#

# This file adds inline T4 instruction support to OpenSSL upstream code.

#

Index: Configure

===================================================================

diff -ru openssl-1.0.1e/Configure openssl-1.0.1e/Configure

--- openssl-1.0.1e/Configure 2011-05-24 17:02:24.000000000 -0700

+++ openssl-1.0.1e/Configure 2011-07-27 10:48:17.817470000 -0700

@@ -135,7 +135,7 @@

 my $x86_64_asm="x86_64cpuid.o:x86_64-gcc.o x86_64-mont.o x86_64-mont5.o x86_64-gf2m.o modexp512-x86_64.o::aes-x86_64.o vpaes-x86_64.o bsaes-x86_64.o aesni-x86_64.o aesni-sha1-x86_64.o::md5-x86_64.o:sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o rc4-md5-x86_64.o:::wp-x86_64.o:cmll-x86_64.o cmll_misc.o:ghash-x86_64.o:";

 my $ia64_asm="ia64cpuid.o:bn-ia64.o ia64-mont.o::aes_core.o aes_cbc.o aes-ia64.o::md5-ia64.o:sha1-ia64.o sha256-ia64.o sha512-ia64.o::rc4-ia64.o rc4_skey.o:::::ghash-ia64.o::void";

-my $sparcv9_asm="sparcv9cap.o sparccpuid.o:bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o:des_enc-sparc.o fcrypt_b.o:aes_core.o aes_cbc.o aes-sparcv9.o:::sha1-sparcv9.o sha256-sparcv9.o sha512-sparcv9.o:::::::ghash-sparcv9.o::void";

 my $sparcv8_asm=":sparcv8.o:des_enc-sparc.o fcrypt_b.o:::::::::::::void";

 my $alpha_asm="alphacpuid.o:bn_asm.o alpha-mont.o:::::sha1-alpha.o:::::::ghash-alpha.o::void";

 my $mips32_asm=":bn-mips.o::aes_cbc.o aes-mips.o:::sha1-mips.o sha256-mips.o::::::::";

Index: crypto/sparccpuid.S

===================================================================

diff -ru openssl-1.0.1e/crypto/sparccpuid.S openssl-1.0.1e/crypto/sparccpuid.S

--- openssl-1.0.1e/crypto/sparccpuid.S 2011-05-24 17:02:24.000000000 -0700

+++ openssl-1.0.1e/crypto/sparccpuid.S 2011-07-27 10:48:17.817470000 -0700

 !	UltraSPARC IIe		7

 !	UltraSPARC III		7

 !	UltraSPARC T1		24

+!	SPARC T4		65(*)

+! (*)	result has lesser to do with VIS instruction latencies, rdtick

+!	appears that slow, but it does the trick in sense that FP and

+!	VIS code paths are still slower than integer-only ones.

 ! Numbers for T2 and SPARC64 V-VII are more than welcomed.

 !

 .global	_sparcv9_vis1_instrument

 .align	8

 _sparcv9_vis1_instrument:

+	.word	0x81b00d80	!fxor	%f0,%f0,%f0

+	.word	0x85b08d82	!fxor	%f2,%f2,%f2

 	.word	0x91410000	!rd	%tick,%o0

 	.word	0x81b00d80	!fxor	%f0,%f0,%f0

 	.word	0x85b08d82	!fxor	%f2,%f2,%f2

 .type	_sparcv9_fmadd_probe,#function

 .size	_sparcv9_fmadd_probe,.-_sparcv9_fmadd_probe

+.global	_sparcv9_rdcfr

+.align	8

+_sparcv9_rdcfr:

+	retl

+	.word	0x91468000	!rd	%asr26,%o0

+.type	_sparcv9_rdcfr,#function

+.size	_sparcv9_rdcfr,.-_sparcv9_rdcfr

+

+.global	_sparcv9_vis3_probe

+.align	8

+_sparcv9_vis3_probe:

+	retl

+	.word	0x81b022a0	!xmulx	%g0,%g0,%g0

+.type	_sparcv9_vis3_probe,#function

+.size	_sparcv9_vis3_probe,.-_sparcv9_vis3_probe

+

+.global	_sparcv9_random

+.align	8

+_sparcv9_random:

+	retl

+	.word	0x91b002a0	!random	%o0

+.type	_sparcv9_random,#function

+.size	_sparcv9_random,.-_sparcv9_vis3_probe

+

 .global	OPENSSL_cleanse

 .align	32

 OPENSSL_cleanse:

Index: crypto/sparcv9cap.c

===================================================================

diff -ru openssl-1.0.1e/crypto/sparcv9cap.c openssl-1.0.1e/crypto/sparcv9cap.c

--- openssl-1.0.1e/crypto/sparcv9cap.c 2011-05-24 17:02:24.000000000 -0700

+++ openssl-1.0.1e/crypto/sparcv9cap.c 2011-07-27 10:48:17.817470000 -0700

 #include <sys/time.h>

 #include <openssl/bn.h>

-#define SPARCV9_TICK_PRIVILEGED	(1<<0)

