Merge branch 'new_ramloader' into nocache_plus_newramloader, plus manual merge into udp version.

Raw ethernet, ISE 10 -- Passes timing, works
UDP, ISE 10 -- barely fails timing, works
ISE 12 -- both  fail timing, not tested yet.
* new_ramloader:
  experimental mods to make ram loader fully synchronous.  Based on IJB's work

2 files changed, 252 insertions, 220 deletions

diff --git a/usrp2/control_lib/ram_harvard.v b/usrp2/control_lib/ram_harvard.v
index 3c00f87c7..948f9b36f 100644
--- a/usrp2/control_lib/ram_harvard.v
+++ b/usrp2/control_lib/ram_harvard.v
@@ -13,10 +13,9 @@ module ram_harvard
      // Firmware download port.
      input [AWIDTH-1:0] ram_loader_adr_i,
      input [31:0] ram_loader_dat_i,
-     input ram_loader_stb_i,
      input [3:0] ram_loader_sel_i,
+     input ram_loader_stb_i,
      input ram_loader_we_i,
-     output ram_loader_ack_o,
      input ram_loader_done_i,    
      // Instruction fetch port.
      input [AWIDTH-1:0] if_adr,
@@ -35,9 +34,6 @@ module ram_harvard
    reg 	   ack_d1;
    reg 	   stb_d1;
    
-   
-   assign 	 ram_loader_ack_o = ram_loader_stb_i;
-
    dpram32 #(.AWIDTH(AWIDTH),.RAM_SIZE(RAM_SIZE)) 
    sys_ram
      (.clk(wb_clk_i),
@@ -46,7 +42,8 @@ module ram_harvard
       .dat1_o(if_data),
       .we1_i(ram_loader_done_i ? 1'b0 : ram_loader_we_i),
       .en1_i(ram_loader_done_i ? 1'b1 : ram_loader_stb_i),
-      .sel1_i(ram_loader_done_i ? 4'hF : ram_loader_sel_i),
+      //.sel1_i(ram_loader_done_i ? 4'hF : ram_loader_sel_i),
+      .sel1_i(ram_loader_sel_i), // Sel is only for writes anyway
       .adr2_i(dwb_adr_i),
       .dat2_i(dwb_dat_i),
       .dat2_o(dwb_dat_o),
@@ -68,6 +65,5 @@ module ram_harvard
        stb_d1 <= 0;
      else
        stb_d1 <= dwb_stb_i;
-   
 
-endmodule // ram_harv_cache
+endmodule // ram_harvard
diff --git a/usrp2/control_lib/ram_loader.v b/usrp2/control_lib/ram_loader.v
index cb67de739..c53ea7aa7 100644
--- a/usrp2/control_lib/ram_loader.v
+++ b/usrp2/control_lib/ram_loader.v
@@ -1,225 +1,261 @@
+module ram_loader
+  #(parameter AWIDTH=16, RAM_SIZE=16384)
+    (
+     // Wishbone I/F and clock domain
+     input wb_clk,
+     input dsp_clk,
+     input ram_loader_rst,
+     output wire [31:0] wb_dat,
+     output wire [AWIDTH-1:0] wb_adr,
+     output wb_stb,
+     output reg [3:0] wb_sel,
+     output wb_we,
+     output reg ram_loader_done,
+     // CPLD signals and clock domain
+     input cpld_clk,
+     input cpld_din,
+     output reg cpld_start,
+     output reg cpld_mode,
+     output reg cpld_done,
+     input cpld_detached
+     );
 
