Add basic XER support

The carry is currently internal to execute1. We don't handle any of
the other XER fields.

This creates type called "xer_common_t" that contains the commonly
used XER bits (CA, CA32, SO, OV, OV32).

The value is stored in the CR file (though it could be a separate
module). The rest of the bits will be implemented as a separate
SPR and the two parts reconciled in mfspr/mtspr in latter commits.

We always read XER in decode2 (there is little point not to)
and send it down all pipeline branches as it will be needed in
writeback for all type of instructions when CR0:SO needs to be
updated (such forms exist for all pipeline branches even if we don't
yet implement them).

To avoid having to track XER hazards, we forward it back in EX1. This
assumes that other pipeline branches that can modify it (mult and div)
are running single issue for now.

One additional hazard to beware of is an XER:SO modifying instruction
in EX1 followed immediately by a store conditional. Due to our writeback
latency, the store will go down the LSU with the previous XER value,
thus the stcx. will set CR0:SO using an obsolete SO value.

I doubt there exist any code relying on this behaviour being correct
but we should account for it regardless, possibly by ensuring that
stcx. remain single issue initially, or later by adding some minimal
tracking or moving the LSU into the same pipeline as execute.

Missing some obscure XER affecting instructions like addex or mcrxrx.

[paulus@ozlabs.org - fix CA32 and OV32 for OP_ADD, fix order of
 arguments to set_ov]

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>

common.vhdl

@@ -12,15 +12,28 @@ package common is
 
     function decode_spr_num(insn: std_ulogic_vector(31 downto 0)) return spr_num_t;
 
+    constant SPR_XER   : spr_num_t := 1;
     constant SPR_LR    : spr_num_t := 8;
     constant SPR_CTR   : spr_num_t := 9;
     constant SPR_TB    : spr_num_t := 268;
 
+    -- The XER is split: the common bits (CA, OV, SO, OV32 and CA32) are
+    -- in the CR file as a kind of CR extension (with a separate write
+    -- control). The rest is stored as a fast SPR.
+    type xer_common_t is record
+	ca : std_ulogic;
+	ca32 : std_ulogic;
+	ov : std_ulogic;
+	ov32 : std_ulogic;
+	so : std_ulogic;
+    end record;
+    constant xerc_init : xer_common_t := (others => '0');
+
+    -- This needs to die...
     type ctrl_t is record
 	lr: std_ulogic_vector(63 downto 0);
 	ctr: std_ulogic_vector(63 downto 0);
 	tb: std_ulogic_vector(63 downto 0);
-	carry: std_ulogic;
     end record;
 
     type Fetch1ToIcacheType is record
@@ -64,8 +77,10 @@ package common is
 	read_data2: std_ulogic_vector(63 downto 0);
 	read_data3: std_ulogic_vector(63 downto 0);
 	cr: std_ulogic_vector(31 downto 0);
+	xerc: xer_common_t;
 	lr: std_ulogic;
 	rc: std_ulogic;
+	oe: std_ulogic;
 	invert_a: std_ulogic;
 	invert_out: std_ulogic;
 	input_carry: carry_in_t;
@@ -78,9 +93,9 @@ package common is
 	data_len: std_ulogic_vector(3 downto 0);
     end record;
     constant Decode2ToExecute1Init : Decode2ToExecute1Type :=
-	(valid => '0', insn_type => OP_ILLEGAL, lr => '0', rc => '0', invert_a => '0',
+	(valid => '0', insn_type => OP_ILLEGAL, lr => '0', rc => '0', oe => '0', invert_a => '0',
 	 invert_out => '0', input_carry => ZERO, output_carry => '0', input_cr => '0', output_cr => '0',
-	 is_32bit => '0', is_signed => '0', others => (others => '0'));
+	 is_32bit => '0', is_signed => '0', xerc => xerc_init, others => (others => '0'));
 
     type Decode2ToMultiplyType is record
 	valid: std_ulogic;
@@ -89,8 +104,13 @@ package common is
 	data1: std_ulogic_vector(64 downto 0);
 	data2: std_ulogic_vector(64 downto 0);
 	rc: std_ulogic;
+	oe: std_ulogic;
+	is_32bit: std_ulogic;
+	xerc: xer_common_t;
     end record;
-    constant Decode2ToMultiplyInit : Decode2ToMultiplyType := (valid => '0', insn_type => OP_ILLEGAL, rc => '0', others => (others => '0'));
+    constant Decode2ToMultiplyInit : Decode2ToMultiplyType := (valid => '0', insn_type => OP_ILLEGAL, rc => '0',
+							       oe => '0', is_32bit => '0', xerc => xerc_init,
+							       others => (others => '0'));
 
