Merge pull request #158 from paulusmack/excpath

Fix exception stuff so we make timing again

common.vhdl

@@ -263,9 +263,12 @@ package common is
 	write_cr_data : std_ulogic_vector(31 downto 0);
 	write_xerc_enable : std_ulogic;
 	xerc : xer_common_t;
+        exc_write_enable : std_ulogic;
+        exc_write_reg : gspr_index_t;
+        exc_write_data : std_ulogic_vector(63 downto 0);
     end record;
     constant Execute1ToWritebackInit : Execute1ToWritebackType := (valid => '0', rc => '0', write_enable => '0',
-								   write_cr_enable => '0',
+								   write_cr_enable => '0', exc_write_enable => '0',
 								   write_xerc_enable => '0', xerc => xerc_init,
 								   others => (others => '0'));
 

decode1.vhdl

@@ -42,6 +42,7 @@ architecture behaviour of decode1 is
 		15 =>       (ALU,    OP_ADD,       RA_OR_ZERO, CONST_SI_HI, NONE, RT,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- addis
 		28 =>       (ALU,    OP_AND,       NONE,       CONST_UI,    RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', ONE,  '0', '0'), -- andi.
 		29 =>       (ALU,    OP_AND,       NONE,       CONST_UI_HI, RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', ONE,  '0', '0'), -- andis.
+                 0 =>       (ALU,    OP_ATTN,      NONE,       NONE,        NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- attn
 		18 =>       (ALU,    OP_B,         NONE,       CONST_LI,    NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '1', '0'), -- b
 		16 =>       (ALU,    OP_BC,        SPR,        CONST_BD,    NONE, SPR , '1', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '1', '0'), -- bc
 		11 =>       (ALU,    OP_CMP,       RA,         CONST_SI,    NONE, NONE, '0', '1', '1', '0', ONE,  '0', NONE, '0', '0', '0', '0', '0', '1', NONE, '0', '0'), -- cmpi
@@ -60,6 +61,7 @@ architecture behaviour of decode1 is
 		20 =>       (ALU,    OP_RLC,       RA,         CONST_SH32,  RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '1', '0', RC,   '0', '0'), -- rlwimi
 		21 =>       (ALU,    OP_RLC,       NONE,       CONST_SH32,  RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '1', '0', RC,   '0', '0'), -- rlwinm
 		23 =>       (ALU,    OP_RLC,       NONE,       RB,          RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '1', '0', RC,   '0', '0'), -- rlwnm
+                17 =>       (ALU,    OP_SC,        NONE,       NONE,        NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- sc
 		38 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- stb
 		39 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- stbu
 		44 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- sth
@@ -67,8 +69,8 @@ architecture behaviour of decode1 is
 		36 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- stw
 		37 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- stwu
 		 8 =>       (ALU,    OP_ADD,       RA,         CONST_SI,    NONE, RT,   '0', '0', '1', '0', ONE,  '1', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- subfic
-		 2 =>       (ALU,    OP_TDI,       RA,         CONST_SI,    NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- tdi
+		 2 =>       (ALU,    OP_TRAP,      RA,         CONST_SI,    NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- tdi
-		--PPC_TWI 3
+		 3 =>       (ALU,    OP_TRAP,      RA,         CONST_SI,    NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '1', '0', NONE, '0', '1'), -- twi
 		26 =>       (ALU,    OP_XOR,       NONE,       CONST_UI,    RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- xori
 		27 =>       (ALU,    OP_XOR,       NONE,       CONST_UI_HI, RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- xoris
 		others   => illegal_inst
@@ -103,7 +105,7 @@ architecture behaviour of decode1 is
 		--                 unit     internal      in1         in2          in3   out   CR   CR   inv  inv  cry   cry  ldst  BR   sgn  upd  rsrv 32b  sgn  rc    lk   sgl
 		--                               op                                            in   out   A   out  in    out  len        ext                                 pipe
                 -- mcrf; and cr logical ops
-		2#000#    =>       (ALU,    OP_MCRF,      NONE,       NONE,        NONE, NONE, '1', '1', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'),
+		2#000#    =>       (ALU,    OP_CROP,      NONE,       NONE,        NONE, NONE, '1', '1', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'),
 		-- addpcis not implemented yet
 		2#001#    =>       (ALU,    OP_ILLEGAL,   NONE,       NONE,        NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', RC,   '0', '1'),
                 -- bclr, bcctr, bctar
@@ -317,8 +319,8 @@ architecture behaviour of decode1 is
 		2#0011001000#  =>       (ALU,    OP_ADD,       RA,         NONE,        NONE, RT,   '0', '0', '1', '0', CA,   '1', NONE, '0', '0', '0', '0', '0', '0', RC,   '0', '0'), -- subfze
 		2#1011001000#  =>       (ALU,    OP_ADD,       RA,         NONE,        NONE, RT,   '0', '0', '1', '0', CA,   '1', NONE, '0', '0', '0', '0', '0', '0', RC,   '0', '0'), -- subfzeo
 		2#1001010110#  =>       (ALU,    OP_NOP,       NONE,       NONE,        NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- sync
-		-- 2#0001000100# td
+		2#0001000100#  =>       (ALU,    OP_TRAP,      RA,         RB,          NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- td
-		2#0000000100#  =>       (ALU,    OP_TW,        RA,         RB,          NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- tw
+		2#0000000100#  =>       (ALU,    OP_TRAP,      RA,         RB,          NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '1', '0', NONE, '0', '1'), -- tw
 		2#0100111100#  =>       (ALU,    OP_XOR,       NONE,       RB,          RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', RC,   '0', '0'), -- xor
 		others => illegal_inst
 	);
@@ -342,10 +344,7 @@ architecture behaviour of decode1 is
 
