1
0
Fork 0

Compare commits

...

35 Commits

Author SHA1 Message Date
morgan 6134ad5794 cxp upconn: refactor idle into its submodule 2024-06-28 17:24:13 +08:00
morgan dbf7ac1cb9 cxp upconn : fix idle word lag behind tx_wordcount 2024-06-28 15:46:13 +08:00
morgan 8ba7793b24 cxp upconn: fix word count issue 2024-06-28 12:46:35 +08:00
morgan 481162430c cxp upconn: add word & char boundary transmission 2024-06-28 12:02:15 +08:00
morgan 4eed5e99f4 cxp upconn: put encoder with tx_fifos 2024-06-27 16:50:15 +08:00
morgan 98f4d4cea4 cxp upconn: refactor fifo into submodule 2024-06-27 12:51:20 +08:00
morgan 14a9184fee cxp upcxp: add tx_enable & startup sequence 2024-06-25 15:29:18 +08:00
morgan ae4bfc40e5 cxp upconn: add priority packet 2024-06-24 14:59:49 +08:00
morgan cdc7294e99 cxp upconn: add sys_clk_freq parameter to set pll 2024-06-20 16:20:27 +08:00
morgan e6550c68cf cxp upconn: add sink to txphy & cleanup 2024-06-20 15:57:37 +08:00
morgan 00e5d32d45 cxp upconn: add priority lv1 fifo & refactor 2024-06-20 14:56:28 +08:00
morgan 779084d5dc cxp upconn: setup fifo 2024-06-20 10:57:05 +08:00
morgan f43c8b8bac cxp upconn: rename bits to tx_bitcount 2024-06-18 16:42:17 +08:00
morgan 6636339701 cxp upconn: refactor serial fsm as sync logic 2024-06-18 16:40:40 +08:00
morgan bd66659883 cxp upconn: write state fsm refactor 2024-06-18 15:58:53 +08:00
morgan 4de58e0d52 cxp upconn: use singlencoder & fix disparity bug 2024-06-18 14:40:58 +08:00
morgan 7c60fb5776 cxp upconn: init encoder 2024-06-18 12:26:10 +08:00
morgan bbf9e37867 cxp upconn: rename high speed upconn to bitrate2x 2024-06-18 12:24:31 +08:00
morgan d36afd6f7b CXP upconn: add sys reset signal to pll 2024-06-17 16:51:55 +08:00
morgan cb1ec7f62a CXP upconn: remove unused code 2024-06-17 14:46:49 +08:00
morgan 48d3a9cd4a CXP up connection gw: init 2024-06-17 13:19:37 +08:00
morgan 96a052513d zc706: use new CXP file name 2024-06-14 17:18:26 +08:00
morgan b2c6c20426 cxp: refactor to its separte module 2024-06-14 17:17:56 +08:00
morgan 5fcf5cb70f cxp gtx: rename to cxp_downconn 2024-06-14 17:17:16 +08:00
morgan 47e1a83519 cxp fmc: rename file 2024-06-14 17:16:51 +08:00
morgan d592825284 CXP gtx: rename to CXP DownConn 2024-06-14 17:07:58 +08:00
morgan b52589bd5f CXP CLK alignment: clock domain naming refactor 2024-06-13 16:56:36 +08:00
morgan 9c69167f7f CXP gtx: Clock domain naming refactor and remove cxp_gtx 2024-06-13 16:56:16 +08:00
morgan 8ac71b37a9 CXP GTX: init 2024-06-12 14:50:38 +08:00
morgan ce0e14879c Gateware: ZC706 CXP GTX setup 2024-06-05 13:02:52 +08:00
morgan df7feb3b17 Gateware: CXP_GTX init 2024-06-05 13:02:52 +08:00
morgan f5c604bbb5 zc706: add USER LED to allow compilation 2024-06-05 13:02:29 +08:00
morgan d61d7a5a95 zc706: add CXP FMC variant 2024-06-05 13:02:29 +08:00
morgan 681d7400c7 FMCIO: add cxp_4r adepter io 2024-06-05 13:02:29 +08:00
morgan 9f2f392728 flake: add CXP_FMC variant build options 2024-06-05 13:02:29 +08:00
6 changed files with 976 additions and 4 deletions

