Kernel+Userland: Make the stack alignment comply with the System V ABI

The System V ABI for both x86 and x86_64 requires that the stack pointer
is 16-byte aligned on entry. Previously we did not align the stack
pointer properly.

As far as "main" was concerned the stack alignment was correct even
without this patch due to how the C++ _start function and the kernel
interacted, i.e. the kernel misaligned the stack as far as the ABI
was concerned but that misalignment (read: it was properly aligned for
a regular function call - but misaligned in terms of what the ABI
dictates) was actually expected by our _start function.

Kernel/Syscalls/execve.cpp

@@ -147,10 +147,10 @@ static KResultOr<FlatPtr> make_userspace_context_for_main_thread([[maybe_unused]
     regs.rsi = argv;
     regs.rdx = envp;
 #endif
-    push_on_new_stack(0); // return address
 
-    VERIFY((new_sp + sizeof(void*)) % 16 == 0);
+    VERIFY(new_sp % 16 == 0);
 
+    // FIXME: The way we're setting up the stack and passing arguments to the entry point isn't ABI-compliant
     return new_sp;
 }
 

Userland/DynamicLoader/CMakeLists.txt

@@ -10,10 +10,10 @@ file(GLOB LIBC_SOURCES2 "../Libraries/LibC/*/*.cpp")
 
 if ("${SERENITY_ARCH}" STREQUAL "i686")
     file(GLOB LIBC_SOURCES3 "../Libraries/LibC/arch/i386/*.S")
-    set(ELF_SOURCES ${ELF_SOURCES} ../Libraries/LibELF/Arch/i386/plt_trampoline.S)
+    set(ELF_SOURCES ${ELF_SOURCES} ../Libraries/LibELF/Arch/i386/entry.S ../Libraries/LibELF/Arch/i386/plt_trampoline.S)
 elseif ("${SERENITY_ARCH}" STREQUAL "x86_64")
     file(GLOB LIBC_SOURCES3 "../Libraries/LibC/arch/x86_64/*.S")
-    set(ELF_SOURCES ${ELF_SOURCES} ../Libraries/LibELF/Arch/x86_64/plt_trampoline.S)
+    set(ELF_SOURCES ${ELF_SOURCES} ../Libraries/LibELF/Arch/x86_64/entry.S ../Libraries/LibELF/Arch/x86_64/plt_trampoline.S)
 endif()
 
 file(GLOB LIBSYSTEM_SOURCES "../Libraries/LibSystem/*.cpp")

Userland/DynamicLoader/main.cpp

@@ -94,9 +94,15 @@ this helper program directly.
 extern "C" {
 
 // The compiler expects a previous declaration
-void _start(int, char**, char**);
+void _entry(int, char**, char**);
 
-void _start(int argc, char** argv, char** envp)
+asm(
+    ".globl _start\n"
+    "_start:\n"
+    "push $0\n"
+    "jmp _entry@plt\n");
+
+void _entry(int argc, char** argv, char** envp)
 {
     char** env;
     for (env = envp; *env; ++env) {

Userland/Libraries/LibC/CMakeLists.txt

@@ -64,10 +64,10 @@ file(GLOB ELF_SOURCES CONFIGURE_DEPENDS "../LibELF/*.cpp")
 
 if ("${SERENITY_ARCH}" STREQUAL "i686")
     set(ASM_SOURCES "arch/i386/setjmp.S")
-    set(ELF_SOURCES ${ELF_SOURCES} ../LibELF/Arch/i386/plt_trampoline.S)
+    set(ELF_SOURCES ${ELF_SOURCES} ../LibELF/Arch/i386/entry.S ../LibELF/Arch/i386/plt_trampoline.S)
 elseif ("${SERENITY_ARCH}" STREQUAL "x86_64")
     set(ASM_SOURCES "arch/x86_64/setjmp.S")
-    set(ELF_SOURCES ${ELF_SOURCES} ../LibELF/Arch/x86_64/plt_trampoline.S)
+    set(ELF_SOURCES ${ELF_SOURCES} ../LibELF/Arch/x86_64/entry.S ../LibELF/Arch/x86_64/plt_trampoline.S)
 endif()
 
