| 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>
|
|
msach@76
|
11 #include <inttypes.h>
|
|
Me@50
|
12 #include <sys/time.h>
|
|
Me@0
|
13
|
|
Me@0
|
14 #include "VMS.h"
|
|
msach@77
|
15 #include "ProcrContext.h"
|
|
nengel@197
|
16 #include "../../C_Libraries/Queue_impl/PrivateQueue.h"
|
|
nengel@197
|
17 #include "../../C_Libraries/Histogram/Histogram.h"
|
|
Me@0
|
18
|
|
Nina@109
|
19 #include <unistd.h>
|
|
Nina@109
|
20 #include <fcntl.h>
|
|
Nina@109
|
21 #include <linux/types.h>
|
|
engelhardt@108
|
22 #include <linux/perf_event.h>
|
|
Nina@109
|
23 #include <errno.h>
|
|
Nina@109
|
24 #include <sys/syscall.h>
|
|
Nina@109
|
25 #include <linux/prctl.h>
|
|
Nina@109
|
26
|
|
Me@0
|
27
|
|
Me@26
|
28 #define thdAttrs NULL
|
|
Me@26
|
29
|
|
Me@22
|
30 //===========================================================================
|
|
Me@22
|
31 void
|
|
Me@22
|
32 shutdownFn( void *dummy, VirtProcr *dummy2 );
|
|
Me@22
|
33
|
|
Me@31
|
34 SchedSlot **
|
|
Me@31
|
35 create_sched_slots();
|
|
Me@22
|
36
|
|
Me@28
|
37 void
|
|
Me@28
|
38 create_masterEnv();
|
|
Me@28
|
39
|
|
Me@28
|
40 void
|
|
Me@28
|
41 create_the_coreLoop_OS_threads();
|
|
Me@28
|
42
|
|
Me@50
|
43 MallocProlog *
|
|
Me@50
|
44 create_free_list();
|
|
Me@50
|
45
|
|
Me@53
|
46 void
|
|
Me@53
|
47 endOSThreadFn( void *initData, VirtProcr *animatingPr );
|
|
Me@50
|
48
|
|
Me@26
|
49 pthread_mutex_t suspendLock = PTHREAD_MUTEX_INITIALIZER;
|
|
Me@26
|
50 pthread_cond_t suspend_cond = PTHREAD_COND_INITIALIZER;
|
|
Me@26
|
51
|
|
Me@22
|
52 //===========================================================================
|
|
Me@22
|
53
|
|
Me@0
|
54 /*Setup has two phases:
|
|
Me@0
|
55 * 1) Semantic layer first calls init_VMS, which creates masterEnv, and puts
|
|
Me@8
|
56 * the master virt procr into the work-queue, ready for first "call"
|
|
Me@8
|
57 * 2) Semantic layer then does its own init, which creates the seed virt
|
|
Me@8
|
58 * procr inside the semantic layer, ready to schedule it when
|
|
Me@0
|
59 * asked by the first run of the masterLoop.
|
|
Me@0
|
60 *
|
|
Me@0
|
61 *This part is bit weird because VMS really wants to be "always there", and
|
|
Me@0
|
62 * have applications attach and detach.. for now, this VMS is part of
|
|
Me@0
|
63 * the app, so the VMS system starts up as part of running the app.
|
|
Me@0
|
64 *
|
|
Me@8
|
65 *The semantic layer is isolated from the VMS internals by making the
|
|
Me@8
|
66 * semantic layer do setup to a state that it's ready with its
|
|
Me@8
|
67 * initial virt procrs, ready to schedule them to slots when the masterLoop
|
|
Me@0
|
68 * asks. Without this pattern, the semantic layer's setup would
|
|
Me@8
|
69 * have to modify slots directly to assign the initial virt-procrs, and put
|
|
Me@31
|
70 * them into the readyToAnimateQ itself, breaking the isolation completely.
|
|
Me@0
|
71 *
|
|
Me@0
|
72 *
|
|
Me@8
|
73 *The semantic layer creates the initial virt procr(s), and adds its
|
|
Me@8
|
74 * own environment to masterEnv, and fills in the pointers to
|
|
Me@0
|
75 * the requestHandler and slaveScheduler plug-in functions
|
|
Me@8
|
76 */
|
|
Me@8
|
77
|
|
Me@8
|
78 /*This allocates VMS data structures, populates the master VMSProc,
|
|
Me@0
|
79 * and master environment, and returns the master environment to the semantic
|
|
Me@0
|
80 * layer.
|
|
Me@0
|
81 */
|
|
Me@8
|
82 void
|
|
Me@8
|
83 VMS__init()
|
|
Me@28
|
84 {
|
|
Me@28
|
85 create_masterEnv();
|
|
Me@28
|
86 create_the_coreLoop_OS_threads();
|
|
Me@28
|
87 }
|
|
Me@28
|
88
|
|
msach@71
|
89 #ifdef SEQUENTIAL
|
|
msach@71
|
90
|
|
Me@28
|
91 /*To initialize the sequential version, just don't create the threads
|
|
Me@28
|
92 */
|
|
Me@28
|
93 void
|
|
Me@28
|
94 VMS__init_Seq()
|
|
Me@28
|
95 {
|
|
Me@28
|
96 create_masterEnv();
|
|
Me@28
|
97 }
|
|
Me@28
|
98
|
|
msach@71
|
99 #endif
|
|
msach@71
|
100
|
|
Me@28
|
101 void
|
|
Me@28
|
102 create_masterEnv()
|
|
Me@31
|
103 { MasterEnv *masterEnv;
|
|
Me@55
|
104 VMSQueueStruc **readyToAnimateQs;
|
|
Me@31
|
105 int coreIdx;
|
|
Me@31
|
106 VirtProcr **masterVPs;
|
|
Me@31
|
107 SchedSlot ***allSchedSlots; //ptr to array of ptrs
|
|
Me@53
|
108
|
|
Me@53
|
109
|
|
Me@31
|
110 //Make the master env, which holds everything else
|
|
Me@1
|
111 _VMSMasterEnv = malloc( sizeof(MasterEnv) );
|
|
Me@53
|
112
|
|
Me@53
|
113 //Very first thing put into the master env is the free-list, seeded
|
|
Me@53
|
114 // with a massive initial chunk of memory.
|
|
Me@53
|
115 //After this, all other mallocs are VMS__malloc.
|
|
Me@53
|
116 _VMSMasterEnv->freeListHead = VMS_ext__create_free_list();
|
|
Me@53
|
117
|
|
Me@65
|
118
|
|
Me@65
|
119 //============================= MEASUREMENT STUFF ========================
|
|
Me@65
|
120 #ifdef MEAS__TIME_MALLOC
|
|
msach@160
|
121 _VMSMasterEnv->mallocTimeHist = makeFixedBinHistExt( 100, 0, 30,
|
|
msach@79
|
122 "malloc_time_hist");
|
|
msach@160
|
123 _VMSMasterEnv->freeTimeHist = makeFixedBinHistExt( 100, 0, 30,
|
|
msach@79
|
124 "free_time_hist");
|
|
Me@65
|
125 #endif
|
|
Me@68
|
126 #ifdef MEAS__TIME_PLUGIN
|
|
msach@160
|
127 _VMSMasterEnv->reqHdlrLowTimeHist = makeFixedBinHistExt( 100, 0, 200,
|
|
msach@79
|
128 "plugin_low_time_hist");
|
|
msach@160
|
129 _VMSMasterEnv->reqHdlrHighTimeHist = makeFixedBinHistExt( 100, 0, 200,
|
|
msach@79
|
130 "plugin_high_time_hist");
|
|
Me@68
|
131 #endif
|
|
Me@65
|
132 //========================================================================
|
|
Me@65
|
133
|
|
Me@53
|
134 //===================== Only VMS__malloc after this ====================
|
|
msach@69
|
135 masterEnv = (MasterEnv*)_VMSMasterEnv;
|
|
Me@31
|
136
|
|
Me@31
|
137 //Make a readyToAnimateQ for each core loop
|
|
Me@55
|
138 readyToAnimateQs = VMS__malloc( NUM_CORES * sizeof(VMSQueueStruc *) );
|
|
Me@53
|
139 masterVPs = VMS__malloc( NUM_CORES * sizeof(VirtProcr *) );
|
|
Me@0
|
140
|
|
Me@31
|
141 //One array for each core, 3 in array, core's masterVP scheds all
|
|
Me@53
|
142 allSchedSlots = VMS__malloc( NUM_CORES * sizeof(SchedSlot *) );
|
|
Me@0
|
143
|
|
Me@53
|
144 _VMSMasterEnv->numProcrsCreated = 0; //used by create procr
|
|
Me@31
|
145 for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
|
|
Me@53
|
146 {
|
|
Me@55
|
147 readyToAnimateQs[ coreIdx ] = makeVMSQ();
|
|
Me@31
|
148
|
|
Me@50
|
149 //Q: should give masterVP core-specific info as its init data?
