"""
from nmigen import (Elaboratable, Module, Signal, ClockSignal, ResetSignal,
- ClockDomain, DomainRenamer)
+ ClockDomain, DomainRenamer, Mux)
from nmigen.cli import rtlil
from nmigen.cli import main
import sys
from soc.decoder.power_decoder import create_pdecode
-from soc.decoder.power_decoder2 import PowerDecode2
+from soc.decoder.power_decoder2 import PowerDecode2, SVP64PrefixDecoder
+from soc.decoder.decode2execute1 import IssuerDecode2ToOperand
from soc.decoder.decode2execute1 import Data
from soc.experiment.testmem import TestMemory # test only for instructions
-from soc.regfile.regfiles import StateRegs
+from soc.regfile.regfiles import StateRegs, FastRegs
from soc.simple.core import NonProductionCore
from soc.config.test.test_loadstore import TestMemPspec
from soc.config.ifetch import ConfigFetchUnit
from soc.decoder.power_enums import MicrOp
from soc.debug.dmi import CoreDebug, DMIInterface
+from soc.debug.jtag import JTAG
+from soc.config.pinouts import get_pinspecs
from soc.config.state import CoreState
from soc.interrupts.xics import XICS_ICP, XICS_ICS
from soc.bus.simple_gpio import SimpleGPIO
+from soc.clock.select import ClockSelect
+from soc.clock.dummypll import DummyPLL
+
from nmutil.util import rising_edge
-class TestIssuer(Elaboratable):
+class TestIssuerInternal(Elaboratable):
"""TestIssuer - reads instructions from TestMemory and issues them
efficiency and speed is not the main goal here: functional correctness is.
"""
def __init__(self, pspec):
+ # JTAG interface. add this right at the start because if it's
+ # added it *modifies* the pspec, by adding enable/disable signals
+ # for parts of the rest of the core
+ self.jtag_en = hasattr(pspec, "debug") and pspec.debug == 'jtag'
+ if self.jtag_en:
+ subset = {'uart', 'mtwi', 'eint', 'gpio', 'mspi0', 'mspi1',
+ 'pwm', 'sd0', 'sdr'}
+ self.jtag = JTAG(get_pinspecs(subset=subset))
+ # add signals to pspec to enable/disable icache and dcache
+ # (or data and intstruction wishbone if icache/dcache not included)
+ # https://bugs.libre-soc.org/show_bug.cgi?id=520
+ # TODO: do we actually care if these are not domain-synchronised?
+ # honestly probably not.
+ pspec.wb_icache_en = self.jtag.wb_icache_en
+ pspec.wb_dcache_en = self.jtag.wb_dcache_en
+
# add interrupt controller?
self.xics = hasattr(pspec, "xics") and pspec.xics == True
if self.xics:
self.simple_gpio = SimpleGPIO()
self.gpio_o = self.simple_gpio.gpio_o
- # main instruction core
+ # main instruction core25
self.core = core = NonProductionCore(pspec)
- # instruction decoder
+ # instruction decoder. goes into Trap Record
pdecode = create_pdecode()
- self.pdecode2 = PowerDecode2(pdecode) # decoder
+ self.cur_state = CoreState("cur") # current state (MSR/PC/EINT)
+ self.pdecode2 = PowerDecode2(pdecode, state=self.cur_state,
+ opkls=IssuerDecode2ToOperand)
+ self.svp64 = SVP64PrefixDecoder() # for decoding SVP64 prefix
# Test Instruction memory
self.imem = ConfigFetchUnit(pspec).fu
self.busy_o = Signal(reset_less=True)
self.memerr_o = Signal(reset_less=True)
- # FAST regfile read /write ports for PC and MSR
+ # FAST regfile read /write ports for PC, MSR, DEC/TB
staterf = self.core.regs.rf['state']
self.state_r_pc = staterf.r_ports['cia'] # PC rd
self.state_w_pc = staterf.w_ports['d_wr1'] # PC wr
