1/*************************************************************************************** 2* Copyright (c) 2020-2021 Institute of Computing Technology, Chinese Academy of Sciences 3* Copyright (c) 2020-2021 Peng Cheng Laboratory 4* 5* XiangShan is licensed under Mulan PSL v2. 6* You can use this software according to the terms and conditions of the Mulan PSL v2. 7* You may obtain a copy of Mulan PSL v2 at: 8* http://license.coscl.org.cn/MulanPSL2 9* 10* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, 11* EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, 12* MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE. 13* 14* See the Mulan PSL v2 for more details. 15***************************************************************************************/ 16 17package xiangshan.mem 18 19import org.chipsalliance.cde.config.Parameters 20import chisel3._ 21import chisel3.util._ 22import utils._ 23import utility._ 24import xiangshan._ 25import xiangshan.backend._ 26import xiangshan.backend.fu.fpu._ 27import xiangshan.backend.rob.RobLsqIO 28import xiangshan.cache._ 29import xiangshan.cache.mmu._ 30import xiangshan.frontend.FtqPtr 31import xiangshan.ExceptionNO._ 32import xiangshan.mem.mdp._ 33import xiangshan.backend.Bundles.{DynInst, MemExuOutput, MemMicroOpRbExt} 34import xiangshan.backend.rob.RobPtr 35 36class LqPtr(implicit p: Parameters) extends CircularQueuePtr[LqPtr]( 37 p => p(XSCoreParamsKey).VirtualLoadQueueSize 38){ 39} 40 41object LqPtr { 42 def apply(f: Bool, v: UInt)(implicit p: Parameters): LqPtr = { 43 val ptr = Wire(new LqPtr) 44 ptr.flag := f 45 ptr.value := v 46 ptr 47 } 48} 49 50trait HasLoadHelper { this: XSModule => 51 def rdataHelper(uop: DynInst, rdata: UInt): UInt = { 52 val fpWen = uop.fpWen 53 LookupTree(uop.fuOpType, List( 54 LSUOpType.lb -> SignExt(rdata(7, 0) , XLEN), 55 LSUOpType.lh -> SignExt(rdata(15, 0), XLEN), 56 /* 57 riscv-spec-20191213: 12.2 NaN Boxing of Narrower Values 58 Any operation that writes a narrower result to an f register must write 59 all 1s to the uppermost FLEN−n bits to yield a legal NaN-boxed value. 60 */ 61 LSUOpType.lw -> Mux(fpWen, FPU.box(rdata, FPU.S), SignExt(rdata(31, 0), XLEN)), 62 LSUOpType.ld -> Mux(fpWen, FPU.box(rdata, FPU.D), SignExt(rdata(63, 0), XLEN)), 63 LSUOpType.lbu -> ZeroExt(rdata(7, 0) , XLEN), 64 LSUOpType.lhu -> ZeroExt(rdata(15, 0), XLEN), 65 LSUOpType.lwu -> ZeroExt(rdata(31, 0), XLEN), 66 )) 67 } 68 69 def rdataVecHelper(alignedType: UInt, rdata: UInt): UInt = { 70 LookupTree(alignedType, List( 71 "b00".U -> ZeroExt(rdata(7, 0), VLEN), 72 "b01".U -> ZeroExt(rdata(15, 0), VLEN), 73 "b10".U -> ZeroExt(rdata(31, 0), VLEN), 74 "b11".U -> ZeroExt(rdata(63, 0), VLEN) 75 )) 76 } 77} 78 79class LqEnqIO(implicit p: Parameters) extends MemBlockBundle { 80 val canAccept = Output(Bool()) 81 val sqCanAccept = Input(Bool()) 82 val needAlloc = Vec(LSQEnqWidth, Input(Bool())) 83 val req = Vec(LSQEnqWidth, Flipped(ValidIO(new DynInst))) 84 val resp = Vec(LSQEnqWidth, Output(new LqPtr)) 85} 86 87class LqTriggerIO(implicit p: Parameters) extends XSBundle { 88 val hitLoadAddrTriggerHitVec = Input(Vec(TriggerNum, Bool())) 89 val lqLoadAddrTriggerHitVec = Output(Vec(TriggerNum, Bool())) 90} 91 92class LoadQueueTopDownIO(implicit p: Parameters) extends XSBundle { 93 val robHeadVaddr = Flipped(Valid(UInt(VAddrBits.W))) 94 val robHeadTlbReplay = Output(Bool()) 95 val robHeadTlbMiss = Output(Bool()) 96 val robHeadLoadVio = Output(Bool()) 97 val robHeadLoadMSHR = Output(Bool()) 98 val robHeadMissInDTlb = Input(Bool()) 99 val robHeadOtherReplay = Output(Bool()) 100} 101 102class LoadQueue(implicit p: Parameters) extends XSModule 103 with HasDCacheParameters 104 with HasCircularQueuePtrHelper 105 with HasLoadHelper 106 with HasPerfEvents 107{ 108 val io = IO(new Bundle() { 109 val redirect = Flipped(Valid(new Redirect)) 110 val enq = new LqEnqIO 111 val ldu = new Bundle() { 112 val stld_nuke_query = Vec(LoadPipelineWidth, Flipped(new LoadNukeQueryIO)) // from load_s2 113 val ldld_nuke_query = Vec(LoadPipelineWidth, Flipped(new LoadNukeQueryIO)) // from load_s2 114 val ldin = Vec(LoadPipelineWidth, Flipped(Decoupled(new LqWriteBundle))) // from load_s3 115 } 116 val sta = new Bundle() { 117 val storeAddrIn = Vec(StorePipelineWidth, Flipped(Valid(new LsPipelineBundle))) // from store_s1 118 val vecStoreAddrIn = Vec(StorePipelineWidth, Flipped(Valid(new LsPipelineBundle))) // from store_s1 119 } 120 val std = new Bundle() { 121 val storeDataIn = Vec(StorePipelineWidth, Flipped(Valid(new MemExuOutput))) // from store_s0, store data, send to sq from rs 122 } 123 val sq = new Bundle() { 124 val stAddrReadySqPtr = Input(new SqPtr) 125 val stAddrReadyVec = Input(Vec(StoreQueueSize, Bool())) 126 val stDataReadySqPtr = Input(new SqPtr) 127 val stDataReadyVec = Input(Vec(StoreQueueSize, Bool())) 128 val stIssuePtr = Input(new SqPtr) 129 val sqEmpty = Input(Bool()) 130 } 131 val ldout = Vec(LoadPipelineWidth, DecoupledIO(new MemExuOutput)) 132 val ld_raw_data = Vec(LoadPipelineWidth, Output(new LoadDataFromLQBundle)) 133 val replay = Vec(LoadPipelineWidth, Decoupled(new LsPipelineBundle)) 134 val refill = Flipped(ValidIO(new Refill)) 135 val tl_d_channel = Input(new DcacheToLduForwardIO) 136 val release = Flipped(Valid(new Release)) 137 val nuke_rollback = Output(Valid(new Redirect)) 138 val nack_rollback = Output(Valid(new Redirect)) 139 val rob = Flipped(new RobLsqIO) 140 val uncache = new UncacheWordIO 141 val exceptionAddr = new ExceptionAddrIO 142 val lqFull = Output(Bool()) 143 val lqDeq = Output(UInt(log2Up(CommitWidth + 1).W)) 144 val lqCancelCnt = Output(UInt(log2Up(VirtualLoadQueueSize+1).W)) 145 val lq_rep_full = Output(Bool()) 146 val tlbReplayDelayCycleCtrl = Vec(4, Input(UInt(ReSelectLen.W))) 