/////////////////////////////////////////// // ifu.sv // // Written: David_Harris@hmc.edu 9 January 2021 // Modified: // // Purpose: Instrunction Fetch Unit // PC, branch prediction, instruction cache // // 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. //////////////////////////////////////////////////////////////////////////////////////////////// `include "wally-config.vh" module ifu ( input logic clk, reset, input logic StallF, StallD, StallE, StallM, StallW, input logic FlushD, FlushE, FlushM, FlushW, output logic IFUStallF, // IFU stalsl pipeline during a multicycle operation // Command from CPU input logic InvalidateICacheM, // Clears all instruction cache valid bits input logic CSRWriteFenceM, // CSR write or fence instruction, PCNextF = the next valid PC (typically PCE) // Bus interface output logic [`PA_BITS-1:0] IFUHADDR, // Bus address from IFU to EBU input logic [`XLEN-1:0] HRDATA, // Bus read data from IFU to EBU input logic IFUHREADY, // Bus ready from IFU to EBU output logic IFUHWRITE, // Bus write operation from IFU to EBU output logic [2:0] IFUHSIZE, // Bus operation size from IFU to EBU output logic [2:0] IFUHBURST, // Bus burst from IFU to EBU output logic [1:0] IFUHTRANS, // Bus transaction type from IFU to EBU output logic [`XLEN-1:0] PCF, // Fetch stage instruction address // Execute output logic [`XLEN-1:0] PCLinkE, // The address following the branch instruction. (AKA Fall through address) input logic PCSrcE, // Executation stage branch is taken input logic [`XLEN-1:0] IEUAdrE, // The branch/jump target address output logic [`XLEN-1:0] PCE, // Execution stage instruction address output logic BPPredWrongE, // Prediction is wrong // Mem output logic CommittedF, // I$ or bus memory operation started, delay interrupts input logic [`XLEN-1:0] UnalignedPCNextF, // The next PCF, but not aligned to 2 bytes. output logic [`XLEN-1:0] PCNext2F, // Selected PC between branch prediction and next valid PC if CSRWriteFence output logic [31:0] InstrD, // The decoded instruction in Decode stage output logic [31:0] InstrM, // The decoded instruction in Memory stage output logic [`XLEN-1:0] PCM, // Memory stage instruction address // branch predictor output logic [3:0] InstrClassM, // The valid instruction class. 1-hot encoded as jalr, ret, jr (not ret), j, br output logic DirPredictionWrongM, // Prediction direction is wrong output logic BTBPredPCWrongM, // Prediction target wrong output logic RASPredPCWrongM, // RAS prediction is wrong output logic PredictionInstrClassWrongM, // Class prediction is wrong // Faults input logic IllegalBaseInstrFaultD, // Illegal non-compressed instruction output logic InstrPageFaultF, // Instruction page fault output logic IllegalIEUInstrFaultD, // Illegal instruction including compressed output logic InstrMisalignedFaultM, // Branch target not aligned to 4 bytes if no compressed allowed (2 bytes if allowed) // mmu management input logic [1:0] PrivilegeModeW, // Priviledge mode in Writeback stage input logic [`XLEN-1:0] PTE, // Hardware page table walker (HPTW) writes Page table entry (PTE) to ITLB input logic [1:0] PageType, // Hardware page table walker (HPTW) writes PageType to ITLB input logic ITLBWriteF, // Writes PTE and PageType to ITLB input logic [`XLEN-1:0] SATP_REGW, // Location of the root page table and page table configuration input logic STATUS_MXR, // Status CSR: make executable page readable input logic STATUS_SUM, // Status CSR: Supervisor access to user memory input logic STATUS_MPRV, // Status CSR: modify machine privilege input logic [1:0] STATUS_MPP, // Status CSR: previous machine privilege level input logic sfencevmaM, // Virtual