3 not in any way intended for production use. this runs a FSM that:
5 * reads the Program Counter from StateRegs
6 * reads an instruction from a fixed-size Test Memory
7 * issues it to the Simple Core
8 * waits for it to complete
10 * does it all over again
12 the purpose of this module is to verify the functional correctness
13 of the Function Units in the absolute simplest and clearest possible
14 way, and to at provide something that can be further incrementally
18 from nmigen
import (Elaboratable
, Module
, Signal
, ClockSignal
, ResetSignal
,
19 ClockDomain
, DomainRenamer
)
20 from nmigen
.cli
import rtlil
21 from nmigen
.cli
import main
24 from soc
.decoder
.power_decoder
import create_pdecode
25 from soc
.decoder
.power_decoder2
import PowerDecode2
26 from soc
.decoder
.decode2execute1
import Data
27 from soc
.experiment
.testmem
import TestMemory
# test only for instructions
28 from soc
.regfile
.regfiles
import StateRegs
, FastRegs
29 from soc
.simple
.core
import NonProductionCore
30 from soc
.config
.test
.test_loadstore
import TestMemPspec
31 from soc
.config
.ifetch
import ConfigFetchUnit
32 from soc
.decoder
.power_enums
import MicrOp
33 from soc
.debug
.dmi
import CoreDebug
, DMIInterface
34 from soc
.config
.state
import CoreState
35 from soc
.interrupts
.xics
import XICS_ICP
, XICS_ICS
36 from soc
.bus
.simple_gpio
import SimpleGPIO
38 from nmutil
.util
import rising_edge
41 class TestIssuer(Elaboratable
):
42 """TestIssuer - reads instructions from TestMemory and issues them
44 efficiency and speed is not the main goal here: functional correctness is.
46 def __init__(self
, pspec
):
48 # add interrupt controller?
49 self
.xics
= hasattr(pspec
, "xics") and pspec
.xics
== True
51 self
.xics_icp
= XICS_ICP()
52 self
.xics_ics
= XICS_ICS()
53 self
.int_level_i
= self
.xics_ics
.int_level_i
55 # add GPIO peripheral?
56 self
.gpio
= hasattr(pspec
, "gpio") and pspec
.gpio
== True
58 self
.simple_gpio
= SimpleGPIO()
59 self
.gpio_o
= self
.simple_gpio
.gpio_o
61 # main instruction core
62 self
.core
= core
= NonProductionCore(pspec
)
65 pdecode
= create_pdecode()
66 self
.pdecode2
= PowerDecode2(pdecode
) # decoder
68 # Test Instruction memory
69 self
.imem
= ConfigFetchUnit(pspec
).fu
70 # one-row cache of instruction read
71 self
.iline
= Signal(64) # one instruction line
72 self
.iprev_adr
= Signal(64) # previous address: if different, do read
75 self
.dbg
= CoreDebug()
77 # instruction go/monitor
78 self
.pc_o
= Signal(64, reset_less
=True)
79 self
.pc_i
= Data(64, "pc_i") # set "ok" to indicate "please change me"
80 self
.core_bigendian_i
= Signal()
81 self
.busy_o
= Signal(reset_less
=True)
82 self
.memerr_o
= Signal(reset_less
=True)
84 # FAST regfile read /write ports for PC, MSR, DEC/TB
85 staterf
= self
.core
.regs
.rf
['state']
86 self
.state_r_pc
= staterf
.r_ports
['cia'] # PC rd
87 self
.state_w_pc
= staterf
.w_ports
['d_wr1'] # PC wr
88 self
.state_r_msr
= staterf
.r_ports
['msr'] # MSR rd
90 # DMI interface access
91 intrf
= self
.core
.regs
.rf
['int']
92 crrf
= self
.core
.regs
.rf
['cr']
93 xerrf
= self
.core
.regs
.rf
['xer']
94 self
.int_r
= intrf
.r_ports
['dmi'] # INT read
95 self
.cr_r
= crrf
.r_ports
['full_cr_dbg'] # CR read
96 self
.xer_r
= xerrf
.r_ports
['full_xer'] # XER read
98 # hack method of keeping an eye on whether branch/trap set the PC
99 self
.state_nia
= self
.core
.regs
.rf
['state'].w_ports
['nia']
100 self
.state_nia
.wen
.name
= 'state_nia_wen'
102 def elaborate(self
, platform
):
104 comb
, sync
= m
.d
.comb
, m
.d
.sync
106 m
.submodules
.core
= core
= DomainRenamer("coresync")(self
.core
)
107 m
.submodules
