outbin.c

开源的nasm编译器源码,研究编译器原理很有帮且

/* outbin.c output routines for the Netwide Assembler to produce
 *    flat-form binary files
 *
 * 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.
 */

/* This is the extended version of NASM's original binary output
 * format.  It is backward compatible with the original BIN format,
 * and contains support for multiple sections and advanced section
 * ordering.
 *
 * Feature summary:
 *
 * - Users can create an arbitrary number of sections; they are not
 *   limited to just ".text", ".data", and ".bss".
 *
 * - Sections can be either progbits or nobits type.
 *
 * - You can specify that they be aligned at a certian boundary
 *   following the previous section ("align="), or positioned at an
 *   arbitrary byte-granular location ("start=").
 *
 * - You can specify a "virtual" start address for a section, which
 *   will be used for the calculation for all address references
 *   with respect to that section ("vstart=").
 *
 * - The ORG directive, as well as the section/segment directive
 *   arguments ("align=", "start=", "vstart="), can take a critical
 *   expression as their value.  For example: "align=(1 << 12)".
 *
 * - You can generate map files using the 'map' directive.
 *
 */

/* Uncomment the following define if you want sections to adapt
 * their progbits/nobits state depending on what type of
 * instructions are issued, rather than defaulting to progbits.
 * Note that this behavior violates the specification.

#define ABIN_SMART_ADAPT

*/

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

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

#ifdef OF_BIN

struct ofmt *bin_get_ofmt(); /* Prototype goes here since no header file. */

static FILE *fp, *rf = NULL;
static efunc error;

/* Section flags keep track of which attributes the user has defined. */
#define START_DEFINED       0x001
#define ALIGN_DEFINED       0x002
#define FOLLOWS_DEFINED     0x004
#define VSTART_DEFINED      0x008
#define VALIGN_DEFINED      0x010
#define VFOLLOWS_DEFINED    0x020
#define TYPE_DEFINED        0x040
#define TYPE_PROGBITS       0x080
#define TYPE_NOBITS         0x100

/* This struct is used to keep track of symbols for map-file generation. */
static struct bin_label
{  char * name;
   struct bin_label *next;
} *no_seg_labels, **nsl_tail;

static struct Section
{  char *name;
   struct SAA * contents;
   long length;          /* section length in bytes */

/* Section attributes */
   int flags;            /* see flag definitions above */
   unsigned long align;  /* section alignment */
   unsigned long valign; /* notional section alignment */
   unsigned long start;  /* section start address */
   unsigned long vstart; /* section virtual start address */
   char *follows;        /* the section that this one will follow */
   char *vfollows;       /* the section that this one will notionally follow */
   long start_index;     /* NASM section id for non-relocated version */
   long vstart_index;    /* the NASM section id */

   struct bin_label *labels;  /* linked-list of label handles for map output. */
   struct bin_label **labels_end;  /* Holds address of end of labels list. */
   struct Section *ifollows;  /* Points to previous section (implicit follows). */
   struct Section *next;  /* This links sections with a defined start address. */

/* The extended bin format allows for sections to have a "virtual"
 * start address.  This is accomplished by creating two sections:
 * one beginning at the Load Memory Address and the other beginning
 * at the Virtual Memory Address.  The LMA section is only used to
 * define the section.<section_name>.start label, but there isn't
 * any other good way for us to handle that label.
 */

} *sections, *last_section;

static struct Reloc {
    struct Reloc *next;
    long posn;
    long bytes;
    long secref;
    long secrel;
    struct Section *target;
} *relocs, **reloctail;

extern char *stdscan_bufptr;
extern int lookup_label (char *label, long *segment, long *offset);

static unsigned char format_mode;  /* 0 = original bin, 1 = extended bin */
static long current_section;       /* only really needed if format_mode = 0 */
static unsigned long origin;
static int origin_defined;

/* Stuff we need for map-file generation. */
#define MAP_ORIGIN       1
#define MAP_SUMMARY      2
#define MAP_SECTIONS     4
#define MAP_SYMBOLS      8
static int map_control = 0;
static char *infile, *outfile;

static const char *bin_stdmac[] = {
   "%define __SECT__ [section .text]",
   "%imacro org 1+.nolist",
   "[org %1]",
   "%endmacro",
   "%macro __NASM_CDecl__ 1",
   "%endmacro",
   NULL
};

static void add_reloc(struct Section *s, long bytes, long secref, long secrel)
{   struct Reloc *r;

    r = *reloctail = nasm_malloc(sizeof(struct Reloc));
    reloctail = &r->next;
    r->next = NULL;
    r->posn = s->length;
    r->bytes = bytes;
    r->secref = secref;
    r->secrel = secrel;
    r->target = s;
}

static struct Section *find_section_by_name(const char *name)
{  struct Section *s;

   for (s = sections; s; s = s->next)
      if (!strcmp(s->name,name)) break;
   return s;
}

static struct Section *find_section_by_index(long index)
{  struct Section *s;

   for (s = sections; s; s = s->next)
      if ((index == s->vstart_index) || (index == s->start_index))
        break;
   return s;
}

static struct Section * create_section(char *name)
{  /* Create a new section. */
   last_section->next = nasm_malloc(sizeof(struct Section));
   last_section->next->ifollows = last_section;
   last_section = last_section->next;
   last_section->labels = NULL;
   last_section->labels_end = &(last_section->labels);

   /* Initialize section attributes. */
   last_section->name = nasm_strdup(name);
   last_section->contents = saa_init(1L);
   last_section->follows = last_section->vfollows = 0;
   last_section->length = 0;
   last_section->flags = 0;
   last_section->next = NULL;

   /* Register our sections with NASM. */
   last_section->vstart_index = seg_alloc();
   last_section->start_index  = seg_alloc();
   return last_section;
}

static void bin_cleanup (int debuginfo)
{  struct Section *g, **gp;
   struct Section *gs = NULL, **gsp;
   struct Section *s, **sp;
   struct Section *nobits = NULL, **nt;
   struct Section * last_progbits;
   struct bin_label *l;
   struct Reloc *r;
   unsigned long pend;
   int h;

#ifdef DEBUG
   fprintf(stdout, "bin_cleanup: Sections were initially referenced in this order:\n");
   for (h = 0, s = sections; s; h++, s = s->next)
      fprintf(stdout, "%i. %s\n", h, s->name);
#endif

   /* Assembly has completed, so now we need to generate the output file.
    * Step 1: Separate progbits and nobits sections into separate lists.
    * Step 2: Sort the progbits sections into their output order.
    * Step 3: Compute start addresses for all progbits sections.
    * Step 4: Compute vstart addresses for all sections.
    * Step 5: Apply relocations.
    * Step 6: Write the sections' data to the output file.
    * Step 7: Generate the map file.
    * Step 8: Release all allocated memory.
    */

   /* To do: Smart section-type adaptation could leave some empty sections
    * without a defined type (progbits/nobits).  Won't fix now since this
    * feature will be disabled.  */


   /* Step 1: Split progbits and nobits sections into separate lists. */

   nt = &nobits;
   /* Move nobits sections into a separate list.  Also pre-process nobits
    * sections' attributes. */
   for (sp = &sections->next, s = sections->next; s; s = *sp)
   {  /* Skip progbits sections. */
      if (s->flags & TYPE_PROGBITS)
      {  sp = &s->next; continue;
      }
      /* Do some special pre-processing on nobits sections' attributes. */
      if (s->flags & (START_DEFINED | ALIGN_DEFINED | FOLLOWS_DEFINED))
      {  /* Check for a mixture of real and virtual section attributes. */
         if (s->flags & (VSTART_DEFINED | VALIGN_DEFINED | VFOLLOWS_DEFINED))
            error(ERR_FATAL, "cannot mix real and virtual attributes"
               " in nobits section (%s)", s->name);
         /* Real and virtual attributes mean the same thing for nobits sections. */
         if (s->flags & START_DEFINED)
         {  s->vstart = s->start; s->flags |= VSTART_DEFINED;
         }
         if (s->flags & ALIGN_DEFINED)
         {  s->valign = s->align; s->flags |= VALIGN_DEFINED;
         }
         if (s->flags & FOLLOWS_DEFINED)
         {  s->vfollows = s->follows; s->flags |= VFOLLOWS_DEFINED;
            s->flags &= ~FOLLOWS_DEFINED;
         }
      }
      /* Every section must have a start address. */
      if (s->flags & VSTART_DEFINED)
      {  s->start = s->vstart; s->flags |= START_DEFINED;
      }
      /* Move the section into the nobits list. */
      *sp = s->next; s->next = NULL;
      *nt = s; nt = &s->next;
   }

   /* Step 2: Sort the progbits sections into their output order. */

   /* In Step 2 we move around sections in groups.  A group
    * begins with a section (group leader) that has a user-
    * defined start address or follows section.  The remainder
    * of the group is made up of the sections that implicitly
    * follow the group leader (i.e., they were defined after
    * the group leader and were not given an explicit start
    * address or follows section by the user). */

   /* For anyone attempting to read this code:
    * g (group) points to a group of sections, the first one of which has
    *   a user-defined start address or follows section.
    * gp (g previous) holds the location of the pointer to g.
    * gs (g scan) is a temp variable that we use to scan to the end of the group.
    * gsp (gs previous) holds the location of the pointer to gs.
    * nt (nobits tail) points to the nobits section-list tail.
    */

   /* Link all 'follows' groups to their proper position.  To do
    * this we need to know three things: the start of the group
    * to relocate (g), the section it is following (s), and the
    * end of the group we're relocating (gs). */
   for (gp = &sections, g = sections; g; g = gs)
   {  /* Find the next follows group that is out of place (g). */
      if (!(g->flags & FOLLOWS_DEFINED))
      {  while (g->next)
         {  if ((g->next->flags & FOLLOWS_DEFINED) &&
               strcmp(g->name, g->next->follows)) break;
            g = g->next;
         }
         if (!g->next) break;
         gp = &g->next; g = g->next;
      }
      /* Find the section that this group follows (s). */
      for (sp = &sections, s = sections;
         s && strcmp(s->name, g->follows);
         sp = &s->next, s = s->next);
      if (!s) error(ERR_FATAL, "section %s follows an invalid or"
         " unknown section (%s)", g->name, g->follows);
      if (s->next && (s->next->flags & FOLLOWS_DEFINED) &&
         !strcmp(s->name, s->next->follows))
         error(ERR_FATAL, "sections %s and %s can't both follow"
            " section %s", g->name, s->next->name, s->name);
      /* Find the end of the current follows group (gs). */
      for (gsp = &g->next, gs = g->next;
         gs && (gs != s) && !(gs->flags & START_DEFINED);
         gsp = &gs->next, gs = gs->next)
      {  if (gs->next && (gs->next->flags & FOLLOWS_DEFINED) &&
            strcmp(gs->name, gs->next->follows))
         {  gsp = &gs->next; gs = gs->next; break;
         }
      }
      /* Re-link the group after its follows section. */
      *gsp = s->next; s->next = g;
      *gp = gs;
   }

   /* Link all 'start' groups to their proper position.  Once
    * again we need to know g, s, and gs (see above).  The main
    * difference is we already know g since we sort by moving
    * groups from the 'unsorted' list into a 'sorted' list (g