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
, Mux
, Const
)
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
, SVP64PrefixDecoder
26 from soc
.decoder
.decode2execute1
import IssuerDecode2ToOperand
27 from soc
.decoder
.decode2execute1
import Data
28 from soc
.experiment
.testmem
import TestMemory
# test only for instructions
29 from soc
.regfile
.regfiles
import StateRegs
, FastRegs
30 from soc
.simple
.core
import NonProductionCore
31 from soc
.config
.test
.test_loadstore
import TestMemPspec
32 from soc
.config
.ifetch
import ConfigFetchUnit
33 from soc
.decoder
.power_enums
import MicrOp
34 from soc
.debug
.dmi
import CoreDebug
, DMIInterface
35 from soc
.debug
.jtag
import JTAG
36 from soc
.config
.pinouts
import get_pinspecs
37 from soc
.config
.state
import CoreState
38 from soc
.interrupts
.xics
import XICS_ICP
, XICS_ICS
39 from soc
.bus
.simple_gpio
import SimpleGPIO
40 from soc
.bus
.SPBlock512W64B8W
import SPBlock512W64B8W
41 from soc
.clock
.select
import ClockSelect
42 from soc
.clock
.dummypll
import DummyPLL
43 from soc
.sv
.svstate
import SVSTATERec
46 from nmutil
.util
import rising_edge
48 def get_insn(f_instr_o
, pc
):
49 if f_instr_o
.width
== 32:
52 # 64-bit: bit 2 of pc decides which word to select
53 return f_instr_o
.word_select(pc
[2], 32)
56 class TestIssuerInternal(Elaboratable
):
57 """TestIssuer - reads instructions from TestMemory and issues them
59 efficiency and speed is not the main goal here: functional correctness is.
61 def __init__(self
, pspec
):
63 # JTAG interface. add this right at the start because if it's
64 # added it *modifies* the pspec, by adding enable/disable signals
65 # for parts of the rest of the core
66 self
.jtag_en
= hasattr(pspec
, "debug") and pspec
.debug
== 'jtag'
68 subset
= {'uart', 'mtwi', 'eint', 'gpio', 'mspi0', 'mspi1',
70 self
.jtag
= JTAG(get_pinspecs(subset
=subset
))
71 # add signals to pspec to enable/disable icache and dcache
72 # (or data and intstruction wishbone if icache/dcache not included)
73 # https://bugs.libre-soc.org/show_bug.cgi?id=520
74 # TODO: do we actually care if these are not domain-synchronised?
75 # honestly probably not.
76 pspec
.wb_icache_en
= self
.jtag
.wb_icache_en
77 pspec
.wb_dcache_en
= self
.jtag
.wb_dcache_en
78 self
.wb_sram_en
= self
.jtag
.wb_sram_en
80 self
.wb_sram_en
= Const(1)
83 self
.sram4x4k
= (hasattr(pspec
, "sram4x4kblock") and
84 pspec
.sram4x4kblock
== True)
88 self
.sram4k
.append(SPBlock512W64B8W(name
="sram4k_%d" % i
,
91 # add interrupt controller?
92 self
.xics
= hasattr(pspec
, "xics") and pspec
.xics
== True
94 self
.xics_icp
= XICS_ICP()
95 self
.xics_ics
= XICS_ICS()
96 self
.int_level_i
= self
.xics_ics
.int_level_i
98 # add GPIO peripheral?
