hw/ip/spi_device/rtl/spi_readcmd.sv

// Copyright lowRISC contributors.
// Licensed under the Apache License, Version 2.0, see LICENSE for details.
// SPDX-License-Identifier: Apache-2.0
//
// SPI Flash Read Commands Processing block
//
// This module processes read commands and Mailbox address region.
//
/*
  Details:

  # Process

  Command Parser sends command at 7th bit. The data valids at the negedge of SCK
  or later. But the data valid in the posedge of 8th SCK.

  Read command process block sees the opcode and transits to Output mode or IO
  mode depends on the opcode. If Fast Dual I/O or Fast Quad I/O, it moves to
  I/O states to receive address in dual or quad lines. In any of those two, the
  protocol assumes M byte and dummy cycle in between.

  When moves to the address states regardless of Output command or I/O command,
  based on 3B/4B config, the address count register is reset to 23 (for 3B) or
  31 (for 4B).

  ## Address handling

  ### Address for Output commands

  As address comes through IO0 only, the state machine shifts the address
  one-by-one and decrement the address counter register by 1.

  When it reaches to the 4B address (addr[2]), it triggers DPSRAM state machine
  to fetch data from DPSRAM. And when it receives addr[0], at posedge the state
  machine moves to appropriate command state.

  ### Address for I/O commands

  Address comes through IO0,1 in dual I/O command, or IO0:3 in quad I/O command.
  The logic latches 2 or 4bit at a time and when it reaches valid 4B address, it
  triggers DPSRAM state machine to fetch data from DPSRAM. It, then, moves to
  M byte state. The value is not used in current design.

  ### Mailbox when the mode is enabled

  If the address falls into mailbox address range while stacking address into a
  register, the state turns on mailbox selection bit. Then all out-going
  requests to DPSRAM for read point to Mailbox section not Read Buffer section.

  ## Normal command

  Normal command sends read data right after the address phase. Right after
  moved to the Normal command state, state machine push proper byte into
  synchronous FIFO. The FIFO has pass parameter enabled to get the data within
  a cycle.

  ## Fast Read command

  Moving into Fast Read command state is identical to the Normal read command
  state. The address state sends read request when it reaches 4B granularity.
  Then when the state moves to this Fast Read command state, the machine waits
  preconfigured dummy cycle. The default dummy cycle is 8. Then it follows same
  step as Normal read command state.

  ## Dual/ Quad Output command

  The state is similar to the Fast Read command. The dummy cycle for Dual/ Quad
  are 4 and 2 by default. But they can be configured by the software. The states
  send DPSRAM read request at every 2nd cycle in a byte for Dual command, 1st
  cycle in a byte (two beats) for Quad command.

  ## Dual/ Quad I/O command

  I/O commands have M byte prior to the dummy cycles. Read command module does
  not support continuous read feature that is described in M byte. So, Dual &
  Quad I/O commands state waits M byte then moves to Dual/ Quad output command
  states.

*/

`include "prim_assert.sv"

module spi_readcmd
  import spi_device_pkg::*;
#(
  // Read command configurations
  // Base address: Index of DPSRAM
  // Buffer size: # of indices assigned to Read Buffer.
  //    This is used to calculate double buffering and threshold.
  // Mailbox Base Addr: Beginning Index of Mailbox in DPSRAM
  // Mailbox Size: Mailbox buffer size (# of indices)
  parameter sram_addr_t  ReadBaseAddr    = spi_device_pkg::SramReadBufferIdx,
  parameter int unsigned ReadBufferDepth = spi_device_pkg::SramMsgDepth,
  parameter sram_addr_t  MailboxBaseAddr = spi_device_pkg::SramMailboxIdx,
  parameter int unsigned MailboxDepth    = spi_device_pkg::SramMailboxDepth
) (
  input clk_i,
  input rst_ni,

  input clk_out_i, // Output clock (inverted SPI_CLK)

  input sys_rst_ni, // System reset for buffer tracking pointers

  // From Command Parser
  // DpReadCmd activates readcmd module, and it sees opcode to determine its
  // mode inside (Normal/ Fast/ Dual/ Quad/ DualIO/ QuadIO)
  input sel_datapath_e sel_dp_i,
  input spi_byte_t     opcode_i,

