This eliminates a path where the inputs to r1.wb.dat and r1.wb.sel
depend on req_op, which depends on the TLB and cache hit detection.
In fact they only need to depend on the nature of the request in
r0.req (i.e. DCBZ, store, cacheable load, or non-cacheable load).
This sets them at the beginning of the code for IDLE state rather
than inside the req_op case statement.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This moves the incrementing or decrementing of r1.acks_pending
to the cycle after a strobe is output or an ack is seen on the
wishbone, and simplifies the logic that determines whether the
cycle is now complete. This means that the path from seeing
req_op equal to OP_STORE_HIT or OP_STORE_MISS to setting r1.state
and r1.cyc now just involves the stbs_done bit rather than a more
complex calculation involving the possibly incremented r1.acks_pending.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This makes d_out.valid and m_out.done come directly from registers in
order to improve timing. The inputs to the registers are set by the
same conditions that cause r1.hit_load_valid, r1.slow_valid,
r1.error_done and r1.stcx_fail to be set.
Note that the STORE_WAIT_ACK state doesn't test r1.mmu_req but assumes
that the request came from loadstore1. This is because we normally
have r1.full = 0 in this state, which means that r1.mmu_req can
change at any time.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This adds "if LOG_LENGTH > 0 generate" to the places in the core
where log output data is latched, so that when LOG_LENGTH = 0 we
don't create the logic to collect the data which won't be stored.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This improves timing by setting r1.wb.{adr,dat,sel} to the next
request when doing a write cycle on the wishbone before we know
whether the next request has a TLB and cache hit or not, i.e.
without depending on req_op. r1.wb.stb still depends on req_op.
This contains a workaround for what is probably a bug elsewhere,
in that changing r1.wb.sel unconditionally once we see stall=0
from the wishbone causes incorrect behaviour. Making it
conditional on there being a valid following request appears
to fix the problem.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This puts the inputs to the TLB PLRU through a register stage, so
the TLB PLRU update is done in the cycle after the TLB tag
matching rather than the same cycle. This improves timing.
The PLRU output is only used when writing the TLB in response to
a tlbwe request from the MMU, and that doesn't happen within one
cycle of a virtual-mode load or store, so the fact that the
tlb victim way information is delayed by one cycle doesn't
create any problems.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This does the PLRU update based on r1.cache_hit and r1.hit_way rather
than req_op and req_hit_way, which means there is now a register
between the TLB and cache tag lookup and the PLRU update, which should
help with timing.
The PLRU victim selection now becomes valid one cycle later, in the
cycle where r1.write_tag = 1. We now have replace_way coming from
the PLRU when r1.write_tag = 1 and from r1.store_way at other times,
and we use that instead of r1.store_way in situations where we need
it to be valid in the first cycle of the RELOAD_WAIT_ACK state.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This uses the machinery we already had for comparing the real address
of a new request with the tag of a previous request (r1.reload_tag)
to get better timing on comparing the address of a second store with
the one in progress. The comparison is now on the set size rather
than the page size, but since set size can't be larger than the page
size (and usually will equal the page size), that is OK.
The same comparison can also be used to tell when we can satisfy
a load miss during a cache line refill.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
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>
This logs 256 bits of data per cycle to a ring buffer in BRAM. The
data collected can be read out through 2 new SPRs or through the
debug interface.
The new SPRs are LOG_ADDR (724) and LOG_DATA (725). LOG_ADDR contains
the buffer write pointer in the upper 32 bits (in units of entries,
i.e. 32 bytes) and the read pointer in the lower 32 bits (in units of
doublewords, i.e. 8 bytes). Reading LOG_DATA gives the doubleword
from the buffer at the read pointer and increments the read pointer.
Setting bit 31 of LOG_ADDR inhibits the trace log system from writing
to the log buffer, so the contents are stable and can be read.
There are two new debug addresses which function similarly to the
LOG_ADDR and LOG_DATA SPRs. The log is frozen while either or both of
the LOG_ADDR SPR bit 31 or the debug LOG_ADDR register bit 31 are set.
The buffer defaults to 2048 entries, i.e. 64kB. The size is set by
the LOG_LENGTH generic on the core_debug module. Software can
determine the length of the buffer because the length is ORed into the
buffer write pointer in the upper 32 bits of LOG_ADDR. Hence the
length of the buffer can be calculated as 1 << (31 - clz(LOG_ADDR)).
There is a program to format the log entries in a somewhat readable
fashion in scripts/fmt_log/fmt_log.c. The log_entry struct in that
file describes the layout of the bits in the log entries.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
The global wr_en signal is causing Vivado to generate two TDP (True Dual Port)
block RAMs instead of one SDP (Simple Dual Port) for each cache way. Remove
it and instead apply a AND to the individual byte write enables.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
This fixes a bug where a store that hits in the dcache immediately
following a dcbz has its write to the cache RAM suppressed (but not
its write to memory). If a load to the same location comes along
before the cache line gets replaced, the load will return incorrect
data.
Fixes: 4db1676ef8 ("dcache: Don't assert on dcbz cache hit")
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Slbia (with IH=7) is used in the Linux kernel to flush the ERATs
(our iTLB/dTLB), so make it do that.