View File

@ -341,6 +341,7 @@
{ {
inherit fastnumbers artiq-netboot ramda migen-axi binutils-arm; inherit fastnumbers artiq-netboot ramda migen-axi binutils-arm;
} // } //
(board-package-set { target = "zc706"; variant = "cxp_fmc"; }) //
(board-package-set { target = "zc706"; variant = "nist_clock"; }) // (board-package-set { target = "zc706"; variant = "nist_clock"; }) //
(board-package-set { target = "zc706"; variant = "nist_clock_master"; }) // (board-package-set { target = "zc706"; variant = "nist_clock_master"; }) //
(board-package-set { target = "zc706"; variant = "nist_clock_master_100mhz"; }) // (board-package-set { target = "zc706"; variant = "nist_clock_master_100mhz"; }) //

118
src/gateware/cxp.py Normal file
View File

@ -0,0 +1,118 @@
from migen import *
from misoc.interconnect.csr import *
from artiq.gateware.drtio.core import TransceiverInterface, ChannelInterface
from cxp_downconn import CXP_DownConn
class CXP(Module, AutoCSR):
def __init__(self, refclk, pads, sys_clk_freq, debug_sma):
self.nchannels = 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.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()
# # #
# single CXP
self.submodules.gtx = gtx = CXP_DownConn(refclk, pads, sys_clk_freq, tx_mode="single", rx_mode="single")
# ! loopback for debugging
self.sync += gtx.loopback_mode.eq(self.loopback_mode.storage)
# ! 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.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)
]
# TODO: add low speed SERDES

View File

@ -0,0 +1,87 @@
from migen.build.generic_platform import *
fmc_adapter_io = [
# CoaXPress high speed link
("CXP_HS", 0,
Subsignal("txp", Pins("HPC:DP0_C2M_P")),
Subsignal("txn", Pins("HPC:DP0_C2M_N")),
Subsignal("rxp", Pins("HPC:DP0_M2C_P")),
Subsignal("rxn", Pins("HPC:DP0_M2C_N")),
),
("CXP_HS", 1,
Subsignal("txp", Pins("HPC:DP1_C2M_P")),
Subsignal("txn", Pins("HPC:DP1_C2M_N")),
Subsignal("rxp", Pins("HPC:DP1_M2C_P")),
Subsignal("rxn", Pins("HPC:DP1_M2C_N")),
),
("CXP_HS", 2,
Subsignal("txp", Pins("HPC:DP2_C2M_P")),
Subsignal("txn", Pins("HPC:DP2_C2M_N")),