99 self
.gpio
= hasattr(pspec
, "gpio") and pspec
.gpio
== True
101 self
.simple_gpio
= SimpleGPIO()
102 self
.gpio_o
= self
.simple_gpio
.gpio_o
104 # main instruction core25
105 self
.core
= core
= NonProductionCore(pspec
)
107 # instruction decoder. goes into Trap Record
108 pdecode
= create_pdecode()
109 self
.cur_state
= CoreState("cur") # current state (MSR/PC/EINT/SVSTATE)
110 self
.pdecode2
= PowerDecode2(pdecode
, state
=self
.cur_state
,
111 opkls
=IssuerDecode2ToOperand
)
112 self
.svp64
= SVP64PrefixDecoder() # for decoding SVP64 prefix
114 # Test Instruction memory
115 self
.imem
= ConfigFetchUnit(pspec
).fu
116 # one-row cache of instruction read
117 self
.iline
= Signal(64) # one instruction line
118 self
.iprev_adr
= Signal(64) # previous address: if different, do read
121 self
.dbg
= CoreDebug()
123 # instruction go/monitor
124 self
.pc_o
= Signal(64, reset_less
=True)
125 self
.pc_i
= Data(64, "pc_i") # set "ok" to indicate "please change me"
126 self
.svstate_i
= Data(32, "svstate_i") # ditto
127 self
.core_bigendian_i
= Signal()
128 self
.busy_o
= Signal(reset_less
=True)
129 self
.memerr_o
= Signal(reset_less
=True)
131 # STATE regfile read /write ports for PC, MSR, SVSTATE
132 staterf
= self
.core
.regs
.rf
['state']
133 self
.state_r_pc
= staterf
.r_ports
['cia'] # PC rd
134 self
.state_w_pc
= staterf
.w_ports
['d_wr1'] # PC wr
135 self
.state_r_msr
= staterf
.r_ports
['msr'] # MSR rd
136 self
.state_r_sv
= staterf
.r_ports
['sv'] # SVSTATE rd
137 self
.state_w_sv
= staterf
.w_ports
['sv'] # SVSTATE wr
139 # DMI interface access
140 intrf
= self
.core
.regs
.rf
['int']
141 crrf
= self
.core
.regs
.rf
['cr']
142 xerrf
= self
.core
.regs
.rf
['xer']
143 self
.int_r
= intrf
.r_ports
['dmi'] # INT read
144 self
.cr_r
= crrf
.r_ports
['full_cr_dbg'] # CR read
145 self
.xer_r
= xerrf
.r_ports
['full_xer'] # XER read
147 # hack method of keeping an eye on whether branch/trap set the PC
148 self
.state_nia
= self
.core
.regs
.rf
['state'].w_ports
['nia']
149 self
.state_nia
.wen
.name
= 'state_nia_wen'
151 def fetch_fsm(self
, m
, core
, pc
, svstate
, nia
,
152 fetch_pc_ready_o
, fetch_pc_valid_i
,
153 fetch_insn_valid_o
, fetch_insn_ready_i
):
155 this FSM performs fetch of raw instruction data, partial-decodes
156 it 32-bit at a time to detect SVP64 prefixes, and will optionally
157 read a 2nd 32-bit quantity if that occurs.
161 pdecode2
= self
.pdecode2
163 cur_state
= self
.cur_state
164 dec_opcode_i
= pdecode2
.dec
.raw_opcode_in
# raw opcode
166 msr_read
= Signal(reset
=1)
168 with m
.FSM(name
='fetch_fsm'):
171 with m
.State("IDLE"):
172 comb
+= fetch_pc_ready_o
.eq(1)
173 with m
.If(fetch_pc_valid_i
):
174 # instruction allowed to go: start by reading the PC
175 # capture the PC and also drop it into Insn Memory
176 # we have joined a pair of combinatorial memory
177 # lookups together. this is Generally Bad.
178 comb
+= self
.imem
.a_pc_i
.eq(pc
)
179 comb
+= self
.imem
.a_valid_i
.eq(1)
180 comb
+= self
.imem
.f_valid_i
.eq(1)
181 sync
+= cur_state
.pc
.eq(pc
)
182 sync
+= cur_state
.svstate
.eq(svstate
) # and svstate
184 # initiate read of MSR. arrives one clock later
185 comb
+= self
.state_r_msr
.ren
.eq(1 << StateRegs
.MSR
)
186 sync
+= msr_read
.eq(0)
188 m
.next
= "INSN_READ" # move to "wait for bus" phase
190 # dummy pause to find out why simulation is not keeping up
191 with m
.State("INSN_READ"):
192 # one cycle later, msr/sv read arrives. valid only once.
193 with m
.If(~msr_read
):
194 sync
+= msr_read
.eq(1) # yeah don't read it again
195 sync
+= cur_state
.msr
.eq(self
.state_r_msr
.data_o
)
196 with m
.If(self
.imem
.f_busy_o
): # zzz...
