(pspec.microwatt_compat == True))
self.alt_reset = Signal(reset_less=True) # not connected yet (microwatt)
+ if self.microwatt_compat:
+ self.microwatt_old = False
+ self.microwatt_debug = True # set to False when using an FPGA
+
# test is SVP64 is to be enabled
self.svp64_en = hasattr(pspec, "svp64") and (pspec.svp64 == True)
self.state_r_pc = staterf.r_ports['cia'] # PC rd
self.state_r_sv = staterf.r_ports['sv'] # SVSTATE rd
- self.state_w_msr = staterf.w_ports['msr'] # MSR wr
+ self.state_w_msr = staterf.w_ports['d_wr2'] # MSR wr
self.state_w_pc = staterf.w_ports['d_wr1'] # PC wr
self.state_w_sv = staterf.w_ports['sv'] # SVSTATE wr
# DMI interface access
intrf = self.core.regs.rf['int']
+ fastrf = self.core.regs.rf['fast']
crrf = self.core.regs.rf['cr']
xerrf = self.core.regs.rf['xer']
- self.int_r = intrf.r_ports['dmi'] # INT read
- self.cr_r = crrf.r_ports['full_cr_dbg'] # CR read
- self.xer_r = xerrf.r_ports['full_xer'] # XER read
+ self.int_r = intrf.r_ports['dmi'] # INT DMI read
+ self.cr_r = crrf.r_ports['full_cr_dbg'] # CR DMI read
+ self.xer_r = xerrf.r_ports['full_xer'] # XER DMI read
+ self.fast_r = fastrf.r_ports['dmi'] # FAST DMI read
if self.svp64_en:
# for predication
# hack method of keeping an eye on whether branch/trap set the PC
self.state_nia = self.core.regs.rf['state'].w_ports['nia']
self.state_nia.wen.name = 'state_nia_wen'
+ # and whether SPR pipeline sets DEC or TB
+ self.state_spr = self.core.regs.rf['state'].w_ports['state1']
# pulse to synchronize the simulator at instruction end
self.insn_done = Signal()
self.srcmask = Signal(64)
self.dstmask = Signal(64)
+ # sigh, the wishbone addresses are not wishbone-compliant
+ # in old versions of microwatt, tplaten_3d_game is a new one
+ if self.microwatt_compat:
+ self.ibus_adr = Signal(32, name='wishbone_insn_out.adr')
+ self.dbus_adr = Signal(32, name='wishbone_data_out.adr')
+
+ # add an output of the PC and instruction, and whether it was requested
+ # this is for verilator debug purposes
+ if self.microwatt_compat:
+ self.nia = Signal(64)
+ self.msr_o = Signal(64)
+ self.nia_req = Signal(1)
+ self.insn = Signal(32)
+ self.ldst_req = Signal(1)
+ self.ldst_addr = Signal(1)
+
+ # for pausing dec/tb during an SPR pipeline event, this
+ # ensures that an SPR write (mtspr) to TB or DEC does not
+ # get overwritten by the DEC/TB FSM
+ self.pause_dec_tb = Signal()
+
def setup_peripherals(self, m):
comb, sync = m.d.comb, m.d.sync
m.submodules.core = core = self.core
else:
m.submodules.core = core = csd(self.core)
+
# this _so_ needs sorting out. ICache is added down inside
# LoadStore1 and is already a submodule of LoadStore1
if not isinstance(self.imem, ICache):
m.submodules.imem = imem = csd(self.imem)
- if self.microwatt_compat:
- m.submodules.dbg = dbg = self.dbg
- else:
- m.submodules.dbg = dbg = dbd(self.dbg)
+
+ # set up JTAG Debug Module (in correct domain)
+ m.submodules.dbg = dbg = dbd(self.dbg)
if self.jtag_en:
m.submodules.jtag = jtag = dbd(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)
+ # fixup the clocks in microwatt-compat mode (but leave resets alone
+ # so that microwatt soc.vhdl can pull a reset on the core or DMI
+ # can do it, just like in TestIssuer)
+ if self.microwatt_compat:
+ intclk = ClockSignal(self.core_domain)
+ dbgclk = ClockSignal(self.dbg_domain)
+ if self.core_domain != 'sync':
+ comb += intclk.eq(ClockSignal())
+ if self.dbg_domain != 'sync':
+ comb += dbgclk.eq(ClockSignal())
+
+ # if using old version of microwatt
+ # drop the first 3 bits of the incoming wishbone addresses
+ if self.microwatt_compat:
+ ibus = self.imem.ibus
+ dbus = self.core.l0.cmpi.wb_bus()
+ if self.microwatt_old:
+ comb += self.ibus_adr.eq(Cat(Const(0, 3), ibus.adr))
+ comb += self.dbus_adr.eq(Cat(Const(0, 3), dbus.adr))
+ else:
+ comb += self.ibus_adr.eq(ibus.adr)
+ comb += self.dbus_adr.eq(dbus.adr)
+ if self.microwatt_debug:
+ # microwatt verilator debug purposes
+ pi = self.core.l0.cmpi.pi.pi
+ comb += self.ldst_req.eq(pi.addr_ok_o)
+ comb += self.ldst_addr.eq(pi.addr)
+
cur_state = self.cur_state
# 4x 4k SRAM blocks. these simply "exist", they get routed in litex
if self.svp64_en:
m.submodules.svp64 = svp64 = csd(self.svp64)
- # convenience
- dmi, d_reg, d_cr, d_xer, = dbg.dmi, dbg.d_gpr, dbg.d_cr, dbg.d_xer
- intrf = self.core.regs.rf['int']
-
# clock delay power-on reset
cd_por = ClockDomain(reset_less=True)
cd_sync = ClockDomain()
dbg_sync = ClockDomain(self.dbg_domain)
m.domains += dbg_sync
+ # create a delay, but remember it is in the power-on-reset clock domain!
ti_rst = Signal(reset_less=True)
delay = Signal(range(4), reset=3)
+ stop_delay = Signal(range(16), reset=5)
with m.If(delay != 0):
- m.d.por += delay.eq(delay - 1)
+ m.d.por += delay.eq(delay - 1) # decrement... in POR domain!
+ with m.If(stop_delay != 0):
+ m.d.por += stop_delay.eq(stop_delay - 1) # likewise
comb += cd_por.clk.eq(ClockSignal())
# power-on reset delay
else:
with m.If(delay != 0 | dbg.core_rst_o):
comb += core_rst.eq(1)
+ with m.If(stop_delay != 0):
+ # run DMI core-stop as well but on an extra couple of cycles
+ comb += dbg.core_stopped_i.eq(1)
# connect external reset signal to DMI Reset
if self.dbg_domain != "sync":
comb += pdecode2.dec.bigendian.eq(self.core_bigendian_i)
# temporary hack: says "go" immediately for both address gen and ST
+ # XXX: st.go_i is set to 1 cycle delay to reduce combinatorial chains
l0 = core.l0
ldst = core.fus.fus['ldst0']
st_go_edge = rising_edge(m, ldst.st.rel_o)
# link addr-go direct to rel
m.d.comb += ldst.ad.go_i.eq(ldst.ad.rel_o)
- m.d.comb += ldst.st.go_i.eq(st_go_edge) # link store-go to rising rel
+ m.d.sync += ldst.st.go_i.eq(st_go_edge) # link store-go to rising rel
def do_dmi(self, m, dbg):
"""deals with DMI debug requests
comb = m.d.comb
sync = m.d.sync
dmi, d_reg, d_cr, d_xer, = dbg.dmi, dbg.d_gpr, dbg.d_cr, dbg.d_xer
+ d_fast = dbg.d_fast
intrf = self.core.regs.rf['int']
+ fastrf = self.core.regs.rf['fast']
with m.If(d_reg.req): # request for regfile access being made
# TODO: error-check this
comb += d_reg.data.eq(self.int_r.o_data)
comb += d_reg.ack.eq(1)
+ # fast regfile
+ with m.If(d_fast.req): # request for regfile access being made
+ if fastrf.unary:
+ comb += self.fast_r.ren.eq(1 << d_fast.addr)
+ else:
+ comb += self.fast_r.addr.eq(d_fast.addr)
+ comb += self.fast_r.ren.eq(1)
+ d_fast_delay = Signal()
+ sync += d_fast_delay.eq(d_fast.req)
+ with m.If(d_fast_delay):
+ # data arrives one clock later
+ comb += d_fast.data.eq(self.fast_r.o_data)
+ comb += d_fast.ack.eq(1)
+
# sigh same thing for CR debug
with m.If(d_cr.req): # request for regfile access being made
comb += self.cr_r.ren.eq(0b11111111) # enable all
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
+ see v3.0B p1097-1099 for Timer 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
+ state_rf = self.core.regs.rf['state']
+ state_r_dectb = state_rf.r_ports['issue'] # DEC/TB
+ state_w_dectb = state_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"
+ comb += state_r_dectb.ren.eq(1<<StateRegs.DEC)
+ with m.If(~self.pause_dec_tb):
+ m.next = "DEC_WRITE"
# waits for DEC read to arrive (1 cycle), updates with new value
+ # respects if dec/tb writing has been paused
with m.State("DEC_WRITE"):
- new_dec = Signal(64)
- # TODO: MSR.LPCR 32-bit decrement mode
- comb += new_dec.eq(fast_r_dectb.o_data - 1)
- comb += fast_w_dectb.addr.eq(FastRegs.DEC)
- comb += fast_w_dectb.wen.eq(1)
- comb += fast_w_dectb.i_data.eq(new_dec)
- sync += spr_dec.eq(new_dec) # copy into cur_state for decoder