|
|
msach@76
|
150 masterVPs[ coreIdx ] = VMS__create_procr( (VirtProcrFnPtr)&masterLoop, (void*)masterEnv );
|
|
Me@31
|
151 masterVPs[ coreIdx ]->coreAnimatedBy = coreIdx;
|
|
nengel@193
|
152 masterVPs[ coreIdx ]->isMasterVP = TRUE;
|
|
Me@31
|
153 allSchedSlots[ coreIdx ] = create_sched_slots(); //makes for one core
|
|
Me@53
|
154 _VMSMasterEnv->numMasterInARow[ coreIdx ] = 0;
|
|
Me@55
|
155 _VMSMasterEnv->workStealingGates[ coreIdx ] = NULL;
|
|
Me@31
|
156 }
|
|
Me@31
|
157 _VMSMasterEnv->readyToAnimateQs = readyToAnimateQs;
|
|
Me@31
|
158 _VMSMasterEnv->masterVPs = masterVPs;
|
|
Me@50
|
159 _VMSMasterEnv->masterLock = UNLOCKED;
|
|
Me@31
|
160 _VMSMasterEnv->allSchedSlots = allSchedSlots;
|
|
Me@55
|
161 _VMSMasterEnv->workStealingLock = UNLOCKED;
|
|
Me@28
|
162
|
|
Me@12
|
163
|
|
Me@31
|
164 //Aug 19, 2010: no longer need to place initial masterVP into queue
|
|
Me@31
|
165 // because coreLoop now controls -- animates its masterVP when no work
|
|
Me@31
|
166
|
|
Me@30
|
167
|
|
Me@50
|
168 //============================= MEASUREMENT STUFF ========================
|
|
Me@50
|
169 #ifdef STATS__TURN_ON_PROBES
|
|
Me@50
|
170 _VMSMasterEnv->dynIntervalProbesInfo =
|
|
msach@69
|
171 makePrivDynArrayOfSize( (void***)&(_VMSMasterEnv->intervalProbes), 200);
|
|
Me@30
|
172
|
|
Me@53
|
173 _VMSMasterEnv->probeNameHashTbl = makeHashTable( 1000, &VMS__free );
|
|
Me@53
|
174
|
|
Me@53
|
175 //put creation time directly into master env, for fast retrieval
|
|
Me@50
|
176 struct timeval timeStamp;
|
|
Me@50
|
177 gettimeofday( &(timeStamp), NULL);
|
|
Me@50
|
178 _VMSMasterEnv->createPtInSecs =
|
|
Me@50
|
179 timeStamp.tv_sec +(timeStamp.tv_usec/1000000.0);
|
|
Me@50
|
180 #endif
|
|
Me@65
|
181 #ifdef MEAS__TIME_MASTER_LOCK
|
|
Me@65
|
182 _VMSMasterEnv->masterLockLowTimeHist = makeFixedBinHist( 50, 0, 2,
|
|
Me@65
|
183 "master lock low time hist");
|
|
Me@68
|
184 _VMSMasterEnv->masterLockHighTimeHist = makeFixedBinHist( 50, 0, 100,
|
|
Me@65
|
185 "master lock high time hist");
|
|
Me@65
|
186 #endif
|
|
Me@68
|
187
|
|
msach@76
|
188 MakeTheMeasHists();
|
|
engelhardt@108
|
189
|
|
Nina@166
|
190
|
|
Nina@166
|
191 #ifdef DETECT_LOOP_GRAPH
|
|
nengel@177
|
192
|
|
Nina@129
|
193 #endif
|
|
Nina@166
|
194
|
|
Nina@109
|
195 #ifdef MEAS__PERF_COUNTERS
|
|
engelhardt@108
|
196 struct perf_event_attr hw_event;
|
|
Nina@109
|
197 memset(&hw_event,0,sizeof(hw_event));
|
|
engelhardt@108
|
198 hw_event.type = PERF_TYPE_HARDWARE;
|
|
Nina@109
|
199 hw_event.size = sizeof(hw_event);
|
|
engelhardt@108
|
200 hw_event.disabled = 1;
|
|
Nina@109
|
201 hw_event.freq = 0;
|
|
engelhardt@108
|
202 hw_event.inherit = 1; /* children inherit it */
|
|
engelhardt@108
|
203 hw_event.pinned = 1; /* must always be on PMU */
|
|
engelhardt@108
|
204 hw_event.exclusive = 0; /* only group on PMU */
|
|
engelhardt@108
|
205 hw_event.exclude_user = 0; /* don't count user */
|
|
nengel@211
|
206 hw_event.exclude_kernel = 0; /* ditto kernel */
|
|
nengel@211
|
207 hw_event.exclude_hv = 0; /* ditto hypervisor */
|
|
engelhardt@108
|
208 hw_event.exclude_idle = 0; /* don't count when idle */
|
|
engelhardt@108
|
209 hw_event.mmap = 0; /* include mmap data */
|
|
engelhardt@108
|
210 hw_event.comm = 0; /* include comm data */
|
|
engelhardt@108
|
211
|
|
Nina@109
|
212
|
|
engelhardt@108
|
213 for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
|
|
engelhardt@108
|
214 {
|
|
Nina@109
|
215 hw_event.config = 0x0000000000000000; //cycles
|
|
engelhardt@108
|
216 _VMSMasterEnv->cycles_counter_fd[coreIdx] = syscall(__NR_perf_event_open, &hw_event,
|
|
engelhardt@108
|
217 0,//pid_t pid,
|
|
Nina@167
|
218 coreIdx,//int cpu,
|
|
engelhardt@108
|
219 -1,//int group_fd,
|
|
engelhardt@108
|
220 0//unsigned long flags
|
|
engelhardt@108
|
221 );
|
|
Nina@109
|
222 if (_VMSMasterEnv->cycles_counter_fd[coreIdx]<0){
|
|
Nina@109
|
223 fprintf(stderr,"On core %d: ",coreIdx);
|
|
Nina@109
|
224 perror("Failed to open cycles counter");
|
|
Nina@109
|
225 }
|
|
Nina@109
|
226 hw_event.config = 0x0000000000000001; //instrs
|
|
engelhardt@108
|
227 _VMSMasterEnv->instrs_counter_fd[coreIdx] = syscall(__NR_perf_event_open, &hw_event,
|
|
engelhardt@108
|
228 0,//pid_t pid,
|
|
Nina@167
|
229 coreIdx,//int cpu,
|
|
engelhardt@108
|
230 -1,//int group_fd,
|
|
engelhardt@108
|
231 0//unsigned long flags
|
|
engelhardt@108
|
232 );
|
|
Nina@109
|
233 if (_VMSMasterEnv->instrs_counter_fd[coreIdx]<0){
|
|
Nina@109
|
234 fprintf(stderr,"On core %d: ",coreIdx);
|
|
Nina@109
|
235 perror("Failed to open instrs counter");
|
|
Nina@109
|
236 }
|
|
engelhardt@108
|
237 }
|
|
nengel@211
|
238 //uint64 tmpc,tmpi;
|
|
nengel@211
|
239 //saveCyclesAndInstrs(0,tmpc,tmpi);
|
|
nengel@211
|
240 //printf("Start: cycles = %llu, instrs = %llu\n",tmpc,tmpi);
|
|
engelhardt@108
|
241 #endif
|
|
engelhardt@108
|
242
|
|
Me@50
|
243 //========================================================================
|
|
Me@38
|
244
|
|
Me@0
|
245 }
|
|
Me@0
|
246
|
|
Me@31
|
247 SchedSlot **
|
|
Me@31
|
248 create_sched_slots()
|
|
Me@31
|
249 { SchedSlot **schedSlots;
|
|
Me@0
|
250 int i;
