|
|
|
library ieee;
|
|
|
|
use ieee.std_logic_1164.all;
|
|
|
|
use ieee.numeric_std.all;
|
|
|
|
|
|
|
|
library work;
|
|
|
|
use work.decode_types.all;
|
|
|
|
use work.common.all;
|
|
|
|
|
|
|
|
-- 2 cycle LSU
|
|
|
|
-- We calculate the address in the first cycle
|
|
|
|
|
|
|
|
entity loadstore1 is
|
|
|
|
generic (
|
|
|
|
-- Non-zero to enable log data collection
|
|
|
|
LOG_LENGTH : natural := 0
|
|
|
|
);
|
|
|
|
port (
|
|
|
|
clk : in std_ulogic;
|
|
|
|
rst : in std_ulogic;
|
|
|
|
|
|
|
|
l_in : in Execute1ToLoadstore1Type;
|
|
|
|
e_out : out Loadstore1ToExecute1Type;
|
|
|
|
l_out : out Loadstore1ToWritebackType;
|
|
|
|
|
|
|
|
d_out : out Loadstore1ToDcacheType;
|
|
|
|
d_in : in DcacheToLoadstore1Type;
|
|
|
|
|
|
|
|
m_out : out Loadstore1ToMmuType;
|
|
|
|
m_in : in MmuToLoadstore1Type;
|
|
|
|
|
|
|
|
dc_stall : in std_ulogic;
|
|
|
|
|
|
|
|
log_out : out std_ulogic_vector(9 downto 0)
|
|
|
|
);
|
|
|
|
end loadstore1;
|
|
|
|
|
|
|
|
-- Note, we don't currently use the stall output from the dcache because
|
|
|
|
-- we know it can take two requests without stalling when idle, we are
|
|
|
|
-- its only user, and we know it never stalls when idle.
|
|
|
|
|
|
|
|
architecture behave of loadstore1 is
|
|
|
|
|
|
|
|
-- State machine for unaligned loads/stores
|
|
|
|
type state_t is (IDLE, -- ready for instruction
|
|
|
|
SECOND_REQ, -- send 2nd request of unaligned xfer
|
|
|
|
ACK_WAIT, -- waiting for ack from dcache
|
|
|
|
MMU_LOOKUP, -- waiting for MMU to look up translation
|
|
|
|
TLBIE_WAIT, -- waiting for MMU to finish doing a tlbie
|
|
|
|
COMPLETE -- extra cycle to complete an operation
|
|
|
|
);
|
|
|
|
|
|
|
|
type reg_stage_t is record
|
|
|
|
-- latch most of the input request
|
|
|
|
load : std_ulogic;
|
|
|
|
tlbie : std_ulogic;
|
|
|
|
dcbz : std_ulogic;
|
|
|
|
mfspr : std_ulogic;
|
|
|
|
addr : std_ulogic_vector(63 downto 0);
|
|
|
|
store_data : std_ulogic_vector(63 downto 0);
|
|
|
|
load_data : std_ulogic_vector(63 downto 0);
|
|
|
|
write_reg : gpr_index_t;
|
|
|
|
length : std_ulogic_vector(3 downto 0);
|
|
|
|
byte_reverse : std_ulogic;
|
|
|
|
sign_extend : std_ulogic;
|
|
|
|
update : std_ulogic;
|
|
|
|
update_reg : gpr_index_t;
|
|
|
|
xerc : xer_common_t;
|
|
|
|
reserve : std_ulogic;
|
|
|
|
rc : std_ulogic;
|
|
|
|
nc : std_ulogic; -- non-cacheable access
|
|
|
|
virt_mode : std_ulogic;
|
|
|
|
priv_mode : std_ulogic;
|
|
|
|
state : state_t;
|
|
|
|
dwords_done : std_ulogic;
|
|
|
|
last_dword : std_ulogic;
|
|
|
|
first_bytes : std_ulogic_vector(7 downto 0);
|
|
|
|
second_bytes : std_ulogic_vector(7 downto 0);
|
|
|
|
dar : std_ulogic_vector(63 downto 0);
|
|
|
|
dsisr : std_ulogic_vector(31 downto 0);
|
Add TLB to icache
This adds a direct-mapped TLB to the icache, with 64 entries by default.
Execute1 now sends a "virt_mode" signal from MSR[IR] to fetch1 along
with redirects to indicate whether instruction addresses should be
translated through the TLB, and fetch1 sends that on to icache.
Similarly a "priv_mode" signal is sent to indicate the privilege
mode for instruction fetches. This means that changes to MSR[IR]
or MSR[PR] don't take effect until the next redirect, meaning an
isync, rfid, branch, etc.
The icache uses a hash of the effective address (i.e. next instruction
address) to index the TLB. The hash is an XOR of three fields of the
address; with a 64-entry TLB, the fields are bits 12--17, 18--23 and
24--29 of the address. TLB invalidations simply invalidate the
indexed TLB entry without checking the contents.
If the icache detects a TLB miss with virt_mode=1, it will send a
fetch_failed indication through fetch2 to decode1, which will turn it
into a special OP_FETCH_FAILED opcode with unit=LDST. That will get
sent down to loadstore1 which will currently just raise a Instruction
Storage Interrupt (0x400) exception.
One bit in the PTE obtained from the TLB is used to check whether an
instruction access is allowed -- the privilege bit (bit 3). If bit 3
is 1 and priv_mode=0, then a fetch_failed indication is sent down to
fetch2 and to decode1, which generates an OP_FETCH_FAILED. Any PTEs
with PTE bit 0 (EAA[3]) clear or bit 8 (R) clear should not be put
into the iTLB since such PTEs would not allow execution by any
context.
