From 44c6cbf6323ef6768fe2cbfded3e29a164db89f4 Mon Sep 17 00:00:00 2001
From: Goetz Lindenmaier <goetz@openjdk.org>
Date: Mon, 12 Oct 2026 06:48:09 +0000
Subject: [PATCH 1/3] 8393195: Update Libpng to 1.6.59

Backport-of: 51d43c00d204e36ae6548bc085cdea23d970d23b
---
 src/java.desktop/share/legal/libpng.md        |  4 ++-
 .../native/libsplashscreen/libpng/CHANGES     | 13 +++++++
 .../native/libsplashscreen/libpng/README      |  2 +-
 .../share/native/libsplashscreen/libpng/png.c |  4 +--
 .../share/native/libsplashscreen/libpng/png.h | 14 ++++----
 .../native/libsplashscreen/libpng/pngconf.h   |  2 +-
 .../libsplashscreen/libpng/pnglibconf.h       |  2 +-
 .../native/libsplashscreen/libpng/pngread.c   |  6 ++--
 .../native/libsplashscreen/libpng/pngrtran.c  |  1 +
 .../native/libsplashscreen/libpng/pngrutil.c  | 36 +++++++++++--------
 .../native/libsplashscreen/libpng/pngset.c    |  2 ++
 11 files changed, 56 insertions(+), 30 deletions(-)

diff --git a/src/java.desktop/share/legal/libpng.md b/src/java.desktop/share/legal/libpng.md
index 4c17a52a7ef2..aa0ecb68b3be 100644
--- a/src/java.desktop/share/legal/libpng.md
+++ b/src/java.desktop/share/legal/libpng.md
@@ -1,4 +1,4 @@
-## libpng v1.6.58
+## libpng v1.6.59
 
 ### libpng License
 <pre>
@@ -160,6 +160,7 @@ Authors, for copyright and licensing purposes.
  * Adam Richter
  * Alexander Smorkalov
  * Andreas Dilger
+ * Anthony Hurtado
  * Chris Blume
  * Cosmin Truta
  * Dave Martindale
@@ -197,6 +198,7 @@ Authors, for copyright and licensing purposes.
  * Vadim Barkov
  * Willem van Schaik
  * Yuelin Wang (王跃林)
+ * Yuki Sekiguchi
  * Zhijie Liang
  * Apple Inc.
     - Zixu Wang (王子旭)
diff --git a/src/java.desktop/share/native/libsplashscreen/libpng/CHANGES b/src/java.desktop/share/native/libsplashscreen/libpng/CHANGES
index 208ca6eaf17d..cd637fb98803 100644
--- a/src/java.desktop/share/native/libsplashscreen/libpng/CHANGES
+++ b/src/java.desktop/share/native/libsplashscreen/libpng/CHANGES
@@ -6385,6 +6385,19 @@ Version 1.6.58 [April 15, 2026]
     in-place.
     (Reported by ralfjunker <ralfjunker@users.noreply.github.com>.)
 
+Version 1.6.59 [September 28, 2026]
+  Fixed CVE-2026-46675 (medium severity):
+    Use-after-free of zlib input in `png_read_end` after incomplete zTXt, iTXt
+    or iCCP decompression.
+    (Reported independently by Ze Sheng and
+    <JasonHonKL@users.noreply.github.com>.)
+  Fixed a regression introduced in version 1.6.47 that caused libpng to reject
+    hIST chunks in their correct position, after PLTE.
+    (Contributed by Yuki Sekiguchi.)
+  Prevented a double free of `png_struct` members after an allocation failure.
+    (Contributed by Anthony Hurtado.)
+  Applied fixes and updates to the CMake build.
+  Adopted the REUSE Specification for licensing the CI files.
 
 Send comments/corrections/commendations to png-mng-implement at lists.sf.net.
 Subscription is required; visit
diff --git a/src/java.desktop/share/native/libsplashscreen/libpng/README b/src/java.desktop/share/native/libsplashscreen/libpng/README
index 75a92ab72330..9fd4e425c4f1 100644
--- a/src/java.desktop/share/native/libsplashscreen/libpng/README
+++ b/src/java.desktop/share/native/libsplashscreen/libpng/README
@@ -1,4 +1,4 @@
-README for libpng version 1.6.58
+README for libpng version 1.6.59
 ================================
 
 See the note about version numbers near the top of `png.h`.
diff --git a/src/java.desktop/share/native/libsplashscreen/libpng/png.c b/src/java.desktop/share/native/libsplashscreen/libpng/png.c
index 12f7ce53f34a..40381ac37738 100644
--- a/src/java.desktop/share/native/libsplashscreen/libpng/png.c
+++ b/src/java.desktop/share/native/libsplashscreen/libpng/png.c
@@ -42,7 +42,7 @@
 #include "pngpriv.h"
 
 /* Generate a compiler error if there is an old png.h in the search path. */
-typedef png_libpng_version_1_6_58 Your_png_h_is_not_version_1_6_58;
+typedef png_libpng_version_1_6_59 Your_png_h_is_not_version_1_6_59;
 
 /* Sanity check the chunks definitions - PNG_KNOWN_CHUNKS from pngpriv.h and the
  * corresponding macro definitions.  This causes a compile time failure if
@@ -849,7 +849,7 @@ png_get_copyright(png_const_structrp png_ptr)
    return PNG_STRING_COPYRIGHT
 #else
    return PNG_STRING_NEWLINE \
-      "libpng version 1.6.58" PNG_STRING_NEWLINE \
+      "libpng version 1.6.59" PNG_STRING_NEWLINE \
       "Copyright (c) 2018-2026 Cosmin Truta" PNG_STRING_NEWLINE \
       "Copyright (c) 1998-2002,2004,2006-2018 Glenn Randers-Pehrson" \
       PNG_STRING_NEWLINE \
diff --git a/src/java.desktop/share/native/libsplashscreen/libpng/png.h b/src/java.desktop/share/native/libsplashscreen/libpng/png.h
index 86a7c8ca1b02..d5c65de77dc6 100644
--- a/src/java.desktop/share/native/libsplashscreen/libpng/png.h
+++ b/src/java.desktop/share/native/libsplashscreen/libpng/png.h
@@ -29,7 +29,7 @@
  * However, the following notice accompanied the original version of this
  * file and, per its terms, should not be removed:
  *
- * libpng version 1.6.58
+ * libpng version 1.6.59
  *
  * Copyright (c) 2018-2026 Cosmin Truta
  * Copyright (c) 1998-2002,2004,2006-2018 Glenn Randers-Pehrson
@@ -43,7 +43,7 @@
  *   libpng versions 0.89, June 1996, through 0.96, May 1997: Andreas Dilger
  *   libpng versions 0.97, January 1998, through 1.6.35, July 2018:
  *     Glenn Randers-Pehrson
- *   libpng versions 1.6.36, December 2018, through 1.6.58, April 2026:
+ *   libpng versions 1.6.36, December 2018, through 1.6.59, September 2026:
  *     Cosmin Truta
  *   See also "Contributing Authors", below.
  */
@@ -267,7 +267,7 @@
  *    ...
  *    1.5.30                  15    10530  15.so.15.30[.0]
  *    ...
- *    1.6.58                  16    10658  16.so.16.58[.0]
+ *    1.6.59                  16    10659  16.so.16.59[.0]
  *
  *    Henceforth the source version will match the shared-library major and
  *    minor numbers; the shared-library major version number will be used for
@@ -303,7 +303,7 @@
  */
 
