Github_Repos/cvw
1
0
Fork 1
mirror of https://github.com/openhwgroup/cvw synced 2025-02-11 06:05:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #1004 from rosethompson/ignorerequest

Browse Source 
Modified the LSU and HPTW to reduce the complexity of the IgnoreRequestTLB logic
This commit is contained in:
David Harris 2024-10-13 21:56:14 -07:00 committed by GitHub
commit 61a5aeaa03
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 182 additions and 40 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/lsu/atomic.sv
View File

@ -39,7 +39,7 @@ module atomic import cvw::*; #(parameter cvw_t P) (
input logic [2:0] LSUFunct3M, // IEU or HPTW memory operation size
input logic [1:0] LSUAtomicM, // 10: AMO operation, select AMOResultM as the writedata output, 01: LR/SC operation
input logic [1:0] PreLSURWM, // IEU or HPTW Read/Write signal
input logic IgnoreRequest, // On FlushM or TLB miss ignore memory operation
input logic LSUFlushW, // On FlushM or TLB miss ignore memory operation
output logic [P.XLEN-1:0] IMAWriteDataM, // IEU, HPTW, or AMO write data
output logic SquashSCW, // Store conditional failed disable write to GPR
output logic [1:0] LSURWM // IEU or HPTW Read/Write signal gated by LR/SC
@ -57,7 +57,7 @@ module atomic import cvw::*; #(parameter cvw_t P) (
// LRSC unit
if (P.ZALRSC_SUPPORTED) begin
assign MemReadM = PreLSURWM[1] & ~IgnoreRequest;
assign MemReadM = PreLSURWM[1] & ~LSUFlushW;
lrsc #(P) lrsc(.clk, .reset, .StallW, .MemReadM, .PreLSURWM, .LSUAtomicM, .PAdrM, .SquashSCW, .LSURWM);
end else begin
assign SquashSCW = 0;

