input clk,
input Nrst,
+ input flush,
+
/* bus interface */
output reg [31:0] busaddr,
output reg rd_req,
input rw_wait,
output reg [31:0] wr_data,
input [31:0] rd_data,
+ output reg [2:0] data_size,
/* regfile interface */
output reg [3:0] st_read,
output reg cp_req,
input cp_ack,
input cp_busy,
+ output reg cp_rnw, /* 1 = read from CP, 0 = write to CP */
+ input [31:0] cp_read,
+ output reg [31:0] cp_write,
/* stage inputs */
input inbubble,
output reg out_write_reg = 1'b0,
output reg [3:0] out_write_num = 4'bxxxx,
output reg [31:0] out_write_data = 32'hxxxxxxxx,
- output reg [31:0] out_spsr = 32'hxxxxxxxx,
- output reg [31:0] out_cpsr = 32'hxxxxxxxx
+ output reg [31:0] outspsr = 32'hxxxxxxxx,
+ output reg [31:0] outcpsr = 32'hxxxxxxxx
);
reg [31:0] addr, raddr, prev_raddr, next_regdata, next_outcpsr;
+ reg [31:0] prevaddr;
reg [3:0] next_regsel, cur_reg, prev_reg;
- reg next_writeback, next_notdone, next_inc_next;
+ reg next_writeback;
+
+ reg next_outbubble;
+ reg next_write_reg;
+ reg [3:0] next_write_num;
+ reg [31:0] next_write_data;
+
+ reg [2:0] lsr_state = 3'b001, next_lsr_state;
reg [31:0] align_s1, align_s2, align_rddata;
- wire next_outbubble;
- wire next_write_reg;
- wire [3:0] next_write_num;
- wire [31:0] next_write_data;
+ reg [2:0] lsrh_state = 3'b001, next_lsrh_state;
+ reg [31:0] lsrh_rddata;
+ reg [15:0] lsrh_rddata_s1;
+ reg [7:0] lsrh_rddata_s2;
reg [15:0] regs, next_regs;
- reg started = 1'b0, next_started;
+ reg [3:0] lsm_state = 4'b0001, next_lsm_state;
reg [5:0] offset, prev_offset, offset_sel;
- reg notdone = 1'b0;
- reg inc_next = 1'b0;
+ reg [31:0] swp_oldval, next_swp_oldval;
+ reg [1:0] swp_state = 2'b01, next_swp_state;
always @(posedge clk)
begin
out_write_reg <= next_write_reg;
out_write_num <= next_write_num;
out_write_data <= next_write_data;
- notdone <= next_notdone;
- inc_next <= next_inc_next;
regs <= next_regs;
prev_reg <= cur_reg;
- started <= next_started;
- prev_offset <= offset;
+ if (!rw_wait)
+ prev_offset <= offset;
prev_raddr <= raddr;
- out_cpsr <= next_outcpsr;
- out_spsr <= spsr;
+ outcpsr <= next_outcpsr;
+ outspsr <= spsr;
+ swp_state <= next_swp_state;
+ lsm_state <= next_lsm_state;
+ lsr_state <= next_lsr_state;
+ lsrh_state <= next_lsrh_state;
+ prevaddr <= addr;
end
+
+ reg delayedflush = 0;
+ always @(posedge clk)
+ if (flush && outstall /* halp! I can't do it now, maybe later? */)
+ delayedflush <= 1;
+ else if (!outstall /* anything has been handled this time around */)
+ delayedflush <= 0;
always @(*)
begin
- addr = 32'hxxxxxxxx;
+ addr = prevaddr;
raddr = 32'hxxxxxxxx;
rd_req = 1'b0;
wr_req = 1'b0;
wr_data = 32'hxxxxxxxx;
busaddr = 32'hxxxxxxxx;
+ data_size = 3'bxxx;
outstall = 1'b0;
- next_notdone = 1'b0;
next_write_reg = write_reg;
next_write_num = write_num;
next_write_data = write_data;
- next_inc_next = 1'b0;
next_outbubble = inbubble;
- outstall = 1'b0;
- next_regs = 16'b0;
- next_started = started;
+ next_regs = regs;
cp_req = 1'b0;
+ cp_rnw = 1'bx;
+ cp_write = 32'hxxxxxxxx;
offset = prev_offset;
- next_outcpsr = started ? out_cpsr : cpsr;
+ next_outcpsr = lsm_state == 4'b0010 ? outcpsr : cpsr;
