merge 0.9.7b with local changes; crank majors for libssl/libcrypto

15 files changed, 500 insertions, 707 deletions

diff --git a/src/lib/libcrypto/bn/Makefile.ssl b/src/lib/libcrypto/bn/Makefile.ssl
index 6a479726c4..fa17d3c7d8 100644
--- a/src/lib/libcrypto/bn/Makefile.ssl
+++ b/src/lib/libcrypto/bn/Makefile.ssl
@@ -23,14 +23,6 @@ BN_ASM=		bn_asm.o
 
 CFLAGS= $(INCLUDES) $(CFLAG)
 
-# We let the C compiler driver to take care of .s files. This is done in
-# order to be excused from maintaining a separate set of architecture
-# dependent assembler flags. E.g. if you throw -mcpu=ultrasparc at SPARC
-# gcc, then the driver will automatically translate it to -xarch=v8plus
-# and pass it down to assembler.
-AS=$(CC) -c
-ASFLAGS=$(CFLAGS)
-
 GENERAL=Makefile
 TEST=bntest.c exptest.c
 APPS=
@@ -73,22 +65,11 @@ lib:	$(LIBOBJ)
         @touch lib
 
 # elf
-asm/bn86-elf.o: asm/bn86unix.cpp
-        $(CPP) -DELF -x c asm/bn86unix.cpp | as -o asm/bn86-elf.o
-
-asm/co86-elf.o: asm/co86unix.cpp
-        $(CPP) -DELF -x c asm/co86unix.cpp | as -o asm/co86-elf.o
+asm/bn86-elf.s: asm/bn-586.pl ../perlasm/x86asm.pl
+        (cd asm; $(PERL) bn-586.pl elf $(CFLAGS) > bn86-elf.s)
 
-# solaris
-asm/bn86-sol.o: asm/bn86unix.cpp
-        $(CC) -E -DSOL asm/bn86unix.cpp | sed 's/^#.*//' > asm/bn86-sol.s
-        as -o asm/bn86-sol.o asm/bn86-sol.s
-        rm -f asm/bn86-sol.s
-
-asm/co86-sol.o: asm/co86unix.cpp
-        $(CC) -E -DSOL asm/co86unix.cpp | sed 's/^#.*//' > asm/co86-sol.s
-        as -o asm/co86-sol.o asm/co86-sol.s
-        rm -f asm/co86-sol.s
+asm/co86-elf.s: asm/co-586.pl ../perlasm/x86asm.pl
+        (cd asm; $(PERL) co-586.pl elf $(CFLAGS) > co86-elf.s)
 
 # a.out
 asm/bn86-out.o: asm/bn86unix.cpp
@@ -136,6 +117,8 @@ asm/ia64-cpp.o:	asm/ia64.S
         $(CC) $(ASFLAGS) -c -o asm/ia64-cpp.o /tmp/ia64.$$$$.s; \
         rm -f /tmp/ia64.$$$$.s
 
+asm/x86_64-gcc.o: asm/x86_64-gcc.c
+
 files:
         $(PERL) $(TOP)/util/files.pl Makefile.ssl >> $(TOP)/MINFO
 
@@ -169,14 +152,14 @@ lint:
         lint -DLINT $(INCLUDES) $(SRC)>fluff
 
 depend:
-        $(MAKEDEPEND) $(CFLAG) $(INCLUDES) $(DEPFLAG) $(PROGS) $(LIBSRC)
+        $(MAKEDEPEND) -- $(CFLAG) $(INCLUDES) $(DEPFLAG) -- $(PROGS) $(LIBSRC)
 
 dclean:
         $(PERL) -pe 'if (/^# DO NOT DELETE THIS LINE/) {print; exit(0);}' $(MAKEFILE) >Makefile.new
         mv -f Makefile.new $(MAKEFILE)
 
 clean:
-        rm -f asm/co86unix.cpp asm/bn86unix.cpp *.o */*.o *.obj lib tags core .pure .nfs* *.old *.bak fluff bn_asm.s
+        rm -f asm/co86unix.cpp asm/bn86unix.cpp asm/*-elf.* *.o */*.o *.obj lib tags core .pure .nfs* *.old *.bak fluff bn_asm.s
 
 # DO NOT DELETE THIS LINE -- make depend depends on it.
 
diff --git a/src/lib/libcrypto/bn/asm/ia64.S b/src/lib/libcrypto/bn/asm/ia64.S
index ae56066310..7dfda85566 100644
--- a/src/lib/libcrypto/bn/asm/ia64.S
+++ b/src/lib/libcrypto/bn/asm/ia64.S
@@ -1,6 +1,6 @@
 .explicit
 .text
-.ident  "ia64.S, Version 1.1"
+.ident  "ia64.S, Version 2.0"
 .ident  "IA-64 ISA artwork by Andy Polyakov <appro@fy.chalmers.se>"
 
 //
@@ -13,6 +13,35 @@
 // disclaimed.
 // ====================================================================
 //
+// Version 2.x is Itanium2 re-tune. Few words about how Itanum2 is
+// different from Itanium to this module viewpoint. Most notably, is it
+// "wider" than Itanium? Can you experience loop scalability as
+// discussed in commentary sections? Not really:-( Itanium2 has 6
+// integer ALU ports, i.e. it's 2 ports wider, but it's not enough to
+// spin twice as fast, as I need 8 IALU ports. Amount of floating point
+// ports is the same, i.e. 2, while I need 4. In other words, to this
+// module Itanium2 remains effectively as "wide" as Itanium. Yet it's
+// essentially different in respect to this module, and a re-tune was
+// required. Well, because some intruction latencies has changed. Most
+// noticeably those intensively used:
+//
+//                      Itanium Itanium2
+//      ldf8            9       6               L2 hit
+//      ld8             2       1               L1 hit
+//      getf            2       5
+//      xma[->getf]     7[+1]   4[+0]
+//      add[->st8]      1[+1]   1[+0]
+//
+// What does it mean? You might ratiocinate that the original code
+// should run just faster... Because sum of latencies is smaller...
+// Wrong! Note that getf latency increased. This means that if a loop is
+// scheduled for lower latency (and they are), then it will suffer from
+// stall condition and the code will therefore turn anti-scalable, e.g.
+// original bn_mul_words spun at 5*n or 2.5 times slower than expected
+// on Itanium2! What to do? Reschedule loops for Itanium2? But then
+// Itanium would exhibit anti-scalability. So I've chosen to reschedule
+// for worst latency for every instruction aiming for best *all-round*
+// performance.  
 
 // Q.   How much faster does it get?
 // A.   Here is the output from 'openssl speed rsa dsa' for vanilla
@@ -149,12 +178,27 @@ bn_add_words:
         brp.loop.imp    .L_bn_add_words_ctop,.L_bn_add_words_cend-16
                                         }
         .body
-{ .mib; mov             r14=r32                 // rp
+{ .mib;
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
+        addp4           r14=0,r32               // rp
+#else
+        mov             r14=r32                 // rp
+#endif
         mov             r9=pr           };;
-{ .mii; mov             r15=r33                 // ap
+{ .mii;
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
+        addp4           r15=0,r33               // ap
+#else
+        mov             r15=r33                 // ap
+#endif
         mov             ar.lc=r10
         mov             ar.ec=6         }
-{ .mib; mov             r16=r34                 // bp
+{ .mib;
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
+        addp4           r16=0,r34               // bp
+#else
+        mov             r16=r34                 // bp
+#endif
         mov             pr.rot=1<<16    };;
 
 .L_bn_add_words_ctop:
@@ -174,7 +218,7 @@ bn_add_words:
 
 { .mii;
 (p59)   add             r8=1,r8         // return value
-        mov             pr=r9,-1
+        mov             pr=r9,0x1ffff
         mov             ar.lc=r3        }
 { .mbb; nop.b           0x0
         br.ret.sptk.many        b0      };;
@@ -202,12 +246,27 @@ bn_sub_words:
         brp.loop.imp    .L_bn_sub_words_ctop,.L_bn_sub_words_cend-16
                                         }
         .body
-{ .mib; mov             r14=r32                 // rp
+{ .mib;
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
+        addp4           r14=0,r32               // rp
+#else
+        mov             r14=r32                 // rp
+#endif
         mov             r9=pr           };;
-{ .mii; mov             r15=r33                 // ap
+{ .mii;
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
+        addp4           r15=0,r33               // ap
+#else
+        mov             r15=r33                 // ap
+#endif
         mov             ar.lc=r10
         mov             ar.ec=6         }
-{ .mib; mov             r16=r34                 // bp
+{ .mib;
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
+        addp4           r16=0,r34               // bp
+#else
+        mov             r16=r34                 // bp
+#endif
         mov             pr.rot=1<<16    };;
 
 .L_bn_sub_words_ctop:
@@ -227,7 +286,7 @@ bn_sub_words:
 
 { .mii;
 (p59)   add             r8=1,r8         // return value
-        mov             pr=r9,-1
+        mov             pr=r9,0x1ffff
         mov             ar.lc=r3        }
 { .mbb; nop.b           0x0
         br.ret.sptk.many        b0      };;
@@ -253,7 +312,7 @@ bn_mul_words:
 #ifdef XMA_TEMPTATION
 { .mfi; alloc           r2=ar.pfs,4,0,0,0       };;
 #else
-{ .mfi; alloc           r2=ar.pfs,4,4,0,8       };;
+{ .mfi; alloc           r2=ar.pfs,4,12,0,16     };;
 #endif
 { .mib; mov             r8=r0                   // return value
         cmp4.le         p6,p0=r34,r0
@@ -266,24 +325,30 @@ bn_mul_words:
 
         .body
 { .mib; setf.sig        f8=r35  // w
-        mov             pr.rot=0x400001<<16
-                        // ------^----- serves as (p48) at first (p26)
+        mov             pr.rot=0x800001<<16
+                        // ------^----- serves as (p50) at first (p27)
         brp.loop.imp    .L_bn_mul_words_ctop,.L_bn_mul_words_cend-16
                                         }
 
 #ifndef XMA_TEMPTATION
 
-{ .mii; mov             r14=r32 // rp
-        mov             r15=r33 // ap
+{ .mii;
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
+        addp4           r14=0,r32       // rp
+        addp4           r15=0,r33       // ap
+#else
+        mov             r14=r32         // rp
+        mov             r15=r33         // ap
+#endif
         mov             ar.lc=r10       }
-{ .mii; mov             r39=0   // serves as r33 at first (p26)
-        mov             ar.ec=12        };;
+{ .mii; mov             r40=0   // serves as r35 at first (p27)
+        mov             ar.ec=13        };;
 
