def elaborate(self, platform):
m = Module()
if self.has_tx:
m.d.comb += self.tx_t.oe.eq(1)
if self.invert_tx:
m.d.comb += self.tx_t.o.eq(~self.tx_o)
else:
m.d.comb += self.tx_t.o.eq(self.tx_o)
if self.has_rx:
if self.invert_rx:
m.submodules += FFSynchronizer(~self.rx_t.i,
self.rx_i,
reset=1)
else:
m.submodules += FFSynchronizer(self.rx_t.i, self.rx_i, reset=1)
return m
def elaborate(self, platform):
m = Module()
# Keep track of how many cycles we'll keep our PHY in reset.
# This is larger than any requirement, in order to work with a broad swathe of PHYs,
# in case a PHY other than the TUSB1310A ever makes it to the market.
cycles_in_reset = int(5e-6 * 50e6)
cycles_spent_in_reset = Signal(range(cycles_in_reset + 1))
# Create versions of our phy_status signals that are observable:
# 1) as an asynchronous inputs for startup pulses
# 2) as a single-cycle pulse, for power-state-change notifications
phy_status = Signal()
# Convert our PHY status signal into a simple, sync-domain signal.
m.submodules += FFSynchronizer(self.phy_status[0] | self.phy_status[1],
phy_status),
with m.FSM():
# STARTUP_RESET -- post configuration, we'll reset the PIPE PHY.
# This is distinct from the PHY's built-in power-on-reset, as we run this
# on every FPGA configuration.
with m.State("STARTUP_RESET"):
m.d.comb += [
self.reset.eq(1),
]
# Once we've extended past a reset time, we can move on.
m.d.sync += cycles_spent_in_reset.eq(cycles_spent_in_reset + 1)
with m.If(cycles_spent_in_reset == cycles_in_reset):
m.next = "DETECT_PHY_STARTUP"
# DETECT_PHY_STARTUP -- post-reset, the PHY should drive its status line high.
# We'll wait for this to happen, so we can track the PHY's progress.
with m.State("DETECT_PHY_STARTUP"):
with m.If(phy_status):
m.next = "WAIT_FOR_STARTUP"
# WAIT_FOR_STARTUP -- we've now detected that the PHY is starting up.
# We'll wait for that startup signal to be de-asserted, indicating that the PHY is ready.
with m.State("WAIT_FOR_STARTUP"):
# For now, we'll start up in P0. This will change once we implement proper RxDetect.
with m.If(~phy_status):
m.next = "READY"
# READY -- our PHY is all started up and ready for use.
# For now, we'll remain here until we're reset.
with m.State("READY"):
m.d.comb += self.ready.eq(1)
return m
def _synchronize_(self, m, output, o_domain="sync", stages=2):
""" Creates a synchronized copy of this interface's I/O. """
# Synchronize our inputs...
m.submodules += [
FFSynchronizer(self.sck,
output.sck,
o_domain=o_domain,
stages=stages),
FFSynchronizer(self.sdi,
output.sdi,
o_domain=o_domain,
stages=stages),
FFSynchronizer(self.cs,
output.cs,
o_domain=o_domain,
stages=stages),
]
# ... and connect our output directly through.
m.d.comb += self.sdo.eq(output.sdo)
def elaborate(self, platform):
m = Module()
pads = self._pads
m.d.comb += [
pads.sbwtck_t.oe.eq(1),
pads.sbwtck_t.o.eq(self.sbwtck),
pads.sbwtdio_t.oe.eq(~self.sbwtd_z),
pads.sbwtdio_t.o.eq(self.sbwtd_o),
]
m.submodules += [
FFSynchronizer(pads.sbwtdio_t.i, self.sbwtd_i),
]
return m
def elaborate(self, platform):
m = Module()
m.d.comb += [
self.pads.clk_t.oe.eq(1),
self.pads.clk_t.o.eq(self.clk),
]
m.submodules += [
FFSynchronizer(self.pads.din_t.i, self.din),
]
if hasattr(self.pads, "osc_t"):
m.d.comb += [
self.pads.osc_t.oe.eq(1),
self.pads.osc_t.o.eq(self.osc),
]
return m
def elaborate(self, platform):
m = Module()
m.submodules += self.analyzer
pins_i = Signal.like(self.pads.i_t.i)
pins_r = Signal.like(self.pads.i_t.i)
m.submodules += FFSynchronizer(self.pads.i_t.i, pins_i)
m.d.sync += pins_r.eq(pins_i)
m.d.comb += [
self.event_source.data.eq(pins_i),
self.event_source.trigger.eq(pins_i != pins_r)
]
return m
def elaborate(self, platform):
m = Module()
m.d.comb += [
self.pads.clock_t.o.eq(0),
self.pads.clock_t.oe.eq(~self.clock_o),
self.pads.data_t.o.eq(0),
self.pads.data_t.oe.eq(~self.data_o),
]
m.submodules += [
FFSynchronizer(self.pads.clock_t.i, self.clock_i, reset=1),
FFSynchronizer(self.pads.data_t.i, self.data_i, reset=1),
]
clock_s = Signal(reset=1)
clock_r = Signal(reset=1)
m.d.sync += [
clock_s.eq(self.clock_i),
clock_r.eq(clock_s),
self.falling.eq(clock_r & ~clock_s),
self.rising.eq(~clock_r & clock_s),
]
return m
def elaborate(self, platform) -> Module:
"""assemble the module"""
m = Module()
comb = m.d.comb
sync = m.d.sync
nrzi = Signal()
nrzi_prev = Signal()
got_edge = Signal()
m.submodules.cdc = FFSynchronizer(self.nrzi_in, nrzi)
sync += nrzi_prev.eq(nrzi)
comb += got_edge.eq(nrzi_prev ^ nrzi)
# we are looking for 10 non changing bits
# and those will be ~900ns long @48kHz
# and if we clock at not more than 100MHz
# the counter will run up to 900ns/10ns = 90
# so 7 bits will suffice for the counter
sync_counter = DividingCounter(divisor=12, width=7)
m.submodules.sync_counter = sync_counter
bit_time = sync_counter.divided_counter_out
with m.FSM():
with m.State("SYNC"):
comb += self.running.eq(0)
sync += [
self.data_out.eq(0),
self.data_out_en.eq(0),
sync_counter.reset_in.eq(0)
]
self.find_bit_timings(m, sync_counter, got_edge)
with m.State("DECODE"):
comb += self.running.eq(1)
self.decode_nrzi(m, bit_time, got_edge, sync_counter)
return m
def elaborate(self, platform):
m = Module()
# The ECP5 SerDes uses a simple feedback mechanism to keep its FIFO clocks in sync
# with the FPGA's fabric. Accordingly, we'll need to capture the output clocks and then
# pass them back to the SerDes; this allows the placer to handle clocking correctly, allows us
# to attach clock constraints for analysis, and allows us to use these clocks for -very- simple tasks.
txoutclk = Signal()
rxoutclk = Signal()
# Internal state.
rx_los = Signal()
rx_lol = Signal()
rx_lsm = Signal()
rx_align = Signal()
rx_bus = Signal(24)
tx_lol = Signal()
tx_bus = Signal(24)
#
# Clock domain crossing.
#
tx_produce_square_wave = Signal()
tx_produce_pattern = Signal()
tx_pattern = Signal(20)
m.submodules += [
# Transmit control synchronization.
FFSynchronizer(self.tx_produce_square_wave, tx_produce_square_wave, o_domain="tx"),
FFSynchronizer(self.tx_produce_pattern, tx_produce_pattern, o_domain="tx"),
FFSynchronizer(self.tx_pattern, tx_pattern, o_domain="tx"),
# Receive control synchronization.
FFSynchronizer(self.rx_align, rx_align, o_domain="rx"),
FFSynchronizer(rx_los, self.rx_idle, o_domain="sync"),
]
#
# Clocking / reset control.
#
# The SerDes needs to be brought up gradually; we'll do that here.
m.submodules.reset_sequencer = reset = ECP5ResetSequencer()
m.d.comb += [
reset.tx_pll_locked .eq(~tx_lol),
reset.rx_pll_locked .eq(~rx_lol)
]
# Create a local transmit domain, for our transmit-side hardware.
m.domains.tx = ClockDomain()
m.d.comb += ClockSignal("tx").eq(txoutclk)
m.submodules += [
ResetSynchronizer(ResetSignal("sync"), domain="tx"),
FFSynchronizer(~ResetSignal("tx"), self.tx_ready)
]
# Create the same setup, buf for the receive side.
m.domains.rx = ClockDomain()
m.d.comb += ClockSignal("rx").eq(rxoutclk)
m.submodules += [
ResetSynchronizer(ResetSignal("sync"), domain="rx"),
FFSynchronizer(~ResetSignal("rx"), self.rx_ready)
]
