xref: /XiangShan/src/main/scala/xiangshan/cache/mmu/PageTableWalker.scala (revision 3088616cbf0793407bb68460b2db89b7de80c12a)
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 guestFault = hptw_pageFault || hptw_accessFault || gvpn_gpf
179  val hpaddr = Cat(hptw_resp.genPPNS2(get_pn(gpaddr)), get_off(gpaddr))
180  val fake_h_resp = 0.U.asTypeOf(new HptwResp)
181  fake_h_resp.entry.tag := get_pn(gpaddr)
182  fake_h_resp.entry.vmid.map(_ := io.csr.hgatp.vmid)
183  fake_h_resp.gpf := true.B
184
185  val pte_valid = RegInit(false.B)  // avoid l1tlb pf from stage1 when gpf happens in the first s2xlate in PTW
186  val fake_pte = 0.U.asTypeOf(new PteBundle())
187  fake_pte.perm.v := false.B // tell L1TLB this is fake pte
188  fake_pte.perm.r := true.B
189  fake_pte.perm.w := true.B
190  fake_pte.perm.x := true.B
191  fake_pte.perm.a := true.B
192  fake_pte.perm.d := true.B
193  fake_pte.ppn := ppn(ppnLen - 1, 0)
194  fake_pte.ppn_high := ppn(ptePPNLen - 1, ppnLen)
195
196  io.req.ready := idle
197  val ptw_resp = Wire(new PtwMergeResp)
198  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)
199
200  val normal_resp = idle === false.B && mem_addr_update && !last_s2xlate && (guestFault || (w_mem_resp && find_pte) || (s_pmp_check && accessFault) || onlyS2xlate )
201  val stageHit_resp = idle === false.B && hptw_resp_stage2
202  io.resp.valid := Mux(stage1Hit, stageHit_resp, normal_resp)
203  io.resp.bits.source := source
204  io.resp.bits.resp := Mux(stage1Hit || (l3Hit || l2Hit) && guestFault && !pte_valid, stage1, ptw_resp)
205  io.resp.bits.h_resp := Mux(gvpn_gpf, fake_h_resp, hptw_resp)
206  io.resp.bits.s2xlate := req_s2xlate
207
208  io.llptw.valid := s_llptw_req === false.B && to_find_pte && !accessFault && !guestFault
209  io.llptw.bits.req_info.source := source
210  io.llptw.bits.req_info.vpn := vpn
211  io.llptw.bits.req_info.s2xlate := req_s2xlate
212  io.llptw.bits.ppn := DontCare
213
214  io.pmp.req.valid := DontCare // samecycle, do not use valid
215  io.pmp.req.bits.addr := Mux(s2xlate, hpaddr, mem_addr)
216  io.pmp.req.bits.size := 3.U // TODO: fix it
217  io.pmp.req.bits.cmd := TlbCmd.read
218
219  mem.req.valid := s_mem_req === false.B && !mem.mask && !accessFault && s_pmp_check
220  mem.req.bits.addr := Mux(s2xlate, hpaddr, mem_addr)
221  mem.req.bits.id := FsmReqID.U(bMemID.W)
222  mem.req.bits.hptw_bypassed := false.B
223
224  io.refill.req_info.s2xlate := req_s2xlate
225  io.refill.req_info.vpn := vpn
226  io.refill.level := level
227  io.refill.req_info.source := source
228
229  io.hptw.req.valid := !s_hptw_req || !s_last_hptw_req
230  io.hptw.req.bits.id := FsmReqID.U(bMemID.W)
231  io.hptw.req.bits.gvpn := get_pn(gpaddr)
232  io.hptw.req.bits.source := source
233
234  when (io.req.fire && io.req.bits.stage1Hit){
235    idle := false.B
236    req_s2xlate := io.req.bits.req_info.s2xlate
237    s_last_hptw_req := false.B
238    hptw_resp_stage2 := false.B
239    last_s2xlate := false.B
240    hptw_pageFault := false.B
241    hptw_accessFault := false.B
242  }
243
244  when (io.resp.fire && stage1Hit){
245    idle := true.B
246  }
247
248  when (io.req.fire && !io.req.bits.stage1Hit){
249    val req = io.req.bits
250    if (EnableSv48) {
251      when (mode === Sv48) {
252        level := Mux(req.l2Hit, 1.U, Mux(req.l3Hit.get, 2.U, 3.U))
253        af_level := Mux(req.l2Hit, 1.U, Mux(req.l3Hit.get, 2.U, 3.U))
254        gpf_level := Mux(req.l2Hit, 2.U, Mux(req.l3Hit.get, 3.U, 0.U))
255        ppn := Mux(req.l2Hit || req.l3Hit.get, io.req.bits.ppn, satp.ppn)
256        l3Hit := req.l3Hit.get
257      } .otherwise {
258        level := Mux(req.l2Hit, 1.U, 2.U)
259        af_level := Mux(req.l2Hit, 1.U, 2.U)
260        gpf_level := 0.U
261        ppn := Mux(req.l2Hit, io.req.bits.ppn, satp.ppn)
262        l3Hit := false.B
263      }
264    } else {
265      level := Mux(req.l2Hit, 1.U, 2.U)
266      af_level := Mux(req.l2Hit, 1.U, 2.U)
267      gpf_level := 0.U
268      ppn := Mux(req.l2Hit, io.req.bits.ppn, satp.ppn)
269      l3Hit := false.B
270    }
271    vpn := io.req.bits.req_info.vpn
272    l2Hit := req.l2Hit
273    accessFault := false.B
274    idle := false.B
275    hptw_pageFault := false.B
276    hptw_accessFault := false.B
277    pte_valid := false.B
278    req_s2xlate := io.req.bits.req_info.s2xlate
279    when(io.req.bits.req_info.s2xlate === onlyStage2){
280      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
281      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)
282      last_s2xlate := false.B
283      when(check_gpa_high_fail){
284        mem_addr_update := true.B
285      }.otherwise{
286        s_last_hptw_req := false.B
287      }
288    }.elsewhen(io.req.bits.req_info.s2xlate === allStage){
289      last_s2xlate := true.B
290      s_hptw_req := false.B
291    }.otherwise {
292      last_s2xlate := false.B
293      s_pmp_check := false.B
294    }
295  }
296
297  when(io.hptw.req.fire && s_hptw_req === false.B){
298    s_hptw_req := true.B
299    w_hptw_resp := false.B
300  }
301
302  when(io.hptw.resp.fire && w_hptw_resp === false.B) {
303    w_hptw_resp := true.B
304    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)
305    hptw_pageFault := io.hptw.resp.bits.h_resp.gpf || g_perm_fail
306    hptw_accessFault := io.hptw.resp.bits.h_resp.gaf
307    hptw_resp := io.hptw.resp.bits.h_resp
308    hptw_resp.gpf := io.hptw.resp.bits.h_resp.gpf || g_perm_fail
309    when(!(g_perm_fail || io.hptw.resp.bits.h_resp.gpf || io.hptw.resp.bits.h_resp.gaf)) {
310      s_pmp_check := false.B
311    }
312  }
313
314  when(io.hptw.req.fire && s_last_hptw_req === false.B) {
315    w_last_hptw_resp := false.B
316    s_last_hptw_req := true.B
317  }
318
319  when (io.hptw.resp.fire && w_last_hptw_resp === false.B && stage1Hit){
320    w_last_hptw_resp := true.B
321    hptw_resp_stage2 := true.B
322    hptw_resp := io.hptw.resp.bits.h_resp
323  }
324
325  when(io.hptw.resp.fire && w_last_hptw_resp === false.B && !stage1Hit){
326    hptw_pageFault := io.hptw.resp.bits.h_resp.gpf
327    hptw_accessFault := io.hptw.resp.bits.h_resp.gaf
328    hptw_resp := io.hptw.resp.bits.h_resp
329    w_last_hptw_resp := true.B
330    mem_addr_update := true.B
331    last_s2xlate := false.B
332  }
333
334  when(sent_to_pmp && mem_addr_update === false.B){
335    s_mem_req := false.B
336    s_pmp_check := true.B
337  }
338
339  when(accessFault && idle === false.B){
340    s_pmp_check := true.B
341    s_mem_req := true.B
342    w_mem_resp := true.B
343    s_llptw_req := true.B
344    s_hptw_req := true.B
345    w_hptw_resp := true.B
346    s_last_hptw_req := true.B
347    w_last_hptw_resp := true.B
348    mem_addr_update := true.B
349    last_s2xlate := false.B
350  }
351
352  when(guestFault && idle === false.B){
353    s_pmp_check := true.B
354    s_mem_req := true.B
355    w_mem_resp := true.B
356    s_llptw_req := true.B
357    s_hptw_req := true.B
358    w_hptw_resp := true.B
359    s_last_hptw_req := true.B
360    w_last_hptw_resp := true.B
361    mem_addr_update := true.B
362    last_s2xlate := false.B
363  }
364
365  when (mem.req.fire){
366    s_mem_req := true.B
367    w_mem_resp := false.B
368  }
369
370  when(mem.resp.fire && w_mem_resp === false.B){
371    w_mem_resp := true.B
372    af_level := af_level - 1.U
373    s_llptw_req := false.B
374    mem_addr_update := true.B
375    gpf_level := Mux(mode === Sv39 && !pte_valid && !(l3Hit || l2Hit), gpf_level - 2.U, gpf_level - 1.U)
376    pte_valid := true.B
377  }
378
379  when(mem_addr_update){
380    when(level >= 2.U && !onlyS2xlate && !(guestFault || find_pte || accessFault)) {
381      level := levelNext
382      when(s2xlate){
383        s_hptw_req := false.B
384      }.otherwise{
385        s_mem_req := false.B
386      }
387      s_llptw_req := true.B
388      mem_addr_update := false.B
389    }.elsewhen(io.llptw.valid){
390      when(io.llptw.fire) {
391        idle := true.B
392        s_llptw_req := true.B
393        mem_addr_update := false.B
394        last_s2xlate := false.B
395      }
396      finish := true.B
397    }.elsewhen(s2xlate && last_s2xlate === true.B) {
398      when(accessFault || pageFault || ppn_af){
399        last_s2xlate := false.B
400      }.otherwise{
401        s_last_hptw_req := false.B
402        mem_addr_update := false.B
403      }
404    }.elsewhen(io.resp.valid){
405      when(io.resp.fire) {
406        idle := true.B
407        s_llptw_req := true.B
408        mem_addr_update := false.B
409        accessFault := false.B
410      }
411      finish := true.B
412    }
413  }
414
415
416  when (flush) {
417    idle := true.B
418    s_pmp_check := true.B
419    s_mem_req := true.B
420    s_llptw_req := true.B
421    w_mem_resp := true.B
422    accessFault := false.B
423    mem_addr_update := false.B
424    s_hptw_req := true.B
425    w_hptw_resp := true.B