+ lsrh_rddata = 32'hxxxxxxxx;
+ lsrh_rddata_s1 = 16'hxxxx;
+ lsrh_rddata_s2 = 8'hxx;
+ next_lsm_state = lsm_state;
+ next_lsr_state = lsr_state;
+ next_lsrh_state = lsrh_state;
+ next_swp_oldval = swp_oldval;
+ next_swp_state = swp_state;
+ cur_reg = prev_reg;
+ /* XXX shit not given about endianness */
casez(insn)
- `DECODE_LDRSTR_UNDEFINED: begin end
- `DECODE_LDRSTR: begin
- if (!inbubble) begin
- next_outbubble = rw_wait;
- outstall = rw_wait | notdone;
-
- addr = insn[23] ? op0 + op1 : op0 - op1; /* up/down select */
- raddr = insn[24] ? op0 : addr; /* pre/post increment */
- busaddr = {raddr[31:2], 2'b0};
+ `DECODE_ALU_SWP: if(!inbubble) begin
+ outstall = rw_wait;
+ next_outbubble = rw_wait;
+ busaddr = {op0[31:2], 2'b0};
+ data_size = insn[22] ? 3'b001 : 3'b100;
+ case(swp_state)
+ 2'b01: begin
+ rd_req = 1'b1;
+ outstall = 1'b1;
+ if(!rw_wait) begin
+ next_swp_state = 2'b10;
+ next_swp_oldval = rd_data;
+ end
+ $display("SWP: read stage");
+ end
+ 2'b10: begin
+ wr_req = 1'b1;
+ wr_data = insn[22] ? {4{op1[7:0]}} : op1;
+ next_write_reg = 1'b1;
+ next_write_num = insn[15:12];
+ next_write_data = insn[22] ? {24'b0, swp_oldval[7:0]} : swp_oldval;
+ if(!rw_wait)
+ next_swp_state = 2'b01;
+ $display("SWP: write stage");
+ end
+ default: begin end
+ endcase
+ end
+ `DECODE_ALU_MULT: begin end
+ `DECODE_ALU_HDATA_REG,
+ `DECODE_ALU_HDATA_IMM: if(!inbubble) begin
+ next_outbubble = rw_wait;
+ outstall = rw_wait;
+ addr = insn[23] ? op0 + op1 : op0 - op1; /* up/down select */
+ raddr = insn[24] ? op0 : addr; /* pre/post increment */
+ busaddr = raddr;
+ /* rotate to correct position */
+ case(insn[6:5])
+ 2'b00: begin end /* swp */
+ 2'b01: begin /* unsigned half */
+ wr_data = {2{op2[15:0]}}; /* XXX need to store halfword */
+ data_size = 3'b010;
+ lsrh_rddata = {16'b0, raddr[1] ? rd_data[31:16] : rd_data[15:0]};
+ end
+ 2'b10: begin /* signed byte */
+ wr_data = {4{op2[7:0]}};
+ data_size = 3'b001;
+ lsrh_rddata_s1 = raddr[1] ? rd_data[31:16] : rd_data[15:0];
+ lsrh_rddata_s2 = raddr[0] ? lsrh_rddata_s1[15:8] : lsrh_rddata_s1[7:0];
+ lsrh_rddata = {{24{lsrh_rddata_s2[7]}}, lsrh_rddata_s2};
+ end
+ 2'b11: begin /* signed half */
+ wr_data = {2{op2[15:0]}};
+ data_size = 3'b010;
+ lsrh_rddata = raddr[1] ? {{16{rd_data[31]}}, rd_data[31:16]} : {{16{rd_data[15]}}, rd_data[15:0]};
+ end
+ endcase
+
+ case(lsrh_state)
+ 3'b001: begin
rd_req = insn[20];
wr_req = ~insn[20];
-
- /* rotate to correct position */
- align_s1 = raddr[1] ? {rd_data[15:0], rd_data[31:16]} : rd_data;
- align_s2 = raddr[0] ? {align_s1[7:0], align_s1[31:8]} : align_s1;
- /* select byte or word */
- align_rddata = insn[22] ? {24'b0, align_s2[7:0]} : align_s2;
-
- if(!insn[20]) begin
- wr_data = insn[22] ? {4{op2[7:0]}} : op2; /* XXX need to actually store just a byte */
- end
- else if(!inc_next) begin
+ next_write_num = insn[15:12];
+ next_write_data = lsrh_rddata;
+ if(insn[20]) begin
next_write_reg = 1'b1;
- next_write_num = insn[15:12];
+ end
+ if(insn[21] | !insn[24]) begin
+ outstall = 1'b1;
+ if(!rw_wait)
+ next_lsrh_state = 3'b010;
+ end
+ $display("ALU_LDRSTRH: rd_req %d, wr_req %d", rd_req, wr_req);
+ end
+ 3'b010: begin
+ next_outbubble = 1'b0;
+ next_write_reg = 1'b1;
+ next_write_num = insn[19:16];
+ next_write_data = addr;
+ next_lsrh_state = 3'b100;
+ end
+ 3'b100: begin
+ outstall = 0;
+ next_lsrh_state = 3'b001;
+ end
+ default: begin end
+ endcase
+
+ if ((lsrh_state == 3'b001) && flush) begin /* Reject it. */
+ outstall = 1'b0;
+ next_lsrh_state = 3'b001;
+ end
+ end
+ `DECODE_LDRSTR_UNDEFINED: begin end
+ `DECODE_LDRSTR: if(!inbubble) begin
+ next_outbubble = rw_wait;
+ outstall = rw_wait;
+ addr = insn[23] ? op0 + op1 : op0 - op1; /* up/down select */
+ raddr = insn[24] ? addr : op0; /* pre/post increment */
+ busaddr = raddr;
+ /* rotate to correct position */
+ align_s1 = raddr[1] ? {rd_data[15:0], rd_data[31:16]} : rd_data;
+ align_s2 = raddr[0] ? {align_s1[7:0], align_s1[31:8]} : align_s1;
+ /* select byte or word */
+ align_rddata = insn[22] ? {24'b0, align_s2[7:0]} : align_s2;
+ wr_data = insn[22] ? {4{op2[7:0]}} : op2; /* XXX need to actually store just a byte */
+ data_size = insn[22] ? 3'b001 : 3'b100;
+ case(lsr_state)
+ 3'b001: begin
+ rd_req = insn[20] /* L */;
+ wr_req = ~insn[20] /* L */;
+ next_write_reg = insn[20] /* L */;
+ next_write_num = insn[15:12];
+ if(insn[20] /* L */) begin
next_write_data = align_rddata;
- next_inc_next = 1'b1;
end
- else if(insn[21]) begin
- next_write_reg = 1'b1;
- next_write_num = insn[19:16];
- next_write_data = addr;
+ if(insn[21] /* W */ | !insn[24] /* P */) begin
+ outstall = 1'b1;
+ if(!rw_wait)
+ next_lsr_state = 3'b010;
end
- next_notdone = rw_wait & insn[20] & insn[21];
+ $display("LDRSTR: rd_req %d, wr_req %d, raddr %08x, wait %d", rd_req, wr_req, raddr, rw_wait);
+ end
+ 3'b010: begin
+ outstall = 1;
+ next_outbubble = 0;
+ next_write_reg = 1'b1;
+ next_write_num = insn[19:16];
+ next_write_data = addr;
+ next_lsr_state = 3'b100;
+ end
+ 3'b100: begin
+ outstall = 0;
+ next_lsr_state = 3'b001;
+ end
+ default: begin end
+ endcase
+
+ if ((lsr_state == 3'b001) && flush) begin /* Reject it. */
+ outstall = 1'b0;
+ next_lsr_state = 3'b001;
end
end
- `DECODE_LDMSTM: begin
- rd_req = insn[20];
- wr_req = ~insn[20];
- if(!started) begin
+ /* XXX ldm/stm incorrect in that stupid case where one of the listed regs is the base reg */
+ `DECODE_LDMSTM: if(!inbubble) begin
+ outstall = rw_wait;
+ next_outbubble = rw_wait;
+ data_size = 3'b100;
+ case(lsm_state)
+ 4'b0001: begin
// next_regs = insn[23] ? op1[15:0] : op1[0:15];
/** verilator can suck my dick */
- next_regs = insn[23] ? op1[15:0] : {op1[0], op1[1], op1[2], op1[3], op1[4], op1[5], op1[6], op1[7],
- op1[8], op1[9], op1[10], op1[11], op1[12], op1[13], op1[14], op1[15]};
+ $display("LDMSTM: Round 1: base register: %08x, reg list %b", op0, op1[15:0]);
+ next_regs = insn[23] /* U */ ? op1[15:0] : {op1[0], op1[1], op1[2], op1[3], op1[4], op1[5], op1[6], op1[7],
+ op1[8], op1[9], op1[10], op1[11], op1[12], op1[13], op1[14], op1[15]};
offset = 6'b0;
- next_started = 1'b1;
- end
- else if(inc_next) begin
- if(insn[21]) begin
- next_write_reg = 1'b1;
- next_write_num = insn[19:16];