Tlbie operations get sent from mmu to icache over a new connection.
Unfortunately the privileged instruction tests are broken for now.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
|
|
|
instr_fault : std_ulogic;
|
|
|
|
align_intr : std_ulogic;
|
|
|
|
sprval : std_ulogic_vector(63 downto 0);
|
|
|
|
busy : std_ulogic;
|
|
|
|
wait_dcache : std_ulogic;
|
|
|
|
wait_mmu : std_ulogic;
|
|
|
|
do_update : std_ulogic;
|
|
|
|
extra_cycle : std_ulogic;
|
|
|
|
mode_32bit : std_ulogic;
|
|
|
|
end record;
|
|
|
|
|
|
|
|
type byte_sel_t is array(0 to 7) of std_ulogic;
|
|
|
|
subtype byte_trim_t is std_ulogic_vector(1 downto 0);
|
|
|
|
type trim_ctl_t is array(0 to 7) of byte_trim_t;
|
|
|
|
|
|
|
|
signal r, rin : reg_stage_t;
|
|
|
|
signal lsu_sum : std_ulogic_vector(63 downto 0);
|
|
|
|
|
|
|
|
-- Generate byte enables from sizes
|
|
|
|
function length_to_sel(length : in std_logic_vector(3 downto 0)) return std_ulogic_vector is
|
|
|
|
begin
|
|
|
|
case length is
|
|
|
|
when "0001" =>
|
|
|
|
return "00000001";
|
|
|
|
when "0010" =>
|
|
|
|
return "00000011";
|
|
|
|
when "0100" =>
|
|
|
|
return "00001111";
|
|
|
|
when "1000" =>
|
|
|
|
return "11111111";
|
|
|
|
when others =>
|
|
|
|
return "00000000";
|
|
|
|
end case;
|
|
|
|
end function length_to_sel;
|
|
|
|
|
|
|
|
-- Calculate byte enables
|
|
|
|
-- This returns 16 bits, giving the select signals for two transfers,
|
|
|
|
-- to account for unaligned loads or stores
|
|
|
|
function xfer_data_sel(size : in std_logic_vector(3 downto 0);
|
|
|
|
address : in std_logic_vector(2 downto 0))
|
|
|
|
return std_ulogic_vector is
|
|
|
|
variable longsel : std_ulogic_vector(15 downto 0);
|
|
|
|
begin
|
|
|
|
longsel := "00000000" & length_to_sel(size);
|
|
|
|
return std_ulogic_vector(shift_left(unsigned(longsel),
|
|
|
|
to_integer(unsigned(address))));
|
|
|
|
end function xfer_data_sel;
|
|
|
|
|
|
|
|
begin
|
|
|
|
-- Calculate the address in the first cycle
|
|
|
|
lsu_sum <= std_ulogic_vector(unsigned(l_in.addr1) + unsigned(l_in.addr2)) when l_in.valid = '1' else (others => '0');
|
|
|
|
|
|
|
|
loadstore1_0: process(clk)
|
|
|
|
begin
|
|
|
|
if rising_edge(clk) then
|
|
|
|
if rst = '1' then
|
|
|
|
r.state <= IDLE;
|
|
|
|
r.busy <= '0';
|
|
|
|
r.do_update <= '0';
|
|
|
|
else
|
|
|
|
r <= rin;
|
|
|
|
end if;
|
|
|
|
end if;
|
|
|
|
end process;
|
|
|
|
|
|
|
|
loadstore1_1: process(all)
|
|
|
|
variable v : reg_stage_t;
|
|
|
|
variable brev_lenm1 : unsigned(2 downto 0);
|
|
|
|
variable byte_offset : unsigned(2 downto 0);
|
|
|
|
variable j : integer;
|
|
|
|
variable k : unsigned(2 downto 0);
|
|
|
|
variable kk : unsigned(3 downto 0);
|
|
|
|
variable long_sel : std_ulogic_vector(15 downto 0);
|
|
|
|
variable byte_sel : std_ulogic_vector(7 downto 0);
|
|
|
|
variable req : std_ulogic;
|
|
|
|
variable busy : std_ulogic;
|
|
|
|
variable addr : std_ulogic_vector(63 downto 0);
|
|
|
|
variable maddr : std_ulogic_vector(63 downto 0);
|
|
|
|
variable wdata : std_ulogic_vector(63 downto 0);
|
|
|
|
variable write_enable : std_ulogic;
|
|
|
|
variable do_update : std_ulogic;
|
|
|
|
variable done : std_ulogic;
|
|
|
|
variable data_permuted : std_ulogic_vector(63 downto 0);
|
|
|
|
variable data_trimmed : std_ulogic_vector(63 downto 0);
|
|
|
|
variable store_data : std_ulogic_vector(63 downto 0);
|
|
|
|
variable use_second : byte_sel_t;
|
|
|
|
variable trim_ctl : trim_ctl_t;
|
|
|
|
variable negative : std_ulogic;
|
|
|
|
variable sprn : std_ulogic_vector(9 downto 0);
|
|
|
|
variable exception : std_ulogic;
|
|
|
|
variable next_addr : std_ulogic_vector(63 downto 0);
|
|
|
|
variable mmureq : std_ulogic;
|
|
|
|
variable dsisr : std_ulogic_vector(31 downto 0);
|
MMU: Implement radix page table machinery
This adds the necessary machinery to the MMU for it to do radix page
table walks. The core elements are a shifter that can shift the
address right by between 0 and 47 bits, a mask generator that can
generate a mask of between 5 and 16 bits, a final mask generator,
and new states in the state machine.
