gprofng/libcollector/hwprofile.c

   1 /* Copyright (C) 2021 Free Software Foundation, Inc.
   2    Contributed by Oracle.
   3
   4    This file is part of GNU Binutils.
   5
   6    This program is free software; you can redistribute it and/or modify
   7    it under the terms of the GNU General Public License as published by
   8    the Free Software Foundation; either version 3, or (at your option)
   9    any later version.
  10
  11    This program is distributed in the hope that it will be useful,
  12    but WITHOUT ANY WARRANTY; without even the implied warranty of
  13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14    GNU General Public License for more details.
  15
  16    You should have received a copy of the GNU General Public License
  17    along with this program; if not, write to the Free Software
  18    Foundation, 51 Franklin Street - Fifth Floor, Boston,
  19    MA 02110-1301, USA.  */
  20
  21 /* Hardware counter profiling */
  22
  23 #include "config.h"
  24 #include <alloca.h>
  25 #include <dlfcn.h>
  26 #include <stdlib.h>
  27 #include <stdio.h>
  28 #include <unistd.h>
  29 #include <errno.h>
  30 #include <sys/syscall.h>
  31 #include <signal.h>
  32 #include <ucontext.h>
  33
  34 #include "gp-defs.h"
  35 #define _STRING_H 1  /* XXX MEZ: temporary workaround */
  36 #include "hwcdrv.h"
  37 #include "collector_module.h"
  38 #include "gp-experiment.h"
  39 #include "libcol_util.h"
  40 #include "hwprofile.h"
  41 #include "ABS.h"
  42 #include "tsd.h"
  43
  44 /* TprintfT(<level>,...) definitions.  Adjust per module as needed */
  45 #define DBG_LT0 0 // for high-level configuration, unexpected errors/warnings
  46 #define DBG_LT1 1 // for configuration details, warnings
  47 #define DBG_LT2 2
  48 #define DBG_LT3 3
  49 #define DBG_LT4 4
  50 #define DBG_LT5 5
  51
  52 #define  SD_OFF 0       /* before start or after close she shut down process */
  53 #define  SD_PENDING 1   /* before running real_detach_experiment() */
  54 #define  SD_COMPLETE 2  /* after running real_detach_experiment() */
  55
  56 static int init_interface (CollectorInterface*);
  57 static int open_experiment (const char *);
  58 static int start_data_collection (void);
  59 static int stop_data_collection (void);
  60 static int close_experiment (void);
  61 static int detach_experiment (void);
  62 static int real_detach_experiment (void);
  63
  64 static ModuleInterface module_interface ={
  65   SP_HWCNTR_FILE,           /* description */
  66   init_interface,           /* initInterface */
  67   open_experiment,          /* openExperiment */
  68   start_data_collection,    /* startDataCollection */
  69   stop_data_collection,     /* stopDataCollection */
  70   close_experiment,         /* closeExperiment */
  71   detach_experiment         /* detachExperiment (fork child) */
  72 };
  73
  74 static CollectorInterface *collector_interface = NULL;
  75
  76
  77 /*---------------------------------------------------------------------------*/
  78 /* compile options and workarounds */
  79
  80 /* Solaris: We set ITIMER_REALPROF to ensure that counters get started on
  81  *      LWPs that existed before the collector initialization.
  82  *
  83  * In addition, if the appropriate #define's are set, we check for:
  84  *      lost-hw-overflow -- the HW counters rollover, but the overflow
  85  *            interrupt is not generated (counters keep running)
  86  *      lost-sigemt -- the interrupt is received by the kernel,
  87  *            which stops the counters, but the kernel fails
  88  *            to deliver the signal.
  89  */
  90
  91 /*---------------------------------------------------------------------------*/
  92 /* typedefs */
  93
  94 typedef enum {
  95   HWCMODE_OFF,       /* before start or after close */
  96   HWCMODE_SUSPEND,  /* stop_data_collection called */
  97   HWCMODE_ACTIVE,   /* counters are defined and after start_data_collection() */
  98   HWCMODE_ABORT     /* fatal error occured. Log a message, stop recording */
  99 } hwc_mode_t;
 100
 101 /*---------------------------------------------------------------------------*/
 102 /* prototypes */
 103 static void init_ucontexts (void);
 104 static int hwc_initialize_handlers (void);
 105 static void collector_record_counter (ucontext_t*,
 106                                       int timecvt,
 107                                       ABST_type, hrtime_t,
 108                                       unsigned, uint64_t);
 109 static void collector_hwc_ABORT (int errnum, const char *msg);
 110 static void hwclogwrite0 ();
 111 static void hwclogwrite (Hwcentry *);
 112 static void set_hwc_mode (hwc_mode_t);
 113 static void collector_sigemt_handler (int sig, siginfo_t *si, void *puc);
 114
 115 /*---------------------------------------------------------------------------*/
 116 /* static variables */
 117
 118 /* --- user counter selections and options */
 119 static int hwcdef_has_memspace;     /* true to indicate use of extened packets */
 120 static unsigned hwcdef_cnt;         /* number of *active* hardware counters */
 121 static unsigned hwcdef_num_sampling_ctrdefs; /* ctrs that use sampling */
 122 static unsigned hwcdef_num_overflow_ctrdefs; /* ctrs that use overflow */
 123 static Hwcentry **hwcdef;           /* HWC definitions */
 124 static int cpcN_cpuver = CPUVER_UNDEFINED;
 125 static int hwcdrv_inited;           /* Don't call hwcdrv_init() in fork_child */
 126 static hwcdrv_api_t *hwc_driver = NULL;
 127 static unsigned hwprofile_tsd_key = COLLECTOR_TSD_INVALID_KEY;
 128 static int hwprofile_tsd_sz = 0;
 129 static volatile hwc_mode_t hwc_mode = HWCMODE_OFF;
 130 static volatile unsigned int nthreads_in_sighandler = 0;
 131 static volatile unsigned int sd_state = SD_OFF;
 132
 133 /* --- experiment logging state */
 134 static CollectorModule expr_hndl = COLLECTOR_MODULE_ERR;
 135 static ucontext_t expr_dummy_uc;        // used for hacked "collector" frames
 136 static ucontext_t expr_out_of_range_uc; // used for "out-of-range" frames
 137 static ucontext_t expr_frozen_uc;       // used for "frozen" frames
 138 static ucontext_t expr_nopc_uc;         // used for not-program-related frames
 139 static ucontext_t expr_lostcounts_uc;   // used for lost_counts frames
 140
 141 /* --- signal handler state */
 142 static struct sigaction old_sigemt_handler;  //overwritten in fork-child
 143
 144 /*---------------------------------------------------------------------------*/
 145 /* macros */
 146 #define COUNTERS_ENABLED()  (hwcdef_cnt)
 147 #define gethrtime           collector_interface->getHiResTime
 148
 149 #ifdef DEBUG
 150 #define Tprintf(...)   if (collector_interface) collector_interface->writeDebugInfo( 0, __VA_ARGS__ )
 151 #define TprintfT(...)  if (collector_interface) collector_interface->writeDebugInfo( 1, __VA_ARGS__ )
 152 #else
 153 #define Tprintf(...)
