xref: /XiangShan/src/main/scala/xiangshan/cache/mmu/PageTableWalker.scala (revision 7ab45173575ae360e1987ddf002d0241f690da81)
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 xiangshan._
23import xiangshan.cache.{HasDCacheParameters, MemoryOpConstants}
24import utils._
25import utility._
26import freechips.rocketchip.diplomacy.{LazyModule, LazyModuleImp}
27import freechips.rocketchip.tilelink._
28import xiangshan.backend.fu.{PMPReqBundle, PMPRespBundle}
29
30/** Page Table Walk is divided into two parts
31  * One,   PTW: page walk for pde, except for leaf entries, one by one
32  * Two, LLPTW: page walk for pte, only the leaf entries(4KB), in parallel
33  */
34
35
36/** PTW : page table walker
37  * a finite state machine
38  * only take 1GB and 2MB page walks
39  * or in other words, except the last level(leaf)
40  **/
41class PTWIO()(implicit p: Parameters) extends MMUIOBaseBundle with HasPtwConst {
42  val req = Flipped(DecoupledIO(new Bundle {
43    val req_info = new L2TlbInnerBundle()
44    val l3Hit = if (EnableSv48) Some(new Bool()) else None
45    val l2Hit = Bool()
46    val ppn = UInt(ptePPNLen.W)
47    val stage1Hit = Bool()
48    val stage1 = new PtwMergeResp
49  }))
50  val resp = DecoupledIO(new Bundle {
51    val source = UInt(bSourceWidth.W)
52    val s2xlate = UInt(2.W)
53    val resp = new PtwMergeResp
54    val h_resp = new HptwResp
55  })
56
57  val llptw = DecoupledIO(new LLPTWInBundle())
58  // NOTE: llptw change from "connect to llptw" to "connect to page cache"
59  // to avoid corner case that caused duplicate entries
60
61  val hptw = new Bundle {
62    val req = DecoupledIO(new Bundle {
63      val source = UInt(bSourceWidth.W)
64      val id = UInt(log2Up(l2tlbParams.llptwsize).W)
65      val gvpn = UInt(ptePPNLen.W)
66    })
67    val resp = Flipped(Valid(new Bundle {
68      val h_resp = Output(new HptwResp)
69    }))
70  }
71  val mem = new Bundle {
72    val req = DecoupledIO(new L2TlbMemReqBundle())
73    val resp = Flipped(ValidIO(UInt(XLEN.W)))
74    val mask = Input(Bool())
75  }
76  val pmp = new Bundle {
77    val req = ValidIO(new PMPReqBundle())
78    val resp = Flipped(new PMPRespBundle())
79  }
80
81  val refill = Output(new Bundle {
82    val req_info = new L2TlbInnerBundle()
83    val level = UInt(log2Up(Level + 1).W)
84  })
85}
86
87class PTW()(implicit p: Parameters) extends XSModule with HasPtwConst with HasPerfEvents {
88  val io = IO(new PTWIO)
89  val sfence = io.sfence
90  val mem = io.mem
91  val req_s2xlate = Reg(UInt(2.W))
92  val enableS2xlate = req_s2xlate =/= noS2xlate
93  val onlyS1xlate = req_s2xlate === onlyStage1
94  val onlyS2xlate = req_s2xlate === onlyStage2
95
96  val satp = Wire(new TlbSatpBundle())
97  when (io.req.fire) {
98    satp := Mux(io.req.bits.req_info.s2xlate =/= noS2xlate, io.csr.vsatp, io.csr.satp)
99  } .otherwise {
100    satp := Mux(enableS2xlate, io.csr.vsatp, io.csr.satp)
101  }
102
103  val mode = satp.mode
104  val hgatp = io.csr.hgatp
105  val flush = io.sfence.valid || io.csr.satp.changed || io.csr.vsatp.changed || io.csr.hgatp.changed
106  val s2xlate = enableS2xlate && !onlyS1xlate
107  val level = RegInit(3.U(log2Up(Level + 1).W))
108  val af_level = RegInit(3.U(log2Up(Level + 1).W)) // access fault return this level
109  val gpf_level = RegInit(3.U(log2Up(Level + 1).W))
110  val ppn = Reg(UInt(ptePPNLen.W))
111  val vpn = Reg(UInt(vpnLen.W)) // vpn or gvpn(onlyS2xlate)
112  val levelNext = level - 1.U
113  val l3Hit = Reg(Bool())
114  val l2Hit = Reg(Bool())
115  val pte = mem.resp.bits.asTypeOf(new PteBundle())
116
117  // s/w register
118  val s_pmp_check = RegInit(true.B)
119  val s_mem_req = RegInit(true.B)
120  val s_llptw_req = RegInit(true.B)
121  val w_mem_resp = RegInit(true.B)
122  val s_hptw_req = RegInit(true.B)
123  val w_hptw_resp = RegInit(true.B)
124  val s_last_hptw_req = RegInit(true.B)
125  val w_last_hptw_resp = RegInit(true.B)
126  // for updating "level"
127  val mem_addr_update = RegInit(false.B)
128
129  val idle = RegInit(true.B)
130  val finish = WireInit(false.B)
131  val sent_to_pmp = idle === false.B && (s_pmp_check === false.B || mem_addr_update) && !finish
132
133  val pageFault = pte.isPf(level)
134  val accessFault = RegEnable(io.pmp.resp.ld || io.pmp.resp.mmio, false.B, sent_to_pmp)
135
136  val hptw_pageFault = RegInit(false.B)
137  val hptw_accessFault = RegInit(false.B)
138  val last_s2xlate = RegInit(false.B)
139  val stage1Hit = RegEnable(io.req.bits.stage1Hit, io.req.fire)
140  val stage1 = RegEnable(io.req.bits.stage1, io.req.fire)
141  val hptw_resp_stage2 = Reg(Bool())
142
143  val ppn_af = Mux(enableS2xlate, Mux(onlyS1xlate, pte.isAf() && !pte.isStage1Gpf(io.csr.vsatp.mode), false.B), pte.isAf()) // In two-stage address translation, stage 1 ppn is a vpn for host, so don't need to check ppn_high
144  val find_pte = pte.isLeaf() || ppn_af || pageFault
145  val to_find_pte = level === 1.U && find_pte === false.B
146  val source = RegEnable(io.req.bits.req_info.source, io.req.fire)
147
148  val l3addr = Wire(UInt(PAddrBits.W))
149  val l2addr = Wire(UInt(PAddrBits.W))
150  val l1addr = Wire(UInt(PAddrBits.W))
151  val mem_addr = Wire(UInt(PAddrBits.W))
152
153  l3addr := MakeAddr(satp.ppn, getVpnn(vpn, 3))
154  if (EnableSv48) {
155    when (mode === Sv48) {
156      l2addr := MakeAddr(Mux(l3Hit, ppn, pte.getPPN()), getVpnn(vpn, 2))
157    } .otherwise {
158      l2addr := MakeAddr(satp.ppn, getVpnn(vpn, 2))
159    }
160  } else {
161    l2addr := MakeAddr(satp.ppn, getVpnn(vpn, 2))
162  }
163  l1addr := MakeAddr(Mux(l2Hit, ppn, pte.getPPN()), getVpnn(vpn, 1))
164  mem_addr := Mux(af_level === 3.U, l3addr, Mux(af_level === 2.U, l2addr, l1addr))
165
166  val hptw_resp = Reg(new HptwResp)
167  val gpaddr = MuxCase(mem_addr, Seq(
168    stage1Hit -> Cat(stage1.genPPN(), 0.U(offLen.W)),
169    onlyS2xlate -> Cat(vpn, 0.U(offLen.W)),
170    !s_last_hptw_req -> Cat(MuxLookup(level, pte.getPPN())(Seq(
171      3.U -> Cat(pte.getPPN()(ptePPNLen - 1, vpnnLen * 3), vpn(vpnnLen * 3 - 1, 0)),
172      2.U -> Cat(pte.getPPN()(ptePPNLen - 1, vpnnLen * 2), vpn(vpnnLen * 2 - 1, 0)),
173      1.U -> Cat(pte.getPPN()(ptePPNLen - 1, vpnnLen), vpn(vpnnLen - 1, 0)
174    ))),
175    0.U(offLen.W))
176  ))
177  val gvpn_gpf = Mux(enableS2xlate && io.csr.hgatp.mode === Sv39x4, gpaddr(gpaddr.getWidth - 1, GPAddrBitsSv39x4) =/= 0.U, Mux(enableS2xlate && io.csr.hgatp.mode === Sv48x4, gpaddr(gpaddr.getWidth - 1, GPAddrBitsSv48x4) =/= 0.U, false.B))
178  val check_g_perm_fail = RegInit(false.B)
179  val guestFault = hptw_pageFault || hptw_accessFault || check_g_perm_fail || gvpn_gpf
180  val hpaddr = Cat(hptw_resp.genPPNS2(get_pn(gpaddr)), get_off(gpaddr))
181  val fake_h_resp = 0.U.asTypeOf(new HptwResp)
182  fake_h_resp.entry.tag := get_pn(gpaddr)
183  fake_h_resp.entry.vmid.map(_ := io.csr.hgatp.vmid)
184  fake_h_resp.gpf := true.B
185
186  val pte_valid = RegInit(false.B)  // avoid l1tlb pf from stage1 when gpf happens in the first s2xlate in PTW
187  val fake_pte = 0.U.asTypeOf(new PteBundle())
188  fake_pte.perm.v := false.B // tell L1TLB this is fake pte
189  fake_pte.perm.r := true.B
190  fake_pte.perm.w := true.B
191  fake_pte.perm.x := true.B
192  fake_pte.perm.a := true.B
193  fake_pte.perm.d := true.B
194  fake_pte.ppn := ppn(ppnLen - 1, 0)
195  fake_pte.ppn_high := ppn(ptePPNLen - 1, ppnLen)
196
197  io.req.ready := idle
198  val ptw_resp = Wire(new PtwMergeResp)
199  ptw_resp.apply(Mux(pte_valid, pageFault && !accessFault && !ppn_af, false.B), accessFault || ppn_af, Mux(accessFault, af_level, Mux(guestFault, gpf_level, level)), Mux(pte_valid, pte, fake_pte), vpn, satp.asid, hgatp.vmid, vpn(sectortlbwidth - 1, 0), not_super = false)
200
201  val normal_resp = idle === false.B && mem_addr_update && !last_s2xlate && (guestFault || (w_mem_resp && find_pte) || (s_pmp_check && accessFault) || onlyS2xlate )
202  val stageHit_resp = idle === false.B && hptw_resp_stage2
203  io.resp.valid := Mux(stage1Hit, stageHit_resp, normal_resp)
204  io.resp.bits.source := source
205  io.resp.bits.resp := Mux(stage1Hit || (l3Hit || l2Hit) && guestFault && !pte_valid, stage1, ptw_resp)
206  io.resp.bits.h_resp := Mux(gvpn_gpf, fake_h_resp, hptw_resp)
207  io.resp.bits.s2xlate := req_s2xlate
208
209  io.llptw.valid := s_llptw_req === false.B && to_find_pte && !accessFault && !guestFault
210  io.llptw.bits.req_info.source := source
211  io.llptw.bits.req_info.vpn := vpn
212  io.llptw.bits.req_info.s2xlate := req_s2xlate
213  io.llptw.bits.ppn := DontCare
214
215  io.pmp.req.valid := DontCare // samecycle, do not use valid
216  io.pmp.req.bits.addr := Mux(s2xlate, hpaddr, mem_addr)
217  io.pmp.req.bits.size := 3.U // TODO: fix it
218  io.pmp.req.bits.cmd := TlbCmd.read
219
220  mem.req.valid := s_mem_req === false.B && !mem.mask && !accessFault && s_pmp_check
221  mem.req.bits.addr := Mux(s2xlate, hpaddr, mem_addr)
222  mem.req.bits.id := FsmReqID.U(bMemID.W)
223  mem.req.bits.hptw_bypassed := false.B
224
225  io.refill.req_info.s2xlate := req_s2xlate
226  io.refill.req_info.vpn := vpn
227  io.refill.level := level
228  io.refill.req_info.source := source
229
230  io.hptw.req.valid := !s_hptw_req || !s_last_hptw_req
231  io.hptw.req.bits.id := FsmReqID.U(bMemID.W)
232  io.hptw.req.bits.gvpn := get_pn(gpaddr)
233  io.hptw.req.bits.source := source
234
235  when (io.req.fire && io.req.bits.stage1Hit){
236    idle := false.B
237    req_s2xlate := io.req.bits.req_info.s2xlate
238    s_last_hptw_req := false.B
239    hptw_resp_stage2 := false.B
240    last_s2xlate := false.B
241    check_g_perm_fail := false.B
242    hptw_pageFault := false.B
243    hptw_accessFault := false.B
244  }
245
246  when (io.resp.fire && stage1Hit){
247    idle := true.B
248  }
249
250  when (io.req.fire && !io.req.bits.stage1Hit){
251    val req = io.req.bits
252    if (EnableSv48) {
253      when (mode === Sv48) {
254        level := Mux(req.l2Hit, 1.U, Mux(req.l3Hit.get, 2.U, 3.U))
255        af_level := Mux(req.l2Hit, 1.U, Mux(req.l3Hit.get, 2.U, 3.U))
256        gpf_level := Mux(req.l2Hit, 2.U, Mux(req.l3Hit.get, 3.U, 0.U))
257        ppn := Mux(req.l2Hit || req.l3Hit.get, io.req.bits.ppn, satp.ppn)
258        l3Hit := req.l3Hit.get
259      } .otherwise {
260        level := Mux(req.l2Hit, 1.U, 2.U)
261        af_level := Mux(req.l2Hit, 1.U, 2.U)
262        gpf_level := 0.U
263        ppn := Mux(req.l2Hit, io.req.bits.ppn, satp.ppn)
264        l3Hit := false.B
265      }
266    } else {
267      level := Mux(req.l2Hit, 1.U, 2.U)
268      af_level := Mux(req.l2Hit, 1.U, 2.U)
269      gpf_level := 0.U
270      ppn := Mux(req.l2Hit, io.req.bits.ppn, satp.ppn)
271      l3Hit := false.B
272    }
273    vpn := io.req.bits.req_info.vpn
274    l2Hit := req.l2Hit
275    accessFault := false.B
276    idle := false.B
277    hptw_pageFault := false.B
278    hptw_accessFault := false.B
279    pte_valid := false.B
280    check_g_perm_fail := false.B
281    req_s2xlate := io.req.bits.req_info.s2xlate
282    when(io.req.bits.req_info.s2xlate === onlyStage2){
283      val onlys2_gpaddr = Cat(io.req.bits.req_info.vpn, 0.U(offLen.W)) // is 50 bits, don't need to check high bits when sv48x4 is enabled
284      val check_gpa_high_fail = Mux(io.req.bits.req_info.s2xlate === onlyStage2 && io.csr.hgatp.mode === Sv39x4, onlys2_gpaddr(onlys2_gpaddr.getWidth - 1, GPAddrBitsSv39x4) =/= 0.U, false.B)
285      last_s2xlate := false.B
286      when(check_gpa_high_fail){
287        mem_addr_update := true.B
288      }.otherwise{
289        s_last_hptw_req := false.B
290      }
291    }.elsewhen(io.req.bits.req_info.s2xlate === allStage){
292      last_s2xlate := true.B