(The final mask generator is used for transferring bits of the
original address into the resulting TLB entry when the leaf PTE
corresponds to a page size larger than 4kB.)
The hardware does not implement a partition table or a process table.
Software is expected to load the appropriate process table entry
into a new SPR called PGTBL0, SPR 720. The contents should be
formatted as described in Book III section 5.7.6.2 of the Power ISA
v3.0B. PGTBL0 is set to 0 on hard reset. At present, the top two bits
of the address (the quadrant) are ignored.
There is currently no caching of any step in the translation process
or of the final result, other than the entry created in the dTLB.
That entry is a 4k page entry even if the leaf PTE found in the walk
corresponds to a larger page size.
This implementation can handle almost any page table layout and any
page size. The RTS field (in PGTBL0) can have any value between 0
and 31, corresponding to a total address space size between 2^31
and 2^62 bytes. The RPDS field of PGTBL0 can be any value between
5 and 16, except that a value of 0 is taken to disable radix page
table walking (for use when one is using software loading of TLB
entries). The NLS field of the page directory entries can have any
value between 5 and 16. The minimum page size is 4kB, meaning that
the sum of RPDS and the NLS values of the PDEs found on the path to
a leaf PTE must be less than or equal to RTS + 31 - 12.
The PGTBL0 SPR is in the mmu module; thus this adds a path for
loadstore1 to read and write SPRs in mmu. This adds code in dcache
to service doubleword read requests from the MMU, as well as requests
to write dTLB entries.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
|
|
|
variable mmu_mtspr : std_ulogic;
|
Add TLB to icache
This adds a direct-mapped TLB to the icache, with 64 entries by default.
Execute1 now sends a "virt_mode" signal from MSR[IR] to fetch1 along
with redirects to indicate whether instruction addresses should be
translated through the TLB, and fetch1 sends that on to icache.
Similarly a "priv_mode" signal is sent to indicate the privilege
mode for instruction fetches. This means that changes to MSR[IR]
or MSR[PR] don't take effect until the next redirect, meaning an
isync, rfid, branch, etc.
The icache uses a hash of the effective address (i.e. next instruction
address) to index the TLB. The hash is an XOR of three fields of the
address; with a 64-entry TLB, the fields are bits 12--17, 18--23 and
24--29 of the address. TLB invalidations simply invalidate the
indexed TLB entry without checking the contents.
If the icache detects a TLB miss with virt_mode=1, it will send a
fetch_failed indication through fetch2 to decode1, which will turn it
into a special OP_FETCH_FAILED opcode with unit=LDST. That will get
sent down to loadstore1 which will currently just raise a Instruction
Storage Interrupt (0x400) exception.
One bit in the PTE obtained from the TLB is used to check whether an
instruction access is allowed -- the privilege bit (bit 3). If bit 3
is 1 and priv_mode=0, then a fetch_failed indication is sent down to
fetch2 and to decode1, which generates an OP_FETCH_FAILED. Any PTEs
with PTE bit 0 (EAA[3]) clear or bit 8 (R) clear should not be put
into the iTLB since such PTEs would not allow execution by any
context.
Tlbie operations get sent from mmu to icache over a new connection.
Unfortunately the privileged instruction tests are broken for now.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
|
|
|
variable itlb_fault : std_ulogic;
|
|
|
|
variable misaligned : std_ulogic;
|
|
|
|
begin
|
|
|
|
v := r;
|
|
|
|
req := '0';
|
|
|
|
v.mfspr := '0';
|
MMU: Implement radix page table machinery
This adds the necessary machinery to the MMU for it to do radix page
table walks. The core elements are a shifter that can shift the
address right by between 0 and 47 bits, a mask generator that can
generate a mask of between 5 and 16 bits, a final mask generator,
and new states in the state machine.
(The final mask generator is used for transferring bits of the
original address into the resulting TLB entry when the leaf PTE
corresponds to a page size larger than 4kB.)
The hardware does not implement a partition table or a process table.
Software is expected to load the appropriate process table entry
into a new SPR called PGTBL0, SPR 720. The contents should be
formatted as described in Book III section 5.7.6.2 of the Power ISA
v3.0B. PGTBL0 is set to 0 on hard reset. At present, the top two bits
of the address (the quadrant) are ignored.
There is currently no caching of any step in the translation process
or of the final result, other than the entry created in the dTLB.
That entry is a 4k page entry even if the leaf PTE found in the walk
corresponds to a larger page size.
This implementation can handle almost any page table layout and any
page size. The RTS field (in PGTBL0) can have any value between 0
and 31, corresponding to a total address space size between 2^31
and 2^62 bytes. The RPDS field of PGTBL0 can be any value between
5 and 16, except that a value of 0 is taken to disable radix page
table walking (for use when one is using software loading of TLB
entries). The NLS field of the page directory entries can have any
value between 5 and 16. The minimum page size is 4kB, meaning that
the sum of RPDS and the NLS values of the PDEs found on the path to
a leaf PTE must be less than or equal to RTS + 31 - 12.
The PGTBL0 SPR is in the mmu module; thus this adds a path for
loadstore1 to read and write SPRs in mmu. This adds code in dcache
to service doubleword read requests from the MMU, as well as requests
to write dTLB entries.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
|
|
|
mmu_mtspr := '0';
|
Add TLB to icache
This adds a direct-mapped TLB to the icache, with 64 entries by default.