 /* Version information for png.h - this should match the version in png.c */
-#define PNG_LIBPNG_VER_STRING "1.6.58"
+#define PNG_LIBPNG_VER_STRING "1.6.59"
 #define PNG_HEADER_VERSION_STRING " libpng version " PNG_LIBPNG_VER_STRING "\n"
 
 /* The versions of shared library builds should stay in sync, going forward */
@@ -314,7 +314,7 @@
 /* These should match the first 3 components of PNG_LIBPNG_VER_STRING: */
 #define PNG_LIBPNG_VER_MAJOR   1
 #define PNG_LIBPNG_VER_MINOR   6
-#define PNG_LIBPNG_VER_RELEASE 58
+#define PNG_LIBPNG_VER_RELEASE 59
 
 /* This should be zero for a public release, or non-zero for a
  * development version.
@@ -345,7 +345,7 @@
  * From version 1.0.1 it is:
  * XXYYZZ, where XX=major, YY=minor, ZZ=release
  */
-#define PNG_LIBPNG_VER 10658 /* 1.6.58 */
+#define PNG_LIBPNG_VER 10659 /* 1.6.59 */
 
 /* Library configuration: these options cannot be changed after
  * the library has been built.
@@ -455,7 +455,7 @@ extern "C" {
 /* This triggers a compiler error in png.c, if png.c and png.h
  * do not agree upon the version number.
  */
-typedef char *png_libpng_version_1_6_58;
+typedef char *png_libpng_version_1_6_59;
 
 /* Basic control structions.  Read libpng-manual.txt or libpng.3 for more info.
  *
diff --git a/src/java.desktop/share/native/libsplashscreen/libpng/pngconf.h b/src/java.desktop/share/native/libsplashscreen/libpng/pngconf.h
index e035a155803a..d57f2406ab5d 100644
--- a/src/java.desktop/share/native/libsplashscreen/libpng/pngconf.h
+++ b/src/java.desktop/share/native/libsplashscreen/libpng/pngconf.h
@@ -29,7 +29,7 @@
  * However, the following notice accompanied the original version of this
  * file and, per its terms, should not be removed:
  *
- * libpng version 1.6.58
+ * libpng version 1.6.59
  *
  * Copyright (c) 2018-2026 Cosmin Truta
  * Copyright (c) 1998-2002,2004,2006-2016,2018 Glenn Randers-Pehrson
diff --git a/src/java.desktop/share/native/libsplashscreen/libpng/pnglibconf.h b/src/java.desktop/share/native/libsplashscreen/libpng/pnglibconf.h
index 46cba507f65a..6e7bcfd308d6 100644
--- a/src/java.desktop/share/native/libsplashscreen/libpng/pnglibconf.h
+++ b/src/java.desktop/share/native/libsplashscreen/libpng/pnglibconf.h
@@ -31,7 +31,7 @@
  * However, the following notice accompanied the original version of this
  * file and, per its terms, should not be removed:
  */
-/* libpng version 1.6.58 */
+/* libpng version 1.6.59 */
 
 /* Copyright (c) 2018-2026 Cosmin Truta */
 /* Copyright (c) 1998-2002,2004,2006-2018 Glenn Randers-Pehrson */
diff --git a/src/java.desktop/share/native/libsplashscreen/libpng/pngread.c b/src/java.desktop/share/native/libsplashscreen/libpng/pngread.c
index 70df18926f50..56aa16d41cfa 100644
--- a/src/java.desktop/share/native/libsplashscreen/libpng/pngread.c
+++ b/src/java.desktop/share/native/libsplashscreen/libpng/pngread.c
@@ -711,8 +711,10 @@ png_read_end(png_structrp png_ptr, png_inforp info_ptr)
    if (png_ptr == NULL)
       return;
 
-   /* If png_read_end is called in the middle of reading the rows there may
-    * still be pending IDAT data and an owned zstream.  Deal with this here.
+   /* If png_read_end is called in the middle of reading the rows,
+    * there may still be pending IDAT data and an owned zstream.
+    * If it is called before row reading starts, there is pending
+    * IDAT data but no owned zstream.
     */
 #ifdef PNG_HANDLE_AS_UNKNOWN_SUPPORTED
    if (png_chunk_unknown_handling(png_ptr, png_IDAT) == 0)
diff --git a/src/java.desktop/share/native/libsplashscreen/libpng/pngrtran.c b/src/java.desktop/share/native/libsplashscreen/libpng/pngrtran.c
index 99050080e9d1..ef1fb0430973 100644
--- a/src/java.desktop/share/native/libsplashscreen/libpng/pngrtran.c
+++ b/src/java.desktop/share/native/libsplashscreen/libpng/pngrtran.c
@@ -543,6 +543,7 @@ png_set_quantize(png_structrp png_ptr, png_colorp palette,
        * this function more than once per png_struct.
        */
       png_free(png_ptr, png_ptr->quantize_index);
+      png_ptr->quantize_index = NULL;
       png_ptr->quantize_index = (png_bytep)png_malloc(png_ptr,
           PNG_MAX_PALETTE_LENGTH);
       for (i = 0; i < PNG_MAX_PALETTE_LENGTH; i++)
diff --git a/src/java.desktop/share/native/libsplashscreen/libpng/pngrutil.c b/src/java.desktop/share/native/libsplashscreen/libpng/pngrutil.c
index 4712dfd418ac..57afbcdde686 100644
--- a/src/java.desktop/share/native/libsplashscreen/libpng/pngrutil.c
+++ b/src/java.desktop/share/native/libsplashscreen/libpng/pngrutil.c
@@ -434,6 +434,18 @@ png_read_buffer(png_structrp png_ptr, png_alloc_size_t new_size)
 }
 #endif /* READ_iCCP|iTXt|pCAL|sCAL|sPLT|tEXt|zTXt|eXIf|SEQUENTIAL_READ */
 
+/* Detach the zstream from the input and output buffers left by
+ * the current or a previous owner, and possibly deallocated since.
+ */
+static void
+png_inflate_detach_buffers(png_structrp png_ptr)
+{
+   png_ptr->zstream.next_in = NULL;
+   png_ptr->zstream.avail_in = 0;
+   png_ptr->zstream.next_out = NULL;
+   png_ptr->zstream.avail_out = 0;
+}
+
 /* png_inflate_claim: claim the zstream for some nefarious purpose that involves
  * decompression.  Returns Z_OK on success, else a zlib error code.  It checks
  * the owner but, in final release builds, just issues a warning if some other
@@ -494,13 +506,7 @@ png_inflate_claim(png_structrp png_ptr, png_uint_32 owner)
 
 #endif /* ZLIB_VERNUM >= 0x1240 */
 
-      /* Set this for safety, just in case the previous owner left pointers to
-       * memory allocations.
-       */
-      png_ptr->zstream.next_in = NULL;
-      png_ptr->zstream.avail_in = 0;
-      png_ptr->zstream.next_out = NULL;
-      png_ptr->zstream.avail_out = 0;
+      png_inflate_detach_buffers(png_ptr);
 
       if ((png_ptr->flags & PNG_FLAG_ZSTREAM_INITIALIZED) != 0)
       {
@@ -837,7 +843,8 @@ png_decompress_chunk(png_structrp png_ptr,
          else if (ret == Z_OK)
             ret = PNG_UNEXPECTED_ZLIB_RETURN;
 
