You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
microwatt/dcache.vhdl

1703 lines
65 KiB
VHDL

--
-- Set associative dcache write-through
--
--
library ieee;
use ieee.std_logic_1164.all;
use ieee.numeric_std.all;
library work;
use work.utils.all;
use work.common.all;
use work.helpers.all;
use work.wishbone_types.all;
entity dcache is
generic (
-- Line size in bytes
LINE_SIZE : positive := 64;
-- Number of lines in a set
NUM_LINES : positive := 32;
-- Number of ways
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
NUM_WAYS : positive := 4;
-- L1 DTLB entries per set
TLB_SET_SIZE : positive := 64;
-- L1 DTLB number of sets
TLB_NUM_WAYS : positive := 2;
-- L1 DTLB log_2(page_size)
TLB_LG_PGSZ : positive := 12;
-- Non-zero to enable log data collection
LOG_LENGTH : natural := 0
);
port (
clk : in std_ulogic;
rst : in std_ulogic;
d_in : in Loadstore1ToDcacheType;
d_out : out DcacheToLoadstore1Type;
m_in : in MmuToDcacheType;
m_out : out DcacheToMmuType;
snoop_in : in wishbone_master_out := wishbone_master_out_init;
stall_out : out std_ulogic;
wishbone_out : out wishbone_master_out;
wishbone_in : in wishbone_slave_out;
events : out DcacheEventType;
log_out : out std_ulogic_vector(19 downto 0)
);
end entity dcache;
architecture rtl of dcache is
-- BRAM organisation: We never access more than wishbone_data_bits at
-- a time so to save resources we make the array only that wide, and
-- use consecutive indices to make a cache "line"
--
-- ROW_SIZE is the width in bytes of the BRAM (based on WB, so 64-bits)
constant ROW_SIZE : natural := wishbone_data_bits / 8;
-- ROW_PER_LINE is the number of row (wishbone transactions) in a line
constant ROW_PER_LINE : natural := LINE_SIZE / ROW_SIZE;
-- BRAM_ROWS is the number of rows in BRAM needed to represent the full
-- dcache
constant BRAM_ROWS : natural := NUM_LINES * ROW_PER_LINE;
-- Bit fields counts in the address
-- ROW_BITS is the number of bits to select a row
constant ROW_BITS : natural := log2(BRAM_ROWS);
-- ROW_LINEBITS is the number of bits to select a row within a line
constant ROW_LINEBITS : natural := log2(ROW_PER_LINE);
-- LINE_OFF_BITS is the number of bits for the offset in a cache line
constant LINE_OFF_BITS : natural := log2(LINE_SIZE);
-- ROW_OFF_BITS is the number of bits for the offset in a row
constant ROW_OFF_BITS : natural := log2(ROW_SIZE);
-- INDEX_BITS is the number if bits to select a cache line
constant INDEX_BITS : natural := log2(NUM_LINES);
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
-- SET_SIZE_BITS is the log base 2 of the set size
constant SET_SIZE_BITS : natural := LINE_OFF_BITS + INDEX_BITS;
-- TAG_BITS is the number of bits of the tag part of the address
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
constant TAG_BITS : natural := REAL_ADDR_BITS - SET_SIZE_BITS;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
-- TAG_WIDTH is the width in bits of each way of the tag RAM
constant TAG_WIDTH : natural := TAG_BITS + 7 - ((TAG_BITS + 7) mod 8);
-- WAY_BITS is the number of bits to select a way
constant WAY_BITS : natural := log2(NUM_WAYS);
-- Example of layout for 32 lines of 64 bytes:
--
-- .. tag |index| line |
-- .. | row | |
-- .. | |---| | ROW_LINEBITS (3)
-- .. | |--- - --| LINE_OFF_BITS (6)
-- .. | |- --| ROW_OFF_BITS (3)
-- .. |----- ---| | ROW_BITS (8)
-- .. |-----| | INDEX_BITS (5)
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
-- .. --------| | TAG_BITS (45)
subtype row_t is unsigned(ROW_BITS-1 downto 0);
subtype index_t is unsigned(INDEX_BITS-1 downto 0);
subtype way_t is unsigned(WAY_BITS-1 downto 0);
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
subtype row_in_line_t is unsigned(ROW_LINEBITS-1 downto 0);
-- The cache data BRAM organized as described above for each way
subtype cache_row_t is std_ulogic_vector(wishbone_data_bits-1 downto 0);
-- The cache tags LUTRAM has a row per set. Vivado is a pain and will
-- not handle a clean (commented) definition of the cache tags as a 3d
-- memory. For now, work around it by putting all the tags
subtype cache_tag_t is std_logic_vector(TAG_BITS-1 downto 0);
-- type cache_tags_set_t is array(way_t) of cache_tag_t;
-- type cache_tags_array_t is array(0 to NUM_LINES-1) of cache_tags_set_t;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
constant TAG_RAM_WIDTH : natural := TAG_WIDTH * NUM_WAYS;
subtype cache_tags_set_t is std_logic_vector(TAG_RAM_WIDTH-1 downto 0);
type cache_tags_array_t is array(0 to NUM_LINES-1) of cache_tags_set_t;
-- The cache valid bits
subtype cache_way_valids_t is std_ulogic_vector(NUM_WAYS-1 downto 0);
type cache_valids_t is array(0 to NUM_LINES-1) of cache_way_valids_t;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
type row_per_line_valid_t is array(0 to ROW_PER_LINE - 1) of std_ulogic;
-- Storage. Hopefully implemented in LUTs
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
signal cache_tags : cache_tags_array_t;
signal cache_tag_set : cache_tags_set_t;
signal cache_valids : cache_valids_t;
attribute ram_style : string;
attribute ram_style of cache_tags : signal is "distributed";
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
-- L1 TLB.
