X-Git-Url: https://git.openssl.org/gitweb/?p=openssl.git;a=blobdiff_plain;f=crypto%2Fppccap.c;h=6a4f4a4c41e9dace194c07393bb255ebc63567ea;hp=74af4732b5fa8dc3905d74ec568657292b22bf06;hb=4e0e4d293753a67e5771c98ec312a37ca54cc3b0;hpb=1125245997dac232a0c0867b6c858cda4e549c6d

diff --git a/crypto/ppccap.c b/crypto/ppccap.c
index 74af4732b5..6a4f4a4c41 100644
--- a/crypto/ppccap.c
+++ b/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>
 
@@ -73,16 +79,88 @@ void sha512_block_data_order(void *ctx, const void *inp, size_t len)
         sha512_block_ppc(ctx, inp, len);
 }
 
+#ifndef OPENSSL_NO_CHACHA
+void ChaCha20_ctr32_int(unsigned char *out, const unsigned char *inp,
+                        size_t len, const unsigned int key[8],
+                        const unsigned int counter[4]);
+void ChaCha20_ctr32_vmx(unsigned char *out, const unsigned char *inp,
+                        size_t len, const unsigned int key[8],
+                        const unsigned int counter[4]);
+void ChaCha20_ctr32(unsigned char *out, const unsigned char *inp,
+                    size_t len, const unsigned int key[8],
+                    const unsigned int counter[4])
+{
+    OPENSSL_ppccap_P & PPC_ALTIVEC
+        ? ChaCha20_ctr32_vmx(out, inp, len, key, counter)
+        : ChaCha20_ctr32_int(out, inp, len, key, counter);
+}
+#endif
+
+#ifndef OPENSSL_NO_POLY1305
+void poly1305_init_int(void *ctx, const unsigned char key[16]);
+void poly1305_blocks(void *ctx, const unsigned char *inp, size_t len,
+                         unsigned int padbit);
+void poly1305_emit(void *ctx, unsigned char mac[16],
+                       const unsigned int nonce[4]);
+void poly1305_init_fpu(void *ctx, const unsigned char key[16]);
+void poly1305_blocks_fpu(void *ctx, const unsigned char *inp, size_t len,
+                         unsigned int padbit);
+void poly1305_emit_fpu(void *ctx, unsigned char mac[16],
+                       const unsigned int nonce[4]);
+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);
+        func[0] = poly1305_blocks_fpu;
+        func[1] = poly1305_emit_fpu;
+    } else {
+        poly1305_init_int(ctx,key);
+        func[0] = poly1305_blocks;
+        func[1] = poly1305_emit;
+    }
+    return 1;
+}
+#endif
+
 static sigjmp_buf ill_jmp;
 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);
+void OPENSSL_madd300_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)
+#define HWCAP_ARCH_3_00         (1U << 23)
+
+# if defined(__GNUC__) && __GNUC__>=2
+__attribute__ ((constructor))
+# endif
 void OPENSSL_cpuid_setup(void)
 {
     char *e;
@@ -94,16 +172,6 @@ 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 +180,8 @@ 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 +191,76 @@ 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;
+
+    if (__power_set(0xffffffffU<<17))           /* POWER9 and later */
+        OPENSSL_ppccap_P |= PPC_MADD300;
+
+    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;
+        }
+
+        if (hwcap & HWCAP_ARCH_3_00) {
+            OPENSSL_ppccap_P |= PPC_MADD300;
+        }
+
+        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 +269,24 @@ 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) {
@@ -154,6 +298,11 @@ void OPENSSL_cpuid_setup(void)
         }
     }
 
+    if (sigsetjmp(ill_jmp, 1) == 0) {
+        OPENSSL_madd300_probe();
+        OPENSSL_ppccap_P |= PPC_MADD300;
+    }
+
     sigaction(SIGILL, &ill_oact, NULL);
     sigprocmask(SIG_SETMASK, &oset, NULL);
 }