summaryrefslogtreecommitdiff
path: root/src/crypto/bn/asm/x86_64-mont.pl
diff options
context:
space:
mode:
Diffstat (limited to 'src/crypto/bn/asm/x86_64-mont.pl')
-rwxr-xr-xsrc/crypto/bn/asm/x86_64-mont.pl316
1 files changed, 230 insertions, 86 deletions
diff --git a/src/crypto/bn/asm/x86_64-mont.pl b/src/crypto/bn/asm/x86_64-mont.pl
index 60e0111a..5775f658 100755
--- a/src/crypto/bn/asm/x86_64-mont.pl
+++ b/src/crypto/bn/asm/x86_64-mont.pl
@@ -84,6 +84,10 @@ $code=<<___;
.type bn_mul_mont,\@function,6
.align 16
bn_mul_mont:
+.cfi_startproc
+ mov ${num}d,${num}d
+ mov %rsp,%rax
+.cfi_def_cfa_register %rax
test \$3,${num}d
jnz .Lmul_enter
cmp \$8,${num}d
@@ -102,20 +106,50 @@ $code.=<<___;
.align 16
.Lmul_enter:
push %rbx
+.cfi_push %rbx
push %rbp
+.cfi_push %rbp
push %r12
+.cfi_push %r12
push %r13
+.cfi_push %r13
push %r14
+.cfi_push %r14
push %r15
+.cfi_push %r15
- mov ${num}d,${num}d
- lea 2($num),%r10
+ neg $num
mov %rsp,%r11
- neg %r10
- lea (%rsp,%r10,8),%rsp # tp=alloca(8*(num+2))
- and \$-1024,%rsp # minimize TLB usage
+ lea -16(%rsp,$num,8),%r10 # future alloca(8*(num+2))
+ neg $num # restore $num
+ and \$-1024,%r10 # minimize TLB usage
- mov %r11,8(%rsp,$num,8) # tp[num+1]=%rsp
+ # An OS-agnostic version of __chkstk.
+ #
+ # Some OSes (Windows) insist on stack being "wired" to
+ # physical memory in strictly sequential manner, i.e. if stack
+ # allocation spans two pages, then reference to farmost one can
+ # be punishable by SEGV. But page walking can do good even on
+ # other OSes, because it guarantees that villain thread hits
+ # the guard page before it can make damage to innocent one...
+ sub %r10,%r11
+ and \$-4096,%r11
+ lea (%r10,%r11),%rsp
+ mov (%rsp),%r11
+ cmp %r10,%rsp
+ ja .Lmul_page_walk
+ jmp .Lmul_page_walk_done
+
+.align 16
+.Lmul_page_walk:
+ lea -4096(%rsp),%rsp
+ mov (%rsp),%r11
+ cmp %r10,%rsp
+ ja .Lmul_page_walk
+.Lmul_page_walk_done:
+
+ mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp
+.cfi_cfa_expression %rsp+8,$num,8,mul,plus,deref,+8
.Lmul_body:
mov $bp,%r12 # reassign $bp
___
@@ -265,36 +299,46 @@ $code.=<<___;
mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i]
mov 8($ap,$i,8),%rax # tp[i+1]
lea 1($i),$i # i++
- dec $j # doesn't affect CF!
+ dec $j # doesnn't affect CF!
