v.stride;
    break;
  case 0x4038: // MLCADDRESS0
  case 0x406c: // MLCADDRESS1
    d = pollux.mlc_stl_adr;
    break;
  // wiz_lib reads:
  //  ???? ???? YXBA DURiLe ???? VdVuMS LR?? ????
  // |     GPIOC[31:16]    |    GPIOB[31:16]     |
  case 0xa058: // GPIOBPAD
    d =  (pollux.btn_state >> 1) & 0x0100;
    d |= (pollux.btn_state << 1) & 0x0200;
    d |= (pollux.btn_state >> 3) & 0x0080;
    d |= (pollux.btn_state >> 5) & 0x0040;
    d |= (pollux.btn_state >> 6) & 0x0c00;
    d <<= 16;
    d = ~d;
    break;
  case 0xa098: // GPIOCPAD
    pollux.btn_state = host_read_btns();
    d =  (pollux.btn_state >> 8) & 0x00f0;
    d |= (pollux.btn_state >> 1) & 0x0008;
    d |= (pollux.btn_state << 2) & 0x0004;
    d |= (pollux.btn_state >> 5) & 0x0002;
    d |= (pollux.btn_state >> 2) & 0x0001;
    d <<= 16;
    d = ~d;
    break;
  default:
    *handled = 0;
    break;
  }

  return d;
}

static u32 xread8(u32 a)
{
  iolog("r8 ", a, 0, 8);
  iolog_unh("r8 ", a, 0, 8);
  return 0;
}

static u32 xread16(u32 a)
{
  static u32 fudge, old_a;
  u32 d = 0, t;

  if ((a & 0xffff0000) == 0x7f000000) {
    u32 a_ = a & 0xffff;
    switch (a_) {
    case 0x0910: // FPLL
    case 0x0912:
      d = 0x9407;
      break;
    // minilib reads as:
    //  0000 P000 VuVd00 0000 YXBA RLSeSt 0Ri0D 0Le0U
    // |        GPIOD        |GPIOC[8:15]|GPIOM[0:7] |
    // /dev/GPIO:
    //             ... 0PVdVu ...
    case 0x1184: // GPIOC
      d = ~mmsp2.btn_state & 0xff00;
      d |= 0x00ff;
      break;
    case 0x1186: // GPIOD
      t = ~mmsp2.btn_state;
      d  = (t >> 9)  & 0x0080;
      d |= (t >> 11) & 0x0040;
      d |= (t >> 7)  & 0x0800;
      d |= 0x373b;
      break;
    case 0x1198: // GPIOM
      mmsp2.btn_state = host_read_btns();
      d = ~mmsp2.btn_state & 0xff;
      d |= 0x01aa;
      break;
    case 0x1836: // reserved
      d = 0x2330;
      break;
    case 0x2816: // DPC_X_MAX
      d = 319;
      break;
    case 0x2818: // DPC_Y_MAX
      d = 239;
      break;
    case 0x28da:
      d = mmsp2.mlc_stl_cntl;
      break;
    case 0x290e:
    case 0x2912:
      d = mmsp2.mlc_stl_adrl;
      break;
    case 0x2910:
    case 0x2914:
      d = mmsp2.mlc_stl_adrh;
      break;
    case 0x2958:
      d = mmsp2.mlc_stl_pallt_a;
      break;

    default:
      d = xread32_io_cmn(a_, &t);
      if (!t)
        goto unk;
      if (!(a_ & 2))
        d >>= 16;
      break;
    }
    goto out;
  }

unk:
  if (a == old_a) {
    d = fudge;
    fudge = ~fudge;
  }
  old_a = a;
  iolog_unh("r16", a, d & 0xffff, 16);

out:
  d &= 0xffff;
  iolog("r16", a, d, 16);
  return d;
}

static u32 xread32(u32 a)
{
  u32 d = 0;
  if ((a & 0xfff00000) == 0x7f000000) {
    u32 a_ = a & 0xffff;
    struct timespec ts;
    u64 t64;
    u32 t;