-// Adapted from VHDL code in spi_boot by Arnim Legauer
-//  Added a full wishbone master interface (32-bit)
-
-module ram_loader #(parameter AWIDTH=16, RAM_SIZE=16384)
-  (input clk_i, input rst_i,
-   // CPLD Interface
-   input cfg_clk_i, input cfg_data_i,
-   output start_o, output mode_o, output done_o,
-   input detached_i,
-   // Wishbone interface
-   output wire [31:0] wb_dat_o,
-   output reg [AWIDTH-1:0] wb_adr_o,
-   output wb_stb_o,
-   output wb_cyc_o,
-   output reg [3:0] wb_sel_o,
-   output reg wb_we_o,
-   input wb_ack_i,
-   output ram_loader_done_o);
+   localparam S0 = 0;
+   localparam S1 = 1;
+   localparam S2 = 2;
+   localparam S3 = 3;
+   localparam S4 = 4;
+   localparam S5 = 5;
+   localparam S6 = 6;
+   localparam RESET = 7;
    
-   //  FSM to control start signal, clocked on main clock
-   localparam FSM1_WAIT_DETACH = 2'b00;
-   localparam FSM1_CHECK_NO_DONE = 2'b01;
-   localparam FSM1_WAIT_DONE = 2'b10;
-   
-   reg [1:0]  start_fsm_q, start_fsm_s;
-   reg 	      start_q, enable_q, start_s, enable_s;
-   reg 	      done_q, done_s;
+   localparam WB_IDLE = 0;
+   localparam WB_WRITE = 1;
+  
+
+   reg [AWIDTH+2:0] count;       // 3 LSB's count bits in, the MSB's generate the Wishbone address
+   reg [6:0] 	    shift_reg;
+   reg [7:0] 	    data_reg;
+   reg 		    sampled_clk;
+   reg 		    sampled_clk_meta;
+   reg 		    sampled_din;
+   reg 		    inc_count;
+   reg 		    load_data_reg;
+   reg 		    shift;  
+   reg 		    wb_state, wb_next_state;
+   reg [2:0] 	    state, next_state;
+    
+   //
+   // CPLD clock doesn't free run and is approximately 12.5MHz.
+   // Use 50MHz Wishbone clock to sample it as a signal and avoid having
+   // an extra clock domain for no reason.
+   //
+
+   always @(posedge dsp_clk or posedge ram_loader_rst)
+      if (ram_loader_rst)
+	begin
+	   sampled_clk_meta <= 1'b0;
+	   sampled_clk <= 1'b0;
+	   sampled_din <= 1'b0;
+	   count <= 'h7FFF8;  // Initialize so that address will be 0 when first byte fully received.
+	   data_reg <= 0;
+	   shift_reg <= 0;
+	end
+      else 
+	begin
+	   sampled_clk_meta <= cpld_clk;
+	   sampled_clk <= sampled_clk_meta;
+	   sampled_din <= cpld_din;
+	   if (inc_count)
+	     count <= count + 1'b1;
+	   if (load_data_reg)
+	     data_reg <= {shift_reg,sampled_din};
+	   if (shift)
+	     shift_reg <= {shift_reg[5:0],sampled_din};	   
+	end // else: !if(ram_loader_rst)
    
