Merge branch 'main' of https://github.com/openhwgroup/cvw

.gitignore

@@ -112,4 +112,6 @@ sim/results-error/
 sim/test1.rep
 sim/vsim.log
 tests/coverage/*.elf
-*.elf.memfile
+*.elf.memfile
+sim/*Cache.log
+sim/branch

bin/CacheSim.py

@@ -0,0 +1,241 @@
+#!/usr/bin/env python3
+
+###########################################
+## CacheSim.py
+##
+## Written: lserafini@hmc.edu
+## Created: 27 March 2023
+## Modified: 5 April 2023
+##
+## Purpose: Simulate a L1 D$ or I$ for comparison with Wally
+##
+## A component of the CORE-V-WALLY configurable RISC-V project.
+##
+## Copyright (C) 2021-23 Harvey Mudd College & Oklahoma State University
+##
+## SPDX-License-Identifier: Apache-2.0 WITH SHL-2.1
+##
+## Licensed under the Solderpad Hardware License v 2.1 (the “License”); you may not use this file 
+## except in compliance with the License, or, at your option, the Apache License version 2.0. You 
+## may obtain a copy of the License at
+##
+## https:##solderpad.org/licenses/SHL-2.1/
+##
+## Unless required by applicable law or agreed to in writing, any work distributed under the 
+## License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, 
+## either express or implied. See the License for the specific language governing permissions 
+## and limitations under the License.
+################################################################################################
+
+# how to invoke this simulator: 
+# CacheSim.py <number of lines> <number of ways> <length of physical address> <length of tag> -f <log file> (-v)
+# so the default invocation for rv64gc is 'CacheSim.py 64 4 56 44 -f <log file>'
+# the log files to run this simulator on can be generated from testbench.sv
+# by setting I_CACHE_ADDR_LOGGER and/or D_CACHE_ADDR_LOGGER to 1 before running tests.
+# I (Lim) recommend logging a single set of tests (such as wally64priv) at a time.
+# This helps avoid unexpected logger behavior.
+# With verbose mode off, the simulator only reports mismatches between its and Wally's behavior.
+# With verbose mode on, the simulator logs each access into the cache.
+
+import sys
+import math
+import argparse
+import os
+
+class CacheLine:
+    def __init__(self):
+        self.tag = 0
+        self.valid = False
+        self.dirty = False
+    
+    def __str__(self):
+        string = "(V: " + str(self.valid) + ", D: " + str(self.dirty)
+        string +=  ", Tag: " + str(hex(self.tag)) + ")"
+        return string
+    
+    def __repr__(self):
+        return self.__str__()
+
+class Cache:
+    def __init__(self, numsets, numways, addrlen, taglen):
+        self.numways = numways
+        self.numsets = numsets
+
+        self.addrlen = addrlen
+        self.taglen = taglen
+        self.setlen = int(math.log(numsets, 2))
+        self.offsetlen = self.addrlen - self.taglen - self.setlen
+
+        self.ways = []
+        for i in range(numways):
+            self.ways.append([])
+            for j in range(numsets):
+                self.ways[i].append(CacheLine())
+        
+        self.pLRU = []
+        for i in range(self.numsets):
+            self.pLRU.append([0]*(self.numways-1))
+    
+    # flushes the cache by setting all dirty bits to False
+    def flush(self):
+        for way in self.ways:
+            for line in way:
+                line.dirty = False
+    
+    # invalidates the cache by setting all valid bits to False
+    def invalidate(self):
+        for way in self.ways:
+            for line in way:
+                line.valid = False
+    
+    # resets the pLRU to a fresh 2-D array of 0s
+    def clear_pLRU(self):
+        self.pLRU = []
+        for i in range(self.numsets):
+            self.pLRU.append([0]*(self.numways-1))
+    
+    # splits the given address into tag, set, and offset
+    def splitaddr(self, addr):
+        # no need for offset in the sim, but it's here for debug
+        tag = addr >> (self.setlen + self.offsetlen) & int('1'*self.taglen, 2)
+        setnum = (addr >> self.offsetlen) & int('1'*self.setlen, 2)
+        offset = addr & int('1'*self.offsetlen, 2)
+        return tag, setnum, offset
+    
+    # performs a cache access with the given address.
+    # returns a character representing the outcome:
+    # H/M/E/D - hit, miss, eviction, or eviction with writeback
+    def cacheaccess(self, addr, write=False):
+        tag, setnum, _ = self.splitaddr(addr)
+
+        # check our ways to see if we have a hit
+        for waynum in range(self.numways):
+            line = self.ways[waynum][setnum]
+            if line.tag == tag and line.valid:
+                line.dirty = line.dirty or write
+                self.update_pLRU(waynum, setnum)
+                return 'H'
+
+        # we didn't hit, but we may not need to evict.
+        # check for an empty way line.
+        for waynum in range(self.numways):
+            line = self.ways[waynum][setnum]
+            if not line.valid:
+                line.tag = tag
+                line.valid = True
+                line.dirty = write
+                self.update_pLRU(waynum, setnum)
+                return 'M'
+        
+        # we need to evict. Select a victim and overwrite.
+        victim = self.getvictimway(setnum)
+        line = self.ways[victim][setnum]
+        prevdirty = line.dirty
+        line.tag = tag
+        line.valid = True   # technically redundant
+        line.dirty = write
+        self.update_pLRU(victim, setnum)
+        return 'D' if prevdirty else 'E'
+
+    # updates the psuedo-LRU tree for the given set
+    # with an access to the given way
+    def update_pLRU(self, waynum, setnum):
+        if self.numways == 1:
+            return
+        
+        tree = self.pLRU[setnum]
+        bottomrow = (self.numways - 1)//2
+        index = (waynum // 2) + bottomrow
+        tree[index] = int(not (waynum % 2))
+        while index > 0:
+            parent = (index-1) // 2
+            tree[parent] = index % 2 
+            index = parent
+
+    # uses the psuedo-LRU tree to select
+    # a victim way from the given set
+    # returns the victim way as an integer
+    def getvictimway(self, setnum):
+        if self.numways == 1:
+            return 0
+        
+        tree = self.pLRU[setnum]
+        index = 0
+        bottomrow = (self.numways - 1) // 2 #first index on the bottom row of the tree
+        while index < bottomrow:
+            if tree[index] == 0:
+                # Go to the left child
+                index = index*2 + 1
+            else: #tree[index] == 1
+                # Go to the right child
+                index = index*2 + 2     
+        
+        victim = (index - bottomrow)*2
+        if tree[index] == 1:
+            victim += 1
+        
+        return victim
+    
+    def __str__(self):
+        string = ""
+        for i in range(self.numways):
+            string += "Way " + str(i) + ": "
+            for line in self.ways[i]:
+                string += str(line) + ", "
+            string += "\n\n"
+        return string
+
+    def __repr__(self):
+        return self.__str__()
+    
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Simulates a L1 cache.")
+    parser.add_argument('numlines', type=int, help="The number of lines per way (a power of 2)", metavar="L")
+    parser.add_argument('numways', type=int, help="The number of ways (a power of 2)", metavar='W')
+    parser.add_argument('addrlen', type=int, help="Length of the address in bits (a power of 2)", metavar="A")
+    parser.add_argument('taglen', type=int, help="Length of the tag in bits", metavar="T")
+    parser.add_argument('-f', "--file", required=True, help="Log file to simulate from")
+    parser.add_argument('-v', "--verbose", action='store_true', help="verbose/full-trace mode")
+
+    args = parser.parse_args()
+    cache = Cache(args.numlines, args.numways, args.addrlen, args.taglen)
+    #numtests = -1
+    extfile = os.path.expanduser(args.file)
+    with open(extfile, "r") as f:
+        for ln in f:
+            ln = ln.strip()
+            lninfo = ln.split()
+            if len(lninfo) < 3: #non-address line
+                if len(lninfo) > 0 and (lninfo[0] == 'BEGIN' or lninfo[0] == 'TRAIN'):
+                    # currently BEGIN and END traces aren't being recorded correctly
+                    # trying TRAIN clears instead
+                    cache.invalidate() # a new test is starting, so 'empty' the cache
+                    cache.clear_pLRU()
+                    #numtests +=1
+                    if args.verbose:
+                        print("New Test")
+                        
+            else:
+                if lninfo[1] == 'F':
+                    cache.flush()
+                    if args.verbose:
+                        print("F")
+                elif lninfo[1] == 'I':
+                    cache.invalidate()
+                    if args.verbose:
+                        print("I")
+                else:
+                    addr = int(lninfo[0], 16)
+                    iswrite = lninfo[1] == 'W' or lninfo[1] == 'A'
+                    result = cache.cacheaccess(addr, iswrite)
+                    if args.verbose:
+                        tag, setnum, offset = cache.splitaddr(addr)
+                        print(hex(addr), hex(tag), hex(setnum), hex(offset), lninfo[2], result)
+                    if not result == lninfo[2]:
+                        print("Result mismatch at address", lninfo[0], ". Wally:", lninfo[2],", Sim:", result) #, "in test", numtests)
+                        
+                        
+
+    
+