m.submodules.core = core = DomainRenamer("coresync")(self.core)
m.submodules.imem = imem = self.imem
m.submodules.dbg = dbg = self.dbg
+ if self.jtag_en:
+ m.submodules.jtag = jtag = self.jtag
+ # TODO: UART2GDB mux, here, from external pin
+ # see https://bugs.libre-soc.org/show_bug.cgi?id=499
+ sync += dbg.dmi.connect_to(jtag.dmi)
- # current state (MSR/PC at the moment
- cur_state = CoreState("cur")
+ cur_state = self.cur_state
# XICS interrupt handler
if self.xics:
if self.gpio:
m.submodules.simple_gpio = simple_gpio = self.simple_gpio
- # connect one GPIO output to ICS bit 5 (like in microwatt soc.vhdl)
- if self.gpio and self.xics:
- comb += self.int_level_i[5].eq(simple_gpio.gpio_o[0])
+ # connect one GPIO output to ICS bit 15 (like in microwatt soc.vhdl)
+ # XXX causes litex ECP5 test to get wrong idea about input and output
+ # (but works with verilator sim *sigh*)
+ #if self.gpio and self.xics:
+ # comb += self.int_level_i[15].eq(simple_gpio.gpio_o[0])
# instruction decoder
pdecode = create_pdecode()
m.submodules.dec2 = pdecode2 = self.pdecode2
+ m.submodules.svp64 = svp64 = self.svp64
# convenience
dmi, d_reg, d_cr, d_xer, = dbg.dmi, dbg.d_gpr, dbg.d_cr, dbg.d_xer
core_sync = ClockDomain("coresync")
m.domains += cd_por, cd_sync, core_sync
+ ti_rst = Signal(reset_less=True)
delay = Signal(range(4), reset=3)
with m.If(delay != 0):
m.d.por += delay.eq(delay - 1)
comb += cd_por.clk.eq(ClockSignal())
- comb += core_sync.clk.eq(ClockSignal())
- # power-on reset delay
- comb += core.core_reset_i.eq(delay != 0 | dbg.core_rst_o)
+
+ # power-on reset delay
+ core_rst = ResetSignal("coresync")
+ comb += ti_rst.eq(delay != 0 | dbg.core_rst_o | ResetSignal())
+ comb += core_rst.eq(ti_rst)
# busy/halted signals from core
comb += self.busy_o.eq(core.busy_o)
comb += self.pc_o.eq(cur_state.pc)
ilatch = Signal(32)
- # next instruction (+4 on current)
+ # address of the next instruction, in the absence of a branch
+ # depends on the instruction size
nia = Signal(64, reset_less=True)
- comb += nia.eq(cur_state.pc + 4)
# read the PC
pc = Signal(64, reset_less=True)
dec_opcode_i = pdecode2.dec.raw_opcode_in # raw opcode
insn_type = core.e.do.insn_type
- dec_state = pdecode2.state
+
+ # handshake signals between fetch and decode/execute
+ # fetch FSM can run as soon as the PC is valid
+ fetch_pc_valid_i = Signal()
+ fetch_pc_ready_o = Signal()
+ # when done, deliver the instruction to the next FSM
+ fetch_insn_o = Signal(32, reset_less=True)
+ fetch_insn_valid_o = Signal()
+ fetch_insn_ready_i = Signal()
# actually use a nmigen FSM for the first time (w00t)
# this FSM is perhaps unusual in that it detects conditions
# then "holds" information, combinatorially, for the core
# (as opposed to using sync - which would be on a clock's delay)
# this includes the actual opcode, valid flags and so on.
- with m.FSM() as fsm:
+ with m.FSM(name='fetch_fsm'):
# waiting (zzz)
with m.State("IDLE"):
- sync += pc_changed.eq(0)
- sync += core.e.eq(0)
- with m.If(~dbg.core_stop_o & ~core.core_reset_i):
- # instruction allowed to go: start by reading the PC
- # capture the PC and also drop it into Insn Memory
- # we have joined a pair of combinatorial memory
- # lookups together. this is Generally Bad.