426    s_last_hptw_req := true.B
427    w_last_hptw_resp := true.B
428  }
429
430
431  XSDebug(p"[ptw] level:${level} notFound:${pageFault}\n")
432
433  // perf
434  XSPerfAccumulate("fsm_count", io.req.fire)
435  for (i <- 0 until PtwWidth) {
436    XSPerfAccumulate(s"fsm_count_source${i}", io.req.fire && io.req.bits.req_info.source === i.U)
437  }
438  XSPerfAccumulate("fsm_busy", !idle)
439  XSPerfAccumulate("fsm_idle", idle)
440  XSPerfAccumulate("resp_blocked", io.resp.valid && !io.resp.ready)
441  XSPerfAccumulate("ptw_ppn_af", io.resp.fire && ppn_af)
442  XSPerfAccumulate("mem_count", mem.req.fire)
443  XSPerfAccumulate("mem_cycle", BoolStopWatch(mem.req.fire, mem.resp.fire, true))
444  XSPerfAccumulate("mem_blocked", mem.req.valid && !mem.req.ready)
445
446  TimeOutAssert(!idle, timeOutThreshold, "page table walker time out")
447
448  val perfEvents = Seq(
449    ("fsm_count         ", io.req.fire                                     ),
450    ("fsm_busy          ", !idle                                             ),
451    ("fsm_idle          ", idle                                              ),
452    ("resp_blocked      ", io.resp.valid && !io.resp.ready                   ),
453    ("mem_count         ", mem.req.fire                                    ),
454    ("mem_cycle         ", BoolStopWatch(mem.req.fire, mem.resp.fire, true)),
455    ("mem_blocked       ", mem.req.valid && !mem.req.ready                   ),
456  )
457  generatePerfEvent()
458}
459
460/*========================= LLPTW ==============================*/
461
462/** LLPTW : Last Level Page Table Walker
463  * the page walker that only takes 4KB(last level) page walk.
464  **/
465
466class LLPTWInBundle(implicit p: Parameters) extends XSBundle with HasPtwConst {
467  val req_info = Output(new L2TlbInnerBundle())
468  val ppn = Output(UInt(ptePPNLen.W))
469}
470
471class LLPTWIO(implicit p: Parameters) extends MMUIOBaseBundle with HasPtwConst {
472  val in = Flipped(DecoupledIO(new LLPTWInBundle()))
473  val out = DecoupledIO(new Bundle {
474    val req_info = Output(new L2TlbInnerBundle())
475    val id = Output(UInt(bMemID.W))
476    val h_resp = Output(new HptwResp)
477    val first_s2xlate_fault = Output(Bool()) // Whether the first stage 2 translation occurs pf/af
478    val af = Output(Bool())
479  })
480  val mem = new Bundle {
481    val req = DecoupledIO(new L2TlbMemReqBundle())
482    val resp = Flipped(Valid(new Bundle {
483      val id = Output(UInt(log2Up(l2tlbParams.llptwsize).W))
484      val value = Output(UInt(blockBits.W))
485    }))
486    val enq_ptr = Output(UInt(log2Ceil(l2tlbParams.llptwsize).W))
487    val buffer_it = Output(Vec(l2tlbParams.llptwsize, Bool()))
488    val refill = Output(new L2TlbInnerBundle())
489    val req_mask = Input(Vec(l2tlbParams.llptwsize, Bool()))
490    val flush_latch = Input(Vec(l2tlbParams.llptwsize, Bool()))
491  }
492  val cache = DecoupledIO(new L2TlbInnerBundle())
493  val pmp = new Bundle {
494    val req = Valid(new PMPReqBundle())
495    val resp = Flipped(new PMPRespBundle())
496  }
497  val hptw = new Bundle {
498    val req = DecoupledIO(new Bundle{
499      val source = UInt(bSourceWidth.W)
500      val id = UInt(log2Up(l2tlbParams.llptwsize).W)
501      val gvpn = UInt(ptePPNLen.W)
502    })
503    val resp = Flipped(Valid(new Bundle {
504      val id = Output(UInt(log2Up(l2tlbParams.llptwsize).W))
505      val h_resp = Output(new HptwResp)
506    }))
507  }
508}
509
510class LLPTWEntry(implicit p: Parameters) extends XSBundle with HasPtwConst {
511  val req_info = new L2TlbInnerBundle()
512  val ppn = UInt(ptePPNLen.W)
513  val wait_id = UInt(log2Up(l2tlbParams.llptwsize).W)
514  val af = Bool()
515  val hptw_resp = new HptwResp()
516  val first_s2xlate_fault = Output(Bool())
517}
518
519
520class LLPTW(implicit p: Parameters) extends XSModule with HasPtwConst with HasPerfEvents {
521  val io = IO(new LLPTWIO())
522  val enableS2xlate = io.in.bits.req_info.s2xlate =/= noS2xlate
523  val satp = Mux(enableS2xlate, io.csr.vsatp, io.csr.satp)
524
525  val flush = io.sfence.valid || io.csr.satp.changed || io.csr.vsatp.changed || io.csr.hgatp.changed
526  val entries = RegInit(VecInit(Seq.fill(l2tlbParams.llptwsize)(0.U.asTypeOf(new LLPTWEntry()))))
527  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)
528  val state = RegInit(VecInit(Seq.fill(l2tlbParams.llptwsize)(state_idle)))
