diff --git a/qemu/accel/tcg/translate-all.c b/qemu/accel/tcg/translate-all.c
index 17a46d45..9c862be0 100644
--- a/qemu/accel/tcg/translate-all.c
+++ b/qemu/accel/tcg/translate-all.c
@@ -1335,6 +1335,8 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
     tb->cs_base = cs_base;
     tb->flags = flags;
     tb->cflags = cflags;
+    tb->trace_vcpu_dstate = 0;
+    tcg_ctx->tb_cflags = cflags;
 
 #ifdef CONFIG_PROFILER
     tcg_ctx->tb_count1++; /* includes aborted translations because of
diff --git a/qemu/tcg/tcg-op.c b/qemu/tcg/tcg-op.c
index b5321acb..8f4aa44f 100644
--- a/qemu/tcg/tcg-op.c
+++ b/qemu/tcg/tcg-op.c
@@ -97,7 +97,7 @@ void tcg_gen_op6(TCGContext *ctx, TCGOpcode opc, TCGArg a1, TCGArg a2,
 
 void tcg_gen_mb(TCGContext *ctx, TCGBar mb_type)
 {
-    if (ctx->uc->parallel_cpus) {
+    if (ctx->tb_cflags & CF_PARALLEL) {
         tcg_gen_op1(ctx, INDEX_op_mb, mb_type);
     }
 }
@@ -2819,7 +2819,7 @@ void tcg_gen_atomic_cmpxchg_i32(TCGContext *s,
 {
     memop = tcg_canonicalize_memop(memop, 0, 0);
 
-    if (!s->uc->parallel_cpus) {
+    if (!(s->tb_cflags & CF_PARALLEL)) {
         TCGv_i32 t1 = tcg_temp_new_i32(s);
         TCGv_i32 t2 = tcg_temp_new_i32(s);
 
@@ -2864,7 +2864,7 @@ void tcg_gen_atomic_cmpxchg_i64(TCGContext *s,
 {
     memop = tcg_canonicalize_memop(memop, 1, 0);
 
-    if (!s->uc->parallel_cpus) {
+    if (!(s->tb_cflags & CF_PARALLEL)) {
         TCGv_i64 t1 = tcg_temp_new_i64(s);
         TCGv_i64 t2 = tcg_temp_new_i64(s);
 
@@ -3071,7 +3071,7 @@ GEN_ATOMIC_TABLE(NAME)                                                  \
 void tcg_gen_atomic_##NAME##_i32                                        \
     (TCGContext *s, TCGv_i32 ret, TCGv addr, TCGv_i32 val, TCGArg idx, TCGMemOp memop) \
 {                                                                       \
-    if (s->uc->parallel_cpus) {                                         \
+    if (s->tb_cflags & CF_PARALLEL) {                                   \
         do_atomic_op_i32(s, ret, addr, val, idx, memop, table_##NAME);  \
     } else {                                                            \
         do_nonatomic_op_i32(s, ret, addr, val, idx, memop, NEW,         \
@@ -3081,7 +3081,7 @@ void tcg_gen_atomic_##NAME##_i32                                        \
 void tcg_gen_atomic_##NAME##_i64                                        \
     (TCGContext *s, TCGv_i64 ret, TCGv addr, TCGv_i64 val, TCGArg idx, TCGMemOp memop) \
 {                                                                       \
-    if (s->uc->parallel_cpus) {                                         \
+    if (s->tb_cflags & CF_PARALLEL) {                                   \
         do_atomic_op_i64(s, ret, addr, val, idx, memop, table_##NAME);  \
     } else {                                                            \
         do_nonatomic_op_i64(s, ret, addr, val, idx, memop, NEW,         \
diff --git a/qemu/tcg/tcg.h b/qemu/tcg/tcg.h
index 3a796773..bc5e03de 100644
--- a/qemu/tcg/tcg.h
+++ b/qemu/tcg/tcg.h
@@ -731,6 +731,7 @@ struct TCGContext {
     uintptr_t *tb_jmp_target_addr; /* tb->jmp_target_arg if !direct_jump */
 
     TCGRegSet reserved_regs;
+    uint32_t tb_cflags; /* cflags of the current TB */
     intptr_t current_frame_offset;
     intptr_t frame_start;
     intptr_t frame_end;