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
))
90 # add interrupt controller?
91 self
.xics
= hasattr(pspec
, "xics") and pspec
.xics
== True
93 self
.xics_icp
= XICS_ICP()
94 self
.xics_ics
= XICS_ICS()
95 self
.int_level_i
= self
.xics_ics
.int_level_i
97 # add GPIO peripheral?
98 self
.gpio
= hasattr(pspec
, "gpio") and pspec
.gpio
== True
100 self
.simple_gpio
= SimpleGPIO()
101 self
.gpio_o
= self
.simple_gpio
.gpio_o
103 # main instruction core25
104 self
.core
= core
= NonProductionCore(pspec
)
106 # instruction decoder. goes into Trap Record
107 pdecode
= create_pdecode()
108 self
.cur_state
= CoreState("cur") # current state (MSR/PC/EINT/SVSTATE)
109 self
.pdecode2
= PowerDecode2(pdecode
, state
=self
.cur_state
,
110 opkls
=IssuerDecode2ToOperand
)
111 self
.svp64
= SVP64PrefixDecoder() # for decoding SVP64 prefix
113 # Test Instruction memory
114 self
.imem
= ConfigFetchUnit(pspec
).fu
115 # one-row cache of instruction read
116 self
.iline
= Signal(64) # one instruction line
117 self
.iprev_adr
= Signal(64) # previous address: if different, do read
120 self
.dbg
= CoreDebug()
122 # instruction go/monitor
123 self
.pc_o
= Signal(64, reset_less
=True)
124 self
.pc_i
= Data(64, "pc_i") # set "ok" to indicate "please change me"
125 self
.svstate_i
= Data(32, "svstate_i") # ditto
126 self
.core_bigendian_i
= Signal()
127 self
.busy_o
= Signal(reset_less
=True)
128 self
.memerr_o
= Signal(reset_less
=True)
130 # STATE regfile read /write ports for PC, MSR, SVSTATE
131 staterf
= self
.core
.regs
.rf
['state']
132 self
.state_r_pc
= staterf
.r_ports
['cia'] # PC rd
133 self
.state_w_pc
= staterf
.w_ports
['d_wr1'] # PC wr
134 self
.state_r_msr
= staterf
.r_ports
['msr'] # MSR rd
135 self
.state_r_sv
= staterf
.r_ports
['sv'] # SVSTATE rd
136 self
.state_w_sv
= staterf
.w_ports
['sv'] # SVSTATE wr
138 # DMI interface access
139 intrf
= self
.core
.regs
.rf
['int']
140 crrf
= self
.core
.regs
.rf
['cr']
141 xerrf
= self
.core
.regs
.rf
['xer']
142 self
.int_r
= intrf
.r_ports
['dmi'] # INT read
143 self
.cr_r
= crrf
.r_ports
['full_cr_dbg'] # CR read
144 self
.xer_r
= xerrf
.r_ports
['full_xer'] # XER read
146 # hack method of keeping an eye on whether branch/trap set the PC
147 self
.state_nia
= self
.core
.regs
.rf
['state'].w_ports
['nia']
148 self
.state_nia
.wen
.name
= 'state_nia_wen'
150 def fetch_fsm(self
, m
, core
, pc
, svstate
, pc_changed
, insn_done
,
151 fetch_pc_ready_o
, fetch_pc_valid_i
,
152 fetch_insn_valid_o
, fetch_insn_ready_i
):
154 this FSM performs fetch of raw instruction data, partial-decodes
155 it 32-bit at a time to detect SVP64 prefixes, and will optionally
156 read a 2nd 32-bit quantity if that occurs.
