crypto/ppccap.c: SIGILL-free processor capabilities detection on MacOS X.

[openssl.git] / crypto / ppccap.c

diff --git a/crypto/ppccap.c b/crypto/ppccap.c
index 712ee799d6dfd830140df627f19d5e8525baaafd..2f7cd8e6e7634429e0fc28a8935244f257ce5683 100644 (file)
--- a/crypto/ppccap.c
+++ b/crypto/ppccap.c
@@ -22,6 +22,10 @@
 #  define __power_set(a) (_system_configuration.implementation & (a))
 # endif
 #endif
+#if defined(__APPLE__) && defined(__MACH__)
+# include <sys/types.h>
+# include <sys/sysctl.h>
+#endif
 #include <openssl/crypto.h>
 #include <openssl/bn.h>
 
@@ -35,38 +39,24 @@ static sigset_t all_masked;
 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
                 const BN_ULONG *np, const BN_ULONG *n0, int num)
 {
-    int bn_mul_mont_fpu64(BN_ULONG *rp, const BN_ULONG *ap,
-                          const BN_ULONG *bp, const BN_ULONG *np,
-                          const BN_ULONG *n0, int num);
     int bn_mul_mont_int(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
                         const BN_ULONG *np, const BN_ULONG *n0, int num);
+    int bn_mul4x_mont_int(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
+                          const BN_ULONG *np, const BN_ULONG *n0, int num);
 
-    if (sizeof(size_t) == 4) {
-# if 1 || (defined(__APPLE__) && defined(__MACH__))
-        if (num >= 8 && (num & 3) == 0 && (OPENSSL_ppccap_P & PPC_FPU64))
-            return bn_mul_mont_fpu64(rp, ap, bp, np, n0, num);
-# else
-        /*
-         * boundary of 32 was experimentally determined on Linux 2.6.22,
-         * might have to be adjusted on AIX...
-         */
-        if (num >= 32 && (num & 3) == 0 && (OPENSSL_ppccap_P & PPC_FPU64)) {
-            sigset_t oset;
-            int ret;
-
-            sigprocmask(SIG_SETMASK, &all_masked, &oset);
-            ret = bn_mul_mont_fpu64(rp, ap, bp, np, n0, num);
-            sigprocmask(SIG_SETMASK, &oset, NULL);
-
-            return ret;
-        }
-# endif
-    } else if ((OPENSSL_ppccap_P & PPC_FPU64))
-        /*
-         * this is a "must" on POWER6, but run-time detection is not
-         * implemented yet...
-         */
-        return bn_mul_mont_fpu64(rp, ap, bp, np, n0, num);
+    if (num < 4)
+        return 0;
+
+    if ((num & 3) == 0)
+        return bn_mul4x_mont_int(rp, ap, bp, np, n0, num);
+
+    /*
+     * There used to be [optional] call to bn_mul_mont_fpu64 here,
+     * but above subroutine is faster on contemporary processors.
+     * Formulation means that there might be old processors where
+     * FPU code path would be faster, POWER6 perhaps, but there was
+     * no opportunity to figure it out...
+     */
 
     return bn_mul_mont_int(rp, ap, bp, np, n0, num);
 }
@@ -119,11 +109,11 @@ void poly1305_emit_fpu(void *ctx, unsigned char mac[16],
 int poly1305_init(void *ctx, const unsigned char key[16], void *func[2])
 {
     if (sizeof(size_t) == 4 && (OPENSSL_ppccap_P & PPC_FPU)) {
-        poly1305_init_fpu(ctx,key);
+        poly1305_init_fpu(ctx, key);
         func[0] = poly1305_blocks_fpu;
         func[1] = poly1305_emit_fpu;
     } else {
-        poly1305_init_int(ctx,key);
+        poly1305_init_int(ctx, key);
         func[0] = poly1305_blocks;
         func[1] = poly1305_emit;
     }
@@ -131,6 +121,30 @@ int poly1305_init(void *ctx, const unsigned char key[16], void *func[2])
 }
 #endif
 
+#ifdef ECP_NISTZ256_ASM
+void ecp_nistz256_mul_mont(unsigned long res[4], const unsigned long a[4],
+                           const unsigned long b[4]);
+
+void ecp_nistz256_to_mont(unsigned long res[4], const unsigned long in[4]);
+void ecp_nistz256_to_mont(unsigned long res[4], const unsigned long in[4])
+{
+    static const unsigned long RR[] = { 0x0000000000000003U,
+                                        0xfffffffbffffffffU,
+                                        0xfffffffffffffffeU,
+                                        0x00000004fffffffdU };
+
+    ecp_nistz256_mul_mont(res, in, RR);
+}
+
+void ecp_nistz256_from_mont(unsigned long res[4], const unsigned long in[4]);
+void ecp_nistz256_from_mont(unsigned long res[4], const unsigned long in[4])
+{
+    static const unsigned long one[] = { 1, 0, 0, 0 };
+
+    ecp_nistz256_mul_mont(res, in, one);
+}
+#endif
+
 static sigjmp_buf ill_jmp;
 static void ill_handler(int sig)
 {
@@ -230,6 +244,28 @@ void OPENSSL_cpuid_setup(void)
 # endif
 #endif
 
+#if defined(__APPLE__) && defined(__MACH__)
+    OPENSSL_ppccap_P |= PPC_FPU;
+
+    {
+        int val;
+        size_t len = sizeof(val);
+
+        if (sysctlbyname("hw.optional.64bitops", &val, &len, NULL, 0) == 0) {
+            if (val)
+                OPENSSL_ppccap_P |= PPC_FPU64;
+        }
+
+        len = sizeof(val);
+        if (sysctlbyname("hw.optional.altivec", &val, &len, NULL, 0) == 0) {
+            if (val)
+                OPENSSL_ppccap_P |= PPC_ALTIVEC;
+        }
+
+        return;
+    }
+#endif
+
     if (getauxval != NULL) {
         unsigned long hwcap = getauxval(HWCAP);