constant TLB_SET_BITS : natural := log2(TLB_SET_SIZE);
constant TLB_WAY_BITS : natural := log2(TLB_NUM_WAYS);
constant TLB_EA_TAG_BITS : natural := 64 - (TLB_LG_PGSZ + TLB_SET_BITS);
constant TLB_TAG_WAY_BITS : natural := TLB_NUM_WAYS * TLB_EA_TAG_BITS;
constant TLB_PTE_BITS : natural := 64;
constant TLB_PTE_WAY_BITS : natural := TLB_NUM_WAYS * TLB_PTE_BITS;
subtype tlb_way_t is integer range 0 to TLB_NUM_WAYS - 1;
subtype tlb_way_sig_t is unsigned(TLB_WAY_BITS-1 downto 0);
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
subtype tlb_index_t is integer range 0 to TLB_SET_SIZE - 1;
subtype tlb_index_sig_t is unsigned(TLB_SET_BITS-1 downto 0);
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
subtype tlb_way_valids_t is std_ulogic_vector(TLB_NUM_WAYS-1 downto 0);
type tlb_valids_t is array(tlb_index_t) of tlb_way_valids_t;
subtype tlb_tag_t is std_ulogic_vector(TLB_EA_TAG_BITS - 1 downto 0);
subtype tlb_way_tags_t is std_ulogic_vector(TLB_TAG_WAY_BITS-1 downto 0);
type tlb_tags_t is array(tlb_index_t) of tlb_way_tags_t;
subtype tlb_pte_t is std_ulogic_vector(TLB_PTE_BITS - 1 downto 0);
subtype tlb_way_ptes_t is std_ulogic_vector(TLB_PTE_WAY_BITS-1 downto 0);
type tlb_ptes_t is array(tlb_index_t) of tlb_way_ptes_t;
type hit_way_set_t is array(tlb_way_t) of way_t;
signal dtlb_valids : tlb_valids_t;
signal dtlb_tags : tlb_tags_t;
signal dtlb_ptes : tlb_ptes_t;
attribute ram_style of dtlb_tags : signal is "distributed";
attribute ram_style of dtlb_ptes : signal is "distributed";
-- Record for storing permission, attribute, etc. bits from a PTE
type perm_attr_t is record
reference : std_ulogic;
changed : std_ulogic;
nocache : std_ulogic;
priv : std_ulogic;
rd_perm : std_ulogic;
wr_perm : std_ulogic;
end record;
function extract_perm_attr(pte : std_ulogic_vector(TLB_PTE_BITS - 1 downto 0)) return perm_attr_t is
variable pa : perm_attr_t;
begin
pa.reference := pte(8);
pa.changed := pte(7);
pa.nocache := pte(5);
pa.priv := pte(3);
pa.rd_perm := pte(2);
pa.wr_perm := pte(1);
return pa;
end;
constant real_mode_perm_attr : perm_attr_t := (nocache => '0', others => '1');
-- Type of operation on a "valid" input
type op_t is (OP_NONE,
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
OP_BAD, -- NC cache hit, TLB miss, prot/RC failure
OP_STCX_FAIL, -- conditional store w/o reservation
OP_LOAD_HIT, -- Cache hit on load
OP_LOAD_MISS, -- Load missing cache
OP_LOAD_NC, -- Non-cachable load
OP_STORE_HIT, -- Store hitting cache
OP_STORE_MISS); -- Store missing cache
-- Cache state machine
type state_t is (IDLE, -- Normal load hit processing
RELOAD_WAIT_ACK, -- Cache reload wait ack
STORE_WAIT_ACK, -- Store wait ack
NC_LOAD_WAIT_ACK);-- Non-cachable load wait ack
--
-- Dcache operations:
--
-- In order to make timing, we use the BRAMs with an output buffer,
-- which means that the BRAM output is delayed by an extra cycle.
--
-- Thus, the dcache has a 2-stage internal pipeline for cache hits
-- with no stalls. Stores also complete in 2 cycles in most
-- circumstances.
--
-- A request proceeds through the pipeline as follows.
--
-- Cycle 0: Request is received from loadstore or mmu if either
-- d_in.valid or m_in.valid is 1 (not both). In this cycle portions
-- of the address are presented to the TLB tag RAM and data RAM
-- and the cache tag RAM and data RAM.
--
-- Clock edge between cycle 0 and cycle 1:
-- Request is stored in r0 (assuming r0_full was 0). TLB tag and
-- data RAMs are read, and the cache tag RAM is read. (Cache data
-- comes out a cycle later due to its output register, giving the
-- whole of cycle 1 to read the cache data RAM.)
--
-- Cycle 1: TLB and cache tag matching is done, the real address
-- (RA) for the access is calculated, and the type of operation is
-- determined (the OP_* values above). This gives the TLB way for
-- a TLB hit, and the cache way for a hit or the way to replace
-- for a load miss.
--
-- Clock edge between cycle 1 and cycle 2:
-- Request is stored in r1 (assuming r1.full was 0)
-- The state machine transitions out of IDLE state for a load miss,
-- a store, a dcbz, or a non-cacheable load. r1.full is set to 1
-- for a load miss, dcbz or non-cacheable load but not a store.
--
-- Cycle 2: Completion signals are asserted for a load hit,
-- a store (excluding dcbz), a TLB operation, a conditional
-- store which failed due to no matching reservation, or an error
-- (cache hit on non-cacheable operation, TLB miss, or protection
-- fault).
--
-- For a load miss, store, or dcbz, the state machine initiates
-- a wishbone cycle, which takes at least 2 cycles. For a store,
-- if another store comes in with the same cache tag (therefore
-- in the same 4k page), it can be added on to the existing cycle,
-- subject to some constraints.
-- While r1.full = 1, no new requests can go from r0 to r1, but
-- requests can come in to r0 and be satisfied if they are
-- cacheable load hits or stores with the same cache tag.
--
-- Writing to the cache data RAM is done at the clock edge
-- at the end of cycle 2 for a store hit (excluding dcbz).
-- Stores that miss are not written to the cache data RAM
-- but just stored through to memory.
-- Dcbz is done like a cache miss, but the wishbone cycle
-- is a write rather than a read, and zeroes are written to
-- the cache data RAM. Thus dcbz will allocate the line in
-- the cache as well as zeroing memory.
--
-- Since stores are written to the cache data RAM at the end of
-- cycle 2, and loads can come in and hit on the data just stored,
-- there is a two-stage bypass from store data to load data to
-- make sure that loads always see previously-stored data even
-- if it has not yet made it to the cache data RAM.
--
-- Load misses read the requested dword of the cache line first in
-- the memory read request and then cycle around through the other
-- dwords. The load is completed on the cycle after the requested
-- dword comes back from memory (using a forwarding path, rather
-- than going via the cache data RAM). We maintain an array of
-- valid bits per dword for the line being refilled so that
-- subsequent load requests to the same line can be completed as
-- soon as the necessary data comes in from memory, without
-- waiting for the whole line to be read.
-- Stage 0 register, basically contains just the latched request
type reg_stage_0_t is record
req : Loadstore1ToDcacheType;
tlbie : std_ulogic; -- indicates a tlbie request (from MMU)
doall : std_ulogic; -- with tlbie, indicates flush whole TLB
tlbld : std_ulogic; -- indicates a TLB load request (from MMU)
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_req : std_ulogic; -- indicates source of request
d_valid : std_ulogic; -- indicates req.data is valid now
end record;
signal r0 : reg_stage_0_t;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
signal r0_full : std_ulogic;
type mem_access_request_t is record
op : op_t;
valid : std_ulogic;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
dcbz : std_ulogic;
real_addr : real_addr_t;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
data : std_ulogic_vector(63 downto 0);
byte_sel : std_ulogic_vector(7 downto 0);
hit_way : way_t;
same_tag : std_ulogic;
mmu_req : std_ulogic;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
end record;
-- First stage register, contains state for stage 1 of load hits
-- and for the state machine used by all other operations
--
type reg_stage_1_t is record
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
-- Info about the request
full : std_ulogic; -- have uncompleted request
mmu_req : std_ulogic; -- request is from MMU
req : mem_access_request_t;
-- Cache hit state
hit_way : way_t;
hit_load_valid : std_ulogic;
hit_index : index_t;
cache_hit : std_ulogic;
-- TLB hit state
tlb_hit : std_ulogic;
tlb_hit_way : tlb_way_sig_t;
tlb_hit_index : tlb_index_sig_t;
-- data buffer for data forwarded from writes to reads
forward_data : std_ulogic_vector(63 downto 0);
forward_tag : cache_tag_t;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
forward_sel : std_ulogic_vector(7 downto 0);
forward_valid : std_ulogic;
forward_row : row_t;
data_out : std_ulogic_vector(63 downto 0);
-- Cache miss state (reload state machine)
state : state_t;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
dcbz : std_ulogic;
write_bram : std_ulogic;
write_tag : std_ulogic;
slow_valid : std_ulogic;
wb : wishbone_master_out;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
reload_tag : cache_tag_t;
store_way : way_t;
store_row : row_t;
store_index : index_t;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
end_row_ix : row_in_line_t;
rows_valid : row_per_line_valid_t;
acks_pending : unsigned(2 downto 0);
inc_acks : std_ulogic;
dec_acks : std_ulogic;
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
-- Signals to complete (possibly with error)
ls_valid : std_ulogic;
ls_error : std_ulogic;
mmu_done : std_ulogic;
mmu_error : std_ulogic;
cache_paradox : std_ulogic;
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
-- Signal to complete a failed stcx.
stcx_fail : std_ulogic;
end record;
signal r1 : reg_stage_1_t;
signal ev : DcacheEventType;
-- Reservation information
--
type reservation_t is record
valid : std_ulogic;
addr : std_ulogic_vector(63 downto LINE_OFF_BITS);
end record;
signal reservation : reservation_t;
-- Async signals on incoming request
signal req_index : index_t;
signal req_hit_way : way_t;
signal req_tag : cache_tag_t;
signal req_op : op_t;
signal req_data : std_ulogic_vector(63 downto 0);
signal req_same_tag : std_ulogic;
signal req_go : std_ulogic;
dcache: Trim one cycle from the load hit path Currently we don't get the result from a load that hits in the dcache until the fourth cycle after the instruction was presented to loadstore1. This trims this back to 3 cycles by taking the low order bits of the address generated in loadstore1 into dcache directly (not via the output register of loadstore1) and using them to address the read port of the dcache data RAM. We use the lower 12 address bits here in the expectation that any reasonable data cache design will have a set size of 4kB or less in order to avoid the aliasing problems that can arise with a virtually-indexed physically-tagged cache if the set size is greater than the smallest page size provided by the MMU. With this we can get rid of r2 and drive the signals going to writeback from r1, since the load hit data is now available one cycle earlier. We need a multiplexer on the read address of the data cache RAM in order to handle the second doubleword of an unaligned access. One small complication is that we now need an extra cycle in the case of an unaligned load which misses in the data cache and which reads the 2nd-last and last doublewords of a cache line. This is the reason for the PRE_NEXT_DWORD state; if we just go straight to NEXT_DWORD then we end up having the write of the last doubleword of the cache line and the read of that same doubleword occurring in the same cycle, which means we read stale data rather than the just-fetched data. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
signal early_req_row : row_t;
signal early_rd_valid : std_ulogic;
signal cancel_store : std_ulogic;
signal set_rsrv : std_ulogic;
signal clear_rsrv : std_ulogic;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
signal r0_valid : std_ulogic;
signal r0_stall : std_ulogic;
signal fwd_same_tag : std_ulogic;
signal use_forward_st : std_ulogic;
signal use_forward_rl : std_ulogic;
signal use_forward2 : std_ulogic;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
-- Cache RAM interface
type cache_ram_out_t is array(0 to NUM_WAYS-1) of cache_row_t;
signal cache_out : cache_ram_out_t;
signal ram_wr_data : cache_row_t;
signal ram_wr_select : std_ulogic_vector(ROW_SIZE - 1 downto 0);
-- PLRU output interface
type plru_out_t is array(0 to NUM_LINES-1) of std_ulogic_vector(WAY_BITS-1 downto 0);
signal plru_victim : plru_out_t;
signal replace_way : way_t;
-- Wishbone read/write/cache write formatting signals
signal bus_sel : std_ulogic_vector(7 downto 0);
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
-- TLB signals
signal tlb_tag_way : tlb_way_tags_t;
signal tlb_pte_way : tlb_way_ptes_t;
signal tlb_valid_way : tlb_way_valids_t;
signal tlb_req_index : tlb_index_sig_t;
signal tlb_read_valid : std_ulogic;
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
signal tlb_hit : std_ulogic;
signal tlb_hit_way : tlb_way_sig_t;
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
signal pte : tlb_pte_t;
signal ra : real_addr_t;
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
signal valid_ra : std_ulogic;
signal perm_attr : perm_attr_t;
signal rc_ok : std_ulogic;
signal perm_ok : std_ulogic;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
signal access_ok : std_ulogic;
signal tlb_miss : std_ulogic;
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
-- TLB PLRU output interface
type tlb_plru_out_t is array(tlb_index_t) of std_ulogic_vector(TLB_WAY_BITS-1 downto 0);
signal tlb_plru_victim : tlb_plru_out_t;
signal snoop_tag_set : cache_tags_set_t;
signal snoop_valid : std_ulogic;
signal snoop_wrtag : cache_tag_t;
signal snoop_index : index_t;
--
-- Helper functions to decode incoming requests
--
-- Return the cache line index (tag index) for an address
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
function get_index(addr: std_ulogic_vector) return index_t is
begin
return unsigned(addr(SET_SIZE_BITS - 1 downto LINE_OFF_BITS));
end;
-- Return the cache row index (data memory) for an address
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
function get_row(addr: std_ulogic_vector) return row_t is
begin
return unsigned(addr(SET_SIZE_BITS - 1 downto ROW_OFF_BITS));
end;
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
-- Return the index of a row within a line
function get_row_of_line(row: row_t) return row_in_line_t is
begin
return row(ROW_LINEBITS-1 downto 0);
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
end;
-- Returns whether this is the last row of a line
function is_last_row_wb_addr(addr: wishbone_addr_type; last: row_in_line_t) return boolean is
begin
return unsigned(addr(LINE_OFF_BITS - ROW_OFF_BITS - 1 downto 0)) = last;
end;
-- Returns whether this is the last row of a line
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
function is_last_row(row: row_t; last: row_in_line_t) return boolean is
begin
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
return get_row_of_line(row) = last;
end;
-- Return the address of the next row in the current cache line
function next_row_wb_addr(addr: wishbone_addr_type) return std_ulogic_vector is
variable row_idx : std_ulogic_vector(ROW_LINEBITS-1 downto 0);
variable result : wishbone_addr_type;
begin
-- Is there no simpler way in VHDL to generate that 3 bits adder ?
row_idx := addr(ROW_LINEBITS - 1 downto 0);
row_idx := std_ulogic_vector(unsigned(row_idx) + 1);
result := addr;
result(ROW_LINEBITS - 1 downto 0) := row_idx;
return result;
end;
-- Return the next row in the current cache line. We use a dedicated
-- function in order to limit the size of the generated adder to be
-- only the bits within a cache line (3 bits with default settings)
--
function next_row(row: row_t) return row_t is
variable row_v : std_ulogic_vector(ROW_BITS-1 downto 0);
variable row_idx : std_ulogic_vector(ROW_LINEBITS-1 downto 0);
variable result : std_ulogic_vector(ROW_BITS-1 downto 0);
begin
row_v := std_ulogic_vector(row);
row_idx := row_v(ROW_LINEBITS-1 downto 0);
row_v(ROW_LINEBITS-1 downto 0) := std_ulogic_vector(unsigned(row_idx) + 1);
return unsigned(row_v);
end;
-- Get the tag value from the address
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
function get_tag(addr: std_ulogic_vector) return cache_tag_t is
begin
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
return addr(REAL_ADDR_BITS - 1 downto SET_SIZE_BITS);
end;
-- Read a tag from a tag memory row
function read_tag(way: integer; tagset: cache_tags_set_t) return cache_tag_t is
begin