|
|
Me@0
|
251
|
|
Me@53
|
252 schedSlots = VMS__malloc( NUM_SCHED_SLOTS * sizeof(SchedSlot *) );
|
|
Me@8
|
253
|
|
Me@1
|
254 for( i = 0; i < NUM_SCHED_SLOTS; i++ )
|
|
Me@0
|
255 {
|
|
Me@53
|
256 schedSlots[i] = VMS__malloc( sizeof(SchedSlot) );
|
|
Me@8
|
257
|
|
Me@1
|
258 //Set state to mean "handling requests done, slot needs filling"
|
|
Me@8
|
259 schedSlots[i]->workIsDone = FALSE;
|
|
Me@8
|
260 schedSlots[i]->needsProcrAssigned = TRUE;
|
|
Me@0
|
261 }
|
|
Me@31
|
262 return schedSlots;
|
|
Me@31
|
263 }
|
|
Me@31
|
264
|
|
Me@31
|
265
|
|
Me@31
|
266 void
|
|
Me@31
|
267 freeSchedSlots( SchedSlot **schedSlots )
|
|
Me@31
|
268 { int i;
|
|
Me@31
|
269 for( i = 0; i < NUM_SCHED_SLOTS; i++ )
|
|
Me@31
|
270 {
|
|
Me@53
|
271 VMS__free( schedSlots[i] );
|
|
Me@31
|
272 }
|
|
Me@53
|
273 VMS__free( schedSlots );
|
|
Me@0
|
274 }
|
|
Me@0
|
275
|
|
Me@8
|
276
|
|
Me@28
|
277 void
|
|
Me@28
|
278 create_the_coreLoop_OS_threads()
|
|
Me@28
|
279 {
|
|
Me@28
|
280 //========================================================================
|
|
Me@28
|
281 // Create the Threads
|
|
Me@28
|
282 int coreIdx, retCode;
|
|
Me@28
|
283
|
|
Me@28
|
284 //Need the threads to be created suspended, and wait for a signal
|
|
Me@28
|
285 // before proceeding -- gives time after creating to initialize other
|
|
Me@28
|
286 // stuff before the coreLoops set off.
|
|
Me@28
|
287 _VMSMasterEnv->setupComplete = 0;
|
|
Me@28
|
288
|
|
Me@28
|
289 //Make the threads that animate the core loops
|
|
Me@28
|
290 for( coreIdx=0; coreIdx < NUM_CORES; coreIdx++ )
|
|
nengel@211
|
291 { coreLoopThdParams[coreIdx] = VMS__malloc( sizeof(ThdParams) + CACHE_LINE ); //make sure there is no false sharing
|
|
Me@28
|
292 coreLoopThdParams[coreIdx]->coreNum = coreIdx;
|
|
nengel@211
|
293 coreLoopThdParams[coreIdx]->sent_ctr = 0;
|
|
nengel@211
|
294 coreLoopThdParams[coreIdx]->ret_tsc = 0;
|
|
Me@28
|
295
|
|
Me@28
|
296 retCode =
|
|
Me@28
|
297 pthread_create( &(coreLoopThdHandles[coreIdx]),
|
|
Me@28
|
298 thdAttrs,
|
|
Me@28
|
299 &coreLoop,
|
|
Me@28
|
300 (void *)(coreLoopThdParams[coreIdx]) );
|
|
Me@50
|
301 if(retCode){printf("ERROR creating thread: %d\n", retCode); exit(1);}
|
|
Me@28
|
302 }
|
|
nengel@211
|
303 prctl(PR_TASK_PERF_EVENTS_ENABLE);
|
|
Me@28
|
304 }
|
|
Me@28
|
305
|
|
Me@0
|
306 /*Semantic layer calls this when it want the system to start running..
|
|
Me@0
|
307 *
|
|
Me@24
|
308 *This starts the core loops running then waits for them to exit.
|
|
Me@0
|
309 */
|
|
Me@12
|
310 void
|
|
Me@24
|
311 VMS__start_the_work_then_wait_until_done()
|
|
Me@12
|
312 { int coreIdx;
|
|
Me@24
|
313 //Start the core loops running
|
|
Me@25
|
314
|
|
Me@25
|
315 //tell the core loop threads that setup is complete
|
|
Me@25
|
316 //get lock, to lock out any threads still starting up -- they'll see
|
|
Me@25
|
317 // that setupComplete is true before entering while loop, and so never
|
|
Me@25
|
318 // wait on the condition
|
|
Me@26
|
319 pthread_mutex_lock( &suspendLock );
|
|
Me@25
|
320 _VMSMasterEnv->setupComplete = 1;
|
|
Me@26
|
321 pthread_mutex_unlock( &suspendLock );
|
|
Me@26
|
322 pthread_cond_broadcast( &suspend_cond );
|
|
Me@25
|
323
|
|
Me@25
|
324
|
|
Me@24
|
325 //wait for all to complete
|
|
Me@8
|
326 for( coreIdx=0; coreIdx < NUM_CORES; coreIdx++ )
|
|
Me@8
|
327 {
|
|
Me@25
|
328 pthread_join( coreLoopThdHandles[coreIdx], NULL );
|
|
Me@24
|
329 }
|
|
Me@25
|
330
|
|
Me@24
|
331 //NOTE: do not clean up VMS env here -- semantic layer has to have
|
|
Me@24
|
332 // a chance to clean up its environment first, then do a call to free
|
|
Me@24
|
333 // the Master env and rest of VMS locations
|
|
Me@8
|
334 }
|
|
Me@0
|
335
|
|
msach@70
|
336 #ifdef SEQUENTIAL
|
|
Me@28
|
337 /*Only difference between version with an OS thread pinned to each core and
|
|
Me@28
|
338 * the sequential version of VMS is VMS__init_Seq, this, and coreLoop_Seq.
|
|
Me@28
|
339 */
|
|
Me@28
|
340 void
|
|
Me@28
|
341 VMS__start_the_work_then_wait_until_done_Seq()
|
|
Me@28
|
342 {
|
|
Me@28
|
343 //Instead of un-suspending threads, just call the one and only
|
|
Me@28
|
344 // core loop (sequential version), in the main thread.
|
|
Me@28
|
345 coreLoop_Seq( NULL );
|
|
msach@75
|
346 flushRegisters();
|
|
Me@28
|
347
|
|
Me@28
|
348 }
|
|
msach@70
|
349 #endif
|
|
Me@28
|
350
|
|
Me@50
|
351 inline VirtProcr *
|
|
Me@50
|
352 VMS__create_procr( VirtProcrFnPtr fnPtr, void *initialData )
|
|
Me@50
|
353 { VirtProcr *newPr;
|
|
msach@76
|
354 void *stackLocs;
|
|
Me@50
|
355
|
|
Me@50
|
356 newPr = VMS__malloc( sizeof(VirtProcr) );
|
|
Me@50
|
357 stackLocs = VMS__malloc( VIRT_PROCR_STACK_SIZE );
|
|
Me@50
|
358 if( stackLocs == 0 )
|
|
Me@50
|
359 { perror("VMS__malloc stack"); exit(1); }
|
|
Me@50
|
360
|
|
msach@69
|
361 return create_procr_helper( newPr, fnPtr, initialData, stackLocs );
|
|
Me@50
|
362 }
|
|
Me@50
|
363
|
|
Me@50
|
364 /* "ext" designates that it's for use outside the VMS system -- should only
|
|
Me@50
|
365 * be called from main thread or other thread -- never from code animated by
|
|
Me@50
|
366 * a VMS virtual processor.
