Fix crash at startup in SSE4.2 code when running a 32 bits Windows build

There is no guarantee that a g_malloc'ed memory block will be aligned on a 128 bits boundary
Instead use a static variable definition (at the cost of exposing the HAVE_SSE4_2 compilation flag in ws_mempbrk.h)

Change-Id: I661bf479a9d458d64c96bafc940c519d29a4780b
Reviewed-on: https://code.wireshark.org/review/7070
Petri-Dish: Pascal Quantin <pascal.quantin@gmail.com>
Tested-by: Petri Dish Buildbot <buildbot-no-reply@wireshark.org>
Reviewed-by: Anders Broman <a.broman58@gmail.com>
Reviewed-by: Pascal Quantin <pascal.quantin@gmail.com>

8 files changed, 33 insertions, 35 deletions

diff --git a/epan/dissectors/packet-cups.c b/epan/dissectors/packet-cups.c
index 255bf970c5..42359bb741 100644
--- a/epan/dissectors/packet-cups.c
+++ b/epan/dissectors/packet-cups.c
@@ -104,7 +104,7 @@ static gint ett_cups = -1;
 static gint ett_cups_ptype = -1;
 
 /* patterns used for tvb_pbrk_pattern_guint8 */
-static tvb_pbrk_pattern pbrk_whitespace = INIT_PBRK_PATTERN;
+static tvb_pbrk_pattern pbrk_whitespace;
 
 /* This protocol is heavily related to IPP, but it is CUPS-specific
    and non-standard. */
diff --git a/epan/dissectors/packet-irc.c b/epan/dissectors/packet-irc.c
index de201c235e..2d0b66f479 100644
--- a/epan/dissectors/packet-irc.c
+++ b/epan/dissectors/packet-irc.c
@@ -76,7 +76,7 @@ static expert_field ei_irc_tag_data_invalid = EI_INIT;
 /* This must be a null-terminated string */
 static const guint8 TAG_DELIMITER[] = {0x01, 0x00};
 /* patterns used for tvb_pbrk_pattern_guint8 */
-static tvb_pbrk_pattern pbrk_tag_delimiter = INIT_PBRK_PATTERN;
+static tvb_pbrk_pattern pbrk_tag_delimiter;
 
 
 #define TCP_PORT_IRC            6667
diff --git a/epan/dissectors/packet-megaco.c b/epan/dissectors/packet-megaco.c
index 84bd157f56..91af614297 100644
--- a/epan/dissectors/packet-megaco.c
+++ b/epan/dissectors/packet-megaco.c
@@ -140,8 +140,8 @@ static dissector_handle_t megaco_text_handle;
 static int megaco_tap = -1;
 
 /* patterns used for tvb_pbrk_pattern_guint8 */
-static tvb_pbrk_pattern pbrk_whitespace = INIT_PBRK_PATTERN;
-static tvb_pbrk_pattern pbrk_braces = INIT_PBRK_PATTERN;
+static tvb_pbrk_pattern pbrk_whitespace;
+static tvb_pbrk_pattern pbrk_braces;
 
 
 /*
diff --git a/epan/dissectors/packet-sdp.c b/epan/dissectors/packet-sdp.c
index cf7897dda7..166b8c30a1 100644
--- a/epan/dissectors/packet-sdp.c
+++ b/epan/dissectors/packet-sdp.c
@@ -198,8 +198,8 @@ static expert_field ei_sdp_invalid_line_space  = EI_INIT;
 static expert_field ei_sdp_invalid_conversion = EI_INIT;
 
 /* patterns used for tvb_pbrk_pattern_guint8 */
-static tvb_pbrk_pattern pbrk_digits = INIT_PBRK_PATTERN;
-static tvb_pbrk_pattern pbrk_alpha = INIT_PBRK_PATTERN;
+static tvb_pbrk_pattern pbrk_digits;
+static tvb_pbrk_pattern pbrk_alpha;
 