160 pdecode2
= self
.pdecode2
162 cur_state
= self
.cur_state
164 # latches copy of raw fetched instruction
165 fetch_insn_o
= Signal(32, reset_less
=True)
166 dec_opcode_i
= pdecode2
.dec
.raw_opcode_in
# raw opcode
167 sync
+= dec_opcode_i
.eq(fetch_insn_o
) # actual opcode
169 msr_read
= Signal(reset
=1)
171 # address of the next instruction, in the absence of a branch
172 # depends on the instruction size
173 nia
= Signal(64, reset_less
=True)
175 with m
.FSM(name
='fetch_fsm'):
178 with m
.State("IDLE"):
179 comb
+= fetch_pc_ready_o
.eq(1)
180 with m
.If(fetch_pc_valid_i
):
181 # instruction allowed to go: start by reading the PC
182 # capture the PC and also drop it into Insn Memory
183 # we have joined a pair of combinatorial memory
184 # lookups together. this is Generally Bad.
185 comb
+= self
.imem
.a_pc_i
.eq(pc
)
186 comb
+= self
.imem
.a_valid_i
.eq(1)
187 comb
+= self
.imem
.f_valid_i
.eq(1)
188 sync
+= cur_state
.pc
.eq(pc
)
189 sync
+= cur_state
.svstate
.eq(svstate
) # and svstate
191 # initiate read of MSR. arrives one clock later
192 comb
+= self
.state_r_msr
.ren
.eq(1 << StateRegs
.MSR
)
193 sync
+= msr_read
.eq(0)
195 m
.next
= "INSN_READ" # move to "wait for bus" phase
197 # dummy pause to find out why simulation is not keeping up
198 with m
.State("INSN_READ"):
199 # one cycle later, msr/sv read arrives. valid only once.
200 with m
.If(~msr_read
):
201 sync
+= msr_read
.eq(1) # yeah don't read it again
202 sync
+= cur_state
.msr
.eq(self
.state_r_msr
.data_o
)
203 with m
.If(self
.imem
.f_busy_o
): # zzz...
204 # busy: stay in wait-read
205 comb
+= self
.imem
.a_valid_i
.eq(1)
206 comb
+= self
.imem
.f_valid_i
.eq(1)
208 # not busy: instruction fetched
209 insn
= get_insn(self
.imem
.f_instr_o
, cur_state
.pc
)
210 # decode the SVP64 prefix, if any
211 comb
+= svp64
.raw_opcode_in
.eq(insn
)
212 comb
+= svp64
.bigendian
.eq(self
.core_bigendian_i
)
213 # pass the decoded prefix (if any) to PowerDecoder2
214 sync
+= pdecode2
.sv_rm
.eq(svp64
.svp64_rm
)
215 # calculate the address of the following instruction
216 insn_size
= Mux(svp64
.is_svp64_mode
, 8, 4)
217 sync
+= nia
.eq(cur_state
.pc
+ insn_size
)
218 with m
.If(~svp64
.is_svp64_mode
):
219 # with no prefix, store the instruction
220 # and hand it directly to the next FSM
221 comb
+= fetch_insn_o
.eq(insn
)
222 m
.next
= "INSN_READY"
224 # fetch the rest of the instruction from memory
225 comb
+= self
.imem
.a_pc_i
.eq(cur_state
.pc
+ 4)
226 comb
+= self
.imem
.a_valid_i
.eq(1)
227 comb
+= self
.imem
.f_valid_i
.eq(1)
228 m
.next
= "INSN_READ2"
230 with m
.State("INSN_READ2"):
231 with m
.If(self
.imem
.f_busy_o
): # zzz...
232 # busy: stay in wait-read
233 comb
+= self
.imem
.a_valid_i
.eq(1)
234 comb
+= self
.imem
.f_valid_i
.eq(1)
236 # not busy: instruction fetched
237 insn
= get_insn(self
.imem
.f_instr_o
, cur_state
.pc
+4)
238 comb
+= fetch_insn_o
.eq(insn
)
239 m
.next
= "INSN_READY"
241 with m
.State("INSN_READY"):
242 # hand over the instruction, to be decoded
243 comb
+= fetch_insn_valid_o
.eq(1)
244 with m
.If(fetch_insn_ready_i
):
247 # code-morph: moving the actual PC-setting out of "execute"
248 # so that it's easier to move this into an "issue" FSM.
