o * ppc.h (ppc_cpu_t): Typedef to uint64_t.

[deliverable/binutils-gdb.git] / gdb / arm-tdep.c

/* Common target dependent code for GDB on ARM systems.

   Copyright (C) 1988, 1989, 1991, 1992, 1993, 1995, 1996, 1998, 1999, 2000,
   2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009
   Free Software Foundation, Inc.

   This file is part of GDB.

   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 3 of the License, or
   (at your option) any later version.

   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program.  If not, see <http://www.gnu.org/licenses/>.  */

#include <ctype.h>              /* XXX for isupper () */

#include "defs.h"
#include "frame.h"
#include "inferior.h"
#include "gdbcmd.h"
#include "gdbcore.h"
#include "gdb_string.h"
#include "dis-asm.h"            /* For register styles. */
#include "regcache.h"
#include "doublest.h"
#include "value.h"
#include "arch-utils.h"
#include "osabi.h"
#include "frame-unwind.h"
#include "frame-base.h"
#include "trad-frame.h"
#include "objfiles.h"
#include "dwarf2-frame.h"
#include "gdbtypes.h"
#include "prologue-value.h"
#include "target-descriptions.h"
#include "user-regs.h"

#include "arm-tdep.h"
#include "gdb/sim-arm.h"

#include "elf-bfd.h"
#include "coff/internal.h"
#include "elf/arm.h"

#include "gdb_assert.h"
#include "vec.h"

static int arm_debug;

/* Macros for setting and testing a bit in a minimal symbol that marks
   it as Thumb function.  The MSB of the minimal symbol's "info" field
   is used for this purpose.

   MSYMBOL_SET_SPECIAL  Actually sets the "special" bit.
   MSYMBOL_IS_SPECIAL   Tests the "special" bit in a minimal symbol.  */

#define MSYMBOL_SET_SPECIAL(msym)                                       \
        MSYMBOL_TARGET_FLAG_1 (msym) = 1

#define MSYMBOL_IS_SPECIAL(msym)                                \
        MSYMBOL_TARGET_FLAG_1 (msym)

/* Per-objfile data used for mapping symbols.  */
static const struct objfile_data *arm_objfile_data_key;

struct arm_mapping_symbol
{
  bfd_vma value;
  char type;
};
typedef struct arm_mapping_symbol arm_mapping_symbol_s;
DEF_VEC_O(arm_mapping_symbol_s);

struct arm_per_objfile
{
  VEC(arm_mapping_symbol_s) **section_maps;
};

/* The list of available "set arm ..." and "show arm ..." commands.  */
static struct cmd_list_element *setarmcmdlist = NULL;
static struct cmd_list_element *showarmcmdlist = NULL;

/* The type of floating-point to use.  Keep this in sync with enum
   arm_float_model, and the help string in _initialize_arm_tdep.  */
static const char *fp_model_strings[] =
{
  "auto",
  "softfpa",
  "fpa",
  "softvfp",
  "vfp",
  NULL
};

/* A variable that can be configured by the user.  */
static enum arm_float_model arm_fp_model = ARM_FLOAT_AUTO;
static const char *current_fp_model = "auto";

/* The ABI to use.  Keep this in sync with arm_abi_kind.  */
static const char *arm_abi_strings[] =
{
  "auto",
  "APCS",
  "AAPCS",
  NULL
};

/* A variable that can be configured by the user.  */
static enum arm_abi_kind arm_abi_global = ARM_ABI_AUTO;
static const char *arm_abi_string = "auto";

/* The execution mode to assume.  */
static const char *arm_mode_strings[] =
  {
    "auto",
    "arm",
    "thumb"
  };

static const char *arm_fallback_mode_string = "auto";
static const char *arm_force_mode_string = "auto";

/* Number of different reg name sets (options).  */
static int num_disassembly_options;

/* The standard register names, and all the valid aliases for them.  */
static const struct
{
  const char *name;
  int regnum;
} arm_register_aliases[] = {
  /* Basic register numbers.  */
  { "r0", 0 },
  { "r1", 1 },
  { "r2", 2 },
  { "r3", 3 },
  { "r4", 4 },
  { "r5", 5 },
  { "r6", 6 },
  { "r7", 7 },
  { "r8", 8 },
  { "r9", 9 },
  { "r10", 10 },
  { "r11", 11 },
  { "r12", 12 },
  { "r13", 13 },
  { "r14", 14 },
  { "r15", 15 },
  /* Synonyms (argument and variable registers).  */
  { "a1", 0 },
  { "a2", 1 },
  { "a3", 2 },
  { "a4", 3 },
  { "v1", 4 },
  { "v2", 5 },
  { "v3", 6 },
  { "v4", 7 },
  { "v5", 8 },
  { "v6", 9 },
  { "v7", 10 },
  { "v8", 11 },
  /* Other platform-specific names for r9.  */
  { "sb", 9 },
  { "tr", 9 },
  /* Special names.  */
  { "ip", 12 },
  { "sp", 13 },
  { "lr", 14 },
  { "pc", 15 },
  /* Names used by GCC (not listed in the ARM EABI).  */
  { "sl", 10 },
  { "fp", 11 },
  /* A special name from the older ATPCS.  */
  { "wr", 7 },
};

static const char *const arm_register_names[] =
{"r0",  "r1",  "r2",  "r3",     /*  0  1  2  3 */
 "r4",  "r5",  "r6",  "r7",     /*  4  5  6  7 */
 "r8",  "r9",  "r10", "r11",    /*  8  9 10 11 */
 "r12", "sp",  "lr",  "pc",     /* 12 13 14 15 */
 "f0",  "f1",  "f2",  "f3",     /* 16 17 18 19 */
 "f4",  "f5",  "f6",  "f7",     /* 20 21 22 23 */
 "fps", "cpsr" };               /* 24 25       */

/* Valid register name styles.  */
static const char **valid_disassembly_styles;

/* Disassembly style to use. Default to "std" register names.  */
static const char *disassembly_style;

/* This is used to keep the bfd arch_info in sync with the disassembly
   style.  */
static void set_disassembly_style_sfunc(char *, int,
                                         struct cmd_list_element *);
static void set_disassembly_style (void);

static void convert_from_extended (const struct floatformat *, const void *,
                                   void *, int);
static void convert_to_extended (const struct floatformat *, void *,
                                 const void *, int);

static void arm_neon_quad_read (struct gdbarch *gdbarch,
                                struct regcache *regcache,
                                int regnum, gdb_byte *buf);
static void arm_neon_quad_write (struct gdbarch *gdbarch,
                                 struct regcache *regcache,
                                 int regnum, const gdb_byte *buf);

struct arm_prologue_cache
{
  /* The stack pointer at the time this frame was created; i.e. the
     caller's stack pointer when this function was called.  It is used
     to identify this frame.  */
  CORE_ADDR prev_sp;

  /* The frame base for this frame is just prev_sp - frame size.
     FRAMESIZE is the distance from the frame pointer to the
     initial stack pointer.  */

  int framesize;

  /* The register used to hold the frame pointer for this frame.  */
  int framereg;

  /* Saved register offsets.  */
  struct trad_frame_saved_reg *saved_regs;
};

/* Architecture version for displaced stepping.  This effects the behaviour of
   certain instructions, and really should not be hard-wired.  */

#define DISPLACED_STEPPING_ARCH_VERSION         5

/* Addresses for calling Thumb functions have the bit 0 set.
   Here are some macros to test, set, or clear bit 0 of addresses.  */
#define IS_THUMB_ADDR(addr)     ((addr) & 1)
#define MAKE_THUMB_ADDR(addr)   ((addr) | 1)
#define UNMAKE_THUMB_ADDR(addr) ((addr) & ~1)

/* Set to true if the 32-bit mode is in use.  */

int arm_apcs_32 = 1;

/* Determine if FRAME is executing in Thumb mode.  */

static int
arm_frame_is_thumb (struct frame_info *frame)
{
  CORE_ADDR cpsr;

  /* Every ARM frame unwinder can unwind the T bit of the CPSR, either
     directly (from a signal frame or dummy frame) or by interpreting
     the saved LR (from a prologue or DWARF frame).  So consult it and
     trust the unwinders.  */
  cpsr = get_frame_register_unsigned (frame, ARM_PS_REGNUM);

  return (cpsr & CPSR_T) != 0;
}

/* Callback for VEC_lower_bound.  */

static inline int
arm_compare_mapping_symbols (const struct arm_mapping_symbol *lhs,
                             const struct arm_mapping_symbol *rhs)
{
  return lhs->value < rhs->value;
}

/* Determine if the program counter specified in MEMADDR is in a Thumb
   function.  This function should be called for addresses unrelated to
   any executing frame; otherwise, prefer arm_frame_is_thumb.  */

static int
arm_pc_is_thumb (CORE_ADDR memaddr)
{
  struct obj_section *sec;
  struct minimal_symbol *sym;

  /* If bit 0 of the address is set, assume this is a Thumb address.  */
  if (IS_THUMB_ADDR (memaddr))
    return 1;

  /* If the user wants to override the symbol table, let him.  */
  if (strcmp (arm_force_mode_string, "arm") == 0)
    return 0;
  if (strcmp (arm_force_mode_string, "thumb") == 0)
    return 1;

  /* If there are mapping symbols, consult them.  */
  sec = find_pc_section (memaddr);
  if (sec != NULL)
    {
      struct arm_per_objfile *data;
      VEC(arm_mapping_symbol_s) *map;
      struct arm_mapping_symbol map_key = { memaddr - obj_section_addr (sec),
                                            0 };
      unsigned int idx;

      data = objfile_data (sec->objfile, arm_objfile_data_key);
      if (data != NULL)
        {
          map = data->section_maps[sec->the_bfd_section->index];
          if (!VEC_empty (arm_mapping_symbol_s, map))
            {
              struct arm_mapping_symbol *map_sym;

              idx = VEC_lower_bound (arm_mapping_symbol_s, map, &map_key,
                                     arm_compare_mapping_symbols);

              /* VEC_lower_bound finds the earliest ordered insertion
                 point.  If the following symbol starts at this exact
                 address, we use that; otherwise, the preceding
                 mapping symbol covers this address.  */
              if (idx < VEC_length (arm_mapping_symbol_s, map))
                {
                  map_sym = VEC_index (arm_mapping_symbol_s, map, idx);
                  if (map_sym->value == map_key.value)
                    return map_sym->type == 't';
                }

              if (idx > 0)
                {
                  map_sym = VEC_index (arm_mapping_symbol_s, map, idx - 1);
                  return map_sym->type == 't';
                }
            }
        }
    }

  /* Thumb functions have a "special" bit set in minimal symbols.  */
  sym = lookup_minimal_symbol_by_pc (memaddr);
  if (sym)
    return (MSYMBOL_IS_SPECIAL (sym));

  /* If the user wants to override the fallback mode, let them.  */
  if (strcmp (arm_fallback_mode_string, "arm") == 0)
    return 0;
  if (strcmp (arm_fallback_mode_string, "thumb") == 0)
    return 1;

  /* If we couldn't find any symbol, but we're talking to a running
     target, then trust the current value of $cpsr.  This lets
     "display/i $pc" always show the correct mode (though if there is
     a symbol table we will not reach here, so it still may not be
     displayed in the mode it will be executed).  */
  if (target_has_registers)
    return arm_frame_is_thumb (get_current_frame ());

  /* Otherwise we're out of luck; we assume ARM.  */
  return 0;
}

/* Remove useless bits from addresses in a running program.  */
static CORE_ADDR
arm_addr_bits_remove (struct gdbarch *gdbarch, CORE_ADDR val)
{
  if (arm_apcs_32)
    return UNMAKE_THUMB_ADDR (val);
  else
    return (val & 0x03fffffc);
}

/* When reading symbols, we need to zap the low bit of the address,
   which may be set to 1 for Thumb functions.  */
static CORE_ADDR
arm_smash_text_address (struct gdbarch *gdbarch, CORE_ADDR val)
{
  return val & ~1;
}

/* Analyze a Thumb prologue, looking for a recognizable stack frame
   and frame pointer.  Scan until we encounter a store that could
   clobber the stack frame unexpectedly, or an unknown instruction.  */

static CORE_ADDR
thumb_analyze_prologue (struct gdbarch *gdbarch,
                        CORE_ADDR start, CORE_ADDR limit,
                        struct arm_prologue_cache *cache)
{
  enum bfd_endian byte_order_for_code = gdbarch_byte_order_for_code (gdbarch);
  int i;
  pv_t regs[16];
  struct pv_area *stack;
  struct cleanup *back_to;
  CORE_ADDR offset;

  for (i = 0; i < 16; i++)
    regs[i] = pv_register (i, 0);
  stack = make_pv_area (ARM_SP_REGNUM, gdbarch_addr_bit (gdbarch));
  back_to = make_cleanup_free_pv_area (stack);

  while (start < limit)
    {
      unsigned short insn;

      insn = read_memory_unsigned_integer (start, 2, byte_order_for_code);

      if ((insn & 0xfe00) == 0xb400)            /* push { rlist } */
        {
          int regno;
          int mask;

          if (pv_area_store_would_trash (stack, regs[ARM_SP_REGNUM]))
            break;

          /* Bits 0-7 contain a mask for registers R0-R7.  Bit 8 says
             whether to save LR (R14).  */
          mask = (insn & 0xff) | ((insn & 0x100) << 6);

          /* Calculate offsets of saved R0-R7 and LR.  */
          for (regno = ARM_LR_REGNUM; regno >= 0; regno--)
            if (mask & (1 << regno))
              {
                regs[ARM_SP_REGNUM] = pv_add_constant (regs[ARM_SP_REGNUM],
                                                       -4);
                pv_area_store (stack, regs[ARM_SP_REGNUM], 4, regs[regno]);
              }
        }
      else if ((insn & 0xff00) == 0xb000)       /* add sp, #simm  OR  
                                                   sub sp, #simm */
        {
          offset = (insn & 0x7f) << 2;          /* get scaled offset */
          if (insn & 0x80)                      /* Check for SUB.  */
            regs[ARM_SP_REGNUM] = pv_add_constant (regs[ARM_SP_REGNUM],
                                                   -offset);
          else
            regs[ARM_SP_REGNUM] = pv_add_constant (regs[ARM_SP_REGNUM],
                                                   offset);
        }
      else if ((insn & 0xff00) == 0xaf00)       /* add r7, sp, #imm */
        regs[THUMB_FP_REGNUM] = pv_add_constant (regs[ARM_SP_REGNUM],
                                                 (insn & 0xff) << 2);
      else if ((insn & 0xff00) == 0x4600)       /* mov hi, lo or mov lo, hi */
        {
          int dst_reg = (insn & 0x7) + ((insn & 0x80) >> 4);
          int src_reg = (insn & 0x78) >> 3;
          regs[dst_reg] = regs[src_reg];
        }
      else if ((insn & 0xf800) == 0x9000)       /* str rd, [sp, #off] */
        {
          /* Handle stores to the stack.  Normally pushes are used,
             but with GCC -mtpcs-frame, there may be other stores
             in the prologue to create the frame.  */
          int regno = (insn >> 8) & 0x7;
          pv_t addr;

          offset = (insn & 0xff) << 2;
          addr = pv_add_constant (regs[ARM_SP_REGNUM], offset);

          if (pv_area_store_would_trash (stack, addr))
            break;

          pv_area_store (stack, addr, 4, regs[regno]);
        }
      else
        {
          /* We don't know what this instruction is.  We're finished
             scanning.  NOTE: Recognizing more safe-to-ignore
             instructions here will improve support for optimized
             code.  */
          break;
        }

      start += 2;
    }

  if (cache == NULL)
    {
      do_cleanups (back_to);
      return start;
    }

  if (pv_is_register (regs[ARM_FP_REGNUM], ARM_SP_REGNUM))
    {
      /* Frame pointer is fp.  Frame size is constant.  */
      cache->framereg = ARM_FP_REGNUM;
      cache->framesize = -regs[ARM_FP_REGNUM].k;
    }
  else if (pv_is_register (regs[THUMB_FP_REGNUM], ARM_SP_REGNUM))
    {
      /* Frame pointer is r7.  Frame size is constant.  */
      cache->framereg = THUMB_FP_REGNUM;
      cache->framesize = -regs[THUMB_FP_REGNUM].k;
    }
  else if (pv_is_register (regs[ARM_SP_REGNUM], ARM_SP_REGNUM))
    {
      /* Try the stack pointer... this is a bit desperate.  */
      cache->framereg = ARM_SP_REGNUM;
      cache->framesize = -regs[ARM_SP_REGNUM].k;
    }
  else
    {
      /* We're just out of luck.  We don't know where the frame is.  */
      cache->framereg = -1;
      cache->framesize = 0;
    }

  for (i = 0; i < 16; i++)
    if (pv_area_find_reg (stack, gdbarch, i, &offset))
      cache->saved_regs[i].addr = offset;

  do_cleanups (back_to);
  return start;
}

/* Advance the PC across any function entry prologue instructions to
   reach some "real" code.

   The APCS (ARM Procedure Call Standard) defines the following
   prologue:

   mov          ip, sp
   [stmfd       sp!, {a1,a2,a3,a4}]
   stmfd        sp!, {...,fp,ip,lr,pc}
   [stfe        f7, [sp, #-12]!]
   [stfe        f6, [sp, #-12]!]
   [stfe        f5, [sp, #-12]!]
   [stfe        f4, [sp, #-12]!]
   sub fp, ip, #nn @@ nn == 20 or 4 depending on second insn */

static CORE_ADDR
arm_skip_prologue (struct gdbarch *gdbarch, CORE_ADDR pc)
{
  enum bfd_endian byte_order_for_code = gdbarch_byte_order_for_code (gdbarch);
  unsigned long inst;
  CORE_ADDR skip_pc;
  CORE_ADDR func_addr, limit_pc;
  struct symtab_and_line sal;

  /* If we're in a dummy frame, don't even try to skip the prologue.  */
  if (deprecated_pc_in_call_dummy (gdbarch, pc))
    return pc;

  /* See if we can determine the end of the prologue via the symbol table.
     If so, then return either PC, or the PC after the prologue, whichever
     is greater.  */
  if (find_pc_partial_function (pc, NULL, &func_addr, NULL))
    {
      CORE_ADDR post_prologue_pc
        = skip_prologue_using_sal (gdbarch, func_addr);
      if (post_prologue_pc != 0)
        return max (pc, post_prologue_pc);
    }

  /* Can't determine prologue from the symbol table, need to examine
     instructions.  */

  /* Find an upper limit on the function prologue using the debug
     information.  If the debug information could not be used to provide
     that bound, then use an arbitrary large number as the upper bound.  */
  /* Like arm_scan_prologue, stop no later than pc + 64. */
  limit_pc = skip_prologue_using_sal (gdbarch, pc);
  if (limit_pc == 0)
    limit_pc = pc + 64;          /* Magic.  */


  /* Check if this is Thumb code.  */
  if (arm_pc_is_thumb (pc))
    return thumb_analyze_prologue (gdbarch, pc, limit_pc, NULL);

  for (skip_pc = pc; skip_pc < limit_pc; skip_pc += 4)
    {
      inst = read_memory_unsigned_integer (skip_pc, 4, byte_order_for_code);

      /* "mov ip, sp" is no longer a required part of the prologue.  */
      if (inst == 0xe1a0c00d)                   /* mov ip, sp */
        continue;

      if ((inst & 0xfffff000) == 0xe28dc000)    /* add ip, sp #n */
        continue;

      if ((inst & 0xfffff000) == 0xe24dc000)    /* sub ip, sp #n */
        continue;

      /* Some prologues begin with "str lr, [sp, #-4]!".  */
      if (inst == 0xe52de004)                   /* str lr, [sp, #-4]! */
        continue;

      if ((inst & 0xfffffff0) == 0xe92d0000)    /* stmfd sp!,{a1,a2,a3,a4} */
        continue;

      if ((inst & 0xfffff800) == 0xe92dd800)    /* stmfd sp!,{fp,ip,lr,pc} */
        continue;

      /* Any insns after this point may float into the code, if it makes
         for better instruction scheduling, so we skip them only if we
         find them, but still consider the function to be frame-ful.  */

      /* We may have either one sfmfd instruction here, or several stfe
         insns, depending on the version of floating point code we
         support.  */
      if ((inst & 0xffbf0fff) == 0xec2d0200)    /* sfmfd fn, <cnt>, [sp]! */
        continue;

      if ((inst & 0xffff8fff) == 0xed6d0103)    /* stfe fn, [sp, #-12]! */
        continue;

      if ((inst & 0xfffff000) == 0xe24cb000)    /* sub fp, ip, #nn */
        continue;

      if ((inst & 0xfffff000) == 0xe24dd000)    /* sub sp, sp, #nn */
        continue;

      if ((inst & 0xffffc000) == 0xe54b0000 ||  /* strb r(0123),[r11,#-nn] */
          (inst & 0xffffc0f0) == 0xe14b00b0 ||  /* strh r(0123),[r11,#-nn] */
          (inst & 0xffffc000) == 0xe50b0000)    /* str  r(0123),[r11,#-nn] */
        continue;

      if ((inst & 0xffffc000) == 0xe5cd0000 ||  /* strb r(0123),[sp,#nn] */
          (inst & 0xffffc0f0) == 0xe1cd00b0 ||  /* strh r(0123),[sp,#nn] */
          (inst & 0xffffc000) == 0xe58d0000)    /* str  r(0123),[sp,#nn] */
        continue;

      /* Un-recognized instruction; stop scanning.  */
      break;
    }

  return skip_pc;               /* End of prologue */
}

/* *INDENT-OFF* */
/* Function: thumb_scan_prologue (helper function for arm_scan_prologue)
   This function decodes a Thumb function prologue to determine:
     1) the size of the stack frame
     2) which registers are saved on it
     3) the offsets of saved regs
     4) the offset from the stack pointer to the frame pointer

   A typical Thumb function prologue would create this stack frame
   (offsets relative to FP)
     old SP ->  24  stack parameters
                20  LR
                16  R7
     R7 ->       0  local variables (16 bytes)
     SP ->     -12  additional stack space (12 bytes)
   The frame size would thus be 36 bytes, and the frame offset would be
   12 bytes.  The frame register is R7. 
   
   The comments for thumb_skip_prolog() describe the algorithm we use
   to detect the end of the prolog.  */
/* *INDENT-ON* */

static void
thumb_scan_prologue (struct gdbarch *gdbarch, CORE_ADDR prev_pc,
                     CORE_ADDR block_addr, struct arm_prologue_cache *cache)
{
  CORE_ADDR prologue_start;
  CORE_ADDR prologue_end;
  CORE_ADDR current_pc;

  if (find_pc_partial_function (block_addr, NULL, &prologue_start,
                                &prologue_end))
    {
      struct symtab_and_line sal = find_pc_line (prologue_start, 0);

      if (sal.line == 0)                /* no line info, use current PC  */
        prologue_end = prev_pc;
      else if (sal.end < prologue_end)  /* next line begins after fn end */
        prologue_end = sal.end;         /* (probably means no prologue)  */
    }
  else
    /* We're in the boondocks: we have no idea where the start of the
       function is.  */
    return;

  prologue_end = min (prologue_end, prev_pc);

  thumb_analyze_prologue (gdbarch, prologue_start, prologue_end, cache);
}

/* This function decodes an ARM function prologue to determine:
   1) the size of the stack frame
   2) which registers are saved on it
   3) the offsets of saved regs
   4) the offset from the stack pointer to the frame pointer
   This information is stored in the "extra" fields of the frame_info.

   There are two basic forms for the ARM prologue.  The fixed argument
   function call will look like:

   mov    ip, sp
   stmfd  sp!, {fp, ip, lr, pc}
   sub    fp, ip, #4
   [sub sp, sp, #4]

   Which would create this stack frame (offsets relative to FP):
   IP ->   4    (caller's stack)
   FP ->   0    PC (points to address of stmfd instruction + 8 in callee)
   -4   LR (return address in caller)
   -8   IP (copy of caller's SP)
   -12  FP (caller's FP)
   SP -> -28    Local variables

   The frame size would thus be 32 bytes, and the frame offset would be
   28 bytes.  The stmfd call can also save any of the vN registers it
   plans to use, which increases the frame size accordingly.

   Note: The stored PC is 8 off of the STMFD instruction that stored it
   because the ARM Store instructions always store PC + 8 when you read
   the PC register.

   A variable argument function call will look like:

   mov    ip, sp
   stmfd  sp!, {a1, a2, a3, a4}
   stmfd  sp!, {fp, ip, lr, pc}
   sub    fp, ip, #20

   Which would create this stack frame (offsets relative to FP):
   IP ->  20    (caller's stack)
   16  A4
   12  A3
   8  A2
   4  A1
   FP ->   0    PC (points to address of stmfd instruction + 8 in callee)
   -4   LR (return address in caller)
   -8   IP (copy of caller's SP)
   -12  FP (caller's FP)
   SP -> -28    Local variables

   The frame size would thus be 48 bytes, and the frame offset would be
   28 bytes.

   There is another potential complication, which is that the optimizer
   will try to separate the store of fp in the "stmfd" instruction from
   the "sub fp, ip, #NN" instruction.  Almost anything can be there, so
   we just key on the stmfd, and then scan for the "sub fp, ip, #NN"...

   Also, note, the original version of the ARM toolchain claimed that there
   should be an

   instruction at the end of the prologue.  I have never seen GCC produce
   this, and the ARM docs don't mention it.  We still test for it below in
   case it happens...

 */

static void
arm_scan_prologue (struct frame_info *this_frame,
                   struct arm_prologue_cache *cache)
{
  struct gdbarch *gdbarch = get_frame_arch (this_frame);
  enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
  enum bfd_endian byte_order_for_code = gdbarch_byte_order_for_code (gdbarch);
  int regno;
  CORE_ADDR prologue_start, prologue_end, current_pc;
  CORE_ADDR prev_pc = get_frame_pc (this_frame);
  CORE_ADDR block_addr = get_frame_address_in_block (this_frame);
  pv_t regs[ARM_FPS_REGNUM];
  struct pv_area *stack;
  struct cleanup *back_to;
  CORE_ADDR offset;

  /* Assume there is no frame until proven otherwise.  */
  cache->framereg = ARM_SP_REGNUM;
  cache->framesize = 0;

  /* Check for Thumb prologue.  */
  if (arm_frame_is_thumb (this_frame))
    {
      thumb_scan_prologue (gdbarch, prev_pc, block_addr, cache);
      return;
    }

  /* Find the function prologue.  If we can't find the function in
     the symbol table, peek in the stack frame to find the PC.  */
  if (find_pc_partial_function (block_addr, NULL, &prologue_start,
                                &prologue_end))
    {
      /* One way to find the end of the prologue (which works well
         for unoptimized code) is to do the following:

            struct symtab_and_line sal = find_pc_line (prologue_start, 0);

            if (sal.line == 0)
              prologue_end = prev_pc;
            else if (sal.end < prologue_end)
              prologue_end = sal.end;

         This mechanism is very accurate so long as the optimizer
         doesn't move any instructions from the function body into the
         prologue.  If this happens, sal.end will be the last
         instruction in the first hunk of prologue code just before
         the first instruction that the scheduler has moved from
         the body to the prologue.

         In order to make sure that we scan all of the prologue
         instructions, we use a slightly less accurate mechanism which
         may scan more than necessary.  To help compensate for this
         lack of accuracy, the prologue scanning loop below contains
         several clauses which'll cause the loop to terminate early if
         an implausible prologue instruction is encountered.  
         
