arch: Pass faults by const reference where possible

This patch changes how faults are passed between methods in an attempt
to copy as few reference-counting pointer instances as possible. This
should avoid unecessary copies being created, contributing to the
increment/decrement of the reference counters.

src/arch/arm/stage2_lookup.cc

@@ -171,7 +171,7 @@ Stage2LookUp::mergeTe(RequestPtr req, BaseTLB::Mode mode)
 }
 
 void
-Stage2LookUp::finish(Fault _fault, RequestPtr req,
+Stage2LookUp::finish(const Fault &_fault, RequestPtr req,
     ThreadContext *tc, BaseTLB::Mode mode)
 {
     fault = _fault;

src/arch/arm/stage2_lookup.hh

@@ -97,7 +97,7 @@ class Stage2LookUp : public BaseTLB::Translation
 
     void markDelayed() {}
 
-    void finish(Fault fault, RequestPtr req, ThreadContext *tc,
+    void finish(const Fault &fault, RequestPtr req, ThreadContext *tc,
                 BaseTLB::Mode mode);
 };
 

src/arch/arm/stage2_mmu.cc

@@ -114,8 +114,8 @@ Stage2MMU::Stage2Translation::Stage2Translation(Stage2MMU &_parent,
 }
 
 void
-Stage2MMU::Stage2Translation::finish(Fault _fault, RequestPtr req, ThreadContext *tc,
+Stage2MMU::Stage2Translation::finish(const Fault &_fault, RequestPtr req,
-    BaseTLB::Mode mode)
+                                     ThreadContext *tc, BaseTLB::Mode mode)
 {
     fault = _fault;
 

src/arch/arm/stage2_mmu.hh

@@ -78,7 +78,7 @@ class Stage2MMU : public SimObject
         markDelayed() {}
 
         void
-        finish(Fault fault, RequestPtr req, ThreadContext *tc,
+        finish(const Fault &fault, RequestPtr req, ThreadContext *tc,
                BaseTLB::Mode mode);
 
         void setVirt(Addr vaddr, int size, Request::Flags flags, int masterId)

src/cpu/checker/cpu.hh

@@ -426,7 +426,7 @@ class Checker : public CheckerCPU
     void switchOut();
     void takeOverFrom(BaseCPU *oldCPU);
 
-    void advancePC(Fault fault);
+    void advancePC(const Fault &fault);
 
     void verify(DynInstPtr &inst);
 

src/cpu/checker/cpu_impl.hh

@@ -69,7 +69,7 @@ using namespace TheISA;
 
 template <class Impl>
 void
-Checker<Impl>::advancePC(Fault fault)
+Checker<Impl>::advancePC(const Fault &fault)
 {
     if (fault != NoFault) {
         curMacroStaticInst = StaticInst::nullStaticInstPtr;

src/cpu/inorder/cpu.cc

@@ -128,8 +128,8 @@ InOrderCPU::TickEvent::description() const
 }
 
 InOrderCPU::CPUEvent::CPUEvent(InOrderCPU *_cpu, CPUEventType e_type,
-                               Fault fault, ThreadID _tid, DynInstPtr inst,
+                               const Fault &fault, ThreadID _tid,
-                               CPUEventPri event_pri)
+                               DynInstPtr inst, CPUEventPri event_pri)
     : Event(event_pri), cpu(_cpu)
 {
     setEvent(e_type, fault, _tid, inst);
@@ -910,7 +910,7 @@ InOrderCPU::getInterrupts()
 }
 
 void
-InOrderCPU::processInterrupts(Fault interrupt)
+InOrderCPU::processInterrupts(const Fault &interrupt)
 {
     // Check for interrupts here.  For now can copy the code that
     // exists within isa_fullsys_traits.hh.  Also assume that thread 0
@@ -928,7 +928,7 @@ InOrderCPU::processInterrupts(Fault interrupt)
 }
 
 void
-InOrderCPU::trapContext(Fault fault, ThreadID tid, DynInstPtr inst,
+InOrderCPU::trapContext(const Fault &fault, ThreadID tid, DynInstPtr inst,
                         Cycles delay)
 {
     scheduleCpuEvent(Trap, fault, tid, inst, delay);
@@ -936,7 +936,7 @@ InOrderCPU::trapContext(Fault fault, ThreadID tid, DynInstPtr inst,
 }
 