250 # ok here we are not reading the branch unit. TODO
251 # this just blithely overwrites whatever pipeline
253 core_busy_o
= core
.busy_o
# core is busy
254 with m
.If(insn_done
& (~pc_changed
) & (~core_busy_o
)):
255 comb
+= self
.state_w_pc
.wen
.eq(1<<StateRegs
.PC
)
256 comb
+= self
.state_w_pc
.data_i
.eq(nia
)
258 def issue_fsm(self
, m
, core
, pc_changed
, sv_changed
,
260 fetch_pc_ready_o
, fetch_pc_valid_i
,
261 fetch_insn_valid_o
, fetch_insn_ready_i
,
262 exec_insn_valid_i
, exec_insn_ready_o
,
263 exec_pc_valid_o
, exec_pc_ready_i
):
266 decode / issue FSM. this interacts with the "fetch" FSM
267 through fetch_insn_ready/valid (incoming) and fetch_pc_ready/valid
268 (outgoing). also interacts with the "execute" FSM
269 through exec_insn_ready/valid (outgoing) and exec_pc_ready/valid
271 SVP64 RM prefixes have already been set up by the
272 "fetch" phase, so execute is fairly straightforward.
277 pdecode2
= self
.pdecode2
278 cur_state
= self
.cur_state
281 dec_opcode_i
= pdecode2
.dec
.raw_opcode_in
# raw opcode
283 # for updating svstate (things like srcstep etc.)
284 update_svstate
= Signal() # set this (below) if updating
285 new_svstate
= SVSTATERec("new_svstate")
286 comb
+= new_svstate
.eq(cur_state
.svstate
)
288 with m
.FSM(name
="issue_fsm"):
290 # Wait on "core stop" release, at reset
291 with m
.State("WAIT_RESET"):
292 with m
.If(~dbg
.core_stop_o
& ~core_rst
):
293 m
.next
= "INSN_FETCH"
295 comb
+= core
.core_stopped_i
.eq(1)
296 comb
+= dbg
.core_stopped_i
.eq(1)
298 # go fetch the instruction at the current PC
299 # at this point, there is no instruction running, that
300 # could inadvertently update the PC.
301 with m
.State("INSN_FETCH"):
302 # TODO: update PC here, before fetch
303 comb
+= fetch_pc_valid_i
.eq(1)
304 with m
.If(fetch_pc_ready_o
):
307 # decode the instruction when it arrives
308 with m
.State("INSN_WAIT"):
309 comb
+= fetch_insn_ready_i
.eq(1)
310 with m
.If(fetch_insn_valid_o
):
311 # decode the instruction
312 sync
+= core
.e
.eq(pdecode2
.e
)
313 sync
+= core
.state
.eq(cur_state
)
314 sync
+= core
.raw_insn_i
.eq(dec_opcode_i
)
315 sync
+= core
.bigendian_i
.eq(self
.core_bigendian_i
)
316 # TODO: loop into INSN_FETCH if it's a vector instruction
317 # and VL == 0. this because VL==0 is a for-loop
318 # from 0 to 0 i.e. always, always a NOP.
319 m
.next
= "INSN_EXECUTE" # move to "execute"
321 with m
.State("INSN_EXECUTE"):
322 comb
+= exec_insn_valid_i
.eq(1)
323 with m
.If(exec_insn_ready_o
):
324 m
.next
= "EXECUTE_WAIT"
326 with m
.State("EXECUTE_WAIT"):
327 # wait on "core stop" release, at instruction end
328 with m
.If(~dbg
.core_stop_o
& ~core_rst
):
329 comb
+= exec_pc_ready_i
.eq(1)
330 with m
.If(exec_pc_valid_o
):
331 # TODO: update SRCSTEP here (in new_svstate)
332 # and set update_svstate to True *as long as*
333 # PC / SVSTATE was not modified. that's an
334 # exception (or setvl was called)
335 # TODO: loop into INSN_EXECUTE if it's a vector
336 # instruction and SRCSTEP != VL-1 and
337 # PowerDecoder.no_out_vec is True
338 # unless PC / SVSTATE was modified, in that
339 # case do go back to INSN_FETCH.
340 m
.next
= "INSN_FETCH"
342 comb
+= core
.core_stopped_i
.eq(1)
343 comb
+= dbg
.core_stopped_i
.eq(1)
345 # check if svstate needs updating: if so, write it to State Regfile
346 with m
.If(update_svstate
):
347 comb
+= self
.state_w_sv
.wen
.eq(1<<StateRegs
.SVSTATE
)
348 comb
+= self
.state_w_sv
.data_i
.eq(new_svstate
)
349 sync
+= cur_state
.svstate
.eq(new_svstate
) # for next clock
351 def execute_fsm(self
, m
, core
, insn_done
, pc_changed
, sv_changed
,
352 exec_insn_valid_i
, exec_insn_ready_o
,
353 exec_pc_valid_o
, exec_pc_ready_i
):
356 execute FSM. this interacts with the "issue" FSM
357 through exec_insn_ready/valid (incoming) and exec_pc_ready/valid
358 (outgoing). SVP64 RM prefixes have already been set up by the
359 "issue" phase, so execute is fairly straightforward.
364 pdecode2
= self
.pdecode2
368 core_busy_o
= core
.busy_o
# core is busy
369 core_ivalid_i
= core
.ivalid_i
# instruction is valid
370 core_issue_i
= core
.issue_i
# instruction is issued
371 insn_type
= core
.e
.do
.insn_type
# instruction MicroOp type
373 with m
.FSM(name
="exec_fsm"):
375 # waiting for instruction bus (stays there until not busy)
376 with m
.State("INSN_START"):
377 comb
+= exec_insn_ready_o
.eq(1)
378 with m
.If(exec_insn_valid_i
):
379 comb
+= core_ivalid_i
.eq(1) # instruction is valid
380 comb
+= core_issue_i
.eq(1) # and issued
381 m
.next
= "INSN_ACTIVE" # move to "wait completion"
383 # instruction started: must wait till it finishes
384 with m
.State("INSN_ACTIVE"):
385 with m
.If(insn_type
!= MicrOp
.OP_NOP
):
386 comb
+= core_ivalid_i
.eq(1) # instruction is valid
387 # note changes to PC and SVSTATE
388 with m
.If(self
.state_nia
.wen
& (1<<StateRegs
.SVSTATE
)):
389 sync
+= sv_changed
.eq(1)
390 with m
.If(self
.state_nia
.wen
& (1<<StateRegs
.PC
)):
391 sync
+= pc_changed
.eq(1)
392 with m
.If(~core_busy_o
): # instruction done!
393 comb
+= insn_done
.eq(1)
395 sync
+= core
.raw_insn_i
.eq(0)
396 sync
+= core
.bigendian_i
.eq(0)
397 sync
+= sv_changed
.eq(0)
398 sync
+= pc_changed
.eq(0)
399 comb
+= exec_pc_valid_o
.eq(1)
400 with m
.If(exec_pc_ready_i
):
401 m
.next
= "INSN_START" # back to fetch
403 def elaborate(self
, platform
):
405 comb
, sync
= m
.d
.comb
, m
.d
.sync
407 m
.submodules
.core
= core
= DomainRenamer("coresync")(self
.core
)
408 m
.submodules
.imem
= imem
= self
.imem
409 m
.submodules
.dbg
= dbg
= self
.dbg
411 m
.submodules
.jtag
= jtag
= self
.jtag
412 # TODO: UART2GDB mux, here, from external pin
413 # see https://bugs.libre-soc.org/show_bug.cgi?id=499
414 sync
+= dbg
.dmi
.connect_to(jtag
.dmi
)
416 cur_state
= self
.cur_state
418 # 4x 4k SRAM blocks. these simply "exist", they get routed in litex
420 for i
, sram
in enumerate(self
.sram4k
):
421 m
.submodules
["sram4k_%d" % i
] = sram
422 comb
+= sram
.enable
.eq(self
.wb_sram_en
)
424 # XICS interrupt handler
426 m
.submodules
.xics_icp
= icp
= self
.xics_icp
427 m
.submodules
.xics_ics
= ics
= self
.xics_ics
428 comb
+= icp
.ics_i
.eq(ics
.icp_o
) # connect ICS to ICP
429 sync
+= cur_state
.eint
.eq(icp
.core_irq_o
) # connect ICP to core
431 # GPIO test peripheral
433 m
.submodules
.simple_gpio
= simple_gpio
= self
.simple_gpio
435 # connect one GPIO output to ICS bit 15 (like in microwatt soc.vhdl)
436 # XXX causes litex ECP5 test to get wrong idea about input and output
437 # (but works with verilator sim *sigh*)
438 #if self.gpio and self.xics:
439 # comb += self.int_level_i[15].eq(simple_gpio.gpio_o[0])
441 # instruction decoder
442 pdecode
= create_pdecode()
443 m
.submodules
.dec2
= pdecode2
= self
.pdecode2
444 m
.submodules
.svp64
= svp64
= self
.svp64
447 dmi
, d_reg
, d_cr
, d_xer
, = dbg
.dmi
, dbg
.d_gpr
, dbg
.d_cr
, dbg
.d_xer
448 intrf
= self
.core
.regs
.rf
['int']
450 # clock delay power-on reset
451 cd_por
= ClockDomain(reset_less
=True)
452 cd_sync
= ClockDomain()
453 core_sync
= ClockDomain("coresync")
454 m
.domains
+= cd_por
, cd_sync
, core_sync
456 ti_rst
= Signal(reset_less
=True)
457 delay
= Signal(range(4), reset
=3)
458 with m
.If(delay
!= 0):
459 m
.d
.por
+= delay
.eq(delay
- 1)
460 comb
+= cd_por
.clk
.eq(ClockSignal())
462 # power-on reset delay
463 core_rst
= ResetSignal("coresync")
464 comb
+= ti_rst
.eq(delay
!= 0 | dbg
.core_rst_o |
ResetSignal())
465 comb
+= core_rst
.eq(ti_rst
)
467 # busy/halted signals from core
468 comb
+= self
.busy_o
.eq(core
.busy_o
)
469 comb
+= pdecode2
.dec
.bigendian
.eq(self
.core_bigendian_i
)
471 # temporary hack: says "go" immediately for both address gen and ST
473 ldst
= core
.fus
.fus
['ldst0']
474 st_go_edge
= rising_edge(m
, ldst
.st
.rel_o
)
475 m
.d
.comb
+= ldst
.ad
.go_i
.eq(ldst
.ad
.rel_o
) # link addr-go direct to rel
476 m
.d
.comb
+= ldst
.st
.go_i
.eq(st_go_edge
) # link store-go to rising rel
478 # PC and instruction from I-Memory
479 comb
+= self
.pc_o
.eq(cur_state
.pc
)
480 pc_changed
= Signal() # note write to PC
481 sv_changed
= Signal() # note write to SVSTATE
482 insn_done
= Signal() # fires just once
485 pc
= Signal(64, reset_less
=True)
486 pc_ok_delay
= Signal()
487 sync
+= pc_ok_delay
.eq(~self
.pc_i
.ok
)
488 with m
.If(self
.pc_i
.ok
):
489 # incoming override (start from pc_i)
490 comb
+= pc
.eq(self
.pc_i
.data
)
492 # otherwise read StateRegs regfile for PC...
