5aa8788e071ebf7756c5167e8761d5102c606b2b
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 # test is SVP64 is to be enabled
64 self
.svp64_en
= hasattr(pspec
, "svp64") and (pspec
.svp64
== True)
66 # JTAG interface. add this right at the start because if it's
67 # added it *modifies* the pspec, by adding enable/disable signals
68 # for parts of the rest of the core
69 self
.jtag_en
= hasattr(pspec
, "debug") and pspec
.debug
== 'jtag'
71 subset
= {'uart', 'mtwi', 'eint', 'gpio', 'mspi0', 'mspi1',
73 self
.jtag
= JTAG(get_pinspecs(subset
=subset
))
74 # add signals to pspec to enable/disable icache and dcache
75 # (or data and intstruction wishbone if icache/dcache not included)
76 # https://bugs.libre-soc.org/show_bug.cgi?id=520
77 # TODO: do we actually care if these are not domain-synchronised?
78 # honestly probably not.
79 pspec
.wb_icache_en
= self
.jtag
.wb_icache_en
80 pspec
.wb_dcache_en
= self
.jtag
.wb_dcache_en
81 self
.wb_sram_en
= self
.jtag
.wb_sram_en
83 self
.wb_sram_en
= Const(1)
86 self
.sram4x4k
= (hasattr(pspec
, "sram4x4kblock") and
87 pspec
.sram4x4kblock
== True)
91 self
.sram4k
.append(SPBlock512W64B8W(name
="sram4k_%d" % i
,
94 # add interrupt controller?
95 self
.xics
= hasattr(pspec
, "xics") and pspec
.xics
== True
97 self
.xics_icp
= XICS_ICP()
98 self
.xics_ics
= XICS_ICS()
99 self
.int_level_i
= self
.xics_ics
.int_level_i
101 # add GPIO peripheral?
102 self
.gpio
= hasattr(pspec
, "gpio") and pspec
.gpio
== True
104 self
.simple_gpio
= SimpleGPIO()
105 self
.gpio_o
= self
.simple_gpio
.gpio_o
107 # main instruction core25
108 self
.core
= core
= NonProductionCore(pspec
)
110 # instruction decoder. goes into Trap Record
111 pdecode
= create_pdecode()
112 self
.cur_state
= CoreState("cur") # current state (MSR/PC/EINT/SVSTATE)
113 self
.pdecode2
= PowerDecode2(pdecode
, state
=self
.cur_state
,
114 opkls
=IssuerDecode2ToOperand
,
115 svp64_en
=self
.svp64_en
)
117 self
.svp64
= SVP64PrefixDecoder() # for decoding SVP64 prefix
119 # Test Instruction memory
120 self
.imem
= ConfigFetchUnit(pspec
).fu
121 # one-row cache of instruction read
122 self
.iline
= Signal(64) # one instruction line
123 self
.iprev_adr
= Signal(64) # previous address: if different, do read
126 self
.dbg
= CoreDebug()
128 # instruction go/monitor
129 self
.pc_o
= Signal(64, reset_less
=True)
130 self
.pc_i
= Data(64, "pc_i") # set "ok" to indicate "please change me"
131 self
.svstate_i
= Data(32, "svstate_i") # ditto
132 self
.core_bigendian_i
= Signal()
133 self
.busy_o
= Signal(reset_less
=True)
134 self
.memerr_o
= Signal(reset_less
=True)
136 # STATE regfile read /write ports for PC, MSR, SVSTATE
137 staterf
= self
.core
.regs
.rf
['state']
138 self
.state_r_pc
= staterf
.r_ports
['cia'] # PC rd
139 self
.state_w_pc
= staterf
.w_ports
['d_wr1'] # PC wr
140 self
.state_r_msr
= staterf
.r_ports
['msr'] # MSR rd
141 self
.state_r_sv
= staterf
.r_ports
['sv'] # SVSTATE rd
142 self
.state_w_sv
= staterf
.w_ports
['sv'] # SVSTATE wr
144 # DMI interface access
145 intrf
= self
.core
.regs
.rf
['int']
146 crrf
= self
.core
.regs
.rf
['cr']
147 xerrf
= self
.core
.regs
.rf
['xer']
148 self
.int_r
= intrf
.r_ports
['dmi'] # INT read
149 self
.cr_r
= crrf
.r_ports
['full_cr_dbg'] # CR read
150 self
.xer_r
= xerrf
.r_ports
['full_xer'] # XER read
152 # hack method of keeping an eye on whether branch/trap set the PC
153 self
.state_nia
= self
.core
.regs
.rf
['state'].w_ports
['nia']
154 self
.state_nia
.wen
.name
= 'state_nia_wen'
156 def fetch_fsm(self
, m
, core
, pc
, svstate
, nia
,
157 fetch_pc_ready_o
, fetch_pc_valid_i
,
158 fetch_insn_valid_o
, fetch_insn_ready_i
):
160 this FSM performs fetch of raw instruction data, partial-decodes
161 it 32-bit at a time to detect SVP64 prefixes, and will optionally
162 read a 2nd 32-bit quantity if that occurs.
