The current dcache will not update the PLRU on a cache miss which is later
satisfied during the reload process. Thus subsequent misses will potentially
evict the same cache line. The same issue happens with dcbz which are
treated more/less as load misses.
This fixes it by triggering a PLRU update when r1.choose_victim, which is
set on a miss for one cycle to snapshot the PLRU output. This means we will
update the PLRU on the same cycle as we capture its output, which is fine
(the new value will be visible on the next cycle).
That way, a "miss" will result in a PLRU update to reflect that the entry
being refilled is actually used (and will be used to serve subsequent
load operations from the same cache line while being refilled).
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
It bitrotted... more signals need to be initialized. This also adds
a lot more accesses with different timing conditions allowing to
test cases of hit during reloads, hit with reload formward, hit on idle
cache etc...
It also exposes a bug where the cache miss caused by the read of 0x140
uses the same victim way as previous cache miss of 0x40 (same index).
This bug will need to be fixed separately, but at least this exposes it.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Starting from 5e7612eb4, OpenOCD identifies itself as 0.12.
This causes Microwatt's flash-arty script to fail. Because neither
the cfg nor the proxy bitstream are affected, we can keep treating
everything as indistinguishable from 0.11. This patch simply tests
for "0.12" as an alias; it would probably be better to replace this
confusing terminology with something like "single-tap/multi-tap".
Signed-off-by: Boris Shingarov <shingarov@labware.com>
This includes the cable configuration, additions to the Python script,
and the jtagspi proxy bitstream. The single-tap version is not included
because 0.10 supported only 3-byte addresses which is unusable on the
s25fl256s anyway.
Signed-off-by: Boris Shingarov <shingarov@labware.com>
As has been done for the L1 dcache and icache, this puts the L2 cache
PLRU state into a little RAM and has a single copy of the logic to
calculate the pseudo-LRU way and update the PLRU state.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Rather than having update and decode logic for each individual PLRU
as well as a register to store the current PLRU state, we now put the
PLRU state in a little RAM, which will typically use LUT RAM on FPGAs,
and have just a single copy of the logic to calculate the pseudo-LRU
way and to update the PLRU state.
The PLRU RAM that apples to the data storage (as opposed to the TLB)
is read asynchronously in the cycle after the cache tag matching is
done. At the end of that cycle the PLRU RAM entry is updated if the
access was a cache hit, or a victim way is calculated and stored if
the access was a cache miss. It is possible that a cache miss doesn't
start being handled until later, in which case the stored victim way
is used later when the miss gets handled.
Similarly for the TLB PLRU, the RAM is read asynchronously in the
cycle after a TLB lookup is done, and either updated at the end of
that cycle (for a hit), or a victim is chosen and stored for when the
TLB miss is satisfied.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Rather than having update and decode logic for each individual PLRU
as well as a register to store the current PLRU state, we now put the
PLRU state in a little RAM, which will typically use LUT RAM on FPGAs,
and have just a single copy of the logic to calculate the pseudo-LRU
way and to update the PLRU state. This logic is in the plrufn module
and is just combinatorial logic. A new module was created for this as
other parts of the system are still using plru.vhdl.
The PLRU RAM in the icache is read asynchronously in the cycle
after the cache tag matching is done. At the end of that cycle the
PLRU RAM entry is updated if the access was a cache hit, or a victim
way is calculated and stored if the access was a cache miss and
miss handling is starting in this cycle.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This fixes a couple of build warnings in litedram-wrapper-l2.vhdl
litedram/extras/litedram-wrapper-l2.vhdl:552:17⚠️ declaration of "i" hides constant "i" [-Whide]
for i in 0 to ROW_SIZE-1 loop
^
litedram/extras/litedram-wrapper-l2.vhdl:1129:9⚠️ declaration of "litedram_trace" hides generic "litedram_trace" [-Whide]
litedram_trace: litedram_trace_stub;
^
It also cleans up the runtime metavalue warnings
Signed-off-by: Michael Neuling <mikey@neuling.org>
We disabled --trace by default, so we need to stop linking verilated_vcd_c.o
as it doesn't exist in that case.
While at it, make a Makefile variable to enable/disable verilator tracing
and add a couple of generics to those test benches to control tracing
in the L2 and in litedram.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
The Antmicro Artix DC-SCM uses the following FTDI part:
0403:6011 Future Technology Devices International, Ltd FT4232H Quad HS USB-UART/FIFO IC
To use:
$ openocd/flash-arty -c antmicro-artix-dc-scm -f a100 -t bin -a 0x300000 ~/u-boot
Signed-off-by: Joel Stanley <joel@jms.id.au>
As with the DRAM configuration, the DC-SCM board uses the same PHY as
the Nexys Video and works with it's generated VHDL.
Signed-off-by: Joel Stanley <joel@jms.id.au>
This uses the exact same gateware as the nexys video, since the DRAM
connection is identical to the nexys video down to the pin assignments
on the FPGA. The only minor difference is that the DRAM chip on the
dc-scm is a MT41K256M16TW vs. a ...HA part on the nexys video.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
[joel: rebase and tweaks]
Signed-off-by: Joel Stanley <joel@jms.id.au>
works with:
fusesoc build --target=antmicro-artix-dc-scm microwatt --ram_init_file=../hello_world/hello_world.hex
Signed-off-by: Michael Neuling <mikey@neuling.org>
[joel: Fixes and updates]
Signed-off-by: Joel Stanley <joel@jms.id.au>
It also stores the dirty status so that's known.
This does some Makefile tricks so that we only rebuild when the git
hash changes. This avoids rebuilding the world every time we run
make.
Also adds fusesoc generator, so that should continue to work as
before.
Signed-off-by: Dan Horák <dan@danny.cz>
Signed-off-by: Michael Neuling <mikey@neuling.org>
Jacob Lifshay found a couple of issues with the PLRU implementation:
- The tree array is one bit too long. This is harmless as this bit is never
accessed and thus should be optimized out
- The PLRU read is using the wrong nodes when going down the tree, which leads
to incorrect results.
This fixes it and improves the test bench a bit. I have verified the expected
output using a hand-written tree states, observed the mismatch with the
current implementation and verified the fix.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
As in dcache, this changes most signals declared with integer type to
be unsigned bit vectors instead. Some code has been rearranged to do
to_integer() or equality comparisons only when the relevant signals
should be well defined. Non-fatal asserts have been sprinkled
throughout to assist with determining the cause of warnings from
library functions (primarily NUMERIC_STD.TO_INTEGER and
NUMERIC_STD."=").
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Among other changes, this makes the things that were previously
declared as signals of integer base type to be unsigned, since
unsigned can carry metavalues, and hence we can get the checking for
metavalues closer to the uses and therefore restrict the checking to
the situations where the signal really ought to be well defined.
We now have a couple more signals that indicate request validity to
help with that.
Non-fatal asserts have been sprinkled throughout to assist with
determining the cause of warnings from library functions (primarily
NUMERIC_STD.TO_INTEGER and NUMERIC_STD."=").
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This adds a "not HAS_FPU" condition in a few places to make it obvious
that logic to interface to the divide unit is not included when we
have an FPU.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Now that we have a 33 bit x 33 bit signed multiplier in execute1,
there is really no need for the 16 bit multiplier. The coremark
results are just as good without it as with it. This removes the
option for the sake of simplicity.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This fixes a bug in the BTC where entries created for a given address
when MSR[IR] = 0 are used when MSR[IR] = 1 and vice-versa. The fix is
to include r.virt_mode (which mirrors MSR[IR]) in the tag portion of
the BTC.
Fixes: 0fb207be60 ("fetch1: Implement a simple branch target cache", 2020-12-19)
Reported-by: Anton Blanchard <anton@linux.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This adds an 'is_signed' signal to MultiplyInputType to indicate
whether the data1 and data2 fields are to be interpreted as signed or
unsigned numbers.
The 'not_result' field is replaced by a 'subtract' field which
provides a more intuitive interface for requesting that the product be
subtracted from the addend rather than added, i.e. subtract = 1 gives
C - A * B, vs. subtract = 0 giving C + A * B. (Previously the users
of the multipliers got the same effect by complementing the addend and
setting not_result = 1.)
The is_32bit field is removed because it is no longer used now that we
have a separate 32-bit multiplier.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Benjamin Herrenschmidt
Michael Neuling
Paul Mackerras