DynamicLoader: Fix TLS on 32 bit platform

There were two problems with my previous implementation
- TLS was not allocated if nothing used it. There is a fallback
  initialization in _init_libc, but this was not enough if one of the
  init functions tried to access errno.
- __tls_get_addr was not resolved. If __tls_get_addr was called through
  a plt entry, everything would just break :(

userspace/libraries/LibC/CMakeLists.txt

@@ -64,10 +64,6 @@ target_compile_definitions(objlibc PRIVATE __arch=${BANAN_ARCH})
 target_compile_options(objlibc PRIVATE -O2 -g -Wstack-usage=512 -fno-exceptions -fpic -nolibc)
 target_compile_options(objlibc PUBLIC -Wall -Wextra -Werror -Wno-error=stack-usage=)
 
-if("${BANAN_ARCH}" STREQUAL "i686")
-	target_compile_definitions(objlibc PRIVATE __disable_thread_local_storage)
-endif()
-
 function(add_crtx crtx)
 	add_custom_target(${crtx}
 		COMMAND ${CMAKE_CXX_COMPILER} -c -o ${CMAKE_INSTALL_LIBDIR}/${crtx}.o ${CMAKE_CURRENT_SOURCE_DIR}/arch/${BANAN_ARCH}/${crtx}.S

userspace/libraries/LibC/pthread.cpp

@@ -105,7 +105,6 @@ void pthread_cleanup_push(void (*routine)(void*), void* arg)
 	uthread->cleanup_stack = cleanup;
 }
 
-#if not __disable_thread_local_storage
 static thread_local struct {
 	void* value;
 	void (*destructor)(void*);
@@ -159,7 +158,6 @@ int pthread_setspecific(pthread_key_t key, const void* value)
 	s_pthread_keys[key].value = const_cast<void*>(value);
 	return 0;
 }
-#endif
 
 int pthread_attr_destroy(pthread_attr_t* attr)
 {
@@ -413,7 +411,6 @@ void pthread_exit(void* value_ptr)
 	while (uthread->cleanup_stack)
 		pthread_cleanup_pop(1);
 
-#if not __disable_thread_local_storage
 	for (size_t iteration = 0; iteration < PTHREAD_DESTRUCTOR_ITERATIONS; iteration++)
 	{
 		bool called = false;
@@ -431,7 +428,6 @@ void pthread_exit(void* value_ptr)
 		if (!called)
 			break;
 	}
-#endif
 
 	free_uthread(uthread);
 	syscall(SYS_PTHREAD_EXIT, value_ptr);
@@ -1199,7 +1195,6 @@ int pthread_barrier_wait(pthread_barrier_t* barrier)
 	return 0;
 }
 
-#if not __disable_thread_local_storage
 struct tls_index
 {
 	unsigned long int ti_module;
@@ -1214,7 +1209,6 @@ extern "C" void* __tls_get_addr(tls_index* ti)
 #if ARCH(i686)
 extern "C" void* __attribute__((__regparm__(1))) ___tls_get_addr(tls_index* ti)
 {
-	return reinterpret_cast<void*>(get_uthread()->dtv[ti->ti_module] + ti->ti_offset);
+	return reinterpret_cast<void*>(_get_uthread()->dtv[ti->ti_module] + ti->ti_offset);
 }
 #endif
-#endif

userspace/programs/DynamicLoader/main.cpp

@@ -338,7 +338,6 @@ static void handle_tls_relocation(const LoadedElf& elf, const RelocT& reloc)
 	if (elf.tls_addr == nullptr)
 		print_error_and_exit("tls relocation without tls", 0);
 
-
 #if defined(__x86_64__)
 	switch (ELF64_R_TYPE(reloc.r_info))
 	{
@@ -607,7 +606,27 @@ static void relocate_elf(LoadedElf& elf, bool lazy_load)
 				#error "unsupported architecture"
 #endif
 
-				*reinterpret_cast<uintptr_t*>(elf.base + offset) += elf.base;
+				bool do_relocation = false;
+
+				if (const uint32_t symbol_index = ELF_R_SYM(info))
+				{
+					const auto& symbol = *reinterpret_cast<ElfNativeSymbol*>(elf.symtab + symbol_index * elf.syment);
+					const char* symbol_name = reinterpret_cast<const char*>(elf.strtab + symbol.st_name);
+					if (strcmp(symbol_name, "__tls_get_addr") == 0 || strcmp(symbol_name, "___tls_get_addr") == 0)
+						do_relocation = true;
+				}
+
+				if (!do_relocation)
+					*reinterpret_cast<uintptr_t*>(elf.base + offset) += elf.base;
+				else switch (elf.pltrel)
+				{
+					case DT_REL:
+						handle_relocation(elf, reinterpret_cast<ElfNativeRelocation*>(elf.jmprel)[i], true);
+						break;
+					case DT_RELA:
+						handle_relocation(elf, reinterpret_cast<ElfNativeRelocationA*>(elf.jmprel)[i], true);
+						break;
+				}
 			}
 		}
 	}
@@ -1063,9 +1082,6 @@ static MasterTLS initialize_master_tls()
 
 static void initialize_tls(MasterTLS master_tls)
 {
-	if (master_tls.addr == nullptr)
-		return;
-
 	const size_t tls_size = master_tls.size
 		+ sizeof(uthread)
 		+ (master_tls.module_count + 1) * sizeof(uintptr_t);