Execute1 now sends a "virt_mode" signal from MSR[IR] to fetch1 along
with redirects to indicate whether instruction addresses should be
translated through the TLB, and fetch1 sends that on to icache.
Similarly a "priv_mode" signal is sent to indicate the privilege
mode for instruction fetches. This means that changes to MSR[IR]
or MSR[PR] don't take effect until the next redirect, meaning an
isync, rfid, branch, etc.
The icache uses a hash of the effective address (i.e. next instruction
address) to index the TLB. The hash is an XOR of three fields of the
address; with a 64-entry TLB, the fields are bits 12--17, 18--23 and
24--29 of the address. TLB invalidations simply invalidate the
indexed TLB entry without checking the contents.
If the icache detects a TLB miss with virt_mode=1, it will send a
fetch_failed indication through fetch2 to decode1, which will turn it
into a special OP_FETCH_FAILED opcode with unit=LDST. That will get
sent down to loadstore1 which will currently just raise a Instruction
Storage Interrupt (0x400) exception.
One bit in the PTE obtained from the TLB is used to check whether an
instruction access is allowed -- the privilege bit (bit 3). If bit 3
is 1 and priv_mode=0, then a fetch_failed indication is sent down to
fetch2 and to decode1, which generates an OP_FETCH_FAILED. Any PTEs
with PTE bit 0 (EAA[3]) clear or bit 8 (R) clear should not be put
into the iTLB since such PTEs would not allow execution by any
context.
Tlbie operations get sent from mmu to icache over a new connection.
Unfortunately the privileged instruction tests are broken for now.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
|
|
|
itlb_fault := '0';
|
|
|
|
sprn := std_ulogic_vector(to_unsigned(decode_spr_num(l_in.insn), 10));
|
|
|
|
dsisr := (others => '0');
|
|
|
|
mmureq := '0';
|
|
|
|
|
|
|
|
write_enable := '0';
|
|
|
|
|
|
|
|
do_update := r.do_update;
|
|
|
|
v.do_update := '0';
|
|
|
|
|
|
|
|
-- load data formatting
|
|
|
|
byte_offset := unsigned(r.addr(2 downto 0));
|
|
|
|
brev_lenm1 := "000";
|
|
|
|
if r.byte_reverse = '1' then
|
|
|
|
brev_lenm1 := unsigned(r.length(2 downto 0)) - 1;
|
|
|
|
end if;
|
|
|
|
|
|
|
|
-- shift and byte-reverse data bytes
|
|
|
|
for i in 0 to 7 loop
|
|
|
|
kk := ('0' & (to_unsigned(i, 3) xor brev_lenm1)) + ('0' & byte_offset);
|
|
|
|
use_second(i) := kk(3);
|
|
|
|
j := to_integer(kk(2 downto 0)) * 8;
|
|
|
|
data_permuted(i * 8 + 7 downto i * 8) := d_in.data(j + 7 downto j);
|
|
|
|
end loop;
|
|
|
|
|
|
|
|
-- Work out the sign bit for sign extension.
|
|
|
|
-- For unaligned loads crossing two dwords, the sign bit is in the
|
|
|
|
-- first dword for big-endian (byte_reverse = 1), or the second dword
|
|
|
|
-- for little-endian.
|
|
|
|
if r.dwords_done = '1' and r.byte_reverse = '1' then
|
|
|
|
negative := (r.length(3) and r.load_data(63)) or
|
|
|
|
(r.length(2) and r.load_data(31)) or
|
|
|
|
(r.length(1) and r.load_data(15)) or
|
|
|
|
(r.length(0) and r.load_data(7));
|
|
|
|
else
|
|
|
|
negative := (r.length(3) and data_permuted(63)) or
|
|
|
|
(r.length(2) and data_permuted(31)) or
|
|
|
|
(r.length(1) and data_permuted(15)) or
|
|
|
|
(r.length(0) and data_permuted(7));
|
|
|
|
end if;
|
|
|
|
|
|
|
|
-- trim and sign-extend
|
|
|
|
for i in 0 to 7 loop
|
|
|
|
if i < to_integer(unsigned(r.length)) then
|
|
|
|
if r.dwords_done = '1' then
|
|
|
|
trim_ctl(i) := '1' & not use_second(i);
|
|
|
|
else
|
|
|
|
trim_ctl(i) := "10";
|
|
|
|
end if;
|
|
|
|
else
|
|
|
|
trim_ctl(i) := '0' & (negative and r.sign_extend);
|
|
|
|
end if;
|
|
|
|
case trim_ctl(i) is
|
|
|
|
when "11" =>
|
|
|
|
data_trimmed(i * 8 + 7 downto i * 8) := r.load_data(i * 8 + 7 downto i * 8);
|
|
|
|
when "10" =>
|
|
|
|
data_trimmed(i * 8 + 7 downto i * 8) := data_permuted(i * 8 + 7 downto i * 8);
|
|
|
|
when "01" =>
|
|
|
|
data_trimmed(i * 8 + 7 downto i * 8) := x"FF";
|
|
|
|
when others =>
|
|
|
|
data_trimmed(i * 8 + 7 downto i * 8) := x"00";
|
|
|
|
end case;
|
|
|
|
end loop;
|
|
|
|
|
|
|
|
-- Byte reversing and rotating for stores
|
|
|
|
-- Done in the first cycle (when l_in.valid = 1)
|
|
|
|
store_data := r.store_data;
|
|
|
|
if l_in.valid = '1' then
|
|
|
|
byte_offset := unsigned(lsu_sum(2 downto 0));
|
|
|
|
brev_lenm1 := "000";
|
|
|
|
if l_in.byte_reverse = '1' then
|
|
|
|
brev_lenm1 := unsigned(l_in.length(2 downto 0)) - 1;
|
|
|
|
end if;
|
|
|
|
for i in 0 to 7 loop
|
|
|
|
k := (to_unsigned(i, 3) - byte_offset) xor brev_lenm1;
|
|
|
|
j := to_integer(k) * 8;
|
|
|
|
store_data(i * 8 + 7 downto i * 8) := l_in.data(j + 7 downto j);
|
|
|
|
end loop;
|
|
|
|
end if;
|
|
|
|
v.store_data := store_data;
|
|
|
|
|
|
|
|
-- compute (addr + 8) & ~7 for the second doubleword when unaligned
|
|
|
|
next_addr := std_ulogic_vector(unsigned(r.addr(63 downto 3)) + 1) & "000";
|
|
|
|
|
|
|
|
-- Busy calculation.
|
|
|
|
-- We need to minimize the delay from clock to busy valid because it
|
|
|
|
-- gates the start of execution of the next instruction.
|
|
|
|
busy := r.busy and not ((r.wait_dcache and d_in.valid) or (r.wait_mmu and m_in.done));
|
|
|
|
v.busy := busy;
|
|
|
|
|
|
|
|
done := '0';
|
|
|
|
if r.state /= IDLE and busy = '0' then
|
|
|
|
done := '1';
|
|
|
|
end if;
|
|
|
|
exception := '0';
|
|
|
|
|
|
|
|
if r.dwords_done = '1' or r.state = SECOND_REQ then
|
|
|
|
addr := next_addr;
|
|
|
|
byte_sel := r.second_bytes;
|
|
|
|
else
|
|
|
|
addr := r.addr;
|
|
|
|
byte_sel := r.first_bytes;
|
|
|
|
end if;
|
|
|
|
if r.mode_32bit = '1' then
|
|
|
|
addr(63 downto 32) := (others => '0');
|
|
|
|
end if;
|
|
|
|
maddr := addr;
|
|
|
|
|
|
|
|
case r.state is
|
|
|
|
when IDLE =>
|
|
|
|
|
|
|
|
when SECOND_REQ =>
|
|
|
|
req := '1';
|
|
|
|
v.state := ACK_WAIT;
|
|
|
|
v.last_dword := '0';
|
|
|
|
|
|
|
|
when ACK_WAIT =>
|
|
|
|
if d_in.error = '1' then
|
|
|
|
-- dcache will discard the second request if it
|
|
|
|
-- gets an error on the 1st of two requests
|
|
|
|
if d_in.cache_paradox = '1' then
|
|
|
|
-- signal an interrupt straight away
|
|
|
|
exception := '1';
|
|
|
|
dsisr(63 - 38) := not r.load;
|
|
|
|
-- XXX there is no architected bit for this
|
|
|
|
dsisr(63 - 35) := d_in.cache_paradox;
|
|
|
|
else
|
|
|
|
-- Look up the translation for TLB miss
|
|
|
|
-- and also for permission error and RC error
|
|
|
|
-- in case the PTE has been updated.
|
|
|
|
mmureq := '1';
|
|
|
|
v.state := MMU_LOOKUP;
|
|
|
|
end if;
|
|
|
|
end if;
|
|
|
|
if d_in.valid = '1' then
|
|
|
|
if r.last_dword = '0' then
|
|
|
|
v.dwords_done := '1';
|
|
|
|
v.last_dword := '1';
|
|
|
|
if r.load = '1' then
|
|
|
|
v.load_data := data_permuted;
|
|
|
|
end if;
|
|
|
|
else
|
|
|
|
write_enable := r.load;
|
|
|
|
if r.extra_cycle = '1' then
|
|
|
|
-- loads with rA update need an extra cycle
|
|
|
|
v.state := COMPLETE;
|
|
|
|
v.do_update := r.update;
|
|
|
|
else
|
|
|
|
-- stores write back rA update in this cycle
|
|
|
|
do_update := r.update;
|
|
|
|
end if;
|
|
|
|
v.busy := '0';
|
|
|
|
end if;
|
|
|
|
end if;
|
|
|
|
-- r.wait_dcache gets set one cycle after we come into ACK_WAIT state,
|
|
|
|
-- which is OK because the dcache always takes at least two cycles.
