Move & split opensslconf.h.in
[openssl.git] / crypto / x86cpuid.pl
index da3e5bc..8677232 100644 (file)
@@ -1,10 +1,13 @@
 #!/usr/bin/env perl
 
-push(@INC,"perlasm");
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+push(@INC, "${dir}perlasm", "perlasm");
 require "x86asm.pl";
 
 &asm_init($ARGV[0],"x86cpuid");
 
+for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
+
 &function_begin("OPENSSL_ia32_cpuid");
        &xor    ("edx","edx");
        &pushf  ();
@@ -16,13 +19,140 @@ require "x86asm.pl";
        &pushf  ();
        &pop    ("eax");
        &xor    ("ecx","eax");
+       &xor    ("eax","eax");
        &bt     ("ecx",21);
        &jnc    (&label("nocpuid"));
+       &mov    ("esi",&wparam(0));
+       &mov    (&DWP(8,"esi"),"eax");  # clear 3rd word
+       &cpuid  ();
+       &mov    ("edi","eax");          # max value for standard query level
+
+       &xor    ("eax","eax");
+       &cmp    ("ebx",0x756e6547);     # "Genu"
+       &setne  (&LB("eax"));
+       &mov    ("ebp","eax");
+       &cmp    ("edx",0x49656e69);     # "ineI"
+       &setne  (&LB("eax"));
+       &or     ("ebp","eax");
+       &cmp    ("ecx",0x6c65746e);     # "ntel"
+       &setne  (&LB("eax"));
+       &or     ("ebp","eax");          # 0 indicates Intel CPU
+       &jz     (&label("intel"));
+
+       &cmp    ("ebx",0x68747541);     # "Auth"
+       &setne  (&LB("eax"));
+       &mov    ("esi","eax");
+       &cmp    ("edx",0x69746E65);     # "enti"
+       &setne  (&LB("eax"));
+       &or     ("esi","eax");
+       &cmp    ("ecx",0x444D4163);     # "cAMD"
+       &setne  (&LB("eax"));
+       &or     ("esi","eax");          # 0 indicates AMD CPU
+       &jnz    (&label("intel"));
+
+       # AMD specific
+       &mov    ("eax",0x80000000);
+       &cpuid  ();
+       &cmp    ("eax",0x80000001);
+       &jb     (&label("intel"));
+       &mov    ("esi","eax");
+       &mov    ("eax",0x80000001);
+       &cpuid  ();
+       &or     ("ebp","ecx");
+       &and    ("ebp",1<<11|1);        # isolate XOP bit
+       &cmp    ("esi",0x80000008);
+       &jb     (&label("intel"));
+
+       &mov    ("eax",0x80000008);
+       &cpuid  ();
+       &movz   ("esi",&LB("ecx"));     # number of cores - 1
+       &inc    ("esi");                # number of cores
+
        &mov    ("eax",1);
+       &xor    ("ecx","ecx");
        &cpuid  ();
+       &bt     ("edx",28);
+       &jnc    (&label("generic"));
+       &shr    ("ebx",16);
+       &and    ("ebx",0xff);
+       &cmp    ("ebx","esi");
+       &ja     (&label("generic"));
+       &and    ("edx",0xefffffff);     # clear hyper-threading bit
+       &jmp    (&label("generic"));
+       
+&set_label("intel");
+       &cmp    ("edi",7);
+       &jb     (&label("cacheinfo"));
+
+       &mov    ("esi",&wparam(0));
+       &mov    ("eax",7);
+       &xor    ("ecx","ecx");
+       &cpuid  ();
+       &mov    (&DWP(8,"esi"),"ebx");
+
+&set_label("cacheinfo");
+       &cmp    ("edi",4);
+       &mov    ("edi",-1);
+       &jb     (&label("nocacheinfo"));
+
+       &mov    ("eax",4);
+       &mov    ("ecx",0);              # query L1D
+       &cpuid  ();
+       &mov    ("edi","eax");
+       &shr    ("edi",14);
+       &and    ("edi",0xfff);          # number of cores -1 per L1D
+
+&set_label("nocacheinfo");
+       &mov    ("eax",1);
+       &xor    ("ecx","ecx");
+       &cpuid  ();
+       &and    ("edx",0xbfefffff);     # force reserved bits #20, #30 to 0
+       &cmp    ("ebp",0);
+       &jne    (&label("notintel"));
+       &or     ("edx",1<<30);          # set reserved bit#30 on Intel CPUs
+       &and    (&HB("eax"),15);        # familiy ID
+       &cmp    (&HB("eax"),15);        # P4?
