gem5/cpu/simple/cpu.cc
Kevin Lim f15e492375 Steps towards setting up the infrastructure to allow the new CPU model to work in full system mode.
The major change is renaming the old ExecContext to CPUExecContext, and creating two new classes, ExecContext (an abstract class), and ProxyExecContext (a templated class that derives from ExecContext).

Code outside of the CPU continues to use ExecContext as normal (other than not being able to access variables within the XC).  The CPU uses the CPUExecContext, or however else it stores its own state.  It then creates a ProxyExecContext, templated on the class used to hold its state.  This proxy is passed to any code outside of the CPU that needs to access the XC.  This allows code outside of the CPU to use the ExecContext interface to access any state needed, without knowledge of how that state is laid out.

Note that these changes will not compile without the accompanying revision to automatically rename the shadow registers.

SConscript:
    Include new file, cpu_exec_context.cc.
arch/alpha/alpha_linux_process.cc:
arch/alpha/alpha_memory.cc:
arch/alpha/alpha_tru64_process.cc:
arch/alpha/arguments.cc:
arch/alpha/isa/decoder.isa:
arch/alpha/stacktrace.cc:
arch/alpha/vtophys.cc:
base/remote_gdb.cc:
cpu/intr_control.cc:
    Avoid directly accessing objects within the XC.
arch/alpha/ev5.cc:
    Avoid directly accessing objects within the XC.

    KernelStats have been moved to the BaseCPU instead of the XC.
arch/alpha/isa_traits.hh:
    Remove clearIprs().  It wasn't used very often and it did not work well with the proxy ExecContext.
cpu/base.cc:
    Place kernel stats within the BaseCPU instead of the ExecContext.

    For now comment out the profiling code sampling until its exact location is decided upon.
cpu/base.hh:
    Kernel stats are now in the BaseCPU instead of the ExecContext.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
cpu/memtest/memtest.cc:
cpu/memtest/memtest.hh:
    Changes to support rename of old ExecContext to CPUExecContext.  See changeset for more details.
cpu/exetrace.cc:
    Remove unneeded include of exec_context.hh.
cpu/intr_control.hh:
cpu/o3/alpha_cpu_builder.cc:
    Remove unneeded include of exec_context.hh
cpu/o3/alpha_cpu.hh:
cpu/o3/alpha_cpu_impl.hh:
cpu/o3/cpu.cc:
cpu/o3/cpu.hh:
cpu/simple/cpu.cc:
cpu/simple/cpu.hh:
    Changes to support rename of old ExecContext to CPUExecContext.  See changeset for more details.

    Also avoid accessing anything directly from the XC.
cpu/pc_event.cc:
    Avoid accessing objects directly from the XC.
dev/tsunami_cchip.cc:
    Avoid accessing objects directly within the XC>
kern/freebsd/freebsd_system.cc:
kern/linux/linux_system.cc:
kern/linux/linux_threadinfo.hh:
kern/tru64/dump_mbuf.cc:
kern/tru64/tru64.hh:
kern/tru64/tru64_events.cc:
sim/syscall_emul.cc:
sim/syscall_emul.hh:
    Avoid accessing objects directly within the XC.
kern/kernel_stats.cc:
kern/kernel_stats.hh:
    Kernel stats no longer exist within the XC.
kern/system_events.cc:
    Avoid accessing objects directly within the XC.  Also kernel stats are now in the BaseCPU.
sim/process.cc:
sim/process.hh:
    Avoid accessing regs directly within an ExecContext.  Instead use a CPUExecContext to initialize the registers and copy them over.
cpu/cpu_exec_context.cc:
    Rename old ExecContext to CPUExecContext.  This is used by the old CPU models to store any necessary architectural state.  Also include the ProxyExecContext, which is used to access the CPUExecContext's state in code outside of the CPU.
cpu/cpu_exec_context.hh:
    Rename old ExecContext to CPUExecContext.  This is used by the old CPU models to store any necessary architectural state.  Also include the ProxyExecContext, which is used to access the CPUExecContext's state in code outside of the CPU.

    Remove kernel stats from the ExecContext.
sim/pseudo_inst.cc:
    Kernel stats now live within the CPU.

    Avoid accessing objects directly within the XC.

--HG--
rename : cpu/exec_context.cc => cpu/cpu_exec_context.cc
rename : cpu/exec_context.hh => cpu/cpu_exec_context.hh
extra : convert_revision : a75393a8945c80cca225b5e9d9c22a16609efb85
2006-03-04 15:18:40 -05:00

944 lines
25 KiB
C++

/*
* Copyright (c) 2002-2005 The Regents of The University of Michigan
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are
* met: redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer;
* redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution;
* neither the name of the copyright holders nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include <cmath>
#include <cstdio>
#include <cstdlib>
#include <iostream>
#include <iomanip>
#include <list>
#include <sstream>
#include <string>
#include "base/cprintf.hh"
#include "base/inifile.hh"
#include "base/loader/symtab.hh"
#include "base/misc.hh"
#include "base/pollevent.hh"
#include "base/range.hh"
#include "base/stats/events.hh"
#include "base/trace.hh"
#include "cpu/base.hh"
#include "cpu/cpu_exec_context.hh"
#include "cpu/exec_context.hh"
#include "cpu/exetrace.hh"
#include "cpu/profile.hh"
#include "cpu/sampler/sampler.hh"
#include "cpu/simple/cpu.hh"
#include "cpu/smt.hh"
#include "cpu/static_inst.hh"
#include "kern/kernel_stats.hh"
#include "mem/base_mem.hh"
#include "mem/mem_interface.hh"
#include "sim/byteswap.hh"
#include "sim/builder.hh"
#include "sim/debug.hh"
#include "sim/host.hh"
#include "sim/sim_events.hh"
#include "sim/sim_object.hh"
#include "sim/stats.hh"
#if FULL_SYSTEM
#include "base/remote_gdb.hh"
#include "mem/functional/memory_control.hh"
#include "mem/functional/physical.hh"
#include "sim/system.hh"
#include "targetarch/alpha_memory.hh"
#include "targetarch/stacktrace.hh"
#include "targetarch/vtophys.hh"
#else // !FULL_SYSTEM
#include "mem/functional/functional.hh"
#endif // FULL_SYSTEM
using namespace std;
//The SimpleCPU does alpha only
using namespace AlphaISA;
SimpleCPU::TickEvent::TickEvent(SimpleCPU *c, int w)
: Event(&mainEventQueue, CPU_Tick_Pri), cpu(c), width(w)
{
}
void
SimpleCPU::init()
{
BaseCPU::init();
#if FULL_SYSTEM
for (int i = 0; i < execContexts.size(); ++i) {
ExecContext *xc = execContexts[i];
// initialize CPU, including PC
TheISA::initCPU(xc, xc->readCpuId());
}
#endif
}
void
SimpleCPU::TickEvent::process()
{
int count = width;
do {
cpu->tick();
} while (--count > 0 && cpu->status() == Running);
}
const char *
SimpleCPU::TickEvent::description()
{
return "SimpleCPU tick event";
}
SimpleCPU::CacheCompletionEvent::CacheCompletionEvent(SimpleCPU *_cpu)
: Event(&mainEventQueue), cpu(_cpu)
{
}
void SimpleCPU::CacheCompletionEvent::process()
{
cpu->processCacheCompletion();
}
const char *
SimpleCPU::CacheCompletionEvent::description()
{
return "SimpleCPU cache completion event";
}
SimpleCPU::SimpleCPU(Params *p)
: BaseCPU(p), tickEvent(this, p->width), cpuXC(NULL),
cacheCompletionEvent(this)
{
_status = Idle;
#if FULL_SYSTEM
cpuXC = new CPUExecContext(this, 0, p->system, p->itb, p->dtb, p->mem);
#else
cpuXC = new CPUExecContext(this, /* thread_num */ 0, p->process,
/* asid */ 0);
#endif // !FULL_SYSTEM
xcProxy = cpuXC->getProxy();
icacheInterface = p->icache_interface;
dcacheInterface = p->dcache_interface;
memReq = new MemReq();
memReq->xc = xcProxy;
memReq->asid = 0;
memReq->data = new uint8_t[64];
numInst = 0;
startNumInst = 0;
numLoad = 0;
startNumLoad = 0;
lastIcacheStall = 0;
lastDcacheStall = 0;
execContexts.push_back(xcProxy);
}
SimpleCPU::~SimpleCPU()
{
}
void
SimpleCPU::switchOut(Sampler *s)
{
sampler = s;
if (status() == DcacheMissStall) {
DPRINTF(Sampler,"Outstanding dcache access, waiting for completion\n");
_status = DcacheMissSwitch;
}
else {
_status = SwitchedOut;
if (tickEvent.scheduled())
tickEvent.squash();
sampler->signalSwitched();
}
}
void
SimpleCPU::takeOverFrom(BaseCPU *oldCPU)
{
BaseCPU::takeOverFrom(oldCPU);
assert(!tickEvent.scheduled());
// if any of this CPU's ExecContexts are active, mark the CPU as
// running and schedule its tick event.
for (int i = 0; i < execContexts.size(); ++i) {
ExecContext *xc = execContexts[i];
if (xc->status() == ExecContext::Active && _status != Running) {
_status = Running;
tickEvent.schedule(curTick);
}
}
}
void
SimpleCPU::activateContext(int thread_num, int delay)
{
assert(thread_num == 0);
assert(cpuXC);
assert(_status == Idle);
notIdleFraction++;
scheduleTickEvent(delay);
_status = Running;
}
void
SimpleCPU::suspendContext(int thread_num)
{
assert(thread_num == 0);
assert(cpuXC);
assert(_status == Running);
notIdleFraction--;
unscheduleTickEvent();
_status = Idle;
}
void
SimpleCPU::deallocateContext(int thread_num)
{
// for now, these are equivalent
suspendContext(thread_num);
}
void
SimpleCPU::haltContext(int thread_num)
{
// for now, these are equivalent
suspendContext(thread_num);
}
void
SimpleCPU::regStats()
{
using namespace Stats;
BaseCPU::regStats();
numInsts
.name(name() + ".num_insts")
.desc("Number of instructions executed")
;
numMemRefs
.name(name() + ".num_refs")
.desc("Number of memory references")
;
notIdleFraction
.name(name() + ".not_idle_fraction")
.desc("Percentage of non-idle cycles")
;
idleFraction
.name(name() + ".idle_fraction")
.desc("Percentage of idle cycles")
;
icacheStallCycles
.name(name() + ".icache_stall_cycles")
.desc("ICache total stall cycles")
.prereq(icacheStallCycles)
;
dcacheStallCycles
.name(name() + ".dcache_stall_cycles")
.desc("DCache total stall cycles")
.prereq(dcacheStallCycles)
;
idleFraction = constant(1.0) - notIdleFraction;
}
void
SimpleCPU::resetStats()
{
startNumInst = numInst;
notIdleFraction = (_status != Idle);
}
void
SimpleCPU::serialize(ostream &os)
{
BaseCPU::serialize(os);
SERIALIZE_ENUM(_status);
SERIALIZE_SCALAR(inst);
nameOut(os, csprintf("%s.xc", name()));
cpuXC->serialize(os);
nameOut(os, csprintf("%s.tickEvent", name()));
tickEvent.serialize(os);
nameOut(os, csprintf("%s.cacheCompletionEvent", name()));
cacheCompletionEvent.serialize(os);
}
void
SimpleCPU::unserialize(Checkpoint *cp, const string &section)
{
BaseCPU::unserialize(cp, section);
UNSERIALIZE_ENUM(_status);
UNSERIALIZE_SCALAR(inst);
cpuXC->unserialize(cp, csprintf("%s.xc", section));
tickEvent.unserialize(cp, csprintf("%s.tickEvent", section));
cacheCompletionEvent
.unserialize(cp, csprintf("%s.cacheCompletionEvent", section));
}
void
change_thread_state(int thread_number, int activate, int priority)
{
}
Fault
SimpleCPU::copySrcTranslate(Addr src)
{
static bool no_warn = true;
int blk_size = (dcacheInterface) ? dcacheInterface->getBlockSize() : 64;
// Only support block sizes of 64 atm.
assert(blk_size == 64);
int offset = src & (blk_size - 1);
// Make sure block doesn't span page
if (no_warn &&
(src & PageMask) != ((src + blk_size) & PageMask) &&
(src >> 40) != 0xfffffc) {
warn("Copied block source spans pages %x.", src);
no_warn = false;
}
memReq->reset(src & ~(blk_size - 1), blk_size);
// translate to physical address
Fault fault = cpuXC->translateDataReadReq(memReq);
assert(fault != AlignmentFault);
if (fault == NoFault) {
cpuXC->copySrcAddr = src;
cpuXC->copySrcPhysAddr = memReq->paddr + offset;
} else {
cpuXC->copySrcAddr = 0;
cpuXC->copySrcPhysAddr = 0;
}
return fault;
}
Fault
SimpleCPU::copy(Addr dest)
{
static bool no_warn = true;
int blk_size = (dcacheInterface) ? dcacheInterface->getBlockSize() : 64;
// Only support block sizes of 64 atm.