     type Decode2ToDividerType is record
 	valid: std_ulogic;
@@ -102,8 +122,13 @@ package common is
 	is_extended: std_ulogic;
 	is_modulus: std_ulogic;
 	rc: std_ulogic;
+	oe: std_ulogic;
+	xerc: xer_common_t;
     end record;
-    constant Decode2ToDividerInit: Decode2ToDividerType := (valid => '0', is_signed => '0', is_32bit => '0', is_extended => '0', is_modulus => '0', rc => '0', others => (others => '0'));
+    constant Decode2ToDividerInit: Decode2ToDividerType := (valid => '0', is_signed => '0', is_32bit => '0',
+							    is_extended => '0', is_modulus => '0',
+							    rc => '0', oe => '0', xerc => xerc_init,
+							    others => (others => '0'));
 
     type Decode2ToRegisterFileType is record
 	read1_enable : std_ulogic;
@@ -126,6 +151,7 @@ package common is
 
     type CrFileToDecode2Type is record
 	read_cr_data   : std_ulogic_vector(31 downto 0);
+	read_xerc_data : xer_common_t;
     end record;
 
     type Execute1ToFetch1Type is record
@@ -146,8 +172,11 @@ package common is
 	sign_extend : std_ulogic;			-- do we need to sign extend?
 	update : std_ulogic;				-- is this an update instruction?
 	update_reg : std_ulogic_vector(4 downto 0);	-- if so, the register to update
+	xerc : xer_common_t;
     end record;
-    constant Decode2ToLoadstore1Init : Decode2ToLoadstore1Type := (valid => '0', load => '0', byte_reverse => '0', sign_extend => '0', update => '0', others => (others => '0'));
+    constant Decode2ToLoadstore1Init : Decode2ToLoadstore1Type := (valid => '0', load => '0', byte_reverse => '0',
+								   sign_extend => '0', update => '0', xerc => xerc_init,
+								   others => (others => '0'));
 
     type Loadstore1ToDcacheType is record
 	valid : std_ulogic;
@@ -161,6 +190,7 @@ package common is
 	sign_extend : std_ulogic;
 	update : std_ulogic;
 	update_reg : std_ulogic_vector(4 downto 0);
+	xerc : xer_common_t;
     end record;
 
     type DcacheToWritebackType is record
@@ -173,8 +203,11 @@ package common is
 	sign_extend : std_ulogic;
 	byte_reverse : std_ulogic;
 	second_word : std_ulogic;
+	xerc : xer_common_t;
     end record;
-    constant DcacheToWritebackInit : DcacheToWritebackType := (valid => '0', write_enable => '0', sign_extend => '0', byte_reverse => '0', second_word => '0', others => (others => '0'));
+    constant DcacheToWritebackInit : DcacheToWritebackType := (valid => '0', write_enable => '0', sign_extend => '0',
+							       byte_reverse => '0', second_word => '0', xerc => xerc_init,
+							       others => (others => '0'));
 
     type Execute1ToWritebackType is record
 	valid: std_ulogic;
@@ -186,9 +219,14 @@ package common is
 	write_cr_enable : std_ulogic;
 	write_cr_mask : std_ulogic_vector(7 downto 0);
 	write_cr_data : std_ulogic_vector(31 downto 0);
+	write_xerc_enable : std_ulogic;
+	xerc : xer_common_t;
 	sign_extend: std_ulogic;
     end record;
-    constant Execute1ToWritebackInit : Execute1ToWritebackType := (valid => '0', rc => '0', write_enable => '0', write_cr_enable => '0', sign_extend => '0', others => (others => '0'));
+    constant Execute1ToWritebackInit : Execute1ToWritebackType := (valid => '0', rc => '0', write_enable => '0',
+								   write_cr_enable => '0', sign_extend => '0',
+								   write_xerc_enable => '0', xerc => xerc_init,
+								   others => (others => '0'));
 
