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 <stdarg.h>
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <string.h>
  37
  38 #include <mLib/macros.h>
  39
  40 #include "dispatch.h"
  41
  42 /*----- Intel x86/AMD64 feature probing -----------------------------------*/
  43
  44 #ifdef CPUFAM_X86
  45
  46 #  define EFLAGS_ID (1u << 21)
  47 #  define CPUID1D_SSE2 (1u << 26)
  48 #  define CPUID1D_FXSR (1u << 24)
  49 #  define CPUID1C_AESNI (1u << 25)
  50
  51 struct cpuid { unsigned a, b, c, d; };
  52
  53 /* --- @cpuid@ --- *
  54  *
  55  * Arguments:   @struct cpuid *cc@ = where to write the result
  56  *              @unsigned a, c@ = EAX and ECX registers to set
  57  *
  58  * Returns:     ---
  59  *
  60  * Use:         Minimal C wrapper around the x86 `CPUID' instruction.  Checks
  61  *              that the instruction is actually available before invoking
  62  *              it; fills the output structure with zero if it's not going to
  63  *              work.
  64  */
  65
  66 #ifdef __GNUC__
  67 static __inline__ unsigned getflags(void)
  68   { unsigned f; __asm__ ("pushf; popl %0" : "=g" (f)); return (f); }
  69 static __inline__ unsigned setflags(unsigned f)
  70 {
  71   unsigned ff;
  72   __asm__ ("pushf; pushl %1; popf; pushf; popl %0; popf"
  73            : "=g" (ff)
  74            : "g" (f));
  75   return (ff);
  76 }
  77 #endif
  78
  79 static void cpuid(struct cpuid *cc, unsigned a, unsigned c)
  80 {
  81 #ifdef __GNUC__
  82   unsigned f;
  83 #endif
  84
  85   cc->a = cc->b = cc->c = cc->d = 0;
  86
  87 #ifdef __GNUC__
  88   /* Stupid dance to detect whether the CPUID instruction is available. */
  89   f = getflags();
  90   if (!(setflags(f |  EFLAGS_ID) & EFLAGS_ID) ||
  91         setflags(f & ~EFLAGS_ID) & EFLAGS_ID) {
  92     dispatch_debug("CPUID instruction not available");
  93     return;
  94   }
  95   setflags(f);
  96
  97   /* Alas, EBX is magical in PIC code, so abuse ESI instead.  This isn't
  98    * pretty, but it works.
  99    */
 100   __asm__ ("pushl %%ebx; cpuid; movl %%ebx, %%esi; popl %%ebx"
 101            : "=a" (cc->a), "=S" (cc->b), "=c" (cc->c), "=d" (cc->d)
 102            : "a" (a) , "c" (c));
 103 #else
 104   dispatch_debug("GNU inline assembler not available; can't CPUID");
 105 #endif
 106 }
 107
 108 static unsigned cpuid_maxleaf(void)
 109   { struct cpuid c; cpuid(&c, 0, 0); return (c.a); }
 110
 111 /* --- @cpuid_features_p@ --- *
 112  *
 113  * Arguments:   @unsigned dbits@ = bits to check in EDX
 114  *              @unsigned cbits@ = bits to check in ECX
 115  *
 116  * Returns:     Nonzero if all the requested bits are set in the CPUID result
 117  *              on leaf 1.
 118  */
 119
 120 static int cpuid_features_p(unsigned dbits, unsigned cbits)
 121 {
 122   struct cpuid c;
 123   if (cpuid_maxleaf() < 1) return (0);
 124   cpuid(&c, 1, 0);
 125   return ((c.d & dbits) == dbits && (c.c & cbits) == cbits);
 126 }
 127
 128 /* --- @xmm_registers_available_p@ --- *
 129  *
 130  * Arguments:   ---
 131  *
 132  * Returns:     Nonzero if the operating system has made the XMM registers
 133  *              available for use.
 134  */
 135
 136 static int xmm_registers_available_p(void)
 137 {
 138 #ifdef __GNUC__
 139   unsigned f;
 140   /* This hack is by Agner Fog.  Use FXSAVE/FXRSTOR to figure out whether the
 141    * XMM registers are actually alive.