sim/imperas.ic

@@ -4,24 +4,36 @@
 --showcommands
 
 # Core settings
+--override cpu/priv_version=1.12 
+--override cpu/user_version=20191213
+# arch
+--override cpu/mimpid=0x100
+--override refRoot/cpu/tvec_align=64
+
+#  clarify
+#--override refRoot/cpu/mtvec_sext=F
+
+--override cpu/tval_ii_code=T
+
+#--override cpu/time_undefined=T
+#--override cpu/cycle_undefined=T
+#--override cpu/instret_undefined=T
+#--override cpu/hpmcounter_undefined=T
+
+--override cpu/reset_address=0x80000000
+
 --override cpu/unaligned=F
 --override cpu/ignore_non_leaf_DAU=1
-#--override cpu/wfi_is_nop=T
+--override cpu/wfi_is_nop=T
---override cpu/mimpid=0x100
 --override cpu/misa_Extensions_mask=0x0
+#--override cpu/updatePTEA=T
+#--override cpu/updatePTED=T
+--override cpu/Sstc=T
 
+# THIS NEEDS FIXING to 16
 --override cpu/PMP_registers=16
 --override cpu/PMP_undefined=T
 
-# Wally-specific non-default configuraiton
---override refRoot/cpu/Sstc=T 
---override cpu/add_implicit_Extensions=B 
---override cpu/bitmanip_version=1.0.0
-
-# Illegal instruction should not contain the bit pattern
-# illegal pmp read contained this
-# --override cpu/tval_ii_code=F
-
 # PMA Settings 
 # 'r': read access allowed
 # 'w': write access allowed
@@ -51,19 +63,11 @@
 #-override  refRoot/cpu/cv/cover=basic
 #-override  refRoot/cpu/cv/extensions=RV32I
 
-
-
 # Add Imperas simulator application instruction tracing
---override cpu/show_c_prefix=T
-
---trace --tracechange --traceshowicount --tracemode -tracemem ASX --monitornetschange --traceafter 800000
-
-# Exceptions and pagetables debug
---override cpu/debugflags=6
-
-# Turn on verbose output for Imperas simulator and Model
 --verbose
---override  cpu/verbose=1
+--trace --tracechange --traceshowicount --tracemode -tracemem ASX --monitornetschange --traceafter 0
+--override cpu/debugflags=6 --override cpu/verbose=1
+--override cpu/show_c_prefix=T
 
