xref: /XiangShan/src/main/scala/xiangshan/cache/mmu/TLB.scala (revision 8a020714df826c6ac860308700137855ecd6ba07)
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.cache.mmu
18
19import org.chipsalliance.cde.config.Parameters
20import chisel3._
21import chisel3.util._
22import difftest._
23import freechips.rocketchip.util.SRAMAnnotation
24import xiangshan._
25import utils._
26import utility._
27import xiangshan.backend.fu.{PMPChecker, PMPReqBundle, PMPConfig => XSPMPConfig}
28import xiangshan.backend.rob.RobPtr
29import xiangshan.backend.fu.util.HasCSRConst
30import freechips.rocketchip.rocket.PMPConfig
31
32/** TLB module
33  * support block request and non-block request io at the same time
34  * return paddr at next cycle, then go for pmp/pma check
35  * @param Width: The number of requestors
36  * @param Block: Blocked or not for each requestor ports
37  * @param q: TLB Parameters, like entry number, each TLB has its own parameters
38  * @param p: XiangShan Paramemters, like XLEN
39  */
40
41class TLB(Width: Int, nRespDups: Int = 1, Block: Seq[Boolean], q: TLBParameters)(implicit p: Parameters) extends TlbModule
42  with HasCSRConst
43  with HasPerfEvents
44{
45  val io = IO(new TlbIO(Width, nRespDups, q))
46
47  val req = io.requestor.map(_.req)
48  val resp = io.requestor.map(_.resp)
49  val ptw = io.ptw
50  val pmp = io.pmp
51  val refill_to_mem = io.refill_to_mem
52
53  /** Sfence.vma & Svinval
54    * Sfence.vma will 1. flush old entries 2. flush inflight 3. flush pipe
55    * Svinval will 1. flush old entries 2. flush inflight
56    * So, Svinval will not flush pipe, which means
57    * it should not drop reqs from pipe and should return right resp
58    */
59  val sfence = DelayN(io.sfence, q.fenceDelay)
60  val csr = io.csr
61  val satp = DelayN(io.csr.satp, q.fenceDelay)
62  val vsatp = DelayN(io.csr.vsatp, q.fenceDelay)
63  val hgatp = DelayN(io.csr.hgatp, q.fenceDelay)
64
65  val flush_mmu = DelayN(sfence.valid || csr.satp.changed || csr.vsatp.changed || csr.hgatp.changed, q.fenceDelay)
66  val mmu_flush_pipe = DelayN(sfence.valid && sfence.bits.flushPipe, q.fenceDelay) // for svinval, won't flush pipe
67  val flush_pipe = io.flushPipe
68  val redirect = io.redirect
69  val req_in = req
70  val req_out = req.map(a => RegEnable(a.bits, a.fire))
71  val req_out_v = (0 until Width).map(i => ValidHold(req_in(i).fire && !req_in(i).bits.kill, resp(i).fire, flush_pipe(i)))
72
73  val isHyperInst = (0 until Width).map(i => req_out_v(i) && req_out(i).hyperinst)
74
75  // ATTENTION: csr and flush from backend are delayed. csr should not be later than flush.
76  // because, csr will influence tlb behavior.
77  val ifecth = if (q.fetchi) true.B else false.B
78  val mode_tmp = if (q.useDmode) csr.priv.dmode else csr.priv.imode
79  val mode = (0 until Width).map(i => Mux(isHyperInst(i), csr.priv.spvp, mode_tmp))
80  val virt_in = csr.priv.virt
81  val virt_out = req.map(a => RegEnable(csr.priv.virt, a.fire))
82  val sum = (0 until Width).map(i => Mux(virt_out(i) || isHyperInst(i), io.csr.priv.vsum, io.csr.priv.sum))
83  val mxr = (0 until Width).map(i => Mux(virt_out(i) || isHyperInst(i), io.csr.priv.vmxr || io.csr.priv.mxr, io.csr.priv.mxr))
84  val req_in_s2xlate = (0 until Width).map(i => MuxCase(noS2xlate, Seq(
85      (!(virt_in || req_in(i).bits.hyperinst)) -> noS2xlate,
86      (csr.vsatp.mode =/= 0.U && csr.hgatp.mode =/= 0.U) -> allStage,
87      (csr.vsatp.mode === 0.U) -> onlyStage2,