 154 #define TprintfT(...)
 155 #endif
 156
 157
 158 /*---------------------------------------------------------------------------*/
 159
 160 /* Initialization routines */
 161 static hwcdrv_api_t *
 162 get_hwc_driver ()
 163 {
 164   if (hwc_driver == NULL)
 165     hwc_driver = __collector_get_hwcdrv ();
 166   return hwc_driver;
 167 }
 168
 169 static void init_module () __attribute__ ((constructor));
 170 static void
 171 init_module ()
 172 {
 173   __collector_dlsym_guard = 1;
 174   RegModuleFunc reg_module = (RegModuleFunc) dlsym (RTLD_DEFAULT, "__collector_register_module");
 175   __collector_dlsym_guard = 0;
 176   if (reg_module == NULL)
 177     {
 178       TprintfT (0, "hwprofile: init_module FAILED - reg_module = NULL\n");
 179       return;
 180     }
 181   expr_hndl = reg_module (&module_interface);
 182   if (expr_hndl == COLLECTOR_MODULE_ERR)
 183     {
 184       TprintfT (0, "hwprofile: ERROR: handle not created.\n");
 185       if (collector_interface)
 186         collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">data handle not created</event>\n",
 187                                        SP_JCMD_CERROR, COL_ERROR_HWCINIT);
 188     }
 189 }
 190
 191 static int
 192 init_interface (CollectorInterface *_collector_interface)
 193 {
 194   collector_interface = _collector_interface;
 195   return COL_ERROR_NONE;
 196 }
 197
 198 static void *
 199 hwprofile_get_tsd ()
 200 {
 201   return collector_interface->getKey (hwprofile_tsd_key);
 202 }
 203
 204 static int
 205 open_experiment (const char *exp)
 206 {
 207   if (collector_interface == NULL)
 208     {
 209       TprintfT (0, "hwprofile: ERROR: collector_interface is null.\n");
 210       return COL_ERROR_HWCINIT;
 211     }
 212   const char *params = collector_interface->getParams ();
 213   while (params)
 214     {
 215       if (__collector_strStartWith (params, "h:*") == 0)
 216         {
 217           /* HWC counters set by default */
 218           collector_interface->writeLog ("<%s %s=\"1\"/>\n",
 219                                          SP_TAG_SETTING, SP_JCMD_HWC_DEFAULT);
 220           params += 3;
 221           break;
 222         }
 223       else if (__collector_strStartWith (params, "h:") == 0)
 224         {
 225           params += 2;
 226           break;
 227         }
 228       params = CALL_UTIL (strchr)(params, ';');
 229       if (params)
 230         params++;
 231     }
 232   if (params == NULL)  /* HWC profiling not specified */
 233     return COL_ERROR_HWCINIT;
 234   char *s = CALL_UTIL (strchr)(params, (int) ';');
 235   int sz = s ? s - params : CALL_UTIL (strlen)(params);
 236   char *defstring = (char*) alloca (sz + 1);
 237   CALL_UTIL (strlcpy)(defstring, params, sz + 1);
 238   TprintfT (0, "hwprofile: open_experiment %s -- %s\n", exp, defstring);
 239
 240   int err = COL_ERROR_NONE;
 241   /* init counter library */
 242   if (!hwcdrv_inited)
 243     { /* do not call hwcdrv_init() from fork-child */
 244       hwcdrv_inited = 1;
 245       get_hwc_driver ();
 246       if (hwc_driver->hwcdrv_init (collector_hwc_ABORT, &hwprofile_tsd_sz) == 0)
 247         {
 248           collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s</event>\n",
 249                                          SP_JCMD_CERROR, COL_ERROR_HWCINIT, defstring);
 250           TprintfT (0, "hwprofile: ERROR: hwcfuncs_init() failed\n");
 251           return COL_ERROR_HWCINIT;
 252         }
 253
 254       if (hwc_driver->hwcdrv_enable_mt (hwprofile_get_tsd))
 255         {
 256           // It is OK to call hwcdrv_enable_mt() before tsd key is created
 257           collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s</event>\n",
 258                                          SP_JCMD_CERROR, COL_ERROR_HWCINIT, defstring);
 259           TprintfT (0, "hwprofile: ERROR: hwcdrv_enable_mt() failed\n");
 260           return COL_ERROR_HWCINIT;
 261         }
 262
 263       hwc_driver->hwcdrv_get_info (&cpcN_cpuver, NULL, NULL, NULL, NULL);
 264       if (cpcN_cpuver < 0)
 265         {
 266           collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s</event>\n",
 267                                          SP_JCMD_CERROR, COL_ERROR_HWCINIT, defstring);
 268           TprintfT (0, "hwprofile: ERROR: hwcdrv_get_info() failed\n");
 269           return COL_ERROR_HWCINIT;
 270         }
 271     }
 272
 273   if (hwprofile_tsd_sz)
 274     {
 275       hwprofile_tsd_key = collector_interface->createKey (hwprofile_tsd_sz, NULL, NULL);
 276       if (hwprofile_tsd_key == COLLECTOR_TSD_INVALID_KEY)
 277         {
 278           collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s</event>\n",
