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 /*----- Main code ---------------------------------------------------------*/
  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
  48 struct cpuid { unsigned a, b, c, d; };
  49
  50 /* --- @cpuid@ --- *
  51  *
  52  * Arguments:   @struct cpuid *cc@ = where to write the result
  53  *              @unsigned a, c@ = EAX and ECX registers to set
  54  *
  55  * Returns:     ---
  56  *
  57  * Use:         Minimal C wrapper around the x86 `CPUID' instruction.  Checks
  58  *              that the instruction is actually available before invoking
  59  *              it; fills the output structure with zero if it's not going to
  60  *              work.
  61  */
  62
  63 #ifdef __GNUC__
  64 static __inline__ unsigned getflags(void)
  65   { unsigned f; __asm__ ("pushf; popl %0" : "=g" (f)); return (f); }
  66 static __inline__ unsigned setflags(unsigned f)
  67 {
  68   unsigned ff;
  69   __asm__ ("pushf; pushl %1; popf; pushf; popl %0; popf"
  70            : "=g" (ff)
  71            : "g" (f));
  72   return (ff);
  73 }
  74 #endif
  75
  76 static void cpuid(struct cpuid *cc, unsigned a, unsigned c)
  77 {
  78 #ifdef __GNUC__
  79   unsigned f;
  80 #endif
  81
  82   cc->a = cc->b = cc->c = cc->d = 0;
  83
  84 #ifdef __GNUC__
  85   /* Stupid dance to detect whether the CPUID instruction is available. */
  86   f = getflags();
  87   if (!(setflags(f |  EFLAGS_ID) & EFLAGS_ID)) return;
  88   if (  setflags(f & ~EFLAGS_ID) & EFLAGS_ID ) return;
  89   setflags(f);
  90
  91   /* Alas, EBX is magical in PIC code, so abuse ESI instead.  This isn't
  92    * pretty, but it works.
  93    */
  94   __asm__ ("pushl %%ebx; cpuid; movl %%ebx, %%esi; popl %%ebx"
  95            : "=a" (cc->a), "=S" (cc->b), "=c" (cc->c), "=d" (cc->d)
  96            : "a" (a) , "c" (c));
  97 #endif
  98 }
  99
 100 static unsigned cpuid_maxleaf(void)
 101   { struct cpuid c; cpuid(&c, 0, 0); return (c.a); }
 102
 103 static int cpuid_features_p(unsigned dbits, unsigned cbits)
 104 {
 105   struct cpuid c;
 106   if (cpuid_maxleaf() < 1) return (0);
 107   cpuid(&c, 1, 0);
 108   return ((c.d & dbits) == dbits && (c.c & cbits) == cbits);
 109 }
 110
 111 static int xmm_registers_available_p(void)
 112 {
 113 #ifdef __GNUC__
 114   unsigned f;
 115   /* This hack is by Agner Fog.  Use FXSAVE/FXRSTOR to figure out whether the
 116    * XMM registers are actually alive.
 117    */
 118   if (!cpuid_features_p(CPUID1D_FXSR, 0)) return (0);
 119   __asm__ ("movl %%esp, %%edx; subl $512, %%esp; andl $~15, %%esp\n"
 120            "fxsave (%%esp)\n"
 121            "movl 160(%%esp), %%eax; xorl $0xaaaa5555, 160(%%esp)\n"
 122            "fxrstor (%%esp); fxsave (%%esp)\n"
 123            "movl 160(%%esp), %%ecx; movl %%eax, 160(%%esp)\n"
 124            "fxrstor (%%esp); movl %%edx, %%esp\n"
 125            "xorl %%ecx, %%eax"
 126            : "=a" (f)
 127            : /* no inputs */
 128            : "%ecx", "%edx");
 129   return (f);
 130 #else
 131   return (0);
 132 #endif
 133 }
 134
 135 #endif
 136
 137 /* --- @check_env@ --- *
 138  *
 139  * Arguments:   @const char *ftok@ = feature token
 140  *
 141  * Returns:     Zero if the feature is forced off; positive if it's forced
 142  *              on; negative if the user hasn't decided.
 143  *
 144  * Use:         Checks the environment variable `CATACOMB_CPUFEAT' for the
 145  *              feature token @ftok@.  The variable, if it exists, should be
 146  *              a space-separated sequence of `+tok' and `-tok' items.  These
 147  *              tokens may end in `*', which matches any suffix.
 148  */
 149
 150 static int IGNORABLE check_env(const char *ftok)
 151 {
 152   const char *p, *q, *pp;
 153   int d;
 154
 155   p = getenv("CATACOMB_CPUFEAT");
 156   if (!p) return (-1);
 157
 158   for (;;) {
 159     while (isspace((unsigned char)*p)) p++;
 160     if (!*p) return (-1);
 161     switch (*p) {
 162       case '+': d = +1; p++; break;
 163       case '-': d =  0; p++; break;
 164       default:  d = -1;      break;
 165     }
 166     for (q = p; *q && !isspace((unsigned char)*q); q++);
 167     if (d >= 0) {
 168       for (pp = ftok; p < q && *pp && *p == *pp; p++, pp++);
 169       if ((p == q && !*pp) || (*p == '*' && p + 1 == q)) return (d);
 170     }
 171     p = q;
 172   }
 173   return (-1);
 174 }
 175
 176 /* --- @cpu_feature_p@ --- *
 177  *
 178  * Arguments:   @unsigned feat@ = a @CPUFEAT_...@ code
 179  *
 180  * Returns:     Nonzero if the feature is available.
 181  */
 182
 183 #include <stdio.h>
 184
 185 int cpu_feature_p(int feat)
 186 {
 187   int IGNORABLE f;
 188   IGNORE(f);
 189 #define CHECK_ENV(ftok)                                                 \
 190   do { if ((f = check_env(ftok)) >= 0) return (f); } while (0)
 191
 192   switch (feat) {
 193 #ifdef CPUFAM_X86
 194     case CPUFEAT_X86_SSE2: {
 195       CHECK_ENV("x86:sse2");
 196       return (xmm_registers_available_p() &&
 197               cpuid_features_p(CPUID1D_SSE2, 0));
 198     }
 199 #endif
 200     default:
 201       return (0);
 202   }
 203 #undef CHECK_ENV
 204 }
 205
 206 /*----- That's all, folks -------------------------------------------------*/