166 pdecode2
= self
.pdecode2
167 cur_state
= self
.cur_state
168 dec_opcode_i
= pdecode2
.dec
.raw_opcode_in
# raw opcode
170 msr_read
= Signal(reset
=1)
172 with m
.FSM(name
='fetch_fsm'):
175 with m
.State("IDLE"):
176 comb
+= fetch_pc_ready_o
.eq(1)
177 with m
.If(fetch_pc_valid_i
):
178 # instruction allowed to go: start by reading the PC
179 # capture the PC and also drop it into Insn Memory
180 # we have joined a pair of combinatorial memory
181 # lookups together. this is Generally Bad.
182 comb
+= self
.imem
.a_pc_i
.eq(pc
)
183 comb
+= self
.imem
.a_valid_i
.eq(1)
184 comb
+= self
.imem
.f_valid_i
.eq(1)
185 sync
+= cur_state
.pc
.eq(pc
)
186 sync
+= cur_state
.svstate
.eq(svstate
) # and svstate
188 # initiate read of MSR. arrives one clock later
189 comb
+= self
.state_r_msr
.ren
.eq(1 << StateRegs
.MSR
)
190 sync
+= msr_read
.eq(0)
192 m
.next
= "INSN_READ" # move to "wait for bus" phase
194 # dummy pause to find out why simulation is not keeping up
195 with m
.State("INSN_READ"):
196 # one cycle later, msr/sv read arrives. valid only once.
197 with m
.If(~msr_read
):
198 sync
+= msr_read
.eq(1) # yeah don't read it again
199 sync
+= cur_state
.msr
.eq(self
.state_r_msr
.data_o
)
200 with m
.If(self
.imem
.f_busy_o
): # zzz...
201 # busy: stay in wait-read
202 comb
+= self
.imem
.a_valid_i
.eq(1)
203 comb
+= self
.imem
.f_valid_i
.eq(1)
205 # not busy: instruction fetched
206 insn
= get_insn(self
.imem
.f_instr_o
, cur_state
.pc
)
209 # decode the SVP64 prefix, if any
210 comb
+= svp64
.raw_opcode_in
.eq(insn
)
211 comb
+= svp64
.bigendian
.eq(self
.core_bigendian_i
)
212 # pass the decoded prefix (if any) to PowerDecoder2
213 sync
+= pdecode2
.sv_rm
.eq(svp64
.svp64_rm
)
214 # calculate the address of the following instruction
215 insn_size
= Mux(svp64
.is_svp64_mode
, 8, 4)
216 sync
+= nia
.eq(cur_state
.pc
+ insn_size
)
217 with m
.If(~svp64
.is_svp64_mode
):
218 # with no prefix, store the instruction
219 # and hand it directly to the next FSM
220 sync
+= dec_opcode_i
.eq(insn
)
221 m
.next
= "INSN_READY"
223 # fetch the rest of the instruction from memory
224 comb
+= self
.imem
.a_pc_i
.eq(cur_state
.pc
+ 4)
225 comb
+= self
.imem
.a_valid_i
.eq(1)
226 comb
+= self
.imem
.f_valid_i
.eq(1)
227 m
.next
= "INSN_READ2"
229 # not SVP64 - 32-bit only
230 sync
+= nia
.eq(cur_state
.pc
+ 4)
231 sync
+= dec_opcode_i
.eq(insn
)
232 m
.next
= "INSN_READY"
234 with m
.State("INSN_READ2"):
235 with m
.If(self
.imem
.f_busy_o
): # zzz...
