gprofng/libcollector/hwprofile.c

   1 /* Copyright (C) 2021-2023 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
  33 #include "gp-defs.h"
  34 #define _STRING_H 1  /* XXX MEZ: temporary workaround */
  35 #include "hwcdrv.h"
  36 #include "collector_module.h"
  37 #include "gp-experiment.h"
  38 #include "libcol_util.h"
  39 #include "hwprofile.h"
  40 #include "ABS.h"
  41 #include "tsd.h"
  42
  43 /* TprintfT(<level>,...) definitions.  Adjust per module as needed */
  44 #define DBG_LT0 0 // for high-level configuration, unexpected errors/warnings
  45 #define DBG_LT1 1 // for configuration details, warnings
  46 #define DBG_LT2 2
  47 #define DBG_LT3 3
  48 #define DBG_LT4 4
  49 #define DBG_LT5 5
  50
  51 #define  SD_OFF 0       /* before start or after close she shut down process */
  52 #define  SD_PENDING 1   /* before running real_detach_experiment() */
  53 #define  SD_COMPLETE 2  /* after running real_detach_experiment() */
  54
  55 static int init_interface (CollectorInterface*);
  56 static int open_experiment (const char *);
  57 static int start_data_collection (void);
  58 static int stop_data_collection (void);
  59 static int close_experiment (void);
  60 static int detach_experiment (void);
  61 static int real_detach_experiment (void);
  62
  63 static ModuleInterface module_interface ={
  64   SP_HWCNTR_FILE,           /* description */
  65   init_interface,           /* initInterface */
  66   open_experiment,          /* openExperiment */
  67   start_data_collection,    /* startDataCollection */
  68   stop_data_collection,     /* stopDataCollection */
  69   close_experiment,         /* closeExperiment */
  70   detach_experiment         /* detachExperiment (fork child) */
  71 };
  72
  73 static CollectorInterface *collector_interface = NULL;
  74
  75
  76 /*---------------------------------------------------------------------------*/
  77 /* compile options and workarounds */
  78
  79 /* Solaris: We set ITIMER_REALPROF to ensure that counters get started on
  80  *      LWPs that existed before the collector initialization.
  81  *
  82  * In addition, if the appropriate #define's are set, we check for:
  83  *      lost-hw-overflow -- the HW counters rollover, but the overflow
  84  *            interrupt is not generated (counters keep running)
  85  *      lost-sigemt -- the interrupt is received by the kernel,
  86  *            which stops the counters, but the kernel fails
  87  *            to deliver the signal.
  88  */
  89
  90 /*---------------------------------------------------------------------------*/
  91 /* typedefs */
  92
  93 typedef enum {
  94   HWCMODE_OFF,       /* before start or after close */
  95   HWCMODE_SUSPEND,  /* stop_data_collection called */
  96   HWCMODE_ACTIVE,   /* counters are defined and after start_data_collection() */
  97   HWCMODE_ABORT     /* fatal error occured. Log a message, stop recording */
  98 } hwc_mode_t;
  99
 100 /*---------------------------------------------------------------------------*/
 101 /* prototypes */
 102 static void init_ucontexts (void);
 103 static int hwc_initialize_handlers (void);
 104 static void collector_record_counter (ucontext_t*,
 105                                       int timecvt,
 106                                       ABST_type, hrtime_t,
 107                                       unsigned, uint64_t);
 108 static void collector_hwc_ABORT (int errnum, const char *msg);
 109 static void hwclogwrite0 ();
 110 static void hwclogwrite (Hwcentry *);
 111 static void set_hwc_mode (hwc_mode_t);
 112 static void collector_sigemt_handler (int sig, siginfo_t *si, void *puc);
 113
 114 /*---------------------------------------------------------------------------*/
 115 /* static variables */
 116
 117 /* --- user counter selections and options */
 118 static int hwcdef_has_memspace;     /* true to indicate use of extened packets */
 119 static unsigned hwcdef_cnt;         /* number of *active* hardware counters */
 120 static unsigned hwcdef_num_sampling_ctrdefs; /* ctrs that use sampling */
 121 static unsigned hwcdef_num_overflow_ctrdefs; /* ctrs that use overflow */
 122 static Hwcentry **hwcdef;           /* HWC definitions */
 123 static int cpcN_cpuver = CPUVER_UNDEFINED;
 124 static int hwcdrv_inited;           /* Don't call hwcdrv_init() in fork_child */
 125 static hwcdrv_api_t *hwc_driver = NULL;
 126 static unsigned hwprofile_tsd_key = COLLECTOR_TSD_INVALID_KEY;
 127 static int hwprofile_tsd_sz = 0;
 128 static volatile hwc_mode_t hwc_mode = HWCMODE_OFF;
 129 static volatile unsigned int nthreads_in_sighandler = 0;
 130 static volatile unsigned int sd_state = SD_OFF;
 131
 132 /* --- experiment logging state */
 133 static CollectorModule expr_hndl = COLLECTOR_MODULE_ERR;
 134 static ucontext_t expr_dummy_uc;        // used for hacked "collector" frames
 135 static ucontext_t expr_out_of_range_uc; // used for "out-of-range" frames
 136 static ucontext_t expr_frozen_uc;       // used for "frozen" frames
 137 static ucontext_t expr_nopc_uc;         // used for not-program-related frames
 138 static ucontext_t expr_lostcounts_uc;   // used for lost_counts frames
 139
 140 /* --- signal handler state */
 141 static struct sigaction old_sigemt_handler;  //overwritten in fork-child
 142
 143 /*---------------------------------------------------------------------------*/
 144 /* macros */
 145 #define COUNTERS_ENABLED()  (hwcdef_cnt)
 146 #define gethrtime           collector_interface->getHiResTime
 147
 148 #ifdef DEBUG
 149 #define Tprintf(...)   if (collector_interface) collector_interface->writeDebugInfo( 0, __VA_ARGS__ )
 150 #define TprintfT(...)  if (collector_interface) collector_interface->writeDebugInfo( 1, __VA_ARGS__ )
 151 #else
 152 #define Tprintf(...)