assert(blk_size == 64);
uint8_t data[blk_size];
//assert(cpuXC->copySrcAddr);
int offset = dest & (blk_size - 1);
// Make sure block doesn't span page
if (no_warn &&
(dest & PageMask) != ((dest + blk_size) & PageMask) &&
(dest >> 40) != 0xfffffc) {
no_warn = false;
warn("Copied block destination spans pages %x. ", dest);
}
memReq->reset(dest & ~(blk_size -1), blk_size);
// translate to physical address
Fault fault = cpuXC->translateDataWriteReq(memReq);
assert(fault != AlignmentFault);
if (fault == NoFault) {
Addr dest_addr = memReq->paddr + offset;
// Need to read straight from memory since we have more than 8 bytes.
memReq->paddr = cpuXC->copySrcPhysAddr;
cpuXC->mem->read(memReq, data);
memReq->paddr = dest_addr;
cpuXC->mem->write(memReq, data);
if (dcacheInterface) {
memReq->cmd = Copy;
memReq->completionEvent = NULL;
memReq->paddr = cpuXC->copySrcPhysAddr;
memReq->dest = dest_addr;
memReq->size = 64;
memReq->time = curTick;
memReq->flags &= ~INST_READ;
dcacheInterface->access(memReq);
}
}
return fault;
}
// precise architected memory state accessor macros
template <class T>
Fault
SimpleCPU::read(Addr addr, T &data, unsigned flags)
{
if (status() == DcacheMissStall || status() == DcacheMissSwitch) {
Fault fault = cpuXC->read(memReq,data);
if (traceData) {
traceData->setAddr(addr);
}
return fault;
}
memReq->reset(addr, sizeof(T), flags);
// translate to physical address
Fault fault = cpuXC->translateDataReadReq(memReq);
// if we have a cache, do cache access too
if (fault == NoFault && dcacheInterface) {
memReq->cmd = Read;
memReq->completionEvent = NULL;
memReq->time = curTick;
memReq->flags &= ~INST_READ;
MemAccessResult result = dcacheInterface->access(memReq);
// Ugly hack to get an event scheduled *only* if the access is
// a miss. We really should add first-class support for this
// at some point.
if (result != MA_HIT && dcacheInterface->doEvents()) {
memReq->completionEvent = &cacheCompletionEvent;
lastDcacheStall = curTick;
unscheduleTickEvent();
_status = DcacheMissStall;
} else {
// do functional access
fault = cpuXC->read(memReq, data);
}
} else if(fault == NoFault) {
// do functional access
fault = cpuXC->read(memReq, data);
}
if (!dcacheInterface && (memReq->flags & UNCACHEABLE))
recordEvent("Uncached Read");
return fault;
}
#ifndef DOXYGEN_SHOULD_SKIP_THIS
template
Fault
SimpleCPU::read(Addr addr, uint64_t &data, unsigned flags);
template
Fault
SimpleCPU::read(Addr addr, uint32_t &data, unsigned flags);
template
Fault
SimpleCPU::read(Addr addr, uint16_t &data, unsigned flags);
template
Fault
SimpleCPU::read(Addr addr, uint8_t &data, unsigned flags);
#endif //DOXYGEN_SHOULD_SKIP_THIS
template<>
Fault
SimpleCPU::read(Addr addr, double &data, unsigned flags)
{
return read(addr, *(uint64_t*)&data, flags);
}
template<>
Fault
SimpleCPU::read(Addr addr, float &data, unsigned flags)
{
return read(addr, *(uint32_t*)&data, flags);
}
template<>
Fault
SimpleCPU::read(Addr addr, int32_t &data, unsigned flags)
{
return read(addr, (uint32_t&)data, flags);
}
template <class T>
Fault
SimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res)
{
memReq->reset(addr, sizeof(T), flags);
// translate to physical address
Fault fault = cpuXC->translateDataWriteReq(memReq);
// do functional access
if (fault == NoFault)
fault = cpuXC->write(memReq, data);
if (fault == NoFault && dcacheInterface) {
memReq->cmd = Write;
memcpy(memReq->data,(uint8_t *)&data,memReq->size);
memReq->completionEvent = NULL;
memReq->time = curTick;
memReq->flags &= ~INST_READ;
MemAccessResult result = dcacheInterface->access(memReq);
// Ugly hack to get an event scheduled *only* if the access is
// a miss. We really should add first-class support for this
// at some point.
