annotate VMS.c @ 71:5ff1631c26ed

working O3 version
author Merten Sach <msach@mailbox.tu-berlin.de>
date Mon, 30 May 2011 18:28:41 +0200
parents 11bfe9d136ed
children f6990e1ba998
rev   line source
Me@0 1 /*
Me@38 2 * Copyright 2010 OpenSourceStewardshipFoundation
Me@0 3 *
Me@0 4 * Licensed under BSD
Me@0 5 */
Me@0 6
Me@0 7 #include <stdio.h>
Me@0 8 #include <stdlib.h>
Me@50 9 #include <string.h>
Me@0 10 #include <malloc.h>
Me@50 11 #include <sys/time.h>
Me@0 12
Me@0 13 #include "VMS.h"
msach@71 14 #include "SwitchAnimators.h"
Me@0 15 #include "Queue_impl/BlockingQueue.h"
Me@38 16 #include "Histogram/Histogram.h"
Me@0 17
Me@0 18
Me@26 19 #define thdAttrs NULL
Me@26 20
Me@22 21 //===========================================================================
Me@22 22 void
Me@22 23 shutdownFn( void *dummy, VirtProcr *dummy2 );
Me@22 24
Me@31 25 SchedSlot **
Me@31 26 create_sched_slots();
Me@22 27
Me@28 28 void
Me@28 29 create_masterEnv();
Me@28 30
Me@28 31 void
Me@28 32 create_the_coreLoop_OS_threads();
Me@28 33
Me@50 34 MallocProlog *
Me@50 35 create_free_list();
Me@50 36
Me@53 37 void
Me@53 38 endOSThreadFn( void *initData, VirtProcr *animatingPr );
Me@50 39
Me@26 40 pthread_mutex_t suspendLock = PTHREAD_MUTEX_INITIALIZER;
Me@26 41 pthread_cond_t suspend_cond = PTHREAD_COND_INITIALIZER;
Me@26 42
Me@22 43 //===========================================================================
Me@22 44
Me@0 45 /*Setup has two phases:
Me@0 46 * 1) Semantic layer first calls init_VMS, which creates masterEnv, and puts
Me@8 47 * the master virt procr into the work-queue, ready for first "call"
Me@8 48 * 2) Semantic layer then does its own init, which creates the seed virt
Me@8 49 * procr inside the semantic layer, ready to schedule it when
Me@0 50 * asked by the first run of the masterLoop.
Me@0 51 *
Me@0 52 *This part is bit weird because VMS really wants to be "always there", and
Me@0 53 * have applications attach and detach.. for now, this VMS is part of
Me@0 54 * the app, so the VMS system starts up as part of running the app.
Me@0 55 *
Me@8 56 *The semantic layer is isolated from the VMS internals by making the
Me@8 57 * semantic layer do setup to a state that it's ready with its
Me@8 58 * initial virt procrs, ready to schedule them to slots when the masterLoop
Me@0 59 * asks. Without this pattern, the semantic layer's setup would
Me@8 60 * have to modify slots directly to assign the initial virt-procrs, and put
Me@31 61 * them into the readyToAnimateQ itself, breaking the isolation completely.
Me@0 62 *
Me@0 63 *
Me@8 64 *The semantic layer creates the initial virt procr(s), and adds its
Me@8 65 * own environment to masterEnv, and fills in the pointers to
Me@0 66 * the requestHandler and slaveScheduler plug-in functions
Me@8 67 */
Me@8 68
Me@8 69 /*This allocates VMS data structures, populates the master VMSProc,
Me@0 70 * and master environment, and returns the master environment to the semantic
Me@0 71 * layer.
Me@0 72 */
Me@8 73 void
Me@8 74 VMS__init()
Me@28 75 {
Me@28 76 create_masterEnv();
Me@28 77 create_the_coreLoop_OS_threads();
Me@28 78 }
Me@28 79
msach@71 80 #ifdef SEQUENTIAL
msach@71 81
Me@28 82 /*To initialize the sequential version, just don't create the threads
Me@28 83 */
Me@28 84 void
Me@28 85 VMS__init_Seq()
Me@28 86 {
Me@28 87 create_masterEnv();
Me@28 88 }
Me@28 89
msach@71 90 #endif
msach@71 91
Me@28 92 void
Me@28 93 create_masterEnv()
Me@31 94 { MasterEnv *masterEnv;
Me@55 95 VMSQueueStruc **readyToAnimateQs;
Me@31 96 int coreIdx;
Me@31 97 VirtProcr **masterVPs;
Me@31 98 SchedSlot ***allSchedSlots; //ptr to array of ptrs
Me@53 99
Me@53 100
Me@31 101 //Make the master env, which holds everything else
Me@1 102 _VMSMasterEnv = malloc( sizeof(MasterEnv) );
Me@53 103
Me@53 104 //Very first thing put into the master env is the free-list, seeded
Me@53 105 // with a massive initial chunk of memory.
Me@53 106 //After this, all other mallocs are VMS__malloc.