 #define SDP_RTP_PROTO       0x00000001
 #define SDP_SRTP_PROTO      0x00000002
diff --git a/epan/dissectors/packet-sip.c b/epan/dissectors/packet-sip.c
index f34a157295..9fa6932045 100644
--- a/epan/dissectors/packet-sip.c
+++ b/epan/dissectors/packet-sip.c
@@ -238,15 +238,15 @@ static expert_field ei_sip_odd_register_response = EI_INIT;
 static expert_field ei_sip_sipsec_malformed = EI_INIT;
 
 /* patterns used for tvb_pbrk_pattern_guint8 */
-static tvb_pbrk_pattern pbrk_comma_semi = INIT_PBRK_PATTERN;
-static tvb_pbrk_pattern pbrk_whitespace = INIT_PBRK_PATTERN;
-static tvb_pbrk_pattern pbrk_param_end = INIT_PBRK_PATTERN;
-static tvb_pbrk_pattern pbrk_param_end_colon_brackets = INIT_PBRK_PATTERN;
-static tvb_pbrk_pattern pbrk_header_end_dquote = INIT_PBRK_PATTERN;
-static tvb_pbrk_pattern pbrk_quotes = INIT_PBRK_PATTERN;
-static tvb_pbrk_pattern pbrk_tab_sp_fslash = INIT_PBRK_PATTERN;
-static tvb_pbrk_pattern pbrk_addr_end = INIT_PBRK_PATTERN;
-static tvb_pbrk_pattern pbrk_via_param_end = INIT_PBRK_PATTERN;
+static tvb_pbrk_pattern pbrk_comma_semi;
+static tvb_pbrk_pattern pbrk_whitespace;
+static tvb_pbrk_pattern pbrk_param_end;
+static tvb_pbrk_pattern pbrk_param_end_colon_brackets;
+static tvb_pbrk_pattern pbrk_header_end_dquote;
+static tvb_pbrk_pattern pbrk_quotes;
+static tvb_pbrk_pattern pbrk_tab_sp_fslash;
+static tvb_pbrk_pattern pbrk_addr_end;
+static tvb_pbrk_pattern pbrk_via_param_end;
 
 
 /* PUBLISH method added as per http://www.ietf.org/internet-drafts/draft-ietf-sip-publish-01.txt */
diff --git a/epan/tvbuff.c b/epan/tvbuff.c
index d37859acbc..4be4f1a1fa 100644
--- a/epan/tvbuff.c
+++ b/epan/tvbuff.c
@@ -3039,7 +3039,7 @@ tvb_get_nstringz0(tvbuff_t *tvb, const gint offset, const guint bufsize, guint8*
 }
 
 
-static tvb_pbrk_pattern pbrk_crlf = INIT_PBRK_PATTERN;
+static tvb_pbrk_pattern pbrk_crlf;
 /*
  * Given a tvbuff, an offset into the tvbuff, and a length that starts
  * at that offset (which may be -1 for "all the way to the end of the
@@ -3164,7 +3164,7 @@ tvb_find_line_end(tvbuff_t *tvb, const gint offset, int len, gint *next_offset,
 	return linelen;
 }
 
-static tvb_pbrk_pattern pbrk_crlf_dquote = INIT_PBRK_PATTERN;
+static tvb_pbrk_pattern pbrk_crlf_dquote;
 /*
  * Given a tvbuff, an offset into the tvbuff, and a length that starts
  * at that offset (which may be -1 for "all the way to the end of the
diff --git a/wsutil/ws_mempbrk.h b/wsutil/ws_mempbrk.h
index 708d53cfff..40c7e5595c 100644
--- a/wsutil/ws_mempbrk.h
+++ b/wsutil/ws_mempbrk.h
@@ -24,28 +24,30 @@
 
 #include "ws_symbol_export.h"
 
+#ifdef HAVE_SSE4_2
+#include <emmintrin.h>
+#endif
+
 /** The pattern object used for tvb_pbrk_pattern_guint8().
  */
 typedef struct {
     gchar patt[256];
+#ifdef HAVE_SSE4_2
     gboolean use_sse42;
-    void *mask;
+    __m128i mask;
+#endif
 } tvb_pbrk_pattern;
 
