mdw@git.distorted.org.uk Git - catacomb/blob - base/dispatch.c

   1 /* -*-c-*-
   2  *
   3  * CPU-specific dispatch
   4  *
   5  * (c) 2015 Straylight/Edgeware
   6  */
   7
   8 /*----- Licensing notice --------------------------------------------------*
   9  *
  10  * This file is part of Catacomb.
  11  *
  12  * Catacomb is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU Library General Public License as
  14  * published by the Free Software Foundation; either version 2 of the
  15  * License, or (at your option) any later version.
  16  *
  17  * Catacomb is distributed in the hope that it will be useful,
  18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20  * GNU Library General Public License for more details.
  21  *
  22  * You should have received a copy of the GNU Library General Public
  23  * License along with Catacomb; if not, write to the Free
  24  * Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
  25  * MA 02111-1307, USA.
  26  */
  27
  28 /*----- Header files ------------------------------------------------------*/
  29
  30 #include "config.h"
  31
  32 #include <ctype.h>
  33 #include <stdlib.h>
  34 #include <string.h>
  35
  36 #include <mLib/macros.h>
  37
  38 #include "dispatch.h"
  39
  40 /*----- Intel x86/AMD64 feature probing -----------------------------------*/
  41
  42 #ifdef CPUFAM_X86
  43
  44 #define EFLAGS_ID (1u << 21)
  45 #define CPUID1D_SSE2 (1u << 26)
  46 #define CPUID1D_FXSR (1u << 24)
  47 #define CPUID1C_AESNI (1u << 25)
  48
  49 struct cpuid { unsigned a, b, c, d; };
  50
  51 /* --- @cpuid@ --- *
  52  *
  53  * Arguments:   @struct cpuid *cc@ = where to write the result
  54  *              @unsigned a, c@ = EAX and ECX registers to set
  55  *
  56  * Returns:     ---
  57  *
  58  * Use:         Minimal C wrapper around the x86 `CPUID' instruction.  Checks
  59  *              that the instruction is actually available before invoking
  60  *              it; fills the output structure with zero if it's not going to
  61  *              work.
  62  */
  63
  64 #ifdef __GNUC__
  65 static __inline__ unsigned getflags(void)
  66   { unsigned f; __asm__ ("pushf; popl %0" : "=g" (f)); return (f); }
  67 static __inline__ unsigned setflags(unsigned f)
  68 {
  69   unsigned ff;
  70   __asm__ ("pushf; pushl %1; popf; pushf; popl %0; popf"
  71            : "=g" (ff)
  72            : "g" (f));
  73   return (ff);
  74 }
  75 #endif
  76
  77 static void cpuid(struct cpuid *cc, unsigned a, unsigned c)
  78 {
  79 #ifdef __GNUC__
  80   unsigned f;
  81 #endif
  82
  83   cc->a = cc->b = cc->c = cc->d = 0;
  84
  85 #ifdef __GNUC__
  86   /* Stupid dance to detect whether the CPUID instruction is available. */
  87   f = getflags();
  88   if (!(setflags(f |  EFLAGS_ID) & EFLAGS_ID)) return;
  89   if (  setflags(f & ~EFLAGS_ID) & EFLAGS_ID ) return;
  90   setflags(f);
  91
  92   /* Alas, EBX is magical in PIC code, so abuse ESI instead.  This isn't
  93    * pretty, but it works.
  94    */
  95   __asm__ ("pushl %%ebx; cpuid; movl %%ebx, %%esi; popl %%ebx"
  96            : "=a" (cc->a), "=S" (cc->b), "=c" (cc->c), "=d" (cc->d)
  97            : "a" (a) , "c" (c));
  98 #endif
  99 }
 100
 101 static unsigned cpuid_maxleaf(void)
 102   { struct cpuid c; cpuid(&c, 0, 0); return (c.a); }
 103
 104 /* --- @cpuid_features_p@ --- *
 105  *
 106  * Arguments:   @unsigned dbits@ = bits to check in EDX
 107  *              @unsigned cbits@ = bits to check in ECX
 108  *
 109  * Returns:     Nonzero if all the requested bits are set in the CPUID result
 110  *              on leaf 1.