         --                                       unit     internal      in1         in2          in3   out   CR   CR   inv  inv  cry   cry  ldst  BR   sgn  upd  rsrv 32b  sgn  rc    lk   sgl
         --                                                      op                                           in   out   A   out  in    out  len        ext                                 pipe
-        constant attn_instr    : decode_rom_t := (ALU,    OP_ATTN,      NONE,       NONE,        NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '1');
 	constant nop_instr     : decode_rom_t := (ALU,    OP_NOP,       NONE,       NONE,        NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0');
-	constant sc_instr     :  decode_rom_t := (ALU,    OP_SC,        NONE,       NONE,        NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0');
-        constant sim_cfg_instr : decode_rom_t := (ALU,    OP_SIM_CONFIG,NONE,       NONE,        NONE, RT,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '1');
 
 begin
 	decode1_0: process(clk)
@@ -403,15 +402,6 @@ begin
                 elsif std_match(f_in.insn, "01100000000000000000000000000000") then
                         report "PPC_nop";
                         v.decode := nop_instr;
-                elsif std_match(f_in.insn, "010001--------------0000000---1-") then
-                        report "PPC_sc";
-                        v.decode := sc_instr;
-                elsif std_match(f_in.insn, "000001---------------0000000011-") then
-                        report "PPC_SIM_CONFIG";
-                        v.decode := sim_cfg_instr;
-                elsif std_match(f_in.insn, "000000---------------0100000000-") then
-                        report "PPC_attn";
-                        v.decode := attn_instr;
 
                 else
                         v.decode := major_decode_rom_array(to_integer(majorop));

decode_types.vhdl

@@ -5,20 +5,19 @@ package decode_types is
     type insn_type_t is (OP_ILLEGAL, OP_NOP, OP_ADD,
 			 OP_ADDPCIS, OP_AND, OP_ATTN, OP_B, OP_BC, OP_BCREG,
 			 OP_BPERM, OP_CMP, OP_CMPB, OP_CMPEQB, OP_CMPRB,
-			 OP_CNTZ, OP_CRAND,
+			 OP_CNTZ, OP_CROP,
-			 OP_CRANDC, OP_CREQV, OP_CRNAND, OP_CRNOR, OP_CROR, OP_CRORC,
+			 OP_DARN, OP_DCBF, OP_DCBST, OP_DCBT, OP_DCBTST,
-			 OP_CRXOR, OP_DARN, OP_DCBF, OP_DCBST, OP_DCBT, OP_DCBTST,
 			 OP_DCBZ, OP_DIV, OP_DIVE, OP_EXTS,
 			 OP_EXTSWSLI, OP_ICBI, OP_ICBT, OP_ISEL, OP_ISYNC,
-			 OP_LOAD, OP_STORE, OP_MADDHD, OP_MADDHDU, OP_MADDLD, OP_MCRF,
+			 OP_LOAD, OP_STORE, OP_MADDHD, OP_MADDHDU, OP_MADDLD,
 			 OP_MCRXR, OP_MCRXRX, OP_MFCR, OP_MFMSR, OP_MFSPR, OP_MOD,
 			 OP_MTCRF, OP_MTMSRD, OP_MTSPR, OP_MUL_L64,
 			 OP_MUL_H64, OP_MUL_H32, OP_OR,
 			 OP_POPCNT, OP_PRTY, OP_RFID,
 			 OP_RLC, OP_RLCL, OP_RLCR, OP_SC, OP_SETB,
 			 OP_SHL, OP_SHR,
-			 OP_SYNC, OP_TD, OP_TDI, OP_TW,
+			 OP_SYNC, OP_TRAP,
-			 OP_TWI, OP_XOR, OP_SIM_CONFIG
+			 OP_XOR
 			 );
     type input_reg_a_t is (NONE, RA, RA_OR_ZERO, SPR);
     type input_reg_b_t is (NONE, RB, CONST_UI, CONST_SI, CONST_SI_HI, CONST_UI_HI, CONST_LI, CONST_BD, CONST_DS, CONST_M1, CONST_SH, CONST_SH32, SPR);

execute1.vhdl

@@ -237,6 +237,9 @@ begin
         variable lv : Execute1ToLoadstore1Type;
 	variable irq_valid : std_ulogic;
 	variable exception : std_ulogic;
+        variable exception_nextpc : std_ulogic;
+        variable trapval : std_ulogic_vector(4 downto 0);
+        variable illegal : std_ulogic;
     begin
 	result := (others => '0');
 	result_with_carry := (others => '0');
@@ -386,24 +389,30 @@ begin
 
 	ctrl_tmp.irq_state <= WRITE_SRR0;
 	exception := '0';
+        illegal := '0';
+        exception_nextpc := '0';
+        v.e.exc_write_enable := '0';
+        v.e.exc_write_reg := fast_spr_num(SPR_SRR0);
+        v.e.exc_write_data := e_in.nia;
 
 	if ctrl.irq_state = WRITE_SRR1 then
-	    v.e.write_reg := fast_spr_num(SPR_SRR1);
+	    v.e.exc_write_reg := fast_spr_num(SPR_SRR1);
-	    result := ctrl.srr1;
+	    v.e.exc_write_data := ctrl.srr1;
-	    result_en := '1';
+            v.e.exc_write_enable := '1';
 	    ctrl_tmp.msr(63 - 48) <= '0'; -- clear EE
 	    f_out.redirect <= '1';
 	    f_out.redirect_nia <= ctrl.irq_nia;
-	    v.e.valid := '1';
+	    v.e.valid := e_in.valid;
 	    report "Writing SRR1: " & to_hstring(ctrl.srr1);
 
 	elsif irq_valid = '1' then
 	    -- we need two cycles to write srr0 and 1
 	    -- will need more when we have to write DSISR, DAR and HIER
-	    exception := '1';
+            -- Don't deliver the interrupt until we have a valid instruction
+            -- coming in, so we have a valid NIA to put in SRR0.
+	    exception := e_in.valid;
 	    ctrl_tmp.irq_nia <= std_logic_vector(to_unsigned(16#900#, 64));
 	    ctrl_tmp.srr1 <= msr_copy(ctrl.msr);
-	    result := e_in.nia;
 
