SPARCv9 assembler pack: refine CPU detection on Linux, fix for "unaligned

opcodes detected in executable segment" error [from HEAD], original from Andy.

5 files changed, 87 insertions, 3 deletions

diff --git a/crypto/sha/asm/sha1-sparcv9.pl b/crypto/sha/asm/sha1-sparcv9.pl
index 8306fc88cc..5c161cecd6 100644
--- a/crypto/sha/asm/sha1-sparcv9.pl
+++ b/crypto/sha/asm/sha1-sparcv9.pl
@@ -276,6 +276,7 @@ $code.=<<___;
 .type	sha1_block_data_order,#function
 .size	sha1_block_data_order,(.-sha1_block_data_order)
 .asciz	"SHA1 block transform for SPARCv9, CRYPTOGAMS by <appro\@openssl.org>"
+.align	4
 ___
 
 $code =~ s/\`([^\`]*)\`/eval $1/gem;
diff --git a/crypto/sha/asm/sha1-sparcv9a.pl b/crypto/sha/asm/sha1-sparcv9a.pl
index 15eb854bad..85e8d68086 100644
--- a/crypto/sha/asm/sha1-sparcv9a.pl
+++ b/crypto/sha/asm/sha1-sparcv9a.pl
@@ -539,6 +539,7 @@ $code.=<<___;
 .type	sha1_block_data_order,#function
 .size	sha1_block_data_order,(.-sha1_block_data_order)
 .asciz	"SHA1 block transform for SPARCv9a, CRYPTOGAMS by <appro\@openssl.org>"
+.align	4
 ___
 
 # Purpose of these subroutines is to explicitly encode VIS instructions,
diff --git a/crypto/sha/asm/sha512-sparcv9.pl b/crypto/sha/asm/sha512-sparcv9.pl
index 54241aab50..ec5d78135e 100644
--- a/crypto/sha/asm/sha512-sparcv9.pl
+++ b/crypto/sha/asm/sha512-sparcv9.pl
@@ -586,6 +586,7 @@ $code.=<<___;
 .type	sha${label}_block_data_order,#function
 .size	sha${label}_block_data_order,(.-sha${label}_block_data_order)
 .asciz	"SHA${label} block transform for SPARCv9, CRYPTOGAMS by <appro\@openssl.org>"
+.align	4
 ___
 
 $code =~ s/\`([^\`]*)\`/eval $1/gem;
diff --git a/crypto/sparccpuid.S b/crypto/sparccpuid.S
index aa8b11efc9..1140b5c559 100644
--- a/crypto/sparccpuid.S
+++ b/crypto/sparccpuid.S
@@ -225,13 +225,33 @@ _sparcv9_rdtick:
 	xor	%o0,%o0,%o0
 	.word	0x91410000	!rd	%tick,%o0
 	retl
-	.word	0x93323020	!srlx	%o2,32,%o1
+	.word	0x93323020	!srlx	%o0,32,%o1
 .notick:
 	retl
 	xor	%o1,%o1,%o1
 .type	_sparcv9_rdtick,#function
 .size	_sparcv9_rdtick,.-_sparcv9_rdtick
 
+.global	_sparcv9_rdwrasi
+.align	8
+_sparcv9_rdwrasi:
+	.word	0x9340c000	!rd	%asi,%o1
+	.word	0x87820000	!wr	%o0,%g0,%asi
+	retl
+	mov	%o1,%o0
+.type	_sparcv9_rdwrasi,#function
+.size	_sparcv9_rdwrasi,.-_sparcv9_rdwrasi
+
+.global	_sparcv9_vis1_probe
+.align	8
+_sparcv9_vis1_probe:
+	.word	0x81b00c20	!fzeros	%f0
+	.word	0xc19ba002+BIAS	!ldda	[%sp+BIAS+2]%asi,%f0
+	retl
+	nop
+.type	_sparcv9_vis1_probe,#function
+.size	_sparcv9_vis1_probe,.-_sparcv9_vis1_probe
+
 .global	OPENSSL_cleanse
 .align	32
 OPENSSL_cleanse:
diff --git a/crypto/sparcv9cap.c b/crypto/sparcv9cap.c
index 5f31d20bd0..1d820f4c0a 100644
--- a/crypto/sparcv9cap.c
+++ b/crypto/sparcv9cap.c
@@ -1,6 +1,8 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <setjmp.h>
+#include <signal.h>
 #include <sys/time.h>
 #include <openssl/bn.h>
 