  // SRAM access
  output logic       sram_req_o,
  output logic       sram_we_o,
  output sram_addr_t sram_addr_o,
  output sram_data_t sram_wdata_o,
  input              sram_rvalid_i,
  input  sram_data_t sram_rdata_i,
  input  sram_err_t  sram_rerror_i,

  // Interface: SPI to Parallel
  input               s2p_valid_i,
  input spi_byte_t    s2p_byte_i,
  input [BitCntW-1:0] s2p_bitcnt_i,

  // Interface: Parallel to SPI
  // Should be latched to clk_out_i
  output logic      p2s_valid_o,
  output spi_byte_t p2s_byte_o,
  input  logic      p2s_sent_i,

  // Configurations:
  //    All configs come from peripheral clock domain.
  //    No CDC exists in between.
  input spi_mode_e spi_mode_i,
  input      [2:0] fastread_dummy_i,
  input      [2:0] dualread_dummy_i, // Dual I/O
  input      [2:0] quadread_dummy_i, // Quad I/O

  // Double buffering
  input [$clog2(ReadBufferDepth)-2:0] readbuf_threshold_i,

  // The command mode is 4B mode. Every read command receives 4B address
  input addr_4b_en_i,

  // Features
  input mailbox_en_i,

  // Mailbox Address base address
  // Only allow compile-time fixed size (1kB by default)
  input [31:0] mailbox_addr_i,
  //input [31:0] mailbox_mask_i,

  // Indicator to take SPI line in Passthrough mode
  output mailbox_assumed_o,

  output io_mode_e io_mode_o,

  // Read watermark event occurs when current address exceeds the threshold cfg
  // value for the first time after hit the current read buffer (1kB granularity).
  output read_watermark_o

);

  logic unused_threshold;
  assign unused_threshold = ^readbuf_threshold_i;
  assign read_watermark_o = 1'b 0;

  logic unused_p2s_sent ;
  assign unused_p2s_sent = p2s_sent_i;

  spi_mode_e unused_spi_mode ; // will be used for passthrough for output enable
  assign unused_spi_mode = spi_mode_i;

  // TODO: Implement
  assign mailbox_assumed_o = 1'b 0;

  sram_err_t unused_sram_rerr;
  assign unused_sram_rerr = sram_rerror_i;

  /////////////////
  // Definitions //
  /////////////////
  typedef enum logic [3:0] {
    // Reset: Wait until the datapath for Read command is activated by cmdparse
    MainReset,

    // Address
    //
    //   In this state, the FSM stacks incoming byte from s2p up to 3B or 4B
    //   based on the config by SW. When FSM moves from Reset to this state,
    //   it sets the IO mode to Single/ Dual/ Quad depending on the incoming
    //   SPI command.
    //
    //   In this state, when it stacks addr[2] bit, which is 3rd from last in
    //   Single IO, 2nd from the last in Dual IO, and the last in Quad IO, it
    //   triggers Fetch State Machine to fetch data from Dual Port SRAM in
    //   SPI_DEVICE.
    //
    //   For I/O commands, the state moves to MByte state.
    //   For Fast commands, the state moves to Dummy state.
    //   For Normal Read command, the state moves to Output state directly.
    MainAddress,

    // MByte
    //
    //   IO commands (Dual I/O, Quad I/O) has M byte following Address. It is
    //   normally used for "continuous read" operation. This logic does not
    //   support the feature, so ignoring it then moves to Dummy state.
    MainMByte,

    // Dummy: Wait until it reaches dummy cycles, then moves to Output state
    MainDummy,