dcache: Reduce latencies and improve timing This implements various improvements to the dcache with the aim of making it go faster. - We can now execute operations that don't need to access main memory (cacheable loads that hit in the cache and TLB operations) as soon as any previous operation has completed, without waiting for the state machine to become idle. - Cache line refills start with the doubleword that is needed to satisfy the load that initiated them. - Cacheable loads that miss return their data and complete as soon as the requested doubleword comes back from memory; they don't wait for the refill to finish. - We now have per-doubleword valid bits for the cache line being refilled, meaning that if a load comes in for a line that is in the process of being refilled, we can return the data and complete it within a couple of cycles of the doubleword coming in from memory. - There is now a bypass path for data being written to the cache RAM so that we can do a store hit followed immediately by a load hit to the same doubleword. This also makes the data from a refill available to load hits one cycle earlier than it would be otherwise. - Stores complete in the cycle where their wishbone operation is initiated, without waiting for the wishbone cycle to complete. - During the wishbone cycle for a store, if another store comes in that is to the same page, and we don't have a stall from the wishbone, we can send out the write for the second store in the same wishbone cycle and without going through the IDLE state first. We limit it to 7 outstanding writes that have not yet been acknowledged. - The cache tag RAM is now read on a clock edge rather than being combinatorial for reading. Its width is rounded up to a multiple of 8 bits per way so that byte enables can be used for writing individual tags. - The cache tag RAM is now written a cycle later than previously, in order to ease timing. - Data for a store hit is now written one cycle later than previously. This eases timing since we don't have to get through the tag matching and on to the write enable within a single cycle. The 2-stage bypass path means we can still handle a load hit on either of the two cycles after the store and return the correct data. (A load hit 3 or more cycles later will get the correct data from the BRAM.) - Operations can sit in r0 while there is an uncompleted operation in r1. Once the operation in r1 is completed, the operation in r0 spends one cycle in r0 for TLB/cache tag lookup and then gets put into r1.req. This can happen before r1 gets to the IDLE state. Some operations can then be completed before r1 gets to the IDLE state - a load miss to the cache line being refilled, or a store to the same page as a previous store. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
return tagset(way * TAG_WIDTH + TAG_BITS - 1 downto way * TAG_WIDTH);
end;
dcache: Implement data TLB This adds a TLB to dcache, providing the ability to translate addresses for loads and stores. No protection mechanism has been implemented yet. The MSR_DR bit controls whether addresses are translated through the TLB. The TLB is a fixed-pagesize, set-associative cache. Currently the page size is 4kB and the TLB is 2-way set associative with 64 entries per set. This implements the tlbie instruction. RB bits 10 and 11 control whether the whole TLB is invalidated (if either bit is 1) or just a single entry corresponding to the effective page number in bits 12-63 of RB. As an extension until we get a hardware page table walk, a tlbie instruction with RB bits 9-11 set to 001 will load an entry into the TLB. The TLB entry value is in RS in the format of a radix PTE. Currently there is no proper handling of TLB misses. The load or store will not be performed but no interrupt is generated. In order to make timing at 100MHz on the Arty A7-100, we compare the real address from each way of the TLB with the tag from each way of the cache in parallel (requiring # TLB ways * # cache ways comparators). Then the result is selected based on which way hit in the TLB. That avoids a timing path going through the TLB EA comparators, the multiplexer that selects the RA, and the cache tag comparators. The hack where addresses of the form 0xc------- are marked as cache-inhibited is kept for now but restricted to real-mode accesses. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
-- Read a TLB tag from a TLB tag memory row
function read_tlb_tag(way: tlb_way_t; tags: tlb_way_tags_t) return tlb_tag_t is
variable j : integer;
begin
j := way * TLB_EA_TAG_BITS;
return tags(j + TLB_EA_TAG_BITS - 1 downto j);
end;
-- Write a TLB tag to a TLB tag memory row
procedure write_tlb_tag(way: tlb_way_t; tags: inout tlb_way_tags_t;
tag: tlb_tag_t) is
variable j : integer;
begin
j := way * TLB_EA_TAG_BITS;
tags(j + TLB_EA_TAG_BITS - 1 downto j) := tag;
end;
-- Read a PTE from a TLB PTE memory row
function read_tlb_pte(way: tlb_way_t; ptes: tlb_way_ptes_t) return tlb_pte_t is
variable j : integer;
begin
j := way * TLB_PTE_BITS;
return ptes(j + TLB_PTE_BITS - 1 downto j);
end;
procedure write_tlb_pte(way: tlb_way_t; ptes: inout tlb_way_ptes_t; newpte: tlb_pte_t) is
variable j : integer;