88      (csr.hgatp.mode === 0.U) -> onlyStage1
89    )))
90  val req_out_s2xlate = (0 until Width).map(i => MuxCase(noS2xlate, Seq(
91    (!(virt_out(i) || isHyperInst(i))) -> noS2xlate,
92    (csr.vsatp.mode =/= 0.U && csr.hgatp.mode =/= 0.U) -> allStage,
93    (csr.vsatp.mode === 0.U) -> onlyStage2,
94    (csr.hgatp.mode === 0.U) -> onlyStage1
95  )))
96  val need_gpa = RegInit(false.B)
97  val need_gpa_robidx = Reg(new RobPtr)
98  val need_gpa_vpn = Reg(UInt(vpnLen.W))
99  val need_gpa_gvpn = Reg(UInt(ptePPNLen.W))
100  val resp_gpa_refill = RegInit(false.B)
101  val hasGpf = Wire(Vec(Width, Bool()))
102
103  val Sv39Enable = satp.mode === 8.U
104  val Sv48Enable = satp.mode === 9.U
105  val Sv39x4Enable = vsatp.mode === 8.U || hgatp.mode === 8.U
106  val Sv48x4Enable = vsatp.mode === 9.U || hgatp.mode === 9.U
107  val vmEnable = (0 until Width).map(i => !(isHyperInst(i) || virt_out(i)) && (
108    if (EnbaleTlbDebug) (Sv39Enable || Sv48Enable)
109    else (Sv39Enable || Sv48Enable) && (mode(i) < ModeM))
110  )
111  val s2xlateEnable = (0 until Width).map(i => (isHyperInst(i) || virt_out(i)) && (Sv39x4Enable || Sv48x4Enable) && (mode(i) < ModeM))
112  val portTranslateEnable = (0 until Width).map(i => (vmEnable(i) || s2xlateEnable(i)) && RegEnable(!req(i).bits.no_translate, req(i).valid))
113
114
115  val refill = ptw.resp.fire && !(ptw.resp.bits.getGpa) && !flush_mmu
116  refill_to_mem := DontCare
117  val entries = Module(new TlbStorageWrapper(Width, q, nRespDups))
118  entries.io.base_connect(sfence, csr, satp)
119  if (q.outReplace) { io.replace <> entries.io.replace }
120  for (i <- 0 until Width) {
121    entries.io.r_req_apply(io.requestor(i).req.valid, get_pn(req_in(i).bits.vaddr), i, req_in_s2xlate(i))
122    entries.io.w_apply(refill, ptw.resp.bits)
123    // TODO: RegNext enable:req.valid
124    resp(i).bits.debug.isFirstIssue := RegEnable(req(i).bits.debug.isFirstIssue, req(i).valid)
125    resp(i).bits.debug.robIdx := RegEnable(req(i).bits.debug.robIdx, req(i).valid)
126  }
127
128  // read TLB, get hit/miss, paddr, perm bits
129  val readResult = (0 until Width).map(TLBRead(_))
130  val hitVec = readResult.map(_._1)
131  val missVec = readResult.map(_._2)
132  val pmp_addr = readResult.map(_._3)
133  val perm = readResult.map(_._4)
134  val g_perm = readResult.map(_._5)
135  val pbmt = readResult.map(_._6)
136  val g_pbmt = readResult.map(_._7)
137  // check pmp use paddr (for timing optization, use pmp_addr here)
138  // check permisson
139  (0 until Width).foreach{i =>
140    when (RegNext(req(i).bits.no_translate)) {
141      pmp_check(req(i).bits.pmp_addr, req_out(i).size, req_out(i).cmd, i)
142    } .otherwise {
143      pmp_check(pmp_addr(i), req_out(i).size, req_out(i).cmd, i)
144    }
145    for (d <- 0 until nRespDups) {
146      pbmt_check(i, d, pbmt(i)(d), g_pbmt(i)(d), req_out_s2xlate(i))
147      perm_check(perm(i)(d), req_out(i).cmd, i, d, g_perm(i)(d), req_out(i).hlvx, req_out_s2xlate(i))
148    }
149    hasGpf(i) := resp(i).bits.excp(0).gpf.ld || resp(i).bits.excp(0).gpf.st || resp(i).bits.excp(0).gpf.instr
150  }
151
152  // handle block or non-block io
153  // for non-block io, just return the above result, send miss to ptw
154  // for block io, hold the request, send miss to ptw,
155  //   when ptw back, return the result
156  (0 until Width) foreach {i =>
157    if (Block(i)) handle_block(i)
158    else handle_nonblock(i)
159  }
160  io.ptw.resp.ready := true.B
161
