[PATCH] D75326: [entry values] X86: Describe effects of MOV{8,16}ri (PR45053)

[David Stenberg via Phabricator via llvm-commits]

dstenb added a comment.

In D75326#1898465 <https://reviews.llvm.org/D75326#1898465>, @vsk wrote:

> If OP_piece in an entry value really /is/ an issue, please confirm, and I'll update the x86 describeLoadedValue to just return 'None' for mov{8,16}ri (it doesn't seem worth it to me to do anything more complicated).


I applied this patch on top of 5900d3f2e94f710d73a89931953ce0a3d928c70d <https://reviews.llvm.org/rG5900d3f2e94f710d73a89931953ce0a3d928c70d>. I was unable to get the IR reproducer listed here up and running, but I tried it out with the following example:

caller.c:

  extern int call(int);
  int main() {
    call(0x11223344);
    return 0;
  }

callee.c:

  int global;
  int call(int param) {
    global = param; // side effect to keep param.
    asm __volatile("movl $0xdeadbeef, %%edi" : : : "rdi"); // clobber reg.
    return 0; // print param here.
  }

in which I changed the parameter setup to:

  $edi = MOV32ri 287454020, debug-location !15 ; 0x11223344
  $di = MOV16ri 21862, debug-location !15 ; 0x5566                                                                                                                                                                                                                                                                           
  CALL64pcrel32 @call, csr_64, implicit $rsp, implicit $ssp, implicit killed $edi, implicit-def $rsp, implicit-def $ssp, implicit-def dead $eax, debug-location !15 

producing the following call site parameter entry:

  0x00000050:       DW_TAG_GNU_call_site_parameter
                      DW_AT_location      (DW_OP_reg5 RDI)
                      DW_AT_GNU_call_site_value   (DW_OP_constu 0x5566, DW_OP_stack_value, DW_OP_piece 0x2)

and the following location list for the parameter in the callee:

  DW_AT_location        (0x00000000: 
     [0x0000000000201110, 0x000000000020111b): DW_OP_reg5 RDI
     [0x000000000020111b, 0x000000000020111e): DW_OP_GNU_entry_value(DW_OP_reg5 RDI), DW_OP_stack_value)

When I print that in GDB 8.2.1, `0x5566` is printed instead of `0x11225566` which is the actual parameter value:

  (gdb) print /x param
  $1 = 0x5566
  (gdb) disas
  Dump of assembler code for function call:
     0x0000000000201110 <+0>:	mov    %edi,0x2eea(%rip)        # 0x204000 <global>
     0x0000000000201116 <+6>:	mov    $0xdeadbeef,%edi
  => 0x000000000020111b <+11>:	xor    %eax,%eax
     0x000000000020111d <+13>:	retq   
  End of assembler dump.

With LLDB trunk (5900d3f2e94f710d73a89931953ce0a3d928c70d <https://reviews.llvm.org/rG5900d3f2e94f710d73a89931953ce0a3d928c70d>), the clobbered register value (0xdeadbeef) is printed, even though the address is covered by an entry value (compare the PC to the location list printed above):

  (lldb) print param
  (int) $0 = 0xdeadbeef
  (lldb) disas
  a.out`call:
      0x201110 <+0>:  movl   %edi, 0x2eea(%rip)        ; global
      0x201116 <+6>:  movl   $0xdeadbeef, %edi         ; imm = 0xDEADBEEF 
  ->  0x20111b <+11>: xorl   %eax, %eax
      0x20111d <+13>: retq   

I'm not very familiar with LLDB (especially when using entry values), so sorry if I have overlooked something and that is only a user error!


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D75326/new/

https://reviews.llvm.org/D75326