529
530  val is_emptys = state.map(_ === state_idle)
531  val is_mems = state.map(_ === state_mem_req)
532  val is_waiting = state.map(_ === state_mem_waiting)
533  val is_having = state.map(_ === state_mem_out)
534  val is_cache = state.map(_ === state_cache)
535  val is_hptw_req = state.map(_ === state_hptw_req)
536  val is_last_hptw_req = state.map(_ === state_last_hptw_req)
537  val is_hptw_resp = state.map(_ === state_hptw_resp)
538  val is_last_hptw_resp = state.map(_ === state_last_hptw_resp)
539
540  val full = !ParallelOR(is_emptys).asBool
541  val enq_ptr = ParallelPriorityEncoder(is_emptys)
542
543  val mem_ptr = ParallelPriorityEncoder(is_having) // TODO: optimize timing, bad: entries -> ptr -> entry
544  val mem_arb = Module(new RRArbiterInit(new LLPTWEntry(), l2tlbParams.llptwsize))
545  for (i <- 0 until l2tlbParams.llptwsize) {
546    mem_arb.io.in(i).bits := entries(i)
547    mem_arb.io.in(i).valid := is_mems(i) && !io.mem.req_mask(i)
548  }
549
550  // process hptw requests in serial
551  val hyper_arb1 = Module(new RRArbiterInit(new LLPTWEntry(), l2tlbParams.llptwsize))
552  for (i <- 0 until l2tlbParams.llptwsize) {
553    hyper_arb1.io.in(i).bits := entries(i)
554    hyper_arb1.io.in(i).valid := is_hptw_req(i) && !(Cat(is_hptw_resp).orR) && !(Cat(is_last_hptw_resp).orR)
555  }
556  val hyper_arb2 = Module(new RRArbiterInit(new LLPTWEntry(), l2tlbParams.llptwsize))
557  for(i <- 0 until l2tlbParams.llptwsize) {
558    hyper_arb2.io.in(i).bits := entries(i)
559    hyper_arb2.io.in(i).valid := is_last_hptw_req(i) && !(Cat(is_hptw_resp).orR) && !(Cat(is_last_hptw_resp).orR)
560  }
561
562  val cache_ptr = ParallelMux(is_cache, (0 until l2tlbParams.llptwsize).map(_.U(log2Up(l2tlbParams.llptwsize).W)))
563
564  // duplicate req
565  // to_wait: wait for the last to access mem, set to mem_resp
566  // to_cache: the last is back just right now, set to mem_cache
567  val dup_vec = state.indices.map(i =>
568    dup(io.in.bits.req_info.vpn, entries(i).req_info.vpn) && io.in.bits.req_info.s2xlate === entries(i).req_info.s2xlate
569  )
570  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
571  val dup_vec_wait = dup_vec.zip(is_waiting).map{case (d, w) => d && w} // dup with "mem_waiting" entries, sending mem req already
572  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
573  val dup_vec_last_hptw = dup_vec.zipWithIndex.map{case (d, i) => d && (is_last_hptw_req(i) || is_last_hptw_resp(i))}
574  val wait_id = Mux(dup_req_fire, mem_arb.io.chosen, ParallelMux(dup_vec_wait zip entries.map(_.wait_id)))
575  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
576  val to_wait = Cat(dup_vec_wait).orR || dup_req_fire
577  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))
578  val to_cache = Cat(dup_vec_having).orR || Cat(dup_vec_last_hptw).orR
579  val to_hptw_req = io.in.bits.req_info.s2xlate === allStage
580  val to_last_hptw_req = dup_wait_resp && entries(io.mem.resp.bits.id).req_info.s2xlate === allStage
581  val last_hptw_req_id = io.mem.resp.bits.id
582  val req_paddr = MakeAddr(io.in.bits.ppn(ppnLen-1, 0), getVpnn(io.in.bits.req_info.vpn, 0))
583  val req_hpaddr = MakeAddr(entries(last_hptw_req_id).hptw_resp.genPPNS2(get_pn(req_paddr)), getVpnn(io.in.bits.req_info.vpn, 0))
584  val index =  Mux(entries(last_hptw_req_id).req_info.s2xlate === allStage, req_hpaddr, req_paddr)(log2Up(l2tlbParams.blockBytes)-1, log2Up(XLEN/8))
585  val last_hptw_req_ppn = io.mem.resp.bits.value.asTypeOf(Vec(blockBits / XLEN, new PteBundle()))(index).getPPN()
586  XSError(RegNext(dup_req_fire && Cat(dup_vec_wait).orR, init = false.B), "mem req but some entries already waiting, should not happed")
587
588  XSError(io.in.fire && ((to_mem_out && to_cache) || (to_wait && to_cache)), "llptw enq, to cache conflict with to mem")
589  val mem_resp_hit = RegInit(VecInit(Seq.fill(l2tlbParams.llptwsize)(false.B)))
590  val enq_state_normal = MuxCase(state_addr_check, Seq(
591    to_mem_out -> state_mem_out, // same to the blew, but the mem resp now
592    to_last_hptw_req -> state_last_hptw_req,
593    to_wait -> state_mem_waiting,
594    to_cache -> state_cache,
595    to_hptw_req -> state_hptw_req
596  ))
597  val enq_state = Mux(from_pre(io.in.bits.req_info.source) && enq_state_normal =/= state_addr_check, state_idle, enq_state_normal)
598  when (io.in.fire) {
599    // if prefetch req does not need mem access, just give it up.