+       &jne    (&label("notintel"));
+       &or     ("edx",1<<20);          # set reserved bit#20 to engage RC4_CHAR
+&set_label("notintel");
+       &bt     ("edx",28);             # test hyper-threading bit
+       &jnc    (&label("generic"));
+       &and    ("edx",0xefffffff);
+       &cmp    ("edi",0);
+       &je     (&label("generic"));
+
+       &or     ("edx",0x10000000);
+       &shr    ("ebx",16);
+       &cmp    (&LB("ebx"),1);
+       &ja     (&label("generic"));
+       &and    ("edx",0xefffffff);     # clear hyper-threading bit if not
+
+&set_label("generic");
+       &and    ("ebp",1<<11);          # isolate AMD XOP flag
+       &and    ("ecx",0xfffff7ff);     # force 11th bit to 0
+       &mov    ("esi","edx");
+       &or     ("ebp","ecx");          # merge AMD XOP flag
+
+       &bt     ("ecx",27);             # check OSXSAVE bit
+       &jnc    (&label("clear_avx"));
+       &xor    ("ecx","ecx");
+       &data_byte(0x0f,0x01,0xd0);     # xgetbv
+       &and    ("eax",6);
+       &cmp    ("eax",6);
+       &je     (&label("done"));
+       &cmp    ("eax",2);
+       &je     (&label("clear_avx"));
+&set_label("clear_xmm");
+       &and    ("ebp",0xfdfffffd);     # clear AESNI and PCLMULQDQ bits
+       &and    ("esi",0xfeffffff);     # clear FXSR
+&set_label("clear_avx");
+       &and    ("ebp",0xefffe7ff);     # clear AVX, FMA and AMD XOP bits
+       &mov    ("edi",&wparam(0));
+       &and    (&DWP(8,"edi"),0xffffffdf);     # clear AVX2
+&set_label("done");
+       &mov    ("eax","esi");
+       &mov    ("edx","ebp");
 &set_label("nocpuid");
-       &mov    ("eax","edx");
-       &mov    ("edx","ecx");
 &function_end("OPENSSL_ia32_cpuid");
 
 &external_label("OPENSSL_ia32cap_P");
@@ -38,6 +168,322 @@ require "x86asm.pl";
        &ret    ();
 &function_end_B("OPENSSL_rdtsc");
 
+# This works in Ring 0 only [read DJGPP+MS-DOS+privileged DPMI host],
+# but it's safe to call it on any [supported] 32-bit platform...
+# Just check for [non-]zero return value...
+&function_begin_B("OPENSSL_instrument_halt","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
+       &picmeup("ecx","OPENSSL_ia32cap_P");
+       &bt     (&DWP(0,"ecx"),4);
+       &jnc    (&label("nohalt"));     # no TSC
+
+       &data_word(0x9058900e);         # push %cs; pop %eax
+       &and    ("eax",3);
+       &jnz    (&label("nohalt"));     # not enough privileges
+
+       &pushf  ();
+       &pop    ("eax");
+       &bt     ("eax",9);
+       &jnc    (&label("nohalt"));     # interrupts are disabled
+
+       &rdtsc  ();
+       &push   ("edx");
+       &push   ("eax");
+       &halt   ();
+       &rdtsc  ();
+
+       &sub    ("eax",&DWP(0,"esp"));
+       &sbb    ("edx",&DWP(4,"esp"));
+       &add    ("esp",8);
+       &ret    ();
+
+&set_label("nohalt");
+       &xor    ("eax","eax");
+       &xor    ("edx","edx");
+       &ret    ();
+&function_end_B("OPENSSL_instrument_halt");
+
+# Essentially there is only one use for this function. Under DJGPP:
+#
+#      #include <go32.h>
+#      ...