293      s_hptw_req := false.B
294    }.otherwise {
295      last_s2xlate := false.B
296      s_pmp_check := false.B
297    }
298  }
299
300  when(io.hptw.req.fire && s_hptw_req === false.B){
301    s_hptw_req := true.B
302    w_hptw_resp := false.B
303  }
304
305  when(io.hptw.resp.fire && w_hptw_resp === false.B) {
306    hptw_pageFault := io.hptw.resp.bits.h_resp.gpf
307    hptw_accessFault := io.hptw.resp.bits.h_resp.gaf
308    hptw_resp := io.hptw.resp.bits.h_resp
309    w_hptw_resp := true.B
310    val g_perm_fail = !io.hptw.resp.bits.h_resp.entry.perm.get.r && !(io.csr.priv.mxr && io.hptw.resp.bits.h_resp.entry.perm.get.x)
311    check_g_perm_fail := g_perm_fail
312    when(!(g_perm_fail || io.hptw.resp.bits.h_resp.gpf || io.hptw.resp.bits.h_resp.gaf)) {
313      s_pmp_check := false.B
314    }
315  }
316
317  when(io.hptw.req.fire && s_last_hptw_req === false.B) {
318    w_last_hptw_resp := false.B
319    s_last_hptw_req := true.B
320  }
321
322  when (io.hptw.resp.fire && w_last_hptw_resp === false.B && stage1Hit){
323    w_last_hptw_resp := true.B
324    hptw_resp_stage2 := true.B
325    hptw_resp := io.hptw.resp.bits.h_resp
326  }
327
328  when(io.hptw.resp.fire && w_last_hptw_resp === false.B && !stage1Hit){
329    hptw_pageFault := io.hptw.resp.bits.h_resp.gpf
330    hptw_accessFault := io.hptw.resp.bits.h_resp.gaf
331    hptw_resp := io.hptw.resp.bits.h_resp
332    w_last_hptw_resp := true.B
333    mem_addr_update := true.B
334    last_s2xlate := false.B
335  }
336
337  when(sent_to_pmp && mem_addr_update === false.B){
338    s_mem_req := false.B
339    s_pmp_check := true.B
340  }
341
342  when(accessFault && idle === false.B){
343    s_pmp_check := true.B
344    s_mem_req := true.B
345    w_mem_resp := true.B
346    s_llptw_req := true.B
347    s_hptw_req := true.B
348    w_hptw_resp := true.B
349    s_last_hptw_req := true.B
350    w_last_hptw_resp := true.B
351    mem_addr_update := true.B
352    last_s2xlate := false.B
353  }
354
355  when(guestFault && idle === false.B){
356    s_pmp_check := true.B
357    s_mem_req := true.B
358    w_mem_resp := true.B
359    s_llptw_req := true.B
360    s_hptw_req := true.B
361    w_hptw_resp := true.B
362    s_last_hptw_req := true.B
363    w_last_hptw_resp := true.B
364    mem_addr_update := true.B
365    last_s2xlate := false.B
366  }
367
368  when (mem.req.fire){
369    s_mem_req := true.B
370    w_mem_resp := false.B
371  }
372
373  when(mem.resp.fire && w_mem_resp === false.B){
374    w_mem_resp := true.B
375    af_level := af_level - 1.U
376    s_llptw_req := false.B
377    mem_addr_update := true.B
378    gpf_level := Mux(mode === Sv39 && !pte_valid && !(l3Hit || l2Hit), gpf_level - 2.U, gpf_level - 1.U)
379    pte_valid := true.B
380  }
381
382  when(mem_addr_update){
383    when(level >= 2.U && !onlyS2xlate && !(guestFault || find_pte || accessFault)) {
384      level := levelNext
385      when(s2xlate){
386        s_hptw_req := false.B
387      }.otherwise{
388        s_mem_req := false.B
389      }
390      s_llptw_req := true.B
391      mem_addr_update := false.B
392    }.elsewhen(io.llptw.valid){
393      when(io.llptw.fire) {
394        idle := true.B
395        s_llptw_req := true.B
396        mem_addr_update := false.B
397        last_s2xlate := false.B
398      }
399      finish := true.B
400    }.elsewhen(s2xlate && last_s2xlate === true.B) {
401      when(accessFault || pageFault || ppn_af){
402        last_s2xlate := false.B
403      }.otherwise{
404        s_last_hptw_req := false.B
405        mem_addr_update := false.B
406      }
407    }.elsewhen(io.resp.valid){
408      when(io.resp.fire) {
409        idle := true.B
410        s_llptw_req := true.B
411        mem_addr_update := false.B
412        accessFault := false.B
413      }
414      finish := true.B
415    }
416  }
417
418
419  when (flush) {
420    idle := true.B
421    s_pmp_check := true.B
422    s_mem_req := true.B
423    s_llptw_req := true.B
424    w_mem_resp := true.B
425    accessFault := false.B
426    mem_addr_update := false.B
427    s_hptw_req := true.B
428    w_hptw_resp := true.B
429    s_last_hptw_req := true.B
430    w_last_hptw_resp := true.B
431    check_g_perm_fail := false.B
432  }
433
434
435  XSDebug(p"[ptw] level:${level} notFound:${pageFault}\n")
436
437  // perf
438  XSPerfAccumulate("fsm_count", io.req.fire)
439  for (i <- 0 until PtwWidth) {
440    XSPerfAccumulate(s"fsm_count_source${i}", io.req.fire && io.req.bits.req_info.source === i.U)
441  }
442  XSPerfAccumulate("fsm_busy", !idle)
443  XSPerfAccumulate("fsm_idle", idle)
444  XSPerfAccumulate("resp_blocked", io.resp.valid && !io.resp.ready)
445  XSPerfAccumulate("ptw_ppn_af", io.resp.fire && ppn_af)
446  XSPerfAccumulate("mem_count", mem.req.fire)
447  XSPerfAccumulate("mem_cycle", BoolStopWatch(mem.req.fire, mem.resp.fire, true))
448  XSPerfAccumulate("mem_blocked", mem.req.valid && !mem.req.ready)
449
450  TimeOutAssert(!idle, timeOutThreshold, "page table walker time out")
451
452  val perfEvents = Seq(
453    ("fsm_count         ", io.req.fire                                     ),
454    ("fsm_busy          ", !idle                                             ),
455    ("fsm_idle          ", idle                                              ),
456    ("resp_blocked      ", io.resp.valid && !io.resp.ready                   ),
457    ("mem_count         ", mem.req.fire                                    ),
458    ("mem_cycle         ", BoolStopWatch(mem.req.fire, mem.resp.fire, true)),
459    ("mem_blocked       ", mem.req.valid && !mem.req.ready                   ),
460  )
461  generatePerfEvent()
462}
463
464/*========================= LLPTW ==============================*/
465
466/** LLPTW : Last Level Page Table Walker
467  * the page walker that only takes 4KB(last level) page walk.
