Ensure stack is always aligned with 8 bytes

In dynamic linking mode, consider external functions may use data types
such as 'long long int' or other equivalents. The program may encounter
a Bus error if external functions access 8-byte data on the stack that
is not properly aligned.

To prevent the aforementioned issue, these changes adjust the Arm code
generator to ensure that the program's stack is always aligned with 8
bytes.

Since the RISC-V architecture does not yet support dynamic linking, the
corresponding code generator is not modified in this commit.

Author: DrXiao

src/arm-codegen.c

@@ -219,8 +219,9 @@ void emit_ph2_ir(ph2_ir_t *ph2_ir)
     switch (ph2_ir->op) {
     case OP_define:
         emit(__sw(__AL, __lr, __sp, -4));
-        emit(__movw(__AL, __r8, ph2_ir->src0 + 4));
-        emit(__movt(__AL, __r8, ph2_ir->src0 + 4));
+        ofs = ALIGN_UP(ph2_ir->src0 + 4, 8);
+        emit(__movw(__AL, __r8, ofs));
+        emit(__movt(__AL, __r8, ofs));
         emit(__sub_r(__AL, __sp, __sp, __r8));
         return;
     case OP_load_constant:
@@ -342,8 +343,9 @@ void emit_ph2_ir(ph2_ir_t *ph2_ir)
             emit(__mov_r(__AL, __r0, __r0));
         else
             emit(__mov_r(__AL, __r0, rn));
-        emit(__movw(__AL, __r8, ph2_ir->src1 + 4));
-        emit(__movt(__AL, __r8, ph2_ir->src1 + 4));
+        ofs = ALIGN_UP(ph2_ir->src1 + 4, 8);
+        emit(__movw(__AL, __r8, ofs));
+        emit(__movt(__AL, __r8, ofs));
         emit(__add_r(__AL, __sp, __sp, __r8));
         emit(__lw(__AL, __lr, __sp, -4));
         emit(__bx(__AL, __lr));
@@ -486,6 +488,8 @@ void emit_ph2_ir(ph2_ir_t *ph2_ir)
 void plt_generate(void);
 void code_generate(void)
 {
+    int ofs;
+
     if (dynlink) {
         plt_generate();
         /* Call __libc_start_main() */
@@ -521,8 +525,9 @@ void code_generate(void)
     /* For both static and dynamic linking, we need to set up the stack
      * and call the main function.
      * */
-    emit(__movw(__AL, __r8, GLOBAL_FUNC->stack_size));
-    emit(__movt(__AL, __r8, GLOBAL_FUNC->stack_size));
+    ofs = ALIGN_UP(GLOBAL_FUNC->stack_size, 8);
+    emit(__movw(__AL, __r8, ofs));
+    emit(__movt(__AL, __r8, ofs));
     emit(__sub_r(__AL, __sp, __sp, __r8));
     emit(__mov_r(__AL, __r12, __sp));
 
@@ -533,8 +538,8 @@ void code_generate(void)
                  56)); /* PC+8: skip exit (24) + syscall (36) + ret (4) - 8 */
 
         /* exit - only for static linking */
-        emit(__movw(__AL, __r8, GLOBAL_FUNC->stack_size));
-        emit(__movt(__AL, __r8, GLOBAL_FUNC->stack_size));
+        emit(__movw(__AL, __r8, ofs));
+        emit(__movt(__AL, __r8, ofs));
         emit(__add_r(__AL, __sp, __sp, __r8));
         emit(__mov_r(__AL, __r0, __r0));
         emit(__mov_i(__AL, __r7, 1));
@@ -568,8 +573,8 @@ void code_generate(void)
              * will return to __libc_start_main. */
             emit(__b(__AL, MAIN_BB->elf_offset - elf_code->size));
         } else {
-            emit(__movw(__AL, __r8, GLOBAL_FUNC->stack_size));
-            emit(__movt(__AL, __r8, GLOBAL_FUNC->stack_size));
+            emit(__movw(__AL, __r8, ofs));
+            emit(__movt(__AL, __r8, ofs));
             emit(__add_r(__AL, __r8, __r12, __r8));
             emit(__lw(__AL, __r0, __r8, 0));
             emit(__add_i(__AL, __r1, __r8, 4));

src/defs.h

@@ -91,6 +91,10 @@
 #define HOST_PTR_SIZE __SIZEOF_POINTER__
 #endif
 
+#ifndef ALIGN_UP
+#define ALIGN_UP(val, align) (((val) + (align) - 1) & ~((align) - 1))
+#endif
+
 /* Common data structures */
 typedef struct arena_block {
     char *memory;