mirror of
https://github.com/netwide-assembler/nasm.git
synced 2024-11-27 08:10:07 +08:00
2e53f27e9d
There is absolutely no reason not to include <string.h> globally, and with the inline function for mempcpy() we need it there anyway. Signed-off-by: H. Peter Anvin <hpa@zytor.com>
1392 lines
44 KiB
C
1392 lines
44 KiB
C
/* ----------------------------------------------------------------------- *
|
|
*
|
|
* Copyright 1996-2014 The NASM Authors - All Rights Reserved
|
|
* See the file AUTHORS included with the NASM distribution for
|
|
* the specific copyright holders.
|
|
*
|
|
* Redistribution and use in source and binary forms, with or without
|
|
* modification, are permitted provided that the following
|
|
* conditions are met:
|
|
*
|
|
* * Redistributions of source code must retain the above copyright
|
|
* notice, this list of conditions and the following disclaimer.
|
|
* * Redistributions in binary form must reproduce the above
|
|
* copyright notice, this list of conditions and the following
|
|
* disclaimer in the documentation and/or other materials provided
|
|
* with the distribution.
|
|
*
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
|
|
* CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
|
|
* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
|
|
* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
|
|
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
|
|
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
|
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
|
|
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
|
|
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
|
|
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
|
|
* OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
|
|
* EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
*
|
|
* ----------------------------------------------------------------------- */
|
|
|
|
/*
|
|
* ldrdf.c - RDOFF Object File linker/loader main program.
|
|
*/
|
|
|
|
/*
|
|
* TODO:
|
|
* - enhance search of required export symbols in libraries (now depends
|
|
* on modules order in library)
|
|
* - keep a cache of symbol names in each library module so
|
|
* we don't have to constantly recheck the file
|
|
* - general performance improvements
|
|
*
|
|
* BUGS & LIMITATIONS: this program doesn't support multiple code, data
|
|
* or bss segments, therefore for 16 bit programs whose code, data or BSS
|
|
* segment exceeds 64K in size, it will not work. This program probably
|
|
* won't work if compiled by a 16 bit compiler. Try DJGPP if you're running
|
|
* under DOS. '#define STINGY_MEMORY' may help a little.
|
|
*/
|
|
|
|
#include "compiler.h"
|
|
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
|
|
#include "rdfutils.h"
|
|
#include "symtab.h"
|
|
#include "collectn.h"
|
|
#include "rdlib.h"
|
|
#include "segtab.h"
|
|
#include "nasmlib.h"
|
|
|
|
#define LDRDF_VERSION "1.08"
|
|
|
|
/* #define STINGY_MEMORY */
|
|
|
|
/* =======================================================================
|
|
* Types & macros that are private to this program
|
|
*/
|
|
|
|
struct segment_infonode {
|
|
int dest_seg; /* output segment to be placed into, -1 to
|
|
skip linking this segment */
|
|
int32_t reloc; /* segment's relocation factor */
|
|
};
|
|
|
|
struct modulenode {
|
|
rdffile f; /* the RDOFF file structure */
|
|
struct segment_infonode seginfo[RDF_MAXSEGS]; /* what are we doing
|
|
with each segment? */
|
|
void *header;
|
|
char *name;
|
|
struct modulenode *next;
|
|
int32_t bss_reloc;
|
|
};
|
|
|
|
#include "ldsegs.h"
|
|
|
|
/* ==========================================================================
|
|
* Function prototypes of private utility functions
|
|
*/
|
|
|
|
void processmodule(const char *filename, struct modulenode *mod);
|
|
int allocnewseg(uint16_t type, uint16_t reserved);
|
|
int findsegment(uint16_t type, uint16_t reserved);
|
|
void symtab_add(const char *symbol, int segment, int32_t offset);
|
|
int symtab_get(const char *symbol, int *segment, int32_t *offset);
|
|
|
|
/* =========================================================================
|
|
* Global data structures.
|
|
*/
|
|
|
|
/* a linked list of modules that will be included in the output */
|
|
struct modulenode *modules = NULL;
|
|
struct modulenode *lastmodule = NULL;
|
|
|
|
/* a linked list of libraries to be searched for unresolved imported symbols */
|
|
struct librarynode *libraries = NULL;
|
|
struct librarynode *lastlib = NULL;
|
|
|
|
/* the symbol table */
|
|
void *symtab = NULL;
|
|
|
|
/* objects search path */
|
|
char *objpath = NULL;
|
|
|
|
/* libraries search path */
|
|
char *libpath = NULL;
|
|
|
|
/* file to embed as a generic record */
|
|
char *generic_rec_file = NULL;
|
|
|
|
/* module name to be added at the beginning of output file */
|
|
char *modname_specified = NULL;
|
|
|
|
/* the header of the output file, built up stage by stage */
|
|
rdf_headerbuf *newheader = NULL;
|
|
|
|
/* The current state of segment allocation, including information about
|
|
* which output segment numbers have been allocated, and their types and
|
|
* amount of data which has already been allocated inside them.
|
|
*/
|
|
struct SegmentHeaderRec outputseg[RDF_MAXSEGS];
|
|
int nsegs = 0;
|
|
int32_t bss_length;
|
|
|
|
/* global options which affect how the program behaves */
|
|
struct ldrdfoptions {
|
|
int verbose;
|
|
int align;
|
|
int dynalink;
|
|
int strip;
|
|
int respfile;
|
|
int stderr_redir;
|
|
int objpath;
|
|
int libpath;
|
|
} options;
|
|
|
|
int errorcount = 0; /* determines main program exit status */
|
|
|
|
/* =========================================================================
|
|
* Utility functions
|
|
*/
|
|
|
|
/*
|
|
* initsegments()
|
|
*
|
|
* sets up segments 0, 1, and 2, the initial code data and bss segments
|
|
*/
|
|
static void initsegments(void)
|
|
{
|
|
nsegs = 3;
|
|
outputseg[0].type = 1;
|
|
outputseg[0].number = 0;
|
|
outputseg[0].reserved = 0;
|
|
outputseg[0].length = 0;
|
|
outputseg[1].type = 2;
|
|
outputseg[1].number = 1;
|
|
outputseg[1].reserved = 0;
|
|
outputseg[1].length = 0;
|
|
outputseg[2].type = 0xFFFF; /* reserved segment type */
|
|
outputseg[2].number = 2;
|
|
outputseg[2].reserved = 0;
|
|
outputseg[2].length = 0;
|
|
bss_length = 0;
|
|
}
|
|
|
|
/*
|
|
* loadmodule()
|
|
*
|
|
* Determine the characteristics of a module, and decide what to do with
|
|
* each segment it contains (including determining destination segments and
|
|
* relocation factors for segments that are kept).