|
|
Me@50
|
367 */
|
|
Me@50
|
368 inline VirtProcr *
|
|
Me@50
|
369 VMS_ext__create_procr( VirtProcrFnPtr fnPtr, void *initialData )
|
|
Me@50
|
370 { VirtProcr *newPr;
|
|
Me@50
|
371 char *stackLocs;
|
|
Me@50
|
372
|
|
Me@50
|
373 newPr = malloc( sizeof(VirtProcr) );
|
|
Me@50
|
374 stackLocs = malloc( VIRT_PROCR_STACK_SIZE );
|
|
Me@50
|
375 if( stackLocs == 0 )
|
|
Me@50
|
376 { perror("malloc stack"); exit(1); }
|
|
Me@50
|
377
|
|
msach@69
|
378 return create_procr_helper( newPr, fnPtr, initialData, stackLocs );
|
|
Me@50
|
379 }
|
|
Me@50
|
380
|
|
Me@8
|
381
|
|
Me@64
|
382 /*Anticipating multi-tasking
|
|
Me@64
|
383 */
|
|
Me@64
|
384 void *
|
|
Me@64
|
385 VMS__give_sem_env_for( VirtProcr *animPr )
|
|
Me@64
|
386 {
|
|
Me@64
|
387 return _VMSMasterEnv->semanticEnv;
|
|
Me@64
|
388 }
|
|
Me@64
|
389 //===========================================================================
|
|
Me@26
|
390 /*there is a label inside this function -- save the addr of this label in
|
|
Me@0
|
391 * the callingPr struc, as the pick-up point from which to start the next
|
|
Me@0
|
392 * work-unit for that procr. If turns out have to save registers, then
|
|
Me@0
|
393 * save them in the procr struc too. Then do assembly jump to the CoreLoop's
|
|
Me@0
|
394 * "done with work-unit" label. The procr struc is in the request in the
|
|
Me@0
|
395 * slave that animated the just-ended work-unit, so all the state is saved
|
|
Me@0
|
396 * there, and will get passed along, inside the request handler, to the
|
|
Me@0
|
397 * next work-unit for that procr.
|
|
Me@0
|
398 */
|
|
Me@8
|
399 void
|
|
Me@38
|
400 VMS__suspend_procr( VirtProcr *animatingPr )
|
|
Me@55
|
401 {
|
|
Me@30
|
402
|
|
Me@30
|
403 //The request to master will cause this suspended virt procr to get
|
|
Me@30
|
404 // scheduled again at some future point -- to resume, core loop jumps
|
|
Me@30
|
405 // to the resume point (below), which causes restore of saved regs and
|
|
Me@30
|
406 // "return" from this call.
|
|
msach@71
|
407 //animatingPr->nextInstrPt = &&ResumePt;
|
|
Me@30
|
408
|
|
Me@30
|
409 //return ownership of the virt procr and sched slot to Master virt pr
|
|
Me@38
|
410 animatingPr->schedSlot->workIsDone = TRUE;
|
|
Me@1
|
411
|
|
Me@41
|
412 //=========================== Measurement stuff ========================
|
|
Me@38
|
413 #ifdef MEAS__TIME_STAMP_SUSP
|
|
Me@41
|
414 //record time stamp: compare to time-stamp recorded below
|
|
Me@38
|
415 saveLowTimeStampCountInto( animatingPr->preSuspTSCLow );
|
|
Me@38
|
416 #endif
|
|
nengel@186
|
417 #ifdef MEAS__PERF_COUNTERS
|
|
nengel@186
|
418 //start work
|
|
nengel@186
|
419 uint64 cycles,instrs;
|
|
nengel@186
|
420 saveCyclesAndInstrs(animatingPr->coreAnimatedBy,cycles, instrs);
|
|
nengel@193
|
421 (*(_VMSMasterEnv->counterHandler))(HwResponderInvocation_start,animatingPr->procrID,animatingPr->numTimesScheduled,animatingPr,cycles,instrs);
|
|
nengel@186
|
422 #endif
|
|
Me@41
|
423 //=======================================================================
|
|
Me@30
|
424
|
|
msach@71
|
425 switchToCoreLoop(animatingPr);
|
|
msach@71
|
426 flushRegisters();
|
|
Me@55
|
427
|
|
Me@55
|
428 //=======================================================================
|
|
Me@30
|
429
|
|
Me@38
|
430 #ifdef MEAS__TIME_STAMP_SUSP
|
|
Me@41
|
431 //NOTE: only take low part of count -- do sanity check when take diff
|
|
Me@38
|
432 saveLowTimeStampCountInto( animatingPr->postSuspTSCLow );
|
|
Me@38
|
433 #endif
|
|
Me@30
|
434
|
|
Me@0
|
435 return;
|
|
Me@0
|
436 }
|
|
Me@0
|
437
|
|
Me@22
|
438
|
|
Me@22
|
439
|
|
Me@50
|
440 /*For this implementation of VMS, it may not make much sense to have the
|
|
Me@50
|
441 * system of requests for creating a new processor done this way.. but over
|
|
Me@50
|
442 * the scope of single-master, multi-master, mult-tasking, OS-implementing,
|
|
Me@50
|
443 * distributed-memory, and so on, this gives VMS implementation a chance to
|
|
Me@50
|
444 * do stuff before suspend, in the AppVP, and in the Master before the plugin
|
|
Me@50
|
445 * is called, as well as in the lang-lib before this is called, and in the
|
|
Me@50
|
446 * plugin. So, this gives both VMS and language implementations a chance to
|
|
Me@50
|
447 * intercept at various points and do order-dependent stuff.
|
|
Me@50
|
448 *Having a standard VMSNewPrReqData struc allows the language to create and
|
|
Me@50
|
449 * free the struc, while VMS knows how to get the newPr if it wants it, and
|
|
Me@50
|
450 * it lets the lang have lang-specific data related to creation transported
|
|
Me@50
|
451 * to the plugin.
|
|
Me@50
|
452 */
|
|
msach@160
|
453 void
|
|
msach@160
|
454 VMS__send_create_procr_req( void *semReqData, VirtProcr *reqstingPr )
|
|
Me@50
|
455 { VMSReqst req;
|
|
Me@50
|
456
|
|
Me@50
|
457 req.reqType = createReq;
|
|
Me@50
|
458 req.semReqData = semReqData;
|
|
Me@50
|
459 req.nextReqst = reqstingPr->requests;
|
|
Me@50
|
460 reqstingPr->requests = &req;
|
|
Me@50
|
461
|
|
Me@50
|
462 VMS__suspend_procr( reqstingPr );
|
|
Me@50
|
463 }
|
|
Me@50
|
464
|
|
Me@22
|
465
|
|
Me@38
|
466 /*
|
|
Me@22
|
467 *This adds a request to dissipate, then suspends the processor so that the
|
|
Me@22
|
468 * request handler will receive the request. The request handler is what
|
|
Me@22
|
469 * does the work of freeing memory and removing the processor from the
|
|
Me@22
|
470 * semantic environment's data structures.
|
|
Me@22
|
471 *The request handler also is what figures out when to shutdown the VMS
|
|
Me@22
|
472 * system -- which causes all the core loop threads to die, and returns from
|
|
Me@22
|
473 * the call that started up VMS to perform the work.