40
src/lsu/lsu.sv
View File

@ -110,8 +110,7 @@ module lsu import cvw::*; #(parameter cvw_t P) (
logic GatedStallW; // Hazard unit StallW gated when SelHPTW = 1
logic BusStall; // Bus interface busy with multicycle operation
logic LSUBusStallM; // Bus interface busy with multicycle operation masked by IgnoreRequestTLB
logic LSUBusStallM; // Bus interface busy with multicycle operation masked by HPTWFlushW
logic HPTWStall; // HPTW busy with multicycle operation
logic DCacheBusStallM; // Cache or bus stall
logic CacheBusHPWTStall; // Cache, bus, or hptw is requesting a stall
@ -146,8 +145,8 @@ module lsu import cvw::*; #(parameter cvw_t P) (
logic DTLBWriteM; // Writes PTE and PageType to DTLB
logic LSULoadAccessFaultM; // Load acces fault
logic LSUStoreAmoAccessFaultM; // Store access fault
logic IgnoreRequestTLB; // On either ITLB or DTLB miss, ignore miss so HPTW can handle
logic IgnoreRequest; // On FlushM or TLB miss ignore memory operation
logic HPTWFlushW; // HPTW needs to flush operation
logic LSUFlushW; // HPTW or hazard unit flushes operation
logic SelDTIM; // Select DTIM rather than bus or D$
logic [P.XLEN-1:0] WriteDataZM;
logic LSULoadPageFaultM, LSUStoreAmoPageFaultM;
@ -200,7 +199,7 @@ module lsu import cvw::*; #(parameter cvw_t P) (
.WriteDataM(WriteDataZM), .Funct3M, .LSUFunct3M, .Funct7M, .LSUFunct7M,
.IEUAdrExtM, .PTE, .IHWriteDataM, .PageType, .PreLSURWM, .LSUAtomicM,
.IHAdrM, .HPTWStall, .SelHPTW,
.IgnoreRequestTLB, .LSULoadAccessFaultM, .LSUStoreAmoAccessFaultM,
.HPTWFlushW, .LSULoadAccessFaultM, .LSUStoreAmoAccessFaultM,
.LoadAccessFaultM, .StoreAmoAccessFaultM, .HPTWInstrAccessFaultF,
.LoadPageFaultM, .StoreAmoPageFaultM, .LSULoadPageFaultM, .LSUStoreAmoPageFaultM, .HPTWInstrPageFaultF
);
@ -215,7 +214,7 @@ module lsu import cvw::*; #(parameter cvw_t P) (
assign StoreAmoAccessFaultM = LSUStoreAmoAccessFaultM;
assign LoadPageFaultM = LSULoadPageFaultM;
assign StoreAmoPageFaultM = LSUStoreAmoPageFaultM;
assign {HPTWStall, SelHPTW, PTE, PageType, DTLBWriteM, ITLBWriteF, IgnoreRequestTLB} = '0;
assign {HPTWStall, SelHPTW, PTE, PageType, DTLBWriteM, ITLBWriteF, HPTWFlushW} = '0;
assign {HPTWInstrAccessFaultF, HPTWInstrPageFaultF} = '0;
end
@ -274,7 +273,7 @@ module lsu import cvw::*; #(parameter cvw_t P) (
// Pause IEU memory request if TLB miss. After TLB fill, replay request.
// Discard memory request on pipeline flush
assign IgnoreRequest = IgnoreRequestTLB | FlushW;
assign LSUFlushW = HPTWFlushW | FlushW;
if (P.DTIM_SUPPORTED) begin : dtim
logic [P.PA_BITS-1:0] DTIMAdr;
@ -285,7 +284,7 @@ module lsu import cvw::*; #(parameter cvw_t P) (
assign DTIMMemRWM = SelDTIM ? LSURWM : 0;
dtim #(P) dtim(.clk, .reset, .ce(~GatedStallW),
.MemRWM(DTIMMemRWM),
.DTIMAdr, .FlushW(IgnoreRequest), .WriteDataM(LSUWriteDataM),
.DTIMAdr, .FlushW(LSUFlushW), .WriteDataM(LSUWriteDataM),
.ReadDataWordM(DTIMReadDataWordM[P.LLEN-1:0]), .ByteMaskM(ByteMaskM));
end else
assign DTIMReadDataWordM = '0;
@ -309,8 +308,6 @@ module lsu import cvw::*; #(parameter cvw_t P) (
logic CacheableOrFlushCacheM; // Memory address is cacheable or operation is a cache flush
logic [1:0] CacheRWM; // Cache read (10), write (01), AMO (11)
logic FlushDCache; // Suppress d cache flush if there is an ITLB miss.
logic CacheStall;
logic [1:0] CacheBusRWTemp;
logic BusCMOZero;
logic [3:0] CacheCMOpM;
logic BusAtomic;
@ -331,29 +328,26 @@ module lsu import cvw::*; #(parameter cvw_t P) (
cache #(.P(P), .PA_BITS(P.PA_BITS), .XLEN(P.XLEN), .LINELEN(P.DCACHE_LINELENINBITS), .NUMSETS(P.DCACHE_WAYSIZEINBYTES*8/LINELEN),
.NUMWAYS(P.DCACHE_NUMWAYS), .LOGBWPL(LLENLOGBWPL), .WORDLEN(CACHEWORDLEN), .MUXINTERVAL(P.LLEN), .READ_ONLY_CACHE(0)) dcache(
.clk, .reset, .Stall(GatedStallW & ~SelSpillE), .SelBusBeat, .FlushStage(IgnoreRequest),