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-unknown-warning-option -DSERENITY_LIBC_BUILD")

Userland/Libraries/LibC/crt0.cpp

@@ -19,9 +19,15 @@ extern u32 __stack_chk_guard;
 int main(int, char**, char**);
 
 // Tell the compiler that this may be called from somewhere else.
-int _start(int argc, char** argv, char** env);
+int _entry(int argc, char** argv, char** env);
 
-int _start(int argc, char** argv, char** env)
+asm(
+    ".globl _start\n"
+    "_start:\n"
+    "push $0\n"
+    "jmp _entry@plt\n");
+
+int _entry(int argc, char** argv, char** env)
 {
     u32 original_stack_chk = __stack_chk_guard;
     arc4random_buf(&__stack_chk_guard, sizeof(__stack_chk_guard));

Userland/Libraries/LibELF/Arch/i386/entry.S

@@ -0,0 +1,28 @@
+/*
+ * Copyright (c) 2021, Gunnar Beutner <gbeutner@serenityos.org>
+ *
+ * SPDX-License-Identifier: BSD-2-Clause
+ */
+
+    .align 4
+    .globl _invoke_entry
+    .hidden _invoke_entry
+    .type _invoke_entry,@function
+_invoke_entry: # (argc, argv, envp, entry)
+    addl $4, %esp # return address
+    popl %edi # argc
+    popl %esi # argv
+    popl %edx # envp
+    popl %ecx # entry
+
+    // The System V ABI for x86 and x86_64 prescribes that the stack pointer is 16-byte aligned
+    andl $~15, %esp
+
+    // We're going to push three arguments so we need to align the stack for that
+    subl $4, %esp
+
+    // FIXME: The way we're setting up the stack and passing arguments to the entry point isn't ABI-compliant
+    pushl %edx
+    pushl %esi
+    pushl %edi
+    jmp *%ecx

Userland/Libraries/LibELF/Arch/x86_64/entry.S

@@ -0,0 +1,16 @@
+/*
+ * Copyright (c) 2021, Gunnar Beutner <gbeutner@serenityos.org>
+ *
+ * SPDX-License-Identifier: BSD-2-Clause
+ */
+
+    .align 4
+    .globl _invoke_entry
+    .hidden _invoke_entry
+    .type _invoke_entry,@function
+_invoke_entry: # (argc, argv, envp, entry)
+    // The System V ABI for x86 and x86_64 prescribes that the stack pointer is 16-byte aligned
+    andq $~15, %rsp
+
+    // FIXME: The way we're setting up the stack and passing arguments to the entry point isn't ABI-compliant
+    jmp *%rcx

Userland/Libraries/LibELF/DynamicLinker.cpp

@@ -41,6 +41,8 @@ using LibCExitFunction = void (*)(int);
 using DlIteratePhdrCallbackFunction = int (*)(struct dl_phdr_info*, size_t, void*);
 using DlIteratePhdrFunction = int (*)(DlIteratePhdrCallbackFunction, void*);
 
+extern "C" [[noreturn]] void _invoke_entry(int argc, char** argv, char** envp, EntryPointFunction entry);
+
 static size_t s_current_tls_offset = 0;
 static size_t s_total_tls_size = 0;
 static size_t s_allocated_tls_block_size = 0;
@@ -550,14 +552,8 @@ void ELF::DynamicLinker::linker_main(String&& main_program_name, int main_progra
     if (s_do_breakpoint_trap_before_entry) {
         asm("int3");
     }
-    rc = entry_point_function(argc, argv, envp);
-    dbgln_if(DYNAMIC_LOAD_DEBUG, "rc: {}", rc);
-    if (s_libc_exit != nullptr) {
-        s_libc_exit(rc);
-    } else {
-        _exit(rc);
-    }
 
+    _invoke_entry(argc, argv, envp, entry_point_function);
     VERIFY_NOT_REACHED();
 }