diff --git a/qemu/target/arm/helper.c b/qemu/target/arm/helper.c
index eb3e0de6..7959e4f7 100644
--- a/qemu/target/arm/helper.c
+++ b/qemu/target/arm/helper.c
@@ -10585,9 +10585,9 @@ VFP_CONV_FIX_A64(uq, s, 32, 64, uint64)
 /* Set the current fp rounding mode and return the old one.
  * The argument is a softfloat float_round_ value.
  */
-uint32_t HELPER(set_rmode)(uint32_t rmode, CPUARMState *env)
+uint32_t HELPER(set_rmode)(uint32_t rmode, void *fpstp)
 {
-    float_status *fp_status = &env->vfp.fp_status;
+    float_status *fp_status = fpstp;
 
     uint32_t prev_rmode = get_float_rounding_mode(fp_status);
     set_float_rounding_mode(rmode, fp_status);
diff --git a/qemu/target/arm/helper.h b/qemu/target/arm/helper.h
index b0bfc092..a9823b36 100644
--- a/qemu/target/arm/helper.h
+++ b/qemu/target/arm/helper.h
@@ -169,7 +169,7 @@ DEF_HELPER_3(vfp_uhtod, f64, i64, i32, ptr)
 DEF_HELPER_3(vfp_ultod, f64, i64, i32, ptr)
 DEF_HELPER_3(vfp_uqtod, f64, i64, i32, ptr)
 
-DEF_HELPER_FLAGS_2(set_rmode, TCG_CALL_NO_RWG, i32, i32, env)
+DEF_HELPER_FLAGS_2(set_rmode, TCG_CALL_NO_RWG, i32, i32, ptr)
 DEF_HELPER_FLAGS_2(set_neon_rmode, TCG_CALL_NO_RWG, i32, i32, env)
 
 DEF_HELPER_2(vfp_fcvt_f16_to_f32, f32, i32, env)
diff --git a/qemu/target/arm/translate-a64.c b/qemu/target/arm/translate-a64.c
index fe2a7606..2a71f273 100644
--- a/qemu/target/arm/translate-a64.c
+++ b/qemu/target/arm/translate-a64.c
@@ -4723,10 +4723,10 @@ static void handle_fp_1src_single(DisasContext *s, int opcode, int rd, int rn)
     {
         TCGv_i32 tcg_rmode = tcg_const_i32(tcg_ctx, arm_rmode_to_sf(opcode & 7));
 
-        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, fpst);
         gen_helper_rints(tcg_ctx, tcg_res, tcg_op, fpst);
 
-        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, fpst);
         tcg_temp_free_i32(tcg_ctx, tcg_rmode);
         break;
     }
@@ -4783,10 +4783,10 @@ static void handle_fp_1src_double(DisasContext *s, int opcode, int rd, int rn)
     {
         TCGv_i32 tcg_rmode = tcg_const_i32(tcg_ctx, arm_rmode_to_sf(opcode & 7));
 
-        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, fpst);
         gen_helper_rintd(tcg_ctx, tcg_res, tcg_op, fpst);
 
-        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, fpst);
         tcg_temp_free_i32(tcg_ctx, tcg_rmode);
         break;
     }
@@ -5320,7 +5320,7 @@ static void handle_fpfpcvt(DisasContext *s, int rd, int rn, int opcode,
 
         tcg_rmode = tcg_const_i32(tcg_ctx, arm_rmode_to_sf(rmode));
 
-        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_fpstatus);
 
         if (is_double) {
             TCGv_i64 tcg_double = read_fp_dreg(s, rn);
@@ -5367,7 +5367,7 @@ static void handle_fpfpcvt(DisasContext *s, int rd, int rn, int opcode,
             tcg_temp_free_i32(tcg_ctx, tcg_single);
         }
 
-        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_fpstatus);
         tcg_temp_free_i32(tcg_ctx, tcg_rmode);
 
         if (!sf) {
@@ -7109,8 +7109,8 @@ static void handle_simd_shift_fpint_conv(DisasContext *s, bool is_scalar,
     assert(!(is_scalar && is_q));
 
     tcg_rmode = tcg_const_i32(tcg_ctx, arm_rmode_to_sf(FPROUNDING_ZERO));
-    gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
     tcg_fpstatus = get_fpstatus_ptr(tcg_ctx, false);
+    gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_fpstatus);
     tcg_shift = tcg_const_i32(tcg_ctx, fracbits);
 
     if (is_double) {
@@ -7154,7 +7154,7 @@ static void handle_simd_shift_fpint_conv(DisasContext *s, bool is_scalar,
 
     tcg_temp_free_ptr(tcg_ctx, tcg_fpstatus);
     tcg_temp_free_i32(tcg_ctx, tcg_shift);
-    gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+    gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_fpstatus);
     tcg_temp_free_i32(tcg_ctx, tcg_rmode);
 }
 