-// This loop spins in 2*(n+11) ticks. It's scheduled for data in L2
-// cache (i.e. 9 ticks away) as floating point load/store instructions
+// This loop spins in 2*(n+12) ticks. It's scheduled for data in Itanium
+// L2 cache (i.e. 9 ticks away) as floating point load/store instructions
 // bypass L1 cache and L2 latency is actually best-case scenario for
-// ldf8. The loop is not scalable and shall run in 2*(n+11) even on
-// "wider" IA-64 implementations. It's a trade-off here. n+22 loop
+// ldf8. The loop is not scalable and shall run in 2*(n+12) even on
+// "wider" IA-64 implementations. It's a trade-off here. n+24 loop
 // would give us ~5% in *overall* performance improvement on "wider"
 // IA-64, but would hurt Itanium for about same because of longer
 // epilogue. As it's a matter of few percents in either case I've
@@ -291,25 +356,25 @@ bn_mul_words:
 // this very instruction sequence in bn_mul_add_words loop which in
 // turn is scalable).
 .L_bn_mul_words_ctop:
-{ .mfi; (p25)   getf.sig        r36=f49                 // low
-        (p21)   xmpy.lu         f45=f37,f8
-        (p27)   cmp.ltu         p52,p48=r39,r38 }
+{ .mfi; (p25)   getf.sig        r36=f52                 // low
+        (p21)   xmpy.lu         f48=f37,f8
+        (p28)   cmp.ltu         p54,p50=r41,r39 }
 { .mfi; (p16)   ldf8            f32=[r15],8
-        (p21)   xmpy.hu         f38=f37,f8
+        (p21)   xmpy.hu         f40=f37,f8
         (p0)    nop.i           0x0             };;
-{ .mii; (p26)   getf.sig        r32=f43                 // high
-        .pred.rel       "mutex",p48,p52
-        (p48)   add             r38=r37,r33             // (p26)
-        (p52)   add             r38=r37,r33,1   }       // (p26)
-{ .mfb; (p27)   st8             [r14]=r39,8
+{ .mii; (p25)   getf.sig        r32=f44                 // high
+        .pred.rel       "mutex",p50,p54
+        (p50)   add             r40=r38,r35             // (p27)
+        (p54)   add             r40=r38,r35,1   }       // (p27)
+{ .mfb; (p28)   st8             [r14]=r41,8
         (p0)    nop.f           0x0
         br.ctop.sptk    .L_bn_mul_words_ctop    };;
 .L_bn_mul_words_cend:
 
 { .mii; nop.m           0x0
-.pred.rel       "mutex",p49,p53
-(p49)   add             r8=r34,r0
-(p53)   add             r8=r34,r0,1     }
+.pred.rel       "mutex",p51,p55
+(p51)   add             r8=r36,r0
+(p55)   add             r8=r36,r0,1     }
 { .mfb; nop.m   0x0
         nop.f   0x0
         nop.b   0x0                     }
@@ -344,7 +409,7 @@ bn_mul_words:
 #endif  // XMA_TEMPTATION
 
 { .mii; nop.m           0x0
-        mov             pr=r9,-1
+        mov             pr=r9,0x1ffff
         mov             ar.lc=r3        }
 { .mfb; rum             1<<5            // clear um.mfh
         nop.f           0x0
@@ -376,59 +441,69 @@ bn_mul_add_words:
 
         .body
 { .mib; setf.sig        f8=r35  // w
-        mov             pr.rot=0x400001<<16
-                        // ------^----- serves as (p48) at first (p26)
+        mov             pr.rot=0x800001<<16
+                        // ------^----- serves as (p50) at first (p27)
         brp.loop.imp    .L_bn_mul_add_words_ctop,.L_bn_mul_add_words_cend-16
                                         }
-{ .mii; mov             r14=r32 // rp
-        mov             r15=r33 // ap
+{ .mii;
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
+        addp4           r14=0,r32       // rp
+        addp4           r15=0,r33       // ap
+#else
+        mov             r14=r32         // rp
+        mov             r15=r33         // ap
+#endif
         mov             ar.lc=r10       }
-{ .mii; mov             r39=0   // serves as r33 at first (p26)
-        mov             r18=r32 // rp copy
-        mov             ar.ec=14        };;
+{ .mii; mov             r40=0   // serves as r35 at first (p27)
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
+        addp4           r18=0,r32       // rp copy
+#else
+        mov             r18=r32         // rp copy
+#endif
+        mov             ar.ec=15        };;
 
-// This loop spins in 3*(n+13) ticks on Itanium and should spin in
-// 2*(n+13) on "wider" IA-64 implementations (to be verified with new
+// This loop spins in 3*(n+14) ticks on Itanium and should spin in
+// 2*(n+14) on "wider" IA-64 implementations (to be verified with new
 // µ-architecture manuals as they become available). As usual it's
 // possible to compress the epilogue, down to 10 in this case, at the
 // cost of scalability. Compressed (and therefore non-scalable) loop
-// running at 3*(n+10) would buy you ~10% on Itanium but take ~35%
+// running at 3*(n+11) would buy you ~10% on Itanium but take ~35%
 // from "wider" IA-64 so let it be scalable! Special attention was
 // paid for having the loop body split at 64-byte boundary. ld8 is
 // scheduled for L1 cache as the data is more than likely there.
 // Indeed, bn_mul_words has put it there a moment ago:-)
 .L_bn_mul_add_words_ctop:
-{ .mfi; (p25)   getf.sig        r36=f49                 // low
-        (p21)   xmpy.lu         f45=f37,f8
-        (p27)   cmp.ltu         p52,p48=r39,r38 }
+{ .mfi; (p25)   getf.sig        r36=f52                 // low
+        (p21)   xmpy.lu         f48=f37,f8
+        (p28)   cmp.ltu         p54,p50=r41,r39 }
 { .mfi; (p16)   ldf8            f32=[r15],8
-        (p21)   xmpy.hu         f38=f37,f8
-        (p27)   add             r43=r43,r39     };;
-{ .mii; (p26)   getf.sig        r32=f43                 // high
-        .pred.rel       "mutex",p48,p52
-        (p48)   add             r38=r37,r33             // (p26)
-        (p52)   add             r38=r37,r33,1   }       // (p26)
-{ .mfb; (p27)   cmp.ltu.unc     p56,p0=r43,r39
+        (p21)   xmpy.hu         f40=f37,f8
+        (p28)   add             r45=r45,r41     };;
+{ .mii; (p25)   getf.sig        r32=f44                 // high
+        .pred.rel       "mutex",p50,p54
+        (p50)   add             r40=r38,r35             // (p27)
+        (p54)   add             r40=r38,r35,1   }       // (p27)
+{ .mfb; (p28)   cmp.ltu.unc     p60,p0=r45,r41
         (p0)    nop.f           0x0
         (p0)    nop.b           0x0             }
-{ .mii; (p26)   ld8             r42=[r18],8
-        (p58)   cmp.eq.or       p57,p0=-1,r44
-        (p58)   add             r44=1,r44       }
-{ .mfb; (p29)   st8             [r14]=r45,8
+{ .mii; (p27)   ld8             r44=[r18],8
+        (p62)   cmp.eq.or       p61,p0=-1,r46
+        (p62)   add             r46=1,r46       }
+{ .mfb; (p30)   st8             [r14]=r47,8
         (p0)    nop.f           0x0
         br.ctop.sptk    .L_bn_mul_add_words_ctop};;
 .L_bn_mul_add_words_cend:
 
 { .mii; nop.m           0x0
-.pred.rel       "mutex",p51,p55
-(p51)   add             r8=r36,r0
-(p55)   add             r8=r36,r0,1     }
+.pred.rel       "mutex",p53,p57
+(p53)   add             r8=r38,r0
+(p57)   add             r8=r38,r0,1     }
 { .mfb; nop.m   0x0
         nop.f   0x0
         nop.b   0x0                     };;
 { .mii;
-(p59)   add             r8=1,r8
-        mov             pr=r9,-1
+(p63)   add             r8=1,r8
+        mov             pr=r9,0x1ffff
         mov             ar.lc=r3        }
 { .mfb; rum             1<<5            // clear um.mfh
         nop.f           0x0
@@ -461,6 +536,10 @@ bn_sqr_words:
         mov     r9=pr                   };;
 
         .body
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
+{ .mii; addp4           r32=0,r32
+        addp4           r33=0,r33       };;
+#endif
 { .mib;
         mov             pr.rot=1<<16
         brp.loop.imp    .L_bn_sqr_words_ctop,.L_bn_sqr_words_cend-16
@@ -492,7 +571,7 @@ bn_sqr_words:
 .L_bn_sqr_words_cend:
 
 { .mii; nop.m           0x0
-        mov             pr=r9,-1
+        mov             pr=r9,0x1ffff
         mov             ar.lc=r3        }
 { .mfb; rum             1<<5            // clear um.mfh
         nop.f           0x0
@@ -526,7 +605,14 @@ bn_sqr_comba8:
         .prologue
         .fframe 0
         .save   ar.pfs,r2
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
 { .mii; alloc   r2=ar.pfs,2,1,0,0
+        addp4   r33=0,r33
+        addp4   r32=0,r32               };;
+{ .mii;
+#else
+{ .mii; alloc   r2=ar.pfs,2,1,0,0
+#endif
         mov     r34=r33
         add     r14=8,r33               };;
         .body
@@ -587,7 +673,14 @@ bn_mul_comba8:
         .prologue
         .fframe 0
         .save   ar.pfs,r2
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
 { .mii; alloc   r2=ar.pfs,3,0,0,0
+        addp4   r33=0,r33
+        addp4   r34=0,r34               };;
+{ .mii; addp4   r32=0,r32
+#else
+{ .mii; alloc   r2=ar.pfs,3,0,0,0
+#endif
         add     r14=8,r33
         add     r17=8,r34               }
         .body
@@ -1138,7 +1231,14 @@ bn_sqr_comba4:
         .prologue
         .fframe 0
         .save   ar.pfs,r2
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
+{ .mii; alloc   r2=ar.pfs,2,1,0,0
+        addp4   r32=0,r32
+        addp4   r33=0,r33               };;
+{ .mii;
+#else
 { .mii; alloc   r2=ar.pfs,2,1,0,0
+#endif
         mov     r34=r33
         add     r14=8,r33               };;
         .body
@@ -1164,7 +1264,14 @@ bn_mul_comba4:
         .prologue
         .fframe 0
         .save   ar.pfs,r2
+#if defined(_HPUX_SOURCE) && defined(_ILP32)
+{ .mii; alloc   r2=ar.pfs,3,0,0,0
+        addp4   r33=0,r33
+        addp4   r34=0,r34               };;
+{ .mii; addp4   r32=0,r32
+#else
 { .mii; alloc   r2=ar.pfs,3,0,0,0
+#endif
         add     r14=8,r33
         add     r17=8,r34               }
         .body
@@ -1464,7 +1571,7 @@ bn_div_words:
         or      r8=r8,r33
         mov     ar.pfs=r2               };;
 { .mii; shr.u   r9=H,I                  // remainder if anybody wants it
-        mov     pr=r10,-1               }
+        mov     pr=r10,0x1ffff          }
 { .mfb; br.ret.sptk.many        b0      };;
 