236 # busy: stay in wait-read
237 comb
+= self
.imem
.a_valid_i
.eq(1)
238 comb
+= self
.imem
.f_valid_i
.eq(1)
240 # not busy: instruction fetched
241 insn
= get_insn(self
.imem
.f_instr_o
, cur_state
.pc
+4)
242 sync
+= dec_opcode_i
.eq(insn
)
243 m
.next
= "INSN_READY"
245 with m
.State("INSN_READY"):
246 # hand over the instruction, to be decoded
247 comb
+= fetch_insn_valid_o
.eq(1)
248 with m
.If(fetch_insn_ready_i
):
251 def issue_fsm(self
, m
, core
, pc_changed
, sv_changed
, nia
,
253 fetch_pc_ready_o
, fetch_pc_valid_i
,
254 fetch_insn_valid_o
, fetch_insn_ready_i
,
255 exec_insn_valid_i
, exec_insn_ready_o
,
256 exec_pc_valid_o
, exec_pc_ready_i
):
259 decode / issue FSM. this interacts with the "fetch" FSM
260 through fetch_insn_ready/valid (incoming) and fetch_pc_ready/valid
261 (outgoing). also interacts with the "execute" FSM
262 through exec_insn_ready/valid (outgoing) and exec_pc_ready/valid
264 SVP64 RM prefixes have already been set up by the
265 "fetch" phase, so execute is fairly straightforward.
270 pdecode2
= self
.pdecode2
271 cur_state
= self
.cur_state
274 dec_opcode_i
= pdecode2
.dec
.raw_opcode_in
# raw opcode
276 # for updating svstate (things like srcstep etc.)
277 update_svstate
= Signal() # set this (below) if updating
278 new_svstate
= SVSTATERec("new_svstate")
279 comb
+= new_svstate
.eq(cur_state
.svstate
)
281 with m
.FSM(name
="issue_fsm"):
283 # go fetch the instruction at the current PC
284 # at this point, there is no instruction running, that
285 # could inadvertently update the PC.
286 with m
.State("INSN_FETCH"):
287 # wait on "core stop" release, before next fetch
288 # need to do this here, in case we are in a VL==0 loop
289 with m
.If(~dbg
.core_stop_o
& ~core_rst
):
290 comb
+= fetch_pc_valid_i
.eq(1)
291 with m
.If(fetch_pc_ready_o
):
294 comb
+= core
.core_stopped_i
.eq(1)
295 comb
+= dbg
.core_stopped_i
.eq(1)
296 # while stopped, allow updating the PC and SVSTATE
297 with m
.If(self
.pc_i
.ok
):
298 comb
+= self
.state_w_pc
.wen
.eq(1 << StateRegs
.PC
)
299 comb
+= self
.state_w_pc
.data_i
.eq(self
.pc_i
.data
)
300 sync
+= pc_changed
.eq(1)
301 with m
.If(self
.svstate_i
.ok
):
302 comb
+= new_svstate
.eq(self
.svstate_i
.data
)
303 comb
+= update_svstate
.eq(1)
304 sync
+= sv_changed
.eq(1)
306 # decode the instruction when it arrives
307 with m
.State("INSN_WAIT"):
308 comb
+= fetch_insn_ready_i
.eq(1)
309 with m
.If(fetch_insn_valid_o
):
310 # decode the instruction
311 sync
+= core
.e
.eq(pdecode2
.e
)
312 sync
+= core
.state
.eq(cur_state
)
313 sync
+= core
.raw_insn_i
.eq(dec_opcode_i
)
314 sync
+= core
.bigendian_i
.eq(self
.core_bigendian_i
)
315 # loop into INSN_FETCH if it's a vector instruction
316 # and VL == 0. this because VL==0 is a for-loop
317 # from 0 to 0 i.e. always, always a NOP.
318 cur_vl
= cur_state
.svstate
.vl
319 with m
.If(~pdecode2
.no_out_vec
& (cur_vl
== 0)):
320 # update the PC before fetching the next instruction
321 # since we are in a VL==0 loop, no instruction was
322 # executed that we could be overwriting
323 comb
+= self
.state_w_pc
.wen
.eq(1 << StateRegs
.PC
)
324 comb
+= self
.state_w_pc
.data_i
.eq(nia
)
325 m
.next
= "INSN_FETCH"
327 m
.next
= "INSN_EXECUTE" # move to "execute"
329 with m
.State("INSN_EXECUTE"):
330 comb
+= exec_insn_valid_i
.eq(1)
331 with m
.If(exec_insn_ready_o
):
332 m
.next
= "EXECUTE_WAIT"
334 with m
.State("EXECUTE_WAIT"):
335 # wait on "core stop" release, at instruction end
336 # need to do this here, in case we are in a VL>1 loop
337 with m
.If(~dbg
.core_stop_o
& ~core_rst
):
338 comb
+= exec_pc_ready_i
.eq(1)
339 with m
.If(exec_pc_valid_o
):
340 # precalculate srcstep+1
341 next_srcstep
= Signal
.like(cur_state
.svstate
.srcstep
)
342 comb
+= next_srcstep
.eq(cur_state
.svstate
.srcstep
+1)
343 # was this the last loop iteration?