 # Store simulator output to logfile
 --output imperas.log

src/hazard/hazard.sv

@@ -36,7 +36,7 @@ module hazard (
   input logic  FCvtIntStallD, FPUStallD,
   input logic  DivBusyE, FDivBusyE,
   input logic  EcallFaultM, BreakpointFaultM,
-  input logic  WFIStallM,
+  input logic  wfiM, IntPendingM,
   // Stall & flush outputs
   output logic StallF, StallD, StallE, StallM, StallW,
   output logic FlushD, FlushE, FlushM, FlushW
@@ -45,6 +45,12 @@ module hazard (
   logic                                       StallFCause, StallDCause, StallECause, StallMCause, StallWCause;
   logic                                       LatestUnstalledD, LatestUnstalledE, LatestUnstalledM, LatestUnstalledW;
   logic                                       FlushDCause, FlushECause, FlushMCause, FlushWCause;
+
+  logic WFIStallM, WFIInterruptedM;
+
+  // WFI logic
+  assign WFIStallM = wfiM & ~IntPendingM;         // WFI waiting for an interrupt or timeout
+  assign WFIInterruptedM = wfiM & IntPendingM;    // WFI detects a pending interrupt.  Retire WFI; trap if interrupt is enabled.
   
   // stalls and flushes
   // loads: stall for one cycle if the subsequent instruction depends on the load
@@ -68,7 +74,7 @@ module hazard (
   assign FlushDCause = TrapM | RetM | CSRWriteFenceM | BPWrongE;
   assign FlushECause = TrapM | RetM | CSRWriteFenceM |(BPWrongE & ~(DivBusyE | FDivBusyE));
   assign FlushMCause = TrapM | RetM | CSRWriteFenceM;
-  assign FlushWCause = TrapM;
+  assign FlushWCause = TrapM & ~WFIInterruptedM;
 
   // Stall causes
   //  Most data depenency stalls are identified in the decode stage

src/lsu/lsu.sv

@@ -62,7 +62,7 @@ module lsu (
   output logic                LoadPageFaultM, StoreAmoPageFaultM,   // Page fault exceptions
   output logic                LoadMisalignedFaultM,                 // Load address misaligned fault
   output logic                LoadAccessFaultM,                     // Load access fault (PMA)
-  output logic                HPTWInstrAccessFaultM,                // HPTW generated access fault during instruction fetch
+  output logic                HPTWInstrAccessFaultF,                // HPTW generated access fault during instruction fetch
   // cpu hazard unit (trap)
   output logic                StoreAmoMisalignedFaultM,             // Store or AMO address misaligned fault
   output logic                StoreAmoAccessFaultM,                 // Store or AMO access fault
@@ -159,7 +159,7 @@ module lsu (
       .IEUAdrExtM, .PTE, .IHWriteDataM, .PageType, .PreLSURWM, .LSUAtomicM,
       .IHAdrM, .HPTWStall, .SelHPTW,
       .IgnoreRequestTLB, .LSULoadAccessFaultM, .LSUStoreAmoAccessFaultM, 
-      .LoadAccessFaultM, .StoreAmoAccessFaultM, .HPTWInstrAccessFaultM);
+      .LoadAccessFaultM, .StoreAmoAccessFaultM, .HPTWInstrAccessFaultF);
   end else begin // No HPTW, so signals are not multiplexed
     assign PreLSURWM = MemRWM; 
     assign IHAdrM = IEUAdrExtM;
@@ -170,7 +170,7 @@ module lsu (
     assign LoadAccessFaultM = LSULoadAccessFaultM;
     assign StoreAmoAccessFaultM = LSUStoreAmoAccessFaultM;   
     assign {HPTWStall, SelHPTW, PTE, PageType, DTLBWriteM, ITLBWriteF, IgnoreRequestTLB} = '0;
-    assign HPTWInstrAccessFaultM = '0;
+    assign HPTWInstrAccessFaultF = '0;
    end
 
   // CommittedM indicates the cache, bus, or HPTW are busy with a multiple cycle operation.

src/mmu/hptw.sv

@@ -64,7 +64,7 @@ module hptw (
   output logic             SelHPTW,
   output logic             HPTWStall,
   input  logic             LSULoadAccessFaultM, LSUStoreAmoAccessFaultM, 
-  output logic             LoadAccessFaultM, StoreAmoAccessFaultM, HPTWInstrAccessFaultM
+  output logic             LoadAccessFaultM, StoreAmoAccessFaultM, HPTWInstrAccessFaultF
 );
 
   typedef enum logic [3:0] {L0_ADR, L0_RD, 
@@ -98,12 +98,25 @@ module hptw (
   logic [1:0]              HPTWRW;
   logic [2:0]              HPTWSize; // 32 or 64 bit access
   statetype                WalkerState, NextWalkerState, InitialWalkerState;
+  logic                    HPTWLoadAccessFault, HPTWStoreAmoAccessFault, HPTWInstrAccessFault;
+  logic                    HPTWLoadAccessFaultDelay, HPTWStoreAmoAccessFaultDelay, HPTWInstrAccessFaultDelay;
+  logic                    HPTWAccessFaultDelay;
+  logic                    TakeHPTWFault, TakeHPTWFaultDelay;
 