    // Output
    //
    //   After passing the dummy cycle or skipping it for Normal Read command,
    //   The state drives SPI interface to send data. This state always assumes
    //   the data is ready in the sync fifo (see instance below). The Fetch
    //   Machine always prepares the data in advance up to designated level.
    //   The prefetch does not affect the watermark that SW has configured. The
    //   event occurs only when the byte is indeed sent through SPI lines.
    //
    //   This state machine watches the sent address and raises an event if it
    //   exceeds the level. This feature is not turned on for Mailbox address.
    MainOutput,

    // Error: Wait until CSb deasserted
    MainError
  } main_st_e;
  main_st_e main_st, main_st_d;

  /////////////
  // Signals //
  /////////////

  // Address shift & latch
  logic addr_ready_in_word;
  logic addr_latched;
  logic addr_shift_en;
  logic addr_latch_en;
  logic addr_inc; // increase address by 1 word

  logic [31:0] addr_q, addr_d;
  // Indicator of 3B or 4B mode: addr_4b_en_i

  // Dummy counter
  logic dummycnt_eq_zero;
  logic load_dummycnt;
  logic [2:0] dummycnt;

  // IO Mode selection

  // SRAM signals
  // Compare addr_d[SramAw-1:2] and mailbox_addr_i with mailbox_mask_i. If the
  // value falls into mailbox, set this.  Even this is set, it only uses when
  // sram address is sent.
  logic addr_in_mailbox;

  logic [31:0] mailbox_masked_addr;

  // TODO: implement
  // logic [31:0] buffer_masked_addr;

  // Double buffering signals
  logic readbuf_idx; // 0 or 1
  logic readbuf_flip; // if this is asserted, readbuf_idx flips.

  // bit count within a word
  logic bitcnt_update;
  logic bitcnt_dec;
  logic [4:0] bitcnt; // count down from 31 or partial for first unaligned

  // FIFO
  logic unused_fifo_rvalid, fifo_pop;
  sram_data_t fifo_rdata;

  // offset_update: latch addr_d[1:0] into fifo_byteoffset when the statemachine
  // moves to Output stage.
  logic       offset_update;
  logic [1:0] fifo_byteoffset; // the byte position in SRAM word
  logic [7:0] fifo_byte [4]; // converting word sram data into a SPI byte
  logic [7:0] p2s_byte;
  logic       p2s_valid_inclk;

  //////////////
  // Datapath //
  //////////////

  // Address Shifting.
  // the data comes from S2P module, which sends valid signal with spi_byte_t.
  // So, if the logic sees `valid` only, it can't latch the exact word address.
  // This logic latches in byte based until the 2nd last byte. Then merging
  // the incoming byte (premature) to organize 4B address.

  always_ff @(posedge clk_i or negedge rst_ni) begin
    if (!rst_ni) begin
      addr_q <= '0;
    end else if (addr_latch_en) begin
      addr_q <= addr_d;
    end
  end

  always_comb begin
    addr_d = '0; // default value. In 3B mode, upper most byte is 0
    addr_latch_en = 1'b0;

    // TODO: Handle the case of IO command
    if (addr_ready_in_word) begin
      // Return word based address, but should not latch
      addr_d = {addr_q[31:8], s2p_byte_i[5:0], 2'b00};
    end else if (addr_shift_en && s2p_valid_i) begin
      // Latch
      addr_d = {addr_q[23:0], s2p_byte_i[7:0]};
      addr_latch_en = 1'b 1;
    end else if (addr_inc) begin
      // Increase the address to next
      addr_d = {addr_q[31:2] + 1'b1, 2'b00};
      addr_latch_en = 1'b 1;
    end
  end

  // Check bitcnt and assert addr_ready_in_word
  always_comb begin
    addr_ready_in_word = 1'b 0;

    // TODO: handle the QuadIO case. `+6` should be `+8`.
    // meaning the SRAM request only can be sent at the end of address beat
    if (addr_4b_en_i) begin
      /// 8bit for Opcode + 24 bit for upper 3Byte + 6bit for addr[7:2]
      addr_ready_in_word = (s2p_bitcnt_i == BitCntW'(8+24+6)) ? 1'b 1 : 1'b 0;
    end else begin
      // 3B
      addr_ready_in_word = (s2p_bitcnt_i == BitCntW'(8+16+6)) ? 1'b 1 : 1'b 0;
    end
  end

  always_comb begin
    addr_latched = 1'b 0;

    // TODO: handle the QuadIO case.
    if (addr_4b_en_i) begin
      addr_latched = (s2p_bitcnt_i == BitCntW'(8+32)) ? 1'b 1 : 1'b 0;
    end else begin
      addr_latched = (s2p_bitcnt_i == BitCntW'(8+24)) ? 1'b 1 : 1'b 0;
    end
  end

  // Dummy Counter
  always_ff @(posedge clk_i or negedge rst_ni) begin
    if (!rst_ni) begin
      dummycnt <= '0;
    end else if (load_dummycnt) begin
      // load quad_io
      if (opcode_i == CmdReadQuadIO) begin
        dummycnt <= quadread_dummy_i;
      end else if (opcode_i == CmdReadDualIO) begin
        dummycnt <= dualread_dummy_i;
      end else begin
        // Using normal fastread dummy cycle
        dummycnt <= fastread_dummy_i;
      end
    end else if (!dummycnt_eq_zero) begin
      dummycnt <= dummycnt - 1'b 1;
    end
  end
  assign dummycnt_eq_zero = ~|dummycnt;

  // FIFO bit count
  always_ff @(posedge clk_i or negedge rst_ni) begin
    if (!rst_ni) begin
      bitcnt <= '0;
    end else if (bitcnt_update) begin
      unique case (addr_d[1:0])
        2'b 00: bitcnt <= 5'h 1f;
        2'b 01: bitcnt <= 5'h 17;
        2'b 10: bitcnt <= 5'h 0f;
        2'b 11: bitcnt <= 5'h 07;
        default: bitcnt <= 5'h 1f;
      endcase
    end else if (bitcnt_dec) begin
      bitcnt <= bitcnt - 1'b 1;
    end
  end

  //= BEGIN: SRAM Datapath ====================================================
  // Main FSM trigger this datapath. This logic calculates the correct address

  // Address conversion
  // Convert into masked address
  localparam int unsigned MailboxAw = $clog2(MailboxDepth);
  localparam logic [31:0] MailboxMask = {{30-MailboxAw{1'b1}}, {2+MailboxAw{1'b0}}};

  assign mailbox_masked_addr = addr_d & MailboxMask;

  // Only valid when logic sends SRAM request
  assign addr_in_mailbox = (mailbox_masked_addr == mailbox_addr_i);

  // internal addr is the address that the logic tracks.
  // the first address comes from host system and then the internal logic
  // manages the address to follow.

  logic sram_req;
  logic [SramAw-1:0] sram_addr;

  always_comb begin
    sram_addr = '0;
    if (mailbox_en_i && addr_in_mailbox) begin
      sram_addr = MailboxBaseAddr | sram_addr_t'(addr_d[2+:MailboxAw]);
    end else begin
      // Read Buffer Address
      // TODO: Double buffering
      sram_addr = ReadBaseAddr | sram_addr_t'({readbuf_idx, addr_d[2+:$clog2(ReadBufferDepth)]});
    end
  end

  assign sram_addr_o = sram_addr;

  assign sram_req_o = sram_req;
  assign sram_we_o = 1'b 0;    // always read
  assign sram_wdata_o = '0;    // always read

  //- END:   SRAM Datapath ----------------------------------------------------

  //= BEGIN: FIFO to P2S datapath =============================================
  assign fifo_byte = '{fifo_rdata.data[7:0],   fifo_rdata.data[15:8],
                       fifo_rdata.data[23:16], fifo_rdata.data[31:24]};
  // TODO: addr_inc should not affect this until it is accepted.
  assign p2s_byte = fifo_byte[fifo_byteoffset];