-         /* Release the claimed stream */
+         /* Release the claimed stream. */
+         png_inflate_detach_buffers(png_ptr);
          png_ptr->zowner = 0;
       }
 
@@ -1520,6 +1527,7 @@ png_handle_iCCP(png_structrp png_ptr, png_inforp info_ptr, png_uint_32 length)
 
                                     if (errmsg == NULL)
                                     {
+                                       png_inflate_detach_buffers(png_ptr);
                                        png_ptr->zowner = 0;
                                        return handled_ok;
                                     }
@@ -1546,7 +1554,8 @@ png_handle_iCCP(png_structrp png_ptr, png_inforp info_ptr, png_uint_32 length)
                else /* profile truncated */
                   errmsg = png_ptr->zstream.msg;
 
-               /* Release the stream */
+               /* Release the claimed stream. */
+               png_inflate_detach_buffers(png_ptr);
                png_ptr->zowner = 0;
             }
 
@@ -3091,7 +3100,7 @@ read_chunks[PNG_INDEX_unknown] =
 #  define CDiTXt  NoCheck,    6U,      0, hIHDR,        1
       /* Allocates 'length+1'; checked in the handler */
 #  define CDbKGD       6U,    1U,  hIDAT, hIHDR,        0
-#  define CDhIST    1024U,    0U,  hPLTE, hIHDR,        0
+#  define CDhIST    1024U,    0U,  hIDAT, hPLTE,        0
 #  define CDpHYs       9U,    9U,  hIDAT, hIHDR,        0
 #  define CDsPLT  NoCheck,    3U,  hIDAT, hIHDR,        1
       /* Allocates 'length+1'; checked in the handler */
@@ -4366,11 +4375,7 @@ png_read_finish_IDAT(png_structrp png_ptr)
     */
    if (png_ptr->zowner == png_IDAT)
    {
-      /* Always do this; the pointers otherwise point into the read buffer. */
-      png_ptr->zstream.next_in = NULL;
-      png_ptr->zstream.avail_in = 0;
-
-      /* Now we no longer own the zstream. */
+      png_inflate_detach_buffers(png_ptr);
       png_ptr->zowner = 0;
 
       /* The slightly weird semantics of the sequential IDAT reading is that we
@@ -4633,6 +4638,7 @@ defined(PNG_USER_TRANSFORM_PTR_SUPPORTED)
    {
       png_free(png_ptr, png_ptr->big_row_buf);
       png_free(png_ptr, png_ptr->big_prev_row);
+      png_ptr->big_row_buf = png_ptr->big_prev_row = NULL;
 
       if (png_ptr->interlaced != 0)
          png_ptr->big_row_buf = (png_bytep)png_calloc(png_ptr,
diff --git a/src/java.desktop/share/native/libsplashscreen/libpng/pngset.c b/src/java.desktop/share/native/libsplashscreen/libpng/pngset.c
index 29082a6be089..d1a3fff77c2d 100644
--- a/src/java.desktop/share/native/libsplashscreen/libpng/pngset.c
+++ b/src/java.desktop/share/native/libsplashscreen/libpng/pngset.c
@@ -840,6 +840,7 @@ png_set_PLTE(png_structrp png_ptr, png_inforp info_ptr,
     * regardless of num_palette.
     */
    png_free(png_ptr, png_ptr->palette);
+   png_ptr->palette = NULL;
    png_ptr->palette = png_voidcast(png_colorp, png_calloc(png_ptr,
        PNG_MAX_PALETTE_LENGTH * (sizeof (png_color))));
    info_ptr->palette = png_voidcast(png_colorp, png_calloc(png_ptr,
@@ -1252,6 +1253,7 @@ png_set_tRNS(png_structrp png_ptr, png_inforp info_ptr,
            * functions (e.g. png_do_expand_palette).
            */
           png_free(png_ptr, png_ptr->trans_alpha);
+          png_ptr->trans_alpha = NULL;
           png_ptr->trans_alpha = png_voidcast(png_bytep,
               png_malloc(png_ptr, PNG_MAX_PALETTE_LENGTH));
           memset(png_ptr->trans_alpha, 0xff, PNG_MAX_PALETTE_LENGTH);

From ec6f1aebd042d42adf3025aa21df5bf0480dc7fa Mon Sep 17 00:00:00 2001
From: David Briemann <dbriemann@openjdk.org>
Date: Mon, 7 Sep 2026 06:05:51 +0000
Subject: [PATCH 2/3] 8390212: (dc) On AIX, DatagramChannel.join can join IPv4
 multicast group with IPv6 socket

Backport-of: 05eaf783b270ec5cd901704c5e16b6201de7a9e4
---
 src/java.base/unix/native/libnio/ch/Net.c | 20 +++++++++++++++++---
 test/jdk/ProblemList.txt                  |  5 +----
 2 files changed, 18 insertions(+), 7 deletions(-)

diff --git a/src/java.base/unix/native/libnio/ch/Net.c b/src/java.base/unix/native/libnio/ch/Net.c
index b9ab0baf5413..9b6f8f5294d9 100644
--- a/src/java.base/unix/native/libnio/ch/Net.c
+++ b/src/java.base/unix/native/libnio/ch/Net.c
@@ -91,6 +91,11 @@ static void initGroupSourceReq(JNIEnv* env, jbyteArray group, jint index,
 
 #ifdef _AIX
 
+static jboolean isIPv4MappedGroup(struct group_source_req *req) {
+    struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&req->gsr_group;
+    return IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr) ? JNI_TRUE : JNI_FALSE;
+}
+
 /*
  * Checks whether or not "socket extensions for multicast source filters" is supported.
  * Returns JNI_TRUE if it is supported, JNI_FALSE otherwise
@@ -220,7 +225,7 @@ Java_sun_nio_ch_Net_shouldSetBothIPv4AndIPv6Options0(JNIEnv* env, jclass cl)
 JNIEXPORT jboolean JNICALL
 Java_sun_nio_ch_Net_canIPv6SocketJoinIPv4Group0(JNIEnv* env, jclass cl)
 {
-#if defined(__linux__) || defined(__APPLE__)
+#if defined(__linux__) || defined(__APPLE__) || defined(_AIX)
     /* IPv6 sockets can join IPv4 multicast groups */
     return JNI_TRUE;
 #else
@@ -232,7 +237,7 @@ Java_sun_nio_ch_Net_canIPv6SocketJoinIPv4Group0(JNIEnv* env, jclass cl)
 JNIEXPORT jboolean JNICALL
 Java_sun_nio_ch_Net_canJoin6WithIPv4Group0(JNIEnv* env, jclass cl)
 {
-#if defined(__APPLE__)
+#if defined(__APPLE__) || defined(_AIX)
     /* IPV6_ADD_MEMBERSHIP can be used to join IPv4 multicast groups */
     return JNI_TRUE;
 #else
@@ -751,6 +756,11 @@ Java_sun_nio_ch_Net_joinOrDrop6(JNIEnv *env, jobject this, jboolean join, jobjec
     if (n < 0) {
         if (join && (errno == ENOPROTOOPT || errno == EOPNOTSUPP))
             return IOS_UNAVAILABLE;
+#ifdef _AIX
+        // AIX rejects MCAST_*_SOURCE_GROUP for IPv4-mapped groups with EINVAL
+        if (source != NULL && errno == EINVAL && isIPv4MappedGroup(&req))
+            return IOS_UNAVAILABLE;
+#endif
         handleSocketErrorWithMessage(env, errno, "setsockopt failed");
     }
     return 0;
@@ -775,6 +785,11 @@ Java_sun_nio_ch_Net_blockOrUnblock6(JNIEnv *env, jobject this, jboolean block, j
     if (n < 0) {
         if (block && (errno == ENOPROTOOPT || errno == EOPNOTSUPP))
             return IOS_UNAVAILABLE;
+#ifdef _AIX
+        // AIX rejects MCAST_BLOCK/UNBLOCK_SOURCE for IPv4-mapped groups with EINVAL
+        if (errno == EINVAL && isIPv4MappedGroup(&req))
+            return IOS_UNAVAILABLE;
+#endif
         handleSocketError(env, errno);
     }
     return 0;
@@ -971,4 +986,3 @@ Java_sun_nio_ch_Net_sendOOB(JNIEnv* env, jclass this, jobject fdo, jbyte b)
     int n = send(fdval(env, fdo), (const void*)&b, 1, MSG_OOB);
     return convertReturnVal(env, n, JNI_FALSE);
 }
-
diff --git a/test/jdk/ProblemList.txt b/test/jdk/ProblemList.txt
index 0b610816bd45..788c04f7267c 100644
--- a/test/jdk/ProblemList.txt
+++ b/test/jdk/ProblemList.txt
@@ -573,17 +573,14 @@ javax/management/remote/mandatory/subjectDelegation/SubjectDelegation1Test.java
 java/net/DatagramSocket/DatagramSocketExample.java              8144003,8308807 macosx-all,aix-ppc64
 java/net/DatagramSocket/DatagramSocketMulticasting.java         8144003,8308807 macosx-all,aix-ppc64
 