#
# Core SerDes instantiation.
#
serdes_params = dict(
# DCU — power management
p_D_MACROPDB = "0b1",
p_D_IB_PWDNB = "0b1", # undocumented (required for RX)
p_D_TXPLL_PWDNB = "0b1",
i_D_FFC_MACROPDB = 1,
# DCU — reset
i_D_FFC_MACRO_RST = ResetSignal("sync"),
i_D_FFC_DUAL_RST = ResetSignal("sync"),
# DCU — clocking
i_D_REFCLKI = self._pll.refclk,
o_D_FFS_PLOL = tx_lol,
p_D_REFCK_MODE = {
25: "0b100",
20: "0b000",
16: "0b010",
10: "0b001",
8: "0b011"}[self._pll.config["mult"]],
p_D_TX_MAX_RATE = "5.0", # 5.0 Gbps
p_D_TX_VCO_CK_DIV = {
32: "0b111",
16: "0b110",
8: "0b101",
4: "0b100",
2: "0b010",
1: "0b000"}[1], # DIV/1
p_D_BITCLK_LOCAL_EN = "0b1", # Use clock from local PLL
# Clock multiplier unit configuration
p_D_CMUSETBIASI = "0b00", # begin undocumented (10BSER sample code used)
p_D_CMUSETI4CPP = "0d3",
p_D_CMUSETI4CPZ = "0d3",
p_D_CMUSETI4VCO = "0b00",
p_D_CMUSETICP4P = "0b01",
p_D_CMUSETICP4Z = "0b101",
p_D_CMUSETINITVCT = "0b00",
p_D_CMUSETISCL4VCO = "0b000",
p_D_CMUSETP1GM = "0b000",
p_D_CMUSETP2AGM = "0b000",
p_D_CMUSETZGM = "0b000",
p_D_SETIRPOLY_AUX = "0b01",
p_D_SETICONST_AUX = "0b01",
p_D_SETIRPOLY_CH = "0b01",
p_D_SETICONST_CH = "0b10",
p_D_SETPLLRC = "0d1",
p_D_RG_EN = "0b0",
p_D_RG_SET = "0b00",
p_D_REQ_ISET = "0b011",
p_D_PD_ISET = "0b11", # end undocumented
# DCU — FIFOs
p_D_LOW_MARK = "0d4", # Clock compensation FIFO low water mark (mean=8)
p_D_HIGH_MARK = "0d12", # Clock compensation FIFO high water mark (mean=8)
# CHX common ---------------------------------------------------------------------------
# CHX — protocol
p_CHX_PROTOCOL = "10BSER",
p_CHX_UC_MODE = "0b1",
p_CHX_ENC_BYPASS = "******", # Use the 8b10b encoder
p_CHX_DEC_BYPASS = "******", # Use the 8b10b decoder
# CHX receive --------------------------------------------------------------------------
# CHX RX — power management
p_CHX_RPWDNB = "0b1",
i_CHX_FFC_RXPWDNB = 1,
# CHX RX — reset
i_CHX_FFC_RRST = ~self.rx_enable | reset.serdes_rx_reset,
i_CHX_FFC_LANE_RX_RST = ~self.rx_enable | reset.pcs_reset,
# CHX RX — input
i_CHX_HDINP = self._rx_pads.p,
i_CHX_HDINN = self._rx_pads.n,
p_CHX_REQ_EN = "0b1", # Enable equalizer
p_CHX_REQ_LVL_SET = "0b01",
p_CHX_RX_RATE_SEL = "0d09", # Equalizer pole position
p_CHX_RTERM_RX = {
"5k-ohms": "0b00000",
"80-ohms": "0b00001",
"75-ohms": "0b00100",
"70-ohms": "0b00110",
"60-ohms": "0b01011",
"50-ohms": "0b10011",
"46-ohms": "0b11001",
"wizard-50-ohms": "0d22"}["wizard-50-ohms"],
p_CHX_RXIN_CM = "0b11", # CMFB (wizard value used)
p_CHX_RXTERM_CM = "0b10", # RX Input (wizard value used)
# CHX RX — clocking
i_CHX_RX_REFCLK = self._pll.refclk,
o_CHX_FF_RX_PCLK = rxoutclk,
i_CHX_FF_RXI_CLK = ClockSignal("rx"),
p_CHX_CDR_MAX_RATE = "5.0", # 5.0 Gbps
p_CHX_RX_DCO_CK_DIV = {
32: "0b111",
16: "0b110",
8: "0b101",
4: "0b100",
2: "0b010",
1: "0b000"}[1], # DIV/1
p_CHX_RX_GEAR_MODE = "0b1", # 1:2 gearbox
p_CHX_FF_RX_H_CLK_EN = "0b1", # enable DIV/2 output clock
p_CHX_FF_RX_F_CLK_DIS = "0b1", # disable DIV/1 output clock
p_CHX_SEL_SD_RX_CLK = "0b1", # FIFO driven by recovered clock
p_CHX_AUTO_FACQ_EN = "0b1", # undocumented (wizard value used)
p_CHX_AUTO_CALIB_EN = "0b1", # undocumented (wizard value used)
p_CHX_PDEN_SEL = "0b0", # phase detector disabled on LOS
p_CHX_DCOATDCFG = "0b00", # begin undocumented (sample code used)
p_CHX_DCOATDDLY = "0b00",
p_CHX_DCOBYPSATD = "0b1",
p_CHX_DCOCALDIV = "0b000",
p_CHX_DCOCTLGI = "0b011",
p_CHX_DCODISBDAVOID = "0b0",
p_CHX_DCOFLTDAC = "0b00",
p_CHX_DCOFTNRG = "0b001",
p_CHX_DCOIOSTUNE = "0b010",
p_CHX_DCOITUNE = "0b00",
p_CHX_DCOITUNE4LSB = "0b010",
p_CHX_DCOIUPDNX2 = "0b1",
p_CHX_DCONUOFLSB = "0b100",
p_CHX_DCOSCALEI = "0b01",
p_CHX_DCOSTARTVAL = "0b010",
p_CHX_DCOSTEP = "0b11", # end undocumented
# CHX RX — loss of signal
o_CHX_FFS_RLOS = rx_los,
p_CHX_RLOS_SEL = "0b1",
p_CHX_RX_LOS_EN = "0b0",
p_CHX_RX_LOS_LVL = "0b101", # Lattice "TBD" (wizard value used)
p_CHX_RX_LOS_CEQ = "0b11", # Lattice "TBD" (wizard value used)