197 # busy: stay in wait-read
198 comb
+= self
.imem
.a_valid_i
.eq(1)
199 comb
+= self
.imem
.f_valid_i
.eq(1)
201 # not busy: instruction fetched
202 insn
= get_insn(self
.imem
.f_instr_o
, cur_state
.pc
)
203 # decode the SVP64 prefix, if any
204 comb
+= svp64
.raw_opcode_in
.eq(insn
)
205 comb
+= svp64
.bigendian
.eq(self
.core_bigendian_i
)
206 # pass the decoded prefix (if any) to PowerDecoder2
207 sync
+= pdecode2
.sv_rm
.eq(svp64
.svp64_rm
)
208 # calculate the address of the following instruction
209 insn_size
= Mux(svp64
.is_svp64_mode
, 8, 4)
210 sync
+= nia
.eq(cur_state
.pc
+ insn_size
)
211 with m
.If(~svp64
.is_svp64_mode
):
212 # with no prefix, store the instruction
213 # and hand it directly to the next FSM
214 sync
+= dec_opcode_i
.eq(insn
)
215 m
.next
= "INSN_READY"
217 # fetch the rest of the instruction from memory
218 comb
+= self
.imem
.a_pc_i
.eq(cur_state
.pc
+ 4)
219 comb
+= self
.imem
.a_valid_i
.eq(1)
220 comb
+= self
.imem
.f_valid_i
.eq(1)
221 m
.next
= "INSN_READ2"
223 with m
.State("INSN_READ2"):
224 with m
.If(self
.imem
.f_busy_o
): # zzz...
225 # busy: stay in wait-read
226 comb
+= self
.imem
.a_valid_i
.eq(1)
227 comb
+= self
.imem
.f_valid_i
.eq(1)
229 # not busy: instruction fetched
230 insn
= get_insn(self
.imem
.f_instr_o
, cur_state
.pc
+4)
231 sync
+= dec_opcode_i
.eq(insn
)
232 m
.next
= "INSN_READY"
234 with m
.State("INSN_READY"):
235 # hand over the instruction, to be decoded
236 comb
+= fetch_insn_valid_o
.eq(1)
237 with m
.If(fetch_insn_ready_i
):
240 def issue_fsm(self
, m
, core
, pc_changed
, sv_changed
, nia
,
242 fetch_pc_ready_o
, fetch_pc_valid_i
,
243 fetch_insn_valid_o
, fetch_insn_ready_i
,
244 exec_insn_valid_i
, exec_insn_ready_o
,
245 exec_pc_valid_o
, exec_pc_ready_i
):
248 decode / issue FSM. this interacts with the "fetch" FSM
249 through fetch_insn_ready/valid (incoming) and fetch_pc_ready/valid
250 (outgoing). also interacts with the "execute" FSM
251 through exec_insn_ready/valid (outgoing) and exec_pc_ready/valid
253 SVP64 RM prefixes have already been set up by the
254 "fetch" phase, so execute is fairly straightforward.
259 pdecode2
= self
.pdecode2
260 cur_state
= self
.cur_state
263 dec_opcode_i
= pdecode2
.dec
.raw_opcode_in
# raw opcode
265 # for updating svstate (things like srcstep etc.)
266 update_svstate
= Signal() # set this (below) if updating
267 new_svstate
= SVSTATERec("new_svstate")
268 comb
+= new_svstate
.eq(cur_state
.svstate
)
270 with m
.FSM(name
="issue_fsm"):
272 # Wait on "core stop" release, at reset
273 with m
.State("WAIT_RESET"):
274 with m
.If(~dbg
.core_stop_o
& ~core_rst
):
275 m
.next
= "INSN_FETCH"
277 comb
+= core
.core_stopped_i
.eq(1)
278 comb
+= dbg
.core_stopped_i
.eq(1)
280 # go fetch the instruction at the current PC
281 # at this point, there is no instruction running, that
282 # could inadvertently update the PC.
283 with m
.State("INSN_FETCH"):
284 # TODO: update PC here, before fetch
285 comb
+= fetch_pc_valid_i
.eq(1)
286 with m
.If(fetch_pc_ready_o
):
289 # decode the instruction when it arrives
290 with m
.State("INSN_WAIT"):
291 comb
+= fetch_insn_ready_i
.eq(1)
292 with m
.If(fetch_insn_valid_o
):
293 # decode the instruction
294 sync
+= core
.e
.eq(pdecode2
.e
)
295 sync
+= core
.state
.eq(cur_state
)
296 sync
+= core
.raw_insn_i
.eq(dec_opcode_i
)
297 sync
+= core
.bigendian_i
.eq(self
.core_bigendian_i
)
298 # TODO: loop into INSN_FETCH if it's a vector instruction
299 # and VL == 0. this because VL==0 is a for-loop
300 # from 0 to 0 i.e. always, always a NOP.