Me@53 107 _VMSMasterEnv->freeListHead = VMS_ext__create_free_list();
Me@53 108
Me@65 109
Me@65 110 //============================= MEASUREMENT STUFF ========================
Me@65 111 #ifdef MEAS__TIME_MALLOC
Me@68 112 _VMSMasterEnv->mallocTimeHist = makeFixedBinHistExt( 50, 0, 100,
Me@65 113 "malloc time hist");
Me@68 114 _VMSMasterEnv->freeTimeHist = makeFixedBinHistExt( 50, 0, 100,
Me@65 115 "free time hist");
Me@65 116 #endif
Me@68 117 #ifdef MEAS__TIME_PLUGIN
Me@68 118 _VMSMasterEnv->reqHdlrLowTimeHist = makeFixedBinHistExt( 50, 0, 10,
Me@68 119 "plugin low time hist");
Me@68 120 _VMSMasterEnv->reqHdlrHighTimeHist = makeFixedBinHistExt( 50, 0, 100,
Me@68 121 "plugin high time hist");
Me@68 122 #endif
Me@65 123 //========================================================================
Me@65 124
Me@53 125 //===================== Only VMS__malloc after this ====================
msach@69 126 masterEnv = (MasterEnv*)_VMSMasterEnv;
Me@31 127
Me@31 128 //Make a readyToAnimateQ for each core loop
Me@55 129 readyToAnimateQs = VMS__malloc( NUM_CORES * sizeof(VMSQueueStruc *) );
Me@53 130 masterVPs = VMS__malloc( NUM_CORES * sizeof(VirtProcr *) );
Me@0 131
Me@31 132 //One array for each core, 3 in array, core's masterVP scheds all
Me@53 133 allSchedSlots = VMS__malloc( NUM_CORES * sizeof(SchedSlot *) );
Me@0 134
Me@53 135 _VMSMasterEnv->numProcrsCreated = 0; //used by create procr
Me@31 136 for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
Me@53 137 {
Me@55 138 readyToAnimateQs[ coreIdx ] = makeVMSQ();
Me@31 139
Me@50 140 //Q: should give masterVP core-specific info as its init data?
Me@53 141 masterVPs[ coreIdx ] = VMS__create_procr( &masterLoop, masterEnv );
Me@31 142 masterVPs[ coreIdx ]->coreAnimatedBy = coreIdx;
Me@31 143 allSchedSlots[ coreIdx ] = create_sched_slots(); //makes for one core
Me@53 144 _VMSMasterEnv->numMasterInARow[ coreIdx ] = 0;
Me@55 145 _VMSMasterEnv->workStealingGates[ coreIdx ] = NULL;
Me@31 146 }
Me@31 147 _VMSMasterEnv->readyToAnimateQs = readyToAnimateQs;
Me@31 148 _VMSMasterEnv->masterVPs = masterVPs;
Me@50 149 _VMSMasterEnv->masterLock = UNLOCKED;
Me@31 150 _VMSMasterEnv->allSchedSlots = allSchedSlots;
Me@55 151 _VMSMasterEnv->workStealingLock = UNLOCKED;
Me@28 152
Me@12 153
Me@31 154 //Aug 19, 2010: no longer need to place initial masterVP into queue
Me@31 155 // because coreLoop now controls -- animates its masterVP when no work
Me@31 156
Me@30 157
Me@50 158 //============================= MEASUREMENT STUFF ========================
Me@50 159 #ifdef STATS__TURN_ON_PROBES
Me@50 160 _VMSMasterEnv->dynIntervalProbesInfo =
msach@69 161 makePrivDynArrayOfSize( (void***)&(_VMSMasterEnv->intervalProbes), 200);
Me@30 162
Me@53 163 _VMSMasterEnv->probeNameHashTbl = makeHashTable( 1000, &VMS__free );
Me@53 164
Me@53 165 //put creation time directly into master env, for fast retrieval
Me@50 166 struct timeval timeStamp;
Me@50 167 gettimeofday( &(timeStamp), NULL);
Me@50 168 _VMSMasterEnv->createPtInSecs =
Me@50 169 timeStamp.tv_sec +(timeStamp.tv_usec/1000000.0);
Me@50 170 #endif
Me@65 171 #ifdef MEAS__TIME_MASTER_LOCK
Me@65 172 _VMSMasterEnv->masterLockLowTimeHist = makeFixedBinHist( 50, 0, 2,
Me@65 173 "master lock low time hist");
Me@68 174 _VMSMasterEnv->masterLockHighTimeHist = makeFixedBinHist( 50, 0, 100,
Me@65 175 "master lock high time hist");
Me@65 176 #endif
Me@68 177
Me@68 178 MakeTheMeasHists
Me@50 179 //========================================================================
Me@38 180
Me@0 181 }
Me@0 182
Me@31 183 SchedSlot **
Me@31 184 create_sched_slots()
Me@31 185 { SchedSlot **schedSlots;
Me@0 186 int i;
Me@0 187
Me@53 188 schedSlots = VMS__malloc( NUM_SCHED_SLOTS * sizeof(SchedSlot *) );
Me@8 189
Me@1 190 for( i = 0; i < NUM_SCHED_SLOTS; i++ )
Me@0 191 {
Me@53 192 schedSlots[i] = VMS__malloc( sizeof(SchedSlot) );
Me@8 193
Me@1 194 //Set state to mean "handling requests done, slot needs filling"
Me@8 195 schedSlots[i]->workIsDone = FALSE;
Me@8 196 schedSlots[i]->needsProcrAssigned = TRUE;
Me@0 197 }
Me@31 198 return schedSlots;
Me@31 199 }
Me@31 200
Me@31 201
Me@31 202 void
Me@31 203 freeSchedSlots( SchedSlot **schedSlots )
Me@31 204 { int i;
Me@31 205 for( i = 0; i < NUM_SCHED_SLOTS; i++ )
Me@31 206 {
Me@53 207 VMS__free( schedSlots[i] );
Me@31 208 }
Me@53 209 VMS__free( schedSlots );
Me@0 210 }
Me@0 211
Me@8 212
Me@28 213 void
Me@28 214 create_the_coreLoop_OS_threads()
Me@28 215 {
Me@28 216 //========================================================================
Me@28 217 // Create the Threads
Me@28 218 int coreIdx, retCode;
Me@28 219
Me@28 220 //Need the threads to be created suspended, and wait for a signal
Me@28 221 // before proceeding -- gives time after creating to initialize other
Me@28 222 // stuff before the coreLoops set off.