         The expression
         
              prologue_start + 64
            
         is a suitable endpoint since it accounts for the largest
         possible prologue plus up to five instructions inserted by
         the scheduler.  */
         
      if (prologue_end > prologue_start + 64)
        {
          prologue_end = prologue_start + 64;   /* See above.  */
        }
    }
  else
    {
      /* We have no symbol information.  Our only option is to assume this
         function has a standard stack frame and the normal frame register.
         Then, we can find the value of our frame pointer on entrance to
         the callee (or at the present moment if this is the innermost frame).
         The value stored there should be the address of the stmfd + 8.  */
      CORE_ADDR frame_loc;
      LONGEST return_value;

      frame_loc = get_frame_register_unsigned (this_frame, ARM_FP_REGNUM);
      if (!safe_read_memory_integer (frame_loc, 4, byte_order, &return_value))
        return;
      else
        {
          prologue_start = gdbarch_addr_bits_remove 
                             (gdbarch, return_value) - 8;
          prologue_end = prologue_start + 64;   /* See above.  */
        }
    }

  if (prev_pc < prologue_end)
    prologue_end = prev_pc;

  /* Now search the prologue looking for instructions that set up the
     frame pointer, adjust the stack pointer, and save registers.

     Be careful, however, and if it doesn't look like a prologue,
     don't try to scan it.  If, for instance, a frameless function
     begins with stmfd sp!, then we will tell ourselves there is
     a frame, which will confuse stack traceback, as well as "finish" 
     and other operations that rely on a knowledge of the stack
     traceback.

     In the APCS, the prologue should start with  "mov ip, sp" so
     if we don't see this as the first insn, we will stop.  

     [Note: This doesn't seem to be true any longer, so it's now an
     optional part of the prologue.  - Kevin Buettner, 2001-11-20]

     [Note further: The "mov ip,sp" only seems to be missing in
     frameless functions at optimization level "-O2" or above,
     in which case it is often (but not always) replaced by
     "str lr, [sp, #-4]!".  - Michael Snyder, 2002-04-23]  */

  for (regno = 0; regno < ARM_FPS_REGNUM; regno++)
    regs[regno] = pv_register (regno, 0);
  stack = make_pv_area (ARM_SP_REGNUM, gdbarch_addr_bit (gdbarch));
  back_to = make_cleanup_free_pv_area (stack);

  for (current_pc = prologue_start;
       current_pc < prologue_end;
       current_pc += 4)
    {
      unsigned int insn
        = read_memory_unsigned_integer (current_pc, 4, byte_order_for_code);

      if (insn == 0xe1a0c00d)           /* mov ip, sp */
        {
          regs[ARM_IP_REGNUM] = regs[ARM_SP_REGNUM];
          continue;
        }
      else if ((insn & 0xfffff000) == 0xe28dc000) /* add ip, sp #n */
        {
          unsigned imm = insn & 0xff;                   /* immediate value */
          unsigned rot = (insn & 0xf00) >> 7;           /* rotate amount */
          imm = (imm >> rot) | (imm << (32 - rot));
          regs[ARM_IP_REGNUM] = pv_add_constant (regs[ARM_SP_REGNUM], imm);
          continue;
        }
      else if ((insn & 0xfffff000) == 0xe24dc000) /* sub ip, sp #n */
        {
          unsigned imm = insn & 0xff;                   /* immediate value */
          unsigned rot = (insn & 0xf00) >> 7;           /* rotate amount */
          imm = (imm >> rot) | (imm << (32 - rot));
          regs[ARM_IP_REGNUM] = pv_add_constant (regs[ARM_SP_REGNUM], -imm);
          continue;
        }
      else if (insn == 0xe52de004)      /* str lr, [sp, #-4]! */
        {
          if (pv_area_store_would_trash (stack, regs[ARM_SP_REGNUM]))
            break;
          regs[ARM_SP_REGNUM] = pv_add_constant (regs[ARM_SP_REGNUM], -4);
          pv_area_store (stack, regs[ARM_SP_REGNUM], 4, regs[ARM_LR_REGNUM]);
          continue;
        }
      else if ((insn & 0xffff0000) == 0xe92d0000)
        /* stmfd sp!, {..., fp, ip, lr, pc}
           or
           stmfd sp!, {a1, a2, a3, a4}  */
        {
          int mask = insn & 0xffff;

          if (pv_area_store_would_trash (stack, regs[ARM_SP_REGNUM]))
            break;

          /* Calculate offsets of saved registers.  */
          for (regno = ARM_PC_REGNUM; regno >= 0; regno--)
            if (mask & (1 << regno))
              {
                regs[ARM_SP_REGNUM] = pv_add_constant (regs[ARM_SP_REGNUM], -4);
                pv_area_store (stack, regs[ARM_SP_REGNUM], 4, regs[regno]);
              }
        }
      else if ((insn & 0xffffc000) == 0xe54b0000 ||     /* strb rx,[r11,#-n] */
               (insn & 0xffffc0f0) == 0xe14b00b0 ||     /* strh rx,[r11,#-n] */
               (insn & 0xffffc000) == 0xe50b0000)       /* str  rx,[r11,#-n] */
        {
          /* No need to add this to saved_regs -- it's just an arg reg.  */
          continue;
        }
      else if ((insn & 0xffffc000) == 0xe5cd0000 ||     /* strb rx,[sp,#n] */
               (insn & 0xffffc0f0) == 0xe1cd00b0 ||     /* strh rx,[sp,#n] */
               (insn & 0xffffc000) == 0xe58d0000)       /* str  rx,[sp,#n] */
        {
          /* No need to add this to saved_regs -- it's just an arg reg.  */
          continue;
        }
      else if ((insn & 0xfffff000) == 0xe24cb000)       /* sub fp, ip #n */
        {
          unsigned imm = insn & 0xff;                   /* immediate value */
          unsigned rot = (insn & 0xf00) >> 7;           /* rotate amount */
          imm = (imm >> rot) | (imm << (32 - rot));
          regs[ARM_FP_REGNUM] = pv_add_constant (regs[ARM_IP_REGNUM], -imm);
        }
      else if ((insn & 0xfffff000) == 0xe24dd000)       /* sub sp, sp #n */
        {
          unsigned imm = insn & 0xff;                   /* immediate value */
          unsigned rot = (insn & 0xf00) >> 7;           /* rotate amount */
          imm = (imm >> rot) | (imm << (32 - rot));
          regs[ARM_SP_REGNUM] = pv_add_constant (regs[ARM_SP_REGNUM], -imm);
        }
      else if ((insn & 0xffff7fff) == 0xed6d0103        /* stfe f?, [sp, -#c]! */
               && gdbarch_tdep (gdbarch)->have_fpa_registers)
        {
          if (pv_area_store_would_trash (stack, regs[ARM_SP_REGNUM]))
            break;

          regs[ARM_SP_REGNUM] = pv_add_constant (regs[ARM_SP_REGNUM], -12);
          regno = ARM_F0_REGNUM + ((insn >> 12) & 0x07);
          pv_area_store (stack, regs[ARM_SP_REGNUM], 12, regs[regno]);
        }
      else if ((insn & 0xffbf0fff) == 0xec2d0200        /* sfmfd f0, 4, [sp!] */
               && gdbarch_tdep (gdbarch)->have_fpa_registers)
        {
          int n_saved_fp_regs;
          unsigned int fp_start_reg, fp_bound_reg;

          if (pv_area_store_would_trash (stack, regs[ARM_SP_REGNUM]))
            break;

          if ((insn & 0x800) == 0x800)          /* N0 is set */
            {
              if ((insn & 0x40000) == 0x40000)  /* N1 is set */
                n_saved_fp_regs = 3;
              else
                n_saved_fp_regs = 1;
            }
          else
            {
              if ((insn & 0x40000) == 0x40000)  /* N1 is set */
                n_saved_fp_regs = 2;
              else
                n_saved_fp_regs = 4;
            }

          fp_start_reg = ARM_F0_REGNUM + ((insn >> 12) & 0x7);
          fp_bound_reg = fp_start_reg + n_saved_fp_regs;
          for (; fp_start_reg < fp_bound_reg; fp_start_reg++)
            {
              regs[ARM_SP_REGNUM] = pv_add_constant (regs[ARM_SP_REGNUM], -12);
              pv_area_store (stack, regs[ARM_SP_REGNUM], 12,
                             regs[fp_start_reg++]);
            }
        }
      else if ((insn & 0xf0000000) != 0xe0000000)
        break;                  /* Condition not true, exit early */
      else if ((insn & 0xfe200000) == 0xe8200000)       /* ldm? */
        break;                  /* Don't scan past a block load */
      else
        /* The optimizer might shove anything into the prologue,
           so we just skip what we don't recognize.  */
        continue;
    }

  /* The frame size is just the distance from the frame register
     to the original stack pointer.  */
  if (pv_is_register (regs[ARM_FP_REGNUM], ARM_SP_REGNUM))
    {
      /* Frame pointer is fp.  */
      cache->framereg = ARM_FP_REGNUM;
      cache->framesize = -regs[ARM_FP_REGNUM].k;
    }
  else if (pv_is_register (regs[ARM_SP_REGNUM], ARM_SP_REGNUM))
    {
      /* Try the stack pointer... this is a bit desperate.  */
      cache->framereg = ARM_SP_REGNUM;
      cache->framesize = -regs[ARM_SP_REGNUM].k;
    }
  else
    {
      /* We're just out of luck.  We don't know where the frame is.  */
      cache->framereg = -1;
      cache->framesize = 0;
    }

  for (regno = 0; regno < ARM_FPS_REGNUM; regno++)
    if (pv_area_find_reg (stack, gdbarch, regno, &offset))
      cache->saved_regs[regno].addr = offset;

  do_cleanups (back_to);
}

static struct arm_prologue_cache *
arm_make_prologue_cache (struct frame_info *this_frame)
{
  int reg;
  struct arm_prologue_cache *cache;
  CORE_ADDR unwound_fp;

  cache = FRAME_OBSTACK_ZALLOC (struct arm_prologue_cache);
  cache->saved_regs = trad_frame_alloc_saved_regs (this_frame);

  arm_scan_prologue (this_frame, cache);

  unwound_fp = get_frame_register_unsigned (this_frame, cache->framereg);
  if (unwound_fp == 0)
    return cache;

  cache->prev_sp = unwound_fp + cache->framesize;

  /* Calculate actual addresses of saved registers using offsets
     determined by arm_scan_prologue.  */
  for (reg = 0; reg < gdbarch_num_regs (get_frame_arch (this_frame)); reg++)
    if (trad_frame_addr_p (cache->saved_regs, reg))
      cache->saved_regs[reg].addr += cache->prev_sp;

  return cache;
}

/* Our frame ID for a normal frame is the current function's starting PC
   and the caller's SP when we were called.  */

static void
arm_prologue_this_id (struct frame_info *this_frame,
                      void **this_cache,
                      struct frame_id *this_id)
{
  struct arm_prologue_cache *cache;
  struct frame_id id;
  CORE_ADDR pc, func;

  if (*this_cache == NULL)
    *this_cache = arm_make_prologue_cache (this_frame);
  cache = *this_cache;

  /* This is meant to halt the backtrace at "_start".  */
  pc = get_frame_pc (this_frame);
  if (pc <= gdbarch_tdep (get_frame_arch (this_frame))->lowest_pc)
    return;

  /* If we've hit a wall, stop.  */
  if (cache->prev_sp == 0)
    return;

  func = get_frame_func (this_frame);
  id = frame_id_build (cache->prev_sp, func);
  *this_id = id;
}

static struct value *
arm_prologue_prev_register (struct frame_info *this_frame,
                            void **this_cache,
                            int prev_regnum)
{
  struct gdbarch *gdbarch = get_frame_arch (this_frame);
  struct arm_prologue_cache *cache;

  if (*this_cache == NULL)
    *this_cache = arm_make_prologue_cache (this_frame);
  cache = *this_cache;

  /* If we are asked to unwind the PC, then we need to return the LR
     instead.  The prologue may save PC, but it will point into this
     frame's prologue, not the next frame's resume location.  Also
     strip the saved T bit.  A valid LR may have the low bit set, but
     a valid PC never does.  */
  if (prev_regnum == ARM_PC_REGNUM)
    {
      CORE_ADDR lr;

      lr = frame_unwind_register_unsigned (this_frame, ARM_LR_REGNUM);
      return frame_unwind_got_constant (this_frame, prev_regnum,
                                        arm_addr_bits_remove (gdbarch, lr));
    }

  /* SP is generally not saved to the stack, but this frame is
     identified by the next frame's stack pointer at the time of the call.
     The value was already reconstructed into PREV_SP.  */
  if (prev_regnum == ARM_SP_REGNUM)
    return frame_unwind_got_constant (this_frame, prev_regnum, cache->prev_sp);

  /* The CPSR may have been changed by the call instruction and by the
     called function.  The only bit we can reconstruct is the T bit,
     by checking the low bit of LR as of the call.  This is a reliable
     indicator of Thumb-ness except for some ARM v4T pre-interworking
     Thumb code, which could get away with a clear low bit as long as
     the called function did not use bx.  Guess that all other
     bits are unchanged; the condition flags are presumably lost,
     but the processor status is likely valid.  */
  if (prev_regnum == ARM_PS_REGNUM)
    {
      CORE_ADDR lr, cpsr;

      cpsr = get_frame_register_unsigned (this_frame, prev_regnum);
      lr = frame_unwind_register_unsigned (this_frame, ARM_LR_REGNUM);
      if (IS_THUMB_ADDR (lr))
        cpsr |= CPSR_T;
      else
        cpsr &= ~CPSR_T;
      return frame_unwind_got_constant (this_frame, prev_regnum, cpsr);
    }

  return trad_frame_get_prev_register (this_frame, cache->saved_regs,
                                       prev_regnum);
}

struct frame_unwind arm_prologue_unwind = {
  NORMAL_FRAME,
  arm_prologue_this_id,
  arm_prologue_prev_register,
  NULL,
  default_frame_sniffer
};

static struct arm_prologue_cache *
arm_make_stub_cache (struct frame_info *this_frame)
{
  int reg;
  struct arm_prologue_cache *cache;
  CORE_ADDR unwound_fp;

  cache = FRAME_OBSTACK_ZALLOC (struct arm_prologue_cache);
  cache->saved_regs = trad_frame_alloc_saved_regs (this_frame);

  cache->prev_sp = get_frame_register_unsigned (this_frame, ARM_SP_REGNUM);

  return cache;
}

/* Our frame ID for a stub frame is the current SP and LR.  */

static void
arm_stub_this_id (struct frame_info *this_frame,
                  void **this_cache,
                  struct frame_id *this_id)
{
  struct arm_prologue_cache *cache;

  if (*this_cache == NULL)
    *this_cache = arm_make_stub_cache (this_frame);
  cache = *this_cache;

  *this_id = frame_id_build (cache->prev_sp, get_frame_pc (this_frame));
}

static int
arm_stub_unwind_sniffer (const struct frame_unwind *self,
                         struct frame_info *this_frame,
                         void **this_prologue_cache)
{
  CORE_ADDR addr_in_block;
  char dummy[4];

  addr_in_block = get_frame_address_in_block (this_frame);
  if (in_plt_section (addr_in_block, NULL)
      || target_read_memory (get_frame_pc (this_frame), dummy, 4) != 0)
    return 1;

  return 0;
}

struct frame_unwind arm_stub_unwind = {
  NORMAL_FRAME,
  arm_stub_this_id,
  arm_prologue_prev_register,
  NULL,
  arm_stub_unwind_sniffer
};

static CORE_ADDR
arm_normal_frame_base (struct frame_info *this_frame, void **this_cache)
{
  struct arm_prologue_cache *cache;

  if (*this_cache == NULL)
    *this_cache = arm_make_prologue_cache (this_frame);
  cache = *this_cache;

  return cache->prev_sp - cache->framesize;
}

struct frame_base arm_normal_base = {
  &arm_prologue_unwind,
  arm_normal_frame_base,
  arm_normal_frame_base,
  arm_normal_frame_base
};

/* Assuming THIS_FRAME is a dummy, return the frame ID of that
   dummy frame.  The frame ID's base needs to match the TOS value
   saved by save_dummy_frame_tos() and returned from
   arm_push_dummy_call, and the PC needs to match the dummy frame's
   breakpoint.  */

static struct frame_id
arm_dummy_id (struct gdbarch *gdbarch, struct frame_info *this_frame)
{
  return frame_id_build (get_frame_register_unsigned (this_frame, ARM_SP_REGNUM),
                         get_frame_pc (this_frame));
}

/* Given THIS_FRAME, find the previous frame's resume PC (which will
   be used to construct the previous frame's ID, after looking up the
   containing function).  */

static CORE_ADDR
arm_unwind_pc (struct gdbarch *gdbarch, struct frame_info *this_frame)
{
  CORE_ADDR pc;
  pc = frame_unwind_register_unsigned (this_frame, ARM_PC_REGNUM);
  return arm_addr_bits_remove (gdbarch, pc);
}

static CORE_ADDR
arm_unwind_sp (struct gdbarch *gdbarch, struct frame_info *this_frame)
{
  return frame_unwind_register_unsigned (this_frame, ARM_SP_REGNUM);
}

static struct value *
arm_dwarf2_prev_register (struct frame_info *this_frame, void **this_cache,
                          int regnum)
{
  struct gdbarch * gdbarch = get_frame_arch (this_frame);
  CORE_ADDR lr, cpsr;

  switch (regnum)
    {
    case ARM_PC_REGNUM:
      /* The PC is normally copied from the return column, which
         describes saves of LR.  However, that version may have an
         extra bit set to indicate Thumb state.  The bit is not
         part of the PC.  */
      lr = frame_unwind_register_unsigned (this_frame, ARM_LR_REGNUM);
      return frame_unwind_got_constant (this_frame, regnum,
                                        arm_addr_bits_remove (gdbarch, lr));

    case ARM_PS_REGNUM:
      /* Reconstruct the T bit; see arm_prologue_prev_register for details.  */
      cpsr = get_frame_register_unsigned (this_frame, regnum);
      lr = frame_unwind_register_unsigned (this_frame, ARM_LR_REGNUM);
      if (IS_THUMB_ADDR (lr))
        cpsr |= CPSR_T;
      else
        cpsr &= ~CPSR_T;
      return frame_unwind_got_constant (this_frame, regnum, cpsr);

    default:
      internal_error (__FILE__, __LINE__,
                      _("Unexpected register %d"), regnum);
    }
}

static void
arm_dwarf2_frame_init_reg (struct gdbarch *gdbarch, int regnum,
                           struct dwarf2_frame_state_reg *reg,
                           struct frame_info *this_frame)
{
  switch (regnum)
    {
    case ARM_PC_REGNUM:
    case ARM_PS_REGNUM:
      reg->how = DWARF2_FRAME_REG_FN;
      reg->loc.fn = arm_dwarf2_prev_register;
      break;
    case ARM_SP_REGNUM:
      reg->how = DWARF2_FRAME_REG_CFA;
      break;
    }
}

/* When arguments must be pushed onto the stack, they go on in reverse
   order.  The code below implements a FILO (stack) to do this.  */

struct stack_item
{
  int len;
  struct stack_item *prev;
  void *data;
};

static struct stack_item *
push_stack_item (struct stack_item *prev, void *contents, int len)
{
  struct stack_item *si;
  si = xmalloc (sizeof (struct stack_item));
  si->data = xmalloc (len);
  si->len = len;
  si->prev = prev;
  memcpy (si->data, contents, len);
  return si;
}

static struct stack_item *
pop_stack_item (struct stack_item *si)
{
  struct stack_item *dead = si;
  si = si->prev;
  xfree (dead->data);
  xfree (dead);
  return si;
}


/* Return the alignment (in bytes) of the given type.  */

static int
arm_type_align (struct type *t)
{
  int n;
  int align;
  int falign;

  t = check_typedef (t);
  switch (TYPE_CODE (t))
    {
    default:
      /* Should never happen.  */
      internal_error (__FILE__, __LINE__, _("unknown type alignment"));
      return 4;

    case TYPE_CODE_PTR:
    case TYPE_CODE_ENUM:
    case TYPE_CODE_INT:
    case TYPE_CODE_FLT:
    case TYPE_CODE_SET:
    case TYPE_CODE_RANGE:
    case TYPE_CODE_BITSTRING:
    case TYPE_CODE_REF:
    case TYPE_CODE_CHAR:
    case TYPE_CODE_BOOL:
      return TYPE_LENGTH (t);

    case TYPE_CODE_ARRAY:
    case TYPE_CODE_COMPLEX:
      /* TODO: What about vector types?  */
      return arm_type_align (TYPE_TARGET_TYPE (t));

    case TYPE_CODE_STRUCT:
    case TYPE_CODE_UNION:
      align = 1;
      for (n = 0; n < TYPE_NFIELDS (t); n++)
        {
          falign = arm_type_align (TYPE_FIELD_TYPE (t, n));
          if (falign > align)
            align = falign;
        }
      return align;
    }
}

/* Possible base types for a candidate for passing and returning in
   VFP registers.  */

enum arm_vfp_cprc_base_type
{
  VFP_CPRC_UNKNOWN,
  VFP_CPRC_SINGLE,
  VFP_CPRC_DOUBLE,
  VFP_CPRC_VEC64,
  VFP_CPRC_VEC128
};

/* The length of one element of base type B.  */

static unsigned
arm_vfp_cprc_unit_length (enum arm_vfp_cprc_base_type b)
{
  switch (b)
    {
    case VFP_CPRC_SINGLE:
      return 4;
    case VFP_CPRC_DOUBLE:
      return 8;
    case VFP_CPRC_VEC64:
      return 8;
    case VFP_CPRC_VEC128:
      return 16;
    default:
      internal_error (__FILE__, __LINE__, _("Invalid VFP CPRC type: %d."),
                      (int) b);
    }
}

/* The character ('s', 'd' or 'q') for the type of VFP register used
   for passing base type B.  */

static int
arm_vfp_cprc_reg_char (enum arm_vfp_cprc_base_type b)
{
  switch (b)
    {
    case VFP_CPRC_SINGLE:
      return 's';
    case VFP_CPRC_DOUBLE:
      return 'd';
    case VFP_CPRC_VEC64:
      return 'd';
    case VFP_CPRC_VEC128:
      return 'q';
    default:
      internal_error (__FILE__, __LINE__, _("Invalid VFP CPRC type: %d."),
                      (int) b);
    }
}

/* Determine whether T may be part of a candidate for passing and
   returning in VFP registers, ignoring the limit on the total number
   of components.  If *BASE_TYPE is VFP_CPRC_UNKNOWN, set it to the
   classification of the first valid component found; if it is not
   VFP_CPRC_UNKNOWN, all components must have the same classification
   as *BASE_TYPE.  If it is found that T contains a type not permitted
   for passing and returning in VFP registers, a type differently
   classified from *BASE_TYPE, or two types differently classified
   from each other, return -1, otherwise return the total number of
   base-type elements found (possibly 0 in an empty structure or
   array).  Vectors and complex types are not currently supported,
   matching the generic AAPCS support.  */

static int
arm_vfp_cprc_sub_candidate (struct type *t,
                            enum arm_vfp_cprc_base_type *base_type)
{
  t = check_typedef (t);
  switch (TYPE_CODE (t))
    {
    case TYPE_CODE_FLT:
      switch (TYPE_LENGTH (t))
        {
        case 4:
          if (*base_type == VFP_CPRC_UNKNOWN)
            *base_type = VFP_CPRC_SINGLE;
          else if (*base_type != VFP_CPRC_SINGLE)
            return -1;
          return 1;

        case 8:
          if (*base_type == VFP_CPRC_UNKNOWN)
            *base_type = VFP_CPRC_DOUBLE;
          else if (*base_type != VFP_CPRC_DOUBLE)
            return -1;
          return 1;

        default:
          return -1;
        }
      break;

    case TYPE_CODE_ARRAY:
      {
        int count;
        unsigned unitlen;
        count = arm_vfp_cprc_sub_candidate (TYPE_TARGET_TYPE (t), base_type);
        if (count == -1)
          return -1;
        if (TYPE_LENGTH (t) == 0)
          {
            gdb_assert (count == 0);
            return 0;
          }
        else if (count == 0)
          return -1;
        unitlen = arm_vfp_cprc_unit_length (*base_type);
        gdb_assert ((TYPE_LENGTH (t) % unitlen) == 0);
        return TYPE_LENGTH (t) / unitlen;
      }
      break;

    case TYPE_CODE_STRUCT:
      {
        int count = 0;
        unsigned unitlen;
        int i;
        for (i = 0; i < TYPE_NFIELDS (t); i++)
          {
            int sub_count = arm_vfp_cprc_sub_candidate (TYPE_FIELD_TYPE (t, i),
                                                        base_type);
            if (sub_count == -1)
              return -1;
            count += sub_count;
          }
        if (TYPE_LENGTH (t) == 0)
          {
            gdb_assert (count == 0);
            return 0;
          }
        else if (count == 0)
          return -1;
        unitlen = arm_vfp_cprc_unit_length (*base_type);
        if (TYPE_LENGTH (t) != unitlen * count)
          return -1;
        return count;
      }

    case TYPE_CODE_UNION:
      {
        int count = 0;
        unsigned unitlen;
        int i;
        for (i = 0; i < TYPE_NFIELDS (t); i++)
          {
            int sub_count = arm_vfp_cprc_sub_candidate (TYPE_FIELD_TYPE (t, i),
                                                        base_type);
            if (sub_count == -1)
              return -1;
            count = (count > sub_count ? count : sub_count);
          }
        if (TYPE_LENGTH (t) == 0)
          {
            gdb_assert (count == 0);
            return 0;
          }
        else if (count == 0)
          return -1;
        unitlen = arm_vfp_cprc_unit_length (*base_type);
        if (TYPE_LENGTH (t) != unitlen * count)
          return -1;
        return count;
      }

    default:
      break;
    }

  return -1;
}

/* Determine whether T is a VFP co-processor register candidate (CPRC)
   if passed to or returned from a non-variadic function with the VFP
   ABI in effect.  Return 1 if it is, 0 otherwise.  If it is, set
   *BASE_TYPE to the base type for T and *COUNT to the number of
   elements of that base type before returning.  */

static int
arm_vfp_call_candidate (struct type *t, enum arm_vfp_cprc_base_type *base_type,
                        int *count)
{
  enum arm_vfp_cprc_base_type b = VFP_CPRC_UNKNOWN;
  int c = arm_vfp_cprc_sub_candidate (t, &b);
  if (c <= 0 || c > 4)
    return 0;
  *base_type = b;
  *count = c;
  return 1;
}

/* Return 1 if the VFP ABI should be used for passing arguments to and
   returning values from a function of type FUNC_TYPE, 0
   otherwise.  */

static int
arm_vfp_abi_for_function (struct gdbarch *gdbarch, struct type *func_type)
{
  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
  /* Variadic functions always use the base ABI.  Assume that functions
     without debug info are not variadic.  */
  if (func_type && TYPE_VARARGS (check_typedef (func_type)))
    return 0;
  /* The VFP ABI is only supported as a variant of AAPCS.  */
  if (tdep->arm_abi != ARM_ABI_AAPCS)
    return 0;
  return gdbarch_tdep (gdbarch)->fp_model == ARM_FLOAT_VFP;
}

/* We currently only support passing parameters in integer registers, which
   conforms with GCC's default model, and VFP argument passing following
   the VFP variant of AAPCS.  Several other variants exist and
   we should probably support some of them based on the selected ABI.  */

static CORE_ADDR
arm_push_dummy_call (struct gdbarch *gdbarch, struct value *function,
                     struct regcache *regcache, CORE_ADDR bp_addr, int nargs,
                     struct value **args, CORE_ADDR sp, int struct_return,
                     CORE_ADDR struct_addr)
{
  enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
  int argnum;
  int argreg;
  int nstack;
  struct stack_item *si = NULL;
  int use_vfp_abi;
  struct type *ftype;
  unsigned vfp_regs_free = (1 << 16) - 1;

  /* Determine the type of this function and whether the VFP ABI
     applies.  */
  ftype = check_typedef (value_type (function));
  if (TYPE_CODE (ftype) == TYPE_CODE_PTR)
    ftype = check_typedef (TYPE_TARGET_TYPE (ftype));
  use_vfp_abi = arm_vfp_abi_for_function (gdbarch, ftype);

  /* Set the return address.  For the ARM, the return breakpoint is
     always at BP_ADDR.  */
  /* XXX Fix for Thumb.  */
  regcache_cooked_write_unsigned (regcache, ARM_LR_REGNUM, bp_addr);

  /* Walk through the list of args and determine how large a temporary
     stack is required.  Need to take care here as structs may be
     passed on the stack, and we have to to push them.  */
  nstack = 0;

  argreg = ARM_A1_REGNUM;
  nstack = 0;

  /* The struct_return pointer occupies the first parameter
     passing register.  */
  if (struct_return)
    {
      if (arm_debug)
        fprintf_unfiltered (gdb_stdlog, "struct return in %s = %s\n",
                            gdbarch_register_name (gdbarch, argreg),
                            paddress (gdbarch, struct_addr));
      regcache_cooked_write_unsigned (regcache, argreg, struct_addr);
      argreg++;
    }

  for (argnum = 0; argnum < nargs; argnum++)
    {
      int len;
      struct type *arg_type;
      struct type *target_type;
      enum type_code typecode;
      bfd_byte *val;
      int align;
      enum arm_vfp_cprc_base_type vfp_base_type;
      int vfp_base_count;
      int may_use_core_reg = 1;

      arg_type = check_typedef (value_type (args[argnum]));
      len = TYPE_LENGTH (arg_type);
      target_type = TYPE_TARGET_TYPE (arg_type);
      typecode = TYPE_CODE (arg_type);
      val = value_contents_writeable (args[argnum]);

      align = arm_type_align (arg_type);
      /* Round alignment up to a whole number of words.  */
      align = (align + INT_REGISTER_SIZE - 1) & ~(INT_REGISTER_SIZE - 1);
      /* Different ABIs have different maximum alignments.  */
      if (gdbarch_tdep (gdbarch)->arm_abi == ARM_ABI_APCS)
        {
          /* The APCS ABI only requires word alignment.  */
          align = INT_REGISTER_SIZE;
        }
      else
        {
          /* The AAPCS requires at most doubleword alignment.  */
          if (align > INT_REGISTER_SIZE * 2)
            align = INT_REGISTER_SIZE * 2;
        }

      if (use_vfp_abi
          && arm_vfp_call_candidate (arg_type, &vfp_base_type,
                                     &vfp_base_count))
        {
          int regno;
          int unit_length;
          int shift;
          unsigned mask;

          /* Because this is a CPRC it cannot go in a core register or
             cause a core register to be skipped for alignment.
             Either it goes in VFP registers and the rest of this loop
             iteration is skipped for this argument, or it goes on the
             stack (and the stack alignment code is correct for this
             case).  */
          may_use_core_reg = 0;

          unit_length = arm_vfp_cprc_unit_length (vfp_base_type);
          shift = unit_length / 4;
          mask = (1 << (shift * vfp_base_count)) - 1;
          for (regno = 0; regno < 16; regno += shift)
            if (((vfp_regs_free >> regno) & mask) == mask)
              break;

          if (regno < 16)
            {
              int reg_char;
              int reg_scaled;
              int i;

              vfp_regs_free &= ~(mask << regno);
              reg_scaled = regno / shift;
              reg_char = arm_vfp_cprc_reg_char (vfp_base_type);
              for (i = 0; i < vfp_base_count; i++)
                {
                  char name_buf[4];
                  int regnum;
                  if (reg_char == 'q')
                    arm_neon_quad_write (gdbarch, regcache, reg_scaled + i,
                                         val + i * unit_length);
                  else
                    {
                      sprintf (name_buf, "%c%d", reg_char, reg_scaled + i);
                      regnum = user_reg_map_name_to_regnum (gdbarch, name_buf,
                                                            strlen (name_buf));
                      regcache_cooked_write (regcache, regnum,
                                             val + i * unit_length);
                    }
                }
              continue;
            }
          else
            {
              /* This CPRC could not go in VFP registers, so all VFP
                 registers are now marked as used.  */
              vfp_regs_free = 0;
            }
        }

      /* Push stack padding for dowubleword alignment.  */
      if (nstack & (align - 1))
        {
          si = push_stack_item (si, val, INT_REGISTER_SIZE);
          nstack += INT_REGISTER_SIZE;
        }
      
      /* Doubleword aligned quantities must go in even register pairs.  */
      if (may_use_core_reg
          && argreg <= ARM_LAST_ARG_REGNUM
          && align > INT_REGISTER_SIZE
          && argreg & 1)
        argreg++;

      /* If the argument is a pointer to a function, and it is a
         Thumb function, create a LOCAL copy of the value and set
         the THUMB bit in it.  */
      if (TYPE_CODE_PTR == typecode
          && target_type != NULL
          && TYPE_CODE_FUNC == TYPE_CODE (target_type))
        {
          CORE_ADDR regval = extract_unsigned_integer (val, len, byte_order);
          if (arm_pc_is_thumb (regval))
            {
              val = alloca (len);
              store_unsigned_integer (val, len, byte_order,
                                      MAKE_THUMB_ADDR (regval));
            }
        }

      /* Copy the argument to general registers or the stack in
         register-sized pieces.  Large arguments are split between
         registers and stack.  */
      while (len > 0)
        {
          int partial_len = len < INT_REGISTER_SIZE ? len : INT_REGISTER_SIZE;

          if (may_use_core_reg && argreg <= ARM_LAST_ARG_REGNUM)
            {
              /* The argument is being passed in a general purpose
                 register.  */
              CORE_ADDR regval
                = extract_unsigned_integer (val, partial_len, byte_order);
              if (byte_order == BFD_ENDIAN_BIG)
                regval <<= (INT_REGISTER_SIZE - partial_len) * 8;
              if (arm_debug)
                fprintf_unfiltered (gdb_stdlog, "arg %d in %s = 0x%s\n",
                                    argnum,
                                    gdbarch_register_name
                                      (gdbarch, argreg),
                                    phex (regval, INT_REGISTER_SIZE));
              regcache_cooked_write_unsigned (regcache, argreg, regval);
              argreg++;
            }
          else
            {
              /* Push the arguments onto the stack.  */
              if (arm_debug)
                fprintf_unfiltered (gdb_stdlog, "arg %d @ sp + %d\n",
                                    argnum, nstack);
              si = push_stack_item (si, val, INT_REGISTER_SIZE);
              nstack += INT_REGISTER_SIZE;
            }
              
          len -= partial_len;
          val += partial_len;
        }
    }
  /* If we have an odd number of words to push, then decrement the stack
     by one word now, so first stack argument will be dword aligned.  */
  if (nstack & 4)
    sp -= 4;

  while (si)
    {
      sp -= si->len;
      write_memory (sp, si->data, si->len);
      si = pop_stack_item (si);
    }

  /* Finally, update teh SP register.  */
  regcache_cooked_write_unsigned (regcache, ARM_SP_REGNUM, sp);

  return sp;
}


/* Always align the frame to an 8-byte boundary.  This is required on
   some platforms and harmless on the rest.  */

static CORE_ADDR
arm_frame_align (struct gdbarch *gdbarch, CORE_ADDR sp)
{
  /* Align the stack to eight bytes.  */
  return sp & ~ (CORE_ADDR) 7;
}

static void
print_fpu_flags (int flags)
{
  if (flags & (1 << 0))
    fputs ("IVO ", stdout);
  if (flags & (1 << 1))
    fputs ("DVZ ", stdout);
  if (flags & (1 << 2))
    fputs ("OFL ", stdout);
  if (flags & (1 << 3))
    fputs ("UFL ", stdout);
  if (flags & (1 << 4))
    fputs ("INX ", stdout);
  putchar ('\n');
}

/* Print interesting information about the floating point processor
   (if present) or emulator.  */
static void
arm_print_float_info (struct gdbarch *gdbarch, struct ui_file *file,
                      struct frame_info *frame, const char *args)
{
  unsigned long status = get_frame_register_unsigned (frame, ARM_FPS_REGNUM);
  int type;

  type = (status >> 24) & 127;
  if (status & (1 << 31))
    printf (_("Hardware FPU type %d\n"), type);
  else
    printf (_("Software FPU type %d\n"), type);
  /* i18n: [floating point unit] mask */
  fputs (_("mask: "), stdout);
  print_fpu_flags (status >> 16);
  /* i18n: [floating point unit] flags */
  fputs (_("flags: "), stdout);
  print_fpu_flags (status);
}

/* Construct the ARM extended floating point type.  */
static struct type *
arm_ext_type (struct gdbarch *gdbarch)
{
  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);

  if (!tdep->arm_ext_type)
    tdep->arm_ext_type
      = arch_float_type (gdbarch, -1, "builtin_type_arm_ext",
                         floatformats_arm_ext);

  return tdep->arm_ext_type;
}

static struct type *
arm_neon_double_type (struct gdbarch *gdbarch)
{
  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);

  if (tdep->neon_double_type == NULL)
    {
      struct type *t, *elem;

      t = arch_composite_type (gdbarch, "__gdb_builtin_type_neon_d",
                               TYPE_CODE_UNION);
      elem = builtin_type (gdbarch)->builtin_uint8;
      append_composite_type_field (t, "u8", init_vector_type (elem, 8));
      elem = builtin_type (gdbarch)->builtin_uint16;
      append_composite_type_field (t, "u16", init_vector_type (elem, 4));
      elem = builtin_type (gdbarch)->builtin_uint32;
      append_composite_type_field (t, "u32", init_vector_type (elem, 2));
      elem = builtin_type (gdbarch)->builtin_uint64;
      append_composite_type_field (t, "u64", elem);
      elem = builtin_type (gdbarch)->builtin_float;
      append_composite_type_field (t, "f32", init_vector_type (elem, 2));
      elem = builtin_type (gdbarch)->builtin_double;
      append_composite_type_field (t, "f64", elem);

      TYPE_VECTOR (t) = 1;
      TYPE_NAME (t) = "neon_d";
      tdep->neon_double_type = t;
    }

  return tdep->neon_double_type;
}

/* FIXME: The vector types are not correctly ordered on big-endian
   targets.  Just as s0 is the low bits of d0, d0[0] is also the low
   bits of d0 - regardless of what unit size is being held in d0.  So
   the offset of the first uint8 in d0 is 7, but the offset of the
   first float is 4.  This code works as-is for little-endian
   targets.  */

static struct type *
arm_neon_quad_type (struct gdbarch *gdbarch)
{
  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);

  if (tdep->neon_quad_type == NULL)
    {
      struct type *t, *elem;

      t = arch_composite_type (gdbarch, "__gdb_builtin_type_neon_q",
                               TYPE_CODE_UNION);
      elem = builtin_type (gdbarch)->builtin_uint8;
      append_composite_type_field (t, "u8", init_vector_type (elem, 16));
      elem = builtin_type (gdbarch)->builtin_uint16;
      append_composite_type_field (t, "u16", init_vector_type (elem, 8));
      elem = builtin_type (gdbarch)->builtin_uint32;
      append_composite_type_field (t, "u32", init_vector_type (elem, 4));
      elem = builtin_type (gdbarch)->builtin_uint64;
      append_composite_type_field (t, "u64", init_vector_type (elem, 2));
      elem = builtin_type (gdbarch)->builtin_float;
      append_composite_type_field (t, "f32", init_vector_type (elem, 4));
      elem = builtin_type (gdbarch)->builtin_double;
      append_composite_type_field (t, "f64", init_vector_type (elem, 2));

      TYPE_VECTOR (t) = 1;
      TYPE_NAME (t) = "neon_q";
      tdep->neon_quad_type = t;
    }

  return tdep->neon_quad_type;
}

/* Return the GDB type object for the "standard" data type of data in
   register N.  */

static struct type *
arm_register_type (struct gdbarch *gdbarch, int regnum)
{
  int num_regs = gdbarch_num_regs (gdbarch);

  if (gdbarch_tdep (gdbarch)->have_vfp_pseudos
      && regnum >= num_regs && regnum < num_regs + 32)
    return builtin_type (gdbarch)->builtin_float;

  if (gdbarch_tdep (gdbarch)->have_neon_pseudos
      && regnum >= num_regs + 32 && regnum < num_regs + 32 + 16)
    return arm_neon_quad_type (gdbarch);

  /* If the target description has register information, we are only
     in this function so that we can override the types of
     double-precision registers for NEON.  */
  if (tdesc_has_registers (gdbarch_target_desc (gdbarch)))
    {
      struct type *t = tdesc_register_type (gdbarch, regnum);

      if (regnum >= ARM_D0_REGNUM && regnum < ARM_D0_REGNUM + 32
          && TYPE_CODE (t) == TYPE_CODE_FLT
          && gdbarch_tdep (gdbarch)->have_neon)
        return arm_neon_double_type (gdbarch);
      else
        return t;
    }

  if (regnum >= ARM_F0_REGNUM && regnum < ARM_F0_REGNUM + NUM_FREGS)
    {
      if (!gdbarch_tdep (gdbarch)->have_fpa_registers)
        return builtin_type (gdbarch)->builtin_void;

      return arm_ext_type (gdbarch);
    }
  else if (regnum == ARM_SP_REGNUM)
    return builtin_type (gdbarch)->builtin_data_ptr;
  else if (regnum == ARM_PC_REGNUM)
    return builtin_type (gdbarch)->builtin_func_ptr;
  else if (regnum >= ARRAY_SIZE (arm_register_names))
    /* These registers are only supported on targets which supply
       an XML description.  */
    return builtin_type (gdbarch)->builtin_int0;
  else
    return builtin_type (gdbarch)->builtin_uint32;
}

/* Map a DWARF register REGNUM onto the appropriate GDB register
   number.  */

static int
arm_dwarf_reg_to_regnum (struct gdbarch *gdbarch, int reg)
{
  /* Core integer regs.  */
  if (reg >= 0 && reg <= 15)
    return reg;

  /* Legacy FPA encoding.  These were once used in a way which
     overlapped with VFP register numbering, so their use is
     discouraged, but GDB doesn't support the ARM toolchain
     which used them for VFP.  */
  if (reg >= 16 && reg <= 23)
    return ARM_F0_REGNUM + reg - 16;

  /* New assignments for the FPA registers.  */
  if (reg >= 96 && reg <= 103)
    return ARM_F0_REGNUM + reg - 96;

  /* WMMX register assignments.  */
  if (reg >= 104 && reg <= 111)
    return ARM_WCGR0_REGNUM + reg - 104;

  if (reg >= 112 && reg <= 127)
    return ARM_WR0_REGNUM + reg - 112;

  if (reg >= 192 && reg <= 199)
    return ARM_WC0_REGNUM + reg - 192;

  /* VFP v2 registers.  A double precision value is actually
     in d1 rather than s2, but the ABI only defines numbering
     for the single precision registers.  This will "just work"
     in GDB for little endian targets (we'll read eight bytes,
     starting in s0 and then progressing to s1), but will be
     reversed on big endian targets with VFP.  This won't
     be a problem for the new Neon quad registers; you're supposed
     to use DW_OP_piece for those.  */
  if (reg >= 64 && reg <= 95)
    {
      char name_buf[4];

      sprintf (name_buf, "s%d", reg - 64);
      return user_reg_map_name_to_regnum (gdbarch, name_buf,
                                          strlen (name_buf));
    }

  /* VFP v3 / Neon registers.  This range is also used for VFP v2
     registers, except that it now describes d0 instead of s0.  */
  if (reg >= 256 && reg <= 287)
    {
      char name_buf[4];

      sprintf (name_buf, "d%d", reg - 256);
      return user_reg_map_name_to_regnum (gdbarch, name_buf,
                                          strlen (name_buf));
    }

  return -1;
}

/* Map GDB internal REGNUM onto the Arm simulator register numbers.  */
static int
arm_register_sim_regno (struct gdbarch *gdbarch, int regnum)
{
  int reg = regnum;
  gdb_assert (reg >= 0 && reg < gdbarch_num_regs (gdbarch));

  if (regnum >= ARM_WR0_REGNUM && regnum <= ARM_WR15_REGNUM)
    return regnum - ARM_WR0_REGNUM + SIM_ARM_IWMMXT_COP0R0_REGNUM;

  if (regnum >= ARM_WC0_REGNUM && regnum <= ARM_WC7_REGNUM)
    return regnum - ARM_WC0_REGNUM + SIM_ARM_IWMMXT_COP1R0_REGNUM;

  if (regnum >= ARM_WCGR0_REGNUM && regnum <= ARM_WCGR7_REGNUM)
    return regnum - ARM_WCGR0_REGNUM + SIM_ARM_IWMMXT_COP1R8_REGNUM;

  if (reg < NUM_GREGS)
    return SIM_ARM_R0_REGNUM + reg;
  reg -= NUM_GREGS;

  if (reg < NUM_FREGS)
    return SIM_ARM_FP0_REGNUM + reg;
  reg -= NUM_FREGS;

  if (reg < NUM_SREGS)
    return SIM_ARM_FPS_REGNUM + reg;
  reg -= NUM_SREGS;

  internal_error (__FILE__, __LINE__, _("Bad REGNUM %d"), regnum);
}

/* NOTE: cagney/2001-08-20: Both convert_from_extended() and
   convert_to_extended() use floatformat_arm_ext_littlebyte_bigword.
   It is thought that this is is the floating-point register format on
   little-endian systems.  */

static void
convert_from_extended (const struct floatformat *fmt, const void *ptr,
                       void *dbl, int endianess)
{
  DOUBLEST d;

  if (endianess == BFD_ENDIAN_BIG)
    floatformat_to_doublest (&floatformat_arm_ext_big, ptr, &d);
  else
    floatformat_to_doublest (&floatformat_arm_ext_littlebyte_bigword,
                             ptr, &d);
  floatformat_from_doublest (fmt, &d, dbl);
}

static void
convert_to_extended (const struct floatformat *fmt, void *dbl, const void *ptr,
                     int endianess)
{
  DOUBLEST d;

  floatformat_to_doublest (fmt, ptr, &d);
  if (endianess == BFD_ENDIAN_BIG)
    floatformat_from_doublest (&floatformat_arm_ext_big, &d, dbl);
  else
    floatformat_from_doublest (&floatformat_arm_ext_littlebyte_bigword,
                               &d, dbl);
}

static int
condition_true (unsigned long cond, unsigned long status_reg)
{
  if (cond == INST_AL || cond == INST_NV)
    return 1;

  switch (cond)
    {
    case INST_EQ:
      return ((status_reg & FLAG_Z) != 0);
    case INST_NE:
      return ((status_reg & FLAG_Z) == 0);
    case INST_CS:
      return ((status_reg & FLAG_C) != 0);
    case INST_CC:
      return ((status_reg & FLAG_C) == 0);
    case INST_MI:
      return ((status_reg & FLAG_N) != 0);
    case INST_PL:
      return ((status_reg & FLAG_N) == 0);
    case INST_VS:
      return ((status_reg & FLAG_V) != 0);
    case INST_VC:
      return ((status_reg & FLAG_V) == 0);
    case INST_HI:
      return ((status_reg & (FLAG_C | FLAG_Z)) == FLAG_C);
    case INST_LS:
      return ((status_reg & (FLAG_C | FLAG_Z)) != FLAG_C);
    case INST_GE:
      return (((status_reg & FLAG_N) == 0) == ((status_reg & FLAG_V) == 0));
    case INST_LT:
      return (((status_reg & FLAG_N) == 0) != ((status_reg & FLAG_V) == 0));
    case INST_GT:
      return (((status_reg & FLAG_Z) == 0) &&
              (((status_reg & FLAG_N) == 0) == ((status_reg & FLAG_V) == 0)));
    case INST_LE:
      return (((status_reg & FLAG_Z) != 0) ||
              (((status_reg & FLAG_N) == 0) != ((status_reg & FLAG_V) == 0)));
    }
  return 1;
}

/* Support routines for single stepping.  Calculate the next PC value.  */
#define submask(x) ((1L << ((x) + 1)) - 1)
#define bit(obj,st) (((obj) >> (st)) & 1)
#define bits(obj,st,fn) (((obj) >> (st)) & submask ((fn) - (st)))
#define sbits(obj,st,fn) \
  ((long) (bits(obj,st,fn) | ((long) bit(obj,fn) * ~ submask (fn - st))))
#define BranchDest(addr,instr) \
  ((CORE_ADDR) (((long) (addr)) + 8 + (sbits (instr, 0, 23) << 2)))
#define ARM_PC_32 1

static unsigned long
shifted_reg_val (struct frame_info *frame, unsigned long inst, int carry,
                 unsigned long pc_val, unsigned long status_reg)
{
  unsigned long res, shift;
  int rm = bits (inst, 0, 3);
  unsigned long shifttype = bits (inst, 5, 6);

  if (bit (inst, 4))
    {
      int rs = bits (inst, 8, 11);
      shift = (rs == 15 ? pc_val + 8
                        : get_frame_register_unsigned (frame, rs)) & 0xFF;
    }
  else
    shift = bits (inst, 7, 11);

  res = (rm == 15
         ? ((pc_val | (ARM_PC_32 ? 0 : status_reg))
            + (bit (inst, 4) ? 12 : 8))
         : get_frame_register_unsigned (frame, rm));

  switch (shifttype)
    {
    case 0:                     /* LSL */
      res = shift >= 32 ? 0 : res << shift;
      break;

    case 1:                     /* LSR */
      res = shift >= 32 ? 0 : res >> shift;
      break;

    case 2:                     /* ASR */
      if (shift >= 32)
        shift = 31;
      res = ((res & 0x80000000L)
             ? ~((~res) >> shift) : res >> shift);
      break;

    case 3:                     /* ROR/RRX */
      shift &= 31;
      if (shift == 0)
        res = (res >> 1) | (carry ? 0x80000000L : 0);
      else
        res = (res >> shift) | (res << (32 - shift));
      break;
    }

  return res & 0xffffffff;
}

/* Return number of 1-bits in VAL.  */

static int
bitcount (unsigned long val)
{
  int nbits;
  for (nbits = 0; val != 0; nbits++)
    val &= val - 1;             /* delete rightmost 1-bit in val */
  return nbits;
}

static CORE_ADDR
thumb_get_next_pc (struct frame_info *frame, CORE_ADDR pc)
{
  struct gdbarch *gdbarch = get_frame_arch (frame);
  enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
  enum bfd_endian byte_order_for_code = gdbarch_byte_order_for_code (gdbarch);
  unsigned long pc_val = ((unsigned long) pc) + 4;      /* PC after prefetch */
  unsigned short inst1;
  CORE_ADDR nextpc = pc + 2;            /* default is next instruction */
  unsigned long offset;

  inst1 = read_memory_unsigned_integer (pc, 2, byte_order_for_code);

  if ((inst1 & 0xff00) == 0xbd00)       /* pop {rlist, pc} */
    {
      CORE_ADDR sp;

      /* Fetch the saved PC from the stack.  It's stored above
         all of the other registers.  */
      offset = bitcount (bits (inst1, 0, 7)) * INT_REGISTER_SIZE;
      sp = get_frame_register_unsigned (frame, ARM_SP_REGNUM);
      nextpc = read_memory_unsigned_integer (sp + offset, 4, byte_order);
      nextpc = gdbarch_addr_bits_remove (gdbarch, nextpc);
      if (nextpc == pc)
        error (_("Infinite loop detected"));
    }
  else if ((inst1 & 0xf000) == 0xd000)  /* conditional branch */
    {
      unsigned long status = get_frame_register_unsigned (frame, ARM_PS_REGNUM);
      unsigned long cond = bits (inst1, 8, 11);
      if (cond != 0x0f && condition_true (cond, status))    /* 0x0f = SWI */
        nextpc = pc_val + (sbits (inst1, 0, 7) << 1);
    }
  else if ((inst1 & 0xf800) == 0xe000)  /* unconditional branch */
    {
      nextpc = pc_val + (sbits (inst1, 0, 10) << 1);
    }
  else if ((inst1 & 0xf800) == 0xf000)  /* long branch with link, and blx */
    {
      unsigned short inst2;
      inst2 = read_memory_unsigned_integer (pc + 2, 2, byte_order_for_code);
      offset = (sbits (inst1, 0, 10) << 12) + (bits (inst2, 0, 10) << 1);
      nextpc = pc_val + offset;
      /* For BLX make sure to clear the low bits.  */
      if (bits (inst2, 11, 12) == 1)
        nextpc = nextpc & 0xfffffffc;
    }
  else if ((inst1 & 0xff00) == 0x4700)  /* bx REG, blx REG */
    {
      if (bits (inst1, 3, 6) == 0x0f)
        nextpc = pc_val;
      else
        nextpc = get_frame_register_unsigned (frame, bits (inst1, 3, 6));

      nextpc = gdbarch_addr_bits_remove (gdbarch, nextpc);
      if (nextpc == pc)
        error (_("Infinite loop detected"));
    }

  return nextpc;
}

CORE_ADDR
arm_get_next_pc (struct frame_info *frame, CORE_ADDR pc)
{
  struct gdbarch *gdbarch = get_frame_arch (frame);
  enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
  enum bfd_endian byte_order_for_code = gdbarch_byte_order_for_code (gdbarch);
  unsigned long pc_val;
  unsigned long this_instr;
  unsigned long status;
  CORE_ADDR nextpc;

  if (arm_frame_is_thumb (frame))
    return thumb_get_next_pc (frame, pc);

  pc_val = (unsigned long) pc;
  this_instr = read_memory_unsigned_integer (pc, 4, byte_order_for_code);

  status = get_frame_register_unsigned (frame, ARM_PS_REGNUM);
  nextpc = (CORE_ADDR) (pc_val + 4);    /* Default case */

  if (bits (this_instr, 28, 31) == INST_NV)
    switch (bits (this_instr, 24, 27))
      {
      case 0xa:
      case 0xb:
        {
          /* Branch with Link and change to Thumb.  */
          nextpc = BranchDest (pc, this_instr);
          nextpc |= bit (this_instr, 24) << 1;

          nextpc = gdbarch_addr_bits_remove (gdbarch, nextpc);
          if (nextpc == pc)
            error (_("Infinite loop detected"));
          break;
        }
      case 0xc:
      case 0xd:
      case 0xe:
        /* Coprocessor register transfer.  */
        if (bits (this_instr, 12, 15) == 15)
          error (_("Invalid update to pc in instruction"));
        break;
      }
  else if (condition_true (bits (this_instr, 28, 31), status))
    {
      switch (bits (this_instr, 24, 27))
        {
        case 0x0:
        case 0x1:                       /* data processing */
        case 0x2:
        case 0x3:
          {
            unsigned long operand1, operand2, result = 0;
            unsigned long rn;
            int c;

            if (bits (this_instr, 12, 15) != 15)
              break;

            if (bits (this_instr, 22, 25) == 0
                && bits (this_instr, 4, 7) == 9)        /* multiply */
              error (_("Invalid update to pc in instruction"));

