tofino: add nopaxos tofino P4 code and controller

experiments/pyexps/nopaxos.py

@@ -25,7 +25,7 @@ import simbricks.simulators as sim
 import simbricks.nodeconfig as node
 
 host_configs = ['qemu', 'gt']
-seq_configs = ['swseq', 'ehseq']
+seq_configs = ['swseq', 'ehseq', 'tofino']
 nic_configs = ['ib', 'cb', 'cv']
 proto_configs = ['vr', 'nopaxos']
 num_client_configs = [1, 2, 3, 4, 5, 6, 7, 8, 10, 12]
@@ -41,7 +41,10 @@ for proto_config in proto_configs:
             for seq_config in seq_configs:
                 for nic_config in nic_configs:
                     e = exp.Experiment(proto_config + '-'  + host_config + '-' + nic_config + '-' + seq_config + f'-{num_c}')
-                    net = sim.NS3SequencerNet()
+                    if seq_config == 'tofino':
+                        net = sim.TofinoNet()
+                    else:
+                        net = sim.NS3SequencerNet()
                     net.sync_period = sync_period
                     net.opt = link_rate_opt + link_latency_opt
                     e.add_network(net)

sims/tofino/p4/nopaxos.p4

+/* -*- P4_16 -*- */
+
+#include <core.p4>
+#include <tna.p4>
+
+typedef bit<48> mac_addr_t;
+typedef bit<32> ipv4_addr_t;
+
+typedef bit<16> sess_num_t;
+typedef bit<64> msg_num_t;
+typedef bit<32> reg_val_t;
+typedef bit<8> reg_key_t;
+
+#define ETHERTYPE_TPID    0x8100
+#define ETHERTYPE_IPV4  0x0800
+#define ETHERTYPE_PKTGEN 16w0x7777
+#define IP_PROTOCOL_UDP 17
+#define IP_PROTOCOL_TCP 6
+#define MCAST_DST_PORT 22222
+
+const int MAC_TABLE_SIZE        = 65536;
+const bit<3> L2_LEARN_DIGEST = 1;
+
+/*************************************************************************
+ ***********************  H E A D E R S  *********************************
+ *************************************************************************/
+
+/*  Define all the headers the program will recognize             */
+/*  The actual sets of headers processed by each gress can differ */
+
+/* Standard ethernet header */
+header ethernet_h {
+    bit<48>   dst_addr;
+    bit<48>   src_addr;
+    bit<16>   ether_type;
+}
+
+header vlan_tag_h {
+    bit<3>   pcp;
+    bit<1>   cfi;
+    bit<12>  vid;
+    bit<16>  ether_type;
+}
+
+header ipv4_h {
+    bit<4>   version;
+    bit<4>   ihl;
+    bit<8>   diffserv;
+    bit<16>  total_len;
+    bit<16>  identification;
+    bit<3>   flags;
+    bit<13>  frag_offset;
+    bit<8>   ttl;
+    bit<8>   protocol;
+    bit<16>  hdr_checksum;
+    bit<32>  src_addr;
+    bit<32>  dst_addr;
+}
+
+header udp_h {
+    bit<16>  src_port;
+    bit<16>  dst_port;
+    bit<16>  len;
+    bit<16>  checksum;
+}
+
+header nopaxos_h {
+    bit<32>         is_frag;
+    bit<16>         header_size;
+    // meta_data from here
+    sess_num_t      sess_num;
+    msg_num_t       msg_num;
+}
+
+/******  G L O B A L   I N G R E S S   M E T A D A T A  *********/
+
+struct my_ingress_headers_t {
+    ethernet_h  ethernet;
+    ipv4_h      ipv4;
+    udp_h       udp;
+    nopaxos_h   nopaxos;
+}
+
+struct my_ingress_metadata_t {
+    bit<9> mac_move;
+    bit<1> is_static;
+    bit<1> smac_hit;
+    PortId_t ingress_port;
+}
+
+/***********************  P A R S E R  **************************/
+parser IngressParser(
+    packet_in pkt,
+    out my_ingress_headers_t hdr,
+    out my_ingress_metadata_t meta,
+    out ingress_intrinsic_metadata_t ig_intr_md){
+
+    /* This is a mandatory state, required by Tofino Architecture */
+    state start {
+        pkt.extract(ig_intr_md);
+        pkt.advance(PORT_METADATA_SIZE);
+        transition meta_init;
+    }
+
+    state meta_init {
+        meta.mac_move = 0;
+        meta.is_static = 0;
+        meta.smac_hit = 0;
+        meta.ingress_port = ig_intr_md.ingress_port;
+        transition parse_ethernet;
+    }
+
+    state parse_ethernet {
+        pkt.extract(hdr.ethernet);
+        transition select(hdr.ethernet.ether_type){
+            ETHERTYPE_IPV4: parse_ipv4;
+            default: accept;
+        }
+    }
+
+    state parse_ipv4 {
+        pkt.extract(hdr.ipv4);
