Re: [uml-user] GDB and UML

[Venkatesh P. R. <vra...@co...>]

On Tue, 2003-01-28 at 09:26, Livio Baldini Soares wrote:
>   Hello Venkatesh!
>=20
> Venkatesh Prasad Ranganath writes:
> > Hi,
> >=20
> > I am using UML 2.4.19-48um and trying to debug modules.  The uml file
> > system is RH7.2.  The problem is that I am using umlgdb and everything
> > works fine till the expect magic kicks in on inserting a module.  The
> > dump of *module_list for some reason has init and cleanup members of
> > module_list are 0.  However, the examples on the web have some non-zero
> > values for the these variables.  It is the same even after adding the
> > symbols. However, on continuing and checking again (after insmod
> > completes) init and cleanup members of module_list are non-zero.  Also,
> > notice that the name_user is garbage.  What may be going wrong? =20
>=20
>   What GCC version  are you using? With GCC  2.95 things work normally
> (as described  on the web pages),  but I'm currently  working with GCC
> 3.2 and the same "problems" you describe happen here too.
>=20
>   The workaround is to let the kernel load the module and only stop it
> afterwards:
>=20
> Venkatesh Prasad Ranganath writes:
> > (gdb) c
> > Continuing.
> >=20
> >  *** Module =F8=EA=FF\237\025=C7\004\b\210=F9\005\b loaded ***=20
> > Breakpoint 1, sys_init_module (name_user=3D0x9fffccd8
> > "=F8=EA=FF\237\025=C7\004\b\210=F9\005\b", mod_user=3D0x80a6500)
> >     at module.c:349
> > 349             char *name, *n_name, *name_tmp =3D NULL;
> > (gdb) finish
> > "finish" not meaningful in the outermost frame.
> =20
>   I  don't  know why  this  happens,  probably  GCC 3.2  inlined  some
> functions which were not in GCC 2.95.=20
>=20
>    Forget about it and `c`ontinue the kernel, let the module load, and
> Ctrl-C GDB to stop it. Like this:
>=20
> | Breakpoint 4, sys_init_module (name_user=3D0xa1097c6c "\214|\t=A1B=F3\r=
 ,<# ",=20
> |    mod_user=3D0x80c08d0)
> |    at /home/livio/kernel/versions/linux-2.4.20.kfs/include/linux/sched.=
h:745
> | 745             if (cap_raised(current->cap_effective, cap))
> | (gdb) finish
> | "finish" not meaningful in the outermost frame.
> | (gdb) c
> | Continuing.
>=20
> | Program received signal SIGINT, Interrupt.
> | 0xa018e341 in nanosleep ()
> | (gdb)
>=20
>   And then do as you would normally do (print the module_list, add the
> module symbols, etc).  Works For Me ;-)
>=20
> Venkatesh Prasad Ranganath writes:
> > print *module_list
> > $2 =3D {size_of_struct =3D 96, next =3D 0xa01f4540, name =3D 0xa4820060
> > "ip_tables", size =3D 15424, uc =3D {usecount =3D {
>    ^^^^^^^^^
>=20
>   The name of the module has  been correcly set. But since you haven't
> let sys_init_module() finish executing, the other members haven't been
> filled  in. They  will  be correctly  filled  out if  you execute  the
> procedure I have described.

Thanks for the reply.  I figured out the same. =20

Just FYI: I also fixed umlgdb to take this into account.

--=20

Venkatesh Prasad Ranganath,
Dept. Computing and Information Science,
Kansas State University, US.
web: http://www.cis.ksu.edu/~rvprasad