jnz .Lsub
sbb \$0,%rax # handle upmost overflow bit
xor $i,$i
+ and %rax,$ap
+ not %rax
+ mov $rp,$np
+ and %rax,$np
mov $num,$j # j=num
+ or $np,$ap # ap=borrow?tp:rp
.align 16
.Lcopy: # copy or in-place refresh
- mov (%rsp,$i,8),$ap
- mov ($rp,$i,8),$np
- xor $np,$ap # conditional select:
- and %rax,$ap # ((ap ^ np) & %rax) ^ np
- xor $np,$ap # ap = borrow?tp:rp
+ mov ($ap,$i,8),%rax
mov $i,(%rsp,$i,8) # zap temporary vector
- mov $ap,($rp,$i,8) # rp[i]=tp[i]
+ mov %rax,($rp,$i,8) # rp[i]=tp[i]
lea 1($i),$i
sub \$1,$j
jnz .Lcopy
mov 8(%rsp,$num,8),%rsi # restore %rsp
+.cfi_def_cfa %rsi,8
mov \$1,%rax
- mov (%rsi),%r15
- mov 8(%rsi),%r14
- mov 16(%rsi),%r13
- mov 24(%rsi),%r12
- mov 32(%rsi),%rbp
- mov 40(%rsi),%rbx
- lea 48(%rsi),%rsp
+ mov -48(%rsi),%r15
+.cfi_restore %r15
+ mov -40(%rsi),%r14
+.cfi_restore %r14
+ mov -32(%rsi),%r13
+.cfi_restore %r13
+ mov -24(%rsi),%r12
+.cfi_restore %r12
+ mov -16(%rsi),%rbp
+.cfi_restore %rbp
+ mov -8(%rsi),%rbx
+.cfi_restore %rbx
+ lea (%rsi),%rsp
+.cfi_def_cfa_register %rsp
.Lmul_epilogue:
ret
+.cfi_endproc
.size bn_mul_mont,.-bn_mul_mont
___
{{{
@@ -304,6 +348,10 @@ $code.=<<___;
.type bn_mul4x_mont,\@function,6
.align 16
bn_mul4x_mont:
+.cfi_startproc
+ mov ${num}d,${num}d
+ mov %rsp,%rax
+.cfi_def_cfa_register %rax
.Lmul4x_enter:
___
$code.=<<___ if ($addx);
@@ -313,20 +361,41 @@ $code.=<<___ if ($addx);
___
$code.=<<___;
push %rbx
+.cfi_push %rbx
push %rbp
+.cfi_push %rbp
push %r12
+.cfi_push %r12
push %r13
+.cfi_push %r13
push %r14
+.cfi_push %r14
push %r15
+.cfi_push %r15
- mov ${num}d,${num}d
- lea 4($num),%r10
+ neg $num
mov %rsp,%r11
- neg %r10
- lea (%rsp,%r10,8),%rsp # tp=alloca(8*(num+4))
- and \$-1024,%rsp # minimize TLB usage
+ lea -32(%rsp,$num,8),%r10 # future alloca(8*(num+4))
+ neg $num # restore
+ and \$-1024,%r10 # minimize TLB usage
- mov %r11,8(%rsp,$num,8) # tp[num+1]=%rsp
+ sub %r10,%r11
+ and \$-4096,%r11
+ lea (%r10,%r11),%rsp
+ mov (%rsp),%r11
+ cmp %r10,%rsp
+ ja .Lmul4x_page_walk
+ jmp .Lmul4x_page_walk_done
+
+.Lmul4x_page_walk:
+ lea -4096(%rsp),%rsp
+ mov (%rsp),%r11
+ cmp %r10,%rsp
+ ja .Lmul4x_page_walk
+.Lmul4x_page_walk_done:
+
+ mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp
+.cfi_cfa_expression %rsp+8,$num,8,mul,plus,deref,+8
.Lmul4x_body:
mov $rp,16(%rsp,$num,8) # tp[num+2]=$rp
mov %rdx,%r12 # reassign $bp
@@ -633,9 +702,11 @@ ___
my @ri=("%rax","%rdx",$m0,$m1);
$code.=<<___;
mov 16(%rsp,$num,8),$rp # restore $rp
+ lea -4($num),$j
mov 0(%rsp),@ri[0] # tp[0]
+ pxor %xmm0,%xmm0
mov 8(%rsp),@ri[1] # tp[1]
- shr \$2,$num # num/=4
+ shr \$2,$j # j=num/4-1
lea (%rsp),$ap # borrow ap for tp
xor $i,$i # i=0 and clear CF!
