Split bignum code out of the sparcv9cap.c

Fixes #15978

Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16019)

8 files changed, 99 insertions, 8 deletions

diff --git a/crypto/bn/asm/sparct4-mont.pl b/crypto/bn/asm/sparct4-mont.pl
index 2df872b4a8..6bc98530ce 100755
--- a/crypto/bn/asm/sparct4-mont.pl
+++ b/crypto/bn/asm/sparct4-mont.pl
@@ -86,7 +86,10 @@ require "sparcv9_modes.pl";
 $output = pop and open STDOUT,">$output";
 
 $code.=<<___;
-#include "sparc_arch.h"
+#ifndef __ASSEMBLER__
+# define __ASSEMBLER__ 1
+#endif
+#include "crypto/sparc_arch.h"
 
 #ifdef	__arch64__
 .register	%g2,#scratch
diff --git a/crypto/bn/asm/sparcv9-gf2m.pl b/crypto/bn/asm/sparcv9-gf2m.pl
index 41ab8039cc..28d1ed7426 100644
--- a/crypto/bn/asm/sparcv9-gf2m.pl
+++ b/crypto/bn/asm/sparcv9-gf2m.pl
@@ -38,7 +38,10 @@ $tab="%l0";
 ($lo,$hi,$b)=("%g1",$a8,"%o7"); $a=$lo;
 
 $code.=<<___;
-#include <sparc_arch.h>
+#ifndef __ASSEMBLER__
+# define __ASSEMBLER__ 1
+#endif
+#include "crypto/sparc_arch.h"
 
 #ifdef __arch64__
 .register	%g2,#scratch
diff --git a/crypto/bn/asm/sparcv9-mont.pl b/crypto/bn/asm/sparcv9-mont.pl
index 6005334790..ac93ef9ee6 100644
--- a/crypto/bn/asm/sparcv9-mont.pl
+++ b/crypto/bn/asm/sparcv9-mont.pl
@@ -83,7 +83,10 @@ $tpj="%l7";
 $fname="bn_mul_mont_int";
 
 $code=<<___;
-#include "sparc_arch.h"
+#ifndef __ASSEMBLER__
+# define __ASSEMBLER__ 1
+#endif
+#include "crypto/sparc_arch.h"
 
 .section	".text",#alloc,#execinstr
 
diff --git a/crypto/bn/asm/sparcv9a-mont.pl b/crypto/bn/asm/sparcv9a-mont.pl
index 2a4255ed69..062a017072 100755
--- a/crypto/bn/asm/sparcv9a-mont.pl
+++ b/crypto/bn/asm/sparcv9a-mont.pl
@@ -126,7 +126,10 @@ $nhia="%f56"; $nhib="%f58"; $nhic="%f60"; $nhid="%f62";
 $ASI_FL16_P=0xD2;	# magic ASI value to engage 16-bit FP load
 
 $code=<<___;
-#include "sparc_arch.h"
+#ifndef __ASSEMBLER__
+# define __ASSEMBLER__ 1
+#endif
+#include "crypto/sparc_arch.h"
 
 .section	".text",#alloc,#execinstr
 
diff --git a/crypto/bn/asm/vis3-mont.pl b/crypto/bn/asm/vis3-mont.pl
index 6ee64e28f6..6649c34727 100644
--- a/crypto/bn/asm/vis3-mont.pl
+++ b/crypto/bn/asm/vis3-mont.pl
@@ -31,7 +31,10 @@ $frame = "STACK_FRAME";
 $bias = "STACK_BIAS";
 
 $code.=<<___;
-#include "sparc_arch.h"
+#ifndef __ASSEMBLER__
+# define __ASSEMBLER__ 1
+#endif
+#include "crypto/sparc_arch.h"
 
 #ifdef	__arch64__
 .register	%g2,#scratch
diff --git a/crypto/bn/bn_exp.c b/crypto/bn/bn_exp.c
index 2ee85aa0cf..5329cd12a9 100644
--- a/crypto/bn/bn_exp.c
+++ b/crypto/bn/bn_exp.c
@@ -29,8 +29,7 @@
 