-   always @(posedge clk_i or posedge rst_i)
-     if(rst_i)
-       begin
-	  start_fsm_q <= FSM1_WAIT_DETACH;
-	  start_q <= 1'b0;
-	  enable_q <= 1'b0;
-       end
+	   
+   always @(posedge dsp_clk or posedge ram_loader_rst)
+     if (ram_loader_rst)
+       state <= RESET;
      else
-       begin
-	  start_fsm_q <= start_fsm_s;
-	  enable_q <= enable_s;
-	  start_q <= start_s;
-       end // else: !if(rst_i)
-   
+       state <= next_state;
+
+
    always @*
-     case(start_fsm_q)
-       FSM1_WAIT_DETACH:
-	 if(detached_i == 1'b1)
-	   begin
-	      start_fsm_s <= FSM1_CHECK_NO_DONE;
-	      enable_s <= 1'b1;
-	      start_s <= 1'b1;
-	   end
-	 else
-	   begin
-	      start_fsm_s <= FSM1_WAIT_DETACH;
-	      enable_s <= enable_q;
-	      start_s <= start_q;
-	   end // else: !if(detached_i == 1'b1)
-       FSM1_CHECK_NO_DONE:
-	 if(~done_q)
-	   begin
-	      start_fsm_s  <= FSM1_WAIT_DONE;
-	      enable_s <= enable_q;
-	      start_s <= start_q;
-	   end
-	 else
-	   begin
-	      start_fsm_s  <= FSM1_CHECK_NO_DONE;
-	      enable_s <= enable_q;
-	      start_s <= start_q;
-	   end // else: !if(~done_q)
-       FSM1_WAIT_DONE:
-	 if(done_q)
-	   begin
-	      start_fsm_s  <= FSM1_WAIT_DETACH;
-	      enable_s <= 1'b0;
-	      start_s <= 1'b0;
-	   end
-	 else
-	   begin
-	      start_fsm_s  <= FSM1_WAIT_DONE;
-	      enable_s <= enable_q;
-	      start_s <= start_q;
-	   end // else: !if(done_q)
-       default:
-	 begin
-	    start_fsm_s  <= FSM1_WAIT_DETACH;
-	    enable_s <= enable_q;
-	    start_s <= start_q;
-	 end // else: !if(done_q)
-     endcase // case(start_fsm_q)
-   
-   //  FSM running on data clock
-
-   localparam FSM2_IDLE = 3'b000;
-   localparam FSM2_WE_ON = 3'b001;
-   localparam FSM2_WE_OFF = 3'b010;
-   localparam FSM2_INC_ADDR1 = 3'b011;
-   localparam FSM2_INC_ADDR2 = 3'b100;
-   localparam FSM2_FINISHED = 3'b101;
-   
-   reg [AWIDTH-1:0] addr_q;
-   reg [7:0] 	    shift_dat_q, ser_dat_q;
-   reg [2:0] 	    bit_q, fsm_q, fsm_s;
-   reg 		    bit_ovfl_q, ram_we_s, ram_we_q, mode_q, mode_s, inc_addr_s;
-   
-   always @(posedge cfg_clk_i or posedge rst_i)
-     if(rst_i)
-       begin
-	  addr_q <= 0;
-	  shift_dat_q <= 8'd0;
-	  ser_dat_q <= 8'd0;
-	  bit_q <= 3'd0;
-	  bit_ovfl_q <= 1'b0;
-	  fsm_q <= FSM2_IDLE;
-	  ram_we_q <= 1'b0;
-	  done_q <= 1'b0;
-	  mode_q <= 1'b0;
-       end
+     begin
+	// Defaults
+	next_state = state;
+	cpld_start = 1'b0;
+	shift = 1'b0;
+	inc_count = 0;
+	load_data_reg = 1'b0;
+	ram_loader_done = 1'b0;
+	cpld_mode = 1'b0;
+	cpld_done = 1'b1;
+	
+	
+	
+	case (state) //synthesis parallel_case full_case
+	  // After reset wait until CPLD indicates its detached.
+	  RESET: begin		     
+	     if (cpld_detached)
+	       next_state = S0;
+	     else
+	       next_state = RESET;
+	  end
+
+	  // Assert cpld_start to signal the CPLD its to start sending serial clock and data.
+	  // Assume cpld_clk is low as we transition into search for first rising edge
+	  S0: begin
+	     cpld_start = 1'b1;	 
+	     cpld_done = 1'b0;	 
+	     if (~cpld_detached)
+	       next_state = S2;
+	     else
+	       next_state = S0;   
+	  end
+	  
+	  //
+	  S1: begin
+	     cpld_start = 1'b1;	   
+	     cpld_done = 1'b0;	 
+	     if (sampled_clk)
+	       begin
+		  // Found rising edge on cpld_clk.
+		  if (count[2:0] == 3'b111)
+		    // Its the last bit of a byte, send it out to the Wishbone bus.
+		    begin
+		       load_data_reg = 1'b1;
+		       inc_count = 1'b1;
+		    end
+		  else 
+	          // Shift databit into LSB of shift register and increment count
+		     begin
+		       shift = 1'b1;
+		       inc_count = 1'b1;
+		     end // else: !if(count[2:0] == 3'b111)
+		  next_state = S2;
+	       end // if (sampled_clk)
+	     else
+	       next_state = S1;
+	  end // case: S1
+	  
+	  //
+	  S2: begin
+	     cpld_start = 1'b1;	    
+	     cpld_done = 1'b0;
+	     if (~sampled_clk)
+	       // Found negative edge of clock
+	       if (count[AWIDTH+2:3] == RAM_SIZE-1) // NOTE need to change this constant
+		 // All firmware now downloaded
+		 next_state = S3;
+	       else
+		 next_state = S1;
+	     else
+	       next_state = S2;
+	  end // case: S2
+	  
+	  // Now that terminal count is reached and all firmware is downloaded signal CPLD that download is done 
+	  // and that mode is now SPI mode.
+	  S3: begin
+	     if (sampled_clk)
+	       begin
+		  cpld_mode = 1'b1;
+		  cpld_done = 1'b1;
+		  next_state = S4;
+	       end
+	     else
+	       next_state = S3;	     
+	  end
+
+	  // Search for negedge of cpld_clk whilst keeping done sequence asserted.
+	  // Keep done assserted 
+	  S4: begin
+	     cpld_mode = 1'b1;
+	     cpld_done = 1'b1;
+	     if (~sampled_clk)
+	       next_state = S5;
+	     else
+	       next_state = S4;
+	  end
+
+	  // Search for posedge of cpld_clk whilst keeping done sequence asserted.
+	  S5: begin
+	     cpld_mode = 1'b1;
+	     cpld_done = 1'b1;
+	     if (sampled_clk)
+	       next_state = S6;
+	     else	      
+	       next_state = S5;	       
+	  end
+
+	  // Stay in this state until reset/power down
+	  S6: begin
+	     ram_loader_done = 1'b1;
+	     cpld_done = 1'b1;
+	     cpld_mode = 1'b1;
+	     next_state = S6;
+	  end
+
+	endcase // case(state)
+     end
+
+   always @(posedge dsp_clk or posedge ram_loader_rst)
+     if (ram_loader_rst)
+       wb_state <= WB_IDLE;
      else
-       begin
-	  if(inc_addr_s)
-	    addr_q <= addr_q + 1;
-	  if(enable_q)
-	    begin
-	       bit_q <= bit_q + 1;
-	       bit_ovfl_q <= (bit_q == 3'd7);
-	       shift_dat_q[0] <= cfg_data_i;
-	       shift_dat_q[7:1] <= shift_dat_q[6:0];
-	    end
-	  if(bit_ovfl_q)
-	    ser_dat_q <= shift_dat_q;
-
-	  fsm_q <= fsm_s;
-
-	  ram_we_q <= ram_we_s;
-
-	  if(done_s)
-	    done_q <= 1'b1;
-	  mode_q <= mode_s;
-       end // else: !if(rst_i)
+       wb_state <= wb_next_state;
 