-#define SPARCV9_PREFER_FPU	(1<<1)

-#define SPARCV9_VIS1		(1<<2)

-#define SPARCV9_VIS2		(1<<3)	/* reserved */

-#define SPARCV9_FMADD		(1<<4)	/* reserved for SPARC64 V */

+#include "sparc_arch.h"

+#if defined(__GNUC__) && defined(__linux)

+__attribute__((visibility("hidden")))

+#endif

 #ifndef	_BOOT

-static int OPENSSL_sparcv9cap_P=SPARCV9_TICK_PRIVILEGED;

+unsigned int OPENSSL_sparcv9cap_P[2]={SPARCV9_TICK_PRIVILEGED,0};

 #else

-static int OPENSSL_sparcv9cap_P = SPARCV9_VIS1;

+unsigned int OPENSSL_sparcv9cap_P[2]={SPARCV9_VIS1,0};

 #endif

 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np,const BN_ULONG *n0, int num)

 	int bn_mul_mont_int(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np,const BN_ULONG *n0, int num);

-	    (OPENSSL_sparcv9cap_P&(SPARCV9_PREFER_FPU|SPARCV9_VIS1)) ==

 unsigned long	_sparcv9_vis1_instrument(void);

 void		_sparcv9_vis2_probe(void);

 void		_sparcv9_fmadd_probe(void);

+unsigned long	_sparcv9_rdcfr(void);

+void		_sparcv9_vis3_probe(void);

+unsigned long	_sparcv9_random(void);

+size_t 		_sparcv9_vis1_instrument_bus(unsigned int *,size_t);

+size_t		_sparcv9_vis1_instrument_bus2(unsigned int *,size_t,size_t);

 #ifndef _BOOT

 unsigned long OPENSSL_rdtsc(void)

 	{

-	if (OPENSSL_sparcv9cap_P&SPARCV9_TICK_PRIVILEGED)

+	if (OPENSSL_sparcv9cap_P[0]&SPARCV9_TICK_PRIVILEGED)

 #if defined(__sun) && defined(__SVR4)

 		return gethrtime();

 #else

 	}

 #endif

+size_t OPENSSL_instrument_bus(unsigned int *out,size_t cnt)

+	{

+			SPARCV9_BLK)

+		return _sparcv9_vis1_instrument_bus(out,cnt);

+	else

+		return 0;

+	}

+

+size_t OPENSSL_instrument_bus2(unsigned int *out,size_t cnt,size_t max)

+	{

+			SPARCV9_BLK)

+		return _sparcv9_vis1_instrument_bus2(out,cnt,max);

+	else

+		return 0;

+	}

+

+

 #if defined(_BOOT)

 /*

  * Hardcoding sparc capabilities for wanboot.

  */

 void OPENSSL_cpuid_setup(void)

 	{

-	OPENSSL_sparcv9cap_P = SPARCV9_VIS1;

+	OPENSSL_sparcv9cap_P[0] = SPARCV9_VIS1;

 	}

 #elif 0 && defined(__sun) && defined(__SVR4)

 	if (!strcmp (name,"SUNW,UltraSPARC") ||

 	    !strncmp(name,"SUNW,UltraSPARC-I",17))  /* covers II,III,IV */

 		{

-		OPENSSL_sparcv9cap_P |= SPARCV9_PREFER_FPU|SPARCV9_VIS1;

+		OPENSSL_sparcv9cap_P[0] |= SPARCV9_PREFER_FPU|SPARCV9_VIS1;

 		/* %tick is privileged only on UltraSPARC-I/II, but not IIe */

 		if (name[14]!='\0' && name[17]!='\0' && name[18]!='\0')

-			OPENSSL_sparcv9cap_P &= ~SPARCV9_TICK_PRIVILEGED;

+			OPENSSL_sparcv9cap_P[0] &= ~SPARCV9_TICK_PRIVILEGED;

 		return DI_WALK_TERMINATE;

 		}

 	/* This is expected to catch remaining UltraSPARCs, such as T1 */

 	else if (!strncmp(name,"SUNW,UltraSPARC",15))

 		{

-		OPENSSL_sparcv9cap_P &= ~SPARCV9_TICK_PRIVILEGED;

+		OPENSSL_sparcv9cap_P[0] &= ~SPARCV9_TICK_PRIVILEGED;

 		return DI_WALK_TERMINATE;

 		}

 	if ((e=getenv("OPENSSL_sparcv9cap")))

 		{

-		OPENSSL_sparcv9cap_P=strtoul(e,NULL,0);

+		OPENSSL_sparcv9cap_P[0]=strtoul(e,NULL,0);

 		return;

 		}

 		{

 		if (strcmp(si,"sun4v"))

 			/* FPU is preferred for all CPUs, but US-T1/2 */

-			OPENSSL_sparcv9cap_P |= SPARCV9_PREFER_FPU;

+			OPENSSL_sparcv9cap_P[0] |= SPARCV9_PREFER_FPU;

 		}

 	if (sysinfo(SI_ISALIST,si,sizeof(si))>0)

 		{

 		if (strstr(si,"+vis"))

-			OPENSSL_sparcv9cap_P |= SPARCV9_VIS1;

+			OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS1|SPARCV9_BLK;

 		if (strstr(si,"+vis2"))

 			{

-			OPENSSL_sparcv9cap_P |= SPARCV9_VIS2;

-			OPENSSL_sparcv9cap_P &= ~SPARCV9_TICK_PRIVILEGED;

+			OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS2;

+			OPENSSL_sparcv9cap_P[0] &= ~SPARCV9_TICK_PRIVILEGED;

 			return;

 			}

 		}

 	if ((e=getenv("OPENSSL_sparcv9cap")))

 		{

-		OPENSSL_sparcv9cap_P=strtoul(e,NULL,0);

+		OPENSSL_sparcv9cap_P[0]=strtoul(e,NULL,0);

+		if ((e=strchr(e,':')))

+			OPENSSL_sparcv9cap_P[1]=strtoul(e+1,NULL,0);

 		return;

 		}

 	/* Initial value, fits UltraSPARC-I&II... */

-	OPENSSL_sparcv9cap_P = SPARCV9_PREFER_FPU|SPARCV9_TICK_PRIVILEGED;

+	OPENSSL_sparcv9cap_P[0] = SPARCV9_PREFER_FPU|SPARCV9_TICK_PRIVILEGED;

 	sigfillset(&all_masked);

 	sigdelset(&all_masked,SIGILL);

 	if (sigsetjmp(common_jmp,1) == 0)

 		{

 		_sparcv9_rdtick();

-		OPENSSL_sparcv9cap_P &= ~SPARCV9_TICK_PRIVILEGED;

+		OPENSSL_sparcv9cap_P[0] &= ~SPARCV9_TICK_PRIVILEGED;

 		}

 	if (sigsetjmp(common_jmp,1) == 0)

 		{

 		_sparcv9_vis1_probe();

-		OPENSSL_sparcv9cap_P |= SPARCV9_VIS1;

+		OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS1|SPARCV9_BLK;

 		/* detect UltraSPARC-Tx, see sparccpud.S for details... */

 		if (_sparcv9_vis1_instrument() >= 12)

-			OPENSSL_sparcv9cap_P &= ~(SPARCV9_VIS1|SPARCV9_PREFER_FPU);

+			OPENSSL_sparcv9cap_P[0] &= ~(SPARCV9_VIS1|SPARCV9_PREFER_FPU);

 		else

 			{

 			_sparcv9_vis2_probe();

-			OPENSSL_sparcv9cap_P |= SPARCV9_VIS2;

+			OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS2;

 			}

 		}

 	if (sigsetjmp(common_jmp,1) == 0)

 		{

 		_sparcv9_fmadd_probe();

-		OPENSSL_sparcv9cap_P |= SPARCV9_FMADD;

+		OPENSSL_sparcv9cap_P[0] |= SPARCV9_FMADD;

 		}

+	/*

+	 * VIS3 flag is tested independently from VIS1, unlike VIS2 that is,

+	 * because VIS3 defines even integer instructions.

+	 */

+	if (sigsetjmp(common_jmp,1) == 0)

+		{

+		_sparcv9_vis3_probe();

+		OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS3;

+		}

+

+	if (sigsetjmp(common_jmp,1) == 0)

+		{

+		(void)_sparcv9_random();

+		OPENSSL_sparcv9cap_P[0] |= SPARCV9_RANDOM;

+		}

+

+	/*

+	 * In wait for better solution _sparcv9_rdcfr is masked by

+	 * VIS3 flag, because it goes to uninterruptable endless

+	 * loop on UltraSPARC II running Solaris. Things might be

+	 * different on Linux...

+	 */

+	if ((OPENSSL_sparcv9cap_P[0]&SPARCV9_VIS3) &&

+	    sigsetjmp(common_jmp,1) == 0)

+		{

+		OPENSSL_sparcv9cap_P[1] = (unsigned int)_sparcv9_rdcfr();

+		}

+

 	sigaction(SIGBUS,&bus_oact,NULL);

 	sigaction(SIGILL,&ill_oact,NULL);

Index: crypto/md5/Makefile

===================================================================

diff -ru openssl-1.0.1e/crypto/md5/Makefile openssl-1.0.1e/crypto/md5/Makefile

--- openssl-1.0.1e/crypto/md5/Makefile    2011-05-24 17:02:24.000000000 -0700

+++ openssl-1.0.1e/crypto/md5/Makefile    2011-07-27 10:48:17.817470000 -0700

@@ -52,6 +52,9 @@

 	$(CC) $(CFLAGS) -E asm/md5-ia64.S | \

 	$(PERL) -ne 's/;\s+/;\n/g; print;' > $@

+md5-sparcv9.S:	asm/md5-sparcv9.pl

+	$(PERL) asm/md5-sparcv9.pl $@ $(CFLAGS)

+

 files:

 	$(PERL) $(TOP)/util/files.pl Makefile >> $(TOP)/MINFO

Index: crypto/md5/md5_locl.h