2009-12-23 23:48:20

by Stephen Rothwell

[permalink] [raw]
Subject: linux-next: kgdb tree build warning

Hi Jason,

Today's linux-next build (x86_64_allmodconfig) produced this warning:

kernel/debug/kdb/kdb_bp.c: In function 'kdb_bp':
kernel/debug/kdb/kdb_bp.c:434: warning: the frame size of 32880 bytes is larger than 2048 bytes

Introduced by commit 16f40ed7bd0ad29dcde7a05b8a07f5ba10937277 ("kdb: core
for kgdb back end").

--
Cheers,
Stephen Rothwell [email protected]
http://www.canb.auug.org.au/~sfr/


Attachments:
(No filename) (427.00 B)
(No filename) (198.00 B)
Download all attachments

2009-12-24 04:09:07

by Jason Wessel

[permalink] [raw]
Subject: Re: linux-next: kgdb tree build warning

Stephen Rothwell wrote:
> Hi Jason,
>
> Today's linux-next build (x86_64_allmodconfig) produced this warning:
>
> kernel/debug/kdb/kdb_bp.c: In function 'kdb_bp':
> kernel/debug/kdb/kdb_bp.c:434: warning: the frame size of 32880 bytes is larger than 2048 bytes
>
> Introduced by commit 16f40ed7bd0ad29dcde7a05b8a07f5ba10937277 ("kdb: core
> for kgdb back end").
>
>

Using gcc 4.3, I did not see this error. What compiler or compiler
options did you use to get this?

I assume it means the stack space was exhausted and a casual glance at
the code did not yield anything obvious like a large array on the stack
or anything. It is conceivable that there is an inline function that is
the source of the problem, but I would like to be able to duplicate it.

Thanks,
Jason.

2009-12-24 05:21:26

by Stephen Rothwell

[permalink] [raw]
Subject: Re: linux-next: kgdb tree build warning

Hi Jason,

On Wed, 23 Dec 2009 22:09:00 -0600 Jason Wessel <[email protected]> wrote:
>
> Stephen Rothwell wrote:
> >
> > Today's linux-next build (x86_64_allmodconfig) produced this warning:
> >
> > kernel/debug/kdb/kdb_bp.c: In function 'kdb_bp':
> > kernel/debug/kdb/kdb_bp.c:434: warning: the frame size of 32880 bytes is larger than 2048 bytes
> >
> > Introduced by commit 16f40ed7bd0ad29dcde7a05b8a07f5ba10937277 ("kdb: core
> > for kgdb back end").
>
> Using gcc 4.3, I did not see this error. What compiler or compiler
> options did you use to get this?

I use a gcc 4.4.0 cross sompiler for all the x86_64 builds. No special
compiler options apart from those provided by the allmodconfig build.
It looks like the compiler gets -Wframe-larger-than=${CONFIG_FRAME_WARN}
passed to it (and presumably CONFIG_FRAME_WARN is 2048).

> I assume it means the stack space was exhausted and a casual glance at
> the code did not yield anything obvious like a large array on the stack
> or anything. It is conceivable that there is an inline function that is
> the source of the problem, but I would like to be able to duplicate it.

OK.

That function has a kdb_bp_t on its stack which contains an NR_CPUS array
of pointers. NR_CPUS is 4096 (in this build) and pointers are 8 bytes,
so there is 32K right there :-(

--
Cheers,
Stephen Rothwell [email protected]
http://www.canb.auug.org.au/~sfr/


Attachments:
(No filename) (1.40 kB)
(No filename) (198.00 B)
Download all attachments

2009-12-24 14:44:56

by Jason Wessel

[permalink] [raw]
Subject: Re: linux-next: kgdb tree build warning

Stephen Rothwell wrote:
> Hi Jason,
>
> On Wed, 23 Dec 2009 22:09:00 -0600 Jason Wessel <[email protected]> wrote:
>
>> Stephen Rothwell wrote:
>>
>>> Today's linux-next build (x86_64_allmodconfig) produced this warning:
>>>
>>> kernel/debug/kdb/kdb_bp.c: In function 'kdb_bp':
>>> kernel/debug/kdb/kdb_bp.c:434: warning: the frame size of 32880 bytes is larger than 2048 bytes
>>>
>>>
> That function has a kdb_bp_t on its stack which contains an NR_CPUS array
> of pointers. NR_CPUS is 4096 (in this build) and pointers are 8 bytes,
> so there is 32K right there :-(
>
>

Thank you for the details. I see that I can see the same problem if I
set CONFIG_MAXSMP and use CONFIG_FRAME_WARN=2048. It also allowed me to
test that the problem is really fixed. :-)

I had talked with Martin a while back and he gave me some patches which
cleaned up the breakpoint code (kdb_bp.c). The hw breakpoint support
was implemented a different way in the new kdb/kgdb. You stumbled upon
something that was intended to already be removed from the source.

As a side point you pointed me to an area where I could remove some
other code that should not be there because the debug core is already
handling the operations. The net result is this problem is fixed, and 2
more variables are gone out of that struct.


- int bp_cpu; /* Cpu # (if bp_global == 0) */
- kdbhard_bp_t *bp_hard[NR_CPUS]; /* Hardware breakpoint
structure */
- int bp_adjust; /* Adjustment to PC for real
- * instruction */

Thanks,
Jason.