cxp downconn fw: doc clean up

cxp downconn: add QPLL with its reset CSR
cxp downconn: set QPLL as CLK source for GTX
cxp downconn: remove PLL_reset signal from tx_init to prevent race
condition
cxp downconn: rename cxp_gtx to GTX

src/gateware/cxp.py

@@ -2,131 +2,20 @@ from migen import *
 from misoc.interconnect.csr import *
 from misoc.cores.liteeth_mini.mac.crc import LiteEthMACCRCEngine
 
-from artiq.gateware.drtio.core import ChannelInterface
-
 from cxp_downconn import CXP_DownConn
 from cxp_upconn import CXP_UpConn
 
 class CXP(Module, AutoCSR):
     def __init__(self, refclk, pads, sys_clk_freq, debug_sma, pmod_pads):
-        nchannels = len(pads)
-        self.rx_start_init = CSRStorage()
-        self.rx_restart = CSRStatus()
-        self.rx_bypass_clk_alignment = CSRStorage()
-
-        self.tx_start_init = CSRStorage()
-        self.tx_restart = CSRStorage()
-
-        self.loopback_mode = CSRStorage(3)
-
-        self.txinit_phaligndone = CSRStatus()
-        self.rxinit_phaligndone = CSRStatus()
-        self.rx_ready = CSRStatus()
-
-        self.data_0 = CSRStorage(8)
-        self.data_1 = CSRStorage(8)
-        self.control_bit_0 = CSRStorage()
-        self.control_bit_1 = CSRStorage()
-        self.encoded_0 = CSRStatus(10)
-        self.encoded_1 = CSRStatus(10)
-
-        self.rxdata_0 = CSRStatus(10)
-        self.rxdata_1 = CSRStatus(10)
-        self.decoded_data_0 = CSRStatus(8)
-        self.decoded_data_1 = CSRStatus(8)
-        self.decoded_k_0 = CSRStatus()
-        self.decoded_k_1 = CSRStatus()
-
-        # # #
-
         self.submodules.crc = CXP_CRC(8)
-
         # FIFOs with transmission priority
         # 0: Trigger packet
         # 1: IO acknowledgment for trigger packet
         # 2: All other packets
-
         self.submodules.upconn = CXP_UpConn(debug_sma, sys_clk_freq, pmod_pads)
-        
 
-        # single & master tx_mode can lock with rx in loopback 
-        self.submodules.gtx = gtx = CXP_DownConn(refclk, pads, sys_clk_freq, tx_mode="single", rx_mode="single")
-        
-        # DEBUG:loopback 
-        self.sync += gtx.loopback_mode.eq(self.loopback_mode.storage) 
+        self.submodules.downconn = CXP_DownConn(refclk, pads, sys_clk_freq, debug_sma, pmod_pads)
 
-        # # ! debug sma
-        # self.specials += [
-        #     Instance("OBUF", i_I=gtx.rxoutclk, o_O=debug_sma.p_tx),
-        #     Instance("OBUF", i_I=gtx.cd_cxp_gtx_tx.clk, o_O=debug_sma.n_rx)
-        # ]
-
-        self.comb += [
-            self.txinit_phaligndone.status.eq(self.gtx.tx_init.Xxphaligndone),
-            self.rxinit_phaligndone.status.eq(self.gtx.rx_init.Xxphaligndone),
-            self.rx_ready.status.eq(self.gtx.rx_ready),
-        ]
-
-        self.sync.cxp_gtx_tx += [
-            self.gtx.encoder.d[0].eq(self.data_0.storage),
-            self.gtx.encoder.k[0].eq(self.control_bit_0.storage),
-            self.encoded_0.status.eq(self.gtx.encoder.output[0]),
-
-            self.gtx.encoder.d[1].eq(self.data_1.storage),
-            self.gtx.encoder.k[1].eq(self.control_bit_1.storage),
-            self.encoded_1.status.eq(self.gtx.encoder.output[1]),
-        ]
-        self.sync.cxp_gtx_rx += [
-            self.rxdata_0.status.eq(self.gtx.decoders[0].input),
-            self.decoded_data_0.status.eq(self.gtx.decoders[0].d),
-            self.decoded_k_0.status.eq(self.gtx.decoders[0].k),
-
-            self.rxdata_1.status.eq(self.gtx.decoders[1].input),
-            self.decoded_data_1.status.eq(self.gtx.decoders[1].d),
-            self.decoded_k_1.status.eq(self.gtx.decoders[1].k),
-        ]
-
-        # TODO: rip encoder & rx clockalignment out of CXP_GTX 
-
-        # TODO: use expose encoder & decoder from CXP
-        # encoder.k = 1 if sending control bit, different calculation
-        # encoder.d = data 8 bit 
-
-
-
-        channel_interface = ChannelInterface(gtx.encoder, gtx.decoders)
-        self.comb += channel_interface.rx_ready.eq(gtx.rx_ready)
-        channel_interfaces = []
-        channel_interfaces.append(channel_interface)
-
-        # TransceiverInterface, just adding cxp_rx_<num>
-        self.stable_clkin = CSRStorage()
-        self.txenable = CSRStorage(len(channel_interfaces))
-        for i in range(len(channel_interfaces)):
-            name = "cxp_gtx_rx" + str(i)
-            setattr(self.clock_domains, "cd_"+name, ClockDomain(name=name))
-        self.channels = channel_interfaces
-
-
-        # TODO: add tx_phase_alignment for multi CXP
-        # The TX phase alignment will fail with a wrong TXUSRCLK frequency
-
-
-        self.comb += [
-            gtx.rx_init.clk_path_ready.eq(self.rx_start_init.storage),
-
-            gtx.tx_init.clk_path_ready.eq(self.tx_start_init.storage),
-            gtx.txenable.eq(self.txenable.storage[0]),
-            gtx.tx_restart.eq(self.tx_restart.storage),
-        ]
-
-        # TODO: Connect multilane cxp_tx 
-
-        # TODO: Connect slave i's `cxp_gtx_rx` clock to `cxp_gtx_rxi` clock
-        self.comb += [
-            getattr(self, "cd_cxp_gtx_rx" + str(0)).clk.eq(self.gtx.cd_cxp_gtx_rx.clk),
-            getattr(self, "cd_cxp_gtx_rx" + str(0)).rst.eq(self.gtx.cd_cxp_gtx_rx.rst)
-        ]
 
