bn/asm/rsaz-avx2.pl: constant-time gather procedure.
authorAndy Polyakov <appro@openssl.org>
Mon, 25 Jan 2016 22:25:40 +0000 (23:25 +0100)
committerMatt Caswell <matt@openssl.org>
Tue, 1 Mar 2016 13:34:22 +0000 (13:34 +0000)
Performance penalty is 2%.

CVE-2016-0702

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
crypto/bn/asm/rsaz-avx2.pl
crypto/bn/asm/rsaz-x86_64.pl

index 3b6ccf8..712a77f 100755 (executable)
@@ -443,7 +443,7 @@ $TEMP2 = $B2;
 $TEMP3 = $Y1;
 $TEMP4 = $Y2;
 $code.=<<___;
-       #we need to fix indexes 32-39 to avoid overflow
+       # we need to fix indices 32-39 to avoid overflow
        vmovdqu         32*8(%rsp), $ACC8               # 32*8-192($tp0),
        vmovdqu         32*9(%rsp), $ACC1               # 32*9-192($tp0)
        vmovdqu         32*10(%rsp), $ACC2              # 32*10-192($tp0)
@@ -1592,68 +1592,128 @@ rsaz_1024_scatter5_avx2:
 .type  rsaz_1024_gather5_avx2,\@abi-omnipotent
 .align 32
 rsaz_1024_gather5_avx2:
+       vzeroupper
+       mov     %rsp,%r11
 ___
 $code.=<<___ if ($win64);
        lea     -0x88(%rsp),%rax
-       vzeroupper
 .LSEH_begin_rsaz_1024_gather5:
        # I can't trust assembler to use specific encoding:-(
-       .byte   0x48,0x8d,0x60,0xe0             #lea    -0x20(%rax),%rsp
-       .byte   0xc5,0xf8,0x29,0x70,0xe0        #vmovaps %xmm6,-0x20(%rax)
-       .byte   0xc5,0xf8,0x29,0x78,0xf0        #vmovaps %xmm7,-0x10(%rax)
-       .byte   0xc5,0x78,0x29,0x40,0x00        #vmovaps %xmm8,0(%rax)
-       .byte   0xc5,0x78,0x29,0x48,0x10        #vmovaps %xmm9,0x10(%rax)
-       .byte   0xc5,0x78,0x29,0x50,0x20        #vmovaps %xmm10,0x20(%rax)
-       .byte   0xc5,0x78,0x29,0x58,0x30        #vmovaps %xmm11,0x30(%rax)
-       .byte   0xc5,0x78,0x29,0x60,0x40        #vmovaps %xmm12,0x40(%rax)
-       .byte   0xc5,0x78,0x29,0x68,0x50        #vmovaps %xmm13,0x50(%rax)
-       .byte   0xc5,0x78,0x29,0x70,0x60        #vmovaps %xmm14,0x60(%rax)
-       .byte   0xc5,0x78,0x29,0x78,0x70        #vmovaps %xmm15,0x70(%rax)
+       .byte   0x48,0x8d,0x60,0xe0             # lea   -0x20(%rax),%rsp
+       .byte   0xc5,0xf8,0x29,0x70,0xe0        # vmovaps %xmm6,-0x20(%rax)
+       .byte   0xc5,0xf8,0x29,0x78,0xf0        # vmovaps %xmm7,-0x10(%rax)
+       .byte   0xc5,0x78,0x29,0x40,0x00        # vmovaps %xmm8,0(%rax)
+       .byte   0xc5,0x78,0x29,0x48,0x10        # vmovaps %xmm9,0x10(%rax)
+       .byte   0xc5,0x78,0x29,0x50,0x20        # vmovaps %xmm10,0x20(%rax)
+       .byte   0xc5,0x78,0x29,0x58,0x30        # vmovaps %xmm11,0x30(%rax)
+       .byte   0xc5,0x78,0x29,0x60,0x40        # vmovaps %xmm12,0x40(%rax)
+       .byte   0xc5,0x78,0x29,0x68,0x50        # vmovaps %xmm13,0x50(%rax)
+       .byte   0xc5,0x78,0x29,0x70,0x60        # vmovaps %xmm14,0x60(%rax)