Subsignal("rxp", Pins("HPC:DP2_M2C_P")),
Subsignal("rxn", Pins("HPC:DP2_M2C_n")),
),
("CXP_HS", 3,
Subsignal("txp", Pins("HPC:DP3_C2M_P")),
Subsignal("txn", Pins("HPC:DP3_C2M_N")),
Subsignal("rxp", Pins("HPC:DP3_M2C_P")),
Subsignal("rxn", Pins("HPC:DP3_M2C_N")),
),
# CoaXPress low speed link
("CXP_LS", 0, Pins("HPC:LA00_CC_P"), IOStandard("LVCMOS25")),
("CXP_LS", 1, Pins("HPC:LA01_CC_N"), IOStandard("LVCMOS25")),
("CXP_LS", 2, Pins("HPC:LA01_CC_P"), IOStandard("LVCMOS25")),
("CXP_LS", 3, Pins("HPC:LA02_N"), IOStandard("LVCMOS25")),
# CoaXPress green and red LED
("CXP_LED", 0,
Subsignal("green", Pins("HPC:LA11_P"), IOStandard("LVCMOS25")),
Subsignal("red", Pins("HPC:LA11_N"), IOStandard("LVCMOS25")),
),
("CXP_LED", 1,
Subsignal("green", Pins("HPC:LA12_P"), IOStandard("LVCMOS25")),
Subsignal("red", Pins("HPC:LA12_N"), IOStandard("LVCMOS25")),
),
("CXP_LED", 2,
Subsignal("green", Pins("HPC:LA13_P"), IOStandard("LVCMOS25")),
Subsignal("red", Pins("HPC:LA13_N"), IOStandard("LVCMOS25")),
),
("CXP_LED", 3,
Subsignal("green", Pins("HPC:LA14_P"), IOStandard("LVCMOS25")),
Subsignal("red", Pins("HPC:LA14_N"), IOStandard("LVCMOS25")),
),
# Power over CoaXPress
("PoCXP", 0,
Subsignal("enable", Pins("HPC:LA21_N"), IOStandard("LVCMOS25")),
Subsignal("alert", Pins("HPC:LA18_CC_P"), IOStandard("LVCMOS25")),
),
("PoCXP", 1,
Subsignal("enable", Pins("HPC:LA21_P"), IOStandard("LVCMOS25")),
Subsignal("alert", Pins("HPC:LA19_N"), IOStandard("LVCMOS25")),
),
("PoCXP", 2,
Subsignal("enable", Pins("HPC:LA22_N"), IOStandard("LVCMOS25")),
Subsignal("alert", Pins("HPC:LA19_P"), IOStandard("LVCMOS25")),
),
("PoCXP", 3,
Subsignal("enable", Pins("HPC:LA22_P"), IOStandard("LVCMOS25")),
Subsignal("alert", Pins("HPC:LA20_N"), IOStandard("LVCMOS25")),
),
("i2c_fmc", 0,
Subsignal("scl", Pins("HPC:IIC_SCL")),
Subsignal("sda", Pins("HPC:IIC_SDA")),
IOStandard("LVCMOS25")
),
("3V3", 0, Pins("HPC:PG_M2C")),
("GND", 0, Pins("HPC:PRSNT_M2C_L HPC:CLK0_M2C_P")),
("VADJ", 0, Pins("HPC:GBTCLK1_M2C_N CLK0_M2C_N")),
("clk125_fmc", 0,
Subsignal("p", Pins("HPC:GBTCLK0_M2C_P")),
Subsignal("n", Pins("HPC:GBTCLK0_M2C_n")),
),
]

