Wade Fife
ea0b8f4012
Adds a 266.667 MHz compute engine clock and uses it instead of the radio clock for the DDC and DUC. Using a faster clock than the radio allows the DDC and DUC to drain after a stall in the data flow. Original-commit: ab93cb0929b9536b6d6cbfdc4813e6ba169053ae
287 lines
14 KiB
Tcl
287 lines
14 KiB
Tcl
#
|
|
# Copyright 2021 Ettus Research, a National Instruments Brand
|
|
#
|
|
# SPDX-License-Identifier: LGPL-3.0-or-later
|
|
#
|
|
# Description:
|
|
# Common timing constraints for X410.
|
|
#
|
|
|
|
###############################################################################
|
|
# Motherboard Clocks
|
|
###############################################################################
|
|
|
|
# 10/25 MHz reference clock from rear panel connector.
|
|
# Constrain to the fastest possible clock rate.
|
|
set ref_clk_period 40.00
|
|
create_clock -name ref_clk -period $ref_clk_period [get_ports BASE_REFCLK_FPGA_P]
|
|
|
|
# MGT Clocks
|
|
# Clock Reference | Frequency | Purpose
|
|
# MGT_REFCLK_LMK0 | 156.25/125 MHz | 10 GbE
|
|
# MGT_REFCLK_LMK1 | 100.00 MHz | Reserved
|
|
# MGT_REFCLK_LMK2 | 100.00 MHz | Reserved
|
|
# MGT_REFCLK_LMK3 | 156.25/125 MHz | 10 GbE
|
|
create_clock -name mgt_ref_0 -period 6.400 [get_ports MGT_REFCLK_LMK0_P]
|
|
create_clock -name mgt_ref_1 -period 10.000 [get_ports MGT_REFCLK_LMK1_P]
|
|
create_clock -name mgt_ref_2 -period 10.400 [get_ports MGT_REFCLK_LMK2_P]
|
|
create_clock -name mgt_ref_3 -period 6.400 [get_ports MGT_REFCLK_LMK3_P]
|
|
|
|
# Virtual clocks for constraining misc. I/Os.
|
|
create_clock -name async_in_clk -period 50.00
|
|
create_clock -name async_out_clk -period 50.00
|
|
|
|
|
|
###############################################################################
|
|
# Aliases for auto-generated clocks
|
|
###############################################################################
|
|
|
|
# Name the PS clocks. These are originally declared in the PS8 IP block.
|
|
# Create the clocks based on the PS PLCLK pins.
|
|
# This generates critical warnings in the OSS flow because the clocks were already
|
|
# define and we are completely rewriting the old clock definition... this is OK.
|
|
create_clock -name clk100 -period 10.000 \
|
|
[get_pins -of_objects [get_cells -hierarchical {*PS8_i}] -filter {NAME =~ *PLCLK[0]}]
|
|
|
|
create_clock -name clk40 -period 25.000 \
|
|
[get_pins -of_objects [get_cells -hierarchical {*PS8_i}] -filter {NAME =~ *PLCLK[1]}]
|
|
|
|
create_clock -name clk166 -period 6.000 \
|
|
[get_pins -of_objects [get_cells -hierarchical {*PS8_i}] -filter {NAME =~ *PLCLK[2]}]
|
|
|
|
create_clock -name clk200 -period 5.000 \
|
|
[get_pins -of_objects [get_cells -hierarchical {*PS8_i}] -filter {NAME =~ *PLCLK[3]}]
|
|
|
|
create_generated_clock -name ce_clk [get_pins {ce_clk_gen_i/CLKOUT0}]
|
|
|
|
###############################################################################
|
|
# Sync to DB synthesizer sync CPLD input
|
|
###############################################################################
|
|
|
|
# synth_sync_hold_requirement and synth_sync_setup_requirement are shared
|
|
# between the FPGA and DB CPLD. The values are set in shared_constants.sdc
|
|
set synth_sync_ports [get_ports {DB0_SYNTH_SYNC DB1_SYNTH_SYNC}]
|
|
set_output_delay -clock [get_clocks pll_ref_clk] -min -$synth_sync_hold_requirement $synth_sync_ports
|
|
set_output_delay -clock [get_clocks pll_ref_clk] -max $synth_sync_setup_requirement $synth_sync_ports