+       .byte   0xc5,0x78,0x29,0x78,0x70        # vmovaps %xmm15,0x70(%rax)
 ___
 $code.=<<___;
-       lea     .Lgather_table(%rip),%r11
-       mov     $power,%eax
-       and     \$3,$power
-       shr     \$2,%eax                        # cache line number
-       shl     \$4,$power                      # offset within cache line
-
-       vmovdqu         -32(%r11),%ymm7         # .Lgather_permd
-       vpbroadcastb    8(%r11,%rax), %xmm8
-       vpbroadcastb    7(%r11,%rax), %xmm9
-       vpbroadcastb    6(%r11,%rax), %xmm10
-       vpbroadcastb    5(%r11,%rax), %xmm11
-       vpbroadcastb    4(%r11,%rax), %xmm12
-       vpbroadcastb    3(%r11,%rax), %xmm13
-       vpbroadcastb    2(%r11,%rax), %xmm14
-       vpbroadcastb    1(%r11,%rax), %xmm15
-
-       lea     64($inp,$power),$inp
-       mov     \$64,%r11                       # size optimization
-       mov     \$9,%eax
-       jmp     .Loop_gather_1024
+       lea     -0x100(%rsp),%rsp
+       and     \$-32, %rsp
+       lea     .Linc(%rip), %r10
+       lea     -128(%rsp),%rax                 # control u-op density
+
+       vmovd           $power, %xmm4
+       vmovdqa         (%r10),%ymm0
+       vmovdqa         32(%r10),%ymm1
+       vmovdqa         64(%r10),%ymm5
+       vpbroadcastd    %xmm4,%ymm4
+
+       vpaddd          %ymm5, %ymm0, %ymm2
+       vpcmpeqd        %ymm4, %ymm0, %ymm0
+       vpaddd          %ymm5, %ymm1, %ymm3
+       vpcmpeqd        %ymm4, %ymm1, %ymm1
+       vmovdqa         %ymm0, 32*0+128(%rax)
+       vpaddd          %ymm5, %ymm2, %ymm0
+       vpcmpeqd        %ymm4, %ymm2, %ymm2
+       vmovdqa         %ymm1, 32*1+128(%rax)
+       vpaddd          %ymm5, %ymm3, %ymm1
+       vpcmpeqd        %ymm4, %ymm3, %ymm3
+       vmovdqa         %ymm2, 32*2+128(%rax)
+       vpaddd          %ymm5, %ymm0, %ymm2
+       vpcmpeqd        %ymm4, %ymm0, %ymm0
+       vmovdqa         %ymm3, 32*3+128(%rax)
+       vpaddd          %ymm5, %ymm1, %ymm3
+       vpcmpeqd        %ymm4, %ymm1, %ymm1
+       vmovdqa         %ymm0, 32*4+128(%rax)
+       vpaddd          %ymm5, %ymm2, %ymm8
+       vpcmpeqd        %ymm4, %ymm2, %ymm2
+       vmovdqa         %ymm1, 32*5+128(%rax)
+       vpaddd          %ymm5, %ymm3, %ymm9
+       vpcmpeqd        %ymm4, %ymm3, %ymm3
+       vmovdqa         %ymm2, 32*6+128(%rax)
+       vpaddd          %ymm5, %ymm8, %ymm10
+       vpcmpeqd        %ymm4, %ymm8, %ymm8
+       vmovdqa         %ymm3, 32*7+128(%rax)
+       vpaddd          %ymm5, %ymm9, %ymm11
+       vpcmpeqd        %ymm4, %ymm9, %ymm9
+       vpaddd          %ymm5, %ymm10, %ymm12
+       vpcmpeqd        %ymm4, %ymm10, %ymm10
+       vpaddd          %ymm5, %ymm11, %ymm13
+       vpcmpeqd        %ymm4, %ymm11, %ymm11
+       vpaddd          %ymm5, %ymm12, %ymm14
+       vpcmpeqd        %ymm4, %ymm12, %ymm12