|
|
Me@22
|
474 *
|
|
Me@22
|
475 *This form is a bit misleading to understand if one is trying to figure out
|
|
Me@22
|
476 * how VMS works -- it looks like a normal function call, but inside it
|
|
Me@22
|
477 * sends a request to the request handler and suspends the processor, which
|
|
Me@22
|
478 * jumps out of the VMS__dissipate_procr function, and out of all nestings
|
|
Me@22
|
479 * above it, transferring the work of dissipating to the request handler,
|
|
Me@22
|
480 * which then does the actual work -- causing the processor that animated
|
|
Me@22
|
481 * the call of this function to disappear and the "hanging" state of this
|
|
Me@22
|
482 * function to just poof into thin air -- the virtual processor's trace
|
|
Me@22
|
483 * never returns from this call, but instead the virtual processor's trace
|
|
Me@22
|
484 * gets suspended in this call and all the virt processor's state disap-
|
|
Me@22
|
485 * pears -- making that suspend the last thing in the virt procr's trace.
|
|
Me@8
|
486 */
|
|
msach@160
|
487 void
|
|
msach@160
|
488 VMS__send_dissipate_req( VirtProcr *procrToDissipate )
|
|
Me@50
|
489 { VMSReqst req;
|
|
Me@22
|
490
|
|
Me@50
|
491 req.reqType = dissipate;
|
|
Me@50
|
492 req.nextReqst = procrToDissipate->requests;
|
|
Me@50
|
493 procrToDissipate->requests = &req;
|
|
Me@50
|
494
|
|
Me@22
|
495 VMS__suspend_procr( procrToDissipate );
|
|
Me@50
|
496 }
|
|
Me@50
|
497
|
|
Me@50
|
498
|
|
Me@50
|
499 /* "ext" designates that it's for use outside the VMS system -- should only
|
|
Me@50
|
500 * be called from main thread or other thread -- never from code animated by
|
|
Me@50
|
501 * a VMS virtual processor.
|
|
Me@50
|
502 *
|
|
Me@50
|
503 *Use this version to dissipate VPs created outside the VMS system.
|
|
Me@50
|
504 */
|
|
Me@50
|
505 void
|
|
Me@50
|
506 VMS_ext__dissipate_procr( VirtProcr *procrToDissipate )
|
|
Me@50
|
507 {
|
|
Me@50
|
508 //NOTE: initialData was given to the processor, so should either have
|
|
Me@50
|
509 // been alloc'd with VMS__malloc, or freed by the level above animPr.
|
|
Me@50
|
510 //So, all that's left to free here is the stack and the VirtProcr struc
|
|
Me@50
|
511 // itself
|
|
Me@50
|
512 //Note, should not stack-allocate initial data -- no guarantee, in
|
|
Me@50
|
513 // general that creating processor will outlive ones it creates.
|
|
Me@50
|
514 free( procrToDissipate->startOfStack );
|
|
Me@50
|
515 free( procrToDissipate );
|
|
Me@50
|
516 }
|
|
Me@50
|
517
|
|
Me@22
|
518
|
|
Me@22
|
519
|
|
Me@53
|
520 /*This call's name indicates that request is malloc'd -- so req handler
|
|
Me@53
|
521 * has to free any extra requests tacked on before a send, using this.
|
|
Me@53
|
522 *
|
|
Me@53
|
523 * This inserts the semantic-layer's request data into standard VMS carrier
|
|
Me@53
|
524 * request data-struct that is mallocd. The sem request doesn't need to
|
|
Me@53
|
525 * be malloc'd if this is called inside the same call chain before the
|
|
Me@53
|
526 * send of the last request is called.
|
|
Me@53
|
527 *
|
|
Me@53
|
528 *The request handler has to call VMS__free_VMSReq for any of these
|
|
Me@22
|
529 */
|
|
Me@22
|
530 inline void
|
|
Me@53
|
531 VMS__add_sem_request_in_mallocd_VMSReqst( void *semReqData,
|
|
Me@53
|
532 VirtProcr *callingPr )
|
|
Me@53
|
533 { VMSReqst *req;
|
|
Me@22
|
534
|
|
Me@53
|
535 req = VMS__malloc( sizeof(VMSReqst) );
|
|
Me@53
|
536 req->reqType = semantic;
|
|
Me@53
|
537 req->semReqData = semReqData;
|
|
Me@53
|
538 req->nextReqst = callingPr->requests;
|
|
Me@53
|
539 callingPr->requests = req;
|
|
Me@22
|
540 }
|
|
Me@22
|
541
|
|
Me@50
|
542 /*This inserts the semantic-layer's request data into standard VMS carrier
|
|
Me@50
|
543 * request data-struct is allocated on stack of this call & ptr to it sent
|
|
Me@50
|
544 * to plugin
|
|
Me@50
|
545 *Then it does suspend, to cause request to be sent.
|
|
Me@50
|
546 */
|
|
msach@160
|
547 inline void
|
|
msach@160
|
548 VMS__send_sem_request( void *semReqData, VirtProcr *callingPr )
|
|
Me@50
|
549 { VMSReqst req;
|
|
Me@22
|
550
|
|
Me@50
|
551 req.reqType = semantic;
|
|
Me@50
|
552 req.semReqData = semReqData;
|
|
Me@50
|
553 req.nextReqst = callingPr->requests;
|
|
Me@50
|
554 callingPr->requests = &req;
|
|
Me@50
|
555
|
|
Me@50
|
556 VMS__suspend_procr( callingPr );
|
|
Me@50
|
557 }
|
|
Me@50
|
558
|
|
Me@50
|
559
|
|
msach@160
|
560 inline void
|
|
msach@160
|
561 VMS__send_VMSSem_request( void *semReqData, VirtProcr *callingPr )
|
|
Me@50
|
562 { VMSReqst req;
|
|
Me@50
|
563
|
|
Me@50
|
564 req.reqType = VMSSemantic;
|
|
Me@50
|
565 req.semReqData = semReqData;
|
|
Me@50
|
566 req.nextReqst = callingPr->requests; //gab any other preceeding
|
|
Me@50
|
567 callingPr->requests = &req;
|
|
Me@50
|
568
|
|
Me@50
|
569 VMS__suspend_procr( callingPr );
|
|
Me@50
|
570 }
|
|
Me@50
|
571
|
|
Me@50
|
572
|
|
Me@50
|
573 /*
|
|
Me@38
|
574 */
|
|
Me@24
|
575 VMSReqst *
|
|
Me@50
|
576 VMS__take_next_request_out_of( VirtProcr *procrWithReq )
|
|
Me@31
|
577 { VMSReqst *req;
|
|
Me@31
|
578
|
|
Me@31
|
579 req = procrWithReq->requests;
|
|
Me@38
|
580 if( req == NULL ) return NULL;
|
|
Me@31
|
581
|
|
Me@31
|
582 procrWithReq->requests = procrWithReq->requests->nextReqst;
|
|
Me@50
|
583 return req;
|
|
Me@24
|
584 }
|
|
Me@22
|
585
|
|
Me@24
|
586
|
|
Me@24
|
587 inline void *
|
|
Me@24
|
588 VMS__take_sem_reqst_from( VMSReqst *req )
|
|
Me@24
|
589 {
|
|
Me@24
|
590 return req->semReqData;
|
|
Me@24
|
591 }
|
|
Me@24
|
592
|
|
Me@24
|
593
|
|
Me@24
|
594
|
|
Me@50
|
595 /* This is for OS requests and VMS infrastructure requests, such as to create
|
|
Me@50
|
596 * a probe -- a probe is inside the heart of VMS-core, it's not part of any
|
|
Me@50
|
597 * language -- but it's also a semantic thing that's triggered from and used
|
|
Me@50
|
598 * in the application.. so it crosses abstractions.. so, need some special
|
|
Me@50
|
599 * pattern here for handling such requests.
|
|
Me@52
|
600 * Doing this just like it were a second language sharing VMS-core.