memory address fence, invalidate TLB entries output logic ITLBMissF, // ITLB miss causes HPTW (hardware pagetable walker) walk output logic InstrDAPageFaultF, // ITLB hit needs to update dirty or access bits input var logic [7:0] PMPCFG_ARRAY_REGW[`PMP_ENTRIES-1:0], // PMP configuration from privileged unit input var logic [`XLEN-1:0] PMPADDR_ARRAY_REGW[`PMP_ENTRIES-1:0], // PMP address from privileged unit output logic InstrAccessFaultF, // Instruction access fault output logic ICacheAccess, // Report I$ read to performance counters output logic ICacheMiss // Report I$ miss to performance counters ); localparam [31:0] nop = 32'h00000013; // instruction for NOP logic [`XLEN-1:0] PCNextF; // Next PCF, selected from Branch predictor, Privilege, or PC+2/4 logic BranchMisalignedFaultE; // Branch target not aligned to 4 bytes if no compressed allowed (2 bytes if allowed) logic [`XLEN-1:0] PCPlus2or4F; // PCF + 2 (CompressedF) or PCF + 4 (Non-compressed) logic [`XLEN-1:0] PCNextFSpill; // Next PCF after possible + 2 to handle spill logic [`XLEN-1:0] PCFSpill; // PCF with possible + 2 to handle spill logic [`XLEN-1:0] PCLinkD; // PCF2or4F delayed 1 cycle. This is next PC after a control flow instruction (br or j) logic [`XLEN-1:2] PCPlus4F; // PCPlus4F is always PCF + 4. Fancy way to compute PCPlus2or4F logic [`XLEN-1:0] PCD; // Decode stage instruction address logic [`XLEN-1:0] NextValidPCE; // The PC of the next valid instruction in the pipeline after csr write or fence logic [`PA_BITS-1:0] PCPF; // Physical address after address translation logic [`XLEN+1:0] PCFExt; // logic [31:0] IROMInstrF; // Instruction from the IROM logic [31:0] ICacheInstrF; // Instruction from the I$ logic [31:0] InstrRawF; // Instruction from the IROM, I$, or bus logic CompressedF; // The fetched instruction is compressed logic [31:0] PostSpillInstrRawF; // Fetch instruction after merge two halves of spill logic [31:0] InstrRawD; // Non-decompressed instruction in the Decode stage logic [1:0] IFURWF; // IFU alreays read IFURWF = 10 logic [31:0] InstrE; // Instruction in the Execution stage logic [31:0] NextInstrD, NextInstrE; // Instruction into the next stage after possible stage flush logic CacheableF; // PMA indicates instruction address is cacheable logic SelNextSpillF; // In a spill, stall pipeline and gate local stallF logic BusStall; // Bus interface busy with multicycle operation logic ICacheStallF; // I$ busy with multicycle operation logic IFUCacheBusStallD; // EIther I$ or bus busy with multicycle operation logic GatedStallD; // StallD gated by selected next spill // branch predictor signal logic [`XLEN-1:0] PCNext1F; // Branch predictor next PCF logic BusCommittedF; // Bus memory operation in flight, delay interrupts logic CacheCommittedF; // I$ memory operation started, delay interrupts logic SelIROM; // PMA indicates instruction address is in the IROM assign PCFExt = {2'b00, PCFSpill}; ///////////////////////////////////////////////////////////////////////////////////////////// // Spill Support ///////////////////////////////////////////////////////////////////////////////////////////// if(`C_SUPPORTED) begin : Spill spill #(`ICACHE) spill(.clk, .reset, .StallD, .FlushD, .PCF, .PCPlus4F, .PCNextF, .InstrRawF, .InstrDAPageFaultF, .IFUCacheBusStallD, .ITLBMissF, .PCNextFSpill, .PCFSpill, .SelNextSpillF, .PostSpillInstrRawF, .CompressedF); end else begin : NoSpill assign PCNextFSpill = PCNextF; assign PCFSpill = PCF; assign PostSpillInstrRawF = InstrRawF; assign {SelNextSpillF, CompressedF} = 0; end //////////////////////////////////////////////////////////////////////////////////////////////// // Memory management //////////////////////////////////////////////////////////////////////////////////////////////// if(`ZICSR_SUPPORTED == 1) begin : immu /////////////////////////////////////////// // sfence.vma causes TLB flushes /////////////////////////////////////////// // sets ITLBFlush to pulse for one cycle of the sfence.vma instruction // In this instr we want to flush the tlb and then do a pagetable walk to update the itlb and continue the program. // But we're still in the stalled sfence instruction, so if itlbflushf == sfencevmaM, tlbflush would never drop and // the tlbwrite would never take place after the pagetable walk. by adding in ~StallMQ, we are able to drop itlbflush // after a cycle AND pulse it for another cycle on any further back-to-back sfences. logic StallMQ, TLBFlush; flopr #(1) StallMReg(.clk, .reset, .d(StallM), .q(StallMQ)); assign TLBFlush = sfencevmaM & ~StallMQ; mmu #(.TLB_ENTRIES(`ITLB_ENTRIES), .IMMU(1)) immu(.clk, .reset, .SATP_REGW, .STATUS_MXR, .STATUS_SUM, .STATUS_MPRV, .STATUS_MPP, .PrivilegeModeW, .DisableTranslation(1'b0), .VAdr(PCFExt), .Size(2'b10), .PTE(PTE), .PageTypeWriteVal(PageType), .TLBWrite(ITLBWriteF), .TLBFlush, .PhysicalAddress(PCPF), .TLBMiss(ITLBMissF), .Cacheable(CacheableF), .Idempotent(), .SelTIM(SelIROM), .InstrAccessFaultF, .LoadAccessFaultM(), .StoreAmoAccessFaultM(), .InstrPageFaultF, .LoadPageFaultM(), .StoreAmoPageFaultM(), .LoadMisalignedFaultM(), .StoreAmoMisalignedFaultM(), .DAPageFault(InstrDAPageFaultF), .AtomicAccessM(1'b0),.ExecuteAccessF(1'b1), .WriteAccessM(1'b0), .ReadAccessM(1'b0), .PMPCFG_ARRAY_REGW, .PMPADDR_ARRAY_REGW); end else begin assign {ITLBMissF, InstrAccessFaultF, InstrPageFaultF, InstrDAPageFaultF} = '0; assign PCPF = PCFExt[`PA_BITS-1:0]; assign CacheableF = '1; assign SelIROM = '0; end //////////////////////////////////////////////////////////////////////////////////////////////// // Memory //////////////////////////////////////////////////////////////////////////////////////////////// // CommittedM tells the CPU's privileged unit the current instruction // in the memory stage is a memory operaton and that memory operation is either completed // or is partially executed. Partially completed memory operations need to prevent an interrupts. // There is not a clean way to restore back to a partial executed instruction. CommiteedM will // delay the interrupt until the LSU is in a clean state. assign CommittedF = CacheCommittedF | BusCommittedF; logic IgnoreRequest; assign IgnoreRequest = ITLBMissF | FlushD; // The IROM uses untranslated addresses, so it is not compatible with virtual memory. if (`IROM_SUPPORTED) begin : irom logic IROMce; assign IROMce = ~GatedStallD | reset; assign IFURWF = 2'b10; irom irom(.clk, .ce(IROMce), .Adr(PCNextFSpill[`XLEN-1:0]), .IROMInstrF); end else begin assign IFURWF = 2'b10; end if (`BUS) begin : bus // **** must fix words per line vs beats per line as in lsu. localparam integer WORDSPERLINE = `ICACHE ? `ICACHE_LINELENINBITS/`XLEN : 1; localparam integer LOGBWPL = `ICACHE ? $clog2(WORDSPERLINE) : 1; if(`ICACHE) begin : icache localparam integer LINELEN = `ICACHE ? `ICACHE_LINELENINBITS : `XLEN; localparam integer LLENPOVERAHBW = `LLEN / `AHBW; // Number of AHB beats in a LLEN word. AHBW cannot be larger than LLEN. (implementation limitation) logic [LINELEN-1:0] FetchBuffer; logic [`PA_BITS-1:0] ICacheBusAdr; logic ICacheBusAck; logic [1:0] CacheBusRW, BusRW, CacheRWF; assign BusRW = ~ITLBMissF & ~CacheableF & ~SelIROM ? IFURWF : '0; assign CacheRWF = ~ITLBMissF & CacheableF & ~SelIROM ? IFURWF : '0; cache #(.LINELEN(`ICACHE_LINELENINBITS), .NUMLINES(`ICACHE_WAYSIZEINBYTES*8/`ICACHE_LINELENINBITS), .NUMWAYS(`ICACHE_NUMWAYS), .LOGBWPL(LOGBWPL), .WORDLEN(32), .MUXINTERVAL(16), .DCACHE(0)) icache(.clk, .reset, .FlushStage(FlushD), .Stall(GatedStallD), .FetchBuffer, .CacheBusAck(ICacheBusAck), .CacheBusAdr(ICacheBusAdr), .CacheStall(ICacheStallF), .CacheBusRW, .ReadDataWord(ICacheInstrF), .SelHPTW('0), .CacheMiss(ICacheMiss), .CacheAccess(ICacheAccess), .ByteMask('0), .BeatCount('0), .SelBusBeat('0), .CacheWriteData('0), .CacheRW(CacheRWF), .CacheAtomic('0), .FlushCache('0), .NextAdr(PCNextFSpill[11:0]), .PAdr(PCPF), .CacheCommitted(CacheCommittedF), .InvalidateCache(InvalidateICacheM)); ahbcacheinterface #(WORDSPERLINE, LOGBWPL, LINELEN, LLENPOVERAHBW) ahbcacheinterface(.HCLK(clk), .HRESETn(~reset), .HRDATA, .Flush(FlushD), .CacheBusRW, .HSIZE(IFUHSIZE), .HBURST(IFUHBURST), .HTRANS(IFUHTRANS), .HWSTRB(), .Funct3(3'b010), .HADDR(IFUHADDR), .HREADY(IFUHREADY), .HWRITE(IFUHWRITE), .CacheBusAdr(ICacheBusAdr), .BeatCount(), .Cacheable(CacheableF), .SelBusBeat(), .WriteDataM('0), .CacheBusAck(ICacheBusAck), .HWDATA(), .CacheableOrFlushCacheM(1'b0), .CacheReadDataWordM('0), .FetchBuffer, .PAdr(PCPF), .BusRW, .Stall(GatedStallD), .BusStall, .BusCommitted(BusCommittedF)); mux3 #(32) UnCachedDataMux(.d0(ICacheInstrF), .d1(FetchBuffer[32-1:0]), .d2(IROMInstrF), .s({SelIROM, ~CacheableF}), .y(InstrRawF[31:0])); end else begin : passthrough assign IFUHADDR = PCPF; logic [31:0] FetchBuffer; logic [1:0] BusRW; assign BusRW = ~ITLBMissF & ~SelIROM ? IFURWF : '0; assign IFUHSIZE = 3'b010; ahbinterface #(0) ahbinterface(.HCLK(clk), .Flush(FlushD), .HRESETn(~reset), .HREADY(IFUHREADY), .HRDATA(HRDATA), .HTRANS(IFUHTRANS), .HWRITE(IFUHWRITE), .HWDATA(), .HWSTRB(), .BusRW, .ByteMask(), .WriteData('0), .Stall(GatedStallD), .BusStall, .BusCommitted(BusCommittedF), .FetchBuffer(FetchBuffer)); assign CacheCommittedF = '0; if(`IROM_SUPPORTED) mux2 #(32) UnCachedDataMux2(FetchBuffer, IROMInstrF, SelIROM, InstrRawF); else assign InstrRawF = FetchBuffer; assign IFUHBURST = 3'b0; assign {ICacheMiss, ICacheAccess, ICacheStallF} = '0; end end else begin : nobus // block: bus assign {BusStall, CacheCommittedF} = '0; assign {ICacheStallF, ICacheMiss, ICacheAccess} = '0; assign InstrRawF = IROMInstrF; end assign IFUCacheBusStallD = ICacheStallF | BusStall; assign IFUStallF = IFUCacheBusStallD | SelNextSpillF; assign GatedStallD = StallD & ~SelNextSpillF; flopenl #(32) AlignedInstrRawDFlop(clk, reset | FlushD, ~StallD, PostSpillInstrRawF, nop, InstrRawD); //////////////////////////////////////////////////////////////////////////////////////////////// // PCNextF logic //////////////////////////////////////////////////////////////////////////////////////////////// if(`ZICSR_SUPPORTED | `ZIFENCEI_SUPPORTED) mux2 #(`XLEN) pcmux2(.d0(PCNext1F), .d1(NextValidPCE), .s(CSRWriteFenceM),.y(PCNext2F)); else assign PCNext2F = PCNext1F; assign PCNextF = {UnalignedPCNextF[`XLEN-1:1], 1'b0}; // hart-SPEC p. 21 about 16-bit alignment flopenl #(`XLEN) pcreg(clk, reset, ~StallF, PCNextF, `RESET_VECTOR, PCF); // pcadder // add 2 or 4 to the PC, based on whether the instruction is 16 bits or 32 assign PCPlus4F = PCF[`XLEN-1:2] + 1; // add 4 to PC // choose PC+2 or PC+4 based on CompressedF, which arrives later. // Speeds up critical path as compared to selecting adder input based on CompressedF // *** consider gating PCPlus4F to provide the reset. /* -----\/----- EXCLUDED -----\/----- assign PCPlus2or4F[0] = '0; assign PCPlus2or4F[1] = ~reset & (CompressedF ^ PCF[1]); assign PCPlus2or4F[`XLEN-1:2] = reset ? '0 : CompressedF & ~PCF[1] ? PCF[`XLEN-1:2] : PCPlus4F; -----/\----- EXCLUDED -----/\----- */ /* -----\/----- EXCLUDED -----\/----- assign PCPlus2or4F[1:0] = reset ? 