/* * Copyright 2010 OpenSourceStewardshipFoundation * * Licensed under BSD */ #include #include #include "VMS.h" #include "ProcrContext.h" //=========================================================================== void inline stealWorkInto( SchedSlot *currSlot, VMSQueueStruc *readyToAnimateQ, VirtProcr *masterPr ); //=========================================================================== /*This code is animated by the virtual Master processor. * *Polls each sched slot exactly once, hands any requests made by a newly * done slave to the "request handler" plug-in function * *Any slots that need a virt procr assigned are given to the "schedule" * plug-in function, which tries to assign a virt procr (slave) to it. * *When all slots needing a processor have been given to the schedule plug-in, * a fraction of the procrs successfully scheduled are put into the * work queue, then a continuation of this function is put in, then the rest * of the virt procrs that were successfully scheduled. * *The first thing the continuation does is busy-wait until the previous * animation completes. This is because an (unlikely) continuation may * sneak through queue before previous continuation is done putting second * part of scheduled slaves in, which is the only race condition. * */ /*May 29, 2010 -- birth a Master during init so that first core loop to * start running gets it and does all the stuff for a newly born -- * from then on, will be doing continuation, but do suspension self * directly at end of master loop *So VMS__init just births the master virtual processor same way it births * all the others -- then does any extra setup needed and puts it into the * work queue. *However means have to make masterEnv a global static volatile the same way * did with readyToAnimateQ in core loop. -- for performance, put the * jump to the core loop directly in here, and have it directly jump back. * * *Aug 18, 2010 -- Going to a separate MasterVP for each core, to see if this * avoids the suspected bug in the system stack that causes bizarre faults * at random places in the system code. * *So, this function is coupled to each of the MasterVPs, -- meaning this * function can't rely on a particular stack and frame -- each MasterVP that * animates this function has a different one. * *At this point, the masterLoop does not write itself into the queue anymore, * instead, the coreLoop acquires the masterLock when it has nothing to * animate, and then animates its own masterLoop. However, still try to put * several AppVPs into the queue to amortize the startup cost of switching * to the MasterVP. Note, don't have to worry about latency of requests much * because most requests generate work for same core -- only latency issue * is case when other cores starved and one core's requests generate work * for them -- so keep max in queue to 3 or 4.. */ void masterLoop( void *initData, VirtProcr *animatingPr ) { int32 slotIdx, numSlotsFilled; VirtProcr *schedVirtPr; SchedSlot *currSlot, **schedSlots; MasterEnv *masterEnv; VMSQueueStruc *readyToAnimateQ; SlaveScheduler slaveScheduler; RequestHandler requestHandler; void *semanticEnv; int32 thisCoresIdx; VirtProcr *masterPr; volatile VirtProcr *volatileMasterPr; volatileMasterPr = animatingPr; masterPr = (VirtProcr*)volatileMasterPr; //used to force re-define after jmp //First animation of each MasterVP will in turn animate this part // of setup code.. (VP creator sets up the stack as if this function // was called normally, but actually get here by jmp) //So, setup values about stack ptr, jmp pt and all that //masterPr->nextInstrPt = &&masterLoopStartPt; //Note, got rid of writing the stack and frame ptr up here, because // only one // core can ever animate a given MasterVP, so don't need to communicate // new frame and stack ptr to the MasterVP storage before a second // version of that MasterVP can get animated on a different core. //Also got rid of the busy-wait. //masterLoopStartPt: while(1){ //============================= MEASUREMENT STUFF ======================== #ifdef MEAS__TIME_MASTER //Total Master time includes one coreloop time -- just assume the core // loop time is same for Master as for AppVPs, even though it may be // smaller due to higher predictability of the fixed jmp. saveLowTimeStampCountInto( masterPr->startMasterTSCLow ); #endif //======================================================================== masterEnv = (MasterEnv*)_VMSMasterEnv; //GCC may optimize so doesn't always re-define from frame-storage masterPr = (VirtProcr*)volatileMasterPr; //just to make sure after jmp thisCoresIdx = masterPr->coreAnimatedBy; readyToAnimateQ = masterEnv->readyToAnimateQs[thisCoresIdx]; schedSlots = masterEnv->allSchedSlots[thisCoresIdx]; requestHandler = masterEnv->requestHandler; slaveScheduler = masterEnv->slaveScheduler; semanticEnv = masterEnv->semanticEnv; //Poll each slot's Done flag numSlotsFilled = 0; for( slotIdx = 0; slotIdx < NUM_SCHED_SLOTS; slotIdx++) { currSlot = schedSlots[ slotIdx ]; if( currSlot->workIsDone ) { currSlot->workIsDone = FALSE; currSlot->needsProcrAssigned = TRUE; //process requests from slave to master //====================== MEASUREMENT STUFF =================== #ifdef MEAS__TIME_PLUGIN int32 startStamp1, endStamp1; saveLowTimeStampCountInto( startStamp1 ); #endif //============================================================ (*requestHandler)( currSlot->procrAssignedToSlot, semanticEnv ); //====================== MEASUREMENT STUFF =================== #ifdef MEAS__TIME_PLUGIN saveLowTimeStampCountInto( endStamp1 ); addIntervalToHist( startStamp1, endStamp1, _VMSMasterEnv->reqHdlrLowTimeHist ); addIntervalToHist( startStamp1, endStamp1, _VMSMasterEnv->reqHdlrHighTimeHist ); #endif //============================================================ } if( currSlot->needsProcrAssigned ) { //give slot a new virt procr schedVirtPr = (*slaveScheduler)( semanticEnv, thisCoresIdx ); if( schedVirtPr != NULL ) { currSlot->procrAssignedToSlot = schedVirtPr; schedVirtPr->schedSlot = currSlot; currSlot->needsProcrAssigned = FALSE; numSlotsFilled += 1; writeVMSQ( schedVirtPr, readyToAnimateQ ); } } } #ifdef USE_WORK_STEALING //If no slots filled, means no more work, look for work to steal. if( numSlotsFilled == 0 ) { gateProtected_stealWorkInto( currSlot, readyToAnimateQ, masterPr ); } #endif #ifdef MEAS__TIME_MASTER saveLowTimeStampCountInto( masterPr->endMasterTSCLow ); #endif masterSwitchToCoreLoop(animatingPr); flushRegisters(); }//MasterLoop } /*This has a race condition -- the coreloops are accessing their own queues * at the same time that this work-stealer on a different core is trying to */ void inline stealWorkInto( SchedSlot *currSlot, VMSQueueStruc *readyToAnimateQ, VirtProcr *masterPr ) { VirtProcr *stolenPr; int32 coreIdx, i; VMSQueueStruc *currQ; stolenPr = NULL; coreIdx = masterPr->coreAnimatedBy; for( i = 0; i < NUM_CORES -1; i++ ) { if( coreIdx >= NUM_CORES -1 ) { coreIdx = 0; } else { coreIdx++; } currQ = _VMSMasterEnv->readyToAnimateQs[coreIdx]; if( numInVMSQ( currQ ) > 0 ) { stolenPr = readVMSQ (currQ ); break; } } if( stolenPr != NULL ) { currSlot->procrAssignedToSlot = stolenPr; stolenPr->schedSlot = currSlot; currSlot->needsProcrAssigned = FALSE; writeVMSQ( stolenPr, readyToAnimateQ ); } } /*This algorithm makes the common case fast. Make the coreloop passive, * and show its progress. Make the stealer control a gate that coreloop * has to pass. *To avoid interference, only one stealer at a time. Use a global * stealer-lock. * *The pattern is based on a gate -- stealer shuts the gate, then monitors * to be sure any already past make it all the way out, before starting. *So, have a "progress" measure just before the gate, then have two after it, * one is in a "waiting room" outside the gate, the other is at the exit. *Then, the stealer first shuts the gate, then checks the progress measure * outside it, then looks to see if the progress measure at the exit is the * same. If yes, it knows the protected area is empty 'cause no other way * to get in and the last to get in also exited. *If the progress measure at the exit is not the same, then the stealer goes * into a loop checking both the waiting-area and the exit progress-measures * until one of them shows the same as the measure outside the gate. Might * as well re-read the measure outside the gate each go around, just