[RFA] Pb stepping inside DLL function

[RFA] Pb stepping inside DLL function

[Joel Brobecker]

Hello,

We have the following little C program (a.c):

        #include <stdio.h>
        
        extern _stdcall int sub ();
        
        main ()
        {
          int a = 0;
          
          a = sub ();
          printf ("%d\n", a);
        }

``sub()'' is a function provided by a DLL. It has all the necessary
debug information. The following transcript shows that GDB is currently
unable to step into sub():

        (gdb) start
        Breakpoint 1 at 0x4012d2: file a.c, line 7.
        Starting program: /[...]/a.exe 
        main () at a.c:7
        7         int a = 0;
        (gdb) step
        9         a = sub ();
        (gdb) p a
        $1 = 0
        (gdb) step
        10        printf ("%d\n", a);
        (gdb) print a
        $2 = 5

During the step, GDB lands inside sub@0, which is the trampoline
for sub(), but doesn't realize it. The fix is to teach GDB how to
recognize them, and hand to find where they will eventually take us.
For that, I found a function that was actually dead but did exactly
what I needed, so I reused it.

2004-06-23  Joel Brobecker  <brobecker@gnat.com>

        * i386-cygwin-tdep.c (i386-cygwin-tdep.c): New function.
        (i386_cygwin_in_solib_call_trampoline): New function.
        (i386_cygwin_init_abi): Initialize the in_solib_call_trampoline
        and skip_trampoline_code gdbarch methods.

Tested on x86-windows (XP), no regression.
OK to apply?

Thanks,
-- 
Joel

Re: [RFA] Pb stepping inside DLL function

[Joel Brobecker]

[-- Attachment #1: Type: text/plain, Size: 1482 bytes --]

[Sigh... With the patch, this time...]

Hello,

We have the following little C program (a.c):

        #include <stdio.h>
        
        extern _stdcall int sub ();
        
        main ()
        {
          int a = 0;
          
          a = sub ();
          printf ("%d\n", a);
        }

``sub()'' is a function provided by a DLL. It has all the necessary
debug information. The following transcript shows that GDB is currently
unable to step into sub():

        (gdb) start
        Breakpoint 1 at 0x4012d2: file a.c, line 7.
        Starting program: /[...]/a.exe 
        main () at a.c:7
        7         int a = 0;
        (gdb) step
        9         a = sub ();
        (gdb) p a
        $1 = 0
        (gdb) step
        10        printf ("%d\n", a);
        (gdb) print a
        $2 = 5

During the step, GDB lands inside sub@0, which is the trampoline
for sub(), but doesn't realize it. The fix is to teach GDB how to
recognize them, and hand to find where they will eventually take us.
For that, I found a function that was actually dead but did exactly
what I needed, so I reused it.

2004-06-23  Joel Brobecker  <brobecker@gnat.com>

        * i386-cygwin-tdep.c (i386-cygwin-tdep.c): New function.
        (i386_cygwin_in_solib_call_trampoline): New function.
        (i386_cygwin_init_abi): Initialize the in_solib_call_trampoline
        and skip_trampoline_code gdbarch methods.

Tested on x86-windows (XP), no regression.
OK to apply?

Thanks,
-- 
Joel

[-- Attachment #2: step-dll.diff --]
[-- Type: text/plain, Size: 1032 bytes --]

Index: i386-cygwin-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/i386-cygwin-tdep.c,v
retrieving revision 1.5
diff -u -p -r1.5 i386-cygwin-tdep.c
--- i386-cygwin-tdep.c	30 Apr 2004 21:13:58 -0000	1.5
+++ i386-cygwin-tdep.c	24 Jun 2004 00:12:25 -0000
@@ -26,11 +26,27 @@
 
 #include "i386-tdep.h"
 