.clk, .reset, .Stall(GatedStallW & ~SelSpillE), .SelBusBeat, .FlushStage(LSUFlushW),
.CacheRW(CacheRWM),
.FlushCache(FlushDCache), .NextSet(IEUAdrExtE[11:0]), .PAdr(PAdrM),
.ByteMask(ByteMaskSpillM), .BeatCount(BeatCount[AHBWLOGBWPL-1:AHBWLOGBWPL-LLENLOGBWPL]),
.WriteData(LSUWriteDataSpillM), .SelHPTW,
.CacheStall, .CacheMiss(DCacheMiss), .CacheAccess(DCacheAccess),
.CacheStall(DCacheStallM), .CacheMiss(DCacheMiss), .CacheAccess(DCacheAccess),
.CacheCommitted(DCacheCommittedM),
.CacheBusAdr(DCacheBusAdr), .ReadDataWord(DCacheReadDataWordM),
.FetchBuffer, .CacheBusRW(CacheBusRWTemp),
.FetchBuffer, .CacheBusRW(CacheBusRW),
.CacheBusAck(DCacheBusAck), .InvalidateCache(1'b0), .CMOpM(CacheCMOpM));
assign DCacheStallM = CacheStall & ~IgnoreRequestTLB;
assign CacheBusRW = CacheBusRWTemp;
ahbcacheinterface #(.P(P), .BEATSPERLINE(BEATSPERLINE), .AHBWLOGBWPL(AHBWLOGBWPL), .LINELEN(LINELEN), .LLENPOVERAHBW(LLENPOVERAHBW), .READ_ONLY_CACHE(0)) ahbcacheinterface(
.HCLK(clk), .HRESETn(~reset), .Flush(IgnoreRequest),
.HCLK(clk), .HRESETn(~reset), .Flush(LSUFlushW),
.HRDATA, .HWDATA(LSUHWDATA), .HWSTRB(LSUHWSTRB),
.HSIZE(LSUHSIZE), .HBURST(LSUHBURST), .HTRANS(LSUHTRANS), .HWRITE(LSUHWRITE), .HREADY(LSUHREADY),
.BeatCount, .SelBusBeat, .CacheReadDataWordM(DCacheReadDataWordM[P.LLEN-1:0]), .WriteDataM(LSUWriteDataM),
.Funct3(LSUFunct3M), .HADDR(LSUHADDR), .CacheBusAdr(DCacheBusAdr), .CacheBusRW, .BusAtomic, .BusCMOZero, .CacheableOrFlushCacheM,
.CacheBusAck(DCacheBusAck), .FetchBuffer, .PAdr(PAdrM),
.Cacheable(CacheableOrFlushCacheM), .BusRW, .Stall(GatedStallW),
.BusStall, .BusCommitted(BusCommittedM));
.BusStall(LSUBusStallM), .BusCommitted(BusCommittedM));
mux3 #(P.LLEN) UnCachedDataMux(.d0(DCacheReadDataWordSpillM), .d1({LLENPOVERAHBW{FetchBuffer[P.XLEN-1:0]}}),
.d2({{P.LLEN-P.XLEN{1'b0}}, DTIMReadDataWordM[P.XLEN-1:0]}),
@ -366,10 +360,10 @@ module lsu import cvw::*; #(parameter cvw_t P) (
assign LSUHADDR = PAdrM;
assign LSUHSIZE = LSUFunct3M;
ahbinterface #(P.XLEN, 1'b1) ahbinterface(.HCLK(clk), .HRESETn(~reset), .Flush(IgnoreRequest), .HREADY(LSUHREADY),
ahbinterface #(P.XLEN, 1'b1) ahbinterface(.HCLK(clk), .HRESETn(~reset), .Flush(LSUFlushW), .HREADY(LSUHREADY),
.HRDATA(HRDATA), .HTRANS(LSUHTRANS), .HWRITE(LSUHWRITE), .HWDATA(LSUHWDATA),
.HWSTRB(LSUHWSTRB), .BusRW, .BusAtomic(AtomicM[1]), .ByteMask(ByteMaskM[P.XLEN/8-1:0]), .WriteData(LSUWriteDataM[P.XLEN-1:0]),
.Stall(GatedStallW), .BusStall, .BusCommitted(BusCommittedM), .FetchBuffer(FetchBuffer));
.Stall(GatedStallW), .BusStall(LSUBusStallM), .BusCommitted(BusCommittedM), .FetchBuffer(FetchBuffer));
// Mux between the 2 sources of read data, 0: Bus, 1: DTIM
if(P.DTIM_SUPPORTED) mux2 #(P.XLEN) ReadDataMux2(FetchBuffer, DTIMReadDataWordM[P.XLEN-1:0], SelDTIM, ReadDataWordMuxM[P.XLEN-1:0]);
@ -381,12 +375,10 @@ module lsu import cvw::*; #(parameter cvw_t P) (
assign {LSUHWDATA, LSUHADDR, LSUHWRITE, LSUHSIZE, LSUHBURST, LSUHTRANS, LSUHWSTRB} = '0;
assign DCacheReadDataWordM = '0;
assign ReadDataWordMuxM = DTIMReadDataWordM;
assign {BusStall, BusCommittedM} = '0;
assign {LSUBusStallM, BusCommittedM} = '0;
assign {DCacheMiss, DCacheAccess} = '0;
assign {DCacheStallM, DCacheCommittedM} = '0;
end
assign LSUBusStallM = BusStall & ~IgnoreRequestTLB;
/////////////////////////////////////////////////////////////////////////////////////////////
// Atomic operations
@ -394,7 +386,7 @@ module lsu import cvw::*; #(parameter cvw_t P) (
if (P.ZAAMO_SUPPORTED | P.ZALRSC_SUPPORTED) begin:atomic
atomic #(P) atomic(.clk, .reset, .StallW, .ReadDataM(ReadDataM[P.XLEN-1:0]), .IHWriteDataM, .PAdrM,
.LSUFunct7M, .LSUFunct3M, .LSUAtomicM, .PreLSURWM, .IgnoreRequest,
.LSUFunct7M, .LSUFunct3M, .LSUAtomicM, .PreLSURWM, .LSUFlushW,
.IMAWriteDataM, .SquashSCW, .LSURWM);
end else begin:lrsc
assign SquashSCW = 1'b0;

