vax.c

早期freebsd实现

	  fragP->fr_fix += 1 + 4;
	  fix_new (fragP, old_fr_fix + 1, 4, fragP->fr_symbol, 0,
		   fragP->fr_offset, 1);
	  frag_wane (fragP);
	}
      break;

    default:
      break;
    }
  return (fragP->fr_var + fragP->fr_fix - old_fr_fix);
}				/* md_estimate_size_before_relax() */

/*
 *			md_convert_frag();
 *
 * Called after relax() is finished.
 * In:	Address of frag.
 *	fr_type == rs_machine_dependent.
 *	fr_subtype is what the address relaxed to.
 *
 * Out:	Any fixSs and constants are set up.
 *	Caller will turn frag into a ".space 0".
 */
void
md_convert_frag (fragP)
     register fragS *fragP;
{
  register char *addressP;	/* -> _var to change. */
  register char *opcodeP;	/* -> opcode char(s) to change. */
  register short int length_code;	/* 2=long 1=word 0=byte */
  register short int extension;	/* Size of relaxed address. */
  /* Added to fr_fix: incl. ALL var chars. */
  register symbolS *symbolP;
  register long int where;
  register long int address_of_var;
  /* Where, in file space, is _var of *fragP? */
  register long int target_address;
  /* Where, in file space, does addr point? */

  know (fragP->fr_type == rs_machine_dependent);
  length_code = fragP->fr_subtype & 3;	/* depends on ENCODE_RELAX() */
  know (length_code >= 0 && length_code < 3);
  where = fragP->fr_fix;
  addressP = fragP->fr_literal + where;
  opcodeP = fragP->fr_opcode;
  symbolP = fragP->fr_symbol;
  know (symbolP);
  target_address = symbolP->sy_value + fragP->fr_offset;
  address_of_var = fragP->fr_address + where;
  switch (fragP->fr_subtype)
    {
    case ENCODE_RELAX (STATE_PC_RELATIVE, STATE_BYTE):
      know (*addressP == 0 || *addressP == 0x10);	/* '@' bit. */
      addressP[0] |= 0xAF;	/* Byte displacement. */
      addressP[1] = target_address - (address_of_var + 2);
      extension = 2;
      break;

    case ENCODE_RELAX (STATE_PC_RELATIVE, STATE_WORD):
      know (*addressP == 0 || *addressP == 0x10);	/* '@' bit. */
      addressP[0] |= 0xCF;	/* Word displacement. */
      md_number_to_chars (addressP + 1, target_address - (address_of_var + 3), 2);
      extension = 3;
      break;

    case ENCODE_RELAX (STATE_PC_RELATIVE, STATE_LONG):
      know (*addressP == 0 || *addressP == 0x10);	/* '@' bit. */
      addressP[0] |= 0xEF;	/* Long word displacement. */
      md_number_to_chars (addressP + 1, target_address - (address_of_var + 5), 4);
      extension = 5;
      break;

    case ENCODE_RELAX (STATE_CONDITIONAL_BRANCH, STATE_BYTE):
      addressP[0] = target_address - (address_of_var + 1);
      extension = 1;
      break;

    case ENCODE_RELAX (STATE_CONDITIONAL_BRANCH, STATE_WORD):
      opcodeP[0] ^= 1;		/* Reverse sense of test. */
      addressP[0] = 3;
      addressP[1] = VAX_BRB + VAX_WIDEN_WORD;
      md_number_to_chars (addressP + 2, target_address - (address_of_var + 4), 2);
      extension = 4;
      break;

    case ENCODE_RELAX (STATE_CONDITIONAL_BRANCH, STATE_LONG):
      opcodeP[0] ^= 1;		/* Reverse sense of test. */
      addressP[0] = 6;
      addressP[1] = VAX_JMP;
      addressP[2] = VAX_PC_RELATIVE_MODE;
      md_number_to_chars (addressP + 3, target_address, 4);
      extension = 7;
      break;

    case ENCODE_RELAX (STATE_ALWAYS_BRANCH, STATE_BYTE):
      addressP[0] = target_address - (address_of_var + 1);
      extension = 1;
      break;

    case ENCODE_RELAX (STATE_ALWAYS_BRANCH, STATE_WORD):
      opcodeP[0] += VAX_WIDEN_WORD;	/* brb -> brw, bsbb -> bsbw */
      md_number_to_chars (addressP, target_address - (address_of_var + 2), 2);
      extension = 2;
      break;