 void
-InOrderCPU::trap(Fault fault, ThreadID tid, DynInstPtr inst)
+InOrderCPU::trap(const Fault &fault, ThreadID tid, DynInstPtr inst)
 {
     fault->invoke(tcBase(tid), inst->staticInst);
     removePipelineStalls(tid);
@@ -970,7 +970,7 @@ InOrderCPU::squashDueToMemStall(int stage_num, InstSeqNum seq_num,
 }
 
 void
-InOrderCPU::scheduleCpuEvent(CPUEventType c_event, Fault fault,
+InOrderCPU::scheduleCpuEvent(CPUEventType c_event, const Fault &fault,
                              ThreadID tid, DynInstPtr inst, 
                              Cycles delay, CPUEventPri event_pri)
 {
@@ -1847,7 +1847,7 @@ InOrderCPU::wakeup()
 }
 
 void
-InOrderCPU::syscallContext(Fault fault, ThreadID tid, DynInstPtr inst,
+InOrderCPU::syscallContext(const Fault &fault, ThreadID tid, DynInstPtr inst,
                            Cycles delay)
 {
     // Syscall must be non-speculative, so squash from last stage

src/cpu/inorder/cpu.hh

@@ -263,11 +263,11 @@ class InOrderCPU : public BaseCPU
         
       public:
         /** Constructs a CPU event. */
-        CPUEvent(InOrderCPU *_cpu, CPUEventType e_type, Fault fault,
+        CPUEvent(InOrderCPU *_cpu, CPUEventType e_type, const Fault &fault,
                  ThreadID _tid, DynInstPtr inst, CPUEventPri event_pri);
 
         /** Set Type of Event To Be Scheduled */
-        void setEvent(CPUEventType e_type, Fault _fault, ThreadID _tid,
+        void setEvent(CPUEventType e_type, const Fault &_fault, ThreadID _tid,
                       DynInstPtr _inst)
         {
             fault = _fault;
@@ -291,7 +291,8 @@ class InOrderCPU : public BaseCPU
     };
 
     /** Schedule a CPU Event */
-    void scheduleCpuEvent(CPUEventType cpu_event, Fault fault, ThreadID tid,
+    void scheduleCpuEvent(CPUEventType cpu_event, const Fault &fault,
+                          ThreadID tid,
                           DynInstPtr inst, Cycles delay = Cycles(0),
                           CPUEventPri event_pri = InOrderCPU_Pri);
 
@@ -471,7 +472,7 @@ class InOrderCPU : public BaseCPU
     Fault getInterrupts();
 
     /** Processes any an interrupt fault. */
-    void processInterrupts(Fault interrupt);
+    void processInterrupts(const Fault &interrupt);
 
     /** Halts the CPU. */
     void halt() { panic("Halt not implemented!\n"); }
@@ -483,18 +484,18 @@ class InOrderCPU : public BaseCPU
     bool validDataAddr(Addr addr) { return true; }
 
     /** Schedule a syscall on the CPU */
-    void syscallContext(Fault fault, ThreadID tid, DynInstPtr inst,
+    void syscallContext(const Fault &fault, ThreadID tid, DynInstPtr inst,
                         Cycles delay = Cycles(0));
 
     /** Executes a syscall.*/
     void syscall(int64_t callnum, ThreadID tid);
 
     /** Schedule a trap on the CPU */
-    void trapContext(Fault fault, ThreadID tid, DynInstPtr inst,
+    void trapContext(const Fault &fault, ThreadID tid, DynInstPtr inst,
                      Cycles delay = Cycles(0));
 
     /** Perform trap to Handle Given Fault */
-    void trap(Fault fault, ThreadID tid, DynInstPtr inst);
+    void trap(const Fault &fault, ThreadID tid, DynInstPtr inst);
 