This moves the logic to work out whether to flush a single entry
or the whole TLB from dcache and icache into mmu. We now invalidate
all dTLB and iTLB entries when the AP (actual pagesize) field of
RB is non-zero on a tlbie[l], as well as when IS is non-zero.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This removes the hack where the tlbie instruction could be used to
load entries directly into the dTLB, because we don't report the
correct DSISR values for accesses that hit software-loaded dTLB
entries and have privilege or permission errors.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This is required by the architecture. It means that the error bits
reported in DSISR or SRR1 now come from the permission/RC check done
on the refetched PTE rather than the TLB entry. Unfortunately that
somewhat breaks the software-loaded TLB mode of operation in that
DSISR/SRR1 always report no PTE rather than permission error or
RC failure.
This also restructures the loadstore1 state machine a bit, combining
the FIRST_ACK_WAIT and LAST_ACK_WAIT states into a single state and
the MMU_LOOKUP_1ST and MMU_LOOKUP_LAST states likewise. We now have a
'dwords_done' bit to say whether the first transfer of two (for an
unaligned access) has been done.
The cache paradox error (where a non-cacheable access finds a hit in
the cache) is now the only cause of DSI from the dcache. This should
probably be a machine check rather than DSI in fact.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
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>
This adds a new module to implement an MMU. At the moment it doesn't
do very much. Tlbie instructions now get sent by loadstore1 to mmu,
which sends them to dcache, rather than loadstore1 sending them
directly to dcache. TLB misses from dcache now get sent by loadstore1
to mmu, which currently just returns an error. Loadstore1 then
generates a DSI in response to the error return from mmu.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This adds logic to the dcache to check the permissions encoded in
the PTE that it gets from the dTLB. The bits that are checked are:
R must be 1
C must be 1 for a store
EAA(0) - if this is 1, MSR[PR] must be 0
EAA(2) must be 1 for a store
EAA(1) | EAA(2) must be 1 for a load
In addition, ATT(0) is used to indicate a cache-inhibited access.
This now implements DSISR bits 36, 38 and 45.
(Bit numbers above correspond to the ISA, i.e. using big-endian
numbering.)
MSR[PR] is now conveyed to loadstore1 for use in permission checking.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This adds a path from loadstore1 back to execute1 for reporting
errors, and machinery in execute1 for generating data storage
interrupts at vector 0x300.
If dcache is given two requests in successive cycles and the
first encounters an error (e.g. a TLB miss), it will now cancel
the second request.
Loadstore1 now responds to errors reported by dcache by sending
an exception signal to execute1 and returning to the idle state.
Execute1 then writes SRR0 and SRR1 and jumps to the 0x300 Data
Storage Interrupt vector. DAR and DSISR are held in loadstore1.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
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>
We can hit the assert for req_op = OP_STORE_HIT and reloading in the
case of dcbz, since it looks like a store. Therefore we need to
exclude that case from the assert.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This adds logic to dcache and loadstore1 to implement dcbz. For now
it zeroes a single cache line (by default 64 bytes), not 128 bytes
like IBM Power processors do.
The dcbz operation is performed much like a load miss, except that
we are writing zeroes to memory instead of reading. As each ack
comes back, we write zeroes to the BRAM instead of data from memory.
In this way we zero the line in memory and also zero the line of
cache memory, establishing the line in the cache if it wasn't already
resident. If it was already resident then we overwrite the existing
line in the cache.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
So that the dcache could in future be used by an MMU, this moves
logic to do with data formatting, rA updates for update-form
instructions, and handling of unaligned loads and stores out of
dcache and into loadstore1. For now, dcache connects only to
loadstore1, and loadstore1 now has the connection to writeback.
Dcache generates a stall signal to loadstore1 which indicates that
the request presented in the current cycle was not accepted and
should be presented again. However, loadstore1 doesn't currently
use it because we know that we can never hit the circumstances
where it might be set.
For unaligned transfers, loadstore1 generates two requests to
dcache back-to-back, and then waits to see two acks back from
dcache (cycles where d_in.valid is true).
Loadstore1 now has a FSM for tracking how many acks we are
expecting from dcache and for doing the rA update cycles when
necessary. Handling for reservations and conditional stores is
still in dcache.
Loadstore1 now generates its own stall signal back to decode2,
so we no longer need the logic in execute1 that generated the stall
for the first two cycles.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Since we removed one cycle from the load hit case, we actually no
longer need the extra cycle provided by having the LOAD_UPDATE
state. Therefore this makes the load hit case in the IDLE and
NEXT_DWORD states go to LOAD_UPDATE2 rather than LOAD_UPDATE.
Then we remove LOAD_UPDATE and then rename LOAD_UPDATE2 to
LOAD_UPDATE.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
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>
This involves plumbing the (existing) 'reserve' and 'rc' bits in
the decode tables down to dcache, and 'rc' and 'store_done' bits
from dcache to writeback.
It turns out that we had 'RC' set in the 'rc' column for several
ordinary stores and for the attn instruction. This corrects them
to 'NONE', and sets the 'rc' column to 'ONE' for the conditional
stores.
In writeback we now have logic to set CR0 when the input from dcache
has rc = 1.
In dcache we have the reservation itself, which has a valid bit
and the address down to cache line granularity. We don't currently
store the reservation length. For a store conditional which fails,
we set a 'cancel_store' signal which inhibits the write to the
cache and prevents the state machine from starting a bus cycle or
going to the STORE_WAIT_ACK state. Instead we set r1.stcx_fail
which causes the instruction to complete in the next cycle with
rc=1 and store_done=0.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
For an unaligned load or store, we do the first doubleword (dword) of
the transfer as normal, but then go to a new NEXT_DWORD state of the
state machine to do the cache tag lookup for the second dword of the
transfer. From the NEXT_DWORD state we have much the same transitions
to other states as from the IDLE state (the transitions for OP_LOAD_HIT
are a bit different but almost identical for the other op values).
We now do the preparation of the data to be written in loadstore1,
that is, byte reversal if necessary and rotation by a number of
bytes based on the low 3 bits of the address. We do rotation not
shifting so we have the bytes that need to go into the second
doubleword in the right place in the low bytes of the data sent to
dcache. The rotation and byte reversal are done in a single step
with one multiplexer per byte by setting the select inputs for each
byte appropriately.
This also fixes writeback to not write the register value until it
has received both pieces of an unaligned load value.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
The obscure bug is that a non-cacheable load with update would never
do the update and would never complete the instruction. This is fixed
by making state NC_LOAD_WAIT_ACK go to LOAD_UPDATE2 if r1.req.update
is set.
The slow load forms with update can go to LOAD_UPDATE2 at the end
rather than LOAD_UPDATE, thus saving a cycle. Loads with a cache
hit need the LOAD_UPDATE state in the third cycle since they are
not writing back until the 4th cycle, when the state is LOAD_UPDATE2.
Slow loads (cacheable loads that miss and non-cacheable loads)
currently go to LOAD_UPDATE in the cycle after they see
r1.wb.ack = 1 for the last time, but that cycle is the cycle where
they write back, and the following cycle does nothing. Going to
LOAD_UPDATE2 in those cases saves a cycle and makes them consistent
with the load hit case.
The logic in the RELOAD_WAIT_ACK case doesn't need to check
r1.req.load = '1' since we only ever use RELOAD_WAIT_ACK for loads.
There are also some whitespace fixes and a typo fix.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
The carry is currently internal to execute1. We don't handle any of
the other XER fields.
This creates type called "xer_common_t" that contains the commonly
used XER bits (CA, CA32, SO, OV, OV32).
The value is stored in the CR file (though it could be a separate
module). The rest of the bits will be implemented as a separate
SPR and the two parts reconciled in mfspr/mtspr in latter commits.
We always read XER in decode2 (there is little point not to)
and send it down all pipeline branches as it will be needed in
writeback for all type of instructions when CR0:SO needs to be
updated (such forms exist for all pipeline branches even if we don't
yet implement them).
To avoid having to track XER hazards, we forward it back in EX1. This
assumes that other pipeline branches that can modify it (mult and div)
are running single issue for now.
One additional hazard to beware of is an XER:SO modifying instruction
in EX1 followed immediately by a store conditional. Due to our writeback
latency, the store will go down the LSU with the previous XER value,
thus the stcx. will set CR0:SO using an obsolete SO value.
I doubt there exist any code relying on this behaviour being correct
but we should account for it regardless, possibly by ensuring that
stcx. remain single issue initially, or later by adding some minimal
tracking or moving the LSU into the same pipeline as execute.
Missing some obscure XER affecting instructions like addex or mcrxrx.
[paulus@ozlabs.org - fix CA32 and OV32 for OP_ADD, fix order of
arguments to set_ov]
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
All that needs to be changed now is the size in wishbone_types.vhdl
and the address decoder in soc.vhdl
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
We used the variable "way" in the wrong state in the cache when
updating a line valid bit after the end of the wishbone transactions,
we need to use the latched "store_way".
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Clearly separate the 2 stages of load hits, improve naming and
comments, clarify the writeback controls etc...
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
This makes the BRAMs use an output buffer, introducing an extra
cycle latency. Without this, Vivado won't make timing at 100Mhz.
We stash all the necessary response data in delayed latches, the
extra cycle is NOT a state in the state machine, thus it's fully
pipelined and doesn't involve stalling.
This introduces an extra non-pipelined cycle for loads with update
to avoid collision on the writeback output between the now delayed
load data and the register update. We could avoid it by moving
the register update in the pipeline bubble created by the extra
update state, but it's a bit trickier, so I leave that for a latter
optimization.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
This replaces loadstore2 with a dcache
The dcache unit is losely based on the icache one (same basic cache
layout), but has some significant logic additions to deal with stores,
loads with update, non-cachable accesses and other differences due to
operating in the execution part of the pipeline rather than the fetch
part.
The cache is store-through, though a hit with an existing line will
update the line rather than invalidate it.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Benjamin Herrenschmidt