- m.next = "TB_READ"
+ with m.If(self.pause_dec_tb):
+ # if paused, return to reading
+ m.next = "DEC_READ"
+ with m.Else():
+ new_dec = Signal(64)
+ # TODO: MSR.LPCR 32-bit decrement mode
+ comb += new_dec.eq(state_r_dectb.o_data - 1)
+ comb += state_w_dectb.wen.eq(1<<StateRegs.DEC)
+ comb += state_w_dectb.i_data.eq(new_dec)
+ # copy to cur_state for decoder, for an interrupt
+ sync += spr_dec.eq(new_dec)
+ 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"
+ comb += state_r_dectb.ren.eq(1<<StateRegs.TB)
+ with m.If(~self.pause_dec_tb):
+ m.next = "TB_WRITE"
# waits for read TB to arrive, initiates write of current TB
+ # respects if dec/tb writing has been paused
with m.State("TB_WRITE"):
- new_tb = Signal(64)
- comb += new_tb.eq(fast_r_dectb.o_data + 1)
- comb += fast_w_dectb.addr.eq(FastRegs.TB)
- comb += fast_w_dectb.wen.eq(1)
- comb += fast_w_dectb.i_data.eq(new_tb)
- m.next = "DEC_READ"
+ with m.If(self.pause_dec_tb):
+ # if paused, return to reading
+ m.next = "TB_READ"
+ with m.Else():
+ new_tb = Signal(64)
+ comb += new_tb.eq(state_r_dectb.o_data + 1)
+ comb += state_w_dectb.wen.eq(1<<StateRegs.TB)
+ comb += state_w_dectb.i_data.eq(new_tb)
+ m.next = "DEC_READ"
return m
# reset current state if core reset requested
with m.If(core_rst):
m.d.sync += self.cur_state.eq(0)
+ # and, sigh, set configured values, which are also done in regfile
+ # XXX ??? what the hell is the shift for??
+ m.d.sync += self.cur_state.pc.eq(self.core.pc_at_reset)
+ m.d.sync += self.cur_state.msr.eq(self.core.msr_at_reset)
# check halted condition: requested PC to execute matches DMI stop addr
# and immediately stop. address of 0xffff_ffff_ffff_ffff can never
def external_ports(self):
if self.microwatt_compat:
ports = [self.core.o.core_terminate_o,
+ self.ext_irq,
self.alt_reset, # not connected yet
+ self.nia, self.insn, self.nia_req, self.msr_o,
+ self.ldst_req, self.ldst_addr,
ClockSignal(),
ResetSignal(),
]
ports += list(self.dbg.dmi.ports())
# for dbus/ibus microwatt, exclude err btw and cti
for name, sig in self.imem.ibus.fields.items():
- if name not in ['err', 'bte', 'cti']:
+ if name not in ['err', 'bte', 'cti', 'adr']:
ports.append(sig)
for name, sig in self.core.l0.cmpi.wb_bus().fields.items():
- if name not in ['err', 'bte', 'cti']:
+ if name not in ['err', 'bte', 'cti', 'adr']:
ports.append(sig)
+ # microwatt non-compliant with wishbone
+ ports.append(self.ibus_adr)
+ ports.append(self.dbus_adr)
return ports
ports = self.pc_i.ports()
return list(self)
+class TestIssuerInternal(TestIssuerBase):
+ """TestIssuer - reads instructions from TestMemory and issues them
-# Fetch Finite State Machine.
-# WARNING: there are currently DriverConflicts but it's actually working.
-# TODO, here: everything that is global in nature, information from the
-# main TestIssuerInternal, needs to move to either ispec() or ospec().
-# not only that: TestIssuerInternal.imem can entirely move into here
-# because imem is only ever accessed inside the FetchFSM.
