Initial load of files

This commit is contained in:
rsnikhil
2019-03-26 14:49:40 -04:00
parent bc62f17032
commit ee24a93944
1008 changed files with 354221 additions and 224 deletions

977
src_Core/CPU/Core.bsv Normal file
View File

@@ -0,0 +1,977 @@
// Copyright (c) 2017 Massachusetts Institute of Technology
//
// Permission is hereby granted, free of charge, to any person
// obtaining a copy of this software and associated documentation
// files (the "Software"), to deal in the Software without
// restriction, including without limitation the rights to use, copy,
// modify, merge, publish, distribute, sublicense, and/or sell copies
// of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be
// included in all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
// MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
// BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
// ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
// CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
// SOFTWARE.
// Portions Copyright (c) Bluespec, Inc.
`include "ProcConfig.bsv"
import Vector::*;
import BuildVector::*;
import DefaultValue::*;
import ClientServer::*;
import GetPut::*;
import Assert::*;
import Cntrs::*;
import ConfigReg::*;
import FIFO::*;
import Fifo::*;
import Ehr::*;
import Connectable::*;
import Types::*;
import ProcTypes::*;
import CacheUtils::*;
import TlbTypes::*;
import SynthParam::*;
import VerificationPacket::*;
import Performance::*;
import HasSpecBits::*;
import Exec::*;
import FetchStage::*;
import ITlb::*;
import DTlb::*;
import L2Tlb::*;
import TlbConnect::*;
import EpochManager::*;
import PhysRFile::*;
import RFileSynth::*;
import RenamingTable::*;
import ReorderBuffer::*;
import ReorderBufferSynth::*;
import Scoreboard::*;
import ScoreboardSynth::*;
import SpecTagManager::*;
import Fpu::*;
import MulDiv::*;
import ReservationStationEhr::*;
import ReservationStationAlu::*;
import ReservationStationMem::*;
import ReservationStationFpuMulDiv::*;
import AluExePipeline::*;
import FpuMulDivExePipeline::*;
import MemExePipeline::*;
import SplitLSQ::*;
import StoreBuffer::*;
import GlobalSpecUpdate::*;
import CCTypes::*;
import L1CoCache::*;
import L1Bank::*;
import IBank::*;
import MMIOCore::*;
import RenameStage::*;
import CommitStage::*;
import Bypass::*;
import CsrFile :: *;
interface CoreReq;
method Action start(
Addr startpc,
Addr toHostAddr, Addr fromHostAddr
);
method Action perfReq(PerfLocation loc, PerfType t);
endinterface
interface CoreIndInv;
method ActionValue#(ProcPerfResp) perfResp;
method ActionValue#(void) terminate;
endinterface
interface CoreDeadlock;
interface Get#(L1DCRqStuck) dCacheCRqStuck;
interface Get#(L1DPRqStuck) dCachePRqStuck;
interface Get#(L1ICRqStuck) iCacheCRqStuck;
interface Get#(L1IPRqStuck) iCachePRqStuck;
interface Get#(RenameStuck) renameInstStuck;
interface Get#(RenameStuck) renameCorrectPathStuck;
interface Get#(CommitStuck) commitInstStuck;
interface Get#(CommitStuck) commitUserInstStuck;
interface Get#(void) checkStarted;
endinterface
interface CoreRenameDebug;
interface Get#(RenameErrInfo) renameErr;
endinterface
interface Core;
// core request & indication
interface CoreReq coreReq;
interface CoreIndInv coreIndInv;
// coherent caches to LLC
interface ChildCacheToParent#(L1Way, void) dCacheToParent;
interface ChildCacheToParent#(L1Way, void) iCacheToParent;
// DMA to LLC
interface TlbMemClient tlbToMem;
// MMIO
interface MMIOCoreToPlatform mmioToPlatform;
// stats enable
method ActionValue#(Bool) sendDoStats;
method Action recvDoStats(Bool x);
// detect deadlock: only in use when macro CHECK_DEADLOCK is defined
interface CoreDeadlock deadlock;
// debug rename
interface CoreRenameDebug renameDebug;
// Bluespec: external interrupt requests targeting Machine and Supervisor modes
method Action setMEIP (Bit #(1) v);
method Action setSEIP (Bit #(1) v);
endinterface
// fixpoint to instantiate modules
interface CoreFixPoint;
interface Vector#(AluExeNum, AluExePipeline) aluExeIfc;
interface Vector#(FpuMulDivExeNum, FpuMulDivExePipeline) fpuMulDivExeIfc;
interface MemExePipeline memExeIfc;
method Action killAll; // kill everything: used by commit stage
interface Reg#(Bool) doStatsIfc;
endinterface
(* synthesize *)
module mkCore#(CoreId coreId)(Core);
let verbose = True;
Reg#(Bool) outOfReset <- mkReg(False);
rule rl_outOfReset if (!outOfReset);
$fwrite(stderr, "mkProc came out of reset\n");
outOfReset <= True;
endrule
Reg#(Bool) started <- mkReg(False);
// front end
FetchStage fetchStage <- mkFetchStage;
ITlb iTlb = fetchStage.iTlbIfc;
ICoCache iMem = fetchStage.iMemIfc;
// back end
RFileSynth rf <- mkRFileSynth;
// Bluespec: CsrFile including external interrupt request methods
CsrFile csrf <- mkCsrFile(zeroExtend(coreId)); // hartid in CSRF should be core id
RegRenamingTable regRenamingTable <- mkRegRenamingTable;
EpochManager epochManager <- mkEpochManager;
SpecTagManager specTagManager <- mkSpecTagManager;
ReorderBufferSynth rob <- mkReorderBufferSynth;
// We have two scoreboards: one conservative and other aggressive
// - Aggressive sb is checked at rename stage, so inst after rename may be issued early
// - Conservative sb is checked at reg read stage, to ensure correctness
// Every pipeline should set both sb if it needs to write reg
// - Conservative sb is set when data is written into rf
// - Aggressive sb is set when pipeline sends out wakeup for reservation staion
// Note that wakeup can be sent early if it knows when the data will be produced
ScoreboardCons sbCons <- mkScoreboardCons; // conservative sb
ScoreboardAggr sbAggr <- mkScoreboardAggr; // aggressive sb
// MMIO: need to track in flight CSR inst or interrupt; note we can at most
// 1 CSR inst or 1 interrupt in ROB, so just use 1 bit track it. Commit
// stage use port 0 to reset this, and Rename stage use port 1 to set this.
Ehr#(2, Bool) csrInstOrInterruptInflight <- mkEhr(False);
Reg#(Bool) csrInstOrInterruptInflight_commit = csrInstOrInterruptInflight[0];
Reg#(Bool) csrInstOrInterruptInflight_rename = csrInstOrInterruptInflight[1];
MMIOCoreInput mmioInIfc = (interface MMIOCoreInput;
interface fetch = fetchStage.mmioIfc;
method getMSIP = csrf.getMSIP;
method setMSIP = csrf.setMSIP;
method setMTIP = csrf.setMTIP;
method noInflightCSRInstOrInterrupt = !csrInstOrInterruptInflight[0];
method setTime = csrf.setTime;
endinterface);
MMIOCore mmio <- mkMMIOCore(mmioInIfc);
// fix point module to instantiate other function units
module mkCoreFixPoint#(CoreFixPoint fix)(CoreFixPoint);
// spec update
Vector#(AluExeNum, SpeculationUpdate) aluSpecUpdate;
for(Integer i = 0; i < valueof(AluExeNum); i = i+1) begin
aluSpecUpdate[i] = fix.aluExeIfc[i].specUpdate;
end
Vector#(FpuMulDivExeNum, SpeculationUpdate) fpuMulDivSpecUpdate;
for(Integer i = 0; i < valueof(FpuMulDivExeNum); i = i+1) begin
fpuMulDivSpecUpdate[i] = fix.fpuMulDivExeIfc[i].specUpdate;
end
GlobalSpecUpdate#(CorrectSpecPortNum, ConflictWrongSpecPortNum) globalSpecUpdate <- mkGlobalSpecUpdate(
joinSpeculationUpdate(
append(append(vec(regRenamingTable.specUpdate,
specTagManager.specUpdate,
fix.memExeIfc.specUpdate), aluSpecUpdate), fpuMulDivSpecUpdate)
),
rob.specUpdate
);
// whether perf data is collected
Reg#(Bool) doStatsReg <- mkConfigReg(False);
// redirect func
//function Action redirectFunc(Addr trap_pc, Maybe#(SpecTag) spec_tag, InstTag inst_tag );
//action
// if (verbose) $fdisplay(stdout, "[redirect_action] new pc = 0x%8x, spec_tag = ", trap_pc, fshow(spec_tag));
// epochManager.redirect;
// fetchStage.redirect(trap_pc);
// if (spec_tag matches tagged Valid .valid_spec_tag) begin
// globalSpecUpdate.incorrectSpec(valid_spec_tag, inst_tag);
// end
//endaction
//endfunction
// write aggressive elements + wakupe reservation stations
function Action writeAggr(Integer wrAggrPort, PhyRIndx dst);
action
sbAggr.setReady[wrAggrPort].put(dst);
for(Integer i = 0; i < valueof(AluExeNum); i = i+1) begin
fix.aluExeIfc[i].rsAluIfc.setRegReady[wrAggrPort].put(Valid (dst));
end
for(Integer i = 0; i < valueof(FpuMulDivExeNum); i = i+1) begin
fix.fpuMulDivExeIfc[i].rsFpuMulDivIfc.setRegReady[wrAggrPort].put(Valid (dst));
end
fix.memExeIfc.rsMemIfc.setRegReady[wrAggrPort].put(Valid (dst));
endaction
endfunction
// write conservative elements
function Action writeCons(Integer wrConsPort, PhyRIndx dst, Data data);
action
rf.write[wrConsPort].wr(dst, data);
sbCons.setReady[wrConsPort].put(dst);
endaction
endfunction
Vector#(AluExeNum, FIFO#(FetchTrainBP)) trainBPQ <- replicateM(mkFIFO);
Vector#(AluExeNum, AluExePipeline) aluExe;
for(Integer i = 0; i < valueof(AluExeNum); i = i+1) begin
Vector#(2, SendBypass) sendBypassIfc; // exe and finish
for(Integer sendPort = 0; sendPort < 2; sendPort = sendPort + 1) begin
sendBypassIfc[sendPort] = (interface SendBypass;
method Action send(PhyRIndx dst, Data data);
// broadcast bypass
Integer recvPort = valueof(AluExeNum) * sendPort + i;
for(Integer j = 0; j < valueof(FpuMulDivExeNum); j = j+1) begin
fix.fpuMulDivExeIfc[j].recvBypass[recvPort].recv(dst, data);
end
fix.memExeIfc.recvBypass[recvPort].recv(dst, data);
for(Integer j = 0; j < valueof(AluExeNum); j = j+1) begin
fix.aluExeIfc[j].recvBypass[recvPort].recv(dst, data);
end
endmethod
endinterface);
end
let aluExeInput = (interface AluExeInput;
method sbCons_lazyLookup = sbCons.lazyLookup[aluRdPort(i)].get;
method rf_rd1 = rf.read[aluRdPort(i)].rd1;
method rf_rd2 = rf.read[aluRdPort(i)].rd2;
method csrf_rd = csrf.rd;
method rob_getPC = rob.getOrigPC[i].get;
method rob_getPredPC = rob.getOrigPredPC[i].get;
method rob_setExecuted = rob.setExecuted_doFinishAlu[i].set;
method fetch_train_predictors = toPut(trainBPQ[i]).put;
method setRegReadyAggr = writeAggr(aluWrAggrPort(i));
interface sendBypass = sendBypassIfc;
method writeRegFile = writeCons(aluWrConsPort(i));
method Action redirect(Addr new_pc, SpecTag spec_tag, InstTag inst_tag);
if (verbose) begin
$display("[ALU redirect - %d] ", i, fshow(new_pc),
"; ", fshow(spec_tag), "; ", fshow(inst_tag));
end
epochManager.incrementEpoch;
fetchStage.redirect(new_pc);