 153 #define TprintfT(...)
 154 #endif
 155
 156
 157 /*---------------------------------------------------------------------------*/
 158
 159 /* Initialization routines */
 160 static hwcdrv_api_t *
 161 get_hwc_driver ()
 162 {
 163   if (hwc_driver == NULL)
 164     hwc_driver = __collector_get_hwcdrv ();
 165   return hwc_driver;
 166 }
 167
 168 static void init_module () __attribute__ ((constructor));
 169 static void
 170 init_module ()
 171 {
 172   __collector_dlsym_guard = 1;
 173   RegModuleFunc reg_module = (RegModuleFunc) dlsym (RTLD_DEFAULT, "__collector_register_module");
 174   __collector_dlsym_guard = 0;
 175   if (reg_module == NULL)
 176     {
 177       TprintfT (0, "hwprofile: init_module FAILED - reg_module = NULL\n");
 178       return;
 179     }
 180   expr_hndl = reg_module (&module_interface);
 181   if (expr_hndl == COLLECTOR_MODULE_ERR)
 182     {
 183       TprintfT (0, "hwprofile: ERROR: handle not created.\n");
 184       if (collector_interface)
 185         collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">data handle not created</event>\n",
 186                                        SP_JCMD_CERROR, COL_ERROR_HWCINIT);
 187     }
 188 }
 189
 190 static int
 191 init_interface (CollectorInterface *_collector_interface)
 192 {
 193   collector_interface = _collector_interface;
 194   return COL_ERROR_NONE;
 195 }
 196
 197 static void *
 198 hwprofile_get_tsd ()
 199 {
 200   return collector_interface->getKey (hwprofile_tsd_key);
 201 }
 202
 203 static int
 204 open_experiment (const char *exp)
 205 {
 206   if (collector_interface == NULL)
 207     {
 208       TprintfT (0, "hwprofile: ERROR: collector_interface is null.\n");
 209       return COL_ERROR_HWCINIT;
 210     }
 211   const char *params = collector_interface->getParams ();
 212   while (params)
 213     {
 214       if (__collector_strStartWith (params, "h:*") == 0)
 215         {
 216           /* HWC counters set by default */
 217           collector_interface->writeLog ("<%s %s=\"1\"/>\n",
 218                                          SP_TAG_SETTING, SP_JCMD_HWC_DEFAULT);
 219           params += 3;
 220           break;
 221         }
 222       else if (__collector_strStartWith (params, "h:") == 0)
 223         {
 224           params += 2;
 225           break;
 226         }
 227       params = CALL_UTIL (strchr)(params, ';');
 228       if (params)
 229         params++;
 230     }
 231   if (params == NULL)  /* HWC profiling not specified */
 232     return COL_ERROR_HWCINIT;
 233   char *s = CALL_UTIL (strchr)(params, (int) ';');
 234   int sz = s ? s - params : CALL_UTIL (strlen)(params);
 235   char *defstring = (char*) alloca (sz + 1);
 236   CALL_UTIL (strlcpy)(defstring, params, sz + 1);
 237   TprintfT (0, "hwprofile: open_experiment %s -- %s\n", exp, defstring);
 238
 239   int err = COL_ERROR_NONE;
 240   /* init counter library */
 241   if (!hwcdrv_inited)
 242     { /* do not call hwcdrv_init() from fork-child */
 243       hwcdrv_inited = 1;
 244       get_hwc_driver ();
 245       if (hwc_driver->hwcdrv_init (collector_hwc_ABORT, &hwprofile_tsd_sz) == 0)
 246         {
 247           collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s</event>\n",
 248                                          SP_JCMD_CERROR, COL_ERROR_HWCINIT, defstring);
 249           TprintfT (0, "hwprofile: ERROR: hwcfuncs_init() failed\n");
 250           return COL_ERROR_HWCINIT;
 251         }
 252
 253       if (hwc_driver->hwcdrv_enable_mt (hwprofile_get_tsd))
 254         {
 255           // It is OK to call hwcdrv_enable_mt() before tsd key is created
 256           collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s</event>\n",
 257                                          SP_JCMD_CERROR, COL_ERROR_HWCINIT, defstring);
 258           TprintfT (0, "hwprofile: ERROR: hwcdrv_enable_mt() failed\n");
 259           return COL_ERROR_HWCINIT;
 260         }
 261
 262       hwc_driver->hwcdrv_get_info (&cpcN_cpuver, NULL, NULL, NULL, NULL);
 263       if (cpcN_cpuver < 0)
 264         {
 265           collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s</event>\n",
 266                                          SP_JCMD_CERROR, COL_ERROR_HWCINIT, defstring);
 267           TprintfT (0, "hwprofile: ERROR: hwcdrv_get_info() failed\n");
 268           return COL_ERROR_HWCINIT;
 269         }
 270     }
 271
 272   if (hwprofile_tsd_sz)
 273     {
 274       hwprofile_tsd_key = collector_interface->createKey (hwprofile_tsd_sz, NULL, NULL);
 275       if (hwprofile_tsd_key == COLLECTOR_TSD_INVALID_KEY)
 276         {
 277           collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s</event>\n",