    switch (a_) {
    case 0x0a00: // TCOUNT, 1/7372800s
      g_clock_gettime_raw(CLOCK_REALTIME, &ts);
      t64 = (u64)ts.tv_sec * 1000000000 + ts.tv_nsec;
      // t * 7372800.0 / 1000000000 * 0x100000000 ~= t * 31665935
      t64 *= 31665935;
      d = t64 >> 32;
      break;

    default:
      d = xread32_io_cmn(a_, &t);
      if (!t)
        goto unh;
      break;
    }
    goto out;
  }
  if ((a & 0xfff00000) == 0x7f100000) {
    u32 *bl = &blitter.dstctrl;
    u32 a_ = a & 0xfff;
    if (a_ < 0x40) {
      d = bl[a_ / 4];
      if (a_ == 0x34)
        d = 0; // not busy
      goto out;
    }
  }

unh:
  iolog_unh("r32", a, d, 32);

out:
  iolog("r32", a, d, 32);
  return d;
}

static void xwrite8(u32 a, u32 d)
{
  iolog("w8 ", a, d, 8);
  iolog_unh("w8 ", a, d, 8);
}

static void xwrite16(u32 a, u32 d)
{
  iolog("w16", a, d, 16);
  if ((a & 0xfff00000) == 0x7f000000) {
    u32 a_ = a & 0xffff;
    switch (a_) {
      case 0x28da: {
        int mode;
        mmsp2.mlc_stl_cntl = d | 0xaa;
        mode = (d >> 9) & 3;
        mmsp2.v.bpp = mode ? mode * 8 : 4;
        break;
      }
      case 0x290c:
        mmsp2.v.stride = d;
        return;
      case 0x290e:
      case 0x2910:
        // odd addresses don't affect LCD. What about TV?
        return;
      case 0x2912:
        mmsp2.mlc_stl_adrl = d;
        return;
      case 0x2914:
        mmsp2.mlc_stl_adrh = d;
        if (mmsp2.mlc_stl_adr != mmsp2.old_mlc_stl_adr) {
          // ask for refresh
          fb_sync_thread_futex = 1;
          g_futex_raw(&fb_sync_thread_futex, FUTEX_WAKE, 1, NULL);
        }
        mmsp2.old_mlc_stl_adr = mmsp2.mlc_stl_adr;
        return;
      case 0x2958:
        mmsp2.mlc_stl_pallt_a = d & 0x1ff;
        return;
      case 0x295a:
        mmsp2.mlc_stl_pallt_d[mmsp2.mlc_stl_pallt_a++] = d;
        mmsp2.mlc_stl_pallt_a &= 0x1ff;
        mmsp2.v.dirty_pal = DIRTY_PAL_MMSP2;
        return;
    }
  }
  iolog_unh("w16", a, d, 16);
}

static void xwrite32(u32 a, u32 d)
{
  iolog("w32", a, d, 32);

  if ((a & 0xfff00000) == 0x7f000000) {
    u32 a_ = a & 0xffff;
    switch (a_) {
    // Wiz
    case 0x4024: // MLCCONTROL0
    case 0x4058: // MLCCONTROL1
      pollux.mlccontrol = d;
      if (!(d & 0x20))
        return; // layer not enabled
      if ((d >> 16) == 0x443A)
        pollux.v.bpp = 8;
      else
        pollux.v.bpp = 16;
      return;
    case 0x402c: // MLCVSTRIDE0
    case 0x4060: // MLCVSTRIDE1
      pollux.v.stride = d;
      return;
    case 0x4038: // MLCADDRESS0
    case 0x406c: // MLCADDRESS1
      pollux.mlc_stl_adr = d;
      if (d != mmsp2.old_mlc_stl_adr) {
        // ask for refresh
        fb_sync_thread_futex = 1;
        g_futex_raw(&fb_sync_thread_futex, FUTEX_WAKE, 1, NULL);
      }
      mmsp2.old_mlc_stl_adr = d;
      return;
    case 0x403c: // MLCPALETTE0
    case 0x4070: // MLCPALETTE1
      pollux.mlcpalette[d >> 24] = d;
      pollux.v.dirty_pal = DIRTY_PAL_POLLUX;
      return;
    }
  }
  if ((a & 0xfff00000) == 0x7f100000) {
    u32 *bl = &blitter.dstctrl;
    u32 a_ = a & 0xfff;
    if (a_ < 0x40) {
      bl[a_ / 4] = d;
      if (a_ == 0x34 && (d & 1))
        blitter_do();
      return;
    }
  }
  iolog_unh("w32", a, d, 32);
}