 	elsif e_in.valid = '1' then
 
@@ -419,29 +428,33 @@ 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
-		exception := '1';
+		illegal := '1';
-		ctrl_tmp.irq_nia <= std_logic_vector(to_unsigned(16#700#, 64));
-		ctrl_tmp.srr1 <= msr_copy(ctrl.msr);
-		-- Since we aren't doing Hypervisor emulation assist (0xe40) we
-		-- set bit 44 to indicate we have an illegal
-		ctrl_tmp.srr1(63 - 44) <= '1';
-		result := e_in.nia;
-		report "illegal";
 	    when OP_SC =>
-		-- FIXME Assume everything is SC (not SCV) for now
+		-- 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';
                     ctrl_tmp.irq_nia <= std_logic_vector(to_unsigned(16#C00#, 64));
                     ctrl_tmp.srr1 <= msr_copy(ctrl.msr);
-		result := std_logic_vector(unsigned(e_in.nia) + 4);
                     report "sc";
+                else
+                    illegal := '1';
+                end if;
 	    when OP_ATTN =>
+                -- check bits 1-10 of the instruction to make sure it's attn
+                -- if not then it is illegal
+                if e_in.insn(10 downto 1) = "0100000000" then
                     terminate_out <= '1';
                     report "ATTN";
+                else
+                    illegal := '1';
+                end if;
 	    when OP_NOP =>
 		-- Do nothing
-	    when OP_ADD | OP_CMP =>
+	    when OP_ADD | OP_CMP | OP_TRAP =>
 		if e_in.invert_a = '0' then
 		    a_inv := a_in;
 		else
@@ -463,18 +476,18 @@ begin
                     end if;
                     result_en := '1';
                 else
-                    -- CMP and CMPL instructions
+                    -- trap, CMP and CMPL instructions
                     -- Note, we have done RB - RA, not RA - RB
-                    bf := insn_bf(e_in.insn);
+                    if e_in.insn_type = OP_CMP then
                         l := insn_l(e_in.insn);
-                    v.e.write_cr_enable := '1';
+                    else
-                    crnum := to_integer(unsigned(bf));
+                        l := not e_in.is_32bit;
-                    v.e.write_cr_mask := num_to_fxm(crnum);
+                    end if;
                     zerolo := not (or (a_in(31 downto 0) xor b_in(31 downto 0)));
                     zerohi := not (or (a_in(63 downto 32) xor b_in(63 downto 32)));
                     if zerolo = '1' and (l = '0' or zerohi = '1') then
                         -- values are equal
-                        newcrf := "001" & v.e.xerc.so;
+                        trapval := "00100";
                     else
                         if l = '1' then
                             -- 64-bit comparison
@@ -489,19 +502,41 @@ begin
                             -- Subtraction might overflow, but
                             -- comparison is clear from MSB difference.
                             -- for signed, 0 is greater; for unsigned, 1 is greater
-                            a_lt := msb_a xnor e_in.is_signed;
+                            trapval := msb_a & msb_b & '0' & msb_b & msb_a;
                         else
                             -- Subtraction cannot overflow since MSBs are equal.
                             -- carry = 1 indicates RA is smaller (signed or unsigned)
                             a_lt := (not l and carry_32) or (l and carry_64);
+                            trapval := a_lt & not a_lt & '0' & a_lt & not a_lt;
+                        end if;
                     end if;
-                        newcrf := a_lt & not a_lt & '0' & v.e.xerc.so;
+                    if e_in.insn_type = OP_CMP then
+                        if e_in.is_signed = '1' then
+                            newcrf := trapval(4 downto 2) & v.e.xerc.so;
+                        else
+                            newcrf := trapval(1 downto 0) & trapval(2) & v.e.xerc.so;
                         end if;
+                        bf := insn_bf(e_in.insn);
+                        crnum := to_integer(unsigned(bf));
+                        v.e.write_cr_enable := '1';
+                        v.e.write_cr_mask := num_to_fxm(crnum);
                         for i in 0 to 7 loop
                             lo := i*4;
                             hi := lo + 3;
                             v.e.write_cr_data(hi downto lo) := newcrf;
                         end loop;
+                    else
+                        -- trap instructions (tw, twi, td, tdi)
+                        if or (trapval and insn_to(e_in.insn)) = '1' then
+                            -- generate trap-type program interrupt
+                            exception := '1';
+                            ctrl_tmp.irq_nia <= std_logic_vector(to_unsigned(16#700#, 64));
+                            ctrl_tmp.srr1 <= msr_copy(ctrl.msr);
+                            -- set bit 46 to say trap occurred
+                            ctrl_tmp.srr1(63 - 46) <= '1';
+                            report "trap";
+                        end if;
+                    end if;
                 end if;
 	    when OP_AND | OP_OR | OP_XOR =>
 		result := logical_result;
@@ -578,7 +613,7 @@ begin
 		    result := b_in;
 		end if;
 		result_en := '1';
-	    when OP_MCRF =>
+	    when OP_CROP =>
 		cr_op := insn_cr(e_in.insn);
 		report "CR OP " & to_hstring(cr_op);
 		if cr_op(0) = '0' then -- MCRF
@@ -715,15 +750,6 @@ begin
 		    set_carry(v.e, rotator_carry, rotator_carry);
 		end if;
 		result_en := '1';
-	    when OP_SIM_CONFIG =>
-		-- bit 0 was used to select the microwatt console, which
-		-- we no longer support.
-		result := x"0000000000000000";
-		result_en := '1';
-
-	    when OP_TDI =>
-		-- Keep our test cases happy for now, ignore trap instructions
-		report "OP_TDI FIXME";
 
 	    when OP_ISYNC =>
 		f_out.redirect <= '1';
@@ -817,14 +843,22 @@ begin
 	    end if;
 	end if;
 
+        if illegal = '1' then
+            exception := '1';
+            ctrl_tmp.irq_nia <= std_logic_vector(to_unsigned(16#700#, 64));
+            ctrl_tmp.srr1 <= msr_copy(ctrl.msr);
+            -- Since we aren't doing Hypervisor emulation assist (0xe40) we
+            -- set bit 44 to indicate we have an illegal
+            ctrl_tmp.srr1(63 - 44) <= '1';
+            report "illegal";
+        end if;
 	if exception = '1' then
-	    v.e.write_reg := fast_spr_num(SPR_SRR0);
+            v.e.exc_write_enable := '1';
-	    if e_in.valid = '1' then
+            if exception_nextpc = '1' then
-		result_en := '1';
+                v.e.exc_write_data := next_nia;
-		ctrl_tmp.irq_state <= WRITE_SRR1;
-		stall_out <= '1';
-		v.e.valid := '0';
             end if;
+            ctrl_tmp.irq_state <= WRITE_SRR1;
+            v.e.valid := '1';
 	end if;
 
 	v.e.write_data := result;