[PATCH] D114818: [libunwind] Fix unwind_leaffunction test

[Leonard Chan via Phabricator via llvm-commits]

leonardchan added a comment.

In D114818#3162689 <https://reviews.llvm.org/D114818#3162689>, @ldionne wrote:

> In D114818#3162675 <https://reviews.llvm.org/D114818#3162675>, @vvereschaka wrote:
>
>> Hi @ldionne, @leonardchan
>>
>> I ran these changes on Linux (Ubuntu 18.04)/Aarch64 board. The test gets failed because of missed 'main':
>>
>>   ubuntu at jetson8:/tmp$ ./t.tmp.exe 
>>   info.dli_sname: _Z14signal_handleri
>>   info.dli_sname: __kernel_rt_sigreturn
>>   info.dli_sname: _Z18crashing_leaf_funcv
>>   info.dli_sname: __libc_start_main

This is what I'm running into also.

> Do we need to apply `__attribute__((noinline))` to `main`? Can you try that out and let me know if that helps?

Locally `__attribute__((noinline))` on main doesn't seem to change anything for me. I checked the unwind info and made sure it's emitted correctly. The CIE for `crashing_leaf_func` is

  [0x202918] CIE length=20
    version: 1
    augmentation: zR
    code_alignment_factor: 1
    data_alignment_factor: -4
    return_address_register: 30
  
    Program:
  DW_CFA_def_cfa: reg31 +0
  DW_CFA_nop:
  DW_CFA_nop:
  DW_CFA_nop:
  DW_CFA_nop:

and it's FDE is

  [0x202970] FDE length=20 cie=[0x202918]
    initial_location: 0x2134ac
    address_range: 0x8 (end : 0x2134b4)
  
    Program:
  DW_CFA_nop:
  DW_CFA_nop:
  DW_CFA_nop:
  DW_CFA_nop:
  DW_CFA_nop:
  DW_CFA_nop:
  DW_CFA_nop:

which I think just indicates that libunwind should use x30 for finding the return address. If I stop right before the `__builtin_trap()`, it looks like x30 points back into main (right after the call to `crashing_leaf_func`):

  x30            0x2134fc 2176252
  ...
  (gdb) info symbol 0x2134fc
  main + 60 in section .text of /home/leonardchan/llvm-monorepo/llvm-build-1-master-fuchsia-toolchain/runtimes/runtimes-aarch64-unknown-linux-gnu-bins/Output/unwind_leaffunction.pass.cpp.dir/t.tmp.exe
  ...
  Leonard Chan, 9 min
  (gdb) disas 0x2134fc
  Dump of assembler code for function main:
  ...
     0x00000000002134f8 <+56>:    bl      0x2134ac <_Z18crashing_leaf_funcv>
     0x00000000002134fc <+60>:    mov     w0, #0xfffffffe                 // #-2
  ...

One thing I also noticed is that if we add another function call, something like:

  void crashing_leaf_func() {
    __builtin_trap();
  }
  
  void func() {
    crashing_leaf_func();
  }
  
  int main() {
    signal(SIGTRAP, signal_handler);
    signal(SIGILL, signal_handler);
    func();
    return 2;
  }

Then the test passes, but it looks like unwinder skips past `func` (it jumps from `_Z18crashing_leaf_funcv` to `main`). Perhaps all this hints at an underlying issue in libunwind?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D114818/new/

https://reviews.llvm.org/D114818