- comb += self.imem.a_pc_i.eq(pc)
- comb += self.imem.a_valid_i.eq(1)
- comb += self.imem.f_valid_i.eq(1)
- sync += cur_state.pc.eq(pc)
-
- # initiate read of MSR
- comb += self.state_r_msr.ren.eq(1<<StateRegs.MSR)
- sync += msr_read.eq(0)
-
- m.next = "INSN_READ" # move to "wait for bus" phase
+ with m.If(~dbg.core_stop_o & ~core_rst):
+ comb += fetch_pc_ready_o.eq(1)
+ with m.If(fetch_pc_valid_i):
+ # instruction allowed to go: start by reading the PC
+ # capture the PC and also drop it into Insn Memory
+ # we have joined a pair of combinatorial memory
+ # lookups together. this is Generally Bad.
+ comb += self.imem.a_pc_i.eq(pc)
+ comb += self.imem.a_valid_i.eq(1)
+ comb += self.imem.f_valid_i.eq(1)
+ sync += cur_state.pc.eq(pc)
+
+ # initiate read of MSR. arrives one clock later
+ comb += self.state_r_msr.ren.eq(1 << StateRegs.MSR)
+ sync += msr_read.eq(0)
+
+ m.next = "INSN_READ" # move to "wait for bus" phase
with m.Else():
comb += core.core_stopped_i.eq(1)
comb += dbg.core_stopped_i.eq(1)
# dummy pause to find out why simulation is not keeping up
with m.State("INSN_READ"):
- # one cycle later, msr read arrives
+ # one cycle later, msr read arrives. valid only once.
with m.If(~msr_read):
- sync += msr_read.eq(1)
+ sync += msr_read.eq(1) # yeah don't read it again
sync += cur_state.msr.eq(self.state_r_msr.data_o)
with m.If(self.imem.f_busy_o): # zzz...
# busy: stay in wait-read
insn = f_instr_o
else:
insn = f_instr_o.word_select(cur_state.pc[2], 32)
- comb += dec_opcode_i.eq(insn) # actual opcode
- comb += dec_state.eq(cur_state)
+ # decode the SVP64 prefix, if any
+ comb += svp64.raw_opcode_in.eq(insn)
+ comb += svp64.bigendian.eq(self.core_bigendian_i)
+ # pass the decoded prefix (if any) to PowerDecoder2
+ sync += pdecode2.sv_rm.eq(svp64.svp64_rm)
+ # calculate the address of the following instruction
+ insn_size = Mux(svp64.is_svp64_mode, 8, 4)
+ sync += nia.eq(cur_state.pc + insn_size)
+ with m.If(~svp64.is_svp64_mode):
+ # with no prefix, store the instruction
+ # and hand it directly to the next FSM
+ sync += fetch_insn_o.eq(insn)
+ m.next = "INSN_READY"
+ with m.Else():
+ # fetch the rest of the instruction from memory
+ comb += self.imem.a_pc_i.eq(cur_state.pc + 4)
+ comb += self.imem.a_valid_i.eq(1)
+ comb += self.imem.f_valid_i.eq(1)
+ m.next = "INSN_READ2"
+
+ with m.State("INSN_READ2"):
+ with m.If(self.imem.f_busy_o): # zzz...
+ # busy: stay in wait-read
+ comb += self.imem.a_valid_i.eq(1)
+ comb += self.imem.f_valid_i.eq(1)
+ with m.Else():
+ # not busy: instruction fetched
+ f_instr_o = self.imem.f_instr_o
+ if f_instr_o.width == 32:
+ insn = f_instr_o
+ else:
+ insn = f_instr_o.word_select((cur_state.pc+4)[2], 32)
+ sync += fetch_insn_o.eq(insn)
+ m.next = "INSN_READY"
+
+ with m.State("INSN_READY"):
+ # hand over the instruction, to be decoded
+ comb += fetch_insn_valid_o.eq(1)
+ with m.If(fetch_insn_ready_i):
+ m.next = "IDLE"
+
+ # decode / issue / execute FSM
+ with m.FSM():
+
+ # go fetch the instruction at the current PC
+ # at this point, there is no instruction running, that
+ # could inadvertently update the PC.