|
|
|
|
###############################################################################
|
|
# SPI to MB CPLD (PL)
|
|
# This interface is defined as system synchronous to pll_ref_clk.
|
|
###############################################################################
|
|
|
|
# Both CPLD and FPGA use PLL reference clock from a common clock chip.
|
|
# The traces from that clock chip to the ICs are not length matched. Assume a
|
|
# worst case clock difference of 0.5 ns at the IC inputs. There is no direction
|
|
# defined. The clock can arrive faster or slower at one IC.
|
|
set pl_clock_diff 0.500
|
|
|
|
# The longest trace on the PL SPI interface is (assuming 170.0 ps/in)
|
|
# Longest trace | Trace length | Trace delay
|
|
# PL_CPLD_MISO | 3.863 in | 0.657 ns
|
|
set pl_spi_board_delay 0.657
|
|
|
|
# Output delay timings of the MB CPLD design, which still meet timing
|
|
set pl_spi_cpld_min_out -1.000
|
|
set pl_spi_cpld_max_out 8.000
|
|
|
|
set spi_in_port [get_ports {PL_CPLD_MISO}]
|
|
set_input_delay -clock [get_clocks pll_ref_clk] \
|
|
-min [expr {- $pl_spi_cpld_min_out - $pl_clock_diff}] \
|
|
$spi_in_port
|
|
set_input_delay -clock [get_clocks pll_ref_clk] \
|
|
-max [expr {$pll_ref_clk_period - $pl_spi_cpld_max_out + $pl_spi_board_delay + $pl_clock_diff}] \
|
|
$spi_in_port
|
|
|
|
|
|
###############################################################################
|
|
# 10 GbE
|
|
###############################################################################
|
|
|
|
# These are the exceptions from "xge_pcs_pma_exceptions.xdc" which are to be
|
|
# used when not using the example design.
|
|
#
|
|
# "clk100" used here is the clock that's connected to the "dclk" input in the core.
|
|
set_max_delay -from [get_clocks -of_objects [get_pins -hierarchical -filter {NAME =~ */channel_inst/*_CHANNEL_PRIM_INST/RXOUTCLK}]] -to [get_clocks -of_objects [get_pins -hierarchical -filter {NAME =~ */channel_inst/*_CHANNEL_PRIM_INST/TXOUTCLK}]] -datapath_only 6.40
|
|
set_max_delay -from [get_clocks -of_objects [get_pins -hierarchical -filter {NAME =~ */channel_inst/*_CHANNEL_PRIM_INST/TXOUTCLK}]] -to [get_clocks -of_objects [get_pins -hierarchical -filter {NAME =~ */channel_inst/*_CHANNEL_PRIM_INST/RXOUTCLK}]] -datapath_only 6.40
|
|
set_max_delay -from [get_clocks -of_objects [get_pins -hierarchical -filter {NAME =~ */channel_inst/*_CHANNEL_PRIM_INST/RXOUTCLK}]] -to [get_clocks clk100] -datapath_only 6.40
|
|
set_max_delay -from [get_clocks -of_objects [get_pins -hierarchical -filter {NAME =~ */channel_inst/*_CHANNEL_PRIM_INST/TXOUTCLK}]] -to [get_clocks clk100] -datapath_only 6.40
|
|
set_max_delay -from [get_clocks clk100] -to [get_clocks -of_objects [get_pins -hierarchical -filter {NAME =~ */channel_inst/*_CHANNEL_PRIM_INST/TXOUTCLK}]] -datapath_only 10.000
|
|
set_max_delay -from [get_clocks clk100] -to [get_clocks -of_objects [get_pins -hierarchical -filter {NAME =~ */channel_inst/*_CHANNEL_PRIM_INST/RXOUTCLK}]] -datapath_only 10.000
|
|
|
|
|
|
###############################################################################
|
|
# DIO
|
|
# Those GPIO pins are considered asynchronous paths. The user has to add
|
|
# constraints in case required. Therefore not setting false_paths from / to
|
|
# user logic to allow user generated timing constraints to be applied.
|
|
###############################################################################
|
|
|
|
# Ignore paths from "slow" PS interface to not interfere with user constraints.
|
|
set dio_ports [get_ports {DIOA_FPGA[*] DIOB_FPGA[*]}]
|
|
set dio_registers [get_cells -hierarchical -filter {NAME =~ *x4xx_dio_i* && IS_SEQUENTIAL && IS_PRIMITIVE}]
