Pass mtspr/mfspr to MMU-related SPRs down to loadstore1

This arranges for some mfspr and mtspr to get sent to loadstore1
instead of being handled in execute1.  In particular, DAR and DSISR
are handled this way.  They are therefore "slow" SPRs.

While we're at it, fix the spelling of HEIR and remove mention of
DAR and DSISR from the comments in execute1.

Signed-off-by: Paul Mackerras <paulus@ozlabs.org>

common.vhdl

@@ -24,6 +24,8 @@ package common is
     constant SPR_XER    : spr_num_t := 1;
     constant SPR_LR     : spr_num_t := 8;
     constant SPR_CTR    : spr_num_t := 9;
+    constant SPR_DSISR  : spr_num_t := 18;
+    constant SPR_DAR    : spr_num_t := 19;
     constant SPR_TB     : spr_num_t := 268;
     constant SPR_DEC    : spr_num_t := 22;
     constant SPR_SRR0   : spr_num_t := 26;
@@ -214,7 +216,7 @@ package common is
 
     type Execute1ToLoadstore1Type is record
 	valid : std_ulogic;
-        op : insn_type_t;                               -- what ld/st op to do
+        op : insn_type_t;                               -- what ld/st or m[tf]spr to do
 	addr1 : std_ulogic_vector(63 downto 0);
 	addr2 : std_ulogic_vector(63 downto 0);
 	data : std_ulogic_vector(63 downto 0);		-- data to write, unused for read
@@ -228,10 +230,12 @@ package common is
 	xerc : xer_common_t;
         reserve : std_ulogic;                           -- set for larx/stcx.
         rc : std_ulogic;                                -- set for stcx.
+        spr_num : spr_num_t;                            -- SPR number for mfspr/mtspr
     end record;
     constant Execute1ToLoadstore1Init : Execute1ToLoadstore1Type := (valid => '0', op => OP_ILLEGAL, ci => '0', byte_reverse => '0',
                                                                      sign_extend => '0', update => '0', xerc => xerc_init,
-                                                                     reserve => '0', rc => '0', others => (others => '0'));
+                                                                     reserve => '0', rc => '0',
+                                                                     spr_num => 0, others => (others => '0'));
 
     type Loadstore1ToDcacheType is record
 	valid : std_ulogic;

decode1.vhdl

@@ -363,6 +363,7 @@ begin
 		variable v : Decode1ToDecode2Type;
                 variable majorop : major_opcode_t;
                 variable op_19_bits: std_ulogic_vector(2 downto 0);
+                variable sprn : spr_num_t;
 	begin
 		v := r;
 
@@ -429,10 +430,17 @@ begin
 			end if;
 		    end if;
 		elsif v.decode.insn_type = OP_MFSPR or v.decode.insn_type = OP_MTSPR then
-		    v.ispr1 := fast_spr_num(decode_spr_num(f_in.insn));
+                    sprn := decode_spr_num(f_in.insn);
+		    v.ispr1 := fast_spr_num(sprn);
 		    -- Make slow SPRs single issue
 		    if is_fast_spr(v.ispr1) = '0' then
 			v.decode.sgl_pipe := '1';
+                        -- send MMU-related SPRs to loadstore1
+                        case sprn is
+                        when SPR_DAR | SPR_DSISR =>
+                            v.decode.unit := LDST;
+                        when others =>
+                        end case;
 		    end if;
 		elsif v.decode.insn_type = OP_RFID then
 		    report "PPC RFID";

execute1.vhdl

@@ -463,7 +463,7 @@ begin
 
 	elsif irq_valid = '1' and e_in.valid = '1' then
 	    -- we need two cycles to write srr0 and 1
-	    -- will need more when we have to write DSISR, DAR and HIER
+	    -- will need more when we have to write HEIR
             -- Don't deliver the interrupt until we have a valid instruction
             -- coming in, so we have a valid NIA to put in SRR0.
 	    exception := '1';
@@ -494,13 +494,12 @@ begin
 