globalSpecUpdate.incorrectSpec(False, spec_tag, inst_tag);
endmethod
method correctSpec = globalSpecUpdate.correctSpec[finishAluCorrectSpecPort(i)].put;
method doStats = doStatsReg._read;
endinterface);
aluExe[i] <- mkAluExePipeline(aluExeInput);
// truly call fetch method to train branch predictor
rule doFetchTrainBP;
let train <- toGet(trainBPQ[i]).get;
fetchStage.train_predictors(
train.pc, train.nextPc, train.iType, train.taken,
train.dpTrain, train.mispred
);
endrule
end
Vector#(FpuMulDivExeNum, FpuMulDivExePipeline) fpuMulDivExe;
for(Integer i = 0; i < valueof(FpuMulDivExeNum); i = i+1) begin
let fpuMulDivExeInput = (interface FpuMulDivExeInput;
method sbCons_lazyLookup = sbCons.lazyLookup[fpuMulDivRdPort(i)].get;
method rf_rd1 = rf.read[fpuMulDivRdPort(i)].rd1;
method rf_rd2 = rf.read[fpuMulDivRdPort(i)].rd2;
method rf_rd3 = rf.read[fpuMulDivRdPort(i)].rd3;
method csrf_rd = csrf.rd;
method rob_setExecuted = rob.setExecuted_doFinishFpuMulDiv[i].set;
method Action writeRegFile(PhyRIndx dst, Data data);
writeAggr(fpuMulDivWrAggrPort(i), dst);
writeCons(fpuMulDivWrConsPort(i), dst, data);
endmethod
method conflictWrongSpec = globalSpecUpdate.conflictWrongSpec[finishFpuMulDivConflictWrongSpecPort(i)].put(?);
method doStats = doStatsReg._read;
endinterface);
fpuMulDivExe[i] <- mkFpuMulDivExePipeline(fpuMulDivExeInput);
end
let memExeInput = (interface MemExeInput;
method sbCons_lazyLookup = sbCons.lazyLookup[memRdPort].get;
method rf_rd1 = rf.read[memRdPort].rd1;
method rf_rd2 = rf.read[memRdPort].rd2;
method csrf_rd = csrf.rd;
method rob_getPC = rob.getOrigPC[valueof(AluExeNum)].get; // last getPC port
method rob_setExecuted_doFinishMem = rob.setExecuted_doFinishMem;
method rob_setExecuted_deqLSQ = rob.setExecuted_deqLSQ;
method isMMIOAddr = mmio.isMMIOAddr;
method mmioReq = mmio.dataReq;
method mmioRespVal = mmio.dataRespVal;
method mmioRespDeq = mmio.dataRespDeq;
method setRegReadyAggr_mem = writeAggr(memWrAggrPort);
method setRegReadyAggr_forward = writeAggr(forwardWrAggrPort);
method writeRegFile = writeCons(memWrConsPort);
method doStats = doStatsReg._read;
endinterface);
let memExe <- mkMemExePipeline(memExeInput);
interface aluExeIfc = aluExe;
interface fpuMulDivExeIfc = fpuMulDivExe;
interface memExeIfc = memExe;
method Action killAll;
globalSpecUpdate.incorrectSpec(True, ?, ?);
endmethod
interface doStatsIfc = doStatsReg;
endmodule
CoreFixPoint coreFix <- moduleFix(mkCoreFixPoint);
Vector#(AluExeNum, ReservationStationAlu) reservationStationAlu;
for(Integer i = 0; i < valueof(AluExeNum); i = i+1) begin
reservationStationAlu[i] = coreFix.aluExeIfc[i].rsAluIfc;
end
Vector#(FpuMulDivExeNum, ReservationStationFpuMulDiv) reservationStationFpuMulDiv;
for(Integer i = 0; i < valueof(FpuMulDivExeNum); i = i+1) begin
reservationStationFpuMulDiv[i] = coreFix.fpuMulDivExeIfc[i].rsFpuMulDivIfc;
end
ReservationStationMem reservationStationMem = coreFix.memExeIfc.rsMemIfc;
DTlbSynth dTlb = coreFix.memExeIfc.dTlbIfc;
SplitLSQ lsq = coreFix.memExeIfc.lsqIfc;
StoreBuffer stb = coreFix.memExeIfc.stbIfc;
DCoCache dMem = coreFix.memExeIfc.dMemIfc;
// L2 TLB
L2Tlb l2Tlb <- mkL2Tlb;
mkTlbConnect(iTlb.toParent, dTlb.toParent, l2Tlb.toChildren);
// flags to flush
Reg#(Bool) flush_tlbs <- mkReg(False);
Reg#(Bool) update_vm_info <- mkReg(False);
Reg#(Bool) flush_reservation <- mkReg(False);
`ifdef SECURITY
Reg#(Bool) flush_caches <- mkReg(False);
Reg#(Bool) flush_brpred <- mkReg(False);
`else
Reg#(Bool) flush_caches <- mkReadOnlyReg(False);
Reg#(Bool) flush_brpred <- mkReadOnlyReg(False);
`endif
`ifdef SELF_INV_CACHE
Reg#(Bool) reconcile_i <- mkReg(False);
`else
Reg#(Bool) reconcile_i <- mkReadOnlyReg(False);
`endif
`ifdef SELF_INV_CACHE
`ifdef SYSTEM_SELF_INV_L1D
Reg#(Bool) reconcile_d <- mkReg(False);
`else // !SYSTEM_SELF_INV_L1D
Reg#(Bool) reconcile_d <- mkReadOnlyReg(False);
`endif // SYSTEM_SELF_INV_L1D
`else // !SELF_INV_CACHE
Reg#(Bool) reconcile_d <- mkReadOnlyReg(False);
`endif // SELF_INV_CACHE
// performance counters
Reg#(Bool) doStats = coreFix.doStatsIfc; // whether data is collected
`ifdef PERF_COUNT
// OOO execute stag (in AluExePipeline and MemExePipeline)
// commit stage (many in CommitStage.bsv)
// cycle
Count#(Data) cycleCnt <- mkCount(0);
// buffer/tags size
Count#(Data) ldqFullCycles <- mkCount(0);
Count#(Data) stqFullCycles <- mkCount(0);
Count#(Data) robFullCycles <- mkCount(0);
Count#(Data) aluRS0FullCycles <- mkCount(0);
Count#(Data) aluRS1FullCycles <- mkCount(0);
Count#(Data) fpuMulDivRSFullCycles <- mkCount(0);
Count#(Data) memRSFullCycles <- mkCount(0);
Count#(Data) epochFullCycles <- mkCount(0);
Count#(Data) specTagFullCycles <- mkCount(0);
// FIFOs to connect performance counters
FIFO#(ExeStagePerfType) exePerfReqQ <- mkFIFO1;
FIFO#(ComStagePerfType) comPerfReqQ <- mkFIFO1;
FIFO#(CoreSizePerfType) sizePerfReqQ <- mkFIFO1;
Fifo#(1, PerfResp#(ExeStagePerfType)) exePerfRespQ <- mkCFFifo;
Fifo#(1, PerfResp#(ComStagePerfType)) comPerfRespQ <- mkCFFifo;
Fifo#(1, PerfResp#(CoreSizePerfType)) sizePerfRespQ <- mkCFFifo;
// FIFO of perf resp
FIFO#(ProcPerfResp) perfRespQ <- mkFIFO1;
`endif
// FIFO of perf req
FIFO#(ProcPerfReq) perfReqQ <- mkFIFO1;
// -- End of performance counters
`ifdef CHECK_DEADLOCK
// when to start deadlock checking
Reg#(Bool) startDeadlockCheck <- mkReg(False);
FIFO#(void) deadlockCheckStartedQ <- mkFIFO;
rule doStartDeadlockCheck(!startDeadlockCheck && started);
startDeadlockCheck <= True;
deadlockCheckStartedQ.enq(?);
endrule
`endif
// Rename stage
let renameInput = (interface RenameInput;
interface fetchIfc = fetchStage;
interface robIfc = rob;
interface rtIfc = regRenamingTable;
interface sbConsIfc = sbCons;
interface sbAggrIfc = sbAggr;
interface csrfIfc = csrf;
interface emIfc = epochManager;
interface smIfc = specTagManager;
interface rsAluIfc = reservationStationAlu;
interface rsFpuMulDivIfc = reservationStationFpuMulDiv;
interface rsMemIfc = reservationStationMem;
interface lsqIfc = lsq;
method pendingMMIOPRq = mmio.hasPendingPRq;
method issueCsrInstOrInterrupt = csrInstOrInterruptInflight_rename._write(True);
method Bool checkDeadlock;
`ifdef CHECK_DEADLOCK
return startDeadlockCheck;
`else
return False;
`endif
endmethod
method doStats = coreFix.doStatsIfc._read;
endinterface);
RenameStage renameStage <- mkRenameStage(renameInput);
// commit stage
let commitInput = (interface CommitInput;
interface robIfc = rob;
interface rtIfc = regRenamingTable;
interface csrfIfc = csrf;
method stbEmpty = stb.isEmpty;
method stqEmpty = lsq.stqEmpty;
method lsqSetAtCommit = lsq.setAtCommit;
method tlbNoPendingReq = iTlb.noPendingReq && dTlb.noPendingReq;
method setFlushTlbs = flush_tlbs._write(True);
method setUpdateVMInfo = update_vm_info._write(True);
method setFlushReservation = flush_reservation._write(True);
method setFlushBrPred = flush_brpred._write(True);
method setFlushCaches = flush_caches._write(True);
method setReconcileI = reconcile_i._write(True);
method setReconcileD = reconcile_d._write(True);
method killAll = coreFix.killAll;
method redirectPc = fetchStage.redirect;
method setFetchWaitRedirect = fetchStage.setWaitRedirect;
method incrementEpoch = epochManager.incrementEpoch;
method commitCsrInstOrInterrupt = csrInstOrInterruptInflight_commit._write(False);
method doStats = coreFix.doStatsIfc._read;
method Bool checkDeadlock;
`ifdef CHECK_DEADLOCK
return startDeadlockCheck;
`else
return False;
`endif
endmethod
endinterface);
CommitStage commitStage <- mkCommitStage(commitInput);
// send rob enq time to reservation stations
(* fire_when_enabled, no_implicit_conditions *)
rule sendRobEnqTime;
InstTime t = rob.getEnqTime;
reservationStationMem.setRobEnqTime(t);
for(Integer i = 0; i < valueof(FpuMulDivExeNum); i = i+1) begin
reservationStationFpuMulDiv[i].setRobEnqTime(t);
end
for(Integer i = 0; i < valueof(AluExeNum); i = i+1) begin
reservationStationAlu[i].setRobEnqTime(t);
end
endrule
// preempt has 2 functions here
// 1. break scheduling cycles
// 2. XXX since csrf is configReg now, we should not let this rule fire together with doCommit
// because we read csrf here and write csrf in doCommit
// TODO We can use wires to catch flush / updateVM enable sigals, because
// there cannot be any instruction in pipeline (there can be poisoned inst
// which cannot change CSR or link reg in D$), so doCommit cannot fire.
// MMIO manager may change pending interrupt bits, but will not affect VM
// info.
(* preempts = "prepareCachesAndTlbs, commitStage.doCommitTrap_handle" *)
(* preempts = "prepareCachesAndTlbs, commitStage.doCommitSystemInst" *)
rule prepareCachesAndTlbs(flush_reservation || flush_tlbs || update_vm_info);
if (flush_reservation) begin
flush_reservation <= False;
dMem.resetLinkAddr;
end
if (flush_tlbs) begin
flush_tlbs <= False;
iTlb.flush;
dTlb.flush;
end
if (update_vm_info) begin
update_vm_info <= False;
let vmI = csrf.vmI;
let vmD = csrf.vmD;
iTlb.updateVMInfo(vmI);
dTlb.updateVMInfo(vmD);
l2Tlb.updateVMInfo(vmI, vmD);
end
endrule
`ifdef SECURITY
// Use wires to capture flush regs and empty signals. This is ok because
// there cannot be any activity to make empty -> not-empty or need-flush ->
// no-need-flush when we are trying to flush.
PulseWire doFlushCaches <- mkPulseWire;
PulseWire doFlushBrPred <- mkPulseWire;
rule setDoFlushCaches(flush_caches && fetchStage.emptyForFlush && lsq.noWrongPathLoads);
doFlushCaches.send;
endrule
rule setDoFlushBrPred(flush_brpred && fetchStage.emptyForFlush);
doFlushBrPred.send;
endrule
// security flush cache: need to wait for wrong path loads or inst fetches
// to finish
rule flushCaches(doFlushCaches);
flush_caches <= False;
iMem.flush;
dMem.flush;
endrule
// security flush branch predictors: wait for wrong path inst fetches to
// finish
rule flushBrPred(doFlushBrPred);
flush_brpred <= False;
fetchStage.flush_predictors;
endrule
`endif
`ifdef SELF_INV_CACHE
// Use wires to capture flush regs and empty signals. This is ok because
// there cannot be any activity to make empty -> not-empty or need-flush ->
// no-need-flush when we are trying to flush.