 279                                          SP_JCMD_CERROR, COL_ERROR_HWCINIT, defstring);
 280           TprintfT (0, "hwprofile: ERROR: TSD createKey failed\n");
 281           return COL_ERROR_HWCINIT;
 282         }
 283     }
 284   hwcdef_cnt = 0;
 285   hwcdef_has_memspace = 0;
 286
 287   /* create counters based on hwcdef[] */
 288   err = __collector_hwcfuncs_bind_descriptor (defstring);
 289   if (err)
 290     {
 291       err = err == HWCFUNCS_ERROR_HWCINIT ? COL_ERROR_HWCINIT : COL_ERROR_HWCARGS;
 292       collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s</event>\n",
 293                                      SP_JCMD_CERROR, err, defstring);
 294       TprintfT (0, "hwprofile: ERROR: open_experiment() failed, RC=%d \n", err);
 295       return err;
 296     }
 297
 298   /* generate an array of counter structures for each requested counter */
 299   hwcdef = __collector_hwcfuncs_get_ctrs (&hwcdef_cnt);
 300   hwcdef_num_sampling_ctrdefs = hwcdef_num_overflow_ctrdefs = 0;
 301   int idx;
 302   for (idx = 0; idx < hwcdef_cnt; idx++)
 303     {
 304       if (HWCENTRY_USES_SAMPLING (hwcdef[idx]))
 305         {
 306           hwcdef_num_sampling_ctrdefs++;
 307         }
 308       else
 309         {
 310           hwcdef_num_overflow_ctrdefs++;
 311         }
 312     }
 313
 314   init_ucontexts ();
 315
 316   /* initialize the SIGEMT handler, and the periodic HWC checker */
 317   err = hwc_initialize_handlers ();
 318   if (err != COL_ERROR_NONE)
 319     {
 320       hwcdef_cnt = 0;
 321       TprintfT (0, "hwprofile: ERROR: open_experiment() failed, RC=%d \n", err);
 322       /* log written by hwc_initialize_handlers() */
 323       return err;
 324     }
 325
 326   for (idx = 0; idx < hwcdef_cnt; idx++)
 327     if (ABST_BACKTRACK_ENABLED (hwcdef[idx]->memop))
 328       hwcdef_has_memspace = 1;
 329
 330   /* record the hwc definitions in the log, based on the counter array */
 331   hwclogwrite0 ();
 332   for (idx = 0; idx < hwcdef_cnt; idx++)
 333     hwclogwrite (hwcdef[idx]);
 334   return COL_ERROR_NONE;
 335 }
 336
 337 int
 338 __collector_ext_hwc_lwp_init ()
 339 {
 340   return get_hwc_driver ()->hwcdrv_lwp_init ();
 341 }
 342
 343 void
 344 __collector_ext_hwc_lwp_fini ()
 345 {
 346   get_hwc_driver ()->hwcdrv_lwp_fini ();
 347 }
 348
 349 int
 350 __collector_ext_hwc_lwp_suspend ()
 351 {
 352   return get_hwc_driver ()->hwcdrv_lwp_suspend ();
 353 }
 354
 355 int
 356 __collector_ext_hwc_lwp_resume ()
 357 {
 358   return get_hwc_driver ()->hwcdrv_lwp_resume ();
 359 }
 360
 361 /* Dummy routine, used to provide a context for non-program related profiles */
 362 void
 363 __collector_not_program_related () { }
 364
 365 /* Dummy routine, used to provide a context for lost counts (perf_events) */
 366 void
 367 __collector_hwc_samples_lost () { }
 368
 369 /* Dummy routine, used to provide a context */
 370 void
 371 __collector_hwcs_frozen () { }
 372
 373 /* Dummy routine, used to provide a context */
 374 void
 375 __collector_hwcs_out_of_range () { }
 376 /* initialize some structures */
 377 static void
 378 init_ucontexts (void)
 379 {
 380   /* initialize dummy context for "collector" frames */
 381   getcontext (&expr_dummy_uc);
 382   SETFUNCTIONCONTEXT (&expr_dummy_uc, NULL);
 383
 384   /* initialize dummy context for "out-of-range" frames */
 385   getcontext (&expr_out_of_range_uc);
 386   SETFUNCTIONCONTEXT (&expr_out_of_range_uc, &__collector_hwcs_out_of_range);
 387
 388   /* initialize dummy context for "frozen" frames */
 389   getcontext (&expr_frozen_uc);
 390   SETFUNCTIONCONTEXT (&expr_frozen_uc, &__collector_hwcs_frozen);
 391
 392   /* initialize dummy context for non-program-related frames */
 393   getcontext (&expr_nopc_uc);
 394   SETFUNCTIONCONTEXT (&expr_nopc_uc, &__collector_not_program_related);
 395
 396   /* initialize dummy context for lost-counts-related frames */
 397   getcontext (&expr_lostcounts_uc);
 398   SETFUNCTIONCONTEXT (&expr_lostcounts_uc, &__collector_hwc_samples_lost);
 399 }
 400 /* initialize the signal handler */
 401 static int
 402 hwc_initialize_handlers (void)
 403 {
 404   /* install the signal handler for SIGEMT */
 405   struct sigaction oact;
 406   if (__collector_sigaction (HWCFUNCS_SIGNAL, NULL, &oact) != 0)
 407     {
 408       TprintfT (0, "hwc_initialize_handlers(): ERROR: hwc_initialize_handlers(): __collector_sigaction() failed to get oact\n");
 409       collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">old handler could not be determined</event>\n", SP_JCMD_CERROR, COL_ERROR_HWCINIT);
 410       return COL_ERROR_HWCINIT;
 411     }
 412   if (oact.sa_sigaction == collector_sigemt_handler)
