Tobias Platen [Tue, 1 Feb 2022 18:13:33 +0000 (18:13 +0000)]
correct path for make target microwatt_external_core
Luke Kenneth Casson Leighton [Mon, 31 Jan 2022 20:38:45 +0000 (20:38 +0000)]
fix bug in itlb_valid SRLatch set/reset, a bit weird but it works
Luke Kenneth Casson Leighton [Mon, 31 Jan 2022 16:09:12 +0000 (16:09 +0000)]
whoops tlb_valids in ICache is a combinatorial-get/set
set SRLatch sync=False
Luke Kenneth Casson Leighton [Mon, 31 Jan 2022 15:57:56 +0000 (15:57 +0000)]
convert TLBValidArray in ICache to SRLatch
Luke Kenneth Casson Leighton [Mon, 31 Jan 2022 15:47:06 +0000 (15:47 +0000)]
add microwatt external core build target to Makefile
Luke Kenneth Casson Leighton [Mon, 31 Jan 2022 15:29:44 +0000 (15:29 +0000)]
use an SRLatch for cache_valids, at least it reduces graphviz size
Luke Kenneth Casson Leighton [Mon, 31 Jan 2022 15:06:56 +0000 (15:06 +0000)]
use Memory for cache tags in dcache
Luke Kenneth Casson Leighton [Mon, 31 Jan 2022 15:06:36 +0000 (15:06 +0000)]
use Memory for cache_tags in icache
Luke Kenneth Casson Leighton [Mon, 31 Jan 2022 15:05:37 +0000 (15:05 +0000)]
doh
Luke Kenneth Casson Leighton [Mon, 31 Jan 2022 15:02:37 +0000 (15:02 +0000)]
remove dummy trap pipeline
Luke Kenneth Casson Leighton [Mon, 31 Jan 2022 14:59:45 +0000 (14:59 +0000)]
remove combinatorial loop from MultiCompUnit
actually not a loop due to an SRLatch but synth tools still think it is
Luke Kenneth Casson Leighton [Sun, 30 Jan 2022 22:18:26 +0000 (22:18 +0000)]
break out cache_tags and cache_valids (again) this time in dcache
Luke Kenneth Casson Leighton [Sun, 30 Jan 2022 22:08:38 +0000 (22:08 +0000)]
remove CacheTagArray in icache.py
Luke Kenneth Casson Leighton [Sun, 30 Jan 2022 22:03:24 +0000 (22:03 +0000)]
create Memory for Cache Tags in I-Cache
another huge reduction in number of LUT4s, uses (again) a
combinatorial-read sync-write
Luke Kenneth Casson Leighton [Sun, 30 Jan 2022 21:30:34 +0000 (21:30 +0000)]
remove unneeded parameter
Luke Kenneth Casson Leighton [Sun, 30 Jan 2022 21:28:52 +0000 (21:28 +0000)]
add Array of CacheValids back in, so as to reduce LUT4 usage
Luke Kenneth Casson Leighton [Sun, 30 Jan 2022 21:11:09 +0000 (21:11 +0000)]
tagset is a local Signal in ICache
Luke Kenneth Casson Leighton [Sun, 30 Jan 2022 21:10:55 +0000 (21:10 +0000)]
identify combinatorial loop signals in MultiCompUnit, TODO resolve
Luke Kenneth Casson Leighton [Sun, 30 Jan 2022 18:16:37 +0000 (18:16 +0000)]
use nmigen Memory in I-Cache for TLB Lookups
surprisingly this makes the Libre-SOC core *50% faster* than microwatt
when running under verilator, despite only being a FSM
Luke Kenneth Casson Leighton [Sun, 30 Jan 2022 17:49:17 +0000 (17:49 +0000)]
put itlb_valid back, ready for conversion to Memory, in ICache
Luke Kenneth Casson Leighton [Sun, 30 Jan 2022 16:51:41 +0000 (16:51 +0000)]
convert CacheRAM to Memory, acts much faster now
Luke Kenneth Casson Leighton [Sat, 29 Jan 2022 16:43:56 +0000 (16:43 +0000)]
explanatory comment when page hit is the same for stores
Luke Kenneth Casson Leighton [Sat, 29 Jan 2022 14:32:23 +0000 (14:32 +0000)]
use right offset in dcache wb address
happened to be the same value but best to be safe, eh?
Luke Kenneth Casson Leighton [Sat, 29 Jan 2022 13:48:19 +0000 (13:48 +0000)]
re-examining dcache.vhdl, still did not get the store-page
address quite right
Luke Kenneth Casson Leighton [Sat, 29 Jan 2022 10:17:42 +0000 (10:17 +0000)]
bug in dcache.py where when two stores occur in the same real page
the address is corrupted.
Luke Kenneth Casson Leighton [Fri, 28 Jan 2022 19:33:13 +0000 (19:33 +0000)]
in LoadStore1 capture the address for misaligned dual ld/sts in
a different way.
something very strange going on with misaligned stores: the address
is advancing far too far under certain circumstances (by 128) which
could just be an MMU / PTE lookup to a different table.
Luke Kenneth Casson Leighton [Fri, 28 Jan 2022 03:19:40 +0000 (03:19 +0000)]
sort out misaligned store in LoadStore1
Luke Kenneth Casson Leighton [Thu, 27 Jan 2022 10:49:14 +0000 (10:49 +0000)]
for second aligned request truncate address to nearest dword
this ensures that DAR gets set correctly if a pagefault 0x300 occurs
Luke Kenneth Casson Leighton [Tue, 25 Jan 2022 00:43:56 +0000 (00:43 +0000)]
add license and copyright header to dcache.py,
extracted authors from git history for the file, but made sure to
credit the original dcache.vhdl as being from microwatt and its
license being CC4
Luke Kenneth Casson Leighton [Tue, 25 Jan 2022 00:42:44 +0000 (00:42 +0000)]
LDSTException now passing bits of SRR1 around to the Trap Pipeline
the actual (former) value of SRR1 is not what is supposed to be used:
the use of the variable "srr1" is a moniker from microwatt
Luke Kenneth Casson Leighton [Mon, 24 Jan 2022 21:23:49 +0000 (21:23 +0000)]
comments
Luke Kenneth Casson Leighton [Mon, 24 Jan 2022 14:11:07 +0000 (14:11 +0000)]
hmm there seems to have been an error in DTLB Read,
where if a write *and* a read occurred at the same time, the old
DTLB-valid entry was given. add similar "forwarding" that is used in
Memory. DTLB-valid is actually a register not a Memory, where the
DTLB way/tags are a Memory, hence the bug
Luke Kenneth Casson Leighton [Mon, 24 Jan 2022 11:58:03 +0000 (11:58 +0000)]
bool test on traptype to
ensure two conditions are properly ANDed
also copy correct bits of SRR over, but there is an additional
bug here that needs to be fixed: Exception class needs to pass over
the bottom 16 LSBs of SRR1
Luke Kenneth Casson Leighton [Sun, 23 Jan 2022 11:05:23 +0000 (11:05 +0000)]
looked in soc.vhdl in microwatt and the parameters are 64 cache
lines. this would not be important if it was not explicitly in
the linux-5.7 device-tree file
Luke Kenneth Casson Leighton [Sun, 23 Jan 2022 11:04:25 +0000 (11:04 +0000)]
add debug output of whether stall occurs on dcache
Luke Kenneth Casson Leighton [Sat, 22 Jan 2022 15:19:00 +0000 (15:19 +0000)]
missed setting of r0_full to zero in dcache. not encountered as
a bug but would have done in future
Luke Kenneth Casson Leighton [Fri, 21 Jan 2022 19:26:07 +0000 (19:26 +0000)]
skip ilang data in branch test_pipe_caller.py
Luke Kenneth Casson Leighton [Fri, 21 Jan 2022 19:21:57 +0000 (19:21 +0000)]
attempting to get compunit and test_pipe_caller unit tests
up and running again.
grrr
Luke Kenneth Casson Leighton [Fri, 21 Jan 2022 00:12:39 +0000 (00:12 +0000)]
sigh, monitor DEC/TB StateRegs "properly" so that the Issuer DEC/TB FSM
does not end up in a race condition with the SPR pipeline for writing
to DEC or TB
Luke Kenneth Casson Leighton [Fri, 21 Jan 2022 00:11:53 +0000 (00:11 +0000)]
whoops fix bug in setting of DEC/TB (State) in test_core.py
Luke Kenneth Casson Leighton [Thu, 20 Jan 2022 18:38:20 +0000 (18:38 +0000)]
whoops MFSPR DEC/TB was reading from FastRegs not StateRegs
also TBU
Luke Kenneth Casson Leighton [Wed, 19 Jan 2022 17:46:13 +0000 (17:46 +0000)]
whoops forgot to enable fast-reg read in DMI
Luke Kenneth Casson Leighton [Wed, 19 Jan 2022 17:18:35 +0000 (17:18 +0000)]
ISI (0x400) trap is the only one that puts memory-based exception
info into SRR1, not *all* memory-based exceptions
Luke Kenneth Casson Leighton [Wed, 19 Jan 2022 17:17:40 +0000 (17:17 +0000)]
comments
Luke Kenneth Casson Leighton [Wed, 19 Jan 2022 12:16:25 +0000 (12:16 +0000)]
move DEC and TB into StateRegs, to make room in FastRegs
also has the advantage that DEC and TB could generate an accurate interrupt
Luke Kenneth Casson Leighton [Tue, 18 Jan 2022 18:15:17 +0000 (18:15 +0000)]
add support for DMI debug read of FAST Regfile SPRs
this to be able to do a side-by-side compare against microwatt
single-stepping
Luke Kenneth Casson Leighton [Tue, 18 Jan 2022 16:28:49 +0000 (16:28 +0000)]
comments on SRR1 in trap
Luke Kenneth Casson Leighton [Tue, 18 Jan 2022 13:42:12 +0000 (13:42 +0000)]
preserve bits of SRR1 on a TRAP (including all interrupts) which in
turn means that PowerDecoder2 has to read SRR1
Luke Kenneth Casson Leighton [Mon, 17 Jan 2022 18:00:52 +0000 (18:00 +0000)]
fix hrfid and mtmsrd so that it is identical to microwatt
both allow MSR.ME to be set, which walks linux-5.7 along a different
codepath particularly for 0x900 exception handling
Luke Kenneth Casson Leighton [Mon, 17 Jan 2022 17:59:59 +0000 (17:59 +0000)]
connect up DEC/TB FSM pauser from core to Issuer
Luke Kenneth Casson Leighton [Mon, 17 Jan 2022 12:01:17 +0000 (12:01 +0000)]
comments
Luke Kenneth Casson Leighton [Mon, 17 Jan 2022 12:00:36 +0000 (12:00 +0000)]
whitespace
Luke Kenneth Casson Leighton [Mon, 17 Jan 2022 11:59:56 +0000 (11:59 +0000)]
add pause_dec_tb signal (not very sophisticated) to Core
TODO, detect MTSPR and DEC/TB SPR being written to, but for now just
detect an entire SPR pipeline
Luke Kenneth Casson Leighton [Mon, 17 Jan 2022 11:50:50 +0000 (11:50 +0000)]
add signal for pausing the DEC/TB FSM to IssuerBase
there is a potential issue with the DEC SPR that needs solving,
and there is a race condition where an mtspr DEC/TB could get
overwritten
adding a "pause" mechanism to the FSM should solve that
Luke Kenneth Casson Leighton [Sun, 16 Jan 2022 16:29:13 +0000 (16:29 +0000)]
raise interrupt on misaligned atomic LDST
Luke Kenneth Casson Leighton [Sun, 16 Jan 2022 16:28:49 +0000 (16:28 +0000)]
pass over store_done correctly from dcache over PortInterface
into LDSTCompUnit so that it can set CR0 correctly on stdcx. etc.
Luke Kenneth Casson Leighton [Sun, 16 Jan 2022 09:08:52 +0000 (09:08 +0000)]
add CR0 to LDSTCompUnit, for reporting if LR/SC store is done
Luke Kenneth Casson Leighton [Sun, 16 Jan 2022 08:54:22 +0000 (08:54 +0000)]
remove PortInterface mmu_done signal,
add store_done
Luke Kenneth Casson Leighton [Sat, 15 Jan 2022 21:47:18 +0000 (21:47 +0000)]
forgot name on dcache Reservation
Luke Kenneth Casson Leighton [Sat, 15 Jan 2022 20:56:35 +0000 (20:56 +0000)]
pass over atomic signals to dcache from loadstore.