   // map hptw access faults onto either the original LSU load/store fault or instruction access fault
   assign LSUAccessFaultM       = LSULoadAccessFaultM | LSUStoreAmoAccessFaultM;
-  assign LoadAccessFaultM      = WalkerState == IDLE ? LSULoadAccessFaultM : LSUAccessFaultM & DTLBWalk & MemRWM[1] & ~MemRWM[0];
+  assign HPTWLoadAccessFault   = LSUAccessFaultM & DTLBWalk & MemRWM[1] & ~MemRWM[0];
-  assign StoreAmoAccessFaultM  = WalkerState == IDLE ? LSUStoreAmoAccessFaultM : LSUAccessFaultM & DTLBWalk & MemRWM[0];
+  assign HPTWStoreAmoAccessFault = LSUAccessFaultM & DTLBWalk & MemRWM[0];
-  assign HPTWInstrAccessFaultM = WalkerState == IDLE ? 1'b0: LSUAccessFaultM & ~DTLBWalk;
+  assign HPTWInstrAccessFault    = LSUAccessFaultM & ~DTLBWalk;
+
+  flopr #(4) HPTWAccesFaultReg(clk, reset, {TakeHPTWFault, HPTWLoadAccessFault, HPTWStoreAmoAccessFault, HPTWInstrAccessFault},
+                               {TakeHPTWFaultDelay, HPTWLoadAccessFaultDelay, HPTWStoreAmoAccessFaultDelay, HPTWInstrAccessFaultDelay});
+
+  assign TakeHPTWFault = WalkerState != IDLE;
+  
+  assign LoadAccessFaultM      = TakeHPTWFaultDelay ? HPTWLoadAccessFaultDelay : LSULoadAccessFaultM;
+  assign StoreAmoAccessFaultM  = TakeHPTWFaultDelay ? HPTWStoreAmoAccessFaultDelay : LSUStoreAmoAccessFaultM;
+  assign HPTWInstrAccessFaultF = TakeHPTWFaultDelay ? HPTWInstrAccessFaultDelay : 1'b0;
 
   // Extract bits from CSRs and inputs
   assign SvMode = SATP_REGW[`XLEN-1:`XLEN-`SVMODE_BITS];
@@ -247,22 +260,26 @@ module hptw (
   flopenl #(.TYPE(statetype)) WalkerStateReg(clk, reset | FlushW, 1'b1, NextWalkerState, IDLE, WalkerState); 
   always_comb 
     case (WalkerState)
-      IDLE:       if (TLBMiss & ~DCacheStallM)                        NextWalkerState = InitialWalkerState;
+      IDLE:       if (TLBMiss & ~DCacheStallM & ~HPTWAccessFaultDelay) NextWalkerState = InitialWalkerState;
                   else                                                NextWalkerState = IDLE;
       L3_ADR:                                                         NextWalkerState = L3_RD; // first access in SV48
       L3_RD:      if (DCacheStallM)                                   NextWalkerState = L3_RD;
+                  else if(LSUAccessFaultM)                            NextWalkerState = IDLE;
                   else                                                NextWalkerState = L2_ADR;
       L2_ADR:     if (InitialWalkerState == L2_ADR | ValidNonLeafPTE) NextWalkerState = L2_RD; // first access in SV39
                   else                                                NextWalkerState = LEAF;
       L2_RD:      if (DCacheStallM)                                   NextWalkerState = L2_RD;
+                  else if(LSUAccessFaultM)                            NextWalkerState = IDLE;
                   else                                                NextWalkerState = L1_ADR;
       L1_ADR:     if (InitialWalkerState == L1_ADR | ValidNonLeafPTE) NextWalkerState = L1_RD; // first access in SV32
                   else                                                NextWalkerState = LEAF;  
       L1_RD:      if (DCacheStallM)                                   NextWalkerState = L1_RD;
+                  else if(LSUAccessFaultM)                            NextWalkerState = IDLE;
                   else                                                NextWalkerState = L0_ADR;
       L0_ADR:     if (ValidNonLeafPTE)                                NextWalkerState = L0_RD;
                   else                                                NextWalkerState = LEAF;
       L0_RD:      if (DCacheStallM)                                   NextWalkerState = L0_RD;
+                  else if(LSUAccessFaultM)                            NextWalkerState = IDLE;
                   else                                                NextWalkerState = LEAF;
       LEAF:       if (`SVADU_SUPPORTED & HPTWUpdateDA)                NextWalkerState = UPDATE_PTE;
                   else                                                NextWalkerState = IDLE;
@@ -273,7 +290,8 @@ module hptw (
 
   assign IgnoreRequestTLB = WalkerState == IDLE & TLBMiss;
   assign SelHPTW = WalkerState != IDLE;
-  assign HPTWStall = (WalkerState != IDLE) | (WalkerState == IDLE & TLBMiss);
+  assign HPTWAccessFaultDelay = HPTWLoadAccessFaultDelay | HPTWStoreAmoAccessFaultDelay | HPTWInstrAccessFaultDelay;
+  assign HPTWStall = (WalkerState != IDLE) | (WalkerState == IDLE & TLBMiss & ~(HPTWAccessFaultDelay));
 
   assign ITLBMissOrUpdateDAF = ITLBMissF | (`SVADU_SUPPORTED & InstrUpdateDAF);
   assign DTLBMissOrUpdateDAM = DTLBMissM | (`SVADU_SUPPORTED & DataUpdateDAM);  

src/privileged/privileged.sv

@@ -65,7 +65,7 @@ module privileged (
   // fault sources                                                         
   input  logic             InstrAccessFaultF,                              // instruction access fault
   input  logic             LoadAccessFaultM, StoreAmoAccessFaultM,         // load or store access fault
-  input  logic             HPTWInstrAccessFaultM,                          // hardware page table access fault while fetching instruction PTE
+  input  logic             HPTWInstrAccessFaultF,                          // hardware page table access fault while fetching instruction PTE
   input  logic             InstrPageFaultF,                                // page faults
   input  logic             LoadPageFaultM, StoreAmoPageFaultM,             // page faults
   input  logic             InstrMisalignedFaultM,                          // misaligned instruction fault
@@ -93,7 +93,7 @@ module privileged (
   output logic             BigEndianM,                                     // Use big endian in current privilege mode
   // Fault outputs                                                         
   output logic             BreakpointFaultM, EcallFaultM,                  // breakpoint and Ecall traps should retire
-  output logic             WFIStallM                                       // Stall in Memory stage for WFI until interrupt or timeout
+  output logic             wfiM, IntPendingM                               // Stall in Memory stage for WFI until interrupt pending or timeout
 );                                                                         
                                                                            
   logic [3:0]              CauseM;                                         // trap cause
@@ -110,10 +110,10 @@ module privileged (
   logic [11:0]             MIP_REGW, MIE_REGW;                             // interrupt pending and enable bits
   logic [1:0]              NextPrivilegeModeM;                             // next privilege mode based on trap or return
   logic                    DelegateM;                                      // trap should be delegated
-  logic                    wfiM;                                           // wait for interrupt instruction
-  logic                    IntPendingM;                                    // interrupt is pending, even if not enabled.  ends wfi
   logic                    InterruptM;                                     // interrupt occuring
   logic                    ExceptionM;                                     // Memory stage instruction caused a fault
+  logic                    HPTWInstrAccessFaultM;                          // Hardware page table access fault while fetching instruction PTE
+  
  