if (result != MA_HIT && dcacheInterface->doEvents()) {
memReq->completionEvent = &cacheCompletionEvent;
lastDcacheStall = curTick;
unscheduleTickEvent();
_status = DcacheMissStall;
}
}
if (res && (fault == NoFault))
*res = memReq->result;
if (!dcacheInterface && (memReq->flags & UNCACHEABLE))
recordEvent("Uncached Write");
return fault;
}
#ifndef DOXYGEN_SHOULD_SKIP_THIS
template
Fault
SimpleCPU::write(uint64_t data, Addr addr, unsigned flags, uint64_t *res);
template
Fault
SimpleCPU::write(uint32_t data, Addr addr, unsigned flags, uint64_t *res);
template
Fault
SimpleCPU::write(uint16_t data, Addr addr, unsigned flags, uint64_t *res);
template
Fault
SimpleCPU::write(uint8_t data, Addr addr, unsigned flags, uint64_t *res);
#endif //DOXYGEN_SHOULD_SKIP_THIS
template<>
Fault
SimpleCPU::write(double data, Addr addr, unsigned flags, uint64_t *res)
{
return write(*(uint64_t*)&data, addr, flags, res);
}
template<>
Fault
SimpleCPU::write(float data, Addr addr, unsigned flags, uint64_t *res)
{
return write(*(uint32_t*)&data, addr, flags, res);
}
template<>
Fault
SimpleCPU::write(int32_t data, Addr addr, unsigned flags, uint64_t *res)
{
return write((uint32_t)data, addr, flags, res);
}
#if FULL_SYSTEM
Addr
SimpleCPU::dbg_vtophys(Addr addr)
{
return vtophys(xcProxy, addr);
}
#endif // FULL_SYSTEM
void
SimpleCPU::processCacheCompletion()
{
switch (status()) {
case IcacheMissStall:
icacheStallCycles += curTick - lastIcacheStall;
_status = IcacheMissComplete;
scheduleTickEvent(1);
break;
case DcacheMissStall:
if (memReq->cmd.isRead()) {
curStaticInst->execute(this,traceData);
if (traceData)
traceData->finalize();
}
dcacheStallCycles += curTick - lastDcacheStall;
_status = Running;
scheduleTickEvent(1);
break;
case DcacheMissSwitch:
if (memReq->cmd.isRead()) {
curStaticInst->execute(this,traceData);
if (traceData)
traceData->finalize();
}
_status = SwitchedOut;
sampler->signalSwitched();
case SwitchedOut:
// If this CPU has been switched out due to sampling/warm-up,
// ignore any further status changes (e.g., due to cache
// misses outstanding at the time of the switch).
return;
default:
panic("SimpleCPU::processCacheCompletion: bad state");
break;
}
}
#if FULL_SYSTEM
void
SimpleCPU::post_interrupt(int int_num, int index)
{
BaseCPU::post_interrupt(int_num, index);
if (cpuXC->status() == ExecContext::Suspended) {
DPRINTF(IPI,"Suspended Processor awoke\n");
cpuXC->activate();
}
}
#endif // FULL_SYSTEM
/* start simulation, program loaded, processor precise state initialized */
void
SimpleCPU::tick()
{
numCycles++;
traceData = NULL;
Fault fault = NoFault;
#if FULL_SYSTEM
if (checkInterrupts && check_interrupts() && !cpuXC->inPalMode() &&
status() != IcacheMissComplete) {
int ipl = 0;
int summary = 0;
checkInterrupts = false;
if (cpuXC->readMiscReg(IPR_SIRR)) {
for (int i = INTLEVEL_SOFTWARE_MIN;
i < INTLEVEL_SOFTWARE_MAX; i++) {
if (cpuXC->readMiscReg(IPR_SIRR) & (ULL(1) << i)) {
// See table 4-19 of 21164 hardware reference
ipl = (i - INTLEVEL_SOFTWARE_MIN) + 1;
summary |= (ULL(1) << i);
}
}
}
uint64_t interrupts = cpuXC->cpu->intr_status();
for (int i = INTLEVEL_EXTERNAL_MIN;
i < INTLEVEL_EXTERNAL_MAX; i++) {
if (interrupts & (ULL(1) << i)) {
// See table 4-19 of 21164 hardware reference
ipl = i;
summary |= (ULL(1) << i);
}
}
if (cpuXC->readMiscReg(IPR_ASTRR))
panic("asynchronous traps not implemented\n");
if (ipl && ipl > cpuXC->readMiscReg(IPR_IPLR)) {
cpuXC->setMiscReg(IPR_ISR, summary);
cpuXC->setMiscReg(IPR_INTID, ipl);
cpuXC->ev5_trap(InterruptFault);
DPRINTF(Flow, "Interrupt! IPLR=%d ipl=%d summary=%x\n",
cpuXC->readMiscReg(IPR_IPLR), ipl, summary);
}
}
#endif
// maintain $r0 semantics
cpuXC->setIntReg(ZeroReg, 0);
#ifdef TARGET_ALPHA
cpuXC->setFloatRegDouble(ZeroReg, 0.0);
#endif // TARGET_ALPHA
if (status() == IcacheMissComplete) {
// We've already fetched an instruction and were stalled on an
// I-cache miss. No need to fetch it again.
