[Patch H8/300] : Fix for crashing of gdb with" info f" command

[Patch H8/300] : Fix for crashing of gdb with" info f" command

[Sherry Samuel]

Hi,
	GDB6.1 for H8/300 targets crashes with "info f" command with a breakpoint in main. 

**********************************************************************************************
Try the code below with h8300-elf toolchain
----------------Start of bug.c ---------------------
1 int main()
2 {
3 int i;
4
5 i=0;
6 }
----------------End of bug.c ------------------------

Compile with

h8300-elf-gcc  bug.c -g

Load the a.out file produced into gdb simulator. Put a break point in line 2 of main and run gdb. 
When the control reaches the breakpoint at main, try to get the frame information by the "info f" 
command. The gdb crashes.

While just before entering main, the frame pointer register is not set. It contains 0x00. At this point 
if we try to take the frame information by "info f" command the gdb crashes. After the frame pointer 
register is set, gdb will not crash by the "info f" command.

This bug is not observed for gdb6.1 for SH targets.

**************************************************************************************************

Please find the patch below which fixes the same.

******************CHANGELOG TEXT*******************************************************************
2004-06-18  Sherry Samuel  <SherryS@KPITCummins.com>

	* h8300-tdep.c (h8300_frame_init_saved_regs) : Check whether frame base 
	is null.	
***************************************************************************************************

--- gdb-6.1/gdb/h8300-tdep.c.old	2004-06-17 15:57:18.000000000 +0530
+++ gdb-6.1/gdb/h8300-tdep.c	2004-06-17 17:08:54.000000000 +0530
@@ -500,8 +500,8 @@ h8300_frame_init_saved_regs (struct fram
 	  CORE_ADDR limit = (sal.end && sal.end < get_frame_pc (fi))
 	    ? sal.end : get_frame_pc (fi);
 	  /* This will fill in fields in fi. */
-
-
+	  if(!get_frame_base (fi))//check whether frame is null.
+	  	return null_frame_id;//if null return
 
 	  h8300_examine_prologue (func_addr, limit, get_frame_base (fi),
 				  deprecated_get_frame_saved_regs (fi), fi);


***************************************************************************************

	
Regards,
Sherry Samuel,
KPIT Cummins InfoSystems Ltd.
Pune, India

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Free download of GNU based tool-chains for Renesas' SH and H8 Series.
The following site also offers free technical support to its users. 
Visit http://www.kpitgnutools.com for details. 
Latest versions of KPIT GNU tools are released on June 1, 2004.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 

Re: [Patch H8/300] : Fix for crashing of gdb with" info f" command

[Andrew Cagney]

> ******************CHANGELOG TEXT*******************************************************************
> 2004-06-18  Sherry Samuel  <SherryS@KPITCummins.com>
> 
> 	* h8300-tdep.c (h8300_frame_init_saved_regs) : Check whether frame base 
> 	is null.	

Are you sure this patch is against 6.1 or a current GDB?  The function 
h8300_frame_init_saved_regs has a void return type.

Also, you might want to read over:
   Post GDB 6.2, require new frame code
   http://sources.redhat.com/ml/gdb/2004-06/msg00199.html
As this will affect the h8/300.

Andrew

> 
> --- gdb-6.1/gdb/h8300-tdep.c.old	2004-06-17 15:57:18.000000000 +0530
> +++ gdb-6.1/gdb/h8300-tdep.c	2004-06-17 17:08:54.000000000 +0530
> @@ -500,8 +500,8 @@ h8300_frame_init_saved_regs (struct fram
>  	  CORE_ADDR limit = (sal.end && sal.end < get_frame_pc (fi))
>  	    ? sal.end : get_frame_pc (fi);
>  	  /* This will fill in fields in fi. */
> -
> -
> +	  if(!get_frame_base (fi))//check whether frame is null.
> +	  	return null_frame_id;//if null return
>  
>  	  h8300_examine_prologue (func_addr, limit, get_frame_base (fi),
>  				  deprecated_get_frame_saved_regs (fi), fi);

RE: [Patch H8/300] : Fix for crashing of gdb with" info f" command

[Sherry Samuel]

Hi
Reference:http://sources.redhat.com/ml/gdb-patches/2004-06/msg00428.html

In the latest GDB(gdb+dejagnu-20040630) snap shot, the "info f" problem described in the above link is fixed. Referring through the links you have suggested, it is evident that h8300 target is not in the list of the new GDB frame structure. Is it possible to know what should be done from our side to update H8300 targets to the new frame work? Can you suggest how to go about it?