            /* BX <reg>, BLX <reg> */
            if (bits (this_instr, 4, 27) == 0x12fff1
                || bits (this_instr, 4, 27) == 0x12fff3)
              {
                rn = bits (this_instr, 0, 3);
                result = (rn == 15) ? pc_val + 8
                                    : get_frame_register_unsigned (frame, rn);
                nextpc = (CORE_ADDR) gdbarch_addr_bits_remove
                                       (gdbarch, result);

                if (nextpc == pc)
                  error (_("Infinite loop detected"));

                return nextpc;
              }

            /* Multiply into PC */
            c = (status & FLAG_C) ? 1 : 0;
            rn = bits (this_instr, 16, 19);
            operand1 = (rn == 15) ? pc_val + 8
                                  : get_frame_register_unsigned (frame, rn);

            if (bit (this_instr, 25))
              {
                unsigned long immval = bits (this_instr, 0, 7);
                unsigned long rotate = 2 * bits (this_instr, 8, 11);
                operand2 = ((immval >> rotate) | (immval << (32 - rotate)))
                  & 0xffffffff;
              }
            else                /* operand 2 is a shifted register */
              operand2 = shifted_reg_val (frame, this_instr, c, pc_val, status);

            switch (bits (this_instr, 21, 24))
              {
              case 0x0: /*and */
                result = operand1 & operand2;
                break;

              case 0x1: /*eor */
                result = operand1 ^ operand2;
                break;

              case 0x2: /*sub */
                result = operand1 - operand2;
                break;

              case 0x3: /*rsb */
                result = operand2 - operand1;
                break;

              case 0x4: /*add */
                result = operand1 + operand2;
                break;

              case 0x5: /*adc */
                result = operand1 + operand2 + c;
                break;

              case 0x6: /*sbc */
                result = operand1 - operand2 + c;
                break;

              case 0x7: /*rsc */
                result = operand2 - operand1 + c;
                break;

              case 0x8:
              case 0x9:
              case 0xa:
              case 0xb: /* tst, teq, cmp, cmn */
                result = (unsigned long) nextpc;
                break;

              case 0xc: /*orr */
                result = operand1 | operand2;
                break;

              case 0xd: /*mov */
                /* Always step into a function.  */
                result = operand2;
                break;

              case 0xe: /*bic */
                result = operand1 & ~operand2;
                break;

              case 0xf: /*mvn */
                result = ~operand2;
                break;
              }
            nextpc = (CORE_ADDR) gdbarch_addr_bits_remove
                                   (gdbarch, result);

            if (nextpc == pc)
              error (_("Infinite loop detected"));
            break;
          }

        case 0x4:
        case 0x5:               /* data transfer */
        case 0x6:
        case 0x7:
          if (bit (this_instr, 20))
            {
              /* load */
              if (bits (this_instr, 12, 15) == 15)
                {
                  /* rd == pc */
                  unsigned long rn;
                  unsigned long base;

                  if (bit (this_instr, 22))
                    error (_("Invalid update to pc in instruction"));

                  /* byte write to PC */
                  rn = bits (this_instr, 16, 19);
                  base = (rn == 15) ? pc_val + 8
                                    : get_frame_register_unsigned (frame, rn);
                  if (bit (this_instr, 24))
                    {
                      /* pre-indexed */
                      int c = (status & FLAG_C) ? 1 : 0;
                      unsigned long offset =
                      (bit (this_instr, 25)
                       ? shifted_reg_val (frame, this_instr, c, pc_val, status)
                       : bits (this_instr, 0, 11));

                      if (bit (this_instr, 23))
                        base += offset;
                      else
                        base -= offset;
                    }
                  nextpc = (CORE_ADDR) read_memory_integer ((CORE_ADDR) base,
                                                            4, byte_order);

                  nextpc = gdbarch_addr_bits_remove (gdbarch, nextpc);

                  if (nextpc == pc)
                    error (_("Infinite loop detected"));
                }
            }
          break;

        case 0x8:
        case 0x9:               /* block transfer */
          if (bit (this_instr, 20))
            {
              /* LDM */
              if (bit (this_instr, 15))
                {
                  /* loading pc */
                  int offset = 0;

                  if (bit (this_instr, 23))
                    {
                      /* up */
                      unsigned long reglist = bits (this_instr, 0, 14);
                      offset = bitcount (reglist) * 4;
                      if (bit (this_instr, 24))         /* pre */
                        offset += 4;
                    }
                  else if (bit (this_instr, 24))
                    offset = -4;

                  {
                    unsigned long rn_val =
                    get_frame_register_unsigned (frame,
                                                 bits (this_instr, 16, 19));
                    nextpc =
                      (CORE_ADDR) read_memory_integer ((CORE_ADDR) (rn_val
                                                                  + offset),
                                                       4, byte_order);
                  }
                  nextpc = gdbarch_addr_bits_remove
                             (gdbarch, nextpc);
                  if (nextpc == pc)
                    error (_("Infinite loop detected"));
                }
            }
          break;

        case 0xb:               /* branch & link */
        case 0xa:               /* branch */
          {
            nextpc = BranchDest (pc, this_instr);

            nextpc = gdbarch_addr_bits_remove (gdbarch, nextpc);
            if (nextpc == pc)
              error (_("Infinite loop detected"));
            break;
          }

        case 0xc:
        case 0xd:
        case 0xe:               /* coproc ops */
        case 0xf:               /* SWI */
          break;

        default:
          fprintf_filtered (gdb_stderr, _("Bad bit-field extraction\n"));
          return (pc);
        }
    }

  return nextpc;
}

/* single_step() is called just before we want to resume the inferior,
   if we want to single-step it but there is no hardware or kernel
   single-step support.  We find the target of the coming instruction
   and breakpoint it.  */

int
arm_software_single_step (struct frame_info *frame)
{
  struct gdbarch *gdbarch = get_frame_arch (frame);

  /* NOTE: This may insert the wrong breakpoint instruction when
     single-stepping over a mode-changing instruction, if the
     CPSR heuristics are used.  */

  CORE_ADDR next_pc = arm_get_next_pc (frame, get_frame_pc (frame));
  insert_single_step_breakpoint (gdbarch, next_pc);

  return 1;
}

/* ARM displaced stepping support.

   Generally ARM displaced stepping works as follows:

   1. When an instruction is to be single-stepped, it is first decoded by
      arm_process_displaced_insn (called from arm_displaced_step_copy_insn).
      Depending on the type of instruction, it is then copied to a scratch
      location, possibly in a modified form.  The copy_* set of functions
      performs such modification, as necessary. A breakpoint is placed after
      the modified instruction in the scratch space to return control to GDB.
      Note in particular that instructions which modify the PC will no longer
      do so after modification.

   2. The instruction is single-stepped, by setting the PC to the scratch
      location address, and resuming.  Control returns to GDB when the
      breakpoint is hit.

   3. A cleanup function (cleanup_*) is called corresponding to the copy_*
      function used for the current instruction.  This function's job is to
      put the CPU/memory state back to what it would have been if the
      instruction had been executed unmodified in its original location.  */

/* NOP instruction (mov r0, r0).  */
#define ARM_NOP                         0xe1a00000

/* Helper for register reads for displaced stepping.  In particular, this
   returns the PC as it would be seen by the instruction at its original
   location.  */

ULONGEST
displaced_read_reg (struct regcache *regs, CORE_ADDR from, int regno)
{
  ULONGEST ret;

  if (regno == 15)
    {
      if (debug_displaced)
        fprintf_unfiltered (gdb_stdlog, "displaced: read pc value %.8lx\n",
                            (unsigned long) from + 8);
      return (ULONGEST) from + 8;  /* Pipeline offset.  */
    }
  else
    {
      regcache_cooked_read_unsigned (regs, regno, &ret);
      if (debug_displaced)
        fprintf_unfiltered (gdb_stdlog, "displaced: read r%d value %.8lx\n",
                            regno, (unsigned long) ret);
      return ret;
    }
}

static int
displaced_in_arm_mode (struct regcache *regs)
{
  ULONGEST ps;

  regcache_cooked_read_unsigned (regs, ARM_PS_REGNUM, &ps);

  return (ps & CPSR_T) == 0;
}

/* Write to the PC as from a branch instruction.  */

static void
branch_write_pc (struct regcache *regs, ULONGEST val)
{
  if (displaced_in_arm_mode (regs))
    /* Note: If bits 0/1 are set, this branch would be unpredictable for
       architecture versions < 6.  */
    regcache_cooked_write_unsigned (regs, ARM_PC_REGNUM, val & ~(ULONGEST) 0x3);
  else
    regcache_cooked_write_unsigned (regs, ARM_PC_REGNUM, val & ~(ULONGEST) 0x1);
}

/* Write to the PC as from a branch-exchange instruction.  */

static void
bx_write_pc (struct regcache *regs, ULONGEST val)
{
  ULONGEST ps;

  regcache_cooked_read_unsigned (regs, ARM_PS_REGNUM, &ps);

  if ((val & 1) == 1)
    {
      regcache_cooked_write_unsigned (regs, ARM_PS_REGNUM, ps | CPSR_T);
      regcache_cooked_write_unsigned (regs, ARM_PC_REGNUM, val & 0xfffffffe);
    }
  else if ((val & 2) == 0)
    {
      regcache_cooked_write_unsigned (regs, ARM_PS_REGNUM,
                                      ps & ~(ULONGEST) CPSR_T);
      regcache_cooked_write_unsigned (regs, ARM_PC_REGNUM, val);
    }
  else
    {
      /* Unpredictable behaviour.  Try to do something sensible (switch to ARM
          mode, align dest to 4 bytes).  */
      warning (_("Single-stepping BX to non-word-aligned ARM instruction."));
      regcache_cooked_write_unsigned (regs, ARM_PS_REGNUM,
                                      ps & ~(ULONGEST) CPSR_T);
      regcache_cooked_write_unsigned (regs, ARM_PC_REGNUM, val & 0xfffffffc);
    }
}

/* Write to the PC as if from a load instruction.  */

static void
load_write_pc (struct regcache *regs, ULONGEST val)
{
  if (DISPLACED_STEPPING_ARCH_VERSION >= 5)
    bx_write_pc (regs, val);
  else
    branch_write_pc (regs, val);
}

/* Write to the PC as if from an ALU instruction.  */

static void
alu_write_pc (struct regcache *regs, ULONGEST val)
{
  if (DISPLACED_STEPPING_ARCH_VERSION >= 7 && displaced_in_arm_mode (regs))
    bx_write_pc (regs, val);
  else
    branch_write_pc (regs, val);
}

/* Helper for writing to registers for displaced stepping.  Writing to the PC
   has a varying effects depending on the instruction which does the write:
   this is controlled by the WRITE_PC argument.  */

void
displaced_write_reg (struct regcache *regs, struct displaced_step_closure *dsc,
                     int regno, ULONGEST val, enum pc_write_style write_pc)
{
  if (regno == 15)
    {
      if (debug_displaced)
        fprintf_unfiltered (gdb_stdlog, "displaced: writing pc %.8lx\n",
                            (unsigned long) val);
      switch (write_pc)
        {
        case BRANCH_WRITE_PC:
          branch_write_pc (regs, val);
          break;

        case BX_WRITE_PC:
          bx_write_pc (regs, val);
          break;

        case LOAD_WRITE_PC:
          load_write_pc (regs, val);
          break;

        case ALU_WRITE_PC:
          alu_write_pc (regs, val);
          break;

        case CANNOT_WRITE_PC:
          warning (_("Instruction wrote to PC in an unexpected way when "
                     "single-stepping"));
          break;

        default:
          internal_error (__FILE__, __LINE__,
                          _("Invalid argument to displaced_write_reg"));
        }

      dsc->wrote_to_pc = 1;
    }
  else
    {
      if (debug_displaced)
        fprintf_unfiltered (gdb_stdlog, "displaced: writing r%d value %.8lx\n",
                            regno, (unsigned long) val);
      regcache_cooked_write_unsigned (regs, regno, val);
    }
}

/* This function is used to concisely determine if an instruction INSN
   references PC.  Register fields of interest in INSN should have the
   corresponding fields of BITMASK set to 0b1111.  The function returns return 1
   if any of these fields in INSN reference the PC (also 0b1111, r15), else it
   returns 0.  */

static int
insn_references_pc (uint32_t insn, uint32_t bitmask)
{
  uint32_t lowbit = 1;

  while (bitmask != 0)
    {
      uint32_t mask;

      for (; lowbit && (bitmask & lowbit) == 0; lowbit <<= 1)
        ;

      if (!lowbit)
        break;

      mask = lowbit * 0xf;

      if ((insn & mask) == mask)
        return 1;

      bitmask &= ~mask;
    }

  return 0;
}

/* The simplest copy function.  Many instructions have the same effect no
   matter what address they are executed at: in those cases, use this.  */

static int
copy_unmodified (struct gdbarch *gdbarch ATTRIBUTE_UNUSED, uint32_t insn,
                 const char *iname, struct displaced_step_closure *dsc)
{
  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying insn %.8lx, "
                        "opcode/class '%s' unmodified\n", (unsigned long) insn,
                        iname);

  dsc->modinsn[0] = insn;

  return 0;
}

/* Preload instructions with immediate offset.  */

static void
cleanup_preload (struct gdbarch *gdbarch ATTRIBUTE_UNUSED,
                 struct regcache *regs, struct displaced_step_closure *dsc)
{
  displaced_write_reg (regs, dsc, 0, dsc->tmp[0], CANNOT_WRITE_PC);
  if (!dsc->u.preload.immed)
    displaced_write_reg (regs, dsc, 1, dsc->tmp[1], CANNOT_WRITE_PC);
}

static int
copy_preload (struct gdbarch *gdbarch, uint32_t insn, struct regcache *regs,
              struct displaced_step_closure *dsc)
{
  unsigned int rn = bits (insn, 16, 19);
  ULONGEST rn_val;
  CORE_ADDR from = dsc->insn_addr;

  if (!insn_references_pc (insn, 0x000f0000ul))
    return copy_unmodified (gdbarch, insn, "preload", dsc);

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying preload insn %.8lx\n",
                        (unsigned long) insn);

  /* Preload instructions:

     {pli/pld} [rn, #+/-imm]
     ->
     {pli/pld} [r0, #+/-imm].  */

  dsc->tmp[0] = displaced_read_reg (regs, from, 0);
  rn_val = displaced_read_reg (regs, from, rn);
  displaced_write_reg (regs, dsc, 0, rn_val, CANNOT_WRITE_PC);

  dsc->u.preload.immed = 1;

  dsc->modinsn[0] = insn & 0xfff0ffff;

  dsc->cleanup = &cleanup_preload;

  return 0;
}

/* Preload instructions with register offset.  */

static int
copy_preload_reg (struct gdbarch *gdbarch, uint32_t insn, struct regcache *regs,
                  struct displaced_step_closure *dsc)
{
  unsigned int rn = bits (insn, 16, 19);
  unsigned int rm = bits (insn, 0, 3);
  ULONGEST rn_val, rm_val;
  CORE_ADDR from = dsc->insn_addr;

  if (!insn_references_pc (insn, 0x000f000ful))
    return copy_unmodified (gdbarch, insn, "preload reg", dsc);

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying preload insn %.8lx\n",
                        (unsigned long) insn);

  /* Preload register-offset instructions:

     {pli/pld} [rn, rm {, shift}]
     ->
     {pli/pld} [r0, r1 {, shift}].  */

  dsc->tmp[0] = displaced_read_reg (regs, from, 0);
  dsc->tmp[1] = displaced_read_reg (regs, from, 1);
  rn_val = displaced_read_reg (regs, from, rn);
  rm_val = displaced_read_reg (regs, from, rm);
  displaced_write_reg (regs, dsc, 0, rn_val, CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, 1, rm_val, CANNOT_WRITE_PC);

  dsc->u.preload.immed = 0;

  dsc->modinsn[0] = (insn & 0xfff0fff0) | 0x1;

  dsc->cleanup = &cleanup_preload;

  return 0;
}

/* Copy/cleanup coprocessor load and store instructions.  */

static void
cleanup_copro_load_store (struct gdbarch *gdbarch ATTRIBUTE_UNUSED,
                          struct regcache *regs,
                          struct displaced_step_closure *dsc)
{
  ULONGEST rn_val = displaced_read_reg (regs, dsc->insn_addr, 0);

  displaced_write_reg (regs, dsc, 0, dsc->tmp[0], CANNOT_WRITE_PC);

  if (dsc->u.ldst.writeback)
    displaced_write_reg (regs, dsc, dsc->u.ldst.rn, rn_val, LOAD_WRITE_PC);
}

static int
copy_copro_load_store (struct gdbarch *gdbarch, uint32_t insn,
                       struct regcache *regs,
                       struct displaced_step_closure *dsc)
{
  unsigned int rn = bits (insn, 16, 19);
  ULONGEST rn_val;
  CORE_ADDR from = dsc->insn_addr;

  if (!insn_references_pc (insn, 0x000f0000ul))
    return copy_unmodified (gdbarch, insn, "copro load/store", dsc);

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying coprocessor "
                        "load/store insn %.8lx\n", (unsigned long) insn);

  /* Coprocessor load/store instructions:

     {stc/stc2} [<Rn>, #+/-imm]  (and other immediate addressing modes)
     ->
     {stc/stc2} [r0, #+/-imm].

     ldc/ldc2 are handled identically.  */

  dsc->tmp[0] = displaced_read_reg (regs, from, 0);
  rn_val = displaced_read_reg (regs, from, rn);
  displaced_write_reg (regs, dsc, 0, rn_val, CANNOT_WRITE_PC);

  dsc->u.ldst.writeback = bit (insn, 25);
  dsc->u.ldst.rn = rn;

  dsc->modinsn[0] = insn & 0xfff0ffff;

  dsc->cleanup = &cleanup_copro_load_store;

  return 0;
}

/* Clean up branch instructions (actually perform the branch, by setting
   PC).  */

static void
cleanup_branch (struct gdbarch *gdbarch ATTRIBUTE_UNUSED, struct regcache *regs,
                struct displaced_step_closure *dsc)
{
  ULONGEST from = dsc->insn_addr;
  uint32_t status = displaced_read_reg (regs, from, ARM_PS_REGNUM);
  int branch_taken = condition_true (dsc->u.branch.cond, status);
  enum pc_write_style write_pc = dsc->u.branch.exchange
                                 ? BX_WRITE_PC : BRANCH_WRITE_PC;

  if (!branch_taken)
    return;

  if (dsc->u.branch.link)
    {
      ULONGEST pc = displaced_read_reg (regs, from, 15);
      displaced_write_reg (regs, dsc, 14, pc - 4, CANNOT_WRITE_PC);
    }

  displaced_write_reg (regs, dsc, 15, dsc->u.branch.dest, write_pc);
}

/* Copy B/BL/BLX instructions with immediate destinations.  */

static int
copy_b_bl_blx (struct gdbarch *gdbarch ATTRIBUTE_UNUSED, uint32_t insn,
               struct regcache *regs, struct displaced_step_closure *dsc)
{
  unsigned int cond = bits (insn, 28, 31);
  int exchange = (cond == 0xf);
  int link = exchange || bit (insn, 24);
  CORE_ADDR from = dsc->insn_addr;
  long offset;

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying %s immediate insn "
                        "%.8lx\n", (exchange) ? "blx" : (link) ? "bl" : "b",
                        (unsigned long) insn);

  /* Implement "BL<cond> <label>" as:

     Preparation: cond <- instruction condition
     Insn: mov r0, r0  (nop)
     Cleanup: if (condition true) { r14 <- pc; pc <- label }.

     B<cond> similar, but don't set r14 in cleanup.  */

  if (exchange)
    /* For BLX, set bit 0 of the destination.  The cleanup_branch function will
       then arrange the switch into Thumb mode.  */
    offset = (bits (insn, 0, 23) << 2) | (bit (insn, 24) << 1) | 1;
  else
    offset = bits (insn, 0, 23) << 2;

  if (bit (offset, 25))
    offset = offset | ~0x3ffffff;

  dsc->u.branch.cond = cond;
  dsc->u.branch.link = link;
  dsc->u.branch.exchange = exchange;
  dsc->u.branch.dest = from + 8 + offset;

  dsc->modinsn[0] = ARM_NOP;

  dsc->cleanup = &cleanup_branch;

  return 0;
}

/* Copy BX/BLX with register-specified destinations.  */

static int
copy_bx_blx_reg (struct gdbarch *gdbarch ATTRIBUTE_UNUSED, uint32_t insn,
                 struct regcache *regs, struct displaced_step_closure *dsc)
{
  unsigned int cond = bits (insn, 28, 31);
  /* BX:  x12xxx1x
     BLX: x12xxx3x.  */
  int link = bit (insn, 5);
  unsigned int rm = bits (insn, 0, 3);
  CORE_ADDR from = dsc->insn_addr;

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying %s register insn "
                        "%.8lx\n", (link) ? "blx" : "bx", (unsigned long) insn);

  /* Implement {BX,BLX}<cond> <reg>" as:

     Preparation: cond <- instruction condition
     Insn: mov r0, r0 (nop)
     Cleanup: if (condition true) { r14 <- pc; pc <- dest; }.

     Don't set r14 in cleanup for BX.  */

  dsc->u.branch.dest = displaced_read_reg (regs, from, rm);

  dsc->u.branch.cond = cond;
  dsc->u.branch.link = link;
  dsc->u.branch.exchange = 1;

  dsc->modinsn[0] = ARM_NOP;

  dsc->cleanup = &cleanup_branch;

  return 0;
}

/* Copy/cleanup arithmetic/logic instruction with immediate RHS. */

static void
cleanup_alu_imm (struct gdbarch *gdbarch ATTRIBUTE_UNUSED,
                 struct regcache *regs, struct displaced_step_closure *dsc)
{
  ULONGEST rd_val = displaced_read_reg (regs, dsc->insn_addr, 0);
  displaced_write_reg (regs, dsc, 0, dsc->tmp[0], CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, 1, dsc->tmp[1], CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, dsc->rd, rd_val, ALU_WRITE_PC);
}

static int
copy_alu_imm (struct gdbarch *gdbarch, uint32_t insn, struct regcache *regs,
              struct displaced_step_closure *dsc)
{
  unsigned int rn = bits (insn, 16, 19);
  unsigned int rd = bits (insn, 12, 15);
  unsigned int op = bits (insn, 21, 24);
  int is_mov = (op == 0xd);
  ULONGEST rd_val, rn_val;
  CORE_ADDR from = dsc->insn_addr;

  if (!insn_references_pc (insn, 0x000ff000ul))
    return copy_unmodified (gdbarch, insn, "ALU immediate", dsc);

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying immediate %s insn "
                        "%.8lx\n", is_mov ? "move" : "ALU",
                        (unsigned long) insn);

  /* Instruction is of form:

     <op><cond> rd, [rn,] #imm

     Rewrite as:

     Preparation: tmp1, tmp2 <- r0, r1;
                  r0, r1 <- rd, rn
     Insn: <op><cond> r0, r1, #imm
     Cleanup: rd <- r0; r0 <- tmp1; r1 <- tmp2
  */

  dsc->tmp[0] = displaced_read_reg (regs, from, 0);
  dsc->tmp[1] = displaced_read_reg (regs, from, 1);
  rn_val = displaced_read_reg (regs, from, rn);
  rd_val = displaced_read_reg (regs, from, rd);
  displaced_write_reg (regs, dsc, 0, rd_val, CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, 1, rn_val, CANNOT_WRITE_PC);
  dsc->rd = rd;

  if (is_mov)
    dsc->modinsn[0] = insn & 0xfff00fff;
  else
    dsc->modinsn[0] = (insn & 0xfff00fff) | 0x10000;

  dsc->cleanup = &cleanup_alu_imm;

  return 0;
}

/* Copy/cleanup arithmetic/logic insns with register RHS.  */

static void
cleanup_alu_reg (struct gdbarch *gdbarch ATTRIBUTE_UNUSED,
                 struct regcache *regs, struct displaced_step_closure *dsc)
{
  ULONGEST rd_val;
  int i;

  rd_val = displaced_read_reg (regs, dsc->insn_addr, 0);

  for (i = 0; i < 3; i++)
    displaced_write_reg (regs, dsc, i, dsc->tmp[i], CANNOT_WRITE_PC);

  displaced_write_reg (regs, dsc, dsc->rd, rd_val, ALU_WRITE_PC);
}

static int
copy_alu_reg (struct gdbarch *gdbarch, uint32_t insn, struct regcache *regs,
              struct displaced_step_closure *dsc)
{
  unsigned int rn = bits (insn, 16, 19);
  unsigned int rm = bits (insn, 0, 3);
  unsigned int rd = bits (insn, 12, 15);
  unsigned int op = bits (insn, 21, 24);
  int is_mov = (op == 0xd);
  ULONGEST rd_val, rn_val, rm_val;
  CORE_ADDR from = dsc->insn_addr;

  if (!insn_references_pc (insn, 0x000ff00ful))
    return copy_unmodified (gdbarch, insn, "ALU reg", dsc);

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying reg %s insn %.8lx\n",
                        is_mov ? "move" : "ALU", (unsigned long) insn);

  /* Instruction is of form:

     <op><cond> rd, [rn,] rm [, <shift>]

     Rewrite as:

     Preparation: tmp1, tmp2, tmp3 <- r0, r1, r2;
                  r0, r1, r2 <- rd, rn, rm
     Insn: <op><cond> r0, r1, r2 [, <shift>]
     Cleanup: rd <- r0; r0, r1, r2 <- tmp1, tmp2, tmp3
  */

  dsc->tmp[0] = displaced_read_reg (regs, from, 0);
  dsc->tmp[1] = displaced_read_reg (regs, from, 1);
  dsc->tmp[2] = displaced_read_reg (regs, from, 2);
  rd_val = displaced_read_reg (regs, from, rd);
  rn_val = displaced_read_reg (regs, from, rn);
  rm_val = displaced_read_reg (regs, from, rm);
  displaced_write_reg (regs, dsc, 0, rd_val, CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, 1, rn_val, CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, 2, rm_val, CANNOT_WRITE_PC);
  dsc->rd = rd;

  if (is_mov)
    dsc->modinsn[0] = (insn & 0xfff00ff0) | 0x2;
  else
    dsc->modinsn[0] = (insn & 0xfff00ff0) | 0x10002;

  dsc->cleanup = &cleanup_alu_reg;

  return 0;
}

/* Cleanup/copy arithmetic/logic insns with shifted register RHS.  */

static void
cleanup_alu_shifted_reg (struct gdbarch *gdbarch ATTRIBUTE_UNUSED,
                         struct regcache *regs,
                         struct displaced_step_closure *dsc)
{
  ULONGEST rd_val = displaced_read_reg (regs, dsc->insn_addr, 0);
  int i;

  for (i = 0; i < 4; i++)
    displaced_write_reg (regs, dsc, i, dsc->tmp[i], CANNOT_WRITE_PC);

  displaced_write_reg (regs, dsc, dsc->rd, rd_val, ALU_WRITE_PC);
}

static int
copy_alu_shifted_reg (struct gdbarch *gdbarch, uint32_t insn,
                      struct regcache *regs, struct displaced_step_closure *dsc)
{
  unsigned int rn = bits (insn, 16, 19);
  unsigned int rm = bits (insn, 0, 3);
  unsigned int rd = bits (insn, 12, 15);
  unsigned int rs = bits (insn, 8, 11);
  unsigned int op = bits (insn, 21, 24);
  int is_mov = (op == 0xd), i;
  ULONGEST rd_val, rn_val, rm_val, rs_val;
  CORE_ADDR from = dsc->insn_addr;

  if (!insn_references_pc (insn, 0x000fff0ful))
    return copy_unmodified (gdbarch, insn, "ALU shifted reg", dsc);

