LCOV - code coverage report
Current view: top level - src - hwf-x86.c (source / functions) Hit Total Coverage
Test: coverage.info Lines: 54 71 76.1 %
Date: 2016-09-12 12:56:58 Functions: 5 5 100.0 %

          Line data    Source code
       1             : /* hwf-x86.c - Detect hardware features - x86 part
       2             :  * Copyright (C) 2007, 2011, 2012  Free Software Foundation, Inc.
       3             :  * Copyright (C) 2012  Jussi Kivilinna
       4             :  *
       5             :  * This file is part of Libgcrypt.
       6             :  *
       7             :  * Libgcrypt is free software; you can redistribute it and/or modify
       8             :  * it under the terms of the GNU Lesser General Public License as
       9             :  * published by the Free Software Foundation; either version 2.1 of
      10             :  * the License, or (at your option) any later version.
      11             :  *
      12             :  * Libgcrypt is distributed in the hope that it will be useful,
      13             :  * but WITHOUT ANY WARRANTY; without even the implied warranty of
      14             :  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
      15             :  * GNU Lesser General Public License for more details.
      16             :  *
      17             :  * You should have received a copy of the GNU Lesser General Public
      18             :  * License along with this program; if not, see <http://www.gnu.org/licenses/>.
      19             :  */
      20             : 
      21             : #include <config.h>
      22             : #include <stdio.h>
      23             : #include <stdlib.h>
      24             : #include <string.h>
      25             : #include <stdarg.h>
      26             : #include <unistd.h>
      27             : 
      28             : #include "g10lib.h"
      29             : #include "hwf-common.h"
      30             : 
      31             : #if !defined (__i386__) && !defined (__x86_64__)
      32             : # error Module build for wrong CPU.
      33             : #endif
      34             : 
      35             : /* We use the next macro to decide whether we can test for certain
      36             :    features.  */
      37             : #undef HAS_X86_CPUID
      38             : 
      39             : #if defined (__i386__) && SIZEOF_UNSIGNED_LONG == 4 && defined (__GNUC__)
      40             : # define HAS_X86_CPUID 1
      41             : 
      42             : static int
      43             : is_cpuid_available(void)
      44             : {
      45             :   int has_cpuid = 0;
      46             : 
      47             :   /* Detect the CPUID feature by testing some undefined behaviour (16
      48             :      vs 32 bit pushf/popf). */
      49             :   asm volatile
      50             :     ("pushf\n\t"                 /* Copy flags to EAX.  */
      51             :      "popl %%eax\n\t"
      52             :      "movl %%eax, %%ecx\n\t"     /* Save flags into ECX.  */
      53             :      "xorl $0x200000, %%eax\n\t" /* Toggle ID bit and copy it to the flags.  */
      54             :      "pushl %%eax\n\t"
      55             :      "popf\n\t"
      56             :      "pushf\n\t"                 /* Copy changed flags again to EAX.  */
      57             :      "popl %%eax\n\t"
      58             :      "pushl %%ecx\n\t"           /* Restore flags from ECX.  */
      59             :      "popf\n\t"
      60             :      "xorl %%eax, %%ecx\n\t"     /* Compare flags against saved flags.  */
      61             :      "jz .Lno_cpuid%=\n\t"       /* Toggling did not work, thus no CPUID.  */
      62             :      "movl $1, %0\n"             /* Worked. true -> HAS_CPUID.  */
      63             :      ".Lno_cpuid%=:\n\t"
      64             :      : "+r" (has_cpuid)
      65             :      :
      66             :      : "%eax", "%ecx", "cc"
      67             :      );
      68             : 
      69             :   return has_cpuid;
      70             : }
      71             : 
      72             : static void
      73             : get_cpuid(unsigned int in, unsigned int *eax, unsigned int *ebx,
      74             :           unsigned int *ecx, unsigned int *edx)
