* Written by Erik Andersen <andersen@lineo.com>
* and Ron Alder <alder@lineo.com>
*
+ * Modified by Bryan Rittmeyer <bryan@ixiacom.com> to support SH4
+ * and (theoretically) SH3. Note that there is still no true
+ * multiple architecture support. You just get SH3|SH4|i386, despite
+ * the mention of ARM and m68k--which may or may not work (but
+ * almost certainly do not, due to at least MATCH_MACHINE). I have
+ * only tested SH4 in little endian mode.
+ *
* Based almost entirely on the Linux modutils-2.3.11 implementation.
* Copyright 1996, 1997 Linux International.
* New implementation contributed by Richard Henderson <rth@tamu.edu>
#ifndef MODUTILS_MODULE_H
#define MODUTILS_MODULE_H 1
-#ident "$Id: insmod.c,v 1.23 2000/09/22 00:38:07 andersen Exp $"
+#ident "$Id: insmod.c,v 1.24 2000/09/24 03:44:29 andersen Exp $"
/* This file contains the structures used by the 2.0 and 2.1 kernels.
We do not use the kernel headers directly because we do not wish
#ifndef MODUTILS_OBJ_H
#define MODUTILS_OBJ_H 1
-#ident "$Id: insmod.c,v 1.23 2000/09/22 00:38:07 andersen Exp $"
+#ident "$Id: insmod.c,v 1.24 2000/09/24 03:44:29 andersen Exp $"
/* The relocatable object is manipulated using elfin types. */
/* Machine-specific elf macros for i386 et al. */
+/* the SH changes have only been tested on the SH4 in =little endian= mode */
+/* I'm not sure about big endian, so let's warn: */
+
+#if (defined(__SH4__) || defined(__SH3__)) && defined(__BIG_ENDIAN__)
+#error insmod.c may require changes for use on big endian SH4/SH3
+#endif
+
+/* it may or may not work on the SH1/SH2... So let's error on those
+ also */
+#if (defined(__sh__) && (!(defined(__SH3__) || defined(__SH4__))))
+#error insmod.c may require changes for non-SH3/SH4 use
+#endif
+
#define ELFCLASSM ELFCLASS32
#define ELFDATAM ELFDATA2LSB
-#define MATCH_MACHINE(x) (x == EM_386 || x == EM_486)
+
+#if defined(__sh__)
+
+#define MATCH_MACHINE(x) (x == EM_SH)
+#define SHT_RELM SHT_RELA
+#define Elf32_RelM Elf32_Rela
+
+#else
+
+/* presumably we can use these for anything but the SH */
+/* this is the previous behavior, but it does result in
+ insmod.c being broken on anything except i386 */
+
+#define MATCH_MACHINE(x) (x == EM_386 || x == EM_486)
#define SHT_RELM SHT_REL
#define Elf32_RelM Elf32_Rel
+#endif
#ifndef ElfW
# if ELFCLASSM == ELFCLASS32
/*======================================================================*/
-struct i386_got_entry {
+/* previously, these were named i386_* but since we could be
+ compiling for the sh, I've renamed them to the more general
+ arch_* These structures are the same between the x86 and SH,
+ and we can't support anything else right now anyway. In the
+ future maybe they should be #if defined'd */
+
+struct arch_got_entry {
int offset;
unsigned offset_done:1;
unsigned reloc_done:1;
};
-struct i386_file {
+struct arch_file {
struct obj_file root;
struct obj_section *got;
};
-struct i386_symbol {
+struct arch_symbol {
struct obj_symbol root;
- struct i386_got_entry gotent;
+ struct arch_got_entry gotent;
};
-
struct external_module {
const char *name;
ElfW(Addr) addr;
extern int delete_module(const char *);
#endif
+/* This is kind of troublesome. See, we don't actually support
+ the m68k or the arm the same way we support i386 and (now)
+ sh. In doing my SH patch, I just assumed that whatever works
+ for i386 also works for m68k and arm since currently insmod.c
+ does nothing special for them. If this isn't true, the below
+ line is rather misleading IMHO, and someone should either
+ change it or add more proper architecture-dependent support
+ for these boys.