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying shifted reg %s insn "
                        "%.8lx\n", is_mov ? "move" : "ALU",
                        (unsigned long) insn);

  /* Instruction is of form:

     <op><cond> rd, [rn,] rm, <shift> rs

     Rewrite as:

     Preparation: tmp1, tmp2, tmp3, tmp4 <- r0, r1, r2, r3
                  r0, r1, r2, r3 <- rd, rn, rm, rs
     Insn: <op><cond> r0, r1, r2, <shift> r3
     Cleanup: tmp5 <- r0
              r0, r1, r2, r3 <- tmp1, tmp2, tmp3, tmp4
              rd <- tmp5
  */

  for (i = 0; i < 4; i++)
    dsc->tmp[i] = displaced_read_reg (regs, from, i);

  rd_val = displaced_read_reg (regs, from, rd);
  rn_val = displaced_read_reg (regs, from, rn);
  rm_val = displaced_read_reg (regs, from, rm);
  rs_val = displaced_read_reg (regs, from, rs);
  displaced_write_reg (regs, dsc, 0, rd_val, CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, 1, rn_val, CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, 2, rm_val, CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, 3, rs_val, CANNOT_WRITE_PC);
  dsc->rd = rd;

  if (is_mov)
    dsc->modinsn[0] = (insn & 0xfff000f0) | 0x302;
  else
    dsc->modinsn[0] = (insn & 0xfff000f0) | 0x10302;

  dsc->cleanup = &cleanup_alu_shifted_reg;

  return 0;
}

/* Clean up load instructions.  */

static void
cleanup_load (struct gdbarch *gdbarch ATTRIBUTE_UNUSED, struct regcache *regs,
              struct displaced_step_closure *dsc)
{
  ULONGEST rt_val, rt_val2 = 0, rn_val;
  CORE_ADDR from = dsc->insn_addr;

  rt_val = displaced_read_reg (regs, from, 0);
  if (dsc->u.ldst.xfersize == 8)
    rt_val2 = displaced_read_reg (regs, from, 1);
  rn_val = displaced_read_reg (regs, from, 2);

  displaced_write_reg (regs, dsc, 0, dsc->tmp[0], CANNOT_WRITE_PC);
  if (dsc->u.ldst.xfersize > 4)
    displaced_write_reg (regs, dsc, 1, dsc->tmp[1], CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, 2, dsc->tmp[2], CANNOT_WRITE_PC);
  if (!dsc->u.ldst.immed)
    displaced_write_reg (regs, dsc, 3, dsc->tmp[3], CANNOT_WRITE_PC);

  /* Handle register writeback.  */
  if (dsc->u.ldst.writeback)
    displaced_write_reg (regs, dsc, dsc->u.ldst.rn, rn_val, CANNOT_WRITE_PC);
  /* Put result in right place.  */
  displaced_write_reg (regs, dsc, dsc->rd, rt_val, LOAD_WRITE_PC);
  if (dsc->u.ldst.xfersize == 8)
    displaced_write_reg (regs, dsc, dsc->rd + 1, rt_val2, LOAD_WRITE_PC);
}

/* Clean up store instructions.  */

static void
cleanup_store (struct gdbarch *gdbarch ATTRIBUTE_UNUSED, struct regcache *regs,
               struct displaced_step_closure *dsc)
{
  CORE_ADDR from = dsc->insn_addr;
  ULONGEST rn_val = displaced_read_reg (regs, from, 2);

  displaced_write_reg (regs, dsc, 0, dsc->tmp[0], CANNOT_WRITE_PC);
  if (dsc->u.ldst.xfersize > 4)
    displaced_write_reg (regs, dsc, 1, dsc->tmp[1], CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, 2, dsc->tmp[2], CANNOT_WRITE_PC);
  if (!dsc->u.ldst.immed)
    displaced_write_reg (regs, dsc, 3, dsc->tmp[3], CANNOT_WRITE_PC);
  if (!dsc->u.ldst.restore_r4)
    displaced_write_reg (regs, dsc, 4, dsc->tmp[4], CANNOT_WRITE_PC);

  /* Writeback.  */
  if (dsc->u.ldst.writeback)
    displaced_write_reg (regs, dsc, dsc->u.ldst.rn, rn_val, CANNOT_WRITE_PC);
}

/* Copy "extra" load/store instructions.  These are halfword/doubleword
   transfers, which have a different encoding to byte/word transfers.  */

static int
copy_extra_ld_st (struct gdbarch *gdbarch, uint32_t insn, int unpriveleged,
                  struct regcache *regs, struct displaced_step_closure *dsc)
{
  unsigned int op1 = bits (insn, 20, 24);
  unsigned int op2 = bits (insn, 5, 6);
  unsigned int rt = bits (insn, 12, 15);
  unsigned int rn = bits (insn, 16, 19);
  unsigned int rm = bits (insn, 0, 3);
  char load[12]     = {0, 1, 0, 1, 1, 1, 1, 1, 0, 1, 0, 1};
  char bytesize[12] = {2, 2, 2, 2, 8, 1, 8, 1, 8, 2, 8, 2};
  int immed = (op1 & 0x4) != 0;
  int opcode;
  ULONGEST rt_val, rt_val2 = 0, rn_val, rm_val = 0;
  CORE_ADDR from = dsc->insn_addr;

  if (!insn_references_pc (insn, 0x000ff00ful))
    return copy_unmodified (gdbarch, insn, "extra load/store", dsc);

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying %sextra load/store "
                        "insn %.8lx\n", unpriveleged ? "unpriveleged " : "",
                        (unsigned long) insn);

  opcode = ((op2 << 2) | (op1 & 0x1) | ((op1 & 0x4) >> 1)) - 4;

  if (opcode < 0)
    internal_error (__FILE__, __LINE__,
                    _("copy_extra_ld_st: instruction decode error"));

  dsc->tmp[0] = displaced_read_reg (regs, from, 0);
  dsc->tmp[1] = displaced_read_reg (regs, from, 1);
  dsc->tmp[2] = displaced_read_reg (regs, from, 2);
  if (!immed)
    dsc->tmp[3] = displaced_read_reg (regs, from, 3);

  rt_val = displaced_read_reg (regs, from, rt);
  if (bytesize[opcode] == 8)
    rt_val2 = displaced_read_reg (regs, from, rt + 1);
  rn_val = displaced_read_reg (regs, from, rn);
  if (!immed)
    rm_val = displaced_read_reg (regs, from, rm);

  displaced_write_reg (regs, dsc, 0, rt_val, CANNOT_WRITE_PC);
  if (bytesize[opcode] == 8)
    displaced_write_reg (regs, dsc, 1, rt_val2, CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, 2, rn_val, CANNOT_WRITE_PC);
  if (!immed)
    displaced_write_reg (regs, dsc, 3, rm_val, CANNOT_WRITE_PC);

  dsc->rd = rt;
  dsc->u.ldst.xfersize = bytesize[opcode];
  dsc->u.ldst.rn = rn;
  dsc->u.ldst.immed = immed;
  dsc->u.ldst.writeback = bit (insn, 24) == 0 || bit (insn, 21) != 0;
  dsc->u.ldst.restore_r4 = 0;

  if (immed)
    /* {ldr,str}<width><cond> rt, [rt2,] [rn, #imm]
        ->
       {ldr,str}<width><cond> r0, [r1,] [r2, #imm].  */
    dsc->modinsn[0] = (insn & 0xfff00fff) | 0x20000;
  else
    /* {ldr,str}<width><cond> rt, [rt2,] [rn, +/-rm]
        ->
       {ldr,str}<width><cond> r0, [r1,] [r2, +/-r3].  */
    dsc->modinsn[0] = (insn & 0xfff00ff0) | 0x20003;

  dsc->cleanup = load[opcode] ? &cleanup_load : &cleanup_store;

  return 0;
}

/* Copy byte/word loads and stores.  */

static int
copy_ldr_str_ldrb_strb (struct gdbarch *gdbarch, uint32_t insn,
                        struct regcache *regs,
                        struct displaced_step_closure *dsc, int load, int byte,
                        int usermode)
{
  int immed = !bit (insn, 25);
  unsigned int rt = bits (insn, 12, 15);
  unsigned int rn = bits (insn, 16, 19);
  unsigned int rm = bits (insn, 0, 3);  /* Only valid if !immed.  */
  ULONGEST rt_val, rn_val, rm_val = 0;
  CORE_ADDR from = dsc->insn_addr;

  if (!insn_references_pc (insn, 0x000ff00ful))
    return copy_unmodified (gdbarch, insn, "load/store", dsc);

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying %s%s insn %.8lx\n",
                        load ? (byte ? "ldrb" : "ldr")
                             : (byte ? "strb" : "str"), usermode ? "t" : "",
                        (unsigned long) insn);

  dsc->tmp[0] = displaced_read_reg (regs, from, 0);
  dsc->tmp[2] = displaced_read_reg (regs, from, 2);
  if (!immed)
    dsc->tmp[3] = displaced_read_reg (regs, from, 3);
  if (!load)
    dsc->tmp[4] = displaced_read_reg (regs, from, 4);

  rt_val = displaced_read_reg (regs, from, rt);
  rn_val = displaced_read_reg (regs, from, rn);
  if (!immed)
    rm_val = displaced_read_reg (regs, from, rm);

  displaced_write_reg (regs, dsc, 0, rt_val, CANNOT_WRITE_PC);
  displaced_write_reg (regs, dsc, 2, rn_val, CANNOT_WRITE_PC);
  if (!immed)
    displaced_write_reg (regs, dsc, 3, rm_val, CANNOT_WRITE_PC);

  dsc->rd = rt;
  dsc->u.ldst.xfersize = byte ? 1 : 4;
  dsc->u.ldst.rn = rn;
  dsc->u.ldst.immed = immed;
  dsc->u.ldst.writeback = bit (insn, 24) == 0 || bit (insn, 21) != 0;

  /* To write PC we can do:

     scratch+0:  str pc, temp  (*temp = scratch + 8 + offset)
     scratch+4:  ldr r4, temp
     scratch+8:  sub r4, r4, pc  (r4 = scratch + 8 + offset - scratch - 8 - 8)
     scratch+12: add r4, r4, #8  (r4 = offset)
     scratch+16: add r0, r0, r4
     scratch+20: str r0, [r2, #imm] (or str r0, [r2, r3])
     scratch+24: <temp>

     Otherwise we don't know what value to write for PC, since the offset is
     architecture-dependent (sometimes PC+8, sometimes PC+12).  */

  if (load || rt != 15)
    {
      dsc->u.ldst.restore_r4 = 0;

      if (immed)
        /* {ldr,str}[b]<cond> rt, [rn, #imm], etc.
           ->
           {ldr,str}[b]<cond> r0, [r2, #imm].  */
        dsc->modinsn[0] = (insn & 0xfff00fff) | 0x20000;
      else
        /* {ldr,str}[b]<cond> rt, [rn, rm], etc.
           ->
           {ldr,str}[b]<cond> r0, [r2, r3].  */
        dsc->modinsn[0] = (insn & 0xfff00ff0) | 0x20003;
    }
  else
    {
      /* We need to use r4 as scratch.  Make sure it's restored afterwards.  */
      dsc->u.ldst.restore_r4 = 1;

      dsc->modinsn[0] = 0xe58ff014;  /* str pc, [pc, #20].  */
      dsc->modinsn[1] = 0xe59f4010;  /* ldr r4, [pc, #16].  */
      dsc->modinsn[2] = 0xe044400f;  /* sub r4, r4, pc.  */
      dsc->modinsn[3] = 0xe2844008;  /* add r4, r4, #8.  */
      dsc->modinsn[4] = 0xe0800004;  /* add r0, r0, r4.  */

      /* As above.  */
      if (immed)
        dsc->modinsn[5] = (insn & 0xfff00fff) | 0x20000;
      else
        dsc->modinsn[5] = (insn & 0xfff00ff0) | 0x20003;

      dsc->modinsn[6] = 0x0;  /* breakpoint location.  */
      dsc->modinsn[7] = 0x0;  /* scratch space.  */

      dsc->numinsns = 6;
    }

  dsc->cleanup = load ? &cleanup_load : &cleanup_store;

  return 0;
}

/* Cleanup LDM instructions with fully-populated register list.  This is an
   unfortunate corner case: it's impossible to implement correctly by modifying
   the instruction.  The issue is as follows: we have an instruction,

   ldm rN, {r0-r15}

   which we must rewrite to avoid loading PC.  A possible solution would be to
   do the load in two halves, something like (with suitable cleanup
   afterwards):

   mov r8, rN
   ldm[id][ab] r8!, {r0-r7}
   str r7, <temp>
   ldm[id][ab] r8, {r7-r14}
   <bkpt>

   but at present there's no suitable place for <temp>, since the scratch space
   is overwritten before the cleanup routine is called.  For now, we simply
   emulate the instruction.  */

static void
cleanup_block_load_all (struct gdbarch *gdbarch, struct regcache *regs,
                        struct displaced_step_closure *dsc)
{
  ULONGEST from = dsc->insn_addr;
  int inc = dsc->u.block.increment;
  int bump_before = dsc->u.block.before ? (inc ? 4 : -4) : 0;
  int bump_after = dsc->u.block.before ? 0 : (inc ? 4 : -4);
  uint32_t regmask = dsc->u.block.regmask;
  int regno = inc ? 0 : 15;
  CORE_ADDR xfer_addr = dsc->u.block.xfer_addr;
  int exception_return = dsc->u.block.load && dsc->u.block.user
                         && (regmask & 0x8000) != 0;
  uint32_t status = displaced_read_reg (regs, from, ARM_PS_REGNUM);
  int do_transfer = condition_true (dsc->u.block.cond, status);
  enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);

  if (!do_transfer)
    return;

  /* If the instruction is ldm rN, {...pc}^, I don't think there's anything
     sensible we can do here.  Complain loudly.  */
  if (exception_return)
    error (_("Cannot single-step exception return"));

  /* We don't handle any stores here for now.  */
  gdb_assert (dsc->u.block.load != 0);

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: emulating block transfer: "
                        "%s %s %s\n", dsc->u.block.load ? "ldm" : "stm",
                        dsc->u.block.increment ? "inc" : "dec",
                        dsc->u.block.before ? "before" : "after");

  while (regmask)
    {
      uint32_t memword;

      if (inc)
        while (regno <= 15 && (regmask & (1 << regno)) == 0)
          regno++;
      else
        while (regno >= 0 && (regmask & (1 << regno)) == 0)
          regno--;

      xfer_addr += bump_before;

      memword = read_memory_unsigned_integer (xfer_addr, 4, byte_order);
      displaced_write_reg (regs, dsc, regno, memword, LOAD_WRITE_PC);

      xfer_addr += bump_after;

      regmask &= ~(1 << regno);
    }

  if (dsc->u.block.writeback)
    displaced_write_reg (regs, dsc, dsc->u.block.rn, xfer_addr,
                         CANNOT_WRITE_PC);
}

/* Clean up an STM which included the PC in the register list.  */

static void
cleanup_block_store_pc (struct gdbarch *gdbarch, struct regcache *regs,
                        struct displaced_step_closure *dsc)
{
  ULONGEST from = dsc->insn_addr;
  uint32_t status = displaced_read_reg (regs, from, ARM_PS_REGNUM);
  int store_executed = condition_true (dsc->u.block.cond, status);
  CORE_ADDR pc_stored_at, transferred_regs = bitcount (dsc->u.block.regmask);
  CORE_ADDR stm_insn_addr;
  uint32_t pc_val;
  long offset;
  enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);

  /* If condition code fails, there's nothing else to do.  */
  if (!store_executed)
    return;

  if (dsc->u.block.increment)
    {
      pc_stored_at = dsc->u.block.xfer_addr + 4 * transferred_regs;

      if (dsc->u.block.before)
         pc_stored_at += 4;
    }
  else
    {
      pc_stored_at = dsc->u.block.xfer_addr;

      if (dsc->u.block.before)
         pc_stored_at -= 4;
    }

  pc_val = read_memory_unsigned_integer (pc_stored_at, 4, byte_order);
  stm_insn_addr = dsc->scratch_base;
  offset = pc_val - stm_insn_addr;

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: detected PC offset %.8lx for "
                        "STM instruction\n", offset);

  /* Rewrite the stored PC to the proper value for the non-displaced original
     instruction.  */
  write_memory_unsigned_integer (pc_stored_at, 4, byte_order,
                                 dsc->insn_addr + offset);
}

/* Clean up an LDM which includes the PC in the register list.  We clumped all
   the registers in the transferred list into a contiguous range r0...rX (to
   avoid loading PC directly and losing control of the debugged program), so we
   must undo that here.  */

static void
cleanup_block_load_pc (struct gdbarch *gdbarch ATTRIBUTE_UNUSED,
                       struct regcache *regs,
                       struct displaced_step_closure *dsc)
{
  ULONGEST from = dsc->insn_addr;
  uint32_t status = displaced_read_reg (regs, from, ARM_PS_REGNUM);
  int load_executed = condition_true (dsc->u.block.cond, status), i;
  unsigned int mask = dsc->u.block.regmask, write_reg = 15;
  unsigned int regs_loaded = bitcount (mask);
  unsigned int num_to_shuffle = regs_loaded, clobbered;

  /* The method employed here will fail if the register list is fully populated
     (we need to avoid loading PC directly).  */
  gdb_assert (num_to_shuffle < 16);

  if (!load_executed)
    return;

  clobbered = (1 << num_to_shuffle) - 1;

  while (num_to_shuffle > 0)
    {
      if ((mask & (1 << write_reg)) != 0)
        {
          unsigned int read_reg = num_to_shuffle - 1;

          if (read_reg != write_reg)
            {
              ULONGEST rval = displaced_read_reg (regs, from, read_reg);
              displaced_write_reg (regs, dsc, write_reg, rval, LOAD_WRITE_PC);
              if (debug_displaced)
                fprintf_unfiltered (gdb_stdlog, _("displaced: LDM: move "
                                    "loaded register r%d to r%d\n"), read_reg,
                                    write_reg);
            }
          else if (debug_displaced)
            fprintf_unfiltered (gdb_stdlog, _("displaced: LDM: register "
                                "r%d already in the right place\n"),
                                write_reg);

          clobbered &= ~(1 << write_reg);

          num_to_shuffle--;
        }

      write_reg--;
    }

  /* Restore any registers we scribbled over.  */
  for (write_reg = 0; clobbered != 0; write_reg++)
    {
      if ((clobbered & (1 << write_reg)) != 0)
        {
          displaced_write_reg (regs, dsc, write_reg, dsc->tmp[write_reg],
                               CANNOT_WRITE_PC);
          if (debug_displaced)
            fprintf_unfiltered (gdb_stdlog, _("displaced: LDM: restored "
                                "clobbered register r%d\n"), write_reg);
          clobbered &= ~(1 << write_reg);
        }
    }

  /* Perform register writeback manually.  */
  if (dsc->u.block.writeback)
    {
      ULONGEST new_rn_val = dsc->u.block.xfer_addr;

      if (dsc->u.block.increment)
        new_rn_val += regs_loaded * 4;
      else
        new_rn_val -= regs_loaded * 4;

      displaced_write_reg (regs, dsc, dsc->u.block.rn, new_rn_val,
                           CANNOT_WRITE_PC);
    }
}

/* Handle ldm/stm, apart from some tricky cases which are unlikely to occur
   in user-level code (in particular exception return, ldm rn, {...pc}^).  */

static int
copy_block_xfer (struct gdbarch *gdbarch, uint32_t insn, struct regcache *regs,
                 struct displaced_step_closure *dsc)
{
  int load = bit (insn, 20);
  int user = bit (insn, 22);
  int increment = bit (insn, 23);
  int before = bit (insn, 24);
  int writeback = bit (insn, 21);
  int rn = bits (insn, 16, 19);
  CORE_ADDR from = dsc->insn_addr;

  /* Block transfers which don't mention PC can be run directly out-of-line.  */
  if (rn != 15 && (insn & 0x8000) == 0)
    return copy_unmodified (gdbarch, insn, "ldm/stm", dsc);

  if (rn == 15)
    {
      warning (_("displaced: Unpredictable LDM or STM with base register r15"));
      return copy_unmodified (gdbarch, insn, "unpredictable ldm/stm", dsc);
    }

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying block transfer insn "
                        "%.8lx\n", (unsigned long) insn);

  dsc->u.block.xfer_addr = displaced_read_reg (regs, from, rn);
  dsc->u.block.rn = rn;

  dsc->u.block.load = load;
  dsc->u.block.user = user;
  dsc->u.block.increment = increment;
  dsc->u.block.before = before;
  dsc->u.block.writeback = writeback;
  dsc->u.block.cond = bits (insn, 28, 31);

  dsc->u.block.regmask = insn & 0xffff;

  if (load)
    {
      if ((insn & 0xffff) == 0xffff)
        {
          /* LDM with a fully-populated register list.  This case is
             particularly tricky.  Implement for now by fully emulating the
             instruction (which might not behave perfectly in all cases, but
             these instructions should be rare enough for that not to matter
             too much).  */
          dsc->modinsn[0] = ARM_NOP;

          dsc->cleanup = &cleanup_block_load_all;
        }
      else
        {
          /* LDM of a list of registers which includes PC.  Implement by
             rewriting the list of registers to be transferred into a
             contiguous chunk r0...rX before doing the transfer, then shuffling
             registers into the correct places in the cleanup routine.  */
          unsigned int regmask = insn & 0xffff;
          unsigned int num_in_list = bitcount (regmask), new_regmask, bit = 1;
          unsigned int to = 0, from = 0, i, new_rn;

          for (i = 0; i < num_in_list; i++)
            dsc->tmp[i] = displaced_read_reg (regs, from, i);

          /* Writeback makes things complicated.  We need to avoid clobbering
             the base register with one of the registers in our modified
             register list, but just using a different register can't work in
             all cases, e.g.:

               ldm r14!, {r0-r13,pc}

             which would need to be rewritten as:

               ldm rN!, {r0-r14}

             but that can't work, because there's no free register for N.

             Solve this by turning off the writeback bit, and emulating
             writeback manually in the cleanup routine.  */

          if (writeback)
            insn &= ~(1 << 21);

          new_regmask = (1 << num_in_list) - 1;

          if (debug_displaced)
            fprintf_unfiltered (gdb_stdlog, _("displaced: LDM r%d%s, "
                                "{..., pc}: original reg list %.4x, modified "
                                "list %.4x\n"), rn, writeback ? "!" : "",
                                (int) insn & 0xffff, new_regmask);

          dsc->modinsn[0] = (insn & ~0xffff) | (new_regmask & 0xffff);

          dsc->cleanup = &cleanup_block_load_pc;
        }
    }
  else
    {
      /* STM of a list of registers which includes PC.  Run the instruction
         as-is, but out of line: this will store the wrong value for the PC,
         so we must manually fix up the memory in the cleanup routine.
         Doing things this way has the advantage that we can auto-detect
         the offset of the PC write (which is architecture-dependent) in
         the cleanup routine.  */
      dsc->modinsn[0] = insn;

      dsc->cleanup = &cleanup_block_store_pc;
    }

  return 0;
}

/* Cleanup/copy SVC (SWI) instructions.  These two functions are overridden
   for Linux, where some SVC instructions must be treated specially.  */

static void
cleanup_svc (struct gdbarch *gdbarch ATTRIBUTE_UNUSED, struct regcache *regs,
             struct displaced_step_closure *dsc)
{
  CORE_ADDR from = dsc->insn_addr;
  CORE_ADDR resume_addr = from + 4;

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: cleanup for svc, resume at "
                        "%.8lx\n", (unsigned long) resume_addr);

  displaced_write_reg (regs, dsc, ARM_PC_REGNUM, resume_addr, BRANCH_WRITE_PC);
}

static int
copy_svc (struct gdbarch *gdbarch, uint32_t insn, CORE_ADDR to,
          struct regcache *regs, struct displaced_step_closure *dsc)
{
  CORE_ADDR from = dsc->insn_addr;

  /* Allow OS-specific code to override SVC handling.  */
  if (dsc->u.svc.copy_svc_os)
    return dsc->u.svc.copy_svc_os (gdbarch, insn, to, regs, dsc);

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying svc insn %.8lx\n",
                        (unsigned long) insn);

