Getting an error while trying to implement a simple P4 PNA program.

[rupesh-chiluka-marvell]

Program:

#include <core.p4>
#include <bmv2/pna.p4>

typedef bit<48>  EthernetAddress;
// const PortId_t dest_port = (PortId_t) 1;

header ethernet_t {
    EthernetAddress dstAddr;
    EthernetAddress srcAddr;
    bit<16>         etherType;
}

struct metadata_t {
}

struct headers_t {
    ethernet_t eth;
}

parser MainParserImpl(
    packet_in pkt,
    out   headers_t  hdr,
    inout metadata_t meta,
    in    pna_main_parser_input_metadata_t istd)
{
    state start {
        pkt.extract(hdr.eth);
        transition accept;
    }
}

control MainControlImpl(
    inout headers_t  hdr,
    inout metadata_t meta,
    in    pna_main_input_metadata_t  istd,
    inout pna_main_output_metadata_t ostd)
{
    action dummy() {
        // do nothing
    }

    table dummy_table {
        key = {}
        actions = { dummy; }
        default_action = dummy;
    }

    action send_packet() {
        // do nothing
    }

    table send_packet_table {
        key = {
            hdr.eth.dstAddr: exact;
        }
        actions = { send_packet; }
        default_action = send_packet;
    }

    apply {
        bool is_recirculated = istd.recirculated;
        dummy_table.apply();
        if (is_recirculated) {
            send_packet_table.apply();
        }
    }
}

control MainDeparserImpl(
    packet_out pkt,
    inout headers_t hdr,
    in    metadata_t meta,
    in    pna_main_output_metadata_t ostd)
{
    apply {
        pkt.emit(hdr.eth);
    }
}

PNA_NIC(
    MainParserImpl(),
    MainControlImpl(),
    MainDeparserImpl()
    ) main;

sudo gdb --args pna_nic  --interface 0@veth0 --interface 1@veth2 --interface 2@veth4 pna-sample-program.json  --l
og-console

Ouptut:

[18:07:04.786] [bmv2] [D] [thread 609629] [0.0] [cxt 0] Processing packet received on port 0
[18:07:04.786] [bmv2] [D] [thread 609629] [0.0] [cxt 0] Parser 'main_parser': start
[18:07:04.786] [bmv2] [D] [thread 609629] [0.0] [cxt 0] Parser 'main_parser' entering state 'start'
[18:07:04.786] [bmv2] [D] [thread 609629] [0.0] [cxt 0] Extracting header 'eth'
[18:07:04.786] [bmv2] [D] [thread 609629] [0.0] [cxt 0] Parser state 'start' has no switch, going to default next state
[18:07:04.786] [bmv2] [T] [thread 609629] [0.0] [cxt 0] Bytes parsed: 14
[18:07:04.786] [bmv2] [D] [thread 609629] [0.0] [cxt 0] Parser 'main_parser': end
[18:07:04.786] [bmv2] [D] [thread 609629] [0.0] [cxt 0] Pipeline 'main_control': start
[18:07:04.786] [bmv2] [T] [thread 609629] [0.0] [cxt 0] Applying table 'MainControlImpl.dummy_table'
[18:07:04.786] [bmv2] [D] [thread 609629] [0.0] [cxt 0] Looking up key:

[18:07:04.786] [bmv2] [D] [thread 609629] [0.0] [cxt 0] Table 'MainControlImpl.dummy_table': miss
[18:07:04.786] [bmv2] [D] [thread 609629] [0.0] [cxt 0] Action entry is MainControlImpl.dummy - 
[18:07:04.786] [bmv2] [T] [thread 609629] [0.0] [cxt 0] Action MainControlImpl.dummy

Thread 7 "pna_nic" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fffef7fe640 (LWP 609629)]
0x00005555556633aa in std::_Bit_reference::operator bool (this=<synthetic pointer>) at /usr/include/c++/11/bits/stl_bvector.h:87
87          { return !!(*_M_p & _M_mask); }

(gdb) bt
#0  0x00005555556633aa in std::_Bit_reference::operator bool (this=<synthetic pointer>) at /usr/include/c++/11/bits/stl_bvector.h:87
p4lang/behavioral-model#1  bm::ExpressionTemps::pop_bool (this=<optimized out>) at expressions.cpp:425
p4lang/behavioral-model#2  bm::Expression::eval_bool (this=this@entry=0x5555557cf658, phv=..., locals=std::vector of length 0, capacity 0) at expressions.cpp:832
p4lang/behavioral-model#3  0x000055555564373a in bm::Conditional::eval (phv=..., this=0x5555557cf620) at ../../include/bm/bm_sim/conditionals.h:44
p4lang/behavioral-model#4  bm::Conditional::operator() (this=0x5555557cf620, pkt=0x7fffe0000e00) at conditionals.cpp:38
p4lang/behavioral-model#5  0x00005555556ba999 in bm::Pipeline::apply (this=this@entry=0x5555557dc160, pkt=0x7fffe0000e00) at pipeline.cpp:44
p4lang/behavioral-model#6  0x00005555555f1463 in bm::pna::PnaNic::main_thread (this=0x5555557ca2e0) at /usr/include/c++/11/bits/unique_ptr.h:173
p4lang/behavioral-model#7  0x00007ffff72dc253 in ?? () from /lib/x86_64-linux-gnu/libstdc++.so.6
p4lang/behavioral-model#8  0x00007ffff6e94ac3 in start_thread (arg=<optimized out>) at ./nptl/pthread_create.c:442
p4lang/behavioral-model#9  0x00007ffff6f26850 in clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:81

[jafingerhut]

Rupesh, is this a new PNA BMv2 test program you are working on, hoping to modify the code so that it works? If so, great!