  // TODO: fifo_byteoffset
  always_ff @(posedge clk_i or negedge rst_ni) begin
    if (!rst_ni) begin
      fifo_byteoffset <= '0;
    end else if (offset_update) begin
      fifo_byteoffset <= addr_d[1:0];
    end else if (p2s_valid_inclk && bitcnt[2:0] == 0) begin
      // at the MainOutput state, when it sends a byte, increase offset
      fifo_byteoffset <= fifo_byteoffset + 1'b 1;
    end
  end

  // outclk latch
  // can't put async fifo. DC constraint should have half clk datapath
  always_ff @(posedge clk_out_i or negedge rst_ni) begin
    if (!rst_ni) begin
      p2s_valid_o <= 1'b 0;
      p2s_byte_o  <= '0 ;
    end else begin
      p2s_valid_o <= p2s_valid_inclk;
      p2s_byte_o  <= p2s_byte;
    end
  end
  //- END:   FIFO to P2S datapath ---------------------------------------------

  //= BEGIN: Double Buffering =================================================

  // Readbuf Index:
  //
  // this signal is not reset by CSb. The value should be alive throughout the
  // CSb event. (last_access_addr too)

  always_ff @(posedge clk_i or negedge sys_rst_ni) begin
    if (!sys_rst_ni) begin
      readbuf_idx <= 1'b 0;
    end else if (readbuf_flip) begin
      // readbuf_flip happens when the module completes the second to the last
      // byte of a buffer through SPI. There will be a chance that will be
      // cancelled by de-asserting CSb. This logic does not guarantee to cover
      // that corner case. It expects to complete a byte transfer if it sends
      // the first beat of the byte.
      readbuf_idx <= ~readbuf_idx;
    end
  end

  // readbuf_flip
  // TODO: implement. Below is temporary.
  assign readbuf_flip = (main_st == MainOutput && addr_q[9:0] == '1);

  //- END:   Double Buffering -------------------------------------------------

  ///////////////////
  // State Machine //
  ///////////////////

  //= BEGIN: Main state machine ===============================================
  always_ff @(posedge clk_i or negedge rst_ni) begin
    if (!rst_ni) begin
      main_st <= MainReset;
    end else begin
      main_st <= main_st_d;
    end
  end

  always_comb begin
    main_st_d = main_st;

    load_dummycnt = 1'b 0;

    // address control
    addr_inc = 1'b 0;
    sram_req = 1'b 0;
    addr_shift_en = 1'b 0;

    p2s_valid_inclk = 1'b 0;
    fifo_pop        = 1'b 0;

    offset_update = 1'b 0;

    bitcnt_update = 1'b 0;
    bitcnt_dec = 1'b 0;

    io_mode_o = SingleIO;

    unique case (main_st)
      MainReset: begin
        if (sel_dp_i == DpReadCmd) begin
          // Any readcommand goes to MainAddress state to latch address
          // 3B, 4B handles inside address
          main_st_d = MainAddress;
        end
      end

      MainAddress: begin
        addr_shift_en = 1'b 1;

        if (addr_ready_in_word) begin
          // TODO: Send Address request. No need to move the state
        end

        if (addr_latched) begin
          // update bitcnt. If input address is not word aligned, bitcnt
          // could be 23, 15, or 7
          bitcnt_update = 1'b 1;

          // latch addr_d[1:0] to fifo_byteoffset;
          offset_update = 1'b 1;

          // Move to next based on Commands
          unique case (opcode_i) inside
            // Move to Output command
            // Assume FIFO entry is not empty in this case
            CmdReadData: main_st_d = MainOutput;

            // Dummy cycle for commands other than IO
            CmdReadFast, CmdReadDual, CmdReadQuad: begin
              main_st_d = MainDummy;
              // TODO: Reset dummy counter
              load_dummycnt = 1'b 1;
            end

            // MByte for IO commands
            CmdReadDualIO, CmdReadQuadIO: begin
              main_st_d = MainMByte;
            end

            default: begin
              // Error if opcode does not match
              main_st_d = MainError;

