This commit contains a patch from Stefan Allius <allius@atecom.com> to change

how uClibc handles _init and _fini, allowing shared lib constructors and
destructors to initialize things in the correct sequence.  Stefan ported the SH
architecture.  I then ported x86, arm, and mips.  x86 and arm are working fine,
but I don't think I quite got things correct for mips.

11 files changed, 111 insertions, 75 deletions

diff --git a/ldso/ldso/arm/elfinterp.c b/ldso/ldso/arm/elfinterp.c
index 4f7e6bd3f..84b94550b 100644
--- a/ldso/ldso/arm/elfinterp.c
+++ b/ldso/ldso/arm/elfinterp.c
@@ -236,8 +236,7 @@ _dl_parse(struct elf_resolve *tpnt, struct dyn_elf *scope,
 		}
 		else if (res >0)
 		{
-			_dl_dprintf(2, "%s: can't resolve symbol '%s'\n",
-					_dl_progname, strtab + symtab[symtab_index].st_name);
+			_dl_dprintf(2, "can't resolve symbol\n");
 			goof += res;
 		}
 	  }
diff --git a/ldso/ldso/dl-elf.c b/ldso/ldso/dl-elf.c
index 4259a4292..199726ff3 100644
--- a/ldso/ldso/dl-elf.c
+++ b/ldso/ldso/dl-elf.c
@@ -318,16 +318,13 @@ struct elf_resolve *_dl_load_elf_shared_library(int secure,
 	Elf32_Dyn *dpnt;
 	struct elf_resolve *tpnt;
 	elf_phdr *ppnt;
-	int piclib;
 	char *status;
-	int flags;
 	char header[4096];
 	unsigned long dynamic_info[24];
 	unsigned long *lpnt;
 	unsigned long libaddr;
 	unsigned long minvma = 0xffffffff, maxvma = 0;
-	int i;
-	int infile;
+	int i, flags, piclib, infile;
 
 	/* If this file is already loaded, skip this step */
 	tpnt = _dl_check_hashed_files(libname);
@@ -544,28 +541,32 @@ struct elf_resolve *_dl_load_elf_shared_library(int secure,
 #if defined(__mips__)
 	{
 		
-		int i = 1;
+		int indx = 1;
 		Elf32_Dyn *dpnt = (Elf32_Dyn *) dynamic_addr;
 
 		while(dpnt->d_tag) {
 			dpnt++;
-			i++;
+			indx++;
 		}
-		dynamic_size = i;
+		dynamic_size = indx;
 	}
 #endif
 
-	for (i = 0; i < dynamic_size; i++) 
 	{
-		if (dpnt->d_tag > DT_JMPREL) {
+		unsigned long indx;
+
+		for (indx = 0; indx < dynamic_size; indx++) 
+		{
+			if (dpnt->d_tag > DT_JMPREL) {
+				dpnt++;
+				continue;
+			}
+			dynamic_info[dpnt->d_tag] = dpnt->d_un.d_val;
+			if (dpnt->d_tag == DT_TEXTREL || SVR4_BUGCOMPAT)
+				dynamic_info[DT_TEXTREL] = 1;
 			dpnt++;
-			continue;
-		}
-		dynamic_info[dpnt->d_tag] = dpnt->d_un.d_val;
-		if (dpnt->d_tag == DT_TEXTREL || SVR4_BUGCOMPAT)
-			dynamic_info[DT_TEXTREL] = 1;
-		dpnt++;
-	};
+		};
+	}
 