 // Unsigned 64 by 32 (well, by 64 for the moment) bit integer division
diff --git a/src/lib/libcrypto/bn/asm/pa-risc2.s b/src/lib/libcrypto/bn/asm/pa-risc2.s
index af9730d062..f3b16290eb 100644
--- a/src/lib/libcrypto/bn/asm/pa-risc2.s
+++ b/src/lib/libcrypto/bn/asm/pa-risc2.s
@@ -747,8 +747,8 @@ bn_div_words
         .PROC
         .EXPORT bn_div_words,ENTRY,PRIV_LEV=3,ARGW0=GR,ARGW1=GR,ARGW2=GR,ARGW3=GR,RTNVAL=GR,LONG_RETURN
         .IMPORT BN_num_bits_word,CODE
-        .IMPORT __iob,DATA
-        .IMPORT fprintf,CODE
+        ;--- not PIC    .IMPORT __iob,DATA
+        ;--- not PIC    .IMPORT fprintf,CODE
         .IMPORT abort,CODE
         .IMPORT $$div2U,MILLICODE
         .CALLINFO CALLER,FRAME=144,ENTRY_GR=%r9,SAVE_RP,ARGS_SAVED,ORDERING_AWARE
@@ -844,12 +844,12 @@ $0006001A
         MOVIB,TR        2,%r8,$0006001C ;offset 0xa18
         EXTRD,U %r3,63,32,%r7   ;offset 0xa1c
 $D2
-        ADDIL   LR'__iob-$global$,%r27,%r1      ;offset 0xa20
-        LDIL    LR'C$7,%r21     ;offset 0xa24
-        LDO     RR'__iob-$global$+32(%r1),%r26  ;offset 0xa28
-        .CALL   ARGW0=GR,ARGW1=GR,ARGW2=GR,RTNVAL=GR    ;in=24,25,26;out=28;
-        B,L     fprintf,%r2     ;offset 0xa2c
-        LDO     RR'C$7(%r21),%r25       ;offset 0xa30
+        ;--- not PIC    ADDIL   LR'__iob-$global$,%r27,%r1      ;offset 0xa20
+        ;--- not PIC    LDIL    LR'C$7,%r21     ;offset 0xa24
+        ;--- not PIC    LDO     RR'__iob-$global$+32(%r1),%r26  ;offset 0xa28
+        ;--- not PIC    .CALL   ARGW0=GR,ARGW1=GR,ARGW2=GR,RTNVAL=GR    ;in=24,25,26;out=28;
+        ;--- not PIC    B,L     fprintf,%r2     ;offset 0xa2c
+        ;--- not PIC    LDO     RR'C$7(%r21),%r25       ;offset 0xa30
         .CALL           ;
         B,L     abort,%r2       ;offset 0xa34
         NOP             ;offset 0xa38
@@ -1605,14 +1605,14 @@ bn_mul_comba4
         .PROCEND        
 
 
-        .SPACE  $TEXT$
-        .SUBSPA $CODE$
-        .SPACE  $PRIVATE$,SORT=16
-        .IMPORT $global$,DATA
-        .SPACE  $TEXT$
-        .SUBSPA $CODE$
-        .SUBSPA $LIT$,ACCESS=0x2c
-C$7
-        .ALIGN  8
-        .STRINGZ        "Division would overflow (%d)\n"
+;--- not PIC    .SPACE  $TEXT$
+;--- not PIC    .SUBSPA $CODE$
+;--- not PIC    .SPACE  $PRIVATE$,SORT=16
+;--- not PIC    .IMPORT $global$,DATA
+;--- not PIC    .SPACE  $TEXT$
+;--- not PIC    .SUBSPA $CODE$
+;--- not PIC    .SUBSPA $LIT$,ACCESS=0x2c
+;--- not PIC    C$7
+;--- not PIC    .ALIGN  8
+;--- not PIC    .STRINGZ        "Division would overflow (%d)\n"
         .END
diff --git a/src/lib/libcrypto/bn/asm/vms.mar b/src/lib/libcrypto/bn/asm/vms.mar
index 465f2774b6..aefab15cdb 100644
--- a/src/lib/libcrypto/bn/asm/vms.mar
+++ b/src/lib/libcrypto/bn/asm/vms.mar
@@ -1,4 +1,4 @@
-        .title  vax_bn_mul_add_word  unsigned multiply & add, 32*32+32+32=>64
+        .title  vax_bn_mul_add_words  unsigned multiply & add, 32*32+32+32=>64
 ;
 ; w.j.m. 15-jan-1999
 ;
@@ -59,7 +59,7 @@ w=16 ;(AP)	w	by value (input)
         movl    r6,r0                   ; return c
         ret
 
-        .title  vax_bn_mul_word  unsigned multiply & add, 32*32+32=>64
+        .title  vax_bn_mul_words  unsigned multiply & add, 32*32+32=>64
 ;
 ; w.j.m. 15-jan-1999
 ;
@@ -172,147 +172,175 @@ n=12 ;(AP)	n	by value (input)
 ; }
 ;
 ; Using EDIV would be very easy, if it didn't do signed calculations.
-; Therefore, som extra things have to happen around it.  The way to
-; handle that is to shift all operands right one step (basically dividing
-; them by 2) and handle the different cases depending on what the lowest
-; bit of each operand was.
+; 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).
 ;
-; To start with, let's define the following:
+; 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.
 ;
-; a' = l & 1
-; a2 = <h,l> >> 1       # UNSIGNED shift!
-; b' = d & 1
-; b2 = d >> 1           # UNSIGNED shift!
+; 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).
 ;
-; Now, use EDIV to calculate a quotient and a remainder:
+; 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.
 ;
-; q'' = a2/b2
-; r'' = a2 - q''*b2
+; The complete algorithm looks like this:
 ;
-; If b' is 0, the quotient is already correct, we just need to adjust the
-; remainder:
+; 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
 ;
-; if (b' == 0)
-;   {
-;     r = 2*r'' + a'
-;     q = q''
-;   }
-;
-; If b' is 1, we need to do other adjustements.  The first thought is the
-; following (note that r' will not always have the right value, but an
-; adjustement follows further down):
-;
-; if (b' == 1)
-;   {
-;     q' = q''
-;     r' = a - q'*b
-;
-; However, one can note the folowing relationship:
-;
-;                         r'' = a2 - q''*b2
-;                  =>   2*r'' = 2*a2 - 2*q''*b2
-;                             = { a = 2*a2 + a', b = 2*b2 + b' = 2*b2 + 1,
-;                                 q' = q'' }
-;                             = a - a' - q'*(b - 1)
-;                             = a - q'*b - a' + q'
-;                             = r' - a' + q'
-;                  =>     r'  = 2*r'' - q' + a'
+; r'    = r >> 29
+; if (d' >= 0)
+;   q'  = q >> 29
+;   q   = q << 3
+; else
+;   q'  = q >> 30
+;   q   = q << 2
+; r     = (r << 3) + l'
 ;
-; This enables us to use r'' instead of discarding and calculating another
-; modulo:
-;
-; if (b' == 1)
+; if (d' < 0)
 ;   {
-;     q' = q''
-;     r' = (r'' << 1) - q' + a'
-;
-; Now, all we have to do is adjust r', because it might be < 0:
-;
-;     while (r' < 0)
+;     [r',r] = [r',r] - q
+;     while ([r',r] < 0)
 ;       {
-;         r' = r' + b
-;         q' = q' - 1
+;         [r',r] = [r',r] + d
+;         [q',q] = [q',q] - 1
 ;       }
 ;   }
 ;
-; return q'
+; while ([r',r] >= d')
+;   {
+;     [r',r] = [r',r] - d'
+;     [q',q] = [q',q] + 1
+;   }
+;
+; return q
 
 h=4 ;(AP)       h       by value (input)
 l=8 ;(AP)       l       by value (input)
 d=12 ;(AP)      d       by value (input)
 
-;aprim=r5
-;a2=r6
-;a20=r6
-;a21=r7
-;bprim=r8
-;b2=r9
-;qprim=r10      ; initially used as q''
-;rprim=r11      ; initially used as r''
-
+;r2 = l, q
+;r3 = h, r
+;r4 = d
+;r5 = l'
+;r6 = r'
+;r7 = d'
+;r8 = q'
 
         .psect  code,nowrt
 
-.entry  bn_div_words,^m<r2,r3,r4,r5,r6,r7,r8,r9,r10,r11>
+.entry  bn_div_words,^m<r2,r3,r4,r5,r6,r7,r8>
         movl    l(ap),r2
         movl    h(ap),r3
         movl    d(ap),r4
 
-        movl    #0,r5
-        movl    #0,r8
-        movl    #0,r0
-;       movl    #0,r1
+        bicl3   #^XFFFFFFF8,r2,r5 ; l' = l & 7
+        bicl3   #^X00000007,r2,r2
 
-        rotl    #-1,r2,r6       ; a20 = l >> 1 (almost)
-        rotl    #-1,r3,r7       ; a21 = h >> 1 (almost)
-        rotl    #-1,r4,r9       ; b2 = d >> 1 (almost)
+        bicl3   #^XFFFFFFF8,r3,r6
+        bicl3   #^X00000007,r3,r3
+        
+        addl    r6,r2
 
