Re: symtab/1253: [regression] bad backtrace for '<function called from gdb>'

[Michael Elizabeth Chastain <mec at shout dot net>]

The following reply was made to PR symtab/1253; it has been noted by GNATS.

From: Michael Elizabeth Chastain <mec@shout.net>
To: drow@mvista.com
Cc: gdb-gnats@sources.redhat.com
Subject: Re: symtab/1253: [regression] bad backtrace for '<function called from gdb>'
Date: Fri, 1 Aug 2003 23:32:59 -0400

 Okay some more clue droppings ...
 
 The problem is localized inside "sleep".  If I set a breakpoint
 on "sleep" and then "stepi" through it, gdb has some funky ideas
 of where the stack is.
 
 The code for "sleep" is:
 
   0x420ae310 <sleep>:     push   %ebp
   0x420ae311 <sleep+1>:   xor    %ecx,%ecx
   0x420ae313 <sleep+3>:   mov    %esp,%ebp
   0x420ae315 <sleep+5>:   push   %edi
   0x420ae316 <sleep+6>:   xor    %edx,%edx
   0x420ae318 <sleep+8>:   lea    0xfffffe50(%ebp),%edi
   0x420ae31e <sleep+14>:  push   %esi
   0x420ae31f <sleep+15>:  push   %ebx
   0x420ae320 <sleep+16>:  call   0x4201575d <__i686.get_pc_thunk.bx>
   0x420ae325 <sleep+21>:  add    $0x7bfab,%ebx
   0x420ae32b <sleep+27>:  sub    $0x1cc,%esp
   0x420ae331 <sleep+33>:  mov    %ecx,0xfffffe54(%ebp)
   0x420ae337 <sleep+39>:  mov    0x8(%ebp),%eax
   0x420ae33a <sleep+42>:  test   %eax,%eax
   0x420ae33c <sleep+44>:  mov    %eax,0xfffffe50(%ebp)
   0x420ae342 <sleep+50>:  je     0x420ae45c <sleep+332>
 
 Note that the stack does not get adjusted until <sleep+27>!
 
 This is PIC-related.
 
 When I link with -static, I get a non-PIC version of sleep:
 
   0x0804eb1c <sleep+0>:   push   %ebp
   0x0804eb1d <sleep+1>:   mov    %esp,%ebp
   0x0804eb1f <sleep+3>:   push   %edi
   0x0804eb20 <sleep+4>:   push   %esi
   0x0804eb21 <sleep+5>:   push   %ebx
   0x0804eb22 <sleep+6>:   sub    $0x1ac,%esp
   0x0804eb28 <sleep+12>:  mov    0x8(%ebp),%eax
   0x0804eb2b <sleep+15>:  xor    %edx,%edx
   0x0804eb2d <sleep+17>:  test   %eax,%eax
   0x0804eb2f <sleep+19>:  mov    %eax,0xfffffe50(%ebp)
   0x0804eb35 <sleep+25>:  lea    0xfffffe50(%ebp),%esi
   0x0804eb3b <sleep+31>:  movl   $0x0,0xfffffe54(%ebp)
   0x0804eb45 <sleep+41>:  je     0x804ec13 <sleep+247>
 
 And then the backtrace is all good:
 
   (gdb) bt
   #0  subroutine (in=14) at mi-syn-frame.c:53
   #1  0x0804826e in handler (sig=14) at mi-syn-frame.c:43
   #2  <signal handler called>
   #3  0x0804eca1 in nanosleep ()
   #4  0x0804ebf3 in sleep ()
   #5  0x080482cb in have_a_very_merry_interrupt () at mi-syn-frame.c:64
   #6  <function called from gdb>
   #7  main () at mi-syn-frame.c:14
 
 So I think the problem really is: backtracing through a function
 which is compiled PIC, so that it has a more elaborate prolog.
 It might be important that the function is in a shared library
 and doesn't have debug info.
 
 I'll dig some more.
 
 Michael C