Thanks in advance.


Sherry Samuel,
KPIT Cummins InfoSystems Ltd.
Pune, India

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Free download of GNU based tool-chains for Renesas' SH and H8 Series.
The following site also offers free technical support to its users. 
Visit http://www.kpitgnutools.com for details. 
Latest versions of KPIT GNU tools are released on June 1, 2004.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 





-----Original Message-----
From: Andrew Cagney [mailto:cagney@gnu.org]
Sent: Wednesday, June 23, 2004 1:39 AM
To: Sherry Samuel
Cc: gdb-patches@sources.redhat.com
Subject: Re: [Patch H8/300] : Fix for crashing of gdb with" info f"
command



> ******************CHANGELOG TEXT*******************************************************************
> 2004-06-18  Sherry Samuel  <SherryS@KPITCummins.com>
> 
> 	* h8300-tdep.c (h8300_frame_init_saved_regs) : Check whether frame base 
> 	is null.	

Are you sure this patch is against 6.1 or a current GDB?  The function 
h8300_frame_init_saved_regs has a void return type.

Also, you might want to read over:
   Post GDB 6.2, require new frame code
   http://sources.redhat.com/ml/gdb/2004-06/msg00199.html
As this will affect the h8/300.

Andrew

> 
> --- gdb-6.1/gdb/h8300-tdep.c.old	2004-06-17 15:57:18.000000000 +0530
> +++ gdb-6.1/gdb/h8300-tdep.c	2004-06-17 17:08:54.000000000 +0530
> @@ -500,8 +500,8 @@ h8300_frame_init_saved_regs (struct fram
>  	  CORE_ADDR limit = (sal.end && sal.end < get_frame_pc (fi))
>  	    ? sal.end : get_frame_pc (fi);
>  	  /* This will fill in fields in fi. */
> -
> -
> +	  if(!get_frame_base (fi))//check whether frame is null.
> +	  	return null_frame_id;//if null return
>  
>  	  h8300_examine_prologue (func_addr, limit, get_frame_base (fi),
>  				  deprecated_get_frame_saved_regs (fi), fi);

Re: [Patch H8/300] : Fix for crashing of gdb with" info f" command

[Andrew Cagney]

> Hi
> Reference:http://sources.redhat.com/ml/gdb-patches/2004-06/msg00428.html
> 
> In the latest GDB(gdb+dejagnu-20040630) snap shot, the "info f" problem described in the above link is fixed. Referring through the links you have suggested, it is evident that h8300 target is not in the list of the new GDB frame structure. Is it possible to know what should be done from our side to update H8300 targets to the new frame work? Can you suggest how to go about it?
> 
> Thanks in advance.

Right,

A very simplistic cookbook is as follows:

- #if 0 all the set_gdbarch_deprecated ... lines
- build gdb
- try gdb on a simple program
   break main, info registers, stepi, nexti, ...
- fix an internal error due to missing method
- repeat

eventually you'll get gdb kind of working, at that point repeat the 
above using the testsuite.

It turns out that for architectures like this that use all the old 
deprecated methods, its quicker / easier to clear the deck and implement 
a new (borrowing heavily from the old code of course).

It also turns out that basic operations (break main; run, stepi, info 
registers) don't even need a frame unwinder - so that can be left as a 
separate pass.

For the unwinder, you need to convert the frame_saved_regs code into the 
frame_unwind_cache code (seen in more up-to-date architectures).

As for contributing this, perhaphs we can just incrementally submit the 
above in realtime?

Andrew