-        tstl    r6
-        bgeq    1$
-        xorl2   #^X80000000,r6  ; fixup a20 so highest bit is 0
-        incl    r5              ; a' = 1
-1$:
-        tstl    r7
-        bgeq    2$
-        xorl2   #^X80000000,r6  ; fixup a20 so highest bit is 1,
-                                ; since that's what was lowest in a21
-        xorl2   #^X80000000,r7  ; fixup a21 so highest bit is 1
-2$:
-        tstl    r9
+        rotl    #-3,r2,r2       ; l = l >> 3
+        rotl    #-3,r3,r3       ; h = h >> 3
+                
+        movl    r4,r7           ; d' = d
+
+        movl    #0,r6           ; r' = 0
+        movl    #0,r8           ; q' = 0
+
+        tstl    r4
         beql    666$            ; Uh-oh, the divisor is 0...
-        bgtr    3$
-        xorl2   #^X80000000,r9  ; fixup b2 so highest bit is 0
-        incl    r8              ; b' = 1
-3$:
-        tstl    r9
-        bneq    4$              ; if b2 is 0, we know that b' is 1
-        tstl    r3
-        bneq    666$            ; if higher half isn't 0, we overflow
-        movl    r2,r10          ; otherwise, we have our result
-        brb     42$             ; This is a success, really.
-4$:
-        ediv    r9,r6,r10,r11
-
-        tstl    r8
-        bneq    5$              ; If b' != 0, go to the other part
-;       addl3   r11,r11,r1
-;       addl2   r5,r1
-        brb     42$
-5$:
-        ashl    #1,r11,r11
-        subl2   r10,r11
-        addl2   r5,r11
-        bgeq    7$
-6$:
-        decl    r10
-        addl2   r4,r11
-        blss    6$
-7$:
-;       movl    r11,r1
+        bgtr    1$
+        rotl    #-1,r4,r4       ; If d is negative, shift it right.
+        bicl2   #^X80000000,r4  ; Since d is then a large number, the
+                                ; lowest bit is insignificant
+                                ; (contradict that, and I'll fix the problem!)
+1$:     
+        ediv    r4,r2,r2,r3     ; Do the actual division
+
+        tstl    r2
+        bgeq    3$
+        mnegl   r2,r2           ; if q < 0, negate it
+3$:     
+        tstl    r7
+        blss    4$
+        rotl    #3,r2,r2        ;   q = q << 3
+        bicl3   #^XFFFFFFF8,r2,r8 ;    q' gets the high bits from q
+        bicl3   #^X00000007,r2,r2
+        bsb     41$
+4$:                             ; else
+        rotl    #2,r2,r2        ;   q = q << 2
+        bicl3   #^XFFFFFFFC,r2,r8 ;   q' gets the high bits from q
+        bicl3   #^X00000003,r2,r2
+41$:
+        rotl    #3,r3,r3        ; r = r << 3
+        bicl3   #^XFFFFFFF8,r3,r6 ; r' gets the high bits from r
+        bicl3   #^X00000007,r3,r3
+        addl    r5,r3           ; r = r + l'
+
+        tstl    r7
+        bgeq    5$
+        bitl    #1,r7
+        beql    5$              ; if d' < 0 && d' & 1
+        subl    r2,r3           ;   [r',r] = [r',r] - [q',q]
+        sbwc    r8,r6
+45$:
+        bgeq    5$              ;   while r < 0
+        decl    r2              ;     [q',q] = [q',q] - 1
+        sbwc    #0,r8
+        addl    r7,r3           ;     [r',r] = [r',r] + d'
+        adwc    #0,r6
+        brb     45$
+
+; The return points are placed in the middle to keep a short distance from
+; all the branch points
 42$:
-        movl    r10,r0
+;       movl    r3,r1
+        movl    r2,r0
+        ret
 666$:
+        movl    #^XFFFFFFFF,r0
         ret
+
+5$:
+        tstl    r6
+        bneq    6$
+        cmpl    r3,r7
+        blssu   42$             ; while [r',r] >= d'
+6$:
+        subl    r7,r3           ;   [r',r] = [r',r] - d'
+        sbwc    #0,r6
+        incl    r2              ;   [q',q] = [q',q] + 1
+        adwc    #0,r8
+        brb     5$      
 
         .title  vax_bn_add_words  unsigned add of two arrays
 ;
diff --git a/src/lib/libcrypto/bn/bn.h b/src/lib/libcrypto/bn/bn.h
index b40682f831..3da6d8ced9 100644
--- a/src/lib/libcrypto/bn/bn.h
+++ b/src/lib/libcrypto/bn/bn.h
@@ -248,6 +248,8 @@ typedef struct bn_blinding_st
         BIGNUM *A;
         BIGNUM *Ai;
         BIGNUM *mod; /* just a reference */
+        unsigned long thread_id; /* added in OpenSSL 0.9.6j and 0.9.7b;
+                                  * used only by crypto/rsa/rsa_eay.c, rsa_lib.c */
         } BN_BLINDING;
 
 /* Used for montgomery multiplication */
diff --git a/src/lib/libcrypto/bn/bn_div.c b/src/lib/libcrypto/bn/bn_div.c
index f9a095e3b3..580d1201bc 100644
--- a/src/lib/libcrypto/bn/bn_div.c
+++ b/src/lib/libcrypto/bn/bn_div.c
@@ -150,6 +150,20 @@ int BN_div(BIGNUM *dv, BIGNUM *rem, const BIGNUM *m, const BIGNUM *d,
             q;                                  \
         })
 #  define REMAINDER_IS_ALREADY_CALCULATED
+#  elif defined(__x86_64) && defined(SIXTY_FOUR_BIT_LONG)
+   /*
+    * Same story here, but it's 128-bit by 64-bit division. Wow!
+    *                                   <appro@fy.chalmers.se>
+    */
+#  define bn_div_words(n0,n1,d0)                \
+        ({  asm volatile (                      \
+                "divq   %4"                     \
+                : "=a"(q), "=d"(rem)            \
+                : "a"(n1), "d"(n0), "g"(d0)     \
+                : "cc");                        \
+            q;                                  \
+        })
+#  define REMAINDER_IS_ALREADY_CALCULATED
 #  endif /* __<cpu> */
 # endif /* __GNUC__ */
 #endif /* OPENSSL_NO_ASM */
@@ -268,6 +282,11 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
                         q=(BN_ULONG)(((((BN_ULLONG)n0)<<BN_BITS2)|n1)/d0);
 #else
                         q=bn_div_words(n0,n1,d0);
+#ifdef BN_DEBUG_LEVITTE
+                        fprintf(stderr,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\
+X) -> 0x%08X\n",
+                                n0, n1, d0, q);
+#endif
 #endif
 
 #ifndef REMAINDER_IS_ALREADY_CALCULATED
@@ -292,11 +311,18 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
                         BN_ULONG t2l,t2h,ql,qh;
 
                         q=bn_div_words(n0,n1,d0);
+#ifdef BN_DEBUG_LEVITTE
+                        fprintf(stderr,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\
+X) -> 0x%08X\n",
+                                n0, n1, d0, q);
+#endif
 #ifndef REMAINDER_IS_ALREADY_CALCULATED
                         rem=(n1-q*d0)&BN_MASK2;
 #endif
 
-#ifdef BN_UMULT_HIGH
+#if defined(BN_UMULT_LOHI)
+                        BN_UMULT_LOHI(t2l,t2h,d1,q);
+#elif defined(BN_UMULT_HIGH)
                         t2l = d1 * q;
                         t2h = BN_UMULT_HIGH(d1,q);
 #else
diff --git a/src/lib/libcrypto/bn/bn_lcl.h b/src/lib/libcrypto/bn/bn_lcl.h
index 8a4dba375a..5614bc6164 100644
--- a/src/lib/libcrypto/bn/bn_lcl.h
+++ b/src/lib/libcrypto/bn/bn_lcl.h
@@ -230,6 +230,21 @@ struct bignum_ctx
              : "r"(a), "r"(b));         \
         ret;                    })
 #  endif        /* compiler */
+# elif defined(__x86_64) && defined(SIXTY_FOUR_BIT_LONG)
+#  if defined(__GNUC__)
+#   define BN_UMULT_HIGH(a,b)   ({      \
+        register BN_ULONG ret,discard;  \
+        asm ("mulq      %3"             \
+             : "=a"(discard),"=d"(ret)  \
+             : "a"(a), "g"(b)           \
+             : "cc");                   \
+        ret;                    })
+#   define BN_UMULT_LOHI(low,high,a,b)  \
+        asm ("mulq      %3"             \
+                : "=a"(low),"=d"(high)  \
+                : "a"(a),"g"(b)         \
+                : "cc");
+#  endif
 # endif         /* cpu */
 #endif          /* OPENSSL_NO_ASM */
 
@@ -337,7 +352,7 @@ struct bignum_ctx
 
 #define LBITS(a)        ((a)&BN_MASK2l)
 #define HBITS(a)        (((a)>>BN_BITS4)&BN_MASK2l)
-#define L2HBITS(a)      ((BN_ULONG)((a)&BN_MASK2l)<<BN_BITS4)
+#define L2HBITS(a)      (((a)<<BN_BITS4)&BN_MASK2)
 
 #define LLBITS(a)       ((a)&BN_MASKl)
 #define LHBITS(a)       (((a)>>BN_BITS2)&BN_MASKl)
@@ -353,7 +368,7 @@ struct bignum_ctx
         lt=(bl)*(lt); \
         m1=(bl)*(ht); \
         ht =(bh)*(ht); \
-        m=(m+m1)&BN_MASK2; if (m < m1) ht+=L2HBITS(1L); \
+        m=(m+m1)&BN_MASK2; if (m < m1) ht+=L2HBITS((BN_ULONG)1); \
         ht+=HBITS(m); \
         m1=L2HBITS(m); \
         lt=(lt+m1)&BN_MASK2; if (lt < m1) ht++; \
@@ -418,20 +433,19 @@ void bn_sqr_comba4(BN_ULONG *r,const BN_ULONG *a);
 int bn_cmp_words(const BN_ULONG *a,const BN_ULONG *b,int n);
 int bn_cmp_part_words(const BN_ULONG *a, const BN_ULONG *b,
         int cl, int dl);
+#if 0
+/* bn_mul.c rollback <appro> */
 void bn_mul_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,int n2,
         int dna,int dnb,BN_ULONG *t);
 void bn_mul_part_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,
         int n,int tna,int tnb,BN_ULONG *t);
+#endif
 void bn_sqr_recursive(BN_ULONG *r,const BN_ULONG *a, int n2, BN_ULONG *t);
 void bn_mul_low_normal(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b, int n);
 void bn_mul_low_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,int n2,
         BN_ULONG *t);
 void bn_mul_high(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,BN_ULONG *l,int n2,
         BN_ULONG *t);
-BN_ULONG bn_add_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
-        int cl, int dl);
-BN_ULONG bn_sub_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
-        int cl, int dl);
 