-java/net/MulticastSocket/B6427403.java                          8308807 aix-ppc64
 java/net/MulticastSocket/IPMulticastIF.java                     8308807 aix-ppc64
 java/net/MulticastSocket/JoinLeave.java                         8308807 aix-ppc64
-java/net/MulticastSocket/MulticastAddresses.java                8308807 aix-ppc64
 java/net/MulticastSocket/NoLoopbackPackets.java                 7122846,8308807 macosx-all,aix-ppc64
 java/net/MulticastSocket/NoSetNetworkInterface.java             8308807 aix-ppc64
-java/net/MulticastSocket/Promiscuous.java                       8308807 aix-ppc64
 java/net/MulticastSocket/SetGetNetworkInterfaceTest.java        8308807 aix-ppc64
 java/net/MulticastSocket/SetLoopbackMode.java                   7122846,8308807 macosx-all,aix-ppc64
 java/net/MulticastSocket/SetOutgoingIf.java                     8308807 aix-ppc64
-java/net/MulticastSocket/Test.java                              7145658,8308807 macosx-all,aix-ppc64
+java/net/MulticastSocket/Test.java                              7145658 macosx-all
 
 ############################################################################
 

From 398dc70c4e3d5feee7d6bdab0c9a10ecb4483d34 Mon Sep 17 00:00:00 2001
From: Martin Doerr <mdoerr@openjdk.org>
Date: Fri, 4 Sep 2026 12:18:30 +0000
Subject: [PATCH 3/3] 8390765: [PPC64] AES Crypto intrinsics may access memory
 beyond the key array

Reviewed-by: mbaesken
Backport-of: 5e150f7439333a8d476cac350d41224e0d6d02ab
---
 src/hotspot/cpu/ppc/assembler_ppc.hpp        |  34 +-
 src/hotspot/cpu/ppc/assembler_ppc.inline.hpp | 110 +++++++
 src/hotspot/cpu/ppc/stubGenerator_ppc.cpp    | 319 ++++---------------
 3 files changed, 198 insertions(+), 265 deletions(-)

diff --git a/src/hotspot/cpu/ppc/assembler_ppc.hpp b/src/hotspot/cpu/ppc/assembler_ppc.hpp
index 7eda53749fb2..1765f636e6ff 100644
--- a/src/hotspot/cpu/ppc/assembler_ppc.hpp
+++ b/src/hotspot/cpu/ppc/assembler_ppc.hpp
@@ -551,6 +551,10 @@ class Assembler : public AbstractAssembler {
     STXVL_OPCODE   = (31u << OPCODE_SHIFT |  397u << 1),
     LXVD2X_OPCODE  = (31u << OPCODE_SHIFT |  844u << 1),
     STXVD2X_OPCODE = (31u << OPCODE_SHIFT |  972u << 1),
+    LXVW4X_OPCODE  = (31u << OPCODE_SHIFT |  780u << 1),
+    STXVW4X_OPCODE = (31u << OPCODE_SHIFT |  908u << 1),
+    LXVB16X_OPCODE = (31u << OPCODE_SHIFT |  876u << 1),
+    STXVB16X_OPCODE= (31u << OPCODE_SHIFT | 1004u << 1),
     MTVSRD_OPCODE  = (31u << OPCODE_SHIFT |  179u << 1),
     MTVSRDD_OPCODE = (31u << OPCODE_SHIFT |  435u << 1),
     MTVSRWZ_OPCODE = (31u << OPCODE_SHIFT |  243u << 1),
@@ -2377,14 +2381,27 @@ class Assembler : public AbstractAssembler {
   inline void mfvscr(   VectorRegister d);
 
   // Vector-Scalar (VSX) instructions.
-  inline void lxv(      VectorSRegister d, int si16, Register a);
-  inline void stxv(     VectorSRegister d, int si16, Register a);
-  inline void lxvl(     VectorSRegister d, Register a, Register b);
-  inline void stxvl(    VectorSRegister d, Register a, Register b);
+  // Power8
   inline void lxvd2x(   VectorSRegister d, Register a);
   inline void lxvd2x(   VectorSRegister d, Register a, Register b);
   inline void stxvd2x(  VectorSRegister d, Register a);
   inline void stxvd2x(  VectorSRegister d, Register a, Register b);
+  inline void lxvw4x(   VectorSRegister d, Register a);
+  inline void lxvw4x(   VectorSRegister d, Register a, Register b);
+  inline void stxvw4x(  VectorSRegister d, Register a);
+  inline void stxvw4x(  VectorSRegister d, Register a, Register b);
+
+  // Power9
+  inline void lxvb16x(  VectorSRegister d, Register a);
+  inline void lxvb16x(  VectorSRegister d, Register a, Register b);
+  inline void stxvb16x( VectorSRegister d, Register a);
+  inline void stxvb16x( VectorSRegister d, Register a, Register b);
+
+  inline void lxv(      VectorSRegister d, int si16, Register a);
+  inline void stxv(     VectorSRegister d, int si16, Register a);
+  inline void lxvl(     VectorSRegister d, Register a, Register b);
+  inline void stxvl(    VectorSRegister d, Register a, Register b);
+
   inline void mtvrwz(   VectorRegister  d, Register a);
   inline void mfvrwz(   Register        a, VectorRegister d);
   inline void mtvrd(    VectorRegister  d, Register a);
@@ -2579,6 +2596,15 @@ class Assembler : public AbstractAssembler {
   inline void vec_perm(VectorRegister first_dest, VectorRegister second, VectorRegister perm);
   inline void vec_perm(VectorRegister dest, VectorRegister first, VectorRegister second, VectorRegister perm);
 
+  // Load/Store unaligned vectors with offs (multiple of 16). Byte versions require vp for Power8 LE.
+  inline void load_byte_vector_unaligned(VectorRegister dest, int offs, Register base, Register tmp,
+                                         VectorRegister vp); // vp should be pre-computed (see generator below)
+  inline void store_byte_vector_unaligned(VectorRegister val, int offs, Register base, Register tmp,
+                                          VectorRegister vp, VectorRegister vtmp = vnoreg); // clobbers val if no vtmp provided
+  inline void compute_vp_for_byte_vector_unaligned(VectorRegister dest, VectorRegister vtmp);
+  inline void load_word_vector_unaligned(VectorRegister dest, int offs, Register base, Register tmp);
+  inline void store_word_vector_unaligned(VectorRegister val, int offs, Register base, Register tmp);
+
   // RegisterOrConstant versions.
   // These emitters choose between the versions using two registers and
   // those with register and immediate, depending on the content of roc.
diff --git a/src/hotspot/cpu/ppc/assembler_ppc.inline.hpp b/src/hotspot/cpu/ppc/assembler_ppc.inline.hpp
index 4aa767174a8e..c1ce1454c212 100644
--- a/src/hotspot/cpu/ppc/assembler_ppc.inline.hpp
+++ b/src/hotspot/cpu/ppc/assembler_ppc.inline.hpp
@@ -828,6 +828,14 @@ inline void Assembler::lxvd2x(  VectorSRegister d, Register s1)              { e
 inline void Assembler::lxvd2x(  VectorSRegister d, Register s1, Register s2) { emit_int32( LXVD2X_OPCODE  | vsrt(d) | ra0mem(s1) | rb(s2)); }
 inline void Assembler::stxvd2x( VectorSRegister d, Register s1)              { emit_int32( STXVD2X_OPCODE | vsrs(d) | ra(0) | rb(s1)); }
 inline void Assembler::stxvd2x( VectorSRegister d, Register s1, Register s2) { emit_int32( STXVD2X_OPCODE | vsrs(d) | ra0mem(s1) | rb(s2)); }
+inline void Assembler::lxvw4x(  VectorSRegister d, Register s1)              { emit_int32( LXVW4X_OPCODE  | vsrt(d) | ra(0) | rb(s1)); }
+inline void Assembler::lxvw4x(  VectorSRegister d, Register s1, Register s2) { emit_int32( LXVW4X_OPCODE  | vsrt(d) | ra0mem(s1) | rb(s2)); }
+inline void Assembler::stxvw4x( VectorSRegister d, Register s1)              { emit_int32( STXVW4X_OPCODE | vsrs(d) | ra(0) | rb(s1)); }
+inline void Assembler::stxvw4x( VectorSRegister d, Register s1, Register s2) { emit_int32( STXVW4X_OPCODE | vsrs(d) | ra0mem(s1) | rb(s2)); }
+inline void Assembler::lxvb16x( VectorSRegister d, Register s1)              { emit_int32( LXVB16X_OPCODE | vsrt(d) | ra(0) | rb(s1)); }
+inline void Assembler::lxvb16x( VectorSRegister d, Register s1, Register s2) { emit_int32( LXVB16X_OPCODE | vsrt(d) | ra0mem(s1) | rb(s2)); }
+inline void Assembler::stxvb16x(VectorSRegister d, Register s1)              { emit_int32( STXVB16X_OPCODE| vsrs(d) | ra(0) | rb(s1)); }
+inline void Assembler::stxvb16x(VectorSRegister d, Register s1, Register s2) { emit_int32( STXVB16X_OPCODE| vsrs(d) | ra0mem(s1) | rb(s2)); }
 inline void Assembler::mtvsrd(  VectorSRegister d, Register a)               { emit_int32( MTVSRD_OPCODE  | vsrt(d)  | ra(a)); }
 inline void Assembler::mtvsrdd( VectorSRegister d, Register a, Register b)   { emit_int32( MTVSRDD_OPCODE | vsrt(d)  | ra(a) | rb(b)); }
 inline void Assembler::mfvsrd(  Register d, VectorSRegister a)               { emit_int32( MFVSRD_OPCODE  | vsrs(a)  | ra(d)); }
@@ -1188,6 +1196,108 @@ inline void Assembler::vec_perm(VectorRegister dest, VectorRegister first, Vecto
 #endif
 }
 
+inline void Assembler::load_byte_vector_unaligned(VectorRegister dest, int offs, Register base, Register tmp,
+                                                  VectorRegister vp) {
+  VectorSRegister vsr = dest->to_vsr();
+  if (PowerArchitecturePPC64 >= 9) {
+#if !defined(VM_LITTLE_ENDIAN)
+    lxv(vsr, offs, base); // all vector load/store instructions use the same byte order on BE
+#else
+    if (offs == 0) {
+      lxvb16x(vsr, base);
+    } else {
+      li(tmp, offs);
+      lxvb16x(vsr, base, tmp);
+    }
+#endif
+  } else { // Power8 only supports very limited instructions
+    if (offs == 0) {
+      lxvd2x(vsr, base);
+    } else {
+      li(tmp, offs);
+      lxvd2x(vsr, base, tmp);
+    }
+#if defined(VM_LITTLE_ENDIAN)
+    // need to swap bytes in both double-words
+    vperm(dest, dest, dest, vp);
+#endif
+  }
+}
+
+inline void Assembler::store_byte_vector_unaligned(VectorRegister val, int offs, Register base, Register tmp,
+                                                   VectorRegister vp, VectorRegister vtmp) {
+  VectorSRegister vsr = val->to_vsr();
+  if (PowerArchitecturePPC64 >= 9) {
+#if !defined(VM_LITTLE_ENDIAN)
+    stxv(vsr, offs, base); // all vector load/store instructions use the same byte order on BE
+#else
+    if (offs == 0) {
+      stxvb16x(vsr, base);
+    } else {
+      li(tmp, offs);
+      stxvb16x(vsr, base, tmp);
+    }
+#endif
+  } else { // Power8 only supports very limited instructions
+#if defined(VM_LITTLE_ENDIAN)
+    // need to swap bytes in both double-words
+    if (vtmp != vnoreg) {
+      vperm(vtmp, val, val, vp);
+      vsr = vtmp->to_vsr();
+    } else {
+      vperm(val, val, val, vp); // clobbers val!
+    }
+#endif
+    if (offs == 0) {
+      stxvd2x(vsr, base);
+    } else {
+      li(tmp, offs);
+      stxvd2x(vsr, base, tmp);
+    }
+  }
+}
+
+inline void Assembler::compute_vp_for_byte_vector_unaligned(VectorRegister dest, VectorRegister vtmp) {
+#if defined(VM_LITTLE_ENDIAN)
+  if (PowerArchitecturePPC64 < 9) {
+    li(R0, 0);
+    vspltisb(vtmp, 7);      // vtmp = [7, ..., 7]
+    lvsl(dest, R0);         // dest = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]
+    vxor(dest, dest, vtmp); // dest = [7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8]
+  }
+#endif
+}
+
+inline void Assembler::load_word_vector_unaligned(VectorRegister dest, int offs, Register base, Register tmp) {
+  VectorSRegister vsr = dest->to_vsr();
+#if !defined(VM_LITTLE_ENDIAN)
+  if (PowerArchitecturePPC64 >= 9) {
+    lxv(vsr, offs, base); // all vector load/store instructions use the same byte order on BE
+  } else
+#endif
+  if (offs == 0) {
+    lxvw4x(vsr, base);
+  } else {
+    li(tmp, offs);
+    lxvw4x(vsr, base, tmp);
+  }
+}
+
+inline void Assembler::store_word_vector_unaligned(VectorRegister val, int offs, Register base, Register tmp) {
+  VectorSRegister vsr = val->to_vsr();
+#if !defined(VM_LITTLE_ENDIAN)
+  if (PowerArchitecturePPC64 >= 9) {
+    stxv(vsr, offs, base); // all vector load/store instructions use the same byte order on BE
+  } else
+#endif
+  if (offs == 0) {
+    stxvw4x(vsr, base);
+  } else {
+    li(tmp, offs);
+    stxvw4x(vsr, base, tmp);
+  }
+}
+
 inline void Assembler::load_const(Register d, void* x, Register tmp) {
    load_const(d, (long)x, tmp);
 }
diff --git a/src/hotspot/cpu/ppc/stubGenerator_ppc.cpp b/src/hotspot/cpu/ppc/stubGenerator_ppc.cpp
index 58f984e74b8f..5ad5f57958e4 100644
--- a/src/hotspot/cpu/ppc/stubGenerator_ppc.cpp
+++ b/src/hotspot/cpu/ppc/stubGenerator_ppc.cpp
@@ -2810,10 +2810,8 @@ class StubGenerator: public StubCodeGenerator {
     Register to             = R4_ARG2;  // destination array address
     Register key            = R5_ARG3;  // round key array
 