301 m
.next
= "INSN_EXECUTE" # move to "execute"
303 with m
.State("INSN_EXECUTE"):
304 comb
+= exec_insn_valid_i
.eq(1)
305 with m
.If(exec_insn_ready_o
):
306 m
.next
= "EXECUTE_WAIT"
308 with m
.State("EXECUTE_WAIT"):
309 # wait on "core stop" release, at instruction end
310 with m
.If(~dbg
.core_stop_o
& ~core_rst
):
311 comb
+= exec_pc_ready_i
.eq(1)
312 with m
.If(exec_pc_valid_o
):
313 # precalculate srcstep+1
314 next_srcstep
= Signal
.like(cur_state
.svstate
.srcstep
)
315 comb
+= next_srcstep
.eq(cur_state
.svstate
.srcstep
+1)
316 # was this the last loop iteration?
318 cur_vl
= cur_state
.svstate
.vl
319 comb
+= is_last
.eq(next_srcstep
== cur_vl
)
321 # if either PC or SVSTATE were changed by the previous
322 # instruction, go directly back to Fetch, without
323 # updating either PC or SVSTATE
324 with m
.If(pc_changed | sv_changed
):
325 m
.next
= "INSN_FETCH"
327 # also return to Fetch, when no output was a vector
328 # (regardless of SRCSTEP and VL), or when the last
329 # instruction was really the last one of the VL loop
330 with m
.Elif(pdecode2
.no_out_vec | is_last
):
331 # before going back to fetch, update the PC state
332 # register with the NIA.
333 # ok here we are not reading the branch unit.
334 # TODO: this just blithely overwrites whatever
335 # pipeline updated the PC
336 comb
+= self
.state_w_pc
.wen
.eq(1 << StateRegs
.PC
)
337 comb
+= self
.state_w_pc
.data_i
.eq(nia
)
338 # reset SRCSTEP before returning to Fetch
339 with m
.If(~pdecode2
.no_out_vec
):
340 comb
+= new_svstate
.srcstep
.eq(0)
341 comb
+= update_svstate
.eq(1)
342 m
.next
= "INSN_FETCH"
344 # returning to Execute? then, first update SRCSTEP
346 comb
+= new_svstate
.srcstep
.eq(next_srcstep
)
347 comb
+= update_svstate
.eq(1)
351 comb
+= core
.core_stopped_i
.eq(1)
352 comb
+= dbg
.core_stopped_i
.eq(1)
354 # need to decode the instruction again, after updating SRCSTEP
355 # in the previous state.
356 # mostly a copy of INSN_WAIT, but without the actual wait
357 with m
.State("DECODE_SV"):
358 # decode the instruction
359 sync
+= core
.e
.eq(pdecode2
.e
)
360 sync
+= core
.state
.eq(cur_state
)
361 sync
+= core
.bigendian_i
.eq(self
.core_bigendian_i
)
362 m
.next
= "INSN_EXECUTE" # move to "execute"
364 # check if svstate needs updating: if so, write it to State Regfile
365 with m
.If(update_svstate
):
366 comb
+= self
.state_w_sv
.wen
.eq(1<<StateRegs
.SVSTATE
)
367 comb
+= self
.state_w_sv
.data_i
.eq(new_svstate
)
368 sync
+= cur_state
.svstate
.eq(new_svstate
) # for next clock
370 def execute_fsm(self
, m
, core
, insn_done
, pc_changed
, sv_changed
,
371 exec_insn_valid_i
, exec_insn_ready_o
,
372 exec_pc_valid_o
, exec_pc_ready_i
):
375 execute FSM. this interacts with the "issue" FSM
376 through exec_insn_ready/valid (incoming) and exec_pc_ready/valid
377 (outgoing). SVP64 RM prefixes have already been set up by the
378 "issue" phase, so execute is fairly straightforward.