162  /************************  main body above | method/log/perf below ****************************/
163  def TLBRead(i: Int) = {
164    val (e_hit, e_ppn, e_perm, e_g_perm, e_s2xlate, e_pbmt, e_g_pbmt) = entries.io.r_resp_apply(i)
165    val (p_hit, p_ppn, p_pbmt, p_perm, p_gvpn, p_g_pbmt, p_g_perm, p_s2xlate) = ptw_resp_bypass(get_pn(req_in(i).bits.vaddr), req_in_s2xlate(i))
166    val enable = portTranslateEnable(i)
167    val isOnlys2xlate = req_out_s2xlate(i) === onlyStage2
168    val need_gpa_vpn_hit = need_gpa_vpn === get_pn(req_out(i).vaddr)
169    val isitlb = TlbCmd.isExec(req_out(i).cmd)
170
171    when (!isitlb && need_gpa_robidx.needFlush(redirect) || isitlb && flush_pipe(i)){
172      need_gpa := false.B
173      resp_gpa_refill := false.B
174      need_gpa_vpn := 0.U
175    }.elsewhen (req_out_v(i) && !p_hit && !(resp_gpa_refill && need_gpa_vpn_hit) && !isOnlys2xlate && hasGpf(i) && need_gpa === false.B && !io.requestor(i).req_kill) {
176      need_gpa := true.B
177      need_gpa_vpn := get_pn(req_out(i).vaddr)
178      resp_gpa_refill := false.B
179      need_gpa_robidx := req_out(i).debug.robIdx
180    }.elsewhen (ptw.resp.fire && need_gpa && need_gpa_vpn === ptw.resp.bits.getVpn(need_gpa_vpn)) {
181      need_gpa_gvpn := ptw.resp.bits.s1.genPPN(need_gpa_vpn)
182      resp_gpa_refill := true.B
183    }
184
185    when (req_out_v(i) && hasGpf(i) && resp_gpa_refill && need_gpa_vpn_hit ){
186      need_gpa := false.B
187    }
188
189    TimeOutAssert(need_gpa && !resp_gpa_refill, timeOutThreshold, s"port${i} need gpa long time not refill.")
190
191    val hit = e_hit || p_hit
192    val miss = (!hit && enable) || hasGpf(i) && !p_hit && !(resp_gpa_refill && need_gpa_vpn_hit) && !isOnlys2xlate
193    hit.suggestName(s"hit_read_${i}")
194    miss.suggestName(s"miss_read_${i}")
195
196    val vaddr = SignExt(req_out(i).vaddr, PAddrBits)
197    resp(i).bits.miss := miss
198    resp(i).bits.ptwBack := ptw.resp.fire
199    resp(i).bits.memidx := RegEnable(req_in(i).bits.memidx, req_in(i).valid)
200
201    val ppn = WireInit(VecInit(Seq.fill(nRespDups)(0.U(ppnLen.W))))
202    val pbmt = WireInit(VecInit(Seq.fill(nRespDups)(0.U(ptePbmtLen.W))))
203    val perm = WireInit(VecInit(Seq.fill(nRespDups)(0.U.asTypeOf(new TlbPermBundle))))
204    val gvpn = WireInit(VecInit(Seq.fill(nRespDups)(0.U(vpnLen.W))))
205    val g_pbmt = WireInit(VecInit(Seq.fill(nRespDups)(0.U(ptePbmtLen.W))))
206    val g_perm = WireInit(VecInit(Seq.fill(nRespDups)(0.U.asTypeOf(new TlbPermBundle))))
207    val r_s2xlate = WireInit(VecInit(Seq.fill(nRespDups)(0.U(2.W))))
208    for (d <- 0 until nRespDups) {
209      ppn(d) := Mux(p_hit, p_ppn, e_ppn(d))
210      pbmt(d) := Mux(p_hit, p_pbmt, e_pbmt(d))
211      perm(d) := Mux(p_hit, p_perm, e_perm(d))
212      gvpn(d) :=  Mux(hasGpf(i), Mux(p_hit, p_gvpn, need_gpa_gvpn), 0.U)
213      g_pbmt(d) := Mux(p_hit, p_g_pbmt, e_g_pbmt(d))
214      g_perm(d) := Mux(p_hit, p_g_perm, e_g_perm(d))
215      r_s2xlate(d) := Mux(p_hit, p_s2xlate, e_s2xlate(d))
216      val paddr = Cat(ppn(d), get_off(req_out(i).vaddr))
217      val gpaddr = Cat(gvpn(d), get_off(req_out(i).vaddr))
218      resp(i).bits.paddr(d) := Mux(enable, paddr, vaddr)
219      resp(i).bits.gpaddr(d) := Mux(r_s2xlate(d) === onlyStage2, vaddr, gpaddr)
220    }
221
222    XSDebug(req_out_v(i), p"(${i.U}) hit:${hit} miss:${miss} ppn:${Hexadecimal(ppn(0))} perm:${perm(0)}\n")
223
224    val pmp_paddr = resp(i).bits.paddr(0)
225
226    (hit, miss, pmp_paddr, perm, g_perm, pbmt, g_pbmt)