    case ENCODE_RELAX (STATE_ALWAYS_BRANCH, STATE_LONG):
      opcodeP[0] += VAX_WIDEN_LONG;	/* brb -> jmp, bsbb -> jsb */
      addressP[0] = VAX_PC_RELATIVE_MODE;
      md_number_to_chars (addressP + 1, target_address - (address_of_var + 5), 4);
      extension = 5;
      break;

    case ENCODE_RELAX (STATE_COMPLEX_BRANCH, STATE_WORD):
      md_number_to_chars (addressP, target_address - (address_of_var + 2), 2);
      extension = 2;
      break;

    case ENCODE_RELAX (STATE_COMPLEX_BRANCH, STATE_LONG):
      addressP[0] = 2;
      addressP[1] = 0;
      addressP[2] = VAX_BRB;
      addressP[3] = 6;
      addressP[4] = VAX_JMP;
      addressP[5] = VAX_PC_RELATIVE_MODE;
      md_number_to_chars (addressP + 6, target_address, 4);
      extension = 10;
      break;

    case ENCODE_RELAX (STATE_COMPLEX_HOP, STATE_BYTE):
      addressP[0] = target_address - (address_of_var + 1);
      extension = 1;
      break;

    case ENCODE_RELAX (STATE_COMPLEX_HOP, STATE_WORD):
      addressP[0] = 2;
      addressP[1] = VAX_BRB;
      addressP[2] = 3;
      addressP[3] = VAX_BRW;
      md_number_to_chars (addressP + 4, target_address - (address_of_var + 6), 2);
      extension = 6;
      break;

    case ENCODE_RELAX (STATE_COMPLEX_HOP, STATE_LONG):
      addressP[0] = 2;
      addressP[1] = VAX_BRB;
      addressP[2] = 6;
      addressP[3] = VAX_JMP;
      addressP[4] = VAX_PC_RELATIVE_MODE;
      md_number_to_chars (addressP + 5, target_address, 4);
      extension = 9;
      break;

    default:
      BAD_CASE (fragP->fr_subtype);
      break;
    }
  fragP->fr_fix += extension;
}

/* the bit-field entries in the relocation_info struct plays hell
   with the byte-order problems of cross-assembly.  So as a hack,
   I added this mach. dependent ri twiddler.  Ugly, but it gets
   you there. -KWK */
/* on vax: first 4 bytes are normal unsigned long, next three bytes
   are symbolnum, least sig. byte first.  Last byte is broken up with
   the upper nibble as nuthin, bit 3 as extern, bits 2 & 1 as length, and
   bit 0 as pcrel. */
void 
md_ri_to_chars (ri_p, ri)
     struct relocation_info *ri_p, ri;
{
  unsigned char the_bytes[8];

  /* this is easy */
  md_number_to_chars (the_bytes, ri.r_address, sizeof (ri.r_address));
  /* now the fun stuff */
  the_bytes[6] = (ri.r_symbolnum >> 16) & 0x0ff;
  the_bytes[5] = (ri.r_symbolnum >> 8) & 0x0ff;
  the_bytes[4] = ri.r_symbolnum & 0x0ff;
  the_bytes[7] = (((ri.r_extern << 3) & 0x08) | ((ri.r_length << 1) & 0x06) |
		  ((ri.r_pcrel << 0) & 0x01)) & 0x0F;
  /* now put it back where you found it */
  bcopy (the_bytes, (char *) ri_p, sizeof (struct relocation_info));
}



 /* JF this used to be a separate file */
/* vax_ins_parse.c - a part for a VAX assembler */

/* Copyright (C) 1987 Free Software Foundtation, Inc */

/*
 *       BUGS, GRIPES,  APOLOGIA, etc.
 *
 * The opcode table 'votstrs' needs to be sorted on opcode frequency.
 * That is, AFTER we hash it with hash_...(), we want most-used opcodes
 * to come out of the hash table faster.
 *
 * I am sorry to inflict
 * yet another VAX assembler on the world, but RMS says we must
 * do everything from scratch, to prevent pin-heads restricting
 * this software.
 */

/*
 * This is a vaguely modular set of routines in C to parse VAX
 * assembly code using DEC mnemonics. It is NOT un*x specific.
 *
 * The idea here is that the assembler has taken care of all:
 *   labels
 *   macros
 *   listing
 *   pseudo-ops
 *   line continuation
 *   comments
 *   condensing any whitespace down to exactly one space
 * and all we have to do is parse 1 line into a vax instruction
 * partially formed. We will accept a line, and deliver:
 *   an error message (hopefully empty)
 *   a skeleton VAX instruction (tree structure)
 *   textual pointers to all the operand expressions
 *   a warning message that notes a silly operand (hopefully empty)
 */