 class CXP_CRC(Module, AutoCSR):
     width = 32

src/gateware/cxp_downconn.py

@@ -1,6 +1,5 @@
 from migen import *
 from migen.genlib.resetsync import AsyncResetSynchronizer
-from migen.genlib.cdc import MultiReg, PulseSynchronizer
 
 from misoc.cores.code_8b10b import Encoder, Decoder
 from misoc.interconnect.csr import *
@@ -8,140 +7,352 @@ from misoc.interconnect.csr import *
 from artiq.gateware.drtio.transceiver.gtx_7series_init import *
 
 from operator import add
-from math import ceil
 from functools import reduce
 
-# Changes the phase of the transceiver RX clock to align the comma to
-# the LSBs of RXDATA, fixing the latency.
-#
-# This is implemented by repeatedly resetting the transceiver until it
-# gives out the correct phase. Each reset gives a random phase.
-#
-# If Xilinx had designed the GTX transceiver correctly, RXSLIDE_MODE=PMA
-# would achieve this faster and in a cleaner way. But:
-#  * the phase jumps are of 2 UI at every second RXSLIDE pulse, instead
-#    of 1 UI at every pulse. It is unclear what the latency becomes.
-#  * RXSLIDE_MODE=PMA cannot be used with the RX buffer bypassed.
-# Those design flaws make RXSLIDE_MODE=PMA yet another broken and useless
-# transceiver "feature".
-#
+class CXP_DownConn(Module, AutoCSR):
+    def __init__(self, refclk, pads, sys_clk_freq, debug_sma, pmod_pads):
+        self.rx_start_init = CSRStorage()
+        self.rx_restart = CSR()
+
+        self.tx_start_init = CSRStorage()
+        self.tx_restart = CSR()
+        self.txenable = CSRStorage()
+
+
+        self.txinit_phaligndone = CSRStatus()
+        self.rxinit_phaligndone = CSRStatus()
+        self.rx_ready = CSRStatus()
+
+        self.tx_div = CSRStorage(3)
+        self.rx_div = CSRStorage(3)
+
+        self.qpll_reset = CSR()
+        self.qpll_locked = CSRStatus()
+
+        # # #
+
+        self.submodules.qpll = qpll = QPLL(refclk, sys_clk_freq)
+
+        # single & master tx_mode can lock with rx in loopback 
+        self.submodules.gtx = gtx = GTX(self.qpll, pads, sys_clk_freq, tx_mode="single", rx_mode="single")
+
+        # TEST: txusrclk alignment
+        # 1) use GTREFCLK with TXSYSCLKSEL = 0b10 -> still inconsistant
+        # 2) tie qpllPDEN with ~qpll.reset, -> inconsistant
+        # 3) seems like tx_init gtxXreset fall too soon (cplllock hold hi too long) <--- this a cross clk domain issue :<
+        
+        self.sync += [
+            # PLL
+            qpll.reset.eq(self.qpll_reset.re),
+            self.qpll_locked.status.eq(qpll.lock),
+
+            # GTX
+            self.txinit_phaligndone.status.eq(gtx.tx_init.Xxphaligndone),
+            self.rxinit_phaligndone.status.eq(gtx.rx_init.Xxphaligndone),
+            self.rx_ready.status.eq(gtx.rx_ready),
+
+            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),
+            # gtx.rx_alignment_en.eq(self.rx_data_alignment.storage),
+
+            # GTX DRP
+            gtx.tx_rate.eq(self.tx_div.storage),
+            gtx.rx_rate.eq(self.rx_div.storage),
+        ]
+
+        # 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.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.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: IO SMA & PMOD
+        self.specials += [
+            Instance("OBUF", i_I=gtx.rxoutclk, 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=gtx.clk_aligner.rxslide, o_O=pmod_pads[0]),
+            Instance("OBUF", i_I=gtx.clk_aligner.ready, o_O=pmod_pads[1]),
+            # Instance("OBUF", i_I=gtx.tx_init.Xxdlysresetdone , o_O=pmod_pads[2]),
+            # Instance("OBUF", i_I=gtx.tx_init.Xxphaligndone , o_O=pmod_pads[3]),
+            # Instance("OBUF", i_I=, o_O=pmod_pads[4]),
+            # Instance("OBUF", i_I=, o_O=pmod_pads[5]),
+            # Instance("OBUF", i_I=, o_O=pmod_pads[6]),
+            # Instance("OBUF", i_I=, o_O=pmod_pads[7]),
+        ]
+
+        # DEBUG: datain
+        counter_max = 2
+        counter = Signal(max=counter_max)
+
+        self.data_0 = CSRStorage(8)
+        self.data_1 = CSRStorage(8)
+        self.data_2 = CSRStorage(8)
+        self.data_3 = CSRStorage(8)
+        self.control_bit_0 = CSRStorage()
+        self.control_bit_1 = CSRStorage()
+        self.control_bit_2 = CSRStorage()
+        self.control_bit_3 = CSRStorage()
+        self.encoded_0 = CSRStatus(10)
+        self.encoded_1 = CSRStatus(10)
+
+        self.rxdata_0 = CSRStatus(10)
+        self.rxdata_1 = CSRStatus(10)
+        self.decoded_data_0 = CSRStatus(8)
+        self.decoded_data_1 = CSRStatus(8)
+        self.decoded_k_0 = CSRStatus()
+        self.decoded_k_1 = CSRStatus()
+
+        
+        self.sync.cxp_gtx_tx += [
+            If(counter == 0,
+                self.gtx.encoder.d[0].eq(self.data_0.storage),
+                self.gtx.encoder.k[0].eq(self.control_bit_0.storage),
+                self.gtx.encoder.d[1].eq(self.data_1.storage),
+                self.gtx.encoder.k[1].eq(self.control_bit_1.storage),
+                counter.eq(counter+1),
+            ).Elif(counter == 1,
+                self.gtx.encoder.d[0].eq(self.data_2.storage),
+                self.gtx.encoder.k[0].eq(self.control_bit_2.storage),
+                self.gtx.encoder.d[1].eq(self.data_3.storage),
+                self.gtx.encoder.k[1].eq(self.control_bit_3.storage),
+                counter.eq(0),
+            ),
+            self.encoded_0.status.eq(self.gtx.encoder.output[0]),
+            self.encoded_1.status.eq(self.gtx.encoder.output[1]),
+        ]
+
+        self.sync.cxp_gtx_rx += [
+            self.rxdata_0.status.eq(self.gtx.decoders[0].input),
+            self.decoded_data_0.status.eq(self.gtx.decoders[0].d),
+            self.decoded_k_0.status.eq(self.gtx.decoders[0].k),
+
+            self.rxdata_1.status.eq(self.gtx.decoders[1].input),
+            self.decoded_data_1.status.eq(self.gtx.decoders[1].d),
+            self.decoded_k_1.status.eq(self.gtx.decoders[1].k),
+        ]
+
+
+        # The TX phase alignment will fail with a wrong TXUSRCLK frequency
+
+
+
+        # NOTE: No need to connect cxp_gtx_tx, we don't use tx anyway (just for loopback)
+
+        # TODO: Connect slave cxp_gtx_rx clock tgt
+        # checkout channel interfaces & drtio_gtx
+        # checkout GTPTXPhaseAlignement for inspiration
+
+class QPLL(Module):
+    def __init__(self, refclk, sys_clk_freq):
+        self.clk = Signal()
+        self.refclk = Signal()
+        self.lock = Signal()
+        self.reset = Signal()
+
+        # # #
+
+        # WARNING: VCO cannot do 12.5GHz on ZC706
+        # VCO freq = sys*qpll_fbdiv
+        # PLL output = VCO/2
+        qpll_fbdiv = 0b0100100000
+        qpll_fbdiv_ratio = 1
+
+        fbdiv_real = 80
+        refclk_div = 1
+        self.Xxout_div = 8
+
+        self.tx_usrclk_freq = (sys_clk_freq*fbdiv_real/self.Xxout_div)/20
+        
+        # QPLL reset
+        pll_reset_cycles = ceil(sys_clk_freq/125e6)
+        pll_reset_timer = WaitTimer(pll_reset_cycles)
+        self.submodules += pll_reset_timer
+
+        reset = Signal()
+        startup_fsm = FSM(reset_state="IDLE")
+        self.submodules += startup_fsm
+
+        startup_fsm.act("IDLE",
+            If(self.reset, NextState("RESET_PLL"))
+        )
+        startup_fsm.act("RESET_PLL",
+            reset.eq(1),
+            pll_reset_timer.wait.eq(1),
+            If(pll_reset_timer.done, NextState("IDLE"))
+        )
+
+        self.specials += [
+            Instance("GTXE2_COMMON",
+                i_QPLLREFCLKSEL=0b001,
+                i_GTREFCLK0=refclk,         
+
+                i_QPLLPD=0,
+                i_QPLLRESET=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,
+            )
+        ]
+
 # Warning: Xilinx transceivers are LSB first, and comma needs to be flipped
 # compared to the usual 8b10b binary representation.
-class CXP_BruteforceClockAligner(Module):
-    def __init__(self, comma, sys_clk_freq, check_period):
+class Manual_Aligner(Module):
+    def __init__(self, comma, check_cycles=20000):
+
+        self.rxslide = Signal()
         self.rxdata = Signal(20)
-        self.restart = Signal()
 