p_CHX_RX_LOS_HYST_EN = "0b1",
# CHX RX — loss of lock
o_CHX_FFS_RLOL = rx_lol,
# CHX RX — link state machine
# Note that Lattice Diamond needs these in their provided bases (and string lengths!).
# Changing their bases will work with the open toolchain, but will make Diamond mad.
i_CHX_FFC_SIGNAL_DETECT = rx_align,
o_CHX_FFS_LS_SYNC_STATUS= rx_lsm,
p_CHX_ENABLE_CG_ALIGN = "0b1",
p_CHX_UDF_COMMA_MASK = "0x0ff", # compare the 8 lsbs
p_CHX_UDF_COMMA_A = "0x003", # "0b0000000011", # K28.1, K28.5 and K28.7
p_CHX_UDF_COMMA_B = "0x07c", # "0b0001111100", # K28.1, K28.5 and K28.7
p_CHX_CTC_BYPASS = "******", # bypass CTC FIFO
p_CHX_MIN_IPG_CNT = "0b11", # minimum interpacket gap of 4
p_CHX_MATCH_2_ENABLE = "0b0", # 2 character skip matching
p_CHX_MATCH_4_ENABLE = "0b0", # 4 character skip matching
p_CHX_CC_MATCH_1 = "0x000",
p_CHX_CC_MATCH_2 = "0x000",
p_CHX_CC_MATCH_3 = "0x000",
p_CHX_CC_MATCH_4 = "0x000",
# CHX RX — data
**{"o_CHX_FF_RX_D_%d" % n: rx_bus[n] for n in range(len(rx_bus))},
# CHX transmit -------------------------------------------------------------------------
# CHX TX — power management
p_CHX_TPWDNB = "0b1",
i_CHX_FFC_TXPWDNB = 1,
# CHX TX — reset
i_D_FFC_TRST = ~self.tx_enable | reset.serdes_tx_reset,
i_CHX_FFC_LANE_TX_RST = ~self.tx_enable | reset.pcs_reset,
# CHX TX — output
o_CHX_HDOUTP = self._tx_pads.p,
o_CHX_HDOUTN = self._tx_pads.n,
p_CHX_TXAMPLITUDE = "0d1000", # 1000 mV
p_CHX_RTERM_TX = {
"5k-ohms": "0b00000",
"80-ohms": "0b00001",
"75-ohms": "0b00100",
"70-ohms": "0b00110",
"60-ohms": "0b01011",
"50-ohms": "0b10011",
"46-ohms": "0b11001",
"wizard-50-ohms": "0d19"}["50-ohms"],
p_CHX_TDRV_SLICE0_CUR = "0b011", # 400 uA
p_CHX_TDRV_SLICE0_SEL = "0b01", # main data
p_CHX_TDRV_SLICE1_CUR = "0b000", # 100 uA
p_CHX_TDRV_SLICE1_SEL = "0b00", # power down
p_CHX_TDRV_SLICE2_CUR = "0b11", # 3200 uA
p_CHX_TDRV_SLICE2_SEL = "0b01", # main data
p_CHX_TDRV_SLICE3_CUR = "0b10", # 2400 uA
p_CHX_TDRV_SLICE3_SEL = "0b01", # main data
p_CHX_TDRV_SLICE4_CUR = "0b00", # 800 uA
p_CHX_TDRV_SLICE4_SEL = "0b00", # power down
p_CHX_TDRV_SLICE5_CUR = "0b00", # 800 uA
p_CHX_TDRV_SLICE5_SEL = "0b00", # power down
# CHX TX — clocking
o_CHX_FF_TX_PCLK = txoutclk,
i_CHX_FF_TXI_CLK = ClockSignal("tx"),
p_CHX_TX_GEAR_MODE = "0b1", # 1:2 gearbox
p_CHX_FF_TX_H_CLK_EN = "0b1", # enable DIV/2 output clock
p_CHX_FF_TX_F_CLK_DIS = "0b1", # disable DIV/1 output clock
# CHX TX — data
**{"i_CHX_FF_TX_D_%d" % n: tx_bus[n] for n in range(len(tx_bus))},
# SCI interface.
#**{"i_D_SCIWDATA%d" % n: sci.sci_wdata[n] for n in range(8)},
#**{"i_D_SCIADDR%d" % n: sci.sci_addr[n] for n in range(6)},
#**{"o_D_SCIRDATA%d" % n: sci.sci_rdata[n] for n in range(8)},
#i_D_SCIENAUX = sci.dual_sel,
#i_D_SCISELAUX = sci.dual_sel,
#i_CHX_SCIEN = sci.chan_sel,
#i_CHX_SCISEL = sci.chan_sel,
#i_D_SCIRD = sci.sci_rd,
#i_D_SCIWSTN = sci.sci_wrn,
# Out-of-band signaling Rx support.
p_CHX_LDR_RX2CORE_SEL = "0b1", # Enables low-speed out-of-band input.
o_CHX_LDR_RX2CORE = self.rx_gpio,
# Out-of-band signaling Tx support.
p_CHX_LDR_CORE2TX_SEL = "0b0", # Uses CORE2TX_EN to enable out-of-band output.
i_CHX_LDR_CORE2TX = self.tx_gpio,
i_CHX_FFC_LDR_CORE2TX_EN = self.tx_gpio_en
)
# Translate the 'CHX' string to the correct channel name in each of our SerDes parameters,
# and create our SerDes instance.
serdes_params = {k.replace("CHX", f"CH{self._channel}"):v for (k,v) in serdes_params.items()}
m.submodules.serdes = serdes = Instance("DCUA", **serdes_params)
# Bind our SerDes to the correct location inside the FPGA.
serdes.attrs["LOC"] = "DCU{}".format(self._dual)
serdes.attrs["CHAN"] = "CH{}".format(self._channel)
serdes.attrs["BEL"] = "X42/Y71/DCU"
#
# TX and RX datapaths (SerDes <-> stream conversion)
#
sink = self.sink
source = self.source
m.d.comb += [
# Grab our received data directly from our SerDes; modifying things to match the
# SerDes Rx bus layout, which squishes status signals between our two geared words.
source.data[0: 8] .eq(rx_bus[ 0: 8]),
source.data[8:16] .eq(rx_bus[12:20]),
source.ctrl[0] .eq(rx_bus[8]),
source.ctrl[1] .eq(rx_bus[20]),
source.valid .eq(1),
# Stick the data we'd like to transmit into the SerDes; again modifying things to match
# the transmit bus layout.
tx_bus[ 0: 8] .eq(sink.data[0: 8]),
tx_bus[12:20] .eq(sink.data[8:16]),
tx_bus[8] .eq(sink.ctrl[0]),
tx_bus[20] .eq(sink.ctrl[1]),
sink.ready .eq(1)
]
return m
def elaborate(self, platform: Platform) -> Module:
m = Module()
# Do TX CDC
# FFSynchronizer
if False:
m.submodules += FFSynchronizer(Cat(self.tx_symbol,
self.tx_set_disp, self.tx_disp,
self.tx_e_idle),
Cat(self.__lane.tx_symbol,
self.__lane.tx_set_disp,
self.__lane.tx_disp,
self.__lane.tx_e_idle),
o_domain="tx",
stages=4)
# No CDC
# TODO: Check if this actually works
if False:
m.d.comb += Cat(self.__lane.tx_symbol, self.__lane.tx_set_disp,
self.__lane.tx_disp, self.__lane.tx_e_idle).eq(
Cat(self.tx_symbol, self.tx_set_disp,
self.tx_disp, self.tx_e_idle))
# AsyncFIFOBuffered
if True:
tx_fifo = m.submodules.tx_fifo = AsyncFIFOBuffered(
width=self.ratio * 12, depth=8, r_domain="tx", w_domain="rx")
m.d.comb += tx_fifo.w_data.eq(
Cat(self.tx_symbol, self.tx_set_disp, self.tx_disp,
self.tx_e_idle))
m.d.comb += Cat(self.__lane.tx_symbol, self.__lane.tx_set_disp,
self.__lane.tx_disp,
self.__lane.tx_e_idle).eq(tx_fifo.r_data)
m.d.comb += tx_fifo.r_en.eq(1)
m.d.comb += tx_fifo.w_en.eq(1)
# Testing symbols
if False:
m.d.comb += self.__lane.tx_symbol.eq(Cat(Ctrl.COM, D(10, 2)))
self.slip = SymbolSlip(symbol_size=10,
word_size=self.__lane.ratio,
comma=Cat(Ctrl.COM, 1))
m.submodules += self.slip
m.d.comb += [
self.slip.en.eq(self.rx_align),
self.slip.i.eq(
Cat((self.__lane.rx_symbol.word_select(n, 9),
self.__lane.rx_valid[n])
for n in range(self.__lane.ratio))),
self.rx_symbol.eq(
Cat(
Part(self.slip.o, 10 * n, 9)
for n in range(self.__lane.ratio))),
self.rx_valid.eq(
Cat(self.slip.o[10 * n + 9]
for n in range(self.__lane.ratio))),
]
return m
def elaborate(self, platform):
m = Module()
sync = m.d.sync
comb = m.d.comb
self.BAR = Signal()
sync += self.BAR.eq(~self.BAR)
self.tms = tms = Signal.like(self.port.tms)
self.tck = tck = Signal.like(self.port.tck)
self.tdi = tdi = Signal.like(self.port.tdi)
self.tdo = tdo = Signal.like(self.port.tdo)
m.submodules += [
FFSynchronizer(self.port.tms, tms),
FFSynchronizer(self.port.tck, tck),
FFSynchronizer(self.port.tdi, tdi),
]
prev_tck = Signal()
self.rising_tck = rising_tck = Signal()
self.falling_tck = falling_tck = Signal()
sync += prev_tck.eq(tck)
comb += [
rising_tck.eq((~prev_tck) & tck),
falling_tck.eq(prev_tck & (~tck)),
]
self.ir = Signal(JtagIR)
assert self.ir.width == 5 # Spike
self.dr = Signal(max([len(v) for _, v in self.regs.items()]))
self.DATA_WRITE = Signal(debug_module_register_len(JtagIR.DMI))
self.DATA_READ = Signal(debug_module_register_len(JtagIR.DMI))
self.DMI_WRITE = Signal(32)
# TODO
for ir, record in self.regs.items():
sync += record.update.eq(0)