 	    when OP_ILLEGAL =>
 		-- we need two cycles to write srr0 and 1
-		-- will need more when we have to write DSISR, DAR and HIER
+		-- will need more when we have to write HEIR
 		illegal := '1';
 	    when OP_SC =>
 		-- check bit 1 of the instruction is 1 so we know this is sc;
                 -- 0 would mean scv, so generate an illegal instruction interrupt
 		-- we need two cycles to write srr0 and 1
-		-- will need more when we have to write DSISR, DAR and HIER
                 if e_in.insn(1) = '1' then
                     exception := '1';
                     exception_nextpc := '1';
@@ -983,6 +982,7 @@ begin
         lv.xerc := v.e.xerc;
         lv.reserve := e_in.reserve;
         lv.rc := e_in.rc;
+        lv.spr_num := decode_spr_num(e_in.insn);
         -- decode l*cix and st*cix instructions here
         if e_in.insn(31 downto 26) = "011111" and e_in.insn(10 downto 9) = "11" and
             e_in.insn(5 downto 1) = "10101" then

loadstore1.vhdl

@@ -59,6 +59,8 @@ architecture behave of loadstore1 is
         nc           : std_ulogic;              -- non-cacheable access
         state        : state_t;
         second_bytes : std_ulogic_vector(7 downto 0);
+        dar          : std_ulogic_vector(63 downto 0);
+        dsisr        : std_ulogic_vector(31 downto 0);
     end record;
 
     type byte_sel_t is array(0 to 7) of std_ulogic;
@@ -135,6 +137,9 @@ begin
         variable use_second : byte_sel_t;
         variable trim_ctl : trim_ctl_t;
         variable negative : std_ulogic;
+        variable mfspr : std_ulogic;
+        variable sprn : std_ulogic_vector(9 downto 0);
+        variable sprval : std_ulogic_vector(63 downto 0);
     begin
         v := r;
         req := '0';
@@ -142,6 +147,8 @@ begin
         done := '0';
         byte_sel := (others => '0');
         addr := lsu_sum;
+        mfspr := '0';
+        sprval := (others => '0');      -- avoid inferred latches
 
         write_enable := '0';
         do_update := '0';
@@ -200,11 +207,38 @@ begin
             if l_in.valid = '1' then
                 v.load := '0';
                 v.dcbz := '0';
-                if l_in.op = OP_LOAD then
+                case l_in.op is
+                when OP_STORE =>
+                    req := '1';
+                when OP_LOAD =>
+                    req := '1';
                     v.load := '1';
-                elsif l_in.op = OP_DCBZ then
+                when OP_DCBZ =>
+                    req := '1';
                     v.dcbz := '1';
-                end if;
+                when OP_MFSPR =>
+                    done := '1';
+                    mfspr := '1';
+                    -- partial decode on SPR number should be adequate given
+                    -- the restricted set that get sent down this path
+                    sprn := std_ulogic_vector(to_unsigned(l_in.spr_num, 10));
+                    if sprn(0) = '0' then
+                        sprval := x"00000000" & r.dsisr;
+                    else
+                        sprval := r.dar;
+                    end if;
+                when OP_MTSPR =>
+                    done := '1';
+                    sprn := std_ulogic_vector(to_unsigned(l_in.spr_num, 10));
+                    if sprn(0) = '0' then
+                        v.dsisr := l_in.data(31 downto 0);
+                    else
+                        v.dar := l_in.data;
+                    end if;
+                when others =>
+                    assert false report "unknown op sent to loadstore1";
+                end case;
+
                 v.addr := lsu_sum;
                 v.write_reg := l_in.write_reg;
                 v.length := l_in.length;
@@ -246,12 +280,13 @@ begin
                     v.store_data(j + 7 downto j) := l_in.data(i * 8 + 7 downto i * 8);
                 end loop;
 
-                req := '1';
-                stall := '1';
-                if long_sel(15 downto 8) = "00000000" then
-                    v.state := LAST_ACK_WAIT;
-                else
-                    v.state := SECOND_REQ;
+                if req = '1' then
+                    stall := '1';
+                    if long_sel(15 downto 8) = "00000000" then
+                        v.state := LAST_ACK_WAIT;
+                    else
+                        v.state := SECOND_REQ;
+                    end if;
                 end if;
             end if;
 
@@ -308,7 +343,11 @@ begin
         -- Multiplex either cache data to the destination GPR or
         -- the address for the rA update.
         l_out.valid <= done;
-        if do_update = '1' then
+        if mfspr = '1' then
+            l_out.write_enable <= '1';
+            l_out.write_reg <= l_in.write_reg;
+            l_out.write_data <= sprval;
+        elsif do_update = '1' then
             l_out.write_enable <= '1';
             l_out.write_reg <= r.update_reg;
             l_out.write_data <= r.addr;