 	/* If the TEXTREL is set, this means that we need to make the pages
 	   writable before we perform relocations.  Do this now. They get set
diff --git a/ldso/ldso/i386/elfinterp.c b/ldso/ldso/i386/elfinterp.c
index 2caaa2413..2d08bf406 100644
--- a/ldso/ldso/i386/elfinterp.c
+++ b/ldso/ldso/i386/elfinterp.c
@@ -230,8 +230,7 @@ _dl_parse(struct elf_resolve *tpnt, struct dyn_elf *scope,
 		}
 		else if (res >0)
 		{
-			_dl_dprintf(2, "%s: can't resolve symbol '%s'\n",
-					_dl_progname, strtab + symtab[symtab_index].st_name);
+			_dl_dprintf(2, "can't resolve symbol\n");
 			goof += res;
 		}
 	  }
diff --git a/ldso/ldso/ldso.c b/ldso/ldso/ldso.c
index ff13675aa..b81c84092 100644
--- a/ldso/ldso/ldso.c
+++ b/ldso/ldso/ldso.c
@@ -303,12 +303,12 @@ LD_BOOT(unsigned long args)
   __asm__("\tmove %0, $28\n\tsubu %0,%0,0x7ff0\n\t":"=r"(got));
 #elif defined(__sh__)
   __asm__(
-"	   mov.l    1f, %0"
-"	   mova     1f, r0"
-"	   bra      2f"
-"	   add r0,  %0"
-"	   .balign  4"
-"1:	   .long    _GLOBAL_OFFSET_TABLE_"
+"       mov.l    1f, %0\n"
+"       mova     1f, r0\n"
+"       bra      2f\n"
+"       add r0,  %0\n"
+"       .balign  4\n"
+"1:     .long    _GLOBAL_OFFSET_TABLE_\n"
 "2:" : "=r" (got) : : "r0");
 #elif defined(__cris__)
   __asm__("\tmove.d $pc,%0\n\tsub.d .:GOTOFF,%0\n\t":"=r"(got));
@@ -629,6 +629,13 @@ LD_BOOT(unsigned long args)
 	START();
 }
 
+#if defined (SUPPORT_LD_DEBUG)
+static void debug_fini (int status, void *arg)
+{
+	(void)status;
+	_dl_dprintf(_dl_debug_file,"\ncalling fini: %s\n\n", (const char*)arg);
+}
+#endif    
 
 static void _dl_get_ready_to_run(struct elf_resolve *tpnt, struct elf_resolve *app_tpnt, 
 		unsigned long load_addr, unsigned long *hash_addr, Elf32_auxv_t auxvt[AT_EGID + 1], 
@@ -642,7 +649,9 @@ static void _dl_get_ready_to_run(struct elf_resolve *tpnt, struct elf_resolve *a
 	struct elf_resolve *tpnt1;
 	unsigned long brk_addr, *lpnt;
 	int (*_dl_atexit) (void *);
-
+#if defined (SUPPORT_LD_DEBUG)
+	int (*_dl_on_exit) (void (*FUNCTION)(int STATUS, void *ARG),void*);
+#endif
 
 	/* Now we have done the mandatory linking of some things.  We are now
 	   free to start using global variables, since these things have all been
@@ -1196,6 +1205,10 @@ static void _dl_get_ready_to_run(struct elf_resolve *tpnt, struct elf_resolve *a
 	}
 #endif
 	_dl_atexit = (int (*)(void *)) (intptr_t) _dl_find_hash("atexit", NULL, NULL, symbolrel);
+#if defined (SUPPORT_LD_DEBUG)
+	_dl_on_exit = (int (*)(void (*)(int, void *),void*)) 
+		(intptr_t) _dl_find_hash("on_exit", NULL, NULL, symbolrel);
+#endif
 