+        transition select(hdr.ipv4.protocol){
+            IP_PROTOCOL_UDP: parse_udp;
+            default: accept;
+        }
+    }
+
+    state parse_udp {
+        pkt.extract(hdr.udp);
+        transition select(hdr.udp.dst_port) {
+            MCAST_DST_PORT: parse_nopaxos;
+            default: accept;
+        }
+    }
+
+    state parse_nopaxos {
+        pkt.extract(hdr.nopaxos);
+        transition accept;
+    }
+}
+
+/***************** M A T C H - A C T I O N  *********************/
+
+control Ingress(
+    /* User */
+    inout my_ingress_headers_t                       hdr,
+    inout my_ingress_metadata_t                      meta,
+    /* Intrinsic */
+    in    ingress_intrinsic_metadata_t               ig_intr_md,
+    in    ingress_intrinsic_metadata_from_parser_t   ig_prsr_md,
+    inout ingress_intrinsic_metadata_for_deparser_t  ig_dprsr_md,
+    inout ingress_intrinsic_metadata_for_tm_t        ig_tm_md)
+{
+    Register<reg_val_t, reg_key_t>(1) reg_cnt; // value, key
+    RegisterAction<reg_val_t, reg_key_t, reg_val_t>(reg_cnt)
+    sequencing = {
+        void apply(inout reg_val_t register_data, out reg_val_t seq_num) {
+            register_data = register_data + 1;
+            seq_num = register_data;
+        }
+    };
+
+    action send(PortId_t port) {
+        ig_tm_md.ucast_egress_port = port;
+    }
+
+    action drop() {
+        ig_dprsr_md.drop_ctl = 1;
+    }
+
+    action smac_hit(PortId_t port, bit<1> is_static) {
+        meta.mac_move  = ig_intr_md.ingress_port ^ port;
+        meta.smac_hit  = 1;
+        meta.is_static = is_static;
+    }
+
+    action smac_miss() { }
+
+    action smac_drop() {
+        drop(); exit;
+    }
+
+    @idletime_precision(3)
+    table smac {
+        key = {
+            hdr.ethernet.src_addr : exact;
+        }
+        actions = {
+            smac_hit; smac_miss; smac_drop;
+        }
+        size                 = MAC_TABLE_SIZE;
+        const default_action = smac_miss();
+        idle_timeout         = true;
+    }
+
+    action mac_learn_notify() {
+        ig_dprsr_md.digest_type = L2_LEARN_DIGEST;
+    }
+
+    table smac_results {
+        key = {
+            meta.mac_move  : ternary;
+            meta.is_static : ternary;
+            meta.smac_hit  : ternary;
+        }
+        actions = {
+            mac_learn_notify; NoAction; smac_drop;
+        }
+        const entries = {
+            ( _, _, 0) : mac_learn_notify();
+            ( 0, _, 1) : NoAction();
+            ( _, 0, 1) : mac_learn_notify();
+            ( _, 1, 1) : smac_drop();
+        }
+    }
+
+    action dmac_unicast(PortId_t port) {
+        send(port);
+    }
+
+    action dmac_miss() {
+        ig_tm_md.mcast_grp_a = 1;
+    }
+
+    action dmac_drop() {
+        drop();
+        exit;
+    }
+
+    table dmac {
+        key = {
+            hdr.ethernet.dst_addr : exact;
+        }
+        actions = {
+            dmac_unicast; dmac_miss; dmac_drop;
+        }
+        size           = MAC_TABLE_SIZE;
+        default_action = dmac_miss();
+    }
+
+    apply {
+        ig_tm_md.bypass_egress = 1w1;
+
+        smac.apply();
+        smac_results.apply();
+
+        if(hdr.udp.isValid() && hdr.udp.dst_port == MCAST_DST_PORT) {
+            hdr.udp.checksum = 0;
+            hdr.nopaxos.msg_num = (msg_num_t)sequencing.execute(0);
+            hdr.nopaxos.sess_num = (sess_num_t)0;
+            ig_tm_md.mcast_grp_a = 2;
+        } else {
+            switch (dmac.apply().action_run) {
+                dmac_unicast: { /* Unicast source pruning */
+                    if (ig_intr_md.ingress_port == ig_tm_md.ucast_egress_port) {
+                      drop();
+                    }
+                }
+            }
+        }
+    }
+
+}  // End of SwitchIngressControl
+
+/*********************  D E P A R S E R  ************************/
+
+/* This struct is needed for proper digest receive API generation */
+struct l2_digest_t {
+    bit<48> src_mac;
+    bit<9>  ingress_port;
+    bit<9>  mac_move;
+    bit<1>  is_static;
+    bit<1>  smac_hit;
+}
+
+control IngressDeparser(packet_out pkt,
+    /* User */
+    inout my_ingress_headers_t                       hdr,
+    in    my_ingress_metadata_t                      meta,
+    /* Intrinsic */
+    in    ingress_intrinsic_metadata_for_deparser_t  ig_dprsr_md)
+{
+    Digest <l2_digest_t>() l2_digest;
+
+    apply {
+        if (ig_dprsr_md.digest_type == L2_LEARN_DIGEST) {
+            l2_digest.pack({
+                    hdr.ethernet.src_addr,
+                    meta.ingress_port,
+                    meta.mac_move,
+                    meta.is_static,
+                    meta.smac_hit });
+        }
+
+        pkt.emit(hdr);
+    }
+}
+
+
+/*************************************************************************
+ ****************  E G R E S S   P R O C E S S I N G   *******************
+ *************************************************************************/
+
+/***********************  H E A D E R S  ************************/
+
+struct my_egress_headers_t {
+    ethernet_h   ethernet;
+    vlan_tag_h   vlan_tag;
+    ipv4_h       ipv4;
+}
+
+/********  G L O B A L   E G R E S S   M E T A D A T A  *********/
+
+struct my_egress_metadata_t {
+}
+
+/***********************  P A R S E R  **************************/
+
+parser EgressParser(packet_in        pkt,
+    /* User */
+    out my_egress_headers_t          hdr,
+    out my_egress_metadata_t         meta,
+    /* Intrinsic */
+    out egress_intrinsic_metadata_t  eg_intr_md)
+{
+    /* This is a mandatory state, required by Tofino Architecture */
+    state start {
+        pkt.extract(eg_intr_md);
+        transition parse_ethernet;
+    }
+
+    state parse_ethernet {
+        pkt.extract(hdr.ethernet);
+        transition select(hdr.ethernet.ether_type) {
+            ETHERTYPE_TPID:  parse_vlan_tag;
+            ETHERTYPE_IPV4:  parse_ipv4;
+            default: accept;
+        }
+    }
+
+    state parse_vlan_tag {
+        pkt.extract(hdr.vlan_tag);
+        transition select(hdr.vlan_tag.ether_type) {
+            ETHERTYPE_IPV4:  parse_ipv4;
+            default: accept;
+        }
+    }
+
+    state parse_ipv4 {
+        pkt.extract(hdr.ipv4);
+        transition accept;
+    }
+}
+
+/***************** M A T C H - A C T I O N  *********************/
+
+control Egress(
+    /* User */
+    inout my_egress_headers_t                          hdr,
+    inout my_egress_metadata_t                         meta,
+    /* Intrinsic */
+    in    egress_intrinsic_metadata_t                  eg_intr_md,
+    in    egress_intrinsic_metadata_from_parser_t      eg_prsr_md,
+    inout egress_intrinsic_metadata_for_deparser_t     eg_dprsr_md,
+    inout egress_intrinsic_metadata_for_output_port_t  eg_oport_md)
+{
+    apply {
+    }
+}
+
+    /*********************  D E P A R S E R  ************************/
+
+control EgressDeparser(packet_out pkt,
+    /* User */
+    inout my_egress_headers_t                       hdr,
+    in    my_egress_metadata_t                      meta,
+    /* Intrinsic */
+    in    egress_intrinsic_metadata_for_deparser_t  eg_dprsr_md)
+{
+    apply {
+        pkt.emit(hdr);
+    }
+}
+
+/************ F I N A L   P A C K A G E ******************************/
+Pipeline(IngressParser(),
+         Ingress(),
+         IngressDeparser(),
+         EgressParser(),
+         Egress(),
+         EgressDeparser()
+         ) pipe;
+
+Switch(pipe) main;

sims/tofino/p4/nopaxos_setup.py

+from bfrtcli import *
+
+class l2_switch():
+    #
+    # Helper Functions to deal with ports
+    #
+    def devport(self, pipe, port):
+        return ((pipe & 3) << 7) | (port & 0x7F)
+    def pipeport(self,dp):
+        return ((dp & 0x180) >> 7, (dp & 0x7F))
+    def mcport(self, pipe, port):
+        return pipe * 72 + port
+    def devport_to_mcport(self, dp):
+        return self.mcport(*self.pipeport(dp))
+
+
+    # This is a useful bfrt_python function that should potentially allow one
+    # to quickly clear all the logical tables (including the fixed ones) in
+    #  their data plane program.
+    #
+    # This function  can clear all P4 tables and later other fixed objects
+    # (once proper BfRt support is added). As of SDE-9.2.0 the support is mixed.
+    # As a result the function contains some workarounds.
+    def clear_all(self, verbose=True, batching=True, clear_ports=False):
+
+        table_list = bfrt.info(return_info=True, print_info=False)
+
+        # Remove port tables from the list
+        port_types = ['PORT_CFG',      'PORT_FRONT_PANEL_IDX_INFO',
+                      'PORT_HDL_INFO', 'PORT_STR_INFO']
+
+        if not clear_ports:
+            for table in list(table_list):
+                if table['type'] in port_types:
+                    table_list.remove(table)
+
+                    # The order is important. We do want to clear from the top,
+                    # i.e. delete objects that use other objects. For example,
+                    # table entries use selector groups and selector groups
+                    # use action profile members.
+                    #
+                    # Same is true for the fixed tables. However, the list of
+                    # table types grows, so we will first clean the tables we
+                    # know and then clear the rest
+        for table_types in (['MATCH_DIRECT', 'MATCH_INDIRECT_SELECTOR'],
+                            ['SELECTOR'],
+                            ['ACTION_PROFILE'],
+                            ['PRE_MGID'],
+                            ['PRE_ECMP'],
+                            ['PRE_NODE'],
+                            []):         # This is catch-all
+            for table in list(table_list):
+                if table['type'] in table_types or len(table_types) == 0:
+                    try:
+                        if verbose:
+                            print("Clearing table {:<40} ... ".
+                                  format(table['full_name']),
+                                  end='', flush=True)
+                        table['node'].clear(batch=batching)
+                        table_list.remove(table)
+                        if verbose:
+                            print('Done')
+                        use_entry_list = False
+                    except:
+                        use_entry_list = True
+
+                    # Some tables do not support clear(). Thus we'll try
+                    # to get a list of entries and clear them one-by-one
+                    if use_entry_list:
+                        try:
+                            if batching:
+                                bfrt.batch_begin()
+
+                            # This line can result in an exception,
+                            # since # not all tables support get()
+                            entry_list = table['node'].get(regex=True,
+                                                           return_ents=True,
+                                                           print_ents=False)
+
+                            # Not every table supports delete() method.
+                            # For those tables we'll try to push in an
+                            # entry with everything being zeroed out
+                            has_delete = hasattr(table['node'], 'delete')
+
+                            if entry_list != -1:
+                                if has_delete:
+                                    for entry in entry_list:
+                                        entry.remove()
+                                else:
+                                    clear_entry = table['node'].entry()
+                                    for entry in entry_list:
+                                        entry.data = clear_entry.data
+                                        # We can still have an exception
+                                        # here, since not all tables
+                                        # support add()/mod()
+                                        entry.push()
+                                if verbose:
+                                    print('Done')
+                            else:
+                                print('Empty')
+                            table_list.remove(table)
+
+                        except BfRtTableError as e:
+                            print('Empty')
+                            table_list.remove(table)
+
+                        except Exception as e:
+                            # We can have in a number of ways: no get(),
+                            # no add() etc. Another reason is that the
+                            # table is read-only.
+                            if verbose:
+                                print("Failed")
+                        finally:
+                            if batching:
+                                bfrt.batch_end()
+        bfrt.complete_operations()
+
+    def __init__(self, default_ttl=60000):
+        self.p4 = bfrt.nopaxos.pipe
+        self.all_ports  = [port.key[b'$DEV_PORT']
+                           for port in bfrt.port.port.get(regex=1,
+                                                          return_ents=True,
+                                                          print_ents=False)]
+        self.l2_age_ttl = default_ttl
+
+    def setup(self):
+        self.clear_all()
+        self.__init__()
+
+        # Enable learning on SMAC
+        print("Initializing learning on SMAC ... ", end='', flush=True)
+        try:
+            self.p4.IngressDeparser.l2_digest.callback_deregister()
+        except:
+            pass
+        self.p4.IngressDeparser.l2_digest.callback_register(self.learning_cb)
+        print("Done")
+
+        # Enable aging on SMAC
+        print("Inializing Aging on SMAC ... ", end='', flush=True)
+        self.p4.Ingress.smac.idle_table_set_notify(enable=False,
+                                                   callback=None)
+
+        self.p4.Ingress.smac.idle_table_set_notify(enable=True,
+                                                   callback=self.aging_cb,
+                                                   interval = 10000,
+                                                   min_ttl  = 10000,
+                                                   max_ttl  = 60000)
+        print("Done")
+
+    @staticmethod
+    def aging_cb(dev_id, pipe_id, direction, parser_id, entry):
+        smac = bfrt.nopaxos.pipe.Ingress.smac
+        dmac = bfrt.nopaxos.pipe.Ingress.dmac
+
+        mac_addr = entry.key[b'hdr.ethernet.src_addr']
+
+        print("Aging out: MAC: {}".format(mac(mac_addr)))
+
+        entry.remove() # from smac
+        try:
+            dmac.delete(dst_addr=mac_addr)
+        except:
+            print("WARNING: Could not find the matching DMAC entry")
+
+    @staticmethod
+    def learning_cb(dev_id, pipe_id, direction, parser_id, session, msg):
+        smac = bfrt.nopaxos.pipe.Ingress.smac
+        dmac = bfrt.nopaxos.pipe.Ingress.dmac
+
+        for digest in msg:
+            port     = digest["ingress_port"]
+            mac_move = digest["mac_move"]
+            mac_addr  = digest["src_mac"]
+
+            old_port = port ^ mac_move # Because mac_move = ingress_port ^ port
+
+            print("MAC: {},  Port={}".format(
+                mac(mac_addr), port), end="")
+
+            if mac_move != 0:
+                print("(Move from port={})".format(old_port))
+            else:
+                print("(New)")
+
+            # Since we do not have access to self, we have to use
+            # the hardcoded value for the TTL :(
+            smac.entry_with_smac_hit(src_addr=mac_addr,
+                                     port=port,
+                                     is_static=False,
+                                     ENTRY_TTL=60000).push()
+            dmac.entry_with_dmac_unicast(dst_addr=mac_addr,
+                                         port=port).push()
+        return 0
+
+def set_mcast(num_groups=1, num_sequencers=1):
+    all_ports = [p for p in range(8)]
+    mcast_ports = [0, 1, 2]
+    # Broadcast
+    bfrt.pre.node.entry(MULTICAST_NODE_ID=0, MULTICAST_RID=0,
+        MULTICAST_LAG_ID=[], DEV_PORT=all_ports).push()
+    bfrt.pre.mgid.entry(MGID=1, MULTICAST_NODE_ID=[0],
+            MULTICAST_NODE_L1_XID_VALID=[False],
+                MULTICAST_NODE_L1_XID=[0]).push()
+    # Multicast
+    bfrt.pre.node.entry(MULTICAST_NODE_ID=1, MULTICAST_RID=1,
+        MULTICAST_LAG_ID=[], DEV_PORT=mcast_ports).push()
+    bfrt.pre.mgid.entry(MGID=2, MULTICAST_NODE_ID=[1],
+        MULTICAST_NODE_L1_XID_VALID=[False],
+            MULTICAST_NODE_L1_XID=[0]).push()
+
+
+### Setup L2 learning
+sl2 = l2_switch(default_ttl=10000)
+sl2.setup()
+bfrt.complete_operations()
+
+p4 = bfrt.nopaxos.pipe
+num_groups = 1
+num_sequencers = 1
+
+set_mcast(num_groups, num_sequencers)
+
+### Register initialization
+p4.Ingress.reg_cnt.mod(0, 0)
+bfrt.complete_operations()
+
+### Register print out
+print("""******************* SETUP RESULTS *****************""")
+print ("\n reg_cnt:")
+p4.Ingress.reg_cnt.get(REGISTER_INDEX=0, from_hw=True)