2'b00 : CompressedF ? PCF[1] ? 2'b00 : 2'b10 : PCF[1:0]; -----/\----- EXCLUDED -----/\----- */ // *** There is actually a bug in the regression test. We fetched an address which returns data with // an X. This version of the code does not die because if CompressedF is an X it just defaults to the last // option. The above code would work, but propagates the x. always_comb if(reset) PCPlus2or4F = '0; else if (CompressedF) // add 2 if (PCF[1]) PCPlus2or4F = {PCPlus4F, 2'b00}; else PCPlus2or4F = {PCF[`XLEN-1:2], 2'b10}; else PCPlus2or4F = {PCPlus4F, PCF[1:0]}; // add 4 //////////////////////////////////////////////////////////////////////////////////////////////// // Branch and Jump Predictor //////////////////////////////////////////////////////////////////////////////////////////////// if (`BPRED_ENABLED) begin : bpred bpred bpred(.clk, .reset, .StallF, .StallD, .StallE, .StallM, .StallW, .FlushD, .FlushE, .FlushM, .FlushW, .InstrD, .PCNextF, .PCPlus2or4F, .PCNext1F, .PCE, .PCM, .PCSrcE, .IEUAdrE, .PCF, .NextValidPCE, .PCD, .PCLinkE, .InstrClassM, .BPPredWrongE, .PostSpillInstrRawF, .DirPredictionWrongM, .BTBPredPCWrongM, .RASPredPCWrongM, .PredictionInstrClassWrongM); end else begin : bpred mux2 #(`XLEN) pcmux1(.d0(PCPlus2or4F), .d1(IEUAdrE), .s(PCSrcE), .y(PCNext1F)); assign BPPredWrongE = PCSrcE; assign {InstrClassM, DirPredictionWrongM, BTBPredPCWrongM, RASPredPCWrongM, PredictionInstrClassWrongM} = '0; assign NextValidPCE = PCE; end //////////////////////////////////////////////////////////////////////////////////////////////// // Decode stage pipeline register and compressed instruction decoding. //////////////////////////////////////////////////////////////////////////////////////////////// // Decode stage pipeline register and logic flopenrc #(`XLEN) PCDReg(clk, reset, FlushD, ~StallD, PCF, PCD); // expand 16-bit compressed instructions to 32 bits if (`C_SUPPORTED) begin logic IllegalCompInstrD; decompress decomp(.InstrRawD, .InstrD, .IllegalCompInstrD); assign IllegalIEUInstrFaultD = IllegalBaseInstrFaultD | IllegalCompInstrD; // illegal if bad 32 or 16-bit instr end else begin assign InstrD = InstrRawD; assign IllegalIEUInstrFaultD = IllegalBaseInstrFaultD; end // Misaligned PC logic // Instruction address misalignement only from br/jal(r) instructions. // instruction address misalignment is generated by the target of control flow instructions, not // the fetch itself. // xret and Traps both cannot produce instruction misaligned. // xret: mepc is an MXLEN-bit read/write register formatted as shown in Figure 3.21. // The low bit of mepc (mepc[0]) is always zero. On implementations that support // only IALIGN=32, the two low bits (mepc[1:0]) are always zero. // Spec 3.1.14 // Traps: Can’t happen. The bottom two bits of MTVEC are ignored so the trap always is to a multiple of 4. See 3.1.7 of the privileged spec. assign BranchMisalignedFaultE = (IEUAdrE[1] & ~`C_SUPPORTED) & PCSrcE; flopenr #(1) InstrMisalginedReg(clk, reset, ~StallM, BranchMisalignedFaultE, InstrMisalignedFaultM); // Instruction and PC/PCLink pipeline registers // Cannot use flopenrc for Instr(E/M) as it resets to NOP not 0. mux2 #(32) FlushInstrEMux(InstrD, nop, FlushE, NextInstrD); mux2 #(32) FlushInstrMMux(InstrE, nop, FlushM, NextInstrE); flopenr #(32) InstrEReg(clk, reset, ~StallE, NextInstrD, InstrE); flopenr #(32) InstrMReg(clk, reset, ~StallM, NextInstrE, InstrM); flopenr #(`XLEN) PCEReg(clk, reset, ~StallE, PCD, PCE); flopenr #(`XLEN) PCMReg(clk, reset, ~StallM, PCE, PCM); flopenr #(`XLEN) PCPDReg(clk, reset, ~StallD, PCPlus2or4F, PCLinkD); flopenr #(`XLEN) PCPEReg(clk, reset, ~StallE, PCLinkD, PCLinkE); endmodule