-class FetchFSM(ControlBase):
- def __init__(self, allow_overlap, svp64_en, imem, core_rst,
- pdecode2, cur_state,
- dbg, core, svstate, nia, is_svp64_mode):
- self.allow_overlap = allow_overlap
- self.svp64_en = svp64_en
- self.imem = imem
- self.core_rst = core_rst
- self.pdecode2 = pdecode2
- self.cur_state = cur_state
- self.dbg = dbg
- self.core = core
- self.svstate = svstate
- self.nia = nia
- self.is_svp64_mode = is_svp64_mode
-
- # set up pipeline ControlBase and allocate i/o specs
- # (unusual: normally done by the Pipeline API)
- super().__init__(stage=self)
- self.p.i_data, self.n.o_data = self.new_specs(None)
- self.i, self.o = self.p.i_data, self.n.o_data
-
- # next 3 functions are Stage API Compliance
- def setup(self, m, i):
- pass
-
- def ispec(self):
- return FetchInput()
-
- def ospec(self):
- return FetchOutput()
+ efficiency and speed is not the main goal here: functional correctness
+ and code clarity is. optimisations (which almost 100% interfere with
+ easy understanding) come later.
+ """
- def elaborate(self, platform):
+ def fetch_fsm(self, m, dbg, core, core_rst, nia, is_svp64_mode,
+ fetch_pc_o_ready, fetch_pc_i_valid,
+ fetch_insn_o_valid, fetch_insn_i_ready):
"""fetch FSM
this FSM performs fetch of raw instruction data, partial-decodes
it 32-bit at a time to detect SVP64 prefixes, and will optionally
read a 2nd 32-bit quantity if that occurs.
"""
- m = super().elaborate(platform)
-
- dbg = self.dbg
- core = self.core
- pc = self.i.pc
- msr = self.i.msr
- svstate = self.svstate
- nia = self.nia
- is_svp64_mode = self.is_svp64_mode
- fetch_pc_o_ready = self.p.o_ready
- fetch_pc_i_valid = self.p.i_valid
- fetch_insn_o_valid = self.n.o_valid
- fetch_insn_i_ready = self.n.i_ready
-
comb = m.d.comb
sync = m.d.sync
pdecode2 = self.pdecode2
cur_state = self.cur_state
- dec_opcode_o = pdecode2.dec.raw_opcode_in # raw opcode
+ dec_opcode_i = pdecode2.dec.raw_opcode_in # raw opcode
+ pc, msr, svstate = cur_state.pc, cur_state.msr, cur_state.svstate
# also note instruction fetch failed
if hasattr(core, "icache"):
with m.FSM(name='fetch_fsm'):
+ # allow fetch to not run at startup due to I-Cache reset not
+ # having time to settle. power-on-reset holds dbg.core_stopped_i
+ with m.State("PRE_IDLE"):
+ with m.If(~dbg.core_stopped_i & ~dbg.core_stop_o & ~core_rst):
+ m.next = "IDLE"
+
# waiting (zzz)
with m.State("IDLE"):
# fetch allowed if not failed and stopped but not stepping
comb += self.imem.a_pc_i.eq(pc)
comb += self.imem.a_i_valid.eq(1)
comb += self.imem.f_i_valid.eq(1)
- # transfer state to output
- sync += cur_state.pc.eq(pc)
- sync += cur_state.svstate.eq(svstate) # and svstate
- sync += cur_state.msr.eq(msr) # and msr
-
m.next = "INSN_READ" # move to "wait for bus" phase
# dummy pause to find out why simulation is not keeping up
with m.If(~svp64.is_svp64_mode):
# with no prefix, store the instruction
# and hand it directly to the next FSM
- sync += dec_opcode_o.eq(insn)
+ sync += dec_opcode_i.eq(insn)
m.next = "INSN_READY"
with m.Else():
# fetch the rest of the instruction from memory
else:
# not SVP64 - 32-bit only
sync += nia.eq(cur_state.pc + 4)
- sync += dec_opcode_o.eq(insn)
+ sync += dec_opcode_i.eq(insn)
+ if self.microwatt_compat:
+ # for verilator debug purposes
+ comb += self.insn.eq(insn)
+ comb += self.nia.eq(cur_state.pc)
+ comb += self.msr_o.eq(cur_state.msr)
+ comb += self.nia_req.eq(1)
m.next = "INSN_READY"
with m.State("INSN_READ2"):
insn = self.imem.f_instr_o
else:
insn = get_insn(self.imem.f_instr_o, cur_state.pc+4)
- sync += dec_opcode_o.eq(insn)
+ sync += dec_opcode_i.eq(insn)
m.next = "INSN_READY"
# TODO: probably can start looking at pdecode2.rm_dec
# here or maybe even in INSN_READ state, if svp64_mode
with m.If(fetch_insn_i_ready):
m.next = "IDLE"
- # whatever was done above, over-ride it if core reset is held
- with m.If(self.core_rst):
- sync += nia.eq(0)
-
- return m
-
-
-class TestIssuerInternal(TestIssuerBase):
- """TestIssuer - reads instructions from TestMemory and issues them
-
- efficiency and speed is not the main goal here: functional correctness
- and code clarity is. optimisations (which almost 100% interfere with
- easy understanding) come later.
