crypto/ppccap.c: add SIGILL-free processor capability detection code.

Reviewed-by: NKurt Roeckx <kurt@openssl.org>

crypto/ppc_arch.h

@@ -3,8 +3,14 @@
 
 extern unsigned int OPENSSL_ppccap_P;
 
+/*
+ * Flags' usage can appear ambiguous, because they are set rather
+ * to reflect OpenSSL performance preferences than actual processor
+ * capabilities.
+ */
 # define PPC_FPU64       (1<<0)
 # define PPC_ALTIVEC     (1<<1)
 # define PPC_CRYPTO207   (1<<2)
+# define PPC_FPU         (1<<3)
 
 #endif

crypto/ppccap.c

@@ -7,6 +7,12 @@
 #if defined(__linux) || defined(_AIX)
 # include <sys/utsname.h>
 #endif
+#if defined(_AIX53)     /* defined even on post-5.3 */
+# include <sys/systemcfg.h>
+# if !defined(__power_set)
+#  define __power_set(a) (_system_configuration.implementation & (a))
+# endif
+#endif
 #include <openssl/crypto.h>
 #include <openssl/bn.h>
 
@@ -79,10 +85,37 @@ static void ill_handler(int sig)
     siglongjmp(ill_jmp, sig);
 }
 
+void OPENSSL_fpu_probe(void);
 void OPENSSL_ppc64_probe(void);
 void OPENSSL_altivec_probe(void);
 void OPENSSL_crypto207_probe(void);
 
+/*
+ * Use a weak reference to getauxval() so we can use it if it is available
+ * but don't break the build if it is not. Note that this is *link-time*
+ * feature detection, not *run-time*. In other words if we link with
+ * symbol present, it's expected to be present even at run-time.
+ */
+#if defined(__GNUC__) && __GNUC__>=2 && defined(__ELF__)
+extern unsigned long getauxval(unsigned long type) __attribute__ ((weak));
+#else
+static unsigned long (*getauxval) (unsigned long) = NULL;
+#endif
+
+/* I wish <sys/auxv.h> was universally available */
+#define HWCAP                   16      /* AT_HWCAP */
+#define HWCAP_PPC64             (1U << 30)
+#define HWCAP_ALTIVEC           (1U << 28)
+#define HWCAP_FPU               (1U << 27)
+#define HWCAP_POWER6_EXT        (1U << 9)
+#define HWCAP_VSX               (1U << 7)
+
+#define HWCAP2                  26      /* AT_HWCAP2 */
+#define HWCAP_VEC_CRYPTO        (1U << 25)
+
+# if defined(__GNUC__) && __GNUC__>=2
+__attribute__ ((constructor))
+# endif
 void OPENSSL_cpuid_setup(void)
 {
     char *e;
@@ -94,16 +127,6 @@ void OPENSSL_cpuid_setup(void)
         return;
     trigger = 1;
 
-    sigfillset(&all_masked);
-    sigdelset(&all_masked, SIGILL);
-    sigdelset(&all_masked, SIGTRAP);
-#ifdef SIGEMT
-    sigdelset(&all_masked, SIGEMT);
-#endif
-    sigdelset(&all_masked, SIGFPE);
-    sigdelset(&all_masked, SIGBUS);
-    sigdelset(&all_masked, SIGSEGV);
-
     if ((e = getenv("OPENSSL_ppccap"))) {
         OPENSSL_ppccap_P = strtoul(e, NULL, 0);
         return;
@@ -112,6 +135,8 @@ void OPENSSL_cpuid_setup(void)
     OPENSSL_ppccap_P = 0;
 
 #if defined(_AIX)
+    OPENSSL_ppccap_P |= PPC_FPU;
+
     if (sizeof(size_t) == 4) {
         struct utsname uts;
 # if defined(_SC_AIX_KERNEL_BITMODE)
@@ -121,7 +146,69 @@ void OPENSSL_cpuid_setup(void)
         if (uname(&uts) != 0 || atoi(uts.version) < 6)
             return;
     }
+
+# if defined(__power_set)
+    /*
+     * Value used in __power_set is a single-bit 1<<n one denoting
+     * specific processor class. Incidentally 0xffffffff<<n can be
+     * used to denote specific processor and its successors.
+     */
+    if (sizeof(size_t) == 4) {
+        /* In 32-bit case PPC_FPU64 is always fastest [if option] */
+        if (__power_set(0xffffffffU<<13))       /* POWER5 and later */
+            OPENSSL_ppccap_P |= PPC_FPU64;
+    } else {
+        /* In 64-bit case PPC_FPU64 is fastest only on POWER6 */
+        if (__power_set(0x1U<<14))              /* POWER6 */
+            OPENSSL_ppccap_P |= PPC_FPU64;
+    }
+
+    if (__power_set(0xffffffffU<<14))           /* POWER6 and later */
+        OPENSSL_ppccap_P |= PPC_ALTIVEC;
+
+    if (__power_set(0xffffffffU<<16))           /* POWER8 and later */
+        OPENSSL_ppccap_P |= PPC_CRYPTO207;
+
+    return;
+# endif
+#endif
+
+    if (getauxval != NULL) {
+        unsigned long hwcap = getauxval(HWCAP);
+
+        if (hwcap & HWCAP_FPU) {
+	    OPENSSL_ppccap_P |= PPC_FPU;
+
+            if (sizeof(size_t) == 4) {
+                /* In 32-bit case PPC_FPU64 is always fastest [if option] */
+                if (hwcap & HWCAP_PPC64)
+                    OPENSSL_ppccap_P |= PPC_FPU64;
+            } else {
+                /* In 64-bit case PPC_FPU64 is fastest only on POWER6 */
+                if (hwcap & HWCAP_POWER6_EXT)
+                    OPENSSL_ppccap_P |= PPC_FPU64;
+            }
+        }
+
+        if (hwcap & HWCAP_ALTIVEC) {
+            OPENSSL_ppccap_P |= PPC_ALTIVEC;
+
+            if ((hwcap & HWCAP_VSX) && (getauxval(HWCAP2) & HWCAP_VEC_CRYPTO))
+                OPENSSL_ppccap_P |= PPC_CRYPTO207;
+        }
+
+        return;
+    }
+
+    sigfillset(&all_masked);
+    sigdelset(&all_masked, SIGILL);
+    sigdelset(&all_masked, SIGTRAP);
+#ifdef SIGEMT
+    sigdelset(&all_masked, SIGEMT);
 #endif
+    sigdelset(&all_masked, SIGFPE);
+    sigdelset(&all_masked, SIGBUS);
+    sigdelset(&all_masked, SIGSEGV);
 
     memset(&ill_act, 0, sizeof(ill_act));
     ill_act.sa_handler = ill_handler;
@@ -130,19 +217,24 @@ void OPENSSL_cpuid_setup(void)
     sigprocmask(SIG_SETMASK, &ill_act.sa_mask, &oset);
     sigaction(SIGILL, &ill_act, &ill_oact);
 
-    if (sizeof(size_t) == 4) {
+    if (sigsetjmp(ill_jmp,1) == 0) {
+        OPENSSL_fpu_probe();
+        OPENSSL_ppccap_P |= PPC_FPU;
+
+        if (sizeof(size_t) == 4) {
 #ifdef __linux
-        struct utsname uts;
-        if (uname(&uts) == 0 && strcmp(uts.machine, "ppc64") == 0)
+            struct utsname uts;
+            if (uname(&uts) == 0 && strcmp(uts.machine, "ppc64") == 0)
 #endif
-            if (sigsetjmp(ill_jmp, 1) == 0) {
-                OPENSSL_ppc64_probe();
-                OPENSSL_ppccap_P |= PPC_FPU64;
-            }
-    } else {
-        /*
-         * Wanted code detecting POWER6 CPU and setting PPC_FPU64
-         */
+                if (sigsetjmp(ill_jmp, 1) == 0) {
+                    OPENSSL_ppc64_probe();
+                    OPENSSL_ppccap_P |= PPC_FPU64;
+                }
+        } else {
+            /*
+             * Wanted code detecting POWER6 CPU and setting PPC_FPU64
+             */
+        }
     }
 
     if (sigsetjmp(ill_jmp, 1) == 0) {