 111  */
 112
 113 static int cpuid_features_p(unsigned dbits, unsigned cbits)
 114 {
 115   struct cpuid c;
 116   if (cpuid_maxleaf() < 1) return (0);
 117   cpuid(&c, 1, 0);
 118   return ((c.d & dbits) == dbits && (c.c & cbits) == cbits);
 119 }
 120
 121 /* --- @xmm_registers_available_p@ --- *
 122  *
 123  * Arguments:   ---
 124  *
 125  * Returns:     Nonzero if the operating system has made the XMM registers
 126  *              available for use.
 127  */
 128
 129 static int xmm_registers_available_p(void)
 130 {
 131 #ifdef __GNUC__
 132   unsigned f;
 133   /* This hack is by Agner Fog.  Use FXSAVE/FXRSTOR to figure out whether the
 134    * XMM registers are actually alive.
 135    */
 136   if (!cpuid_features_p(CPUID1D_FXSR, 0)) return (0);
 137   __asm__ ("movl %%esp, %%edx; subl $512, %%esp; andl $~15, %%esp\n"
 138            "fxsave (%%esp)\n"
 139            "movl 160(%%esp), %%eax; xorl $0xaaaa5555, 160(%%esp)\n"
 140            "fxrstor (%%esp); fxsave (%%esp)\n"
 141            "movl 160(%%esp), %%ecx; movl %%eax, 160(%%esp)\n"
 142            "fxrstor (%%esp); movl %%edx, %%esp\n"
 143            "xorl %%ecx, %%eax"
 144            : "=a" (f)
 145            : /* no inputs */
 146            : "%ecx", "%edx");
 147   return (f);
 148 #else
 149   return (0);
 150 #endif
 151 }
 152
 153 #endif
 154
 155 /*----- External interface ------------------------------------------------*/
 156
 157 /* --- @check_env@ --- *
 158  *
 159  * Arguments:   @const char *ftok@ = feature token
 160  *
 161  * Returns:     Zero if the feature is forced off; positive if it's forced
 162  *              on; negative if the user hasn't decided.
 163  *
 164  * Use:         Checks the environment variable `CATACOMB_CPUFEAT' for the
 165  *              feature token @ftok@.  The variable, if it exists, should be
 166  *              a space-separated sequence of `+tok' and `-tok' items.  These
 167  *              tokens may end in `*', which matches any suffix.
 168  */
 169
 170 static int IGNORABLE check_env(const char *ftok)
 171 {
 172   const char *p, *q, *pp;
 173   int d;
 174
 175   p = getenv("CATACOMB_CPUFEAT");
 176   if (!p) return (-1);
 177
 178   for (;;) {
 179     while (isspace((unsigned char)*p)) p++;
 180     if (!*p) return (-1);
 181     switch (*p) {
 182       case '+': d = +1; p++; break;
 183       case '-': d =  0; p++; break;
 184       default:  d = -1;      break;
 185     }
 186     for (q = p; *q && !isspace((unsigned char)*q); q++);
 187     if (d >= 0) {
 188       for (pp = ftok; p < q && *pp && *p == *pp; p++, pp++);
 189       if ((p == q && !*pp) || (*p == '*' && p + 1 == q)) return (d);
 190     }
 191     p = q;
 192   }
 193   return (-1);
 194 }
 195
 196 /* --- @cpu_feature_p@ --- *
 197  *
 198  * Arguments:   @unsigned feat@ = a @CPUFEAT_...@ code
 199  *
 200  * Returns:     Nonzero if the feature is available.
 201  */
 202
 203 #include <stdio.h>
 204
 205 int cpu_feature_p(int feat)
 206 {
 207   int IGNORABLE f;
 208   IGNORE(f);
 209 #define CHECK_ENV(ftok)                                                 \
 210   do { if ((f = check_env(ftok)) >= 0) return (f); } while (0)
 211
 212   switch (feat) {
 213 #ifdef CPUFAM_X86
 214     case CPUFEAT_X86_SSE2: {
 215       CHECK_ENV("x86:sse2");
 216       return (xmm_registers_available_p() &&
 217               cpuid_features_p(CPUID1D_SSE2, 0));
 218     }
 219     case CPUFEAT_X86_AESNI: {
 220       check_env("x86:aesni");
 221       return (xmm_registers_available_p() &&
 222               cpuid_features_p(CPUID1D_SSE2, CPUID1C_AESNI));
 223     }
 224 #endif
 225     default:
 226       return (0);
 227   }
 228 #undef CHECK_ENV
 229 }
 230
 231 /*----- That's all, folks -------------------------------------------------*/