600    // so there will be at most 1 + FilterSize entries that needs re-access page cache
601    // so 2 + FilterSize is enough to avoid dead-lock
602    state(enq_ptr) := enq_state
603    entries(enq_ptr).req_info := io.in.bits.req_info
604    entries(enq_ptr).ppn := Mux(to_last_hptw_req, last_hptw_req_ppn, io.in.bits.ppn)
605    entries(enq_ptr).wait_id := Mux(to_wait, wait_id, enq_ptr)
606    entries(enq_ptr).af := false.B
607    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))
608    entries(enq_ptr).first_s2xlate_fault := false.B
609    mem_resp_hit(enq_ptr) := to_mem_out || to_last_hptw_req
610  }
611
612  val enq_ptr_reg = RegNext(enq_ptr)
613  val need_addr_check = GatedValidRegNext(enq_state === state_addr_check && io.in.fire && !flush)
614
615  val hasHptwResp = ParallelOR(state.map(_ === state_hptw_resp)).asBool
616  val hptw_resp_ptr_reg = RegNext(io.hptw.resp.bits.id)
617  val hptw_need_addr_check = RegNext(hasHptwResp && io.hptw.resp.fire && !flush) && state(hptw_resp_ptr_reg) === state_addr_check
618
619  val ptes = io.mem.resp.bits.value.asTypeOf(Vec(blockBits / XLEN, new PteBundle()))
620  val gpaddr = MakeGPAddr(entries(hptw_resp_ptr_reg).ppn, getVpnn(entries(hptw_resp_ptr_reg).req_info.vpn, 0))
621  val hptw_resp = entries(hptw_resp_ptr_reg).hptw_resp
622  val hpaddr = Cat(hptw_resp.genPPNS2(get_pn(gpaddr)), get_off(gpaddr))
623  val addr = RegEnable(MakeAddr(io.in.bits.ppn(ppnLen - 1, 0), getVpnn(io.in.bits.req_info.vpn, 0)), io.in.fire)
624  io.pmp.req.valid := need_addr_check || hptw_need_addr_check
625  io.pmp.req.bits.addr := Mux(hptw_need_addr_check, hpaddr, addr)
626  io.pmp.req.bits.cmd := TlbCmd.read
627  io.pmp.req.bits.size := 3.U // TODO: fix it
628  val pmp_resp_valid = io.pmp.req.valid // same cycle
629  when (pmp_resp_valid) {
630    // NOTE: when pmp resp but state is not addr check, then the entry is dup with other entry, the state was changed before
631    //       when dup with the req-ing entry, set to mem_waiting (above codes), and the ld must be false, so dontcare
632    val ptr = Mux(hptw_need_addr_check, hptw_resp_ptr_reg, enq_ptr_reg);
633    val accessFault = io.pmp.resp.ld || io.pmp.resp.mmio
634    entries(ptr).af := accessFault
635    state(ptr) := Mux(accessFault, state_mem_out, state_mem_req)
636  }
637
638  when (mem_arb.io.out.fire) {
639    for (i <- state.indices) {
640      when (state(i) =/= state_idle && state(i) =/= state_mem_out && state(i) =/= state_last_hptw_req && state(i) =/= state_last_hptw_resp
641      && entries(i).req_info.s2xlate === mem_arb.io.out.bits.req_info.s2xlate
642      && dup(entries(i).req_info.vpn, mem_arb.io.out.bits.req_info.vpn)) {
643        // NOTE: "dup enq set state to mem_wait" -> "sending req set other dup entries to mem_wait"
644        state(i) := state_mem_waiting
645        entries(i).hptw_resp := entries(mem_arb.io.chosen).hptw_resp
646        entries(i).wait_id := mem_arb.io.chosen
647      }
648    }
649  }
650  when (io.mem.resp.fire) {
651    state.indices.map{i =>
652      when (state(i) === state_mem_waiting && io.mem.resp.bits.id === entries(i).wait_id) {
653        val req_paddr = MakeAddr(entries(i).ppn, getVpnn(entries(i).req_info.vpn, 0))
654        val req_hpaddr = MakeAddr(entries(i).hptw_resp.genPPNS2(get_pn(req_paddr)), getVpnn(entries(i).req_info.vpn, 0))
655        val index =  Mux(entries(i).req_info.s2xlate === allStage, req_hpaddr, req_paddr)(log2Up(l2tlbParams.blockBytes)-1, log2Up(XLEN/8))
656        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))
657                , state_last_hptw_req, state_mem_out)
658        mem_resp_hit(i) := true.B
659        entries(i).ppn := ptes(index).getPPN() // for last stage 2 translation
660        // when onlystage1, gpf has higher priority
661        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()))
662        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)
663      }
664    }
665  }
666
667  when (hyper_arb1.io.out.fire) {
668    for (i <- state.indices) {
669      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) {
670        state(i) := state_hptw_resp
671        entries(i).wait_id := hyper_arb1.io.chosen
672      }
673    }
674  }
675
676  when (hyper_arb2.io.out.fire) {
677    for (i <- state.indices) {