493 comb
+= self
.state_r_pc
.ren
.eq(1<<StateRegs
.PC
)
494 # ... but on a 1-clock delay
495 with m
.If(pc_ok_delay
):
496 comb
+= pc
.eq(self
.state_r_pc
.data_o
)
499 svstate
= Signal(64, reset_less
=True)
500 svstate_ok_delay
= Signal()
501 sync
+= svstate_ok_delay
.eq(~self
.svstate_i
.ok
)
502 with m
.If(self
.svstate_i
.ok
):
503 # incoming override (start from svstate__i)
504 comb
+= svstate
.eq(self
.svstate_i
.data
)
506 # otherwise read StateRegs regfile for SVSTATE...
507 comb
+= self
.state_r_sv
.ren
.eq(1 << StateRegs
.SVSTATE
)
508 # ... but on a 1-clock delay
509 with m
.If(svstate_ok_delay
):
510 comb
+= svstate
.eq(self
.state_r_sv
.data_o
)
512 # don't write pc every cycle
513 comb
+= self
.state_w_pc
.wen
.eq(0)
514 comb
+= self
.state_w_pc
.data_i
.eq(0)
516 # don't read msr every cycle
517 comb
+= self
.state_r_msr
.ren
.eq(0)
519 # connect up debug signals
520 # TODO comb += core.icache_rst_i.eq(dbg.icache_rst_o)
521 comb
+= dbg
.terminate_i
.eq(core
.core_terminate_o
)
522 comb
+= dbg
.state
.pc
.eq(pc
)
523 comb
+= dbg
.state
.svstate
.eq(svstate
)
524 comb
+= dbg
.state
.msr
.eq(cur_state
.msr
)
526 # there are *TWO* FSMs, one fetch (32/64-bit) one decode/execute.
527 # these are the handshake signals between fetch and decode/execute
529 # fetch FSM can run as soon as the PC is valid
530 fetch_pc_valid_i
= Signal() # Execute tells Fetch "start next read"
531 fetch_pc_ready_o
= Signal() # Fetch Tells SVSTATE "proceed"
533 # fetch FSM hands over the instruction to be decoded / issued
534 fetch_insn_valid_o
= Signal()
535 fetch_insn_ready_i
= Signal()
537 # issue FSM delivers the instruction to the be executed
538 exec_insn_valid_i
= Signal()
539 exec_insn_ready_o
= Signal()
541 # execute FSM, hands over the PC/SVSTATE back to the issue FSM
542 exec_pc_valid_o
= Signal()
543 exec_pc_ready_i
= Signal()
545 # actually use a nmigen FSM for the first time (w00t)
546 # this FSM is perhaps unusual in that it detects conditions
547 # then "holds" information, combinatorially, for the core
548 # (as opposed to using sync - which would be on a clock's delay)
549 # this includes the actual opcode, valid flags and so on.
551 self
.fetch_fsm(m
, core
, pc
, svstate
, pc_changed
, insn_done
,
552 fetch_pc_ready_o
, fetch_pc_valid_i
,
553 fetch_insn_valid_o
, fetch_insn_ready_i
)
555 # TODO: an SVSTATE-based for-loop FSM that goes in between
556 # fetch pc/insn ready/valid and advances SVSTATE.srcstep
557 # until it reaches VL-1 or PowerDecoder2.no_out_vec is True.
558 self
.issue_fsm(m
, core
, pc_changed
, sv_changed
,
560 fetch_pc_ready_o
, fetch_pc_valid_i
,
561 fetch_insn_valid_o
, fetch_insn_ready_i
,
562 exec_insn_valid_i
, exec_insn_ready_o
,
563 exec_pc_ready_i
, exec_pc_valid_o
)
565 self
.execute_fsm(m
, core
, insn_done
, pc_changed
, sv_changed
,
566 exec_insn_valid_i
, exec_insn_ready_o
,
567 exec_pc_ready_i
, exec_pc_valid_o
)
569 # this bit doesn't have to be in the FSM: connect up to read
570 # regfiles on demand from DMI
571 with m
.If(d_reg
.req
): # request for regfile access being made
572 # TODO: error-check this
573 # XXX should this be combinatorial? sync better?