468  **/
469
470class LLPTWInBundle(implicit p: Parameters) extends XSBundle with HasPtwConst {
471  val req_info = Output(new L2TlbInnerBundle())
472  val ppn = Output(UInt(ptePPNLen.W))
473}
474
475class LLPTWIO(implicit p: Parameters) extends MMUIOBaseBundle with HasPtwConst {
476  val in = Flipped(DecoupledIO(new LLPTWInBundle()))
477  val out = DecoupledIO(new Bundle {
478    val req_info = Output(new L2TlbInnerBundle())
479    val id = Output(UInt(bMemID.W))
480    val h_resp = Output(new HptwResp)
481    val first_s2xlate_fault = Output(Bool()) // Whether the first stage 2 translation occurs pf/af
482    val af = Output(Bool())
483  })
484  val mem = new Bundle {
485    val req = DecoupledIO(new L2TlbMemReqBundle())
486    val resp = Flipped(Valid(new Bundle {
487      val id = Output(UInt(log2Up(l2tlbParams.llptwsize).W))
488      val value = Output(UInt(blockBits.W))
489    }))
490    val enq_ptr = Output(UInt(log2Ceil(l2tlbParams.llptwsize).W))
491    val buffer_it = Output(Vec(l2tlbParams.llptwsize, Bool()))
492    val refill = Output(new L2TlbInnerBundle())
493    val req_mask = Input(Vec(l2tlbParams.llptwsize, Bool()))
494    val flush_latch = Input(Vec(l2tlbParams.llptwsize, Bool()))
495  }
496  val cache = DecoupledIO(new L2TlbInnerBundle())
497  val pmp = new Bundle {
498    val req = Valid(new PMPReqBundle())
499    val resp = Flipped(new PMPRespBundle())
500  }
501  val hptw = new Bundle {
502    val req = DecoupledIO(new Bundle{
503      val source = UInt(bSourceWidth.W)
504      val id = UInt(log2Up(l2tlbParams.llptwsize).W)
505      val gvpn = UInt(ptePPNLen.W)
506    })
507    val resp = Flipped(Valid(new Bundle {
508      val id = Output(UInt(log2Up(l2tlbParams.llptwsize).W))
509      val h_resp = Output(new HptwResp)
510    }))
511  }
512}
513
514class LLPTWEntry(implicit p: Parameters) extends XSBundle with HasPtwConst {
515  val req_info = new L2TlbInnerBundle()
516  val ppn = UInt(ptePPNLen.W)
517  val wait_id = UInt(log2Up(l2tlbParams.llptwsize).W)
518  val af = Bool()
519  val hptw_resp = new HptwResp()
520  val first_s2xlate_fault = Output(Bool())
521}
522
523
524class LLPTW(implicit p: Parameters) extends XSModule with HasPtwConst with HasPerfEvents {
525  val io = IO(new LLPTWIO())
526  val enableS2xlate = io.in.bits.req_info.s2xlate =/= noS2xlate
527  val satp = Mux(enableS2xlate, io.csr.vsatp, io.csr.satp)
528
529  val flush = io.sfence.valid || io.csr.satp.changed || io.csr.vsatp.changed || io.csr.hgatp.changed
530  val entries = RegInit(VecInit(Seq.fill(l2tlbParams.llptwsize)(0.U.asTypeOf(new LLPTWEntry()))))
531  val state_idle :: state_hptw_req :: state_hptw_resp :: state_addr_check :: state_mem_req :: state_mem_waiting :: state_mem_out :: state_last_hptw_req :: state_last_hptw_resp :: state_cache :: Nil = Enum(10)
532  val state = RegInit(VecInit(Seq.fill(l2tlbParams.llptwsize)(state_idle)))
533
534  val is_emptys = state.map(_ === state_idle)
535  val is_mems = state.map(_ === state_mem_req)
536  val is_waiting = state.map(_ === state_mem_waiting)
537  val is_having = state.map(_ === state_mem_out)
538  val is_cache = state.map(_ === state_cache)
539  val is_hptw_req = state.map(_ === state_hptw_req)
540  val is_last_hptw_req = state.map(_ === state_last_hptw_req)
541  val is_hptw_resp = state.map(_ === state_hptw_resp)
542  val is_last_hptw_resp = state.map(_ === state_last_hptw_resp)
543
544  val full = !ParallelOR(is_emptys).asBool
545  val enq_ptr = ParallelPriorityEncoder(is_emptys)
546
547  val mem_ptr = ParallelPriorityEncoder(is_having) // TODO: optimize timing, bad: entries -> ptr -> entry
548  val mem_arb = Module(new RRArbiterInit(new LLPTWEntry(), l2tlbParams.llptwsize))
549  for (i <- 0 until l2tlbParams.llptwsize) {
550    mem_arb.io.in(i).bits := entries(i)
551    mem_arb.io.in(i).valid := is_mems(i) && !io.mem.req_mask(i)
552  }
553
554  // process hptw requests in serial
555  val hyper_arb1 = Module(new RRArbiterInit(new LLPTWEntry(), l2tlbParams.llptwsize))
556  for (i <- 0 until l2tlbParams.llptwsize) {
557    hyper_arb1.io.in(i).bits := entries(i)
558    hyper_arb1.io.in(i).valid := is_hptw_req(i) && !(Cat(is_hptw_resp).orR) && !(Cat(is_last_hptw_resp).orR)
559  }
560  val hyper_arb2 = Module(new RRArbiterInit(new LLPTWEntry(), l2tlbParams.llptwsize))
561  for(i <- 0 until l2tlbParams.llptwsize) {
562    hyper_arb2.io.in(i).bits := entries(i)
563    hyper_arb2.io.in(i).valid := is_last_hptw_req(i) && !(Cat(is_hptw_resp).orR) && !(Cat(is_last_hptw_resp).orR)
564  }
565
566  val cache_ptr = ParallelMux(is_cache, (0 until l2tlbParams.llptwsize).map(_.U(log2Up(l2tlbParams.llptwsize).W)))
567
568  // duplicate req
569  // to_wait: wait for the last to access mem, set to mem_resp
570  // to_cache: the last is back just right now, set to mem_cache
571  val dup_vec = state.indices.map(i =>
572    dup(io.in.bits.req_info.vpn, entries(i).req_info.vpn) && io.in.bits.req_info.s2xlate === entries(i).req_info.s2xlate
573  )
574  val dup_req_fire = mem_arb.io.out.fire && dup(io.in.bits.req_info.vpn, mem_arb.io.out.bits.req_info.vpn) && io.in.bits.req_info.s2xlate === mem_arb.io.out.bits.req_info.s2xlate // dup with the req fire entry