+ with m.State("INSN_FETCH"):
+ comb += fetch_pc_valid_i.eq(1)
+ with m.If(fetch_pc_ready_o):
+ m.next = "INSN_WAIT"
+
+ # decode the instruction when it arrives
+ with m.State("INSN_WAIT"):
+ comb += fetch_insn_ready_i.eq(1)
+ with m.If(fetch_insn_valid_o):
+ # decode the instruction
+ # TODO, before issuing new instruction first
+ # check if it's SVP64. (svp64.is_svp64_mode set)
+ # if yes, record the svp64_rm, put that into
+ # pdecode2.sv_rm, then read another 32 bits (INSN_FETCH2?)
+ comb += dec_opcode_i.eq(fetch_insn_o) # actual opcode
sync += core.e.eq(pdecode2.e)
+ sync += core.state.eq(cur_state)
+ sync += core.raw_insn_i.eq(dec_opcode_i)
+ sync += core.bigendian_i.eq(self.core_bigendian_i)
sync += ilatch.eq(insn) # latch current insn
# also drop PC and MSR into decode "state"
m.next = "INSN_START" # move to "start"
with m.State("INSN_START"):
comb += core_ivalid_i.eq(1) # instruction is valid
comb += core_issue_i.eq(1) # and issued
-
+ sync += pc_changed.eq(0)
m.next = "INSN_ACTIVE" # move to "wait completion"
comb += self.state_w_pc.wen.eq(1<<StateRegs.PC)
comb += self.state_w_pc.data_i.eq(nia)
sync += core.e.eq(0)
- m.next = "IDLE" # back to idle
+ sync += core.raw_insn_i.eq(0)
+ sync += core.bigendian_i.eq(0)
+ m.next = "INSN_FETCH" # back to fetch
# this bit doesn't have to be in the FSM: connect up to read
# regfiles on demand from DMI
comb += d_xer.data.eq(self.xer_r.data_o)
comb += d_xer.ack.eq(1)
+ # DEC and TB inc/dec FSM
+ self.tb_dec_fsm(m, cur_state.dec)
+
+ return m
+
+ def tb_dec_fsm(self, m, spr_dec):
+ """tb_dec_fsm
+
+ this is a FSM for updating either dec or tb. it runs alternately
+ DEC, TB, DEC, TB. note that SPR pipeline could have written a new
+ value to DEC, however the regfile has "passthrough" on it so this
+ *should* be ok.
+
+ see v3.0B p1097-1099 for Timeer Resource and p1065 and p1076
+ """
+
+ comb, sync = m.d.comb, m.d.sync
+ fast_rf = self.core.regs.rf['fast']
+ fast_r_dectb = fast_rf.r_ports['issue'] # DEC/TB
+ fast_w_dectb = fast_rf.w_ports['issue'] # DEC/TB
+
+ with m.FSM() as fsm:
+
+ # initiates read of current DEC
+ with m.State("DEC_READ"):
+ comb += fast_r_dectb.addr.eq(FastRegs.DEC)
+ comb += fast_r_dectb.ren.eq(1)
+ m.next = "DEC_WRITE"
+
+ # waits for DEC read to arrive (1 cycle), updates with new value
+ with m.State("DEC_WRITE"):
+ new_dec = Signal(64)
+ # TODO: MSR.LPCR 32-bit decrement mode
+ comb += new_dec.eq(fast_r_dectb.data_o - 1)
+ comb += fast_w_dectb.addr.eq(FastRegs.DEC)
+ comb += fast_w_dectb.wen.eq(1)
+ comb += fast_w_dectb.data_i.eq(new_dec)
+ sync += spr_dec.eq(new_dec) # copy into cur_state for decoder
+ m.next = "TB_READ"
+
+ # initiates read of current TB
+ with m.State("TB_READ"):
+ comb += fast_r_dectb.addr.eq(FastRegs.TB)