 #undef SPARC_T4_MONT
 #if defined(OPENSSL_BN_ASM_MONT) && (defined(__sparc__) || defined(__sparc))
-# include "sparc_arch.h"
-extern unsigned int OPENSSL_sparcv9cap_P[];
+# include "crypto/sparc_arch.h"
 # define SPARC_T4_MONT
 #endif
 
diff --git a/crypto/bn/bn_sparc.c b/crypto/bn/bn_sparc.c
new file mode 100644
index 0000000000..a810c3b1fa
--- /dev/null
+++ b/crypto/bn/bn_sparc.c
@@ -0,0 +1,77 @@
+/*
+ * Copyright 2005-2021 The OpenSSL Project Authors. All Rights Reserved.
+ *
+ * Licensed under the Apache License 2.0 (the "License").  You may not use
+ * this file except in compliance with the License.  You can obtain a copy
+ * in the file LICENSE in the source distribution or at
+ * https://www.openssl.org/source/license.html
+ */
+
+#include <stdlib.h>
+#include <openssl/bn.h>
+#include "internal/cryptlib.h"
+#include "crypto/sparc_arch.h"
+#include "bn_local.h"    /* for definition of bn_mul_mont */
+
+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_vis3(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_fpu(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);
+
+    if (!(num & 1) && num >= 6) {
+        if ((num & 15) == 0 && num <= 64 &&
+            (OPENSSL_sparcv9cap_P[1] & (CFR_MONTMUL | CFR_MONTSQR)) ==
+            (CFR_MONTMUL | CFR_MONTSQR)) {
+            typedef int (*bn_mul_mont_f) (BN_ULONG *rp, const BN_ULONG *ap,
+                                          const BN_ULONG *bp,
+                                          const BN_ULONG *np,
+                                          const BN_ULONG *n0);
+            int bn_mul_mont_t4_8(BN_ULONG *rp, const BN_ULONG *ap,
+                                 const BN_ULONG *bp, const BN_ULONG *np,
+                                 const BN_ULONG *n0);
+            int bn_mul_mont_t4_16(BN_ULONG *rp, const BN_ULONG *ap,
+                                  const BN_ULONG *bp, const BN_ULONG *np,
+                                  const BN_ULONG *n0);
+            int bn_mul_mont_t4_24(BN_ULONG *rp, const BN_ULONG *ap,
+                                  const BN_ULONG *bp, const BN_ULONG *np,
+                                  const BN_ULONG *n0);
+            int bn_mul_mont_t4_32(BN_ULONG *rp, const BN_ULONG *ap,
+                                  const BN_ULONG *bp, const BN_ULONG *np,
+                                  const BN_ULONG *n0);
+            static const bn_mul_mont_f funcs[4] = {
+                bn_mul_mont_t4_8, bn_mul_mont_t4_16,
+                bn_mul_mont_t4_24, bn_mul_mont_t4_32
+            };
+            bn_mul_mont_f worker = funcs[num / 16 - 1];
+
+            if ((*worker) (rp, ap, bp, np, n0))
+                return 1;
+            /* retry once and fall back */
+            if ((*worker) (rp, ap, bp, np, n0))
+                return 1;
+            return bn_mul_mont_vis3(rp, ap, bp, np, n0, num);
+        }
+        if ((OPENSSL_sparcv9cap_P[0] & SPARCV9_VIS3))
+            return bn_mul_mont_vis3(rp, ap, bp, np, n0, num);
+        else if (num >= 8 &&
+                 /*
+                  * bn_mul_mont_fpu doesn't use FMADD, we just use the
+                  * flag to detect when FPU path is preferable in cases
+                  * when current heuristics is unreliable. [it works
+                  * out because FMADD-capable processors where FPU
+                  * code path is undesirable are also VIS3-capable and
+                  * VIS3 code path takes precedence.]
+                  */
+                 ( (OPENSSL_sparcv9cap_P[0] & SPARCV9_FMADD) ||
+                   (OPENSSL_sparcv9cap_P[0] &
+                    (SPARCV9_PREFER_FPU | SPARCV9_VIS1)) ==
+                   (SPARCV9_PREFER_FPU | SPARCV9_VIS1) ))
+            return bn_mul_mont_fpu(rp, ap, bp, np, n0, num);
+    }
+    return bn_mul_mont_int(rp, ap, bp, np, n0, num);
+}
diff --git a/crypto/bn/build.info b/crypto/bn/build.info
index 40d7db9f50..9330274aef 100644
--- a/crypto/bn/build.info
+++ b/crypto/bn/build.info
@@ -40,7 +40,7 @@ IF[{- !$disabled{asm} -}]
   ENDIF
 
   $BNASM_sparcv9=asm/sparcv8plus.S sparcv9-mont.S sparcv9a-mont.S vis3-mont.S \
-          sparct4-mont.S
+          sparct4-mont.S bn_sparc.c
   $BNDEF_sparcv9=OPENSSL_BN_ASM_MONT
   $BNASM_sparcv9_ec2m=sparcv9-gf2m.S
   $BNDEF_sparcv9_ec2m=OPENSSL_BN_ASM_GF2m