     type MultiplyToWritebackType is record
 	valid: std_ulogic;
@@ -196,9 +234,14 @@ package common is
 	write_reg_enable : std_ulogic;
 	write_reg_nr: std_ulogic_vector(4 downto 0);
 	write_reg_data: std_ulogic_vector(63 downto 0);
+	write_xerc_enable : std_ulogic;
+	xerc : xer_common_t;
 	rc: std_ulogic;
     end record;
-    constant MultiplyToWritebackInit : MultiplyToWritebackType := (valid => '0', write_reg_enable => '0', rc => '0', others => (others => '0'));
+    constant MultiplyToWritebackInit : MultiplyToWritebackType := (valid => '0', write_reg_enable => '0',
+								   rc => '0', write_xerc_enable => '0',
+								   xerc => xerc_init,
+								   others => (others => '0'));
 
     type DividerToWritebackType is record
 	valid: std_ulogic;
@@ -206,9 +249,14 @@ package common is
 	write_reg_enable : std_ulogic;
 	write_reg_nr: std_ulogic_vector(4 downto 0);
 	write_reg_data: std_ulogic_vector(63 downto 0);
+	write_xerc_enable : std_ulogic;
+	xerc : xer_common_t;
 	rc: std_ulogic;
     end record;
-    constant DividerToWritebackInit : DividerToWritebackType := (valid => '0', write_reg_enable => '0', rc => '0', others => (others => '0'));
+    constant DividerToWritebackInit : DividerToWritebackType := (valid => '0', write_reg_enable => '0',
+								 rc => '0', write_xerc_enable => '0',
+								 xerc => xerc_init,
+								 others => (others => '0'));
 
     type WritebackToRegisterFileType is record
 	write_reg : std_ulogic_vector(4 downto 0);
@@ -221,9 +269,12 @@ package common is
 	write_cr_enable : std_ulogic;
 	write_cr_mask : std_ulogic_vector(7 downto 0);
 	write_cr_data : std_ulogic_vector(31 downto 0);
+	write_xerc_enable : std_ulogic;
+	write_xerc_data : xer_common_t;
     end record;
-    constant WritebackToCrFileInit : WritebackToCrFileType := (write_cr_enable => '0', others => (others => '0'));
-
+    constant WritebackToCrFileInit : WritebackToCrFileType := (write_cr_enable => '0', write_xerc_enable => '0',
+							       write_xerc_data => xerc_init,
+							       others => (others => '0'));
 end common;
 
 package body common is

cr_file.vhdl

@@ -18,7 +18,9 @@ end entity cr_file;
 
 architecture behaviour of cr_file is
     signal crs : std_ulogic_vector(31 downto 0) := (others => '0');
-    signal crs_updated : std_ulogic_vector(31 downto 0) := (others => '0');
+    signal crs_updated : std_ulogic_vector(31 downto 0);
+    signal xerc : xer_common_t := xerc_init;
+    signal xerc_updated : xer_common_t;
 begin
     cr_create_0: process(all)
         variable hi, lo : integer := 0;
@@ -35,6 +37,13 @@ begin
         end loop;
 
         crs_updated <= cr_tmp;
+
+	if w_in.write_xerc_enable = '1' then
+	    xerc_updated <= w_in.write_xerc_data;
+	else
+	    xerc_updated <= xerc;
+	end if;
+
     end process;
 
     -- synchronous writes
@@ -45,6 +54,10 @@ begin
                 report "Writing " & to_hstring(w_in.write_cr_data) & " to CR mask " & to_hstring(w_in.write_cr_mask);
 		crs <= crs_updated;
             end if;
+	    if w_in.write_xerc_enable = '1' then
+                report "Writing XERC";
+		xerc <= xerc_updated;
+	    end if;
         end if;
     end process;
 
@@ -56,5 +69,6 @@ begin
             report "Reading CR " & to_hstring(crs_updated);
         end if;
         d_out.read_cr_data <= crs_updated;
+        d_out.read_xerc_data <= xerc_updated;
     end process;
 end architecture behaviour;

dcache.vhdl

@@ -185,6 +185,7 @@ architecture rtl of dcache is
 	length         : std_ulogic_vector(3 downto 0);
 	sign_extend    : std_ulogic;
 	byte_reverse   : std_ulogic;
+	xerc           : xer_common_t;
     end record;
 
     signal r2 : reg_stage_2_t;
@@ -469,6 +470,7 @@ begin
 	d_out.sign_extend <= r2.sign_extend;
 	d_out.byte_reverse <= r2.byte_reverse;
 	d_out.second_word <= '0';
+	d_out.xerc <= r2.xerc;
 
 	-- We have a valid load or store hit or we just completed a slow
 	-- op such as a load miss, a NC load or a store
@@ -518,6 +520,7 @@ begin
 		d_out.sign_extend <= r1.req.sign_extend;
 		d_out.byte_reverse <= r1.req.byte_reverse;
 		d_out.write_len <= r1.req.length;
+		d_out.xerc <= r1.req.xerc;
 	    end if;
 