.imem
= imem
= self
.imem
108 m
.submodules
.dbg
= dbg
= self
.dbg
110 # current state (MSR/PC at the moment
111 cur_state
= CoreState("cur")
113 # XICS interrupt handler
115 m
.submodules
.xics_icp
= icp
= self
.xics_icp
116 m
.submodules
.xics_ics
= ics
= self
.xics_ics
117 comb
+= icp
.ics_i
.eq(ics
.icp_o
) # connect ICS to ICP
118 sync
+= cur_state
.eint
.eq(icp
.core_irq_o
) # connect ICP to core
120 # GPIO test peripheral
122 m
.submodules
.simple_gpio
= simple_gpio
= self
.simple_gpio
124 # connect one GPIO output to ICS bit 15 (like in microwatt soc.vhdl)
125 if self
.gpio
and self
.xics
:
126 comb
+= self
.int_level_i
[15].eq(simple_gpio
.gpio_o
[0])
128 # instruction decoder
129 pdecode
= create_pdecode()
130 m
.submodules
.dec2
= pdecode2
= self
.pdecode2
133 dmi
, d_reg
, d_cr
, d_xer
, = dbg
.dmi
, dbg
.d_gpr
, dbg
.d_cr
, dbg
.d_xer
134 intrf
= self
.core
.regs
.rf
['int']
136 # clock delay power-on reset
137 cd_por
= ClockDomain(reset_less
=True)
138 cd_sync
= ClockDomain()
139 core_sync
= ClockDomain("coresync")
140 m
.domains
+= cd_por
, cd_sync
, core_sync
142 delay
= Signal(range(4), reset
=3)
143 with m
.If(delay
!= 0):
144 m
.d
.por
+= delay
.eq(delay
- 1)
145 comb
+= cd_por
.clk
.eq(ClockSignal())
146 comb
+= core_sync
.clk
.eq(ClockSignal())
147 # power-on reset delay
148 comb
+= core
.core_reset_i
.eq(delay
!= 0 | dbg
.core_rst_o
)
150 # busy/halted signals from core
151 comb
+= self
.busy_o
.eq(core
.busy_o
)
152 comb
+= pdecode2
.dec
.bigendian
.eq(self
.core_bigendian_i
)
154 # temporary hack: says "go" immediately for both address gen and ST
156 ldst
= core
.fus
.fus
['ldst0']
157 st_go_edge
= rising_edge(m
, ldst
.st
.rel_o
)
158 m
.d
.comb
+= ldst
.ad
.go_i
.eq(ldst
.ad
.rel_o
) # link addr-go direct to rel
159 m
.d
.comb
+= ldst
.st
.go_i
.eq(st_go_edge
) # link store-go to rising rel
161 # PC and instruction from I-Memory
162 pc_changed
= Signal() # note write to PC
163 comb
+= self
.pc_o
.eq(cur_state
.pc
)
166 # next instruction (+4 on current)
167 nia
= Signal(64, reset_less
=True)
168 comb
+= nia
.eq(cur_state
.pc
+ 4)
171 pc
= Signal(64, reset_less
=True)
172 pc_ok_delay
= Signal()
173 sync
+= pc_ok_delay
.eq(~self
.pc_i
.ok
)
174 with m
.If(self
.pc_i
.ok
):
175 # incoming override (start from pc_i)
176 comb
+= pc
.eq(self
.pc_i
.data
)
178 # otherwise read StateRegs regfile for PC...
179 comb
+= self
.state_r_pc
.ren
.eq(1<<StateRegs
.PC
)
180 # ... but on a 1-clock delay
181 with m
.If(pc_ok_delay
):
182 comb
+= pc
.eq(self
.state_r_pc
.data_o
)
184 # don't write pc every cycle
185 comb
+= self
.state_w_pc
.wen
.eq(0)
186 comb
+= self
.state_w_pc
.data_i
.eq(0)
188 # don't read msr every cycle
189 comb
+= self
.state_r_msr
.ren
.eq(0)
190 msr_read
= Signal(reset
=1)
192 # connect up debug signals
193 # TODO comb += core.icache_rst_i.eq(dbg.icache_rst_o)
194 comb
+= dbg
.terminate_i
.eq(core
.core_terminate_o
)
195 comb
+= dbg
.state
.pc
.eq(pc
)
196 #comb += dbg.state.pc.eq(cur_state.pc)
197 comb
+= dbg
.state
.msr
.eq(cur_state
.msr
)
200 core_busy_o
= core
.busy_o
# core is busy
201 core_ivalid_i
= core
.ivalid_i
# instruction is valid
202 core_issue_i
= core
.issue_i
# instruction is issued
203 dec_opcode_i
= pdecode2
.dec
.raw_opcode_in
# raw opcode
205 insn_type
= core
.e
.do
.insn_type
206 dec_state
= pdecode2
.state
208 # actually use a nmigen FSM for the first time (w00t)
209 # this FSM is perhaps unusual in that it detects conditions
210 # then "holds" information, combinatorially, for the core
211 # (as opposed to using sync - which would be on a clock's delay)
212 # this includes the actual opcode, valid flags and so on.
216 with m
.State("IDLE"):
217 sync
+= pc_changed
.eq(0)
219 with m
.If(~dbg
.core_stop_o
& ~core
.core_reset_i
):
220 # instruction allowed to go: start by reading the PC
221 # capture the PC and also drop it into Insn Memory
222 # we have joined a pair of combinatorial memory
223 # lookups together. this is Generally Bad.
224 comb
+= self
.imem
.a_pc_i
.eq(pc
)
225 comb
+= self
.imem
.a_valid_i
.eq(1)
226 comb
+= self
.imem
.f_valid_i
.eq(1)
227 sync
+= cur_state
.pc
.eq(pc
)
229 # initiate read of MSR. arrives one clock later
230 comb
+= self
.state_r_msr
.ren
.eq(1<<StateRegs
.MSR
)
231 sync
+= msr_read
.eq(0)
233 m
.next
= "INSN_READ" # move to "wait for bus" phase
235 comb
+= core
.core_stopped_i
.eq(1)
236 comb
+= dbg
.core_stopped_i
.eq(1)
238 # dummy pause to find out why simulation is not keeping up
239 with m
.State("INSN_READ"):
240 # one cycle later, msr read arrives. valid only once.
241 with m
.If(~msr_read
):
242 sync
+= msr_read
.eq(1) # yeah don't read it again
243 sync
+= cur_state
.msr
.eq(self
.state_r_msr
.data_o
)
244 with m
.If(self
.imem
.f_busy_o
): # zzz...
245 # busy: stay in wait-read
246 comb
+= self
.imem
.a_valid_i
.eq(1)
247 comb
+= self
.imem
.f_valid_i
.eq(1)
249 # not busy: instruction fetched
250 f_instr_o
= self
.imem
.f_instr_o
251 if f_instr_o
.width
== 32:
254 insn
= f_instr_o
.word_select(cur_state
.pc
[2], 32)
255 comb
+= dec_opcode_i
.eq(insn
) # actual opcode
256 comb
+= dec_state
.eq(cur_state
)
257 sync
+= core
.e
.eq(pdecode2
.e
)
258 sync
+= ilatch
.eq(insn
) # latch current insn
259 # also drop PC and MSR into decode "state"
260 m
.next
= "INSN_START" # move to "start"
262 # waiting for instruction bus (stays there until not busy)
263 with m
.State("INSN_START"):
264 comb
+= core_ivalid_i
.eq(1) # instruction is valid
265 comb
+= core_issue_i
.eq(1) # and issued
267 m
.next
= "INSN_ACTIVE" # move to "wait completion"
269 # instruction started: must wait till it finishes
270 with m
.State("INSN_ACTIVE"):
271 with m
.If(insn_type
!= MicrOp
.OP_NOP
):
272 comb
+= core_ivalid_i
.eq(1) # instruction is valid
273 with m
.If(self
.state_nia
.wen
& (1<<StateRegs
.PC
)):
274 sync
+= pc_changed
.eq(1)
275 with m
.If(~core_busy_o
): # instruction done!
276 # ok here we are not reading the branch unit. TODO
277 # this just blithely overwrites whatever pipeline
279 with m
.If(~pc_changed
):
280 comb
+= self
.state_w_pc
.wen
.eq(1<<StateRegs
.PC
)
281 comb
+= self
.state_w_pc
.data_i
.eq(nia
)
283 m
.next
= "IDLE" # back to idle
285 # this bit doesn't have to be in the FSM: connect up to read
286 # regfiles on demand from DMI
287 with m
.If(d_reg
.req
): # request for regfile access being made
288 # TODO: error-check this
289 # XXX should this be combinatorial? sync better?
