From eed2a4fee9870f69f4e05ab60ac52e5662070206 Mon Sep 17 00:00:00 2001 From: morgan Date: Wed, 5 Jun 2024 13:01:17 +0800 Subject: [PATCH] cxp downconn gw: add gtx up to 12.5Gbps testing: add txusrclk mmcm & loopback mode testing: add debug output testing: send comma in the middle of long packet to maintain lock downconn: don't put IDLE into fifo downconn: add GTX and QPLL support downconn: add DRP for GTX and QPLL to support all CXP linerates GTX: add gtx with mmcm for TXUSRCLK freq requirement GTX: add loopback mode parameter for testing GTX: add gtx with 40bits internal width GTX: use built-in comma aligner GTX: add comma checker to ensure comma is aligner on highest linerate GTX: set QPLL as CLK source for GTX --- src/gateware/cxp_downconn.py | 836 +++++++++++++++++++++++++++++++++++ 1 file changed, 836 insertions(+) create mode 100644 src/gateware/cxp_downconn.py diff --git a/src/gateware/cxp_downconn.py b/src/gateware/cxp_downconn.py new file mode 100644 index 0000000..40e9938 --- /dev/null +++ b/src/gateware/cxp_downconn.py @@ -0,0 +1,836 @@ +from migen import * +from migen.genlib.cdc import MultiReg +from migen.genlib.resetsync import AsyncResetSynchronizer + +from misoc.cores.code_8b10b import Encoder, Decoder +from misoc.interconnect.csr import * +from misoc.interconnect import stream + +from artiq.gateware.drtio.transceiver.gtx_7series_init import * +from cxp_pipeline import downconn_layout + +from functools import reduce +from operator import add + +class CXP_DownConn_PHY(Module, AutoCSR): + def __init__(self, refclk, pads, sys_clk_freq, debug_sma, pmod_pads): + nconn = len(pads) + self.rx_start_init = CSRStorage() + self.rx_restart = CSR() + + self.tx_start_init = CSRStorage() + self.tx_restart = CSR() + self.txenable = CSRStorage() + + self.rx_ready = CSRStatus(nconn) + + self.qpll_reset = CSR() + self.qpll_locked = CSRStatus() + + self.gtxs = [] + # # # + + self.submodules.qpll = qpll = QPLL(refclk, sys_clk_freq) + self.sync += [ + qpll.reset.eq(self.qpll_reset.re), + self.qpll_locked.status.eq(qpll.lock), + ] + + + + for i in range(nconn): + if i != 0: + break + gtx = GTX(self.qpll, pads[i], sys_clk_freq, tx_mode="single", rx_mode="single") + self.gtxs.append(gtx) + setattr(self.submodules, "gtx"+str(i), gtx) + + # TODO: add extension gtx connections + # TODO: add connection interface + + # TODO: Connect slave cxp_gtx_rx clock tgt + # checkout channel interfaces & drtio_gtx + # GTPTXPhaseAlignement for inspiration + + # Connect all GTX connections' DRP + + self.gtx_daddr = CSRStorage(9) + self.gtx_dread = CSR() + self.gtx_din_stb = CSR() + self.gtx_din = CSRStorage(16) + + self.gtx_dout = CSRStatus(16) + self.gtx_dready = CSR() + + for gtx in self.gtxs: + self.sync += [ + + gtx.txenable.eq(self.txenable.storage[0]), + gtx.tx_restart.eq(self.tx_restart.re), + gtx.rx_restart.eq(self.rx_restart.re), + gtx.tx_init.clk_path_ready.eq(self.tx_start_init.storage), + gtx.rx_init.clk_path_ready.eq(self.rx_start_init.storage), + ] + + self.comb += gtx.dclk.eq(ClockSignal("sys")) + self.sync += [ + gtx.den.eq(0), + gtx.dwen.eq(0), + If(self.gtx_dread.re, + gtx.den.eq(1), + gtx.daddr.eq(self.gtx_daddr.storage), + ).Elif(self.gtx_din_stb.re, + gtx.den.eq(1), + gtx.dwen.eq(1), + gtx.daddr.eq(self.gtx_daddr.storage), + gtx.din.eq(self.gtx_din.storage), + ), + ] + + # TODO: deal with 4 GTX instance of outpus + for n, gtx in enumerate(self.gtxs): + self.sync += [ + self.rx_ready.status[n].eq(gtx.rx_ready), + If(gtx.dready, + self.gtx_dready.w.eq(1), + self.gtx_dout.status.eq(gtx.dout), + ), + If(self.gtx_dready.re, + self.gtx_dready.w.eq(0), + ), + ] + + self.sources = [] + + for n, gtx in enumerate(self.gtxs): + # DEBUG: remove cdc fifo + # gtx rx -> fifo out -> cdc out + + fifo_out = stream.AsyncFIFO(downconn_layout, 512) + self.submodules += ClockDomainsRenamer({"write": "cxp_gtx_rx", "read": "sys"})(fifo_out) + self.sources.append(fifo_out) + + for i in range(4): + self.sync.cxp_gtx_rx += [ + fifo_out.sink.stb.eq(0), + # don't store idle word in fifo + If((gtx.rx_ready & fifo_out.sink.ack & ~((gtx.decoders[0].d == 0xBC) & (gtx.decoders[0].k == 1))), + fifo_out.sink.stb.eq(1), + fifo_out.sink.data[i*8:(i*8)+8].eq(gtx.decoders[i].d), + fifo_out.sink.k[i].eq(gtx.decoders[i].k), + ), + ] + + + + # DEBUG: tx of gtx is not used in CXP + # DEBUG: txusrclk PLL DRG + + self.txpll_reset = CSRStorage() + self.pll_daddr = CSRStorage(7) + self.pll_dclk = CSRStorage() + self.pll_den = CSRStorage() + self.pll_din = CSRStorage(16) + self.pll_dwen = CSRStorage() + + self.txpll_locked = CSRStatus() + self.pll_dout = CSRStatus(16) + self.pll_dready = CSRStatus() + + self.txinit_phaligndone = CSRStatus() + self.rxinit_phaligndone = CSRStatus() + + self.tx_stb = CSRStorage() + self.sinks = [] + + for n, gtx in enumerate(self.gtxs): + self.comb += [ + gtx.txpll_reset.eq(self.txpll_reset.storage), + gtx.pll_daddr.eq(self.pll_daddr.storage), + gtx.pll_dclk.eq(self.pll_dclk.storage), + gtx.pll_den.eq(self.pll_den.storage), + gtx.pll_din.eq(self.pll_din.storage), + gtx.pll_dwen.eq(self.pll_dwen.storage), + + self.txinit_phaligndone.status.eq(gtx.tx_init.Xxphaligndone), + self.rxinit_phaligndone.status.eq(gtx.rx_init.Xxphaligndone), self.txpll_locked.status.eq(gtx.txpll_locked), + self.pll_dout.status.eq(gtx.pll_dout), + self.pll_dready.status.eq(gtx.pll_dready), + ] + + # DEBUG:loopback + self.loopback_mode = CSRStorage(3) + self.comb += gtx.loopback_mode.eq(self.loopback_mode.storage) + + # DEBUG: datain + # fw -> fifo (sys) -> cdc fifo -> gtx tx + + fifo_in = stream.AsyncFIFO(downconn_layout, 512) + self.submodules += ClockDomainsRenamer({"write": "sys", "read": "cxp_gtx_tx"})(fifo_in) + self.sinks.append(fifo_in) + + # TODO: why there this send an extra 0xFB word + txstb = Signal() + self.specials += MultiReg(self.tx_stb.storage, txstb, odomain="cxp_gtx_tx") + + word_count = Signal(max=100) + + # JANK: fix the every 98th word got eaten + # cnt 97 98 99 0 + # out fifo[97] IDLE IDLE fifo[99] + # ack 1 0 0 1 + self.sync.cxp_gtx_tx += [ + fifo_in.source.ack.eq(0), + + If(word_count == 99, + word_count.eq(word_count.reset), + ).Else( + If(fifo_in.source.stb & txstb, + If(word_count != 98, fifo_in.source.ack.eq(1)), + word_count.eq(word_count + 1), + ) + ) + ] + + # NOTE: prevent the first word send twice due to stream stb delay + self.comb += [ + If((fifo_in.source.stb & fifo_in.source.ack & (word_count != 99)), + gtx.encoder.d[0].eq(fifo_in.source.data[:8]), + gtx.encoder.d[1].eq(fifo_in.source.data[8:16]), + gtx.encoder.d[2].eq(fifo_in.source.data[16:24]), + gtx.encoder.d[3].eq(fifo_in.source.data[24:]), + gtx.encoder.k[0].eq(fifo_in.source.k[0]), + gtx.encoder.k[1].eq(fifo_in.source.k[1]), + gtx.encoder.k[2].eq(fifo_in.source.k[2]), + gtx.encoder.k[3].eq(fifo_in.source.k[3]), + ).Else( + # NOTE: IDLE WORD + gtx.encoder.d[0].eq(0xBC), + gtx.encoder.k[0].eq(1), + gtx.encoder.d[1].eq(0x3C), + gtx.encoder.k[1].eq(1), + gtx.encoder.d[2].eq(0x3C), + gtx.encoder.k[2].eq(1), + gtx.encoder.d[3].eq(0xB5), + gtx.encoder.k[3].eq(0), + ) + ] + + # DEBUG: IO SMA & PMOD + if n == 0: + self.specials += [ + # Instance("OBUF", i_I=gtx.cd_cxp_gtx_rx.clk, o_O=debug_sma.p_tx), + # Instance("OBUF", i_I=gtx.cd_cxp_gtx_tx.clk, o_O=debug_sma.n_rx), + + # # pmod 0-7 pin + # Instance("OBUF", i_I=txstb, o_O=pmod_pads[0]), + # Instance("OBUF", i_I=fifo_in.source.stb, o_O=pmod_pads[1]), + # Instance("OBUF", i_I=fifo_in.source.ack, o_O=pmod_pads[2]), + # Instance("OBUF", i_I=gtx.comma_checker.aligner_en, o_O=pmod_pads[3]), + # Instance("OBUF", i_I=gtx.comma_checker.check_reset, o_O=pmod_pads[4]), + # Instance("OBUF", i_I=gtx.comma_checker.has_comma, o_O=pmod_pads[5]), + # Instance("OBUF", i_I=gtx.comma_checker.has_error, o_O=pmod_pads[6]), + # Instance("OBUF", i_I=gtx.comma_checker.ready_sys, o_O=pmod_pads[7]), + + # Instance("OBUF", i_I=gtx.dclk, o_O=pmod_pads[0]), + # Instance("OBUF", i_I=gtx.den, o_O=pmod_pads[1]), + # Instance("OBUF", i_I=gtx.dwen, o_O=pmod_pads[2]), + # Instance("OBUF", i_I=gtx.dready, o_O=pmod_pads[3]), + ] + +class QPLL(Module, AutoCSR): + def __init__(self, refclk, sys_clk_freq): + self.clk = Signal() + self.refclk = Signal() + self.lock = Signal() + self.reset = Signal() + + self.daddr = CSRStorage(8) + self.dread = CSR() + self.din_stb = CSR() + self.din = CSRStorage(16) + + self.dout = CSRStatus(16) + self.dready = CSR() + + # # # + + # VCO @ 10GHz, linerate = 1.25Gbps + # feedback divider = 80 + qpll_fbdiv = 0b0100100000 + qpll_fbdiv_ratio = 1 + refclk_div = 1 + self.Xxout_div = 8 + + # DEBUG: txuserclk + fbdiv_real = 80 + self.tx_usrclk_freq = (sys_clk_freq*fbdiv_real/self.Xxout_div)/40 + + dready = Signal() + self.specials += [ + Instance("GTXE2_COMMON", + i_QPLLREFCLKSEL=0b001, + i_GTREFCLK0=refclk, + + i_QPLLPD=0, + i_QPLLRESET=self.reset, + i_QPLLLOCKEN=1, + o_QPLLLOCK=self.lock, + o_QPLLOUTCLK=self.clk, + o_QPLLOUTREFCLK=self.refclk, + + # See UG476 (v1.12.1) Table 2-16 + p_QPLL_FBDIV=qpll_fbdiv, + p_QPLL_FBDIV_RATIO=qpll_fbdiv_ratio, + p_QPLL_REFCLK_DIV=refclk_div, + + # From 7 Series FPGAs Transceivers Wizard + p_BIAS_CFG=0x0000040000001000, + p_COMMON_CFG=0x00000000, + p_QPLL_CFG=0x0680181, + p_QPLL_CLKOUT_CFG=0b0000, + p_QPLL_COARSE_FREQ_OVRD=0b010000, + p_QPLL_COARSE_FREQ_OVRD_EN=0b0, + p_QPLL_CP=0b0000011111, + p_QPLL_CP_MONITOR_EN=0b0, + p_QPLL_DMONITOR_SEL=0b0, + p_QPLL_FBDIV_MONITOR_EN= 