2013-03-27 00:59:10 +04:00
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
# Implement f a s t S H A - 2 5 6 w i t h A V X 2 i n s t r u c t i o n s . ( x86 _ 6 4 )
#
# Copyright ( C ) 2 0 1 3 I n t e l C o r p o r a t i o n .
#
# Authors :
# James G u i l f o r d < j a m e s . g u i l f o r d @intel.com>
# Kirk Y a p < k i r k . s . y a p @intel.com>
# Tim C h e n < t i m . c . c h e n @linux.intel.com>
#
# This s o f t w a r e i s a v a i l a b l e t o y o u u n d e r a c h o i c e o f o n e o f t w o
# licenses. Y o u m a y c h o o s e t o b e l i c e n s e d u n d e r t h e t e r m s o f t h e G N U
# General P u b l i c L i c e n s e ( G P L ) V e r s i o n 2 , a v a i l a b l e f r o m t h e f i l e
# COPYING i n t h e m a i n d i r e c t o r y o f t h i s s o u r c e t r e e , o r t h e
# OpenIB. o r g B S D l i c e n s e b e l o w :
#
# Redistribution a n d u s e i n s o u r c e a n d b i n a r y f o r m s , w i t h o r
# without m o d i f i c a t i o n , a r e p e r m i t t e d p r o v i d e d t h a t t h e f o l l o w i n g
# conditions a r e m e t :
#
# - Redistributions o f s o u r c e c o d e m u s t r e t a i n t h e a b o v e
# copyright n o t i c e , t h i s l i s t o f c o n d i t i o n s a n d t h e f o l l o w i n g
# disclaimer.
#
# - Redistributions i n b i n a r y f o r m m u s t r e p r o d u c e t h e a b o v e
# copyright n o t i c e , t h i s l i s t o f c o n d i t i o n s a n d t h e f o l l o w i n g
# disclaimer i n t h e d o c u m e n t a t i o n a n d / o r o t h e r m a t e r i a l s
# provided w i t h t h e d i s t r i b u t i o n .
#
# THE S O F T W A R E I S P R O V I D E D " A S I S " , W I T H O U T W A R R A N T Y O F A N Y K I N D ,
# EXPRESS O R I M P L I E D , I N C L U D I N G B U T N O T L I M I T E D T O T H E W A R R A N T I E S O F
# MERCHANTABILITY, F I T N E S S F O R A P A R T I C U L A R P U R P O S E A N D
# NONINFRINGEMENT. I N N O E V E N T S H A L L T H E A U T H O R S O R C O P Y R I G H T H O L D E R S
# BE L I A B L E F O R A N Y C L A I M , D A M A G E S O R O T H E R L I A B I L I T Y , W H E T H E R I N A N
# ACTION O F C O N T R A C T , T O R T O R O T H E R W I S E , A R I S I N G F R O M , O U T O F O R I N
# CONNECTION W I T H T H E S O F T W A R E O R T H E U S E O R O T H E R D E A L I N G S I N T H E
# SOFTWARE.
#
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
#
# This c o d e i s d e s c r i b e d i n a n I n t e l W h i t e - P a p e r :
# " Fast S H A - 2 5 6 I m p l e m e n t a t i o n s o n I n t e l A r c h i t e c t u r e P r o c e s s o r s "
#
# To f i n d i t , s u r f t o h t t p : / / w w w . i n t e l . c o m / p / e n _ U S / e m b e d d e d
# and s e a r c h f o r t h a t t i t l e .
#
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
# This c o d e s c h e d u l e s 2 b l o c k s a t a t i m e , w i t h 4 l a n e s p e r b l o c k
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
# include < l i n u x / l i n k a g e . h >
# # assume b u f f e r s n o t a l i g n e d
# define V M O V D Q v m o v d q u
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # Define M a c r o s
# addm [ m e m ] , r e g
# Add r e g t o m e m u s i n g r e g - m e m a d d a n d s t o r e
.macro addm p1 p2
add \ p1 , \ p2
mov \ p2 , \ p1
.endm
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
X0 = % y m m 4
X1 = % y m m 5
X2 = % y m m 6
X3 = % y m m 7
# XMM v e r s i o n s o f a b o v e
XWORD0 = % x m m 4
XWORD1 = % x m m 5
XWORD2 = % x m m 6
XWORD3 = % x m m 7
XTMP0 = % y m m 0
XTMP1 = % y m m 1
XTMP2 = % y m m 2
XTMP3 = % y m m 3
XTMP4 = % y m m 8
XFER = % y m m 9
XTMP5 = % y m m 1 1
SHUF_ 0 0 B A = % y m m 1 0 # s h u f f l e x B x A - > 00 B A
SHUF_ D C 0 0 = % y m m 1 2 # s h u f f l e x D x C - > D C 00
BYTE_ F L I P _ M A S K = % y m m 1 3
X_ B Y T E _ F L I P _ M A S K = % x m m 1 3 # X M M v e r s i o n o f B Y T E _ F L I P _ M A S K
NUM_ B L K S = % r d x # 3 r d a r g
2015-04-09 13:55:47 +03:00
INP = % r s i # 2 n d a r g
CTX = % r d i # 1 s t a r g
2013-03-27 00:59:10 +04:00
c = % e c x
d = % r8 d
e = % e d x # c l o b b e r s N U M _ B L K S
2015-04-09 13:55:47 +03:00
y3 = % e s i # c l o b b e r s I N P
2013-03-27 00:59:10 +04:00
SRND = C T X # S R N D i s s a m e r e g i s t e r a s C T X
a = % e a x
b = % e b x
f = % r9 d
g = % r10 d
h = % r11 d
old_ h = % r11 d
T1 = % r12 d
y0 = % r13 d
y1 = % r14 d
y2 = % r15 d
_ XFER_ S I Z E = 2 * 6 4 * 4 # 2 b l o c k s , 6 4 r o u n d s , 4 b y t e s / r o u n d
_ XMM_ S A V E _ S I Z E = 0
_ INP_ E N D _ S I Z E = 8