+       vpaddd          %ymm5, %ymm13, %ymm15
+       vpcmpeqd        %ymm4, %ymm13, %ymm13
+       vpcmpeqd        %ymm4, %ymm14, %ymm14
+       vpcmpeqd        %ymm4, %ymm15, %ymm15
+
+       vmovdqa -32(%r10),%ymm7                 # .Lgather_permd
+       lea     128($inp), $inp
+       mov     \$9,$power
 
-.align 32
 .Loop_gather_1024:
-       vpand           -64($inp),              %xmm8,%xmm0
-       vpand           ($inp),                 %xmm9,%xmm1
-       vpand           64($inp),               %xmm10,%xmm2
-       vpand           ($inp,%r11,2),          %xmm11,%xmm3
-        vpor                                   %xmm0,%xmm1,%xmm1
-       vpand           64($inp,%r11,2),        %xmm12,%xmm4
-        vpor                                   %xmm2,%xmm3,%xmm3
-       vpand           ($inp,%r11,4),          %xmm13,%xmm5
-        vpor                                   %xmm1,%xmm3,%xmm3
-       vpand           64($inp,%r11,4),        %xmm14,%xmm6
-        vpor                                   %xmm4,%xmm5,%xmm5
-       vpand           -128($inp,%r11,8),      %xmm15,%xmm2
-       lea             ($inp,%r11,8),$inp
-        vpor                                   %xmm3,%xmm5,%xmm5
-        vpor                                   %xmm2,%xmm6,%xmm6
-        vpor                                   %xmm5,%xmm6,%xmm6
-       vpermd          %ymm6,%ymm7,%ymm6
-       vmovdqu         %ymm6,($out)
+       vmovdqa         32*0-128($inp), %ymm0
+       vmovdqa         32*1-128($inp), %ymm1
+       vmovdqa         32*2-128($inp), %ymm2
+       vmovdqa         32*3-128($inp), %ymm3
+       vpand           32*0+128(%rax), %ymm0,  %ymm0
+       vpand           32*1+128(%rax), %ymm1,  %ymm1
+       vpand           32*2+128(%rax), %ymm2,  %ymm2
+       vpor            %ymm0, %ymm1, %ymm4
+       vpand           32*3+128(%rax), %ymm3,  %ymm3
+       vmovdqa         32*4-128($inp), %ymm0
+       vmovdqa         32*5-128($inp), %ymm1
+       vpor            %ymm2, %ymm3, %ymm5
+       vmovdqa         32*6-128($inp), %ymm2
+       vmovdqa         32*7-128($inp), %ymm3
+       vpand           32*4+128(%rax), %ymm0,  %ymm0
+       vpand           32*5+128(%rax), %ymm1,  %ymm1
+       vpand           32*6+128(%rax), %ymm2,  %ymm2
+       vpor            %ymm0, %ymm4, %ymm4
+       vpand           32*7+128(%rax), %ymm3,  %ymm3
+       vpand           32*8-128($inp), %ymm8,  %ymm0
+       vpor            %ymm1, %ymm5, %ymm5
+       vpand           32*9-128($inp), %ymm9,  %ymm1
+       vpor            %ymm2, %ymm4, %ymm4
+       vpand           32*10-128($inp),%ymm10, %ymm2
+       vpor            %ymm3, %ymm5, %ymm5
+       vpand           32*11-128($inp),%ymm11, %ymm3
+       vpor            %ymm0, %ymm4, %ymm4
+       vpand           32*12-128($inp),%ymm12, %ymm0
+       vpor            %ymm1, %ymm5, %ymm5
+       vpand           32*13-128($inp),%ymm13, %ymm1
+       vpor            %ymm2, %ymm4, %ymm4
+       vpand           32*14-128($inp),%ymm14, %ymm2
+       vpor            %ymm3, %ymm5, %ymm5
+       vpand           32*15-128($inp),%ymm15, %ymm3
+       lea             32*16($inp), $inp
+       vpor            %ymm0, %ymm4, %ymm4
+       vpor            %ymm1, %ymm5, %ymm5
+       vpor            %ymm2, %ymm4, %ymm4
+       vpor            %ymm3, %ymm5, %ymm5
+
+       vpor            %ymm5, %ymm4, %ymm4
+       vextracti128    \$1, %ymm4, %xmm5       # upper half is cleared
+       vpor            %xmm4, %xmm5, %xmm5
+       vpermd          %ymm5,%ymm7,%ymm5
+       vmovdqu         %ymm5,($out)
        lea             32($out),$out
-       dec     %eax
+       dec     $power
        jnz     .Loop_gather_1024
 
        vpxor   %ymm0,%ymm0,%ymm0
@@ -1661,20 +1721,20 @@ $code.=<<___;
        vzeroupper
 ___
 $code.=<<___ if ($win64);
-       movaps  (%rsp),%xmm6
-       movaps  0x10(%rsp),%xmm7
-       movaps  0x20(%rsp),%xmm8
-       movaps  0x30(%rsp),%xmm9
-       movaps  0x40(%rsp),%xmm10
-       movaps  0x50(%rsp),%xmm11
-       movaps  0x60(%rsp),%xmm12
-       movaps  0x70(%rsp),%xmm13
-       movaps  0x80(%rsp),%xmm14
-       movaps  0x90(%rsp),%xmm15
-       lea     0xa8(%rsp),%rsp
+       movaps  -0xa8(%r11),%xmm6
+       movaps  -0x98(%r11),%xmm7
+       movaps  -0x88(%r11),%xmm8
+       movaps  -0x78(%r11),%xmm9
+       movaps  -0x68(%r11),%xmm10
+       movaps  -0x58(%r11),%xmm11
+       movaps  -0x48(%r11),%xmm12
+       movaps  -0x38(%r11),%xmm13
+       movaps  -0x28(%r11),%xmm14
+       movaps  -0x18(%r11),%xmm15
 .LSEH_end_rsaz_1024_gather5:
 ___
 $code.=<<___;
+       lea     (%r11),%rsp
        ret
 .size  rsaz_1024_gather5_avx2,.-rsaz_1024_gather5_avx2
 ___
@@ -1708,8 +1768,10 @@ $code.=<<___;
        .long   0,2,4,6,7,7,7,7
 .Lgather_permd:
        .long   0,7,1,7,2,7,3,7
-.Lgather_table:
-       .byte   0,0,0,0,0,0,0,0, 0xff,0,0,0,0,0,0,0
+.Linc:
+       .long   0,0,0,0, 1,1,1,1
+       .long   2,2,2,2, 3,3,3,3
+       .long   4,4,4,4, 4,4,4,4
 .align 64
 ___
 
@@ -1837,18 +1899,19 @@ rsaz_se_handler:
        .rva    rsaz_se_handler
        .rva    .Lmul_1024_body,.Lmul_1024_epilogue
 .LSEH_info_rsaz_1024_gather5:
-       .byte   0x01,0x33,0x16,0x00
-       .byte   0x36,0xf8,0x09,0x00     #vmovaps 0x90(rsp),xmm15
-       .byte   0x31,0xe8,0x08,0x00     #vmovaps 0x80(rsp),xmm14
-       .byte   0x2c,0xd8,0x07,0x00     #vmovaps 0x70(rsp),xmm13
-       .byte   0x27,0xc8,0x06,0x00     #vmovaps 0x60(rsp),xmm12
-       .byte   0x22,0xb8,0x05,0x00     #vmovaps 0x50(rsp),xmm11
-       .byte   0x1d,0xa8,0x04,0x00     #vmovaps 0x40(rsp),xmm10