383 pdecode2
= self
.pdecode2
387 core_busy_o
= core
.busy_o
# core is busy
388 core_ivalid_i
= core
.ivalid_i
# instruction is valid
389 core_issue_i
= core
.issue_i
# instruction is issued
390 insn_type
= core
.e
.do
.insn_type
# instruction MicroOp type
392 with m
.FSM(name
="exec_fsm"):
394 # waiting for instruction bus (stays there until not busy)
395 with m
.State("INSN_START"):
396 comb
+= exec_insn_ready_o
.eq(1)
397 with m
.If(exec_insn_valid_i
):
398 comb
+= core_ivalid_i
.eq(1) # instruction is valid
399 comb
+= core_issue_i
.eq(1) # and issued
400 sync
+= sv_changed
.eq(0)
401 sync
+= pc_changed
.eq(0)
402 m
.next
= "INSN_ACTIVE" # move to "wait completion"
404 # instruction started: must wait till it finishes
405 with m
.State("INSN_ACTIVE"):
406 with m
.If(insn_type
!= MicrOp
.OP_NOP
):
407 comb
+= core_ivalid_i
.eq(1) # instruction is valid
408 # note changes to PC and SVSTATE
409 with m
.If(self
.state_nia
.wen
& (1<<StateRegs
.SVSTATE
)):
410 sync
+= sv_changed
.eq(1)
411 with m
.If(self
.state_nia
.wen
& (1<<StateRegs
.PC
)):
412 sync
+= pc_changed
.eq(1)
413 with m
.If(~core_busy_o
): # instruction done!
414 comb
+= insn_done
.eq(1)
415 comb
+= exec_pc_valid_o
.eq(1)
416 with m
.If(exec_pc_ready_i
):
417 m
.next
= "INSN_START" # back to fetch
419 def elaborate(self
, platform
):
421 comb
, sync
= m
.d
.comb
, m
.d
.sync
423 m
.submodules
.core
= core
= DomainRenamer("coresync")(self
.core
)
424 m
.submodules
.imem
= imem
= self
.imem
425 m
.submodules
.dbg
= dbg
= self
.dbg
427 m
.submodules
.jtag
= jtag
= self
.jtag
428 # TODO: UART2GDB mux, here, from external pin
429 # see https://bugs.libre-soc.org/show_bug.cgi?id=499
430 sync
+= dbg
.dmi
.connect_to(jtag
.dmi
)
432 cur_state
= self
.cur_state
434 # 4x 4k SRAM blocks. these simply "exist", they get routed in litex
436 for i
, sram
in enumerate(self
.sram4k
):
437 m
.submodules
["sram4k_%d" % i
] = sram
438 comb
+= sram
.enable
.eq(self
.wb_sram_en
)
440 # XICS interrupt handler
442 m
.submodules
.xics_icp
= icp
= self
.xics_icp
443 m
.submodules
.xics_ics
= ics
= self
.xics_ics
444 comb
+= icp
.ics_i
.eq(ics
.icp_o
) # connect ICS to ICP
445 sync
+= cur_state
.eint
.eq(icp
.core_irq_o
) # connect ICP to core
447 # GPIO test peripheral
449 m
.submodules
.simple_gpio
= simple_gpio
= self
.simple_gpio
451 # connect one GPIO output to ICS bit 15 (like in microwatt soc.vhdl)
452 # XXX causes litex ECP5 test to get wrong idea about input and output
453 # (but works with verilator sim *sigh*)
454 #if self.gpio and self.xics:
455 # comb += self.int_level_i[15].eq(simple_gpio.gpio_o[0])
457 # instruction decoder
458 pdecode
= create_pdecode()
459 m
.submodules
.dec2
= pdecode2
= self
.pdecode2
460 m
.submodules
.svp64
= svp64
= self
.svp64
463 dmi
, d_reg
, d_cr
, d_xer
, = dbg
.dmi
, dbg
.d_gpr
, dbg
.d_cr
, dbg
.d_xer
464 intrf
= self
.core
.regs
.rf
['int']
466 # clock delay power-on reset
467 cd_por
= ClockDomain(reset_less
=True)
468 cd_sync
= ClockDomain()
469 core_sync
= ClockDomain("coresync")
470 m
.domains
+= cd_por
, cd_sync
, core_sync
472 ti_rst
= Signal(reset_less
=True)
473 delay
= Signal(range(4), reset
=3)
474 with m
.If(delay
!= 0):
475 m
.d
.por
+= delay
.eq(delay
- 1)
476 comb
+= cd_por
.clk
.eq(ClockSignal())
478 # power-on reset delay
479 core_rst
= ResetSignal("coresync")
480 comb
+= ti_rst
.eq(delay
!= 0 | dbg
.core_rst_o |
ResetSignal())
481 comb
+= core_rst
.eq(ti_rst
)
483 # busy/halted signals from core
484 comb
+= self
.busy_o
.eq(core
.busy_o
)
485 comb
+= pdecode2
.dec
.bigendian
.eq(self
.core_bigendian_i
)
487 # temporary hack: says "go" immediately for both address gen and ST
489 ldst
= core
.fus
.fus
['ldst0']
490 st_go_edge
= rising_edge(m
, ldst
.st
.rel_o
)
491 m
.d
.comb
+= ldst
.ad
.go_i
.eq(ldst
.ad
.rel_o
) # link addr-go direct to rel
492 m
.d
.comb
+= ldst
.st
.go_i
.eq(st_go_edge
) # link store-go to rising rel
494 # PC and instruction from I-Memory
495 comb
+= self
.pc_o
.eq(cur_state
.pc
)
496 pc_changed
= Signal() # note write to PC
497 sv_changed
= Signal() # note write to SVSTATE
498 insn_done
= Signal() # fires just once
501 pc
= Signal(64, reset_less
=True)
502 pc_ok_delay
= Signal()
503 sync
+= pc_ok_delay
.eq(~self
.pc_i
.ok
)
504 with m
.If(self
.pc_i
.ok
):
505 # incoming override (start from pc_i)
506 comb
+= pc
.eq(self
.pc_i
.data
)
508 # otherwise read StateRegs regfile for PC...
509 comb
+= self
.state_r_pc
.ren
.eq(1<<StateRegs
.PC
)
510 # ... but on a 1-clock delay
511 with m
.If(pc_ok_delay
):
512 comb
+= pc
.eq(self
.state_r_pc
.data_o
)
515 svstate
= Signal(64, reset_less
=True)
516 svstate_ok_delay
= Signal()
517 sync
+= svstate_ok_delay
.eq(~self
.svstate_i
.ok
)
518 with m
.If(self
.svstate_i
.ok
):
519 # incoming override (start from svstate__i)
520 comb
+= svstate
.eq(self
.svstate_i
.data
)
522 # otherwise read StateRegs regfile for SVSTATE...
523 comb
+= self
.state_r_sv
.ren
.eq(1 << StateRegs
.SVSTATE
)
524 # ... but on a 1-clock delay
525 with m
.If(svstate_ok_delay
):
526 comb
+= svstate
.eq(self
.state_r_sv
.data_o
)
528 # don't write pc every cycle
529 comb
+= self
.state_w_pc
.wen
.eq(0)
530 comb
+= self
.state_w_pc
.data_i
.eq(0)
532 # don't read msr every cycle
533 comb
+= self
.state_r_msr
.ren
.eq(0)
535 # address of the next instruction, in the absence of a branch
536 # depends on the instruction size
537 nia
= Signal(64, reset_less
=True)
539 # connect up debug signals
540 # TODO comb += core.icache_rst_i.eq(dbg.icache_rst_o)
541 comb
+= dbg
.terminate_i
.eq(core
.core_terminate_o
)
542 comb
+= dbg
.state
.pc
.eq(pc
)
543 comb
+= dbg
.state
.svstate
.eq(svstate
)
544 comb
+= dbg
.state
.msr
.eq(cur_state
.msr
)
546 # there are *TWO* FSMs, one fetch (32/64-bit) one decode/execute.
547 # these are the handshake signals between fetch and decode/execute
549 # fetch FSM can run as soon as the PC is valid
550 fetch_pc_valid_i
= Signal() # Execute tells Fetch "start next read"
551 fetch_pc_ready_o
= Signal() # Fetch Tells SVSTATE "proceed"
553 # fetch FSM hands over the instruction to be decoded / issued
554 fetch_insn_valid_o
= Signal()
555 fetch_insn_ready_i
= Signal()
557 # issue FSM delivers the instruction to the be executed
558 exec_insn_valid_i
= Signal()
559 exec_insn_ready_o
= Signal()
561 # execute FSM, hands over the PC/SVSTATE back to the issue FSM
562 exec_pc_valid_o
= Signal()
563 exec_pc_ready_i
= Signal()
565 # actually use a nmigen FSM for the first time (w00t)
566 # this FSM is perhaps unusual in that it detects conditions
567 # then "holds" information, combinatorially, for the core
568 # (as opposed to using sync - which would be on a clock's delay)
569 # this includes the actual opcode, valid flags and so on.
571 self
.fetch_fsm(m
, core
, pc
, svstate
, nia
,
572 fetch_pc_ready_o
, fetch_pc_valid_i
,
573 fetch_insn_valid_o
, fetch_insn_ready_i
)
575 # TODO: an SVSTATE-based for-loop FSM that goes in between
576 # fetch pc/insn ready/valid and advances SVSTATE.srcstep
577 # until it reaches VL-1 or PowerDecoder2.no_out_vec is True.