|
|
|
|
v.wait_dcache := r.last_dword and not r.extra_cycle;
|
|
|
|
|
|
|
|
when MMU_LOOKUP =>
|
|
|
|
if m_in.done = '1' then
|
|
|
|
if r.instr_fault = '0' then
|
|
|
|
-- retry the request now that the MMU has installed a TLB entry
|
|
|
|
req := '1';
|
|
|
|
if r.last_dword = '0' then
|
|
|
|
v.state := SECOND_REQ;
|
|
|
|
else
|
|
|
|
v.state := ACK_WAIT;
|
|
|
|
end if;
|
|
|
|
end if;
|
|
|
|
end if;
|
|
|
|
if m_in.err = '1' then
|
|
|
|
exception := '1';
|
|
|
|
dsisr(63 - 33) := m_in.invalid;
|
|
|
|
dsisr(63 - 36) := m_in.perm_error;
|
|
|
|
dsisr(63 - 38) := not r.load;
|
|
|
|
dsisr(63 - 44) := m_in.badtree;
|
|
|
|
dsisr(63 - 45) := m_in.rc_error;
|
|
|
|
end if;
|
|
|
|
|
|
|
|
when TLBIE_WAIT =>
|
|
|
|
|
|
|
|
when COMPLETE =>
|
|
|
|
exception := r.align_intr;
|
|
|
|
|
|
|
|
end case;
|
|
|
|
|
|
|
|
if done = '1' or exception = '1' then
|
|
|
|
v.state := IDLE;
|
|
|
|
v.busy := '0';
|
|
|
|
end if;
|
|
|
|
|
|
|
|
-- Note that l_in.valid is gated with busy inside execute1
|
|
|
|
if l_in.valid = '1' then
|
|
|
|
v.addr := lsu_sum;
|
|
|
|
v.mode_32bit := l_in.mode_32bit;
|
|
|
|
v.load := '0';
|
|
|
|
v.dcbz := '0';
|
|
|
|
v.tlbie := '0';
|
|
|
|
v.instr_fault := '0';
|
|
|
|
v.align_intr := '0';
|
|
|
|
v.dwords_done := '0';
|
|
|
|
v.last_dword := '1';
|
|
|
|
v.write_reg := l_in.write_reg;
|
|
|
|
v.length := l_in.length;
|
|
|
|
v.byte_reverse := l_in.byte_reverse;
|
|
|
|
v.sign_extend := l_in.sign_extend;
|
|
|
|
v.update := l_in.update;
|
|
|
|
v.update_reg := l_in.update_reg;
|
|
|
|
v.xerc := l_in.xerc;
|
|
|
|
v.reserve := l_in.reserve;
|
|
|
|
v.rc := l_in.rc;
|
|
|
|
v.nc := l_in.ci;
|
|
|
|
v.virt_mode := l_in.virt_mode;
|
|
|
|
v.priv_mode := l_in.priv_mode;
|
|
|
|
v.wait_dcache := '0';
|
|
|
|
v.wait_mmu := '0';
|
|
|
|
v.do_update := '0';
|
|
|
|
v.extra_cycle := '0';
|
|
|
|
|
|
|
|
addr := lsu_sum;
|
|
|
|
if l_in.mode_32bit = '1' then
|
|
|
|
addr(63 downto 32) := (others => '0');
|
|
|
|
end if;
|
|
|
|
maddr := l_in.addr2; -- address from RB for tlbie
|
|
|
|
|
|
|
|
-- XXX Temporary hack. Mark the op as non-cachable if the address
|
|
|
|
-- is the form 0xc------- for a real-mode access.
|
|
|
|
if lsu_sum(31 downto 28) = "1100" and l_in.virt_mode = '0' then
|
|
|
|
v.nc := '1';
|
|
|
|
end if;
|
|
|
|
|
|
|
|
-- Do length_to_sel and work out if we are doing 2 dwords
|
|
|
|
long_sel := xfer_data_sel(l_in.length, v.addr(2 downto 0));
|
|
|
|
byte_sel := long_sel(7 downto 0);
|
|
|
|
v.first_bytes := byte_sel;
|
|
|
|
v.second_bytes := long_sel(15 downto 8);
|
|
|
|
|
|
|
|
-- check alignment for larx/stcx
|
|
|
|
misaligned := or (std_ulogic_vector(unsigned(l_in.length(2 downto 0)) - 1) and addr(2 downto 0));
|
|
|
|
v.align_intr := l_in.reserve and misaligned;
|
|
|
|
|
|
|
|
case l_in.op is
|
|
|
|
when OP_STORE =>
|
|
|
|
req := '1';
|
|
|
|
when OP_LOAD =>
|
|
|
|
req := '1';
|
|
|
|
v.load := '1';
|
|
|
|
-- Allow an extra cycle for RA update on loads
|
|
|
|
v.extra_cycle := l_in.update;
|
|
|
|
when OP_DCBZ =>
|
|
|
|
v.align_intr := v.nc;
|
|
|
|
req := '1';
|
|
|
|
v.dcbz := '1';
|
|
|
|
when OP_TLBIE =>
|
|
|
|
mmureq := '1';
|
|
|
|
v.tlbie := '1';
|
|
|
|
v.state := TLBIE_WAIT;
|
|
|
|
v.wait_mmu := '1';
|
|
|
|
when OP_MFSPR =>
|
|
|
|
v.mfspr := '1';
|
|
|
|
-- partial decode on SPR number should be adequate given
|
|
|
|
-- the restricted set that get sent down this path
|
|
|
|
if sprn(9) = '0' and sprn(5) = '0' then
|
|
|
|
if sprn(0) = '0' then
|
|
|
|
v.sprval := x"00000000" & r.dsisr;
|
|
|
|
else
|
|
|
|
v.sprval := r.dar;
|
|
|
|
end if;
|
|
|
|
else
|
|
|
|
-- reading one of the SPRs in the MMU
|
|
|
|
v.sprval := m_in.sprval;
|
|
|
|
end if;
|
|
|
|
v.state := COMPLETE;
|
|
|
|
when OP_MTSPR =>
|
|
|
|
if sprn(9) = '0' and sprn(5) = '0' then
|
|
|
|
if sprn(0) = '0' then
|
|
|
|
v.dsisr := l_in.data(31 downto 0);
|
|
|
|
else
|
|
|
|
v.dar := l_in.data;
|
|
|
|
end if;
|
|
|
|
v.state := COMPLETE;
|
|
|
|
else
|
|
|
|
-- writing one of the SPRs in the MMU
|
|
|
|
mmu_mtspr := '1';
|
|
|
|
v.state := TLBIE_WAIT;
|
|
|
|
v.wait_mmu := '1';
|
|
|
|
end if;
|
|
|
|
when OP_FETCH_FAILED =>
|
|
|
|
-- send it to the MMU to do the radix walk
|
|
|
|
maddr := l_in.nia;
|
|
|
|
v.instr_fault := '1';
|
|
|
|
mmureq := '1';
|
|
|
|
v.state := MMU_LOOKUP;