PulseWire doReconcileI <- mkPulseWire;
// We don't really need to wait for fetch to be empty, but just in case we
// back pressure I TLB because I$ is reconciling.
rule setDoReconcileI(reconcile_i && fetchStage.emptyForFlush);
doReconcileI.send;
endrule
rule reconcileI(doReconcileI);
reconcile_i <= False;
iMem.reconcile;
endrule
`ifdef SYSTEM_SELF_INV_L1D
PulseWire doReconcileD <- mkPulseWire;
Reg#(Bool) waitReconcileD <- mkReg(False);
// We don't really need to wait for lsq empty, but just in case
rule setDoReconcileD(reconcile_d && lsq.noWrongPathLoads);
doReconcileD.send;
endrule
rule startReconcileD(doReconcileD && !waitReconcileD);
coreFix.memExeIfc.reconcile.request.put(?);
waitReconcileD <= True;
endrule
rule completeReconcileD(waitReconcileD);
let unused <- coreFix.memExeIfc.reconcile.response.get;
waitReconcileD <= False;
reconcile_d <= False;
endrule
`endif // SYSTEM_SELF_INV_L1D
`endif // SELF_INV_CACHE
rule readyToFetch(
!flush_reservation && !flush_tlbs && !update_vm_info
&& iTlb.flush_done && dTlb.flush_done
`ifdef SECURITY
&& !flush_caches && !flush_brpred
&& iMem.flush_done && dMem.flush_done
&& fetchStage.flush_predictors_done
`endif
`ifdef SELF_INV_CACHE
&& !reconcile_i && iMem.reconcile_done
`ifdef SYSTEM_SELF_INV_L1D
&& !reconcile_d
`endif
`endif
);
fetchStage.done_flushing();
endrule
`ifdef PERF_COUNT
// incr cycle count
(* fire_when_enabled, no_implicit_conditions *)
rule incCycleCnt(doStats);
cycleCnt.incr(1);
endrule
// incr buffer full cycles
(* fire_when_enabled, no_implicit_conditions *)
rule incLdQFull(doStats && lsq.ldqFull_ehrPort0);
ldqFullCycles.incr(1);
endrule
(* fire_when_enabled, no_implicit_conditions *)
rule incStQFull(doStats && lsq.stqFull_ehrPort0);
stqFullCycles.incr(1);
endrule
(* fire_when_enabled, no_implicit_conditions *)
rule incROBFull(doStats && rob.isFull_ehrPort0);
robFullCycles.incr(1);
endrule
(* fire_when_enabled, no_implicit_conditions *)
rule incAluRS0Full(doStats && reservationStationAlu[0].isFull_ehrPort0);
aluRS0FullCycles.incr(1);
endrule
(* fire_when_enabled, no_implicit_conditions *)
rule incAluRS1Full(doStats && reservationStationAlu[1].isFull_ehrPort0);
aluRS1FullCycles.incr(1);
endrule
(* fire_when_enabled, no_implicit_conditions *)
rule incFpuMulDivRSFull(doStats && reservationStationFpuMulDiv[0].isFull_ehrPort0);
fpuMulDivRSFullCycles.incr(1);
endrule
(* fire_when_enabled, no_implicit_conditions *)
rule incMemRSFull(doStats && reservationStationMem.isFull_ehrPort0);
memRSFullCycles.incr(1);
endrule
(* fire_when_enabled, no_implicit_conditions *)
rule incEpochFull(doStats && epochManager.isFull_ehrPort0);
epochFullCycles.incr(1);
endrule
(* fire_when_enabled, no_implicit_conditions *)
rule incSpecTagFull(doStats && specTagManager.isFull_ehrPort0);
specTagFullCycles.incr(1);
endrule
// broadcast whether we should collect data
rule broadcastDoStats;
let stats = csrf.doPerfStats;
doStats <= stats;
iMem.perf.setStatus(stats);
dMem.perf.setStatus(stats);
iTlb.perf.setStatus(stats);
dTlb.perf.setStatus(stats);
l2Tlb.perf.setStatus(stats);
fetchStage.perf.setStatus(stats);
if(stats && !doStats) begin
$display("[stats] enabled");
end
else if(!stats && doStats) begin
$display("[stats] disabled");
end
endrule
// dispatch perf req
rule dispathPerfReq;
perfReqQ.deq;
let r = perfReqQ.first;
case(r.loc)
ICache: begin
iMem.perf.req(unpack(truncate(r.pType)));
end
DCache: begin
dMem.perf.req(unpack(truncate(r.pType)));
end
ITlb: begin
iTlb.perf.req(unpack(truncate(r.pType)));
end
DTlb: begin
dTlb.perf.req(unpack(truncate(r.pType)));
end
L2Tlb: begin
l2Tlb.perf.req(unpack(truncate(r.pType)));
end
DecStage: begin
fetchStage.perf.req(unpack(truncate(r.pType)));
end
ExeStage: begin
exePerfReqQ.enq(unpack(truncate(r.pType)));
end
ComStage: begin
comPerfReqQ.enq(unpack(truncate(r.pType)));
end
CoreSize: begin
sizePerfReqQ.enq(unpack(truncate(r.pType)));
end
default: begin
$fwrite(stderr, "[WARNING] unrecognzied perf req location ", fshow(r.loc), "\n");
doAssert(False, "unknown perf location");
end
endcase
endrule
// handle perf req: exe stage
rule readPerfCnt_Exe;
function Data getAluCnt(ExeStagePerfType pType);
Data cnt = 0;
for(Integer i = 0; i < valueof(AluExeNum); i = i+1) begin
cnt = cnt + coreFix.aluExeIfc[i].getPerf(pType);
end
return cnt;
endfunction
function Data getFpuMulDivCnt(ExeStagePerfType pType);
Data cnt = 0;
for(Integer i = 0; i < valueof(FpuMulDivExeNum); i = i+1) begin
cnt = cnt + coreFix.fpuMulDivExeIfc[i].getPerf(pType);
end
return cnt;
endfunction
let pType <- toGet(exePerfReqQ).get;
Data data = (case(pType)
SupRenameCnt, SpecNoneCycles, SpecNonMemCycles: renameStage.getPerf(pType);
ExeRedirectBr, ExeRedirectJr, ExeRedirectOther: getAluCnt(pType);
ExeTlbExcep, ExeScSuccessCnt,
ExeLrScAmoAcqCnt, ExeLrScAmoRelCnt,
ExeFenceAcqCnt, ExeFenceRelCnt, ExeFenceCnt,
ExeLdStallByLd, ExeLdStallBySt, ExeLdStallBySB,
ExeLdForward, ExeLdMemLat, ExeStMemLat,
ExeLdToUseLat, ExeLdToUseCnt: coreFix.memExeIfc.getPerf(pType);
ExeIntMulCnt, ExeIntDivCnt,
ExeFpFmaCnt, ExeFpDivCnt, ExeFpSqrtCnt: getFpuMulDivCnt(pType);
default: 0;
endcase);
exePerfRespQ.enq(PerfResp {
pType: pType,
data: data
});
endrule
// handle perf req: com stage
rule readPerfCnt_Com;
let pType <- toGet(comPerfReqQ).get;
Data data = (case(pType)
CycleCnt: cycleCnt;
default: commitStage.getPerf(pType);
endcase);
comPerfRespQ.enq(PerfResp {
pType: pType,
data: data
});
endrule
// handle perf req: core size
rule readPerfCnt_Size;
let pType <- toGet(sizePerfReqQ).get;
Data data = (case(pType)
LdQFullCycles: ldqFullCycles;
StQFullCycles: stqFullCycles;
ROBFullCycles: robFullCycles;
AluRS0FullCycles: aluRS0FullCycles;
AluRS1FullCycles: aluRS1FullCycles;
FpuMulDivRSFullCycles: fpuMulDivRSFullCycles;
MemRSFullCycles: memRSFullCycles;
EpochFullCycles: epochFullCycles;
SpecTagFullCycles: specTagFullCycles;
default: 0;
endcase);
sizePerfRespQ.enq(PerfResp {
pType: pType,
data: data
});
endrule
// gather perf resp
rule gatherPerfResp;
Maybe#(ProcPerfResp) resp = Invalid;
if(iMem.perf.respValid) begin
let r <- iMem.perf.resp;
resp = Valid(ProcPerfResp {
loc: ICache,
pType: zeroExtend(pack(r.pType)),
data: r.data
});
end
else if(dMem.perf.respValid) begin
let r <- dMem.perf.resp;
resp = Valid(ProcPerfResp {
loc: DCache,
pType: zeroExtend(pack(r.pType)),
data: r.data
});
end
else if(iTlb.perf.respValid) begin
let r <- iTlb.perf.resp;
resp = Valid(ProcPerfResp {
loc: ITlb,
pType: zeroExtend(pack(r.pType)),
data: r.data
});
end
else if(dTlb.perf.respValid) begin
let r <- dTlb.perf.resp;
resp = Valid(ProcPerfResp {
loc: DTlb,
pType: zeroExtend(pack(r.pType)),
data: r.data
});
end
else if(l2Tlb.perf.respValid) begin
let r <- l2Tlb.perf.resp;
resp = Valid(ProcPerfResp {
loc: L2Tlb,
pType: zeroExtend(pack(r.pType)),
data: r.data
});
end
else if(fetchStage.perf.respValid) begin
let r <- fetchStage.perf.resp;
resp = Valid(ProcPerfResp {
loc: DecStage,
pType: zeroExtend(pack(r.pType)),
data: r.data
});
end
else if(exePerfRespQ.notEmpty) begin
let r <- toGet(exePerfRespQ).get;
resp = Valid(ProcPerfResp {
loc: ExeStage,
pType: zeroExtend(pack(r.pType)),
data: r.data
});
end
else if(comPerfRespQ.notEmpty) begin
let r <- toGet(comPerfRespQ).get;
resp = Valid(ProcPerfResp {
loc: ComStage,
pType: zeroExtend(pack(r.pType)),
data: r.data
});
end
else if(sizePerfRespQ.notEmpty) begin
let r <- toGet(sizePerfRespQ).get;
resp = Valid (ProcPerfResp {
loc: CoreSize,
pType: zeroExtend(pack(r.pType)),
data: r.data
});
end
// enq to resp Q
if(resp matches tagged Valid .r) begin
perfRespQ.enq(r);
end
endrule
`endif
interface CoreReq coreReq;
method Action start(
Bit#(64) startpc,
Addr toHostAddr, Addr fromHostAddr
);
fetchStage.start(startpc);
started <= True;
mmio.setHtifAddrs(toHostAddr, fromHostAddr);
// start rename debug
commitStage.startRenameDebug;
endmethod
method Action perfReq(PerfLocation loc, PerfType t);
perfReqQ.enq(ProcPerfReq {
loc: loc,
pType: t
});
endmethod
endinterface
interface CoreIndInv coreIndInv;
method ActionValue#(ProcPerfResp) perfResp;
`ifdef PERF_COUNT
perfRespQ.deq;
return perfRespQ.first;
`else
perfReqQ.deq;
let r = perfReqQ.first;
return ProcPerfResp {
loc: r.loc,
pType: r.pType,
data: 0
};
`endif
endmethod
method terminate = csrf.terminate;
endinterface
interface dCacheToParent = dMem.to_parent;
interface iCacheToParent = iMem.to_parent;
interface tlbToMem = l2Tlb.toMem;
interface mmioToPlatform = mmio.toP;
method sendDoStats = csrf.sendDoStats;
method recvDoStats = csrf.recvDoStats;
// deadlock check
interface CoreDeadlock deadlock;
interface dCacheCRqStuck = dMem.cRqStuck;
interface dCachePRqStuck = dMem.pRqStuck;
interface iCacheCRqStuck = iMem.cRqStuck;
interface iCachePRqStuck = iMem.pRqStuck;
interface renameInstStuck = renameStage.renameInstStuck;
interface renameCorrectPathStuck = renameStage.renameCorrectPathStuck;
interface commitInstStuck = commitStage.commitInstStuck;
interface commitUserInstStuck = commitStage.commitUserInstStuck;
`ifdef CHECK_DEADLOCK
interface checkStarted = toGet(deadlockCheckStartedQ);
`else
interface checkStarted = nullGet;
`endif
endinterface
// rename debug
interface CoreRenameDebug renameDebug;
interface renameErr = commitStage.renameErr;
endinterface
// Bluespec: external interrupt requests targeting Machine and Supervisor modes
method Action setMEIP (v) = csrf.setMEIP (v);
method Action setSEIP (v) = csrf.setSEIP (v);
endmodule

842
src_Core/CPU/CsrFile.bsv Normal file
View File

@@ -0,0 +1,842 @@
// Copyright (c) 2017 Massachusetts Institute of Technology
//
// Permission is hereby granted, free of charge, to any person
// obtaining a copy of this software and associated documentation
// files (the "Software"), to deal in the Software without
// restriction, including without limitation the rights to use, copy,
// modify, merge, publish, distribute, sublicense, and/or sell copies
// of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be
// included in all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
// MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
// BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
// ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
// CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
// SOFTWARE.
// Portions Copyright (c) Bluespec, Inc.
`include "ProcConfig.bsv"
import Types::*;
import ProcTypes::*;
import DefaultValue::*;
import ConcatReg::*;
import ConfigReg::*;
import Ehr::*;
import Fifo::*;
import Vector::*;
import FIFO::*;
import GetPut::*;
import BuildVector::*;
//import TRNG::*;
interface CsrFile;
// Read
method Data rd(CSR csr);
// normal write by CSRXXX inst to any CSR
method Action csrInstWr(CSR csr, Data x);
// normal write by FPU inst to FPU CSR
method Bool fpuInstNeedWr(Bit#(5) fflags, Bool fpu_dirty);
method Action fpuInstWr(Bit#(5) fflags); // FPU must become dirty
// Methods for handling traps
method Maybe#(Interrupt) pending_interrupt;
method ActionValue#(Addr) trap(Trap t, Addr pc, Addr faultAddr);
method ActionValue#(Addr) sret;
method ActionValue#(Addr) mret;
// Outputs for CSRs that the rest of the processor needs to know about
method VMInfo vmI;
method VMInfo vmD;
method CsrDecodeInfo decodeInfo;
// Updating minstret CSR outside of normal CSR write instructions. This
// increment will see the effect of normal CSR write.
method Action incInstret(SupCnt x);
// update copy of mtime
method Action setTime(Data t);
// MSIP/MTIP bits for external world (e.g., for MMIO and timer interrupt).