 413     {
 414       /* signal handler is already in place; we are probably in a fork-child */
 415       TprintfT (DBG_LT1, "hwc_initialize_handlers(): hwc_initialize_handlers() collector_sigemt_handler already installed\n");
 416     }
 417   else
 418     {
 419       /* set our signal handler */
 420       struct sigaction c_act;
 421       CALL_UTIL (memset)(&c_act, 0, sizeof c_act);
 422       sigemptyset (&c_act.sa_mask);
 423       sigaddset (&c_act.sa_mask, SIGPROF); /* block SIGPROF delivery in handler */
 424       /* XXXX should probably also block sample_sig & pause_sig */
 425       c_act.sa_sigaction = collector_sigemt_handler;  /* note: used to set sa_handler instead */
 426       c_act.sa_flags = SA_RESTART | SA_SIGINFO;
 427       if (__collector_sigaction (HWCFUNCS_SIGNAL, &c_act, &old_sigemt_handler) != 0)
 428         {
 429           TprintfT (0, "hwc_initialize_handlers(): ERROR: hwc_initialize_handlers(): __collector_sigaction() failed to set cact\n");
 430           collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">event handler could not be installed</event>\n", SP_JCMD_CERROR, COL_ERROR_HWCINIT);
 431           return COL_ERROR_HWCINIT;
 432         }
 433     }
 434   return COL_ERROR_NONE;
 435 }
 436
 437 static int
 438 close_experiment (void)
 439 {
 440   /* note: stop_data_collection() should have already been called by
 441    * collector_close_experiment()
 442    */
 443   if (!COUNTERS_ENABLED ())
 444     return COL_ERROR_NONE;
 445   detach_experiment ();
 446
 447   /* cpc or libperfctr may still generate sigemts for a while */
 448   /* verify that SIGEMT handler is still installed */
 449   /* (still required with sigaction interposition and management,
 450      since interposition is not done for attach experiments)
 451    */
 452   struct sigaction curr;
 453   if (__collector_sigaction (HWCFUNCS_SIGNAL, NULL, &curr) == -1)
 454     {
 455       TprintfT (0, "hwprofile close_experiment: ERROR: hwc sigaction check failed: errno=%d\n", errno);
 456     }
 457   else if (curr.sa_sigaction != collector_sigemt_handler)
 458     {
 459       TprintfT (DBG_LT1, "hwprofile close_experiment: WARNING: collector sigemt handler replaced by 0x%p!\n", curr.sa_handler);
 460       (void) collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">0x%p</event>\n",
 461                                             SP_JCMD_CWARN, COL_WARN_SIGEMT, curr.sa_handler);
 462     }
 463   else
 464     TprintfT (DBG_LT1, "hwprofile close_experiment: collector sigemt handler integrity verified!\n");
 465   TprintfT (0, "hwprofile: close_experiment\n");
 466   return 0;
 467 }
 468
 469 static int
 470 detach_experiment (void)
 471 {
 472   /* fork child.  Clean up state but don't write to experiment */
 473   /* note: stop_data_collection() has already been called by the fork_prologue */
 474   // detach_experiment() can be called asynchronously
 475   // from anywhere, even from within a sigemt handler
 476   // via DBX detach.
 477   // Important: stop_data_collection() _must_ be called
 478   // before detach_experiment() is called.
 479   if (!COUNTERS_ENABLED ())
 480     return COL_ERROR_NONE;
 481   TprintfT (0, "hwprofile: detach_experiment()\n");
 482   if (SD_OFF != __collector_cas_32 (&sd_state, SD_OFF, SD_PENDING))
 483     return 0;
 484   // one and only one call should ever make it here here.
 485   if (hwc_mode == HWCMODE_ACTIVE)
 486     {
 487       TprintfT (0, "hwprofile: ERROR: stop_data_collection() should have been called before detach_experiment()\n");
 488       stop_data_collection ();
 489     }
 490
 491   // Assumption: The only calls to sigemt_handler
 492   // we should see at this point
 493   // will be those that were already in-flight before
 494   // stop_new_sigemts() was called.
 495   if (nthreads_in_sighandler > 0)
 496     {
 497       // sigemt handlers should see
 498       // SD_PENDING and should call real_detach_experiment()
 499       // when the last handler is finished.
 500       TprintfT (DBG_LT1, "hwprofile: detach in the middle of signal handler.\n");
 501       return 0;
 502     }
 503
 504   // If we get here, there should be no remaining
 505   // sigemt handlers.  However,  we don't really know
 506   // if there were ever any in flight, so call
 507   // real_detach_experiment() here:
 508   return real_detach_experiment (); // multiple calls to this OK
 509 }
 510
 511 static int
 512 real_detach_experiment (void)
 513 {
 514   /*multiple calls to this routine are OK.*/
 515   if (SD_PENDING != __collector_cas_32 (&sd_state, SD_PENDING, SD_COMPLETE))
 516     return 0;
 517   // only the first caller to this routine should get here.
 518   hwcdef_cnt = 0; /* since now deinstalled */
 519   hwcdef = NULL;
 520   set_hwc_mode (HWCMODE_OFF);
 521   if (SD_COMPLETE != __collector_cas_32 (&sd_state, SD_COMPLETE, SD_OFF))
 522     {
 523       TprintfT (0, "hwprofile: ERROR: unexpected sd_state in real_detach_experiment()\n");
 524       sd_state = SD_OFF;
 525     }
 526   hwprofile_tsd_key = COLLECTOR_TSD_INVALID_KEY;
 527   TprintfT (DBG_LT0, "hwprofile: real_detach_experiment() detached from experiment.\n");
 528   return 0;
 529 }
 530
 531 /*---------------------------------------------------------------------------*/
 532 /* Record counter values. */
 533
 534 /* <value> should already be adjusted to be "zero-based" (counting up from 0).*/
 535 static void
 536 collector_record_counter_internal (ucontext_t *ucp, int timecvt,
 537                                    ABST_type ABS_memop, hrtime_t time,
 538                                    unsigned tag, uint64_t value, uint64_t pc,
 539                                    uint64_t va, uint64_t latency,
 540                                    uint64_t data_source)
 541 {
 542   MHwcntr_packet pckt;
 543   CALL_UTIL (memset)(&pckt, 0, sizeof ( MHwcntr_packet));
 544   pckt.comm.tstamp = time;
 545   pckt.tag = tag;
 546   if (timecvt > 1)
 547     {
 548       if (HWCVAL_HAS_ERR (value))
 549         {
 550           value = HWCVAL_CLR_ERR (value);
 551           value *= timecvt;
 552           value = HWCVAL_SET_ERR (value);
 553         }
 554       else
 555         value *= timecvt;
 556     }
 557   pckt.interval = value;
 558   pckt.comm.type = HW_PCKT;
 559   pckt.comm.tsize = sizeof (Hwcntr_packet);
 560   TprintfT (DBG_LT4, "hwprofile: %llu sample %lld tag %u recorded\n",
 561             (unsigned long long) time, (long long) value, tag);
 562   if (ABS_memop == ABST_NOPC)
 563     ucp = &expr_nopc_uc;
 564   pckt.comm.frinfo = collector_interface->getFrameInfo (expr_hndl, pckt.comm.tstamp, FRINFO_FROM_UC, ucp);
 565   collector_interface->writeDataRecord (expr_hndl, (Common_packet*) & pckt);
 566 }
 567
 568 static void
 569 collector_record_counter (ucontext_t *ucp, int timecvt, ABST_type ABS_memop,
 570                           hrtime_t time, unsigned tag, uint64_t value)
 571 {
 572   collector_record_counter_internal (ucp, timecvt, ABS_memop, time, tag, value,
 573                                      HWCFUNCS_INVALID_U64, HWCFUNCS_INVALID_U64,
 574                                      HWCFUNCS_INVALID_U64, HWCFUNCS_INVALID_U64);
 575 }
 576
 577
 578 /*---------------------------------------------------------------------------*/
 579 /* Signal handlers */
 580
 581 /* SIGEMT -- relayed from libcpc, when the counter overflows */
 582
 583 /*   Generates the appropriate event or events, and resets the counters */
 584 static void
 585 collector_sigemt_handler (int sig, siginfo_t *si, void *puc)
 586 {
 587   int rc;
 588   hwc_event_t sample, lost_samples;
 589   if (sig != HWCFUNCS_SIGNAL)
 590     {
 591       TprintfT (0, "hwprofile: ERROR: %s: unexpected signal %d\n", "collector_sigemt_handler", sig);
 592       return;
 593     }
 594   if (!COUNTERS_ENABLED ())
 595     { /* apparently deinstalled */
 596       TprintfT (0, "hwprofile: WARNING: SIGEMT detected after close_experiment()\n");
 597       /* kills future sigemts since hwcdrv_sighlr_restart() not called */
 598       return;
 599     }
 600
 601   /* Typically, we expect HWC overflow signals to come from the kernel: si_code > 0.
 602    * On Linux, however, dbx might be "forwarding" a signal using tkill()/tgkill().
 603    * For more information on what si_code values can be expected on Linux, check:
 604    *     cmn_components/Collector_Interface/hwcdrv_pcl.c     hwcdrv_overflow()
 605    *     cmn_components/Collector_Interface/hwcdrv_perfctr.c hdrv_perfctr_overflow()
 606    */
 607   if (puc == NULL || si == NULL || (si->si_code <= 0 && si->si_code != SI_TKILL))
 608     {
 609       TprintfT (DBG_LT3, "hwprofile: collector_sigemt_handler SIG%02d\n", sig);
 610       if (old_sigemt_handler.sa_handler == SIG_DFL)
 611         __collector_SIGDFL_handler (HWCFUNCS_SIGNAL);
 612       else if (old_sigemt_handler.sa_handler != SIG_IGN &&
 613          old_sigemt_handler.sa_sigaction != &collector_sigemt_handler)
 614         {
 615           /* Redirect the signal to the previous signal handler */
 616           (old_sigemt_handler.sa_sigaction)(sig, si, puc);
 617           TprintfT (DBG_LT1, "hwprofile: collector_sigemt_handler SIG%02d redirected to original handler\n", sig);
 618         }
 619       return;
 620     }
 621   rc = get_hwc_driver ()->hwcdrv_overflow (si, &sample, &lost_samples);
 622   if (rc)
 623     {
 624       /* hwcdrv_sighlr_restart() should not be called */
 625       TprintfT (0, "hwprofile: ERROR: collector_sigemt_handler: hwcdrv_overflow() failed\n");
 626       return;
 627     }
 628
 629   if (hwc_mode == HWCMODE_ACTIVE)
 630     {
 631       /* record the event only if counters are active */
 632       /* The following has been copied from dispatcher.c */
 633 #if ARCH(SPARC)
 634       /* 23340823 signal handler third argument should point to a ucontext_t */
 635       /* Convert sigcontext to ucontext_t on sparc-Linux */
 636       ucontext_t uctxmem;
 637       struct sigcontext *sctx = (struct sigcontext*) puc;
 638       ucontext_t *uctx = &uctxmem;
 639       uctx->uc_link = NULL;
 640 #if WSIZE(32)
 641       uctx->uc_mcontext.gregs[REG_PC] = sctx->si_regs.pc;
 642       __collector_memcpy (&uctx->uc_mcontext.gregs[3],
 643                           sctx->si_regs.u_regs,
 644                           sizeof (sctx->si_regs.u_regs));
 645 #else
 646       uctx->uc_mcontext.mc_gregs[MC_PC] = sctx->sigc_regs.tpc;
 647       __collector_memcpy (&uctx->uc_mcontext.mc_gregs[3],
 648                           sctx->sigc_regs.u_regs,
 649                           sizeof (sctx->sigc_regs.u_regs));
 650 #endif /* WSIZE() */
 651 #else
 652       ucontext_t *uctx = (ucontext_t*) puc;
 653 #endif /* ARCH() */
 654
 655       for (int ii = 0; ii < hwcdef_cnt; ii++)
 656         if (lost_samples.ce_pic[ii])
 657           collector_record_counter (&expr_lostcounts_uc, hwcdef[ii]->timecvt,
 658                                     hwcdef[ii]->memop, lost_samples.ce_hrt,
 659                                     hwcdef[ii]->sort_order, lost_samples.ce_pic[ii]);
 660       for (int ii = 0; ii < hwcdef_cnt; ii++)
 661         if (sample.ce_pic[ii])
 662           collector_record_counter (uctx, hwcdef[ii]->timecvt,
 663                                     hwcdef[ii]->memop, sample.ce_hrt,
 664                                     hwcdef[ii]->sort_order, sample.ce_pic[ii]);
 665     }
 666   rc = get_hwc_driver ()->hwcdrv_sighlr_restart (NULL);
 667 }
 668 /*      SIGPROF -- not installed as handler, but
 669  *      __collector_ext_hwc_check: called by (SIGPROF) dispatcher.
