add support for init/finit (constructors and destructors)

this is mainly in hopes of supporting c++ (not yet possible for other
reasons) but will also help applications/libraries which use (and more
often, abuse) the gcc __attribute__((__constructor__)) feature in "C"
code.

x86_64 and arm versions of the new startup asm are untested and may
have minor problems.

crt/arm/crt1.s

@@ -1,13 +1,16 @@
+.weak _init
+.weak _fini
 .global _start
 _start:
 	mov fp,#0
 	mov lr,#0
 	ldr a2,[sp],#4
 	mov a3,sp
-	mov a4,#0
+	ldr a4,=_fini
 	str fp,[sp,#-4]!
 	str a1,[sp,#-4]!
-	str fp,[sp,#-4]!
+	str a4,[sp,#-4]!
+	ldr a4,=_init
 	ldr a1,=main
 	bl __libc_start_main
 1:	b 1b

crt/arm/crti.s

@@ -0,0 +1,7 @@
+.section .init
+.global _init
+_init:
+
+.section .fini
+.global _fini
+_fini:

crt/arm/crtn.s

@@ -0,0 +1,9 @@
+.section .init
+	tst lr,#1
+	moveq pc,lr
+	bx lr
+
+.section .fini
+	tst lr,#1
+	moveq pc,lr
+	bx lr

crt/i386/crt1.s

@@ -1,3 +1,5 @@
+.weak _init
+.weak _fini
 .text
 .global _start
 _start:
@@ -8,8 +10,10 @@ _start:
 	pushl %esp
 	pushl %esp
 	pushl %edx
-	pushl %ebp
+	call 1f
-	pushl %ebp
+1:	addl $[_fini-.],(%esp)
+	call 1f
+1:	addl $[_init-.],(%esp)
 	pushl %eax
 	pushl %ecx
 	call 1f

crt/i386/crti.s

@@ -0,0 +1,7 @@
+.section .init
+.global _init
+_init:
+
+.section .fini
+.global _fini
+_fini:

crt/i386/crtn.s

@@ -0,0 +1,5 @@
+.section .init
+	ret
+
+.section .fini
+	ret

crt/x86_64/crt1.s

@@ -1,4 +1,6 @@
 /* Written 2011 Nicholas J. Kain, released as Public Domain */
+.weak _init
+.weak _fini
 .text
 .global _start
 _start:
@@ -9,8 +11,8 @@ _start:
 	andq $-16,%rsp  /* align stack pointer */
 	push %rax       /* 8th arg: glibc ABI compatible */
 	push %rsp       /* 7th arg: glibc ABI compatible */
-	xor %r8,%r8     /* 5th arg: always 0 */
+	mov $_fini,%r8  /* 5th arg: fini/dtors function */
-	xor %rcx,%rcx   /* 4th arg: always 0 */
+	mov $_init,%rcx /* 4th arg: init/ctors function */
 	mov $main,%rdi  /* 1st arg: application entry ip */
 	call __libc_start_main /* musl init will run the program */
-.L0:	jmp .L0
+1:	jmp 1b

crt/x86_64/crti.s

@@ -0,0 +1,7 @@
+.section .init
+.global _init
+_init:
+
+.section .fini
+.global _fini
+_fini:

crt/x86_64/crtn.s

@@ -0,0 +1,5 @@
+.section .init
+	ret
+
+.section .fini
+	ret

src/ldso/dynlink.c

@@ -49,6 +49,7 @@ struct dso
 	ino_t ino;
 	int global;
 	int relocated;
+	int constructed;
 	struct dso **deps;
 	char *name;
 	char buf[];
@@ -471,6 +472,20 @@ static size_t find_dyn(Phdr *ph, size_t cnt, size_t stride)
 	return 0;
 }
 
+static void do_init_fini(struct dso *p)
+{
+	size_t dyn[DYN_CNT] = {0};
+	for (; p; p=p->prev) {
+		if (p->constructed) return;
+		decode_vec(p->dynv, dyn, DYN_CNT);
+		if (dyn[0] & (1<<DT_FINI))
+			atexit((void (*)(void))(p->base + dyn[DT_FINI]));
+		if (dyn[0] & (1<<DT_INIT))
+			((void (*)(void))(p->base + dyn[DT_INIT]))();
+		p->constructed = 1;
+	}
+}
+
 void *__dynlink(int argc, char **argv)
 {
 	size_t *auxv, aux[AUX_CNT] = {0};
@@ -520,6 +535,7 @@ void *__dynlink(int argc, char **argv)
 	}
 	app->name = argv[0];
 	app->global = 1;
+	app->constructed = 1;
 	app->dynv = (void *)(app->base + find_dyn(
 		(void *)aux[AT_PHDR], aux[AT_PHNUM], aux[AT_PHENT]));
 	decode_dyn(app);
@@ -577,6 +593,9 @@ void *__dynlink(int argc, char **argv)
 	 * error. If the dynamic loader (dlopen) will not be used, free
 	 * all memory used by the dynamic linker. */
 	runtime = 1;
+
+	do_init_fini(tail);
+
 	if (!rtld_used) {
 		free_all(head);
 		free(sys_path);