 	    -- If it's a store or a non-update load form, complete now
@@ -539,6 +542,7 @@ begin
 	    d_out.write_len <= "1000";
 	    d_out.sign_extend <= '0';
 	    d_out.byte_reverse <= '0';
+	    d_out.xerc <= r1.req.xerc;
 
 	    -- If it was a load, this completes the operation (load with
 	    -- update case).

decode2.vhdl

@@ -131,6 +131,22 @@ architecture behaviour of decode2 is
 		end case;
 	end;
 
+	-- For now, use "rc" in the decode table to decide whether oe exists.
+	-- This is not entirely correct architecturally: For mulhd and
+	-- mulhdu, the OE field is reserved. It remains to be seen what an
+	-- actual POWER9 does if we set it on those instructions, for now we
+	-- test that further down when assigning to the multiplier oe input.
+	--
+	function decode_oe (t : rc_t; insn_in : std_ulogic_vector(31 downto 0)) return std_ulogic is
+	begin
+		case t is
+		when RC =>
+			return insn_oe(insn_in);
+		when OTHERS =>
+			return '0';
+		end case;
+	end;
+
 	-- issue control signals
 	signal control_valid_in : std_ulogic;
 	signal control_valid_out : std_ulogic;
@@ -255,7 +271,9 @@ begin
                 v.e.read_data3 := decoded_reg_c.data;
 		v.e.write_reg := decode_output_reg(d_in.decode.output_reg_a, d_in.insn);
 		v.e.rc := decode_rc(d_in.decode.rc, d_in.insn);
+		v.e.oe := decode_oe(d_in.decode.rc, d_in.insn);
 		v.e.cr := c_in.read_cr_data;
+		v.e.xerc := c_in.read_xerc_data;
                 v.e.invert_a := d_in.decode.invert_a;
                 v.e.invert_out := d_in.decode.invert_out;
 		v.e.input_carry := d_in.decode.input_carry;
@@ -274,6 +292,11 @@ begin
 		mul_b := decoded_reg_b.data;
 		v.m.write_reg := decode_output_reg(d_in.decode.output_reg_a, d_in.insn);
 		v.m.rc := decode_rc(d_in.decode.rc, d_in.insn);
+		v.m.xerc := c_in.read_xerc_data;
+		if v.m.insn_type = OP_MUL_L64 then
+		  v.m.oe := decode_oe(d_in.decode.rc, d_in.insn);
+		end if;
+		v.m.is_32bit := d_in.decode.is_32bit;
 
 		if d_in.decode.is_32bit = '1' then
 			if d_in.decode.is_signed = '1' then
@@ -337,6 +360,8 @@ begin
                         end if;
                 end if;
                 v.d.rc := decode_rc(d_in.decode.rc, d_in.insn);
+		v.d.xerc := c_in.read_xerc_data;
+		v.d.oe := decode_oe(d_in.decode.rc, d_in.insn);
 
 		-- load/store unit
 		v.l.update_reg := decoded_reg_a.reg;
@@ -355,6 +380,7 @@ begin
 		v.l.byte_reverse := d_in.decode.byte_reverse;
 		v.l.sign_extend := d_in.decode.sign_extend;
 		v.l.update := d_in.decode.update;
+		v.l.xerc := c_in.read_xerc_data;
 
 		-- issue control
 		control_valid_in <= d_in.valid;

divider.vhdl

@@ -36,7 +36,8 @@ architecture behaviour of divider is
     signal overflow   : std_ulogic;
     signal ovf32      : std_ulogic;
     signal did_ovf    : std_ulogic;
-
+    signal oe         : std_ulogic;
+    signal xerc       : xer_common_t;
 begin
     divider_0: process(clk)
     begin
@@ -62,6 +63,8 @@ begin
                 is_32bit <= d_in.is_32bit;
                 is_signed <= d_in.is_signed;
                 rc <= d_in.rc;
+                oe <= d_in.oe;
+		xerc <= d_in.xerc;
                 count <= "1111111";
                 running <= '1';
                 overflow <= '0';
@@ -147,13 +150,25 @@ begin
     divider_out: process(clk)
     begin
         if rising_edge(clk) then
+	    d_out.valid <= '0';
             d_out.write_reg_data <= oresult;
+	    d_out.write_reg_enable <= '0';
+	    d_out.write_xerc_enable <= '0';
+	    d_out.xerc <= xerc;
             if count = "1000000" then
                 d_out.valid <= '1';
                 d_out.write_reg_enable <= '1';
-            else
-                d_out.valid <= '0';
-                d_out.write_reg_enable <= '0';
+		d_out.write_xerc_enable <= oe;
+
+		-- We must test oe because the RC update code in writeback
+		-- will use the xerc value to set CR0:SO so we must not clobber
+		-- xerc if OE wasn't set.
+		--
+		if oe = '1' then
+		    d_out.xerc.ov <= did_ovf;
+		    d_out.xerc.ov32 <= did_ovf;
+		    d_out.xerc.so <= xerc.so or did_ovf;
+		end if;
             end if;
         end if;
     end process;

execute1.vhdl

@@ -31,14 +31,13 @@ end entity execute1;
 
 architecture behaviour of execute1 is
     type reg_type is record
-	--f : Execute1ToFetch1Type;
 	e : Execute1ToWritebackType;
     end record;
 
     signal r, rin : reg_type;
 
-    signal ctrl: ctrl_t := (carry => '0', others => (others => '0'));
-    signal ctrl_tmp: ctrl_t := (carry => '0', others => (others => '0'));
+    signal ctrl: ctrl_t := (others => (others => '0'));
+    signal ctrl_tmp: ctrl_t := (others => (others => '0'));
 
     signal right_shift, rot_clear_left, rot_clear_right: std_ulogic;
     signal rotator_result: std_ulogic_vector(63 downto 0);
@@ -46,17 +45,46 @@ architecture behaviour of execute1 is
     signal logical_result: std_ulogic_vector(63 downto 0);
     signal countzero_result: std_ulogic_vector(63 downto 0);
 
-    function decode_input_carry (carry_sel : carry_in_t; ca_in : std_ulogic) return std_ulogic is
+    procedure set_carry(e: inout Execute1ToWritebackType;
+			carry32 : in std_ulogic;
+			carry : in std_ulogic) is
     begin
-	case carry_sel is
+	e.xerc.ca32 := carry32;
+	e.xerc.ca := carry;
+	e.write_xerc_enable := '1';
+    end;
+
+    procedure set_ov(e: inout Execute1ToWritebackType;
+		     ov   : in std_ulogic;
+		     ov32 : in std_ulogic) is
+    begin
+	e.xerc.ov32 := ov32;
+	e.xerc.ov := ov;
+	if ov = '1' then
+	    e.xerc.so := '1';
+	end if;
+	e.write_xerc_enable := '1';
+    end;
+
+    function calc_ov(msb_a : std_ulogic; msb_b: std_ulogic;
+		     ca: std_ulogic; msb_r: std_ulogic) return std_ulogic is
+    begin
+	return (ca xor msb_r) and not (msb_a xor msb_b);
+    end;
+
+    function decode_input_carry(ic : carry_in_t;
+				xerc : xer_common_t) return std_ulogic is
+    begin
+	case ic is
 	when ZERO =>
 	    return '0';
 	when CA =>
-	    return ca_in;
+	    return xerc.ca;
 	when ONE =>
 	    return '1';
 	end case;
     end;
+
 begin
 
     rotator_0: entity work.rotator
@@ -117,6 +145,7 @@ begin
 	variable bf, bfa : std_ulogic_vector(2 downto 0);
 	variable l : std_ulogic;
 	variable next_nia : std_ulogic_vector(63 downto 0);
+        variable carry_32, carry_64 : std_ulogic;
     begin
 	result := (others => '0');
 	result_with_carry := (others => '0');
@@ -125,7 +154,41 @@ begin
 
 	v := r;
 	v.e := Execute1ToWritebackInit;
-	--v.f := Execute1ToFetch1TypeInit;
+
+	-- XER forwarding. To avoid having to track XER hazards, we
+	-- use the previously latched value.
+	--
+	-- If the XER was modified by a multiply or a divide, those are
+	-- single issue, we'll get the up to date value from decode2 from
+	-- the register file.
+	--
+	-- If it was modified by an instruction older than the previous
+	-- one in EX1, it will have also hit writeback and will be up
+	-- to date in decode2.
+	--
+	-- That leaves us with the case where it was updated by the previous
+	-- instruction in EX1. In that case, we can forward it back here.
+	--
+	-- This will break if we allow pipelining of multiply and divide,
+	-- but ideally, those should go via EX1 anyway and run as a state
+	-- machine from here.
+	--
+	-- One additional hazard to beware of is an XER:SO modifying instruction
+	-- in EX1 followed immediately by a store conditional. Due to our
+	-- writeback latency, the store will go down the LSU with the previous
+	-- XER value, thus the stcx. will set CR0:SO using an obsolete SO value.
+	--
+	-- We will need to handle that if we ever make stcx. not single issue
+	--
+	-- We always pass a valid XER value downto writeback even when
+	-- we aren't updating it, in order for XER:SO -> CR0:SO transfer
+	-- to work for RC instructions.
+	--
+	if r.e.write_xerc_enable = '1' then
+	    v.e.xerc := r.e.xerc;
+	else
+	    v.e.xerc := e_in.xerc;
+	end if;
 
 	ctrl_tmp <= ctrl;
 	-- FIXME: run at 512MHz not core freq
@@ -163,10 +226,18 @@ begin
 		else
 		    a_inv := not e_in.read_data1;
 		end if;
-		result_with_carry := ppc_adde(a_inv, e_in.read_data2, decode_input_carry(e_in.input_carry, ctrl.carry));
+		result_with_carry := ppc_adde(a_inv, e_in.read_data2,
+					      decode_input_carry(e_in.input_carry, v.e.xerc));
 		result := result_with_carry(63 downto 0);
-		if e_in.output_carry then
-		    ctrl_tmp.carry <= result_with_carry(64);
+                carry_32 := result(32) xor a_inv(32) xor e_in.read_data2(32);
+                carry_64 := result_with_carry(64);
+		if e_in.output_carry = '1' then
+		    set_carry(v.e, carry_32, carry_64);
+		end if;
+		if e_in.oe = '1' then
+		    set_ov(v.e,
+			   calc_ov(a_inv(63), e_in.read_data2(63), carry_64, result_with_carry(63)),
+			   calc_ov(a_inv(31), e_in.read_data2(31), carry_32, result_with_carry(31)));
 		end if;
 		result_en := '1';
 	    when OP_AND | OP_OR | OP_XOR =>
@@ -270,6 +341,13 @@ begin
 		end loop;
 	    when OP_MFSPR =>
 		case decode_spr_num(e_in.insn) is
+		when SPR_XER =>
+		    result := ( 63-32 => v.e.xerc.so,
+				63-33 => v.e.xerc.ov,
+				63-34 => v.e.xerc.ca,
+				63-44 => v.e.xerc.ov32,
+				63-45 => v.e.xerc.ca32,
+				others => '0');
 		when SPR_CTR =>
 		    result := ctrl.ctr;
 		when SPR_LR =>
@@ -310,6 +388,13 @@ begin
 		v.e.write_cr_data := e_in.read_data3(31 downto 0);
 	    when OP_MTSPR =>
 		case decode_spr_num(e_in.insn) is
+		when SPR_XER =>
+		    v.e.xerc.so := e_in.read_data3(63-32);
+		    v.e.xerc.ov := e_in.read_data3(63-33);
+		    v.e.xerc.ca := e_in.read_data3(63-34);
+		    v.e.xerc.ov32 := e_in.read_data3(63-44);
+		    v.e.xerc.ca32 := e_in.read_data3(63-45);
+		    v.e.write_xerc_enable := '1';
 		when SPR_CTR =>
 		    ctrl_tmp.ctr <= e_in.read_data3;
 		when SPR_LR =>
@@ -334,7 +419,7 @@ begin
 	    when OP_RLC | OP_RLCL | OP_RLCR | OP_SHL | OP_SHR =>
 		result := rotator_result;
 		if e_in.output_carry = '1' then
-		    ctrl_tmp.carry <= rotator_carry;
+		    set_carry(v.e, rotator_carry, rotator_carry);
 		end if;
 		result_en := '1';
 	    when OP_SIM_CONFIG =>

insn_helpers.vhdl

@@ -16,6 +16,7 @@ package insn_helpers is
     function insn_lk (insn_in : std_ulogic_vector) return std_ulogic;
     function insn_aa (insn_in : std_ulogic_vector) return std_ulogic;
     function insn_rc (insn_in : std_ulogic_vector) return std_ulogic;
+    function insn_oe (insn_in : std_ulogic_vector) return std_ulogic;
     function insn_bd (insn_in : std_ulogic_vector) return std_ulogic_vector;
     function insn_bf (insn_in : std_ulogic_vector) return std_ulogic_vector;
     function insn_bfa (insn_in : std_ulogic_vector) return std_ulogic_vector;
@@ -103,6 +104,11 @@ package body insn_helpers is
         return insn_in(0);
     end;
 