  /* Preparation: none.
     Insn: unmodified svc.
     Cleanup: pc <- insn_addr + 4.  */

  dsc->modinsn[0] = insn;

  dsc->cleanup = &cleanup_svc;
  /* Pretend we wrote to the PC, so cleanup doesn't set PC to the next
     instruction.  */
  dsc->wrote_to_pc = 1;

  return 0;
}

/* Copy undefined instructions.  */

static int
copy_undef (struct gdbarch *gdbarch ATTRIBUTE_UNUSED, uint32_t insn,
            struct displaced_step_closure *dsc)
{
  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying undefined insn %.8lx\n",
                        (unsigned long) insn);

  dsc->modinsn[0] = insn;

  return 0;
}

/* Copy unpredictable instructions.  */

static int
copy_unpred (struct gdbarch *gdbarch ATTRIBUTE_UNUSED, uint32_t insn,
             struct displaced_step_closure *dsc)
{
  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copying unpredictable insn "
                        "%.8lx\n", (unsigned long) insn);

  dsc->modinsn[0] = insn;

  return 0;
}

/* The decode_* functions are instruction decoding helpers.  They mostly follow
   the presentation in the ARM ARM.  */

static int
decode_misc_memhint_neon (struct gdbarch *gdbarch, uint32_t insn,
                          struct regcache *regs,
                          struct displaced_step_closure *dsc)
{
  unsigned int op1 = bits (insn, 20, 26), op2 = bits (insn, 4, 7);
  unsigned int rn = bits (insn, 16, 19);

  if (op1 == 0x10 && (op2 & 0x2) == 0x0 && (rn & 0xe) == 0x0)
    return copy_unmodified (gdbarch, insn, "cps", dsc);
  else if (op1 == 0x10 && op2 == 0x0 && (rn & 0xe) == 0x1)
    return copy_unmodified (gdbarch, insn, "setend", dsc);
  else if ((op1 & 0x60) == 0x20)
    return copy_unmodified (gdbarch, insn, "neon dataproc", dsc);
  else if ((op1 & 0x71) == 0x40)
    return copy_unmodified (gdbarch, insn, "neon elt/struct load/store", dsc);
  else if ((op1 & 0x77) == 0x41)
    return copy_unmodified (gdbarch, insn, "unallocated mem hint", dsc);
  else if ((op1 & 0x77) == 0x45)
    return copy_preload (gdbarch, insn, regs, dsc);  /* pli.  */
  else if ((op1 & 0x77) == 0x51)
    {
      if (rn != 0xf)
        return copy_preload (gdbarch, insn, regs, dsc);  /* pld/pldw.  */
      else
        return copy_unpred (gdbarch, insn, dsc);
    }
  else if ((op1 & 0x77) == 0x55)
    return copy_preload (gdbarch, insn, regs, dsc);  /* pld/pldw.  */
  else if (op1 == 0x57)
    switch (op2)
      {
      case 0x1: return copy_unmodified (gdbarch, insn, "clrex", dsc);
      case 0x4: return copy_unmodified (gdbarch, insn, "dsb", dsc);
      case 0x5: return copy_unmodified (gdbarch, insn, "dmb", dsc);
      case 0x6: return copy_unmodified (gdbarch, insn, "isb", dsc);
      default: return copy_unpred (gdbarch, insn, dsc);
      }
  else if ((op1 & 0x63) == 0x43)
    return copy_unpred (gdbarch, insn, dsc);
  else if ((op2 & 0x1) == 0x0)
    switch (op1 & ~0x80)
      {
      case 0x61:
        return copy_unmodified (gdbarch, insn, "unallocated mem hint", dsc);
      case 0x65:
        return copy_preload_reg (gdbarch, insn, regs, dsc);  /* pli reg.  */
      case 0x71: case 0x75:
        /* pld/pldw reg.  */
        return copy_preload_reg (gdbarch, insn, regs, dsc);
      case 0x63: case 0x67: case 0x73: case 0x77:
        return copy_unpred (gdbarch, insn, dsc);
      default:
        return copy_undef (gdbarch, insn, dsc);
      }
  else
    return copy_undef (gdbarch, insn, dsc);  /* Probably unreachable.  */
}

static int
decode_unconditional (struct gdbarch *gdbarch, uint32_t insn,
                      struct regcache *regs, struct displaced_step_closure *dsc)
{
  if (bit (insn, 27) == 0)
    return decode_misc_memhint_neon (gdbarch, insn, regs, dsc);
  /* Switch on bits: 0bxxxxx321xxx0xxxxxxxxxxxxxxxxxxxx.  */
  else switch (((insn & 0x7000000) >> 23) | ((insn & 0x100000) >> 20))
    {
    case 0x0: case 0x2:
      return copy_unmodified (gdbarch, insn, "srs", dsc);

    case 0x1: case 0x3:
      return copy_unmodified (gdbarch, insn, "rfe", dsc);

    case 0x4: case 0x5: case 0x6: case 0x7:
      return copy_b_bl_blx (gdbarch, insn, regs, dsc);

    case 0x8:
      switch ((insn & 0xe00000) >> 21)
        {
        case 0x1: case 0x3: case 0x4: case 0x5: case 0x6: case 0x7:
          /* stc/stc2.  */
          return copy_copro_load_store (gdbarch, insn, regs, dsc);

        case 0x2:
          return copy_unmodified (gdbarch, insn, "mcrr/mcrr2", dsc);

        default:
          return copy_undef (gdbarch, insn, dsc);
        }

    case 0x9:
      {
         int rn_f = (bits (insn, 16, 19) == 0xf);
        switch ((insn & 0xe00000) >> 21)
          {
          case 0x1: case 0x3:
            /* ldc/ldc2 imm (undefined for rn == pc).  */
            return rn_f ? copy_undef (gdbarch, insn, dsc)
                        : copy_copro_load_store (gdbarch, insn, regs, dsc);

          case 0x2:
            return copy_unmodified (gdbarch, insn, "mrrc/mrrc2", dsc);

          case 0x4: case 0x5: case 0x6: case 0x7:
            /* ldc/ldc2 lit (undefined for rn != pc).  */
            return rn_f ? copy_copro_load_store (gdbarch, insn, regs, dsc)
                        : copy_undef (gdbarch, insn, dsc);

          default:
            return copy_undef (gdbarch, insn, dsc);
          }
      }

    case 0xa:
      return copy_unmodified (gdbarch, insn, "stc/stc2", dsc);

    case 0xb:
      if (bits (insn, 16, 19) == 0xf)
        /* ldc/ldc2 lit.  */
        return copy_copro_load_store (gdbarch, insn, regs, dsc);
      else
        return copy_undef (gdbarch, insn, dsc);

    case 0xc:
      if (bit (insn, 4))
        return copy_unmodified (gdbarch, insn, "mcr/mcr2", dsc);
      else
        return copy_unmodified (gdbarch, insn, "cdp/cdp2", dsc);

    case 0xd:
      if (bit (insn, 4))
        return copy_unmodified (gdbarch, insn, "mrc/mrc2", dsc);
      else
        return copy_unmodified (gdbarch, insn, "cdp/cdp2", dsc);

    default:
      return copy_undef (gdbarch, insn, dsc);
    }
}

/* Decode miscellaneous instructions in dp/misc encoding space.  */

static int
decode_miscellaneous (struct gdbarch *gdbarch, uint32_t insn,
                      struct regcache *regs, struct displaced_step_closure *dsc)
{
  unsigned int op2 = bits (insn, 4, 6);
  unsigned int op = bits (insn, 21, 22);
  unsigned int op1 = bits (insn, 16, 19);

  switch (op2)
    {
    case 0x0:
      return copy_unmodified (gdbarch, insn, "mrs/msr", dsc);

    case 0x1:
      if (op == 0x1)  /* bx.  */
        return copy_bx_blx_reg (gdbarch, insn, regs, dsc);
      else if (op == 0x3)
        return copy_unmodified (gdbarch, insn, "clz", dsc);
      else
        return copy_undef (gdbarch, insn, dsc);

    case 0x2:
      if (op == 0x1)
        /* Not really supported.  */
        return copy_unmodified (gdbarch, insn, "bxj", dsc);
      else
        return copy_undef (gdbarch, insn, dsc);

    case 0x3:
      if (op == 0x1)
        return copy_bx_blx_reg (gdbarch, insn, regs, dsc);  /* blx register.  */
      else
        return copy_undef (gdbarch, insn, dsc);

    case 0x5:
      return copy_unmodified (gdbarch, insn, "saturating add/sub", dsc);

    case 0x7:
      if (op == 0x1)
        return copy_unmodified (gdbarch, insn, "bkpt", dsc);
      else if (op == 0x3)
        /* Not really supported.  */
        return copy_unmodified (gdbarch, insn, "smc", dsc);

    default:
      return copy_undef (gdbarch, insn, dsc);
    }
}

static int
decode_dp_misc (struct gdbarch *gdbarch, uint32_t insn, struct regcache *regs,
                struct displaced_step_closure *dsc)
{
  if (bit (insn, 25))
    switch (bits (insn, 20, 24))
      {
      case 0x10:
        return copy_unmodified (gdbarch, insn, "movw", dsc);

      case 0x14:
        return copy_unmodified (gdbarch, insn, "movt", dsc);

      case 0x12: case 0x16:
        return copy_unmodified (gdbarch, insn, "msr imm", dsc);

      default:
        return copy_alu_imm (gdbarch, insn, regs, dsc);
      }
  else
    {
      uint32_t op1 = bits (insn, 20, 24), op2 = bits (insn, 4, 7);

      if ((op1 & 0x19) != 0x10 && (op2 & 0x1) == 0x0)
        return copy_alu_reg (gdbarch, insn, regs, dsc);
      else if ((op1 & 0x19) != 0x10 && (op2 & 0x9) == 0x1)
        return copy_alu_shifted_reg (gdbarch, insn, regs, dsc);
      else if ((op1 & 0x19) == 0x10 && (op2 & 0x8) == 0x0)
        return decode_miscellaneous (gdbarch, insn, regs, dsc);
      else if ((op1 & 0x19) == 0x10 && (op2 & 0x9) == 0x8)
        return copy_unmodified (gdbarch, insn, "halfword mul/mla", dsc);
      else if ((op1 & 0x10) == 0x00 && op2 == 0x9)
        return copy_unmodified (gdbarch, insn, "mul/mla", dsc);
      else if ((op1 & 0x10) == 0x10 && op2 == 0x9)
        return copy_unmodified (gdbarch, insn, "synch", dsc);
      else if (op2 == 0xb || (op2 & 0xd) == 0xd)
        /* 2nd arg means "unpriveleged".  */
        return copy_extra_ld_st (gdbarch, insn, (op1 & 0x12) == 0x02, regs,
                                 dsc);
    }

  /* Should be unreachable.  */
  return 1;
}

static int
decode_ld_st_word_ubyte (struct gdbarch *gdbarch, uint32_t insn,
                         struct regcache *regs,
                         struct displaced_step_closure *dsc)
{
  int a = bit (insn, 25), b = bit (insn, 4);
  uint32_t op1 = bits (insn, 20, 24);
  int rn_f = bits (insn, 16, 19) == 0xf;

  if ((!a && (op1 & 0x05) == 0x00 && (op1 & 0x17) != 0x02)
      || (a && (op1 & 0x05) == 0x00 && (op1 & 0x17) != 0x02 && !b))
    return copy_ldr_str_ldrb_strb (gdbarch, insn, regs, dsc, 0, 0, 0);
  else if ((!a && (op1 & 0x17) == 0x02)
            || (a && (op1 & 0x17) == 0x02 && !b))
    return copy_ldr_str_ldrb_strb (gdbarch, insn, regs, dsc, 0, 0, 1);
  else if ((!a && (op1 & 0x05) == 0x01 && (op1 & 0x17) != 0x03)
            || (a && (op1 & 0x05) == 0x01 && (op1 & 0x17) != 0x03 && !b))
    return copy_ldr_str_ldrb_strb (gdbarch, insn, regs, dsc, 1, 0, 0);
  else if ((!a && (op1 & 0x17) == 0x03)
           || (a && (op1 & 0x17) == 0x03 && !b))
    return copy_ldr_str_ldrb_strb (gdbarch, insn, regs, dsc, 1, 0, 1);
  else if ((!a && (op1 & 0x05) == 0x04 && (op1 & 0x17) != 0x06)
            || (a && (op1 & 0x05) == 0x04 && (op1 & 0x17) != 0x06 && !b))
    return copy_ldr_str_ldrb_strb (gdbarch, insn, regs, dsc, 0, 1, 0);
  else if ((!a && (op1 & 0x17) == 0x06)
           || (a && (op1 & 0x17) == 0x06 && !b))
    return copy_ldr_str_ldrb_strb (gdbarch, insn, regs, dsc, 0, 1, 1);
  else if ((!a && (op1 & 0x05) == 0x05 && (op1 & 0x17) != 0x07)
           || (a && (op1 & 0x05) == 0x05 && (op1 & 0x17) != 0x07 && !b))
    return copy_ldr_str_ldrb_strb (gdbarch, insn, regs, dsc, 1, 1, 0);
  else if ((!a && (op1 & 0x17) == 0x07)
           || (a && (op1 & 0x17) == 0x07 && !b))
    return copy_ldr_str_ldrb_strb (gdbarch, insn, regs, dsc, 1, 1, 1);

  /* Should be unreachable.  */
  return 1;
}

static int
decode_media (struct gdbarch *gdbarch, uint32_t insn,
              struct displaced_step_closure *dsc)
{
  switch (bits (insn, 20, 24))
    {
    case 0x00: case 0x01: case 0x02: case 0x03:
      return copy_unmodified (gdbarch, insn, "parallel add/sub signed", dsc);

    case 0x04: case 0x05: case 0x06: case 0x07:
      return copy_unmodified (gdbarch, insn, "parallel add/sub unsigned", dsc);

    case 0x08: case 0x09: case 0x0a: case 0x0b:
    case 0x0c: case 0x0d: case 0x0e: case 0x0f:
      return copy_unmodified (gdbarch, insn,
                              "decode/pack/unpack/saturate/reverse", dsc);

    case 0x18:
      if (bits (insn, 5, 7) == 0)  /* op2.  */
         {
          if (bits (insn, 12, 15) == 0xf)
            return copy_unmodified (gdbarch, insn, "usad8", dsc);
          else
            return copy_unmodified (gdbarch, insn, "usada8", dsc);
        }
      else
         return copy_undef (gdbarch, insn, dsc);

    case 0x1a: case 0x1b:
      if (bits (insn, 5, 6) == 0x2)  /* op2[1:0].  */
        return copy_unmodified (gdbarch, insn, "sbfx", dsc);
      else
        return copy_undef (gdbarch, insn, dsc);

    case 0x1c: case 0x1d:
      if (bits (insn, 5, 6) == 0x0)  /* op2[1:0].  */
         {
          if (bits (insn, 0, 3) == 0xf)
            return copy_unmodified (gdbarch, insn, "bfc", dsc);
          else
            return copy_unmodified (gdbarch, insn, "bfi", dsc);
        }
      else
        return copy_undef (gdbarch, insn, dsc);

    case 0x1e: case 0x1f:
      if (bits (insn, 5, 6) == 0x2)  /* op2[1:0].  */
        return copy_unmodified (gdbarch, insn, "ubfx", dsc);
      else
        return copy_undef (gdbarch, insn, dsc);
    }

  /* Should be unreachable.  */
  return 1;
}

static int
decode_b_bl_ldmstm (struct gdbarch *gdbarch, int32_t insn,
                    struct regcache *regs, struct displaced_step_closure *dsc)
{
  if (bit (insn, 25))
    return copy_b_bl_blx (gdbarch, insn, regs, dsc);
  else
    return copy_block_xfer (gdbarch, insn, regs, dsc);
}

static int
decode_ext_reg_ld_st (struct gdbarch *gdbarch, uint32_t insn,
                      struct regcache *regs, struct displaced_step_closure *dsc)
{
  unsigned int opcode = bits (insn, 20, 24);

  switch (opcode)
    {
    case 0x04: case 0x05:  /* VFP/Neon mrrc/mcrr.  */
      return copy_unmodified (gdbarch, insn, "vfp/neon mrrc/mcrr", dsc);

    case 0x08: case 0x0a: case 0x0c: case 0x0e:
    case 0x12: case 0x16:
      return copy_unmodified (gdbarch, insn, "vfp/neon vstm/vpush", dsc);

    case 0x09: case 0x0b: case 0x0d: case 0x0f:
    case 0x13: case 0x17:
      return copy_unmodified (gdbarch, insn, "vfp/neon vldm/vpop", dsc);

    case 0x10: case 0x14: case 0x18: case 0x1c:  /* vstr.  */
    case 0x11: case 0x15: case 0x19: case 0x1d:  /* vldr.  */
      /* Note: no writeback for these instructions.  Bit 25 will always be
         zero though (via caller), so the following works OK.  */
      return copy_copro_load_store (gdbarch, insn, regs, dsc);
    }

  /* Should be unreachable.  */
  return 1;
}

static int
decode_svc_copro (struct gdbarch *gdbarch, uint32_t insn, CORE_ADDR to,
                  struct regcache *regs, struct displaced_step_closure *dsc)
{
  unsigned int op1 = bits (insn, 20, 25);
  int op = bit (insn, 4);
  unsigned int coproc = bits (insn, 8, 11);
  unsigned int rn = bits (insn, 16, 19);

  if ((op1 & 0x20) == 0x00 && (op1 & 0x3a) != 0x00 && (coproc & 0xe) == 0xa)
    return decode_ext_reg_ld_st (gdbarch, insn, regs, dsc);
  else if ((op1 & 0x21) == 0x00 && (op1 & 0x3a) != 0x00
           && (coproc & 0xe) != 0xa)
    /* stc/stc2.  */
    return copy_copro_load_store (gdbarch, insn, regs, dsc);
  else if ((op1 & 0x21) == 0x01 && (op1 & 0x3a) != 0x00
           && (coproc & 0xe) != 0xa)
    /* ldc/ldc2 imm/lit.  */
    return copy_copro_load_store (gdbarch, insn, regs, dsc);
  else if ((op1 & 0x3e) == 0x00)
    return copy_undef (gdbarch, insn, dsc);
  else if ((op1 & 0x3e) == 0x04 && (coproc & 0xe) == 0xa)
    return copy_unmodified (gdbarch, insn, "neon 64bit xfer", dsc);
  else if (op1 == 0x04 && (coproc & 0xe) != 0xa)
    return copy_unmodified (gdbarch, insn, "mcrr/mcrr2", dsc);
  else if (op1 == 0x05 && (coproc & 0xe) != 0xa)
    return copy_unmodified (gdbarch, insn, "mrrc/mrrc2", dsc);
  else if ((op1 & 0x30) == 0x20 && !op)
    {
      if ((coproc & 0xe) == 0xa)
        return copy_unmodified (gdbarch, insn, "vfp dataproc", dsc);
      else
        return copy_unmodified (gdbarch, insn, "cdp/cdp2", dsc);
    }
  else if ((op1 & 0x30) == 0x20 && op)
    return copy_unmodified (gdbarch, insn, "neon 8/16/32 bit xfer", dsc);
  else if ((op1 & 0x31) == 0x20 && op && (coproc & 0xe) != 0xa)
    return copy_unmodified (gdbarch, insn, "mcr/mcr2", dsc);
  else if ((op1 & 0x31) == 0x21 && op && (coproc & 0xe) != 0xa)
    return copy_unmodified (gdbarch, insn, "mrc/mrc2", dsc);
  else if ((op1 & 0x30) == 0x30)
    return copy_svc (gdbarch, insn, to, regs, dsc);
  else
    return copy_undef (gdbarch, insn, dsc);  /* Possibly unreachable.  */
}

void
arm_process_displaced_insn (struct gdbarch *gdbarch, uint32_t insn,
                            CORE_ADDR from, CORE_ADDR to, struct regcache *regs,
                            struct displaced_step_closure *dsc)
{
  int err = 0;

  if (!displaced_in_arm_mode (regs))
    error (_("Displaced stepping is only supported in ARM mode"));

  /* Most displaced instructions use a 1-instruction scratch space, so set this
     here and override below if/when necessary.  */
  dsc->numinsns = 1;
  dsc->insn_addr = from;
  dsc->scratch_base = to;
  dsc->cleanup = NULL;
  dsc->wrote_to_pc = 0;

  if ((insn & 0xf0000000) == 0xf0000000)
    err = decode_unconditional (gdbarch, insn, regs, dsc);
  else switch (((insn & 0x10) >> 4) | ((insn & 0xe000000) >> 24))
    {
    case 0x0: case 0x1: case 0x2: case 0x3:
      err = decode_dp_misc (gdbarch, insn, regs, dsc);
      break;

    case 0x4: case 0x5: case 0x6:
      err = decode_ld_st_word_ubyte (gdbarch, insn, regs, dsc);
      break;

    case 0x7:
      err = decode_media (gdbarch, insn, dsc);
      break;

    case 0x8: case 0x9: case 0xa: case 0xb:
      err = decode_b_bl_ldmstm (gdbarch, insn, regs, dsc);
      break;

    case 0xc: case 0xd: case 0xe: case 0xf:
      err = decode_svc_copro (gdbarch, insn, to, regs, dsc);
      break;
    }

  if (err)
    internal_error (__FILE__, __LINE__,
                    _("arm_process_displaced_insn: Instruction decode error"));
}

/* Actually set up the scratch space for a displaced instruction.  */

void
arm_displaced_init_closure (struct gdbarch *gdbarch, CORE_ADDR from,
                            CORE_ADDR to, struct displaced_step_closure *dsc)
{
  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
  unsigned int i;
  enum bfd_endian byte_order_for_code = gdbarch_byte_order_for_code (gdbarch);

  /* Poke modified instruction(s).  */
  for (i = 0; i < dsc->numinsns; i++)
    {
      if (debug_displaced)
        fprintf_unfiltered (gdb_stdlog, "displaced: writing insn %.8lx at "
                            "%.8lx\n", (unsigned long) dsc->modinsn[i],
                            (unsigned long) to + i * 4);
      write_memory_unsigned_integer (to + i * 4, 4, byte_order_for_code,
                                     dsc->modinsn[i]);
    }

  /* Put breakpoint afterwards.  */
  write_memory (to + dsc->numinsns * 4, tdep->arm_breakpoint,
                tdep->arm_breakpoint_size);

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: copy %s->%s: ",
                        paddress (gdbarch, from), paddress (gdbarch, to));
}

/* Entry point for copying an instruction into scratch space for displaced
   stepping.  */

struct displaced_step_closure *
arm_displaced_step_copy_insn (struct gdbarch *gdbarch,
                              CORE_ADDR from, CORE_ADDR to,
                              struct regcache *regs)
{
  struct displaced_step_closure *dsc
    = xmalloc (sizeof (struct displaced_step_closure));
  enum bfd_endian byte_order_for_code = gdbarch_byte_order_for_code (gdbarch);
  uint32_t insn = read_memory_unsigned_integer (from, 4, byte_order_for_code);

  if (debug_displaced)
    fprintf_unfiltered (gdb_stdlog, "displaced: stepping insn %.8lx "
                        "at %.8lx\n", (unsigned long) insn,
                        (unsigned long) from);

  arm_process_displaced_insn (gdbarch, insn, from, to, regs, dsc);
  arm_displaced_init_closure (gdbarch, from, to, dsc);

  return dsc;
}

/* Entry point for cleaning things up after a displaced instruction has been
   single-stepped.  */

void
arm_displaced_step_fixup (struct gdbarch *gdbarch,
                          struct displaced_step_closure *dsc,
                          CORE_ADDR from, CORE_ADDR to,
                          struct regcache *regs)
{
  if (dsc->cleanup)
    dsc->cleanup (gdbarch, regs, dsc);

  if (!dsc->wrote_to_pc)
    regcache_cooked_write_unsigned (regs, ARM_PC_REGNUM, dsc->insn_addr + 4);
}

#include "bfd-in2.h"
#include "libcoff.h"

static int
gdb_print_insn_arm (bfd_vma memaddr, disassemble_info *info)
{
  if (arm_pc_is_thumb (memaddr))
    {
      static asymbol *asym;
      static combined_entry_type ce;
      static struct coff_symbol_struct csym;
      static struct bfd fake_bfd;
      static bfd_target fake_target;

      if (csym.native == NULL)
        {
          /* Create a fake symbol vector containing a Thumb symbol.
             This is solely so that the code in print_insn_little_arm() 
             and print_insn_big_arm() in opcodes/arm-dis.c will detect
             the presence of a Thumb symbol and switch to decoding
             Thumb instructions.  */

          fake_target.flavour = bfd_target_coff_flavour;
          fake_bfd.xvec = &fake_target;
          ce.u.syment.n_sclass = C_THUMBEXTFUNC;
          csym.native = &ce;
          csym.symbol.the_bfd = &fake_bfd;
          csym.symbol.name = "fake";
          asym = (asymbol *) & csym;
        }

      memaddr = UNMAKE_THUMB_ADDR (memaddr);
      info->symbols = &asym;
    }
  else
    info->symbols = NULL;

  if (info->endian == BFD_ENDIAN_BIG)
    return print_insn_big_arm (memaddr, info);
  else
    return print_insn_little_arm (memaddr, info);
}

/* The following define instruction sequences that will cause ARM
   cpu's to take an undefined instruction trap.  These are used to
   signal a breakpoint to GDB.
   
   The newer ARMv4T cpu's are capable of operating in ARM or Thumb
   modes.  A different instruction is required for each mode.  The ARM
   cpu's can also be big or little endian.  Thus four different
   instructions are needed to support all cases.
   
   Note: ARMv4 defines several new instructions that will take the
   undefined instruction trap.  ARM7TDMI is nominally ARMv4T, but does
   not in fact add the new instructions.  The new undefined
   instructions in ARMv4 are all instructions that had no defined
   behaviour in earlier chips.  There is no guarantee that they will
   raise an exception, but may be treated as NOP's.  In practice, it
   may only safe to rely on instructions matching:
   
   3 3 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1 
   1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
   C C C C 0 1 1 x x x x x x x x x x x x x x x x x x x x 1 x x x x
   
   Even this may only true if the condition predicate is true. The
   following use a condition predicate of ALWAYS so it is always TRUE.
   
   There are other ways of forcing a breakpoint.  GNU/Linux, RISC iX,
   and NetBSD all use a software interrupt rather than an undefined
   instruction to force a trap.  This can be handled by by the
   abi-specific code during establishment of the gdbarch vector.  */

#define ARM_LE_BREAKPOINT {0xFE,0xDE,0xFF,0xE7}
#define ARM_BE_BREAKPOINT {0xE7,0xFF,0xDE,0xFE}
#define THUMB_LE_BREAKPOINT {0xbe,0xbe}
#define THUMB_BE_BREAKPOINT {0xbe,0xbe}

static const char arm_default_arm_le_breakpoint[] = ARM_LE_BREAKPOINT;
static const char arm_default_arm_be_breakpoint[] = ARM_BE_BREAKPOINT;
static const char arm_default_thumb_le_breakpoint[] = THUMB_LE_BREAKPOINT;
static const char arm_default_thumb_be_breakpoint[] = THUMB_BE_BREAKPOINT;

/* Determine the type and size of breakpoint to insert at PCPTR.  Uses
   the program counter value to determine whether a 16-bit or 32-bit
   breakpoint should be used.  It returns a pointer to a string of
   bytes that encode a breakpoint instruction, stores the length of
   the string to *lenptr, and adjusts the program counter (if
   necessary) to point to the actual memory location where the
   breakpoint should be inserted.  */

static const unsigned char *
arm_breakpoint_from_pc (struct gdbarch *gdbarch, CORE_ADDR *pcptr, int *lenptr)
{
  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);

  if (arm_pc_is_thumb (*pcptr))
    {
      *pcptr = UNMAKE_THUMB_ADDR (*pcptr);
      *lenptr = tdep->thumb_breakpoint_size;
      return tdep->thumb_breakpoint;
    }
  else
    {
      *lenptr = tdep->arm_breakpoint_size;
      return tdep->arm_breakpoint;
    }
}

/* Extract from an array REGBUF containing the (raw) register state a
   function return value of type TYPE, and copy that, in virtual
   format, into VALBUF.  */

static void
arm_extract_return_value (struct type *type, struct regcache *regs,
                          gdb_byte *valbuf)
{
  struct gdbarch *gdbarch = get_regcache_arch (regs);
  enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);

  if (TYPE_CODE_FLT == TYPE_CODE (type))
    {
      switch (gdbarch_tdep (gdbarch)->fp_model)
        {
        case ARM_FLOAT_FPA:
          {
            /* The value is in register F0 in internal format.  We need to
               extract the raw value and then convert it to the desired
               internal type.  */
            bfd_byte tmpbuf[FP_REGISTER_SIZE];

            regcache_cooked_read (regs, ARM_F0_REGNUM, tmpbuf);
            convert_from_extended (floatformat_from_type (type), tmpbuf,
                                   valbuf, gdbarch_byte_order (gdbarch));
          }
          break;

        case ARM_FLOAT_SOFT_FPA:
        case ARM_FLOAT_SOFT_VFP:
          /* ARM_FLOAT_VFP can arise if this is a variadic function so
             not using the VFP ABI code.  */
        case ARM_FLOAT_VFP:
          regcache_cooked_read (regs, ARM_A1_REGNUM, valbuf);
          if (TYPE_LENGTH (type) > 4)
            regcache_cooked_read (regs, ARM_A1_REGNUM + 1,
                                  valbuf + INT_REGISTER_SIZE);
          break;

        default:
          internal_error
            (__FILE__, __LINE__,
             _("arm_extract_return_value: Floating point model not supported"));
          break;
        }
    }
  else if (TYPE_CODE (type) == TYPE_CODE_INT
           || TYPE_CODE (type) == TYPE_CODE_CHAR
           || TYPE_CODE (type) == TYPE_CODE_BOOL
           || TYPE_CODE (type) == TYPE_CODE_PTR
           || TYPE_CODE (type) == TYPE_CODE_REF
           || TYPE_CODE (type) == TYPE_CODE_ENUM)
    {
      /* If the the type is a plain integer, then the access is
         straight-forward.  Otherwise we have to play around a bit more.  */
      int len = TYPE_LENGTH (type);
      int regno = ARM_A1_REGNUM;
      ULONGEST tmp;

      while (len > 0)
        {
          /* By using store_unsigned_integer we avoid having to do
             anything special for small big-endian values.  */
          regcache_cooked_read_unsigned (regs, regno++, &tmp);
          store_unsigned_integer (valbuf, 
                                  (len > INT_REGISTER_SIZE
                                   ? INT_REGISTER_SIZE : len),
                                  byte_order, tmp);
          len -= INT_REGISTER_SIZE;
          valbuf += INT_REGISTER_SIZE;
        }
    }
  else
    {
      /* For a structure or union the behaviour is as if the value had
         been stored to word-aligned memory and then loaded into 
         registers with 32-bit load instruction(s).  */
      int len = TYPE_LENGTH (type);
      int regno = ARM_A1_REGNUM;
      bfd_byte tmpbuf[INT_REGISTER_SIZE];

      while (len > 0)
        {
          regcache_cooked_read (regs, regno++, tmpbuf);
          memcpy (valbuf, tmpbuf,
                  len > INT_REGISTER_SIZE ? INT_REGISTER_SIZE : len);
          len -= INT_REGISTER_SIZE;
          valbuf += INT_REGISTER_SIZE;
        }
    }
}


/* Will a function return an aggregate type in memory or in a
   register?  Return 0 if an aggregate type can be returned in a
   register, 1 if it must be returned in memory.  */

static int
arm_return_in_memory (struct gdbarch *gdbarch, struct type *type)
{
  int nRc;
  enum type_code code;

  CHECK_TYPEDEF (type);

  /* In the ARM ABI, "integer" like aggregate types are returned in
     registers.  For an aggregate type to be integer like, its size
     must be less than or equal to INT_REGISTER_SIZE and the
     offset of each addressable subfield must be zero.  Note that bit
     fields are not addressable, and all addressable subfields of
     unions always start at offset zero.