227  }
228
229  def pmp_check(addr: UInt, size: UInt, cmd: UInt, idx: Int): Unit = {
230    pmp(idx).valid := resp(idx).valid
231    pmp(idx).bits.addr := addr
232    pmp(idx).bits.size := size
233    pmp(idx).bits.cmd := cmd
234  }
235
236  def pbmt_check(idx: Int, d: Int, pbmt: UInt, g_pbmt: UInt, s2xlate: UInt):Unit = {
237    val onlyS1 = s2xlate === onlyStage1 || s2xlate === noS2xlate
238    resp(idx).bits.pbmt(d) := Mux(
239      portTranslateEnable(idx),
240      Mux(onlyS1, pbmt, g_pbmt),
241      0.U
242    )
243  }
244
245  // for timing optimization, pmp check is divided into dynamic and static
246  def perm_check(perm: TlbPermBundle, cmd: UInt, idx: Int, nDups: Int, g_perm: TlbPermBundle, hlvx: Bool, s2xlate: UInt) = {
247    // dynamic: superpage (or full-connected reg entries) -> check pmp when translation done
248    // static: 4K pages (or sram entries) -> check pmp with pre-checked results
249    val hasS2xlate = s2xlate =/= noS2xlate
250    val onlyS1 = s2xlate === onlyStage1
251    val onlyS2 = s2xlate === onlyStage2
252    val af = perm.af || (hasS2xlate && g_perm.af)
253
254    // Stage 1 perm check
255    val pf = perm.pf
256    val ldUpdate = !perm.a && TlbCmd.isRead(cmd) && !TlbCmd.isAmo(cmd) // update A/D through exception
257    val stUpdate = (!perm.a || !perm.d) && (TlbCmd.isWrite(cmd) || TlbCmd.isAmo(cmd)) // update A/D through exception
258    val instrUpdate = !perm.a && TlbCmd.isExec(cmd) // update A/D through exception
259    val modeCheck = !(mode(idx) === ModeU && !perm.u || mode(idx) === ModeS && perm.u && (!sum(idx) || ifecth))
260    val ldPermFail = !(modeCheck && Mux(hlvx, perm.x, perm.r || mxr(idx) && perm.x))
261    val stPermFail = !(modeCheck && perm.w)
262    val instrPermFail = !(modeCheck && perm.x)
263    val ldPf = (ldPermFail || pf) && (TlbCmd.isRead(cmd) && !TlbCmd.isAmo(cmd))
264    val stPf = (stPermFail || pf) && (TlbCmd.isWrite(cmd) || TlbCmd.isAmo(cmd))
265    val instrPf = (instrPermFail || pf) && TlbCmd.isExec(cmd)
266    val s1_valid = portTranslateEnable(idx) && !onlyS2
267
268    // Stage 2 perm check
269    val gpf = g_perm.pf
270    val g_ldUpdate = !g_perm.a && TlbCmd.isRead(cmd) && !TlbCmd.isAmo(cmd)
271    val g_stUpdate = (!g_perm.a || !g_perm.d) && (TlbCmd.isWrite(cmd) || TlbCmd.isAmo(cmd))
272    val g_instrUpdate = !g_perm.a && TlbCmd.isExec(cmd)
273    val g_ldPermFail = !Mux(hlvx, g_perm.x, (g_perm.r || io.csr.priv.mxr && g_perm.x))
274    val g_stPermFail = !g_perm.w
275    val g_instrPermFail = !g_perm.x
276    val ldGpf = (g_ldPermFail || gpf) && (TlbCmd.isRead(cmd) && !TlbCmd.isAmo(cmd))
277    val stGpf = (g_stPermFail || gpf) && (TlbCmd.isWrite(cmd) || TlbCmd.isAmo(cmd))
278    val instrGpf = (g_instrPermFail || gpf) && TlbCmd.isExec(cmd)
279    val s2_valid = hasS2xlate && !onlyS1 && portTranslateEnable(idx)
280
281    val fault_valid = s1_valid || s2_valid
282
283    // when pf and gpf can't happens simultaneously
284    val hasPf = (ldPf || ldUpdate || stPf || stUpdate || instrPf || instrUpdate) && s1_valid && !af
285    resp(idx).bits.excp(nDups).pf.ld := (ldPf || ldUpdate) && s1_valid && !af
286    resp(idx).bits.excp(nDups).pf.st := (stPf || stUpdate) && s1_valid && !af
287    resp(idx).bits.excp(nDups).pf.instr := (instrPf || instrUpdate) && s1_valid && !af
288    // NOTE: pf need && with !af, page fault has higher priority than access fault
289    // but ptw may also have access fault, then af happens, the translation is wrong.