 	/*
 	 * OK, fix one more thing - set up the debug_addr structure to point
@@ -1246,6 +1259,12 @@ static void _dl_get_ready_to_run(struct elf_resolve *tpnt, struct elf_resolve *a
 		}
 		if (_dl_atexit && tpnt->dynamic_info[DT_FINI]) {
 			(*_dl_atexit) (tpnt->loadaddr + tpnt->dynamic_info[DT_FINI]);
+#if defined (SUPPORT_LD_DEBUG)
+			if(_dl_debug && _dl_on_exit)
+			{
+				(*_dl_on_exit)(debug_fini, tpnt->libname);
+			}
+#endif
 		}
 #ifdef LD_DEBUG
 		else {
@@ -1318,7 +1337,10 @@ int _dl_fixup(struct elf_resolve *tpnt)
 					tpnt->dynamic_info[DT_PLTRELSZ], 0);
 	}
 #if defined (SUPPORT_LD_DEBUG)
-	if(_dl_debug) _dl_dprintf(_dl_debug_file,"\nrelocation processing: %s; finished\n\n", tpnt->libname);	
+	if(_dl_debug) {
+		_dl_dprintf(_dl_debug_file,"\nrelocation processing: %s", tpnt->libname);     
+		_dl_dprintf(_dl_debug_file,"; finished\n\n");
+	}
 #endif    
 	return goof;
 }
diff --git a/ldso/ldso/readelflib1.c b/ldso/ldso/readelflib1.c
index 4259a4292..199726ff3 100644
--- a/ldso/ldso/readelflib1.c
+++ b/ldso/ldso/readelflib1.c
@@ -318,16 +318,13 @@ struct elf_resolve *_dl_load_elf_shared_library(int secure,
 	Elf32_Dyn *dpnt;
 	struct elf_resolve *tpnt;
 	elf_phdr *ppnt;
-	int piclib;
 	char *status;
-	int flags;
 	char header[4096];
 	unsigned long dynamic_info[24];
 	unsigned long *lpnt;
 	unsigned long libaddr;
 	unsigned long minvma = 0xffffffff, maxvma = 0;
-	int i;
-	int infile;
+	int i, flags, piclib, infile;
 
 	/* If this file is already loaded, skip this step */
 	tpnt = _dl_check_hashed_files(libname);
@@ -544,28 +541,32 @@ struct elf_resolve *_dl_load_elf_shared_library(int secure,
 #if defined(__mips__)
 	{
 		
-		int i = 1;
+		int indx = 1;
 		Elf32_Dyn *dpnt = (Elf32_Dyn *) dynamic_addr;
 
 		while(dpnt->d_tag) {
 			dpnt++;
-			i++;
+			indx++;
 		}
-		dynamic_size = i;
+		dynamic_size = indx;
 	}
 #endif
 
-	for (i = 0; i < dynamic_size; i++) 
 	{
-		if (dpnt->d_tag > DT_JMPREL) {
+		unsigned long indx;
+
+		for (indx = 0; indx < dynamic_size; indx++) 
+		{
+			if (dpnt->d_tag > DT_JMPREL) {
+				dpnt++;
+				continue;
+			}
+			dynamic_info[dpnt->d_tag] = dpnt->d_un.d_val;
+			if (dpnt->d_tag == DT_TEXTREL || SVR4_BUGCOMPAT)
+				dynamic_info[DT_TEXTREL] = 1;
 			dpnt++;
-			continue;
-		}
-		dynamic_info[dpnt->d_tag] = dpnt->d_un.d_val;
-		if (dpnt->d_tag == DT_TEXTREL || SVR4_BUGCOMPAT)
-			dynamic_info[DT_TEXTREL] = 1;
-		dpnt++;
-	};
+		};
+	}
 
 	/* If the TEXTREL is set, this means that we need to make the pages
 	   writable before we perform relocations.  Do this now. They get set
diff --git a/ldso/ldso/sh/boot1_arch.h b/ldso/ldso/sh/boot1_arch.h
index 798121dc0..40d6a0dd4 100644
--- a/ldso/ldso/sh/boot1_arch.h
+++ b/ldso/ldso/sh/boot1_arch.h
@@ -7,15 +7,14 @@ asm("\
 	.globl	_dl_boot
 _dl_boot:
         mov	r15, r4
-        mov.l	.L_dl_boot2, r1
-	mova	.L_dl_boot2, r0
-	add	r1, r0
-	jsr	@r0
-	 add	#4, r4
+	mov.l   .L_dl_boot2, r0
+	bsrf    r0
+	add	#4, r4
+.jmp_loc:
 	jmp	@r0
 	 mov    #0, r4        /* call _start with arg == 0 */
 .L_dl_boot2:\n\
-	.long	_dl_boot2-.\n\
+	.long   _dl_boot2-.jmp_loc\n\
 	.previous\n\
 ");
 