578 self
.issue_fsm(m
, core
, pc_changed
, sv_changed
, nia
,
580 fetch_pc_ready_o
, fetch_pc_valid_i
,
581 fetch_insn_valid_o
, fetch_insn_ready_i
,
582 exec_insn_valid_i
, exec_insn_ready_o
,
583 exec_pc_ready_i
, exec_pc_valid_o
)
585 self
.execute_fsm(m
, core
, insn_done
, pc_changed
, sv_changed
,
586 exec_insn_valid_i
, exec_insn_ready_o
,
587 exec_pc_ready_i
, exec_pc_valid_o
)
589 # this bit doesn't have to be in the FSM: connect up to read
590 # regfiles on demand from DMI
591 with m
.If(d_reg
.req
): # request for regfile access being made
592 # TODO: error-check this
593 # XXX should this be combinatorial? sync better?
595 comb
+= self
.int_r
.ren
.eq(1<<d_reg
.addr
)
597 comb
+= self
.int_r
.addr
.eq(d_reg
.addr
)
598 comb
+= self
.int_r
.ren
.eq(1)
599 d_reg_delay
= Signal()
600 sync
+= d_reg_delay
.eq(d_reg
.req
)
601 with m
.If(d_reg_delay
):
602 # data arrives one clock later
603 comb
+= d_reg
.data
.eq(self
.int_r
.data_o
)
604 comb
+= d_reg
.ack
.eq(1)
606 # sigh same thing for CR debug
607 with m
.If(d_cr
.req
): # request for regfile access being made
608 comb
+= self
.cr_r
.ren
.eq(0b11111111) # enable all
609 d_cr_delay
= Signal()
610 sync
+= d_cr_delay
.eq(d_cr
.req
)
611 with m
.If(d_cr_delay
):
612 # data arrives one clock later
613 comb
+= d_cr
.data
.eq(self
.cr_r
.data_o
)
614 comb
+= d_cr
.ack
.eq(1)
617 with m
.If(d_xer
.req
): # request for regfile access being made
618 comb
+= self
.xer_r
.ren
.eq(0b111111) # enable all
619 d_xer_delay
= Signal()
620 sync
+= d_xer_delay
.eq(d_xer
.req
)
621 with m
.If(d_xer_delay
):
622 # data arrives one clock later
623 comb
+= d_xer
.data
.eq(self
.xer_r
.data_o
)
624 comb
+= d_xer
.ack
.eq(1)
626 # DEC and TB inc/dec FSM. copy of DEC is put into CoreState,
627 # (which uses that in PowerDecoder2 to raise 0x900 exception)
628 self
.tb_dec_fsm(m
, cur_state
.dec
)
632 def tb_dec_fsm(self
, m
, spr_dec
):
635 this is a FSM for updating either dec or tb. it runs alternately
636 DEC, TB, DEC, TB. note that SPR pipeline could have written a new
637 value to DEC, however the regfile has "passthrough" on it so this
640 see v3.0B p1097-1099 for Timeer Resource and p1065 and p1076
643 comb
, sync
= m
.d
.comb
, m
.d
.sync
644 fast_rf
= self
.core
.regs
.rf
['fast']
645 fast_r_dectb
= fast_rf
.r_ports
['issue'] # DEC/TB
646 fast_w_dectb
= fast_rf
.w_ports
['issue'] # DEC/TB
650 # initiates read of current DEC
651 with m
.State("DEC_READ"):
652 comb
+= fast_r_dectb
.addr
.eq(FastRegs
.DEC
)
653 comb
+= fast_r_dectb
.ren
.eq(1)
656 # waits for DEC read to arrive (1 cycle), updates with new value
657 with m
.State("DEC_WRITE"):
659 # TODO: MSR.LPCR 32-bit decrement mode
660 comb
+= new_dec
.eq(fast_r_dectb
.data_o
- 1)
661 comb
+= fast_w_dectb
.addr
.eq(FastRegs
.DEC
)
662 comb
+= fast_w_dectb
.wen
.eq(1)
663 comb
+= fast_w_dectb
.data_i
.eq(new_dec
)
664 sync
+= spr_dec
.eq(new_dec
) # copy into cur_state for decoder
667 # initiates read of current TB
668 with m
.State("TB_READ"):
669 comb
+= fast_r_dectb
.addr
.eq(FastRegs
.TB
)
670 comb
+= fast_r_dectb
.ren
.eq(1)
673 # waits for read TB to arrive, initiates write of current TB
674 with m
.State("TB_WRITE"):
676 comb
+= new_tb
