sysdeps/mach/hurd/profil.c

   1 /* Low-level statistical profiling support function.  Mach/Hurd version.
   2    Copyright (C) 1995, 1996, 1997, 2000, 2002, 2005 Free Software Foundation, Inc.
   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 <sys/types.h>
  21 #include <unistd.h>
  22 #include <errno.h>
  23 #include <hurd.h>
  24 #include <mach/mach4.h>
  25 #include <mach/pc_sample.h>
  26 #include <cthreads.h>
  27 #include <assert.h>
  28 #include <libc-internal.h>
  29
  30
  31 #define MAX_PC_SAMPLES  512     /* XXX ought to be exported in kernel hdr */
  32
  33 static thread_t profile_thread = MACH_PORT_NULL;
  34 static u_short *samples;
  35 static size_t maxsamples;
  36 static size_t pc_offset;
  37 static size_t sample_scale;
  38 static sampled_pc_seqno_t seqno;
  39 static spin_lock_t lock = SPIN_LOCK_INITIALIZER;
  40 static mach_msg_timeout_t collector_timeout; /* ms between collections.  */
  41 static int profile_tick;
  42
  43 /* Reply port used by profiler thread */
  44 static mach_port_t profil_reply_port;
  45
  46 /* Forwards */
  47 static kern_return_t profil_task_get_sampled_pcs (mach_port_t,
  48                                                   sampled_pc_seqno_t *,
  49                                                   sampled_pc_array_t,
  50                                                   mach_msg_type_number_t *);
  51 static void fetch_samples (void);
  52 static void profile_waiter (void);
  53
  54 /* Enable statistical profiling, writing samples of the PC into at most
  55    SIZE bytes of SAMPLE_BUFFER; every processor clock tick while profiling
  56    is enabled, the system examines the user PC and increments
  57    SAMPLE_BUFFER[((PC - OFFSET) / 2) * SCALE / 65536].  If SCALE is zero,
  58    disable profiling.  Returns zero on success, -1 on error.  */
  59
  60 static error_t
  61 update_waiter (u_short *sample_buffer, size_t size, size_t offset, u_int scale)
  62 {
  63   error_t err;
  64
  65   if (profile_thread == MACH_PORT_NULL)
  66     {
  67       /* Set up the profiling collector thread.  */
  68       err = __thread_create (__mach_task_self (), &profile_thread);
  69       if (! err)
  70         err = __mach_setup_thread (__mach_task_self (), profile_thread,
  71                                    &profile_waiter, NULL, NULL);
  72     }
  73   else
  74     err = 0;
  75
  76   if (! err)
  77     {
  78       err = __task_enable_pc_sampling (__mach_task_self (), &profile_tick,
  79                                        SAMPLED_PC_PERIODIC);
  80       if (!err && sample_scale == 0)
  81         /* Profiling was not turned on, so the collector thread was
  82            suspended.  Resume it.  */
  83         err = __thread_resume (profile_thread);
  84       if (! err)
  85         {
  86           samples = sample_buffer;
  87           maxsamples = size / sizeof *sample_buffer;
  88           pc_offset = offset;
  89           sample_scale = scale;
  90           /* Calculate a good period for the collector thread.  From TICK
  91              and the kernel buffer size we get the length of time it takes
  92              to fill the buffer; translate that to milliseconds for
  93              mach_msg, and chop it in half for general lag factor.  */
  94           collector_timeout = MAX_PC_SAMPLES * profile_tick / 1000 / 2;
  95         }
  96     }
  97
  98   return err;
  99 }
 100
 101 int
 102 __profile_frequency (void)
 103 {
 104   return profile_tick;
 105 }
 106 libc_hidden_def (__profile_frequency)
 107
 108 int
 109 __profil (u_short *sample_buffer, size_t size, size_t offset, u_int scale)
 110 {
 111   error_t err;
 112
 113   __spin_lock (&lock);
 114
 115   if (scale == 0)
 116     {
 117       /* Disable profiling.  */
 118       int count;
 119
 120       if (profile_thread != MACH_PORT_NULL)
 121         __thread_suspend (profile_thread);
 122
 123       /* Fetch the last set of samples */
 124       if (sample_scale)
 125         fetch_samples ();
 126
 127       err = __task_disable_pc_sampling (__mach_task_self (), &count);
 128       sample_scale = 0;
 129       seqno = 0;
 130     }
 131   else
 132     err = update_waiter (sample_buffer, size, offset, scale);
 133
 134   __spin_unlock (&lock);
 135
 136   return err ? __hurd_fail (err) : 0;
 137 }
 138 weak_alias (__profil, profil)
 139