         self.ready = Signal()
 
-        check_max_val = ceil(check_period*sys_clk_freq)
-        check_counter = Signal(max=check_max_val+1)
-        check = Signal()
-        reset_check_counter = Signal()
-        self.sync += [
-            check.eq(0),
-            If(reset_check_counter,
-                check_counter.eq(check_max_val)
-            ).Else(
-                If(check_counter == 0,
-                    check.eq(1),
-                    check_counter.eq(check_max_val)
-                ).Else(
-                    check_counter.eq(check_counter-1)
-                )
-            )
-        ]
+        # # #
 
-        checks_reset = PulseSynchronizer("sys", "cxp_gtx_rx")
-        self.submodules += checks_reset
+        checks_reset = Signal()
+        error_seen = Signal()
+        comma_seen = Signal()
+
+        rx1cnt = Signal(max=11)
 
         comma_n = ~comma & 0b1111111111
-        comma_seen_rxclk = Signal()
-        comma_seen = Signal()
-        comma_seen_rxclk.attr.add("no_retiming")
-        self.specials += MultiReg(comma_seen_rxclk, comma_seen)
-        self.sync.cxp_gtx_rx += \
-            If(checks_reset.o,
-                comma_seen_rxclk.eq(0)
-            ).Elif((self.rxdata[:10] == comma) | (self.rxdata[:10] == comma_n),
-                comma_seen_rxclk.eq(1)
-            )
-
-        error_seen_rxclk = Signal()
-        error_seen = Signal()
-        error_seen_rxclk.attr.add("no_retiming")
-        self.specials += MultiReg(error_seen_rxclk, error_seen)
-        rx1cnt = Signal(max=11)
         self.sync.cxp_gtx_rx += [
             rx1cnt.eq(reduce(add, [self.rxdata[i] for i in range(10)])),
-            If(checks_reset.o,
-                error_seen_rxclk.eq(0)
+            If(checks_reset,
+                error_seen.eq(0)
             ).Elif((rx1cnt != 4) & (rx1cnt != 5) & (rx1cnt != 6),
-                error_seen_rxclk.eq(1)
+                error_seen.eq(1)
+            ),
+            If(checks_reset,
+                comma_seen.eq(0)
+            ).Elif((self.rxdata[:10] == comma) | (self.rxdata[:10] == comma_n),
+                comma_seen.eq(1)
             )
         ]
 
-        fsm = FSM(reset_state="WAIT_COMMA")
+        # minimum of 32 RXUSRCLK2 cycles are required between two RXSLIDE pulses
+        slide_timer = ClockDomainsRenamer("cxp_gtx_rx")(WaitTimer(64))
+        self.submodules += slide_timer
+        counter = Signal(reset=check_cycles-1, max=check_cycles)
+
+        fsm = ClockDomainsRenamer("cxp_gtx_rx")(FSM(reset_state="IDLE"))
         self.submodules += fsm
 
