s390x assembly pack: cache capability query results.

IBM argues that in certain scenarios capability query is really
expensive. At the same time it's asserted that query results can
be safely cached, because disabling CPACF is incompatible with
reboot-free operation.

Reviewed-by: Tim Hudson <tjh@openssl.org>

crypto/aes/asm/aes-s390x.pl

@@ -818,13 +818,9 @@ $code.=<<___ if (!$softonly);
 	tmhl	%r0,0x4000	# check for message-security assist
 	jz	.Lekey_internal
 
-	lghi	%r0,0		# query capability vector
+	llihh	%r0,0x8000
-	la	%r1,16($sp)
+	srlg	%r0,%r0,0(%r5)
-	.long	0xb92f0042	# kmc %r4,%r2
+	ng	%r0,48(%r1)	# check kmc capability vector
-
-	llihh	%r1,0x8000
-	srlg	%r1,%r1,0(%r5)
-	ng	%r1,16($sp)
 	jz	.Lekey_internal
 
 	lmg	%r0,%r1,0($inp)	# just copy 128 bits...
@@ -1444,13 +1440,10 @@ $code.=<<___ if (0);	######### kmctr code was measured to be ~12% slower
 
 	llgfr	$s0,%r0
 	lgr	$s1,%r1
-	lghi	%r0,0
+	larl	%r1,OPENSSL_s390xcap_P
-	la	%r1,16($sp)
-	.long	0xb92d2042	# kmctr %r4,%r2,%r2
-
 	llihh	%r0,0x8000	# check if kmctr supports the function code
 	srlg	%r0,%r0,0($s0)
-	ng	%r0,16($sp)
+	ng	%r0,64(%r1)	# check kmctr capability vector
 	lgr	%r0,$s0
 	lgr	%r1,$s1
 	jz	.Lctr32_km_loop
@@ -1597,12 +1590,10 @@ $code.=<<___ if(1);
 	llgfr	$s0,%r0			# put aside the function code
 	lghi	$s1,0x7f
 	nr	$s1,%r0
-	lghi	%r0,0			# query capability vector
+	larl	%r1,OPENSSL_s390xcap_P
-	la	%r1,$tweak-16($sp)
+	llihh	%r0,0x8000
-	.long	0xb92e0042		# km %r4,%r2
+	srlg	%r0,%r0,32($s1)		# check for 32+function code
-	llihh	%r1,0x8000
+	ng	%r0,32(%r1)		# check km capability vector
-	srlg	%r1,%r1,32($s1)		# check for 32+function code
-	ng	%r1,$tweak-16($sp)
 	lgr	%r0,$s0			# restore the function code
 	la	%r1,0($key1)		# restore $key1
 	jz	.Lxts_km_vanilla
@@ -2229,7 +2220,7 @@ ___
 }
 $code.=<<___;
 .string	"AES for s390x, CRYPTOGAMS by <appro\@openssl.org>"
-.comm	OPENSSL_s390xcap_P,16,8
+.comm	OPENSSL_s390xcap_P,80,8
 ___
 
 $code =~ s/\`([^\`]*)\`/eval $1/gem;

crypto/modes/asm/ghash-s390x.pl

@@ -85,9 +85,7 @@ $code.=<<___ if(!$softonly && 0);	# hardware is slow for single block...
 	tmhl	%r0,0x4000	# check for message-security-assist
 	jz	.Lsoft_gmult
 	lghi	%r0,0
-	la	%r1,16($sp)
+	lg	%r1,24(%r1)	# load second word of kimd capabilities vector
-	.long	0xb93e0004	# kimd %r0,%r4
-	lg	%r1,24($sp)
 	tmhh	%r1,0x4000	# check for function 65
 	jz	.Lsoft_gmult
 	stg	%r0,16($sp)	# arrange 16 bytes of zero input