sync += record.capture.eq(0)
with m.FSM() as jtag_fsm:
with m.State("TEST-LOGIC-RESET"):
with m.If(rising_tck & ~tms):
sync += self.ir.eq(self.ir_reset)
m.next = "RUN-TEST-IDLE"
with m.State("RUN-TEST-IDLE"):
with m.If(rising_tck & tms):
m.next = "SELECT-DR-SCAN"
with m.State("SELECT-DR-SCAN"):
with m.If(rising_tck):
with m.If(tms):
m.next = "SELECT-IR-SCAN" # IR path
with m.Else():
m.next = "CAPTURE-DR" # DR path
# DR path
with m.State("CAPTURE-DR"):
with m.Switch(self.ir):
for ir, record in self.regs.items():
with m.Case(ir):
sync += self.DATA_READ.eq(record.r)
sync += self.dr.eq(record.r)
sync += record.capture.eq(rising_tck)
with m.If(rising_tck):
with m.If(tms):
m.next = "EXIT1-DR"
with m.Else():
m.next = "SHIFT-DR"
with m.State("SHIFT-DR"):
with m.If(falling_tck):
sync += self.port.tdo.eq(self.dr[0])
with m.Switch(self.ir):
for ir, record in self.regs.items():
with m.Case(ir):
with m.If(rising_tck):
sync += self.dr.eq(
Cat(self.dr[1:len(record.r)], tdi))
# below is not enough, as it may effect in garbage
# sync += self.dr.eq(Cat(self.dr[1:], tdi))
with m.If(rising_tck & tms):
m.next = "EXIT1-DR"
with m.State("EXIT1-DR"):
sync += self.port.tdo.eq(0) # TODO
with m.If(rising_tck):
with m.If(tms):
m.next = "UPDATE-DR"
with m.Else():
m.next = "PAUSE-DR"
with m.State("PAUSE-DR"):
with m.If(rising_tck & tms):
m.next = "EXIT2-DR"
with m.State("EXIT2-DR"):
with m.If(rising_tck):
with m.If(tms):
m.next = "UPDATE-DR"
with m.Else():
m.next = "SHIFT-DR"
with m.State("UPDATE-DR"):
with m.Switch(self.ir):
for ir, record in self.regs.items():
with m.Case(ir):
sync += self.DATA_WRITE.eq(self.dr)
with m.If(ir == JtagIR.DMI):
sync += self.DMI_WRITE.eq(self.dr[2:34])
sync += record.w.eq(self.dr)
sync += record.update.eq(falling_tck)
with m.If(rising_tck):
with m.If(tms):
m.next = "SELECT-DR-SCAN"
with m.Else():
m.next = "RUN-TEST-IDLE"
# IR path
with m.State("SELECT-IR-SCAN"):
with m.If(rising_tck):
with m.If(tms):
m.next = "TEST-LOGIC-RESET"
with m.Else():
m.next = "CAPTURE-IR"
with m.State("CAPTURE-IR"):
sync += self.ir.eq(JtagIR.IDCODE)
with m.If(rising_tck):
with m.If(tms):
m.next = "EXIT1-IR"
with m.Else():
m.next = "SHIFT-IR"
with m.State("SHIFT-IR"):
with m.If(falling_tck):
sync += self.port.tdo.eq(self.ir[0])
with m.If(rising_tck):
sync += self.ir.eq(Cat(self.ir[1:], tdi))
comb += tdo.eq(self.ir[0])
with m.If(rising_tck & tms):
m.next = "EXIT1-IR"
with m.State("EXIT1-IR"):
sync += self.port.tdo.eq(0) # TODO
with m.If(rising_tck):
with m.If(tms):
m.next = "UPDATE-IR"
with m.Else():
m.next = "PAUSE-IR"
with m.State("PAUSE-IR"):
with m.If(rising_tck & tms):
m.next = "EXIT2-IR"
with m.State("EXIT2-IR"):
with m.If(rising_tck & tms):
m.next = "UPDATE-IR"
with m.State("UPDATE-IR"):
with m.If(rising_tck):
with m.If(tms):
m.next = "SELECT-IR-SCAN"
with m.Else():
m.next = "RUN-TEST-IDLE"
return m
def elaborate(self, platform):
m = Module()
pads = self._pads
if hasattr(pads, "ce_t"):
m.d.comb += [
pads.ce_t.oe.eq(1),
pads.ce_t.o.eq(~self.ce),
]
if hasattr(pads, "oe_t"):
m.d.comb += [
pads.oe_t.oe.eq(1),
pads.oe_t.o.eq(~self.oe),
]
if hasattr(pads, "we_t"):
m.d.comb += [
pads.we_t.oe.eq(1),
pads.we_t.o.eq(~self.we),
]
m.d.comb += [
pads.dq_t.oe.eq(~self.oe),
pads.dq_t.o.eq(self.d),
]
m.submodules += FFSynchronizer(pads.dq_t.i, self.q)
m.d.comb += [
pads.a_t.oe.eq(1),
pads.a_t.o.eq(self.a), # directly drive low bits
]
if hasattr(pads, "a_clk_t") and hasattr(pads, "a_si_t"):
a_clk = Signal(reset=1)
a_si = Signal()
a_lat = Signal(reset=0) if hasattr(pads, "a_lat_t") else None
m.d.comb += [
pads.a_clk_t.oe.eq(1),
pads.a_clk_t.o.eq(a_clk),
pads.a_si_t.oe.eq(1),
pads.a_si_t.o.eq(a_si),
]
if a_lat is not None:
m.d.comb += [pads.a_lat_t.oe.eq(1), pads.a_lat_t.o.eq(a_lat)]
# "sa" is the sliced|shifted address, referring to the top-most bits
sa_input = self.a[len(pads.a_t.o):]
# This represents a buffer of those high address bits,
# not to be confused with the latch pin.
sa_latch = Signal(self.a_bits - len(pads.a_t.o))
sh_cyc = math.ceil(platform.default_clk_frequency / self._sh_freq)
timer = Signal(range(sh_cyc), reset=sh_cyc - 1)
count = Signal(range(len(sa_latch) + 1))
first = Signal(reset=1)
with m.FSM():
with m.State("READY"):
m.d.sync += first.eq(0)
with m.If((sa_latch == sa_input) & ~first):
m.d.comb += self.rdy.eq(1)
with m.Else():
m.d.sync += count.eq(len(sa_latch))
m.d.sync += sa_latch.eq(sa_input)
m.next = "SHIFT"
with m.State("SHIFT"):
with m.If(timer == 0):
m.d.sync += timer.eq(timer.reset)
m.d.sync += a_clk.eq(~a_clk)
with m.If(a_clk):
m.d.sync += a_si.eq(sa_latch[-1])
m.d.sync += count.eq(count - 1)
with m.Else():
m.d.sync += sa_latch.eq(sa_latch.rotate_left(1))
with m.If(count == 0):
if a_lat is None:
m.next = "READY"
else:
m.next = "LATCH-1"
with m.Else():
m.d.sync += timer.eq(timer - 1)
if a_lat is not None:
with m.State("LATCH-1"):
m.d.sync += a_lat.eq(1)
with m.If(timer == 0):
m.d.sync += timer.eq(timer.reset)
m.next = "LATCH-2"
with m.Else():
m.d.sync += timer.eq(timer - 1)
with m.State("LATCH-2"):
with m.If(timer == 0):
m.d.sync += timer.eq(timer.reset)
m.d.sync += a_lat.eq(0)
m.next = "READY"
with m.Else():
m.d.sync += timer.eq(timer - 1)
else:
m.d.comb += self.rdy.eq(1)
return m
def elaborate(self, platform):
m = Module()
sync_pulse = Signal(8)
da_reset_shifter = Signal()
da_reset_bitstuff = Signal(
) # Need to reset the bit stuffer 1 cycle after the shifter.
stall = Signal()
# These signals are set during the sync pulse
sp_reset_bitstuff = Signal()
sp_reset_shifter = Signal()
sp_bit = Signal()
sp_o_data_strobe = Signal()
# 12MHz domain
bitstuff_valid_data = Signal()
# Keep a Gray counter around to smoothly transition between states
state_gray = Signal(2)
state_data = Signal()
state_sync = Signal()
#
# Transmit gearing.
#
m.submodules.shifter = shifter = TxShifter(width=8)
m.d.comb += [
shifter.i_data.eq(self.i_data_payload),
shifter.i_enable.eq(~stall),
shifter.i_clear.eq(da_reset_shifter | sp_reset_shifter)
]
#
# Bit-stuffing and NRZI.
#
bitstuff = ResetInserter(da_reset_bitstuff)(TxBitstuffer())
m.submodules.bitstuff = bitstuff
m.submodules.nrzi = nrzi = TxNRZIEncoder()
#
# Transmit controller.
#
m.d.comb += [
# Send a data strobe when we're two bits from the end of the sync pulse.
# This is because the pipeline takes two bit times, and we want to ensure the pipeline
# has spooled up enough by the time we're there.
bitstuff.i_data.eq(shifter.o_data),
stall.eq(bitstuff.o_stall),
sp_bit.eq(sync_pulse[0]),
sp_reset_bitstuff.eq(sync_pulse[0]),
# The shifter has one clock cycle of latency, so reset it
# one cycle before the end of the sync byte.
sp_reset_shifter.eq(sync_pulse[1]),
sp_o_data_strobe.eq(sync_pulse[5]),
state_data.eq(state_gray[0] & state_gray[1]),
state_sync.eq(state_gray[0] & ~state_gray[1]),
self.fit_oe.eq(state_data | state_sync),
self.fit_dat.eq((state_data & shifter.o_data & ~bitstuff.o_stall)
| sp_bit),
self.o_data_strobe.eq(state_data & shifter.o_get & ~stall
& self.i_oe),
]
# If we reset the shifter, then o_empty will go high on the next cycle.
#
m.d.usb += [
# If the shifter runs out of data, percolate the "reset" signal to the
# shifter, and then down to the bitstuffer.
# da_reset_shifter.eq(~stall & shifter.o_empty & ~da_stalled_reset),
# da_stalled_reset.eq(da_reset_shifter),
# da_reset_bitstuff.eq(~stall & da_reset_shifter),
bitstuff_valid_data.eq(~stall & shifter.o_get & self.i_oe),
]
with m.FSM(domain="usb"):
with m.State('IDLE'):
with m.If(self.i_oe):
m.d.usb += [sync_pulse.eq(1 << 7), state_gray.eq(0b01)]
m.next = "SEND_SYNC"
with m.Else():
m.d.usb += state_gray.eq(0b00)
with m.State('SEND_SYNC'):
m.d.usb += sync_pulse.eq(sync_pulse >> 1)
with m.If(sync_pulse[0]):
m.d.usb += state_gray.eq(0b11)
m.next = "SEND_DATA"
with m.Else():
m.d.usb += state_gray.eq(0b01)
with m.State('SEND_DATA'):
with m.If(~self.i_oe & shifter.o_empty & ~bitstuff.o_stall):
with m.If(bitstuff.o_will_stall):
m.next = 'STUFF_LAST_BIT'
with m.Else():
m.d.usb += state_gray.eq(0b10)
m.next = 'IDLE'
with m.Else():
m.d.usb += state_gray.eq(0b11)
with m.State('STUFF_LAST_BIT'):
m.d.usb += state_gray.eq(0b10)
m.next = 'IDLE'
# 48MHz domain
# NRZI encoding
nrzi_dat = Signal()
nrzi_oe = Signal()
# Cross the data from the 12MHz domain to the 48MHz domain
cdc_dat = FFSynchronizer(self.fit_dat,
nrzi_dat,
o_domain="usb_io",
stages=3)
cdc_oe = FFSynchronizer(self.fit_oe,
nrzi_oe,
o_domain="usb_io",
stages=3)
m.submodules += [cdc_dat, cdc_oe]
m.d.comb += [
nrzi.i_valid.eq(self.i_bit_strobe),
nrzi.i_data.eq(nrzi_dat),
nrzi.i_oe.eq(nrzi_oe),
self.o_usbp.eq(nrzi.o_usbp),
self.o_usbn.eq(nrzi.o_usbn),
self.o_oe.eq(nrzi.o_oe),
]
return m
from amaranth import *
from amaranth.lib.cdc import FFSynchronizer
from amaranth.cli import main
i, o = Signal(name="i"), Signal(name="o")
m = Module()
m.submodules += FFSynchronizer(i, o)
if __name__ == "__main__":
main(m, ports=[i, o])
def elaborate(self, platform):
m = Module()
#
# Reference clock selection.
#
# If we seem to have a raw pin record, we'll assume we're being passed the external REFCLK.
# We'll instantiate an instance that captures the reference clock signal.
if hasattr(self._refclk, 'p'):
refclk = Signal()
m.submodules.refclk_input = refclk_in = Instance("EXTREFB",
i_REFCLKP = self._refclk.p,
i_REFCLKN = self._refclk.n,
o_REFCLKO = refclk,
p_REFCK_PWDNB = "0b1",
p_REFCK_RTERM = "0b1", # 100 Ohm
)
refclk_in.attrs["LOC"] = f"EXTREF{self._refclk_num}"
# Otherwise, we'll accept the reference clock directly.
else:
refclk = self._refclk
#
# Raw serdes.
#
pll_config = ECP5SerDesPLLConfiguration(refclk, refclk_freq=self._refclk_frequency, linerate=5e9)
serdes = ECP5SerDes(
pll_config = pll_config,
tx_pads = self._tx_pads,
rx_pads = self._rx_pads,
channel = self._channel,
)
m.submodules.serdes = serdes
m.d.comb += [
serdes.train_equalizer .eq(self.train_equalizer),
self.ready .eq(serdes.tx_ready & serdes.rx_ready)
]
#
# Transmit datapath.
#
m.submodules.tx_datapath = tx_datapath = TransmitPreprocessing()
m.d.comb += [
serdes.tx_idle .eq(self.tx_idle),
serdes.tx_enable .eq(self.enable),
tx_datapath.sink .stream_eq(self.sink),
serdes.sink .stream_eq(tx_datapath.source),
serdes.tx_gpio_en .eq(self.use_tx_as_gpio),
serdes.tx_gpio .eq(self.tx_gpio)
]
#
# Receive datapath.
#
m.submodules.rx_datapath = rx_datapath = ReceivePostprocessing()
m.d.comb += [
self.rx_idle .eq(serdes.rx_idle),
serdes.rx_enable .eq(self.enable),
serdes.rx_align .eq(self.rx_align),
rx_datapath.align .eq(self.rx_align),
rx_datapath.sink .stream_eq(serdes.source),
self.source .stream_eq(rx_datapath.source)
]
# Pass through a synchronized version of our SerDes' rx-gpio.
m.submodules += FFSynchronizer(serdes.rx_gpio, self.rx_gpio, o_domain="fast")
#
# LFPS Detection
#
m.submodules.lfps_detector = lfps_detector = LFPSSquareWaveDetector(self._fast_clock_frequency)
m.d.comb += [
lfps_detector.rx_gpio .eq(self.rx_gpio),
self.lfps_signaling_detected .eq(lfps_detector.present)
]
# debug signals
m.d.comb += [
self.raw_rx_data.eq(serdes.source.data),
self.raw_rx_ctrl.eq(serdes.source.ctrl),
]
return m
def elaborate(self, platform: Platform) -> Module:
m = Module()
lane = self.lane
m.submodules += lane # Add the PCIe Lane as a submodule
# TODO: Change this for 5 Gbit/s
platform.add_clock_constraint(
self.rx_clk, (500e6 if self.speed_5GTps else 250e6) / self.gearing
) # For NextPNR, set the maximum clock frequency such that errors are given
platform.add_clock_constraint(
self.tx_clk, (500e6 if self.speed_5GTps else 250e6) / self.gearing)
# RX and TX clock input signals, these go to the SERDES.
rx_clk_i = Signal()
tx_clk_i = Signal()
# RX and TX clock output signals, these come from the SERDES.
rx_clk_o = Signal()
tx_clk_o = Signal()
# Connect RX and TX SERDES clock inputs to the RX clock output
m.d.comb += rx_clk_i.eq(rx_clk_o)
m.d.comb += tx_clk_i.eq(tx_clk_o)
# Clocks exposed by this module are the clock for rx symbol input and tx symbol output, usually they should be frequency-locked but have variable phase offset.
m.d.comb += self.rx_clk.eq(rx_clk_o)
m.d.comb += self.tx_clk.eq(tx_clk_o)
if self.fabric_clk:
m.d.comb += self.ref_clk.eq(ClockSignal())
else:
# The clock input, on the Versa board this comes from the ispCLOCK IC
m.submodules.extref0 = Instance(
"EXTREFB",
o_REFCLKO=self.
ref_clk, # The reference clock is output to ref_clk, it is not really accessible as a signal, since it only exists within the SERDES
p_REFCK_PWDNB="0b1",
p_REFCK_RTERM="0b1", # 100 Ohm
p_REFCK_DCBIAS_EN="0b0",
)
m.submodules.extref0.attrs["LOC"] = "EXTREF0" # Locate it
if self.gearing == 1: # Different gearing compatibility!
# If it is 1:1, only the first symbol has data
m.d.comb += [
lane.rx_symbol.eq(self.rx_bus[0:9]),
lane.rx_valid.eq(
self.rx_bus[0:9] != K(14, 7)
), # SERDES outputs K14.7 when there are coding errors
self.tx_bus.eq(
Cat(lane.tx_symbol[0:9], lane.tx_set_disp[0],
lane.tx_disp[0], lane.tx_e_idle[0]))
]
else:
# For 1:2, the output symbols get composed from both symbols, structure of rx_data and tx_data is shown on page 8/9 of TN1261
m.d.comb += [
lane.rx_symbol.eq(Cat(self.rx_bus[0:9], self.rx_bus[12:21])),
lane.rx_valid.eq(
Cat(self.rx_bus[0:9] != K(14, 7),
self.rx_bus[12:21] != K(14, 7))),
self.tx_bus.eq(
Cat(lane.tx_symbol[0:9], lane.tx_set_disp[0],
lane.tx_disp[0], lane.tx_e_idle[0],
lane.tx_symbol[9:18], lane.tx_set_disp[1],
lane.tx_disp[1], lane.tx_e_idle[1])),
]
#
vals_ch_write = [ # LSB first
#[0x16, "----0---"],
#[0x18, "------10"],
#[0x04, "----1--1"], # CTC bypass and lsm_disable
#[0x05, "----00--"], # Disanble skip match
#[0x1A, "---1----"], # pden_sel
#[0x1B, "------00"], # los_en
#[0x18, "----1---"],
]
vals_du_write = [ # LSB first
#[0x0B, "11-----0"],
#[0x18, "----1---"],
]
vals_ch_read = [
#[0x37, self.debug.dco_status], # Read address 16 to test1
[0x0B, self.debug.dco_status], # Read address 16 to test1
]
vals_du_read = []
# SCI interface. (from LUNA https://github.com/greatscottgadgets/luna/blob/main/luna/gateware/interface/serdes_phy/backends/ecp5.py)
m.submodules.sci = sci = ECP5SerDesConfigInterface()
m.submodules.sci_ctrl = sci_ctrl = ECP5SerDesConfigController(
sci, vals_ch_write, vals_du_write, vals_ch_read, vals_du_read)
# RX signals and their domain-crossed parts
rx_los = Signal() # Loss of Signal
rx_los_s = Signal()
rx_lol = Signal() # RX Loss of Lock
rx_lol_s = Signal()
rx_lsm = Signal() # Sync state machine status
rx_lsm_s = Signal()
rx_inv = Signal() # Invert RX
rx_det = Signal() # RX detected
# TX signals
tx_lol = Signal() # TX PLL Loss of Lock
tx_lol_s = Signal()
# Reset Signals
serdes_tx_reset = Signal()
serdes_rx_reset = Signal()
pcs_reset = Signal()
cnt = Signal(8)
with m.FSM(domain="rx"): # Inspirations taken from LUNA
with m.State("init"):
m.d.comb += [
serdes_tx_reset.eq(1),
serdes_rx_reset.eq(1),
pcs_reset.eq(1),
lane.reset_done.eq(0),
]
m.d.rx += cnt.eq(0)
with m.If(~self.lane.reset):
m.next = "start-tx"
with m.State("start-tx"):
m.d.comb += [
serdes_tx_reset.eq(0),
serdes_rx_reset.eq(1),
pcs_reset.eq(1),
lane.reset_done.eq(0),
]
m.d.rx += cnt.eq(cnt + 1)
with m.If(~tx_lol_s | (cnt > 200)):
m.d.rx += cnt.eq(0)
m.next = "start-rx"
with m.State("start-rx"):
m.d.comb += [
serdes_tx_reset.eq(0),
serdes_rx_reset.eq(0),
pcs_reset.eq(1),
lane.reset_done.eq(0),
]
m.d.rx += cnt.eq(cnt + 1)
with m.If(~rx_lol_s | (cnt > 200)):
m.next = "start-pcs-done"
with m.State("start-pcs-done"):
m.d.comb += [
serdes_tx_reset.eq(0),
serdes_rx_reset.eq(0),
pcs_reset.eq(0),
lane.reset_done.eq(1),
]
with m.If(self.lane.reset):
m.next = "init"
# Clock domain crossing for status signals and tx data
m.submodules += [
FFSynchronizer(rx_los, rx_los_s, o_domain="rx"),
FFSynchronizer(rx_lol, rx_lol_s, o_domain="rx"),
FFSynchronizer(rx_lsm, rx_lsm_s, o_domain="rx"),
FFSynchronizer(tx_lol, tx_lol_s, o_domain="rx"),
# FFSynchronizer(self.tx_bus, tx_bus_s, o_domain="rx"),
]
# Connect the signals to the lanes signals
m.d.comb += [
rx_inv.eq(lane.rx_invert),
rx_det.eq(lane.rx_align),
lane.rx_present.eq(~rx_los_s),
lane.rx_locked.eq(~rx_lol_s),
lane.rx_aligned.eq(rx_lsm_s),
lane.tx_locked.eq(~tx_lol_s)
]
pcie_det_en = Signal() # Enable lane detection
pcie_ct = Signal() # Scan enable flag
pcie_done = Signal() # Scan finished flag
pcie_done_s = Signal()
pcie_con = Signal() # PCIe lane connected
pcie_con_s = Signal()
det_timer = Signal(range(16)) # Detection Timer
# Clock domain crossing for PCIe detection signals
m.submodules += [
FFSynchronizer(pcie_done, pcie_done_s, o_domain="tx"),
FFSynchronizer(pcie_con, pcie_con_s, o_domain="tx")
]
with m.FSM(domain="tx", reset="START"):
with m.State("START"):
# Before starting a Receiver Detection test, the transmitter must be put into
# electrical idle by setting the tx_idle_ch#_c input high. The Receiver Detection
# test can begin 120 ns after tx_elec_idle is set high by driving the appropriate
# pci_det_en_ch#_c high.
m.d.tx += det_timer.eq(15)
#m.d.tx += lane.det_valid.eq(0)
with m.If(lane.det_enable):
m.next = "SET-DETECT-H"
with m.State("SET-DETECT-H"):
# 1. The user drives pcie_det_en high, putting the corresponding TX driver into
# receiver detect mode. [...] The TX driver takes some time to enter this state
# so the pcie_det_en must be driven high for at least 120ns before pcie_ct
# is asserted.
with m.If(det_timer == 0):
m.d.tx += pcie_det_en.eq(1)
m.d.tx += det_timer.eq(15)
m.next = "SET-STROBE-H"
with m.Else():
m.d.tx += det_timer.eq(det_timer - 1)
with m.State("SET-STROBE-H"):
# 2. The user drives pcie_ct high for four byte clocks.
with m.If(det_timer == 0):
m.d.tx += pcie_ct.eq(1)
m.d.tx += det_timer.eq(3)
m.next = "SET-STROBE-L"
with m.Else():
m.d.tx += det_timer.eq(det_timer - 1)
with m.State("SET-STROBE-L"):
# 3. SERDES drives the corresponding pcie_done low.
# (this happens asynchronously, so we're going to observe a few samples of pcie_done
# as high)
with m.If(det_timer == 0):
m.d.tx += pcie_ct.eq(0)
m.next = "WAIT-DONE-L"
with m.Else():
m.d.tx += det_timer.eq(det_timer - 1)
with m.State("WAIT-DONE-L"):
with m.If(~pcie_done_s):
m.next = "WAIT-DONE-H"
with m.State("WAIT-DONE-H"):
with m.If(pcie_done_s):
#m.d.tx += lane.det_status.eq(pcie_con_s) TODO: Figure this out
#m.d.tx += lane.det_status.eq(pcie_con_s)
m.d.tx += lane.det_status.eq(1)
m.next = "DONE"
with m.State("DONE"):
m.d.tx += lane.det_valid.eq(1)
with m.If(~lane.det_enable):
m.next = "START"
with m.Else():
m.next = "DONE"
gearing_str = "0b0" if self.gearing == 1 else "0b1" # Automatically select value based on gearing
dcu_config = {
"p_D_MACROPDB": "0b1",
"p_D_IB_PWDNB":
"0b1", # undocumented, seems to be "input buffer power down"
"p_D_TXPLL_PWDNB": "0b1",
"i_D_FFC_MACROPDB": 1,
# DCU — reset
"i_D_FFC_MACRO_RST": 0,
"i_D_FFC_DUAL_RST": 0,
"i_D_FFC_TRST": serdes_tx_reset,
# DCU — clocking
"i_D_REFCLKI": self.ref_clk,
"o_D_FFS_PLOL": tx_lol,
"p_D_REFCK_MODE": "0b100", # 25x ref_clk
"p_D_TX_MAX_RATE":
"5.0" if self.speed_5GTps else "2.5", # How many Gbps
"p_D_TX_VCO_CK_DIV": "0b000", # DIV/1
"p_D_BITCLK_LOCAL_EN":
"0b1", # undocumented (PCIe sample code used)
"p_D_SYNC_LOCAL_EN": "0b1",
"p_D_BITCLK_FROM_ND_EN": "0b0",
# DCU — unknown
"p_D_CMUSETBIASI":
"0b00", # begin undocumented (PCIe sample code used)
"p_D_CMUSETI4CPP": "0d3", # 0d4 in Yumewatari
"p_D_CMUSETI4CPZ": "0b101", # 0d3 in Yumewatari
"p_D_CMUSETI4VCO": "0b00",
"p_D_CMUSETICP4P": "0b01",
"p_D_CMUSETICP4Z": "0b101",
"p_D_CMUSETINITVCT": "0b00",
"p_D_CMUSETISCL4VCO": "0b000",
"p_D_CMUSETP1GM": "0b000",
"p_D_CMUSETP2AGM": "0b000",
#"p_D_CMUSETZGM" :"0b100",
#"p_D_SETIRPOLY_AUX" :"0b10",
"p_D_CMUSETZGM": "0b000",
"p_D_SETIRPOLY_AUX": "0b00",
"p_D_SETICONST_AUX": "0b01",
#"p_D_SETIRPOLY_CH" :"0b10",
#"p_D_SETICONST_CH" :"0b10",
"p_D_SETIRPOLY_CH": "0b00",
"p_D_SETICONST_CH": "0b00",
"p_D_SETPLLRC": "0d1",
"p_D_RG_EN": "0b1",
"p_D_RG_SET": "0b00", # end undocumented
# DCU — FIFOs
"p_D_LOW_MARK": "0d4",
"p_D_HIGH_MARK": "0d12",
}
ch_config = {
# CH0 — protocol
"p_CHx_PROTOCOL": "PCIe",
"p_CHx_PCIE_MODE": "0b1",
# RX CH — power management
"p_CHx_RPWDNB": "0b1",
"i_CHx_FFC_RXPWDNB": 1,
# RX CH — reset
"i_CHx_FFC_RRST": serdes_rx_reset,
"i_CHx_FFC_LANE_RX_RST": pcs_reset,
# RX CH — input
"i_CHx_FFC_SB_INV_RX": rx_inv,
#"p_CHx_REQ_EN" :"0b1",
#"p_CHx_RX_RATE_SEL" :"0d8", # LUNA uses 0d09 here
#"p_CHx_REQ_LVL_SET" :"0b00", # LUNA uses 0b01 (9 dB) here
"p_CHx_RTERM_RX":
"0d22", # 50 Ohm (wizard value used, does not match datasheet)
"p_CHx_RXIN_CM": "0b11", # CMFB (wizard value used)
"p_CHx_RXTERM_CM": "0b10", # RX Input terminate to GND
# RX CH — clocking
"i_CHx_RX_REFCLK": self.ref_clk,
"o_CHx_FF_RX_PCLK": rx_clk_o,
"i_CHx_FF_RXI_CLK": rx_clk_i,
"p_CHx_RX_GEAR_MODE": gearing_str, # 1:2 gearbox
"p_CHx_FF_RX_H_CLK_EN": gearing_str, # enable DIV/2 output clock
"p_CHx_FF_RX_F_CLK_DIS": gearing_str, # disable DIV/1 output clock
"i_CHx_FFC_RATE_MODE_RX":
self.divide_clk, # Divide by 2 when set to 1
"p_CHx_AUTO_FACQ_EN": "0b1", # undocumented (wizard value used)
"p_CHx_AUTO_CALIB_EN": "0b1", # undocumented (wizard value used)
#"p_CHx_BAND_THRESHOLD" :"0b00",
"p_CHx_CDR_MAX_RATE":
"5.0" if self.speed_5GTps else "2.5", # How many Gbps
"p_CHx_RX_DCO_CK_DIV": "0b000", # DIV/1
"p_CHx_PDEN_SEL": "0b1", # phase detector disabled on ~LOS
"p_CHx_SEL_SD_RX_CLK": "0b1", # FIFO driven by recovered clock
#"p_CHx_SEL_SD_RX_CLK" :"0b0", # FIFO driven by FF_EBRD_CLK
"p_CHx_CTC_BYPASS": "******", # bypass CTC FIFO
# FIFO bridge clocking
"i_CHx_FF_EBRD_CLK": tx_clk_i,
"p_CHx_TXDEPRE": "DISABLED",
"p_CHx_TXDEPOST": "DISABLED",
"p_CHx_DCOATDCFG":
"0b00", # begin undocumented (PCIe sample code used)
"p_CHx_DCOATDDLY": "0b00",
"p_CHx_DCOBYPSATD": "0b1",
#"p_CHx_DCOCALDIV :"0b010",
#"p_CHx_DCOCTLGI :"0b011",
#"p_CHx_DCODISBDAVOID :"0b1",
#"p_CHx_DCOFLTDAC :"0b00",
#"p_CHx_DCOFTNRG :"0b010",
#"p_CHx_DCOIOSTUNE :"0b010",
"p_CHx_DCOCALDIV": "0b001",
"p_CHx_DCOCTLGI": "0b010",
"p_CHx_DCODISBDAVOID": "0b0",
"p_CHx_DCOFLTDAC": "0b01",
"p_CHx_DCOFTNRG": "0b111",
"p_CHx_DCOIOSTUNE": "0b000",
"p_CHx_DCOITUNE": "0b00",
#"p_CHx_DCOITUNE4LSB :"0b010",
"p_CHx_DCOITUNE4LSB": "0b111",
"p_CHx_DCOIUPDNX2": "0b1",
"p_CHx_DCONUOFLSB": "0b101",
#"p_CHx_DCOSCALEI :"0b01",
#"p_CHx_DCOSTARTVAL :"0b010",
#"p_CHx_DCOSTEP :"0b11", # end undocumented
"p_CHx_DCOSCALEI": "0b00",
"p_CHx_DCOSTARTVAL": "0b000",
"p_CHx_DCOSTEP": "0b00", # end undocumented
#"p_CHx_DCOATDCFG" : "0b00", # begin undocumented (sample code used) (from LUNA https://github.com/greatscottgadgets/luna/blob/0a0393528519d68bfd8dcb2b771502134f331923/luna/gateware/interface/serdes_phy/backends/ecp5.py)
#"p_CHx_DCOATDDLY" : "0b00",
#"p_CHx_DCOBYPSATD" : "0b1",
#"p_CHx_DCOCALDIV" : "0b000",
#"p_CHx_DCOCTLGI" : "0b011",
#"p_CHx_DCODISBDAVOID" : "0b0",
#"p_CHx_DCOFLTDAC" : "0b00",
#"p_CHx_DCOFTNRG" : "0b001",
#"p_CHx_DCOIOSTUNE" : "0b010",
#"p_CHx_DCOITUNE" : "0b00",
#"p_CHx_DCOITUNE4LSB" : "0b010",
#"p_CHx_DCOIUPDNX2" : "0b1",
#"p_CHx_DCONUOFLSB" : "0b100",
#"p_CHx_DCOSCALEI" : "0b01",
#"p_CHx_DCOSTARTVAL" : "0b010",
#"p_CHx_DCOSTEP" : "0b11", # end undocumented
# RX CH — link state machine
"i_CHx_FFC_SIGNAL_DETECT":
rx_det, # WARNING: If 0, then no symbol lock happens
"o_CHx_FFS_LS_SYNC_STATUS": rx_lsm,
"p_CHx_ENABLE_CG_ALIGN": "0b1",
"p_CHx_UDF_COMMA_MASK": "0x3ff", # compare all 10 bits
"p_CHx_UDF_COMMA_A":
"0x283", # K28.5 inverted, encoded in reversed order
"p_CHx_UDF_COMMA_B": "0x17C", # K28.5, encoded in reversed order
"p_CHx_MIN_IPG_CNT": "0b11", # minimum interpacket gap of 4
"p_CHx_MATCH_4_ENABLE": "0b1", # 4 character skip matching
"p_CHx_CC_MATCH_1": "0x1BC", # K28.5 Comma
"p_CHx_CC_MATCH_2": "0x11C", # K28.0 Skip
"p_CHx_CC_MATCH_3": "0x11C", # K28.0 Skip
"p_CHx_CC_MATCH_4": "0x11C", # K28.0 Skip
# RX CH — loss of signal
"o_CHx_FFS_RLOS": rx_los,
"p_CHx_RLOS_SEL":
"0b1", # Maybe try if values from LUNA are better here?
"p_CHx_RX_LOS_EN": "0b1",
"p_CHx_RX_LOS_LVL": "0b100", # Lattice "TBD" (wizard value used)
"p_CHx_RX_LOS_CEQ": "0b11", # Lattice "TBD" (wizard value used)
"p_CHx_RX_LOS_HYST_EN": "0b0",
# RX CH — loss of lock
"o_CHx_FFS_RLOL": rx_lol,
# RX CH — data
**{
"o_CHx_FF_RX_D_%d" % n: self.rx_bus[n]
for n in range(self.rx_bus.width)
}, # Connect outputs to RX data signals
"p_CHx_DEC_BYPASS": "******", # Bypass 8b10b?
# TX CH — power management
#"p_CHx_TPWDNB" :"0b1",
"p_CHx_TPWDNB": "0b0",
"i_CHx_FFC_TXPWDNB": 1,
# TX CH — reset
"i_CHx_FFC_LANE_TX_RST": pcs_reset,
# TX CH — output
"p_CHx_TXAMPLITUDE": "0d1000", # 1000 mV
"p_CHx_RTERM_TX": "0d19", # 50 Ohm
"p_CHx_TDRV_SLICE0_CUR": "0b011", # 400 uA
"p_CHx_TDRV_SLICE0_SEL": "0b01", # main data
"p_CHx_TDRV_SLICE1_CUR": "0b000", # 100 uA
"p_CHx_TDRV_SLICE1_SEL": "0b00", # power down
"p_CHx_TDRV_SLICE2_CUR": "0b11", # 3200 uA
"p_CHx_TDRV_SLICE2_SEL": "0b01", # main data
"p_CHx_TDRV_SLICE3_CUR": "0b11", # 3200 uA
"p_CHx_TDRV_SLICE3_SEL": "0b01", # main data
"p_CHx_TDRV_SLICE4_CUR": "0b11", # 3200 uA
"p_CHx_TDRV_SLICE4_SEL": "0b01", # main data
"p_CHx_TDRV_SLICE5_CUR": "0b00", # 800 uA
"p_CHx_TDRV_SLICE5_SEL": "0b00", # power down
# TX CH — clocking
"o_CHx_FF_TX_PCLK": tx_clk_o, # Output from SERDES
"i_CHx_FF_TXI_CLK": tx_clk_i, # Input to SERDES
"p_CHx_TX_GEAR_MODE": gearing_str, # 1:2 gearbox
"p_CHx_FF_TX_H_CLK_EN": gearing_str, # disable DIV/1 output clock
"p_CHx_FF_TX_F_CLK_DIS": gearing_str, # enable DIV/2 output clock
"i_CHx_FFC_RATE_MODE_TX":
self.divide_clk, # Divide by 2 when set to 1
# TX CH — data
**{
"o_CHx_FF_TX_D_%d" % n: self.tx_bus[n]
for n in range(self.tx_bus.width)
}, # Connect TX SERDES inputs to the signals
"p_CHx_ENC_BYPASS": "******", # Bypass 8b10b
# CHx DET
"i_CHx_FFC_PCIE_DET_EN": pcie_det_en,
"i_CHx_FFC_PCIE_CT": pcie_ct,
"o_CHx_FFS_PCIE_DONE": pcie_done,
"o_CHx_FFS_PCIE_CON": pcie_con,
# Bit Slip
"i_CHx_FFC_CDR_EN_BITSLIP": self.slip,
#"i_CHx_FFC_FB_LOOPBACK" : 3,
# SCI interface, see pages 52-55 in TN1261
**{"i_D_SCIWDATA%d" % n: sci.sci_wdata[n]
for n in range(8)}, # Data in
**{"i_D_SCIADDR%d" % n: sci.sci_addr[n]
for n in range(6)}, # Address
**{"o_D_SCIRDATA%d" % n: sci.sci_rdata[n]
for n in range(8)}, # Data out
"i_D_SCIENAUX": sci.dual_sel, # Select dual registers
"i_D_SCISELAUX": sci.dual_sel,
"i_CH1_SCIEN": sci.chan_sel, # Select channel registers
"i_CH1_SCISEL": sci.chan_sel,
"i_D_SCIRD": sci.sci_rd, # Read
"i_D_SCIWSTN": sci.sci_wrn, # Write
}
modified_ch_config = {}
for key in ch_config:
modified_ch_config[key.replace(
"CHx", "CH0" if self.CH == 0 else "CH1")] = ch_config[key]
m.submodules.dcu0 = Instance("DCUA", **dcu_config,
**modified_ch_config)
m.submodules.dcu0.attrs["LOC"] = "DCU0" if self.DCU == 0 else "DCU1"
# Needed for Lattice Diamond, Trellis does not need this.
m.submodules.dcu0.attrs["CHAN"] = "CH0" if self.CH == 0 else "CH1"
m.submodules.dcu0.attrs["BEL"] = "X42/Y71/DCU"
return m
def elaborate(self, platform):
m = Module()
# Memory ports.
write_port = self.mem.write_port()
read_port = self.mem.read_port(domain="sync")
m.submodules += [write_port, read_port]
# If necessary, create synchronized versions of the relevant signals.
if self.samples_pretrigger >= 2:
delayed_inputs = Signal.like(self.inputs)
m.submodules += FFSynchronizer(self.inputs,
delayed_inputs,
stages=self.samples_pretrigger)
elif self.samples_pretrigger == 1:
delayed_inputs = Signal.like(self.inputs)
m.d.sync += delayed_inputs.eq(self.inputs)
else:
delayed_inputs = self.inputs
# Counter that keeps track of our write position.
write_position = Signal(range(0, self.sample_depth))
# Set up our write port to capture the input signals,
# and our read port to provide the output.
m.d.comb += [
write_port.data.eq(delayed_inputs),
write_port.addr.eq(write_position),
self.captured_sample.eq(read_port.data),
read_port.addr.eq(self.captured_sample_number)
]
# Don't sample unless our FSM asserts our sample signal explicitly.
m.d.sync += write_port.en.eq(0)
with m.FSM(name="ila_state") as fsm:
m.d.comb += self.sampling.eq(~fsm.ongoing("IDLE"))
# IDLE: wait for the trigger strobe
with m.State('IDLE'):
with m.If(self.trigger):
m.next = 'SAMPLE'
# Grab a sample as our trigger is asserted.
m.d.sync += [
write_port.en.eq(1),
write_position.eq(0),
self.complete.eq(0),
]
# SAMPLE: do our sampling
with m.State('SAMPLE'):
# Sample until we run out of samples.
m.d.sync += [
write_port.en.eq(1),
write_position.eq(write_position + 1),
]
# If this is the last sample, we're done. Finish up.
with m.If(write_position + 1 == self.sample_depth):
m.next = "IDLE"
m.d.sync += [self.complete.eq(1), write_port.en.eq(0)]
# Convert our sync domain to the domain requested by the user, if necessary.
if self.domain != "sync":
m = DomainRenamer(self.domain)(m)
return m
def elaborate(self, platform):
m = Module()
pins = self._pins
in_fifo = self._in_fifo
out_fifo = self._out_fifo
jtag_oe = Signal(len(pins))
jtag_o = Signal(len(pins))
jtag_i = Signal(len(pins))
m.d.comb += [
Cat(pin.oe for pin in pins).eq(jtag_oe),
Cat(pin.o for pin in pins).eq(jtag_o),
]
m.submodules += FFSynchronizer(Cat(pin.i for pin in pins), jtag_i)
timer = Signal(range(self._period_cyc))
cmd = Signal(8)
data = Signal(16)
with m.FSM():
with m.State("RECV-COMMAND"):
with m.If(out_fifo.readable):
m.d.comb += out_fifo.re.eq(1)
m.d.sync += cmd.eq(out_fifo.dout)
with m.If(out_fifo.dout == CMD_W):
m.d.sync += timer.eq(self._period_cyc - 1)
m.next = "WAIT"
with m.Elif(out_fifo.dout == CMD_I):
m.next = "SAMPLE"
with m.Else():
m.next = "RECV-DATA-1"
with m.State("RECV-DATA-1"):
with m.If(out_fifo.readable):
m.d.comb += out_fifo.re.eq(1)
m.d.sync += data[0:8].eq(out_fifo.dout)
m.next = "RECV-DATA-2"
with m.State("RECV-DATA-2"):
with m.If(out_fifo.readable):
m.d.comb += out_fifo.re.eq(1)
m.d.sync += data[8:16].eq(out_fifo.dout)
m.next = "DRIVE"
with m.State("DRIVE"):
with m.If(cmd == CMD_OE):
m.d.sync += jtag_oe.eq(data)
with m.Elif(cmd == CMD_O):
m.d.sync += jtag_o.eq( data)
with m.Elif(cmd == CMD_L):
m.d.sync += jtag_o.eq(~data & jtag_o)
with m.Elif(cmd == CMD_H):
m.d.sync += jtag_o.eq( data | jtag_o)
m.next = "RECV-COMMAND"
with m.State("WAIT"):
with m.If(timer == 0):
m.next = "RECV-COMMAND"
with m.Else():
m.d.sync += timer.eq(timer - 1)
with m.State("SAMPLE"):
m.d.sync += data.eq(jtag_i)
m.next = "SEND-DATA-1"
with m.State("SEND-DATA-1"):
with m.If(in_fifo.writable):
m.d.comb += in_fifo.we.eq(1)
m.d.comb += in_fifo.din.eq(data[0:8])
m.next = "SEND-DATA-2"
with m.State("SEND-DATA-2"):
with m.If(in_fifo.writable):
m.d.comb += in_fifo.we.eq(1)
m.d.comb += in_fifo.din.eq(data[8:16])
m.next = "RECV-COMMAND"
return m
def create_synchronizer(signal, output):
return FFSynchronizer(signal, output, o_domain=o_domain, stages=stages)