|
|
Me@52
|
601 *
|
|
Me@50
|
602 * This is called from the language's request handler when it sees a request
|
|
Me@50
|
603 * of type VMSSemReq
|
|
Me@52
|
604 *
|
|
Me@52
|
605 * TODO: Later change this, to give probes their own separate plugin & have
|
|
Me@52
|
606 * VMS-core steer the request to appropriate plugin
|
|
Me@52
|
607 * Do the same for OS calls -- look later at it..
|
|
Me@50
|
608 */
|
|
Me@50
|
609 void inline
|
|
Me@50
|
610 VMS__handle_VMSSemReq( VMSReqst *req, VirtProcr *requestingPr, void *semEnv,
|
|
Me@50
|
611 ResumePrFnPtr resumePrFnPtr )
|
|
Me@50
|
612 { VMSSemReq *semReq;
|
|
Me@50
|
613 IntervalProbe *newProbe;
|
|
Me@24
|
614
|
|
Me@50
|
615 semReq = req->semReqData;
|
|
Me@24
|
616
|
|
Me@50
|
617 newProbe = VMS__malloc( sizeof(IntervalProbe) );
|
|
Me@65
|
618 newProbe->nameStr = VMS__strDup( semReq->nameStr );
|
|
Me@50
|
619 newProbe->hist = NULL;
|
|
Me@50
|
620 newProbe->schedChoiceWasRecorded = FALSE;
|
|
Me@53
|
621
|
|
Me@53
|
622 //This runs in masterVP, so no race-condition worries
|
|
Me@50
|
623 newProbe->probeID =
|
|
Me@50
|
624 addToDynArray( newProbe, _VMSMasterEnv->dynIntervalProbesInfo );
|
|
Me@50
|
625
|
|
Me@53
|
626 requestingPr->dataRetFromReq = newProbe;
|
|
Me@50
|
627
|
|
Me@50
|
628 (*resumePrFnPtr)( requestingPr, semEnv );
|
|
Me@22
|
629 }
|
|
Me@22
|
630
|
|
Me@22
|
631
|
|
Me@22
|
632
|
|
Me@24
|
633 /*This must be called by the request handler plugin -- it cannot be called
|
|
Me@24
|
634 * from the semantic library "dissipate processor" function -- instead, the
|
|
Me@50
|
635 * semantic layer has to generate a request, and the plug-in calls this
|
|
Me@24
|
636 * function.
|
|
Me@24
|
637 *The reason is that this frees the virtual processor's stack -- which is
|
|
Me@24
|
638 * still in use inside semantic library calls!
|
|
Me@24
|
639 *
|
|
Me@24
|
640 *This frees or recycles all the state owned by and comprising the VMS
|
|
Me@24
|
641 * portion of the animating virtual procr. The request handler must first
|
|
Me@24
|
642 * free any semantic data created for the processor that didn't use the
|
|
Me@24
|
643 * VMS_malloc mechanism. Then it calls this, which first asks the malloc
|
|
Me@24
|
644 * system to disown any state that did use VMS_malloc, and then frees the
|
|
Me@24
|
645 * statck and the processor-struct itself.
|
|
Me@24
|
646 *If the dissipated processor is the sole (remaining) owner of VMS__malloc'd
|
|
Me@24
|
647 * state, then that state gets freed (or sent to recycling) as a side-effect
|
|
Me@24
|
648 * of dis-owning it.
|
|
Me@24
|
649 */
|
|
Me@24
|
650 void
|
|
Me@53
|
651 VMS__dissipate_procr( VirtProcr *animatingPr )
|
|
Me@24
|
652 {
|
|
Me@24
|
653 //dis-own all locations owned by this processor, causing to be freed
|
|
Me@24
|
654 // any locations that it is (was) sole owner of
|
|
Me@29
|
655 //TODO: implement VMS__malloc system, including "give up ownership"
|
|
Me@24
|
656
|
|
Me@24
|
657
|
|
Me@24
|
658 //NOTE: initialData was given to the processor, so should either have
|
|
Me@24
|
659 // been alloc'd with VMS__malloc, or freed by the level above animPr.
|
|
Me@24
|
660 //So, all that's left to free here is the stack and the VirtProcr struc
|
|
Me@24
|
661 // itself
|
|
Me@50
|
662 //Note, should not stack-allocate initial data -- no guarantee, in
|
|
Me@50
|
663 // general that creating processor will outlive ones it creates.
|
|
Me@50
|
664 VMS__free( animatingPr->startOfStack );
|
|
Me@50
|
665 VMS__free( animatingPr );
|
|
Me@24
|
666 }
|
|
Me@24
|
667
|
|
Me@24
|
668
|
|
Me@53
|
669 //TODO: look at architecting cleanest separation between request handler
|
|
Me@29
|
670 // and master loop, for dissipate, create, shutdown, and other non-semantic
|
|
Me@29
|
671 // requests. Issue is chain: one removes requests from AppVP, one dispatches
|
|
Me@29
|
672 // on type of request, and one handles each type.. but some types require
|
|
Me@29
|
673 // action from both request handler and master loop -- maybe just give the
|
|
Me@29
|
674 // request handler calls like: VMS__handle_X_request_type
|
|
Me@24
|
675
|
|
Me@29
|
676
|
|
Me@29
|
677 /*This is called by the semantic layer's request handler when it decides its
|
|
Me@29
|
678 * time to shut down the VMS system. Calling this causes the core loop OS
|
|
Me@29
|
679 * threads to exit, which unblocks the entry-point function that started up
|
|
Me@29
|
680 * VMS, and allows it to grab the result and return to the original single-
|
|
Me@29
|
681 * threaded application.
|
|
Me@22
|
682 *
|
|
Me@29
|
683 *The _VMSMasterEnv is needed by this shut down function, so the create-seed-
|
|
Me@29
|
684 * and-wait function has to free a bunch of stuff after it detects the
|
|
Me@29
|
685 * threads have all died: the masterEnv, the thread-related locations,
|
|
Me@29
|
686 * masterVP any AppVPs that might still be allocated and sitting in the
|
|
Me@29
|
687 * semantic environment, or have been orphaned in the _VMSWorkQ.
|
|
Me@29
|
688 *
|
|
Me@53
|
689 *NOTE: the semantic plug-in is expected to use VMS__malloc to get all the
|
|
Me@29
|
690 * locations it needs, and give ownership to masterVP. Then, they will be
|
|
Me@53
|
691 * automatically freed.
|
|
Me@22
|
692 *
|
|
Me@29
|
693 *In here,create one core-loop shut-down processor for each core loop and put
|
|
Me@31
|
694 * them all directly into the readyToAnimateQ.
|
|
Me@29
|
695 *Note, this function can ONLY be called after the semantic environment no
|
|
Me@29
|
696 * longer cares if AppVPs get animated after the point this is called. In
|
|
Me@29
|
697 * other words, this can be used as an abort, or else it should only be
|
|
Me@29
|
698 * called when all AppVPs have finished dissipate requests -- only at that
|
|
Me@29
|
699 * point is it sure that all results have completed.