_ INP_ S I Z E = 8
_ CTX_ S I Z E = 8
_ XFER = 0
_ XMM_ S A V E = _ X F E R + _ X F E R _ S I Z E
_ INP_ E N D = _ X M M _ S A V E + _ X M M _ S A V E _ S I Z E
_ INP = _ I N P _ E N D + _ I N P _ E N D _ S I Z E
_ CTX = _ I N P + _ I N P _ S I Z E
2021-02-24 19:29:22 +03:00
STACK_ S I Z E = _ C T X + _ C T X _ S I Z E
2013-03-27 00:59:10 +04:00
# rotate_ X s
# Rotate v a l u e s o f s y m b o l s X 0 . . . X 3
.macro rotate_Xs
X_ = X 0
X0 = X 1
X1 = X 2
X2 = X 3
X3 = X _
.endm
# ROTATE_ A R G S
# Rotate v a l u e s o f s y m b o l s a . . . h
.macro ROTATE_ARGS
old_ h = h
TMP_ = h
h = g
g = f
f = e
e = d
d = c
c = b
b = a
a = T M P _
.endm
.macro FOUR_ROUNDS_AND_SCHED disp
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # RND N + 0 ## # # # # # # # # # # # # # # # # # # # # # # # # # #
mov a , y 3 # y 3 = a # M A J A
rorx $ 2 5 , e , y 0 # y 0 = e > > 2 5 # S 1 A
rorx $ 1 1 , e , y 1 # y 1 = e > > 1 1 # S 1 B
addl \ d i s p ( % r s p , S R N D ) , h # h = k + w + h # - -
or c , y 3 # y 3 = a | c # M A J A
vpalignr $ 4 , X 2 , X 3 , X T M P 0 # X T M P 0 = W [ - 7 ]
mov f , y 2 # y 2 = f # C H
rorx $ 1 3 , a , T 1 # T 1 = a > > 1 3 # S 0 B
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) # S 1
xor g , y 2 # y 2 = f ^ g # C H
vpaddd X 0 , X T M P 0 , X T M P 0 # X T M P 0 = W [ - 7 ] + W [ - 1 6 ] # y 1 = ( e > > 6 ) # S 1
rorx $ 6 , e , y 1 # y 1 = ( e > > 6 ) # S 1
and e , y 2 # y 2 = ( f ^ g ) & e # C H
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) ^ ( e > > 6 ) # S 1
rorx $ 2 2 , a , y 1 # y 1 = a > > 2 2 # S 0 A
add h , d # d = k + w + h + d # - -
and b , y 3 # y 3 = ( a | c ) & b # M A J A
vpalignr $ 4 , X 0 , X 1 , X T M P 1 # X T M P 1 = W [ - 1 5 ]
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) # S 0
rorx $ 2 , a , T 1 # T 1 = ( a > > 2 ) # S 0
xor g , y 2 # y 2 = C H = ( ( f ^ g ) & e ) ^ g # C H
vpsrld $ 7 , X T M P 1 , X T M P 2
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) ^ ( a > > 2 ) # S 0
mov a , T 1 # T 1 = a # M A J B
and c , T 1 # T 1 = a & c # M A J B
add y 0 , y 2 # y 2 = S 1 + C H # - -
vpslld $ ( 3 2 - 7 ) , X T M P 1 , X T M P 3
or T 1 , y 3 # y 3 = M A J = ( a | c ) & b ) | ( a & c ) # M A J
add y 1 , h # h = k + w + h + S 0 # - -
add y 2 , d # d = k + w + h + d + S 1 + C H = d + t 1 # - -
vpor X T M P 2 , X T M P 3 , X T M P 3 # X T M P 3 = W [ - 1 5 ] r o r 7
vpsrld $ 1 8 , X T M P 1 , X T M P 2
add y 2 , h # h = k + w + h + S 0 + S 1 + C H = t 1 + S 0 # - -
add y 3 , h # h = t 1 + S 0 + M A J # - -
ROTATE_ A R G S
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # RND N + 1 ## # # # # # # # # # # # # # # # # # # # # # # # # # #
mov a , y 3 # y 3 = a # M A J A
rorx $ 2 5 , e , y 0 # y 0 = e > > 2 5 # S 1 A
rorx $ 1 1 , e , y 1 # y 1 = e > > 1 1 # S 1 B
offset = \ d i s p + 1 * 4
addl o f f s e t ( % r s p , S R N D ) , h # h = k + w + h # - -
or c , y 3 # y 3 = a | c # M A J A
vpsrld $ 3 , X T M P 1 , X T M P 4 # X T M P 4 = W [ - 1 5 ] > > 3
mov f , y 2 # y 2 = f # C H
rorx $ 1 3 , a , T 1 # T 1 = a > > 1 3 # S 0 B
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) # S 1
xor g , y 2 # y 2 = f ^ g # C H
rorx $ 6 , e , y 1 # y 1 = ( e > > 6 ) # S 1
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) ^ ( e > > 6 ) # S 1
rorx $ 2 2 , a , y 1 # y 1 = a > > 2 2 # S 0 A
and e , y 2 # y 2 = ( f ^ g ) & e # C H
add h , d # d = k + w + h + d # - -
vpslld $ ( 3 2 - 1 8 ) , X T M P 1 , X T M P 1
and b , y 3 # y 3 = ( a | c ) & b # M A J A
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) # S 0
vpxor X T M P 1 , X T M P 3 , X T M P 3
rorx $ 2 , a , T 1 # T 1 = ( a > > 2 ) # S 0
xor g , y 2 # y 2 = C H = ( ( f ^ g ) & e ) ^ g # C H
vpxor X T M P 2 , X T M P 3 , X T M P 3 # X T M P 3 = W [ - 1 5 ] r o r 7 ^ W [ - 1 5 ] r o r 1 8
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) ^ ( a > > 2 ) # S 0
mov a , T 1 # T 1 = a # M A J B
and c , T 1 # T 1 = a & c # M A J B
add y 0 , y 2 # y 2 = S 1 + C H # - -
vpxor X T M P 4 , X T M P 3 , X T M P 1 # X T M P 1 = s0
vpshufd $ 0 b11 1 1 1 0 1 0 , X 3 , X T M P 2 # X T M P 2 = W [ - 2 ] { B B A A }
or T 1 , y 3 # y 3 = M A J = ( a | c ) & b ) | ( a & c ) # M A J
add y 1 , h # h = k + w + h + S 0 # - -
vpaddd X T M P 1 , X T M P 0 , X T M P 0 # X T M P 0 = W [ - 1 6 ] + W [ - 7 ] + s0
add y 2 , d # d = k + w + h + d + S 1 + C H = d + t 1 # - -
add y 2 , h # h = k + w + h + S 0 + S 1 + C H = t 1 + S 0 # - -
add y 3 , h # h = t 1 + S 0 + M A J # - -
vpsrld $ 1 0 , X T M P 2 , X T M P 4 # X T M P 4 = W [ - 2 ] > > 1 0 { B B A A }
