VMS/VMS_Implementations/VMS_impls/VMS__MC_shared_impl: 125f0f90a0d3 SchedulingMaster.c

view SchedulingMaster.c @ 228:125f0f90a0d3

integrating holistic model in common ancestor - not finished yet

author	Nina Engelhardt <nengel@mailbox.tu-berlin.de>
date	Thu, 15 Mar 2012 18:28:29 +0100
parents	b0b93147adfb
children

line source

1 /*

3 *

4 * Licensed under BSD

5 */

9 #include <stdio.h>

10 #include <stddef.h>

12 #include "VMS.h"

15 //===========================================================================

16 void inline

17 stealWorkInto( SchedSlot *currSlot, VMSQueueStruc *readyToAnimateQ,

18 SlaveVP *masterVP );

20 //===========================================================================

24 /*This code is animated by the virtual Master processor.

25 *

26 *Polls each sched slot exactly once, hands any requests made by a newly

27 * done slave to the "request handler" plug-in function

28 *

29 *Any slots that need a Slv assigned are given to the "assign"

30 * plug-in function, which tries to assign a Slv (slave) to it.

31 *

32 *When all slots needing a processor have been given to the assign plug-in,

33 * a fraction of the slaves successfully assigned are put into the

34 * work queue, then a continuation of this function is put in, then the rest

35 * of the Slvs that were successfully assigned.

36 *

37 *The first thing the continuation does is busy-wait until the previous

38 * animation completes. This is because an (unlikely) continuation may

39 * sneak through queue before previous continuation is done putting second

40 * part of assigned slaves in, which is the only race condition.

41 *

42 */

44 /*May 29, 2010 -- birth a Master during init so that first core controller to

45 * start running gets it and does all the stuff for a newly born --

46 * from then on, will be doing continuation, but do suspension self

47 * directly at end of master loop

48 *So VMS_WL__init just births the master virtual processor same way it births

49 * all the others -- then does any extra setup needed and puts it into the

50 * work queue.

51 *However means have to make masterEnv a global static volatile the same way

52 * did with readyToAnimateQ in core controller. -- for performance, put the

53 * jump to the core controller directly in here, and have it directly jump back.

54 *

55 *

56 *Aug 18, 2010 -- Going to a separate MasterVP for each core, to see if this

57 * avoids the suspected bug in the system stack that causes bizarre faults

58 * at random places in the system code.

59 *

60 *So, this function is coupled to each of the MasterVPs, -- meaning this

61 * function can't rely on a particular stack and frame -- each MasterVP that

62 * animates this function has a different one.

63 *

64 *At this point, the schedulingMaster does not write itself into the queue anymore,

65 * instead, the coreCtlr acquires the masterLock when it has nothing to

66 * animate, and then animates its own schedulingMaster. However, still try to put

67 * several AppSlvs into the queue to amortize the startup cost of switching

68 * to the MasterVP. Note, don't have to worry about latency of requests much

69 * because most requests generate work for same core -- only latency issue

70 * is case when other cores starved and one core's requests generate work

71 * for them -- so keep max in queue to 3 or 4..

72 */

73 void schedulingMaster( void *initData, SlaveVP *animatingSlv )

74 {

75 int32 slotIdx, numSlotsFilled;

76 SlaveVP *schedSlaveVP;

77 SchedSlot *currSlot, **schedSlots;

78 MasterEnv *masterEnv;

79 VMSQueueStruc *readyToAnimateQ;

81 SlaveAssigner slaveAssigner;

82 RequestHandler requestHandler;

83 void *semanticEnv;

85 int32 thisCoresIdx;

86 SlaveVP *masterVP;

87 volatile SlaveVP *volatileMasterVP;

89 volatileMasterVP = animatingSlv;

90 masterVP = (SlaveVP*)volatileMasterVP; //used to force re-define after jmp

92 //First animation of each MasterVP will in turn animate this part

93 // of setup code.. (Slv creator sets up the stack as if this function

94 // was called normally, but actually get here by jmp)

95 //So, setup values about stack ptr, jmp pt and all that

96 //masterVP->resumeInstrPtr = &&schedulingMasterStartPt;

99 //Note, got rid of writing the stack and frame ptr up here, because

100 // only one

101 // core can ever animate a given MasterVP, so don't need to communicate

102 // new frame and stack ptr to the MasterVP storage before a second

103 // version of that MasterVP can get animated on a different core.

104 //Also got rid of the busy-wait.