+static CORE_ADDR
+i386_cygwin_skip_trampoline_code (CORE_ADDR pc)
+{
+  return i386_pe_skip_trampoline_code (pc, NULL);
+}
+
+static int
+i386_cygwin_in_solib_call_trampoline (CORE_ADDR pc, char *name)
+{
+  return (i386_pe_skip_trampoline_code (pc, name) != 0);
+}
+
 static void
 i386_cygwin_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
 {
   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
 
+  set_gdbarch_in_solib_call_trampoline (gdbarch,
+                                        i386_cygwin_in_solib_call_trampoline);
+  set_gdbarch_skip_trampoline_code (gdbarch, i386_cygwin_skip_trampoline_code);
+
   tdep->struct_return = reg_struct_return;
 }
 

Re: [RFA] Pb stepping inside DLL function

[Christopher Faylor]

On Wed, Jun 23, 2004 at 05:23:05PM -0700, Joel Brobecker wrote:
>2004-06-23  Joel Brobecker  <brobecker@gnat.com>
>
>        * i386-cygwin-tdep.c (i386-cygwin-tdep.c): New function.
>        (i386_cygwin_in_solib_call_trampoline): New function.
>        (i386_cygwin_init_abi): Initialize the in_solib_call_trampoline
>        and skip_trampoline_code gdbarch methods.
>
>Tested on x86-windows (XP), no regression.
>OK to apply?

Sorry for the delay, Joel.

Yes, please apply.

Thanks.

cgf

Re: [RFA] Pb stepping inside DLL function

[Joel Brobecker]

> On Wed, Jun 23, 2004 at 05:23:05PM -0700, Joel Brobecker wrote:
> >2004-06-23  Joel Brobecker  <brobecker@gnat.com>
> >
> >        * i386-cygwin-tdep.c (i386-cygwin-tdep.c): New function.
> >        (i386_cygwin_in_solib_call_trampoline): New function.
> >        (i386_cygwin_init_abi): Initialize the in_solib_call_trampoline
> >        and skip_trampoline_code gdbarch methods.
> >
> >Tested on x86-windows (XP), no regression.
> >OK to apply?
> 
> Sorry for the delay, Joel.
> 
> Yes, please apply.

Thank you, applied. (and no worries for the delay, a delay of a week
or even two weeks is perfectly acceptable to me)

-- 
Joel

Re: [RFA] Pb stepping inside DLL function

[Michael Snyder]

Joel, did you forget to attach the patch?   ;-)

Joel Brobecker wrote:
> Hello,
> 
> We have the following little C program (a.c):
> 
>         #include <stdio.h>
>         
>         extern _stdcall int sub ();
>         
>         main ()
>         {
>           int a = 0;
>           
>           a = sub ();
>           printf ("%d\n", a);
>         }
> 
> ``sub()'' is a function provided by a DLL. It has all the necessary
> debug information. The following transcript shows that GDB is currently
> unable to step into sub():
> 
>         (gdb) start
>         Breakpoint 1 at 0x4012d2: file a.c, line 7.
>         Starting program: /[...]/a.exe 
>         main () at a.c:7
>         7         int a = 0;
>         (gdb) step
>         9         a = sub ();
>         (gdb) p a
>         $1 = 0
>         (gdb) step
>         10        printf ("%d\n", a);
>         (gdb) print a
>         $2 = 5
> 
> During the step, GDB lands inside sub@0, which is the trampoline
> for sub(), but doesn't realize it. The fix is to teach GDB how to
> recognize them, and hand to find where they will eventually take us.
> For that, I found a function that was actually dead but did exactly
> what I needed, so I reused it.
> 
> 2004-06-23  Joel Brobecker  <brobecker@gnat.com>
> 
>         * i386-cygwin-tdep.c (i386-cygwin-tdep.c): New function.
>         (i386_cygwin_in_solib_call_trampoline): New function.
>         (i386_cygwin_init_abi): Initialize the in_solib_call_trampoline
>         and skip_trampoline_code gdbarch methods.
> 
> Tested on x86-windows (XP), no regression.
> OK to apply?
> 
> Thanks,