// XXX These methods should only be called when the processor backend
// pipeline does not contain any CSRXXX inst or corresponding interrupt
// inst (the inst which is turned into an interrupt). This ensures that
// CSRXXX and interrupt are handled atomically. MSIP/MTIP should not
// affect other insts (e.g., address translation of loads/stores),
// synchronous exceptions and other types of interrupts.
method Bit#(1) getMSIP;
method Action setMSIP(Bit#(1) v);
method Action setMTIP(Bit#(1) v);
// Bluespec: external interrupts targeting machine and supervisor modes
method Action setMEIP (Bit #(1) v);
method Action setSEIP (Bit #(1) v);
// performance stats is collected or not
method Bool doPerfStats;
// send/recv updates on stats CSR globally
method ActionValue#(Bool) sendDoStats;
method Action recvDoStats(Bool s);
// terminate
method ActionValue#(void) terminate;
endinterface
// Fancy Reg functions
function Reg#(Bit#(n)) truncateReg(Reg#(Bit#(m)) r) provisos (Add#(a__,n,m));
return (interface Reg;
method Bit#(n) _read = truncate(r._read);
method Action _write(Bit#(n) x) = r._write({truncateLSB(r._read), x});
endinterface);
endfunction
function Reg#(Bit#(n)) truncateRegLSB(Reg#(Bit#(m)) r) provisos (Add#(a__,n,m));
return (interface Reg;
method Bit#(n) _read = truncateLSB(r._read);
method Action _write(Bit#(n) x) = r._write({x, truncate(r._read)});
endinterface);
endfunction
function Reg#(Bit#(n)) zeroExtendReg(Reg#(Bit#(m)) r) provisos (Add#(a__,m,n));
return (interface Reg;
method Bit#(n) _read = zeroExtend(r._read);
method Action _write(Bit#(n) x) = r._write(truncate(x));
endinterface);
endfunction
function Reg#(t) readOnlyReg(t r);
return (interface Reg;
method t _read = r;
method Action _write(t x) = noAction;
endinterface);
endfunction
// module version of readOnlyReg for convenience
module mkReadOnlyReg#(t x)(Reg#(t));
return readOnlyReg(x);
endmodule
function Reg#(t) regFromReadOnly(ReadOnly#(t) r);
return (interface Reg;
method t _read = r._read;
method Action _write(t x);
noAction;
endmethod
endinterface);
endfunction
function Reg#(t) addWriteSideEffect(Reg#(t) r, Action a);
return (interface Reg;
method t _read = r._read;
method Action _write(t x);
r._write(x);
a;
endmethod
endinterface);
endfunction
function Bool has_csr_permission(CSR csr, Bit#(2) prv, Bool write);
Bit#(12) csr_index = pack(csr);
return ((prv >= csr_index[9:8]) && (!write || (csr_index[11:10] != 2'b11)));
endfunction
// non-standard terminate CSR
interface Terminate;
interface Reg#(Data) reg_ifc;
method ActionValue#(void) terminate;
endinterface
module mkTerminate(Terminate);
FIFO#(void) terminateQ <- mkFIFO1;
interface Reg reg_ifc;
method Action _write(Data x);
terminateQ.enq(?);
$display(
"[Terminate CSR] being written (val = %x), ",
"send terminate signal to host", x
);
endmethod
method Data _read = 0;
endinterface
method terminate = toGet(terminateQ).get;
endmodule
// stats CSR: there is only one copy in the whole multiprocessor, so any write
// to stats CSR will be broadcasted
interface StatsCsr;
interface Reg#(Data) reg_ifc;
method Bool doPerfStats;
// send/recv updates on stats CSR globally
method ActionValue#(Bool) sendDoStats;
method Action recvDoStats(Bool s);
endinterface
module mkStatsCsr(StatsCsr);
Reg#(Bool) doStats <- mkConfigReg(False);
FIFO#(Bool) writeQ <- mkFIFO1;
interface Reg reg_ifc;
method Data _read = zeroExtend(pack(doStats));
method Action _write(Data x);
writeQ.enq(unpack(truncate(x)));
endmethod
endinterface
method Bool doPerfStats = doStats;
method ActionValue#(Bool) sendDoStats;
writeQ.deq;
return writeQ.first;
endmethod
method Action recvDoStats(Bool s);
doStats <= s;
endmethod
endmodule
// same as EHR except that read port 0 is not ordered with other methods. Read
// port 1 will still get bypassing from write port 0.
module mkConfigEhr#(t init)(Ehr#(n, t)) provisos(Bits#(t, w));
Ehr#(n, t) data <- mkEhr(init);
Wire#(t) read <- mkBypassWire;
(* fire_when_enabled, no_implicit_conditions *)
rule setRead;
read <= data[0];
endrule
Ehr#(n, t) ifc = ?;
ifc[0] = (interface Reg;
method _read = read._read;
method _write = data[0]._write;
endinterface);
for(Integer i = 1; i < valueOf(n); i = i+1) begin
ifc[i] = (interface Reg;
method _read = data[i]._read;
method _write = data[i]._write;
endinterface);
end
return ifc;
endmodule
module mkCsrFile #(Data hartid)(CsrFile);
RiscVISASubset isa = defaultValue;
// To save from bypassing logic, CSR reads will get stale value
let mkCsrReg = mkConfigReg;
let mkCsrEhr = mkConfigEhr;
// current prv level (this is not a csr...)
Reg#(Bit#(2)) prv_reg <- mkCsrReg(prvM);
// Machine level CSRs
// mstatus
Reg#(Bit#(2)) xs_reg <- mkReadOnlyReg(0); // XXX no extension
Reg#(Bit#(2)) fs_reg <- (isa.f || isa.d) ? mkCsrReg(0) : mkReadOnlyReg(0);
Reg#(Bit#(1)) sd_reg = readOnlyReg(
((xs_reg == 2'b11) || (fs_reg == 2'b11)) ? 1 : 0
);
Reg#(Bit#(2)) sxl_reg = readOnlyReg(getXLBits);
Reg#(Bit#(2)) uxl_reg = readOnlyReg(getXLBits);
Reg#(Bit#(1)) tsr_reg <- mkCsrReg(0);
Reg#(Bit#(1)) tw_reg <- mkCsrReg(0);
Reg#(Bit#(1)) tvm_reg <- mkCsrReg(0);
Reg#(Bit#(1)) mxr_reg <- mkCsrReg(0);
Reg#(Bit#(1)) sum_reg <- mkCsrReg(0);
Reg#(Bit#(1)) mprv_reg <- mkCsrReg(0);
Reg#(Bit#(2)) mpp_reg <- mkCsrReg(0);
Reg#(Bit#(1)) spp_reg <- mkCsrReg(0);
Vector#(4, Reg#(Bit#(2))) prev_prv_vec = vec(
// prev_prv_vec[x]: privilege mode before trapping into mode x
readOnlyReg(prvU), // upp
concatReg2(readOnlyReg(1'b0), spp_reg), // spp
readOnlyReg(2'b0), // reserved
mpp_reg
);
Vector#(4, Reg#(Bit#(1))) ie_vec = replicate(
readOnlyReg(0) // ie_vec[x]: interrupt enable for mode x
);
ie_vec[prvU] <- mkCsrReg(0);
ie_vec[prvS] <- mkCsrReg(0);
ie_vec[prvM] <- mkCsrReg(0);
Vector#(4, Reg#(Bit#(1))) prev_ie_vec = replicate(
readOnlyReg(0) // prev_ie_vec[x]: ie_vec[x] before trapping into mode x
);
prev_ie_vec[prvU] <- mkCsrReg(0);
prev_ie_vec[prvS] <- mkCsrReg(0);
prev_ie_vec[prvM] <- mkCsrReg(0);
Reg#(Data) mstatus_csr = concatReg24(
sd_reg, readOnlyReg(27'b0), sxl_reg, uxl_reg, readOnlyReg(9'b0),
tsr_reg, tw_reg, tvm_reg, mxr_reg, sum_reg, mprv_reg, xs_reg, fs_reg,
mpp_reg, readOnlyReg(2'b0), spp_reg,
prev_ie_vec[prvM], readOnlyReg(1'b0),
prev_ie_vec[prvS], prev_ie_vec[prvU],
ie_vec[prvM], readOnlyReg(1'b0),
ie_vec[prvS], ie_vec[prvU]
);
// misa
Reg#(Data) misa_csr = readOnlyReg({getXLBits, 36'b0, getExtensionBits(isa)});
// medeleg: some exceptions don't exist, fix corresponding bits to 0
Reg#(Bit#(1)) medeleg_15_reg <- mkCsrReg(0); // cause 15
Reg#(Bit#(3)) medeleg_13_11_reg <- mkCsrReg(0); // case 13-11
Reg#(Bit#(10)) medeleg_9_0_reg <- mkCsrReg(0); // cause 9-0
Reg#(Data) medeleg_csr = concatReg6(
readOnlyReg(48'b0), medeleg_15_reg,
readOnlyReg(1'b0), medeleg_13_11_reg,
readOnlyReg(1'b0), medeleg_9_0_reg
);
// mideleg: some interrupts don't exist, fix corresponding bits to 0
Reg#(Bit#(1)) mideleg_11_reg <- mkCsrReg(0);
Reg#(Bit#(3)) mideleg_9_7_reg <- mkCsrReg(0);
Reg#(Bit#(3)) mideleg_5_3_reg <- mkCsrReg(0);
Reg#(Bit#(2)) mideleg_1_0_reg <- mkCsrReg(0);
Reg#(Data) mideleg_csr = concatReg8(
readOnlyReg(52'b0), mideleg_11_reg,
readOnlyReg(1'b0), mideleg_9_7_reg,
readOnlyReg(1'b0), mideleg_5_3_reg,
readOnlyReg(1'b0), mideleg_1_0_reg
);
// mie
Vector#(4, Reg#(Bit#(1))) external_int_en_vec = replicate(readOnlyReg(0));
external_int_en_vec[prvU] <- mkCsrReg(0);
external_int_en_vec[prvS] <- mkCsrReg(0);
external_int_en_vec[prvM] <- mkCsrReg(0);
Vector#(4, Reg#(Bit#(1))) timer_int_en_vec = replicate(readOnlyReg(0));
timer_int_en_vec[prvU] <- mkCsrReg(0);
timer_int_en_vec[prvS] <- mkCsrReg(0);
timer_int_en_vec[prvM] <- mkCsrReg(0);
Vector#(4, Reg#(Bit#(1))) software_int_en_vec = replicate(readOnlyReg(0));
software_int_en_vec[prvU] <- mkCsrReg(0);
software_int_en_vec[prvS] <- mkCsrReg(0);
software_int_en_vec[prvM] <- mkCsrReg(0);
Reg#(Data) mie_csr = concatReg13(
readOnlyReg(52'b0),
external_int_en_vec[prvM], readOnlyReg(1'b0),
external_int_en_vec[prvS], external_int_en_vec[prvU],
timer_int_en_vec[prvM], readOnlyReg(1'b0),
timer_int_en_vec[prvS], timer_int_en_vec[prvU],
software_int_en_vec[prvM], readOnlyReg(1'b0),
software_int_en_vec[prvS], software_int_en_vec[prvU]
);
// mtvec
Reg#(Bit#(62)) mtvec_base_hi_reg <- mkCsrReg(0); // this is BASE[63:2]
Reg#(Bit#(1)) mtvec_mode_low_reg <- mkCsrReg(0); // this is MODE[0]
Reg#(Data) mtvec_csr = concatReg3(
mtvec_base_hi_reg, readOnlyReg(1'b0), mtvec_mode_low_reg
);
// mcounteren
Reg#(Bit#(1)) mcounteren_ir_reg <- mkCsrReg(0);
Reg#(Bit#(1)) mcounteren_tm_reg <- mkCsrReg(0);
Reg#(Bit#(1)) mcounteren_cy_reg <- mkCsrReg(0);
Reg#(Data) mcounteren_csr = concatReg5(
readOnlyReg(32'b0),
readOnlyReg(29'b0), // hpmcounter 3-31 not accessible in S mode
mcounteren_ir_reg, mcounteren_tm_reg, mcounteren_cy_reg
);
// mscratch
Reg#(Data) mscratch_csr <- mkCsrReg(0);
// mepc: FIXME Since we don't have C extension, mepc should be 4-byte
// aligned. However, spike is not checking this, so we don't implement it.
