Clean up TLS_SLOT_BIONIC_PREINIT usage a bit

- It is only needed for dynamic executables, so move the initialization
   out of __libc_init_main_thread and just before the solib constructor
   calls. For static executables, the slot was initialized, then never
   used or cleared. Instead, leave it clear.

 - For static executables, __libc_init_main_thread already initialized the
   stack guard, so remove the redundant __init_thread_stack_guard call.

 - Simplify the slot access/clearing a bit in __libc_preinit.

 - Remove the "__libc_init_common() will change the TLS area so the old one
   won't be accessible anyway." comment. AFAICT, it's incorrect -- the
   main thread's TLS area in a dynamic executable is initialized to a
   static pthread_internal_t object in the linker, then reused by libc.so.

Test: adb shell /data/nativetest/bionic-unit-tests/bionic-unit-tests
Test: adb shell /data/nativetest/bionic-unit-tests-static/bionic-unit-tests-static
Change-Id: Ie2da6f5be3ad563fa65b38eaadf8ba6ecc6a64b6

libc/bionic/__libc_init_main_thread.cpp

@@ -81,8 +81,7 @@ void __libc_init_main_thread(KernelArgumentBlock& args) {
 
   // We don't want to free the main thread's stack even when the main thread exits
   // because things like environment variables with global scope live on it.
-  // We also can't free the pthread_internal_t itself, since that lives on the main
-  // thread's stack rather than on the heap.
+  // We also can't free the pthread_internal_t itself, since it is a static variable.
   // The main thread has no mmap allocated space for stack or pthread_internal_t.
   main_thread.mmap_size = 0;
 
@@ -102,9 +101,5 @@ void __libc_init_main_thread(KernelArgumentBlock& args) {
 
   __init_thread(&main_thread);
 
-  // Store a pointer to the kernel argument block in a TLS slot to be
-  // picked up by the libc constructor.
-  main_thread.tls[TLS_SLOT_BIONIC_PREINIT] = &args;
-
   __init_alternate_signal_stack(&main_thread);
 }

libc/bionic/libc_init_dynamic.cpp

@@ -85,14 +85,11 @@ static void __libc_preinit_impl(KernelArgumentBlock& args) {
 // to run before any others (such as the jemalloc constructor), and lower
 // is better (http://b/68046352).
 __attribute__((constructor(1))) static void __libc_preinit() {
-  // Read the kernel argument block pointer from TLS.
+  // Read the kernel argument block pointer from TLS, then clear the slot so no
+  // other initializer sees its value.
   void** tls = __get_tls();
-  KernelArgumentBlock** args_slot = &reinterpret_cast<KernelArgumentBlock**>(tls)[TLS_SLOT_BIONIC_PREINIT];
-  KernelArgumentBlock* args = *args_slot;
-
-  // 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.
-  *args_slot = NULL;
+  KernelArgumentBlock* args = static_cast<KernelArgumentBlock*>(tls[TLS_SLOT_BIONIC_PREINIT]);
+  tls[TLS_SLOT_BIONIC_PREINIT] = nullptr;
 
   // The linker has initialized its copy of the global stack_chk_guard, and filled in the main
   // thread's TLS slot with that value. Initialize the local global stack guard with its value.
@@ -102,9 +99,8 @@ __attribute__((constructor(1))) static void __libc_preinit() {
 }
 
 // This function is called from the executable's _start entry point
-// (see arch-$ARCH/bionic/crtbegin_dynamic.S), which is itself
-// called by the dynamic linker after it has loaded all shared
-// libraries the executable depends on.
+// (see arch-$ARCH/bionic/crtbegin.c), which is itself called by the dynamic
+// linker after it has loaded all shared libraries the executable depends on.
 //
 // Note that the dynamic linker has also run all constructors in the
 // executable at this point.

libc/bionic/libc_init_static.cpp

@@ -86,10 +86,9 @@ __noreturn void __libc_init(void* raw_args,
   BIONIC_STOP_UNWIND;
 
   KernelArgumentBlock args(raw_args);
-  __libc_init_main_thread(args);
 
   // Initializing the globals requires TLS to be available for errno.
-  __init_thread_stack_guard(__get_thread());
+  __libc_init_main_thread(args);
   __libc_init_globals(args);
 
   __libc_init_AT_SECURE(args);

linker/linker_main.cpp

@@ -414,6 +414,10 @@ static ElfW(Addr) __linker_init_post_relocation(KernelArgumentBlock& args) {
 
   if (!get_cfi_shadow()->InitialLinkDone(solist)) __linker_cannot_link(g_argv[0]);
 
+  // Store a pointer to the kernel argument block in a TLS slot to be
+  // picked up by the libc constructor.
+  __get_tls()[TLS_SLOT_BIONIC_PREINIT] = &args;
+
   si->call_pre_init_constructors();
 
   /* After the prelink_image, the si->load_bias is initialized.