Fix the C library runtime initialization order.

This allows libc.so to run the C runtime initializer as soon as the
dynamic linker loads the shared library, i.e. before any other initializers
(e.g. static C++ constructors in other shared libraries the executable depends
on).

This also removes the bug where the initializers from the executable itself
were run twice: once by the dynamic linker, and another time by __libc_init
as defined by libc_init_dynamic.c
diff --git a/libc/bionic/libc_init_common.c b/libc/bionic/libc_init_common.c
index de4919d..c77c162 100644
--- a/libc/bionic/libc_init_common.c
+++ b/libc/bionic/libc_init_common.c
@@ -39,23 +39,6 @@
 #include <bionic_tls.h>
 #include <errno.h>
 
-extern void _init(void);
-extern void _fini(void);
-
-static void call_array(void(**list)())
-{
-    // First element is -1, list is null-terminated
-    while (*++list) {
-        (*list)();
-    }
-}
-
-static void __bionic_do_global_dtors(structors_array_t const * const p)
-{
-    call_array(p->fini_array);
-    //_fini();
-}
-
 extern unsigned __get_sp(void);
 extern pid_t    gettid(void);
 
@@ -69,23 +52,23 @@
 
 int __system_properties_init(void);
 
-void __libc_init_common(uintptr_t *elfdata,
-                       void (*onexit)(void),
-                       int (*slingshot)(int, char**, char**),
-                       structors_array_t const * const structors,
-                       void (*pre_ctor_hook)())
-{
-    pthread_internal_t thread;
-    pthread_attr_t thread_attr;
-    void *tls_area[BIONIC_TLS_SLOTS];
-    int argc;
-    char **argv, **envp, **envend;
-    struct auxentry *auxentry;
-    unsigned int page_size = 0, page_shift = 0;
+#ifdef MALLOCK_LEAK_CHECK
+void malloc_debug_init(void);
+#endif
 
-    /* The main thread's stack has empirically shown to be 84k */
+void __libc_init_common(uintptr_t *elfdata)
+{
+    int     argc = *elfdata;
+    char**  argv = (char**)(elfdata + 1);
+    char**  envp = argv + argc + 1;
+
+    pthread_attr_t             thread_attr;
+    static pthread_internal_t  thread;
+    static void*               tls_area[BIONIC_TLS_SLOTS];
+
+    /* setup pthread runtime and maint thread descriptor */
     unsigned stacktop = (__get_sp() & ~(PAGE_SIZE - 1)) + PAGE_SIZE;
-    unsigned stacksize = 128 * 1024; //84 * 1024;
+    unsigned stacksize = 128 * 1024;
     unsigned stackbottom = stacktop - stacksize;
 
     pthread_attr_init(&thread_attr);
@@ -93,30 +76,20 @@
     _init_thread(&thread, gettid(), &thread_attr, (void*)stackbottom);
     __init_tls(tls_area, &thread);
 
-    argc = (int) *elfdata++;
-    argv = (char**) elfdata;
-    envp = argv+(argc+1);
-    environ = envp;
-
-    __progname = argv[0] ? argv[0] : "<unknown>";
-
+    /* clear errno - requires TLS area */
     errno = 0;
 
+    /* set program name */
+    __progname = argv[0] ? argv[0] : "<unknown>";
+
+    /* setup environment pointer */
+    environ = envp;
+
+    /* setup system properties - requires environment */
     __system_properties_init();
 
-    if (pre_ctor_hook) pre_ctor_hook();
-
-    // XXX: we should execute the .fini_array upon exit
-
-    // pre-init array.
-    // XXX: I'm not sure what's the different with the init array.
-    call_array(structors->preinit_array);
-
-    // for compatibility with non-eabi binary, call the .ctors section
-    call_array(structors->ctors_array);
-
-    // call static constructors
-    call_array(structors->init_array);
-
-    exit(slingshot(argc, argv, envp));
+    /* setup malloc leak checker, requires system properties */
+#if MALLOC_LEAK_CHECK
+    malloc_debug_init();
+#endif
 }
diff --git a/libc/bionic/libc_init_common.h b/libc/bionic/libc_init_common.h
index bbc82e4..8663c61 100644
--- a/libc/bionic/libc_init_common.h
+++ b/libc/bionic/libc_init_common.h
@@ -38,10 +38,6 @@
     void (**ctors_array)(void);
 } structors_array_t;
 
-extern __noreturn void __libc_init_common(uintptr_t *elfdata,
-                       void (*onexit)(void),
-                       int (*slingshot)(int, char**, char**),
-                       structors_array_t const * const structors,
-                       void (*pre_ctor_hook)());
+extern void __libc_init_common(uintptr_t *elfdata);
 
 #endif
diff --git a/libc/bionic/libc_init_dynamic.c b/libc/bionic/libc_init_dynamic.c
index 8cf24b4..b8e1078 100644
--- a/libc/bionic/libc_init_dynamic.c
+++ b/libc/bionic/libc_init_dynamic.c
@@ -26,41 +26,74 @@
  * SUCH DAMAGE.
  */
 /*
- * libc_init_static.c
+ * libc_init_dynamic.c
  *
- * This function takes the raw data block set up by the ELF loader
- * in the kernel and parses it.  It is invoked by crt0.S which makes
- * any necessary adjustments and passes calls this function using
- * the standard C calling convention.
+ * This source files provides two important functions for dynamic
+ * executables:
  *
- * The arguments are:
- *  uintptr_t *elfdata	 -- The ELF loader data block; usually from the stack.
- *                          Basically a pointer to argc.
- *  void (*onexit)(void) -- Function to install into onexit
+ * - a C runtime initializer (__libc_preinit), which is called by
+ *   the dynamic linker when libc.so is loaded. This happens before
+ *   any other initializer (e.g. static C++ constructors in other
+ *   shared libraries the program depends on).
+ *
+ * - a program launch function (__libc_init), which is called after
+ *   all dynamic linking has been performed. Technically, it is called
+ *   from arch-$ARCH/bionic/crtbegin_dynamic.S which is itself called
+ *   by the dynamic linker after all libraries have been loaded and
+ *   initialized.
  */
 