// Set status to running; tick event will get rescheduled if
// necessary at end of tick() function.
_status = Running;
}
else {
// Try to fetch an instruction
// set up memory request for instruction fetch
#if FULL_SYSTEM
#define IFETCH_FLAGS(pc) ((pc) & 1) ? PHYSICAL : 0
#else
#define IFETCH_FLAGS(pc) 0
#endif
memReq->cmd = Read;
memReq->reset(cpuXC->readPC() & ~3, sizeof(uint32_t),
IFETCH_FLAGS(cpuXC->readPC()));
fault = cpuXC->translateInstReq(memReq);
if (fault == NoFault)
fault = cpuXC->mem->read(memReq, inst);
if (icacheInterface && fault == NoFault) {
memReq->completionEvent = NULL;
memReq->time = curTick;
memReq->flags |= INST_READ;
MemAccessResult result = icacheInterface->access(memReq);
// Ugly hack to get an event scheduled *only* if the access is
// a miss. We really should add first-class support for this
// at some point.
if (result != MA_HIT && icacheInterface->doEvents()) {
memReq->completionEvent = &cacheCompletionEvent;
lastIcacheStall = curTick;
unscheduleTickEvent();
_status = IcacheMissStall;
return;
}
}
}
// If we've got a valid instruction (i.e., no fault on instruction
// fetch), then execute it.
if (fault == NoFault) {
// keep an instruction count
numInst++;
numInsts++;
// check for instruction-count-based events
comInstEventQueue[0]->serviceEvents(numInst);
// decode the instruction
inst = gtoh(inst);
curStaticInst = StaticInst::decode(inst);
traceData = Trace::getInstRecord(curTick, xcProxy, this, curStaticInst,
cpuXC->readPC());
#if FULL_SYSTEM
cpuXC->setInst(inst);
#endif // FULL_SYSTEM
cpuXC->func_exe_inst++;
fault = curStaticInst->execute(this, traceData);
#if FULL_SYSTEM
if (system->kernelBinning->fnbin) {
assert(kernelStats);
system->kernelBinning->execute(xcProxy, inst);
}
if (cpuXC->profile) {
bool usermode =
(cpuXC->readMiscReg(AlphaISA::IPR_DTB_CM) & 0x18) != 0;
cpuXC->profilePC = usermode ? 1 : cpuXC->readPC();
ProfileNode *node = cpuXC->profile->consume(xcProxy, inst);
if (node)
cpuXC->profileNode = node;
}
#endif
if (curStaticInst->isMemRef()) {
numMemRefs++;
}
if (curStaticInst->isLoad()) {
++numLoad;
comLoadEventQueue[0]->serviceEvents(numLoad);
}
// If we have a dcache miss, then we can't finialize the instruction
// trace yet because we want to populate it with the data later
if (traceData &&
!(status() == DcacheMissStall && memReq->cmd.isRead())) {
traceData->finalize();
}
traceFunctions(cpuXC->readPC());
} // if (fault == NoFault)
if (fault != NoFault) {
#if FULL_SYSTEM
cpuXC->ev5_trap(fault);
#else // !FULL_SYSTEM
fatal("fault (%d) detected @ PC 0x%08p", fault, cpuXC->readPC());
#endif // FULL_SYSTEM
}
else {
// go to the next instruction
cpuXC->setPC(cpuXC->readNextPC());
cpuXC->setNextPC(cpuXC->readNextPC() + sizeof(MachInst));
}
#if FULL_SYSTEM
Addr oldpc;
do {
oldpc = cpuXC->readPC();
system->pcEventQueue.service(xcProxy);
} while (oldpc != cpuXC->readPC());
#endif
assert(status() == Running ||
status() == Idle ||
status() == DcacheMissStall);
if (status() == Running && !tickEvent.scheduled())
tickEvent.schedule(curTick + cycles(1));
}
////////////////////////////////////////////////////////////////////////
//
// SimpleCPU Simulation Object
//
BEGIN_DECLARE_SIM_OBJECT_PARAMS(SimpleCPU)
Param<Counter> max_insts_any_thread;
Param<Counter> max_insts_all_threads;
Param<Counter> max_loads_any_thread;
Param<Counter> max_loads_all_threads;
#if FULL_SYSTEM
SimObjectParam<AlphaITB *> itb;
SimObjectParam<AlphaDTB *> dtb;
SimObjectParam<FunctionalMemory *> mem;
SimObjectParam<System *> system;
Param<int> cpu_id;
Param<Tick> profile;
#else
SimObjectParam<Process *> workload;
#endif // FULL_SYSTEM
Param<int> clock;
SimObjectParam<BaseMem *> icache;
SimObjectParam<BaseMem *> dcache;
Param<bool> defer_registration;
Param<int> width;
Param<bool> function_trace;
Param<Tick> function_trace_start;
END_DECLARE_SIM_OBJECT_PARAMS(SimpleCPU)
BEGIN_INIT_SIM_OBJECT_PARAMS(SimpleCPU)
INIT_PARAM(max_insts_any_thread,
"terminate when any thread reaches this inst count"),
INIT_PARAM(max_insts_all_threads,
"terminate when all threads have reached this inst count"),
INIT_PARAM(max_loads_any_thread,
"terminate when any thread reaches this load count"),
INIT_PARAM(max_loads_all_threads,
"terminate when all threads have reached this load count"),
#if FULL_SYSTEM
INIT_PARAM(itb, "Instruction TLB"),
INIT_PARAM(dtb, "Data TLB"),
INIT_PARAM(mem, "memory"),
INIT_PARAM(system, "system object"),
INIT_PARAM(cpu_id, "processor ID"),
INIT_PARAM(profile, ""),
#else
INIT_PARAM(workload, "processes to run"),
#endif // FULL_SYSTEM
INIT_PARAM(clock, "clock speed"),
INIT_PARAM(icache, "L1 instruction cache object"),
INIT_PARAM(dcache, "L1 data cache object"),
INIT_PARAM(defer_registration, "defer system registration (for sampling)"),
INIT_PARAM(width, "cpu width"),
INIT_PARAM(function_trace, "Enable function trace"),
INIT_PARAM(function_trace_start, "Cycle to start function trace")
END_INIT_SIM_OBJECT_PARAMS(SimpleCPU)
CREATE_SIM_OBJECT(SimpleCPU)
{
SimpleCPU::Params *params = new SimpleCPU::Params();
params->name = getInstanceName();
params->numberOfThreads = 1;
params->max_insts_any_thread = max_insts_any_thread;
params->max_insts_all_threads = max_insts_all_threads;
params->max_loads_any_thread = max_loads_any_thread;
params->max_loads_all_threads = max_loads_all_threads;
params->deferRegistration = defer_registration;
params->clock = clock;
params->functionTrace = function_trace;
params->functionTraceStart = function_trace_start;
params->icache_interface = (icache) ? icache->getInterface() : NULL;
params->dcache_interface = (dcache) ? dcache->getInterface() : NULL;
params->width = width;
#if FULL_SYSTEM
params->itb = itb;
params->dtb = dtb;
params->mem = mem;
params->system = system;
params->cpu_id = cpu_id;
params->profile = profile;
#else
params->process = workload;
#endif
SimpleCPU *cpu = new SimpleCPU(params);
return cpu;
}
REGISTER_SIM_OBJECT("SimpleCPU", SimpleCPU)