O3CPU: Undo Gabe's changes to remove hwrei and simpalcheck from O3 CPU. Removing hwrei causes

the instruction after the hwrei to be fetched before the ITB/DTB_CM register is updated in a call pal
call sys and thus the translation fails because the user is attempting to access a super page address.

Minimally, it seems as though some sort of fetch stall or refetch after a hwrei is required. I think
this works currently because the hwrei uses the exec context interface, and the o3 stalls when that occurs.

Additionally, these changes don't update the LOCK register and probably break ll/sc. Both o3 changes were
removed since a great deal of manual patching would be required to only remove the hwrei change.

src/arch/alpha/ev5.cc

@@ -547,3 +547,53 @@ copyIprs(ThreadContext *src, ThreadContext *dest)
 }
 
 } // namespace AlphaISA
+
+#if FULL_SYSTEM
+
+using namespace AlphaISA;
+
+Fault
+SimpleThread::hwrei()
+{
+    if (!(readPC() & 0x3))
+        return new UnimplementedOpcodeFault;
+
+    setNextPC(readMiscRegNoEffect(IPR_EXC_ADDR));
+
+    if (!misspeculating()) {
+        if (kernelStats)
+            kernelStats->hwrei();
+    }
+
+    // FIXME: XXX check for interrupts? XXX
+    return NoFault;
+}
+
+/**
+ * Check for special simulator handling of specific PAL calls.
+ * If return value is false, actual PAL call will be suppressed.
+ */
+bool
+SimpleThread::simPalCheck(int palFunc)
+{
+    if (kernelStats)
+        kernelStats->callpal(palFunc, tc);
+
+    switch (palFunc) {
+      case PAL::halt:
+        halt();
+        if (--System::numSystemsRunning == 0)
+            exitSimLoop("all cpus halted");
+        break;
+
+      case PAL::bpt:
+      case PAL::bugchk:
+        if (system->breakpoint())
+            return false;
+        break;
+    }
+
+    return true;
+}
+
+#endif // FULL_SYSTEM

src/arch/alpha/isa/decoder.isa

@@ -698,28 +698,7 @@ decode OPCODE default Unknown::unknown() {
         else {
             // check to see if simulator wants to do something special
             // on this PAL call (including maybe suppress it)
-            
-            bool dopal = true;
-
-            ThreadContext * tc = xc->tcBase();
-            AlphaISA::Kernel::Statistics * kernelStats = tc->getKernelStats();
-            System * system = tc->getSystemPtr();
-            if (kernelStats)
-                kernelStats->callpal(palFunc, tc);
-
-            switch (palFunc) {
-              case PAL::halt:
-                tc->halt();
-                if (--System::numSystemsRunning == 0)
-                    exitSimLoop("all cpus halted");
-                break;
-
-              case PAL::bpt:
-              case PAL::bugchk:
-                if (system->breakpoint())
-                    dopal = false;
-                break;
-            }
+            bool dopal = xc->simPalCheck(palFunc);
 
             if (dopal) {
                 xc->setMiscReg(IPR_EXC_ADDR, NPC);
@@ -807,16 +786,7 @@ decode OPCODE default Unknown::unknown() {
     format BasicOperate {
         0x1e: decode PALMODE {
             0: OpcdecFault::hw_rei();
-            1: hw_rei({{
-                NPC = ExcAddr;
-                ThreadContext * tc = xc->tcBase();
-                if (!tc->misspeculating()) {
-                    AlphaISA::Kernel::Statistics * kernelStats =
-                        tc->getKernelStats();
-                    if (kernelStats)
-                        kernelStats->hwrei();
-                }
-            }}, IsSerializing, IsSerializeBefore);
+            1:hw_rei({{ xc->hwrei(); }}, IsSerializing, IsSerializeBefore);
         }
 
         // M5 special opcodes use the reserved 0x01 opcode space

src/arch/alpha/isa/main.isa

@@ -69,8 +69,6 @@ output exec {{
 #include <math.h>
 
 #if FULL_SYSTEM
-#include "arch/alpha/kernel_stats.hh"
-#include "arch/alpha/osfpal.hh"
 #include "sim/pseudo_inst.hh"
 #endif
 #include "arch/alpha/ipr.hh"
@@ -189,7 +187,6 @@ def operands {{
     'Runiq': ('ControlReg', 'uq', 'MISCREG_UNIQ', None, 1),
     'FPCR':  ('ControlReg', 'uq', 'MISCREG_FPCR', None, 1),
     'IntrFlag': ('ControlReg', 'uq', 'MISCREG_INTR', None, 1),
-    'ExcAddr': ('ControlReg', 'uq', 'IPR_EXC_ADDR', None, 1),
     # The next two are hacks for non-full-system call-pal emulation
     'R0':  ('IntReg', 'uq', '0', None, 1),
     'R16': ('IntReg', 'uq', '16', None, 1),

src/arch/x86/bios/IntelMP.py

@@ -86,15 +86,6 @@ class X86IntelMPConfigTable(SimObject):
     ext_entries = VectorParam.X86IntelMPExtConfigEntry([],
             'extended configuration table entries')
 
-    def add_entry(self, entry):
-        if isinstance(entry, X86IntelMPBaseConfigEntry):
-            self.base_entries.append(entry)
-        elif isinstance(entry, X86IntelMPExtConfigEntry):
-            self.base_entries.append(entry)
-        else:
-            panic("Don't know what type of Intel MP entry %s is." \
-                    % entry.__class__.__name__)
-
 class X86IntelMPBaseConfigEntry(SimObject):
     type = 'X86IntelMPBaseConfigEntry'
     cxx_class = 'X86ISA::IntelMP::BaseConfigEntry'

src/cpu/checker/cpu.hh

@@ -336,7 +336,9 @@ class CheckerCPU : public BaseCPU
     void translateDataReadReq(Request *req);
 
 #if FULL_SYSTEM
+    Fault hwrei() { return thread->hwrei(); }
     void ev5_trap(Fault fault) { fault->invoke(tc); }
+    bool simPalCheck(int palFunc) { return thread->simPalCheck(palFunc); }
 #else
     // Assume that the normal CPU's call to syscall was successful.
     // The checker's state would have already been updated by the syscall.

src/cpu/exec_context.hh

@@ -143,7 +143,17 @@ class ExecContext {
      * given flags. */
     void writeHint(Addr addr, int size, unsigned flags);
 
-#if !FULL_SYSTEM
+#if FULL_SYSTEM
+    /** Somewhat Alpha-specific function that handles returning from
+     * an error or interrupt. */
+    Fault hwrei();
+
+    /**
+     * Check for special simulator handling of specific PAL calls.  If
+     * return value is false, actual PAL call will be suppressed.
+     */
+    bool simPalCheck(int palFunc);
+#else
     /** Executes a syscall specified by the callnum. */
     void syscall(int64_t callnum);
 #endif

src/cpu/o3/cpu.cc

@@ -53,6 +53,10 @@
 #include "cpu/checker/cpu.hh"
 #endif
 
+#if THE_ISA == ALPHA_ISA
+#include "arch/alpha/osfpal.hh"
+#endif
+
 class BaseCPUParams;
 
 using namespace TheISA;
@@ -901,6 +905,47 @@ FullO3CPU<Impl>::post_interrupt(int int_num, int index)
     }
 }
 
+template <class Impl>
+Fault
+FullO3CPU<Impl>::hwrei(unsigned tid)
+{
+#if THE_ISA == ALPHA_ISA
+    // Need to clear the lock flag upon returning from an interrupt.
+    this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid);
+
+    this->thread[tid]->kernelStats->hwrei();
+
+    // FIXME: XXX check for interrupts? XXX
+#endif
+    return NoFault;
+}
+
+template <class Impl>
+bool
+FullO3CPU<Impl>::simPalCheck(int palFunc, unsigned tid)
+{
+#if THE_ISA == ALPHA_ISA
+    if (this->thread[tid]->kernelStats)
+        this->thread[tid]->kernelStats->callpal(palFunc,
+                                                this->threadContexts[tid]);
+
+    switch (palFunc) {
+      case PAL::halt:
+        halt();
+        if (--System::numSystemsRunning == 0)
+            exitSimLoop("all cpus halted");
+        break;
+
+      case PAL::bpt:
+      case PAL::bugchk:
+        if (this->system->breakpoint())
+            return false;
+        break;
+    }
+#endif
+    return true;
+}
+
 template <class Impl>
 Fault
 FullO3CPU<Impl>::getInterrupts()

src/cpu/o3/cpu.hh

@@ -414,6 +414,11 @@ class FullO3CPU : public BaseO3CPU
     /** Posts an interrupt. */
     void post_interrupt(int int_num, int index);
 
+    /** HW return from error interrupt. */
+    Fault hwrei(unsigned tid);
+
+    bool simPalCheck(int palFunc, unsigned tid);
+
     /** Returns the Fault for any valid interrupt. */
     Fault getInterrupts();
 

src/cpu/o3/dyn_inst.hh

@@ -168,8 +168,11 @@ class BaseO3DynInst : public BaseDynInst<Impl>
     }
 
 #if FULL_SYSTEM
+    /** Calls hardware return from error interrupt. */
+    Fault hwrei();
     /** Traps to handle specified fault. */
     void trap(Fault fault);
+    bool simPalCheck(int palFunc);
 #else
     /** Calls a syscall. */
     void syscall(int64_t callnum);

src/cpu/o3/dyn_inst_impl.hh

@@ -124,12 +124,44 @@ BaseO3DynInst<Impl>::completeAcc(PacketPtr pkt)
 }
 
 #if FULL_SYSTEM
+template <class Impl>
+Fault
+BaseO3DynInst<Impl>::hwrei()
+{
+#if THE_ISA == ALPHA_ISA
+    // Can only do a hwrei when in pal mode.
+    if (!(this->readPC() & 0x3))
+        return new AlphaISA::UnimplementedOpcodeFault;
+
+    // Set the next PC based on the value of the EXC_ADDR IPR.
+    this->setNextPC(this->cpu->readMiscRegNoEffect(AlphaISA::IPR_EXC_ADDR,
+                                           this->threadNumber));
+
+    // Tell CPU to clear any state it needs to if a hwrei is taken.
+    this->cpu->hwrei(this->threadNumber);
+#else
+
+#endif
+    // FIXME: XXX check for interrupts? XXX
+    return NoFault;
+}
+
 template <class Impl>
 void
 BaseO3DynInst<Impl>::trap(Fault fault)
 {
     this->cpu->trap(fault, this->threadNumber);
 }
+
+template <class Impl>
+bool
+BaseO3DynInst<Impl>::simPalCheck(int palFunc)
+{
+#if THE_ISA != ALPHA_ISA
+    panic("simPalCheck called, but PAL only exists in Alpha!\n");
+#endif
+    return this->cpu->simPalCheck(palFunc, this->threadNumber);
+}
 #else
 template <class Impl>
 void

src/cpu/ozone/cpu.hh

@@ -507,6 +507,8 @@ class OzoneCPU : public BaseCPU
     void dumpInsts() { frontEnd->dumpInsts(); }
 
 #if FULL_SYSTEM
+    Fault hwrei();
+    bool simPalCheck(int palFunc);
     void processInterrupts();
 #else
     void syscall(uint64_t &callnum);

src/cpu/ozone/cpu_impl.hh

@@ -668,6 +668,21 @@ OzoneCPU<Impl>::setSyscallReturn(SyscallReturn return_value, int tid)
     }
 }
 #else
+template <class Impl>
+Fault
+OzoneCPU<Impl>::hwrei()
+{
+    // Need to move this to ISA code
+    // May also need to make this per thread
+
+    lockFlag = false;
+    lockAddrList.clear();
+    thread.kernelStats->hwrei();
+
+    // FIXME: XXX check for interrupts? XXX
+    return NoFault;
+}
+
 template <class Impl>
 void
 OzoneCPU<Impl>::processInterrupts()
@@ -685,6 +700,31 @@ OzoneCPU<Impl>::processInterrupts()
         interrupt->invoke(thread.getTC());
     }
 }
+
+template <class Impl>
+bool
+OzoneCPU<Impl>::simPalCheck(int palFunc)
+{
+    // Need to move this to ISA code
+    // May also need to make this per thread
+    thread.kernelStats->callpal(palFunc, tc);
+
+    switch (palFunc) {
+      case PAL::halt:
+        haltContext(thread.readTid());
+        if (--System::numSystemsRunning == 0)
+            exitSimLoop("all cpus halted");
+        break;
+
+      case PAL::bpt:
+      case PAL::bugchk:
+        if (system->breakpoint())
+            return false;
+        break;
+    }
+
+    return true;
+}
 #endif
 
 template <class Impl>

src/cpu/ozone/dyn_inst.hh

@@ -240,7 +240,9 @@ class OzoneDynInst : public BaseDynInst<Impl>
     void setMiscReg(int misc_reg, const MiscReg &val);
 
 #if FULL_SYSTEM
+    Fault hwrei();
     void trap(Fault fault);
+    bool simPalCheck(int palFunc);
 #else
     void syscall(uint64_t &callnum);
 #endif

src/cpu/ozone/dyn_inst_impl.hh

@@ -248,12 +248,34 @@ OzoneDynInst<Impl>::setMiscReg(int misc_reg, const MiscReg &val)
 
 #if FULL_SYSTEM
 
+template <class Impl>
+Fault
+OzoneDynInst<Impl>::hwrei()
+{
+    if (!(this->readPC() & 0x3))
+        return new AlphaISA::UnimplementedOpcodeFault;
+
+    this->setNextPC(this->thread->readMiscRegNoEffect(AlphaISA::IPR_EXC_ADDR));
+
+    this->cpu->hwrei();
+
+    // FIXME: XXX check for interrupts? XXX
+    return NoFault;
+}
+
 template <class Impl>
 void
 OzoneDynInst<Impl>::trap(Fault fault)
 {
     fault->invoke(this->thread->getTC());
 }
+
+template <class Impl>
+bool
+OzoneDynInst<Impl>::simPalCheck(int palFunc)
+{
+    return this->cpu->simPalCheck(palFunc);
+}
 #else
 template <class Impl>
 void

src/cpu/simple/base.hh

@@ -413,7 +413,9 @@ class BaseSimpleCPU : public BaseCPU
     //Fault CacheOp(uint8_t Op, Addr EA);
 
 #if FULL_SYSTEM
+    Fault hwrei() { return thread->hwrei(); }
     void ev5_trap(Fault fault) { fault->invoke(tc); }
+    bool simPalCheck(int palFunc) { return thread->simPalCheck(palFunc); }
 #else
     void syscall(int64_t callnum) { thread->syscall(callnum); }
 #endif

src/cpu/simple_thread.hh

@@ -185,6 +185,10 @@ class SimpleThread : public ThreadState
 
     void dumpFuncProfile();
 
+    Fault hwrei();
+
+    bool simPalCheck(int palFunc);
+
 #endif
 
     /*******************************************