@@ -643,7 +714,6 @@ $code.=<<___;
mov 16($ap),@ri[2] # tp[2]
mov 24($ap),@ri[3] # tp[3]
sbb 8($np),@ri[1]
- lea -1($num),$j # j=num/4-1
jmp .Lsub4x
.align 16
.Lsub4x:
@@ -671,50 +741,58 @@ $code.=<<___;
mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
sbb \$0,@ri[0] # handle upmost overflow bit
- mov @ri[0],%xmm0
- punpcklqdq %xmm0,%xmm0 # extend mask to 128 bits
mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
xor $i,$i # i=0
-
- mov $num,$j
- pxor %xmm5,%xmm5
+ and @ri[0],$ap
+ not @ri[0]
+ mov $rp,$np
+ and @ri[0],$np
+ lea -4($num),$j
+ or $np,$ap # ap=borrow?tp:rp
+ shr \$2,$j # j=num/4-1
+
+ movdqu ($ap),%xmm1
+ movdqa %xmm0,(%rsp)
+ movdqu %xmm1,($rp)
jmp .Lcopy4x
.align 16
-.Lcopy4x: # copy or in-place refresh
- movdqu (%rsp,$i),%xmm2
- movdqu 16(%rsp,$i),%xmm4
- movdqu ($rp,$i),%xmm1
- movdqu 16($rp,$i),%xmm3
- pxor %xmm1,%xmm2 # conditional select
- pxor %xmm3,%xmm4
- pand %xmm0,%xmm2
- pand %xmm0,%xmm4
- pxor %xmm1,%xmm2
- pxor %xmm3,%xmm4
- movdqu %xmm2,($rp,$i)
- movdqu %xmm4,16($rp,$i)
- movdqa %xmm5,(%rsp,$i) # zap temporary vectors
- movdqa %xmm5,16(%rsp,$i)
-
+.Lcopy4x: # copy or in-place refresh
+ movdqu 16($ap,$i),%xmm2
+ movdqu 32($ap,$i),%xmm1
+ movdqa %xmm0,16(%rsp,$i)
+ movdqu %xmm2,16($rp,$i)
+ movdqa %xmm0,32(%rsp,$i)
+ movdqu %xmm1,32($rp,$i)
lea 32($i),$i
dec $j
jnz .Lcopy4x
- shl \$2,$num
+ movdqu 16($ap,$i),%xmm2
+ movdqa %xmm0,16(%rsp,$i)
+ movdqu %xmm2,16($rp,$i)
___
}
$code.=<<___;
mov 8(%rsp,$num,8),%rsi # restore %rsp
+.cfi_def_cfa %rsi, 8
mov \$1,%rax
- mov (%rsi),%r15
- mov 8(%rsi),%r14
- mov 16(%rsi),%r13
- mov 24(%rsi),%r12
- mov 32(%rsi),%rbp
- mov 40(%rsi),%rbx
- lea 48(%rsi),%rsp
+ mov -48(%rsi),%r15
+.cfi_restore %r15
+ mov -40(%rsi),%r14
+.cfi_restore %r14
+ mov -32(%rsi),%r13
+.cfi_restore %r13
+ mov -24(%rsi),%r12
+.cfi_restore %r12
+ mov -16(%rsi),%rbp
+.cfi_restore %rbp
+ mov -8(%rsi),%rbx
+.cfi_restore %rbx
+ lea (%rsi),%rsp
+.cfi_def_cfa_register %rsp
.Lmul4x_epilogue:
ret
+.cfi_endproc
.size bn_mul4x_mont,.-bn_mul4x_mont
___
}}}
@@ -742,14 +820,23 @@ $code.=<<___;
.type bn_sqr8x_mont,\@function,6
.align 32
bn_sqr8x_mont:
-.Lsqr8x_enter:
+.cfi_startproc
mov %rsp,%rax
+.cfi_def_cfa_register %rax
+.Lsqr8x_enter:
push %rbx
+.cfi_push %rbx
push %rbp
+.cfi_push %rbp
push %r12
+.cfi_push %r12
push %r13
+.cfi_push %r13
push %r14
+.cfi_push %r14
push %r15
+.cfi_push %r15
+.Lsqr8x_prologue:
mov ${num}d,%r10d
shl \$3,${num}d # convert $num to bytes
@@ -762,30 +849,49 @@ bn_sqr8x_mont:
# do its job.
#
lea -64(%rsp,$num,2),%r11
+ mov %rsp,%rbp
mov ($n0),$n0 # *n0
sub $aptr,%r11
and \$4095,%r11
cmp %r11,%r10
jb .Lsqr8x_sp_alt
- sub %r11,%rsp # align with $aptr
- lea -64(%rsp,$num,2),%rsp # alloca(frame+2*$num)
+ sub %r11,%rbp # align with $aptr
+ lea -64(%rbp,$num,2),%rbp # future alloca(frame+2*$num)
jmp .Lsqr8x_sp_done
.align 32
.Lsqr8x_sp_alt:
lea 4096-64(,$num,2),%r10 # 4096-frame-2*$num
- lea -64(%rsp,$num,2),%rsp # alloca(frame+2*$num)
+ lea -64(%rbp,$num,2),%rbp # future alloca(frame+2*$num)
sub %r10,%r11
mov \$0,%r10
cmovc %r10,%r11
- sub %r11,%rsp
+ sub %r11,%rbp
.Lsqr8x_sp_done:
- and \$-64,%rsp
+ and \$-64,%rbp
+ mov %rsp,%r11
+ sub %rbp,%r11
+ and \$-4096,%r11
+ lea (%rbp,%r11),%rsp
+ mov (%rsp),%r10
+ cmp %rbp,%rsp
+ ja .Lsqr8x_page_walk
+ jmp .Lsqr8x_page_walk_done
+
+.align 16
+.Lsqr8x_page_walk:
+ lea -4096(%rsp),%rsp
+ mov (%rsp),%r10
+ cmp %rbp,%rsp
+ ja .Lsqr8x_page_walk
+.Lsqr8x_page_walk_done:
+
mov $num,%r10
neg $num
mov $n0, 32(%rsp)
mov %rax, 40(%rsp) # save original %rsp
+.cfi_cfa_expression %rsp+40,deref,+8
.Lsqr8x_body:
movq $nptr, %xmm2 # save pointer to modulus
@@ -855,6 +961,7 @@ $code.=<<___;
pxor %xmm0,%xmm0
pshufd \$0,%xmm1,%xmm1
mov 40(%rsp),%rsi # restore %rsp
+.cfi_def_cfa %rsi,8
jmp .Lsqr8x_cond_copy
.align 32
@@ -884,14 +991,22 @@ $code.=<<___;
mov \$1,%rax
mov -48(%rsi),%r15
+.cfi_restore %r15
mov -40(%rsi),%r14
+.cfi_restore %r14
mov -32(%rsi),%r13
+.cfi_restore %r13
mov -24(%rsi),%r12
+.cfi_restore %r12
mov -16(%rsi),%rbp
+.cfi_restore %rbp
mov -8(%rsi),%rbx
+.cfi_restore %rbx
lea (%rsi),%rsp
+.cfi_def_cfa_register %rsp
.Lsqr8x_epilogue:
ret
+.cfi_endproc
.size bn_sqr8x_mont,.-bn_sqr8x_mont
___
}}}
@@ -903,23 +1018,48 @@ $code.=<<___;
.type bn_mulx4x_mont,\@function,6
.align 32
bn_mulx4x_mont:
-.Lmulx4x_enter:
+.cfi_startproc
mov %rsp,%rax
+.cfi_def_cfa_register %rax
+.Lmulx4x_enter:
push %rbx
+.cfi_push %rbx
push %rbp
+.cfi_push %rbp
push %r12
+.cfi_push %r12
push %r13
+.cfi_push %r13
push %r14
+.cfi_push %r14
push %r15
+.cfi_push %r15
+.Lmulx4x_prologue:
shl \$3,${num}d # convert $num to bytes
- .byte 0x67
xor %r10,%r10
sub $num,%r10 # -$num
mov ($n0),$n0 # *n0
- lea -72(%rsp,%r10),%rsp # alloca(frame+$num+8)
+ lea -72(%rsp,%r10),%rbp # future alloca(frame+$num+8)
+ and \$-128,%rbp
+ mov %rsp,%r11
+ sub %rbp,%r11
+ and \$-4096,%r11
+ lea (%rbp,%r11),%rsp
+ mov (%rsp),%r10
+ cmp %rbp,%rsp
+ ja .Lmulx4x_page_walk
+ jmp .Lmulx4x_page_walk_done
+
+.align 16
+.Lmulx4x_page_walk:
+ lea -4096(%rsp),%rsp
+ mov (%rsp),%r10
+ cmp %rbp,%rsp
+ ja .Lmulx4x_page_walk
+.Lmulx4x_page_walk_done:
+
lea ($bp,$num),%r10
- and \$-128,%rsp
##############################################################
