outcoff.c

开源的nasm编译器源码,研究编译器原理很有帮且

/* outcoff.c	output routines for the Netwide Assembler to produce
 *		COFF object files (for DJGPP and Win32)
 *
 * The Netwide Assembler is copyright (C) 1996 Simon Tatham and
 * Julian Hall. All rights reserved. The software is
 * redistributable under the licence given in the file "Licence"
 * distributed in the NASM archive.
 */

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <ctype.h>
#include <time.h>

#include "nasm.h"
#include "nasmlib.h"
#include "outform.h"

#if defined(OF_COFF) || defined(OF_WIN32)

/*
 * Notes on COFF:
 *
 * (0) When I say `standard COFF' below, I mean `COFF as output and
 * used by DJGPP'. I assume DJGPP gets it right.
 *
 * (1) Win32 appears to interpret the term `relative relocation'
 * differently from standard COFF. Standard COFF understands a
 * relative relocation to mean that during relocation you add the
 * address of the symbol you're referencing, and subtract the base
 * address of the section you're in. Win32 COFF, by contrast, seems
 * to add the address of the symbol and then subtract the address
 * of THE BYTE AFTER THE RELOCATED DWORD. Hence the two formats are
 * subtly incompatible.
 *
 * (2) Win32 doesn't bother putting any flags in the header flags
 * field (at offset 0x12 into the file).
 *
 * (3) Win32 uses some extra flags into the section header table:
 * it defines flags 0x80000000 (writable), 0x40000000 (readable)
 * and 0x20000000 (executable), and uses them in the expected
 * combinations. It also defines 0x00100000 through 0x00700000 for
 * section alignments of 1 through 64 bytes.
 *
 * (4) Both standard COFF and Win32 COFF seem to use the DWORD
 * field directly after the section name in the section header
 * table for something strange: they store what the address of the
 * section start point _would_ be, if you laid all the sections end
 * to end starting at zero. Dunno why. Microsoft's documentation
 * lists this field as "Virtual Size of Section", which doesn't
 * seem to fit at all. In fact, Win32 even includes non-linked
 * sections such as .drectve in this calculation.
 *
 * (5) Standard COFF does something very strange to common
 * variables: the relocation point for a common variable is as far
 * _before_ the variable as its size stretches out _after_ it. So
 * we must fix up common variable references. Win32 seems to be
 * sensible on this one.
 */

/* Flag which version of COFF we are currently outputting. */
static int win32;

struct Reloc {
    struct Reloc *next;
    long address;		       /* relative to _start_ of section */
    long symbol;		       /* symbol number */
    enum {
	SECT_SYMBOLS,
	ABS_SYMBOL,
	REAL_SYMBOLS
    } symbase;			       /* relocation for symbol number :) */
    int relative;		       /* TRUE or FALSE */
};

struct Symbol {
    char name[9];
    long strpos;		       /* string table position of name */
    int section;		       /* section number where it's defined
					* - in COFF codes, not NASM codes */
    int is_global;		       /* is it a global symbol or not? */
    long value;			       /* address, or COMMON variable size */
};

static FILE *coffp;
static efunc error;
static char coff_infile[FILENAME_MAX];

struct Section {
    struct SAA *data;
    unsigned long len;
    int nrelocs;
    long index;
    struct Reloc *head, **tail;
    unsigned long flags;	       /* section flags */
    char name[9];
    long pos, relpos;
};

#define TEXT_FLAGS (win32 ? 0x60500020L : 0x20L)
#define DATA_FLAGS (win32 ? 0xC0300040L : 0x40L)
#define BSS_FLAGS (win32 ? 0xC0300080L : 0x80L)
#define INFO_FLAGS 0x00100A00L
#define RDATA_FLAGS (win32 ? 0x40400040L : 0x40L)

#define SECT_DELTA 32
static struct Section **sects;
static int nsects, sectlen;

static struct SAA *syms;
static unsigned long nsyms;

static long def_seg;

static int initsym;

static struct RAA *bsym, *symval;

static struct SAA *strs;
static unsigned long strslen;

static void coff_gen_init(FILE *, efunc);
static void coff_sect_write (struct Section *, const unsigned char *,
			     unsigned long);
static void coff_write (void);
static void coff_section_header (char *, long, long, long, long, int, long);
static void coff_write_relocs (struct Section *);
static void coff_write_symbols (void);