678      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) {
679        state(i) := state_last_hptw_resp
680        entries(i).wait_id := hyper_arb2.io.chosen
681      }
682    }
683  }
684
685  when (io.hptw.resp.fire) {
686    for (i <- state.indices) {
687      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) {
688        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)
689        when (check_g_perm_fail || io.hptw.resp.bits.h_resp.gaf || io.hptw.resp.bits.h_resp.gpf) {
690          state(i) := state_mem_out
691          entries(i).hptw_resp := io.hptw.resp.bits.h_resp
692          entries(i).hptw_resp.gpf := io.hptw.resp.bits.h_resp.gpf || check_g_perm_fail
693          entries(i).first_s2xlate_fault := io.hptw.resp.bits.h_resp.gaf || io.hptw.resp.bits.h_resp.gpf
694        }.otherwise{ // change the entry that is waiting hptw resp
695          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))
696          val waiting_index = ParallelMux(need_to_waiting_vec zip entries.map(_.wait_id))
697          state(i) := Mux(Cat(need_to_waiting_vec).orR, state_mem_waiting, state_addr_check)
698          entries(i).hptw_resp := io.hptw.resp.bits.h_resp
699          entries(i).wait_id := Mux(Cat(need_to_waiting_vec).orR, waiting_index, entries(i).wait_id)
700          //To do: change the entry that is having the same hptw req
701        }
702      }
703      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) {
704        state(i) := state_mem_out
705        entries(i).hptw_resp := io.hptw.resp.bits.h_resp
706        //To do: change the entry that is having the same hptw req
707      }
708    }
709  }
710  when (io.out.fire) {
711    assert(state(mem_ptr) === state_mem_out)
712    state(mem_ptr) := state_idle
713  }
714  mem_resp_hit.map(a => when (a) { a := false.B } )
715
716  when (io.cache.fire) {
717    state(cache_ptr) := state_idle
718  }
719  XSError(io.out.fire && io.cache.fire && (mem_ptr === cache_ptr), "mem resp and cache fire at the same time at same entry")
720
721  when (flush) {
722    state.map(_ := state_idle)
723  }
724
725  io.in.ready := !full
726
727  io.out.valid := ParallelOR(is_having).asBool
728  io.out.bits.req_info := entries(mem_ptr).req_info
729  io.out.bits.id := mem_ptr
730  io.out.bits.af := entries(mem_ptr).af
731  io.out.bits.h_resp := entries(mem_ptr).hptw_resp
732  io.out.bits.first_s2xlate_fault := entries(mem_ptr).first_s2xlate_fault
733
734  val hptw_req_arb = Module(new Arbiter(new Bundle{
735      val source = UInt(bSourceWidth.W)
736      val id = UInt(log2Up(l2tlbParams.llptwsize).W)
737      val ppn = UInt(ptePPNLen.W)
738    } , 2))
739  // first stage 2 translation
740  hptw_req_arb.io.in(0).valid := hyper_arb1.io.out.valid
741  hptw_req_arb.io.in(0).bits.source := hyper_arb1.io.out.bits.req_info.source
742  hptw_req_arb.io.in(0).bits.ppn := hyper_arb1.io.out.bits.ppn
743  hptw_req_arb.io.in(0).bits.id := hyper_arb1.io.chosen
744  hyper_arb1.io.out.ready := hptw_req_arb.io.in(0).ready
745  // last stage 2 translation
746  hptw_req_arb.io.in(1).valid := hyper_arb2.io.out.valid
747  hptw_req_arb.io.in(1).bits.source := hyper_arb2.io.out.bits.req_info.source
748  hptw_req_arb.io.in(1).bits.ppn := hyper_arb2.io.out.bits.ppn
749  hptw_req_arb.io.in(1).bits.id := hyper_arb2.io.chosen
750  hyper_arb2.io.out.ready := hptw_req_arb.io.in(1).ready
751  hptw_req_arb.io.out.ready := io.hptw.req.ready
752  io.hptw.req.valid := hptw_req_arb.io.out.fire && !flush
753  io.hptw.req.bits.gvpn := hptw_req_arb.io.out.bits.ppn
754  io.hptw.req.bits.id := hptw_req_arb.io.out.bits.id
755  io.hptw.req.bits.source := hptw_req_arb.io.out.bits.source
756
757  io.mem.req.valid := mem_arb.io.out.valid && !flush
758  val mem_paddr = MakeAddr(mem_arb.io.out.bits.ppn, getVpnn(mem_arb.io.out.bits.req_info.vpn, 0))
759  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))
760  io.mem.req.bits.addr := Mux(mem_arb.io.out.bits.req_info.s2xlate === allStage, mem_hpaddr, mem_paddr)
761  io.mem.req.bits.id := mem_arb.io.chosen
762  io.mem.req.bits.hptw_bypassed := false.B
763  mem_arb.io.out.ready := io.mem.req.ready
764  val mem_refill_id = RegNext(io.mem.resp.bits.id(log2Up(l2tlbParams.llptwsize)-1, 0))
765  io.mem.refill := entries(mem_refill_id).req_info