View File

@ -0,0 +1,427 @@
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 *
from artiq.gateware.drtio.core import TransceiverInterface, ChannelInterface
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".
#
# 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, tx_clk_freq, check_period=6e-3):
self.rxdata = Signal(20)
self.restart = Signal()
self.ready = Signal()
check_max_val = ceil(check_period*tx_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
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)
).Elif((rx1cnt != 4) & (rx1cnt != 5) & (rx1cnt != 6),
error_seen_rxclk.eq(1)
)
]
fsm = FSM(reset_state="WAIT_COMMA")
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.
If(comma_seen,
NextState("WAIT_NOERROR")
).Else(
self.restart.eq(1)
),
checks_reset.i.eq(1)
)
)
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("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")
)
)
class CXP_DownConn(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"):
assert tx_mode in ["single", "master", "slave"]
assert rx_mode in ["single", "master", "slave"]
cpll_div = 4
pll_div = int(40/cpll_div)
self.rx_restart = Signal()
self.rx_bypass_clk_alignment = Signal()
self.tx_restart = Signal()
self.loopback_mode = Signal(3)
self.txenable = Signal()
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
self.txoutclk = Signal()
self.rxoutclk = Signal()
# # #
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)
self.comb += [
cpllreset.eq(tx_init.cpllreset),
tx_init.cplllock.eq(cplllock),
rx_init.cplllock.eq(cplllock)
]
txdata = Signal(20)
rxdata = Signal(20)
# Note: the following parameters were set after consulting AR45360
self.specials += \
Instance("GTXE2_CHANNEL",
# PMA Attributes
p_PMA_RSV=0x00018480,
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,
# 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,
# 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=20,
p_TX_INT_DATAWIDTH=0,
i_TXCHARDISPMODE=Cat(txdata[9], txdata[19]),
i_TXCHARDISPVAL=Cat(txdata[8], txdata[18]),
i_TXDATA=Cat(txdata[:8], txdata[10:18]),
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=0, # RXLPMEN = 0: DFE mode is enabled
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=20,
p_RX_INT_DATAWIDTH=0,
o_RXDISPERR=Cat(rxdata[9], rxdata[19]),
o_RXCHARISK=Cat(rxdata[8], rxdata[18]),
o_RXDATA=Cat(rxdata[:8], rxdata[10:18]),
# 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_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="PCS",
p_RX_SIG_VALID_DLY=10,
# 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_1040_0020, # DFE @ <= 6.6Gb/s, 8B/10B encoded data, CDR setting < +/- 200ppm
# (See UG476 (v1.12.1), p.205)
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,
# ! 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()
txpll_reset = Signal()
txpll_locked = Signal()
txoutclk_buf = Signal()
txpll_clkout = Signal()
self.specials += [
Instance("PLLE2_ADV",
p_BANDWIDTH="HIGH",
o_LOCKED=txpll_locked,
i_RST=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,
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,
# TODO: DRP for line rate change
),
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)
]
# 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.decoders[0].input.eq(rxdata[:10]),
self.decoders[1].input.eq(rxdata[10:])
]
clock_aligner = CXP_BruteforceClockAligner(0b0101111100, sys_clk_freq, check_period=1e-3)
self.submodules += clock_aligner
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)
]