Reg#(Data) mepc_csr <- mkCsrReg(0);
// mcause
Reg#(Bit#(1)) mcause_interrupt_reg <- mkCsrReg(0);
Reg#(Bit#(4)) mcause_code_reg <- mkCsrReg(0);
Reg#(Data) mcause_csr = concatReg3(
mcause_interrupt_reg, readOnlyReg(59'b0), mcause_code_reg
);
// mtval (mbadaddr in spike)
Reg#(Data) mtval_csr <- mkCsrReg(0);
// mip
Vector#(4, Reg#(Bit#(1))) external_int_pend_vec = replicate(readOnlyReg(0));
external_int_pend_vec[prvU] <- mkCsrReg(0);
external_int_pend_vec[prvS] <- mkCsrReg(0);
external_int_pend_vec[prvM] <- mkCsrReg(0);
Vector#(4, Reg#(Bit#(1))) timer_int_pend_vec = replicate(readOnlyReg(0));
timer_int_pend_vec[prvU] <- mkCsrReg(0);
timer_int_pend_vec[prvS] <- mkCsrReg(0);
timer_int_pend_vec[prvM] <- mkCsrReg(0);
Vector#(4, Reg#(Bit#(1))) software_int_pend_vec = replicate(readOnlyReg(0));
software_int_pend_vec[prvU] <- mkCsrReg(0);
software_int_pend_vec[prvS] <- mkCsrReg(0);
software_int_pend_vec[prvM] <- mkCsrReg(0);
Reg#(Data) mip_csr = concatReg13(
readOnlyReg(52'b0),
external_int_pend_vec[prvM], readOnlyReg(1'b0),
external_int_pend_vec[prvS], external_int_pend_vec[prvU],
readOnlyReg(timer_int_pend_vec[prvM]), // MTIP is read-only to software
readOnlyReg(1'b0),
timer_int_pend_vec[prvS], timer_int_pend_vec[prvU],
software_int_pend_vec[prvM], readOnlyReg(1'b0),
software_int_pend_vec[prvS], software_int_pend_vec[prvU]
);
// minstret
Ehr#(2, Data) minstret_ehr <- mkCsrEhr(0);
Reg#(Data) minstret_csr = minstret_ehr[0];
// mcycle
Ehr#(2, Data) mcycle_ehr <- mkCsrEhr(0);
Reg#(Data) mcycle_csr = mcycle_ehr[0];
// mvendorid
Reg#(Data) mvendorid_csr = readOnlyReg(0);
// marchid
Reg#(Data) marchid_csr = readOnlyReg(0);
// mimpid
Reg#(Data) mimpid_csr = readOnlyReg(0);
// mhartid
Reg#(Data) mhartid_csr = readOnlyReg(hartid);
// Supervisor level CSRs
// sstatus: restricted view of mstatus
Reg#(Data) sstatus_csr = concatReg17(
sd_reg, readOnlyReg(29'b0), uxl_reg, readOnlyReg(12'b0),
mxr_reg, sum_reg, readOnlyReg(1'b0), xs_reg, fs_reg,
readOnlyReg(4'b0), spp_reg,
readOnlyReg(2'b0), prev_ie_vec[prvS], prev_ie_vec[prvU],
readOnlyReg(2'b0), ie_vec[prvS], ie_vec[prvU]
);
// sie: restricted view of mie
Reg#(Data) sie_csr = concatReg9(
readOnlyReg(54'b0),
external_int_en_vec[prvS], external_int_en_vec[prvU],
readOnlyReg(2'b0),
timer_int_en_vec[prvS], timer_int_en_vec[prvU],
readOnlyReg(2'b0),
software_int_en_vec[prvS], software_int_en_vec[prvU]
);
// stvec
Reg#(Bit#(62)) stvec_base_hi_reg <- mkCsrReg(0); // BASE[63:2]
Reg#(Bit#(1)) stvec_mode_low_reg <- mkCsrReg(0); // MODE[0]
Reg#(Data) stvec_csr = concatReg3(
stvec_base_hi_reg, readOnlyReg(1'b0), stvec_mode_low_reg
);
// scounteren
Reg#(Bit#(1)) scounteren_ir_reg <- mkCsrReg(0);
Reg#(Bit#(1)) scounteren_tm_reg <- mkCsrReg(0);
Reg#(Bit#(1)) scounteren_cy_reg <- mkCsrReg(0);
Reg#(Data) scounteren_csr = concatReg5(
readOnlyReg(32'b0),
readOnlyReg(29'b0), // hpmcounter 3-31 not accessible in U mode
scounteren_ir_reg, scounteren_tm_reg, scounteren_cy_reg
);
// sscratch
Reg#(Data) sscratch_csr <- mkCsrReg(0);
// sepc: FIXME Since we don't have C extension, sepc should be 4-byte
// aligned. However, spike is not checking this, so we don't implement it.
Reg#(Data) sepc_csr <- mkCsrReg(0);
// scause
Reg#(Bit#(1)) scause_interrupt_reg <- mkCsrReg(0);
Reg#(Bit#(4)) scause_code_reg <- mkCsrReg(0);
Reg#(Data) scause_csr = concatReg3(
scause_interrupt_reg, readOnlyReg(59'b0), scause_code_reg
);
// stval (sbadaddr in spike)
Reg#(Data) stval_csr <- mkCsrReg(0);
// sip: restricted view of mip
Reg#(Data) sip_csr = concatReg9(
readOnlyReg(54'b0),
external_int_pend_vec[prvS], external_int_pend_vec[prvU],
readOnlyReg(2'b0),
timer_int_pend_vec[prvS], timer_int_pend_vec[prvU],
readOnlyReg(2'b0),
software_int_pend_vec[prvS], software_int_pend_vec[prvU]
);
// satp (sptbr in spike): FIXME we only support Bare and Sv39, so we hack
// the encoding of mode[3:0] field. Only mode[3] is relevant, other bits
// are always 0
Reg#(Bit#(1)) vm_mode_sv39_reg <- mkCsrReg(0);
Reg#(Bit#(4)) vm_mode_reg = concatReg2(vm_mode_sv39_reg, readOnlyReg(3'b0));
Reg#(Asid) asid_reg <- mkCsrReg(0);
Reg#(Bit#(16)) full_asid_reg = zeroExtendReg(asid_reg);
Reg#(Bit#(44)) ppn_reg <- mkCsrReg(0);
Reg#(Data) satp_csr = concatReg3(vm_mode_reg, full_asid_reg, ppn_reg);
// User level CSRs
// According to spike, any write to fflags/frm/fcsr will set fs_reg as
// dirty, regardless of whether the write truly changes value or not.
// Besides, any non-zero FP exception flags will also make fs_reg dirty.
// fflags: if we directly change fflags_reg (instead of fflags_csr), then
// we must set fs_reg manually
Reg#(Bit#(5)) fflags_reg <- mkCsrReg(0);
Reg#(Data) fflags_csr = addWriteSideEffect(
zeroExtendReg(fflags_reg), fs_reg._write(2'b11)
);
// frm: if we directly change frm_reg (instead of frm_csr), then we must
// set fs_reg manually
Reg#(Bit#(3)) frm_reg <- mkCsrReg(0);
Reg#(Data) frm_csr = addWriteSideEffect(
zeroExtendReg(frm_reg), fs_reg._write(2'b11)
);
// fcsr
Reg#(Data) fcsr_csr = addWriteSideEffect(
zeroExtendReg(concatReg2(frm_reg, fflags_reg)), fs_reg._write(2'b11)
);
// cycle
Reg#(Data) cycle_csr = readOnlyReg(mcycle_csr);
// time
Reg#(Data) time_reg <- mkCsrReg(0);
Reg#(Data) time_csr = readOnlyReg(time_reg);
// instret
Reg#(Data) instret_csr = readOnlyReg(minstret_csr);
// terminate (non-standard)
Terminate terminate_module <- mkTerminate;
Reg#(Data) terminate_csr = terminate_module.reg_ifc;
// whether performance stats is collected
StatsCsr stats_module <- mkStatsCsr;
Reg#(Data) stats_csr = stats_module.reg_ifc;
`ifdef SECURITY
// sanctum machine CSRs
// ### Enclave virtual base and mask
// (per-core) registers
// ( defines a virtual region for which enclave page tables are used in
// place of OS-controlled page tables)
// (machine-mode non-standard read/write)
Reg#(Data) mevbase_csr <- mkCsrReg(maxBound); // impossible base & mask,
Reg#(Data) mevmask_csr <- mkCsrReg(0); // so no enclave accesses are possible
// ### Enclave page table base
// (per core) register
// ( pointer to a separate page table data structure used to translate enclave
// virtual addresses)
// (machine-mode non-standard read/write)
Reg#(Bit#(44)) eppn_reg <- mkCsrReg(0);
Reg#(Data) meatp_csr = zeroExtendReg(eppn_reg);
// ### DRAM bitmap
// (per core) registers (OS and Enclave)
// ( white-lists the DRAM regions the core is allowed to access via OS and
// enclave virtual addresses)
// (machine-mode non-standard read/write)
Reg#(Data) mmrbm_csr <- mkCsrReg(maxBound);
Reg#(Data) memrbm_csr <- mkCsrReg(0);
// ### Protected region base and mask
// (per core) registers (OS and Enclave)
// ( these are used to prevent address translation into a specific range of
// physical addresses, for example to protect the security monitor from all software)
// (machine-mode non-standard read/write)
Reg#(Data) mparbase_csr <- mkCsrReg(maxBound);
Reg#(Data) mparmask_csr <- mkCsrReg(0);
Reg#(Data) meparbase_csr <- mkCsrReg(0);
Reg#(Data) meparmask_csr <- mkCsrReg(0);
// ### Turn on/off speculation
Reg#(Bit#(2)) mspec_reg <- mkCsrReg(mSpecAll);
Reg#(Data) mspec_csr = zeroExtendReg(mspec_reg);
// sanctum user CSR
// ### true random number
// For now, we skip secure boot, keep TRNG = 0
Reg#(Data) trng_csr <- mkReadOnlyReg(0); //mkTRNG;
`endif
rule incCycle;
mcycle_ehr[1] <= mcycle_ehr[1] + 1;
endrule
// Function for getting a csr given an index
function Reg#(Data) get_csr(CSR csr);
return (case (csr)
// User CSRs
CSRfflags: fflags_csr;
CSRfrm: frm_csr;
CSRfcsr: fcsr_csr;
CSRcycle: cycle_csr;
CSRtime: time_csr;
CSRinstret: instret_csr;
CSRterminate: terminate_csr;
CSRstats: stats_csr;
// Supervisor CSRs
CSRsstatus: sstatus_csr;
CSRsie: sie_csr;
CSRstvec: stvec_csr;
CSRscounteren: scounteren_csr;
CSRsscratch: sscratch_csr;
CSRsepc: sepc_csr;
CSRscause: scause_csr;
CSRstval: stval_csr;
CSRsip: sip_csr;
CSRsatp: satp_csr;
// Machine CSRs
CSRmstatus: mstatus_csr;
CSRmisa: misa_csr;
CSRmedeleg: medeleg_csr;
CSRmideleg: mideleg_csr;
CSRmie: mie_csr;
CSRmtvec: mtvec_csr;
CSRmcounteren: mcounteren_csr;
CSRmscratch: mscratch_csr;
CSRmepc: mepc_csr;
CSRmcause: mcause_csr;
CSRmtval: mtval_csr;
CSRmip: mip_csr;
CSRmcycle: mcycle_csr;
CSRminstret: minstret_csr;
CSRmvendorid: mvendorid_csr;
CSRmarchid: marchid_csr;
CSRmimpid: mimpid_csr;
CSRmhartid: mhartid_csr;
`ifdef SECURITY
CSRmevbase: mevbase_csr;
CSRmevmask: mevmask_csr;
CSRmeatp: meatp_csr;
CSRmmrbm: mmrbm_csr;
CSRmemrbm: memrbm_csr;
CSRmparbase: mparbase_csr;
CSRmparmask: mparmask_csr;
CSRmeparbase: meparbase_csr;
CSRmeparmask: meparmask_csr;
CSRmspec: mspec_csr;
CSRtrng: trng_csr;
`endif
default: readOnlyReg(64'b0);
endcase);
endfunction
method Data rd(CSR csr);
return get_csr(csr)._read;
endmethod
method Action csrInstWr(CSR csr, Data x);
get_csr(csr)._write(x);
endmethod
method Bool fpuInstNeedWr(Bit#(5) fflags, Bool fpu_dirty);
Bool fflags_change = (fflags & fflags_reg) != fflags;
// we need to set fs_reg as dirty in two cases
// 1. FP reg is written (i.e., fpu_dirty)
// 2. FP exception (i.e., fflags) is non-zero (try to match spike)
Bool need_set_dirty = fs_reg != 2'b11 && (fpu_dirty || fflags != 0);
return fflags_change || need_set_dirty;
endmethod
method Action fpuInstWr(Bit#(5) fflags);
fs_reg <= 2'b11; // FPU must be dirty
fflags_reg <= fflags_reg | fflags;
endmethod
method Maybe#(Interrupt) pending_interrupt;
// first get all the pending interrupts
Bit#(InterruptNum) pend_ints = truncate(mie_csr & mip_csr);
// now find out all the truly enabled interrupts (that needs handling)
Bit#(InterruptNum) enabled_ints = 0;
// check interrupts that needs to be handled at M mode: all interrupts