      75             : {
      76             :   unsigned int regs[4];
      77             : 
      78             :   asm volatile
      79             :     ("pushl %%ebx\n\t"           /* Save GOT register.  */
      80             :      "movl %1, %%ebx\n\t"
      81             :      "cpuid\n\t"
      82             :      "movl %%ebx, %1\n\t"
      83             :      "popl %%ebx\n\t"            /* Restore GOT register. */
      84             :      : "=a" (regs[0]), "=D" (regs[1]), "=c" (regs[2]), "=d" (regs[3])
      85             :      : "0" (in), "1" (0), "2" (0), "3" (0)
      86             :      : "cc"
      87             :      );
      88             : 
      89             :   if (eax)
      90             :     *eax = regs[0];
      91             :   if (ebx)
      92             :     *ebx = regs[1];
      93             :   if (ecx)
      94             :     *ecx = regs[2];
      95             :   if (edx)
      96             :     *edx = regs[3];
      97             : }
      98             : 
      99             : #if defined(ENABLE_AVX_SUPPORT) || defined(ENABLE_AVX2_SUPPORT)
     100             : static unsigned int
     101             : get_xgetbv(void)
     102             : {
     103             :   unsigned int t_eax, t_edx;
     104             : 
     105             :   asm volatile
     106             :     ("xgetbv\n\t"
     107             :      : "=a" (t_eax), "=d" (t_edx)
     108             :      : "c" (0)
     109             :     );
     110             : 
     111             :   return t_eax;
     112             : }
     113             : #endif /* ENABLE_AVX_SUPPORT || ENABLE_AVX2_SUPPORT */
     114             : 
     115             : #endif /* i386 && GNUC */
     116             : 
     117             : 
     118             : #if defined (__x86_64__) && defined (__GNUC__)
     119             : # define HAS_X86_CPUID 1
     120             : 
     121             : static int
     122          32 : is_cpuid_available(void)
     123             : {
     124          32 :   return 1;
     125             : }
     126             : 
     127             : static void
     128          96 : get_cpuid(unsigned int in, unsigned int *eax, unsigned int *ebx,
     129             :           unsigned int *ecx, unsigned int *edx)
     130             : {
     131             :   unsigned int regs[4];
     132             : 
     133          96 :   asm volatile
     134             :     ("cpuid\n\t"
     135             :      : "=a" (regs[0]), "=b" (regs[1]), "=c" (regs[2]), "=d" (regs[3])
     136             :      : "0" (in), "1" (0), "2" (0), "3" (0)
     137             :      : "cc"
     138             :      );
     139             : 
     140          96 :   if (eax)
     141          64 :     *eax = regs[0];
     142          96 :   if (ebx)
     143          64 :     *ebx = regs[1];
     144          96 :   if (ecx)
     145          64 :     *ecx = regs[2];
     146          96 :   if (edx)
     147          32 :     *edx = regs[3];
     148          96 : }
     149             : 
     150             : #if defined(ENABLE_AVX_SUPPORT) || defined(ENABLE_AVX2_SUPPORT)
     151             : static unsigned int
     152          32 : get_xgetbv(void)
     153             : {
     154             :   unsigned int t_eax, t_edx;
     155             : 
     156          32 :   asm volatile
     157             :     ("xgetbv\n\t"
     158             :      : "=a" (t_eax), "=d" (t_edx)
     159             :      : "c" (0)
     160             :     );
     161             : 
     162          32 :   return t_eax;
     163             : }
     164             : #endif /* ENABLE_AVX_SUPPORT || ENABLE_AVX2_SUPPORT */
     165             : 
     166             : #endif /* x86-64 && GNUC */
     167             : 
     168             : 
     169             : #ifdef HAS_X86_CPUID
     170             : static unsigned int
     171          32 : detect_x86_gnuc (void)
     172             : {
     173             :   char vendor_id[12+1];
     174             :   unsigned int features;
     175          32 :   unsigned int os_supports_avx_avx2_registers = 0;
     176             :   unsigned int max_cpuid_level;
     177             :   unsigned int fms, family, model;