147 val l2_hint = Input(Valid(new L2ToL1Hint())) 148 val tlb_hint = Flipped(new TlbHintIO) 149 val lqEmpty = Output(Bool()) 150 151 val vecWriteback = Flipped(ValidIO(new MemExuOutput(isVector = true))) 152 val lqDeqPtr = Output(new LqPtr) 153 154 val debugTopDown = new LoadQueueTopDownIO 155 }) 156 157 val loadQueueRAR = Module(new LoadQueueRAR) // read-after-read violation 158 val loadQueueRAW = Module(new LoadQueueRAW) // read-after-write violation 159 val loadQueueReplay = Module(new LoadQueueReplay) // enqueue if need replay 160 val virtualLoadQueue = Module(new VirtualLoadQueue) // control state 161 val exceptionBuffer = Module(new LqExceptionBuffer) // exception buffer 162 val uncacheBuffer = Module(new UncacheBuffer) // uncache buffer 163 164 /** 165 * LoadQueueRAR 166 */ 167 loadQueueRAR.io.redirect <> io.redirect 168 loadQueueRAR.io.release <> io.release 169 loadQueueRAR.io.ldWbPtr <> virtualLoadQueue.io.ldWbPtr 170 for (w <- 0 until LoadPipelineWidth) { 171 loadQueueRAR.io.query(w).req <> io.ldu.ldld_nuke_query(w).req // from load_s1 172 loadQueueRAR.io.query(w).resp <> io.ldu.ldld_nuke_query(w).resp // to load_s2 173 loadQueueRAR.io.query(w).revoke := io.ldu.ldld_nuke_query(w).revoke // from load_s3 174 } 175 176 /** 177 * LoadQueueRAW 178 */ 179 loadQueueRAW.io.redirect <> io.redirect 180 loadQueueRAW.io.storeIn <> io.sta.storeAddrIn 181 loadQueueRAW.io.vecStoreIn <> io.sta.vecStoreAddrIn 182 loadQueueRAW.io.stAddrReadySqPtr <> io.sq.stAddrReadySqPtr 183 loadQueueRAW.io.stIssuePtr <> io.sq.stIssuePtr 184 for (w <- 0 until LoadPipelineWidth) { 185 loadQueueRAW.io.query(w).req <> io.ldu.stld_nuke_query(w).req // from load_s1 186 loadQueueRAW.io.query(w).resp <> io.ldu.stld_nuke_query(w).resp // to load_s2 187 loadQueueRAW.io.query(w).revoke := io.ldu.stld_nuke_query(w).revoke // from load_s3 188 } 189 190 /** 191 * VirtualLoadQueue 192 */ 193 virtualLoadQueue.io.redirect <> io.redirect 194 virtualLoadQueue.io.enq <> io.enq 195 virtualLoadQueue.io.ldin <> io.ldu.ldin // from load_s3 196 virtualLoadQueue.io.lqFull <> io.lqFull 197 virtualLoadQueue.io.lqDeq <> io.lqDeq 198 virtualLoadQueue.io.lqCancelCnt <> io.lqCancelCnt 199 virtualLoadQueue.io.lqEmpty <> io.lqEmpty 200 virtualLoadQueue.io.vecWriteback <> io.vecWriteback 201 virtualLoadQueue.io.ldWbPtr <> io.lqDeqPtr 202 203 /** 204 * Load queue exception buffer 205 */ 206 exceptionBuffer.io.redirect <> io.redirect 207 for ((buff, w) <- exceptionBuffer.io.req.zipWithIndex) { 208 buff.valid := io.ldu.ldin(w).valid // from load_s3 209 buff.bits := io.ldu.ldin(w).bits 210 } 211 io.exceptionAddr <> exceptionBuffer.io.exceptionAddr 212 213 /** 214 * Load uncache buffer 215 */ 216 uncacheBuffer.io.redirect <> io.redirect 217 uncacheBuffer.io.ldout <> io.ldout 218 uncacheBuffer.io.ld_raw_data <> io.ld_raw_data 219 uncacheBuffer.io.rob <> io.rob 220 uncacheBuffer.io.uncache <> io.uncache 221 for ((buff, w) <- uncacheBuffer.io.req.zipWithIndex) { 222 buff.valid := io.ldu.ldin(w).valid // from load_s3 223 buff.bits := io.ldu.ldin(w).bits // from load_s3 224 } 225 226 227 io.nuke_rollback := loadQueueRAW.io.rollback 228 io.nack_rollback := uncacheBuffer.io.rollback 229 230 /* <------- DANGEROUS: Don't change sequence here ! -------> */ 231 232 /** 233 * LoadQueueReplay 234 */ 235 loadQueueReplay.io.redirect <> io.redirect 236 loadQueueReplay.io.enq <> io.ldu.ldin // from load_s3 237 loadQueueReplay.io.storeAddrIn <> io.sta.storeAddrIn // from store_s1 238 loadQueueReplay.io.storeDataIn <> io.std.storeDataIn // from store_s0 239 loadQueueReplay.io.replay <> io.replay 240 loadQueueReplay.io.refill <> io.refill 241 loadQueueReplay.io.tl_d_channel <> io.tl_d_channel 242 loadQueueReplay.io.stAddrReadySqPtr <> io.sq.stAddrReadySqPtr 243 loadQueueReplay.io.stAddrReadyVec <> io.sq.stAddrReadyVec 244 loadQueueReplay.io.stDataReadySqPtr <> io.sq.stDataReadySqPtr 245 loadQueueReplay.io.stDataReadyVec <> io.sq.stDataReadyVec 246 loadQueueReplay.io.sqEmpty <> io.sq.sqEmpty 247 loadQueueReplay.io.lqFull <> io.lq_rep_full 248 loadQueueReplay.io.ldWbPtr <> virtualLoadQueue.io.ldWbPtr 249 loadQueueReplay.io.rarFull <> loadQueueRAR.io.lqFull 250 loadQueueReplay.io.rawFull <> loadQueueRAW.io.lqFull 251 loadQueueReplay.io.l2_hint <> io.l2_hint 252 loadQueueReplay.io.tlb_hint <> io.tlb_hint 253 loadQueueReplay.io.tlbReplayDelayCycleCtrl <> io.tlbReplayDelayCycleCtrl 254 255 loadQueueReplay.io.debugTopDown <> io.debugTopDown 256 257 val full_mask = Cat(loadQueueRAR.io.lqFull, loadQueueRAW.io.lqFull, loadQueueReplay.io.lqFull) 258 XSPerfAccumulate("full_mask_000", full_mask === 0.U) 259 XSPerfAccumulate("full_mask_001", full_mask === 1.U) 260 XSPerfAccumulate("full_mask_010", full_mask === 2.U) 261 XSPerfAccumulate("full_mask_011", full_mask === 3.U) 262 XSPerfAccumulate("full_mask_100", full_mask === 4.U) 263 XSPerfAccumulate("full_mask_101", full_mask === 5.U) 264 XSPerfAccumulate("full_mask_110", full_mask === 6.U) 265 XSPerfAccumulate("full_mask_111", full_mask === 7.U) 266 XSPerfAccumulate("nuke_rollback", io.nuke_rollback.valid) 267 XSPerfAccumulate("nack_rollabck", io.nack_rollback.valid) 268 269 // perf cnt 270 val perfEvents = Seq(virtualLoadQueue, loadQueueRAR, loadQueueRAW, loadQueueReplay).flatMap(_.getPerfEvents) ++ 271 Seq( 272 ("full_mask_000", full_mask === 0.U), 273 ("full_mask_001", full_mask === 1.U), 274 ("full_mask_010", full_mask === 2.U), 275 ("full_mask_011", full_mask === 3.U), 276 ("full_mask_100", full_mask === 4.U), 277 ("full_mask_101", full_mask === 5.U), 278 ("full_mask_110", full_mask === 6.U), 279 ("full_mask_111", full_mask === 7.U), 280 ("nuke_rollback", io.nuke_rollback.valid), 281 ("nack_rollback", io.nack_rollback.valid) 282 ) 283 generatePerfEvent() 284 // end 285}