/*
 *		E D I T   H I S T O R Y
 *
 * 17may86 Dean Elsner. Bug if line ends immediately after opcode.
 * 30apr86 Dean Elsner. New vip_op() uses arg block so change call.
 *  6jan86 Dean Elsner. Crock vip_begin() to call vip_op_defaults().
 *  2jan86 Dean Elsner. Invent synthetic opcodes.
 *	Widen vax_opcodeT to 32 bits. Use a bit for VIT_OPCODE_SYNTHETIC,
 *	which means this is not a real opcode, it is like a macro; it will
 *	be relax()ed into 1 or more instructions.
 *	Use another bit for VIT_OPCODE_SPECIAL if the op-code is not optimised
 *	like a regular branch instruction. Option added to vip_begin():
 *	exclude	synthetic opcodes. Invent synthetic_votstrs[].
 * 31dec85 Dean Elsner. Invent vit_opcode_nbytes.
 *	Also make vit_opcode into a char[]. We now have n-byte vax opcodes,
 *	so caller's don't have to know the difference between a 1-byte & a
 *	2-byte op-code. Still need vax_opcodeT concept, so we know how
 *	big an object must be to hold an op.code.
 * 30dec85 Dean Elsner. Widen typedef vax_opcodeT in "vax-inst.h"
 *	because vax opcodes may be 16 bits. Our crufty C compiler was
 *	happily initialising 8-bit vot_codes with 16-bit numbers!
 *	(Wouldn't the 'phone company like to compress data so easily!)
 * 29dec85 Dean Elsner. New static table vax_operand_width_size[].
 *	Invented so we know hw many bytes a "I^#42" needs in its immediate
 *	operand. Revised struct vop in "vax-inst.h": explicitly include
 *	byte length of each operand, and it's letter-code datum type.
 * 17nov85 Dean Elsner. Name Change.
 *	Due to ar(1) truncating names, we learned the hard way that
 *	"vax-inst-parse.c" -> "vax-inst-parse." dropping the "o" off
 *	the archived object name. SO... we shortened the name of this
 *	source file, and changed the makefile.
 */

/* #include <stdio.h> JF for one big happy file */

/* JF #include "vax-inst.h"		/* define the tree we parse it into */


static char *op_hash = NULL;	/* handle of the OPCODE hash table */
 /* NULL means any use before vip_begin() */
 /* will crash */

/*
 * In:	1 character, from "bdfghloqpw" being the data-type of an operand
 *	of a vax instruction.
 *
 * Out:	the length of an operand of that type, in bytes.
 *	Special branch operands types "-?!" have length 0.
 */

static const short int vax_operand_width_size[256] =
{

#define _ 0
  _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _,
  _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _,
  _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _,
  _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _,
  _, _, 1, _, 8, _, 4, 8, 16, _, _, _, 4, _, _, 16,	/* ..b.d.fgh...l..o */
  _, 8, _, _, _, _, _, 2, _, _, _, _, _, _, _, _,	/* .q.....w........ */
  _, _, 1, _, 8, _, 4, 8, 16, _, _, _, 4, _, _, 16,	/* ..b.d.fgh...l..o */
  _, 8, _, _, _, _, _, 2, _, _, _, _, _, _, _, _,	/* .q.....w........ */
  _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _,
  _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _,
  _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _,
  _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _,
  _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _,
  _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _,
  _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _,
  _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _};
#undef _

/*
 * This perversion encodes all the vax opcodes as a bunch of strings.
 * RMS says we should build our hash-table at run-time. Hmm.
 * Please would someone arrange these in decreasing frequency of opcode?
 * Because of the way hash_...() works, the most frequently used opcode
 * should be textually first and so on.
 *
 * Input for this table was 'vax.opcodes', awk(1)ed by 'vax.opcodes.c.awk' .
 * So change 'vax.opcodes', then re-generate this table.
 */

#include "vax-opcode.h"