345 cur_vl
= cur_state
.svstate
.vl
346 comb
+= is_last
.eq(next_srcstep
== cur_vl
)
348 # if either PC or SVSTATE were changed by the previous
349 # instruction, go directly back to Fetch, without
350 # updating either PC or SVSTATE
351 with m
.If(pc_changed | sv_changed
):
352 m
.next
= "INSN_FETCH"
354 # also return to Fetch, when no output was a vector
355 # (regardless of SRCSTEP and VL), or when the last
356 # instruction was really the last one of the VL loop
357 with m
.Elif(pdecode2
.no_out_vec | is_last
):
358 # before going back to fetch, update the PC state
359 # register with the NIA.
360 # ok here we are not reading the branch unit.
361 # TODO: this just blithely overwrites whatever
362 # pipeline updated the PC
363 comb
+= self
.state_w_pc
.wen
.eq(1 << StateRegs
.PC
)
364 comb
+= self
.state_w_pc
.data_i
.eq(nia
)
365 # reset SRCSTEP before returning to Fetch
366 with m
.If(~pdecode2
.no_out_vec
):
367 comb
+= new_svstate
.srcstep
.eq(0)
368 comb
+= update_svstate
.eq(1)
369 m
.next
= "INSN_FETCH"
371 # returning to Execute? then, first update SRCSTEP
373 comb
+= new_svstate
.srcstep
.eq(next_srcstep
)
374 comb
+= update_svstate
.eq(1)
378 comb
+= core
.core_stopped_i
.eq(1)
379 comb
+= dbg
.core_stopped_i
.eq(1)
380 # while stopped, allow updating the PC and SVSTATE
381 with m
.If(self
.pc_i
.ok
):
382 comb
+= self
.state_w_pc
.wen
.eq(1 << StateRegs
.PC
)
383 comb
+= self
.state_w_pc
.data_i
.eq(self
.pc_i
.data
)
384 sync
+= pc_changed
.eq(1)
385 with m
.If(self
.svstate_i
.ok
):
386 comb
+= new_svstate
.eq(self
.svstate_i
.data
)
387 comb
+= update_svstate
.eq(1)
388 sync
+= sv_changed
.eq(1)
390 # need to decode the instruction again, after updating SRCSTEP
391 # in the previous state.
392 # mostly a copy of INSN_WAIT, but without the actual wait
393 with m
.State("DECODE_SV"):
394 # decode the instruction
395 sync
+= core
.e
.eq(pdecode2
.e
)
396 sync
+= core
.state
.eq(cur_state
)
397 sync
+= core
.bigendian_i
.eq(self
.core_bigendian_i
)
398 m
.next
= "INSN_EXECUTE" # move to "execute"
400 # check if svstate needs updating: if so, write it to State Regfile
401 with m
.If(update_svstate
):
402 comb
+= self
.state_w_sv
.wen
.eq(1<<StateRegs
.SVSTATE
)
403 comb
+= self
.state_w_sv
.data_i
.eq(new_svstate
)
404 sync
+= cur_state
.svstate
.eq(new_svstate
) # for next clock
406 def execute_fsm(self
, m
, core
, pc_changed
, sv_changed
,
407 exec_insn_valid_i
, exec_insn_ready_o
,
408 exec_pc_valid_o
, exec_pc_ready_i
):
411 execute FSM. this interacts with the "issue" FSM
412 through exec_insn_ready/valid (incoming) and exec_pc_ready/valid
413 (outgoing). SVP64 RM prefixes have already been set up by the
414 "issue" phase, so execute is fairly straightforward.