|
|
*/
|
|
static void loadmodule(const char *filename)
|
|
{
|
|
if (options.verbose)
|
|
printf("loading `%s'\n", filename);
|
|
|
|
/* allocate a new module entry on the end of the modules list */
|
|
if (!modules) {
|
|
modules = nasm_malloc(sizeof(*modules));
|
|
lastmodule = modules;
|
|
} else {
|
|
lastmodule->next = nasm_malloc(sizeof(*modules));
|
|
lastmodule = lastmodule->next;
|
|
}
|
|
|
|
if (!lastmodule) {
|
|
fprintf(stderr, "ldrdf: out of memory\n");
|
|
exit(1);
|
|
}
|
|
|
|
/* open the file using 'rdfopen', which returns nonzero on error */
|
|
if (rdfopen(&lastmodule->f, filename) != 0) {
|
|
rdfperror("ldrdf", filename);
|
|
exit(1);
|
|
}
|
|
|
|
/*
|
|
* store information about the module, and determine what segments
|
|
* it contains, and what we should do with them (determine relocation
|
|
* factor if we decide to keep them)
|
|
*/
|
|
lastmodule->header = NULL;
|
|
lastmodule->name = nasm_strdup(filename);
|
|
lastmodule->next = NULL;
|
|
|
|
processmodule(filename, lastmodule);
|
|
}
|
|
|
|
/*
|
|
* processmodule()
|
|
*
|
|
* step through each segment, determine what exactly we're doing with
|
|
* it, and if we intend to keep it, determine (a) which segment to
|
|
* put it in and (b) whereabouts in that segment it will end up.
|
|
* (b) is fairly easy, because we're now keeping track of how big each
|
|
* segment in our output file is...
|
|
*/
|
|
void processmodule(const char *filename, struct modulenode *mod)
|
|
{
|
|
struct segconfig sconf;
|
|
int seg, outseg;
|
|
void *header;
|
|
rdfheaderrec *hr;
|
|
int32_t bssamount = 0;
|
|
int bss_was_referenced = 0;
|
|
|
|
memset(&sconf, 0, sizeof sconf);
|
|
|
|
for (seg = 0; seg < mod->f.nsegs; seg++) {
|
|
/*
|
|
* get the segment configuration for this type from the segment
|
|
* table. getsegconfig() is a macro, defined in ldsegs.h.
|
|
*/
|
|
getsegconfig(sconf, mod->f.seg[seg].type);
|
|
|
|
if (options.verbose > 1) {
|
|
printf("%s %04x [%04x:%10s] ", filename,
|
|
mod->f.seg[seg].number, mod->f.seg[seg].type,
|
|
sconf.typedesc);
|
|
}
|
|
/*
|
|
* sconf->dowhat tells us what to do with a segment of this type.
|
|
*/
|
|
switch (sconf.dowhat) {
|
|
case SEG_IGNORE:
|
|
/*
|
|
* Set destination segment to -1, to indicate that this segment
|
|
* should be ignored for the purpose of output, ie it is left
|
|
* out of the linked executable.
|
|
*/
|
|
mod->seginfo[seg].dest_seg = -1;
|
|
if (options.verbose > 1)
|
|
printf("IGNORED\n");
|
|
break;
|
|
|
|
case SEG_NEWSEG:
|
|
/*
|
|
* The configuration tells us to create a new segment for
|
|
* each occurrence of this segment type.
|
|
*/
|
|
outseg = allocnewseg(sconf.mergetype,
|
|
mod->f.seg[seg].reserved);
|
|
mod->seginfo[seg].dest_seg = outseg;
|
|
mod->seginfo[seg].reloc = 0;
|
|
outputseg[outseg].length = mod->f.seg[seg].length;
|
|
if (options.verbose > 1)
|
|
printf("=> %04x:%08"PRIx32" (+%04"PRIx32")\n", outseg,
|
|
mod->seginfo[seg].reloc, mod->f.seg[seg].length);
|
|
break;
|
|
|
|
case SEG_MERGE:
|
|
/*
|
|
* The configuration tells us to merge the segment with
|
|
* a previously existing segment of type 'sconf.mergetype',
|
|
* if one exists. Otherwise a new segment is created.
|
|
* This is handled transparently by 'findsegment()'.
|
|
*/
|
|
outseg = findsegment(sconf.mergetype,
|
|
mod->f.seg[seg].reserved);
|
|
mod->seginfo[seg].dest_seg = outseg;
|
|
|
|
/*
|
|
* We need to add alignment to these segments.
|
|
*/
|
|
if (outputseg[outseg].length % options.align != 0)
|
|
outputseg[outseg].length +=
|
|
options.align -
|
|
(outputseg[outseg].length % options.align);
|
|
|
|
mod->seginfo[seg].reloc = outputseg[outseg].length;
|
|
outputseg[outseg].length += mod->f.seg[seg].length;
|
|
|
|
if (options.verbose > 1)
|
|
printf("=> %04x:%08"PRIx32" (+%04"PRIx32")\n", outseg,
|
|
mod->seginfo[seg].reloc, mod->f.seg[seg].length);
|
|
}
|
|
|
|
}
|
|
|
|
/*
|
|
* extract symbols from the header, and dump them into the
|
|
* symbol table
|
|
*/
|
|
header = nasm_malloc(mod->f.header_len);
|
|
if (!header) {
|
|
fprintf(stderr, "ldrdf: not enough memory\n");
|
|
exit(1);
|
|
}
|
|
if (rdfloadseg(&mod->f, RDOFF_HEADER, header)) {
|
|
rdfperror("ldrdf", filename);
|
|
exit(1);
|
|
}
|
|
|
|
while ((hr = rdfgetheaderrec(&mod->f))) {
|
|
switch (hr->type) {
|
|
case RDFREC_IMPORT: /* imported symbol */
|
|
case RDFREC_FARIMPORT:
|
|
/* Define with seg = -1 */
|
|
symtab_add(hr->i.label, -1, 0);
|
|
break;
|
|
|
|
case RDFREC_GLOBAL:{ /* exported symbol */
|
|
int destseg;
|
|
int32_t destreloc;
|
|
|
|
if (hr->e.segment == 2) {
|
|
bss_was_referenced = 1;
|
|
destreloc = bss_length;
|
|
if (destreloc % options.align != 0)
|
|
destreloc +=
|
|
options.align - (destreloc % options.align);
|
|
destseg = 2;
|
|
} else {
|
|
if ((destseg =
|
|
mod->seginfo[(int)hr->e.segment].dest_seg) == -1)
|
|
continue;
|
|
destreloc = mod->seginfo[(int)hr->e.segment].reloc;
|
|
}
|
|
symtab_add(hr->e.label, destseg, destreloc + hr->e.offset);
|
|
break;
|
|
}
|
|
|
|
case RDFREC_BSS: /* BSS reservation */
|
|
/*
|
|
* first, amalgamate all BSS reservations in this module
|
|
* into one, because we allow this in the output format.
|
|
*/
|
|
bssamount += hr->b.amount;
|
|
break;
|
|
|
|
case RDFREC_COMMON:{ /* Common variable */
|
|
symtabEnt *ste = symtabFind(symtab, hr->c.label);
|
|
|
|
/* Is the symbol already in the table? */
|
|
if (ste)
|
|
break;
|
|
|
|
/* Align the variable */
|
|
if (bss_length % hr->c.align != 0)
|
|
bss_length += hr->c.align - (bss_length % hr->c.align);
|
|
if (options.verbose > 1) {
|
|
printf("%s %04x common '%s' => 0002:%08"PRIx32" (+%04"PRIx32")\n",
|
|
filename, hr->c.segment, hr->c.label,
|
|
bss_length, hr->c.size);
|
|
}
|
|
|
|
symtab_add(hr->c.label, 2, bss_length);
|
|
mod->bss_reloc = bss_length;
|
|
bss_length += hr->c.size;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
if (bssamount != 0 || bss_was_referenced) {
|
|
/*
|
|
* handle the BSS segment - first pad the existing bss length
|
|
* to the correct alignment, then store the length in bss_reloc
|
|
* for this module. Then add this module's BSS length onto
|
|
* bss_length.
|
|
*/
|
|
if (bss_length % options.align != 0)
|
|
bss_length += options.align - (bss_length % options.align);
|
|
|
|
mod->bss_reloc = bss_length;
|
|
if (options.verbose > 1) {
|
|
printf("%s 0002 [ BSS] => 0002:%08"PRIx32" (+%04"PRIx32")\n",
|
|
filename, bss_length, bssamount);
|
|
}
|
|
bss_length += bssamount;
|
|
}
|
|
#ifdef STINGY_MEMORY
|
|
/*
|
|
* we free the header buffer here, to save memory later.
|
|
* this isn't efficient, but probably halves the memory usage
|
|
* of this program...
|
|
*/
|
|
mod->f.header_loc = NULL;
|
|
nasm_free(header);
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
/*
|
|
* Return 1 if a given module is in the list, 0 otherwise.
|
|
*/
|
|
static int lookformodule(const char *name)
|
|
{
|
|
struct modulenode *curr = modules;
|
|
|
|
while (curr) {
|
|
if (!strcmp(name, curr->name))
|
|
return 1;
|
|
curr = curr->next;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
/*
|
|
* allocnewseg()
|
|
* findsegment()
|
|
*
|
|
* These functions manipulate the array of output segments, and are used
|
|
* by processmodule(). allocnewseg() allocates a segment in the array,
|
|
* initialising it to be empty. findsegment() first scans the array for
|
|
* a segment of the type requested, and if one isn't found allocates a
|
|
* new one.
|
|
*/
|
|
int allocnewseg(uint16_t type, uint16_t reserved)
|
|
{
|
|
outputseg[nsegs].type = type;
|
|
outputseg[nsegs].number = nsegs;
|
|
outputseg[nsegs].reserved = reserved;
|
|
outputseg[nsegs].length = 0;
|
|
outputseg[nsegs].offset = 0;
|
|
outputseg[nsegs].data = NULL;
|
|
|
|
return nsegs++;
|
|
}
|
|
|
|
int findsegment(uint16_t type, uint16_t reserved)
|
|
{
|
|
int i;
|
|
|
|
for (i = 0; i < nsegs; i++)
|
|
if (outputseg[i].type == type)
|
|
return i;
|
|
|
|
return allocnewseg(type, reserved);
|
|
}
|
|
|
|
/*
|
|
* symtab_add()
|
|
*
|
|
* inserts a symbol into the global symbol table, which associates symbol
|
|
* names either with addresses, or a marker that the symbol hasn't been
|
|
* resolved yet, or possibly that the symbol has been defined as
|
|
* contained in a dynamic [load time/run time] linked library.
|
|
*
|
|
* segment = -1 => not yet defined
|
|
* segment = -2 => defined as dll symbol
|
|
*
|
|
* If the symbol is already defined, and the new segment >= 0, then
|
|
* if the original segment was < 0 the symbol is redefined, otherwise
|
|
* a duplicate symbol warning is issued. If new segment == -1, this
|
|
* routine won't change a previously existing symbol. It will change
|
|
* to segment = -2 only if the segment was previously < 0.
|
|
*/
|
|
void symtab_add(const char *symbol, int segment, int32_t offset)
|
|
{
|
|
symtabEnt *ste;
|
|
|
|
ste = symtabFind(symtab, symbol);
|
|
if (ste) {
|
|
if (ste->segment >= 0) {
|
|
/*
|
|
* symbol previously defined
|
|
*/
|
|
if (segment < 0)
|
|
return;
|
|
fprintf(error_file, "warning: `%s' redefined\n", symbol);
|
|
return;
|
|
}
|
|
|
|
/*
|
|
* somebody wanted the symbol, and put an undefined symbol
|
|
* marker into the table
|
|
*/
|
|
if (segment == -1)
|
|
return;
|
|
/*
|
|
* we have more information now - update the symbol's entry
|
|
*/
|
|
ste->segment = segment;
|
|
ste->offset = offset;
|
|
ste->flags = 0;
|
|
return;
|
|
}
|
|
/*
|
|
* this is the first declaration of this symbol
|
|
*/
|
|
ste = nasm_malloc(sizeof(symtabEnt));
|
|
if (!ste) {
|
|
fprintf(stderr, "ldrdf: out of memory\n");
|
|
exit(1);
|
|
}
|
|
ste->name = nasm_strdup(symbol);
|
|
ste->segment = segment;
|
|
ste->offset = offset;
|
|
ste->flags = 0;
|
|
symtabInsert(symtab, ste);
|
|
}
|
|
|
|
/*
|
|
* symtab_get()
|
|
*
|
|
* Retrieves the values associated with a symbol. Undefined symbols
|
|
* are assumed to have -1:0 associated. Returns 1 if the symbol was
|
|
* successfully located.
|
|
*/
|
|
int symtab_get(const char *symbol, int *segment, int32_t *offset)
|
|
{
|
|
symtabEnt *ste = symtabFind(symtab, symbol);
|
|
if (!ste) {
|
|
*segment = -1;
|
|
*offset = 0;
|
|
return 0;
|
|
} else {
|
|
*segment = ste->segment;
|
|
*offset = ste->offset;
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* add_library()
|
|
*
|
|
* checks that a library can be opened and is in the correct format,
|
|
* then adds it to the linked list of libraries.
|
|
*/
|
|
static void add_library(const char *name)
|
|
{
|
|
if (rdl_verify(name)) {
|
|
rdl_perror("ldrdf", name);
|
|
errorcount++;
|
|
return;
|
|
}
|
|
if (!libraries) {
|
|
lastlib = libraries = nasm_malloc(sizeof(*libraries));
|
|
if (!libraries) {
|
|
fprintf(stderr, "ldrdf: out of memory\n");
|
|
exit(1);
|
|
}
|
|
} else {
|
|
lastlib->next = nasm_malloc(sizeof(*libraries));
|
|
if (!lastlib->next) {
|
|
fprintf(stderr, "ldrdf: out of memory\n");
|
|
exit(1);
|
|
}
|
|
lastlib = lastlib->next;
|
|
}
|
|
lastlib->next = NULL;
|
|
if (rdl_open(lastlib, name)) {
|
|
rdl_perror("ldrdf", name);
|
|
errorcount++;
|
|
return;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* search_libraries()
|
|
*
|
|
* scans through the list of libraries, attempting to match symbols
|
|
* defined in library modules against symbols that are referenced but
|
|
* not defined (segment = -1 in the symbol table)
|
|
*
|
|
* returns 1 if any extra library modules are included, indicating that
|
|
* another pass through the library list should be made (possibly).
|
|
*/
|
|
static int search_libraries(void)
|
|
{
|
|
struct librarynode *cur;
|
|
rdffile f;
|
|
int i;
|
|
void *header;
|
|
int segment;
|
|
int32_t offset;
|
|
int doneanything = 0, pass = 1, keepfile;
|
|
rdfheaderrec *hr;
|
|
|
|
cur = libraries;
|
|
|
|
while (cur) {
|
|
if (options.verbose > 2)
|
|
printf("scanning library `%s', pass %d...\n", cur->name, pass);
|
|
|
|
for (i = 0; rdl_openmodule(cur, i, &f) == 0; i++) {
|
|
if (pass == 2 && lookformodule(f.name))
|
|
continue;
|
|
|
|
if (options.verbose > 3)
|
|
printf(" looking in module `%s'\n", f.name);
|
|
|
|
header = nasm_malloc(f.header_len);
|
|
if (!header) {
|
|
fprintf(stderr, "ldrdf: not enough memory\n");
|
|
exit(1);
|
|
}
|
|
if (rdfloadseg(&f, RDOFF_HEADER, header)) {
|
|
rdfperror("ldrdf", f.name);
|
|
errorcount++;
|
|
return 0;
|
|
}
|
|
|
|
keepfile = 0;
|
|
|
|
while ((hr = rdfgetheaderrec(&f))) {
|
|
/* We're only interested in exports, so skip others */
|
|
if (hr->type != RDFREC_GLOBAL)
|
|
continue;
|
|
|
|
/*
|
|
* If the symbol is marked as SYM_GLOBAL, somebody will be
|
|
* definitely interested in it..
|
|
*/
|
|
if ((hr->e.flags & SYM_GLOBAL) == 0) {
|
|
/*
|
|
* otherwise the symbol is just public. Find it in
|
|
* the symbol table. If the symbol isn't defined, we
|
|
* aren't interested, so go on to the next.
|
|
* If it is defined as anything but -1, we're also not
|
|
* interested. But if it is defined as -1, insert this
|
|
* module into the list of modules to use, and go
|
|
* immediately on to the next module...
|
|
*/
|
|
if (!symtab_get(hr->e.label, &segment, &offset)
|
|
|| segment != -1)
|
|
continue;
|
|
}
|
|
|
|
doneanything = 1;
|
|
keepfile = 1;
|
|
|
|
/*
|
|
* as there are undefined symbols, we can assume that
|
|
* there are modules on the module list by the time
|
|
* we get here.
|
|
*/
|
|
lastmodule->next = nasm_malloc(sizeof(*lastmodule->next));
|
|
if (!lastmodule->next) {
|
|
fprintf(stderr, "ldrdf: not enough memory\n");
|
|
exit(1);
|
|
}
|
|
lastmodule = lastmodule->next;
|
|
memcpy(&lastmodule->f, &f, sizeof(f));
|
|
lastmodule->name = nasm_strdup(f.name);
|
|
lastmodule->next = NULL;
|
|
processmodule(f.name, lastmodule);
|
|
break;
|
|
}
|
|
if (!keepfile) {
|
|
nasm_free(f.name);
|
|
f.name = NULL;
|
|
f.fp = NULL;
|
|
}
|
|
}
|
|
if (rdl_error != 0 && rdl_error != RDL_ENOTFOUND)
|
|
rdl_perror("ldrdf", cur->name);
|
|
|
|
cur = cur->next;
|
|
if (cur == NULL && pass == 1) {
|
|
cur = libraries;
|
|
pass++;
|
|
}
|
|
}
|
|
|
|
return doneanything;
|
|
}
|
|
|
|
/*
|
|
* write_output()
|
|
*
|
|
* this takes the linked list of modules, and walks through it, merging
|
|
* all the modules into a single output module, and then writes this to a
|
|
* file.
|
|
*/
|
|
static void write_output(const char *filename)
|
|
{
|
|
FILE *f;
|
|
rdf_headerbuf *rdfheader;
|
|
struct modulenode *cur;
|
|
int i, n, availableseg, seg, localseg, isrelative;
|
|
void *header;
|
|
rdfheaderrec *hr, newrec;
|
|
symtabEnt *se;
|
|
segtab segs;
|
|
int32_t offset;
|
|
uint8_t *data;
|
|
|
|
if ((f = fopen(filename, "wb")) == NULL) {
|
|
fprintf(stderr, "ldrdf: couldn't open %s for output\n", filename);
|
|
exit(1);
|
|
}
|
|
if ((rdfheader = rdfnewheader()) == NULL) {
|
|
fprintf(stderr, "ldrdf: out of memory\n");
|
|
exit(1);
|
|
}
|
|
|
|
/*
|
|
* If '-g' option was given, first record in output file will be a
|
|
* `generic' record, filled with a given file content.
|
|
* This can be useful, for example, when constructing multiboot
|
|
* compliant kernels.
|
|
*/
|
|
if (generic_rec_file) {
|
|
FILE *ff;
|
|
|
|
if (options.verbose)
|
|
printf("\nadding generic record from binary file %s\n",
|
|
generic_rec_file);
|
|
|
|
hr = (rdfheaderrec *) nasm_malloc(sizeof(struct GenericRec));
|
|
if ((ff = fopen(generic_rec_file, "r")) == NULL) {
|
|
fprintf(stderr, "ldrdf: couldn't open %s for input\n",
|
|
generic_rec_file);
|
|
exit(1);
|
|
}
|
|
n = fread(hr->g.data, 1, sizeof(hr->g.data), ff);
|
|
fseek(ff, 0, SEEK_END);
|
|
if (ftell(ff) > (long)sizeof(hr->g.data)) {
|
|
fprintf(error_file,
|
|
"warning: maximum generic record size is %u, "
|
|
"rest of file ignored\n",
|
|
(unsigned int)sizeof(hr->g.data));
|
|
}
|
|
fclose(ff);
|
|
|
|
hr->g.type = RDFREC_GENERIC;
|
|
hr->g.reclen = n;
|
|
rdfaddheader(rdfheader, hr);
|
|
nasm_free(hr);
|
|
}
|
|
|
|
/*
|
|
* Add module name record if `-mn' option was given
|
|
*/
|
|
if (modname_specified) {
|
|
n = strlen(modname_specified);
|
|
|
|
if ((n < 1) || (n >= MODLIB_NAME_MAX)) {
|
|
fprintf(stderr, "ldrdf: invalid length of module name `%s'\n",
|
|
modname_specified);
|
|
exit(1);
|
|
}
|
|
|
|
if (options.verbose)
|
|
printf("\nadding module name record %s\n", modname_specified);
|
|
|
|
hr = (rdfheaderrec *) nasm_malloc(sizeof(struct ModRec));
|
|
hr->m.type = RDFREC_MODNAME;
|
|
hr->m.reclen = n + 1;
|
|
strcpy(hr->m.modname, modname_specified);
|
|
rdfaddheader(rdfheader, hr);
|
|
nasm_free(hr);
|
|
}
|
|
|
|
|
|
if (options.verbose)
|
|
printf("\nbuilding output module (%d segments)\n", nsegs);
|
|
|
|
/*
|
|
* Allocate the memory for the segments. We may be better off
|
|
* building the output module one segment at a time when running
|
|
* under 16 bit DOS, but that would be a slower way of doing this.
|
|
* And you could always use DJGPP...
|
|
*/
|
|
for (i = 0; i < nsegs; i++) {
|
|
outputseg[i].data = NULL;
|
|
if (!outputseg[i].length)
|
|
continue;
|
|
outputseg[i].data = nasm_malloc(outputseg[i].length);
|
|
if (!outputseg[i].data) {
|
|
fprintf(stderr, "ldrdf: out of memory\n");
|
|
exit(1);
|
|
}
|
|
}
|
|
|
|
/*
|
|
* initialise availableseg, used to allocate segment numbers for
|
|
* imported and exported labels...
|
|
*/
|
|
availableseg = nsegs;
|
|
|
|
/*
|
|
* Step through the modules, performing required actions on each one
|
|
*/
|
|
for (cur = modules; cur; cur = cur->next) {
|
|
/*
|
|
* Read the actual segment contents into the correct places in
|
|
* the newly allocated segments
|
|
*/
|
|
|
|
for (i = 0; i < cur->f.nsegs; i++) {
|
|
int dest = cur->seginfo[i].dest_seg;
|
|
|
|
if (dest == -1)
|
|
continue;
|
|
if (rdfloadseg(&cur->f, i,
|
|
outputseg[dest].data + cur->seginfo[i].reloc)) {
|
|
rdfperror("ldrdf", cur->name);
|
|
exit(1);
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Perform fixups, and add new header records where required
|
|
*/
|
|
|
|
header = nasm_malloc(cur->f.header_len);
|
|
if (!header) {
|
|
fprintf(stderr, "ldrdf: out of memory\n");
|
|
exit(1);
|
|
}
|
|
|
|
if (cur->f.header_loc)
|
|
rdfheaderrewind(&cur->f);
|
|
else if (rdfloadseg(&cur->f, RDOFF_HEADER, header)) {
|
|
rdfperror("ldrdf", cur->name);
|
|
exit(1);
|
|
}
|
|
|
|
/*
|
|
* we need to create a local segment number -> location
|
|
* table for the segments in this module.
|
|
*/
|
|
init_seglocations(&segs);
|
|
for (i = 0; i < cur->f.nsegs; i++) {
|
|
add_seglocation(&segs, cur->f.seg[i].number,
|
|
cur->seginfo[i].dest_seg,
|
|
cur->seginfo[i].reloc);
|
|
}
|
|
/*
|
|
* and the BSS segment (doh!)
|
|
*/
|
|
add_seglocation(&segs, 2, 2, cur->bss_reloc);
|
|
|
|
while ((hr = rdfgetheaderrec(&cur->f))) {
|
|
switch (hr->type) {
|
|
case RDFREC_RELOC: /* relocation record - need to do a fixup */
|
|
/*
|
|
* First correct the offset stored in the segment from
|
|
* the start of the segment (which may well have changed).
|
|
*
|
|
* To do this we add to the number stored the relocation
|
|
* factor associated with the segment that contains the
|
|
* target segment.
|
|
*
|
|
* The relocation could be a relative relocation, in which
|
|
* case we have to first subtract the amount we've relocated
|
|
* the containing segment by.
|
|
*/
|
|
if (!get_seglocation(&segs, hr->r.refseg, &seg, &offset)) {
|
|
fprintf(stderr,
|
|
"%s: reloc to undefined segment %04x\n",
|
|
cur->name, (int)hr->r.refseg);
|
|
errorcount++;
|
|
break;
|
|
}
|
|
|
|
isrelative =
|
|
(hr->r.segment & RDOFF_RELATIVEMASK) ==
|
|
RDOFF_RELATIVEMASK;
|
|
hr->r.segment &= (RDOFF_RELATIVEMASK - 1);
|
|
|
|
if (hr->r.segment == 2 ||
|
|
(localseg =
|
|
rdffindsegment(&cur->f, hr->r.segment)) == -1) {
|
|
fprintf(stderr, "%s: reloc from %s segment (%d)\n",
|
|
cur->name,
|
|
hr->r.segment == 2 ? "BSS" : "unknown",
|
|
hr->r.segment);
|
|
errorcount++;
|
|
break;
|
|
}
|
|
|
|
if (hr->r.length != 1 && hr->r.length != 2 &&
|
|
hr->r.length != 4) {
|
|
fprintf(stderr, "%s: nonstandard length reloc "
|
|
"(%d bytes)\n", cur->name, hr->r.length);
|
|
errorcount++;
|
|
break;
|
|
}
|
|
|
|
/*
|
|
* okay, now the relocation is in the segment pointed to by
|
|
* cur->seginfo[localseg], and we know everything else is
|
|
* okay to go ahead and do the relocation
|
|
*/
|
|
data = outputseg[cur->seginfo[localseg].dest_seg].data;
|
|
data += cur->seginfo[localseg].reloc + hr->r.offset;
|
|
|
|
/*
|
|
* data now points to the reference that needs
|
|
* relocation. Calculate the relocation factor.
|
|
* Factor is:
|
|
* offset of referred object in segment [in offset]
|
|
* (- relocation of localseg, if ref is relative)
|
|
* For simplicity, the result is stored in 'offset'.
|
|
* Then add 'offset' onto the value at data.
|
|
*/
|
|
|
|
if (isrelative)
|
|
offset -= cur->seginfo[localseg].reloc;
|
|
switch (hr->r.length) {
|
|
case 1:
|
|
offset += *data;
|
|
if (offset < -127 || offset > 128)
|
|
fprintf(error_file,
|
|
"warning: relocation out of range "
|
|
"at %s(%02x:%08"PRIx32")\n", cur->name,
|
|
(int)hr->r.segment, hr->r.offset);
|
|
*data = (char)offset;
|
|
break;
|
|
case 2:
|
|
offset += *(int16_t *)data;
|
|
if (offset < -32767 || offset > 32768)
|
|
fprintf(error_file,
|
|
"warning: relocation out of range "
|
|
"at %s(%02x:%08"PRIx32")\n", cur->name,
|
|
(int)hr->r.segment, hr->r.offset);
|
|
*(int16_t *)data = (int16_t)offset;
|
|
break;
|
|
case 4:
|
|
*(int32_t *)data += offset;
|
|
/* we can't easily detect overflow on this one */
|
|
break;
|
|
}
|
|
|
|
/*
|
|
* If the relocation was relative between two symbols in
|
|
* the same segment, then we're done.
|
|
*
|
|
* Otherwise, we need to output a new relocation record
|
|
* with the references updated segment and offset...
|
|
*/
|
|
if (!isrelative || cur->seginfo[localseg].dest_seg != seg) {
|
|
hr->r.segment = cur->seginfo[localseg].dest_seg;
|
|
hr->r.offset += cur->seginfo[localseg].reloc;
|
|
hr->r.refseg = seg;
|
|
if (isrelative)
|
|
hr->r.segment += RDOFF_RELATIVEMASK;
|
|
rdfaddheader(rdfheader, hr);
|
|
}
|
|
break;
|
|
|
|
case RDFREC_IMPORT: /* import symbol */
|
|
case RDFREC_FARIMPORT:
|
|
/*
|
|
* scan the global symbol table for the symbol
|
|
* and associate its location with the segment number
|
|
* for this module
|
|
*/
|
|
se = symtabFind(symtab, hr->i.label);
|
|
if (!se || se->segment == -1) {
|
|
if (!options.dynalink && !(hr->i.flags & SYM_IMPORT)) {
|
|
fprintf(error_file,
|
|
"error: unresolved reference to `%s'"
|
|
" in module `%s'\n", hr->i.label,
|
|
cur->name);
|
|
errorcount++;
|
|
}
|
|
/*
|
|
* we need to allocate a segment number for this
|
|
* symbol, and store it in the symbol table for
|
|
* future reference
|
|
*/
|
|
if (!se) {
|
|
se = nasm_malloc(sizeof(*se));
|
|
if (!se) {
|
|
fprintf(stderr, "ldrdf: out of memory\n");
|
|
exit(1);
|
|
}
|
|
se->name = nasm_strdup(hr->i.label);
|
|
se->flags = 0;
|
|
se->segment = availableseg++;
|
|
se->offset = 0;
|
|
symtabInsert(symtab, se);
|
|
} else {
|
|
se->segment = availableseg++;
|
|
se->offset = 0;
|
|
}
|
|
/*
|
|
* output a header record that imports it to the
|
|
* recently allocated segment number...
|
|
*/
|
|
newrec = *hr;
|
|
newrec.i.segment = se->segment;
|
|
rdfaddheader(rdfheader, &newrec);
|
|
}
|
|
|
|
add_seglocation(&segs, hr->i.segment, se->segment,
|
|
se->offset);
|
|
break;
|
|
|
|
case RDFREC_GLOBAL: /* export symbol */
|
|
/*
|
|
* need to insert an export for this symbol into the new
|
|
* header, unless we're stripping symbols. Even if we're
|
|
* stripping, put the symbol if it's marked as SYM_GLOBAL.
|
|
*/
|
|
if (options.strip && !(hr->e.flags & SYM_GLOBAL))
|
|
break;
|
|
|
|
if (hr->e.segment == 2) {
|
|
seg = 2;
|
|
offset = cur->bss_reloc;
|
|
} else {
|
|
localseg = rdffindsegment(&cur->f, hr->e.segment);
|
|
if (localseg == -1) {
|
|
fprintf(stderr, "%s: exported symbol `%s' from "
|
|
"unrecognised segment\n", cur->name,
|
|
hr->e.label);
|
|
errorcount++;
|
|
break;
|
|
}
|
|
offset = cur->seginfo[localseg].reloc;
|
|
seg = cur->seginfo[localseg].dest_seg;
|
|
}
|
|
|
|
hr->e.segment = seg;
|
|
hr->e.offset += offset;
|
|
rdfaddheader(rdfheader, hr);
|
|
break;
|
|
|
|
case RDFREC_MODNAME: /* module name */
|
|
/*
|
|
* Insert module name record if export symbols
|
|
* are not stripped.
|
|
* If module name begins with '$' - insert it anyway.
|
|
*/
|
|
if (options.strip && hr->m.modname[0] != '$')
|
|
break;
|
|
rdfaddheader(rdfheader, hr);
|
|
break;
|
|
|
|
case RDFREC_DLL: /* DLL name */
|
|
/*
|
|
* Insert DLL name if it begins with '$'
|
|
*/
|
|
if (hr->d.libname[0] != '$')
|
|
break;
|
|
rdfaddheader(rdfheader, hr);
|
|
break;
|
|
|
|
case RDFREC_SEGRELOC: /* segment fixup */
|
|
/*
|
|
* modify the segment numbers if necessary, and
|
|
* pass straight through to the output module header
|
|
*
|
|
* *** FIXME ***
|
|
*/
|
|
if (hr->r.segment == 2) {
|
|
fprintf(stderr, "%s: segment fixup in BSS section\n",
|
|
cur->name);
|
|
errorcount++;
|
|
break;
|
|
}
|
|
localseg = rdffindsegment(&cur->f, hr->r.segment);
|
|
if (localseg == -1) {
|
|
fprintf(stderr, "%s: segment fixup in unrecognised"
|
|
" segment (%d)\n", cur->name, hr->r.segment);
|
|
errorcount++;
|
|
break;
|
|
}
|
|
hr->r.segment = cur->seginfo[localseg].dest_seg;
|
|
hr->r.offset += cur->seginfo[localseg].reloc;
|
|
|
|
if (!get_seglocation(&segs, hr->r.refseg, &seg, &offset)) {
|
|
fprintf(stderr, "%s: segment fixup to undefined "
|
|
"segment %04x\n", cur->name,
|
|
(int)hr->r.refseg);
|
|
errorcount++;
|
|
break;
|
|
}
|
|
hr->r.refseg = seg;
|
|
rdfaddheader(rdfheader, hr);
|
|
break;
|
|
|
|
case RDFREC_COMMON: /* Common variable */
|
|
/* Is this symbol already in the table? */
|
|
se = symtabFind(symtab, hr->c.label);
|
|
if (!se) {
|
|
printf("%s is not in symtab yet\n", hr->c.label);
|
|
break;
|
|
}
|
|
/* Add segment location */
|
|
add_seglocation(&segs, hr->c.segment, se->segment,
|
|
se->offset);
|
|
break;
|
|
}
|
|
}
|
|
|
|
nasm_free(header);
|
|
done_seglocations(&segs);
|
|
|
|
}
|
|
|
|
/*
|
|
* combined BSS reservation for the entire results
|
|
*/
|
|
newrec.type = RDFREC_BSS;
|
|
newrec.b.reclen = 4;
|
|
newrec.b.amount = bss_length;
|
|
rdfaddheader(rdfheader, &newrec);
|
|
|
|
/*
|
|
* Write the header
|
|
*/
|
|
for (i = 0; i < nsegs; i++) {
|
|
if (i == 2)
|
|
continue;
|
|
rdfaddsegment(rdfheader, outputseg[i].length);
|
|
}
|
|
|
|
rdfwriteheader(f, rdfheader);
|
|
rdfdoneheader(rdfheader);
|
|
|
|
/*
|
|
* Step through the segments, one at a time, writing out into
|
|
* the output file
|
|
*/
|
|
for (i = 0; i < nsegs; i++) {
|
|
if (i == 2)
|
|
continue;
|
|
|
|
fwriteint16_t(outputseg[i].type, f);
|
|
fwriteint16_t(outputseg[i].number, f);
|
|
fwriteint16_t(outputseg[i].reserved, f);
|
|
fwriteint32_t(outputseg[i].length, f);
|
|
nasm_write(outputseg[i].data, outputseg[i].length, f);
|
|
}
|
|
|
|
fwritezero(10, f);
|
|
}
|
|
|
|
/* =========================================================================
|
|
* Main program
|
|
*/
|
|
|
|
static void usage(void)
|
|
{
|
|
printf("usage:\n"
|
|
" ldrdf [options] object modules ... [-llibrary ...]\n"
|
|
" ldrdf -r\n"
|
|
"options:\n"
|
|
" -v[=n] increase verbosity by 1, or set it to n\n"
|
|
" -a nn set segment alignment value (default 16)\n"
|
|
" -s strip public symbols\n"
|
|
" -dy Unix-style dynamic linking\n"
|
|
" -o name write output in file 'name'\n"
|
|
" -j path specify objects search path\n"
|
|
" -L path specify libraries search path\n"
|
|
" -g file embed 'file' as a first header record with type 'generic'\n"
|
|
" -mn name add module name record at the beginning of output file\n");
|
|
exit(0);
|
|
}
|
|
|
|
int main(int argc, char **argv)
|
|
{
|
|
char *outname = "aout.rdf";
|
|
int moduleloaded = 0;
|
|
char *respstrings[128] = { 0, };
|
|
|
|
rdoff_init();
|
|
|
|
options.verbose = 0;
|
|
options.align = 16;
|
|
options.dynalink = 0;
|
|
options.strip = 0;
|
|
|
|
error_file = stderr;
|
|
|
|
argc--, argv++;
|
|
if (argc == 0)
|
|
usage();
|
|
while (argc && *argv && **argv == '-' && argv[0][1] != 'l') {
|
|
switch (argv[0][1]) {
|
|
case 'r':
|
|
printf("ldrdf (linker for RDF files) version " LDRDF_VERSION
|
|
"\n");
|
|
printf("RDOFF2 revision %s\n", RDOFF2_REVISION);
|
|
exit(0);
|
|
case 'v':
|
|
if (argv[0][2] == '=') {
|
|
options.verbose = argv[0][3] - '0';
|
|
if (options.verbose < 0 || options.verbose > 9) {
|
|
fprintf(stderr,
|
|
"ldrdf: verbosity level must be a number"
|
|
" between 0 and 9\n");
|
|
exit(1);
|
|
}
|
|
} else
|
|
options.verbose++;
|
|
break;
|
|
case 'a':
|
|
options.align = atoi(argv[1]);
|
|
if (options.align <= 0) {
|
|
fprintf(stderr,
|
|
"ldrdf: -a expects a positive number argument\n");
|
|
exit(1);
|
|
}
|
|
argv++, argc--;
|
|
break;
|
|
case 's':
|
|
options.strip = 1;
|
|
break;
|
|
case 'd':
|
|
if (argv[0][2] == 'y')
|
|
options.dynalink = 1;
|
|
break;
|
|
case 'm':
|
|
if (argv[0][2] == 'n') {
|
|
modname_specified = argv[1];
|
|
argv++, argc--;
|
|
if (!argc) {
|
|
fprintf(stderr, "ldrdf: -mn expects a module name\n");
|
|
exit(1);
|
|
}
|
|
}
|
|
break;
|
|
case 'o':
|
|
outname = argv[1];
|
|
argv++, argc--;
|
|
break;
|
|
case 'j':
|
|
if (!objpath) {
|
|
options.objpath = 1;
|
|
objpath = argv[1];
|
|
argv++, argc--;
|
|
break;
|
|
} else {
|
|
fprintf(stderr,
|
|
"ldrdf: more than one objects search path specified\n");
|
|
exit(1);
|
|
}
|
|
case 'L':
|
|
if (!libpath) {
|
|
options.libpath = 1;
|
|
libpath = argv[1];
|
|
argv++, argc--;
|
|
break;
|
|
} else {
|
|
fprintf(stderr,
|
|
"ldrdf: more than one libraries search path specified\n");
|
|
exit(1);
|
|
}
|
|
case '@':{
|
|
int i = 0;
|
|
char buf[256];
|
|
FILE *f;
|
|
|
|
options.respfile = 1;
|
|
if (argv[1] != NULL)
|
|
f = fopen(argv[1], "r");
|
|
else {
|
|
fprintf(stderr,
|
|
"ldrdf: no response file name specified\n");
|
|
exit(1);
|
|
}
|
|
|
|
if (f == NULL) {
|
|
fprintf(stderr,
|
|
"ldrdf: unable to open response file\n");
|
|
exit(1);
|
|
}
|
|
|
|
argv++, argc--;
|
|
while (fgets(buf, sizeof(buf), f) != NULL) {
|
|
char *p;
|
|
if (buf[0] == '\n')
|
|
continue;
|
|
if ((p = strchr(buf, '\n')) != NULL)
|
|
*p = '\0';
|
|
if (i >= 128) {
|
|
fclose(f);
|
|
fprintf(stderr, "ldrdf: too many input files\n");
|
|
exit(1);
|
|
}
|
|
*(respstrings + i) = nasm_strdup(buf);
|
|
argc++, i++;
|
|
}
|
|
fclose(f);
|
|
break;
|
|
}
|
|
case '2':
|
|
options.stderr_redir = 1;
|
|
error_file = stdout;
|
|
break;
|
|
case 'g':
|
|
generic_rec_file = argv[1];
|
|
argv++, argc--;
|
|
if (!argc) {
|
|
fprintf(stderr, "ldrdf: -g expects a file name\n");
|
|
exit(1);
|
|
}
|
|
break;
|
|
default:
|
|
usage();
|
|
}
|
|
argv++, argc--;
|
|
}
|
|
|
|
if (options.verbose > 4) {
|
|
printf("ldrdf invoked with options:\n");
|
|
printf(" section alignment: %d bytes\n", options.align);
|
|
printf(" output name: `%s'\n", outname);
|
|
if (options.strip)
|
|
printf(" strip symbols\n");
|
|
if (options.dynalink)
|
|
printf(" Unix-style dynamic linking\n");
|
|
if (options.objpath)
|
|
printf(" objects search path: %s\n", objpath);
|
|
if (options.libpath)
|
|
printf(" libraries search path: %s\n", libpath);
|
|
printf("\n");
|
|
}
|
|
|
|
symtab = symtabNew();
|
|
initsegments();
|
|
|
|
if (!symtab) {
|
|
fprintf(stderr, "ldrdf: out of memory\n");
|
|
exit(1);
|
|
}
|
|
|
|
while (argc) {
|
|
if (!*argv)
|
|
argv = respstrings;
|
|
if (!*argv)
|
|
break;
|
|
if (!strncmp(*argv, "-l", 2)) {
|
|
if (libpath && (argv[0][2] != '/'))
|
|
add_library(nasm_strcat(libpath, *argv + 2));
|
|
else
|
|
add_library(*argv + 2);
|
|
} else {
|
|
if (objpath && (argv[0][0] != '/'))
|
|
loadmodule(nasm_strcat(objpath, *argv));
|
|
else
|
|
loadmodule(*argv);
|
|
moduleloaded = 1;
|
|
}
|
|
argv++, argc--;
|
|
}
|
|
|
|
if (!moduleloaded) {
|
|
printf("ldrdf: nothing to do. ldrdf -h for usage\n");
|
|
return 0;
|
|
}
|
|
|
|
search_libraries();
|
|
|
|
if (options.verbose > 2) {
|
|
printf("symbol table:\n");
|
|
symtabDump(symtab, stdout);
|
|
}
|
|
|
|
write_output(outname);
|
|
|
|
if (errorcount > 0) {
|
|
remove(outname);
|
|
exit(1);
|
|
}
|
|
return 0;
|
|
}
|