#define LINKPAGE_SIZE 0x1000
#define LINKPAGE_COUNT 4
#define LINKPAGE_ALLOC (LINKPAGE_SIZE * LINKPAGE_COUNT)

struct op_context {
  u32 pc;
  u32 op;
  u32 code[0];
};

struct op_linkpage {
  void (*handler)(struct op_context *op_ctx);
  u32 code[0];
};

struct op_stackframe {
  u32 saved_regs[15];
  u32 cpsr;
};

static struct op_linkpage *g_linkpage;
static u32 *g_code_ptr;
static int g_linkpage_count;

enum opcond {
  C_EQ, C_NE, C_CS, C_CC, C_MI, C_PL, C_VS, C_VC,
  C_HI, C_LS, C_GE, C_LT, C_GT, C_LE, C_AL,
};
enum cpsr_cond {
  CPSR_N = (1u << 31),
  CPSR_Z = (1u << 30),
  CPSR_C = (1u << 29),
  CPSR_V = (1u << 28),
};

#define BIT_SET(v, b) (v & (1 << (b)))

void emu_handle_op(struct op_context *op_ctx, struct op_stackframe *sframe)
{
  u32 *regs = sframe->saved_regs;
  u32 cpsr = sframe->cpsr;
  u32 op = op_ctx->op;
  u32 t, shift, ret, addr;
  int i, rn, rd, cond;

  cond = (op & 0xf0000000) >> 28;
  rd = (op & 0x0000f000) >> 12;
  rn = (op & 0x000f0000) >> 16;

  if (cond != 0x0e) {
    switch (cond) {
    case C_EQ: if ( (cpsr & CPSR_Z)) break; return;
    case C_NE: if (!(cpsr & CPSR_Z)) break; return;
    case C_CS: if ( (cpsr & CPSR_C)) break; return;
    case C_CC: if (!(cpsr & CPSR_C)) break; return;
    case C_MI: if ( (cpsr & CPSR_N)) break; return;
    case C_PL: if (!(cpsr & CPSR_N)) break; return;
    case C_VS: if ( (cpsr & CPSR_V)) break; return;
    case C_VC: if (!(cpsr & CPSR_V)) break; return;
    default:
      goto unhandled;
    }
  }

  if ((op & 0x0f200090) == 0x01000090) { // AM3: LDRH, STRH
    if (!BIT_SET(op, 5)) // !H
      goto unhandled;
    if (BIT_SET(op, 6) && !BIT_SET(op, 20)) // S && !L
      goto unhandled;

    if (BIT_SET(op, 22))                // imm offset
      t = ((op & 0xf00) >> 4) | (op & 0x0f);
    else                                // reg offset
      t = regs[op & 0x000f];

    if (!BIT_SET(op, 23))
      t = -t;
    addr = regs[rn] + t;

    if (BIT_SET(op, 20)) { // Load
      ret = xread16(addr);
      if (BIT_SET(op, 6)) { // S
        ret <<= 16;
        ret = (signed int)ret >> 16;
      }
      regs[rd] = ret;
    }
    else
      xwrite16(addr, regs[rd]);
  }
  else if ((op & 0x0c000000) == 0x04000000) { // load/store word/byte
    if (BIT_SET(op, 21))
      goto unhandled;                   // unprivileged
    if (BIT_SET(op, 25)) {              // reg offs
      if (BIT_SET(op, 4))
        goto unhandled;                 // nah it's media