-        fsm.act("WAIT_COMMA",
-            If(check,
-                # Errors are still OK at this stage, as the transceiver
-                # has just been reset and may output garbage data.
+        fsm.act("IDLE",
+            slide_timer.wait.eq(1),
+            If(slide_timer.done, 
                 If(comma_seen,
-                    NextState("WAIT_NOERROR")
+                    NextState("READY"),    
                 ).Else(
-                    self.restart.eq(1)
-                ),
-                checks_reset.i.eq(1)
+                    NextState("SLIDING")
+                )
             )
         )
-        fsm.act("WAIT_NOERROR",
-            If(check,
-                If(comma_seen & ~error_seen,
-                    NextState("READY")
-                ).Else(
-                    self.restart.eq(1),
-                    NextState("WAIT_COMMA")
-                ),
-                checks_reset.i.eq(1)
-            )
+
+        fsm.act("SLIDING",
+            self.rxslide.eq(1),
+            checks_reset.eq(1),
+            NextState("IDLE")
         )
+
         fsm.act("READY",
-            reset_check_counter.eq(1),
             self.ready.eq(1),
-            If(error_seen,
-                checks_reset.i.eq(1),
-                self.restart.eq(1),
-                NextState("WAIT_COMMA")
+            If(counter == 0,
+                NextValue(counter, check_cycles - 1),
+                If(error_seen,
+                    NextState("IDLE"),
+                )
+            ).Else(
+                NextValue(counter, counter - 1),
             )
         )
+        
 
-
-
-class CXP_DownConn(Module):
+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, refclk, pads, sys_clk_freq, tx_mode="single", rx_mode="single"):
+    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"]
 
-        cpll_div = 4
-        pll_div = int(40/cpll_div)
+        # linerate = USRCLK * datawidth
+        pll_fbout_mult = 8
+        txusr_pll_div = pll_fbout_mult*sys_clk_freq/qpll.tx_usrclk_freq # 20 is datawidth
 
-        self.rx_restart = Signal()
         self.tx_restart = Signal()
+        self.rx_restart = Signal()
         self.loopback_mode = Signal(3)
 
         self.txenable = Signal()
+        self.rx_ready = Signal()
+
+        self.tx_rate = Signal(3)
+        self.rx_rate = Signal(3)
+
         self.submodules.encoder = ClockDomainsRenamer("cxp_gtx_tx")(Encoder(2, True))
         self.submodules.decoders = [ClockDomainsRenamer("cxp_gtx_rx")(
             (Decoder(True))) for _ in range(2)]
-        self.rx_ready = Signal()
+
 
         # transceiver direct clock outputs
         # useful to specify clock constraints in a way palatable to Vivado
@@ -150,20 +361,23 @@ class CXP_DownConn(Module):
 
         # # #
 
-        cpllreset = Signal()
-        cplllock = Signal()
         # TX generates cxp_tx clock, init must be in system domain
-        self.submodules.tx_init = tx_init = GTXInit(sys_clk_freq, False, mode=tx_mode)
-        # RX receives restart commands from RTIO domain
-        self.submodules.rx_init = rx_init = GTXInit(sys_clk_freq, True, mode=rx_mode)
+        # 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(500e6, 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 += [
-            cpllreset.eq(tx_init.cpllreset),
-            tx_init.cplllock.eq(cplllock),
-            rx_init.cplllock.eq(cplllock)
+            tx_init.cplllock.eq(qpll.lock),
+            rx_init.cplllock.eq(qpll.lock)
         ]
 
         txdata = Signal(20)
         rxdata = Signal(20)
+
+        rxslide = Signal()
         # Note: the following parameters were set after consulting AR45360
         self.specials += \
             Instance("GTXE2_CHANNEL",
@@ -181,29 +395,25 @@ class CXP_DownConn(Module):
                 p_PD_TRANS_TIME_FROM_P2=0x3c,
                 p_PD_TRANS_TIME_NONE_P2=0x3c,
                 p_PD_TRANS_TIME_TO_P2=0x64,
+                i_CPLLPD=1,
 
-                # CPLL
-                p_CPLL_CFG=0xBC07DC,
-                p_CPLL_FBDIV=cpll_div,
-                p_CPLL_FBDIV_45=5,
-                p_CPLL_REFCLK_DIV=1,
-                p_RXOUT_DIV=2,
-                p_TXOUT_DIV=2,
-                p_CPLL_INIT_CFG=0x00001E,
-                p_CPLL_LOCK_CFG=0x01E8,
-                i_CPLLRESET=cpllreset,
-                i_CPLLPD=cpllreset,
-                o_CPLLLOCK=cplllock,
-                i_CPLLLOCKEN=1,
-                i_CPLLREFCLKSEL=0b001,
-                i_TSTIN=2**20-1,
-                i_GTREFCLK0=refclk,
+                # Dynamic Tx/Rx divider
+                i_TXRATE=self.tx_rate,
+                i_RXRATE=self.rx_rate,
+
+                # 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_TXSYSCLKSEL=0b00,
                 i_TXOUTCLKSEL=0b11,
 
                 # TX Startup/Reset
@@ -274,7 +484,7 @@ class CXP_DownConn(Module):
 
                 # RX clock
                 i_RXDDIEN=1,
-                i_RXSYSCLKSEL=0b00,
+                # i_RXSYSCLKSEL=0b00,
                 i_RXOUTCLKSEL=0b010,
                 o_RXOUTCLK=self.rxoutclk,
                 i_RXUSRCLK=ClockSignal("cxp_gtx_rx"),
@@ -297,16 +507,22 @@ class CXP_DownConn(Module):
                 # RX Byte and Word Alignment Attributes
                 p_ALIGN_COMMA_DOUBLE="FALSE",
                 p_ALIGN_COMMA_ENABLE=0b1111111111,
-                p_ALIGN_COMMA_WORD=1,
-                p_ALIGN_MCOMMA_DET="TRUE",
+                p_ALIGN_COMMA_WORD=2,
+                p_ALIGN_MCOMMA_DET="FALSE",
                 p_ALIGN_MCOMMA_VALUE=0b1010000011,
-                p_ALIGN_PCOMMA_DET="TRUE",
+                p_ALIGN_PCOMMA_DET="FALSE",
                 p_ALIGN_PCOMMA_VALUE=0b0101111100,
                 p_SHOW_REALIGN_COMMA="FALSE",
                 p_RXSLIDE_AUTO_WAIT=7,
                 p_RXSLIDE_MODE="PCS",
                 p_RX_SIG_VALID_DLY=10,
 
+                # Manual Word Alignment
+                i_RXPCOMMAALIGNEN=0,
+                i_RXMCOMMAALIGNEN=0,
+                i_RXCOMMADETEN=1, # enable word alignment, but breaks rxrestart if gtxXxreset hold too short
+                i_RXSLIDE=rxslide,
+
                 # RX 8B/10B Decoder Attributes
                 p_RX_DISPERR_SEQ_MATCH="FALSE",
                 p_DEC_MCOMMA_DETECT="TRUE",
@@ -374,31 +590,47 @@ class CXP_DownConn(Module):
         # 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()
-        txpll_reset = Signal()
-        txpll_locked = 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=txpll_locked,
-                i_RST=txpll_reset,
+                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=10, p_DIVCLK_DIVIDE=1,
+                p_CLKFBOUT_MULT=pll_fbout_mult, p_DIVCLK_DIVIDE=1,
                 i_CLKFBIN=txpll_fb_clk, o_CLKFBOUT=txpll_fb_clk, 
 
-                # 156.25MHz
-                p_CLKOUT0_DIVIDE=pll_div, p_CLKOUT0_PHASE=0.0, o_CLKOUT0=txpll_clkout,
+                # frequency = linerate/20
+                p_CLKOUT0_DIVIDE=txusr_pll_div, p_CLKOUT0_PHASE=0.0, o_CLKOUT0=txpll_clkout,
 
-                # TODO: DRP for line rate change
+                # 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, ~txpll_locked & ~tx_init.done)
+            AsyncResetSynchronizer(self.cd_cxp_gtx_tx, ~self.txpll_locked & ~tx_init.done)
         ]
 
         # RX clocking
@@ -415,12 +647,13 @@ class CXP_DownConn(Module):
             self.decoders[1].input.eq(rxdata[10:])
         ]
 
-        # 6e-3 is too slow for 3.25Gbps line rate
-        clock_aligner = CXP_BruteforceClockAligner(0b0101111100, sys_clk_freq, check_period=1e-2)
-        self.submodules += clock_aligner
+
+        self.submodules.clk_aligner = clk_aligner = Manual_Aligner(0b0101111100)
         self.comb += [
-            clock_aligner.rxdata.eq(rxdata),
-            rx_init.restart.eq(clock_aligner.restart),
-            self.rx_ready.eq(clock_aligner.ready),
-            tx_init.restart.eq(self.tx_restart)
+            clk_aligner.rxdata.eq(rxdata),
+            rxslide.eq(clk_aligner.rxslide),
+            self.rx_ready.eq(clk_aligner.ready),
+
+            rx_init.restart.eq(self.rx_restart),
+            tx_init.restart.eq(self.tx_restart),
         ]

src/gateware/cxp_upconn.py

@@ -162,22 +162,22 @@ class CXP_UpConn(Module, AutoCSR):
            p3.eq(self.tx_idle.word_idx == 3),
         ] 
         self.specials += [
-            # debug sma
-            Instance("OBUF", i_I=o, o_O=pads.p_tx),
-            Instance("OBUF", i_I=self.cd_cxp_upconn.clk, o_O=pads.n_rx),
+            # # debug sma
+            # Instance("OBUF", i_I=o, o_O=pads.p_tx),
+            # Instance("OBUF", i_I=self.cd_cxp_upconn.clk, o_O=pads.n_rx),
 
-            # pmod 0-7 pin
-            Instance("OBUF", i_I=o, o_O=pmod[0]),
-            Instance("OBUF", i_I=self.cd_cxp_upconn.clk, o_O=pmod[1]),
-            Instance("OBUF", i_I=~self.tx_fifos.pe.n, o_O=pmod[2]),
-            Instance("OBUF", i_I=ninth_bit, o_O=pmod[3]),
-            Instance("OBUF", i_I=word_bound, o_O=pmod[4]),
-            Instance("OBUF", i_I=idling, o_O=pmod[5]),
-            # Instance("OBUF", i_I=self.tx_fifos.source_ack[0], o_O=pmod[6]),
-            # Instance("OBUF", i_I=self.tx_fifos.source_ack[2], o_O=pmod[6]),
-            # Instance("OBUF", i_I=self.tx_fifos.source_ack[1], o_O=pmod[7]),
-            Instance("OBUF", i_I=p0, o_O=pmod[6]),
-            Instance("OBUF", i_I=p3, o_O=pmod[7]),
+            # # pmod 0-7 pin
+            # Instance("OBUF", i_I=o, o_O=pmod[0]),
+            # Instance("OBUF", i_I=self.cd_cxp_upconn.clk, o_O=pmod[1]),
+            # Instance("OBUF", i_I=~self.tx_fifos.pe.n, o_O=pmod[2]),
+            # Instance("OBUF", i_I=ninth_bit, o_O=pmod[3]),
+            # Instance("OBUF", i_I=word_bound, o_O=pmod[4]),
+            # Instance("OBUF", i_I=idling, o_O=pmod[5]),
+            # # Instance("OBUF", i_I=self.tx_fifos.source_ack[0], o_O=pmod[6]),
+            # # Instance("OBUF", i_I=self.tx_fifos.source_ack[2], o_O=pmod[6]),
+            # # Instance("OBUF", i_I=self.tx_fifos.source_ack[1], o_O=pmod[7]),
+            # Instance("OBUF", i_I=p0, o_O=pmod[6]),
+            # Instance("OBUF", i_I=p3, o_O=pmod[7]),
         ]
         self.symbol0 = CSR(9)
         self.symbol1 = CSR(9)

src/libboard_artiq/src/cxp_downconn.rs

@@ -4,67 +4,431 @@ use log::info;
 
 use crate::pl::csr;
 
-pub fn main(timer: &mut GlobalTimer) {
+pub struct CXP_DownConn_Settings {
+    pub rxdiv: u8,
+    pub qpll_fbdiv: u8,
+}
+
+#[derive(Clone, Copy, Debug)]
+#[allow(non_camel_case_types)]
+pub enum CXP_SPEED {
+    CXP_1,
+    CXP_2,
+    CXP_3,
+    CXP_5,
+    CXP_6,
+    CXP_10,
+    CXP_12,
+}
+
+fn loopback_testing(timer: &mut GlobalTimer, data: u8, control_bit: u8) {
     unsafe {
-        info!("turning on pmc loopback mode...");
-        csr::cxp::loopback_mode_write(0b010); // Near-End PMA Loopback
+        // send K28_5 for CDR to align
+        const K28_5: u8 = 0xBC;
+        const K28_1: u8 = 0x3C;
+        const D21_5: u8 = 21 | 5 << 5;
 
