s390x assembler pack update.

[oweals/openssl.git] / crypto / aes / asm / aes-sparcv9.pl

diff --git a/crypto/aes/asm/aes-sparcv9.pl b/crypto/aes/asm/aes-sparcv9.pl
index 30f38d7a5a2ec060b97513cf16939c3ca146434b..095711ca710dae03d423063e7a62c8de086bf517 100755 (executable)
--- a/crypto/aes/asm/aes-sparcv9.pl
+++ b/crypto/aes/asm/aes-sparcv9.pl
@@ -6,7 +6,7 @@
 # forms are granted according to the OpenSSL license.
 # ====================================================================
 #
-# Version 1.0
+# Version 1.1
 #
 # The major reason for undertaken effort was to mitigate the hazard of
 # cache-timing attack. This is [currently and initially!] addressed in
@@ -16,6 +16,9 @@
 # is an initial draft and one should expect more countermeasures to
 # be implemented...
 #
+# Version 1.1 prefetches T[ed]4 in order to mitigate attack on last
+# round.
+#
 # Even though performance was not the primary goal [on the contrary,
 # extra shifts "induced" by compressed S-box and longer loop epilogue
 # "induced" by scheduling for L2 have negative effect on performance],
@@ -78,7 +81,7 @@ ___
 $code.=<<___;
 .section       ".text",#alloc,#execinstr
 
-.align 64
+.align 256
 AES_Te:
 ___
 &_data_word(
@@ -364,20 +367,28 @@ _sparcv9_AES_encrypt:
        ld      [$key+28],$t3                   !
                srlx    $acc9,8,$acc9
                xor     $acc5,$s1,$s1
+       ldx     [$tbl+2048+0],%g0               ! prefetch te4
                srlx    $acc10,16,$acc10
                xor     $acc6,$s1,$s1
+       ldx     [$tbl+2048+32],%g0              ! prefetch te4
                srlx    $acc11,24,$acc11
                xor     $acc7,$s1,$s1
+       ldx     [$tbl+2048+64],%g0              ! prefetch te4
                srlx    $acc13,8,$acc13
                xor     $acc8,$s2,$s2
+       ldx     [$tbl+2048+96],%g0              ! prefetch te4
                srlx    $acc14,16,$acc14        !
                xor     $acc9,$s2,$s2
+       ldx     [$tbl+2048+128],%g0             ! prefetch te4
                srlx    $acc15,24,$acc15
                xor     $acc10,$s2,$s2
+       ldx     [$tbl+2048+160],%g0             ! prefetch te4
        srl     $s0,21,$acc0
                xor     $acc11,$s2,$s2
+       ldx     [$tbl+2048+192],%g0             ! prefetch te4
                xor     $acc12,$acc14,$acc14
                xor     $acc13,$s3,$s3
+       ldx     [$tbl+2048+224],%g0             ! prefetch te4
        srl     $s1,13,$acc1                    !
                xor     $acc14,$s3,$s3
                xor     $acc15,$s3,$s3
@@ -511,11 +522,10 @@ AES_encrypt:
        ld      [%i0+8],%o2
        ld      [%i0+12],%o3
 
-1:     call    .+8
        mov     %i2,%o5
-       sub     %o7,1b-AES_Te,%o4
-       call    _sparcv9_AES_encrypt
        nop
+1:     call    _sparcv9_AES_encrypt
+       sub     %o7,1b-AES_Te,%o4
 
        st      %o0,[%i1+0]
        st      %o1,[%i1+4]
@@ -572,11 +582,10 @@ AES_encrypt:
        or      %l7,%l6,%l6
        or      %l4,%l6,%o3
 
-1:     call    .+8
        mov     %i2,%o5
-       sub     %o7,1b-AES_Te,%o4
-       call    _sparcv9_AES_encrypt
        nop
+1:     call    _sparcv9_AES_encrypt
+       sub     %o7,1b-AES_Te,%o4
 
        srl     %o0,24,%l0
        srl     %o0,16,%l1
@@ -618,7 +627,7 @@ AES_encrypt:
 ___
 
 $code.=<<___;
-.align 64
+.align 256
 AES_Td:
 ___
 &_data_word(
@@ -904,20 +913,28 @@ _sparcv9_AES_decrypt:
        ld      [$key+28],$t3                   !
                srlx    $acc9,8,$acc9
                xor     $acc5,$s1,$s1
+       ldx     [$tbl+2048+0],%g0               ! prefetch td4
                srlx    $acc10,16,$acc10
                xor     $acc6,$s1,$s1
+       ldx     [$tbl+2048+32],%g0              ! prefetch td4
                srlx    $acc11,24,$acc11
                xor     $acc7,$s1,$s1
+       ldx     [$tbl+2048+64],%g0              ! prefetch td4
                srlx    $acc13,8,$acc13
                xor     $acc8,$s2,$s2
+       ldx     [$tbl+2048+96],%g0              ! prefetch td4
                srlx    $acc14,16,$acc14        !
                xor     $acc9,$s2,$s2
+       ldx     [$tbl+2048+128],%g0             ! prefetch td4
                srlx    $acc15,24,$acc15
                xor     $acc10,$s2,$s2
+       ldx     [$tbl+2048+160],%g0             ! prefetch td4
        srl     $s0,21,$acc0
                xor     $acc11,$s2,$s2
+       ldx     [$tbl+2048+192],%g0             ! prefetch td4
                xor     $acc12,$acc14,$acc14
                xor     $acc13,$s3,$s3
+       ldx     [$tbl+2048+224],%g0             ! prefetch td4
        and     $acc0,2040,$acc0                !
                xor     $acc14,$s3,$s3
                xor     $acc15,$s3,$s3
@@ -1051,11 +1068,10 @@ AES_decrypt:
        ld      [%i0+8],%o2
        ld      [%i0+12],%o3
 
-1:     call    .+8
        mov     %i2,%o5
-       sub     %o7,1b-AES_Td,%o4
-       call    _sparcv9_AES_decrypt
        nop
+1:     call    _sparcv9_AES_decrypt
+       sub     %o7,1b-AES_Td,%o4
 
        st      %o0,[%i1+0]
        st      %o1,[%i1+4]
@@ -1112,11 +1128,10 @@ AES_decrypt:
        or      %l7,%l6,%l6
        or      %l4,%l6,%o3
 
-1:     call    .+8
        mov     %i2,%o5
-       sub     %o7,1b-AES_Td,%o4
-       call    _sparcv9_AES_decrypt
        nop
+1:     call    _sparcv9_AES_decrypt
+       sub     %o7,1b-AES_Td,%o4
 
        srl     %o0,24,%l0
        srl     %o0,16,%l1