+    function insn_oe (insn_in : std_ulogic_vector) return std_ulogic is
+    begin
+        return insn_in(10);
+    end;
+
     function insn_bd (insn_in : std_ulogic_vector) return std_ulogic_vector is
     begin
         return insn_in(15 downto 2);

loadstore1.vhdl

@@ -47,6 +47,7 @@ begin
         v.sign_extend := l_in.sign_extend;
         v.update := l_in.update;
         v.update_reg := l_in.update_reg;
+	v.xerc := l_in.xerc;
 
 	-- XXX Temporary hack. Mark the op as non-cachable if the address
 	-- is the form 0xc-------

multiply.vhdl

@@ -27,8 +27,17 @@ architecture behaviour of multiply is
         data      : signed(129 downto 0);
         write_reg : std_ulogic_vector(4 downto 0);
         rc        : std_ulogic;
+	oe        : std_ulogic;
+	is_32bit  : std_ulogic;
+	xerc      : xer_common_t;
     end record;
-    constant MultiplyPipelineStageInit : multiply_pipeline_stage := (valid => '0', insn_type => OP_ILLEGAL, rc => '0', data => (others => '0'), others => (others => '0'));
+    constant MultiplyPipelineStageInit : multiply_pipeline_stage := (valid => '0',
+								     insn_type => OP_ILLEGAL,
+								     rc => '0', oe => '0',
+								     is_32bit => '0',
+								     xerc => xerc_init,
+								     data => (others => '0'),
+								     others => (others => '0'));
 
     type multiply_pipeline_type is array(0 to PIPELINE_DEPTH-1) of multiply_pipeline_stage;
     constant MultiplyPipelineInit : multiply_pipeline_type := (others => MultiplyPipelineStageInit);
@@ -51,6 +60,7 @@ begin
         variable v : reg_type;
         variable d : std_ulogic_vector(129 downto 0);
         variable d2 : std_ulogic_vector(63 downto 0);
+	variable ov : std_ulogic;
     begin
         v := r;
 
@@ -61,16 +71,26 @@ begin
         v.multiply_pipeline(0).data := signed(m.data1) * signed(m.data2);
         v.multiply_pipeline(0).write_reg := m.write_reg;
         v.multiply_pipeline(0).rc := m.rc;
+        v.multiply_pipeline(0).oe := m.oe;
+        v.multiply_pipeline(0).is_32bit := m.is_32bit;
+        v.multiply_pipeline(0).xerc := m.xerc;
 
         loop_0: for i in 1 to PIPELINE_DEPTH-1 loop
             v.multiply_pipeline(i) := r.multiply_pipeline(i-1);
         end loop;
 
         d := std_ulogic_vector(v.multiply_pipeline(PIPELINE_DEPTH-1).data);
+	ov := '0';
 
+	-- TODO: Handle overflows
         case_0: case v.multiply_pipeline(PIPELINE_DEPTH-1).insn_type is
             when OP_MUL_L64 =>
                 d2 := d(63 downto 0);
+		if v.multiply_pipeline(PIPELINE_DEPTH-1).is_32bit = '1' then
+		    ov := (or d(63 downto 31)) and not (and d(63 downto 31));
+		else
+		    ov := (or d(127 downto 63)) and not (and d(127 downto 63));
+		end if;
             when OP_MUL_H32 =>
                 d2 := d(63 downto 32) & d(63 downto 32);
             when OP_MUL_H64 =>
@@ -82,11 +102,24 @@ begin
 
         m_out.write_reg_data <= d2;
         m_out.write_reg_nr <= v.multiply_pipeline(PIPELINE_DEPTH-1).write_reg;
+	m_out.xerc <= v.multiply_pipeline(PIPELINE_DEPTH-1).xerc;
 