34
src/mmu/hptw.sv
View File

@ -58,7 +58,7 @@ module hptw import cvw::*; #(parameter cvw_t P) (
output logic [1:0] LSUAtomicM,
output logic [2:0] LSUFunct3M,
output logic [6:0] LSUFunct7M,
output logic IgnoreRequestTLB,
output logic HPTWFlushW,
output logic SelHPTW,
output logic HPTWStall,
input logic LSULoadAccessFaultM, LSUStoreAmoAccessFaultM,
@ -105,6 +105,7 @@ module hptw import cvw::*; #(parameter cvw_t P) (
logic TakeHPTWFault;
logic PBMTFaultM;
logic HPTWFaultM;
logic ResetPTE;
// map hptw access faults onto either the original LSU load/store fault or instruction access fault
assign LSUAccessFaultM = LSULoadAccessFaultM | LSUStoreAmoAccessFaultM;
@ -143,7 +144,7 @@ module hptw import cvw::*; #(parameter cvw_t P) (
// State flops
flopenr #(1) TLBMissMReg(clk, reset, StartWalk, DTLBMissOrUpdateDAM, DTLBWalk); // when walk begins, record whether it was for DTLB (or record 0 for ITLB)
assign PRegEn = HPTWRW[1] & ~DCacheBusStallM | UpdatePTE;
flopenr #(P.XLEN) PTEReg(clk, reset, PRegEn, NextPTE, PTE); // Capture page table entry from data cache
flopenr #(P.XLEN) PTEReg(clk, ResetPTE, PRegEn, NextPTE, PTE); // Capture page table entry from data cache
// Assign PTE descriptors common across all XLEN values
// For non-leaf PTEs, D, A, U bits are reserved and ignored. They do not cause faults while walking the page table
@ -274,23 +275,26 @@ module hptw import cvw::*; #(parameter cvw_t P) (
IDLE: if (TLBMissOrUpdateDA) NextWalkerState = InitialWalkerState;
else NextWalkerState = IDLE;
L3_ADR: NextWalkerState = L3_RD; // First access in SV48
L3_RD: if (DCacheBusStallM) NextWalkerState = L3_RD;
else if (HPTWFaultM) NextWalkerState = FAULT;
L3_RD: if (HPTWFaultM) NextWalkerState = FAULT;
else if (DCacheBusStallM) NextWalkerState = L3_RD;
else NextWalkerState = L2_ADR;
L2_ADR: if (InitialWalkerState == L2_ADR | ValidNonLeafPTE) NextWalkerState = L2_RD; // First access in SV39
L2_ADR: if (HPTWFaultM) NextWalkerState = FAULT;
else if (InitialWalkerState == L2_ADR | ValidNonLeafPTE) NextWalkerState = L2_RD; // First access in SV39
else NextWalkerState = LEAF;
L2_RD: if (DCacheBusStallM) NextWalkerState = L2_RD;
else if (HPTWFaultM) NextWalkerState = FAULT;
L2_RD: if (HPTWFaultM) NextWalkerState = FAULT;
else if (DCacheBusStallM) NextWalkerState = L2_RD;
else NextWalkerState = L1_ADR;
L1_ADR: if (InitialWalkerState == L1_ADR | ValidNonLeafPTE) NextWalkerState = L1_RD; // First access in SV32
L1_ADR: if (HPTWFaultM) NextWalkerState = FAULT;
else if (InitialWalkerState == L1_ADR | ValidNonLeafPTE) NextWalkerState = L1_RD; // First access in SV32
else NextWalkerState = LEAF;
L1_RD: if (DCacheBusStallM) NextWalkerState = L1_RD;
else if (HPTWFaultM) NextWalkerState = FAULT;
L1_RD: if (HPTWFaultM) NextWalkerState = FAULT;
else if (DCacheBusStallM) NextWalkerState = L1_RD;
else NextWalkerState = L0_ADR;
L0_ADR: if (ValidNonLeafPTE) NextWalkerState = L0_RD;
L0_ADR: if (HPTWFaultM) NextWalkerState = FAULT;
else if (ValidNonLeafPTE) NextWalkerState = L0_RD;
else NextWalkerState = LEAF;
L0_RD: if (DCacheBusStallM) NextWalkerState = L0_RD;
else if (HPTWFaultM) NextWalkerState = FAULT;
L0_RD: if (HPTWFaultM) NextWalkerState = FAULT;
else if (DCacheBusStallM) NextWalkerState = L0_RD;
else NextWalkerState = LEAF;
LEAF: if (P.SVADU_SUPPORTED & HPTWUpdateDA) NextWalkerState = UPDATE_PTE;
else NextWalkerState = IDLE;
@ -300,7 +304,9 @@ module hptw import cvw::*; #(parameter cvw_t P) (
default: NextWalkerState = IDLE; // Should never be reached
endcase // case (WalkerState)
assign IgnoreRequestTLB = (WalkerState == IDLE & TLBMissOrUpdateDA) | (HPTWFaultM); // If hptw request has pmp/a fault suppress bus access.
assign HPTWFlushW = (WalkerState == IDLE & TLBMissOrUpdateDA) | (WalkerState != IDLE & HPTWFaultM);
assign ResetPTE = reset | (NextWalkerState == IDLE);
assign SelHPTW = WalkerState != IDLE;
assign HPTWStall = (WalkerState != IDLE & WalkerState != FAULT) | (WalkerState == IDLE & TLBMissOrUpdateDA);