-        // enable cxp gtx clock domains
-        csr::cxp::tx_start_init_write(1);
-        csr::cxp::rx_start_init_write(1);
+        // NOTE: testing with IDLE word + manual comma alignment
+        // 62.5MHz   OK
+        // 125MHz    OK
+        // 156.25MHz
+        // 250MHz    OK
+        // 312.5MHz
+        // 500MHz    ?? havn't change CDR config yet
+        const LEN: usize = 4;
+        const DATA: [[u8; LEN]; 2] = [
+            // [K28_5, K28_1, K28_5, K28_1],
+            // [1, 1, 1, 1],
+            [K28_5, K28_1, K28_1, D21_5],
+            [1, 1, 1, 0],
+        ];
 
-        info!("waiting for QPLL/CPLL to lock...");
+        csr::cxp::downconn_data_0_write(DATA[0][0]);
+        csr::cxp::downconn_data_1_write(DATA[0][1]);
+        csr::cxp::downconn_data_2_write(DATA[0][2]);
+        csr::cxp::downconn_data_3_write(DATA[0][3]);
+
+        csr::cxp::downconn_control_bit_0_write(DATA[1][0]);
+        csr::cxp::downconn_control_bit_1_write(DATA[1][1]);
+        csr::cxp::downconn_control_bit_2_write(DATA[1][2]);
+        csr::cxp::downconn_control_bit_3_write(DATA[1][3]);
+
+        info!("waiting for tx&rx setup...");
         timer.delay_us(50_000);
-        info!("tx_phaligndone = {} ", csr::cxp::txinit_phaligndone_read(),);
+        info!(
+            "tx_phaligndone = {} | rx_phaligndone = {}",
+            csr::cxp::downconn_txinit_phaligndone_read(),
+            csr::cxp::downconn_rxinit_phaligndone_read(),
+        );
 
         // enable txdata tranmission thought MGTXTXP, required by PMA loopback
-        csr::cxp::txenable_write(1);
+        csr::cxp::downconn_txenable_write(1);
 
-        loopback_testing(timer, 0x00, 0);
+        info!("waiting for rx to align...");
+        timer.delay_us(50_000);
+        // while csr::cxp::downconn_rx_ready_read() != 1 {}
+        // info!("rx ready!");
+
+        // csr::cxp::data_3_write(data);
+        // csr::cxp::control_bit_3_write(control_bit);
+        // println!(
+        //     "data[0] = {:#04x} control bit = {:#b} encoded = {:#012b}",
+        //     csr::cxp::downconn_data_0_read(),
+        //     csr::cxp::downconn_control_bit_0_read(),
+        //     csr::cxp::downconn_encoded_0_read(),
+        // );
+        // println!(
+        //     "data[1] = {:#04x} control bit = {:#b} encoded = {:#012b}",
+        //     csr::cxp::downconn_data_1_read(),
+        //     csr::cxp::downconn_control_bit_1_read(),
+        //     csr::cxp::downconn_encoded_1_read(),
+        // );
+
+        // for _ in 0..20 {
+        loop {
+            // timer.delay_us(100);
+            println!(
+                "data = {:#022b} | rx ready = {}",
+                (csr::cxp::downconn_rxdata_0_read() as u32 | ((csr::cxp::downconn_rxdata_1_read() as u32) << 10)),
+                csr::cxp::downconn_rx_ready_read()
+            );
+            timer.delay_us(1_000_000);
+            // println!(
+            //     "data[0] = {:#012b} data[1] = {:#012b}",
+            //     csr::cxp::rxdata_0_read(),
+            //     csr::cxp::rxdata_1_read(),
+            // );
+            // println!(
+            //     "decoded_data[0] = {:#04x} decoded_k[0] = {:#b} decoded_data[1] = {:#04x} decoded_k[1] = {:#b}",
+            //     csr::cxp::downconn_decoded_data_0_read(),
+            //     csr::cxp::downconn_decoded_k_0_read(),
+            //     csr::cxp::downconn_decoded_data_1_read(),
+            //     csr::cxp::downconn_decoded_k_1_read(),
+            // );
+        }
+    }
+}
+
+pub fn setup(timer: &mut GlobalTimer, speed: CXP_SPEED) {
+    unsafe {
+        info!("turning on pmc loopback mode...");
+        csr::cxp::downconn_loopback_mode_write(0b010); // Near-End PMA Loopback
+
+        // QPLL setup
+        csr::cxp::downconn_qpll_reset_write(1);
+        info!("waiting for QPLL/CPLL to lock...");
+        while csr::cxp::downconn_qpll_locked_read() != 1 {}
+        info!("QPLL locked");
+
+        // tx/rx setup
+        csr::cxp::downconn_tx_start_init_write(1);
+        csr::cxp::downconn_rx_start_init_write(1);
+
+        info!("waiting for tx & rx setup...");
+        timer.delay_us(50_000);
+        info!(
+            "tx_phaligndone = {} | rx_phaligndone = {}",
+            csr::cxp::downconn_txinit_phaligndone_read(),
+            csr::cxp::downconn_rxinit_phaligndone_read(),
+        );
+        println!("==============================================================================");
     }
 
-    fn loopback_testing(timer: &mut GlobalTimer, data: u8, control_bit: u8) {
+    change_linerate(timer, speed);
+
+    loopback_testing(timer, 0x00, 0);
+}
+
+pub fn change_linerate(timer: &mut GlobalTimer, speed: CXP_SPEED) {
+    info!("Changing datarate to {:?}", speed);
+    // DEBUG: DRP pll for TXUSRCLK = freq(linerate)/20
+    let settings = txusrclk::get_txusrclk_config(speed);
+    txusrclk::setup(timer, settings);
+
+    // TODO: set QPLL_FBDIV via DRP
+    change_qpll_settings(speed);
+
+    unsafe {
+        csr::cxp::downconn_qpll_reset_write(1);
+        info!("waiting for QPLL/CPLL to lock...");
+        while csr::cxp::downconn_qpll_locked_read() != 1 {}
+        info!("QPLL locked");
+    }
+
+    unsafe {
+        csr::cxp::downconn_tx_restart_write(1);
+        csr::cxp::downconn_rx_restart_write(1);
+    }
+}
+
+fn change_qpll_settings(speed: CXP_SPEED) {
+    let divider = match speed {
+        CXP_SPEED::CXP_1 => 0b100,                      // Divided by 8
+        CXP_SPEED::CXP_2 | CXP_SPEED::CXP_3 => 0b011,   // Divided by 4
+        CXP_SPEED::CXP_5 | CXP_SPEED::CXP_6 => 0b010,   // Divided by 2
+        CXP_SPEED::CXP_10 | CXP_SPEED::CXP_12 => 0b010, // Divided by 1
+    };
+
+    unsafe {
+        csr::cxp::downconn_rx_div_write(divider);
+        csr::cxp::downconn_tx_div_write(divider);
+    }
+}
+
+pub mod txusrclk {
+    use super::*;
+
+    pub struct PLLSetting {
+        pub clkout0_reg1: u16,  //0x08
+        pub clkout0_reg2: u16,  //0x09
+        pub clkfbout_reg1: u16, //0x14
+        pub clkfbout_reg2: u16, //0x15
+        pub div_reg: u16,       //0x16
+        pub lock_reg1: u16,     //0x18
+        pub lock_reg2: u16,     //0x19
+        pub lock_reg3: u16,     //0x1A
+        pub power_reg: u16,     //0x28
+        pub filt_reg1: u16,     //0x4E
+        pub filt_reg2: u16,     //0x4F
+    }
+
+    fn one_clock_cycle() {
         unsafe {
-            // send K28_5 for CDR to align
-            const K28_5: u8 = 0xBC;
-            csr::cxp::data_0_write(K28_5);
-            csr::cxp::control_bit_0_write(1);
-            csr::cxp::data_1_write(K28_5);
-            csr::cxp::control_bit_1_write(1);
+            csr::cxp::downconn_pll_dclk_write(1);
+            csr::cxp::downconn_pll_dclk_write(0);
+        }
+    }
 
-            info!("waiting for rx to align...");
-            while csr::cxp::rx_ready_read() != 1 {}
-            info!("rx ready!");
+    fn set_addr(address: u8) {
+        unsafe {
+            csr::cxp::downconn_pll_daddr_write(address);
+        }
+    }
 
-            csr::cxp::data_1_write(data);
-            csr::cxp::control_bit_1_write(control_bit);
-            println!(
-                "data[0] = {:#04x} control bit = {:#b} encoded = {:#012b}",
-                csr::cxp::data_0_read(),
-                csr::cxp::control_bit_0_read(),
-                csr::cxp::encoded_0_read(),
-            );
-            println!(
-                "data[1] = {:#04x} control bit = {:#b} encoded = {:#012b}",
-                csr::cxp::data_1_read(),
-                csr::cxp::control_bit_1_read(),
-                csr::cxp::encoded_1_read(),
-            );
+    fn set_data(value: u16) {
+        unsafe {
+            csr::cxp::downconn_pll_din_write(value);
+        }
+    }
 
-            for _ in 0..20 {
-                timer.delay_us(100);
-                // println!(
-                //     "data[0] = {:#012b} data[1] = {:#012b}",
-                //     csr::cxp::rxdata_0_read(),
-                //     csr::cxp::rxdata_1_read(),
-                // );
-                println!(
-                    "decoded_data[0] = {:#04x} decoded_k[0] = {:#b} decoded_data[1] = {:#04x} decoded_k[1] = {:#b}",
-                    csr::cxp::decoded_data_0_read(),
-                    csr::cxp::decoded_k_0_read(),
-                    csr::cxp::decoded_data_1_read(),
-                    csr::cxp::decoded_k_1_read(),
-                );
+    fn set_enable(en: bool) {
+        unsafe {
+            let val = if en { 1 } else { 0 };
+            csr::cxp::downconn_pll_den_write(val);
+        }
+    }
+
+    fn set_write_enable(en: bool) {
+        unsafe {
+            let val = if en { 1 } else { 0 };
+            csr::cxp::downconn_pll_dwen_write(val);
+        }
+    }
+
+    fn get_data() -> u16 {
+        unsafe { csr::cxp::downconn_pll_dout_read() }
+    }
+
+    fn drp_ready() -> bool {
+        unsafe { csr::cxp::downconn_pll_dready_read() == 1 }
+    }
+
+    #[allow(dead_code)]
+    fn read(address: u8) -> u16 {
+        set_addr(address);
+        set_enable(true);
+        // Set DADDR on the mmcm and assert DEN for one clock cycle
+        one_clock_cycle();
+
+        set_enable(false);
+        while !drp_ready() {
+            // keep the clock signal until data is ready
+            one_clock_cycle();
+        }
+        get_data()
+    }
+
+    fn write(address: u8, value: u16) {
+        set_addr(address);
+        set_data(value);
+        set_write_enable(true);
+        set_enable(true);
+        // Set DADDR, DI on the mmcm and assert DWE, DEN for one clock cycle
+        one_clock_cycle();
+
+        set_write_enable(false);
+        set_enable(false);
+        while !drp_ready() {
+            // keep the clock signal until write is finished
+            one_clock_cycle();
+        }
+    }
+
+    fn reset(rst: bool) {
+        unsafe {
+            let val = if rst { 1 } else { 0 };
+            csr::cxp::downconn_txpll_reset_write(val)
+        }
+    }
+
+    pub fn setup(timer: &mut GlobalTimer, settings: PLLSetting) {
+        if false {
+            info!("0x08 = {:#06x}", read(0x08));
+            info!("0x09 = {:#06x}", read(0x09));
+            info!("0x14 = {:#06x}", read(0x14));
+            info!("0x15 = {:#06x}", read(0x15));
+            info!("0x16 = {:#06x}", read(0x16));
+            info!("0x18 = {:#06x}", read(0x18));
+            info!("0x19 = {:#06x}", read(0x19));
+            info!("0x1A = {:#06x}", read(0x1A));
+            info!("0x28 = {:#06x}", read(0x28));
+            info!("0x4E = {:#06x}", read(0x4E));
+            info!("0x4F = {:#06x}", read(0x4F));
+        } else {
+            // Based on "DRP State Machine" from XAPP888
+            // hold reset HIGH during pll config
+            reset(true);
+            write(0x08, settings.clkout0_reg1);
+            write(0x09, settings.clkout0_reg2);
+            write(0x14, settings.clkfbout_reg1);
+            write(0x15, settings.clkfbout_reg2);
+            write(0x16, settings.div_reg);
+            write(0x18, settings.lock_reg1);
+            write(0x19, settings.lock_reg2);
+            write(0x1A, settings.lock_reg3);
+            write(0x28, settings.power_reg);
+            write(0x4E, settings.filt_reg1);
+            write(0x4F, settings.filt_reg2);
+            reset(false);
+
+            // wait for the pll to lock
+            timer.delay_us(100);
+
+            let locked = unsafe { csr::cxp::downconn_txpll_locked_read() == 1 };
+            info!("txusrclk locked = {}", locked);
+        }
+    }
+
+    pub fn get_txusrclk_config(speed: CXP_SPEED) -> PLLSetting {
+        match speed {
+            CXP_SPEED::CXP_1 => {
+                // CLKFBOUT_MULT = 8, DIVCLK_DIVIDE = 1 , CLKOUT0_DIVIDE = 16
+                // TXUSRCLK=62.5MHz
+                PLLSetting {
+                    clkout0_reg1: 0x1208,  //0x08
+                    clkout0_reg2: 0x0000,  //0x09