      t = regs[op & 0x000f];
      shift = (op & 0x0f80) >> 7;
      switch ((op & 0x0060) >> 5) {
        case 0: t = t << shift; break;
        case 1: t = t >> (shift + 1); break;
        case 2: t = (signed int)t >> (shift + 1); break;
        case 3: goto unhandled; // I'm just lazy
      }
    }
    else                                // imm offs
      t = op & 0x0fff;

    if (!BIT_SET(op, 23))
      t = -t;

    addr = regs[rn];
    if (BIT_SET(op, 24))   // pre-indexed
      addr += t;
    if (!BIT_SET(op, 24) || BIT_SET(op, 21))
      regs[rn] += t;       // writeback

    if (BIT_SET(op, 20)) { // Load
      if (BIT_SET(op, 22)) // Byte
        ret = xread8(addr);
      else
        ret = xread32(addr);
      regs[rd] = ret;
    }
    else {
      if (BIT_SET(op, 22)) // Byte
        xwrite8(addr, regs[rd]);
      else
        xwrite32(addr, regs[rd]);
    }
  }
  else
    goto unhandled;

#if 0
  if (addr != addr_check) {
    fprintf(stderr, "bad calculated addr: %08x vs %08x\n", addr, addr_check);
    abort();
  }
#endif
  return;

unhandled:
  err("unhandled IO op %08x @ %08x\n", op, op_ctx->pc);
  for (i = 0; i < 8-1; i++)
    err(" r%d=%08x  r%-2d=%08x\n", i, regs[i], i+8, regs[i+8]);
  err(" r%d=%08x cpsr=%08x\n", i, regs[i], cpsr);
  abort();
}

static u32 make_offset12(u32 *pc, u32 *target)
{
  int lp_offs, u = 1;

  lp_offs = (char *)target - (char *)pc - 2*4;
  if (lp_offs < 0) {
    lp_offs = -lp_offs;
    u = 0;
  }
  if (lp_offs >= LINKPAGE_SIZE) {
    err("linkpage too far: %d\n", lp_offs);
    abort();
  }

  return (u << 23) | lp_offs;
}

static u32 make_jmp(u32 *pc, u32 *target, int bl)
{
  int jmp_val;

  jmp_val = target - pc - 2;
  if (jmp_val < (int)0xff000000 || jmp_val > 0x00ffffff) {
    err("jump out of range (%p -> %p)\n", pc, target);
    abort();
  }

  return 0xea000000 | (bl << 24) | (jmp_val & 0x00ffffff);
}

static void emit_op(u32 op)
{
  *g_code_ptr++ = op;
}

static void emit_op_io(u32 op, u32 *target)
{
  op |= make_offset12(g_code_ptr, target);
  emit_op(op);
}

static void init_linkpage(void)
{
  g_linkpage->handler = emu_call_handle_op;
  g_code_ptr = g_linkpage->code;
}

static void segv_sigaction(int num, siginfo_t *info, void *ctx)
{
  struct ucontext *context = ctx;
  u32 *regs = (u32 *)&context->uc_mcontext.arm_r0;
  u32 *pc = (u32 *)regs[15];
  struct op_context *op_ctx;
  int i, lp_size;

  if (((regs[15] ^ (u32)&segv_sigaction) & 0xff000000) == 0 ||         // PC is in our segment or
      (((regs[15] ^ (u32)g_linkpage) & ~(LINKPAGE_ALLOC - 1)) == 0) || // .. in linkpage
      ((long)info->si_addr & 0xffe00000) != 0x7f000000)                // faulting not where expected
  {
    // real crash - time to die
    err("segv %d %p @ %08x\n", info->si_code, info->si_addr, regs[15]);
    for (i = 0; i < 8; i++)
      dbg(" r%d=%08x r%-2d=%08x\n", i, regs[i], i+8, regs[i+8]);
    signal(num, SIG_DFL);
    raise(num);
    return;
  }
  segvlog("segv %d %p @ %08x\n", info->si_code, info->si_addr, regs[15]);

  // spit PC and op
  op_ctx = (void *)g_code_ptr;
  op_ctx->pc = (u32)pc;
  op_ctx->op = *pc;
  g_code_ptr = &op_ctx->code[0];