- next_write_data = insn[23] ? op0 + {26'b0, prev_offset} : op0 - {26'b0, prev_offset};
- end
- next_started = 1'b0;
+ outstall = 1'b1;
+ next_lsm_state = 4'b0010;
end
- else if(rw_wait) begin
- next_regs = regs;
- cur_reg = prev_reg;
- raddr = prev_raddr;
- end
- else begin
+ 4'b0010: begin
+ rd_req = insn[20];
+ wr_req = ~insn[20];
casez(regs)
16'b???????????????1: begin
cur_reg = 4'h0;
next_regs = 16'b0;
end
endcase
- cur_reg = insn[23] ? 4'hF - cur_reg : cur_reg;
+ cur_reg = insn[23] ? cur_reg : 4'hF - cur_reg;
if(cur_reg == 4'hF && insn[22]) begin
next_outcpsr = spsr;
end
+
offset = prev_offset + 6'h4;
offset_sel = insn[24] ? offset : prev_offset;
raddr = insn[23] ? op0 + {26'b0, offset_sel} : op0 - {26'b0, offset_sel};
-
if(insn[20]) begin
- next_write_reg = 1'b1;
+ next_write_reg = !rw_wait;
next_write_num = cur_reg;
next_write_data = rd_data;
end
+ if (rw_wait) begin
+ next_regs = regs;
+ cur_reg = prev_reg; /* whoops, do this one again */
+ end
st_read = cur_reg;
- wr_data = st_data;
+ wr_data = (cur_reg == 4'hF) ? (pc + 12) : st_data;
+ busaddr = raddr;
+
+ $display("LDMSTM: Stage 2: Writing: regs %b, next_regs %b, reg %d, wr_data %08x, addr %08x", regs, next_regs, cur_reg, wr_data, busaddr);
+
+ outstall = 1'b1;
- next_inc_next = next_regs == 16'b0;
- next_notdone = ~next_inc_next | rw_wait;
- busaddr = {raddr[31:2], 2'b0};
+ if(next_regs == 16'b0) begin
+ next_lsm_state = 4'b0100;
+ end
+ end
+ 4'b0100: begin
+ outstall = 1;
+ next_outbubble = 0;
+ next_write_reg = insn[21] /* writeback */;
+ next_write_num = insn[19:16];
+ next_write_data = insn[23] ? op0 + {26'b0, prev_offset} : op0 - {26'b0, prev_offset};
+ next_lsm_state = 4'b1000;
+ $display("LDMSTM: Stage 3: Writing back");
+ end
+ 4'b1000: begin
+ outstall = 0;
+ next_lsm_state = 4'b0001;
+ end
+ default: $stop;
+ endcase
+ if ((lsm_state == 4'b0001) && flush) begin /* Reject it. */
+ outstall = 1'b0;
+ next_lsm_state = 4'b0001;
+ end
+ $display("LDMSTM: Decoded, bubble %d, insn %08x, lsm state %b -> %b, stall %d", inbubble, insn, lsm_state, next_lsm_state, outstall);
+ end
+ `DECODE_LDCSTC: if(!inbubble) begin
+ $display("WARNING: Unimplemented LDCSTC");
+ end
+ `DECODE_CDP: if(!inbubble) begin
+ cp_req = 1;
+ if (cp_busy) begin
+ outstall = 1;
+ next_outbubble = 1;
+ end
+ if (!cp_ack) begin
+ /* XXX undefined instruction trap */
+ $display("WARNING: Possible CDP undefined instruction");
+ end
+ end
+ `DECODE_MRCMCR: if(!inbubble) begin
+ cp_req = 1;
+ cp_rnw = insn[20] /* L */;
+ if (insn[20] == 0 /* store to coprocessor */)
+ cp_write = op0;
+ else begin
+ if (insn[15:12] != 4'hF /* Fuck you ARM */) begin
+ next_write_reg = 1'b1;
+ next_write_num = insn[15:12];
+ next_write_data = cp_read;
+ end else
+ next_outcpsr = {cp_read[31:28], cpsr[27:0]};
+ end
+ if (cp_busy) begin
+ outstall = 1;
+ next_outbubble = 1;
+ end
+ if (!cp_ack) begin
+ $display("WARNING: Possible MRCMCR undefined instruction: cp_ack %d, cp_busy %d",cp_ack, cp_busy);
end
+ $display("MRCMCR: ack %d, busy %d", cp_ack, cp_busy);
end
default: begin end
endcase
+
+ if ((flush || delayedflush) && !outstall)
+ next_outbubble = 1'b1;
end
endmodule