 278                                          SP_JCMD_CERROR, COL_ERROR_HWCINIT, defstring);
 279           TprintfT (0, "hwprofile: ERROR: TSD createKey failed\n");
 280           return COL_ERROR_HWCINIT;
 281         }
 282     }
 283   hwcdef_cnt = 0;
 284   hwcdef_has_memspace = 0;
 285
 286   /* create counters based on hwcdef[] */
 287   err = __collector_hwcfuncs_bind_descriptor (defstring);
 288   if (err)
 289     {
 290       err = err == HWCFUNCS_ERROR_HWCINIT ? COL_ERROR_HWCINIT : COL_ERROR_HWCARGS;
 291       collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s</event>\n",
 292                                      SP_JCMD_CERROR, err, defstring);
 293       TprintfT (0, "hwprofile: ERROR: open_experiment() failed, RC=%d \n", err);
 294       return err;
 295     }
 296
 297   /* generate an array of counter structures for each requested counter */
 298   hwcdef = __collector_hwcfuncs_get_ctrs (&hwcdef_cnt);
 299   hwcdef_num_sampling_ctrdefs = hwcdef_num_overflow_ctrdefs = 0;
 300   int idx;
 301   for (idx = 0; idx < hwcdef_cnt; idx++)
 302     {
 303       if (HWCENTRY_USES_SAMPLING (hwcdef[idx]))
 304         {
 305           hwcdef_num_sampling_ctrdefs++;
 306         }
 307       else
 308         {
 309           hwcdef_num_overflow_ctrdefs++;
 310         }
 311     }
 312
 313   init_ucontexts ();
 314
 315   /* initialize the SIGEMT handler, and the periodic HWC checker */
 316   err = hwc_initialize_handlers ();
 317   if (err != COL_ERROR_NONE)
 318     {
 319       hwcdef_cnt = 0;
 320       TprintfT (0, "hwprofile: ERROR: open_experiment() failed, RC=%d \n", err);
 321       /* log written by hwc_initialize_handlers() */
 322       return err;
 323     }
 324
 325   for (idx = 0; idx < hwcdef_cnt; idx++)
 326     if (ABST_BACKTRACK_ENABLED (hwcdef[idx]->memop))
 327       hwcdef_has_memspace = 1;
 328
 329   /* record the hwc definitions in the log, based on the counter array */
 330   hwclogwrite0 ();
 331   for (idx = 0; idx < hwcdef_cnt; idx++)
 332     hwclogwrite (hwcdef[idx]);
 333   return COL_ERROR_NONE;
 334 }
 335
 336 int
 337 __collector_ext_hwc_lwp_init ()
 338 {
 339   return get_hwc_driver ()->hwcdrv_lwp_init ();
 340 }
 341
 342 void
 343 __collector_ext_hwc_lwp_fini ()
 344 {
 345   get_hwc_driver ()->hwcdrv_lwp_fini ();
 346 }
 347
 348 int
 349 __collector_ext_hwc_lwp_suspend ()
 350 {
 351   return get_hwc_driver ()->hwcdrv_lwp_suspend ();
 352 }
 353
 354 int
 355 __collector_ext_hwc_lwp_resume ()
 356 {
 357   return get_hwc_driver ()->hwcdrv_lwp_resume ();
 358 }
 359
 360 /* Dummy routine, used to provide a context for non-program related profiles */
 361 void
 362 __collector_not_program_related () { }
 363
 364 /* Dummy routine, used to provide a context for lost counts (perf_events) */
 365 void
 366 __collector_hwc_samples_lost () { }
 367
 368 /* Dummy routine, used to provide a context */
 369 void
 370 __collector_hwcs_frozen () { }
 371
 372 /* Dummy routine, used to provide a context */
 373 void
 374 __collector_hwcs_out_of_range () { }
 375 /* initialize some structures */
 376 static void
 377 init_ucontexts (void)
 378 {
 379   /* initialize dummy context for "collector" frames */
 380   CALL_UTIL (getcontext) (&expr_dummy_uc);
 381   SETFUNCTIONCONTEXT (&expr_dummy_uc, NULL);
 382
 383   /* initialize dummy context for "out-of-range" frames */
 384   CALL_UTIL (getcontext) (&expr_out_of_range_uc);
 385   SETFUNCTIONCONTEXT (&expr_out_of_range_uc, &__collector_hwcs_out_of_range);
 386
 387   /* initialize dummy context for "frozen" frames */
 388   CALL_UTIL (getcontext) (&expr_frozen_uc);
 389   SETFUNCTIONCONTEXT (&expr_frozen_uc, &__collector_hwcs_frozen);
 390
 391   /* initialize dummy context for non-program-related frames */
 392   CALL_UTIL (getcontext) (&expr_nopc_uc);
 393   SETFUNCTIONCONTEXT (&expr_nopc_uc, &__collector_not_program_related);
 394
 395   /* initialize dummy context for lost-counts-related frames */
 396   CALL_UTIL (getcontext) (&expr_lostcounts_uc);
 397   SETFUNCTIONCONTEXT (&expr_lostcounts_uc, &__collector_hwc_samples_lost);
 398 }
 399 /* initialize the signal handler */
 400 static int
 401 hwc_initialize_handlers (void)
 402 {
 403   /* install the signal handler for SIGEMT */
 404   struct sigaction oact;
 405   if (__collector_sigaction (HWCFUNCS_SIGNAL, NULL, &oact) != 0)
 406     {
 407       TprintfT (0, "hwc_initialize_handlers(): ERROR: hwc_initialize_handlers(): __collector_sigaction() failed to get oact\n");
 408       collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">old handler could not be determined</event>\n", SP_JCMD_CERROR, COL_ERROR_HWCINIT);
 409       return COL_ERROR_HWCINIT;
 410     }
 411   if (oact.sa_sigaction == collector_sigemt_handler)