// are by default handled at M mode unless it is delegated in
// mideleg_csr, we just need to ignore those interrupts
if(prv_reg < prvM || (prv_reg == prvM && ie_vec[prvM] == 1)) begin
enabled_ints = pend_ints & ~truncate(mideleg_csr);
end
// check interrupts that needs to be handled at S mode only if no
// interrupt needs to be handled at M mode: interrupts handled at S
// mode must be delegated in mideleg_csr
if (enabled_ints == 0 &&
(prv_reg < prvS || (prv_reg == prvS && ie_vec[prvS] == 1))) begin
enabled_ints = pend_ints & truncate(mideleg_csr);
end
// According to spike, return the interrupt bit at LSB
function Bool isEnabled(Integer i) = (enabled_ints[i] == 1);
Vector#(InterruptNum, Integer) idxVec = genVector;
if(find(isEnabled, idxVec) matches tagged Valid .i) begin
return Valid (unpack(fromInteger(i)));
end
else begin
return Invalid;
end
endmethod
method ActionValue#(Addr) trap(Trap t, Addr pc, Addr addr);
// figure out trap cause & trap val
Bit#(1) cause_interrupt = 0;
Bit#(4) cause_code = 0;
Data trap_val = 0;
case(t) matches
tagged Exception .e: begin
cause_code = pack(e);
trap_val = (case(e)
InstAddrMisaligned, InstAccessFault,
Breakpoint, InstPageFault: return pc;
LoadAddrMisaligned, LoadAccessFault,
StoreAddrMisaligned, StoreAccessFault,
LoadPageFault, StorePageFault: return addr;
default: return 0;
endcase);
end
tagged Interrupt .i: begin
cause_code = pack(i);
cause_interrupt = 1;
end
endcase
// function to figure out next PC
function Addr getNextPc(Bit#(1) mode_low, Bit#(62) base_hi);
Addr base = {base_hi, 2'b0};
if(mode_low == 1 && cause_interrupt == 1) begin
// vector jump: only for interrupt
return base + zeroExtend({cause_code, 2'b0});
end
else begin // direct jump
return base;
end
endfunction
// check if trap is delegated
Bool deleg = prv_reg <= prvS && (case(t) matches
tagged Exception .e: return medeleg_csr[pack(e)] == 1;
tagged Interrupt .i: return mideleg_csr[pack(i)] == 1;
default: return False;
endcase);
// handle the trap
if(deleg) begin // handle in S mode
// ie/prv stack
prev_prv_vec[prvS] <= prv_reg;
prv_reg <= prvS;
prev_ie_vec[prvS] <= ie_vec[prvS];
ie_vec[prvS] <= 0;
// record trap info
sepc_csr <= pc;
scause_interrupt_reg <= cause_interrupt;
scause_code_reg <= cause_code;
stval_csr <= trap_val;
// return next pc
return getNextPc(stvec_mode_low_reg, stvec_base_hi_reg);
end
else begin
// ie/prv stack
prev_prv_vec[prvM] <= prv_reg;
prv_reg <= prvM;
prev_ie_vec[prvM] <= ie_vec[prvM];
ie_vec[prvM] <= 0;
// record trap info
mepc_csr <= pc;
mcause_interrupt_reg <= cause_interrupt;
mcause_code_reg <= cause_code;
mtval_csr <= trap_val;
// return next pc
return getNextPc(mtvec_mode_low_reg, mtvec_base_hi_reg);
end
// XXX yield load reservation should be done outside this method
endmethod
method ActionValue#(Addr) mret;
prv_reg <= prev_prv_vec[prvM];
prev_prv_vec[prvM] <= prvU;
ie_vec[prvM] <= prev_ie_vec[prvM];
prev_ie_vec[prvM] <= 1;
return mepc_csr;
endmethod
method ActionValue#(Addr) sret;
prv_reg <= prev_prv_vec[prvS];
prev_prv_vec[prvS] <= prvU;
ie_vec[prvS] <= prev_ie_vec[prvS];
prev_ie_vec[prvS] <= 1;
return sepc_csr;
endmethod
method VMInfo vmI;
// for inst fetch, NO need to consider MPRV
Bit#(2) prv = prv_reg;
return VMInfo {
prv: prv,
asid: asid_reg,
sv39: prv < prvM && vm_mode_sv39_reg == 1,
exeReadable: mxr_reg == 1,
userAccessibleByS: sum_reg == 1,
basePPN: ppn_reg
`ifdef SECURITY
, sanctum_evbase: mevbase_csr,
sanctum_evmask: mevmask_csr,
sanctum_ebasePPN: eppn_reg,
sanctum_mrbm: mmrbm_csr,
sanctum_emrbm: memrbm_csr,
sanctum_parbase: mparbase_csr,
sanctum_parmask: mparmask_csr,
sanctum_eparbase: meparbase_csr,
sanctum_eparmask: meparmask_csr,
// enclave / security monitor should never execute instructions
// from untrusted shared region
sanctum_authShared: False
`endif
};
endmethod
method VMInfo vmD;
// for load/store, need to consider MPRV
Bit#(2) prv = (mprv_reg == 1) ? prev_prv_vec[prvM] : prv_reg;
return VMInfo {
prv: prv,
asid: asid_reg,
sv39: prv < prvM && vm_mode_sv39_reg == 1,
exeReadable: mxr_reg == 1,
userAccessibleByS: sum_reg == 1,
basePPN: ppn_reg
`ifdef SECURITY
, sanctum_evbase: mevbase_csr,
sanctum_evmask: mevmask_csr,
sanctum_ebasePPN: eppn_reg,
sanctum_mrbm: mmrbm_csr,
sanctum_emrbm: memrbm_csr,
sanctum_parbase: mparbase_csr,
sanctum_parmask: mparmask_csr,
sanctum_eparbase: meparbase_csr,
sanctum_eparmask: meparmask_csr,
// enclave / security monitor can read/write untrusted shared
// region when speculation is off (either by mspec CSR or in M
// mode)
// XXX Because of the effects of mprv, we have to use prv_reg here
// instead of prv. Otherwise, we may be in M mode, but prv=S, and
// still forbid shared accesses
sanctum_authShared: mspec_reg != mSpecAll || prv_reg == prvM
`endif
};
endmethod
method CsrDecodeInfo decodeInfo = CsrDecodeInfo {
frm: frm_reg,
fEnabled: fs_reg != 0,
prv: prv_reg,
trapVM: tvm_reg == 1,
timeoutWait: tw_reg == 1,
trapSret: tsr_reg == 1,
cycleReadableByS: mcounteren_cy_reg == 1,
cycleReadableByU: mcounteren_cy_reg == 1 && scounteren_cy_reg == 1,
instretReadableByS: mcounteren_ir_reg == 1,
instretReadableByU: mcounteren_ir_reg == 1 && scounteren_ir_reg == 1,
timeReadableByS: mcounteren_tm_reg == 1,
timeReadableByU: mcounteren_tm_reg == 1 && scounteren_tm_reg == 1
};
method Action incInstret(SupCnt x);
minstret_ehr[1] <= minstret_ehr[1] + zeroExtend(x);
endmethod
method Action setTime(Data t);
time_reg <= t;
endmethod
method getMSIP = software_int_pend_vec[prvM]._read;
method setMSIP = software_int_pend_vec[prvM]._write;
method setMTIP = timer_int_pend_vec[prvM]._write;
// Bluespec: external interrupts targeting machine and supervisor modes
method Action setMEIP (Bit #(1) v);
external_int_pend_vec[prvM] <= v;
endmethod
method Action setSEIP (Bit #(1) v);
external_int_pend_vec[prvS] <= v;
endmethod
method terminate = terminate_module.terminate;
// performance stats
method doPerfStats = stats_module.doPerfStats;
method sendDoStats = stats_module.sendDoStats;
method recvDoStats = stats_module.recvDoStats;
endmodule

View File

@@ -0,0 +1,277 @@
package LLC_AXI4_Adapter;
// ================================================================
// BSV lib imports
import ConfigReg :: *;
import Assert :: *;
import FIFOF :: *;
import Vector :: *;
// ----------------
// BSV additional libs
import GetPut_Aux :: *;
import Cur_Cycle :: *;
import Semi_FIFOF :: *;
import CreditCounter :: *;
// ================================================================
// Project imports
// ----------------
// From MIT RISCY-OOO
import Types :: *;
import CacheUtils :: *;
import CCTypes :: *;
// ----------------
// From Bluespec Pipes
import AXI4_Types :: *;
import Fabric_Defs :: *;
// ================================================================
interface LLC_AXI4_Adapter_IFC;
method Action reset;
// Fabric master interface for memory
interface AXI4_Master_IFC #(Wd_Id, Wd_Addr, Wd_Data, Wd_User) mem_master;
endinterface
// ================================================================
module mkLLC_AXi4_Adapter #(MemFifoClient #(idT, childT) llc)
(LLC_AXI4_Adapter_IFC)
provisos(Bits#(idT, a__),
Bits#(childT, b__),
FShow#(ToMemMsg#(idT, childT)),
FShow#(MemRsMsg#(idT, childT)),
Add#(SizeOf#(Line), 0, 512)); // assert Line sz = 512
// Verbosity: 0: quiet; 1: LLC transactions; 2: loop detail
Integer verbosity = 2;
Reg #(Bit #(4)) cfg_verbosity <- mkConfigReg (fromInteger (verbosity));
// ================================================================
// Fabric request/response
AXI4_Master_Xactor_IFC #(Wd_Id, Wd_Addr, Wd_Data, Wd_User) master_xactor <- mkAXI4_Master_Xactor_2;
// For discarding write-responses
CreditCounter_IFC #(4) ctr_wr_rsps_pending <- mkCreditCounter; // Max 15 writes outstanding
// ================================================================
// Functions to interact with the fabric
// Send a read-request into the fabric
function Action fa_fabric_send_read_req (Fabric_Addr addr);
action
AXI4_Size size = axsize_8;
let mem_req_rd_addr = AXI4_Rd_Addr {arid: fabric_default_id,
araddr: addr,
arlen: 0, // burst len = arlen+1
arsize: size,
arburst: fabric_default_burst,
arlock: fabric_default_lock,
arcache: fabric_default_arcache,
arprot: fabric_default_prot,
arqos: fabric_default_qos,
arregion: fabric_default_region,
aruser: fabric_default_user};
master_xactor.i_rd_addr.enq (mem_req_rd_addr);
// Debugging
if (cfg_verbosity > 1) begin
$display (" ", fshow (mem_req_rd_addr));
end
endaction
endfunction
// Send a write-request into the fabric
function Action fa_fabric_send_write_req (Fabric_Addr addr, Fabric_Strb strb, Bit #(64) st_val);
action
AXI4_Size size = axsize_8;
let mem_req_wr_addr = AXI4_Wr_Addr {awid: fabric_default_id,
awaddr: addr,
awlen: 0, // burst len = awlen+1
awsize: size,
awburst: fabric_default_burst,
awlock: fabric_default_lock,
awcache: fabric_default_awcache,
awprot: fabric_default_prot,
awqos: fabric_default_qos,
awregion: fabric_default_region,
awuser: fabric_default_user};
let mem_req_wr_data = AXI4_Wr_Data {wid: fabric_default_id,
wdata: st_val,
wstrb: strb,
wlast: True,
wuser: fabric_default_user};
master_xactor.i_wr_addr.enq (mem_req_wr_addr);
master_xactor.i_wr_data.enq (mem_req_wr_data);
// Expect a fabric response
ctr_wr_rsps_pending.incr;
// Debugging
if (cfg_verbosity > 1) begin
$display (" To fabric: ", fshow (mem_req_wr_addr));
$display (" ", fshow (mem_req_wr_data));
end
endaction
endfunction
// ================================================================
// Handle read requests and responses
// Don't do reads while writes are outstanding.