-/** The value to use when initializing a tvb_pbrk_pattern variable.
- * For example:
- *    static tvb_pbrk_pattern pbrk_mypattern = INIT_PBRK_PATTERN;
- */
-#define INIT_PBRK_PATTERN { { 0 }, FALSE, NULL }
-
 /** Compile the pattern for the needles to find using tvb_pbrk_pattern_guint8().
  */
 WS_DLL_PUBLIC void tvb_pbrk_compile(tvb_pbrk_pattern* pattern, const gchar *needles);
 
 WS_DLL_PUBLIC const guint8 *tvb_pbrk_exec(const guint8* haystack, size_t haystacklen, const tvb_pbrk_pattern* pattern, guchar *found_needle);
 
+#ifdef HAVE_SSE4_2
 void ws_mempbrk_sse42_compile(tvb_pbrk_pattern* pattern, const gchar *needles);
 const char *ws_mempbrk_sse42_exec(const char* haystack, size_t haystacklen, const tvb_pbrk_pattern* pattern, guchar *found_needle);
+#endif
 
 const guint8 *ws_mempbrk_exec(const guint8* haystack, size_t haystacklen, const tvb_pbrk_pattern* pattern, guchar *found_needle);
 
diff --git a/wsutil/ws_mempbrk_sse42.c b/wsutil/ws_mempbrk_sse42.c
index da37a665aa..4fd3ef0701 100644
--- a/wsutil/ws_mempbrk_sse42.c
+++ b/wsutil/ws_mempbrk_sse42.c
@@ -68,11 +68,8 @@ ws_mempbrk_sse42_compile(tvb_pbrk_pattern* pattern, const gchar *needles)
     pattern->use_sse42 = ws_cpuid_sse42() && (length <= 16);
 
     if (pattern->use_sse42) {
-        __m128i *pmask = NULL;
-        pattern->mask = g_malloc(sizeof(__m128i));
-        pmask = (__m128i *) pattern->mask;
-        *pmask = _mm_setzero_si128();
-        memcpy(pmask, needles, length);
+        pattern->mask = _mm_setzero_si128();
+        memcpy(&(pattern->mask), needles, length);
     }
 }
 
@@ -112,7 +109,6 @@ const char *
 ws_mempbrk_sse42_exec(const char *s, size_t slen, const tvb_pbrk_pattern* pattern, guchar *found_needle)
 {
   const char *aligned;
-  __m128i *pmask = (__m128i *) pattern->mask;
   int offset;
 
   offset = (int) ((size_t) s & 15);
@@ -122,9 +118,9 @@ ws_mempbrk_sse42_exec(const char *s, size_t slen, const tvb_pbrk_pattern* patter
       /* Check partial string. cast safe it's 16B aligned */
       __m128i value = __m128i_shift_right (_mm_load_si128 (cast_128aligned__m128i(aligned)), offset);
 
-      int length = _mm_cmpistri (*pmask, value, 0x2);
+      int length = _mm_cmpistri (pattern->mask, value, 0x2);
       /* No need to check ZFlag since ZFlag is always 1.  */
-      int cflag = _mm_cmpistrc (*pmask, value, 0x2);
+      int cflag = _mm_cmpistrc (pattern->mask, value, 0x2);
       /* XXX: why does this compare value with value? */
       int idx = _mm_cmpistri (value, value, 0x3a);
 
@@ -149,9 +145,9 @@ ws_mempbrk_sse42_exec(const char *s, size_t slen, const tvb_pbrk_pattern* patter
   while (slen >= 16)
     {
       __m128i value = _mm_load_si128 (cast_128aligned__m128i(aligned));
-      int idx = _mm_cmpistri (*pmask, value, 0x2);
-      int cflag = _mm_cmpistrc (*pmask, value, 0x2);
-      int zflag = _mm_cmpistrz (*pmask, value, 0x2);
+      int idx = _mm_cmpistri (pattern->mask, value, 0x2);
+      int cflag = _mm_cmpistrc (pattern->mask, value, 0x2);
+      int zflag = _mm_cmpistrz (pattern->mask, value, 0x2);
 
       if (cflag) {
         if (found_needle)