diff --git a/ldso/ldso/sh/dl-startup.h b/ldso/ldso/sh/dl-startup.h
index 798121dc0..40d6a0dd4 100644
--- a/ldso/ldso/sh/dl-startup.h
+++ b/ldso/ldso/sh/dl-startup.h
@@ -7,15 +7,14 @@ asm("\
 	.globl	_dl_boot
 _dl_boot:
         mov	r15, r4
-        mov.l	.L_dl_boot2, r1
-	mova	.L_dl_boot2, r0
-	add	r1, r0
-	jsr	@r0
-	 add	#4, r4
+	mov.l   .L_dl_boot2, r0
+	bsrf    r0
+	add	#4, r4
+.jmp_loc:
 	jmp	@r0
 	 mov    #0, r4        /* call _start with arg == 0 */
 .L_dl_boot2:\n\
-	.long	_dl_boot2-.\n\
+	.long   _dl_boot2-.jmp_loc\n\
 	.previous\n\
 ");
 
diff --git a/ldso/ldso/sh/dl-sysdep.h b/ldso/ldso/sh/dl-sysdep.h
index a0ff05fdb..cf8820e4e 100644
--- a/ldso/ldso/sh/dl-sysdep.h
+++ b/ldso/ldso/sh/dl-sysdep.h
@@ -34,13 +34,14 @@
 #define PERFORM_BOOTSTRAP_RELOC(RELP,REL,SYMBOL,LOAD)		\
 	switch(ELF32_R_TYPE((RELP)->r_info)){			\
 	case R_SH_REL32:					\
-		*(REL) += (RELP)->r_addend - (LOAD);		\
+		*(REL)  = (SYMBOL) + (RELP)->r_addend		\
+			    - (unsigned long)(REL);		\
 		break;						\
 	case R_SH_DIR32:					\
-		*(REL) += (SYMBOL) + (RELP)->r_addend;		\
+		*(REL)  = (SYMBOL) + (RELP)->r_addend;		\
 		break;						\
 	case R_SH_RELATIVE:					\
-		*(REL) += (LOAD);				\
+		*(REL)  = (LOAD) + (RELP)->r_addend;		\
 		break;						\
 	case R_SH_NONE:						\
 		break;						\
diff --git a/ldso/ldso/sh/elfinterp.c b/ldso/ldso/sh/elfinterp.c
index 3e0bb87be..d94389ef7 100644
--- a/ldso/ldso/sh/elfinterp.c
+++ b/ldso/ldso/sh/elfinterp.c
@@ -236,8 +236,7 @@ _dl_parse(struct elf_resolve *tpnt, struct dyn_elf *scope,
 		}
 		else if (res >0)
 		{
-			_dl_dprintf(2, "%s: can't resolve symbol '%s'\n",
-					_dl_progname, strtab + symtab[symtab_index].st_name);
+			_dl_dprintf(2, "can't resolve symbol\n");
 			goof += res;
 		}
 	  }
@@ -289,17 +288,17 @@ _dl_do_reloc (struct elf_resolve *tpnt,struct dyn_elf *scope,
 			/* handled later on */
 			break;
 		case R_SH_DIR32:
-			*reloc_addr += symbol_addr + rpnt->r_addend;
+			*reloc_addr = symbol_addr + rpnt->r_addend;
 			break;
 		case R_SH_JMP_SLOT:
 			*reloc_addr = symbol_addr + rpnt->r_addend;
 			break;
 		case R_SH_REL32:
-			*reloc_addr += rpnt->r_addend -
-					(unsigned long) tpnt->loadaddr;
+			*reloc_addr = symbol_addr + rpnt->r_addend -
+					(unsigned long) reloc_addr;
 			break;
 		case R_SH_RELATIVE:
-			*reloc_addr += (unsigned long) tpnt->loadaddr;
+			*reloc_addr = (unsigned long) tpnt->loadaddr + rpnt->r_addend;
 			break;
 		default:
 			return -1; /*call _dl_exit(1) */
diff --git a/ldso/ldso/sh/ld_sysdep.h b/ldso/ldso/sh/ld_sysdep.h
index a0ff05fdb..cf8820e4e 100644
--- a/ldso/ldso/sh/ld_sysdep.h
+++ b/ldso/ldso/sh/ld_sysdep.h
@@ -34,13 +34,14 @@
 #define PERFORM_BOOTSTRAP_RELOC(RELP,REL,SYMBOL,LOAD)		\
 	switch(ELF32_R_TYPE((RELP)->r_info)){			\
 	case R_SH_REL32:					\
-		*(REL) += (RELP)->r_addend - (LOAD);		\
+		*(REL)  = (SYMBOL) + (RELP)->r_addend		\
+			    - (unsigned long)(REL);		\
 		break;						\
 	case R_SH_DIR32:					\
-		*(REL) += (SYMBOL) + (RELP)->r_addend;		\
+		*(REL)  = (SYMBOL) + (RELP)->r_addend;		\
 		break;						\
 	case R_SH_RELATIVE:					\
-		*(REL) += (LOAD);				\
+		*(REL)  = (LOAD) + (RELP)->r_addend;		\
 		break;						\
 	case R_SH_NONE:						\
 		break;						\
diff --git a/ldso/ldso/sh/resolve.S b/ldso/ldso/sh/resolve.S
index 4d8eee6c2..7fef6d77f 100644
--- a/ldso/ldso/sh/resolve.S
+++ b/ldso/ldso/sh/resolve.S
@@ -3,6 +3,7 @@
  */
 