+
+ -- Bryan Rittmeyer <bryan@ixiacom.com> */
+
#if defined(__i386__) || defined(__m68k__) || defined(__arm__)
/* Jump through hoops to fixup error return codes */
#define __NR__create_module __NR_create_module
struct obj_file *arch_new_file(void)
{
- struct i386_file *f;
+ struct arch_file *f;
f = xmalloc(sizeof(*f));
f->got = NULL;
return &f->root;
struct obj_symbol *arch_new_symbol(void)
{
- struct i386_symbol *sym;
+ struct arch_symbol *sym;
sym = xmalloc(sizeof(*sym));
memset(&sym->gotent, 0, sizeof(sym->gotent));
return &sym->root;
}
+
enum obj_reloc
arch_apply_relocation(struct obj_file *f,
struct obj_section *targsec,
struct obj_section *symsec,
struct obj_symbol *sym,
+#if defined(__sh__)
+ Elf32_Rela * rel, Elf32_Addr v)
+#else
Elf32_Rel * rel, Elf32_Addr v)
+#endif
{
- struct i386_file *ifile = (struct i386_file *) f;
- struct i386_symbol *isym = (struct i386_symbol *) sym;
+ struct arch_file *ifile = (struct arch_file *) f;
+ struct arch_symbol *isym = (struct arch_symbol *) sym;
Elf32_Addr *loc = (Elf32_Addr *) (targsec->contents + rel->r_offset);
Elf32_Addr dot = targsec->header.sh_addr + rel->r_offset;
enum obj_reloc ret = obj_reloc_ok;
switch (ELF32_R_TYPE(rel->r_info)) {
+
+/* even though these constants seem to be the same for
+ the i386 and the sh, we "#if define" them for clarity
+ and in case that ever changes */
+#if defined(__sh__)
+ case R_SH_NONE:
+#else
case R_386_NONE:
+#endif
break;
+#if defined(__sh__)
+ case R_SH_DIR32:
+#else
case R_386_32:
+#endif
*loc += v;
break;
+#if defined(__sh__)
+ case R_SH_REL32:
+#else
case R_386_PLT32:
case R_386_PC32:
+#endif
*loc += v - dot;
break;
+#if defined(__sh__)
+ case R_SH_PLT32:
+ *loc = v - dot;
+ break;
+#endif
+
+
+#if defined(__sh__)
+ case R_SH_GLOB_DAT:
+ case R_SH_JMP_SLOT:
+ *loc = v;
+ break;
+#else
case R_386_GLOB_DAT:
case R_386_JMP_SLOT:
*loc = v;
break;
+#endif
- case R_386_RELATIVE:
+#if defined(__sh__)
+ case R_SH_RELATIVE:
+ *loc += f->baseaddr + rel->r_addend;
+ break;
+#else
+ case R_386_RELATIVE:
*loc += f->baseaddr;
break;
+#endif
+#if defined(__sh__)
+ case R_SH_GOTPC:
+ assert(got != 0);
+ *loc += got - dot + rel->r_addend;;
+ break;
+#else
case R_386_GOTPC:
assert(got != 0);
*loc += got - dot;
break;
+#endif
+#if defined(__sh__)
+ case R_SH_GOT32:
+ assert(isym != NULL);
+ if (!isym->gotent.reloc_done) {
+ isym->gotent.reloc_done = 1;
+ *(Elf32_Addr *) (ifile->got->contents + isym->gotent.offset) =
+ v;
+ }
+ *loc += isym->gotent.offset + rel->r_addend;
+ break;
+#else
case R_386_GOT32:
assert(isym != NULL);
if (!isym->gotent.reloc_done) {
}
*loc += isym->gotent.offset;
break;
+#endif
+#if defined(__sh__)
+ case R_SH_GOTOFF:
+#else
case R_386_GOTOFF:
+#endif
assert(got != 0);
*loc += v - got;
break;
int arch_create_got(struct obj_file *f)
{
- struct i386_file *ifile = (struct i386_file *) f;
+ struct arch_file *ifile = (struct arch_file *) f;
int i, n, offset = 0, gotneeded = 0;
n = ifile->root.header.e_shnum;
for (i = 0; i < n; ++i) {
struct obj_section *relsec, *symsec, *strsec;
+#if defined(__sh__)
+ Elf32_Rela *rel, *relend;
+#else
Elf32_Rel *rel, *relend;
+#endif
Elf32_Sym *symtab;
const char *strtab;
symsec = ifile->root.sections[relsec->header.sh_link];
strsec = ifile->root.sections[symsec->header.sh_link];
+
+#if defined(__sh__)
+ rel = (Elf32_Rela *) relsec->contents;
+ relend = rel + (relsec->header.sh_size / sizeof(Elf32_Rela));
+#else
rel = (Elf32_Rel *) relsec->contents;
relend = rel + (relsec->header.sh_size / sizeof(Elf32_Rel));
+#endif
symtab = (Elf32_Sym *) symsec->contents;
strtab = (const char *) strsec->contents;
for (; rel < relend; ++rel) {
Elf32_Sym *extsym;
- struct i386_symbol *intsym;
+ struct arch_symbol *intsym;
const char *name;
switch (ELF32_R_TYPE(rel->r_info)) {
+#if defined(__sh__)
+ case R_SH_GOTPC:
+ case R_SH_GOTOFF:
+#else
case R_386_GOTPC:
case R_386_GOTOFF:
+#endif
gotneeded = 1;
default:
continue;
+#if defined(__sh__)
+ case R_SH_GOT32:
+#else
case R_386_GOT32:
+#endif
break;
}
else
name = f->sections[extsym->st_shndx]->name;
intsym =
- (struct i386_symbol *) obj_find_symbol(&ifile->root, name);
+ (struct arch_symbol *) obj_find_symbol(&ifile->root, name);
if (!intsym->gotent.offset_done) {
intsym->gotent.offset_done = 1;
* Written by Erik Andersen <andersen@lineo.com>
* and Ron Alder <alder@lineo.com>
*
+ * Modified by Bryan Rittmeyer <bryan@ixiacom.com> to support SH4
+ * and (theoretically) SH3. Note that there is still no true
+ * multiple architecture support. You just get SH3|SH4|i386, despite
+ * the mention of ARM and m68k--which may or may not work (but
+ * almost certainly do not, due to at least MATCH_MACHINE). I have
+ * only tested SH4 in little endian mode.
+ *
* Based almost entirely on the Linux modutils-2.3.11 implementation.
* Copyright 1996, 1997 Linux International.
* New implementation contributed by Richard Henderson <rth@tamu.edu>
#ifndef MODUTILS_MODULE_H
#define MODUTILS_MODULE_H 1
-#ident "$Id: insmod.c,v 1.23 2000/09/22 00:38:07 andersen Exp $"
+#ident "$Id: insmod.c,v 1.24 2000/09/24 03:44:29 andersen Exp $"
/* This file contains the structures used by the 2.0 and 2.1 kernels.