Me@28 223 _VMSMasterEnv->setupComplete = 0;
Me@28 224
Me@28 225 //Make the threads that animate the core loops
Me@28 226 for( coreIdx=0; coreIdx < NUM_CORES; coreIdx++ )
Me@53 227 { coreLoopThdParams[coreIdx] = VMS__malloc( sizeof(ThdParams) );
Me@28 228 coreLoopThdParams[coreIdx]->coreNum = coreIdx;
Me@28 229
Me@28 230 retCode =
Me@28 231 pthread_create( &(coreLoopThdHandles[coreIdx]),
Me@28 232 thdAttrs,
Me@28 233 &coreLoop,
Me@28 234 (void *)(coreLoopThdParams[coreIdx]) );
Me@50 235 if(retCode){printf("ERROR creating thread: %d\n", retCode); exit(1);}
Me@28 236 }
Me@28 237 }
Me@28 238
Me@0 239 /*Semantic layer calls this when it want the system to start running..
Me@0 240 *
Me@24 241 *This starts the core loops running then waits for them to exit.
Me@0 242 */
Me@12 243 void
Me@24 244 VMS__start_the_work_then_wait_until_done()
Me@12 245 { int coreIdx;
Me@24 246 //Start the core loops running
Me@25 247
Me@25 248 //tell the core loop threads that setup is complete
Me@25 249 //get lock, to lock out any threads still starting up -- they'll see
Me@25 250 // that setupComplete is true before entering while loop, and so never
Me@25 251 // wait on the condition
Me@26 252 pthread_mutex_lock( &suspendLock );
Me@25 253 _VMSMasterEnv->setupComplete = 1;
Me@26 254 pthread_mutex_unlock( &suspendLock );
Me@26 255 pthread_cond_broadcast( &suspend_cond );
Me@25 256
Me@25 257
Me@24 258 //wait for all to complete
Me@8 259 for( coreIdx=0; coreIdx < NUM_CORES; coreIdx++ )
Me@8 260 {
Me@25 261 pthread_join( coreLoopThdHandles[coreIdx], NULL );
Me@24 262 }
Me@25 263
Me@24 264 //NOTE: do not clean up VMS env here -- semantic layer has to have
Me@24 265 // a chance to clean up its environment first, then do a call to free
Me@24 266 // the Master env and rest of VMS locations
Me@8 267 }
Me@0 268
msach@71 269 #ifdef SEQUENTIAL
Me@28 270 /*Only difference between version with an OS thread pinned to each core and
Me@28 271 * the sequential version of VMS is VMS__init_Seq, this, and coreLoop_Seq.
Me@28 272 */
Me@28 273 void
Me@28 274 VMS__start_the_work_then_wait_until_done_Seq()
Me@28 275 {
Me@28 276 //Instead of un-suspending threads, just call the one and only
Me@28 277 // core loop (sequential version), in the main thread.