+	-- Generate OV/OV32/SO when OE=1
         if v.multiply_pipeline(PIPELINE_DEPTH-1).valid = '1' then
             m_out.valid <= '1';
             m_out.write_reg_enable <= '1';
             m_out.rc <= v.multiply_pipeline(PIPELINE_DEPTH-1).rc;
+            m_out.write_xerc_enable <= v.multiply_pipeline(PIPELINE_DEPTH-1).oe;
+
+	    -- We must test oe because the RC update code in writeback
+	    -- will use the xerc value to set CR0:SO so we must not clobber
+	    -- xerc if OE wasn't set.
+	    --
+	    if v.multiply_pipeline(PIPELINE_DEPTH-1).oe = '1' then
+		m_out.xerc.ov <= ov;
+		m_out.xerc.ov32 <= ov;
+		m_out.xerc.so <= v.multiply_pipeline(PIPELINE_DEPTH-1).xerc.so or ov;
+	    end if;
         end if;
 
         rin <= v;

writeback.vhdl

@@ -62,6 +62,8 @@ begin
         variable w : std_ulogic_vector(0 downto 0);
         variable j : integer;
         variable k : unsigned(3 downto 0);
+	variable cf: std_ulogic_vector(3 downto 0);
+	variable xe: xer_common_t;
     begin
         x := "" & e_in.valid;
         y := "" & l_in.valid;
@@ -81,6 +83,11 @@ begin
         z := "" & (d_in.valid and d_in.rc);
         assert (to_integer(unsigned(w)) + to_integer(unsigned(x)) + to_integer(unsigned(y)) + to_integer(unsigned(z))) <= 1 severity failure;
 
+        x := "" & e_in.write_xerc_enable;
+        y := "" & m_in.write_xerc_enable;
+        z := "" & D_in.write_xerc_enable;
+        assert (to_integer(unsigned(x)) + to_integer(unsigned(y)) + to_integer(unsigned(z))) <= 1 severity failure;
+
         w_out <= WritebackToRegisterFileInit;
         c_out <= WritebackToCrFileInit;
 
@@ -96,12 +103,12 @@ begin
         partial_write <= '0';
         sign_extend <= '0';
         second_word <= '0';
-        data_in <= e_in.write_data;
+	xe := e_in.xerc;
 
         if e_in.write_enable = '1' then
             w_out.write_reg <= e_in.write_reg;
-            data_in <= e_in.write_data;
             w_out.write_enable <= '1';
+	    data_in <= e_in.write_data;
             data_len <= unsigned(e_in.write_len);
             sign_extend <= e_in.sign_extend;
             rc <= e_in.rc;
@@ -113,6 +120,11 @@ begin
             c_out.write_cr_data <= e_in.write_cr_data;
         end if;
 
+	if e_in.write_xerc_enable = '1' then
+            c_out.write_xerc_enable <= '1';
+            c_out.write_xerc_data <= e_in.xerc;
+	end if;
+
 	if l_in.write_enable = '1' then
             w_out.write_reg <= l_in.write_reg;
             data_in <= l_in.write_data;
@@ -127,6 +139,7 @@ begin
             if l_in.valid = '0' and (data_len + byte_offset > 8) then
                 partial_write <= '1';
             end if;
+	    xe := l_in.xerc;
         end if;
 
         if m_in.write_reg_enable = '1' then
@@ -134,6 +147,12 @@ begin
             w_out.write_reg <= m_in.write_reg_nr;
             data_in <= m_in.write_reg_data;
             rc <= m_in.rc;
+	    xe := m_in.xerc;
+        end if;
+
+	if m_in.write_xerc_enable = '1' then
+            c_out.write_xerc_enable <= '1';
+            c_out.write_xerc_data <= m_in.xerc;
 	end if;
 
         if d_in.write_reg_enable = '1' then
@@ -141,6 +160,12 @@ begin
             w_out.write_reg <= d_in.write_reg_nr;
             data_in <= d_in.write_reg_data;
             rc <= d_in.rc;
+	    xe := d_in.xerc;
+        end if;
+
+	if d_in.write_xerc_enable = '1' then
+            c_out.write_xerc_enable <= '1';
+            c_out.write_xerc_data <= d_in.xerc;
 	end if;
 
         -- shift and byte-reverse data bytes
@@ -193,17 +218,15 @@ begin
         -- deliver to regfile
         w_out.write_data <= data_trimmed;
 
-        -- test value against 0 and set CR0 if requested
+        -- Perform CR0 update for RC forms
         if rc = '1' then
             c_out.write_cr_enable <= '1';
             c_out.write_cr_mask <= num_to_fxm(0);
-            if negative = '1' then
-                c_out.write_cr_data <= x"80000000";
-            elsif zero = '0' then
-                c_out.write_cr_data <= x"40000000";
-            else
-                c_out.write_cr_data <= x"20000000";
-            end if;
+	    cf(3) := negative;
+	    cf(2) := not negative and not zero;
+	    cf(1) := zero;
+	    cf(0) := xe.so;
+	    c_out.write_cr_data(31 downto 28) <= cf;
         end if;
     end process;
 end;