ROTATE_ A R G S
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # RND N + 2 ## # # # # # # # # # # # # # # # # # # # # # # # # # #
mov a , y 3 # y 3 = a # M A J A
rorx $ 2 5 , e , y 0 # y 0 = e > > 2 5 # S 1 A
offset = \ d i s p + 2 * 4
addl o f f s e t ( % r s p , S R N D ) , h # h = k + w + h # - -
vpsrlq $ 1 9 , X T M P 2 , X T M P 3 # X T M P 3 = W [ - 2 ] r o r 1 9 { x B x A }
rorx $ 1 1 , e , y 1 # y 1 = e > > 1 1 # S 1 B
or c , y 3 # y 3 = a | c # M A J A
mov f , y 2 # y 2 = f # C H
xor g , y 2 # y 2 = f ^ g # C H
rorx $ 1 3 , a , T 1 # T 1 = a > > 1 3 # S 0 B
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) # S 1
vpsrlq $ 1 7 , X T M P 2 , X T M P 2 # X T M P 2 = W [ - 2 ] r o r 1 7 { x B x A }
and e , y 2 # y 2 = ( f ^ g ) & e # C H
rorx $ 6 , e , y 1 # y 1 = ( e > > 6 ) # S 1
vpxor X T M P 3 , X T M P 2 , X T M P 2
add h , d # d = k + w + h + d # - -
and b , y 3 # y 3 = ( a | c ) & b # M A J A
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) ^ ( e > > 6 ) # S 1
rorx $ 2 2 , a , y 1 # y 1 = a > > 2 2 # S 0 A
vpxor X T M P 2 , X T M P 4 , X T M P 4 # X T M P 4 = s1 { x B x A }
xor g , y 2 # y 2 = C H = ( ( f ^ g ) & e ) ^ g # C H
vpshufb S H U F _ 0 0 B A , X T M P 4 , X T M P 4 # X T M P 4 = s1 { 0 0 B A }
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) # S 0
rorx $ 2 , a ,T 1 # T 1 = ( a > > 2 ) # S 0
vpaddd X T M P 4 , X T M P 0 , X T M P 0 # X T M P 0 = { . . . , . . . , W [ 1 ] , W [ 0 ] }
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) ^ ( a > > 2 ) # S 0
mov a , T 1 # T 1 = a # M A J B
and c , T 1 # T 1 = a & c # M A J B
add y 0 , y 2 # y 2 = S 1 + C H # - -
vpshufd $ 0 b01 0 1 0 0 0 0 , X T M P 0 , X T M P 2 # X T M P 2 = W [ - 2 ] { D D C C }
or T 1 , y 3 # y 3 = M A J = ( a | c ) & b ) | ( a & c ) # M A J
add y 1 ,h # h = k + w + h + S 0 # - -
add y 2 ,d # d = k + w + h + d + S 1 + C H = d + t 1 # - -
add y 2 ,h # h = k + w + h + S 0 + S 1 + C H = t 1 + S 0 # - -
add y 3 ,h # h = t 1 + S 0 + M A J # - -
ROTATE_ A R G S
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # RND N + 3 ## # # # # # # # # # # # # # # # # # # # # # # # # # #
mov a , y 3 # y 3 = a # M A J A
rorx $ 2 5 , e , y 0 # y 0 = e > > 2 5 # S 1 A
rorx $ 1 1 , e , y 1 # y 1 = e > > 1 1 # S 1 B
offset = \ d i s p + 3 * 4
addl o f f s e t ( % r s p , S R N D ) , h # h = k + w + h # - -
or c , y 3 # y 3 = a | c # M A J A
vpsrld $ 1 0 , X T M P 2 , X T M P 5 # X T M P 5 = W [ - 2 ] > > 1 0 { D D C C }
mov f , y 2 # y 2 = f # C H
rorx $ 1 3 , a , T 1 # T 1 = a > > 1 3 # S 0 B
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) # S 1
xor g , y 2 # y 2 = f ^ g # C H
vpsrlq $ 1 9 , X T M P 2 , X T M P 3 # X T M P 3 = W [ - 2 ] r o r 1 9 { x D x C }
rorx $ 6 , e , y 1 # y 1 = ( e > > 6 ) # S 1
and e , y 2 # y 2 = ( f ^ g ) & e # C H
add h , d # d = k + w + h + d # - -
and b , y 3 # y 3 = ( a | c ) & b # M A J A
vpsrlq $ 1 7 , X T M P 2 , X T M P 2 # X T M P 2 = W [ - 2 ] r o r 1 7 { x D x C }
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) ^ ( e > > 6 ) # S 1
xor g , y 2 # y 2 = C H = ( ( f ^ g ) & e ) ^ g # C H
vpxor X T M P 3 , X T M P 2 , X T M P 2
rorx $ 2 2 , a , y 1 # y 1 = a > > 2 2 # S 0 A
add y 0 , y 2 # y 2 = S 1 + C H # - -
vpxor X T M P 2 , X T M P 5 , X T M P 5 # X T M P 5 = s1 { x D x C }
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) # S 0
add y 2 , d # d = k + w + h + d + S 1 + C H = d + t 1 # - -
rorx $ 2 , a , T 1 # T 1 = ( a > > 2 ) # S 0
vpshufb S H U F _ D C 0 0 , X T M P 5 , X T M P 5 # X T M P 5 = s1 { D C 0 0 }
vpaddd X T M P 0 , X T M P 5 , X 0 # X 0 = { W [ 3 ] , W [ 2 ] , W [ 1 ] , W [ 0 ] }
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) ^ ( a > > 2 ) # S 0
mov a , T 1 # T 1 = a # M A J B
and c , T 1 # T 1 = a & c # M A J B
or T 1 , y 3 # y 3 = M A J = ( a | c ) & b ) | ( a & c ) # M A J
add y 1 , h # h = k + w + h + S 0 # - -
add y 2 , h # h = k + w + h + S 0 + S 1 + C H = t 1 + S 0 # - -
add y 3 , h # h = t 1 + S 0 + M A J # - -
ROTATE_ A R G S
rotate_ X s
.endm
.macro DO_4ROUNDS disp
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # RND N + 0 ## # # # # # # # # # # # # # # # # # # # # # # # # #
mov f , y 2 # y 2 = f # C H
rorx $ 2 5 , e , y 0 # y 0 = e > > 2 5 # S 1 A
rorx $ 1 1 , e , y 1 # y 1 = e > > 1 1 # S 1 B
xor g , y 2 # y 2 = f ^ g # C H
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) # S 1
rorx $ 6 , e , y 1 # y 1 = ( e > > 6 ) # S 1
and e , y 2 # y 2 = ( f ^ g ) & e # C H
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) ^ ( e > > 6 ) # S 1