575 comb
+= self
.int_r
.ren
.eq(1<<d_reg
.addr
)
577 comb
+= self
.int_r
.addr
.eq(d_reg
.addr
)
578 comb
+= self
.int_r
.ren
.eq(1)
579 d_reg_delay
= Signal()
580 sync
+= d_reg_delay
.eq(d_reg
.req
)
581 with m
.If(d_reg_delay
):
582 # data arrives one clock later
583 comb
+= d_reg
.data
.eq(self
.int_r
.data_o
)
584 comb
+= d_reg
.ack
.eq(1)
586 # sigh same thing for CR debug
587 with m
.If(d_cr
.req
): # request for regfile access being made
588 comb
+= self
.cr_r
.ren
.eq(0b11111111) # enable all
589 d_cr_delay
= Signal()
590 sync
+= d_cr_delay
.eq(d_cr
.req
)
591 with m
.If(d_cr_delay
):
592 # data arrives one clock later
593 comb
+= d_cr
.data
.eq(self
.cr_r
.data_o
)
594 comb
+= d_cr
.ack
.eq(1)
597 with m
.If(d_xer
.req
): # request for regfile access being made
598 comb
+= self
.xer_r
.ren
.eq(0b111111) # enable all
599 d_xer_delay
= Signal()
600 sync
+= d_xer_delay
.eq(d_xer
.req
)
601 with m
.If(d_xer_delay
):
602 # data arrives one clock later
603 comb
+= d_xer
.data
.eq(self
.xer_r
.data_o
)
604 comb
+= d_xer
.ack
.eq(1)
606 # DEC and TB inc/dec FSM. copy of DEC is put into CoreState,
607 # (which uses that in PowerDecoder2 to raise 0x900 exception)
608 self
.tb_dec_fsm(m
, cur_state
.dec
)
612 def tb_dec_fsm(self
, m
, spr_dec
):
615 this is a FSM for updating either dec or tb. it runs alternately
616 DEC, TB, DEC, TB. note that SPR pipeline could have written a new
617 value to DEC, however the regfile has "passthrough" on it so this
620 see v3.0B p1097-1099 for Timeer Resource and p1065 and p1076
623 comb
, sync
= m
.d
.comb
, m
.d
.sync
624 fast_rf
= self
.core
.regs
.rf
['fast']
625 fast_r_dectb
= fast_rf
.r_ports
['issue'] # DEC/TB
626 fast_w_dectb
= fast_rf
.w_ports
['issue'] # DEC/TB
630 # initiates read of current DEC
631 with m
.State("DEC_READ"):
632 comb
+= fast_r_dectb
.addr
.eq(FastRegs
.DEC
)
633 comb
+= fast_r_dectb
.ren
.eq(1)
636 # waits for DEC read to arrive (1 cycle), updates with new value
637 with m
.State("DEC_WRITE"):
639 # TODO: MSR.LPCR 32-bit decrement mode
640 comb
+= new_dec
.eq(fast_r_dectb
.data_o
- 1)
641 comb
+= fast_w_dectb
.addr
.eq(FastRegs
.DEC
)
642 comb
+= fast_w_dectb
.wen
.eq(1)
643 comb
+= fast_w_dectb
.data_i
.eq(new_dec
)
644 sync
+= spr_dec
.eq(new_dec
) # copy into cur_state for decoder
647 # initiates read of current TB
648 with m
.State("TB_READ"):
649 comb
+= fast_r_dectb
.addr
.eq(FastRegs
.TB
)
650 comb
+= fast_r_dectb
.ren
.eq(1)
653 # waits for read TB to arrive, initiates write of current TB
654 with m
.State("TB_WRITE"):
656 comb
+= new_tb
.eq(fast_r_dectb
.data_o
+ 1)
657 comb
+= fast_w_dectb
.addr
.eq(FastRegs
.TB
)
658 comb
+= fast_w_dectb
.wen
.eq(1)
659 comb
+= fast_w_dectb
.data_i
.eq(new_tb
)
665 yield from self
.pc_i
.ports()
668 yield from self
.core
.ports()
669 yield from self
.imem
.ports()
670 yield self
.core_bigendian_i
676 def external_ports(self