291
src/gateware/cxp_upconn.py Normal file
View File

@ -0,0 +1,291 @@
from migen import *
from migen.genlib.resetsync import AsyncResetSynchronizer
from migen.genlib.coding import PriorityEncoder
from misoc.cores.code_8b10b import SingleEncoder
from misoc.interconnect import stream
from misoc.interconnect.csr import *
class CXP_UpConn(Module, AutoCSR):
def __init__(self, pads, sys_clk_freq, pmod, nfifos=3, fifo_depth=32):
self.clock_domains.cd_cxp_upconn = ClockDomain()
self.clk_reset = CSRStorage(reset=1)
self.bitrate2x_enable = CSRStorage()
self.tx_enable = CSRStorage()
# # #
pll_locked = Signal()
pll_fb_clk = Signal()
pll_cxpclk = Signal()
pll_cxpclk2x = Signal()
self.specials += [
Instance("PLLE2_ADV",
p_BANDWIDTH="HIGH",
o_LOCKED=pll_locked,
i_RST=ResetSignal("sys"),
p_CLKIN1_PERIOD=1e9/sys_clk_freq, # ns
i_CLKIN1=ClockSignal("sys"),
# VCO @ 1.25GHz
p_CLKFBOUT_MULT=1.25e9/sys_clk_freq, p_DIVCLK_DIVIDE=1,
i_CLKFBIN=pll_fb_clk, o_CLKFBOUT=pll_fb_clk,
# 20.83MHz (48ns)
p_CLKOUT0_DIVIDE=60, p_CLKOUT0_PHASE=0.0, o_CLKOUT0=pll_cxpclk,
# 41.66MHz (24ns) for downconnection over 6.25Gpbs
p_CLKOUT1_DIVIDE=30, p_CLKOUT1_PHASE=0.0, o_CLKOUT1=pll_cxpclk2x,
),
Instance("BUFGMUX",
i_I0=pll_cxpclk,
i_I1=pll_cxpclk2x,
i_S=self.bitrate2x_enable.storage,
o_O=self.cd_cxp_upconn.clk
),
AsyncResetSynchronizer(self.cd_cxp_upconn, ~pll_locked | self.clk_reset.storage)
]
self.submodules.fsm = ClockDomainsRenamer("cxp_upconn")(FSM(reset_state="WAIT_TX_ENABLE"))
self.submodules.tx_fifos = TxFIFOs(nfifos, fifo_depth)
self.submodules.tx_idle = TxIdle()
o = Signal()
tx_en = Signal()
tx_bitcount = Signal(max=10)
tx_wordcount = Signal()
tx_reg = Signal(10)
disp = Signal()
tx_wordcount = Signal(max=4)
priority = Signal(max=nfifos)
idling = Signal()
# startup sequence
self.fsm.act("WAIT_TX_ENABLE",
NextValue(self.tx_idle.disp_in, 0),
NextValue(self.tx_idle.word_idx, 0),
If(self.tx_enable.storage,
NextValue(tx_wordcount, 0),
NextValue(tx_bitcount, 0),
NextState("LOAD_CHAR")
)
)
self.fsm.act("LOAD_CHAR",
NextValue(idling, 1),
NextValue(self.tx_idle.source_ack, 1),
NextValue(tx_reg, self.tx_idle.source_data),
NextValue(disp, self.tx_idle.disp_out),
NextState("START_TX")
)
self.fsm.act("START_TX",
tx_en.eq(1),
If((~self.tx_enable.storage) & (tx_wordcount == 3),
NextState("WAIT_TX_ENABLE")
)
)
# CXP 2.1 section 9.2.4
# Higher priority packet can be inserted into a lower priority packet during transmission
# Priority lv 0 can be inserted in char boundary of the packet
# Priority lv 1-2 need to be inserted in word boundary of the packet
self.sync.cxp_upconn += [
self.tx_fifos.disp_in.eq(disp),
self.tx_idle.disp_in.eq(disp),
If(tx_en,
o.eq(tx_reg[0]),
tx_reg.eq(Cat(tx_reg[1:], 0)),
tx_bitcount.eq(tx_bitcount + 1),
# char boundary
If(tx_bitcount == 9,
tx_bitcount.eq(0),
If((~self.tx_fifos.pe.n) & (self.tx_fifos.pe.o == 0),
# trigger packet can interrupt at char level
tx_reg.eq(self.tx_fifos.source_data[0]),
self.tx_fifos.source_ack[0].eq(1),
disp.eq(self.tx_fifos.disp_out[0]),
).Else(
# priority zero doesn't contribute to word count
If(tx_wordcount != 3,
tx_wordcount.eq(tx_wordcount + 1),
).Else(
tx_wordcount.eq(0),
),
# new word boundary
If(tx_wordcount == 3,
If(~self.tx_fifos.pe.n,
idling.eq(0),
priority.eq(self.tx_fifos.pe.o),
self.tx_fifos.source_ack[self.tx_fifos.pe.o].eq(1),
tx_reg.eq(self.tx_fifos.source_data[self.tx_fifos.pe.o]),