 670  *       Periodical check of integrity of HWC count/signal mechanism,
 671  *       as required for various chip/system bugs/workarounds.
 672  */
 673 void
 674 __collector_ext_hwc_check (siginfo_t *info, ucontext_t *vcontext) { }
 675
 676 /*---------------------------------------------------------------------------*/
 677 int
 678 collector_sigemt_sigaction (const struct sigaction *nact,
 679                             struct sigaction *oact)
 680 {
 681   struct sigaction oact_check;
 682   /* Error codes and messages that refer to HWC are tricky.
 683    * E.g., HWC profiling might not even be on;  we might
 684    * encounter an error here simply because the user is
 685    * trying to set a handler for a signal that happens to
 686    * be HWCFUNCS_SIGNAL, which we aren't even using.
 687    */
 688   if (__collector_sigaction (HWCFUNCS_SIGNAL, NULL, &oact_check) != 0)
 689     {
 690       TprintfT (0, "hwprofile: ERROR: collector_sigemt_sigaction(): request to set handler for signal %d, but check on existing handler failed\n", HWCFUNCS_SIGNAL);
 691       collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">old handler for signal %d could not be determined</event>\n", SP_JCMD_CERROR, COL_ERROR_HWCINIT, HWCFUNCS_SIGNAL);
 692       return COL_ERROR_HWCINIT;
 693     }
 694
 695   if (oact_check.sa_sigaction == collector_sigemt_handler)
 696     {
 697       /* dispatcher is in place, so nact/oact apply to old_sigemt_handler */
 698       if (oact != NULL)
 699         {
 700           oact->sa_handler = old_sigemt_handler.sa_handler;
 701           oact->sa_mask = old_sigemt_handler.sa_mask;
 702           oact->sa_flags = old_sigemt_handler.sa_flags;
 703         }
 704       if (nact != NULL)
 705         {
 706           old_sigemt_handler.sa_handler = nact->sa_handler;
 707           old_sigemt_handler.sa_mask = nact->sa_mask;
 708           old_sigemt_handler.sa_flags = nact->sa_flags;
 709         }
 710       return COL_ERROR_NONE;
 711     }
 712   else /* no dispatcher in place, so just act like normal sigaction() */
 713     return __collector_sigaction (HWCFUNCS_SIGNAL, nact, oact);
 714 }
 715
 716 static void
 717 collector_hwc_ABORT (int errnum, const char *msg)
 718 {
 719   TprintfT (0, "hwprofile: collector_hwc_ABORT: [%d] %s\n", errnum, msg);
 720   if (hwc_mode == HWCMODE_ABORT) /* HWC collection already aborted! */
 721     return;
 722   set_hwc_mode (HWCMODE_ABORT); /* set global flag to disable handlers and indicate abort */
 723
 724   /* Write the error message to the experiment */
 725   collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s: errno=%d</event>\n",
 726                                  SP_JCMD_CERROR, COL_ERROR_HWCFAIL, msg, errnum);
 727
 728 #ifdef REAL_DEBUG
 729   abort ();
 730 #else
 731   TprintfT (0, "hwprofile: Continuing without HWC collection...\n");
 732 #endif
 733 }
 734
 735 static int
 736 start_data_collection (void)
 737 {
 738   hwc_mode_t old_mode = hwc_mode;
 739   if (!COUNTERS_ENABLED ())
 740     return COL_ERROR_NONE;
 741   TprintfT (0, "hwprofile: start_data_collection (hwc_mode=%d)\n", old_mode);
 742   switch (old_mode)
 743     {
 744     case HWCMODE_OFF:
 745       if (get_hwc_driver ()->hwcdrv_start ())
 746         {
 747           TprintfT (0, "hwprofile: ERROR: start_data_collection() failed in hwcdrv_start()\n");
 748           collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s: errno=%d</event>\n",
 749                                          SP_JCMD_CERROR, COL_ERROR_HWCFAIL,
 750                                          "start_data_collection()", errno);
 751           return COL_ERROR_HWCINIT;
 752         }
 753       set_hwc_mode (HWCMODE_ACTIVE); /* start handling events on signals */
 754       break;
 755     case HWCMODE_SUSPEND:
 756       if (get_hwc_driver ()->hwcdrv_lwp_resume ())
 757         {
 758           TprintfT (0, "hwprofile: ERROR: start_data_collection() failed in hwcdrv_lwp_resume()\n");
 759           /* ignore errors from lwp_resume() */
 760         }
 761       set_hwc_mode (HWCMODE_ACTIVE); /* start handling events on signals */
 762       break;
 763     default:
 764       TprintfT (0, "hwprofile: ERROR: start_data_collection() invalid mode\n");
 765       return COL_ERROR_HWCINIT;
 766     }
 767   return COL_ERROR_NONE;
 768 }
 769
 770 static int
 771 stop_data_collection (void)
 772 {
 773   hwc_mode_t old_mode = hwc_mode;
 774   if (!COUNTERS_ENABLED ())
 775     return COL_ERROR_NONE;
 776   TprintfT (0, "hwprofile: stop_data_collection (hwc_mode=%d)\n", old_mode);
 777   switch (old_mode)
 778     {
 779     case HWCMODE_SUSPEND:
 780       return COL_ERROR_NONE;
 781     case HWCMODE_ACTIVE:
 782       set_hwc_mode (HWCMODE_SUSPEND); /* stop handling signals */
 783       break;
 784     default:
 785       /* Don't change the mode, but attempt to suspend anyway... */
 786       break;
 787     }
 788
 789   if (get_hwc_driver ()->hwcdrv_lwp_suspend ())
 790     /* ignore errors from lwp_suspend() */
 791     TprintfT (0, "hwprofile: ERROR: stop_data_collection() failed in hwcdrv_lwp_suspend()\n");
 792
 793   /*
 794    * hwcdrv_lwp_suspend() cannot guarantee that all SIGEMTs will stop
 795    * but hwc_mode will prevent logging and counters will overflow once
 796    * then stay frozen.