 #ifdef  __cplusplus
 }
diff --git a/src/lib/libcrypto/bn/bn_lib.c b/src/lib/libcrypto/bn/bn_lib.c
index ce2ae78419..463463cfcb 100644
--- a/src/lib/libcrypto/bn/bn_lib.c
+++ b/src/lib/libcrypto/bn/bn_lib.c
@@ -263,12 +263,12 @@ void BN_clear_free(BIGNUM *a)
         if (a == NULL) return;
         if (a->d != NULL)
                 {
-                memset(a->d,0,a->dmax*sizeof(a->d[0]));
+                OPENSSL_cleanse(a->d,a->dmax*sizeof(a->d[0]));
                 if (!(BN_get_flags(a,BN_FLG_STATIC_DATA)))
                         OPENSSL_free(a->d);
                 }
         i=BN_get_flags(a,BN_FLG_MALLOCED);
-        memset(a,0,sizeof(BIGNUM));
+        OPENSSL_cleanse(a,sizeof(BIGNUM));
         if (i)
                 OPENSSL_free(a);
         }
diff --git a/src/lib/libcrypto/bn/bn_mul.c b/src/lib/libcrypto/bn/bn_mul.c
index b03458d002..cb93ac3356 100644
--- a/src/lib/libcrypto/bn/bn_mul.c
+++ b/src/lib/libcrypto/bn/bn_mul.c
@@ -56,325 +56,10 @@
  * [including the GNU Public Licence.]
  */
 
-#ifndef BN_DEBUG
-# undef NDEBUG /* avoid conflicting definitions */
-# define NDEBUG
-#endif
-
 #include <stdio.h>
-#include <assert.h>
 #include "cryptlib.h"
 #include "bn_lcl.h"
 
-#if defined(OPENSSL_NO_ASM) || !(defined(__i386) || defined(__i386__)) || defined(__DJGPP__) /* Assembler implementation exists only for x86 */
-/* Here follows specialised variants of bn_add_words() and
-   bn_sub_words().  They have the property performing operations on
-   arrays of different sizes.  The sizes of those arrays is expressed through
-   cl, which is the common length ( basicall, min(len(a),len(b)) ), and dl,
-   which is the delta between the two lengths, calculated as len(a)-len(b).
-   All lengths are the number of BN_ULONGs...  For the operations that require
-   a result array as parameter, it must have the length cl+abs(dl).
-   These functions should probably end up in bn_asm.c as soon as there are
-   assembler counterparts for the systems that use assembler files.  */
-
-BN_ULONG bn_sub_part_words(BN_ULONG *r,
-        const BN_ULONG *a, const BN_ULONG *b,
-        int cl, int dl)
-        {
-        BN_ULONG c, t;
-
-        assert(cl >= 0);
-        c = bn_sub_words(r, a, b, cl);
-
-        if (dl == 0)
-                return c;
-
-        r += cl;
-        a += cl;
-        b += cl;
-
-        if (dl < 0)
-                {
-#ifdef BN_COUNT
-                fprintf(stderr, "  bn_sub_part_words %d + %d (dl < 0, c = %d)\n", cl, dl, c);
-#endif
-                for (;;)
-                        {
-                        t = b[0];
-                        r[0] = (0-t-c)&BN_MASK2;
-                        if (t != 0) c=1;
-                        if (++dl >= 0) break;
-
-                        t = b[1];
-                        r[1] = (0-t-c)&BN_MASK2;
-                        if (t != 0) c=1;
-                        if (++dl >= 0) break;
-
-                        t = b[2];
-                        r[2] = (0-t-c)&BN_MASK2;
-                        if (t != 0) c=1;
-                        if (++dl >= 0) break;
-
-                        t = b[3];
-                        r[3] = (0-t-c)&BN_MASK2;
-                        if (t != 0) c=1;
-                        if (++dl >= 0) break;
-
-                        b += 4;
-                        r += 4;
-                        }
-                }
-        else
-                {
-                int save_dl = dl;
-#ifdef BN_COUNT
-                fprintf(stderr, "  bn_sub_part_words %d + %d (dl > 0, c = %d)\n", cl, dl, c);
-#endif
-                while(c)
-                        {
-                        t = a[0];
-                        r[0] = (t-c)&BN_MASK2;
-                        if (t != 0) c=0;
-                        if (--dl <= 0) break;
-
-                        t = a[1];
-                        r[1] = (t-c)&BN_MASK2;
-                        if (t != 0) c=0;
-                        if (--dl <= 0) break;
-
-                        t = a[2];
-                        r[2] = (t-c)&BN_MASK2;
-                        if (t != 0) c=0;
-                        if (--dl <= 0) break;
-
-                        t = a[3];
-                        r[3] = (t-c)&BN_MASK2;
-                        if (t != 0) c=0;
-                        if (--dl <= 0) break;
-
-                        save_dl = dl;
-                        a += 4;
-                        r += 4;
-                        }
-                if (dl > 0)
-                        {
-#ifdef BN_COUNT
-                        fprintf(stderr, "  bn_sub_part_words %d + %d (dl > 0, c == 0)\n", cl, dl);
-#endif
-                        if (save_dl > dl)
-                                {
-                                switch (save_dl - dl)
-                                        {
-                                case 1:
-                                        r[1] = a[1];
-                                        if (--dl <= 0) break;
-                                case 2:
-                                        r[2] = a[2];
-                                        if (--dl <= 0) break;
-                                case 3:
-                                        r[3] = a[3];
-                                        if (--dl <= 0) break;
-                                        }
-                                a += 4;
-                                r += 4;
-                                }
-                        }
-                if (dl > 0)
-                        {
-#ifdef BN_COUNT
-                        fprintf(stderr, "  bn_sub_part_words %d + %d (dl > 0, copy)\n", cl, dl);
-#endif
-                        for(;;)
-                                {
-                                r[0] = a[0];
-                                if (--dl <= 0) break;
-                                r[1] = a[1];
-                                if (--dl <= 0) break;
-                                r[2] = a[2];
-                                if (--dl <= 0) break;
-                                r[3] = a[3];
-                                if (--dl <= 0) break;
-
-                                a += 4;
-                                r += 4;
-                                }
-                        }
-                }
-        return c;
-        }
-#endif
-
-BN_ULONG bn_add_part_words(BN_ULONG *r,
-        const BN_ULONG *a, const BN_ULONG *b,
-        int cl, int dl)
-        {
-        BN_ULONG c, l, t;
-
-        assert(cl >= 0);
-        c = bn_add_words(r, a, b, cl);
-
-        if (dl == 0)
-                return c;
-
-        r += cl;
-        a += cl;
-        b += cl;
-
-        if (dl < 0)
-                {
-                int save_dl = dl;
-#ifdef BN_COUNT
-                fprintf(stderr, "  bn_add_part_words %d + %d (dl < 0, c = %d)\n", cl, dl, c);
-#endif
-                while (c)
-                        {
-                        l=(c+b[0])&BN_MASK2;
-                        c=(l < c);
-                        r[0]=l;
-                        if (++dl >= 0) break;
-
-                        l=(c+b[1])&BN_MASK2;
-                        c=(l < c);
-                        r[1]=l;
-                        if (++dl >= 0) break;
-
-                        l=(c+b[2])&BN_MASK2;
-                        c=(l < c);
-                        r[2]=l;
-                        if (++dl >= 0) break;
-
-                        l=(c+b[3])&BN_MASK2;
-                        c=(l < c);
-                        r[3]=l;
-                        if (++dl >= 0) break;
-
-                        save_dl = dl;
-                        b+=4;
-                        r+=4;
-                        }
-                if (dl < 0)
-                        {
-#ifdef BN_COUNT
-                        fprintf(stderr, "  bn_add_part_words %d + %d (dl < 0, c == 0)\n", cl, dl);
-#endif
-                        if (save_dl < dl)
-                                {
-                                switch (dl - save_dl)
-                                        {
-                                case 1:
-                                        r[1] = b[1];
-                                        if (++dl >= 0) break;
-                                case 2:
-                                        r[2] = b[2];
-                                        if (++dl >= 0) break;
-                                case 3:
-                                        r[3] = b[3];
-                                        if (++dl >= 0) break;
-                                        }
-                                b += 4;
-                                r += 4;
-                                }
-                        }
-                if (dl < 0)
-                        {
-#ifdef BN_COUNT
-                        fprintf(stderr, "  bn_add_part_words %d + %d (dl < 0, copy)\n", cl, dl);
-#endif
-                        for(;;)
-                                {
-                                r[0] = b[0];
-                                if (++dl >= 0) break;
-                                r[1] = b[1];
-                                if (++dl >= 0) break;
-                                r[2] = b[2];
-                                if (++dl >= 0) break;
-                                r[3] = b[3];
-                                if (++dl >= 0) break;
-
-                                b += 4;
-                                r += 4;
-                                }
-                        }
-                }
-        else
-                {
-                int save_dl = dl;
-#ifdef BN_COUNT
-                fprintf(stderr, "  bn_add_part_words %d + %d (dl > 0)\n", cl, dl);
-#endif
-                while (c)
-                        {
-                        t=(a[0]+c)&BN_MASK2;
-                        c=(t < c);
-                        r[0]=t;
-                        if (--dl <= 0) break;
-
-                        t=(a[1]+c)&BN_MASK2;
-                        c=(t < c);
-                        r[1]=t;
-                        if (--dl <= 0) break;
-
-                        t=(a[2]+c)&BN_MASK2;
-                        c=(t < c);
-                        r[2]=t;
-                        if (--dl <= 0) break;
-
-                        t=(a[3]+c)&BN_MASK2;
-                        c=(t < c);
-                        r[3]=t;
-                        if (--dl <= 0) break;
-
-                        save_dl = dl;
-                        a+=4;
-                        r+=4;
-                        }
-#ifdef BN_COUNT
-                fprintf(stderr, "  bn_add_part_words %d + %d (dl > 0, c == 0)\n", cl, dl);
-#endif
-                if (dl > 0)
-                        {
-                        if (save_dl > dl)
-                                {
-                                switch (save_dl - dl)
-                                        {
-                                case 1:
-                                        r[1] = a[1];
-                                        if (--dl <= 0) break;
-                                case 2:
-                                        r[2] = a[2];
-                                        if (--dl <= 0) break;
-                                case 3:
-                                        r[3] = a[3];
-                                        if (--dl <= 0) break;
-                                        }
-                                a += 4;
-                                r += 4;
-                                }
-                        }
-                if (dl > 0)
-                        {
-#ifdef BN_COUNT
-                        fprintf(stderr, "  bn_add_part_words %d + %d (dl > 0, copy)\n", cl, dl);
-#endif
-                        for(;;)
-                                {
-                                r[0] = a[0];
-                                if (--dl <= 0) break;
-                                r[1] = a[1];
-                                if (--dl <= 0) break;
-                                r[2] = a[2];
-                                if (--dl <= 0) break;
-                                r[3] = a[3];
-                                if (--dl <= 0) break;
-
-                                a += 4;
-                                r += 4;
-                                }
-                        }
-                }
-        return c;
-        }
-
 #ifdef BN_RECURSION
 /* Karatsuba recursive multiplication algorithm
  * (cf. Knuth, The Art of Computer Programming, Vol. 2) */
@@ -390,15 +75,14 @@ BN_ULONG bn_add_part_words(BN_ULONG *r,
  * a[1]*b[1]
  */
 void bn_mul_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
-        int dna, int dnb, BN_ULONG *t)
+             BN_ULONG *t)
         {
         int n=n2/2,c1,c2;
-        int tna=n+dna, tnb=n+dnb;
         unsigned int neg,zero;
         BN_ULONG ln,lo,*p;
 
