On amd64 OPENSSL_cpuid_setup and OPENSSL_ia32cap_P are now hidden so we don't

have to go through the PLT/GOT to get at them anymore.  In fact going through
the GOT now fails since we no longer have a GOT entry for OPENSSL_ia32cap_P.

Fixes the problem spotted by jasper@ and sthen@.  Based on a diff from mikeb@
who did most of the actual work of tracking down the issue.

ok millert@, mikeb@

8 files changed, 10 insertions, 18 deletions

diff --git a/src/lib/libcrypto/aes/asm/aes-x86_64.pl b/src/lib/libcrypto/aes/asm/aes-x86_64.pl
index 027b4ae2e5..48fa857d5b 100755
--- a/src/lib/libcrypto/aes/asm/aes-x86_64.pl
+++ b/src/lib/libcrypto/aes/asm/aes-x86_64.pl
@@ -1678,7 +1678,7 @@ AES_cbc_encrypt:
         lea     .LAES_Td(%rip),$sbox
 .Lcbc_picked_te:
 
-        mov     PIC_GOT(OPENSSL_ia32cap_P),%r10d
+        mov     OPENSSL_ia32cap_P(%rip),%r10d
         cmp     \$$speed_limit,%rdx
         jb      .Lcbc_slow_prologue
         test    \$15,%rdx
diff --git a/src/lib/libcrypto/perlasm/x86_64-xlate.pl b/src/lib/libcrypto/perlasm/x86_64-xlate.pl
index 50ff91e2f0..56d9b64b6f 100755
--- a/src/lib/libcrypto/perlasm/x86_64-xlate.pl
+++ b/src/lib/libcrypto/perlasm/x86_64-xlate.pl
@@ -215,8 +215,7 @@ my %globals;
         undef   $ret;
 
         # optional * ---vvv--- appears in indirect jmp/call
-        if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/ &&
-            !($line =~ /^PIC_(GOT|PLT)/)) {
+        if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/) {
             $self->{asterisk} = $1;
             $self->{label} = $2;
             ($self->{base},$self->{index},$self->{scale})=split(/,/,$3);
@@ -773,8 +772,6 @@ my $rdrand = sub {
     }
 };
 
-print "#include <machine/asm.h>\n";
-
 if ($nasm) {
     print <<___;
 default rel
diff --git a/src/lib/libcrypto/rc4/asm/rc4-x86_64.pl b/src/lib/libcrypto/rc4/asm/rc4-x86_64.pl
index ac2c05074e..d6eac205e9 100755
--- a/src/lib/libcrypto/rc4/asm/rc4-x86_64.pl
+++ b/src/lib/libcrypto/rc4/asm/rc4-x86_64.pl
@@ -442,7 +442,7 @@ private_RC4_set_key:
         xor     %r10,%r10
         xor     %r11,%r11
 
-        mov     PIC_GOT(OPENSSL_ia32cap_P),$idx#d
+        mov     OPENSSL_ia32cap_P(%rip),$idx#d
         bt      \$20,$idx#d     # RC4_CHAR?
         jc      .Lc1stloop
         jmp     .Lw1stloop
@@ -506,7 +506,7 @@ private_RC4_set_key:
 .align  16
 RC4_options:
         lea     .Lopts(%rip),%rax
-        mov     PIC_GOT(OPENSSL_ia32cap_P),%edx
+        mov     OPENSSL_ia32cap_P(%rip),%edx
         bt      \$20,%edx
         jc      .L8xchar
         bt      \$30,%edx
diff --git a/src/lib/libcrypto/x86_64cpuid.pl b/src/lib/libcrypto/x86_64cpuid.pl
index 58e9bedcf0..7b7b93b223 100644
--- a/src/lib/libcrypto/x86_64cpuid.pl
+++ b/src/lib/libcrypto/x86_64cpuid.pl
@@ -17,11 +17,10 @@ open STDOUT,"| $^X $xlate $flavour $output";
                                  ("%rdi","%rsi","%rdx","%rcx"); # Unix order
 
 print<<___;
-#include <machine/asm.h>
 .extern         OPENSSL_cpuid_setup
 .hidden         OPENSSL_cpuid_setup
 .section        .init
-        call    PIC_PLT(OPENSSL_cpuid_setup)
+        call    OPENSSL_cpuid_setup
 
 .hidden OPENSSL_ia32cap_P
 .comm   OPENSSL_ia32cap_P,8,4
diff --git a/src/lib/libssl/src/crypto/aes/asm/aes-x86_64.pl b/src/lib/libssl/src/crypto/aes/asm/aes-x86_64.pl
index 027b4ae2e5..48fa857d5b 100755
--- a/src/lib/libssl/src/crypto/aes/asm/aes-x86_64.pl
+++ b/src/lib/libssl/src/crypto/aes/asm/aes-x86_64.pl
@@ -1678,7 +1678,7 @@ AES_cbc_encrypt:
         lea     .LAES_Td(%rip),$sbox
 .Lcbc_picked_te:
 
-        mov     PIC_GOT(OPENSSL_ia32cap_P),%r10d
+        mov     OPENSSL_ia32cap_P(%rip),%r10d
         cmp     \$$speed_limit,%rdx
         jb      .Lcbc_slow_prologue
         test    \$15,%rdx
diff --git a/src/lib/libssl/src/crypto/perlasm/x86_64-xlate.pl b/src/lib/libssl/src/crypto/perlasm/x86_64-xlate.pl
index 50ff91e2f0..56d9b64b6f 100755
--- a/src/lib/libssl/src/crypto/perlasm/x86_64-xlate.pl
+++ b/src/lib/libssl/src/crypto/perlasm/x86_64-xlate.pl
@@ -215,8 +215,7 @@ my %globals;
         undef   $ret;
 
         # optional * ---vvv--- appears in indirect jmp/call
-        if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/ &&
-            !($line =~ /^PIC_(GOT|PLT)/)) {
+        if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/) {
             $self->{asterisk} = $1;
             $self->{label} = $2;
             ($self->{base},$self->{index},$self->{scale})=split(/,/,$3);
@@ -773,8 +772,6 @@ my $rdrand = sub {
     }
 };
 
-print "#include <machine/asm.h>\n";
-
 if ($nasm) {
     print <<___;
 default rel
diff --git a/src/lib/libssl/src/crypto/rc4/asm/rc4-x86_64.pl b/src/lib/libssl/src/crypto/rc4/asm/rc4-x86_64.pl
index ac2c05074e..d6eac205e9 100755
--- a/src/lib/libssl/src/crypto/rc4/asm/rc4-x86_64.pl
+++ b/src/lib/libssl/src/crypto/rc4/asm/rc4-x86_64.pl
@@ -442,7 +442,7 @@ private_RC4_set_key:
         xor     %r10,%r10
         xor     %r11,%r11
 
-        mov     PIC_GOT(OPENSSL_ia32cap_P),$idx#d
+        mov     OPENSSL_ia32cap_P(%rip),$idx#d
         bt      \$20,$idx#d     # RC4_CHAR?
         jc      .Lc1stloop
         jmp     .Lw1stloop
@@ -506,7 +506,7 @@ private_RC4_set_key:
 .align  16
 RC4_options:
         lea     .Lopts(%rip),%rax
-        mov     PIC_GOT(OPENSSL_ia32cap_P),%edx
+        mov     OPENSSL_ia32cap_P(%rip),%edx
         bt      \$20,%edx
         jc      .L8xchar
         bt      \$30,%edx
diff --git a/src/lib/libssl/src/crypto/x86_64cpuid.pl b/src/lib/libssl/src/crypto/x86_64cpuid.pl
index 58e9bedcf0..7b7b93b223 100644
--- a/src/lib/libssl/src/crypto/x86_64cpuid.pl
+++ b/src/lib/libssl/src/crypto/x86_64cpuid.pl
@@ -17,11 +17,10 @@ open STDOUT,"| $^X $xlate $flavour $output";
                                  ("%rdi","%rsi","%rdx","%rcx"); # Unix order
 
 print<<___;
-#include <machine/asm.h>
 .extern         OPENSSL_cpuid_setup
 .hidden         OPENSSL_cpuid_setup
 .section        .init
-        call    PIC_PLT(OPENSSL_cpuid_setup)
+        call    OPENSSL_cpuid_setup
 
 .hidden OPENSSL_ia32cap_P
 .comm   OPENSSL_ia32cap_P,8,4