105

106

107 //schedulingMasterStartPt:

108 while(1){

109

110 MEAS__Capture_Pre_Master_Point

111

112 masterEnv = (MasterEnv*)_VMSMasterEnv;

113

114 //GCC may optimize so doesn't always re-define from frame-storage

115 masterVP = (SlaveVP*)volatileMasterVP; //just to make sure after jmp

116 thisCoresIdx = masterVP->coreAnimatedBy;

117 schedSlots = masterEnv->allSchedSlots[thisCoresIdx];

118

119 requestHandler = masterEnv->requestHandler;

120 slaveAssigner = masterEnv->slaveAssigner;

121 semanticEnv = masterEnv->semanticEnv;

122

123

124 //Poll each slot's Done flag

125 numSlotsFilled = 0;

126 for( slotIdx = 0; slotIdx < NUM_SCHED_SLOTS; slotIdx++)

127 {

128 currSlot = schedSlots[ slotIdx ];

129

130 if( currSlot->workIsDone )

131 {

132 currSlot->workIsDone = FALSE;

133 currSlot->needsSlaveAssigned = TRUE;

134

135 MEAS__startReqHdlr;

136

137 //process the requests made by the slave (held inside slave struc)

138 (*requestHandler)( currSlot->slaveAssignedToSlot, semanticEnv );

139

140 MEAS__endReqHdlr;

141 }

142 if( currSlot->needsSlaveAssigned )

143 { //give slot a new Slv

144 schedSlaveVP =

145 (*slaveAssigner)( semanticEnv, thisCoresIdx, currSlot );

146

147 if( schedSlaveVP != NULL )

148 { currSlot->slaveAssignedToSlot = schedSlaveVP;

149 schedSlaveVP->schedSlot = currSlot;

150 currSlot->needsSlaveAssigned = FALSE;

151 numSlotsFilled += 1;

152 }

153 }

154 }

155

156

157 #ifdef SYS__TURN_ON_WORK_STEALING

158 //If no slots filled, means no more work, look for work to steal.

159 if( numSlotsFilled == 0 )

160 { gateProtected_stealWorkInto( currSlot, readyToAnimateQ, masterVP );

161 }

162 #endif

163

164 MEAS__Capture_Post_Master_Point;

165

166 masterSwitchToCoreCtlr(animatingSlv);

167 flushRegisters();

168 }//MasterLoop

169

170

171 }

172

173

174

175 /*This has a race condition -- the coreloops are accessing their own queues

176 * at the same time that this work-stealer on a different core is trying to

177 */

178 void inline

179 stealWorkInto( SchedSlot *currSlot, VMSQueueStruc *readyToAnimateQ,

180 SlaveVP *masterVP )

181 {

182 SlaveVP *stolenSlv;

183 int32 coreIdx, i;

184 VMSQueueStruc *currQ;

185

186 stolenSlv = NULL;

187 coreIdx = masterVP->coreAnimatedBy;

188 for( i = 0; i < NUM_CORES -1; i++ )

189 {

190 if( coreIdx >= NUM_CORES -1 )

191 { coreIdx = 0;

192 }

193 else

194 { coreIdx++;

195 }

196 //TODO: fix this for coreCtlr scans slots

197 // currQ = _VMSMasterEnv->readyToAnimateQs[coreIdx];

198 if( numInVMSQ( currQ ) > 0 )

199 { stolenSlv = readVMSQ (currQ );

200 break;

201 }

202 }

203

204 if( stolenSlv != NULL )

205 { currSlot->slaveAssignedToSlot = stolenSlv;

206 stolenSlv->schedSlot = currSlot;

207 currSlot->needsSlaveAssigned = FALSE;

208

209 writeVMSQ( stolenSlv, readyToAnimateQ );

210 }

211 }

212

213 /*This algorithm makes the common case fast. Make the coreloop passive,

214 * and show its progress. Make the stealer control a gate that coreloop

215 * has to pass.

216 *To avoid interference, only one stealer at a time. Use a global

217 * stealer-lock.

218 *

219 *The pattern is based on a gate -- stealer shuts the gate, then monitors

220 * to be sure any already past make it all the way out, before starting.

221 *So, have a "progress" measure just before the gate, then have two after it,

222 * one is in a "waiting room" outside the gate, the other is at the exit.

223 *Then, the stealer first shuts the gate, then checks the progress measure

224 * outside it, then looks to see if the progress measure at the exit is the

225 * same. If yes, it knows the protected area is empty 'cause no other way

226 * to get in and the last to get in also exited.

227 *If the progress measure at the exit is not the same, then the stealer goes

228 * into a loop checking both the waiting-area and the exit progress-measures

229 * until one of them shows the same as the measure outside the gate. Might

230 * as well re-read the measure outside the gate each go around, just to be

231 * sure. It is guaranteed that one of the two will eventually match the one

232 * outside the gate.

233 *

234 *Here's an informal proof of correctness:

235 *The gate can be closed at any point, and have only four cases:

236 * 1) coreloop made it past the gate-closing but not yet past the exit

237 * 2) coreloop made it past the pre-gate progress update but not yet past

238 * the gate,

239 * 3) coreloop is right before the pre-gate update

240 * 4) coreloop is past the exit and far from the pre-gate update.

241 *

242 * Covering the cases in reverse order,

243 * 4) is not a problem -- stealer will read pre-gate progress, see that it

244 * matches exit progress, and the gate is closed, so stealer can proceed.

245 * 3) stealer will read pre-gate progress just after coreloop updates it..

246 * so stealer goes into a loop until the coreloop causes wait-progress

247 * to match pre-gate progress, so then stealer can proceed

248 * 2) same as 3..

249 * 1) stealer reads pre-gate progress, sees that it's different than exit,

250 * so goes into loop until exit matches pre-gate, now it knows coreloop

251 * is not in protected and cannot get back in, so can proceed.

252 *

253 *Implementation for the stealer:

254 *

255 *First, acquire the stealer lock -- only cores with no work to do will

256 * compete to steal, so not a big performance penalty having only one --

257 * will rarely have multiple stealers in a system with plenty of work -- and

258 * in a system with little work, it doesn't matter.

259 *

260 *Note, have single-reader, single-writer pattern for all variables used to

261 * communicate between stealer and victims

262 *

263 *So, scan the queues of the core controllers, until find non-empty. Each core

264 * has its own list that it scans. The list goes in order from closest to

265 * furthest core, so it steals first from close cores. Later can add

266 * taking info from the app about overlapping footprints, and scan all the

267 * others then choose work with the most footprint overlap with the contents

268 * of this core's cache.

269 *

270 *Now, have a victim want to take work from. So, shut the gate in that

271 * coreloop, by setting the "gate closed" var on its stack to TRUE.

272 *Then, read the core's pre-gate progress and compare to the core's exit

273 * progress.

274 *If same, can proceed to take work from the coreloop's queue. When done,

275 * write FALSE to gate closed var.

276 *If different, then enter a loop that reads the pre-gate progress, then

277 * compares to exit progress then to wait progress. When one of two

278 * matches, proceed. Take work from the coreloop's queue. When done,

279 * write FALSE to the gate closed var.

280 *

281 */

282 void inline

283 gateProtected_stealWorkInto( SchedSlot *currSlot,

284 VMSQueueStruc *myReadyToAnimateQ,

285 SlaveVP *masterVP )

286 {

287 SlaveVP *stolenSlv;

288 int32 coreIdx, i, haveAVictim, gotLock;

289 VMSQueueStruc *victimsQ;

290

291 volatile GateStruc *vicGate;

292 int32 coreMightBeInProtected;

293

294

295

296 //see if any other cores have work available to steal

297 haveAVictim = FALSE;

298 coreIdx = masterVP->coreAnimatedBy;

299 for( i = 0; i < NUM_CORES -1; i++ )

300 {

301 if( coreIdx >= NUM_CORES -1 )

302 { coreIdx = 0;

303 }

304 else

305 { coreIdx++;

306 }

307 //TODO: fix this for coreCtlr scans slots

308 // victimsQ = _VMSMasterEnv->readyToAnimateQs[coreIdx];

309 if( numInVMSQ( victimsQ ) > 0 )

310 { haveAVictim = TRUE;

311 vicGate = _VMSMasterEnv->workStealingGates[ coreIdx ];

312 break;

313 }

314 }

315 if( !haveAVictim ) return; //no work to steal, exit

316

317 //have a victim core, now get the stealer-lock

318 gotLock =__sync_bool_compare_and_swap( &(_VMSMasterEnv->workStealingLock),

319 UNLOCKED, LOCKED );

320 if( !gotLock ) return; //go back to core controller, which will re-start master

321

322

323 //====== Start Gate-protection =======

324 vicGate->gateClosed = TRUE;

325 coreMightBeInProtected= vicGate->preGateProgress != vicGate->exitProgress;

326 while( coreMightBeInProtected )

327 { //wait until sure

328 if( vicGate->preGateProgress == vicGate->waitProgress )

329 coreMightBeInProtected = FALSE;

330 if( vicGate->preGateProgress == vicGate->exitProgress )

331 coreMightBeInProtected = FALSE;

332 }

333

334 stolenSlv = readVMSQ ( victimsQ );

335

336 vicGate->gateClosed = FALSE;

337 //======= End Gate-protection =======

338

339

340 if( stolenSlv != NULL ) //victim could have been in protected and taken

341 { currSlot->slaveAssignedToSlot = stolenSlv;

342 stolenSlv->schedSlot = currSlot;

343 currSlot->needsSlaveAssigned = FALSE;

344

345 writeVMSQ( stolenSlv, myReadyToAnimateQ );

346 }

347

348 //unlock the work stealing lock

349 _VMSMasterEnv->workStealingLock = UNLOCKED;

350 }

Mercurial > cgi-bin > hgwebdir.cgi > VMS > VMS_Implementations > VMS_impls > VMS__MC_shared_impl

view SchedulingMaster.c @ 228:125f0f90a0d3