disp.eq(self.tx_fifos.disp_out[self.tx_fifos.pe.o]),
).Else(
idling.eq(1),
self.tx_idle.source_ack.eq(1),
tx_reg.eq(self.tx_idle.source_data),
disp.eq(self.tx_idle.disp_out),
)
).Else(
If(~idling,
self.tx_fifos.source_ack[priority].eq(1),
tx_reg.eq(self.tx_fifos.source_data[priority]),
disp.eq(self.tx_fifos.disp_out[priority]),
).Else(
self.tx_idle.source_ack.eq(1),
tx_reg.eq(self.tx_idle.source_data),
disp.eq(self.tx_idle.disp_out),
)
),
)
)
).Else(
o.eq(0)
)
]
# DEBUG: remove pads
self.encoded_data = CSRStatus(10)
self.sync.cxp_upconn +=[
If(tx_bitcount == 0,
self.encoded_data.status.eq(tx_reg),
)
]
ninth_bit = Signal()
word_bound = Signal()
p0 = Signal()
p3 = Signal()
self.comb += [
ninth_bit.eq(tx_bitcount == 9),
word_bound.eq(tx_wordcount == 3),
p0.eq(self.tx_idle.word_idx == 0),
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),
# 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)
self.symbol2 = CSR(9)
self.sync += [
self.tx_fifos.sink_stb[0].eq(self.symbol0.re),
self.tx_fifos.sink_data[0].eq(self.symbol0.r),
self.tx_fifos.sink_stb[1].eq(self.symbol1.re),
self.tx_fifos.sink_data[1].eq(self.symbol1.r),
self.tx_fifos.sink_stb[2].eq(self.symbol2.re),
self.tx_fifos.sink_data[2].eq(self.symbol2.r),
]
class TxFIFOs(Module):
def __init__(self, nfifos, fifo_depth):
self.disp_in = Signal()
self.disp_out = Array(Signal() for _ in range(nfifos))
self.sink_stb = Signal(nfifos)
self.sink_ack = Signal(nfifos)
self.sink_data = [Signal(9) for _ in range(nfifos)]
self.source_ack = Array(Signal() for _ in range(nfifos))
self.source_data = Array(Signal(10) for _ in range(nfifos))
# # #
source_stb = Signal(nfifos)
for i in range(nfifos):
cdr = ClockDomainsRenamer({"write": "sys", "read": "cxp_upconn"})
fifo = cdr(stream.AsyncFIFO([("data", 9)], fifo_depth))
encoder = ClockDomainsRenamer("cxp_upconn")(SingleEncoder(True))
setattr(self.submodules, "tx_fifo" + str(i), fifo)
setattr(self.submodules, "tx_encoder" + str(i), encoder)
self.sync += [
fifo.sink.stb.eq(self.sink_stb[i]),
self.sink_ack[i].eq(fifo.sink.ack),
fifo.sink.data.eq(self.sink_data[i]),
]
self.sync.cxp_upconn += [
encoder.d.eq(fifo.source.data[:8]),
encoder.k.eq(fifo.source.data[8]),
encoder.disp_in.eq(self.disp_in),
self.disp_out[i].eq(encoder.disp_out),
source_stb[i].eq(fifo.source.stb),
fifo.source.ack.eq(self.source_ack[i]),
self.source_data[i].eq(encoder.output),
# reset ack after asserted
If(self.source_ack[i], self.source_ack[i].eq(0)),
]
# FIFOs transmission priority
self.submodules.pe = PriorityEncoder(nfifos)
self.comb += self.pe.i.eq(source_stb)
class TxIdle(Module):
def __init__(self):
self.disp_in = Signal()
self.disp_out = Signal()
self.word_idx = Signal(max=4)
self.source_ack = Signal()
self.source_data = Signal(10)
# # #
# CXP 2.1 section 9.2.5
IDLE_CHARS = Array([
#[data, k]
[0b10111100, 1], #K28.5
[0b00111100, 1], #K28.1
[0b00111100, 1], #K28.1
[0b10111100, 0], #D28.5
])
encoder = ClockDomainsRenamer("cxp_upconn")(SingleEncoder(True))
self.submodules += encoder
self.sync.cxp_upconn += [
encoder.d.eq(IDLE_CHARS[self.word_idx][0]),
encoder.k.eq(IDLE_CHARS[self.word_idx][1]),
encoder.disp_in.eq(self.disp_in),
self.disp_out.eq(encoder.disp_out),
self.source_data.eq(encoder.output),
If(self.source_ack,
# reset after asserted
self.source_ack.eq(0),
If(self.word_idx != 3,
self.word_idx.eq(self.word_idx + 1),
).Else(
self.word_idx.eq(0),
)
),
]