  // emit jump to code ptr
  *pc = make_jmp(pc, g_code_ptr, 0);

  // generate code:
  emit_op   (0xe50d0000 + 0xf00 - 4 * 0);                        // str r0, [sp, #(-0xf00 + r0_offs)]
  emit_op   (0xe50de000 + 0xf00 - 4 * 14);                       // str lr, [sp, #(-0xf00 + lr_offs)]
  emit_op   (0xe24f0000 + (g_code_ptr - (u32 *)op_ctx + 2) * 4); // sub r0, pc, #op_ctx
  emit_op   (0xe1a0e00f);                                        // mov lr, pc
  emit_op_io(0xe51ff000, (u32 *)&g_linkpage->handler);           // ldr pc, =handle_op
  emit_op   (0xe51de000 + 0xf00 - 4 * 14);                       // ldr lr, [sp, #(-0xf00 + lr_offs)]
  emit_op   (make_jmp(g_code_ptr, pc + 1, 0));                   // jmp <back>

  // sync caches
  sys_cacheflush(pc, pc + 1);
  sys_cacheflush(g_linkpage, g_code_ptr);

  lp_size = (char *)g_code_ptr - (char *)g_linkpage;
  segvlog("code #%d %d/%d\n", g_linkpage_count, lp_size, LINKPAGE_SIZE);

  if (lp_size + 13*4 > LINKPAGE_SIZE) {
    g_linkpage_count++;
    if (g_linkpage_count >= LINKPAGE_COUNT) {
      err("too many linkpages needed\n");
      abort();
    }
    g_linkpage = (void *)((char *)g_linkpage + LINKPAGE_SIZE);
    init_linkpage();
  }
  //handle_op(regs[15], op, regs, (u32)info->si_addr);
  //regs[15] += 4;
}

void emu_init(void *map_bottom)
{
  sigaction_t segv_action = {
    .sa_sigaction = segv_sigaction,
    .sa_flags = SA_SIGINFO,
  };
  void *pret;
  int ret;

#ifdef PND
  if (geteuid() == 0) {
    err("don't try to run as root, device registers or memory "
        "might get trashed crashing the OS or even damaging the device.\n");
    exit(1);
  }
#endif

  g_linkpage = (void *)(((u32)map_bottom - LINKPAGE_ALLOC) & ~0xfff);
  pret = mmap(g_linkpage, LINKPAGE_ALLOC, PROT_READ|PROT_WRITE,
              MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, -1, 0);
  if (pret != g_linkpage) {
    perror(PFX "mmap linkpage");
    exit(1);
  }
  log("linkpages @ %p\n", g_linkpage);
  init_linkpage();

  // host stuff
  ret = host_init();
  if (ret != 0) {
    err("can't init host\n");
    exit(1);
  }

  ret = host_video_init(NULL, 0);
  if (ret != 0) {
    err("can't init host video\n");
    exit(1);
  }

  // TODO: check if this really fails on Wiz..
  mmsp2.umem = mmap(NULL, 0x2000000, PROT_READ|PROT_WRITE|PROT_EXEC,
                    MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
#ifdef WIZ
  if (mmsp2.umem == MAP_FAILED) {
    // we are short on memmory on Wiz, need special handling
    extern void *host_mmap_upper(void);
    mmsp2.umem = host_mmap_upper();
  }
#endif
  if (mmsp2.umem == MAP_FAILED) {
    perror(PFX "mmap upper mem");
    exit(1);
  }

  pret = mmap(NULL, THREAD_STACK_SIZE, PROT_READ|PROT_WRITE|PROT_EXEC,
              MAP_PRIVATE|MAP_ANONYMOUS|MAP_GROWSDOWN, -1, 0);
  if (mmsp2.umem == MAP_FAILED) {
    perror(PFX "mmap thread stack");
    exit(1);
  }
  fb_sync_thread_futex = 1;
  ret = g_clone(CLONE_VM | CLONE_FS | CLONE_FILES
                | CLONE_SIGHAND | CLONE_THREAD,
                (char *)pret + THREAD_STACK_SIZE, 0, 0, 0,
                fb_sync_thread);
  if (ret == 0 || ret == -1) {
    perror(PFX "start fb thread");
    exit(1);
  }
  g_futex_raw(&fb_sync_thread_futex, FUTEX_WAIT, 1, NULL);

