diff --git a/crypto/bn/asm/ppc64-mont.pl b/crypto/bn/asm/ppc64-mont.pl
index 985a2b27447edd89a815efce5e722753a65e8e1b..51e6bbb3397c69920b5c68134466905f208d0d05 100644
--- a/crypto/bn/asm/ppc64-mont.pl
+++ b/crypto/bn/asm/ppc64-mont.pl
@@ -56,10 +56,10 @@
 # that asynchronous signals have to be blocked upon entry to this
 # subroutine. Signal masking (and complementary unmasking) has quite
 # an impact on performance, naturally larger for shorter keys. It's
-# so severe that shorter key performance can be as low as 1/3 of
+# so severe that 512-bit key performance can be as low as 1/3 of
 # expected one. This is why this routine can be engaged for longer
 # key operations only, see crypto/ppccap.c for further details.
-# Alternative is to break dependance on upper halves on GPRs...
+# Alternative is to break dependence on upper halves on GPRs...
 # MacOS X is an exception from this and doesn't require signal
 # masking, and that's where above improvement coefficients were
 # collected.
diff --git a/crypto/ppccap.c b/crypto/ppccap.c
index 85901598ddef0f2fd0ed08ca9899fde9eadf93fe..2c3e1f2fcbe2c0da53c3780255d5fb11602bda8d 100644
--- a/crypto/ppccap.c
+++ b/crypto/ppccap.c
@@ -25,7 +25,7 @@ int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_U
 #else
 		/* boundary of 32 was experimentally determined on
 		   Linux 2.6.22, might have to be adjusted on AIX... */
-		if ((num>=32) && (OPENSSL_ppccap_P&PPC_FPU64))
+		if (num>=32 && (num&3)==0 && (OPENSSL_ppccap_P&PPC_FPU64))
 			{
 			sigset_t oset;
 			int ret;
@@ -50,6 +50,8 @@ int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_U
 static sigjmp_buf ill_jmp;
 static void ill_handler (int sig) { siglongjmp(ill_jmp,sig); }
 
+void OPENSSL_ppc64_probe(void);
+
 void OPENSSL_cpuid_setup(void)
 	{
 	char *e;