crypto/s390xcpuid.S

@@ -5,14 +5,46 @@
 .align	16
 OPENSSL_s390x_facilities:
 	lghi	%r0,0
-	larl	%r2,OPENSSL_s390xcap_P
+	larl	%r4,OPENSSL_s390xcap_P
-	stg	%r0,8(%r2)
+	stg	%r0,8(%r4)	# wipe capability vectors
-	.long	0xb2b02000	# stfle	0(%r2)
+	stg	%r0,16(%r4)
+	stg	%r0,24(%r4)
+	stg	%r0,32(%r4)
+	stg	%r0,40(%r4)
+	stg	%r0,48(%r4)
+	stg	%r0,56(%r4)
+	stg	%r0,64(%r4)
+	stg	%r0,72(%r4)
+
+	.long	0xb2b04000	# stfle	0(%r4)
 	brc	8,.Ldone
 	lghi	%r0,1
-	.long	0xb2b02000	# stfle 0(%r2)
+	.long	0xb2b04000	# stfle 0(%r4)
 .Ldone:
-	lg	%r2,0(%r2)
+	lmg	%r2,%r3,0(%r4)
+	tmhl	%r2,0x4000	# check for message-security-assist
+	jz	.Lret
+
+	lghi	%r0,0		# query kimd capabilities
+	la	%r1,16(%r4)
+	.long	0xb93e0002	# kimd %r0,%r2
+
+	lghi	%r0,0		# query km capability vector
+	la	%r1,32(%r4)
+	.long	0xb92e0042	# km %r4,%r2
+
+	lghi	%r0,0		# query kmc capability vector
+	la	%r1,48(%r4)
+	.long	0xb92f0042	# kmc %r4,%r2
+
+	tmhh	%r3,0x0004	# check for message-security-assist-4
+	jz	.Lret
+
+	lghi	%r0,0		# query kmctr capability vector
+	la	%r1,64(%r4)
+	.long	0xb92d2042	# kmctr %r4,%r2,%r2
+
+.Lret:
 	br	%r14
 .size	OPENSSL_s390x_facilities,.-OPENSSL_s390x_facilities
 
@@ -112,4 +144,4 @@ OPENSSL_instrument_bus2:
 .section	.init
 	brasl	%r14,OPENSSL_cpuid_setup
 
-.comm	OPENSSL_s390xcap_P,16,8
+.comm	OPENSSL_s390xcap_P,80,8

crypto/sha/asm/sha1-s390x.pl

@@ -168,10 +168,7 @@ $code.=<<___ if ($kimdfunc);
 	lg	%r0,0(%r1)
 	tmhl	%r0,0x4000	# check for message-security assist
 	jz	.Lsoftware
-	lghi	%r0,0
+	lg	%r0,16(%r1)	# check kimd capabilities
-	la	%r1,`2*$SIZE_T`($sp)
-	.long	0xb93e0002	# kimd %r0,%r2
-	lg	%r0,`2*$SIZE_T`($sp)
 	tmhh	%r0,`0x8000>>$kimdfunc`
 	jz	.Lsoftware
 	lghi	%r0,$kimdfunc
@@ -238,7 +235,7 @@ $code.=<<___;
 	br	%r14
 .size	sha1_block_data_order,.-sha1_block_data_order
 .string	"SHA1 block transform for s390x, CRYPTOGAMS by <appro\@openssl.org>"
-.comm	OPENSSL_s390xcap_P,16,8
+.comm	OPENSSL_s390xcap_P,80,8
 ___
 
 $code =~ s/\`([^\`]*)\`/eval $1/gem;

crypto/sha/asm/sha512-s390x.pl

@@ -240,10 +240,7 @@ $code.=<<___ if ($kimdfunc);
 	lg	%r0,0(%r1)
 	tmhl	%r0,0x4000	# check for message-security assist
 	jz	.Lsoftware
-	lghi	%r0,0
+	lg	%r0,16(%r1)	# check kimd capabilities
-	la	%r1,`2*$SIZE_T`($sp)
-	.long	0xb93e0002	# kimd %r0,%r2
-	lg	%r0,`2*$SIZE_T`($sp)
 	tmhh	%r0,`0x8000>>$kimdfunc`
 	jz	.Lsoftware
 	lghi	%r0,$kimdfunc
@@ -311,7 +308,7 @@ $code.=<<___;
 	br	%r14
 .size	$Func,.-$Func
 .string	"SHA${label} block transform for s390x, CRYPTOGAMS by <appro\@openssl.org>"
-.comm	OPENSSL_s390xcap_P,16,8
+.comm	OPENSSL_s390xcap_P,80,8
 ___
 
 $code =~ s/\`([^\`]*)\`/eval $1/gem;