We do not use the kernel headers directly because we do not wish
#ifndef MODUTILS_OBJ_H
#define MODUTILS_OBJ_H 1
-#ident "$Id: insmod.c,v 1.23 2000/09/22 00:38:07 andersen Exp $"
+#ident "$Id: insmod.c,v 1.24 2000/09/24 03:44:29 andersen Exp $"
/* The relocatable object is manipulated using elfin types. */
/* Machine-specific elf macros for i386 et al. */
+/* the SH changes have only been tested on the SH4 in =little endian= mode */
+/* I'm not sure about big endian, so let's warn: */
+
+#if (defined(__SH4__) || defined(__SH3__)) && defined(__BIG_ENDIAN__)
+#error insmod.c may require changes for use on big endian SH4/SH3
+#endif
+
+/* it may or may not work on the SH1/SH2... So let's error on those
+ also */
+#if (defined(__sh__) && (!(defined(__SH3__) || defined(__SH4__))))
+#error insmod.c may require changes for non-SH3/SH4 use
+#endif
+
#define ELFCLASSM ELFCLASS32
#define ELFDATAM ELFDATA2LSB
-#define MATCH_MACHINE(x) (x == EM_386 || x == EM_486)
+
+#if defined(__sh__)
+
+#define MATCH_MACHINE(x) (x == EM_SH)
+#define SHT_RELM SHT_RELA
+#define Elf32_RelM Elf32_Rela
+
+#else
+
+/* presumably we can use these for anything but the SH */
+/* this is the previous behavior, but it does result in
+ insmod.c being broken on anything except i386 */
+
+#define MATCH_MACHINE(x) (x == EM_386 || x == EM_486)
#define SHT_RELM SHT_REL
#define Elf32_RelM Elf32_Rel
+#endif
#ifndef ElfW
# if ELFCLASSM == ELFCLASS32
/*======================================================================*/
-struct i386_got_entry {
+/* previously, these were named i386_* but since we could be
+ compiling for the sh, I've renamed them to the more general
+ arch_* These structures are the same between the x86 and SH,
+ and we can't support anything else right now anyway. In the
+ future maybe they should be #if defined'd */
+
+struct arch_got_entry {
int offset;
unsigned offset_done:1;
unsigned reloc_done:1;
};
-struct i386_file {
+struct arch_file {
struct obj_file root;
struct obj_section *got;
};
-struct i386_symbol {
+struct arch_symbol {
struct obj_symbol root;
- struct i386_got_entry gotent;
+ struct arch_got_entry gotent;
};
-
struct external_module {
const char *name;
ElfW(Addr) addr;
extern int delete_module(const char *);
#endif
+/* This is kind of troublesome. See, we don't actually support
+ the m68k or the arm the same way we support i386 and (now)
+ sh. In doing my SH patch, I just assumed that whatever works
+ for i386 also works for m68k and arm since currently insmod.c
+ does nothing special for them. If this isn't true, the below
+ line is rather misleading IMHO, and someone should either
+ change it or add more proper architecture-dependent support
+ for these boys.
+
+ -- Bryan Rittmeyer <bryan@ixiacom.com> */
+
#if defined(__i386__) || defined(__m68k__) || defined(__arm__)
/* Jump through hoops to fixup error return codes */
#define __NR__create_module __NR_create_module
struct obj_file *arch_new_file(void)
{
- struct i386_file *f;
+ struct arch_file *f;
f = xmalloc(sizeof(*f));
f->got = NULL;
return &f->root;
struct obj_symbol *arch_new_symbol(void)
{
- struct i386_symbol *sym;
+ struct arch_symbol *sym;
sym = xmalloc(sizeof(*sym));
memset(&sym->gotent, 0, sizeof(sym->gotent));
return &sym->root;
}
+
enum obj_reloc
arch_apply_relocation(struct obj_file *f,
struct obj_section *targsec,
struct obj_section *symsec,
struct obj_symbol *sym,
+#if defined(__sh__)
+ Elf32_Rela * rel, Elf32_Addr v)
+#else
Elf32_Rel * rel, Elf32_Addr v)
+#endif