-       .byte   0x18,0x98,0x03,0x00     #vmovaps 0x30(rsp),xmm9
-       .byte   0x13,0x88,0x02,0x00     #vmovaps 0x20(rsp),xmm8
-       .byte   0x0e,0x78,0x01,0x00     #vmovaps 0x10(rsp),xmm7
-       .byte   0x09,0x68,0x00,0x00     #vmovaps 0x00(rsp),xmm6
-       .byte   0x04,0x01,0x15,0x00     #sub    rsp,0xa8
+       .byte   0x01,0x36,0x17,0x0b
+       .byte   0x36,0xf8,0x09,0x00     # vmovaps 0x90(rsp),xmm15
+       .byte   0x31,0xe8,0x08,0x00     # vmovaps 0x80(rsp),xmm14
+       .byte   0x2c,0xd8,0x07,0x00     # vmovaps 0x70(rsp),xmm13
+       .byte   0x27,0xc8,0x06,0x00     # vmovaps 0x60(rsp),xmm12
+       .byte   0x22,0xb8,0x05,0x00     # vmovaps 0x50(rsp),xmm11
+       .byte   0x1d,0xa8,0x04,0x00     # vmovaps 0x40(rsp),xmm10
+       .byte   0x18,0x98,0x03,0x00     # vmovaps 0x30(rsp),xmm9
+       .byte   0x13,0x88,0x02,0x00     # vmovaps 0x20(rsp),xmm8
+       .byte   0x0e,0x78,0x01,0x00     # vmovaps 0x10(rsp),xmm7
+       .byte   0x09,0x68,0x00,0x00     # vmovaps 0x00(rsp),xmm6
+       .byte   0x04,0x01,0x15,0x00     # sub     rsp,0xa8
+       .byte   0x00,0xb3,0x00,0x00     # set_frame r11
 ___
 }
 
index d1ea454..3290054 100755 (executable)
@@ -2332,17 +2332,17 @@ se_handler:
        .rva    .Lmul_by_one_body,.Lmul_by_one_epilogue         # HandlerData[]
 .LSEH_info_rsaz_512_gather4:
        .byte   0x01,0x46,0x16,0x00
-       .byte   0x46,0xf8,0x09,0x00     #vmovaps 0x90(rsp),xmm15
-       .byte   0x3d,0xe8,0x08,0x00     #vmovaps 0x80(rsp),xmm14
-       .byte   0x34,0xd8,0x07,0x00     #vmovaps 0x70(rsp),xmm13
-       .byte   0x2e,0xc8,0x06,0x00     #vmovaps 0x60(rsp),xmm12
-       .byte   0x28,0xb8,0x05,0x00     #vmovaps 0x50(rsp),xmm11
-       .byte   0x22,0xa8,0x04,0x00     #vmovaps 0x40(rsp),xmm10
-       .byte   0x1c,0x98,0x03,0x00     #vmovaps 0x30(rsp),xmm9
-       .byte   0x16,0x88,0x02,0x00     #vmovaps 0x20(rsp),xmm8
-       .byte   0x10,0x78,0x01,0x00     #vmovaps 0x10(rsp),xmm7
-       .byte   0x0b,0x68,0x00,0x00     #vmovaps 0x00(rsp),xmm6
-       .byte   0x07,0x01,0x15,0x00     #sub     rsp,0xa8
+       .byte   0x46,0xf8,0x09,0x00     # vmovaps 0x90(rsp),xmm15
+       .byte   0x3d,0xe8,0x08,0x00     # vmovaps 0x80(rsp),xmm14
+       .byte   0x34,0xd8,0x07,0x00     # vmovaps 0x70(rsp),xmm13
+       .byte   0x2e,0xc8,0x06,0x00     # vmovaps 0x60(rsp),xmm12
+       .byte   0x28,0xb8,0x05,0x00     # vmovaps 0x50(rsp),xmm11
+       .byte   0x22,0xa8,0x04,0x00     # vmovaps 0x40(rsp),xmm10
+       .byte   0x1c,0x98,0x03,0x00     # vmovaps 0x30(rsp),xmm9
+       .byte   0x16,0x88,0x02,0x00     # vmovaps 0x20(rsp),xmm8
+       .byte   0x10,0x78,0x01,0x00     # vmovaps 0x10(rsp),xmm7
+       .byte   0x0b,0x68,0x00,0x00     # vmovaps 0x00(rsp),xmm6
+       .byte   0x07,0x01,0x15,0x00     # sub     rsp,0xa8
 ___
 }