     This function is based on the behaviour of GCC 2.95.1.
     See: gcc/arm.c: arm_return_in_memory() for details.

     Note: All versions of GCC before GCC 2.95.2 do not set up the
     parameters correctly for a function returning the following
     structure: struct { float f;}; This should be returned in memory,
     not a register.  Richard Earnshaw sent me a patch, but I do not
     know of any way to detect if a function like the above has been
     compiled with the correct calling convention.  */

  /* All aggregate types that won't fit in a register must be returned
     in memory.  */
  if (TYPE_LENGTH (type) > INT_REGISTER_SIZE)
    {
      return 1;
    }

  /* The AAPCS says all aggregates not larger than a word are returned
     in a register.  */
  if (gdbarch_tdep (gdbarch)->arm_abi != ARM_ABI_APCS)
    return 0;

  /* The only aggregate types that can be returned in a register are
     structs and unions.  Arrays must be returned in memory.  */
  code = TYPE_CODE (type);
  if ((TYPE_CODE_STRUCT != code) && (TYPE_CODE_UNION != code))
    {
      return 1;
    }

  /* Assume all other aggregate types can be returned in a register.
     Run a check for structures, unions and arrays.  */
  nRc = 0;

  if ((TYPE_CODE_STRUCT == code) || (TYPE_CODE_UNION == code))
    {
      int i;
      /* Need to check if this struct/union is "integer" like.  For
         this to be true, its size must be less than or equal to
         INT_REGISTER_SIZE and the offset of each addressable
         subfield must be zero.  Note that bit fields are not
         addressable, and unions always start at offset zero.  If any
         of the subfields is a floating point type, the struct/union
         cannot be an integer type.  */

      /* For each field in the object, check:
         1) Is it FP? --> yes, nRc = 1;
         2) Is it addressable (bitpos != 0) and
         not packed (bitsize == 0)?
         --> yes, nRc = 1  
       */

      for (i = 0; i < TYPE_NFIELDS (type); i++)
        {
          enum type_code field_type_code;
          field_type_code = TYPE_CODE (check_typedef (TYPE_FIELD_TYPE (type, i)));

          /* Is it a floating point type field?  */
          if (field_type_code == TYPE_CODE_FLT)
            {
              nRc = 1;
              break;
            }

          /* If bitpos != 0, then we have to care about it.  */
          if (TYPE_FIELD_BITPOS (type, i) != 0)
            {
              /* Bitfields are not addressable.  If the field bitsize is 
                 zero, then the field is not packed.  Hence it cannot be
                 a bitfield or any other packed type.  */
              if (TYPE_FIELD_BITSIZE (type, i) == 0)
                {
                  nRc = 1;
                  break;
                }
            }
        }
    }

  return nRc;
}

/* Write into appropriate registers a function return value of type
   TYPE, given in virtual format.  */

static void
arm_store_return_value (struct type *type, struct regcache *regs,
                        const gdb_byte *valbuf)
{
  struct gdbarch *gdbarch = get_regcache_arch (regs);
  enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);

  if (TYPE_CODE (type) == TYPE_CODE_FLT)
    {
      char buf[MAX_REGISTER_SIZE];

      switch (gdbarch_tdep (gdbarch)->fp_model)
        {
        case ARM_FLOAT_FPA:

          convert_to_extended (floatformat_from_type (type), buf, valbuf,
                               gdbarch_byte_order (gdbarch));
          regcache_cooked_write (regs, ARM_F0_REGNUM, buf);
          break;

        case ARM_FLOAT_SOFT_FPA:
        case ARM_FLOAT_SOFT_VFP:
          /* ARM_FLOAT_VFP can arise if this is a variadic function so
             not using the VFP ABI code.  */
        case ARM_FLOAT_VFP:
          regcache_cooked_write (regs, ARM_A1_REGNUM, valbuf);
          if (TYPE_LENGTH (type) > 4)
            regcache_cooked_write (regs, ARM_A1_REGNUM + 1, 
                                   valbuf + INT_REGISTER_SIZE);
          break;

        default:
          internal_error
            (__FILE__, __LINE__,
             _("arm_store_return_value: Floating point model not supported"));
          break;
        }
    }
  else if (TYPE_CODE (type) == TYPE_CODE_INT
           || TYPE_CODE (type) == TYPE_CODE_CHAR
           || TYPE_CODE (type) == TYPE_CODE_BOOL
           || TYPE_CODE (type) == TYPE_CODE_PTR
           || TYPE_CODE (type) == TYPE_CODE_REF
           || TYPE_CODE (type) == TYPE_CODE_ENUM)
    {
      if (TYPE_LENGTH (type) <= 4)
        {
          /* Values of one word or less are zero/sign-extended and
             returned in r0.  */
          bfd_byte tmpbuf[INT_REGISTER_SIZE];
          LONGEST val = unpack_long (type, valbuf);

          store_signed_integer (tmpbuf, INT_REGISTER_SIZE, byte_order, val);
          regcache_cooked_write (regs, ARM_A1_REGNUM, tmpbuf);
        }
      else
        {
          /* Integral values greater than one word are stored in consecutive
             registers starting with r0.  This will always be a multiple of
             the regiser size.  */
          int len = TYPE_LENGTH (type);
          int regno = ARM_A1_REGNUM;

          while (len > 0)
            {
              regcache_cooked_write (regs, regno++, valbuf);
              len -= INT_REGISTER_SIZE;
              valbuf += INT_REGISTER_SIZE;
            }
        }
    }
  else
    {
      /* For a structure or union the behaviour is as if the value had
         been stored to word-aligned memory and then loaded into 
         registers with 32-bit load instruction(s).  */
      int len = TYPE_LENGTH (type);
      int regno = ARM_A1_REGNUM;
      bfd_byte tmpbuf[INT_REGISTER_SIZE];

      while (len > 0)
        {
          memcpy (tmpbuf, valbuf,
                  len > INT_REGISTER_SIZE ? INT_REGISTER_SIZE : len);
          regcache_cooked_write (regs, regno++, tmpbuf);
          len -= INT_REGISTER_SIZE;
          valbuf += INT_REGISTER_SIZE;
        }
    }
}


/* Handle function return values.  */

static enum return_value_convention
arm_return_value (struct gdbarch *gdbarch, struct type *func_type,
                  struct type *valtype, struct regcache *regcache,
                  gdb_byte *readbuf, const gdb_byte *writebuf)
{
  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
  enum arm_vfp_cprc_base_type vfp_base_type;
  int vfp_base_count;

  if (arm_vfp_abi_for_function (gdbarch, func_type)
      && arm_vfp_call_candidate (valtype, &vfp_base_type, &vfp_base_count))
    {
      int reg_char = arm_vfp_cprc_reg_char (vfp_base_type);
      int unit_length = arm_vfp_cprc_unit_length (vfp_base_type);
      int i;
      for (i = 0; i < vfp_base_count; i++)
        {
          if (reg_char == 'q')
            {
              if (writebuf)
                arm_neon_quad_write (gdbarch, regcache, i,
                                     writebuf + i * unit_length);

              if (readbuf)
                arm_neon_quad_read (gdbarch, regcache, i,
                                    readbuf + i * unit_length);
            }
          else
            {
              char name_buf[4];
              int regnum;

              sprintf (name_buf, "%c%d", reg_char, i);
              regnum = user_reg_map_name_to_regnum (gdbarch, name_buf,
                                                    strlen (name_buf));
              if (writebuf)
                regcache_cooked_write (regcache, regnum,
                                       writebuf + i * unit_length);
              if (readbuf)
                regcache_cooked_read (regcache, regnum,
                                      readbuf + i * unit_length);
            }
        }
      return RETURN_VALUE_REGISTER_CONVENTION;
    }

  if (TYPE_CODE (valtype) == TYPE_CODE_STRUCT
      || TYPE_CODE (valtype) == TYPE_CODE_UNION
      || TYPE_CODE (valtype) == TYPE_CODE_ARRAY)
    {
      if (tdep->struct_return == pcc_struct_return
          || arm_return_in_memory (gdbarch, valtype))
        return RETURN_VALUE_STRUCT_CONVENTION;
    }

  if (writebuf)
    arm_store_return_value (valtype, regcache, writebuf);

  if (readbuf)
    arm_extract_return_value (valtype, regcache, readbuf);

  return RETURN_VALUE_REGISTER_CONVENTION;
}


static int
arm_get_longjmp_target (struct frame_info *frame, CORE_ADDR *pc)
{
  struct gdbarch *gdbarch = get_frame_arch (frame);
  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
  enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
  CORE_ADDR jb_addr;
  char buf[INT_REGISTER_SIZE];
  
  jb_addr = get_frame_register_unsigned (frame, ARM_A1_REGNUM);

  if (target_read_memory (jb_addr + tdep->jb_pc * tdep->jb_elt_size, buf,
                          INT_REGISTER_SIZE))
    return 0;

  *pc = extract_unsigned_integer (buf, INT_REGISTER_SIZE, byte_order);
  return 1;
}

/* Recognize GCC and GNU ld's trampolines.  If we are in a trampoline,
   return the target PC.  Otherwise return 0.  */

CORE_ADDR
arm_skip_stub (struct frame_info *frame, CORE_ADDR pc)
{
  char *name;
  int namelen;
  CORE_ADDR start_addr;

  /* Find the starting address and name of the function containing the PC.  */
  if (find_pc_partial_function (pc, &name, &start_addr, NULL) == 0)
    return 0;

  /* If PC is in a Thumb call or return stub, return the address of the
     target PC, which is in a register.  The thunk functions are called
     _call_via_xx, where x is the register name.  The possible names
     are r0-r9, sl, fp, ip, sp, and lr.  */
  if (strncmp (name, "_call_via_", 10) == 0)
    {
      /* Use the name suffix to determine which register contains the
         target PC.  */
      static char *table[15] =
      {"r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7",
       "r8", "r9", "sl", "fp", "ip", "sp", "lr"
      };
      int regno;
      int offset = strlen (name) - 2;

      for (regno = 0; regno <= 14; regno++)
        if (strcmp (&name[offset], table[regno]) == 0)
          return get_frame_register_unsigned (frame, regno);
    }

  /* GNU ld generates __foo_from_arm or __foo_from_thumb for
     non-interworking calls to foo.  We could decode the stubs
     to find the target but it's easier to use the symbol table.  */
  namelen = strlen (name);
  if (name[0] == '_' && name[1] == '_'
      && ((namelen > 2 + strlen ("_from_thumb")
           && strncmp (name + namelen - strlen ("_from_thumb"), "_from_thumb",
                       strlen ("_from_thumb")) == 0)
          || (namelen > 2 + strlen ("_from_arm")
              && strncmp (name + namelen - strlen ("_from_arm"), "_from_arm",
                          strlen ("_from_arm")) == 0)))
    {
      char *target_name;
      int target_len = namelen - 2;
      struct minimal_symbol *minsym;
      struct objfile *objfile;
      struct obj_section *sec;

      if (name[namelen - 1] == 'b')
        target_len -= strlen ("_from_thumb");
      else
        target_len -= strlen ("_from_arm");

      target_name = alloca (target_len + 1);
      memcpy (target_name, name + 2, target_len);
      target_name[target_len] = '\0';

      sec = find_pc_section (pc);
      objfile = (sec == NULL) ? NULL : sec->objfile;
      minsym = lookup_minimal_symbol (target_name, NULL, objfile);
      if (minsym != NULL)
        return SYMBOL_VALUE_ADDRESS (minsym);
      else
        return 0;
    }

  return 0;                     /* not a stub */
}

static void
set_arm_command (char *args, int from_tty)
{
  printf_unfiltered (_("\
\"set arm\" must be followed by an apporpriate subcommand.\n"));
  help_list (setarmcmdlist, "set arm ", all_commands, gdb_stdout);
}

static void
show_arm_command (char *args, int from_tty)
{
  cmd_show_list (showarmcmdlist, from_tty, "");
}

static void
arm_update_current_architecture (void)
{
  struct gdbarch_info info;

  /* If the current architecture is not ARM, we have nothing to do.  */
  if (gdbarch_bfd_arch_info (target_gdbarch)->arch != bfd_arch_arm)
    return;

  /* Update the architecture.  */
  gdbarch_info_init (&info);

  if (!gdbarch_update_p (info))
    internal_error (__FILE__, __LINE__, "could not update architecture");
}

static void
set_fp_model_sfunc (char *args, int from_tty,
                    struct cmd_list_element *c)
{
  enum arm_float_model fp_model;

  for (fp_model = ARM_FLOAT_AUTO; fp_model != ARM_FLOAT_LAST; fp_model++)
    if (strcmp (current_fp_model, fp_model_strings[fp_model]) == 0)
      {
        arm_fp_model = fp_model;
        break;
      }

  if (fp_model == ARM_FLOAT_LAST)
    internal_error (__FILE__, __LINE__, _("Invalid fp model accepted: %s."),
                    current_fp_model);

  arm_update_current_architecture ();
}

static void
show_fp_model (struct ui_file *file, int from_tty,
               struct cmd_list_element *c, const char *value)
{
  struct gdbarch_tdep *tdep = gdbarch_tdep (target_gdbarch);

  if (arm_fp_model == ARM_FLOAT_AUTO
      && gdbarch_bfd_arch_info (target_gdbarch)->arch == bfd_arch_arm)
    fprintf_filtered (file, _("\
The current ARM floating point model is \"auto\" (currently \"%s\").\n"),
                      fp_model_strings[tdep->fp_model]);
  else
    fprintf_filtered (file, _("\
The current ARM floating point model is \"%s\".\n"),
                      fp_model_strings[arm_fp_model]);
}

static void
arm_set_abi (char *args, int from_tty,
             struct cmd_list_element *c)
{
  enum arm_abi_kind arm_abi;

  for (arm_abi = ARM_ABI_AUTO; arm_abi != ARM_ABI_LAST; arm_abi++)
    if (strcmp (arm_abi_string, arm_abi_strings[arm_abi]) == 0)
      {
        arm_abi_global = arm_abi;
        break;
      }

  if (arm_abi == ARM_ABI_LAST)
    internal_error (__FILE__, __LINE__, _("Invalid ABI accepted: %s."),
                    arm_abi_string);

  arm_update_current_architecture ();
}

static void
arm_show_abi (struct ui_file *file, int from_tty,
             struct cmd_list_element *c, const char *value)
{
  struct gdbarch_tdep *tdep = gdbarch_tdep (target_gdbarch);

  if (arm_abi_global == ARM_ABI_AUTO
      && gdbarch_bfd_arch_info (target_gdbarch)->arch == bfd_arch_arm)
    fprintf_filtered (file, _("\
The current ARM ABI is \"auto\" (currently \"%s\").\n"),
                      arm_abi_strings[tdep->arm_abi]);
  else
    fprintf_filtered (file, _("The current ARM ABI is \"%s\".\n"),
                      arm_abi_string);
}

static void
arm_show_fallback_mode (struct ui_file *file, int from_tty,
                        struct cmd_list_element *c, const char *value)
{
  struct gdbarch_tdep *tdep = gdbarch_tdep (target_gdbarch);

  fprintf_filtered (file, _("\
The current execution mode assumed (when symbols are unavailable) is \"%s\".\n"),
                    arm_fallback_mode_string);
}

static void
arm_show_force_mode (struct ui_file *file, int from_tty,
                     struct cmd_list_element *c, const char *value)
{
  struct gdbarch_tdep *tdep = gdbarch_tdep (target_gdbarch);

  fprintf_filtered (file, _("\
The current execution mode assumed (even when symbols are available) is \"%s\".\n"),
                    arm_force_mode_string);
}

/* If the user changes the register disassembly style used for info
   register and other commands, we have to also switch the style used
   in opcodes for disassembly output.  This function is run in the "set
   arm disassembly" command, and does that.  */

static void
set_disassembly_style_sfunc (char *args, int from_tty,
                              struct cmd_list_element *c)
{
  set_disassembly_style ();
}
\f
/* Return the ARM register name corresponding to register I.  */
static const char *
arm_register_name (struct gdbarch *gdbarch, int i)
{
  const int num_regs = gdbarch_num_regs (gdbarch);

  if (gdbarch_tdep (gdbarch)->have_vfp_pseudos
      && i >= num_regs && i < num_regs + 32)
    {
      static const char *const vfp_pseudo_names[] = {
        "s0", "s1", "s2", "s3", "s4", "s5", "s6", "s7",
        "s8", "s9", "s10", "s11", "s12", "s13", "s14", "s15",
        "s16", "s17", "s18", "s19", "s20", "s21", "s22", "s23",
        "s24", "s25", "s26", "s27", "s28", "s29", "s30", "s31",
      };

      return vfp_pseudo_names[i - num_regs];
    }

  if (gdbarch_tdep (gdbarch)->have_neon_pseudos
      && i >= num_regs + 32 && i < num_regs + 32 + 16)
    {
      static const char *const neon_pseudo_names[] = {
        "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7",
        "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15",
      };

      return neon_pseudo_names[i - num_regs - 32];
    }

  if (i >= ARRAY_SIZE (arm_register_names))
    /* These registers are only supported on targets which supply
       an XML description.  */
    return "";

  return arm_register_names[i];
}

static void
set_disassembly_style (void)
{
  int current;

  /* Find the style that the user wants.  */
  for (current = 0; current < num_disassembly_options; current++)
    if (disassembly_style == valid_disassembly_styles[current])
      break;
  gdb_assert (current < num_disassembly_options);

  /* Synchronize the disassembler.  */
  set_arm_regname_option (current);
}

/* Test whether the coff symbol specific value corresponds to a Thumb
   function.  */

static int
coff_sym_is_thumb (int val)
{
  return (val == C_THUMBEXT ||
          val == C_THUMBSTAT ||
          val == C_THUMBEXTFUNC ||
          val == C_THUMBSTATFUNC ||
          val == C_THUMBLABEL);
}

/* arm_coff_make_msymbol_special()
   arm_elf_make_msymbol_special()
   
   These functions test whether the COFF or ELF symbol corresponds to
   an address in thumb code, and set a "special" bit in a minimal
   symbol to indicate that it does.  */
   
static void
arm_elf_make_msymbol_special(asymbol *sym, struct minimal_symbol *msym)
{
  /* Thumb symbols are of type STT_LOPROC, (synonymous with
     STT_ARM_TFUNC).  */
  if (ELF_ST_TYPE (((elf_symbol_type *)sym)->internal_elf_sym.st_info)
      == STT_LOPROC)
    MSYMBOL_SET_SPECIAL (msym);
}

static void
arm_coff_make_msymbol_special(int val, struct minimal_symbol *msym)
{
  if (coff_sym_is_thumb (val))
    MSYMBOL_SET_SPECIAL (msym);
}

static void
arm_objfile_data_free (struct objfile *objfile, void *arg)
{
  struct arm_per_objfile *data = arg;
  unsigned int i;

  for (i = 0; i < objfile->obfd->section_count; i++)
    VEC_free (arm_mapping_symbol_s, data->section_maps[i]);
}

static void
arm_record_special_symbol (struct gdbarch *gdbarch, struct objfile *objfile,
                           asymbol *sym)
{
  const char *name = bfd_asymbol_name (sym);
  struct arm_per_objfile *data;
  VEC(arm_mapping_symbol_s) **map_p;
  struct arm_mapping_symbol new_map_sym;

  gdb_assert (name[0] == '$');
  if (name[1] != 'a' && name[1] != 't' && name[1] != 'd')
    return;

  data = objfile_data (objfile, arm_objfile_data_key);
  if (data == NULL)
    {
      data = OBSTACK_ZALLOC (&objfile->objfile_obstack,
                             struct arm_per_objfile);
      set_objfile_data (objfile, arm_objfile_data_key, data);
      data->section_maps = OBSTACK_CALLOC (&objfile->objfile_obstack,
                                           objfile->obfd->section_count,
                                           VEC(arm_mapping_symbol_s) *);
    }
  map_p = &data->section_maps[bfd_get_section (sym)->index];

  new_map_sym.value = sym->value;
  new_map_sym.type = name[1];

  /* Assume that most mapping symbols appear in order of increasing
     value.  If they were randomly distributed, it would be faster to
     always push here and then sort at first use.  */
  if (!VEC_empty (arm_mapping_symbol_s, *map_p))
    {
      struct arm_mapping_symbol *prev_map_sym;

      prev_map_sym = VEC_last (arm_mapping_symbol_s, *map_p);
      if (prev_map_sym->value >= sym->value)
        {
          unsigned int idx;
          idx = VEC_lower_bound (arm_mapping_symbol_s, *map_p, &new_map_sym,
                                 arm_compare_mapping_symbols);
          VEC_safe_insert (arm_mapping_symbol_s, *map_p, idx, &new_map_sym);
          return;
        }
    }

  VEC_safe_push (arm_mapping_symbol_s, *map_p, &new_map_sym);
}

static void
arm_write_pc (struct regcache *regcache, CORE_ADDR pc)
{
  regcache_cooked_write_unsigned (regcache, ARM_PC_REGNUM, pc);

  /* If necessary, set the T bit.  */
  if (arm_apcs_32)
    {
      ULONGEST val;
      regcache_cooked_read_unsigned (regcache, ARM_PS_REGNUM, &val);
      if (arm_pc_is_thumb (pc))
        regcache_cooked_write_unsigned (regcache, ARM_PS_REGNUM, val | CPSR_T);
      else
        regcache_cooked_write_unsigned (regcache, ARM_PS_REGNUM,
                                        val & ~(ULONGEST) CPSR_T);
    }
}

/* Read the contents of a NEON quad register, by reading from two
   double registers.  This is used to implement the quad pseudo
   registers, and for argument passing in case the quad registers are
   missing; vectors are passed in quad registers when using the VFP
   ABI, even if a NEON unit is not present.  REGNUM is the index of
   the quad register, in [0, 15].  */

static void
arm_neon_quad_read (struct gdbarch *gdbarch, struct regcache *regcache,
                    int regnum, gdb_byte *buf)
{
  char name_buf[4];
  gdb_byte reg_buf[8];
  int offset, double_regnum;

  sprintf (name_buf, "d%d", regnum << 1);
  double_regnum = user_reg_map_name_to_regnum (gdbarch, name_buf,
                                               strlen (name_buf));

  /* d0 is always the least significant half of q0.  */
  if (gdbarch_byte_order (gdbarch) == BFD_ENDIAN_BIG)
    offset = 8;
  else
    offset = 0;

  regcache_raw_read (regcache, double_regnum, reg_buf);
  memcpy (buf + offset, reg_buf, 8);

  offset = 8 - offset;
  regcache_raw_read (regcache, double_regnum + 1, reg_buf);
  memcpy (buf + offset, reg_buf, 8);
}

static void
arm_pseudo_read (struct gdbarch *gdbarch, struct regcache *regcache,
                 int regnum, gdb_byte *buf)
{
  const int num_regs = gdbarch_num_regs (gdbarch);
  char name_buf[4];
  gdb_byte reg_buf[8];
  int offset, double_regnum;

  gdb_assert (regnum >= num_regs);
  regnum -= num_regs;

  if (gdbarch_tdep (gdbarch)->have_neon_pseudos && regnum >= 32 && regnum < 48)
    /* Quad-precision register.  */
    arm_neon_quad_read (gdbarch, regcache, regnum - 32, buf);
  else
    {
      /* Single-precision register.  */
      gdb_assert (regnum < 32);

      /* s0 is always the least significant half of d0.  */
      if (gdbarch_byte_order (gdbarch) == BFD_ENDIAN_BIG)
        offset = (regnum & 1) ? 0 : 4;
      else
        offset = (regnum & 1) ? 4 : 0;

      sprintf (name_buf, "d%d", regnum >> 1);
      double_regnum = user_reg_map_name_to_regnum (gdbarch, name_buf,
                                                   strlen (name_buf));

      regcache_raw_read (regcache, double_regnum, reg_buf);
      memcpy (buf, reg_buf + offset, 4);
    }
}

/* Store the contents of BUF to a NEON quad register, by writing to
   two double registers.  This is used to implement the quad pseudo
   registers, and for argument passing in case the quad registers are
   missing; vectors are passed in quad registers when using the VFP
   ABI, even if a NEON unit is not present.  REGNUM is the index
   of the quad register, in [0, 15].  */

static void
arm_neon_quad_write (struct gdbarch *gdbarch, struct regcache *regcache,
                     int regnum, const gdb_byte *buf)
{
  char name_buf[4];
  gdb_byte reg_buf[8];
  int offset, double_regnum;

  sprintf (name_buf, "d%d", regnum << 1);
  double_regnum = user_reg_map_name_to_regnum (gdbarch, name_buf,
                                               strlen (name_buf));

  /* d0 is always the least significant half of q0.  */
  if (gdbarch_byte_order (gdbarch) == BFD_ENDIAN_BIG)
    offset = 8;
  else
    offset = 0;

  regcache_raw_write (regcache, double_regnum, buf + offset);
  offset = 8 - offset;
  regcache_raw_write (regcache, double_regnum + 1, buf + offset);
}

static void
arm_pseudo_write (struct gdbarch *gdbarch, struct regcache *regcache,
                  int regnum, const gdb_byte *buf)
{
  const int num_regs = gdbarch_num_regs (gdbarch);
  char name_buf[4];
  gdb_byte reg_buf[8];
  int offset, double_regnum;

  gdb_assert (regnum >= num_regs);
  regnum -= num_regs;

  if (gdbarch_tdep (gdbarch)->have_neon_pseudos && regnum >= 32 && regnum < 48)
    /* Quad-precision register.  */
    arm_neon_quad_write (gdbarch, regcache, regnum - 32, buf);
  else
    {
      /* Single-precision register.  */
      gdb_assert (regnum < 32);

      /* s0 is always the least significant half of d0.  */
      if (gdbarch_byte_order (gdbarch) == BFD_ENDIAN_BIG)
        offset = (regnum & 1) ? 0 : 4;
      else
        offset = (regnum & 1) ? 4 : 0;

      sprintf (name_buf, "d%d", regnum >> 1);
      double_regnum = user_reg_map_name_to_regnum (gdbarch, name_buf,
                                                   strlen (name_buf));

      regcache_raw_read (regcache, double_regnum, reg_buf);
      memcpy (reg_buf + offset, buf, 4);
      regcache_raw_write (regcache, double_regnum, reg_buf);
    }
}

static struct value *
value_of_arm_user_reg (struct frame_info *frame, const void *baton)
{
  const int *reg_p = baton;
  return value_of_register (*reg_p, frame);
}
\f
static enum gdb_osabi
arm_elf_osabi_sniffer (bfd *abfd)
{
  unsigned int elfosabi;
  enum gdb_osabi osabi = GDB_OSABI_UNKNOWN;

  elfosabi = elf_elfheader (abfd)->e_ident[EI_OSABI];

  if (elfosabi == ELFOSABI_ARM)
    /* GNU tools use this value.  Check note sections in this case,
       as well.  */
    bfd_map_over_sections (abfd,
                           generic_elf_osabi_sniff_abi_tag_sections, 
                           &osabi);

  /* Anything else will be handled by the generic ELF sniffer.  */
  return osabi;
}

\f
/* Initialize the current architecture based on INFO.  If possible,
   re-use an architecture from ARCHES, which is a list of
   architectures already created during this debugging session.