 # ifdef BN_COUNT
-        fprintf(stderr," bn_mul_recursive %d * %d\n",n2,n2);
+        printf(" bn_mul_recursive %d * %d\n",n2,n2);
 # endif
 # ifdef BN_MUL_COMBA
 #  if 0
@@ -408,40 +92,34 @@ void bn_mul_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
                 return;
                 }
 #  endif
-        /* Only call bn_mul_comba 8 if n2 == 8 and the
-         * two arrays are complete [steve]
-         */
-        if (n2 == 8 && dna == 0 && dnb == 0)
+        if (n2 == 8)
                 {
                 bn_mul_comba8(r,a,b);
                 return; 
                 }
 # endif /* BN_MUL_COMBA */
-        /* Else do normal multiply */
         if (n2 < BN_MUL_RECURSIVE_SIZE_NORMAL)
                 {
-                bn_mul_normal(r,a,n2+dna,b,n2+dnb);
-                if ((dna + dnb) < 0)
-                        memset(&r[2*n2 + dna + dnb], 0,
-                                sizeof(BN_ULONG) * -(dna + dnb));
+                /* This should not happen */
+                bn_mul_normal(r,a,n2,b,n2);
                 return;
                 }
         /* r=(a[0]-a[1])*(b[1]-b[0]) */
-        c1=bn_cmp_part_words(a,&(a[n]),tna,n-tna);
-        c2=bn_cmp_part_words(&(b[n]),b,tnb,tnb-n);
+        c1=bn_cmp_words(a,&(a[n]),n);
+        c2=bn_cmp_words(&(b[n]),b,n);
         zero=neg=0;
         switch (c1*3+c2)
                 {
         case -4:
-                bn_sub_part_words(t,      &(a[n]),a,      tna,tna-n); /* - */
-                bn_sub_part_words(&(t[n]),b,      &(b[n]),tnb,n-tnb); /* - */
+                bn_sub_words(t,      &(a[n]),a,      n); /* - */
+                bn_sub_words(&(t[n]),b,      &(b[n]),n); /* - */
                 break;
         case -3:
                 zero=1;
                 break;
         case -2:
-                bn_sub_part_words(t,      &(a[n]),a,      tna,tna-n); /* - */
-                bn_sub_part_words(&(t[n]),&(b[n]),b,      tnb,tnb-n); /* + */
+                bn_sub_words(t,      &(a[n]),a,      n); /* - */
+                bn_sub_words(&(t[n]),&(b[n]),b,      n); /* + */
                 neg=1;
                 break;
         case -1:
@@ -450,22 +128,21 @@ void bn_mul_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
                 zero=1;
                 break;
         case 2:
-                bn_sub_part_words(t,      a,      &(a[n]),tna,n-tna); /* + */
-                bn_sub_part_words(&(t[n]),b,      &(b[n]),tnb,n-tnb); /* - */
+                bn_sub_words(t,      a,      &(a[n]),n); /* + */
+                bn_sub_words(&(t[n]),b,      &(b[n]),n); /* - */
                 neg=1;
                 break;
         case 3:
                 zero=1;
                 break;
         case 4:
-                bn_sub_part_words(t,      a,      &(a[n]),tna,n-tna);
-                bn_sub_part_words(&(t[n]),&(b[n]),b,      tnb,tnb-n);
+                bn_sub_words(t,      a,      &(a[n]),n);
+                bn_sub_words(&(t[n]),&(b[n]),b,      n);
                 break;
                 }
 
 # ifdef BN_MUL_COMBA
-        if (n == 4 && dna == 0 && dnb == 0) /* XXX: bn_mul_comba4 could take
-                                               extra args to do this well */
+        if (n == 4)
                 {
                 if (!zero)
                         bn_mul_comba4(&(t[n2]),t,&(t[n]));
@@ -475,9 +152,7 @@ void bn_mul_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
                 bn_mul_comba4(r,a,b);
                 bn_mul_comba4(&(r[n2]),&(a[n]),&(b[n]));
                 }
-        else if (n == 8 && dna == 0 && dnb == 0) /* XXX: bn_mul_comba8 could
-                                                    take extra args to do this
-                                                    well */
+        else if (n == 8)
                 {
                 if (!zero)
                         bn_mul_comba8(&(t[n2]),t,&(t[n]));
@@ -492,11 +167,11 @@ void bn_mul_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
                 {
                 p= &(t[n2*2]);
                 if (!zero)
-                        bn_mul_recursive(&(t[n2]),t,&(t[n]),n,0,0,p);
+                        bn_mul_recursive(&(t[n2]),t,&(t[n]),n,p);
                 else
                         memset(&(t[n2]),0,n2*sizeof(BN_ULONG));
-                bn_mul_recursive(r,a,b,n,0,0,p);
-                bn_mul_recursive(&(r[n2]),&(a[n]),&(b[n]),n,dna,dnb,p);
+                bn_mul_recursive(r,a,b,n,p);
+                bn_mul_recursive(&(r[n2]),&(a[n]),&(b[n]),n,p);
                 }
 
         /* t[32] holds (a[0]-a[1])*(b[1]-b[0]), c1 is the sign
@@ -545,39 +220,39 @@ void bn_mul_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
 
 /* n+tn is the word length
  * t needs to be n*4 is size, as does r */
-void bn_mul_part_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n,
-             int tna, int tnb, BN_ULONG *t)
+void bn_mul_part_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int tn,
+             int n, BN_ULONG *t)
         {
         int i,j,n2=n*2;
         unsigned int c1,c2,neg,zero;
         BN_ULONG ln,lo,*p;
 
 # ifdef BN_COUNT
-        fprintf(stderr," bn_mul_part_recursive (%d+%d) * (%d+%d)\n",
-                tna, n, tnb, n);
+        printf(" bn_mul_part_recursive %d * %d\n",tn+n,tn+n);
 # endif
         if (n < 8)
                 {
-                bn_mul_normal(r,a,n+tna,b,n+tnb);
+                i=tn+n;
+                bn_mul_normal(r,a,i,b,i);
                 return;
                 }
 