291 comb
+= self
.int_r
.ren
.eq(1<<d_reg
.addr
)
293 comb
+= self
.int_r
.addr
.eq(d_reg
.addr
)
294 comb
+= self
.int_r
.ren
.eq(1)
295 d_reg_delay
= Signal()
296 sync
+= d_reg_delay
.eq(d_reg
.req
)
297 with m
.If(d_reg_delay
):
298 # data arrives one clock later
299 comb
+= d_reg
.data
.eq(self
.int_r
.data_o
)
300 comb
+= d_reg
.ack
.eq(1)
302 # sigh same thing for CR debug
303 with m
.If(d_cr
.req
): # request for regfile access being made
304 comb
+= self
.cr_r
.ren
.eq(0b11111111) # enable all
305 d_cr_delay
= Signal()
306 sync
+= d_cr_delay
.eq(d_cr
.req
)
307 with m
.If(d_cr_delay
):
308 # data arrives one clock later
309 comb
+= d_cr
.data
.eq(self
.cr_r
.data_o
)
310 comb
+= d_cr
.ack
.eq(1)
313 with m
.If(d_xer
.req
): # request for regfile access being made
314 comb
+= self
.xer_r
.ren
.eq(0b111111) # enable all
315 d_xer_delay
= Signal()
316 sync
+= d_xer_delay
.eq(d_xer
.req
)
317 with m
.If(d_xer_delay
):
318 # data arrives one clock later
319 comb
+= d_xer
.data
.eq(self
.xer_r
.data_o
)
320 comb
+= d_xer
.ack
.eq(1)
322 # DEC and TB inc/dec FSM
323 self
.tb_dec_fsm(m
, cur_state
.dec
)
327 def tb_dec_fsm(self
, m
, spr_dec
):
330 this is a FSM for updating either dec or tb. it runs alternately
331 DEC, TB, DEC, TB. note that SPR pipeline could have written a new
332 value to DEC, however the regfile has "passthrough" on it so this
335 see v3.0B p1097-1099 for Timeer Resource and p1065 and p1076
338 comb
, sync
= m
.d
.comb
, m
.d
.sync
339 fast_rf
= self
.core
.regs
.rf
['fast']
340 fast_r_dectb
= fast_rf
.r_ports
['issue'] # DEC/TB
341 fast_w_dectb
= fast_rf
.w_ports
['issue'] # DEC/TB
345 # initiates read of current DEC
346 with m
.State("DEC_READ"):
347 comb
+= fast_r_dectb
.addr
.eq(FastRegs
.DEC
)
348 comb
+= fast_r_dectb
.ren
.eq(1)
351 # waits for DEC read to arrive (1 cycle), updates with new value
352 with m
.State("DEC_WRITE"):
354 # TODO: MSR.LPCR 32-bit decrement mode
355 comb
+= new_dec
.eq(fast_r_dectb
.data_o
- 1)
356 comb
+= fast_w_dectb
.addr
.eq(FastRegs
.DEC
)
357 comb
+= fast_w_dectb
.wen
.eq(1)
358 comb
+= fast_w_dectb
.data_i
.eq(new_dec
)
359 sync
+= spr_dec
.eq(new_dec
) # copy into cur_state for decoder
362 # initiates read of current TB
363 with m
.State("TB_READ"):
364 comb
+= fast_r_dectb
.addr
.eq(FastRegs
.TB
)
365 comb
+= fast_r_dectb
.ren
.eq(1)
368 # waits for read TB to arrive, initiates write of current TB
369 with m
.State("TB_WRITE"):
371 comb
+= new_tb
.eq(fast_r_dectb
.data_o
+ 1)
372 comb
+= fast_w_dectb
.addr
.eq(FastRegs
.TB
)
373 comb
+= fast_w_dectb
.wen
.eq(1)
374 comb
+= fast_w_dectb
.data_i
.eq(new_tb
)
380 yield from self
.pc_i
.ports()
383 yield from self
.core
.ports()
384 yield from self
.imem
.ports()
385 yield self
.core_bigendian_i
391 def external_ports(self
):
392 ports
= self
.pc_i
.ports()
393 ports
+= [self
.pc_o
, self
.memerr_o
, self
.core_bigendian_i
, self
.busy_o
,
394 ClockSignal(), ResetSignal(),
396 ports
+= list(self
.dbg
.dmi
.ports())
397 ports
+= list(self
.imem
.ibus
.fields
.values())
398 ports
+= list(self
.core
.l0
.cmpi
.lsmem
.lsi
.slavebus
.fields
.values())
401 ports
+= list(self
.xics_icp
.bus
.fields
.values())
402 ports
+= list(self
.xics_ics
.bus
.fields
.values())
403 ports
.append(self
.int_level_i
)
406 ports
+= list(self
.simple_gpio
.bus
.fields
.values())
407 ports
.append(self
.gpio_o
)
415 if __name__
== '__main__':
416 units
= {'alu': 1, 'cr': 1, 'branch': 1, 'trap': 1, 'logical': 1,
422 pspec
= TestMemPspec(ldst_ifacetype
='bare_wb',
423 imem_ifacetype
='bare_wb',
428 dut
= TestIssuer(pspec
)
429 vl
= main(dut
, ports
=dut
.ports(), name
="test_issuer")
431 if len(sys
.argv
) == 1:
432 vl
= rtlil
.convert(dut
, ports
=dut
.external_ports(), name
="test_issuer")
433 with
open("test_issuer.il", "w") as f
: