downconn fw: update csr & add doc

testing tx: add buffer to improve timing
testing tx: send idle more frequently
downconn GW: remove unneeded stb signal
downconn GW: add docs

src/gateware/cxp.py

@@ -165,11 +165,6 @@ class DownConn_Interface(Module, AutoCSR):
         ]
 
 
-        # DEBUG: tx loopback fifo control
-        self.tx_stb = CSRStorage()
-        self.sync += phy.tx_stb_sys.eq(self.tx_stb.storage) 
-   
-
         # DEBUG: Transmission Pipeline
         # 
         #  rtio pak  ----+
@@ -183,20 +178,14 @@ class DownConn_Interface(Module, AutoCSR):
 
         # DEBUG: TX pipeline
         self.submodules.bootstrap_loopback = bootstrap_loopback = TX_Bootstrap()
-        self.submodules.mux = mux = stream.Multiplexer(word_layout, 2)
         self.submodules.pak_wrp = pak_wrp = Packet_Wrapper()
         self.submodules.trig_ack = trig_ack = Trigger_ACK_Inserter()
 
         self.ack = CSR()
-        self.mux_sel = CSRStorage()
 
         self.sync += trig_ack.stb.eq(self.ack.re),
-        self.comb += [
-            bootstrap_loopback.source.connect(mux.sink0),
-            mux.sel.eq(self.mux_sel.storage),
-        ]
 
-        tx_pipeline = [mux , pak_wrp, trig_ack, phy]
+        tx_pipeline = [bootstrap_loopback, pak_wrp, trig_ack, phy]
         for s, d in zip(tx_pipeline, tx_pipeline[1:]):
             self.comb += s.source.connect(d.sink)
 
@@ -205,17 +194,20 @@ class DownConn_Interface(Module, AutoCSR):
 
         # Receiver Pipeline WIP
         # 
-        #        32                32
-        # PHY ---/---> CDC FIFO ---/---> trigger ack ------> packet  ------> debug buffer
-        #                                checker             decoder
+        #        32             32+8(dchar)
+        # PHY ---/---> dchar   -----/-----> trigger ack ------>  packet  ------> CDC FIFO ------> debug buffer
+        #              decoder              checker              decoder
         #
         cdr = ClockDomainsRenamer("cxp_gtx_rx")
 
+        # decode all incoming data as duplicate char and inject the result into the bus for downstream modules
+        self.submodules.dchar_decoder = dchar_decoder = cdr(Duplicated_Char_Decoder())
+
         # Priority level 1 packet - Trigger ack packet
-        self.submodules.trig_ack_checker = trig_ack_checker = cdr(CXP_Trig_Ack_Checker())
+        self.submodules.trig_ack_checker = trig_ack_checker = cdr(Trigger_Ack_Checker())
         
         self.submodules.trig_ack_ps = trig_ack_ps = PulseSynchronizer("cxp_gtx_rx", "sys")
-        self.comb += trig_ack_ps.i.eq(trig_ack_checker.ack)
+        self.sync.cxp_gtx_rx += trig_ack_ps.i.eq(trig_ack_checker.ack)
 
         self.trig_ack = Signal()
         self.trig_clr = Signal()
@@ -239,7 +231,7 @@ class DownConn_Interface(Module, AutoCSR):
         test_err_ps = PulseSynchronizer("cxp_gtx_rx", "sys")
         buffer_err_ps = PulseSynchronizer("cxp_gtx_rx", "sys")
         self.submodules += decode_err_ps, test_err_ps, buffer_err_ps
-        self.comb += [
+        self.sync.cxp_gtx_rx += [
             decode_err_ps.i.eq(bootstrap.decode_err),
             test_err_ps.i.eq(bootstrap.test_err),
             buffer_err_ps.i.eq(bootstrap.buffer_err),
@@ -282,23 +274,11 @@ class DownConn_Interface(Module, AutoCSR):
         ]
         
 
-        # DEBUG: remove this cdc fifo
-        cdc_fifo = stream.AsyncFIFO(word_layout, 512)
+        cdc_fifo = stream.AsyncFIFO(word_layout_dchar, 512)
         self.submodules += ClockDomainsRenamer({"write": "cxp_gtx_rx", "read": "sys"})(cdc_fifo)
         self.submodules.debug_out = debug_out = RX_Debug_Buffer()
 
-        self.dmux_sel = CSRStorage()
-        self.submodules.dmux = dmux = stream.Demultiplexer(word_layout, 2)
-
-
-        self.comb += [
-            dmux.source0.connect(bootstrap.sink),
-            dmux.source1.connect(cdc_fifo.sink),
-            cdc_fifo.source.connect(debug_out.sink),
-            dmux.sel.eq(self.dmux_sel.storage),
-        ]
-
-        rx_pipeline = [phy, trig_ack_checker, dmux]
+        rx_pipeline = [phy, dchar_decoder, trig_ack_checker, bootstrap, cdc_fifo, debug_out]
         for s, d in zip(rx_pipeline, rx_pipeline[1:]):
             self.comb += s.source.connect(d.sink)
         
@@ -319,8 +299,8 @@ class DownConn_Interface(Module, AutoCSR):
             # Instance("OBUF", i_I=phy.gtx.cd_cxp_gtx_rx.clk, o_O=debug_sma.p_tx),
             # Instance("OBUF", i_I=, o_O=debug_sma.p_rx),
             # # pmod 0-7 pin
-            Instance("OBUF", i_I=bootstrap.test_err, o_O=pmod_pads[0]),
-            Instance("OBUF", i_I=pak_start, o_O=pmod_pads[1]),
+            # Instance("OBUF", i_I=bootstrap.test_err, o_O=pmod_pads[0]),
+            # Instance("OBUF", i_I=pak_start, o_O=pmod_pads[1]),
             # Instance("OBUF", i_I=fifo_in.source.ack, o_O=pmod_pads[2]),
             # Instance("OBUF", i_I=gtx.comma_checker.aligner_en, o_O=pmod_pads[3]),
             # Instance("OBUF", i_I=gtx.comma_checker.check_reset, o_O=pmod_pads[4]),

src/gateware/cxp_downconn.py

@@ -56,9 +56,12 @@ class Receiver(Module):
        
         self.source = stream.Endpoint(word_layout)
 
+        data_valid = Signal()
         self.sync.cxp_gtx_rx += [
+            data_valid.eq(gtx.comma_checker.rxfsm.ongoing("READY")),
+
             self.source.stb.eq(0),
-            If(gtx.rx_ready & self.source.ack & ~((gtx.decoders[0].d == 0xBC) & (gtx.decoders[0].k == 1)),
+            If(data_valid & self.source.ack & ~((gtx.decoders[0].d == 0xBC) & (gtx.decoders[0].k == 1)),
                 self.source.stb.eq(1),
                 self.source.data.eq(Cat(gtx.decoders[i].d for i in range(4))),
                 self.source.k.eq(Cat(gtx.decoders[i].k for i in range(4))),
@@ -66,17 +69,22 @@ class Receiver(Module):
         ]
 
         # DEBUG: tx fifos for loopback
-        # fw -> fifo (sys) -> cdc fifo -> gtx tx
+        # fw -> -> cdc fifo -> buffered fifo -> gtx tx
         
-        tx_fifo = stream.AsyncFIFO(word_layout, 512)
-        self.submodules += ClockDomainsRenamer({"write": "sys", "read": "cxp_gtx_tx"})(tx_fifo)
-        self.sink = tx_fifo.sink
-    
-        self.tx_stb_sys = Signal()
-        txstb = Signal()
-        self.specials += MultiReg(self.tx_stb_sys, txstb, odomain="cxp_gtx_tx")
+        cdc_fifo = stream.AsyncFIFO(word_layout, 512)
+        self.submodules += ClockDomainsRenamer({"write": "sys", "read": "cxp_gtx_tx"})(cdc_fifo)
+        self.sink = cdc_fifo.sink
 
-        word_count = Signal(max=100)
+        # fix timing violation
+        cdr = ClockDomainsRenamer("cxp_gtx_tx")
+        self.submodules.buf = tx_fifo = cdr(stream.SyncFIFO(word_layout, 2, buffered=True))
+
+        self.comb += [
+            cdc_fifo.source.connect(tx_fifo.sink),
+        ]
+    
+        idle_period = 50 # press in word
+        word_count = Signal(max=idle_period)
 
         # JANK: fix the every 98th word got eaten
         # cnt      97    98   99       0
@@ -85,11 +93,11 @@ class Receiver(Module):
         self.sync.cxp_gtx_tx += [
             tx_fifo.source.ack.eq(0),
 
-            If(word_count == 99,
+            If(word_count == idle_period-1,
                 word_count.eq(word_count.reset),
             ).Else(
-                If(tx_fifo.source.stb & txstb,
-                    If(word_count != 98, tx_fifo.source.ack.eq(1)),
+                If(tx_fifo.source.stb,
+                    If(word_count != idle_period-2, tx_fifo.source.ack.eq(1)),
                     word_count.eq(word_count + 1),
                 )
             )
@@ -97,7 +105,7 @@ class Receiver(Module):
 
         # NOTE: prevent the first word send twice due to stream stb delay 
         self.comb += [
-            If((tx_fifo.source.stb & tx_fifo.source.ack & (word_count != 99)),
+            If((tx_fifo.source.stb & tx_fifo.source.ack & (word_count != idle_period-1)),
                 gtx.encoder.d[0].eq(tx_fifo.source.data[:8]),
                 gtx.encoder.d[1].eq(tx_fifo.source.data[8:16]),
                 gtx.encoder.d[2].eq(tx_fifo.source.data[16:24]),

src/gateware/cxp_pipeline.py

@@ -4,12 +4,23 @@ from misoc.interconnect.csr import *
 from misoc.interconnect import stream
 from misoc.cores.liteeth_mini.mac.crc import LiteEthMACCRCEngine, LiteEthMACCRCChecker
 
+from functools import reduce
+from itertools import combinations
+from operator import or_, and_
+
 char_width = 8
 char_layout = [("data", char_width), ("k", char_width//8)]
 
 word_dw = 32
 word_layout = [("data", word_dw), ("k", word_dw//8)]
 
+word_layout_dchar = [
+    ("data", word_dw),
+    ("k", word_dw//8), 
+    ("dchar", char_width), 
+    ("dchar_k", char_width//8)
+]
+
 buffer_count = 4
 buffer_depth = 512
 
@@ -261,7 +272,7 @@ class TX_Bootstrap(Module, AutoCSR):
 
 class RX_Debug_Buffer(Module,AutoCSR):
     def __init__(self):
-        self.submodules.buf_out = buf_out = stream.SyncFIFO(word_layout, 128)
+        self.submodules.buf_out = buf_out = stream.SyncFIFO(word_layout_dchar, 128)
         self.sink = buf_out.sink
 
         self.inc = CSR()
@@ -277,21 +288,64 @@ class RX_Debug_Buffer(Module,AutoCSR):
             self.dout_valid.status.eq(buf_out.source.stb),
         ]
 
-class Duplicate_Majority_Voter(Module):
-    def __init__(self, char_4x, k_4x):
-        assert char_4x.nbits == 32
-        assert k_4x.nbits == 4
-        
-        # Section 9.2.2.1 (CXP-001-2021)
-        # decoder should immune to single bit errors when handling duplicated characters
-        self.char = Signal(char_width)
-        self.k = Signal()
+class Duplicated_Char_Decoder(Module):
+    def __init__(self):
+        self.sink = stream.Endpoint(word_layout)
+        self.buffer = stream.Endpoint(word_layout)
+        self.source = stream.Endpoint(word_layout_dchar)
 
-        a, b, c, d = [char_4x[i*8:(i+1)*8] for i in range(4)]
-        a_k, b_k, c_k, d_k = [k_4x[i:(i+1)] for i in range(4)]
-        self.comb += [
-            self.char.eq(a&b&c | a&b&d | a&c&d | b&c&d),
-            self.k.eq(a_k&b_k&c_k | a_k&b_k&d_k | a_k&c_k&d_k | b_k&c_k&d_k),
+        # # #
+
+
+        # For duplicated characters, an error correction method (e.g. majority voting) is required to meet the CXP spec:
+        # RX decoder should immune to single bit errors when handling duplicated characters - Section 9.2.2.1 (CXP-001-2021)
+        # 
+        #
+        #                                 32
+        #             +--->  buffer  -----/-----+
+        #         32  |                         |            32+8(dchar)
+        # sink ---/---+                         ---> source -----/-----> downstream
+        #             |                8(dchar) |                        decoders
+        #             +---> majority -----/-----+
+        #                   voting
+        # 
+        #
+        # Due to the tight setup/hold time requiremnt for 12.5Gbps CXP, the voting logic cannot be implemented as combinational logic
+        # Hence, a pipeline approach is needed to avoid any s/h violation, where the majority voting result are pre-calculate and injected into the bus immediate after the PHY.
+        # And any downstream modules can access the voting result without implementing the voting logic inside the decoder
+
+        self.sync += [
+            self.sink.ack.eq(self.buffer.ack),
+            self.buffer.stb.eq(self.sink.stb),
+            If(self.sink.stb,
+                self.buffer.data.eq(self.sink.data),
+                self.buffer.k.eq(self.sink.k),
+            ),
+        ]
+
+        # cycle 1 - calculate ABC, ABD, ACD & BCD  
+        char = [[self.sink.data[i*8:(i+1)*8], self.sink.k[i]] for i in range(4)]
+        voters = [Record([("data", 8), ("k", 1)]) for _ in range(4)]
+
+        for i, comb in enumerate(combinations(char, 3)):
+            self.sync += [
+                If(self.sink.stb,
+                    voters[i].data.eq(reduce(and_, [code[0] for code in comb])),
+                    voters[i].k.eq(reduce(and_, [code[1] for code in comb])),
+                )
+            ]
+
+
+        # cycle 2 - inject the voting result 
+        self.sync += [
+            self.buffer.ack.eq(self.source.ack),
+            self.source.stb.eq(self.buffer.stb),
+            If(self.buffer.stb,
+                self.source.data.eq(self.buffer.data),
+                self.source.k.eq(self.buffer.k),
+                self.source.dchar.eq(Replicate(reduce(or_, [v.data for v in voters]), 4)),
+                self.source.dchar_k.eq(Replicate(reduce(or_, [v.k for v in voters]), 4)),
+            ),
         ]
 
 @FullMemoryWE()
@@ -314,15 +368,14 @@ class RX_Bootstrap(Module):
             "heartbeat": 0x09,
         }
 
-        self.sink = stream.Endpoint(word_layout)
-        self.source = stream.Endpoint(word_layout)
+        self.sink = stream.Endpoint(word_layout_dchar)
+        self.source = stream.Endpoint(word_layout_dchar)
 
         self.submodules.fsm = fsm = FSM(reset_state="IDLE")
-        self.submodules.voter = voter = Duplicate_Majority_Voter(self.sink.data, self.sink.k)
 
         fsm.act("IDLE",
             self.sink.ack.eq(1),
-            If((self.sink.stb & (voter.char == KCode["pak_start"]) & (voter.k == 1)),
+            If((self.sink.stb & (self.sink.dchar == KCode["pak_start"]) & (self.sink.dchar_k == 1)),
                 NextState("DECODE"),
             )
         )
@@ -333,9 +386,9 @@ class RX_Bootstrap(Module):
         fsm.act("DECODE",
             self.sink.ack.eq(1),
             If(self.sink.stb,
-                NextValue(self.packet_type, voter.char),
+                NextValue(self.packet_type, self.sink.dchar),
 
-                Case(voter.char, {
+                Case(self.sink.dchar, {
                     type["data_stream"]: NextState("STREAMING"),
                     type["test_packet"]: [
                         NextValue(cnt, cnt.reset),
@@ -363,7 +416,7 @@ class RX_Bootstrap(Module):
         )
         # For stream data packet
         fsm.act("STREAMING",
-            If((self.sink.stb & (voter.char == KCode["pak_end"]) & (voter.k == 1)),
+            If((self.sink.stb & (self.sink.dchar == KCode["pak_end"]) & (self.sink.dchar_k == 1)),
                 # discard K29,7
                 self.sink.ack.eq(1),
                 NextState("IDLE")
@@ -378,7 +431,7 @@ class RX_Bootstrap(Module):
         fsm.act("VERIFY_TEST_PATTERN",
             self.sink.ack.eq(1),
             If(self.sink.stb,
-                If(((voter.char == KCode["pak_end"]) & (voter.k == 1)),
+                If(((self.sink.dchar == KCode["pak_end"]) & (self.sink.dchar_k == 1)),
                     NextState("IDLE"),
                 ).Else(
                     If(((self.sink.data != Cat(cnt, cnt+1, cnt+2, cnt+3))),  
@@ -405,7 +458,6 @@ class RX_Bootstrap(Module):
         self.comb += [
             mem_port.adr[:addr_nbits].eq(addr),
             mem_port.adr[addr_nbits:].eq(write_ptr),
-            mem_port.dat_w.eq(self.sink.data),
         ]
 
         # For control ack, event packet
@@ -413,10 +465,11 @@ class RX_Bootstrap(Module):
             mem_port.we.eq(0),
             self.sink.ack.eq(1),
             If(self.sink.stb,
-                If(((voter.char == KCode["pak_end"]) & (voter.k == 1)),
+                If(((self.sink.dchar == KCode["pak_end"]) & (self.sink.dchar_k == 1)),
                     NextState("MOVE_BUFFER_PTR"),
                 ).Else(
                     mem_port.we.eq(1),
+                    mem_port.dat_w.eq(self.sink.data),
                     NextValue(addr, addr + 1),
                     If(addr == buffer_depth - 1,
                         # discard the packet
@@ -439,10 +492,10 @@ class RX_Bootstrap(Module):
             NextState("IDLE"),
         )
 
-class CXP_Trig_Ack_Checker(Module, AutoCSR):
+class Trigger_Ack_Checker(Module, AutoCSR):
     def __init__(self):
-        self.sink = stream.Endpoint(word_layout)
-        self.source = stream.Endpoint(word_layout)
+        self.sink = stream.Endpoint(word_layout_dchar)
+        self.source = stream.Endpoint(word_layout_dchar)
 
         self.ack = Signal()
 
@@ -450,10 +503,8 @@ class CXP_Trig_Ack_Checker(Module, AutoCSR):
 
         self.submodules.fsm = fsm = FSM(reset_state="COPY")
 
-        self.submodules.voter = voter = Duplicate_Majority_Voter(self.sink.data, self.sink.k)
-
         fsm.act("COPY",
-            If((self.sink.stb & (voter.char == KCode["io_ack"]) & (voter.k == 1)),
+            If((self.sink.stb & (self.sink.dchar == KCode["io_ack"]) & (self.sink.dchar_k == 1)),
                 # discard K28,6
                 self.sink.ack.eq(1),
                 NextState("CHECK_ACK")
@@ -467,7 +518,7 @@ class CXP_Trig_Ack_Checker(Module, AutoCSR):
                 NextState("COPY"),
                 # discard the word after K28,6
                 self.sink.ack.eq(1),
-                If((voter.char == 0x01) & (voter.k == 0),
+                If((self.sink.dchar == 0x01) & (self.sink.dchar_k == 0),
                     self.ack.eq(1),
                 )
             )

src/libboard_artiq/src/cxp_downconn.rs

@@ -24,9 +24,9 @@ pub fn loopback_testing(channel: usize, timer: &mut GlobalTimer, speed: cxp_phys
         while (CXP[channel].downconn_rx_ready_read)() != 1 {}
         info!("rx ready!");
 
-        (CXP[channel].downconn_tx_stb_write)(1);
         cxp_proto::downconn_send_test_packet(channel);
 
+        // FIXME: why test + trig ack doesn't work well for rx??
         cxp_proto::downconn_debug_send_trig_ack(channel);
 
         const DATA_MAXSIZE: usize = 253;
@@ -51,17 +51,22 @@ pub fn loopback_testing(channel: usize, timer: &mut GlobalTimer, speed: cxp_phys
         .expect("loopback gtx tx error");
 
         timer.delay_us(1000); // wait packet has arrive at RX async fifo
-        (CXP[channel].downconn_tx_stb_write)(0);
 
-        info!("trig ack = {}", (CXP[channel].downconn_trigger_ack_read)());
-        (CXP[channel].downconn_trigger_ack_write)(1);
-        info!("after clr trig ack = {}", (CXP[channel].downconn_trigger_ack_read)());
+        if (CXP[channel].downconn_trigger_ack_read)() == 1 {
+            (CXP[channel].downconn_trigger_ack_write)(1);
+            info!("trig ack and cleared");
+        }
+
+        if (CXP[channel].downconn_bootstrap_decoder_err_read)() == 1 {
+            info!("!!!!!!!DECODER ERROR!!!!!!! and cleared");
+            (CXP[channel].downconn_bootstrap_decoder_err_write)(1);
+        }
+
+        if (CXP[channel].downconn_bootstrap_test_err_read)() == 1 {
+            info!("!!!!!!!TEST ERROR!!!!!!! and cleared");
+            (CXP[channel].downconn_bootstrap_test_err_write)(1);
+        }
 
-        info!(
-            "decoder error = {}",
-            (CXP[channel].downconn_bootstrap_decoder_err_read)()
-        );
-        info!("test error = {}", (CXP[channel].downconn_bootstrap_test_err_read)());
         info!("packet type = {:#06X}", (CXP[channel].downconn_packet_type_read)());
 
         cxp_proto::receive(channel).expect("loopback gtx rx error");
@@ -82,6 +87,8 @@ pub fn loopback_testing(channel: usize, timer: &mut GlobalTimer, speed: cxp_phys
                 break;
             }
         }
-        cxp_proto::print_packetu32(&pak_arr, &k_arr);
+
+        info!("rx ready = {}", (CXP[channel].downconn_rx_ready_read)());
+        // cxp_proto::print_packetu32(&pak_arr, &k_arr);
     }
 }