+#      i=OPENSSL_far_spin(_dos_ds,0x46c);
+#      ...
+# to obtain the number of spins till closest timer interrupt.
+
+&function_begin_B("OPENSSL_far_spin");
+       &pushf  ();
+       &pop    ("eax");
+       &bt     ("eax",9);
+       &jnc    (&label("nospin"));     # interrupts are disabled
+
+       &mov    ("eax",&DWP(4,"esp"));
+       &mov    ("ecx",&DWP(8,"esp"));
+       &data_word (0x90d88e1e);        # push %ds, mov %eax,%ds
+       &xor    ("eax","eax");
+       &mov    ("edx",&DWP(0,"ecx"));
+       &jmp    (&label("spin"));
+
+       &align  (16);
+&set_label("spin");
+       &inc    ("eax");
+       &cmp    ("edx",&DWP(0,"ecx"));
+       &je     (&label("spin"));
+
+       &data_word (0x1f909090);        # pop   %ds
+       &ret    ();
+
+&set_label("nospin");
+       &xor    ("eax","eax");
+       &xor    ("edx","edx");
+       &ret    ();
+&function_end_B("OPENSSL_far_spin");
+
+&function_begin_B("OPENSSL_wipe_cpu","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
+       &xor    ("eax","eax");
+       &xor    ("edx","edx");
+       &picmeup("ecx","OPENSSL_ia32cap_P");
+       &mov    ("ecx",&DWP(0,"ecx"));
+       &bt     (&DWP(0,"ecx"),1);
+       &jnc    (&label("no_x87"));
+       if ($sse2) {
+               &and    ("ecx",1<<26|1<<24);    # check SSE2 and FXSR bits
+               &cmp    ("ecx",1<<26|1<<24);
+               &jne    (&label("no_sse2"));
+               &pxor   ("xmm0","xmm0");
+               &pxor   ("xmm1","xmm1");
+               &pxor   ("xmm2","xmm2");
+               &pxor   ("xmm3","xmm3");
+               &pxor   ("xmm4","xmm4");
+               &pxor   ("xmm5","xmm5");
+               &pxor   ("xmm6","xmm6");
+               &pxor   ("xmm7","xmm7");
+       &set_label("no_sse2");
+       }
+       # just a bunch of fldz to zap the fp/mm bank followed by finit...
+       &data_word(0xeed9eed9,0xeed9eed9,0xeed9eed9,0xeed9eed9,0x90e3db9b);
+&set_label("no_x87");
+       &lea    ("eax",&DWP(4,"esp"));
+       &ret    ();
+&function_end_B("OPENSSL_wipe_cpu");
+
+&function_begin_B("OPENSSL_atomic_add");
+       &mov    ("edx",&DWP(4,"esp"));  # fetch the pointer, 1st arg
+       &mov    ("ecx",&DWP(8,"esp"));  # fetch the increment, 2nd arg
+       &push   ("ebx");
+       &nop    ();
+       &mov    ("eax",&DWP(0,"edx"));
+&set_label("spin");
+       &lea    ("ebx",&DWP(0,"eax","ecx"));
+       &nop    ();
+       &data_word(0x1ab10ff0); # lock; cmpxchg %ebx,(%edx)     # %eax is envolved and is always reloaded
+       &jne    (&label("spin"));
+       &mov    ("eax","ebx");  # OpenSSL expects the new value
+       &pop    ("ebx");
+       &ret    ();
+&function_end_B("OPENSSL_atomic_add");
+
+# This function can become handy under Win32 in situations when
+# we don't know which calling convention, __stdcall or __cdecl(*),
+# indirect callee is using. In C it can be deployed as
+#
+#ifdef OPENSSL_CPUID_OBJ
+#      type OPENSSL_indirect_call(void *f,...);
+#      ...
+#      OPENSSL_indirect_call(func,[up to $max arguments]);
+#endif
+#
+# (*)  it's designed to work even for __fastcall if number of
+#      arguments is 1 or 2!