290    // In this case, pf has lower priority than af
291
292    resp(idx).bits.excp(nDups).gpf.ld := (ldGpf || g_ldUpdate) && s2_valid && !af && !hasPf
293    resp(idx).bits.excp(nDups).gpf.st := (stGpf || g_stUpdate) && s2_valid && !af && !hasPf
294    resp(idx).bits.excp(nDups).gpf.instr := (instrGpf || g_instrUpdate) && s2_valid && !af && !hasPf
295
296    resp(idx).bits.excp(nDups).af.ld    := af && TlbCmd.isRead(cmd) && fault_valid
297    resp(idx).bits.excp(nDups).af.st    := af && TlbCmd.isWrite(cmd) && fault_valid
298    resp(idx).bits.excp(nDups).af.instr := af && TlbCmd.isExec(cmd) && fault_valid
299
300
301  }
302
303  def handle_nonblock(idx: Int): Unit = {
304    io.requestor(idx).resp.valid := req_out_v(idx)
305    io.requestor(idx).req.ready := io.requestor(idx).resp.ready // should always be true
306    XSError(!io.requestor(idx).resp.ready, s"${q.name} port ${idx} is non-block, resp.ready must be true.B")
307
308    val req_need_gpa = hasGpf(idx)
309    val req_s2xlate = Wire(UInt(2.W))
310    req_s2xlate := MuxCase(noS2xlate, Seq(
311      (!(virt_out(idx) || req_out(idx).hyperinst)) -> noS2xlate,
312      (csr.vsatp.mode =/= 0.U && csr.hgatp.mode =/= 0.U) -> allStage,
313      (csr.vsatp.mode === 0.U) -> onlyStage2,
314      (csr.hgatp.mode === 0.U || req_need_gpa) -> onlyStage1
315    ))
316
317    val ptw_just_back = ptw.resp.fire && req_s2xlate === ptw.resp.bits.s2xlate && ptw.resp.bits.hit(get_pn(req_out(idx).vaddr), io.csr.satp.asid, io.csr.vsatp.asid, io.csr.hgatp.vmid, true, false)
318    // TODO: RegNext enable: ptw.resp.valid ? req.valid
319    val ptw_resp_bits_reg = RegEnable(ptw.resp.bits, ptw.resp.valid)
320    val ptw_already_back = GatedValidRegNext(ptw.resp.fire) && req_s2xlate === ptw_resp_bits_reg.s2xlate && ptw_resp_bits_reg.hit(get_pn(req_out(idx).vaddr), io.csr.satp.asid, io.csr.vsatp.asid, io.csr.hgatp.vmid, allType = true)
321    io.ptw.req(idx).valid := req_out_v(idx) && missVec(idx) && !(ptw_just_back || ptw_already_back) // TODO: remove the regnext, timing
322    io.tlbreplay(idx) := req_out_v(idx) && missVec(idx) && (ptw_just_back || ptw_already_back)
323    when (io.requestor(idx).req_kill && GatedValidRegNext(io.requestor(idx).req.fire)) {
324      io.ptw.req(idx).valid := false.B
325      io.tlbreplay(idx) := true.B
326    }
327    io.ptw.req(idx).bits.vpn := get_pn(req_out(idx).vaddr)
328    io.ptw.req(idx).bits.s2xlate := req_s2xlate
329    io.ptw.req(idx).bits.getGpa := req_need_gpa && hitVec(idx)
330    io.ptw.req(idx).bits.memidx := req_out(idx).memidx
331  }
332
333  def handle_block(idx: Int): Unit = {
334    // three valid: 1.if exist a entry; 2.if sent to ptw; 3.unset resp.valid
335    io.requestor(idx).req.ready := !req_out_v(idx) || io.requestor(idx).resp.fire
336    // req_out_v for if there is a request, may long latency, fixme
337
338    // miss request entries
339    val req_need_gpa = hasGpf(idx)
340    val miss_req_vpn = get_pn(req_out(idx).vaddr)
341    val miss_req_memidx = req_out(idx).memidx
342    val miss_req_s2xlate = Wire(UInt(2.W))
343    miss_req_s2xlate := MuxCase(noS2xlate, Seq(
344      (!(virt_out(idx) || req_out(idx).hyperinst)) -> noS2xlate,
345      (csr.vsatp.mode =/= 0.U && csr.hgatp.mode =/= 0.U) -> allStage,
346      (csr.vsatp.mode === 0.U) -> onlyStage2,
347      (csr.hgatp.mode === 0.U || req_need_gpa) -> onlyStage1
348    ))
349    val miss_req_s2xlate_reg = RegEnable(miss_req_s2xlate, io.ptw.req(idx).fire)
350    val hasS2xlate = miss_req_s2xlate_reg =/= noS2xlate
