use bn_asm_vax.S (from netbsd); test + ok by miod

use asm code for i386, except for the CBC code, because
it is not clean PIC code.
add <machime/asm.h> support to x86unix.pl
tested by: nick (on 30386), henning, djm, tedu, jmc and more;
no shlib minor crank necessary, only internal symbols changed.

18 files changed, 588 insertions, 25 deletions

diff --git a/src/lib/libcrypto/bf/asm/bf-586.pl b/src/lib/libcrypto/bf/asm/bf-586.pl
index b556642c94..b5a4760d09 100644
--- a/src/lib/libcrypto/bf/asm/bf-586.pl
+++ b/src/lib/libcrypto/bf/asm/bf-586.pl
@@ -18,7 +18,7 @@ $tmp4="edx";
 
 &BF_encrypt("BF_encrypt",1);
 &BF_encrypt("BF_decrypt",0);
-&cbc("BF_cbc_encrypt","BF_encrypt","BF_decrypt",1,4,5,3,-1,-1);
+&cbc("BF_cbc_encrypt","BF_encrypt","BF_decrypt",1,4,5,3,-1,-1) unless $main'openbsd;
 &asm_finish();
 
 sub BF_encrypt
diff --git a/src/lib/libcrypto/bn/asm/bn-586.pl b/src/lib/libcrypto/bn/asm/bn-586.pl
index 33f6125920..9a78f63be1 100644
--- a/src/lib/libcrypto/bn/asm/bn-586.pl
+++ b/src/lib/libcrypto/bn/asm/bn-586.pl
@@ -11,7 +11,7 @@ require "x86asm.pl";
 &bn_div_words("bn_div_words");
 &bn_add_words("bn_add_words");
 &bn_sub_words("bn_sub_words");
-&bn_sub_part_words("bn_sub_part_words");
+&bn_sub_part_words("bn_sub_part_words") unless $main'openbsd;
 
 &asm_finish();
 
diff --git a/src/lib/libcrypto/cast/asm/cast-586.pl b/src/lib/libcrypto/cast/asm/cast-586.pl
index 6be0bfe572..0ed55d1905 100644
--- a/src/lib/libcrypto/cast/asm/cast-586.pl
+++ b/src/lib/libcrypto/cast/asm/cast-586.pl
@@ -28,7 +28,7 @@ $S4="CAST_S_table3";
 
 &CAST_encrypt("CAST_encrypt",1);
 &CAST_encrypt("CAST_decrypt",0);
-&cbc("CAST_cbc_encrypt","CAST_encrypt","CAST_decrypt",1,4,5,3,-1,-1);
+&cbc("CAST_cbc_encrypt","CAST_encrypt","CAST_decrypt",1,4,5,3,-1,-1) unless $main'openbsd;
 
 &asm_finish();
 
diff --git a/src/lib/libcrypto/cast/c_enc.c b/src/lib/libcrypto/cast/c_enc.c
index 0fe2cffecc..e80f65b698 100644
--- a/src/lib/libcrypto/cast/c_enc.c
+++ b/src/lib/libcrypto/cast/c_enc.c
@@ -59,6 +59,7 @@
 #include <openssl/cast.h>
 #include "cast_lcl.h"
 
+#ifndef OPENBSD_CAST_ASM
 void CAST_encrypt(CAST_LONG *data, CAST_KEY *key)
         {
         register CAST_LONG l,r,*k,t;
@@ -122,6 +123,7 @@ void CAST_decrypt(CAST_LONG *data, CAST_KEY *key)
         data[1]=l&0xffffffffL;
         data[0]=r&0xffffffffL;
         }
+#endif
 
 void CAST_cbc_encrypt(const unsigned char *in, unsigned char *out, long length,
              CAST_KEY *ks, unsigned char *iv, int enc)
diff --git a/src/lib/libcrypto/des/asm/des-586.pl b/src/lib/libcrypto/des/asm/des-586.pl
index b75d3c6b3a..60d577cc8d 100644
--- a/src/lib/libcrypto/des/asm/des-586.pl
+++ b/src/lib/libcrypto/des/asm/des-586.pl
@@ -22,10 +22,14 @@ $R="esi";
 &external_label("DES_SPtrans");
 &DES_encrypt("DES_encrypt1",1);
 &DES_encrypt("DES_encrypt2",0);
-&DES_encrypt3("DES_encrypt3",1);
-&DES_encrypt3("DES_decrypt3",0);
-&cbc("DES_ncbc_encrypt","DES_encrypt1","DES_encrypt1",0,4,5,3,5,-1);
-&cbc("DES_ede3_cbc_encrypt","DES_encrypt3","DES_decrypt3",0,6,7,3,4,5);
+
+if (!$main'openbsd)
+        {
+        &DES_encrypt3("DES_encrypt3",1);
+        &DES_encrypt3("DES_decrypt3",0);
+        &cbc("DES_ncbc_encrypt","DES_encrypt1","DES_encrypt1",0,4,5,3,5,-1);
+        &cbc("DES_ede3_cbc_encrypt","DES_encrypt3","DES_decrypt3",0,6,7,3,4,5);
+        }
 
 &asm_finish();
 
diff --git a/src/lib/libcrypto/des/des_enc.c b/src/lib/libcrypto/des/des_enc.c
index 1c37ab96d3..4f09804c44 100644
--- a/src/lib/libcrypto/des/des_enc.c
+++ b/src/lib/libcrypto/des/des_enc.c
@@ -58,6 +58,7 @@
 
