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.frontend 18 19import chipsalliance.rocketchip.config.Parameters 20import chisel3._ 21import chisel3.util._ 22import freechips.rocketchip.rocket.RVCDecoder 23import xiangshan._ 24import xiangshan.cache.mmu._ 25import xiangshan.frontend.icache._ 26import utils._ 27import xiangshan.backend.fu.{PMPReqBundle, PMPRespBundle} 28 29trait HasInstrMMIOConst extends HasXSParameter with HasIFUConst{ 30 def mmioBusWidth = 64 31 def mmioBusBytes = mmioBusWidth / 8 32 def maxInstrLen = 32 33} 34 35trait HasIFUConst extends HasXSParameter{ 36 def addrAlign(addr: UInt, bytes: Int, highest: Int): UInt = Cat(addr(highest-1, log2Ceil(bytes)), 0.U(log2Ceil(bytes).W)) 37 def fetchQueueSize = 2 38 39 def getBasicBlockIdx( pc: UInt, start: UInt ): UInt = { 40 val byteOffset = pc - start 41 (byteOffset - instBytes.U)(log2Ceil(PredictWidth),instOffsetBits) 42 } 43} 44 45class IfuToFtqIO(implicit p:Parameters) extends XSBundle { 46 val pdWb = Valid(new PredecodeWritebackBundle) 47} 48 49class FtqInterface(implicit p: Parameters) extends XSBundle { 50 val fromFtq = Flipped(new FtqToIfuIO) 51 val toFtq = new IfuToFtqIO 52} 53 54class UncacheInterface(implicit p: Parameters) extends XSBundle { 55 val fromUncache = Flipped(DecoupledIO(new InsUncacheResp)) 56 val toUncache = DecoupledIO( new InsUncacheReq ) 57} 58class NewIFUIO(implicit p: Parameters) extends XSBundle { 59 val ftqInter = new FtqInterface 60 val icacheInter = Flipped(new IFUICacheIO) 61 val icacheStop = Output(Bool()) 62 val icachePerfInfo = Input(new ICachePerfInfo) 63 val toIbuffer = Decoupled(new FetchToIBuffer) 64 val uncacheInter = new UncacheInterface 65 val frontendTrigger = Flipped(new FrontendTdataDistributeIO) 66 val csrTriggerEnable = Input(Vec(4, Bool())) 67 val rob_commits = Flipped(Vec(CommitWidth, Valid(new RobCommitInfo))) 68 val iTLBInter = new TlbRequestIO 69 val pmp = new ICachePMPBundle 70} 71 72// record the situation in which fallThruAddr falls into 73// the middle of an RVI inst 74class LastHalfInfo(implicit p: Parameters) extends XSBundle { 75 val valid = Bool() 76 val middlePC = UInt(VAddrBits.W) 77 def matchThisBlock(startAddr: UInt) = valid && middlePC === startAddr 78} 79 80class IfuToPreDecode(implicit p: Parameters) extends XSBundle { 81 val data = if(HasCExtension) Vec(PredictWidth + 1, UInt(16.W)) else Vec(PredictWidth, UInt(32.W)) 82 val frontendTrigger = new FrontendTdataDistributeIO 83 val csrTriggerEnable = Vec(4, Bool()) 84 val pc = Vec(PredictWidth, UInt(VAddrBits.W)) 85} 86 87 88class IfuToPredChecker(implicit p: Parameters) extends XSBundle { 89 val ftqOffset = Valid(UInt(log2Ceil(PredictWidth).W)) 90 val jumpOffset = Vec(PredictWidth, UInt(XLEN.W)) 91 val target = UInt(VAddrBits.W) 92 val instrRange = Vec(PredictWidth, Bool()) 93 val instrValid = Vec(PredictWidth, Bool()) 94 val pds = Vec(PredictWidth, new PreDecodeInfo) 95 val pc = Vec(PredictWidth, UInt(VAddrBits.W)) 96} 97 98class NewIFU(implicit p: Parameters) extends XSModule 99 with HasICacheParameters 100 with HasIFUConst 101 with HasPdConst 102 with HasCircularQueuePtrHelper 103 with HasPerfEvents 104{ 105 val io = IO(new NewIFUIO) 106 val (toFtq, fromFtq) = (io.ftqInter.toFtq, io.ftqInter.fromFtq) 107 val fromICache = io.icacheInter.resp 108 val (toUncache, fromUncache) = (io.uncacheInter.toUncache , io.uncacheInter.fromUncache) 109 110 def isCrossLineReq(start: UInt, end: UInt): Bool = start(blockOffBits) ^ end(blockOffBits) 111 112 def isLastInCacheline(addr: UInt): Bool = addr(blockOffBits - 1, 1) === 0.U 113 114 class TlbExept(implicit p: Parameters) extends XSBundle{ 115 val pageFault = Bool() 116 val accessFault = Bool() 117 val mmio = Bool() 118 } 119 120 val preDecoders = Seq.fill(4){ Module(new PreDecode) } 121 122 val predChecker = Module(new PredChecker) 123 val frontendTrigger = Module(new FrontendTrigger) 124 val (checkerIn, checkerOutStage1, checkerOutStage2) = (predChecker.io.in, predChecker.io.out.stage1Out,predChecker.io.out.stage2Out) 125 126 io.iTLBInter.req_kill := false.B 127 io.iTLBInter.resp.ready := true.B 128 129 /** 130 ****************************************************************************** 131 * IFU Stage 0 132 * - send cacheline fetch request to ICacheMainPipe 133 ****************************************************************************** 134 */ 135 136 val f0_valid = fromFtq.req.valid 137 val f0_ftq_req = fromFtq.req.bits 138 val f0_doubleLine = fromFtq.req.bits.crossCacheline 139 val f0_vSetIdx = VecInit(get_idx((f0_ftq_req.startAddr)), get_idx(f0_ftq_req.nextlineStart)) 140 val f0_fire = fromFtq.req.fire() 141 142 val f0_flush, f1_flush, f2_flush, f3_flush = WireInit(false.B) 143 val from_bpu_f0_flush, from_bpu_f1_flush, from_bpu_f2_flush, from_bpu_f3_flush = WireInit(false.B) 144 145 from_bpu_f0_flush := fromFtq.flushFromBpu.shouldFlushByStage2(f0_ftq_req.ftqIdx) || 146 fromFtq.flushFromBpu.shouldFlushByStage3(f0_ftq_req.ftqIdx) 147 148 val wb_redirect , mmio_redirect, backend_redirect= WireInit(false.B) 149 val f3_wb_not_flush = WireInit(false.B) 150 151 backend_redirect := fromFtq.redirect.valid 152 f3_flush := backend_redirect || (wb_redirect && !f3_wb_not_flush) 153 f2_flush := backend_redirect || mmio_redirect || wb_redirect 154 f1_flush := f2_flush || from_bpu_f1_flush 155 f0_flush := f1_flush || from_bpu_f0_flush 156 157 val f1_ready, f2_ready, f3_ready = WireInit(false.B) 158 159 fromFtq.req.ready := f1_ready && io.icacheInter.icacheReady 160 161 /** <PERF> f0 fetch bubble */ 162 163 XSPerfAccumulate("fetch_bubble_ftq_not_valid", !fromFtq.req.valid && fromFtq.req.ready ) 164 // XSPerfAccumulate("fetch_bubble_pipe_stall", f0_valid && toICache(0).ready && toICache(1).ready && !f1_ready ) 165 // XSPerfAccumulate("fetch_bubble_icache_0_busy", f0_valid && !toICache(0).ready ) 166 // XSPerfAccumulate("fetch_bubble_icache_1_busy", f0_valid && !toICache(1).ready ) 167 XSPerfAccumulate("fetch_flush_backend_redirect", backend_redirect ) 168 XSPerfAccumulate("fetch_flush_wb_redirect", wb_redirect ) 169 XSPerfAccumulate("fetch_flush_bpu_f1_flush", from_bpu_f1_flush ) 170 XSPerfAccumulate("fetch_flush_bpu_f0_flush", from_bpu_f0_flush ) 171 172 173 /** 174 ****************************************************************************** 175 * IFU Stage 1 176 * - calculate pc/half_pc/cut_ptr for every instruction 177 ****************************************************************************** 178 */ 179 180 val f1_valid = RegInit(false.B) 181 val f1_ftq_req = RegEnable(f0_ftq_req, f0_fire) 182 // val f1_situation = RegEnable(f0_situation, f0_fire) 183 val f1_doubleLine = RegEnable(f0_doubleLine, f0_fire) 184 val f1_vSetIdx = RegEnable(f0_vSetIdx, f0_fire) 185 val f1_fire = f1_valid && f2_ready 186 187 f1_ready := f1_fire || !f1_valid 188 189 from_bpu_f1_flush := fromFtq.flushFromBpu.shouldFlushByStage3(f1_ftq_req.ftqIdx) && f1_valid 190 // from_bpu_f1_flush := false.B 191 192 when(f1_flush) {f1_valid := false.B} 193 .elsewhen(f0_fire && !f0_flush) {f1_valid := true.B} 194 .elsewhen(f1_fire) {f1_valid := false.B} 195 196 val f1_pc = VecInit((0 until PredictWidth).map(i => f1_ftq_req.startAddr + (i * 2).U)) 197 val f1_half_snpc = VecInit((0 until PredictWidth).map(i => f1_ftq_req.startAddr + ((i+2) * 2).U)) 198 val f1_cut_ptr = if(HasCExtension) VecInit((0 until PredictWidth + 1).map(i => Cat(0.U(1.W), f1_ftq_req.startAddr(blockOffBits-1, 1)) + i.U )) 199 else VecInit((0 until PredictWidth).map(i => Cat(0.U(1.W), f1_ftq_req.startAddr(blockOffBits-1, 2)) + i.U )) 200 201 /** 202 ****************************************************************************** 203 * IFU Stage 2 204 * - icache response data (latched for pipeline stop) 205 * - generate exceprion bits for every instruciton (page fault/access fault/mmio) 206 * - generate predicted instruction range (1 means this instruciton is in this fetch packet) 207 * - cut data from cachlines to packet instruction code 208 * - instruction predecode and RVC expand 209 ****************************************************************************** 210 */ 211 212 val icacheRespAllValid = WireInit(false.B) 213 214 val f2_valid = RegInit(false.B) 215 val f2_ftq_req = RegEnable(f1_ftq_req, f1_fire) 216 // val f2_situation = RegEnable(f1_situation, f1_fire) 217 val f2_doubleLine = RegEnable(f1_doubleLine, f1_fire) 218 val f2_vSetIdx = RegEnable(f1_vSetIdx, f1_fire) 219 val f2_fire = f2_valid && f3_ready && icacheRespAllValid 220 221 f2_ready := f2_fire || !f2_valid 222 //TODO: addr compare may be timing critical 223 val f2_icache_all_resp_wire = fromICache(0).valid && (fromICache(0).bits.vaddr === f2_ftq_req.startAddr) && ((fromICache(1).valid && (fromICache(1).bits.vaddr === f2_ftq_req.nextlineStart)) || !f2_doubleLine) 224 val f2_icache_all_resp_reg = RegInit(false.B) 225 226 icacheRespAllValid := f2_icache_all_resp_reg || f2_icache_all_resp_wire 227 228 io.icacheStop := !f3_ready 229 230 when(f2_flush) {f2_icache_all_resp_reg := false.B} 231 .elsewhen(f2_valid && f2_icache_all_resp_wire && !f3_ready) {f2_icache_all_resp_reg := true.B} 232 .elsewhen(f2_fire && f2_icache_all_resp_reg) {f2_icache_all_resp_reg := false.B} 233 234 when(f2_flush) {f2_valid := false.B} 235 .elsewhen(f1_fire && !f1_flush) {f2_valid := true.B } 236 .elsewhen(f2_fire) {f2_valid := false.B} 237 238 // val f2_cache_response_data = ResultHoldBypass(valid = f2_icache_all_resp_wire, data = VecInit(fromICache.map(_.bits.readData))) 239 val f2_cache_response_reg_data = VecInit(fromICache.map(_.bits.registerData)) 240 val f2_cache_response_sram_data = VecInit(fromICache.map(_.bits.sramData)) 241 val f2_cache_response_select = VecInit(fromICache.map(_.bits.select)) 242 243 244 val f2_except_pf = VecInit((0 until PortNumber).map(i => fromICache(i).bits.tlbExcp.pageFault)) 245 val f2_except_af = VecInit((0 until PortNumber).map(i => fromICache(i).bits.tlbExcp.accessFault)) 246 val f2_mmio = fromICache(0).bits.tlbExcp.mmio && !fromICache(0).bits.tlbExcp.accessFault && 247 !fromICache(0).bits.tlbExcp.pageFault 248 249 val f2_pc = RegEnable(f1_pc, f1_fire) 250 val f2_half_snpc = RegEnable(f1_half_snpc, f1_fire) 251 val f2_cut_ptr = RegEnable(f1_cut_ptr, f1_fire) 252 253 val f2_resend_vaddr = RegEnable(f1_ftq_req.startAddr + 2.U, f1_fire) 254 255 def isNextLine(pc: UInt, startAddr: UInt) = { 256 startAddr(blockOffBits) ^ pc(blockOffBits) 257 } 258 259 def isLastInLine(pc: UInt) = { 260 pc(blockOffBits - 1, 0) === "b111110".U 261 } 262 263 val f2_foldpc = VecInit(f2_pc.map(i => XORFold(i(VAddrBits-1,1), MemPredPCWidth))) 264 val f2_jump_range = Fill(PredictWidth, !f2_ftq_req.ftqOffset.valid) | Fill(PredictWidth, 1.U(1.W)) >> ~f2_ftq_req.ftqOffset.bits 265 val f2_ftr_range = Fill(PredictWidth, f2_ftq_req.ftqOffset.valid) | Fill(PredictWidth, 1.U(1.W)) >> ~getBasicBlockIdx(f2_ftq_req.nextStartAddr, f2_ftq_req.startAddr) 266 val f2_instr_range = f2_jump_range & f2_ftr_range 267 val f2_pf_vec = VecInit((0 until PredictWidth).map(i => (!isNextLine(f2_pc(i), f2_ftq_req.startAddr) && f2_except_pf(0) || isNextLine(f2_pc(i), f2_ftq_req.startAddr) && f2_doubleLine && f2_except_pf(1)))) 268 val f2_af_vec = VecInit((0 until PredictWidth).map(i => (!isNextLine(f2_pc(i), f2_ftq_req.startAddr) && f2_except_af(0) || isNextLine(f2_pc(i), f2_ftq_req.startAddr) && f2_doubleLine && f2_except_af(1)))) 269 270 val f2_paddrs = VecInit((0 until PortNumber).map(i => fromICache(i).bits.paddr)) 271 val f2_perf_info = io.icachePerfInfo 272 273 def cut(cacheline: UInt, cutPtr: Vec[UInt]) : Vec[UInt] ={ 274 require(HasCExtension) 275 // if(HasCExtension){ 276 val partCacheline = cacheline((blockBytes * 8 * 2 * 3) / 4 - 1, 0) 277 val result = Wire(Vec(PredictWidth + 1, UInt(16.W))) 278 val dataVec = cacheline.asTypeOf(Vec(blockBytes * 3 /4, UInt(16.W))) //47 16-bit data vector 279 (0 until PredictWidth + 1).foreach( i => 280 result(i) := dataVec(cutPtr(i)) //the max ptr is 3*blockBytes/4-1 281 ) 282 result 283 // } else { 284 // val result = Wire(Vec(PredictWidth, UInt(32.W)) ) 285 // val dataVec = cacheline.asTypeOf(Vec(blockBytes * 2/ 4, UInt(32.W))) 286 // (0 until PredictWidth).foreach( i => 287 // result(i) := dataVec(cutPtr(i)) 288 // ) 289 // result 290 // } 291 } 292 293 val f2_data_2_cacheline = Wire(Vec(4, UInt((2 * blockBits).W))) 294 f2_data_2_cacheline(0) := Cat(f2_cache_response_reg_data(1) , f2_cache_response_reg_data(0)) 295 f2_data_2_cacheline(1) := Cat(f2_cache_response_reg_data(1) , f2_cache_response_sram_data(0)) 296 f2_data_2_cacheline(2) := Cat(f2_cache_response_sram_data(1) , f2_cache_response_reg_data(0)) 297 f2_data_2_cacheline(3) := Cat(f2_cache_response_sram_data(1) , f2_cache_response_sram_data(0)) 298 299 val f2_cut_data = VecInit(f2_data_2_cacheline.map(data => cut( data, f2_cut_ptr ))) 300 301 val f2_predecod_ptr = Wire(UInt(2.W)) 302 f2_predecod_ptr := Cat(f2_cache_response_select(1),f2_cache_response_select(0)) 303 304 /** predecode (include RVC expander) */ 305 // preDecoderRegIn.data := f2_reg_cut_data 306 // preDecoderRegInIn.frontendTrigger := io.frontendTrigger 307 // preDecoderRegInIn.csrTriggerEnable := io.csrTriggerEnable 308 // preDecoderRegIn.pc := f2_pc 309 310 val preDecoderOut = Mux1H(UIntToOH(f2_predecod_ptr), preDecoders.map(_.io.out)) 311 for(i <- 0 until 4){ 312 val preDecoderIn = preDecoders(i).io.in 313 preDecoderIn.data := f2_cut_data(i) 314 preDecoderIn.frontendTrigger := io.frontendTrigger 315 preDecoderIn.csrTriggerEnable := io.csrTriggerEnable 316 preDecoderIn.pc := f2_pc 317 } 318 319 //val f2_expd_instr = preDecoderOut.expInstr 320 val f2_instr = preDecoderOut.instr 321 val f2_pd = preDecoderOut.pd 322 val f2_jump_offset = preDecoderOut.jumpOffset 323 val f2_hasHalfValid = preDecoderOut.hasHalfValid 324 val f2_crossPageFault = VecInit((0 until PredictWidth).map(i => isLastInLine(f2_pc(i)) && !f2_except_pf(0) && f2_doubleLine && f2_except_pf(1) && !f2_pd(i).isRVC )) 325 326 XSPerfAccumulate("fetch_bubble_icache_not_resp", f2_valid && !icacheRespAllValid ) 327 328 329 /** 330 ****************************************************************************** 331 * IFU Stage 3 332 * - handle MMIO instruciton 333 * -send request to Uncache fetch Unit 334 * -every packet include 1 MMIO instruction 335 * -MMIO instructions will stop fetch pipeline until commiting from RoB 336 * -flush to snpc (send ifu_redirect to Ftq) 337 * - Ibuffer enqueue 338 * - check predict result in Frontend (jalFault/retFault/notCFIFault/invalidTakenFault/targetFault) 339 * - handle last half RVI instruction 340 ****************************************************************************** 341 */ 342 343 val f3_valid = RegInit(false.B) 344 val f3_ftq_req = RegEnable(f2_ftq_req, f2_fire) 345 // val f3_situation = RegEnable(f2_situation, f2_fire) 346 val f3_doubleLine = RegEnable(f2_doubleLine, f2_fire) 347 val f3_fire = io.toIbuffer.fire() 348 349 f3_ready := f3_fire || !f3_valid 350 351 val f3_cut_data = RegEnable(next = f2_cut_data(f2_predecod_ptr), enable=f2_fire) 352 353 val f3_except_pf = RegEnable(f2_except_pf, f2_fire) 354 val f3_except_af = RegEnable(f2_except_af, f2_fire) 355 val f3_mmio = RegEnable(f2_mmio , f2_fire) 356 357 //val f3_expd_instr = RegEnable(next = f2_expd_instr, enable = f2_fire) 358 val f3_instr = RegEnable(next = f2_instr, enable = f2_fire) 359 val f3_expd_instr = VecInit((0 until PredictWidth).map{ i => 360 val expander = Module(new RVCExpander) 361 expander.io.in := f3_instr(i) 362 expander.io.out.bits 363 }) 364 365 val f3_pd = RegEnable(next = f2_pd, enable = f2_fire) 366 val f3_jump_offset = RegEnable(next = f2_jump_offset, enable = f2_fire) 367 val f3_af_vec = RegEnable(next = f2_af_vec, enable = f2_fire) 368 val f3_pf_vec = RegEnable(next = f2_pf_vec , enable = f2_fire) 369 val f3_pc = RegEnable(next = f2_pc, enable = f2_fire) 370 val f3_half_snpc = RegEnable(next = f2_half_snpc, enable = f2_fire) 371 val f3_instr_range = RegEnable(next = f2_instr_range, enable = f2_fire) 372 val f3_foldpc = RegEnable(next = f2_foldpc, enable = f2_fire) 373 val f3_crossPageFault = RegEnable(next = f2_crossPageFault, enable = f2_fire) 374 val f3_hasHalfValid = RegEnable(next = f2_hasHalfValid, enable = f2_fire) 375 val f3_except = VecInit((0 until 2).map{i => f3_except_pf(i) || f3_except_af(i)}) 376 val f3_has_except = f3_valid && (f3_except_af.reduce(_||_) || f3_except_pf.reduce(_||_)) 377 val f3_pAddrs = RegEnable(f2_paddrs, f2_fire) 378 val f3_resend_vaddr = RegEnable(f2_resend_vaddr, f2_fire) 379 380 when(f3_valid && !f3_ftq_req.ftqOffset.valid){ 381 assert(f3_ftq_req.startAddr + 32.U >= f3_ftq_req.nextStartAddr , "More tha 32 Bytes fetch is not allowed!") 382 } 383 384 /*** MMIO State Machine***/ 385 val f3_mmio_data = Reg(Vec(2, UInt(16.W))) 386 val mmio_is_RVC = RegInit(false.B) 387 val mmio_resend_addr =RegInit(0.U(PAddrBits.W)) 388 val mmio_resend_af = RegInit(false.B) 389 val mmio_resend_pf = RegInit(false.B) 390 391 392 val m_idle :: m_sendReq :: m_waitResp :: m_sendTLB :: m_tlbResp :: m_sendPMP :: m_resendReq :: m_waitResendResp :: m_waitCommit :: m_commited :: Nil = Enum(10) 393 val mmio_state = RegInit(m_idle) 394 395 val f3_req_is_mmio = f3_mmio && f3_valid 396 val mmio_commit = VecInit(io.rob_commits.map{commit => commit.valid && commit.bits.ftqIdx === f3_ftq_req.ftqIdx && commit.bits.ftqOffset === 0.U}).asUInt.orR 397 val f3_mmio_req_commit = f3_req_is_mmio && mmio_state === m_commited 398 399 val f3_mmio_to_commit = f3_req_is_mmio && mmio_state === m_waitCommit 400 val f3_mmio_to_commit_next = RegNext(f3_mmio_to_commit) 401 val f3_mmio_can_go = f3_mmio_to_commit && !f3_mmio_to_commit_next 402 403 val fromFtqRedirectReg = RegNext(fromFtq.redirect) 404 val f3_ftq_flush_self = fromFtqRedirectReg.valid && RedirectLevel.flushItself(fromFtqRedirectReg.bits.level) 405 val f3_ftq_flush_by_older = fromFtqRedirectReg.valid && isBefore(fromFtqRedirectReg.bits.ftqIdx, f3_ftq_req.ftqIdx) 406 407 val f3_need_not_flush = f3_req_is_mmio && fromFtqRedirectReg.valid && !f3_ftq_flush_self && !f3_ftq_flush_by_older 408 409 when(f3_flush && !f3_need_not_flush) {f3_valid := false.B} 410 .elsewhen(f2_fire && !f2_flush ) {f3_valid := true.B } 411 .elsewhen(io.toIbuffer.fire() && !f3_req_is_mmio) {f3_valid := false.B} 412 .elsewhen{f3_req_is_mmio && f3_mmio_req_commit} {f3_valid := false.B} 413 414 val f3_mmio_use_seq_pc = RegInit(false.B) 415 416 val (redirect_ftqIdx, redirect_ftqOffset) = (fromFtqRedirectReg.bits.ftqIdx,fromFtqRedirectReg.bits.ftqOffset) 417 val redirect_mmio_req = fromFtqRedirectReg.valid && redirect_ftqIdx === f3_ftq_req.ftqIdx && redirect_ftqOffset === 0.U 418 419 when(RegNext(f2_fire && !f2_flush) && f3_req_is_mmio) { f3_mmio_use_seq_pc := true.B } 420 .elsewhen(redirect_mmio_req) { f3_mmio_use_seq_pc := false.B } 421 422 f3_ready := Mux(f3_req_is_mmio, io.toIbuffer.ready && f3_mmio_req_commit || !f3_valid , io.toIbuffer.ready || !f3_valid) 423 424 425 switch(mmio_state){ 426 is(m_idle){ 427 when(f3_req_is_mmio){ 428 mmio_state := m_sendReq 429 } 430 } 431 432 is(m_sendReq){ 433 mmio_state := Mux(toUncache.fire(), m_waitResp, m_sendReq ) 434 } 435 436 is(m_waitResp){ 437 when(fromUncache.fire()){ 438 val isRVC = fromUncache.bits.data(1,0) =/= 3.U 439 val needResend = !isRVC && f3_pAddrs(0)(2,1) === 3.U 440 mmio_state := Mux(needResend, m_sendTLB , m_waitCommit) 441 442 mmio_is_RVC := isRVC 443 f3_mmio_data(0) := fromUncache.bits.data(15,0) 444 f3_mmio_data(1) := fromUncache.bits.data(31,16) 445 } 446 } 447 448 is(m_sendTLB){ 449 when( io.iTLBInter.req.valid && !io.iTLBInter.resp.bits.miss ){ 450 mmio_state := m_tlbResp 451 } 452 } 453 454 is(m_tlbResp){ 455 val tlbExept = io.iTLBInter.resp.bits.excp(0).pf.instr || 456 io.iTLBInter.resp.bits.excp(0).af.instr 457 mmio_state := Mux(tlbExept,m_waitCommit,m_sendPMP) 458 mmio_resend_addr := io.iTLBInter.resp.bits.paddr(0) 459 mmio_resend_af := mmio_resend_af || io.iTLBInter.resp.bits.excp(0).af.instr 460 mmio_resend_pf := mmio_resend_pf || io.iTLBInter.resp.bits.excp(0).pf.instr 461 } 462 463 is(m_sendPMP){ 464 val pmpExcpAF = io.pmp.resp.instr || !io.pmp.resp.mmio 465 mmio_state := Mux(pmpExcpAF, m_waitCommit , m_resendReq) 466 mmio_resend_af := pmpExcpAF 467 } 468 469 is(m_resendReq){ 470 mmio_state := Mux(toUncache.fire(), m_waitResendResp, m_resendReq ) 471 } 472 473 is(m_waitResendResp){ 474 when(fromUncache.fire()){ 475 mmio_state := m_waitCommit 476 f3_mmio_data(1) := fromUncache.bits.data(15,0) 477 } 478 } 479 480 is(m_waitCommit){ 481 when(mmio_commit){ 482 mmio_state := m_commited 483 } 484 } 485 486 //normal mmio instruction 487 is(m_commited){ 488 mmio_state := m_idle 489 mmio_is_RVC := false.B 490 mmio_resend_addr := 0.U 491 } 492 } 493 494 //exception or flush by older branch prediction 495 when(f3_ftq_flush_self || f3_ftq_flush_by_older) { 496 mmio_state := m_idle 497 mmio_is_RVC := false.B 498 mmio_resend_addr := 0.U 499 mmio_resend_af := false.B 500 f3_mmio_data.map(_ := 0.U) 501 } 502 503 toUncache.valid := ((mmio_state === m_sendReq) || (mmio_state === m_resendReq)) && f3_req_is_mmio 504 toUncache.bits.addr := Mux((mmio_state === m_resendReq), mmio_resend_addr, f3_pAddrs(0)) 505 fromUncache.ready := true.B 506 507 io.iTLBInter.req.valid := (mmio_state === m_sendTLB) && f3_req_is_mmio 508 io.iTLBInter.req.bits.size := 3.U 509 io.iTLBInter.req.bits.vaddr := f3_resend_vaddr 510 io.iTLBInter.req.bits.debug.pc := f3_resend_vaddr 511 512 io.iTLBInter.req.bits.kill := false.B // IFU use itlb for mmio, doesn't need sync, set it to false 513 io.iTLBInter.req.bits.cmd := TlbCmd.exec 514 io.iTLBInter.req.bits.debug.robIdx := DontCare 515 io.iTLBInter.req.bits.debug.isFirstIssue := DontCare 516 517 io.pmp.req.valid := (mmio_state === m_sendPMP) && f3_req_is_mmio 518 io.pmp.req.bits.addr := mmio_resend_addr 519 io.pmp.req.bits.size := 3.U 520 io.pmp.req.bits.cmd := TlbCmd.exec 521 522 val f3_lastHalf = RegInit(0.U.asTypeOf(new LastHalfInfo)) 523 524 val f3_predecode_range = VecInit(preDecoderOut.pd.map(inst => inst.valid)).asUInt 525 val f3_mmio_range = VecInit((0 until PredictWidth).map(i => if(i ==0) true.B else false.B)) 526 val f3_instr_valid = Wire(Vec(PredictWidth, Bool())) 527 528 /*** prediction result check ***/ 529 checkerIn.ftqOffset := f3_ftq_req.ftqOffset 530 checkerIn.jumpOffset := f3_jump_offset 531 checkerIn.target := f3_ftq_req.nextStartAddr 532 checkerIn.instrRange := f3_instr_range.asTypeOf(Vec(PredictWidth, Bool())) 533 checkerIn.instrValid := f3_instr_valid.asTypeOf(Vec(PredictWidth, Bool())) 534 checkerIn.pds := f3_pd 535 checkerIn.pc := f3_pc 536 537 /*** handle half RVI in the last 2 Bytes ***/ 538 539 def hasLastHalf(idx: UInt) = { 540 //!f3_pd(idx).isRVC && checkerOutStage1.fixedRange(idx) && f3_instr_valid(idx) && !checkerOutStage1.fixedTaken(idx) && !checkerOutStage2.fixedMissPred(idx) && ! f3_req_is_mmio 541 !f3_pd(idx).isRVC && checkerOutStage1.fixedRange(idx) && f3_instr_valid(idx) && !checkerOutStage1.fixedTaken(idx) && ! f3_req_is_mmio 542 } 543 544 val f3_last_validIdx = ~ParallelPriorityEncoder(checkerOutStage1.fixedRange.reverse) 545 546 val f3_hasLastHalf = hasLastHalf((PredictWidth - 1).U) 547 val f3_false_lastHalf = hasLastHalf(f3_last_validIdx) 548 val f3_false_snpc = f3_half_snpc(f3_last_validIdx) 549 550 val f3_lastHalf_mask = VecInit((0 until PredictWidth).map( i => if(i ==0) false.B else true.B )).asUInt() 551 val f3_lastHalf_disable = RegInit(false.B) 552 553 when(f3_flush || (f3_fire && f3_lastHalf_disable)){ 554 f3_lastHalf_disable := false.B 555 } 556 557 when (f3_flush) { 558 f3_lastHalf.valid := false.B 559 }.elsewhen (f3_fire) { 560 f3_lastHalf.valid := f3_hasLastHalf && !f3_lastHalf_disable 561 f3_lastHalf.middlePC := f3_ftq_req.nextStartAddr 562 } 563 564 f3_instr_valid := Mux(f3_lastHalf.valid,f3_hasHalfValid ,VecInit(f3_pd.map(inst => inst.valid))) 565 566 /*** frontend Trigger ***/ 567 frontendTrigger.io.pds := f3_pd 568 frontendTrigger.io.pc := f3_pc 569 frontendTrigger.io.data := f3_cut_data 570 571 frontendTrigger.io.frontendTrigger := io.frontendTrigger 572 frontendTrigger.io.csrTriggerEnable := io.csrTriggerEnable 573 574 val f3_triggered = frontendTrigger.io.triggered 575 576 /*** send to Ibuffer ***/ 577 578 io.toIbuffer.valid := f3_valid && (!f3_req_is_mmio || f3_mmio_can_go) && !f3_flush 579 io.toIbuffer.bits.instrs := f3_expd_instr 580 io.toIbuffer.bits.valid := f3_instr_valid.asUInt 581 io.toIbuffer.bits.enqEnable := checkerOutStage1.fixedRange.asUInt & f3_instr_valid.asUInt 582 io.toIbuffer.bits.pd := f3_pd 583 io.toIbuffer.bits.ftqPtr := f3_ftq_req.ftqIdx 584 io.toIbuffer.bits.pc := f3_pc 585 io.toIbuffer.bits.ftqOffset.zipWithIndex.map{case(a, i) => a.bits := i.U; a.valid := checkerOutStage1.fixedTaken(i) && !f3_req_is_mmio} 586 io.toIbuffer.bits.foldpc := f3_foldpc 587 io.toIbuffer.bits.ipf := VecInit(f3_pf_vec.zip(f3_crossPageFault).map{case (pf, crossPF) => pf || crossPF}) 588 io.toIbuffer.bits.acf := f3_af_vec 589 io.toIbuffer.bits.crossPageIPFFix := f3_crossPageFault 590 io.toIbuffer.bits.triggered := f3_triggered 591 592 when(f3_lastHalf.valid){ 593 io.toIbuffer.bits.enqEnable := checkerOutStage1.fixedRange.asUInt & f3_instr_valid.asUInt & f3_lastHalf_mask 594 io.toIbuffer.bits.valid := f3_lastHalf_mask & f3_instr_valid.asUInt 595 } 596 597 /** external predecode for MMIO instruction */ 598 when(f3_req_is_mmio){ 599 val inst = Cat(f3_mmio_data(1), f3_mmio_data(0)) 600 val currentIsRVC = isRVC(inst) 601 602 val brType::isCall::isRet::Nil = brInfo(inst) 603 val jalOffset = jal_offset(inst, currentIsRVC) 604 val brOffset = br_offset(inst, currentIsRVC) 605 606 io.toIbuffer.bits.instrs (0) := new RVCDecoder(inst, XLEN).decode.bits 607 608 io.toIbuffer.bits.pd(0).valid := true.B 609 io.toIbuffer.bits.pd(0).isRVC := currentIsRVC 610 io.toIbuffer.bits.pd(0).brType := brType 611 io.toIbuffer.bits.pd(0).isCall := isCall 612 io.toIbuffer.bits.pd(0).isRet := isRet 613 614 io.toIbuffer.bits.acf(0) := mmio_resend_af 615 io.toIbuffer.bits.ipf(0) := mmio_resend_pf 616 io.toIbuffer.bits.crossPageIPFFix(0) := mmio_resend_pf 617 618 io.toIbuffer.bits.enqEnable := f3_mmio_range.asUInt 619 } 620 621 622 //Write back to Ftq 623 val f3_cache_fetch = f3_valid && !(f2_fire && !f2_flush) 624 val finishFetchMaskReg = RegNext(f3_cache_fetch) 625 626 val mmioFlushWb = Wire(Valid(new PredecodeWritebackBundle)) 627 val f3_mmio_missOffset = Wire(ValidUndirectioned(UInt(log2Ceil(PredictWidth).W))) 628 f3_mmio_missOffset.valid := f3_req_is_mmio 629 f3_mmio_missOffset.bits := 0.U 630 631 mmioFlushWb.valid := (f3_req_is_mmio && mmio_state === m_waitCommit && RegNext(fromUncache.fire()) && f3_mmio_use_seq_pc) 632 mmioFlushWb.bits.pc := f3_pc 633 mmioFlushWb.bits.pd := f3_pd 634 mmioFlushWb.bits.pd.zipWithIndex.map{case(instr,i) => instr.valid := f3_mmio_range(i)} 635 mmioFlushWb.bits.ftqIdx := f3_ftq_req.ftqIdx 636 mmioFlushWb.bits.ftqOffset := f3_ftq_req.ftqOffset.bits 637 mmioFlushWb.bits.misOffset := f3_mmio_missOffset 638 mmioFlushWb.bits.cfiOffset := DontCare 639 mmioFlushWb.bits.target := Mux(mmio_is_RVC, f3_ftq_req.startAddr + 2.U , f3_ftq_req.startAddr + 4.U) 640 mmioFlushWb.bits.jalTarget := DontCare 641 mmioFlushWb.bits.instrRange := f3_mmio_range 642 643 mmio_redirect := (f3_req_is_mmio && mmio_state === m_waitCommit && RegNext(fromUncache.fire()) && f3_mmio_use_seq_pc) 644 645 XSPerfAccumulate("fetch_bubble_ibuffer_not_ready", io.toIbuffer.valid && !io.toIbuffer.ready ) 646 647 648 /** 649 ****************************************************************************** 650 * IFU Write Back Stage 651 * - write back predecode information to Ftq to update 652 * - redirect if found fault prediction 653 * - redirect if has false hit last half (last PC is not start + 32 Bytes, but in the midle of an notCFI RVI instruction) 654 ****************************************************************************** 655 */ 656 657 val wb_valid = RegNext(RegNext(f2_fire && !f2_flush) && !f3_req_is_mmio && !f3_flush) 658 val wb_ftq_req = RegNext(f3_ftq_req) 659 660 val wb_check_result_stage1 = RegNext(checkerOutStage1) 661 val wb_check_result_stage2 = checkerOutStage2 662 val wb_instr_range = RegNext(io.toIbuffer.bits.enqEnable) 663 val wb_pc = RegNext(f3_pc) 664 val wb_pd = RegNext(f3_pd) 665 val wb_instr_valid = RegNext(f3_instr_valid) 666 667 /* false hit lastHalf */ 668 val wb_lastIdx = RegNext(f3_last_validIdx) 669 val wb_false_lastHalf = RegNext(f3_false_lastHalf) && wb_lastIdx =/= (PredictWidth - 1).U 670 val wb_false_target = RegNext(f3_false_snpc) 671 672 val wb_half_flush = wb_false_lastHalf 673 val wb_half_target = wb_false_target 674 675 /* false oversize */ 676 val lastIsRVC = wb_instr_range.asTypeOf(Vec(PredictWidth,Bool())).last && wb_pd.last.isRVC 677 val lastIsRVI = wb_instr_range.asTypeOf(Vec(PredictWidth,Bool()))(PredictWidth - 2) && !wb_pd(PredictWidth - 2).isRVC 678 val lastTaken = wb_check_result_stage1.fixedTaken.last 679 680 f3_wb_not_flush := wb_ftq_req.ftqIdx === f3_ftq_req.ftqIdx && f3_valid && wb_valid 681 682 /** if a req with a last half but miss predicted enters in wb stage, and this cycle f3 stalls, 683 * we set a flag to notify f3 that the last half flag need not to be set. 684 */ 685 //f3_fire is after wb_valid 686 when(wb_valid && RegNext(f3_hasLastHalf,init = false.B) 687 && wb_check_result_stage2.fixedMissPred(PredictWidth - 1) && !f3_fire && !RegNext(f3_fire,init = false.B) 688 ){ 689 f3_lastHalf_disable := true.B 690 } 691 692 //wb_valid and f3_fire are in same cycle 693 when(wb_valid && RegNext(f3_hasLastHalf,init = false.B) 694 && wb_check_result_stage2.fixedMissPred(PredictWidth - 1) && f3_fire 695 ){ 696 f3_lastHalf.valid := false.B 697 } 698 699 val checkFlushWb = Wire(Valid(new PredecodeWritebackBundle)) 700 checkFlushWb.valid := wb_valid 701 checkFlushWb.bits.pc := wb_pc 702 checkFlushWb.bits.pd := wb_pd 703 checkFlushWb.bits.pd.zipWithIndex.map{case(instr,i) => instr.valid := wb_instr_valid(i)} 704 checkFlushWb.bits.ftqIdx := wb_ftq_req.ftqIdx 705 checkFlushWb.bits.ftqOffset := wb_ftq_req.ftqOffset.bits 706 checkFlushWb.bits.misOffset.valid := ParallelOR(wb_check_result_stage2.fixedMissPred) || wb_half_flush 707 checkFlushWb.bits.misOffset.bits := Mux(wb_half_flush, wb_lastIdx, ParallelPriorityEncoder(wb_check_result_stage2.fixedMissPred)) 708 checkFlushWb.bits.cfiOffset.valid := ParallelOR(wb_check_result_stage1.fixedTaken) 709 checkFlushWb.bits.cfiOffset.bits := ParallelPriorityEncoder(wb_check_result_stage1.fixedTaken) 710 checkFlushWb.bits.target := Mux(wb_half_flush, wb_half_target, wb_check_result_stage2.fixedTarget(ParallelPriorityEncoder(wb_check_result_stage2.fixedMissPred))) 711 checkFlushWb.bits.jalTarget := wb_check_result_stage2.fixedTarget(ParallelPriorityEncoder(VecInit(wb_pd.zip(wb_instr_valid).map{case (pd, v) => v && pd.isJal }))) 712 checkFlushWb.bits.instrRange := wb_instr_range.asTypeOf(Vec(PredictWidth, Bool())) 713 714 toFtq.pdWb := Mux(wb_valid, checkFlushWb, mmioFlushWb) 715 716 wb_redirect := checkFlushWb.bits.misOffset.valid && wb_valid 717 718 /*write back flush type*/ 719 val checkFaultType = wb_check_result_stage2.faultType 720 val checkJalFault = wb_valid && checkFaultType.map(_.isjalFault).reduce(_||_) 721 val checkRetFault = wb_valid && checkFaultType.map(_.isRetFault).reduce(_||_) 722 val checkTargetFault = wb_valid && checkFaultType.map(_.istargetFault).reduce(_||_) 723 val checkNotCFIFault = wb_valid && checkFaultType.map(_.notCFIFault).reduce(_||_) 724 val checkInvalidTaken = wb_valid && checkFaultType.map(_.invalidTakenFault).reduce(_||_) 725 726 727 XSPerfAccumulate("predecode_flush_jalFault", checkJalFault ) 728 XSPerfAccumulate("predecode_flush_retFault", checkRetFault ) 729 XSPerfAccumulate("predecode_flush_targetFault", checkTargetFault ) 730 XSPerfAccumulate("predecode_flush_notCFIFault", checkNotCFIFault ) 731 XSPerfAccumulate("predecode_flush_incalidTakenFault", checkInvalidTaken ) 732 733 when(checkRetFault){ 734 XSDebug("startAddr:%x nextstartAddr:%x taken:%d takenIdx:%d\n", 735 wb_ftq_req.startAddr, wb_ftq_req.nextStartAddr, wb_ftq_req.ftqOffset.valid, wb_ftq_req.ftqOffset.bits) 736 } 737 738 /** performance counter */ 739 val f3_perf_info = RegEnable(f2_perf_info, f2_fire) 740 val f3_req_0 = io.toIbuffer.fire() 741 val f3_req_1 = io.toIbuffer.fire() && f3_doubleLine 742 val f3_hit_0 = io.toIbuffer.fire() && f3_perf_info.bank_hit(0) 743 val f3_hit_1 = io.toIbuffer.fire() && f3_doubleLine & f3_perf_info.bank_hit(1) 744 val f3_hit = f3_perf_info.hit 745 val perfEvents = Seq( 746 ("frontendFlush ", wb_redirect ), 747 ("ifu_req ", io.toIbuffer.fire() ), 748 ("ifu_miss ", io.toIbuffer.fire() && !f3_perf_info.hit ), 749 ("ifu_req_cacheline_0 ", f3_req_0 ), 750 ("ifu_req_cacheline_1 ", f3_req_1 ), 751 ("ifu_req_cacheline_0_hit ", f3_hit_1 ), 752 ("ifu_req_cacheline_1_hit ", f3_hit_1 ), 753 ("only_0_hit ", f3_perf_info.only_0_hit && io.toIbuffer.fire() ), 754 ("only_0_miss ", f3_perf_info.only_0_miss && io.toIbuffer.fire() ), 755 ("hit_0_hit_1 ", f3_perf_info.hit_0_hit_1 && io.toIbuffer.fire() ), 756 ("hit_0_miss_1 ", f3_perf_info.hit_0_miss_1 && io.toIbuffer.fire() ), 757 ("miss_0_hit_1 ", f3_perf_info.miss_0_hit_1 && io.toIbuffer.fire() ), 758 ("miss_0_miss_1 ", f3_perf_info.miss_0_miss_1 && io.toIbuffer.fire() ), 759 ) 760 generatePerfEvent() 761 762 XSPerfAccumulate("ifu_req", io.toIbuffer.fire() ) 763 XSPerfAccumulate("ifu_miss", io.toIbuffer.fire() && !f3_hit ) 764 XSPerfAccumulate("ifu_req_cacheline_0", f3_req_0 ) 765 XSPerfAccumulate("ifu_req_cacheline_1", f3_req_1 ) 766 XSPerfAccumulate("ifu_req_cacheline_0_hit", f3_hit_0 ) 767 XSPerfAccumulate("ifu_req_cacheline_1_hit", f3_hit_1 ) 768 XSPerfAccumulate("frontendFlush", wb_redirect ) 769 XSPerfAccumulate("only_0_hit", f3_perf_info.only_0_hit && io.toIbuffer.fire() ) 770 XSPerfAccumulate("only_0_miss", f3_perf_info.only_0_miss && io.toIbuffer.fire() ) 771 XSPerfAccumulate("hit_0_hit_1", f3_perf_info.hit_0_hit_1 && io.toIbuffer.fire() ) 772 XSPerfAccumulate("hit_0_miss_1", f3_perf_info.hit_0_miss_1 && io.toIbuffer.fire() ) 773 XSPerfAccumulate("miss_0_hit_1", f3_perf_info.miss_0_hit_1 && io.toIbuffer.fire() ) 774 XSPerfAccumulate("miss_0_miss_1", f3_perf_info.miss_0_miss_1 && io.toIbuffer.fire() ) 775 XSPerfAccumulate("hit_0_except_1", f3_perf_info.hit_0_except_1 && io.toIbuffer.fire() ) 776 XSPerfAccumulate("miss_0_except_1", f3_perf_info.miss_0_except_1 && io.toIbuffer.fire() ) 777 XSPerfAccumulate("except_0", f3_perf_info.except_0 && io.toIbuffer.fire() ) 778} 779 780