Are you looking for help on why this error occurs, or do you already know the root cause?

[rupesh-chiluka-marvell]

Rupesh, is this a new PNA BMv2 test program you are working on, hoping to modify the code so that it works? If so, great!

Yes

Are you looking for help on why this error occurs, or do you already know the root cause?

Looking for the help

[jafingerhut]

Is the JSON file output by the compiler something anyone should be able to reproduce from the source code and latest version of p4c source code? And the error occurs with latest behavioral-model code published? Or does the JSON file require local changes to p4c that only you have, and/or the error only occur with locally modified behavioral-model code? In either case, attaching the file pna-sample-program.json to a comment might make it easier for someone else to reproduce the issue.

[antoninbas]

In either case, attaching the file pna-sample-program.json to a comment might make it easier for someone else to reproduce the issue.

Not to mention that there is a pretty strong chance that this is caused by an invalid / unsupported JSON file.

[rupesh-chiluka-marvell]

Is the JSON file output by the compiler something anyone should be able to reproduce from the source code and latest version of p4c source code?

Yes

And the error occurs with latest behavioral-model code published?

Yes

Or does the JSON file require local changes to p4c that only you have, and/or the error only occur with locally modified behavioral-model code?

No local changes are made in this case

[rupesh-chiluka-marvell]

JSON file:
pna_example_program.json

[antoninbas]

The expression for the conditional present in the JSON is not supported:

      "conditionals" : [
        {
          "name" : "node_3",
          "id" : 0,
          "source_info" : {
            "filename" : "pna-demo-recirculation.p4",
            "line" : 96,
            "column" : 12,
            "source_fragment" : "is_recirculated"
          },
          "expression" : {
            "type" : "field",
            "value" : ["pna_main_input_metadata", "recirculated"]
          },
          "false_next" : null,
          "true_next" : "MainControlImpl.send_packet_table"
        }
      ]

The pna_main_input_metadata.recirculated field cannot be used directly as a boolean value. There are 2 possibilities: either the expression is changed to use an explicit comparison (pna_main_input_metadata.recirculated == 1) or the d2b (data-to-bool conversion) unary operator is used.

This seems to be an issue in the PNA p4c backend, as I would expect such a simple program to work well for simple_switch / v1model. You can probably write the same program for v1model, and check what the JSON looks like (and make sure it is accepted by simple_switch).

I think this issue should be transferred to p4c.

[rupesh-chiluka-marvell]

(pna_main_input_metadata.recirculated == (bool) 1) didn't work.

This seems to be an issue in the PNA p4c backend, as I would expect such a simple program to work well for simple_switch / v1model.

Yes.

I think this issue should be transferred to p4c.

Ok

[antoninbas]

(pna_main_input_metadata.recirculated == (bool) 1) didn't work.

I didn't suggest modifying the P4 program, that was very unlikely to work. JSON generation needs to be fixed.

[rupesh-chiluka-marvell]

Yes, after changing to explicit conversion in the JSON, it is working fine.

"expression" : {
  "type" : "expression",
  "value" : {
    "op" : "==",
    "left" : {
      "type" : "field",
      "value" : ["pna_main_input_metadata", "recirculated"]
    },
    "right" : {
      "type" : "hexstr",
      "value" : 1
    }
  }
},

[rupesh-chiluka-marvell]

I am having the same problem with the below PSA program, and it is working after changing the JSON as per the above suggestion.

psa-example-program.p4.txt

[jafingerhut]

The PSA back end for BMv2 is probably in a similarly unfinished state as the PNA back end for BMv2. There are many, many small simple PSA programs that either do not compile using p4c-bm2-psa, or they compile but do not execute correctly. I would recommend_not_ using the PSA back end for BMv2 as a reference for how to do things.

The v1model back end for BMv2 is by (I believe) a wide margin the most mature, featureful, and well tested open source p4c back end.

[jafingerhut]

I guess now that the Tofino back end is open source, it might beat the v1model back end now, by those measures. I would still recommend referring to how the v1model back end does things when trying to learn reasonable ways to work with BMv2 as the target.

[rupesh-chiluka-marvell]

Ok. Thanks everyone for very quick response :)

[rupesh-chiluka-marvell]

Should I close this issue or leave it open till the bug is fixed?

[jafingerhut]

I think it is fine to leave the issue open until the bug is fixed.

[fruffy]

@rupesh-chiluka-marvell Are you interested in implementing a fix for this in the PNA back end?

[rupesh-chiluka-marvell]

Yes. I would like to 👍.

[fruffy]

Great! Feel free to make a PR at any time.