419 pdecode2
= self
.pdecode2
422 core_busy_o
= core
.busy_o
# core is busy
423 core_ivalid_i
= core
.ivalid_i
# instruction is valid
424 core_issue_i
= core
.issue_i
# instruction is issued
425 insn_type
= core
.e
.do
.insn_type
# instruction MicroOp type
427 with m
.FSM(name
="exec_fsm"):
429 # waiting for instruction bus (stays there until not busy)
430 with m
.State("INSN_START"):
431 comb
+= exec_insn_ready_o
.eq(1)
432 with m
.If(exec_insn_valid_i
):
433 comb
+= core_ivalid_i
.eq(1) # instruction is valid
434 comb
+= core_issue_i
.eq(1) # and issued
435 sync
+= sv_changed
.eq(0)
436 sync
+= pc_changed
.eq(0)
437 m
.next
= "INSN_ACTIVE" # move to "wait completion"
439 # instruction started: must wait till it finishes
440 with m
.State("INSN_ACTIVE"):
441 with m
.If(insn_type
!= MicrOp
.OP_NOP
):
442 comb
+= core_ivalid_i
.eq(1) # instruction is valid
443 # note changes to PC and SVSTATE
444 with m
.If(self
.state_nia
.wen
& (1<<StateRegs
.SVSTATE
)):
445 sync
+= sv_changed
.eq(1)
446 with m
.If(self
.state_nia
.wen
& (1<<StateRegs
.PC
)):
447 sync
+= pc_changed
.eq(1)
448 with m
.If(~core_busy_o
): # instruction done!
449 comb
+= exec_pc_valid_o
.eq(1)
450 with m
.If(exec_pc_ready_i
):
451 m
.next
= "INSN_START" # back to fetch
453 def elaborate(self
, platform
):
455 comb
, sync
= m
.d
.comb
, m
.d
.sync
457 m
.submodules
.core
= core
= DomainRenamer("coresync")(self
.core
)
458 m
.submodules
.imem
= imem
= self
.imem
459 m
.submodules
.dbg
= dbg
= self
.dbg
461 m
.submodules
.jtag
= jtag
= self
.jtag
462 # TODO: UART2GDB mux, here, from external pin
463 # see https://bugs.libre-soc.org/show_bug.cgi?id=499
464 sync
+= dbg
.dmi
.connect_to(jtag
.dmi
)
466 cur_state
= self
.cur_state
468 # 4x 4k SRAM blocks. these simply "exist", they get routed in litex
470 for i
, sram
in enumerate(self
.sram4k
):
471 m
.submodules
["sram4k_%d" % i
] = sram
472 comb
+= sram
.enable
.eq(self
.wb_sram_en
)
474 # XICS interrupt handler
476 m
.submodules
.xics_icp
= icp
= self
.xics_icp
477 m
.submodules
.xics_ics
= ics
= self
.xics_ics
478 comb
+= icp
.ics_i
.eq(ics
.icp_o
) # connect ICS to ICP
479 sync
+= cur_state
.eint
.eq(icp
.core_irq_o
) # connect ICP to core
481 # GPIO test peripheral
483 m
.submodules
.simple_gpio
= simple_gpio
= self
.simple_gpio
485 # connect one GPIO output to ICS bit 15 (like in microwatt soc.vhdl)
486 # XXX causes litex ECP5 test to get wrong idea about input and output
487 # (but works with verilator sim *sigh*)
488 #if self.gpio and self.xics:
489 # comb += self.int_level_i[15].eq(simple_gpio.gpio_o[0])
491 # instruction decoder
492 pdecode
= create_pdecode()
493 m
.submodules
.dec2
= pdecode2
= self
.pdecode2
495 m
.submodules
.svp64
= svp64
= self
.svp64
498 dmi
, d_reg
, d_cr
, d_xer
, = dbg
.dmi
, dbg
.d_gpr
, dbg
.d_cr
, dbg
.d_xer
499 intrf
= self
.core
.regs
.rf
['int']
501 # clock delay power-on reset
502 cd_por
= ClockDomain(reset_less
=True)
503 cd_sync
= ClockDomain()
504 core_sync
= ClockDomain("coresync")
505 m
.domains
+= cd_por
, cd_sync
, core_sync
507 ti_rst
= Signal(reset_less
=True)
508 delay
= Signal(range(4), reset
=3)
509 with m
.If(delay
!= 0):
510 m
.d
.por
+= delay
.eq(delay
- 1)
511 comb
+= cd_por
.clk
.eq(ClockSignal())
513 # power-on reset delay
514 core_rst
= ResetSignal("coresync")
515 comb
+= ti_rst
.eq(delay
!= 0 | dbg
.core_rst_o |
ResetSignal())
516 comb
+= core_rst
.eq(ti_rst
)
518 # busy/halted signals from core
519 comb
+= self
.busy_o
.eq(core
.busy_o
)
520 comb
+= pdecode2
.dec
.bigendian
.eq(self
.core_bigendian_i
)
522 # temporary hack: says "go" immediately for both address gen and ST
524 ldst
= core
.fus
.fus
['ldst0']
525 st_go_edge
= rising_edge(m
, ldst
.st
.rel_o
)
526 m
.d
.comb
+= ldst
.ad
.go_i
.eq(ldst
.ad
.rel_o
) # link addr-go direct to rel
527 m
.d
.comb
+= ldst
.st
.go_i
.eq(st_go_edge
) # link store-go to rising rel
529 # PC and instruction from I-Memory
530 comb
+= self
.pc_o
.eq(cur_state
.pc
)
531 pc_changed
= Signal() # note write to PC
532 sv_changed
= Signal() # note write to SVSTATE
535 pc
= Signal(64, reset_less
=True)
536 pc_ok_delay
= Signal()
537 sync
+= pc_ok_delay
.eq(~self
.pc_i
.ok
)
538 with m
.If(self
.pc_i
.ok
):
539 # incoming override (start from pc_i)
540 comb
+= pc
.eq(self
.pc_i
.data
)
542 # otherwise read StateRegs regfile for PC...