     /** Schedule thread activation on the CPU */
     void activateContext(ThreadID tid, Cycles delay = Cycles(0));

src/cpu/inorder/inorder_dyn_inst.cc

@@ -298,7 +298,7 @@ InOrderDynInst::hwrei()
 
 
 void
-InOrderDynInst::trap(Fault fault)
+InOrderDynInst::trap(const Fault &fault)
 {
     this->cpu->trap(fault, this->threadNumber, this);
 }

src/cpu/inorder/inorder_dyn_inst.hh

@@ -524,7 +524,7 @@ class InOrderDynInst : public ExecContext, public RefCounted
     /** Calls hardware return from error interrupt. */
     Fault hwrei();
     /** Traps to handle specified fault. */
-    void trap(Fault fault);
+    void trap(const Fault &fault);
     bool simPalCheck(int palFunc);
     short syscallNum;
 

src/cpu/inorder/resource.hh

@@ -104,7 +104,7 @@ class Resource {
     virtual void instGraduated(InstSeqNum seq_num, ThreadID tid) { }
 
     /** Post-processsing for Trap Generated from this instruction */
-    virtual void trap(Fault fault, ThreadID tid, DynInstPtr inst) { }
+    virtual void trap(const Fault &fault, ThreadID tid, DynInstPtr inst) { }
 
     /** Request usage of this resource. Returns a ResourceRequest object
      *  with all the necessary resource information

src/cpu/inorder/resource_pool.cc

@@ -206,7 +206,7 @@ ResourcePool::squash(DynInstPtr inst, int res_idx, InstSeqNum done_seq_num,
 }
 
 void
-ResourcePool::trap(Fault fault, ThreadID tid, DynInstPtr inst)
+ResourcePool::trap(const Fault &fault, ThreadID tid, DynInstPtr inst)
 {
     DPRINTF(Resource, "[tid:%i] Broadcasting Trap to all "
             "resources.\n", tid);

src/cpu/inorder/resource_pool.hh

@@ -193,7 +193,7 @@ class ResourcePool {
     void instGraduated(InstSeqNum seq_num, ThreadID tid);
 
     /** Broadcast trap to all resources */
-    void trap(Fault fault, ThreadID tid, DynInstPtr inst);
+    void trap(const Fault &fault, ThreadID tid, DynInstPtr inst);
 
     /** The number of instructions available that a resource can
      *  can still process.

src/cpu/inorder/resources/cache_unit.cc

@@ -405,7 +405,7 @@ CacheUnit::doTLBAccess(DynInstPtr inst, CacheReqPtr cache_req, int acc_size,
 }
 
 void
-CacheUnit::trap(Fault fault, ThreadID tid, DynInstPtr inst)
+CacheUnit::trap(const Fault &fault, ThreadID tid, DynInstPtr inst)
 {
     tlbBlocked[tid] = false;
 }

src/cpu/inorder/resources/cache_unit.hh

@@ -113,7 +113,7 @@ class CacheUnit : public Resource
 
     bool processSquash(CacheReqPacket *cache_pkt);
 
-    void trap(Fault fault, ThreadID tid, DynInstPtr inst);
+    void trap(const Fault &fault, ThreadID tid, DynInstPtr inst);
 
     void recvRetry();
     

src/cpu/inorder/resources/fetch_seq_unit.cc

@@ -304,7 +304,7 @@ FetchSeqUnit::suspendThread(ThreadID tid)
 }
 
 void
-FetchSeqUnit::trap(Fault fault, ThreadID tid, DynInstPtr inst)
+FetchSeqUnit::trap(const Fault &fault, ThreadID tid, DynInstPtr inst)
 {
     pcValid[tid] = true;
     pc[tid] = cpu->pcState(tid);

src/cpu/inorder/resources/fetch_seq_unit.hh

@@ -71,7 +71,7 @@ class FetchSeqUnit : public Resource {
                 InstSeqNum squash_seq_num, ThreadID tid);
 
     /** Update to correct PC from a trap */
-    void trap(Fault fault, ThreadID tid, DynInstPtr inst);
+    void trap(const Fault &fault, ThreadID tid, DynInstPtr inst);
 
   protected:
     unsigned instSize;

src/cpu/inorder/resources/fetch_unit.cc

@@ -574,7 +574,7 @@ FetchUnit::squashCacheRequest(CacheReqPtr req_ptr)
 }
 
 void
-FetchUnit::trap(Fault fault, ThreadID tid, DynInstPtr inst)
+FetchUnit::trap(const Fault &fault, ThreadID tid, DynInstPtr inst)
 {
     //@todo: per thread?
     decoder[tid]->reset();

src/cpu/inorder/resources/fetch_unit.hh

@@ -87,7 +87,7 @@ class FetchUnit : public CacheUnit
     /** Executes one of the commands from the "Command" enum */
     void execute(int slot_num);
 
-    void trap(Fault fault, ThreadID tid, DynInstPtr inst);
+    void trap(const Fault &fault, ThreadID tid, DynInstPtr inst);
 
     TheISA::Decoder *decoder[ThePipeline::MaxThreads];
 

src/cpu/minor/fetch1.cc

@@ -204,8 +204,8 @@ Fetch1::FetchRequest::makePacket()
 }
 
 void
-Fetch1::FetchRequest::finish(
+Fetch1::FetchRequest::finish(const Fault &fault_, RequestPtr request_,
-    Fault fault_, RequestPtr request_, ThreadContext *tc, BaseTLB::Mode mode)
+                             ThreadContext *tc, BaseTLB::Mode mode)
 {
     fault = fault_;
 

src/cpu/minor/fetch1.hh

@@ -163,8 +163,8 @@ class Fetch1 : public Named
         /** Interface for ITLB responses.  Populates self and then passes
          *  the request on to the ports' handleTLBResponse member
          *  function */
-        void finish(Fault fault_, RequestPtr request_, ThreadContext *tc,
+        void finish(const Fault &fault_, RequestPtr request_,
-            BaseTLB::Mode mode);
+                    ThreadContext *tc, BaseTLB::Mode mode);
 
       public:
         FetchRequest(Fetch1 &fetch_, InstId id_, TheISA::PCState pc_) :

src/cpu/minor/lsq.cc

@@ -226,8 +226,8 @@ LSQ::clearMemBarrier(MinorDynInstPtr inst)
 }
 
 void
-LSQ::SingleDataRequest::finish(Fault fault_, RequestPtr request_,
+LSQ::SingleDataRequest::finish(const Fault &fault_, RequestPtr request_,
-    ThreadContext *tc, BaseTLB::Mode mode)
+                               ThreadContext *tc, BaseTLB::Mode mode)
 {
     fault = fault_;
 
@@ -273,8 +273,8 @@ LSQ::SingleDataRequest::retireResponse(PacketPtr packet_)
 }
 
 void
-LSQ::SplitDataRequest::finish(Fault fault_, RequestPtr request_,
+LSQ::SplitDataRequest::finish(const Fault &fault_, RequestPtr request_,
-    ThreadContext *tc, BaseTLB::Mode mode)
+                              ThreadContext *tc, BaseTLB::Mode mode)
 {
     fault = fault_;
 

src/cpu/minor/lsq.hh

@@ -268,8 +268,8 @@ class LSQ : public Named
     {
       protected:
         /** TLB interace */
-        void finish(Fault fault_, RequestPtr request_, ThreadContext *tc,
+        void finish(const Fault &fault_, RequestPtr request_,
-            BaseTLB::Mode mode)
+                    ThreadContext *tc, BaseTLB::Mode mode)
         { }
 
       public:
@@ -329,8 +329,8 @@ class LSQ : public Named
     {
       protected:
         /** TLB interace */
-        void finish(Fault fault_, RequestPtr request_, ThreadContext *tc,
+        void finish(const Fault &fault_, RequestPtr request_,
-            BaseTLB::Mode mode);
+                    ThreadContext *tc, BaseTLB::Mode mode);
 
         /** Has my only packet been sent to the memory system but has not
          *  yet been responded to */
@@ -415,8 +415,8 @@ class LSQ : public Named
 
       protected:
         /** TLB response interface */
-        void finish(Fault fault_, RequestPtr request_, ThreadContext *tc,
+        void finish(const Fault &fault_, RequestPtr request_,
-            BaseTLB::Mode mode);
+                    ThreadContext *tc, BaseTLB::Mode mode);
 
       public:
         SplitDataRequest(LSQ &port_, MinorDynInstPtr inst_,

src/cpu/o3/cpu.cc

@@ -1095,7 +1095,7 @@ FullO3CPU<Impl>::getInterrupts()
 
 template <class Impl>
 void
-FullO3CPU<Impl>::processInterrupts(Fault interrupt)
+FullO3CPU<Impl>::processInterrupts(const Fault &interrupt)
 {
     // Check for interrupts here.  For now can copy the code that
     // exists within isa_fullsys_traits.hh.  Also assume that thread 0
@@ -1112,7 +1112,7 @@ FullO3CPU<Impl>::processInterrupts(Fault interrupt)
 
 template <class Impl>
 void
-FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst)
+FullO3CPU<Impl>::trap(const Fault &fault, ThreadID tid, StaticInstPtr inst)
 {
     // Pass the thread's TC into the invoke method.
     fault->invoke(this->threadContexts[tid], inst);

src/cpu/o3/cpu.hh

@@ -498,7 +498,7 @@ class FullO3CPU : public BaseO3CPU
     { return globalSeqNum++; }
 
     /** Traps to handle given fault. */
-    void trap(Fault fault, ThreadID tid, StaticInstPtr inst);
+    void trap(const Fault &fault, ThreadID tid, StaticInstPtr inst);
 
     /** HW return from error interrupt. */
     Fault hwrei(ThreadID tid);
@@ -509,7 +509,7 @@ class FullO3CPU : public BaseO3CPU
     Fault getInterrupts();
 
     /** Processes any an interrupt fault. */
-    void processInterrupts(Fault interrupt);
+    void processInterrupts(const Fault &interrupt);
 
     /** Halts the CPU. */
     void halt() { panic("Halt not implemented!\n"); }

src/cpu/o3/dyn_inst.hh

@@ -230,7 +230,7 @@ class BaseO3DynInst : public BaseDynInst<Impl>
     /** Calls hardware return from error interrupt. */
     Fault hwrei();
     /** Traps to handle specified fault. */
-    void trap(Fault fault);
+    void trap(const Fault &fault);
     bool simPalCheck(int palFunc);
 
     /** Emulates a syscall. */

src/cpu/o3/dyn_inst_impl.hh

@@ -225,7 +225,7 @@ BaseO3DynInst<Impl>::hwrei()
 
 template <class Impl>
 void
-BaseO3DynInst<Impl>::trap(Fault fault)
+BaseO3DynInst<Impl>::trap(const Fault &fault)
 {
     this->cpu->trap(fault, this->threadNumber, this->staticInst);
 }

src/cpu/o3/fetch.hh

@@ -100,7 +100,7 @@ class DefaultFetch
         {}
 
         void
-        finish(Fault fault, RequestPtr req, ThreadContext *tc,
+        finish(const Fault &fault, RequestPtr req, ThreadContext *tc,
                BaseTLB::Mode mode)
         {
             assert(mode == BaseTLB::Execute);
@@ -294,7 +294,7 @@ class DefaultFetch
      * @return Any fault that occured.
      */
     bool fetchCacheLine(Addr vaddr, ThreadID tid, Addr pc);
-    void finishTranslation(Fault fault, RequestPtr mem_req);
+    void finishTranslation(const Fault &fault, RequestPtr mem_req);
 
 
     /** Check if an interrupt is pending and that we need to handle

src/cpu/o3/fetch_impl.hh

@@ -633,7 +633,7 @@ DefaultFetch<Impl>::fetchCacheLine(Addr vaddr, ThreadID tid, Addr pc)
 
 template <class Impl>
 void
-DefaultFetch<Impl>::finishTranslation(Fault fault, RequestPtr mem_req)
+DefaultFetch<Impl>::finishTranslation(const Fault &fault, RequestPtr mem_req)
 {
     ThreadID tid = mem_req->threadId();
     Addr fetchBufferBlockPC = mem_req->getVaddr();

src/cpu/ozone/dyn_inst_impl.hh

@@ -255,7 +255,7 @@ OzoneDynInst<Impl>::hwrei()
 
 template <class Impl>
 void
-OzoneDynInst<Impl>::trap(Fault fault)
+OzoneDynInst<Impl>::trap(const Fault &fault)
 {
     fault->invoke(this->thread->getTC());
 }

src/cpu/simple/base.cc

@@ -555,7 +555,7 @@ BaseSimpleCPU::postExecute()
 }
 
 void
-BaseSimpleCPU::advancePC(Fault fault)
+BaseSimpleCPU::advancePC(const Fault &fault)
 {
     const bool branching(thread->pcState().branching());
 

src/cpu/simple/base.hh

@@ -168,7 +168,7 @@ class BaseSimpleCPU : public BaseCPU, public ExecContext
     void setupFetchRequest(Request *req);
     void preExecute();
     void postExecute();
-    void advancePC(Fault fault);
+    void advancePC(const Fault &fault);
 
     virtual void deallocateContext(ThreadID thread_num);
     virtual void haltContext(ThreadID thread_num);

src/cpu/simple/timing.cc

@@ -328,7 +328,7 @@ TimingSimpleCPU::sendSplitData(RequestPtr req1, RequestPtr req2,
 }
 
 void
-TimingSimpleCPU::translationFault(Fault fault)
+TimingSimpleCPU::translationFault(const Fault &fault)
 {
     // fault may be NoFault in cases where a fault is suppressed,
     // for instance prefetches.
@@ -576,7 +576,8 @@ TimingSimpleCPU::fetch()
 
 
 void
-TimingSimpleCPU::sendFetch(Fault fault, RequestPtr req, ThreadContext *tc)
+TimingSimpleCPU::sendFetch(const Fault &fault, RequestPtr req,
+                           ThreadContext *tc)
 {
     if (fault == NoFault) {
         DPRINTF(SimpleCPU, "Sending fetch for addr %#x(pa: %#x)\n",
@@ -608,7 +609,7 @@ TimingSimpleCPU::sendFetch(Fault fault, RequestPtr req, ThreadContext *tc)
 
 
 void
-TimingSimpleCPU::advanceInst(Fault fault)
+TimingSimpleCPU::advanceInst(const Fault &fault)
 {
     if (_status == Faulting)
         return;

src/cpu/simple/timing.hh

@@ -123,7 +123,7 @@ class TimingSimpleCPU : public BaseSimpleCPU
         }
 
         void
-        finish(Fault fault, RequestPtr req, ThreadContext *tc,
+        finish(const Fault &fault, RequestPtr req, ThreadContext *tc,
                BaseTLB::Mode mode)
         {
             cpu->sendFetch(fault, req, tc);
@@ -135,7 +135,7 @@ class TimingSimpleCPU : public BaseSimpleCPU
     void sendSplitData(RequestPtr req1, RequestPtr req2, RequestPtr req,
                        uint8_t *data, bool read);
 
-    void translationFault(Fault fault);
+    void translationFault(const Fault &fault);
 
     void buildPacket(PacketPtr &pkt, RequestPtr req, bool read);
     void buildSplitPacket(PacketPtr &pkt1, PacketPtr &pkt2,
@@ -280,10 +280,10 @@ class TimingSimpleCPU : public BaseSimpleCPU
                    Addr addr, unsigned flags, uint64_t *res);
 
     void fetch();
-    void sendFetch(Fault fault, RequestPtr req, ThreadContext *tc);
+    void sendFetch(const Fault &fault, RequestPtr req, ThreadContext *tc);
     void completeIfetch(PacketPtr );
     void completeDataAccess(PacketPtr pkt);
-    void advanceInst(Fault fault);
+    void advanceInst(const Fault &fault);
 
     /** This function is used by the page table walker to determine if it could
      * translate the a pending request or if the underlying request has been

src/cpu/translation.hh

@@ -111,7 +111,7 @@ class WholeTranslationState
      * request to make it easier to access them later on.
      */
     bool
-    finish(Fault fault, int index)
+    finish(const Fault &fault, int index)
     {
         assert(outstanding);
         faults[index] = fault;
@@ -249,7 +249,7 @@ class DataTranslation : public BaseTLB::Translation
      * translation is complete if the state says so.
      */
     void
-    finish(Fault fault, RequestPtr req, ThreadContext *tc,
+    finish(const Fault &fault, RequestPtr req, ThreadContext *tc,
            BaseTLB::Mode mode)
     {
         assert(state);

src/sim/tlb.hh

@@ -104,8 +104,8 @@ class BaseTLB : public SimObject
          * be responsible for cleaning itself up which will happen in this
          * function. Once it's called, the object is no longer valid.
          */
-        virtual void finish(Fault fault, RequestPtr req, ThreadContext *tc,
+        virtual void finish(const Fault &fault, RequestPtr req,
-                            Mode mode) = 0;
+                            ThreadContext *tc, Mode mode) = 0;
 
         /** This function is used by the page table walker to determine if it
          * should translate the a pending request or if the underlying request