-    Register keylen         = R8;
-    Register temp           = R9;
-    Register keypos         = R10;
-    Register fifteen        = R12;
+    Register keylen         = R6;
+    Register tmp            = R7;
 
     VectorRegister vRet     = VR0;
 
@@ -2822,68 +2820,27 @@ class StubGenerator: public StubCodeGenerator {
     VectorRegister vKey3    = VR3;
     VectorRegister vKey4    = VR4;
 
-    VectorRegister fromPerm = VR5;
-    VectorRegister keyPerm  = VR6;
-    VectorRegister toPerm   = VR7;
-    VectorRegister fSplt    = VR8;
+    VectorRegister vp       = VR6;  // permute vector for byte vector accesses on P8 LE
 
-    VectorRegister vTmp1    = VR9;
-    VectorRegister vTmp2    = VR10;
-    VectorRegister vTmp3    = VR11;
-    VectorRegister vTmp4    = VR12;
-
-    __ li              (fifteen, 15);
+    __ compute_vp_for_byte_vector_unaligned(vp, /*temp*/ vRet);
 
     // load unaligned from[0-15] to vRet
-    __ lvx             (vRet, from);
-    __ lvx             (vTmp1, fifteen, from);
-    __ lvsl            (fromPerm, from);
-#ifdef VM_LITTLE_ENDIAN
-    __ vspltisb        (fSplt, 0x0f);
-    __ vxor            (fromPerm, fromPerm, fSplt);
-#endif
-    __ vperm           (vRet, vRet, vTmp1, fromPerm);
+    __ load_byte_vector_unaligned(vRet, 0, from, tmp, vp);
+
+    // load the 1st round key to vKey1
+    __ load_word_vector_unaligned(vKey1, 0, key, tmp);
 
     // load keylen (44 or 52 or 60)
     __ lwz             (keylen, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT), key);
 
-    // to load keys
-    __ load_perm       (keyPerm, key);
-#ifdef VM_LITTLE_ENDIAN
-    __ vspltisb        (vTmp2, -16);
-    __ vrld            (keyPerm, keyPerm, vTmp2);
-    __ vrld            (keyPerm, keyPerm, vTmp2);
-    __ vsldoi          (keyPerm, keyPerm, keyPerm, 8);
-#endif
-
-    // load the 1st round key to vTmp1
-    __ lvx             (vTmp1, key);
-    __ li              (keypos, 16);
-    __ lvx             (vKey1, keypos, key);
-    __ vec_perm        (vTmp1, vKey1, keyPerm);
-
     // 1st round
-    __ vxor            (vRet, vRet, vTmp1);
-
-    // load the 2nd round key to vKey1
-    __ li              (keypos, 32);
-    __ lvx             (vKey2, keypos, key);
-    __ vec_perm        (vKey1, vKey2, keyPerm);
-
-    // load the 3rd round key to vKey2
-    __ li              (keypos, 48);
-    __ lvx             (vKey3, keypos, key);
-    __ vec_perm        (vKey2, vKey3, keyPerm);
-
-    // load the 4th round key to vKey3
-    __ li              (keypos, 64);
-    __ lvx             (vKey4, keypos, key);
-    __ vec_perm        (vKey3, vKey4, keyPerm);
+    __ vxor            (vRet, vRet, vKey1);
 
-    // load the 5th round key to vKey4
-    __ li              (keypos, 80);
-    __ lvx             (vTmp1, keypos, key);
-    __ vec_perm        (vKey4, vTmp1, keyPerm);
+    // load the 2nd - 5th round key to vKey1 - vKey4
+    __ load_word_vector_unaligned(vKey1, 16, key, tmp);
+    __ load_word_vector_unaligned(vKey2, 32, key, tmp);
+    __ load_word_vector_unaligned(vKey3, 48, key, tmp);
+    __ load_word_vector_unaligned(vKey4, 64, key, tmp);
 
     // 2nd - 5th rounds
     __ vcipher         (vRet, vRet, vKey1);
@@ -2891,25 +2848,11 @@ class StubGenerator: public StubCodeGenerator {
     __ vcipher         (vRet, vRet, vKey3);
     __ vcipher         (vRet, vRet, vKey4);
 
-    // load the 6th round key to vKey1
-    __ li              (keypos, 96);
-    __ lvx             (vKey2, keypos, key);
-    __ vec_perm        (vKey1, vTmp1, vKey2, keyPerm);
-
-    // load the 7th round key to vKey2
-    __ li              (keypos, 112);
-    __ lvx             (vKey3, keypos, key);
-    __ vec_perm        (vKey2, vKey3, keyPerm);
-
-    // load the 8th round key to vKey3
-    __ li              (keypos, 128);
-    __ lvx             (vKey4, keypos, key);
-    __ vec_perm        (vKey3, vKey4, keyPerm);
-
-    // load the 9th round key to vKey4
-    __ li              (keypos, 144);
-    __ lvx             (vTmp1, keypos, key);
-    __ vec_perm        (vKey4, vTmp1, keyPerm);
+    // load the 6th - 9th round key to vKey1 - vKey4
+    __ load_word_vector_unaligned(vKey1, 80, key, tmp);
+    __ load_word_vector_unaligned(vKey2, 96, key, tmp);
+    __ load_word_vector_unaligned(vKey3, 112, key, tmp);
+    __ load_word_vector_unaligned(vKey4, 128, key, tmp);
 
     // 6th - 9th rounds
     __ vcipher         (vRet, vRet, vKey1);
@@ -2917,15 +2860,9 @@ class StubGenerator: public StubCodeGenerator {
     __ vcipher         (vRet, vRet, vKey3);
     __ vcipher         (vRet, vRet, vKey4);
 