 	.text
+	.globl	_dl_linux_resolver
 	.globl	_dl_linux_resolve
 	.type	_dl_linux_resolve, @function
 	.balign	16
@@ -31,13 +32,27 @@ _dl_linux_resolve:
 	fmov.s	fr4, @-r15
 #endif
 	sts.l	pr, @-r15
+/* Note - The PLT entries have been "optimised" not to use r2.  r2 is used by
+   GCC to return the address of large structures, so it should not be
+   corrupted here.  This does mean however, that those PLTs does not conform
+   to the SH PIC ABI.  That spec says that r0 contains the type of the PLT
+   and r2 contains the GOT id.  The GNU Plt version stores the GOT id in r0 and
+   ignores the type.  We can easily detect this difference however,
+   since the type will always be 0 or 8, and the GOT ids will always be
+   greater than or equal to 12.
 
-	mov	r2, r4		! link map address
-
-	mov.l	3f, r0
-	jsr	@r0		! Call resolver
-	 mov	r1, r5		! Reloc offset
-
+   Found in binutils/bfd/elf32-sh.c by Stefan Allius <allius@atecom.com>
+ */
+	mov     #8 ,r5
+	cmp/gt  r5, r0
+	bt      1f
+	mov     r2, r0          ! link map address in r2 (SH PIC ABI)
+1:
+	mov     r0, r4          ! link map address in r0 (GNUs PLT)
+	mov.l   3f, r5
+	bsrf    r5
+	mov	r1, r5		! Reloc offset
+.jmp_loc:
 	lds.l	@r15+, pr	! Get register content back
 
 #ifdef HAVE_FPU
@@ -64,6 +79,6 @@ _dl_linux_resolve:
 
 	.balign	4
 3:
-	.long	_dl_linux_resolver
+	.long   _dl_linux_resolver - .jmp_loc
 	.size	_dl_linux_resolve, . - _dl_linux_resolve