Fix Armv8.1-M x4 xor_bytes alignment fallback

Fixes #1759.

Signed-off-by: Brendan Moran <brendan.moran@arm.com>

Author: bremoran

dev/fips202/armv81m/mve.h

@@ -49,6 +49,30 @@ static MLK_INLINE int mlk_keccakf1600_xor_bytes_x4_native(
     const uint8_t *data2, const uint8_t *data3, unsigned offset,
     unsigned length)
 {
+  unsigned offset_mod8 = offset & 7u;
+  unsigned byte_prefix = 0;
+  unsigned main_length;
+
+  if (offset_mod8 != 0u)
+  {
+    byte_prefix = 8u - offset_mod8;
+    if (byte_prefix > length)
+    {
+      byte_prefix = length;
+    }
+  }
+
+  main_length = length - byte_prefix;
+  if (main_length >= 8u &&
+      ((((uintptr_t)data0 + byte_prefix) |
+        ((uintptr_t)data1 + byte_prefix) |
+        ((uintptr_t)data2 + byte_prefix) |
+        ((uintptr_t)data3 + byte_prefix)) &
+       3u) != 0u)
+  {
+    return MLK_NATIVE_FUNC_FALLBACK;
+  }
+
   mlk_keccak_f1600_x4_state_xor_bytes(state, data0, data1, data2, data3, offset,
                                       length);
   return MLK_NATIVE_FUNC_SUCCESS;

mlkem/src/fips202/native/armv81m/mve.h

@@ -49,6 +49,30 @@ static MLK_INLINE int mlk_keccakf1600_xor_bytes_x4_native(
     const uint8_t *data2, const uint8_t *data3, unsigned offset,
     unsigned length)
 {
+  unsigned offset_mod8 = offset & 7u;
+  unsigned byte_prefix = 0;
+  unsigned main_length;
+
+  if (offset_mod8 != 0u)
+  {
+    byte_prefix = 8u - offset_mod8;
+    if (byte_prefix > length)
+    {
+      byte_prefix = length;
+    }
+  }
+
+  main_length = length - byte_prefix;
+  if (main_length >= 8u &&
+      ((((uintptr_t)data0 + byte_prefix) |
+        ((uintptr_t)data1 + byte_prefix) |
+        ((uintptr_t)data2 + byte_prefix) |
+        ((uintptr_t)data3 + byte_prefix)) &
+       3u) != 0u)
+  {
+    return MLK_NATIVE_FUNC_FALLBACK;
+  }
+
   mlk_keccak_f1600_x4_state_xor_bytes(state, data0, data1, data2, data3, offset,
                                       length);
   return MLK_NATIVE_FUNC_SUCCESS;

test/baremetal/platform/m33-an524/exec_wrapper.py

@@ -45,11 +45,17 @@ def err(msg, **kwargs):
     result = subprocess.run(qemu_cmd, encoding="utf-8", capture_output=True)
 finally:
     os.unlink(args_file)
+
+output = result.stdout + result.stderr
 if result.returncode != 0:
     err("FAIL!")
     err(f"{qemu_cmd} failed with error code {result.returncode}")
     err(result.stderr)
     exit(1)
 
+if "ERROR" in output or "FAIL:" in output or "FAILED" in output:
+    err(output, end="")
+    exit(1)
+
 for line in result.stdout.splitlines():
     print(line)

test/baremetal/platform/m55-an547/exec_wrapper.py

@@ -38,11 +38,17 @@ def err(msg, **kwargs):
     result = subprocess.run(qemu_cmd, encoding="utf-8", capture_output=True)
 finally:
     os.unlink(args_file)
+
+output = result.stdout + result.stderr
 if result.returncode != 0:
     err("FAIL!")
     err(f"{qemu_cmd} failed with error code {result.returncode}")
     err(result.stderr)
     exit(1)
 
+if "ERROR" in output or "FAIL:" in output or "FAILED" in output:
+    err(output, end="")
+    exit(1)
+
 for line in result.stdout.splitlines():
     print(line)