  // defaults
  mmsp2.mlc_stl_adr = 0x03101000; // fb2 is at 0x03381000
  mmsp2.mlc_stl_cntl = 0x4ab; // 16bpp, region 1 active
  mmsp2.v.width = 320;
  mmsp2.v.height = 240;
  mmsp2.v.stride = 320*2;
  mmsp2.v.bpp = 16;
  mmsp2.v.dirty_pal = 1;

  sigemptyset(&segv_action.sa_mask);
  sigaction(SIGSEGV, &segv_action, NULL);
}

long emu_read_gpiodev(void *buf, int count)
{
  if (count <= 0) {
    err("gpiodev read %d?\n", count);
    return -EINVAL;
  }
  if (count > 4)
    count = 4;

  mmsp2.btn_state = host_read_btns();
  memcpy(buf, &mmsp2.btn_state, count);
  return count;
}

static long emu_mmap_dev(unsigned int length, int prot, int flags, unsigned int offset)
{
  u8 *umem, *umem_end;

  // SoC regs
  if ((offset & ~0x1ffff) == 0xc0000000) {
    return g_mmap2_raw((void *)0x7f000000, length, PROT_NONE,
      MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_NORESERVE, -1, 0);
  }
  // MMSP2 blitter
  if ((offset & ~0xffff) == 0xe0020000) {
    return g_mmap2_raw((void *)0x7f100000, length, PROT_NONE,
      MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_NORESERVE, -1, 0);
  }
  // upper mem
  if ((offset & 0xfe000000) != 0x02000000) {
    err("unexpected devmem mmap @ %08x\n", offset);
    return -EINVAL;
  }

  umem = uppermem_lookup(offset, &umem_end);
  if (umem + length > umem_end)
    err("warning: uppermem @ %08x overflows by %d bytes\n",
        offset, umem + length - umem_end);

  dbg("upper mem @ %08x %x = %p\n", offset, length, umem);
  return (long)umem;
}

long emu_do_mmap(unsigned int length, int prot, int flags, int fd,
  unsigned int offset)
{
  if (fd == FAKEDEV_MEM)
    return emu_mmap_dev(length, prot, flags, offset);

  if (fd == FAKEDEV_FB0)
    return emu_mmap_dev(length, prot, flags, offset + 0x03101000);

  if (fd == FAKEDEV_FB1)
    return emu_mmap_dev(length, prot, flags, offset + 0x03381000);

  err("bad/ni mmap(?, %d, %x, %x, %d, %08x)\n", length, prot, flags, fd, offset);
  return -EINVAL;
}

long emu_do_munmap(void *addr, unsigned int length)
{
  u8 *p = addr;

  // don't allow to unmap upper mem
  if ((u8 *)mmsp2.umem <= p && p < (u8 *)mmsp2.umem + 0x2000000) {
    dbg("ignoring munmap: %p %x\n", addr, length);
    return 0;
  }

  return -EAGAIN;
}

static void emu_sound_open(int fd)
{
#ifdef PND
  int ret, frag;

  // set default buffer size to 16 * 1K
  frag = (16<<16) | 10; // 16K
  ret = g_ioctl_raw(fd, SNDCTL_DSP_SETFRAGMENT, &frag);
  if (ret != 0)
    err("snd ioctl SETFRAGMENT %08x: %d\n", frag, ret);
#endif
}

static long emu_sound_ioctl(int fd, int request, void *argp)
{
  int *arg = argp;

#if 0
  dbg("snd ioctl(%d, %08x, %p)", fd, request, argp);
  if (arg != NULL)
    dbg_c(" [%d]", *arg);
  dbg_c("\n");
#endif