.eq(fast_r_dectb
.data_o
+ 1)
677 comb
+= fast_w_dectb
.addr
.eq(FastRegs
.TB
)
678 comb
+= fast_w_dectb
.wen
.eq(1)
679 comb
+= fast_w_dectb
.data_i
.eq(new_tb
)
685 yield from self
.pc_i
.ports()
688 yield from self
.core
.ports()
689 yield from self
.imem
.ports()
690 yield self
.core_bigendian_i
696 def external_ports(self
):
697 ports
= self
.pc_i
.ports()
698 ports
+= [self
.pc_o
, self
.memerr_o
, self
.core_bigendian_i
, self
.busy_o
,
702 ports
+= list(self
.jtag
.external_ports())
704 # don't add DMI if JTAG is enabled
705 ports
+= list(self
.dbg
.dmi
.ports())
707 ports
+= list(self
.imem
.ibus
.fields
.values())
708 ports
+= list(self
.core
.l0
.cmpi
.lsmem
.lsi
.slavebus
.fields
.values())
711 for sram
in self
.sram4k
:
712 ports
+= list(sram
.bus
.fields
.values())
715 ports
+= list(self
.xics_icp
.bus
.fields
.values())
716 ports
+= list(self
.xics_ics
.bus
.fields
.values())
717 ports
.append(self
.int_level_i
)
720 ports
+= list(self
.simple_gpio
.bus
.fields
.values())
721 ports
.append(self
.gpio_o
)
729 class TestIssuer(Elaboratable
):
730 def __init__(self
, pspec
):
731 self
.ti
= TestIssuerInternal(pspec
)
733 self
.pll
= DummyPLL()
735 # PLL direct clock or not
736 self
.pll_en
= hasattr(pspec
, "use_pll") and pspec
.use_pll
738 self
.pll_18_o
= Signal(reset_less
=True)
740 def elaborate(self
, platform
):
744 # TestIssuer runs at direct clock
745 m
.submodules
.ti
= ti
= self
.ti
746 cd_int
= ClockDomain("coresync")
749 # ClockSelect runs at PLL output internal clock rate
750 m
.submodules
.pll
= pll
= self
.pll
752 # add clock domains from PLL
753 cd_pll
= ClockDomain("pllclk")
756 # PLL clock established. has the side-effect of running clklsel
757 # at the PLL's speed (see DomainRenamer("pllclk") above)
758 pllclk
= ClockSignal("pllclk")
759 comb
+= pllclk
.eq(pll
.clk_pll_o
)
761 # wire up external 24mhz to PLL
762 comb
+= pll
.clk_24_i
.eq(ClockSignal())
764 # output 18 mhz PLL test signal
765 comb
+= self
.pll_18_o
.eq(pll
.pll_18_o
)
767 # now wire up ResetSignals. don't mind them being in this domain
768 pll_rst
= ResetSignal("pllclk")
769 comb
+= pll_rst
.eq(ResetSignal())
771 # internal clock is set to selector clock-out. has the side-effect of
772 # running TestIssuer at this speed (see DomainRenamer("intclk") above)
773 intclk
= ClockSignal("coresync")
775 comb
+= intclk
.eq(pll
.clk_pll_o
)
777 comb
+= intclk
.eq(ClockSignal())
782 return list(self
.ti
.ports()) + list(self
.pll
.ports()) + \
783 [ClockSignal(), ResetSignal()]
785 def external_ports(self
):
786 ports
= self
.ti
.external_ports()
787 ports
.append(ClockSignal())
788 ports
.append(ResetSignal())
790 ports
.append(self
.pll
.clk_sel_i
)
791 ports
.append(self
.pll_18_o
)
792 ports
.append(self
.pll
.pll_lck_o
)
796 if __name__
== '__main__':
797 units
= {'alu': 1, 'cr': 1, 'branch': 1, 'trap': 1, 'logical': 1,
803 pspec
= TestMemPspec(ldst_ifacetype
='bare_wb',
804 imem_ifacetype
='bare_wb',
809 dut
= TestIssuer(pspec
)
810 vl
= main(dut
, ports
=dut
.ports(), name
="test_issuer")
812 if len(sys
.argv
) == 1:
813 vl
= rtlil
.convert(dut
, ports
=dut
.external_ports(), name
="test_issuer")
814 with
open("test_issuer.il", "w") as f
: