[Bug target/126938] New: [17 Regression] ICE in s390_function_arg_advance since r17-3213

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126938

            Bug ID: 126938
           Summary: [17 Regression] ICE in s390_function_arg_advance since
                    r17-3213
           Product: gcc
           Version: 17.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: stefansf at gcc dot gnu.org
                CC: jakub at redhat dot com
  Target Milestone: ---
            Target: s390x-*

Starting with r17-3213 the following tests fail on s390:

FAIL: g++.dg/reflect/crash18.C  -std=c++26 (internal compiler error: in
s390_function_arg_advance, at config/s390/s390.cc:13290)
FAIL: g++.dg/reflect/crash18.C  -std=c++26 (test for excess errors)
FAIL: g++.dg/reflect/crash18.C  -std=c++29 (internal compiler error: in
s390_function_arg_advance, at config/s390/s390.cc:13290)
FAIL: g++.dg/reflect/crash18.C  -std=c++29 (test for excess errors)
FAIL: g++.dg/reflect/parm3.C  -std=c++26 (internal compiler error: in
s390_function_arg_advance, at config/s390/s390.cc:13290)
FAIL: g++.dg/reflect/parm3.C  -std=c++26 (test for excess errors)
FAIL: g++.dg/reflect/parm3.C  -std=c++29 (internal compiler error: in
s390_function_arg_advance, at config/s390/s390.cc:13290)
FAIL: g++.dg/reflect/parm3.C  -std=c++29 (test for excess errors)

In s390_function_arg_advance we are now faced with

(gdb) call debug (arg.type)
 <meta_type 0x3fff76d5738 decltype(^^int) unsigned DI
    size <integer_cst 0x3fff7502ea0 type <integer_type 0x3fff751b0a8
bitsizetype> constant 64>
    unit-size <integer_cst 0x3fff7502eb8 type <integer_type 0x3fff751b000
sizetype> constant 8>
    align:64 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0x3fff76d5738>

This is because now in gimplify_parameters() we do not skip the meta_type
anymore because passed_mode != VOIDmode:

  FOR_EACH_VEC_ELT (fnargs, i, parm)
    {
      struct assign_parm_data_one data;

      /* Extract the type of PARM; adjust it according to ABI.  */
      assign_parm_find_data_types (&all, parm, &data);

      /* Early out for errors and void parameters.  */
      if (data.passed_mode == VOIDmode || DECL_SIZE (parm) == NULL)
        continue;

      /* Update info on where next arg arrives in registers.  */
      targetm.calls.function_arg_advance (all.args_so_far, data.arg);

I'm wondering what the contract here is.  An error is already thrown by the FE,
though, still we proceed.  Should the back end deal with meta_type?