 412     {
 413       /* signal handler is already in place; we are probably in a fork-child */
 414       TprintfT (DBG_LT1, "hwc_initialize_handlers(): hwc_initialize_handlers() collector_sigemt_handler already installed\n");
 415     }
 416   else
 417     {
 418       /* set our signal handler */
 419       struct sigaction c_act;
 420       CALL_UTIL (memset)(&c_act, 0, sizeof c_act);
 421       sigemptyset (&c_act.sa_mask);
 422       sigaddset (&c_act.sa_mask, SIGPROF); /* block SIGPROF delivery in handler */
 423       /* XXXX should probably also block sample_sig & pause_sig */
 424       c_act.sa_sigaction = collector_sigemt_handler;  /* note: used to set sa_handler instead */
 425       c_act.sa_flags = SA_RESTART | SA_SIGINFO;
 426       if (__collector_sigaction (HWCFUNCS_SIGNAL, &c_act, &old_sigemt_handler) != 0)
 427         {
 428           TprintfT (0, "hwc_initialize_handlers(): ERROR: hwc_initialize_handlers(): __collector_sigaction() failed to set cact\n");
 429           collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">event handler could not be installed</event>\n", SP_JCMD_CERROR, COL_ERROR_HWCINIT);
 430           return COL_ERROR_HWCINIT;
 431         }
 432     }
 433   return COL_ERROR_NONE;
 434 }
 435
 436 static int
 437 close_experiment (void)
 438 {
 439   /* note: stop_data_collection() should have already been called by
 440    * collector_close_experiment()
 441    */
 442   if (!COUNTERS_ENABLED ())
 443     return COL_ERROR_NONE;
 444   detach_experiment ();
 445
 446   /* cpc or libperfctr may still generate sigemts for a while */
 447   /* verify that SIGEMT handler is still installed */
 448   /* (still required with sigaction interposition and management,
 449      since interposition is not done for attach experiments)
 450    */
 451   struct sigaction curr;
 452   if (__collector_sigaction (HWCFUNCS_SIGNAL, NULL, &curr) == -1)
 453     {
 454       TprintfT (0, "hwprofile close_experiment: ERROR: hwc sigaction check failed: errno=%d\n", errno);
 455     }
 456   else if (curr.sa_sigaction != collector_sigemt_handler)
 457     {
 458       TprintfT (DBG_LT1, "hwprofile close_experiment: WARNING: collector sigemt handler replaced by 0x%p!\n", curr.sa_handler);
 459       (void) collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">0x%p</event>\n",
 460                                             SP_JCMD_CWARN, COL_WARN_SIGEMT, curr.sa_handler);
 461     }
 462   else
 463     TprintfT (DBG_LT1, "hwprofile close_experiment: collector sigemt handler integrity verified!\n");
 464   TprintfT (0, "hwprofile: close_experiment\n");
 465   return 0;
 466 }
 467
 468 static int
 469 detach_experiment (void)
 470 {
 471   /* fork child.  Clean up state but don't write to experiment */
 472   /* note: stop_data_collection() has already been called by the fork_prologue */
 473   // detach_experiment() can be called asynchronously
 474   // from anywhere, even from within a sigemt handler
 475   // via DBX detach.
 476   // Important: stop_data_collection() _must_ be called
 477   // before detach_experiment() is called.
 478   if (!COUNTERS_ENABLED ())
 479     return COL_ERROR_NONE;
 480   TprintfT (0, "hwprofile: detach_experiment()\n");
 481   if (SD_OFF != __collector_cas_32 (&sd_state, SD_OFF, SD_PENDING))
 482     return 0;
 483   // one and only one call should ever make it here here.
 484   if (hwc_mode == HWCMODE_ACTIVE)
 485     {
 486       TprintfT (0, "hwprofile: ERROR: stop_data_collection() should have been called before detach_experiment()\n");
 487       stop_data_collection ();
 488     }
 489
 490   // Assumption: The only calls to sigemt_handler
 491   // we should see at this point
 492   // will be those that were already in-flight before
 493   // stop_new_sigemts() was called.
 494   if (nthreads_in_sighandler > 0)
 495     {
 496       // sigemt handlers should see
 497       // SD_PENDING and should call real_detach_experiment()
 498       // when the last handler is finished.
 499       TprintfT (DBG_LT1, "hwprofile: detach in the middle of signal handler.\n");
 500       return 0;
 501     }
 502
 503   // If we get here, there should be no remaining
 504   // sigemt handlers.  However,  we don't really know
 505   // if there were ever any in flight, so call
 506   // real_detach_experiment() here:
 507   return real_detach_experiment (); // multiple calls to this OK
 508 }
 509
 510 static int
 511 real_detach_experiment (void)
 512 {
 513   /*multiple calls to this routine are OK.*/
 514   if (SD_PENDING != __collector_cas_32 (&sd_state, SD_PENDING, SD_COMPLETE))
 515     return 0;
 516   // only the first caller to this routine should get here.
