diff --git a/qemu/target-arm/cpu.h b/qemu/target-arm/cpu.h
index d00846d3..443f0257 100644
--- a/qemu/target-arm/cpu.h
+++ b/qemu/target-arm/cpu.h
@@ -1961,6 +1961,8 @@ static inline bool arm_singlestep_active(CPUARMState *env)
  */
 #define ARM_TBFLAG_NS_SHIFT         19
 #define ARM_TBFLAG_NS_MASK          (1 << ARM_TBFLAG_NS_SHIFT)
+#define ARM_TBFLAG_BE_DATA_SHIFT    20
+#define ARM_TBFLAG_BE_DATA_MASK     (1 << ARM_TBFLAG_BE_DATA_SHIFT)
 
 /* Bit usage when in AArch64 state: currently we have no A64 specific bits */
 
@@ -1991,6 +1993,8 @@ static inline bool arm_singlestep_active(CPUARMState *env)
     (((F) & ARM_TBFLAG_XSCALE_CPAR_MASK) >> ARM_TBFLAG_XSCALE_CPAR_SHIFT)
 #define ARM_TBFLAG_NS(F) \
     (((F) & ARM_TBFLAG_NS_MASK) >> ARM_TBFLAG_NS_SHIFT)
+#define ARM_TBFLAG_BE_DATA(F) \
+    (((F) & ARM_TBFLAG_BE_DATA_MASK) >> ARM_TBFLAG_BE_DATA_SHIFT)
 
 static inline bool bswap_code(bool sctlr_b)
 {
@@ -2091,6 +2095,40 @@ static inline bool arm_sctlr_b(CPUARMState *env)
         (env->cp15.sctlr_el[1] & SCTLR_B) != 0;
 }
 
+/* Return true if the processor is in big-endian mode. */
+static inline bool arm_cpu_data_is_big_endian(CPUARMState *env)
+{
+    int cur_el;
+
+    /* In 32bit endianness is determined by looking at CPSR's E bit */
+    if (!is_a64(env)) {
+        return
+#ifdef CONFIG_USER_ONLY
+            /* In system mode, BE32 is modelled in line with the
+             * architecture (as word-invariant big-endianness), where loads
+             * and stores are done little endian but from addresses which
+             * are adjusted by XORing with the appropriate constant. So the
+             * endianness to use for the raw data access is not affected by
+             * SCTLR.B.
+             * In user mode, however, we model BE32 as byte-invariant
+             * big-endianness (because user-only code cannot tell the
+             * difference), and so we need to use a data access endianness
+             * that depends on SCTLR.B.
+             */
+            arm_sctlr_b(env) ||
+#endif
+                ((env->uncached_cpsr & CPSR_E) ? 1 : 0);
+    }
+
+    cur_el = arm_current_el(env);
+
+    if (cur_el == 0) {
+        return (env->cp15.sctlr_el[1] & SCTLR_E0E) != 0;
+    }
+
+    return (env->cp15.sctlr_el[cur_el] & SCTLR_EE) != 0;
+}
+
 static inline void cpu_get_tb_cpu_state(CPUARMState *env, target_ulong *pc,
                                         target_ulong *cs_base, int *flags)
 {
@@ -2135,6 +2173,9 @@ static inline void cpu_get_tb_cpu_state(CPUARMState *env, target_ulong *pc,
             }
         }
     }
+    if (arm_cpu_data_is_big_endian(env)) {
+        *flags |= ARM_TBFLAG_BE_DATA_MASK;
+    }
     *flags |= fp_exception_el(env) << ARM_TBFLAG_FPEXC_EL_SHIFT;
 
     *cs_base = 0;
@@ -2167,40 +2208,6 @@ static inline uint64_t *aa64_vfp_qreg(CPUARMState *env, unsigned regno)
     return &env->vfp.regs[2 * regno];
 }
 
-/* Return true if the processor is in big-endian mode. */
-static inline bool arm_cpu_data_is_big_endian(CPUARMState *env)
-{
-    int cur_el;
-
-    /* In 32bit endianness is determined by looking at CPSR's E bit */
-    if (!is_a64(env)) {
-        return
-#ifdef CONFIG_USER_ONLY
-            /* In system mode, BE32 is modelled in line with the
-             * architecture (as word-invariant big-endianness), where loads
-             * and stores are done little endian but from addresses which
-             * are adjusted by XORing with the appropriate constant. So the
-             * endianness to use for the raw data access is not affected by
-             * SCTLR.B.
-             * In user mode, however, we model BE32 as byte-invariant
-             * big-endianness (because user-only code cannot tell the
-             * difference), and so we need to use a data access endianness
-             * that depends on SCTLR.B.
-             */
-            arm_sctlr_b(env) ||
-#endif
-                ((env->uncached_cpsr & CPSR_E) ? 1 : 0);
-    }
-
-    cur_el = arm_current_el(env);
-
-    if (cur_el == 0) {
-        return (env->cp15.sctlr_el[1] & SCTLR_E0E) != 0;
-    }
-
-    return (env->cp15.sctlr_el[cur_el] & SCTLR_EE) != 0;
-}
-
 #include "exec/exec-all.h"
 
 enum {
diff --git a/qemu/target-arm/translate-a64.c b/qemu/target-arm/translate-a64.c
index cf117749..d9f6aeb7 100644
--- a/qemu/target-arm/translate-a64.c
+++ b/qemu/target-arm/translate-a64.c
@@ -11261,7 +11261,7 @@ void gen_intermediate_code_a64(ARMCPU *cpu, TranslationBlock *tb)
                                !arm_el_is_aa64(env, 3);
     dc->thumb = 0;
     dc->sctlr_b = 0;
-    dc->be_data = MO_TE;
+    dc->be_data = ARM_TBFLAG_BE_DATA(tb->flags) ? MO_BE : MO_LE;
     dc->condexec_mask = 0;
     dc->condexec_cond = 0;
     dc->mmu_idx = ARM_TBFLAG_MMUIDX(tb->flags);
diff --git a/qemu/target-arm/translate.c b/qemu/target-arm/translate.c
index 3c10d8d2..e66ffa16 100644
--- a/qemu/target-arm/translate.c
+++ b/qemu/target-arm/translate.c
@@ -11490,7 +11490,7 @@ void gen_intermediate_code(CPUARMState *env, TranslationBlock *tb)
                                !arm_el_is_aa64(env, 3);
     dc->thumb = ARM_TBFLAG_THUMB(tb->flags);    // qq
     dc->sctlr_b = ARM_TBFLAG_SCTLR_B(tb->flags);
-    dc->be_data = MO_TE;
+    dc->be_data = ARM_TBFLAG_BE_DATA(tb->flags) ? MO_BE : MO_LE;
     dc->condexec_mask = (ARM_TBFLAG_CONDEXEC(tb->flags) & 0xf) << 1;
     dc->condexec_cond = ARM_TBFLAG_CONDEXEC(tb->flags) >> 4;
     dc->mmu_idx = ARM_TBFLAG_MMUIDX(tb->flags);