+&function_begin_B("OPENSSL_indirect_call");
+       {
+       my ($max,$i)=(7,);      # $max has to be chosen as 4*n-1
+                               # in order to preserve eventual
+                               # stack alignment
+       &push   ("ebp");
+       &mov    ("ebp","esp");
+       &sub    ("esp",$max*4);
+       &mov    ("ecx",&DWP(12,"ebp"));
+       &mov    (&DWP(0,"esp"),"ecx");
+       &mov    ("edx",&DWP(16,"ebp"));
+       &mov    (&DWP(4,"esp"),"edx");
+       for($i=2;$i<$max;$i++)
+               {
+               # Some copies will be redundant/bogus...
+               &mov    ("eax",&DWP(12+$i*4,"ebp"));
+               &mov    (&DWP(0+$i*4,"esp"),"eax");
+               }
+       &call_ptr       (&DWP(8,"ebp"));# make the call...
+       &mov    ("esp","ebp");  # ... and just restore the stack pointer
+                               # without paying attention to what we called,
+                               # (__cdecl *func) or (__stdcall *one).
+       &pop    ("ebp");
+       &ret    ();
+       }
+&function_end_B("OPENSSL_indirect_call");
+
+&function_begin_B("OPENSSL_cleanse");
+       &mov    ("edx",&wparam(0));
+       &mov    ("ecx",&wparam(1));
+       &xor    ("eax","eax");
+       &cmp    ("ecx",7);
+       &jae    (&label("lot"));
+       &cmp    ("ecx",0);
+       &je     (&label("ret"));
+&set_label("little");
+       &mov    (&BP(0,"edx"),"al");
+       &sub    ("ecx",1);
+       &lea    ("edx",&DWP(1,"edx"));
+       &jnz    (&label("little"));
+&set_label("ret");
+       &ret    ();
+
+&set_label("lot",16);
+       &test   ("edx",3);
+       &jz     (&label("aligned"));
+       &mov    (&BP(0,"edx"),"al");
+       &lea    ("ecx",&DWP(-1,"ecx"));
+       &lea    ("edx",&DWP(1,"edx"));
+       &jmp    (&label("lot"));
+&set_label("aligned");
+       &mov    (&DWP(0,"edx"),"eax");
+       &lea    ("ecx",&DWP(-4,"ecx"));
+       &test   ("ecx",-4);
+       &lea    ("edx",&DWP(4,"edx"));
+       &jnz    (&label("aligned"));
+       &cmp    ("ecx",0);
+       &jne    (&label("little"));
+       &ret    ();
+&function_end_B("OPENSSL_cleanse");
+
+{
+my $lasttick = "esi";
+my $lastdiff = "ebx";
+my $out = "edi";
+my $cnt = "ecx";
+my $max = "ebp";
+
+&function_begin("OPENSSL_instrument_bus");
+    &mov       ("eax",0);
+    if ($sse2) {
+       &picmeup("edx","OPENSSL_ia32cap_P");
+       &bt     (&DWP(0,"edx"),4);
+       &jnc    (&label("nogo"));       # no TSC
+       &bt     (&DWP(0,"edx"),19);
+       &jnc    (&label("nogo"));       # no CLFLUSH
+
+       &mov    ($out,&wparam(0));      # load arguments
+       &mov    ($cnt,&wparam(1));
+
+       # collect 1st tick
+       &rdtsc  ();
+       &mov    ($lasttick,"eax");      # lasttick = tick
+       &mov    ($lastdiff,0);          # lastdiff = 0
+       &clflush(&DWP(0,$out));
+       &data_byte(0xf0);               # lock
+       &add    (&DWP(0,$out),$lastdiff);
+       &jmp    (&label("loop"));
+
+&set_label("loop",16);
+       &rdtsc  ();
+       &mov    ("edx","eax");          # put aside tick (yes, I neglect edx)
+       &sub    ("eax",$lasttick);      # diff
+       &mov    ($lasttick,"edx");      # lasttick = tick
