x86_64/i386 and illumos/solaris/SunOS compatibility fixes.

* SunOS linker doesn't support --gc-sections
 * Add portable_endian.h with __sun version of htole64 and le64toh
   * Replace portable endian code in shake.c with inclusion of portable_endian.h
   * Replace portable endian code in word.h with inclusion of portable_endian.h
 * Add explicit extern reference to word.h for posix_memalign when __sun defined
 * Replace references to u_int*_t with uint*_t
 * rdtsc call in shake.c was only working on 32-bit i386
 * rdtsc call in bench_decaf.cxx was inaccurate on 64-bit x86_64 when clang absent
 * Fix two signed/unsigned comparison errors in test_decaf.cxx

Makefile

@@ -237,6 +237,9 @@ $(BUILD_LIB)/libdecaf.so.1: $(LIBCOMPONENTS)
 ifeq ($(UNAME),Darwin)
 	libtool -macosx_version_min $(MACOSX_VERSION_MIN) -dynamic -dead_strip -lc -x -o $@ \
 		  $(LIBCOMPONENTS)
+else ifeq ($(UNAME),SunOS)
+	$(LD) $(LDFLAGS) -shared -Wl,-soname,`basename $@` -o $@ $(LIBCOMPONENTS)
+	strip --discard-all $@
 else
 	$(LD) $(LDFLAGS) -shared -Wl,-soname,`basename $@` -Wl,--gc-sections -o $@ $(LIBCOMPONENTS)
 	strip --discard-all $@

src/include/portable_endian.h

@@ -0,0 +1,37 @@
+/* Subset of Mathias Panzenböck's portable endian code, public domain */
+
+#ifndef __PORTABLE_ENDIAN_H__
+#define __PORTABLE_ENDIAN_H__
+
+#if defined(__linux__) || defined(__CYGWIN__)
+#	include <endian.h>
+#elif defined(__OpenBSD__)
+#	include <sys/endian.h>
+#elif defined(__APPLE__)
+#	include <libkern/OSByteOrder.h>
+#	define htole64(x) OSSwapHostToLittleInt64(x)
+#	define le64toh(x) OSSwapLittleToHostInt64(x)
+#elif defined(__NetBSD__) || defined(__FreeBSD__) || defined(__DragonFly__)
+#	include <sys/endian.h>
+#	define le64toh(x) letoh64(x)
+#elif defined(__sun) && defined(__SVR4)
+#	include <sys/byteorder.h>
+#	define htole64(x) LE_64(x)
+#	define le64toh(x) LE_64(x)
+#elif defined(_WIN16) || defined(_WIN32) || defined(_WIN64) || defined(__WINDOWS__)
+#	include <winsock2.h>
+#	include <sys/param.h>
+#	if BYTE_ORDER == LITTLE_ENDIAN
+#		define htole64(x) (x)
+#		define le64toh(x) (x)
+#	elif BYTE_ORDER == BIG_ENDIAN
+#		define htole64(x) __builtin_bswap64(x)
+#		define le64toh(x) __builtin_bswap64(x)
+#	else
+#		error byte order not supported
+#	endif
+#else
+#	error platform not supported
+#endif
+
+#endif // __PORTABLE_ENDIAN_H__

src/include/word.h

@@ -9,6 +9,9 @@
 #define _XOPEN_SOURCE 600
 #define __STDC_WANT_LIB_EXT1__ 1 /* for memset_s */
 #include <string.h>
+#if defined(__sun) && defined(__SVR4)
+extern int posix_memalign(void **, size_t, size_t);
+#endif
 
 #include <assert.h>
 #include <stdint.h>
@@ -16,13 +19,11 @@
 
 #include <decaf/common.h>
 
-
-#ifndef __APPLE__
 #ifndef _BSD_SOURCE
 #define _BSD_SOURCE 1
 #endif
-#include <endian.h>
+#include "portable_endian.h"
-#endif
 
 #include <stdlib.h>
 #include <sys/types.h>
@@ -170,12 +171,6 @@ typedef struct {
     #define br_is_zero word_is_zero
 #endif
 
-
-#ifdef __APPLE__
-    static INLINE uint64_t htole64 (uint64_t x) { return x; }
-    static INLINE uint64_t letoh64 (uint64_t x) { return x; }
-#endif
-
 /**
  * Really call memset, in a way that prevents the compiler from optimizing it out.
  * @param p The object to zeroize.