+   reg do_write;
+   wire empty, full;
+   
    always @*
      begin
-	inc_addr_s <= 1'b0;
-	ram_we_s <= 1'b0;
-	done_s <= 1'b0;
-	fsm_s <= FSM2_IDLE;
-	mode_s <= 1'b0;
-
-	case(fsm_q)
-	  FSM2_IDLE :
-	    if(start_q)
-	      if(bit_ovfl_q)
-		fsm_s <= FSM2_WE_ON;
-	  FSM2_WE_ON:
-	    begin
-	       ram_we_s <= 1'b1;
-	       fsm_s <= FSM2_WE_OFF;
-	    end
-	  FSM2_WE_OFF:
-	    begin
-	       ram_we_s <= 1'b1;
-	       fsm_s <= FSM2_INC_ADDR1;
-	    end
-	  FSM2_INC_ADDR1:
-	    fsm_s <= FSM2_INC_ADDR2;
-	  FSM2_INC_ADDR2:
-	    if(addr_q == (RAM_SIZE-1))
-	    //if(&addr_q)
-	      begin
-		 fsm_s <= FSM2_FINISHED;
-		 done_s <= 1'b1;
-		 mode_s <= 1'b1;
-	      end
-	    else
-	      begin
-		 inc_addr_s <= 1'b1;
-		 fsm_s <= FSM2_IDLE;
-	      end // else: !if(&addr_q)
-	  FSM2_FINISHED:
-	    begin
-	       fsm_s <= FSM2_FINISHED;
-	       mode_s <= 1'b1;
-	    end
-	endcase // case(fsm_q)
+	wb_next_state = wb_state;
+	do_write = 1'b0;
+	
+	case (wb_state) //synthesis full_case parallel_case
+	  //
+	  WB_IDLE: begin
+	     if (load_data_reg)
+	       // Data reg will load ready to write wishbone @ next clock edge
+	       wb_next_state  =  WB_WRITE;
+	     else
+	       wb_next_state = WB_IDLE;
+	  end
+
+	  // Drive address and data onto wishbone.
+	  WB_WRITE: begin
+      	     do_write = 1'b1;
+	     if (~full)	       
+	       wb_next_state =  WB_IDLE;		      
+	     else
+	       wb_next_state = WB_WRITE;	       
+	  end
+
+	endcase // case(wb_state)
      end // always @ *
    