+                    clkfbout_reg1: 0x1104, //0x14
+                    clkfbout_reg2: 0x0000, //0x15
+                    div_reg: 0x1041,       //0x16
+                    lock_reg1: 0x03e8,     //0x18
+                    lock_reg2: 0x5801,     //0x19
+                    lock_reg3: 0xdbe9,     //0x1A
+                    power_reg: 0x0000,     //0x28
+                    filt_reg1: 0x9808,     //0x4E
+                    filt_reg2: 0x9100,     //0x4F
+                }
+            }
+            CXP_SPEED::CXP_2 => {
+                // CLKFBOUT_MULT = 8, DIVCLK_DIVIDE = 1 , CLKOUT0_DIVIDE = 8
+                // TXUSRCLK=125MHz
+                PLLSetting {
+                    clkout0_reg1: 0x1104,  //0x08
+                    clkout0_reg2: 0x0000,  //0x09
+                    clkfbout_reg1: 0x1104, //0x14
+                    clkfbout_reg2: 0x0000, //0x15
+                    div_reg: 0x1041,       //0x16
+                    lock_reg1: 0x03e8,     //0x18
+                    lock_reg2: 0x5801,     //0x19
+                    lock_reg3: 0xdbe9,     //0x1A
+                    power_reg: 0x0000,     //0x28
+                    filt_reg1: 0x9808,     //0x4E
+                    filt_reg2: 0x9100,     //0x4F
+                }
+            }
+            CXP_SPEED::CXP_3 => {
+                // CLKFBOUT_MULT = 10, DIVCLK_DIVIDE = 1 , CLKOUT0_DIVIDE = 8
+                // TXUSRCLK=125MHz
+                PLLSetting {
+                    clkout0_reg1: 0x1104,  //0x08
+                    clkout0_reg2: 0x0000,  //0x09
+                    clkfbout_reg1: 0x1145, //0x14
+                    clkfbout_reg2: 0x0000, //0x15
+                    div_reg: 0x1041,       //0x16
+                    lock_reg1: 0x03e8,     //0x18
+                    lock_reg2: 0x7001,     //0x19
+                    lock_reg3: 0xf3e9,     //0x1A
+                    power_reg: 0x0000,     //0x28
+                    filt_reg1: 0x9908,     //0x4E
+                    filt_reg2: 0x1900,     //0x4F
+                }
+            }
+            CXP_SPEED::CXP_5 => {
+                // CLKFBOUT_MULT = 8, DIVCLK_DIVIDE = 1 , CLKOUT0_DIVIDE = 4
+                // TXUSRCLK=250MHz
+                PLLSetting {
+                    clkout0_reg1: 0x1082,  //0x08
+                    clkout0_reg2: 0x0000,  //0x09
+                    clkfbout_reg1: 0x1104, //0x14
+                    clkfbout_reg2: 0x0000, //0x15
+                    div_reg: 0x1041,       //0x16
+                    lock_reg1: 0x03e8,     //0x18
+                    lock_reg2: 0x5801,     //0x19
+                    lock_reg3: 0xdbe9,     //0x1A
+                    power_reg: 0x0000,     //0x28
+                    filt_reg1: 0x9808,     //0x4E
+                    filt_reg2: 0x9100,     //0x4F
+                }
+            }
+            CXP_SPEED::CXP_6 => {
+                // CLKFBOUT_MULT = 10, DIVCLK_DIVIDE = 1 , CLKOUT0_DIVIDE = 4
+                // TXUSRCLK=312.5MHz
+                PLLSetting {
+                    clkout0_reg1: 0x1082,  //0x08
+                    clkout0_reg2: 0x0000,  //0x09
+                    clkfbout_reg1: 0x1145, //0x14
+                    clkfbout_reg2: 0x0000, //0x15
+                    div_reg: 0x1041,       //0x16
+                    lock_reg1: 0x03e8,     //0x18
+                    lock_reg2: 0x7001,     //0x19
+                    lock_reg3: 0xf3e9,     //0x1A
+                    power_reg: 0x0000,     //0x28
+                    filt_reg1: 0x9908,     //0x4E
+                    filt_reg2: 0x1900,     //0x4F
+                }
+            }
+            CXP_SPEED::CXP_10 => {
+                // CLKFBOUT_MULT = 8, DIVCLK_DIVIDE = 1 , CLKOUT0_DIVIDE = 2
+                // TXUSRCLK=500MHz
+                PLLSetting {
+                    clkout0_reg1: 0x1041,  //0x08
+                    clkout0_reg2: 0x0000,  //0x09
+                    clkfbout_reg1: 0x1104, //0x14
+                    clkfbout_reg2: 0x0000, //0x15
+                    div_reg: 0x1041,       //0x16
+                    lock_reg1: 0x03e8,     //0x18
+                    lock_reg2: 0x5801,     //0x19
+                    lock_reg3: 0xdbe9,     //0x1A
+                    power_reg: 0x0000,     //0x28
+                    filt_reg1: 0x9808,     //0x4E
+                    filt_reg2: 0x9100,     //0x4F
+                }
+            }
+            CXP_SPEED::CXP_12 => {
+                // CLKFBOUT_MULT = 10, DIVCLK_DIVIDE = 1 , CLKOUT0_DIVIDE = 2
+                // TXUSRCLK=625MHz
+                PLLSetting {
+                    clkout0_reg1: 0x1041,  //0x08
+                    clkout0_reg2: 0x0000,  //0x09
+                    clkfbout_reg1: 0x1145, //0x14
+                    clkfbout_reg2: 0x0000, //0x15
+                    div_reg: 0x1041,       //0x16
+                    lock_reg1: 0x03e8,     //0x18
+                    lock_reg2: 0x7001,     //0x19
+                    lock_reg3: 0xf3e9,     //0x1A
+                    power_reg: 0x0000,     //0x28
+                    filt_reg1: 0x9908,     //0x4E
+                    filt_reg2: 0x1900,     //0x4F
+                }
             }
         }
     }