rorx $ 1 3 , a , T 1 # T 1 = a > > 1 3 # S 0 B
xor g , y 2 # y 2 = C H = ( ( f ^ g ) & e ) ^ g # C H
rorx $ 2 2 , a , y 1 # y 1 = a > > 2 2 # S 0 A
mov a , y 3 # y 3 = a # M A J A
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) # S 0
rorx $ 2 , a , T 1 # T 1 = ( a > > 2 ) # S 0
addl \ d i s p ( % r s p , S R N D ) , h # h = k + w + h # - -
or c , y 3 # y 3 = a | c # M A J A
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) ^ ( a > > 2 ) # S 0
mov a , T 1 # T 1 = a # M A J B
and b , y 3 # y 3 = ( a | c ) & b # M A J A
and c , T 1 # T 1 = a & c # M A J B
add y 0 , y 2 # y 2 = S 1 + C H # - -
add h , d # d = k + w + h + d # - -
or T 1 , y 3 # y 3 = M A J = ( a | c ) & b ) | ( a & c ) # M A J
add y 1 , h # h = k + w + h + S 0 # - -
add y 2 , d # d = k + w + h + d + S 1 + C H = d + t 1 # - -
ROTATE_ A R G S
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # RND N + 1 ## # # # # # # # # # # # # # # # # # # # # # # # # #
add y 2 , o l d _ h # h = k + w + h + S 0 + S 1 + C H = t 1 + S 0 # - -
mov f , y 2 # y 2 = f # C H
rorx $ 2 5 , e , y 0 # y 0 = e > > 2 5 # S 1 A
rorx $ 1 1 , e , y 1 # y 1 = e > > 1 1 # S 1 B
xor g , y 2 # y 2 = f ^ g # C H
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) # S 1
rorx $ 6 , e , y 1 # y 1 = ( e > > 6 ) # S 1
and e , y 2 # y 2 = ( f ^ g ) & e # C H
add y 3 , o l d _ h # h = t 1 + S 0 + M A J # - -
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) ^ ( e > > 6 ) # S 1
rorx $ 1 3 , a , T 1 # T 1 = a > > 1 3 # S 0 B
xor g , y 2 # y 2 = C H = ( ( f ^ g ) & e ) ^ g # C H
rorx $ 2 2 , a , y 1 # y 1 = a > > 2 2 # S 0 A
mov a , y 3 # y 3 = a # M A J A
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) # S 0
rorx $ 2 , a , T 1 # T 1 = ( a > > 2 ) # S 0
offset = 4 * 1 + \ d i s p
addl o f f s e t ( % r s p , S R N D ) , h # h = k + w + h # - -
or c , y 3 # y 3 = a | c # M A J A
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) ^ ( a > > 2 ) # S 0
mov a , T 1 # T 1 = a # M A J B
and b , y 3 # y 3 = ( a | c ) & b # M A J A
and c , T 1 # T 1 = a & c # M A J B
add y 0 , y 2 # y 2 = S 1 + C H # - -
add h , d # d = k + w + h + d # - -
or T 1 , y 3 # y 3 = M A J = ( a | c ) & b ) | ( a & c ) # M A J
add y 1 , h # h = k + w + h + S 0 # - -
add y 2 , d # d = k + w + h + d + S 1 + C H = d + t 1 # - -
ROTATE_ A R G S
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # RND N + 2 ## # # # # # # # # # # # # # # # # # # # # # # # # # # # #
add y 2 , o l d _ h # h = k + w + h + S 0 + S 1 + C H = t 1 + S 0 # - -
mov f , y 2 # y 2 = f # C H
rorx $ 2 5 , e , y 0 # y 0 = e > > 2 5 # S 1 A
rorx $ 1 1 , e , y 1 # y 1 = e > > 1 1 # S 1 B
xor g , y 2 # y 2 = f ^ g # C H
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) # S 1
rorx $ 6 , e , y 1 # y 1 = ( e > > 6 ) # S 1
and e , y 2 # y 2 = ( f ^ g ) & e # C H
add y 3 , o l d _ h # h = t 1 + S 0 + M A J # - -
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) ^ ( e > > 6 ) # S 1
rorx $ 1 3 , a , T 1 # T 1 = a > > 1 3 # S 0 B
xor g , y 2 # y 2 = C H = ( ( f ^ g ) & e ) ^ g # C H
rorx $ 2 2 , a , y 1 # y 1 = a > > 2 2 # S 0 A
mov a , y 3 # y 3 = a # M A J A
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) # S 0
rorx $ 2 , a , T 1 # T 1 = ( a > > 2 ) # S 0
offset = 4 * 2 + \ d i s p
addl o f f s e t ( % r s p , S R N D ) , h # h = k + w + h # - -
or c , y 3 # y 3 = a | c # M A J A
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) ^ ( a > > 2 ) # S 0
mov a , T 1 # T 1 = a # M A J B
and b , y 3 # y 3 = ( a | c ) & b # M A J A
and c , T 1 # T 1 = a & c # M A J B
add y 0 , y 2 # y 2 = S 1 + C H # - -
add h , d # d = k + w + h + d # - -
or T 1 , y 3 # y 3 = M A J = ( a | c ) & b ) | ( a & c ) # M A J
add y 1 , h # h = k + w + h + S 0 # - -
add y 2 , d # d = k + w + h + d + S 1 + C H = d + t 1 # - -
ROTATE_ A R G S
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # RND N + 3 ## # # # # # # # # # # # # # # # # # # # # # # # # #
add y 2 , o l d _ h # h = k + w + h + S 0 + S 1 + C H = t 1 + S 0 # - -
mov f , y 2 # y 2 = f # C H
rorx $ 2 5 , e , y 0 # y 0 = e > > 2 5 # S 1 A
rorx $ 1 1 , e , y 1 # y 1 = e > > 1 1 # S 1 B
xor g , y 2 # y 2 = f ^ g # C H
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) # S 1
rorx $ 6 , e , y 1 # y 1 = ( e > > 6 ) # S 1
and e , y 2 # y 2 = ( f ^ g ) & e # C H
add y 3 , o l d _ h # h = t 1 + S 0 + M A J # - -
xor y 1 , y 0 # y 0 = ( e > > 2 5 ) ^ ( e > > 1 1 ) ^ ( e > > 6 ) # S 1
rorx $ 1 3 , a , T 1 # T 1 = a > > 1 3 # S 0 B
xor g , y 2 # y 2 = C H = ( ( f ^ g ) & e ) ^ g # C H
rorx $ 2 2 , a , y 1 # y 1 = a > > 2 2 # S 0 A
mov a , y 3 # y 3 = a # M A J A
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) # S 0
rorx $ 2 , a , T 1 # T 1 = ( a > > 2 ) # S 0
offset = 4 * 3 + \ d i s p
addl o f f s e t ( % r s p , S R N D ) , h # h = k + w + h # - -
or c , y 3 # y 3 = a | c # M A J A
xor T 1 , y 1 # y 1 = ( a > > 2 2 ) ^ ( a > > 1 3 ) ^ ( a > > 2 ) # S 0
mov a , T 1 # T 1 = a # M A J B
and b , y 3 # y 3 = ( a | c ) & b # M A J A
and c , T 1 # T 1 = a & c # M A J B
add y 0 , y 2 # y 2 = S 1 + C H # - -
add h , d # d = k + w + h + d # - -
or T 1 , y 3 # y 3 = M A J = ( a | c ) & b ) | ( a & c ) # M A J
add y 1 , h # h = k + w + h + S 0 # - -
add y 2 , d # d = k + w + h + d + S 1 + C H = d + t 1 # - -
add y 2 , h # h = k + w + h + S 0 + S 1 + C H = t 1 + S 0 # - -
add y 3 , h # h = t 1 + S 0 + M A J # - -
ROTATE_ A R G S
.endm
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
2020-01-15 06:57:29 +03:00
# # void s h a25 6 _ t r a n s f o r m _ r o r x ( s t r u c t s h a25 6 _ s t a t e * s t a t e , c o n s t u 8 * d a t a , i n t b l o c k s )
# # arg 1 : p o i n t e r t o s t a t e
2015-04-09 13:55:47 +03:00
# # arg 2 : p o i n t e r t o i n p u t d a t a
2013-03-27 00:59:10 +04:00
# # arg 3 : N u m b l o c k s
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
.text
2019-10-11 14:51:04 +03:00
SYM_ F U N C _ S T A R T ( s h a25 6 _ t r a n s f o r m _ r o r x )
2013-03-27 00:59:10 +04:00
.align 32
pushq % r b x
pushq % r12
pushq % r13
pushq % r14
pushq % r15
2021-02-24 19:29:22 +03:00
push % r b p
mov % r s p , % r b p
2013-03-27 00:59:10 +04:00
subq $ S T A C K _ S I Z E , % r s p
and $ - 3 2 , % r s p # a l i g n r s p t o 32 b y t e b o u n d a r y
shl $ 6 , N U M _ B L K S # c o n v e r t t o b y t e s
jz d o n e _ h a s h
lea - 6 4 ( I N P , N U M _ B L K S ) , N U M _ B L K S # p o i n t e r t o l a s t b l o c k
mov N U M _ B L K S , _ I N P _ E N D ( % r s p )
cmp N U M _ B L K S , I N P
je o n l y _ o n e _ b l o c k
# # load i n i t i a l d i g e s t
mov ( C T X ) , a
mov 4 * 1 ( C T X ) , b
mov 4 * 2 ( C T X ) , c
mov 4 * 3 ( C T X ) , d
mov 4 * 4 ( C T X ) , e
mov 4 * 5 ( C T X ) , f
mov 4 * 6 ( C T X ) , g
mov 4 * 7 ( C T X ) , h
vmovdqa P S H U F F L E _ B Y T E _ F L I P _ M A S K ( % r i p ) , B Y T E _ F L I P _ M A S K
vmovdqa _ S H U F _ 0 0 B A ( % r i p ) , S H U F _ 0 0 B A
vmovdqa _ S H U F _ D C 0 0 ( % r i p ) , S H U F _ D C 0 0
mov C T X , _ C T X ( % r s p )
loop0 :
# # Load f i r s t 1 6 d w o r d s f r o m t w o b l o c k s
VMOVDQ 0 * 3 2 ( I N P ) ,X T M P 0
VMOVDQ 1 * 3 2 ( I N P ) ,X T M P 1
VMOVDQ 2 * 3 2 ( I N P ) ,X T M P 2
VMOVDQ 3 * 3 2 ( I N P ) ,X T M P 3
# # byte s w a p d a t a
vpshufb B Y T E _ F L I P _ M A S K , X T M P 0 , X T M P 0
vpshufb B Y T E _ F L I P _ M A S K , X T M P 1 , X T M P 1
vpshufb B Y T E _ F L I P _ M A S K , X T M P 2 , X T M P 2
vpshufb B Y T E _ F L I P _ M A S K , X T M P 3 , X T M P 3
# # transpose d a t a i n t o h i g h / l o w h a l v e s
vperm2 i 1 2 8 $ 0 x20 , X T M P 2 , X T M P 0 , X 0
vperm2 i 1 2 8 $ 0 x31 , X T M P 2 , X T M P 0 , X 1
vperm2 i 1 2 8 $ 0 x20 , X T M P 3 , X T M P 1 , X 2
vperm2 i 1 2 8 $ 0 x31 , X T M P 3 , X T M P 1 , X 3
last_block_enter :
add $ 6 4 , I N P
mov I N P , _ I N P ( % r s p )
# # schedule 4 8 i n p u t d w o r d s , b y d o i n g 3 r o u n d s o f 1 2 e a c h
xor S R N D , S R N D
.align 16
loop1 :
2017-09-18 22:42:08 +03:00
vpaddd K 2 5 6 + 0 * 3 2 ( S R N D ) , X 0 , X F E R
2013-03-27 00:59:10 +04:00
vmovdqa X F E R , 0 * 3 2 + _ X F E R ( % r s p , S R N D )
FOUR_ R O U N D S _ A N D _ S C H E D _ X F E R + 0 * 3 2
2017-09-18 22:42:08 +03:00
vpaddd K 2 5 6 + 1 * 3 2 ( S R N D ) , X 0 , X F E R
2013-03-27 00:59:10 +04:00
vmovdqa X F E R , 1 * 3 2 + _ X F E R ( % r s p , S R N D )
FOUR_ R O U N D S _ A N D _ S C H E D _ X F E R + 1 * 3 2
2017-09-18 22:42:08 +03:00
vpaddd K 2 5 6 + 2 * 3 2 ( S R N D ) , X 0 , X F E R
2013-03-27 00:59:10 +04:00
vmovdqa X F E R , 2 * 3 2 + _ X F E R ( % r s p , S R N D )
FOUR_ R O U N D S _ A N D _ S C H E D _ X F E R + 2 * 3 2
2017-09-18 22:42:08 +03:00
vpaddd K 2 5 6 + 3 * 3 2 ( S R N D ) , X 0 , X F E R
2013-03-27 00:59:10 +04:00
vmovdqa X F E R , 3 * 3 2 + _ X F E R ( % r s p , S R N D )
FOUR_ R O U N D S _ A N D _ S C H E D _ X F E R + 3 * 3 2
add $ 4 * 3 2 , S R N D
cmp $ 3 * 4 * 3 2 , S R N D
jb l o o p1
loop2 :
# # Do l a s t 1 6 r o u n d s w i t h n o s c h e d u l i n g
2017-09-18 22:42:08 +03:00
vpaddd K 2 5 6 + 0 * 3 2 ( S R N D ) , X 0 , X F E R
2013-03-27 00:59:10 +04:00
vmovdqa X F E R , 0 * 3 2 + _ X F E R ( % r s p , S R N D )
