diff --git a/core/arch/aarchxx/mangle.c b/core/arch/aarchxx/mangle.c
index b60bf2907ad..a2e7a92f700 100644
--- a/core/arch/aarchxx/mangle.c
+++ b/core/arch/aarchxx/mangle.c
@@ -603,6 +603,9 @@ insert_push_all_registers(dcontext_t *dcontext, clean_call_info_t *cci,
                                   DR_REG_LIST_LENGTH_ARM, DR_REG_LIST_ARM));
     }
     dstack_offs += 15 * XSP_SZ;
+
+    /* Make dstack_offs 8-byte algined, as we only accounted for 17 4-byte slots. */
+    dstack_offs += XSP_SZ;
     ASSERT(cci->skip_save_flags    ||
            cci->num_simd_skip != 0 ||
            cci->num_regs_skip != 0 ||
diff --git a/core/arch/emit_utils_shared.c b/core/arch/emit_utils_shared.c
index 38a76aaf5de..880aadd6fdc 100644
--- a/core/arch/emit_utils_shared.c
+++ b/core/arch/emit_utils_shared.c
@@ -5011,10 +5011,10 @@ emit_new_thread_dynamo_start(dcontext_t *dcontext, byte *pc)
                                        SCRATCH_REG0 _IF_AARCH64(false));
 # ifndef AARCH64
     /* put pre-push xsp into priv_mcontext_t.xsp slot */
-    ASSERT(offset == sizeof(priv_mcontext_t));
+    ASSERT(offset == get_clean_call_switch_stack_size());
     APP(&ilist, XINST_CREATE_add_2src
         (dcontext, opnd_create_reg(SCRATCH_REG0),
-         opnd_create_reg(REG_XSP), OPND_CREATE_INT32(sizeof(priv_mcontext_t))));
+         opnd_create_reg(REG_XSP), OPND_CREATE_INT32(offset)));
     APP(&ilist, XINST_CREATE_store
         (dcontext, OPND_CREATE_MEMPTR(REG_XSP, offsetof(priv_mcontext_t, xsp)),
          opnd_create_reg(SCRATCH_REG0)));
diff --git a/core/arch/mangle_shared.c b/core/arch/mangle_shared.c
index b11dd64c7f8..26debfbdfbc 100644
--- a/core/arch/mangle_shared.c
+++ b/core/arch/mangle_shared.c
@@ -297,7 +297,8 @@ prepare_for_clean_call(dcontext_t *dcontext, clean_call_info_t *cci,
     ASSERT(cci->skip_save_flags    ||
            cci->num_simd_skip != 0 ||
            cci->num_regs_skip != 0 ||
-           dstack_offs == sizeof(priv_mcontext_t) + clean_call_beyond_mcontext());
+           (int) dstack_offs == (get_clean_call_switch_stack_size() +
+                                  clean_call_beyond_mcontext()));
     return dstack_offs;
 }