   // track the current privilege level
   privmode privmode(.clk, .reset, .StallW, .TrapM, .mretM, .sretM, .DelegateM,
@@ -144,8 +144,8 @@ module privileged (
 
   // pipeline early-arriving trap sources
   privpiperegs ppr(.clk, .reset, .StallD, .StallE, .StallM, .FlushD, .FlushE, .FlushM,
-    .InstrPageFaultF, .InstrAccessFaultF, .IllegalIEUFPUInstrD, 
+    .InstrPageFaultF, .InstrAccessFaultF, .HPTWInstrAccessFaultF, .IllegalIEUFPUInstrD, 
-    .InstrPageFaultM, .InstrAccessFaultM, .IllegalIEUFPUInstrM);
+    .InstrPageFaultM, .InstrAccessFaultM, .HPTWInstrAccessFaultM, .IllegalIEUFPUInstrM);
 
   // trap logic
   trap trap(.reset,
@@ -156,7 +156,7 @@ module privileged (
     .mretM, .sretM, .PrivilegeModeW, 
     .MIP_REGW, .MIE_REGW, .MIDELEG_REGW, .MEDELEG_REGW, .STATUS_MIE, .STATUS_SIE,
     .InstrValidM, .CommittedM, .CommittedF,
-    .TrapM, .RetM, .wfiM, .InterruptM, .ExceptionM, .IntPendingM, .DelegateM, .WFIStallM, .CauseM);
+    .TrapM, .RetM, .wfiM, .InterruptM, .ExceptionM, .IntPendingM, .DelegateM, .CauseM);
 endmodule
 
 

src/privileged/privpiperegs.sv

@@ -33,24 +33,26 @@ module privpiperegs (
   input  logic         StallD, StallE, StallM,
   input  logic         FlushD, FlushE, FlushM,
   input  logic         InstrPageFaultF, InstrAccessFaultF,  // instruction faults
+  input  logic         HPTWInstrAccessFaultF,               // hptw fault during instruction page fetch
   input  logic         IllegalIEUFPUInstrD,                 // illegal IEU instruction decoded
   output logic         InstrPageFaultM, InstrAccessFaultM,  // delayed instruction faults
-  output logic         IllegalIEUFPUInstrM                  // delayed illegal IEU instruction
+  output logic         IllegalIEUFPUInstrM,                 // delayed illegal IEU instruction
+  output logic         HPTWInstrAccessFaultM                // hptw fault during instruction page fetch
 );
 
   // Delayed fault signals
-  logic                InstrPageFaultD, InstrAccessFaultD;
+  logic                InstrPageFaultD, InstrAccessFaultD, HPTWInstrAccessFaultD;
-  logic                InstrPageFaultE, InstrAccessFaultE;
+  logic                InstrPageFaultE, InstrAccessFaultE, HPTWInstrAccessFaultE;
   logic                IllegalIEUFPUInstrE; 
 
   // pipeline fault signals
-  flopenrc #(2) faultregD(clk, reset, FlushD, ~StallD,
+  flopenrc #(3) faultregD(clk, reset, FlushD, ~StallD,
-                  {InstrPageFaultF, InstrAccessFaultF},
+                  {InstrPageFaultF, InstrAccessFaultF, HPTWInstrAccessFaultF},
-                  {InstrPageFaultD, InstrAccessFaultD});
+                  {InstrPageFaultD, InstrAccessFaultD, HPTWInstrAccessFaultD});
-  flopenrc #(3) faultregE(clk, reset, FlushE, ~StallE,
+  flopenrc #(4) faultregE(clk, reset, FlushE, ~StallE,
-                  {IllegalIEUFPUInstrD, InstrPageFaultD, InstrAccessFaultD}, 
+                  {IllegalIEUFPUInstrD, InstrPageFaultD, InstrAccessFaultD, HPTWInstrAccessFaultD}, 
-                  {IllegalIEUFPUInstrE, InstrPageFaultE, InstrAccessFaultE});
+                  {IllegalIEUFPUInstrE, InstrPageFaultE, InstrAccessFaultE, HPTWInstrAccessFaultE});
-  flopenrc #(3) faultregM(clk, reset, FlushM, ~StallM,
+  flopenrc #(4) faultregM(clk, reset, FlushM, ~StallM,
-                  {IllegalIEUFPUInstrE, InstrPageFaultE, InstrAccessFaultE},
+                  {IllegalIEUFPUInstrE, InstrPageFaultE, InstrAccessFaultE, HPTWInstrAccessFaultE},
-                  {IllegalIEUFPUInstrM, InstrPageFaultM, InstrAccessFaultM});
+                  {IllegalIEUFPUInstrM, InstrPageFaultM, InstrAccessFaultM, HPTWInstrAccessFaultM});
-endmodule
+endmodule

src/privileged/trap.sv

@@ -48,7 +48,6 @@ module trap (
   output logic                 ExceptionM,                                      // exception is occurring
   output logic                 IntPendingM,                                     // Interrupt is pending, might occur if enabled
   output logic                 DelegateM,                                       // Delegate trap to supervisor handler
-  output logic                 WFIStallM,                                       // Stall due to WFI instruction
   output logic [3:0]           CauseM                                           // trap cause
 );
 