-/*
- * Several Linux ABIs don't pass the onexit pointer, and the ones that
- * do never use it.  Therefore, unless USE_ONEXIT is defined, we just
- * ignore the onexit pointer.
- */
-/* #define USE_ONEXIT */
-
 #include <stddef.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <stdint.h>
 #include <elf.h>
-#include "pthread_internal.h"
 #include "atexit.h"
 #include "libc_init_common.h"
+#include <bionic_tls.h>
 
 extern void malloc_debug_init();
 
+/* We flag the __libc_preinit function as a constructor to ensure
+ * that its address is listed in libc.so's .init_array section.
+ * This ensures that the function is called by the dynamic linker
+ * as soon as the shared library is loaded.
+ */
+void __attribute__((constructor)) __libc_prenit(void);
+
+void __libc_prenit(void)
+{
+    /* Read the ELF data pointer form a special slot of the
+     * TLS area, then call __libc_init_common with it.
+     *
+     * Note that:
+     * - we clear the slot so no other initializer sees its value.
+     * - __libc_init_common() will change the TLS area so the old one
+     *   won't be accessible anyway.
+     */
+    void**      tls_area = (void**)__get_tls();
+    unsigned*   elfdata   = tls_area[TLS_SLOT_BIONIC_PREINIT];
+
+    tls_area[TLS_SLOT_BIONIC_PREINIT] = NULL;
+
+    __libc_init_common(elfdata);
+}
+
 __noreturn void __libc_init(uintptr_t *elfdata,
                        void (*onexit)(void),
                        int (*slingshot)(int, char**, char**),
                        structors_array_t const * const structors)
 {
-    __libc_init_common(elfdata, onexit, slingshot, structors, malloc_debug_init);
+    /* When we reach this point, all initializers have been already
+     * run by the dynamic linker, so ignore 'structors'.
+     */
+    int     argc = (int)*elfdata;
+    char**  argv = (char**)(elfdata + 1);
+    char**  envp = argv + argc + 1;
+
+    /* Several Linux ABIs don't pass the onexit pointer, and the ones that
+     * do never use it.  Therefore, we ignore it.
+     */
+
+    exit(slingshot(argc, argv, envp));
 }
diff --git a/libc/bionic/libc_init_static.c b/libc/bionic/libc_init_static.c
index ec463f7..d097b6b 100644
--- a/libc/bionic/libc_init_static.c
+++ b/libc/bionic/libc_init_static.c
@@ -28,24 +28,15 @@
 /*
  * libc_init_static.c
  *
- * This function takes the raw data block set up by the ELF loader
- * in the kernel and parses it.  It is invoked by crt0.S which makes
- * any necessary adjustments and passes calls this function using
- * the standard C calling convention.
+ * The program startup function __libc_init() defined here is
+ * used for static executables only (i.e. those that don't depend
+ * on shared libraries). It is called from arch-$ARCH/bionic/crtbegin_static.S
+ * which is directly invoked by the kernel when the program is launched.
  *
- * The arguments are:
- *  uintptr_t *elfdata	 -- The ELF loader data block; usually from the stack.
- *                          Basically a pointer to argc.
- *  void (*onexit)(void) -- Function to install into onexit
+ * The 'structors' parameter contains pointers to various initializer
+ * arrays that must be run before the program's 'main' routine is launched.
  */
 
-/*
- * Several Linux ABIs don't pass the onexit pointer, and the ones that
- * do never use it.  Therefore, unless USE_ONEXIT is defined, we just
- * ignore the onexit pointer.
- */
-/* #define USE_ONEXIT */
-
 #include <stddef.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -58,19 +49,41 @@
 #include <bionic_tls.h>
 #include <errno.h>
 
+static void call_array(void(**list)())
+{
+    // First element is -1, list is null-terminated
+    while (*++list) {
+        (*list)();
+    }
+}
+
 __noreturn void __libc_init(uintptr_t *elfdata,
                        void (*onexit)(void),
                        int (*slingshot)(int, char**, char**),
                        structors_array_t const * const structors)
 {
-/* 
- * To enable malloc checks for statically linked programs, add
- * "WITH_MALLOC_CHECK_LIBC_A := true" in device/buildspec.mk
- */
-#ifdef MALLOC_LEAK_CHECK
-    extern void malloc_debug_init();
-    __libc_init_common(elfdata, onexit, slingshot, structors, malloc_debug_init);
-#else
-    __libc_init_common(elfdata, onexit, slingshot, structors, NULL);
-#endif
+    int  argc;
+    char **argv, **envp;
+
+    /* Initialize the C runtime environment */
+    __libc_init_common(elfdata);
+
+    /* Several Linux ABIs don't pass the onexit pointer, and the ones that
+     * do never use it.  Therefore, we ignore it.
+     */
+
+    /* pre-init array. */
+    call_array(structors->preinit_array);
+
+    /* .ctors section initializers, for non-arm-eabi ABIs */
+    call_array(structors->ctors_array);
+
+    // call static constructors
+    call_array(structors->init_array);
+
+    argc = (int) *elfdata;
+    argv = (char**)(elfdata + 1);
+    envp = argv + argc + 1;
+
+    exit(slingshot(argc, argv, envp));
 }