- """
def fetch_predicate_fsm(self, m,
pred_insn_i_valid, pred_insn_o_ready,
fetch_failed = Const(0, 1)
flush_needed = False
+ sync += fetch_pc_i_valid.eq(0)
+
with m.FSM(name="issue_fsm"):
+ with m.State("PRE_IDLE"):
+ with m.If(~dbg.core_stop_o & ~core_rst):
+ m.next = "ISSUE_START"
+
# sync with the "fetch" phase which is reading the instruction
# at this point, there is no instruction running, that
# could inadvertently update the PC.
# wait on "core stop" release, before next fetch
# need to do this here, in case we are in a VL==0 loop
with m.If(~dbg.core_stop_o & ~core_rst):
- comb += fetch_pc_i_valid.eq(1) # tell fetch to start
+ sync += fetch_pc_i_valid.eq(1) # tell fetch to start
+ sync += cur_state.pc.eq(dbg.state.pc)
+ sync += cur_state.svstate.eq(dbg.state.svstate)
+ sync += cur_state.msr.eq(dbg.state.msr)
with m.If(fetch_pc_o_ready): # fetch acknowledged us
m.next = "INSN_WAIT"
with m.Else():
# instruction started: must wait till it finishes
with m.State("INSN_ACTIVE"):
- # note changes to MSR, PC and SVSTATE
- # XXX oops, really must monitor *all* State Regfile write
- # ports looking for changes!
+ # note changes to MSR, PC and SVSTATE, and DEC/TB
+ # these last two are done together, and passed to the
+ # DEC/TB FSM
with m.If(self.state_nia.wen & (1 << StateRegs.SVSTATE)):
sync += self.sv_changed.eq(1)
with m.If(self.state_nia.wen & (1 << StateRegs.MSR)):
sync += self.msr_changed.eq(1)
with m.If(self.state_nia.wen & (1 << StateRegs.PC)):
sync += self.pc_changed.eq(1)
+ with m.If((self.state_spr.wen &
+ ((1 << StateRegs.DEC) | (1 << StateRegs.TB))).bool()):
+ comb += self.pause_dec_tb.eq(1)
with m.If(~core_busy_o): # instruction done!
comb += exec_pc_o_valid.eq(1)
with m.If(exec_pc_i_ready):
# Issue is where the VL for-loop # lives. the ready/valid
# signalling is used to communicate between the four.
- # set up Fetch FSM
- fetch = FetchFSM(self.allow_overlap, self.svp64_en,
- self.imem, core_rst, pdecode2, cur_state,
- dbg, core,
- dbg.state.svstate, # combinatorially same
- nia, is_svp64_mode)
- m.submodules.fetch = fetch
- # connect up in/out data to existing Signals
- comb += fetch.p.i_data.pc.eq(dbg.state.pc) # combinatorially same
- comb += fetch.p.i_data.msr.eq(dbg.state.msr) # combinatorially same
- # and the ready/valid signalling
- comb += fetch_pc_o_ready.eq(fetch.p.o_ready)
- comb += fetch.p.i_valid.eq(fetch_pc_i_valid)
- comb += fetch_insn_o_valid.eq(fetch.n.o_valid)
- comb += fetch.n.i_ready.eq(fetch_insn_i_ready)
+ self.fetch_fsm(m, dbg, core, core_rst, nia, is_svp64_mode,
+ fetch_pc_o_ready, fetch_pc_i_valid,
+ fetch_insn_o_valid, fetch_insn_i_ready)
self.issue_fsm(m, core, nia,
dbg, core_rst, is_svp64_mode,
exec_insn_i_valid, exec_insn_o_ready,
exec_pc_o_valid, exec_pc_i_ready)
+ # whatever was done above, over-ride it if core reset is held.
+ # set NIA to pc_at_reset
+ with m.If(core_rst):
+ sync += nia.eq(self.core.pc_at_reset)
+
return m
}
pspec = TestMemPspec(ldst_ifacetype='bare_wb',
imem_ifacetype='bare_wb',
- addr_wid=48,
+ addr_wid=64,
mask_wid=8,
reg_wid=64,
units=units)
projects / soc.git / blobdiff