):
677 ports
= self
.pc_i
.ports()
678 ports
+= [self
.pc_o
, self
.memerr_o
, self
.core_bigendian_i
, self
.busy_o
,
682 ports
+= list(self
.jtag
.external_ports())
684 # don't add DMI if JTAG is enabled
685 ports
+= list(self
.dbg
.dmi
.ports())
687 ports
+= list(self
.imem
.ibus
.fields
.values())
688 ports
+= list(self
.core
.l0
.cmpi
.lsmem
.lsi
.slavebus
.fields
.values())
691 for sram
in self
.sram4k
:
692 ports
+= list(sram
.bus
.fields
.values())
695 ports
+= list(self
.xics_icp
.bus
.fields
.values())
696 ports
+= list(self
.xics_ics
.bus
.fields
.values())
697 ports
.append(self
.int_level_i
)
700 ports
+= list(self
.simple_gpio
.bus
.fields
.values())
701 ports
.append(self
.gpio_o
)
709 class TestIssuer(Elaboratable
):
710 def __init__(self
, pspec
):
711 self
.ti
= TestIssuerInternal(pspec
)
713 self
.pll
= DummyPLL()
715 # PLL direct clock or not
716 self
.pll_en
= hasattr(pspec
, "use_pll") and pspec
.use_pll
718 self
.pll_18_o
= Signal(reset_less
=True)
720 def elaborate(self
, platform
):
724 # TestIssuer runs at direct clock
725 m
.submodules
.ti
= ti
= self
.ti
726 cd_int
= ClockDomain("coresync")
729 # ClockSelect runs at PLL output internal clock rate
730 m
.submodules
.pll
= pll
= self
.pll
732 # add clock domains from PLL
733 cd_pll
= ClockDomain("pllclk")
736 # PLL clock established. has the side-effect of running clklsel
737 # at the PLL's speed (see DomainRenamer("pllclk") above)
738 pllclk
= ClockSignal("pllclk")
739 comb
+= pllclk
.eq(pll
.clk_pll_o
)
741 # wire up external 24mhz to PLL
742 comb
+= pll
.clk_24_i
.eq(ClockSignal())
744 # output 18 mhz PLL test signal
745 comb
+= self
.pll_18_o
.eq(pll
.pll_18_o
)
747 # now wire up ResetSignals. don't mind them being in this domain
748 pll_rst
= ResetSignal("pllclk")
749 comb
+= pll_rst
.eq(ResetSignal())
751 # internal clock is set to selector clock-out. has the side-effect of
752 # running TestIssuer at this speed (see DomainRenamer("intclk") above)
753 intclk
= ClockSignal("coresync")
755 comb
+= intclk
.eq(pll
.clk_pll_o
)
757 comb
+= intclk
.eq(ClockSignal())
762 return list(self
.ti
.ports()) + list(self
.pll
.ports()) + \
763 [ClockSignal(), ResetSignal()]
765 def external_ports(self
):
766 ports
= self
.ti
.external_ports()
767 ports
.append(ClockSignal())
768 ports
.append(ResetSignal())
770 ports
.append(self
.pll
.clk_sel_i
)
771 ports
.append(self
.pll_18_o
)
772 ports
.append(self
.pll
.pll_lck_o
)
776 if __name__
== '__main__':
777 units
= {'alu': 1, 'cr': 1, 'branch': 1, 'trap': 1, 'logical': 1,
783 pspec
= TestMemPspec(ldst_ifacetype
='bare_wb',
784 imem_ifacetype
='bare_wb',
789 dut
= TestIssuer(pspec
)
790 vl
= main(dut
, ports
=dut
.ports(), name
="test_issuer")
792 if len(sys
.argv
) == 1:
793 vl
= rtlil
.convert(dut
, ports
=dut
.external_ports(), name
="test_issuer")
794 with
open("test_issuer.il", "w") as f
: