From c355df5bfea757604113104c99998fb232539a5d Mon Sep 17 00:00:00 2001 From: Korey Sewell Date: Fri, 7 Jul 2006 04:06:26 -0400 Subject: Fix so that O3CPU doesnt segfault on exit. Major thing was to not execute commit if there are no active threads in CPU. src/cpu/o3/alpha/thread_context.hh: call deallocate instead of deallocateContext src/cpu/o3/commit_impl.hh: dont run commit stage if there are no instructions src/cpu/o3/cpu.cc: add deallocate event, deactivateThread function, and edit deallocateContext. src/cpu/o3/cpu.hh: add deallocate event and add optional delay to deallocateContext src/cpu/o3/thread_context.hh: optional delay for deallocate src/cpu/o3/thread_context_impl.hh: edit DPRINTFs to say Thread Context instead of Alpha TC src/cpu/thread_context.hh: optional delay src/sim/syscall_emul.hh: name stuff --HG-- extra : convert_revision : f4033e1f66b3043d30ad98dcc70d8b193dea70b6 --- src/cpu/o3/alpha/thread_context.hh | 15 +-- src/cpu/o3/commit_impl.hh | 3 + src/cpu/o3/cpu.cc | 249 ++++++++++++++++++++++--------------- src/cpu/o3/cpu.hh | 50 +++++++- src/cpu/o3/thread_context.hh | 2 +- src/cpu/o3/thread_context_impl.hh | 16 ++- src/cpu/thread_context.hh | 4 +- 7 files changed, 218 insertions(+), 121 deletions(-) (limited to 'src/cpu') diff --git a/src/cpu/o3/alpha/thread_context.hh b/src/cpu/o3/alpha/thread_context.hh index 78b0ee788..ad52b0d2e 100644 --- a/src/cpu/o3/alpha/thread_context.hh +++ b/src/cpu/o3/alpha/thread_context.hh @@ -70,18 +70,19 @@ class AlphaTC : public O3ThreadContext { panic("Not supported on Alpha!"); } - // This function exits the thread context in the CPU and returns - // 1 if the CPU has no more active threads (meaning it's OK to exit); - // Used in syscall-emulation mode when a thread executes the 'exit' - // syscall. + /** This function exits the thread context in the CPU and returns + * 1 if the CPU has no more active threads (meaning it's OK to exit); + * Used in syscall-emulation mode when a thread executes the 'exit' + * syscall. + */ virtual int exit() { - this->cpu->deallocateContext(this->thread->readTid()); + this->deallocate(); // If there are still threads executing in the system if (this->cpu->numActiveThreads()) - return 0; + return 0; // don't exit simulation else - return 1; + return 1; // exit simulation } }; diff --git a/src/cpu/o3/commit_impl.hh b/src/cpu/o3/commit_impl.hh index 06b8e8a95..53d247e97 100644 --- a/src/cpu/o3/commit_impl.hh +++ b/src/cpu/o3/commit_impl.hh @@ -562,6 +562,9 @@ DefaultCommit::tick() return; } + if ((*activeThreads).size() <= 0) + return; + list::iterator threads = (*activeThreads).begin(); // Check if any of the threads are done squashing. Change the diff --git a/src/cpu/o3/cpu.cc b/src/cpu/o3/cpu.cc index c88146fa6..0a564169a 100644 --- a/src/cpu/o3/cpu.cc +++ b/src/cpu/o3/cpu.cc @@ -114,6 +114,36 @@ FullO3CPU::ActivateThreadEvent::description() return "FullO3CPU \"Activate Thread\" event"; } +template +FullO3CPU::DeallocateContextEvent::DeallocateContextEvent() + : Event(&mainEventQueue, CPU_Tick_Pri) +{ +} + +template +void +FullO3CPU::DeallocateContextEvent::init(int thread_num, + FullO3CPU *thread_cpu) +{ + tid = thread_num; + cpu = thread_cpu; +} + +template +void +FullO3CPU::DeallocateContextEvent::process() +{ + cpu->deactivateThread(tid); + cpu->removeThread(tid); +} + +template +const char * +FullO3CPU::DeallocateContextEvent::description() +{ + return "FullO3CPU \"Deallocate Context\" event"; +} + template FullO3CPU::FullO3CPU(Params *params) : BaseO3CPU(params), @@ -459,6 +489,118 @@ FullO3CPU::init() commit.setThreads(thread); } +template +void +FullO3CPU::activateThread(unsigned tid) +{ + list::iterator isActive = find( + activeThreads.begin(), activeThreads.end(), tid); + + if (isActive == activeThreads.end()) { + DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", + tid); + + activeThreads.push_back(tid); + } +} + +template +void +FullO3CPU::deactivateThread(unsigned tid) +{ + //Remove From Active List, if Active + list::iterator thread_it = + find(activeThreads.begin(), activeThreads.end(), tid); + + if (thread_it != activeThreads.end()) { + DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", + tid); + activeThreads.erase(thread_it); + } +} + +template +void +FullO3CPU::activateContext(int tid, int delay) +{ + // Needs to set each stage to running as well. + if (delay){ + DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " + "on cycle %d\n", tid, curTick + cycles(delay)); + scheduleActivateThreadEvent(tid, delay); + } else { + activateThread(tid); + } + + if(lastActivatedCycle < curTick) { + scheduleTickEvent(delay); + + // Be sure to signal that there's some activity so the CPU doesn't + // deschedule itself. + activityRec.activity(); + fetch.wakeFromQuiesce(); + + lastActivatedCycle = curTick; + + _status = Running; + } +} + +template +void +FullO3CPU::deallocateContext(int tid, int delay) +{ + // Schedule removal of thread data from CPU + if (delay){ + DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " + "on cycle %d\n", tid, curTick + cycles(delay)); + scheduleDeallocateContextEvent(tid, delay); + } else { + deactivateThread(tid); + removeThread(tid); + } +} + +template +void +FullO3CPU::suspendContext(int tid) +{ + DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); + unscheduleTickEvent(); + _status = Idle; +/* + //Remove From Active List, if Active + list::iterator isActive = find( + activeThreads.begin(), activeThreads.end(), tid); + + if (isActive != activeThreads.end()) { + DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", + tid); + activeThreads.erase(isActive); + } +*/ +} + +template +void +FullO3CPU::haltContext(int tid) +{ + DPRINTF(O3CPU,"[tid:%i]: Halting Thread Context", tid); +/* + //Remove From Active List, if Active + list::iterator isActive = find( + activeThreads.begin(), activeThreads.end(), tid); + + if (isActive != activeThreads.end()) { + DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", + tid); + activeThreads.erase(isActive); + + removeThread(tid); + } +*/ +} + template void FullO3CPU::insertThread(unsigned tid) @@ -511,7 +653,7 @@ template void FullO3CPU::removeThread(unsigned tid) { - DPRINTF(O3CPU,"[tid:%i] Removing thread from CPU."); + DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU."); // Copy Thread Data From RegFile // If thread is suspended, it might be re-allocated @@ -537,6 +679,8 @@ FullO3CPU::removeThread(unsigned tid) fetch.squash(0,tid); decode.squash(tid); rename.squash(tid); + iew.squash(tid); + commit.rob->squash(commit.rob->readHeadInst(tid)->seqNum, tid); assert(iew.ldstQueue.getCount(tid) == 0); @@ -600,113 +744,12 @@ FullO3CPU::activateWhenReady(int tid) //blocks fetch contextSwitch = true; + //@todo: dont always add to waitlist //do waitlist cpuWaitList.push_back(tid); } } -template -void -FullO3CPU::activateThread(unsigned int tid) -{ - list::iterator isActive = find( - activeThreads.begin(), activeThreads.end(), tid); - - if (isActive == activeThreads.end()) { - DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", - tid); - - activeThreads.push_back(tid); - } -} - - -template -void -FullO3CPU::activateContext(int tid, int delay) -{ - // Needs to set each stage to running as well. - if (delay){ - DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " - "on cycle %d\n", tid, curTick + cycles(delay)); - scheduleActivateThreadEvent(tid, delay); - } else { - activateThread(tid); - } - - if(lastActivatedCycle < curTick) { - scheduleTickEvent(delay); - - // Be sure to signal that there's some activity so the CPU doesn't - // deschedule itself. - activityRec.activity(); - fetch.wakeFromQuiesce(); - - lastActivatedCycle = curTick; - - _status = Running; - } -} - -template -void -FullO3CPU::suspendContext(int tid) -{ - DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); - unscheduleTickEvent(); - _status = Idle; -/* - //Remove From Active List, if Active - list::iterator isActive = find( - activeThreads.begin(), activeThreads.end(), tid); - - if (isActive != activeThreads.end()) { - DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", - tid); - activeThreads.erase(isActive); - } -*/ -} - -template -void -FullO3CPU::deallocateContext(int tid) -{ - DPRINTF(O3CPU,"[tid:%i]: Deallocating Thread Context", tid); - - //Remove From Active List, if Active - list::iterator thread_it = - find(activeThreads.begin(), activeThreads.end(), tid); - - if (thread_it != activeThreads.end()) { - DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", - tid); - activeThreads.erase(thread_it); - - removeThread(tid); - } -} - -template -void -FullO3CPU::haltContext(int tid) -{ - DPRINTF(O3CPU,"[tid:%i]: Halting Thread Context", tid); -/* - //Remove From Active List, if Active - list::iterator isActive = find( - activeThreads.begin(), activeThreads.end(), tid); - - if (isActive != activeThreads.end()) { - DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", - tid); - activeThreads.erase(isActive); - - removeThread(tid); - } -*/ -} - template void FullO3CPU::switchOut() diff --git a/src/cpu/o3/cpu.hh b/src/cpu/o3/cpu.hh index bd0451601..476b5ffb3 100644 --- a/src/cpu/o3/cpu.hh +++ b/src/cpu/o3/cpu.hh @@ -197,6 +197,49 @@ class FullO3CPU : public BaseO3CPU /** The tick event used for scheduling CPU ticks. */ ActivateThreadEvent activateThreadEvent[Impl::MaxThreads]; + class DeallocateContextEvent : public Event + { + private: + /** Number of Thread to Activate */ + int tid; + + /** Pointer to the CPU. */ + FullO3CPU *cpu; + + public: + /** Constructs the event. */ + DeallocateContextEvent(); + + /** Initialize Event */ + void init(int thread_num, FullO3CPU *thread_cpu); + + /** Processes the event, calling activateThread() on the CPU. */ + void process(); + + /** Returns the description of the event. */ + const char *description(); + }; + + /** Schedule cpu to deallocate thread context.*/ + void scheduleDeallocateContextEvent(int tid, int delay) + { + // Schedule thread to activate, regardless of its current state. + if (deallocateContextEvent[tid].squashed()) + deallocateContextEvent[tid].reschedule(curTick + cycles(delay)); + else if (!deallocateContextEvent[tid].scheduled()) + deallocateContextEvent[tid].schedule(curTick + cycles(delay)); + } + + /** Unschedule thread deallocation in CPU */ + void unscheduleDeallocateContextEvent(int tid) + { + if (deallocateContextEvent[tid].scheduled()) + deallocateContextEvent[tid].squash(); + } + + /** The tick event used for scheduling CPU ticks. */ + DeallocateContextEvent deallocateContextEvent[Impl::MaxThreads]; + public: /** Constructs a CPU with the given parameters. */ FullO3CPU(Params *params); @@ -219,7 +262,10 @@ class FullO3CPU : public BaseO3CPU { return activeThreads.size(); } /** Add Thread to Active Threads List */ - void activateThread(unsigned int tid); + void activateThread(unsigned tid); + + /** Remove Thread from Active Threads List */ + void deactivateThread(unsigned tid); /** Setup CPU to insert a thread's context */ void insertThread(unsigned tid); @@ -247,7 +293,7 @@ class FullO3CPU : public BaseO3CPU /** Remove Thread from Active Threads List && * Remove Thread Context from CPU. */ - void deallocateContext(int tid); + void deallocateContext(int tid, int delay = 1); /** Remove Thread from Active Threads List && * Remove Thread Context from CPU. diff --git a/src/cpu/o3/thread_context.hh b/src/cpu/o3/thread_context.hh index d097ee63e..df8d1a6d8 100755 --- a/src/cpu/o3/thread_context.hh +++ b/src/cpu/o3/thread_context.hh @@ -112,7 +112,7 @@ class O3ThreadContext : public ThreadContext virtual void suspend(); /** Set the status to Unallocated. */ - virtual void deallocate(); + virtual void deallocate(int delay = 0); /** Set the status to Halted. */ virtual void halt(); diff --git a/src/cpu/o3/thread_context_impl.hh b/src/cpu/o3/thread_context_impl.hh index cfb71f623..bf8cbf850 100755 --- a/src/cpu/o3/thread_context_impl.hh +++ b/src/cpu/o3/thread_context_impl.hh @@ -115,7 +115,8 @@ template void O3ThreadContext::activate(int delay) { - DPRINTF(O3CPU, "Calling activate on AlphaTC\n"); + DPRINTF(O3CPU, "Calling activate on Thread Context %d\n", + getThreadNum()); if (thread->status() == ThreadContext::Active) return; @@ -139,7 +140,8 @@ template void O3ThreadContext::suspend() { - DPRINTF(O3CPU, "Calling suspend on AlphaTC\n"); + DPRINTF(O3CPU, "Calling suspend on Thread Context %d\n", + getThreadNum()); if (thread->status() == ThreadContext::Suspended) return; @@ -163,22 +165,24 @@ O3ThreadContext::suspend() template void -O3ThreadContext::deallocate() +O3ThreadContext::deallocate(int delay) { - DPRINTF(O3CPU, "Calling deallocate on AlphaTC\n"); + DPRINTF(O3CPU, "Calling deallocate on Thread Context %d\n", + getThreadNum()); if (thread->status() == ThreadContext::Unallocated) return; thread->setStatus(ThreadContext::Unallocated); - cpu->deallocateContext(thread->readTid()); + cpu->deallocateContext(thread->readTid(), delay); } template void O3ThreadContext::halt() { - DPRINTF(O3CPU, "Calling halt on AlphaTC\n"); + DPRINTF(O3CPU, "Calling halt on Thread Context %d\n", + getThreadNum()); if (thread->status() == ThreadContext::Halted) return; diff --git a/src/cpu/thread_context.hh b/src/cpu/thread_context.hh index 70d705144..e019e22bc 100644 --- a/src/cpu/thread_context.hh +++ b/src/cpu/thread_context.hh @@ -143,7 +143,7 @@ class ThreadContext virtual void suspend() = 0; /// Set the status to Unallocated. - virtual void deallocate() = 0; + virtual void deallocate(int delay = 0) = 0; /// Set the status to Halted. virtual void halt() = 0; @@ -318,7 +318,7 @@ class ProxyThreadContext : public ThreadContext void suspend() { actualTC->suspend(); } /// Set the status to Unallocated. - void deallocate() { actualTC->deallocate(); } + void deallocate(int delay = 0) { actualTC->deallocate(); } /// Set the status to Halted. void halt() { actualTC->halt(); } -- cgit v1.2.3 From 74d4d671386252d0d5f570b69fc63cb332757580 Mon Sep 17 00:00:00 2001 From: Korey Sewell Date: Fri, 7 Jul 2006 15:58:03 -0400 Subject: Minor fix for SMT Hello Worlds to finish correctly. Still, there is a problem with the LSQ and indexing out of range in the buffer. I havent nailed down the fix yet, but it's coming ... src/cpu/o3/commit_impl.hh: add space to DPRINT src/cpu/o3/cpu.cc: add newline to DPRINT src/cpu/o3/rob.hh: src/cpu/o3/rob_impl.hh: Each thread needs it's own squashedSeqNum for the case where they are both squashing at the same time and they dont write over each other's squash number. --HG-- extra : convert_revision : 2155421a8b5b20e4544eea3d3c53d3e715465fa6 --- src/cpu/o3/commit_impl.hh | 2 +- src/cpu/o3/cpu.cc | 2 +- src/cpu/o3/rob.hh | 2 +- src/cpu/o3/rob_impl.hh | 14 ++++++++------ 4 files changed, 11 insertions(+), 9 deletions(-) (limited to 'src/cpu') diff --git a/src/cpu/o3/commit_impl.hh b/src/cpu/o3/commit_impl.hh index dc2c8cbbb..e1f8e1f1e 100644 --- a/src/cpu/o3/commit_impl.hh +++ b/src/cpu/o3/commit_impl.hh @@ -585,7 +585,7 @@ DefaultCommit::tick() commitStatus[tid] = Running; } else { DPRINTF(Commit,"[tid:%u]: Still Squashing, cannot commit any" - "insts this cycle.\n", tid); + " insts this cycle.\n", tid); rob->doSquash(tid); toIEW->commitInfo[tid].robSquashing = true; wroteToTimeBuffer = true; diff --git a/src/cpu/o3/cpu.cc b/src/cpu/o3/cpu.cc index ec02a3929..c46276d5a 100644 --- a/src/cpu/o3/cpu.cc +++ b/src/cpu/o3/cpu.cc @@ -653,7 +653,7 @@ template void FullO3CPU::removeThread(unsigned tid) { - DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU."); + DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); // Copy Thread Data From RegFile // If thread is suspended, it might be re-allocated diff --git a/src/cpu/o3/rob.hh b/src/cpu/o3/rob.hh index 6f8080ef4..7cd5a5143 100644 --- a/src/cpu/o3/rob.hh +++ b/src/cpu/o3/rob.hh @@ -308,7 +308,7 @@ class ROB private: /** The sequence number of the squashed instruction. */ - InstSeqNum squashedSeqNum; + InstSeqNum squashedSeqNum[Impl::MaxThreads]; /** Is the ROB done squashing. */ bool doneSquashing[Impl::MaxThreads]; diff --git a/src/cpu/o3/rob_impl.hh b/src/cpu/o3/rob_impl.hh index d9978b17f..1b9f666b8 100644 --- a/src/cpu/o3/rob_impl.hh +++ b/src/cpu/o3/rob_impl.hh @@ -41,10 +41,10 @@ ROB::ROB(unsigned _numEntries, unsigned _squashWidth, : numEntries(_numEntries), squashWidth(_squashWidth), numInstsInROB(0), - squashedSeqNum(0), numThreads(_numThreads) { for (int tid=0; tid < numThreads; tid++) { + squashedSeqNum[tid] = 0; doneSquashing[tid] = true; threadEntries[tid] = 0; } @@ -352,11 +352,11 @@ void ROB::doSquash(unsigned tid) { DPRINTF(ROB, "[tid:%u]: Squashing instructions until [sn:%i].\n", - tid, squashedSeqNum); + tid, squashedSeqNum[tid]); assert(squashIt[tid] != instList[tid].end()); - if ((*squashIt[tid])->seqNum < squashedSeqNum) { + if ((*squashIt[tid])->seqNum < squashedSeqNum[tid]) { DPRINTF(ROB, "[tid:%u]: Done squashing instructions.\n", tid); @@ -371,7 +371,7 @@ ROB::doSquash(unsigned tid) for (int numSquashed = 0; numSquashed < squashWidth && squashIt[tid] != instList[tid].end() && - (*squashIt[tid])->seqNum > squashedSeqNum; + (*squashIt[tid])->seqNum > squashedSeqNum[tid]; ++numSquashed) { DPRINTF(ROB, "[tid:%u]: Squashing instruction PC %#x, seq num %i.\n", @@ -408,7 +408,7 @@ ROB::doSquash(unsigned tid) // Check if ROB is done squashing. - if ((*squashIt[tid])->seqNum <= squashedSeqNum) { + if ((*squashIt[tid])->seqNum <= squashedSeqNum[tid]) { DPRINTF(ROB, "[tid:%u]: Done squashing instructions.\n", tid); @@ -520,7 +520,7 @@ ROB::squash(InstSeqNum squash_num,unsigned tid) doneSquashing[tid] = false; - squashedSeqNum = squash_num; + squashedSeqNum[tid] = squash_num; if (!instList[tid].empty()) { InstIt tail_thread = instList[tid].end(); @@ -544,6 +544,7 @@ ROB::readHeadInst() } } */ + template typename Impl::DynInstPtr ROB::readHeadInst(unsigned tid) @@ -558,6 +559,7 @@ ROB::readHeadInst(unsigned tid) return dummyInst; } } + /* template uint64_t -- cgit v1.2.3