sysdeps/posix/sprofil.c

   1 /* Copyright (C) 2001, 2002, 2003 Free Software Foundation, Inc.
   2    Contributed by David Mosberger-Tang <davidm@hpl.hp.com>.
   3    This file is part of the GNU C Library.
   4
   5    The GNU C Library is free software; you can redistribute it and/or
   6    modify it under the terms of the GNU Lesser General Public
   7    License as published by the Free Software Foundation; either
   8    version 2.1 of the License, or (at your option) any later version.
   9
  10    The GNU C Library is distributed in the hope that it will be useful,
  11    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13    Lesser General Public License for more details.
  14
  15    You should have received a copy of the GNU Lesser General Public
  16    License along with the GNU C Library; if not, write to the Free
  17    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  18    02111-1307 USA.  */
  19
  20 #include <assert.h>
  21 #include <signal.h>
  22 #include <stdlib.h>
  23 #include <string.h>
  24 #include <unistd.h>
  25
  26 #include <sys/time.h>
  27 #include <sys/profil.h>
  28
  29 #ifndef SIGPROF
  30 # include <sysdeps/generic/sprofil.c>
  31 #else
  32
  33 #include <libc-internal.h>
  34
  35 struct region
  36   {
  37     size_t offset;
  38     size_t nsamples;
  39     unsigned int scale;
  40     union
  41       {
  42         void *vp;
  43         unsigned short *us;
  44         unsigned int *ui;
  45       }
  46     sample;
  47     size_t start;
  48     size_t end;
  49   };
  50
  51 struct prof_info
  52   {
  53     unsigned int num_regions;
  54     struct region *region;
  55     struct region *last, *overflow;
  56     struct itimerval saved_timer;
  57     struct sigaction saved_action;
  58   };
  59
  60 static unsigned int overflow_counter;
  61
  62 static struct region default_overflow_region =
  63   {
  64     .offset     = 0,
  65     .nsamples   = 1,
  66     .scale      = 2,
  67     .sample     = { &overflow_counter },
  68     .start      = 0,
  69     .end        = ~(size_t) 0
  70   };
  71
  72 static struct prof_info prof_info;
  73
  74 static unsigned long int
  75 pc_to_index (size_t pc, size_t offset, unsigned int scale, int prof_uint)
  76 {
  77   size_t i = (pc - offset) / (prof_uint ? sizeof (int) : sizeof (short));
  78
  79   if (sizeof (unsigned long long int) > sizeof (size_t))
  80     return (unsigned long long int) i * scale / 65536;
  81   else
  82     return i / 65536 * scale + i % 65536 * scale / 65536;
  83 }
  84
  85 static inline size_t
  86 index_to_pc (unsigned long int n, size_t offset, unsigned int scale,
  87              int prof_uint)
  88 {
  89   size_t pc, bin_size = (prof_uint ? sizeof (int) : sizeof (short));
  90
  91   if (sizeof (unsigned long long int) > sizeof (size_t))
  92     pc = offset + (unsigned long long int) n * bin_size * 65536ull / scale;
  93   else
  94     pc = (offset + n * bin_size / scale * 65536
  95           + n * bin_size % scale * 65536 / scale);
  96
  97   if (pc_to_index (pc, offset, scale, prof_uint) < n)
  98     /* Adjust for rounding error.  */
  99     ++pc;
 100
 101   assert (pc_to_index (pc - 1, offset, scale, prof_uint) < n
 102           && pc_to_index (pc, offset, scale, prof_uint) >= n);
 103
 104   return pc;
 105 }
 106
 107 static void
 108 profil_count (void *pcp, int prof_uint)
 109 {
 110   struct region *region, *r = prof_info.last;
 111   size_t lo, hi, mid, pc = (unsigned long int) pcp;
 112   unsigned long int i;
 113
 114   /* Fast path: pc is in same region as before.  */
 115   if (pc >= r->start && pc < r->end)
 116     region = r;
 117   else
 118     {
 119       /* Slow path: do a binary search for the right region.  */
 120       lo = 0; hi = prof_info.num_regions - 1;
 121       while (lo <= hi)
 122         {
 123           mid = (lo + hi) / 2;
 124
 125           r = prof_info.region + mid;
 126           if (pc >= r->start && pc < r->end)
 127             {
 128               prof_info.last = r;
 129               region = r;
 130               break;
 131             }
 132
 133           if (pc < r->start)
 134             hi = mid - 1;
 135           else
 136             lo = mid + 1;
 137         }
 138
 139       /* No matching region: increment overflow count.  There is no point
 140          in updating the cache here, as it won't hit anyhow.  */
 141       region = prof_info.overflow;
 142     }
 143
 144   i = pc_to_index (pc, region->offset, region->scale, prof_uint);
 145   if (i < r->nsamples)
 146     {
 147       if (prof_uint)
 148         {
 149           if (r->sample.ui[i] < (unsigned int) ~0)
 150             ++r->sample.ui[i];
 151         }
 152       else
 153         {
 154           if (r->sample.us[i] < (unsigned short) ~0)
 155             ++r->sample.us[i];
 156         }
 157     }
 158   else
 159     {
 160       if (prof_uint)
 161         ++prof_info.overflow->sample.ui[0];
 162       else
 163         ++prof_info.overflow->sample.us[0];
 164     }
 165 }
 166
 167 static inline void
 168 profil_count_ushort (void *pcp)
 169 {
 170   profil_count (pcp, 0);
 171 }
 172
 173 static inline void
 174 profil_count_uint (void *pcp)
 175 {
 176   profil_count (pcp, 1);
 177 }
 178
 179 /* Get the machine-dependent definition of `profil_counter', the signal
 180    handler for SIGPROF.  It calls `profil_count' (above) with the PC of the
 181    interrupted code.  */
 182 #define profil_counter          profil_counter_ushort
 183 #define profil_count(pc)        profil_count (pc, 0)
 184 #include "profil-counter.h"
 185
 186 #undef profil_counter
 187 #undef profil_count
 188
 189 #define profil_counter          profil_counter_uint
 190 #define profil_count(pc)        profil_count (pc, 1)
 191 #include "profil-counter.h"
 192
 193 static int
 194 insert (int i, unsigned long int start, unsigned long int end, struct prof *p,
 195         int prof_uint)
 196 {
 197   struct region *r;
 198   size_t to_copy;
 199
 200   if (start >= end)
 201     return 0;           /* don't bother with empty regions */
 202
 203   if (prof_info.num_regions == 0)
 204     r = malloc (sizeof (*r));
 205   else
 206     r = realloc (prof_info.region, (prof_info.num_regions + 1) * sizeof (*r));
 207   if (r == NULL)
 208     return -1;
 209
 210   to_copy = prof_info.num_regions - i;
 211   if (to_copy > 0)
 212     memmove (r + i + 1, r + i, to_copy * sizeof (*r));
 213
 214   r[i].offset = p->pr_off;
 215   r[i].nsamples = p->pr_size / (prof_uint ? sizeof (int) : sizeof (short));
 216   r[i].scale = p->pr_scale;
 217   r[i].sample.vp = p->pr_base;
 218   r[i].start = start;
 219   r[i].end = end;
 220
 221   prof_info.region = r;
 222   ++prof_info.num_regions;
 223
 224   if (p->pr_off == 0 && p->pr_scale == 2)
 225     prof_info.overflow = r;
 226
 227   return 0;
 228 }
 229
 230 /* Add a new profiling region.  If the new region overlaps with
 231    existing ones, this may add multiple subregions so that the final
 232    data structure is free of overlaps.  The absence of overlaps makes
 233    it possible to use a binary search in profil_count().  Note that
 234    this function depends on new regions being presented in DECREASING
 235    ORDER of starting address.  */
 236
 237 static int
 238 add_region (struct prof *p, int prof_uint)
 239 {
 240   unsigned long int nsamples;
 241   size_t start, end;
 242   unsigned int i;
 243
 244   if (p->pr_scale < 2)
 245     return 0;
 246
 247   nsamples = p->pr_size / (prof_uint ? sizeof (int) : sizeof (short));
 248
 249   start = p->pr_off;
 250   end = index_to_pc (nsamples, p->pr_off, p->pr_scale, prof_uint);
 251
 252   /* Merge with existing regions.  */
 253   for (i = 0; i < prof_info.num_regions; ++i)
 254     {
 255       if (start < prof_info.region[i].start)
 256         {
 257           if (end < prof_info.region[i].start)
 258             break;
 259           else if (insert (i, start, prof_info.region[i].start, p, prof_uint)
 260                    < 0)
 261             return -1;
 262         }
 263       start = prof_info.region[i].end;
 264     }
 265   return insert (i, start, end, p, prof_uint);
 266 }
 267
 268 static int
 269 pcmp (const void *left, const void *right)
 270 {
 271   struct prof *l = *(struct prof **) left;
 272   struct prof *r = *(struct prof **) right;
 273
 274   if (l->pr_off < r->pr_off)
 275     return 1;
 276   else if (l->pr_off > r->pr_off)
 277     return -1;
 278   return 0;
 279 }
 280
 281 int
 282 __sprofil (struct prof *profp, int profcnt, struct timeval *tvp,
 283            unsigned int flags)
 284 {
 285   struct prof *p[profcnt];
 286   struct itimerval timer;
 287   struct sigaction act;
 288   int i;
 289
 290   if (tvp != NULL)
 291     {
 292       /* Return profiling period.  */
 293       unsigned long int t = 1000000 / __profile_frequency ();
 294       tvp->tv_sec  = t / 1000000;
 295       tvp->tv_usec = t % 1000000;
 296     }
 297
 298   if (prof_info.num_regions > 0)
 299     {
 300       /* Disable profiling.  */
 301       if (__setitimer (ITIMER_PROF, &prof_info.saved_timer, NULL) < 0)
 302         return -1;
 303
 304       if (__sigaction (SIGPROF, &prof_info.saved_action, NULL) < 0)
 305         return -1;
 306
 307       free (prof_info.region);
 308       return 0;
 309     }
 310
 311   prof_info.num_regions = 0;
 312   prof_info.region = NULL;
 313   prof_info.overflow = &default_overflow_region;
 314
 315   for (i = 0; i < profcnt; ++i)
 316     p[i] = profp + i;
 317
 318   /* Sort in order of decreasing starting address: */
 319   qsort (p, profcnt, sizeof (p[0]), pcmp);
 320
 321   /* Add regions in order of decreasing starting address: */
 322   for (i = 0; i < profcnt; ++i)
 323     if (add_region (p[i], (flags & PROF_UINT) != 0) < 0)
 324       {
 325         if (prof_info.region)
 326           free (prof_info.region);
 327         prof_info.num_regions = 0;
 328         prof_info.region = NULL;
 329         return -1;
 330       }
 331
 332   if (prof_info.num_regions == 0)
 333     return 0;
 334
 335   prof_info.last = prof_info.region;
 336
 337   /* Install SIGPROF handler.  */
 338   if (flags & PROF_UINT)
 339     act.sa_handler = (sighandler_t) &profil_counter_uint;
 340   else
 341     act.sa_handler = (sighandler_t) &profil_counter_ushort;
 342   act.sa_flags = SA_RESTART;
 343   __sigfillset (&act.sa_mask);
 344   if (__sigaction (SIGPROF, &act, &prof_info.saved_action) < 0)
 345     return -1;
 346
 347   /* Setup profiling timer.  */
 348   timer.it_value.tv_sec  = 0;
 349   timer.it_value.tv_usec = 1;
 350   timer.it_interval = timer.it_value;
 351   return __setitimer (ITIMER_PROF, &timer, &prof_info.saved_timer);
 352 }
 353
 354 weak_alias (__sprofil, sprofil)
 355
 356 #endif /* SIGPROF */