766  io.mem.refill.s2xlate := entries(mem_refill_id).req_info.s2xlate
767  io.mem.buffer_it := mem_resp_hit
768  io.mem.enq_ptr := enq_ptr
769
770  io.cache.valid := Cat(is_cache).orR
771  io.cache.bits := ParallelMux(is_cache, entries.map(_.req_info))
772
773  XSPerfAccumulate("llptw_in_count", io.in.fire)
774  XSPerfAccumulate("llptw_in_block", io.in.valid && !io.in.ready)
775  for (i <- 0 until 7) {
776    XSPerfAccumulate(s"enq_state${i}", io.in.fire && enq_state === i.U)
777  }
778  for (i <- 0 until (l2tlbParams.llptwsize + 1)) {
779    XSPerfAccumulate(s"util${i}", PopCount(is_emptys.map(!_)) === i.U)
780    XSPerfAccumulate(s"mem_util${i}", PopCount(is_mems) === i.U)
781    XSPerfAccumulate(s"waiting_util${i}", PopCount(is_waiting) === i.U)
782  }
783  XSPerfAccumulate("mem_count", io.mem.req.fire)
784  XSPerfAccumulate("mem_cycle", PopCount(is_waiting) =/= 0.U)
785  XSPerfAccumulate("blocked_in", io.in.valid && !io.in.ready)
786
787  for (i <- 0 until l2tlbParams.llptwsize) {
788    TimeOutAssert(state(i) =/= state_idle, timeOutThreshold, s"missqueue time out no out ${i}")
789  }
790
791  val perfEvents = Seq(
792    ("tlbllptw_incount           ", io.in.fire               ),
793    ("tlbllptw_inblock           ", io.in.valid && !io.in.ready),
794    ("tlbllptw_memcount          ", io.mem.req.fire          ),
795    ("tlbllptw_memcycle          ", PopCount(is_waiting)       ),
796  )
797  generatePerfEvent()
798}
799
800/*========================= HPTW ==============================*/
801
802/** HPTW : Hypervisor Page Table Walker
803  * the page walker take the virtual machine's page walk.
804  * guest physical address translation, guest physical address -> host physical address
805  **/
806class HPTWIO()(implicit p: Parameters) extends MMUIOBaseBundle with HasPtwConst {
807  val req = Flipped(DecoupledIO(new Bundle {
808    val source = UInt(bSourceWidth.W)
809    val id = UInt(log2Up(l2tlbParams.llptwsize).W)
810    val gvpn = UInt(gvpnLen.W)
811    val ppn = UInt(ppnLen.W)
812    val l3Hit = if (EnableSv48) Some(new Bool()) else None
813    val l2Hit = Bool()
814    val l1Hit = Bool()
815    val bypassed = Bool() // if bypass, don't refill
816  }))
817  val resp = DecoupledIO(new Bundle {
818    val source = UInt(bSourceWidth.W)
819    val resp = Output(new HptwResp())
820    val id = Output(UInt(bMemID.W))
821  })
822
823  val mem = new Bundle {
824    val req = DecoupledIO(new L2TlbMemReqBundle())
825    val resp = Flipped(ValidIO(UInt(XLEN.W)))
826    val mask = Input(Bool())
827  }
828  val refill = Output(new Bundle {
829    val req_info = new L2TlbInnerBundle()
830    val level = UInt(log2Up(Level + 1).W)
831  })
832  val pmp = new Bundle {
833    val req = ValidIO(new PMPReqBundle())
834    val resp = Flipped(new PMPRespBundle())
835  }
836}
837
838class HPTW()(implicit p: Parameters) extends XSModule with HasPtwConst {
839  val io = IO(new HPTWIO)
840  val hgatp = io.csr.hgatp
841  val sfence = io.sfence
842  val flush = sfence.valid || hgatp.changed || io.csr.satp.changed || io.csr.vsatp.changed
843  val mode = hgatp.mode
844
845  val level = RegInit(3.U(log2Up(Level + 1).W))
846  val af_level = RegInit(3.U(log2Up(Level + 1).W)) // access fault return this level
847  val gpaddr = Reg(UInt(GPAddrBits.W))
848  val req_ppn = Reg(UInt(ppnLen.W))
849  val vpn = gpaddr(GPAddrBits-1, offLen)
850  val levelNext = level - 1.U
851  val l3Hit = Reg(Bool())
852  val l2Hit = Reg(Bool())
853  val l1Hit = Reg(Bool())
854  val bypassed = Reg(Bool())
855//  val pte = io.mem.resp.bits.MergeRespToPte()
856  val pte = io.mem.resp.bits.asTypeOf(new PteBundle().cloneType)
857  val ppn_l3 = Mux(l3Hit, req_ppn, pte.ppn)
858  val ppn_l2 = Mux(l2Hit, req_ppn, pte.ppn)
859  val ppn_l1 = Mux(l1Hit, req_ppn, pte.ppn)
860  val ppn = Wire(UInt(PAddrBits.W))
861  val p_pte = MakeAddr(ppn, getVpnn(vpn, level))
862  val pg_base = Wire(UInt(PAddrBits.W))
863  val mem_addr = Wire(UInt(PAddrBits.W))
864  if (EnableSv48) {
865    when (mode === Sv48) {
866      ppn := Mux(af_level === 2.U, ppn_l3, Mux(af_level === 1.U, ppn_l2, ppn_l1)) // for l2, l1 and l3
867      pg_base := MakeGPAddr(hgatp.ppn, getGVpnn(vpn, 3.U, mode = Sv48)) // for l3
868      mem_addr := Mux(af_level === 3.U, pg_base, p_pte)
869    } .otherwise {
870      ppn := Mux(af_level === 1.U, ppn_l2, ppn_l1) //for l1 and l2
871      pg_base := MakeGPAddr(hgatp.ppn, getGVpnn(vpn, 2.U, mode = Sv39))
872      mem_addr := Mux(af_level === 2.U, pg_base, p_pte)
873    }
874  } else {
875    ppn := Mux(af_level === 1.U, ppn_l2, ppn_l1) //for l1 and l2
876    pg_base := MakeGPAddr(hgatp.ppn, getGVpnn(vpn, 2.U, mode = Sv39))
877    mem_addr := Mux(af_level === 2.U, pg_base, p_pte)
878  }
879
880  //s/w register
881  val s_pmp_check = RegInit(true.B)
882  val s_mem_req = RegInit(true.B)
883  val w_mem_resp = RegInit(true.B)
884  val idle = RegInit(true.B)
885  val mem_addr_update = RegInit(false.B)
886  val finish = WireInit(false.B)
887
888  val sent_to_pmp = !idle && (!s_pmp_check || mem_addr_update) && !finish
889  val pageFault = pte.isGpf(level) || (!pte.isLeaf() && level === 0.U)
890  val accessFault = RegEnable(io.pmp.resp.ld || io.pmp.resp.mmio, sent_to_pmp)
891
892  val ppn_af = pte.isAf()
893  val find_pte = pte.isLeaf() || ppn_af || pageFault
894
895  val resp_valid = !idle && mem_addr_update && ((w_mem_resp && find_pte) || (s_pmp_check && accessFault))
896  val id = Reg(UInt(log2Up(l2tlbParams.llptwsize).W))
897  val source = RegEnable(io.req.bits.source, io.req.fire)
898
899  io.req.ready := idle
900  val resp = Wire(new HptwResp())
901  resp.apply(pageFault && !accessFault && !ppn_af, accessFault || ppn_af, Mux(accessFault, af_level, level), pte, vpn, hgatp.vmid)
902  io.resp.valid := resp_valid
903  io.resp.bits.id := id
904  io.resp.bits.resp := resp
905  io.resp.bits.source := source
906
907  io.pmp.req.valid := DontCare
908  io.pmp.req.bits.addr := mem_addr
909  io.pmp.req.bits.size := 3.U
910  io.pmp.req.bits.cmd := TlbCmd.read
911
912  io.mem.req.valid := !s_mem_req && !io.mem.mask && !accessFault && s_pmp_check
913  io.mem.req.bits.addr := mem_addr
914  io.mem.req.bits.id := HptwReqId.U(bMemID.W)
915  io.mem.req.bits.hptw_bypassed := bypassed
916
917  io.refill.req_info.vpn := vpn
918  io.refill.level := level
919  io.refill.req_info.source := source
920  io.refill.req_info.s2xlate := onlyStage2
921  when (idle){
922    when(io.req.fire){
923      bypassed := io.req.bits.bypassed
924      idle := false.B
925      gpaddr := Cat(io.req.bits.gvpn, 0.U(offLen.W))
926      accessFault := false.B
927      s_pmp_check := false.B
928      id := io.req.bits.id
929      req_ppn := io.req.bits.ppn
930      if (EnableSv48) {
931        when (mode === Sv48) {
932          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)))
933          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)))
934          l3Hit := io.req.bits.l3Hit.get
935        } .otherwise {
936          level := Mux(io.req.bits.l1Hit, 0.U, Mux(io.req.bits.l2Hit, 1.U, 2.U))
937          af_level := Mux(io.req.bits.l1Hit, 0.U, Mux(io.req.bits.l2Hit, 1.U, 2.U))
938          l3Hit := false.B
939        }
940      } else {
941        level := Mux(io.req.bits.l1Hit, 0.U, Mux(io.req.bits.l2Hit, 1.U, 2.U))
942        af_level := Mux(io.req.bits.l1Hit, 0.U, Mux(io.req.bits.l2Hit, 1.U, 2.U))
943        l3Hit := false.B
944      }
945      l2Hit := io.req.bits.l2Hit
946      l1Hit := io.req.bits.l1Hit
947    }
948  }
949
950  when(sent_to_pmp && !mem_addr_update){
951    s_mem_req := false.B
952    s_pmp_check := true.B
953  }
954
955  when(accessFault && !idle){
956    s_pmp_check := true.B
957    s_mem_req := true.B
958    w_mem_resp := true.B
959    mem_addr_update := true.B
960  }
961
962  when(io.mem.req.fire){
963    s_mem_req := true.B
964    w_mem_resp := false.B
965  }
966
967  when(io.mem.resp.fire && !w_mem_resp){
968    w_mem_resp := true.B
969    af_level := af_level - 1.U
970    mem_addr_update := true.B
971  }
972
973  when(mem_addr_update){
974    when(!(find_pte || accessFault)){
975      level := levelNext
976      s_mem_req := false.B
977      mem_addr_update := false.B
978    }.elsewhen(resp_valid){
979      when(io.resp.fire){
980        idle := true.B
981        mem_addr_update := false.B
982        accessFault := false.B
983      }
984      finish := true.B
985    }
986  }
987   when (flush) {
988    idle := true.B
989    s_pmp_check := true.B
990    s_mem_req := true.B
991    w_mem_resp := true.B
992    accessFault := false.B
993    mem_addr_update := false.B
994  }
995}
996