[PATCH] Fix i386 memmove issue [BZ #22644]

[Andrew Senkevich <andrew.n.senkevich@gmail.com>]

Hi,

in BZ #22644 was localized bug in i386 memmove-sse2-unaligned. This
patch adds fix and testcase. Any comments?

        * sysdeps/i386/i686/multiarch/memcpy-sse2-unaligned.S: Fixed
branch conditions.
        * string/test-memmove.c (do_test2): New testcase.

diff --git a/sysdeps/i386/i686/multiarch/memcpy-sse2-unaligned.S
b/sysdeps/i386/i686/multiarch/memcpy-sse2-unaligned.S
index 9c3bbe7..9aa17de 100644
--- a/sysdeps/i386/i686/multiarch/memcpy-sse2-unaligned.S
+++ b/sysdeps/i386/i686/multiarch/memcpy-sse2-unaligned.S
@@ -72,7 +72,7 @@ ENTRY (MEMCPY)
        cmp     %edx, %eax

 # ifdef USE_AS_MEMMOVE
-       jg      L(check_forward)
+       ja      L(check_forward)

 L(mm_len_0_or_more_backward):
 /* Now do checks for lengths. We do [0..16], [16..32], [32..64], [64..128]
@@ -81,7 +81,7 @@ L(mm_len_0_or_more_backward):
        jbe     L(mm_len_0_16_bytes_backward)

        cmpl    $32, %ecx
-       jg      L(mm_len_32_or_more_backward)
+       ja      L(mm_len_32_or_more_backward)

 /* Copy [0..32] and return.  */
        movdqu  (%eax), %xmm0
@@ -92,7 +92,7 @@ L(mm_len_0_or_more_backward):

 L(mm_len_32_or_more_backward):
        cmpl    $64, %ecx
-       jg      L(mm_len_64_or_more_backward)
+       ja      L(mm_len_64_or_more_backward)

 /* Copy [0..64] and return.  */
        movdqu  (%eax), %xmm0
@@ -107,7 +107,7 @@ L(mm_len_32_or_more_backward):

 L(mm_len_64_or_more_backward):
        cmpl    $128, %ecx
-       jg      L(mm_len_128_or_more_backward)
+       ja      L(mm_len_128_or_more_backward)

 /* Copy [0..128] and return.  */
        movdqu  (%eax), %xmm0
@@ -132,7 +132,7 @@ L(mm_len_128_or_more_backward):
        add     %ecx, %eax
        cmp     %edx, %eax
        movl    SRC(%esp), %eax
-       jle     L(forward)
+       jbe     L(forward)
        PUSH (%esi)
        PUSH (%edi)
        PUSH (%ebx)
@@ -269,7 +269,7 @@ L(check_forward):
        add     %edx, %ecx
        cmp     %eax, %ecx
        movl    LEN(%esp), %ecx
-       jle     L(forward)
+       jbe     L(forward)

 /* Now do checks for lengths. We do [0..16], [0..32], [0..64], [0..128]
        separately.  */

diff --git a/string/test-memmove.c b/string/test-memmove.c
index edc7a4c..8dc152b
--- a/string/test-memmove.c
+++ b/string/test-memmove.c
@@ -245,6 +245,49 @@ do_random_tests (void)
     }
 }

+#if __SIZEOF_POINTER__ == 4
+static void
+do_test2 (void)
+{
+  uint32_t i;
+  uint32_t num = 0x20000000;
+  uint32_t * large_buf = mmap (0, sizeof(uint32_t) * num, PROT_READ |
PROT_WRITE,
+        MAP_PRIVATE | MAP_ANON, -1, 0);
+  if (large_buf == MAP_FAILED)
+    error (EXIT_FAILURE, errno, "large mmap failed");
+
+  if (!((uint32_t)(large_buf) < (0x80000000 - 128) && (0x80000000 +
128) < (uint32_t)(&large_buf[num])))
+    {
+      error (0, 0,"allocated large memory doesn't cross 0x80000000 boundary");
+      ret = 1;
+      return;
+    }
+
+  int bytes_move = 0x80000000 - (uint32_t)large_buf;
+  int arr_size = bytes_move / sizeof(uint32_t);
+
+  FOR_EACH_IMPL (impl, 0)
+    {
+      for (i = 0; i < arr_size; i++)
+        large_buf[i] = i;
+
+      uint32_t * dst = &large_buf[33];
+
+      CALL (impl, (char *)dst, (char *)large_buf, bytes_move);
+
+      for (i = 0; i < arr_size; i++)
+        {
+          if (dst[i] != i)
+     {
+       error (0, 0, "Wrong result in function %s dst \"%p\" src
\"%p\" offset \"%d\"",
+      impl->name, dst, large_buf, i);
+       ret = 1;
+              return;
+     }
+ }
+}
+#endif
+
 int
 test_main (void)
 {
@@ -284,6 +327,11 @@ test_main (void)
     }

   do_random_tests ();
+
+#if __SIZEOF_POINTER__ == 4
+  do_test2 ();
+#endif
+
   return ret;
 }


--
WBR,
Andrew