|
|
|
|
v.wait_mmu := '1';
|
|
|
|
when others =>
|
|
|
|
assert false report "unknown op sent to loadstore1";
|
|
|
|
end case;
|
|
|
|
|
|
|
|
if req = '1' then
|
|
|
|
if v.align_intr = '1' then
|
|
|
|
v.state := COMPLETE;
|
|
|
|
elsif long_sel(15 downto 8) = "00000000" then
|
|
|
|
v.state := ACK_WAIT;
|
|
|
|
else
|
|
|
|
v.state := SECOND_REQ;
|
|
|
|
end if;
|
|
|
|
end if;
|
|
|
|
|
|
|
|
v.busy := req or mmureq or mmu_mtspr;
|
|
|
|
end if;
|
|
|
|
|
|
|
|
-- Update outputs to dcache
|
|
|
|
d_out.valid <= req and not v.align_intr;
|
|
|
|
d_out.load <= v.load;
|
|
|
|
d_out.dcbz <= v.dcbz;
|
|
|
|
d_out.nc <= v.nc;
|
|
|
|
d_out.reserve <= v.reserve;
|
|
|
|
d_out.addr <= addr;
|
|
|
|
d_out.data <= store_data;
|
|
|
|
d_out.byte_sel <= byte_sel;
|
|
|
|
d_out.virt_mode <= v.virt_mode;
|
|
|
|
d_out.priv_mode <= v.priv_mode;
|
|
|
|
|
|
|
|
-- Update outputs to MMU
|
|
|
|
m_out.valid <= mmureq;
|
|
|
|
m_out.iside <= v.instr_fault;
|
|
|
|
m_out.load <= r.load;
|
|
|
|
m_out.priv <= r.priv_mode;
|
|
|
|
m_out.tlbie <= v.tlbie;
|
MMU: Implement radix page table machinery
This adds the necessary machinery to the MMU for it to do radix page
table walks. The core elements are a shifter that can shift the
address right by between 0 and 47 bits, a mask generator that can
generate a mask of between 5 and 16 bits, a final mask generator,
and new states in the state machine.
(The final mask generator is used for transferring bits of the
original address into the resulting TLB entry when the leaf PTE
corresponds to a page size larger than 4kB.)
The hardware does not implement a partition table or a process table.
Software is expected to load the appropriate process table entry
into a new SPR called PGTBL0, SPR 720. The contents should be
formatted as described in Book III section 5.7.6.2 of the Power ISA
v3.0B. PGTBL0 is set to 0 on hard reset. At present, the top two bits
of the address (the quadrant) are ignored.
There is currently no caching of any step in the translation process
or of the final result, other than the entry created in the dTLB.
That entry is a 4k page entry even if the leaf PTE found in the walk
corresponds to a larger page size.
This implementation can handle almost any page table layout and any
page size. The RTS field (in PGTBL0) can have any value between 0
and 31, corresponding to a total address space size between 2^31
and 2^62 bytes. The RPDS field of PGTBL0 can be any value between
5 and 16, except that a value of 0 is taken to disable radix page
table walking (for use when one is using software loading of TLB
entries). The NLS field of the page directory entries can have any
value between 5 and 16. The minimum page size is 4kB, meaning that
the sum of RPDS and the NLS values of the PDEs found on the path to
a leaf PTE must be less than or equal to RTS + 31 - 12.
The PGTBL0 SPR is in the mmu module; thus this adds a path for
loadstore1 to read and write SPRs in mmu. This adds code in dcache
to service doubleword read requests from the MMU, as well as requests
to write dTLB entries.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
|
|
|
m_out.mtspr <= mmu_mtspr;
|
|
|
|
m_out.sprn <= sprn;
|
|
|
|
m_out.addr <= maddr;
|
|
|
|
m_out.slbia <= l_in.insn(7);
|
|
|
|
m_out.rs <= l_in.data;
|
|
|
|
|
|
|
|
-- Update outputs to writeback
|
|
|
|
-- Multiplex either cache data to the destination GPR or
|
|
|
|
-- the address for the rA update.
|
|
|
|
l_out.valid <= done;
|
|
|
|
if r.mfspr = '1' then
|
|
|
|
l_out.write_enable <= '1';
|
|
|
|
l_out.write_reg <= r.write_reg;
|
|
|
|
l_out.write_data <= r.sprval;
|
|
|
|
elsif do_update = '1' then
|
|
|
|
l_out.write_enable <= '1';
|
|
|
|
l_out.write_reg <= r.update_reg;
|
|
|
|
l_out.write_data <= r.addr;
|
|
|
|
else
|
|
|
|
l_out.write_enable <= write_enable;
|
|
|
|
l_out.write_reg <= r.write_reg;
|
|
|
|
l_out.write_data <= data_trimmed;
|
|
|
|
end if;
|
|
|
|
l_out.xerc <= r.xerc;
|
|
|
|
l_out.rc <= r.rc and done;
|
|
|
|
l_out.store_done <= d_in.store_done;
|
|
|
|
|
|
|
|
-- update exception info back to execute1
|
|
|
|
e_out.busy <= busy;
|
|
|
|
e_out.exception <= exception;
|
|
|
|
e_out.alignment <= r.align_intr;
|
Add TLB to icache
This adds a direct-mapped TLB to the icache, with 64 entries by default.