 140 /* Fetch PC samples.  This function must be very careful not to depend
 141    on Hurd threadvar variables.  We arrange that by using a special
 142    stub arranged for at the end of this file. */
 143 static void
 144 fetch_samples (void)
 145 {
 146   sampled_pc_t pc_samples[MAX_PC_SAMPLES];
 147   mach_msg_type_number_t nsamples, i;
 148   error_t err;
 149
 150   nsamples = MAX_PC_SAMPLES;
 151
 152   err = profil_task_get_sampled_pcs (__mach_task_self (), &seqno,
 153                                      pc_samples, &nsamples);
 154   if (err)
 155     {
 156       static error_t special_profil_failure;
 157       static volatile int a, b, c;
 158
 159       special_profil_failure = err;
 160       a = 1;
 161       b = 0;
 162       while (1)
 163         c = a / b;
 164     }
 165
 166   for (i = 0; i < nsamples; ++i)
 167     {
 168       /* Do arithmetic in long long to avoid overflow problems. */
 169       long long pc_difference = pc_samples[i].pc - pc_offset;
 170       size_t idx = ((pc_difference / 2) * sample_scale) / 65536;
 171       if (idx < maxsamples)
 172         ++samples[idx];
 173     }
 174 }
 175
 176
 177 /* This function must be very careful not to depend on Hurd threadvar
 178    variables.  We arrange that by using special stubs arranged for at the
 179    end of this file. */
 180 static void
 181 profile_waiter (void)
 182 {
 183   mach_msg_header_t msg;
 184   mach_port_t timeout_reply_port;
 185
 186   profil_reply_port = __mach_reply_port ();
 187   timeout_reply_port = __mach_reply_port ();
 188
 189   while (1)
 190     {
 191       __spin_lock (&lock);
 192
 193       fetch_samples ();
 194
 195       __spin_unlock (&lock);
 196
 197       __mach_msg (&msg, MACH_RCV_MSG|MACH_RCV_TIMEOUT, 0, sizeof msg,
 198                   timeout_reply_port, collector_timeout, MACH_PORT_NULL);
 199     }
 200 }
 201 \f
 202 /* Fork interaction */
 203
 204 /* Before fork, lock the interlock so that we are in a clean state. */
 205 static void
 206 fork_profil_prepare (void)
 207 {
 208   __spin_lock (&lock);
 209 }
 210 text_set_element (_hurd_fork_prepare_hook, fork_profil_prepare);
 211
 212 /* In the parent, unlock the interlock once fork is complete. */
 213 static void
 214 fork_profil_parent (void)
 215 {
 216   __spin_unlock (&lock);
 217 }
 218 text_set_element (_hurd_fork_parent_hook, fork_profil_parent);
 219
 220 /* In the childs, unlock the interlock, and start a profiling thread up
 221    if necessary. */
 222 static void
 223 fork_profil_child (void)
 224 {
 225   u_short *sb;
 226   size_t n, o, ss;
 227   error_t err;
 228
 229   __spin_unlock (&lock);
 230
 231   if (profile_thread != MACH_PORT_NULL)
 232     {
 233       __mach_port_deallocate (__mach_task_self (), profile_thread);
 234       profile_thread = MACH_PORT_NULL;
 235     }
 236
 237   sb = samples;
 238   samples = NULL;
 239   n = maxsamples;
 240   maxsamples = 0;
 241   o = pc_offset;
 242   pc_offset = 0;
 243   ss = sample_scale;
 244   sample_scale = 0;
 245
 246   if (ss != 0)
 247     {
 248       err = update_waiter (sb, n * sizeof *sb, o, ss);
 249       assert_perror (err);
 250     }
 251 }
 252 text_set_element (_hurd_fork_child_hook, fork_profil_child);
 253
 254
 255 \f
 256
 257 /* Special RPC stubs for profile_waiter are made by including the normal
 258    source code, with special CPP state to prevent it from doing the
 259    usual thing. */
 260
 261 /* Include these first; then our #define's will take full effect, not
 262    being overridden. */
 263 #include <mach/mig_support.h>
 264
 265 /* This need not do anything; it is always associated with errors, which
 266    are fatal in profile_waiter anyhow. */
 267 #define __mig_put_reply_port(foo)
 268
 269 /* Use our static variable instead of the usual threadvar mechanism for
 270    this. */
 271 #define __mig_get_reply_port() profil_reply_port
 272
 273 /* Make the functions show up as static */
 274 #define mig_external static
 275
 276 /* Turn off the attempt to generate ld aliasing records. */
 277 #undef weak_alias
 278 #define weak_alias(a,b)
 279
 280 /* And change their names to avoid confusing disasters. */
 281 #define __vm_deallocate_rpc profil_vm_deallocate
 282 #define __task_get_sampled_pcs profil_task_get_sampled_pcs
 283
 284 /* And include the source code */
 285 #include <../mach/RPC_task_get_sampled_pcs.c>