-   assign start_o = start_q;
-   assign mode_o = mode_q;
-   assign done_o = start_q ? done_q : 1'b1;
-   wire [AWIDTH-1:0] ram_addr = addr_q;
-   wire [7:0] ram_data = ser_dat_q;
-   assign ram_loader_done_o = (fsm_q == FSM2_FINISHED);
-   
-   // wishbone master, only writes
-   reg [7:0] dat_holder;
-   assign    wb_dat_o = {4{dat_holder}};
-   assign    wb_stb_o = wb_we_o;
-   assign    wb_cyc_o = wb_we_o;
+   wire [1:0] count_out;
+   wire [7:0] data_out;
+
+   fifo_xlnx_16x40_2clk crossclk
+     (.rst(ram_loader_rst),
+      .wr_clk(dsp_clk), .din({count[4:3],count[AWIDTH+2:3],data_reg}), .wr_en(do_write), .full(full),
+      .rd_clk(wb_clk), .dout({count_out,wb_adr,data_out}), .rd_en(~empty), .empty(empty));
+
+   assign wb_dat = {4{data_out}};
+
+   always @*
+     case(count_out[1:0]) //synthesis parallel_case full_case
+       2'b00 : wb_sel = 4'b1000;
+       2'b01 : wb_sel = 4'b0100;
+       2'b10 : wb_sel = 4'b0010;
+       2'b11 : wb_sel = 4'b0001;
+     endcase
+
+   assign wb_we = ~empty;
+   assign wb_stb = ~empty;
    
-   always @(posedge clk_i or posedge rst_i)
-     if(rst_i)
-       begin
-	  dat_holder <= 8'd0;
-	  wb_adr_o <= 0;
-	  wb_sel_o <= 4'b0000;
-	  wb_we_o <= 1'b0;
-       end
-     else if(ram_we_q)
-       begin
-	  dat_holder <= ram_data;
-	  wb_adr_o <= ram_addr;
-	  wb_we_o <= 1'b1;
-	  case(ram_addr[1:0])   // Big Endian
-	    2'b00 : wb_sel_o <= 4'b1000;
-	    2'b01 : wb_sel_o <= 4'b0100;
-	    2'b10 : wb_sel_o <= 4'b0010;
-	    2'b11 : wb_sel_o <= 4'b0001;
-	  endcase // case(ram_addr[1:0])
-       end // if (ram_we_q)
-     else if(wb_ack_i)
-       wb_we_o <= 1'b0;
-      
 endmodule // ram_loader