+       &mov    ($lastdiff,"eax");      # lastdiff = diff
+       &clflush(&DWP(0,$out));
+       &data_byte(0xf0);               # lock
+       &add    (&DWP(0,$out),"eax");   # accumulate diff
+       &lea    ($out,&DWP(4,$out));    # ++$out
+       &sub    ($cnt,1);               # --$cnt
+       &jnz    (&label("loop"));
+
+       &mov    ("eax",&wparam(1));
+&set_label("nogo");
+    }
+&function_end("OPENSSL_instrument_bus");
+
+&function_begin("OPENSSL_instrument_bus2");
+    &mov       ("eax",0);
+    if ($sse2) {
+       &picmeup("edx","OPENSSL_ia32cap_P");
+       &bt     (&DWP(0,"edx"),4);
+       &jnc    (&label("nogo"));       # no TSC
+       &bt     (&DWP(0,"edx"),19);
+       &jnc    (&label("nogo"));       # no CLFLUSH
+
+       &mov    ($out,&wparam(0));      # load arguments
+       &mov    ($cnt,&wparam(1));
+       &mov    ($max,&wparam(2));
+
+       &rdtsc  ();                     # collect 1st tick
+       &mov    ($lasttick,"eax");      # lasttick = tick
+       &mov    ($lastdiff,0);          # lastdiff = 0
+
+       &clflush(&DWP(0,$out));
+       &data_byte(0xf0);               # lock
+       &add    (&DWP(0,$out),$lastdiff);
+
+       &rdtsc  ();                     # collect 1st diff
+       &mov    ("edx","eax");          # put aside tick (yes, I neglect edx)
+       &sub    ("eax",$lasttick);      # diff
+       &mov    ($lasttick,"edx");      # lasttick = tick
+       &mov    ($lastdiff,"eax");      # lastdiff = diff
+       &jmp    (&label("loop2"));
+
+&set_label("loop2",16);
+       &clflush(&DWP(0,$out));
+       &data_byte(0xf0);               # lock
+       &add    (&DWP(0,$out),"eax");   # accumulate diff
+
+       &sub    ($max,1);
+       &jz     (&label("done2"));
+
+       &rdtsc  ();
+       &mov    ("edx","eax");          # put aside tick (yes, I neglect edx)
+       &sub    ("eax",$lasttick);      # diff
+       &mov    ($lasttick,"edx");      # lasttick = tick
+       &cmp    ("eax",$lastdiff);
+       &mov    ($lastdiff,"eax");      # lastdiff = diff
+       &mov    ("edx",0);
+       &setne  ("dl");
+       &sub    ($cnt,"edx");           # conditional --$cnt
+       &lea    ($out,&DWP(0,$out,"edx",4));    # conditional ++$out
+       &jnz    (&label("loop2"));
+
+&set_label("done2");
+       &mov    ("eax",&wparam(1));
+       &sub    ("eax",$cnt);
+&set_label("nogo");
+    }
+&function_end("OPENSSL_instrument_bus2");
+}
+
+&function_begin_B("OPENSSL_ia32_rdrand");
+       &mov    ("ecx",8);
+&set_label("loop");
+       &rdrand ("eax");
+       &jc     (&label("break"));
+       &loop   (&label("loop"));
+&set_label("break");
+       &cmp    ("eax",0);
+       &cmove  ("eax","ecx");
+       &ret    ();
+&function_end_B("OPENSSL_ia32_rdrand");
+
+&function_begin_B("OPENSSL_ia32_rdseed");
+       &mov    ("ecx",8);
+&set_label("loop");
+       &rdseed ("eax");
+       &jc     (&label("break"));
+       &loop   (&label("loop"));
+&set_label("break");
+       &cmp    ("eax",0);
+       &cmove  ("eax","ecx");
+       &ret    ();
+&function_end_B("OPENSSL_ia32_rdseed");
+
 &initseg("OPENSSL_cpuid_setup");
 
+&hidden("OPENSSL_cpuid_setup");
+&hidden("OPENSSL_ia32cap_P");
+
 &asm_finish();