 797    */
 798   /*   There may still be pending SIGEMTs so don't reset the SIG_DFL handler.
 799    */
 800   /* see comment in dispatcher.c */
 801   /* ret = __collector_sigaction( SIGEMT, &old_sigemt_handler, NULL ); */
 802   return COL_ERROR_NONE;
 803 }
 804
 805 /*---------------------------------------------------------------------------*/
 806
 807 /* utilities */
 808 static void
 809 set_hwc_mode (hwc_mode_t md)
 810 {
 811   TprintfT (DBG_LT1, "hwprofile: set_hwc_mode(%d)\n", md);
 812   hwc_mode = md;
 813 }
 814
 815 int
 816 __collector_ext_hwc_active ()
 817 {
 818   return (hwc_mode == HWCMODE_ACTIVE);
 819 }
 820
 821 static void
 822 hwclogwrite0 ()
 823 {
 824   collector_interface->writeLog ("<profdata fname=\"%s\"/>\n",
 825                                  module_interface.description);
 826   /* Record Hwcntr_packet description */
 827   Hwcntr_packet *pp = NULL;
 828   collector_interface->writeLog ("<profpckt kind=\"%d\" uname=\"" STXT ("Hardware counter profiling data") "\">\n", HW_PCKT);
 829   collector_interface->writeLog ("    <field name=\"LWPID\" uname=\"" STXT ("Lightweight process id") "\" offset=\"%d\" type=\"%s\"/>\n",
 830                                  &pp->comm.lwp_id, sizeof (pp->comm.lwp_id) == 4 ? "INT32" : "INT64");
 831   collector_interface->writeLog ("    <field name=\"THRID\" uname=\"" STXT ("Thread number") "\" offset=\"%d\" type=\"%s\"/>\n",
 832                                  &pp->comm.thr_id, sizeof (pp->comm.thr_id) == 4 ? "INT32" : "INT64");
 833   collector_interface->writeLog ("    <field name=\"CPUID\" uname=\"" STXT ("CPU id") "\" offset=\"%d\" type=\"%s\"/>\n",
 834                                  &pp->comm.cpu_id, sizeof (pp->comm.cpu_id) == 4 ? "INT32" : "INT64");
 835   collector_interface->writeLog ("    <field name=\"TSTAMP\" uname=\"" STXT ("High resolution timestamp") "\" offset=\"%d\" type=\"%s\"/>\n",
 836                                  &pp->comm.tstamp, sizeof (pp->comm.tstamp) == 4 ? "INT32" : "INT64");
 837   collector_interface->writeLog ("    <field name=\"FRINFO\" offset=\"%d\" type=\"%s\"/>\n",
 838                                  &pp->comm.frinfo, sizeof (pp->comm.frinfo) == 4 ? "INT32" : "INT64");
 839   collector_interface->writeLog ("    <field name=\"HWCTAG\" uname=\"" STXT ("Hardware counter index") "\" offset=\"%d\" type=\"%s\"/>\n",
 840                                  &pp->tag, sizeof (pp->tag) == 4 ? "INT32" : "INT64");
 841   collector_interface->writeLog ("    <field name=\"HWCINT\" uname=\"" STXT ("Hardware counter interval") "\" offset=\"%d\" type=\"%s\"/>\n",
 842                                  &pp->interval, sizeof (pp->interval) == 4 ? "INT32" : "INT64");
 843   collector_interface->writeLog ("</profpckt>\n");
 844   if (hwcdef_has_memspace)
 845     {
 846       /* Record MHwcntr_packet description */
 847       MHwcntr_packet *xpp = NULL;
 848       collector_interface->writeLog ("<profpckt kind=\"%d\" uname=\"" STXT ("Hardware counter profiling data") "\">\n", MHWC_PCKT);
 849       collector_interface->writeLog ("    <field name=\"LWPID\" uname=\"" STXT ("Lightweight process id") "\" offset=\"%d\" type=\"%s\"/>\n",
 850                                      &xpp->comm.lwp_id, sizeof (xpp->comm.lwp_id) == 4 ? "INT32" : "INT64");
 851       collector_interface->writeLog ("    <field name=\"THRID\" uname=\"" STXT ("Thread number") "\" offset=\"%d\" type=\"%s\"/>\n",
 852                                      &xpp->comm.thr_id, sizeof (xpp->comm.thr_id) == 4 ? "INT32" : "INT64");
 853       collector_interface->writeLog ("    <field name=\"CPUID\" uname=\"" STXT ("CPU id") "\" offset=\"%d\" type=\"%s\"/>\n",
 854                                      &xpp->comm.cpu_id, sizeof (xpp->comm.cpu_id) == 4 ? "INT32" : "INT64");
 855       collector_interface->writeLog ("    <field name=\"TSTAMP\" uname=\"" STXT ("High resolution timestamp") "\" offset=\"%d\" type=\"%s\"/>\n",
 856                                      &xpp->comm.tstamp, sizeof (xpp->comm.tstamp) == 4 ? "INT32" : "INT64");
 857       collector_interface->writeLog ("    <field name=\"FRINFO\" offset=\"%d\" type=\"%s\"/>\n",
 858                                      &xpp->comm.frinfo, sizeof (xpp->comm.frinfo) == 4 ? "INT32" : "INT64");
 859       collector_interface->writeLog ("    <field name=\"HWCTAG\" uname=\"" STXT ("Hardware counter index") "\" offset=\"%d\" type=\"%s\"/>\n",
