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)

25 files changed, 148 insertions, 213 deletions

diff --git a/crypto/aes/asm/aes-sparcv9.pl b/crypto/aes/asm/aes-sparcv9.pl
index 9b9a795dbf..cb17a75f24 100755
--- a/crypto/aes/asm/aes-sparcv9.pl
+++ b/crypto/aes/asm/aes-sparcv9.pl
@@ -82,7 +82,10 @@ sub _data_word()
 }
 
 $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/aes/asm/aesfx-sparcv9.pl b/crypto/aes/asm/aesfx-sparcv9.pl
index b40f3919b4..86f2c442ed 100644
--- a/crypto/aes/asm/aesfx-sparcv9.pl
+++ b/crypto/aes/asm/aesfx-sparcv9.pl
@@ -39,7 +39,10 @@ $output = pop and open STDOUT,">$output";
 my ($inp,$out,$key,$rounds,$tmp,$mask) = map("%o$_",(0..5));
 
 $code.=<<___;
-#include "sparc_arch.h"
+#ifndef __ASSEMBLER__
+# define __ASSEMBLER__ 1
+#endif
+#include "crypto/sparc_arch.h"
 
 #define LOCALS (STACK_BIAS+STACK_FRAME)
 
diff --git a/crypto/aes/asm/aest4-sparcv9.pl b/crypto/aes/asm/aest4-sparcv9.pl
index 6b7a6eb403..bb73f24915 100644
--- a/crypto/aes/asm/aest4-sparcv9.pl
+++ b/crypto/aes/asm/aest4-sparcv9.pl
@@ -91,7 +91,10 @@ $::evp=1;	# if $evp is set to 0, script generates module with
 my ($inp,$out,$key,$rounds,$tmp,$mask)=map("%o$_",(0..5));
 
 $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/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
diff --git a/crypto/camellia/asm/cmllt4-sparcv9.pl b/crypto/camellia/asm/cmllt4-sparcv9.pl
index cbb626f00b..3f71e4c2fc 100644
--- a/crypto/camellia/asm/cmllt4-sparcv9.pl
+++ b/crypto/camellia/asm/cmllt4-sparcv9.pl
@@ -66,7 +66,10 @@ $::evp=1;	# if $evp is set to 0, script generates module with
 my ($inp,$out,$key,$rounds,$tmp,$mask)=map("%o$_",(0..5));
 
 $code=<<___;
-#include "sparc_arch.h"
+#ifndef __ASSEMBLER__
+# define __ASSEMBLER__ 1
+#endif
+#include "crypto/sparc_arch.h"
 
 .text
 
diff --git a/crypto/des/asm/dest4-sparcv9.pl b/crypto/des/asm/dest4-sparcv9.pl
index eec9716074..09298583e4 100644
--- a/crypto/des/asm/dest4-sparcv9.pl
+++ b/crypto/des/asm/dest4-sparcv9.pl
@@ -37,7 +37,10 @@ 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/ec/asm/ecp_nistz256-sparcv9.pl b/crypto/ec/asm/ecp_nistz256-sparcv9.pl
index 5815896fa8..19821cdc85 100755
--- a/crypto/ec/asm/ecp_nistz256-sparcv9.pl
+++ b/crypto/ec/asm/ecp_nistz256-sparcv9.pl
@@ -34,7 +34,10 @@
 $output = pop and open STDOUT,">$output";
 
 $code.=<<___;
-#include "sparc_arch.h"
+#ifndef __ASSEMBLER__
+# define __ASSEMBLER__ 1
+#endif
+#include "crypto/sparc_arch.h"
 
 #define LOCALS	(STACK_BIAS+STACK_FRAME)
 #ifdef	__arch64__
diff --git a/crypto/evp/e_camellia.c b/crypto/evp/e_camellia.c
index b2f59847d4..4f1f4822dc 100644
--- a/crypto/evp/e_camellia.c
+++ b/crypto/evp/e_camellia.c
@@ -46,8 +46,7 @@ typedef struct {
 #if defined(AES_ASM) && (defined(__sparc) || defined(__sparc__))
 /* ---------^^^ this is not a typo, just a way to detect that
  * assembler support was in general requested... */
-# include "sparc_arch.h"
-
+# include "crypto/sparc_arch.h"
 
 static int cmll_t4_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
                             const unsigned char *iv, int enc)
diff --git a/crypto/evp/e_des.c b/crypto/evp/e_des.c
index 51a2f9d353..cd6e5af8d0 100644
--- a/crypto/evp/e_des.c
+++ b/crypto/evp/e_des.c
@@ -37,9 +37,7 @@ typedef struct {
 # if defined(AES_ASM) && (defined(__sparc) || defined(__sparc__))
 /* ----------^^^ this is not a typo, just a way to detect that
  * assembler support was in general requested... */
-#  include "sparc_arch.h"
-
-extern unsigned int OPENSSL_sparcv9cap_P[];
+#  include "crypto/sparc_arch.h"
 
 #  define SPARC_DES_CAPABLE       (OPENSSL_sparcv9cap_P[1] & CFR_DES)
 
diff --git a/crypto/evp/e_des3.c b/crypto/evp/e_des3.c
index 746709fa41..1e15918344 100644
--- a/crypto/evp/e_des3.c
+++ b/crypto/evp/e_des3.c
@@ -40,9 +40,7 @@ typedef struct {
 # if defined(AES_ASM) && (defined(__sparc) || defined(__sparc__))
 /* ---------^^^ this is not a typo, just a way to detect that
  * assembler support was in general requested... */
-#  include "sparc_arch.h"
-
-extern unsigned int OPENSSL_sparcv9cap_P[];
+#  include "crypto/sparc_arch.h"
 
 #  define SPARC_DES_CAPABLE       (OPENSSL_sparcv9cap_P[1] & CFR_DES)
 
diff --git a/crypto/md5/asm/md5-sparcv9.pl b/crypto/md5/asm/md5-sparcv9.pl
index 64e53a003f..a17d0b6bf5 100644
--- a/crypto/md5/asm/md5-sparcv9.pl
+++ b/crypto/md5/asm/md5-sparcv9.pl
@@ -203,7 +203,10 @@ ___
 }
 
 $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/modes/asm/ghash-sparcv9.pl b/crypto/modes/asm/ghash-sparcv9.pl
index c844235c0a..d3fa2a68fe 100644
--- a/crypto/modes/asm/ghash-sparcv9.pl
+++ b/crypto/modes/asm/ghash-sparcv9.pl
@@ -80,7 +80,10 @@ $inp="%i2";
 $len="%i3";
 
 $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/modes/gcm128.c b/crypto/modes/gcm128.c
index 8317fd48e4..e7e719fc0e 100644
--- a/crypto/modes/gcm128.c
+++ b/crypto/modes/gcm128.c
@@ -674,10 +674,9 @@ void gcm_ghash_v8(u64 Xi[2], const u128 Htable[16], const u8 *inp,
                   size_t len);
 #  endif
 # elif defined(__sparc__) || defined(__sparc)
-#  include "sparc_arch.h"
+#  include "crypto/sparc_arch.h"
 #  define GHASH_ASM_SPARC
 #  define GCM_FUNCREF_4BIT
-extern unsigned int OPENSSL_sparcv9cap_P[];
 void gcm_init_vis3(u128 Htable[16], const u64 Xi[2]);
 void gcm_gmult_vis3(u64 Xi[2], const u128 Htable[16]);
 void gcm_ghash_vis3(u64 Xi[2], const u128 Htable[16], const u8 *inp,
diff --git a/crypto/poly1305/asm/poly1305-sparcv9.pl b/crypto/poly1305/asm/poly1305-sparcv9.pl
index c9db450b48..b20517f57b 100755
--- a/crypto/poly1305/asm/poly1305-sparcv9.pl
+++ b/crypto/poly1305/asm/poly1305-sparcv9.pl
@@ -52,7 +52,10 @@ my ($h0,$h1,$h2,$h3, $t0,$t1,$t2)	= map("%o$_",(0..5,7));
 my ($d0,$d1,$d2,$d3)			= map("%g$_",(1..4));
 
 $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/sha/asm/sha1-sparcv9.pl b/crypto/sha/asm/sha1-sparcv9.pl
index c3b82d0cfd..bbca9ed21a 100644
--- a/crypto/sha/asm/sha1-sparcv9.pl
+++ b/crypto/sha/asm/sha1-sparcv9.pl
@@ -187,7 +187,10 @@ ___
 }
 
 $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/sha/asm/sha512-sparcv9.pl b/crypto/sha/asm/sha512-sparcv9.pl
index 1619f14387..a77ae7bed7 100644
--- a/crypto/sha/asm/sha512-sparcv9.pl
+++ b/crypto/sha/asm/sha512-sparcv9.pl
@@ -394,7 +394,10 @@ ___
 } if ($SZ==8);
 
 $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/sparc_arch.h b/crypto/sparc_arch.h
deleted file mode 100644
index f57f0e145b..0000000000
--- a/crypto/sparc_arch.h
+++ /dev/null
@@ -1,118 +0,0 @@
-/*
- * Copyright 2012-2016 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
- */
-
-#ifndef OSSL_CRYPTO_SPARC_ARCH_H
-# define OSSL_CRYPTO_SPARC_ARCH_H
-
-# define SPARCV9_TICK_PRIVILEGED (1<<0)
-# define SPARCV9_PREFER_FPU      (1<<1)
-# define SPARCV9_VIS1            (1<<2)
-# define SPARCV9_VIS2            (1<<3)/* reserved */
-# define SPARCV9_FMADD           (1<<4)
-# define SPARCV9_BLK             (1<<5)/* VIS1 block copy */
-# define SPARCV9_VIS3            (1<<6)
-# define SPARCV9_RANDOM          (1<<7)
-# define SPARCV9_64BIT_STACK     (1<<8)
-# define SPARCV9_FJAESX          (1<<9)/* Fujitsu SPARC64 X AES */
-# define SPARCV9_FJDESX          (1<<10)/* Fujitsu SPARC64 X DES, reserved */
-# define SPARCV9_FJHPCACE        (1<<11)/* Fujitsu HPC-ACE, reserved */
-# define SPARCV9_IMA             (1<<13)/* reserved */
-# define SPARCV9_VIS4            (1<<14)/* reserved */
-
-/*
- * OPENSSL_sparcv9cap_P[1] is copy of Compatibility Feature Register,
- * %asr26, SPARC-T4 and later. There is no SPARCV9_CFR bit in
- * OPENSSL_sparcv9cap_P[0], as %cfr copy is sufficient...
- */
-# define CFR_AES         0x00000001/* Supports AES opcodes */
-# define CFR_DES         0x00000002/* Supports DES opcodes */
-# define CFR_KASUMI      0x00000004/* Supports KASUMI opcodes */
-# define CFR_CAMELLIA    0x00000008/* Supports CAMELLIA opcodes */
-# define CFR_MD5         0x00000010/* Supports MD5 opcodes */
-# define CFR_SHA1        0x00000020/* Supports SHA1 opcodes */
-# define CFR_SHA256      0x00000040/* Supports SHA256 opcodes */
-# define CFR_SHA512      0x00000080/* Supports SHA512 opcodes */
-# define CFR_MPMUL       0x00000100/* Supports MPMUL opcodes */
-# define CFR_MONTMUL     0x00000200/* Supports MONTMUL opcodes */
-# define CFR_MONTSQR     0x00000400/* Supports MONTSQR opcodes */
-# define CFR_CRC32C      0x00000800/* Supports CRC32C opcodes */
-# define CFR_XMPMUL      0x00001000/* Supports XMPMUL opcodes */
-# define CFR_XMONTMUL    0x00002000/* Supports XMONTMUL opcodes */
-# define CFR_XMONTSQR    0x00004000/* Supports XMONTSQR opcodes */
-
-# if defined(OPENSSL_PIC) && !defined(__PIC__)
-#  define __PIC__
-# endif
-
-# if defined(__SUNPRO_C) && defined(__sparcv9) && !defined(__arch64__)
-#  define __arch64__
-# endif
-
-# define SPARC_PIC_THUNK(reg)    \
-        .align  32;             \
-.Lpic_thunk:                    \
-        jmp     %o7 + 8;        \
-         add    %o7, reg, reg;
-
-# define SPARC_PIC_THUNK_CALL(reg)                       \
-        sethi   %hi(_GLOBAL_OFFSET_TABLE_-4), reg;      \
-        call    .Lpic_thunk;                            \
-         or     reg, %lo(_GLOBAL_OFFSET_TABLE_+4), reg;
-
-# if 1
-#  define SPARC_SETUP_GOT_REG(reg)       SPARC_PIC_THUNK_CALL(reg)
-# else
-#  define SPARC_SETUP_GOT_REG(reg)       \
-        sethi   %hi(_GLOBAL_OFFSET_TABLE_-4), reg;      \
-        call    .+8;                                    \
-        or      reg,%lo(_GLOBAL_OFFSET_TABLE_+4), reg;  \
-        add     %o7, reg, reg
-# endif
-
-# if defined(__arch64__)
-
-#  define SPARC_LOAD_ADDRESS(SYM, reg)   \
-        setx    SYM, %o7, reg;
-#  define LDPTR          ldx
-#  define SIZE_T_CC      %xcc
-#  define STACK_FRAME    192
-#  define STACK_BIAS     2047
-#  define STACK_7thARG   (STACK_BIAS+176)
-
-# else
-
-#  define SPARC_LOAD_ADDRESS(SYM, reg)   \
-        set     SYM, reg;
-#  define LDPTR          ld
-#  define SIZE_T_CC      %icc
-#  define STACK_FRAME    112
-#  define STACK_BIAS     0
-#  define STACK_7thARG   92
-#  define SPARC_LOAD_ADDRESS_LEAF(SYM,reg,tmp) SPARC_LOAD_ADDRESS(SYM,reg)
-
-# endif
-
-# ifdef __PIC__
-#  undef SPARC_LOAD_ADDRESS
-#  undef SPARC_LOAD_ADDRESS_LEAF
-#  define SPARC_LOAD_ADDRESS(SYM, reg)   \
-        SPARC_SETUP_GOT_REG(reg);       \
-        sethi   %hi(SYM), %o7;          \
-        or      %o7, %lo(SYM), %o7;     \
-        LDPTR   [reg + %o7], reg;
-# endif
-
-# ifndef SPARC_LOAD_ADDRESS_LEAF
-#  define SPARC_LOAD_ADDRESS_LEAF(SYM, reg, tmp) \
-        mov     %o7, tmp;                       \
-        SPARC_LOAD_ADDRESS(SYM, reg)            \
-        mov     tmp, %o7;
-# endif
-
-#endif                          /* OSSL_CRYPTO_SPARC_ARCH_H */
diff --git a/crypto/sparcv9cap.c b/crypto/sparcv9cap.c
index d93429f4c3..53c0c3e045 100644
--- a/crypto/sparcv9cap.c
+++ b/crypto/sparcv9cap.c
@@ -16,78 +16,13 @@
 #include <unistd.h>
 #include <openssl/bn.h>
 #include "internal/cryptlib.h"
-#include "bn/bn_local.h"    /* for definition of bn_mul_mont */
-
-#include "sparc_arch.h"
+#include "crypto/sparc_arch.h"
 
 #if defined(__GNUC__) && defined(__linux)
 __attribute__ ((visibility("hidden")))
 #endif
 unsigned int OPENSSL_sparcv9cap_P[2] = { SPARCV9_TICK_PRIVILEGED, 0 };
 
-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);
-}
-
 unsigned long _sparcv9_rdtick(void);
 void _sparcv9_vis1_probe(void);
 unsigned long _sparcv9_vis1_instrument(void);