@@ -74,7 +73,6 @@ module trap (
   assign InterruptM = (|ValidIntsM) & InstrValidM; // suppress interrupt if the memory system has partially processed a request.
   assign DelegateM = `S_SUPPORTED & (InterruptM ? MIDELEG_REGW[CauseM] : MEDELEG_REGW[CauseM]) & 
                      (PrivilegeModeW == `U_MODE | PrivilegeModeW == `S_MODE);
-  assign WFIStallM = wfiM & ~IntPendingM;
 
   ///////////////////////////////////////////
   // Trigger Traps and RET

src/wally/wallypipelinedcore.sv

@@ -106,7 +106,7 @@ module wallypipelinedcore (
   logic [1:0]                    PrivilegeModeW;
   logic [`XLEN-1:0]              PTE;
   logic [1:0]                    PageType;
-  logic                          sfencevmaM, WFIStallM;
+  logic                          sfencevmaM;
   logic                          SelHPTW;
 
   // PMA checker signals
@@ -146,7 +146,7 @@ module wallypipelinedcore (
   logic                          RASPredPCWrongM;
   logic                          IClassWrongM;
   logic [3:0]                    InstrClassM;
-  logic                          InstrAccessFaultF, HPTWInstrAccessFaultM;
+  logic                          InstrAccessFaultF, HPTWInstrAccessFaultF;
   logic [2:0]                    LSUHSIZE;
   logic [2:0]                    LSUHBURST;
   logic [1:0]                    LSUHTRANS;
@@ -162,7 +162,8 @@ module wallypipelinedcore (
   logic                          CommittedF;
   logic                          BranchD, BranchE, JumpD, JumpE;
   logic                          DCacheStallM, ICacheStallF;
-  
+  logic                          wfiM, IntPendingM;
+
   // instruction fetch unit: PC, branch prediction, instruction cache
   ifu ifu(.clk, .reset,
     .StallF, .StallD, .StallE, .StallM, .StallW, .FlushD, .FlushE, .FlushM, .FlushW,
@@ -236,7 +237,7 @@ module wallypipelinedcore (
     .StoreAmoPageFaultM, // connects to privilege
     .LoadMisalignedFaultM, // connects to privilege
     .LoadAccessFaultM,         // connects to privilege
-    .HPTWInstrAccessFaultM,         // connects to privilege
+    .HPTWInstrAccessFaultF,         // connects to privilege
     .StoreAmoMisalignedFaultM, // connects to privilege
     .StoreAmoAccessFaultM,     // connects to privilege
     .InstrUpdateDAF,
@@ -265,7 +266,7 @@ module wallypipelinedcore (
     .FCvtIntStallD, .FPUStallD,
     .DivBusyE, .FDivBusyE,
     .EcallFaultM, .BreakpointFaultM,
-    .WFIStallM,
+    .wfiM, .IntPendingM,
     // Stall & flush outputs
     .StallF, .StallD, .StallE, .StallM, .StallW,
     .FlushD, .FlushE, .FlushM, .FlushW);    
@@ -288,17 +289,18 @@ module wallypipelinedcore (
       .LoadMisalignedFaultM, .StoreAmoMisalignedFaultM,
       .MTimerInt, .MExtInt, .SExtInt, .MSwInt,
       .MTIME_CLINT, .IEUAdrM, .SetFflagsM,
-      .InstrAccessFaultF, .HPTWInstrAccessFaultM, .LoadAccessFaultM, .StoreAmoAccessFaultM, .SelHPTW,
+      .InstrAccessFaultF, .HPTWInstrAccessFaultF, .LoadAccessFaultM, .StoreAmoAccessFaultM, .SelHPTW,
       .PrivilegeModeW, .SATP_REGW,
       .STATUS_MXR, .STATUS_SUM, .STATUS_MPRV, .STATUS_MPP, .STATUS_FS,
       .PMPCFG_ARRAY_REGW, .PMPADDR_ARRAY_REGW, 
-      .FRM_REGW,.BreakpointFaultM, .EcallFaultM, .WFIStallM, .BigEndianM);
+      .FRM_REGW,.BreakpointFaultM, .EcallFaultM, .wfiM, .IntPendingM, .BigEndianM);
   end else begin
     assign CSRReadValW = 0;
     assign UnalignedPCNextF = PC2NextF;
     assign RetM = 0;
     assign TrapM = 0;
-    assign WFIStallM = 0;
+    assign wfiM = 0;
+    assign IntPendingM = 0;
     assign sfencevmaM = 0;
     assign BigEndianM = 0;
   end

testbench/testbench-linux-imperas.sv

@@ -413,10 +413,11 @@ module testbench;
         end
       end
 
-      always @(dut.core.MTimerInt) void'(rvvi.net_push("MTimerInterrupt",    dut.core.MTimerInt));
+      always @(dut.core.MTimerInt)   void'(rvvi.net_push("MTimerInterrupt",    dut.core.MTimerInt));
-      always @(dut.core.MExtInt)   void'(rvvi.net_push("MExternalInterrupt", dut.core.MExtInt));
+      always @(dut.core.MExtInt)     void'(rvvi.net_push("MExternalInterrupt", dut.core.MExtInt));
-      always @(dut.core.SExtInt)   void'(rvvi.net_push("SExternalInterrupt", dut.core.SExtInt));
+      always @(dut.core.SExtInt)     void'(rvvi.net_push("SExternalInterrupt", dut.core.SExtInt));
-      always @(dut.core.MSwInt)    void'(rvvi.net_push("MSWInterrupt",       dut.core.MSwInt));
+      always @(dut.core.MSwInt)      void'(rvvi.net_push("MSWInterrupt",       dut.core.MSwInt));
+      always @(dut.core.priv.priv.csr.csrs.csrs.STimerInt) void'(rvvi.net_push("STimerInterrupt", dut.core.priv.priv.csr.csrs.csrs.STimerInt));
 
       final begin
         void'(rvviRefShutdown());

testbench/testbench.sv

@@ -480,7 +480,7 @@ logic [3:0] dummy;
 	  assign EndSample = DCacheFlushStart & ~DCacheFlushDone;
 