 860                                      &xpp->tag, sizeof (xpp->tag) == 4 ? "INT32" : "INT64");
 861       collector_interface->writeLog ("    <field name=\"HWCINT\" uname=\"" STXT ("Hardware counter interval") "\" offset=\"%d\" type=\"%s\"/>\n",
 862                                      &xpp->interval, sizeof (xpp->interval) == 4 ? "INT32" : "INT64");
 863       collector_interface->writeLog ("    <field name=\"VADDR\" uname=\"" STXT ("Virtual address (data)") "\" offset=\"%d\" type=\"%s\"/>\n",
 864                                      &xpp->ea_vaddr, sizeof (xpp->ea_vaddr) == 4 ? "UINT32" : "UINT64");
 865       collector_interface->writeLog ("    <field name=\"PADDR\" uname=\"" STXT ("Physical address (data)") "\" offset=\"%d\" type=\"%s\"/>\n",
 866                                      &xpp->ea_paddr, sizeof (xpp->ea_paddr) == 4 ? "UINT32" : "UINT64");
 867       collector_interface->writeLog ("    <field name=\"VIRTPC\" uname=\"" STXT ("Virtual address (instruction)") "\" offset=\"%d\" type=\"%s\"/>\n",
 868                                      &xpp->pc_vaddr, sizeof (xpp->pc_vaddr) == 4 ? "UINT32" : "UINT64");
 869       collector_interface->writeLog ("    <field name=\"PHYSPC\" uname=\"" STXT ("Physical address (instruction)") "\" offset=\"%d\" type=\"%s\"/>\n",
 870                                      &xpp->pc_paddr, sizeof (xpp->pc_paddr) == 4 ? "UINT32" : "UINT64");
 871       collector_interface->writeLog ("    <field name=\"EA_PAGESIZE\" uname=\"" STXT ("Page size (data)") "\" offset=\"%d\" type=\"%s\"/>\n",
 872                                      &xpp->ea_pagesz, sizeof (xpp->ea_pagesz) == 4 ? "INT32" : "INT64");
 873       collector_interface->writeLog ("    <field name=\"PC_PAGESIZE\" uname=\"" STXT ("Page size (instruction)") "\" offset=\"%d\" type=\"%s\"/>\n",
 874                                      &xpp->pc_pagesz, sizeof (xpp->pc_pagesz) == 4 ? "INT32" : "INT64");
 875       collector_interface->writeLog ("    <field name=\"EA_LGRP\" uname=\"" STXT ("Page locality group (data)") "\" offset=\"%d\" type=\"%s\"/>\n",
 876                                      &xpp->ea_lgrp, sizeof (xpp->ea_lgrp) == 4 ? "INT32" : "INT64");
 877       collector_interface->writeLog ("    <field name=\"PC_LGRP\" uname=\"" STXT ("Page locality group (instruction)") "\" offset=\"%d\" type=\"%s\"/>\n",
 878                                      &xpp->pc_lgrp, sizeof (xpp->pc_lgrp) == 4 ? "INT32" : "INT64");
 879       collector_interface->writeLog ("    <field name=\"LWP_LGRP_HOME\" uname=\"" STXT ("LWP home lgroup id") "\" offset=\"%d\" type=\"%s\"/>\n",
 880                                      &xpp->lgrp_lwp, sizeof (xpp->lgrp_lwp) == 4 ? "INT32" : "INT64");
 881       collector_interface->writeLog ("    <field name=\"PS_LGRP_HOME\" uname=\"" STXT ("Process home lgroup id") "\" offset=\"%d\" type=\"%s\"/>\n",
 882                                      &xpp->lgrp_ps, sizeof (xpp->lgrp_ps) == 4 ? "INT32" : "INT64");
 883       collector_interface->writeLog ("    <field name=\"MEM_LAT\" uname=\"" STXT ("Memory Latency Cycles") "\" offset=\"%d\" type=\"%s\"/>\n",
 884                                      &xpp->latency, sizeof (xpp->latency) == 4 ? "INT32" : "INT64");
 885       collector_interface->writeLog ("    <field name=\"MEM_SRC\" uname=\"" STXT ("Memory Data Source") "\" offset=\"%d\" type=\"%s\"/>\n",
 886                                      &xpp->data_source, sizeof (xpp->data_source) == 4 ? "INT32" : "INT64");
 887       collector_interface->writeLog ("</profpckt>\n");
 888     }
 889 }
 890
 891 static void
 892 hwclogwrite (Hwcentry * ctr)
 893 {
 894   TprintfT (DBG_LT1, "hwprofile: writeLog(%s %u %s %d %u %d)\n",
 895             SP_JCMD_HW_COUNTER, cpcN_cpuver, ctr->name ? ctr->name : "NULL",
 896             ctr->val, ctr->sort_order, ctr->memop);
 897   collector_interface->writeLog ("<profile name=\"%s\"", SP_JCMD_HW_COUNTER);
 898   collector_interface->writeLog (" cpuver=\"%u\"", cpcN_cpuver);
 899   collector_interface->writeLog (" hwcname=\"%s\"", ctr->name);
 900   collector_interface->writeLog (" int_name=\"%s\"", ctr->int_name);
 901   collector_interface->writeLog (" interval=\"%d\"", ctr->val);
 902   collector_interface->writeLog (" tag=\"%u\"", ctr->sort_order);
 903   collector_interface->writeLog (" memop=\"%d\"", ctr->memop);
 904   collector_interface->writeLog ("/>\n");
 905 }