Segfault with certain typedef names

[s-holst]

Consider this verilog module:

module m;
    typedef enum { FALSE, TRUE } u;
    //typedef u [0:1] yeah;  // "yeah" works
    typedef u [0:1] crash;  // "crash" segfaults
endmodule

This crashes synlig in systemverilog_plugin::synlig_simplify and this crash seems to depend on the name of the second typedef.

Here is the crash in GDB:

$ gdb /cad/synlig/bin/synlig
GNU gdb (Ubuntu 12.1-0ubuntu1~22.04.2) 12.1
Copyright (C) 2022 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.
Type "show copying" and "show warranty" for details.
This GDB was configured as "x86_64-linux-gnu".
Type "show configuration" for configuration details.
For bug reporting instructions, please see:
<https://www.gnu.org/software/gdb/bugs/>.
Find the GDB manual and other documentation resources online at:
    <http://www.gnu.org/software/gdb/documentation/>.

For help, type "help".
Type "apropos word" to search for commands related to "word"...
Reading symbols from /cad/synlig/bin/synlig...
(No debugging symbols found in /cad/synlig/bin/synlig)
(gdb) r
Starting program: /cad/synlig/bin/synlig 
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".

synlig> read_systemverilog -debug check.sv


1. Executing SystemVerilog frontend.
[INF:CM0023] Creating log file "/.../slpp_all/surelog.log".
[INF:CP0300] Compilation...
[INF:CP0303] /.../check.sv:1:1: Compile module "work@m".
[INF:CP0302] Compile class "work@mailbox".
[INF:CP0302] Compile class "work@process".
[INF:CP0302] Compile class "work@semaphore".
[INF:EL0526] Design Elaboration...
[NTE:EL0503] /.../check.sv:1:1: Top level module "work@m".
[NTE:EL0508] Nb Top level modules: 1.
[NTE:EL0509] Max instance depth: 1.
[NTE:EL0510] Nb instances: 1.
[NTE:EL0511] Nb leaf instances: 1.
[INF:UH0706] Creating UHDM Model...
[INF:UH0707] Elaborating UHDM...
[  FATAL] : 0
[ SYNTAX] : 0
[  ERROR] : 0
[WARNING] : 0
[   NOTE] : 5
Object 'work@m' of type 'design'
  Object 'builtin' of type 'package'
  Object 'work@m' of type 'module_inst'
    Object 'crash' of type 'packed_array_typespec'
      Object '' of type 'range'
        Object '' of type 'constant'
        Object '' of type 'constant'
      Object 'u' of type 'enum_typespec'
        Object 'FALSE' of type 'enum_const'
        Object 'TRUE' of type 'enum_const'
    Object 'u' of type 'enum_typespec'
      Object 'FALSE' of type 'enum_const'
      Object 'TRUE' of type 'enum_const'

Program received signal SIGSEGV, Segmentation fault.
0x00005555577b4200 in systemverilog_plugin::synlig_simplify(Yosys::AST::AstNode*, bool, bool, bool, int, int, bool, bool) ()
(gdb) 

I didn't manage to build synlig with debug symbols yet.
If you need more info, please let me know.

[kamilrakoczy]

Thanks for the report.

We are always building synlig with debug symbols, but we are stripping them on installation.

It looks like the problem here is related to order in which typedefs are parsed.
This order is based on name of the typedef.
So, when you have typedef with name u that is used in typedef named yeah, it works ok, as first we are parsing u and then yeah.
In second case, first we are parsing crash and then u. In this case, we recursively executing synlig_simplify until we are out of stack.

Are you willing to investigate it further to find out the root cause for it?

[s-holst]

Thanks. I got debug symbols now and can confirm a infinite recursion here:

synlig/src/mods/yosys_ast/synlig_simplify.cc

Line 1647 in dab70c6

while (synlig_simplify(template_node, const_fold, at_zero, in_lvalue, stage, width_hint, sign_hint, in_param)) {

I can try to look into that, but I'm new to this code and don't know how long it will take (also busy with lots of other things). So, no promises ;)

[s-holst]

Some observations from the debugger:
First crash gets resolved, then the u as dependency.
This operation involves convert_packed_unpacked_range (some sort of multi range workaround) that in turn calls resolve_wiretype that calls synlig_simplify. Everything works until this point.
Then, u gets processed again. Now convert_packed_unpacked_range -> resolve_wiretype -> synlig_simplify, and synlig_simplify is calling itself indefinitely.

I'm too unfamiliar with this code to get any deeper insight or come up with a fix with reasonable effort by myself.

Any help is very welcome

[kamilrakoczy]

Hi, @s-holst thanks for looking into it.

It turned out, that we were always overriding typedef name when processing packed enums (but we should do it only in case of anonymous enums). We fixed this in: #2670

While doing it, we noticed different issue: accessing packed enums doesn't work as expected, e.g.:

typedef enum { FALSE, TRUE } u;
typedef u [0:1] yeah;
yeah x;
assign x[0] = TRUE;
assign x[1] = FALSE;  

will be translated to:

logic [63:0] x;
assign x[1:0] = TRUE;
assign x[3:2] = FALSE;

but it should be changed to:

logic [63:0] x;
assign x[31:0] = TRUE;
assign x[63:32] = FALSE;

as a temporary workaround, you can wrap typedef definition into struct, such access works ok:

typedef enum { FALSE, TRUE } u;
typedef struct packed {
	u [1:0] e;
} yeah;

yeah x;
assign x.e[0] = TRUE;
assign x.e[1] = FALSE;

[alaindargelas]

@s-holst this is supported here: alainmarcel/uhdm2rtlil#175