 517   hwcdef_cnt = 0; /* since now deinstalled */
 518   hwcdef = NULL;
 519   set_hwc_mode (HWCMODE_OFF);
 520   if (SD_COMPLETE != __collector_cas_32 (&sd_state, SD_COMPLETE, SD_OFF))
 521     {
 522       TprintfT (0, "hwprofile: ERROR: unexpected sd_state in real_detach_experiment()\n");
 523       sd_state = SD_OFF;
 524     }
 525   hwprofile_tsd_key = COLLECTOR_TSD_INVALID_KEY;
 526   TprintfT (DBG_LT0, "hwprofile: real_detach_experiment() detached from experiment.\n");
 527   return 0;
 528 }
 529
 530 /*---------------------------------------------------------------------------*/
 531 /* Record counter values. */
 532
 533 /* <value> should already be adjusted to be "zero-based" (counting up from 0).*/
 534 static void
 535 collector_record_counter_internal (ucontext_t *ucp, int timecvt,
 536                                    ABST_type ABS_memop, hrtime_t time,
 537                                    unsigned tag, uint64_t value, uint64_t pc,
 538                                    uint64_t va, uint64_t latency,
 539                                    uint64_t data_source)
 540 {
 541   MHwcntr_packet pckt;
 542   CALL_UTIL (memset)(&pckt, 0, sizeof ( MHwcntr_packet));
 543   pckt.comm.tstamp = time;
 544   pckt.tag = tag;
 545   if (timecvt > 1)
 546     {
 547       if (HWCVAL_HAS_ERR (value))
 548         {
 549           value = HWCVAL_CLR_ERR (value);
 550           value *= timecvt;
 551           value = HWCVAL_SET_ERR (value);
 552         }
 553       else
 554         value *= timecvt;
 555     }
 556   pckt.interval = value;
 557   pckt.comm.type = HW_PCKT;
 558   pckt.comm.tsize = sizeof (Hwcntr_packet);
 559   TprintfT (DBG_LT4, "hwprofile: %llu sample %lld tag %u recorded\n",
 560             (unsigned long long) time, (long long) value, tag);
 561   if (ABS_memop == ABST_NOPC)
 562     ucp = &expr_nopc_uc;
 563   pckt.comm.frinfo = collector_interface->getFrameInfo (expr_hndl, pckt.comm.tstamp, FRINFO_FROM_UC, ucp);
 564   collector_interface->writeDataRecord (expr_hndl, (Common_packet*) & pckt);
 565 }
 566
 567 static void
 568 collector_record_counter (ucontext_t *ucp, int timecvt, ABST_type ABS_memop,
 569                           hrtime_t time, unsigned tag, uint64_t value)
 570 {
 571   collector_record_counter_internal (ucp, timecvt, ABS_memop, time, tag, value,
 572                                      HWCFUNCS_INVALID_U64, HWCFUNCS_INVALID_U64,
 573                                      HWCFUNCS_INVALID_U64, HWCFUNCS_INVALID_U64);
 574 }
 575
 576
 577 /*---------------------------------------------------------------------------*/
 578 /* Signal handlers */
 579
 580 /* SIGEMT -- relayed from libcpc, when the counter overflows */
 581
 582 /*   Generates the appropriate event or events, and resets the counters */
 583 static void
 584 collector_sigemt_handler (int sig, siginfo_t *si, void *puc)
 585 {
 586   int rc;
 587   hwc_event_t sample, lost_samples;
 588   if (sig != HWCFUNCS_SIGNAL)
 589     {
 590       TprintfT (0, "hwprofile: ERROR: %s: unexpected signal %d\n", "collector_sigemt_handler", sig);
 591       return;
 592     }
 593   if (!COUNTERS_ENABLED ())
 594     { /* apparently deinstalled */
 595       TprintfT (0, "hwprofile: WARNING: SIGEMT detected after close_experiment()\n");
 596       /* kills future sigemts since hwcdrv_sighlr_restart() not called */
 597       return;
 598     }
 599
 600   /* Typically, we expect HWC overflow signals to come from the kernel: si_code > 0.
 601    * On Linux, however, dbx might be "forwarding" a signal using tkill()/tgkill().
 602    * For more information on what si_code values can be expected on Linux, check:
 603    *     cmn_components/Collector_Interface/hwcdrv_pcl.c     hwcdrv_overflow()
 604    *     cmn_components/Collector_Interface/hwcdrv_perfctr.c hdrv_perfctr_overflow()
 605    */
 606   if (puc == NULL || si == NULL || (si->si_code <= 0 && si->si_code != SI_TKILL))
 607     {
 608       TprintfT (DBG_LT3, "hwprofile: collector_sigemt_handler SIG%02d\n", sig);
 609       if (old_sigemt_handler.sa_handler == SIG_DFL)
 610         __collector_SIGDFL_handler (HWCFUNCS_SIGNAL);
 611       else if (old_sigemt_handler.sa_handler != SIG_IGN &&
 612          old_sigemt_handler.sa_sigaction != &collector_sigemt_handler)
 613         {
 614           /* Redirect the signal to the previous signal handler */
 615           (old_sigemt_handler.sa_sigaction)(sig, si, puc);
 616           TprintfT (DBG_LT1, "hwprofile: collector_sigemt_handler SIG%02d redirected to original handler\n", sig);
 617         }
 618       return;
 619     }
 620   rc = get_hwc_driver ()->hwcdrv_overflow (si, &sample, &lost_samples);
 621   if (rc)
 622     {
 623       /* hwcdrv_sighlr_restart() should not be called */
 624       TprintfT (0, "hwprofile: ERROR: collector_sigemt_handler: hwcdrv_overflow() failed\n");
 625       return;
 626     }
 627
 628   if (hwc_mode == HWCMODE_ACTIVE)
 629     {
 630       /* record the event only if counters are active */
 631       /* The following has been copied from dispatcher.c */
 632 #if ARCH(SPARC)
 633       /* 23340823 signal handler third argument should point to a ucontext_t */
 634       /* Convert sigcontext to ucontext_t on sparc-Linux */
 635       ucontext_t uctxmem;
 636       struct sigcontext *sctx = (struct sigcontext*) puc;
 637       ucontext_t *uctx = &uctxmem;
 638       uctx->uc_link = NULL;
 639 #if WSIZE(32)
 640       uctx->uc_mcontext.gregs[REG_PC] = sctx->si_regs.pc;
 641       __collector_memcpy (&uctx->uc_mcontext.gregs[3],
 642                           sctx->si_regs.u_regs,
 643                           sizeof (sctx->si_regs.u_regs));
 644 #else
 645       uctx->uc_mcontext.mc_gregs[MC_PC] = sctx->sigc_regs.tpc;
 646       __collector_memcpy (&uctx->uc_mcontext.mc_gregs[3],
 647                           sctx->sigc_regs.u_regs,
 648                           sizeof (sctx->sigc_regs.u_regs));
 649 #endif /* WSIZE() */
 650 #else
 651       ucontext_t *uctx = (ucontext_t*) puc;
 652 #endif /* ARCH() */
 653
 654       for (int ii = 0; ii < hwcdef_cnt; ii++)
 655         if (lost_samples.ce_pic[ii])
 656           collector_record_counter (&expr_lostcounts_uc, hwcdef[ii]->timecvt,
 657                                     hwcdef[ii]->memop, lost_samples.ce_hrt,
 658                                     hwcdef[ii]->sort_order, lost_samples.ce_pic[ii]);
 659       for (int ii = 0; ii < hwcdef_cnt; ii++)
 660         if (sample.ce_pic[ii])
 661           collector_record_counter (uctx, hwcdef[ii]->timecvt,
 662                                     hwcdef[ii]->memop, sample.ce_hrt,
 663                                     hwcdef[ii]->sort_order, sample.ce_pic[ii]);
 664     }
 665   rc = get_hwc_driver ()->hwcdrv_sighlr_restart (NULL);
 666 }
 667 /*      SIGPROF -- not installed as handler, but
 668  *      __collector_ext_hwc_check: called by (SIGPROF) dispatcher.
 669  *       Periodical check of integrity of HWC count/signal mechanism,
 670  *       as required for various chip/system bugs/workarounds.
 671  */
 672 void
 673 __collector_ext_hwc_check (siginfo_t *info, ucontext_t *vcontext) { }
 674
 675 /*---------------------------------------------------------------------------*/
 676 int
 677 collector_sigemt_sigaction (const struct sigaction *nact,
 678                             struct sigaction *oact)
 679 {
 680   struct sigaction oact_check;
 681   /* Error codes and messages that refer to HWC are tricky.
 682    * E.g., HWC profiling might not even be on;  we might
 683    * encounter an error here simply because the user is
 684    * trying to set a handler for a signal that happens to
 685    * be HWCFUNCS_SIGNAL, which we aren't even using.
 686    */
 687   if (__collector_sigaction (HWCFUNCS_SIGNAL, NULL, &oact_check) != 0)
 688     {
 689       TprintfT (0, "hwprofile: ERROR: collector_sigemt_sigaction(): request to set handler for signal %d, but check on existing handler failed\n", HWCFUNCS_SIGNAL);
 690       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);
 691       return COL_ERROR_HWCINIT;
 692     }
 693
 694   if (oact_check.sa_sigaction == collector_sigemt_handler)
 695     {
 696       /* dispatcher is in place, so nact/oact apply to old_sigemt_handler */
 697       if (oact != NULL)
 698         {
 699           oact->sa_handler = old_sigemt_handler.sa_handler;
 700           oact->sa_mask = old_sigemt_handler.sa_mask;
 701           oact->sa_flags = old_sigemt_handler.sa_flags;
 702         }
 703       if (nact != NULL)
 704         {
 705           old_sigemt_handler.sa_handler = nact->sa_handler;
 706           old_sigemt_handler.sa_mask = nact->sa_mask;
 707           old_sigemt_handler.sa_flags = nact->sa_flags;
 708         }
 709       return COL_ERROR_NONE;
 710     }
 711   else /* no dispatcher in place, so just act like normal sigaction() */
 712     return __collector_sigaction (HWCFUNCS_SIGNAL, nact, oact);
 713 }
 714
 715 static void
 716 collector_hwc_ABORT (int errnum, const char *msg)
 717 {
 718   TprintfT (0, "hwprofile: collector_hwc_ABORT: [%d] %s\n", errnum, msg);
 719   if (hwc_mode == HWCMODE_ABORT) /* HWC collection already aborted! */
 720     return;
 721   set_hwc_mode (HWCMODE_ABORT); /* set global flag to disable handlers and indicate abort */
 722
 723   /* Write the error message to the experiment */
 724   collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s: errno=%d</event>\n",
 725                                  SP_JCMD_CERROR, COL_ERROR_HWCFAIL, msg, errnum);
 726
 727 #ifdef REAL_DEBUG
 728   abort ();
 729 #else
 730   TprintfT (0, "hwprofile: Continuing without HWC collection...\n");
 731 #endif
 732 }
 733
 734 static int
 735 start_data_collection (void)
 736 {
 737   hwc_mode_t old_mode = hwc_mode;
 738   if (!COUNTERS_ENABLED ())
 739     return COL_ERROR_NONE;
 740   TprintfT (0, "hwprofile: start_data_collection (hwc_mode=%d)\n", old_mode);
 741   switch (old_mode)
 742     {
 743     case HWCMODE_OFF:
 744       if (get_hwc_driver ()->hwcdrv_start ())
 745         {
 746           TprintfT (0, "hwprofile: ERROR: start_data_collection() failed in hwcdrv_start()\n");
 747           collector_interface->writeLog ("<event kind=\"%s\" id=\"%d\">%s: errno=%d</event>\n",
 748                                          SP_JCMD_CERROR, COL_ERROR_HWCFAIL,
 749                                          "start_data_collection()", errno);
 750           return COL_ERROR_HWCINIT;
 751         }
 752       set_hwc_mode (HWCMODE_ACTIVE); /* start handling events on signals */
 753       break;
 754     case HWCMODE_SUSPEND:
 755       if (get_hwc_driver ()->hwcdrv_lwp_resume ())
 756         {
 757           TprintfT (0, "hwprofile: ERROR: start_data_collection() failed in hwcdrv_lwp_resume()\n");
 758           /* ignore errors from lwp_resume() */
 759         }
 760       set_hwc_mode (HWCMODE_ACTIVE); /* start handling events on signals */
 761       break;
 762     default:
 763       TprintfT (0, "hwprofile: ERROR: start_data_collection() invalid mode\n");
 764       return COL_ERROR_HWCINIT;
 765     }
 766   return COL_ERROR_NONE;
 767 }
 768
 769 static int
 770 stop_data_collection (void)
 771 {
 772   hwc_mode_t old_mode = hwc_mode;
 773   if (!COUNTERS_ENABLED ())
 774     return COL_ERROR_NONE;
 775   TprintfT (0, "hwprofile: stop_data_collection (hwc_mode=%d)\n", old_mode);
 776   switch (old_mode)
 777     {
 778     case HWCMODE_SUSPEND:
 779       return COL_ERROR_NONE;
 780     case HWCMODE_ACTIVE:
 781       set_hwc_mode (HWCMODE_SUSPEND); /* stop handling signals */
 782       break;
 783     default:
 784       /* Don't change the mode, but attempt to suspend anyway... */
 785       break;
 786     }
 787
 788   if (get_hwc_driver ()->hwcdrv_lwp_suspend ())
 789     /* ignore errors from lwp_suspend() */
 790     TprintfT (0, "hwprofile: ERROR: stop_data_collection() failed in hwcdrv_lwp_suspend()\n");
 791
 792   /*
 793    * hwcdrv_lwp_suspend() cannot guarantee that all SIGEMTs will stop
 794    * but hwc_mode will prevent logging and counters will overflow once
 795    * then stay frozen.
 796    */
 797   /*   There may still be pending SIGEMTs so don't reset the SIG_DFL handler.
 798    */
 799   /* see comment in dispatcher.c */
 800   /* ret = __collector_sigaction( SIGEMT, &old_sigemt_handler, NULL ); */
 801   return COL_ERROR_NONE;
 802 }
 803
 804 /*---------------------------------------------------------------------------*/
 805
 806 /* utilities */
 807 static void
 808 set_hwc_mode (hwc_mode_t md)
 809 {
 810   TprintfT (DBG_LT1, "hwprofile: set_hwc_mode(%d)\n", md);
 811   hwc_mode = md;
 812 }
 813
 814 int
 815 __collector_ext_hwc_active ()
 816 {
 817   return (hwc_mode == HWCMODE_ACTIVE);
 818 }
 819
 820 static void
 821 hwclogwrite0 ()
 822 {
 823   collector_interface->writeLog ("<profdata fname=\"%s\"/>\n",
 824                                  module_interface.description);
 825   /* Record Hwcntr_packet description */
 826   Hwcntr_packet *pp = NULL;
 827   collector_interface->writeLog ("<profpckt kind=\"%d\" uname=\"" STXT ("Hardware counter profiling data") "\">\n", HW_PCKT);
 828   collector_interface->writeLog ("    <field name=\"LWPID\" uname=\"" STXT ("Lightweight process id") "\" offset=\"%d\" type=\"%s\"/>\n",
 829                                  &pp->comm.lwp_id, sizeof (pp->comm.lwp_id) == 4 ? "INT32" : "INT64");
 830   collector_interface->writeLog ("    <field name=\"THRID\" uname=\"" STXT ("Thread number") "\" offset=\"%d\" type=\"%s\"/>\n",
 831                                  &pp->comm.thr_id, sizeof (pp->comm.thr_id) == 4 ? "INT32" : "INT64");
 832   collector_interface->writeLog ("    <field name=\"CPUID\" uname=\"" STXT ("CPU id") "\" offset=\"%d\" type=\"%s\"/>\n",
 833                                  &pp->comm.cpu_id, sizeof (pp->comm.cpu_id) == 4 ? "INT32" : "INT64");
 834   collector_interface->writeLog ("    <field name=\"TSTAMP\" uname=\"" STXT ("High resolution timestamp") "\" offset=\"%d\" type=\"%s\"/>\n",
 835                                  &pp->comm.tstamp, sizeof (pp->comm.tstamp) == 4 ? "INT32" : "INT64");
 836   collector_interface->writeLog ("    <field name=\"FRINFO\" offset=\"%d\" type=\"%s\"/>\n",
 837                                  &pp->comm.frinfo, sizeof (pp->comm.frinfo) == 4 ? "INT32" : "INT64");
 838   collector_interface->writeLog ("    <field name=\"HWCTAG\" uname=\"" STXT ("Hardware counter index") "\" offset=\"%d\" type=\"%s\"/>\n",
 839                                  &pp->tag, sizeof (pp->tag) == 4 ? "INT32" : "INT64");
 840   collector_interface->writeLog ("    <field name=\"HWCINT\" uname=\"" STXT ("Hardware counter interval") "\" offset=\"%d\" type=\"%s\"/>\n",
 841                                  &pp->interval, sizeof (pp->interval) == 4 ? "INT32" : "INT64");
 842   collector_interface->writeLog ("</profpckt>\n");
 843   if (hwcdef_has_memspace)
 844     {
 845       /* Record MHwcntr_packet description */
 846       MHwcntr_packet *xpp = NULL;
 847       collector_interface->writeLog ("<profpckt kind=\"%d\" uname=\"" STXT ("Hardware counter profiling data") "\">\n", MHWC_PCKT);
 848       collector_interface->writeLog ("    <field name=\"LWPID\" uname=\"" STXT ("Lightweight process id") "\" offset=\"%d\" type=\"%s\"/>\n",
 849                                      &xpp->comm.lwp_id, sizeof (xpp->comm.lwp_id) == 4 ? "INT32" : "INT64");
 850       collector_interface->writeLog ("    <field name=\"THRID\" uname=\"" STXT ("Thread number") "\" offset=\"%d\" type=\"%s\"/>\n",
 851                                      &xpp->comm.thr_id, sizeof (xpp->comm.thr_id) == 4 ? "INT32" : "INT64");
 852       collector_interface->writeLog ("    <field name=\"CPUID\" uname=\"" STXT ("CPU id") "\" offset=\"%d\" type=\"%s\"/>\n",
 853                                      &xpp->comm.cpu_id, sizeof (xpp->comm.cpu_id) == 4 ? "INT32" : "INT64");
 854       collector_interface->writeLog ("    <field name=\"TSTAMP\" uname=\"" STXT ("High resolution timestamp") "\" offset=\"%d\" type=\"%s\"/>\n",
 855                                      &xpp->comm.tstamp, sizeof (xpp->comm.tstamp) == 4 ? "INT32" : "INT64");
 856       collector_interface->writeLog ("    <field name=\"FRINFO\" offset=\"%d\" type=\"%s\"/>\n",
 857                                      &xpp->comm.frinfo, sizeof (xpp->comm.frinfo) == 4 ? "INT32" : "INT64");
 858       collector_interface->writeLog ("    <field name=\"HWCTAG\" uname=\"" STXT ("Hardware counter index") "\" offset=\"%d\" type=\"%s\"/>\n",
 859                                      &xpp->tag, sizeof (xpp->tag) == 4 ? "INT32" : "INT64");
 860       collector_interface->writeLog ("    <field name=\"HWCINT\" uname=\"" STXT ("Hardware counter interval") "\" offset=\"%d\" type=\"%s\"/>\n",
 861                                      &xpp->interval, sizeof (xpp->interval) == 4 ? "INT32" : "INT64");
 862       collector_interface->writeLog ("    <field name=\"VADDR\" uname=\"" STXT ("Virtual address (data)") "\" offset=\"%d\" type=\"%s\"/>\n",
 863                                      &xpp->ea_vaddr, sizeof (xpp->ea_vaddr) == 4 ? "UINT32" : "UINT64");
 864       collector_interface->writeLog ("    <field name=\"PADDR\" uname=\"" STXT ("Physical address (data)") "\" offset=\"%d\" type=\"%s\"/>\n",
 865                                      &xpp->ea_paddr, sizeof (xpp->ea_paddr) == 4 ? "UINT32" : "UINT64");
 866       collector_interface->writeLog ("    <field name=\"VIRTPC\" uname=\"" STXT ("Virtual address (instruction)") "\" offset=\"%d\" type=\"%s\"/>\n",
 867                                      &xpp->pc_vaddr, sizeof (xpp->pc_vaddr) == 4 ? "UINT32" : "UINT64");
 868       collector_interface->writeLog ("    <field name=\"PHYSPC\" uname=\"" STXT ("Physical address (instruction)") "\" offset=\"%d\" type=\"%s\"/>\n",
 869                                      &xpp->pc_paddr, sizeof (xpp->pc_paddr) == 4 ? "UINT32" : "UINT64");
 870       collector_interface->writeLog ("    <field name=\"EA_PAGESIZE\" uname=\"" STXT ("Page size (data)") "\" offset=\"%d\" type=\"%s\"/>\n",
 871                                      &xpp->ea_pagesz, sizeof (xpp->ea_pagesz) == 4 ? "INT32" : "INT64");
 872       collector_interface->writeLog ("    <field name=\"PC_PAGESIZE\" uname=\"" STXT ("Page size (instruction)") "\" offset=\"%d\" type=\"%s\"/>\n",
 873                                      &xpp->pc_pagesz, sizeof (xpp->pc_pagesz) == 4 ? "INT32" : "INT64");
 874       collector_interface->writeLog ("    <field name=\"EA_LGRP\" uname=\"" STXT ("Page locality group (data)") "\" offset=\"%d\" type=\"%s\"/>\n",
 875                                      &xpp->ea_lgrp, sizeof (xpp->ea_lgrp) == 4 ? "INT32" : "INT64");
 876       collector_interface->writeLog ("    <field name=\"PC_LGRP\" uname=\"" STXT ("Page locality group (instruction)") "\" offset=\"%d\" type=\"%s\"/>\n",
 877                                      &xpp->pc_lgrp, sizeof (xpp->pc_lgrp) == 4 ? "INT32" : "INT64");
 878       collector_interface->writeLog ("    <field name=\"LWP_LGRP_HOME\" uname=\"" STXT ("LWP home lgroup id") "\" offset=\"%d\" type=\"%s\"/>\n",
 879                                      &xpp->lgrp_lwp, sizeof (xpp->lgrp_lwp) == 4 ? "INT32" : "INT64");
 880       collector_interface->writeLog ("    <field name=\"PS_LGRP_HOME\" uname=\"" STXT ("Process home lgroup id") "\" offset=\"%d\" type=\"%s\"/>\n",
 881                                      &xpp->lgrp_ps, sizeof (xpp->lgrp_ps) == 4 ? "INT32" : "INT64");
 882       collector_interface->writeLog ("    <field name=\"MEM_LAT\" uname=\"" STXT ("Memory Latency Cycles") "\" offset=\"%d\" type=\"%s\"/>\n",
 883                                      &xpp->latency, sizeof (xpp->latency) == 4 ? "INT32" : "INT64");
 884       collector_interface->writeLog ("    <field name=\"MEM_SRC\" uname=\"" STXT ("Memory Data Source") "\" offset=\"%d\" type=\"%s\"/>\n",
 885                                      &xpp->data_source, sizeof (xpp->data_source) == 4 ? "INT32" : "INT64");
 886       collector_interface->writeLog ("</profpckt>\n");
 887     }
 888 }
 889
 890 static void
 891 hwclogwrite (Hwcentry * ctr)
 892 {
 893   TprintfT (DBG_LT1, "hwprofile: writeLog(%s %u %s %d %u %d)\n",
 894             SP_JCMD_HW_COUNTER, cpcN_cpuver, ctr->name ? ctr->name : "NULL",
 895             ctr->val, ctr->sort_order, ctr->memop);
 896   collector_interface->writeLog ("<profile name=\"%s\"", SP_JCMD_HW_COUNTER);
 897   collector_interface->writeLog (" cpuver=\"%u\"", cpcN_cpuver);
 898   collector_interface->writeLog (" hwcname=\"%s\"", ctr->name);
 899   collector_interface->writeLog (" int_name=\"%s\"", ctr->int_name);
 900   collector_interface->writeLog (" interval=\"%d\"", ctr->val);
 901   collector_interface->writeLog (" tag=\"%u\"", ctr->sort_order);
 902   collector_interface->writeLog (" memop=\"%d\"", ctr->memop);
 903   collector_interface->writeLog ("/>\n");
 904 }