 	  flop #(1) BeginReg(clk, StartSampleFirst, BeginDelayed);
-	  assign Begin = StartSampleFirst & ~ BeginDelayed;
+	  assign Begin = StartSampleFirst & ~BeginDelayed;
 
 	end
 	
@@ -555,12 +555,16 @@ logic [3:0] dummy;
 end
 
 
-  if (`ICACHE_SUPPORTED && `I_CACHE_ADDR_LOGGER) begin
+  if (`ICACHE_SUPPORTED && `I_CACHE_ADDR_LOGGER) begin : ICacheLogger
     int    file;
 	string LogFile;
 	logic  resetD, resetEdge;
     logic  Enable;
-    assign Enable = ~dut.core.StallD & ~dut.core.FlushD & dut.core.ifu.bus.icache.CacheRWF[1] & ~reset;
+    // assign Enable = ~dut.core.StallD & ~dut.core.FlushD & dut.core.ifu.bus.icache.CacheRWF[1] & ~reset;
+    
+    // this version of Enable allows for accurate eviction logging.
+    // Likely needs further improvement.
+    assign Enable = dut.core.ifu.bus.icache.icache.cachefsm.LRUWriteEn & ~reset;
 	flop #(1) ResetDReg(clk, reset, resetD);
 	assign resetEdge = ~reset & resetD;
     initial begin
@@ -568,50 +572,64 @@ end
       file = $fopen(LogFile, "w");
 	  $fwrite(file, "BEGIN %s\n", memfilename);
 	end
-    string HitMissString;
+    string AccessTypeString, HitMissString;
-    assign HitMissString = dut.core.ifu.bus.icache.icache.CacheHit ? "H" : "M";
+    assign HitMissString = dut.core.ifu.bus.icache.icache.CacheHit ? "H" :
+                           dut.core.ifu.bus.icache.icache.vict.cacheLRU.AllValid ? "E" : "M";
+    assign AccessTypeString = dut.core.ifu.InvalidateICacheM ? "I" : "R";
     always @(posedge clk) begin
 	  if(resetEdge) $fwrite(file, "TRAIN\n");
 	  if(Begin) $fwrite(file, "BEGIN %s\n", memfilename);
 	  if(Enable) begin  // only log i cache reads
-	    $fwrite(file, "%h R %s\n", dut.core.ifu.PCPF, HitMissString);
+	    $fwrite(file, "%h %s %s\n", dut.core.ifu.PCPF, AccessTypeString, HitMissString);
 	  end
 	  if(EndSample) $fwrite(file, "END %s\n", memfilename);
     end
   end
 
-  if (`DCACHE_SUPPORTED && `D_CACHE_ADDR_LOGGER) begin
+
+  if (`DCACHE_SUPPORTED && `D_CACHE_ADDR_LOGGER) begin : DCacheLogger
     int    file;
 	string LogFile;
 	logic  resetD, resetEdge;
-    string HitMissString;
+    logic  Enabled;
+    string AccessTypeString, HitMissString;
+
 	flop #(1) ResetDReg(clk, reset, resetD);
 	assign resetEdge = ~reset & resetD;
-    assign HitMissString = dut.core.lsu.bus.dcache.dcache.CacheHit ? "H" : "M";
+    assign HitMissString = dut.core.lsu.bus.dcache.dcache.CacheHit ? "H" :
+                           (!dut.core.lsu.bus.dcache.dcache.vict.cacheLRU.AllValid) ? "M" :
+                           dut.core.lsu.bus.dcache.dcache.LineDirty ? "D" : "E";
+    assign AccessTypeString = dut.core.lsu.bus.dcache.FlushDCache ? "F" :
+                              dut.core.lsu.bus.dcache.CacheAtomicM[1] ? "A" :
+                              dut.core.lsu.bus.dcache.CacheRWM == 2'b10 ? "R" : 
+                              dut.core.lsu.bus.dcache.CacheRWM == 2'b01 ? "W" :
+                              "NULL";
+    // assign Enabled = (dut.core.lsu.bus.dcache.dcache.cachefsm.CurrState == 0) &
+    //                  ~dut.core.lsu.bus.dcache.dcache.cachefsm.FlushStage &
+    //                  (AccessTypeString != "NULL");
+
+    // This version of enable allows for accurate eviction logging. 
+    // Likely needs further improvement.
+    assign Enabled = dut.core.lsu.bus.dcache.dcache.cachefsm.LRUWriteEn &
+                     ~dut.core.lsu.bus.dcache.dcache.cachefsm.FlushStage &
+                     (AccessTypeString != "NULL");
+
     initial begin
-	  LogFile = $psprintf("DCache.log");
+	    LogFile = $psprintf("DCache.log");
       file = $fopen(LogFile, "w");
-	  $fwrite(file, "BEGIN %s\n", memfilename);
+	    $fwrite(file, "BEGIN %s\n", memfilename);
-	end
+	  end
     always @(posedge clk) begin
 	  if(resetEdge) $fwrite(file, "TRAIN\n");
 	  if(Begin) $fwrite(file, "BEGIN %s\n", memfilename);
-	  if(~dut.core.StallW & ~dut.core.FlushW & dut.core.InstrValidM) begin
+	  if(Enabled) begin
-        if(dut.core.lsu.bus.dcache.CacheRWM == 2'b10) begin
+	    $fwrite(file, "%h %s %s\n", dut.core.lsu.PAdrM, AccessTypeString, HitMissString);
-	      $fwrite(file, "%h R %s\n", dut.core.lsu.PAdrM, HitMissString);
-        end else if (dut.core.lsu.bus.dcache.CacheRWM == 2'b01) begin
-	      $fwrite(file, "%h W %s\n", dut.core.lsu.PAdrM, HitMissString);
-        end else if (dut.core.lsu.bus.dcache.CacheAtomicM[1] == 1'b1) begin // *** This may change
-	      $fwrite(file, "%h A %s\n", dut.core.lsu.PAdrM, HitMissString);
-        end else if (dut.core.lsu.bus.dcache.FlushDCache) begin
-	      $fwrite(file, "%h F %s\n", dut.core.lsu.PAdrM, HitMissString);
-        end
 	  end
 	  if(EndSample) $fwrite(file, "END %s\n", memfilename);
     end
   end
 