351    val onlyS2 = miss_req_s2xlate_reg === onlyStage2
352    val hit_s1 = io.ptw.resp.bits.s1.hit(miss_req_vpn, Mux(hasS2xlate, io.csr.vsatp.asid, io.csr.satp.asid), io.csr.hgatp.vmid, allType = true, false, hasS2xlate)
353    val hit_s2 = io.ptw.resp.bits.s2.hit(miss_req_vpn, io.csr.hgatp.vmid)
354    val hit = Mux(onlyS2, hit_s2, hit_s1) && io.ptw.resp.valid && miss_req_s2xlate_reg === io.ptw.resp.bits.s2xlate
355
356    val new_coming_valid = WireInit(false.B)
357    new_coming_valid := req_in(idx).fire && !req_in(idx).bits.kill && !flush_pipe(idx)
358    val new_coming = GatedValidRegNext(new_coming_valid)
359    val miss_wire = new_coming && missVec(idx)
360    val miss_v = ValidHoldBypass(miss_wire, resp(idx).fire, flush_pipe(idx))
361    val miss_req_v = ValidHoldBypass(miss_wire || (miss_v && flush_mmu && !mmu_flush_pipe),
362      io.ptw.req(idx).fire || resp(idx).fire, flush_pipe(idx))
363
364    // when ptw resp, check if hit, reset miss_v, resp to lsu/ifu
365    resp(idx).valid := req_out_v(idx) && !(miss_v && portTranslateEnable(idx))
366    when (io.ptw.resp.fire && hit && req_out_v(idx) && portTranslateEnable(idx)) {
367      val stage1 = io.ptw.resp.bits.s1
368      val stage2 = io.ptw.resp.bits.s2
369      val s2xlate = io.ptw.resp.bits.s2xlate
370      resp(idx).valid := true.B
371      resp(idx).bits.miss := false.B
372      val s1_paddr = Cat(stage1.genPPN(get_pn(req_out(idx).vaddr)), get_off(req_out(idx).vaddr))
373      val s2_paddr = Cat(stage2.genPPNS2(get_pn(req_out(idx).vaddr)), get_off(req_out(idx).vaddr))
374      for (d <- 0 until nRespDups) {
375        resp(idx).bits.paddr(d) := Mux(s2xlate =/= noS2xlate, s2_paddr, s1_paddr)
376        resp(idx).bits.gpaddr(d) := s1_paddr
377        pbmt_check(idx, d, io.ptw.resp.bits.s1.entry.pbmt, io.ptw.resp.bits.s2.entry.pbmt, s2xlate)
378        perm_check(stage1, req_out(idx).cmd, idx, d, stage2, req_out(idx).hlvx, s2xlate)
379      }
380      pmp_check(resp(idx).bits.paddr(0), req_out(idx).size, req_out(idx).cmd, idx)
381
382      // NOTE: the unfiltered req would be handled by Repeater
383    }
384    assert(RegNext(!resp(idx).valid || resp(idx).ready, true.B), "when tlb resp valid, ready should be true, must")
385    assert(RegNext(req_out_v(idx) || !(miss_v || miss_req_v), true.B), "when not req_out_v, should not set miss_v/miss_req_v")
386
387    val ptw_req = io.ptw.req(idx)
388    ptw_req.valid := miss_req_v
389    ptw_req.bits.vpn := miss_req_vpn
390    ptw_req.bits.s2xlate := miss_req_s2xlate
391    ptw_req.bits.getGpa := req_need_gpa && hitVec(idx)
392    ptw_req.bits.memidx := miss_req_memidx
393
394    io.tlbreplay(idx) := false.B
395
396    // NOTE: when flush pipe, tlb should abandon last req
397    // however, some outside modules like icache, dont care flushPipe, and still waiting for tlb resp
398    // just resp valid and raise page fault to go through. The pipe(ifu) will abandon it.
399    if (!q.outsideRecvFlush) {
400      when (req_out_v(idx) && flush_pipe(idx) && portTranslateEnable(idx)) {
401        resp(idx).valid := true.B
402        for (d <- 0 until nRespDups) {
403          resp(idx).bits.pbmt(d) := 0.U
404          resp(idx).bits.excp(d).pf.ld := true.B // sfence happened, pf for not to use this addr
405          resp(idx).bits.excp(d).pf.st := true.B
406          resp(idx).bits.excp(d).pf.instr := true.B
407        }
408      }
409    }
410  }
411
412  // when ptw resp, tlb at refill_idx maybe set to miss by force.
413  // Bypass ptw resp to check.
414  def ptw_resp_bypass(vpn: UInt, s2xlate: UInt) = {
415    // TODO: RegNext enable: ptw.resp.valid
416    val hasS2xlate = s2xlate =/= noS2xlate
417    val onlyS2 = s2xlate === onlyStage2
418    val onlyS1 = s2xlate === onlyStage1
419    val s2xlate_hit = s2xlate === ptw.resp.bits.s2xlate
420    val resp_hit = ptw.resp.bits.hit(vpn, io.csr.satp.asid, io.csr.vsatp.asid, io.csr.hgatp.vmid, true, false)
421    val p_hit = GatedValidRegNext(resp_hit && io.ptw.resp.fire && s2xlate_hit)
422    val ppn_s1 = ptw.resp.bits.s1.genPPN(vpn)
423    val gvpn = Mux(onlyS2, vpn, ppn_s1)
424    val ppn_s2 = ptw.resp.bits.s2.genPPNS2(gvpn)
425    val p_ppn = RegEnable(Mux(hasS2xlate, ppn_s2, ppn_s1), io.ptw.resp.fire)
426    val p_pbmt = RegEnable(ptw.resp.bits.s1.entry.pbmt,io.ptw.resp.fire)
427    val p_perm = RegEnable(ptwresp_to_tlbperm(ptw.resp.bits.s1), io.ptw.resp.fire)
428    val p_gvpn = RegEnable(Mux(onlyS2, ptw.resp.bits.s2.entry.tag, ppn_s1), io.ptw.resp.fire)
429    val p_g_pbmt = RegEnable(ptw.resp.bits.s2.entry.pbmt,io.ptw.resp.fire)
430    val p_g_perm = RegEnable(hptwresp_to_tlbperm(ptw.resp.bits.s2), io.ptw.resp.fire)
431    val p_s2xlate = RegEnable(ptw.resp.bits.s2xlate, io.ptw.resp.fire)
432    (p_hit, p_ppn, p_pbmt, p_perm, p_gvpn, p_g_pbmt, p_g_perm, p_s2xlate)
433  }
434
435  // assert
436  for(i <- 0 until Width) {
437    TimeOutAssert(req_out_v(i) && !resp(i).valid, timeOutThreshold, s"{q.name} port{i} long time no resp valid.")
438  }
439
440  // perf event
441  val result_ok = req_in.map(a => GatedValidRegNext(a.fire))
442  val perfEvents =
443    Seq(
444      ("access", PopCount((0 until Width).map{i => if (Block(i)) io.requestor(i).req.fire else portTranslateEnable(i) && result_ok(i) })),
445      ("miss  ", PopCount((0 until Width).map{i => if (Block(i)) portTranslateEnable(i) && result_ok(i) && missVec(i) else ptw.req(i).fire })),
446    )
447  generatePerfEvent()
448
449  // perf log
450  for (i <- 0 until Width) {
451    if (Block(i)) {
452      XSPerfAccumulate(s"access${i}",result_ok(i) && portTranslateEnable(i))
453      XSPerfAccumulate(s"miss${i}", result_ok(i) && missVec(i))
454    } else {
455      XSPerfAccumulate("first_access" + Integer.toString(i, 10), result_ok(i) && portTranslateEnable(i) && RegEnable(req(i).bits.debug.isFirstIssue, req(i).valid))
456      XSPerfAccumulate("access" + Integer.toString(i, 10), result_ok(i) && portTranslateEnable(i))
457      XSPerfAccumulate("first_miss" + Integer.toString(i, 10), result_ok(i) && portTranslateEnable(i) && missVec(i) && RegEnable(req(i).bits.debug.isFirstIssue, req(i).valid))
458      XSPerfAccumulate("miss" + Integer.toString(i, 10), result_ok(i) && portTranslateEnable(i) && missVec(i))
459    }
460  }
461  XSPerfAccumulate("ptw_resp_count", ptw.resp.fire)
462  XSPerfAccumulate("ptw_resp_pf_count", ptw.resp.fire && ptw.resp.bits.s1.pf)
463
464  // Log
465  for(i <- 0 until Width) {
466    XSDebug(req(i).valid, p"req(${i.U}): (${req(i).valid} ${req(i).ready}) ${req(i).bits}\n")
467    XSDebug(resp(i).valid, p"resp(${i.U}): (${resp(i).valid} ${resp(i).ready}) ${resp(i).bits}\n")
468  }
469
470  XSDebug(io.sfence.valid, p"Sfence: ${io.sfence}\n")
471  XSDebug(ParallelOR(req_out_v) || ptw.resp.valid, p"vmEnable:${vmEnable} hit:${Binary(VecInit(hitVec).asUInt)} miss:${Binary(VecInit(missVec).asUInt)}\n")
472  for (i <- ptw.req.indices) {
473    XSDebug(ptw.req(i).fire, p"L2TLB req:${ptw.req(i).bits}\n")
474  }
475  XSDebug(ptw.resp.valid, p"L2TLB resp:${ptw.resp.bits} (v:${ptw.resp.valid}r:${ptw.resp.ready}) \n")
476
477  println(s"${q.name}: page: ${q.NWays} ${q.Associative} ${q.Replacer.get}")
478
479  if (env.EnableDifftest) {
480    for (i <- 0 until Width) {
481      val pf = io.requestor(i).resp.bits.excp(0).pf.instr || io.requestor(i).resp.bits.excp(0).pf.st || io.requestor(i).resp.bits.excp(0).pf.ld
482      val gpf = io.requestor(i).resp.bits.excp(0).gpf.instr || io.requestor(i).resp.bits.excp(0).gpf.st || io.requestor(i).resp.bits.excp(0).gpf.ld
483      val af = io.requestor(i).resp.bits.excp(0).af.instr || io.requestor(i).resp.bits.excp(0).af.st || io.requestor(i).resp.bits.excp(0).af.ld
484      val difftest = DifftestModule(new DiffL1TLBEvent)
485      difftest.coreid := io.hartId
486      difftest.valid := RegNext(io.requestor(i).req.fire) && !io.requestor(i).req_kill && io.requestor(i).resp.fire && !io.requestor(i).resp.bits.miss && !pf && !af && !gpf && portTranslateEnable(i)
487      if (!Seq("itlb", "ldtlb", "sttlb").contains(q.name)) {
488        difftest.valid := false.B
489      }
490      difftest.index := TLBDiffId(p(XSCoreParamsKey).HartId).U
491      difftest.vpn := RegEnable(get_pn(req_in(i).bits.vaddr), req_in(i).valid)
492      difftest.ppn := get_pn(io.requestor(i).resp.bits.paddr(0))
493      difftest.satp := Cat(io.csr.satp.mode, io.csr.satp.asid, io.csr.satp.ppn)
494      difftest.vsatp := Cat(io.csr.vsatp.mode, io.csr.vsatp.asid, io.csr.vsatp.ppn)
495      difftest.hgatp := Cat(io.csr.hgatp.mode, io.csr.hgatp.vmid, io.csr.hgatp.ppn)
496      val req_need_gpa = gpf
497      val req_s2xlate = Wire(UInt(2.W))
498      req_s2xlate := MuxCase(noS2xlate, Seq(
499        (!RegNext(virt_in || req_in(i).bits.hyperinst)) -> noS2xlate,
500        (vsatp.mode =/= 0.U && hgatp.mode =/= 0.U) -> allStage,
501        (vsatp.mode === 0.U) -> onlyStage2,
502        (hgatp.mode === 0.U || req_need_gpa) -> onlyStage1
503      ))
504      difftest.s2xlate := req_s2xlate
505    }
506  }
507}
508
509object TLBDiffId {
510  var i: Int = 0
511  var lastHartId: Int = -1
512  def apply(hartId: Int): Int = {
513    if (lastHartId != hartId) {
514      i = 0
515      lastHartId = hartId
516    }
517    i += 1
518    i - 1
519  }
520}
521
522class TLBNonBlock(Width: Int, nRespDups: Int = 1, q: TLBParameters)(implicit p: Parameters) extends TLB(Width, nRespDups, Seq.fill(Width)(false), q)
523class TLBBLock(Width: Int, nRespDups: Int = 1, q: TLBParameters)(implicit p: Parameters) extends TLB(Width, nRespDups, Seq.fill(Width)(true), q)
524
525class TlbReplace(Width: Int, q: TLBParameters)(implicit p: Parameters) extends TlbModule {
526  val io = IO(new TlbReplaceIO(Width, q))
527
528  if (q.Associative == "fa") {
529    val re = ReplacementPolicy.fromString(q.Replacer, q.NWays)
530    re.access(io.page.access.map(_.touch_ways))
531    io.page.refillIdx := re.way
532  } else { // set-acco && plru
533    val re = ReplacementPolicy.fromString(q.Replacer, q.NSets, q.NWays)
534    re.access(io.page.access.map(_.sets), io.page.access.map(_.touch_ways))
535    io.page.refillIdx := { if (q.NWays == 1) 0.U else re.way(io.page.chosen_set) }
536  }
537}
538