Execute1 now sends a "virt_mode" signal from MSR[IR] to fetch1 along
with redirects to indicate whether instruction addresses should be
translated through the TLB, and fetch1 sends that on to icache.
Similarly a "priv_mode" signal is sent to indicate the privilege
mode for instruction fetches. This means that changes to MSR[IR]
or MSR[PR] don't take effect until the next redirect, meaning an
isync, rfid, branch, etc.
The icache uses a hash of the effective address (i.e. next instruction
address) to index the TLB. The hash is an XOR of three fields of the
address; with a 64-entry TLB, the fields are bits 12--17, 18--23 and
24--29 of the address. TLB invalidations simply invalidate the
indexed TLB entry without checking the contents.
If the icache detects a TLB miss with virt_mode=1, it will send a
fetch_failed indication through fetch2 to decode1, which will turn it
into a special OP_FETCH_FAILED opcode with unit=LDST. That will get
sent down to loadstore1 which will currently just raise a Instruction
Storage Interrupt (0x400) exception.
One bit in the PTE obtained from the TLB is used to check whether an
instruction access is allowed -- the privilege bit (bit 3). If bit 3
is 1 and priv_mode=0, then a fetch_failed indication is sent down to
fetch2 and to decode1, which generates an OP_FETCH_FAILED. Any PTEs
with PTE bit 0 (EAA[3]) clear or bit 8 (R) clear should not be put
into the iTLB since such PTEs would not allow execution by any
context.
Tlbie operations get sent from mmu to icache over a new connection.
Unfortunately the privileged instruction tests are broken for now.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
|
|
|
e_out.instr_fault <= r.instr_fault;
|
|
|
|
e_out.invalid <= m_in.invalid;
|
|
|
|
e_out.badtree <= m_in.badtree;
|
|
|
|
e_out.perm_error <= m_in.perm_error;
|
|
|
|
e_out.rc_error <= m_in.rc_error;
|
|
|
|
e_out.segment_fault <= m_in.segerr;
|
Add TLB to icache
This adds a direct-mapped TLB to the icache, with 64 entries by default.
Execute1 now sends a "virt_mode" signal from MSR[IR] to fetch1 along
with redirects to indicate whether instruction addresses should be
translated through the TLB, and fetch1 sends that on to icache.
Similarly a "priv_mode" signal is sent to indicate the privilege
mode for instruction fetches. This means that changes to MSR[IR]
or MSR[PR] don't take effect until the next redirect, meaning an
isync, rfid, branch, etc.
The icache uses a hash of the effective address (i.e. next instruction
address) to index the TLB. The hash is an XOR of three fields of the
address; with a 64-entry TLB, the fields are bits 12--17, 18--23 and
24--29 of the address. TLB invalidations simply invalidate the
indexed TLB entry without checking the contents.
If the icache detects a TLB miss with virt_mode=1, it will send a
fetch_failed indication through fetch2 to decode1, which will turn it
into a special OP_FETCH_FAILED opcode with unit=LDST. That will get
sent down to loadstore1 which will currently just raise a Instruction
Storage Interrupt (0x400) exception.
One bit in the PTE obtained from the TLB is used to check whether an
instruction access is allowed -- the privilege bit (bit 3). If bit 3
is 1 and priv_mode=0, then a fetch_failed indication is sent down to
fetch2 and to decode1, which generates an OP_FETCH_FAILED. Any PTEs
with PTE bit 0 (EAA[3]) clear or bit 8 (R) clear should not be put
into the iTLB since such PTEs would not allow execution by any
context.
Tlbie operations get sent from mmu to icache over a new connection.
Unfortunately the privileged instruction tests are broken for now.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
|
|
|
if exception = '1' and r.instr_fault = '0' then
|
|
|
|
v.dar := addr;
|
|
|
|
if m_in.segerr = '0' and r.align_intr = '0' then
|
|
|
|
v.dsisr := dsisr;
|
|
|
|
end if;
|
|
|
|
end if;
|
|
|
|
|
|
|
|
-- Update registers
|
|
|
|
rin <= v;
|
|
|
|
|
|
|
|
end process;
|
|
|
|
|
|
|
|
l1_log: if LOG_LENGTH > 0 generate
|
|
|
|
signal log_data : std_ulogic_vector(9 downto 0);
|
|
|
|
begin
|
|
|
|
ls1_log: process(clk)
|
|
|
|
begin
|
|
|
|
if rising_edge(clk) then
|
|
|
|
log_data <= e_out.busy &
|
|
|
|
e_out.exception &
|
|
|
|
l_out.valid &
|
|
|
|
m_out.valid &
|
|
|
|
d_out.valid &
|
|
|
|
m_in.done &
|
|
|
|
r.dwords_done &
|
|
|
|
std_ulogic_vector(to_unsigned(state_t'pos(r.state), 3));
|
|
|
|
end if;
|
|
|
|
end process;
|
|
|
|
log_out <= log_data;
|
|
|
|
end generate;
|
|
|
|
|
|
|
|
end;
|