kern/misc: Implement faster grub_memcpy() for aligned buffers

When both "dest" and "src" are aligned, copying the data in grub_addr_t
sized chunks is more efficient than a byte-by-byte copy.

Also tweak __aeabi_memcpy(), __aeabi_memcpy4(), and __aeabi_memcpy8(),
since grub_memcpy() is not inline anymore.

Optimization for unaligned buffers was omitted to maintain code
simplicity and readability. The current chunk-copy optimization
for aligned buffers already provides a noticeable performance
improvement (*) for Argon2 keyslot decryption.

  (*) On my system, for a LUKS2 keyslot configured with a 1 GB Argon2
      memory requirement, this patch reduces the decryption time from
      22 seconds to 12 seconds.

Signed-off-by: Gary Lin <glin@suse.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>

grub-core/boot/decompressor/minilib.c

@@ -50,6 +50,12 @@ grub_memmove (void *dest, const void *src, grub_size_t n)
   return dest;
 }
 
+void *
+grub_memcpy (void *dest, const void *src, grub_size_t n)
+{
+  return grub_memmove (dest, src, n);
+}
+
 int
 grub_memcmp (const void *s1, const void *s2, grub_size_t n)
 {

grub-core/kern/compiler-rt.c

@@ -24,7 +24,7 @@
 void * GRUB_BUILTIN_ATTR
 memcpy (void *dest, const void *src, grub_size_t n)
 {
-	return grub_memmove (dest, src, n);
+	return grub_memcpy (dest, src, n);
 }
 void * GRUB_BUILTIN_ATTR
 memmove (void *dest, const void *src, grub_size_t n)
@@ -372,11 +372,11 @@ grub_int32_t
 __aeabi_idiv (grub_int32_t a, grub_int32_t b)
   __attribute__ ((alias ("__divsi3")));
 void *__aeabi_memcpy (void *dest, const void *src, grub_size_t n)
-  __attribute__ ((alias ("grub_memcpy")));
+  __attribute__ ((alias ("memcpy")));
 void *__aeabi_memcpy4 (void *dest, const void *src, grub_size_t n)
-  __attribute__ ((alias ("grub_memcpy")));
+  __attribute__ ((alias ("memcpy")));
 void *__aeabi_memcpy8 (void *dest, const void *src, grub_size_t n)
-  __attribute__ ((alias ("grub_memcpy")));
+  __attribute__ ((alias ("memcpy")));
 void *__aeabi_memset (void *s, int c, grub_size_t n)
   __attribute__ ((alias ("memset")));
 

grub-core/kern/misc.c

@@ -99,6 +99,37 @@ grub_memmove (void *dest, const void *src, grub_size_t n)
   return dest;
 }
 
+static void *
+__memcpy_aligned (void *dest, const void *src, grub_size_t n)
+{
+  grub_addr_t *dw = (grub_addr_t *) dest;
+  const grub_addr_t *sw = (const grub_addr_t *) src;
+  grub_uint8_t *d;
+  const grub_uint8_t *s;
+
+  for (; n >= sizeof (grub_addr_t); n -= sizeof (grub_addr_t))
+    *dw++ = *sw++;
+
+  d = (grub_uint8_t *) dw;
+  s = (const grub_uint8_t *) sw;
+  for (; n > 0; n--)
+    *d++ = *s++;
+
+  return dest;
+}
+
+void *
+grub_memcpy (void *dest, const void *src, grub_size_t n)
+{
+  /* Check if dest and src are aligned and n >= sizeof(grub_addr_t). */
+  if (((grub_addr_t) dest & (sizeof (grub_addr_t) - 1)) == 0 &&
+      ((grub_addr_t) src & (sizeof (grub_addr_t) - 1)) == 0 &&
+      n >= sizeof (grub_addr_t))
+    return __memcpy_aligned (dest, src, n);
+
+  return grub_memmove (dest, src, n);
+}
+
 char *
 grub_strcpy (char *dest, const char *src)
 {

include/grub/misc.h

@@ -38,6 +38,7 @@
 #define grub_dprintf(condition, ...) grub_real_dprintf(GRUB_FILE, __FUNCTION__, __LINE__, condition, __VA_ARGS__)
 
 void *EXPORT_FUNC(grub_memmove) (void *dest, const void *src, grub_size_t n);
+void *EXPORT_FUNC(grub_memcpy) (void *dest, const void *src, grub_size_t n);
 char *EXPORT_FUNC(grub_strcpy) (char *dest, const char *src);
 
 static inline char *
@@ -103,13 +104,6 @@ grub_strlcpy (char *dest, const char *src, grub_size_t size)
   return res;
 }
 
-/* XXX: If grub_memmove is too slow, we must implement grub_memcpy.  */
-static inline void *
-grub_memcpy (void *dest, const void *src, grub_size_t n)
-{
-  return grub_memmove (dest, src, n);
-}
-
 #if defined(__x86_64__) && !defined (GRUB_UTIL)
 #if defined (__MINGW32__) || defined (__CYGWIN__) || defined (__MINGW64__)
 #define GRUB_ASM_ATTR __attribute__ ((sysv_abi))