-    // load the 10th round key to vKey1
-    __ li              (keypos, 160);
-    __ lvx             (vKey2, keypos, key);
-    __ vec_perm        (vKey1, vTmp1, vKey2, keyPerm);
-
-    // load the 11th round key to vKey2
-    __ li              (keypos, 176);
-    __ lvx             (vTmp1, keypos, key);
-    __ vec_perm        (vKey2, vTmp1, keyPerm);
+    // load the 10th - 11th round key to vKey1 - vKey2
+    __ load_word_vector_unaligned(vKey1, 144, key, tmp);
+    __ load_word_vector_unaligned(vKey2, 160, key, tmp);
 
     // if all round keys are loaded, skip next 4 rounds
     __ cmpwi           (CCR0, keylen, 44);
@@ -2935,15 +2872,9 @@ class StubGenerator: public StubCodeGenerator {
     __ vcipher         (vRet, vRet, vKey1);
     __ vcipher         (vRet, vRet, vKey2);
 
-    // load the 12th round key to vKey1
-    __ li              (keypos, 192);
-    __ lvx             (vKey2, keypos, key);
-    __ vec_perm        (vKey1, vTmp1, vKey2, keyPerm);
-
-    // load the 13th round key to vKey2
-    __ li              (keypos, 208);
-    __ lvx             (vTmp1, keypos, key);
-    __ vec_perm        (vKey2, vTmp1, keyPerm);
+    // load the 12th - 13th round key to vKey1 - vKey2
+    __ load_word_vector_unaligned(vKey1, 176, key, tmp);
+    __ load_word_vector_unaligned(vKey2, 192, key, tmp);
 
     // if all round keys are loaded, skip next 2 rounds
     __ cmpwi           (CCR0, keylen, 52);
@@ -2958,15 +2889,9 @@ class StubGenerator: public StubCodeGenerator {
     __ vcipher         (vRet, vRet, vKey1);
     __ vcipher         (vRet, vRet, vKey2);
 
-    // load the 14th round key to vKey1
-    __ li              (keypos, 224);
-    __ lvx             (vKey2, keypos, key);
-    __ vec_perm        (vKey1, vTmp1, vKey2, keyPerm);
-
-    // load the 15th round key to vKey2
-    __ li              (keypos, 240);
-    __ lvx             (vTmp1, keypos, key);
-    __ vec_perm        (vKey2, vTmp1, keyPerm);
+    // load the 14th - 15th round key to vKey1 - vKey2
+    __ load_word_vector_unaligned(vKey1, 208, key, tmp);
+    __ load_word_vector_unaligned(vKey2, 224, key, tmp);
 
     __ bind(L_doLast);
 
@@ -2974,23 +2899,8 @@ class StubGenerator: public StubCodeGenerator {
     __ vcipher         (vRet, vRet, vKey1);
     __ vcipherlast     (vRet, vRet, vKey2);
 
-#ifdef VM_LITTLE_ENDIAN
-    // toPerm = 0x0F0E0D0C0B0A09080706050403020100
-    __ lvsl            (toPerm, keypos); // keypos is a multiple of 16
-    __ vxor            (toPerm, toPerm, fSplt);
-
-    // Swap Bytes
-    __ vperm           (vRet, vRet, vRet, toPerm);
-#endif
-
     // store result (unaligned)
-    // Note: We can't use a read-modify-write sequence which touches additional Bytes.
-    Register lo = temp, hi = fifteen; // Reuse
-    __ vsldoi          (vTmp1, vRet, vRet, 8);
-    __ mfvrd           (hi, vRet);
-    __ mfvrd           (lo, vTmp1);
-    __ std             (hi, 0 LITTLE_ENDIAN_ONLY(+ 8), to);
-    __ std             (lo, 0 BIG_ENDIAN_ONLY(+ 8), to);
+    __ store_byte_vector_unaligned(vRet, 0, to, tmp, vp);
 
     __ blr();
 
@@ -3017,10 +2927,8 @@ class StubGenerator: public StubCodeGenerator {
     Register to             = R4_ARG2;  // destination array address
     Register key            = R5_ARG3;  // round key array
 
-    Register keylen         = R8;
-    Register temp           = R9;
-    Register keypos         = R10;
-    Register fifteen        = R12;
+    Register keylen         = R6;
+    Register tmp            = R7;
 
     VectorRegister vRet     = VR0;
 
@@ -3030,41 +2938,16 @@ class StubGenerator: public StubCodeGenerator {
     VectorRegister vKey4    = VR4;
     VectorRegister vKey5    = VR5;
 
-    VectorRegister fromPerm = VR6;
-    VectorRegister keyPerm  = VR7;
-    VectorRegister toPerm   = VR8;
-    VectorRegister fSplt    = VR9;
-
-    VectorRegister vTmp1    = VR10;
-    VectorRegister vTmp2    = VR11;
-    VectorRegister vTmp3    = VR12;
-    VectorRegister vTmp4    = VR13;
+    VectorRegister vp       = VR6;  // permute vector for byte vector accesses on P8 LE
 
-    __ li              (fifteen, 15);
+    __ compute_vp_for_byte_vector_unaligned(vp, /*temp*/ vRet);
 
     // load unaligned from[0-15] to vRet
-    __ lvx             (vRet, from);
-    __ lvx             (vTmp1, fifteen, from);
-    __ lvsl            (fromPerm, from);
-#ifdef VM_LITTLE_ENDIAN
-    __ vspltisb        (fSplt, 0x0f);
-    __ vxor            (fromPerm, fromPerm, fSplt);
-#endif
-    __ vperm           (vRet, vRet, vTmp1, fromPerm); // align [and byte swap in LE]
+    __ load_byte_vector_unaligned(vRet, 0, from, tmp, vp);
 
     // load keylen (44 or 52 or 60)
     __ lwz             (keylen, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT), key);
 
-    // to load keys
-    __ load_perm       (keyPerm, key);
-#ifdef VM_LITTLE_ENDIAN
-    __ vxor            (vTmp2, vTmp2, vTmp2);
-    __ vspltisb        (vTmp2, -16);
-    __ vrld            (keyPerm, keyPerm, vTmp2);
-    __ vrld            (keyPerm, keyPerm, vTmp2);
-    __ vsldoi          (keyPerm, keyPerm, keyPerm, 8);
-#endif
-
     __ cmpwi           (CCR0, keylen, 44);
     __ beq             (CCR0, L_do44);
 
@@ -3076,32 +2959,12 @@ class StubGenerator: public StubCodeGenerator {
     __ bne             (CCR0, L_error);
 #endif
 
-    // load the 15th round key to vKey1
-    __ li              (keypos, 240);
-    __ lvx             (vKey1, keypos, key);
-    __ li              (keypos, 224);
-    __ lvx             (vKey2, keypos, key);
-    __ vec_perm        (vKey1, vKey2, vKey1, keyPerm);
-
-    // load the 14th round key to vKey2
-    __ li              (keypos, 208);
-    __ lvx             (vKey3, keypos, key);
-    __ vec_perm        (vKey2, vKey3, vKey2, keyPerm);
-
-    // load the 13th round key to vKey3
-    __ li              (keypos, 192);
-    __ lvx             (vKey4, keypos, key);
-    __ vec_perm        (vKey3, vKey4, vKey3, keyPerm);
-
-    // load the 12th round key to vKey4
-    __ li              (keypos, 176);
-    __ lvx             (vKey5, keypos, key);
-    __ vec_perm        (vKey4, vKey5, vKey4, keyPerm);
-
-    // load the 11th round key to vKey5
-    __ li              (keypos, 160);
-    __ lvx             (vTmp1, keypos, key);
-    __ vec_perm        (vKey5, vTmp1, vKey5, keyPerm);
+    // load the 15th - 11th round key to vKey1 - vKey5
+    __ load_word_vector_unaligned(vKey1, 224, key, tmp);
+    __ load_word_vector_unaligned(vKey2, 208, key, tmp);
+    __ load_word_vector_unaligned(vKey3, 192, key, tmp);
+    __ load_word_vector_unaligned(vKey4, 176, key, tmp);
+    __ load_word_vector_unaligned(vKey5, 160, key, tmp);
 