1
testbench/tests.vh
View File

@ -60,6 +60,7 @@ string coverage64gc[] = '{
"tlbTP",
"tlbMisaligned",
"hptwAccessFault",
"nonleafpbmtfault",
"amoAccessFault",
"floatmisc",
"ifuCamlineWrite",

143
tests/coverage/nonleafpbmtfault.S Normal file
View File

@ -0,0 +1,143 @@
///////////////////////////////////////////
// hptwAccessFault.S
//
// Written: Rose Thompson rose@rosethompson.net
//
// Purpose: Force the HPTW to walk a page table with non-leaf non-zero PBMT bits. This will generate
// a load or store/amo page fault based on the original access type.
//
// A component of the CORE-V-WALLY configurable RISC-V project.
// https://github.com/openhwgroup/cvw
//
// Copyright (C) 2021-24 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.
////////////////////////////////////////////////////////////////////////////////////////////////
// load code to initalize stack, handle interrupts, terminate
#include "WALLY-init-lib.h"
# run-elf.bash find this in project description
main:
# Page table root address at 0x80010000
li t5, 0x9000000000080010
csrw satp, t5
# sfence.vma x0, x0
# switch to supervisor mode
li a0, 1
ecall
li t5, 0
li t2, 0x1000
li t0, 0x8000001000
lw t1, 0(t0) # valid virtual address, valid physical address, but invalid PBMT in middle of page table.
li t1, 0x00008067
add t0, t0, t2
sw t1, 0(t0) # valid virtual address, valid physical address, but invalid PBMT in middle of page table.
fence.I
finished:
j done
.data
.align 16
# Page table situated at 0x80010000
pagetable:
.8byte 0x200044C1
.8byte 0x200044C1
.align 12
.8byte 0x40000040200048C1
.8byte 0x00000000200048C1
.8byte 0x00000000200048C1
.align 12
.8byte 0x0000000020004CC1
.align 12
#80000000
.8byte 0x200000CF
.8byte 0x200004CF
.8byte 0x200008CF
.8byte 0x20000CCF
.8byte 0x200010CF
.8byte 0x200014CF
.8byte 0x200018CF
.8byte 0x20001CCF
.8byte 0x200020CF
.8byte 0x200024CF
.8byte 0x200028CF
.8byte 0x20002CCF
.8byte 0x200030CF
.8byte 0x200034CF
.8byte 0x200038CF
.8byte 0x20003CCF
.8byte 0x200040CF
.8byte 0x200044CF
.8byte 0x200048CF
.8byte 0x20004CCF
.8byte 0x200050CF
.8byte 0x200054CF
.8byte 0x200058CF
.8byte 0x20005CCF
.8byte 0x200060CF
.8byte 0x200064CF
.8byte 0x200068CF
.8byte 0x20006CCF
.8byte 0x200070CF
.8byte 0x200074CF
.8byte 0x200078CF
.8byte 0x20007CCF
.8byte 0x200080CF
.8byte 0x200084CF
.8byte 0x200088CF
.8byte 0x20008CCF
.8byte 0x200090CF
.8byte 0x200094CF
.8byte 0x200098CF
.8byte 0x20009CCF
.8byte 0x2000A0CF
.8byte 0x2000A4CF
.8byte 0x2000A8CF
.8byte 0x2000ACCF
.8byte 0x2000B0CF
.8byte 0x2000B4CF
.8byte 0x2000B8CF
.8byte 0x2000BCCF
.8byte 0x2000C0CF
.8byte 0x2000C4CF
.8byte 0x2000C8CF
.8byte 0x2000CCCF
.8byte 0x2000D0CF
.8byte 0x2000D4CF