// Each 512b cache line takes 8 beats, each handling 64 bits
Reg #(Bit #(3)) rg_rd_req_beat <- mkReg (0);
Reg #(Bit #(3)) rg_rd_rsp_beat <- mkReg (0);
FIFOF #(LdMemRq #(idT, childT)) f_pending_reads <- mkFIFOF;
Reg #(Bit #(512)) rg_cline <- mkRegU;
rule rl_handle_read_req (llc.toM.first matches tagged Ld .ld
&&& (ctr_wr_rsps_pending.value == 0));
if ((cfg_verbosity > 0) && (rg_rd_req_beat == 0)) begin
$display ("%0d: LLC_AXI4_Adapter.rl_handle_read_req: Ld request from LLC to memory: beat %0d",
cur_cycle, rg_rd_req_beat);
$display (" ", fshow (ld));
end
Addr line_addr = { ld.addr [63:6], 6'h0 }; // Addr of containing cache line
Addr offset = zeroExtend ( { rg_rd_req_beat, 3'b_000 } ); // Addr offset of 64b word for this beat
fa_fabric_send_read_req (line_addr | offset);
if (rg_rd_req_beat == 0)
f_pending_reads.enq (ld);
if (rg_rd_req_beat == 7)
llc.toM.deq;
rg_rd_req_beat <= rg_rd_req_beat + 1;
endrule
rule rl_handle_read_rsps;
let mem_rsp <- pop_o (master_xactor.o_rd_data);
if (cfg_verbosity > 1) begin
$display ("%0d: LLC_AXI4_Adapter.rl_handle_read_rsps: beat %0d ", cur_cycle, rg_rd_rsp_beat);
$display (" ", fshow (mem_rsp));
end
if (mem_rsp.rresp != axi4_resp_okay) begin
// TODO: need to raise a non-maskable interrupt (NMI) here
$display ("%0d: LLC_AXI4_Adapter.rl_handle_read_rsp: fabric response error; exit", cur_cycle);
$display (" ", fshow (mem_rsp));
$finish (1);
end
// Shift next 64 bits from fabric into the cache line being assembled
let new_cline = { mem_rsp.rdata, rg_cline [511:64] };
if (rg_rd_rsp_beat == 7) begin
let ldreq <- pop (f_pending_reads);
MemRsMsg #(idT, childT) resp = MemRsMsg {data: unpack (new_cline),
child: ldreq.child,
id: ldreq.id};
llc.rsFromM.enq (resp);
if (cfg_verbosity > 1)
$display (" Response to LLC: ", fshow (resp));
end
rg_cline <= new_cline;
rg_rd_rsp_beat <= rg_rd_rsp_beat + 1;
endrule
// ================================================================
// Handle write requests and responses
// Each 512b cache line takes 8 beats, each handling 64 bits
Reg #(Bit #(3)) rg_wr_req_beat <- mkReg (0);
Reg #(Bit #(3)) rg_wr_rsp_beat <- mkReg (0);
FIFOF #(WbMemRs) f_pending_writes <- mkFIFOF;
rule rl_handle_write_req (llc.toM.first matches tagged Wb .wb);
if ((cfg_verbosity > 0) && (rg_wr_req_beat == 0)) begin
$display ("%d: LLC_AXI4_Adapter.rl_handle_write_req: Wb request from LLC to memory:", cur_cycle);
$display (" ", fshow (wb));
end
Addr line_addr = { wb.addr [63:6], 6'h0 }; // Addr of containing cache line
Line line_data = wb.data;
Vector #(8, Bit #(8)) line_bes = unpack (pack (wb.byteEn));
Addr offset = zeroExtend ( { rg_wr_req_beat, 3'b_000 } ); // Addr offset of 64b word for this beat
Bit #(64) data64 = line_data [rg_wr_req_beat];
Bit #(8) strb8 = line_bes [rg_wr_req_beat];
fa_fabric_send_write_req (line_addr | offset, strb8, data64);
if (rg_wr_req_beat == 0)
f_pending_writes.enq (wb);
if (rg_wr_req_beat == 7)
llc.toM.deq;
rg_wr_req_beat <= rg_wr_req_beat + 1;
endrule
// ----------------
// Discard write-responses from the fabric
rule rl_discard_write_rsp;
let wr_resp <- pop_o (master_xactor.o_wr_resp);
if (cfg_verbosity > 1) begin
$display ("%0d: LLC_AXI4_Adapter.rl_discard_write_rsp: beat %0d ", cur_cycle, rg_wr_rsp_beat);
$display (" ", fshow (wr_resp));
end
if (ctr_wr_rsps_pending.value == 0) begin
$display ("%0d: ERROR: LLC_AXI4_Adapter.rl_discard_write_rsp: unexpected Wr response (ctr_wr_rsps_pending.value == 0)",
cur_cycle);
$display (" ", fshow (wr_resp));
$finish (1); // Assertion failure
end
ctr_wr_rsps_pending.decr;
if (wr_resp.bresp != axi4_resp_okay) begin
// TODO: need to raise a non-maskable interrupt (NMI) here
$display ("%0d: LLC_AXI4_Adapter.rl_discard_write_rsp: fabric response error: exit", cur_cycle);
$display (" ", fshow (wr_resp));
$finish (1);
end
if (rg_wr_rsp_beat == 7) begin
let wrreq <- pop (f_pending_writes);
// LLC does not expect any response for writes
end
rg_wr_rsp_beat <= rg_wr_rsp_beat + 1;
endrule
// ================================================================
// INTERFACE
method Action reset;
ctr_wr_rsps_pending.clear;
endmethod
// Fabric interface for memory
interface mem_master = master_xactor.axi_side;
endmodule
// ================================================================
endpackage

File diff suppressed because it is too large Load Diff

View File

@@ -0,0 +1,253 @@
package MMIO_AXI4_Adapter;
// ================================================================
// BSV lib imports
import Assert :: *;
import ConfigReg :: *;
import FIFOF :: *;
import GetPut :: *;
import ClientServer :: *;
// ----------------
// BSV additional libs
import GetPut_Aux :: *;
import Cur_Cycle :: *;
import Semi_FIFOF :: *;
import CreditCounter :: *;
// ================================================================
// Project imports
// ----------------
// From MIT RISCY-OOO
import ProcTypes :: *;
// ----------------
// From Bluespec Pipes
import AXI4_Types :: *;
import Fabric_Defs :: *;
// ================================================================
interface MMIO_AXI4_Adapter_IFC;
method Action reset;
interface Server #(MMIOCRq, MMIODataPRs) core_side;
// Fabric master interface for IO
interface AXI4_Master_IFC #(Wd_Id, Wd_Addr, Wd_Data, Wd_User) mmio_master;
endinterface
// ================================================================
module mkMMIO_AXI4_Adapter (MMIO_AXI4_Adapter_IFC);
// Verbosity: 0: quiet; 1: transactions
Integer verbosity = 2;
Reg #(Bit #(4)) cfg_verbosity <- mkConfigReg (fromInteger (verbosity));
// ================================================================
// Requests from and responses to core
FIFOF #(MMIOCRq) f_reqs_from_core <- mkFIFOF;
FIFOF #(MMIODataPRs) f_rsps_to_core <- mkFIFOF;
// ================================================================
// Fabric request/response
AXI4_Master_Xactor_IFC #(Wd_Id, Wd_Addr, Wd_Data, Wd_User) master_xactor <- mkAXI4_Master_Xactor_2;
// For discarding write-responses
CreditCounter_IFC #(4) ctr_wr_rsps_pending <- mkCreditCounter; // Max 15 writes outstanding
// ================================================================
// Functions to interact with the fabric
// Send a read-request into the fabric
function Action fa_fabric_send_read_req (Fabric_Addr addr);
action
AXI4_Size size = axsize_8;
let mem_req_rd_addr = AXI4_Rd_Addr {arid: fabric_default_id,
araddr: addr,
arlen: 0, // burst len = arlen+1
arsize: size,
arburst: fabric_default_burst,
arlock: fabric_default_lock,
arcache: fabric_default_arcache,
arprot: fabric_default_prot,
arqos: fabric_default_qos,
arregion: fabric_default_region,
aruser: fabric_default_user};
master_xactor.i_rd_addr.enq (mem_req_rd_addr);
// Debugging
if (cfg_verbosity > 0) begin
$display (" ", fshow (mem_req_rd_addr));
end
endaction
endfunction
// Send a write-request into the fabric
function Action fa_fabric_send_write_req (Fabric_Addr addr, Fabric_Strb strb, Bit #(64) st_val);
action
AXI4_Size size = axsize_8;
let mem_req_wr_addr = AXI4_Wr_Addr {awid: fabric_default_id,
awaddr: addr,
awlen: 0, // burst len = awlen+1
awsize: size,
awburst: fabric_default_burst,
awlock: fabric_default_lock,
awcache: fabric_default_awcache,
awprot: fabric_default_prot,
awqos: fabric_default_qos,
awregion: fabric_default_region,
awuser: fabric_default_user};
let mem_req_wr_data = AXI4_Wr_Data {wid: fabric_default_id,
wdata: st_val,
wstrb: strb,
wlast: True,
wuser: fabric_default_user};
master_xactor.i_wr_addr.enq (mem_req_wr_addr);
master_xactor.i_wr_data.enq (mem_req_wr_data);
// Expect a fabric response
ctr_wr_rsps_pending.incr;
// Debugging
if (cfg_verbosity > 0) begin
$display (" To fabric: ", fshow (mem_req_wr_addr));
$display (" ", fshow (mem_req_wr_data));
end
endaction
endfunction
// ================================================================
// Handle read requests and responses.
// Don't do a read while a write is outstanding.
// This is just an adapter from MMIOCRq/MMIODataPRs to AXI4
rule rl_handle_read_req (f_reqs_from_core.first.func matches Ld
&&& (ctr_wr_rsps_pending.value == 0));
let req <- pop (f_reqs_from_core);
if (cfg_verbosity > 0) begin
$display ("%0d: MMIO_AXI4_Adapter.rl_handle_read_req: Ld request", cur_cycle);
$display (" ", fshow (req));
end
fa_fabric_send_read_req (req.addr);
endrule
// ----------------
rule rl_handle_read_rsps;
let mem_rsp <- pop_o (master_xactor.o_rd_data);
if (cfg_verbosity > 0) begin
$display ("%0d: MMIO_AXI4_Adapter.rl_handle_read_rsps ", cur_cycle);
$display (" ", fshow (mem_rsp));
end
if (mem_rsp.rresp != axi4_resp_okay) begin
// TODO: need to raise a non-maskable interrupt (NMI) here
$display ("%0d: MMIO_AXI4_Adapter.rl_handle_read_rsp: fabric response error; exit", cur_cycle);
$display (" ", fshow (mem_rsp));
$finish (1);
end
let rsp = MMIODataPRs {valid: True, data: mem_rsp.rdata};
f_rsps_to_core.enq (rsp);
if (cfg_verbosity > 0)
$display (" Response MMIO to core: ", fshow (rsp));
endrule
// ================================================================
// Handle write requests and responses
rule rl_handle_write_req (f_reqs_from_core.first.func matches St);
let req <- pop (f_reqs_from_core);
if (cfg_verbosity > 0) begin
$display ("%d: MMIO_AXI4_Adapter.rl_handle_write_req: St request:", cur_cycle);
$display (" ", fshow (req));
end
fa_fabric_send_write_req (req.addr, pack (req.byteEn), req.data);
endrule
// ----------------
// Discard write-responses from the fabric
rule rl_discard_write_rsp;
let wr_resp <- pop_o (master_xactor.o_wr_resp);
if (cfg_verbosity > 0) begin
$display ("%0d: MMIO_AXI4_Adapter.rl_discard_write_rsp", cur_cycle);
$display (" ", fshow (wr_resp));
end
if (ctr_wr_rsps_pending.value == 0) begin
$display ("%0d: ERROR: MMIO_AXI4_Adapter.rl_discard_write_rsp: unexpected Wr response (ctr_wr_rsps_pending.value == 0)",
cur_cycle);
$display (" ", fshow (wr_resp));
$finish (1); // Assertion failure
end
ctr_wr_rsps_pending.decr;
if (wr_resp.bresp != axi4_resp_okay) begin
// TODO: need to raise a non-maskable interrupt (NMI) here
$display ("%0d: MMIO_AXI4_Adapter.rl_discard_write_rsp: fabric response error: exit", cur_cycle);
$display (" ", fshow (wr_resp));
$finish (1);
end
else begin
let rsp = MMIODataPRs {valid: True, data: 0};
f_rsps_to_core.enq (rsp);
end
endrule
// ================================================================
// This adapter should only receive Ld/St requests, no Inst or AMO reqs.
function Bool fn_is_Ld_or_St (MMIOCRq req);
return case (req.func) matches
Ld : True;
St : True;
default: False;
endcase;
endfunction
rule rl_handle_non_Ld_St (! fn_is_Ld_or_St (f_reqs_from_core.first));
let req <- pop (f_reqs_from_core);
$display ("%0d: ERROR: MMIO_AXI4_Adapter.rl_handle_non_Ld_St",
cur_cycle);
$display (" ", fshow (req));
$finish (1); // Assertion failure
endrule
// ================================================================
// INTERFACE
method Action reset;
ctr_wr_rsps_pending.clear;
endmethod
interface Server core_side = toGPServer (f_reqs_from_core, f_rsps_to_core);
// Fabric master interface for IO
interface mmio_master = master_xactor.axi_side;
endmodule
// ================================================================
endpackage

358
src_Core/CPU/Proc.bsv Normal file
View File

@@ -0,0 +1,358 @@
package Proc;
// Copyright (c) 2018 Massachusetts Institute of Technology
//
// Permission is hereby granted, free of charge, to any person
// obtaining a copy of this software and associated documentation
// files (the "Software"), to deal in the Software without
// restriction, including without limitation the rights to use, copy,
// modify, merge, publish, distribute, sublicense, and/or sell copies
// of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be
// included in all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
// MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
// BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
// ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
// CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
// SOFTWARE.
// Portions Copyright (c) 2019 Bluespec, Inc.
// ================================================================
// BSV lib imports
import Vector::*;
import GetPut::*;
import ClientServer::*;
import Connectable::*;
import FIFOF :: *;
import ConfigReg :: *;
// ----------------
// BSV additional libs
import GetPut_Aux :: *;
// ================================================================
// Project imports
// ----------------
// From MIT RISCY-OOO
import Types::*;
import ProcTypes::*;
import L1CoCache::*;
import L2Tlb::*;
import CCTypes::*;
import CacheUtils::*;
import LLCache::*;
import MemLoader::*;
import L1LLConnect::*;
import LLCDmaConnect::*;
import MMIOAddrs::*;
import MMIOCore::*;
import DramCommon::*;
import Performance::*;
// ----------------
// From McStriiv
import AXI4_Types :: *;
import Fabric_Defs :: *;
import Core :: *;
import Proc_IFC :: *;
import MMIOPlatform :: *;
import LLC_AXI4_Adapter :: *;
import MMIO_AXI4_Adapter :: *;
// ================================================================
(* synthesize *)
module mkProc (Proc_IFC);
// cores
Vector#(CoreNum, Core) core = ?;
for(Integer i = 0; i < valueof(CoreNum); i = i+1) begin
core[i] <- mkCore(fromInteger(i));
end
// ----------------
// Verbosity control for debugging
// Verbosity: 0=quiet; 1=instruction trace; 2=more detail
Reg #(Bit #(4)) cfg_verbosity <- mkConfigReg (0);
// ----------------
// Reset requests and responses (TODO: to be implemented)
FIFOF #(Bit #(0)) f_reset_reqs <- mkFIFOF;
FIFOF #(Bit #(0)) f_reset_rsps <- mkFIFOF;
// ----------------
// Communication to/from External debug module (TODO: to be implemented)
`ifdef INCLUDE_GDB_CONTROL
// Debugger run-control
FIFOF #(Bool) f_run_halt_reqs <- mkFIFOF;
FIFOF #(Bool) f_run_halt_rsps <- mkFIFOF;
// Stop-request from debugger (e.g., GDB ^C or Dsharp 'stop')
Reg #(Bool) rg_stop_req <- mkReg (False);
// Count instrs after step-request from debugger (via dcsr.step)
Reg #(Bit #(1)) rg_step_count <- mkReg (0);
// Debugger GPR read/write request/response
FIFOF #(MemoryRequest #(5, XLEN)) f_gpr_reqs <- mkFIFOF1;
FIFOF #(MemoryResponse #( XLEN)) f_gpr_rsps <- mkFIFOF1;
`ifdef ISA_F
// Debugger FPR read/write request/response
FIFOF #(MemoryRequest #(5, FLEN)) f_fpr_reqs <- mkFIFOF1;
FIFOF #(MemoryResponse #( FLEN)) f_fpr_rsps <- mkFIFOF1;
`endif
// Debugger CSR read/write request/response
FIFOF #(MemoryRequest #(12, XLEN)) f_csr_reqs <- mkFIFOF1;
FIFOF #(MemoryResponse #( XLEN)) f_csr_rsps <- mkFIFOF1;
`endif
// ----------------
// Tandem Verification (TODO: to be implemented)
`ifdef INCLUDE_TANDEM_VERIF
FIFOF #(Trace_Data) f_trace_data <- mkFIFOF;
`endif
// ----------------
// MMIO
MMIO_AXI4_Adapter_IFC mmio_axi4_adapter <- mkMMIO_AXI4_Adapter;
// MMIO platform
Vector#(CoreNum, MMIOCoreToPlatform) mmioToP;
for(Integer i = 0; i < valueof(CoreNum); i = i+1) begin
mmioToP[i] = core[i].mmioToPlatform;
end
MMIOPlatform mmioPlatform <- mkMMIOPlatform (mmioToP,
mmio_axi4_adapter.core_side);
// last level cache
LLCache llc <- mkLLCache;
// connect LLC to L1 caches
Vector#(L1Num, ChildCacheToParent#(L1Way, void)) l1 = ?;
for(Integer i = 0; i < valueof(CoreNum); i = i+1) begin
l1[i] = core[i].dCacheToParent;
l1[i + valueof(CoreNum)] = core[i].iCacheToParent;
end
mkL1LLConnect(llc.to_child, l1);
// ================================================================
// LLC's DMA connections
// Core's tlbToMem
Vector#(CoreNum, TlbMemClient) tlbToMem = ?;
for(Integer i = 0; i < valueof(CoreNum); i = i+1) begin
tlbToMem[i] = core[i].tlbToMem;
end
// Stub out memLoader (TODO: can be Debug Module's access)
let memLoaderStub = interface MemLoaderMemClient;
interface memReq = nullFifoDeq;
interface respSt = nullFifoEnq;
endinterface;
mkLLCDmaConnect(llc.dma, memLoaderStub, tlbToMem);
// ================================================================
// interface LLC to AXI4
LLC_AXI4_Adapter_IFC llc_axi4_adapter <- mkLLC_AXi4_Adapter (llc.to_mem);
// ================================================================
// Connect stats
for(Integer i = 0; i < valueof(CoreNum); i = i+1) begin
rule broadcastStats;
Bool doStats <- core[i].sendDoStats;
for(Integer j = 0; j < valueof(CoreNum); j = j+1) begin
core[j].recvDoStats(doStats);
end
llc.perf.setStatus(doStats);
endrule
end
// ================================================================
// Stub out deadlock and renameDebug interfaces
for(Integer j = 0; j < valueof(CoreNum); j = j+1) begin
rule rl_dummy1;
let x <- core[j].deadlock.dCacheCRqStuck.get;
endrule
rule rl_dummy2;
let x <- core[j].deadlock.dCachePRqStuck.get;
endrule
rule rl_dummy3;
let x <- core[j].deadlock.iCacheCRqStuck.get;
endrule
rule rl_dummy4;
let x <- core[j].deadlock.iCachePRqStuck.get;
endrule
rule rl_dummy5;
let x <- core[j].deadlock.renameInstStuck.get;
endrule
rule rl_dummy6;
let x <- core[j].deadlock.renameCorrectPathStuck.get;
endrule
rule rl_dummy7;
let x <- core[j].deadlock.commitInstStuck.get;
endrule
rule rl_dummy8;
let x <- core[j].deadlock.commitUserInstStuck.get;
endrule
rule rl_dummy9;
let x <- core[j].deadlock.checkStarted.get;
endrule
rule rl_dummy20;
let x <- core[j].renameDebug.renameErr.get;
endrule
end
// ================================================================
// Reset
rule rl_reset;
let x <- pop (f_reset_reqs);
llc_axi4_adapter.reset;
mmio_axi4_adapter.reset;
f_reset_rsps.enq (?);
endrule
// ----------------
// Termination detection
for(Integer i = 0; i < valueof(CoreNum); i = i+1) begin
rule rl_terminate;
let x <- core[i].coreIndInv.terminate;
$display ("Core %d terminated", i);
endrule
end
// Print out values written 'tohost'
rule rl_tohost;
let x <- mmioPlatform.to_host;
$display ("mmioPlatform.rl_tohost: 0x%0x (= %0d)", x, x);
if (x != 0) begin
// Standard RISC-V ISA tests finish by writing a value tohost with x[0]==1.
// Further when x[63:1]==0, all tests within the program pass,
// otherwise x[63:1] = the test within the program that failed.
let failed_testnum = (x >> 1);
if (failed_testnum == 0)
$display ("PASS");
else
$display ("FAIL %0d", failed_testnum);
$finish (0);
end
endrule
// ================================================================
// ================================================================
// ================================================================
// INTERFACE
// Reset
interface Server hart0_server_reset = toGPServer (f_reset_reqs, f_reset_rsps);
// ----------------
// Start the cores running
method Action start (Addr startpc, Addr tohostAddr, Addr fromhostAddr);
action
for(Integer i = 0; i < valueof(CoreNum); i = i+1)
core[i].coreReq.start (startpc, tohostAddr, fromhostAddr);
endaction
mmioPlatform.start (tohostAddr, fromhostAddr);
$display ("Proc.start: startpc = 0x%0h, tohostAddr = 0x%0h, fromhostAddr = %0h",
startpc, tohostAddr, fromhostAddr);
endmethod
// ----------------
// SoC fabric connections
// Fabric master interface for memory (from LLC)
interface master0 = llc_axi4_adapter.mem_master;
// Fabric master interface for IO (from MMIOPlatform)
interface master1 = mmio_axi4_adapter.mmio_master;
// ----------------
// External interrupts
method Action m_external_interrupt_req (x);
core[0].setMEIP (pack (x));
endmethod
method Action s_external_interrupt_req (x);
core[0].setSEIP (pack (x));
endmethod
// ----------------
// Non-maskable interrupt
// TODO: fixup: NMIs should send CPU to an NMI vector (TBD in SoC_Map)
method Action non_maskable_interrupt_req (Bool set_not_clear) = noAction;
// ----------------
// For tracing
method Action set_verbosity (Bit #(4) verbosity);
cfg_verbosity <= verbosity;
endmethod
// ----------------
// Optional interface to Tandem Verifier
`ifdef INCLUDE_TANDEM_VERIF
interface Get trace_data_out = toGet (f_trace_data);
`endif
// ----------------
// Optional interface to Debug Module
`ifdef INCLUDE_GDB_CONTROL
// run-control, other
interface Server hart0_server_run_halt = toGPServer (f_run_halt_reqs, f_run_halt_rsps);
interface Put hart0_put_other_req;
method Action put (Bit #(4) req);
cfg_verbosity <= req;
endmethod
endinterface
// GPR access
interface MemoryServer hart0_gpr_mem_server = toGPServer (f_gpr_reqs, f_gpr_rsps);
`ifdef ISA_F
// FPR access
interface MemoryServer hart0_fpr_mem_server = toGPServer (f_fpr_reqs, f_fpr_rsps);
`endif
// CSR access
interface MemoryServer hart0_csr_mem_server = toGPServer (f_csr_reqs, f_csr_rsps);
`endif
endmodule
endpackage

99
src_Core/CPU/Proc_IFC.bsv Normal file
View File

@@ -0,0 +1,99 @@
// Copyright (c) 2016-2019 Bluespec, Inc. All Rights Reserved
package Proc_IFC;
// ================================================================
// BSV library imports
import Memory :: *;
import GetPut :: *;
import ClientServer :: *;
// ================================================================
// Project imports
import ISA_Decls :: *;
`ifdef INCLUDE_TANDEM_VERIF
import TV_Info :: *;
`endif
import AXI4_Types :: *;
import Fabric_Defs :: *;
// ================================================================
// CPU interface
// Note: this Proc_IFC is similar, but not identical to CPU_IFC for Piccolo and Flute
// Specifically, it removes interfaces for software and timer,
// because the RISCY-OOO mkProc contains those elements.
interface Proc_IFC;
// Reset
interface Server #(Token, Token) hart0_server_reset;
// ----------------
// Start the cores running
method Action start (Addr startpc, Addr tohostAddr, Addr fromhostAddr);
// ----------------
// SoC fabric connections
// Fabric master interface for memory (from LLC)
interface AXI4_Master_IFC #(Wd_Id, Wd_Addr, Wd_Data, Wd_User) master0;
// Fabric master interface for IO (from MMIOPlatform)
interface AXI4_Master_IFC #(Wd_Id, Wd_Addr, Wd_Data, Wd_User) master1;
// ----------------
// External interrupts
(* always_ready, always_enabled *)
method Action m_external_interrupt_req (Bool set_not_clear);
(* always_ready, always_enabled *)
method Action s_external_interrupt_req (Bool set_not_clear);
// ----------------
// Non-maskable interrupt
(* always_ready, always_enabled *)
method Action non_maskable_interrupt_req (Bool set_not_clear);
// ----------------
// Set core's verbosity
method Action set_verbosity (Bit #(4) verbosity);
// ----------------
// Optional interface to Tandem Verifier
`ifdef INCLUDE_TANDEM_VERIF
interface Get #(Trace_Data) trace_data_out;
`endif
// ----------------
// Optional interface to Debug Module
`ifdef INCLUDE_GDB_CONTROL
// run-control, other
interface Server #(Bool, Bool) hart0_server_run_halt;
interface Put #(Bit #(4)) hart0_put_other_req;
// GPR access
interface MemoryServer #(5, XLEN) hart0_gpr_mem_server;
`ifdef ISA_F
// FPR access
interface MemoryServer #(5, FLEN) hart0_fpr_mem_server;
`endif
// CSR access
interface MemoryServer #(12, XLEN) hart0_csr_mem_server;
`endif
endinterface
// ================================================================
endpackage