+ comb += fast_r_dectb.ren.eq(1)
+ m.next = "TB_WRITE"
+
+ # waits for read TB to arrive, initiates write of current TB
+ with m.State("TB_WRITE"):
+ new_tb = Signal(64)
+ comb += new_tb.eq(fast_r_dectb.data_o + 1)
+ comb += fast_w_dectb.addr.eq(FastRegs.TB)
+ comb += fast_w_dectb.wen.eq(1)
+ comb += fast_w_dectb.data_i.eq(new_tb)
+ m.next = "DEC_READ"
+
return m
def __iter__(self):
def external_ports(self):
ports = self.pc_i.ports()
ports += [self.pc_o, self.memerr_o, self.core_bigendian_i, self.busy_o,
- ClockSignal(), ResetSignal(),
]
- ports += list(self.dbg.dmi.ports())
+
+ if self.jtag_en:
+ ports += list(self.jtag.external_ports())
+ else:
+ # don't add DMI if JTAG is enabled
+ ports += list(self.dbg.dmi.ports())
+
ports += list(self.imem.ibus.fields.values())
ports += list(self.core.l0.cmpi.lsmem.lsi.slavebus.fields.values())
return list(self)
+class TestIssuer(Elaboratable):
+ def __init__(self, pspec):
+ self.ti = TestIssuerInternal(pspec)
+
+ self.pll = DummyPLL()
+
+ # PLL direct clock or not
+ self.pll_en = hasattr(pspec, "use_pll") and pspec.use_pll
+ if self.pll_en:
+ self.pll_18_o = Signal(reset_less=True)
+
+ def elaborate(self, platform):
+ m = Module()
+ comb = m.d.comb
+
+ # TestIssuer runs at direct clock
+ m.submodules.ti = ti = self.ti
+ cd_int = ClockDomain("coresync")
+
+ if self.pll_en:
+ # ClockSelect runs at PLL output internal clock rate
+ m.submodules.pll = pll = self.pll
+
+ # add clock domains from PLL
+ cd_pll = ClockDomain("pllclk")
+ m.domains += cd_pll
+
+ # PLL clock established. has the side-effect of running clklsel
+ # at the PLL's speed (see DomainRenamer("pllclk") above)
+ pllclk = ClockSignal("pllclk")
+ comb += pllclk.eq(pll.clk_pll_o)
+
+ # wire up external 24mhz to PLL
+ comb += pll.clk_24_i.eq(ClockSignal())
+
+ # output 18 mhz PLL test signal
+ comb += self.pll_18_o.eq(pll.pll_18_o)
+
+ # now wire up ResetSignals. don't mind them being in this domain
+ pll_rst = ResetSignal("pllclk")
+ comb += pll_rst.eq(ResetSignal())
+
+ # internal clock is set to selector clock-out. has the side-effect of
+ # running TestIssuer at this speed (see DomainRenamer("intclk") above)
+ intclk = ClockSignal("coresync")
+ if self.pll_en:
+ comb += intclk.eq(pll.clk_pll_o)
+ else:
+ comb += intclk.eq(ClockSignal())
+
+ return m
+
+ def ports(self):
+ return list(self.ti.ports()) + list(self.pll.ports()) + \
+ [ClockSignal(), ResetSignal()]
+
+ def external_ports(self):
+ ports = self.ti.external_ports()
+ ports.append(ClockSignal())
+ ports.append(ResetSignal())
+ if self.pll_en:
+ ports.append(self.pll.clk_sel_i)
+ ports.append(self.pll_18_o)
+ ports.append(self.pll.pll_lck_o)
+ return ports
+
+
if __name__ == '__main__':
units = {'alu': 1, 'cr': 1, 'branch': 1, 'trap': 1, 'logical': 1,
'spr': 1,