@@ -9,6 +11,7 @@
 #define SPARCV9_VIS1		(1<<2)
 #define SPARCV9_VIS2		(1<<3)	/* reserved */
 #define SPARCV9_FMADD		(1<<4)	/* reserved for SPARC64 V */
+
 static int OPENSSL_sparcv9cap_P=SPARCV9_TICK_PRIVILEGED;
 
 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)
@@ -23,10 +26,12 @@ int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_U
 		return bn_mul_mont_int(rp,ap,bp,np,n0,num);
 	}
 
+unsigned long	_sparcv9_rdtick(void);
+unsigned long	_sparcv9_rdwrasi(unsigned long);
+void		_sparcv9_vis1_probe(void);
+
 unsigned long OPENSSL_rdtsc(void)
 	{
-	unsigned long _sparcv9_rdtick(void);
-
 	if (OPENSSL_sparcv9cap_P&SPARCV9_TICK_PRIVILEGED)
 #if defined(__sun) && defined(__SVR4)
 		return gethrtime();
@@ -137,9 +142,16 @@ void OPENSSL_cpuid_setup(void)
 
 #else
 
+static sigjmp_buf common_jmp;
+static void common_handler(int sig) { siglongjmp(common_jmp,sig); }
+
 void OPENSSL_cpuid_setup(void)
 	{
 	char *e;
+	struct sigaction	common_act,ill_oact,bus_oact;
+	sigset_t		all_masked,oset;
+	unsigned long		oasi;
+	int			sig;
  
 	if ((e=getenv("OPENSSL_sparcv9cap")))
 		{
@@ -149,6 +161,55 @@ void OPENSSL_cpuid_setup(void)
 
 	/* For now we assume that the rest supports UltraSPARC-I* only */
 	OPENSSL_sparcv9cap_P |= SPARCV9_PREFER_FPU|SPARCV9_VIS1;
+
+	sigfillset(&all_masked);
+	sigdelset(&all_masked,SIGILL);
+	sigdelset(&all_masked,SIGTRAP);
+#ifdef SIGEMT
+	sigdelset(&all_masked,SIGEMT);
+#endif
+	sigdelset(&all_masked,SIGFPE);
+	sigdelset(&all_masked,SIGBUS);
+	sigdelset(&all_masked,SIGSEGV);
+	sigprocmask(SIG_SETMASK,&all_masked,&oset);
+
+	memset(&common_act,0,sizeof(common_act));
+	common_act.sa_handler = common_handler;
+	common_act.sa_mask    = all_masked;
+
+	sigaction(SIGILL,&common_act,&ill_oact);
+	sigaction(SIGBUS,&common_act,&bus_oact);/* T1 fails 16-bit ldda */
+	oasi = _sparcv9_rdwrasi(0xD2);		/* ASI_FL16_P */
+	if ((sig=sigsetjmp(common_jmp,0)) == 0)
+		{
+		_sparcv9_vis1_probe();
+		OPENSSL_sparcv9cap_P |= SPARCV9_VIS1;
+		}
+	else if (sig == SIGBUS)			/* T1 fails 16-bit ldda */
+		{
+		OPENSSL_sparcv9cap_P &= ~SPARCV9_PREFER_FPU;
+		}
+	else
+		{
+		OPENSSL_sparcv9cap_P &= ~SPARCV9_VIS1;
+		}
+	_sparcv9_rdwrasi(oasi);
+	sigaction(SIGBUS,&bus_oact,NULL);
+	sigaction(SIGILL,&ill_oact,NULL);
+
+	sigaction(SIGILL,&common_act,&ill_oact);
+	if (sigsetjmp(common_jmp,0) == 0)
+		{
+		_sparcv9_rdtick();
+		OPENSSL_sparcv9cap_P &= ~SPARCV9_TICK_PRIVILEGED;
+		}
+	else
+		{
+		OPENSSL_sparcv9cap_P |= SPARCV9_TICK_PRIVILEGED;
+		}
+	sigaction(SIGILL,&ill_oact,NULL);
+
+	sigprocmask(SIG_SETMASK,&oset,NULL);
 	}
 
 #endif