@@ -8438,8 +8438,8 @@ static void disas_simd_scalar_two_reg_misc(DisasContext *s, uint32_t insn)
 
     if (is_fcvt) {
         tcg_rmode = tcg_const_i32(tcg_ctx, arm_rmode_to_sf(rmode));
-        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
         tcg_fpstatus = get_fpstatus_ptr(tcg_ctx, false);
+        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_fpstatus);
     } else {
         tcg_rmode = NULL;
         tcg_fpstatus = NULL;
@@ -8504,7 +8504,7 @@ static void disas_simd_scalar_two_reg_misc(DisasContext *s, uint32_t insn)
     }
 
     if (is_fcvt) {
-        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_fpstatus);
         tcg_temp_free_i32(tcg_ctx, tcg_rmode);
         tcg_temp_free_ptr(tcg_ctx, tcg_fpstatus);
     }
@@ -10835,14 +10835,14 @@ static void disas_simd_two_reg_misc(DisasContext *s, uint32_t insn)
         return;
     }
 
-    if (need_fpstatus) {
+    if (need_fpstatus || need_rmode) {
         tcg_fpstatus = get_fpstatus_ptr(tcg_ctx, false);
     } else {
         tcg_fpstatus = NULL;
     }
     if (need_rmode) {
         tcg_rmode = tcg_const_i32(tcg_ctx, arm_rmode_to_sf(rmode));
-        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_fpstatus);
     } else {
         tcg_rmode = NULL;
     }
@@ -11084,7 +11084,7 @@ static void disas_simd_two_reg_misc(DisasContext *s, uint32_t insn)
     clear_vec_high(s, is_q, rd);
 
     if (need_rmode) {
-        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_fpstatus);
         tcg_temp_free_i32(tcg_ctx, tcg_rmode);
     }
     if (need_fpstatus) {
diff --git a/qemu/target/arm/translate.c b/qemu/target/arm/translate.c
index 8749019e..0681840d 100644
--- a/qemu/target/arm/translate.c
+++ b/qemu/target/arm/translate.c
@@ -3250,7 +3250,7 @@ static int handle_vrint(DisasContext *s, uint32_t insn, uint32_t rd, uint32_t rm
     TCGv_i32 tcg_rmode;
 
     tcg_rmode = tcg_const_i32(tcg_ctx, arm_rmode_to_sf(rounding));
-    gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+    gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, fpst);
 
     if (dp) {
         TCGv_i64 tcg_op;
@@ -3274,7 +3274,7 @@ static int handle_vrint(DisasContext *s, uint32_t insn, uint32_t rd, uint32_t rm
         tcg_temp_free_i32(tcg_ctx, tcg_res);
     }
 
-    gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+    gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, fpst);
     tcg_temp_free_i32(tcg_ctx, tcg_rmode);
 
     tcg_temp_free_ptr(tcg_ctx, fpst);
@@ -3292,7 +3292,7 @@ static int handle_vcvt(DisasContext *s, uint32_t insn, uint32_t rd, uint32_t rm,
     tcg_shift = tcg_const_i32(tcg_ctx, 0);
 
     tcg_rmode = tcg_const_i32(tcg_ctx, arm_rmode_to_sf(rounding));
-    gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+    gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, fpst);
 
     if (dp) {
         TCGv_i64 tcg_double, tcg_res;
@@ -3330,7 +3330,7 @@ static int handle_vcvt(DisasContext *s, uint32_t insn, uint32_t rd, uint32_t rm,
         tcg_temp_free_i32(tcg_ctx, tcg_single);
     }
 
-    gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+    gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, fpst);
     tcg_temp_free_i32(tcg_ctx, tcg_rmode);
 
     tcg_temp_free_i32(tcg_ctx, tcg_shift);
@@ -4001,13 +4001,13 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                         TCGv_ptr fpst = get_fpstatus_ptr(s, 0);
                         TCGv_i32 tcg_rmode;
                         tcg_rmode = tcg_const_i32(tcg_ctx, float_round_to_zero);
-                        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+                        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, fpst);
                         if (dp) {
                             gen_helper_rintd(tcg_ctx, tcg_ctx->cpu_F0d, tcg_ctx->cpu_F0d, fpst);
                         } else {
                             gen_helper_rints(tcg_ctx, tcg_ctx->cpu_F0s, tcg_ctx->cpu_F0s, fpst);
                         }
-                        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, tcg_ctx->cpu_env);
+                        gen_helper_set_rmode(tcg_ctx, tcg_rmode, tcg_rmode, fpst);
                         tcg_temp_free_i32(tcg_ctx, tcg_rmode);
                         tcg_temp_free_ptr(tcg_ctx, fpst);
                         break;