 142    */
 143   if (!cpuid_features_p(CPUID1D_FXSR, 0)) return (0);
 144   __asm__ ("movl %%esp, %%edx; subl $512, %%esp; andl $~15, %%esp\n"
 145            "fxsave (%%esp)\n"
 146            "movl 160(%%esp), %%eax; xorl $0xaaaa5555, 160(%%esp)\n"
 147            "fxrstor (%%esp); fxsave (%%esp)\n"
 148            "movl 160(%%esp), %%ecx; movl %%eax, 160(%%esp)\n"
 149            "fxrstor (%%esp); movl %%edx, %%esp\n"
 150            "xorl %%ecx, %%eax"
 151            : "=a" (f)
 152            : /* no inputs */
 153            : "%ecx", "%edx");
 154   return (f);
 155 #else
 156   dispatch_debug("GNU inline assembler not available; can't check for XMM");
 157   return (0);
 158 #endif
 159 }
 160
 161 #endif
 162
 163 /*----- External interface ------------------------------------------------*/
 164
 165 /* --- @dispatch_debug@ --- *
 166  *
 167  * Arguments:   @const char *fmt@ = a format string
 168  *              @...@ = additional arguments
 169  *
 170  * Returns:     ---
 171  *
 172  * Use:         Writes a formatted message to standard output if dispatch
 173  *              debugging is enabled.
 174  */
 175
 176 void dispatch_debug(const char *fmt, ...)
 177 {
 178   va_list ap;
 179   const char *e = getenv("CATACOMB_CPUDISPATCH_DEBUG");
 180
 181   if (e && *e != 'n' && *e != '0') {
 182     va_start(ap, fmt);
 183     fputs("Catacomb CPUDISPATCH: ", stderr);
 184     vfprintf(stderr, fmt, ap);
 185     fputc('\n', stderr);
 186     va_end(ap);
 187   }
 188 }
 189
 190 /* --- @check_env@ --- *
 191  *
 192  * Arguments:   @const char *ftok@ = feature token
 193  *
 194  * Returns:     Zero if the feature is forced off; positive if it's forced
 195  *              on; negative if the user hasn't decided.
 196  *
 197  * Use:         Checks the environment variable `CATACOMB_CPUFEAT' for the
 198  *              feature token @ftok@.  The variable, if it exists, should be
 199  *              a space-separated sequence of `+tok' and `-tok' items.  These
 200  *              tokens may end in `*', which matches any suffix.
 201  */
 202
 203 static int IGNORABLE check_env(const char *ftok)
 204 {
 205   const char *p, *q, *pp;
 206   int d;
 207
 208   p = getenv("CATACOMB_CPUFEAT");
 209   if (!p) return (-1);
 210
 211   for (;;) {
 212     while (isspace((unsigned char)*p)) p++;
 213     if (!*p) return (-1);
 214     switch (*p) {
 215       case '+': d = +1; p++; break;
 216       case '-': d =  0; p++; break;
 217       default:  d = -1;      break;
 218     }
 219     for (q = p; *q && !isspace((unsigned char)*q); q++);
 220     if (d >= 0) {
 221       for (pp = ftok; p < q && *pp && *p == *pp; p++, pp++);
 222       if ((p == q && !*pp) || (*p == '*' && p + 1 == q)) return (d);
 223     }
 224     p = q;
 225   }
 226   return (-1);
 227 }
 228
 229 /* --- @cpu_feature_p@ --- *
 230  *
 231  * Arguments:   @unsigned feat@ = a @CPUFEAT_...@ code
 232  *
 233  * Returns:     Nonzero if the feature is available.
 234  */
 235
 236 #include <stdio.h>
 237
 238 static int IGNORABLE
 239   feat_debug(const char *ftok, const char *check, int verdict)
 240 {
 241   if (verdict >= 0) {
 242     dispatch_debug("feature `%s': %s -> %s", ftok, check,
 243                    verdict ? "available" : "absent");
 244   }
 245   return (verdict);
 246 }
 247
 248 int cpu_feature_p(int feat)
 249 {
 250   int IGNORABLE f;
 251   IGNORE(f);
 252 #define CASE_CPUFEAT(feat, ftok, cond) case CPUFEAT_##feat:             \
 253   if ((f = feat_debug(ftok, "environment override",                     \
 254                       check_env(ftok))) >= 0)                           \
 255     return (f);                                                         \
 256   else                                                                  \
 257     return (feat_debug(ftok, "runtime probe", cond));
 258
 259   switch (feat) {
 260 #ifdef CPUFAM_X86
 261     CASE_CPUFEAT(X86_SSE2, "x86:sse2",
 262                  xmm_registers_available_p() &&
 263                  cpuid_features_p(CPUID1D_SSE2, 0));
 264     CASE_CPUFEAT(X86_AESNI, "x86:aesni",
 265                  xmm_registers_available_p() &&
 266                  cpuid_features_p(CPUID1D_SSE2, CPUID1C_AESNI));
 267 #endif
 268     default:
 269       dispatch_debug("denying unknown feature %d", feat);
 270       return (0);
 271   }
 272 #undef CASE_CPUFEAT
 273 }
 274
 275 /*----- That's all, folks -------------------------------------------------*/