to be * sure. It is guaranteed that one of the two will eventually match the one * outside the gate. * *Here's an informal proof of correctness: *The gate can be closed at any point, and have only four cases: * 1) coreloop made it past the gate-closing but not yet past the exit * 2) coreloop made it past the pre-gate progress update but not yet past * the gate, * 3) coreloop is right before the pre-gate update * 4) coreloop is past the exit and far from the pre-gate update. * * Covering the cases in reverse order, * 4) is not a problem -- stealer will read pre-gate progress, see that it * matches exit progress, and the gate is closed, so stealer can proceed. * 3) stealer will read pre-gate progress just after coreloop updates it.. * so stealer goes into a loop until the coreloop causes wait-progress * to match pre-gate progress, so then stealer can proceed * 2) same as 3.. * 1) stealer reads pre-gate progress, sees that it's different than exit, * so goes into loop until exit matches pre-gate, now it knows coreloop * is not in protected and cannot get back in, so can proceed. * *Implementation for the stealer: * *First, acquire the stealer lock -- only cores with no work to do will * compete to steal, so not a big performance penalty having only one -- * will rarely have multiple stealers in a system with plenty of work -- and * in a system with little work, it doesn't matter. * *Note, have single-reader, single-writer pattern for all variables used to * communicate between stealer and victims * *So, scan the queues of the core loops, until find non-empty. Each core * has its own list that it scans. The list goes in order from closest to * furthest core, so it steals first from close cores. Later can add * taking info from the app about overlapping footprints, and scan all the * others then choose work with the most footprint overlap with the contents * of this core's cache. * *Now, have a victim want to take work from. So, shut the gate in that * coreloop, by setting the "gate closed" var on its stack to TRUE. *Then, read the core's pre-gate progress and compare to the core's exit * progress. *If same, can proceed to take work from the coreloop's queue. When done, * write FALSE to gate closed var. *If different, then enter a loop that reads the pre-gate progress, then * compares to exit progress then to wait progress. When one of two * matches, proceed. Take work from the coreloop's queue. When done, * write FALSE to the gate closed var. * */ void inline gateProtected_stealWorkInto( SchedSlot *currSlot, VMSQueueStruc *myReadyToAnimateQ, VirtProcr *masterPr ) { VirtProcr *stolenPr; int32 coreIdx, i, haveAVictim, gotLock; VMSQueueStruc *victimsQ; volatile GateStruc *vicGate; int32 coreMightBeInProtected; //see if any other cores have work available to steal haveAVictim = FALSE; coreIdx = masterPr->coreAnimatedBy; for( i = 0; i < NUM_CORES -1; i++ ) { if( coreIdx >= NUM_CORES -1 ) { coreIdx = 0; } else { coreIdx++; } victimsQ = _VMSMasterEnv->readyToAnimateQs[coreIdx]; if( numInVMSQ( victimsQ ) > 0 ) { haveAVictim = TRUE; vicGate = _VMSMasterEnv->workStealingGates[ coreIdx ]; break; } } if( !haveAVictim ) return; //no work to steal, exit //have a victim core, now get the stealer-lock gotLock =__sync_bool_compare_and_swap( &(_VMSMasterEnv->workStealingLock), UNLOCKED, LOCKED ); if( !gotLock ) return; //go back to core loop, which will re-start master //====== Start Gate-protection ======= vicGate->gateClosed = TRUE; coreMightBeInProtected= vicGate->preGateProgress != vicGate->exitProgress; while( coreMightBeInProtected ) { //wait until sure if( vicGate->preGateProgress == vicGate->waitProgress ) coreMightBeInProtected = FALSE; if( vicGate->preGateProgress == vicGate->exitProgress ) coreMightBeInProtected = FALSE; } stolenPr = readVMSQ ( victimsQ ); vicGate->gateClosed = FALSE; //======= End Gate-protection ======= if( stolenPr != NULL ) //victim could have been in protected and taken { currSlot->procrAssignedToSlot = stolenPr; stolenPr->schedSlot = currSlot; currSlot->needsProcrAssigned = FALSE; writeVMSQ( stolenPr, myReadyToAnimateQ ); } //unlock the work stealing lock _VMSMasterEnv->workStealingLock = UNLOCKED; }