static void coff_win32_init(FILE *fp,  efunc errfunc,
			    ldfunc ldef, evalfunc eval) 
{
    win32 = TRUE;
    (void) ldef;		       /* placate optimisers */
    coff_gen_init(fp, errfunc);
}

static void coff_std_init(FILE *fp, efunc errfunc, ldfunc ldef, evalfunc eval) 
{
    win32 = FALSE;
    (void) ldef;		       /* placate optimisers */
    coff_gen_init(fp, errfunc);
}

static void coff_gen_init(FILE *fp, efunc errfunc) 
{

    coffp = fp;
    error = errfunc;
    sects = NULL;
    nsects = sectlen = 0;
    syms = saa_init((long)sizeof(struct Symbol));
    nsyms = 0;
    bsym = raa_init();
    symval = raa_init();
    strs = saa_init(1L);
    strslen = 0;
    def_seg = seg_alloc();
}

static void coff_cleanup(int debuginfo) 
{
    struct Reloc *r;
    int i;

    (void) debuginfo;

    coff_write();
    fclose (coffp);
    for (i=0; i<nsects; i++) {
	if (sects[i]->data)
	    saa_free (sects[i]->data);
	while (sects[i]->head) {
	    r = sects[i]->head;
	    sects[i]->head = sects[i]->head->next;
	    nasm_free (r);
	}
	nasm_free (sects[i]);
    }
    nasm_free (sects);
    saa_free (syms);
    raa_free (bsym);
    raa_free (symval);
    saa_free (strs);
}

static int coff_make_section (char *name, unsigned long flags) 
{
    struct Section *s;

    s = nasm_malloc (sizeof(*s));

    if (flags != BSS_FLAGS)
	s->data = saa_init (1L);
    else
	s->data = NULL;
    s->head = NULL;
    s->tail = &s->head;
    s->len = 0;
    s->nrelocs = 0;
    if (!strcmp(name, ".text"))
	s->index = def_seg;
    else
	s->index = seg_alloc();
    strncpy (s->name, name, 8);
    s->name[8] = '\0';
    s->flags = flags;

    if (nsects >= sectlen)
	sects = nasm_realloc (sects, (sectlen += SECT_DELTA)*sizeof(*sects));
    sects[nsects++] = s;

    return nsects-1;
}

static long coff_section_names (char *name, int pass, int *bits) 
{
    char *p;
    unsigned long flags, align_and = ~0L, align_or = 0L;
    int i;

    /*
     * Default is 32 bits.
     */
    if (!name)
	*bits = 32;

    if (!name)
	return def_seg;

    p = name;
    while (*p && !isspace(*p)) p++;
    if (*p) *p++ = '\0';
    if (strlen(name) > 8) {
	error (ERR_WARNING, "COFF section names limited to 8 characters:"
	       " truncating");
	name[8] = '\0';
    }
    flags = 0;

    while (*p && isspace(*p)) p++;
    while (*p) {
	char *q = p;
	while (*p && !isspace(*p)) p++;
	if (*p) *p++ = '\0';
	while (*p && isspace(*p)) p++;

	if (!nasm_stricmp(q, "code") || !nasm_stricmp(q, "text")) {
	    flags = TEXT_FLAGS;
	} else if (!nasm_stricmp(q, "data")) {
	    flags = DATA_FLAGS;
	} else if (!nasm_stricmp(q, "rdata")) {
	    if (win32)
	    flags = RDATA_FLAGS;
	    else {
		flags = DATA_FLAGS;    /* gotta do something */
		error (ERR_NONFATAL, "standard COFF does not support"
		       " read-only data sections");
	    }
	} else if (!nasm_stricmp(q, "bss")) {
	    flags = BSS_FLAGS;
	} else if (!nasm_stricmp(q, "info")) {
	    if (win32)
		flags = INFO_FLAGS;
	    else {
		flags = DATA_FLAGS;    /* gotta do something */
		error (ERR_NONFATAL, "standard COFF does not support"
		       " informational sections");
	    }
	} else if (!nasm_strnicmp(q,"align=",6)) {
	    if (!win32)
		error (ERR_NONFATAL, "standard COFF does not support"
		       " section alignment specification");
	    else {
		if (q[6+strspn(q+6,"0123456789")])
		    error(ERR_NONFATAL, "argument to `align' is not numeric");
		else {
		    unsigned int align = atoi(q+6);
		    if (!align || ((align-1) & align))
			error(ERR_NONFATAL, "argument to `align' is not a"
			      " power of two");
		    else if (align > 64)
			error(ERR_NONFATAL, "Win32 cannot align sections"
			      " to better than 64-byte boundaries");
		    else {
			align_and = ~0x00F00000L;
			align_or = (align == 1 ? 0x00100000L :
				    align == 2 ? 0x00200000L :
				    align == 4 ? 0x00300000L :
				    align == 8 ? 0x00400000L :
				    align == 16 ? 0x00500000L :
				    align == 32 ? 0x00600000L : 0x00700000L);
		    }
		}
	    }
	}
    }