              // TODO: Error push?
              // TODO: Define Error handling in Programmer's Guide
            end

          endcase
        end
      end

      MainMByte: begin
        if (s2p_valid_i) begin
          main_st_d = MainDummy;

          load_dummycnt = 1'b 1;
        end
      end

      MainDummy: begin
        if (dummycnt_eq_zero) begin
          main_st_d = MainOutput;
        end
      end

      MainOutput: begin
        bitcnt_dec = 1'b 1;

        // Note: p2s accepts the byte and latch inside at the first beat.
        // So, it is safe to change the data at the next cycle.
        p2s_valid_inclk = 1'b 1;

        // DeadEnd until CSb deasserted
        // Change Mode based on the input opcode
        // Return data from FIFO
        unique case (opcode_i) inside
          CmdReadData, CmdReadFast:   io_mode_o = SingleIO;
          CmdReadDual, CmdReadDualIO: io_mode_o = DualIO;
          CmdReadQuad, CmdReadQuadIO: io_mode_o = QuadIO;
          default: io_mode_o = SingleIO;
        endcase

        // if 2 bits left, increase the address
        // TODO: Dual or Quad output I/O handling
        if (bitcnt == 5'h 2) begin
          addr_inc = 1'b 1;
          sram_req = 1'b 1;
        end

        if (bitcnt == 5'h 0) begin
          // sent all words
          bitcnt_update = 1'b 1;
          // TODO: FIFO pop here?
          fifo_pop = 1'b 1;
        end
      end

      MainError: begin
        main_st_d = MainError;
      end

      default: begin
        main_st_d = MainReset;
      end
    endcase
  end
  //- END:   Main state machine -----------------------------------------------

  ///////////////
  // Instances //
  ///////////////

  // FIFO for read data from DPSRAM
  logic unused_full;
  logic [1:0] unused_depth;
  prim_fifo_sync #(
    .Width             ($bits(sram_data_t)),
    .Pass              (1'b1),
    .Depth             (2),
    .OutputZeroIfEmpty (1'b0)
  ) u_fifo (
    .clk_i,
    .rst_ni,

    .clr_i   ( 1'b0),

    .wvalid_i (sram_rvalid_i),
    .wready_o (), // not used
    .wdata_i  (sram_rdata_i),

    .rvalid_o (unused_fifo_rvalid),
    .rready_i (fifo_pop),
    .rdata_o  (fifo_rdata),

    .full_o   (unused_full),
    .depth_o  (unused_depth)
  );

  // TODO: Handle SRAM integrity errors
  sram_err_t unused_sram_rerror;
  assign unused_sram_rerror = sram_rerror_i;

  // Assertions //
  // FIFO should not overflow. The Main state machine shall send request only
  // when it needs the data within 2 cycles
  `ASSERT(NotOverflow_A, sram_req_o && !sram_we_o |-> !unused_full)

  // SRAM access always read
  `ASSERT(SramReadOnly_A, sram_req_o |-> !sram_we_o)

  // SRAM data should return in next cycle
  `ASSUME(SramDataReturnRequirement_M, sram_req_o && !sram_we_o |=> sram_rvalid_i)

  // TODO: When main state machine returns data to SPI (via p2s), the FIFO shall
  // not be empty
  `ASSERT(NotEmpty_A, p2s_valid_inclk |-> (unused_depth != 0))

  // There's chance to addr_latched and addr_Ready_in_word happens at the same cycle.
  // That should be QuadIO command only.
  `ASSUME(QuadIOLatchAndWordReady_M,
          addr_latched && addr_ready_in_word |-> opcode_i == CmdReadQuadIO)

  // `addr_inc` should not be asserted in Address phase
  `ASSERT(AddrIncNotAssertInAddressState_A, addr_inc |-> main_st != MainAddress)

  // Assume mailbox addr config is mailbox size.
  // As depth is # of entries and the entry is word, the size should add 2bits
  `ASSUME(MailboxSizeMatch_M, mailbox_addr_i[$clog2(MailboxDepth)+1:0] == '0)


endmodule : spi_readcmd