|
|
Me@22
|
700 */
|
|
Me@22
|
701 void
|
|
Me@53
|
702 VMS__shutdown()
|
|
Me@8
|
703 { int coreIdx;
|
|
Me@14
|
704 VirtProcr *shutDownPr;
|
|
Me@22
|
705
|
|
Me@29
|
706 //create the shutdown processors, one for each core loop -- put them
|
|
Me@31
|
707 // directly into the Q -- each core will die when gets one
|
|
Me@8
|
708 for( coreIdx=0; coreIdx < NUM_CORES; coreIdx++ )
|
|
Me@50
|
709 { //Note, this is running in the master
|
|
Me@29
|
710 shutDownPr = VMS__create_procr( &endOSThreadFn, NULL );
|
|
nengel@193
|
711 shutDownPr->isShutdownVP = TRUE;
|
|
Me@55
|
712 writeVMSQ( shutDownPr, _VMSMasterEnv->readyToAnimateQs[coreIdx] );
|
|
Me@8
|
713 }
|
|
Nina@109
|
714 #ifdef MEAS__PERF_COUNTERS
|
|
Nina@109
|
715 uint64 tmpc,tmpi;
|
|
Nina@109
|
716 saveCyclesAndInstrs(0,tmpc,tmpi);
|
|
nengel@211
|
717 //printf("End: cycles = %llu, instrs = %llu\n",tmpc,tmpi);
|
|
Nina@109
|
718 /*
|
|
Nina@109
|
719 for( coreIdx=0; coreIdx < NUM_CORES; coreIdx++ ){
|
|
Nina@109
|
720 close(_VMSMasterEnv->cycles_counter_fd[coreIdx]);
|
|
Nina@109
|
721 close(_VMSMasterEnv->instrs_counter_fd[coreIdx]);
|
|
Nina@109
|
722 }
|
|
Nina@109
|
723 */
|
|
Nina@109
|
724 #endif
|
|
Me@12
|
725 }
|
|
Me@12
|
726
|
|
Me@12
|
727
|
|
Me@29
|
728 /*Am trying to be cute, avoiding IF statement in coreLoop that checks for
|
|
Me@29
|
729 * a special shutdown procr. Ended up with extra-complex shutdown sequence.
|
|
Me@29
|
730 *This function has the sole purpose of setting the stack and framePtr
|
|
Me@29
|
731 * to the coreLoop's stack and framePtr.. it does that then jumps to the
|
|
Me@29
|
732 * core loop's shutdown point -- might be able to just call Pthread_exit
|
|
Me@30
|
733 * from here, but am going back to the pthread's stack and setting everything
|
|
Me@29
|
734 * up just as if it never jumped out, before calling pthread_exit.
|
|
Me@29
|
735 *The end-point of core loop will free the stack and so forth of the
|
|
Me@29
|
736 * processor that animates this function, (this fn is transfering the
|
|
Me@29
|
737 * animator of the AppVP that is in turn animating this function over
|
|
Me@29
|
738 * to core loop function -- note that this slices out a level of virtual
|
|
Me@29
|
739 * processors).
|
|
Me@29
|
740 */
|
|
Me@29
|
741 void
|
|
Me@29
|
742 endOSThreadFn( void *initData, VirtProcr *animatingPr )
|
|
msach@71
|
743 {
|
|
msach@75
|
744 #ifdef SEQUENTIAL
|
|
msach@75
|
745 asmTerminateCoreLoopSeq(animatingPr);
|
|
msach@75
|
746 #else
|
|
msach@71
|
747 asmTerminateCoreLoop(animatingPr);
|
|
msach@75
|
748 #endif
|
|
Me@30
|
749 }
|
|
Me@29
|
750
|
|
Me@29
|
751
|
|
Me@53
|
752 /*This is called from the startup & shutdown
|
|
Me@24
|
753 */
|
|
Me@24
|
754 void
|
|
Me@53
|
755 VMS__cleanup_at_end_of_shutdown()
|
|
Me@31
|
756 {
|
|
msach@78
|
757 //unused
|
|
msach@78
|
758 //VMSQueueStruc **readyToAnimateQs;
|
|
msach@78
|
759 //int coreIdx;
|
|
msach@78
|
760 //VirtProcr **masterVPs;
|
|
msach@78
|
761 //SchedSlot ***allSchedSlots; //ptr to array of ptrs
|
|
Me@31
|
762
|
|
Me@65
|
763 //Before getting rid of everything, print out any measurements made
|
|
Nina@109
|
764 //forAllInDynArrayDo( _VMSMasterEnv->measHistsInfo, (DynArrayFnPtr)&printHist );
|
|
Nina@109
|
765 //forAllInDynArrayDo( _VMSMasterEnv->measHistsInfo, (DynArrayFnPtr)&saveHistToFile);
|
|
nengel@182
|
766 //forAllInDynArrayDo( _VMSMasterEnv->measHistsInfo, &freeHist );
|
|
Nina@167
|
767
|
|
nengel@211
|
768 prctl(PR_TASK_PERF_EVENTS_DISABLE);
|
|
Me@65
|
769 #ifdef MEAS__TIME_PLUGIN
|
|
Me@68
|
770 printHist( _VMSMasterEnv->reqHdlrLowTimeHist );
|
|
msach@84
|
771 saveHistToFile( _VMSMasterEnv->reqHdlrLowTimeHist );
|
|
Me@68
|
772 printHist( _VMSMasterEnv->reqHdlrHighTimeHist );
|
|
msach@79
|
773 saveHistToFile( _VMSMasterEnv->reqHdlrHighTimeHist );
|
|
Me@68
|
774 freeHistExt( _VMSMasterEnv->reqHdlrLowTimeHist );
|
|
Me@68
|
775 freeHistExt( _VMSMasterEnv->reqHdlrHighTimeHist );
|
|
Me@65
|
776 #endif
|
|
Me@65
|
777 #ifdef MEAS__TIME_MALLOC
|
|
Me@65
|
778 printHist( _VMSMasterEnv->mallocTimeHist );
|
|
msach@79
|
779 saveHistToFile( _VMSMasterEnv->mallocTimeHist );
|
|
Me@65
|
780 printHist( _VMSMasterEnv->freeTimeHist );
|
|
msach@79
|
781 saveHistToFile( _VMSMasterEnv->freeTimeHist );
|
|
Me@65
|
782 freeHistExt( _VMSMasterEnv->mallocTimeHist );
|
|
Me@65
|
783 freeHistExt( _VMSMasterEnv->freeTimeHist );
|
|
Me@65
|
784 #endif
|
|
Me@65
|
785 #ifdef MEAS__TIME_MASTER_LOCK
|
|
Me@65
|
786 printHist( _VMSMasterEnv->masterLockLowTimeHist );
|
|
Me@65
|
787 printHist( _VMSMasterEnv->masterLockHighTimeHist );
|
|
Me@65
|
788 #endif
|
|
Me@65
|
789 #ifdef MEAS__TIME_MASTER
|
|
Me@65
|
790 printHist( _VMSMasterEnv->pluginTimeHist );
|
|
Me@65
|
791 for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
|
|
Me@65
|
792 {
|
|
Me@65
|
793 freeVMSQ( readyToAnimateQs[ coreIdx ] );
|
|
Me@65
|
794 //master VPs were created external to VMS, so use external free
|
|
Me@65
|
795 VMS__dissipate_procr( masterVPs[ coreIdx ] );
|
|
Me@65
|
796
|
|
Me@65
|
797 freeSchedSlots( allSchedSlots[ coreIdx ] );
|
|
Me@65
|
798 }
|
|
Me@65
|
799 #endif
|
|
Me@65
|
800 #ifdef MEAS__TIME_STAMP_SUSP
|
|
Me@65
|
801 printHist( _VMSMasterEnv->pluginTimeHist );
|
|
Me@65
|
802 for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
|
|
Me@65
|
803 {
|
|
Me@65
|
804 freeVMSQ( readyToAnimateQs[ coreIdx ] );
|
|
Me@65
|
805 //master VPs were created external to VMS, so use external free
|
|
Me@65
|
806 VMS__dissipate_procr( masterVPs[ coreIdx ] );
|
|
Me@65
|
807
|
|
Me@65
|
808 freeSchedSlots( allSchedSlots[ coreIdx ] );
|
|
Me@65
|
809 }
|
|
Me@65
|
810 #endif
|
|
Me@65
|
811
|
|
Me@53
|
812 //All the environment data has been allocated with VMS__malloc, so just
|
|
Me@53
|
813 // free its internal big-chunk and all inside it disappear.