  /* People set strange frag settings on GP2X, which even manage
   * to break audio on pandora (causes writes to fail).
   * Catch this and set to something that works. */
  switch(request) {
    case SNDCTL_DSP_SETFRAGMENT: {
      int bsize, frag, frag_cnt;
      long ret;

      if (arg == NULL)
        break;

      frag = *arg & 0xffff;
      frag_cnt = *arg >> 16;
      bsize = frag_cnt << frag;
      if (frag < 10 || bsize < 4096*4 || bsize > 4096*4*2) {
        /*
         * ~4ms. gpSP wants small buffers or else it stutters
         * because of it's audio thread sync stuff
         * XXX: hardcoding, as low samplerates will result in small fragment size,
         * which itself causes ALSA stall and hangs the program.
         * Also some apps change samplerate without reopening /dev/dsp,
         * which causes ALSA to reject SNDCTL_DSP_SETFRAGMENT.
         */
        bsize = 44100 / 250 * 4;

        for (frag = 0; bsize; bsize >>= 1, frag++)
          ;

        frag_cnt = 16;
      }

      frag |= frag_cnt << 16;
      ret = g_ioctl_raw(fd, SNDCTL_DSP_SETFRAGMENT, &frag);
      if (ret != 0)
        err("snd ioctl SETFRAGMENT %08x: %ld\n", frag, ret);
      // indicate success even if we fail (because of ALSA mostly),
      // things like MikMod will bail out otherwise.
      return 0;
    }
    case SNDCTL_DSP_SYNC:
      // Franxis tends to use sync/write loops, bad idea under ALSA
      return 0;
    default:
      break;
  }

  return g_ioctl_raw(fd, request, argp);
}

long emu_do_ioctl(int fd, int request, void *argp)
{
  if (fd == emu_interesting_fds[IFD_SOUND].fd)
    return emu_sound_ioctl(fd, request, argp);

  switch (fd) {
  /* *********************** */
  case FAKEDEV_FB0:
  case FAKEDEV_FB1:
    if (argp == NULL)
      goto fail;

    switch (request) {
      case FBIOGET_FSCREENINFO: {
        struct fb_fix_screeninfo *fix = argp;

        memset(fix, 0, sizeof(*fix));
        strcpy(fix->id, "mmsp2_RGB0");
        fix->type         = FB_TYPE_PACKED_PIXELS;
        fix->accel        = FB_ACCEL_NONE;
        fix->visual       = FB_VISUAL_TRUECOLOR;
        fix->line_length  = 320*2;
        fix->smem_start   = (fd == FAKEDEV_FB0) ? 0x03101000 : 0x03381000;
        fix->smem_len     = 320*240*2;
        return 0;
      }
      case FBIOGET_VSCREENINFO: {
        struct fb_var_screeninfo *var = argp;
        static const struct fb_bitfield fbb_red   = { offset: 11, length: 5, };
        static const struct fb_bitfield fbb_green = { offset:  5, length: 6, };
        static const struct fb_bitfield fbb_blue  = { offset:  0, length: 5, };

        memset(var, 0, sizeof(*var));
        var->activate     = FB_ACTIVATE_NOW;
        var->xres         =
        var->xres_virtual = 320;
        var->yres         =
        var->yres_virtual = 240;
        var->width        =
        var->height       = -1;
        var->vmode        = FB_VMODE_NONINTERLACED;
        var->bits_per_pixel = 16;
        var->red          = fbb_red;
        var->green        = fbb_green;
        var->blue         = fbb_blue;
        return 0;
      }
      case FBIOPUT_VSCREENINFO: {
        struct fb_var_screeninfo *var = argp;
        dbg(" put vscreen: %dx%d@%d\n", var->xres, var->yres, var->bits_per_pixel);
        if (var->xres != 320 || var->yres != 240 || var->bits_per_pixel != 16)
          return -1;
        return 0;
      }
    }

  /* *********************** */
  case FAKEDEV_TTY0:
    // fake tty0 to make GPH SDL happy
    if (request == 0x4b46) // KDGKBENT
      return -1;
    return 0;
  }

fail:
  err("bad/ni ioctl(%d, %08x, %p)\n", fd, request, argp);
  return -EINVAL;
}

struct dev_fd_t emu_interesting_fds[] = {
  [IFD_SOUND] = { "/dev/dsp", -1, emu_sound_open },
  { NULL, 0, NULL },
};

static const struct {
  const char *from;
  const char *to;
} path_map[] = {
  { "/mnt/tmp", "./tmp" },
};

static const char *wrap_path(const char *path)
{
  char *buff, *p;
  size_t size;
  int i, len;

  // do only path mapping for now
  for (i = 0; i < ARRAY_SIZE(path_map); i++) {
    p = strstr(path, path_map[i].from);
    if (p != NULL) {
      size = strlen(path) + strlen(path_map[i].to) + 1;
      buff = malloc(size);
      if (buff == NULL)
        break;
      len = p - path;
      strncpy(buff, path, len);
      snprintf(buff + len, size - len, "%s%s", path_map[i].to,
        path + len + strlen(path_map[i].from));
      dbg("mapped path \"%s\" -> \"%s\"\n", path, buff);
      return buff;
    }
  }

  return path;
}

static void wrap_path_free(const char *w_path, const char *old_path)
{
  if (w_path != old_path)
    free((void *)w_path);
}

void *emu_do_fopen(const char *path, const char *mode)
{
  const char *w_path;
  FILE *ret;

  w_path = wrap_path(path);
  ret = fopen(w_path, mode);
  wrap_path_free(w_path, path);

  return ret;
}

// FIXME: threads..
int emu_do_system(const char *command)
{
  static char tmp_path[512];
  int need_ginge = 0;
  const char *p2;
  char *p;
  int ret;

  if (command == NULL)
    return -1;

  for (p2 = command; *p2 && isspace(*p2); p2++)
    ;

  if (*p2 == '.') // relative path?
    need_ginge = 1;
  else if (*p2 == '/' && strncmp(p2, "/bin", 4) && strncmp(p2, "/lib", 4)
           && strncmp(p2, "/sbin", 4) && strncmp(p2, "/usr", 4))
    // absolute path, but not a system command
    need_ginge = 1;

  p2 = wrap_path(command);
  if (need_ginge) {
    make_local_path(tmp_path, sizeof(tmp_path), "ginge_prep");
    p = tmp_path + strlen(tmp_path);

    snprintf(p, sizeof(tmp_path) - (p - tmp_path), " --nomenu %s", p2);
  }
  else
    snprintf(tmp_path, sizeof(tmp_path), "%s", p2);
  wrap_path_free(p2, command);

  dbg("system: \"%s\"\n", tmp_path);

  // the app might want the screen too..
  fb_thread_pause();
  ret = system(tmp_path);
  fb_thread_resume();
  return ret;
}

int emu_do_execve(const char *filename, char *const argv[], char *const envp[])
{
  const char **new_argv;
  char *prep_path;
  int i, ret, argc;

  if (filename == NULL)
    return -1;

  if (strstr(filename, "/gp2xmenu") != NULL)
    exit(0);

  for (i = 0; argv[i] != NULL; i++)
    ;
  argc = i + 1;

  new_argv = calloc(argc + 2, sizeof(new_argv[0]));
  if (new_argv == NULL)
    return -1;

  prep_path = malloc(512);
  if (prep_path == NULL)
    return -1;

  make_local_path(prep_path, 512, "ginge_prep");
  new_argv[0] = prep_path;
  new_argv[1] = "--nomenu";
  new_argv[2] = wrap_path(filename);

  if (argv[0] != NULL)
    for (i = 1; argv[i] != NULL; i++)
      new_argv[i + 2] = argv[i];

  dbg("execve \"%s\" %s \"%s\"\n", new_argv[0], new_argv[1], new_argv[2]);
  ret = execve(new_argv[0], (char **)new_argv, envp);
  perror("execve");
  return ret;
}

// vim:shiftwidth=2:expandtab