does not do everything yet: load-quad for example is not included
Luke Kenneth Casson Leighton [Sat, 15 Jan 2022 18:44:24 +0000 (18:44 +0000)]
try using req.op in RELOAD_WAIT_ACK to detect whether request
can complete next cycle
Luke Kenneth Casson Leighton [Sat, 15 Jan 2022 14:21:50 +0000 (14:21 +0000)]
pass atomic reserve through from PortInterface to DCache
not yet doing anything with it, so should be fine
Luke Kenneth Casson Leighton [Sat, 15 Jan 2022 14:06:44 +0000 (14:06 +0000)]
add atomic LR/SC signal to LDSTCompUnit
Luke Kenneth Casson Leighton [Sat, 15 Jan 2022 14:04:55 +0000 (14:04 +0000)]
add reserve (atomic) signal to LDST data structures including PortInterface
Luke Kenneth Casson Leighton [Sat, 15 Jan 2022 14:03:02 +0000 (14:03 +0000)]
tidyup PortInterface
Luke Kenneth Casson Leighton [Sat, 15 Jan 2022 12:36:52 +0000 (12:36 +0000)]
workaround for bug in dcache where the r1.req waiting to be deployed
was interfering with the current state being executed
http://lists.libre-soc.org/pipermail/libre-soc-dev/2022-January/004358.html
Luke Kenneth Casson Leighton [Sat, 15 Jan 2022 12:35:12 +0000 (12:35 +0000)]
enable both linux-5.7 tests
Luke Kenneth Casson Leighton [Fri, 14 Jan 2022 14:02:19 +0000 (14:02 +0000)]
split out CacheTag Record to separate structure
Luke Kenneth Casson Leighton [Fri, 14 Jan 2022 03:02:39 +0000 (03:02 +0000)]
update how d_valid is handled
Luke Kenneth Casson Leighton [Fri, 14 Jan 2022 01:42:07 +0000 (01:42 +0000)]
missed setting r1.store_way and r1.store_row in STORE_WAIT_ACK state
Luke Kenneth Casson Leighton [Fri, 14 Jan 2022 01:28:17 +0000 (01:28 +0000)]
Revert "dcache 2nd stage (r1) should only indicate not-busy"
This reverts commit
a03aefb1e8ae7d6110a328b57f1336890ebee469.
Luke Kenneth Casson Leighton [Fri, 14 Jan 2022 01:27:09 +0000 (01:27 +0000)]
second test for linux-5.7
Luke Kenneth Casson Leighton [Wed, 12 Jan 2022 20:09:03 +0000 (20:09 +0000)]
add allow-overlap option to issuer_verilog.py
Luke Kenneth Casson Leighton [Wed, 12 Jan 2022 12:11:12 +0000 (12:11 +0000)]
dcache 2nd stage (r1) should only indicate not-busy
(r1.full) when all the ACKs of a cache-line fill have been processed
doing this too early results in r0 being pushed into r1 whilst
ACKs are still outstanding, and their completion corrupts the
operation that should not have been put into r1 in the first place
Luke Kenneth Casson Leighton [Wed, 12 Jan 2022 11:22:24 +0000 (11:22 +0000)]
fix issue with priv_mode not being passed correctly to MMU
on instruction load
Luke Kenneth Casson Leighton [Wed, 12 Jan 2022 11:21:40 +0000 (11:21 +0000)]
fix issue with d_valid in dcache, was not being set properly
Luke Kenneth Casson Leighton [Mon, 10 Jan 2022 23:03:25 +0000 (23:03 +0000)]
LoadStore1 priv_mode was not being correctly picked up by the MMU
priv_mode needs to come from the original LD/ST request (or the
fetch), which was not happening
Luke Kenneth Casson Leighton [Sun, 9 Jan 2022 23:40:34 +0000 (23:40 +0000)]
grab the LDST request address for microwatt verilator debug purposes
Luke Kenneth Casson Leighton [Sun, 9 Jan 2022 23:29:10 +0000 (23:29 +0000)]
add linux-5.7 unit test which showed a silly error:
LDST requests through PortInterface were truncated to 48 bits,
where linux uses the top 2 bits of an address for VM/guest (Quadrant 0-3)