575  val dup_vec_wait = dup_vec.zip(is_waiting).map{case (d, w) => d && w} // dup with "mem_waiting" entries, sending mem req already
576  val dup_vec_having = dup_vec.zipWithIndex.map{case (d, i) => d && is_having(i)} // dup with the "mem_out" entry recv the data just now
577  val dup_vec_last_hptw = dup_vec.zipWithIndex.map{case (d, i) => d && (is_last_hptw_req(i) || is_last_hptw_resp(i))}
578  val wait_id = Mux(dup_req_fire, mem_arb.io.chosen, ParallelMux(dup_vec_wait zip entries.map(_.wait_id)))
579  val dup_wait_resp = io.mem.resp.fire && VecInit(dup_vec_wait)(io.mem.resp.bits.id) && !io.mem.flush_latch(io.mem.resp.bits.id) // dup with the entry that data coming next cycle
580  val to_wait = Cat(dup_vec_wait).orR || dup_req_fire
581  val to_mem_out = dup_wait_resp && ((entries(io.mem.resp.bits.id).req_info.s2xlate === noS2xlate) || (entries(io.mem.resp.bits.id).req_info.s2xlate === onlyStage1))
582  val to_cache = Cat(dup_vec_having).orR || Cat(dup_vec_last_hptw).orR
583  val to_hptw_req = io.in.bits.req_info.s2xlate === allStage
584  val to_last_hptw_req = dup_wait_resp && entries(io.mem.resp.bits.id).req_info.s2xlate === allStage
585  val last_hptw_req_id = io.mem.resp.bits.id
586  val req_paddr = MakeAddr(io.in.bits.ppn(ppnLen-1, 0), getVpnn(io.in.bits.req_info.vpn, 0))
587  val req_hpaddr = MakeAddr(entries(last_hptw_req_id).hptw_resp.genPPNS2(get_pn(req_paddr)), getVpnn(io.in.bits.req_info.vpn, 0))
588  val index =  Mux(entries(last_hptw_req_id).req_info.s2xlate === allStage, req_hpaddr, req_paddr)(log2Up(l2tlbParams.blockBytes)-1, log2Up(XLEN/8))
589  val last_hptw_req_ppn = io.mem.resp.bits.value.asTypeOf(Vec(blockBits / XLEN, new PteBundle()))(index).getPPN()
590  XSError(RegNext(dup_req_fire && Cat(dup_vec_wait).orR, init = false.B), "mem req but some entries already waiting, should not happed")
591
592  XSError(io.in.fire && ((to_mem_out && to_cache) || (to_wait && to_cache)), "llptw enq, to cache conflict with to mem")
593  val mem_resp_hit = RegInit(VecInit(Seq.fill(l2tlbParams.llptwsize)(false.B)))
594  val enq_state_normal = MuxCase(state_addr_check, Seq(
595    to_mem_out -> state_mem_out, // same to the blew, but the mem resp now
596    to_last_hptw_req -> state_last_hptw_req,
597    to_wait -> state_mem_waiting,
598    to_cache -> state_cache,
599    to_hptw_req -> state_hptw_req
600  ))
601  val enq_state = Mux(from_pre(io.in.bits.req_info.source) && enq_state_normal =/= state_addr_check, state_idle, enq_state_normal)
602  when (io.in.fire) {
603    // if prefetch req does not need mem access, just give it up.
604    // so there will be at most 1 + FilterSize entries that needs re-access page cache
605    // so 2 + FilterSize is enough to avoid dead-lock
606    state(enq_ptr) := enq_state
607    entries(enq_ptr).req_info := io.in.bits.req_info
608    entries(enq_ptr).ppn := Mux(to_last_hptw_req, last_hptw_req_ppn, io.in.bits.ppn)
609    entries(enq_ptr).wait_id := Mux(to_wait, wait_id, enq_ptr)
610    entries(enq_ptr).af := false.B
611    entries(enq_ptr).hptw_resp := Mux(to_last_hptw_req, entries(last_hptw_req_id).hptw_resp, Mux(to_wait, entries(wait_id).hptw_resp, entries(enq_ptr).hptw_resp))
612    entries(enq_ptr).first_s2xlate_fault := false.B
613    mem_resp_hit(enq_ptr) := to_mem_out || to_last_hptw_req
614  }
615
616  val enq_ptr_reg = RegNext(enq_ptr)
617  val need_addr_check = GatedValidRegNext(enq_state === state_addr_check && io.in.fire && !flush)
618
619  val hasHptwResp = ParallelOR(state.map(_ === state_hptw_resp)).asBool
620  val hptw_resp_ptr_reg = RegNext(io.hptw.resp.bits.id)
621  val hptw_need_addr_check = RegNext(hasHptwResp && io.hptw.resp.fire && !flush) && state(hptw_resp_ptr_reg) === state_addr_check
622
623  val ptes = io.mem.resp.bits.value.asTypeOf(Vec(blockBits / XLEN, new PteBundle()))
624  val gpaddr = MakeGPAddr(entries(hptw_resp_ptr_reg).ppn, getVpnn(entries(hptw_resp_ptr_reg).req_info.vpn, 0))
625  val hptw_resp = entries(hptw_resp_ptr_reg).hptw_resp
626  val hpaddr = Cat(hptw_resp.genPPNS2(get_pn(gpaddr)), get_off(gpaddr))
627  val addr = RegEnable(MakeAddr(io.in.bits.ppn(ppnLen - 1, 0), getVpnn(io.in.bits.req_info.vpn, 0)), io.in.fire)
628  io.pmp.req.valid := need_addr_check || hptw_need_addr_check
629  io.pmp.req.bits.addr := Mux(hptw_need_addr_check, hpaddr, addr)
630  io.pmp.req.bits.cmd := TlbCmd.read
631  io.pmp.req.bits.size := 3.U // TODO: fix it
632  val pmp_resp_valid = io.pmp.req.valid // same cycle
633  when (pmp_resp_valid) {
634    // NOTE: when pmp resp but state is not addr check, then the entry is dup with other entry, the state was changed before
635    //       when dup with the req-ing entry, set to mem_waiting (above codes), and the ld must be false, so dontcare
636    val ptr = Mux(hptw_need_addr_check, hptw_resp_ptr_reg, enq_ptr_reg);
637    val accessFault = io.pmp.resp.ld || io.pmp.resp.mmio
638    entries(ptr).af := accessFault
639    state(ptr) := Mux(accessFault, state_mem_out, state_mem_req)
640  }
641
642  when (mem_arb.io.out.fire) {
643    for (i <- state.indices) {
644      when (state(i) =/= state_idle && state(i) =/= state_mem_out && state(i) =/= state_last_hptw_req && state(i) =/= state_last_hptw_resp
645      && entries(i).req_info.s2xlate === mem_arb.io.out.bits.req_info.s2xlate