0b0, + p_QPLL_INIT_CFG=0x000006, + p_QPLL_LOCK_CFG=0x21E8, + p_QPLL_LPF=0b1111, + + # Reserved, values cannot be modified + i_BGBYPASSB=0b1, + i_BGMONITORENB=0b1, + i_BGPDB=0b1, + i_BGRCALOVRD=0b11111, + i_RCALENB=0b1, + i_QPLLRSVD1=0b0, + i_QPLLRSVD2=0b11111, + + # Dynamic Reconfiguration Ports + i_DRPADDR=self.daddr.storage, + i_DRPCLK=ClockSignal("sys"), + i_DRPEN=(self.dread.re | self.din_stb.re), + i_DRPWE=self.din_stb.re, + i_DRPDI=self.din.storage, + o_DRPDO=self.dout.status, + o_DRPRDY=dready, + ) + ] + + self.sync += [ + If(dready, + self.dready.w.eq(1), + ), + If(self.dready.re, + self.dready.w.eq(0), + ), + ] + +# Warning: Xilinx transceivers are LSB first, and comma needs to be flipped +# compared to the usual 8b10b binary representation. +class Comma_Checker(Module): + def __init__(self, comma, reset_period=10_000_000): + self.data = Signal(20) + self.comma_aligned = Signal() + self.comma_realigned = Signal() + self.comma_det = Signal() + + self.aligner_en = Signal() + self.ready_sys = Signal() + self.restart_sys = Signal() + + # # # + + + # periodically reset rx until rx is connected and receiving valid data + # as after connecting RXP/RXN, the whole RX need to be reset + + reset_counter = Signal(reset=reset_period-1, max=reset_period) + self.sync += [ + self.restart_sys.eq(0), + If(~self.ready_sys, + If(reset_counter == 0, + reset_counter.eq(reset_counter.reset), + self.restart_sys.eq(1), + ).Else( + reset_counter.eq(reset_counter - 1), + ) + ) + ] + + + # Data and comma checker + # From UG476 (v1.12.1) p.228 + # The built-in RXBYTEISALIGNED can be falsely asserted at linerate higher than 5Gbps + # The validity of data and comma needed to be checked externally + + comma_n = ~comma & 0b1111111111 + + # DEBUG: remove after use + self.has_comma = Signal() + self.has_error = Signal() + + comma_seen = Signal() + error_seen = Signal() + one_counts = Signal(max=11) + + # From CXP-001-2021 section 9.2.5.1 + # For high speed connection an IDLE word shall be transmitted at least once every 100 words + counter_period = 200 + + counter = Signal(reset=counter_period-1, max=counter_period) + check_reset = Signal() + check = Signal() + + self.sync.cxp_gtx_rx += [ + If(check_reset, + counter.eq(counter.reset), + check.eq(0), + ).Elif(counter == 0, + check.eq(1), + ).Else( + counter.eq(counter - 1), + ), + + If(check_reset, + comma_seen.eq(0), + ).Elif((self.data[:10] == comma) | (self.data[:10] == comma_n), + comma_seen.eq(1) + ), + + one_counts.eq(reduce(add, [self.data[i] for i in range(10)])), + If(check_reset, + error_seen.eq(0), + ).Elif((one_counts != 4) & (one_counts != 5) & (one_counts != 6), + error_seen.eq(1), + ), + + # DEBUG: + self.has_comma.eq(0), + If((self.data[:10] == comma) | (self.data[:10] == comma_n), + self.has_comma.eq(1), + ), + + self.has_error.eq(0), + If((one_counts != 4) & (one_counts != 5) & (one_counts != 6), + self.has_error.eq(1), + ), + ] + + # DEBUG: expose signal + self.check_reset = Signal() + self.comb +=[ + self.check_reset.eq(check_reset), + ] + + self.submodules.rxfsm = rxfsm = ClockDomainsRenamer("cxp_gtx_rx")(FSM(reset_state="WAIT_COMMA")) + + rxfsm.act("WAIT_COMMA", + If(self.comma_det, + NextState("ALIGNING"), + ) + ) + + rxfsm.act("ALIGNING", + If(self.comma_aligned & (~self.comma_realigned), + NextState("WAIT_ALIGNED_DATA"), + ).Else( + self.aligner_en.eq(1), + ) + ) + + # wait for the aligned data to arrive at the FPGA RX interface + # as there is a delay before the data is avaiable after RXBYTEISALIGNED is asserted + self.submodules.timer = timer = ClockDomainsRenamer("cxp_gtx_rx")(WaitTimer(10_000)) + + rxfsm.act("WAIT_ALIGNED_DATA", + timer.wait.eq(1), + If(timer.done, + check_reset.eq(1), + NextState("CHECKING"), + ) + ) + + rxfsm.act("CHECKING", + If(check, + check_reset.eq(1), + If(comma_seen & (~error_seen), + NextState("READY"), + ).Else( + NextState("WAIT_COMMA") + ) + ) + ) + + ready = Signal() + self.specials += MultiReg(ready, self.ready_sys) + rxfsm.act("READY", + ready.eq(1), + If(check, + check_reset.eq(1), + If(~(comma_seen & (~error_seen)), + NextState("WAIT_COMMA"), + ) + ) + ) + + +class GTX(Module): + # Settings: + # * GTX reference clock @ 125MHz + # * GTX data width = 20 + # * GTX PLL frequency @ 3.125GHz + # * GTX line rate (TX & RX) @ 3.125Gb/s + # * GTX TX/RX USRCLK @ PLL/datawidth = 156MHz + def __init__(self, qpll, pads, sys_clk_freq, tx_mode="single", rx_mode="single"): + assert tx_mode in ["single", "master", "slave"] + assert rx_mode in ["single", "master", "slave"] + + # linerate = USRCLK * datawidth + pll_fbout_mult = 8 + txusr_pll_div = pll_fbout_mult*sys_clk_freq/qpll.tx_usrclk_freq + + self.tx_restart = Signal() + self.rx_restart = Signal() + self.loopback_mode = Signal(3) + + self.txenable = Signal() + self.rx_ready = Signal() + + # Dynamic Reconfiguration Ports + self.daddr = Signal(9) + self.dclk = Signal() + self.den = Signal() + self.dwen = Signal() + self.din = Signal(16) + self.dout = Signal(16) + self.dready = Signal() + + self.submodules.encoder = ClockDomainsRenamer("cxp_gtx_tx")(Encoder(4, True)) + self.submodules.decoders = [ClockDomainsRenamer("cxp_gtx_rx")( + (Decoder(True))) for _ in range(4)] + + + # transceiver direct clock outputs + # useful to specify clock constraints in a way palatable to Vivado + self.txoutclk = Signal() + self.rxoutclk = Signal() + + # # # + + # TX generates cxp_tx clock, init must be in system domain + # FIXME: 500e6 is used to fix Xx reset by holding gtxXxreset for a couple cycle more + self.submodules.tx_init = tx_init = GTXInit(500e6, False, mode=tx_mode) + self.submodules.rx_init = rx_init = GTXInit(sys_clk_freq, True, mode=rx_mode) + + # RX receives restart commands from txusrclk domain + # self.submodules.rx_init = rx_init = ClockDomainsRenamer("cxp_gtx_tx")(GTXInit(500e6, True, mode=rx_mode)) + + self.comb += [ + tx_init.cplllock.eq(qpll.lock), + rx_init.cplllock.eq(qpll.lock) + ] + + txdata = Signal(40) + rxdata = Signal(40) + + comma_aligned = Signal() + comma_realigned = Signal() + comma_det = Signal() + comma_aligner_en = Signal() + # Note: the following parameters were set after consulting AR45360 + self.specials += \ + Instance("GTXE2_CHANNEL", + # PMA Attributes + p_PMA_RSV=0x001E7080, + p_PMA_RSV2=0x2050, # PMA_RSV2[5] = 0: Eye scan feature disabled + p_PMA_RSV3=0, + p_PMA_RSV4=1, # PMA_RSV[4],RX_CM_TRIM[2:0] = 0b1010: Common mode 800mV + p_RX_BIAS_CFG=0b000000000100, + p_RX_OS_CFG=0b0000010000000, + p_RX_CLK25_DIV=5, + p_TX_CLK25_DIV=5, + + # Power-Down Attributes + p_PD_TRANS_TIME_FROM_P2=0x3c, + p_PD_TRANS_TIME_NONE_P2=0x3c, + p_PD_TRANS_TIME_TO_P2=0x64, + i_CPLLPD=1, + + # QPLL + i_QPLLCLK=qpll.clk, + i_QPLLREFCLK=qpll.refclk, + p_RXOUT_DIV=qpll.Xxout_div, + p_TXOUT_DIV=qpll.Xxout_div, + i_RXSYSCLKSEL=0b11, # use QPLL & QPLL's REFCLK + i_TXSYSCLKSEL=0b11, # use QPLL & CPLL's REFCLK + + # TX clock + p_TXBUF_EN="FALSE", + p_TX_XCLK_SEL="TXUSR", + o_TXOUTCLK=self.txoutclk, + # i_TXSYSCLKSEL=0b00, + i_TXOUTCLKSEL=0b11, + + # TX Startup/Reset + i_TXPHDLYRESET=0, + i_TXDLYBYPASS=0, + i_TXPHALIGNEN=1 if tx_mode != "single" else 0, + i_GTTXRESET=tx_init.gtXxreset, + o_TXRESETDONE=tx_init.Xxresetdone, + i_TXDLYSRESET=tx_init.Xxdlysreset, + o_TXDLYSRESETDONE=tx_init.Xxdlysresetdone, + i_TXPHINIT=tx_init.txphinit if tx_mode != "single" else 0, + o_TXPHINITDONE=tx_init.txphinitdone if tx_mode != "single" else Signal(), + i_TXPHALIGN=tx_init.Xxphalign if tx_mode != "single" else 0, + i_TXDLYEN=tx_init.Xxdlyen if tx_mode != "single" else 0, + o_TXPHALIGNDONE=tx_init.Xxphaligndone, + i_TXUSERRDY=tx_init.Xxuserrdy, + p_TXPMARESET_TIME=1, + p_TXPCSRESET_TIME=1, + i_TXINHIBIT=~self.txenable, + + # TX data + p_TX_DATA_WIDTH=40, + p_TX_INT_DATAWIDTH=1, # 1 if a line rate is greater than 6.6 Gbps + i_TXCHARDISPMODE=Cat(txdata[9], txdata[19], txdata[29], txdata[39]), + i_TXCHARDISPVAL=Cat(txdata[8], txdata[18], txdata[28], txdata[38]), + i_TXDATA=Cat(txdata[:8], txdata[10:18], txdata[20:28], txdata[30:38]), + i_TXUSRCLK=ClockSignal("cxp_gtx_tx"), + i_TXUSRCLK2=ClockSignal("cxp_gtx_tx"), + + # TX electrical + i_TXBUFDIFFCTRL=0b100, + i_TXDIFFCTRL=0b1000, + + # RX Startup/Reset + i_RXPHDLYRESET=0, + i_RXDLYBYPASS=0, + i_RXPHALIGNEN=1 if rx_mode != "single" else 0, + i_GTRXRESET=rx_init.gtXxreset, + o_RXRESETDONE=rx_init.Xxresetdone, + i_RXDLYSRESET=rx_init.Xxdlysreset, + o_RXDLYSRESETDONE=rx_init.Xxdlysresetdone, + i_RXPHALIGN=rx_init.Xxphalign if rx_mode != "single" else 0, + i_RXDLYEN=rx_init.Xxdlyen if rx_mode != "single" else 0, + o_RXPHALIGNDONE=rx_init.Xxphaligndone, + i_RXUSERRDY=rx_init.Xxuserrdy, + p_RXPMARESET_TIME=1, + p_RXPCSRESET_TIME=1, + + # RX AFE + p_RX_DFE_XYD_CFG=0, + p_RX_CM_SEL=0b11, # RX_CM_SEL = 0b11: Common mode is programmable + p_RX_CM_TRIM=0b010, # PMA_RSV[4],RX_CM_TRIM[2:0] = 0b1010: Common mode 800mV + i_RXDFEXYDEN=1, + i_RXDFEXYDHOLD=0, + i_RXDFEXYDOVRDEN=0, + i_RXLPMEN=1, # RXLPMEN = 1: LPM mode is enable for non scramble 8b10b data + p_RXLPM_HF_CFG=0b00000011110000, + p_RXLPM_LF_CFG=0b00000011110000, + + p_RX_DFE_GAIN_CFG=0x0207EA, + p_RX_DFE_VP_CFG=0b00011111100000011, + p_RX_DFE_UT_CFG=0b10001000000000000, + p_RX_DFE_KL_CFG=0b0000011111110, + p_RX_DFE_KL_CFG2=0x3788140A, + p_RX_DFE_H2_CFG=0b000110000000, + p_RX_DFE_H3_CFG=0b000110000000, + p_RX_DFE_H4_CFG=0b00011100000, + p_RX_DFE_H5_CFG=0b00011100000, + p_RX_DFE_LPM_CFG=0x0904, # RX_DFE_LPM_CFG = 0x0904: linerate <= 6.6Gb/s + # = 0x0104: linerate > 6.6Gb/s + + # RX clock + i_RXDDIEN=1, + # i_RXSYSCLKSEL=0b00, + i_RXOUTCLKSEL=0b010, + o_RXOUTCLK=self.rxoutclk, + i_RXUSRCLK=ClockSignal("cxp_gtx_rx"), + i_RXUSRCLK2=ClockSignal("cxp_gtx_rx"), + + # RX Clock Correction Attributes + p_CLK_CORRECT_USE="FALSE", + p_CLK_COR_SEQ_1_1=0b0100000000, + p_CLK_COR_SEQ_2_1=0b0100000000, + p_CLK_COR_SEQ_1_ENABLE=0b1111, + p_CLK_COR_SEQ_2_ENABLE=0b1111, + + # RX data + p_RX_DATA_WIDTH=40, + p_RX_INT_DATAWIDTH=1, # 1 if a line rate is greater than 6.6 Gbps + o_RXDISPERR=Cat(rxdata[9], rxdata[19], rxdata[29], rxdata[39]), + o_RXCHARISK=Cat(rxdata[8], rxdata[18], rxdata[28], rxdata[38]), + o_RXDATA=Cat(rxdata[:8], rxdata[10:18], rxdata[20:28], rxdata[30:38]), + + # RX Byte and Word Alignment Attributes + p_ALIGN_COMMA_DOUBLE="FALSE", + p_ALIGN_COMMA_ENABLE=0b1111111111, + p_ALIGN_COMMA_WORD=4, # align comma to rxdata[:10] only + p_ALIGN_MCOMMA_DET="TRUE", + p_ALIGN_MCOMMA_VALUE=0b1010000011, + p_ALIGN_PCOMMA_DET="TRUE", + p_ALIGN_PCOMMA_VALUE=0b0101111100, + p_SHOW_REALIGN_COMMA="FALSE", + p_RXSLIDE_AUTO_WAIT=7, + p_RXSLIDE_MODE="OFF", + p_RX_SIG_VALID_DLY=10, + i_RXPCOMMAALIGNEN=comma_aligner_en, + i_RXMCOMMAALIGNEN=comma_aligner_en, + i_RXCOMMADETEN=1, + i_RXSLIDE=0, + o_RXBYTEISALIGNED=comma_aligned, + o_RXBYTEREALIGN=comma_realigned, + o_RXCOMMADET=comma_det, + + # RX 8B/10B Decoder Attributes + p_RX_DISPERR_SEQ_MATCH="FALSE", + p_DEC_MCOMMA_DETECT="TRUE", + p_DEC_PCOMMA_DETECT="TRUE", + p_DEC_VALID_COMMA_ONLY="FALSE", + + # RX Buffer Attributes + p_RXBUF_ADDR_MODE="FAST", + p_RXBUF_EIDLE_HI_CNT=0b1000, + p_RXBUF_EIDLE_LO_CNT=0b0000, + p_RXBUF_EN="FALSE", + p_RX_BUFFER_CFG=0b000000, + p_RXBUF_RESET_ON_CB_CHANGE="TRUE", + p_RXBUF_RESET_ON_COMMAALIGN="FALSE", + p_RXBUF_RESET_ON_EIDLE="FALSE", # RXBUF_RESET_ON_EIDLE = FALSE: OOB is disabled + p_RXBUF_RESET_ON_RATE_CHANGE="TRUE", + p_RXBUFRESET_TIME=0b00001, + p_RXBUF_THRESH_OVFLW=61, + p_RXBUF_THRESH_OVRD="FALSE", + p_RXBUF_THRESH_UNDFLW=4, + p_RXDLY_CFG=0x001F, + p_RXDLY_LCFG=0x030, + p_RXDLY_TAP_CFG=0x0000, + p_RXPH_CFG=0xC00002, + p_RXPHDLY_CFG=0x084020, + p_RXPH_MONITOR_SEL=0b00000, + p_RX_XCLK_SEL="RXUSR", + p_RX_DDI_SEL=0b000000, + p_RX_DEFER_RESET_BUF_EN="TRUE", + + # CDR Attributes + p_RXCDR_CFG=0x03_0000_23FF_1008_0020, # LPM @ 0.5G-1.5625G , 8B/10B encoded data, CDR setting < +/- 200ppm + # (See UG476 (v1.12.1), p.206) + p_RXCDR_FR_RESET_ON_EIDLE=0b0, + p_RXCDR_HOLD_DURING_EIDLE=0b0, + p_RXCDR_PH_RESET_ON_EIDLE=0b0, + p_RXCDR_LOCK_CFG=0b010101, + + # Pads + i_GTXRXP=pads.rxp, + i_GTXRXN=pads.rxn, + o_GTXTXP=pads.txp, + o_GTXTXN=pads.txn, + + # Dynamic Reconfiguration Ports + p_IS_DRPCLK_INVERTED=0b0, + i_DRPADDR=self.daddr, + i_DRPCLK=self.dclk, + i_DRPEN=self.den, + i_DRPWE=self.dwen, + i_DRPDI=self.din, + o_DRPDO=self.dout, + o_DRPRDY=self.dready, + + # ! loopback for debugging + i_LOOPBACK = self.loopback_mode, + p_TX_LOOPBACK_DRIVE_HIZ = "FALSE", + p_RXPRBS_ERR_LOOPBACK = 0b0, + + # Other parameters + p_PCS_RSVD_ATTR=( + (tx_mode != "single") << 1 | # PCS_RSVD_ATTR[1] = 0: TX Single Lane Auto Mode + # = 1: TX Manual Mode + (rx_mode != "single") << 2 | # [2] = 0: RX Single Lane Auto Mode + # = 1: RX Manual Mode + 0 << 8 # [8] = 0: OOB is disabled + ), + i_RXELECIDLEMODE=0b11, # RXELECIDLEMODE = 0b11: OOB is disabled + p_RX_DFE_LPM_HOLD_DURING_EIDLE=0b0, + p_ES_EYE_SCAN_EN="TRUE", # Must be TRUE for GTX + ) + + + # TX clocking + # A PLL is used to generate the correct frequency for TXUSRCLK (UG476 Equation 3-1) + self.clock_domains.cd_cxp_gtx_tx = ClockDomain() + txpll_fb_clk = Signal() + txoutclk_buf = Signal() + txpll_clkout = Signal() + + self.txpll_reset = Signal() + self.pll_daddr = Signal(7) + self.pll_dclk = Signal() + self.pll_den = Signal() + self.pll_din = Signal(16) + self.pll_dwen = Signal() + + self.txpll_locked = Signal() + self.pll_dout = Signal(16) + self.pll_dready = Signal() + self.specials += [ + Instance("PLLE2_ADV", + p_BANDWIDTH="HIGH", + o_LOCKED=self.txpll_locked, + i_RST=self.txpll_reset, + + p_CLKIN1_PERIOD=1e9/sys_clk_freq, # ns + i_CLKIN1=txoutclk_buf, + + # VCO @ 1.25GHz + p_CLKFBOUT_MULT=pll_fbout_mult, p_DIVCLK_DIVIDE=1, + i_CLKFBIN=txpll_fb_clk, o_CLKFBOUT=txpll_fb_clk, + + # frequency = linerate/40 + p_CLKOUT0_DIVIDE=txusr_pll_div, p_CLKOUT0_PHASE=0.0, o_CLKOUT0=txpll_clkout, + + # Dynamic Reconfiguration Ports + i_DADDR = self.pll_daddr, + i_DCLK = self.pll_dclk, + i_DEN = self.pll_den, + i_DI = self.pll_din, + i_DWE = self.pll_dwen, + o_DO = self.pll_dout, + o_DRDY = self.pll_dready, + ), + Instance("BUFG", i_I=self.txoutclk, o_O=txoutclk_buf), + Instance("BUFG", i_I=txpll_clkout, o_O=self.cd_cxp_gtx_tx.clk), + AsyncResetSynchronizer(self.cd_cxp_gtx_tx, ~self.txpll_locked & ~tx_init.done) + ] + + # RX clocking + # the CDR matches the required frequency for RXUSRCLK, no need for PLL + self.clock_domains.cd_cxp_gtx_rx = ClockDomain() + self.specials += [ + Instance("BUFG", i_I=self.rxoutclk, o_O=self.cd_cxp_gtx_rx.clk), + AsyncResetSynchronizer(self.cd_cxp_gtx_rx, ~rx_init.done) + ] + + self.comb += [ + txdata.eq(Cat(self.encoder.output[0], self.encoder.output[1], self.encoder.output[2], self.encoder.output[3])), + self.decoders[0].input.eq(rxdata[:10]), + self.decoders[1].input.eq(rxdata[10:20]), + self.decoders[2].input.eq(rxdata[20:30]), + self.decoders[3].input.eq(rxdata[30:]), + ] + + + self.submodules.comma_checker = comma_checker = Comma_Checker(0b0101111100) + self.comb += [ + comma_checker.data.eq(rxdata), + comma_checker.comma_aligned.eq(comma_aligned), + comma_checker.comma_realigned.eq(comma_realigned), + comma_checker.comma_det.eq(comma_det), + comma_aligner_en.eq(comma_checker.aligner_en), + self.rx_ready.eq(comma_checker.ready_sys), + + rx_init.restart.eq(self.rx_restart | comma_checker.restart_sys), + tx_init.restart.eq(self.tx_restart), + ]