{
- struct i386_file *ifile = (struct i386_file *) f;
- struct i386_symbol *isym = (struct i386_symbol *) sym;
+ struct arch_file *ifile = (struct arch_file *) f;
+ struct arch_symbol *isym = (struct arch_symbol *) sym;
Elf32_Addr *loc = (Elf32_Addr *) (targsec->contents + rel->r_offset);
Elf32_Addr dot = targsec->header.sh_addr + rel->r_offset;
enum obj_reloc ret = obj_reloc_ok;
switch (ELF32_R_TYPE(rel->r_info)) {
+
+/* even though these constants seem to be the same for
+ the i386 and the sh, we "#if define" them for clarity
+ and in case that ever changes */
+#if defined(__sh__)
+ case R_SH_NONE:
+#else
case R_386_NONE:
+#endif
break;
+#if defined(__sh__)
+ case R_SH_DIR32:
+#else
case R_386_32:
+#endif
*loc += v;
break;
+#if defined(__sh__)
+ case R_SH_REL32:
+#else
case R_386_PLT32:
case R_386_PC32:
+#endif
*loc += v - dot;
break;
+#if defined(__sh__)
+ case R_SH_PLT32:
+ *loc = v - dot;
+ break;
+#endif
+
+
+#if defined(__sh__)
+ case R_SH_GLOB_DAT:
+ case R_SH_JMP_SLOT:
+ *loc = v;
+ break;
+#else
case R_386_GLOB_DAT:
case R_386_JMP_SLOT:
*loc = v;
break;
+#endif
- case R_386_RELATIVE:
+#if defined(__sh__)
+ case R_SH_RELATIVE:
+ *loc += f->baseaddr + rel->r_addend;
+ break;
+#else
+ case R_386_RELATIVE:
*loc += f->baseaddr;
break;
+#endif
+#if defined(__sh__)
+ case R_SH_GOTPC:
+ assert(got != 0);
+ *loc += got - dot + rel->r_addend;;
+ break;
+#else
case R_386_GOTPC:
assert(got != 0);
*loc += got - dot;
break;
+#endif
+#if defined(__sh__)
+ case R_SH_GOT32:
+ assert(isym != NULL);
+ if (!isym->gotent.reloc_done) {
+ isym->gotent.reloc_done = 1;
+ *(Elf32_Addr *) (ifile->got->contents + isym->gotent.offset) =
+ v;
+ }
+ *loc += isym->gotent.offset + rel->r_addend;
+ break;
+#else
case R_386_GOT32:
assert(isym != NULL);
if (!isym->gotent.reloc_done) {
}
*loc += isym->gotent.offset;
break;
+#endif
+#if defined(__sh__)
+ case R_SH_GOTOFF:
+#else
case R_386_GOTOFF:
+#endif
assert(got != 0);
*loc += v - got;
break;
int arch_create_got(struct obj_file *f)
{
- struct i386_file *ifile = (struct i386_file *) f;
+ struct arch_file *ifile = (struct arch_file *) f;
int i, n, offset = 0, gotneeded = 0;
n = ifile->root.header.e_shnum;
for (i = 0; i < n; ++i) {
struct obj_section *relsec, *symsec, *strsec;
+#if defined(__sh__)
+ Elf32_Rela *rel, *relend;
+#else
Elf32_Rel *rel, *relend;
+#endif
Elf32_Sym *symtab;
const char *strtab;
symsec = ifile->root.sections[relsec->header.sh_link];
strsec = ifile->root.sections[symsec->header.sh_link];
+
+#if defined(__sh__)
+ rel = (Elf32_Rela *) relsec->contents;
+ relend = rel + (relsec->header.sh_size / sizeof(Elf32_Rela));
+#else
rel = (Elf32_Rel *) relsec->contents;
relend = rel + (relsec->header.sh_size / sizeof(Elf32_Rel));
+#endif
symtab = (Elf32_Sym *) symsec->contents;
strtab = (const char *) strsec->contents;
for (; rel < relend; ++rel) {
Elf32_Sym *extsym;
- struct i386_symbol *intsym;
+ struct arch_symbol *intsym;
const char *name;
switch (ELF32_R_TYPE(rel->r_info)) {
+#if defined(__sh__)
+ case R_SH_GOTPC:
+ case R_SH_GOTOFF:
+#else
case R_386_GOTPC:
case R_386_GOTOFF:
+#endif
gotneeded = 1;
default:
continue;
+#if defined(__sh__)
+ case R_SH_GOT32:
+#else
case R_386_GOT32:
+#endif
break;
}
else
name = f->sections[extsym->st_shndx]->name;
intsym =
- (struct i386_symbol *) obj_find_symbol(&ifile->root, name);
+ (struct arch_symbol *) obj_find_symbol(&ifile->root, name);
if (!intsym->gotent.offset_done) {
intsym->gotent.offset_done = 1;
projects / oweals / busybox.git / commitdiff