DO_ 4 R O U N D S _ X F E R + 0 * 3 2
2017-09-18 22:42:08 +03:00
vpaddd K 2 5 6 + 1 * 3 2 ( S R N D ) , X 1 , X F E R
2013-03-27 00:59:10 +04:00
vmovdqa X F E R , 1 * 3 2 + _ X F E R ( % r s p , S R N D )
DO_ 4 R O U N D S _ X F E R + 1 * 3 2
add $ 2 * 3 2 , S R N D
vmovdqa X 2 , X 0
vmovdqa X 3 , X 1
cmp $ 4 * 4 * 3 2 , S R N D
jb l o o p2
mov _ C T X ( % r s p ) , C T X
mov _ I N P ( % r s p ) , I N P
addm ( 4 * 0 ) ( C T X ) ,a
addm ( 4 * 1 ) ( C T X ) ,b
addm ( 4 * 2 ) ( C T X ) ,c
addm ( 4 * 3 ) ( C T X ) ,d
addm ( 4 * 4 ) ( C T X ) ,e
addm ( 4 * 5 ) ( C T X ) ,f
addm ( 4 * 6 ) ( C T X ) ,g
addm ( 4 * 7 ) ( C T X ) ,h
cmp _ I N P _ E N D ( % r s p ) , I N P
ja d o n e _ h a s h
# # # # Do s e c o n d b l o c k u s i n g p r e v i o u s l y s c h e d u l e d r e s u l t s
xor S R N D , S R N D
.align 16
loop3 :
DO_ 4 R O U N D S _ X F E R + 0 * 3 2 + 1 6
DO_ 4 R O U N D S _ X F E R + 1 * 3 2 + 1 6
add $ 2 * 3 2 , S R N D
cmp $ 4 * 4 * 3 2 , S R N D
jb l o o p3
mov _ C T X ( % r s p ) , C T X
mov _ I N P ( % r s p ) , I N P
add $ 6 4 , I N P
addm ( 4 * 0 ) ( C T X ) ,a
addm ( 4 * 1 ) ( C T X ) ,b
addm ( 4 * 2 ) ( C T X ) ,c
addm ( 4 * 3 ) ( C T X ) ,d
addm ( 4 * 4 ) ( C T X ) ,e
addm ( 4 * 5 ) ( C T X ) ,f
addm ( 4 * 6 ) ( C T X ) ,g
addm ( 4 * 7 ) ( C T X ) ,h
cmp _ I N P _ E N D ( % r s p ) , I N P
jb l o o p0
ja d o n e _ h a s h
do_last_block :
VMOVDQ 0 * 1 6 ( I N P ) ,X W O R D 0
VMOVDQ 1 * 1 6 ( I N P ) ,X W O R D 1
VMOVDQ 2 * 1 6 ( I N P ) ,X W O R D 2
VMOVDQ 3 * 1 6 ( I N P ) ,X W O R D 3
vpshufb X _ B Y T E _ F L I P _ M A S K , X W O R D 0 , X W O R D 0
vpshufb X _ B Y T E _ F L I P _ M A S K , X W O R D 1 , X W O R D 1
vpshufb X _ B Y T E _ F L I P _ M A S K , X W O R D 2 , X W O R D 2
vpshufb X _ B Y T E _ F L I P _ M A S K , X W O R D 3 , X W O R D 3
jmp l a s t _ b l o c k _ e n t e r
only_one_block :
# # load i n i t i a l d i g e s t
mov ( 4 * 0 ) ( C T X ) ,a
mov ( 4 * 1 ) ( C T X ) ,b
mov ( 4 * 2 ) ( C T X ) ,c
mov ( 4 * 3 ) ( C T X ) ,d
mov ( 4 * 4 ) ( C T X ) ,e
mov ( 4 * 5 ) ( C T X ) ,f
mov ( 4 * 6 ) ( C T X ) ,g
mov ( 4 * 7 ) ( C T X ) ,h
vmovdqa P S H U F F L E _ B Y T E _ F L I P _ M A S K ( % r i p ) , B Y T E _ F L I P _ M A S K
vmovdqa _ S H U F _ 0 0 B A ( % r i p ) , S H U F _ 0 0 B A
vmovdqa _ S H U F _ D C 0 0 ( % r i p ) , S H U F _ D C 0 0
mov C T X , _ C T X ( % r s p )
jmp d o _ l a s t _ b l o c k
done_hash :
2021-02-24 19:29:22 +03:00
mov % r b p , % r s p
pop % r b p
2013-03-27 00:59:10 +04:00
popq % r15
popq % r14
popq % r13
popq % r12
popq % r b x
ret
2019-10-11 14:51:04 +03:00
SYM_ F U N C _ E N D ( s h a25 6 _ t r a n s f o r m _ r o r x )
2013-03-27 00:59:10 +04:00
crypto: x86 - make constants readonly, allow linker to merge them
A lot of asm-optimized routines in arch/x86/crypto/ keep its
constants in .data. This is wrong, they should be on .rodata.
Mnay of these constants are the same in different modules.
For example, 128-bit shuffle mask 0x000102030405060708090A0B0C0D0E0F
exists in at least half a dozen places.
There is a way to let linker merge them and use just one copy.
The rules are as follows: mergeable objects of different sizes
should not share sections. You can't put them all in one .rodata
section, they will lose "mergeability".
GCC puts its mergeable constants in ".rodata.cstSIZE" sections,
or ".rodata.cstSIZE.<object_name>" if -fdata-sections is used.
This patch does the same:
.section .rodata.cst16.SHUF_MASK, "aM", @progbits, 16
It is important that all data in such section consists of
16-byte elements, not larger ones, and there are no implicit
use of one element from another.
When this is not the case, use non-mergeable section:
.section .rodata[.VAR_NAME], "a", @progbits
This reduces .data by ~15 kbytes:
text data bss dec hex filename
11097415 2705840 2630712 16433967 fac32f vmlinux-prev.o
11112095 2690672 2630712 16433479 fac147 vmlinux.o
Merged objects are visible in System.map:
ffffffff81a28810 r POLY
ffffffff81a28810 r POLY
ffffffff81a28820 r TWOONE
ffffffff81a28820 r TWOONE
ffffffff81a28830 r PSHUFFLE_BYTE_FLIP_MASK <- merged regardless of
ffffffff81a28830 r SHUF_MASK <------------- the name difference
ffffffff81a28830 r SHUF_MASK
ffffffff81a28830 r SHUF_MASK
..
ffffffff81a28d00 r K512 <- merged three identical 640-byte tables
ffffffff81a28d00 r K512
ffffffff81a28d00 r K512
Use of object names in section name suffixes is not strictly necessary,
but might help if someday link stage will use garbage collection
to eliminate unused sections (ld --gc-sections).
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Josh Poimboeuf <jpoimboe@redhat.com>
CC: Xiaodong Liu <xiaodong.liu@intel.com>
CC: Megha Dey <megha.dey@intel.com>
CC: linux-crypto@vger.kernel.org
CC: x86@kernel.org
CC: linux-kernel@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-20 00:33:04 +03:00
.section .rodata .cst512 .K256 , " aM" , @progbits, 512
2013-03-27 00:59:10 +04:00
.align 64
K256 :
.long 0 x4 2 8 a2 f98 ,0 x71 3 7 4 4 9 1 ,0 x b5 c0 f b c f ,0 x e 9 b5 d b a5
.long 0 x4 2 8 a2 f98 ,0 x71 3 7 4 4 9 1 ,0 x b5 c0 f b c f ,0 x e 9 b5 d b a5
.long 0 x3 9 5 6 c25 b ,0 x59 f11 1 f1 ,0 x92 3 f82 a4 ,0 x a b1 c5 e d5
.long 0 x3 9 5 6 c25 b ,0 x59 f11 1 f1 ,0 x92 3 f82 a4 ,0 x a b1 c5 e d5
.long 0 xd8 0 7 a a98 ,0 x12 8 3 5 b01 ,0 x24 3 1 8 5 b e ,0 x55 0 c7 d c3
.long 0 xd8 0 7 a a98 ,0 x12 8 3 5 b01 ,0 x24 3 1 8 5 b e ,0 x55 0 c7 d c3
.long 0 x7 2 b e 5 d74 ,0 x80 d e b1 f e ,0 x9 b d c06 a7 ,0 x c19 b f17 4
.long 0 x7 2 b e 5 d74 ,0 x80 d e b1 f e ,0 x9 b d c06 a7 ,0 x c19 b f17 4
.long 0 xe4 9 b69 c1 ,0 x e f b e 4 7 8 6 ,0 x0 f c19 d c6 ,0 x24 0 c a1 c c
.long 0 xe4 9 b69 c1 ,0 x e f b e 4 7 8 6 ,0 x0 f c19 d c6 ,0 x24 0 c a1 c c
.long 0 x2 d e 9 2 c6 f ,0 x4 a74 8 4 a a ,0 x5 c b0 a9 d c ,0 x76 f98 8 d a
.long 0 x2 d e 9 2 c6 f ,0 x4 a74 8 4 a a ,0 x5 c b0 a9 d c ,0 x76 f98 8 d a
.long 0 x9 8 3 e 5 1 5 2 ,0 x a83 1 c66 d ,0 x b00 3 2 7 c8 ,0 x b f59 7 f c7
.long 0 x9 8 3 e 5 1 5 2 ,0 x a83 1 c66 d ,0 x b00 3 2 7 c8 ,0 x b f59 7 f c7
.long 0 xc6 e 0 0 b f3 ,0 x d5 a79 1 4 7 ,0 x06 c a63 5 1 ,0 x14 2 9 2 9 6 7
.long 0 xc6 e 0 0 b f3 ,0 x d5 a79 1 4 7 ,0 x06 c a63 5 1 ,0 x14 2 9 2 9 6 7
.long 0 x2 7 b70 a85 ,0 x2 e 1 b21 3 8 ,0 x4 d2 c6 d f c ,0 x53 3 8 0 d13
.long 0 x2 7 b70 a85 ,0 x2 e 1 b21 3 8 ,0 x4 d2 c6 d f c ,0 x53 3 8 0 d13
.long 0 x6 5 0 a73 5 4 ,0 x76 6 a0 a b b ,0 x81 c2 c92 e ,0 x92 7 2 2 c85
.long 0 x6 5 0 a73 5 4 ,0 x76 6 a0 a b b ,0 x81 c2 c92 e ,0 x92 7 2 2 c85
.long 0 xa2 b f e 8 a1 ,0 x a81 a66 4 b ,0 x c24 b8 b70 ,0 x c76 c51 a3
.long 0 xa2 b f e 8 a1 ,0 x a81 a66 4 b ,0 x c24 b8 b70 ,0 x c76 c51 a3
.long 0 xd1 9 2 e 8 1 9 ,0 x d69 9 0 6 2 4 ,0 x f40 e 3 5 8 5 ,0 x10 6 a a07 0
.long 0 xd1 9 2 e 8 1 9 ,0 x d69 9 0 6 2 4 ,0 x f40 e 3 5 8 5 ,0 x10 6 a a07 0
.long 0 x1 9 a4 c11 6 ,0 x1 e 3 7 6 c08 ,0 x27 4 8 7 7 4 c ,0 x34 b0 b c b5
.long 0 x1 9 a4 c11 6 ,0 x1 e 3 7 6 c08 ,0 x27 4 8 7 7 4 c ,0 x34 b0 b c b5
.long 0 x3 9 1 c0 c b3 ,0 x4 e d8 a a4 a ,0 x5 b9 c c a4 f ,0 x68 2 e 6 f f3
.long 0 x3 9 1 c0 c b3 ,0 x4 e d8 a a4 a ,0 x5 b9 c c a4 f ,0 x68 2 e 6 f f3
.long 0 x7 4 8 f82 e e ,0 x78 a56 3 6 f ,0 x84 c87 8 1 4 ,0 x8 c c70 2 0 8
.long 0 x7 4 8 f82 e e ,0 x78 a56 3 6 f ,0 x84 c87 8 1 4 ,0 x8 c c70 2 0 8
.long 0 x9 0 b e f f f a ,0 x a45 0 6 c e b ,0 x b e f9 a3 f7 ,0 x c67 1 7 8 f2
.long 0 x9 0 b e f f f a ,0 x a45 0 6 c e b ,0 x b e f9 a3 f7 ,0 x c67 1 7 8 f2
crypto: x86 - make constants readonly, allow linker to merge them
A lot of asm-optimized routines in arch/x86/crypto/ keep its
constants in .data. This is wrong, they should be on .rodata.
Mnay of these constants are the same in different modules.
For example, 128-bit shuffle mask 0x000102030405060708090A0B0C0D0E0F
exists in at least half a dozen places.
There is a way to let linker merge them and use just one copy.
The rules are as follows: mergeable objects of different sizes
should not share sections. You can't put them all in one .rodata
section, they will lose "mergeability".
GCC puts its mergeable constants in ".rodata.cstSIZE" sections,
or ".rodata.cstSIZE.<object_name>" if -fdata-sections is used.
This patch does the same:
.section .rodata.cst16.SHUF_MASK, "aM", @progbits, 16
It is important that all data in such section consists of
16-byte elements, not larger ones, and there are no implicit
use of one element from another.
When this is not the case, use non-mergeable section:
.section .rodata[.VAR_NAME], "a", @progbits
This reduces .data by ~15 kbytes:
text data bss dec hex filename
11097415 2705840 2630712 16433967 fac32f vmlinux-prev.o
11112095 2690672 2630712 16433479 fac147 vmlinux.o
Merged objects are visible in System.map:
ffffffff81a28810 r POLY
ffffffff81a28810 r POLY
ffffffff81a28820 r TWOONE
ffffffff81a28820 r TWOONE
ffffffff81a28830 r PSHUFFLE_BYTE_FLIP_MASK <- merged regardless of
ffffffff81a28830 r SHUF_MASK <------------- the name difference
ffffffff81a28830 r SHUF_MASK
ffffffff81a28830 r SHUF_MASK
..
ffffffff81a28d00 r K512 <- merged three identical 640-byte tables
ffffffff81a28d00 r K512
ffffffff81a28d00 r K512
Use of object names in section name suffixes is not strictly necessary,
but might help if someday link stage will use garbage collection
to eliminate unused sections (ld --gc-sections).
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Josh Poimboeuf <jpoimboe@redhat.com>
CC: Xiaodong Liu <xiaodong.liu@intel.com>
CC: Megha Dey <megha.dey@intel.com>
CC: linux-crypto@vger.kernel.org
CC: x86@kernel.org
CC: linux-kernel@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-20 00:33:04 +03:00
.section .rodata .cst32 .PSHUFFLE_BYTE_FLIP_MASK , " aM" , @progbits, 32
.align 32
2013-03-27 00:59:10 +04:00
PSHUFFLE_BYTE_FLIP_MASK :
.octa 0 x0 c0 d0 e 0 f08 0 9 0 a0 b04 0 5 0 6 0 7 0 0 0 1 0 2 0 3 ,0 x0 c0 d0 e 0 f08 0 9 0 a0 b04 0 5 0 6 0 7 0 0 0 1 0 2 0 3
# shuffle x B x A - > 0 0 B A
crypto: x86 - make constants readonly, allow linker to merge them
A lot of asm-optimized routines in arch/x86/crypto/ keep its
constants in .data. This is wrong, they should be on .rodata.
Mnay of these constants are the same in different modules.
For example, 128-bit shuffle mask 0x000102030405060708090A0B0C0D0E0F
exists in at least half a dozen places.
There is a way to let linker merge them and use just one copy.
The rules are as follows: mergeable objects of different sizes
should not share sections. You can't put them all in one .rodata
section, they will lose "mergeability".
GCC puts its mergeable constants in ".rodata.cstSIZE" sections,
or ".rodata.cstSIZE.<object_name>" if -fdata-sections is used.
This patch does the same:
.section .rodata.cst16.SHUF_MASK, "aM", @progbits, 16
It is important that all data in such section consists of
16-byte elements, not larger ones, and there are no implicit
use of one element from another.
When this is not the case, use non-mergeable section:
.section .rodata[.VAR_NAME], "a", @progbits
This reduces .data by ~15 kbytes:
text data bss dec hex filename
11097415 2705840 2630712 16433967 fac32f vmlinux-prev.o
11112095 2690672 2630712 16433479 fac147 vmlinux.o
Merged objects are visible in System.map:
ffffffff81a28810 r POLY
ffffffff81a28810 r POLY
ffffffff81a28820 r TWOONE
ffffffff81a28820 r TWOONE
ffffffff81a28830 r PSHUFFLE_BYTE_FLIP_MASK <- merged regardless of
ffffffff81a28830 r SHUF_MASK <------------- the name difference
ffffffff81a28830 r SHUF_MASK
ffffffff81a28830 r SHUF_MASK
..
ffffffff81a28d00 r K512 <- merged three identical 640-byte tables
ffffffff81a28d00 r K512
ffffffff81a28d00 r K512
Use of object names in section name suffixes is not strictly necessary,
but might help if someday link stage will use garbage collection
to eliminate unused sections (ld --gc-sections).
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Josh Poimboeuf <jpoimboe@redhat.com>
CC: Xiaodong Liu <xiaodong.liu@intel.com>
CC: Megha Dey <megha.dey@intel.com>
CC: linux-crypto@vger.kernel.org
CC: x86@kernel.org
CC: linux-kernel@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-20 00:33:04 +03:00
.section .rodata .cst32 ._SHUF_00BA , " aM" , @progbits, 32
.align 32
2013-03-27 00:59:10 +04:00
_SHUF_00BA :
.octa 0 xFFFFFFFFFFFFFFFF0 b0 a09 0 8 0 3 0 2 0 1 0 0 ,0 x F F F F F F F F F F F F F F F F 0 b0 a09 0 8 0 3 0 2 0 1 0 0
# shuffle x D x C - > D C 0 0
crypto: x86 - make constants readonly, allow linker to merge them
A lot of asm-optimized routines in arch/x86/crypto/ keep its
constants in .data. This is wrong, they should be on .rodata.
Mnay of these constants are the same in different modules.
For example, 128-bit shuffle mask 0x000102030405060708090A0B0C0D0E0F
exists in at least half a dozen places.
There is a way to let linker merge them and use just one copy.
The rules are as follows: mergeable objects of different sizes
should not share sections. You can't put them all in one .rodata
section, they will lose "mergeability".
GCC puts its mergeable constants in ".rodata.cstSIZE" sections,
or ".rodata.cstSIZE.<object_name>" if -fdata-sections is used.
This patch does the same:
.section .rodata.cst16.SHUF_MASK, "aM", @progbits, 16
It is important that all data in such section consists of
16-byte elements, not larger ones, and there are no implicit
use of one element from another.
When this is not the case, use non-mergeable section:
.section .rodata[.VAR_NAME], "a", @progbits
This reduces .data by ~15 kbytes:
text data bss dec hex filename
11097415 2705840 2630712 16433967 fac32f vmlinux-prev.o
11112095 2690672 2630712 16433479 fac147 vmlinux.o
Merged objects are visible in System.map:
ffffffff81a28810 r POLY
ffffffff81a28810 r POLY
ffffffff81a28820 r TWOONE
ffffffff81a28820 r TWOONE
ffffffff81a28830 r PSHUFFLE_BYTE_FLIP_MASK <- merged regardless of
ffffffff81a28830 r SHUF_MASK <------------- the name difference
ffffffff81a28830 r SHUF_MASK
ffffffff81a28830 r SHUF_MASK
..
ffffffff81a28d00 r K512 <- merged three identical 640-byte tables
ffffffff81a28d00 r K512
ffffffff81a28d00 r K512
Use of object names in section name suffixes is not strictly necessary,
but might help if someday link stage will use garbage collection
to eliminate unused sections (ld --gc-sections).
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Josh Poimboeuf <jpoimboe@redhat.com>
CC: Xiaodong Liu <xiaodong.liu@intel.com>
CC: Megha Dey <megha.dey@intel.com>
CC: linux-crypto@vger.kernel.org
CC: x86@kernel.org
CC: linux-kernel@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-20 00:33:04 +03:00
.section .rodata .cst32 ._SHUF_DC00 , " aM" , @progbits, 32
.align 32
2013-03-27 00:59:10 +04:00
_SHUF_DC00 :
.octa 0 x0 b0 a09 0 8 0 3 0 2 0 1 0 0 F F F F F F F F F F F F F F F F ,0 x0 b0 a09 0 8 0 3 0 2 0 1 0 0 F F F F F F F F F F F F F F F F