543 comb
+= self
.state_r_pc
.ren
.eq(1<<StateRegs
.PC
)
544 # ... but on a 1-clock delay
545 with m
.If(pc_ok_delay
):
546 comb
+= pc
.eq(self
.state_r_pc
.data_o
)
549 svstate
= Signal(64, reset_less
=True)
550 svstate_ok_delay
= Signal()
551 sync
+= svstate_ok_delay
.eq(~self
.svstate_i
.ok
)
552 with m
.If(self
.svstate_i
.ok
):
553 # incoming override (start from svstate__i)
554 comb
+= svstate
.eq(self
.svstate_i
.data
)
556 # otherwise read StateRegs regfile for SVSTATE...
557 comb
+= self
.state_r_sv
.ren
.eq(1 << StateRegs
.SVSTATE
)
558 # ... but on a 1-clock delay
559 with m
.If(svstate_ok_delay
):
560 comb
+= svstate
.eq(self
.state_r_sv
.data_o
)
562 # don't write pc every cycle
563 comb
+= self
.state_w_pc
.wen
.eq(0)
564 comb
+= self
.state_w_pc
.data_i
.eq(0)
566 # don't read msr every cycle
567 comb
+= self
.state_r_msr
.ren
.eq(0)
569 # address of the next instruction, in the absence of a branch
570 # depends on the instruction size
571 nia
= Signal(64, reset_less
=True)
573 # connect up debug signals
574 # TODO comb += core.icache_rst_i.eq(dbg.icache_rst_o)
575 comb
+= dbg
.terminate_i
.eq(core
.core_terminate_o
)
576 comb
+= dbg
.state
.pc
.eq(pc
)
577 comb
+= dbg
.state
.svstate
.eq(svstate
)
578 comb
+= dbg
.state
.msr
.eq(cur_state
.msr
)
580 # there are *THREE* FSMs, fetch (32/64-bit) issue, decode/execute.
581 # these are the handshake signals between fetch and decode/execute
583 # fetch FSM can run as soon as the PC is valid
584 fetch_pc_valid_i
= Signal() # Execute tells Fetch "start next read"
585 fetch_pc_ready_o
= Signal() # Fetch Tells SVSTATE "proceed"
587 # fetch FSM hands over the instruction to be decoded / issued
588 fetch_insn_valid_o
= Signal()
589 fetch_insn_ready_i
= Signal()
591 # issue FSM delivers the instruction to the be executed
592 exec_insn_valid_i
= Signal()
593 exec_insn_ready_o
= Signal()
595 # execute FSM, hands over the PC/SVSTATE back to the issue FSM
596 exec_pc_valid_o
= Signal()
597 exec_pc_ready_i
= Signal()
599 # the FSMs here are perhaps unusual in that they detect conditions
600 # then "hold" information, combinatorially, for the core
601 # (as opposed to using sync - which would be on a clock's delay)
602 # this includes the actual opcode, valid flags and so on.
