diff --git a/tcc.h b/tcc.h
index 3f8647d..eeeb01c 100644
--- a/tcc.h
+++ b/tcc.h
@@ -1036,7 +1036,8 @@ struct filespec {
 #define VT_STATIC  0x00002000  /* static variable */
 #define VT_TYPEDEF 0x00004000  /* typedef definition */
 #define VT_INLINE  0x00008000  /* inline definition */
-/* currently unused: 0x000[1248]0000  */
+#define VT_PACKED  0x00010000  /* packed struct */
+/* currently unused: 0x000[248]0000  */
 
 #define VT_STRUCT_SHIFT 20     /* shift for bitfield shift values (32 - 2*6) */
 #define VT_STRUCT_MASK (((1U << (6+6)) - 1) << VT_STRUCT_SHIFT | VT_BITFIELD)
diff --git a/tccgen.c b/tccgen.c
index 33a91f8..39bb682 100644
--- a/tccgen.c
+++ b/tccgen.c
@@ -2316,11 +2316,18 @@ ST_FUNC int gv(int rc)
     int bit_pos, bit_size, size, align;
 
     /* NOTE: get_reg can modify vstack[] */
-    if (vtop->type.t & VT_BITFIELD) {
+    if (vtop->type.t & (VT_BITFIELD | VT_PACKED)) {
         CType type;
 
-        bit_pos = BIT_POS(vtop->type.t);
-        bit_size = BIT_SIZE(vtop->type.t);
+	if (vtop->type.t & VT_BITFIELD) {
+            bit_pos = BIT_POS(vtop->type.t);
+            bit_size = BIT_SIZE(vtop->type.t);
+	}
+	else {
+            size = type_size(&vtop->type, &align);
+            bit_pos = 0;
+            bit_size = size * 8;
+	}
         /* remove bit field info to avoid loops */
         vtop->type.t &= ~VT_STRUCT_MASK;
 
@@ -2329,12 +2336,17 @@ ST_FUNC int gv(int rc)
         if ((vtop->type.t & VT_BTYPE) == VT_BOOL)
             type.t |= VT_UNSIGNED;
 
-        r = adjust_bf(vtop, bit_pos, bit_size);
-
-        if ((vtop->type.t & VT_BTYPE) == VT_LLONG)
-            type.t |= VT_LLONG;
-        else
-            type.t |= VT_INT;
+        if (vtop->type.t & VT_PACKED) {
+            type.t |= VT_BYTE;
+	    r = VT_STRUCT;
+	}
+	else {
+            r = adjust_bf(vtop, bit_pos, bit_size);
+             if ((vtop->type.t & VT_BTYPE) == VT_LLONG)
+                type.t |= VT_LLONG;
+            else
+                type.t |= VT_INT;
+	}
 
         if (r == VT_STRUCT) {
             load_packed_bf(&type, bit_pos, bit_size);
@@ -2528,7 +2540,7 @@ static void gv_dup(void)
     t = vtop->type.t;
 #if PTR_SIZE == 4
     if ((t & VT_BTYPE) == VT_LLONG) {
-        if (t & VT_BITFIELD) {
+        if (t & (VT_BITFIELD | VT_PACKED)) {
             gv(RC_INT);
             t = vtop->type.t;
         }
@@ -3679,7 +3691,7 @@ static void gen_cast(CType *type)
         force_charshort_cast();
 
     /* bitfields first get cast to ints */
-    if (vtop->type.t & VT_BITFIELD)
+    if (vtop->type.t & (VT_BITFIELD | VT_PACKED))
         gv(RC_INT);
 
     dbt = type->t & (VT_BTYPE | VT_UNSIGNED);
@@ -4151,14 +4163,21 @@ ST_FUNC void vstore(void)
             gfunc_call(3);
         /* leave source on stack */
 
-    } else if (ft & VT_BITFIELD) {
+    } else if (ft & (VT_BITFIELD | VT_PACKED)) {
         /* bitfield store handling */
 
         /* save lvalue as expression result (example: s.b = s.a = n;) */
         vdup(), vtop[-1] = vtop[-2];
 
-        bit_pos = BIT_POS(ft);
-        bit_size = BIT_SIZE(ft);
+	if (ft & VT_BITFIELD) {
+            bit_pos = BIT_POS(ft);
+            bit_size = BIT_SIZE(ft);
+	}
+	else {
+            size = type_size(&vtop[-1].type, &align);
+            bit_pos = 0;
+            bit_size = size * 8;
+	}
         /* remove bit field info to avoid loops */
         vtop[-1].type.t = ft & ~VT_STRUCT_MASK;
 
@@ -4166,11 +4185,17 @@ ST_FUNC void vstore(void)
             gen_cast(&vtop[-1].type);
             vtop[-1].type.t = (vtop[-1].type.t & ~VT_BTYPE) | (VT_BYTE | VT_UNSIGNED);
         }
-        r = adjust_bf(vtop - 1, bit_pos, bit_size);
+        r = ft & VT_PACKED ? VT_STRUCT : adjust_bf(vtop - 1, bit_pos, bit_size);
         if (dbt != VT_BOOL) {
             gen_cast(&vtop[-1].type);
             dbt = vtop[-1].type.t & VT_BTYPE;
-        }
+	}
+
+	if (ft & VT_PACKED) {
+            vtop->type.t = vtop->type.t & ~VT_PACKED;
+            vtop[-1].type.t = (vtop[-1].type.t & ~(VT_BTYPE | VT_PACKED)) | VT_BYTE;
+	}
+
         if (r == VT_STRUCT) {
             store_packed_bf(bit_pos, bit_size);
         } else {
@@ -4603,6 +4628,10 @@ static void struct_layout(CType *type, AttributeDef *ad)
                 if (pcc && pragma_pack < a)
                     a = 0;
             }
+#ifdef TCC_TARGET_ARM
+	    if (packed)
+		f->type.t |= VT_PACKED;
+#endif
         }
         /* some individual align was specified */
         if (a)
diff --git a/tests/tests2/96_nodata_wanted.c b/tests/tests2/96_nodata_wanted.c
index 1a4cd33..4e95603 100644
--- a/tests/tests2/96_nodata_wanted.c
+++ b/tests/tests2/96_nodata_wanted.c
@@ -58,10 +58,7 @@ ts1:;
         void *p = (void*)&main;
         char cc[] = "static string";
         double d = 8.0;
-        struct
-#ifndef __arm__
-        __attribute__((packed))
-#endif
+        struct __attribute__((packed))
         {
             unsigned x : 12;
             unsigned char y : 7;