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