/*
 * This is a table of optional op-codes. All of them represent
 * 'synthetic' instructions that seem popular.
 *
 * Here we make some pseudo op-codes. Every code has a bit set to say
 * it is synthetic. This lets you catch them if you want to
 * ban these opcodes. They are mnemonics for "elastic" instructions
 * that are supposed to assemble into the fewest bytes needed to do a
 * branch, or to do a conditional branch, or whatever.
 *
 * The opcode is in the usual place [low-order n*8 bits]. This means
 * that if you mask off the bucky bits, the usual rules apply about
 * how long the opcode is.
 *
 * All VAX branch displacements come at the end of the instruction.
 * For simple branches (1-byte opcode + 1-byte displacement) the last
 * operand is coded 'b?' where the "data type" '?' is a clue that we
 * may reverse the sense of the branch (complement lowest order bit)
 * and branch around a jump. This is by far the most common case.
 * That is why the VIT_OPCODE_SYNTHETIC bit is set: it says this is
 * a 0-byte op-code followed by 2 or more bytes of operand address.
 *
 * If the op-code has VIT_OPCODE_SPECIAL set, then we have a more unusual
 * case.
 *
 * For JBSB & JBR the treatment is the similar, except (1) we have a 'bw'
 * option before (2) we can directly JSB/JMP because there is no condition.
 * These operands have 'b-' as their access/data type.
 *
 * That leaves a bunch of random opcodes: JACBx, JxOBxxx. In these
 * cases, we do the same idea. JACBxxx are all marked with a 'b!'
 * JAOBxxx & JSOBxxx are marked with a 'b:'.
 *
 */
#if (VIT_OPCODE_SYNTHETIC != 0x80000000)
You have just broken the encoding below, which assumes the sign bit
 means 'I am an imaginary instruction'.
#endif

#if (VIT_OPCODE_SPECIAL != 0x40000000)
 You have just broken the encoding below, which assumes the 0x40 M bit means
 'I am not to be "optimised" the way normal branches are'.
#endif

static const struct vot
 synthetic_votstrs[] =
{
  {"jbsb",
   {"b-", 0xC0000010}},		/* BSD 4.2 */
 /* jsb used already */
  {"jbr",
   {"b-", 0xC0000011}},		/* BSD 4.2 */
  {"jr",
   {"b-", 0xC0000011}},		/* consistent */
  {"jneq",
   {"b?", 0x80000012}},
  {"jnequ",
   {"b?", 0x80000012}},
  {"jeql",
   {"b?", 0x80000013}},
  {"jeqlu",
   {"b?", 0x80000013}},
  {"jgtr",
   {"b?", 0x80000014}},
  {"jleq",
   {"b?", 0x80000015}},
 /* un-used opcodes here */
  {"jgeq",
   {"b?", 0x80000018}},
  {"jlss",
   {"b?", 0x80000019}},
  {"jgtru",
   {"b?", 0x8000001a}},
  {"jlequ",
   {"b?", 0x8000001b}},
  {"jvc",
   {"b?", 0x8000001c}},
  {"jvs",
   {"b?", 0x8000001d}},
  {"jgequ",
   {"b?", 0x8000001e}},
  {"jcc",
   {"b?", 0x8000001e}},
  {"jlssu",
   {"b?", 0x8000001f}},
  {"jcs",
   {"b?", 0x8000001f}},

  {"jacbw",
   {"rwrwmwb!", 0xC000003d}},
  {"jacbf",
   {"rfrfmfb!", 0xC000004f}},
  {"jacbd",
   {"rdrdmdb!", 0xC000006f}},
  {"jacbb",
   {"rbrbmbb!", 0xC000009d}},
  {"jacbl",
   {"rlrlmlb!", 0xC00000f1}},
  {"jacbg",
   {"rgrgmgb!", 0xC0004ffd}},
  {"jacbh",
   {"rhrhmhb!", 0xC0006ffd}},

  {"jbs",
   {"rlvbb?", 0x800000e0}},
  {"jbc",
   {"rlvbb?", 0x800000e1}},
  {"jbss",
   {"rlvbb?", 0x800000e2}},
  {"jbcs",
   {"rlvbb?", 0x800000e3}},
  {"jbsc",
   {"rlvbb?", 0x800000e4}},
  {"jbcc",
   {"rlvbb?", 0x800000e5}},
  {"jbssi",
   {"rlvbb?", 0x800000e6}},
  {"jbcci",
   {"rlvbb?", 0x800000e7}},
  {"jlbs",
   {"rlb?", 0x800000e8}},	/* JF changed from rlvbb? */
  {"jlbc",
   {"rlb?", 0x800000e9}},	/* JF changed from rlvbb? */

  {"jaoblss",
   {"rlmlb:", 0xC00000f2}},
  {"jaobleq",
   {"rlmlb:", 0xC00000f3}},
  {"jsobgeq",
   {"mlb:", 0xC00000f4}},	/* JF was rlmlb: */
  {"jsobgtr",
   {"mlb:", 0xC00000f5}},	/* JF was rlmlb: */

/* CASEx has no branch addresses in our conception of it. */