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.

6 files changed, 45 insertions, 31 deletions

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