 #include "des_locl.h"
 
+#ifndef OPENBSD_DES_ASM
 void DES_encrypt1(DES_LONG *data, DES_key_schedule *ks, int enc)
         {
         register DES_LONG l,r,t,u;
@@ -246,6 +247,7 @@ void DES_encrypt2(DES_LONG *data, DES_key_schedule *ks, int enc)
         data[1]=ROTATE(r,3)&0xffffffffL;
         l=r=t=u=0;
         }
+#endif
 
 void DES_encrypt3(DES_LONG *data, DES_key_schedule *ks1,
                   DES_key_schedule *ks2, DES_key_schedule *ks3)
diff --git a/src/lib/libcrypto/perlasm/x86asm.pl b/src/lib/libcrypto/perlasm/x86asm.pl
index 1cb96e914a..7c675e3ced 100644
--- a/src/lib/libcrypto/perlasm/x86asm.pl
+++ b/src/lib/libcrypto/perlasm/x86asm.pl
@@ -18,9 +18,13 @@ sub main'asm_init
         ($type,$fn,$i386)=@_;
         $filename=$fn;
 
-        $elf=$cpp=$sol=$aout=$win32=$gaswin=0;
+        $elf=$cpp=$sol=$aout=$win32=$gaswin=$openbsd=0;
         if (    ($type eq "elf"))
                 { $elf=1; require "x86unix.pl"; }
+        elsif ( ($type eq "openbsd-elf"))
+                { $openbsd=$elf=1; require "x86unix.pl"; }
+        elsif ( ($type eq "openbsd-a.out"))
+                { $openbsd=1; require "x86unix.pl"; }
         elsif ( ($type eq "a.out"))
                 { $aout=1; require "x86unix.pl"; }
         elsif ( ($type eq "gaswin"))
@@ -43,6 +47,8 @@ Pick one target type from
         cpp     - format so x86unix.cpp can be used
         win32   - Windows 95/Windows NT
         win32n  - Windows 95/Windows NT NASM format
+        openbsd-elf     - OpenBSD elf
+        openbsd-a.out   - OpenBSD a.out
 EOF
                 exit(1);
                 }
diff --git a/src/lib/libcrypto/perlasm/x86unix.pl b/src/lib/libcrypto/perlasm/x86unix.pl
index 10b669bf04..53ad5f4927 100644
--- a/src/lib/libcrypto/perlasm/x86unix.pl
+++ b/src/lib/libcrypto/perlasm/x86unix.pl
@@ -15,6 +15,12 @@ sub main'asm_get_output { return(@out); }
 sub main'get_labels { return(@labels); }
 sub main'external_label { push(@labels,@_); }
 
+if ($main'openbsd)
+        {
+        $com_start='/*';
+        $com_end='*/';
+        }
+
 if ($main'cpp)
         {
         $align="ALIGN";
@@ -276,6 +282,9 @@ sub main'file
         {
         local($file)=@_;
 
+        if ($main'openbsd)
+                { push(@out,"#include <machine/asm.h>\n"); return; }
+
         local($tmp)=<<"EOF";
         .file   "$file.s"
         .version        "01.01"
@@ -291,6 +300,9 @@ sub main'function_begin
         &main'external_label($func);
         $func=$under.$func;
 
+        if ($main'openbsd)
+                { push (@out, "\nENTRY($func)\n"); goto skip; }
+
         local($tmp)=<<"EOF";
 .text
         .align $align
@@ -303,6 +315,7 @@ EOF
                 { $tmp=push(@out,"\t.def\t$func;\t.scl\t2;\t.type\t32;\t.endef\n"); }
         else    { $tmp=push(@out,"\t.type\t$func,\@function\n"); }
         push(@out,"$func:\n");
+skip:
         $tmp=<<"EOF";
         pushl   %ebp
         pushl   %ebx
@@ -321,6 +334,9 @@ sub main'function_begin_B
         &main'external_label($func);
         $func=$under.$func;
 
+        if ($main'openbsd)
+                { push(@out, "\nENTRY($func)\n"); goto skip; }
+
         local($tmp)=<<"EOF";
 .text
         .align $align
@@ -333,6 +349,7 @@ EOF
                 { $tmp=push(@out,"\t.def\t$func;\t.scl\t2;\t.type\t32;\t.endef\n"); }
         else    { push(@out,"\t.type    $func,\@function\n"); }
         push(@out,"$func:\n");
+skip:
         $stack=4;
         }
 
@@ -429,7 +446,8 @@ sub main'swtmp
 
 sub main'comment
         {
-        if ($main'elf)  # GNU and SVR4 as'es use different comment delimiters,
+        if (!$main'openbsd && $main'elf)
+                        # GNU and SVR4 as'es use different comment delimiters,
                 {       # so we just skip comments...
                 push(@out,"\n");
                 return;
@@ -460,7 +478,10 @@ sub main'set_label
                 $label{$_[0]}=".${label}${_[0]}";
                 $label++;
                 }
-        push(@out,".align $align\n") if ($_[1] != 0);
+        if ($main'openbsd)
+                { push(@out,"_ALIGN_TEXT\n") if ($_[1] != 0); }
+        else
+                { push(@out,".align $align\n") if ($_[1] != 0); }
         push(@out,"$label{$_[0]}:\n");
         }
 
@@ -569,6 +590,16 @@ sub main'picmeup
 ___
                 push(@out,$tmp);
                 }
+        elsif ($main'openbsd)
+                {
+                push(@out, "#ifdef PIC\n");
+                push(@out, "\tPIC_PROLOGUE\n");
+                &main'mov($dst,"PIC_GOT($sym)");
+                push(@out, "\tPIC_EPILOGUE\n");
+                push(@out, "#else\n");
+                &main'lea($dst,&main'DWP($sym));
+                push(@out, "#endif\n");
+                }
         elsif ($main'pic && ($main'elf || $main'aout))
                 {
                 push(@out,"\t.align\t8\n");
diff --git a/src/lib/libssl/crypto/Makefile b/src/lib/libssl/crypto/Makefile
index dead23398f..18c744d706 100644
--- a/src/lib/libssl/crypto/Makefile
+++ b/src/lib/libssl/crypto/Makefile
@@ -1,4 +1,4 @@
-# $OpenBSD: Makefile,v 1.31 2003/06/02 08:45:58 markus Exp $
+# $OpenBSD: Makefile,v 1.32 2003/11/18 12:39:05 markus Exp $
 
 LIB=    crypto
 
@@ -53,17 +53,17 @@ SRCS+=	cbc_cksm.c cbc_enc.c cfb64enc.c cfb_enc.c \
         des_old.c read2pwd.c
 SRCS+=  rc2_ecb.c rc2_skey.c rc2_cbc.c rc2cfb64.c
 SRCS+=  rc2ofb64.c                                                      
-SRCS+=  rc4_skey.c rc4_enc.c
+SRCS+=  rc4_skey.c
 #SRCS+= rc5_skey.c rc5_ecb.c rc5cfb64.c rc5cfb64.c      
 #SRCS+= rc5ofb64.c rc5_enc.c
 #SRCS+= i_cbc.c i_cfb64.c i_ofb64.c i_ecb.c             
 #SRCS+= i_skey.c                                                        
-SRCS+=  bf_skey.c bf_ecb.c bf_cfb64.c bf_ofb64.c bf_enc.c               
+SRCS+=  bf_skey.c bf_ecb.c bf_cfb64.c bf_ofb64.c
 SRCS+=  c_skey.c c_ecb.c c_cfb64.c c_ofb64.c c_enc.c    
 SRCS+=  bn_add.c bn_div.c bn_exp.c bn_lib.c
 SRCS+=  bn_mul.c bn_print.c bn_rand.c bn_shift.c
 SRCS+=  bn_word.c bn_blind.c bn_gcd.c bn_prime.c bn_err.c
-SRCS+=  bn_sqr.c bn_recp.c bn_mont.c bn_mpi.c bn_asm.c bn_mod.c
+SRCS+=  bn_sqr.c bn_recp.c bn_mont.c bn_mpi.c bn_mod.c
 SRCS+=  bn_exp2.c bn_ctx.c
 SRCS+=  bn_sqrt.c bn_kron.c
 SRCS+=  rsa_eay.c rsa_gen.c rsa_lib.c rsa_sign.c
@@ -151,7 +151,8 @@ SRCS+= ec_err.c ec_mult.c ecp_nist.c ecp_smpl.c
         ${LCRYPTO_SRC}/pkcs7 ${LCRYPTO_SRC}/x509v3 ${LCRYPTO_SRC}/pkcs12 \
         ${LCRYPTO_SRC}/comp ${LCRYPTO_SRC}/txt_db ${LCRYPTO_SRC}/md4 \
         ${LCRYPTO_SRC}/engine ${LCRYPTO_SRC}/dso ${LCRYPTO_SRC}/ui \
-        ${LCRYPTO_SRC}/ocsp ${LCRYPTO_SRC}/ec ${LCRYPTO_SRC}/aes ${LCRYPTO_SRC}
+        ${LCRYPTO_SRC}/ocsp ${LCRYPTO_SRC}/ec ${LCRYPTO_SRC}/aes ${LCRYPTO_SRC} \
+        ${.CURDIR}/arch/${MACHINE_ARCH}
 
 HDRS=\
         crypto/aes/aes.h \
@@ -260,6 +261,42 @@ des_enc.po:
         ${CC} ${CFLAGS} -O1 ${CPPFLAGS} -c ${.IMPSRC} -o $@
 .endif  
 
+.if (${MACHINE_ARCH} == "i386")
+CFLAGS+= -DMD5_ASM
+CFLAGS+= -DSHA1_ASM
+CFLAGS+= -DRMD160_ASM
+CFLAGS+= -DOPENBSD_CAST_ASM
+CFLAGS+= -DOPENBSD_DES_ASM
+SSLASM=\
+        bf bf-586 \
+        bn bn-586 \
+        bn co-586 \
+        cast cast-586 \
+        des des-586 \
+        md5 md5-586 \
+        rc4 rc4-586 \
+        ripemd rmd-586 \
+        sha sha1-586
+.for dir f in ${SSLASM}
+SRCS+=  ${f}.S
+GENERATED+=${f}.S
+${f}.S: ${LCRYPTO_SRC}/${dir}/asm/${f}.pl ${LCRYPTO_SRC}/perlasm/x86unix.pl
+        /usr/bin/perl -I${LCRYPTO_SRC}/perlasm:${LCRYPTO_SRC}/${dir}/asm \
+                ${LCRYPTO_SRC}/${dir}/asm/${f}.pl openbsd-elf 386 > ${.TARGET}
+.endfor
+SRCS+= bf_cbc.c 
+.else
+.if (${MACHINE_ARCH} == "vax")
+SRCS+= bf_enc.c
+SRCS+= bn_asm_vax.S
+SRCS+= rc4_enc.c        
+.else
+SRCS+= bf_enc.c 
+SRCS+= bn_asm.c
+SRCS+= rc4_enc.c        
+.endif
+.endif
+
 all beforedepend: ${GENERATED}
 
 .include <bsd.lib.mk>
diff --git a/src/lib/libssl/crypto/arch/vax/bn_asm_vax.S b/src/lib/libssl/crypto/arch/vax/bn_asm_vax.S
new file mode 100644
index 0000000000..bd067a55d3
--- /dev/null
+++ b/src/lib/libssl/crypto/arch/vax/bn_asm_vax.S
@@ -0,0 +1,436 @@
+#       $OpenBSD: bn_asm_vax.S,v 1.1 2003/11/18 12:39:05 markus Exp $
+#       $NetBSD: bn_asm_vax.S,v 1.1 2003/11/03 10:22:28 ragge Exp $
+
+#include <machine/asm.h>
+
+# w.j.m. 15-jan-1999
+#
+# it's magic ...
+#
+# ULONG bn_mul_add_words(ULONG r[],ULONG a[],int n,ULONG w) {
+#       ULONG c = 0;
+#       int i;
+#       for(i = 0; i < n; i++) <c,r[i]> := r[i] + c + a[i] * w ;
+#       return c;
+# }
+
+ENTRY(bn_mul_add_words,R6)
+        movl    4(ap),r2                # *r
+        movl    8(ap),r3                # *a
+        movl    12(ap),r4               # n
+        movl    16(ap),r5               # w
+        clrl    r6                      # return value ("carry")
+
+0:      emul    r5,(r3),(r2),r0 # w * a[0] + r[0] -> r0
+
+        # fixup for "negative" r[]
+        tstl    (r2)
+        bgeq    1f
+        incl    r1                      # add 1 to highword
+
+1:      # add saved carry to result
+        addl2   r6,r0
+        adwc    $0,r1
+
+        # combined fixup for "negative" w, a[]
+        tstl    r5              # if w is negative...
+        bgeq    1f
+        addl2   (r3),r1         # ...add a[0] again to highword
+1:      tstl    (r3)            # if a[0] is negative...
+        bgeq    1f
+        addl2   r5,r1           # ...add w again to highword
+1:
+        movl    r0,(r2)+        # save low word in dest & advance *r
+        addl2   $4,r3           # advance *a
+        movl    r1,r6           # high word in r6 for return value
+
+        sobgtr  r4,0b           # loop?
+
+        movl    r6,r0
+        ret
+
+#       .title  vax_bn_mul_words  unsigned multiply & add, 32*32+32=>64
+#;
+#; w.j.m. 15-jan-1999
+#;
+#; it's magic ...
+#;
+#; ULONG bn_mul_words(ULONG r[],ULONG a[],int n,ULONG w) {
+#;      ULONG c = 0;
+#;      int i;
+#;      for(i = 0; i < num; i++) <c,r[i]> := a[i] * w + c ;
+#;      return(c);
+#; }
+#
+
+ENTRY(bn_mul_words,R6)
+        movl    4(ap),r2                # *r
+        movl    8(ap),r3                # *a
+        movl    12(ap),r4               # n
+        movl    16(ap),r5               # w
+        clrl    r6                      # carry
+
+0:      emul    r5,(r3),r6,r0           # w * a[0] + carry -> r0
+
+        # fixup for "negative" carry
+        tstl    r6
+        bgeq    1f
+        incl    r1
+
+1:      # combined fixup for "negative" w, a[]
+        tstl    r5
+        bgeq    1f
+        addl2   (r3),r1
+1:      tstl    (r3)
+        bgeq    1f
+        addl2   r5,r1
+
+1:      movl    r0,(r2)+
+        addl2   $4,r3
+        movl    r1,r6
+
+        sobgtr  r4,0b
+
+        movl    r6,r0
+        ret
+
+
+
+#       .title  vax_bn_sqr_words  unsigned square, 32*32=>64
+#;
+#; w.j.m. 15-jan-1999
+#;
+#; it's magic ...
+#;
+#; void bn_sqr_words(ULONG r[],ULONG a[],int n) {
+#;      int i;
+#;      for(i = 0; i < n; i++) <r[2*i+1],r[2*i]> := a[i] * a[i] ;
+#; }
+#
+
+ENTRY(bn_sqr_words,0)
+        movl    4(ap),r2                # r
+        movl    8(ap),r3                # a
+        movl    12(ap),r4               # n
+
+0:      movl    (r3)+,r5                # r5 = a[] & advance
+
+        emul    r5,r5,$0,r0             # a[0] * a[0] + 0 -> r0
+
+        # fixup for "negative" a[]
+        tstl    r5
+        bgeq    1f
+        addl2   r5,r1
+        addl2   r5,r1
+
+1:      movq    r0,(r2)+                # store 64-bit result
+
+        sobgtr  r4,0b                   # loop
+
+        ret
+
+
+#       .title  vax_bn_div_words  unsigned divide
+#;
+#; Richard Levitte 20-Nov-2000
+#;
+#; ULONG bn_div_words(ULONG h, ULONG l, ULONG d)
+#; {
+#;      return ((ULONG)((((ULLONG)h)<<32)|l) / (ULLONG)d);
+#; }
+#;
+#; Using EDIV would be very easy, if it didn't do signed calculations.
+#; Any time any of the input numbers are signed, there are problems,
+#; usually with integer overflow, at which point it returns useless
+#; data (the quotient gets the value of l, and the remainder becomes 0).
+#;
+#; If it was just for the dividend, it would be very easy, just divide
+#; it by 2 (unsigned), do the division, multiply the resulting quotient
+#; and remainder by 2, add the bit that was dropped when dividing by 2
+#; to the remainder, and do some adjustment so the remainder doesn't
+#; end up larger than the divisor.  For some cases when the divisor is
+#; negative (from EDIV's point of view, i.e. when the highest bit is set),
+#; dividing the dividend by 2 isn't enough, and since some operations
+#; might generate integer overflows even when the dividend is divided by
+#; 4 (when the high part of the shifted down dividend ends up being exactly
+#; half of the divisor, the result is the quotient 0x80000000, which is
+#; negative...) it needs to be divided by 8.  Furthermore, the divisor needs
+#; to be divided by 2 (unsigned) as well, to avoid more problems with the sign.
+#; In this case, a little extra fiddling with the remainder is required.
+#;
+#; So, the simplest way to handle this is always to divide the dividend
+#; by 8, and to divide the divisor by 2 if it's highest bit is set.
+#; After EDIV has been used, the quotient gets multiplied by 8 if the
+#; original divisor was positive, otherwise 4.  The remainder, oddly
+#; enough, is *always* multiplied by 8.
+#; NOTE: in the case mentioned above, where the high part of the shifted
+#; down dividend ends up being exactly half the shifted down divisor, we
+#; end up with a 33 bit quotient.  That's no problem however, it usually
+#; means we have ended up with a too large remainder as well, and the
+#; problem is fixed by the last part of the algorithm (next paragraph).
+#;
+#; The routine ends with comparing the resulting remainder with the
+#; original divisor and if the remainder is larger, subtract the
+#; original divisor from it, and increase the quotient by 1.  This is
+#; done until the remainder is smaller than the divisor.
+#;
+#; The complete algorithm looks like this:
+#;
+#; d'    = d
+#; l'    = l & 7
+#; [h,l] = [h,l] >> 3
+#; [q,r] = floor([h,l] / d)     # This is the EDIV operation
+#; if (q < 0) q = -q            # I doubt this is necessary any more
+#;
+#; r'    = r >> 29
+#; if (d' >= 0)
+#;   q'  = q >> 29
+#;   q   = q << 3
+#; else
+#;   q'  = q >> 30
+#;   q   = q << 2
+#; r     = (r << 3) + l'
+#;
+#; if (d' < 0)
+#;   {
+#;     [r',r] = [r',r] - q
+#;     while ([r',r] < 0)
+#;       {
+#;         [r',r] = [r',r] + d
+#;         [q',q] = [q',q] - 1
+#;       }
+#;   }
+#;
+#; while ([r',r] >= d')
+#;   {
+#;     [r',r] = [r',r] - d'
+#;     [q',q] = [q',q] + 1
+#;   }
+#;
+#; return q
+#
+#;r2 = l, q
+#;r3 = h, r
+#;r4 = d
+#;r5 = l'
+#;r6 = r'
+#;r7 = d'
+#;r8 = q'
+#
+
+ENTRY(bn_div_words,R6|R7|R8)
+        movl    4(ap),r3                # h
+        movl    8(ap),r2                # l
+        movl    12(ap),r4               # d
+
+        bicl3   $-8,r2,r5               # l' = l & 7
+        bicl3   $7,r2,r2
+
+        bicl3   $-8,r3,r6
+        bicl3   $7,r3,r3
+
+        addl2   r6,r2
+
+        rotl    $-3,r2,r2               # l = l >> 3
+        rotl    $-3,r3,r3               # h = h >> 3
+
+        movl    r4,r7                   # d' = d
+
+        clrl    r6                      # r' = 0
+        clrl    r8                      # q' = 0
+
+        tstl    r4
+        beql    0f                      # Uh-oh, the divisor is 0...
+        bgtr    1f
+        rotl    $-1,r4,r4       # If d is negative, shift it right.
+        bicl2   $0x80000000,r4  # Since d is then a large number, the
+                                # lowest bit is insignificant
+                                # (contradict that, and I'll fix the problem!)
+1:
+        ediv    r4,r2,r2,r3             # Do the actual division
+
+        tstl    r2
+        bgeq    1f
+        mnegl   r2,r2           # if q < 0, negate it
+1:
+        tstl    r7
+        blss    1f
+        rotl    $3,r2,r2        #   q = q << 3
+        bicl3   $-8,r2,r8       #   q' gets the high bits from q
+        bicl3   $7,r2,r2
+        brb     2f
+
+1:                              # else
+        rotl    $2,r2,r2        #   q = q << 2
+        bicl3   $-4,r2,r8       #   q' gets the high bits from q
+        bicl3   $3,r2,r2
+2:
+        rotl    $3,r3,r3        # r = r << 3
+        bicl3   $-8,r3,r6       # r' gets the high bits from r
+        bicl3   $7,r3,r3
+        addl2   r5,r3           # r = r + l'
+
+        tstl    r7
+        bgeq    5f
+        bitl    $1,r7
+        beql    5f              # if d' < 0 && d' & 1
+        subl2   r2,r3           #   [r',r] = [r',r] - [q',q]
+        sbwc    r8,r6
+3:
+        bgeq    5f              #   while r < 0
+        decl    r2              #     [q',q] = [q',q] - 1
+        sbwc    $0,r8
+        addl2   r7,r3           #     [r',r] = [r',r] + d'
+        adwc    $0,r6
+        brb     3b
+
+# The return points are placed in the middle to keep a short distance from
+# all the branch points
+1:
+#       movl    r3,r1
+        movl    r2,r0
+        ret
+0:
+        movl    $-1,r0
+        ret
+5:
+        tstl    r6
+        bneq    6f
+        cmpl    r3,r7
+        blssu   1b              # while [r',r] >= d'
+6:
+        subl2   r7,r3           #   [r',r] = [r',r] - d'
+        sbwc    $0,r6
+        incl    r2              #   [q',q] = [q',q] + 1
+        adwc    $0,r8
+        brb     5b
+
+
+
+#       .title  vax_bn_add_words  unsigned add of two arrays
+#;
+#; Richard Levitte 20-Nov-2000
+#;
+#; ULONG bn_add_words(ULONG r[], ULONG a[], ULONG b[], int n) {
+#;      ULONG c = 0;
+#;      int i;
+#;      for (i = 0; i < n; i++) <c,r[i]> = a[i] + b[i] + c;
+#;      return(c);
+#; }
+#
+
+ENTRY(bn_add_words,0)
+        movl    4(ap),r2        # r
+        movl    8(ap),r3        # a
+        movl    12(ap),r4       # b
+        movl    16(ap),r5       # n
+        clrl    r0
+
+        tstl    r5
+        bleq    1f
+
+0:      movl    (r3)+,r1        # carry untouched
+        adwc    (r4)+,r1        # carry used and touched
+        movl    r1,(r2)+        # carry untouched
+        sobgtr  r5,0b           # carry untouched
+
+        adwc    $0,r0
+1:      ret
+
+#;
+#; Richard Levitte 20-Nov-2000
+#;
+#; ULONG bn_sub_words(ULONG r[], ULONG a[], ULONG b[], int n) {
+#;      ULONG c = 0;
+#;      int i;
+#;      for (i = 0; i < n; i++) <c,r[i]> = a[i] - b[i] - c;
+#;      return(c);
+#; }
+#
+
+ENTRY(bn_sub_words,R6)
+        movl    4(ap),r2        # r
+        movl    8(ap),r3        # a
+        movl    12(ap),r4       # b
+        movl    16(ap),r5       # n
+        clrl    r0
+
+        tstl    r5
+        bleq    1f
+
+0:      movl    (r3)+,r6        # carry untouched
+        sbwc    (r4)+,r6        # carry used and touched
+        movl    r6,(r2)+        # carry untouched
+        sobgtr  r5,0b           # carry untouched
+
+1:      adwc    $0,r0
+        ret
+
+#
+#       Ragge 20-Sep-2003
+#
+#       Multiply a vector of 4/8 longword by another.
+#       Uses two loops and 16/64 emuls.
+#
+
+ENTRY(bn_mul_comba4,R6|R7|R8|R9)
+        movl    $4,r9           # 4*4
+        brb     6f
+
+ENTRY(bn_mul_comba8,R6|R7|R8|R9)
+        movl    $8,r9           # 8*8
+
+6:      movl    8(ap),r3        # a[]
+        movl    12(ap),r7       # b[]
+        brb     5f
+
+ENTRY(bn_sqr_comba4,R6|R7|R8|R9)
+        movl    $4,r9           # 4*4
+        brb 0f
+
+ENTRY(bn_sqr_comba8,R6|R7|R8|R9)
+        movl    $8,r9           # 8*8
+
+0:
+        movl    8(ap),r3        # a[]
+        movl    r3,r7           # a[]
+
+5:      movl    4(ap),r5        # r[]
+        movl    r9,r8
+
+        clrq    (r5)            # clear destinatino, for add.
+        clrq    8(r5)
+        clrq    16(r5)          # these only needed for comba8
+        clrq    24(r5)
+
+2:      clrl    r4              # carry
+        movl    r9,r6           # inner loop count
+        movl    (r7)+,r2        # value to multiply with
+
+1:      emul    r2,(r3),r4,r0
+        tstl    r4
+        bgeq    3f
+        incl    r1
+3:      tstl    r2
+        bgeq    3f
+        addl2   (r3),r1
+3:      tstl    (r3)
+        bgeq    3f
+        addl2   r2,r1
+
+3:      addl2   r0,(r5)+        # add to destination
+        adwc    $0,r1           # remember carry
+        movl    r1,r4           # add carry in next emul
+        addl2   $4,r3
+        sobgtr  r6,1b
+
+        movl    r4,(r5)         # save highest add result
+
+        ashl    $2,r9,r4
+        subl2   r4,r3
+        subl2   $4,r4
+        subl2   r4,r5
+
+        sobgtr  r8,2b
+
+        ret
diff --git a/src/lib/libssl/src/crypto/bf/asm/bf-586.pl b/src/lib/libssl/src/crypto/bf/asm/bf-586.pl
index b556642c94..b5a4760d09 100644
--- a/src/lib/libssl/src/crypto/bf/asm/bf-586.pl
+++ b/src/lib/libssl/src/crypto/bf/asm/bf-586.pl
@@ -18,7 +18,7 @@ $tmp4="edx";
 
 &BF_encrypt("BF_encrypt",1);
 &BF_encrypt("BF_decrypt",0);
-&cbc("BF_cbc_encrypt","BF_encrypt","BF_decrypt",1,4,5,3,-1,-1);
+&cbc("BF_cbc_encrypt","BF_encrypt","BF_decrypt",1,4,5,3,-1,-1) unless $main'openbsd;
 &asm_finish();
 
 sub BF_encrypt
diff --git a/src/lib/libssl/src/crypto/bn/asm/bn-586.pl b/src/lib/libssl/src/crypto/bn/asm/bn-586.pl
index 33f6125920..9a78f63be1 100644
--- a/src/lib/libssl/src/crypto/bn/asm/bn-586.pl
+++ b/src/lib/libssl/src/crypto/bn/asm/bn-586.pl
@@ -11,7 +11,7 @@ require "x86asm.pl";
 &bn_div_words("bn_div_words");
 &bn_add_words("bn_add_words");
 &bn_sub_words("bn_sub_words");
-&bn_sub_part_words("bn_sub_part_words");
+&bn_sub_part_words("bn_sub_part_words") unless $main'openbsd;
 
 &asm_finish();
 
diff --git a/src/lib/libssl/src/crypto/cast/asm/cast-586.pl b/src/lib/libssl/src/crypto/cast/asm/cast-586.pl
index 6be0bfe572..0ed55d1905 100644
--- a/src/lib/libssl/src/crypto/cast/asm/cast-586.pl
+++ b/src/lib/libssl/src/crypto/cast/asm/cast-586.pl
@@ -28,7 +28,7 @@ $S4="CAST_S_table3";
 
 &CAST_encrypt("CAST_encrypt",1);
 &CAST_encrypt("CAST_decrypt",0);
-&cbc("CAST_cbc_encrypt","CAST_encrypt","CAST_decrypt",1,4,5,3,-1,-1);
+&cbc("CAST_cbc_encrypt","CAST_encrypt","CAST_decrypt",1,4,5,3,-1,-1) unless $main'openbsd;
 
 &asm_finish();
 
diff --git a/src/lib/libssl/src/crypto/cast/c_enc.c b/src/lib/libssl/src/crypto/cast/c_enc.c
index 0fe2cffecc..e80f65b698 100644
--- a/src/lib/libssl/src/crypto/cast/c_enc.c
+++ b/src/lib/libssl/src/crypto/cast/c_enc.c
@@ -59,6 +59,7 @@
 #include <openssl/cast.h>
 #include "cast_lcl.h"
 
+#ifndef OPENBSD_CAST_ASM
 void CAST_encrypt(CAST_LONG *data, CAST_KEY *key)
         {
         register CAST_LONG l,r,*k,t;
@@ -122,6 +123,7 @@ void CAST_decrypt(CAST_LONG *data, CAST_KEY *key)
         data[1]=l&0xffffffffL;
         data[0]=r&0xffffffffL;
         }
+#endif
 
 void CAST_cbc_encrypt(const unsigned char *in, unsigned char *out, long length,
              CAST_KEY *ks, unsigned char *iv, int enc)
diff --git a/src/lib/libssl/src/crypto/des/asm/des-586.pl b/src/lib/libssl/src/crypto/des/asm/des-586.pl
index b75d3c6b3a..60d577cc8d 100644
--- a/src/lib/libssl/src/crypto/des/asm/des-586.pl
+++ b/src/lib/libssl/src/crypto/des/asm/des-586.pl
@@ -22,10 +22,14 @@ $R="esi";
 &external_label("DES_SPtrans");
 &DES_encrypt("DES_encrypt1",1);
 &DES_encrypt("DES_encrypt2",0);
-&DES_encrypt3("DES_encrypt3",1);
-&DES_encrypt3("DES_decrypt3",0);
-&cbc("DES_ncbc_encrypt","DES_encrypt1","DES_encrypt1",0,4,5,3,5,-1);
-&cbc("DES_ede3_cbc_encrypt","DES_encrypt3","DES_decrypt3",0,6,7,3,4,5);
+
+if (!$main'openbsd)
+        {
+        &DES_encrypt3("DES_encrypt3",1);
+        &DES_encrypt3("DES_decrypt3",0);
+        &cbc("DES_ncbc_encrypt","DES_encrypt1","DES_encrypt1",0,4,5,3,5,-1);
+        &cbc("DES_ede3_cbc_encrypt","DES_encrypt3","DES_decrypt3",0,6,7,3,4,5);
+        }
 
 &asm_finish();
 
diff --git a/src/lib/libssl/src/crypto/des/des_enc.c b/src/lib/libssl/src/crypto/des/des_enc.c
index 1c37ab96d3..4f09804c44 100644
--- a/src/lib/libssl/src/crypto/des/des_enc.c
+++ b/src/lib/libssl/src/crypto/des/des_enc.c
@@ -58,6 +58,7 @@
 
 #include "des_locl.h"
 
+#ifndef OPENBSD_DES_ASM
 void DES_encrypt1(DES_LONG *data, DES_key_schedule *ks, int enc)
         {
         register DES_LONG l,r,t,u;
@@ -246,6 +247,7 @@ void DES_encrypt2(DES_LONG *data, DES_key_schedule *ks, int enc)
         data[1]=ROTATE(r,3)&0xffffffffL;
         l=r=t=u=0;
         }
+#endif
 
 void DES_encrypt3(DES_LONG *data, DES_key_schedule *ks1,
                   DES_key_schedule *ks2, DES_key_schedule *ks3)
diff --git a/src/lib/libssl/src/crypto/perlasm/x86asm.pl b/src/lib/libssl/src/crypto/perlasm/x86asm.pl
index 1cb96e914a..7c675e3ced 100644
--- a/src/lib/libssl/src/crypto/perlasm/x86asm.pl
+++ b/src/lib/libssl/src/crypto/perlasm/x86asm.pl
@@ -18,9 +18,13 @@ sub main'asm_init
         ($type,$fn,$i386)=@_;
         $filename=$fn;
 
-        $elf=$cpp=$sol=$aout=$win32=$gaswin=0;
+        $elf=$cpp=$sol=$aout=$win32=$gaswin=$openbsd=0;
         if (    ($type eq "elf"))
                 { $elf=1; require "x86unix.pl"; }
+        elsif ( ($type eq "openbsd-elf"))
+                { $openbsd=$elf=1; require "x86unix.pl"; }
+        elsif ( ($type eq "openbsd-a.out"))
+                { $openbsd=1; require "x86unix.pl"; }
         elsif ( ($type eq "a.out"))
                 { $aout=1; require "x86unix.pl"; }
         elsif ( ($type eq "gaswin"))
@@ -43,6 +47,8 @@ Pick one target type from
         cpp     - format so x86unix.cpp can be used
         win32   - Windows 95/Windows NT
         win32n  - Windows 95/Windows NT NASM format
+        openbsd-elf     - OpenBSD elf
+        openbsd-a.out   - OpenBSD a.out
 EOF
                 exit(1);
                 }
diff --git a/src/lib/libssl/src/crypto/perlasm/x86unix.pl b/src/lib/libssl/src/crypto/perlasm/x86unix.pl
index 10b669bf04..53ad5f4927 100644
--- a/src/lib/libssl/src/crypto/perlasm/x86unix.pl
+++ b/src/lib/libssl/src/crypto/perlasm/x86unix.pl
@@ -15,6 +15,12 @@ sub main'asm_get_output { return(@out); }
 sub main'get_labels { return(@labels); }
 sub main'external_label { push(@labels,@_); }
 
+if ($main'openbsd)
+        {
+        $com_start='/*';
+        $com_end='*/';
+        }
+
 if ($main'cpp)
         {
         $align="ALIGN";
@@ -276,6 +282,9 @@ sub main'file
         {
         local($file)=@_;
 
+        if ($main'openbsd)
+                { push(@out,"#include <machine/asm.h>\n"); return; }
+
         local($tmp)=<<"EOF";
         .file   "$file.s"
         .version        "01.01"
@@ -291,6 +300,9 @@ sub main'function_begin
         &main'external_label($func);
         $func=$under.$func;
 
+        if ($main'openbsd)
+                { push (@out, "\nENTRY($func)\n"); goto skip; }
+
         local($tmp)=<<"EOF";
 .text
         .align $align
@@ -303,6 +315,7 @@ EOF
                 { $tmp=push(@out,"\t.def\t$func;\t.scl\t2;\t.type\t32;\t.endef\n"); }
         else    { $tmp=push(@out,"\t.type\t$func,\@function\n"); }
         push(@out,"$func:\n");
+skip:
         $tmp=<<"EOF";
         pushl   %ebp
         pushl   %ebx
@@ -321,6 +334,9 @@ sub main'function_begin_B
         &main'external_label($func);
         $func=$under.$func;
 
+        if ($main'openbsd)
+                { push(@out, "\nENTRY($func)\n"); goto skip; }
+
         local($tmp)=<<"EOF";
 .text
         .align $align
@@ -333,6 +349,7 @@ EOF
                 { $tmp=push(@out,"\t.def\t$func;\t.scl\t2;\t.type\t32;\t.endef\n"); }
         else    { push(@out,"\t.type    $func,\@function\n"); }
         push(@out,"$func:\n");
+skip:
         $stack=4;
         }
 
@@ -429,7 +446,8 @@ sub main'swtmp
 
 sub main'comment
         {
-        if ($main'elf)  # GNU and SVR4 as'es use different comment delimiters,
+        if (!$main'openbsd && $main'elf)
+                        # GNU and SVR4 as'es use different comment delimiters,
                 {       # so we just skip comments...
                 push(@out,"\n");
                 return;
@@ -460,7 +478,10 @@ sub main'set_label
                 $label{$_[0]}=".${label}${_[0]}";
                 $label++;
                 }
-        push(@out,".align $align\n") if ($_[1] != 0);
+        if ($main'openbsd)
+                { push(@out,"_ALIGN_TEXT\n") if ($_[1] != 0); }
+        else
+                { push(@out,".align $align\n") if ($_[1] != 0); }
         push(@out,"$label{$_[0]}:\n");
         }
 
@@ -569,6 +590,16 @@ sub main'picmeup
 ___
                 push(@out,$tmp);
                 }
+        elsif ($main'openbsd)
+                {
+                push(@out, "#ifdef PIC\n");
+                push(@out, "\tPIC_PROLOGUE\n");
+                &main'mov($dst,"PIC_GOT($sym)");
+                push(@out, "\tPIC_EPILOGUE\n");
+                push(@out, "#else\n");
+                &main'lea($dst,&main'DWP($sym));
+                push(@out, "#endif\n");
+                }
         elsif ($main'pic && ($main'elf || $main'aout))
                 {
                 push(@out,"\t.align\t8\n");