Me@28 278 coreLoop_Seq( NULL );
Me@28 279
Me@28 280 }
msach@71 281 #endif
Me@0 282
Me@8 283 /*Create stack, then create __cdecl structure on it and put initialData and
Me@8 284 * pointer to the new structure instance into the parameter positions on
Me@8 285 * the stack
Me@8 286 *Then put function pointer into nextInstrPt -- the stack is setup in std
Me@8 287 * call structure, so jumping to function ptr is same as a GCC generated
Me@8 288 * function call
Me@8 289 *No need to save registers on old stack frame, because there's no old
Me@8 290 * animator state to return to --
Me@8 291 *
Me@8 292 */
Me@50 293 inline VirtProcr *
Me@50 294 create_procr_helper( VirtProcr *newPr, VirtProcrFnPtr fnPtr,
Me@50 295 void *initialData, char *stackLocs )
Me@50 296 {
Me@50 297 char *stackPtr;
Me@8 298
Me@53 299 newPr->startOfStack = stackLocs;
Me@53 300 newPr->procrID = _VMSMasterEnv->numProcrsCreated++;
Me@53 301 newPr->nextInstrPt = fnPtr;
Me@53 302 newPr->initialData = initialData;
Me@53 303 newPr->requests = NULL;
Me@53 304 newPr->schedSlot = NULL;
Me@8 305
Me@14 306 //fnPtr takes two params -- void *initData & void *animProcr
Me@8 307 //alloc stack locations, make stackPtr be the highest addr minus room
Me@14 308 // for 2 params + return addr. Return addr (NULL) is in loc pointed to
Me@14 309 // by stackPtr, initData at stackPtr + 4 bytes, animatingPr just above
Me@22 310 stackPtr = ( (char *)stackLocs + VIRT_PROCR_STACK_SIZE - 0x10 );
Me@50 311
Me@8 312 //setup __cdecl on stack -- coreloop will switch to stackPtr before jmp
Me@22 313 *( (int *)stackPtr + 2 ) = (int) newPr; //rightmost param -- 32bit pointer
Me@14 314 *( (int *)stackPtr + 1 ) = (int) initialData; //next param to left
Me@8 315 newPr->stackPtr = stackPtr; //core loop will switch to this, then
Me@8 316 newPr->framePtr = stackPtr; //suspend loop will save new stack & frame ptr
Me@8 317
Me@50 318 //============================= MEASUREMENT STUFF ========================
Me@50 319 #ifdef STATS__TURN_ON_PROBES
Me@50 320 struct timeval timeStamp;
Me@50 321 gettimeofday( &(timeStamp), NULL);
Me@54 322 newPr->createPtInSecs = timeStamp.tv_sec +(timeStamp.tv_usec/1000000.0) -
Me@54 323 _VMSMasterEnv->createPtInSecs;
Me@50 324 #endif
Me@50 325 //========================================================================
Me@50 326
Me@8 327 return newPr;
Me@8 328 }
Me@8 329
Me@50 330 inline VirtProcr *
Me@50 331 VMS__create_procr( VirtProcrFnPtr fnPtr, void *initialData )
Me@50 332 { VirtProcr *newPr;
Me@50 333 char *stackLocs;
Me@50 334
Me@50 335 newPr = VMS__malloc( sizeof(VirtProcr) );
Me@50 336 stackLocs = VMS__malloc( VIRT_PROCR_STACK_SIZE );
Me@50 337 if( stackLocs == 0 )
Me@50 338 { perror("VMS__malloc stack"); exit(1); }
Me@50 339
Me@50 340 return create_procr_helper( newPr, fnPtr, initialData, stackLocs );
Me@50 341 }
Me@50 342
Me@50 343 /* "ext" designates that it's for use outside the VMS system -- should only
Me@50 344 * be called from main thread or other thread -- never from code animated by
Me@50 345 * a VMS virtual processor.
Me@50 346 */
Me@50 347 inline VirtProcr *
Me@50 348 VMS_ext__create_procr( VirtProcrFnPtr fnPtr, void *initialData )
Me@50 349 { VirtProcr *newPr;
Me@50 350 char *stackLocs;
Me@50 351
Me@50 352 newPr = malloc( sizeof(VirtProcr) );
Me@50 353 stackLocs = malloc( VIRT_PROCR_STACK_SIZE );
Me@50 354 if( stackLocs == 0 )
Me@50 355 { perror("malloc stack"); exit(1); }
Me@50 356
Me@50 357 return create_procr_helper( newPr, fnPtr, initialData, stackLocs );
Me@50 358 }
Me@50 359
Me@8 360
Me@64 361 /*Anticipating multi-tasking
Me@64 362 */
Me@64 363 void *
Me@64 364 VMS__give_sem_env_for( VirtProcr *animPr )
Me@64 365 {
Me@64 366 return _VMSMasterEnv->semanticEnv;
Me@64 367 }
Me@64 368 //===========================================================================
Me@26 369 /*there is a label inside this function -- save the addr of this label in
Me@0 370 * the callingPr struc, as the pick-up point from which to start the next
Me@0 371 * work-unit for that procr. If turns out have to save registers, then
Me@0 372 * save them in the procr struc too. Then do assembly jump to the CoreLoop's
Me@0 373 * "done with work-unit" label. The procr struc is in the request in the
Me@0 374 * slave that animated the just-ended work-unit, so all the state is saved
Me@0 375 * there, and will get passed along, inside the request handler, to the
Me@0 376 * next work-unit for that procr.
Me@0 377 */
Me@8 378 void
Me@38 379 VMS__suspend_procr( VirtProcr *animatingPr )
Me@55 380 {
Me@0 381
Me@14 382 //The request to master will cause this suspended virt procr to get
Me@14 383 // scheduled again at some future point -- to resume, core loop jumps
Me@14 384 // to the resume point (below), which causes restore of saved regs and
Me@14 385 // "return" from this call.
msach@71 386 //animatingPr->nextInstrPt = &&ResumePt;
Me@1 387
Me@1 388 //return ownership of the virt procr and sched slot to Master virt pr
Me@38 389 animatingPr->schedSlot->workIsDone = TRUE;
Me@1 390
Me@41 391 //=========================== Measurement stuff ========================
Me@38 392 #ifdef MEAS__TIME_STAMP_SUSP
Me@41 393 //record time stamp: compare to time-stamp recorded below
Me@38 394 saveLowTimeStampCountInto( animatingPr->preSuspTSCLow );
Me@38 395 #endif
Me@41 396 //=======================================================================
Me@38 397
msach@71 398 switchToCoreLoop(animatingPr);
msach@71 399 flushRegisters();
Me@55 400
Me@55 401 //=======================================================================
msach@71 402
Me@38 403 #ifdef MEAS__TIME_STAMP_SUSP
Me@41 404 //NOTE: only take low part of count -- do sanity check when take diff
Me@38 405 saveLowTimeStampCountInto( animatingPr->postSuspTSCLow );
Me@38 406 #endif
Me@38 407
Me@0 408 return;
Me@0 409 }
Me@0 410
Me@22 411
Me@22 412
Me@50 413 /*For this implementation of VMS, it may not make much sense to have the
Me@50 414 * system of requests for creating a new processor done this way.. but over
Me@50 415 * the scope of single-master, multi-master, mult-tasking, OS-implementing,
Me@50 416 * distributed-memory, and so on, this gives VMS implementation a chance to
Me@50 417 * do stuff before suspend, in the AppVP, and in the Master before the plugin
Me@50 418 * is called, as well as in the lang-lib before this is called, and in the
Me@50 419 * plugin. So, this gives both VMS and language implementations a chance to
Me@50 420 * intercept at various points and do order-dependent stuff.