src/shake.c

@@ -23,33 +23,7 @@
 #include <fcntl.h>
 #include <unistd.h>
 
-/* Subset of Mathias Panzenböck's portable endian code, public domain */
+#include "portable_endian.h"
-#if defined(__linux__) || defined(__CYGWIN__)
-#	include <endian.h>
-#elif defined(__OpenBSD__)
-#	include <sys/endian.h>
-#elif defined(__APPLE__)
-#	include <libkern/OSByteOrder.h>
-#	define htole64(x) OSSwapHostToLittleInt64(x)
-#	define le64toh(x) OSSwapLittleToHostInt64(x)
-#elif defined(__NetBSD__) || defined(__FreeBSD__) || defined(__DragonFly__)
-#	include <sys/endian.h>
-#	define le64toh(x) letoh64(x)
-#elif defined(_WIN16) || defined(_WIN32) || defined(_WIN64) || defined(__WINDOWS__)
-#	include <winsock2.h>
-#	include <sys/param.h>
-#	if BYTE_ORDER == LITTLE_ENDIAN
-#		define htole64(x) (x)
-#		define le64toh(x) (x)
-#	elif BYTE_ORDER == BIG_ENDIAN
-#		define htole64(x) __builtin_bswap64(x)
-#		define le64toh(x) __builtin_bswap64(x)
-#	else
-#		error byte order not supported
-#	endif
-#else
-#	error platform not supported
-#endif
 
 /* The internal, non-opaque definition of the decaf_sponge struct. */
 typedef union {
@@ -292,7 +266,7 @@ static void get_cpu_entropy(uint8_t *entropy, size_t len) {
 # if (defined(__i386__) || defined(__x86_64__))
     static char tested = 0, have_rdrand = 0;
     if (!tested) {
-        u_int32_t a,b,c,d;
+        uint32_t a,b,c,d;
         a=1; __asm__("cpuid" : "+a"(a), "=b"(b), "=c"(c), "=d"(d));
         have_rdrand = (c>>30)&1;
         tested = 1;
@@ -314,9 +288,20 @@ static void get_cpu_entropy(uint8_t *entropy, size_t len) {
             *eo ^= out;
         }
     } else if (len>=8) {
-        uint64_t out;
+#ifndef __has_builtin
-        __asm__ __volatile__ ("rdtsc" : "=A"(out));
+#define __has_builtin(X) 0
-        *(uint64_t*) entropy ^= out;
+#endif
+#if defined(__clang__) && __has_builtin(__builtin_readcyclecounter)
+        *(uint64_t*) entropy ^= __builtin_readcyclecounter();
+#elif defined(__x86_64__)
+        uint32_t lobits, hibits;
+        __asm__ __volatile__ ("rdtsc" : "=a"(lobits), "=d"(hibits));
+        *(uint64_t*) entropy ^= (lobits | ((uint64_t)(hibits) << 32));
+#elif defined(__i386__)
+        uint64_t __value;
+        __asm__ __volatile__ ("rdtsc" : "=A"(__value));
+        *(uint64_t*) entropy ^= __value;
+#endif
     }
 
 #else

test/bench_decaf.cxx

@@ -43,11 +43,17 @@ static double now(void) {
 #define rdtsc __builtin_readcyclecounter
 #else
 static inline uint64_t rdtsc(void) {
-  u_int64_t out = 0;
+# if defined(__x86_64__)
-# if (defined(__i386__) || defined(__x86_64__))
+    uint32_t lobits, hibits;
-    __asm__ __volatile__ ("rdtsc" : "=A"(out));
+    __asm__ __volatile__ ("rdtsc" : "=a"(lobits), "=d"(hibits));
+    return (lobits | ((uint64_t)(hibits) << 32));
+# elif defined(__i386__)
+    uint64_t __value;
+    __asm__ __volatile__ ("rdtsc" : "=A"(__value));
+    return __value;
+# else
+    return 0;
 # endif
-  return out;
 }
 #endif
 

test/test_decaf.cxx

@@ -208,7 +208,7 @@ static void test_elligator() {
     SecureBuffer *alts2[NHINTS];
     bool successes2[NHINTS];
 
-    for (int i=0; i<NTESTS/10 && (i<10 || test.passing_now); i++) {
+    for (unsigned int i=0; i<NTESTS/10 && (i<10 || test.passing_now); i++) {
         size_t len =  (i % (2*Point::HASH_BYTES + 3));
         SecureBuffer b1(len);
         if (i!=Point::HASH_BYTES) rng.read(b1); /* special test case */
@@ -222,7 +222,7 @@ static void test_elligator() {
         
         
         Point s = Point::from_hash(b1), ss=s;
-        for (int j=0; j<(i&3); j++) ss = ss.debugging_torque();
+        for (unsigned int j=0; j<(i&3); j++) ss = ss.debugging_torque();
         
         ss = ss.debugging_pscale(rng);