         /* r=(a[0]-a[1])*(b[1]-b[0]) */
-        c1=bn_cmp_part_words(a,&(a[n]),tna,n-tna);
-        c2=bn_cmp_part_words(&(b[n]),b,tnb,tnb-n);
+        c1=bn_cmp_words(a,&(a[n]),n);
+        c2=bn_cmp_words(&(b[n]),b,n);
         zero=neg=0;
         switch (c1*3+c2)
                 {
         case -4:
-                bn_sub_part_words(t,      &(a[n]),a,      tna,tna-n); /* - */
-                bn_sub_part_words(&(t[n]),b,      &(b[n]),tnb,n-tnb); /* - */
+                bn_sub_words(t,      &(a[n]),a,      n); /* - */
+                bn_sub_words(&(t[n]),b,      &(b[n]),n); /* - */
                 break;
         case -3:
                 zero=1;
                 /* break; */
         case -2:
-                bn_sub_part_words(t,      &(a[n]),a,      tna,tna-n); /* - */
-                bn_sub_part_words(&(t[n]),&(b[n]),b,      tnb,tnb-n); /* + */
+                bn_sub_words(t,      &(a[n]),a,      n); /* - */
+                bn_sub_words(&(t[n]),&(b[n]),b,      n); /* + */
                 neg=1;
                 break;
         case -1:
@@ -586,16 +261,16 @@ void bn_mul_part_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n,
                 zero=1;
                 /* break; */
         case 2:
-                bn_sub_part_words(t,      a,      &(a[n]),tna,n-tna); /* + */
-                bn_sub_part_words(&(t[n]),b,      &(b[n]),tnb,n-tnb); /* - */
+                bn_sub_words(t,      a,      &(a[n]),n); /* + */
+                bn_sub_words(&(t[n]),b,      &(b[n]),n); /* - */
                 neg=1;
                 break;
         case 3:
                 zero=1;
                 /* break; */
         case 4:
-                bn_sub_part_words(t,      a,      &(a[n]),tna,n-tna);
-                bn_sub_part_words(&(t[n]),&(b[n]),b,      tnb,tnb-n);
+                bn_sub_words(t,      a,      &(a[n]),n);
+                bn_sub_words(&(t[n]),&(b[n]),b,      n);
                 break;
                 }
                 /* The zero case isn't yet implemented here. The speedup
@@ -614,59 +289,54 @@ void bn_mul_part_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n,
                 {
                 bn_mul_comba8(&(t[n2]),t,&(t[n]));
                 bn_mul_comba8(r,a,b);
-                bn_mul_normal(&(r[n2]),&(a[n]),tna,&(b[n]),tnb);
-                memset(&(r[n2+tna+tnb]),0,sizeof(BN_ULONG)*(n2-tna-tnb));
+                bn_mul_normal(&(r[n2]),&(a[n]),tn,&(b[n]),tn);
+                memset(&(r[n2+tn*2]),0,sizeof(BN_ULONG)*(n2-tn*2));
                 }
         else
                 {
                 p= &(t[n2*2]);
-                bn_mul_recursive(&(t[n2]),t,&(t[n]),n,0,0,p);
-                bn_mul_recursive(r,a,b,n,0,0,p);
+                bn_mul_recursive(&(t[n2]),t,&(t[n]),n,p);
+                bn_mul_recursive(r,a,b,n,p);
                 i=n/2;
                 /* If there is only a bottom half to the number,
                  * just do it */
-                if (tna > tnb)
-                        j = tna - i;
-                else
-                        j = tnb - i;
+                j=tn-i;
                 if (j == 0)
                         {
-                        bn_mul_recursive(&(r[n2]),&(a[n]),&(b[n]),
-                                i,tna-i,tnb-i,p);
+                        bn_mul_recursive(&(r[n2]),&(a[n]),&(b[n]),i,p);
                         memset(&(r[n2+i*2]),0,sizeof(BN_ULONG)*(n2-i*2));
                         }
                 else if (j > 0) /* eg, n == 16, i == 8 and tn == 11 */
                                 {
                                 bn_mul_part_recursive(&(r[n2]),&(a[n]),&(b[n]),
-                                        i,tna-i,tnb-i,p);
-                                memset(&(r[n2+tna+tnb]),0,
-                                        sizeof(BN_ULONG)*(n2-tna-tnb));
+                                        j,i,p);
+                                memset(&(r[n2+tn*2]),0,
+                                        sizeof(BN_ULONG)*(n2-tn*2));
                                 }
                 else /* (j < 0) eg, n == 16, i == 8 and tn == 5 */
                         {
                         memset(&(r[n2]),0,sizeof(BN_ULONG)*n2);
-                        if (tna < BN_MUL_RECURSIVE_SIZE_NORMAL
-                                && tnb < BN_MUL_RECURSIVE_SIZE_NORMAL)
+                        if (tn < BN_MUL_RECURSIVE_SIZE_NORMAL)
                                 {
-                                bn_mul_normal(&(r[n2]),&(a[n]),tna,&(b[n]),tnb);
+                                bn_mul_normal(&(r[n2]),&(a[n]),tn,&(b[n]),tn);
                                 }
                         else
                                 {
                                 for (;;)
                                         {
                                         i/=2;
-                                        if (i < tna && i < tnb)
+                                        if (i < tn)
                                                 {
                                                 bn_mul_part_recursive(&(r[n2]),
                                                         &(a[n]),&(b[n]),
-                                                        i,tna-i,tnb-i,p);
+                                                        tn-i,i,p);
                                                 break;
                                                 }
-                                        else if (i <= tna && i <= tnb)
+                                        else if (i == tn)
                                                 {
                                                 bn_mul_recursive(&(r[n2]),
                                                         &(a[n]),&(b[n]),
-                                                        i,tna-i,tnb-i,p);
+                                                        i,p);
                                                 break;
                                                 }
                                         }
@@ -727,10 +397,10 @@ void bn_mul_low_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
         int n=n2/2;
 
 # ifdef BN_COUNT
-        fprintf(stderr," bn_mul_low_recursive %d * %d\n",n2,n2);
+        printf(" bn_mul_low_recursive %d * %d\n",n2,n2);
 # endif
 
-        bn_mul_recursive(r,a,b,n,0,0,&(t[0]));
+        bn_mul_recursive(r,a,b,n,&(t[0]));
         if (n >= BN_MUL_LOW_RECURSIVE_SIZE_NORMAL)
                 {
                 bn_mul_low_recursive(&(t[0]),&(a[0]),&(b[n]),n,&(t[n2]));
@@ -761,7 +431,7 @@ void bn_mul_high(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, BN_ULONG *l, int n2,
         BN_ULONG ll,lc,*lp,*mp;
 
 # ifdef BN_COUNT
-        fprintf(stderr," bn_mul_high %d * %d\n",n2,n2);
+        printf(" bn_mul_high %d * %d\n",n2,n2);
 # endif
         n=n2/2;
 
@@ -814,8 +484,8 @@ void bn_mul_high(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, BN_ULONG *l, int n2,
         else
 # endif
                 {
-                bn_mul_recursive(&(t[0]),&(r[0]),&(r[n]),n,0,0,&(t[n2]));
-                bn_mul_recursive(r,&(a[n]),&(b[n]),n,0,0,&(t[n2]));
+                bn_mul_recursive(&(t[0]),&(r[0]),&(r[n]),n,&(t[n2]));
+                bn_mul_recursive(r,&(a[n]),&(b[n]),n,&(t[n2]));
                 }
 
         /* s0 == low(al*bl)
@@ -940,19 +610,19 @@ void bn_mul_high(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, BN_ULONG *l, int n2,
 
 int BN_mul(BIGNUM *r, const BIGNUM *a, const BIGNUM *b, BN_CTX *ctx)
         {
-        int ret=0;
         int top,al,bl;
         BIGNUM *rr;
+        int ret = 0;
 #if defined(BN_MUL_COMBA) || defined(BN_RECURSION)
         int i;
 #endif
 #ifdef BN_RECURSION
-        BIGNUM *t=NULL;
-        int j=0,k;
+        BIGNUM *t;
+        int j,k;
 #endif
 
 #ifdef BN_COUNT
-        fprintf(stderr,"BN_mul %d * %d\n",a->top,b->top);
+        printf("BN_mul %d * %d\n",a->top,b->top);
 #endif
 
         bn_check_top(a);
@@ -1005,55 +675,21 @@ int BN_mul(BIGNUM *r, const BIGNUM *a, const BIGNUM *b, BN_CTX *ctx)
 #ifdef BN_RECURSION
         if ((al >= BN_MULL_SIZE_NORMAL) && (bl >= BN_MULL_SIZE_NORMAL))
                 {
-                if (i >= -1 && i <= 1)
+                if (i == 1 && !BN_get_flags(b,BN_FLG_STATIC_DATA) && bl<b->dmax)
                         {
-                        int sav_j =0;
-                        /* Find out the power of two lower or equal
-                           to the longest of the two numbers */
-                        if (i >= 0)
-                                {
-                                j = BN_num_bits_word((BN_ULONG)al);
-                                }
-                        if (i == -1)
-                                {
-                                j = BN_num_bits_word((BN_ULONG)bl);
-                                }
-                        sav_j = j;
-                        j = 1<<(j-1);
-                        assert(j <= al || j <= bl);
-                        k = j+j;
-                        t = BN_CTX_get(ctx);
-                        if (al > j || bl > j)
-                                {
-                                bn_wexpand(t,k*4);
-                                bn_wexpand(rr,k*4);
-                                bn_mul_part_recursive(rr->d,a->d,b->d,
-                                        j,al-j,bl-j,t->d);
-                                }
-                        else    /* al <= j || bl <= j */
-                                {
-                                bn_wexpand(t,k*2);
-                                bn_wexpand(rr,k*2);
-                                bn_mul_recursive(rr->d,a->d,b->d,
-                                        j,al-j,bl-j,t->d);
-                                }
-                        rr->top=top;
-                        goto end;
-                        }
-#if 0
-                if (i == 1 && !BN_get_flags(b,BN_FLG_STATIC_DATA))
-                        {
-                        BIGNUM *tmp_bn = (BIGNUM *)b;
-                        if (bn_wexpand(tmp_bn,al) == NULL) goto err;
-                        tmp_bn->d[bl]=0;
+#if 0   /* tribute to const-ification, bl<b->dmax above covers for this */
+                        if (bn_wexpand(b,al) == NULL) goto err;
+#endif
+                        b->d[bl]=0;
                         bl++;
                         i--;
                         }
-                else if (i == -1 && !BN_get_flags(a,BN_FLG_STATIC_DATA))
+                else if (i == -1 && !BN_get_flags(a,BN_FLG_STATIC_DATA) && al<a->dmax)
                         {
-                        BIGNUM *tmp_bn = (BIGNUM *)a;
-                        if (bn_wexpand(tmp_bn,bl) == NULL) goto err;
-                        tmp_bn->d[al]=0;
+#if 0   /* tribute to const-ification, al<a->dmax above covers for this */
+                        if (bn_wexpand(a,bl) == NULL) goto err;
+#endif
+                        a->d[al]=0;
                         al++;
                         i++;
                         }
@@ -1070,17 +706,26 @@ int BN_mul(BIGNUM *r, const BIGNUM *a, const BIGNUM *b, BN_CTX *ctx)
                                 if (bn_wexpand(t,k*2) == NULL) goto err;
                                 if (bn_wexpand(rr,k*2) == NULL) goto err;
                                 bn_mul_recursive(rr->d,a->d,b->d,al,t->d);
+                                rr->top=top;
+                                goto end;
                                 }
+#if 0   /* tribute to const-ification, rsa/dsa performance is not affected */
                         else
                                 {
-                                if (bn_wexpand(t,k*4) == NULL) goto err;
-                                if (bn_wexpand(rr,k*4) == NULL) goto err;
+                                if (bn_wexpand(a,k) == NULL ) goto err;
+                                if (bn_wexpand(b,k) == NULL ) goto err;
+                                if (bn_wexpand(t,k*4) == NULL ) goto err;
+                                if (bn_wexpand(rr,k*4) == NULL ) goto err;
+                                for (i=a->top; i<k; i++)
+                                        a->d[i]=0;
+                                for (i=b->top; i<k; i++)
+                                        b->d[i]=0;
                                 bn_mul_part_recursive(rr->d,a->d,b->d,al-j,j,t->d);
                                 }
                         rr->top=top;
                         goto end;
-                        }
 #endif
+                        }
                 }
 #endif /* BN_RECURSION */
         if (bn_wexpand(rr,top) == NULL) goto err;
@@ -1103,7 +748,7 @@ void bn_mul_normal(BN_ULONG *r, BN_ULONG *a, int na, BN_ULONG *b, int nb)
         BN_ULONG *rr;
 
