[PATCH v4 22/30] target/arm: Enforce alignment for VLDn/VSTn (multiple)

[Richard Henderson]

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/translate-neon.c.inc | 27 ++++++++++++++++++++++-----
 1 file changed, 22 insertions(+), 5 deletions(-)

diff --git a/target/arm/translate-neon.c.inc b/target/arm/translate-neon.c.inc
index 9c2b076027..e706c37c80 100644
--- a/target/arm/translate-neon.c.inc
+++ b/target/arm/translate-neon.c.inc
@@ -429,7 +429,7 @@ static bool trans_VLDST_multiple(DisasContext *s, 
arg_VLDST_multiple *a)
 {
     /* Neon load/store multiple structures */
     int nregs, interleave, spacing, reg, n;
-    MemOp endian = s->be_data;
+    MemOp mop, align, endian;
     int mmu_idx = get_mem_index(s);
     int size = a->size;
     TCGv_i64 tmp64;
@@ -473,20 +473,36 @@ static bool trans_VLDST_multiple(DisasContext *s, 
arg_VLDST_multiple *a)
     }
 
     /* For our purposes, bytes are always little-endian.  */
+    endian = s->be_data;
     if (size == 0) {
         endian = MO_LE;
     }
+
+    /* Enforce alignment requested by the instruction */
+    if (a->align) {
+        align = pow2_align(a->align + 2); /* 4 ** a->align */
+    } else {
+        align = s->align_mem ? MO_ALIGN : 0;
+    }
+
     /*
      * Consecutive little-endian elements from a single register
      * can be promoted to a larger little-endian operation.
      */
     if (interleave == 1 && endian == MO_LE) {
+        /* Retain any natural alignment. */
+        if (align == MO_ALIGN) {
+            align = pow2_align(size);
+        }
         size = 3;
     }
+
     tmp64 = tcg_temp_new_i64();
     addr = tcg_temp_new_i32();
     tmp = tcg_const_i32(1 << size);
     load_reg_var(s, addr, a->rn);
+
+    mop = endian | size | align;
     for (reg = 0; reg < nregs; reg++) {
         for (n = 0; n < 8 >> size; n++) {
             int xs;
@@ -494,15 +510,16 @@ static bool trans_VLDST_multiple(DisasContext *s, 
arg_VLDST_multiple *a)
                 int tt = a->vd + reg + spacing * xs;
 
                 if (a->l) {
-                    gen_aa32_ld_internal_i64(s, tmp64, addr, mmu_idx,
-                                             endian | size);
+                    gen_aa32_ld_internal_i64(s, tmp64, addr, mmu_idx, mop);
                     neon_store_element64(tt, n, size, tmp64);
                 } else {
                     neon_load_element64(tmp64, tt, n, size);
-                    gen_aa32_st_internal_i64(s, tmp64, addr, mmu_idx,
-                                             endian | size);
+                    gen_aa32_st_internal_i64(s, tmp64, addr, mmu_idx, mop);
                 }
                 tcg_gen_add_i32(addr, addr, tmp);
+
+                /* Subsequent memory operations inherit alignment */
+                mop &= ~MO_AMASK;
             }
         }
     }
-- 
2.25.1