    for (i=0; i<nsects; i++)
	if (!strcmp(name, sects[i]->name))
	    break;
    if (i == nsects) {
	if (!flags) {
	    if (!strcmp(name, ".data"))
		flags = DATA_FLAGS;
	    else if (!strcmp(name, ".rdata"))
		flags = RDATA_FLAGS;
	    else if (!strcmp(name, ".bss"))
		flags = BSS_FLAGS;
	    else
		flags = TEXT_FLAGS;
	}
	i = coff_make_section (name, flags);
	if (flags)
	    sects[i]->flags = flags;
	sects[i]->flags &= align_and;
	sects[i]->flags |= align_or;
    } else if (pass == 1) {
	if (flags)
	    error (ERR_WARNING, "section attributes ignored on"
		   " redeclaration of section `%s'", name);
    }

    return sects[i]->index;
}

static void coff_deflabel (char *name, long segment, long offset,
			   int is_global, char *special) 
{
    int pos = strslen+4;
    struct Symbol *sym;

    if (special)
	error (ERR_NONFATAL, "binary format does not support any"
	       " special symbol types");

    if (name[0] == '.' && name[1] == '.' && name[2] != '@') {
	error (ERR_NONFATAL, "unrecognised special symbol `%s'", name);
	return;
    }

    if (strlen(name) > 8) {
	saa_wbytes (strs, name, (long)(1+strlen(name)));
	strslen += 1+strlen(name);
    } else
	pos = -1;

    sym = saa_wstruct (syms);

    sym->strpos = pos;
    if (pos == -1)
	strcpy (sym->name, name);
    sym->is_global = !!is_global;
    if (segment == NO_SEG)
	sym->section = -1;      /* absolute symbol */
    else {
	int i;
	sym->section = 0;
	for (i=0; i<nsects; i++)
	    if (segment == sects[i]->index) {
		sym->section = i+1;
		break;
	    }
	if (!sym->section)
	    sym->is_global = TRUE;
    }
    if (is_global == 2)
	sym->value = offset;
    else
	sym->value = (sym->section == 0 ? 0 : offset);

    /*
     * define the references from external-symbol segment numbers
     * to these symbol records.
     */
    if (sym->section == 0)
	bsym = raa_write (bsym, segment, nsyms);

    if (segment != NO_SEG)
	symval = raa_write (symval, segment, sym->section ? 0 : sym->value);

    nsyms++;
}

static long coff_add_reloc (struct Section *sect, long segment,
			    int relative) 
{
    struct Reloc *r;

    r = *sect->tail = nasm_malloc(sizeof(struct Reloc));
    sect->tail = &r->next;
    r->next = NULL;

    r->address = sect->len;
    if (segment == NO_SEG)
	r->symbol = 0, r->symbase = ABS_SYMBOL;
    else {
	int i;
	r->symbase = REAL_SYMBOLS;
	for (i=0; i<nsects; i++)
	    if (segment == sects[i]->index) {
		r->symbol = i*2;
		r->symbase = SECT_SYMBOLS;
		break;
	    }
	if (r->symbase == REAL_SYMBOLS)
	    r->symbol = raa_read (bsym, segment);
    }
    r->relative = relative;

    sect->nrelocs++;

    /*
     * Return the fixup for standard COFF common variables.
     */
    if (r->symbase == REAL_SYMBOLS && !win32)
	return raa_read (symval, segment);
    else
	return 0;
}

static void coff_out (long segto, const void *data, unsigned long type,
		      long segment, long wrt) 
{
    struct Section *s;
    long realbytes = type & OUT_SIZMASK;
    unsigned char mydata[4], *p;
    int i;

    if (wrt != NO_SEG) {
	wrt = NO_SEG;		       /* continue to do _something_ */
	error (ERR_NONFATAL, "WRT not supported by COFF output formats");
    }

    type &= OUT_TYPMASK;

    /*
     * handle absolute-assembly (structure definitions)
     */