Re: [lkcd-devel] gcc-3.1 considered harmful

[Andreas H. <AHE...@de...>]

I have seen this problem some days ago, and have a fix for it.
I think only one if-condition has to been changed in kl_stabs.c ...

It is not a problem of gcc-3.1.

IMHO, the stab type entry is correct.

cleanup:(2,13)=*(2,14)=f(2,15)=(2,15),384,32

It is a nested type definition:

cleanup is name of type (2,13)
(2,13) is of type pointer to (2,14)
(2,14) is a function type with return value (2,15)
(2,15) is of type void

BTW, the C void type is defined as itself.

The problem is that currently lcrash is not able to handle
nested type defintions for type void correctly.


(I will provide the fix asap ;-)


Regards,

Andreas




--
Linux for eServer Development
Tel :  +49-7031-16-4640
Notes mail :  Andreas Herrmann/GERMANY/IBM@IBMDE
email :  ahe...@de...



|---------+-------------------------------------->
|         |           SL Baur                    |
|         |           <st...@bs...|
|         |           >                          |
|         |           Sent by:                   |
|         |           lkc...@li...ur|
|         |           ceforge.net                |
|         |                                      |
|         |                                      |
|         |           04/10/02 09:37 AM          |
|         |           Please respond to SL Baur  |
|         |                                      |
|---------+-------------------------------------->
  >------------------------------------------------------------------------------------------------------------------------|
  |                                                                                                                        |
  |       To:       lkc...@li...                                                                       |
  |       cc:                                                                                                              |
  |       Subject:  [lkcd-devel] gcc-3.1 considered harmful                                                                |
  |                                                                                                                        |
  |                                                                                                                        |
  >------------------------------------------------------------------------------------------------------------------------|




gcc-3.1 appears to have a problem with stab generation.  I
attempted to run lcrash on a kernel and Kerntypes compiled with
gcc-3.1 and got the following:

get_embedded_type: stab type references itself!
get_embedded_type: stab type references itself!
get_embedded_type: stab type references itself!
 [repeated]
Segmentation fault (core dumped)
#

A stack backtrace shows that it is infinitely recursing into
find_embedded_type.

The stab entry that it is choking on:
cleanup:(2,13)=*(2,14)=f(2,15)=(2,15),384,32

(from struct module in module.h) is clearly bad, but I think the
error could be handled more gracefully.  The attached one line
patch stops the recursion and lcrash exits.  I get really nervous
when programs run by root can overflow their stacks when fed bad
data.




#### stacktrace has been removed from this note on April 10 2002 by Andreas
Herrmann
#### stab_patch has been removed from this note on April 10 2002 by Andreas
Herrmann