|
|
set_false_path -from $dio_registers -to $dio_ports
|
|
set_false_path -from $dio_ports -to $dio_registers
|
|
|
|
|
|
###############################################################################
|
|
# PPS
|
|
###############################################################################
|
|
|
|
# The TRIG_IO port may be driven by either the PPS in BRC domain to
|
|
# enable direct sync between 2 devices, or by any other user logic.
|
|
# When PPS is exported through Trigger I/O, timing must be analyzed
|
|
# to ensure determinism in the PPS exporting.
|
|
# But, when other user logic drives TRIG_IO, then the port should be
|
|
# treated as asynchronous (or close to async. at least).
|
|
# To achieve this conditional timing analysis, the following trick is
|
|
# used:
|
|
# 1. A virtual copy of ref_clk is created for I/O timing - virtual_ref_clk
|
|
# 2. Set output_delay constraints to assign a clock to the TRIG_IO port.
|
|
# 3. A set_max_delay constraint is used to time the output path to TRIG_IO
|
|
# set_max_delay makes the timing constraint driver agnostic, and as long
|
|
# as the critical output delay is met for driving PPS through TRIG_IO,
|
|
# we should be fine as this requirement is relatively loose.
|
|
|
|
# 1) Creating copy of ref_clk to only analyze timing to TRIG_IO port (output)
|
|
# when output is driven by ref_clk (PPS generation in ref_clk domain).
|
|
create_clock -name virtual_ref_clk -period $ref_clk_period
|
|
|
|
# Trigger IO port is used as output for the PPS signal
|
|
# TRIG_IO_1V8 trace length MB = 4.050 + 1.190 inch = 5.240 inch
|
|
# TRIG_IO_1V8 trace length DB = 2.401 + 0.120 + 0.457 + 0.261 inch = 3.239 inch
|
|
# TRIG_IO buffer max switching time = 3.3
|
|
set trig_max_out_delay [expr {8.479 * 0.17 + 3.3}]
|
|
|
|
# Set minimum output delay hold time to a small amount to grant external
|
|
# devices some hold time. Delay should be simple to achieve as there is no PLL
|
|
# in the clocking path and some combinatorial logic.
|
|
set trig_min_out_delay 2.000
|
|
|
|
# 2) set_output_delay for assigning clocks to TRIG_IO. Use zero for delay to
|
|
# avoid adding extra delay requirements on top of the set_max|min_delay
|
|
# constraints below.
|
|
set_output_delay -clock [get_clocks virtual_ref_clk] 0.0 [get_ports {TRIG_IO}]
|
|
|
|
# 3) Min and max delays make constraining driver agnostic. We just make sure
|
|
# the critical timing for PPS export is met though.
|
|
set_max_delay -through [get_port {TRIG_IO}] -to [get_clocks {virtual_ref_clk}] \
|
|
[expr {$ref_clk_period - $trig_max_out_delay}]
|
|
set_min_delay -through [get_port {TRIG_IO}] -to [get_clocks {virtual_ref_clk}] \
|
|
$trig_min_out_delay
|
|
|
|
# Treat TRIG_IO input as asynchronous.
|
|
set_false_path -from [get_ports {TRIG_IO}]
|
|
# For documentation purposes, these are the input max/min delays for TRIG_IO:
|
|
# - Input delay assuming zero trace delay and TRIG_IO buffer min switching
|
|
# time (B->A) = 0.1 ns.
|
|
# - TRIG_IO buffer max switching time (B->A = input) = 3.7 ns + same trace
|
|
# length as for output (8.479).
|
|
|
|
# Assuming no delay on external clock distribution.
|
|
|
|
# Account for the PPS min output delay only (for the case two X410 are directly
|
|
# connected to each other).
|
|
set pps_min_in_delay $trig_min_out_delay
|
|
|
|
# PPS_IN trace length DB = 0.535 + 0.133 + 0.117 + 0.061 + 2.745 inch = 3.591 inch
|
|
# PPS_IN trace length MB = 5.726 inch
|
|
# PPS switch max propagation delay = 3.6
|
|
|
|
# Assume 50% of the clock period is used for external PPS clock distribution as
|
|
# the PPS out is used to synchronize one X410 (master) with another X410
|
|
# (slave) the PPS out (trig_io) delay is added to the PPS input.
|
|
set pps_max_in_delay [expr {9.317 * 0.17 + 3.6 + 0.5 * $ref_clk_period + $trig_max_out_delay}]
|
|
|
|
# Apply PPS input constraints.
|
|
set_input_delay -clock [get_clocks ref_clk] -min $pps_min_in_delay [get_ports {PPS_IN}]
|
|
set_input_delay -clock [get_clocks ref_clk] -max $pps_max_in_delay [get_ports {PPS_IN}]
|
|
|
|
# PPS clock domain crossing BRC -> PRC on the aligned edge.
|
|
# Use a data path of half PLL reference clock period to make sure the value is
|
|
# captured without metastability.
|
|
set_max_delay -from [get_cells -hierarchical pps_delayed_brc_reg] \
|
|
-to [get_clocks pll_ref_clk*] [expr {$pll_ref_clk_period/2}]