646      && dup(entries(i).req_info.vpn, mem_arb.io.out.bits.req_info.vpn)) {
647        // NOTE: "dup enq set state to mem_wait" -> "sending req set other dup entries to mem_wait"
648        state(i) := state_mem_waiting
649        entries(i).hptw_resp := entries(mem_arb.io.chosen).hptw_resp
650        entries(i).wait_id := mem_arb.io.chosen
651      }
652    }
653  }
654  when (io.mem.resp.fire) {
655    state.indices.map{i =>
656      when (state(i) === state_mem_waiting && io.mem.resp.bits.id === entries(i).wait_id) {
657        val req_paddr = MakeAddr(entries(i).ppn, getVpnn(entries(i).req_info.vpn, 0))
658        val req_hpaddr = MakeAddr(entries(i).hptw_resp.genPPNS2(get_pn(req_paddr)), getVpnn(entries(i).req_info.vpn, 0))
659        val index =  Mux(entries(i).req_info.s2xlate === allStage, req_hpaddr, req_paddr)(log2Up(l2tlbParams.blockBytes)-1, log2Up(XLEN/8))
660        state(i) := Mux(entries(i).req_info.s2xlate === allStage && !(ptes(index).isPf(0.U) || !ptes(index).isLeaf() || ptes(index).isAf() || ptes(index).isStage1Gpf(io.csr.vsatp.mode))
661                , state_last_hptw_req, state_mem_out)
662        mem_resp_hit(i) := true.B
663        entries(i).ppn := ptes(index).getPPN() // for last stage 2 translation
664        // when onlystage1, gpf has higher priority
665        entries(i).af := Mux(entries(i).req_info.s2xlate === allStage, false.B, Mux(entries(i).req_info.s2xlate === onlyStage1, ptes(index).isAf() && !ptes(index).isStage1Gpf(io.csr.vsatp.mode), ptes(index).isAf()))
666        entries(i).hptw_resp.gpf := Mux(entries(i).req_info.s2xlate === allStage || entries(i).req_info.s2xlate === onlyStage1, ptes(index).isStage1Gpf(io.csr.vsatp.mode), false.B)
667      }
668    }
669  }
670
671  when (hyper_arb1.io.out.fire) {
672    for (i <- state.indices) {
673      when (state(i) === state_hptw_req && entries(i).ppn === hyper_arb1.io.out.bits.ppn && entries(i).req_info.s2xlate === allStage && hyper_arb1.io.chosen === i.U) {
674        state(i) := state_hptw_resp
675        entries(i).wait_id := hyper_arb1.io.chosen
676      }
677    }
678  }
679
680  when (hyper_arb2.io.out.fire) {
681    for (i <- state.indices) {
682      when (state(i) === state_last_hptw_req && entries(i).ppn === hyper_arb2.io.out.bits.ppn && entries(i).req_info.s2xlate === allStage && hyper_arb2.io.chosen === i.U) {
683        state(i) := state_last_hptw_resp
684        entries(i).wait_id := hyper_arb2.io.chosen
685      }
686    }
687  }
688
689  when (io.hptw.resp.fire) {
690    for (i <- state.indices) {
691      when (state(i) === state_hptw_resp && io.hptw.resp.bits.id === entries(i).wait_id && io.hptw.resp.bits.h_resp.entry.tag === entries(i).ppn) {
692        val check_g_perm_fail = !io.hptw.resp.bits.h_resp.entry.perm.get.r && !(io.csr.priv.mxr && io.hptw.resp.bits.h_resp.entry.perm.get.x)
693        when (check_g_perm_fail || io.hptw.resp.bits.h_resp.gaf || io.hptw.resp.bits.h_resp.gpf) {
694          state(i) := state_mem_out
695          entries(i).hptw_resp := io.hptw.resp.bits.h_resp
696          entries(i).hptw_resp.gpf := io.hptw.resp.bits.h_resp.gpf || check_g_perm_fail
697          entries(i).first_s2xlate_fault := io.hptw.resp.bits.h_resp.gaf || io.hptw.resp.bits.h_resp.gpf
698        }.otherwise{ // change the entry that is waiting hptw resp
699          val need_to_waiting_vec = state.indices.map(i => state(i) === state_mem_waiting && dup(entries(i).req_info.vpn, entries(io.hptw.resp.bits.id).req_info.vpn))
700          val waiting_index = ParallelMux(need_to_waiting_vec zip entries.map(_.wait_id))
701          state(i) := Mux(Cat(need_to_waiting_vec).orR, state_mem_waiting, state_addr_check)
702          entries(i).hptw_resp := io.hptw.resp.bits.h_resp
703          entries(i).wait_id := Mux(Cat(need_to_waiting_vec).orR, waiting_index, entries(i).wait_id)
704          //To do: change the entry that is having the same hptw req
705        }
706      }
707      when (state(i) === state_last_hptw_resp && io.hptw.resp.bits.id === entries(i).wait_id && io.hptw.resp.bits.h_resp.entry.tag === entries(i).ppn) {
708        state(i) := state_mem_out
709        entries(i).hptw_resp := io.hptw.resp.bits.h_resp
710        //To do: change the entry that is having the same hptw req
711      }
712    }
713  }
714  when (io.out.fire) {
715    assert(state(mem_ptr) === state_mem_out)
716    state(mem_ptr) := state_idle
717  }
718  mem_resp_hit.map(a => when (a) { a := false.B } )
719
720  when (io.cache.fire) {
721    state(cache_ptr) := state_idle
722  }
723  XSError(io.out.fire && io.cache.fire && (mem_ptr === cache_ptr), "mem resp and cache fire at the same time at same entry")
724
725  when (flush) {
726    state.map(_ := state_idle)
727  }
728
729  io.in.ready := !full
730
731  io.out.valid := ParallelOR(is_having).asBool
732  io.out.bits.req_info := entries(mem_ptr).req_info
733  io.out.bits.id := mem_ptr
734  io.out.bits.af := entries(mem_ptr).af
735  io.out.bits.h_resp := entries(mem_ptr).hptw_resp
736  io.out.bits.first_s2xlate_fault := entries(mem_ptr).first_s2xlate_fault
737
738  val hptw_req_arb = Module(new Arbiter(new Bundle{
739      val source = UInt(bSourceWidth.W)
740      val id = UInt(log2Up(l2tlbParams.llptwsize).W)