     // 1st - 5th rounds
     __ vxor            (vRet, vRet, vKey1);
@@ -3115,22 +2978,10 @@ class StubGenerator: public StubCodeGenerator {
     __ align(32);
     __ bind            (L_do52);
 
-    // load the 13th round key to vKey1
-    __ li              (keypos, 208);
-    __ lvx             (vKey1, keypos, key);
-    __ li              (keypos, 192);
-    __ lvx             (vKey2, keypos, key);
-    __ vec_perm        (vKey1, vKey2, vKey1, keyPerm);
-
-    // load the 12th round key to vKey2
-    __ li              (keypos, 176);
-    __ lvx             (vKey3, keypos, key);
-    __ vec_perm        (vKey2, vKey3, vKey2, keyPerm);
-
-    // load the 11th round key to vKey3
-    __ li              (keypos, 160);
-    __ lvx             (vTmp1, keypos, key);
-    __ vec_perm        (vKey3, vTmp1, vKey3, keyPerm);
+    // load the 13th - 11th round key to vKey1 - vKey3
+    __ load_word_vector_unaligned(vKey1, 192, key, tmp);
+    __ load_word_vector_unaligned(vKey2, 176, key, tmp);
+    __ load_word_vector_unaligned(vKey3, 160, key, tmp);
 
     // 1st - 3rd rounds
     __ vxor            (vRet, vRet, vKey1);
@@ -3143,41 +2994,19 @@ class StubGenerator: public StubCodeGenerator {
     __ bind            (L_do44);
 
     // load the 11th round key to vKey1
-    __ li              (keypos, 176);
-    __ lvx             (vKey1, keypos, key);
-    __ li              (keypos, 160);
-    __ lvx             (vTmp1, keypos, key);
-    __ vec_perm        (vKey1, vTmp1, vKey1, keyPerm);
+    __ load_word_vector_unaligned(vKey1, 160, key, tmp);
 
     // 1st round
     __ vxor            (vRet, vRet, vKey1);
 
     __ bind            (L_doLast);
 
-    // load the 10th round key to vKey1
-    __ li              (keypos, 144);
-    __ lvx             (vKey2, keypos, key);
-    __ vec_perm        (vKey1, vKey2, vTmp1, keyPerm);
-
-    // load the 9th round key to vKey2
-    __ li              (keypos, 128);
-    __ lvx             (vKey3, keypos, key);
-    __ vec_perm        (vKey2, vKey3, vKey2, keyPerm);
-
-    // load the 8th round key to vKey3
-    __ li              (keypos, 112);
-    __ lvx             (vKey4, keypos, key);
-    __ vec_perm        (vKey3, vKey4, vKey3, keyPerm);
-
-    // load the 7th round key to vKey4
-    __ li              (keypos, 96);
-    __ lvx             (vKey5, keypos, key);
-    __ vec_perm        (vKey4, vKey5, vKey4, keyPerm);
-
-    // load the 6th round key to vKey5
-    __ li              (keypos, 80);
-    __ lvx             (vTmp1, keypos, key);
-    __ vec_perm        (vKey5, vTmp1, vKey5, keyPerm);
+    // load the 10th - 6th round key to vKey1 - vKey5
+    __ load_word_vector_unaligned(vKey1, 144, key, tmp);
+    __ load_word_vector_unaligned(vKey2, 128, key, tmp);
+    __ load_word_vector_unaligned(vKey3, 112, key, tmp);
+    __ load_word_vector_unaligned(vKey4, 96, key, tmp);
+    __ load_word_vector_unaligned(vKey5, 80, key, tmp);
 
     // last 10th - 6th rounds
     __ vncipher        (vRet, vRet, vKey1);
@@ -3186,29 +3015,12 @@ class StubGenerator: public StubCodeGenerator {
     __ vncipher        (vRet, vRet, vKey4);
     __ vncipher        (vRet, vRet, vKey5);
 
-    // load the 5th round key to vKey1
-    __ li              (keypos, 64);
-    __ lvx             (vKey2, keypos, key);
-    __ vec_perm        (vKey1, vKey2, vTmp1, keyPerm);
-
-    // load the 4th round key to vKey2
-    __ li              (keypos, 48);
-    __ lvx             (vKey3, keypos, key);
-    __ vec_perm        (vKey2, vKey3, vKey2, keyPerm);
-
-    // load the 3rd round key to vKey3
-    __ li              (keypos, 32);
-    __ lvx             (vKey4, keypos, key);
-    __ vec_perm        (vKey3, vKey4, vKey3, keyPerm);
-
-    // load the 2nd round key to vKey4
-    __ li              (keypos, 16);
-    __ lvx             (vKey5, keypos, key);
-    __ vec_perm        (vKey4, vKey5, vKey4, keyPerm);
-
-    // load the 1st round key to vKey5
-    __ lvx             (vTmp1, key);
-    __ vec_perm        (vKey5, vTmp1, vKey5, keyPerm);
+    // load the 5th - 1st round key to vKey1 - vKey5
+    __ load_word_vector_unaligned(vKey1, 64, key, tmp);
+    __ load_word_vector_unaligned(vKey2, 48, key, tmp);
+    __ load_word_vector_unaligned(vKey3, 32, key, tmp);
+    __ load_word_vector_unaligned(vKey4, 16, key, tmp);
+    __ load_word_vector_unaligned(vKey5, 0, key, tmp);
 
     // last 5th - 1th rounds
     __ vncipher        (vRet, vRet, vKey1);
@@ -3217,23 +3029,8 @@ class StubGenerator: public StubCodeGenerator {
     __ vncipher        (vRet, vRet, vKey4);
     __ vncipherlast    (vRet, vRet, vKey5);
 
-#ifdef VM_LITTLE_ENDIAN
-    // toPerm = 0x0F0E0D0C0B0A09080706050403020100
-    __ lvsl            (toPerm, keypos); // keypos is a multiple of 16
-    __ vxor            (toPerm, toPerm, fSplt);
-
-    // Swap Bytes
-    __ vperm           (vRet, vRet, vRet, toPerm);
-#endif
-
     // store result (unaligned)
-    // Note: We can't use a read-modify-write sequence which touches additional Bytes.
-    Register lo = temp, hi = fifteen; // Reuse
-    __ vsldoi          (vTmp1, vRet, vRet, 8);
-    __ mfvrd           (hi, vRet);
-    __ mfvrd           (lo, vTmp1);
-    __ std             (hi, 0 LITTLE_ENDIAN_ONLY(+ 8), to);
-    __ std             (lo, 0 BIG_ENDIAN_ONLY(+ 8), to);
+    __ store_byte_vector_unaligned(vRet, 0, to, tmp, vp);
 
     __ blr();