|
|
|
|
|
|
###############################################################################
|
|
# LMK sync
|
|
###############################################################################
|
|
|
|
# The timings are derived from simulation.
|
|
|
|
# Clock Buffer ADCLK944 -> FPGA.
|
|
set buffer_to_fpga_min_clk_delay 0.997
|
|
set buffer_to_fpga_max_clk_delay 1.154
|
|
|
|
# Clock Buffer ADCLK944 -> Sample clock PLL (LMK04832).
|
|
set buffer_to_spll_min_clk_delay 0.000
|
|
set buffer_to_spll_max_clk_delay 0.014
|
|
|
|
# FPGA -> Sample clock PLL SYNC input.
|
|
set fpga_to_spll_min_clk_delay 0.381
|
|
set fpga_to_spll_max_clk_delay 0.460
|
|
|
|
# Sample clock PLL requirements.
|
|
set lmk_sync_input_hold 4.000
|
|
set lmk_sync_input_setup 4.000
|
|
|
|
set lmk_sync_output_max_delay [expr {$fpga_to_spll_max_clk_delay + $buffer_to_fpga_max_clk_delay + \
|
|
$lmk_sync_input_setup - $buffer_to_spll_min_clk_delay}]
|
|
set lmk_sync_output_min_delay [expr {$fpga_to_spll_min_clk_delay + $buffer_to_fpga_min_clk_delay - \
|
|
$buffer_to_spll_max_clk_delay - $lmk_sync_input_hold}]
|
|
set_output_delay -clock ref_clk -max $lmk_sync_output_max_delay [get_ports {LMK_SYNC}]
|
|
set_output_delay -clock ref_clk -min $lmk_sync_output_min_delay [get_ports {LMK_SYNC}]
|
|
|
|
|
|
###############################################################################
|
|
# x4xx_ps_rfdc_bd
|
|
###############################################################################
|
|
|
|
# The calibration_muxes component contains a clock crossing from some GPIO
|
|
# component instances that are synchronous to a configuration clock and ending
|
|
# in some AXI registers synchronous to data clock. The GPIO registers are
|
|
# essentially constant. When they are changing (due to a register write), the
|
|
# latching registers can definitely become metastable, so the software must
|
|
# ensure that the corrupted data appears at a safe time.
|
|
set gpio_regs [get_pins -of [get_cells -filter {IS_SEQUENTIAL && NAME =~ *rfdc/calibration_muxes/axi_gpio*} -hier] -filter {IS_CLOCK}]
|
|
set mux_regs [get_cells -hier -filter {IS_SEQUENTIAL && NAME =~ *rfdc/calibration_muxes/gpio_to_axis_mux*}]
|
|
set_false_path -from $gpio_regs -to $mux_regs
|
|
|
|
|
|
###############################################################################
|
|
# Misc Constraints
|
|
###############################################################################
|
|
|
|
# Double synchronizer false paths.
|
|
set_false_path -to [get_pins -hierarchical -filter {NAME =~ */synchronizer_false_path/stages[0].value_reg[0][*]/D}]
|
|
set_false_path -to [get_pins -hierarchical -filter {NAME =~ */rf_reset_controller*/*_ms_reg/D}]
|
|
set_false_path -to [get_pins -hierarchical -filter {NAME =~ */rfdc/rf_nco_reset_0/*_ms*/D}]
|
|
|
|
# GTY_RCV_CLK_* is driven by a OBUFDS_GTE4 buffer, which has an asynchronous
|
|
# clock-enable pin.
|
|
# By experimentation, it was observed that explicitly setting a false_path to
|
|
# this pin improved timing.
|
|
set gty_rcv_clk_buff_ceb [get_pins -of_objects [get_cells -of_objects [all_fanin -flat -startpoints_only [get_ports {GTY_RCV_CLK_P}]]] -filter {NAME=~ "*CEB"}]
|
|
set_false_path -from [get_clocks {clk40}] -through $gty_rcv_clk_buff_ceb
|
|
|
|
|
|
###############################################################################
|
|
# Asynchronous / misc. I/O constraints
|
|
# Loosely constrain these to prevent warnings in Vivado.
|
|
# Using set_input_delay associates the I/Os to a clock group, but
|
|
# set_(min|max)_delay overwrites the setup/hold analysis values.
|
|
###############################################################################
|
|
|
|
set async_inputs [get_ports {FPGA_AUX_REF}]
|
|
|
|
set_input_delay -clock [get_clocks async_in_clk] 0.000 $async_inputs
|
|
set_max_delay -from $async_inputs 50.000
|
|
set_min_delay -from $async_inputs 0.000
|
|
|
|
|
|
set async_outputs [get_ports {FABRIC_CLK_OUT_P PPS_LED}]
|
|
|
|
set_output_delay -clock [get_clocks async_out_clk] 0.000 $async_outputs
|
|
set_max_delay -to $async_outputs 50.000
|
|
set_min_delay -to $async_outputs 0.000
|