View File

@ -25,6 +25,7 @@ import analyzer
import acpki import acpki
import drtio_aux_controller import drtio_aux_controller
import zynq_clocking import zynq_clocking
import cxp_4r_fmc, cxp
from config import write_csr_file, write_mem_file, write_rustc_cfg_file from config import write_csr_file, write_mem_file, write_rustc_cfg_file
class SMAClkinForward(Module): class SMAClkinForward(Module):
@ -138,7 +139,7 @@ class ZC706(SoCCore):
platform.add_extension(si5324_fmc33) platform.add_extension(si5324_fmc33)
self.comb += platform.request("si5324_33").rst_n.eq(1) self.comb += platform.request("si5324_33").rst_n.eq(1)
cdr_clk = Signal() self.cdr_clk = Signal()
cdr_clk_buf = Signal() cdr_clk_buf = Signal()
si5324_out = platform.request("si5324_clkout") si5324_out = platform.request("si5324_clkout")
platform.add_period_constraint(si5324_out.p, 8.0) platform.add_period_constraint(si5324_out.p, 8.0)
@ -146,11 +147,11 @@ class ZC706(SoCCore):
Instance("IBUFDS_GTE2", Instance("IBUFDS_GTE2",
i_CEB=0, i_CEB=0,
i_I=si5324_out.p, i_IB=si5324_out.n, i_I=si5324_out.p, i_IB=si5324_out.n,
o_O=cdr_clk, o_O=self.cdr_clk,
p_CLKCM_CFG="TRUE", p_CLKCM_CFG="TRUE",
p_CLKRCV_TRST="TRUE", p_CLKRCV_TRST="TRUE",
p_CLKSWING_CFG=3), p_CLKSWING_CFG=3),
Instance("BUFG", i_I=cdr_clk, o_O=cdr_clk_buf) Instance("BUFG", i_I=self.cdr_clk, o_O=cdr_clk_buf)
] ]
self.config["HAS_SI5324"] = None self.config["HAS_SI5324"] = None
self.config["SI5324_AS_SYNTHESIZER"] = None self.config["SI5324_AS_SYNTHESIZER"] = None
@ -648,6 +649,47 @@ class _NIST_QC2_RTIO:
self.add_rtio(rtio_channels) self.add_rtio(rtio_channels)
class _CXP_FMC_RTIO():
"""
CoaXpress FMC with 4 CXP channel and 1 SMA trigger
"""
def __init__(self):
platform = self.platform
platform.add_extension(cxp_4r_fmc.fmc_adapter_io)
platform.add_extension(leds_fmc33)
debug_sma = [
("user_sma_clock_33", 0,
Subsignal("p_tx", Pins("AD18"), IOStandard("LVCMOS33")),
Subsignal("n_rx", Pins("AD19"), IOStandard("LVCMOS33")),
),
]
platform.add_extension(debug_sma)
clk_freq = 125e6
self.submodules.cxp_gtx = cxp.CXP(
refclk=self.cdr_clk,
pads=platform.request("CXP_HS", 0),
sys_clk_freq=clk_freq,
debug_sma=platform.request("user_sma_clock_33")
)
self.csr_devices.append("cxp_gtx")
rtio_channels = []
# FIXME remove this placeholder RTIO channel
# There are too few RTIO channels and cannot be compiled (adr width issue of the lane distributor)
# see https://github.com/m-labs/artiq/pull/2158 for similar issue
print("USER LED at RTIO channel 0x{:06x}".format(len(rtio_channels)))
phy = ttl_simple.Output(self.platform.request("user_led_33", 0))
self.submodules += phy
rtio_channels.append(rtio.Channel.from_phy(phy))
self.config["HAS_RTIO_LOG"] = None
rtio_channels.append(rtio.LogChannel())
self.config["RTIO_LOG_CHANNEL"] = len(rtio_channels)
self.add_rtio(rtio_channels)
class NIST_CLOCK(ZC706, _NIST_CLOCK_RTIO): class NIST_CLOCK(ZC706, _NIST_CLOCK_RTIO):
def __init__(self, acpki, drtio100mhz): def __init__(self, acpki, drtio100mhz):
ZC706.__init__(self, acpki) ZC706.__init__(self, acpki)
@ -680,8 +722,14 @@ class NIST_QC2_Satellite(_SatelliteBase, _NIST_QC2_RTIO):
_SatelliteBase.__init__(self, acpki, drtio100mhz) _SatelliteBase.__init__(self, acpki, drtio100mhz)
_NIST_QC2_RTIO.__init__(self) _NIST_QC2_RTIO.__init__(self)
class CXP_FMC(ZC706, _CXP_FMC_RTIO):
def __init__(self, acpki, drtio100mhz):
ZC706.__init__(self, acpki)
# self.submodules += SMAClkinForward(self.platform)
_CXP_FMC_RTIO.__init__(self)
VARIANTS = {cls.__name__.lower(): cls for cls in [NIST_CLOCK, NIST_CLOCK_Master, NIST_CLOCK_Satellite, VARIANTS = {cls.__name__.lower(): cls for cls in [NIST_CLOCK, NIST_CLOCK_Master, NIST_CLOCK_Satellite,
NIST_QC2, NIST_QC2_Master, NIST_QC2_Satellite]} NIST_QC2, NIST_QC2_Master, NIST_QC2_Satellite, CXP_FMC]}
def main(): def main():
parser = argparse.ArgumentParser( parser = argparse.ArgumentParser(