 #ifdef BN_COUNT
-        fprintf(stderr," bn_mul_normal %d * %d\n",na,nb);
+        printf(" bn_mul_normal %d * %d\n",na,nb);
 #endif
 
         if (na < nb)
@@ -1116,13 +761,7 @@ void bn_mul_normal(BN_ULONG *r, BN_ULONG *a, int na, BN_ULONG *b, int nb)
 
                 }
         rr= &(r[na]);
-        if (nb <= 0)
-                {
-                (void)bn_mul_words(r,a,na,0);
-                return;
-                }
-        else
-                rr[0]=bn_mul_words(r,a,na,b[0]);
+        rr[0]=bn_mul_words(r,a,na,b[0]);
 
         for (;;)
                 {
@@ -1143,7 +782,7 @@ void bn_mul_normal(BN_ULONG *r, BN_ULONG *a, int na, BN_ULONG *b, int nb)
 void bn_mul_low_normal(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n)
         {
 #ifdef BN_COUNT
-        fprintf(stderr," bn_mul_low_normal %d * %d\n",n,n);
+        printf(" bn_mul_low_normal %d * %d\n",n,n);
 #endif
         bn_mul_words(r,a,n,b[0]);
 
diff --git a/src/lib/libcrypto/bn/bn_prime.c b/src/lib/libcrypto/bn/bn_prime.c
index 918b9237c6..e072d9255c 100644
--- a/src/lib/libcrypto/bn/bn_prime.c
+++ b/src/lib/libcrypto/bn/bn_prime.c
@@ -140,6 +140,7 @@ BIGNUM *BN_generate_prime(BIGNUM *ret, int bits, int safe,
         BN_CTX *ctx;
         int checks = BN_prime_checks_for_size(bits);
 
+        BN_init(&t);
         ctx=BN_CTX_new();
         if (ctx == NULL) goto err;
         if (ret == NULL)
@@ -148,7 +149,6 @@ BIGNUM *BN_generate_prime(BIGNUM *ret, int bits, int safe,
                 }
         else
                 rnd=ret;
-        BN_init(&t);
 loop: 
         /* make a random number and set the top and bottom bits */
         if (add == NULL)
diff --git a/src/lib/libcrypto/bn/bn_rand.c b/src/lib/libcrypto/bn/bn_rand.c
index 9e08ccd22e..893c9d2af9 100644
--- a/src/lib/libcrypto/bn/bn_rand.c
+++ b/src/lib/libcrypto/bn/bn_rand.c
@@ -201,7 +201,7 @@ static int bnrand(int pseudorand, BIGNUM *rnd, int bits, int top, int bottom)
 err:
         if (buf != NULL)
                 {
-                memset(buf,0,bytes);
+                OPENSSL_cleanse(buf,bytes);
                 OPENSSL_free(buf);
                 }
         return(ret);
diff --git a/src/lib/libcrypto/bn/bn_word.c b/src/lib/libcrypto/bn/bn_word.c
index cd59baa2c4..988e0ca7b3 100644
--- a/src/lib/libcrypto/bn/bn_word.c
+++ b/src/lib/libcrypto/bn/bn_word.c
@@ -123,7 +123,10 @@ int BN_add_word(BIGNUM *a, BN_ULONG w)
         i=0;
         for (;;)
                 {
-                l=(a->d[i]+(BN_ULONG)w)&BN_MASK2;
+                if (i >= a->top)
+                        l=w;
+                else
+                        l=(a->d[i]+(BN_ULONG)w)&BN_MASK2;
                 a->d[i]=l;
                 if (w > l)
                         w=1;
diff --git a/src/lib/libcrypto/bn/bntest.c b/src/lib/libcrypto/bn/bntest.c
index 8158a67374..3c8c540387 100644
--- a/src/lib/libcrypto/bn/bntest.c
+++ b/src/lib/libcrypto/bn/bntest.c
@@ -68,10 +68,6 @@
 #include <openssl/x509.h>
 #include <openssl/err.h>
 
-#ifdef OPENSSL_SYS_WINDOWS
-#include "../bio/bss_file.c"
-#endif
-
 const int num0 = 100; /* number of tests */
 const int num1 = 50;  /* additional tests for some functions */
 const int num2 = 5;   /* number of tests for slow functions */
@@ -96,11 +92,6 @@ int test_sqrt(BIO *bp,BN_CTX *ctx);
 int rand_neg(void);
 static int results=0;
 
-#ifdef OPENSSL_NO_STDIO
-#define APPS_WIN16
-#include "bss_file.c"
-#endif
-
 static unsigned char lst[]="\xC6\x4F\x43\x04\x2A\xEA\xCA\x6E\x58\x36\x80\x5B\xE8\xC9"
 "\x9B\x04\x5D\x48\x36\xC2\xFD\x16\xC9\x64\xF0";
 
@@ -141,10 +132,10 @@ int main(int argc, char *argv[])
 
 
         ctx=BN_CTX_new();
-        if (ctx == NULL) exit(1);
+        if (ctx == NULL) EXIT(1);
 
         out=BIO_new(BIO_s_file());
-        if (out == NULL) exit(1);
+        if (out == NULL) EXIT(1);
         if (outfile == NULL)
                 {
                 BIO_set_fp(out,stdout,BIO_NOCLOSE);
@@ -154,7 +145,7 @@ int main(int argc, char *argv[])
                 if (!BIO_write_filename(out,outfile))
                         {
                         perror(outfile);
-                        exit(1);
+                        EXIT(1);
                         }
                 }
 
@@ -238,14 +229,14 @@ int main(int argc, char *argv[])
         BIO_free(out);
 
 /**/
-        exit(0);
+        EXIT(0);
 err:
         BIO_puts(out,"1\n"); /* make sure the Perl script fed by bc notices
                               * the failure, see test_bn in test/Makefile.ssl*/
         BIO_flush(out);
         ERR_load_crypto_strings();
         ERR_print_errors_fp(stderr);
-        exit(1);
+        EXIT(1);
         return(1);
         }
 
@@ -488,7 +479,7 @@ int test_mul(BIO *bp)
         BN_CTX *ctx;
 
         ctx = BN_CTX_new();
-        if (ctx == NULL) exit(1);
+        if (ctx == NULL) EXIT(1);
         
         BN_init(&a);
         BN_init(&b);
@@ -726,7 +717,7 @@ int test_mod_mul(BIO *bp, BN_CTX *ctx)
                         while ((l=ERR_get_error()))
                                 fprintf(stderr,"ERROR:%s\n",
                                         ERR_error_string(l,NULL));
-                        exit(1);
+                        EXIT(1);
                         }
                 if (bp != NULL)
                         {
diff --git a/src/lib/libcrypto/bn/divtest.c b/src/lib/libcrypto/bn/divtest.c
index 13ba86e3c4..d3fc688f33 100644
--- a/src/lib/libcrypto/bn/divtest.c
+++ b/src/lib/libcrypto/bn/divtest.c
@@ -1,7 +1,7 @@
 #include <openssl/bn.h>
 #include <openssl/rand.h>
 
-static int rand(n)
+static int Rand(n)
 {
     unsigned char x[2];
     RAND_pseudo_bytes(x,2);
@@ -26,8 +26,8 @@ main()
     BN_CTX *ctx=BN_CTX_new();
 
     for(;;) {
-        BN_pseudo_rand(a,rand(),0,0);
-        BN_pseudo_rand(b,rand(),0,0);
+        BN_pseudo_rand(a,Rand(),0,0);
+        BN_pseudo_rand(b,Rand(),0,0);
         if (BN_is_zero(b)) continue;
 
         BN_RECP_CTX_set(recp,b,ctx);
diff --git a/src/lib/libcrypto/bn/exptest.c b/src/lib/libcrypto/bn/exptest.c
index 5ca570d1a8..b09cf88705 100644
--- a/src/lib/libcrypto/bn/exptest.c
+++ b/src/lib/libcrypto/bn/exptest.c
@@ -59,13 +59,13 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+
+#include "../e_os.h"
+
 #include <openssl/bio.h>
 #include <openssl/bn.h>
 #include <openssl/rand.h>
 #include <openssl/err.h>
-#ifdef OPENSSL_SYS_WINDOWS
-#include "../bio/bss_file.c"
-#endif
 
 #define NUM_BITS        (BN_BITS*2)
 
@@ -86,7 +86,7 @@ int main(int argc, char *argv[])
         ERR_load_BN_strings();
 
         ctx=BN_CTX_new();
-        if (ctx == NULL) exit(1);
+        if (ctx == NULL) EXIT(1);
         r_mont=BN_new();
         r_recp=BN_new();
         r_simple=BN_new();
@@ -99,7 +99,7 @@ int main(int argc, char *argv[])
 
         out=BIO_new(BIO_s_file());
 
-        if (out == NULL) exit(1);
+        if (out == NULL) EXIT(1);
         BIO_set_fp(out,stdout,BIO_NOCLOSE);
 
         for (i=0; i<200; i++)
@@ -124,7 +124,7 @@ int main(int argc, char *argv[])
                         {
                         printf("BN_mod_exp_mont() problems\n");
                         ERR_print_errors(out);
-                        exit(1);
+                        EXIT(1);
                         }
 
                 ret=BN_mod_exp_recp(r_recp,a,b,m,ctx);
@@ -132,7 +132,7 @@ int main(int argc, char *argv[])
                         {
                         printf("BN_mod_exp_recp() problems\n");
                         ERR_print_errors(out);
-                        exit(1);
+                        EXIT(1);
                         }
 
                 ret=BN_mod_exp_simple(r_simple,a,b,m,ctx);
@@ -140,7 +140,7 @@ int main(int argc, char *argv[])
                         {
                         printf("BN_mod_exp_simple() problems\n");
                         ERR_print_errors(out);
-                        exit(1);
+                        EXIT(1);
                         }
 
                 if (BN_cmp(r_simple, r_mont) == 0
@@ -163,7 +163,7 @@ int main(int argc, char *argv[])
                         printf("\nrecp     ="); BN_print(out,r_recp);
                         printf("\nmont     ="); BN_print(out,r_mont);
                         printf("\n");
-                        exit(1);
+                        EXIT(1);
                         }
                 }
         BN_free(r_mont);
@@ -177,11 +177,11 @@ int main(int argc, char *argv[])
         CRYPTO_mem_leaks(out);
         BIO_free(out);
         printf(" done\n");
-        exit(0);
+        EXIT(0);
 err:
         ERR_load_crypto_strings();
         ERR_print_errors(out);
-        exit(1);
+        EXIT(1);
         return(1);
         }