Me@50 421 *Having a standard VMSNewPrReqData struc allows the language to create and
Me@50 422 * free the struc, while VMS knows how to get the newPr if it wants it, and
Me@50 423 * it lets the lang have lang-specific data related to creation transported
Me@50 424 * to the plugin.
Me@50 425 */
Me@50 426 void
Me@50 427 VMS__send_create_procr_req( void *semReqData, VirtProcr *reqstingPr )
Me@50 428 { VMSReqst req;
Me@50 429
Me@50 430 req.reqType = createReq;
Me@50 431 req.semReqData = semReqData;
Me@50 432 req.nextReqst = reqstingPr->requests;
Me@50 433 reqstingPr->requests = &req;
Me@50 434
Me@50 435 VMS__suspend_procr( reqstingPr );
Me@50 436 }
Me@50 437
Me@22 438
Me@38 439 /*
Me@22 440 *This adds a request to dissipate, then suspends the processor so that the
Me@22 441 * request handler will receive the request. The request handler is what
Me@22 442 * does the work of freeing memory and removing the processor from the
Me@22 443 * semantic environment's data structures.
Me@22 444 *The request handler also is what figures out when to shutdown the VMS
Me@22 445 * system -- which causes all the core loop threads to die, and returns from
Me@22 446 * the call that started up VMS to perform the work.
Me@22 447 *
Me@22 448 *This form is a bit misleading to understand if one is trying to figure out
Me@22 449 * how VMS works -- it looks like a normal function call, but inside it
Me@22 450 * sends a request to the request handler and suspends the processor, which
Me@22 451 * jumps out of the VMS__dissipate_procr function, and out of all nestings
Me@22 452 * above it, transferring the work of dissipating to the request handler,
Me@22 453 * which then does the actual work -- causing the processor that animated
Me@22 454 * the call of this function to disappear and the "hanging" state of this
Me@22 455 * function to just poof into thin air -- the virtual processor's trace
Me@22 456 * never returns from this call, but instead the virtual processor's trace
Me@22 457 * gets suspended in this call and all the virt processor's state disap-
Me@22 458 * pears -- making that suspend the last thing in the virt procr's trace.
Me@8 459 */
Me@8 460 void
Me@53 461 VMS__send_dissipate_req( VirtProcr *procrToDissipate )
Me@50 462 { VMSReqst req;
Me@22 463
Me@50 464 req.reqType = dissipate;
Me@50 465 req.nextReqst = procrToDissipate->requests;
Me@50 466 procrToDissipate->requests = &req;
Me@50 467
Me@22 468 VMS__suspend_procr( procrToDissipate );
Me@50 469 }
Me@50 470
Me@50 471
Me@50 472 /* "ext" designates that it's for use outside the VMS system -- should only
Me@50 473 * be called from main thread or other thread -- never from code animated by
Me@50 474 * a VMS virtual processor.
Me@50 475 *
Me@50 476 *Use this version to dissipate VPs created outside the VMS system.
Me@50 477 */
Me@50 478 void
Me@50 479 VMS_ext__dissipate_procr( VirtProcr *procrToDissipate )
Me@50 480 {
Me@50 481 //NOTE: initialData was given to the processor, so should either have
Me@50 482 // been alloc'd with VMS__malloc, or freed by the level above animPr.
Me@50 483 //So, all that's left to free here is the stack and the VirtProcr struc
Me@50 484 // itself
Me@50 485 //Note, should not stack-allocate initial data -- no guarantee, in
Me@50 486 // general that creating processor will outlive ones it creates.
Me@50 487 free( procrToDissipate->startOfStack );
Me@50 488 free( procrToDissipate );
Me@50 489 }
Me@50 490
Me@22 491
Me@22 492
Me@53 493 /*This call's name indicates that request is malloc'd -- so req handler
Me@53 494 * has to free any extra requests tacked on before a send, using this.
Me@53 495 *
Me@53 496 * This inserts the semantic-layer's request data into standard VMS carrier
Me@53 497 * request data-struct that is mallocd. The sem request doesn't need to
Me@53 498 * be malloc'd if this is called inside the same call chain before the
Me@53 499 * send of the last request is called.
Me@53 500 *
Me@53 501 *The request handler has to call VMS__free_VMSReq for any of these
Me@22 502 */
Me@22 503 inline void
Me@53 504 VMS__add_sem_request_in_mallocd_VMSReqst( void *semReqData,
Me@53 505 VirtProcr *callingPr )
Me@53 506 { VMSReqst *req;
Me@22 507
Me@53 508 req = VMS__malloc( sizeof(VMSReqst) );
Me@53 509 req->reqType = semantic;
Me@53 510 req->semReqData = semReqData;
Me@53 511 req->nextReqst = callingPr->requests;
Me@53 512 callingPr->requests = req;
Me@22 513 }
Me@22 514
Me@50 515 /*This inserts the semantic-layer's request data into standard VMS carrier
Me@50 516 * request data-struct is allocated on stack of this call & ptr to it sent
Me@50 517 * to plugin
Me@50 518 *Then it does suspend, to cause request to be sent.
Me@50 519 */
Me@50 520 inline void
Me@50 521 VMS__send_sem_request( void *semReqData, VirtProcr *callingPr )
Me@50 522 { VMSReqst req;
Me@22 523
Me@50 524 req.reqType = semantic;
Me@50 525 req.semReqData = semReqData;
Me@50 526 req.nextReqst = callingPr->requests;
Me@50 527 callingPr->requests = &req;
Me@50 528
Me@50 529 VMS__suspend_procr( callingPr );
Me@50 530 }
Me@50 531
Me@50 532
Me@50 533 inline void
Me@50 534 VMS__send_VMSSem_request( void *semReqData, VirtProcr *callingPr )
Me@50 535 { VMSReqst req;
Me@50 536
Me@50 537 req.reqType = VMSSemantic;
Me@50 538 req.semReqData = semReqData;
Me@50 539 req.nextReqst = callingPr->requests; //gab any other preceeding
Me@50 540 callingPr->requests = &req;
Me@50 541
Me@50 542 VMS__suspend_procr( callingPr );
Me@50 543 }
Me@50 544
Me@50 545
Me@50 546 /*
Me@38 547 */
Me@24 548 VMSReqst *
Me@50 549 VMS__take_next_request_out_of( VirtProcr *procrWithReq )
Me@31 550 { VMSReqst *req;
Me@31 551
Me@31 552 req = procrWithReq->requests;
Me@38 553 if( req == NULL ) return NULL;
Me@31 554
Me@31 555 procrWithReq->requests = procrWithReq->requests->nextReqst;
Me@50 556 return req;
Me@24 557 }
Me@22 558
Me@24 559
Me@24 560 inline void *
Me@24 561 VMS__take_sem_reqst_from( VMSReqst *req )
Me@24 562 {
Me@24 563 return req->semReqData;
Me@24 564 }
Me@24 565
Me@24 566
Me@24 567
Me@50 568 /* This is for OS requests and VMS infrastructure requests, such as to create
Me@50 569 * a probe -- a probe is inside the heart of VMS-core, it's not part of any
Me@50 570 * language -- but it's also a semantic thing that's triggered from and used
Me@50 571 * in the application.. so it crosses abstractions.. so, need some special
Me@50 572 * pattern here for handling such requests.
Me@52 573 * Doing this just like it were a second language sharing VMS-core.
Me@52 574 *
Me@50 575 * This is called from the language's request handler when it sees a request
Me@50 576 * of type VMSSemReq
Me@52 577 *
Me@52 578 * TODO: Later change this, to give probes their own separate plugin & have
Me@52 579 * VMS-core steer the request to appropriate plugin
Me@52 580 * Do the same for OS calls -- look later at it..
Me@50 581 */
Me@50 582 void inline
Me@50 583 VMS__handle_VMSSemReq( VMSReqst *req, VirtProcr *requestingPr, void *semEnv,
Me@50 584 ResumePrFnPtr resumePrFnPtr )
Me@50 585 { VMSSemReq *semReq;
Me@50 586 IntervalProbe *newProbe;
Me@50 587 int32 nameLen;
Me@24 588
Me@50 589 semReq = req->semReqData;
Me@24 590
Me@50 591 newProbe = VMS__malloc( sizeof(IntervalProbe) );
Me@65 592 newProbe->nameStr = VMS__strDup( semReq->nameStr );
Me@50 593 newProbe->hist = NULL;
Me@50 594 newProbe->schedChoiceWasRecorded = FALSE;
Me@53 595
Me@53 596 //This runs in masterVP, so no race-condition worries
Me@50 597 newProbe->probeID =
Me@50 598 addToDynArray( newProbe, _VMSMasterEnv->dynIntervalProbesInfo );
Me@50 599
Me@53 600 requestingPr->dataRetFromReq = newProbe;
Me@50 601
Me@50 602 (*resumePrFnPtr)( requestingPr, semEnv );
Me@22 603 }
Me@22 604
Me@22 605
Me@22 606
Me@24 607 /*This must be called by the request handler plugin -- it cannot be called
Me@24 608 * from the semantic library "dissipate processor" function -- instead, the
Me@50 609 * semantic layer has to generate a request, and the plug-in calls this
Me@24 610 * function.
Me@24 611 *The reason is that this frees the virtual processor's stack -- which is
Me@24 612 * still in use inside semantic library calls!
Me@24 613 *
Me@24 614 *This frees or recycles all the state owned by and comprising the VMS
Me@24 615 * portion of the animating virtual procr. The request handler must first
Me@24 616 * free any semantic data created for the processor that didn't use the
Me@24 617 * VMS_malloc mechanism. Then it calls this, which first asks the malloc
Me@24 618 * system to disown any state that did use VMS_malloc, and then frees the
Me@24 619 * statck and the processor-struct itself.
Me@24 620 *If the dissipated processor is the sole (remaining) owner of VMS__malloc'd
Me@24 621 * state, then that state gets freed (or sent to recycling) as a side-effect
Me@24 622 * of dis-owning it.
Me@24 623 */
Me@24 624 void
Me@53 625 VMS__dissipate_procr( VirtProcr *animatingPr )
Me@24 626 {
Me@24 627 //dis-own all locations owned by this processor, causing to be freed
Me@24 628 // any locations that it is (was) sole owner of
Me@29 629 //TODO: implement VMS__malloc system, including "give up ownership"
Me@24 630
Me@24 631
Me@24 632 //NOTE: initialData was given to the processor, so should either have
Me@24 633 // been alloc'd with VMS__malloc, or freed by the level above animPr.
Me@24 634 //So, all that's left to free here is the stack and the VirtProcr struc
Me@24 635 // itself
Me@50 636 //Note, should not stack-allocate initial data -- no guarantee, in
Me@50 637 // general that creating processor will outlive ones it creates.
Me@50 638 VMS__free( animatingPr->startOfStack );
Me@50 639 VMS__free( animatingPr );
Me@24 640 }
Me@24 641
Me@24 642
Me@53 643 //TODO: look at architecting cleanest separation between request handler
Me@29 644 // and master loop, for dissipate, create, shutdown, and other non-semantic
Me@29 645 // requests. Issue is chain: one removes requests from AppVP, one dispatches
Me@29 646 // on type of request, and one handles each type.. but some types require
Me@29 647 // action from both request handler and master loop -- maybe just give the
Me@29 648 // request handler calls like: VMS__handle_X_request_type
Me@24 649
Me@29 650
Me@29 651 /*This is called by the semantic layer's request handler when it decides its
Me@29 652 * time to shut down the VMS system. Calling this causes the core loop OS
Me@29 653 * threads to exit, which unblocks the entry-point function that started up
Me@29 654 * VMS, and allows it to grab the result and return to the original single-
Me@29 655 * threaded application.
Me@22 656 *
Me@29 657 *The _VMSMasterEnv is needed by this shut down function, so the create-seed-
Me@29 658 * and-wait function has to free a bunch of stuff after it detects the
Me@29 659 * threads have all died: the masterEnv, the thread-related locations,
Me@29 660 * masterVP any AppVPs that might still be allocated and sitting in the
Me@29 661 * semantic environment, or have been orphaned in the _VMSWorkQ.
Me@29 662 *
Me@53 663 *NOTE: the semantic plug-in is expected to use VMS__malloc to get all the
Me@29 664 * locations it needs, and give ownership to masterVP. Then, they will be
Me@53 665 * automatically freed.
Me@22 666 *
Me@29 667 *In here,create one core-loop shut-down processor for each core loop and put
Me@31 668 * them all directly into the readyToAnimateQ.
Me@29 669 *Note, this function can ONLY be called after the semantic environment no
Me@29 670 * longer cares if AppVPs get animated after the point this is called. In
Me@29 671 * other words, this can be used as an abort, or else it should only be
Me@29 672 * called when all AppVPs have finished dissipate requests -- only at that
Me@29 673 * point is it sure that all results have completed.
Me@22 674 */
Me@22 675 void
Me@53 676 VMS__shutdown()
Me@8 677 { int coreIdx;
Me@14 678 VirtProcr *shutDownPr;
Me@22 679
Me@29 680 //create the shutdown processors, one for each core loop -- put them
Me@31 681 // directly into the Q -- each core will die when gets one
Me@8 682 for( coreIdx=0; coreIdx < NUM_CORES; coreIdx++ )
Me@50 683 { //Note, this is running in the master
Me@29 684 shutDownPr = VMS__create_procr( &endOSThreadFn, NULL );
Me@55 685 writeVMSQ( shutDownPr, _VMSMasterEnv->readyToAnimateQs[coreIdx] );
Me@8 686 }
Me@22 687
Me@12 688 }
Me@12 689
Me@12 690
Me@29 691 /*Am trying to be cute, avoiding IF statement in coreLoop that checks for
Me@29 692 * a special shutdown procr. Ended up with extra-complex shutdown sequence.
Me@29 693 *This function has the sole purpose of setting the stack and framePtr
Me@29 694 * to the coreLoop's stack and framePtr.. it does that then jumps to the
Me@29 695 * core loop's shutdown point -- might be able to just call Pthread_exit
Me@30 696 * from here, but am going back to the pthread's stack and setting everything
Me@29 697 * up just as if it never jumped out, before calling pthread_exit.
Me@29 698 *The end-point of core loop will free the stack and so forth of the
Me@29 699 * processor that animates this function, (this fn is transfering the
Me@29 700 * animator of the AppVP that is in turn animating this function over
Me@29 701 * to core loop function -- note that this slices out a level of virtual
Me@29 702 * processors).
Me@29 703 */
Me@29 704 void
Me@29 705 endOSThreadFn( void *initData, VirtProcr *animatingPr )
msach@71 706 {
msach@71 707 asmTerminateCoreLoop(animatingPr);
Me@29 708 }
Me@29 709
Me@29 710
Me@53 711 /*This is called from the startup & shutdown
Me@24 712 */
Me@24 713 void
Me@53 714 VMS__cleanup_at_end_of_shutdown()
Me@31 715 {
Me@55 716 VMSQueueStruc **readyToAnimateQs;
Me@31 717 int coreIdx;
Me@31 718 VirtProcr **masterVPs;
Me@31 719 SchedSlot ***allSchedSlots; //ptr to array of ptrs
Me@31 720
Me@65 721 //Before getting rid of everything, print out any measurements made
msach@69 722 forAllInDynArrayDo( _VMSMasterEnv->measHistsInfo, (DynArrayFnPtr)&printHist );
Me@68 723 //forAllInDynArrayDo( _VMSMasterEnv->measHistsInfo, &freeHistExt );
Me@65 724 #ifdef MEAS__TIME_PLUGIN
Me@68 725 printHist( _VMSMasterEnv->reqHdlrLowTimeHist );
Me@68 726 printHist( _VMSMasterEnv->reqHdlrHighTimeHist );
Me@68 727 freeHistExt( _VMSMasterEnv->reqHdlrLowTimeHist );
Me@68 728 freeHistExt( _VMSMasterEnv->reqHdlrHighTimeHist );
Me@65 729 #endif
Me@65 730 #ifdef MEAS__TIME_MALLOC
Me@65 731 printHist( _VMSMasterEnv->mallocTimeHist );
Me@65 732 printHist( _VMSMasterEnv->freeTimeHist );
Me@65 733 freeHistExt( _VMSMasterEnv->mallocTimeHist );
Me@65 734 freeHistExt( _VMSMasterEnv->freeTimeHist );
Me@65 735 #endif
Me@65 736 #ifdef MEAS__TIME_MASTER_LOCK
Me@65 737 printHist( _VMSMasterEnv->masterLockLowTimeHist );
Me@65 738 printHist( _VMSMasterEnv->masterLockHighTimeHist );
Me@65 739 #endif
Me@65 740 #ifdef MEAS__TIME_MASTER
Me@65 741 printHist( _VMSMasterEnv->pluginTimeHist );
Me@65 742 for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
Me@65 743 {
Me@65 744 freeVMSQ( readyToAnimateQs[ coreIdx ] );
Me@65 745 //master VPs were created external to VMS, so use external free
Me@65 746 VMS__dissipate_procr( masterVPs[ coreIdx ] );
Me@65 747
Me@65 748 freeSchedSlots( allSchedSlots[ coreIdx ] );
Me@65 749 }
Me@65 750 #endif
Me@65 751 #ifdef MEAS__TIME_STAMP_SUSP
Me@65 752 printHist( _VMSMasterEnv->pluginTimeHist );
Me@65 753 for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
Me@65 754 {
Me@65 755 freeVMSQ( readyToAnimateQs[ coreIdx ] );
Me@65 756 //master VPs were created external to VMS, so use external free
Me@65 757 VMS__dissipate_procr( masterVPs[ coreIdx ] );
Me@65 758
Me@65 759 freeSchedSlots( allSchedSlots[ coreIdx ] );
Me@65 760 }
Me@65 761 #endif
Me@65 762
Me@53 763 //All the environment data has been allocated with VMS__malloc, so just
Me@53 764 // free its internal big-chunk and all inside it disappear.
Me@53 765 /*
Me@31 766 readyToAnimateQs = _VMSMasterEnv->readyToAnimateQs;
Me@31 767 masterVPs = _VMSMasterEnv->masterVPs;
Me@31 768 allSchedSlots = _VMSMasterEnv->allSchedSlots;
Me@31 769
Me@31 770 for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
Me@24 771 {
Me@55 772 freeVMSQ( readyToAnimateQs[ coreIdx ] );
Me@50 773 //master VPs were created external to VMS, so use external free
Me@53 774 VMS__dissipate_procr( masterVPs[ coreIdx ] );
Me@31 775
Me@31 776 freeSchedSlots( allSchedSlots[ coreIdx ] );
Me@24 777 }
Me@31 778
Me@53 779 VMS__free( _VMSMasterEnv->readyToAnimateQs );
Me@53 780 VMS__free( _VMSMasterEnv->masterVPs );
Me@53 781 VMS__free( _VMSMasterEnv->allSchedSlots );
Me@50 782
Me@50 783 //============================= MEASUREMENT STUFF ========================
Me@50 784 #ifdef STATS__TURN_ON_PROBES
Me@53 785 freeDynArrayDeep( _VMSMasterEnv->dynIntervalProbesInfo, &VMS__free_probe);
Me@50 786 #endif
Me@50 787 //========================================================================
Me@53 788 */
Me@53 789 //These are the only two that use system free
Me@53 790 VMS_ext__free_free_list( _VMSMasterEnv->freeListHead );
Me@53 791 free( (void *)_VMSMasterEnv );
Me@24 792 }
Me@24 793
Me@54 794
Me@54 795 //================================
Me@54 796
Me@54 797
Me@54 798 /*Later, improve this -- for now, just exits the application after printing
Me@54 799 * the error message.
Me@54 800 */
Me@54 801 void
Me@54 802 VMS__throw_exception( char *msgStr, VirtProcr *reqstPr, VMSExcp *excpData )
Me@54 803 {
msach@69 804 printf("%s",msgStr);
Me@54 805 fflush(stdin);
Me@54 806 exit(1);
Me@54 807 }
Me@54 808