# Stack layout
# +0 num
@@ -939,6 +1079,7 @@ bn_mulx4x_mont:
mov $n0, 24(%rsp) # save *n0
mov $rp, 32(%rsp) # save $rp
mov %rax,40(%rsp) # save original %rsp
+.cfi_cfa_expression %rsp+40,deref,+8
mov $num,48(%rsp) # inner counter
jmp .Lmulx4x_body
@@ -1188,6 +1329,7 @@ $code.=<<___;
pxor %xmm0,%xmm0
pshufd \$0,%xmm1,%xmm1
mov 40(%rsp),%rsi # restore %rsp
+.cfi_def_cfa %rsi,8
jmp .Lmulx4x_cond_copy
.align 32
@@ -1217,14 +1359,22 @@ $code.=<<___;
mov \$1,%rax
mov -48(%rsi),%r15
+.cfi_restore %r15
mov -40(%rsi),%r14
+.cfi_restore %r14
mov -32(%rsi),%r13
+.cfi_restore %r13
mov -24(%rsi),%r12
+.cfi_restore %r12
mov -16(%rsi),%rbp
+.cfi_restore %rbp
mov -8(%rsi),%rbx
+.cfi_restore %rbx
lea (%rsi),%rsp
+.cfi_def_cfa_register %rsp
.Lmulx4x_epilogue:
ret
+.cfi_endproc
.size bn_mulx4x_mont,.-bn_mulx4x_mont
___
}}}
@@ -1277,22 +1427,8 @@ mul_handler:
mov 192($context),%r10 # pull $num
mov 8(%rax,%r10,8),%rax # pull saved stack pointer
- lea 48(%rax),%rax
-
- mov -8(%rax),%rbx
- mov -16(%rax),%rbp
- mov -24(%rax),%r12
- mov -32(%rax),%r13
- mov -40(%rax),%r14
- mov -48(%rax),%r15
- mov %rbx,144($context) # restore context->Rbx
- mov %rbp,160($context) # restore context->Rbp
- mov %r12,216($context) # restore context->R12
- mov %r13,224($context) # restore context->R13
- mov %r14,232($context) # restore context->R14
- mov %r15,240($context) # restore context->R15
- jmp .Lcommon_seh_tail
+ jmp .Lcommon_pop_regs
.size mul_handler,.-mul_handler
.type sqr_handler,\@abi-omnipotent
@@ -1317,18 +1453,24 @@ sqr_handler:
mov 0(%r11),%r10d # HandlerData[0]
lea (%rsi,%r10),%r10 # end of prologue label
- cmp %r10,%rbx # context->Rip<.Lsqr_body
+ cmp %r10,%rbx # context->Rip<.Lsqr_prologue
jb .Lcommon_seh_tail
+ mov 4(%r11),%r10d # HandlerData[1]
+ lea (%rsi,%r10),%r10 # body label
+ cmp %r10,%rbx # context->Rip<.Lsqr_body
+ jb .Lcommon_pop_regs
+
mov 152($context),%rax # pull context->Rsp
- mov 4(%r11),%r10d # HandlerData[1]
+ mov 8(%r11),%r10d # HandlerData[2]
lea (%rsi,%r10),%r10 # epilogue label
cmp %r10,%rbx # context->Rip>=.Lsqr_epilogue
jae .Lcommon_seh_tail
mov 40(%rax),%rax # pull saved stack pointer
+.Lcommon_pop_regs:
mov -8(%rax),%rbx
mov -16(%rax),%rbp
mov -24(%rax),%r12
@@ -1415,13 +1557,15 @@ $code.=<<___;
.LSEH_info_bn_sqr8x_mont:
.byte 9,0,0,0
.rva sqr_handler
- .rva .Lsqr8x_body,.Lsqr8x_epilogue # HandlerData[]
+ .rva .Lsqr8x_prologue,.Lsqr8x_body,.Lsqr8x_epilogue # HandlerData[]
+.align 8
___
$code.=<<___ if ($addx);
.LSEH_info_bn_mulx4x_mont:
.byte 9,0,0,0
.rva sqr_handler
- .rva .Lmulx4x_body,.Lmulx4x_epilogue # HandlerData[]
+ .rva .Lmulx4x_prologue,.Lmulx4x_body,.Lmulx4x_epilogue # HandlerData[]
+.align 8
___
}
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-20 16:26:10 +0000