Luke Kenneth Casson Leighton [Sat, 8 Jan 2022 17:46:56 +0000 (17:46 +0000)]
fix MMU lookup after 2nd request (misaligned) by also updating the
ldst_r with the next address/byte_sel
Luke Kenneth Casson Leighton [Sat, 8 Jan 2022 17:27:42 +0000 (17:27 +0000)]
add microwatt mmu.bin test5 to show page-fault on misaligned LD
Luke Kenneth Casson Leighton [Sat, 8 Jan 2022 16:49:27 +0000 (16:49 +0000)]
do not clear out ldst request after TLB entry is added
Luke Kenneth Casson Leighton [Sat, 8 Jan 2022 15:42:58 +0000 (15:42 +0000)]
enable microwatt mmu test2
Luke Kenneth Casson Leighton [Sat, 8 Jan 2022 15:41:32 +0000 (15:41 +0000)]
whitespace and use exc is None not exc == None
Luke Kenneth Casson Leighton [Sat, 8 Jan 2022 15:26:03 +0000 (15:26 +0000)]
add a second LD request to dcache which is merged with first,
to implement mis-aligned LD operations
Luke Kenneth Casson Leighton [Sat, 8 Jan 2022 14:10:16 +0000 (14:10 +0000)]
start adding in mis-aligned LD/ST support into LoadStore1
currently not activated or used so will have no effect
Tobias Platen [Sat, 8 Jan 2022 13:30:25 +0000 (14:30 +0100)]
add function test_pi_ld_misalign
Tobias Platen [Fri, 7 Jan 2022 18:07:43 +0000 (19:07 +0100)]
begin testcase for misalign
Luke Kenneth Casson Leighton [Fri, 7 Jan 2022 16:59:59 +0000 (16:59 +0000)]
whitespace
Luke Kenneth Casson Leighton [Fri, 7 Jan 2022 16:57:56 +0000 (16:57 +0000)]
add missing MSRSpec import
Luke Kenneth Casson Leighton [Fri, 7 Jan 2022 12:26:45 +0000 (12:26 +0000)]
add msr_o to issuer in microwatt_compat mode
Luke Kenneth Casson Leighton [Thu, 6 Jan 2022 17:32:46 +0000 (17:32 +0000)]
double the number of lines in the L1 D/I-Cache to match microwatt
early tests halved the number of lines so as to reduce the size of SRAMs
but the issue is that this is mis-matched against the microwatt.dts
device-tree file
Luke Kenneth Casson Leighton [Thu, 6 Jan 2022 17:31:57 +0000 (17:31 +0000)]
add SECOND_REQ state to loadstore.py, not yet implemented
Luke Kenneth Casson Leighton [Wed, 5 Jan 2022 16:32:42 +0000 (16:32 +0000)]
add easy-to-access debug reporting of instruction and PC
for microwatt verilator
Luke Kenneth Casson Leighton [Wed, 5 Jan 2022 16:31:39 +0000 (16:31 +0000)]
use microwatt-specific PLRU due to bug in nmutil version
(needs investigating)
Luke Kenneth Casson Leighton [Tue, 4 Jan 2022 17:19:47 +0000 (17:19 +0000)]
fix DriverConflict over MSR write in Issuer/Core by providing an
extra write-port to StateRegs
Luke Kenneth Casson Leighton [Tue, 4 Jan 2022 17:03:48 +0000 (17:03 +0000)]
remove FetchFSM from TestIssuer (it served its purpose for creating
the Inorder version)
Luke Kenneth Casson Leighton [Mon, 3 Jan 2022 23:37:28 +0000 (23:37 +0000)]
doh, bus-hack was the wrong way round. *output* the address with
3 extra LSBs at the front to fix the wishbone incompatibility
Luke Kenneth Casson Leighton [Mon, 3 Jan 2022 23:28:13 +0000 (23:28 +0000)]
sigh, microwatts wishbone bus usage is non-wishbone-compliant:
the full address (including LSBs) is dropped onto the bus
Luke Kenneth Casson Leighton [Mon, 3 Jan 2022 22:33:39 +0000 (22:33 +0000)]
sigh have to allow external clocks and reset mess even in microwatt-compat
mode. soc.vhdl still needs to be able to pull an external reset OR
DMI needs to be able to instruct the core to do it. hardly surprising