-  if (`BPRED_SUPPORTED) begin
+  if (`BPRED_SUPPORTED) begin : BranchLogger
     if (`BPRED_LOGGER) begin
       string direction;
       int    file;

testbench/testbench_imperas.sv

@@ -198,10 +198,12 @@ module testbench;
       
     end
 
-    always @(dut.core.MTimerInt) void'(rvvi.net_push("MTimerInterrupt",    dut.core.MTimerInt));
+    always @(dut.core.MTimerInt)   void'(rvvi.net_push("MTimerInterrupt",    dut.core.MTimerInt));
-    always @(dut.core.MExtInt)   void'(rvvi.net_push("MExternalInterrupt", dut.core.MExtInt));
+    always @(dut.core.MExtInt)     void'(rvvi.net_push("MExternalInterrupt", dut.core.MExtInt));
-    always @(dut.core.SExtInt)   void'(rvvi.net_push("SExternalInterrupt", dut.core.SExtInt));
+    always @(dut.core.SExtInt)     void'(rvvi.net_push("SExternalInterrupt", dut.core.SExtInt));
-    always @(dut.core.MSwInt)    void'(rvvi.net_push("MSWInterrupt",       dut.core.MSwInt));
+    always @(dut.core.MSwInt)      void'(rvvi.net_push("MSWInterrupt",       dut.core.MSwInt));
+    always @(dut.core.priv.priv.csr.csrs.csrs.STimerInt) void'(rvvi.net_push("STimerInterrupt", dut.core.priv.priv.csr.csrs.csrs.STimerInt));
+
 
     final begin
       void'(rvviRefShutdown());

tests/custom/cacheSimTest/CacheSimTest.py

@@ -0,0 +1,89 @@
+#!/usr/bin/env python3
+
+###########################################
+## CacheSimTest.py
+##
+## Written: lserafini@hmc.edu
+## Created: 4 April 2023
+## Modified: 5 April 2023
+##
+## Purpose: Confirm that the cache simulator behaves as expected. 
+##
+## A component of the CORE-V-WALLY configurable RISC-V project.
+##
+## Copyright (C) 2021-23 Harvey Mudd College & Oklahoma State University
+##
+## SPDX-License-Identifier: Apache-2.0 WITH SHL-2.1
+##
+## Licensed under the Solderpad Hardware License v 2.1 (the “License”); you may not use this file 
+## except in compliance with the License, or, at your option, the Apache License version 2.0. You 
+## may obtain a copy of the License at
+##
+## https:##solderpad.org/licenses/SHL-2.1/
+##
+## Unless required by applicable law or agreed to in writing, any work distributed under the 
+## License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, 
+## either express or implied. See the License for the specific language governing permissions 
+## and limitations under the License.
+################################################################################################
+
+import sys
+import os
+
+sys.path.append(os.path.expanduser("~/cvw/bin"))
+import CacheSim as cs
+
+if __name__ == "__main__":
+    cache = cs.Cache(16, 4, 16, 8)
+    # 0xABCD -> tag: AB, set: C, offset: D
+
+    #address split checking
+    assert (cache.splitaddr(0x1234) == (0x12,0x3,0x4))
+    assert (cache.splitaddr(0x2638) == (0x26,0x3,0x8))
+    assert (cache.splitaddr(0xA3E6) == (0xA3,0xE,0x6))
+    
+    #insert way 0 set C tag AB
+    assert (cache.cacheaccess(0xABCD) == 'M')
+    assert (cache.ways[0][0xC].tag == 0xAB)
+    assert (cache.cacheaccess(0xABCD) == 'H')
+    assert (cache.pLRU[0xC] == [1,1,0])
+
+    #make way 0 set C dirty
+    assert (cache.cacheaccess(0xABCD, True) == 'H')
+
+    #insert way 1 set C tag AC 
+    assert (cache.cacheaccess(0xACCD) == 'M')
+    assert (cache.ways[1][0xC].tag == 0xAC)
+    assert (cache.pLRU[0xC] == [1,0,0])
+
+    #insert way 2 set C tag AD
+    assert (cache.cacheaccess(0xADCD) == 'M')
+    assert (cache.ways[2][0xC].tag == 0xAD)
+    assert (cache.pLRU[0xC] == [0,0,1])
+
+    #insert way 3 set C tag AE 
+    assert (cache.cacheaccess(0xAECD) == 'M')
+    assert (cache.ways[3][0xC].tag == 0xAE)
+    assert (cache.pLRU[0xC] == [0,0,0])
+
+    #misc hit and pLRU checking
+    assert (cache.cacheaccess(0xABCD) == 'H')
+    assert (cache.pLRU[0xC] == [1,1,0])
+    assert (cache.cacheaccess(0xADCD) == 'H')
+    assert (cache.pLRU[0xC] == [0,1,1])
+
+    #evict way 1, now set C has tag AF
+    assert (cache.cacheaccess(0xAFCD) == 'E')
+    assert (cache.ways[1][0xC].tag == 0xAF)
+    assert (cache.pLRU[0xC] == [1,0,1])
+
+    #evict way 3, now set C has tag AC
+    assert (cache.cacheaccess(0xACCD) == 'E')
+    assert (cache.ways[3][0xC].tag == 0xAC)
+    assert (cache.pLRU[0xC] == [0,0,0])
+
+    #evict way 0, now set C has tag EA
+    #this line was dirty, so there was a wb
+    assert (cache.cacheaccess(0xEAC2) == 'D')
+    assert (cache.ways[0][0xC].tag == 0xEA)
+    assert (cache.pLRU[0xC] == [1,1,0])