741      val ppn = UInt(ptePPNLen.W)
742    } , 2))
743  // first stage 2 translation
744  hptw_req_arb.io.in(0).valid := hyper_arb1.io.out.valid
745  hptw_req_arb.io.in(0).bits.source := hyper_arb1.io.out.bits.req_info.source
746  hptw_req_arb.io.in(0).bits.ppn := hyper_arb1.io.out.bits.ppn
747  hptw_req_arb.io.in(0).bits.id := hyper_arb1.io.chosen
748  hyper_arb1.io.out.ready := hptw_req_arb.io.in(0).ready
749  // last stage 2 translation
750  hptw_req_arb.io.in(1).valid := hyper_arb2.io.out.valid
751  hptw_req_arb.io.in(1).bits.source := hyper_arb2.io.out.bits.req_info.source
752  hptw_req_arb.io.in(1).bits.ppn := hyper_arb2.io.out.bits.ppn
753  hptw_req_arb.io.in(1).bits.id := hyper_arb2.io.chosen
754  hyper_arb2.io.out.ready := hptw_req_arb.io.in(1).ready
755  hptw_req_arb.io.out.ready := io.hptw.req.ready
756  io.hptw.req.valid := hptw_req_arb.io.out.fire && !flush
757  io.hptw.req.bits.gvpn := hptw_req_arb.io.out.bits.ppn
758  io.hptw.req.bits.id := hptw_req_arb.io.out.bits.id
759  io.hptw.req.bits.source := hptw_req_arb.io.out.bits.source
760
761  io.mem.req.valid := mem_arb.io.out.valid && !flush
762  val mem_paddr = MakeAddr(mem_arb.io.out.bits.ppn, getVpnn(mem_arb.io.out.bits.req_info.vpn, 0))
763  val mem_hpaddr = MakeAddr(mem_arb.io.out.bits.hptw_resp.genPPNS2(get_pn(mem_paddr)), getVpnn(mem_arb.io.out.bits.req_info.vpn, 0))
764  io.mem.req.bits.addr := Mux(mem_arb.io.out.bits.req_info.s2xlate === allStage, mem_hpaddr, mem_paddr)
765  io.mem.req.bits.id := mem_arb.io.chosen
766  io.mem.req.bits.hptw_bypassed := false.B
767  mem_arb.io.out.ready := io.mem.req.ready
768  val mem_refill_id = RegNext(io.mem.resp.bits.id(log2Up(l2tlbParams.llptwsize)-1, 0))
769  io.mem.refill := entries(mem_refill_id).req_info
770  io.mem.refill.s2xlate := entries(mem_refill_id).req_info.s2xlate
771  io.mem.buffer_it := mem_resp_hit
772  io.mem.enq_ptr := enq_ptr
773
774  io.cache.valid := Cat(is_cache).orR
775  io.cache.bits := ParallelMux(is_cache, entries.map(_.req_info))
776
777  XSPerfAccumulate("llptw_in_count", io.in.fire)
778  XSPerfAccumulate("llptw_in_block", io.in.valid && !io.in.ready)
779  for (i <- 0 until 7) {
780    XSPerfAccumulate(s"enq_state${i}", io.in.fire && enq_state === i.U)
781  }
782  for (i <- 0 until (l2tlbParams.llptwsize + 1)) {
783    XSPerfAccumulate(s"util${i}", PopCount(is_emptys.map(!_)) === i.U)
784    XSPerfAccumulate(s"mem_util${i}", PopCount(is_mems) === i.U)
785    XSPerfAccumulate(s"waiting_util${i}", PopCount(is_waiting) === i.U)
786  }
787  XSPerfAccumulate("mem_count", io.mem.req.fire)
788  XSPerfAccumulate("mem_cycle", PopCount(is_waiting) =/= 0.U)
789  XSPerfAccumulate("blocked_in", io.in.valid && !io.in.ready)
790
791  for (i <- 0 until l2tlbParams.llptwsize) {
792    TimeOutAssert(state(i) =/= state_idle, timeOutThreshold, s"missqueue time out no out ${i}")
793  }
794
795  val perfEvents = Seq(
796    ("tlbllptw_incount           ", io.in.fire               ),
797    ("tlbllptw_inblock           ", io.in.valid && !io.in.ready),
798    ("tlbllptw_memcount          ", io.mem.req.fire          ),
799    ("tlbllptw_memcycle          ", PopCount(is_waiting)       ),
800  )
801  generatePerfEvent()
802}
803
804/*========================= HPTW ==============================*/
805
806/** HPTW : Hypervisor Page Table Walker
807  * the page walker take the virtual machine's page walk.
808  * guest physical address translation, guest physical address -> host physical address
809  **/
810class HPTWIO()(implicit p: Parameters) extends MMUIOBaseBundle with HasPtwConst {
811  val req = Flipped(DecoupledIO(new Bundle {
812    val source = UInt(bSourceWidth.W)
813    val id = UInt(log2Up(l2tlbParams.llptwsize).W)
814    val gvpn = UInt(gvpnLen.W)
815    val ppn = UInt(ppnLen.W)
816    val l3Hit = if (EnableSv48) Some(new Bool()) else None
817    val l2Hit = Bool()
818    val l1Hit = Bool()
819    val bypassed = Bool() // if bypass, don't refill
820  }))
821  val resp = DecoupledIO(new Bundle {
822    val source = UInt(bSourceWidth.W)
823    val resp = Output(new HptwResp())
824    val id = Output(UInt(bMemID.W))
825  })
826
827  val mem = new Bundle {
828    val req = DecoupledIO(new L2TlbMemReqBundle())
829    val resp = Flipped(ValidIO(UInt(XLEN.W)))
830    val mask = Input(Bool())
831  }
832  val refill = Output(new Bundle {
833    val req_info = new L2TlbInnerBundle()
834    val level = UInt(log2Up(Level + 1).W)
835  })
836  val pmp = new Bundle {
837    val req = ValidIO(new PMPReqBundle())
838    val resp = Flipped(new PMPRespBundle())
839  }
840}
841
842class HPTW()(implicit p: Parameters) extends XSModule with HasPtwConst {
843  val io = IO(new HPTWIO)
844  val hgatp = io.csr.hgatp
845  val sfence = io.sfence
846  val flush = sfence.valid || hgatp.changed || io.csr.satp.changed || io.csr.vsatp.changed
847  val mode = hgatp.mode
848
849  val level = RegInit(3.U(log2Up(Level + 1).W))
850  val af_level = RegInit(3.U(log2Up(Level + 1).W)) // access fault return this level
851  val gpaddr = Reg(UInt(GPAddrBits.W))
852  val req_ppn = Reg(UInt(ppnLen.W))
853  val vpn = gpaddr(GPAddrBits-1, offLen)
854  val levelNext = level - 1.U
855  val l3Hit = Reg(Bool())
856  val l2Hit = Reg(Bool())
857  val l1Hit = Reg(Bool())
858  val bypassed = Reg(Bool())
859//  val pte = io.mem.resp.bits.MergeRespToPte()
860  val pte = io.mem.resp.bits.asTypeOf(new PteBundle().cloneType)
861  val ppn_l3 = Mux(l3Hit, req_ppn, pte.ppn)
862  val ppn_l2 = Mux(l2Hit, req_ppn, pte.ppn)
863  val ppn_l1 = Mux(l1Hit, req_ppn, pte.ppn)
864  val ppn = Wire(UInt(PAddrBits.W))
865  val p_pte = MakeAddr(ppn, getVpnn(vpn, level))
866  val pg_base = Wire(UInt(PAddrBits.W))
867  val mem_addr = Wire(UInt(PAddrBits.W))
868  if (EnableSv48) {
869    when (mode === Sv48) {
870      ppn := Mux(af_level === 2.U, ppn_l3, Mux(af_level === 1.U, ppn_l2, ppn_l1)) // for l2, l1 and l3
871      pg_base := MakeGPAddr(hgatp.ppn, getGVpnn(vpn, 3.U, mode = Sv48)) // for l3
872      mem_addr := Mux(af_level === 3.U, pg_base, p_pte)
873    } .otherwise {
874      ppn := Mux(af_level === 1.U, ppn_l2, ppn_l1) //for l1 and l2
875      pg_base := MakeGPAddr(hgatp.ppn, getGVpnn(vpn, 2.U, mode = Sv39))
876      mem_addr := Mux(af_level === 2.U, pg_base, p_pte)
877    }
878  } else {
879    ppn := Mux(af_level === 1.U, ppn_l2, ppn_l1) //for l1 and l2
880    pg_base := MakeGPAddr(hgatp.ppn, getGVpnn(vpn, 2.U, mode = Sv39))
881    mem_addr := Mux(af_level === 2.U, pg_base, p_pte)
882  }
883
884  //s/w register
885  val s_pmp_check = RegInit(true.B)
886  val s_mem_req = RegInit(true.B)
887  val w_mem_resp = RegInit(true.B)
888  val idle = RegInit(true.B)
889  val mem_addr_update = RegInit(false.B)
890  val finish = WireInit(false.B)
891
892  val sent_to_pmp = !idle && (!s_pmp_check || mem_addr_update) && !finish
893  val pageFault = pte.isGpf(level) || (!pte.isLeaf() && level === 0.U)
894  val accessFault = RegEnable(io.pmp.resp.ld || io.pmp.resp.mmio, sent_to_pmp)
895
896  val ppn_af = pte.isAf()
897  val find_pte = pte.isLeaf() || ppn_af || pageFault
898
899  val resp_valid = !idle && mem_addr_update && ((w_mem_resp && find_pte) || (s_pmp_check && accessFault))
900  val id = Reg(UInt(log2Up(l2tlbParams.llptwsize).W))
901  val source = RegEnable(io.req.bits.source, io.req.fire)
902
903  io.req.ready := idle
904  val resp = Wire(new HptwResp())
905  resp.apply(pageFault && !accessFault && !ppn_af, accessFault || ppn_af, Mux(accessFault, af_level, level), pte, vpn, hgatp.vmid)
906  io.resp.valid := resp_valid
907  io.resp.bits.id := id
908  io.resp.bits.resp := resp
909  io.resp.bits.source := source
910
911  io.pmp.req.valid := DontCare
912  io.pmp.req.bits.addr := mem_addr
913  io.pmp.req.bits.size := 3.U
914  io.pmp.req.bits.cmd := TlbCmd.read
915
916  io.mem.req.valid := !s_mem_req && !io.mem.mask && !accessFault && s_pmp_check
917  io.mem.req.bits.addr := mem_addr
918  io.mem.req.bits.id := HptwReqId.U(bMemID.W)
919  io.mem.req.bits.hptw_bypassed := bypassed
920
921  io.refill.req_info.vpn := vpn
922  io.refill.level := level
923  io.refill.req_info.source := source
924  io.refill.req_info.s2xlate := onlyStage2
925  when (idle){
926    when(io.req.fire){
927      bypassed := io.req.bits.bypassed
928      idle := false.B
929      gpaddr := Cat(io.req.bits.gvpn, 0.U(offLen.W))
930      accessFault := false.B
931      s_pmp_check := false.B
932      id := io.req.bits.id
933      req_ppn := io.req.bits.ppn
934      if (EnableSv48) {
935        when (mode === Sv48) {
936          level := Mux(io.req.bits.l1Hit, 0.U, Mux(io.req.bits.l2Hit, 1.U, Mux(io.req.bits.l3Hit.get, 2.U, 3.U)))
937          af_level := Mux(io.req.bits.l1Hit, 0.U, Mux(io.req.bits.l2Hit, 1.U, Mux(io.req.bits.l3Hit.get, 2.U, 3.U)))
938          l3Hit := io.req.bits.l3Hit.get
939        } .otherwise {
940          level := Mux(io.req.bits.l1Hit, 0.U, Mux(io.req.bits.l2Hit, 1.U, 2.U))
941          af_level := Mux(io.req.bits.l1Hit, 0.U, Mux(io.req.bits.l2Hit, 1.U, 2.U))
942          l3Hit := false.B
943        }
944      } else {
945        level := Mux(io.req.bits.l1Hit, 0.U, Mux(io.req.bits.l2Hit, 1.U, 2.U))
946        af_level := Mux(io.req.bits.l1Hit, 0.U, Mux(io.req.bits.l2Hit, 1.U, 2.U))
947        l3Hit := false.B
948      }
949      l2Hit := io.req.bits.l2Hit
950      l1Hit := io.req.bits.l1Hit
951    }
952  }
953
954  when(sent_to_pmp && !mem_addr_update){
955    s_mem_req := false.B
956    s_pmp_check := true.B
957  }
958
959  when(accessFault && !idle){
960    s_pmp_check := true.B
961    s_mem_req := true.B
962    w_mem_resp := true.B
963    mem_addr_update := true.B
964  }
965
966  when(io.mem.req.fire){
967    s_mem_req := true.B
968    w_mem_resp := false.B
969  }
970
971  when(io.mem.resp.fire && !w_mem_resp){
972    w_mem_resp := true.B
973    af_level := af_level - 1.U
974    mem_addr_update := true.B
975  }
976
977  when(mem_addr_update){
978    when(!(find_pte || accessFault)){
979      level := levelNext
980      s_mem_req := false.B
981      mem_addr_update := false.B
982    }.elsewhen(resp_valid){
983      when(io.resp.fire){
984        idle := true.B
985        mem_addr_update := false.B
986        accessFault := false.B
987      }
988      finish := true.B
989    }
990  }
991   when (flush) {
992    idle := true.B
993    s_pmp_check := true.B
994    s_mem_req := true.B
995    w_mem_resp := true.B
996    accessFault := false.B
997    mem_addr_update := false.B
998  }
999}
1000