   Called e.g. at program startup, when reading a core file, and when
   reading a binary file.  */

static struct gdbarch *
arm_gdbarch_init (struct gdbarch_info info, struct gdbarch_list *arches)
{
  struct gdbarch_tdep *tdep;
  struct gdbarch *gdbarch;
  struct gdbarch_list *best_arch;
  enum arm_abi_kind arm_abi = arm_abi_global;
  enum arm_float_model fp_model = arm_fp_model;
  struct tdesc_arch_data *tdesc_data = NULL;
  int i;
  int have_vfp_registers = 0, have_vfp_pseudos = 0, have_neon_pseudos = 0;
  int have_neon = 0;
  int have_fpa_registers = 1;

  /* Check any target description for validity.  */
  if (tdesc_has_registers (info.target_desc))
    {
      /* For most registers we require GDB's default names; but also allow
         the numeric names for sp / lr / pc, as a convenience.  */
      static const char *const arm_sp_names[] = { "r13", "sp", NULL };
      static const char *const arm_lr_names[] = { "r14", "lr", NULL };
      static const char *const arm_pc_names[] = { "r15", "pc", NULL };

      const struct tdesc_feature *feature;
      int valid_p;

      feature = tdesc_find_feature (info.target_desc,
                                    "org.gnu.gdb.arm.core");
      if (feature == NULL)
        return NULL;

      tdesc_data = tdesc_data_alloc ();

      valid_p = 1;
      for (i = 0; i < ARM_SP_REGNUM; i++)
        valid_p &= tdesc_numbered_register (feature, tdesc_data, i,
                                            arm_register_names[i]);
      valid_p &= tdesc_numbered_register_choices (feature, tdesc_data,
                                                  ARM_SP_REGNUM,
                                                  arm_sp_names);
      valid_p &= tdesc_numbered_register_choices (feature, tdesc_data,
                                                  ARM_LR_REGNUM,
                                                  arm_lr_names);
      valid_p &= tdesc_numbered_register_choices (feature, tdesc_data,
                                                  ARM_PC_REGNUM,
                                                  arm_pc_names);
      valid_p &= tdesc_numbered_register (feature, tdesc_data,
                                          ARM_PS_REGNUM, "cpsr");

      if (!valid_p)
        {
          tdesc_data_cleanup (tdesc_data);
          return NULL;
        }

      feature = tdesc_find_feature (info.target_desc,
                                    "org.gnu.gdb.arm.fpa");
      if (feature != NULL)
        {
          valid_p = 1;
          for (i = ARM_F0_REGNUM; i <= ARM_FPS_REGNUM; i++)
            valid_p &= tdesc_numbered_register (feature, tdesc_data, i,
                                                arm_register_names[i]);
          if (!valid_p)
            {
              tdesc_data_cleanup (tdesc_data);
              return NULL;
            }
        }
      else
        have_fpa_registers = 0;

      feature = tdesc_find_feature (info.target_desc,
                                    "org.gnu.gdb.xscale.iwmmxt");
      if (feature != NULL)
        {
          static const char *const iwmmxt_names[] = {
            "wR0", "wR1", "wR2", "wR3", "wR4", "wR5", "wR6", "wR7",
            "wR8", "wR9", "wR10", "wR11", "wR12", "wR13", "wR14", "wR15",
            "wCID", "wCon", "wCSSF", "wCASF", "", "", "", "",
            "wCGR0", "wCGR1", "wCGR2", "wCGR3", "", "", "", "",
          };

          valid_p = 1;
          for (i = ARM_WR0_REGNUM; i <= ARM_WR15_REGNUM; i++)
            valid_p
              &= tdesc_numbered_register (feature, tdesc_data, i,
                                          iwmmxt_names[i - ARM_WR0_REGNUM]);

          /* Check for the control registers, but do not fail if they
             are missing.  */
          for (i = ARM_WC0_REGNUM; i <= ARM_WCASF_REGNUM; i++)
            tdesc_numbered_register (feature, tdesc_data, i,
                                     iwmmxt_names[i - ARM_WR0_REGNUM]);

          for (i = ARM_WCGR0_REGNUM; i <= ARM_WCGR3_REGNUM; i++)
            valid_p
              &= tdesc_numbered_register (feature, tdesc_data, i,
                                          iwmmxt_names[i - ARM_WR0_REGNUM]);

          if (!valid_p)
            {
              tdesc_data_cleanup (tdesc_data);
              return NULL;
            }
        }

      /* If we have a VFP unit, check whether the single precision registers
         are present.  If not, then we will synthesize them as pseudo
         registers.  */
      feature = tdesc_find_feature (info.target_desc,
                                    "org.gnu.gdb.arm.vfp");
      if (feature != NULL)
        {
          static const char *const vfp_double_names[] = {
            "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7",
            "d8", "d9", "d10", "d11", "d12", "d13", "d14", "d15",
            "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23",
            "d24", "d25", "d26", "d27", "d28", "d29", "d30", "d31",
          };

          /* Require the double precision registers.  There must be either
             16 or 32.  */
          valid_p = 1;
          for (i = 0; i < 32; i++)
            {
              valid_p &= tdesc_numbered_register (feature, tdesc_data,
                                                  ARM_D0_REGNUM + i,
                                                  vfp_double_names[i]);
              if (!valid_p)
                break;
            }

          if (!valid_p && i != 16)
            {
              tdesc_data_cleanup (tdesc_data);
              return NULL;
            }

          if (tdesc_unnumbered_register (feature, "s0") == 0)
            have_vfp_pseudos = 1;

          have_vfp_registers = 1;

          /* If we have VFP, also check for NEON.  The architecture allows
             NEON without VFP (integer vector operations only), but GDB
             does not support that.  */
          feature = tdesc_find_feature (info.target_desc,
                                        "org.gnu.gdb.arm.neon");
          if (feature != NULL)
            {
              /* NEON requires 32 double-precision registers.  */
              if (i != 32)
                {
                  tdesc_data_cleanup (tdesc_data);
                  return NULL;
                }

              /* If there are quad registers defined by the stub, use
                 their type; otherwise (normally) provide them with
                 the default type.  */
              if (tdesc_unnumbered_register (feature, "q0") == 0)
                have_neon_pseudos = 1;

              have_neon = 1;
            }
        }
    }

  /* If we have an object to base this architecture on, try to determine
     its ABI.  */

  if (arm_abi == ARM_ABI_AUTO && info.abfd != NULL)
    {
      int ei_osabi, e_flags;

      switch (bfd_get_flavour (info.abfd))
        {
        case bfd_target_aout_flavour:
          /* Assume it's an old APCS-style ABI.  */
          arm_abi = ARM_ABI_APCS;
          break;

        case bfd_target_coff_flavour:
          /* Assume it's an old APCS-style ABI.  */
          /* XXX WinCE?  */
          arm_abi = ARM_ABI_APCS;
          break;

        case bfd_target_elf_flavour:
          ei_osabi = elf_elfheader (info.abfd)->e_ident[EI_OSABI];
          e_flags = elf_elfheader (info.abfd)->e_flags;

          if (ei_osabi == ELFOSABI_ARM)
            {
              /* GNU tools used to use this value, but do not for EABI
                 objects.  There's nowhere to tag an EABI version
                 anyway, so assume APCS.  */
              arm_abi = ARM_ABI_APCS;
            }
          else if (ei_osabi == ELFOSABI_NONE)
            {
              int eabi_ver = EF_ARM_EABI_VERSION (e_flags);

              switch (eabi_ver)
                {
                case EF_ARM_EABI_UNKNOWN:
                  /* Assume GNU tools.  */
                  arm_abi = ARM_ABI_APCS;
                  break;

                case EF_ARM_EABI_VER4:
                case EF_ARM_EABI_VER5:
                  arm_abi = ARM_ABI_AAPCS;
                  /* EABI binaries default to VFP float ordering.
                     They may also contain build attributes that can
                     be used to identify if the VFP argument-passing
                     ABI is in use.  */
                  if (fp_model == ARM_FLOAT_AUTO)
                    {
#ifdef HAVE_ELF
                      switch (bfd_elf_get_obj_attr_int (info.abfd,
                                                        OBJ_ATTR_PROC,
                                                        Tag_ABI_VFP_args))
                        {
                        case 0:
                          /* "The user intended FP parameter/result
                             passing to conform to AAPCS, base
                             variant".  */
                          fp_model = ARM_FLOAT_SOFT_VFP;
                          break;
                        case 1:
                          /* "The user intended FP parameter/result
                             passing to conform to AAPCS, VFP
                             variant".  */
                          fp_model = ARM_FLOAT_VFP;
                          break;
                        case 2:
                          /* "The user intended FP parameter/result
                             passing to conform to tool chain-specific
                             conventions" - we don't know any such
                             conventions, so leave it as "auto".  */
                          break;
                        default:
                          /* Attribute value not mentioned in the
                             October 2008 ABI, so leave it as
                             "auto".  */
                          break;
                        }
#else
                      fp_model = ARM_FLOAT_SOFT_VFP;
#endif
                    }
                  break;

                default:
                  /* Leave it as "auto".  */
                  warning (_("unknown ARM EABI version 0x%x"), eabi_ver);
                  break;
                }
            }

          if (fp_model == ARM_FLOAT_AUTO)
            {
              int e_flags = elf_elfheader (info.abfd)->e_flags;

              switch (e_flags & (EF_ARM_SOFT_FLOAT | EF_ARM_VFP_FLOAT))
                {
                case 0:
                  /* Leave it as "auto".  Strictly speaking this case
                     means FPA, but almost nobody uses that now, and
                     many toolchains fail to set the appropriate bits
                     for the floating-point model they use.  */
                  break;
                case EF_ARM_SOFT_FLOAT:
                  fp_model = ARM_FLOAT_SOFT_FPA;
                  break;
                case EF_ARM_VFP_FLOAT:
                  fp_model = ARM_FLOAT_VFP;
                  break;
                case EF_ARM_SOFT_FLOAT | EF_ARM_VFP_FLOAT:
                  fp_model = ARM_FLOAT_SOFT_VFP;
                  break;
                }
            }

          if (e_flags & EF_ARM_BE8)
            info.byte_order_for_code = BFD_ENDIAN_LITTLE;

          break;

        default:
          /* Leave it as "auto".  */
          break;
        }
    }

  /* If there is already a candidate, use it.  */
  for (best_arch = gdbarch_list_lookup_by_info (arches, &info);
       best_arch != NULL;
       best_arch = gdbarch_list_lookup_by_info (best_arch->next, &info))
    {
      if (arm_abi != ARM_ABI_AUTO
          && arm_abi != gdbarch_tdep (best_arch->gdbarch)->arm_abi)
        continue;

      if (fp_model != ARM_FLOAT_AUTO
          && fp_model != gdbarch_tdep (best_arch->gdbarch)->fp_model)
        continue;

      /* There are various other properties in tdep that we do not
         need to check here: those derived from a target description,
         since gdbarches with a different target description are
         automatically disqualified.  */

      /* Found a match.  */
      break;
    }

  if (best_arch != NULL)
    {
      if (tdesc_data != NULL)
        tdesc_data_cleanup (tdesc_data);
      return best_arch->gdbarch;
    }

  tdep = xcalloc (1, sizeof (struct gdbarch_tdep));
  gdbarch = gdbarch_alloc (&info, tdep);

  /* Record additional information about the architecture we are defining.
     These are gdbarch discriminators, like the OSABI.  */
  tdep->arm_abi = arm_abi;
  tdep->fp_model = fp_model;
  tdep->have_fpa_registers = have_fpa_registers;
  tdep->have_vfp_registers = have_vfp_registers;
  tdep->have_vfp_pseudos = have_vfp_pseudos;
  tdep->have_neon_pseudos = have_neon_pseudos;
  tdep->have_neon = have_neon;

  /* Breakpoints.  */
  switch (info.byte_order_for_code)
    {
    case BFD_ENDIAN_BIG:
      tdep->arm_breakpoint = arm_default_arm_be_breakpoint;
      tdep->arm_breakpoint_size = sizeof (arm_default_arm_be_breakpoint);
      tdep->thumb_breakpoint = arm_default_thumb_be_breakpoint;
      tdep->thumb_breakpoint_size = sizeof (arm_default_thumb_be_breakpoint);

      break;

    case BFD_ENDIAN_LITTLE:
      tdep->arm_breakpoint = arm_default_arm_le_breakpoint;
      tdep->arm_breakpoint_size = sizeof (arm_default_arm_le_breakpoint);
      tdep->thumb_breakpoint = arm_default_thumb_le_breakpoint;
      tdep->thumb_breakpoint_size = sizeof (arm_default_thumb_le_breakpoint);

      break;

    default:
      internal_error (__FILE__, __LINE__,
                      _("arm_gdbarch_init: bad byte order for float format"));
    }

  /* On ARM targets char defaults to unsigned.  */
  set_gdbarch_char_signed (gdbarch, 0);

  /* Note: for displaced stepping, this includes the breakpoint, and one word
     of additional scratch space.  This setting isn't used for anything beside
     displaced stepping at present.  */
  set_gdbarch_max_insn_length (gdbarch, 4 * DISPLACED_MODIFIED_INSNS);

  /* This should be low enough for everything.  */
  tdep->lowest_pc = 0x20;
  tdep->jb_pc = -1;     /* Longjump support not enabled by default.  */

  /* The default, for both APCS and AAPCS, is to return small
     structures in registers.  */
  tdep->struct_return = reg_struct_return;

  set_gdbarch_push_dummy_call (gdbarch, arm_push_dummy_call);
  set_gdbarch_frame_align (gdbarch, arm_frame_align);

  set_gdbarch_write_pc (gdbarch, arm_write_pc);

  /* Frame handling.  */
  set_gdbarch_dummy_id (gdbarch, arm_dummy_id);
  set_gdbarch_unwind_pc (gdbarch, arm_unwind_pc);
  set_gdbarch_unwind_sp (gdbarch, arm_unwind_sp);

  frame_base_set_default (gdbarch, &arm_normal_base);

  /* Address manipulation.  */
  set_gdbarch_smash_text_address (gdbarch, arm_smash_text_address);
  set_gdbarch_addr_bits_remove (gdbarch, arm_addr_bits_remove);

  /* Advance PC across function entry code.  */
  set_gdbarch_skip_prologue (gdbarch, arm_skip_prologue);

  /* Skip trampolines.  */
  set_gdbarch_skip_trampoline_code (gdbarch, arm_skip_stub);

  /* The stack grows downward.  */
  set_gdbarch_inner_than (gdbarch, core_addr_lessthan);

  /* Breakpoint manipulation.  */
  set_gdbarch_breakpoint_from_pc (gdbarch, arm_breakpoint_from_pc);

  /* Information about registers, etc.  */
  set_gdbarch_deprecated_fp_regnum (gdbarch, ARM_FP_REGNUM);    /* ??? */
  set_gdbarch_sp_regnum (gdbarch, ARM_SP_REGNUM);
  set_gdbarch_pc_regnum (gdbarch, ARM_PC_REGNUM);
  set_gdbarch_num_regs (gdbarch, ARM_NUM_REGS);
  set_gdbarch_register_type (gdbarch, arm_register_type);

  /* This "info float" is FPA-specific.  Use the generic version if we
     do not have FPA.  */
  if (gdbarch_tdep (gdbarch)->have_fpa_registers)
    set_gdbarch_print_float_info (gdbarch, arm_print_float_info);

  /* Internal <-> external register number maps.  */
  set_gdbarch_dwarf2_reg_to_regnum (gdbarch, arm_dwarf_reg_to_regnum);
  set_gdbarch_register_sim_regno (gdbarch, arm_register_sim_regno);

  set_gdbarch_register_name (gdbarch, arm_register_name);

  /* Returning results.  */
  set_gdbarch_return_value (gdbarch, arm_return_value);

  /* Disassembly.  */
  set_gdbarch_print_insn (gdbarch, gdb_print_insn_arm);

  /* Minsymbol frobbing.  */
  set_gdbarch_elf_make_msymbol_special (gdbarch, arm_elf_make_msymbol_special);
  set_gdbarch_coff_make_msymbol_special (gdbarch,
                                         arm_coff_make_msymbol_special);
  set_gdbarch_record_special_symbol (gdbarch, arm_record_special_symbol);

  /* Virtual tables.  */
  set_gdbarch_vbit_in_delta (gdbarch, 1);

  /* Hook in the ABI-specific overrides, if they have been registered.  */
  gdbarch_init_osabi (info, gdbarch);

  dwarf2_frame_set_init_reg (gdbarch, arm_dwarf2_frame_init_reg);

  /* Add some default predicates.  */
  frame_unwind_append_unwinder (gdbarch, &arm_stub_unwind);
  dwarf2_append_unwinders (gdbarch);
  frame_unwind_append_unwinder (gdbarch, &arm_prologue_unwind);

  /* Now we have tuned the configuration, set a few final things,
     based on what the OS ABI has told us.  */

  /* If the ABI is not otherwise marked, assume the old GNU APCS.  EABI
     binaries are always marked.  */
  if (tdep->arm_abi == ARM_ABI_AUTO)
    tdep->arm_abi = ARM_ABI_APCS;

  /* We used to default to FPA for generic ARM, but almost nobody
     uses that now, and we now provide a way for the user to force
     the model.  So default to the most useful variant.  */
  if (tdep->fp_model == ARM_FLOAT_AUTO)
    tdep->fp_model = ARM_FLOAT_SOFT_FPA;

  if (tdep->jb_pc >= 0)
    set_gdbarch_get_longjmp_target (gdbarch, arm_get_longjmp_target);

  /* Floating point sizes and format.  */
  set_gdbarch_float_format (gdbarch, floatformats_ieee_single);
  if (tdep->fp_model == ARM_FLOAT_SOFT_FPA || tdep->fp_model == ARM_FLOAT_FPA)
    {
      set_gdbarch_double_format
        (gdbarch, floatformats_ieee_double_littlebyte_bigword);
      set_gdbarch_long_double_format
        (gdbarch, floatformats_ieee_double_littlebyte_bigword);
    }
  else
    {
      set_gdbarch_double_format (gdbarch, floatformats_ieee_double);
      set_gdbarch_long_double_format (gdbarch, floatformats_ieee_double);
    }

  if (have_vfp_pseudos)
    {
      /* NOTE: These are the only pseudo registers used by
         the ARM target at the moment.  If more are added, a
         little more care in numbering will be needed.  */

      int num_pseudos = 32;
      if (have_neon_pseudos)
        num_pseudos += 16;
      set_gdbarch_num_pseudo_regs (gdbarch, num_pseudos);
      set_gdbarch_pseudo_register_read (gdbarch, arm_pseudo_read);
      set_gdbarch_pseudo_register_write (gdbarch, arm_pseudo_write);
    }

  if (tdesc_data)
    {
      set_tdesc_pseudo_register_name (gdbarch, arm_register_name);

      tdesc_use_registers (gdbarch, info.target_desc, tdesc_data);

      /* Override tdesc_register_type to adjust the types of VFP
         registers for NEON.  */
      set_gdbarch_register_type (gdbarch, arm_register_type);
    }

  /* Add standard register aliases.  We add aliases even for those
     nanes which are used by the current architecture - it's simpler,
     and does no harm, since nothing ever lists user registers.  */
  for (i = 0; i < ARRAY_SIZE (arm_register_aliases); i++)
    user_reg_add (gdbarch, arm_register_aliases[i].name,
                  value_of_arm_user_reg, &arm_register_aliases[i].regnum);

  return gdbarch;
}

static void
arm_dump_tdep (struct gdbarch *gdbarch, struct ui_file *file)
{
  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);

  if (tdep == NULL)
    return;

  fprintf_unfiltered (file, _("arm_dump_tdep: Lowest pc = 0x%lx"),
                      (unsigned long) tdep->lowest_pc);
}

extern initialize_file_ftype _initialize_arm_tdep; /* -Wmissing-prototypes */

void
_initialize_arm_tdep (void)
{
  struct ui_file *stb;
  long length;
  struct cmd_list_element *new_set, *new_show;
  const char *setname;
  const char *setdesc;
  const char *const *regnames;
  int numregs, i, j;
  static char *helptext;
  char regdesc[1024], *rdptr = regdesc;
  size_t rest = sizeof (regdesc);

  gdbarch_register (bfd_arch_arm, arm_gdbarch_init, arm_dump_tdep);

  arm_objfile_data_key
    = register_objfile_data_with_cleanup (NULL, arm_objfile_data_free);

  /* Register an ELF OS ABI sniffer for ARM binaries.  */
  gdbarch_register_osabi_sniffer (bfd_arch_arm,
                                  bfd_target_elf_flavour,
                                  arm_elf_osabi_sniffer);

  /* Get the number of possible sets of register names defined in opcodes.  */
  num_disassembly_options = get_arm_regname_num_options ();

  /* Add root prefix command for all "set arm"/"show arm" commands.  */
  add_prefix_cmd ("arm", no_class, set_arm_command,
                  _("Various ARM-specific commands."),
                  &setarmcmdlist, "set arm ", 0, &setlist);

  add_prefix_cmd ("arm", no_class, show_arm_command,
                  _("Various ARM-specific commands."),
                  &showarmcmdlist, "show arm ", 0, &showlist);

  /* Sync the opcode insn printer with our register viewer.  */
  parse_arm_disassembler_option ("reg-names-std");

  /* Initialize the array that will be passed to
     add_setshow_enum_cmd().  */
  valid_disassembly_styles
    = xmalloc ((num_disassembly_options + 1) * sizeof (char *));
  for (i = 0; i < num_disassembly_options; i++)
    {
      numregs = get_arm_regnames (i, &setname, &setdesc, &regnames);
      valid_disassembly_styles[i] = setname;
      length = snprintf (rdptr, rest, "%s - %s\n", setname, setdesc);
      rdptr += length;
      rest -= length;
      /* When we find the default names, tell the disassembler to use
         them.  */
      if (!strcmp (setname, "std"))
        {
          disassembly_style = setname;
          set_arm_regname_option (i);
        }
    }
  /* Mark the end of valid options.  */
  valid_disassembly_styles[num_disassembly_options] = NULL;

  /* Create the help text.  */
  stb = mem_fileopen ();
  fprintf_unfiltered (stb, "%s%s%s",
                      _("The valid values are:\n"),
                      regdesc,
                      _("The default is \"std\"."));
  helptext = ui_file_xstrdup (stb, NULL);
  ui_file_delete (stb);

  add_setshow_enum_cmd("disassembler", no_class,
                       valid_disassembly_styles, &disassembly_style,
                       _("Set the disassembly style."),
                       _("Show the disassembly style."),
                       helptext,
                       set_disassembly_style_sfunc,
                       NULL, /* FIXME: i18n: The disassembly style is \"%s\".  */
                       &setarmcmdlist, &showarmcmdlist);

  add_setshow_boolean_cmd ("apcs32", no_class, &arm_apcs_32,
                           _("Set usage of ARM 32-bit mode."),
                           _("Show usage of ARM 32-bit mode."),
                           _("When off, a 26-bit PC will be used."),
                           NULL,
                           NULL, /* FIXME: i18n: Usage of ARM 32-bit mode is %s.  */
                           &setarmcmdlist, &showarmcmdlist);

  /* Add a command to allow the user to force the FPU model.  */
  add_setshow_enum_cmd ("fpu", no_class, fp_model_strings, &current_fp_model,
                        _("Set the floating point type."),
                        _("Show the floating point type."),
                        _("auto - Determine the FP typefrom the OS-ABI.\n\
softfpa - Software FP, mixed-endian doubles on little-endian ARMs.\n\
fpa - FPA co-processor (GCC compiled).\n\
softvfp - Software FP with pure-endian doubles.\n\
vfp - VFP co-processor."),
                        set_fp_model_sfunc, show_fp_model,
                        &setarmcmdlist, &showarmcmdlist);

  /* Add a command to allow the user to force the ABI.  */
  add_setshow_enum_cmd ("abi", class_support, arm_abi_strings, &arm_abi_string,
                        _("Set the ABI."),
                        _("Show the ABI."),
                        NULL, arm_set_abi, arm_show_abi,
                        &setarmcmdlist, &showarmcmdlist);

  /* Add two commands to allow the user to force the assumed
     execution mode.  */
  add_setshow_enum_cmd ("fallback-mode", class_support,
                        arm_mode_strings, &arm_fallback_mode_string,
                        _("Set the mode assumed when symbols are unavailable."),
                        _("Show the mode assumed when symbols are unavailable."),
                        NULL, NULL, arm_show_fallback_mode,
                        &setarmcmdlist, &showarmcmdlist);
  add_setshow_enum_cmd ("force-mode", class_support,
                        arm_mode_strings, &arm_force_mode_string,
                        _("Set the mode assumed even when symbols are available."),
                        _("Show the mode assumed even when symbols are available."),
                        NULL, NULL, arm_show_force_mode,
                        &setarmcmdlist, &showarmcmdlist);

  /* Debugging flag.  */
  add_setshow_boolean_cmd ("arm", class_maintenance, &arm_debug,
                           _("Set ARM debugging."),
                           _("Show ARM debugging."),
                           _("When on, arm-specific debugging is enabled."),
                           NULL,
                           NULL, /* FIXME: i18n: "ARM debugging is %s.  */
                           &setdebuglist, &showdebuglist);
}