604 # Fetch, then Issue, then Execute. Issue is where the VL for-loop
605 # lives. the ready/valid signalling is used to communicate between
608 self
.fetch_fsm(m
, core
, pc
, svstate
, nia
,
609 fetch_pc_ready_o
, fetch_pc_valid_i
,
610 fetch_insn_valid_o
, fetch_insn_ready_i
)
612 self
.issue_fsm(m
, core
, pc_changed
, sv_changed
, nia
,
614 fetch_pc_ready_o
, fetch_pc_valid_i
,
615 fetch_insn_valid_o
, fetch_insn_ready_i
,
616 exec_insn_valid_i
, exec_insn_ready_o
,
617 exec_pc_valid_o
, exec_pc_ready_i
)
619 self
.execute_fsm(m
, core
, pc_changed
, sv_changed
,
620 exec_insn_valid_i
, exec_insn_ready_o
,
621 exec_pc_valid_o
, exec_pc_ready_i
)
623 # this bit doesn't have to be in the FSM: connect up to read
624 # regfiles on demand from DMI
627 # DEC and TB inc/dec FSM. copy of DEC is put into CoreState,
628 # (which uses that in PowerDecoder2 to raise 0x900 exception)
629 self
.tb_dec_fsm(m
, cur_state
.dec
)
633 def do_dmi(self
, m
, dbg
):
636 dmi
, d_reg
, d_cr
, d_xer
, = dbg
.dmi
, dbg
.d_gpr
, dbg
.d_cr
, dbg
.d_xer
637 intrf
= self
.core
.regs
.rf
['int']
639 with m
.If(d_reg
.req
): # request for regfile access being made
640 # TODO: error-check this
641 # XXX should this be combinatorial? sync better?
643 comb
+= self
.int_r
.ren
.eq(1<<d_reg
.addr
)
645 comb
+= self
.int_r
.addr
.eq(d_reg
.addr
)
646 comb
+= self
.int_r
.ren
.eq(1)
647 d_reg_delay
= Signal()
648 sync
+= d_reg_delay
.eq(d_reg
.req
)
649 with m
.If(d_reg_delay
):
650 # data arrives one clock later
651 comb
+= d_reg
.data
.eq(self
.int_r
.data_o
)
652 comb
+= d_reg
.ack
.eq(1)
654 # sigh same thing for CR debug
655 with m
.If(d_cr
.req
): # request for regfile access being made
656 comb
+= self
.cr_r
.ren
.eq(0b11111111) # enable all
657 d_cr_delay
= Signal()
658 sync
+= d_cr_delay
.eq(d_cr
.req
)
659 with m
.If(d_cr_delay
):
660 # data arrives one clock later
661 comb
+= d_cr
.data
.eq(self
.cr_r
.data_o
)
662 comb
+= d_cr
.ack
.eq(1)
665 with m
.If(d_xer
.req
): # request for regfile access being made
666 comb
+= self
.xer_r
.ren
.eq(0b111111) # enable all
667 d_xer_delay
= Signal()
668 sync
+= d_xer_delay
.eq(d_xer
.req
)
669 with m
.If(d_xer_delay
):
670 # data arrives one clock later
671 comb
+= d_xer
.data
.eq(self
.xer_r
.data_o
)
672 comb
+= d_xer
.ack
.eq(1)
674 def tb_dec_fsm(self
, m
, spr_dec
):
677 this is a FSM for updating either dec or tb. it runs alternately
678 DEC, TB, DEC, TB. note that SPR pipeline could have written a new
679 value to DEC, however the regfile has "passthrough" on it so this
682 see v3.0B p1097-1099 for Timeer Resource and p1065 and p1076
685 comb
, sync
= m
.d
.comb
, m
.d
.sync
686 fast_rf
= self
.core
.regs
.rf
['fast']
687 fast_r_dectb
= fast_rf
.r_ports
['issue'] # DEC/TB
688 fast_w_dectb
= fast_rf
.w_ports
['issue'] # DEC/TB
692 # initiates read of current DEC
693 with m
.State("DEC_READ"):
694 comb
+= fast_r_dectb
.addr
.eq(FastRegs
.DEC
)
695 comb
+= fast_r_dectb
.ren
.eq(1)
698 # waits for DEC read to arrive (1 cycle), updates with new value
699 with m
.State("DEC_WRITE"):
701 # TODO: MSR.LPCR 32-bit decrement mode
702 comb
+= new_dec
.eq(fast_r_dectb
.data_o
- 1)
703 comb
+= fast_w_dectb
.addr
.eq(FastRegs
.DEC
)
704 comb
+= fast_w_dectb
.wen
.eq(1)
705 comb
+= fast_w_dectb
.data_i
.eq(new_dec
)
706 sync
+= spr_dec
.eq(new_dec
) # copy into cur_state for decoder
709 # initiates read of current TB
710 with m
