Rework bn_sqr() to use bn_sqr_words().

Rework some of the squaring code so that it calls bn_sqr_words() and use
this as the integration point for assembly. Convert bn_sqr_normal() to
bn_sqr_words(), which is then used on architectures that do not provide
their own version.

This means that we resume using the assembly version of bn_sqr_words() on
i386, mips64 and powerpc, which can provide considerable performance gains.

ok tb@

2 files changed, 11 insertions, 13 deletions

diff --git a/src/lib/libcrypto/bn/arch/amd64/bn_arch.c b/src/lib/libcrypto/bn/arch/amd64/bn_arch.c
index 9ff8920ca2..e4fbb4cfc3 100644
--- a/src/lib/libcrypto/bn/arch/amd64/bn_arch.c
+++ b/src/lib/libcrypto/bn/arch/amd64/bn_arch.c
@@ -1,4 +1,4 @@
-/*      $OpenBSD: bn_arch.c,v 1.12 2025/08/14 15:29:17 jsing Exp $ */
+/*      $OpenBSD: bn_arch.c,v 1.13 2025/08/30 07:16:06 jsing Exp $ */
 /*
  * Copyright (c) 2023 Joel Sing <jsing@openbsd.org>
  *
@@ -142,16 +142,6 @@ bn_mul_comba8(BN_ULONG *rd, const BN_ULONG *ad, const BN_ULONG *bd)
 }
 #endif
 
-#ifdef HAVE_BN_SQR
-int
-bn_sqr(BIGNUM *r, const BIGNUM *a, int r_len, BN_CTX *ctx)
-{
-        bignum_sqr(r_len, (uint64_t *)r->d, a->top, (const uint64_t *)a->d);
-
-        return 1;
-}
-#endif
-
 #ifdef HAVE_BN_SQR_COMBA4
 void
 bn_sqr_comba4(BN_ULONG *rd, const BN_ULONG *ad)
@@ -191,6 +181,14 @@ bn_sqr_comba8(BN_ULONG *rd, const BN_ULONG *ad)
 }
 #endif
 
+#ifdef HAVE_BN_SQR_WORDS
+void
+bn_sqr_words(BN_ULONG *rd, const BN_ULONG *ad, int a_len)
+{
+        bignum_sqr(a_len * 2, (uint64_t *)rd, a_len, (const uint64_t *)ad);
+}
+#endif
+
 #ifdef HAVE_BN_WORD_CLZ
 int
 bn_word_clz(BN_ULONG w)
diff --git a/src/lib/libcrypto/bn/arch/amd64/bn_arch.h b/src/lib/libcrypto/bn/arch/amd64/bn_arch.h
index 7359f993a7..dd7abd3002 100644
--- a/src/lib/libcrypto/bn/arch/amd64/bn_arch.h
+++ b/src/lib/libcrypto/bn/arch/amd64/bn_arch.h
@@ -1,4 +1,4 @@
-/*      $OpenBSD: bn_arch.h,v 1.16 2025/08/14 15:22:54 jsing Exp $ */
+/*      $OpenBSD: bn_arch.h,v 1.17 2025/08/30 07:16:06 jsing Exp $ */
 /*
  * Copyright (c) 2023 Joel Sing <jsing@openbsd.org>
  *
@@ -36,10 +36,10 @@
 #define HAVE_BN_MUL_COMBA8
 #define HAVE_BN_MUL_WORDS
 
-#define HAVE_BN_SQR
 #define HAVE_BN_SQR_COMBA4
 #define HAVE_BN_SQR_COMBA6
 #define HAVE_BN_SQR_COMBA8
+#define HAVE_BN_SQR_WORDS
 
 #define HAVE_BN_SUB
 #define HAVE_BN_SUB_WORDS