     178          32 :   unsigned int result = 0;
     179             : 
     180             :   (void)os_supports_avx_avx2_registers;
     181             : 
     182          32 :   if (!is_cpuid_available())
     183           0 :     return 0;
     184             : 
     185          32 :   get_cpuid(0, &max_cpuid_level,
     186             :             (unsigned int *)&vendor_id[0],
     187             :             (unsigned int *)&vendor_id[8],
     188             :             (unsigned int *)&vendor_id[4]);
     189          32 :   vendor_id[12] = 0;
     190             : 
     191             :   if (0)
     192             :     ; /* Just to make "else if" and ifdef macros look pretty.  */
     193             : #ifdef ENABLE_PADLOCK_SUPPORT
     194          32 :   else if (!strcmp (vendor_id, "CentaurHauls"))
     195             :     {
     196             :       /* This is a VIA CPU.  Check what PadLock features we have.  */
     197             : 
     198             :       /* Check for extended centaur (EAX).  */
     199           0 :       get_cpuid(0xC0000000, &features, NULL, NULL, NULL);
     200             : 
     201             :       /* Has extended centaur features? */
     202           0 :       if (features > 0xC0000000)
     203             :         {
     204             :            /* Ask for the extended feature flags (EDX). */
     205           0 :            get_cpuid(0xC0000001, NULL, NULL, NULL, &features);
     206             : 
     207             :            /* Test bits 2 and 3 to see whether the RNG exists and is enabled. */
     208           0 :            if ((features & 0x0C) == 0x0C)
     209           0 :              result |= HWF_PADLOCK_RNG;
     210             : 
     211             :            /* Test bits 6 and 7 to see whether the ACE exists and is enabled. */
     212           0 :            if ((features & 0xC0) == 0xC0)
     213           0 :              result |= HWF_PADLOCK_AES;
     214             : 
     215             :            /* Test bits 10 and 11 to see whether the PHE exists and is
     216             :               enabled.  */
     217           0 :            if ((features & 0xC00) == 0xC00)
     218           0 :              result |= HWF_PADLOCK_SHA;
     219             : 
     220             :            /* Test bits 12 and 13 to see whether the MONTMUL exists and is
     221             :               enabled.  */
     222           0 :            if ((features & 0x3000) == 0x3000)
     223           0 :              result |= HWF_PADLOCK_MMUL;
     224             :         }
     225             :     }
     226             : #endif /*ENABLE_PADLOCK_SUPPORT*/
     227          32 :   else if (!strcmp (vendor_id, "GenuineIntel"))
     228             :     {
     229             :       /* This is an Intel CPU.  */
     230          32 :       result |= HWF_INTEL_CPU;
     231             :     }
     232           0 :   else if (!strcmp (vendor_id, "AuthenticAMD"))
     233             :     {
     234             :       /* This is an AMD CPU.  */
     235             :     }
     236             : 
     237             :   /* Detect Intel features, that might also be supported by other
     238             :      vendors.  */
     239             : 
     240             :   /* Get CPU family/model/stepping (EAX) and Intel feature flags (ECX).  */
     241          32 :   get_cpuid(1, &fms, NULL, &features, NULL);
     242             : 
     243          32 :   family = ((fms & 0xf00) >> 8) + ((fms & 0xff00000) >> 20);
     244          32 :   model = ((fms & 0xf0) >> 4) + ((fms & 0xf0000) >> 12);
     245             : 
     246          32 :   if ((result & HWF_INTEL_CPU) && family == 6)
     247             :     {
     248             :       /* These Intel Core processor models have SHLD/SHRD instruction that
     249             :        * can do integer rotation faster actual ROL/ROR instructions. */
     250          32 :       switch (model)
     251             :         {
     252             :         case 0x2A:
     253             :         case 0x2D:
     254             :         case 0x3A:
     255             :         case 0x3C:
     256             :         case 0x3F:
     257             :         case 0x45:
     258             :         case 0x46:
     259             :         case 0x3D:
     260             :         case 0x4F:
     261             :         case 0x56:
     262             :         case 0x47:
     263             :         case 0x4E:
     264             :         case 0x5E:
     265             :         case 0x55:
     266             :         case 0x66:
     267          32 :           result |= HWF_INTEL_FAST_SHLD;
     268          32 :           break;
     269             :         }
     270             :     }
     271             : 
     272             : #ifdef ENABLE_PCLMUL_SUPPORT
     273             :   /* Test bit 1 for PCLMUL.  */
     274          32 :   if (features & 0x00000002)
     275          32 :      result |= HWF_INTEL_PCLMUL;
     276             : #endif
     277             :   /* Test bit 9 for SSSE3.  */
     278          32 :   if (features & 0x00000200)
     279          32 :      result |= HWF_INTEL_SSSE3;
     280             :   /* Test bit 19 for SSE4.1.  */
     281          32 :   if (features & 0x00080000)
     282          32 :      result |= HWF_INTEL_SSE4_1;
     283             : #ifdef ENABLE_AESNI_SUPPORT
     284             :   /* Test bit 25 for AES-NI.  */
     285          32 :   if (features & 0x02000000)
     286          32 :      result |= HWF_INTEL_AESNI;
     287             : #endif /*ENABLE_AESNI_SUPPORT*/
     288             : #if defined(ENABLE_AVX_SUPPORT) || defined(ENABLE_AVX2_SUPPORT)
     289             :   /* Test bit 27 for OSXSAVE (required for AVX/AVX2).  */
     290          32 :   if (features & 0x08000000)
     291             :     {
     292             :       /* Check that OS has enabled both XMM and YMM state support.  */
     293          32 :       if ((get_xgetbv() & 0x6) == 0x6)
     294          32 :         os_supports_avx_avx2_registers = 1;
     295             :     }
     296             : #endif
     297             : #ifdef ENABLE_AVX_SUPPORT
     298             :   /* Test bit 28 for AVX.  */
     299          32 :   if (features & 0x10000000)
     300          32 :     if (os_supports_avx_avx2_registers)
     301          32 :       result |= HWF_INTEL_AVX;
     302             : #endif /*ENABLE_AVX_SUPPORT*/
     303             : #ifdef ENABLE_DRNG_SUPPORT
     304             :   /* Test bit 30 for RDRAND.  */
     305          32 :   if (features & 0x40000000)
     306           0 :      result |= HWF_INTEL_RDRAND;
     307             : #endif /*ENABLE_DRNG_SUPPORT*/
     308             : 
     309             :   /* Check additional Intel feature flags.  Early Intel P5 processors report
     310             :    * too high max_cpuid_level, so don't check level 7 if processor does not
     311             :    * support SSE3 (as cpuid:7 contains only features for newer processors).
     312             :    * Source: http://www.sandpile.org/x86/cpuid.htm  */
     313          32 :   if (max_cpuid_level >= 7 && (features & 0x00000001))
     314             :     {
     315             :       /* Get CPUID:7 contains further Intel feature flags. */
     316          32 :       get_cpuid(7, NULL, &features, NULL, NULL);
     317             : 
     318             :       /* Test bit 8 for BMI2.  */
     319          32 :       if (features & 0x00000100)
     320           0 :           result |= HWF_INTEL_BMI2;
     321             : 
     322             : #ifdef ENABLE_AVX2_SUPPORT
     323             :       /* Test bit 5 for AVX2.  */
     324          32 :       if (features & 0x00000020)
     325           0 :         if (os_supports_avx_avx2_registers)
     326           0 :           result |= HWF_INTEL_AVX2;
     327             : #endif /*ENABLE_AVX_SUPPORT*/
     328             :     }
     329             : 
     330          32 :   return result;
     331             : }
     332             : #endif /* HAS_X86_CPUID */
     333             : 
     334             : 
     335             : unsigned int
     336          32 : _gcry_hwf_detect_x86 (void)
     337             : {
     338             : #if defined (HAS_X86_CPUID)
     339          32 :   return detect_x86_gnuc ();
     340             : #else
     341             :   return 0;
     342             : #endif
     343             : }

Generated by: LCOV version 1.11