|
|
Me@53
|
814 /*
|
|
Me@31
|
815 readyToAnimateQs = _VMSMasterEnv->readyToAnimateQs;
|
|
Me@31
|
816 masterVPs = _VMSMasterEnv->masterVPs;
|
|
Me@31
|
817 allSchedSlots = _VMSMasterEnv->allSchedSlots;
|
|
Me@31
|
818
|
|
Me@31
|
819 for( coreIdx = 0; coreIdx < NUM_CORES; coreIdx++ )
|
|
Me@24
|
820 {
|
|
Me@55
|
821 freeVMSQ( readyToAnimateQs[ coreIdx ] );
|
|
Me@50
|
822 //master VPs were created external to VMS, so use external free
|
|
Me@53
|
823 VMS__dissipate_procr( masterVPs[ coreIdx ] );
|
|
Me@31
|
824
|
|
Me@31
|
825 freeSchedSlots( allSchedSlots[ coreIdx ] );
|
|
Me@24
|
826 }
|
|
Me@31
|
827
|
|
Me@53
|
828 VMS__free( _VMSMasterEnv->readyToAnimateQs );
|
|
Me@53
|
829 VMS__free( _VMSMasterEnv->masterVPs );
|
|
Me@53
|
830 VMS__free( _VMSMasterEnv->allSchedSlots );
|
|
Me@50
|
831
|
|
Me@50
|
832 //============================= MEASUREMENT STUFF ========================
|
|
Me@50
|
833 #ifdef STATS__TURN_ON_PROBES
|
|
Me@53
|
834 freeDynArrayDeep( _VMSMasterEnv->dynIntervalProbesInfo, &VMS__free_probe);
|
|
Me@50
|
835 #endif
|
|
Me@50
|
836 //========================================================================
|
|
Me@53
|
837 */
|
|
Me@53
|
838 //These are the only two that use system free
|
|
Me@53
|
839 VMS_ext__free_free_list( _VMSMasterEnv->freeListHead );
|
|
Me@53
|
840 free( (void *)_VMSMasterEnv );
|
|
Me@24
|
841 }
|
|
Me@24
|
842
|
|
Me@54
|
843
|
|
Me@54
|
844 //================================
|
|
Me@54
|
845
|
|
Me@54
|
846
|
|
Me@54
|
847 /*Later, improve this -- for now, just exits the application after printing
|
|
Me@54
|
848 * the error message.
|
|
Me@54
|
849 */
|
|
Me@54
|
850 void
|
|
Me@54
|
851 VMS__throw_exception( char *msgStr, VirtProcr *reqstPr, VMSExcp *excpData )
|
|
Me@54
|
852 {
|
|
msach@69
|
853 printf("%s",msgStr);
|
|
Me@54
|
854 fflush(stdin);
|
|
Me@54
|
855 exit(1);
|
|
Me@54
|
856 }
|
|
Me@54
|
857
|
|
nengel@211
|
858 __inline__ uint64_t rdtsc(void){
|
|
nengel@211
|
859 uint32_t lo, hi;
|
|
nengel@211
|
860 __asm__ __volatile__ ( // serialize
|
|
nengel@211
|
861 "xorl %%eax,%%eax \n cpuid"
|
|
nengel@211
|
862 ::: "%rax", "%rbx", "%rcx", "%rdx");
|
|
nengel@211
|
863 __asm__ __volatile__ ("rdtsc" : "=a" (lo), "=d" (hi));
|
|
nengel@211
|
864 /* asm volatile("RDTSC;"
|
|
nengel@211
|
865 "movl %%eax, %0;"
|
|
nengel@211
|
866 "movl %%edx, %1;"
|
|
nengel@211
|
867 : "=m" (lo), "=m" (hi)
|
|
nengel@211
|
868 :
|
|
nengel@211
|
869 : "%eax", "%edx"
|
|
nengel@211
|
870 ); */
|
|
nengel@211
|
871 return (uint64_t)hi << 32 | lo;
|
|
nengel@211
|
872 }
|
|
nengel@211
|
873
|
|
nengel@211
|
874 uint64 tsc_offset_send(ThdParams* thisCoresThdParams, uint64 initval){
|
|
nengel@211
|
875 uint64 ret_tsc_curr; //local copy of coreLoopThdParams->ret_tsc
|
|
nengel@211
|
876 uint64 ret_tsc_prev;
|
|
nengel@211
|
877 uint64 local_before;
|
|
nengel@211
|
878 uint64 local_after;
|
|
nengel@211
|
879
|
|
nengel@211
|
880 ret_tsc_prev = initval;
|
|
nengel@211
|
881 ret_tsc_curr = initval;
|
|
nengel@211
|
882 local_before = rdtsc();
|
|
nengel@211
|
883 thisCoresThdParams->sent_ctr++;
|
|
nengel@211
|
884 while(ret_tsc_curr == ret_tsc_prev)
|
|
nengel@211
|
885 ret_tsc_curr = thisCoresThdParams->ret_tsc;
|
|
nengel@211
|
886 local_after = rdtsc();
|
|
nengel@211
|
887 ret_tsc_prev = ret_tsc_curr;
|
|
nengel@211
|
888
|
|
nengel@211
|
889 int i;
|
|
nengel@211
|
890 for(i=0;i<3;++i){
|
|
nengel@211
|
891 local_before = rdtsc();
|
|
nengel@211
|
892 thisCoresThdParams->sent_ctr++;
|
|
nengel@211
|
893 while(ret_tsc_curr == ret_tsc_prev)
|
|
nengel@211
|
894 ret_tsc_curr = thisCoresThdParams->ret_tsc;
|
|
nengel@211
|
895 local_after = rdtsc();
|
|
nengel@211
|
896 int64 midpoint = local_before + (local_after-local_before)/2;
|
|
nengel@211
|
897 int64 difference;
|
|
nengel@211
|
898 if (midpoint > ret_tsc_curr)
|
|
nengel@211
|
899 difference = midpoint - (int64)ret_tsc_curr;
|
|
nengel@211
|
900 else
|
|
nengel@211
|
901 difference = (int64)ret_tsc_curr - midpoint;
|
|
nengel@211
|
902 //printf("TSC: %llu (Core %d) = %llu (Core %d) // difference=%llu\n",midpoint,thisCoresThdParams->coreNum,ret_tsc_curr,thisCoresThdParams->coreNum + 1,difference);
|
|
nengel@211
|
903 ret_tsc_prev = ret_tsc_curr;
|
|
nengel@211
|
904 }
|
|
nengel@211
|
905
|
|
nengel@211
|
906 return ret_tsc_curr;
|
|
nengel@211
|
907 }
|
|
nengel@211
|
908
|
|
nengel@211
|
909 int tsc_offset_resp(ThdParams* sendCoresThdParams, int initialctrval){
|
|
nengel@211
|
910
|
|
nengel@211
|
911 int send_ctr_curr = initialctrval;
|
|
nengel@211
|
912 int send_ctr_prev = initialctrval;
|
|
nengel@211
|
913
|
|
nengel@211
|
914
|
|
nengel@211
|
915 while(send_ctr_curr == send_ctr_prev)
|
|
nengel@211
|
916 send_ctr_curr = sendCoresThdParams->sent_ctr;
|
|
nengel@211
|
917 sendCoresThdParams->ret_tsc = rdtsc();
|
|
nengel@211
|
918 send_ctr_prev = send_ctr_curr;
|
|
nengel@211
|
919
|
|
nengel@211
|
920 int i;
|
|
nengel@211
|
921 for(i=0;i<3;++i){
|
|
nengel@211
|
922 while(send_ctr_curr == send_ctr_prev)
|
|
nengel@211
|
923 send_ctr_curr = sendCoresThdParams->sent_ctr;
|
|
nengel@211
|
924 sendCoresThdParams->ret_tsc = rdtsc();
|
|
nengel@211
|
925 send_ctr_prev = send_ctr_curr;
|
|
nengel@211
|
926 }
|
|
nengel@211
|
927 return send_ctr_curr;
|
|
nengel@211
|
928 } |