.State("TB_READ"):
711 comb
+= fast_r_dectb
.addr
.eq(FastRegs
.TB
)
712 comb
+= fast_r_dectb
.ren
.eq(1)
715 # waits for read TB to arrive, initiates write of current TB
716 with m
.State("TB_WRITE"):
718 comb
+= new_tb
.eq(fast_r_dectb
.data_o
+ 1)
719 comb
+= fast_w_dectb
.addr
.eq(FastRegs
.TB
)
720 comb
+= fast_w_dectb
.wen
.eq(1)
721 comb
+= fast_w_dectb
.data_i
.eq(new_tb
)
727 yield from self
.pc_i
.ports()
730 yield from self
.core
.ports()
731 yield from self
.imem
.ports()
732 yield self
.core_bigendian_i
738 def external_ports(self
):
739 ports
= self
.pc_i
.ports()
740 ports
+= [self
.pc_o
, self
.memerr_o
, self
.core_bigendian_i
, self
.busy_o
,
744 ports
+= list(self
.jtag
.external_ports())
746 # don't add DMI if JTAG is enabled
747 ports
+= list(self
.dbg
.dmi
.ports())
749 ports
+= list(self
.imem
.ibus
.fields
.values())
750 ports
+= list(self
.core
.l0
.cmpi
.lsmem
.lsi
.slavebus
.fields
.values())
753 for sram
in self
.sram4k
:
754 ports
+= list(sram
.bus
.fields
.values())
757 ports
+= list(self
.xics_icp
.bus
.fields
.values())
758 ports
+= list(self
.xics_ics
.bus
.fields
.values())
759 ports
.append(self
.int_level_i
)
762 ports
+= list(self
.simple_gpio
.bus
.fields
.values())
763 ports
.append(self
.gpio_o
)
771 class TestIssuer(Elaboratable
):
772 def __init__(self
, pspec
):
773 self
.ti
= TestIssuerInternal(pspec
)
775 self
.pll
= DummyPLL()
777 # PLL direct clock or not
778 self
.pll_en
= hasattr(pspec
, "use_pll") and pspec
.use_pll
780 self
.pll_18_o
= Signal(reset_less
=True)
782 def elaborate(self
, platform
):
786 # TestIssuer runs at direct clock
787 m
.submodules
.ti
= ti
= self
.ti
788 cd_int
= ClockDomain("coresync")
791 # ClockSelect runs at PLL output internal clock rate
792 m
.submodules
.pll
= pll
= self
.pll
794 # add clock domains from PLL
795 cd_pll
= ClockDomain("pllclk")
798 # PLL clock established. has the side-effect of running clklsel
799 # at the PLL's speed (see DomainRenamer("pllclk") above)
800 pllclk
= ClockSignal("pllclk")
801 comb
+= pllclk
.eq(pll
.clk_pll_o
)
803 # wire up external 24mhz to PLL
804 comb
+= pll
.clk_24_i
.eq(ClockSignal())
806 # output 18 mhz PLL test signal
807 comb
+= self
.pll_18_o
.eq(pll
.pll_18_o
)
809 # now wire up ResetSignals. don't mind them being in this domain
810 pll_rst
= ResetSignal("pllclk")
811 comb
+= pll_rst
.eq(ResetSignal())
813 # internal clock is set to selector clock-out. has the side-effect of
814 # running TestIssuer at this speed (see DomainRenamer("intclk") above)
815 intclk
= ClockSignal("coresync")
817 comb
+= intclk
.eq(pll
.clk_pll_o
)
819 comb
+= intclk
.eq(ClockSignal())
824 return list(self
.ti
.ports()) + list(self
.pll
.ports()) + \
825 [ClockSignal(), ResetSignal()]
827 def external_ports(self
):
828 ports
= self
.ti
.external_ports()
829 ports
.append(ClockSignal())
830 ports
.append(ResetSignal())
832 ports
.append(self
.pll
.clk_sel_i
)
833 ports
.append(self
.pll_18_o
)
834 ports
.append(self
.pll
.pll_lck_o
)
838 if __name__
== '__main__':
839 units
= {'alu': 1, 'cr': 1, 'branch': 1, 'trap': 1, 'logical': 1,
845 pspec
= TestMemPspec(ldst_ifacetype
='bare_wb',
846 imem_ifacetype
='bare_wb',
851 dut
= TestIssuer(pspec
)
852 vl
= main(dut
, ports
=dut
.ports(), name
="test_issuer")
854 if len(sys
.argv
) == 1:
855 vl
= rtlil
.convert(dut
, ports
=dut
.external_ports(), name
="test_issuer")
856 with
open("test_issuer.il", "w") as f
: